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         int retval;
 446
 447         condlog(2, "%s: remove path (uevent)", dev->kernel);
 448         retval = ev_remove_path(dev->kernel, vecs);
 449         if (!retval)
 450                 sysfs_device_put(dev);
 451
 452         return retval;
 453 }
 454
 455 int
 456 ev_remove_path (char * devname, struct vectors * vecs)
 457 {
 458         struct multipath * mpp;
 459         struct path * pp;
 460         int i;
 461         int rm_path = 1;
 462
 463         pp = find_path_by_dev(vecs->pathvec, devname);
 464
 465         if (!pp) {
 466                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 467                 return 1;
 468         }
 469
 470         /*
 471          * avoid referring to the map of an orphanned path
 472          */
 473         if ((mpp = pp->mpp)) {
 474
 475                 /*
 476                  * remove the map IFF removing the last path
 477                  */
 478                 if (pathcount(mpp, PATH_WILD) > 1) {
 479                         vector rpvec = vector_alloc();
 480
 481                         /*
 482                          * transform the mp->pg vector of vectors of paths
 483                          * into a mp->params string to feed the device-mapper
 484                          */
 485                         update_mpp_paths(mpp, vecs->pathvec);
 486                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 487                                 vector_del_slot(mpp->paths, i);
 488
 489                         if (VECTOR_SIZE(mpp->paths) == 0) {
 490                                 char alias[WWID_SIZE];
 491
 492                                 /*
 493                                  * flush_map will fail if the device is open
 494                                  */
 495                                 strncpy(alias, mpp->alias, WWID_SIZE);
 496                                 if (flush_map(mpp, vecs))
 497                                         rm_path = 0;
 498                                 else
 499                                         condlog(3, "%s: removed map after removing"
 500                                                 " multiple paths", alias);
 501                         }
 502                         else {
 503                                 if (setup_map(mpp)) {
 504                                         condlog(0, "%s: failed to setup map for"
 505                                                 " removal of path %s", mpp->alias, devname);
 506                                         free_pathvec(rpvec, KEEP_PATHS);
 507                                         goto out;
 508                                 }
 509                                 /*
 510                                  * reload the map
 511                                  */
 512                                 mpp->action = ACT_RELOAD;
 513                                 if (domap(mpp) <= 0) {
 514                                         condlog(0, "%s: failed in domap for "
 515                                                 "removal of path %s",
 516                                                 mpp->alias, devname);
 517                                         /*
 518                                          * Delete path from pathvec so that
 519                                          * update_mpp_paths wont find it later
 520                                          * when/if another path is removed.
 521                                          */
 522                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 523                                                 vector_del_slot(vecs->pathvec, i);
 524                                         free_path(pp);
 525                                         return 1;
 526                                 }
 527                                 /*
 528                                  * update our state from kernel
 529                                  */
 530                                 if (setup_multipath(vecs, mpp)) {
 531                                         free_pathvec(rpvec, KEEP_PATHS);
 532                                         goto out;
 533                                 }
 534                                 sync_map_state(mpp);
 535
 536                                 condlog(3, "%s: path removed from map %s",
 537                                         devname, mpp->alias);
 538                         }
 539                         free_pathvec(rpvec, KEEP_PATHS);
 540                 }
 541                 else {
 542                         char alias[WWID_SIZE];
 543
 544                         /*
 545                          * flush_map will fail if the device is open
 546                          */
 547                         strncpy(alias, mpp->alias, WWID_SIZE);
 548                         if (flush_map(mpp, vecs))
 549                                 rm_path = 0;
 550                         else
 551                                 condlog(3, "%s: removed map", alias);
 552                 }
 553         }
 554
 555         if (rm_path) {
 556                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 557                         vector_del_slot(vecs->pathvec, i);
 558                 free_path(pp);
 559         }
 560
 561         return 0;
 562
 563 out:
 564         remove_map(mpp, vecs, stop_waiter_thread, 1);
 565         return 1;
 566 }
 567
 568 static int
 569 map_discovery (struct vectors * vecs)
 570 {
