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 int logsink;
  69
  70 /*
  71  * global copy of vecs for use in sig handlers
  72  */
  73 struct vectors * gvecs;
  74
  75 static int
  76 need_switch_pathgroup (struct multipath * mpp, int refresh)
  77 {
  78         struct pathgroup * pgp;
  79         struct path * pp;
  80         unsigned int i, j;
  81
  82         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  83                 return 0;
  84
  85         /*
  86          * Refresh path priority values
  87          */
  88         if (refresh)
  89                 vector_foreach_slot (mpp->pg, pgp, i)
  90                         vector_foreach_slot (pgp->paths, pp, j)
  91                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  92
  93         mpp->bestpg = select_path_group(mpp);
  94
  95         if (mpp->bestpg != mpp->nextpg)
  96                 return 1;
  97
  98         return 0;
  99 }
 100
 101 static void
 102 switch_pathgroup (struct multipath * mpp)
 103 {
 104         mpp->stat_switchgroup++;
 105         dm_switchgroup(mpp->alias, mpp->bestpg);
 106         condlog(2, "%s: switch to path group #%i",
 107                  mpp->alias, mpp->bestpg);
 108 }
 109
 110 static int
 111 coalesce_maps(struct vectors *vecs, vector nmpv)
 112 {
 113         struct multipath * ompp;
 114         vector ompv = vecs->mpvec;
 115         unsigned int i;
 116         int j;
 117
 118         vector_foreach_slot (ompv, ompp, i) {
 119                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 120                         /*
 121                          * remove all current maps not allowed by the
 122                          * current configuration
 123                          */
 124                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 125                                 condlog(0, "%s: unable to flush devmap",
 126                                         ompp->alias);
 127                                 /*
 128                                  * may be just because the device is open
 129                                  */
 130                                 if (!vector_alloc_slot(nmpv))
 131                                         return 1;
 132
 133                                 vector_set_slot(nmpv, ompp);
 134                                 setup_multipath(vecs, ompp);
 135
 136                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 137                                         vector_del_slot(ompv, j);
 138
 139                                 continue;
 140                         }
 141                         else {
 142                                 dm_lib_release();
 143                                 condlog(3, "%s devmap removed", ompp->alias);
 144                         }
 145                 }
 146         }
 147         return 0;
 148 }
 149
 150 static void
 151 sync_map_state(struct multipath *mpp)
 152 {
 153         struct pathgroup *pgp;
 154         struct path *pp;
 155         unsigned int i, j;
 156
 157         vector_foreach_slot (mpp->pg, pgp, i){
 158                 vector_foreach_slot (pgp->paths, pp, j){
 159                         if (pp->state <= PATH_UNCHECKED)
 160                                 continue;
 161                         if ((pp->dmstate == PSTATE_FAILED ||
 162                              pp->dmstate == PSTATE_UNDEF) &&
 163                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 164                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 165                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 166                                   pp->dmstate == PSTATE_UNDEF) &&
 167                                  (pp->state == PATH_DOWN ||
 168                                   pp->state == PATH_SHAKY))
 169                                 dm_fail_path(mpp->alias, pp->dev_t);
 170                 }
 171         }
 172 }
 173
 174 static void
 175 sync_maps_state(vector mpvec)
 176 {
 177         unsigned int i;
 178         struct multipath *mpp;
 179
 180         vector_foreach_slot (mpvec, mpp, i)
 181                 sync_map_state(mpp);
 182 }
 183
 184 static int
 185 flush_map(struct multipath * mpp, struct vectors * vecs)
 186 {
 187         /*
 188          * clear references to this map before flushing so we can ignore
 189          * the spurious uevent we may generate with the dm_flush_map call below
 190          */
 191         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 192                 /*
 193                  * May not really be an error -- if the map was already flushed
 194                  * from the device mapper by dmsetup(8) for instance.
