multipathd/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Christophe Varoqui
   3  * Copyright (c) 2005 Kiyoshi Ueda, NEC
   4  * Copyright (c) 2005 Benjamin Marzinski, Redhat
   5  * Copyright (c) 2005 Edward Goggin, EMC
   6  */
   7 #include <unistd.h>
   8 #include <sys/stat.h>
   9 #include <libdevmapper.h>
  10 #include <wait.h>
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <fcntl.h>
  14 #include <errno.h>
  15 #include <sys/time.h>
  16 #include <sys/resource.h>
  17
  18 /*
  19  * libcheckers
  20  */
  21 #include <checkers.h>
  22
  23 /*
  24  * libmultipath
  25  */
  26 #include <parser.h>
  27 #include <vector.h>
  28 #include <memory.h>
  29 #include <config.h>
  30 #include <util.h>
  31 #include <hwtable.h>
  32 #include <defaults.h>
  33 #include <structs.h>
  34 #include <callout.h>
  35 #include <blacklist.h>
  36 #include <structs_vec.h>
  37 #include <dmparser.h>
  38 #include <devmapper.h>
  39 #include <sysfs.h>
  40 #include <dict.h>
  41 #include <discovery.h>
  42 #include <debug.h>
  43 #include <propsel.h>
  44 #include <uevent.h>
  45 #include <switchgroup.h>
  46 #include <print.h>
  47 #include <configure.h>
  48 #include <prio.h>
  49
  50 #include "main.h"
  51 #include "pidfile.h"
  52 #include "uxlsnr.h"
  53 #include "uxclnt.h"
  54 #include "cli.h"
  55 #include "cli_handlers.h"
  56 #include "lock.h"
  57 #include "waiter.h"
  58
  59 #define FILE_NAME_SIZE 256
  60 #define CMDSIZE 160
  61
  62 #define LOG_MSG(a,b) \
  63         if (strlen(b)) condlog(a, "%s: %s", pp->dev, b);
  64
  65 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  66 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  67
  68 /*
  69  * global copy of vecs for use in sig handlers
  70  */
  71 struct vectors * gvecs;
  72
  73 static int
  74 need_switch_pathgroup (struct multipath * mpp, int refresh)
  75 {
  76         struct pathgroup * pgp;
  77         struct path * pp;
  78         unsigned int i, j;
  79
  80         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  81                 return 0;
  82
  83         /*
  84          * Refresh path priority values
  85          */
  86         if (refresh)
  87                 vector_foreach_slot (mpp->pg, pgp, i)
  88                         vector_foreach_slot (pgp->paths, pp, j)
  89                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  90
  91         mpp->bestpg = select_path_group(mpp);
  92
  93         if (mpp->bestpg != mpp->nextpg)
  94                 return 1;
  95
  96         return 0;
  97 }
  98
  99 static void
 100 switch_pathgroup (struct multipath * mpp)
 101 {
 102         mpp->stat_switchgroup++;
 103         dm_switchgroup(mpp->alias, mpp->bestpg);
 104         condlog(2, "%s: switch to path group #%i",
 105                  mpp->alias, mpp->bestpg);
 106 }
 107
 108 static int
 109 coalesce_maps(struct vectors *vecs, vector nmpv)
 110 {
 111         struct multipath * ompp;
 112         vector ompv = vecs->mpvec;
 113         unsigned int i;
 114         int j;
 115
 116         vector_foreach_slot (ompv, ompp, i) {
 117                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 118                         /*
 119                          * remove all current maps not allowed by the
 120                          * current configuration
 121                          */
 122                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 123                                 condlog(0, "%s: unable to flush devmap",
 124                                         ompp->alias);
 125                                 /*
 126                                  * may be just because the device is open
 127                                  */
 128                                 if (!vector_alloc_slot(nmpv))
 129                                         return 1;
 130
 131                                 vector_set_slot(nmpv, ompp);
 132                                 setup_multipath(vecs, ompp);
 133
 134                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 135                                         vector_del_slot(ompv, j);
 136
 137                                 continue;
 138                         }
 139                         else {
 140                                 dm_lib_release();
 141                                 condlog(3, "%s devmap removed", ompp->alias);
 142                         }
 143                 }
 144         }
 145         return 0;
 146 }
 147
 148 static void
 149 sync_map_state(struct multipath *mpp)
 150 {
 151         struct pathgroup *pgp;
 152         struct path *pp;
 153         unsigned int i, j;
 154
 155         vector_foreach_slot (mpp->pg, pgp, i){
 156                 vector_foreach_slot (pgp->paths, pp, j){
 157                         if (pp->state <= PATH_UNCHECKED)
 158                                 continue;
 159                         if ((pp->dmstate == PSTATE_FAILED ||
 160                              pp->dmstate == PSTATE_UNDEF) &&
 161                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 162                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 163                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 164                                   pp->dmstate == PSTATE_UNDEF) &&
 165                                  (pp->state == PATH_DOWN ||
 166                                   pp->state == PATH_SHAKY))
 167                                 dm_fail_path(mpp->alias, pp->dev_t);
 168                 }
 169         }
 170 }
 171
 172 static void
 173 sync_maps_state(vector mpvec)
 174 {
 175         unsigned int i;
 176         struct multipath *mpp;
 177
 178         vector_foreach_slot (mpvec, mpp, i)
 179                 sync_map_state(mpp);
 180 }
 181
 182 static int
 183 flush_map(struct multipath * mpp, struct vectors * vecs)
 184 {
 185         /*
 186          * clear references to this map before flushing so we can ignore
 187          * the spurious uevent we may generate with the dm_flush_map call below
 188          */
 189         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 190                 /*
 191                  * May not really be an error -- if the map was already flushed
 192                  * from the device mapper by dmsetup(8) for instance.
