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
9 #include <libdevmapper.h>
12 #include <sys/types.h>
19 #include <sysfs/libsysfs.h>
20 #include <sysfs/dlist.h>
26 #include <path_state.h>
37 #include <blacklist.h>
41 #include <structs_vec.h>
43 #include <devmapper.h>
45 #include <discovery.h>
49 #include <switchgroup.h>
50 #include <path_state.h>
52 #include <configure.h>
59 #include "cli_handlers.h"
61 #define FILE_NAME_SIZE 256
64 #define LOG_MSG(a,b) \
66 condlog(a, "%s: %s", pp->dev_t, b); \
67 memset(b, 0, MAX_CHECKER_MSG_SIZE); \
72 fprintf(stderr, "%s:%s(%i) lock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
75 fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
76 pthread_mutex_unlock(a)
77 #define lock_cleanup_pop(a) \
78 fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
79 pthread_cleanup_pop(1);
81 #define lock(a) pthread_mutex_lock(a)
82 #define unlock(a) pthread_mutex_unlock(a)
83 #define lock_cleanup_pop(a) pthread_cleanup_pop(1);
86 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
87 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
92 struct vectors * gvecs; /* global copy of vecs for use in sig handlers */
94 static struct event_thread *
98 struct event_thread * wp;
100 wp = (struct event_thread *)MALLOC(sizeof(struct event_thread));
106 free_waiter (void * data)
108 struct event_thread * wp = (struct event_thread *)data;
111 * indicate in mpp that the wp is already freed storage
113 lock(wp->vecs->lock);
117 * be careful, mpp may already be freed -- null if so
119 wp->mpp->waiter = NULL;
121 condlog(3, "free_waiter, mpp freed before wp=%p,", wp);
123 unlock(wp->vecs->lock);
126 dm_task_destroy(wp->dmt);
132 stop_waiter_thread (struct multipath * mpp, struct vectors * vecs)
134 struct event_thread * wp = (struct event_thread *)mpp->waiter;
137 condlog(3, "%s: no waiter thread", mpp->alias);
140 condlog(2, "%s: stop event checker thread", wp->mapname);
141 pthread_kill((pthread_t)wp->thread, SIGUSR1);
145 cleanup_lock (void * data)
147 unlock((pthread_mutex_t *)data);
151 * mpp->no_path_retry:
152 * -2 (QUEUE) : queue_if_no_path enabled, never turned off
153 * -1 (FAIL) : fail_if_no_path
154 * 0 (UNDEF) : nothing
155 * >0 : queue_if_no_path enabled, turned off after polling n times
158 update_queue_mode_del_path(struct multipath *mpp)
160 if (--mpp->nr_active == 0 && mpp->no_path_retry > 0) {
163 * meaning of +1: retry_tick may be decremented in
164 * checkerloop before starting retry.
