multipathd/main.c

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