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         struct udev *udev = ap;
 843
 844         if (uevent_listen(udev))
 845                 condlog(0, "error starting uevent listener");
 846
 847         return NULL;
 848 }
 849
 850 static void *
 851 uevqloop (void * ap)
 852 {
 853         if (uevent_dispatch(&uev_trigger, ap))
 854                 condlog(0, "error starting uevent dispatcher");
 855
 856         return NULL;
 857 }
 858 static void *
 859 uxlsnrloop (void * ap)
 860 {
 861         if (cli_init())
 862                 return NULL;
 863
 864         set_handler_callback(LIST+PATHS, cli_list_paths);
 865         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 866         set_handler_callback(LIST+MAPS, cli_list_maps);
 867         set_handler_callback(LIST+STATUS, cli_list_status);
 868         set_handler_callback(LIST+DAEMON, cli_list_daemon);
 869         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 870         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 871         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 872         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 873         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 874         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 875         set_handler_callback(LIST+CONFIG, cli_list_config);
 876         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 877         set_handler_callback(LIST+DEVICES, cli_list_devices);
 878         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 879         set_handler_callback(ADD+PATH, cli_add_path);
 880         set_handler_callback(DEL+PATH, cli_del_path);
 881         set_handler_callback(ADD+MAP, cli_add_map);
 882         set_handler_callback(DEL+MAP, cli_del_map);
 883         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 884         set_handler_callback(RECONFIGURE, cli_reconfigure);
 885         set_handler_callback(SUSPEND+MAP, cli_suspend);
 886         set_handler_callback(RESUME+MAP, cli_resume);
 887         set_handler_callback(RESIZE+MAP, cli_resize);
 888         set_handler_callback(RELOAD+MAP, cli_reload);
 889         set_handler_callback(RESET+MAP, cli_reassign);
 890         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 891         set_handler_callback(FAIL+PATH, cli_fail);
 892         set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
 893         set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
 894         set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
 895         set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
 896         set_handler_callback(QUIT, cli_quit);
 897         set_handler_callback(SHUTDOWN, cli_shutdown);
 898         set_handler_callback(GETPRSTATUS+MAP, cli_getprstatus);
 899         set_handler_callback(SETPRSTATUS+MAP, cli_setprstatus);
 900         set_handler_callback(UNSETPRSTATUS+MAP, cli_unsetprstatus);
 901         set_handler_callback(FORCEQ+DAEMON, cli_force_no_daemon_q);
 902         set_handler_callback(RESTOREQ+DAEMON, cli_restore_no_daemon_q);
 903
 904         umask(077);
 905         uxsock_listen(&uxsock_trigger, ap);
 906
 907         return NULL;
 908 }
 909
 910 void
 911 exit_daemon (void)
 912 {
 913         sem_post(&exit_sem);
 914 }
 915
 916 const char *
 917 daemon_status(void)
 918 {
 919         switch (running_state) {
 920         case DAEMON_INIT:
 921                 return "init";
 922         case DAEMON_START:
 923                 return "startup";
 924         case DAEMON_CONFIGURE:
 925                 return "configure";
 926         case DAEMON_RUNNING:
 927                 return "running";
 928         case DAEMON_SHUTDOWN:
 929                 return "shutdown";
 930         }
 931         return NULL;
 932 }
 933
 934 static void
 935 fail_path (struct path * pp, int del_active)
 936 {
 937         if (!pp->mpp)
 938                 return;
 939
 940         condlog(2, "checker failed path %s in map %s",
 941                  pp->dev_t, pp->mpp->alias);
 942
 943         dm_fail_path(pp->mpp->alias, pp->dev_t);
 944         if (del_active)
 945                 update_queue_mode_del_path(pp->mpp);
 946 }
 947
 948 /*
 949  * caller must have locked the path list before calling that function
 950  */
 951 static void
 952 reinstate_path (struct path * pp, int add_active)
 953 {
 954         if (!pp->mpp)
 955                 return;
 956
 957         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 958                 condlog(0, "%s: reinstate failed", pp->dev_t);
 959         else {
 960                 condlog(2, "%s: reinstated", pp->dev_t);
 961                 if (add_active)
 962                         update_queue_mode_add_path(pp->mpp);
 963         }
 964 }
 965
 966 static void
 967 enable_group(struct path * pp)
 968 {
 969         struct pathgroup * pgp;
 970
 971         /*
 972          * if path is added through uev_add_path, pgindex can be unset.
 973          * next update_strings() will set it, upon map reload event.
 974          *
 975          * we can safely return here, because upon map reload, all
 976          * PG will be enabled.
