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 #include <limits.h>
  18 #include <linux/oom.h>
  19 #include <libudev.h>
  20 #include <mpath_persist.h>
  21
  22 /*
  23  * libcheckers
  24  */
  25 #include <checkers.h>
  26
  27 /*
  28  * libmultipath
  29  */
  30 #include <parser.h>
  31 #include <vector.h>
  32 #include <memory.h>
  33 #include <config.h>
  34 #include <util.h>
  35 #include <hwtable.h>
  36 #include <defaults.h>
  37 #include <structs.h>
  38 #include <callout.h>
  39 #include <blacklist.h>
  40 #include <structs_vec.h>
  41 #include <dmparser.h>
  42 #include <devmapper.h>
  43 #include <sysfs.h>
  44 #include <dict.h>
  45 #include <discovery.h>
  46 #include <debug.h>
  47 #include <propsel.h>
  48 #include <uevent.h>
  49 #include <switchgroup.h>
  50 #include <print.h>
  51 #include <configure.h>
  52 #include <prio.h>
  53 #include <pgpolicies.h>
  54 #include <uevent.h>
  55
  56 #include "main.h"
  57 #include "pidfile.h"
  58 #include "uxlsnr.h"
  59 #include "uxclnt.h"
  60 #include "cli.h"
  61 #include "cli_handlers.h"
  62 #include "lock.h"
  63 #include "waiter.h"
  64
  65 #define FILE_NAME_SIZE 256
  66 #define CMDSIZE 160
  67
  68 #define LOG_MSG(a, b) \
  69 do { \
  70         if (pp->offline) \
  71                 condlog(a, "%s: %s - path offline", pp->mpp->alias, pp->dev); \
  72         else if (strlen(b)) \
  73                 condlog(a, "%s: %s - %s", pp->mpp->alias, pp->dev, b); \
  74 } while(0)
  75
  76 struct mpath_event_param
  77 {
  78         char * devname;
  79         struct multipath *mpp;
  80 };
  81
  82 unsigned int mpath_mx_alloc_len;
  83
  84 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  85 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  86
  87 int logsink;
  88 enum daemon_status running_state;
  89 pid_t daemon_pid;
  90
  91 /*
  92  * global copy of vecs for use in sig handlers
  93  */
  94 struct vectors * gvecs;
  95
  96 static int
  97 need_switch_pathgroup (struct multipath * mpp, int refresh)
  98 {
  99         struct pathgroup * pgp;
 100         struct path * pp;
 101         unsigned int i, j;
 102
 103         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 104                 return 0;
 105
 106         /*
 107          * Refresh path priority values
 108          */
 109         if (refresh)
 110                 vector_foreach_slot (mpp->pg, pgp, i)
 111                         vector_foreach_slot (pgp->paths, pp, j)
 112                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 113
 114         mpp->bestpg = select_path_group(mpp);
 115
 116         if (mpp->bestpg != mpp->nextpg)
 117                 return 1;
 118
 119         return 0;
 120 }
 121
 122 static void
 123 switch_pathgroup (struct multipath * mpp)
 124 {
 125         mpp->stat_switchgroup++;
 126         dm_switchgroup(mpp->alias, mpp->bestpg);
 127         condlog(2, "%s: switch to path group #%i",
 128                  mpp->alias, mpp->bestpg);
 129 }
 130
 131 static int
 132 coalesce_maps(struct vectors *vecs, vector nmpv)
 133 {
 134         struct multipath * ompp;
 135         vector ompv = vecs->mpvec;
 136         unsigned int i;
 137         int j;
 138
 139         vector_foreach_slot (ompv, ompp, i) {
 140                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 141                         /*
 142                          * remove all current maps not allowed by the
 143                          * current configuration
 144                          */
 145                         if (dm_flush_map(ompp->alias)) {
 146                                 condlog(0, "%s: unable to flush devmap",
 147                                         ompp->alias);
 148                                 /*
 149                                  * may be just because the device is open
 150                                  */
 151                                 if (!vector_alloc_slot(nmpv))
 152                                         return 1;
 153
 154                                 vector_set_slot(nmpv, ompp);
 155                                 setup_multipath(vecs, ompp);
 156
 157                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 158                                         vector_del_slot(ompv, j);
 159
 160                                 continue;
 161                         }
 162                         else {
 163                                 dm_lib_release();
 164                                 condlog(2, "%s devmap removed", ompp->alias);
 165                         }
 166                 } else if (conf->reassign_maps) {
 167                         condlog(3, "%s: Reassign existing device-mapper"
 168                                 " devices", ompp->alias);
 169                         dm_reassign(ompp->alias);
 170                 }
 171         }
 172         return 0;
 173 }
 174
 175 void
 176 sync_map_state(struct multipath *mpp)
 177 {
 178         struct pathgroup *pgp;
 179         struct path *pp;
 180         unsigned int i, j;
 181
 182         if (!mpp->pg)
 183                 return;
 184
 185         vector_foreach_slot (mpp->pg, pgp, i){
 186                 vector_foreach_slot (pgp->paths, pp, j){
 187                         if (pp->state == PATH_UNCHECKED ||
 188                             pp->state == PATH_WILD)
 189                                 continue;
 190                         if ((pp->dmstate == PSTATE_FAILED ||
 191                              pp->dmstate == PSTATE_UNDEF) &&
 192                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 193                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 194                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 195                                   pp->dmstate == PSTATE_UNDEF) &&
 196                                  (pp->state == PATH_DOWN ||
 197                                   pp->state == PATH_SHAKY))
 198                                 dm_fail_path(mpp->alias, pp->dev_t);
 199                 }
 200         }
 201 }
 202
 203 static void
 204 sync_maps_state(vector mpvec)
 205 {
 206         unsigned int i;
 207         struct multipath *mpp;
 208
 209         vector_foreach_slot (mpvec, mpp, i)
 210                 sync_map_state(mpp);
 211 }
 212
 213 static int
 214 flush_map(struct multipath * mpp, struct vectors * vecs)
 215 {
 216         /*
 217          * clear references to this map before flushing so we can ignore
 218          * the spurious uevent we may generate with the dm_flush_map call below
 219          */
 220         if (dm_flush_map(mpp->alias)) {
 221                 /*
 222                  * May not really be an error -- if the map was already flushed
 223                  * from the device mapper by dmsetup(8) for instance.
