2 * Copyright (C) 2003 Sistina Software Limited.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include <linux/device-mapper.h>
10 #include "dm-path-selector.h"
11 #include "dm-bio-record.h"
12 #include "dm-uevent.h"
14 #include <linux/ctype.h>
15 #include <linux/init.h>
16 #include <linux/mempool.h>
17 #include <linux/module.h>
18 #include <linux/pagemap.h>
19 #include <linux/slab.h>
20 #include <linux/time.h>
21 #include <linux/workqueue.h>
22 #include <scsi/scsi_dh.h>
23 #include <asm/atomic.h>
25 #define DM_MSG_PREFIX "multipath"
26 #define MESG_STR(x) x, sizeof(x)
30 struct list_head list;
32 struct priority_group *pg; /* Owning PG */
33 unsigned is_active; /* Path status */
34 unsigned fail_count; /* Cumulative failure count */
37 struct work_struct deactivate_path;
40 #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
43 * Paths are grouped into Priority Groups and numbered from 1 upwards.
44 * Each has a path selector which controls which path gets used.
46 struct priority_group {
47 struct list_head list;
49 struct multipath *m; /* Owning multipath instance */
50 struct path_selector ps;
52 unsigned pg_num; /* Reference number */
53 unsigned bypassed; /* Temporarily bypass this PG? */
55 unsigned nr_pgpaths; /* Number of paths in PG */
56 struct list_head pgpaths;
59 /* Multipath context */
61 struct list_head list;
66 const char *hw_handler_name;
67 struct work_struct activate_path;
68 struct pgpath *pgpath_to_activate;
69 unsigned nr_priority_groups;
70 struct list_head priority_groups;
71 unsigned pg_init_required; /* pg_init needs calling? */
72 unsigned pg_init_in_progress; /* Only one pg_init allowed at once */
74 unsigned nr_valid_paths; /* Total number of usable paths */
75 struct pgpath *current_pgpath;
76 struct priority_group *current_pg;
77 struct priority_group *next_pg; /* Switch to this PG if set */
78 unsigned repeat_count; /* I/Os left before calling PS again */
80 unsigned queue_io; /* Must we queue all I/O? */
81 unsigned queue_if_no_path; /* Queue I/O if last path fails? */
82 unsigned saved_queue_if_no_path;/* Saved state during suspension */
83 unsigned pg_init_retries; /* Number of times to retry pg_init */
84 unsigned pg_init_count; /* Number of times pg_init called */
86 struct work_struct process_queued_ios;
87 struct bio_list queued_ios;
90 struct work_struct trigger_event;
93 * We must use a mempool of dm_mpath_io structs so that we
94 * can resubmit bios on error.
100 * Context information attached to each bio we process.
103 struct pgpath *pgpath;
104 struct dm_bio_details details;
107 typedef int (*action_fn) (struct pgpath *pgpath);
109 #define MIN_IOS 256 /* Mempool size */
111 static struct kmem_cache *_mpio_cache;
113 static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
114 static void process_queued_ios(struct work_struct *work);
115 static void trigger_event(struct work_struct *work);
116 static void activate_path(struct work_struct *work);
117 static void deactivate_path(struct work_struct *work);
120 /*-----------------------------------------------
121 * Allocation routines
122 *-----------------------------------------------*/
124 static struct pgpath *alloc_pgpath(void)
126 struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
129 pgpath->is_active = 1;
130 INIT_WORK(&pgpath->deactivate_path, deactivate_path);
136 static void free_pgpath(struct pgpath *pgpath)
141 static void deactivate_path(struct work_struct *work)
143 struct pgpath *pgpath =
144 container_of(work, struct pgpath, deactivate_path);
146 blk_abort_queue(pgpath->path.dev->bdev->bd_disk->queue);
149 static struct priority_group *alloc_priority_group(void)
151 struct priority_group *pg;
153 pg = kzalloc(sizeof(*pg), GFP_KERNEL);
156 INIT_LIST_HEAD(&pg->pgpaths);
161 static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
164 struct pgpath *pgpath, *tmp;
165 struct multipath *m = ti->private;
167 list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
168 list_del(&pgpath->list);
169 if (m->hw_handler_name)