 977          */
 978         if (!pp->mpp->pg || !pp->pgindex)
 979                 return;
 980
 981         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 982
 983         if (pgp->status == PGSTATE_DISABLED) {
 984                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 985                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 986         }
 987 }
 988
 989 static void
 990 mpvec_garbage_collector (struct vectors * vecs)
 991 {
 992         struct multipath * mpp;
 993         unsigned int i;
 994
 995         if (!vecs->mpvec)
 996                 return;
 997
 998         vector_foreach_slot (vecs->mpvec, mpp, i) {
 999                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1000                         condlog(2, "%s: remove dead map", mpp->alias);
1001                         remove_map_and_stop_waiter(mpp, vecs, 1);
1002                         i--;
1003                 }
1004         }
1005 }
1006
1007 /* This is called after a path has started working again. It the multipath
1008  * device for this path uses the followover failback type, and this is the
1009  * best pathgroup, and this is the first path in the pathgroup to come back
1010  * up, then switch to this pathgroup */
1011 static int
1012 followover_should_failback(struct path * pp)
1013 {
1014         struct pathgroup * pgp;
1015         struct path *pp1;
1016         int i;
1017
1018         if (pp->mpp->pgfailback != -FAILBACK_FOLLOWOVER ||
1019             !pp->mpp->pg || !pp->pgindex ||
1020             pp->pgindex != pp->mpp->bestpg)
1021                 return 0;
1022
1023         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1024         vector_foreach_slot(pgp->paths, pp1, i) {
1025                 if (pp1 == pp)
1026                         continue;
1027                 if (pp1->chkrstate != PATH_DOWN && pp1->chkrstate != PATH_SHAKY)
1028                         return 0;
1029         }
1030         return 1;
1031 }
1032
1033 static void
1034 defered_failback_tick (vector mpvec)
1035 {
1036         struct multipath * mpp;
1037         unsigned int i;
1038
1039         vector_foreach_slot (mpvec, mpp, i) {
1040                 /*
1041                  * defered failback getting sooner
1042                  */
1043                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1044                         mpp->failback_tick--;
1045
1046                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1047                                 switch_pathgroup(mpp);
1048                 }
1049         }
1050 }
1051
1052 static void
1053 retry_count_tick(vector mpvec)
1054 {
1055         struct multipath *mpp;
1056         unsigned int i;
1057
1058         vector_foreach_slot (mpvec, mpp, i) {
1059                 if (mpp->retry_tick) {
1060                         mpp->stat_total_queueing_time++;
1061                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1062                         if(--mpp->retry_tick == 0) {
1063                                 dm_queue_if_no_path(mpp->alias, 0);
1064                                 condlog(2, "%s: Disable queueing", mpp->alias);
1065                         }
1066                 }
1067         }
1068 }
1069
1070 int update_prio(struct path *pp, int refresh_all)
1071 {
1072         int oldpriority;
1073         struct path *pp1;
1074         struct pathgroup * pgp;
1075         int i, j, changed = 0;
1076
1077         if (refresh_all) {
1078                 vector_foreach_slot (pp->mpp->pg, pgp, i) {
1079                         vector_foreach_slot (pgp->paths, pp1, j) {
1080                                 oldpriority = pp1->priority;
1081                                 pathinfo(pp1, conf->hwtable, DI_PRIO);
1082                                 if (pp1->priority != oldpriority)
1083                                         changed = 1;
1084                         }
1085                 }
1086                 return changed;
1087         }
1088         oldpriority = pp->priority;
1089         pathinfo(pp, conf->hwtable, DI_PRIO);
1090
1091         if (pp->priority == oldpriority)
1092                 return 0;
1093         return 1;
1094 }
1095
1096 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
1097 {
1098         if (reload_map(vecs, mpp, refresh))
1099                 return 1;
1100
1101         dm_lib_release();
1102         if (setup_multipath(vecs, mpp) != 0)
1103                 return 1;
1104         sync_map_state(mpp);
1105
1106         return 0;
1107 }
1108
1109 int
1110 check_path (struct vectors * vecs, struct path * pp)
1111 {
1112         int newstate;
1113         int new_path_up = 0;
1114         int chkr_new_path_up = 0;
1115         int oldchkrstate = pp->chkrstate;
1116
1117         if (!pp->mpp)
1118                 return 0;
1119
1120         if (pp->tick && --pp->tick)
1121                 return 0; /* don't check this path yet */
1122
1123         /*
1124          * provision a next check soonest,
1125          * in case we exit abnormaly from here
1126          */
1127         pp->tick = conf->checkint;
1128
1129         newstate = path_offline(pp);
1130         if (newstate == PATH_REMOVED) {
1131                 condlog(2, "%s: remove path (checker)", pp->dev);
1132                 ev_remove_path(pp, vecs);
1133                 return;
1134         }
1135         if (newstate == PATH_UP)
1136                 newstate = get_state(pp, 1);
1137         else
1138                 checker_clear_message(&pp->checker);
1139
1140         if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1141                 condlog(2, "%s: unusable path", pp->dev);
1142                 pathinfo(pp, conf->hwtable, 0);
1143                 return 1;
1144         }
1145         /*
1146          * Async IO in flight. Keep the previous path state
1147          * and reschedule as soon as possible
1148          */
1149         if (newstate == PATH_PENDING) {
1150                 pp->tick = 1;
1151                 return 0;
1152         }
1153         /*
1154          * Synchronize with kernel state
1155          */
1156         if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1157                 condlog(1, "%s: Could not synchronize with kernel state",
1158                         pp->dev);
1159                 pp->dmstate = PSTATE_UNDEF;
1160         }
1161         pp->chkrstate = newstate;
1162         if (newstate != pp->state) {
1163                 int oldstate = pp->state;
1164                 pp->state = newstate;
1165
1166                 if (strlen(checker_message(&pp->checker)))
1167                         LOG_MSG(1, checker_message(&pp->checker));
1168
1169                 /*
1170                  * upon state change, reset the checkint
1171                  * to the shortest delay
1172                  */
1173                 pp->checkint = conf->checkint;
1174
1175                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1176                         /*
1177                          * proactively fail path in the DM
1178                          */
1179                         if (oldstate == PATH_UP ||
1180                             oldstate == PATH_GHOST)
1181                                 fail_path(pp, 1);
1182                         else
1183                                 fail_path(pp, 0);
1184
1185                         /*
1186                          * cancel scheduled failback
1187                          */
1188                         pp->mpp->failback_tick = 0;
1189
1190                         pp->mpp->stat_path_failures++;
1191                         return 1;
1192                 }
1193
1194                 if(newstate == PATH_UP || newstate == PATH_GHOST){
1195                         if ( pp->mpp && pp->mpp->prflag ){
1196                                 /*
1197                                  * Check Persistent Reservation.
