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 #include <linux/oom.h>
  19 #include <libudev.h>
  20 #ifdef USE_SYSTEMD
  21 #include <systemd/sd-daemon.h>
  22 #endif
  23 #include <semaphore.h>
  24 #include <mpath_cmd.h>
  25 #include <mpath_persist.h>
  26 #include <time.h>
  27
  28 /*
  29  * libcheckers
  30  */
  31 #include <checkers.h>
  32
  33 #ifdef USE_SYSTEMD
  34 static int use_watchdog;
  35 #endif
  36
  37 /*
  38  * libmultipath
  39  */
  40 #include <parser.h>
  41 #include <vector.h>
  42 #include <memory.h>
  43 #include <config.h>
  44 #include <util.h>
  45 #include <hwtable.h>
  46 #include <defaults.h>
  47 #include <structs.h>
  48 #include <blacklist.h>
  49 #include <structs_vec.h>
  50 #include <dmparser.h>
  51 #include <devmapper.h>
  52 #include <sysfs.h>
  53 #include <dict.h>
  54 #include <discovery.h>
  55 #include <debug.h>
  56 #include <propsel.h>
  57 #include <uevent.h>
  58 #include <switchgroup.h>
  59 #include <print.h>
  60 #include <configure.h>
  61 #include <prio.h>
  62 #include <wwids.h>
  63 #include <pgpolicies.h>
  64 #include <uevent.h>
  65 #include <log.h>
  66 #include "prioritizers/alua_rtpg.h"
  67
  68 #include "main.h"
  69 #include "pidfile.h"
  70 #include "uxlsnr.h"
  71 #include "uxclnt.h"
  72 #include "cli.h"
  73 #include "cli_handlers.h"
  74 #include "lock.h"
  75 #include "waiter.h"
  76 #include "wwids.h"
  77
  78 #define FILE_NAME_SIZE 256
  79 #define CMDSIZE 160
  80
  81 #define LOG_MSG(a, b) \
  82 do { \
  83         if (pp->offline) \
  84                 condlog(a, "%s: %s - path offline", pp->mpp->alias, pp->dev); \
  85         else if (strlen(b)) \
  86                 condlog(a, "%s: %s - %s", pp->mpp->alias, pp->dev, b); \
  87 } while(0)
  88
  89 struct mpath_event_param
  90 {
  91         char * devname;
  92         struct multipath *mpp;
  93 };
  94
  95 unsigned int mpath_mx_alloc_len;
  96
  97 int logsink;
  98 enum daemon_status running_state;
  99 pid_t daemon_pid;
 100
 101 static sem_t exit_sem;
 102 /*
 103  * global copy of vecs for use in sig handlers
 104  */
 105 struct vectors * gvecs;
 106
 107 struct udev * udev;
 108
 109 static int
 110 need_switch_pathgroup (struct multipath * mpp, int refresh)
 111 {
 112         struct pathgroup * pgp;
 113         struct path * pp;
 114         unsigned int i, j;
 115
 116         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 117                 return 0;
 118
 119         /*
 120          * Refresh path priority values
 121          */
 122         if (refresh)
 123                 vector_foreach_slot (mpp->pg, pgp, i)
 124                         vector_foreach_slot (pgp->paths, pp, j)
 125                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 126
 127         mpp->bestpg = select_path_group(mpp);
 128
 129         if (mpp->bestpg != mpp->nextpg)
 130                 return 1;
 131
 132         return 0;
 133 }
 134
 135 static void
 136 switch_pathgroup (struct multipath * mpp)
 137 {
 138         mpp->stat_switchgroup++;
 139         dm_switchgroup(mpp->alias, mpp->bestpg);
 140         condlog(2, "%s: switch to path group #%i",
 141                  mpp->alias, mpp->bestpg);
 142 }
 143
 144 static int
 145 coalesce_maps(struct vectors *vecs, vector nmpv)
 146 {
 147         struct multipath * ompp;
 148         vector ompv = vecs->mpvec;
 149         unsigned int i;
 150
 151         vector_foreach_slot (ompv, ompp, i) {
 152                 condlog(3, "%s: coalesce map", ompp->alias);
 153                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 154                         /*
 155                          * remove all current maps not allowed by the
 156                          * current configuration
 157                          */
 158                         if (dm_flush_map(ompp->alias)) {
 159                                 condlog(0, "%s: unable to flush devmap",
 160                                         ompp->alias);
 161                                 /*
 162                                  * may be just because the device is open
 163                                  */
 164                                 if (setup_multipath(vecs, ompp) != 0) {
 165                                         i--;
 166                                         continue;
 167                                 }
 168                                 if (!vector_alloc_slot(nmpv))
 169                                         return 1;
 170
 171                                 vector_set_slot(nmpv, ompp);
 172
 173                                 vector_del_slot(ompv, i);
 174                                 i--;
 175                         }
 176                         else {
 177                                 dm_lib_release();
 178                                 condlog(2, "%s devmap removed", ompp->alias);
 179                         }
 180                 } else if (conf->reassign_maps) {
 181                         condlog(3, "%s: Reassign existing device-mapper"
 182                                 " devices", ompp->alias);
 183                         dm_reassign(ompp->alias);
 184                 }
 185         }
 186         return 0;
 187 }
 188
 189 void
 190 sync_map_state(struct multipath *mpp)
 191 {
 192         struct pathgroup *pgp;
 193         struct path *pp;
 194         unsigned int i, j;
 195
 196         if (!mpp->pg)
 197                 return;
 198
 199         vector_foreach_slot (mpp->pg, pgp, i){
 200                 vector_foreach_slot (pgp->paths, pp, j){
 201                         if (pp->state == PATH_UNCHECKED ||
 202                             pp->state == PATH_WILD ||
 203                             pp->state == PATH_DELAYED)
 204                                 continue;
 205                         if ((pp->dmstate == PSTATE_FAILED ||
 206                              pp->dmstate == PSTATE_UNDEF) &&
 207                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 208                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 209                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 210                                   pp->dmstate == PSTATE_UNDEF) &&
 211                                  (pp->state == PATH_DOWN ||
 212                                   pp->state == PATH_SHAKY))
 213                                 dm_fail_path(mpp->alias, pp->dev_t);
 214                 }
 215         }
 216 }
 217
 218 static void
 219 sync_maps_state(vector mpvec)
 220 {
 221         unsigned int i;
 222         struct multipath *mpp;
 223
 224         vector_foreach_slot (mpvec, mpp, i)
 225                 sync_map_state(mpp);
 226 }
 227
 228 static int
 229 flush_map(struct multipath * mpp, struct vectors * vecs, int nopaths)
 230 {
 231         int r;
 232
 233         if (nopaths)
 234                 r = dm_flush_map_nopaths(mpp->alias, mpp->deferred_remove);
 235         else
 236                 r = dm_flush_map(mpp->alias);
 237         /*
 238          * clear references to this map before flushing so we can ignore
 239          * the spurious uevent we may generate with the dm_flush_map call below
 240          */
 241         if (r) {
 242                 /*
 243                  * May not really be an error -- if the map was already flushed
 244                  * from the device mapper by dmsetup(8) for instance.
