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