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-hw-handler.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 fail_count
; /* Cumulative failure count */
39 #define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path)
42 * Paths are grouped into Priority Groups and numbered from 1 upwards.
43 * Each has a path selector which controls which path gets used.
45 struct priority_group
{
46 struct list_head list
;
48 struct multipath
*m
; /* Owning multipath instance */
49 struct path_selector ps
;
51 unsigned pg_num
; /* Reference number */
52 unsigned bypassed
; /* Temporarily bypass this PG? */
54 unsigned nr_pgpaths
; /* Number of paths in PG */
55 struct list_head pgpaths
;
58 /* Multipath context */
60 struct list_head list
;
65 const char *hw_handler_name
;
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
;
111 static void process_queued_ios(struct work_struct
*work
);
112 static void trigger_event(struct work_struct
*work
);
113 static void pg_init_done(struct dm_path
*, int);
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 m
->mpio_pool
= mempool_create_slab_pool(MIN_IOS
, _mpio_cache
);
195 static void free_multipath(struct multipath
*m
)
197 struct priority_group
*pg
, *tmp
;
199 list_for_each_entry_safe(pg
, tmp
, &m
->priority_groups
, list
) {
201 free_priority_group(pg
, m
->ti
);
204 kfree(m
->hw_handler_name
);
205 mempool_destroy(m
->mpio_pool
);
210 /*-----------------------------------------------
212 *-----------------------------------------------*/
214 static void __switch_pg(struct multipath
*m
, struct pgpath
*pgpath
)
216 m
->current_pg
= pgpath
->pg
;
218 /* Must we initialise the PG first, and queue I/O till it's ready? */
219 if (m
->hw_handler_name
) {
220 m
->pg_init_required
= 1;
223 m
->pg_init_required
= 0;
227 m
->pg_init_count
= 0;
230 static int __choose_path_in_pg(struct multipath
*m
, struct priority_group
*pg
)
232 struct dm_path
*path
;
234 path
= pg
->ps
.type
->select_path(&pg
->ps
, &m
->repeat_count
);
238 m
->current_pgpath
= path_to_pgpath(path
);
240 if (m
->current_pg
!= pg
)
241 __switch_pg(m
, m
->current_pgpath
);
246 static void __choose_pgpath(struct multipath
*m
)
248 struct priority_group
*pg
;
249 unsigned bypassed
= 1;
251 if (!m
->nr_valid_paths
)
254 /* Were we instructed to switch PG? */
258 if (!__choose_path_in_pg(m
, pg
))
262 /* Don't change PG until it has no remaining paths */
263 if (m
->current_pg
&& !__choose_path_in_pg(m
, m
->current_pg
))
267 * Loop through priority groups until we find a valid path.
268 * First time we skip PGs marked 'bypassed'.
269 * Second time we only try the ones we skipped.
272 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
273 if (pg
->bypassed
== bypassed
)
275 if (!__choose_path_in_pg(m
, pg
))
278 } while (bypassed
--);
281 m
->current_pgpath
= NULL
;
282 m
->current_pg
= NULL
;
286 * Check whether bios must be queued in the device-mapper core rather
287 * than here in the target.
289 * m->lock must be held on entry.
291 * If m->queue_if_no_path and m->saved_queue_if_no_path hold the
292 * same value then we are not between multipath_presuspend()
293 * and multipath_resume() calls and we have no need to check
294 * for the DMF_NOFLUSH_SUSPENDING flag.