170 scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev));
171 dm_put_device(ti, pgpath->path.dev);
172 spin_lock_irqsave(&m->lock, flags);
173 if (m->pgpath_to_activate == pgpath)
174 m->pgpath_to_activate = NULL;
175 spin_unlock_irqrestore(&m->lock, flags);
180 static void free_priority_group(struct priority_group *pg,
181 struct dm_target *ti)
183 struct path_selector *ps = &pg->ps;
186 ps->type->destroy(ps);
187 dm_put_path_selector(ps->type);
190 free_pgpaths(&pg->pgpaths, ti);
194 static struct multipath *alloc_multipath(struct dm_target *ti)
198 m = kzalloc(sizeof(*m), GFP_KERNEL);
200 INIT_LIST_HEAD(&m->priority_groups);
201 spin_lock_init(&m->lock);
203 INIT_WORK(&m->process_queued_ios, process_queued_ios);
204 INIT_WORK(&m->trigger_event, trigger_event);
205 INIT_WORK(&m->activate_path, activate_path);
206 m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
218 static void free_multipath(struct multipath *m)
220 struct priority_group *pg, *tmp;
222 list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
224 free_priority_group(pg, m->ti);
227 kfree(m->hw_handler_name);
228 mempool_destroy(m->mpio_pool);
233 /*-----------------------------------------------
235 *-----------------------------------------------*/
237 static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
239 m->current_pg = pgpath->pg;
241 /* Must we initialise the PG first, and queue I/O till it's ready? */
242 if (m->hw_handler_name) {
243 m->pg_init_required = 1;
246 m->pg_init_required = 0;
250 m->pg_init_count = 0;
253 static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg)
255 struct dm_path *path;
257 path = pg->ps.type->select_path(&pg->ps, &m->repeat_count);
261 m->current_pgpath = path_to_pgpath(path);
263 if (m->current_pg != pg)
264 __switch_pg(m, m->current_pgpath);
269 static void __choose_pgpath(struct multipath *m)
271 struct priority_group *pg;
272 unsigned bypassed = 1;
274 if (!m->nr_valid_paths)
277 /* Were we instructed to switch PG? */
281 if (!__choose_path_in_pg(m, pg))
285 /* Don't change PG until it has no remaining paths */
286 if (m->current_pg && !__choose_path_in_pg(m, m->current_pg))
290 * Loop through priority groups until we find a valid path.
291 * First time we skip PGs marked 'bypassed'.
292 * Second time we only try the ones we skipped.
295 list_for_each_entry(pg, &m->priority_groups, list) {
296 if (pg->bypassed == bypassed)
298 if (!__choose_path_in_pg(m, pg))
301 } while (bypassed--);
304 m->current_pgpath = NULL;
305 m->current_pg = NULL;
309 * Check whether bios must be queued in the device-mapper core rather
310 * than here in the target.
312 * m->lock must be held on entry.
314 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
315 * same value then we are not between multipath_presuspend()
316 * and multipath_resume() calls and we have no need to check
317 * for the DMF_NOFLUSH_SUSPENDING flag.
319 static int __must_push_back(struct multipath *m)
321 return (m->queue_if_no_path != m->saved_queue_if_no_path &&
322 dm_noflush_suspending(m->ti));
325 static int map_io(struct multipath *m, struct bio *bio,
326 struct dm_mpath_io *mpio, unsigned was_queued)
328 int r = DM_MAPIO_REMAPPED;
330 struct pgpath *pgpath;
332 spin_lock_irqsave(&m->lock, flags);
334 /* Do we need to select a new pgpath? */
335 if (!m->current_pgpath ||
336 (!m->queue_io && (m->repeat_count && --m->repeat_count == 0)))
339 pgpath = m->current_pgpath;
344 if ((pgpath && m->queue_io) ||
345 (!pgpath && m->queue_if_no_path)) {
346 /* Queue for the daemon to resubmit */
347 bio_list_add(&m->queued_ios, bio);
349 if ((m->pg_init_required && !m->pg_init_in_progress) ||
351 queue_work(kmultipathd, &m->process_queued_ios);
353 r = DM_MAPIO_SUBMITTED;
355 bio->bi_bdev = pgpath->path.dev->bdev;
356 else if (__must_push_back(m))
357 r = DM_MAPIO_REQUEUE;
359 r = -EIO; /* Failed */
361 mpio->pgpath = pgpath;
363 spin_unlock_irqrestore(&m->lock, flags);
369 * If we run out of usable paths, should we queue I/O or error it?
