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 if (ps_argc
> as
->argc
) {
557 dm_put_path_selector(pst
);
558 ti
->error
= "not enough arguments for path selector";
562 r
= pst
->create(&pg
->ps
, ps_argc
, as
->argv
);
564 dm_put_path_selector(pst
);
565 ti
->error
= "path selector constructor failed";
570 consume(as
, ps_argc
);
575 static struct pgpath
*parse_path(struct arg_set
*as
, struct path_selector
*ps
,
576 struct dm_target
*ti
)
580 struct multipath
*m
= ti
->private;
582 /* we need at least a path arg */
584 ti
->error
= "no device given";
585 return ERR_PTR(-EINVAL
);
590 return ERR_PTR(-ENOMEM
);
592 r
= dm_get_device(ti
, shift(as
), ti
->begin
, ti
->len
,
593 dm_table_get_mode(ti
->table
), &p
->path
.dev
);
595 ti
->error
= "error getting device";
599 if (m
->hw_handler_name
) {
600 r
= scsi_dh_attach(bdev_get_queue(p
->path
.dev
->bdev
),
603 dm_put_device(ti
, p
->path
.dev
);
608 r
= ps
->type
->add_path(ps
, &p
->path
, as
->argc
, as
->argv
, &ti
->error
);
610 dm_put_device(ti
, p
->path
.dev
);
621 static struct priority_group
*parse_priority_group(struct arg_set
*as
,
624 static struct param _params
[] = {
625 {1, 1024, "invalid number of paths"},
626 {0, 1024, "invalid number of selector args"}
630 unsigned i
, nr_selector_args
, nr_params
;
631 struct priority_group
*pg
;
632 struct dm_target
*ti
= m
->ti
;
636 ti
->error
= "not enough priority group arguments";
637 return ERR_PTR(-EINVAL
);
640 pg
= alloc_priority_group();
642 ti
->error
= "couldn't allocate priority group";
643 return ERR_PTR(-ENOMEM
);
647 r
= parse_path_selector(as
, pg
, ti
);
654 r
= read_param(_params
, shift(as
), &pg
->nr_pgpaths
, &ti
->error
);
658 r
= read_param(_params
+ 1, shift(as
), &nr_selector_args
, &ti
->error
);
662 nr_params
= 1 + nr_selector_args
;
663 for (i
= 0; i
< pg
->nr_pgpaths
; i
++) {
664 struct pgpath
*pgpath
;
665 struct arg_set path_args
;
667 if (as
->argc
< nr_params
) {
668 ti
->error
= "not enough path parameters";
672 path_args
.argc
= nr_params
;
673 path_args
.argv
= as
->argv
;
675 pgpath
= parse_path(&path_args
, &pg
->ps
, ti
);
676 if (IS_ERR(pgpath
)) {
682 list_add_tail(&pgpath
->list
, &pg
->pgpaths
);
683 consume(as
, nr_params
);
689 free_priority_group(pg
, ti
);
693 static int parse_hw_handler(struct arg_set
*as
, struct multipath
*m
)
696 struct dm_target
*ti
= m
->ti
;
698 static struct param _params
[] = {
699 {0, 1024, "invalid number of hardware handler args"},
702 if (read_param(_params
, shift(as
), &hw_argc
, &ti
->error
))
708 if (hw_argc
> as
->argc
) {
709 ti
->error
= "not enough arguments for hardware handler";
713 m
->hw_handler_name
= kstrdup(shift(as
), GFP_KERNEL
);
714 request_module("scsi_dh_%s", m
->hw_handler_name
);
715 if (scsi_dh_handler_exist(m
->hw_handler_name
) == 0) {
716 ti
->error
= "unknown hardware handler type";
717 kfree(m
->hw_handler_name
);
718 m
->hw_handler_name
= NULL
;
723 DMWARN("Ignoring user-specified arguments for "
724 "hardware handler \"%s\"", m
->hw_handler_name
);
725 consume(as
, hw_argc
- 1);
730 static int parse_features(struct arg_set
*as
, struct multipath
*m
)
734 struct dm_target
*ti
= m
->ti
;
735 const char *param_name
;
737 static struct param _params
[] = {
738 {0, 3, "invalid number of feature args"},
739 {1, 50, "pg_init_retries must be between 1 and 50"},
742 r
= read_param(_params
, shift(as
), &argc
, &ti
->error
);
750 param_name
= shift(as
);
753 if (!strnicmp(param_name
, MESG_STR("queue_if_no_path"))) {
754 r
= queue_if_no_path(m
, 1, 0);