296 static int __must_push_back(struct multipath
*m
)
298 return (m
->queue_if_no_path
!= m
->saved_queue_if_no_path
&&
299 dm_noflush_suspending(m
->ti
));
302 static int map_io(struct multipath
*m
, struct bio
*bio
,
303 struct dm_mpath_io
*mpio
, unsigned was_queued
)
305 int r
= DM_MAPIO_REMAPPED
;
307 struct pgpath
*pgpath
;
309 spin_lock_irqsave(&m
->lock
, flags
);
311 /* Do we need to select a new pgpath? */
312 if (!m
->current_pgpath
||
313 (!m
->queue_io
&& (m
->repeat_count
&& --m
->repeat_count
== 0)))
316 pgpath
= m
->current_pgpath
;
321 if ((pgpath
&& m
->queue_io
) ||
322 (!pgpath
&& m
->queue_if_no_path
)) {
323 /* Queue for the daemon to resubmit */
324 bio_list_add(&m
->queued_ios
, bio
);
326 if ((m
->pg_init_required
&& !m
->pg_init_in_progress
) ||
328 queue_work(kmultipathd
, &m
->process_queued_ios
);
330 r
= DM_MAPIO_SUBMITTED
;
332 bio
->bi_bdev
= pgpath
->path
.dev
->bdev
;
333 else if (__must_push_back(m
))
334 r
= DM_MAPIO_REQUEUE
;
336 r
= -EIO
; /* Failed */
338 mpio
->pgpath
= pgpath
;
340 spin_unlock_irqrestore(&m
->lock
, flags
);
346 * If we run out of usable paths, should we queue I/O or error it?
348 static int queue_if_no_path(struct multipath
*m
, unsigned queue_if_no_path
,
349 unsigned save_old_value
)
353 spin_lock_irqsave(&m
->lock
, flags
);
356 m
->saved_queue_if_no_path
= m
->queue_if_no_path
;
358 m
->saved_queue_if_no_path
= queue_if_no_path
;
359 m
->queue_if_no_path
= queue_if_no_path
;
360 if (!m
->queue_if_no_path
&& m
->queue_size
)
361 queue_work(kmultipathd
, &m
->process_queued_ios
);
363 spin_unlock_irqrestore(&m
->lock
, flags
);
368 /*-----------------------------------------------------------------
369 * The multipath daemon is responsible for resubmitting queued ios.
370 *---------------------------------------------------------------*/
372 static void dispatch_queued_ios(struct multipath
*m
)
376 struct bio
*bio
= NULL
, *next
;
377 struct dm_mpath_io
*mpio
;
378 union map_info
*info
;
380 spin_lock_irqsave(&m
->lock
, flags
);
381 bio
= bio_list_get(&m
->queued_ios
);
382 spin_unlock_irqrestore(&m
->lock
, flags
);
388 info
= dm_get_mapinfo(bio
);
391 r
= map_io(m
, bio
, mpio
, 1);
394 else if (r
== DM_MAPIO_REMAPPED
)
395 generic_make_request(bio
);
396 else if (r
== DM_MAPIO_REQUEUE
)
397 bio_endio(bio
, -EIO
);
403 static void process_queued_ios(struct work_struct
*work
)
405 struct multipath
*m
=
406 container_of(work
, struct multipath
, process_queued_ios
);
407 struct pgpath
*pgpath
= NULL
;
408 unsigned init_required
= 0, must_queue
= 1;
411 spin_lock_irqsave(&m
->lock
, flags
);
416 if (!m
->current_pgpath
)
419 pgpath
= m
->current_pgpath
;
421 if ((pgpath
&& !m
->queue_io
) ||
422 (!pgpath
&& !m
->queue_if_no_path
))
425 if (m
->pg_init_required
&& !m
->pg_init_in_progress
) {
427 m
->pg_init_required
= 0;
428 m
->pg_init_in_progress
= 1;
433 spin_unlock_irqrestore(&m
->lock
, flags
);
436 struct dm_path
*path
= &pgpath
->path
;
437 int ret
= scsi_dh_activate(bdev_get_queue(path
->dev
->bdev
));
438 pg_init_done(path
, ret
);
442 dispatch_queued_ios(m
);
446 * An event is triggered whenever a path is taken out of use.
447 * Includes path failure and PG bypass.