 245                  */
 246                 if (r == 1)
 247                         condlog(0, "%s: can't flush", mpp->alias);
 248                 else {
 249                         condlog(2, "%s: devmap deferred remove", mpp->alias);
 250                         mpp->deferred_remove = DEFERRED_REMOVE_IN_PROGRESS;
 251                 }
 252                 return r;
 253         }
 254         else {
 255                 dm_lib_release();
 256                 condlog(2, "%s: map flushed", mpp->alias);
 257         }
 258
 259         orphan_paths(vecs->pathvec, mpp);
 260         remove_map_and_stop_waiter(mpp, vecs, 1);
 261
 262         return 0;
 263 }
 264
 265 int
 266 update_map (struct multipath *mpp, struct vectors *vecs)
 267 {
 268         int retries = 3;
 269         char params[PARAMS_SIZE] = {0};
 270
 271 retry:
 272         condlog(4, "%s: updating new map", mpp->alias);
 273         if (adopt_paths(vecs->pathvec, mpp, 1)) {
 274                 condlog(0, "%s: failed to adopt paths for new map update",
 275                         mpp->alias);
 276                 retries = -1;
 277                 goto fail;
 278         }
 279         verify_paths(mpp, vecs);
 280         mpp->flush_on_last_del = FLUSH_UNDEF;
 281         mpp->action = ACT_RELOAD;
 282
 283         if (setup_map(mpp, params, PARAMS_SIZE)) {
 284                 condlog(0, "%s: failed to setup new map in update", mpp->alias);
 285                 retries = -1;
 286                 goto fail;
 287         }
 288         if (domap(mpp, params) <= 0 && retries-- > 0) {
 289                 condlog(0, "%s: map_udate sleep", mpp->alias);
 290                 sleep(1);
 291                 goto retry;
 292         }
 293         dm_lib_release();
 294
 295 fail:
 296         if (setup_multipath(vecs, mpp))
 297                 return 1;
 298
 299         sync_map_state(mpp);
 300
 301         if (retries < 0)
 302                 condlog(0, "%s: failed reload in new map update", mpp->alias);
 303         return 0;
 304 }
 305
 306 static int
 307 uev_add_map (struct uevent * uev, struct vectors * vecs)
 308 {
 309         char *alias;
 310         int major = -1, minor = -1, rc;
 311
 312         condlog(3, "%s: add map (uevent)", uev->kernel);
 313         alias = uevent_get_dm_name(uev);
 314         if (!alias) {
 315                 condlog(3, "%s: No DM_NAME in uevent", uev->kernel);
 316                 major = uevent_get_major(uev);
 317                 minor = uevent_get_minor(uev);
 318                 alias = dm_mapname(major, minor);
 319                 if (!alias) {
 320                         condlog(2, "%s: mapname not found for %d:%d",
 321                                 uev->kernel, major, minor);
 322                         return 1;
 323                 }
 324         }
 325         rc = ev_add_map(uev->kernel, alias, vecs);
 326         FREE(alias);
 327         return rc;
 328 }
 329
 330 int
 331 ev_add_map (char * dev, char * alias, struct vectors * vecs)
 332 {
 333         char * refwwid;
 334         struct multipath * mpp;
 335         int map_present;
 336         int r = 1;
 337
 338         map_present = dm_map_present(alias);
 339
 340         if (map_present && !dm_is_mpath(alias)) {
 341                 condlog(4, "%s: not a multipath map", alias);
 342                 return 0;
 343         }
 344
 345         mpp = find_mp_by_alias(vecs->mpvec, alias);
 346
 347         if (mpp) {
 348                 if (mpp->wait_for_udev > 1) {
 349                         if (update_map(mpp, vecs))
 350                         /* setup multipathd removed the map */
 351                                 return 1;
 352                 }
 353                 if (mpp->wait_for_udev) {
 354                         mpp->wait_for_udev = 0;
 355                         if (conf->delayed_reconfig &&
 356                             !need_to_delay_reconfig(vecs)) {
 357                                 condlog(2, "reconfigure (delayed)");
 358                                 reconfigure(vecs);
 359                                 return 0;
 360                         }
 361                 }
 362                 /*
 363                  * Not really an error -- we generate our own uevent
 364                  * if we create a multipath mapped device as a result
 365                  * of uev_add_path
 366                  */
 367                 if (conf->reassign_maps) {
 368                         condlog(3, "%s: Reassign existing device-mapper devices",
 369                                 alias);
 370                         dm_reassign(alias);
 371                 }
 372                 return 0;
 373         }
 374         condlog(2, "%s: adding map", alias);
 375
 376         /*
 377          * now we can register the map
 378          */
 379         if (map_present) {
 380                 if ((mpp = add_map_without_path(vecs, alias))) {
 381                         sync_map_state(mpp);
 382                         condlog(2, "%s: devmap %s registered", alias, dev);
 383                         return 0;
 384                 } else {
 385                         condlog(2, "%s: uev_add_map failed", dev);
 386                         return 1;
 387                 }
 388         }
 389         r = get_refwwid(dev, DEV_DEVMAP, vecs->pathvec, &refwwid);
 390
 391         if (refwwid) {
 392                 r = coalesce_paths(vecs, NULL, refwwid, 0);
 393                 dm_lib_release();
 394         }
 395
 396         if (!r)
 397                 condlog(2, "%s: devmap %s added", alias, dev);
 398         else if (r == 2)
 399                 condlog(2, "%s: uev_add_map %s blacklisted", alias, dev);
 400         else
 401                 condlog(0, "%s: uev_add_map %s failed", alias, dev);
 402
 403         FREE(refwwid);
 404         return r;
 405 }
 406
 407 static int
 408 uev_remove_map (struct uevent * uev, struct vectors * vecs)
 409 {
 410         char *alias;
 411         int minor;
 412         struct multipath *mpp;
 413
 414         condlog(2, "%s: remove map (uevent)", uev->kernel);
 415         alias = uevent_get_dm_name(uev);
 416         if (!alias) {
 417                 condlog(3, "%s: No DM_NAME in uevent, ignoring", uev->kernel);
 418                 return 0;
 419         }
 420         minor = uevent_get_minor(uev);
 421         mpp = find_mp_by_minor(vecs->mpvec, minor);
 422
 423         if (!mpp) {
 424                 condlog(2, "%s: devmap not registered, can't remove",
 425                         uev->kernel);
 426                 goto out;
 427         }
 428         if (strcmp(mpp->alias, alias)) {
 429                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 430                         mpp->alias, mpp->dmi->minor, minor);
 431                 goto out;
 432         }
 433
 434         orphan_paths(vecs->pathvec, mpp);
 435         remove_map_and_stop_waiter(mpp, vecs, 1);
 436 out:
 437         FREE(alias);
 438         return 0;
 439 }
 440
 441 int
 442 ev_remove_map (char * devname, char * alias, int minor, struct vectors * vecs)
 443 {
 444         struct multipath * mpp;
 445
 446         mpp = find_mp_by_minor(vecs->mpvec, minor);
 447
 448         if (!mpp) {
 449                 condlog(2, "%s: devmap not registered, can't remove",
 450                         devname);
 451                 return 1;
 452         }
 453         if (strcmp(mpp->alias, alias)) {
 454                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 455                         mpp->alias, mpp->dmi->minor, minor);
 456                 return 1;
 457         }
 458         return flush_map(mpp, vecs, 0);
 459 }
 460
 461 static int
 462 uev_add_path (struct uevent *uev, struct vectors * vecs)
 463 {
 464         struct path *pp;
 465         int ret = 0, i;
 466
 467         condlog(2, "%s: add path (uevent)", uev->kernel);
 468         if (strstr(uev->kernel, "..") != NULL) {
 469                 /*
 470                  * Don't allow relative device names in the pathvec
 471                  */
 472                 condlog(0, "%s: path name is invalid", uev->kernel);
 473                 return 1;
 474         }
 475
 476         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 477         if (pp) {
 478                 int r;
 479
 480                 condlog(0, "%s: spurious uevent, path already in pathvec",
 481                         uev->kernel);
 482                 if (!pp->mpp && !strlen(pp->wwid)) {
 483                         condlog(3, "%s: reinitialize path", uev->kernel);
 484                         udev_device_unref(pp->udev);
 485                         pp->udev = udev_device_ref(uev->udev);
 486                         r = pathinfo(pp, conf->hwtable,
 487                                      DI_ALL | DI_BLACKLIST);
 488                         if (r == PATHINFO_OK)
 489                                 ret = ev_add_path(pp, vecs);
 490                         else if (r == PATHINFO_SKIPPED) {
 491                                 condlog(3, "%s: remove blacklisted path",
 492                                         uev->kernel);
 493                                 i = find_slot(vecs->pathvec, (void *)pp);
 494                                 if (i != -1)
 495                                         vector_del_slot(vecs->pathvec, i);
 496                                 free_path(pp);
 497                         } else {
 498                                 condlog(0, "%s: failed to reinitialize path",
 499                                         uev->kernel);
 500                                 ret = 1;
 501                         }
 502                 }
 503                 return ret;
 504         }
 505
 506         /*
 507          * get path vital state
 508          */
 509         ret = alloc_path_with_pathinfo(conf->hwtable, uev->udev,
 510                                        DI_ALL, &pp);
 511         if (!pp) {
 512                 if (ret == PATHINFO_SKIPPED)
 513                         return 0;
 514                 condlog(3, "%s: failed to get path info", uev->kernel);
 515                 return 1;
 516         }
 517         ret = store_path(vecs->pathvec, pp);
 518         if (!ret) {
 519                 pp->checkint = conf->checkint;
 520                 ret = ev_add_path(pp, vecs);
 521         } else {
 522                 condlog(0, "%s: failed to store path info, "
 523                         "dropping event",
 524                         uev->kernel);
 525                 free_path(pp);
 526                 ret = 1;
 527         }
 528
 529         return ret;
 530 }
 531
 532 /*
 533  * returns:
 534  * 0: added
 535  * 1: error
 536  */
 537 int
 538 ev_add_path (struct path * pp, struct vectors * vecs)
 539 {
 540         struct multipath * mpp;
 541         char params[PARAMS_SIZE] = {0};
 542         int retries = 3;
 543         int start_waiter = 0;
 544         int ret;
 545
 546         /*
 547          * need path UID to go any further
 548          */
 549         if (strlen(pp->wwid) == 0) {
 550                 condlog(0, "%s: failed to get path uid", pp->dev);
 551                 goto fail; /* leave path added to pathvec */
 552         }
 553         mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 554         if (mpp && mpp->wait_for_udev) {
 555                 mpp->wait_for_udev = 2;
 556                 orphan_path(pp, "waiting for create to complete");
 557                 return 0;
 558         }
 559
 560         pp->mpp = mpp;
 561 rescan:
 562         if (mpp) {
 563                 if (mpp->size != pp->size) {
 564                         condlog(0, "%s: failed to add new path %s, "
 565                                 "device size mismatch",
 566                                 mpp->alias, pp->dev);
 567                         int i = find_slot(vecs->pathvec, (void *)pp);
 568                         if (i != -1)
 569                                 vector_del_slot(vecs->pathvec, i);
 570                         free_path(pp);
 571                         return 1;
 572                 }
 573
 574                 condlog(4,"%s: adopting all paths for path %s",
 575                         mpp->alias, pp->dev);
 576                 if (adopt_paths(vecs->pathvec, mpp, 1))
 577                         goto fail; /* leave path added to pathvec */
 578
 579                 verify_paths(mpp, vecs);
 580                 mpp->flush_on_last_del = FLUSH_UNDEF;
 581                 mpp->action = ACT_RELOAD;
 582         } else {
 583                 if (!should_multipath(pp, vecs->pathvec)) {
 584                         orphan_path(pp, "only one path");
 585                         return 0;
 586                 }
 587                 condlog(4,"%s: creating new map", pp->dev);
 588                 if ((mpp = add_map_with_path(vecs, pp, 1))) {
 589                         mpp->action = ACT_CREATE;
 590                         /*
 591                          * We don't depend on ACT_CREATE, as domap will
 592                          * set it to ACT_NOTHING when complete.
