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
  18 /*
  19  * libcheckers
  20  */
  21 #include <checkers.h>
  22
  23 /*
  24  * libmultipath
  25  */
  26 #include <parser.h>
  27 #include <vector.h>
  28 #include <memory.h>
  29 #include <config.h>
  30 #include <util.h>
  31 #include <hwtable.h>
  32 #include <defaults.h>
  33 #include <structs.h>
  34 #include <callout.h>
  35 #include <blacklist.h>
  36 #include <structs_vec.h>
  37 #include <dmparser.h>
  38 #include <devmapper.h>
  39 #include <sysfs.h>
  40 #include <dict.h>
  41 #include <discovery.h>
  42 #include <debug.h>
  43 #include <propsel.h>
  44 #include <uevent.h>
  45 #include <switchgroup.h>
  46 #include <print.h>
  47 #include <configure.h>
  48 #include <prio.h>
  49
  50 #include "main.h"
  51 #include "pidfile.h"
  52 #include "uxlsnr.h"
  53 #include "uxclnt.h"
  54 #include "cli.h"
  55 #include "cli_handlers.h"
  56 #include "lock.h"
  57 #include "waiter.h"
  58
  59 #define FILE_NAME_SIZE 256
  60 #define CMDSIZE 160
  61
  62 #define LOG_MSG(a,b) \
  63         if (strlen(b)) condlog(a, "%s: %s", pp->dev, b);
  64
  65 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  66 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  67
  68 int logsink;
  69
  70 /*
  71  * global copy of vecs for use in sig handlers
  72  */
  73 struct vectors * gvecs;
  74
  75 static int
  76 need_switch_pathgroup (struct multipath * mpp, int refresh)
  77 {
  78         struct pathgroup * pgp;
  79         struct path * pp;
  80         unsigned int i, j;
  81
  82         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  83                 return 0;
  84
  85         /*
  86          * Refresh path priority values
  87          */
  88         if (refresh)
  89                 vector_foreach_slot (mpp->pg, pgp, i)
  90                         vector_foreach_slot (pgp->paths, pp, j)
  91                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  92
  93         mpp->bestpg = select_path_group(mpp);
  94
  95         if (mpp->bestpg != mpp->nextpg)
  96                 return 1;
  97
  98         return 0;
  99 }
 100
 101 static void
 102 switch_pathgroup (struct multipath * mpp)
 103 {
 104         mpp->stat_switchgroup++;
 105         dm_switchgroup(mpp->alias, mpp->bestpg);
 106         condlog(2, "%s: switch to path group #%i",
 107                  mpp->alias, mpp->bestpg);
 108 }
 109
 110 static int
 111 coalesce_maps(struct vectors *vecs, vector nmpv)
 112 {
 113         struct multipath * ompp;
 114         vector ompv = vecs->mpvec;
 115         unsigned int i;
 116         int j;
 117
 118         vector_foreach_slot (ompv, ompp, i) {
 119                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 120                         /*
 121                          * remove all current maps not allowed by the
 122                          * current configuration
 123                          */
 124                         if (dm_flush_map(ompp->alias)) {
 125                                 condlog(0, "%s: unable to flush devmap",
 126                                         ompp->alias);
 127                                 /*
 128                                  * may be just because the device is open
 129                                  */
 130                                 if (!vector_alloc_slot(nmpv))
 131                                         return 1;
 132
 133                                 vector_set_slot(nmpv, ompp);
 134                                 setup_multipath(vecs, ompp);
 135
 136                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 137                                         vector_del_slot(ompv, j);
 138
 139                                 continue;
 140                         }
 141                         else {
 142                                 dm_lib_release();
 143                                 condlog(2, "%s devmap removed", ompp->alias);
 144                         }
 145                 }
 146         }
 147         return 0;
 148 }
 149
 150 void
 151 sync_map_state(struct multipath *mpp)
 152 {
 153         struct pathgroup *pgp;
 154         struct path *pp;
 155         unsigned int i, j;
 156
 157         if (!mpp->pg)
 158                 return;
 159
 160         vector_foreach_slot (mpp->pg, pgp, i){
 161                 vector_foreach_slot (pgp->paths, pp, j){
 162                         if (pp->state == PATH_UNCHECKED ||
 163                             pp->state == PATH_WILD)
 164                                 continue;
 165                         if ((pp->dmstate == PSTATE_FAILED ||
 166                              pp->dmstate == PSTATE_UNDEF) &&
 167                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 168                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 169                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 170                                   pp->dmstate == PSTATE_UNDEF) &&
 171                                  (pp->state == PATH_DOWN ||
 172                                   pp->state == PATH_SHAKY))
 173                                 dm_fail_path(mpp->alias, pp->dev_t);
 174                 }
 175         }
 176 }
 177
 178 static void
 179 sync_maps_state(vector mpvec)
 180 {
 181         unsigned int i;
 182         struct multipath *mpp;
 183
 184         vector_foreach_slot (mpvec, mpp, i)
 185                 sync_map_state(mpp);
 186 }
 187
 188 static int
 189 flush_map(struct multipath * mpp, struct vectors * vecs)
 190 {
 191         /*
 192          * clear references to this map before flushing so we can ignore
 193          * the spurious uevent we may generate with the dm_flush_map call below
 194          */
 195         if (dm_flush_map(mpp->alias)) {
 196                 /*
 197                  * May not really be an error -- if the map was already flushed
 198                  * from the device mapper by dmsetup(8) for instance.
