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  * libcheckers
  18  */
  19 #include <checkers.h>
  20
  21 /*
  22  * libmultipath
  23  */
  24 #include <parser.h>
  25 #include <vector.h>
  26 #include <memory.h>
  27 #include <config.h>
  28 #include <util.h>
  29 #include <hwtable.h>
  30 #include <defaults.h>
  31 #include <structs.h>
  32 #include <callout.h>
  33 #include <blacklist.h>
  34 #include <structs_vec.h>
  35 #include <dmparser.h>
  36 #include <devmapper.h>
  37 #include <sysfs.h>
  38 #include <dict.h>
  39 #include <discovery.h>
  40 #include <debug.h>
  41 #include <propsel.h>
  42 #include <uevent.h>
  43 #include <switchgroup.h>
  44 #include <print.h>
  45 #include <configure.h>
  46
  47 #include "main.h"
  48 #include "pidfile.h"
  49 #include "uxlsnr.h"
  50 #include "uxclnt.h"
  51 #include "cli.h"
  52 #include "cli_handlers.h"
  53 #include "lock.h"
  54 #include "waiter.h"
  55
  56 #define FILE_NAME_SIZE 256
  57 #define CMDSIZE 160
  58
  59 #define LOG_MSG(a,b) \
  60         if (strlen(b)) condlog(a, "%s: %s", pp->dev, b);
  61
  62 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  63 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  64
  65 /*
  66  * global copy of vecs for use in sig handlers
  67  */
  68 struct vectors * gvecs;
  69
  70 static int
  71 need_switch_pathgroup (struct multipath * mpp, int refresh)
  72 {
  73         struct pathgroup * pgp;
  74         struct path * pp;
  75         unsigned int i, j;
  76
  77         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  78                 return 0;
  79
  80         /*
  81          * Refresh path priority values
  82          */
  83         if (refresh)
  84                 vector_foreach_slot (mpp->pg, pgp, i)
  85                         vector_foreach_slot (pgp->paths, pp, j)
  86                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  87
  88         mpp->bestpg = select_path_group(mpp);
  89
  90         if (mpp->bestpg != mpp->nextpg)
  91                 return 1;
  92
  93         return 0;
  94 }
  95
  96 static void
  97 switch_pathgroup (struct multipath * mpp)
  98 {
  99         mpp->stat_switchgroup++;
 100         dm_switchgroup(mpp->alias, mpp->bestpg);
 101         condlog(2, "%s: switch to path group #%i",
 102                  mpp->alias, mpp->bestpg);
 103 }
 104
 105 static int
 106 coalesce_maps(struct vectors *vecs, vector nmpv)
 107 {
 108         struct multipath * ompp;
 109         vector ompv = vecs->mpvec;
 110         unsigned int i;
 111         int j;
 112
 113         vector_foreach_slot (ompv, ompp, i) {
 114                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 115                         /*
 116                          * remove all current maps not allowed by the
 117                          * current configuration
 118                          */
 119                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 120                                 condlog(0, "%s: unable to flush devmap",
 121                                         ompp->alias);
 122                                 /*
 123                                  * may be just because the device is open
 124                                  */
 125                                 if (!vector_alloc_slot(nmpv))
 126                                         return 1;
 127
 128                                 vector_set_slot(nmpv, ompp);
 129                                 setup_multipath(vecs, ompp);
 130
 131                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 132                                         vector_del_slot(ompv, j);
 133
 134                                 continue;
 135                         }
 136                         else {
 137                                 dm_lib_release();
 138                                 condlog(3, "%s devmap removed", ompp->alias);
 139                         }
 140                 }
 141         }
 142         return 0;
 143 }
 144
 145 static void
 146 sync_map_state(struct multipath *mpp)
 147 {
 148         struct pathgroup *pgp;
 149         struct path *pp;
 150         unsigned int i, j;
 151
 152         vector_foreach_slot (mpp->pg, pgp, i){
 153                 vector_foreach_slot (pgp->paths, pp, j){
 154                         if (pp->state <= PATH_UNCHECKED)
 155                                 continue;
 156                         if ((pp->dmstate == PSTATE_FAILED ||
 157                              pp->dmstate == PSTATE_UNDEF) &&
 158                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 159                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 160                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 161                                   pp->dmstate == PSTATE_UNDEF) &&
 162                                  (pp->state == PATH_DOWN ||
 163                                   pp->state == PATH_SHAKY))
 164                                 dm_fail_path(mpp->alias, pp->dev_t);
 165                 }
 166         }
 167 }
 168
 169 static void
 170 sync_maps_state(vector mpvec)
 171 {
 172         unsigned int i;
 173         struct multipath *mpp;
 174
 175         vector_foreach_slot (mpvec, mpp, i)
 176                 sync_map_state(mpp);
 177 }
 178
 179 static int
 180 flush_map(struct multipath * mpp, struct vectors * vecs)
 181 {
 182         /*
 183          * clear references to this map before flushing so we can ignore
 184          * the spurious uevent we may generate with the dm_flush_map call below
 185          */
 186         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 187                 /*
 188                  * May not really be an error -- if the map was already flushed
 189                  * from the device mapper by dmsetup(8) for instance.