 195                  */
 196                 condlog(0, "%s: can't flush", mpp->alias);
 197                 return 1;
 198         }
 199         else {
 200                 dm_lib_release();
 201                 condlog(3, "%s: devmap removed", mpp->alias);
 202         }
 203
 204         orphan_paths(vecs->pathvec, mpp);
 205         remove_map_and_stop_waiter(mpp, vecs, 1);
 206
 207         return 0;
 208 }
 209
 210 static int
 211 uev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 212 {
 213         condlog(2, "%s: add map (uevent)", dev->kernel);
 214         return ev_add_map(dev, vecs);
 215 }
 216
 217 int
 218 ev_add_map (struct sysfs_device * dev, struct vectors * vecs)
 219 {
 220         char * alias;
 221         char *dev_t;
 222         int major, minor;
 223         char * refwwid;
 224         struct multipath * mpp;
 225         int map_present;
 226         int r = 1;
 227
 228         dev_t = sysfs_attr_get_value(dev->devpath, "dev");
 229
 230         if (!dev_t || sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 231                 return 1;
 232
 233         alias = dm_mapname(major, minor);
 234
 235         if (!alias)
 236                 return 1;
 237
 238         map_present = dm_map_present(alias);
 239
 240         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 241                 condlog(4, "%s: not a multipath map", alias);
 242                 return 0;
 243         }
 244
 245         mpp = find_mp_by_alias(vecs->mpvec, alias);
 246
 247         if (mpp) {
 248                 /*
 249                  * Not really an error -- we generate our own uevent
 250                  * if we create a multipath mapped device as a result
 251                  * of uev_add_path
 252                  */
 253                 condlog(0, "%s: devmap already registered",
 254                         dev->kernel);
 255                 return 0;
 256         }
 257
 258         /*
 259          * now we can register the map
 260          */
 261         if (map_present && (mpp = add_map_without_path(vecs, minor, alias))) {
 262                 sync_map_state(mpp);
 263                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 264                 return 0;
 265         }
 266         refwwid = get_refwwid(dev->kernel, DEV_DEVMAP, vecs->pathvec);
 267
 268         if (refwwid) {
 269                 r = coalesce_paths(vecs, NULL, refwwid);
 270                 dm_lib_release();
 271         }
 272
 273         if (!r)
 274                 condlog(3, "%s: devmap %s added", alias, dev->kernel);
 275         else
 276                 condlog(0, "%s: uev_add_map %s failed", alias, dev->kernel);
 277
 278         FREE(refwwid);
 279         return r;
 280 }
 281
 282 static int
 283 uev_remove_map (struct sysfs_device * dev, struct vectors * vecs)
 284 {
 285         condlog(2, "%s: remove map (uevent)", dev->kernel);
 286         return ev_remove_map(dev->kernel, vecs);
 287 }
 288
 289 int
 290 ev_remove_map (char * devname, struct vectors * vecs)
 291 {
 292         struct multipath * mpp;
 293
 294         mpp = find_mp_by_str(vecs->mpvec, devname);
 295
 296         if (!mpp) {
 297                 condlog(3, "%s: devmap not registered, can't remove",
 298                         devname);
 299                 return 0;
 300         }
 301         flush_map(mpp, vecs);
 302
 303         return 0;
 304 }
 305
 306 static int
 307 uev_umount_map (struct sysfs_device * dev, struct vectors * vecs)
 308 {
 309         struct multipath * mpp;
 310
 311         condlog(2, "%s: umount map (uevent)", dev->kernel);
 312
 313         mpp = find_mp_by_str(vecs->mpvec, dev->kernel);
 314
 315         if (!mpp)
 316                 return 0;
 317
 318         update_mpp_paths(mpp, vecs->pathvec);
 319         verify_paths(mpp, vecs, NULL);
 320
 321         if (!VECTOR_SIZE(mpp->paths))
 322                 flush_map(mpp, vecs);
 323
 324         return 0;
 325 }
 326
 327 static int
 328 uev_add_path (struct sysfs_device * dev, struct vectors * vecs)
 329 {
 330         condlog(2, "%s: add path (uevent)", dev->kernel);
 331         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 332 }
 333
 334
 335 /*
 336  * returns:
 337  * 0: added
 338  * 1: error
 339  * 2: blacklisted
 340  */
 341 int
 342 ev_add_path (char * devname, struct vectors * vecs)
 343 {
 344         struct multipath * mpp;
 345         struct path * pp;
 346         char empty_buff[WWID_SIZE] = {0};
 347
 348         pp = find_path_by_dev(vecs->pathvec, devname);
 349
 350         if (pp) {
 351                 condlog(0, "%s: spurious uevent, path already in pathvec",
 352                         devname);
 353                 if (pp->mpp)
 354                         return 0;
 355         }
 356         else {
 357                 /*
 358                  * get path vital state
 359                  */
 360                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 361                       devname, DI_ALL))) {
 362                         condlog(0, "%s: failed to store path info", devname);
 363                         return 1;
 364                 }
 365                 pp->checkint = conf->checkint;
 366         }
 367
 368         /*
 369          * need path UID to go any further
 370          */
 371         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 372                 condlog(0, "%s: failed to get path uid", devname);
 373                 return 1; /* leave path added to pathvec */
 374         }
 375         if (filter_path(conf, pp) > 0){
 376                 int i = find_slot(vecs->pathvec, (void *)pp);
 377                 if (i != -1)
 378                         vector_del_slot(vecs->pathvec, i);
 379                 free_path(pp);
 380                 return 2;
 381         }
 382         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 383 rescan:
 384         if (mpp) {
 385                 if (adopt_paths(vecs->pathvec, mpp))
 386                         return 1; /* leave path added to pathvec */
 387
 388                 verify_paths(mpp, vecs, NULL);
 389                 mpp->action = ACT_RELOAD;
 390         }
 391         else {
 392                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 393                         mpp->action = ACT_CREATE;
 394                 else
 395                         return 1; /* leave path added to pathvec */
 396         }
 397
 398         /*
 399          * push the map to the device-mapper
 400          */
 401         if (setup_map(mpp)) {
 402                 condlog(0, "%s: failed to setup map for addition of new "
 403                         "path %s", mpp->alias, devname);
 404                 goto out;
 405         }
 406         /*
 407          * reload the map for the multipath mapped device
 408          */
