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
;
151 list_for_each_entry_safe(pgpath
, tmp
, pgpaths
, list
) {
152 list_del(&pgpath
->list
);
153 dm_put_device(ti
, pgpath
->path
.dev
);
158 static void free_priority_group(struct priority_group
*pg
,
159 struct dm_target
*ti
)
161 struct path_selector
*ps
= &pg
->ps
;
164 ps
->type
->destroy(ps
);
165 dm_put_path_selector(ps
->type
);
168 free_pgpaths(&pg
->pgpaths
, ti
);
172 static struct multipath
*alloc_multipath(struct dm_target
*ti
)
176 m
= kzalloc(sizeof(*m
), GFP_KERNEL
);
178 INIT_LIST_HEAD(&m
->priority_groups
);
179 spin_lock_init(&m
->lock
);
181 INIT_WORK(&m
->process_queued_ios
, process_queued_ios
);
182 INIT_WORK(&m
->trigger_event
, trigger_event
);
183 INIT_WORK(&m
->activate_path
, activate_path
);
184 m
->mpio_pool
= mempool_create_slab_pool(MIN_IOS
, _mpio_cache
);
196 static void free_multipath(struct multipath
*m
)
198 struct priority_group
*pg
, *tmp
;
200 list_for_each_entry_safe(pg
, tmp
, &m
->priority_groups
, list
) {
202 free_priority_group(pg
, m
->ti
);
205 kfree(m
->hw_handler_name
);
206 mempool_destroy(m
->mpio_pool
);
211 /*-----------------------------------------------
213 *-----------------------------------------------*/
215 static void __switch_pg(struct multipath
*m
, struct pgpath
*pgpath
)
217 m
->current_pg
= pgpath
->pg
;
219 /* Must we initialise the PG first, and queue I/O till it's ready? */
220 if (m
->hw_handler_name
) {
221 m
->pg_init_required
= 1;
224 m
->pg_init_required
= 0;
228 m
->pg_init_count
= 0;
231 static int __choose_path_in_pg(struct multipath
*m
, struct priority_group
*pg
)
233 struct dm_path
*path
;
235 path
= pg
->ps
.type
->select_path(&pg
->ps
, &m
->repeat_count
);
239 m
->current_pgpath
= path_to_pgpath(path
);
241 if (m
->current_pg
!= pg
)
242 __switch_pg(m
, m
->current_pgpath
);
247 static void __choose_pgpath(struct multipath
*m
)
249 struct priority_group
*pg
;
250 unsigned bypassed
= 1;
252 if (!m
->nr_valid_paths
)
255 /* Were we instructed to switch PG? */
259 if (!__choose_path_in_pg(m
, pg
))
263 /* Don't change PG until it has no remaining paths */
264 if (m
->current_pg
&& !__choose_path_in_pg(m
, m
->current_pg
))
268 * Loop through priority groups until we find a valid path.
269 * First time we skip PGs marked 'bypassed'.
270 * Second time we only try the ones we skipped.
273 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
274 if (pg
->bypassed
== bypassed
)
276 if (!__choose_path_in_pg(m
, pg
))
279 } while (bypassed
--);
282 m
->current_pgpath
= NULL
;
283 m
->current_pg
= NULL
;
287 * Check whether bios must be queued in the device-mapper core rather
288 * than here in the target.
290 * m->lock must be held on entry.
292 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
293 * same value then we are not between multipath_presuspend()
294 * and multipath_resume() calls and we have no need to check
295 * for the DMF_NOFLUSH_SUSPENDING flag.
