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
 718         uxsock_listen(&uxsock_trigger, ap);
 719
 720         return NULL;
 721 }
 722
 723 static int
 724 exit_daemon (int status)
 725 {
 726         if (status != 0)
 727                 fprintf(stderr, "bad exit status. see daemon.log\n");
 728
 729         condlog(3, "unlink pidfile");
 730         unlink(DEFAULT_PIDFILE);
 731
 732         lock(&exit_mutex);
 733         pthread_cond_signal(&exit_cond);
 734         unlock(&exit_mutex);
 735
 736         return status;
 737 }
 738
 739 static void
 740 fail_path (struct path * pp, int del_active)
 741 {
 742         if (!pp->mpp)
 743                 return;
 744
 745         condlog(2, "checker failed path %s in map %s",
 746                  pp->dev_t, pp->mpp->alias);
 747
 748         dm_fail_path(pp->mpp->alias, pp->dev_t);
 749         if (del_active)
 750                 update_queue_mode_del_path(pp->mpp);
 751 }
 752
 753 /*
 754  * caller must have locked the path list before calling that function
 755  */
 756 static void
 757 reinstate_path (struct path * pp, int add_active)
 758 {
 759         if (!pp->mpp)
 760                 return;
 761
 762         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 763                 condlog(0, "%s: reinstate failed", pp->dev_t);
 764         else {
 765                 condlog(2, "%s: reinstated", pp->dev_t);
 766                 if (add_active)
 767                         update_queue_mode_add_path(pp->mpp);
 768         }
 769 }
 770
 771 static void
 772 enable_group(struct path * pp)
 773 {
 774         struct pathgroup * pgp;
 775
 776         /*
 777          * if path is added through uev_add_path, pgindex can be unset.
 778          * next update_strings() will set it, upon map reload event.
 779          *
 780          * we can safely return here, because upon map reload, all
 781          * PG will be enabled.
 782          */
 783         if (!pp->mpp->pg || !pp->pgindex)
 784                 return;
 785
 786         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 787
 788         if (pgp->status == PGSTATE_DISABLED) {
 789                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 790                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 791         }
 792 }
 793
 794 static void
 795 mpvec_garbage_collector (struct vectors * vecs)
 796 {
 797         struct multipath * mpp;
 798         unsigned int i;
 799
 800         vector_foreach_slot (vecs->mpvec, mpp, i) {
 801                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 802                         condlog(2, "%s: remove dead map", mpp->alias);
 803                         remove_map_and_stop_waiter(mpp, vecs, 1);
 804                         i--;
 805                 }
 806         }
 807 }
 808
 809 static void
 810 defered_failback_tick (vector mpvec)
 811 {
 812         struct multipath * mpp;
 813         unsigned int i;
 814
 815         vector_foreach_slot (mpvec, mpp, i) {
 816                 /*
 817                  * defered failback getting sooner
 818                  */
 819                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 820                         mpp->failback_tick--;
 821
 822                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 823                                 switch_pathgroup(mpp);
 824                 }
 825         }
 826 }
 827
 828 static void
 829 retry_count_tick(vector mpvec)
 830 {
 831         struct multipath *mpp;
 832         unsigned int i;
 833
 834         vector_foreach_slot (mpvec, mpp, i) {
 835                 if (mpp->retry_tick) {
 836                         mpp->stat_total_queueing_time++;
 837                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 838                         if(--mpp->retry_tick == 0) {
 839                                 dm_queue_if_no_path(mpp->alias, 0);
 840                                 condlog(2, "%s: Disable queueing", mpp->alias);
 841                         }
 842                 }
 843         }
 844 }
 845
 846 void
 847 check_path (struct vectors * vecs, struct path * pp)
 848 {
 849         int newstate;
 850
 851         if (!pp->mpp)
 852                 return;
 853
 854         if (pp->tick && --pp->tick)
 855                 return; /* don't check this path yet */
 856
 857         /*
 858          * provision a next check soonest,
 859          * in case we exit abnormaly from here
 860          */
 861         pp->tick = conf->checkint;
 862
 863         if (!checker_selected(&pp->checker)) {
 864                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 865                 select_checker(pp);
 866         }
 867         if (!checker_selected(&pp->checker)) {
 868                 condlog(0, "%s: checker is not set", pp->dev);
 869                 return;
 870         }
 871         /*
 872          * Set checker in async mode.
 873          * Honored only by checker implementing the said mode.