 193                  */
 194                 condlog(0, "%s: can't flush", mpp->alias);
 195                 return 1;
 196         }
 197         else {
 198                 dm_lib_release();
 199                 condlog(3, "%s: devmap removed", mpp->alias);
 200         }
 201
 202         orphan_paths(vecs->pathvec, mpp);
 203         remove_map_and_stop_waiter(mpp, vecs, 1);
 204
 205         return 0;
 206 }
 207
 208 static int
 209 uev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 210 {
 211         condlog(2, "%s: add map (uevent)", dev->kernel);
 212         return ev_add_map(dev, vecs);
 213 }
 214
 215 int
 216 ev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 217 {
 218         char * alias;
 219         char *dev_t;
 220         int major, minor;
 221         char * refwwid;
 222         struct multipath * mpp;
 223         int map_present;
 224         int r = 1;
 225
 226         dev_t = sysfs_attr_get_value(dev->devpath, "dev");
 227
 228         if (!dev_t || sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 229                 return 1;
 230
 231         alias = dm_mapname(major, minor);
 232
 233         if (!alias)
 234                 return 1;
 235
 236         map_present = dm_map_present(alias);
 237
 238         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 239                 condlog(4, "%s: not a multipath map", alias);
 240                 return 0;
 241         }
 242
 243         mpp = find_mp_by_alias(vecs->mpvec, alias);
 244
 245         if (mpp) {
 246                 /*
 247                  * Not really an error -- we generate our own uevent
 248                  * if we create a multipath mapped device as a result
 249                  * of uev_add_path
 250                  */
 251                 condlog(0, "%s: devmap already registered",
 252                         dev->kernel);
 253                 return 0;
 254         }
 255
 256         /*
 257          * now we can register the map
 258          */
 259         if (map_present && (mpp = add_map_without_path(vecs, minor, alias))) {
 260                 sync_map_state(mpp);
 261                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 262                 return 0;
 263         }
 264         refwwid = get_refwwid(dev->kernel, DEV_DEVMAP, vecs->pathvec);
 265
 266         if (refwwid) {
 267                 r = coalesce_paths(vecs, NULL, refwwid);
 268                 dm_lib_release();
 269         }
 270
 271         if (!r)
 272                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 273         else
 274                 condlog(0, "%s: uev_add_map %s failed", alias, dev->kernel);
 275
 276         FREE(refwwid);
 277         return r;
 278 }
 279
 280 static int
 281 uev_remove_map (struct sysfs_device * dev, struct vectors * vecs)
 282 {
 283         condlog(2, "%s: remove map (uevent)", dev->kernel);
 284         return ev_remove_map(dev->kernel, vecs);
 285 }
 286
 287 int
 288 ev_remove_map (char * devname, struct vectors * vecs)
 289 {
 290         struct multipath * mpp;
 291
 292         mpp = find_mp_by_str(vecs->mpvec, devname);
 293
 294         if (!mpp) {
 295                 condlog(3, "%s: devmap not registered, can't remove",
 296                         devname);
 297                 return 0;
 298         }
 299         flush_map(mpp, vecs);
 300
 301         return 0;
 302 }
 303
 304 static int
 305 uev_umount_map (struct sysfs_device * dev, struct vectors * vecs)
 306 {
 307         struct multipath * mpp;
 308
 309         condlog(2, "%s: umount map (uevent)", dev->kernel);
 310
 311         mpp = find_mp_by_str(vecs->mpvec, dev->kernel);
 312
 313         if (!mpp)
 314                 return 0;
 315
 316         update_mpp_paths(mpp, vecs->pathvec);
 317         verify_paths(mpp, vecs, NULL);
 318
 319         if (!VECTOR_SIZE(mpp->paths))
 320                 flush_map(mpp, vecs);
 321
 322         return 0;
 323 }
 324
 325 static int
 326 uev_add_path (struct sysfs_device * dev, struct vectors * vecs)
 327 {
 328         condlog(2, "%s: add path (uevent)", dev->kernel);
 329         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 330 }
 331
 332
 333 /*
 334  * returns:
 335  * 0: added
 336  * 1: error
 337  * 2: blacklisted
 338  */
 339 int
 340 ev_add_path (char * devname, struct vectors * vecs)
 341 {
 342         struct multipath * mpp;
 343         struct path * pp;
 344         char empty_buff[WWID_SIZE] = {0};
 345
 346         pp = find_path_by_dev(vecs->pathvec, devname);
 347
 348         if (pp) {
 349                 condlog(0, "%s: spurious uevent, path already in pathvec",
 350                         devname);
 351                 if (pp->mpp)
 352                         return 0;
 353         }
 354         else {
 355                 /*
 356                  * get path vital state
 357                  */
 358                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 359                       devname, DI_ALL))) {
 360                         condlog(0, "%s: failed to store path info", devname);
 361                         return 1;
 362                 }
 363                 pp->checkint = conf->checkint;
 364         }
 365
 366         /*
 367          * need path UID to go any further
 368          */
 369         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 370                 condlog(0, "%s: failed to get path uid", devname);
 371                 return 1; /* leave path added to pathvec */
 372         }
 373         if (filter_path(conf, pp) > 0){
 374                 int i = find_slot(vecs->pathvec, (void *)pp);
 375                 if (i != -1)
 376                         vector_del_slot(vecs->pathvec, i);
 377                 free_path(pp);
 378                 return 2;
 379         }
