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-list.h"
12 #include "dm-bio-record.h"
13 #include "dm-uevent.h"
15 #include <linux/ctype.h>
16 #include <linux/init.h>
17 #include <linux/mempool.h>
18 #include <linux/module.h>
19 #include <linux/pagemap.h>
20 #include <linux/slab.h>
21 #include <linux/time.h>
22 #include <linux/workqueue.h>
23 #include <scsi/scsi_dh.h>
24 #include <asm/atomic.h>
26 #define DM_MSG_PREFIX "multipath"
27 #define MESG_STR(x) x, sizeof(x)
31 struct list_head list
;
33 struct priority_group
*pg
; /* Owning PG */
34 unsigned is_active
; /* Path status */
35 unsigned fail_count
; /* Cumulative failure count */
38 struct work_struct deactivate_path
;
41 #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
44 * Paths are grouped into Priority Groups and numbered from 1 upwards.
45 * Each has a path selector which controls which path gets used.
47 struct priority_group
{
48 struct list_head list
;
50 struct multipath
*m
; /* Owning multipath instance */
51 struct path_selector ps
;
53 unsigned pg_num
; /* Reference number */
54 unsigned bypassed
; /* Temporarily bypass this PG? */
56 unsigned nr_pgpaths
; /* Number of paths in PG */
57 struct list_head pgpaths
;
60 /* Multipath context */
62 struct list_head list
;
67 const char *hw_handler_name
;
68 struct work_struct activate_path
;
69 struct pgpath
*pgpath_to_activate
;
70 unsigned nr_priority_groups
;
71 struct list_head priority_groups
;
72 unsigned pg_init_required
; /* pg_init needs calling? */
73 unsigned pg_init_in_progress
; /* Only one pg_init allowed at once */
75 unsigned nr_valid_paths
; /* Total number of usable paths */
76 struct pgpath
*current_pgpath
;
77 struct priority_group
*current_pg
;
78 struct priority_group
*next_pg
; /* Switch to this PG if set */
79 unsigned repeat_count
; /* I/Os left before calling PS again */
81 unsigned queue_io
; /* Must we queue all I/O? */
82 unsigned queue_if_no_path
; /* Queue I/O if last path fails? */
83 unsigned saved_queue_if_no_path
;/* Saved state during suspension */
84 unsigned pg_init_retries
; /* Number of times to retry pg_init */
85 unsigned pg_init_count
; /* Number of times pg_init called */
87 struct work_struct process_queued_ios
;
88 struct bio_list queued_ios
;
91 struct work_struct trigger_event
;
94 * We must use a mempool of dm_mpath_io structs so that we
95 * can resubmit bios on error.
101 * Context information attached to each bio we process.
104 struct pgpath
*pgpath
;
105 struct dm_bio_details details
;
108 typedef int (*action_fn
) (struct pgpath
*pgpath
);
110 #define MIN_IOS 256 /* Mempool size */
112 static struct kmem_cache
*_mpio_cache
;
114 static struct workqueue_struct
*kmultipathd
, *kmpath_handlerd
;
115 static void process_queued_ios(struct work_struct
*work
);
116 static void trigger_event(struct work_struct
*work
);
117 static void activate_path(struct work_struct
*work
);
118 static void deactivate_path(struct work_struct
*work
);
121 /*-----------------------------------------------
122 * Allocation routines
123 *-----------------------------------------------*/
125 static struct pgpath
*alloc_pgpath(void)
127 struct pgpath
*pgpath
= kzalloc(sizeof(*pgpath
), GFP_KERNEL
);
130 pgpath
->is_active
= 1;
131 INIT_WORK(&pgpath
->deactivate_path
, deactivate_path
);
137 static void free_pgpath(struct pgpath
*pgpath
)
142 static void deactivate_path(struct work_struct
*work
)
144 struct pgpath
*pgpath
=
145 container_of(work
, struct pgpath
, deactivate_path
);
147 blk_abort_queue(pgpath
->path
.dev
->bdev
->bd_disk
->queue
);
150 static struct priority_group
*alloc_priority_group(void)
152 struct priority_group
*pg
;
154 pg
= kzalloc(sizeof(*pg
), GFP_KERNEL
);
157 INIT_LIST_HEAD(&pg
->pgpaths
);
162 static void free_pgpaths(struct list_head
*pgpaths
, struct dm_target
*ti
)
165 struct pgpath
*pgpath
, *tmp
;
166 struct multipath
*m
= ti
->private;
168 list_for_each_entry_safe(pgpath
, tmp
, pgpaths
, list
) {
169 list_del(&pgpath
->list
);
170 if (m
->hw_handler_name
)
171 scsi_dh_detach(bdev_get_queue(pgpath
->path
.dev
->bdev
));
172 dm_put_device(ti