 874          */
 875         checker_set_async(&pp->checker);
 876
 877         newstate = checker_check(&pp->checker);
 878
 879         if (newstate < 0) {
 880                 condlog(2, "%s: unusable path", pp->dev);
 881                 pathinfo(pp, conf->hwtable, 0);
 882                 return;
 883         }
 884         /*
 885          * Async IO in flight. Keep the previous path state
 886          * and reschedule as soon as possible
 887          */
 888         if (newstate == PATH_PENDING) {
 889                 pp->tick = 1;
 890                 return;
 891         }
 892         if (newstate != pp->state) {
 893                 int oldstate = pp->state;
 894                 pp->state = newstate;
 895                 LOG_MSG(1, checker_message(&pp->checker));
 896
 897                 /*
 898                  * upon state change, reset the checkint
 899                  * to the shortest delay
 900                  */
 901                 pp->checkint = conf->checkint;
 902
 903                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY ||
 904                     update_multipath_strings(pp->mpp, vecs->pathvec)) {
 905                         /*
 906                          * proactively fail path in the DM
 907                          */
 908                         if (oldstate == PATH_UP ||
 909                             oldstate == PATH_GHOST)
 910                                 fail_path(pp, 1);
 911                         else
 912                                 fail_path(pp, 0);
 913
 914                         /*
 915                          * cancel scheduled failback
 916                          */
 917                         pp->mpp->failback_tick = 0;
 918
 919                         pp->mpp->stat_path_failures++;
 920                         return;
 921                 }
 922
 923                 /*
 924                  * reinstate this path
 925                  */
 926                 if (oldstate != PATH_UP &&
 927                     oldstate != PATH_GHOST)
 928                         reinstate_path(pp, 1);
 929                 else
 930                         reinstate_path(pp, 0);
 931
 932                 /*
 933                  * schedule [defered] failback
 934                  */
 935                 if (pp->mpp->pgfailback > 0)
 936                         pp->mpp->failback_tick =
 937                                 pp->mpp->pgfailback + 1;
 938                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 939                     need_switch_pathgroup(pp->mpp, 1))
 940                         switch_pathgroup(pp->mpp);
 941
 942                 /*
 943                  * if at least one path is up in a group, and
 944                  * the group is disabled, re-enable it
 945                  */
 946                 if (newstate == PATH_UP)
 947                         enable_group(pp);
 948         }
 949         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 950                 LOG_MSG(4, checker_message(&pp->checker));
 951                 /*
 952                  * double the next check delay.
 953                  * max at conf->max_checkint
 954                  */
 955                 if (pp->checkint < (conf->max_checkint / 2))
 956                         pp->checkint = 2 * pp->checkint;
 957                 else
 958                         pp->checkint = conf->max_checkint;
 959
 960                 pp->tick = pp->checkint;
 961                 condlog(4, "%s: delay next check %is",
 962                                 pp->dev_t, pp->tick);
 963         }
 964         else if (newstate == PATH_DOWN)
 965                 LOG_MSG(2, checker_message(&pp->checker));
 966
 967         pp->state = newstate;
 968
 969         /*
 970          * path prio refreshing
 971          */
 972         condlog(4, "path prio refresh");
 973         pathinfo(pp, conf->hwtable, DI_PRIO);
 974
 975         /*
 976          * pathgroup failback policy
 977          */
 978         if (need_switch_pathgroup(pp->mpp, 0)) {
 979                 if (pp->mpp->pgfailback > 0 &&
 980                     pp->mpp->failback_tick <= 0)
 981                         pp->mpp->failback_tick =
 982                                 pp->mpp->pgfailback + 1;
 983                 else if (pp->mpp->pgfailback ==
 984                                 -FAILBACK_IMMEDIATE)
 985                         switch_pathgroup(pp->mpp);
 986         }
 987 }
 988
 989 static void *
 990 checkerloop (void *ap)
 991 {
 992         struct vectors *vecs;
 993         struct path *pp;
 994         int count = 0;
 995         unsigned int i;
 996
 997         mlockall(MCL_CURRENT | MCL_FUTURE);
 998         vecs = (struct vectors *)ap;
 999         condlog(2, "path checkers start up");
1000
1001         /*
1002          * init the path check interval
