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>
16 #include <sys/resource.h>
36 #include <blacklist.h>
37 #include <structs_vec.h>
39 #include <devmapper.h>
42 #include <discovery.h>
46 #include <switchgroup.h>
48 #include <configure.h>
50 #include <pgpolicies.h>
58 #include "cli_handlers.h"
62 #define FILE_NAME_SIZE 256
65 #define LOG_MSG(a,b) \
66 if (strlen(b)) condlog(a, "%s: %s - %s", pp->mpp->alias, pp->dev, b);
68 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
69 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
72 enum daemon_status running_state;
76 * global copy of vecs for use in sig handlers
78 struct vectors * gvecs;
81 need_switch_pathgroup (struct multipath * mpp, int refresh)
83 struct pathgroup * pgp;
87 if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
91 * Refresh path priority values
94 vector_foreach_slot (mpp->pg, pgp, i)
95 vector_foreach_slot (pgp->paths, pp, j)
96 pathinfo(pp, conf->hwtable, DI_PRIO);
98 mpp->bestpg = select_path_group(mpp);
100 if (mpp->bestpg != mpp->nextpg)
107 switch_pathgroup (struct multipath * mpp)
109 mpp->stat_switchgroup++;
110 dm_switchgroup(mpp->alias, mpp->bestpg);
111 condlog(2, "%s: switch to path group #%i",
112 mpp->alias, mpp->bestpg);
116 coalesce_maps(struct vectors *vecs, vector nmpv)
118 struct multipath * ompp;
119 vector ompv = vecs->mpvec;
123 vector_foreach_slot (ompv, ompp, i) {
124 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
126 * remove all current maps not allowed by the
127 * current configuration
129 if (dm_flush_map(ompp->alias)) {
130 condlog(0, "%s: unable to flush devmap",
133 * may be just because the device is open
135 if (!vector_alloc_slot(nmpv))
138 vector_set_slot(nmpv, ompp);
139 setup_multipath(vecs, ompp);
141 if ((j = find_slot(ompv, (void *)ompp)) != -1)
142 vector_del_slot(ompv, j);
148 condlog(2, "%s devmap removed", ompp->alias);
150 } else if (conf->reassign_maps) {
151 condlog(3, "%s: Reassign existing device-mapper"
152 " devices", ompp->alias);
153 dm_reassign(ompp->alias);
160 sync_map_state(struct multipath *mpp)
162 struct pathgroup *pgp;
169 vector_foreach_slot (mpp->pg, pgp, i){
170 vector_foreach_slot (pgp->paths, pp, j){
171 if (pp->state == PATH_UNCHECKED ||
172 pp->state == PATH_WILD)
174 if ((pp->dmstate == PSTATE_FAILED ||
175 pp->dmstate == PSTATE_UNDEF) &&
176 (pp->state == PATH_UP || pp->state == PATH_GHOST))
177 dm_reinstate_path(mpp->alias, pp->dev_t);
178 else if ((pp->dmstate == PSTATE_ACTIVE ||
179 pp->dmstate == PSTATE_UNDEF) &&
180 (pp->state == PATH_DOWN ||
181 pp->state == PATH_SHAKY))
182 dm_fail_path(mpp->alias, pp->dev_t);
188 sync_maps_state(vector mpvec)
191 struct multipath *mpp;
193 vector_foreach_slot (mpvec, mpp, i)
198 flush_map(struct multipath * mpp, struct vectors * vecs)
201 * clear references to this map before flushing so we can ignore
202 * the spurious uevent we may generate with the dm_flush_map call below
204 if (dm_flush_map(mpp->alias)) {
206 * May not really be an error -- if the map was already flushed
207 * from the device mapper by dmsetup(8) for instance.
