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