 380         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 381 rescan:
 382         if (mpp) {
 383                 if (adopt_paths(vecs->pathvec, mpp))
 384                         return 1; /* leave path added to pathvec */
 385
 386                 verify_paths(mpp, vecs, NULL);
 387                 mpp->action = ACT_RELOAD;
 388         }
 389         else {
 390                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 391                         mpp->action = ACT_CREATE;
 392                 else
 393                         return 1; /* leave path added to pathvec */
 394         }
 395
 396         /*
 397          * push the map to the device-mapper
 398          */
 399         if (setup_map(mpp)) {
 400                 condlog(0, "%s: failed to setup map for addition of new "
 401                         "path %s", mpp->alias, devname);
 402                 goto out;
 403         }
 404         /*
 405          * reload the map for the multipath mapped device
 406          */
 407         if (domap(mpp) <= 0) {
 408                 condlog(0, "%s: failed in domap for addition of new "
 409                         "path %s", mpp->alias, devname);
 410                 /*
 411                  * deal with asynchronous uevents :((
 412                  */
 413                 if (mpp->action == ACT_RELOAD) {
 414                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 415                         sleep(1);
 416                         update_mpp_paths(mpp, vecs->pathvec);
 417                         goto rescan;
 418                 }
 419                 else
 420                         goto out;
 421         }
 422         dm_lib_release();
 423
 424         /*
 425          * update our state from kernel regardless of create or reload
 426          */
 427         if (setup_multipath(vecs, mpp))
 428                 goto out;
 429
 430         sync_map_state(mpp);
 431
 432         if (mpp->action == ACT_CREATE &&
 433             start_waiter_thread(mpp, vecs))
 434                         goto out;
 435
 436         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 437         return 0;
 438
 439 out:
 440         remove_map(mpp, vecs, 1);
 441         return 1;
 442 }
 443
 444 static int
 445 uev_remove_path (struct sysfs_device * dev, struct vectors * vecs)
 446 {
 447         int retval;
 448
 449         condlog(2, "%s: remove path (uevent)", dev->kernel);
 450         retval = ev_remove_path(dev->kernel, vecs);
 451         if (!retval)
 452                 sysfs_device_put(dev);
 453
 454         return retval;
 455 }
 456
 457 int
 458 ev_remove_path (char * devname, struct vectors * vecs)
 459 {
 460         struct multipath * mpp;
 461         struct path * pp;
 462         int i;
 463         int rm_path = 1;
 464
 465         pp = find_path_by_dev(vecs->pathvec, devname);
 466
 467         if (!pp) {
 468                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 469                 return 1;
 470         }
 471
 472         /*
 473          * avoid referring to the map of an orphanned path
 474          */
 475         if ((mpp = pp->mpp)) {
 476
 477                 /*
 478                  * remove the map IFF removing the last path
 479                  */
 480                 if (pathcount(mpp, PATH_WILD) > 1) {
 481                         vector rpvec = vector_alloc();
 482
 483                         /*
 484                          * transform the mp->pg vector of vectors of paths
 485                          * into a mp->params string to feed the device-mapper
 486                          */
 487                         update_mpp_paths(mpp, vecs->pathvec);
 488                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 489                                 vector_del_slot(mpp->paths, i);
 490
 491                         if (VECTOR_SIZE(mpp->paths) == 0) {
 492                                 char alias[WWID_SIZE];
 493
 494                                 /*
 495                                  * flush_map will fail if the device is open
 496                                  */
 497                                 strncpy(alias, mpp->alias, WWID_SIZE);
 498                                 if (flush_map(mpp, vecs))
 499                                         rm_path = 0;
 500                                 else
 501                                         condlog(3, "%s: removed map after removing"
 502                                                 " multiple paths", alias);
 503                         }
 504                         else {
 505                                 if (setup_map(mpp)) {
 506                                         condlog(0, "%s: failed to setup map for"
 507                                                 " removal of path %s", mpp->alias, devname);
 508                                         free_pathvec(rpvec, KEEP_PATHS);
 509                                         goto out;
 510                                 }
 511                                 /*
 512                                  * reload the map
 513                                  */
 514                                 mpp->action = ACT_RELOAD;
 515                                 if (domap(mpp) <= 0) {
 516                                         condlog(0, "%s: failed in domap for "
 517                                                 "removal of path %s",
 518                                                 mpp->alias, devname);
 519                                         /*
 520                                          * Delete path from pathvec so that
 521                                          * update_mpp_paths wont find it later
 522                                          * when/if another path is removed.