 593                          */
 594                         start_waiter = 1;
 595                 }
 596                 else
 597                         goto fail; /* leave path added to pathvec */
 598         }
 599
 600         /* persistent reseravtion check*/
 601         mpath_pr_event_handle(pp);
 602
 603         /*
 604          * push the map to the device-mapper
 605          */
 606         if (setup_map(mpp, params, PARAMS_SIZE)) {
 607                 condlog(0, "%s: failed to setup map for addition of new "
 608                         "path %s", mpp->alias, pp->dev);
 609                 goto fail_map;
 610         }
 611         /*
 612          * reload the map for the multipath mapped device
 613          */
 614 retry:
 615         ret = domap(mpp, params);
 616         if (ret <= 0) {
 617                 if (ret < 0 && retries-- > 0) {
 618                         condlog(0, "%s: retry domap for addition of new "
 619                                 "path %s", mpp->alias, pp->dev);
 620                         sleep(1);
 621                         goto retry;
 622                 }
 623                 condlog(0, "%s: failed in domap for addition of new "
 624                         "path %s", mpp->alias, pp->dev);
 625                 /*
 626                  * deal with asynchronous uevents :((
 627                  */
 628                 if (mpp->action == ACT_RELOAD && retries-- > 0) {
 629                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 630                         sleep(1);
 631                         update_mpp_paths(mpp, vecs->pathvec);
 632                         goto rescan;
 633                 }
 634                 else if (mpp->action == ACT_RELOAD)
 635                         condlog(0, "%s: giving up reload", mpp->alias);
 636                 else
 637                         goto fail_map;
 638         }
 639         dm_lib_release();
 640
 641         /*
 642          * update our state from kernel regardless of create or reload
 643          */
 644         if (setup_multipath(vecs, mpp))
 645                 goto fail; /* if setup_multipath fails, it removes the map */
 646
 647         sync_map_state(mpp);
 648
 649         if ((mpp->action == ACT_CREATE ||
 650              (mpp->action == ACT_NOTHING && start_waiter && !mpp->waiter)) &&
 651             start_waiter_thread(mpp, vecs))
 652                         goto fail_map;
 653
 654         if (retries >= 0) {
 655                 condlog(2, "%s [%s]: path added to devmap %s",
 656                         pp->dev, pp->dev_t, mpp->alias);
 657                 return 0;
 658         }
 659         else
 660                 goto fail;
 661
 662 fail_map:
 663         remove_map(mpp, vecs, 1);
 664 fail:
 665         orphan_path(pp, "failed to add path");
 666         return 1;
 667 }
 668
 669 static int
 670 uev_remove_path (struct uevent *uev, struct vectors * vecs)
 671 {
 672         struct path *pp;
 673
 674         condlog(2, "%s: remove path (uevent)", uev->kernel);
 675         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 676
 677         if (!pp) {
 678                 /* Not an error; path might have been purged earlier */
 679                 condlog(0, "%s: path already removed", uev->kernel);
 680                 return 0;
 681         }
 682
 683         return ev_remove_path(pp, vecs);
 684 }
 685
 686 int
 687 ev_remove_path (struct path *pp, struct vectors * vecs)
 688 {
 689         struct multipath * mpp;
 690         int i, retval = 0;
 691         char params[PARAMS_SIZE] = {0};
 692
 693         /*
 694          * avoid referring to the map of an orphaned path
 695          */
 696         if ((mpp = pp->mpp)) {
 697                 /*
 698                  * transform the mp->pg vector of vectors of paths
 699                  * into a mp->params string to feed the device-mapper
 700                  */
 701                 if (update_mpp_paths(mpp, vecs->pathvec)) {
 702                         condlog(0, "%s: failed to update paths",
 703                                 mpp->alias);
 704                         goto fail;
 705                 }
 706                 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 707                         vector_del_slot(mpp->paths, i);
 708
 709                 /*
 710                  * remove the map IFF removing the last path
 711                  */
 712                 if (VECTOR_SIZE(mpp->paths) == 0) {
 713                         char alias[WWID_SIZE];
 714
 715                         /*
 716                          * flush_map will fail if the device is open
 717                          */
 718                         strncpy(alias, mpp->alias, WWID_SIZE);
 719                         if (mpp->flush_on_last_del == FLUSH_ENABLED) {
 720                                 condlog(2, "%s Last path deleted, disabling queueing", mpp->alias);
 721                                 mpp->retry_tick = 0;
 722                                 mpp->no_path_retry = NO_PATH_RETRY_FAIL;
 723                                 mpp->flush_on_last_del = FLUSH_IN_PROGRESS;
 724                                 dm_queue_if_no_path(mpp->alias, 0);
 725                         }
 726                         if (!flush_map(mpp, vecs, 1)) {
 727                                 condlog(2, "%s: removed map after"
 728                                         " removing all paths",
 729                                         alias);
 730                                 retval = 0;
 731                                 goto out;
 732                         }
 733                         /*
 734                          * Not an error, continue
 735                          */
 736                 }
 737
 738                 if (setup_map(mpp, params, PARAMS_SIZE)) {
 739                         condlog(0, "%s: failed to setup map for"
 740                                 " removal of path %s", mpp->alias, pp->dev);
 741                         goto fail;
 742                 }
 743
 744                 if (mpp->wait_for_udev) {
 745                         mpp->wait_for_udev = 2;
 746                         goto out;
 747                 }
 748
 749                 /*
 750                  * reload the map
 751                  */
 752                 mpp->action = ACT_RELOAD;
 753                 if (domap(mpp, params) <= 0) {
 754                         condlog(0, "%s: failed in domap for "
 755                                 "removal of path %s",
 756                                 mpp->alias, pp->dev);
 757                         retval = 1;
 758                 } else {
 759                         /*
 760                          * update our state from kernel
 761                          */
 762                         if (setup_multipath(vecs, mpp))
 763                                 return 1;
 764                         sync_map_state(mpp);
 765
 766                         condlog(2, "%s [%s]: path removed from map %s",
 767                                 pp->dev, pp->dev_t, mpp->alias);
 768                 }
 769         }
 770
 771 out:
 772         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 773                 vector_del_slot(vecs->pathvec, i);
 774
 775         free_path(pp);
 776
 777         return retval;
 778
 779 fail:
 780         remove_map_and_stop_waiter(mpp, vecs, 1);
 781         return 1;
 782 }
 783
 784 static int
 785 uev_update_path (struct uevent *uev, struct vectors * vecs)
 786 {
 787         int ro, retval = 0;
 788         struct path * pp;
 789
 790         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 791         if (!pp) {
 792                 condlog(0, "%s: spurious uevent, path not found",
 793                         uev->kernel);
 794                 return 1;
 795         }
 796
 797         if (pp->initialized == INIT_REQUESTED_UDEV)
 798                 return uev_add_path(uev, vecs);
 799
 800         ro = uevent_get_disk_ro(uev);
 801
 802         if (ro >= 0) {
 803                 condlog(2, "%s: update path write_protect to '%d' (uevent)",
 804                         uev->kernel, ro);
 805                 if (pp->mpp) {
 806                         if (pp->mpp->wait_for_udev) {
 807                                 pp->mpp->wait_for_udev = 2;
 808                                 return 0;
 809                         }
 810
 811                         retval = reload_map(vecs, pp->mpp, 0);
 812
 813                         condlog(2, "%s: map %s reloaded (retval %d)",
 814                                 uev->kernel, pp->mpp->alias, retval);
 815                 }
 816
 817         }
 818
 819         return retval;
 820 }
 821
 822 static int
 823 map_discovery (struct vectors * vecs)
 824 {
 825         struct multipath * mpp;
 826         unsigned int i;
 827
 828         if (dm_get_maps(vecs->mpvec))
 829                 return 1;
 830
 831         vector_foreach_slot (vecs->mpvec, mpp, i)
 832                 if (setup_multipath(vecs, mpp))
 833                         return 1;
 834
 835         return 0;
 836 }
 837
 838 int
 839 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 840 {
 841         struct vectors * vecs;
 842         int r;
 843
 844         *reply = NULL;
 845         *len = 0;
 846         vecs = (struct vectors *)trigger_data;
 847
 848         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 849         lock(vecs->lock);
 850         pthread_testcancel();
 851
 852         r = parse_cmd(str, reply, len, vecs);
 853
 854         if (r > 0) {
 855                 *reply = STRDUP("fail\n");
 856                 *len = strlen(*reply) + 1;
 857                 r = 1;
 858         }
 859         else if (!r && *len == 0) {
 860                 *reply = STRDUP("ok\n");
 861                 *len = strlen(*reply) + 1;
 862                 r = 0;
 863         }
 864         /* else if (r < 0) leave *reply alone */
 865
 866         lock_cleanup_pop(vecs->lock);
 867         return r;
 868 }
 869
 870 static int
 871 uev_discard(char * devpath)
 872 {
 873         char *tmp;
 874         char a[11], b[11];
 875
 876         /*
 877          * keep only block devices, discard partitions
 878          */
 879         tmp = strstr(devpath, "/block/");
 880         if (tmp == NULL){
 881                 condlog(4, "no /block/ in '%s'", devpath);
 882                 return 1;
 883         }
 884         if (sscanf(tmp, "/block/%10s", a) != 1 ||
 885             sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
 886                 condlog(4, "discard event on %s", devpath);
 887                 return 1;
 888         }
 889         return 0;
 890 }
 891
 892 int
 893 uev_trigger (struct uevent * uev, void * trigger_data)
 894 {
 895         int r = 0;
 896         struct vectors * vecs;
 897
 898         vecs = (struct vectors *)trigger_data;
 899
 900         if (uev_discard(uev->devpath))
 901                 return 0;
 902
 903         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 904         lock(vecs->lock);
 905         pthread_testcancel();
 906
 907         /*
 908          * device map event
 909          * Add events are ignored here as the tables
 910          * are not fully initialised then.
