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