371 static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
372 unsigned save_old_value)
376 spin_lock_irqsave(&m->lock, flags);
379 m->saved_queue_if_no_path = m->queue_if_no_path;
381 m->saved_queue_if_no_path = queue_if_no_path;
382 m->queue_if_no_path = queue_if_no_path;
383 if (!m->queue_if_no_path && m->queue_size)
384 queue_work(kmultipathd, &m->process_queued_ios);
386 spin_unlock_irqrestore(&m->lock, flags);
391 /*-----------------------------------------------------------------
392 * The multipath daemon is responsible for resubmitting queued ios.
393 *---------------------------------------------------------------*/
395 static void dispatch_queued_ios(struct multipath *m)
399 struct bio *bio = NULL, *next;
400 struct dm_mpath_io *mpio;
401 union map_info *info;
403 spin_lock_irqsave(&m->lock, flags);
404 bio = bio_list_get(&m->queued_ios);
405 spin_unlock_irqrestore(&m->lock, flags);
411 info = dm_get_mapinfo(bio);
414 r = map_io(m, bio, mpio, 1);
417 else if (r == DM_MAPIO_REMAPPED)
418 generic_make_request(bio);
419 else if (r == DM_MAPIO_REQUEUE)
420 bio_endio(bio, -EIO);
426 static void process_queued_ios(struct work_struct *work)
428 struct multipath *m =
429 container_of(work, struct multipath, process_queued_ios);
430 struct pgpath *pgpath = NULL;
431 unsigned init_required = 0, must_queue = 1;
434 spin_lock_irqsave(&m->lock, flags);
439 if (!m->current_pgpath)
442 pgpath = m->current_pgpath;
444 if ((pgpath && !m->queue_io) ||
445 (!pgpath && !m->queue_if_no_path))
448 if (m->pg_init_required && !m->pg_init_in_progress && pgpath) {
449 m->pgpath_to_activate = pgpath;
451 m->pg_init_required = 0;
452 m->pg_init_in_progress = 1;
457 spin_unlock_irqrestore(&m->lock, flags);
460 queue_work(kmpath_handlerd, &m->activate_path);
463 dispatch_queued_ios(m);
467 * An event is triggered whenever a path is taken out of use.
468 * Includes path failure and PG bypass.
470 static void trigger_event(struct work_struct *work)
472 struct multipath *m =
473 container_of(work, struct multipath, trigger_event);
475 dm_table_event(m->ti->table);
478 /*-----------------------------------------------------------------
479 * Constructor/argument parsing:
480 * <#multipath feature args> [<arg>]*
481 * <#hw_handler args> [hw_handler [<arg>]*]
483 * <initial priority group>
484 * [<selector> <#selector args> [<arg>]*
485 * <#paths> <#per-path selector args>
486 * [<path> [<arg>]* ]+ ]+
487 *---------------------------------------------------------------*/
494 static int read_param(struct param *param, char *str, unsigned *v, char **error)
497 (sscanf(str, "%u", v) != 1) ||
500 *error = param->error;
512 static char *shift(struct arg_set *as)
526 static void consume(struct arg_set *as, unsigned n)
528 BUG_ON (as->argc < n);
533 static int parse_path_selector(struct arg_set *as, struct priority_group *pg,
534 struct dm_target *ti)
537 struct path_selector_type *pst;
540 static struct param _params[] = {
541 {0, 1024, "invalid number of path selector args"},
544 pst = dm_get_path_selector(shift(as));
546 ti->error = "unknown path selector type";
550 r = read_param(_params, shift(as), &ps_argc, &ti->error);
552 dm_put_path_selector(pst);
556 r = pst->create(&pg->ps, ps_argc, as->argv);
558 dm_put_path_selector(pst);
559 ti->error = "path selector constructor failed";
564 consume(as, ps_argc);
569 static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps,
570 struct dm_target *ti)
574 struct multipath *m = ti->private;
576 /* we need at least a path arg */
578 ti->error = "no device given";
579 return ERR_PTR(-EINVAL);
584 return ERR_PTR(-ENOMEM);
586 r = dm_get_device(ti, shift(as), ti->begin, ti->len,
587 dm_table_get_mode(ti->table), &p->path.dev);
589 ti->error = "error getting device";
593 if (m->hw_handler_name) {
594 r = scsi_dh_attach(bdev_get_queue(p->path.dev->bdev),
597 dm_put_device(ti, p->path.dev);
602 r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
604 dm_put_device(ti, p->path.dev);
615 static struct priority_group *parse_priority_group(struct arg_set *as,
618 static struct param _params[] = {