166 mpp->stat_queueing_timeouts++;
167 mpp->retry_tick = mpp->no_path_retry * conf->checkint + 1;
168 condlog(1, "%s: Entering recovery mode: max_retries=%d",
169 mpp->alias, mpp->no_path_retry);
171 condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
175 update_queue_mode_add_path(struct multipath *mpp)
177 if (mpp->nr_active++ == 0 && mpp->no_path_retry > 0) {
178 /* come back to normal mode from retry mode */
180 dm_queue_if_no_path(mpp->alias, 1);
181 condlog(2, "%s: queue_if_no_path enabled", mpp->alias);
182 condlog(1, "%s: Recovered to normal mode", mpp->alias);
184 condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
188 need_switch_pathgroup (struct multipath * mpp, int refresh)
190 struct pathgroup * pgp;
194 if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
198 * Refresh path priority values
201 vector_foreach_slot (mpp->pg, pgp, i)
202 vector_foreach_slot (pgp->paths, pp, j)
203 pathinfo(pp, conf->hwtable, DI_PRIO);
205 mpp->bestpg = select_path_group(mpp);
207 if (mpp->bestpg != mpp->nextpg)
214 switch_pathgroup (struct multipath * mpp)
216 mpp->stat_switchgroup++;
217 dm_switchgroup(mpp->alias, mpp->bestpg);
218 condlog(2, "%s: switch to path group #%i",
219 mpp->alias, mpp->bestpg);
223 coalesce_maps(struct vectors *vecs, vector nmpv)
225 struct multipath * ompp;
226 vector ompv = vecs->mpvec;
229 condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(ompv));
230 condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(nmpv));
232 vector_foreach_slot (ompv, ompp, i) {
233 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
235 * remove all current maps not allowed by the
236 * current configuration
238 if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
239 condlog(0, "%s: unable to flush devmap",
242 * may be just because the device is open
244 if (!vector_alloc_slot(nmpv))
247 vector_set_slot(nmpv, ompp);
248 setup_multipath(vecs, ompp);
250 if ((j = find_slot(ompv, (void *)ompp)) != -1)
251 vector_del_slot(ompv, j);
257 condlog(3, "%s devmap removed", ompp->alias);
265 update_multipath (struct vectors *vecs, char *mapname)
267 struct multipath *mpp;
268 struct pathgroup *pgp;
273 mpp = find_mp_by_alias(vecs->mpvec, mapname);
278 free_pgvec(mpp->pg, KEEP_PATHS);
281 if (setup_multipath(vecs, mpp))
282 goto out; /* mpp freed in setup_multipath */
285 * compare checkers states with DM states
287 vector_foreach_slot (mpp->pg, pgp, i) {
288 vector_foreach_slot (pgp->paths, pp, j) {
289 if (pp->dmstate != PSTATE_FAILED)
292 if (pp->state != PATH_DOWN) {
293 condlog(2, "%s: mark as failed", pp->dev_t);
294 mpp->stat_path_failures++;
295 pp->state = PATH_DOWN;
296 update_queue_mode_del_path(mpp);
300 * schedule the next check earlier
302 if (pp->tick > conf->checkint)
303 pp->tick = conf->checkint;
310 condlog(0, "failed to update multipath");
315 static sigset_t unblock_signals(void)
320 sigaddset(&set, SIGHUP);
321 sigaddset(&set, SIGUSR1);
322 pthread_sigmask(SIG_UNBLOCK, &set, &old);
327 * returns the reschedule delay
328 * negative means *stop*
331 waiteventloop (struct event_thread * waiter)
337 if (!waiter->event_nr)
338 waiter->event_nr = dm_geteventnr(waiter->mapname);
340 if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT))) {
341 condlog(0, "%s: devmap event #%i dm_task_create error",
342 waiter->mapname, waiter->event_nr);
346 if (!dm_task_set_name(waiter->dmt, waiter->mapname)) {
347 condlog(0, "%s: devmap event #%i dm_task_set_name error",
348 waiter->mapname, waiter->event_nr);
349 dm_task_destroy(waiter->dmt);
353 if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
355 condlog(0, "%s: devmap event #%i dm_task_set_event_nr error",
356 waiter->mapname, waiter->event_nr);
357 dm_task_destroy(waiter->dmt);
361 dm_task_no_open_count(waiter->dmt);
363 /* accept wait interruption */
364 set = unblock_signals();
366 /* interruption spits messages */
370 r = dm_task_run(waiter->dmt);
372 /* wait is over : event or interrupt */
373 pthread_sigmask(SIG_SETMASK, &set, NULL);
376 if (!r) /* wait interrupted by signal */
379 dm_task_destroy(waiter->dmt);
387 condlog(3, "%s: devmap event #%i",
388 waiter->mapname, waiter->event_nr);
393 * 1) a table reload, which means our mpp structure is
394 * obsolete : refresh it through update_multipath()
395 * 2) a path failed by DM : mark as such through
397 * 3) map has gone away : stop the thread.