297 static int __must_push_back(struct multipath
*m
)
299 return (m
->queue_if_no_path
!= m
->saved_queue_if_no_path
&&
300 dm_noflush_suspending(m
->ti
));
303 static int map_io(struct multipath
*m
, struct bio
*bio
,
304 struct dm_mpath_io
*mpio
, unsigned was_queued
)
306 int r
= DM_MAPIO_REMAPPED
;
308 struct pgpath
*pgpath
;
310 spin_lock_irqsave(&m
->lock
, flags
);
312 /* Do we need to select a new pgpath? */
313 if (!m
->current_pgpath
||
314 (!m
->queue_io
&& (m
->repeat_count
&& --m
->repeat_count
== 0)))
317 pgpath
= m
->current_pgpath
;
322 if ((pgpath
&& m
->queue_io
) ||
323 (!pgpath
&& m
->queue_if_no_path
)) {
324 /* Queue for the daemon to resubmit */
325 bio_list_add(&m
->queued_ios
, bio
);
327 if ((m
->pg_init_required
&& !m
->pg_init_in_progress
) ||
329 queue_work(kmultipathd
, &m
->process_queued_ios
);
331 r
= DM_MAPIO_SUBMITTED
;
333 bio
->bi_bdev
= pgpath
->path
.dev
->bdev
;
334 else if (__must_push_back(m
))
335 r
= DM_MAPIO_REQUEUE
;
337 r
= -EIO
; /* Failed */
339 mpio
->pgpath
= pgpath
;
341 spin_unlock_irqrestore(&m
->lock
, flags
);
347 * If we run out of usable paths, should we queue I/O or error it?
349 static int queue_if_no_path(struct multipath
*m
, unsigned queue_if_no_path
,
350 unsigned save_old_value
)
354 spin_lock_irqsave(&m
->lock
, flags
);
357 m
->saved_queue_if_no_path
= m
->queue_if_no_path
;
359 m
->saved_queue_if_no_path
= queue_if_no_path
;
360 m
->queue_if_no_path
= queue_if_no_path
;
361 if (!m
->queue_if_no_path
&& m
->queue_size
)
362 queue_work(kmultipathd
, &m
->process_queued_ios
);
364 spin_unlock_irqrestore(&m
->lock
, flags
);
369 /*-----------------------------------------------------------------
370 * The multipath daemon is responsible for resubmitting queued ios.
371 *---------------------------------------------------------------*/
373 static void dispatch_queued_ios(struct multipath
*m
)
377 struct bio
*bio
= NULL
, *next
;
378 struct dm_mpath_io
*mpio
;
379 union map_info
*info
;
381 spin_lock_irqsave(&m
->lock
, flags
);
382 bio
= bio_list_get(&m
->queued_ios
);
383 spin_unlock_irqrestore(&m
->lock
, flags
);
389 info
= dm_get_mapinfo(bio
);
392 r
= map_io(m
, bio
, mpio
, 1);
395 else if (r
== DM_MAPIO_REMAPPED
)
396 generic_make_request(bio
);
397 else if (r
== DM_MAPIO_REQUEUE
)
398 bio_endio(bio
, -EIO
);
404 static void process_queued_ios(struct work_struct
*work
)
406 struct multipath
*m
=
407 container_of(work
, struct multipath
, process_queued_ios
);
408 struct pgpath
*pgpath
= NULL
;
409 unsigned init_required
= 0, must_queue
= 1;
412 spin_lock_irqsave(&m
->lock
, flags
);
417 if (!m
->current_pgpath
)
420 pgpath
= m
->current_pgpath
;
422 if ((pgpath
&& !m
->queue_io
) ||
423 (!pgpath
&& !m
->queue_if_no_path
))
426 if (m
->pg_init_required
&& !m
->pg_init_in_progress
) {
428 m
->pg_init_required
= 0;
429 m
->pg_init_in_progress
= 1;
434 spin_unlock_irqrestore(&m
->lock
, flags
);
437 queue_work(kmpath_handlerd
, &m
->activate_path
);
440 dispatch_queued_ios(m
);
444 * An event is triggered whenever a path is taken out of use.
445 * Includes path failure and PG bypass.