449 static void trigger_event(struct work_struct
*work
)
451 struct multipath
*m
=
452 container_of(work
, struct multipath
, trigger_event
);
454 dm_table_event(m
->ti
->table
);
457 /*-----------------------------------------------------------------
458 * Constructor/argument parsing:
459 * <#multipath feature args> [<arg>]*
460 * <#hw_handler args> [hw_handler [<arg>]*]
462 * <initial priority group>
463 * [<selector> <#selector args> [<arg>]*
464 * <#paths> <#per-path selector args>
465 * [<path> [<arg>]* ]+ ]+
466 *---------------------------------------------------------------*/
473 static int read_param(struct param
*param
, char *str
, unsigned *v
, char **error
)
476 (sscanf(str
, "%u", v
) != 1) ||
479 *error
= param
->error
;
491 static char *shift(struct arg_set
*as
)
505 static void consume(struct arg_set
*as
, unsigned n
)
507 BUG_ON (as
->argc
< n
);
512 static int parse_path_selector(struct arg_set
*as
, struct priority_group
*pg
,
513 struct dm_target
*ti
)
516 struct path_selector_type
*pst
;
519 static struct param _params
[] = {
520 {0, 1024, "invalid number of path selector args"},
523 pst
= dm_get_path_selector(shift(as
));
525 ti
->error
= "unknown path selector type";
529 r
= read_param(_params
, shift(as
), &ps_argc
, &ti
->error
);
533 r
= pst
->create(&pg
->ps
, ps_argc
, as
->argv
);
535 dm_put_path_selector(pst
);
536 ti
->error
= "path selector constructor failed";
541 consume(as
, ps_argc
);
546 static struct pgpath
*parse_path(struct arg_set
*as
, struct path_selector
*ps
,
547 struct dm_target
*ti
)
552 /* we need at least a path arg */
554 ti
->error
= "no device given";
562 r
= dm_get_device(ti
, shift(as
), ti
->begin
, ti
->len
,
563 dm_table_get_mode(ti
->table
), &p
->path
.dev
);
565 ti
->error
= "error getting device";
569 r
= ps
->type
->add_path(ps
, &p
->path
, as
->argc
, as
->argv
, &ti
->error
);
571 dm_put_device(ti
, p
->path
.dev
);
582 static struct priority_group
*parse_priority_group(struct arg_set
*as
,
585 static struct param _params
[] = {
586 {1, 1024, "invalid number of paths"},
587 {0, 1024, "invalid number of selector args"}
591 unsigned i
, nr_selector_args
, nr_params
;
592 struct priority_group
*pg
;
593 struct dm_target
*ti
= m
->ti
;
597 ti
->error
= "not enough priority group aruments";
601 pg
= alloc_priority_group();
603 ti
->error
= "couldn't allocate priority group";
608 r
= parse_path_selector(as
, pg
, ti
);
615 r
= read_param(_params
, shift(as
), &pg
->nr_pgpaths
, &ti
->error
);
619 r
= read_param(_params
+ 1, shift(as
), &nr_selector_args
, &ti
->error
);
623 nr_params
= 1 + nr_selector_args
;
624 for (i
= 0; i
< pg
->nr_pgpaths
; i
++) {
625 struct pgpath
*pgpath
;
626 struct arg_set path_args
;
628 if (as
->argc
< nr_params
)
631 path_args
.argc
= nr_params
;
632 path_args
.argv
= as
->argv
;
634 pgpath
= parse_path(&path_args
, &pg
->ps
, ti
);
639 list_add_tail(&pgpath
->list
, &pg
->pgpaths
);
640 consume(as
, nr_params
);
646 free_priority_group(pg
, ti
);
650 static int parse_hw_handler(struct arg_set
*as
, struct multipath
*m
)
653 struct dm_target
*ti
= m
->ti
;
655 static struct param _params
[] = {
656 {0, 1024, "invalid number of hardware handler args"},
659 if (read_param(_params
, shift(as
), &hw_argc
, &ti
->error
))
665 m
->hw_handler_name
= kstrdup(shift(as
), GFP_KERNEL
);
666 request_module("scsi_dh_%s", m
->hw_handler_name
);
667 if (scsi_dh_handler_exist(m
->hw_handler_name
) == 0) {
668 ti
->error
= "unknown hardware handler type";
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(kmultipathd
);
799 * Map bios, recording original fields for later in case we have to resubmit
801 static int multipath_map(struct dm_target
*ti
, struct bio
*bio
,
802 union map_info
*map_context
)
805 struct dm_mpath_io
*mpio
;
806 struct multipath
*m
= (struct multipath
*) ti
->private;
808 mpio
= mempool_alloc(m
->mpio_pool
, GFP_NOIO
);
809 dm_bio_record(&mpio
->details
, bio
);
811 map_context
->ptr
= mpio
;
812 bio
->bi_rw
|= (1 << BIO_RW_FAILFAST
);
813 r
= map_io(m
, bio
, mpio
, 0);
814 if (r
< 0 || r
== DM_MAPIO_REQUEUE
)
815 mempool_free(mpio
, m
->mpio_pool
);
821 * Take a path out of use.