 224                  */
 225                 condlog(0, "%s: can't flush", mpp->alias);
 226                 return 1;
 227         }
 228         else {
 229                 dm_lib_release();
 230                 condlog(2, "%s: devmap removed", mpp->alias);
 231         }
 232
 233         orphan_paths(vecs->pathvec, mpp);
 234         remove_map_and_stop_waiter(mpp, vecs, 1);
 235
 236         return 0;
 237 }
 238
 239 static int
 240 uev_add_map (struct uevent * uev, struct vectors * vecs)
 241 {
 242         char *alias;
 243         int major = -1, minor = -1, rc;
 244
 245         condlog(3, "%s: add map (uevent)", uev->kernel);
 246         alias = uevent_get_dm_name(uev);
 247         if (!alias) {
 248                 condlog(3, "%s: No DM_NAME in uevent", uev->kernel);
 249                 major = uevent_get_major(uev);
 250                 minor = uevent_get_minor(uev);
 251                 alias = dm_mapname(major, minor);
 252                 if (!alias) {
 253                         condlog(2, "%s: mapname not found for %d:%d",
 254                                 uev->kernel, major, minor);
 255                         return 1;
 256                 }
 257         }
 258         rc = ev_add_map(uev->kernel, alias, vecs);
 259         FREE(alias);
 260         return rc;
 261 }
 262
 263 int
 264 ev_add_map (char * dev, char * alias, struct vectors * vecs)
 265 {
 266         char * refwwid;
 267         struct multipath * mpp;
 268         int map_present;
 269         int r = 1;
 270
 271         map_present = dm_map_present(alias);
 272
 273         if (map_present && dm_type(alias, TGT_MPATH) <= 0) {
 274                 condlog(4, "%s: not a multipath map", alias);
 275                 return 0;
 276         }
 277
 278         mpp = find_mp_by_alias(vecs->mpvec, alias);
 279
 280         if (mpp) {
 281                 /*
 282                  * Not really an error -- we generate our own uevent
 283                  * if we create a multipath mapped device as a result
 284                  * of uev_add_path
 285                  */
 286                 if (conf->reassign_maps) {
 287                         condlog(3, "%s: Reassign existing device-mapper devices",
 288                                 alias);
 289                         dm_reassign(alias);
 290                 }
 291                 return 0;
 292         }
 293         condlog(2, "%s: adding map", alias);
 294
 295         /*
 296          * now we can register the map
 297          */
 298         if (map_present && (mpp = add_map_without_path(vecs, alias))) {
 299                 sync_map_state(mpp);
 300                 condlog(2, "%s: devmap %s registered", alias, dev);
 301                 return 0;
 302         }
 303         refwwid = get_refwwid(dev, DEV_DEVMAP, vecs->pathvec);
 304
 305         if (refwwid) {
 306                 r = coalesce_paths(vecs, NULL, refwwid, 0);
 307                 dm_lib_release();
 308         }
 309
 310         if (!r)
 311                 condlog(2, "%s: devmap %s added", alias, dev);
 312         else
 313                 condlog(0, "%s: uev_add_map %s failed", alias, dev);
 314
 315         FREE(refwwid);
 316         return r;
 317 }
 318
 319 static int
 320 uev_remove_map (struct uevent * uev, struct vectors * vecs)
 321 {
 322         char *alias;
 323         int minor;
 324         struct multipath *mpp;
 325
 326         condlog(2, "%s: remove map (uevent)", uev->kernel);
 327         alias = uevent_get_dm_name(uev);
 328         if (!alias) {
 329                 condlog(3, "%s: No DM_NAME in uevent, ignoring", uev->kernel);
 330                 return 0;
 331         }
 332         minor = uevent_get_minor(uev);
 333         mpp = find_mp_by_minor(vecs->mpvec, minor);
 334
 335         if (!mpp) {
 336                 condlog(2, "%s: devmap not registered, can't remove",
 337                         uev->kernel);
 338                 goto out;
 339         }
 340         if (strcmp(mpp->alias, alias)) {
 341                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 342                         mpp->alias, mpp->dmi->minor, minor);
 343                 goto out;
 344         }
 345
 346         orphan_paths(vecs->pathvec, mpp);
 347         remove_map_and_stop_waiter(mpp, vecs, 1);
 348 out:
 349         FREE(alias);
 350         return 0;
 351 }
 352
 353 int
 354 ev_remove_map (char * devname, char * alias, int minor, struct vectors * vecs)
 355 {
 356         struct multipath * mpp;
 357
 358         mpp = find_mp_by_minor(vecs->mpvec, minor);
 359
 360         if (!mpp) {
 361                 condlog(2, "%s: devmap not registered, can't remove",
 362                         devname);
 363                 return 0;
 364         }
 365         if (strcmp(mpp->alias, alias)) {
 366                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 367                         mpp->alias, mpp->dmi->minor, minor);
 368                 return 0;
 369         }
 370         return flush_map(mpp, vecs);
 371 }
 372
 373 static int
 374 uev_add_path (struct uevent *uev, struct vectors * vecs)
 375 {
 376         struct path *pp;
 377
 378         condlog(2, "%s: add path (uevent)", uev->kernel);
 379         if (strstr(uev->kernel, "..") != NULL) {
 380                 /*
 381                  * Don't allow relative device names in the pathvec
 382                  */
 383                 condlog(0, "%s: path name is invalid", uev->kernel);
 384                 return 1;
 385         }
 386
 387         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 388         if (pp) {
 389                 condlog(0, "%s: spurious uevent, path already in pathvec",
 390                         uev->kernel);
 391                 if (pp->mpp)
 392                         return 0;
 393         } else {
 394                 struct udev_device *udevice;
 395
 396                 /*
 397                  * get path vital state
 398                  */
 399                 udevice = udev_device_ref(uev->udev);
 400                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 401                                           udevice, DI_ALL))) {
 402                         condlog(0, "%s: failed to store path info",
 403                                 uev->kernel);
 404                         udev_device_unref(udevice);
 405                         return 1;
 406                 }
 407                 pp->checkint = conf->checkint;
 408         }
 409
 410         return (ev_add_path(pp, vecs) != 1)? 0 : 1;
 411 }
 412
 413 /*
 414  * returns:
 415  * 0: added
 416  * 1: error
 417  * 2: blacklisted
 418  */
 419 int
 420 ev_add_path (struct path * pp, struct vectors * vecs)
 421 {
 422         struct multipath * mpp;
 423         char empty_buff[WWID_SIZE] = {0};
 424         char params[PARAMS_SIZE] = {0};
 425         int retries = 3;
 426         int start_waiter = 0;
 427
 428         /*
 429          * need path UID to go any further
 430          */
 431         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 432                 condlog(0, "%s: failed to get path uid", pp->dev);
 433                 goto fail; /* leave path added to pathvec */
 434         }
 435         if (filter_path(conf, pp) > 0){
 436                 int i = find_slot(vecs->pathvec, (void *)pp);
 437                 if (i != -1)
 438                         vector_del_slot(vecs->pathvec, i);
 439                 free_path(pp);
 440                 return 2;
 441         }
 442         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 443 rescan:
 444         if (mpp) {
 445                 if ((!pp->size) || (mpp->size != pp->size)) {
 446                         if (!pp->size)
 447                                 condlog(0, "%s: failed to add new path %s, "
 448                                         "device size is 0",
 449                                         mpp->alias, pp->dev);
 450                         else
 451                                 condlog(0, "%s: failed to add new path %s, "
 452                                         "device size mismatch",
 453                                         mpp->alias, pp->dev);
 454                         int i = find_slot(vecs->pathvec, (void *)pp);
 455                         if (i != -1)
 456                                 vector_del_slot(vecs->pathvec, i);
 457                         free_path(pp);
 458                         return 1;
 459                 }
 460
 461                 condlog(4,"%s: adopting all paths for path %s",
 462                         mpp->alias, pp->dev);
 463                 if (adopt_paths(vecs->pathvec, mpp, 1))
 464                         goto fail; /* leave path added to pathvec */
 465
 466                 verify_paths(mpp, vecs, NULL);
 467                 mpp->flush_on_last_del = FLUSH_UNDEF;
 468                 mpp->action = ACT_RELOAD;
 469         }
 470         else {
 471                 if (!pp->size) {
 472                         condlog(0, "%s: failed to create new map,"
 473                                 " device size is 0 ", pp->dev);
 474                         int i = find_slot(vecs->pathvec, (void *)pp);
 475                         if (i != -1)
 476                                 vector_del_slot(vecs->pathvec, i);
 477                         free_path(pp);
 478                         return 1;
 479                 }
 480
 481                 condlog(4,"%s: creating new map", pp->dev);
 482                 if ((mpp = add_map_with_path(vecs, pp, 1))) {
 483                         mpp->action = ACT_CREATE;
 484                         /*
 485                          * We don't depend on ACT_CREATE, as domap will
 486                          * set it to ACT_NOTHING when complete.