209 condlog(0, "%s: can't flush", mpp->alias);
214 condlog(2, "%s: devmap removed", mpp->alias);
217 orphan_paths(vecs->pathvec, mpp);
218 remove_map_and_stop_waiter(mpp, vecs, 1);
224 uev_add_map (struct uevent * uev, struct vectors * vecs)
227 int major = -1, minor = -1, rc;
229 condlog(2, "%s: add map (uevent)", uev->kernel);
230 alias = uevent_get_dm_name(uev);
232 condlog(3, "%s: No DM_NAME in uevent", uev->kernel);
233 major = uevent_get_major(uev);
234 minor = uevent_get_minor(uev);
235 alias = dm_mapname(major, minor);
237 condlog(2, "%s: mapname not found for %d:%d",
238 uev->kernel, major, minor);
242 rc = ev_add_map(uev->kernel, alias, vecs);
248 ev_add_map (char * dev, char * alias, struct vectors * vecs)
251 struct multipath * mpp;
255 map_present = dm_map_present(alias);
257 if (map_present && dm_type(alias, TGT_MPATH) <= 0) {
258 condlog(4, "%s: not a multipath map", alias);
262 mpp = find_mp_by_alias(vecs->mpvec, alias);
266 * Not really an error -- we generate our own uevent
267 * if we create a multipath mapped device as a result
270 if (conf->reassign_maps) {
271 condlog(3, "%s: Reassign existing device-mapper devices",
280 * now we can register the map
282 if (map_present && (mpp = add_map_without_path(vecs, alias))) {
284 condlog(2, "%s: devmap %s registered", alias, dev);
287 refwwid = get_refwwid(dev, DEV_DEVMAP, vecs->pathvec);
290 r = coalesce_paths(vecs, NULL, refwwid, 0);
295 condlog(2, "%s: devmap %s added", alias, dev);
297 condlog(0, "%s: uev_add_map %s failed", alias, dev);
304 uev_remove_map (struct uevent * uev, struct vectors * vecs)
309 condlog(2, "%s: remove map (uevent)", uev->kernel);
310 alias = uevent_get_dm_name(uev);
312 condlog(3, "%s: No DM_NAME in uevent, ignoring", uev->kernel);
315 minor = uevent_get_minor(uev);
316 rc = ev_remove_map(uev->kernel, alias, minor, vecs);
322 ev_remove_map (char * devname, char * alias, int minor, struct vectors * vecs)
324 struct multipath * mpp;
326 mpp = find_mp_by_minor(vecs->mpvec, minor);
329 condlog(2, "%s: devmap not registered, can't remove",
333 if (strcmp(mpp->alias, alias)) {
334 condlog(2, "%s: minor number mismatch (map %d, event %d)",
335 mpp->alias, mpp->dmi->minor, minor);
338 return flush_map(mpp, vecs);
342 uev_add_path (struct uevent *uev, struct vectors * vecs)
344 struct sysfs_device * dev;
346 dev = sysfs_device_get(uev->devpath);
348 condlog(2, "%s: not found in sysfs", uev->devpath);
351 condlog(2, "%s: add path (uevent)", dev->kernel);
352 return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
362 ev_add_path (char * devname, struct vectors * vecs)
364 struct multipath * mpp;
366 char empty_buff[WWID_SIZE] = {0};
367 char params[PARAMS_SIZE] = {0};
368 int start_waiter = 0;
370 if (strstr(devname, "..") != NULL) {
372 * Don't allow relative device names in the pathvec
374 condlog(0, "%s: path name is invalid", devname);
378 pp = find_path_by_dev(vecs->pathvec, devname);
381 condlog(0, "%s: spurious uevent, path already in pathvec",
388 * get path vital state
390 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
392 condlog(0, "%s: failed to store path info", devname);
395 pp->checkint = conf->checkint;
399 * need path UID to go any further
401 if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
402 condlog(0, "%s: failed to get path uid", devname);
403 goto fail; /* leave path added to pathvec */
405 if (filter_path(conf, pp) > 0){
406 int i = find_slot(vecs->pathvec, (void *)pp);
408 vector_del_slot(vecs->pathvec, i);
412 mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
415 if ((!pp->size) || (mpp->size != pp->size)) {
417 condlog(0, "%s: failed to add new path %s, "
421 condlog(0, "%s: failed to add new path %s, "
422 "device size mismatch",
424 int i = find_slot(vecs->pathvec, (void *)pp);
426 vector_del_slot(vecs->pathvec, i);
431 condlog(4,"%s: adopting all paths for path %s",
432 mpp->alias, pp->dev);
433 if (adopt_paths(vecs->pathvec, mpp, 1))
434 goto fail; /* leave path added to pathvec */
436 verify_paths(mpp, vecs, NULL);
437 mpp->flush_on_last_del = FLUSH_UNDEF;
438 mpp->action = ACT_RELOAD;
442 condlog(0, "%s: failed to create new map,"
443 " %s device size is 0 ", devname, pp->dev);
444 int i = find_slot(vecs->pathvec, (void *)pp);
446 vector_del_slot(vecs->pathvec, i);
451 condlog(4,"%s: creating new map", pp->dev);
452 if ((mpp = add_map_with_path(vecs, pp, 1))) {
453 mpp->action = ACT_CREATE;
455 * We don't depend on ACT_CREATE, as domap will
456 * set it to ACT_NOTHING when complete.
