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         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 229             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 230             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 231                 alias = dm_mapname(major, minor);
 232         else
 233                 alias = STRDUP(devname);
 234
 235         if (!alias)
 236                 return 1;
 237
 238         map_present = dm_map_present(alias);
 239
 240         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 241                 condlog(4, "%s: not a multipath map", alias);
 242                 FREE(alias);
 243                 return 0;
 244         }
 245
 246         mpp = find_mp_by_alias(vecs->mpvec, alias);
 247
 248         if (mpp) {
 249                 /*
 250                  * Not really an error -- we generate our own uevent
 251                  * if we create a multipath mapped device as a result
 252                  * of uev_add_path
 253                  */
 254                 condlog(0, "%s: devmap already registered",
 255                         devname);
 256                 FREE(alias);
 257                 return 0;
 258         }
 259
 260         /*
 261          * now we can register the map
 262          */
 263         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 264                                         start_waiter_thread))) {
 265                 sync_map_state(mpp);
 266                 condlog(3, "%s: devmap %s added", alias, devname);
 267                 return 0;
 268         }
 269         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 270
 271         if (refwwid) {
 272                 r = coalesce_paths(vecs, NULL, refwwid);
 273                 dm_lib_release();
 274         }
 275
 276         if (!r)
 277                 condlog(3, "%s: devmap %s added", alias, devname);
 278         else
 279                 condlog(0, "%s: uev_add_map %s failed", alias, devname);
 280
 281         FREE(refwwid);
 282         FREE(alias);
 283         return r;
 284 }
 285
 286 static int
 287 uev_remove_map (char * devname, struct vectors * vecs)
 288 {
 289         condlog(2, "%s: remove map (uevent)", devname);
 290         return ev_remove_map(devname, vecs);
 291 }
 292
 293 int
 294 ev_remove_map (char * devname, struct vectors * vecs)
 295 {
 296         struct multipath * mpp;
 297
 298         mpp = find_mp_by_str(vecs->mpvec, devname);
 299
 300         if (!mpp) {
 301                 condlog(3, "%s: devmap not registered, can't remove",
 302                         devname);
 303                 return 0;
 304         }
 305         flush_map(mpp, vecs);
 306
 307         return 0;
 308 }
 309
 310 static int
 311 uev_umount_map (char * devname, struct vectors * vecs)
 312 {
 313         struct multipath * mpp;
 314
 315         condlog(2, "%s: umount map (uevent)", devname);
 316
 317         mpp = find_mp_by_str(vecs->mpvec, devname);
 318
 319         if (!mpp)
 320                 return 0;
 321
 322         update_mpp_paths(mpp, vecs->pathvec);
 323         verify_paths(mpp, vecs, NULL);
 324
 325         if (!VECTOR_SIZE(mpp->paths))
 326                 flush_map(mpp, vecs);
 327
 328         return 0;
 329 }
 330
 331 static int
 332 uev_add_path (char * devname, struct vectors * vecs)
 333 {
 334         condlog(2, "%s: add path (uevent)", devname);
 335         return (ev_add_path(devname, vecs) != 1)? 0 : 1;
 336 }
 337
 338
 339 /*
 340  * returns:
 341  * 0: added
 342  * 1: error
 343  * 2: blacklisted
 344  */
 345 int
 346 ev_add_path (char * devname, struct vectors * vecs)
 347 {
 348         struct multipath * mpp;
 349         struct path * pp;
 350         char empty_buff[WWID_SIZE] = {0};
 351
 352         pp = find_path_by_dev(vecs->pathvec, devname);
 353
 354         if (pp) {
 355                 condlog(0, "%s: spurious uevent, path already in pathvec",
 356                         devname);
 357                 if (pp->mpp)
 358                         return 0;
 359         }
 360         else {
 361                 /*
 362                  * get path vital state
 363                  */
 364                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 365                       devname, DI_ALL))) {
 366                         condlog(0, "%s: failed to store path info", devname);
 367                         return 1;
 368                 }
 369                 pp->checkint = conf->checkint;
 370         }
 371
 372         /*
 373          * need path UID to go any further
 374          */
 375         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 376                 condlog(0, "%s: failed to get path uid", devname);
 377                 return 1; /* leave path added to pathvec */
 378         }
 379         if (blacklist_path(conf, pp)){
 380                 int i = find_slot(vecs->pathvec, (void *)pp);
 381                 if (i != -1)
 382                         vector_del_slot(vecs->pathvec, i);
 383                 free_path(pp);
 384                 return 2;
 385         }
 386         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 387 rescan:
 388         if (mpp) {
 389                 if (adopt_paths(vecs->pathvec, mpp))
 390                         return 1; /* leave path added to pathvec */
 391
 392                 verify_paths(mpp, vecs, NULL);