 523                                          */
 524                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 525                                                 vector_del_slot(vecs->pathvec, i);
 526                                         free_path(pp);
 527                                         return 1;
 528                                 }
 529                                 /*
 530                                  * update our state from kernel
 531                                  */
 532                                 if (setup_multipath(vecs, mpp)) {
 533                                         free_pathvec(rpvec, KEEP_PATHS);
 534                                         goto out;
 535                                 }
 536                                 sync_map_state(mpp);
 537
 538                                 condlog(3, "%s: path removed from map %s",
 539                                         devname, mpp->alias);
 540                         }
 541                         free_pathvec(rpvec, KEEP_PATHS);
 542                 }
 543                 else {
 544                         char alias[WWID_SIZE];
 545
 546                         /*
 547                          * flush_map will fail if the device is open
 548                          */
 549                         strncpy(alias, mpp->alias, WWID_SIZE);
 550                         if (flush_map(mpp, vecs))
 551                                 rm_path = 0;
 552                         else
 553                                 condlog(3, "%s: removed map", alias);
 554                 }
 555         }
 556
 557         if (rm_path) {
 558                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 559                         vector_del_slot(vecs->pathvec, i);
 560                 free_path(pp);
 561         }
 562
 563         return 0;
 564
 565 out:
 566         remove_map_and_stop_waiter(mpp, vecs, 1);
 567         return 1;
 568 }
 569
 570 static int
 571 map_discovery (struct vectors * vecs)
 572 {
 573         struct multipath * mpp;
 574         unsigned int i;
 575
 576         if (dm_get_maps(vecs->mpvec, "multipath"))
 577                 return 1;
 578
 579         vector_foreach_slot (vecs->mpvec, mpp, i)
 580                 if (setup_multipath(vecs, mpp))
 581                         return 1;
 582
 583         return 0;
 584 }
 585
 586 int
 587 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 588 {
 589         struct vectors * vecs;
 590         int r;
 591
 592         *reply = NULL;
 593         *len = 0;
 594         vecs = (struct vectors *)trigger_data;
 595
 596         pthread_cleanup_push(cleanup_lock, vecs->lock);
 597         lock(vecs->lock);
 598
 599         r = parse_cmd(str, reply, len, vecs);
 600
 601         if (r > 0) {
 602                 *reply = STRDUP("fail\n");
 603                 *len = strlen(*reply) + 1;
 604                 r = 1;
 605         }
 606         else if (!r && *len == 0) {
 607                 *reply = STRDUP("ok\n");
 608                 *len = strlen(*reply) + 1;
 609                 r = 0;
 610         }
 611         /* else if (r < 0) leave *reply alone */
 612
 613         lock_cleanup_pop(vecs->lock);
 614         return r;
 615 }
 616
 617 static int
 618 uev_discard(char * devpath)
 619 {
 620         char a[10], b[10];
 621
 622         /*
 623          * keep only block devices, discard partitions
 624          */
 625         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 626             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 627                 condlog(4, "discard event on %s", devpath);
 628                 return 1;
 629         }
 630         return 0;
 631 }
 632
 633 int
 634 uev_trigger (struct uevent * uev, void * trigger_data)
 635 {
 636         int r = 0;
 637         struct sysfs_device *sysdev;
 638         struct vectors * vecs;
 639
 640         vecs = (struct vectors *)trigger_data;
 641
 642         if (uev_discard(uev->devpath))
 643                 return 0;
 644
 645         sysdev = sysfs_device_get(uev->devpath);
 646         if(!sysdev)
 647                 return 0;
 648
 649         lock(vecs->lock);
 650
 651         /*
 652          * device map event
 653          * Add events are ignored here as the tables
 654          * are not fully initialised then.
 655          */
 656         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 657                 if (!strncmp(uev->action, "change", 6)) {
 658                         r = uev_add_map(sysdev, vecs);
 659                         goto out;
 660                 }
 661                 if (!strncmp(uev->action, "remove", 6)) {
 662                         r = uev_remove_map(sysdev, vecs);
 663                         goto out;
 664                 }
 665                 if (!strncmp(uev->action, "umount", 6)) {
 666                         r = uev_umount_map(sysdev, vecs);
 667                         goto out;
 668                 }
 669                 goto out;
 670         }
 671
 672         /*
 673          * path add/remove event
 674          */
 675         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 676                            sysdev->kernel) > 0)
 677                 goto out;
 678
 679         if (!strncmp(uev->action, "add", 3)) {
 680                 r = uev_add_path(sysdev, vecs);
 681                 goto out;
 682         }
 683         if (!strncmp(uev->action, "remove", 6)) {
 684                 r = uev_remove_path(sysdev, vecs);
 685                 goto out;
 686         }
 687
 688 out:
 689         unlock(vecs->lock);
 690         return r;
 691 }
 692
 693 static void *
 694 ueventloop (void * ap)
 695 {
 696         if (uevent_listen(&uev_trigger, ap))
 697                 fprintf(stderr, "error starting uevent listener");
 698
 699         return NULL;
 700 }
 701
 702 static void *
 703 uxlsnrloop (void * ap)
 704 {
 705         if (cli_init())
 706                 return NULL;
 707
 708         set_handler_callback(LIST+PATHS, cli_list_paths);
 709         set_handler_callback(LIST+MAPS, cli_list_maps);