 911          */
 912         if (!strncmp(uev->kernel, "dm-", 3)) {
 913                 if (!strncmp(uev->action, "change", 6)) {
 914                         r = uev_add_map(uev, vecs);
 915                         goto out;
 916                 }
 917                 if (!strncmp(uev->action, "remove", 6)) {
 918                         r = uev_remove_map(uev, vecs);
 919                         goto out;
 920                 }
 921                 goto out;
 922         }
 923
 924         /*
 925          * path add/remove event
 926          */
 927         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 928                            uev->kernel) > 0)
 929                 goto out;
 930
 931         if (!strncmp(uev->action, "add", 3)) {
 932                 r = uev_add_path(uev, vecs);
 933                 goto out;
 934         }
 935         if (!strncmp(uev->action, "remove", 6)) {
 936                 r = uev_remove_path(uev, vecs);
 937                 goto out;
 938         }
 939         if (!strncmp(uev->action, "change", 6)) {
 940                 r = uev_update_path(uev, vecs);
 941                 goto out;
 942         }
 943
 944 out:
 945         lock_cleanup_pop(vecs->lock);
 946         return r;
 947 }
 948
 949 static void *
 950 ueventloop (void * ap)
 951 {
 952         struct udev *udev = ap;
 953
 954         if (uevent_listen(udev))
 955                 condlog(0, "error starting uevent listener");
 956
 957         return NULL;
 958 }
 959
 960 static void *
 961 uevqloop (void * ap)
 962 {
 963         if (uevent_dispatch(&uev_trigger, ap))
 964                 condlog(0, "error starting uevent dispatcher");
 965
 966         return NULL;
 967 }
 968 static void *
 969 uxlsnrloop (void * ap)
 970 {
 971         if (cli_init())
 972                 return NULL;
 973
 974         set_handler_callback(LIST+PATHS, cli_list_paths);
 975         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 976         set_handler_callback(LIST+PATHS+RAW+FMT, cli_list_paths_raw);
 977         set_handler_callback(LIST+PATH, cli_list_path);
 978         set_handler_callback(LIST+MAPS, cli_list_maps);
 979         set_handler_callback(LIST+STATUS, cli_list_status);
 980         set_handler_callback(LIST+DAEMON, cli_list_daemon);
 981         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 982         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 983         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 984         set_handler_callback(LIST+MAPS+RAW+FMT, cli_list_maps_raw);
 985         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 986         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 987         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 988         set_handler_callback(LIST+CONFIG, cli_list_config);
 989         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 990         set_handler_callback(LIST+DEVICES, cli_list_devices);
 991         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 992         set_handler_callback(ADD+PATH, cli_add_path);
 993         set_handler_callback(DEL+PATH, cli_del_path);
 994         set_handler_callback(ADD+MAP, cli_add_map);
 995         set_handler_callback(DEL+MAP, cli_del_map);
 996         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 997         set_handler_callback(RECONFIGURE, cli_reconfigure);
 998         set_handler_callback(SUSPEND+MAP, cli_suspend);
 999         set_handler_callback(RESUME+MAP, cli_resume);
1000         set_handler_callback(RESIZE+MAP, cli_resize);
1001         set_handler_callback(RELOAD+MAP, cli_reload);
1002         set_handler_callback(RESET+MAP, cli_reassign);
1003         set_handler_callback(REINSTATE+PATH, cli_reinstate);
1004         set_handler_callback(FAIL+PATH, cli_fail);
1005         set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
1006         set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
1007         set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
1008         set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
1009         set_handler_callback(QUIT, cli_quit);
1010         set_handler_callback(SHUTDOWN, cli_shutdown);
1011         set_handler_callback(GETPRSTATUS+MAP, cli_getprstatus);
1012         set_handler_callback(SETPRSTATUS+MAP, cli_setprstatus);
1013         set_handler_callback(UNSETPRSTATUS+MAP, cli_unsetprstatus);
1014         set_handler_callback(FORCEQ+DAEMON, cli_force_no_daemon_q);
1015         set_handler_callback(RESTOREQ+DAEMON, cli_restore_no_daemon_q);
1016
1017         umask(077);
1018         uxsock_listen(&uxsock_trigger, ap);
1019
1020         return NULL;
1021 }
1022
1023 void
1024 exit_daemon (void)
1025 {
1026         sem_post(&exit_sem);
1027 }
1028
1029 const char *
1030 daemon_status(void)
1031 {
1032         switch (running_state) {
1033         case DAEMON_INIT:
1034                 return "init";
1035         case DAEMON_START:
1036                 return "startup";
1037         case DAEMON_CONFIGURE:
1038                 return "configure";
1039         case DAEMON_RUNNING:
1040                 return "running";
1041         case DAEMON_SHUTDOWN:
1042                 return "shutdown";
1043         }
1044         return NULL;
1045 }
1046
1047 static void
1048 fail_path (struct path * pp, int del_active)
1049 {
1050         if (!pp->mpp)
1051                 return;
1052
1053         condlog(2, "checker failed path %s in map %s",
1054                  pp->dev_t, pp->mpp->alias);
1055
1056         dm_fail_path(pp->mpp->alias, pp->dev_t);
1057         if (del_active)
1058                 update_queue_mode_del_path(pp->mpp);
1059 }
1060
1061 /*
1062  * caller must have locked the path list before calling that function
1063  */
1064 static int
1065 reinstate_path (struct path * pp, int add_active)
1066 {
1067         int ret = 0;
1068
1069         if (!pp->mpp)
1070                 return 0;
1071
1072         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t)) {
1073                 condlog(0, "%s: reinstate failed", pp->dev_t);
1074                 ret = 1;
1075         } else {
1076                 condlog(2, "%s: reinstated", pp->dev_t);
1077                 if (add_active)
1078                         update_queue_mode_add_path(pp->mpp);
1079         }
1080         return ret;
1081 }
1082
1083 static void
1084 enable_group(struct path * pp)
1085 {
1086         struct pathgroup * pgp;
1087
1088         /*
1089          * if path is added through uev_add_path, pgindex can be unset.
1090          * next update_strings() will set it, upon map reload event.
1091          *
1092          * we can safely return here, because upon map reload, all
1093          * PG will be enabled.
1094          */
1095         if (!pp->mpp->pg || !pp->pgindex)
1096                 return;
1097
1098         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1099
1100         if (pgp->status == PGSTATE_DISABLED) {
1101                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1102                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1103         }
1104 }
1105
1106 static void
1107 mpvec_garbage_collector (struct vectors * vecs)
1108 {
1109         struct multipath * mpp;
1110         unsigned int i;
1111
1112         if (!vecs->mpvec)
1113                 return;
1114
1115         vector_foreach_slot (vecs->mpvec, mpp, i) {
1116                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1117                         condlog(2, "%s: remove dead map", mpp->alias);
1118                         remove_map_and_stop_waiter(mpp, vecs, 1);
1119                         i--;
1120                 }
1121         }
1122 }
1123
1124 /* This is called after a path has started working again. It the multipath
1125  * device for this path uses the followover failback type, and this is the
1126  * best pathgroup, and this is the first path in the pathgroup to come back
1127  * up, then switch to this pathgroup */
1128 static int
1129 followover_should_failback(struct path * pp)
1130 {
1131         struct pathgroup * pgp;
1132         struct path *pp1;
1133         int i;
1134
1135         if (pp->mpp->pgfailback != -FAILBACK_FOLLOWOVER ||
1136             !pp->mpp->pg || !pp->pgindex ||
1137             pp->pgindex != pp->mpp->bestpg)
1138                 return 0;
1139
1140         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1141         vector_foreach_slot(pgp->paths, pp1, i) {
1142                 if (pp1 == pp)
1143                         continue;
1144                 if (pp1->chkrstate != PATH_DOWN && pp1->chkrstate != PATH_SHAKY)
1145                         return 0;
1146         }
1147         return 1;
1148 }
1149
1150 static void
1151 missing_uev_wait_tick(struct vectors *vecs)
1152 {
1153         struct multipath * mpp;
1154         unsigned int i;
1155         int timed_out = 0;
1156
1157         vector_foreach_slot (vecs->mpvec, mpp, i) {
1158                 if (mpp->wait_for_udev && --mpp->uev_wait_tick <= 0) {
1159                         timed_out = 1;
1160                         condlog(0, "%s: timeout waiting on creation uevent. enabling reloads", mpp->alias);
1161                         if (mpp->wait_for_udev > 1 && update_map(mpp, vecs)) {
1162                                 /* update_map removed map */
1163                                 i--;
1164                                 continue;
1165                         }
1166                         mpp->wait_for_udev = 0;
1167                 }
1168         }
1169
1170         if (timed_out && conf->delayed_reconfig &&
1171             !need_to_delay_reconfig(vecs)) {
1172                 condlog(2, "reconfigure (delayed)");
1173                 reconfigure(vecs);
1174         }
1175 }
1176
1177 static void
1178 defered_failback_tick (vector mpvec)
1179 {
1180         struct multipath * mpp;
1181         unsigned int i;
1182
1183         vector_foreach_slot (mpvec, mpp, i) {
1184                 /*
1185                  * defered failback getting sooner
1186                  */
1187                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1188                         mpp->failback_tick--;
1189
1190                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1191                                 switch_pathgroup(mpp);
1192                 }
1193         }
1194 }
1195
1196 static void
1197 retry_count_tick(vector mpvec)
1198 {
1199         struct multipath *mpp;
1200         unsigned int i;
1201
1202         vector_foreach_slot (mpvec, mpp, i) {
1203                 if (mpp->retry_tick) {
1204                         mpp->stat_total_queueing_time++;
1205                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1206                         if(--mpp->retry_tick == 0) {
1207                                 dm_queue_if_no_path(mpp->alias, 0);
1208                                 condlog(2, "%s: Disable queueing", mpp->alias);
1209                         }
1210                 }
1211         }
1212 }
1213
1214 int update_prio(struct path *pp, int refresh_all)
1215 {
1216         int oldpriority;
1217         struct path *pp1;
1218         struct pathgroup * pgp;
1219         int i, j, changed = 0;
1220
1221         if (refresh_all) {
1222                 vector_foreach_slot (pp->mpp->pg, pgp, i) {
1223                         vector_foreach_slot (pgp->paths, pp1, j) {
1224                                 oldpriority = pp1->priority;
1225                                 pathinfo(pp1, conf->hwtable, DI_PRIO);
1226                                 if (pp1->priority != oldpriority)
1227                                         changed = 1;
1228                         }
1229                 }
1230                 return changed;
1231         }
1232         oldpriority = pp->priority;
1233         pathinfo(pp, conf->hwtable, DI_PRIO);
1234
1235         if (pp->priority == oldpriority)
1236                 return 0;
1237         return 1;
1238 }
1239
1240 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
1241 {
1242         if (reload_map(vecs, mpp, refresh))
1243                 return 1;
1244
1245         dm_lib_release();
1246         if (setup_multipath(vecs, mpp) != 0)
1247                 return 1;
1248         sync_map_state(mpp);
1249
1250         return 0;
1251 }
1252
1253 /*
1254  * Returns '1' if the path has been checked, '0' otherwise
1255  */
1256 int
1257 check_path (struct vectors * vecs, struct path * pp)
1258 {
1259         int newstate;
1260         int new_path_up = 0;
1261         int chkr_new_path_up = 0;
1262         int add_active;
1263         int disable_reinstate = 0;
1264         int oldchkrstate = pp->chkrstate;
1265
1266         if ((pp->initialized == INIT_OK ||
1267              pp->initialized == INIT_REQUESTED_UDEV) && !pp->mpp)
1268                 return 0;
1269
1270         if (pp->tick && --pp->tick)
1271                 return 0; /* don't check this path yet */
1272
1273         if (!pp->mpp && pp->initialized == INIT_MISSING_UDEV &&
1274             pp->retriggers < conf->retrigger_tries) {
1275                 condlog(2, "%s: triggering change event to reinitialize",
1276                         pp->dev);
1277                 pp->initialized = INIT_REQUESTED_UDEV;
1278                 pp->retriggers++;
1279                 sysfs_attr_set_value(pp->udev, "uevent", "change",
1280                                      strlen("change"));
1281                 return 0;
1282         }
1283
1284         /*
1285          * provision a next check soonest,
1286          * in case we exit abnormaly from here
1287          */
1288         pp->tick = conf->checkint;
1289
1290         newstate = path_offline(pp);
1291         /*
1292          * Wait for uevent for removed paths;
1293          * some LLDDs like zfcp keep paths unavailable
1294          * without sending uevents.
