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