461 goto fail; /* leave path added to pathvec */
465 * push the map to the device-mapper
467 if (setup_map(mpp, params, PARAMS_SIZE)) {
468 condlog(0, "%s: failed to setup map for addition of new "
469 "path %s", mpp->alias, devname);
473 * reload the map for the multipath mapped device
475 if (domap(mpp, params) <= 0) {
476 condlog(0, "%s: failed in domap for addition of new "
477 "path %s", mpp->alias, devname);
479 * deal with asynchronous uevents :((
481 if (mpp->action == ACT_RELOAD) {
482 condlog(0, "%s: uev_add_path sleep", mpp->alias);
484 update_mpp_paths(mpp, vecs->pathvec);
493 * update our state from kernel regardless of create or reload
495 if (setup_multipath(vecs, mpp))
500 if ((mpp->action == ACT_CREATE ||
501 (mpp->action == ACT_NOTHING && start_waiter && !mpp->waiter)) &&
502 start_waiter_thread(mpp, vecs))
505 condlog(2, "%s path added to devmap %s", devname, mpp->alias);
509 remove_map(mpp, vecs, 1);
516 uev_remove_path (struct uevent *uev, struct vectors * vecs)
518 struct sysfs_device * dev;
521 dev = sysfs_device_get(uev->devpath);
523 condlog(2, "%s: not found in sysfs", uev->devpath);
526 condlog(2, "%s: remove path (uevent)", uev->kernel);
527 retval = ev_remove_path(uev->kernel, vecs);
530 sysfs_device_put(dev);
536 ev_remove_path (char * devname, struct vectors * vecs)
538 struct multipath * mpp;
541 char params[PARAMS_SIZE] = {0};
543 pp = find_path_by_dev(vecs->pathvec, devname);
546 /* Not an error; path might have been purged earlier */
547 condlog(0, "%s: path already removed", devname);
552 * avoid referring to the map of an orphaned path
554 if ((mpp = pp->mpp)) {
556 * transform the mp->pg vector of vectors of paths
557 * into a mp->params string to feed the device-mapper
559 if (update_mpp_paths(mpp, vecs->pathvec)) {
560 condlog(0, "%s: failed to update paths",
564 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
565 vector_del_slot(mpp->paths, i);
568 * remove the map IFF removing the last path
570 if (VECTOR_SIZE(mpp->paths) == 0) {
571 char alias[WWID_SIZE];
574 * flush_map will fail if the device is open
576 strncpy(alias, mpp->alias, WWID_SIZE);
577 if (mpp->flush_on_last_del == FLUSH_ENABLED) {
578 condlog(2, "%s Last path deleted, disabling queueing", mpp->alias);
580 mpp->no_path_retry = NO_PATH_RETRY_FAIL;
581 mpp->flush_on_last_del = FLUSH_IN_PROGRESS;
582 dm_queue_if_no_path(mpp->alias, 0);
584 if (!flush_map(mpp, vecs)) {
585 condlog(2, "%s: removed map after"
586 " removing all paths",
592 * Not an error, continue
596 if (setup_map(mpp, params, PARAMS_SIZE)) {
597 condlog(0, "%s: failed to setup map for"
598 " removal of path %s", mpp->alias,
605 mpp->action = ACT_RELOAD;
606 if (domap(mpp, params) <= 0) {
607 condlog(0, "%s: failed in domap for "
608 "removal of path %s",
609 mpp->alias, devname);
613 * update our state from kernel
615 if (setup_multipath(vecs, mpp)) {
620 condlog(2, "%s: path removed from map %s",
621 devname, mpp->alias);
626 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
627 vector_del_slot(vecs->pathvec, i);
634 remove_map_and_stop_waiter(mpp, vecs, 1);
639 uev_update_path (struct uevent *uev, struct vectors * vecs)
641 struct sysfs_device * dev;
644 dev = sysfs_device_get(uev->devpath);
646 condlog(2, "%s: not found in sysfs", uev->devpath);
649 ro = uevent_get_disk_ro(uev);
654 condlog(2, "%s: update path write_protect to '%d' (uevent)",
656 pp = find_path_by_dev(vecs->pathvec, uev->kernel);
658 condlog(0, "%s: spurious uevent, path not found",
663 retval = reload_map(vecs, pp->mpp);
665 condlog(2, "%s: map %s reloaded (retval %d)",
666 uev->kernel, pp->mpp->alias, retval);
670 sysfs_device_put(dev);
676 map_discovery (struct vectors * vecs)
678 struct multipath * mpp;
681 if (dm_get_maps(vecs->mpvec))