1198                                  */
1199                         condlog(2, "%s: checking persistent reservation "
1200                                 "registration", pp->dev);
1201                         mpath_pr_event_handle(pp);
1202                         }
1203                 }
1204
1205                 /*
1206                  * reinstate this path
1207                  */
1208                 if (oldstate != PATH_UP &&
1209                     oldstate != PATH_GHOST)
1210                         reinstate_path(pp, 1);
1211                 else
1212                         reinstate_path(pp, 0);
1213
1214                 new_path_up = 1;
1215
1216                 if (oldchkrstate != PATH_UP && oldchkrstate != PATH_GHOST)
1217                         chkr_new_path_up = 1;
1218
1219                 /*
1220                  * if at least one path is up in a group, and
1221                  * the group is disabled, re-enable it
1222                  */
1223                 if (newstate == PATH_UP)
1224                         enable_group(pp);
1225         }
1226         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1227                 if (pp->dmstate == PSTATE_FAILED ||
1228                     pp->dmstate == PSTATE_UNDEF) {
1229                         /* Clear IO errors */
1230                         reinstate_path(pp, 0);
1231                 } else {
1232                         LOG_MSG(4, checker_message(&pp->checker));
1233                         if (pp->checkint != conf->max_checkint) {
1234                                 /*
1235                                  * double the next check delay.
1236                                  * max at conf->max_checkint
1237                                  */
1238                                 if (pp->checkint < (conf->max_checkint / 2))
1239                                         pp->checkint = 2 * pp->checkint;
1240                                 else
1241                                         pp->checkint = conf->max_checkint;
1242
1243                                 condlog(4, "%s: delay next check %is",
1244                                         pp->dev_t, pp->checkint);
1245                         }
1246                         pp->tick = pp->checkint;
1247                 }
1248         }
1249         else if (newstate == PATH_DOWN &&
1250                  strlen(checker_message(&pp->checker))) {
1251                 if (conf->log_checker_err == LOG_CHKR_ERR_ONCE)
1252                         LOG_MSG(3, checker_message(&pp->checker));
1253                 else
1254                         LOG_MSG(2, checker_message(&pp->checker));
1255         }
1256
1257         pp->state = newstate;
1258
1259         /*
1260          * path prio refreshing
1261          */
1262         condlog(4, "path prio refresh");
1263
1264         if (update_prio(pp, new_path_up) &&
1265             (pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio) &&
1266              pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1267                 update_path_groups(pp->mpp, vecs, !new_path_up);
1268         else if (need_switch_pathgroup(pp->mpp, 0)) {
1269                 if (pp->mpp->pgfailback > 0 &&
1270                     (new_path_up || pp->mpp->failback_tick <= 0))
1271                         pp->mpp->failback_tick =
1272                                 pp->mpp->pgfailback + 1;
1273                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE ||
1274                          (chkr_new_path_up && followover_should_failback(pp)))
1275                         switch_pathgroup(pp->mpp);
1276         }
1277         return 1;
1278 }
1279
1280 static void *
1281 checkerloop (void *ap)
1282 {
1283         struct vectors *vecs;
1284         struct path *pp;
1285         int count = 0;
1286         unsigned int i;
1287
1288         mlockall(MCL_CURRENT | MCL_FUTURE);
1289         vecs = (struct vectors *)ap;
1290         condlog(2, "path checkers start up");
1291
1292         /*
1293          * init the path check interval
1294          */
1295         vector_foreach_slot (vecs->pathvec, pp, i) {
1296                 pp->checkint = conf->checkint;
1297         }
1298
1299         while (1) {
1300                 struct timeval diff_time, start_time, end_time;
1301                 int num_paths = 0;
1302
1303                 if (gettimeofday(&start_time, NULL) != 0)
1304                         start_time.tv_sec = 0;
1305                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1306                 lock(vecs->lock);
1307                 pthread_testcancel();
1308                 condlog(4, "tick");
1309 #ifdef USE_SYSTEMD
1310                 if (conf->watchdog)
1311                         sd_notify(0, "WATCHDOG=1");
1312 #endif
1313                 if (vecs->pathvec) {
1314                         vector_foreach_slot (vecs->pathvec, pp, i) {
1315                                 num_paths += check_path(vecs, pp);
1316                         }
1317                 }
1318                 if (vecs->mpvec) {
1319                         defered_failback_tick(vecs->mpvec);
1320                         retry_count_tick(vecs->mpvec);
1321                 }
1322                 if (count)
1323                         count--;
1324                 else {
1325                         condlog(4, "map garbage collection");
1326                         mpvec_garbage_collector(vecs);
1327                         count = MAPGCINT;
1328                 }
1329
1330                 lock_cleanup_pop(vecs->lock);
1331                 if (start_time.tv_sec &&