447 static void trigger_event(struct work_struct
*work
)
449 struct multipath
*m
=
450 container_of(work
, struct multipath
, trigger_event
);
452 dm_table_event(m
->ti
->table
);
455 /*-----------------------------------------------------------------
456 * Constructor/argument parsing:
457 * <#multipath feature args> [<arg>]*
458 * <#hw_handler args> [hw_handler [<arg>]*]
460 * <initial priority group>
461 * [<selector> <#selector args> [<arg>]*
462 * <#paths> <#per-path selector args>
463 * [<path> [<arg>]* ]+ ]+
464 *---------------------------------------------------------------*/
471 static int read_param(struct param
*param
, char *str
, unsigned *v
, char **error
)
474 (sscanf(str
, "%u", v
) != 1) ||
477 *error
= param
->error
;
489 static char *shift(struct arg_set
*as
)
503 static void consume(struct arg_set
*as
, unsigned n
)
505 BUG_ON (as
->argc
< n
);
510 static int parse_path_selector(struct arg_set
*as
, struct priority_group
*pg
,
511 struct dm_target
*ti
)
514 struct path_selector_type
*pst
;
517 static struct param _params
[] = {
518 {0, 1024, "invalid number of path selector args"},
521 pst
= dm_get_path_selector(shift(as
));
523 ti
->error
= "unknown path selector type";
527 r
= read_param(_params
, shift(as
), &ps_argc
, &ti
->error
);
531 r
= pst
->create(&pg
->ps
, ps_argc
, as
->argv
);
533 dm_put_path_selector(pst
);
534 ti
->error
= "path selector constructor failed";
539 consume(as
, ps_argc
);
544 static struct pgpath
*parse_path(struct arg_set
*as
, struct path_selector
*ps
,
545 struct dm_target
*ti
)
550 /* we need at least a path arg */
552 ti
->error
= "no device given";
560 r
= dm_get_device(ti
, shift(as
), ti
->begin
, ti
->len
,
561 dm_table_get_mode(ti
->table
), &p
->path
.dev
);
563 ti
->error
= "error getting device";
567 r
= ps
->type
->add_path(ps
, &p
->path
, as
->argc
, as
->argv
, &ti
->error
);
569 dm_put_device(ti
, p
->path
.dev
);
580 static struct priority_group
*parse_priority_group(struct arg_set
*as
,
583 static struct param _params
[] = {
584 {1, 1024, "invalid number of paths"},
585 {0, 1024, "invalid number of selector args"}
589 unsigned i
, nr_selector_args
, nr_params
;
590 struct priority_group
*pg
;
591 struct dm_target
*ti
= m
->ti
;
595 ti
->error
= "not enough priority group aruments";
599 pg
= alloc_priority_group();
601 ti
->error
= "couldn't allocate priority group";
606 r
= parse_path_selector(as
, pg
, ti
);
613 r
= read_param(_params
, shift(as
), &pg
->nr_pgpaths
, &ti
->error
);
617 r
= read_param(_params
+ 1, shift(as
), &nr_selector_args
, &ti
->error
);
621 nr_params
= 1 + nr_selector_args
;
622 for (i
= 0; i
< pg
->nr_pgpaths
; i
++) {
623 struct pgpath
*pgpath
;
624 struct arg_set path_args
;
626 if (as
->argc
< nr_params
)
629 path_args
.argc
= nr_params
;
630 path_args
.argv
= as
->argv
;
632 pgpath
= parse_path(&path_args
, &pg
->ps
, ti
);
637 list_add_tail(&pgpath
->list
, &pg
->pgpaths
);
638 consume(as
, nr_params
);
644 free_priority_group(pg
, ti
);
648 static int parse_hw_handler(struct arg_set
*as
, struct multipath
*m
)
651 struct dm_target
*ti
= m
->ti
;
653 static struct param _params
[] = {
654 {0, 1024, "invalid number of hardware handler args"},
657 if (read_param(_params
, shift(as
), &hw_argc
, &ti
->error
))
663 m
->hw_handler_name
= kstrdup(shift(as
), GFP_KERNEL
);
664 request_module("scsi_dh_%s", m
->hw_handler_name
);
665 if (scsi_dh_handler_exist(m
->hw_handler_name
) == 0) {
666 ti
->error
= "unknown hardware handler type";
667 kfree(m
->hw_handler_name
);
668 m
->hw_handler_name
= NULL
;
671 consume(as
, hw_argc
- 1);
676 static int parse_features(struct arg_set
*as
, struct multipath
*m
)
680 struct dm_target
*ti
= m
->ti
;
681 const char *param_name
;
683 static struct param _params
[] = {