 487                          */
 488                         start_waiter = 1;
 489                 }
 490                 else
 491                         goto fail; /* leave path added to pathvec */
 492         }
 493
 494         /* persistent reseravtion check*/
 495         mpath_pr_event_handle(pp);
 496
 497         /*
 498          * push the map to the device-mapper
 499          */
 500         if (setup_map(mpp, params, PARAMS_SIZE)) {
 501                 condlog(0, "%s: failed to setup map for addition of new "
 502                         "path %s", mpp->alias, pp->dev);
 503                 goto fail_map;
 504         }
 505         /*
 506          * reload the map for the multipath mapped device
 507          */
 508         if (domap(mpp, params) <= 0) {
 509                 condlog(0, "%s: failed in domap for addition of new "
 510                         "path %s", mpp->alias, pp->dev);
 511                 /*
 512                  * deal with asynchronous uevents :((
 513                  */
 514                 if (mpp->action == ACT_RELOAD && retries-- > 0) {
 515                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 516                         sleep(1);
 517                         update_mpp_paths(mpp, vecs->pathvec);
 518                         goto rescan;
 519                 }
 520                 else if (mpp->action == ACT_RELOAD)
 521                         condlog(0, "%s: giving up reload", mpp->alias);
 522                 else
 523                         goto fail_map;
 524         }
 525         dm_lib_release();
 526
 527         /*
 528          * update our state from kernel regardless of create or reload
 529          */
 530         if (setup_multipath(vecs, mpp))
 531                 goto fail; /* if setup_multipath fails, it removes the map */
 532
 533         sync_map_state(mpp);
 534
 535         if ((mpp->action == ACT_CREATE ||
 536              (mpp->action == ACT_NOTHING && start_waiter && !mpp->waiter)) &&
 537             start_waiter_thread(mpp, vecs))
 538                         goto fail_map;
 539
 540         if (retries >= 0) {
 541                 condlog(2, "%s path added to devmap %s", pp->dev, mpp->alias);
 542                 return 0;
 543         }
 544         else
 545                 return 1;
 546
 547 fail_map:
 548         remove_map(mpp, vecs, 1);
 549 fail:
 550         orphan_path(pp);
 551         return 1;
 552 }
 553
 554 static int
 555 uev_remove_path (struct uevent *uev, struct vectors * vecs)
 556 {
 557         struct path *pp;
 558
 559         condlog(2, "%s: remove path (uevent)", uev->kernel);
 560         pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 561
 562         if (!pp) {
 563                 /* Not an error; path might have been purged earlier */
 564                 condlog(0, "%s: path already removed", uev->kernel);
 565                 return 0;
 566         }
 567
 568         return ev_remove_path(pp, vecs);
 569 }
 570
 571 int
 572 ev_remove_path (struct path *pp, struct vectors * vecs)
 573 {
 574         struct multipath * mpp;
 575         int i, retval = 0;
 576         char params[PARAMS_SIZE] = {0};
 577
 578         /*
 579          * avoid referring to the map of an orphaned path
 580          */
 581         if ((mpp = pp->mpp)) {
 582                 /*
 583                  * transform the mp->pg vector of vectors of paths
 584                  * into a mp->params string to feed the device-mapper
 585                  */
 586                 if (update_mpp_paths(mpp, vecs->pathvec)) {
 587                         condlog(0, "%s: failed to update paths",
 588                                 mpp->alias);
 589                         goto fail;
 590                 }
 591                 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 592                         vector_del_slot(mpp->paths, i);
 593
 594                 /*
 595                  * remove the map IFF removing the last path
 596                  */
 597                 if (VECTOR_SIZE(mpp->paths) == 0) {
 598                         char alias[WWID_SIZE];
 599
 600                         /*
 601                          * flush_map will fail if the device is open
 602                          */
 603                         strncpy(alias, mpp->alias, WWID_SIZE);
 604                         if (mpp->flush_on_last_del == FLUSH_ENABLED) {
 605                                 condlog(2, "%s Last path deleted, disabling queueing", mpp->alias);
 606                                 mpp->retry_tick = 0;
 607                                 mpp->no_path_retry = NO_PATH_RETRY_FAIL;
 608                                 mpp->flush_on_last_del = FLUSH_IN_PROGRESS;
 609                                 dm_queue_if_no_path(mpp->alias, 0);
 610                         }
 611                         if (!flush_map(mpp, vecs)) {
 612                                 condlog(2, "%s: removed map after"
 613                                         " removing all paths",
 614                                         alias);
 615                                 retval = 0;
 616                                 goto out;
 617                         }
 618                         /*
 619                          * Not an error, continue
 620                          */
 621                 }
 622
 623                 if (setup_map(mpp, params, PARAMS_SIZE)) {
 624                         condlog(0, "%s: failed to setup map for"
 625                                 " removal of path %s", mpp->alias, pp->dev);
 626                         goto fail;
 627                 }
 628                 /*
 629                  * reload the map
 630                  */
 631                 mpp->action = ACT_RELOAD;
 632                 if (domap(mpp, params) <= 0) {
 633                         condlog(0, "%s: failed in domap for "
 634                                 "removal of path %s",
 635                                 mpp->alias, pp->dev);
 636                         retval = 1;
 637                 } else {
 638                         /*
 639                          * update our state from kernel
 640                          */
 641                         if (setup_multipath(vecs, mpp)) {
 642                                 goto fail;
 643                         }
 644                         sync_map_state(mpp);
 645
 646                         condlog(2, "%s: path removed from map %s",
 647                                 pp->dev, mpp->alias);
 648                 }
 649         }
 650
 651 out:
 652         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 653                 vector_del_slot(vecs->pathvec, i);
 654
 655         free_path(pp);
 656
 657         return retval;
 658
 659 fail:
 660         remove_map_and_stop_waiter(mpp, vecs, 1);
 661         return 1;
 662 }
 663
 664 static int
 665 uev_update_path (struct uevent *uev, struct vectors * vecs)
 666 {
 667         int retval, ro;
 668
 669         ro = uevent_get_disk_ro(uev);
 670
 671         if (ro >= 0) {
 672                 struct path * pp;
 673
 674                 condlog(2, "%s: update path write_protect to '%d' (uevent)",
 675                         uev->kernel, ro);
 676                 pp = find_path_by_dev(vecs->pathvec, uev->kernel);
 677                 if (!pp) {
 678                         condlog(0, "%s: spurious uevent, path not found",
 679                                 uev->kernel);
 680                         return 1;
 681                 }
 682                 if (pp->mpp)
 683                         retval = reload_map(vecs, pp->mpp);
 684
 685                 condlog(2, "%s: map %s reloaded (retval %d)",
 686                         uev->kernel, pp->mpp->alias, retval);
 687
 688         }
 689
 690         return retval;
 691 }
 692
 693 static int
 694 map_discovery (struct vectors * vecs)
 695 {
 696         struct multipath * mpp;
 697         unsigned int i;
 698
 699         if (dm_get_maps(vecs->mpvec))
 700                 return 1;
 701
 702         vector_foreach_slot (vecs->mpvec, mpp, i)
 703                 if (setup_multipath(vecs, mpp))
 704                         return 1;
 705
 706         return 0;
 707 }
 708
 709 int
 710 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 711 {
 712         struct vectors * vecs;
 713         int r;
 714
 715         *reply = NULL;
 716         *len = 0;
 717         vecs = (struct vectors *)trigger_data;
 718
 719         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 720         lock(vecs->lock);
 721         pthread_testcancel();
 722
 723         r = parse_cmd(str, reply, len, vecs);
 724
 725         if (r > 0) {
 726                 *reply = STRDUP("fail\n");
 727                 *len = strlen(*reply) + 1;
 728                 r = 1;
 729         }
 730         else if (!r && *len == 0) {
 731                 *reply = STRDUP("ok\n");
 732                 *len = strlen(*reply) + 1;
 733                 r = 0;
 734         }
 735         /* else if (r < 0) leave *reply alone */
 736
 737         lock_cleanup_pop(vecs->lock);
 738         return r;
 739 }
 740
 741 static int
 742 uev_discard(char * devpath)
 743 {
 744         char *tmp;
 745         char a[11], b[11];
 746
 747         /*
 748          * keep only block devices, discard partitions
 749          */
 750         tmp = strstr(devpath, "/block/");
 751         if (tmp == NULL){
 752                 condlog(4, "no /block/ in '%s'", devpath);
 753                 return 1;
 754         }
 755         if (sscanf(tmp, "/block/%10s", a) != 1 ||
 756             sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
 757                 condlog(4, "discard event on %s", devpath);
 758                 return 1;
 759         }
 760         return 0;
 761 }
 762
 763 int
 764 uev_trigger (struct uevent * uev, void * trigger_data)
 765 {
 766         int r = 0;
 767         struct vectors * vecs;
 768
 769         vecs = (struct vectors *)trigger_data;
 770
 771         if (uev_discard(uev->devpath))
 772                 return 0;
 773
 774         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 775         lock(vecs->lock);
 776         pthread_testcancel();
 777
 778         /*
 779          * device map event
 780          * Add events are ignored here as the tables
 781          * are not fully initialised then.