1003          */
1004         vector_foreach_slot (vecs->pathvec, pp, i) {
1005                 pp->checkint = conf->checkint;
1006         }
1007
1008         while (1) {
1009                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1010                 lock(vecs->lock);
1011                 condlog(4, "tick");
1012
1013                 if (vecs->pathvec) {
1014                         vector_foreach_slot (vecs->pathvec, pp, i) {
1015                                 check_path(vecs, pp);
1016                         }
1017                 }
1018                 if (vecs->mpvec) {
1019                         defered_failback_tick(vecs->mpvec);
1020                         retry_count_tick(vecs->mpvec);
1021                 }
1022                 if (count)
1023                         count--;
1024                 else {
1025                         condlog(4, "map garbage collection");
1026                         mpvec_garbage_collector(vecs);
1027                         count = MAPGCINT;
1028                 }
1029
1030                 lock_cleanup_pop(vecs->lock);
1031                 sleep(1);
1032         }
1033         return NULL;
1034 }
1035
1036 int
1037 configure (struct vectors * vecs, int start_waiters)
1038 {
1039         struct multipath * mpp;
1040         struct path * pp;
1041         vector mpvec;
1042         int i;
1043
1044         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1045                 return 1;
1046
1047         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1048                 return 1;
1049
1050         if (!(mpvec = vector_alloc()))
1051                 return 1;
1052
1053         /*
1054          * probe for current path (from sysfs) and map (from dm) sets
1055          */
1056         path_discovery(vecs->pathvec, conf, DI_ALL);
1057
1058         vector_foreach_slot (vecs->pathvec, pp, i){
1059                 if (filter_path(conf, pp) > 0){
1060                         vector_del_slot(vecs->pathvec, i);
1061                         free_path(pp);
1062                         i--;
1063                 }
1064                 else
1065                         pp->checkint = conf->checkint;
1066         }
1067         if (map_discovery(vecs))
1068                 return 1;
1069
1070         /*
1071          * create new set of maps & push changed ones into dm
1072          */
1073         if (coalesce_paths(vecs, mpvec, NULL, 0))
1074                 return 1;
1075
1076         /*
1077          * may need to remove some maps which are no longer relevant
1078          * e.g., due to blacklist changes in conf file
1079          */
1080         if (coalesce_maps(vecs, mpvec))
1081                 return 1;
1082
1083         dm_lib_release();
1084
1085         sync_maps_state(mpvec);
1086
1087         /*
1088          * purge dm of old maps
1089          */
1090         remove_maps(vecs);
1091
1092         /*
1093          * save new set of maps formed by considering current path state
1094          */
1095         vector_free(vecs->mpvec);
1096         vecs->mpvec = mpvec;
1097
1098         /*
1099          * start dm event waiter threads for these new maps
1100          */
1101         vector_foreach_slot(vecs->mpvec, mpp, i) {
1102                 if (setup_multipath(vecs, mpp))
1103                         return 1;
1104                 if (start_waiters)
1105                         if (start_waiter_thread(mpp, vecs))
1106                                 return 1;
1107         }
1108         return 0;
1109 }
1110
1111 int
1112 reconfigure (struct vectors * vecs)
1113 {
1114         struct config * old = conf;
1115
1116         /*
1117          * free old map and path vectors ... they use old conf state
1118          */
1119         if (VECTOR_SIZE(vecs->mpvec))
1120                 remove_maps_and_stop_waiters(vecs);
1121
1122         if (VECTOR_SIZE(vecs->pathvec))
1123                 free_pathvec(vecs->pathvec, FREE_PATHS);
1124
1125         vecs->pathvec = NULL;
1126         conf = NULL;
1127
1128         if (load_config(DEFAULT_CONFIGFILE))
1129                 return 1;
1130
1131         conf->verbosity = old->verbosity;
1132
1133         if (!conf->checkint) {
1134                 conf->checkint = DEFAULT_CHECKINT;
1135                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1136         }
1137         configure(vecs, 1);
1138         free_config(old);
1139         return 0;
1140 }
1141
1142 static struct vectors *
1143 init_vecs (void)
1144 {
1145         struct vectors * vecs;
1146
1147         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1148
1149         if (!vecs)
1150                 return NULL;
1151
1152         vecs->lock =
1153                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1154
1155         if (!vecs->lock)
1156                 goto out;
