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.
9 #include "dm-path-selector.h"
10 #include "dm-bio-list.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 fail_count; /* Cumulative failure count */
38 #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
41 * Paths are grouped into Priority Groups and numbered from 1 upwards.
42 * Each has a path selector which controls which path gets used.
44 struct priority_group {
45 struct list_head list;
47 struct multipath *m; /* Owning multipath instance */
48 struct path_selector ps;
50 unsigned pg_num; /* Reference number */
51 unsigned bypassed; /* Temporarily bypass this PG? */
53 unsigned nr_pgpaths; /* Number of paths in PG */
54 struct list_head pgpaths;
57 /* Multipath context */
59 struct list_head list;
64 const char *hw_handler_name;
65 struct work_struct activate_path;
66 unsigned nr_priority_groups;
67 struct list_head priority_groups;
68 unsigned pg_init_required; /* pg_init needs calling? */
69 unsigned pg_init_in_progress; /* Only one pg_init allowed at once */
71 unsigned nr_valid_paths; /* Total number of usable paths */
72 struct pgpath *current_pgpath;
73 struct priority_group *current_pg;
74 struct priority_group *next_pg; /* Switch to this PG if set */
75 unsigned repeat_count; /* I/Os left before calling PS again */
77 unsigned queue_io; /* Must we queue all I/O? */
78 unsigned queue_if_no_path; /* Queue I/O if last path fails? */
79 unsigned saved_queue_if_no_path;/* Saved state during suspension */
80 unsigned pg_init_retries; /* Number of times to retry pg_init */
81 unsigned pg_init_count; /* Number of times pg_init called */
83 struct work_struct process_queued_ios;
84 struct bio_list queued_ios;
87 struct work_struct trigger_event;
90 * We must use a mempool of dm_mpath_io structs so that we
91 * can resubmit bios on error.
97 * Context information attached to each bio we process.
100 struct pgpath *pgpath;
101 struct dm_bio_details details;
104 typedef int (*action_fn) (struct pgpath *pgpath);
106 #define MIN_IOS 256 /* Mempool size */
108 static struct kmem_cache *_mpio_cache;
110 static struct workqueue_struct *kmultipathd, *kmpath_handlerd;
111 static void process_queued_ios(struct work_struct *work);
112 static void trigger_event(struct work_struct *work);
113 static void activate_path(struct work_struct *work);
116 /*-----------------------------------------------
117 * Allocation routines
118 *-----------------------------------------------*/
120 static struct pgpath *alloc_pgpath(void)
122 struct pgpath *pgpath = kzalloc(sizeof(*pgpath), GFP_KERNEL);
125 pgpath->path.is_active = 1;
130 static void free_pgpath(struct pgpath *pgpath)
135 static struct priority_group *alloc_priority_group(void)
137 struct priority_group *pg;
139 pg = kzalloc(sizeof(*pg), GFP_KERNEL);
142 INIT_LIST_HEAD(&pg->pgpaths);
147 static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti)
149 struct pgpath *pgpath, *tmp;
150 struct multipath *m = ti->private;
152 list_for_each_entry_safe(pgpath, tmp, pgpaths, list) {
153 list_del(&pgpath->list);
154 if (m->hw_handler_name)
155 scsi_dh_detach(bdev_get_queue(pgpath->path.dev->bdev));
156 dm_put_device(ti, pgpath->path.dev);
161 static void free_priority_group(struct priority_group *pg,
162 struct dm_target *ti)
164 struct path_selector *ps = &pg->ps;
167 ps->type->destroy(ps);
168 dm_put_path_selector(ps->type);
171 free_pgpaths(&pg->pgpaths, ti);
175 static struct multipath *alloc_multipath(struct dm_target *ti)
179 m = kzalloc(sizeof(*m), GFP_KERNEL);
181 INIT_LIST_HEAD(&m->priority_groups);
182 spin_lock_init(&m->lock);
184 INIT_WORK(&m->process_queued_ios, process_queued_ios);
185 INIT_WORK(&m->trigger_event, trigger_event);
186 INIT_WORK(&m->activate_path, activate_path);