 199                  */
 200                 condlog(0, "%s: can't flush", mpp->alias);
 201                 return 1;
 202         }
 203         else {
 204                 dm_lib_release();
 205                 condlog(2, "%s: devmap removed", mpp->alias);
 206         }
 207
 208         orphan_paths(vecs->pathvec, mpp);
 209         remove_map_and_stop_waiter(mpp, vecs, 1);
 210
 211         return 0;
 212 }
 213
 214 static int
 215 uev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 216 {
 217         condlog(2, "%s: add map (uevent)", dev->kernel);
 218         return ev_add_map(dev, vecs);
 219 }
 220
 221 int
 222 ev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 223 {
 224         char * alias;
 225         char *dev_t;
 226         int major, minor;
 227         char * refwwid;
 228         struct multipath * mpp;
 229         int map_present;
 230         int r = 1;
 231
 232         dev_t = sysfs_attr_get_value(dev->devpath, "dev");
 233
 234         if (!dev_t || sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 235                 return 1;
 236
 237         alias = dm_mapname(major, minor);
 238
 239         if (!alias)
 240                 return 1;
 241
 242         map_present = dm_map_present(alias);
 243
 244         if (map_present && dm_type(alias, TGT_MPATH) <= 0) {
 245                 condlog(4, "%s: not a multipath map", alias);
 246                 FREE(alias);
 247                 return 0;
 248         }
 249
 250         mpp = find_mp_by_alias(vecs->mpvec, alias);
 251
 252         if (mpp) {
 253                 /*
 254                  * Not really an error -- we generate our own uevent
 255                  * if we create a multipath mapped device as a result
 256                  * of uev_add_path
 257                  */
 258                 condlog(0, "%s: devmap already registered",
 259                         dev->kernel);
 260                 FREE(alias);
 261                 return 0;
 262         }
 263
 264         /*
 265          * now we can register the map
 266          */
 267         if (map_present && (mpp = add_map_without_path(vecs, minor, alias))) {
 268                 sync_map_state(mpp);
 269                 condlog(2, "%s: devmap %s added", alias, dev->kernel);
 270                 return 0;
 271         }
 272         refwwid = get_refwwid(dev->kernel, DEV_DEVMAP, vecs->pathvec);
 273
 274         if (refwwid) {
 275                 r = coalesce_paths(vecs, NULL, refwwid, 0);
 276                 dm_lib_release();
 277         }
 278
 279         if (!r)
 280                 condlog(2, "%s: devmap %s added", alias, dev->kernel);
 281         else
 282                 condlog(0, "%s: uev_add_map %s failed", alias, dev->kernel);
 283
 284         FREE(refwwid);
 285         FREE(alias);
 286         return r;
 287 }
 288
 289 static int
 290 uev_remove_map (struct sysfs_device * dev, struct vectors * vecs)
 291 {
 292         condlog(2, "%s: remove map (uevent)", dev->kernel);
 293         return ev_remove_map(dev->kernel, vecs);
 294 }
 295
 296 int
 297 ev_remove_map (char * devname, struct vectors * vecs)
 298 {
 299         struct multipath * mpp;
 300
 301         mpp = find_mp_by_str(vecs->mpvec, devname);
 302
 303         if (!mpp) {
 304                 condlog(2, "%s: devmap not registered, can't remove",
 305                         devname);
 306                 return 0;
 307         }
 308         flush_map(mpp, vecs);
 309
 310         return 0;
 311 }
 312
 313 static int
 314 uev_umount_map (struct sysfs_device * dev, struct vectors * vecs)
 315 {
 316         struct multipath * mpp;
 317
 318         condlog(2, "%s: umount map (uevent)", dev->kernel);
 319
 320         mpp = find_mp_by_str(vecs->mpvec, dev->kernel);
 321
 322         if (!mpp)
 323                 return 0;
 324
 325         update_mpp_paths(mpp, vecs->pathvec);
 326         verify_paths(mpp, vecs, NULL);
 327
 328         if (!VECTOR_SIZE(mpp->paths))
 329                 flush_map(mpp, vecs);
 330
 331         return 0;
 332 }
 333
 334 static int
 335 uev_add_path (struct sysfs_device * dev, struct vectors * vecs)
 336 {
 337         condlog(2, "%s: add path (uevent)", dev->kernel);
 338         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 339 }
 340
 341
 342 /*
 343  * returns:
 344  * 0: added
 345  * 1: error
 346  * 2: blacklisted
 347  */
 348 int
 349 ev_add_path (char * devname, struct vectors * vecs)
 350 {
 351         struct multipath * mpp;
 352         struct path * pp;
 353         char empty_buff[WWID_SIZE] = {0};
 354
 355         pp = find_path_by_dev(vecs->pathvec, devname);
 356
 357         if (pp) {
 358                 condlog(0, "%s: spurious uevent, path already in pathvec",
 359                         devname);
 360                 if (pp->mpp)
 361                         return 0;
 362         }
 363         else {
 364                 /*
 365                  * get path vital state
 366                  */
 367                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 368                       devname, DI_ALL))) {
 369                         condlog(0, "%s: failed to store path info", devname);
 370                         return 1;
 371                 }
 372                 pp->checkint = conf->checkint;
 373         }
 374
 375         /*
 376          * need path UID to go any further
 377          */
 378         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 379                 condlog(0, "%s: failed to get path uid", devname);
 380                 return 1; /* leave path added to pathvec */
 381         }
 382         if (filter_path(conf, pp) > 0){
 383                 int i = find_slot(vecs->pathvec, (void *)pp);
 384                 if (i != -1)
 385                         vector_del_slot(vecs->pathvec, i);
 386                 free_path(pp);
 387                 return 2;
 388         }
 389         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 390 rescan:
 391         if (mpp) {
 392                 if (adopt_paths(vecs->pathvec, mpp))
 393                         return 1; /* leave path added to pathvec */
 394
 395                 verify_paths(mpp, vecs, NULL);
 396                 mpp->action = ACT_RELOAD;
 397         }
 398         else {
 399                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 400                         mpp->action = ACT_CREATE;
 401                 else
 402                         return 1; /* leave path added to pathvec */
 403         }
 404
 405         /*
 406          * push the map to the device-mapper
 407          */