1295          */
1296         if (newstate == PATH_REMOVED)
1297                 newstate = PATH_DOWN;
1298
1299         if (newstate == PATH_UP)
1300                 newstate = get_state(pp, 1);
1301         else
1302                 checker_clear_message(&pp->checker);
1303
1304         if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1305                 condlog(2, "%s: unusable path", pp->dev);
1306                 pathinfo(pp, conf->hwtable, 0);
1307                 return 1;
1308         }
1309         if (!pp->mpp) {
1310                 if (!strlen(pp->wwid) && pp->initialized != INIT_MISSING_UDEV &&
1311                     (newstate == PATH_UP || newstate == PATH_GHOST)) {
1312                         condlog(2, "%s: add missing path", pp->dev);
1313                         if (pathinfo(pp, conf->hwtable, DI_ALL) == 0) {
1314                                 ev_add_path(pp, vecs);
1315                                 pp->tick = 1;
1316                         }
1317                 }
1318                 return 0;
1319         }
1320         /*
1321          * Async IO in flight. Keep the previous path state
1322          * and reschedule as soon as possible
1323          */
1324         if (newstate == PATH_PENDING) {
1325                 pp->tick = 1;
1326                 return 0;
1327         }
1328         /*
1329          * Synchronize with kernel state
1330          */
1331         if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1332                 condlog(1, "%s: Could not synchronize with kernel state",
1333                         pp->dev);
1334                 pp->dmstate = PSTATE_UNDEF;
1335         }
1336         /* if update_multipath_strings orphaned the path, quit early */
1337         if (!pp->mpp)
1338                 return 0;
1339
1340         if ((newstate == PATH_UP || newstate == PATH_GHOST) &&
1341              pp->wait_checks > 0) {
1342                 if (pp->mpp && pp->mpp->nr_active > 0) {
1343                         pp->state = PATH_DELAYED;
1344                         pp->wait_checks--;
1345                         return 1;
1346                 } else
1347                         pp->wait_checks = 0;
1348         }
1349
1350         /*
1351          * don't reinstate failed path, if its in stand-by
1352          * and if target supports only implicit tpgs mode.
1353          * this will prevent unnecessary i/o by dm on stand-by
1354          * paths if there are no other active paths in map.
1355          */
1356         disable_reinstate = (newstate == PATH_GHOST &&
1357                             pp->mpp->nr_active == 0 &&
1358                             pp->tpgs == TPGS_IMPLICIT) ? 1 : 0;
1359
1360         pp->chkrstate = newstate;
1361         if (newstate != pp->state) {
1362                 int oldstate = pp->state;
1363                 pp->state = newstate;
1364
1365                 if (strlen(checker_message(&pp->checker)))
1366                         LOG_MSG(1, checker_message(&pp->checker));
1367
1368                 /*
1369                  * upon state change, reset the checkint
1370                  * to the shortest delay
1371                  */
1372                 pp->checkint = conf->checkint;
1373
1374                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1375                         /*
1376                          * proactively fail path in the DM
1377                          */
1378                         if (oldstate == PATH_UP ||
1379                             oldstate == PATH_GHOST) {
1380                                 fail_path(pp, 1);
1381                                 if (pp->mpp->delay_wait_checks > 0 &&
1382                                     pp->watch_checks > 0) {
1383                                         pp->wait_checks = pp->mpp->delay_wait_checks;
1384                                         pp->watch_checks = 0;
1385                                 }
1386                         }else
1387                                 fail_path(pp, 0);
1388
1389                         /*
1390                          * cancel scheduled failback
1391                          */
1392                         pp->mpp->failback_tick = 0;
1393
1394                         pp->mpp->stat_path_failures++;
1395                         return 1;
1396                 }
1397
1398                 if(newstate == PATH_UP || newstate == PATH_GHOST){
1399                         if ( pp->mpp && pp->mpp->prflag ){
1400                                 /*
1401                                  * Check Persistent Reservation.
1402                                  */
1403                         condlog(2, "%s: checking persistent reservation "
1404                                 "registration", pp->dev);
1405                         mpath_pr_event_handle(pp);
1406                         }
1407                 }
1408
1409                 /*
1410                  * reinstate this path
1411                  */
1412                 if (oldstate != PATH_UP &&
1413                     oldstate != PATH_GHOST) {
1414                         if (pp->mpp->delay_watch_checks > 0)
1415                                 pp->watch_checks = pp->mpp->delay_watch_checks;
1416                         add_active = 1;
1417                 } else {
1418                         if (pp->watch_checks > 0)
1419                                 pp->watch_checks--;
1420                         add_active = 0;
1421                 }
1422                 if (!disable_reinstate && reinstate_path(pp, add_active)) {
1423                         condlog(3, "%s: reload map", pp->dev);
1424                         ev_add_path(pp, vecs);
1425                         pp->tick = 1;
1426                         return 0;
1427                 }
1428                 new_path_up = 1;
1429
1430                 if (oldchkrstate != PATH_UP && oldchkrstate != PATH_GHOST)
1431                         chkr_new_path_up = 1;
1432
1433                 /*
1434                  * if at least one path is up in a group, and
1435                  * the group is disabled, re-enable it
1436                  */
1437                 if (newstate == PATH_UP)
1438                         enable_group(pp);
1439         }
1440         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1441                 if ((pp->dmstate == PSTATE_FAILED ||
1442                     pp->dmstate == PSTATE_UNDEF) &&
1443                     !disable_reinstate) {
1444                         /* Clear IO errors */
1445                         if (reinstate_path(pp, 0)) {
1446                                 condlog(3, "%s: reload map", pp->dev);
1447                                 ev_add_path(pp, vecs);
1448                                 pp->tick = 1;
1449                                 return 0;
1450                         }
1451                 } else {
1452                         LOG_MSG(4, checker_message(&pp->checker));
1453                         if (pp->checkint != conf->max_checkint) {
1454                                 /*
1455                                  * double the next check delay.