684 {0, 3, "invalid number of feature args"},
685 {1, 50, "pg_init_retries must be between 1 and 50"},
688 r
= read_param(_params
, shift(as
), &argc
, &ti
->error
);
696 param_name
= shift(as
);
699 if (!strnicmp(param_name
, MESG_STR("queue_if_no_path"))) {
700 r
= queue_if_no_path(m
, 1, 0);
704 if (!strnicmp(param_name
, MESG_STR("pg_init_retries")) &&
706 r
= read_param(_params
+ 1, shift(as
),
707 &m
->pg_init_retries
, &ti
->error
);
712 ti
->error
= "Unrecognised multipath feature request";
714 } while (argc
&& !r
);
719 static int multipath_ctr(struct dm_target
*ti
, unsigned int argc
,
722 /* target parameters */
723 static struct param _params
[] = {
724 {1, 1024, "invalid number of priority groups"},
725 {1, 1024, "invalid initial priority group number"},
731 unsigned pg_count
= 0;
732 unsigned next_pg_num
;
737 m
= alloc_multipath(ti
);
739 ti
->error
= "can't allocate multipath";
743 r
= parse_features(&as
, m
);
747 r
= parse_hw_handler(&as
, m
);
751 r
= read_param(_params
, shift(&as
), &m
->nr_priority_groups
, &ti
->error
);
755 r
= read_param(_params
+ 1, shift(&as
), &next_pg_num
, &ti
->error
);
759 /* parse the priority groups */
761 struct priority_group
*pg
;
763 pg
= parse_priority_group(&as
, m
);
769 m
->nr_valid_paths
+= pg
->nr_pgpaths
;
770 list_add_tail(&pg
->list
, &m
->priority_groups
);
772 pg
->pg_num
= pg_count
;
777 if (pg_count
!= m
->nr_priority_groups
) {
778 ti
->error
= "priority group count mismatch";
790 static void multipath_dtr(struct dm_target
*ti
)
792 struct multipath
*m
= (struct multipath
*) ti
->private;
794 flush_workqueue(kmpath_handlerd
);
795 flush_workqueue(kmultipathd
);
800 * Map bios, recording original fields for later in case we have to resubmit
802 static int multipath_map(struct dm_target
*ti
, struct bio
*bio
,
803 union map_info
*map_context
)
806 struct dm_mpath_io
*mpio
;
807 struct multipath
*m
= (struct multipath
*) ti
->private;
809 mpio
= mempool_alloc(m
->mpio_pool
, GFP_NOIO
);
810 dm_bio_record(&mpio
->details
, bio
);
812 map_context
->ptr
= mpio
;
813 bio
->bi_rw
|= (1 << BIO_RW_FAILFAST
);
814 r
= map_io(m
, bio
, mpio
, 0);
815 if (r
< 0 || r
== DM_MAPIO_REQUEUE
)
816 mempool_free(mpio
, m
->mpio_pool
);
822 * Take a path out of use.
824 static int fail_path(struct pgpath
*pgpath
)
827 struct multipath
*m
= pgpath
->pg
->m
;
829 spin_lock_irqsave(&m
->lock
, flags
);
831 if (!pgpath
->path
.is_active
)
834 DMWARN("Failing path %s.", pgpath
->path
.dev
->name
);
836 pgpath
->pg
->ps
.type
->fail_path(&pgpath
->pg
->ps
, &pgpath
->path
);
837 pgpath
->path
.is_active
= 0;
838 pgpath
->fail_count
++;
842 if (pgpath
== m
->current_pgpath
)
843 m
->current_pgpath
= NULL
;
845 dm_path_uevent(DM_UEVENT_PATH_FAILED
, m
->ti
,
846 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
848 queue_work(kmultipathd
, &m
->trigger_event
);
851 spin_unlock_irqrestore(&m
->lock
, flags
);
857 * Reinstate a previously-failed path
859 static int reinstate_path(struct pgpath
*pgpath
)
863 struct multipath
*m
= pgpath
->pg
->m
;
865 spin_lock_irqsave(&m
->lock
, flags
);
867 if (pgpath
->path
.is_active
)
870 if (!pgpath
->pg
->ps
.type
) {
871 DMWARN("Reinstate path not supported by path selector %s",
872 pgpath
->pg
->ps
.type
->name
);
877 r
= pgpath
->pg
->ps
.type
->reinstate_path(&pgpath
->pg
->ps
, &pgpath
->path
);
881 pgpath
->path
.is_active
= 1;
883 m
->current_pgpath
= NULL
;
884 if (!m
->nr_valid_paths
++ && m
->queue_size
)
885 queue_work(kmultipathd
, &m
->process_queued_ios
);
887 dm_path_uevent(DM_UEVENT_PATH_REINSTATED
, m
->ti
,
888 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
890 queue_work(kmultipathd
, &m
->trigger_event
);
893 spin_unlock_irqrestore(&m
->lock
, flags
);
899 * Fail or reinstate all paths that match the provided struct dm_dev.