758 if (!strnicmp(param_name
, MESG_STR("pg_init_retries")) &&
760 r
= read_param(_params
+ 1, shift(as
),
761 &m
->pg_init_retries
, &ti
->error
);
766 ti
->error
= "Unrecognised multipath feature request";
768 } while (argc
&& !r
);
773 static int multipath_ctr(struct dm_target
*ti
, unsigned int argc
,
776 /* target parameters */
777 static struct param _params
[] = {
778 {1, 1024, "invalid number of priority groups"},
779 {1, 1024, "invalid initial priority group number"},
785 unsigned pg_count
= 0;
786 unsigned next_pg_num
;
791 m
= alloc_multipath(ti
);
793 ti
->error
= "can't allocate multipath";
797 r
= parse_features(&as
, m
);
801 r
= parse_hw_handler(&as
, m
);
805 r
= read_param(_params
, shift(&as
), &m
->nr_priority_groups
, &ti
->error
);
809 r
= read_param(_params
+ 1, shift(&as
), &next_pg_num
, &ti
->error
);
813 /* parse the priority groups */
815 struct priority_group
*pg
;
817 pg
= parse_priority_group(&as
, m
);
823 m
->nr_valid_paths
+= pg
->nr_pgpaths
;
824 list_add_tail(&pg
->list
, &m
->priority_groups
);
826 pg
->pg_num
= pg_count
;
831 if (pg_count
!= m
->nr_priority_groups
) {
832 ti
->error
= "priority group count mismatch";
844 static void multipath_dtr(struct dm_target
*ti
)
846 struct multipath
*m
= (struct multipath
*) ti
->private;
848 flush_workqueue(kmpath_handlerd
);
849 flush_workqueue(kmultipathd
);
854 * Map bios, recording original fields for later in case we have to resubmit
856 static int multipath_map(struct dm_target
*ti
, struct bio
*bio
,
857 union map_info
*map_context
)
860 struct dm_mpath_io
*mpio
;
861 struct multipath
*m
= (struct multipath
*) ti
->private;
863 mpio
= mempool_alloc(m
->mpio_pool
, GFP_NOIO
);
864 dm_bio_record(&mpio
->details
, bio
);
866 map_context
->ptr
= mpio
;
867 bio
->bi_rw
|= (1 << BIO_RW_FAILFAST_TRANSPORT
);
868 r
= map_io(m
, bio
, mpio
, 0);
869 if (r
< 0 || r
== DM_MAPIO_REQUEUE
)
870 mempool_free(mpio
, m
->mpio_pool
);
876 * Take a path out of use.
878 static int fail_path(struct pgpath
*pgpath
)
881 struct multipath
*m
= pgpath
->pg
->m
;
883 spin_lock_irqsave(&m
->lock
, flags
);
885 if (!pgpath
->is_active
)
888 DMWARN("Failing path %s.", pgpath
->path
.dev
->name
);
890 pgpath
->pg
->ps
.type
->fail_path(&pgpath
->pg
->ps
, &pgpath
->path
);
891 pgpath
->is_active
= 0;
892 pgpath
->fail_count
++;
896 if (pgpath
== m
->current_pgpath
)
897 m
->current_pgpath
= NULL
;
899 dm_path_uevent(DM_UEVENT_PATH_FAILED
, m
->ti
,
900 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
902 schedule_work(&m
->trigger_event
);
903 queue_work(kmultipathd
, &pgpath
->deactivate_path
);
906 spin_unlock_irqrestore(&m
->lock
, flags
);
912 * Reinstate a previously-failed path
914 static int reinstate_path(struct pgpath
*pgpath
)
918 struct multipath
*m
= pgpath
->pg
->m
;
920 spin_lock_irqsave(&m
->lock
, flags
);
922 if (pgpath
->is_active
)
925 if (!pgpath
->pg
->ps
.type
->reinstate_path
) {
926 DMWARN("Reinstate path not supported by path selector %s",
927 pgpath
->pg
->ps
.type
->name
);
932 r
= pgpath
->pg
->ps
.type
->reinstate_path(&pgpath
->pg
->ps
, &pgpath
->path
);
936 pgpath
->is_active
= 1;
938 m
->current_pgpath
= NULL
;
939 if (!m
->nr_valid_paths
++ && m
->queue_size
)
940 queue_work(kmultipathd
, &m
->process_queued_ios
);
942 dm_path_uevent(DM_UEVENT_PATH_REINSTATED
, m
->ti
,
943 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
945 schedule_work(&m
->trigger_event
);
948 spin_unlock_irqrestore(&m
->lock
, flags
);
954 * Fail or reinstate all paths that match the provided struct dm_dev.