187 m->mpio_pool = mempool_create_slab_pool(MIN_IOS, _mpio_cache);
199 static void free_multipath(struct multipath *m)
201 struct priority_group *pg, *tmp;
203 list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) {
205 free_priority_group(pg, m->ti);
208 kfree(m->hw_handler_name);
209 mempool_destroy(m->mpio_pool);
214 /*-----------------------------------------------
216 *-----------------------------------------------*/
218 static void __switch_pg(struct multipath *m, struct pgpath *pgpath)
220 m->current_pg = pgpath->pg;
222 /* Must we initialise the PG first, and queue I/O till it's ready? */
223 if (m->hw_handler_name) {
224 m->pg_init_required = 1;
227 m->pg_init_required = 0;
231 m->pg_init_count = 0;
234 static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg)
236 struct dm_path *path;
238 path = pg->ps.type->select_path(&pg->ps, &m->repeat_count);
242 m->current_pgpath = path_to_pgpath(path);
244 if (m->current_pg != pg)
245 __switch_pg(m, m->current_pgpath);
250 static void __choose_pgpath(struct multipath *m)
252 struct priority_group *pg;
253 unsigned bypassed = 1;
255 if (!m->nr_valid_paths)
258 /* Were we instructed to switch PG? */
262 if (!__choose_path_in_pg(m, pg))
266 /* Don't change PG until it has no remaining paths */
267 if (m->current_pg && !__choose_path_in_pg(m, m->current_pg))
271 * Loop through priority groups until we find a valid path.
272 * First time we skip PGs marked 'bypassed'.
273 * Second time we only try the ones we skipped.
276 list_for_each_entry(pg, &m->priority_groups, list) {
277 if (pg->bypassed == bypassed)
279 if (!__choose_path_in_pg(m, pg))
282 } while (bypassed--);
285 m->current_pgpath = NULL;
286 m->current_pg = NULL;
290 * Check whether bios must be queued in the device-mapper core rather
291 * than here in the target.
293 * m->lock must be held on entry.
295 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
296 * same value then we are not between multipath_presuspend()
297 * and multipath_resume() calls and we have no need to check
298 * for the DMF_NOFLUSH_SUSPENDING flag.
300 static int __must_push_back(struct multipath *m)
302 return (m->queue_if_no_path != m->saved_queue_if_no_path &&
303 dm_noflush_suspending(m->ti));
306 static int map_io(struct multipath *m, struct bio *bio,
307 struct dm_mpath_io *mpio, unsigned was_queued)
309 int r = DM_MAPIO_REMAPPED;
311 struct pgpath *pgpath;
313 spin_lock_irqsave(&m->lock, flags);
315 /* Do we need to select a new pgpath? */
316 if (!m->current_pgpath ||
317 (!m->queue_io && (m->repeat_count && --m->repeat_count == 0)))
320 pgpath = m->current_pgpath;
325 if ((pgpath && m->queue_io) ||
326 (!pgpath && m->queue_if_no_path)) {
327 /* Queue for the daemon to resubmit */
328 bio_list_add(&m->queued_ios, bio);
330 if ((m->pg_init_required && !m->pg_init_in_progress) ||
332 queue_work(kmultipathd, &m->process_queued_ios);
334 r = DM_MAPIO_SUBMITTED;
336 bio->bi_bdev = pgpath->path.dev->bdev;
337 else if (__must_push_back(m))
338 r = DM_MAPIO_REQUEUE;
340 r = -EIO; /* Failed */
342 mpio->pgpath = pgpath;
344 spin_unlock_irqrestore(&m->lock, flags);
350 * If we run out of usable paths, should we queue I/O or error it?
352 static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path,
353 unsigned save_old_value)
357 spin_lock_irqsave(&m->lock, flags);
360 m->saved_queue_if_no_path = m->queue_if_no_path;
362 m->saved_queue_if_no_path = queue_if_no_path;
363 m->queue_if_no_path = queue_if_no_path;
364 if (!m->queue_if_no_path && m->queue_size)
365 queue_work(kmultipathd, &m->process_queued_ios);
367 spin_unlock_irqrestore(&m->lock, flags);
372 /*-----------------------------------------------------------------
373 * The multipath daemon is responsible for resubmitting queued ios.