619 {1, 1024, "invalid number of paths"},
620 {0, 1024, "invalid number of selector args"}
624 unsigned i, nr_selector_args, nr_params;
625 struct priority_group *pg;
626 struct dm_target *ti = m->ti;
630 ti->error = "not enough priority group arguments";
631 return ERR_PTR(-EINVAL);
634 pg = alloc_priority_group();
636 ti->error = "couldn't allocate priority group";
637 return ERR_PTR(-ENOMEM);
641 r = parse_path_selector(as, pg, ti);
648 r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error);
652 r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error);
656 nr_params = 1 + nr_selector_args;
657 for (i = 0; i < pg->nr_pgpaths; i++) {
658 struct pgpath *pgpath;
659 struct arg_set path_args;
661 if (as->argc < nr_params) {
662 ti->error = "not enough path parameters";
666 path_args.argc = nr_params;
667 path_args.argv = as->argv;
669 pgpath = parse_path(&path_args, &pg->ps, ti);
670 if (IS_ERR(pgpath)) {
676 list_add_tail(&pgpath->list, &pg->pgpaths);
677 consume(as, nr_params);
683 free_priority_group(pg, ti);
687 static int parse_hw_handler(struct arg_set *as, struct multipath *m)
690 struct dm_target *ti = m->ti;
692 static struct param _params[] = {
693 {0, 1024, "invalid number of hardware handler args"},
696 if (read_param(_params, shift(as), &hw_argc, &ti->error))
702 m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL);
703 request_module("scsi_dh_%s", m->hw_handler_name);
704 if (scsi_dh_handler_exist(m->hw_handler_name) == 0) {
705 ti->error = "unknown hardware handler type";
706 kfree(m->hw_handler_name);
707 m->hw_handler_name = NULL;
712 DMWARN("Ignoring user-specified arguments for "
713 "hardware handler \"%s\"", m->hw_handler_name);
714 consume(as, hw_argc - 1);
719 static int parse_features(struct arg_set *as, struct multipath *m)
723 struct dm_target *ti = m->ti;
724 const char *param_name;
726 static struct param _params[] = {
727 {0, 3, "invalid number of feature args"},
728 {1, 50, "pg_init_retries must be between 1 and 50"},
731 r = read_param(_params, shift(as), &argc, &ti->error);
739 param_name = shift(as);
742 if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) {
743 r = queue_if_no_path(m, 1, 0);
747 if (!strnicmp(param_name, MESG_STR("pg_init_retries")) &&
749 r = read_param(_params + 1, shift(as),
750 &m->pg_init_retries, &ti->error);
755 ti->error = "Unrecognised multipath feature request";
757 } while (argc && !r);
762 static int multipath_ctr(struct dm_target *ti, unsigned int argc,
765 /* target parameters */
766 static struct param _params[] = {
767 {1, 1024, "invalid number of priority groups"},
768 {1, 1024, "invalid initial priority group number"},
774 unsigned pg_count = 0;
775 unsigned next_pg_num;
780 m = alloc_multipath(ti);
782 ti->error = "can't allocate multipath";
786 r = parse_features(&as, m);
790 r = parse_hw_handler(&as, m);
794 r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error);
798 r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error);
802 /* parse the priority groups */
804 struct priority_group *pg;
806 pg = parse_priority_group(&as, m);
812 m->nr_valid_paths += pg->nr_pgpaths;
813 list_add_tail(&pg->list, &m->priority_groups);
815 pg->pg_num = pg_count;
820 if (pg_count != m->nr_priority_groups) {
821 ti->error = "priority group count mismatch";
833 static void multipath_dtr(struct dm_target *ti)
835 struct multipath *m = (struct multipath *) ti->private;
837 flush_workqueue(kmpath_handlerd);
838 flush_workqueue(kmultipathd);
843 * Map bios, recording original fields for later in case we have to resubmit
845 static int multipath_map(struct dm_target *ti, struct bio *bio,
846 union map_info *map_context)
849 struct dm_mpath_io *mpio;
850 struct multipath *m = (struct multipath *) ti->private;
852 mpio = mempool_alloc(m->mpio_pool, GFP_NOIO);
853 dm_bio_record(&mpio->details, bio);
855 map_context->ptr = mpio;
856 bio->bi_rw |= (1 << BIO_RW_FAILFAST_TRANSPORT);
857 r = map_io(m, bio, mpio, 0);
858 if (r < 0 || r == DM_MAPIO_REQUEUE)
859 mempool_free(mpio, m->mpio_pool);
865 * Take a path out of use.