823 static int fail_path(struct pgpath
*pgpath
)
826 struct multipath
*m
= pgpath
->pg
->m
;
828 spin_lock_irqsave(&m
->lock
, flags
);
830 if (!pgpath
->path
.is_active
)
833 DMWARN("Failing path %s.", pgpath
->path
.dev
->name
);
835 pgpath
->pg
->ps
.type
->fail_path(&pgpath
->pg
->ps
, &pgpath
->path
);
836 pgpath
->path
.is_active
= 0;
837 pgpath
->fail_count
++;
841 if (pgpath
== m
->current_pgpath
)
842 m
->current_pgpath
= NULL
;
844 dm_path_uevent(DM_UEVENT_PATH_FAILED
, m
->ti
,
845 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
847 queue_work(kmultipathd
, &m
->trigger_event
);
850 spin_unlock_irqrestore(&m
->lock
, flags
);
856 * Reinstate a previously-failed path
858 static int reinstate_path(struct pgpath
*pgpath
)
862 struct multipath
*m
= pgpath
->pg
->m
;
864 spin_lock_irqsave(&m
->lock
, flags
);
866 if (pgpath
->path
.is_active
)
869 if (!pgpath
->pg
->ps
.type
) {
870 DMWARN("Reinstate path not supported by path selector %s",
871 pgpath
->pg
->ps
.type
->name
);
876 r
= pgpath
->pg
->ps
.type
->reinstate_path(&pgpath
->pg
->ps
, &pgpath
->path
);
880 pgpath
->path
.is_active
= 1;
882 m
->current_pgpath
= NULL
;
883 if (!m
->nr_valid_paths
++ && m
->queue_size
)
884 queue_work(kmultipathd
, &m
->process_queued_ios
);
886 dm_path_uevent(DM_UEVENT_PATH_REINSTATED
, m
->ti
,
887 pgpath
->path
.dev
->name
, m
->nr_valid_paths
);
889 queue_work(kmultipathd
, &m
->trigger_event
);
892 spin_unlock_irqrestore(&m
->lock
, flags
);
898 * Fail or reinstate all paths that match the provided struct dm_dev.
900 static int action_dev(struct multipath
*m
, struct dm_dev
*dev
,
904 struct pgpath
*pgpath
;
905 struct priority_group
*pg
;
907 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
908 list_for_each_entry(pgpath
, &pg
->pgpaths
, list
) {
909 if (pgpath
->path
.dev
== dev
)
918 * Temporarily try to avoid having to use the specified PG
920 static void bypass_pg(struct multipath
*m
, struct priority_group
*pg
,
925 spin_lock_irqsave(&m
->lock
, flags
);
927 pg
->bypassed
= bypassed
;
928 m
->current_pgpath
= NULL
;
929 m
->current_pg
= NULL
;
931 spin_unlock_irqrestore(&m
->lock
, flags
);
933 queue_work(kmultipathd
, &m
->trigger_event
);
937 * Switch to using the specified PG from the next I/O that gets mapped
939 static int switch_pg_num(struct multipath
*m
, const char *pgstr
)
941 struct priority_group
*pg
;
945 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
946 (pgnum
> m
->nr_priority_groups
)) {
947 DMWARN("invalid PG number supplied to switch_pg_num");
951 spin_lock_irqsave(&m
->lock
, flags
);
952 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
957 m
->current_pgpath
= NULL
;
958 m
->current_pg
= NULL
;
961 spin_unlock_irqrestore(&m
->lock
, flags
);
963 queue_work(kmultipathd
, &m
->trigger_event
);
968 * Set/clear bypassed status of a PG.