374 *---------------------------------------------------------------*/
376 static void dispatch_queued_ios(struct multipath *m)
380 struct bio *bio = NULL, *next;
381 struct dm_mpath_io *mpio;
382 union map_info *info;
384 spin_lock_irqsave(&m->lock, flags);
385 bio = bio_list_get(&m->queued_ios);
386 spin_unlock_irqrestore(&m->lock, flags);
392 info = dm_get_mapinfo(bio);
395 r = map_io(m, bio, mpio, 1);
398 else if (r == DM_MAPIO_REMAPPED)
399 generic_make_request(bio);
400 else if (r == DM_MAPIO_REQUEUE)
401 bio_endio(bio, -EIO);
407 static void process_queued_ios(struct work_struct *work)
409 struct multipath *m =
410 container_of(work, struct multipath, process_queued_ios);
411 struct pgpath *pgpath = NULL;
412 unsigned init_required = 0, must_queue = 1;
415 spin_lock_irqsave(&m->lock, flags);
420 if (!m->current_pgpath)
423 pgpath = m->current_pgpath;
425 if ((pgpath && !m->queue_io) ||
426 (!pgpath && !m->queue_if_no_path))
429 if (m->pg_init_required && !m->pg_init_in_progress) {
431 m->pg_init_required = 0;
432 m->pg_init_in_progress = 1;
437 spin_unlock_irqrestore(&m->lock, flags);
440 queue_work(kmpath_handlerd, &m->activate_path);
443 dispatch_queued_ios(m);
447 * An event is triggered whenever a path is taken out of use.
448 * Includes path failure and PG bypass.
450 static void trigger_event(struct work_struct *work)
452 struct multipath *m =
453 container_of(work, struct multipath, trigger_event);
455 dm_table_event(m->ti->table);
458 /*-----------------------------------------------------------------
459 * Constructor/argument parsing:
460 * <#multipath feature args> [<arg>]*
461 * <#hw_handler args> [hw_handler [<arg>]*]
463 * <initial priority group>
464 * [<selector> <#selector args> [<arg>]*
465 * <#paths> <#per-path selector args>
466 * [<path> [<arg>]* ]+ ]+
467 *---------------------------------------------------------------*/
474 static int read_param(struct param *param, char *str, unsigned *v, char **error)
477 (sscanf(str, "%u", v) != 1) ||
480 *error = param->error;
492 static char *shift(struct arg_set *as)
506 static void consume(struct arg_set *as, unsigned n)
508 BUG_ON (as->argc < n);
513 static int parse_path_selector(struct arg_set *as, struct priority_group *pg,
514 struct dm_target *ti)
517 struct path_selector_type *pst;
520 static struct param _params[] = {
521 {0, 1024, "invalid number of path selector args"},
524 pst = dm_get_path_selector(shift(as));
526 ti->error = "unknown path selector type";
530 r = read_param(_params, shift(as), &ps_argc, &ti->error);
532 dm_put_path_selector(pst);
536 r = pst->create(&pg->ps, ps_argc, as->argv);
538 dm_put_path_selector(pst);
539 ti->error = "path selector constructor failed";
544 consume(as, ps_argc);
549 static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps,
550 struct dm_target *ti)
554 struct multipath *m = ti->private;
556 /* we need at least a path arg */
558 ti->error = "no device given";
566 r = dm_get_device(ti, shift(as), ti->begin, ti->len,
567 dm_table_get_mode(ti->table), &p->path.dev);
569 ti->error = "error getting device";
573 if (m->hw_handler_name) {
574 r = scsi_dh_attach(bdev_get_queue(p->path.dev->bdev),
577 dm_put_device(ti, p->path.dev);
582 r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error);
584 dm_put_device(ti, p->path.dev);
595 static struct priority_group *parse_priority_group(struct arg_set *as,
598 static struct param _params[] = {
599 {1, 1024, "invalid number of paths"},
600 {0, 1024, "invalid number of selector args"}
604 unsigned i, nr_selector_args, nr_params;
605 struct priority_group *pg;
606 struct dm_target *ti = m->ti;