, pgpath
->path
.dev
);
173 spin_lock_irqsave(&m
->lock
, flags
);
174 if (m
->pgpath_to_activate
== pgpath
)
175 m
->pgpath_to_activate
= NULL
;
176 spin_unlock_irqrestore(&m
->lock
, flags
);
181 static void free_priority_group(struct priority_group
*pg
,
182 struct dm_target
*ti
)
184 struct path_selector
*ps
= &pg
->ps
;
187 ps
->type
->destroy(ps
);
188 dm_put_path_selector(ps
->type
);
191 free_pgpaths(&pg
->pgpaths
, ti
);
195 static struct multipath
*alloc_multipath(struct dm_target
*ti
)
199 m
= kzalloc(sizeof(*m
), GFP_KERNEL
);
201 INIT_LIST_HEAD(&m
->priority_groups
);
202 spin_lock_init(&m
->lock
);
204 INIT_WORK(&m
->process_queued_ios
, process_queued_ios
);
205 INIT_WORK(&m
->trigger_event
, trigger_event
);
206 INIT_WORK(&m
->activate_path
, activate_path
);
207 m
->mpio_pool
= mempool_create_slab_pool(MIN_IOS
, _mpio_cache
);
219 static void free_multipath(struct multipath
*m
)
221 struct priority_group
*pg
, *tmp
;
223 list_for_each_entry_safe(pg
, tmp
, &m
->priority_groups
, list
) {
225 free_priority_group(pg
, m
->ti
);
228 kfree(m
->hw_handler_name
);
229 mempool_destroy(m
->mpio_pool
);
234 /*-----------------------------------------------
236 *-----------------------------------------------*/
238 static void __switch_pg(struct multipath
*m
, struct pgpath
*pgpath
)
240 m
->current_pg
= pgpath
->pg
;
242 /* Must we initialise the PG first, and queue I/O till it's ready? */
243 if (m
->hw_handler_name
) {
244 m
->pg_init_required
= 1;
247 m
->pg_init_required
= 0;
251 m
->pg_init_count
= 0;
254 static int __choose_path_in_pg(struct multipath
*m
, struct priority_group
*pg
)
256 struct dm_path
*path
;
258 path
= pg
->ps
.type
->select_path(&pg
->ps
, &m
->repeat_count
);
262 m
->current_pgpath
= path_to_pgpath(path
);
264 if (m
->current_pg
!= pg
)
265 __switch_pg(m
, m
->current_pgpath
);
270 static void __choose_pgpath(struct multipath
*m
)
272 struct priority_group
*pg
;
273 unsigned bypassed
= 1;
275 if (!m
->nr_valid_paths
)
278 /* Were we instructed to switch PG? */
282 if (!__choose_path_in_pg(m
, pg
))
286 /* Don't change PG until it has no remaining paths */
287 if (m
->current_pg
&& !__choose_path_in_pg(m
, m
->current_pg
))
291 * Loop through priority groups until we find a valid path.
292 * First time we skip PGs marked 'bypassed'.
293 * Second time we only try the ones we skipped.
296 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
297 if (pg
->bypassed
== bypassed
)
299 if (!__choose_path_in_pg(m
, pg
))
302 } while (bypassed
--);
305 m
->current_pgpath
= NULL
;
306 m
->current_pg
= NULL
;
310 * Check whether bios must be queued in the device-mapper core rather
311 * than here in the target.
313 * m->lock must be held on entry.
315 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
316 * same value then we are not between multipath_presuspend()
317 * and multipath_resume() calls and we have no need to check
318 * for the DMF_NOFLUSH_SUSPENDING flag.
320 static int __must_push_back(struct multipath
*m
)
322 return (m
->queue_if_no_path
!= m
->saved_queue_if_no_path
&&
323 dm_noflush_suspending(m
->ti
));
326 static int map_io(struct multipath
*m
, struct bio
*bio
,
327 struct dm_mpath_io
*mpio
, unsigned was_queued
)
329 int r
= DM_MAPIO_REMAPPED
;
331 struct pgpath
*pgpath
;
333 spin_lock_irqsave(&m
->lock
, flags
);
335 /* Do we need to select a new pgpath? */
336 if (!m
->current_pgpath
||
337 (!m
->queue_io
&& (m
->repeat_count
&& --m
->repeat_count
== 0)))
340 pgpath
= m
->current_pgpath
;
345 if ((pgpath
&& m
->queue_io
) ||
346 (!pgpath
&& m
->queue_if_no_path
)) {
347 /* Queue for the daemon to resubmit */
348 bio_list_add(&m
->queued_ios
, bio
);
350 if ((m
->pg_init_required
&& !m
->pg_init_in_progress
) ||
352 queue_work(kmultipathd
, &m
->process_queued_ios
);
354 r
= DM_MAPIO_SUBMITTED
;
356 bio
->bi_bdev
= pgpath
->path
.dev
->bdev
;
357 else if (__must_push_back(m
))
358 r
= DM_MAPIO_REQUEUE
;
360 r
= -EIO
; /* Failed */
362 mpio
->pgpath
= pgpath
;
364 spin_unlock_irqrestore(&m
->lock
, flags
);
370 * If we run out of usable paths, should we queue I/O or error it?
