2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 Errors, Warnings, etc.
36 pr_crit() for error conditions that risk data loss
37 pr_err() for error conditions that are unexpected, like an IO error
38 or internal inconsistency
39 pr_warn() for error conditions that could have been predicated, like
40 adding a device to an array when it has incompatible metadata
41 pr_info() for every interesting, very rare events, like an array starting
42 or stopping, or resync starting or stopping
43 pr_debug() for everything else.
47 #include <linux/sched/signal.h>
48 #include <linux/kthread.h>
49 #include <linux/blkdev.h>
50 #include <linux/badblocks.h>
51 #include <linux/sysctl.h>
52 #include <linux/seq_file.h>
54 #include <linux/poll.h>
55 #include <linux/ctype.h>
56 #include <linux/string.h>
57 #include <linux/hdreg.h>
58 #include <linux/proc_fs.h>
59 #include <linux/random.h>
60 #include <linux/module.h>
61 #include <linux/reboot.h>
62 #include <linux/file.h>
63 #include <linux/compat.h>
64 #include <linux/delay.h>
65 #include <linux/raid/md_p.h>
66 #include <linux/raid/md_u.h>
67 #include <linux/slab.h>
68 #include <linux/percpu-refcount.h>
70 #include <trace/events/block.h>
72 #include "md-bitmap.h"
73 #include "md-cluster.h"
76 static void autostart_arrays(int part
);
79 /* pers_list is a list of registered personalities protected
81 * pers_lock does extra service to protect accesses to
82 * mddev->thread when the mutex cannot be held.
84 static LIST_HEAD(pers_list
);
85 static DEFINE_SPINLOCK(pers_lock
);
87 static struct kobj_type md_ktype
;
89 struct md_cluster_operations
*md_cluster_ops
;
90 EXPORT_SYMBOL(md_cluster_ops
);
91 struct module
*md_cluster_mod
;
92 EXPORT_SYMBOL(md_cluster_mod
);
94 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
95 static struct workqueue_struct
*md_wq
;
96 static struct workqueue_struct
*md_misc_wq
;
98 static int remove_and_add_spares(struct mddev
*mddev
,
99 struct md_rdev
*this);
100 static void mddev_detach(struct mddev
*mddev
);
103 * Default number of read corrections we'll attempt on an rdev
104 * before ejecting it from the array. We divide the read error
105 * count by 2 for every hour elapsed between read errors.
107 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
109 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
110 * is 1000 KB/sec, so the extra system load does not show up that much.
111 * Increase it if you want to have more _guaranteed_ speed. Note that
112 * the RAID driver will use the maximum available bandwidth if the IO
113 * subsystem is idle. There is also an 'absolute maximum' reconstruction
114 * speed limit - in case reconstruction slows down your system despite
117 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
118 * or /sys/block/mdX/md/sync_speed_{min,max}
121 static int sysctl_speed_limit_min
= 1000;
122 static int sysctl_speed_limit_max
= 200000;
123 static inline int speed_min(struct mddev
*mddev
)
125 return mddev
->sync_speed_min
?
126 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
129 static inline int speed_max(struct mddev
*mddev
)
131 return mddev
->sync_speed_max
?
132 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
135 static void * flush_info_alloc(gfp_t gfp_flags
, void *data
)
137 return kzalloc(sizeof(struct flush_info
), gfp_flags
);
139 static void flush_info_free(void *flush_info
, void *data
)
144 static void * flush_bio_alloc(gfp_t gfp_flags
, void *data
)
146 return kzalloc(sizeof(struct flush_bio
), gfp_flags
);
148 static void flush_bio_free(void *flush_bio
, void *data
)
153 static struct ctl_table_header
*raid_table_header
;
155 static struct ctl_table raid_table
[] = {
157 .procname
= "speed_limit_min",
158 .data
= &sysctl_speed_limit_min
,
159 .maxlen
= sizeof(int),
160 .mode
= S_IRUGO
|S_IWUSR
,
161 .proc_handler
= proc_dointvec
,
164 .procname
= "speed_limit_max",
165 .data
= &sysctl_speed_limit_max
,
166 .maxlen
= sizeof(int),
167 .mode
= S_IRUGO
|S_IWUSR
,
168 .proc_handler
= proc_dointvec
,
173 static struct ctl_table raid_dir_table
[] = {
177 .mode
= S_IRUGO
|S_IXUGO
,
183 static struct ctl_table raid_root_table
[] = {
188 .child
= raid_dir_table
,
193 static const struct block_device_operations md_fops
;
195 static int start_readonly
;
198 * The original mechanism for creating an md device is to create
199 * a device node in /dev and to open it. This causes races with device-close.
200 * The preferred method is to write to the "new_array" module parameter.
201 * This can avoid races.
202 * Setting create_on_open to false disables the original mechanism
203 * so all the races disappear.
205 static bool create_on_open
= true;
207 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
210 if (!mddev
|| !bioset_initialized(&mddev
->bio_set
))
211 return bio_alloc(gfp_mask
, nr_iovecs
);
213 return bio_alloc_bioset(gfp_mask
, nr_iovecs
, &mddev
->bio_set
);
215 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
217 static struct bio
*md_bio_alloc_sync(struct mddev
*mddev
)
219 if (!mddev
|| !bioset_initialized(&mddev
->sync_set
))
220 return bio_alloc(GFP_NOIO
, 1);
222 return bio_alloc_bioset(GFP_NOIO
, 1, &mddev
->sync_set
);
226 * We have a system wide 'event count' that is incremented
227 * on any 'interesting' event, and readers of /proc/mdstat
228 * can use 'poll' or 'select' to find out when the event
232 * start array, stop array, error, add device, remove device,
233 * start build, activate spare
235 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
236 static atomic_t md_event_count
;
237 void md_new_event(struct mddev
*mddev
)
239 atomic_inc(&md_event_count
);
240 wake_up(&md_event_waiters
);
242 EXPORT_SYMBOL_GPL(md_new_event
);
245 * Enables to iterate over all existing md arrays
246 * all_mddevs_lock protects this list.
248 static LIST_HEAD(all_mddevs
);
249 static DEFINE_SPINLOCK(all_mddevs_lock
);
252 * iterates through all used mddevs in the system.
253 * We take care to grab the all_mddevs_lock whenever navigating
254 * the list, and to always hold a refcount when unlocked.
255 * Any code which breaks out of this loop while own
256 * a reference to the current mddev and must mddev_put it.
258 #define for_each_mddev(_mddev,_tmp) \
260 for (({ spin_lock(&all_mddevs_lock); \
261 _tmp = all_mddevs.next; \
263 ({ if (_tmp != &all_mddevs) \
264 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
265 spin_unlock(&all_mddevs_lock); \
266 if (_mddev) mddev_put(_mddev); \
267 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
268 _tmp != &all_mddevs;}); \
269 ({ spin_lock(&all_mddevs_lock); \
270 _tmp = _tmp->next;}) \
273 /* Rather than calling directly into the personality make_request function,
274 * IO requests come here first so that we can check if the device is
275 * being suspended pending a reconfiguration.
276 * We hold a refcount over the call to ->make_request. By the time that
277 * call has finished, the bio has been linked into some internal structure
278 * and so is visible to ->quiesce(), so we don't need the refcount any more.
280 static bool is_suspended(struct mddev
*mddev
, struct bio
*bio
)
282 if (mddev
->suspended
)
284 if (bio_data_dir(bio
) != WRITE
)
286 if (mddev
->suspend_lo
>= mddev
->suspend_hi
)
288 if (bio
->bi_iter
.bi_sector
>= mddev
->suspend_hi
)
290 if (bio_end_sector(bio
) < mddev
->suspend_lo
)
295 void md_handle_request(struct mddev
*mddev
, struct bio
*bio
)
299 if (is_suspended(mddev
, bio
)) {
302 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
303 TASK_UNINTERRUPTIBLE
);
304 if (!is_suspended(mddev
, bio
))
310 finish_wait(&mddev
->sb_wait
, &__wait
);
312 atomic_inc(&mddev
->active_io
);
315 if (!mddev
->pers
->make_request(mddev
, bio
)) {
316 atomic_dec(&mddev
->active_io
);
317 wake_up(&mddev
->sb_wait
);
318 goto check_suspended
;
321 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
322 wake_up(&mddev
->sb_wait
);
324 EXPORT_SYMBOL(md_handle_request
);
326 static blk_qc_t
md_make_request(struct request_queue
*q
, struct bio
*bio
)
328 const int rw
= bio_data_dir(bio
);
329 const int sgrp
= op_stat_group(bio_op(bio
));
330 struct mddev
*mddev
= q
->queuedata
;
331 unsigned int sectors
;
333 blk_queue_split(q
, &bio
);
335 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
337 return BLK_QC_T_NONE
;
339 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
340 if (bio_sectors(bio
) != 0)
341 bio
->bi_status
= BLK_STS_IOERR
;
343 return BLK_QC_T_NONE
;
347 * save the sectors now since our bio can
348 * go away inside make_request
350 sectors
= bio_sectors(bio
);
351 /* bio could be mergeable after passing to underlayer */
352 bio
->bi_opf
&= ~REQ_NOMERGE
;
354 md_handle_request(mddev
, bio
);
357 part_stat_inc(&mddev
->gendisk
->part0
, ios
[sgrp
]);
358 part_stat_add(&mddev
->gendisk
->part0
, sectors
[sgrp
], sectors
);
361 return BLK_QC_T_NONE
;
364 /* mddev_suspend makes sure no new requests are submitted
365 * to the device, and that any requests that have been submitted
366 * are completely handled.
367 * Once mddev_detach() is called and completes, the module will be
370 void mddev_suspend(struct mddev
*mddev
)
372 WARN_ON_ONCE(mddev
->thread
&& current
== mddev
->thread
->tsk
);
373 lockdep_assert_held(&mddev
->reconfig_mutex
);
374 if (mddev
->suspended
++)
377 wake_up(&mddev
->sb_wait
);
378 set_bit(MD_ALLOW_SB_UPDATE
, &mddev
->flags
);
379 smp_mb__after_atomic();
380 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
381 mddev
->pers
->quiesce(mddev
, 1);
382 clear_bit_unlock(MD_ALLOW_SB_UPDATE
, &mddev
->flags
);
383 wait_event(mddev
->sb_wait
, !test_bit(MD_UPDATING_SB
, &mddev
->flags
));
385 del_timer_sync(&mddev
->safemode_timer
);
387 EXPORT_SYMBOL_GPL(mddev_suspend
);
389 void mddev_resume(struct mddev
*mddev
)
391 lockdep_assert_held(&mddev
->reconfig_mutex
);
392 if (--mddev
->suspended
)
394 wake_up(&mddev
->sb_wait
);
395 mddev
->pers
->quiesce(mddev
, 0);
397 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
398 md_wakeup_thread(mddev
->thread
);
399 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
401 EXPORT_SYMBOL_GPL(mddev_resume
);
403 int mddev_congested(struct mddev
*mddev
, int bits
)
405 struct md_personality
*pers
= mddev
->pers
;
409 if (mddev
->suspended
)
411 else if (pers
&& pers
->congested
)
412 ret
= pers
->congested(mddev
, bits
);
416 EXPORT_SYMBOL_GPL(mddev_congested
);
417 static int md_congested(void *data
, int bits
)
419 struct mddev
*mddev
= data
;
420 return mddev_congested(mddev
, bits
);
424 * Generic flush handling for md
426 static void submit_flushes(struct work_struct
*ws
)
428 struct flush_info
*fi
= container_of(ws
, struct flush_info
, flush_work
);
429 struct mddev
*mddev
= fi
->mddev
;
430 struct bio
*bio
= fi
->bio
;
432 bio
->bi_opf
&= ~REQ_PREFLUSH
;
433 md_handle_request(mddev
, bio
);
435 mempool_free(fi
, mddev
->flush_pool
);
438 static void md_end_flush(struct bio
*fbio
)
440 struct flush_bio
*fb
= fbio
->bi_private
;
441 struct md_rdev
*rdev
= fb
->rdev
;
442 struct flush_info
*fi
= fb
->fi
;
443 struct bio
*bio
= fi
->bio
;
444 struct mddev
*mddev
= fi
->mddev
;
446 rdev_dec_pending(rdev
, mddev
);
448 if (atomic_dec_and_test(&fi
->flush_pending
)) {
449 if (bio
->bi_iter
.bi_size
== 0) {
450 /* an empty barrier - all done */
452 mempool_free(fi
, mddev
->flush_pool
);
454 INIT_WORK(&fi
->flush_work
, submit_flushes
);
455 queue_work(md_wq
, &fi
->flush_work
);
459 mempool_free(fb
, mddev
->flush_bio_pool
);
463 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
465 struct md_rdev
*rdev
;
466 struct flush_info
*fi
;
468 fi
= mempool_alloc(mddev
->flush_pool
, GFP_NOIO
);
472 atomic_set(&fi
->flush_pending
, 1);
475 rdev_for_each_rcu(rdev
, mddev
)
476 if (rdev
->raid_disk
>= 0 &&
477 !test_bit(Faulty
, &rdev
->flags
)) {
478 /* Take two references, one is dropped
479 * when request finishes, one after
480 * we reclaim rcu_read_lock
483 struct flush_bio
*fb
;
484 atomic_inc(&rdev
->nr_pending
);
485 atomic_inc(&rdev
->nr_pending
);
488 fb
= mempool_alloc(mddev
->flush_bio_pool
, GFP_NOIO
);
492 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
493 bio_set_dev(bi
, rdev
->bdev
);
494 bi
->bi_end_io
= md_end_flush
;
496 bi
->bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
498 atomic_inc(&fi
->flush_pending
);
502 rdev_dec_pending(rdev
, mddev
);
506 if (atomic_dec_and_test(&fi
->flush_pending
)) {
507 if (bio
->bi_iter
.bi_size
== 0) {
508 /* an empty barrier - all done */
510 mempool_free(fi
, mddev
->flush_pool
);
512 INIT_WORK(&fi
->flush_work
, submit_flushes
);
513 queue_work(md_wq
, &fi
->flush_work
);
517 EXPORT_SYMBOL(md_flush_request
);
519 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
521 atomic_inc(&mddev
->active
);
525 static void mddev_delayed_delete(struct work_struct
*ws
);
527 static void mddev_put(struct mddev
*mddev
)
529 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
531 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
532 mddev
->ctime
== 0 && !mddev
->hold_active
) {
533 /* Array is not configured at all, and not held active,
535 list_del_init(&mddev
->all_mddevs
);
538 * Call queue_work inside the spinlock so that
539 * flush_workqueue() after mddev_find will succeed in waiting
540 * for the work to be done.
542 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
543 queue_work(md_misc_wq
, &mddev
->del_work
);
545 spin_unlock(&all_mddevs_lock
);
548 static void md_safemode_timeout(struct timer_list
*t
);
550 void mddev_init(struct mddev
*mddev
)
552 kobject_init(&mddev
->kobj
, &md_ktype
);
553 mutex_init(&mddev
->open_mutex
);
554 mutex_init(&mddev
->reconfig_mutex
);
555 mutex_init(&mddev
->bitmap_info
.mutex
);
556 INIT_LIST_HEAD(&mddev
->disks
);
557 INIT_LIST_HEAD(&mddev
->all_mddevs
);
558 timer_setup(&mddev
->safemode_timer
, md_safemode_timeout
, 0);
559 atomic_set(&mddev
->active
, 1);
560 atomic_set(&mddev
->openers
, 0);
561 atomic_set(&mddev
->active_io
, 0);
562 spin_lock_init(&mddev
->lock
);
563 init_waitqueue_head(&mddev
->sb_wait
);
564 init_waitqueue_head(&mddev
->recovery_wait
);
565 mddev
->reshape_position
= MaxSector
;
566 mddev
->reshape_backwards
= 0;
567 mddev
->last_sync_action
= "none";
568 mddev
->resync_min
= 0;
569 mddev
->resync_max
= MaxSector
;
570 mddev
->level
= LEVEL_NONE
;
572 EXPORT_SYMBOL_GPL(mddev_init
);
574 static struct mddev
*mddev_find(dev_t unit
)
576 struct mddev
*mddev
, *new = NULL
;
578 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
579 unit
&= ~((1<<MdpMinorShift
)-1);
582 spin_lock(&all_mddevs_lock
);
585 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
586 if (mddev
->unit
== unit
) {
588 spin_unlock(&all_mddevs_lock
);
594 list_add(&new->all_mddevs
, &all_mddevs
);
595 spin_unlock(&all_mddevs_lock
);
596 new->hold_active
= UNTIL_IOCTL
;
600 /* find an unused unit number */
601 static int next_minor
= 512;
602 int start
= next_minor
;
606 dev
= MKDEV(MD_MAJOR
, next_minor
);
608 if (next_minor
> MINORMASK
)
610 if (next_minor
== start
) {
611 /* Oh dear, all in use. */
612 spin_unlock(&all_mddevs_lock
);
618 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
619 if (mddev
->unit
== dev
) {
625 new->md_minor
= MINOR(dev
);
626 new->hold_active
= UNTIL_STOP
;
627 list_add(&new->all_mddevs
, &all_mddevs
);
628 spin_unlock(&all_mddevs_lock
);
631 spin_unlock(&all_mddevs_lock
);
633 new = kzalloc(sizeof(*new), GFP_KERNEL
);
638 if (MAJOR(unit
) == MD_MAJOR
)
639 new->md_minor
= MINOR(unit
);
641 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
648 static struct attribute_group md_redundancy_group
;
650 void mddev_unlock(struct mddev
*mddev
)
652 if (mddev
->to_remove
) {
653 /* These cannot be removed under reconfig_mutex as
654 * an access to the files will try to take reconfig_mutex
655 * while holding the file unremovable, which leads to
657 * So hold set sysfs_active while the remove in happeing,
658 * and anything else which might set ->to_remove or my
659 * otherwise change the sysfs namespace will fail with
660 * -EBUSY if sysfs_active is still set.
661 * We set sysfs_active under reconfig_mutex and elsewhere
662 * test it under the same mutex to ensure its correct value
665 struct attribute_group
*to_remove
= mddev
->to_remove
;
666 mddev
->to_remove
= NULL
;
667 mddev
->sysfs_active
= 1;
668 mutex_unlock(&mddev
->reconfig_mutex
);
670 if (mddev
->kobj
.sd
) {
671 if (to_remove
!= &md_redundancy_group
)
672 sysfs_remove_group(&mddev
->kobj
, to_remove
);
673 if (mddev
->pers
== NULL
||
674 mddev
->pers
->sync_request
== NULL
) {
675 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
676 if (mddev
->sysfs_action
)
677 sysfs_put(mddev
->sysfs_action
);
678 mddev
->sysfs_action
= NULL
;
681 mddev
->sysfs_active
= 0;
683 mutex_unlock(&mddev
->reconfig_mutex
);
685 /* As we've dropped the mutex we need a spinlock to
686 * make sure the thread doesn't disappear
688 spin_lock(&pers_lock
);
689 md_wakeup_thread(mddev
->thread
);
690 wake_up(&mddev
->sb_wait
);
691 spin_unlock(&pers_lock
);
693 EXPORT_SYMBOL_GPL(mddev_unlock
);
695 struct md_rdev
*md_find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
697 struct md_rdev
*rdev
;
699 rdev_for_each_rcu(rdev
, mddev
)
700 if (rdev
->desc_nr
== nr
)
705 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu
);
707 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
709 struct md_rdev
*rdev
;
711 rdev_for_each(rdev
, mddev
)
712 if (rdev
->bdev
->bd_dev
== dev
)
718 struct md_rdev
*md_find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
720 struct md_rdev
*rdev
;
722 rdev_for_each_rcu(rdev
, mddev
)
723 if (rdev
->bdev
->bd_dev
== dev
)
728 EXPORT_SYMBOL_GPL(md_find_rdev_rcu
);
730 static struct md_personality
*find_pers(int level
, char *clevel
)
732 struct md_personality
*pers
;
733 list_for_each_entry(pers
, &pers_list
, list
) {
734 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
736 if (strcmp(pers
->name
, clevel
)==0)
742 /* return the offset of the super block in 512byte sectors */
743 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
745 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
746 return MD_NEW_SIZE_SECTORS(num_sectors
);
749 static int alloc_disk_sb(struct md_rdev
*rdev
)
751 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
757 void md_rdev_clear(struct md_rdev
*rdev
)
760 put_page(rdev
->sb_page
);
762 rdev
->sb_page
= NULL
;
767 put_page(rdev
->bb_page
);
768 rdev
->bb_page
= NULL
;
770 badblocks_exit(&rdev
->badblocks
);
772 EXPORT_SYMBOL_GPL(md_rdev_clear
);
774 static void super_written(struct bio
*bio
)
776 struct md_rdev
*rdev
= bio
->bi_private
;
777 struct mddev
*mddev
= rdev
->mddev
;
779 if (bio
->bi_status
) {
780 pr_err("md: super_written gets error=%d\n", bio
->bi_status
);
781 md_error(mddev
, rdev
);
782 if (!test_bit(Faulty
, &rdev
->flags
)
783 && (bio
->bi_opf
& MD_FAILFAST
)) {
784 set_bit(MD_SB_NEED_REWRITE
, &mddev
->sb_flags
);
785 set_bit(LastDev
, &rdev
->flags
);
788 clear_bit(LastDev
, &rdev
->flags
);
790 if (atomic_dec_and_test(&mddev
->pending_writes
))
791 wake_up(&mddev
->sb_wait
);
792 rdev_dec_pending(rdev
, mddev
);
796 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
797 sector_t sector
, int size
, struct page
*page
)
799 /* write first size bytes of page to sector of rdev
800 * Increment mddev->pending_writes before returning
801 * and decrement it on completion, waking up sb_wait
802 * if zero is reached.
803 * If an error occurred, call md_error
811 if (test_bit(Faulty
, &rdev
->flags
))
814 bio
= md_bio_alloc_sync(mddev
);
816 atomic_inc(&rdev
->nr_pending
);
818 bio_set_dev(bio
, rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
);
819 bio
->bi_iter
.bi_sector
= sector
;
820 bio_add_page(bio
, page
, size
, 0);
821 bio
->bi_private
= rdev
;
822 bio
->bi_end_io
= super_written
;
824 if (test_bit(MD_FAILFAST_SUPPORTED
, &mddev
->flags
) &&
825 test_bit(FailFast
, &rdev
->flags
) &&
826 !test_bit(LastDev
, &rdev
->flags
))
828 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
| ff
;
830 atomic_inc(&mddev
->pending_writes
);
834 int md_super_wait(struct mddev
*mddev
)
836 /* wait for all superblock writes that were scheduled to complete */
837 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
838 if (test_and_clear_bit(MD_SB_NEED_REWRITE
, &mddev
->sb_flags
))
843 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
844 struct page
*page
, int op
, int op_flags
, bool metadata_op
)
846 struct bio
*bio
= md_bio_alloc_sync(rdev
->mddev
);
849 if (metadata_op
&& rdev
->meta_bdev
)
850 bio_set_dev(bio
, rdev
->meta_bdev
);
852 bio_set_dev(bio
, rdev
->bdev
);
853 bio_set_op_attrs(bio
, op
, op_flags
);
855 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
856 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
857 (rdev
->mddev
->reshape_backwards
==
858 (sector
>= rdev
->mddev
->reshape_position
)))
859 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
861 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
862 bio_add_page(bio
, page
, size
, 0);
864 submit_bio_wait(bio
);
866 ret
= !bio
->bi_status
;
870 EXPORT_SYMBOL_GPL(sync_page_io
);
872 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
874 char b
[BDEVNAME_SIZE
];
879 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, REQ_OP_READ
, 0, true))
885 pr_err("md: disabled device %s, could not read superblock.\n",
886 bdevname(rdev
->bdev
,b
));
890 static int md_uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
892 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
893 sb1
->set_uuid1
== sb2
->set_uuid1
&&
894 sb1
->set_uuid2
== sb2
->set_uuid2
&&
895 sb1
->set_uuid3
== sb2
->set_uuid3
;
898 static int md_sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
901 mdp_super_t
*tmp1
, *tmp2
;
903 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
904 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
906 if (!tmp1
|| !tmp2
) {
915 * nr_disks is not constant
920 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
927 static u32
md_csum_fold(u32 csum
)
929 csum
= (csum
& 0xffff) + (csum
>> 16);
930 return (csum
& 0xffff) + (csum
>> 16);
933 static unsigned int calc_sb_csum(mdp_super_t
*sb
)
936 u32
*sb32
= (u32
*)sb
;
938 unsigned int disk_csum
, csum
;
940 disk_csum
= sb
->sb_csum
;
943 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
945 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
948 /* This used to use csum_partial, which was wrong for several
949 * reasons including that different results are returned on
950 * different architectures. It isn't critical that we get exactly
951 * the same return value as before (we always csum_fold before
952 * testing, and that removes any differences). However as we
953 * know that csum_partial always returned a 16bit value on
954 * alphas, do a fold to maximise conformity to previous behaviour.
956 sb
->sb_csum
= md_csum_fold(disk_csum
);
958 sb
->sb_csum
= disk_csum
;
964 * Handle superblock details.
965 * We want to be able to handle multiple superblock formats
966 * so we have a common interface to them all, and an array of
967 * different handlers.
968 * We rely on user-space to write the initial superblock, and support
969 * reading and updating of superblocks.
970 * Interface methods are:
971 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
972 * loads and validates a superblock on dev.
973 * if refdev != NULL, compare superblocks on both devices
975 * 0 - dev has a superblock that is compatible with refdev
976 * 1 - dev has a superblock that is compatible and newer than refdev
977 * so dev should be used as the refdev in future
978 * -EINVAL superblock incompatible or invalid
979 * -othererror e.g. -EIO
981 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
982 * Verify that dev is acceptable into mddev.
983 * The first time, mddev->raid_disks will be 0, and data from
984 * dev should be merged in. Subsequent calls check that dev
985 * is new enough. Return 0 or -EINVAL
987 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
988 * Update the superblock for rdev with data in mddev
989 * This does not write to disc.
995 struct module
*owner
;
996 int (*load_super
)(struct md_rdev
*rdev
,
997 struct md_rdev
*refdev
,
999 int (*validate_super
)(struct mddev
*mddev
,
1000 struct md_rdev
*rdev
);
1001 void (*sync_super
)(struct mddev
*mddev
,
1002 struct md_rdev
*rdev
);
1003 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1004 sector_t num_sectors
);
1005 int (*allow_new_offset
)(struct md_rdev
*rdev
,
1006 unsigned long long new_offset
);
1010 * Check that the given mddev has no bitmap.
1012 * This function is called from the run method of all personalities that do not
1013 * support bitmaps. It prints an error message and returns non-zero if mddev
1014 * has a bitmap. Otherwise, it returns 0.
1017 int md_check_no_bitmap(struct mddev
*mddev
)
1019 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1021 pr_warn("%s: bitmaps are not supported for %s\n",
1022 mdname(mddev
), mddev
->pers
->name
);
1025 EXPORT_SYMBOL(md_check_no_bitmap
);
1028 * load_super for 0.90.0
1030 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1032 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1037 * Calculate the position of the superblock (512byte sectors),
1038 * it's at the end of the disk.
1040 * It also happens to be a multiple of 4Kb.
1042 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1044 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1050 bdevname(rdev
->bdev
, b
);
1051 sb
= page_address(rdev
->sb_page
);
1053 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1054 pr_warn("md: invalid raid superblock magic on %s\n", b
);
1058 if (sb
->major_version
!= 0 ||
1059 sb
->minor_version
< 90 ||
1060 sb
->minor_version
> 91) {
1061 pr_warn("Bad version number %d.%d on %s\n",
1062 sb
->major_version
, sb
->minor_version
, b
);
1066 if (sb
->raid_disks
<= 0)
1069 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1070 pr_warn("md: invalid superblock checksum on %s\n", b
);
1074 rdev
->preferred_minor
= sb
->md_minor
;
1075 rdev
->data_offset
= 0;
1076 rdev
->new_data_offset
= 0;
1077 rdev
->sb_size
= MD_SB_BYTES
;
1078 rdev
->badblocks
.shift
= -1;
1080 if (sb
->level
== LEVEL_MULTIPATH
)
1083 rdev
->desc_nr
= sb
->this_disk
.number
;
1089 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1090 if (!md_uuid_equal(refsb
, sb
)) {
1091 pr_warn("md: %s has different UUID to %s\n",
1092 b
, bdevname(refdev
->bdev
,b2
));
1095 if (!md_sb_equal(refsb
, sb
)) {
1096 pr_warn("md: %s has same UUID but different superblock to %s\n",
1097 b
, bdevname(refdev
->bdev
, b2
));
1101 ev2
= md_event(refsb
);
1107 rdev
->sectors
= rdev
->sb_start
;
1108 /* Limit to 4TB as metadata cannot record more than that.