 409         if (domap(mpp) <= 0) {
 410                 condlog(0, "%s: failed in domap for addition of new "
 411                         "path %s", mpp->alias, devname);
 412                 /*
 413                  * deal with asynchronous uevents :((
 414                  */
 415                 if (mpp->action == ACT_RELOAD) {
 416                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 417                         sleep(1);
 418                         update_mpp_paths(mpp, vecs->pathvec);
 419                         goto rescan;
 420                 }
 421                 else
 422                         goto out;
 423         }
 424         dm_lib_release();
 425
 426         /*
 427          * update our state from kernel regardless of create or reload
 428          */
 429         if (setup_multipath(vecs, mpp))
 430                 goto out;
 431
 432         sync_map_state(mpp);
 433
 434         if (mpp->action == ACT_CREATE &&
 435             start_waiter_thread(mpp, vecs))
 436                         goto out;
 437
 438         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 439         return 0;
 440
 441 out:
 442         remove_map(mpp, vecs, 1);
 443         return 1;
 444 }
 445
 446 static int
 447 uev_remove_path (struct sysfs_device * dev, struct vectors * vecs)
 448 {
 449         int retval;
 450
 451         condlog(2, "%s: remove path (uevent)", dev->kernel);
 452         retval = ev_remove_path(dev->kernel, vecs);
 453         if (!retval)
 454                 sysfs_device_put(dev);
 455
 456         return retval;
 457 }
 458
 459 int
 460 ev_remove_path (char * devname, struct vectors * vecs)
 461 {
 462         struct multipath * mpp;
 463         struct path * pp;
 464         int i;
 465         int rm_path = 1;
 466
 467         pp = find_path_by_dev(vecs->pathvec, devname);
 468
 469         if (!pp) {
 470                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 471                 return 1;
 472         }
 473
 474         /*
 475          * avoid referring to the map of an orphanned path
 476          */
 477         if ((mpp = pp->mpp)) {
 478
 479                 /*
 480                  * remove the map IFF removing the last path
 481                  */
 482                 if (pathcount(mpp, PATH_WILD) > 1) {
 483                         vector rpvec = vector_alloc();
 484
 485                         /*
 486                          * transform the mp->pg vector of vectors of paths
 487                          * into a mp->params string to feed the device-mapper
 488                          */
 489                         update_mpp_paths(mpp, vecs->pathvec);
 490                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 491                                 vector_del_slot(mpp->paths, i);
 492
 493                         if (VECTOR_SIZE(mpp->paths) == 0) {
 494                                 char alias[WWID_SIZE];
 495
 496                                 /*
 497                                  * flush_map will fail if the device is open
 498                                  */
 499                                 strncpy(alias, mpp->alias, WWID_SIZE);
 500                                 if (flush_map(mpp, vecs))
 501                                         rm_path = 0;
 502                                 else
 503                                         condlog(3, "%s: removed map after removing"
 504                                                 " multiple paths", alias);
 505                         }
 506                         else {
 507                                 if (setup_map(mpp)) {
 508                                         condlog(0, "%s: failed to setup map for"
 509                                                 " removal of path %s", mpp->alias, devname);
 510                                         free_pathvec(rpvec, KEEP_PATHS);
 511                                         goto out;
 512                                 }
 513                                 /*
 514                                  * reload the map
 515                                  */
 516                                 mpp->action = ACT_RELOAD;
 517                                 if (domap(mpp) <= 0) {
 518                                         condlog(0, "%s: failed in domap for "
 519                                                 "removal of path %s",
 520                                                 mpp->alias, devname);
 521                                         /*
 522                                          * Delete path from pathvec so that
 523                                          * update_mpp_paths wont find it later
 524                                          * when/if another path is removed.
 525                                          */
 526                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 527                                                 vector_del_slot(vecs->pathvec, i);
 528                                         free_path(pp);
 529                                         return 1;
 530                                 }
 531                                 /*
 532                                  * update our state from kernel
 533                                  */
 534                                 if (setup_multipath(vecs, mpp)) {
 535                                         free_pathvec(rpvec, KEEP_PATHS);
 536                                         goto out;
 537                                 }
 538                                 sync_map_state(mpp);
 539
 540                                 condlog(3, "%s: path removed from map %s",
 541                                         devname, mpp->alias);
 542                         }
 543                         free_pathvec(rpvec, KEEP_PATHS);
 544                 }
 545                 else {
 546                         char alias[WWID_SIZE];
 547
 548                         /*
 549                          * flush_map will fail if the device is open
 550                          */
 551                         strncpy(alias, mpp->alias, WWID_SIZE);
 552                         if (flush_map(mpp, vecs))
 553                                 rm_path = 0;
 554                         else
 555                                 condlog(3, "%s: removed map", alias);
 556                 }
 557         }
 558
 559         if (rm_path) {
 560                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 561                         vector_del_slot(vecs->pathvec, i);
 562                 free_path(pp);
 563         }
 564
 565         return 0;
 566
 567 out:
 568         remove_map_and_stop_waiter(mpp, vecs, 1);
 569         return 1;
 570 }
 571
 572 static int
 573 map_discovery (struct vectors * vecs)
 574 {
 575         struct multipath * mpp;