 782          */
 783         if (!strncmp(uev->kernel, "dm-", 3)) {
 784                 if (!strncmp(uev->action, "change", 6)) {
 785                         r = uev_add_map(uev, vecs);
 786                         goto out;
 787                 }
 788                 if (!strncmp(uev->action, "remove", 6)) {
 789                         r = uev_remove_map(uev, vecs);
 790                         goto out;
 791                 }
 792                 goto out;
 793         }
 794
 795         /*
 796          * path add/remove event
 797          */
 798         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 799                            uev->kernel) > 0)
 800                 goto out;
 801
 802         if (!strncmp(uev->action, "add", 3)) {
 803                 r = uev_add_path(uev, vecs);
 804                 goto out;
 805         }
 806         if (!strncmp(uev->action, "remove", 6)) {
 807                 r = uev_remove_path(uev, vecs);
 808                 goto out;
 809         }
 810         if (!strncmp(uev->action, "change", 6)) {
 811                 r = uev_update_path(uev, vecs);
 812                 goto out;
 813         }
 814
 815 out:
 816         lock_cleanup_pop(vecs->lock);
 817         return r;
 818 }
 819
 820 static void *
 821 ueventloop (void * ap)
 822 {
 823         block_signal(SIGUSR1, NULL);
 824         block_signal(SIGHUP, NULL);
 825
 826         if (uevent_listen())
 827                 condlog(0, "error starting uevent listener");
 828
 829         return NULL;
 830 }
 831
 832 static void *
 833 uevqloop (void * ap)
 834 {
 835         block_signal(SIGUSR1, NULL);
 836         block_signal(SIGHUP, NULL);
 837
 838         if (uevent_dispatch(&uev_trigger, ap))
 839                 condlog(0, "error starting uevent dispatcher");
 840
 841         return NULL;
 842 }
 843 static void *
 844 uxlsnrloop (void * ap)
 845 {
 846         block_signal(SIGUSR1, NULL);
 847         block_signal(SIGHUP, NULL);
 848
 849         if (cli_init())
 850                 return NULL;
 851
 852         set_handler_callback(LIST+PATHS, cli_list_paths);
 853         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 854         set_handler_callback(LIST+MAPS, cli_list_maps);
 855         set_handler_callback(LIST+STATUS, cli_list_status);
 856         set_handler_callback(LIST+DAEMON, cli_list_daemon);
 857         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 858         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 859         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 860         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 861         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 862         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 863         set_handler_callback(LIST+CONFIG, cli_list_config);
 864         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 865         set_handler_callback(LIST+DEVICES, cli_list_devices);
 866         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 867         set_handler_callback(ADD+PATH, cli_add_path);
 868         set_handler_callback(DEL+PATH, cli_del_path);
 869         set_handler_callback(ADD+MAP, cli_add_map);
 870         set_handler_callback(DEL+MAP, cli_del_map);
 871         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 872         set_handler_callback(RECONFIGURE, cli_reconfigure);
 873         set_handler_callback(SUSPEND+MAP, cli_suspend);
 874         set_handler_callback(RESUME+MAP, cli_resume);
 875         set_handler_callback(RESIZE+MAP, cli_resize);
 876         set_handler_callback(RELOAD+MAP, cli_reload);
 877         set_handler_callback(RESET+MAP, cli_reassign);
 878         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 879         set_handler_callback(FAIL+PATH, cli_fail);
 880         set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
 881         set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
 882         set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
 883         set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
 884         set_handler_callback(QUIT, cli_quit);
 885         set_handler_callback(SHUTDOWN, cli_shutdown);
 886         set_handler_callback(GETPRSTATUS+MAP, cli_getprstatus);
 887         set_handler_callback(SETPRSTATUS+MAP, cli_setprstatus);
 888         set_handler_callback(UNSETPRSTATUS+MAP, cli_unsetprstatus);
 889
 890         umask(077);
 891         uxsock_listen(&uxsock_trigger, ap);
 892
 893         return NULL;
 894 }
 895
 896 int
 897 exit_daemon (int status)
 898 {
 899         if (status != 0)
 900                 fprintf(stderr, "bad exit status. see daemon.log\n");
 901
 902         condlog(3, "unlink pidfile");
 903         unlink(DEFAULT_PIDFILE);
 904
 905         pthread_mutex_lock(&exit_mutex);
 906         pthread_cond_signal(&exit_cond);
 907         pthread_mutex_unlock(&exit_mutex);
 908
 909         return status;
 910 }
 911
 912 const char *
 913 daemon_status(void)
 914 {
 915         switch (running_state) {
 916         case DAEMON_INIT:
 917                 return "init";
 918         case DAEMON_START:
 919                 return "startup";
 920         case DAEMON_CONFIGURE:
 921                 return "configure";
 922         case DAEMON_RUNNING:
 923                 return "running";
 924         case DAEMON_SHUTDOWN:
 925                 return "shutdown";
 926         }
 927         return NULL;
 928 }
 929
 930 static void
 931 fail_path (struct path * pp, int del_active)
 932 {
 933         if (!pp->mpp)
 934                 return;
 935
 936         condlog(2, "checker failed path %s in map %s",
 937                  pp->dev_t, pp->mpp->alias);
 938
 939         dm_fail_path(pp->mpp->alias, pp->dev_t);
 940         if (del_active)
 941                 update_queue_mode_del_path(pp->mpp);
 942 }
 943
 944 /*
 945  * caller must have locked the path list before calling that function
 946  */
 947 static void
 948 reinstate_path (struct path * pp, int add_active)
 949 {
 950         if (!pp->mpp)
 951                 return;
 952
 953         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 954                 condlog(0, "%s: reinstate failed", pp->dev_t);
 955         else {
 956                 condlog(2, "%s: reinstated", pp->dev_t);
 957                 if (add_active)
 958                         update_queue_mode_add_path(pp->mpp);
 959         }
 960 }
 961
 962 static void
 963 enable_group(struct path * pp)
 964 {
 965         struct pathgroup * pgp;
 966
 967         /*
 968          * if path is added through uev_add_path, pgindex can be unset.
 969          * next update_strings() will set it, upon map reload event.