969 * PGs are numbered upwards from 1 in the order they were declared.
971 static int bypass_pg_num(struct multipath
*m
, const char *pgstr
, int bypassed
)
973 struct priority_group
*pg
;
976 if (!pgstr
|| (sscanf(pgstr
, "%u", &pgnum
) != 1) || !pgnum
||
977 (pgnum
> m
->nr_priority_groups
)) {
978 DMWARN("invalid PG number supplied to bypass_pg");
982 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
987 bypass_pg(m
, pg
, bypassed
);
992 * Should we retry pg_init immediately?
994 static int pg_init_limit_reached(struct multipath
*m
, struct pgpath
*pgpath
)
997 int limit_reached
= 0;
999 spin_lock_irqsave(&m
->lock
, flags
);
1001 if (m
->pg_init_count
<= m
->pg_init_retries
)
1002 m
->pg_init_required
= 1;
1006 spin_unlock_irqrestore(&m
->lock
, flags
);
1008 return limit_reached
;
1012 * pg_init must call this when it has completed its initialisation
1014 void dm_pg_init_complete(struct dm_path
*path
, unsigned err_flags
)
1016 struct pgpath
*pgpath
= path_to_pgpath(path
);
1017 struct priority_group
*pg
= pgpath
->pg
;
1018 struct multipath
*m
= pg
->m
;
1019 unsigned long flags
;
1022 * If requested, retry pg_init until maximum number of retries exceeded.
1023 * If retry not requested and PG already bypassed, always fail the path.
1025 if (err_flags
& MP_RETRY
) {
1026 if (pg_init_limit_reached(m
, pgpath
))
1027 err_flags
|= MP_FAIL_PATH
;
1028 } else if (err_flags
&& pg
->bypassed
)
1029 err_flags
|= MP_FAIL_PATH
;
1031 if (err_flags
& MP_FAIL_PATH
)
1034 if (err_flags
& MP_BYPASS_PG
)
1035 bypass_pg(m
, pg
, 1);
1037 spin_lock_irqsave(&m
->lock
, flags
);
1038 if (err_flags
& ~MP_RETRY
) {
1039 m
->current_pgpath
= NULL
;
1040 m
->current_pg
= NULL
;
1041 } else if (!m
->pg_init_required
)
1044 m
->pg_init_in_progress
= 0;
1045 queue_work(kmultipathd
, &m
->process_queued_ios
);
1046 spin_unlock_irqrestore(&m
->lock
, flags
);
1049 static void pg_init_done(struct dm_path
*path
, int errors
)
1051 struct pgpath
*pgpath
= path_to_pgpath(path
);
1052 struct priority_group
*pg
= pgpath
->pg
;
1053 struct multipath
*m
= pg
->m
;
1054 unsigned long flags
;
1056 /* device or driver problems */
1061 if (!m
->hw_handler_name
) {
1065 DMERR("Cannot failover device because scsi_dh_%s was not "
1066 "loaded.", m
->hw_handler_name
);
1068 * Fail path for now, so we do not ping pong
1072 case SCSI_DH_DEV_TEMP_BUSY
:
1074 * Probably doing something like FW upgrade on the
1075 * controller so try the other pg.