398 * 4) a path reinstate : nothing to do
399 * 5) a switch group : nothing to do
401 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
402 lock(waiter->vecs->lock);
403 r = update_multipath(waiter->vecs, waiter->mapname);
404 lock_cleanup_pop(waiter->vecs->lock);
407 return -1; /* stop the thread */
409 event_nr = dm_geteventnr(waiter->mapname);
411 if (waiter->event_nr == event_nr)
412 return 1; /* upon problem reschedule 1s later */
414 waiter->event_nr = event_nr;
416 return -1; /* never reach there */
420 waitevent (void * et)
423 struct event_thread *waiter;
425 mlockall(MCL_CURRENT | MCL_FUTURE);
427 waiter = (struct event_thread *)et;
428 pthread_cleanup_push(free_waiter, et);
431 r = waiteventloop(waiter);
439 pthread_cleanup_pop(1);
444 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
447 struct event_thread * wp;
452 if (pthread_attr_init(&attr))
455 pthread_attr_setstacksize(&attr, 32 * 1024);
456 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
463 mpp->waiter = (void *)wp;
464 strncpy(wp->mapname, mpp->alias, WWID_SIZE);
468 if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
469 condlog(0, "%s: cannot create event checker", wp->mapname);
472 condlog(2, "%s: event checker started", wp->mapname);
479 condlog(0, "failed to start waiter thread");
484 flush_map(struct multipath * mpp, struct vectors * vecs)
487 * clear references to this map before flushing so we can ignore
488 * the spurious uevent we may generate with the dm_flush_map call below
490 if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
492 * May not really be an error -- if the map was already flushed
493 * from the device mapper by dmsetup(8) for instance.
495 condlog(0, "%s: can't flush", mpp->alias);
500 condlog(3, "%s: devmap removed", mpp->alias);
503 orphan_paths(vecs->pathvec, mpp);
504 remove_map(mpp, vecs, stop_waiter_thread, 1);
510 uev_add_map (char * devname, struct vectors * vecs)
512 condlog(2, "%s: add map (uevent)", devname);
513 return ev_add_map(devname, vecs);
517 ev_add_map (char * devname, struct vectors * vecs)
520 char dev_t[BLK_DEV_SIZE];
523 struct multipath * mpp;
527 if (sscanf(devname, "dm-%d", &minor) == 1 &&
528 !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
529 sscanf(dev_t, "%d:%d", &major, &minor) == 2)
530 alias = dm_mapname(major, minor);
532 alias = STRDUP(devname);
537 map_present = dm_map_present(alias);
539 if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
540 condlog(4, "%s: not a multipath map", alias);
545 mpp = find_mp_by_alias(vecs->mpvec, alias);
549 * Not really an error -- we generate our own uevent
550 * if we create a multipath mapped device as a result
553 condlog(0, "%s: devmap already registered",
560 * now we can register the map
562 if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
563 start_waiter_thread))) {
564 condlog(3, "%s: devmap %s added", alias, devname);
567 refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
570 r = coalesce_paths(vecs, NULL, refwwid);
575 condlog(3, "%s: devmap %s added", alias, devname);
577 condlog(0, "%s: uev_add_map %s failed", alias, devname);
585 uev_remove_map (char * devname, struct vectors * vecs)
587 condlog(2, "%s: remove map (uevent)", devname);
588 return ev_remove_map(devname, vecs);
592 ev_remove_map (char * devname, struct vectors * vecs)
594 struct multipath * mpp;
596 mpp = find_mp_by_str(vecs->mpvec, devname);
599 condlog(3, "%s: devmap not registered, can't remove",
603 flush_map(mpp, vecs);
609 uev_umount_map (char * devname, struct vectors * vecs)
612 struct multipath * mpp;
614 condlog(2, "%s: umount map (uevent)", devname);
616 mpp = find_mp_by_str(vecs->mpvec, devname);