867 static int fail_path(struct pgpath *pgpath)
870 struct multipath *m = pgpath->pg->m;
872 spin_lock_irqsave(&m->lock, flags);
874 if (!pgpath->is_active)
877 DMWARN("Failing path %s.", pgpath->path.dev->name);
879 pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
880 pgpath->is_active = 0;
881 pgpath->fail_count++;
885 if (pgpath == m->current_pgpath)
886 m->current_pgpath = NULL;
888 dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
889 pgpath->path.dev->name, m->nr_valid_paths);
891 schedule_work(&m->trigger_event);
892 queue_work(kmultipathd, &pgpath->deactivate_path);
895 spin_unlock_irqrestore(&m->lock, flags);
901 * Reinstate a previously-failed path
903 static int reinstate_path(struct pgpath *pgpath)
907 struct multipath *m = pgpath->pg->m;
909 spin_lock_irqsave(&m->lock, flags);
911 if (pgpath->is_active)
914 if (!pgpath->pg->ps.type->reinstate_path) {
915 DMWARN("Reinstate path not supported by path selector %s",
916 pgpath->pg->ps.type->name);
921 r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
925 pgpath->is_active = 1;
927 m->current_pgpath = NULL;
928 if (!m->nr_valid_paths++ && m->queue_size)
929 queue_work(kmultipathd, &m->process_queued_ios);
931 dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
932 pgpath->path.dev->name, m->nr_valid_paths);
934 schedule_work(&m->trigger_event);
937 spin_unlock_irqrestore(&m->lock, flags);
943 * Fail or reinstate all paths that match the provided struct dm_dev.
945 static int action_dev(struct multipath *m, struct dm_dev *dev,
949 struct pgpath *pgpath;
950 struct priority_group *pg;
952 list_for_each_entry(pg, &m->priority_groups, list) {
953 list_for_each_entry(pgpath, &pg->pgpaths, list) {
954 if (pgpath->path.dev == dev)
963 * Temporarily try to avoid having to use the specified PG
965 static void bypass_pg(struct multipath *m, struct priority_group *pg,
970 spin_lock_irqsave(&m->lock, flags);
972 pg->bypassed = bypassed;
973 m->current_pgpath = NULL;
974 m->current_pg = NULL;
976 spin_unlock_irqrestore(&m->lock, flags);
978 schedule_work(&m->trigger_event);
982 * Switch to using the specified PG from the next I/O that gets mapped
984 static int switch_pg_num(struct multipath *m, const char *pgstr)
986 struct priority_group *pg;
990 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
991 (pgnum > m->nr_priority_groups)) {
992 DMWARN("invalid PG number supplied to switch_pg_num");
996 spin_lock_irqsave(&m->lock, flags);
997 list_for_each_entry(pg, &m->priority_groups, list) {
1002 m->current_pgpath = NULL;
1003 m->current_pg = NULL;
1006 spin_unlock_irqrestore(&m->lock, flags);
1008 schedule_work(&m->trigger_event);
1013 * Set/clear bypassed status of a PG.
1014 * PGs are numbered upwards from 1 in the order they were declared.
1016 static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed)
1018 struct priority_group *pg;
1021 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
1022 (pgnum > m->nr_priority_groups)) {
1023 DMWARN("invalid PG number supplied to bypass_pg");
1027 list_for_each_entry(pg, &m->priority_groups, list) {
1032 bypass_pg(m, pg, bypassed);
1037 * Should we retry pg_init immediately?