1077 bypass_pg(m
, pg
, 1);
1079 /* TODO: For SCSI_DH_RETRY we should wait a couple seconds */
1081 case SCSI_DH_IMM_RETRY
:
1082 case SCSI_DH_RES_TEMP_UNAVAIL
:
1083 if (pg_init_limit_reached(m
, pgpath
))
1089 * We probably do not want to fail the path for a device
1090 * error, but this is what the old dm did. In future
1091 * patches we can do more advanced handling.
1096 spin_lock_irqsave(&m
->lock
, flags
);
1098 DMERR("Could not failover device. Error %d.", errors
);
1099 m
->current_pgpath
= NULL
;
1100 m
->current_pg
= NULL
;
1101 } else if (!m
->pg_init_required
) {
1106 m
->pg_init_in_progress
= 0;
1107 queue_work(kmultipathd
, &m
->process_queued_ios
);
1108 spin_unlock_irqrestore(&m
->lock
, flags
);
1114 static int do_end_io(struct multipath
*m
, struct bio
*bio
,
1115 int error
, struct dm_mpath_io
*mpio
)
1117 unsigned long flags
;
1120 return 0; /* I/O complete */
1122 if ((error
== -EWOULDBLOCK
) && bio_rw_ahead(bio
))
1125 if (error
== -EOPNOTSUPP
)
1128 spin_lock_irqsave(&m
->lock
, flags
);
1129 if (!m
->nr_valid_paths
) {
1130 if (__must_push_back(m
)) {
1131 spin_unlock_irqrestore(&m
->lock
, flags
);
1132 return DM_ENDIO_REQUEUE
;
1133 } else if (!m
->queue_if_no_path
) {
1134 spin_unlock_irqrestore(&m
->lock
, flags
);
1137 spin_unlock_irqrestore(&m
->lock
, flags
);
1141 spin_unlock_irqrestore(&m
->lock
, flags
);
1144 fail_path(mpio
->pgpath
);
1147 dm_bio_restore(&mpio
->details
, bio
);
1149 /* queue for the daemon to resubmit or fail */
1150 spin_lock_irqsave(&m
->lock
, flags
);
1151 bio_list_add(&m
->queued_ios
, bio
);
1154 queue_work(kmultipathd
, &m
->process_queued_ios
);
1155 spin_unlock_irqrestore(&m
->lock
, flags
);
1157 return DM_ENDIO_INCOMPLETE
; /* io not complete */
1160 static int multipath_end_io(struct dm_target
*ti
, struct bio
*bio
,
1161 int error
, union map_info
*map_context
)
1163 struct multipath
*m
= ti
->private;
1164 struct dm_mpath_io
*mpio
= map_context
->ptr
;
1165 struct pgpath
*pgpath
= mpio
->pgpath
;
1166 struct path_selector
*ps
;
1169 r
= do_end_io(m
, bio
, error
, mpio
);
1171 ps
= &pgpath
->pg
->ps
;
1172 if (ps
->type
->end_io
)
1173 ps
->type
->end_io(ps
, &pgpath
->path
);
1175 if (r
!= DM_ENDIO_INCOMPLETE
)
1176 mempool_free(mpio
, m
->mpio_pool
);
1182 * Suspend can't complete until all the I/O is processed so if
1183 * the last path fails we must error any remaining I/O.
1184 * Note that if the freeze_bdev fails while suspending, the
1185 * queue_if_no_path state is lost - userspace should reset it.
1187 static void multipath_presuspend(struct dm_target
*ti
)
1189 struct multipath
*m
= (struct multipath
*) ti
->private;
1191 queue_if_no_path(m
, 0, 1);
1195 * Restore the queue_if_no_path setting.