 190                  */
 191                 condlog(0, "%s: can't flush", mpp->alias);
 192                 return 1;
 193         }
 194         else {
 195                 dm_lib_release();
 196                 condlog(3, "%s: devmap removed", mpp->alias);
 197         }
 198
 199         orphan_paths(vecs->pathvec, mpp);
 200         remove_map(mpp, vecs, stop_waiter_thread, 1);
 201
 202         return 0;
 203 }
 204
 205 static int
 206 uev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 207 {
 208         condlog(2, "%s: add map (uevent)", dev->kernel);
 209         return ev_add_map(dev, vecs);
 210 }
 211
 212 int
 213 ev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 214 {
 215         char * alias;
 216         char *dev_t;
 217         int major, minor;
 218         char * refwwid;
 219         struct multipath * mpp;
 220         int map_present;
 221         int r = 1;
 222
 223         dev_t = sysfs_attr_get_value(dev->devpath, "dev");
 224
 225         if (!dev_t || sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 226                 return 1;
 227
 228         alias = dm_mapname(major, minor);
 229
 230         if (!alias)
 231                 return 1;
 232
 233         map_present = dm_map_present(alias);
 234
 235         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 236                 condlog(4, "%s: not a multipath map", alias);
 237                 return 0;
 238         }
 239
 240         mpp = find_mp_by_alias(vecs->mpvec, alias);
 241
 242         if (mpp) {
 243                 /*
 244                  * Not really an error -- we generate our own uevent
 245                  * if we create a multipath mapped device as a result
 246                  * of uev_add_path
 247                  */
 248                 condlog(0, "%s: devmap already registered",
 249                         dev->kernel);
 250                 return 0;
 251         }
 252
 253         /*
 254          * now we can register the map
 255          */
 256         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 257                                         start_waiter_thread))) {
 258                 sync_map_state(mpp);
 259                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 260                 return 0;
 261         }
 262         refwwid = get_refwwid(dev->kernel, DEV_DEVMAP, vecs->pathvec);
 263
 264         if (refwwid) {
 265                 r = coalesce_paths(vecs, NULL, refwwid);
 266                 dm_lib_release();
 267         }
 268
 269         if (!r)
 270                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 271         else
 272                 condlog(0, "%s: uev_add_map %s failed", alias, dev->kernel);
 273
 274         FREE(refwwid);
 275         return r;
 276 }
 277
 278 static int
 279 uev_remove_map (struct sysfs_device * dev, struct vectors * vecs)
 280 {
 281         condlog(2, "%s: remove map (uevent)", dev->kernel);
 282         return ev_remove_map(dev->kernel, vecs);
 283 }
 284
 285 int
 286 ev_remove_map (char * devname, struct vectors * vecs)
 287 {
 288         struct multipath * mpp;
 289
 290         mpp = find_mp_by_str(vecs->mpvec, devname);
 291
 292         if (!mpp) {
 293                 condlog(3, "%s: devmap not registered, can't remove",
 294                         devname);
 295                 return 0;
 296         }
 297         flush_map(mpp, vecs);
 298
 299         return 0;
 300 }
 301
 302 static int
 303 uev_umount_map (struct sysfs_device * dev, struct vectors * vecs)
 304 {
 305         struct multipath * mpp;
 306
 307         condlog(2, "%s: umount map (uevent)", dev->kernel);
 308
 309         mpp = find_mp_by_str(vecs->mpvec, dev->kernel);
 310
 311         if (!mpp)
 312                 return 0;
 313
 314         update_mpp_paths(mpp, vecs->pathvec);
 315         verify_paths(mpp, vecs, NULL);
 316
 317         if (!VECTOR_SIZE(mpp->paths))
 318                 flush_map(mpp, vecs);
 319
 320         return 0;
 321 }
 322
 323 static int
 324 uev_add_path (struct sysfs_device * dev, struct vectors * vecs)
 325 {
 326         condlog(2, "%s: add path (uevent)", dev->kernel);
 327         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 328 }
 329
 330
 331 /*
 332  * returns:
 333  * 0: added
 334  * 1: error
 335  * 2: blacklisted
 336  */
 337 int
 338 ev_add_path (char * devname, struct vectors * vecs)
 339 {
 340         struct multipath * mpp;
 341         struct path * pp;
 342         char empty_buff[WWID_SIZE] = {0};
 343
 344         pp = find_path_by_dev(vecs->pathvec, devname);
 345
 346         if (pp) {
 347                 condlog(0, "%s: spurious uevent, path already in pathvec",
 348                         devname);
 349                 if (pp->mpp)
 350                         return 0;
 351         }
 352         else {
 353                 /*
 354                  * get path vital state
 355                  */
 356                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 357                       devname, DI_ALL))) {
 358                         condlog(0, "%s: failed to store path info", devname);
 359                         return 1;
 360                 }
 361                 pp->checkint = conf->checkint;
 362         }
 363
 364         /*
 365          * need path UID to go any further
 366          */
 367         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 368                 condlog(0, "%s: failed to get path uid", devname);
 369                 return 1; /* leave path added to pathvec */
 370         }
 371         if (filter_path(conf, pp)){
 372                 int i = find_slot(vecs->pathvec, (void *)pp);
 373                 if (i != -1)
 374                         vector_del_slot(vecs->pathvec, i);