1456                                  * max at conf->max_checkint
1457                                  */
1458                                 if (pp->checkint < (conf->max_checkint / 2))
1459                                         pp->checkint = 2 * pp->checkint;
1460                                 else
1461                                         pp->checkint = conf->max_checkint;
1462
1463                                 condlog(4, "%s: delay next check %is",
1464                                         pp->dev_t, pp->checkint);
1465                         }
1466                         if (pp->watch_checks > 0)
1467                                 pp->watch_checks--;
1468                         pp->tick = pp->checkint;
1469                 }
1470         }
1471         else if (newstate == PATH_DOWN &&
1472                  strlen(checker_message(&pp->checker))) {
1473                 if (conf->log_checker_err == LOG_CHKR_ERR_ONCE)
1474                         LOG_MSG(3, checker_message(&pp->checker));
1475                 else
1476                         LOG_MSG(2, checker_message(&pp->checker));
1477         }
1478
1479         pp->state = newstate;
1480
1481
1482         if (pp->mpp->wait_for_udev)
1483                 return 1;
1484         /*
1485          * path prio refreshing
1486          */
1487         condlog(4, "path prio refresh");
1488
1489         if (update_prio(pp, new_path_up) &&
1490             (pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio) &&
1491              pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1492                 update_path_groups(pp->mpp, vecs, !new_path_up);
1493         else if (need_switch_pathgroup(pp->mpp, 0)) {
1494                 if (pp->mpp->pgfailback > 0 &&
1495                     (new_path_up || pp->mpp->failback_tick <= 0))
1496                         pp->mpp->failback_tick =
1497                                 pp->mpp->pgfailback + 1;
1498                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE ||
1499                          (chkr_new_path_up && followover_should_failback(pp)))
1500                         switch_pathgroup(pp->mpp);
1501         }
1502         return 1;
1503 }
1504
1505 static void *
1506 checkerloop (void *ap)
1507 {
1508         struct vectors *vecs;
1509         struct path *pp;
1510         int count = 0;
1511         unsigned int i;
1512
1513         mlockall(MCL_CURRENT | MCL_FUTURE);
1514         vecs = (struct vectors *)ap;
1515         condlog(2, "path checkers start up");
1516
1517         /*
1518          * init the path check interval
1519          */
1520         vector_foreach_slot (vecs->pathvec, pp, i) {
1521                 pp->checkint = conf->checkint;
1522         }
1523
1524         while (1) {
1525                 struct timeval diff_time, start_time, end_time;
1526                 int num_paths = 0;
1527
1528                 if (gettimeofday(&start_time, NULL) != 0)
1529                         start_time.tv_sec = 0;
1530                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1531                 lock(vecs->lock);
1532                 pthread_testcancel();
1533                 condlog(4, "tick");
1534 #ifdef USE_SYSTEMD
1535                 if (use_watchdog)
1536                         sd_notify(0, "WATCHDOG=1");
1537 #endif
1538                 if (vecs->pathvec) {
1539                         vector_foreach_slot (vecs->pathvec, pp, i) {
1540                                 num_paths += check_path(vecs, pp);
1541                         }
1542                 }
1543                 if (vecs->mpvec) {
1544                         defered_failback_tick(vecs->mpvec);
1545                         retry_count_tick(vecs->mpvec);
1546                         missing_uev_wait_tick(vecs);
1547                 }
1548                 if (count)
1549                         count--;
1550                 else {
1551                         condlog(4, "map garbage collection");
1552                         mpvec_garbage_collector(vecs);
1553                         count = MAPGCINT;
1554                 }
1555
1556                 lock_cleanup_pop(vecs->lock);
1557                 if (start_time.tv_sec &&
1558                     gettimeofday(&end_time, NULL) == 0 &&
1559                     num_paths) {
1560                         timersub(&end_time, &start_time, &diff_time);
1561                         condlog(3, "checked %d path%s in %lu.%06lu secs",
1562                                 num_paths, num_paths > 1 ? "s" : "",
1563                                 diff_time.tv_sec, diff_time.tv_usec);
1564                 }
1565                 sleep(1);
1566         }
1567         return NULL;
1568 }
1569
1570 int
1571 configure (struct vectors * vecs, int start_waiters)
1572 {
1573         struct multipath * mpp;
1574         struct path * pp;
1575         vector mpvec;
1576         int i, ret;
1577
1578         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1579                 return 1;
1580
1581         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1582                 return 1;
1583
1584         if (!(mpvec = vector_alloc()))
1585                 return 1;
1586
1587         /*
1588          * probe for current path (from sysfs) and map (from dm) sets
1589          */
1590         ret = path_discovery(vecs->pathvec, conf, DI_ALL);
1591         if (ret < 0)
1592                 return 1;
1593
1594         vector_foreach_slot (vecs->pathvec, pp, i){
1595                 if (filter_path(conf, pp) > 0){
1596                         vector_del_slot(vecs->pathvec, i);
1597                         free_path(pp);
1598                         i--;
1599                 }
1600                 else
1601                         pp->checkint = conf->checkint;
1602         }
1603         if (map_discovery(vecs))
1604                 return 1;
1605
1606         /*
1607          * create new set of maps & push changed ones into dm
1608          */
1609         if (coalesce_paths(vecs, mpvec, NULL, 1))
1610                 return 1;
1611
1612         /*
1613          * may need to remove some maps which are no longer relevant
1614          * e.g., due to blacklist changes in conf file
1615          */
1616         if (coalesce_maps(vecs, mpvec))
1617                 return 1;
1618
1619         dm_lib_release();
1620
1621         sync_maps_state(mpvec);
1622         vector_foreach_slot(mpvec, mpp, i){
1623                 remember_wwid(mpp->wwid);
1624                 update_map_pr(mpp);
1625         }
1626
1627         /*
1628          * purge dm of old maps
1629          */
1630         remove_maps(vecs);
1631
1632         /*
1633          * save new set of maps formed by considering current path state
1634          */
1635         vector_free(vecs->mpvec);
1636         vecs->mpvec = mpvec;
1637
1638         /*
1639          * start dm event waiter threads for these new maps
1640          */
1641         vector_foreach_slot(vecs->mpvec, mpp, i) {
1642                 if (setup_multipath(vecs, mpp))
1643                         return 1;
1644                 if (start_waiters)
1645                         if (start_waiter_thread(mpp, vecs))
1646                                 return 1;
1647         }
1648         return 0;
1649 }
1650
1651 int
1652 need_to_delay_reconfig(struct vectors * vecs)
1653 {
1654         struct multipath *mpp;
1655         int i;
1656
1657         if (!VECTOR_SIZE(vecs->mpvec))
1658                 return 0;
1659
1660         vector_foreach_slot(vecs->mpvec, mpp, i) {
1661                 if (mpp->wait_for_udev)
1662                         return 1;
1663         }
1664         return 0;
1665 }
1666
1667 int
1668 reconfigure (struct vectors * vecs)
1669 {
1670         struct config * old = conf;
1671         int retval = 1;
1672
1673         running_state = DAEMON_CONFIGURE;
1674
1675         /*
1676          * free old map and path vectors ... they use old conf state
1677          */
1678         if (VECTOR_SIZE(vecs->mpvec))
1679                 remove_maps_and_stop_waiters(vecs);
1680
1681         if (VECTOR_SIZE(vecs->pathvec))
1682                 free_pathvec(vecs->pathvec, FREE_PATHS);
1683
1684         vecs->pathvec = NULL;
1685         conf = NULL;
1686
1687         /* Re-read any timezone changes */
1688         tzset();
1689
1690         if (!load_config(DEFAULT_CONFIGFILE, udev)) {
1691                 dm_drv_version(conf->version, TGT_MPATH);
1692                 conf->verbosity = old->verbosity;
1693                 conf->bindings_read_only = old->bindings_read_only;
1694                 conf->ignore_new_devs = old->ignore_new_devs;
1695                 conf->daemon = 1;
1696                 configure(vecs, 1);
1697                 free_config(old);
1698                 retval = 0;
1699         }
1700
1701         running_state = DAEMON_RUNNING;
1702
1703         return retval;
1704 }
1705
1706 static struct vectors *
1707 init_vecs (void)
1708 {
1709         struct vectors * vecs;
1710
1711         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1712
1713         if (!vecs)
1714                 return NULL;
1715
1716         vecs->lock.mutex =
1717                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1718
1719         if (!vecs->lock.mutex)
1720                 goto out;
1721
1722         pthread_mutex_init(vecs->lock.mutex, NULL);
1723         vecs->lock.depth = 0;
1724
1725         return vecs;
1726
1727 out:
1728         FREE(vecs);
1729         condlog(0, "failed to init paths");
1730         return NULL;
1731 }
1732
1733 static void *
1734 signal_set(int signo, void (*func) (int))
1735 {
1736         int r;
1737         struct sigaction sig;
1738         struct sigaction osig;
1739
1740         sig.sa_handler = func;
1741         sigemptyset(&sig.sa_mask);
1742         sig.sa_flags = 0;
1743
1744         r = sigaction(signo, &sig, &osig);
1745
1746         if (r < 0)
1747                 return (SIG_ERR);
1748         else
1749                 return (osig.sa_handler);
1750 }
1751
1752 void
1753 handle_signals(void)
1754 {
1755         if (reconfig_sig && running_state == DAEMON_RUNNING) {
1756                 pthread_cleanup_push(cleanup_lock,
1757                                 &gvecs->lock);
1758                 lock(gvecs->lock);
1759                 pthread_testcancel();
1760                 if (need_to_delay_reconfig(gvecs)) {
1761                         conf->delayed_reconfig = 1;
1762                         condlog(2, "delaying reconfigure (signal)");
1763                 }
1764                 else {
1765                         condlog(2, "reconfigure (signal)");
1766                         reconfigure(gvecs);
1767                 }
1768                 lock_cleanup_pop(gvecs->lock);
1769         }
1770         if (log_reset_sig) {
1771                 condlog(2, "reset log (signal)");
1772                 pthread_mutex_lock(&logq_lock);
1773                 log_reset("multipathd");
1774                 pthread_mutex_unlock(&logq_lock);
1775         }
1776         reconfig_sig = 0;
1777         log_reset_sig = 0;
1778 }