1109 * (not needed for Linear and RAID0 as metadata doesn't
1112 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)rdev
->sectors
>= (2ULL << 32) &&
1114 rdev
->sectors
= (sector_t
)(2ULL << 32) - 2;
1116 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1117 /* "this cannot possibly happen" ... */
1125 * validate_super for 0.90.0
1127 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1130 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1131 __u64 ev1
= md_event(sb
);
1133 rdev
->raid_disk
= -1;
1134 clear_bit(Faulty
, &rdev
->flags
);
1135 clear_bit(In_sync
, &rdev
->flags
);
1136 clear_bit(Bitmap_sync
, &rdev
->flags
);
1137 clear_bit(WriteMostly
, &rdev
->flags
);
1139 if (mddev
->raid_disks
== 0) {
1140 mddev
->major_version
= 0;
1141 mddev
->minor_version
= sb
->minor_version
;
1142 mddev
->patch_version
= sb
->patch_version
;
1143 mddev
->external
= 0;
1144 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1145 mddev
->ctime
= sb
->ctime
;
1146 mddev
->utime
= sb
->utime
;
1147 mddev
->level
= sb
->level
;
1148 mddev
->clevel
[0] = 0;
1149 mddev
->layout
= sb
->layout
;
1150 mddev
->raid_disks
= sb
->raid_disks
;
1151 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1152 mddev
->events
= ev1
;
1153 mddev
->bitmap_info
.offset
= 0;
1154 mddev
->bitmap_info
.space
= 0;
1155 /* bitmap can use 60 K after the 4K superblocks */
1156 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1157 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1158 mddev
->reshape_backwards
= 0;
1160 if (mddev
->minor_version
>= 91) {
1161 mddev
->reshape_position
= sb
->reshape_position
;
1162 mddev
->delta_disks
= sb
->delta_disks
;
1163 mddev
->new_level
= sb
->new_level
;
1164 mddev
->new_layout
= sb
->new_layout
;
1165 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1166 if (mddev
->delta_disks
< 0)
1167 mddev
->reshape_backwards
= 1;
1169 mddev
->reshape_position
= MaxSector
;
1170 mddev
->delta_disks
= 0;
1171 mddev
->new_level
= mddev
->level
;
1172 mddev
->new_layout
= mddev
->layout
;
1173 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1176 if (sb
->state
& (1<<MD_SB_CLEAN
))
1177 mddev
->recovery_cp
= MaxSector
;
1179 if (sb
->events_hi
== sb
->cp_events_hi
&&
1180 sb
->events_lo
== sb
->cp_events_lo
) {
1181 mddev
->recovery_cp
= sb
->recovery_cp
;
1183 mddev
->recovery_cp
= 0;
1186 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1187 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1188 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1189 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1191 mddev
->max_disks
= MD_SB_DISKS
;
1193 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1194 mddev
->bitmap_info
.file
== NULL
) {
1195 mddev
->bitmap_info
.offset
=
1196 mddev
->bitmap_info
.default_offset
;
1197 mddev
->bitmap_info
.space
=
1198 mddev
->bitmap_info
.default_space
;
1201 } else if (mddev
->pers
== NULL
) {
1202 /* Insist on good event counter while assembling, except
1203 * for spares (which don't need an event count) */
1205 if (sb
->disks
[rdev
->desc_nr
].state
& (
1206 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1207 if (ev1
< mddev
->events
)
1209 } else if (mddev
->bitmap
) {
1210 /* if adding to array with a bitmap, then we can accept an
1211 * older device ... but not too old.
1213 if (ev1
< mddev
->bitmap
->events_cleared
)
1215 if (ev1
< mddev
->events
)
1216 set_bit(Bitmap_sync
, &rdev
->flags
);
1218 if (ev1
< mddev
->events
)
1219 /* just a hot-add of a new device, leave raid_disk at -1 */
1223 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1224 desc
= sb
->disks
+ rdev
->desc_nr
;
1226 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1227 set_bit(Faulty
, &rdev
->flags
);
1228 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1229 desc->raid_disk < mddev->raid_disks */) {
1230 set_bit(In_sync
, &rdev
->flags
);
1231 rdev
->raid_disk
= desc
->raid_disk
;
1232 rdev
->saved_raid_disk
= desc
->raid_disk
;
1233 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1234 /* active but not in sync implies recovery up to
1235 * reshape position. We don't know exactly where
1236 * that is, so set to zero for now */
1237 if (mddev
->minor_version
>= 91) {
1238 rdev
->recovery_offset
= 0;
1239 rdev
->raid_disk
= desc
->raid_disk
;
1242 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1243 set_bit(WriteMostly
, &rdev
->flags
);
1244 if (desc
->state
& (1<<MD_DISK_FAILFAST
))
1245 set_bit(FailFast
, &rdev
->flags
);
1246 } else /* MULTIPATH are always insync */
1247 set_bit(In_sync
, &rdev
->flags
);
1252 * sync_super for 0.90.0
1254 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1257 struct md_rdev
*rdev2
;
1258 int next_spare
= mddev
->raid_disks
;
1260 /* make rdev->sb match mddev data..
1263 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1264 * 3/ any empty disks < next_spare become removed
1266 * disks[0] gets initialised to REMOVED because
1267 * we cannot be sure from other fields if it has
1268 * been initialised or not.
1271 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1273 rdev
->sb_size
= MD_SB_BYTES
;
1275 sb
= page_address(rdev
->sb_page
);
1277 memset(sb
, 0, sizeof(*sb
));
1279 sb
->md_magic
= MD_SB_MAGIC
;
1280 sb
->major_version
= mddev
->major_version
;
1281 sb
->patch_version
= mddev
->patch_version
;
1282 sb
->gvalid_words
= 0; /* ignored */
1283 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1284 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1285 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1286 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1288 sb
->ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
1289 sb
->level
= mddev
->level
;
1290 sb
->size
= mddev
->dev_sectors
/ 2;
1291 sb
->raid_disks
= mddev
->raid_disks
;
1292 sb
->md_minor
= mddev
->md_minor
;
1293 sb
->not_persistent
= 0;
1294 sb
->utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
1296 sb
->events_hi
= (mddev
->events
>>32);
1297 sb
->events_lo
= (u32
)mddev
->events
;
1299 if (mddev
->reshape_position
== MaxSector
)
1300 sb
->minor_version
= 90;
1302 sb
->minor_version
= 91;
1303 sb
->reshape_position
= mddev
->reshape_position
;
1304 sb
->new_level
= mddev
->new_level
;
1305 sb
->delta_disks
= mddev
->delta_disks
;
1306 sb
->new_layout
= mddev
->new_layout
;
1307 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1309 mddev
->minor_version
= sb
->minor_version
;
1312 sb
->recovery_cp
= mddev
->recovery_cp
;
1313 sb
->cp_events_hi
= (mddev
->events
>>32);
1314 sb
->cp_events_lo
= (u32
)mddev
->events
;
1315 if (mddev
->recovery_cp
== MaxSector
)
1316 sb
->state
= (1<< MD_SB_CLEAN
);
1318 sb
->recovery_cp
= 0;
1320 sb
->layout
= mddev
->layout
;
1321 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1323 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1324 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1326 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1327 rdev_for_each(rdev2
, mddev
) {
1330 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1332 if (rdev2
->raid_disk
>= 0 &&
1333 sb
->minor_version
>= 91)
1334 /* we have nowhere to store the recovery_offset,
1335 * but if it is not below the reshape_position,
1336 * we can piggy-back on that.
1339 if (rdev2
->raid_disk
< 0 ||
1340 test_bit(Faulty
, &rdev2
->flags
))
1343 desc_nr
= rdev2
->raid_disk
;
1345 desc_nr
= next_spare
++;
1346 rdev2
->desc_nr
= desc_nr
;
1347 d
= &sb
->disks
[rdev2
->desc_nr
];
1349 d
->number
= rdev2
->desc_nr
;
1350 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1351 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1353 d
->raid_disk
= rdev2
->raid_disk
;
1355 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1356 if (test_bit(Faulty
, &rdev2
->flags
))
1357 d
->state
= (1<<MD_DISK_FAULTY
);
1358 else if (is_active
) {
1359 d
->state
= (1<<MD_DISK_ACTIVE
);
1360 if (test_bit(In_sync
, &rdev2
->flags
))
1361 d
->state
|= (1<<MD_DISK_SYNC
);
1369 if (test_bit(WriteMostly
, &rdev2
->flags
))
1370 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1371 if (test_bit(FailFast
, &rdev2
->flags
))
1372 d
->state
|= (1<<MD_DISK_FAILFAST
);
1374 /* now set the "removed" and "faulty" bits on any missing devices */
1375 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1376 mdp_disk_t
*d
= &sb
->disks
[i
];
1377 if (d
->state
== 0 && d
->number
== 0) {
1380 d
->state
= (1<<MD_DISK_REMOVED
);
1381 d
->state
|= (1<<MD_DISK_FAULTY
);
1385 sb
->nr_disks
= nr_disks
;
1386 sb
->active_disks
= active
;
1387 sb
->working_disks
= working
;
1388 sb
->failed_disks
= failed
;
1389 sb
->spare_disks
= spare
;
1391 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1392 sb
->sb_csum
= calc_sb_csum(sb
);
1396 * rdev_size_change for 0.90.0
1398 static unsigned long long
1399 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1401 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1402 return 0; /* component must fit device */
1403 if (rdev
->mddev
->bitmap_info
.offset
)
1404 return 0; /* can't move bitmap */
1405 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1406 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1407 num_sectors
= rdev
->sb_start
;
1408 /* Limit to 4TB as metadata cannot record more than that.
1409 * 4TB == 2^32 KB, or 2*2^32 sectors.
1411 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)num_sectors
>= (2ULL << 32) &&
1412 rdev
->mddev
->level
>= 1)
1413 num_sectors
= (sector_t
)(2ULL << 32) - 2;
1415 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1417 } while (md_super_wait(rdev
->mddev
) < 0);
1422 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1424 /* non-zero offset changes not possible with v0.90 */
1425 return new_offset
== 0;
1429 * version 1 superblock
1432 static __le32
calc_sb_1_csum(struct mdp_superblock_1
*sb
)
1436 unsigned long long newcsum
;
1437 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1438 __le32
*isuper
= (__le32
*)sb
;
1440 disk_csum
= sb
->sb_csum
;
1443 for (; size
>= 4; size
-= 4)
1444 newcsum
+= le32_to_cpu(*isuper
++);
1447 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1449 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1450 sb
->sb_csum
= disk_csum
;
1451 return cpu_to_le32(csum
);
1454 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1456 struct mdp_superblock_1
*sb
;
1460 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1464 * Calculate the position of the superblock in 512byte sectors.
1465 * It is always aligned to a 4K boundary and
1466 * depeding on minor_version, it can be:
1467 * 0: At least 8K, but less than 12K, from end of device
1468 * 1: At start of device
1469 * 2: 4K from start of device.
1471 switch(minor_version
) {
1473 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1475 sb_start
&= ~(sector_t
)(4*2-1);
1486 rdev
->sb_start
= sb_start
;
1488 /* superblock is rarely larger than 1K, but it can be larger,
1489 * and it is safe to read 4k, so we do that
1491 ret
= read_disk_sb(rdev
, 4096);
1492 if (ret
) return ret
;
1494 sb
= page_address(rdev
->sb_page
);
1496 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1497 sb
->major_version
!= cpu_to_le32(1) ||
1498 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1499 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1500 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1503 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1504 pr_warn("md: invalid superblock checksum on %s\n",
1505 bdevname(rdev
->bdev
,b
));
1508 if (le64_to_cpu(sb
->data_size
) < 10) {
1509 pr_warn("md: data_size too small on %s\n",
1510 bdevname(rdev
->bdev
,b
));
1515 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1516 /* Some padding is non-zero, might be a new feature */
1519 rdev
->preferred_minor
= 0xffff;
1520 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1521 rdev
->new_data_offset
= rdev
->data_offset
;
1522 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1523 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1524 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1525 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1527 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1528 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1529 if (rdev
->sb_size
& bmask
)
1530 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1533 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1536 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1539 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1542 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1544 if (!rdev
->bb_page
) {
1545 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1549 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1550 rdev
->badblocks
.count
== 0) {
1551 /* need to load the bad block list.
1552 * Currently we limit it to one page.
1558 int sectors
= le16_to_cpu(sb
->bblog_size
);
1559 if (sectors
> (PAGE_SIZE
/ 512))
1561 offset
= le32_to_cpu(sb
->bblog_offset
);
1564 bb_sector
= (long long)offset
;
1565 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1566 rdev
->bb_page
, REQ_OP_READ
, 0, true))
1568 bbp
= (u64
*)page_address(rdev
->bb_page
);
1569 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1570 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1571 u64 bb
= le64_to_cpu(*bbp
);
1572 int count
= bb
& (0x3ff);
1573 u64 sector
= bb
>> 10;
1574 sector
<<= sb
->bblog_shift
;
1575 count
<<= sb
->bblog_shift
;
1578 if (badblocks_set(&rdev
->badblocks
, sector
, count
, 1))
1581 } else if (sb
->bblog_offset
!= 0)
1582 rdev
->badblocks
.shift
= 0;
1584 if ((le32_to_cpu(sb
->feature_map
) &
1585 (MD_FEATURE_PPL
| MD_FEATURE_MULTIPLE_PPLS
))) {
1586 rdev
->ppl
.offset
= (__s16
)le16_to_cpu(sb
->ppl
.offset
);
1587 rdev
->ppl
.size
= le16_to_cpu(sb
->ppl
.size
);
1588 rdev
->ppl
.sector
= rdev
->sb_start
+ rdev
->ppl
.offset
;
1595 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1597 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1598 sb
->level
!= refsb
->level
||
1599 sb
->layout
!= refsb
->layout
||
1600 sb
->chunksize
!= refsb
->chunksize
) {
1601 pr_warn("md: %s has strangely different superblock to %s\n",
1602 bdevname(rdev
->bdev
,b
),
1603 bdevname(refdev
->bdev
,b2
));
1606 ev1
= le64_to_cpu(sb
->events
);
1607 ev2
= le64_to_cpu(refsb
->events
);
1614 if (minor_version
) {
1615 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1616 sectors
-= rdev
->data_offset
;
1618 sectors
= rdev
->sb_start
;
1619 if (sectors
< le64_to_cpu(sb
->data_size
))
1621 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1625 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1627 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1628 __u64 ev1
= le64_to_cpu(sb
->events
);
1630 rdev
->raid_disk
= -1;
1631 clear_bit(Faulty
, &rdev
->flags
);
1632 clear_bit(In_sync
, &rdev
->flags
);
1633 clear_bit(Bitmap_sync
, &rdev
->flags
);
1634 clear_bit(WriteMostly
, &rdev
->flags
);
1636 if (mddev
->raid_disks
== 0) {
1637 mddev
->major_version
= 1;
1638 mddev
->patch_version
= 0;
1639 mddev
->external
= 0;
1640 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1641 mddev
->ctime
= le64_to_cpu(sb
->ctime
);
1642 mddev
->utime
= le64_to_cpu(sb
->utime
);
1643 mddev
->level
= le32_to_cpu(sb
->level
);
1644 mddev
->clevel
[0] = 0;
1645 mddev
->layout
= le32_to_cpu(sb
->layout
);
1646 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1647 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1648 mddev
->events
= ev1
;
1649 mddev
->bitmap_info
.offset
= 0;
1650 mddev
->bitmap_info
.space
= 0;
1651 /* Default location for bitmap is 1K after superblock
1652 * using 3K - total of 4K
1654 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1655 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1656 mddev
->reshape_backwards
= 0;
1658 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1659 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1661 mddev
->max_disks
= (4096-256)/2;
1663 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1664 mddev
->bitmap_info
.file
== NULL
) {
1665 mddev
->bitmap_info
.offset
=
1666 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1667 /* Metadata doesn't record how much space is available.
1668 * For 1.0, we assume we can use up to the superblock
1669 * if before, else to 4K beyond superblock.
1670 * For others, assume no change is possible.
1672 if (mddev
->minor_version
> 0)
1673 mddev
->bitmap_info
.space
= 0;
1674 else if (mddev
->bitmap_info
.offset
> 0)
1675 mddev
->bitmap_info
.space
=
1676 8 - mddev
->bitmap_info
.offset
;
1678 mddev
->bitmap_info
.space
=
1679 -mddev
->bitmap_info
.offset
;
1682 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1683 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1684 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1685 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1686 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1687 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1688 if (mddev
->delta_disks
< 0 ||
1689 (mddev
->delta_disks
== 0 &&
1690 (le32_to_cpu(sb
->feature_map
)
1691 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1692 mddev
->reshape_backwards
= 1;
1694 mddev
->reshape_position
= MaxSector
;
1695 mddev
->delta_disks
= 0;
1696 mddev
->new_level
= mddev
->level
;
1697 mddev
->new_layout
= mddev
->layout
;
1698 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1701 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
)
1702 set_bit(MD_HAS_JOURNAL
, &mddev
->flags
);
1704 if (le32_to_cpu(sb
->feature_map
) &
1705 (MD_FEATURE_PPL
| MD_FEATURE_MULTIPLE_PPLS
)) {
1706 if (le32_to_cpu(sb
->feature_map
) &
1707 (MD_FEATURE_BITMAP_OFFSET
| MD_FEATURE_JOURNAL
))
1709 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_PPL
) &&
1710 (le32_to_cpu(sb
->feature_map
) &
1711 MD_FEATURE_MULTIPLE_PPLS
))
1713 set_bit(MD_HAS_PPL
, &mddev
->flags
);
1715 } else if (mddev
->pers
== NULL
) {
1716 /* Insist of good event counter while assembling, except for
1717 * spares (which don't need an event count) */
1719 if (rdev
->desc_nr
>= 0 &&
1720 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1721 (le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < MD_DISK_ROLE_MAX
||
1722 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) == MD_DISK_ROLE_JOURNAL
))
1723 if (ev1
< mddev
->events
)
1725 } else if (mddev
->bitmap
) {
1726 /* If adding to array with a bitmap, then we can accept an
1727 * older device, but not too old.
1729 if (ev1
< mddev
->bitmap
->events_cleared
)
1731 if (ev1
< mddev
->events
)
1732 set_bit(Bitmap_sync
, &rdev
->flags
);
1734 if (ev1
< mddev
->events
)
1735 /* just a hot-add of a new device, leave raid_disk at -1 */
1738 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1740 if (rdev
->desc_nr
< 0 ||
1741 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1742 role
= MD_DISK_ROLE_SPARE
;
1745 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1747 case MD_DISK_ROLE_SPARE
: /* spare */
1749 case MD_DISK_ROLE_FAULTY
: /* faulty */
1750 set_bit(Faulty
, &rdev
->flags
);
1752 case MD_DISK_ROLE_JOURNAL
: /* journal device */
1753 if (!(le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
)) {
1754 /* journal device without journal feature */
1755 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1758 set_bit(Journal
, &rdev
->flags
);
1759 rdev
->journal_tail
= le64_to_cpu(sb
->journal_tail
);
1760 rdev
->raid_disk
= 0;
1763 rdev
->saved_raid_disk
= role
;
1764 if ((le32_to_cpu(sb
->feature_map
) &
1765 MD_FEATURE_RECOVERY_OFFSET
)) {
1766 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1767 if (!(le32_to_cpu(sb
->feature_map
) &
1768 MD_FEATURE_RECOVERY_BITMAP
))
1769 rdev
->saved_raid_disk
= -1;
1771 set_bit(In_sync
, &rdev
->flags
);
1772 rdev
->raid_disk
= role
;
1775 if (sb
->devflags
& WriteMostly1
)
1776 set_bit(WriteMostly
, &rdev
->flags
);
1777 if (sb
->devflags
& FailFast1
)
1778 set_bit(FailFast
, &rdev
->flags
);
1779 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1780 set_bit(Replacement
, &rdev
->flags
);
1781 } else /* MULTIPATH are always insync */
1782 set_bit(In_sync
, &rdev
->flags
);
1787 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1789 struct mdp_superblock_1
*sb
;
1790 struct md_rdev
*rdev2
;
1792 /* make rdev->sb match mddev and rdev data. */
1794 sb
= page_address(rdev
->sb_page
);
1796 sb
->feature_map
= 0;
1798 sb
->recovery_offset
= cpu_to_le64(0);
1799 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1801 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1802 sb
->events
= cpu_to_le64(mddev
->events
);
1804 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1805 else if (test_bit(MD_JOURNAL_CLEAN
, &mddev
->flags
))
1806 sb
->resync_offset
= cpu_to_le64(MaxSector
);
1808 sb
->resync_offset
= cpu_to_le64(0);
1810 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1812 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1813 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1814 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1815 sb
->level
= cpu_to_le32(mddev
->level
);
1816 sb
->layout
= cpu_to_le32(mddev
->layout
);
1817 if (test_bit(FailFast
, &rdev
->flags
))
1818 sb
->devflags
|= FailFast1
;
1820 sb
->devflags
&= ~FailFast1
;
1822 if (test_bit(WriteMostly
, &rdev
->flags
))
1823 sb
->devflags
|= WriteMostly1
;
1825 sb
->devflags
&= ~WriteMostly1
;
1826 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1827 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1829 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1830 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1831 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1834 if (rdev
->raid_disk
>= 0 && !test_bit(Journal
, &rdev
->flags
) &&
1835 !test_bit(In_sync
, &rdev
->flags
)) {
1837 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1838 sb
->recovery_offset
=
1839 cpu_to_le64(rdev
->recovery_offset
);
1840 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1842 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1844 /* Note: recovery_offset and journal_tail share space */
1845 if (test_bit(Journal
, &rdev
->flags
))
1846 sb
->journal_tail
= cpu_to_le64(rdev
->journal_tail
);
1847 if (test_bit(Replacement
, &rdev
->flags
))
1849 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1851 if (mddev
->reshape_position
!= MaxSector
) {
1852 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1853 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1854 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1855 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1856 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1857 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1858 if (mddev
->delta_disks
== 0 &&
1859 mddev
->reshape_backwards
)
1861 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1862 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1864 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1865 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1866 - rdev
->data_offset
));
1870 if (mddev_is_clustered(mddev
))
1871 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_CLUSTERED
);
1873 if (rdev
->badblocks
.count
== 0)
1874 /* Nothing to do for bad blocks*/ ;
1875 else if (sb
->bblog_offset
== 0)
1876 /* Cannot record bad blocks on this device */
1877 md_error(mddev
, rdev
);
1879 struct badblocks
*bb
= &rdev
->badblocks
;
1880 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1882 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1887 seq
= read_seqbegin(&bb
->lock
);
1889 memset(bbp
, 0xff, PAGE_SIZE
);
1891 for (i
= 0 ; i
< bb
->count
; i
++) {
1892 u64 internal_bb
= p
[i
];
1893 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1894 | BB_LEN(internal_bb
));
1895 bbp
[i
] = cpu_to_le64(store_bb
);
1898 if (read_seqretry(&bb
->lock
, seq
))
1901 bb
->sector
= (rdev
->sb_start
+
1902 (int)le32_to_cpu(sb
->bblog_offset
));
1903 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1908 rdev_for_each(rdev2
, mddev
)
1909 if (rdev2
->desc_nr
+1 > max_dev
)
1910 max_dev
= rdev2
->desc_nr
+1;
1912 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1914 sb
->max_dev
= cpu_to_le32(max_dev
);
1915 rdev
->sb_size
= max_dev
* 2 + 256;
1916 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1917 if (rdev
->sb_size
& bmask
)
1918 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1920 max_dev
= le32_to_cpu(sb
->max_dev
);
1922 for (i
=0; i
<max_dev
;i
++)
1923 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_SPARE
);
1925 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
))
1926 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_JOURNAL
);
1928 if (test_bit(MD_HAS_PPL
, &mddev
->flags
)) {
1929 if (test_bit(MD_HAS_MULTIPLE_PPLS
, &mddev
->flags
))
1931 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS
);
1933 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_PPL
);
1934 sb
->ppl
.offset
= cpu_to_le16(rdev
->ppl
.offset
);
1935 sb
->ppl
.size
= cpu_to_le16(rdev
->ppl
.size
);
1938 rdev_for_each(rdev2
, mddev
) {
1940 if (test_bit(Faulty
, &rdev2
->flags
))
1941 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_FAULTY
);
1942 else if (test_bit(In_sync
, &rdev2
->flags
))
1943 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1944 else if (test_bit(Journal
, &rdev2
->flags
))
1945 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_JOURNAL
);
1946 else if (rdev2
->raid_disk
>= 0)
1947 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1949 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_SPARE
);
1952 sb
->sb_csum
= calc_sb_1_csum(sb
);
1955 static unsigned long long
1956 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1958 struct mdp_superblock_1
*sb
;
1959 sector_t max_sectors
;
1960 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1961 return 0; /* component must fit device */
1962 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1963 return 0; /* too confusing */
1964 if (rdev
->sb_start
< rdev
->data_offset
) {
1965 /* minor versions 1 and 2; superblock before data */
1966 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1967 max_sectors
-= rdev
->data_offset
;
1968 if (!num_sectors
|| num_sectors
> max_sectors
)
1969 num_sectors
= max_sectors
;
1970 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1971 /* minor version 0 with bitmap we can't move */
1974 /* minor version 0; superblock after data */
1976 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1977 sb_start
&= ~(sector_t
)(4*2 - 1);
1978 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1979 if (!num_sectors
|| num_sectors
> max_sectors
)
1980 num_sectors
= max_sectors
;
1981 rdev
->sb_start
= sb_start
;
1983 sb
= page_address(rdev
->sb_page
);
1984 sb
->data_size
= cpu_to_le64(num_sectors
);
1985 sb
->super_offset
= cpu_to_le64(rdev
->sb_start
);
1986 sb
->sb_csum
= calc_sb_1_csum(sb
);
1988 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1990 } while (md_super_wait(rdev
->mddev
) < 0);
1996 super_1_allow_new_offset(struct md_rdev
*rdev
,
1997 unsigned long long new_offset
)
1999 /* All necessary checks on new >= old have been done */
2000 struct bitmap
*bitmap
;
2001 if (new_offset
>= rdev
->data_offset
)
2004 /* with 1.0 metadata, there is no metadata to tread on
2005 * so we can always move back */
2006 if (rdev
->mddev
->minor_version
== 0)
2009 /* otherwise we must be sure not to step on
2010 * any metadata, so stay:
2011 * 36K beyond start of superblock
2012 * beyond end of badblocks
2013 * beyond write-intent bitmap
2015 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
2017 bitmap
= rdev
->mddev
->bitmap
;
2018 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
2019 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
2020 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
2022 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
2028 static struct super_type super_types
[] = {
2031 .owner
= THIS_MODULE
,
2032 .load_super
= super_90_load
,
2033 .validate_super
= super_90_validate
,
2034 .sync_super
= super_90_sync
,
2035 .rdev_size_change
= super_90_rdev_size_change
,
2036 .allow_new_offset
= super_90_allow_new_offset
,
2040 .owner
= THIS_MODULE
,
2041 .load_super
= super_1_load
,
2042 .validate_super
= super_1_validate
,
2043 .sync_super
= super_1_sync
,
2044 .rdev_size_change
= super_1_rdev_size_change
,
2045 .allow_new_offset
= super_1_allow_new_offset
,
2049 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
2051 if (mddev
->sync_super
) {
2052 mddev
->sync_super(mddev
, rdev
);
2056 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
2058 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
2061 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
2063 struct md_rdev
*rdev
, *rdev2
;
2066 rdev_for_each_rcu(rdev
, mddev1
) {
2067 if (test_bit(Faulty
, &rdev
->flags
) ||
2068 test_bit(Journal
, &rdev
->flags
) ||
2069 rdev
->raid_disk
== -1)
2071 rdev_for_each_rcu(rdev2
, mddev2
) {
2072 if (test_bit(Faulty
, &rdev2
->flags
) ||
2073 test_bit(Journal
, &rdev2
->flags
) ||
2074 rdev2
->raid_disk
== -1)
2076 if (rdev
->bdev
->bd_contains
==
2077 rdev2
->bdev
->bd_contains
) {
2087 static LIST_HEAD(pending_raid_disks
);
2090 * Try to register data integrity profile for an mddev
2092 * This is called when an array is started and after a disk has been kicked
2093 * from the array. It only succeeds if all working and active component devices
2094 * are integrity capable with matching profiles.