956 static int action_dev(struct multipath
*m
, struct dm_dev
*dev
,
960 struct pgpath
*pgpath
;
961 struct priority_group
*pg
;
963 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
964 list_for_each_entry(pgpath
, &pg
->pgpaths
, list
) {
965 if (pgpath
->path
.dev
== dev
)
974 * Temporarily try to avoid having to use the specified PG
976 static void bypass_pg(struct multipath
*m
, struct priority_group
*pg
,
981 spin_lock_irqsave(&m
->lock
, flags
);
983 pg
->bypassed
= bypassed
;
984 m
->current_pgpath
= NULL
;
985 m
->current_pg
= NULL
;
987 spin_unlock_irqrestore(&m
->lock
, flags
);
989 schedule_work(&m
->trigger_event
);
993 * Switch to using the specified PG from the next I/O that gets mapped
995 static int switch_pg_num(struct multipath
*m
, const char *pgstr
)
997 struct priority_group
*pg
;
1001 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
1002 (pgnum
> m
->nr_priority_groups
)) {
1003 DMWARN("invalid PG number supplied to switch_pg_num");
1007 spin_lock_irqsave(&m
->lock
, flags
);
1008 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1013 m
->current_pgpath
= NULL
;
1014 m
->current_pg
= NULL
;
1017 spin_unlock_irqrestore(&m
->lock
, flags
);
1019 schedule_work(&m
->trigger_event
);
1024 * Set/clear bypassed status of a PG.
1025 * PGs are numbered upwards from 1 in the order they were declared.
1027 static int bypass_pg_num(struct multipath
*m
, const char *pgstr
, int bypassed
)
1029 struct priority_group
*pg
;
1032 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
1033 (pgnum
> m
->nr_priority_groups
)) {
1034 DMWARN("invalid PG number supplied to bypass_pg");
1038 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1043 bypass_pg(m
, pg
, bypassed
);
1048 * Should we retry pg_init immediately?
1050 static int pg_init_limit_reached(struct multipath
*m
, struct pgpath
*pgpath
)
1052 unsigned long flags
;
1053 int limit_reached
= 0;
1055 spin_lock_irqsave(&m
->lock
, flags
);
1057 if (m
->pg_init_count
<= m
->pg_init_retries
)
1058 m
->pg_init_required
= 1;
1062 spin_unlock_irqrestore(&m
->lock
, flags
);
1064 return limit_reached
;
1067 static void pg_init_done(struct dm_path
*path
, int errors
)
1069 struct pgpath
*pgpath
= path_to_pgpath(path
);
1070 struct priority_group
*pg
= pgpath
->pg
;
1071 struct multipath
*m
= pg
->m
;
1072 unsigned long flags
;
1074 /* device or driver problems */
1079 if (!m
->hw_handler_name
) {
1083 DMERR("Cannot failover device because scsi_dh_%s was not "
1084 "loaded.", m
->hw_handler_name
);
1086 * Fail path for now, so we do not ping pong
1090 case SCSI_DH_DEV_TEMP_BUSY
:
1092 * Probably doing something like FW upgrade on the
1093 * controller so try the other pg.