621 update_mpp_paths(mpp, vecs->pathvec);
622 verify_paths(mpp, vecs, NULL);
624 if (!VECTOR_SIZE(mpp->paths))
625 flush_map(mpp, vecs);
631 uev_add_path (char * devname, struct vectors * vecs)
633 condlog(2, "%s: add path (uevent)", devname);
634 return ev_add_path(devname, vecs);
638 ev_add_path (char * devname, struct vectors * vecs)
640 struct multipath * mpp;
642 char empty_buff[WWID_SIZE] = {0};
644 pp = find_path_by_dev(vecs->pathvec, devname);
647 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
650 * allow reconfig of orphaned path here
652 if (pp->mpp) return 1;
656 * get path vital state
658 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
660 condlog(0, "%s: failed to store path info", devname);
663 pp->checkint = conf->checkint;
667 * need path UID to go any further
669 if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
670 condlog(0, "%s: failed to get path uid", devname);
671 return 1; /* leave path added to pathvec */
674 mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
677 if (adopt_paths(vecs->pathvec, mpp))
678 return 1; /* leave path added to pathvec */
680 verify_paths(mpp, vecs, NULL);
681 condlog(0, "%s: ownership set to %s", pp->dev_t, mpp->alias);
682 mpp->action = ACT_RELOAD;
685 if ((mpp = add_map_with_path(vecs, pp, 1)))
686 mpp->action = ACT_CREATE;
688 return 1; /* leave path added to pathvec */
692 * push the map to the device-mapper
694 if (setup_map(mpp)) {
695 condlog(0, "%s: failed to setup map for addition of new "
696 "path %s", mpp->alias, devname);
700 * reload the map for the multipath mapped device
702 if (domap(mpp) <= 0) {
703 condlog(0, "%s: failed in domap for addition of new "
704 "path %s", mpp->alias, devname);
706 * deal with asynchronous uevents :((
708 if (mpp->action == ACT_RELOAD) {
709 condlog(0, "%s: uev_add_path sleep", mpp->alias);
711 update_mpp_paths(mpp, vecs->pathvec);
720 * update our state from kernel regardless of create or reload
722 if (setup_multipath(vecs, mpp))
725 if (mpp->action == ACT_CREATE &&
726 start_waiter_thread(mpp, vecs))
729 condlog(3, "%s path added to devmap %s", devname, mpp->alias);
733 remove_map(mpp, vecs, NULL, 1);
738 uev_remove_path (char * devname, struct vectors * vecs)
740 condlog(2, "%s: remove path (uevent)", devname);
741 return ev_remove_path(devname, vecs);
745 ev_remove_path (char * devname, struct vectors * vecs)
747 struct multipath * mpp;
752 pp = find_path_by_dev(vecs->pathvec, devname);
755 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
760 * avoid referring to the map of an orphanned path
762 if ((mpp = pp->mpp)) {
765 * remove the map IFF removing the last path
767 if (pathcount(mpp, PATH_WILD) > 1) {
768 vector rpvec = vector_alloc();
771 * transform the mp->pg vector of vectors of paths
772 * into a mp->params string to feed the device-mapper
774 update_mpp_paths(mpp, vecs->pathvec);
775 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
776 vector_del_slot(mpp->paths, i);
778 if (VECTOR_SIZE(mpp->paths) == 0) {
779 char alias[WWID_SIZE];
782 * flush_map will fail if the device is open
784 strncpy(alias, mpp->alias, WWID_SIZE);
785 if (flush_map(mpp, vecs))
788 condlog(3, "%s: removed map after removing"
789 " multiple paths", alias);
792 if (setup_map(mpp)) {
793 condlog(0, "%s: failed to setup map for"
794 " removal of path %s", mpp->alias, devname);
795 free_pathvec(rpvec, KEEP_PATHS);
801 mpp->action = ACT_RELOAD;
802 if (domap(mpp) <= 0) {
803 condlog(0, "%s: failed in domap for "
804 "removal of path %s",
805 mpp->alias, devname);
807 * Delete path from pathvec so that
808 * update_mpp_paths wont find it later
809 * when/if another path is removed.