 408         if (setup_map(mpp)) {
 409                 condlog(0, "%s: failed to setup map for addition of new "
 410                         "path %s", mpp->alias, devname);
 411                 goto out;
 412         }
 413         /*
 414          * reload the map for the multipath mapped device
 415          */
 416         if (domap(mpp) <= 0) {
 417                 condlog(0, "%s: failed in domap for addition of new "
 418                         "path %s", mpp->alias, devname);
 419                 /*
 420                  * deal with asynchronous uevents :((
 421                  */
 422                 if (mpp->action == ACT_RELOAD) {
 423                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 424                         sleep(1);
 425                         update_mpp_paths(mpp, vecs->pathvec);
 426                         goto rescan;
 427                 }
 428                 else
 429                         goto out;
 430         }
 431         dm_lib_release();
 432
 433         /*
 434          * update our state from kernel regardless of create or reload
 435          */
 436         if (setup_multipath(vecs, mpp))
 437                 goto out;
 438
 439         sync_map_state(mpp);
 440
 441         if (mpp->action == ACT_CREATE &&
 442             start_waiter_thread(mpp, vecs))
 443                         goto out;
 444
 445         condlog(2, "%s path added to devmap %s", devname, mpp->alias);
 446         return 0;
 447
 448 out:
 449         remove_map(mpp, vecs, 1);
 450         return 1;
 451 }
 452
 453 static int
 454 uev_remove_path (struct sysfs_device * dev, struct vectors * vecs)
 455 {
 456         int retval;
 457
 458         condlog(2, "%s: remove path (uevent)", dev->kernel);
 459         retval = ev_remove_path(dev->kernel, vecs);
 460         if (!retval)
 461                 sysfs_device_put(dev);
 462
 463         return retval;
 464 }
 465
 466 int
 467 ev_remove_path (char * devname, struct vectors * vecs)
 468 {
 469         struct multipath * mpp;
 470         struct path * pp;
 471         int i, retval = 0;
 472
 473         pp = find_path_by_dev(vecs->pathvec, devname);
 474
 475         if (!pp) {
 476                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 477                 return 1;
 478         }
 479
 480         /*
 481          * avoid referring to the map of an orphaned path
 482          */
 483         if ((mpp = pp->mpp)) {
 484                 /*
 485                  * transform the mp->pg vector of vectors of paths
 486                  * into a mp->params string to feed the device-mapper
 487                  */
 488                 if (update_mpp_paths(mpp, vecs->pathvec)) {
 489                         condlog(0, "%s: failed to update paths",
 490                                 mpp->alias);
 491                         goto out;
 492                 }
 493                 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 494                         vector_del_slot(mpp->paths, i);
 495
 496                 /*
 497                  * remove the map IFF removing the last path
 498                  */
 499                 if (VECTOR_SIZE(mpp->paths) == 0) {
 500                         char alias[WWID_SIZE];
 501
 502                         /*
 503                          * flush_map will fail if the device is open
 504                          */
 505                         strncpy(alias, mpp->alias, WWID_SIZE);
 506                         if (!flush_map(mpp, vecs)) {
 507                                 condlog(2, "%s: removed map after"
 508                                         " removing all paths",
 509                                         alias);
 510                                 free_path(pp);
 511                                 return 0;
 512                         }
 513                         /*
 514                          * Not an error, continue
 515                          */
 516                 }
 517
 518                 if (setup_map(mpp)) {
 519                         condlog(0, "%s: failed to setup map for"
 520                                 " removal of path %s", mpp->alias,
 521                                 devname);
 522                         goto out;
 523                 }
 524                 /*
 525                  * reload the map
 526                  */
 527                 mpp->action = ACT_RELOAD;
 528                 if (domap(mpp) <= 0) {
 529                         condlog(0, "%s: failed in domap for "
 530                                 "removal of path %s",
 531                                 mpp->alias, devname);
 532                         retval = 1;
 533                 } else {
 534                         /*
 535                          * update our state from kernel
 536                          */
 537                         if (setup_multipath(vecs, mpp)) {
 538                                 goto out;
 539                         }
 540                         sync_map_state(mpp);
 541
 542                         condlog(2, "%s: path removed from map %s",
 543                                 devname, mpp->alias);
 544                 }
 545         }
 546
 547         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 548                 vector_del_slot(vecs->pathvec, i);
 549
 550         free_path(pp);
 551
 552         return retval;
 553
 554 out:
 555         remove_map_and_stop_waiter(mpp, vecs, 1);
 556         return 1;
 557 }
 558
 559 static int
 560 map_discovery (struct vectors * vecs)
 561 {
 562         struct multipath * mpp;
 563         unsigned int i;
 564
 565         if (dm_get_maps(vecs->mpvec))
 566                 return 1;
 567
 568         vector_foreach_slot (vecs->mpvec, mpp, i)
 569                 if (setup_multipath(vecs, mpp))
 570                         return 1;
 571
 572         return 0;
 573 }
 574
 575 int
 576 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 577 {
 578         struct vectors * vecs;
 579         int r;
 580
 581         *reply = NULL;
 582         *len = 0;
 583         vecs = (struct vectors *)trigger_data;
 584
 585         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 586         lock(vecs->lock);
 587
 588         r = parse_cmd(str, reply, len, vecs);
 589
 590         if (r > 0) {
 591                 *reply = STRDUP("fail\n");
 592                 *len = strlen(*reply) + 1;
 593                 r = 1;
 594         }
 595         else if (!r && *len == 0) {
 596                 *reply = STRDUP("ok\n");
 597                 *len = strlen(*reply) + 1;