1332                     gettimeofday(&end_time, NULL) == 0 &&
1333                     num_paths) {
1334                         timersub(&end_time, &start_time, &diff_time);
1335                         condlog(3, "checked %d path%s in %lu.%06lu secs",
1336                                 num_paths, num_paths > 1 ? "s" : "",
1337                                 diff_time.tv_sec, diff_time.tv_usec);
1338                 }
1339                 sleep(1);
1340         }
1341         return NULL;
1342 }
1343
1344 int
1345 configure (struct vectors * vecs, int start_waiters)
1346 {
1347         struct multipath * mpp;
1348         struct path * pp;
1349         vector mpvec;
1350         int i;
1351
1352         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1353                 return 1;
1354
1355         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1356                 return 1;
1357
1358         if (!(mpvec = vector_alloc()))
1359                 return 1;
1360
1361         /*
1362          * probe for current path (from sysfs) and map (from dm) sets
1363          */
1364         path_discovery(vecs->pathvec, conf, DI_ALL);
1365
1366         vector_foreach_slot (vecs->pathvec, pp, i){
1367                 if (filter_path(conf, pp) > 0){
1368                         vector_del_slot(vecs->pathvec, i);
1369                         free_path(pp);
1370                         i--;
1371                 }
1372                 else
1373                         pp->checkint = conf->checkint;
1374         }
1375         if (map_discovery(vecs))
1376                 return 1;
1377
1378         /*
1379          * create new set of maps & push changed ones into dm
1380          */
1381         if (coalesce_paths(vecs, mpvec, NULL, 1))
1382                 return 1;
1383
1384         /*
1385          * may need to remove some maps which are no longer relevant
1386          * e.g., due to blacklist changes in conf file
1387          */
1388         if (coalesce_maps(vecs, mpvec))
1389                 return 1;
1390
1391         dm_lib_release();
1392
1393         sync_maps_state(mpvec);
1394         vector_foreach_slot(mpvec, mpp, i){
1395                 remember_wwid(mpp->wwid);
1396                 update_map_pr(mpp);
1397         }
1398
1399         /*
1400          * purge dm of old maps
1401          */
1402         remove_maps(vecs);
1403
1404         /*
1405          * save new set of maps formed by considering current path state
1406          */
1407         vector_free(vecs->mpvec);
1408         vecs->mpvec = mpvec;
1409
1410         /*
1411          * start dm event waiter threads for these new maps
1412          */
1413         vector_foreach_slot(vecs->mpvec, mpp, i) {
1414                 if (setup_multipath(vecs, mpp))
1415                         return 1;
1416                 if (start_waiters)
1417                         if (start_waiter_thread(mpp, vecs))
1418                                 return 1;
1419         }
1420         return 0;
1421 }
1422
1423 int
1424 reconfigure (struct vectors * vecs)
1425 {
1426         struct config * old = conf;
1427         int retval = 1;
1428
1429         /*
1430          * free old map and path vectors ... they use old conf state
1431          */
1432         if (VECTOR_SIZE(vecs->mpvec))
1433                 remove_maps_and_stop_waiters(vecs);
1434
1435         if (VECTOR_SIZE(vecs->pathvec))
1436                 free_pathvec(vecs->pathvec, FREE_PATHS);
1437
1438         vecs->pathvec = NULL;
1439         conf = NULL;
1440
1441         /* Re-read any timezone changes */
1442         tzset();
1443
1444         if (!load_config(DEFAULT_CONFIGFILE, udev)) {
1445                 conf->verbosity = old->verbosity;
1446                 conf->daemon = 1;
1447                 configure(vecs, 1);
1448                 free_config(old);
1449                 retval = 0;
1450         }
1451
1452         return retval;
1453 }
1454
1455 static struct vectors *
1456 init_vecs (void)
1457 {
1458         struct vectors * vecs;
1459
1460         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1461
1462         if (!vecs)
1463                 return NULL;
1464
1465         vecs->lock.mutex =
1466                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1467
1468         if (!vecs->lock.mutex)
1469                 goto out;
1470
1471         pthread_mutex_init(vecs->lock.mutex, NULL);
1472         vecs->lock.depth = 0;
1473
1474         return vecs;
1475
1476 out:
1477         FREE(vecs);
1478         condlog(0, "failed to init paths");
1479         return NULL;
1480 }
1481
1482 static void *
1483 signal_set(int signo, void (*func) (int))
1484 {
1485         int r;
1486         struct sigaction sig;
1487         struct sigaction osig;
1488
1489         sig.sa_handler = func;
1490         sigemptyset(&sig.sa_mask);
1491         sig.sa_flags = 0;
1492
1493         r = sigaction(signo, &sig, &osig);
1494
1495         if (r < 0)
1496                 return (SIG_ERR);
1497         else
1498                 return (osig.sa_handler);
1499 }
1500
1501 void
1502 handle_signals(void)
1503 {
1504         if (reconfig_sig && running_state == DAEMON_RUNNING) {
1505                 condlog(2, "reconfigure (signal)");
1506                 pthread_cleanup_push(cleanup_lock,
1507                                 &gvecs->lock);
1508                 lock(gvecs->lock);
1509                 pthread_testcancel();
1510                 reconfigure(gvecs);
1511                 lock_cleanup_pop(gvecs->lock);
1512         }
1513         if (log_reset_sig) {
1514                 condlog(2, "reset log (signal)");