 375                 free_path(pp);
 376                 return 2;
 377         }
 378         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 379 rescan:
 380         if (mpp) {
 381                 if (adopt_paths(vecs->pathvec, mpp))
 382                         return 1; /* leave path added to pathvec */
 383
 384                 verify_paths(mpp, vecs, NULL);
 385                 mpp->action = ACT_RELOAD;
 386         }
 387         else {
 388                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 389                         mpp->action = ACT_CREATE;
 390                 else
 391                         return 1; /* leave path added to pathvec */
 392         }
 393
 394         /*
 395          * push the map to the device-mapper
 396          */
 397         if (setup_map(mpp)) {
 398                 condlog(0, "%s: failed to setup map for addition of new "
 399                         "path %s", mpp->alias, devname);
 400                 goto out;
 401         }
 402         /*
 403          * reload the map for the multipath mapped device
 404          */
 405         if (domap(mpp) <= 0) {
 406                 condlog(0, "%s: failed in domap for addition of new "
 407                         "path %s", mpp->alias, devname);
 408                 /*
 409                  * deal with asynchronous uevents :((
 410                  */
 411                 if (mpp->action == ACT_RELOAD) {
 412                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 413                         sleep(1);
 414                         update_mpp_paths(mpp, vecs->pathvec);
 415                         goto rescan;
 416                 }
 417                 else
 418                         goto out;
 419         }
 420         dm_lib_release();
 421
 422         /*
 423          * update our state from kernel regardless of create or reload
 424          */
 425         if (setup_multipath(vecs, mpp))
 426                 goto out;
 427
 428         sync_map_state(mpp);
 429
 430         if (mpp->action == ACT_CREATE &&
 431             start_waiter_thread(mpp, vecs))
 432                         goto out;
 433
 434         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 435         return 0;
 436
 437 out:
 438         remove_map(mpp, vecs, NULL, 1);
 439         return 1;
 440 }
 441
 442 static int
 443 uev_remove_path (struct sysfs_device * dev, struct vectors * vecs)
 444 {
 445         condlog(2, "%s: remove path (uevent)", dev->kernel);
 446         return ev_remove_path(dev->kernel, vecs);
 447 }
 448
 449 int
 450 ev_remove_path (char * devname, struct vectors * vecs)
 451 {
 452         struct multipath * mpp;
 453         struct path * pp;
 454         int i;
 455         int rm_path = 1;
 456
 457         pp = find_path_by_dev(vecs->pathvec, devname);
 458
 459         if (!pp) {
 460                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 461                 return 1;
 462         }
 463
 464         /*
 465          * avoid referring to the map of an orphanned path
 466          */
 467         if ((mpp = pp->mpp)) {
 468
 469                 /*
 470                  * remove the map IFF removing the last path
 471                  */
 472                 if (pathcount(mpp, PATH_WILD) > 1) {
 473                         vector rpvec = vector_alloc();
 474
 475                         /*
 476                          * transform the mp->pg vector of vectors of paths
 477                          * into a mp->params string to feed the device-mapper
 478                          */
 479                         update_mpp_paths(mpp, vecs->pathvec);
 480                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 481                                 vector_del_slot(mpp->paths, i);
 482
 483                         if (VECTOR_SIZE(mpp->paths) == 0) {
 484                                 char alias[WWID_SIZE];
 485
 486                                 /*
 487                                  * flush_map will fail if the device is open
 488                                  */
 489                                 strncpy(alias, mpp->alias, WWID_SIZE);
 490                                 if (flush_map(mpp, vecs))
 491                                         rm_path = 0;
 492                                 else
 493                                         condlog(3, "%s: removed map after removing"
 494                                                 " multiple paths", alias);
 495                         }
 496                         else {
 497                                 if (setup_map(mpp)) {
 498                                         condlog(0, "%s: failed to setup map for"
 499                                                 " removal of path %s", mpp->alias, devname);
 500                                         free_pathvec(rpvec, KEEP_PATHS);
 501                                         goto out;
 502                                 }
 503                                 /*
 504                                  * reload the map
 505                                  */
 506                                 mpp->action = ACT_RELOAD;
 507                                 if (domap(mpp) <= 0) {
 508                                         condlog(0, "%s: failed in domap for "
 509                                                 "removal of path %s",
 510                                                 mpp->alias, devname);
 511                                         /*
 512                                          * Delete path from pathvec so that
 513                                          * update_mpp_paths wont find it later
 514                                          * when/if another path is removed.