 598                 r = 0;
 599         }
 600         /* else if (r < 0) leave *reply alone */
 601
 602         lock_cleanup_pop(vecs->lock);
 603         return r;
 604 }
 605
 606 static int
 607 uev_discard(char * devpath)
 608 {
 609         char *tmp;
 610         char a[11], b[11];
 611
 612         /*
 613          * keep only block devices, discard partitions
 614          */
 615         tmp = strstr(devpath, "/block/");
 616         if (tmp == NULL){
 617                 condlog(4, "no /block/ in '%s'", devpath);
 618                 return 1;
 619         }
 620         if (sscanf(tmp, "/block/%10s", a) != 1 ||
 621             sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
 622                 condlog(4, "discard event on %s", devpath);
 623                 return 1;
 624         }
 625         return 0;
 626 }
 627
 628 int
 629 uev_trigger (struct uevent * uev, void * trigger_data)
 630 {
 631         int r = 0;
 632         struct sysfs_device *sysdev;
 633         struct vectors * vecs;
 634
 635         vecs = (struct vectors *)trigger_data;
 636
 637         if (uev_discard(uev->devpath))
 638                 return 0;
 639
 640         sysdev = sysfs_device_get(uev->devpath);
 641         if(!sysdev)
 642                 return 0;
 643
 644         lock(vecs->lock);
 645
 646         /*
 647          * device map event
 648          * Add events are ignored here as the tables
 649          * are not fully initialised then.
 650          */
 651         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 652                 if (!strncmp(uev->action, "change", 6)) {
 653                         r = uev_add_map(sysdev, vecs);
 654                         goto out;
 655                 }
 656                 if (!strncmp(uev->action, "remove", 6)) {
 657                         r = uev_remove_map(sysdev, vecs);
 658                         goto out;
 659                 }
 660                 if (!strncmp(uev->action, "umount", 6)) {
 661                         r = uev_umount_map(sysdev, vecs);
 662                         goto out;
 663                 }
 664                 goto out;
 665         }
 666
 667         /*
 668          * path add/remove event
 669          */
 670         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 671                            sysdev->kernel) > 0)
 672                 goto out;
 673
 674         if (!strncmp(uev->action, "add", 3)) {
 675                 r = uev_add_path(sysdev, vecs);
 676                 goto out;
 677         }
 678         if (!strncmp(uev->action, "remove", 6)) {
 679                 r = uev_remove_path(sysdev, vecs);
 680                 goto out;
 681         }
 682
 683 out:
 684         unlock(vecs->lock);
 685         return r;
 686 }
 687
 688 static void *
 689 ueventloop (void * ap)
 690 {
 691         block_signal(SIGUSR1, NULL);
 692         block_signal(SIGHUP, NULL);
 693
 694         if (uevent_listen(&uev_trigger, ap))
 695                 fprintf(stderr, "error starting uevent listener");
 696
 697         return NULL;
 698 }
 699
 700 static void *
 701 uxlsnrloop (void * ap)
 702 {
 703         block_signal(SIGUSR1, NULL);
 704         block_signal(SIGHUP, NULL);
 705
 706         if (cli_init())
 707                 return NULL;
 708
 709         set_handler_callback(LIST+PATHS, cli_list_paths);
 710         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 711         set_handler_callback(LIST+MAPS, cli_list_maps);
 712         set_handler_callback(LIST+STATUS, cli_list_status);
 713         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 714         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 715         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 716         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 717         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 718         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 719         set_handler_callback(LIST+CONFIG, cli_list_config);
 720         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 721         set_handler_callback(LIST+DEVICES, cli_list_devices);
 722         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 723         set_handler_callback(ADD+PATH, cli_add_path);
 724         set_handler_callback(DEL+PATH, cli_del_path);
 725         set_handler_callback(ADD+MAP, cli_add_map);
 726         set_handler_callback(DEL+MAP, cli_del_map);
 727         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 728         set_handler_callback(RECONFIGURE, cli_reconfigure);
 729         set_handler_callback(SUSPEND+MAP, cli_suspend);
 730         set_handler_callback(RESUME+MAP, cli_resume);
 731         set_handler_callback(RESIZE+MAP, cli_resize);
 732         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 733         set_handler_callback(FAIL+PATH, cli_fail);
 734         set_handler_callback(QUIT, cli_quit);
 735
 736         uxsock_listen(&uxsock_trigger, ap);
 737
 738         return NULL;
 739 }
 740
 741 static int
 742 exit_daemon (int status)
 743 {
 744         if (status != 0)
 745                 fprintf(stderr, "bad exit status. see daemon.log\n");
 746
 747         condlog(3, "unlink pidfile");
 748         unlink(DEFAULT_PIDFILE);
 749
 750         pthread_mutex_lock(&exit_mutex);
 751         pthread_cond_signal(&exit_cond);
 752         pthread_mutex_unlock(&exit_mutex);
 753
 754         return status;
 755 }
 756
 757 static void
 758 fail_path (struct path * pp, int del_active)
 759 {
 760         if (!pp->mpp)
 761                 return;
 762
 763         condlog(2, "checker failed path %s in map %s",
 764                  pp->dev_t, pp->mpp->alias);
 765
 766         dm_fail_path(pp->mpp->alias, pp->dev_t);
 767         if (del_active)
 768                 update_queue_mode_del_path(pp->mpp);
 769 }
 770
 771 /*
 772  * caller must have locked the path list before calling that function
 773  */
 774 static void
 775 reinstate_path (struct path * pp, int add_active)
 776 {
 777         if (!pp->mpp)
 778                 return;
 779
 780         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 781                 condlog(0, "%s: reinstate failed", pp->dev_t);
 782         else {
 783                 condlog(2, "%s: reinstated", pp->dev_t);
 784                 if (add_active)
 785                         update_queue_mode_add_path(pp->mpp);
 786         }
 787 }
 788
 789 static void
 790 enable_group(struct path * pp)
 791 {
 792         struct pathgroup * pgp;
 793
 794         /*
 795          * if path is added through uev_add_path, pgindex can be unset.