684 vector_foreach_slot (vecs->mpvec, mpp, i)
685 if (setup_multipath(vecs, mpp))
692 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
694 struct vectors * vecs;
699 vecs = (struct vectors *)trigger_data;
701 pthread_cleanup_push(cleanup_lock, &vecs->lock);
704 r = parse_cmd(str, reply, len, vecs);
707 *reply = STRDUP("fail\n");
708 *len = strlen(*reply) + 1;
711 else if (!r && *len == 0) {
712 *reply = STRDUP("ok\n");
713 *len = strlen(*reply) + 1;
716 /* else if (r < 0) leave *reply alone */
718 lock_cleanup_pop(vecs->lock);
723 uev_discard(char * devpath)
729 * keep only block devices, discard partitions
731 tmp = strstr(devpath, "/block/");
733 condlog(4, "no /block/ in '%s'", devpath);
736 if (sscanf(tmp, "/block/%10s", a) != 1 ||
737 sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
738 condlog(4, "discard event on %s", devpath);
745 uev_trigger (struct uevent * uev, void * trigger_data)
748 struct vectors * vecs;
750 vecs = (struct vectors *)trigger_data;
752 if (uev_discard(uev->devpath))
759 * Add events are ignored here as the tables
760 * are not fully initialised then.
762 if (!strncmp(uev->kernel, "dm-", 3)) {
763 if (!strncmp(uev->action, "change", 6)) {
764 r = uev_add_map(uev, vecs);
767 if (!strncmp(uev->action, "remove", 6)) {
768 r = uev_remove_map(uev, vecs);
775 * path add/remove event
777 if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
781 if (!strncmp(uev->action, "add", 3)) {
782 r = uev_add_path(uev, vecs);
785 if (!strncmp(uev->action, "remove", 6)) {
786 r = uev_remove_path(uev, vecs);
789 if (!strncmp(uev->action, "change", 6)) {
790 r = uev_update_path(uev, vecs);
800 ueventloop (void * ap)
802 block_signal(SIGUSR1, NULL);
803 block_signal(SIGHUP, NULL);
806 condlog(0, "error starting uevent listener");
814 block_signal(SIGUSR1, NULL);
815 block_signal(SIGHUP, NULL);
817 if (uevent_dispatch(&uev_trigger, ap))
818 condlog(0, "error starting uevent dispatcher");
823 uxlsnrloop (void * ap)
825 block_signal(SIGUSR1, NULL);
826 block_signal(SIGHUP, NULL);
831 set_handler_callback(LIST+PATHS, cli_list_paths);
832 set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
833 set_handler_callback(LIST+MAPS, cli_list_maps);
834 set_handler_callback(LIST+STATUS, cli_list_status);
835 set_handler_callback(LIST+DAEMON, cli_list_daemon);
836 set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
837 set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
838 set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
839 set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
840 set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
841 set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
842 set_handler_callback(LIST+CONFIG, cli_list_config);
843 set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
844 set_handler_callback(LIST+DEVICES, cli_list_devices);
845 set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
846 set_handler_callback(ADD+PATH, cli_add_path);
847 set_handler_callback(DEL+PATH, cli_del_path);
848 set_handler_callback(ADD+MAP, cli_add_map);
849 set_handler_callback(DEL+MAP, cli_del_map);
850 set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
851 set_handler_callback(RECONFIGURE, cli_reconfigure);
852 set_handler_callback(SUSPEND+MAP, cli_suspend);
853 set_handler_callback(RESUME+MAP, cli_resume);
854 set_handler_callback(RESIZE+MAP, cli_resize);
855 set_handler_callback(RELOAD+MAP, cli_reload);
856 set_handler_callback(RESET+MAP, cli_reassign);
857 set_handler_callback(REINSTATE+PATH, cli_reinstate);
858 set_handler_callback(FAIL+PATH, cli_fail);
859 set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
860 set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
861 set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