372 static int queue_if_no_path(struct multipath
*m
, unsigned queue_if_no_path
,
373 unsigned save_old_value
)
377 spin_lock_irqsave(&m
->lock
, flags
);
380 m
->saved_queue_if_no_path
= m
->queue_if_no_path
;
382 m
->saved_queue_if_no_path
= queue_if_no_path
;
383 m
->queue_if_no_path
= queue_if_no_path
;
384 if (!m
->queue_if_no_path
&& m
->queue_size
)
385 queue_work(kmultipathd
, &m
->process_queued_ios
);
387 spin_unlock_irqrestore(&m
->lock
, flags
);
392 /*-----------------------------------------------------------------
393 * The multipath daemon is responsible for resubmitting queued ios.
394 *---------------------------------------------------------------*/
396 static void dispatch_queued_ios(struct multipath
*m
)
400 struct bio
*bio
= NULL
, *next
;
401 struct dm_mpath_io
*mpio
;
402 union map_info
*info
;
404 spin_lock_irqsave(&m
->lock
, flags
);
405 bio
= bio_list_get(&m
->queued_ios
);
406 spin_unlock_irqrestore(&m
->lock
, flags
);
412 info
= dm_get_mapinfo(bio
);
415 r
= map_io(m
, bio
, mpio
, 1);
418 else if (r
== DM_MAPIO_REMAPPED
)
419 generic_make_request(bio
);
420 else if (r
== DM_MAPIO_REQUEUE
)
421 bio_endio(bio
, -EIO
);
427 static void process_queued_ios(struct work_struct
*work
)
429 struct multipath
*m
=
430 container_of(work
, struct multipath
, process_queued_ios
);
431 struct pgpath
*pgpath
= NULL
;
432 unsigned init_required
= 0, must_queue
= 1;
435 spin_lock_irqsave(&m
->lock
, flags
);
440 if (!m
->current_pgpath
)
443 pgpath
= m
->current_pgpath
;
444 m
->pgpath_to_activate
= m
->current_pgpath
;
446 if ((pgpath
&& !m
->queue_io
) ||
447 (!pgpath
&& !m
->queue_if_no_path
))
450 if (m
->pg_init_required
&& !m
->pg_init_in_progress
) {
452 m
->pg_init_required
= 0;
453 m
->pg_init_in_progress
= 1;
458 spin_unlock_irqrestore(&m
->lock
, flags
);
461 queue_work(kmpath_handlerd
, &m
->activate_path
);
464 dispatch_queued_ios(m
);
468 * An event is triggered whenever a path is taken out of use.
469 * Includes path failure and PG bypass.
471 static void trigger_event(struct work_struct
*work
)
473 struct multipath
*m
=
474 container_of(work
, struct multipath
, trigger_event
);
476 dm_table_event(m
->ti
->table
);
479 /*-----------------------------------------------------------------
480 * Constructor/argument parsing:
481 * <#multipath feature args> [<arg>]*
482 * <#hw_handler args> [hw_handler [<arg>]*]
484 * <initial priority group>
485 * [<selector> <#selector args> [<arg>]*
486 * <#paths> <#per-path selector args>
487 * [<path> [<arg>]* ]+ ]+
488 *---------------------------------------------------------------*/
495 static int read_param(struct param
*param
, char *str
, unsigned *v
, char **error
)
498 (sscanf(str
, "%u", v
) != 1) ||
501 *error
= param
->error
;
513 static char *shift(struct arg_set
*as
)
527 static void consume(struct arg_set
*as
, unsigned n
)
529 BUG_ON (as
->argc
< n
);
534 static int parse_path_selector(struct arg_set
*as
, struct priority_group
*pg
,
535 struct dm_target
*ti
)
538 struct path_selector_type
*pst
;
541 static struct param _params
[] = {
542 {0, 1024, "invalid number of path selector args"},
545 pst
= dm_get_path_selector(shift(as
));
547 ti
->error
= "unknown path selector type";
551 r
= read_param(_params
, shift(as
), &ps_argc
, &ti
->error
);
553 dm_put_path_selector(pst
);
557 r
= pst
->create(&pg
->ps
, ps_argc
, as
->argv
);
559 dm_put_path_selector(pst
);
560 ti
->error
= "path selector constructor failed";
565 consume(as
, ps_argc
);
570 static struct pgpath
*parse_path(struct arg_set
*as
, struct path_selector
*ps
,
571 struct dm_target
*ti
)
575 struct multipath
*m
= ti
->private;
577 /* we need at least a path arg */
579 ti
->error
= "no device given";
580 return ERR_PTR(-EINVAL
);
585 return ERR_PTR(-ENOMEM
);
587 r
= dm_get_device(ti
, shift(as
), ti
->begin
, ti
->len
,
588 dm_table_get_mode(ti
->table
), &p
->path
.dev
);
590 ti
->error
= "error getting device";
594 if (m
->hw_handler_name
) {
595 r
= scsi_dh_attach(bdev_get_queue(p
->path
.dev
->bdev
),
598 dm_put_device(ti
, p
->path
.dev
);
603 r
= ps
->type
->add_path(ps
, &p
->path
, as
->argc
, as
->argv
, &ti
->error
);
605 dm_put_device(ti
, p
->path
.dev
);
616 static struct priority_group