1779
1780 static void
1781 sighup (int sig)
1782 {
1783         reconfig_sig = 1;
1784 }
1785
1786 static void
1787 sigend (int sig)
1788 {
1789         exit_daemon();
1790 }
1791
1792 static void
1793 sigusr1 (int sig)
1794 {
1795         log_reset_sig = 1;
1796 }
1797
1798 static void
1799 sigusr2 (int sig)
1800 {
1801         condlog(3, "SIGUSR2 received");
1802 }
1803
1804 static void
1805 signal_init(void)
1806 {
1807         sigset_t set;
1808
1809         sigemptyset(&set);
1810         sigaddset(&set, SIGHUP);
1811         sigaddset(&set, SIGUSR1);
1812         sigaddset(&set, SIGUSR2);
1813         pthread_sigmask(SIG_BLOCK, &set, NULL);
1814
1815         signal_set(SIGHUP, sighup);
1816         signal_set(SIGUSR1, sigusr1);
1817         signal_set(SIGUSR2, sigusr2);
1818         signal_set(SIGINT, sigend);
1819         signal_set(SIGTERM, sigend);
1820         signal(SIGPIPE, SIG_IGN);
1821 }
1822
1823 static void
1824 setscheduler (void)
1825 {
1826         int res;
1827         static struct sched_param sched_param = {
1828                 .sched_priority = 99
1829         };
1830
1831         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1832
1833         if (res == -1)
1834                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1835         return;
1836 }
1837
1838 static void
1839 set_oom_adj (void)
1840 {
1841 #ifdef OOM_SCORE_ADJ_MIN
1842         int retry = 1;
1843         char *file = "/proc/self/oom_score_adj";
1844         int score = OOM_SCORE_ADJ_MIN;
1845 #else
1846         int retry = 0;
1847         char *file = "/proc/self/oom_adj";
1848         int score = OOM_ADJUST_MIN;
1849 #endif
1850         FILE *fp;
1851         struct stat st;
1852         char *envp;
1853
1854         envp = getenv("OOMScoreAdjust");
1855         if (envp) {
1856                 condlog(3, "Using systemd provided OOMScoreAdjust");
1857                 return;
1858         }
1859         do {
1860                 if (stat(file, &st) == 0){
1861                         fp = fopen(file, "w");
1862                         if (!fp) {
1863                                 condlog(0, "couldn't fopen %s : %s", file,
1864                                         strerror(errno));
1865                                 return;
1866                         }
1867                         fprintf(fp, "%i", score);
1868                         fclose(fp);
1869                         return;
1870                 }
1871                 if (errno != ENOENT) {
1872                         condlog(0, "couldn't stat %s : %s", file,
1873                                 strerror(errno));
1874                         return;
1875                 }
1876 #ifdef OOM_ADJUST_MIN
1877                 file = "/proc/self/oom_adj";
1878                 score = OOM_ADJUST_MIN;
1879 #else
1880                 retry = 0;
1881 #endif
1882         } while (retry--);
1883         condlog(0, "couldn't adjust oom score");
1884 }
1885
1886 static int
1887 child (void * param)
1888 {
1889         pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1890         pthread_attr_t log_attr, misc_attr, uevent_attr;
1891         struct vectors * vecs;
1892         struct multipath * mpp;
1893         int i;
1894 #ifdef USE_SYSTEMD
1895         unsigned long checkint;
1896 #endif
1897         int rc;
1898         char *envp;
1899
1900         mlockall(MCL_CURRENT | MCL_FUTURE);
1901         sem_init(&exit_sem, 0, 0);
1902         signal_init();
1903
1904         udev = udev_new();
1905
1906         setup_thread_attr(&misc_attr, 64 * 1024, 1);
1907         setup_thread_attr(&uevent_attr, DEFAULT_UEVENT_STACKSIZE * 1024, 1);
1908         setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1909
1910         if (logsink == 1) {
1911                 setup_thread_attr(&log_attr, 64 * 1024, 0);
1912                 log_thread_start(&log_attr);
1913                 pthread_attr_destroy(&log_attr);
1914         }
1915         if (pidfile_create(DEFAULT_PIDFILE, daemon_pid)) {
1916                 condlog(1, "failed to create pidfile");
1917                 if (logsink == 1)
1918                         log_thread_stop();
1919                 exit(1);
1920         }
1921
1922         running_state = DAEMON_START;
1923
1924 #ifdef USE_SYSTEMD
1925         sd_notify(0, "STATUS=startup");
1926 #endif
1927         condlog(2, "--------start up--------");
1928         condlog(2, "read " DEFAULT_CONFIGFILE);
1929
1930         if (load_config(DEFAULT_CONFIGFILE, udev))
1931                 goto failed;
1932
1933         dm_drv_version(conf->version, TGT_MPATH);
1934         if (init_checkers()) {
1935                 condlog(0, "failed to initialize checkers");
1936                 goto failed;
1937         }
1938         if (init_prio()) {
1939                 condlog(0, "failed to initialize prioritizers");
1940                 goto failed;
1941         }
1942
1943         setlogmask(LOG_UPTO(conf->verbosity + 3));
1944
1945         envp = getenv("LimitNOFILE");
1946
1947         if (envp) {
1948                 condlog(2,"Using systemd provided open fds limit of %s", envp);
1949         } else if (conf->max_fds) {
1950                 struct rlimit fd_limit;
1951
1952                 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1953                         condlog(0, "can't get open fds limit: %s",
1954                                 strerror(errno));
1955                         fd_limit.rlim_cur = 0;
1956                         fd_limit.rlim_max = 0;
1957                 }
1958                 if (fd_limit.rlim_cur < conf->max_fds) {
1959                         fd_limit.rlim_cur = conf->max_fds;
1960                         if (fd_limit.rlim_max < conf->max_fds)
1961                                 fd_limit.rlim_max = conf->max_fds;
1962                         if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1963                                 condlog(0, "can't set open fds limit to "
1964                                         "%lu/%lu : %s",
1965                                         fd_limit.rlim_cur, fd_limit.rlim_max,
1966                                         strerror(errno));
1967                         } else {
1968                                 condlog(3, "set open fds limit to %lu/%lu",
1969                                         fd_limit.rlim_cur, fd_limit.rlim_max);
1970                         }
1971                 }
1972
1973         }
1974
1975         vecs = gvecs = init_vecs();
1976         if (!vecs)
1977                 goto failed;
1978
1979         setscheduler();
1980         set_oom_adj();
1981
1982         conf->daemon = 1;
1983         dm_udev_set_sync_support(0);
1984 #ifdef USE_SYSTEMD
1985         envp = getenv("WATCHDOG_USEC");
1986         if (envp && sscanf(envp, "%lu", &checkint) == 1) {
1987                 /* Value is in microseconds */
1988                 conf->max_checkint = checkint / 1000000;
1989                 /* Rescale checkint */
1990                 if (conf->checkint > conf->max_checkint)
1991                         conf->checkint = conf->max_checkint;
1992                 else
1993                         conf->checkint = conf->max_checkint / 4;
1994                 condlog(3, "enabling watchdog, interval %d max %d",
1995                         conf->checkint, conf->max_checkint);
1996                 use_watchdog = conf->checkint;
1997         }
1998 #endif
1999         /*
2000          * Start uevent listener early to catch events
2001          */
2002         if ((rc = pthread_create(&uevent_thr, &uevent_attr, ueventloop, udev))) {
2003                 condlog(0, "failed to create uevent thread: %d", rc);
2004                 goto failed;
2005         }
2006         pthread_attr_destroy(&uevent_attr);
2007         if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
2008                 condlog(0, "failed to create cli listener: %d", rc);
2009                 goto failed;
2010         }
2011         /*
2012          * fetch and configure both paths and multipaths
2013          */
2014 #ifdef USE_SYSTEMD
2015         sd_notify(0, "STATUS=configure");
2016 #endif
2017         running_state = DAEMON_CONFIGURE;
2018
2019         lock(vecs->lock);
2020         if (configure(vecs, 1)) {
2021                 unlock(vecs->lock);
2022                 condlog(0, "failure during configuration");
2023                 goto failed;
2024         }
2025         unlock(vecs->lock);
2026
2027         /*
2028          * start threads
2029          */
2030         if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
2031                 condlog(0,"failed to create checker loop thread: %d", rc);
2032                 goto failed;
2033         }
2034         if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
2035                 condlog(0, "failed to create uevent dispatcher: %d", rc);
2036                 goto failed;
2037         }
2038         pthread_attr_destroy(&misc_attr);
2039
2040         running_state = DAEMON_RUNNING;
2041 #ifdef USE_SYSTEMD
2042         sd_notify(0, "READY=1\nSTATUS=running");
2043 #endif
2044
2045         /*
2046          * exit path
2047          */
2048         while(sem_wait(&exit_sem) != 0); /* Do nothing */
2049
2050 #ifdef USE_SYSTEMD
2051         sd_notify(0, "STATUS=shutdown");
2052 #endif
2053         running_state = DAEMON_SHUTDOWN;
2054         lock(vecs->lock);
2055         if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
2056                 vector_foreach_slot(vecs->mpvec, mpp, i)
2057                         dm_queue_if_no_path(mpp->alias, 0);
2058         remove_maps_and_stop_waiters(vecs);
2059         unlock(vecs->lock);
2060
2061         pthread_cancel(check_thr);
2062         pthread_cancel(uevent_thr);
2063         pthread_cancel(uxlsnr_thr);
2064         pthread_cancel(uevq_thr);
2065
2066         lock(vecs->lock);
2067         free_pathvec(vecs->pathvec, FREE_PATHS);
2068         vecs->pathvec = NULL;
2069         unlock(vecs->lock);
2070         /* Now all the waitevent threads will start rushing in. */
2071         while (vecs->lock.depth > 0) {
2072                 sleep (1); /* This is weak. */
2073                 condlog(3, "Have %d wait event checkers threads to de-alloc,"
2074                         " waiting...", vecs->lock.depth);
2075         }
2076         pthread_mutex_destroy(vecs->lock.mutex);
2077         FREE(vecs->lock.mutex);
2078         vecs->lock.depth = 0;
2079         vecs->lock.mutex = NULL;
2080         FREE(vecs);
2081         vecs = NULL;
2082
2083         cleanup_checkers();
2084         cleanup_prio();
2085
2086         dm_lib_release();
2087         dm_lib_exit();
2088
2089         /* We're done here */
2090         condlog(3, "unlink pidfile");
2091         unlink(DEFAULT_PIDFILE);
2092
2093         condlog(2, "--------shut down-------");
2094
2095         if (logsink == 1)
2096                 log_thread_stop();
2097
2098         /*
2099          * Freeing config must be done after condlog() and dm_lib_exit(),
2100          * because logging functions like dlog() and dm_write_log()
2101          * reference the config.