 515                                          */
 516                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 517                                                 vector_del_slot(vecs->pathvec, i);
 518                                         free_path(pp);
 519                                         return 1;
 520                                 }
 521                                 /*
 522                                  * update our state from kernel
 523                                  */
 524                                 if (setup_multipath(vecs, mpp)) {
 525                                         free_pathvec(rpvec, KEEP_PATHS);
 526                                         goto out;
 527                                 }
 528                                 sync_map_state(mpp);
 529
 530                                 condlog(3, "%s: path removed from map %s",
 531                                         devname, mpp->alias);
 532                         }
 533                         free_pathvec(rpvec, KEEP_PATHS);
 534                 }
 535                 else {
 536                         char alias[WWID_SIZE];
 537
 538                         /*
 539                          * flush_map will fail if the device is open
 540                          */
 541                         strncpy(alias, mpp->alias, WWID_SIZE);
 542                         if (flush_map(mpp, vecs))
 543                                 rm_path = 0;
 544                         else
 545                                 condlog(3, "%s: removed map", alias);
 546                 }
 547         }
 548
 549         if (rm_path) {
 550                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 551                         vector_del_slot(vecs->pathvec, i);
 552                 free_path(pp);
 553         }
 554
 555         return 0;
 556
 557 out:
 558         remove_map(mpp, vecs, stop_waiter_thread, 1);
 559         return 1;
 560 }
 561
 562 static int
 563 map_discovery (struct vectors * vecs)
 564 {
 565         struct multipath * mpp;
 566         unsigned int i;
 567
 568         if (dm_get_maps(vecs->mpvec, "multipath"))
 569                 return 1;
 570
 571         vector_foreach_slot (vecs->mpvec, mpp, i)
 572                 if (setup_multipath(vecs, mpp))
 573                         return 1;
 574
 575         return 0;
 576 }
 577
 578 int
 579 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 580 {
 581         struct vectors * vecs;
 582         int r;
 583
 584         *reply = NULL;
 585         *len = 0;
 586         vecs = (struct vectors *)trigger_data;
 587
 588         pthread_cleanup_push(cleanup_lock, vecs->lock);
 589         lock(vecs->lock);
 590
 591         r = parse_cmd(str, reply, len, vecs);
 592
 593         if (r > 0) {
 594                 *reply = STRDUP("fail\n");
 595                 *len = strlen(*reply) + 1;
 596                 r = 1;
 597         }
 598         else if (!r && *len == 0) {
 599                 *reply = STRDUP("ok\n");
 600                 *len = strlen(*reply) + 1;
 601                 r = 0;
 602         }
 603         /* else if (r < 0) leave *reply alone */
 604
 605         lock_cleanup_pop(vecs->lock);
 606         return r;
 607 }
 608
 609 static int
 610 uev_discard(char * devpath)
 611 {
 612         char a[10], b[10];
 613
 614         /*
 615          * keep only block devices, discard partitions
 616          */
 617         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 618             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 619                 condlog(4, "discard event on %s", devpath);
 620                 return 1;
 621         }
 622         return 0;
 623 }
 624
 625 int
 626 uev_trigger (struct uevent * uev, void * trigger_data)
 627 {
 628         int r = 0;
 629         struct sysfs_device *sysdev;
 630         struct vectors * vecs;
 631
 632         vecs = (struct vectors *)trigger_data;
 633
 634         if (uev_discard(uev->devpath))
 635                 return 0;
 636
 637         sysdev = sysfs_device_get(uev->devpath);
 638         if(!sysdev)
 639                 return 0;
 640
 641         lock(vecs->lock);
 642
 643         /*
 644          * device map event
 645          * Add events are ignored here as the tables
 646          * are not fully initialised then.
 647          */
 648         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 649                 if (!strncmp(uev->action, "change", 6)) {
 650                         r = uev_add_map(sysdev, vecs);
 651                         goto out;
 652                 }
 653                 if (!strncmp(uev->action, "remove", 6)) {
 654                         r = uev_remove_map(sysdev, vecs);
 655                         goto out;
 656                 }
 657                 if (!strncmp(uev->action, "umount", 6)) {
 658                         r = uev_umount_map(sysdev, vecs);
 659                         goto out;
 660                 }
 661                 goto out;
 662         }
 663
 664         /*
 665          * path add/remove event
 666          */
 667         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 668                            sysdev->kernel) > 0)
 669                 goto out;
 670
 671         if (!strncmp(uev->action, "add", 3)) {
 672                 r = uev_add_path(sysdev, vecs);
 673                 goto out;
 674         }
 675         if (!strncmp(uev->action, "remove", 6)) {
 676                 r = uev_remove_path(sysdev, vecs);
 677                 goto out;
 678         }
 679
 680 out:
 681         unlock(vecs->lock);
 682         return r;
 683 }
 684
 685 static void *
 686 ueventloop (void * ap)
 687 {
 688         if (uevent_listen(&uev_trigger, ap))
 689                 fprintf(stderr, "error starting uevent listener");
 690
 691         return NULL;
 692 }
 693
 694 static void *
 695 uxlsnrloop (void * ap)
 696 {
 697         if (cli_init())
 698                 return NULL;
 699
 700         set_handler_callback(LIST+PATHS, cli_list_paths);
 701         set_handler_callback(LIST+MAPS, cli_list_maps);