 576         unsigned int i;
 577
 578         if (dm_get_maps(vecs->mpvec, "multipath"))
 579                 return 1;
 580
 581         vector_foreach_slot (vecs->mpvec, mpp, i)
 582                 if (setup_multipath(vecs, mpp))
 583                         return 1;
 584
 585         return 0;
 586 }
 587
 588 int
 589 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 590 {
 591         struct vectors * vecs;
 592         int r;
 593
 594         *reply = NULL;
 595         *len = 0;
 596         vecs = (struct vectors *)trigger_data;
 597
 598         pthread_cleanup_push(cleanup_lock, vecs->lock);
 599         lock(vecs->lock);
 600
 601         r = parse_cmd(str, reply, len, vecs);
 602
 603         if (r > 0) {
 604                 *reply = STRDUP("fail\n");
 605                 *len = strlen(*reply) + 1;
 606                 r = 1;
 607         }
 608         else if (!r && *len == 0) {
 609                 *reply = STRDUP("ok\n");
 610                 *len = strlen(*reply) + 1;
 611                 r = 0;
 612         }
 613         /* else if (r < 0) leave *reply alone */
 614
 615         lock_cleanup_pop(vecs->lock);
 616         return r;
 617 }
 618
 619 static int
 620 uev_discard(char * devpath)
 621 {
 622         char a[10], b[10];
 623
 624         /*
 625          * keep only block devices, discard partitions
 626          */
 627         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 628             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 629                 condlog(4, "discard event on %s", devpath);
 630                 return 1;
 631         }
 632         return 0;
 633 }
 634
 635 int
 636 uev_trigger (struct uevent * uev, void * trigger_data)
 637 {
 638         int r = 0;
 639         struct sysfs_device *sysdev;
 640         struct vectors * vecs;
 641
 642         vecs = (struct vectors *)trigger_data;
 643
 644         if (uev_discard(uev->devpath))
 645                 return 0;
 646
 647         sysdev = sysfs_device_get(uev->devpath);
 648         if(!sysdev)
 649                 return 0;
 650
 651         lock(vecs->lock);
 652
 653         /*
 654          * device map event
 655          * Add events are ignored here as the tables
 656          * are not fully initialised then.
 657          */
 658         if (!strncmp(sysdev->kernel, "dm-", 3)) {
 659                 if (!strncmp(uev->action, "change", 6)) {
 660                         r = uev_add_map(sysdev, vecs);
 661                         goto out;
 662                 }
 663                 if (!strncmp(uev->action, "remove", 6)) {
 664                         r = uev_remove_map(sysdev, vecs);
 665                         goto out;
 666                 }
 667                 if (!strncmp(uev->action, "umount", 6)) {
 668                         r = uev_umount_map(sysdev, vecs);
 669                         goto out;
 670                 }
 671                 goto out;
 672         }
 673
 674         /*
 675          * path add/remove event
 676          */
 677         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 678                            sysdev->kernel) > 0)
 679                 goto out;
 680
 681         if (!strncmp(uev->action, "add", 3)) {
 682                 r = uev_add_path(sysdev, vecs);
 683                 goto out;
 684         }
 685         if (!strncmp(uev->action, "remove", 6)) {
 686                 r = uev_remove_path(sysdev, vecs);
 687                 goto out;
 688         }
 689
 690 out:
 691         unlock(vecs->lock);
 692         return r;
 693 }
 694
 695 static void *
 696 ueventloop (void * ap)
 697 {
 698         if (uevent_listen(&uev_trigger, ap))
 699                 fprintf(stderr, "error starting uevent listener");
 700
 701         return NULL;
 702 }
 703
 704 static void *
 705 uxlsnrloop (void * ap)
 706 {
 707         if (cli_init())
 708                 return NULL;
 709
 710         set_handler_callback(LIST+PATHS, cli_list_paths);
 711         set_handler_callback(LIST+MAPS, cli_list_maps);
 712         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 713         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 714         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 715         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 716         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 717         set_handler_callback(LIST+CONFIG, cli_list_config);
 718         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 719         set_handler_callback(LIST+DEVICES, cli_list_devices);
 720         set_handler_callback(ADD+PATH, cli_add_path);
 721         set_handler_callback(DEL+PATH, cli_del_path);
 722         set_handler_callback(ADD+MAP, cli_add_map);
 723         set_handler_callback(DEL+MAP, cli_del_map);
 724         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 725         set_handler_callback(RECONFIGURE, cli_reconfigure);
 726         set_handler_callback(SUSPEND+MAP, cli_suspend);
 727         set_handler_callback(RESUME+MAP, cli_resume);
 728         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 729         set_handler_callback(FAIL+PATH, cli_fail);
 730
 731         uxsock_listen(&uxsock_trigger, ap);
 732
 733         return NULL;
 734 }
 735
 736 static int
 737 exit_daemon (int status)
 738 {
 739         if (status != 0)
 740                 fprintf(stderr, "bad exit status. see daemon.log\n");
 741
 742         condlog(3, "unlink pidfile");
 743         unlink(DEFAULT_PIDFILE);
 744
 745         lock(&exit_mutex);
 746         pthread_cond_signal(&exit_cond);
 747         unlock(&exit_mutex);
 748
 749         return status;
 750 }
 751
 752 static void
 753 fail_path (struct path * pp, int del_active)
 754 {
 755         if (!pp->mpp)
 756                 return;
 757
 758         condlog(2, "checker failed path %s in map %s",
 759                  pp->dev_t, pp->mpp->alias);
 760
 761         dm_fail_path(pp->mpp->alias, pp->dev_t);
 762         if (del_active)
 763                 update_queue_mode_del_path(pp->mpp);
 764 }
 765
 766 /*
 767  * caller must have locked the path list before calling that function
 768  */
 769 static void
 770 reinstate_path (struct path * pp, int add_active)
 771 {
 772         if (!pp->mpp)
 773                 return;
 774
 775         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 776                 condlog(0, "%s: reinstate failed", pp->dev_t);
 777         else {
 778                 condlog(2, "%s: reinstated", pp->dev_t);
 779                 if (add_active)
 780                         update_queue_mode_add_path(pp->mpp);
 781         }
 782 }
 783
 784 static void
 785 enable_group(struct path * pp)
 786 {
 787         struct pathgroup * pgp;
 788
 789         /*
 790          * if path is added through uev_add_path, pgindex can be unset.