1515                 pthread_mutex_lock(&logq_lock);
1516                 log_reset("multipathd");
1517                 pthread_mutex_unlock(&logq_lock);
1518         }
1519         reconfig_sig = 0;
1520         log_reset_sig = 0;
1521 }
1522
1523 static void
1524 sighup (int sig)
1525 {
1526         reconfig_sig = 1;
1527 }
1528
1529 static void
1530 sigend (int sig)
1531 {
1532         exit_daemon();
1533 }
1534
1535 static void
1536 sigusr1 (int sig)
1537 {
1538         log_reset_sig = 1;
1539 }
1540
1541 static void
1542 sigusr2 (int sig)
1543 {
1544         condlog(3, "SIGUSR2 received");
1545 }
1546
1547 static void
1548 signal_init(void)
1549 {
1550         sigset_t set;
1551
1552         sigemptyset(&set);
1553         sigaddset(&set, SIGHUP);
1554         sigaddset(&set, SIGUSR1);
1555         sigaddset(&set, SIGUSR2);
1556         pthread_sigmask(SIG_BLOCK, &set, NULL);
1557
1558         signal_set(SIGHUP, sighup);
1559         signal_set(SIGUSR1, sigusr1);
1560         signal_set(SIGUSR2, sigusr2);
1561         signal_set(SIGINT, sigend);
1562         signal_set(SIGTERM, sigend);
1563         signal(SIGPIPE, SIG_IGN);
1564 }
1565
1566 static void
1567 setscheduler (void)
1568 {
1569         int res;
1570         static struct sched_param sched_param = {
1571                 .sched_priority = 99
1572         };
1573
1574         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1575
1576         if (res == -1)
1577                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1578         return;
1579 }
1580
1581 static void
1582 set_oom_adj (void)
1583 {
1584 #ifdef OOM_SCORE_ADJ_MIN
1585         int retry = 1;
1586         char *file = "/proc/self/oom_score_adj";
1587         int score = OOM_SCORE_ADJ_MIN;
1588 #else
1589         int retry = 0;
1590         char *file = "/proc/self/oom_adj";
1591         int score = OOM_ADJUST_MIN;
1592 #endif
1593         FILE *fp;
1594         struct stat st;
1595         char *envp;
1596
1597         envp = getenv("OOMScoreAdjust");
1598         if (envp) {
1599                 condlog(3, "Using systemd provided OOMScoreAdjust");
1600                 return;
1601         }
1602         do {
1603                 if (stat(file, &st) == 0){
1604                         fp = fopen(file, "w");
1605                         if (!fp) {
1606                                 condlog(0, "couldn't fopen %s : %s", file,
1607                                         strerror(errno));
1608                                 return;
1609                         }
1610                         fprintf(fp, "%i", score);
1611                         fclose(fp);
1612                         return;
1613                 }
1614                 if (errno != ENOENT) {
1615                         condlog(0, "couldn't stat %s : %s", file,
1616                                 strerror(errno));
1617                         return;
1618                 }
1619 #ifdef OOM_ADJUST_MIN
1620                 file = "/proc/self/oom_adj";
1621                 score = OOM_ADJUST_MIN;
1622 #else
1623                 retry = 0;
1624 #endif
1625         } while (retry--);
1626         condlog(0, "couldn't adjust oom score");
1627 }
1628
1629 static int
1630 child (void * param)
1631 {
1632         pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1633         pthread_attr_t log_attr, misc_attr, uevent_attr;
1634         struct vectors * vecs;
1635         struct multipath * mpp;
1636         int i;
1637 #ifdef USE_SYSTEMD
1638         unsigned long checkint;
1639 #endif
1640         int rc, pid_rc;
1641         char *envp;
1642
1643         mlockall(MCL_CURRENT | MCL_FUTURE);
1644         sem_init(&exit_sem, 0, 0);
1645         signal_init();
1646
1647         udev = udev_new();
1648
1649         setup_thread_attr(&misc_attr, 64 * 1024, 1);
1650         setup_thread_attr(&uevent_attr, 128 * 1024, 1);
1651         setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1652
1653         if (logsink == 1) {
1654                 setup_thread_attr(&log_attr, 64 * 1024, 0);
1655                 log_thread_start(&log_attr);
1656                 pthread_attr_destroy(&log_attr);
1657         }
1658
1659         running_state = DAEMON_START;
1660
1661 #ifdef USE_SYSTEMD
1662         sd_notify(0, "STATUS=startup");
1663 #endif
1664         condlog(2, "--------start up--------");
1665         condlog(2, "read " DEFAULT_CONFIGFILE);
1666
1667         if (load_config(DEFAULT_CONFIGFILE, udev))
1668                 goto failed;
1669
1670         if (init_checkers()) {
1671                 condlog(0, "failed to initialize checkers");
1672                 goto failed;
1673         }
1674         if (init_prio()) {
1675                 condlog(0, "failed to initialize prioritizers");
1676                 goto failed;
1677         }
1678
1679         setlogmask(LOG_UPTO(conf->verbosity + 3));
1680
1681         envp = getenv("LimitNOFILE");
1682
1683         if (envp) {
1684                 condlog(2,"Using systemd provided open fds limit of %s", envp);
1685         } else if (conf->max_fds) {
1686                 struct rlimit fd_limit;
1687
1688                 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1689                         condlog(0, "can't get open fds limit: %s",
1690                                 strerror(errno));
1691                         fd_limit.rlim_cur = 0;
1692                         fd_limit.rlim_max = 0;
1693                 }
1694                 if (fd_limit.rlim_cur < conf->max_fds) {
1695                         fd_limit.rlim_cur = conf->max_fds;
1696                         if (fd_limit.rlim_max < conf->max_fds)
