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