1095 bypass_pg(m
, pg
, 1);
1097 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1099 case SCSI_DH_IMM_RETRY
:
1100 case SCSI_DH_RES_TEMP_UNAVAIL
:
1101 if (pg_init_limit_reached(m
, pgpath
))
1107 * We probably do not want to fail the path for a device
1108 * error, but this is what the old dm did. In future
1109 * patches we can do more advanced handling.
1114 spin_lock_irqsave(&m
->lock
, flags
);
1116 DMERR("Could not failover device. Error %d.", errors
);
1117 m
->current_pgpath
= NULL
;
1118 m
->current_pg
= NULL
;
1119 } else if (!m
->pg_init_required
) {
1124 m
->pg_init_in_progress
= 0;
1125 queue_work(kmultipathd
, &m
->process_queued_ios
);
1126 spin_unlock_irqrestore(&m
->lock
, flags
);
1129 static void activate_path(struct work_struct
*work
)
1132 struct multipath
*m
=
1133 container_of(work
, struct multipath
, activate_path
);
1134 struct dm_path
*path
;
1135 unsigned long flags
;
1137 spin_lock_irqsave(&m
->lock
, flags
);
1138 path
= &m
->pgpath_to_activate
->path
;
1139 m
->pgpath_to_activate
= NULL
;
1140 spin_unlock_irqrestore(&m
->lock
, flags
);
1143 ret
= scsi_dh_activate(bdev_get_queue(path
->dev
->bdev
));
1144 pg_init_done(path
, ret
);
1150 static int do_end_io(struct multipath
*m
, struct bio
*bio
,
1151 int error
, struct dm_mpath_io
*mpio
)
1153 unsigned long flags
;
1156 return 0; /* I/O complete */
1158 if ((error
== -EWOULDBLOCK
) && bio_rw_ahead(bio
))
1161 if (error
== -EOPNOTSUPP
)
1164 spin_lock_irqsave(&m
->lock
, flags
);
1165 if (!m
->nr_valid_paths
) {
1166 if (__must_push_back(m
)) {
1167 spin_unlock_irqrestore(&m
->lock
, flags
);
1168 return DM_ENDIO_REQUEUE
;
1169 } else if (!m
->queue_if_no_path
) {
1170 spin_unlock_irqrestore(&m
->lock
, flags
);
1173 spin_unlock_irqrestore(&m
->lock
, flags
);
1177 spin_unlock_irqrestore(&m
->lock
, flags
);
1180 fail_path(mpio
->pgpath
);
1183 dm_bio_restore(&mpio
->details
, bio
);
1185 /* queue for the daemon to resubmit or fail */
1186 spin_lock_irqsave(&m
->lock
, flags
);
1187 bio_list_add(&m
->queued_ios
, bio
);
1190 queue_work(kmultipathd
, &m
->process_queued_ios
);
1191 spin_unlock_irqrestore(&m
->lock
, flags
);
1193 return DM_ENDIO_INCOMPLETE
; /* io not complete */
1196 static int multipath_end_io(struct dm_target
*ti
, struct bio
*bio
,
1197 int error
, union map_info
*map_context
)
1199 struct multipath
*m
= ti
->private;
1200 struct dm_mpath_io
*mpio
= map_context
->ptr
;
1201 struct pgpath
*pgpath
= mpio
->pgpath
;
1202 struct path_selector
*ps
;
1205 r
= do_end_io(m
, bio
, error
, mpio
);
1207 ps
= &pgpath
->pg
->ps
;
1208 if (ps
->type
->end_io
)
1209 ps
->type
->end_io(ps
, &pgpath
->path
);
1211 if (r
!= DM_ENDIO_INCOMPLETE
)
1212 mempool_free(mpio
, m
->mpio_pool
);
1218 * Suspend can't complete until all the I/O is processed so if
1219 * the last path fails we must error any remaining I/O.
1220 * Note that if the freeze_bdev fails while suspending, the
1221 * queue_if_no_path state is lost - userspace should reset it.