1697                                 fd_limit.rlim_max = conf->max_fds;
1698                         if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1699                                 condlog(0, "can't set open fds limit to "
1700                                         "%lu/%lu : %s",
1701                                         fd_limit.rlim_cur, fd_limit.rlim_max,
1702                                         strerror(errno));
1703                         } else {
1704                                 condlog(3, "set open fds limit to %lu/%lu",
1705                                         fd_limit.rlim_cur, fd_limit.rlim_max);
1706                         }
1707                 }
1708
1709         }
1710
1711         vecs = gvecs = init_vecs();
1712         if (!vecs)
1713                 goto failed;
1714
1715         setscheduler();
1716         set_oom_adj();
1717
1718         conf->daemon = 1;
1719         udev_set_sync_support(0);
1720 #ifdef USE_SYSTEMD
1721         envp = getenv("WATCHDOG_USEC");
1722         if (envp && sscanf(envp, "%lu", &checkint) == 1) {
1723                 /* Value is in microseconds */
1724                 conf->max_checkint = checkint / 1000000;
1725                 /* Rescale checkint */
1726                 if (conf->checkint > conf->max_checkint)
1727                         conf->checkint = conf->max_checkint;
1728                 else
1729                         conf->checkint = conf->max_checkint / 4;
1730                 condlog(3, "enabling watchdog, interval %d max %d",
1731                         conf->checkint, conf->max_checkint);
1732                 conf->watchdog = conf->checkint;
1733         }
1734 #endif
1735         /*
1736          * Start uevent listener early to catch events
1737          */
1738         if ((rc = pthread_create(&uevent_thr, &uevent_attr, ueventloop, udev))) {
1739                 condlog(0, "failed to create uevent thread: %d", rc);
1740                 goto failed;
1741         }
1742         pthread_attr_destroy(&uevent_attr);
1743         if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
1744                 condlog(0, "failed to create cli listener: %d", rc);
1745                 goto failed;
1746         }
1747         /*
1748          * fetch and configure both paths and multipaths
1749          */
1750 #ifdef USE_SYSTEMD
1751         sd_notify(0, "STATUS=configure");
1752 #endif
1753         running_state = DAEMON_CONFIGURE;
1754
1755         lock(vecs->lock);
1756         if (configure(vecs, 1)) {
1757                 unlock(vecs->lock);
1758                 condlog(0, "failure during configuration");
1759                 goto failed;
1760         }
1761         unlock(vecs->lock);
1762
1763         /*
1764          * start threads
1765          */
1766         if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
1767                 condlog(0,"failed to create checker loop thread: %d", rc);
1768                 goto failed;
1769         }
1770         if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
1771                 condlog(0, "failed to create uevent dispatcher: %d", rc);
1772                 goto failed;
1773         }
1774         pthread_attr_destroy(&misc_attr);
1775
1776         /* Startup complete, create logfile */
1777         pid_rc = pidfile_create(DEFAULT_PIDFILE, daemon_pid);
1778         /* Ignore errors, we can live without */
1779
1780         running_state = DAEMON_RUNNING;
1781 #ifdef USE_SYSTEMD
1782         sd_notify(0, "READY=1\nSTATUS=running");
1783 #endif
1784
1785         /*
1786          * exit path
1787          */
1788         while(sem_wait(&exit_sem) != 0); /* Do nothing */
1789
1790 #ifdef USE_SYSTEMD
1791         sd_notify(0, "STATUS=shutdown");
1792 #endif
1793         running_state = DAEMON_SHUTDOWN;
1794         lock(vecs->lock);
1795         if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
1796                 vector_foreach_slot(vecs->mpvec, mpp, i)
1797                         dm_queue_if_no_path(mpp->alias, 0);
1798         remove_maps_and_stop_waiters(vecs);
1799         unlock(vecs->lock);
1800
1801         pthread_cancel(check_thr);
1802         pthread_cancel(uevent_thr);
1803         pthread_cancel(uxlsnr_thr);
1804         pthread_cancel(uevq_thr);
1805
1806         lock(vecs->lock);
1807         free_pathvec(vecs->pathvec, FREE_PATHS);
1808         vecs->pathvec = NULL;
1809         unlock(vecs->lock);
1810         /* Now all the waitevent threads will start rushing in. */
1811         while (vecs->lock.depth > 0) {
1812                 sleep (1); /* This is weak. */
1813                 condlog(3, "Have %d wait event checkers threads to de-alloc,"
1814                         " waiting...", vecs->lock.depth);
1815         }
1816         pthread_mutex_destroy(vecs->lock.mutex);
1817         FREE(vecs->lock.mutex);
1818         vecs->lock.depth = 0;
1819         vecs->lock.mutex = NULL;
1820         FREE(vecs);
1821         vecs = NULL;
1822
1823         cleanup_checkers();
1824         cleanup_prio();
1825
1826         dm_lib_release();
1827         dm_lib_exit();
1828
1829         /* We're done here */
1830         if (!pid_rc) {
1831                 condlog(3, "unlink pidfile");
1832                 unlink(DEFAULT_PIDFILE);
1833         }
1834
1835         condlog(2, "--------shut down-------");
1836
1837         if (logsink == 1)
1838                 log_thread_stop();
1839
1840         /*
1841          * Freeing config must be done after condlog() and dm_lib_exit(),
1842          * because logging functions like dlog() and dm_write_log()
1843          * reference the config.