 710         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 711         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 712         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 713         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 714         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 715         set_handler_callback(LIST+CONFIG, cli_list_config);
 716         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 717         set_handler_callback(LIST+DEVICES, cli_list_devices);
 718         set_handler_callback(ADD+PATH, cli_add_path);
 719         set_handler_callback(DEL+PATH, cli_del_path);
 720         set_handler_callback(ADD+MAP, cli_add_map);
 721         set_handler_callback(DEL+MAP, cli_del_map);
 722         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 723         set_handler_callback(RECONFIGURE, cli_reconfigure);
 724         set_handler_callback(SUSPEND+MAP, cli_suspend);
 725         set_handler_callback(RESUME+MAP, cli_resume);
 726         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 727         set_handler_callback(FAIL+PATH, cli_fail);
 728
 729         uxsock_listen(&uxsock_trigger, ap);
 730
 731         return NULL;
 732 }
 733
 734 static int
 735 exit_daemon (int status)
 736 {
 737         if (status != 0)
 738                 fprintf(stderr, "bad exit status. see daemon.log\n");
 739
 740         condlog(3, "unlink pidfile");
 741         unlink(DEFAULT_PIDFILE);
 742
 743         lock(&exit_mutex);
 744         pthread_cond_signal(&exit_cond);
 745         unlock(&exit_mutex);
 746
 747         return status;
 748 }
 749
 750 static void
 751 fail_path (struct path * pp, int del_active)
 752 {
 753         if (!pp->mpp)
 754                 return;
 755
 756         condlog(2, "checker failed path %s in map %s",
 757                  pp->dev_t, pp->mpp->alias);
 758
 759         dm_fail_path(pp->mpp->alias, pp->dev_t);
 760         if (del_active)
 761                 update_queue_mode_del_path(pp->mpp);
 762 }
 763
 764 /*
 765  * caller must have locked the path list before calling that function
 766  */
 767 static void
 768 reinstate_path (struct path * pp, int add_active)
 769 {
 770         if (!pp->mpp)
 771                 return;
 772
 773         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 774                 condlog(0, "%s: reinstate failed", pp->dev_t);
 775         else {
 776                 condlog(2, "%s: reinstated", pp->dev_t);
 777                 if (add_active)
 778                         update_queue_mode_add_path(pp->mpp);
 779         }
 780 }
 781
 782 static void
 783 enable_group(struct path * pp)
 784 {
 785         struct pathgroup * pgp;
 786
 787         /*
 788          * if path is added through uev_add_path, pgindex can be unset.
 789          * next update_strings() will set it, upon map reload event.
 790          *
 791          * we can safely return here, because upon map reload, all
 792          * PG will be enabled.
 793          */
 794         if (!pp->mpp->pg || !pp->pgindex)
 795                 return;
 796
 797         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 798
 799         if (pgp->status == PGSTATE_DISABLED) {
 800                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 801                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 802         }
 803 }
 804
 805 static void
 806 mpvec_garbage_collector (struct vectors * vecs)
 807 {
 808         struct multipath * mpp;
 809         unsigned int i;
 810
 811         vector_foreach_slot (vecs->mpvec, mpp, i) {
 812                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 813                         condlog(2, "%s: remove dead map", mpp->alias);
 814                         remove_map_and_stop_waiter(mpp, vecs, 1);
 815                         i--;
 816                 }
 817         }
 818 }
 819
 820 static void
 821 defered_failback_tick (vector mpvec)
 822 {
 823         struct multipath * mpp;
 824         unsigned int i;
 825
 826         vector_foreach_slot (mpvec, mpp, i) {
 827                 /*
 828                  * defered failback getting sooner
 829                  */
 830                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 831                         mpp->failback_tick--;
 832
 833                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 834                                 switch_pathgroup(mpp);
 835                 }
 836         }
 837 }
 838
 839 static void
 840 retry_count_tick(vector mpvec)
 841 {
 842         struct multipath *mpp;
 843         unsigned int i;
 844
 845         vector_foreach_slot (mpvec, mpp, i) {
 846                 if (mpp->retry_tick) {
 847                         mpp->stat_total_queueing_time++;
 848                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 849                         if(--mpp->retry_tick == 0) {
 850                                 dm_queue_if_no_path(mpp->alias, 0);
 851                                 condlog(2, "%s: Disable queueing", mpp->alias);
 852                         }
 853                 }
 854         }
 855 }
 856
 857 void
 858 check_path (struct vectors * vecs, struct path * pp)
 859 {
 860         int newstate;
 861
 862         if (!pp->mpp)
 863                 return;
 864
 865         if (pp->tick && --pp->tick)
 866                 return; /* don't check this path yet */
 867
 868         /*
 869          * provision a next check soonest,
 870          * in case we exit abnormaly from here
 871          */
 872         pp->tick = conf->checkint;
 873
 874         if (!checker_selected(&pp->checker)) {
 875                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 876                 select_checker(pp);
 877         }
 878         if (!checker_selected(&pp->checker)) {
 879                 condlog(0, "%s: checker is not set", pp->dev);
 880                 return;
 881         }
 882         /*
 883          * Set checker in async mode.
 884          * Honored only by checker implementing the said mode.