811 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
812 vector_del_slot(vecs->pathvec, i);
817 * update our state from kernel
819 if (setup_multipath(vecs, mpp)) {
820 free_pathvec(rpvec, KEEP_PATHS);
824 condlog(3, "%s path removed from devmap %s",
825 devname, mpp->alias);
827 free_pathvec(rpvec, KEEP_PATHS);
830 char alias[WWID_SIZE];
833 * flush_map will fail if the device is open
835 strncpy(alias, mpp->alias, WWID_SIZE);
836 if (flush_map(mpp, vecs))
839 condlog(3, "%s: removed map", alias);
844 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
845 vector_del_slot(vecs->pathvec, i);
852 remove_map(mpp, vecs, stop_waiter_thread, 1);
857 map_discovery (struct vectors * vecs)
860 struct multipath * mpp;
862 if (dm_get_maps(vecs->mpvec, "multipath"))
865 vector_foreach_slot (vecs->mpvec, mpp, i)
866 if (setup_multipath(vecs, mpp))
873 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
875 struct vectors * vecs;
880 vecs = (struct vectors *)trigger_data;
882 pthread_cleanup_push(cleanup_lock, vecs->lock);
885 r = parse_cmd(str, reply, len, vecs);
888 *reply = STRDUP("fail\n");
889 *len = strlen(*reply) + 1;
892 else if (!r && *len == 0) {
893 *reply = STRDUP("ok\n");
894 *len = strlen(*reply) + 1;
897 /* else if (r < 0) leave *reply alone */
899 lock_cleanup_pop(vecs->lock);
904 uev_discard(char * devpath)
909 * keep only block devices, discard partitions
911 if (sscanf(devpath, "/block/%10s", a) != 1 ||
912 sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
913 condlog(4, "discard event on %s", devpath);
920 uev_trigger (struct uevent * uev, void * trigger_data)
924 struct vectors * vecs;
926 vecs = (struct vectors *)trigger_data;
928 if (uev_discard(uev->devpath))
931 basename(uev->devpath, devname);
935 * device map add/remove event
937 if (!strncmp(devname, "dm-", 3)) {
938 if (!strncmp(uev->action, "add", 3)) {
939 r = uev_add_map(devname, vecs);
942 if (!strncmp(uev->action, "remove", 6)) {
943 r = uev_remove_map(devname, vecs);
946 if (!strncmp(uev->action, "umount", 6)) {
947 r = uev_umount_map(devname, vecs);
954 * path add/remove event
956 if (blacklist(conf->blist, devname))
959 if (!strncmp(uev->action, "add", 3)) {
960 r = uev_add_path(devname, vecs);
963 if (!strncmp(uev->action, "remove", 6)) {
964 r = uev_remove_path(devname, vecs);
974 ueventloop (void * ap)
976 if (uevent_listen(&uev_trigger, ap))
977 fprintf(stderr, "error starting uevent listener");
983 uxlsnrloop (void * ap)
988 if (alloc_handlers())
991 add_handler(LIST+PATHS, cli_list_paths);
992 add_handler(LIST+MAPS, cli_list_maps);
993 add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
994 add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
995 add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
996 add_handler(ADD+PATH, cli_add_path);
997 add_handler(DEL+PATH, cli_del_path);
998 add_handler(ADD+MAP, cli_add_map);
999 add_handler(DEL+MAP, cli_del_map);
1000 add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1001 add_handler(RECONFIGURE, cli_reconfigure);
1002 add_handler(SUSPEND+MAP, cli_suspend);
1003 add_handler(RESUME+MAP, cli_resume);
1004 add_handler(REINSTATE+PATH, cli_reinstate);
1005 add_handler(FAIL+PATH, cli_fail);
1007 uxsock_listen(&uxsock_trigger, ap);
1013 exit_daemon (int status)
1016 fprintf(stderr, "bad exit status. see daemon.log\n");
1018 condlog(3, "unlink pidfile");
1019 unlink(DEFAULT_PIDFILE);
1022 pthread_cond_signal(&exit_cond);
1023 unlock(&exit_mutex);
1029 fail_path (struct path * pp)
1034 condlog(2, "checker failed path %s in map %s",
1035 pp->dev_t, pp->mpp->alias);
1037 dm_fail_path(pp->mpp->alias, pp->dev_t);
1038 update_queue_mode_del_path(pp->mpp);
1042 * caller must have locked the path list before calling that function
1045 reinstate_path (struct path * pp)
1050 if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1051 condlog(0, "%s: reinstate failed", pp->dev_t);
1053 condlog(2, "%s: reinstated", pp->dev_t);
1054 update_queue_mode_add_path(pp->mpp);
1059 enable_group(struct path * pp)
1061 struct pathgroup * pgp;
1064 * if path is added through uev_add_path, pgindex can be unset.