610 ti->error = "not enough priority group aruments";
614 pg = alloc_priority_group();
616 ti->error = "couldn't allocate priority group";
621 r = parse_path_selector(as, pg, ti);
628 r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error);
632 r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error);
636 nr_params = 1 + nr_selector_args;
637 for (i = 0; i < pg->nr_pgpaths; i++) {
638 struct pgpath *pgpath;
639 struct arg_set path_args;
641 if (as->argc < nr_params) {
642 ti->error = "not enough path parameters";
646 path_args.argc = nr_params;
647 path_args.argv = as->argv;
649 pgpath = parse_path(&path_args, &pg->ps, ti);
654 list_add_tail(&pgpath->list, &pg->pgpaths);
655 consume(as, nr_params);
661 free_priority_group(pg, ti);
665 static int parse_hw_handler(struct arg_set *as, struct multipath *m)
668 struct dm_target *ti = m->ti;
670 static struct param _params[] = {
671 {0, 1024, "invalid number of hardware handler args"},
674 if (read_param(_params, shift(as), &hw_argc, &ti->error))
680 m->hw_handler_name = kstrdup(shift(as), GFP_KERNEL);
681 request_module("scsi_dh_%s", m->hw_handler_name);
682 if (scsi_dh_handler_exist(m->hw_handler_name) == 0) {
683 ti->error = "unknown hardware handler type";
684 kfree(m->hw_handler_name);
685 m->hw_handler_name = NULL;
688 consume(as, hw_argc - 1);
693 static int parse_features(struct arg_set *as, struct multipath *m)
697 struct dm_target *ti = m->ti;
698 const char *param_name;
700 static struct param _params[] = {
701 {0, 3, "invalid number of feature args"},
702 {1, 50, "pg_init_retries must be between 1 and 50"},
705 r = read_param(_params, shift(as), &argc, &ti->error);
713 param_name = shift(as);
716 if (!strnicmp(param_name, MESG_STR("queue_if_no_path"))) {
717 r = queue_if_no_path(m, 1, 0);
721 if (!strnicmp(param_name, MESG_STR("pg_init_retries")) &&
723 r = read_param(_params + 1, shift(as),
724 &m->pg_init_retries, &ti->error);
729 ti->error = "Unrecognised multipath feature request";
731 } while (argc && !r);
736 static int multipath_ctr(struct dm_target *ti, unsigned int argc,
739 /* target parameters */
740 static struct param _params[] = {
741 {1, 1024, "invalid number of priority groups"},
742 {1, 1024, "invalid initial priority group number"},
748 unsigned pg_count = 0;
749 unsigned next_pg_num;
754 m = alloc_multipath(ti);
756 ti->error = "can't allocate multipath";
760 r = parse_features(&as, m);
764 r = parse_hw_handler(&as, m);
768 r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error);
772 r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error);
776 /* parse the priority groups */
778 struct priority_group *pg;
780 pg = parse_priority_group(&as, m);
786 m->nr_valid_paths += pg->nr_pgpaths;
787 list_add_tail(&pg->list, &m->priority_groups);
789 pg->pg_num = pg_count;
794 if (pg_count != m->nr_priority_groups) {
795 ti->error = "priority group count mismatch";
807 static void multipath_dtr(struct dm_target *ti)
809 struct multipath *m = (struct multipath *) ti->private;
811 flush_workqueue(kmpath_handlerd);
812 flush_workqueue(kmultipathd);
817 * Map bios, recording original fields for later in case we have to resubmit
819 static int multipath_map(struct dm_target *ti, struct bio *bio,
820 union map_info *map_context)
823 struct dm_mpath_io *mpio;
824 struct multipath *m = (struct multipath *) ti->private;
826 mpio = mempool_alloc(m->mpio_pool, GFP_NOIO);
827 dm_bio_record(&mpio->details, bio);
829 map_context->ptr = mpio;
830 bio->bi_rw |= (1 << BIO_RW_FAILFAST);
831 r = map_io(m, bio, mpio, 0);
832 if (r < 0 || r == DM_MAPIO_REQUEUE)
833 mempool_free(mpio, m->mpio_pool);
839 * Take a path out of use.