*parse_priority_group(struct arg_set
*as
,
619 static struct param _params
[] = {
620 {1, 1024, "invalid number of paths"},
621 {0, 1024, "invalid number of selector args"}
625 unsigned i
, nr_selector_args
, nr_params
;
626 struct priority_group
*pg
;
627 struct dm_target
*ti
= m
->ti
;
631 ti
->error
= "not enough priority group arguments";
632 return ERR_PTR(-EINVAL
);
635 pg
= alloc_priority_group();
637 ti
->error
= "couldn't allocate priority group";
638 return ERR_PTR(-ENOMEM
);
642 r
= parse_path_selector(as
, pg
, ti
);
649 r
= read_param(_params
, shift(as
), &pg
->nr_pgpaths
, &ti
->error
);
653 r
= read_param(_params
+ 1, shift(as
), &nr_selector_args
, &ti
->error
);
657 nr_params
= 1 + nr_selector_args
;
658 for (i
= 0; i
< pg
->nr_pgpaths
; i
++) {
659 struct pgpath
*pgpath
;
660 struct arg_set path_args
;
662 if (as
->argc
< nr_params
) {
663 ti
->error
= "not enough path parameters";
667 path_args
.argc
= nr_params
;
668 path_args
.argv
= as
->argv
;
670 pgpath
= parse_path(&path_args
, &pg
->ps
, ti
);
671 if (IS_ERR(pgpath
)) {
677 list_add_tail(&pgpath
->list
, &pg
->pgpaths
);
678 consume(as
, nr_params
);
684 free_priority_group(pg
, ti
);
688 static int parse_hw_handler(struct arg_set
*as
, struct multipath
*m
)
691 struct dm_target
*ti
= m
->ti
;
693 static struct param _params
[] = {
694 {0, 1024, "invalid number of hardware handler args"},
697 if (read_param(_params
, shift(as
), &hw_argc
, &ti
->error
))
703 m
->hw_handler_name
= kstrdup(shift(as
), GFP_KERNEL
);
704 request_module("scsi_dh_%s", m
->hw_handler_name
);
705 if (scsi_dh_handler_exist(m
->hw_handler_name
) == 0) {
706 ti
->error
= "unknown hardware handler type";
707 kfree(m
->hw_handler_name
);
708 m
->hw_handler_name
= NULL
;
711 consume(as
, hw_argc
- 1);
716 static int parse_features(struct arg_set
*as
, struct multipath
*m
)
720 struct dm_target
*ti
= m
->ti
;
721 const char *param_name
;
723 static struct param _params
[] = {
724 {0, 3, "invalid number of feature args"},
725 {1, 50, "pg_init_retries must be between 1 and 50"},
728 r
= read_param(_params
, shift(as
), &argc
, &ti
->error
);
736 param_name
= shift(as
);
739 if (!strnicmp(param_name
, MESG_STR("queue_if_no_path"))) {
740 r
= queue_if_no_path(m
, 1, 0);
744 if (!strnicmp(param_name
, MESG_STR("pg_init_retries")) &&
746 r
= read_param(_params
+ 1, shift(as
),
747 &m
->pg_init_retries
, &ti
->error
);
752 ti
->error
= "Unrecognised multipath feature request";
754 } while (argc
&& !r
);
759 static int multipath_ctr(struct dm_target
*ti
, unsigned int argc
,
762 /* target parameters */
763 static struct param _params
[] = {
764 {1, 1024, "invalid number of priority groups"},
765 {1, 1024, "invalid initial priority group number"},
771 unsigned pg_count
= 0;
772 unsigned next_pg_num
;
777 m
= alloc_multipath(ti
);
779 ti
->error
= "can't allocate multipath";
783 r
= parse_features(&as
, m
);
787 r
= parse_hw_handler(&as
, m
);
791 r
= read_param(_params
, shift(&as
), &m
->nr_priority_groups
, &ti
->error
);
795 r
= read_param(_params
+ 1, shift(&as
), &next_pg_num
, &ti
->error
);
799 /* parse the priority groups */
801 struct priority_group
*pg
;
803 pg
= parse_priority_group(&as
, m
);
809 m
->nr_valid_paths
+= pg
->nr_pgpaths
;
810 list_add_tail(&pg
->list
, &m
->priority_groups
);
812 pg
->pg_num
= pg_count
;
817 if (pg_count
!= m
->nr_priority_groups
) {
818 ti
->error
= "priority group count mismatch";
830 static void multipath_dtr(struct dm_target
*ti
)
832 struct multipath
*m
= (struct multipath
*) ti
->private;
834 flush_workqueue(kmpath_handlerd
);
835 flush_workqueue(kmultipathd
);
840 * Map bios, recording original fields for later in case we have to resubmit
842 static int multipath_map(struct dm_target
*ti
, struct bio
*bio
,
843 union map_info
*map_context
)
846 struct dm_mpath_io
*mpio
;
847 struct multipath
*m
= (struct multipath
*) ti
->private;
849 mpio
= mempool_alloc(m
->mpio_pool
, GFP_NOIO
);
850 dm_bio_record(&mpio
->details
, bio
);
852 map_context
->ptr
= mpio
;
853 bio
->bi_rw
|= (1 << BIO_RW_FAILFAST_TRANSPORT
);
854 r
= map_io(m
, bio
, mpio
, 0);
855 if (r
< 0 || r
== DM_MAPIO_REQUEUE
)
856 mempool_free(mpio
, m
->mpio_pool
);
862 * Take a path out of use.