1157
1158         pthread_mutex_init(vecs->lock, NULL);
1159
1160         return vecs;
1161
1162 out:
1163         FREE(vecs);
1164         condlog(0, "failed to init paths");
1165         return NULL;
1166 }
1167
1168 static void *
1169 signal_set(int signo, void (*func) (int))
1170 {
1171         int r;
1172         struct sigaction sig;
1173         struct sigaction osig;
1174
1175         sig.sa_handler = func;
1176         sigemptyset(&sig.sa_mask);
1177         sig.sa_flags = 0;
1178
1179         r = sigaction(signo, &sig, &osig);
1180
1181         if (r < 0)
1182                 return (SIG_ERR);
1183         else
1184                 return (osig.sa_handler);
1185 }
1186
1187 static void
1188 sighup (int sig)
1189 {
1190         condlog(2, "reconfigure (SIGHUP)");
1191
1192         lock(gvecs->lock);
1193         reconfigure(gvecs);
1194         unlock(gvecs->lock);
1195
1196 #ifdef _DEBUG_
1197         dbg_free_final(NULL);
1198 #endif
1199 }
1200
1201 static void
1202 sigend (int sig)
1203 {
1204         exit_daemon(0);
1205 }
1206
1207 static void
1208 sigusr1 (int sig)
1209 {
1210         condlog(3, "SIGUSR1 received");
1211 }
1212
1213 static void
1214 signal_init(void)
1215 {
1216         signal_set(SIGHUP, sighup);
1217         signal_set(SIGUSR1, sigusr1);
1218         signal_set(SIGINT, sigend);
1219         signal_set(SIGTERM, sigend);
1220         signal(SIGPIPE, SIG_IGN);
1221 }
1222
1223 static void
1224 setscheduler (void)
1225 {
1226         int res;
1227         static struct sched_param sched_param = {
1228                 .sched_priority = 99
1229         };
1230
1231         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1232
1233         if (res == -1)
1234                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1235         return;
1236 }
1237
1238 static void
1239 set_oom_adj (int val)
1240 {
1241         FILE *fp;
1242
1243         fp = fopen("/proc/self/oom_adj", "w");
1244
1245         if (!fp)
1246                 return;
1247
1248         fprintf(fp, "%i", val);
1249         fclose(fp);
1250 }
1251
1252 static int
1253 child (void * param)
1254 {
1255         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1256         pthread_attr_t attr;
1257         struct vectors * vecs;
1258
1259         mlockall(MCL_CURRENT | MCL_FUTURE);
1260
1261         if (logsink)
1262                 log_thread_start();
1263
1264         condlog(2, "--------start up--------");
1265         condlog(2, "read " DEFAULT_CONFIGFILE);
1266
1267         if (load_config(DEFAULT_CONFIGFILE))
1268                 exit(1);
1269
1270         if (init_checkers()) {
1271                 condlog(0, "failed to initialize checkers");
1272                 exit(1);
1273         }
1274         if (init_prio()) {
1275                 condlog(0, "failed to initialize prioritizers");
1276                 exit(1);
1277         }
1278
1279         setlogmask(LOG_UPTO(conf->verbosity + 3));
1280
1281         /*
1282          * fill the voids left in the config file
1283          */
1284         if (!conf->checkint) {
1285                 conf->checkint = DEFAULT_CHECKINT;
1286                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1287         }
1288
1289         if (conf->max_fds) {
1290                 struct rlimit fd_limit;
1291                 if (conf->max_fds > 0) {
1292                         fd_limit.rlim_cur = conf->max_fds;
1293                         fd_limit.rlim_max = conf->max_fds;
1294                 }
1295                 else {
1296                         fd_limit.rlim_cur = RLIM_INFINITY;
1297                         fd_limit.rlim_max = RLIM_INFINITY;
1298                 }
1299                 if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0)
1300                         condlog(0, "can't set open fds limit to %d : %s\n",
1301                                 conf->max_fds, strerror(errno));
1302         }
1303
1304         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1305                 if (logsink)
1306                         log_thread_stop();
1307
1308                 exit(1);
1309         }
1310         signal_init();
1311         setscheduler();
1312         set_oom_adj(-16);
1313         vecs = gvecs = init_vecs();
1314
1315         if (!vecs)
1316                 exit(1);
1317
1318         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1319                 condlog(0, "can not find sysfs mount point");
1320                 exit(1);
1321         }
1322
1323         /*
1324          * fetch and configure both paths and multipaths
1325          */
1326         if (configure(vecs, 1)) {
1327                 condlog(0, "failure during configuration");
1328                 exit(1);
1329         }
1330
1331         /*
1332          * start threads
1333          */
1334         pthread_attr_init(&attr);