2096 int md_integrity_register(struct mddev
*mddev
)
2098 struct md_rdev
*rdev
, *reference
= NULL
;
2100 if (list_empty(&mddev
->disks
))
2101 return 0; /* nothing to do */
2102 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
2103 return 0; /* shouldn't register, or already is */
2104 rdev_for_each(rdev
, mddev
) {
2105 /* skip spares and non-functional disks */
2106 if (test_bit(Faulty
, &rdev
->flags
))
2108 if (rdev
->raid_disk
< 0)
2111 /* Use the first rdev as the reference */
2115 /* does this rdev's profile match the reference profile? */
2116 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
2117 rdev
->bdev
->bd_disk
) < 0)
2120 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2123 * All component devices are integrity capable and have matching
2124 * profiles, register the common profile for the md device.
2126 blk_integrity_register(mddev
->gendisk
,
2127 bdev_get_integrity(reference
->bdev
));
2129 pr_debug("md: data integrity enabled on %s\n", mdname(mddev
));
2130 if (bioset_integrity_create(&mddev
->bio_set
, BIO_POOL_SIZE
)) {
2131 pr_err("md: failed to create integrity pool for %s\n",
2137 EXPORT_SYMBOL(md_integrity_register
);
2140 * Attempt to add an rdev, but only if it is consistent with the current
2143 int md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2145 struct blk_integrity
*bi_mddev
;
2146 char name
[BDEVNAME_SIZE
];
2148 if (!mddev
->gendisk
)
2151 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2153 if (!bi_mddev
) /* nothing to do */
2156 if (blk_integrity_compare(mddev
->gendisk
, rdev
->bdev
->bd_disk
) != 0) {
2157 pr_err("%s: incompatible integrity profile for %s\n",
2158 mdname(mddev
), bdevname(rdev
->bdev
, name
));
2164 EXPORT_SYMBOL(md_integrity_add_rdev
);
2166 static int bind_rdev_to_array(struct md_rdev
*rdev
, struct mddev
*mddev
)
2168 char b
[BDEVNAME_SIZE
];
2172 /* prevent duplicates */
2173 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2176 if ((bdev_read_only(rdev
->bdev
) || bdev_read_only(rdev
->meta_bdev
)) &&
2180 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2181 if (!test_bit(Journal
, &rdev
->flags
) &&
2183 (mddev
->dev_sectors
== 0 || rdev
->sectors
< mddev
->dev_sectors
)) {
2185 /* Cannot change size, so fail
2186 * If mddev->level <= 0, then we don't care
2187 * about aligning sizes (e.g. linear)
2189 if (mddev
->level
> 0)
2192 mddev
->dev_sectors
= rdev
->sectors
;
2195 /* Verify rdev->desc_nr is unique.
2196 * If it is -1, assign a free number, else
2197 * check number is not in use
2200 if (rdev
->desc_nr
< 0) {
2203 choice
= mddev
->raid_disks
;
2204 while (md_find_rdev_nr_rcu(mddev
, choice
))
2206 rdev
->desc_nr
= choice
;
2208 if (md_find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2214 if (!test_bit(Journal
, &rdev
->flags
) &&
2215 mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2216 pr_warn("md: %s: array is limited to %d devices\n",
2217 mdname(mddev
), mddev
->max_disks
);
2220 bdevname(rdev
->bdev
,b
);
2221 strreplace(b
, '/', '!');
2223 rdev
->mddev
= mddev
;
2224 pr_debug("md: bind<%s>\n", b
);
2226 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2229 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2230 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2231 /* failure here is OK */;
2232 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2234 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2235 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2237 /* May as well allow recovery to be retried once */
2238 mddev
->recovery_disabled
++;
2243 pr_warn("md: failed to register dev-%s for %s\n",
2248 static void md_delayed_delete(struct work_struct
*ws
)
2250 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2251 kobject_del(&rdev
->kobj
);
2252 kobject_put(&rdev
->kobj
);
2255 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2257 char b
[BDEVNAME_SIZE
];
2259 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2260 list_del_rcu(&rdev
->same_set
);
2261 pr_debug("md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2263 sysfs_remove_link(&rdev
->kobj
, "block");
2264 sysfs_put(rdev
->sysfs_state
);
2265 rdev
->sysfs_state
= NULL
;
2266 rdev
->badblocks
.count
= 0;
2267 /* We need to delay this, otherwise we can deadlock when
2268 * writing to 'remove' to "dev/state". We also need
2269 * to delay it due to rcu usage.
2272 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2273 kobject_get(&rdev
->kobj
);
2274 queue_work(md_misc_wq
, &rdev
->del_work
);
2278 * prevent the device from being mounted, repartitioned or
2279 * otherwise reused by a RAID array (or any other kernel
2280 * subsystem), by bd_claiming the device.
2282 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2285 struct block_device
*bdev
;
2286 char b
[BDEVNAME_SIZE
];
2288 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2289 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2291 pr_warn("md: could not open %s.\n", __bdevname(dev
, b
));
2292 return PTR_ERR(bdev
);
2298 static void unlock_rdev(struct md_rdev
*rdev
)
2300 struct block_device
*bdev
= rdev
->bdev
;
2302 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2305 void md_autodetect_dev(dev_t dev
);
2307 static void export_rdev(struct md_rdev
*rdev
)
2309 char b
[BDEVNAME_SIZE
];
2311 pr_debug("md: export_rdev(%s)\n", bdevname(rdev
->bdev
,b
));
2312 md_rdev_clear(rdev
);
2314 if (test_bit(AutoDetected
, &rdev
->flags
))
2315 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2318 kobject_put(&rdev
->kobj
);
2321 void md_kick_rdev_from_array(struct md_rdev
*rdev
)
2323 unbind_rdev_from_array(rdev
);
2326 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array
);
2328 static void export_array(struct mddev
*mddev
)
2330 struct md_rdev
*rdev
;
2332 while (!list_empty(&mddev
->disks
)) {
2333 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2335 md_kick_rdev_from_array(rdev
);
2337 mddev
->raid_disks
= 0;
2338 mddev
->major_version
= 0;
2341 static bool set_in_sync(struct mddev
*mddev
)
2343 lockdep_assert_held(&mddev
->lock
);
2344 if (!mddev
->in_sync
) {
2345 mddev
->sync_checkers
++;
2346 spin_unlock(&mddev
->lock
);
2347 percpu_ref_switch_to_atomic_sync(&mddev
->writes_pending
);
2348 spin_lock(&mddev
->lock
);
2349 if (!mddev
->in_sync
&&
2350 percpu_ref_is_zero(&mddev
->writes_pending
)) {
2353 * Ensure ->in_sync is visible before we clear
2357 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
2358 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
2360 if (--mddev
->sync_checkers
== 0)
2361 percpu_ref_switch_to_percpu(&mddev
->writes_pending
);
2363 if (mddev
->safemode
== 1)
2364 mddev
->safemode
= 0;
2365 return mddev
->in_sync
;
2368 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2370 /* Update each superblock (in-memory image), but
2371 * if we are allowed to, skip spares which already
2372 * have the right event counter, or have one earlier
2373 * (which would mean they aren't being marked as dirty
2374 * with the rest of the array)
2376 struct md_rdev
*rdev
;
2377 rdev_for_each(rdev
, mddev
) {
2378 if (rdev
->sb_events
== mddev
->events
||
2380 rdev
->raid_disk
< 0 &&
2381 rdev
->sb_events
+1 == mddev
->events
)) {
2382 /* Don't update this superblock */
2383 rdev
->sb_loaded
= 2;
2385 sync_super(mddev
, rdev
);
2386 rdev
->sb_loaded
= 1;
2391 static bool does_sb_need_changing(struct mddev
*mddev
)
2393 struct md_rdev
*rdev
;
2394 struct mdp_superblock_1
*sb
;
2397 /* Find a good rdev */
2398 rdev_for_each(rdev
, mddev
)
2399 if ((rdev
->raid_disk
>= 0) && !test_bit(Faulty
, &rdev
->flags
))
2402 /* No good device found. */
2406 sb
= page_address(rdev
->sb_page
);
2407 /* Check if a device has become faulty or a spare become active */
2408 rdev_for_each(rdev
, mddev
) {
2409 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
2410 /* Device activated? */
2411 if (role
== 0xffff && rdev
->raid_disk
>=0 &&
2412 !test_bit(Faulty
, &rdev
->flags
))
2414 /* Device turned faulty? */
2415 if (test_bit(Faulty
, &rdev
->flags
) && (role
< 0xfffd))
2419 /* Check if any mddev parameters have changed */
2420 if ((mddev
->dev_sectors
!= le64_to_cpu(sb
->size
)) ||
2421 (mddev
->reshape_position
!= le64_to_cpu(sb
->reshape_position
)) ||
2422 (mddev
->layout
!= le32_to_cpu(sb
->layout
)) ||
2423 (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
)) ||
2424 (mddev
->chunk_sectors
!= le32_to_cpu(sb
->chunksize
)))
2430 void md_update_sb(struct mddev
*mddev
, int force_change
)
2432 struct md_rdev
*rdev
;
2435 int any_badblocks_changed
= 0;
2440 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2445 if (mddev_is_clustered(mddev
)) {
2446 if (test_and_clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
))
2448 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
))
2450 ret
= md_cluster_ops
->metadata_update_start(mddev
);
2451 /* Has someone else has updated the sb */
2452 if (!does_sb_need_changing(mddev
)) {
2454 md_cluster_ops
->metadata_update_cancel(mddev
);
2455 bit_clear_unless(&mddev
->sb_flags
, BIT(MD_SB_CHANGE_PENDING
),
2456 BIT(MD_SB_CHANGE_DEVS
) |
2457 BIT(MD_SB_CHANGE_CLEAN
));
2463 * First make sure individual recovery_offsets are correct
2464 * curr_resync_completed can only be used during recovery.
2465 * During reshape/resync it might use array-addresses rather
2466 * that device addresses.
2468 rdev_for_each(rdev
, mddev
) {
2469 if (rdev
->raid_disk
>= 0 &&
2470 mddev
->delta_disks
>= 0 &&
2471 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
2472 test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
) &&
2473 !test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
2474 !test_bit(Journal
, &rdev
->flags
) &&
2475 !test_bit(In_sync
, &rdev
->flags
) &&
2476 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2477 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2480 if (!mddev
->persistent
) {
2481 clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
2482 clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2483 if (!mddev
->external
) {
2484 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
2485 rdev_for_each(rdev
, mddev
) {
2486 if (rdev
->badblocks
.changed
) {
2487 rdev
->badblocks
.changed
= 0;
2488 ack_all_badblocks(&rdev
->badblocks
);
2489 md_error(mddev
, rdev
);
2491 clear_bit(Blocked
, &rdev
->flags
);
2492 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2493 wake_up(&rdev
->blocked_wait
);
2496 wake_up(&mddev
->sb_wait
);
2500 spin_lock(&mddev
->lock
);
2502 mddev
->utime
= ktime_get_real_seconds();
2504 if (test_and_clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
))
2506 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
))
2507 /* just a clean<-> dirty transition, possibly leave spares alone,
2508 * though if events isn't the right even/odd, we will have to do
2514 if (mddev
->degraded
)
2515 /* If the array is degraded, then skipping spares is both
2516 * dangerous and fairly pointless.
2517 * Dangerous because a device that was removed from the array
2518 * might have a event_count that still looks up-to-date,
2519 * so it can be re-added without a resync.
2520 * Pointless because if there are any spares to skip,
2521 * then a recovery will happen and soon that array won't
2522 * be degraded any more and the spare can go back to sleep then.
2526 sync_req
= mddev
->in_sync
;
2528 /* If this is just a dirty<->clean transition, and the array is clean
2529 * and 'events' is odd, we can roll back to the previous clean state */
2531 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2532 && mddev
->can_decrease_events
2533 && mddev
->events
!= 1) {
2535 mddev
->can_decrease_events
= 0;
2537 /* otherwise we have to go forward and ... */
2539 mddev
->can_decrease_events
= nospares
;
2543 * This 64-bit counter should never wrap.
2544 * Either we are in around ~1 trillion A.C., assuming
2545 * 1 reboot per second, or we have a bug...
2547 WARN_ON(mddev
->events
== 0);
2549 rdev_for_each(rdev
, mddev
) {
2550 if (rdev
->badblocks
.changed
)
2551 any_badblocks_changed
++;
2552 if (test_bit(Faulty
, &rdev
->flags
))
2553 set_bit(FaultRecorded
, &rdev
->flags
);
2556 sync_sbs(mddev
, nospares
);
2557 spin_unlock(&mddev
->lock
);
2559 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2560 mdname(mddev
), mddev
->in_sync
);
2563 blk_add_trace_msg(mddev
->queue
, "md md_update_sb");
2565 md_bitmap_update_sb(mddev
->bitmap
);
2566 rdev_for_each(rdev
, mddev
) {
2567 char b
[BDEVNAME_SIZE
];
2569 if (rdev
->sb_loaded
!= 1)
2570 continue; /* no noise on spare devices */
2572 if (!test_bit(Faulty
, &rdev
->flags
)) {
2573 md_super_write(mddev
,rdev
,
2574 rdev
->sb_start
, rdev
->sb_size
,
2576 pr_debug("md: (write) %s's sb offset: %llu\n",
2577 bdevname(rdev
->bdev
, b
),
2578 (unsigned long long)rdev
->sb_start
);
2579 rdev
->sb_events
= mddev
->events
;
2580 if (rdev
->badblocks
.size
) {
2581 md_super_write(mddev
, rdev
,
2582 rdev
->badblocks
.sector
,
2583 rdev
->badblocks
.size
<< 9,
2585 rdev
->badblocks
.size
= 0;
2589 pr_debug("md: %s (skipping faulty)\n",
2590 bdevname(rdev
->bdev
, b
));
2592 if (mddev
->level
== LEVEL_MULTIPATH
)
2593 /* only need to write one superblock... */
2596 if (md_super_wait(mddev
) < 0)
2598 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2600 if (mddev_is_clustered(mddev
) && ret
== 0)
2601 md_cluster_ops
->metadata_update_finish(mddev
);
2603 if (mddev
->in_sync
!= sync_req
||
2604 !bit_clear_unless(&mddev
->sb_flags
, BIT(MD_SB_CHANGE_PENDING
),
2605 BIT(MD_SB_CHANGE_DEVS
) | BIT(MD_SB_CHANGE_CLEAN
)))
2606 /* have to write it out again */
2608 wake_up(&mddev
->sb_wait
);
2609 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2610 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2612 rdev_for_each(rdev
, mddev
) {
2613 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2614 clear_bit(Blocked
, &rdev
->flags
);
2616 if (any_badblocks_changed
)
2617 ack_all_badblocks(&rdev
->badblocks
);
2618 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2619 wake_up(&rdev
->blocked_wait
);
2622 EXPORT_SYMBOL(md_update_sb
);
2624 static int add_bound_rdev(struct md_rdev
*rdev
)
2626 struct mddev
*mddev
= rdev
->mddev
;
2628 bool add_journal
= test_bit(Journal
, &rdev
->flags
);
2630 if (!mddev
->pers
->hot_remove_disk
|| add_journal
) {
2631 /* If there is hot_add_disk but no hot_remove_disk
2632 * then added disks for geometry changes,
2633 * and should be added immediately.
2635 super_types
[mddev
->major_version
].
2636 validate_super(mddev
, rdev
);
2638 mddev_suspend(mddev
);
2639 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
2641 mddev_resume(mddev
);
2643 md_kick_rdev_from_array(rdev
);
2647 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2649 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2650 if (mddev
->degraded
)
2651 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2652 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2653 md_new_event(mddev
);
2654 md_wakeup_thread(mddev
->thread
);
2658 /* words written to sysfs files may, or may not, be \n terminated.
2659 * We want to accept with case. For this we use cmd_match.
2661 static int cmd_match(const char *cmd
, const char *str
)
2663 /* See if cmd, written into a sysfs file, matches
2664 * str. They must either be the same, or cmd can
2665 * have a trailing newline
2667 while (*cmd
&& *str
&& *cmd
== *str
) {
2678 struct rdev_sysfs_entry
{
2679 struct attribute attr
;
2680 ssize_t (*show
)(struct md_rdev
*, char *);
2681 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2685 state_show(struct md_rdev
*rdev
, char *page
)
2689 unsigned long flags
= READ_ONCE(rdev
->flags
);
2691 if (test_bit(Faulty
, &flags
) ||
2692 (!test_bit(ExternalBbl
, &flags
) &&
2693 rdev
->badblocks
.unacked_exist
))
2694 len
+= sprintf(page
+len
, "faulty%s", sep
);
2695 if (test_bit(In_sync
, &flags
))
2696 len
+= sprintf(page
+len
, "in_sync%s", sep
);
2697 if (test_bit(Journal
, &flags
))
2698 len
+= sprintf(page
+len
, "journal%s", sep
);
2699 if (test_bit(WriteMostly
, &flags
))
2700 len
+= sprintf(page
+len
, "write_mostly%s", sep
);
2701 if (test_bit(Blocked
, &flags
) ||
2702 (rdev
->badblocks
.unacked_exist
2703 && !test_bit(Faulty
, &flags
)))
2704 len
+= sprintf(page
+len
, "blocked%s", sep
);
2705 if (!test_bit(Faulty
, &flags
) &&
2706 !test_bit(Journal
, &flags
) &&
2707 !test_bit(In_sync
, &flags
))
2708 len
+= sprintf(page
+len
, "spare%s", sep
);
2709 if (test_bit(WriteErrorSeen
, &flags
))
2710 len
+= sprintf(page
+len
, "write_error%s", sep
);
2711 if (test_bit(WantReplacement
, &flags
))
2712 len
+= sprintf(page
+len
, "want_replacement%s", sep
);
2713 if (test_bit(Replacement
, &flags
))
2714 len
+= sprintf(page
+len
, "replacement%s", sep
);
2715 if (test_bit(ExternalBbl
, &flags
))
2716 len
+= sprintf(page
+len
, "external_bbl%s", sep
);
2717 if (test_bit(FailFast
, &flags
))
2718 len
+= sprintf(page
+len
, "failfast%s", sep
);
2723 return len
+sprintf(page
+len
, "\n");
2727 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2730 * faulty - simulates an error
2731 * remove - disconnects the device
2732 * writemostly - sets write_mostly
2733 * -writemostly - clears write_mostly
2734 * blocked - sets the Blocked flags
2735 * -blocked - clears the Blocked and possibly simulates an error
2736 * insync - sets Insync providing device isn't active
2737 * -insync - clear Insync for a device with a slot assigned,
2738 * so that it gets rebuilt based on bitmap
2739 * write_error - sets WriteErrorSeen
2740 * -write_error - clears WriteErrorSeen
2741 * {,-}failfast - set/clear FailFast
2744 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2745 md_error(rdev
->mddev
, rdev
);
2746 if (test_bit(Faulty
, &rdev
->flags
))
2750 } else if (cmd_match(buf
, "remove")) {
2751 if (rdev
->mddev
->pers
) {
2752 clear_bit(Blocked
, &rdev
->flags
);
2753 remove_and_add_spares(rdev
->mddev
, rdev
);
2755 if (rdev
->raid_disk
>= 0)
2758 struct mddev
*mddev
= rdev
->mddev
;
2760 if (mddev_is_clustered(mddev
))
2761 err
= md_cluster_ops
->remove_disk(mddev
, rdev
);
2764 md_kick_rdev_from_array(rdev
);
2766 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2767 md_wakeup_thread(mddev
->thread
);
2769 md_new_event(mddev
);
2772 } else if (cmd_match(buf
, "writemostly")) {
2773 set_bit(WriteMostly
, &rdev
->flags
);
2775 } else if (cmd_match(buf
, "-writemostly")) {
2776 clear_bit(WriteMostly
, &rdev
->flags
);
2778 } else if (cmd_match(buf
, "blocked")) {
2779 set_bit(Blocked
, &rdev
->flags
);
2781 } else if (cmd_match(buf
, "-blocked")) {
2782 if (!test_bit(Faulty
, &rdev
->flags
) &&
2783 !test_bit(ExternalBbl
, &rdev
->flags
) &&
2784 rdev
->badblocks
.unacked_exist
) {
2785 /* metadata handler doesn't understand badblocks,
2786 * so we need to fail the device
2788 md_error(rdev
->mddev
, rdev
);
2790 clear_bit(Blocked
, &rdev
->flags
);
2791 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2792 wake_up(&rdev
->blocked_wait
);
2793 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2794 md_wakeup_thread(rdev
->mddev
->thread
);
2797 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2798 set_bit(In_sync
, &rdev
->flags
);
2800 } else if (cmd_match(buf
, "failfast")) {
2801 set_bit(FailFast
, &rdev
->flags
);
2803 } else if (cmd_match(buf
, "-failfast")) {
2804 clear_bit(FailFast
, &rdev
->flags
);
2806 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0 &&
2807 !test_bit(Journal
, &rdev
->flags
)) {
2808 if (rdev
->mddev
->pers
== NULL
) {
2809 clear_bit(In_sync
, &rdev
->flags
);
2810 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2811 rdev
->raid_disk
= -1;
2814 } else if (cmd_match(buf
, "write_error")) {
2815 set_bit(WriteErrorSeen
, &rdev
->flags
);
2817 } else if (cmd_match(buf
, "-write_error")) {
2818 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2820 } else if (cmd_match(buf
, "want_replacement")) {
2821 /* Any non-spare device that is not a replacement can
2822 * become want_replacement at any time, but we then need to
2823 * check if recovery is needed.
2825 if (rdev
->raid_disk
>= 0 &&
2826 !test_bit(Journal
, &rdev
->flags
) &&
2827 !test_bit(Replacement
, &rdev
->flags
))
2828 set_bit(WantReplacement
, &rdev
->flags
);
2829 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2830 md_wakeup_thread(rdev
->mddev
->thread
);
2832 } else if (cmd_match(buf
, "-want_replacement")) {
2833 /* Clearing 'want_replacement' is always allowed.
2834 * Once replacements starts it is too late though.
2837 clear_bit(WantReplacement
, &rdev
->flags
);
2838 } else if (cmd_match(buf
, "replacement")) {
2839 /* Can only set a device as a replacement when array has not
2840 * yet been started. Once running, replacement is automatic
2841 * from spares, or by assigning 'slot'.
2843 if (rdev
->mddev
->pers
)
2846 set_bit(Replacement
, &rdev
->flags
);
2849 } else if (cmd_match(buf
, "-replacement")) {
2850 /* Similarly, can only clear Replacement before start */
2851 if (rdev
->mddev
->pers
)
2854 clear_bit(Replacement
, &rdev
->flags
);
2857 } else if (cmd_match(buf
, "re-add")) {
2858 if (test_bit(Faulty
, &rdev
->flags
) && (rdev
->raid_disk
== -1) &&
2859 rdev
->saved_raid_disk
>= 0) {
2860 /* clear_bit is performed _after_ all the devices
2861 * have their local Faulty bit cleared. If any writes
2862 * happen in the meantime in the local node, they
2863 * will land in the local bitmap, which will be synced
2864 * by this node eventually
2866 if (!mddev_is_clustered(rdev
->mddev
) ||
2867 (err
= md_cluster_ops
->gather_bitmaps(rdev
)) == 0) {
2868 clear_bit(Faulty
, &rdev
->flags
);
2869 err
= add_bound_rdev(rdev
);
2873 } else if (cmd_match(buf
, "external_bbl") && (rdev
->mddev
->external
)) {
2874 set_bit(ExternalBbl
, &rdev
->flags
);
2875 rdev
->badblocks
.shift
= 0;
2877 } else if (cmd_match(buf
, "-external_bbl") && (rdev
->mddev
->external
)) {
2878 clear_bit(ExternalBbl
, &rdev
->flags
);
2882 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2883 return err
? err
: len
;
2885 static struct rdev_sysfs_entry rdev_state
=
2886 __ATTR_PREALLOC(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2889 errors_show(struct md_rdev
*rdev
, char *page
)
2891 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2895 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2900 rv
= kstrtouint(buf
, 10, &n
);
2903 atomic_set(&rdev
->corrected_errors
, n
);
2906 static struct rdev_sysfs_entry rdev_errors
=
2907 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2910 slot_show(struct md_rdev
*rdev
, char *page
)
2912 if (test_bit(Journal
, &rdev
->flags
))
2913 return sprintf(page
, "journal\n");
2914 else if (rdev
->raid_disk
< 0)
2915 return sprintf(page
, "none\n");
2917 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2921 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2926 if (test_bit(Journal
, &rdev
->flags
))
2928 if (strncmp(buf
, "none", 4)==0)
2931 err
= kstrtouint(buf
, 10, (unsigned int *)&slot
);
2935 if (rdev
->mddev
->pers
&& slot
== -1) {
2936 /* Setting 'slot' on an active array requires also
2937 * updating the 'rd%d' link, and communicating
2938 * with the personality with ->hot_*_disk.
2939 * For now we only support removing
2940 * failed/spare devices. This normally happens automatically,
2941 * but not when the metadata is externally managed.
2943 if (rdev
->raid_disk
== -1)
2945 /* personality does all needed checks */
2946 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2948 clear_bit(Blocked
, &rdev
->flags
);
2949 remove_and_add_spares(rdev
->mddev
, rdev
);
2950 if (rdev
->raid_disk
>= 0)
2952 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2953 md_wakeup_thread(rdev
->mddev
->thread
);
2954 } else if (rdev
->mddev
->pers
) {
2955 /* Activating a spare .. or possibly reactivating
2956 * if we ever get bitmaps working here.
2960 if (rdev
->raid_disk
!= -1)
2963 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2966 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2969 if (slot
>= rdev
->mddev
->raid_disks
&&
2970 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2973 rdev
->raid_disk
= slot
;
2974 if (test_bit(In_sync
, &rdev
->flags
))
2975 rdev
->saved_raid_disk
= slot
;
2977 rdev
->saved_raid_disk
= -1;
2978 clear_bit(In_sync
, &rdev
->flags
);
2979 clear_bit(Bitmap_sync
, &rdev
->flags
);
2980 err
= rdev
->mddev
->pers
->
2981 hot_add_disk(rdev
->mddev
, rdev
);
2983 rdev
->raid_disk
= -1;
2986 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2987 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2988 /* failure here is OK */;
2989 /* don't wakeup anyone, leave that to userspace. */
2991 if (slot
>= rdev
->mddev
->raid_disks
&&
2992 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2994 rdev
->raid_disk
= slot
;
2995 /* assume it is working */
2996 clear_bit(Faulty
, &rdev
->flags
);
2997 clear_bit(WriteMostly
, &rdev
->flags
);
2998 set_bit(In_sync
, &rdev
->flags
);
2999 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
3004 static struct rdev_sysfs_entry rdev_slot
=
3005 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
3008 offset_show(struct md_rdev
*rdev
, char *page
)
3010 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
3014 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3016 unsigned long long offset
;
3017 if (kstrtoull(buf
, 10, &offset
) < 0)
3019 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
3021 if (rdev
->sectors
&& rdev
->mddev
->external
)
3022 /* Must set offset before size, so overlap checks
3025 rdev
->data_offset
= offset
;
3026 rdev
->new_data_offset
= offset
;
3030 static struct rdev_sysfs_entry rdev_offset
=
3031 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
3033 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
3035 return sprintf(page
, "%llu\n",
3036 (unsigned long long)rdev
->new_data_offset
);
3039 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
3040 const char *buf
, size_t len
)
3042 unsigned long long new_offset
;
3043 struct mddev
*mddev
= rdev
->mddev
;
3045 if (kstrtoull(buf
, 10, &new_offset
) < 0)
3048 if (mddev
->sync_thread
||
3049 test_bit(MD_RECOVERY_RUNNING
,&mddev
->recovery
))
3051 if (new_offset
== rdev
->data_offset
)
3052 /* reset is always permitted */
3054 else if (new_offset
> rdev
->data_offset
) {
3055 /* must not push array size beyond rdev_sectors */
3056 if (new_offset
- rdev
->data_offset
3057 + mddev
->dev_sectors
> rdev
->sectors
)
3060 /* Metadata worries about other space details. */
3062 /* decreasing the offset is inconsistent with a backwards
3065 if (new_offset
< rdev
->data_offset
&&
3066 mddev
->reshape_backwards
)
3068 /* Increasing offset is inconsistent with forwards
3069 * reshape. reshape_direction should be set to
3070 * 'backwards' first.