 571         struct multipath * mpp;
 572         unsigned int i;
 573
 574         if (dm_get_maps(vecs->mpvec, "multipath"))
 575                 return 1;
 576
 577         vector_foreach_slot (vecs->mpvec, mpp, i)
 578                 if (setup_multipath(vecs, mpp))
 579                         return 1;
 580
 581         return 0;
 582 }
 583
 584 int
 585 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 586 {
 587         struct vectors * vecs;
 588         int r;
 589
 590         *reply = NULL;
 591         *len = 0;
 592         vecs = (struct vectors *)trigger_data;
 593
 594         pthread_cleanup_push(cleanup_lock, vecs->lock);
 595         lock(vecs->lock);
 596
 597         r = parse_cmd(str, reply, len, vecs);
 598
 599         if (r > 0) {
 600                 *reply = STRDUP("fail\n");
 601                 *len = strlen(*reply) + 1;
 602                 r = 1;
 603         }
 604         else if (!r && *len == 0) {
 605                 *reply = STRDUP("ok\n");
 606                 *len = strlen(*reply) + 1;
 607                 r = 0;
 608         }
 609         /* else if (r < 0) leave *reply alone */
 610
 611         lock_cleanup_pop(vecs->lock);
 612         return r;
 613 }
 614
 615 static int
 616 uev_discard(char * devpath)
 617 {
 618         char a[10], b[10];
 619
 620         /*
 621          * keep only block devices, discard partitions
 622          */
 623         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 624             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 625                 condlog(4, "discard event on %s", devpath);
 626                 return 1;
 627         }
 628         return 0;
 629 }
 630
 631 int
 632 uev_trigger (struct uevent * uev, void * trigger_data)
 633 {
 634         int r = 0;
 635         struct sysfs_device *sysdev;
 636         struct vectors * vecs;
 637
 638         vecs = (struct vectors *)trigger_data;
 639
 640         if (uev_discard(uev->devpath))
 641                 return 0;
 642
 643         sysdev = sysfs_device_get(uev->devpath);
 644         if(!sysdev)
 645                 return 0;
 646
 647         lock(vecs->lock);
 648
 649         /*
 650          * device map event
 651          * Add events are ignored here as the tables
 652          * are not fully initialised then.
 653          */
 654         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 655                 if (!strncmp(uev->action, "change", 6)) {
 656                         r = uev_add_map(sysdev, vecs);
 657                         goto out;
 658                 }
 659                 if (!strncmp(uev->action, "remove", 6)) {
 660                         r = uev_remove_map(sysdev, vecs);
 661                         goto out;
 662                 }
 663                 if (!strncmp(uev->action, "umount", 6)) {
 664                         r = uev_umount_map(sysdev, vecs);
 665                         goto out;
 666                 }
 667                 goto out;
 668         }
 669
 670         /*
 671          * path add/remove event
 672          */
 673         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 674                            sysdev->kernel) > 0)
 675                 goto out;
 676
 677         if (!strncmp(uev->action, "add", 3)) {
 678                 r = uev_add_path(sysdev, vecs);
 679                 goto out;
 680         }
 681         if (!strncmp(uev->action, "remove", 6)) {
 682                 r = uev_remove_path(sysdev, vecs);
 683                 goto out;
 684         }
 685
 686 out:
 687         unlock(vecs->lock);
 688         return r;
 689 }
 690
 691 static void *
 692 ueventloop (void * ap)
 693 {
 694         if (uevent_listen(&uev_trigger, ap))
 695                 fprintf(stderr, "error starting uevent listener");
 696
 697         return NULL;
 698 }
 699
 700 static void *
 701 uxlsnrloop (void * ap)
 702 {
 703         if (cli_init())
 704                 return NULL;
 705
 706         set_handler_callback(LIST+PATHS, cli_list_paths);
 707         set_handler_callback(LIST+MAPS, cli_list_maps);
 708         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 709         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 710         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 711         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 712         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 713         set_handler_callback(LIST+CONFIG, cli_list_config);
 714         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 715         set_handler_callback(LIST+DEVICES, cli_list_devices);
 716         set_handler_callback(ADD+PATH, cli_add_path);
 717         set_handler_callback(DEL+PATH, cli_del_path);
 718         set_handler_callback(ADD+MAP, cli_add_map);
 719         set_handler_callback(DEL+MAP, cli_del_map);
 720         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 721         set_handler_callback(RECONFIGURE, cli_reconfigure);
 722         set_handler_callback(SUSPEND+MAP, cli_suspend);
 723         set_handler_callback(RESUME+MAP, cli_resume);
 724         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 725         set_handler_callback(FAIL+PATH, cli_fail);
 726
 727         uxsock_listen(&uxsock_trigger, ap);
 728
 729         return NULL;
 730 }
 731
 732 static int
 733 exit_daemon (int status)
 734 {
 735         if (status != 0)
 736                 fprintf(stderr, "bad exit status. see daemon.log\n");
 737
 738         condlog(3, "unlink pidfile");
 739         unlink(DEFAULT_PIDFILE);