1223 static void multipath_presuspend(struct dm_target
*ti
)
1225 struct multipath
*m
= (struct multipath
*) ti
->private;
1227 queue_if_no_path(m
, 0, 1);
1231 * Restore the queue_if_no_path setting.
1233 static void multipath_resume(struct dm_target
*ti
)
1235 struct multipath
*m
= (struct multipath
*) ti
->private;
1236 unsigned long flags
;
1238 spin_lock_irqsave(&m
->lock
, flags
);
1239 m
->queue_if_no_path
= m
->saved_queue_if_no_path
;
1240 spin_unlock_irqrestore(&m
->lock
, flags
);
1244 * Info output has the following format:
1245 * num_multipath_feature_args [multipath_feature_args]*
1246 * num_handler_status_args [handler_status_args]*
1247 * num_groups init_group_number
1248 * [A|D|E num_ps_status_args [ps_status_args]*
1249 * num_paths num_selector_args
1250 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1252 * Table output has the following format (identical to the constructor string):
1253 * num_feature_args [features_args]*
1254 * num_handler_args hw_handler [hw_handler_args]*
1255 * num_groups init_group_number
1256 * [priority selector-name num_ps_args [ps_args]*
1257 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1259 static int multipath_status(struct dm_target
*ti
, status_type_t type
,
1260 char *result
, unsigned int maxlen
)
1263 unsigned long flags
;
1264 struct multipath
*m
= (struct multipath
*) ti
->private;
1265 struct priority_group
*pg
;
1270 spin_lock_irqsave(&m
->lock
, flags
);
1273 if (type
== STATUSTYPE_INFO
)
1274 DMEMIT("2 %u %u ", m
->queue_size
, m
->pg_init_count
);
1276 DMEMIT("%u ", m
->queue_if_no_path
+
1277 (m
->pg_init_retries
> 0) * 2);
1278 if (m
->queue_if_no_path
)
1279 DMEMIT("queue_if_no_path ");
1280 if (m
->pg_init_retries
)
1281 DMEMIT("pg_init_retries %u ", m
->pg_init_retries
);
1284 if (!m
->hw_handler_name
|| type
== STATUSTYPE_INFO
)
1287 DMEMIT("1 %s ", m
->hw_handler_name
);
1289 DMEMIT("%u ", m
->nr_priority_groups
);
1292 pg_num
= m
->next_pg
->pg_num
;
1293 else if (m
->current_pg
)
1294 pg_num
= m
->current_pg
->pg_num
;
1298 DMEMIT("%u ", pg_num
);
1301 case STATUSTYPE_INFO
:
1302 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1304 state
= 'D'; /* Disabled */
1305 else if (pg
== m
->current_pg
)
1306 state
= 'A'; /* Currently Active */
1308 state
= 'E'; /* Enabled */
1310 DMEMIT("%c ", state
);
1312 if (pg
->ps
.type
->status
)
1313 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1319 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1320 pg
->ps
.type
->info_args
);
1322 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1323 DMEMIT("%s %s %u ", p
->path
.dev
->name
,
1324 p
->is_active
? "A" : "F",
1326 if (pg
->ps
.type
->status
)
1327 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1328 &p
->path
, type
, result
+ sz
,
1334 case STATUSTYPE_TABLE
:
1335 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1336 DMEMIT("%s ", pg
->ps
.type
->name
);
1338 if (pg
->ps
.type
->status
)
1339 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1345 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1346 pg
->ps
.type
->table_args
);
1348 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1349 DMEMIT("%s ", p
->path
.dev
->name
);
1350 if (pg
->ps
.type
->status
)
1351 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1352 &p
->path
, type
, result
+ sz
,
1359 spin_unlock_irqrestore(&m
->lock
, flags
);
1364 static int multipath_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
1368 struct multipath
*m
= (struct multipath
*) ti
->private;
1372 if (!strnicmp(argv
[0], MESG_STR("queue_if_no_path")))