3072 if (new_offset
> rdev
->data_offset
&&
3073 !mddev
->reshape_backwards
)
3076 if (mddev
->pers
&& mddev
->persistent
&&
3077 !super_types
[mddev
->major_version
]
3078 .allow_new_offset(rdev
, new_offset
))
3080 rdev
->new_data_offset
= new_offset
;
3081 if (new_offset
> rdev
->data_offset
)
3082 mddev
->reshape_backwards
= 1;
3083 else if (new_offset
< rdev
->data_offset
)
3084 mddev
->reshape_backwards
= 0;
3088 static struct rdev_sysfs_entry rdev_new_offset
=
3089 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
3092 rdev_size_show(struct md_rdev
*rdev
, char *page
)
3094 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
3097 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
3099 /* check if two start/length pairs overlap */
3107 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
3109 unsigned long long blocks
;
3112 if (kstrtoull(buf
, 10, &blocks
) < 0)
3115 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
3116 return -EINVAL
; /* sector conversion overflow */
3119 if (new != blocks
* 2)
3120 return -EINVAL
; /* unsigned long long to sector_t overflow */
3127 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3129 struct mddev
*my_mddev
= rdev
->mddev
;
3130 sector_t oldsectors
= rdev
->sectors
;
3133 if (test_bit(Journal
, &rdev
->flags
))
3135 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3137 if (rdev
->data_offset
!= rdev
->new_data_offset
)
3138 return -EINVAL
; /* too confusing */
3139 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
3140 if (my_mddev
->persistent
) {
3141 sectors
= super_types
[my_mddev
->major_version
].
3142 rdev_size_change(rdev
, sectors
);
3145 } else if (!sectors
)
3146 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
3148 if (!my_mddev
->pers
->resize
)
3149 /* Cannot change size for RAID0 or Linear etc */
3152 if (sectors
< my_mddev
->dev_sectors
)
3153 return -EINVAL
; /* component must fit device */
3155 rdev
->sectors
= sectors
;
3156 if (sectors
> oldsectors
&& my_mddev
->external
) {
3157 /* Need to check that all other rdevs with the same
3158 * ->bdev do not overlap. 'rcu' is sufficient to walk
3159 * the rdev lists safely.
3160 * This check does not provide a hard guarantee, it
3161 * just helps avoid dangerous mistakes.
3163 struct mddev
*mddev
;
3165 struct list_head
*tmp
;
3168 for_each_mddev(mddev
, tmp
) {
3169 struct md_rdev
*rdev2
;
3171 rdev_for_each(rdev2
, mddev
)
3172 if (rdev
->bdev
== rdev2
->bdev
&&
3174 overlaps(rdev
->data_offset
, rdev
->sectors
,
3187 /* Someone else could have slipped in a size
3188 * change here, but doing so is just silly.
3189 * We put oldsectors back because we *know* it is
3190 * safe, and trust userspace not to race with
3193 rdev
->sectors
= oldsectors
;
3200 static struct rdev_sysfs_entry rdev_size
=
3201 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3203 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3205 unsigned long long recovery_start
= rdev
->recovery_offset
;
3207 if (test_bit(In_sync
, &rdev
->flags
) ||
3208 recovery_start
== MaxSector
)
3209 return sprintf(page
, "none\n");
3211 return sprintf(page
, "%llu\n", recovery_start
);
3214 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3216 unsigned long long recovery_start
;
3218 if (cmd_match(buf
, "none"))
3219 recovery_start
= MaxSector
;
3220 else if (kstrtoull(buf
, 10, &recovery_start
))
3223 if (rdev
->mddev
->pers
&&
3224 rdev
->raid_disk
>= 0)
3227 rdev
->recovery_offset
= recovery_start
;
3228 if (recovery_start
== MaxSector
)
3229 set_bit(In_sync
, &rdev
->flags
);
3231 clear_bit(In_sync
, &rdev
->flags
);
3235 static struct rdev_sysfs_entry rdev_recovery_start
=
3236 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3238 /* sysfs access to bad-blocks list.
3239 * We present two files.
3240 * 'bad-blocks' lists sector numbers and lengths of ranges that
3241 * are recorded as bad. The list is truncated to fit within
3242 * the one-page limit of sysfs.
3243 * Writing "sector length" to this file adds an acknowledged
3245 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3246 * been acknowledged. Writing to this file adds bad blocks
3247 * without acknowledging them. This is largely for testing.
3249 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3251 return badblocks_show(&rdev
->badblocks
, page
, 0);
3253 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3255 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3256 /* Maybe that ack was all we needed */
3257 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3258 wake_up(&rdev
->blocked_wait
);
3261 static struct rdev_sysfs_entry rdev_bad_blocks
=
3262 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3264 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3266 return badblocks_show(&rdev
->badblocks
, page
, 1);
3268 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3270 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3272 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3273 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3276 ppl_sector_show(struct md_rdev
*rdev
, char *page
)
3278 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->ppl
.sector
);
3282 ppl_sector_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3284 unsigned long long sector
;
3286 if (kstrtoull(buf
, 10, §or
) < 0)
3288 if (sector
!= (sector_t
)sector
)
3291 if (rdev
->mddev
->pers
&& test_bit(MD_HAS_PPL
, &rdev
->mddev
->flags
) &&
3292 rdev
->raid_disk
>= 0)
3295 if (rdev
->mddev
->persistent
) {
3296 if (rdev
->mddev
->major_version
== 0)
3298 if ((sector
> rdev
->sb_start
&&
3299 sector
- rdev
->sb_start
> S16_MAX
) ||
3300 (sector
< rdev
->sb_start
&&
3301 rdev
->sb_start
- sector
> -S16_MIN
))
3303 rdev
->ppl
.offset
= sector
- rdev
->sb_start
;
3304 } else if (!rdev
->mddev
->external
) {
3307 rdev
->ppl
.sector
= sector
;
3311 static struct rdev_sysfs_entry rdev_ppl_sector
=
3312 __ATTR(ppl_sector
, S_IRUGO
|S_IWUSR
, ppl_sector_show
, ppl_sector_store
);
3315 ppl_size_show(struct md_rdev
*rdev
, char *page
)
3317 return sprintf(page
, "%u\n", rdev
->ppl
.size
);
3321 ppl_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3325 if (kstrtouint(buf
, 10, &size
) < 0)
3328 if (rdev
->mddev
->pers
&& test_bit(MD_HAS_PPL
, &rdev
->mddev
->flags
) &&
3329 rdev
->raid_disk
>= 0)
3332 if (rdev
->mddev
->persistent
) {
3333 if (rdev
->mddev
->major_version
== 0)
3337 } else if (!rdev
->mddev
->external
) {
3340 rdev
->ppl
.size
= size
;
3344 static struct rdev_sysfs_entry rdev_ppl_size
=
3345 __ATTR(ppl_size
, S_IRUGO
|S_IWUSR
, ppl_size_show
, ppl_size_store
);
3347 static struct attribute
*rdev_default_attrs
[] = {
3352 &rdev_new_offset
.attr
,
3354 &rdev_recovery_start
.attr
,
3355 &rdev_bad_blocks
.attr
,
3356 &rdev_unack_bad_blocks
.attr
,
3357 &rdev_ppl_sector
.attr
,
3358 &rdev_ppl_size
.attr
,
3362 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3364 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3365 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3371 return entry
->show(rdev
, page
);
3375 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3376 const char *page
, size_t length
)
3378 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3379 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3381 struct mddev
*mddev
= rdev
->mddev
;
3385 if (!capable(CAP_SYS_ADMIN
))
3387 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3389 if (rdev
->mddev
== NULL
)
3392 rv
= entry
->store(rdev
, page
, length
);
3393 mddev_unlock(mddev
);
3398 static void rdev_free(struct kobject
*ko
)
3400 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3403 static const struct sysfs_ops rdev_sysfs_ops
= {
3404 .show
= rdev_attr_show
,
3405 .store
= rdev_attr_store
,
3407 static struct kobj_type rdev_ktype
= {
3408 .release
= rdev_free
,
3409 .sysfs_ops
= &rdev_sysfs_ops
,
3410 .default_attrs
= rdev_default_attrs
,
3413 int md_rdev_init(struct md_rdev
*rdev
)
3416 rdev
->saved_raid_disk
= -1;
3417 rdev
->raid_disk
= -1;
3419 rdev
->data_offset
= 0;
3420 rdev
->new_data_offset
= 0;
3421 rdev
->sb_events
= 0;
3422 rdev
->last_read_error
= 0;
3423 rdev
->sb_loaded
= 0;
3424 rdev
->bb_page
= NULL
;
3425 atomic_set(&rdev
->nr_pending
, 0);
3426 atomic_set(&rdev
->read_errors
, 0);
3427 atomic_set(&rdev
->corrected_errors
, 0);
3429 INIT_LIST_HEAD(&rdev
->same_set
);
3430 init_waitqueue_head(&rdev
->blocked_wait
);
3432 /* Add space to store bad block list.
3433 * This reserves the space even on arrays where it cannot
3434 * be used - I wonder if that matters
3436 return badblocks_init(&rdev
->badblocks
, 0);
3438 EXPORT_SYMBOL_GPL(md_rdev_init
);
3440 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3442 * mark the device faulty if:
3444 * - the device is nonexistent (zero size)
3445 * - the device has no valid superblock
3447 * a faulty rdev _never_ has rdev->sb set.
3449 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3451 char b
[BDEVNAME_SIZE
];
3453 struct md_rdev
*rdev
;
3456 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3458 return ERR_PTR(-ENOMEM
);
3460 err
= md_rdev_init(rdev
);
3463 err
= alloc_disk_sb(rdev
);
3467 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3471 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3473 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3475 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3476 bdevname(rdev
->bdev
,b
));
3481 if (super_format
>= 0) {
3482 err
= super_types
[super_format
].
3483 load_super(rdev
, NULL
, super_minor
);
3484 if (err
== -EINVAL
) {
3485 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3486 bdevname(rdev
->bdev
,b
),
3487 super_format
, super_minor
);
3491 pr_warn("md: could not read %s's sb, not importing!\n",
3492 bdevname(rdev
->bdev
,b
));
3502 md_rdev_clear(rdev
);
3504 return ERR_PTR(err
);
3508 * Check a full RAID array for plausibility
3511 static void analyze_sbs(struct mddev
*mddev
)
3514 struct md_rdev
*rdev
, *freshest
, *tmp
;
3515 char b
[BDEVNAME_SIZE
];
3518 rdev_for_each_safe(rdev
, tmp
, mddev
)
3519 switch (super_types
[mddev
->major_version
].
3520 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3527 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3528 bdevname(rdev
->bdev
,b
));
3529 md_kick_rdev_from_array(rdev
);
3532 super_types
[mddev
->major_version
].
3533 validate_super(mddev
, freshest
);
3536 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3537 if (mddev
->max_disks
&&
3538 (rdev
->desc_nr
>= mddev
->max_disks
||
3539 i
> mddev
->max_disks
)) {
3540 pr_warn("md: %s: %s: only %d devices permitted\n",
3541 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3543 md_kick_rdev_from_array(rdev
);
3546 if (rdev
!= freshest
) {
3547 if (super_types
[mddev
->major_version
].
3548 validate_super(mddev
, rdev
)) {
3549 pr_warn("md: kicking non-fresh %s from array!\n",
3550 bdevname(rdev
->bdev
,b
));
3551 md_kick_rdev_from_array(rdev
);
3555 if (mddev
->level
== LEVEL_MULTIPATH
) {
3556 rdev
->desc_nr
= i
++;
3557 rdev
->raid_disk
= rdev
->desc_nr
;
3558 set_bit(In_sync
, &rdev
->flags
);
3559 } else if (rdev
->raid_disk
>=
3560 (mddev
->raid_disks
- min(0, mddev
->delta_disks
)) &&
3561 !test_bit(Journal
, &rdev
->flags
)) {
3562 rdev
->raid_disk
= -1;
3563 clear_bit(In_sync
, &rdev
->flags
);
3568 /* Read a fixed-point number.
3569 * Numbers in sysfs attributes should be in "standard" units where
3570 * possible, so time should be in seconds.
3571 * However we internally use a a much smaller unit such as
3572 * milliseconds or jiffies.
3573 * This function takes a decimal number with a possible fractional
3574 * component, and produces an integer which is the result of
3575 * multiplying that number by 10^'scale'.
3576 * all without any floating-point arithmetic.
3578 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3580 unsigned long result
= 0;
3582 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3585 else if (decimals
< scale
) {
3588 result
= result
* 10 + value
;
3600 while (decimals
< scale
) {
3609 safe_delay_show(struct mddev
*mddev
, char *page
)
3611 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3612 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3615 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3619 if (mddev_is_clustered(mddev
)) {
3620 pr_warn("md: Safemode is disabled for clustered mode\n");
3624 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3627 mddev
->safemode_delay
= 0;
3629 unsigned long old_delay
= mddev
->safemode_delay
;
3630 unsigned long new_delay
= (msec
*HZ
)/1000;
3634 mddev
->safemode_delay
= new_delay
;
3635 if (new_delay
< old_delay
|| old_delay
== 0)
3636 mod_timer(&mddev
->safemode_timer
, jiffies
+1);
3640 static struct md_sysfs_entry md_safe_delay
=
3641 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3644 level_show(struct mddev
*mddev
, char *page
)
3646 struct md_personality
*p
;
3648 spin_lock(&mddev
->lock
);
3651 ret
= sprintf(page
, "%s\n", p
->name
);
3652 else if (mddev
->clevel
[0])
3653 ret
= sprintf(page
, "%s\n", mddev
->clevel
);
3654 else if (mddev
->level
!= LEVEL_NONE
)
3655 ret
= sprintf(page
, "%d\n", mddev
->level
);
3658 spin_unlock(&mddev
->lock
);
3663 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3668 struct md_personality
*pers
, *oldpers
;
3670 void *priv
, *oldpriv
;
3671 struct md_rdev
*rdev
;
3673 if (slen
== 0 || slen
>= sizeof(clevel
))
3676 rv
= mddev_lock(mddev
);
3680 if (mddev
->pers
== NULL
) {
3681 strncpy(mddev
->clevel
, buf
, slen
);
3682 if (mddev
->clevel
[slen
-1] == '\n')
3684 mddev
->clevel
[slen
] = 0;
3685 mddev
->level
= LEVEL_NONE
;
3693 /* request to change the personality. Need to ensure:
3694 * - array is not engaged in resync/recovery/reshape
3695 * - old personality can be suspended
3696 * - new personality will access other array.
3700 if (mddev
->sync_thread
||
3701 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3702 mddev
->reshape_position
!= MaxSector
||
3703 mddev
->sysfs_active
)
3707 if (!mddev
->pers
->quiesce
) {
3708 pr_warn("md: %s: %s does not support online personality change\n",
3709 mdname(mddev
), mddev
->pers
->name
);
3713 /* Now find the new personality */
3714 strncpy(clevel
, buf
, slen
);
3715 if (clevel
[slen
-1] == '\n')
3718 if (kstrtol(clevel
, 10, &level
))
3721 if (request_module("md-%s", clevel
) != 0)
3722 request_module("md-level-%s", clevel
);
3723 spin_lock(&pers_lock
);
3724 pers
= find_pers(level
, clevel
);
3725 if (!pers
|| !try_module_get(pers
->owner
)) {
3726 spin_unlock(&pers_lock
);
3727 pr_warn("md: personality %s not loaded\n", clevel
);
3731 spin_unlock(&pers_lock
);
3733 if (pers
== mddev
->pers
) {
3734 /* Nothing to do! */
3735 module_put(pers
->owner
);
3739 if (!pers
->takeover
) {
3740 module_put(pers
->owner
);
3741 pr_warn("md: %s: %s does not support personality takeover\n",
3742 mdname(mddev
), clevel
);
3747 rdev_for_each(rdev
, mddev
)
3748 rdev
->new_raid_disk
= rdev
->raid_disk
;
3750 /* ->takeover must set new_* and/or delta_disks
3751 * if it succeeds, and may set them when it fails.
3753 priv
= pers
->takeover(mddev
);
3755 mddev
->new_level
= mddev
->level
;
3756 mddev
->new_layout
= mddev
->layout
;
3757 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3758 mddev
->raid_disks
-= mddev
->delta_disks
;
3759 mddev
->delta_disks
= 0;
3760 mddev
->reshape_backwards
= 0;
3761 module_put(pers
->owner
);
3762 pr_warn("md: %s: %s would not accept array\n",
3763 mdname(mddev
), clevel
);
3768 /* Looks like we have a winner */
3769 mddev_suspend(mddev
);
3770 mddev_detach(mddev
);
3772 spin_lock(&mddev
->lock
);
3773 oldpers
= mddev
->pers
;
3774 oldpriv
= mddev
->private;
3776 mddev
->private = priv
;
3777 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3778 mddev
->level
= mddev
->new_level
;
3779 mddev
->layout
= mddev
->new_layout
;
3780 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3781 mddev
->delta_disks
= 0;
3782 mddev
->reshape_backwards
= 0;
3783 mddev
->degraded
= 0;
3784 spin_unlock(&mddev
->lock
);
3786 if (oldpers
->sync_request
== NULL
&&
3788 /* We are converting from a no-redundancy array
3789 * to a redundancy array and metadata is managed
3790 * externally so we need to be sure that writes
3791 * won't block due to a need to transition
3793 * until external management is started.
3796 mddev
->safemode_delay
= 0;
3797 mddev
->safemode
= 0;
3800 oldpers
->free(mddev
, oldpriv
);
3802 if (oldpers
->sync_request
== NULL
&&
3803 pers
->sync_request
!= NULL
) {
3804 /* need to add the md_redundancy_group */
3805 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3806 pr_warn("md: cannot register extra attributes for %s\n",
3808 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3810 if (oldpers
->sync_request
!= NULL
&&
3811 pers
->sync_request
== NULL
) {
3812 /* need to remove the md_redundancy_group */
3813 if (mddev
->to_remove
== NULL
)
3814 mddev
->to_remove
= &md_redundancy_group
;
3817 module_put(oldpers
->owner
);
3819 rdev_for_each(rdev
, mddev
) {
3820 if (rdev
->raid_disk
< 0)
3822 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3823 rdev
->new_raid_disk
= -1;
3824 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3826 sysfs_unlink_rdev(mddev
, rdev
);
3828 rdev_for_each(rdev
, mddev
) {
3829 if (rdev
->raid_disk
< 0)
3831 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3833 rdev
->raid_disk
= rdev
->new_raid_disk
;
3834 if (rdev
->raid_disk
< 0)
3835 clear_bit(In_sync
, &rdev
->flags
);
3837 if (sysfs_link_rdev(mddev
, rdev
))
3838 pr_warn("md: cannot register rd%d for %s after level change\n",
3839 rdev
->raid_disk
, mdname(mddev
));
3843 if (pers
->sync_request
== NULL
) {
3844 /* this is now an array without redundancy, so
3845 * it must always be in_sync
3848 del_timer_sync(&mddev
->safemode_timer
);
3850 blk_set_stacking_limits(&mddev
->queue
->limits
);
3852 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
3853 mddev_resume(mddev
);
3855 md_update_sb(mddev
, 1);
3856 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3857 md_new_event(mddev
);
3860 mddev_unlock(mddev
);
3864 static struct md_sysfs_entry md_level
=
3865 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3868 layout_show(struct mddev
*mddev
, char *page
)
3870 /* just a number, not meaningful for all levels */
3871 if (mddev
->reshape_position
!= MaxSector
&&
3872 mddev
->layout
!= mddev
->new_layout
)
3873 return sprintf(page
, "%d (%d)\n",
3874 mddev
->new_layout
, mddev
->layout
);
3875 return sprintf(page
, "%d\n", mddev
->layout
);
3879 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3884 err
= kstrtouint(buf
, 10, &n
);
3887 err
= mddev_lock(mddev
);
3892 if (mddev
->pers
->check_reshape
== NULL
)
3897 mddev
->new_layout
= n
;
3898 err
= mddev
->pers
->check_reshape(mddev
);
3900 mddev
->new_layout
= mddev
->layout
;
3903 mddev
->new_layout
= n
;
3904 if (mddev
->reshape_position
== MaxSector
)
3907 mddev_unlock(mddev
);
3910 static struct md_sysfs_entry md_layout
=
3911 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3914 raid_disks_show(struct mddev
*mddev
, char *page
)
3916 if (mddev
->raid_disks
== 0)
3918 if (mddev
->reshape_position
!= MaxSector
&&
3919 mddev
->delta_disks
!= 0)
3920 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3921 mddev
->raid_disks
- mddev
->delta_disks
);
3922 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3925 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3928 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3933 err
= kstrtouint(buf
, 10, &n
);
3937 err
= mddev_lock(mddev
);
3941 err
= update_raid_disks(mddev
, n
);
3942 else if (mddev
->reshape_position
!= MaxSector
) {
3943 struct md_rdev
*rdev
;
3944 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3947 rdev_for_each(rdev
, mddev
) {
3949 rdev
->data_offset
< rdev
->new_data_offset
)
3952 rdev
->data_offset
> rdev
->new_data_offset
)
3956 mddev
->delta_disks
= n
- olddisks
;
3957 mddev
->raid_disks
= n
;
3958 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3960 mddev
->raid_disks
= n
;
3962 mddev_unlock(mddev
);
3963 return err
? err
: len
;
3965 static struct md_sysfs_entry md_raid_disks
=
3966 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3969 chunk_size_show(struct mddev
*mddev
, char *page
)
3971 if (mddev
->reshape_position
!= MaxSector
&&
3972 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3973 return sprintf(page
, "%d (%d)\n",
3974 mddev
->new_chunk_sectors
<< 9,
3975 mddev
->chunk_sectors
<< 9);
3976 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3980 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3985 err
= kstrtoul(buf
, 10, &n
);
3989 err
= mddev_lock(mddev
);
3993 if (mddev
->pers
->check_reshape
== NULL
)
3998 mddev
->new_chunk_sectors
= n
>> 9;
3999 err
= mddev
->pers
->check_reshape(mddev
);
4001 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4004 mddev
->new_chunk_sectors
= n
>> 9;
4005 if (mddev
->reshape_position
== MaxSector
)
4006 mddev
->chunk_sectors
= n
>> 9;
4008 mddev_unlock(mddev
);
4011 static struct md_sysfs_entry md_chunk_size
=
4012 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
4015 resync_start_show(struct mddev
*mddev
, char *page
)
4017 if (mddev
->recovery_cp
== MaxSector
)
4018 return sprintf(page
, "none\n");
4019 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
4023 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4025 unsigned long long n
;
4028 if (cmd_match(buf
, "none"))
4031 err
= kstrtoull(buf
, 10, &n
);
4034 if (n
!= (sector_t
)n
)
4038 err
= mddev_lock(mddev
);
4041 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4045 mddev
->recovery_cp
= n
;
4047 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
4049 mddev_unlock(mddev
);
4052 static struct md_sysfs_entry md_resync_start
=
4053 __ATTR_PREALLOC(resync_start
, S_IRUGO
|S_IWUSR
,
4054 resync_start_show
, resync_start_store
);
4057 * The array state can be:
4060 * No devices, no size, no level
4061 * Equivalent to STOP_ARRAY ioctl
4063 * May have some settings, but array is not active
4064 * all IO results in error
4065 * When written, doesn't tear down array, but just stops it
4066 * suspended (not supported yet)
4067 * All IO requests will block. The array can be reconfigured.
4068 * Writing this, if accepted, will block until array is quiescent
4070 * no resync can happen. no superblocks get written.
4071 * write requests fail
4073 * like readonly, but behaves like 'clean' on a write request.
4075 * clean - no pending writes, but otherwise active.
4076 * When written to inactive array, starts without resync
4077 * If a write request arrives then
4078 * if metadata is known, mark 'dirty' and switch to 'active'.
4079 * if not known, block and switch to write-pending
4080 * If written to an active array that has pending writes, then fails.
4082 * fully active: IO and resync can be happening.
4083 * When written to inactive array, starts with resync
4086 * clean, but writes are blocked waiting for 'active' to be written.
4089 * like active, but no writes have been seen for a while (100msec).
4092 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
4093 write_pending
, active_idle
, bad_word
};
4094 static char *array_states
[] = {
4095 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4096 "write-pending", "active-idle", NULL
};
4098 static int match_word(const char *word
, char **list
)
4101 for (n
=0; list
[n
]; n
++)
4102 if (cmd_match(word
, list
[n
]))
4108 array_state_show(struct mddev
*mddev
, char *page
)
4110 enum array_state st
= inactive
;
4121 spin_lock(&mddev
->lock
);
4122 if (test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
))
4124 else if (mddev
->in_sync
)
4126 else if (mddev
->safemode
)
4130 spin_unlock(&mddev
->lock
);
4133 if (list_empty(&mddev
->disks
) &&
4134 mddev
->raid_disks
== 0 &&
4135 mddev
->dev_sectors
== 0)
4140 return sprintf(page
, "%s\n", array_states
[st
]);
4143 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
4144 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
4145 static int do_md_run(struct mddev
*mddev
);
4146 static int restart_array(struct mddev
*mddev
);
4149 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4152 enum array_state st
= match_word(buf
, array_states
);
4154 if (mddev
->pers
&& (st
== active
|| st
== clean
) && mddev
->ro
!= 1) {
4155 /* don't take reconfig_mutex when toggling between
4158 spin_lock(&mddev
->lock
);
4160 restart_array(mddev
);
4161 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
4162 md_wakeup_thread(mddev
->thread
);
4163 wake_up(&mddev
->sb_wait
);
4164 } else /* st == clean */ {
4165 restart_array(mddev
);
4166 if (!set_in_sync(mddev
))
4170 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4171 spin_unlock(&mddev
->lock
);
4174 err
= mddev_lock(mddev
);
4182 /* stopping an active array */
4183 err
= do_md_stop(mddev
, 0, NULL
);
4186 /* stopping an active array */
4188 err
= do_md_stop(mddev
, 2, NULL
);
4190 err
= 0; /* already inactive */
4193 break; /* not supported yet */
4196 err
= md_set_readonly(mddev
, NULL
);
4199 set_disk_ro(mddev
->gendisk
, 1);
4200 err
= do_md_run(mddev
);
4206 err
= md_set_readonly(mddev
, NULL
);
4207 else if (mddev
->ro
== 1)
4208 err
= restart_array(mddev
);
4211 set_disk_ro(mddev
->gendisk
, 0);
4215 err
= do_md_run(mddev
);
4220 err
= restart_array(mddev
);
4223 spin_lock(&mddev
->lock
);
4224 if (!set_in_sync(mddev
))
4226 spin_unlock(&mddev
->lock
);
4232 err
= restart_array(mddev
);
4235 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
4236 wake_up(&mddev
->sb_wait
);
4240 set_disk_ro(mddev
->gendisk
, 0);
4241 err
= do_md_run(mddev
);
4246 /* these cannot be set */
4251 if (mddev
->hold_active
== UNTIL_IOCTL
)
4252 mddev
->hold_active
= 0;
4253 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4255 mddev_unlock(mddev
);
4258 static struct md_sysfs_entry md_array_state
=
4259 __ATTR_PREALLOC(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
4262 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
4263 return sprintf(page
, "%d\n",
4264 atomic_read(&mddev
->max_corr_read_errors
));
4268 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4273 rv
= kstrtouint(buf
, 10, &n
);
4276 atomic_set(&mddev
->max_corr_read_errors
, n
);
4280 static struct md_sysfs_entry max_corr_read_errors
=
4281 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4282 max_corrected_read_errors_store
);
4285 null_show(struct mddev
*mddev
, char *page
)
4291 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4293 /* buf must be %d:%d\n? giving major and minor numbers */
4294 /* The new device is added to the array.
4295 * If the array has a persistent superblock, we read the
4296 * superblock to initialise info and check validity.
4297 * Otherwise, only checking done is that in bind_rdev_to_array,
4298 * which mainly checks size.
4301 int major
= simple_strtoul(buf
, &e
, 10);
4304 struct md_rdev
*rdev
;
4307 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4309 minor
= simple_strtoul(e
+1, &e
, 10);
4310 if (*e
&& *e
!= '\n')
4312 dev
= MKDEV(major
, minor
);
4313 if (major
!= MAJOR(dev
) ||
4314 minor
!= MINOR(dev
))
4317 flush_workqueue(md_misc_wq
);
4319 err
= mddev_lock(mddev
);
4322 if (mddev
->persistent
) {
4323 rdev
= md_import_device(dev
, mddev
->major_version
,
4324 mddev
->minor_version
);
4325 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4326 struct md_rdev
*rdev0
4327 = list_entry(mddev
->disks
.next
,
4328 struct md_rdev
, same_set
);
4329 err
= super_types
[mddev
->major_version
]
4330 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4334 } else if (mddev
->external
)
4335 rdev
= md_import_device(dev
, -2, -1);
4337 rdev
= md_import_device(dev
, -1, -1);
4340 mddev_unlock(mddev
);
4341 return PTR_ERR(rdev
);
4343 err
= bind_rdev_to_array(rdev
, mddev
);
4347 mddev_unlock(mddev
);
4349 md_new_event(mddev
);
4350 return err
? err
: len
;
4353 static struct md_sysfs_entry md_new_device
=
4354 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4357 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4360 unsigned long chunk
, end_chunk
;
4363 err
= mddev_lock(mddev
);
4368 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4370 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4371 if (buf
== end
) break;
4372 if (*end
== '-') { /* range */
4374 end_chunk
= simple_strtoul(buf
, &end
, 0);
4375 if (buf
== end
) break;
4377 if (*end
&& !isspace(*end
)) break;
4378 md_bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4379 buf
= skip_spaces(end
);
4381 md_bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4383 mddev_unlock(mddev
);
4387 static struct md_sysfs_entry md_bitmap
=
4388 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4391 size_show(struct mddev
*mddev
, char *page
)
4393 return sprintf(page
, "%llu\n",
4394 (unsigned long long)mddev
->dev_sectors
/ 2);
4397 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4400 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4402 /* If array is inactive, we can reduce the component size, but
4403 * not increase it (except from 0).