 885          */
 886         checker_set_async(&pp->checker);
 887
 888         newstate = checker_check(&pp->checker);
 889
 890         if (newstate < 0) {
 891                 condlog(2, "%s: unusable path", pp->dev);
 892                 pathinfo(pp, conf->hwtable, 0);
 893                 return;
 894         }
 895         /*
 896          * Async IO in flight. Keep the previous path state
 897          * and reschedule as soon as possible
 898          */
 899         if (newstate == PATH_PENDING) {
 900                 pp->tick = 1;
 901                 return;
 902         }
 903         if (newstate != pp->state) {
 904                 int oldstate = pp->state;
 905                 pp->state = newstate;
 906                 LOG_MSG(1, checker_message(&pp->checker));
 907
 908                 /*
 909                  * upon state change, reset the checkint
 910                  * to the shortest delay
 911                  */
 912                 pp->checkint = conf->checkint;
 913
 914                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY ||
 915                     update_multipath_strings(pp->mpp, vecs->pathvec)) {
 916                         /*
 917                          * proactively fail path in the DM
 918                          */
 919                         if (oldstate == PATH_UP ||
 920                             oldstate == PATH_GHOST)
 921                                 fail_path(pp, 1);
 922                         else
 923                                 fail_path(pp, 0);
 924
 925                         /*
 926                          * cancel scheduled failback
 927                          */
 928                         pp->mpp->failback_tick = 0;
 929
 930                         pp->mpp->stat_path_failures++;
 931                         return;
 932                 }
 933
 934                 /*
 935                  * reinstate this path
 936                  */
 937                 if (oldstate != PATH_UP &&
 938                     oldstate != PATH_GHOST)
 939                         reinstate_path(pp, 1);
 940                 else
 941                         reinstate_path(pp, 0);
 942
 943                 /*
 944                  * schedule [defered] failback
 945                  */
 946                 if (pp->mpp->pgfailback > 0)
 947                         pp->mpp->failback_tick =
 948                                 pp->mpp->pgfailback + 1;
 949                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 950                     need_switch_pathgroup(pp->mpp, 1))
 951                         switch_pathgroup(pp->mpp);
 952
 953                 /*
 954                  * if at least one path is up in a group, and
 955                  * the group is disabled, re-enable it
 956                  */
 957                 if (newstate == PATH_UP)
 958                         enable_group(pp);
 959         }
 960         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 961                 LOG_MSG(4, checker_message(&pp->checker));
 962                 /*
 963                  * double the next check delay.
 964                  * max at conf->max_checkint
 965                  */
 966                 if (pp->checkint < (conf->max_checkint / 2))
 967                         pp->checkint = 2 * pp->checkint;
 968                 else
 969                         pp->checkint = conf->max_checkint;
 970
 971                 pp->tick = pp->checkint;
 972                 condlog(4, "%s: delay next check %is",
 973                                 pp->dev_t, pp->tick);
 974         }
 975         else if (newstate == PATH_DOWN)
 976                 LOG_MSG(2, checker_message(&pp->checker));
 977
 978         pp->state = newstate;
 979
 980         /*
 981          * path prio refreshing
 982          */
 983         condlog(4, "path prio refresh");
 984         pathinfo(pp, conf->hwtable, DI_PRIO);
 985
 986         /*
 987          * pathgroup failback policy
 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
1000 static void *
1001 checkerloop (void *ap)
1002 {
1003         struct vectors *vecs;
1004         struct path *pp;
1005         int count = 0;
1006         unsigned int i;
1007
1008         mlockall(MCL_CURRENT | MCL_FUTURE);
1009         vecs = (struct vectors *)ap;
1010         condlog(2, "path checkers start up");
1011
1012         /*
1013          * init the path check interval
1014          */
1015         vector_foreach_slot (vecs->pathvec, pp, i) {
1016                 pp->checkint = conf->checkint;
1017         }
1018
1019         while (1) {
1020                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1021                 lock(vecs->lock);
1022                 condlog(4, "tick");
1023
1024                 vector_foreach_slot (vecs->pathvec, pp, i) {
1025                         check_path(vecs, pp);
1026                 }
1027                 defered_failback_tick(vecs->mpvec);
1028                 retry_count_tick(vecs->mpvec);
1029
1030                 if (count)
1031                         count--;
1032                 else {
1033                         condlog(4, "map garbage collection");
1034                         mpvec_garbage_collector(vecs);
1035                         count = MAPGCINT;
1036                 }
1037
1038                 lock_cleanup_pop(vecs->lock);
1039                 sleep(1);
1040         }
1041         return NULL;
1042 }
1043
1044 int
1045 configure (struct vectors * vecs, int start_waiters)
1046 {
1047         struct multipath * mpp;
1048         struct path * pp;
1049         vector mpvec;
1050         int i;
1051
1052         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1053                 return 1;
1054
1055         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1056                 return 1;
1057
1058         if (!(mpvec = vector_alloc()))
1059                 return 1;
1060
1061         /*
1062          * probe for current path (from sysfs) and map (from dm) sets
1063          */
1064         path_discovery(vecs->pathvec, conf, DI_ALL);
1065
1066         vector_foreach_slot (vecs->pathvec, pp, i){
1067                 if (filter_path(conf, pp) > 0){