2102          */
2103         free_config(conf);
2104         conf = NULL;
2105         udev_unref(udev);
2106         udev = NULL;
2107 #ifdef _DEBUG_
2108         dbg_free_final(NULL);
2109 #endif
2110
2111 #ifdef USE_SYSTEMD
2112         sd_notify(0, "ERRNO=0");
2113 #endif
2114         exit(0);
2115
2116 failed:
2117 #ifdef USE_SYSTEMD
2118         sd_notify(0, "ERRNO=1");
2119 #endif
2120         exit(1);
2121 }
2122
2123 static int
2124 daemonize(void)
2125 {
2126         int pid;
2127         int dev_null_fd;
2128
2129         if( (pid = fork()) < 0){
2130                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
2131                 return -1;
2132         }
2133         else if (pid != 0)
2134                 return pid;
2135
2136         setsid();
2137
2138         if ( (pid = fork()) < 0)
2139                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
2140         else if (pid != 0)
2141                 _exit(0);
2142
2143         if (chdir("/") < 0)
2144                 fprintf(stderr, "cannot chdir to '/', continuing\n");
2145
2146         dev_null_fd = open("/dev/null", O_RDWR);
2147         if (dev_null_fd < 0){
2148                 fprintf(stderr, "cannot open /dev/null for input & output : %s\n",
2149                         strerror(errno));
2150                 _exit(0);
2151         }
2152
2153         close(STDIN_FILENO);
2154         if (dup(dev_null_fd) < 0) {
2155                 fprintf(stderr, "cannot dup /dev/null to stdin : %s\n",
2156                         strerror(errno));
2157                 _exit(0);
2158         }
2159         close(STDOUT_FILENO);
2160         if (dup(dev_null_fd) < 0) {
2161                 fprintf(stderr, "cannot dup /dev/null to stdout : %s\n",
2162                         strerror(errno));
2163                 _exit(0);
2164         }
2165         close(STDERR_FILENO);
2166         if (dup(dev_null_fd) < 0) {
2167                 fprintf(stderr, "cannot dup /dev/null to stderr : %s\n",
2168                         strerror(errno));
2169                 _exit(0);
2170         }
2171         close(dev_null_fd);
2172         daemon_pid = getpid();
2173         return 0;
2174 }
2175
2176 int
2177 main (int argc, char *argv[])
2178 {
2179         extern char *optarg;
2180         extern int optind;
2181         int arg;
2182         int err;
2183         int foreground = 0;
2184
2185         logsink = 1;
2186         running_state = DAEMON_INIT;
2187         dm_init();
2188
2189         if (getuid() != 0) {
2190                 fprintf(stderr, "need to be root\n");
2191                 exit(1);
2192         }
2193
2194         /* make sure we don't lock any path */
2195         if (chdir("/") < 0)
2196                 fprintf(stderr, "can't chdir to root directory : %s\n",
2197                         strerror(errno));
2198         umask(umask(077) | 022);
2199
2200         conf = alloc_config();
2201
2202         if (!conf)
2203                 exit(1);
2204
2205         while ((arg = getopt(argc, argv, ":dsv:k::Bn")) != EOF ) {
2206         switch(arg) {
2207                 case 'd':
2208                         foreground = 1;
2209                         if (logsink > 0)
2210                                 logsink = 0;
2211                         //debug=1; /* ### comment me out ### */
2212                         break;
2213                 case 'v':
2214                         if (sizeof(optarg) > sizeof(char *) ||
2215                             !isdigit(optarg[0]))
2216                                 exit(1);
2217
2218                         conf->verbosity = atoi(optarg);
2219                         break;
2220                 case 's':
2221                         logsink = -1;
2222                         break;
2223                 case 'k':
2224                         if (load_config(DEFAULT_CONFIGFILE, udev_new()))
2225                                 exit(1);
2226                         uxclnt(optarg, conf->uxsock_timeout);
2227                         exit(0);
2228                 case 'B':
2229                         conf->bindings_read_only = 1;
2230                         break;
2231                 case 'n':
2232                         conf->ignore_new_devs = 1;
2233                         break;
2234                 default:
2235                         ;
2236                 }
2237         }
2238         if (optind < argc) {
2239                 char cmd[CMDSIZE];
2240                 char * s = cmd;
2241                 char * c = s;
2242
2243                 if (load_config(DEFAULT_CONFIGFILE, udev_new()))
2244                         exit(1);
2245                 while (optind < argc) {
2246                         if (strchr(argv[optind], ' '))
2247                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
2248                         else
2249                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
2250                         optind++;
2251                 }
2252                 c += snprintf(c, s + CMDSIZE - c, "\n");
2253                 uxclnt(s, conf->uxsock_timeout);
2254                 exit(0);
2255         }
2256
2257         if (foreground) {
2258                 if (!isatty(fileno(stdout)))
2259                         setbuf(stdout, NULL);
2260                 err = 0;
2261                 daemon_pid = getpid();
2262         } else
2263                 err = daemonize();
2264
2265         if (err < 0)
2266                 /* error */
2267                 exit(1);
2268         else if (err > 0)
2269                 /* parent dies */
2270                 exit(0);
2271         else
2272                 /* child lives */
2273                 return (child(NULL));
2274 }
2275
2276 void *  mpath_pr_event_handler_fn (void * pathp )
2277 {
2278         struct multipath * mpp;
2279         int i,j, ret, isFound;
2280         struct path * pp = (struct path *)pathp;
2281         unsigned char *keyp;
2282         uint64_t prkey;
2283         struct prout_param_descriptor *param;
2284         struct prin_resp *resp;
2285
2286         mpp = pp->mpp;
2287
2288         resp = mpath_alloc_prin_response(MPATH_PRIN_RKEY_SA);
2289         if (!resp){
2290                 condlog(0,"%s Alloc failed for prin response", pp->dev);
2291                 return NULL;
2292         }
2293
2294         ret = prin_do_scsi_ioctl(pp->dev, MPATH_PRIN_RKEY_SA, resp, 0);
2295         if (ret != MPATH_PR_SUCCESS )
2296         {
2297                 condlog(0,"%s : pr in read keys service action failed. Error=%d", pp->dev, ret);
2298                 goto out;
2299         }
2300
2301         condlog(3, " event pr=%d addlen=%d",resp->prin_descriptor.prin_readkeys.prgeneration,
2302                         resp->prin_descriptor.prin_readkeys.additional_length );
2303
2304         if (resp->prin_descriptor.prin_readkeys.additional_length == 0 )
2305         {
2306                 condlog(1, "%s: No key found. Device may not be registered.", pp->dev);
2307                 ret = MPATH_PR_SUCCESS;
2308                 goto out;
2309         }
2310         prkey = 0;
2311         keyp = (unsigned char *)mpp->reservation_key;
2312         for (j = 0; j < 8; ++j) {
2313                 if (j > 0)
2314                         prkey <<= 8;
2315                 prkey |= *keyp;
2316                 ++keyp;
2317         }
2318         condlog(2, "Multipath  reservation_key: 0x%" PRIx64 " ", prkey);
2319
2320         isFound =0;
2321         for (i = 0; i < resp->prin_descriptor.prin_readkeys.additional_length/8; i++ )
2322         {
2323                 condlog(2, "PR IN READKEYS[%d]  reservation key:",i);
2324                 dumpHex((char *)&resp->prin_descriptor.prin_readkeys.key_list[i*8], 8 , -1);
2325                 if (!memcmp(mpp->reservation_key, &resp->prin_descriptor.prin_readkeys.key_list[i*8], 8))
2326                 {
2327                         condlog(2, "%s: pr key found in prin readkeys response", mpp->alias);
2328                         isFound =1;
2329                         break;
2330                 }
2331         }
2332         if (!isFound)
2333         {
2334                 condlog(0, "%s: Either device not registered or ", pp->dev);
2335                 condlog(0, "host is not authorised for registration. Skip path");
2336                 ret = MPATH_PR_OTHER;
2337                 goto out;
2338         }
2339
2340         param= malloc(sizeof(struct prout_param_descriptor));
2341         memset(param, 0 , sizeof(struct prout_param_descriptor));
2342
2343         for (j = 7; j >= 0; --j) {
2344                 param->sa_key[j] = (prkey & 0xff);
2345                 prkey >>= 8;
2346         }
2347         param->num_transportid = 0;
2348
2349         condlog(3, "device %s:%s", pp->dev, pp->mpp->wwid);
2350
2351         ret = prout_do_scsi_ioctl(pp->dev, MPATH_PROUT_REG_IGN_SA, 0, 0, param, 0);
2352         if (ret != MPATH_PR_SUCCESS )
2353         {
2354                 condlog(0,"%s: Reservation registration failed. Error: %d", pp->dev, ret);
2355         }
2356         mpp->prflag = 1;
2357
2358         free(param);
2359 out:
2360         free(resp);
2361         return NULL;
2362 }
2363
2364 int mpath_pr_event_handle(struct path *pp)
2365 {
2366         pthread_t thread;
2367         int rc;
2368         pthread_attr_t attr;
2369         struct multipath * mpp;
2370
2371         mpp = pp->mpp;
2372
2373         if (!mpp->reservation_key)
2374                 return -1;
2375
2376         pthread_attr_init(&attr);
2377         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
2378
2379         rc = pthread_create(&thread, NULL , mpath_pr_event_handler_fn, pp);
2380         if (rc) {
2381                 condlog(0, "%s: ERROR; return code from pthread_create() is %d", pp->dev, rc);
2382                 return -1;
2383         }
2384         pthread_attr_destroy(&attr);
2385         rc = pthread_join(thread, NULL);
2386         return 0;
2387 }
2388