 702         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 703         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 704         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 705         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 706         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 707         set_handler_callback(LIST+CONFIG, cli_list_config);
 708         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 709         set_handler_callback(LIST+DEVICES, cli_list_devices);
 710         set_handler_callback(ADD+PATH, cli_add_path);
 711         set_handler_callback(DEL+PATH, cli_del_path);
 712         set_handler_callback(ADD+MAP, cli_add_map);
 713         set_handler_callback(DEL+MAP, cli_del_map);
 714         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 715         set_handler_callback(RECONFIGURE, cli_reconfigure);
 716         set_handler_callback(SUSPEND+MAP, cli_suspend);
 717         set_handler_callback(RESUME+MAP, cli_resume);
 718         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 719         set_handler_callback(FAIL+PATH, cli_fail);
 720
 721         uxsock_listen(&uxsock_trigger, ap);
 722
 723         return NULL;
 724 }
 725
 726 static int
 727 exit_daemon (int status)
 728 {
 729         if (status != 0)
 730                 fprintf(stderr, "bad exit status. see daemon.log\n");
 731
 732         condlog(3, "unlink pidfile");
 733         unlink(DEFAULT_PIDFILE);
 734
 735         lock(&exit_mutex);
 736         pthread_cond_signal(&exit_cond);
 737         unlock(&exit_mutex);
 738
 739         return status;
 740 }
 741
 742 static void
 743 fail_path (struct path * pp, int del_active)
 744 {
 745         if (!pp->mpp)
 746                 return;
 747
 748         condlog(2, "checker failed path %s in map %s",
 749                  pp->dev_t, pp->mpp->alias);
 750
 751         dm_fail_path(pp->mpp->alias, pp->dev_t);
 752         if (del_active)
 753                 update_queue_mode_del_path(pp->mpp);
 754 }
 755
 756 /*
 757  * caller must have locked the path list before calling that function
 758  */
 759 static void
 760 reinstate_path (struct path * pp, int add_active)
 761 {
 762         if (!pp->mpp)
 763                 return;
 764
 765         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 766                 condlog(0, "%s: reinstate failed", pp->dev_t);
 767         else {
 768                 condlog(2, "%s: reinstated", pp->dev_t);
 769                 if (add_active)
 770                         update_queue_mode_add_path(pp->mpp);
 771         }
 772 }
 773
 774 static void
 775 enable_group(struct path * pp)
 776 {
 777         struct pathgroup * pgp;
 778
 779         /*
 780          * if path is added through uev_add_path, pgindex can be unset.
 781          * next update_strings() will set it, upon map reload event.
 782          *
 783          * we can safely return here, because upon map reload, all
 784          * PG will be enabled.
 785          */
 786         if (!pp->mpp->pg || !pp->pgindex)
 787                 return;
 788
 789         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 790
 791         if (pgp->status == PGSTATE_DISABLED) {
 792                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 793                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 794         }
 795 }
 796
 797 static void
 798 mpvec_garbage_collector (struct vectors * vecs)
 799 {
 800         struct multipath * mpp;
 801         unsigned int i;
 802
 803         vector_foreach_slot (vecs->mpvec, mpp, i) {
 804                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 805                         condlog(2, "%s: remove dead map", mpp->alias);
 806                         remove_map(mpp, vecs, stop_waiter_thread, 1);
 807                         i--;
 808                 }
 809         }
 810 }
 811
 812 static void
 813 defered_failback_tick (vector mpvec)
 814 {
 815         struct multipath * mpp;
 816         unsigned int i;
 817
 818         vector_foreach_slot (mpvec, mpp, i) {
 819                 /*
 820                  * defered failback getting sooner
 821                  */
 822                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 823                         mpp->failback_tick--;
 824
 825                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 826                                 switch_pathgroup(mpp);
 827                 }
 828         }
 829 }
 830
 831 static void
 832 retry_count_tick(vector mpvec)
 833 {
 834         struct multipath *mpp;
 835         unsigned int i;
 836
 837         vector_foreach_slot (mpvec, mpp, i) {
 838                 if (mpp->retry_tick) {
 839                         mpp->stat_total_queueing_time++;
 840                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 841                         if(--mpp->retry_tick == 0) {
 842                                 dm_queue_if_no_path(mpp->alias, 0);
 843                                 condlog(2, "%s: Disable queueing", mpp->alias);
 844                         }
 845                 }
 846         }
 847 }
 848
 849 static void *
 850 checkerloop (void *ap)
 851 {
 852         struct vectors *vecs;
 853         struct path *pp;
 854         int count = 0;
 855         int newstate;
 856         unsigned int i;
 857
 858         mlockall(MCL_CURRENT | MCL_FUTURE);
 859         vecs = (struct vectors *)ap;
 860         condlog(2, "path checkers start up");
 861
 862         /*
 863          * init the path check interval
 864          */
 865         vector_foreach_slot (vecs->pathvec, pp, i) {
 866                 pp->checkint = conf->checkint;
 867         }
 868
 869         while (1) {
 870                 pthread_cleanup_push(cleanup_lock, vecs->lock);
 871                 lock(vecs->lock);
 872                 condlog(4, "tick");
 873
 874                 vector_foreach_slot (vecs->pathvec, pp, i) {
 875                         if (!pp->mpp)
 876                                 continue;
 877
 878                         if (pp->tick && --pp->tick)
 879                                 continue; /* don't check this path yet */
 880
 881                         /*
 882                          * provision a next check soonest,
 883                          * in case we exit abnormaly from here