 970          *
 971          * we can safely return here, because upon map reload, all
 972          * PG will be enabled.
 973          */
 974         if (!pp->mpp->pg || !pp->pgindex)
 975                 return;
 976
 977         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 978
 979         if (pgp->status == PGSTATE_DISABLED) {
 980                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 981                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 982         }
 983 }
 984
 985 static void
 986 mpvec_garbage_collector (struct vectors * vecs)
 987 {
 988         struct multipath * mpp;
 989         unsigned int i;
 990
 991         if (!vecs->mpvec)
 992                 return;
 993
 994         vector_foreach_slot (vecs->mpvec, mpp, i) {
 995                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 996                         condlog(2, "%s: remove dead map", mpp->alias);
 997                         remove_map_and_stop_waiter(mpp, vecs, 1);
 998                         i--;
 999                 }
1000         }
1001 }
1002
1003 static void
1004 defered_failback_tick (vector mpvec)
1005 {
1006         struct multipath * mpp;
1007         unsigned int i;
1008
1009         vector_foreach_slot (mpvec, mpp, i) {
1010                 /*
1011                  * defered failback getting sooner
1012                  */
1013                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1014                         mpp->failback_tick--;
1015
1016                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1017                                 switch_pathgroup(mpp);
1018                 }
1019         }
1020 }
1021
1022 static void
1023 retry_count_tick(vector mpvec)
1024 {
1025         struct multipath *mpp;
1026         unsigned int i;
1027
1028         vector_foreach_slot (mpvec, mpp, i) {
1029                 if (mpp->retry_tick) {
1030                         mpp->stat_total_queueing_time++;
1031                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1032                         if(--mpp->retry_tick == 0) {
1033                                 dm_queue_if_no_path(mpp->alias, 0);
1034                                 condlog(2, "%s: Disable queueing", mpp->alias);
1035                         }
1036                 }
1037         }
1038 }
1039
1040 int update_prio(struct path *pp, int refresh_all)
1041 {
1042         int oldpriority;
1043         struct path *pp1;
1044         struct pathgroup * pgp;
1045         int i, j, changed = 0;
1046
1047         if (refresh_all) {
1048                 vector_foreach_slot (pp->mpp->pg, pgp, i) {
1049                         vector_foreach_slot (pgp->paths, pp1, j) {
1050                                 oldpriority = pp1->priority;
1051                                 pathinfo(pp1, conf->hwtable, DI_PRIO);
1052                                 if (pp1->priority != oldpriority)
1053                                         changed = 1;
1054                         }
1055                 }
1056                 return changed;
1057         }
1058         oldpriority = pp->priority;
1059         pathinfo(pp, conf->hwtable, DI_PRIO);
1060
1061         if (pp->priority == oldpriority)
1062                 return 0;
1063         return 1;
1064 }
1065
1066 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
1067 {
1068         int i;
1069         struct path * pp;
1070         char params[PARAMS_SIZE];
1071
1072         update_mpp_paths(mpp, vecs->pathvec);
1073         if (refresh) {
1074                 vector_foreach_slot (mpp->paths, pp, i)
1075                         pathinfo(pp, conf->hwtable, DI_PRIO);
1076         }
1077         params[0] = '\0';
1078         if (setup_map(mpp, params, PARAMS_SIZE))
1079                 return 1;
1080
1081         mpp->action = ACT_RELOAD;
1082         if (domap(mpp, params) <= 0) {
1083                 condlog(0, "%s: failed to update map : %s", mpp->alias,
1084                         strerror(errno));
1085                 return 1;
1086         }
1087         dm_lib_release();
1088         if (setup_multipath(vecs, mpp) != 0)
1089                 return 1;
1090         sync_map_state(mpp);
1091
1092         return 0;
1093 }
1094
1095 void
1096 check_path (struct vectors * vecs, struct path * pp)
1097 {
1098         int newstate;
1099         int new_path_up = 0;
1100
1101         if (!pp->mpp)
1102                 return;
1103
1104         if (pp->tick && --pp->tick)
1105                 return; /* don't check this path yet */
1106
1107         /*
1108          * provision a next check soonest,
1109          * in case we exit abnormaly from here
1110          */
1111         pp->tick = conf->checkint;
1112
1113         newstate = path_offline(pp);
1114         if (newstate == PATH_UP)
1115                 newstate = get_state(pp, 1);
1116
1117         if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1118                 condlog(2, "%s: unusable path", pp->dev);
1119                 pathinfo(pp, conf->hwtable, 0);
1120                 return;
1121         }
1122         /*
1123          * Async IO in flight. Keep the previous path state
1124          * and reschedule as soon as possible
1125          */
1126         if (newstate == PATH_PENDING) {
1127                 pp->tick = 1;
1128                 return;
1129         }
1130         /*
1131          * Synchronize with kernel state
1132          */
1133         if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1134                 condlog(1, "%s: Could not synchronize with kernel state\n",
1135                         pp->dev);
1136                 pp->dmstate = PSTATE_UNDEF;
1137         }
1138         if (newstate != pp->state) {
1139                 int oldstate = pp->state;
1140                 pp->state = newstate;
1141                 LOG_MSG(1, checker_message(&pp->checker));
1142
1143                 /*
1144                  * upon state change, reset the checkint
1145                  * to the shortest delay
1146                  */
1147                 pp->checkint = conf->checkint;
1148
1149                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1150                         /*
1151                          * proactively fail path in the DM
1152                          */
1153                         if (oldstate == PATH_UP ||
1154                             oldstate == PATH_GHOST)
1155                                 fail_path(pp, 1);
1156                         else
1157                                 fail_path(pp, 0);
1158
1159                         /*
1160                          * cancel scheduled failback
1161                          */
1162                         pp->mpp->failback_tick = 0;
1163
1164                         pp->mpp->stat_path_failures++;
1165                         return;
1166                 }
1167
1168                 if(newstate == PATH_UP || newstate == PATH_GHOST){
1169                         if ( pp->mpp && pp->mpp->prflag ){
1170                                 /*
1171                                  * Check Persistent Reservation.
1172                                  */
1173                         condlog(2, "%s: checking persistent reservation "
1174                                 "registration", pp->dev);
1175                         mpath_pr_event_handle(pp);
1176                         }
1177                 }
1178
1179                 /*
1180                  * reinstate this path
1181                  */
1182                 if (oldstate != PATH_UP &&
1183                     oldstate != PATH_GHOST)
1184                         reinstate_path(pp, 1);
1185                 else
1186                         reinstate_path(pp, 0);
1187
1188                 new_path_up = 1;
1189
1190                 /*
1191                  * if at least one path is up in a group, and
1192                  * the group is disabled, re-enable it
1193                  */
1194                 if (newstate == PATH_UP)
1195                         enable_group(pp);
1196         }
1197         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1198                 if (pp->dmstate == PSTATE_FAILED ||
1199                     pp->dmstate == PSTATE_UNDEF) {
1200                         /* Clear IO errors */
1201                         reinstate_path(pp, 0);
1202                 } else {
1203                         LOG_MSG(4, checker_message(&pp->checker));
1204                         /*
1205                          * double the next check delay.