 796          * next update_strings() will set it, upon map reload event.
 797          *
 798          * we can safely return here, because upon map reload, all
 799          * PG will be enabled.
 800          */
 801         if (!pp->mpp->pg || !pp->pgindex)
 802                 return;
 803
 804         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 805
 806         if (pgp->status == PGSTATE_DISABLED) {
 807                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 808                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 809         }
 810 }
 811
 812 static void
 813 mpvec_garbage_collector (struct vectors * vecs)
 814 {
 815         struct multipath * mpp;
 816         unsigned int i;
 817
 818         vector_foreach_slot (vecs->mpvec, mpp, i) {
 819                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 820                         condlog(2, "%s: remove dead map", mpp->alias);
 821                         remove_map_and_stop_waiter(mpp, vecs, 1);
 822                         i--;
 823                 }
 824         }
 825 }
 826
 827 static void
 828 defered_failback_tick (vector mpvec)
 829 {
 830         struct multipath * mpp;
 831         unsigned int i;
 832
 833         vector_foreach_slot (mpvec, mpp, i) {
 834                 /*
 835                  * defered failback getting sooner
 836                  */
 837                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 838                         mpp->failback_tick--;
 839
 840                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 841                                 switch_pathgroup(mpp);
 842                 }
 843         }
 844 }
 845
 846 static void
 847 retry_count_tick(vector mpvec)
 848 {
 849         struct multipath *mpp;
 850         unsigned int i;
 851
 852         vector_foreach_slot (mpvec, mpp, i) {
 853                 if (mpp->retry_tick) {
 854                         mpp->stat_total_queueing_time++;
 855                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 856                         if(--mpp->retry_tick == 0) {
 857                                 dm_queue_if_no_path(mpp->alias, 0);
 858                                 condlog(2, "%s: Disable queueing", mpp->alias);
 859                         }
 860                 }
 861         }
 862 }
 863
 864 void
 865 check_path (struct vectors * vecs, struct path * pp)
 866 {
 867         int newstate;
 868
 869         if (!pp->mpp)
 870                 return;
 871
 872         if (pp->tick && --pp->tick)
 873                 return; /* don't check this path yet */
 874
 875         /*
 876          * provision a next check soonest,
 877          * in case we exit abnormaly from here
 878          */
 879         pp->tick = conf->checkint;
 880
 881         if (!checker_selected(&pp->checker)) {
 882                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 883                 select_checker(pp);
 884         }
 885         if (!checker_selected(&pp->checker)) {
 886                 condlog(0, "%s: checker is not set", pp->dev);
 887                 return;
 888         }
 889         /*
 890          * Set checker in async mode.
 891          * Honored only by checker implementing the said mode.
 892          */
 893         checker_set_async(&pp->checker);
 894
 895         if (path_offline(pp))
 896                 newstate = PATH_DOWN;
 897         else
 898                 newstate = checker_check(&pp->checker);
 899
 900         if (newstate < 0) {
 901                 condlog(2, "%s: unusable path", pp->dev);
 902                 pathinfo(pp, conf->hwtable, 0);
 903                 return;
 904         }
 905         /*
 906          * Async IO in flight. Keep the previous path state
 907          * and reschedule as soon as possible
 908          */
 909         if (newstate == PATH_PENDING) {
 910                 pp->tick = 1;
 911                 return;
 912         }
 913         if (newstate != pp->state) {
 914                 int oldstate = pp->state;
 915                 pp->state = newstate;
 916                 LOG_MSG(1, checker_message(&pp->checker));
 917
 918                 /*
 919                  * upon state change, reset the checkint
 920                  * to the shortest delay
 921                  */
 922                 pp->checkint = conf->checkint;
 923
 924                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY ||
 925                     update_multipath_strings(pp->mpp, vecs->pathvec)) {
 926                         /*
 927                          * proactively fail path in the DM
 928                          */
 929                         if (oldstate == PATH_UP ||
 930                             oldstate == PATH_GHOST)
 931                                 fail_path(pp, 1);
 932                         else
 933                                 fail_path(pp, 0);
 934
 935                         /*
 936                          * cancel scheduled failback
 937                          */
 938                         pp->mpp->failback_tick = 0;
 939
 940                         pp->mpp->stat_path_failures++;
 941                         return;
 942                 }
 943
 944                 /*
 945                  * reinstate this path
 946                  */
 947                 if (oldstate != PATH_UP &&
 948                     oldstate != PATH_GHOST)
 949                         reinstate_path(pp, 1);
 950                 else
 951                         reinstate_path(pp, 0);
 952
 953                 /*
 954                  * schedule [defered] failback
 955                  */
 956                 if (pp->mpp->pgfailback > 0)
 957                         pp->mpp->failback_tick =
 958                                 pp->mpp->pgfailback + 1;
 959                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 960                     need_switch_pathgroup(pp->mpp, 1))
 961                         switch_pathgroup(pp->mpp);
 962
 963                 /*
 964                  * if at least one path is up in a group, and
 965                  * the group is disabled, re-enable it
 966                  */
 967                 if (newstate == PATH_UP)
 968                         enable_group(pp);
 969         }
 970         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 971                 LOG_MSG(4, checker_message(&pp->checker));
 972                 /*
 973                  * double the next check delay.