1068                         vector_del_slot(vecs->pathvec, i);
1069                         free_path(pp);
1070                         i--;
1071                 }
1072                 else
1073                         pp->checkint = conf->checkint;
1074         }
1075         if (map_discovery(vecs))
1076                 return 1;
1077
1078         /*
1079          * create new set of maps & push changed ones into dm
1080          */
1081         if (coalesce_paths(vecs, mpvec, NULL))
1082                 return 1;
1083
1084         /*
1085          * may need to remove some maps which are no longer relevant
1086          * e.g., due to blacklist changes in conf file
1087          */
1088         if (coalesce_maps(vecs, mpvec))
1089                 return 1;
1090
1091         dm_lib_release();
1092
1093         sync_maps_state(mpvec);
1094
1095         /*
1096          * purge dm of old maps
1097          */
1098         remove_maps(vecs);
1099
1100         /*
1101          * save new set of maps formed by considering current path state
1102          */
1103         vector_free(vecs->mpvec);
1104         vecs->mpvec = mpvec;
1105
1106         /*
1107          * start dm event waiter threads for these new maps
1108          */
1109         vector_foreach_slot(vecs->mpvec, mpp, i) {
1110                 if (setup_multipath(vecs, mpp))
1111                         return 1;
1112                 if (start_waiters)
1113                         if (start_waiter_thread(mpp, vecs))
1114                                 return 1;
1115         }
1116         return 0;
1117 }
1118
1119 int
1120 reconfigure (struct vectors * vecs)
1121 {
1122         struct config * old = conf;
1123
1124         /*
1125          * free old map and path vectors ... they use old conf state
1126          */
1127         if (VECTOR_SIZE(vecs->mpvec))
1128                 remove_maps_and_stop_waiters(vecs);
1129
1130         if (VECTOR_SIZE(vecs->pathvec))
1131                 free_pathvec(vecs->pathvec, FREE_PATHS);
1132
1133         vecs->pathvec = NULL;
1134         conf = NULL;
1135
1136         if (load_config(DEFAULT_CONFIGFILE))
1137                 return 1;
1138
1139         conf->verbosity = old->verbosity;
1140
1141         if (!conf->checkint) {
1142                 conf->checkint = DEFAULT_CHECKINT;
1143                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1144         }
1145         configure(vecs, 1);
1146         free_config(old);
1147         return 0;
1148 }
1149
1150 static struct vectors *
1151 init_vecs (void)
1152 {
1153         struct vectors * vecs;
1154
1155         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1156
1157         if (!vecs)
1158                 return NULL;
1159
1160         vecs->lock =
1161                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1162
1163         if (!vecs->lock)
1164                 goto out;
1165
1166         pthread_mutex_init(vecs->lock, NULL);
1167
1168         return vecs;
1169
1170 out:
1171         FREE(vecs);
1172         condlog(0, "failed to init paths");
1173         return NULL;
1174 }
1175
1176 static void *
1177 signal_set(int signo, void (*func) (int))
1178 {
1179         int r;
1180         struct sigaction sig;
1181         struct sigaction osig;
1182
1183         sig.sa_handler = func;
1184         sigemptyset(&sig.sa_mask);
1185         sig.sa_flags = 0;
1186
1187         r = sigaction(signo, &sig, &osig);
1188
1189         if (r < 0)
1190                 return (SIG_ERR);
1191         else
1192                 return (osig.sa_handler);
1193 }
1194
1195 static void
1196 sighup (int sig)
1197 {
1198         condlog(2, "reconfigure (SIGHUP)");
1199
1200         lock(gvecs->lock);
1201         reconfigure(gvecs);
1202         unlock(gvecs->lock);
1203
1204 #ifdef _DEBUG_
1205         dbg_free_final(NULL);
1206 #endif
1207 }
1208
1209 static void
1210 sigend (int sig)
1211 {
1212         exit_daemon(0);
1213 }
1214
1215 static void
1216 sigusr1 (int sig)
1217 {
1218         condlog(3, "SIGUSR1 received");
1219 }
1220
1221 static void
1222 signal_init(void)
1223 {
1224         signal_set(SIGHUP, sighup);
1225         signal_set(SIGUSR1, sigusr1);
1226         signal_set(SIGINT, sigend);
1227         signal_set(SIGTERM, sigend);
1228         signal(SIGPIPE, SIG_IGN);
1229 }
1230
1231 static void
1232 setscheduler (void)
1233 {
1234         int res;
1235         static struct sched_param sched_param = {
1236                 .sched_priority = 99
1237         };
1238
1239         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1240
1241         if (res == -1)
1242                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1243         return;
1244 }
1245
1246 static void
1247 set_oom_adj (int val)
1248 {
1249         FILE *fp;
1250
1251         fp = fopen("/proc/self/oom_adj", "w");
1252
1253         if (!fp)
1254                 return;
1255
1256         fprintf(fp, "%i", val);
1257         fclose(fp);
1258 }
1259
1260 static int
1261 child (void * param)
1262 {
1263         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1264         pthread_attr_t attr;
1265         struct vectors * vecs;
1266
1267         mlockall(MCL_CURRENT | MCL_FUTURE);
1268
1269         if (logsink)
1270                 log_thread_start();
1271
1272         condlog(2, "--------start up--------");
1273         condlog(2, "read " DEFAULT_CONFIGFILE);
1274
1275         if (init_checkers()) {
1276                 condlog(0, "failed to initialize checkers");
1277                 exit(1);
1278         }
1279         if (init_prio()) {
1280                 condlog(0, "failed to initialize prioritizers");
1281                 exit(1);
1282         }
1283         if (load_config(DEFAULT_CONFIGFILE))
1284                 exit(1);
1285
1286         setlogmask(LOG_UPTO(conf->verbosity + 3));
1287
1288         /*
1289          * fill the voids left in the config file
1290          */
1291         if (!conf->checkint) {
1292                 conf->checkint = DEFAULT_CHECKINT;
1293                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1294         }
1295
1296         if (conf->max_fds) {