1206                          * max at conf->max_checkint
1207                          */
1208                         if (pp->checkint < (conf->max_checkint / 2))
1209                                 pp->checkint = 2 * pp->checkint;
1210                         else
1211                                 pp->checkint = conf->max_checkint;
1212
1213                         pp->tick = pp->checkint;
1214                         condlog(4, "%s: delay next check %is",
1215                                 pp->dev_t, pp->tick);
1216                 }
1217         }
1218         else if (newstate == PATH_DOWN) {
1219                 if (conf->log_checker_err == LOG_CHKR_ERR_ONCE)
1220                         LOG_MSG(3, checker_message(&pp->checker));
1221                 else
1222                         LOG_MSG(2, checker_message(&pp->checker));
1223         }
1224
1225         pp->state = newstate;
1226
1227         /*
1228          * path prio refreshing
1229          */
1230         condlog(4, "path prio refresh");
1231
1232         if (update_prio(pp, new_path_up) &&
1233             (pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio) &&
1234              pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1235                 update_path_groups(pp->mpp, vecs, !new_path_up);
1236         else if (need_switch_pathgroup(pp->mpp, 0)) {
1237                 if (pp->mpp->pgfailback > 0 &&
1238                     (new_path_up || pp->mpp->failback_tick <= 0))
1239                         pp->mpp->failback_tick =
1240                                 pp->mpp->pgfailback + 1;
1241                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1242                         switch_pathgroup(pp->mpp);
1243         }
1244 }
1245
1246 static void *
1247 checkerloop (void *ap)
1248 {
1249         struct vectors *vecs;
1250         struct path *pp;
1251         int count = 0;
1252         unsigned int i;
1253         sigset_t old;
1254
1255         mlockall(MCL_CURRENT | MCL_FUTURE);
1256         vecs = (struct vectors *)ap;
1257         condlog(2, "path checkers start up");
1258
1259         /*
1260          * init the path check interval
1261          */
1262         vector_foreach_slot (vecs->pathvec, pp, i) {
1263                 pp->checkint = conf->checkint;
1264         }
1265
1266         while (1) {
1267                 block_signal(SIGHUP, &old);
1268                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1269                 lock(vecs->lock);
1270                 pthread_testcancel();
1271                 condlog(4, "tick");
1272
1273                 if (vecs->pathvec) {
1274                         vector_foreach_slot (vecs->pathvec, pp, i) {
1275                                 check_path(vecs, pp);
1276                         }
1277                 }
1278                 if (vecs->mpvec) {
1279                         defered_failback_tick(vecs->mpvec);
1280                         retry_count_tick(vecs->mpvec);
1281                 }
1282                 if (count)
1283                         count--;
1284                 else {
1285                         condlog(4, "map garbage collection");
1286                         mpvec_garbage_collector(vecs);
1287                         count = MAPGCINT;
1288                 }
1289
1290                 lock_cleanup_pop(vecs->lock);
1291                 pthread_sigmask(SIG_SETMASK, &old, NULL);
1292                 sleep(1);
1293         }
1294         return NULL;
1295 }
1296
1297 int
1298 configure (struct vectors * vecs, int start_waiters)
1299 {
1300         struct multipath * mpp;
1301         struct path * pp;
1302         vector mpvec;
1303         int i;
1304
1305         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1306                 return 1;
1307
1308         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1309                 return 1;
1310
1311         if (!(mpvec = vector_alloc()))
1312                 return 1;
1313
1314         /*
1315          * probe for current path (from sysfs) and map (from dm) sets
1316          */
1317         path_discovery(vecs->pathvec, conf, DI_ALL);
1318
1319         vector_foreach_slot (vecs->pathvec, pp, i){
1320                 if (filter_path(conf, pp) > 0){
1321                         vector_del_slot(vecs->pathvec, i);
1322                         free_path(pp);
1323                         i--;
1324                 }
1325                 else
1326                         pp->checkint = conf->checkint;
1327         }
1328         if (map_discovery(vecs))
1329                 return 1;
1330
1331         /*
1332          * create new set of maps & push changed ones into dm
1333          */
1334         if (coalesce_paths(vecs, mpvec, NULL, 1))
1335                 return 1;
1336
1337         /*
1338          * may need to remove some maps which are no longer relevant
1339          * e.g., due to blacklist changes in conf file
1340          */
1341         if (coalesce_maps(vecs, mpvec))
1342                 return 1;
1343
1344         dm_lib_release();
1345
1346         sync_maps_state(mpvec);
1347         vector_foreach_slot(mpvec, mpp, i){
1348                 update_map_pr(mpp);
1349         }
1350
1351         /*
1352          * purge dm of old maps
1353          */
1354         remove_maps(vecs);
1355
1356         /*
1357          * save new set of maps formed by considering current path state
1358          */
1359         vector_free(vecs->mpvec);
1360         vecs->mpvec = mpvec;
1361
1362         /*
1363          * start dm event waiter threads for these new maps
1364          */
1365         vector_foreach_slot(vecs->mpvec, mpp, i) {
1366                 if (setup_multipath(vecs, mpp))
1367                         return 1;
1368                 if (start_waiters)
1369                         if (start_waiter_thread(mpp, vecs))
1370                                 return 1;
1371         }
1372         return 0;
1373 }
1374
1375 int
1376 reconfigure (struct vectors * vecs)
1377 {
1378         struct config * old = conf;
1379         int retval = 1;
1380
1381         /*
1382          * free old map and path vectors ... they use old conf state
1383          */
1384         if (VECTOR_SIZE(vecs->mpvec))
1385                 remove_maps_and_stop_waiters(vecs);
1386
1387         if (VECTOR_SIZE(vecs->pathvec))
1388                 free_pathvec(vecs->pathvec, FREE_PATHS);
1389
1390         vecs->pathvec = NULL;
1391         conf = NULL;
1392
1393         if (!load_config(DEFAULT_CONFIGFILE)) {
1394                 conf->verbosity = old->verbosity;
1395                 conf->daemon = 1;
1396                 configure(vecs, 1);
1397                 free_config(old);
1398                 retval = 0;
1399         }
1400
1401         return retval;
1402 }
1403
1404 static struct vectors *
1405 init_vecs (void)
1406 {
1407         struct vectors * vecs;
1408
1409         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1410
1411         if (!vecs)
1412                 return NULL;
1413
1414         vecs->lock.mutex =
1415                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1416
1417         if (!vecs->lock.mutex)
1418                 goto out;
1419
1420         pthread_mutex_init(vecs->lock.mutex, NULL);
1421         vecs->lock.depth = 0;
1422
1423         return vecs;
1424
1425 out:
1426         FREE(vecs);
1427         condlog(0, "failed to init paths");
1428         return NULL;
1429 }
1430
1431 static void *
1432 signal_set(int signo, void (*func) (int))
1433 {
1434         int r;
1435         struct sigaction sig;
1436         struct sigaction osig;
1437
1438         sig.sa_handler = func;
1439         sigemptyset(&sig.sa_mask);
1440         sig.sa_flags = 0;
1441
1442         r = sigaction(signo, &sig, &osig);
1443
1444         if (r < 0)
1445                 return (SIG_ERR);
1446         else
1447                 return (osig.sa_handler);
1448 }
1449
1450 static void
1451 sighup (int sig)
1452 {
1453         condlog(2, "reconfigure (SIGHUP)");
1454
1455         if (running_state != DAEMON_RUNNING)
1456                 return;
1457
1458         lock(gvecs->lock);
1459         reconfigure(gvecs);
1460         unlock(gvecs->lock);
1461
1462 #ifdef _DEBUG_
1463         dbg_free_final(NULL);
1464 #endif
1465 }
1466
1467 static void
1468 sigend (int sig)
1469 {
1470         exit_daemon(0);
1471 }
1472
1473 static void
1474 sigusr1 (int sig)
1475 {