862 set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
863 set_handler_callback(QUIT, cli_quit);
864 set_handler_callback(SHUTDOWN, cli_shutdown);
867 uxsock_listen(&uxsock_trigger, ap);
873 exit_daemon (int status)
876 fprintf(stderr, "bad exit status. see daemon.log\n");
878 condlog(3, "unlink pidfile");
879 unlink(DEFAULT_PIDFILE);
881 pthread_mutex_lock(&exit_mutex);
882 pthread_cond_signal(&exit_cond);
883 pthread_mutex_unlock(&exit_mutex);
891 switch (running_state) {
896 case DAEMON_CONFIGURE:
900 case DAEMON_SHUTDOWN:
907 fail_path (struct path * pp, int del_active)
912 condlog(2, "checker failed path %s in map %s",
913 pp->dev_t, pp->mpp->alias);
915 dm_fail_path(pp->mpp->alias, pp->dev_t);
917 update_queue_mode_del_path(pp->mpp);
921 * caller must have locked the path list before calling that function
924 reinstate_path (struct path * pp, int add_active)
929 if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
930 condlog(0, "%s: reinstate failed", pp->dev_t);
932 condlog(2, "%s: reinstated", pp->dev_t);
934 update_queue_mode_add_path(pp->mpp);
939 enable_group(struct path * pp)
941 struct pathgroup * pgp;
944 * if path is added through uev_add_path, pgindex can be unset.
945 * next update_strings() will set it, upon map reload event.
947 * we can safely return here, because upon map reload, all
948 * PG will be enabled.
950 if (!pp->mpp->pg || !pp->pgindex)
953 pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
955 if (pgp->status == PGSTATE_DISABLED) {
956 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
957 dm_enablegroup(pp->mpp->alias, pp->pgindex);
962 mpvec_garbage_collector (struct vectors * vecs)
964 struct multipath * mpp;
970 vector_foreach_slot (vecs->mpvec, mpp, i) {
971 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
972 condlog(2, "%s: remove dead map", mpp->alias);
973 remove_map_and_stop_waiter(mpp, vecs, 1);
980 defered_failback_tick (vector mpvec)
982 struct multipath * mpp;
985 vector_foreach_slot (mpvec, mpp, i) {
987 * defered failback getting sooner
989 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
990 mpp->failback_tick--;
992 if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
993 switch_pathgroup(mpp);
999 retry_count_tick(vector mpvec)
1001 struct multipath *mpp;
1004 vector_foreach_slot (mpvec, mpp, i) {
1005 if (mpp->retry_tick) {
1006 mpp->stat_total_queueing_time++;
1007 condlog(4, "%s: Retrying.. No active path", mpp->alias);
1008 if(--mpp->retry_tick == 0) {
1009 dm_queue_if_no_path(mpp->alias, 0);
1010 condlog(2, "%s: Disable queueing", mpp->alias);
1016 int update_prio(struct path *pp, int refresh_all)
1020 struct pathgroup * pgp;
1021 int i, j, changed = 0;
1024 vector_foreach_slot (pp->mpp->pg, pgp, i) {
1025 vector_foreach_slot (pgp->paths, pp1, j) {
1026 oldpriority = pp1->priority;
1027 pathinfo(pp1, conf->hwtable, DI_PRIO);
1028 if (pp1->priority != oldpriority)
1034 oldpriority = pp->priority;
1035 pathinfo(pp, conf->hwtable, DI_PRIO);
1037 if (pp->priority == oldpriority)
1042 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
1046 char params[PARAMS_SIZE];
1048 update_mpp_paths(mpp, vecs->pathvec);
1050 vector_foreach_slot (mpp->paths, pp, i)
1051 pathinfo(pp, conf->hwtable, DI_PRIO);
1054 if (setup_map(mpp, params, PARAMS_SIZE))
1057 mpp->action = ACT_RELOAD;
1058 if (domap(mpp, params) <= 0) {
1059 condlog(0, "%s: failed to update map : %s", mpp->alias,
1064 setup_multipath(vecs, mpp);
1065 sync_map_state(mpp);
1071 check_path (struct vectors * vecs, struct path * pp)
1074 int new_path_up = 0;
1079 if (pp->tick && --pp->tick)
1080 return; /* don't check this path yet */
1083 * provision a next check soonest,
1084 * in case we exit abnormaly from here
1086 pp->tick = conf->checkint;
1088 newstate = path_offline(pp);