 740
 741         lock(&exit_mutex);
 742         pthread_cond_signal(&exit_cond);
 743         unlock(&exit_mutex);
 744
 745         return status;
 746 }
 747
 748 static void
 749 fail_path (struct path * pp, int del_active)
 750 {
 751         if (!pp->mpp)
 752                 return;
 753
 754         condlog(2, "checker failed path %s in map %s",
 755                  pp->dev_t, pp->mpp->alias);
 756
 757         dm_fail_path(pp->mpp->alias, pp->dev_t);
 758         if (del_active)
 759                 update_queue_mode_del_path(pp->mpp);
 760 }
 761
 762 /*
 763  * caller must have locked the path list before calling that function
 764  */
 765 static void
 766 reinstate_path (struct path * pp, int add_active)
 767 {
 768         if (!pp->mpp)
 769                 return;
 770
 771         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 772                 condlog(0, "%s: reinstate failed", pp->dev_t);
 773         else {
 774                 condlog(2, "%s: reinstated", pp->dev_t);
 775                 if (add_active)
 776                         update_queue_mode_add_path(pp->mpp);
 777         }
 778 }
 779
 780 static void
 781 enable_group(struct path * pp)
 782 {
 783         struct pathgroup * pgp;
 784
 785         /*
 786          * if path is added through uev_add_path, pgindex can be unset.
 787          * next update_strings() will set it, upon map reload event.
 788          *
 789          * we can safely return here, because upon map reload, all
 790          * PG will be enabled.
 791          */
 792         if (!pp->mpp->pg || !pp->pgindex)
 793                 return;
 794
 795         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 796
 797         if (pgp->status == PGSTATE_DISABLED) {
 798                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 799                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 800         }
 801 }
 802
 803 static void
 804 mpvec_garbage_collector (struct vectors * vecs)
 805 {
 806         struct multipath * mpp;
 807         unsigned int i;
 808
 809         vector_foreach_slot (vecs->mpvec, mpp, i) {
 810                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 811                         condlog(2, "%s: remove dead map", mpp->alias);
 812                         remove_map(mpp, vecs, stop_waiter_thread, 1);
 813                         i--;
 814                 }
 815         }
 816 }
 817
 818 static void
 819 defered_failback_tick (vector mpvec)
 820 {
 821         struct multipath * mpp;
 822         unsigned int i;
 823
 824         vector_foreach_slot (mpvec, mpp, i) {
 825                 /*
 826                  * defered failback getting sooner
 827                  */
 828                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 829                         mpp->failback_tick--;
 830
 831                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 832                                 switch_pathgroup(mpp);
 833                 }
 834         }
 835 }
 836
 837 static void
 838 retry_count_tick(vector mpvec)
 839 {
 840         struct multipath *mpp;
 841         unsigned int i;
 842
 843         vector_foreach_slot (mpvec, mpp, i) {
 844                 if (mpp->retry_tick) {
 845                         mpp->stat_total_queueing_time++;
 846                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 847                         if(--mpp->retry_tick == 0) {
 848                                 dm_queue_if_no_path(mpp->alias, 0);
 849                                 condlog(2, "%s: Disable queueing", mpp->alias);
 850                         }
 851                 }
 852         }
 853 }
 854
 855 static void *
 856 checkerloop (void *ap)
 857 {
 858         struct vectors *vecs;
 859         struct path *pp;
 860         int count = 0;
 861         int newstate;
 862         unsigned int i;
 863
 864         mlockall(MCL_CURRENT | MCL_FUTURE);
 865         vecs = (struct vectors *)ap;
 866         condlog(2, "path checkers start up");
 867
 868         /*
 869          * init the path check interval
 870          */
 871         vector_foreach_slot (vecs->pathvec, pp, i) {
 872                 pp->checkint = conf->checkint;
 873         }
 874
 875         while (1) {
 876                 pthread_cleanup_push(cleanup_lock, vecs->lock);
 877                 lock(vecs->lock);
 878                 condlog(4, "tick");
 879
 880                 vector_foreach_slot (vecs->pathvec, pp, i) {
 881                         if (!pp->mpp)
 882                                 continue;
 883
 884                         if (pp->tick && --pp->tick)
 885                                 continue; /* don't check this path yet */
 886
 887                         /*
 888                          * provision a next check soonest,
 889                          * in case we exit abnormaly from here
 890                          */
 891                         pp->tick = conf->checkint;
 892
 893                         if (!checker_selected(&pp->checker)) {
 894                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 895                                 select_checker(pp);
 896                         }
 897                         if (!checker_selected(&pp->checker)) {
 898                                 condlog(0, "%s: checker is not set", pp->dev);
 899                                 continue;
 900                         }
 901                         /*
 902                          * Set checker in async mode.