1476         condlog(3, "SIGUSR1 received");
1477 }
1478
1479 static void
1480 signal_init(void)
1481 {
1482         signal_set(SIGHUP, sighup);
1483         signal_set(SIGUSR1, sigusr1);
1484         signal_set(SIGINT, sigend);
1485         signal_set(SIGTERM, sigend);
1486         signal(SIGPIPE, SIG_IGN);
1487 }
1488
1489 static void
1490 setscheduler (void)
1491 {
1492         int res;
1493         static struct sched_param sched_param = {
1494                 .sched_priority = 99
1495         };
1496
1497         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1498
1499         if (res == -1)
1500                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1501         return;
1502 }
1503
1504 static void
1505 set_oom_adj (void)
1506 {
1507         int retry = 1;
1508         char *file = "/proc/self/oom_score_adj";
1509         int score = OOM_SCORE_ADJ_MIN;
1510         FILE *fp;
1511         struct stat st;
1512
1513         do {
1514                 if (stat(file, &st) == 0){
1515                         fp = fopen(file, "w");
1516                         if (!fp) {
1517                                 condlog(0, "couldn't fopen %s : %s", file,
1518                                         strerror(errno));
1519                                 return;
1520                         }
1521                         fprintf(fp, "%i", score);
1522                         fclose(fp);
1523                         return;
1524                 }
1525                 if (errno != ENOENT) {
1526                         condlog(0, "couldn't stat %s : %s", file,
1527                                 strerror(errno));
1528                         return;
1529                 }
1530                 file = "/proc/self/oom_adj";
1531                 score = OOM_ADJUST_MIN;
1532         } while (retry--);
1533         condlog(0, "couldn't adjust oom score");
1534 }
1535
1536 static int
1537 child (void * param)
1538 {
1539         pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1540         pthread_attr_t log_attr, misc_attr;
1541         struct vectors * vecs;
1542         struct multipath * mpp;
1543         int i;
1544         int rc;
1545
1546         mlockall(MCL_CURRENT | MCL_FUTURE);
1547
1548         setup_thread_attr(&misc_attr, 64 * 1024, 1);
1549         setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1550
1551         if (logsink) {
1552                 setup_thread_attr(&log_attr, 64 * 1024, 0);
1553                 log_thread_start(&log_attr);
1554                 pthread_attr_destroy(&log_attr);
1555         }
1556
1557         running_state = DAEMON_START;
1558
1559         condlog(2, "--------start up--------");
1560         condlog(2, "read " DEFAULT_CONFIGFILE);
1561
1562         if (load_config(DEFAULT_CONFIGFILE))
1563                 exit(1);
1564
1565         if (init_checkers()) {
1566                 condlog(0, "failed to initialize checkers");
1567                 exit(1);
1568         }
1569         if (init_prio()) {
1570                 condlog(0, "failed to initialize prioritizers");
1571                 exit(1);
1572         }
1573
1574         setlogmask(LOG_UPTO(conf->verbosity + 3));
1575
1576         if (conf->max_fds) {
1577                 struct rlimit fd_limit;
1578
1579                 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1580                         condlog(0, "can't get open fds limit: %s\n",
1581                                 strerror(errno));
1582                         fd_limit.rlim_cur = 0;
1583                         fd_limit.rlim_max = 0;
1584                 }
1585                 if (fd_limit.rlim_cur < conf->max_fds) {
1586                         fd_limit.rlim_cur = conf->max_fds;
1587                         if (fd_limit.rlim_max < conf->max_fds)
1588                                 fd_limit.rlim_max = conf->max_fds;
1589                         if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1590                                 condlog(0, "can't set open fds limit to "
1591                                         "%lu/%lu : %s\n",
1592                                         fd_limit.rlim_cur, fd_limit.rlim_max,
1593                                         strerror(errno));
1594                         } else {
1595                                 condlog(3, "set open fds limit to %lu/%lu\n",
1596                                         fd_limit.rlim_cur, fd_limit.rlim_max);
1597                         }
1598                 }
1599
1600         }
1601
1602         signal_init();
1603         setscheduler();
1604         set_oom_adj();
1605         vecs = gvecs = init_vecs();
1606
1607         if (!vecs)
1608                 exit(1);
1609
1610         conf->daemon = 1;
1611         udev_set_sync_support(0);
1612         /*
1613          * Start uevent listener early to catch events
1614          */
1615         if ((rc = pthread_create(&uevent_thr, &misc_attr, ueventloop, vecs))) {
1616                 condlog(0, "failed to create uevent thread: %d", rc);
1617                 exit(1);
1618         }
1619         if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
1620                 condlog(0, "failed to create cli listener: %d", rc);
1621                 exit(1);
1622         }
1623         /*
1624          * fetch and configure both paths and multipaths
1625          */
1626         lock(vecs->lock);
1627         running_state = DAEMON_CONFIGURE;
1628
1629         if (configure(vecs, 1)) {
1630                 unlock(vecs->lock);
1631                 condlog(0, "failure during configuration");
1632                 exit(1);
1633         }
1634         unlock(vecs->lock);
1635
1636         /*
1637          * start threads
1638          */
1639         if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
1640                 condlog(0,"failed to create checker loop thread: %d", rc);
1641                 exit(1);
1642         }
1643         if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
1644                 condlog(0, "failed to create uevent dispatcher: %d", rc);
1645                 exit(1);
1646         }
1647         pthread_attr_destroy(&misc_attr);
1648
1649         pthread_mutex_lock(&exit_mutex);
1650         /* Startup complete, create logfile */
1651         if (pidfile_create(DEFAULT_PIDFILE, daemon_pid))
1652                 /* Ignore errors, we can live without */
1653                 condlog(1, "failed to create pidfile");
1654
1655         running_state = DAEMON_RUNNING;
1656         pthread_cond_wait(&exit_cond, &exit_mutex);
1657
1658         /*
1659          * exit path
1660          */
1661         running_state = DAEMON_SHUTDOWN;
1662         block_signal(SIGHUP, NULL);
1663         lock(vecs->lock);
1664         if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
1665                 vector_foreach_slot(vecs->mpvec, mpp, i)
1666                         dm_queue_if_no_path(mpp->alias, 0);
1667         remove_maps_and_stop_waiters(vecs);
1668         unlock(vecs->lock);
1669
1670         pthread_cancel(check_thr);
1671         pthread_cancel(uevent_thr);
1672         pthread_cancel(uxlsnr_thr);
1673         pthread_cancel(uevq_thr);
1674
1675         lock(vecs->lock);
1676         free_pathvec(vecs->pathvec, FREE_PATHS);
1677         vecs->pathvec = NULL;
1678         unlock(vecs->lock);
1679         /* Now all the waitevent threads will start rushing in. */
1680         while (vecs->lock.depth > 0) {
1681                 sleep (1); /* This is weak. */
1682                 condlog(3,"Have %d wait event checkers threads to de-alloc, waiting..\n", vecs->lock.depth);
1683         }
1684         pthread_mutex_destroy(vecs->lock.mutex);
1685         FREE(vecs->lock.mutex);
1686         vecs->lock.depth = 0;
1687         vecs->lock.mutex = NULL;
1688         FREE(vecs);
1689         vecs = NULL;
1690
1691         cleanup_checkers();
1692         cleanup_prio();
1693
1694         dm_lib_release();
1695         dm_lib_exit();
1696
1697         /* We're done here */
1698         condlog(3, "unlink pidfile");
1699         unlink(DEFAULT_PIDFILE);
1700
1701         condlog(2, "--------shut down-------");
1702
1703         if (logsink)
1704                 log_thread_stop();
1705
1706         /*
1707          * Freeing config must be done after condlog() and dm_lib_exit(),
1708          * because logging functions like dlog() and dm_write_log()
1709          * reference the config.