1844          */
1845         free_config(conf);
1846         conf = NULL;
1847         udev_unref(udev);
1848         udev = NULL;
1849 #ifdef _DEBUG_
1850         dbg_free_final(NULL);
1851 #endif
1852
1853 #ifdef USE_SYSTEMD
1854         sd_notify(0, "ERRNO=0");
1855 #endif
1856         exit(0);
1857
1858 failed:
1859 #ifdef USE_SYSTEMD
1860         sd_notify(0, "ERRNO=1");
1861 #endif
1862         exit(1);
1863 }
1864
1865 static int
1866 daemonize(void)
1867 {
1868         int pid;
1869         int dev_null_fd;
1870
1871         if( (pid = fork()) < 0){
1872                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1873                 return -1;
1874         }
1875         else if (pid != 0)
1876                 return pid;
1877
1878         setsid();
1879
1880         if ( (pid = fork()) < 0)
1881                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1882         else if (pid != 0)
1883                 _exit(0);
1884
1885         if (chdir("/") < 0)
1886                 fprintf(stderr, "cannot chdir to '/', continuing\n");
1887
1888         dev_null_fd = open("/dev/null", O_RDWR);
1889         if (dev_null_fd < 0){
1890                 fprintf(stderr, "cannot open /dev/null for input & output : %s\n",
1891                         strerror(errno));
1892                 _exit(0);
1893         }
1894
1895         close(STDIN_FILENO);
1896         if (dup(dev_null_fd) < 0) {
1897                 fprintf(stderr, "cannot dup /dev/null to stdin : %s\n",
1898                         strerror(errno));
1899                 _exit(0);
1900         }
1901         close(STDOUT_FILENO);
1902         if (dup(dev_null_fd) < 0) {
1903                 fprintf(stderr, "cannot dup /dev/null to stdout : %s\n",
1904                         strerror(errno));
1905                 _exit(0);
1906         }
1907         close(STDERR_FILENO);
1908         if (dup(dev_null_fd) < 0) {
1909                 fprintf(stderr, "cannot dup /dev/null to stderr : %s\n",
1910                         strerror(errno));
1911                 _exit(0);
1912         }
1913         close(dev_null_fd);
1914         daemon_pid = getpid();
1915         return 0;
1916 }
1917
1918 int
1919 main (int argc, char *argv[])
1920 {
1921         extern char *optarg;
1922         extern int optind;
1923         int arg;
1924         int err;
1925
1926         logsink = 1;
1927         running_state = DAEMON_INIT;
1928         dm_init();
1929
1930         if (getuid() != 0) {
1931                 fprintf(stderr, "need to be root\n");
1932                 exit(1);
1933         }
1934
1935         /* make sure we don't lock any path */
1936         if (chdir("/") < 0)
1937                 fprintf(stderr, "can't chdir to root directory : %s\n",
1938                         strerror(errno));
1939         umask(umask(077) | 022);
1940
1941         conf = alloc_config();
1942
1943         if (!conf)
1944                 exit(1);
1945
1946         while ((arg = getopt(argc, argv, ":dsv:k::")) != EOF ) {
1947         switch(arg) {
1948                 case 'd':
1949                         logsink = 0;
1950                         //debug=1; /* ### comment me out ### */
1951                         break;
1952                 case 'v':
1953                         if (sizeof(optarg) > sizeof(char *) ||
1954                             !isdigit(optarg[0]))
1955                                 exit(1);
1956
1957                         conf->verbosity = atoi(optarg);
1958                         break;
1959                 case 's':
1960                         logsink = -1;
1961                         break;
1962                 case 'k':
1963                         uxclnt(optarg);
1964                         exit(0);
1965                 default:
1966                         ;
1967                 }
1968         }
1969         if (optind < argc) {
1970                 char cmd[CMDSIZE];
1971                 char * s = cmd;
1972                 char * c = s;
1973
1974                 while (optind < argc) {
1975                         if (strchr(argv[optind], ' '))
1976                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1977                         else
1978                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1979                         optind++;
1980                 }
1981                 c += snprintf(c, s + CMDSIZE - c, "\n");
1982                 uxclnt(s);
1983                 exit(0);
1984         }
1985
1986         if (logsink < 1)
1987                 err = 0;
1988         else