1197 static void multipath_resume(struct dm_target
*ti
)
1199 struct multipath
*m
= (struct multipath
*) ti
->private;
1200 unsigned long flags
;
1202 spin_lock_irqsave(&m
->lock
, flags
);
1203 m
->queue_if_no_path
= m
->saved_queue_if_no_path
;
1204 spin_unlock_irqrestore(&m
->lock
, flags
);
1208 * Info output has the following format:
1209 * num_multipath_feature_args [multipath_feature_args]*
1210 * num_handler_status_args [handler_status_args]*
1211 * num_groups init_group_number
1212 * [A|D|E num_ps_status_args [ps_status_args]*
1213 * num_paths num_selector_args
1214 * [path_dev A|F fail_count [selector_args]* ]+ ]+
1216 * Table output has the following format (identical to the constructor string):
1217 * num_feature_args [features_args]*
1218 * num_handler_args hw_handler [hw_handler_args]*
1219 * num_groups init_group_number
1220 * [priority selector-name num_ps_args [ps_args]*
1221 * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+
1223 static int multipath_status(struct dm_target
*ti
, status_type_t type
,
1224 char *result
, unsigned int maxlen
)
1227 unsigned long flags
;
1228 struct multipath
*m
= (struct multipath
*) ti
->private;
1229 struct priority_group
*pg
;
1234 spin_lock_irqsave(&m
->lock
, flags
);
1237 if (type
== STATUSTYPE_INFO
)
1238 DMEMIT("2 %u %u ", m
->queue_size
, m
->pg_init_count
);
1240 DMEMIT("%u ", m
->queue_if_no_path
+
1241 (m
->pg_init_retries
> 0) * 2);
1242 if (m
->queue_if_no_path
)
1243 DMEMIT("queue_if_no_path ");
1244 if (m
->pg_init_retries
)
1245 DMEMIT("pg_init_retries %u ", m
->pg_init_retries
);
1248 if (!m
->hw_handler_name
|| type
== STATUSTYPE_INFO
)
1251 DMEMIT("1 %s ", m
->hw_handler_name
);
1253 DMEMIT("%u ", m
->nr_priority_groups
);
1256 pg_num
= m
->next_pg
->pg_num
;
1257 else if (m
->current_pg
)
1258 pg_num
= m
->current_pg
->pg_num
;
1262 DMEMIT("%u ", pg_num
);
1265 case STATUSTYPE_INFO
:
1266 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1268 state
= 'D'; /* Disabled */
1269 else if (pg
== m
->current_pg
)
1270 state
= 'A'; /* Currently Active */
1272 state
= 'E'; /* Enabled */
1274 DMEMIT("%c ", state
);
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
->info_args
);
1286 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1287 DMEMIT("%s %s %u ", p
->path
.dev
->name
,
1288 p
->path
.is_active
? "A" : "F",
1290 if (pg
->ps
.type
->status
)
1291 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1292 &p
->path
, type
, result
+ sz
,
1298 case STATUSTYPE_TABLE
:
1299 list_for_each_entry(pg
, &m
->priority_groups
, list
) {
1300 DMEMIT("%s ", pg
->ps
.type
->name
);
1302 if (pg
->ps
.type
->status
)
1303 sz
+= pg
->ps
.type
->status(&pg
->ps
, NULL
, type
,
1309 DMEMIT("%u %u ", pg
->nr_pgpaths
,
1310 pg
->ps
.type
->table_args
);
1312 list_for_each_entry(p
, &pg
->pgpaths
, list
) {
1313 DMEMIT("%s ", p
->path
.dev
->name
);
1314 if (pg
->ps
.type
->status
)
1315 sz
+= pg
->ps
.type
->status(&pg
->ps
,
1316 &p
->path
, type
, result
+ sz
,
1323 spin_unlock_irqrestore(&m
->lock
, flags
);
1328 static int multipath_message(struct dm_target
*ti
, unsigned argc
, char **argv
)
1332 struct multipath
*m
= (struct multipath
*) ti
->private;
1336 if (!strnicmp(argv
[0], MESG_STR("queue_if_no_path")))
1337 return queue_if_no_path(m
, 1, 0);