841 static int fail_path(struct pgpath *pgpath)
844 struct multipath *m = pgpath->pg->m;
846 spin_lock_irqsave(&m->lock, flags);
848 if (!pgpath->path.is_active)
851 DMWARN("Failing path %s.", pgpath->path.dev->name);
853 pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path);
854 pgpath->path.is_active = 0;
855 pgpath->fail_count++;
859 if (pgpath == m->current_pgpath)
860 m->current_pgpath = NULL;
862 dm_path_uevent(DM_UEVENT_PATH_FAILED, m->ti,
863 pgpath->path.dev->name, m->nr_valid_paths);
865 queue_work(kmultipathd, &m->trigger_event);
868 spin_unlock_irqrestore(&m->lock, flags);
874 * Reinstate a previously-failed path
876 static int reinstate_path(struct pgpath *pgpath)
880 struct multipath *m = pgpath->pg->m;
882 spin_lock_irqsave(&m->lock, flags);
884 if (pgpath->path.is_active)
887 if (!pgpath->pg->ps.type->reinstate_path) {
888 DMWARN("Reinstate path not supported by path selector %s",
889 pgpath->pg->ps.type->name);
894 r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path);
898 pgpath->path.is_active = 1;
900 m->current_pgpath = NULL;
901 if (!m->nr_valid_paths++ && m->queue_size)
902 queue_work(kmultipathd, &m->process_queued_ios);
904 dm_path_uevent(DM_UEVENT_PATH_REINSTATED, m->ti,
905 pgpath->path.dev->name, m->nr_valid_paths);
907 queue_work(kmultipathd, &m->trigger_event);
910 spin_unlock_irqrestore(&m->lock, flags);
916 * Fail or reinstate all paths that match the provided struct dm_dev.
918 static int action_dev(struct multipath *m, struct dm_dev *dev,
922 struct pgpath *pgpath;
923 struct priority_group *pg;
925 list_for_each_entry(pg, &m->priority_groups, list) {
926 list_for_each_entry(pgpath, &pg->pgpaths, list) {
927 if (pgpath->path.dev == dev)
936 * Temporarily try to avoid having to use the specified PG
938 static void bypass_pg(struct multipath *m, struct priority_group *pg,
943 spin_lock_irqsave(&m->lock, flags);
945 pg->bypassed = bypassed;
946 m->current_pgpath = NULL;
947 m->current_pg = NULL;
949 spin_unlock_irqrestore(&m->lock, flags);
951 queue_work(kmultipathd, &m->trigger_event);
955 * Switch to using the specified PG from the next I/O that gets mapped
957 static int switch_pg_num(struct multipath *m, const char *pgstr)
959 struct priority_group *pg;
963 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
964 (pgnum > m->nr_priority_groups)) {
965 DMWARN("invalid PG number supplied to switch_pg_num");
969 spin_lock_irqsave(&m->lock, flags);
970 list_for_each_entry(pg, &m->priority_groups, list) {
975 m->current_pgpath = NULL;
976 m->current_pg = NULL;
979 spin_unlock_irqrestore(&m->lock, flags);
981 queue_work(kmultipathd, &m->trigger_event);
986 * Set/clear bypassed status of a PG.
987 * PGs are numbered upwards from 1 in the order they were declared.
989 static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed)
991 struct priority_group *pg;
994 if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum ||
995 (pgnum > m->nr_priority_groups)) {
996 DMWARN("invalid PG number supplied to bypass_pg");
1000 list_for_each_entry(pg, &m->priority_groups, list) {
1005 bypass_pg(m, pg, bypassed);
1010 * Should we retry pg_init immediately?
1012 static int pg_init_limit_reached(struct multipath *m, struct pgpath *pgpath)
1014 unsigned long flags;
1015 int limit_reached = 0;
1017 spin_lock_irqsave(&m->lock, flags);
1019 if (m->pg_init_count <= m->pg_init_retries)
1020 m->pg_init_required = 1;
1024 spin_unlock_irqrestore(&m->lock, flags);
1026 return limit_reached;
1029 static void pg_init_done(struct dm_path *path, int errors)
1031 struct pgpath *pgpath = path_to_pgpath(path);
1032 struct priority_group *pg = pgpath->pg;
1033 struct multipath *m = pg->m;
1034 unsigned long flags;
1036 /* device or driver problems */
1041 if (!m->hw_handler_name) {
1045 DMERR("Cannot failover device because scsi_dh_%s was not "
1046 "loaded.", m->hw_handler_name);
1048 * Fail path for now, so we do not ping pong
1052 case SCSI_DH_DEV_TEMP_BUSY:
1054 * Probably doing something like FW upgrade on the
1055 * controller so try the other pg.