 903                          * Honored only by checker implementing the said mode.
 904                          */
 905                         checker_set_async(&pp->checker);
 906
 907                         newstate = checker_check(&pp->checker);
 908
 909                         if (newstate < 0) {
 910                                 condlog(2, "%s: unusable path", pp->dev);
 911                                 pathinfo(pp, conf->hwtable, 0);
 912                                 continue;
 913                         }
 914                         /*
 915                          * Async IO in flight. Keep the previous path state
 916                          * and reschedule as soon as possible
 917                          */
 918                         if (newstate == PATH_PENDING) {
 919                                 pp->tick = 1;
 920                                 continue;
 921                         }
 922                         if (newstate != pp->state) {
 923                                 int oldstate = pp->state;
 924                                 pp->state = newstate;
 925                                 LOG_MSG(1, checker_message(&pp->checker));
 926
 927                                 /*
 928                                  * upon state change, reset the checkint
 929                                  * to the shortest delay
 930                                  */
 931                                 pp->checkint = conf->checkint;
 932
 933                                 if (newstate == PATH_DOWN ||
 934                                     newstate == PATH_SHAKY ||
 935                                     update_multipath_strings(pp->mpp,
 936                                                              vecs->pathvec)) {
 937                                         /*
 938                                          * proactively fail path in the DM
 939                                          */
 940                                         if (oldstate == PATH_UP ||
 941                                             oldstate == PATH_GHOST)
 942                                                 fail_path(pp, 1);
 943                                         else
 944                                                 fail_path(pp, 0);
 945
 946                                         /*
 947                                          * cancel scheduled failback
 948                                          */
 949                                         pp->mpp->failback_tick = 0;
 950
 951                                         pp->mpp->stat_path_failures++;
 952                                         continue;
 953                                 }
 954
 955                                 /*
 956                                  * reinstate this path
 957                                  */
 958                                 if (oldstate != PATH_UP &&
 959                                     oldstate != PATH_GHOST)
 960                                         reinstate_path(pp, 1);
 961                                 else
 962                                         reinstate_path(pp, 0);
 963
 964                                 /*
 965                                  * schedule [defered] failback
 966                                  */
 967                                 if (pp->mpp->pgfailback > 0)
 968                                         pp->mpp->failback_tick =
 969                                                 pp->mpp->pgfailback + 1;
 970                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 971                                     need_switch_pathgroup(pp->mpp, 1))
 972                                         switch_pathgroup(pp->mpp);
 973
 974                                 /*
 975                                  * if at least one path is up in a group, and
 976                                  * the group is disabled, re-enable it
 977                                  */
 978                                 if (newstate == PATH_UP)
 979                                         enable_group(pp);
 980                         }
 981                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 982                                 LOG_MSG(4, checker_message(&pp->checker));
 983                                 /*
 984                                  * double the next check delay.