1338 else if (!strnicmp(argv
[0], MESG_STR("fail_if_no_path")))
1339 return queue_if_no_path(m
, 0, 0);
1345 if (!strnicmp(argv
[0], MESG_STR("disable_group")))
1346 return bypass_pg_num(m
, argv
[1], 1);
1347 else if (!strnicmp(argv
[0], MESG_STR("enable_group")))
1348 return bypass_pg_num(m
, argv
[1], 0);
1349 else if (!strnicmp(argv
[0], MESG_STR("switch_group")))
1350 return switch_pg_num(m
, argv
[1]);
1351 else if (!strnicmp(argv
[0], MESG_STR("reinstate_path")))
1352 action
= reinstate_path
;
1353 else if (!strnicmp(argv
[0], MESG_STR("fail_path")))
1358 r
= dm_get_device(ti
, argv
[1], ti
->begin
, ti
->len
,
1359 dm_table_get_mode(ti
->table
), &dev
);
1361 DMWARN("message: error getting device %s",
1366 r
= action_dev(m
, dev
, action
);
1368 dm_put_device(ti
, dev
);
1373 DMWARN("Unrecognised multipath message received.");
1377 static int multipath_ioctl(struct dm_target
*ti
, struct inode
*inode
,
1378 struct file
*filp
, unsigned int cmd
,
1381 struct multipath
*m
= (struct multipath
*) ti
->private;
1382 struct block_device
*bdev
= NULL
;
1383 unsigned long flags
;
1384 struct file fake_file
= {};
1385 struct dentry fake_dentry
= {};
1388 fake_file
.f_path
.dentry
= &fake_dentry
;
1390 spin_lock_irqsave(&m
->lock
, flags
);
1392 if (!m
->current_pgpath
)
1395 if (m
->current_pgpath
) {
1396 bdev
= m
->current_pgpath
->path
.dev
->bdev
;
1397 fake_dentry
.d_inode
= bdev
->bd_inode
;
1398 fake_file
.f_mode
= m
->current_pgpath
->path
.dev
->mode
;
1406 spin_unlock_irqrestore(&m
->lock
, flags
);
1408 return r
? : blkdev_driver_ioctl(bdev
->bd_inode
, &fake_file
,
1409 bdev
->bd_disk
, cmd
, arg
);
1412 /*-----------------------------------------------------------------
1414 *---------------------------------------------------------------*/
1415 static struct target_type multipath_target
= {
1416 .name
= "multipath",
1417 .version
= {1, 0, 5},
1418 .module
= THIS_MODULE
,
1419 .ctr
= multipath_ctr
,
1420 .dtr
= multipath_dtr
,
1421 .map
= multipath_map
,
1422 .end_io
= multipath_end_io
,
1423 .presuspend
= multipath_presuspend
,
1424 .resume
= multipath_resume
,
1425 .status
= multipath_status
,
1426 .message
= multipath_message
,
1427 .ioctl
= multipath_ioctl
,
1430 static int __init
dm_multipath_init(void)
1434 /* allocate a slab for the dm_ios */
1435 _mpio_cache
= KMEM_CACHE(dm_mpath_io
, 0);
1439 r
= dm_register_target(&multipath_target
);
1441 DMERR("register failed %d", r
);
1442 kmem_cache_destroy(_mpio_cache
);
1446 kmultipathd
= create_workqueue("kmpathd");
1448 DMERR("failed to create workqueue kmpathd");
1449 dm_unregister_target(&multipath_target
);
1450 kmem_cache_destroy(_mpio_cache
);
1454 DMINFO("version %u.%u.%u loaded",
1455 multipath_target
.version
[0], multipath_target
.version
[1],
1456 multipath_target
.version
[2]);
1461 static void __exit
dm_multipath_exit(void)
1465 destroy_workqueue(kmultipathd
);
1467 r
= dm_unregister_target(&multipath_target
);
1469 DMERR("target unregister failed %d", r
);
1470 kmem_cache_destroy(_mpio_cache
);
1473 EXPORT_SYMBOL_GPL(dm_pg_init_complete
);
1475 module_init(dm_multipath_init
);
1476 module_exit(dm_multipath_exit
);
1478 MODULE_DESCRIPTION(DM_NAME
" multipath target");
1479 MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>");
1480 MODULE_LICENSE("GPL");