1057 bypass_pg(m, pg, 1);
1059 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1061 case SCSI_DH_IMM_RETRY:
1062 case SCSI_DH_RES_TEMP_UNAVAIL:
1063 if (pg_init_limit_reached(m, pgpath))
1069 * We probably do not want to fail the path for a device
1070 * error, but this is what the old dm did. In future
1071 * patches we can do more advanced handling.
1076 spin_lock_irqsave(&m->lock, flags);
1078 DMERR("Could not failover device. Error %d.", errors);
1079 m->current_pgpath = NULL;
1080 m->current_pg = NULL;
1081 } else if (!m->pg_init_required) {
1086 m->pg_init_in_progress = 0;
1087 queue_work(kmultipathd, &m->process_queued_ios);
1088 spin_unlock_irqrestore(&m->lock, flags);
1091 static void activate_path(struct work_struct *work)
1094 struct multipath *m =
1095 container_of(work, struct multipath, activate_path);
1096 struct dm_path *path = &m->current_pgpath->path;
1098 ret = scsi_dh_activate(bdev_get_queue(path->dev->bdev));
1099 pg_init_done(path, ret);
1105 static int do_end_io(struct multipath *m, struct bio *bio,
1106 int error, struct dm_mpath_io *mpio)
1108 unsigned long flags;
1111 return 0; /* I/O complete */
1113 if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
1116 if (error == -EOPNOTSUPP)
1119 spin_lock_irqsave(&m->lock, flags);
1120 if (!m->nr_valid_paths) {
1121 if (__must_push_back(m)) {
1122 spin_unlock_irqrestore(&m->lock, flags);
1123 return DM_ENDIO_REQUEUE;
1124 } else if (!m->queue_if_no_path) {
1125 spin_unlock_irqrestore(&m->lock, flags);
1128 spin_unlock_irqrestore(&m->lock, flags);
1132 spin_unlock_irqrestore(&m->lock, flags);
1135 fail_path(mpio->pgpath);
1138 dm_bio_restore(&mpio->details, bio);
1140 /* queue for the daemon to resubmit or fail */
1141 spin_lock_irqsave(&m->lock, flags);
1142 bio_list_add(&m->queued_ios, bio);
1145 queue_work(kmultipathd, &m->process_queued_ios);
1146 spin_unlock_irqrestore(&m->lock, flags);
1148 return DM_ENDIO_INCOMPLETE; /* io not complete */
1151 static int multipath_end_io(struct dm_target *ti, struct bio *bio,
1152 int error, union map_info *map_context)
1154 struct multipath *m = ti->private;
1155 struct dm_mpath_io *mpio = map_context->ptr;
1156 struct pgpath *pgpath = mpio->pgpath;
1157 struct path_selector *ps;
1160 r = do_end_io(m, bio, error, mpio);
1162 ps = &pgpath->pg->ps;
1163 if (ps->type->end_io)
1164 ps->type->end_io(ps, &pgpath->path);
1166 if (r != DM_ENDIO_INCOMPLETE)
1167 mempool_free(mpio, m->mpio_pool);
1173 * Suspend can't complete until all the I/O is processed so if
1174 * the last path fails we must error any remaining I/O.
1175 * Note that if the freeze_bdev fails while suspending, the
1176 * queue_if_no_path state is lost - userspace should reset it.
1178 static void multipath_presuspend(struct dm_target *ti)
1180 struct multipath *m = (struct multipath *) ti->private;
1182 queue_if_no_path(m, 0, 1);
1186 * Restore the queue_if_no_path setting.