 985                                  * max at conf->max_checkint
 986                                  */
 987                                 if (pp->checkint < (conf->max_checkint / 2))
 988                                         pp->checkint = 2 * pp->checkint;
 989                                 else
 990                                         pp->checkint = conf->max_checkint;
 991
 992                                 pp->tick = pp->checkint;
 993                                 condlog(4, "%s: delay next check %is",
 994                                                 pp->dev_t, pp->tick);
 995                         }
 996                         else if (newstate == PATH_DOWN)
 997                                 LOG_MSG(2, checker_message(&pp->checker));
 998
 999                         pp->state = newstate;
1000
1001                         /*
1002                          * path prio refreshing
1003                          */
1004                         condlog(4, "path prio refresh");
1005                         pathinfo(pp, conf->hwtable, DI_PRIO);
1006
1007                         if (need_switch_pathgroup(pp->mpp, 0)) {
1008                                 if (pp->mpp->pgfailback > 0 &&
1009                                     pp->mpp->failback_tick <= 0)
1010                                         pp->mpp->failback_tick =
1011                                                 pp->mpp->pgfailback + 1;
1012                                 else if (pp->mpp->pgfailback ==
1013                                                 -FAILBACK_IMMEDIATE)
1014                                         switch_pathgroup(pp->mpp);
1015                         }
1016                 }
1017                 defered_failback_tick(vecs->mpvec);
1018                 retry_count_tick(vecs->mpvec);
1019
1020                 if (count)
1021                         count--;
1022                 else {
1023                         condlog(4, "map garbage collection");
1024                         mpvec_garbage_collector(vecs);
1025                         count = MAPGCINT;
1026                 }
1027
1028                 lock_cleanup_pop(vecs->lock);
1029                 sleep(1);
1030         }
1031         return NULL;
1032 }
1033
1034 int
1035 configure (struct vectors * vecs, int start_waiters)
1036 {
1037         struct multipath * mpp;
1038         struct path * pp;
1039         vector mpvec;
1040         int i;
1041
1042         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1043                 return 1;
1044
1045         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1046                 return 1;
1047
1048         if (!(mpvec = vector_alloc()))
1049                 return 1;
1050
1051         /*
1052          * probe for current path (from sysfs) and map (from dm) sets
1053          */
1054         path_discovery(vecs->pathvec, conf, DI_ALL);
1055
1056         vector_foreach_slot (vecs->pathvec, pp, i){
1057                 if (filter_path(conf, pp)){
1058                         vector_del_slot(vecs->pathvec, i);
1059                         free_path(pp);
1060                         i--;
1061                 }
1062                 else
1063                         pp->checkint = conf->checkint;
1064         }
1065         if (map_discovery(vecs))
1066                 return 1;
1067
1068         /*
1069          * create new set of maps & push changed ones into dm
1070          */
1071         if (coalesce_paths(vecs, mpvec, NULL))
1072                 return 1;
1073
1074         /*
1075          * may need to remove some maps which are no longer relevant
1076          * e.g., due to blacklist changes in conf file
1077          */
1078         if (coalesce_maps(vecs, mpvec))
1079                 return 1;
1080
1081         dm_lib_release();
1082
1083         sync_maps_state(mpvec);
1084
1085         /*
1086          * purge dm of old maps
1087          */
1088         remove_maps(vecs, NULL);
1089
1090         /*
1091          * save new set of maps formed by considering current path state
1092          */
1093         vector_free(vecs->mpvec);
1094         vecs->mpvec = mpvec;
1095
1096         /*
1097          * start dm event waiter threads for these new maps
1098          */
1099         vector_foreach_slot(vecs->mpvec, mpp, i) {
1100                 if (setup_multipath(vecs, mpp))
1101                         return 1;
1102                 if (start_waiters)
1103                         if (start_waiter_thread(mpp, vecs))
1104                                 return 1;
1105         }
1106         return 0;
1107 }
1108
1109 int
1110 reconfigure (struct vectors * vecs)
1111 {
1112         struct config * old = conf;
1113
1114         /*
1115          * free old map and path vectors ... they use old conf state