 974                  * max at conf->max_checkint
 975                  */
 976                 if (pp->checkint < (conf->max_checkint / 2))
 977                         pp->checkint = 2 * pp->checkint;
 978                 else
 979                         pp->checkint = conf->max_checkint;
 980
 981                 pp->tick = pp->checkint;
 982                 condlog(4, "%s: delay next check %is",
 983                                 pp->dev_t, pp->tick);
 984         }
 985         else if (newstate == PATH_DOWN)
 986                 LOG_MSG(2, checker_message(&pp->checker));
 987
 988         pp->state = newstate;
 989
 990         /*
 991          * path prio refreshing
 992          */
 993         condlog(4, "path prio refresh");
 994         pathinfo(pp, conf->hwtable, DI_PRIO);
 995
 996         /*
 997          * pathgroup failback policy
 998          */
 999         if (need_switch_pathgroup(pp->mpp, 0)) {
1000                 if (pp->mpp->pgfailback > 0 &&
1001                     pp->mpp->failback_tick <= 0)
1002                         pp->mpp->failback_tick =
1003                                 pp->mpp->pgfailback + 1;
1004                 else if (pp->mpp->pgfailback ==
1005                                 -FAILBACK_IMMEDIATE)
1006                         switch_pathgroup(pp->mpp);
1007         }
1008 }
1009
1010 static void *
1011 checkerloop (void *ap)
1012 {
1013         struct vectors *vecs;
1014         struct path *pp;
1015         int count = 0;
1016         unsigned int i;
1017         sigset_t old;
1018
1019         mlockall(MCL_CURRENT | MCL_FUTURE);
1020         vecs = (struct vectors *)ap;
1021         condlog(2, "path checkers start up");
1022
1023         /*
1024          * init the path check interval
1025          */
1026         vector_foreach_slot (vecs->pathvec, pp, i) {
1027                 pp->checkint = conf->checkint;
1028         }
1029
1030         while (1) {
1031                 block_signal(SIGHUP, &old);
1032                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1033                 lock(vecs->lock);
1034                 condlog(4, "tick");
1035
1036                 if (vecs->pathvec) {
1037                         vector_foreach_slot (vecs->pathvec, pp, i) {
1038                                 check_path(vecs, pp);
1039                         }
1040                 }
1041                 if (vecs->mpvec) {
1042                         defered_failback_tick(vecs->mpvec);
1043                         retry_count_tick(vecs->mpvec);
1044                 }
1045                 if (count)
1046                         count--;
1047                 else {
1048                         condlog(4, "map garbage collection");
1049                         mpvec_garbage_collector(vecs);
1050                         count = MAPGCINT;
1051                 }
1052
1053                 lock_cleanup_pop(vecs->lock);
1054                 pthread_sigmask(SIG_SETMASK, &old, NULL);
1055                 sleep(1);
1056         }
1057         return NULL;
1058 }
1059
1060 int
1061 configure (struct vectors * vecs, int start_waiters)
1062 {
1063         struct multipath * mpp;
1064         struct path * pp;
1065         vector mpvec;
1066         int i;
1067
1068         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1069                 return 1;
1070
1071         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1072                 return 1;
1073
1074         if (!(mpvec = vector_alloc()))
1075                 return 1;
1076
1077         /*
1078          * probe for current path (from sysfs) and map (from dm) sets
1079          */
1080         path_discovery(vecs->pathvec, conf, DI_ALL);
1081
1082         vector_foreach_slot (vecs->pathvec, pp, i){
1083                 if (filter_path(conf, pp) > 0){
1084                         vector_del_slot(vecs->pathvec, i);
1085                         free_path(pp);
1086                         i--;
1087                 }
1088                 else
1089                         pp->checkint = conf->checkint;
1090         }
1091         if (map_discovery(vecs))
1092                 return 1;
1093
1094         /*
1095          * create new set of maps & push changed ones into dm
1096          */
1097         if (coalesce_paths(vecs, mpvec, NULL, 0))
1098                 return 1;
1099
1100         /*
1101          * may need to remove some maps which are no longer relevant
1102          * e.g., due to blacklist changes in conf file
1103          */
1104         if (coalesce_maps(vecs, mpvec))
1105                 return 1;
1106
1107         dm_lib_release();
1108
1109         sync_maps_state(mpvec);
1110
1111         /*
1112          * purge dm of old maps
1113          */
1114         remove_maps(vecs);
1115
1116         /*
1117          * save new set of maps formed by considering current path state
1118          */
1119         vector_free(vecs->mpvec);
1120         vecs->mpvec = mpvec;
1121
1122         /*
1123          * start dm event waiter threads for these new maps
1124          */
1125         vector_foreach_slot(vecs->mpvec, mpp, i) {
1126                 if (setup_multipath(vecs, mpp))
1127                         return 1;
1128                 if (start_waiters)
1129                         if (start_waiter_thread(mpp, vecs))
1130                                 return 1;
1131         }
1132         return 0;
1133 }
1134
1135 int
1136 reconfigure (struct vectors * vecs)
1137 {
1138         struct config * old = conf;
1139
1140         /*
1141          * free old map and path vectors ... they use old conf state
1142          */
1143         if (VECTOR_SIZE(vecs->mpvec))
1144                 remove_maps_and_stop_waiters(vecs);
1145
1146         if (VECTOR_SIZE(vecs->pathvec))
1147                 free_pathvec(vecs->pathvec, FREE_PATHS);
1148
1149         vecs->pathvec = NULL;