864 static int fail_path(struct pgpath
*pgpath
)
867 struct multipath
*m
= pgpath
->pg
->m
;
869 spin_lock_irqsave(&m
->lock
, flags
);
871 if (!pgpath
->is_active
)
874 DMWARN("Failing path %s.", pgpath
->path
.dev
->name
);
876 pgpath
->pg
->ps
.type
->fail_path(&pgpath
->pg
->ps
, &pgpath
->path
);
877 pgpath
->is_active
= 0;
878 pgpath
->fail_count
++;
882 if (pgpath
== m
->current_pgpath
)
883 m
->current_pgpath
= NULL
;
885 dm_path_uevent(DM_UEVENT_PATH_FAILED
, m
->ti
,
886 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
888 queue_work(kmultipathd
, &m
->trigger_event
);
889 queue_work(kmultipathd
, &pgpath
->deactivate_path
);
892 spin_unlock_irqrestore(&m
->lock
, flags
);
898 * Reinstate a previously-failed path
900 static int reinstate_path(struct pgpath
*pgpath
)
904 struct multipath
*m
= pgpath
->pg
->m
;
906 spin_lock_irqsave(&m
->lock
, flags
);
908 if (pgpath
->is_active
)
911 if (!pgpath
->pg
->ps
.type
->reinstate_path
) {
912 DMWARN("Reinstate path not supported by path selector %s",
913 pgpath
->pg
->ps
.type
->name
);
918 r
= pgpath
->pg
->ps
.type
->reinstate_path(&pgpath
->pg
->ps
, &pgpath
->path
);
922 pgpath
->is_active
= 1;
924 m
->current_pgpath
= NULL
;
925 if (!m
->nr_valid_paths
++ && m
->queue_size
)
926 queue_work(kmultipathd
, &m
->process_queued_ios
);
928 dm_path_uevent(DM_UEVENT_PATH_REINSTATED
, m
->ti
,
929 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
931 queue_work(kmultipathd
, &m
->trigger_event
);
934 spin_unlock_irqrestore(&m
->lock
, flags
);
940 * Fail or reinstate all paths that match the provided struct dm_dev.
942 static int action_dev(struct multipath
*m
, struct dm_dev
*dev
,
946 struct pgpath
*pgpath
;
947 struct priority_group
*pg
;
949 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
950 list_for_each_entry(pgpath
, &pg
->pgpaths
, list
) {
951 if (pgpath
->path
.dev
== dev
)
960 * Temporarily try to avoid having to use the specified PG
962 static void bypass_pg(struct multipath
*m
, struct priority_group
*pg
,
967 spin_lock_irqsave(&m
->lock
, flags
);
969 pg
->bypassed
= bypassed
;
970 m
->current_pgpath
= NULL
;
971 m
->current_pg
= NULL
;
973 spin_unlock_irqrestore(&m
->lock
, flags
);
975 queue_work(kmultipathd
, &m
->trigger_event
);
979 * Switch to using the specified PG from the next I/O that gets mapped
981 static int switch_pg_num(struct multipath
*m
, const char *pgstr
)
983 struct priority_group
*pg
;
987 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
988 (pgnum
> m
->nr_priority_groups
)) {
989 DMWARN("invalid PG number supplied to switch_pg_num");
993 spin_lock_irqsave(&m
->lock
, flags
);
994 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
999 m
->current_pgpath
= NULL
;
1000 m
->current_pg
= NULL
;
1003 spin_unlock_irqrestore(&m
->lock
, flags
);
1005 queue_work(kmultipathd
, &m
->trigger_event
);
1010 * Set/clear bypassed status of a PG.