901 static int action_dev(struct multipath
*m
, struct dm_dev
*dev
,
905 struct pgpath
*pgpath
;
906 struct priority_group
*pg
;
908 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
909 list_for_each_entry(pgpath
, &pg
->pgpaths
, list
) {
910 if (pgpath
->path
.dev
== dev
)
919 * Temporarily try to avoid having to use the specified PG
921 static void bypass_pg(struct multipath
*m
, struct priority_group
*pg
,
926 spin_lock_irqsave(&m
->lock
, flags
);
928 pg
->bypassed
= bypassed
;
929 m
->current_pgpath
= NULL
;
930 m
->current_pg
= NULL
;
932 spin_unlock_irqrestore(&m
->lock
, flags
);
934 queue_work(kmultipathd
, &m
->trigger_event
);
938 * Switch to using the specified PG from the next I/O that gets mapped
940 static int switch_pg_num(struct multipath
*m
, const char *pgstr
)
942 struct priority_group
*pg
;
946 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
947 (pgnum
> m
->nr_priority_groups
)) {
948 DMWARN("invalid PG number supplied to switch_pg_num");
952 spin_lock_irqsave(&m
->lock
, flags
);
953 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
958 m
->current_pgpath
= NULL
;
959 m
->current_pg
= NULL
;
962 spin_unlock_irqrestore(&m
->lock
, flags
);
964 queue_work(kmultipathd
, &m
->trigger_event
);
969 * Set/clear bypassed status of a PG.
970 * PGs are numbered upwards from 1 in the order they were declared.
972 static int bypass_pg_num(struct multipath
*m
, const char *pgstr
, int bypassed
)
974 struct priority_group
*pg
;
977 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
978 (pgnum
> m
->nr_priority_groups
)) {
979 DMWARN("invalid PG number supplied to bypass_pg");
983 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
988 bypass_pg(m
, pg
, bypassed
);
993 * Should we retry pg_init immediately?
995 static int pg_init_limit_reached(struct multipath
*m
, struct pgpath
*pgpath
)
998 int limit_reached
= 0;
1000 spin_lock_irqsave(&m
->lock
, flags
);
1002 if (m
->pg_init_count
<= m
->pg_init_retries
)
1003 m
->pg_init_required
= 1;
1007 spin_unlock_irqrestore(&m
->lock
, flags
);
1009 return limit_reached
;
1012 static void pg_init_done(struct dm_path
*path
, int errors
)
1014 struct pgpath
*pgpath
= path_to_pgpath(path
);
1015 struct priority_group
*pg
= pgpath
->pg
;
1016 struct multipath
*m
= pg
->m
;
1017 unsigned long flags
;
1019 /* device or driver problems */
1024 if (!m
->hw_handler_name
) {
1028 DMERR("Cannot failover device because scsi_dh_%s was not "
1029 "loaded.", m
->hw_handler_name
);
1031 * Fail path for now, so we do not ping pong
1035 case SCSI_DH_DEV_TEMP_BUSY
:
1037 * Probably doing something like FW upgrade on the
1038 * controller so try the other pg.
1040 bypass_pg(m
, pg
, 1);
1042 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1044 case SCSI_DH_IMM_RETRY
:
1045 case SCSI_DH_RES_TEMP_UNAVAIL
:
1046 if (pg_init_limit_reached(m
, pgpath
))
1052 * We probably do not want to fail the path for a device
1053 * error, but this is what the old dm did. In future
1054 * patches we can do more advanced handling.