 791          * next update_strings() will set it, upon map reload event.
 792          *
 793          * we can safely return here, because upon map reload, all
 794          * PG will be enabled.
 795          */
 796         if (!pp->mpp->pg || !pp->pgindex)
 797                 return;
 798
 799         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 800
 801         if (pgp->status == PGSTATE_DISABLED) {
 802                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 803                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 804         }
 805 }
 806
 807 static void
 808 mpvec_garbage_collector (struct vectors * vecs)
 809 {
 810         struct multipath * mpp;
 811         unsigned int i;
 812
 813         vector_foreach_slot (vecs->mpvec, mpp, i) {
 814                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 815                         condlog(2, "%s: remove dead map", mpp->alias);
 816                         remove_map_and_stop_waiter(mpp, vecs, 1);
 817                         i--;
 818                 }
 819         }
 820 }
 821
 822 static void
 823 defered_failback_tick (vector mpvec)
 824 {
 825         struct multipath * mpp;
 826         unsigned int i;
 827
 828         vector_foreach_slot (mpvec, mpp, i) {
 829                 /*
 830                  * defered failback getting sooner
 831                  */
 832                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 833                         mpp->failback_tick--;
 834
 835                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 836                                 switch_pathgroup(mpp);
 837                 }
 838         }
 839 }
 840
 841 static void
 842 retry_count_tick(vector mpvec)
 843 {
 844         struct multipath *mpp;
 845         unsigned int i;
 846
 847         vector_foreach_slot (mpvec, mpp, i) {
 848                 if (mpp->retry_tick) {
 849                         mpp->stat_total_queueing_time++;
 850                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 851                         if(--mpp->retry_tick == 0) {
 852                                 dm_queue_if_no_path(mpp->alias, 0);
 853                                 condlog(2, "%s: Disable queueing", mpp->alias);
 854                         }
 855                 }
 856         }
 857 }
 858
 859 void
 860 check_path (struct vectors * vecs, struct path * pp)
 861 {
 862         int newstate;
 863
 864         if (!pp->mpp)
 865                 return;
 866
 867         if (pp->tick && --pp->tick)
 868                 return; /* don't check this path yet */
 869
 870         /*
 871          * provision a next check soonest,
 872          * in case we exit abnormaly from here
 873          */
 874         pp->tick = conf->checkint;
 875
 876         if (!checker_selected(&pp->checker)) {
 877                 pathinfo(pp, conf->hwtable, DI_SYSFS);
 878                 select_checker(pp);
 879         }
 880         if (!checker_selected(&pp->checker)) {
 881                 condlog(0, "%s: checker is not set", pp->dev);
 882                 return;
 883         }
 884         /*
 885          * Set checker in async mode.
 886          * Honored only by checker implementing the said mode.
 887          */
 888         checker_set_async(&pp->checker);
 889
 890         newstate = checker_check(&pp->checker);
 891
 892         if (newstate < 0) {
 893                 condlog(2, "%s: unusable path", pp->dev);
 894                 pathinfo(pp, conf->hwtable, 0);
 895                 return;
 896         }
 897         /*
 898          * Async IO in flight. Keep the previous path state
 899          * and reschedule as soon as possible
 900          */
 901         if (newstate == PATH_PENDING) {
 902                 pp->tick = 1;
 903                 return;
 904         }
 905         if (newstate != pp->state) {
 906                 int oldstate = pp->state;
 907                 pp->state = newstate;
 908                 LOG_MSG(1, checker_message(&pp->checker));
 909
 910                 /*
 911                  * upon state change, reset the checkint
 912                  * to the shortest delay
 913                  */
 914                 pp->checkint = conf->checkint;
 915
 916                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY ||
 917                     update_multipath_strings(pp->mpp, vecs->pathvec)) {
 918                         /*
 919                          * proactively fail path in the DM
 920                          */
 921                         if (oldstate == PATH_UP ||
 922                             oldstate == PATH_GHOST)
 923                                 fail_path(pp, 1);
 924                         else
 925                                 fail_path(pp, 0);
 926
 927                         /*
 928                          * cancel scheduled failback
 929                          */
 930                         pp->mpp->failback_tick = 0;
 931
 932                         pp->mpp->stat_path_failures++;
 933                         return;
 934                 }
 935
 936                 /*
 937                  * reinstate this path
 938                  */
 939                 if (oldstate != PATH_UP &&
 940                     oldstate != PATH_GHOST)
 941                         reinstate_path(pp, 1);
 942                 else
 943                         reinstate_path(pp, 0);
 944
 945                 /*
 946                  * schedule [defered] failback
 947                  */
 948                 if (pp->mpp->pgfailback > 0)
 949                         pp->mpp->failback_tick =
 950                                 pp->mpp->pgfailback + 1;
 951                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
 952                     need_switch_pathgroup(pp->mpp, 1))
 953                         switch_pathgroup(pp->mpp);
 954
 955                 /*
 956                  * if at least one path is up in a group, and
 957                  * the group is disabled, re-enable it
 958                  */
 959                 if (newstate == PATH_UP)
 960                         enable_group(pp);
 961         }
 962         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
 963                 LOG_MSG(4, checker_message(&pp->checker));
 964                 /*
 965                  * double the next check delay.