1373 return queue_if_no_path(m
, 1, 0);
1374 else if (!strnicmp(argv
[0], MESG_STR("fail_if_no_path")))
1375 return queue_if_no_path(m
, 0, 0);
1381 if (!strnicmp(argv
[0], MESG_STR("disable_group")))
1382 return bypass_pg_num(m
, argv
[1], 1);
1383 else if (!strnicmp(argv
[0], MESG_STR("enable_group")))
1384 return bypass_pg_num(m
, argv
[1], 0);
1385 else if (!strnicmp(argv
[0], MESG_STR("switch_group")))
1386 return switch_pg_num(m
, argv
[1]);
1387 else if (!strnicmp(argv
[0], MESG_STR("reinstate_path")))
1388 action
= reinstate_path
;
1389 else if (!strnicmp(argv
[0], MESG_STR("fail_path")))
1394 r
= dm_get_device(ti
, argv
[1], ti
->begin
, ti
->len
,
1395 dm_table_get_mode(ti
->table
), &dev
);
1397 DMWARN("message: error getting device %s",
1402 r
= action_dev(m
, dev
, action
);
1404 dm_put_device(ti
, dev
);
1409 DMWARN("Unrecognised multipath message received.");
1413 static int multipath_ioctl(struct dm_target
*ti
, unsigned int cmd
,
1416 struct multipath
*m
= (struct multipath
*) ti
->private;
1417 struct block_device
*bdev
= NULL
;
1419 unsigned long flags
;
1422 spin_lock_irqsave(&m
->lock
, flags
);
1424 if (!m
->current_pgpath
)
1427 if (m
->current_pgpath
) {
1428 bdev
= m
->current_pgpath
->path
.dev
->bdev
;
1429 mode
= m
->current_pgpath
->path
.dev
->mode
;
1437 spin_unlock_irqrestore(&m
->lock
, flags
);
1439 return r
? : __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
1442 /*-----------------------------------------------------------------
1444 *---------------------------------------------------------------*/
1445 static struct target_type multipath_target
= {
1446 .name
= "multipath",
1447 .version
= {1, 0, 5},
1448 .module
= THIS_MODULE
,
1449 .ctr
= multipath_ctr
,
1450 .dtr
= multipath_dtr
,
1451 .map
= multipath_map
,
1452 .end_io
= multipath_end_io
,
1453 .presuspend
= multipath_presuspend
,
1454 .resume
= multipath_resume
,
1455 .status
= multipath_status
,
1456 .message
= multipath_message
,
1457 .ioctl
= multipath_ioctl
,
1460 static int __init
dm_multipath_init(void)
1464 /* allocate a slab for the dm_ios */
1465 _mpio_cache
= KMEM_CACHE(dm_mpath_io
, 0);
1469 r
= dm_register_target(&multipath_target
);
1471 DMERR("register failed %d", r
);
1472 kmem_cache_destroy(_mpio_cache
);
1476 kmultipathd
= create_workqueue("kmpathd");
1478 DMERR("failed to create workqueue kmpathd");
1479 dm_unregister_target(&multipath_target
);
1480 kmem_cache_destroy(_mpio_cache
);
1485 * A separate workqueue is used to handle the device handlers
1486 * to avoid overloading existing workqueue. Overloading the
1487 * old workqueue would also create a bottleneck in the
1488 * path of the storage hardware device activation.
1490 kmpath_handlerd
= create_singlethread_workqueue("kmpath_handlerd");
1491 if (!kmpath_handlerd
) {
1492 DMERR("failed to create workqueue kmpath_handlerd");
1493 destroy_workqueue(kmultipathd
);
1494 dm_unregister_target(&multipath_target
);
1495 kmem_cache_destroy(_mpio_cache
);
1499 DMINFO("version %u.%u.%u loaded",
1500 multipath_target
.version
[0], multipath_target
.version
[1],
1501 multipath_target
.version
[2]);
1506 static void __exit
dm_multipath_exit(void)
1508 destroy_workqueue(kmpath_handlerd
);
1509 destroy_workqueue(kmultipathd
);
1511 dm_unregister_target(&multipath_target
);
1512 kmem_cache_destroy(_mpio_cache
);
1515 module_init(dm_multipath_init
);
1516 module_exit(dm_multipath_exit
);
1518 MODULE_DESCRIPTION(DM_NAME
" multipath target");
1519 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1520 MODULE_LICENSE("GPL");