 393                 mpp->action = ACT_RELOAD;
 394         }
 395         else {
 396                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 397                         mpp->action = ACT_CREATE;
 398                 else
 399                         return 1; /* leave path added to pathvec */
 400         }
 401
 402         /*
 403          * push the map to the device-mapper
 404          */
 405         if (setup_map(mpp)) {
 406                 condlog(0, "%s: failed to setup map for addition of new "
 407                         "path %s", mpp->alias, devname);
 408                 goto out;
 409         }
 410         /*
 411          * reload the map for the multipath mapped device
 412          */
 413         if (domap(mpp) <= 0) {
 414                 condlog(0, "%s: failed in domap for addition of new "
 415                         "path %s", mpp->alias, devname);
 416                 /*
 417                  * deal with asynchronous uevents :((
 418                  */
 419                 if (mpp->action == ACT_RELOAD) {
 420                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 421                         sleep(1);
 422                         update_mpp_paths(mpp, vecs->pathvec);
 423                         goto rescan;
 424                 }
 425                 else
 426                         goto out;
 427         }
 428         dm_lib_release();
 429
 430         /*
 431          * update our state from kernel regardless of create or reload
 432          */
 433         if (setup_multipath(vecs, mpp))
 434                 goto out;
 435
 436         sync_map_state(mpp);
 437
 438         if (mpp->action == ACT_CREATE &&
 439             start_waiter_thread(mpp, vecs))
 440                         goto out;
 441
 442         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 443         return 0;
 444
 445 out:
 446         remove_map(mpp, vecs, NULL, 1);
 447         return 1;
 448 }
 449
 450 static int
 451 uev_remove_path (char * devname, struct vectors * vecs)
 452 {
 453         condlog(2, "%s: remove path (uevent)", devname);
 454         return ev_remove_path(devname, vecs);
 455 }
 456
 457 int
 458 ev_remove_path (char * devname, struct vectors * vecs)
 459 {
 460         struct multipath * mpp;
 461         struct path * pp;
 462         int i;
 463         int rm_path = 1;
 464
 465         pp = find_path_by_dev(vecs->pathvec, devname);
 466
 467         if (!pp) {
 468                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 469                 return 1;
 470         }
 471
 472         /*
 473          * avoid referring to the map of an orphanned path
 474          */
 475         if ((mpp = pp->mpp)) {
 476
 477                 /*
 478                  * remove the map IFF removing the last path
 479                  */
 480                 if (pathcount(mpp, PATH_WILD) > 1) {
 481                         vector rpvec = vector_alloc();
 482
 483                         /*
 484                          * transform the mp->pg vector of vectors of paths
 485                          * into a mp->params string to feed the device-mapper
 486                          */
 487                         update_mpp_paths(mpp, vecs->pathvec);
 488                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 489                                 vector_del_slot(mpp->paths, i);
 490
 491                         if (VECTOR_SIZE(mpp->paths) == 0) {
 492                                 char alias[WWID_SIZE];
 493
 494                                 /*
 495                                  * flush_map will fail if the device is open
 496                                  */
 497                                 strncpy(alias, mpp->alias, WWID_SIZE);
 498                                 if (flush_map(mpp, vecs))
 499                                         rm_path = 0;
 500                                 else
 501                                         condlog(3, "%s: removed map after removing"
 502                                                 " multiple paths", alias);
 503                         }
 504                         else {
 505                                 if (setup_map(mpp)) {
 506                                         condlog(0, "%s: failed to setup map for"
 507                                                 " removal of path %s", mpp->alias, devname);
 508                                         free_pathvec(rpvec, KEEP_PATHS);
 509                                         goto out;
 510                                 }
 511                                 /*
 512                                  * reload the map
 513                                  */
 514                                 mpp->action = ACT_RELOAD;
 515                                 if (domap(mpp) <= 0) {
 516                                         condlog(0, "%s: failed in domap for "
 517                                                 "removal of path %s",
 518                                                 mpp->alias, devname);
 519                                         /*
 520                                          * Delete path from pathvec so that
 521                                          * update_mpp_paths wont find it later
 522                                          * when/if another path is removed.