1089 if (newstate == PATH_UP)
1090 newstate = get_state(pp, 1);
1092 if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1093 condlog(2, "%s: unusable path", pp->dev);
1094 pathinfo(pp, conf->hwtable, 0);
1098 * Async IO in flight. Keep the previous path state
1099 * and reschedule as soon as possible
1101 if (newstate == PATH_PENDING) {
1106 * Synchronize with kernel state
1108 if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1109 condlog(1, "%s: Could not synchronize with kernel state\n",
1111 pp->dmstate = PSTATE_UNDEF;
1113 if (newstate != pp->state) {
1114 int oldstate = pp->state;
1115 pp->state = newstate;
1116 LOG_MSG(1, checker_message(&pp->checker));
1119 * upon state change, reset the checkint
1120 * to the shortest delay
1122 pp->checkint = conf->checkint;
1124 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1126 * proactively fail path in the DM
1128 if (oldstate == PATH_UP ||
1129 oldstate == PATH_GHOST)
1135 * cancel scheduled failback
1137 pp->mpp->failback_tick = 0;
1139 pp->mpp->stat_path_failures++;
1144 * reinstate this path
1146 if (oldstate != PATH_UP &&
1147 oldstate != PATH_GHOST)
1148 reinstate_path(pp, 1);
1150 reinstate_path(pp, 0);
1155 * if at least one path is up in a group, and
1156 * the group is disabled, re-enable it
1158 if (newstate == PATH_UP)
1161 else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1162 if (pp->dmstate == PSTATE_FAILED ||
1163 pp->dmstate == PSTATE_UNDEF) {
1164 /* Clear IO errors */
1165 reinstate_path(pp, 0);
1167 LOG_MSG(4, checker_message(&pp->checker));
1169 * double the next check delay.
1170 * max at conf->max_checkint
1172 if (pp->checkint < (conf->max_checkint / 2))
1173 pp->checkint = 2 * pp->checkint;
1175 pp->checkint = conf->max_checkint;
1177 pp->tick = pp->checkint;
1178 condlog(4, "%s: delay next check %is",
1179 pp->dev_t, pp->tick);
1182 else if (newstate == PATH_DOWN)
1183 LOG_MSG(2, checker_message(&pp->checker));
1185 pp->state = newstate;
1188 * path prio refreshing
1190 condlog(4, "path prio refresh");
1192 if (update_prio(pp, new_path_up) &&
1193 pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio)
1194 update_path_groups(pp->mpp, vecs, !new_path_up);
1195 else if (need_switch_pathgroup(pp->mpp, 0)) {
1196 if (pp->mpp->pgfailback > 0 &&
1197 (new_path_up || pp->mpp->failback_tick <= 0))
1198 pp->mpp->failback_tick =
1199 pp->mpp->pgfailback + 1;
1200 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1201 switch_pathgroup(pp->mpp);
1206 checkerloop (void *ap)
1208 struct vectors *vecs;
1214 mlockall(MCL_CURRENT | MCL_FUTURE);
1215 vecs = (struct vectors *)ap;
1216 condlog(2, "path checkers start up");
1219 * init the path check interval
1221 vector_foreach_slot (vecs->pathvec, pp, i) {
1222 pp->checkint = conf->checkint;
1226 block_signal(SIGHUP, &old);
1227 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1231 if (vecs->pathvec) {
1232 vector_foreach_slot (vecs->pathvec, pp, i) {
1233 check_path(vecs, pp);
1237 defered_failback_tick(vecs->mpvec);
1238 retry_count_tick(vecs->mpvec);
1243 condlog(4, "map garbage collection");
1244 mpvec_garbage_collector(vecs);
1248 lock_cleanup_pop(vecs->lock);
1249 pthread_sigmask(SIG_SETMASK, &old, NULL);
1256 configure (struct vectors * vecs, int start_waiters)
1258 struct multipath * mpp;
1263 if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1266 if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1269 if (!(mpvec = vector_alloc()))
1273 * probe for current path (from sysfs) and map (from dm) sets
1275 path_discovery(vecs->pathvec, conf, DI_ALL);
1277 vector_foreach_slot (vecs->pathvec, pp, i){
1278 if (filter_path(conf, pp) > 0){
1279 vector_del_slot(vecs->pathvec, i);
1284 pp->checkint = conf->checkint;
1286 if (map_discovery(vecs))