1150         conf = NULL;
1151
1152         if (load_config(DEFAULT_CONFIGFILE))
1153                 return 1;
1154
1155         conf->verbosity = old->verbosity;
1156
1157         if (!conf->checkint) {
1158                 conf->checkint = DEFAULT_CHECKINT;
1159                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1160         }
1161         configure(vecs, 1);
1162         free_config(old);
1163         return 0;
1164 }
1165
1166 static struct vectors *
1167 init_vecs (void)
1168 {
1169         struct vectors * vecs;
1170
1171         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1172
1173         if (!vecs)
1174                 return NULL;
1175
1176         vecs->lock.mutex =
1177                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1178
1179         if (!vecs->lock.mutex)
1180                 goto out;
1181
1182         pthread_mutex_init(vecs->lock.mutex, NULL);
1183         vecs->lock.depth = 0;
1184
1185         return vecs;
1186
1187 out:
1188         FREE(vecs);
1189         condlog(0, "failed to init paths");
1190         return NULL;
1191 }
1192
1193 static void *
1194 signal_set(int signo, void (*func) (int))
1195 {
1196         int r;
1197         struct sigaction sig;
1198         struct sigaction osig;
1199
1200         sig.sa_handler = func;
1201         sigemptyset(&sig.sa_mask);
1202         sig.sa_flags = 0;
1203
1204         r = sigaction(signo, &sig, &osig);
1205
1206         if (r < 0)
1207                 return (SIG_ERR);
1208         else
1209                 return (osig.sa_handler);
1210 }
1211
1212 static void
1213 sighup (int sig)
1214 {
1215         condlog(2, "reconfigure (SIGHUP)");
1216
1217         lock(gvecs->lock);
1218         reconfigure(gvecs);
1219         unlock(gvecs->lock);
1220
1221 #ifdef _DEBUG_
1222         dbg_free_final(NULL);
1223 #endif
1224 }
1225
1226 static void
1227 sigend (int sig)
1228 {
1229         exit_daemon(0);
1230 }
1231
1232 static void
1233 sigusr1 (int sig)
1234 {
1235         condlog(3, "SIGUSR1 received");
1236 }
1237
1238 static void
1239 signal_init(void)
1240 {
1241         signal_set(SIGHUP, sighup);
1242         signal_set(SIGUSR1, sigusr1);
1243         signal_set(SIGINT, sigend);
1244         signal_set(SIGTERM, sigend);
1245         signal(SIGPIPE, SIG_IGN);
1246 }
1247
1248 static void
1249 setscheduler (void)
1250 {
1251         int res;
1252         static struct sched_param sched_param = {
1253                 .sched_priority = 99
1254         };
1255
1256         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1257
1258         if (res == -1)
1259                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1260         return;
1261 }
1262
1263 static void
1264 set_oom_adj (int val)
1265 {
1266         FILE *fp;
1267
1268         fp = fopen("/proc/self/oom_adj", "w");
1269
1270         if (!fp)
1271                 return;
1272
1273         fprintf(fp, "%i", val);
1274         fclose(fp);
1275 }
1276
1277 static int
1278 child (void * param)
1279 {
1280         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1281         pthread_attr_t attr;
1282         struct vectors * vecs;
1283
1284         mlockall(MCL_CURRENT | MCL_FUTURE);
1285
1286         if (logsink)
1287                 log_thread_start();
1288
1289         condlog(2, "--------start up--------");
1290         condlog(2, "read " DEFAULT_CONFIGFILE);
1291
1292         if (load_config(DEFAULT_CONFIGFILE))
1293                 exit(1);
1294
1295         if (init_checkers()) {
1296                 condlog(0, "failed to initialize checkers");
1297                 exit(1);
1298         }
1299         if (init_prio()) {
1300                 condlog(0, "failed to initialize prioritizers");
1301                 exit(1);
1302         }
1303
1304         setlogmask(LOG_UPTO(conf->verbosity + 3));
1305
1306         /*
1307          * fill the voids left in the config file
1308          */
1309         if (!conf->checkint) {
1310                 conf->checkint = DEFAULT_CHECKINT;
1311                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1312         }
1313
1314         if (conf->max_fds) {
1315                 struct rlimit fd_limit;
1316                 if (conf->max_fds > 0) {
1317                         fd_limit.rlim_cur = conf->max_fds;
1318                         fd_limit.rlim_max = conf->max_fds;
1319                 }
1320                 else {
1321                         fd_limit.rlim_cur = RLIM_INFINITY;
1322                         fd_limit.rlim_max = RLIM_INFINITY;
1323                 }
1324                 if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0)
1325                         condlog(0, "can't set open fds limit to %d : %s\n",
1326                                 conf->max_fds, strerror(errno));
1327         }
1328
1329         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1330                 if (logsink)
1331                         log_thread_stop();
1332
1333                 exit(1);
1334         }
1335         signal_init();
1336         setscheduler();
1337         set_oom_adj(-16);
1338         vecs = gvecs = init_vecs();
1339
1340         if (!vecs)
1341                 exit(1);
1342
1343         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1344                 condlog(0, "can not find sysfs mount point");
1345                 exit(1);
1346         }
1347         conf->daemon = 1;
1348         /*
1349          * fetch and configure both paths and multipaths
1350          */
1351         if (configure(vecs, 1)) {
1352                 condlog(0, "failure during configuration");
1353                 exit(1);
1354         }
1355         /*
1356          * start threads
1357          */
1358         pthread_attr_init(&attr);
1359         pthread_attr_setstacksize(&attr, 64 * 1024);
1360         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1361