4404 * If array is active, we can try an on-line resize
4407 int err
= strict_blocks_to_sectors(buf
, §ors
);
4411 err
= mddev_lock(mddev
);
4415 err
= update_size(mddev
, sectors
);
4417 md_update_sb(mddev
, 1);
4419 if (mddev
->dev_sectors
== 0 ||
4420 mddev
->dev_sectors
> sectors
)
4421 mddev
->dev_sectors
= sectors
;
4425 mddev_unlock(mddev
);
4426 return err
? err
: len
;
4429 static struct md_sysfs_entry md_size
=
4430 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4432 /* Metadata version.
4434 * 'none' for arrays with no metadata (good luck...)
4435 * 'external' for arrays with externally managed metadata,
4436 * or N.M for internally known formats
4439 metadata_show(struct mddev
*mddev
, char *page
)
4441 if (mddev
->persistent
)
4442 return sprintf(page
, "%d.%d\n",
4443 mddev
->major_version
, mddev
->minor_version
);
4444 else if (mddev
->external
)
4445 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4447 return sprintf(page
, "none\n");
4451 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4456 /* Changing the details of 'external' metadata is
4457 * always permitted. Otherwise there must be
4458 * no devices attached to the array.
4461 err
= mddev_lock(mddev
);
4465 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4467 else if (!list_empty(&mddev
->disks
))
4471 if (cmd_match(buf
, "none")) {
4472 mddev
->persistent
= 0;
4473 mddev
->external
= 0;
4474 mddev
->major_version
= 0;
4475 mddev
->minor_version
= 90;
4478 if (strncmp(buf
, "external:", 9) == 0) {
4479 size_t namelen
= len
-9;
4480 if (namelen
>= sizeof(mddev
->metadata_type
))
4481 namelen
= sizeof(mddev
->metadata_type
)-1;
4482 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4483 mddev
->metadata_type
[namelen
] = 0;
4484 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4485 mddev
->metadata_type
[--namelen
] = 0;
4486 mddev
->persistent
= 0;
4487 mddev
->external
= 1;
4488 mddev
->major_version
= 0;
4489 mddev
->minor_version
= 90;
4492 major
= simple_strtoul(buf
, &e
, 10);
4494 if (e
==buf
|| *e
!= '.')
4497 minor
= simple_strtoul(buf
, &e
, 10);
4498 if (e
==buf
|| (*e
&& *e
!= '\n') )
4501 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4503 mddev
->major_version
= major
;
4504 mddev
->minor_version
= minor
;
4505 mddev
->persistent
= 1;
4506 mddev
->external
= 0;
4509 mddev_unlock(mddev
);
4513 static struct md_sysfs_entry md_metadata
=
4514 __ATTR_PREALLOC(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4517 action_show(struct mddev
*mddev
, char *page
)
4519 char *type
= "idle";
4520 unsigned long recovery
= mddev
->recovery
;
4521 if (test_bit(MD_RECOVERY_FROZEN
, &recovery
))
4523 else if (test_bit(MD_RECOVERY_RUNNING
, &recovery
) ||
4524 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &recovery
))) {
4525 if (test_bit(MD_RECOVERY_RESHAPE
, &recovery
))
4527 else if (test_bit(MD_RECOVERY_SYNC
, &recovery
)) {
4528 if (!test_bit(MD_RECOVERY_REQUESTED
, &recovery
))
4530 else if (test_bit(MD_RECOVERY_CHECK
, &recovery
))
4534 } else if (test_bit(MD_RECOVERY_RECOVER
, &recovery
))
4536 else if (mddev
->reshape_position
!= MaxSector
)
4539 return sprintf(page
, "%s\n", type
);
4543 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4545 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4549 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4550 if (cmd_match(page
, "frozen"))
4551 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4553 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4554 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4555 mddev_lock(mddev
) == 0) {
4556 flush_workqueue(md_misc_wq
);
4557 if (mddev
->sync_thread
) {
4558 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4559 md_reap_sync_thread(mddev
);
4561 mddev_unlock(mddev
);
4563 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4565 else if (cmd_match(page
, "resync"))
4566 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4567 else if (cmd_match(page
, "recover")) {
4568 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4569 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4570 } else if (cmd_match(page
, "reshape")) {
4572 if (mddev
->pers
->start_reshape
== NULL
)
4574 err
= mddev_lock(mddev
);
4576 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4579 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4580 err
= mddev
->pers
->start_reshape(mddev
);
4582 mddev_unlock(mddev
);
4586 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4588 if (cmd_match(page
, "check"))
4589 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4590 else if (!cmd_match(page
, "repair"))
4592 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4593 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4594 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4596 if (mddev
->ro
== 2) {
4597 /* A write to sync_action is enough to justify
4598 * canceling read-auto mode
4601 md_wakeup_thread(mddev
->sync_thread
);
4603 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4604 md_wakeup_thread(mddev
->thread
);
4605 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4609 static struct md_sysfs_entry md_scan_mode
=
4610 __ATTR_PREALLOC(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4613 last_sync_action_show(struct mddev
*mddev
, char *page
)
4615 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4618 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4621 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4623 return sprintf(page
, "%llu\n",
4624 (unsigned long long)
4625 atomic64_read(&mddev
->resync_mismatches
));
4628 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4631 sync_min_show(struct mddev
*mddev
, char *page
)
4633 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4634 mddev
->sync_speed_min
? "local": "system");
4638 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4643 if (strncmp(buf
, "system", 6)==0) {
4646 rv
= kstrtouint(buf
, 10, &min
);
4652 mddev
->sync_speed_min
= min
;
4656 static struct md_sysfs_entry md_sync_min
=
4657 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4660 sync_max_show(struct mddev
*mddev
, char *page
)
4662 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4663 mddev
->sync_speed_max
? "local": "system");
4667 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4672 if (strncmp(buf
, "system", 6)==0) {
4675 rv
= kstrtouint(buf
, 10, &max
);
4681 mddev
->sync_speed_max
= max
;
4685 static struct md_sysfs_entry md_sync_max
=
4686 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4689 degraded_show(struct mddev
*mddev
, char *page
)
4691 return sprintf(page
, "%d\n", mddev
->degraded
);
4693 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4696 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4698 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4702 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4706 if (kstrtol(buf
, 10, &n
))
4709 if (n
!= 0 && n
!= 1)
4712 mddev
->parallel_resync
= n
;
4714 if (mddev
->sync_thread
)
4715 wake_up(&resync_wait
);
4720 /* force parallel resync, even with shared block devices */
4721 static struct md_sysfs_entry md_sync_force_parallel
=
4722 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4723 sync_force_parallel_show
, sync_force_parallel_store
);
4726 sync_speed_show(struct mddev
*mddev
, char *page
)
4728 unsigned long resync
, dt
, db
;
4729 if (mddev
->curr_resync
== 0)
4730 return sprintf(page
, "none\n");
4731 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4732 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4734 db
= resync
- mddev
->resync_mark_cnt
;
4735 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4738 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4741 sync_completed_show(struct mddev
*mddev
, char *page
)
4743 unsigned long long max_sectors
, resync
;
4745 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4746 return sprintf(page
, "none\n");
4748 if (mddev
->curr_resync
== 1 ||
4749 mddev
->curr_resync
== 2)
4750 return sprintf(page
, "delayed\n");
4752 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4753 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4754 max_sectors
= mddev
->resync_max_sectors
;
4756 max_sectors
= mddev
->dev_sectors
;
4758 resync
= mddev
->curr_resync_completed
;
4759 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4762 static struct md_sysfs_entry md_sync_completed
=
4763 __ATTR_PREALLOC(sync_completed
, S_IRUGO
, sync_completed_show
, NULL
);
4766 min_sync_show(struct mddev
*mddev
, char *page
)
4768 return sprintf(page
, "%llu\n",
4769 (unsigned long long)mddev
->resync_min
);
4772 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4774 unsigned long long min
;
4777 if (kstrtoull(buf
, 10, &min
))
4780 spin_lock(&mddev
->lock
);
4782 if (min
> mddev
->resync_max
)
4786 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4789 /* Round down to multiple of 4K for safety */
4790 mddev
->resync_min
= round_down(min
, 8);
4794 spin_unlock(&mddev
->lock
);
4798 static struct md_sysfs_entry md_min_sync
=
4799 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4802 max_sync_show(struct mddev
*mddev
, char *page
)
4804 if (mddev
->resync_max
== MaxSector
)
4805 return sprintf(page
, "max\n");
4807 return sprintf(page
, "%llu\n",
4808 (unsigned long long)mddev
->resync_max
);
4811 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4814 spin_lock(&mddev
->lock
);
4815 if (strncmp(buf
, "max", 3) == 0)
4816 mddev
->resync_max
= MaxSector
;
4818 unsigned long long max
;
4822 if (kstrtoull(buf
, 10, &max
))
4824 if (max
< mddev
->resync_min
)
4828 if (max
< mddev
->resync_max
&&
4830 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4833 /* Must be a multiple of chunk_size */
4834 chunk
= mddev
->chunk_sectors
;
4836 sector_t temp
= max
;
4839 if (sector_div(temp
, chunk
))
4842 mddev
->resync_max
= max
;
4844 wake_up(&mddev
->recovery_wait
);
4847 spin_unlock(&mddev
->lock
);
4851 static struct md_sysfs_entry md_max_sync
=
4852 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4855 suspend_lo_show(struct mddev
*mddev
, char *page
)
4857 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4861 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4863 unsigned long long new;
4866 err
= kstrtoull(buf
, 10, &new);
4869 if (new != (sector_t
)new)
4872 err
= mddev_lock(mddev
);
4876 if (mddev
->pers
== NULL
||
4877 mddev
->pers
->quiesce
== NULL
)
4879 mddev_suspend(mddev
);
4880 mddev
->suspend_lo
= new;
4881 mddev_resume(mddev
);
4885 mddev_unlock(mddev
);
4888 static struct md_sysfs_entry md_suspend_lo
=
4889 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4892 suspend_hi_show(struct mddev
*mddev
, char *page
)
4894 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4898 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4900 unsigned long long new;
4903 err
= kstrtoull(buf
, 10, &new);
4906 if (new != (sector_t
)new)
4909 err
= mddev_lock(mddev
);
4913 if (mddev
->pers
== NULL
)
4916 mddev_suspend(mddev
);
4917 mddev
->suspend_hi
= new;
4918 mddev_resume(mddev
);
4922 mddev_unlock(mddev
);
4925 static struct md_sysfs_entry md_suspend_hi
=
4926 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4929 reshape_position_show(struct mddev
*mddev
, char *page
)
4931 if (mddev
->reshape_position
!= MaxSector
)
4932 return sprintf(page
, "%llu\n",
4933 (unsigned long long)mddev
->reshape_position
);
4934 strcpy(page
, "none\n");
4939 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4941 struct md_rdev
*rdev
;
4942 unsigned long long new;
4945 err
= kstrtoull(buf
, 10, &new);
4948 if (new != (sector_t
)new)
4950 err
= mddev_lock(mddev
);
4956 mddev
->reshape_position
= new;
4957 mddev
->delta_disks
= 0;
4958 mddev
->reshape_backwards
= 0;
4959 mddev
->new_level
= mddev
->level
;
4960 mddev
->new_layout
= mddev
->layout
;
4961 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4962 rdev_for_each(rdev
, mddev
)
4963 rdev
->new_data_offset
= rdev
->data_offset
;
4966 mddev_unlock(mddev
);
4970 static struct md_sysfs_entry md_reshape_position
=
4971 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4972 reshape_position_store
);
4975 reshape_direction_show(struct mddev
*mddev
, char *page
)
4977 return sprintf(page
, "%s\n",
4978 mddev
->reshape_backwards
? "backwards" : "forwards");
4982 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4987 if (cmd_match(buf
, "forwards"))
4989 else if (cmd_match(buf
, "backwards"))
4993 if (mddev
->reshape_backwards
== backwards
)
4996 err
= mddev_lock(mddev
);
4999 /* check if we are allowed to change */
5000 if (mddev
->delta_disks
)
5002 else if (mddev
->persistent
&&
5003 mddev
->major_version
== 0)
5006 mddev
->reshape_backwards
= backwards
;
5007 mddev_unlock(mddev
);
5011 static struct md_sysfs_entry md_reshape_direction
=
5012 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
5013 reshape_direction_store
);
5016 array_size_show(struct mddev
*mddev
, char *page
)
5018 if (mddev
->external_size
)
5019 return sprintf(page
, "%llu\n",
5020 (unsigned long long)mddev
->array_sectors
/2);
5022 return sprintf(page
, "default\n");
5026 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
5031 err
= mddev_lock(mddev
);
5035 /* cluster raid doesn't support change array_sectors */
5036 if (mddev_is_clustered(mddev
)) {
5037 mddev_unlock(mddev
);
5041 if (strncmp(buf
, "default", 7) == 0) {
5043 sectors
= mddev
->pers
->size(mddev
, 0, 0);
5045 sectors
= mddev
->array_sectors
;
5047 mddev
->external_size
= 0;
5049 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
5051 else if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
5054 mddev
->external_size
= 1;
5058 mddev
->array_sectors
= sectors
;
5060 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5061 revalidate_disk(mddev
->gendisk
);
5064 mddev_unlock(mddev
);
5068 static struct md_sysfs_entry md_array_size
=
5069 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
5073 consistency_policy_show(struct mddev
*mddev
, char *page
)
5077 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
)) {
5078 ret
= sprintf(page
, "journal\n");
5079 } else if (test_bit(MD_HAS_PPL
, &mddev
->flags
)) {
5080 ret
= sprintf(page
, "ppl\n");
5081 } else if (mddev
->bitmap
) {
5082 ret
= sprintf(page
, "bitmap\n");
5083 } else if (mddev
->pers
) {
5084 if (mddev
->pers
->sync_request
)
5085 ret
= sprintf(page
, "resync\n");
5087 ret
= sprintf(page
, "none\n");
5089 ret
= sprintf(page
, "unknown\n");
5096 consistency_policy_store(struct mddev
*mddev
, const char *buf
, size_t len
)
5101 if (mddev
->pers
->change_consistency_policy
)
5102 err
= mddev
->pers
->change_consistency_policy(mddev
, buf
);
5105 } else if (mddev
->external
&& strncmp(buf
, "ppl", 3) == 0) {
5106 set_bit(MD_HAS_PPL
, &mddev
->flags
);
5111 return err
? err
: len
;
5114 static struct md_sysfs_entry md_consistency_policy
=
5115 __ATTR(consistency_policy
, S_IRUGO
| S_IWUSR
, consistency_policy_show
,
5116 consistency_policy_store
);
5118 static struct attribute
*md_default_attrs
[] = {
5121 &md_raid_disks
.attr
,
5122 &md_chunk_size
.attr
,
5124 &md_resync_start
.attr
,
5126 &md_new_device
.attr
,
5127 &md_safe_delay
.attr
,
5128 &md_array_state
.attr
,
5129 &md_reshape_position
.attr
,
5130 &md_reshape_direction
.attr
,
5131 &md_array_size
.attr
,
5132 &max_corr_read_errors
.attr
,
5133 &md_consistency_policy
.attr
,
5137 static struct attribute
*md_redundancy_attrs
[] = {
5139 &md_last_scan_mode
.attr
,
5140 &md_mismatches
.attr
,
5143 &md_sync_speed
.attr
,
5144 &md_sync_force_parallel
.attr
,
5145 &md_sync_completed
.attr
,
5148 &md_suspend_lo
.attr
,
5149 &md_suspend_hi
.attr
,
5154 static struct attribute_group md_redundancy_group
= {
5156 .attrs
= md_redundancy_attrs
,
5160 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
5162 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
5163 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
5168 spin_lock(&all_mddevs_lock
);
5169 if (list_empty(&mddev
->all_mddevs
)) {
5170 spin_unlock(&all_mddevs_lock
);
5174 spin_unlock(&all_mddevs_lock
);
5176 rv
= entry
->show(mddev
, page
);
5182 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
5183 const char *page
, size_t length
)
5185 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
5186 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
5191 if (!capable(CAP_SYS_ADMIN
))
5193 spin_lock(&all_mddevs_lock
);
5194 if (list_empty(&mddev
->all_mddevs
)) {
5195 spin_unlock(&all_mddevs_lock
);
5199 spin_unlock(&all_mddevs_lock
);
5200 rv
= entry
->store(mddev
, page
, length
);
5205 static void md_free(struct kobject
*ko
)
5207 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
5209 if (mddev
->sysfs_state
)
5210 sysfs_put(mddev
->sysfs_state
);
5213 del_gendisk(mddev
->gendisk
);
5215 blk_cleanup_queue(mddev
->queue
);
5217 put_disk(mddev
->gendisk
);
5218 percpu_ref_exit(&mddev
->writes_pending
);
5220 bioset_exit(&mddev
->bio_set
);
5221 bioset_exit(&mddev
->sync_set
);
5225 static const struct sysfs_ops md_sysfs_ops
= {
5226 .show
= md_attr_show
,
5227 .store
= md_attr_store
,
5229 static struct kobj_type md_ktype
= {
5231 .sysfs_ops
= &md_sysfs_ops
,
5232 .default_attrs
= md_default_attrs
,
5237 static void mddev_delayed_delete(struct work_struct
*ws
)
5239 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
5241 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
5242 kobject_del(&mddev
->kobj
);
5243 kobject_put(&mddev
->kobj
);
5246 static void no_op(struct percpu_ref
*r
) {}
5248 int mddev_init_writes_pending(struct mddev
*mddev
)
5250 if (mddev
->writes_pending
.percpu_count_ptr
)
5252 if (percpu_ref_init(&mddev
->writes_pending
, no_op
, 0, GFP_KERNEL
) < 0)
5254 /* We want to start with the refcount at zero */
5255 percpu_ref_put(&mddev
->writes_pending
);
5258 EXPORT_SYMBOL_GPL(mddev_init_writes_pending
);
5260 static int md_alloc(dev_t dev
, char *name
)
5263 * If dev is zero, name is the name of a device to allocate with
5264 * an arbitrary minor number. It will be "md_???"
5265 * If dev is non-zero it must be a device number with a MAJOR of
5266 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5267 * the device is being created by opening a node in /dev.
5268 * If "name" is not NULL, the device is being created by
5269 * writing to /sys/module/md_mod/parameters/new_array.
5271 static DEFINE_MUTEX(disks_mutex
);
5272 struct mddev
*mddev
= mddev_find(dev
);
5273 struct gendisk
*disk
;
5282 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
5283 shift
= partitioned
? MdpMinorShift
: 0;
5284 unit
= MINOR(mddev
->unit
) >> shift
;
5286 /* wait for any previous instance of this device to be
5287 * completely removed (mddev_delayed_delete).
5289 flush_workqueue(md_misc_wq
);
5291 mutex_lock(&disks_mutex
);
5297 /* Need to ensure that 'name' is not a duplicate.
5299 struct mddev
*mddev2
;
5300 spin_lock(&all_mddevs_lock
);
5302 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
5303 if (mddev2
->gendisk
&&
5304 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
5305 spin_unlock(&all_mddevs_lock
);
5308 spin_unlock(&all_mddevs_lock
);
5312 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5314 mddev
->hold_active
= UNTIL_STOP
;
5317 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
5320 mddev
->queue
->queuedata
= mddev
;
5322 blk_queue_make_request(mddev
->queue
, md_make_request
);
5323 blk_set_stacking_limits(&mddev
->queue
->limits
);
5325 disk
= alloc_disk(1 << shift
);
5327 blk_cleanup_queue(mddev
->queue
);
5328 mddev
->queue
= NULL
;
5331 disk
->major
= MAJOR(mddev
->unit
);
5332 disk
->first_minor
= unit
<< shift
;
5334 strcpy(disk
->disk_name
, name
);
5335 else if (partitioned
)
5336 sprintf(disk
->disk_name
, "md_d%d", unit
);
5338 sprintf(disk
->disk_name
, "md%d", unit
);
5339 disk
->fops
= &md_fops
;
5340 disk
->private_data
= mddev
;
5341 disk
->queue
= mddev
->queue
;
5342 blk_queue_write_cache(mddev
->queue
, true, true);
5343 /* Allow extended partitions. This makes the
5344 * 'mdp' device redundant, but we can't really
5347 disk
->flags
|= GENHD_FL_EXT_DEVT
;
5348 mddev
->gendisk
= disk
;
5349 /* As soon as we call add_disk(), another thread could get
5350 * through to md_open, so make sure it doesn't get too far
5352 mutex_lock(&mddev
->open_mutex
);
5355 error
= kobject_add(&mddev
->kobj
, &disk_to_dev(disk
)->kobj
, "%s", "md");
5357 /* This isn't possible, but as kobject_init_and_add is marked
5358 * __must_check, we must do something with the result
5360 pr_debug("md: cannot register %s/md - name in use\n",
5364 if (mddev
->kobj
.sd
&&
5365 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
5366 pr_debug("pointless warning\n");
5367 mutex_unlock(&mddev
->open_mutex
);
5369 mutex_unlock(&disks_mutex
);
5370 if (!error
&& mddev
->kobj
.sd
) {
5371 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
5372 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
5378 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
5381 md_alloc(dev
, NULL
);
5385 static int add_named_array(const char *val
, const struct kernel_param
*kp
)
5388 * val must be "md_*" or "mdNNN".
5389 * For "md_*" we allocate an array with a large free minor number, and
5390 * set the name to val. val must not already be an active name.
5391 * For "mdNNN" we allocate an array with the minor number NNN
5392 * which must not already be in use.
5394 int len
= strlen(val
);
5395 char buf
[DISK_NAME_LEN
];
5396 unsigned long devnum
;
5398 while (len
&& val
[len
-1] == '\n')
5400 if (len
>= DISK_NAME_LEN
)
5402 strlcpy(buf
, val
, len
+1);
5403 if (strncmp(buf
, "md_", 3) == 0)
5404 return md_alloc(0, buf
);
5405 if (strncmp(buf
, "md", 2) == 0 &&
5407 kstrtoul(buf
+2, 10, &devnum
) == 0 &&
5408 devnum
<= MINORMASK
)
5409 return md_alloc(MKDEV(MD_MAJOR
, devnum
), NULL
);
5414 static void md_safemode_timeout(struct timer_list
*t
)
5416 struct mddev
*mddev
= from_timer(mddev
, t
, safemode_timer
);
5418 mddev
->safemode
= 1;
5419 if (mddev
->external
)
5420 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5422 md_wakeup_thread(mddev
->thread
);
5425 static int start_dirty_degraded
;
5427 int md_run(struct mddev
*mddev
)
5430 struct md_rdev
*rdev
;
5431 struct md_personality
*pers
;
5433 if (list_empty(&mddev
->disks
))
5434 /* cannot run an array with no devices.. */
5439 /* Cannot run until previous stop completes properly */
5440 if (mddev
->sysfs_active
)
5444 * Analyze all RAID superblock(s)
5446 if (!mddev
->raid_disks
) {
5447 if (!mddev
->persistent
)
5452 if (mddev
->level
!= LEVEL_NONE
)
5453 request_module("md-level-%d", mddev
->level
);
5454 else if (mddev
->clevel
[0])
5455 request_module("md-%s", mddev
->clevel
);
5458 * Drop all container device buffers, from now on
5459 * the only valid external interface is through the md
5462 mddev
->has_superblocks
= false;
5463 rdev_for_each(rdev
, mddev
) {
5464 if (test_bit(Faulty
, &rdev
->flags
))
5466 sync_blockdev(rdev
->bdev
);
5467 invalidate_bdev(rdev
->bdev
);
5468 if (mddev
->ro
!= 1 &&
5469 (bdev_read_only(rdev
->bdev
) ||
5470 bdev_read_only(rdev
->meta_bdev
))) {
5473 set_disk_ro(mddev
->gendisk
, 1);
5477 mddev
->has_superblocks
= true;
5479 /* perform some consistency tests on the device.
5480 * We don't want the data to overlap the metadata,
5481 * Internal Bitmap issues have been handled elsewhere.
5483 if (rdev
->meta_bdev
) {
5484 /* Nothing to check */;
5485 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5486 if (mddev
->dev_sectors
&&
5487 rdev
->data_offset
+ mddev
->dev_sectors
5489 pr_warn("md: %s: data overlaps metadata\n",
5494 if (rdev
->sb_start
+ rdev
->sb_size
/512
5495 > rdev
->data_offset
) {
5496 pr_warn("md: %s: metadata overlaps data\n",
5501 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5504 if (!bioset_initialized(&mddev
->bio_set
)) {
5505 err
= bioset_init(&mddev
->bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
5509 if (!bioset_initialized(&mddev
->sync_set
)) {
5510 err
= bioset_init(&mddev
->sync_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
5514 if (mddev
->flush_pool
== NULL
) {
5515 mddev
->flush_pool
= mempool_create(NR_FLUSH_INFOS
, flush_info_alloc
,
5516 flush_info_free
, mddev
);
5517 if (!mddev
->flush_pool
) {
5522 if (mddev
->flush_bio_pool
== NULL
) {
5523 mddev
->flush_bio_pool
= mempool_create(NR_FLUSH_BIOS
, flush_bio_alloc
,
5524 flush_bio_free
, mddev
);
5525 if (!mddev
->flush_bio_pool
) {
5531 spin_lock(&pers_lock
);
5532 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5533 if (!pers
|| !try_module_get(pers
->owner
)) {
5534 spin_unlock(&pers_lock
);
5535 if (mddev
->level
!= LEVEL_NONE
)
5536 pr_warn("md: personality for level %d is not loaded!\n",
5539 pr_warn("md: personality for level %s is not loaded!\n",
5544 spin_unlock(&pers_lock
);
5545 if (mddev
->level
!= pers
->level
) {
5546 mddev
->level
= pers
->level
;
5547 mddev
->new_level
= pers
->level
;
5549 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5551 if (mddev
->reshape_position
!= MaxSector
&&
5552 pers
->start_reshape
== NULL
) {
5553 /* This personality cannot handle reshaping... */
5554 module_put(pers
->owner
);
5559 if (pers
->sync_request
) {
5560 /* Warn if this is a potentially silly
5563 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5564 struct md_rdev
*rdev2
;
5567 rdev_for_each(rdev
, mddev
)
5568 rdev_for_each(rdev2
, mddev
) {
5570 rdev
->bdev
->bd_contains
==
5571 rdev2
->bdev
->bd_contains
) {
5572 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5574 bdevname(rdev
->bdev
,b
),
5575 bdevname(rdev2
->bdev
,b2
));
5581 pr_warn("True protection against single-disk failure might be compromised.\n");
5584 mddev
->recovery
= 0;
5585 /* may be over-ridden by personality */
5586 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5588 mddev
->ok_start_degraded
= start_dirty_degraded
;
5590 if (start_readonly
&& mddev
->ro
== 0)
5591 mddev
->ro
= 2; /* read-only, but switch on first write */
5593 err
= pers
->run(mddev
);
5595 pr_warn("md: pers->run() failed ...\n");
5596 else if (pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5597 WARN_ONCE(!mddev
->external_size
,
5598 "%s: default size too small, but 'external_size' not in effect?\n",
5600 pr_warn("md: invalid array_size %llu > default size %llu\n",
5601 (unsigned long long)mddev
->array_sectors
/ 2,
5602 (unsigned long long)pers
->size(mddev
, 0, 0) / 2);
5605 if (err
== 0 && pers
->sync_request
&&
5606 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5607 struct bitmap
*bitmap
;
5609 bitmap
= md_bitmap_create(mddev
, -1);
5610 if (IS_ERR(bitmap
)) {
5611 err
= PTR_ERR(bitmap
);
5612 pr_warn("%s: failed to create bitmap (%d)\n",
5613 mdname(mddev
), err
);
5615 mddev
->bitmap
= bitmap
;
5619 mddev_detach(mddev
);
5621 pers
->free(mddev
, mddev
->private);
5622 mddev
->private = NULL
;
5623 module_put(pers
->owner
);
5624 md_bitmap_destroy(mddev
);
5630 rdev_for_each(rdev
, mddev
) {
5631 if (rdev
->raid_disk
>= 0 &&
5632 !blk_queue_nonrot(bdev_get_queue(rdev
->bdev
))) {
5637 if (mddev
->degraded
)
5640 blk_queue_flag_set(QUEUE_FLAG_NONROT
, mddev
->queue
);
5642 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, mddev
->queue
);
5643 mddev
->queue
->backing_dev_info
->congested_data
= mddev
;
5644 mddev
->queue
->backing_dev_info
->congested_fn
= md_congested
;
5646 if (pers
->sync_request
) {
5647 if (mddev
->kobj
.sd
&&
5648 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5649 pr_warn("md: cannot register extra attributes for %s\n",
5651 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5652 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5655 atomic_set(&mddev
->max_corr_read_errors
,
5656 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5657 mddev
->safemode
= 0;
5658 if (mddev_is_clustered(mddev
))
5659 mddev
->safemode_delay
= 0;
5661 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5664 spin_lock(&mddev
->lock
);
5666 spin_unlock(&mddev
->lock
);
5667 rdev_for_each(rdev
, mddev
)
5668 if (rdev
->raid_disk
>= 0)
5669 if (sysfs_link_rdev(mddev
, rdev
))
5670 /* failure here is OK */;
5672 if (mddev
->degraded
&& !mddev
->ro
)
5673 /* This ensures that recovering status is reported immediately
5674 * via sysfs - until a lack of spares is confirmed.