 884                          */
 885                         pp->tick = conf->checkint;
 886
 887                         if (!checker_selected(&pp->checker)) {
 888                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 889                                 select_checker(pp);
 890                         }
 891
 892                         if (!checker_selected(&pp->checker)) {
 893                                 condlog(0, "%s: checker is not set", pp->dev);
 894                                 continue;
 895                         }
 896                         newstate = checker_check(&pp->checker);
 897
 898                         if (newstate < 0) {
 899                                 condlog(2, "%s: unusable path", pp->dev);
 900                                 pathinfo(pp, conf->hwtable, 0);
 901                                 continue;
 902                         }
 903
 904                         if (newstate != pp->state) {
 905                                 int oldstate = pp->state;
 906                                 pp->state = newstate;
 907                                 LOG_MSG(1, checker_message(&pp->checker));
 908
 909                                 /*
 910                                  * upon state change, reset the checkint
 911                                  * to the shortest delay
 912                                  */
 913                                 pp->checkint = conf->checkint;
 914
 915                                 if (newstate == PATH_DOWN ||
 916                                     newstate == PATH_SHAKY ||
 917                                     update_multipath_strings(pp->mpp,
 918                                                              vecs->pathvec)) {
 919                                         /*
 920                                          * proactively fail path in the DM
 921                                          */
 922                                         if (oldstate == PATH_UP ||
 923                                             oldstate == PATH_GHOST)
 924                                                 fail_path(pp, 1);
 925                                         else
 926                                                 fail_path(pp, 0);
 927
 928                                         /*
 929                                          * cancel scheduled failback
 930                                          */
 931                                         pp->mpp->failback_tick = 0;
 932
 933                                         pp->mpp->stat_path_failures++;
 934                                         continue;
 935                                 }
 936
 937                                 /*
 938                                  * reinstate this path
 939                                  */
 940                                 if (oldstate != PATH_UP &&
 941                                     oldstate != PATH_GHOST)
 942                                         reinstate_path(pp, 1);
 943                                 else
 944                                         reinstate_path(pp, 0);
 945
 946                                 /*
 947                                  * schedule [defered] failback
 948                                  */
 949                                 if (pp->mpp->pgfailback > 0)
 950                                         pp->mpp->failback_tick =
 951                                                 pp->mpp->pgfailback + 1;
 952                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 953                                     need_switch_pathgroup(pp->mpp, 1))
 954                                         switch_pathgroup(pp->mpp);
 955
 956                                 /*
 957                                  * if at least one path is up in a group, and
 958                                  * the group is disabled, re-enable it
 959                                  */
 960                                 if (newstate == PATH_UP)
 961                                         enable_group(pp);
 962                         }
 963                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 964                                 LOG_MSG(4, checker_message(&pp->checker));
 965                                 /*
 966                                  * double the next check delay.
 967                                  * max at conf->max_checkint
 968                                  */
 969                                 if (pp->checkint < (conf->max_checkint / 2))
 970                                         pp->checkint = 2 * pp->checkint;
 971                                 else
 972                                         pp->checkint = conf->max_checkint;
 973
 974                                 pp->tick = pp->checkint;
 975                                 condlog(4, "%s: delay next check %is",
 976                                                 pp->dev_t, pp->tick);
 977                         }
 978                         else if (newstate == PATH_DOWN)
 979                                 LOG_MSG(2, checker_message(&pp->checker));
 980
 981                         pp->state = newstate;
 982
 983                         /*
 984                          * path prio refreshing
 985                          */
 986                         condlog(4, "path prio refresh");
 987                         pathinfo(pp, conf->hwtable, DI_PRIO);
 988
 989                         if (need_switch_pathgroup(pp->mpp, 0)) {
 990                                 if (pp->mpp->pgfailback > 0 &&
 991                                     pp->mpp->failback_tick <= 0)
 992                                         pp->mpp->failback_tick =
 993                                                 pp->mpp->pgfailback + 1;
 994                                 else if (pp->mpp->pgfailback ==
 995                                                 -FAILBACK_IMMEDIATE)
 996                                         switch_pathgroup(pp->mpp);
 997                         }
 998                 }
 999                 defered_failback_tick(vecs->mpvec);
1000                 retry_count_tick(vecs->mpvec);
1001
1002                 if (count)
