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