 966                  * max at conf->max_checkint
 967                  */
 968                 if (pp->checkint < (conf->max_checkint / 2))
 969                         pp->checkint = 2 * pp->checkint;
 970                 else
 971                         pp->checkint = conf->max_checkint;
 972
 973                 pp->tick = pp->checkint;
 974                 condlog(4, "%s: delay next check %is",
 975                                 pp->dev_t, pp->tick);
 976         }
 977         else if (newstate == PATH_DOWN)
 978                 LOG_MSG(2, checker_message(&pp->checker));
 979
 980         pp->state = newstate;
 981
 982         /*
 983          * path prio refreshing
 984          */
 985         condlog(4, "path prio refresh");
 986         pathinfo(pp, conf->hwtable, DI_PRIO);
 987
 988         /*
 989          * pathgroup failback policy
 990          */
 991         if (need_switch_pathgroup(pp->mpp, 0)) {
 992                 if (pp->mpp->pgfailback > 0 &&
 993                     pp->mpp->failback_tick <= 0)
 994                         pp->mpp->failback_tick =
 995                                 pp->mpp->pgfailback + 1;
 996                 else if (pp->mpp->pgfailback ==
 997                                 -FAILBACK_IMMEDIATE)
 998                         switch_pathgroup(pp->mpp);
 999         }
1000 }
1001
1002 static void *
1003 checkerloop (void *ap)
1004 {
1005         struct vectors *vecs;
1006         struct path *pp;
1007         int count = 0;
1008         unsigned int i;
1009
1010         mlockall(MCL_CURRENT | MCL_FUTURE);
1011         vecs = (struct vectors *)ap;
1012         condlog(2, "path checkers start up");
1013
1014         /*
1015          * init the path check interval
1016          */
1017         vector_foreach_slot (vecs->pathvec, pp, i) {
1018                 pp->checkint = conf->checkint;
1019         }
1020
1021         while (1) {
1022                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1023                 lock(vecs->lock);
1024                 condlog(4, "tick");
1025
1026                 vector_foreach_slot (vecs->pathvec, pp, i) {
1027                         check_path(vecs, pp);
1028                 }
1029                 defered_failback_tick(vecs->mpvec);
1030                 retry_count_tick(vecs->mpvec);
1031
1032                 if (count)
1033                         count--;
1034                 else {
1035                         condlog(4, "map garbage collection");
1036                         mpvec_garbage_collector(vecs);
1037                         count = MAPGCINT;
1038                 }
1039
1040                 lock_cleanup_pop(vecs->lock);
1041                 sleep(1);
1042         }
1043         return NULL;
1044 }
1045
1046 int
1047 configure (struct vectors * vecs, int start_waiters)
1048 {
1049         struct multipath * mpp;
1050         struct path * pp;
1051         vector mpvec;
1052         int i;
1053
1054         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1055                 return 1;
1056
1057         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1058                 return 1;
1059
1060         if (!(mpvec = vector_alloc()))
1061                 return 1;
1062
1063         /*
1064          * probe for current path (from sysfs) and map (from dm) sets
1065          */
1066         path_discovery(vecs->pathvec, conf, DI_ALL);
1067
1068         vector_foreach_slot (vecs->pathvec, pp, i){
1069                 if (filter_path(conf, pp) > 0){
1070                         vector_del_slot(vecs->pathvec, i);
1071                         free_path(pp);
1072                         i--;
1073                 }
1074                 else
1075                         pp->checkint = conf->checkint;
1076         }
1077         if (map_discovery(vecs))
1078                 return 1;
1079
1080         /*
1081          * create new set of maps & push changed ones into dm
1082          */
1083         if (coalesce_paths(vecs, mpvec, NULL))
1084                 return 1;
1085
1086         /*
1087          * may need to remove some maps which are no longer relevant
1088          * e.g., due to blacklist changes in conf file
1089          */
1090         if (coalesce_maps(vecs, mpvec))
1091                 return 1;
1092
1093         dm_lib_release();
1094
1095         sync_maps_state(mpvec);
1096
1097         /*
1098          * purge dm of old maps
1099          */
1100         remove_maps(vecs);
1101
1102         /*
1103          * save new set of maps formed by considering current path state
1104          */
1105         vector_free(vecs->mpvec);
1106         vecs->mpvec = mpvec;
1107
1108         /*
1109          * start dm event waiter threads for these new maps
1110          */
1111         vector_foreach_slot(vecs->mpvec, mpp, i) {
1112                 if (setup_multipath(vecs, mpp))
1113                         return 1;
1114                 if (start_waiters)
1115                         if (start_waiter_thread(mpp, vecs))
1116                                 return 1;
1117         }
1118         return 0;
1119 }
1120
1121 int
1122 reconfigure (struct vectors * vecs)
1123 {
1124         struct config * old = conf;
1125
1126         /*
1127          * free old map and path vectors ... they use old conf state
1128          */
1129         if (VECTOR_SIZE(vecs->mpvec))