1059 spin_lock_irqsave(&m
->lock
, flags
);
1061 DMERR("Could not failover device. Error %d.", errors
);
1062 m
->current_pgpath
= NULL
;
1063 m
->current_pg
= NULL
;
1064 } else if (!m
->pg_init_required
) {
1069 m
->pg_init_in_progress
= 0;
1070 queue_work(kmultipathd
, &m
->process_queued_ios
);
1071 spin_unlock_irqrestore(&m
->lock
, flags
);
1074 static void activate_path(struct work_struct
*work
)
1077 struct multipath
*m
=
1078 container_of(work
, struct multipath
, activate_path
);
1079 struct dm_path
*path
= &m
->current_pgpath
->path
;
1081 ret
= scsi_dh_activate(bdev_get_queue(path
->dev
->bdev
));
1082 pg_init_done(path
, ret
);
1088 static int do_end_io(struct multipath
*m
, struct bio
*bio
,
1089 int error
, struct dm_mpath_io
*mpio
)
1091 unsigned long flags
;
1094 return 0; /* I/O complete */
1096 if ((error
== -EWOULDBLOCK
) && bio_rw_ahead(bio
))
1099 if (error
== -EOPNOTSUPP
)
1102 spin_lock_irqsave(&m
->lock
, flags
);
1103 if (!m
->nr_valid_paths
) {
1104 if (__must_push_back(m
)) {
1105 spin_unlock_irqrestore(&m
->lock
, flags
);
1106 return DM_ENDIO_REQUEUE
;
1107 } else if (!m
->queue_if_no_path
) {
1108 spin_unlock_irqrestore(&m
->lock
, flags
);
1111 spin_unlock_irqrestore(&m
->lock
, flags
);
1115 spin_unlock_irqrestore(&m
->lock
, flags
);
1118 fail_path(mpio
->pgpath
);
1121 dm_bio_restore(&mpio
->details
, bio
);
1123 /* queue for the daemon to resubmit or fail */
1124 spin_lock_irqsave(&m
->lock
, flags
);
1125 bio_list_add(&m
->queued_ios
, bio
);
1128 queue_work(kmultipathd
, &m
->process_queued_ios
);
1129 spin_unlock_irqrestore(&m
->lock
, flags
);
1131 return DM_ENDIO_INCOMPLETE
; /* io not complete */
1134 static int multipath_end_io(struct dm_target
*ti
, struct bio
*bio
,
1135 int error
, union map_info
*map_context
)
1137 struct multipath
*m
= ti
->private;
1138 struct dm_mpath_io
*mpio
= map_context
->ptr
;
1139 struct pgpath
*pgpath
= mpio
->pgpath
;
1140 struct path_selector
*ps
;
1143 r
= do_end_io(m
, bio
, error
, mpio
);
1145 ps
= &pgpath
->pg
->ps
;
1146 if (ps
->type
->end_io
)
1147 ps
->type
->end_io(ps
, &pgpath
->path
);
1149 if (r
!= DM_ENDIO_INCOMPLETE
)
1150 mempool_free(mpio
, m
->mpio_pool
);
1156 * Suspend can't complete until all the I/O is processed so if
1157 * the last path fails we must error any remaining I/O.
1158 * Note that if the freeze_bdev fails while suspending, the
1159 * queue_if_no_path state is lost - userspace should reset it.
1161 static void multipath_presuspend(struct dm_target
*ti
)
1163 struct multipath
*m
= (struct multipath
*) ti
->private;
1165 queue_if_no_path(m
, 0, 1);
1169 * Restore the queue_if_no_path setting.