 523                                          */
 524                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 525                                                 vector_del_slot(vecs->pathvec, i);
 526                                         free_path(pp);
 527                                         return 1;
 528                                 }
 529                                 /*
 530                                  * update our state from kernel
 531                                  */
 532                                 if (setup_multipath(vecs, mpp)) {
 533                                         free_pathvec(rpvec, KEEP_PATHS);
 534                                         goto out;
 535                                 }
 536                                 sync_map_state(mpp);
 537
 538                                 condlog(3, "%s: path removed from map %s",
 539                                         devname, mpp->alias);
 540                         }
 541                         free_pathvec(rpvec, KEEP_PATHS);
 542                 }
 543                 else {
 544                         char alias[WWID_SIZE];
 545
 546                         /*
 547                          * flush_map will fail if the device is open
 548                          */
 549                         strncpy(alias, mpp->alias, WWID_SIZE);
 550                         if (flush_map(mpp, vecs))
 551                                 rm_path = 0;
 552                         else
 553                                 condlog(3, "%s: removed map", alias);
 554                 }
 555         }
 556
 557         if (rm_path) {
 558                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 559                         vector_del_slot(vecs->pathvec, i);
 560                 free_path(pp);
 561         }
 562
 563         return 0;
 564
 565 out:
 566         remove_map(mpp, vecs, stop_waiter_thread, 1);
 567         return 1;
 568 }
 569
 570 static int
 571 map_discovery (struct vectors * vecs)
 572 {
 573         struct multipath * mpp;
 574         unsigned int i;
 575
 576         if (dm_get_maps(vecs->mpvec, "multipath"))
 577                 return 1;
 578
 579         vector_foreach_slot (vecs->mpvec, mpp, i)
 580                 if (setup_multipath(vecs, mpp))
 581                         return 1;
 582
 583         return 0;
 584 }
 585
 586 int
 587 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 588 {
 589         struct vectors * vecs;
 590         int r;
 591
 592         *reply = NULL;
 593         *len = 0;
 594         vecs = (struct vectors *)trigger_data;
 595
 596         pthread_cleanup_push(cleanup_lock, vecs->lock);
 597         lock(vecs->lock);
 598
 599         r = parse_cmd(str, reply, len, vecs);
 600
 601         if (r > 0) {
 602                 *reply = STRDUP("fail\n");
 603                 *len = strlen(*reply) + 1;
 604                 r = 1;
 605         }
 606         else if (!r && *len == 0) {
 607                 *reply = STRDUP("ok\n");
 608                 *len = strlen(*reply) + 1;
 609                 r = 0;
 610         }
 611         /* else if (r < 0) leave *reply alone */
 612
 613         lock_cleanup_pop(vecs->lock);
 614         return r;
 615 }
 616
 617 static int
 618 uev_discard(char * devpath)
 619 {
 620         char a[10], b[10];
 621
 622         /*
 623          * keep only block devices, discard partitions
 624          */
 625         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 626             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 627                 condlog(4, "discard event on %s", devpath);
 628                 return 1;
 629         }
 630         return 0;
 631 }
 632
 633 int
 634 uev_trigger (struct uevent * uev, void * trigger_data)
 635 {
 636         int r = 0;
 637         char devname[32];
 638         struct vectors * vecs;
 639
 640         vecs = (struct vectors *)trigger_data;
 641
 642         if (uev_discard(uev->devpath))
 643                 return 0;
 644
 645         basename(uev->devpath, devname);
 646         lock(vecs->lock);
 647
 648         /*
 649          * device map add/remove event
 650          */
 651         if (!strncmp(devname, "dm-", 3)) {
 652                 if (!strncmp(uev->action, "add", 3)) {
 653                         r = uev_add_map(devname, vecs);
 654                         goto out;