5676 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5677 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5679 if (mddev
->sb_flags
)
5680 md_update_sb(mddev
, 0);
5682 md_new_event(mddev
);
5683 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5684 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5685 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5689 mempool_destroy(mddev
->flush_bio_pool
);
5690 mddev
->flush_bio_pool
= NULL
;
5691 mempool_destroy(mddev
->flush_pool
);
5692 mddev
->flush_pool
= NULL
;
5696 EXPORT_SYMBOL_GPL(md_run
);
5698 static int do_md_run(struct mddev
*mddev
)
5702 err
= md_run(mddev
);
5705 err
= md_bitmap_load(mddev
);
5707 md_bitmap_destroy(mddev
);
5711 if (mddev_is_clustered(mddev
))
5712 md_allow_write(mddev
);
5714 /* run start up tasks that require md_thread */
5717 md_wakeup_thread(mddev
->thread
);
5718 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5720 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5721 revalidate_disk(mddev
->gendisk
);
5723 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5728 int md_start(struct mddev
*mddev
)
5732 if (mddev
->pers
->start
) {
5733 set_bit(MD_RECOVERY_WAIT
, &mddev
->recovery
);
5734 md_wakeup_thread(mddev
->thread
);
5735 ret
= mddev
->pers
->start(mddev
);
5736 clear_bit(MD_RECOVERY_WAIT
, &mddev
->recovery
);
5737 md_wakeup_thread(mddev
->sync_thread
);
5741 EXPORT_SYMBOL_GPL(md_start
);
5743 static int restart_array(struct mddev
*mddev
)
5745 struct gendisk
*disk
= mddev
->gendisk
;
5746 struct md_rdev
*rdev
;
5747 bool has_journal
= false;
5748 bool has_readonly
= false;
5750 /* Complain if it has no devices */
5751 if (list_empty(&mddev
->disks
))
5759 rdev_for_each_rcu(rdev
, mddev
) {
5760 if (test_bit(Journal
, &rdev
->flags
) &&
5761 !test_bit(Faulty
, &rdev
->flags
))
5763 if (bdev_read_only(rdev
->bdev
))
5764 has_readonly
= true;
5767 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
) && !has_journal
)
5768 /* Don't restart rw with journal missing/faulty */
5773 mddev
->safemode
= 0;
5775 set_disk_ro(disk
, 0);
5776 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev
));
5777 /* Kick recovery or resync if necessary */
5778 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5779 md_wakeup_thread(mddev
->thread
);
5780 md_wakeup_thread(mddev
->sync_thread
);
5781 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5785 static void md_clean(struct mddev
*mddev
)
5787 mddev
->array_sectors
= 0;
5788 mddev
->external_size
= 0;
5789 mddev
->dev_sectors
= 0;
5790 mddev
->raid_disks
= 0;
5791 mddev
->recovery_cp
= 0;
5792 mddev
->resync_min
= 0;
5793 mddev
->resync_max
= MaxSector
;
5794 mddev
->reshape_position
= MaxSector
;
5795 mddev
->external
= 0;
5796 mddev
->persistent
= 0;
5797 mddev
->level
= LEVEL_NONE
;
5798 mddev
->clevel
[0] = 0;
5800 mddev
->sb_flags
= 0;
5802 mddev
->metadata_type
[0] = 0;
5803 mddev
->chunk_sectors
= 0;
5804 mddev
->ctime
= mddev
->utime
= 0;
5806 mddev
->max_disks
= 0;
5808 mddev
->can_decrease_events
= 0;
5809 mddev
->delta_disks
= 0;
5810 mddev
->reshape_backwards
= 0;
5811 mddev
->new_level
= LEVEL_NONE
;
5812 mddev
->new_layout
= 0;
5813 mddev
->new_chunk_sectors
= 0;
5814 mddev
->curr_resync
= 0;
5815 atomic64_set(&mddev
->resync_mismatches
, 0);
5816 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5817 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5818 mddev
->recovery
= 0;
5821 mddev
->degraded
= 0;
5822 mddev
->safemode
= 0;
5823 mddev
->private = NULL
;
5824 mddev
->cluster_info
= NULL
;
5825 mddev
->bitmap_info
.offset
= 0;
5826 mddev
->bitmap_info
.default_offset
= 0;
5827 mddev
->bitmap_info
.default_space
= 0;
5828 mddev
->bitmap_info
.chunksize
= 0;
5829 mddev
->bitmap_info
.daemon_sleep
= 0;
5830 mddev
->bitmap_info
.max_write_behind
= 0;
5831 mddev
->bitmap_info
.nodes
= 0;
5834 static void __md_stop_writes(struct mddev
*mddev
)
5836 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5837 flush_workqueue(md_misc_wq
);
5838 if (mddev
->sync_thread
) {
5839 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5840 md_reap_sync_thread(mddev
);
5843 del_timer_sync(&mddev
->safemode_timer
);
5845 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5846 mddev
->pers
->quiesce(mddev
, 1);
5847 mddev
->pers
->quiesce(mddev
, 0);
5849 md_bitmap_flush(mddev
);
5851 if (mddev
->ro
== 0 &&
5852 ((!mddev
->in_sync
&& !mddev_is_clustered(mddev
)) ||
5854 /* mark array as shutdown cleanly */
5855 if (!mddev_is_clustered(mddev
))
5857 md_update_sb(mddev
, 1);
5861 void md_stop_writes(struct mddev
*mddev
)
5863 mddev_lock_nointr(mddev
);
5864 __md_stop_writes(mddev
);
5865 mddev_unlock(mddev
);
5867 EXPORT_SYMBOL_GPL(md_stop_writes
);
5869 static void mddev_detach(struct mddev
*mddev
)
5871 md_bitmap_wait_behind_writes(mddev
);
5872 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5873 mddev
->pers
->quiesce(mddev
, 1);
5874 mddev
->pers
->quiesce(mddev
, 0);
5876 md_unregister_thread(&mddev
->thread
);
5878 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
5881 static void __md_stop(struct mddev
*mddev
)
5883 struct md_personality
*pers
= mddev
->pers
;
5884 md_bitmap_destroy(mddev
);
5885 mddev_detach(mddev
);
5886 /* Ensure ->event_work is done */
5887 flush_workqueue(md_misc_wq
);
5888 spin_lock(&mddev
->lock
);
5890 spin_unlock(&mddev
->lock
);
5891 pers
->free(mddev
, mddev
->private);
5892 mddev
->private = NULL
;
5893 if (pers
->sync_request
&& mddev
->to_remove
== NULL
)
5894 mddev
->to_remove
= &md_redundancy_group
;
5895 module_put(pers
->owner
);
5896 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5897 if (mddev
->flush_bio_pool
) {
5898 mempool_destroy(mddev
->flush_bio_pool
);
5899 mddev
->flush_bio_pool
= NULL
;
5901 if (mddev
->flush_pool
) {
5902 mempool_destroy(mddev
->flush_pool
);
5903 mddev
->flush_pool
= NULL
;
5907 void md_stop(struct mddev
*mddev
)
5909 /* stop the array and free an attached data structures.
5910 * This is called from dm-raid
5913 bioset_exit(&mddev
->bio_set
);
5914 bioset_exit(&mddev
->sync_set
);
5917 EXPORT_SYMBOL_GPL(md_stop
);
5919 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5924 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5926 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5927 md_wakeup_thread(mddev
->thread
);
5929 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5930 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5931 if (mddev
->sync_thread
)
5932 /* Thread might be blocked waiting for metadata update
5933 * which will now never happen */
5934 wake_up_process(mddev
->sync_thread
->tsk
);
5936 if (mddev
->external
&& test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
))
5938 mddev_unlock(mddev
);
5939 wait_event(resync_wait
, !test_bit(MD_RECOVERY_RUNNING
,
5941 wait_event(mddev
->sb_wait
,
5942 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
5943 mddev_lock_nointr(mddev
);
5945 mutex_lock(&mddev
->open_mutex
);
5946 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5947 mddev
->sync_thread
||
5948 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
)) {
5949 pr_warn("md: %s still in use.\n",mdname(mddev
));
5951 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5952 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5953 md_wakeup_thread(mddev
->thread
);
5959 __md_stop_writes(mddev
);
5965 set_disk_ro(mddev
->gendisk
, 1);
5966 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5967 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5968 md_wakeup_thread(mddev
->thread
);
5969 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5973 mutex_unlock(&mddev
->open_mutex
);
5978 * 0 - completely stop and dis-assemble array
5979 * 2 - stop but do not disassemble array
5981 static int do_md_stop(struct mddev
*mddev
, int mode
,
5982 struct block_device
*bdev
)
5984 struct gendisk
*disk
= mddev
->gendisk
;
5985 struct md_rdev
*rdev
;
5988 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5990 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5991 md_wakeup_thread(mddev
->thread
);
5993 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5994 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5995 if (mddev
->sync_thread
)
5996 /* Thread might be blocked waiting for metadata update
5997 * which will now never happen */
5998 wake_up_process(mddev
->sync_thread
->tsk
);
6000 mddev_unlock(mddev
);
6001 wait_event(resync_wait
, (mddev
->sync_thread
== NULL
&&
6002 !test_bit(MD_RECOVERY_RUNNING
,
6003 &mddev
->recovery
)));
6004 mddev_lock_nointr(mddev
);
6006 mutex_lock(&mddev
->open_mutex
);
6007 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
6008 mddev
->sysfs_active
||
6009 mddev
->sync_thread
||
6010 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
)) {
6011 pr_warn("md: %s still in use.\n",mdname(mddev
));
6012 mutex_unlock(&mddev
->open_mutex
);
6014 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
6015 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6016 md_wakeup_thread(mddev
->thread
);
6022 set_disk_ro(disk
, 0);
6024 __md_stop_writes(mddev
);
6026 mddev
->queue
->backing_dev_info
->congested_fn
= NULL
;
6028 /* tell userspace to handle 'inactive' */
6029 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6031 rdev_for_each(rdev
, mddev
)
6032 if (rdev
->raid_disk
>= 0)
6033 sysfs_unlink_rdev(mddev
, rdev
);
6035 set_capacity(disk
, 0);
6036 mutex_unlock(&mddev
->open_mutex
);
6038 revalidate_disk(disk
);
6043 mutex_unlock(&mddev
->open_mutex
);
6045 * Free resources if final stop
6048 pr_info("md: %s stopped.\n", mdname(mddev
));
6050 if (mddev
->bitmap_info
.file
) {
6051 struct file
*f
= mddev
->bitmap_info
.file
;
6052 spin_lock(&mddev
->lock
);
6053 mddev
->bitmap_info
.file
= NULL
;
6054 spin_unlock(&mddev
->lock
);
6057 mddev
->bitmap_info
.offset
= 0;
6059 export_array(mddev
);
6062 if (mddev
->hold_active
== UNTIL_STOP
)
6063 mddev
->hold_active
= 0;
6065 md_new_event(mddev
);
6066 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6071 static void autorun_array(struct mddev
*mddev
)
6073 struct md_rdev
*rdev
;
6076 if (list_empty(&mddev
->disks
))
6079 pr_info("md: running: ");
6081 rdev_for_each(rdev
, mddev
) {
6082 char b
[BDEVNAME_SIZE
];
6083 pr_cont("<%s>", bdevname(rdev
->bdev
,b
));
6087 err
= do_md_run(mddev
);
6089 pr_warn("md: do_md_run() returned %d\n", err
);
6090 do_md_stop(mddev
, 0, NULL
);
6095 * lets try to run arrays based on all disks that have arrived
6096 * until now. (those are in pending_raid_disks)
6098 * the method: pick the first pending disk, collect all disks with
6099 * the same UUID, remove all from the pending list and put them into
6100 * the 'same_array' list. Then order this list based on superblock
6101 * update time (freshest comes first), kick out 'old' disks and
6102 * compare superblocks. If everything's fine then run it.
6104 * If "unit" is allocated, then bump its reference count
6106 static void autorun_devices(int part
)
6108 struct md_rdev
*rdev0
, *rdev
, *tmp
;
6109 struct mddev
*mddev
;
6110 char b
[BDEVNAME_SIZE
];
6112 pr_info("md: autorun ...\n");
6113 while (!list_empty(&pending_raid_disks
)) {
6116 LIST_HEAD(candidates
);
6117 rdev0
= list_entry(pending_raid_disks
.next
,
6118 struct md_rdev
, same_set
);
6120 pr_debug("md: considering %s ...\n", bdevname(rdev0
->bdev
,b
));
6121 INIT_LIST_HEAD(&candidates
);
6122 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
6123 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
6124 pr_debug("md: adding %s ...\n",
6125 bdevname(rdev
->bdev
,b
));
6126 list_move(&rdev
->same_set
, &candidates
);
6129 * now we have a set of devices, with all of them having
6130 * mostly sane superblocks. It's time to allocate the
6134 dev
= MKDEV(mdp_major
,
6135 rdev0
->preferred_minor
<< MdpMinorShift
);
6136 unit
= MINOR(dev
) >> MdpMinorShift
;
6138 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
6141 if (rdev0
->preferred_minor
!= unit
) {
6142 pr_warn("md: unit number in %s is bad: %d\n",
6143 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
6147 md_probe(dev
, NULL
, NULL
);
6148 mddev
= mddev_find(dev
);
6149 if (!mddev
|| !mddev
->gendisk
) {
6154 if (mddev_lock(mddev
))
6155 pr_warn("md: %s locked, cannot run\n", mdname(mddev
));
6156 else if (mddev
->raid_disks
|| mddev
->major_version
6157 || !list_empty(&mddev
->disks
)) {
6158 pr_warn("md: %s already running, cannot run %s\n",
6159 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
6160 mddev_unlock(mddev
);
6162 pr_debug("md: created %s\n", mdname(mddev
));
6163 mddev
->persistent
= 1;
6164 rdev_for_each_list(rdev
, tmp
, &candidates
) {
6165 list_del_init(&rdev
->same_set
);
6166 if (bind_rdev_to_array(rdev
, mddev
))
6169 autorun_array(mddev
);
6170 mddev_unlock(mddev
);
6172 /* on success, candidates will be empty, on error
6175 rdev_for_each_list(rdev
, tmp
, &candidates
) {
6176 list_del_init(&rdev
->same_set
);
6181 pr_info("md: ... autorun DONE.\n");
6183 #endif /* !MODULE */
6185 static int get_version(void __user
*arg
)
6189 ver
.major
= MD_MAJOR_VERSION
;
6190 ver
.minor
= MD_MINOR_VERSION
;
6191 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
6193 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
6199 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
6201 mdu_array_info_t info
;
6202 int nr
,working
,insync
,failed
,spare
;
6203 struct md_rdev
*rdev
;
6205 nr
= working
= insync
= failed
= spare
= 0;
6207 rdev_for_each_rcu(rdev
, mddev
) {
6209 if (test_bit(Faulty
, &rdev
->flags
))
6213 if (test_bit(In_sync
, &rdev
->flags
))
6215 else if (test_bit(Journal
, &rdev
->flags
))
6216 /* TODO: add journal count to md_u.h */
6224 info
.major_version
= mddev
->major_version
;
6225 info
.minor_version
= mddev
->minor_version
;
6226 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
6227 info
.ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
6228 info
.level
= mddev
->level
;
6229 info
.size
= mddev
->dev_sectors
/ 2;
6230 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
6233 info
.raid_disks
= mddev
->raid_disks
;
6234 info
.md_minor
= mddev
->md_minor
;
6235 info
.not_persistent
= !mddev
->persistent
;
6237 info
.utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
6240 info
.state
= (1<<MD_SB_CLEAN
);
6241 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6242 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
6243 if (mddev_is_clustered(mddev
))
6244 info
.state
|= (1<<MD_SB_CLUSTERED
);
6245 info
.active_disks
= insync
;
6246 info
.working_disks
= working
;
6247 info
.failed_disks
= failed
;
6248 info
.spare_disks
= spare
;
6250 info
.layout
= mddev
->layout
;
6251 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
6253 if (copy_to_user(arg
, &info
, sizeof(info
)))
6259 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
6261 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
6265 file
= kzalloc(sizeof(*file
), GFP_NOIO
);
6270 spin_lock(&mddev
->lock
);
6271 /* bitmap enabled */
6272 if (mddev
->bitmap_info
.file
) {
6273 ptr
= file_path(mddev
->bitmap_info
.file
, file
->pathname
,
6274 sizeof(file
->pathname
));
6278 memmove(file
->pathname
, ptr
,
6279 sizeof(file
->pathname
)-(ptr
-file
->pathname
));
6281 spin_unlock(&mddev
->lock
);
6284 copy_to_user(arg
, file
, sizeof(*file
)))
6291 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
6293 mdu_disk_info_t info
;
6294 struct md_rdev
*rdev
;
6296 if (copy_from_user(&info
, arg
, sizeof(info
)))
6300 rdev
= md_find_rdev_nr_rcu(mddev
, info
.number
);
6302 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
6303 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
6304 info
.raid_disk
= rdev
->raid_disk
;
6306 if (test_bit(Faulty
, &rdev
->flags
))
6307 info
.state
|= (1<<MD_DISK_FAULTY
);
6308 else if (test_bit(In_sync
, &rdev
->flags
)) {
6309 info
.state
|= (1<<MD_DISK_ACTIVE
);
6310 info
.state
|= (1<<MD_DISK_SYNC
);
6312 if (test_bit(Journal
, &rdev
->flags
))
6313 info
.state
|= (1<<MD_DISK_JOURNAL
);
6314 if (test_bit(WriteMostly
, &rdev
->flags
))
6315 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
6316 if (test_bit(FailFast
, &rdev
->flags
))
6317 info
.state
|= (1<<MD_DISK_FAILFAST
);
6319 info
.major
= info
.minor
= 0;
6320 info
.raid_disk
= -1;
6321 info
.state
= (1<<MD_DISK_REMOVED
);
6325 if (copy_to_user(arg
, &info
, sizeof(info
)))
6331 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
6333 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
6334 struct md_rdev
*rdev
;
6335 dev_t dev
= MKDEV(info
->major
,info
->minor
);
6337 if (mddev_is_clustered(mddev
) &&
6338 !(info
->state
& ((1 << MD_DISK_CLUSTER_ADD
) | (1 << MD_DISK_CANDIDATE
)))) {
6339 pr_warn("%s: Cannot add to clustered mddev.\n",
6344 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
6347 if (!mddev
->raid_disks
) {
6349 /* expecting a device which has a superblock */
6350 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
6352 pr_warn("md: md_import_device returned %ld\n",
6354 return PTR_ERR(rdev
);
6356 if (!list_empty(&mddev
->disks
)) {
6357 struct md_rdev
*rdev0
6358 = list_entry(mddev
->disks
.next
,
6359 struct md_rdev
, same_set
);
6360 err
= super_types
[mddev
->major_version
]
6361 .load_super(rdev
, rdev0
, mddev
->minor_version
);
6363 pr_warn("md: %s has different UUID to %s\n",
6364 bdevname(rdev
->bdev
,b
),
6365 bdevname(rdev0
->bdev
,b2
));
6370 err
= bind_rdev_to_array(rdev
, mddev
);
6377 * add_new_disk can be used once the array is assembled
6378 * to add "hot spares". They must already have a superblock
6383 if (!mddev
->pers
->hot_add_disk
) {
6384 pr_warn("%s: personality does not support diskops!\n",
6388 if (mddev
->persistent
)
6389 rdev
= md_import_device(dev
, mddev
->major_version
,
6390 mddev
->minor_version
);
6392 rdev
= md_import_device(dev
, -1, -1);
6394 pr_warn("md: md_import_device returned %ld\n",
6396 return PTR_ERR(rdev
);
6398 /* set saved_raid_disk if appropriate */
6399 if (!mddev
->persistent
) {
6400 if (info
->state
& (1<<MD_DISK_SYNC
) &&
6401 info
->raid_disk
< mddev
->raid_disks
) {
6402 rdev
->raid_disk
= info
->raid_disk
;
6403 set_bit(In_sync
, &rdev
->flags
);
6404 clear_bit(Bitmap_sync
, &rdev
->flags
);
6406 rdev
->raid_disk
= -1;
6407 rdev
->saved_raid_disk
= rdev
->raid_disk
;
6409 super_types
[mddev
->major_version
].
6410 validate_super(mddev
, rdev
);
6411 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
6412 rdev
->raid_disk
!= info
->raid_disk
) {
6413 /* This was a hot-add request, but events doesn't
6414 * match, so reject it.
6420 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
6421 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6422 set_bit(WriteMostly
, &rdev
->flags
);
6424 clear_bit(WriteMostly
, &rdev
->flags
);
6425 if (info
->state
& (1<<MD_DISK_FAILFAST
))
6426 set_bit(FailFast
, &rdev
->flags
);
6428 clear_bit(FailFast
, &rdev
->flags
);
6430 if (info
->state
& (1<<MD_DISK_JOURNAL
)) {
6431 struct md_rdev
*rdev2
;
6432 bool has_journal
= false;
6434 /* make sure no existing journal disk */
6435 rdev_for_each(rdev2
, mddev
) {
6436 if (test_bit(Journal
, &rdev2
->flags
)) {
6441 if (has_journal
|| mddev
->bitmap
) {
6445 set_bit(Journal
, &rdev
->flags
);
6448 * check whether the device shows up in other nodes
6450 if (mddev_is_clustered(mddev
)) {
6451 if (info
->state
& (1 << MD_DISK_CANDIDATE
))
6452 set_bit(Candidate
, &rdev
->flags
);
6453 else if (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)) {
6454 /* --add initiated by this node */
6455 err
= md_cluster_ops
->add_new_disk(mddev
, rdev
);
6463 rdev
->raid_disk
= -1;
6464 err
= bind_rdev_to_array(rdev
, mddev
);
6469 if (mddev_is_clustered(mddev
)) {
6470 if (info
->state
& (1 << MD_DISK_CANDIDATE
)) {
6472 err
= md_cluster_ops
->new_disk_ack(mddev
,
6475 md_kick_rdev_from_array(rdev
);
6479 md_cluster_ops
->add_new_disk_cancel(mddev
);
6481 err
= add_bound_rdev(rdev
);
6485 err
= add_bound_rdev(rdev
);
6490 /* otherwise, add_new_disk is only allowed
6491 * for major_version==0 superblocks
6493 if (mddev
->major_version
!= 0) {
6494 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev
));
6498 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
6500 rdev
= md_import_device(dev
, -1, 0);
6502 pr_warn("md: error, md_import_device() returned %ld\n",
6504 return PTR_ERR(rdev
);
6506 rdev
->desc_nr
= info
->number
;
6507 if (info
->raid_disk
< mddev
->raid_disks
)
6508 rdev
->raid_disk
= info
->raid_disk
;
6510 rdev
->raid_disk
= -1;
6512 if (rdev
->raid_disk
< mddev
->raid_disks
)
6513 if (info
->state
& (1<<MD_DISK_SYNC
))
6514 set_bit(In_sync
, &rdev
->flags
);
6516 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6517 set_bit(WriteMostly
, &rdev
->flags
);
6518 if (info
->state
& (1<<MD_DISK_FAILFAST
))
6519 set_bit(FailFast
, &rdev
->flags
);
6521 if (!mddev
->persistent
) {
6522 pr_debug("md: nonpersistent superblock ...\n");
6523 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6525 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6526 rdev
->sectors
= rdev
->sb_start
;
6528 err
= bind_rdev_to_array(rdev
, mddev
);
6538 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
6540 char b
[BDEVNAME_SIZE
];
6541 struct md_rdev
*rdev
;
6546 rdev
= find_rdev(mddev
, dev
);
6550 if (rdev
->raid_disk
< 0)
6553 clear_bit(Blocked
, &rdev
->flags
);
6554 remove_and_add_spares(mddev
, rdev
);
6556 if (rdev
->raid_disk
>= 0)
6560 if (mddev_is_clustered(mddev
))
6561 md_cluster_ops
->remove_disk(mddev
, rdev
);
6563 md_kick_rdev_from_array(rdev
);
6564 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6566 md_wakeup_thread(mddev
->thread
);
6568 md_update_sb(mddev
, 1);
6569 md_new_event(mddev
);
6573 pr_debug("md: cannot remove active disk %s from %s ...\n",
6574 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6578 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
6580 char b
[BDEVNAME_SIZE
];
6582 struct md_rdev
*rdev
;
6587 if (mddev
->major_version
!= 0) {
6588 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6592 if (!mddev
->pers
->hot_add_disk
) {
6593 pr_warn("%s: personality does not support diskops!\n",
6598 rdev
= md_import_device(dev
, -1, 0);
6600 pr_warn("md: error, md_import_device() returned %ld\n",
6605 if (mddev
->persistent
)
6606 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6608 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6610 rdev
->sectors
= rdev
->sb_start
;
6612 if (test_bit(Faulty
, &rdev
->flags
)) {
6613 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6614 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6619 clear_bit(In_sync
, &rdev
->flags
);
6621 rdev
->saved_raid_disk
= -1;
6622 err
= bind_rdev_to_array(rdev
, mddev
);
6627 * The rest should better be atomic, we can have disk failures
6628 * noticed in interrupt contexts ...
6631 rdev
->raid_disk
= -1;
6633 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6635 md_update_sb(mddev
, 1);
6637 * Kick recovery, maybe this spare has to be added to the
6638 * array immediately.
6640 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6641 md_wakeup_thread(mddev
->thread
);
6642 md_new_event(mddev
);
6650 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
6655 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
6657 if (mddev
->recovery
|| mddev
->sync_thread
)
6659 /* we should be able to change the bitmap.. */
6663 struct inode
*inode
;
6666 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
)
6667 return -EEXIST
; /* cannot add when bitmap is present */
6671 pr_warn("%s: error: failed to get bitmap file\n",
6676 inode
= f
->f_mapping
->host
;
6677 if (!S_ISREG(inode
->i_mode
)) {
6678 pr_warn("%s: error: bitmap file must be a regular file\n",
6681 } else if (!(f
->f_mode
& FMODE_WRITE
)) {
6682 pr_warn("%s: error: bitmap file must open for write\n",
6685 } else if (atomic_read(&inode
->i_writecount
) != 1) {
6686 pr_warn("%s: error: bitmap file is already in use\n",
6694 mddev
->bitmap_info
.file
= f
;
6695 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6696 } else if (mddev
->bitmap
== NULL
)
6697 return -ENOENT
; /* cannot remove what isn't there */
6701 struct bitmap
*bitmap
;
6703 bitmap
= md_bitmap_create(mddev
, -1);
6704 mddev_suspend(mddev
);
6705 if (!IS_ERR(bitmap
)) {
6706 mddev
->bitmap
= bitmap
;
6707 err
= md_bitmap_load(mddev
);
6709 err
= PTR_ERR(bitmap
);
6711 md_bitmap_destroy(mddev
);
6714 mddev_resume(mddev
);
6715 } else if (fd
< 0) {
6716 mddev_suspend(mddev
);
6717 md_bitmap_destroy(mddev
);
6718 mddev_resume(mddev
);
6722 struct file
*f
= mddev
->bitmap_info
.file
;
6724 spin_lock(&mddev
->lock
);
6725 mddev
->bitmap_info
.file
= NULL
;
6726 spin_unlock(&mddev
->lock
);
6735 * set_array_info is used two different ways
6736 * The original usage is when creating a new array.