1011 * PGs are numbered upwards from 1 in the order they were declared.
1013 static int bypass_pg_num(struct multipath
*m
, const char *pgstr
, int bypassed
)
1015 struct priority_group
*pg
;
1018 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
1019 (pgnum
> m
->nr_priority_groups
)) {
1020 DMWARN("invalid PG number supplied to bypass_pg");
1024 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1029 bypass_pg(m
, pg
, bypassed
);
1034 * Should we retry pg_init immediately?
1036 static int pg_init_limit_reached(struct multipath
*m
, struct pgpath
*pgpath
)
1038 unsigned long flags
;
1039 int limit_reached
= 0;
1041 spin_lock_irqsave(&m
->lock
, flags
);
1043 if (m
->pg_init_count
<= m
->pg_init_retries
)
1044 m
->pg_init_required
= 1;
1048 spin_unlock_irqrestore(&m
->lock
, flags
);
1050 return limit_reached
;
1053 static void pg_init_done(struct dm_path
*path
, int errors
)
1055 struct pgpath
*pgpath
= path_to_pgpath(path
);
1056 struct priority_group
*pg
= pgpath
->pg
;
1057 struct multipath
*m
= pg
->m
;
1058 unsigned long flags
;
1060 /* device or driver problems */
1065 if (!m
->hw_handler_name
) {
1069 DMERR("Cannot failover device because scsi_dh_%s was not "
1070 "loaded.", m
->hw_handler_name
);
1072 * Fail path for now, so we do not ping pong
1076 case SCSI_DH_DEV_TEMP_BUSY
:
1078 * Probably doing something like FW upgrade on the
1079 * controller so try the other pg.
1081 bypass_pg(m
, pg
, 1);
1083 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1085 case SCSI_DH_IMM_RETRY
:
1086 case SCSI_DH_RES_TEMP_UNAVAIL
:
1087 if (pg_init_limit_reached(m
, pgpath
))
1093 * We probably do not want to fail the path for a device
1094 * error, but this is what the old dm did. In future
1095 * patches we can do more advanced handling.
1100 spin_lock_irqsave(&m
->lock
, flags
);
1102 DMERR("Could not failover device. Error %d.", errors
);
1103 m
->current_pgpath
= NULL
;
1104 m
->current_pg
= NULL
;
1105 } else if (!m
->pg_init_required
) {
1110 m
->pg_init_in_progress
= 0;
1111 queue_work(kmultipathd
, &m
->process_queued_ios
);
1112 spin_unlock_irqrestore(&m
->lock
, flags
);
1115 static void activate_path(struct work_struct
*work
)
1118 struct multipath
*m
=
1119 container_of(work
, struct multipath
, activate_path
);
1120 struct dm_path
*path
;
1121 unsigned long flags
;
1123 spin_lock_irqsave(&m
->lock
, flags
);
1124 path
= &m
->pgpath_to_activate
->path
;
1125 m
->pgpath_to_activate
= NULL
;
1126 spin_unlock_irqrestore(&m
->lock
, flags
);
1129 ret
= scsi_dh_activate(bdev_get_queue(path
->dev
->bdev
));
1130 pg_init_done(path
, ret
);
1136 static int do_end_io(struct multipath
*m
, struct bio
*bio
,
1137 int error
, struct dm_mpath_io
*mpio
)
1139 unsigned long flags
;
1142 return 0; /* I/O complete */
1144 if ((error
== -EWOULDBLOCK
) && bio_rw_ahead(bio
))
1147 if (error
== -EOPNOTSUPP
)
1150 spin_lock_irqsave(&m
->lock
, flags
);
1151 if (!m
->nr_valid_paths
) {
1152 if (__must_push_back(m
)) {
1153 spin_unlock_irqrestore(&m
->lock
, flags
);
1154 return DM_ENDIO_REQUEUE
;
1155 } else if (!m
->queue_if_no_path
) {
1156 spin_unlock_irqrestore(&m
->lock
, flags
);
1159 spin_unlock_irqrestore(&m
->lock
, flags
);
1163 spin_unlock_irqrestore(&m
->lock
, flags
);
1166 fail_path(mpio
->pgpath
);
1169 dm_bio_restore(&mpio
->details
, bio
);
1171 /* queue for the daemon to resubmit or fail */
1172 spin_lock_irqsave(&m
->lock
, flags
);
1173 bio_list_add(&m
->queued_ios
, bio
);
1176 queue_work(kmultipathd
, &m
->process_queued_ios
);
1177 spin_unlock_irqrestore(&m
->lock
, flags
);
1179 return DM_ENDIO_INCOMPLETE
; /* io not complete */
1182 static int multipath_end_io(struct dm_target
*ti
, struct bio
*bio
,
1183 int error
, union map_info
*map_context
)
1185 struct multipath
*m
= ti
->private;
1186 struct dm_mpath_io
*mpio
= map_context
->ptr
;
1187 struct pgpath
*pgpath
= mpio
->pgpath
;
1188 struct path_selector
*ps
;
1191 r
= do_end_io(m
, bio
, error
, mpio
);
1193 ps
= &pgpath
->pg
->ps
;
1194 if (ps
->type
->end_io
)
1195 ps
->type
->end_io(ps
, &pgpath
->path
);
1197 if (r
!= DM_ENDIO_INCOMPLETE
)
1198 mempool_free(mpio
, m
->mpio_pool
);
1204 * Suspend can't complete until all the I/O is processed so if
1205 * the last path fails we must error any remaining I/O.