1171 static void multipath_resume(struct dm_target
*ti
)
1173 struct multipath
*m
= (struct multipath
*) ti
->private;
1174 unsigned long flags
;
1176 spin_lock_irqsave(&m
->lock
, flags
);
1177 m
->queue_if_no_path
= m
->saved_queue_if_no_path
;
1178 spin_unlock_irqrestore(&m
->lock
, flags
);
1182 * Info output has the following format:
1183 * num_multipath_feature_args [multipath_feature_args]*
1184 * num_handler_status_args [handler_status_args]*
1185 * num_groups init_group_number
1186 * [A|D|E num_ps_status_args [ps_status_args]*
1187 * num_paths num_selector_args
1188 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1190 * Table output has the following format (identical to the constructor string):
1191 * num_feature_args [features_args]*
1192 * num_handler_args hw_handler [hw_handler_args]*
1193 * num_groups init_group_number
1194 * [priority selector-name num_ps_args [ps_args]*
1195 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1197 static int multipath_status(struct dm_target
*ti
, status_type_t type
,
1198 char *result
, unsigned int maxlen
)
1201 unsigned long flags
;
1202 struct multipath
*m
= (struct multipath
*) ti
->private;
1203 struct priority_group
*pg
;
1208 spin_lock_irqsave(&m
->lock
, flags
);
1211 if (type
== STATUSTYPE_INFO
)
1212 DMEMIT("2 %u %u ", m
->queue_size
, m
->pg_init_count
);
1214 DMEMIT("%u ", m
->queue_if_no_path
+
1215 (m
->pg_init_retries
> 0) * 2);
1216 if (m
->queue_if_no_path
)
1217 DMEMIT("queue_if_no_path ");
1218 if (m
->pg_init_retries
)
1219 DMEMIT("pg_init_retries %u ", m
->pg_init_retries
);
1222 if (!m
->hw_handler_name
|| type
== STATUSTYPE_INFO
)
1225 DMEMIT("1 %s ", m
->hw_handler_name
);
1227 DMEMIT("%u ", m
->nr_priority_groups
);
1230 pg_num
= m
->next_pg
->pg_num
;
1231 else if (m
->current_pg
)
1232 pg_num
= m
->current_pg
->pg_num
;
1236 DMEMIT("%u ", pg_num
);
1239 case STATUSTYPE_INFO
:
1240 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1242 state
= 'D'; /* Disabled */
1243 else if (pg
== m
->current_pg
)
1244 state
= 'A'; /* Currently Active */
1246 state
= 'E'; /* Enabled */
1248 DMEMIT("%c ", state
);
1250 if (pg
->ps
.type
->status
)
1251 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1257 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1258 pg
->ps
.type
->info_args
);
1260 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1261 DMEMIT("%s %s %u ", p
->path
.dev
->name
,
1262 p
->path
.is_active
? "A" : "F",
1264 if (pg
->ps
.type
->status
)
1265 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1266 &p
->path
, type
, result
+ sz
,
1272 case STATUSTYPE_TABLE
:
1273 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1274 DMEMIT("%s ", pg
->ps
.type
->name
);
1276 if (pg
->ps
.type
->status
)
1277 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1283 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1284 pg
->ps
.type
->table_args
);
1286 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1287 DMEMIT("%s ", p
->path
.dev
->name
);
1288 if (pg
->ps
.type
->status
)
1289 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1290 &p
->path
, type
, result
+ sz
,
1297 spin_unlock_irqrestore(&m
->lock
, flags
);
1302 static int multipath_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
1306 struct multipath
*m
= (struct multipath
*) ti
->private;
1310 if (!strnicmp(argv
[0], MESG_STR("queue_if_no_path")))
1311 return queue_if_no_path(m
, 1, 0);
1312 else if (!strnicmp(argv
[0], MESG_STR("fail_if_no_path")))
1313 return queue_if_no_path(m
, 0, 0);
1319 if (!strnicmp(argv
[0], MESG_STR("disable_group")))
1320 return bypass_pg_num(m
, argv