 655                 }
 656                 if (!strncmp(uev->action, "remove", 6)) {
 657                         r = uev_remove_map(devname, vecs);
 658                         goto out;
 659                 }
 660                 if (!strncmp(uev->action, "umount", 6)) {
 661                         r = uev_umount_map(devname, vecs);
 662                         goto out;
 663                 }
 664                 goto out;
 665         }
 666
 667         /*
 668          * path add/remove event
 669          */
 670         if (blacklist(conf->blist_devnode, conf->elist_devnode, devname))
 671                 goto out;
 672
 673         if (!strncmp(uev->action, "add", 3)) {
 674                 r = uev_add_path(devname, vecs);
 675                 goto out;
 676         }
 677         if (!strncmp(uev->action, "remove", 6)) {
 678                 r = uev_remove_path(devname, vecs);
 679                 goto out;
 680         }
 681
 682 out:
 683         unlock(vecs->lock);
 684         return r;
 685 }
 686
 687 static void *
 688 ueventloop (void * ap)
 689 {
 690         if (uevent_listen(&uev_trigger, ap))
 691                 fprintf(stderr, "error starting uevent listener");
 692
 693         return NULL;
 694 }
 695
 696 static void *
 697 uxlsnrloop (void * ap)
 698 {
 699         if (load_keys())
 700                 return NULL;
 701
 702         if (alloc_handlers())
 703                 return NULL;
 704
 705         add_handler(LIST+PATHS, cli_list_paths);
 706         add_handler(LIST+MAPS, cli_list_maps);
 707         add_handler(LIST+MAPS+STATUS, cli_list_maps_status);
 708         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
 709         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 710         add_handler(LIST+TOPOLOGY, cli_list_maps_topology);
 711         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 712         add_handler(LIST+CONFIG, cli_list_config);
 713         add_handler(LIST+BLACKLIST, cli_list_blacklist);
 714         add_handler(LIST+DEVICES, cli_list_devices);
 715         add_handler(ADD+PATH, cli_add_path);
 716         add_handler(DEL+PATH, cli_del_path);
 717         add_handler(ADD+MAP, cli_add_map);
 718         add_handler(DEL+MAP, cli_del_map);
 719         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
 720         add_handler(RECONFIGURE, cli_reconfigure);
 721         add_handler(SUSPEND+MAP, cli_suspend);
 722         add_handler(RESUME+MAP, cli_resume);
 723         add_handler(REINSTATE+PATH, cli_reinstate);
 724         add_handler(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 (blacklist_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         if (conf->verbosity > 2)
1073                 vector_foreach_slot(mpvec, mpp, i)
1074                         print_map(mpp);
1075
1076         /*
1077          * purge dm of old maps
1078          */
1079         remove_maps(vecs, NULL);
1080
1081         /*
1082          * save new set of maps formed by considering current path state
1083          */
1084         vector_free(vecs->mpvec);
1085         vecs->mpvec = mpvec;
1086
1087         /*
1088          * start dm event waiter threads for these new maps
1089          */
1090         vector_foreach_slot(vecs->mpvec, mpp, i) {
1091                 if (setup_multipath(vecs, mpp))
1092                         return 1;
1093                 if (start_waiters)
1094                         if (start_waiter_thread(mpp, vecs))
1095                                 return 1;
1096         }
1097         return 0;
1098 }
1099
1100 int
1101 reconfigure (struct vectors * vecs)
1102 {
1103         struct config * old = conf;
1104
1105         /*
1106          * free old map and path vectors ... they use old conf state
1107          */
1108         if (VECTOR_SIZE(vecs->mpvec))
1109                 remove_maps(vecs, stop_waiter_thread);
1110
1111         if (VECTOR_SIZE(vecs->pathvec))
1112                 free_pathvec(vecs->pathvec, FREE_PATHS);
1113
1114         vecs->pathvec = NULL;
1115         conf = NULL;
1116
1117         if (load_config(DEFAULT_CONFIGFILE))
1118                 return 1;
1119
1120         conf->verbosity = old->verbosity;
1121
1122         if (!conf->checkint) {
1123                 conf->checkint = DEFAULT_CHECKINT;
1124                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1125         }
1126         configure(vecs, 1);
1127         free_config(old);