1065 * next update_strings() will set it, upon map reload event.
1067 * we can safely return here, because upon map reload, all
1068 * PG will be enabled.
1070 if (!pp->mpp->pg || !pp->pgindex)
1073 pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1075 if (pgp->status == PGSTATE_DISABLED) {
1076 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1077 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1082 mpvec_garbage_collector (struct vectors * vecs)
1084 struct multipath * mpp;
1087 vector_foreach_slot (vecs->mpvec, mpp, i) {
1088 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1089 condlog(2, "%s: remove dead map", mpp->alias);
1090 remove_map(mpp, vecs, stop_waiter_thread, 1);
1097 defered_failback_tick (vector mpvec)
1099 struct multipath * mpp;
1102 vector_foreach_slot (mpvec, mpp, i) {
1104 * defered failback getting sooner
1106 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1107 mpp->failback_tick--;
1109 if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1110 switch_pathgroup(mpp);
1116 retry_count_tick(vector mpvec)
1118 struct multipath *mpp;
1121 vector_foreach_slot (mpvec, mpp, i) {
1122 if (mpp->retry_tick) {
1123 mpp->stat_total_queueing_time++;
1124 condlog(4, "%s: Retrying.. No active path", mpp->alias);
1125 if(--mpp->retry_tick == 0) {
1126 dm_queue_if_no_path(mpp->alias, 0);
1127 condlog(2, "%s: Disable queueing", mpp->alias);
1134 checkerloop (void *ap)
1136 struct vectors *vecs;
1140 char checker_msg[MAX_CHECKER_MSG_SIZE];
1142 mlockall(MCL_CURRENT | MCL_FUTURE);
1144 memset(checker_msg, 0, MAX_CHECKER_MSG_SIZE);
1145 vecs = (struct vectors *)ap;
1147 condlog(2, "path checkers start up");
1150 * init the path check interval
1152 vector_foreach_slot (vecs->pathvec, pp, i) {
1153 pp->checkint = conf->checkint;
1157 pthread_cleanup_push(cleanup_lock, vecs->lock);
1161 vector_foreach_slot (vecs->pathvec, pp, i) {
1165 if (pp->tick && --pp->tick)
1166 continue; /* don't check this path yet */
1169 * provision a next check soonest,
1170 * in case we exit abnormaly from here
1172 pp->tick = conf->checkint;
1175 pathinfo(pp, conf->hwtable, DI_SYSFS);
1180 condlog(0, "%s: checkfn is void", pp->dev);
1183 newstate = pp->checkfn(pp->fd, checker_msg,
1184 &pp->checker_context);
1187 condlog(2, "%s: unusable path", pp->dev);
1188 pathinfo(pp, conf->hwtable, 0);
1192 if (newstate != pp->state) {
1193 pp->state = newstate;
1194 LOG_MSG(1, checker_msg);
1197 * upon state change, reset the checkint
1198 * to the shortest delay
1200 pp->checkint = conf->checkint;
1202 if (newstate == PATH_DOWN ||
1203 newstate == PATH_SHAKY ||
1204 update_multipath_strings(pp->mpp,
1207 * proactively fail path in the DM
1212 * cancel scheduled failback
1214 pp->mpp->failback_tick = 0;
1216 pp->mpp->stat_path_failures++;
1221 * reinstate this path
1226 * schedule [defered] failback
1228 if (pp->mpp->pgfailback > 0)
1229 pp->mpp->failback_tick =
1230 pp->mpp->pgfailback + 1;
1231 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1232 need_switch_pathgroup(pp->mpp, 1))
1233 switch_pathgroup(pp->mpp);
1236 * if at least one path is up in a group, and
1237 * the group is disabled, re-enable it
1239 if (newstate == PATH_UP)
1242 else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1243 LOG_MSG(4, checker_msg);
1245 * double the next check delay.