1206 * Note that if the freeze_bdev fails while suspending, the
1207 * queue_if_no_path state is lost - userspace should reset it.
1209 static void multipath_presuspend(struct dm_target
*ti
)
1211 struct multipath
*m
= (struct multipath
*) ti
->private;
1213 queue_if_no_path(m
, 0, 1);
1217 * Restore the queue_if_no_path setting.
1219 static void multipath_resume(struct dm_target
*ti
)
1221 struct multipath
*m
= (struct multipath
*) ti
->private;
1222 unsigned long flags
;
1224 spin_lock_irqsave(&m
->lock
, flags
);
1225 m
->queue_if_no_path
= m
->saved_queue_if_no_path
;
1226 spin_unlock_irqrestore(&m
->lock
, flags
);
1230 * Info output has the following format:
1231 * num_multipath_feature_args [multipath_feature_args]*
1232 * num_handler_status_args [handler_status_args]*
1233 * num_groups init_group_number
1234 * [A|D|E num_ps_status_args [ps_status_args]*
1235 * num_paths num_selector_args
1236 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1238 * Table output has the following format (identical to the constructor string):
1239 * num_feature_args [features_args]*
1240 * num_handler_args hw_handler [hw_handler_args]*
1241 * num_groups init_group_number
1242 * [priority selector-name num_ps_args [ps_args]*
1243 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1245 static int multipath_status(struct dm_target
*ti
, status_type_t type
,
1246 char *result
, unsigned int maxlen
)
1249 unsigned long flags
;
1250 struct multipath
*m
= (struct multipath
*) ti
->private;
1251 struct priority_group
*pg
;
1256 spin_lock_irqsave(&m
->lock
, flags
);
1259 if (type
== STATUSTYPE_INFO
)
1260 DMEMIT("2 %u %u ", m
->queue_size
, m
->pg_init_count
);
1262 DMEMIT("%u ", m
->queue_if_no_path
+
1263 (m
->pg_init_retries
> 0) * 2);
1264 if (m
->queue_if_no_path
)
1265 DMEMIT("queue_if_no_path ");
1266 if (m
->pg_init_retries
)
1267 DMEMIT("pg_init_retries %u ", m
->pg_init_retries
);
1270 if (!m
->hw_handler_name
|| type
== STATUSTYPE_INFO
)
1273 DMEMIT("1 %s ", m
->hw_handler_name
);
1275 DMEMIT("%u ", m
->nr_priority_groups
);
1278 pg_num
= m
->next_pg
->pg_num
;
1279 else if (m
->current_pg
)
1280 pg_num
= m
->current_pg
->pg_num
;
1284 DMEMIT("%u ", pg_num
);
1287 case STATUSTYPE_INFO
:
1288 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1290 state
= 'D'; /* Disabled */
1291 else if (pg
== m
->current_pg
)
1292 state
= 'A'; /* Currently Active */
1294 state
= 'E'; /* Enabled */
1296 DMEMIT("%c ", state
);
1298 if (pg
->ps
.type
->status
)
1299 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1305 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1306 pg
->ps
.type
->info_args
);
1308 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1309 DMEMIT("%s %s %u ", p
->path
.dev
->name
,
1310 p
->is_active
? "A" : "F",
1312 if (pg
->ps
.type
->status
)
1313 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1314 &p
->path
, type
, result
+ sz
,
1320 case STATUSTYPE_TABLE
:
1321 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1322 DMEMIT("%s ", pg
->ps
.type
->name
);
1324 if (pg
->ps
.type
->status
)
1325 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1331 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1332 pg
->ps
.type
->table_args
);
1334 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1335 DMEMIT("%s ", p
->path
.dev
->name
);
1336 if (pg
->ps
.type
->status
)
1337 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1338 &p
->path
, type
, result
+ sz
,
1345 spin_unlock_irqrestore(&m
->lock
, flags
);
1350 static int multipath_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
1354 struct multipath
*m
= (struct multipath
*) ti
->private;
1358 if (!strnicmp(argv
[0], MESG_STR("queue_if_no_path")))
1359 return queue_if_no_path(m
, 1, 0);
1360 else if (!strnicmp(argv