1116          */
1117         if (VECTOR_SIZE(vecs->mpvec))
1118                 remove_maps(vecs, stop_waiter_thread);
1119
1120         if (VECTOR_SIZE(vecs->pathvec))
1121                 free_pathvec(vecs->pathvec, FREE_PATHS);
1122
1123         vecs->pathvec = NULL;
1124         conf = NULL;
1125
1126         if (load_config(DEFAULT_CONFIGFILE))
1127                 return 1;
1128
1129         conf->verbosity = old->verbosity;
1130
1131         if (!conf->checkint) {
1132                 conf->checkint = DEFAULT_CHECKINT;
1133                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1134         }
1135         configure(vecs, 1);
1136         free_config(old);
1137         return 0;
1138 }
1139
1140 static struct vectors *
1141 init_vecs (void)
1142 {
1143         struct vectors * vecs;
1144
1145         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1146
1147         if (!vecs)
1148                 return NULL;
1149
1150         vecs->lock =
1151                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1152
1153         if (!vecs->lock)
1154                 goto out;
1155
1156         pthread_mutex_init(vecs->lock, NULL);
1157
1158         return vecs;
1159
1160 out:
1161         FREE(vecs);
1162         condlog(0, "failed to init paths");
1163         return NULL;
1164 }
1165
1166 static void *
1167 signal_set(int signo, void (*func) (int))
1168 {
1169         int r;
1170         struct sigaction sig;
1171         struct sigaction osig;
1172
1173         sig.sa_handler = func;
1174         sigemptyset(&sig.sa_mask);
1175         sig.sa_flags = 0;
1176
1177         r = sigaction(signo, &sig, &osig);
1178
1179         if (r < 0)
1180                 return (SIG_ERR);
1181         else
1182                 return (osig.sa_handler);
1183 }
1184
1185 static void
1186 sighup (int sig)
1187 {
1188         condlog(2, "reconfigure (SIGHUP)");
1189
1190         lock(gvecs->lock);
1191         reconfigure(gvecs);
1192         unlock(gvecs->lock);
1193
1194 #ifdef _DEBUG_
1195         dbg_free_final(NULL);
1196 #endif
1197 }
1198
1199 static void
1200 sigend (int sig)
1201 {
1202         exit_daemon(0);
1203 }
1204
1205 static void
1206 sigusr1 (int sig)
1207 {
1208         condlog(3, "SIGUSR1 received");
1209 }
1210
1211 static void
1212 signal_init(void)
1213 {
1214         signal_set(SIGHUP, sighup);
1215         signal_set(SIGUSR1, sigusr1);
1216         signal_set(SIGINT, sigend);
1217         signal_set(SIGTERM, sigend);
1218         signal(SIGPIPE, SIG_IGN);
1219 }
1220
1221 static void
1222 setscheduler (void)
1223 {
1224         int res;
1225         static struct sched_param sched_param = {
1226                 .sched_priority = 99
1227         };
1228
1229         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1230
1231         if (res == -1)
1232                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1233         return;
1234 }
1235
1236 static void
1237 set_oom_adj (int val)
1238 {
1239         FILE *fp;
1240
1241         fp = fopen("/proc/self/oom_adj", "w");
1242
1243         if (!fp)
1244                 return;
1245
1246         fprintf(fp, "%i", val);
1247         fclose(fp);
1248 }
1249
1250 static int
1251 child (void * param)
1252 {
1253         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1254         pthread_attr_t attr;
1255         struct vectors * vecs;
1256
1257         mlockall(MCL_CURRENT | MCL_FUTURE);
1258
1259         if (logsink)
1260                 log_thread_start();
1261
1262         condlog(2, "--------start up--------");
1263         condlog(2, "read " DEFAULT_CONFIGFILE);
1264
1265         if (load_config(DEFAULT_CONFIGFILE))
1266                 exit(1);
1267
1268         setlogmask(LOG_UPTO(conf->verbosity + 3));
1269
1270         /*
1271          * fill the voids left in the config file
1272          */
1273         if (!conf->checkint) {
1274                 conf->checkint = DEFAULT_CHECKINT;
1275                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1276         }
1277
1278         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1279                 if (logsink)
1280                         log_thread_stop();
1281
1282                 exit(1);
1283         }
1284         signal_init();
1285         setscheduler();
1286         set_oom_adj(-16);
1287         vecs = gvecs = init_vecs();
1288
1289         if (!vecs)
1290                 exit(1);
1291
1292         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1293                 condlog(0, "can not find sysfs mount point");
1294                 exit(1);
1295         }
1296
1297         /*
1298          * fetch and configure both paths and multipaths
1299          */
1300         if (configure(vecs, 1)) {
1301                 condlog(0, "failure during configuration");