1710          */
1711         free_config(conf);
1712         conf = NULL;
1713
1714 #ifdef _DEBUG_
1715         dbg_free_final(NULL);
1716 #endif
1717
1718         exit(0);
1719 }
1720
1721 static int
1722 daemonize(void)
1723 {
1724         int pid;
1725         int dev_null_fd;
1726
1727         if( (pid = fork()) < 0){
1728                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1729                 return -1;
1730         }
1731         else if (pid != 0)
1732                 return pid;
1733
1734         setsid();
1735
1736         if ( (pid = fork()) < 0)
1737                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1738         else if (pid != 0)
1739                 _exit(0);
1740
1741         if (chdir("/") < 0)
1742                 fprintf(stderr, "cannot chdir to '/', continuing\n");
1743
1744         dev_null_fd = open("/dev/null", O_RDWR);
1745         if (dev_null_fd < 0){
1746                 fprintf(stderr, "cannot open /dev/null for input & output : %s\n",
1747                         strerror(errno));
1748                 _exit(0);
1749         }
1750
1751         close(STDIN_FILENO);
1752         dup(dev_null_fd);
1753         close(STDOUT_FILENO);
1754         dup(dev_null_fd);
1755         close(STDERR_FILENO);
1756         dup(dev_null_fd);
1757         close(dev_null_fd);
1758         daemon_pid = getpid();
1759         return 0;
1760 }
1761
1762 int
1763 main (int argc, char *argv[])
1764 {
1765         extern char *optarg;
1766         extern int optind;
1767         int arg;
1768         int err;
1769
1770         logsink = 1;
1771         running_state = DAEMON_INIT;
1772         dm_init();
1773
1774         if (getuid() != 0) {
1775                 fprintf(stderr, "need to be root\n");
1776                 exit(1);
1777         }
1778
1779         /* make sure we don't lock any path */
1780         chdir("/");
1781         umask(umask(077) | 022);
1782
1783         conf = alloc_config();
1784
1785         if (!conf)
1786                 exit(1);
1787
1788         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1789         switch(arg) {
1790                 case 'd':
1791                         logsink = 0;
1792                         //debug=1; /* ### comment me out ### */
1793                         break;
1794                 case 'v':
1795                         if (sizeof(optarg) > sizeof(char *) ||
1796                             !isdigit(optarg[0]))
1797                                 exit(1);
1798
1799                         conf->verbosity = atoi(optarg);
1800                         break;
1801                 case 'k':
1802                         uxclnt(optarg);
1803                         exit(0);
1804                 default:
1805                         ;
1806                 }
1807         }
1808         if (optind < argc) {
1809                 char cmd[CMDSIZE];
1810                 char * s = cmd;
1811                 char * c = s;
1812
1813                 while (optind < argc) {
1814                         if (strchr(argv[optind], ' '))
1815                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1816                         else
1817                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1818                         optind++;
1819                 }
1820                 c += snprintf(c, s + CMDSIZE - c, "\n");
1821                 uxclnt(s);
1822                 exit(0);
1823         }
1824
1825         if (!logsink)
1826                 err = 0;
1827         else
1828                 err = daemonize();
1829
1830         if (err < 0)
1831                 /* error */
1832                 exit(1);
1833         else if (err > 0)
1834                 /* parent dies */
1835                 exit(0);
1836         else
1837                 /* child lives */
1838                 return (child(NULL));
1839 }
1840
1841 void *  mpath_pr_event_handler_fn (void * pathp )
1842 {
1843         struct multipath * mpp;
1844         int i,j, ret, isFound;
1845         struct path * pp = (struct path *)pathp;
1846         unsigned char *keyp;
1847         uint64_t prkey;
1848         struct prout_param_descriptor *param;
1849         struct prin_resp *resp;
1850
1851         mpp = pp->mpp;
1852
1853         resp = mpath_alloc_prin_response(MPATH_PRIN_RKEY_SA);
1854         if (!resp){
1855                 condlog(0,"%s Alloc failed for prin response \n", pp->dev);
1856                 return NULL;
1857         }
1858
1859         ret = prin_do_scsi_ioctl(pp->dev, MPATH_PRIN_RKEY_SA, resp, 0);
1860         if (ret != MPATH_PR_SUCCESS )
1861         {
1862                 condlog(0,"%s : pr in read keys service action failed. Error=%d\n", pp->dev, ret);
1863                 goto out;
1864         }
1865
1866         condlog(3, " event pr=%d addlen=%d\n",resp->prin_descriptor.prin_readkeys.prgeneration,
1867                         resp->prin_descriptor.prin_readkeys.additional_length );
1868
1869         if (resp->prin_descriptor.prin_readkeys.additional_length == 0 )
1870         {
1871                 condlog(1, "%s: No key found. Device may not be registered.", pp->dev);
1872                 ret = MPATH_PR_SUCCESS;
1873                 goto out;
1874         }
1875         prkey = 0;
1876         keyp = (unsigned char *)mpp->reservation_key;
1877         for (j = 0; j < 8; ++j) {
1878                 if (j > 0)
1879                         prkey <<= 8;
1880                 prkey |= *keyp;
1881                 ++keyp;
1882         }
1883         condlog(2, "Multipath  reservation_key: 0x%" PRIx64 " ", prkey);
1884
1885         isFound =0;
1886         for (i = 0; i < resp->prin_descriptor.prin_readkeys.additional_length/8; i++ )
1887         {
1888                 condlog(2, "PR IN READKEYS[%d]  reservation key:\n",i);
1889                 dumpHex((char *)&resp->prin_descriptor.prin_readkeys.key_list[i*8], 8 , -1);
1890                 if (!memcmp(mpp->reservation_key, &resp->prin_descriptor.prin_readkeys.key_list[i*8], 8))
1891                 {
1892                         condlog(2, "%s: pr key found in prin readkeys response", mpp->alias);
1893                         isFound =1;
1894                         break;
1895                 }
1896         }
1897         if (!isFound)
1898         {
1899                 condlog(0, "%s: Either device not registered or ", pp->dev);
1900                 condlog(0, "host is not authorised for registration. Skip path\n");
1901                 ret = MPATH_PR_OTHER;
1902                 goto out;
1903         }
1904
1905         param= malloc(sizeof(struct prout_param_descriptor));
1906         memset(param, 0 , sizeof(struct prout_param_descriptor));
1907
1908         for (j = 7; j >= 0; --j) {
1909                 param->sa_key[j] = (prkey & 0xff);
1910                 prkey >>= 8;
1911         }
1912         param->num_transportid = 0;
1913
1914         condlog(3, "device %s:%s \n", pp->dev, pp->mpp->wwid);
1915
1916         ret = prout_do_scsi_ioctl(pp->dev, MPATH_PROUT_REG_IGN_SA, 0, 0, param, 0);
1917         if (ret != MPATH_PR_SUCCESS )
1918         {
1919                 condlog(0,"%s: Reservation registration failed. Error: %d\n", pp->dev, ret);
1920         }
1921         mpp->prflag = 1;
1922
1923         free(param);
1924 out:
1925         free(resp);
1926         return NULL;
1927 }
1928
1929 int mpath_pr_event_handle(struct path *pp)
1930 {
1931         pthread_t thread;
1932         int rc;
1933         pthread_attr_t attr;
1934         struct multipath * mpp;
1935
1936         mpp = pp->mpp;
1937
1938         if (!mpp->reservation_key)
1939                 return -1;
1940
1941         pthread_attr_init(&attr);
1942         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
1943
1944         rc = pthread_create(&thread, NULL , mpath_pr_event_handler_fn, pp);
1945         if (rc) {
1946                 condlog(0, "%s: ERROR; return code from pthread_create() is %d\n", pp->dev, rc);
1947                 return -1;
1948         }
1949         pthread_attr_destroy(&attr);
1950         rc = pthread_join(thread, NULL);
1951         return 0;
1952 }
1953