1989                 err = daemonize();
1990
1991         if (err < 0)
1992                 /* error */
1993                 exit(1);
1994         else if (err > 0)
1995                 /* parent dies */
1996                 exit(0);
1997         else
1998                 /* child lives */
1999                 return (child(NULL));
2000 }
2001
2002 void *  mpath_pr_event_handler_fn (void * pathp )
2003 {
2004         struct multipath * mpp;
2005         int i,j, ret, isFound;
2006         struct path * pp = (struct path *)pathp;
2007         unsigned char *keyp;
2008         uint64_t prkey;
2009         struct prout_param_descriptor *param;
2010         struct prin_resp *resp;
2011
2012         mpp = pp->mpp;
2013
2014         resp = mpath_alloc_prin_response(MPATH_PRIN_RKEY_SA);
2015         if (!resp){
2016                 condlog(0,"%s Alloc failed for prin response", pp->dev);
2017                 return NULL;
2018         }
2019
2020         ret = prin_do_scsi_ioctl(pp->dev, MPATH_PRIN_RKEY_SA, resp, 0);
2021         if (ret != MPATH_PR_SUCCESS )
2022         {
2023                 condlog(0,"%s : pr in read keys service action failed. Error=%d", pp->dev, ret);
2024                 goto out;
2025         }
2026
2027         condlog(3, " event pr=%d addlen=%d",resp->prin_descriptor.prin_readkeys.prgeneration,
2028                         resp->prin_descriptor.prin_readkeys.additional_length );
2029
2030         if (resp->prin_descriptor.prin_readkeys.additional_length == 0 )
2031         {
2032                 condlog(1, "%s: No key found. Device may not be registered.", pp->dev);
2033                 ret = MPATH_PR_SUCCESS;
2034                 goto out;
2035         }
2036         prkey = 0;
2037         keyp = (unsigned char *)mpp->reservation_key;
2038         for (j = 0; j < 8; ++j) {
2039                 if (j > 0)
2040                         prkey <<= 8;
2041                 prkey |= *keyp;
2042                 ++keyp;
2043         }
2044         condlog(2, "Multipath  reservation_key: 0x%" PRIx64 " ", prkey);
2045
2046         isFound =0;
2047         for (i = 0; i < resp->prin_descriptor.prin_readkeys.additional_length/8; i++ )
2048         {
2049                 condlog(2, "PR IN READKEYS[%d]  reservation key:",i);
2050                 dumpHex((char *)&resp->prin_descriptor.prin_readkeys.key_list[i*8], 8 , -1);
2051                 if (!memcmp(mpp->reservation_key, &resp->prin_descriptor.prin_readkeys.key_list[i*8], 8))
2052                 {
2053                         condlog(2, "%s: pr key found in prin readkeys response", mpp->alias);
2054                         isFound =1;
2055                         break;
2056                 }
2057         }
2058         if (!isFound)
2059         {
2060                 condlog(0, "%s: Either device not registered or ", pp->dev);
2061                 condlog(0, "host is not authorised for registration. Skip path");
2062                 ret = MPATH_PR_OTHER;
2063                 goto out;
2064         }
2065
2066         param= malloc(sizeof(struct prout_param_descriptor));
2067         memset(param, 0 , sizeof(struct prout_param_descriptor));
2068
2069         for (j = 7; j >= 0; --j) {
2070                 param->sa_key[j] = (prkey & 0xff);
2071                 prkey >>= 8;
2072         }
2073         param->num_transportid = 0;
2074
2075         condlog(3, "device %s:%s", pp->dev, pp->mpp->wwid);
2076
2077         ret = prout_do_scsi_ioctl(pp->dev, MPATH_PROUT_REG_IGN_SA, 0, 0, param, 0);
2078         if (ret != MPATH_PR_SUCCESS )
2079         {
2080                 condlog(0,"%s: Reservation registration failed. Error: %d", pp->dev, ret);
2081         }
2082         mpp->prflag = 1;
2083
2084         free(param);
2085 out:
2086         free(resp);
2087         return NULL;
2088 }
2089
2090 int mpath_pr_event_handle(struct path *pp)
2091 {
2092         pthread_t thread;
2093         int rc;
2094         pthread_attr_t attr;
2095         struct multipath * mpp;
2096
2097         mpp = pp->mpp;
2098
2099         if (!mpp->reservation_key)
2100                 return -1;
2101
2102         pthread_attr_init(&attr);
2103         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
2104
2105         rc = pthread_create(&thread, NULL , mpath_pr_event_handler_fn, pp);
2106         if (rc) {
2107                 condlog(0, "%s: ERROR; return code from pthread_create() is %d", pp->dev, rc);
2108                 return -1;
2109         }
2110         pthread_attr_destroy(&attr);
2111         rc = pthread_join(thread, NULL);
2112         return 0;
2113 }
2114