1302                 exit(1);
1303         }
1304
1305         /*
1306          * start threads
1307          */
1308         pthread_attr_init(&attr);
1309         pthread_attr_setstacksize(&attr, 64 * 1024);
1310         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1311
1312         pthread_create(&check_thr, &attr, checkerloop, vecs);
1313         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1314         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1315
1316         pthread_cond_wait(&exit_cond, &exit_mutex);
1317
1318         /*
1319          * exit path
1320          */
1321         lock(vecs->lock);
1322         remove_maps(vecs, stop_waiter_thread);
1323         free_pathvec(vecs->pathvec, FREE_PATHS);
1324
1325         pthread_cancel(check_thr);
1326         pthread_cancel(uevent_thr);
1327         pthread_cancel(uxlsnr_thr);
1328
1329         sysfs_cleanup();
1330
1331         free_keys(keys);
1332         keys = NULL;
1333         free_handlers(handlers);
1334         handlers = NULL;
1335         free_polls();
1336
1337         unlock(vecs->lock);
1338         pthread_mutex_destroy(vecs->lock);
1339         FREE(vecs->lock);
1340         vecs->lock = NULL;
1341         FREE(vecs);
1342         vecs = NULL;
1343         free_config(conf);
1344         conf = NULL;
1345
1346         condlog(2, "--------shut down-------");
1347
1348         if (logsink)
1349                 log_thread_stop();
1350
1351         dm_lib_release();
1352         dm_lib_exit();
1353
1354 #ifdef _DEBUG_
1355         dbg_free_final(NULL);
1356 #endif
1357
1358         exit(0);
1359 }
1360
1361 static int
1362 daemonize(void)
1363 {
1364         int pid;
1365         int in_fd, out_fd;
1366
1367         if( (pid = fork()) < 0){
1368                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1369                 return -1;
1370         }
1371         else if (pid != 0)
1372                 return pid;
1373
1374         setsid();
1375
1376         if ( (pid = fork()) < 0)
1377                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1378         else if (pid != 0)
1379                 _exit(0);
1380
1381         in_fd = open("/dev/null", O_RDONLY);
1382         if (in_fd < 0){
1383                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1384                         strerror(errno));
1385                 _exit(0);
1386         }
1387         out_fd = open("/dev/console", O_WRONLY);
1388         if (out_fd < 0){
1389                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1390                         strerror(errno));
1391                 _exit(0);
1392         }
1393
1394         close(STDIN_FILENO);
1395         dup(in_fd);
1396         close(STDOUT_FILENO);
1397         dup(out_fd);
1398         close(STDERR_FILENO);
1399         dup(out_fd);
1400
1401         close(in_fd);
1402         close(out_fd);
1403         chdir("/");
1404         umask(0);
1405         return 0;
1406 }
1407
1408 int
1409 main (int argc, char *argv[])
1410 {
1411         extern char *optarg;
1412         extern int optind;
1413         int arg;
1414         int err;
1415
1416         logsink = 1;
1417         dm_init();
1418
1419         if (getuid() != 0) {
1420                 fprintf(stderr, "need to be root\n");
1421                 exit(1);
1422         }
1423
1424         /* make sure we don't lock any path */
1425         chdir("/");
1426         umask(umask(077) | 022);
1427
1428         conf = alloc_config();
1429
1430         if (!conf)
1431                 exit(1);
1432
1433         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1434         switch(arg) {
1435                 case 'd':
1436                         logsink = 0;
1437                         //debug=1; /* ### comment me out ### */
1438                         break;
1439                 case 'v':
1440                         if (sizeof(optarg) > sizeof(char *) ||
1441                             !isdigit(optarg[0]))
1442                                 exit(1);
1443
1444                         conf->verbosity = atoi(optarg);
1445                         break;
1446                 case 'k':
1447                         uxclnt(optarg);
1448                         exit(0);
1449                 default:
1450                         ;
1451                 }
1452         }
1453
1454         if (!logsink)
1455                 err = 0;
1456         else
1457                 err = daemonize();
1458
1459         if (err < 0)
1460                 /* error */
1461                 exit(1);
1462         else if (err > 0)
1463                 /* parent dies */
1464                 exit(0);
1465         else
1466                 /* child lives */
1467                 return (child(NULL));
1468 }
1469