1188 static void multipath_resume(struct dm_target *ti)
1190 struct multipath *m = (struct multipath *) ti->private;
1191 unsigned long flags;
1193 spin_lock_irqsave(&m->lock, flags);
1194 m->queue_if_no_path = m->saved_queue_if_no_path;
1195 spin_unlock_irqrestore(&m->lock, flags);
1199 * Info output has the following format:
1200 * num_multipath_feature_args [multipath_feature_args]*
1201 * num_handler_status_args [handler_status_args]*
1202 * num_groups init_group_number
1203 * [A|D|E num_ps_status_args [ps_status_args]*
1204 * num_paths num_selector_args
1205 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1207 * Table output has the following format (identical to the constructor string):
1208 * num_feature_args [features_args]*
1209 * num_handler_args hw_handler [hw_handler_args]*
1210 * num_groups init_group_number
1211 * [priority selector-name num_ps_args [ps_args]*
1212 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1214 static int multipath_status(struct dm_target *ti, status_type_t type,
1215 char *result, unsigned int maxlen)
1218 unsigned long flags;
1219 struct multipath *m = (struct multipath *) ti->private;
1220 struct priority_group *pg;
1225 spin_lock_irqsave(&m->lock, flags);
1228 if (type == STATUSTYPE_INFO)
1229 DMEMIT("2 %u %u ", m->queue_size, m->pg_init_count);
1231 DMEMIT("%u ", m->queue_if_no_path +
1232 (m->pg_init_retries > 0) * 2);
1233 if (m->queue_if_no_path)
1234 DMEMIT("queue_if_no_path ");
1235 if (m->pg_init_retries)
1236 DMEMIT("pg_init_retries %u ", m->pg_init_retries);
1239 if (!m->hw_handler_name || type == STATUSTYPE_INFO)
1242 DMEMIT("1 %s ", m->hw_handler_name);
1244 DMEMIT("%u ", m->nr_priority_groups);
1247 pg_num = m->next_pg->pg_num;
1248 else if (m->current_pg)
1249 pg_num = m->current_pg->pg_num;
1253 DMEMIT("%u ", pg_num);
1256 case STATUSTYPE_INFO:
1257 list_for_each_entry(pg, &m->priority_groups, list) {
1259 state = 'D'; /* Disabled */
1260 else if (pg == m->current_pg)
1261 state = 'A'; /* Currently Active */
1263 state = 'E'; /* Enabled */
1265 DMEMIT("%c ", state);
1267 if (pg->ps.type->status)
1268 sz += pg->ps.type->status(&pg->ps, NULL, type,
1274 DMEMIT("%u %u ", pg->nr_pgpaths,
1275 pg->ps.type->info_args);
1277 list_for_each_entry(p, &pg->pgpaths, list) {
1278 DMEMIT("%s %s %u ", p->path.dev->name,
1279 p->path.is_active ? "A" : "F",
1281 if (pg->ps.type->status)
1282 sz += pg->ps.type->status(&pg->ps,
1283 &p->path, type, result + sz,
1289 case STATUSTYPE_TABLE:
1290 list_for_each_entry(pg, &m->priority_groups, list) {
1291 DMEMIT("%s ", pg->ps.type->name);
1293 if (pg->ps.type->status)
1294 sz += pg->ps.type->status(&pg->ps, NULL, type,
1300 DMEMIT("%u %u ", pg->nr_pgpaths,
1301 pg->ps.type->table_args);
1303 list_for_each_entry(p, &pg->pgpaths, list) {
1304 DMEMIT("%s ", p->path.dev->name);
1305 if (pg->ps.type->status)
1306 sz += pg->ps.type->status(&pg->ps,
1307 &p->path, type, result + sz,
1314 spin_unlock_irqrestore(&m->lock, flags);
1319 static int multipath_message(struct dm_target *ti, unsigned argc, char **argv)
1323 struct multipath *m = (struct multipath *) ti->private;
1327 if (!strnicmp(argv[0], MESG_STR("queue_if_no_path")))
1328 return queue_if_no_path(m, 1, 0);
1329 else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path")))
1330 return queue_if_no_path(m, 0, 0);
1336 if (!strnicmp(argv[0], MESG_STR("disable_group")))
1337 return bypass_pg_num(m, argv[1], 1);
1338 else if (!strnicmp(argv[0], MESG_STR("enable_group")))