1362         pthread_create(&check_thr, &attr, checkerloop, vecs);
1363         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1364         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1365
1366         pthread_cond_wait(&exit_cond, &exit_mutex);
1367
1368         /*
1369          * exit path
1370          */
1371         block_signal(SIGHUP, NULL);
1372         lock(vecs->lock);
1373         remove_maps_and_stop_waiters(vecs);
1374         free_pathvec(vecs->pathvec, FREE_PATHS);
1375
1376         pthread_cancel(check_thr);
1377         pthread_cancel(uevent_thr);
1378         pthread_cancel(uxlsnr_thr);
1379
1380         sysfs_cleanup();
1381
1382         free_keys(keys);
1383         keys = NULL;
1384         free_handlers(handlers);
1385         handlers = NULL;
1386         free_polls();
1387
1388         unlock(vecs->lock);
1389         /* Now all the waitevent threads will start rushing in. */
1390         while (vecs->lock.depth > 0) {
1391                 sleep (1); /* This is weak. */
1392                 condlog(3,"Have %d wait event checkers threads to de-alloc, waiting..\n", vecs->lock.depth);
1393         }
1394         pthread_mutex_destroy(vecs->lock.mutex);
1395         FREE(vecs->lock.mutex);
1396         vecs->lock.depth = 0;
1397         vecs->lock.mutex = NULL;
1398         FREE(vecs);
1399         vecs = NULL;
1400
1401         condlog(2, "--------shut down-------");
1402
1403         if (logsink)
1404                 log_thread_stop();
1405
1406         dm_lib_release();
1407         dm_lib_exit();
1408
1409         cleanup_prio();
1410         cleanup_checkers();
1411         /*
1412          * Freeing config must be done after condlog() and dm_lib_exit(),
1413          * because logging functions like dlog() and dm_write_log()
1414          * reference the config.
1415          */
1416         free_config(conf);
1417         conf = NULL;
1418
1419 #ifdef _DEBUG_
1420         dbg_free_final(NULL);
1421 #endif
1422
1423         exit(0);
1424 }
1425
1426 static int
1427 daemonize(void)
1428 {
1429         int pid;
1430         int in_fd, out_fd;
1431
1432         if( (pid = fork()) < 0){
1433                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1434                 return -1;
1435         }
1436         else if (pid != 0)
1437                 return pid;
1438
1439         setsid();
1440
1441         if ( (pid = fork()) < 0)
1442                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1443         else if (pid != 0)
1444                 _exit(0);
1445
1446         in_fd = open("/dev/null", O_RDONLY);
1447         if (in_fd < 0){
1448                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1449                         strerror(errno));
1450                 _exit(0);
1451         }
1452         out_fd = open("/dev/console", O_WRONLY);
1453         if (out_fd < 0){
1454                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1455                         strerror(errno));
1456                 _exit(0);
1457         }
1458
1459         close(STDIN_FILENO);
1460         dup(in_fd);
1461         close(STDOUT_FILENO);
1462         dup(out_fd);
1463         close(STDERR_FILENO);
1464         dup(out_fd);
1465
1466         close(in_fd);
1467         close(out_fd);
1468         if (chdir("/") < 0)
1469                 fprintf(stderr, "cannot chdir to '/', continuing\n");
1470
1471         return 0;
1472 }
1473
1474 int
1475 main (int argc, char *argv[])
1476 {
1477         extern char *optarg;
1478         extern int optind;
1479         int arg;
1480         int err;
1481
1482         logsink = 1;
1483         dm_init();
1484
1485         if (getuid() != 0) {
1486                 fprintf(stderr, "need to be root\n");
1487                 exit(1);
1488         }
1489
1490         /* make sure we don't lock any path */
1491         chdir("/");
1492         umask(umask(077) | 022);
1493
1494         conf = alloc_config();
1495
1496         if (!conf)
1497                 exit(1);
1498
1499         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1500         switch(arg) {
1501                 case 'd':
1502                         logsink = 0;
1503                         //debug=1; /* ### comment me out ### */
1504                         break;
1505                 case 'v':
1506                         if (sizeof(optarg) > sizeof(char *) ||
1507                             !isdigit(optarg[0]))
1508                                 exit(1);
1509
1510                         conf->verbosity = atoi(optarg);
1511                         break;
1512                 case 'k':
1513                         uxclnt(optarg);
1514                         exit(0);
1515                 default:
1516                         ;
1517                 }
1518         }
1519         if (optind < argc) {
1520                 char cmd[CMDSIZE];
1521                 char * s = cmd;
1522                 char * c = s;
1523
1524                 while (optind < argc) {
1525                         if (strchr(argv[optind], ' '))
1526                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1527                         else
1528                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1529                         optind++;
1530                 }
1531                 c += snprintf(c, s + CMDSIZE - c, "\n");
1532                 uxclnt(s);
1533                 exit(0);
1534         }
1535
1536         if (!logsink)
1537                 err = 0;
1538         else
1539                 err = daemonize();
1540
1541         if (err < 0)
1542                 /* error */
1543                 exit(1);
1544         else if (err > 0)
1545                 /* parent dies */
1546                 exit(0);
1547         else
1548                 /* child lives */
1549                 return (child(NULL));
1550 }
1551