1290 * create new set of maps & push changed ones into dm
1292 if (coalesce_paths(vecs, mpvec, NULL, 1))
1296 * may need to remove some maps which are no longer relevant
1297 * e.g., due to blacklist changes in conf file
1299 if (coalesce_maps(vecs, mpvec))
1304 sync_maps_state(mpvec);
1307 * purge dm of old maps
1312 * save new set of maps formed by considering current path state
1314 vector_free(vecs->mpvec);
1315 vecs->mpvec = mpvec;
1318 * start dm event waiter threads for these new maps
1320 vector_foreach_slot(vecs->mpvec, mpp, i) {
1321 if (setup_multipath(vecs, mpp))
1324 if (start_waiter_thread(mpp, vecs))
1331 reconfigure (struct vectors * vecs)
1333 struct config * old = conf;
1338 * free old map and path vectors ... they use old conf state
1340 if (VECTOR_SIZE(vecs->mpvec))
1341 remove_maps_and_stop_waiters(vecs);
1343 if (VECTOR_SIZE(vecs->pathvec))
1344 free_pathvec(vecs->pathvec, FREE_PATHS);
1346 vecs->pathvec = NULL;
1349 if (!load_config(DEFAULT_CONFIGFILE)) {
1350 conf->verbosity = old->verbosity;
1361 static struct vectors *
1364 struct vectors * vecs;
1366 vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1372 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1374 if (!vecs->lock.mutex)
1377 pthread_mutex_init(vecs->lock.mutex, NULL);
1378 vecs->lock.depth = 0;
1384 condlog(0, "failed to init paths");
1389 signal_set(int signo, void (*func) (int))
1392 struct sigaction sig;
1393 struct sigaction osig;
1395 sig.sa_handler = func;
1396 sigemptyset(&sig.sa_mask);
1399 r = sigaction(signo, &sig, &osig);
1404 return (osig.sa_handler);
1410 condlog(2, "reconfigure (SIGHUP)");
1412 if (running_state != DAEMON_RUNNING)
1418 dbg_free_final(NULL);
1431 condlog(3, "SIGUSR1 received");
1437 signal_set(SIGHUP, sighup);
1438 signal_set(SIGUSR1, sigusr1);
1439 signal_set(SIGINT, sigend);
1440 signal_set(SIGTERM, sigend);
1441 signal(SIGPIPE, SIG_IGN);
1448 static struct sched_param sched_param = {
1449 .sched_priority = 99
1452 res = sched_setscheduler (0, SCHED_RR, &sched_param);
1455 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1460 set_oom_adj (int val)
1464 fp = fopen("/proc/self/oom_adj", "w");
1469 fprintf(fp, "%i", val);
1474 child (void * param)
1476 pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1477 pthread_attr_t log_attr, misc_attr;
1478 struct vectors * vecs;
1479 struct multipath * mpp;
1483 mlockall(MCL_CURRENT | MCL_FUTURE);
1485 setup_thread_attr(&misc_attr, 64 * 1024, 1);
1486 setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1489 setup_thread_attr(&log_attr, 64 * 1024, 0);
1490 log_thread_start(&log_attr);
1491 pthread_attr_destroy(&log_attr);
1494 running_state = DAEMON_START;
1496 condlog(2, "--------start up--------");
1497 condlog(2, "read " DEFAULT_CONFIGFILE);
1499 if (load_config(DEFAULT_CONFIGFILE))
1502 if (init_checkers()) {
1503 condlog(0, "failed to initialize checkers");
1507 condlog(0, "failed to initialize prioritizers");
1511 setlogmask(LOG_UPTO(conf->verbosity + 3));
1514 * fill the voids left in the config file
1516 if (!conf->checkint) {
1517 conf->checkint = DEFAULT_CHECKINT;
1518 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1521 if (conf->max_fds) {
1522 struct rlimit fd_limit;
1524 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1525 condlog(0, "can't get open fds limit: %s\n",
1527 fd_limit.rlim_cur = 0;
1528 fd_limit.rlim_max = 0;
1530 if (fd_limit.rlim_cur < conf->max_fds) {
1531 fd_limit.rlim_cur = conf->max_fds;
1532 if (fd_limit.rlim_max < conf->max_fds)
1533 fd_limit.rlim_max = conf->max_fds;
1534 if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1535 condlog(0, "can't set open fds limit to "
1537 fd_limit.rlim_cur, fd_limit.rlim_max,
1540 condlog(3, "set open fds limit to %lu/%lu\n",