1297                 struct rlimit fd_limit;
1298                 if (conf->max_fds > 0) {
1299                         fd_limit.rlim_cur = conf->max_fds;
1300                         fd_limit.rlim_max = conf->max_fds;
1301                 }
1302                 else {
1303                         fd_limit.rlim_cur = RLIM_INFINITY;
1304                         fd_limit.rlim_max = RLIM_INFINITY;
1305                 }
1306                 if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0)
1307                         condlog(0, "can't set open fds limit to %d : %s\n",
1308                                 conf->max_fds, strerror(errno));
1309         }
1310
1311         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1312                 if (logsink)
1313                         log_thread_stop();
1314
1315                 exit(1);
1316         }
1317         signal_init();
1318         setscheduler();
1319         set_oom_adj(-16);
1320         vecs = gvecs = init_vecs();
1321
1322         if (!vecs)
1323                 exit(1);
1324
1325         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1326                 condlog(0, "can not find sysfs mount point");
1327                 exit(1);
1328         }
1329
1330         /*
1331          * fetch and configure both paths and multipaths
1332          */
1333         if (configure(vecs, 1)) {
1334                 condlog(0, "failure during configuration");
1335                 exit(1);
1336         }
1337
1338         /*
1339          * start threads
1340          */
1341         pthread_attr_init(&attr);
1342         pthread_attr_setstacksize(&attr, 64 * 1024);
1343         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1344
1345         pthread_create(&check_thr, &attr, checkerloop, vecs);
1346         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1347         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1348
1349         pthread_cond_wait(&exit_cond, &exit_mutex);
1350
1351         /*
1352          * exit path
1353          */
1354         lock(vecs->lock);
1355         remove_maps_and_stop_waiters(vecs);
1356         free_pathvec(vecs->pathvec, FREE_PATHS);
1357
1358         pthread_cancel(check_thr);
1359         pthread_cancel(uevent_thr);
1360         pthread_cancel(uxlsnr_thr);
1361
1362         sysfs_cleanup();
1363
1364         free_keys(keys);
1365         keys = NULL;
1366         free_handlers(handlers);
1367         handlers = NULL;
1368         free_polls();
1369
1370         unlock(vecs->lock);
1371         pthread_mutex_destroy(vecs->lock);
1372         FREE(vecs->lock);
1373         vecs->lock = NULL;
1374         FREE(vecs);
1375         vecs = NULL;
1376         free_config(conf);
1377         conf = NULL;
1378
1379         condlog(2, "--------shut down-------");
1380
1381         if (logsink)
1382                 log_thread_stop();
1383
1384         dm_lib_release();
1385         dm_lib_exit();
1386
1387 #ifdef _DEBUG_
1388         dbg_free_final(NULL);
1389 #endif
1390
1391         exit(0);
1392 }
1393
1394 static int
1395 daemonize(void)
1396 {
1397         int pid;
1398         int in_fd, out_fd;
1399
1400         if( (pid = fork()) < 0){
1401                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1402                 return -1;
1403         }
1404         else if (pid != 0)
1405                 return pid;
1406
1407         setsid();
1408
1409         if ( (pid = fork()) < 0)
1410                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1411         else if (pid != 0)
1412                 _exit(0);
1413
1414         in_fd = open("/dev/null", O_RDONLY);
1415         if (in_fd < 0){
1416                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1417                         strerror(errno));
1418                 _exit(0);
1419         }
1420         out_fd = open("/dev/console", O_WRONLY);
1421         if (out_fd < 0){
1422                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1423                         strerror(errno));
1424                 _exit(0);
1425         }
1426
1427         close(STDIN_FILENO);
1428         dup(in_fd);
1429         close(STDOUT_FILENO);
1430         dup(out_fd);
1431         close(STDERR_FILENO);
1432         dup(out_fd);
1433
1434         close(in_fd);
1435         close(out_fd);
1436         chdir("/");
1437         umask(0);
1438         return 0;
1439 }
1440
1441 int
1442 main (int argc, char *argv[])
1443 {
1444         extern char *optarg;
1445         extern int optind;
1446         int arg;
1447         int err;
1448
1449         logsink = 1;
1450         dm_init();
1451
1452         if (getuid() != 0) {
1453                 fprintf(stderr, "need to be root\n");
1454                 exit(1);
1455         }
1456
1457         /* make sure we don't lock any path */
1458         chdir("/");
1459         umask(umask(077) | 022);
1460
1461         conf = alloc_config();
1462
1463         if (!conf)
1464                 exit(1);
1465
1466         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1467         switch(arg) {
1468                 case 'd':
1469                         logsink = 0;
1470                         //debug=1; /* ### comment me out ### */
1471                         break;
1472                 case 'v':
1473                         if (sizeof(optarg) > sizeof(char *) ||
1474                             !isdigit(optarg[0]))
1475                                 exit(1);
1476
1477                         conf->verbosity = atoi(optarg);
1478                         break;
1479                 case 'k':
1480                         uxclnt(optarg);
1481                         exit(0);
1482                 default:
1483                         ;
1484                 }
1485         }
1486
1487         if (!logsink)
1488                 err = 0;
1489         else
1490                 err = daemonize();
1491
1492         if (err < 0)
1493                 /* error */
1494                 exit(1);
1495         else if (err > 0)
1496                 /* parent dies */
1497                 exit(0);
1498         else
1499                 /* child lives */
1500                 return (child(NULL));
1501 }
1502