6737 * In this usage, raid_disks is > 0 and it together with
6738 * level, size, not_persistent,layout,chunksize determine the
6739 * shape of the array.
6740 * This will always create an array with a type-0.90.0 superblock.
6741 * The newer usage is when assembling an array.
6742 * In this case raid_disks will be 0, and the major_version field is
6743 * use to determine which style super-blocks are to be found on the devices.
6744 * The minor and patch _version numbers are also kept incase the
6745 * super_block handler wishes to interpret them.
6747 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6750 if (info
->raid_disks
== 0) {
6751 /* just setting version number for superblock loading */
6752 if (info
->major_version
< 0 ||
6753 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6754 super_types
[info
->major_version
].name
== NULL
) {
6755 /* maybe try to auto-load a module? */
6756 pr_warn("md: superblock version %d not known\n",
6757 info
->major_version
);
6760 mddev
->major_version
= info
->major_version
;
6761 mddev
->minor_version
= info
->minor_version
;
6762 mddev
->patch_version
= info
->patch_version
;
6763 mddev
->persistent
= !info
->not_persistent
;
6764 /* ensure mddev_put doesn't delete this now that there
6765 * is some minimal configuration.
6767 mddev
->ctime
= ktime_get_real_seconds();
6770 mddev
->major_version
= MD_MAJOR_VERSION
;
6771 mddev
->minor_version
= MD_MINOR_VERSION
;
6772 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6773 mddev
->ctime
= ktime_get_real_seconds();
6775 mddev
->level
= info
->level
;
6776 mddev
->clevel
[0] = 0;
6777 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6778 mddev
->raid_disks
= info
->raid_disks
;
6779 /* don't set md_minor, it is determined by which /dev/md* was
6782 if (info
->state
& (1<<MD_SB_CLEAN
))
6783 mddev
->recovery_cp
= MaxSector
;
6785 mddev
->recovery_cp
= 0;
6786 mddev
->persistent
= ! info
->not_persistent
;
6787 mddev
->external
= 0;
6789 mddev
->layout
= info
->layout
;
6790 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6792 if (mddev
->persistent
) {
6793 mddev
->max_disks
= MD_SB_DISKS
;
6795 mddev
->sb_flags
= 0;
6797 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6799 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6800 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6801 mddev
->bitmap_info
.offset
= 0;
6803 mddev
->reshape_position
= MaxSector
;
6806 * Generate a 128 bit UUID
6808 get_random_bytes(mddev
->uuid
, 16);
6810 mddev
->new_level
= mddev
->level
;
6811 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6812 mddev
->new_layout
= mddev
->layout
;
6813 mddev
->delta_disks
= 0;
6814 mddev
->reshape_backwards
= 0;
6819 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6821 lockdep_assert_held(&mddev
->reconfig_mutex
);
6823 if (mddev
->external_size
)
6826 mddev
->array_sectors
= array_sectors
;
6828 EXPORT_SYMBOL(md_set_array_sectors
);
6830 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6832 struct md_rdev
*rdev
;
6834 int fit
= (num_sectors
== 0);
6835 sector_t old_dev_sectors
= mddev
->dev_sectors
;
6837 if (mddev
->pers
->resize
== NULL
)
6839 /* The "num_sectors" is the number of sectors of each device that
6840 * is used. This can only make sense for arrays with redundancy.
6841 * linear and raid0 always use whatever space is available. We can only
6842 * consider changing this number if no resync or reconstruction is
6843 * happening, and if the new size is acceptable. It must fit before the
6844 * sb_start or, if that is <data_offset, it must fit before the size
6845 * of each device. If num_sectors is zero, we find the largest size
6848 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6854 rdev_for_each(rdev
, mddev
) {
6855 sector_t avail
= rdev
->sectors
;
6857 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6858 num_sectors
= avail
;
6859 if (avail
< num_sectors
)
6862 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6864 if (mddev_is_clustered(mddev
))
6865 md_cluster_ops
->update_size(mddev
, old_dev_sectors
);
6866 else if (mddev
->queue
) {
6867 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
6868 revalidate_disk(mddev
->gendisk
);
6874 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6877 struct md_rdev
*rdev
;
6878 /* change the number of raid disks */
6879 if (mddev
->pers
->check_reshape
== NULL
)
6883 if (raid_disks
<= 0 ||
6884 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6886 if (mddev
->sync_thread
||
6887 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6888 mddev
->reshape_position
!= MaxSector
)
6891 rdev_for_each(rdev
, mddev
) {
6892 if (mddev
->raid_disks
< raid_disks
&&
6893 rdev
->data_offset
< rdev
->new_data_offset
)
6895 if (mddev
->raid_disks
> raid_disks
&&
6896 rdev
->data_offset
> rdev
->new_data_offset
)
6900 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6901 if (mddev
->delta_disks
< 0)
6902 mddev
->reshape_backwards
= 1;
6903 else if (mddev
->delta_disks
> 0)
6904 mddev
->reshape_backwards
= 0;
6906 rv
= mddev
->pers
->check_reshape(mddev
);
6908 mddev
->delta_disks
= 0;
6909 mddev
->reshape_backwards
= 0;
6915 * update_array_info is used to change the configuration of an
6917 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6918 * fields in the info are checked against the array.
6919 * Any differences that cannot be handled will cause an error.
6920 * Normally, only one change can be managed at a time.
6922 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6928 /* calculate expected state,ignoring low bits */
6929 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6930 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6932 if (mddev
->major_version
!= info
->major_version
||
6933 mddev
->minor_version
!= info
->minor_version
||
6934 /* mddev->patch_version != info->patch_version || */
6935 mddev
->ctime
!= info
->ctime
||
6936 mddev
->level
!= info
->level
||
6937 /* mddev->layout != info->layout || */
6938 mddev
->persistent
!= !info
->not_persistent
||
6939 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6940 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6941 ((state
^info
->state
) & 0xfffffe00)
6944 /* Check there is only one change */
6945 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6947 if (mddev
->raid_disks
!= info
->raid_disks
)
6949 if (mddev
->layout
!= info
->layout
)
6951 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6958 if (mddev
->layout
!= info
->layout
) {
6960 * we don't need to do anything at the md level, the
6961 * personality will take care of it all.
6963 if (mddev
->pers
->check_reshape
== NULL
)
6966 mddev
->new_layout
= info
->layout
;
6967 rv
= mddev
->pers
->check_reshape(mddev
);
6969 mddev
->new_layout
= mddev
->layout
;
6973 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6974 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6976 if (mddev
->raid_disks
!= info
->raid_disks
)
6977 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6979 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6980 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
) {
6984 if (mddev
->recovery
|| mddev
->sync_thread
) {
6988 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6989 struct bitmap
*bitmap
;
6990 /* add the bitmap */
6991 if (mddev
->bitmap
) {
6995 if (mddev
->bitmap_info
.default_offset
== 0) {
6999 mddev
->bitmap_info
.offset
=
7000 mddev
->bitmap_info
.default_offset
;
7001 mddev
->bitmap_info
.space
=
7002 mddev
->bitmap_info
.default_space
;
7003 bitmap
= md_bitmap_create(mddev
, -1);
7004 mddev_suspend(mddev
);
7005 if (!IS_ERR(bitmap
)) {
7006 mddev
->bitmap
= bitmap
;
7007 rv
= md_bitmap_load(mddev
);
7009 rv
= PTR_ERR(bitmap
);
7011 md_bitmap_destroy(mddev
);
7012 mddev_resume(mddev
);
7014 /* remove the bitmap */
7015 if (!mddev
->bitmap
) {
7019 if (mddev
->bitmap
->storage
.file
) {
7023 if (mddev
->bitmap_info
.nodes
) {
7024 /* hold PW on all the bitmap lock */
7025 if (md_cluster_ops
->lock_all_bitmaps(mddev
) <= 0) {
7026 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7028 md_cluster_ops
->unlock_all_bitmaps(mddev
);
7032 mddev
->bitmap_info
.nodes
= 0;
7033 md_cluster_ops
->leave(mddev
);
7035 mddev_suspend(mddev
);
7036 md_bitmap_destroy(mddev
);
7037 mddev_resume(mddev
);
7038 mddev
->bitmap_info
.offset
= 0;
7041 md_update_sb(mddev
, 1);
7047 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
7049 struct md_rdev
*rdev
;
7052 if (mddev
->pers
== NULL
)
7056 rdev
= md_find_rdev_rcu(mddev
, dev
);
7060 md_error(mddev
, rdev
);
7061 if (!test_bit(Faulty
, &rdev
->flags
))
7069 * We have a problem here : there is no easy way to give a CHS
7070 * virtual geometry. We currently pretend that we have a 2 heads
7071 * 4 sectors (with a BIG number of cylinders...). This drives
7072 * dosfs just mad... ;-)
7074 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
7076 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
7080 geo
->cylinders
= mddev
->array_sectors
/ 8;
7084 static inline bool md_ioctl_valid(unsigned int cmd
)
7089 case GET_ARRAY_INFO
:
7090 case GET_BITMAP_FILE
:
7093 case HOT_REMOVE_DISK
:
7096 case RESTART_ARRAY_RW
:
7098 case SET_ARRAY_INFO
:
7099 case SET_BITMAP_FILE
:
7100 case SET_DISK_FAULTY
:
7103 case CLUSTERED_DISK_NACK
:
7110 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
7111 unsigned int cmd
, unsigned long arg
)
7114 void __user
*argp
= (void __user
*)arg
;
7115 struct mddev
*mddev
= NULL
;
7117 bool did_set_md_closing
= false;
7119 if (!md_ioctl_valid(cmd
))
7124 case GET_ARRAY_INFO
:
7128 if (!capable(CAP_SYS_ADMIN
))
7133 * Commands dealing with the RAID driver but not any
7138 err
= get_version(argp
);
7144 autostart_arrays(arg
);
7151 * Commands creating/starting a new array:
7154 mddev
= bdev
->bd_disk
->private_data
;
7161 /* Some actions do not requires the mutex */
7163 case GET_ARRAY_INFO
:
7164 if (!mddev
->raid_disks
&& !mddev
->external
)
7167 err
= get_array_info(mddev
, argp
);
7171 if (!mddev
->raid_disks
&& !mddev
->external
)
7174 err
= get_disk_info(mddev
, argp
);
7177 case SET_DISK_FAULTY
:
7178 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
7181 case GET_BITMAP_FILE
:
7182 err
= get_bitmap_file(mddev
, argp
);
7187 if (cmd
== ADD_NEW_DISK
)
7188 /* need to ensure md_delayed_delete() has completed */
7189 flush_workqueue(md_misc_wq
);
7191 if (cmd
== HOT_REMOVE_DISK
)
7192 /* need to ensure recovery thread has run */
7193 wait_event_interruptible_timeout(mddev
->sb_wait
,
7194 !test_bit(MD_RECOVERY_NEEDED
,
7196 msecs_to_jiffies(5000));
7197 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
7198 /* Need to flush page cache, and ensure no-one else opens
7201 mutex_lock(&mddev
->open_mutex
);
7202 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
7203 mutex_unlock(&mddev
->open_mutex
);
7207 WARN_ON_ONCE(test_bit(MD_CLOSING
, &mddev
->flags
));
7208 set_bit(MD_CLOSING
, &mddev
->flags
);
7209 did_set_md_closing
= true;
7210 mutex_unlock(&mddev
->open_mutex
);
7211 sync_blockdev(bdev
);
7213 err
= mddev_lock(mddev
);
7215 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7220 if (cmd
== SET_ARRAY_INFO
) {
7221 mdu_array_info_t info
;
7223 memset(&info
, 0, sizeof(info
));
7224 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
7229 err
= update_array_info(mddev
, &info
);
7231 pr_warn("md: couldn't update array info. %d\n", err
);
7236 if (!list_empty(&mddev
->disks
)) {
7237 pr_warn("md: array %s already has disks!\n", mdname(mddev
));
7241 if (mddev
->raid_disks
) {
7242 pr_warn("md: array %s already initialised!\n", mdname(mddev
));
7246 err
= set_array_info(mddev
, &info
);
7248 pr_warn("md: couldn't set array info. %d\n", err
);
7255 * Commands querying/configuring an existing array:
7257 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7258 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7259 if ((!mddev
->raid_disks
&& !mddev
->external
)
7260 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
7261 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
7262 && cmd
!= GET_BITMAP_FILE
) {
7268 * Commands even a read-only array can execute:
7271 case RESTART_ARRAY_RW
:
7272 err
= restart_array(mddev
);
7276 err
= do_md_stop(mddev
, 0, bdev
);
7280 err
= md_set_readonly(mddev
, bdev
);
7283 case HOT_REMOVE_DISK
:
7284 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
7288 /* We can support ADD_NEW_DISK on read-only arrays
7289 * only if we are re-adding a preexisting device.
7290 * So require mddev->pers and MD_DISK_SYNC.
7293 mdu_disk_info_t info
;
7294 if (copy_from_user(&info
, argp
, sizeof(info
)))
7296 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
7297 /* Need to clear read-only for this */
7300 err
= add_new_disk(mddev
, &info
);
7306 if (get_user(ro
, (int __user
*)(arg
))) {
7312 /* if the bdev is going readonly the value of mddev->ro
7313 * does not matter, no writes are coming
7318 /* are we are already prepared for writes? */
7322 /* transitioning to readauto need only happen for
7323 * arrays that call md_write_start
7326 err
= restart_array(mddev
);
7329 set_disk_ro(mddev
->gendisk
, 0);
7336 * The remaining ioctls are changing the state of the
7337 * superblock, so we do not allow them on read-only arrays.
7339 if (mddev
->ro
&& mddev
->pers
) {
7340 if (mddev
->ro
== 2) {
7342 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7343 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7344 /* mddev_unlock will wake thread */
7345 /* If a device failed while we were read-only, we
7346 * need to make sure the metadata is updated now.
7348 if (test_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
)) {
7349 mddev_unlock(mddev
);
7350 wait_event(mddev
->sb_wait
,
7351 !test_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
) &&
7352 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
7353 mddev_lock_nointr(mddev
);
7364 mdu_disk_info_t info
;
7365 if (copy_from_user(&info
, argp
, sizeof(info
)))
7368 err
= add_new_disk(mddev
, &info
);
7372 case CLUSTERED_DISK_NACK
:
7373 if (mddev_is_clustered(mddev
))
7374 md_cluster_ops
->new_disk_ack(mddev
, false);
7380 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
7384 err
= do_md_run(mddev
);
7387 case SET_BITMAP_FILE
:
7388 err
= set_bitmap_file(mddev
, (int)arg
);
7397 if (mddev
->hold_active
== UNTIL_IOCTL
&&
7399 mddev
->hold_active
= 0;
7400 mddev_unlock(mddev
);
7402 if(did_set_md_closing
)
7403 clear_bit(MD_CLOSING
, &mddev
->flags
);
7406 #ifdef CONFIG_COMPAT
7407 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
7408 unsigned int cmd
, unsigned long arg
)
7411 case HOT_REMOVE_DISK
:
7413 case SET_DISK_FAULTY
:
7414 case SET_BITMAP_FILE
:
7415 /* These take in integer arg, do not convert */
7418 arg
= (unsigned long)compat_ptr(arg
);
7422 return md_ioctl(bdev
, mode
, cmd
, arg
);
7424 #endif /* CONFIG_COMPAT */
7426 static int md_open(struct block_device
*bdev
, fmode_t mode
)
7429 * Succeed if we can lock the mddev, which confirms that
7430 * it isn't being stopped right now.
7432 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
7438 if (mddev
->gendisk
!= bdev
->bd_disk
) {
7439 /* we are racing with mddev_put which is discarding this
7443 /* Wait until bdev->bd_disk is definitely gone */
7444 flush_workqueue(md_misc_wq
);
7445 /* Then retry the open from the top */
7446 return -ERESTARTSYS
;
7448 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
7450 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
7453 if (test_bit(MD_CLOSING
, &mddev
->flags
)) {
7454 mutex_unlock(&mddev
->open_mutex
);
7460 atomic_inc(&mddev
->openers
);
7461 mutex_unlock(&mddev
->open_mutex
);
7463 check_disk_change(bdev
);
7470 static void md_release(struct gendisk
*disk
, fmode_t mode
)
7472 struct mddev
*mddev
= disk
->private_data
;
7475 atomic_dec(&mddev
->openers
);
7479 static int md_media_changed(struct gendisk
*disk
)
7481 struct mddev
*mddev
= disk
->private_data
;
7483 return mddev
->changed
;
7486 static int md_revalidate(struct gendisk
*disk
)
7488 struct mddev
*mddev
= disk
->private_data
;
7493 static const struct block_device_operations md_fops
=
7495 .owner
= THIS_MODULE
,
7497 .release
= md_release
,
7499 #ifdef CONFIG_COMPAT
7500 .compat_ioctl
= md_compat_ioctl
,
7502 .getgeo
= md_getgeo
,
7503 .media_changed
= md_media_changed
,
7504 .revalidate_disk
= md_revalidate
,
7507 static int md_thread(void *arg
)
7509 struct md_thread
*thread
= arg
;
7512 * md_thread is a 'system-thread', it's priority should be very
7513 * high. We avoid resource deadlocks individually in each
7514 * raid personality. (RAID5 does preallocation) We also use RR and
7515 * the very same RT priority as kswapd, thus we will never get
7516 * into a priority inversion deadlock.
7518 * we definitely have to have equal or higher priority than
7519 * bdflush, otherwise bdflush will deadlock if there are too
7520 * many dirty RAID5 blocks.
7523 allow_signal(SIGKILL
);
7524 while (!kthread_should_stop()) {
7526 /* We need to wait INTERRUPTIBLE so that
7527 * we don't add to the load-average.
7528 * That means we need to be sure no signals are
7531 if (signal_pending(current
))
7532 flush_signals(current
);
7534 wait_event_interruptible_timeout
7536 test_bit(THREAD_WAKEUP
, &thread
->flags
)
7537 || kthread_should_stop() || kthread_should_park(),
7540 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
7541 if (kthread_should_park())
7543 if (!kthread_should_stop())
7544 thread
->run(thread
);
7550 void md_wakeup_thread(struct md_thread
*thread
)
7553 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
7554 set_bit(THREAD_WAKEUP
, &thread
->flags
);
7555 wake_up(&thread
->wqueue
);
7558 EXPORT_SYMBOL(md_wakeup_thread
);
7560 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
7561 struct mddev
*mddev
, const char *name
)
7563 struct md_thread
*thread
;
7565 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
7569 init_waitqueue_head(&thread
->wqueue
);
7572 thread
->mddev
= mddev
;
7573 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
7574 thread
->tsk
= kthread_run(md_thread
, thread
,
7576 mdname(thread
->mddev
),
7578 if (IS_ERR(thread
->tsk
)) {
7584 EXPORT_SYMBOL(md_register_thread
);
7586 void md_unregister_thread(struct md_thread
**threadp
)
7588 struct md_thread
*thread
= *threadp
;
7591 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
7592 /* Locking ensures that mddev_unlock does not wake_up a
7593 * non-existent thread
7595 spin_lock(&pers_lock
);
7597 spin_unlock(&pers_lock
);
7599 kthread_stop(thread
->tsk
);
7602 EXPORT_SYMBOL(md_unregister_thread
);
7604 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
7606 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
7609 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
7611 mddev
->pers
->error_handler(mddev
,rdev
);
7612 if (mddev
->degraded
)
7613 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7614 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7615 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7616 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7617 md_wakeup_thread(mddev
->thread
);
7618 if (mddev
->event_work
.func
)
7619 queue_work(md_misc_wq
, &mddev
->event_work
);
7620 md_new_event(mddev
);
7622 EXPORT_SYMBOL(md_error
);
7624 /* seq_file implementation /proc/mdstat */
7626 static void status_unused(struct seq_file
*seq
)
7629 struct md_rdev
*rdev
;
7631 seq_printf(seq
, "unused devices: ");
7633 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
7634 char b
[BDEVNAME_SIZE
];
7636 seq_printf(seq
, "%s ",
7637 bdevname(rdev
->bdev
,b
));
7640 seq_printf(seq
, "<none>");
7642 seq_printf(seq
, "\n");
7645 static int status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
7647 sector_t max_sectors
, resync
, res
;
7648 unsigned long dt
, db
;
7651 unsigned int per_milli
;
7653 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7654 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7655 max_sectors
= mddev
->resync_max_sectors
;
7657 max_sectors
= mddev
->dev_sectors
;
7659 resync
= mddev
->curr_resync
;
7661 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7662 /* Still cleaning up */
7663 resync
= max_sectors
;
7664 } else if (resync
> max_sectors
)
7665 resync
= max_sectors
;
7667 resync
-= atomic_read(&mddev
->recovery_active
);
7670 if (test_bit(MD_RESYNCING_REMOTE
, &mddev
->recovery
)) {
7671 struct md_rdev
*rdev
;
7673 rdev_for_each(rdev
, mddev
)
7674 if (rdev
->raid_disk
>= 0 &&
7675 !test_bit(Faulty
, &rdev
->flags
) &&
7676 rdev
->recovery_offset
!= MaxSector
&&
7677 rdev
->recovery_offset
) {
7678 seq_printf(seq
, "\trecover=REMOTE");
7681 if (mddev
->reshape_position
!= MaxSector
)
7682 seq_printf(seq
, "\treshape=REMOTE");
7684 seq_printf(seq
, "\tresync=REMOTE");
7687 if (mddev
->recovery_cp
< MaxSector
) {
7688 seq_printf(seq
, "\tresync=PENDING");
7694 seq_printf(seq
, "\tresync=DELAYED");
7698 WARN_ON(max_sectors
== 0);
7699 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7700 * in a sector_t, and (max_sectors>>scale) will fit in a
7701 * u32, as those are the requirements for sector_div.
7702 * Thus 'scale' must be at least 10
7705 if (sizeof(sector_t
) > sizeof(unsigned long)) {
7706 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
7709 res
= (resync
>>scale
)*1000;
7710 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
7714 int i
, x
= per_milli
/50, y
= 20-x
;
7715 seq_printf(seq
, "[");
7716 for (i
= 0; i
< x
; i
++)
7717 seq_printf(seq
, "=");
7718 seq_printf(seq
, ">");
7719 for (i
= 0; i
< y
; i
++)
7720 seq_printf(seq
, ".");
7721 seq_printf(seq
, "] ");
7723 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
7724 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
7726 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
7728 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
7729 "resync" : "recovery"))),
7730 per_milli
/10, per_milli
% 10,
7731 (unsigned long long) resync
/2,
7732 (unsigned long long) max_sectors
/2);
7735 * dt: time from mark until now
7736 * db: blocks written from mark until now
7737 * rt: remaining time
7739 * rt is a sector_t, so could be 32bit or 64bit.
7740 * So we divide before multiply in case it is 32bit and close
7742 * We scale the divisor (db) by 32 to avoid losing precision
7743 * near the end of resync when the number of remaining sectors
7745 * We then divide rt by 32 after multiplying by db to compensate.
7746 * The '+1' avoids division by zero if db is very small.
7748 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
7750 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
7751 - mddev
->resync_mark_cnt
;
7753 rt
= max_sectors
- resync
; /* number of remaining sectors */
7754 sector_div(rt
, db
/32+1);
7758 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
7759 ((unsigned long)rt
% 60)/6);
7761 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
7765 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
7767 struct list_head
*tmp
;
7769 struct mddev
*mddev
;
7777 spin_lock(&all_mddevs_lock
);
7778 list_for_each(tmp
,&all_mddevs
)
7780 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7782 spin_unlock(&all_mddevs_lock
);
7785 spin_unlock(&all_mddevs_lock
);
7787 return (void*)2;/* tail */
7791 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7793 struct list_head
*tmp
;
7794 struct mddev
*next_mddev
, *mddev
= v
;
7800 spin_lock(&all_mddevs_lock
);
7802 tmp
= all_mddevs
.next
;
7804 tmp
= mddev
->all_mddevs
.next
;
7805 if (tmp
!= &all_mddevs
)
7806 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7808 next_mddev
= (void*)2;
7811 spin_unlock(&all_mddevs_lock
);
7819 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7821 struct mddev
*mddev
= v
;
7823 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7827 static int md_seq_show(struct seq_file
*seq
, void *v
)
7829 struct mddev
*mddev
= v
;
7831 struct md_rdev
*rdev
;
7833 if (v
== (void*)1) {
7834 struct md_personality
*pers
;
7835 seq_printf(seq
, "Personalities : ");
7836 spin_lock(&pers_lock
);
7837 list_for_each_entry(pers
, &pers_list
, list
)
7838 seq_printf(seq
, "[%s] ", pers
->name
);
7840 spin_unlock(&pers_lock
);
7841 seq_printf(seq
, "\n");
7842 seq
->poll_event
= atomic_read(&md_event_count
);
7845 if (v
== (void*)2) {
7850 spin_lock(&mddev
->lock
);
7851 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7852 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7853 mddev
->pers
? "" : "in");
7856 seq_printf(seq
, " (read-only)");
7858 seq_printf(seq
, " (auto-read-only)");
7859 seq_printf(seq
, " %s", mddev
->pers
->name
);
7864 rdev_for_each_rcu(rdev
, mddev
) {
7865 char b
[BDEVNAME_SIZE
];
7866 seq_printf(seq
, " %s[%d]",
7867 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7868 if (test_bit(WriteMostly
, &rdev
->flags
))
7869 seq_printf(seq
, "(W)");
7870 if (test_bit(Journal
, &rdev
->flags
))
7871 seq_printf(seq
, "(J)");
7872 if (test_bit(Faulty
, &rdev
->flags
)) {
7873 seq_printf(seq
, "(F)");
7876 if (rdev
->raid_disk
< 0)
7877 seq_printf(seq
, "(S)"); /* spare */
7878 if (test_bit(Replacement
, &rdev
->flags
))
7879 seq_printf(seq
, "(R)");
7880 sectors
+= rdev
->sectors
;
7884 if (!list_empty(&mddev
->disks
)) {
7886 seq_printf(seq
, "\n %llu blocks",
7887 (unsigned long long)
7888 mddev
->array_sectors
/ 2);
7890 seq_printf(seq
, "\n %llu blocks",
7891 (unsigned long long)sectors
/ 2);
7893 if (mddev
->persistent
) {
7894 if (mddev
->major_version
!= 0 ||
7895 mddev
->minor_version
!= 90) {
7896 seq_printf(seq
," super %d.%d",
7897 mddev
->major_version
,
7898 mddev
->minor_version
);
7900 } else if (mddev
->external
)
7901 seq_printf(seq
, " super external:%s",
7902 mddev
->metadata_type
);
7904 seq_printf(seq
, " super non-persistent");
7907 mddev
->pers
->status(seq
, mddev
);
7908 seq_printf(seq
, "\n ");
7909 if (mddev
->pers
->sync_request
) {
7910 if (status_resync(seq
, mddev
))
7911 seq_printf(seq
, "\n ");
7914 seq_printf(seq
, "\n ");
7916 md_bitmap_status(seq
, mddev
->bitmap
);
7918 seq_printf(seq
, "\n");
7920 spin_unlock(&mddev
->lock
);
7925 static const struct seq_operations md_seq_ops
= {
7926 .start
= md_seq_start
,
7927 .next
= md_seq_next
,
7928 .stop
= md_seq_stop
,
7929 .show
= md_seq_show
,
7932 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7934 struct seq_file
*seq
;
7937 error
= seq_open(file
, &md_seq_ops
);
7941 seq
= file
->private_data
;
7942 seq
->poll_event
= atomic_read(&md_event_count
);
7946 static int md_unloading
;
7947 static __poll_t
mdstat_poll(struct file
*filp
, poll_table
*wait
)
7949 struct seq_file
*seq
= filp
->private_data
;
7953 return EPOLLIN
|EPOLLRDNORM
|EPOLLERR
|EPOLLPRI
;
7954 poll_wait(filp
, &md_event_waiters
, wait
);
7956 /* always allow read */
7957 mask
= EPOLLIN
| EPOLLRDNORM
;
7959 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7960 mask
|= EPOLLERR
| EPOLLPRI
;
7964 static const struct file_operations md_seq_fops
= {
7965 .owner
= THIS_MODULE
,
7966 .open
= md_seq_open
,
7968 .llseek
= seq_lseek
,
7969 .release
= seq_release
,
7970 .poll
= mdstat_poll
,
7973 int register_md_personality(struct md_personality
*p
)
7975 pr_debug("md: %s personality registered for level %d\n",
7977 spin_lock(&pers_lock
);
7978 list_add_tail(&p
->list
, &pers_list
);
7979 spin_unlock(&pers_lock
);
7982 EXPORT_SYMBOL(register_md_personality
);
7984 int unregister_md_personality(struct md_personality
*p
)
7986 pr_debug("md: %s personality unregistered\n", p
->name
);
7987 spin_lock(&pers_lock
);
7988 list_del_init(&p
->list
);
7989 spin_unlock(&pers_lock
);
7992 EXPORT_SYMBOL(unregister_md_personality
);
7994 int register_md_cluster_operations(struct md_cluster_operations
*ops
,
7995 struct module
*module
)
7998 spin_lock(&pers_lock
);
7999 if (md_cluster_ops
!= NULL
)
8002 md_cluster_ops
= ops
;
8003 md_cluster_mod
= module
;
8005 spin_unlock(&pers_lock
);
8008 EXPORT_SYMBOL(register_md_cluster_operations
);
8010 int unregister_md_cluster_operations(void)
8012 spin_lock(&pers_lock
);
8013 md_cluster_ops
= NULL
;
8014 spin_unlock(&pers_lock
);
8017 EXPORT_SYMBOL(unregister_md_cluster_operations
);
8019 int md_setup_cluster(struct mddev
*mddev
, int nodes
)
8021 if (!md_cluster_ops
)
8022 request_module("md-cluster");
8023 spin_lock(&pers_lock
);
8024 /* ensure module won't be unloaded */
8025 if (!md_cluster_ops
|| !try_module_get(md_cluster_mod
)) {
8026 pr_warn("can't find md-cluster module or get it's reference.\n");
8027 spin_unlock(&pers_lock
);
8030 spin_unlock(&pers_lock
);
8032 return md_cluster_ops
->join(mddev
, nodes
);
8035 void md_cluster_stop(struct mddev
*mddev
)
8037 if (!md_cluster_ops
)
8039 md_cluster_ops
->leave(mddev
);
8040 module_put(md_cluster_mod
);
8043 static int is_mddev_idle(struct mddev
*mddev
, int init
)
8045 struct md_rdev
*rdev
;
8051 rdev_for_each_rcu(rdev
, mddev
) {
8052 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
8053 curr_events
= (int)part_stat_read_accum(&disk
->part0
, sectors
) -
8054 atomic_read(&disk
->sync_io
);
8055 /* sync IO will cause sync_io to increase before the disk_stats
8056 * as sync_io is counted when a request starts, and
8057 * disk_stats is counted when it completes.