1541 fd_limit.rlim_cur, fd_limit.rlim_max);
1550 vecs = gvecs = init_vecs();
1555 if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1556 condlog(0, "can not find sysfs mount point");
1560 udev_set_sync_support(0);
1562 * Start uevent listener early to catch events
1564 if ((rc = pthread_create(&uevent_thr, &misc_attr, ueventloop, vecs))) {
1565 condlog(0, "failed to create uevent thread: %d", rc);
1568 if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
1569 condlog(0, "failed to create cli listener: %d", rc);
1573 * fetch and configure both paths and multipaths
1576 running_state = DAEMON_CONFIGURE;
1578 if (configure(vecs, 1)) {
1580 condlog(0, "failure during configuration");
1588 if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
1589 condlog(0,"failed to create checker loop thread: %d", rc);
1592 if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
1593 condlog(0, "failed to create uevent dispatcher: %d", rc);
1596 pthread_attr_destroy(&misc_attr);
1598 pthread_mutex_lock(&exit_mutex);
1599 /* Startup complete, create logfile */
1600 if (pidfile_create(DEFAULT_PIDFILE, daemon_pid))
1601 /* Ignore errors, we can live without */
1602 condlog(1, "failed to create pidfile");
1604 running_state = DAEMON_RUNNING;
1605 pthread_cond_wait(&exit_cond, &exit_mutex);
1610 running_state = DAEMON_SHUTDOWN;
1611 block_signal(SIGHUP, NULL);
1613 if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
1614 vector_foreach_slot(vecs->mpvec, mpp, i)
1615 dm_queue_if_no_path(mpp->alias, 0);
1616 remove_maps_and_stop_waiters(vecs);
1619 pthread_cancel(check_thr);
1620 pthread_cancel(uevent_thr);
1621 pthread_cancel(uxlsnr_thr);
1622 pthread_cancel(uevq_thr);
1627 free_pathvec(vecs->pathvec, FREE_PATHS);
1628 vecs->pathvec = NULL;
1630 /* Now all the waitevent threads will start rushing in. */
1631 while (vecs->lock.depth > 0) {
1632 sleep (1); /* This is weak. */
1633 condlog(3,"Have %d wait event checkers threads to de-alloc, waiting..\n", vecs->lock.depth);
1635 pthread_mutex_destroy(vecs->lock.mutex);
1636 FREE(vecs->lock.mutex);
1637 vecs->lock.depth = 0;
1638 vecs->lock.mutex = NULL;
1648 /* We're done here */
1649 condlog(3, "unlink pidfile");
1650 unlink(DEFAULT_PIDFILE);
1652 condlog(2, "--------shut down-------");
1658 * Freeing config must be done after condlog() and dm_lib_exit(),
1659 * because logging functions like dlog() and dm_write_log()
1660 * reference the config.
1666 dbg_free_final(NULL);
1678 if( (pid = fork()) < 0){
1679 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1687 if ( (pid = fork()) < 0)
1688 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1692 daemon_pid = getpid();
1693 in_fd = open("/dev/null", O_RDONLY);
1695 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1699 out_fd = open("/dev/console", O_WRONLY);
1701 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1706 close(STDIN_FILENO);
1708 close(STDOUT_FILENO);
1710 close(STDERR_FILENO);
1716 fprintf(stderr, "cannot chdir to '/', continuing\n");
1722 main (int argc, char *argv[])
1724 extern char *optarg;
1730 running_state = DAEMON_INIT;
1733 if (getuid() != 0) {
1734 fprintf(stderr, "need to be root\n");
1738 /* make sure we don't lock any path */
1740 umask(umask(077) | 022);
1742 conf = alloc_config();
1747 while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1751 //debug=1; /* ### comment me out ### */
1754 if (sizeof(optarg) > sizeof(char *) ||
1755 !isdigit(optarg[0]))
1758 conf->verbosity = atoi(optarg);
1767 if (optind < argc) {
1772 while (optind < argc) {
1773 if (strchr(argv[optind], ' '))
1774 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1776 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1779 c += snprintf(c, s + CMDSIZE - c, "\n");
1797 return (child(NULL));