1003                         count--;
1004                 else {
1005                         condlog(4, "map garbage collection");
1006                         mpvec_garbage_collector(vecs);
1007                         count = MAPGCINT;
1008                 }
1009
1010                 lock_cleanup_pop(vecs->lock);
1011                 sleep(1);
1012         }
1013         return NULL;
1014 }
1015
1016 int
1017 configure (struct vectors * vecs, int start_waiters)
1018 {
1019         struct multipath * mpp;
1020         struct path * pp;
1021         vector mpvec;
1022         int i;
1023
1024         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1025                 return 1;
1026
1027         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1028                 return 1;
1029
1030         if (!(mpvec = vector_alloc()))
1031                 return 1;
1032
1033         /*
1034          * probe for current path (from sysfs) and map (from dm) sets
1035          */
1036         path_discovery(vecs->pathvec, conf, DI_ALL);
1037
1038         vector_foreach_slot (vecs->pathvec, pp, i){
1039                 if (filter_path(conf, pp)){
1040                         vector_del_slot(vecs->pathvec, i);
1041                         free_path(pp);
1042                         i--;
1043                 }
1044                 else
1045                         pp->checkint = conf->checkint;
1046         }
1047         if (map_discovery(vecs))
1048                 return 1;
1049
1050         /*
1051          * create new set of maps & push changed ones into dm
1052          */
1053         if (coalesce_paths(vecs, mpvec, NULL))
1054                 return 1;
1055
1056         /*
1057          * may need to remove some maps which are no longer relevant
1058          * e.g., due to blacklist changes in conf file
1059          */
1060         if (coalesce_maps(vecs, mpvec))
1061                 return 1;
1062
1063         dm_lib_release();
1064
1065         sync_maps_state(mpvec);
1066
1067         /*
1068          * purge dm of old maps
1069          */
1070         remove_maps(vecs, NULL);
1071
1072         /*
1073          * save new set of maps formed by considering current path state
1074          */
1075         vector_free(vecs->mpvec);
1076         vecs->mpvec = mpvec;
1077
1078         /*
1079          * start dm event waiter threads for these new maps
1080          */
1081         vector_foreach_slot(vecs->mpvec, mpp, i) {
1082                 if (setup_multipath(vecs, mpp))
1083                         return 1;
1084                 if (start_waiters)
1085                         if (start_waiter_thread(mpp, vecs))
1086                                 return 1;
1087         }
1088         return 0;
1089 }
1090
1091 int
1092 reconfigure (struct vectors * vecs)
1093 {
1094         struct config * old = conf;
1095
1096         /*
1097          * free old map and path vectors ... they use old conf state
1098          */
1099         if (VECTOR_SIZE(vecs->mpvec))
1100                 remove_maps(vecs, stop_waiter_thread);
1101
1102         if (VECTOR_SIZE(vecs->pathvec))
1103                 free_pathvec(vecs->pathvec, FREE_PATHS);
1104
1105         vecs->pathvec = NULL;
1106         conf = NULL;
1107
1108         if (load_config(DEFAULT_CONFIGFILE))
1109                 return 1;
1110
1111         conf->verbosity = old->verbosity;
1112
1113         if (!conf->checkint) {
1114                 conf->checkint = DEFAULT_CHECKINT;
1115                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1116         }
1117         configure(vecs, 1);
1118         free_config(old);
1119         return 0;
1120 }
1121
1122 static struct vectors *
1123 init_vecs (void)
1124 {
1125         struct vectors * vecs;
1126
1127         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1128
1129         if (!vecs)
1130                 return NULL;
1131
1132         vecs->lock =
1133                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1134
1135         if (!vecs->lock)
1136                 goto out;
1137
1138         pthread_mutex_init(vecs->lock, NULL);
1139
1140         return vecs;
1141
1142 out:
1143         FREE(vecs);
1144         condlog(0, "failed to init paths");
1145         return NULL;
1146 }
1147
1148 static void *
1149 signal_set(int signo, void (*func) (int))
1150 {
1151         int r;
1152         struct sigaction sig;
1153         struct sigaction osig;
1154
1155         sig.sa_handler = func;
1156         sigemptyset(&sig.sa_mask);
1157         sig.sa_flags = 0;
1158
1159         r = sigaction(signo, &sig, &osig);
1160
1161         if (r < 0)
1162                 return (SIG_ERR);
1163         else
1164                 return (osig.sa_handler);
1165 }
1166
1167 static void
1168 sighup (int sig)
1169 {
1170         condlog(2, "reconfigure (SIGHUP)");
1171
1172         lock(gvecs->lock);
1173         reconfigure(gvecs);
1174         unlock(gvecs->lock);
1175
1176 #ifdef _DEBUG_
1177         dbg_free_final(NULL);
1178 #endif
1179 }
1180
1181 static void
1182 sigend (int sig)
1183 {
1184         exit_daemon(0);
1185 }
1186
1187 static void
1188 sigusr1 (int sig)
1189 {
1190         condlog(3, "SIGUSR1 received");
1191 }
1192
1193 static void
1194 signal_init(void)
1195 {
1196         signal_set(SIGHUP, sighup);
1197         signal_set(SIGUSR1, sigusr1);
1198         signal_set(SIGINT, sigend);
1199         signal_set(SIGTERM, sigend);
1200         signal(SIGPIPE, SIG_IGN);
1201 }
1202
1203 static void
1204 setscheduler (void)
1205 {
1206         int res;
1207         static struct sched_param sched_param = {
1208                 .sched_priority = 99
1209         };
1210
1211         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1212
1213         if (res == -1)
1214                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1215         return;
1216 }
1217
1218 static void
1219 set_oom_adj (int val)
1220 {
1221         FILE *fp;
1222
1223         fp = fopen("/proc/self/oom_adj", "w");
1224
1225         if (!fp)
1226                 return;
1227
1228         fprintf(fp, "%i", val);
1229         fclose(fp);
1230 }
1231
1232 static int
1233 child (void * param)
1234 {
1235         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1236         pthread_attr_t attr;