1335         pthread_attr_setstacksize(&attr, 64 * 1024);
1336         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1337
1338         pthread_create(&check_thr, &attr, checkerloop, vecs);
1339         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1340         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1341
1342         pthread_cond_wait(&exit_cond, &exit_mutex);
1343
1344         /*
1345          * exit path
1346          */
1347         lock(vecs->lock);
1348         remove_maps_and_stop_waiters(vecs);
1349         free_pathvec(vecs->pathvec, FREE_PATHS);
1350
1351         pthread_cancel(check_thr);
1352         pthread_cancel(uevent_thr);
1353         pthread_cancel(uxlsnr_thr);
1354
1355         sysfs_cleanup();
1356
1357         free_keys(keys);
1358         keys = NULL;
1359         free_handlers(handlers);
1360         handlers = NULL;
1361         free_polls();
1362
1363         unlock(vecs->lock);
1364         pthread_mutex_destroy(vecs->lock);
1365         FREE(vecs->lock);
1366         vecs->lock = NULL;
1367         FREE(vecs);
1368         vecs = NULL;
1369
1370         condlog(2, "--------shut down-------");
1371
1372         if (logsink)
1373                 log_thread_stop();
1374
1375         dm_lib_release();
1376         dm_lib_exit();
1377
1378         /*
1379          * Freeing config must be done after condlog() and dm_lib_exit(),
1380          * because logging functions like dlog() and dm_write_log()
1381          * reference the config.
1382          */
1383         free_config(conf);
1384         conf = NULL;
1385
1386 #ifdef _DEBUG_
1387         dbg_free_final(NULL);
1388 #endif
1389
1390         exit(0);
1391 }
1392
1393 static int
1394 daemonize(void)
1395 {
1396         int pid;
1397         int in_fd, out_fd;
1398
1399         if( (pid = fork()) < 0){
1400                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1401                 return -1;
1402         }
1403         else if (pid != 0)
1404                 return pid;
1405
1406         setsid();
1407
1408         if ( (pid = fork()) < 0)
1409                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1410         else if (pid != 0)
1411                 _exit(0);
1412
1413         in_fd = open("/dev/null", O_RDONLY);
1414         if (in_fd < 0){
1415                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1416                         strerror(errno));
1417                 _exit(0);
1418         }
1419         out_fd = open("/dev/console", O_WRONLY);
1420         if (out_fd < 0){
1421                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1422                         strerror(errno));
1423                 _exit(0);
1424         }
1425
1426         close(STDIN_FILENO);
1427         dup(in_fd);
1428         close(STDOUT_FILENO);
1429         dup(out_fd);
1430         close(STDERR_FILENO);
1431         dup(out_fd);
1432
1433         close(in_fd);
1434         close(out_fd);
1435         chdir("/");
1436         umask(0);
1437         return 0;
1438 }
1439
1440 int
1441 main (int argc, char *argv[])
1442 {
1443         extern char *optarg;
1444         extern int optind;
1445         int arg;
1446         int err;
1447
1448         logsink = 1;
1449         dm_init();
1450
1451         if (getuid() != 0) {
1452                 fprintf(stderr, "need to be root\n");
1453                 exit(1);
1454         }
1455
1456         /* make sure we don't lock any path */
1457         chdir("/");
1458         umask(umask(077) | 022);
1459
1460         conf = alloc_config();
1461
1462         if (!conf)
1463                 exit(1);
1464
1465         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1466         switch(arg) {
1467                 case 'd':
1468                         logsink = 0;
1469                         //debug=1; /* ### comment me out ### */
1470                         break;
1471                 case 'v':
1472                         if (sizeof(optarg) > sizeof(char *) ||
1473                             !isdigit(optarg[0]))
1474                                 exit(1);
1475
1476                         conf->verbosity = atoi(optarg);
1477                         break;
1478                 case 'k':
1479                         uxclnt(optarg);
1480                         exit(0);
1481                 default:
1482                         ;
1483                 }
1484         }
1485
1486         if (!logsink)
1487                 err = 0;
1488         else
1489                 err = daemonize();
1490
1491         if (err < 0)
1492                 /* error */
1493                 exit(1);
1494         else if (err > 0)
1495                 /* parent dies */
1496                 exit(0);
1497         else
1498                 /* child lives */
1499                 return (child(NULL));
1500 }
1501