1039 static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
1041 unsigned long flags;
1042 int limit_reached = 0;
1044 spin_lock_irqsave(&m->lock, flags);
1046 if (m->pg_init_count <= m->pg_init_retries)
1047 m->pg_init_required = 1;
1051 spin_unlock_irqrestore(&m->lock, flags);
1053 return limit_reached;
1056 static void pg_init_done(struct dm_path *path, int errors)
1058 struct pgpath *pgpath = path_to_pgpath(path);
1059 struct priority_group *pg = pgpath->pg;
1060 struct multipath *m = pg->m;
1061 unsigned long flags;
1063 /* device or driver problems */
1068 if (!m->hw_handler_name) {
1072 DMERR("Cannot failover device because scsi_dh_%s was not "
1073 "loaded.", m->hw_handler_name);
1075 * Fail path for now, so we do not ping pong
1079 case SCSI_DH_DEV_TEMP_BUSY:
1081 * Probably doing something like FW upgrade on the
1082 * controller so try the other pg.
1084 bypass_pg(m, pg, 1);
1086 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1088 case SCSI_DH_IMM_RETRY:
1089 case SCSI_DH_RES_TEMP_UNAVAIL:
1090 if (pg_init_limit_reached(m, pgpath))
1096 * We probably do not want to fail the path for a device
1097 * error, but this is what the old dm did. In future
1098 * patches we can do more advanced handling.
1103 spin_lock_irqsave(&m->lock, flags);
1105 DMERR("Could not failover device. Error %d.", errors);
1106 m->current_pgpath = NULL;
1107 m->current_pg = NULL;
1108 } else if (!m->pg_init_required) {
1113 m->pg_init_in_progress = 0;
1114 queue_work(kmultipathd, &m->process_queued_ios);
1115 spin_unlock_irqrestore(&m->lock, flags);
1118 static void activate_path(struct work_struct *work)
1121 struct multipath *m =
1122 container_of(work, struct multipath, activate_path);
1123 struct dm_path *path;
1124 unsigned long flags;
1126 spin_lock_irqsave(&m->lock, flags);
1127 path = &m->pgpath_to_activate->path;
1128 m->pgpath_to_activate = NULL;
1129 spin_unlock_irqrestore(&m->lock, flags);
1132 ret = scsi_dh_activate(bdev_get_queue(path->dev->bdev));
1133 pg_init_done(path, ret);
1139 static int do_end_io(struct multipath *m, struct bio *bio,
1140 int error, struct dm_mpath_io *mpio)
1142 unsigned long flags;
1145 return 0; /* I/O complete */
1147 if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
1150 if (error == -EOPNOTSUPP)
1153 spin_lock_irqsave(&m->lock, flags);
1154 if (!m->nr_valid_paths) {
1155 if (__must_push_back(m)) {
1156 spin_unlock_irqrestore(&m->lock, flags);
1157 return DM_ENDIO_REQUEUE;
1158 } else if (!m->queue_if_no_path) {
1159 spin_unlock_irqrestore(&m->lock, flags);
1162 spin_unlock_irqrestore(&m->lock, flags);
1166 spin_unlock_irqrestore(&m->lock, flags);
1169 fail_path(mpio->pgpath);
1172 dm_bio_restore(&mpio->details, bio);
1174 /* queue for the daemon to resubmit or fail */
1175 spin_lock_irqsave(&m->lock, flags);
1176 bio_list_add(&m->queued_ios, bio);
1179 queue_work(kmultipathd, &m->process_queued_ios);
1180 spin_unlock_irqrestore(&m->lock, flags);
1182 return DM_ENDIO_INCOMPLETE; /* io not complete */
1185 static int multipath_end_io(struct dm_target *ti, struct bio *bio,
1186 int error, union map_info *map_context)
1188 struct multipath *m = ti->private;
1189 struct dm_mpath_io *mpio = map_context->ptr;
1190 struct pgpath *pgpath = mpio->pgpath;
1191 struct path_selector *ps;
1194 r = do_end_io(m, bio, error, mpio);
1196 ps = &pgpath->pg->ps;
1197 if (ps->type->end_io)
1198 ps->type->end_io(ps, &pgpath->path);
1200 if (r != DM_ENDIO_INCOMPLETE)
1201 mempool_free(mpio, m->mpio_pool);
1207 * Suspend can't complete until all the I/O is processed so if
1208 * the last path fails we must error any remaining I/O.