1339 return bypass_pg_num(m, argv[1], 0);
1340 else if (!strnicmp(argv[0], MESG_STR("switch_group")))
1341 return switch_pg_num(m, argv[1]);
1342 else if (!strnicmp(argv[0], MESG_STR("reinstate_path")))
1343 action = reinstate_path;
1344 else if (!strnicmp(argv[0], MESG_STR("fail_path")))
1349 r = dm_get_device(ti, argv[1], ti->begin, ti->len,
1350 dm_table_get_mode(ti->table), &dev);
1352 DMWARN("message: error getting device %s",
1357 r = action_dev(m, dev, action);
1359 dm_put_device(ti, dev);
1364 DMWARN("Unrecognised multipath message received.");
1368 static int multipath_ioctl(struct dm_target *ti, struct inode *inode,
1369 struct file *filp, unsigned int cmd,
1372 struct multipath *m = (struct multipath *) ti->private;
1373 struct block_device *bdev = NULL;
1374 unsigned long flags;
1375 struct file fake_file = {};
1376 struct dentry fake_dentry = {};
1379 fake_file.f_path.dentry = &fake_dentry;
1381 spin_lock_irqsave(&m->lock, flags);
1383 if (!m->current_pgpath)
1386 if (m->current_pgpath) {
1387 bdev = m->current_pgpath->path.dev->bdev;
1388 fake_dentry.d_inode = bdev->bd_inode;
1389 fake_file.f_mode = m->current_pgpath->path.dev->mode;
1397 spin_unlock_irqrestore(&m->lock, flags);
1399 return r ? : blkdev_driver_ioctl(bdev->bd_inode, &fake_file,
1400 bdev->bd_disk, cmd, arg);
1403 /*-----------------------------------------------------------------
1405 *---------------------------------------------------------------*/
1406 static struct target_type multipath_target = {
1407 .name = "multipath",
1408 .version = {1, 0, 5},
1409 .module = THIS_MODULE,
1410 .ctr = multipath_ctr,
1411 .dtr = multipath_dtr,
1412 .map = multipath_map,
1413 .end_io = multipath_end_io,
1414 .presuspend = multipath_presuspend,
1415 .resume = multipath_resume,
1416 .status = multipath_status,
1417 .message = multipath_message,
1418 .ioctl = multipath_ioctl,
1421 static int __init dm_multipath_init(void)
1425 /* allocate a slab for the dm_ios */
1426 _mpio_cache = KMEM_CACHE(dm_mpath_io, 0);
1430 r = dm_register_target(&multipath_target);
1432 DMERR("register failed %d", r);
1433 kmem_cache_destroy(_mpio_cache);
1437 kmultipathd = create_workqueue("kmpathd");
1439 DMERR("failed to create workqueue kmpathd");
1440 dm_unregister_target(&multipath_target);
1441 kmem_cache_destroy(_mpio_cache);
1446 * A separate workqueue is used to handle the device handlers
1447 * to avoid overloading existing workqueue. Overloading the
1448 * old workqueue would also create a bottleneck in the
1449 * path of the storage hardware device activation.
1451 kmpath_handlerd = create_singlethread_workqueue("kmpath_handlerd");
1452 if (!kmpath_handlerd) {
1453 DMERR("failed to create workqueue kmpath_handlerd");
1454 destroy_workqueue(kmultipathd);
1455 dm_unregister_target(&multipath_target);
1456 kmem_cache_destroy(_mpio_cache);
1460 DMINFO("version %u.%u.%u loaded",
1461 multipath_target.version[0], multipath_target.version[1],
1462 multipath_target.version[2]);
1467 static void __exit dm_multipath_exit(void)
1471 destroy_workqueue(kmpath_handlerd);
1472 destroy_workqueue(kmultipathd);
1474 r = dm_unregister_target(&multipath_target);
1476 DMERR("target unregister failed %d", r);
1477 kmem_cache_destroy(_mpio_cache);
1480 module_init(dm_multipath_init);
1481 module_exit(dm_multipath_exit);
1483 MODULE_DESCRIPTION(DM_NAME " multipath target");
1484 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1485 MODULE_LICENSE("GPL");