1128         return 0;
1129 }
1130
1131 static struct vectors *
1132 init_vecs (void)
1133 {
1134         struct vectors * vecs;
1135
1136         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1137
1138         if (!vecs)
1139                 return NULL;
1140
1141         vecs->lock =
1142                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1143
1144         if (!vecs->lock)
1145                 goto out;
1146
1147         pthread_mutex_init(vecs->lock, NULL);
1148
1149         return vecs;
1150
1151 out:
1152         FREE(vecs);
1153         condlog(0, "failed to init paths");
1154         return NULL;
1155 }
1156
1157 static void *
1158 signal_set(int signo, void (*func) (int))
1159 {
1160         int r;
1161         struct sigaction sig;
1162         struct sigaction osig;
1163
1164         sig.sa_handler = func;
1165         sigemptyset(&sig.sa_mask);
1166         sig.sa_flags = 0;
1167
1168         r = sigaction(signo, &sig, &osig);
1169
1170         if (r < 0)
1171                 return (SIG_ERR);
1172         else
1173                 return (osig.sa_handler);
1174 }
1175
1176 static void
1177 sighup (int sig)
1178 {
1179         condlog(2, "reconfigure (SIGHUP)");
1180
1181         lock(gvecs->lock);
1182         reconfigure(gvecs);
1183         unlock(gvecs->lock);
1184
1185 #ifdef _DEBUG_
1186         dbg_free_final(NULL);
1187 #endif
1188 }
1189
1190 static void
1191 sigend (int sig)
1192 {
1193         exit_daemon(0);
1194 }
1195
1196 static void
1197 sigusr1 (int sig)
1198 {
1199         condlog(3, "SIGUSR1 received");
1200 }
1201
1202 static void
1203 signal_init(void)
1204 {
1205         signal_set(SIGHUP, sighup);
1206         signal_set(SIGUSR1, sigusr1);
1207         signal_set(SIGINT, sigend);
1208         signal_set(SIGTERM, sigend);
1209         signal(SIGPIPE, SIG_IGN);
1210 }
1211
1212 static void
1213 setscheduler (void)
1214 {
1215         int res;
1216         static struct sched_param sched_param = {
1217                 .sched_priority = 99
1218         };
1219
1220         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1221
1222         if (res == -1)
1223                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1224         return;
1225 }
1226
1227 static void
1228 set_oom_adj (int val)
1229 {
1230         FILE *fp;
1231
1232         fp = fopen("/proc/self/oom_adj", "w");
1233
1234         if (!fp)
1235                 return;
1236
1237         fprintf(fp, "%i", val);
1238         fclose(fp);
1239 }
1240
1241 static int
1242 child (void * param)
1243 {
1244         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1245         pthread_attr_t attr;
1246         struct vectors * vecs;
1247
1248         mlockall(MCL_CURRENT | MCL_FUTURE);
1249
1250         if (logsink)
1251                 log_thread_start();
1252
1253         condlog(2, "--------start up--------");
1254         condlog(2, "read " DEFAULT_CONFIGFILE);
1255
1256         if (load_config(DEFAULT_CONFIGFILE))
1257                 exit(1);
1258
1259         setlogmask(LOG_UPTO(conf->verbosity + 3));
1260
1261         /*
1262          * fill the voids left in the config file
1263          */
1264         if (!conf->checkint) {
1265                 conf->checkint = DEFAULT_CHECKINT;
1266                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1267         }
1268
1269         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1270                 if (logsink)
1271                         log_thread_stop();
1272
1273                 exit(1);
1274         }
1275         signal_init();
1276         setscheduler();
1277         set_oom_adj(-16);
1278         vecs = gvecs = init_vecs();
1279
1280         if (!vecs)
1281                 exit(1);
1282
1283         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1284                 condlog(0, "can not find sysfs mount point");
1285                 exit(1);
1286         }
1287
1288         /*
1289          * fetch and configure both paths and multipaths
1290          */
1291         if (configure(vecs, 1)) {
1292                 condlog(0, "failure during configuration");
1293                 exit(1);
1294         }
1295
1296         /*
1297          * start threads
1298          */
1299         pthread_attr_init(&attr);
1300         pthread_attr_setstacksize(&attr, 64 * 1024);
1301         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1302