1246 * max at conf->max_checkint
1248 if (pp->checkint < (conf->max_checkint / 2))
1249 pp->checkint = 2 * pp->checkint;
1251 pp->checkint = conf->max_checkint;
1253 pp->tick = pp->checkint;
1254 condlog(4, "%s: delay next check %is",
1255 pp->dev_t, pp->tick);
1258 pp->state = newstate;
1261 * path prio refreshing
1263 condlog(4, "path prio refresh");
1264 pathinfo(pp, conf->hwtable, DI_PRIO);
1266 if (need_switch_pathgroup(pp->mpp, 0)) {
1267 if (pp->mpp->pgfailback > 0)
1268 pp->mpp->failback_tick =
1269 pp->mpp->pgfailback + 1;
1270 else if (pp->mpp->pgfailback ==
1271 -FAILBACK_IMMEDIATE)
1272 switch_pathgroup(pp->mpp);
1275 defered_failback_tick(vecs->mpvec);
1276 retry_count_tick(vecs->mpvec);
1281 condlog(4, "map garbage collection");
1282 mpvec_garbage_collector(vecs);
1286 lock_cleanup_pop(vecs->lock);
1293 configure (struct vectors * vecs, int start_waiters)
1295 struct multipath * mpp;
1300 if (!(vecs->pathvec = vector_alloc()))
1303 if (!(vecs->mpvec = vector_alloc()))
1306 if (!(mpvec = vector_alloc()))
1310 * probe for current path (from sysfs) and map (from dm) sets
1312 path_discovery(vecs->pathvec, conf, DI_ALL);
1314 vector_foreach_slot (vecs->pathvec, pp, i)
1315 pp->checkint = conf->checkint;
1317 if (map_discovery(vecs))
1321 * create new set of maps & push changed ones into dm
1323 if (coalesce_paths(vecs, mpvec, NULL))
1327 * may need to remove some maps which are no longer relevant
1328 * e.g., due to blacklist changes in conf file
1330 if (coalesce_maps(vecs, mpvec))
1335 if (conf->verbosity > 2)
1336 vector_foreach_slot(mpvec, mpp, i)
1340 * purge dm of old maps
1342 remove_maps(vecs, NULL);
1345 * save new set of maps formed by considering current path state
1347 vecs->mpvec = mpvec;
1350 * start dm event waiter threads for these new maps
1352 vector_foreach_slot(vecs->mpvec, mpp, i) {
1353 if (setup_multipath(vecs, mpp))
1356 if (start_waiter_thread(mpp, vecs))
1363 reconfigure (struct vectors * vecs)
1365 struct config * old = conf;
1368 * free old map and path vectors ... they use old conf state
1370 if (VECTOR_SIZE(vecs->mpvec))
1371 remove_maps(vecs, stop_waiter_thread);
1373 if (VECTOR_SIZE(vecs->pathvec))
1374 free_pathvec(vecs->pathvec, FREE_PATHS);
1378 if (load_config(DEFAULT_CONFIGFILE))
1381 conf->verbosity = old->verbosity;
1383 if (!conf->checkint) {
1384 conf->checkint = CHECKINT;
1385 conf->max_checkint = MAX_CHECKINT;
1392 static struct vectors *
1395 struct vectors * vecs;
1397 vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1403 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1408 vecs->pathvec = vector_alloc();
1413 vecs->mpvec = vector_alloc();
1418 pthread_mutex_init(vecs->lock, NULL);
1423 vector_free(vecs->pathvec);
1428 condlog(0, "failed to init paths");
1433 signal_set(int signo, void (*func) (int))
1436 struct sigaction sig;
1437 struct sigaction osig;
1439 sig.sa_handler = func;
1440 sigemptyset(&sig.sa_mask);
1443 r = sigaction(signo, &sig, &osig);
1448 return (osig.sa_handler);