1209 * Note that if the freeze_bdev fails while suspending, the
1210 * queue_if_no_path state is lost - userspace should reset it.
1212 static void multipath_presuspend(struct dm_target *ti)
1214 struct multipath *m = (struct multipath *) ti->private;
1216 queue_if_no_path(m, 0, 1);
1220 * Restore the queue_if_no_path setting.
1222 static void multipath_resume(struct dm_target *ti)
1224 struct multipath *m = (struct multipath *) ti->private;
1225 unsigned long flags;
1227 spin_lock_irqsave(&m->lock, flags);
1228 m->queue_if_no_path = m->saved_queue_if_no_path;
1229 spin_unlock_irqrestore(&m->lock, flags);
1233 * Info output has the following format:
1234 * num_multipath_feature_args [multipath_feature_args]*
1235 * num_handler_status_args [handler_status_args]*
1236 * num_groups init_group_number
1237 * [A|D|E num_ps_status_args [ps_status_args]*
1238 * num_paths num_selector_args
1239 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1241 * Table output has the following format (identical to the constructor string):
1242 * num_feature_args [features_args]*
1243 * num_handler_args hw_handler [hw_handler_args]*
1244 * num_groups init_group_number
1245 * [priority selector-name num_ps_args [ps_args]*
1246 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1248 static int multipath_status(struct dm_target *ti, status_type_t type,
1249 char *result, unsigned int maxlen)
1252 unsigned long flags;
1253 struct multipath *m = (struct multipath *) ti->private;
1254 struct priority_group *pg;
1259 spin_lock_irqsave(&m->lock, flags);
1262 if (type == STATUSTYPE_INFO)
1263 DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
1265 DMEMIT("%u ", m->queue_if_no_path +
1266 (m->pg_init_retries > 0) * 2);
1267 if (m->queue_if_no_path)
1268 DMEMIT("queue_if_no_path ");
1269 if (m->pg_init_retries)
1270 DMEMIT("pg_init_retries %u ", m->pg_init_retries);
1273 if (!m->hw_handler_name || type == STATUSTYPE_INFO)
1276 DMEMIT("1 %s ", m->hw_handler_name);
1278 DMEMIT("%u ", m->nr_priority_groups);
1281 pg_num = m->next_pg->pg_num;
1282 else if (m->current_pg)
1283 pg_num = m->current_pg->pg_num;
1287 DMEMIT("%u ", pg_num);
1290 case STATUSTYPE_INFO:
1291 list_for_each_entry(pg, &m->priority_groups, list) {
1293 state = 'D'; /* Disabled */
1294 else if (pg == m->current_pg)
1295 state = 'A'; /* Currently Active */
1297 state = 'E'; /* Enabled */
1299 DMEMIT("%c ", state);
1301 if (pg->ps.type->status)
1302 sz += pg->ps.type->status(&pg->ps, NULL, type,
1308 DMEMIT("%u %u ", pg->nr_pgpaths,
1309 pg->ps.type->info_args);
1311 list_for_each_entry(p, &pg->pgpaths, list) {
1312 DMEMIT("%s %s %u ", p->path.dev->name,
1313 p->is_active ? "A" : "F",
1315 if (pg->ps.type->status)
1316 sz += pg->ps.type->status(&pg->ps,
1317 &p->path, type, result + sz,
1323 case STATUSTYPE_TABLE:
1324 list_for_each_entry(pg, &m->priority_groups, list) {
1325 DMEMIT("%s ", pg->ps.type->name);
1327 if (pg->ps.type->status)
1328 sz += pg->ps.type->status(&pg->ps, NULL, type,
1334 DMEMIT("%u %u ", pg->nr_pgpaths,
1335 pg->ps.type->table_args);
1337 list_for_each_entry(p, &pg->pgpaths, list) {
1338 DMEMIT("%s ", p->path.dev->name);
1339 if (pg->ps.type->status)
1340 sz += pg->ps.type->status(&pg->ps,
1341 &p->path, type, result + sz,
1348 spin_unlock_irqrestore(&m->lock, flags);
1353 static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
1357 struct multipath *m = (struct multipath *) ti->private;
1361 if (!strnicmp(argv[0], MESG_STR("queue_if_no_path")))