1303         pthread_create(&check_thr, &attr, checkerloop, vecs);
1304         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1305         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1306
1307         pthread_cond_wait(&exit_cond, &exit_mutex);
1308
1309         /*
1310          * exit path
1311          */
1312         lock(vecs->lock);
1313         remove_maps(vecs, stop_waiter_thread);
1314         free_pathvec(vecs->pathvec, FREE_PATHS);
1315
1316         pthread_cancel(check_thr);
1317         pthread_cancel(uevent_thr);
1318         pthread_cancel(uxlsnr_thr);
1319
1320         free_keys(keys);
1321         keys = NULL;
1322         free_handlers(handlers);
1323         handlers = NULL;
1324         free_polls();
1325
1326         unlock(vecs->lock);
1327         pthread_mutex_destroy(vecs->lock);
1328         FREE(vecs->lock);
1329         vecs->lock = NULL;
1330         FREE(vecs);
1331         vecs = NULL;
1332         free_config(conf);
1333         conf = NULL;
1334
1335         condlog(2, "--------shut down-------");
1336
1337         if (logsink)
1338                 log_thread_stop();
1339
1340         dm_lib_release();
1341         dm_lib_exit();
1342
1343 #ifdef _DEBUG_
1344         dbg_free_final(NULL);
1345 #endif
1346
1347         exit(0);
1348 }
1349
1350 static int
1351 daemonize(void)
1352 {
1353         int pid;
1354         int in_fd, out_fd;
1355
1356         if( (pid = fork()) < 0){
1357                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1358                 return -1;
1359         }
1360         else if (pid != 0)
1361                 return pid;
1362
1363         setsid();
1364
1365         if ( (pid = fork()) < 0)
1366                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1367         else if (pid != 0)
1368                 _exit(0);
1369
1370         in_fd = open("/dev/null", O_RDONLY);
1371         if (in_fd < 0){
1372                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1373                         strerror(errno));
1374                 _exit(0);
1375         }
1376         out_fd = open("/dev/console", O_WRONLY);
1377         if (out_fd < 0){
1378                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1379                         strerror(errno));
1380                 _exit(0);
1381         }
1382
1383         close(STDIN_FILENO);
1384         dup(in_fd);
1385         close(STDOUT_FILENO);
1386         dup(out_fd);
1387         close(STDERR_FILENO);
1388         dup(out_fd);
1389
1390         close(in_fd);
1391         close(out_fd);
1392         chdir("/");
1393         umask(0);
1394         return 0;
1395 }
1396
1397 int
1398 main (int argc, char *argv[])
1399 {
1400         extern char *optarg;
1401         extern int optind;
1402         int arg;
1403         int err;
1404
1405         logsink = 1;
1406
1407         if (getuid() != 0) {
1408                 fprintf(stderr, "need to be root\n");
1409                 exit(1);
1410         }
1411
1412         /* make sure we don't lock any path */
1413         chdir("/");
1414         umask(umask(077) | 022);
1415
1416         conf = alloc_config();
1417
1418         if (!conf)
1419                 exit(1);
1420
1421         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1422         switch(arg) {
1423                 case 'd':
1424                         logsink = 0;
1425                         //debug=1; /* ### comment me out ### */
1426                         break;
1427                 case 'v':
1428                         if (sizeof(optarg) > sizeof(char *) ||
1429                             !isdigit(optarg[0]))
1430                                 exit(1);
1431
1432                         conf->verbosity = atoi(optarg);
1433                         break;
1434                 case 'k':
1435                         uxclnt(optarg);
1436                         exit(0);
1437                 default:
1438                         ;
1439                 }
1440         }
1441
1442         if (!logsink)
1443                 err = 0;
1444         else
1445                 err = daemonize();
1446
1447         if (err < 0)
1448                 /* error */
1449                 exit(1);
1450         else if (err > 0)
1451                 /* parent dies */
1452                 exit(0);
1453         else
1454                 /* child lives */
1455                 return (child(NULL));
1456 }
1457