1237         struct vectors * vecs;
1238
1239         mlockall(MCL_CURRENT | MCL_FUTURE);
1240
1241         if (logsink)
1242                 log_thread_start();
1243
1244         condlog(2, "--------start up--------");
1245         condlog(2, "read " DEFAULT_CONFIGFILE);
1246
1247         if (load_config(DEFAULT_CONFIGFILE))
1248                 exit(1);
1249
1250         setlogmask(LOG_UPTO(conf->verbosity + 3));
1251
1252         /*
1253          * fill the voids left in the config file
1254          */
1255         if (!conf->checkint) {
1256                 conf->checkint = DEFAULT_CHECKINT;
1257                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1258         }
1259
1260         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1261                 if (logsink)
1262                         log_thread_stop();
1263
1264                 exit(1);
1265         }
1266         signal_init();
1267         setscheduler();
1268         set_oom_adj(-16);
1269         vecs = gvecs = init_vecs();
1270
1271         if (!vecs)
1272                 exit(1);
1273
1274         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1275                 condlog(0, "can not find sysfs mount point");
1276                 exit(1);
1277         }
1278
1279         /*
1280          * fetch and configure both paths and multipaths
1281          */
1282         if (configure(vecs, 1)) {
1283                 condlog(0, "failure during configuration");
1284                 exit(1);
1285         }
1286
1287         /*
1288          * start threads
1289          */
1290         pthread_attr_init(&attr);
1291         pthread_attr_setstacksize(&attr, 64 * 1024);
1292         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1293
1294         pthread_create(&check_thr, &attr, checkerloop, vecs);
1295         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1296         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1297
1298         pthread_cond_wait(&exit_cond, &exit_mutex);
1299
1300         /*
1301          * exit path
1302          */
1303         lock(vecs->lock);
1304         remove_maps(vecs, stop_waiter_thread);
1305         free_pathvec(vecs->pathvec, FREE_PATHS);
1306
1307         pthread_cancel(check_thr);
1308         pthread_cancel(uevent_thr);
1309         pthread_cancel(uxlsnr_thr);
1310
1311         sysfs_cleanup();
1312
1313         free_keys(keys);
1314         keys = NULL;
1315         free_handlers(handlers);
1316         handlers = NULL;
1317         free_polls();
1318
1319         unlock(vecs->lock);
1320         pthread_mutex_destroy(vecs->lock);
1321         FREE(vecs->lock);
1322         vecs->lock = NULL;
1323         FREE(vecs);
1324         vecs = NULL;
1325         free_config(conf);
1326         conf = NULL;
1327
1328         condlog(2, "--------shut down-------");
1329
1330         if (logsink)
1331                 log_thread_stop();
1332
1333         dm_lib_release();
1334         dm_lib_exit();
1335
1336 #ifdef _DEBUG_
1337         dbg_free_final(NULL);
1338 #endif
1339
1340         exit(0);
1341 }
1342
1343 static int
1344 daemonize(void)
1345 {
1346         int pid;
1347         int in_fd, out_fd;
1348
1349         if( (pid = fork()) < 0){
1350                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1351                 return -1;
1352         }
1353         else if (pid != 0)
1354                 return pid;
1355
1356         setsid();
1357
1358         if ( (pid = fork()) < 0)
1359                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1360         else if (pid != 0)
1361                 _exit(0);
1362
1363         in_fd = open("/dev/null", O_RDONLY);
1364         if (in_fd < 0){
1365                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1366                         strerror(errno));
1367                 _exit(0);
1368         }
1369         out_fd = open("/dev/console", O_WRONLY);
1370         if (out_fd < 0){
1371                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1372                         strerror(errno));
1373                 _exit(0);
1374         }
1375
1376         close(STDIN_FILENO);
1377         dup(in_fd);
1378         close(STDOUT_FILENO);
1379         dup(out_fd);
1380         close(STDERR_FILENO);
1381         dup(out_fd);
1382
1383         close(in_fd);
1384         close(out_fd);
1385         chdir("/");
1386         umask(0);
1387         return 0;
1388 }
1389
1390 int
1391 main (int argc, char *argv[])
1392 {
1393         extern char *optarg;
1394         extern int optind;
1395         int arg;
1396         int err;
1397
1398         logsink = 1;
1399         dm_init();
1400
1401         if (getuid() != 0) {
1402                 fprintf(stderr, "need to be root\n");
1403                 exit(1);
1404         }
1405
1406         /* make sure we don't lock any path */
1407         chdir("/");
1408         umask(umask(077) | 022);
1409
1410         conf = alloc_config();
1411
1412         if (!conf)
1413                 exit(1);
1414
1415         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1416         switch(arg) {
1417                 case 'd':
1418                         logsink = 0;
1419                         //debug=1; /* ### comment me out ### */
1420                         break;
1421                 case 'v':
1422                         if (sizeof(optarg) > sizeof(char *) ||
1423                             !isdigit(optarg[0]))
1424                                 exit(1);
1425
1426                         conf->verbosity = atoi(optarg);
1427                         break;
1428                 case 'k':
1429                         uxclnt(optarg);
1430                         exit(0);
1431                 default:
1432                         ;
1433                 }
1434         }
1435
1436         if (!logsink)
1437                 err = 0;
1438         else
1439                 err = daemonize();
1440
1441         if (err < 0)
1442                 /* error */
1443                 exit(1);
1444         else if (err > 0)
1445                 /* parent dies */
1446                 exit(0);
1447         else
1448                 /* child lives */
1449                 return (child(NULL));
1450 }
1451