1362 return queue_if_no_path(m, 1, 0);
1363 else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path")))
1364 return queue_if_no_path(m, 0, 0);
1370 if (!strnicmp(argv[0], MESG_STR("disable_group")))
1371 return bypass_pg_num(m, argv[1], 1);
1372 else if (!strnicmp(argv[0], MESG_STR("enable_group")))
1373 return bypass_pg_num(m, argv[1], 0);
1374 else if (!strnicmp(argv[0], MESG_STR("switch_group")))
1375 return switch_pg_num(m, argv[1]);
1376 else if (!strnicmp(argv[0], MESG_STR("reinstate_path")))
1377 action = reinstate_path;
1378 else if (!strnicmp(argv[0], MESG_STR("fail_path")))
1383 r = dm_get_device(ti, argv[1], ti->begin, ti->len,
1384 dm_table_get_mode(ti->table), &dev);
1386 DMWARN("message: error getting device %s",
1391 r = action_dev(m, dev, action);
1393 dm_put_device(ti, dev);
1398 DMWARN("Unrecognised multipath message received.");
1402 static int multipath_ioctl(struct dm_target *ti, unsigned int cmd,
1405 struct multipath *m = (struct multipath *) ti->private;
1406 struct block_device *bdev = NULL;
1408 unsigned long flags;
1411 spin_lock_irqsave(&m->lock, flags);
1413 if (!m->current_pgpath)
1416 if (m->current_pgpath) {
1417 bdev = m->current_pgpath->path.dev->bdev;
1418 mode = m->current_pgpath->path.dev->mode;
1426 spin_unlock_irqrestore(&m->lock, flags);
1428 return r ? : __blkdev_driver_ioctl(bdev, mode, cmd, arg);
1431 /*-----------------------------------------------------------------
1433 *---------------------------------------------------------------*/
1434 static struct target_type multipath_target = {
1435 .name = "multipath",
1436 .version = {1, 0, 5},
1437 .module = THIS_MODULE,
1438 .ctr = multipath_ctr,
1439 .dtr = multipath_dtr,
1440 .map = multipath_map,
1441 .end_io = multipath_end_io,
1442 .presuspend = multipath_presuspend,
1443 .resume = multipath_resume,
1444 .status = multipath_status,
1445 .message = multipath_message,
1446 .ioctl = multipath_ioctl,
1449 static int __init dm_multipath_init(void)
1453 /* allocate a slab for the dm_ios */
1454 _mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
1458 r = dm_register_target(&multipath_target);
1460 DMERR("register failed %d", r);
1461 kmem_cache_destroy(_mpio_cache);
1465 kmultipathd = create_workqueue("kmpathd");
1467 DMERR("failed to create workqueue kmpathd");
1468 dm_unregister_target(&multipath_target);
1469 kmem_cache_destroy(_mpio_cache);
1474 * A separate workqueue is used to handle the device handlers
1475 * to avoid overloading existing workqueue. Overloading the
1476 * old workqueue would also create a bottleneck in the
1477 * path of the storage hardware device activation.
1479 kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd");
1480 if (!kmpath_handlerd) {
1481 DMERR("failed to create workqueue kmpath_handlerd");
1482 destroy_workqueue(kmultipathd);
1483 dm_unregister_target(&multipath_target);
1484 kmem_cache_destroy(_mpio_cache);
1488 DMINFO("version %u.%u.%u loaded",
1489 multipath_target.version[0], multipath_target.version[1],
1490 multipath_target.version[2]);
1495 static void __exit dm_multipath_exit(void)
1497 destroy_workqueue(kmpath_handlerd);
1498 destroy_workqueue(kmultipathd);
1500 dm_unregister_target(&multipath_target);
1501 kmem_cache_destroy(_mpio_cache);
1504 module_init(dm_multipath_init);
1505 module_exit(dm_multipath_exit);
1507 MODULE_DESCRIPTION(DM_NAME " multipath target");
1508 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1509 MODULE_LICENSE("GPL");