[0], MESG_STR("fail_if_no_path")))
1361 return queue_if_no_path(m
, 0, 0);
1367 if (!strnicmp(argv
[0], MESG_STR("disable_group")))
1368 return bypass_pg_num(m
, argv
[1], 1);
1369 else if (!strnicmp(argv
[0], MESG_STR("enable_group")))
1370 return bypass_pg_num(m
, argv
[1], 0);
1371 else if (!strnicmp(argv
[0], MESG_STR("switch_group")))
1372 return switch_pg_num(m
, argv
[1]);
1373 else if (!strnicmp(argv
[0], MESG_STR("reinstate_path")))
1374 action
= reinstate_path
;
1375 else if (!strnicmp(argv
[0], MESG_STR("fail_path")))
1380 r
= dm_get_device(ti
, argv
[1], ti
->begin
, ti
->len
,
1381 dm_table_get_mode(ti
->table
), &dev
);
1383 DMWARN("message: error getting device %s",
1388 r
= action_dev(m
, dev
, action
);
1390 dm_put_device(ti
, dev
);
1395 DMWARN("Unrecognised multipath message received.");
1399 static int multipath_ioctl(struct dm_target
*ti
, struct inode
*inode
,
1400 struct file
*filp
, unsigned int cmd
,
1403 struct multipath
*m
= (struct multipath
*) ti
->private;
1404 struct block_device
*bdev
= NULL
;
1405 unsigned long flags
;
1406 struct file fake_file
= {};
1407 struct dentry fake_dentry
= {};
1410 fake_file
.f_path
.dentry
= &fake_dentry
;
1412 spin_lock_irqsave(&m
->lock
, flags
);
1414 if (!m
->current_pgpath
)
1417 if (m
->current_pgpath
) {
1418 bdev
= m
->current_pgpath
->path
.dev
->bdev
;
1419 fake_dentry
.d_inode
= bdev
->bd_inode
;
1420 fake_file
.f_mode
= m
->current_pgpath
->path
.dev
->mode
;
1428 spin_unlock_irqrestore(&m
->lock
, flags
);
1430 return r
? : blkdev_driver_ioctl(bdev
->bd_inode
, &fake_file
,
1431 bdev
->bd_disk
, cmd
, arg
);
1434 /*-----------------------------------------------------------------
1436 *---------------------------------------------------------------*/
1437 static struct target_type multipath_target
= {
1438 .name
= "multipath",
1439 .version
= {1, 0, 5},
1440 .module
= THIS_MODULE
,
1441 .ctr
= multipath_ctr
,
1442 .dtr
= multipath_dtr
,
1443 .map
= multipath_map
,
1444 .end_io
= multipath_end_io
,
1445 .presuspend
= multipath_presuspend
,
1446 .resume
= multipath_resume
,
1447 .status
= multipath_status
,
1448 .message
= multipath_message
,
1449 .ioctl
= multipath_ioctl
,
1452 static int __init
dm_multipath_init(void)
1456 /* allocate a slab for the dm_ios */
1457 _mpio_cache
= KMEM_CACHE(dm_mpath_io
, 0);
1461 r
= dm_register_target(&multipath_target
);
1463 DMERR("register failed %d", r
);
1464 kmem_cache_destroy(_mpio_cache
);
1468 kmultipathd
= create_workqueue("kmpathd");
1470 DMERR("failed to create workqueue kmpathd");
1471 dm_unregister_target(&multipath_target
);
1472 kmem_cache_destroy(_mpio_cache
);
1477 * A separate workqueue is used to handle the device handlers
1478 * to avoid overloading existing workqueue. Overloading the
1479 * old workqueue would also create a bottleneck in the
1480 * path of the storage hardware device activation.
1482 kmpath_handlerd
= create_singlethread_workqueue("kmpath_handlerd");
1483 if (!kmpath_handlerd
) {
1484 DMERR("failed to create workqueue kmpath_handlerd");
1485 destroy_workqueue(kmultipathd
);
1486 dm_unregister_target(&multipath_target
);
1487 kmem_cache_destroy(_mpio_cache
);
1491 DMINFO("version %u.%u.%u loaded",
1492 multipath_target
.version
[0], multipath_target
.version
[1],
1493 multipath_target
.version
[2]);
1498 static void __exit
dm_multipath_exit(void)
1502 destroy_workqueue(kmpath_handlerd
);
1503 destroy_workqueue(kmultipathd
);
1505 r
= dm_unregister_target(&multipath_target
);
1507 DMERR("target unregister failed %d", r
);
1508 kmem_cache_destroy(_mpio_cache
);
1511 module_init(dm_multipath_init
);
1512 module_exit(dm_multipath_exit
);
1514 MODULE_DESCRIPTION(DM_NAME
" multipath target");
1515 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1516 MODULE_LICENSE("GPL");