1454 condlog(2, "reconfigure (SIGHUP)");
1458 unlock(gvecs->lock);
1461 dbg_free_final(NULL);
1474 condlog(3, "SIGUSR1 received");
1480 signal_set(SIGHUP, sighup);
1481 signal_set(SIGUSR1, sigusr1);
1482 signal_set(SIGINT, sigend);
1483 signal_set(SIGTERM, sigend);
1484 signal_set(SIGKILL, sigend);
1491 static struct sched_param sched_param = {
1495 res = sched_setscheduler (0, SCHED_RR, &sched_param);
1498 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1503 set_oom_adj (int val)
1507 fp = fopen("/proc/self/oom_adj", "w");
1512 fprintf(fp, "%i", val);
1517 child (void * param)
1519 pthread_t check_thr, uevent_thr, uxlsnr_thr;
1520 pthread_attr_t attr;
1521 struct vectors * vecs;
1523 mlockall(MCL_CURRENT | MCL_FUTURE);
1528 condlog(2, "--------start up--------");
1529 condlog(2, "read " DEFAULT_CONFIGFILE);
1531 if (load_config(DEFAULT_CONFIGFILE))
1534 setlogmask(LOG_UPTO(conf->verbosity + 3));
1537 * fill the voids left in the config file
1539 if (!conf->checkint) {
1540 conf->checkint = CHECKINT;
1541 conf->max_checkint = MAX_CHECKINT;
1544 if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1553 vecs = gvecs = init_vecs();
1558 if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1559 condlog(0, "can not find sysfs mount point");
1564 * fetch and configure both paths and multipaths
1566 if (configure(vecs, 1)) {
1567 condlog(0, "failure during configuration");
1574 pthread_attr_init(&attr);
1575 pthread_attr_setstacksize(&attr, 64 * 1024);
1576 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1578 pthread_create(&check_thr, &attr, checkerloop, vecs);
1579 pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1580 pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1582 pthread_cond_wait(&exit_cond, &exit_mutex);
1588 remove_maps(vecs, stop_waiter_thread);
1589 free_pathvec(vecs->pathvec, FREE_PATHS);
1591 pthread_cancel(check_thr);
1592 pthread_cancel(uevent_thr);
1593 pthread_cancel(uxlsnr_thr);
1597 free_handlers(handlers);
1602 pthread_mutex_destroy(vecs->lock);
1610 condlog(2, "--------shut down-------");
1619 dbg_free_final(NULL);
1631 if( (pid = fork()) < 0){
1632 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1640 if ( (pid = fork()) < 0)
1641 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1645 in_fd = open("/dev/null", O_RDONLY);
1647 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1651 out_fd = open("/dev/console", O_WRONLY);
1653 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1658 close(STDIN_FILENO);
1660 close(STDOUT_FILENO);
1662 close(STDERR_FILENO);
1673 main (int argc, char *argv[])
1675 extern char *optarg;
1682 if (getuid() != 0) {
1683 fprintf(stderr, "need to be root\n");
1687 /* make sure we don't lock any path */
1689 umask(umask(077) | 022);
1691 conf = alloc_config();
1696 while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1700 //debug=1; /* ### comment me out ### */
1703 if (sizeof(optarg) > sizeof(char *) ||
1704 !isdigit(optarg[0]))
1707 conf->verbosity = atoi(optarg);
1730 return (child(NULL));