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