[1], 1);
1321 else if (!strnicmp(argv
[0], MESG_STR("enable_group")))
1322 return bypass_pg_num(m
, argv
[1], 0);
1323 else if (!strnicmp(argv
[0], MESG_STR("switch_group")))
1324 return switch_pg_num(m
, argv
[1]);
1325 else if (!strnicmp(argv
[0], MESG_STR("reinstate_path")))
1326 action
= reinstate_path
;
1327 else if (!strnicmp(argv
[0], MESG_STR("fail_path")))
1332 r
= dm_get_device(ti
, argv
[1], ti
->begin
, ti
->len
,
1333 dm_table_get_mode(ti
->table
), &dev
);
1335 DMWARN("message: error getting device %s",
1340 r
= action_dev(m
, dev
, action
);
1342 dm_put_device(ti
, dev
);
1347 DMWARN("Unrecognised multipath message received.");
1351 static int multipath_ioctl(struct dm_target
*ti
, struct inode
*inode
,
1352 struct file
*filp
, unsigned int cmd
,
1355 struct multipath
*m
= (struct multipath
*) ti
->private;
1356 struct block_device
*bdev
= NULL
;
1357 unsigned long flags
;
1358 struct file fake_file
= {};
1359 struct dentry fake_dentry
= {};
1362 fake_file
.f_path
.dentry
= &fake_dentry
;
1364 spin_lock_irqsave(&m
->lock
, flags
);
1366 if (!m
->current_pgpath
)
1369 if (m
->current_pgpath
) {
1370 bdev
= m
->current_pgpath
->path
.dev
->bdev
;
1371 fake_dentry
.d_inode
= bdev
->bd_inode
;
1372 fake_file
.f_mode
= m
->current_pgpath
->path
.dev
->mode
;
1380 spin_unlock_irqrestore(&m
->lock
, flags
);
1382 return r
? : blkdev_driver_ioctl(bdev
->bd_inode
, &fake_file
,
1383 bdev
->bd_disk
, cmd
, arg
);
1386 /*-----------------------------------------------------------------
1388 *---------------------------------------------------------------*/
1389 static struct target_type multipath_target
= {
1390 .name
= "multipath",
1391 .version
= {1, 0, 5},
1392 .module
= THIS_MODULE
,
1393 .ctr
= multipath_ctr
,
1394 .dtr
= multipath_dtr
,
1395 .map
= multipath_map
,
1396 .end_io
= multipath_end_io
,
1397 .presuspend
= multipath_presuspend
,
1398 .resume
= multipath_resume
,
1399 .status
= multipath_status
,
1400 .message
= multipath_message
,
1401 .ioctl
= multipath_ioctl
,
1404 static int __init
dm_multipath_init(void)
1408 /* allocate a slab for the dm_ios */
1409 _mpio_cache
= KMEM_CACHE(dm_mpath_io
, 0);
1413 r
= dm_register_target(&multipath_target
);
1415 DMERR("register failed %d", r
);
1416 kmem_cache_destroy(_mpio_cache
);
1420 kmultipathd
= create_workqueue("kmpathd");
1422 DMERR("failed to create workqueue kmpathd");
1423 dm_unregister_target(&multipath_target
);
1424 kmem_cache_destroy(_mpio_cache
);
1429 * A separate workqueue is used to handle the device handlers
1430 * to avoid overloading existing workqueue. Overloading the
1431 * old workqueue would also create a bottleneck in the
1432 * path of the storage hardware device activation.
1434 kmpath_handlerd
= create_singlethread_workqueue("kmpath_handlerd");
1435 if (!kmpath_handlerd
) {
1436 DMERR("failed to create workqueue kmpath_handlerd");
1437 destroy_workqueue(kmultipathd
);
1438 dm_unregister_target(&multipath_target
);
1439 kmem_cache_destroy(_mpio_cache
);
1443 DMINFO("version %u.%u.%u loaded",
1444 multipath_target
.version
[0], multipath_target
.version
[1],
1445 multipath_target
.version
[2]);
1450 static void __exit
dm_multipath_exit(void)
1454 destroy_workqueue(kmpath_handlerd
);
1455 destroy_workqueue(kmultipathd
);
1457 r
= dm_unregister_target(&multipath_target
);
1459 DMERR("target unregister failed %d", r
);
1460 kmem_cache_destroy(_mpio_cache
);
1463 module_init(dm_multipath_init
);
1464 module_exit(dm_multipath_exit
);
1466 MODULE_DESCRIPTION(DM_NAME
" multipath target");
1467 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1468 MODULE_LICENSE("GPL");