1130                 remove_maps_and_stop_waiters(vecs);
1131
1132         if (VECTOR_SIZE(vecs->pathvec))
1133                 free_pathvec(vecs->pathvec, FREE_PATHS);
1134
1135         vecs->pathvec = NULL;
1136         conf = NULL;
1137
1138         if (load_config(DEFAULT_CONFIGFILE))
1139                 return 1;
1140
1141         conf->verbosity = old->verbosity;
1142
1143         if (!conf->checkint) {
1144                 conf->checkint = DEFAULT_CHECKINT;
1145                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1146         }
1147         configure(vecs, 1);
1148         free_config(old);
1149         return 0;
1150 }
1151
1152 static struct vectors *
1153 init_vecs (void)
1154 {
1155         struct vectors * vecs;
1156
1157         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1158
1159         if (!vecs)
1160                 return NULL;
1161
1162         vecs->lock =
1163                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1164
1165         if (!vecs->lock)
1166                 goto out;
1167
1168         pthread_mutex_init(vecs->lock, NULL);
1169
1170         return vecs;
1171
1172 out:
1173         FREE(vecs);
1174         condlog(0, "failed to init paths");
1175         return NULL;
1176 }
1177
1178 static void *
1179 signal_set(int signo, void (*func) (int))
1180 {
1181         int r;
1182         struct sigaction sig;
1183         struct sigaction osig;
1184
1185         sig.sa_handler = func;
1186         sigemptyset(&sig.sa_mask);
1187         sig.sa_flags = 0;
1188
1189         r = sigaction(signo, &sig, &osig);
1190
1191         if (r < 0)
1192                 return (SIG_ERR);
1193         else
1194                 return (osig.sa_handler);
1195 }
1196
1197 static void
1198 sighup (int sig)
1199 {
1200         condlog(2, "reconfigure (SIGHUP)");
1201
1202         lock(gvecs->lock);
1203         reconfigure(gvecs);
1204         unlock(gvecs->lock);
1205
1206 #ifdef _DEBUG_
1207         dbg_free_final(NULL);
1208 #endif
1209 }
1210
1211 static void
1212 sigend (int sig)
1213 {
1214         exit_daemon(0);
1215 }
1216
1217 static void
1218 sigusr1 (int sig)
1219 {
1220         condlog(3, "SIGUSR1 received");
1221 }
1222
1223 static void
1224 signal_init(void)
1225 {
1226         signal_set(SIGHUP, sighup);
1227         signal_set(SIGUSR1, sigusr1);
1228         signal_set(SIGINT, sigend);
1229         signal_set(SIGTERM, sigend);
1230         signal(SIGPIPE, SIG_IGN);
1231 }
1232
1233 static void
1234 setscheduler (void)
1235 {
1236         int res;
1237         static struct sched_param sched_param = {
1238                 .sched_priority = 99
1239         };
1240
1241         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1242
1243         if (res == -1)
1244                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1245         return;
1246 }
1247
1248 static void
1249 set_oom_adj (int val)
1250 {
1251         FILE *fp;
1252
1253         fp = fopen("/proc/self/oom_adj", "w");
1254
1255         if (!fp)
1256                 return;
1257
1258         fprintf(fp, "%i", val);
1259         fclose(fp);
1260 }
1261
1262 static int
1263 child (void * param)
1264 {
1265         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1266         pthread_attr_t attr;
1267         struct vectors * vecs;
1268
1269         mlockall(MCL_CURRENT | MCL_FUTURE);
1270
1271         if (logsink)
1272                 log_thread_start();
1273
1274         condlog(2, "--------start up--------");
1275         condlog(2, "read " DEFAULT_CONFIGFILE);
1276
1277         if (load_config(DEFAULT_CONFIGFILE))
1278                 exit(1);
1279
1280         if (init_checkers()) {
1281                 condlog(0, "failed to initialize checkers");
1282                 exit(1);
1283         }
1284         if (init_prio()) {
1285                 condlog(0, "failed to initialize prioritizers");
1286                 exit(1);
1287         }
1288
1289         setlogmask(LOG_UPTO(conf->verbosity + 3));
1290
1291         /*
1292          * fill the voids left in the config file
1293          */
1294         if (!conf->checkint) {
1295                 conf->checkint = DEFAULT_CHECKINT;
1296                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1297         }
1298
1299         if (conf->max_fds) {
1300                 struct rlimit fd_limit;
1301                 if (conf->max_fds > 0) {
1302                         fd_limit.rlim_cur = conf->max_fds;
1303                         fd_limit.rlim_max = conf->max_fds;
1304                 }
1305                 else {
1306                         fd_limit.rlim_cur = RLIM_INFINITY;
1307                         fd_limit.rlim_max = RLIM_INFINITY;
1308                 }
1309                 if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0)
1310                         condlog(0, "can't set open fds limit to %d : %s\n",
1311                                 conf->max_fds, strerror(errno));
1312         }
1313
1314         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1315                 if (logsink)
1316                         log_thread_stop();
1317
1318                 exit(1);
1319         }
1320         signal_init();
1321         setscheduler();
1322         set_oom_adj(-16);
1323         vecs = gvecs = init_vecs();
1324
1325         if (!vecs)
1326                 exit(1);
1327
1328         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1329                 condlog(0, "can not find sysfs mount point");
1330                 exit(1);
1331         }
1332
1333         /*
1334          * fetch and configure both paths and multipaths
1335          */
1336         if (configure(vecs, 1)) {
1337                 condlog(0, "failure during configuration");
1338                 exit(1);
1339         }
1340
1341         /*
1342          * start threads
1343          */
1344         pthread_attr_init(&attr);
1345         pthread_attr_setstacksize(&attr, 64 * 1024);
1346         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1347
1348         pthread_create(&check_thr, &attr, checkerloop, vecs);
1349         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1350         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1351
1352         pthread_cond_wait(&exit_cond, &exit_mutex);
1353
1354         /*
1355          * exit path
1356          */
1357         lock(vecs->lock);
1358         remove_maps_and_stop_waiters(vecs);
1359         free_pathvec(vecs->pathvec, FREE_PATHS);
1360
1361         pthread_cancel(check_thr);
1362         pthread_cancel(uevent_thr);
1363         pthread_cancel(uxlsnr_thr);
1364
1365         sysfs_cleanup();
1366
1367         free_keys(keys);
1368         keys = NULL;
1369         free_handlers(handlers);
1370         handlers = NULL;
1371         free_polls();
1372
1373         unlock(vecs->lock);
1374         pthread_mutex_destroy(vecs->lock);
1375         FREE(vecs->lock);
1376         vecs->lock = NULL;
1377         FREE(vecs);
1378         vecs = NULL;
1379
1380         condlog(2, "--------shut down-------");
1381
1382         if (logsink)
1383                 log_thread_stop();
1384
1385         dm_lib_release();
1386         dm_lib_exit();
1387
1388         /*
1389          * Freeing config must be done after condlog() and dm_lib_exit(),
1390          * because logging functions like dlog() and dm_write_log()
1391          * reference the config.
1392          */
1393         free_config(conf);
1394         conf = NULL;
1395
1396 #ifdef _DEBUG_
1397         dbg_free_final(NULL);
1398 #endif
1399
1400         exit(0);
1401 }
1402
1403 static int
1404 daemonize(void)
1405 {
1406         int pid;
1407         int in_fd, out_fd;
1408
1409         if( (pid = fork()) < 0){
1410                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1411                 return -1;
1412         }
1413         else if (pid != 0)
1414                 return pid;
1415
1416         setsid();
1417
1418         if ( (pid = fork()) < 0)
1419                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1420         else if (pid != 0)
1421                 _exit(0);
1422
1423         in_fd = open("/dev/null", O_RDONLY);
1424         if (in_fd < 0){
1425                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1426                         strerror(errno));
1427                 _exit(0);
1428         }
1429         out_fd = open("/dev/console", O_WRONLY);
1430         if (out_fd < 0){
1431                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1432                         strerror(errno));
1433                 _exit(0);
1434         }
1435
1436         close(STDIN_FILENO);
1437         dup(in_fd);
1438         close(STDOUT_FILENO);
1439         dup(out_fd);
1440         close(STDERR_FILENO);
1441         dup(out_fd);
1442
1443         close(in_fd);
1444         close(out_fd);
1445         chdir("/");
1446         umask(0);
1447         return 0;
1448 }
1449
1450 int
1451 main (int argc, char *argv[])
1452 {
1453         extern char *optarg;
1454         extern int optind;
1455         int arg;
1456         int err;
1457
1458         logsink = 1;
1459         dm_init();
1460
1461         if (getuid() != 0) {
1462                 fprintf(stderr, "need to be root\n");
1463                 exit(1);
1464         }
1465
1466         /* make sure we don't lock any path */
1467         chdir("/");
1468         umask(umask(077) | 022);
1469
1470         conf = alloc_config();
1471
1472         if (!conf)
1473                 exit(1);
1474
1475         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1476         switch(arg) {
1477                 case 'd':
1478                         logsink = 0;
1479                         //debug=1; /* ### comment me out ### */
1480                         break;
1481                 case 'v':
1482                         if (sizeof(optarg) > sizeof(char *) ||
1483                             !isdigit(optarg[0]))
1484                                 exit(1);
1485
1486                         conf->verbosity = atoi(optarg);
1487                         break;
1488                 case 'k':
1489                         uxclnt(optarg);
1490                         exit(0);
1491                 default:
1492                         ;
1493                 }
1494         }
1495
1496         if (!logsink)
1497                 err = 0;
1498         else
1499                 err = daemonize();
1500
1501         if (err < 0)
1502                 /* error */
1503                 exit(1);
1504         else if (err > 0)
1505                 /* parent dies */
1506                 exit(0);
1507         else
1508                 /* child lives */
1509                 return (child(NULL));
1510 }
1511