8058 * So resync activity will cause curr_events to be smaller than
8059 * when there was no such activity.
8060 * non-sync IO will cause disk_stat to increase without
8061 * increasing sync_io so curr_events will (eventually)
8062 * be larger than it was before. Once it becomes
8063 * substantially larger, the test below will cause
8064 * the array to appear non-idle, and resync will slow
8066 * If there is a lot of outstanding resync activity when
8067 * we set last_event to curr_events, then all that activity
8068 * completing might cause the array to appear non-idle
8069 * and resync will be slowed down even though there might
8070 * not have been non-resync activity. This will only
8071 * happen once though. 'last_events' will soon reflect
8072 * the state where there is little or no outstanding
8073 * resync requests, and further resync activity will
8074 * always make curr_events less than last_events.
8077 if (init
|| curr_events
- rdev
->last_events
> 64) {
8078 rdev
->last_events
= curr_events
;
8086 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
8088 /* another "blocks" (512byte) blocks have been synced */
8089 atomic_sub(blocks
, &mddev
->recovery_active
);
8090 wake_up(&mddev
->recovery_wait
);
8092 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8093 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
8094 md_wakeup_thread(mddev
->thread
);
8095 // stop recovery, signal do_sync ....
8098 EXPORT_SYMBOL(md_done_sync
);
8100 /* md_write_start(mddev, bi)
8101 * If we need to update some array metadata (e.g. 'active' flag
8102 * in superblock) before writing, schedule a superblock update
8103 * and wait for it to complete.
8104 * A return value of 'false' means that the write wasn't recorded
8105 * and cannot proceed as the array is being suspend.
8107 bool md_write_start(struct mddev
*mddev
, struct bio
*bi
)
8111 if (bio_data_dir(bi
) != WRITE
)
8114 BUG_ON(mddev
->ro
== 1);
8115 if (mddev
->ro
== 2) {
8116 /* need to switch to read/write */
8118 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8119 md_wakeup_thread(mddev
->thread
);
8120 md_wakeup_thread(mddev
->sync_thread
);
8124 percpu_ref_get(&mddev
->writes_pending
);
8125 smp_mb(); /* Match smp_mb in set_in_sync() */
8126 if (mddev
->safemode
== 1)
8127 mddev
->safemode
= 0;
8128 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8129 if (mddev
->in_sync
|| mddev
->sync_checkers
) {
8130 spin_lock(&mddev
->lock
);
8131 if (mddev
->in_sync
) {
8133 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8134 set_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8135 md_wakeup_thread(mddev
->thread
);
8138 spin_unlock(&mddev
->lock
);
8142 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8143 if (!mddev
->has_superblocks
)
8145 wait_event(mddev
->sb_wait
,
8146 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
) ||
8148 if (test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
)) {
8149 percpu_ref_put(&mddev
->writes_pending
);
8154 EXPORT_SYMBOL(md_write_start
);
8156 /* md_write_inc can only be called when md_write_start() has
8157 * already been called at least once of the current request.
8158 * It increments the counter and is useful when a single request
8159 * is split into several parts. Each part causes an increment and
8160 * so needs a matching md_write_end().
8161 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8162 * a spinlocked region.
8164 void md_write_inc(struct mddev
*mddev
, struct bio
*bi
)
8166 if (bio_data_dir(bi
) != WRITE
)
8168 WARN_ON_ONCE(mddev
->in_sync
|| mddev
->ro
);
8169 percpu_ref_get(&mddev
->writes_pending
);
8171 EXPORT_SYMBOL(md_write_inc
);
8173 void md_write_end(struct mddev
*mddev
)
8175 percpu_ref_put(&mddev
->writes_pending
);
8177 if (mddev
->safemode
== 2)
8178 md_wakeup_thread(mddev
->thread
);
8179 else if (mddev
->safemode_delay
)
8180 /* The roundup() ensures this only performs locking once
8181 * every ->safemode_delay jiffies
8183 mod_timer(&mddev
->safemode_timer
,
8184 roundup(jiffies
, mddev
->safemode_delay
) +
8185 mddev
->safemode_delay
);
8188 EXPORT_SYMBOL(md_write_end
);
8190 /* md_allow_write(mddev)
8191 * Calling this ensures that the array is marked 'active' so that writes
8192 * may proceed without blocking. It is important to call this before
8193 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8194 * Must be called with mddev_lock held.
8196 void md_allow_write(struct mddev
*mddev
)
8202 if (!mddev
->pers
->sync_request
)
8205 spin_lock(&mddev
->lock
);
8206 if (mddev
->in_sync
) {
8208 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8209 set_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8210 if (mddev
->safemode_delay
&&
8211 mddev
->safemode
== 0)
8212 mddev
->safemode
= 1;
8213 spin_unlock(&mddev
->lock
);
8214 md_update_sb(mddev
, 0);
8215 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8216 /* wait for the dirty state to be recorded in the metadata */
8217 wait_event(mddev
->sb_wait
,
8218 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
8220 spin_unlock(&mddev
->lock
);
8222 EXPORT_SYMBOL_GPL(md_allow_write
);
8224 #define SYNC_MARKS 10
8225 #define SYNC_MARK_STEP (3*HZ)
8226 #define UPDATE_FREQUENCY (5*60*HZ)
8227 void md_do_sync(struct md_thread
*thread
)
8229 struct mddev
*mddev
= thread
->mddev
;
8230 struct mddev
*mddev2
;
8231 unsigned int currspeed
= 0,
8233 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
8234 unsigned long mark
[SYNC_MARKS
];
8235 unsigned long update_time
;
8236 sector_t mark_cnt
[SYNC_MARKS
];
8238 struct list_head
*tmp
;
8239 sector_t last_check
;
8241 struct md_rdev
*rdev
;
8242 char *desc
, *action
= NULL
;
8243 struct blk_plug plug
;
8246 /* just incase thread restarts... */
8247 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8248 test_bit(MD_RECOVERY_WAIT
, &mddev
->recovery
))
8250 if (mddev
->ro
) {/* never try to sync a read-only array */
8251 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8255 if (mddev_is_clustered(mddev
)) {
8256 ret
= md_cluster_ops
->resync_start(mddev
);
8260 set_bit(MD_CLUSTER_RESYNC_LOCKED
, &mddev
->flags
);
8261 if (!(test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
8262 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) ||
8263 test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
8264 && ((unsigned long long)mddev
->curr_resync_completed
8265 < (unsigned long long)mddev
->resync_max_sectors
))
8269 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8270 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
8271 desc
= "data-check";
8273 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8274 desc
= "requested-resync";
8278 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
8283 mddev
->last_sync_action
= action
?: desc
;
8285 /* we overload curr_resync somewhat here.
8286 * 0 == not engaged in resync at all
8287 * 2 == checking that there is no conflict with another sync
8288 * 1 == like 2, but have yielded to allow conflicting resync to
8290 * other == active in resync - this many blocks
8292 * Before starting a resync we must have set curr_resync to
8293 * 2, and then checked that every "conflicting" array has curr_resync
8294 * less than ours. When we find one that is the same or higher
8295 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8296 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8297 * This will mean we have to start checking from the beginning again.
8302 int mddev2_minor
= -1;
8303 mddev
->curr_resync
= 2;
8306 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8308 for_each_mddev(mddev2
, tmp
) {
8309 if (mddev2
== mddev
)
8311 if (!mddev
->parallel_resync
8312 && mddev2
->curr_resync
8313 && match_mddev_units(mddev
, mddev2
)) {
8315 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
8316 /* arbitrarily yield */
8317 mddev
->curr_resync
= 1;
8318 wake_up(&resync_wait
);
8320 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
8321 /* no need to wait here, we can wait the next
8322 * time 'round when curr_resync == 2
8325 /* We need to wait 'interruptible' so as not to
8326 * contribute to the load average, and not to
8327 * be caught by 'softlockup'
8329 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
8330 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8331 mddev2
->curr_resync
>= mddev
->curr_resync
) {
8332 if (mddev2_minor
!= mddev2
->md_minor
) {
8333 mddev2_minor
= mddev2
->md_minor
;
8334 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8335 desc
, mdname(mddev
),
8339 if (signal_pending(current
))
8340 flush_signals(current
);
8342 finish_wait(&resync_wait
, &wq
);
8345 finish_wait(&resync_wait
, &wq
);
8348 } while (mddev
->curr_resync
< 2);
8351 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8352 /* resync follows the size requested by the personality,
8353 * which defaults to physical size, but can be virtual size
8355 max_sectors
= mddev
->resync_max_sectors
;
8356 atomic64_set(&mddev
->resync_mismatches
, 0);
8357 /* we don't use the checkpoint if there's a bitmap */
8358 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8359 j
= mddev
->resync_min
;
8360 else if (!mddev
->bitmap
)
8361 j
= mddev
->recovery_cp
;
8363 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)) {
8364 max_sectors
= mddev
->resync_max_sectors
;
8366 * If the original node aborts reshaping then we continue the
8367 * reshaping, so set j again to avoid restart reshape from the
8370 if (mddev_is_clustered(mddev
) &&
8371 mddev
->reshape_position
!= MaxSector
)
8372 j
= mddev
->reshape_position
;
8374 /* recovery follows the physical size of devices */
8375 max_sectors
= mddev
->dev_sectors
;
8378 rdev_for_each_rcu(rdev
, mddev
)
8379 if (rdev
->raid_disk
>= 0 &&
8380 !test_bit(Journal
, &rdev
->flags
) &&
8381 !test_bit(Faulty
, &rdev
->flags
) &&
8382 !test_bit(In_sync
, &rdev
->flags
) &&
8383 rdev
->recovery_offset
< j
)
8384 j
= rdev
->recovery_offset
;
8387 /* If there is a bitmap, we need to make sure all
8388 * writes that started before we added a spare
8389 * complete before we start doing a recovery.
8390 * Otherwise the write might complete and (via
8391 * bitmap_endwrite) set a bit in the bitmap after the
8392 * recovery has checked that bit and skipped that
8395 if (mddev
->bitmap
) {
8396 mddev
->pers
->quiesce(mddev
, 1);
8397 mddev
->pers
->quiesce(mddev
, 0);
8401 pr_info("md: %s of RAID array %s\n", desc
, mdname(mddev
));
8402 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev
));
8403 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8404 speed_max(mddev
), desc
);
8406 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
8409 for (m
= 0; m
< SYNC_MARKS
; m
++) {
8411 mark_cnt
[m
] = io_sectors
;
8414 mddev
->resync_mark
= mark
[last_mark
];
8415 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
8418 * Tune reconstruction:
8420 window
= 32*(PAGE_SIZE
/512);
8421 pr_debug("md: using %dk window, over a total of %lluk.\n",
8422 window
/2, (unsigned long long)max_sectors
/2);
8424 atomic_set(&mddev
->recovery_active
, 0);
8428 pr_debug("md: resuming %s of %s from checkpoint.\n",
8429 desc
, mdname(mddev
));
8430 mddev
->curr_resync
= j
;
8432 mddev
->curr_resync
= 3; /* no longer delayed */
8433 mddev
->curr_resync_completed
= j
;
8434 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8435 md_new_event(mddev
);
8436 update_time
= jiffies
;
8438 blk_start_plug(&plug
);
8439 while (j
< max_sectors
) {
8444 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8445 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
8446 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
8447 > (max_sectors
>> 4)) ||
8448 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
8449 (j
- mddev
->curr_resync_completed
)*2
8450 >= mddev
->resync_max
- mddev
->curr_resync_completed
||
8451 mddev
->curr_resync_completed
> mddev
->resync_max
8453 /* time to update curr_resync_completed */
8454 wait_event(mddev
->recovery_wait
,
8455 atomic_read(&mddev
->recovery_active
) == 0);
8456 mddev
->curr_resync_completed
= j
;
8457 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
8458 j
> mddev
->recovery_cp
)
8459 mddev
->recovery_cp
= j
;
8460 update_time
= jiffies
;
8461 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8462 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8465 while (j
>= mddev
->resync_max
&&
8466 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8467 /* As this condition is controlled by user-space,
8468 * we can block indefinitely, so use '_interruptible'
8469 * to avoid triggering warnings.
8471 flush_signals(current
); /* just in case */
8472 wait_event_interruptible(mddev
->recovery_wait
,
8473 mddev
->resync_max
> j
8474 || test_bit(MD_RECOVERY_INTR
,
8478 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8481 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
);
8483 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8487 if (!skipped
) { /* actual IO requested */
8488 io_sectors
+= sectors
;
8489 atomic_add(sectors
, &mddev
->recovery_active
);
8492 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8496 if (j
> max_sectors
)
8497 /* when skipping, extra large numbers can be returned. */
8500 mddev
->curr_resync
= j
;
8501 mddev
->curr_mark_cnt
= io_sectors
;
8502 if (last_check
== 0)
8503 /* this is the earliest that rebuild will be
8504 * visible in /proc/mdstat
8506 md_new_event(mddev
);
8508 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
8511 last_check
= io_sectors
;
8513 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
8515 int next
= (last_mark
+1) % SYNC_MARKS
;
8517 mddev
->resync_mark
= mark
[next
];
8518 mddev
->resync_mark_cnt
= mark_cnt
[next
];
8519 mark
[next
] = jiffies
;
8520 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
8524 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8528 * this loop exits only if either when we are slower than
8529 * the 'hard' speed limit, or the system was IO-idle for
8531 * the system might be non-idle CPU-wise, but we only care
8532 * about not overloading the IO subsystem. (things like an
8533 * e2fsck being done on the RAID array should execute fast)
8537 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
8538 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
8539 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
8541 if (currspeed
> speed_min(mddev
)) {
8542 if (currspeed
> speed_max(mddev
)) {
8546 if (!is_mddev_idle(mddev
, 0)) {
8548 * Give other IO more of a chance.
8549 * The faster the devices, the less we wait.
8551 wait_event(mddev
->recovery_wait
,
8552 !atomic_read(&mddev
->recovery_active
));
8556 pr_info("md: %s: %s %s.\n",mdname(mddev
), desc
,
8557 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
8558 ? "interrupted" : "done");
8560 * this also signals 'finished resyncing' to md_stop
8562 blk_finish_plug(&plug
);
8563 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
8565 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8566 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8567 mddev
->curr_resync
> 3) {
8568 mddev
->curr_resync_completed
= mddev
->curr_resync
;
8569 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8571 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
);
8573 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
8574 mddev
->curr_resync
> 3) {
8575 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8576 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8577 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
8578 pr_debug("md: checkpointing %s of %s.\n",
8579 desc
, mdname(mddev
));
8580 if (test_bit(MD_RECOVERY_ERROR
,
8582 mddev
->recovery_cp
=
8583 mddev
->curr_resync_completed
;
8585 mddev
->recovery_cp
=
8589 mddev
->recovery_cp
= MaxSector
;
8591 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8592 mddev
->curr_resync
= MaxSector
;
8593 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8594 test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
)) {
8596 rdev_for_each_rcu(rdev
, mddev
)
8597 if (rdev
->raid_disk
>= 0 &&
8598 mddev
->delta_disks
>= 0 &&
8599 !test_bit(Journal
, &rdev
->flags
) &&
8600 !test_bit(Faulty
, &rdev
->flags
) &&
8601 !test_bit(In_sync
, &rdev
->flags
) &&
8602 rdev
->recovery_offset
< mddev
->curr_resync
)
8603 rdev
->recovery_offset
= mddev
->curr_resync
;
8609 /* set CHANGE_PENDING here since maybe another update is needed,
8610 * so other nodes are informed. It should be harmless for normal
8612 set_mask_bits(&mddev
->sb_flags
, 0,
8613 BIT(MD_SB_CHANGE_PENDING
) | BIT(MD_SB_CHANGE_DEVS
));
8615 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8616 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8617 mddev
->delta_disks
> 0 &&
8618 mddev
->pers
->finish_reshape
&&
8619 mddev
->pers
->size
&&
8621 mddev_lock_nointr(mddev
);
8622 md_set_array_sectors(mddev
, mddev
->pers
->size(mddev
, 0, 0));
8623 mddev_unlock(mddev
);
8624 if (!mddev_is_clustered(mddev
)) {
8625 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
8626 revalidate_disk(mddev
->gendisk
);
8630 spin_lock(&mddev
->lock
);
8631 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8632 /* We completed so min/max setting can be forgotten if used. */
8633 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8634 mddev
->resync_min
= 0;
8635 mddev
->resync_max
= MaxSector
;
8636 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8637 mddev
->resync_min
= mddev
->curr_resync_completed
;
8638 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8639 mddev
->curr_resync
= 0;
8640 spin_unlock(&mddev
->lock
);
8642 wake_up(&resync_wait
);
8643 md_wakeup_thread(mddev
->thread
);
8646 EXPORT_SYMBOL_GPL(md_do_sync
);
8648 static int remove_and_add_spares(struct mddev
*mddev
,
8649 struct md_rdev
*this)
8651 struct md_rdev
*rdev
;
8654 bool remove_some
= false;
8656 if (this && test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
8657 /* Mustn't remove devices when resync thread is running */
8660 rdev_for_each(rdev
, mddev
) {
8661 if ((this == NULL
|| rdev
== this) &&
8662 rdev
->raid_disk
>= 0 &&
8663 !test_bit(Blocked
, &rdev
->flags
) &&
8664 test_bit(Faulty
, &rdev
->flags
) &&
8665 atomic_read(&rdev
->nr_pending
)==0) {
8666 /* Faulty non-Blocked devices with nr_pending == 0
8667 * never get nr_pending incremented,
8668 * never get Faulty cleared, and never get Blocked set.
8669 * So we can synchronize_rcu now rather than once per device
8672 set_bit(RemoveSynchronized
, &rdev
->flags
);
8678 rdev_for_each(rdev
, mddev
) {
8679 if ((this == NULL
|| rdev
== this) &&
8680 rdev
->raid_disk
>= 0 &&
8681 !test_bit(Blocked
, &rdev
->flags
) &&
8682 ((test_bit(RemoveSynchronized
, &rdev
->flags
) ||
8683 (!test_bit(In_sync
, &rdev
->flags
) &&
8684 !test_bit(Journal
, &rdev
->flags
))) &&
8685 atomic_read(&rdev
->nr_pending
)==0)) {
8686 if (mddev
->pers
->hot_remove_disk(
8687 mddev
, rdev
) == 0) {
8688 sysfs_unlink_rdev(mddev
, rdev
);
8689 rdev
->saved_raid_disk
= rdev
->raid_disk
;
8690 rdev
->raid_disk
= -1;
8694 if (remove_some
&& test_bit(RemoveSynchronized
, &rdev
->flags
))
8695 clear_bit(RemoveSynchronized
, &rdev
->flags
);
8698 if (removed
&& mddev
->kobj
.sd
)
8699 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
8701 if (this && removed
)
8704 rdev_for_each(rdev
, mddev
) {
8705 if (this && this != rdev
)
8707 if (test_bit(Candidate
, &rdev
->flags
))
8709 if (rdev
->raid_disk
>= 0 &&
8710 !test_bit(In_sync
, &rdev
->flags
) &&
8711 !test_bit(Journal
, &rdev
->flags
) &&
8712 !test_bit(Faulty
, &rdev
->flags
))
8714 if (rdev
->raid_disk
>= 0)
8716 if (test_bit(Faulty
, &rdev
->flags
))
8718 if (!test_bit(Journal
, &rdev
->flags
)) {
8720 ! (rdev
->saved_raid_disk
>= 0 &&
8721 !test_bit(Bitmap_sync
, &rdev
->flags
)))
8724 rdev
->recovery_offset
= 0;
8727 hot_add_disk(mddev
, rdev
) == 0) {
8728 if (sysfs_link_rdev(mddev
, rdev
))
8729 /* failure here is OK */;
8730 if (!test_bit(Journal
, &rdev
->flags
))
8732 md_new_event(mddev
);
8733 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8738 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8742 static void md_start_sync(struct work_struct
*ws
)
8744 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
8746 mddev
->sync_thread
= md_register_thread(md_do_sync
,
8749 if (!mddev
->sync_thread
) {
8750 pr_warn("%s: could not start resync thread...\n",
8752 /* leave the spares where they are, it shouldn't hurt */
8753 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8754 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8755 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8756 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8757 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8758 wake_up(&resync_wait
);
8759 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8761 if (mddev
->sysfs_action
)
8762 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8764 md_wakeup_thread(mddev
->sync_thread
);
8765 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8766 md_new_event(mddev
);
8770 * This routine is regularly called by all per-raid-array threads to
8771 * deal with generic issues like resync and super-block update.
8772 * Raid personalities that don't have a thread (linear/raid0) do not
8773 * need this as they never do any recovery or update the superblock.
8775 * It does not do any resync itself, but rather "forks" off other threads
8776 * to do that as needed.
8777 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8778 * "->recovery" and create a thread at ->sync_thread.
8779 * When the thread finishes it sets MD_RECOVERY_DONE
8780 * and wakeups up this thread which will reap the thread and finish up.
8781 * This thread also removes any faulty devices (with nr_pending == 0).
8783 * The overall approach is:
8784 * 1/ if the superblock needs updating, update it.
8785 * 2/ If a recovery thread is running, don't do anything else.
8786 * 3/ If recovery has finished, clean up, possibly marking spares active.
8787 * 4/ If there are any faulty devices, remove them.
8788 * 5/ If array is degraded, try to add spares devices
8789 * 6/ If array has spares or is not in-sync, start a resync thread.
8791 void md_check_recovery(struct mddev
*mddev
)
8793 if (test_bit(MD_ALLOW_SB_UPDATE
, &mddev
->flags
) && mddev
->sb_flags
) {
8794 /* Write superblock - thread that called mddev_suspend()
8795 * holds reconfig_mutex for us.
8797 set_bit(MD_UPDATING_SB
, &mddev
->flags
);
8798 smp_mb__after_atomic();
8799 if (test_bit(MD_ALLOW_SB_UPDATE
, &mddev
->flags
))
8800 md_update_sb(mddev
, 0);
8801 clear_bit_unlock(MD_UPDATING_SB
, &mddev
->flags
);
8802 wake_up(&mddev
->sb_wait
);
8805 if (mddev
->suspended
)
8809 md_bitmap_daemon_work(mddev
);
8811 if (signal_pending(current
)) {
8812 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
8813 pr_debug("md: %s in immediate safe mode\n",
8815 mddev
->safemode
= 2;
8817 flush_signals(current
);
8820 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
8823 (mddev
->sb_flags
& ~ (1<<MD_SB_CHANGE_PENDING
)) ||
8824 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8825 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8826 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
8827 (mddev
->safemode
== 2
8828 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
8832 if (mddev_trylock(mddev
)) {
8835 if (!mddev
->external
&& mddev
->safemode
== 1)
8836 mddev
->safemode
= 0;
8839 struct md_rdev
*rdev
;
8840 if (!mddev
->external
&& mddev
->in_sync
)
8841 /* 'Blocked' flag not needed as failed devices
8842 * will be recorded if array switched to read/write.
8843 * Leaving it set will prevent the device
8844 * from being removed.
8846 rdev_for_each(rdev
, mddev
)
8847 clear_bit(Blocked
, &rdev
->flags
);
8848 /* On a read-only array we can:
8849 * - remove failed devices
8850 * - add already-in_sync devices if the array itself
8852 * As we only add devices that are already in-sync,
8853 * we can activate the spares immediately.
8855 remove_and_add_spares(mddev
, NULL
);
8856 /* There is no thread, but we need to call
8857 * ->spare_active and clear saved_raid_disk
8859 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8860 md_reap_sync_thread(mddev
);
8861 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8862 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8863 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8867 if (mddev_is_clustered(mddev
)) {
8868 struct md_rdev
*rdev
;
8869 /* kick the device if another node issued a
8872 rdev_for_each(rdev
, mddev
) {
8873 if (test_and_clear_bit(ClusterRemove
, &rdev
->flags
) &&
8874 rdev
->raid_disk
< 0)
8875 md_kick_rdev_from_array(rdev
);
8879 if (!mddev
->external
&& !mddev
->in_sync
) {
8880 spin_lock(&mddev
->lock
);
8882 spin_unlock(&mddev
->lock
);
8885 if (mddev
->sb_flags
)
8886 md_update_sb(mddev
, 0);
8888 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
8889 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
8890 /* resync/recovery still happening */
8891 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8894 if (mddev
->sync_thread
) {
8895 md_reap_sync_thread(mddev
);
8898 /* Set RUNNING before clearing NEEDED to avoid
8899 * any transients in the value of "sync_action".
8901 mddev
->curr_resync_completed
= 0;
8902 spin_lock(&mddev
->lock
);
8903 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8904 spin_unlock(&mddev
->lock
);
8905 /* Clear some bits that don't mean anything, but
8908 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8909 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8911 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8912 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
8914 /* no recovery is running.
8915 * remove any failed drives, then
8916 * add spares if possible.
8917 * Spares are also removed and re-added, to allow
8918 * the personality to fail the re-add.
8921 if (mddev
->reshape_position
!= MaxSector
) {
8922 if (mddev
->pers
->check_reshape
== NULL
||
8923 mddev
->pers
->check_reshape(mddev
) != 0)
8924 /* Cannot proceed */
8926 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8927 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8928 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
8929 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8930 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8931 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8932 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8933 } else if (mddev
->recovery_cp
< MaxSector
) {
8934 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8935 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8936 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
8937 /* nothing to be done ... */
8940 if (mddev
->pers
->sync_request
) {
8942 /* We are adding a device or devices to an array
8943 * which has the bitmap stored on all devices.
8944 * So make sure all bitmap pages get written
8946 md_bitmap_write_all(mddev
->bitmap
);
8948 INIT_WORK(&mddev
->del_work
, md_start_sync
);
8949 queue_work(md_misc_wq
, &mddev
->del_work
);
8953 if (!mddev
->sync_thread
) {
8954 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8955 wake_up(&resync_wait
);
8956 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8958 if (mddev
->sysfs_action
)
8959 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8962 wake_up(&mddev
->sb_wait
);
8963 mddev_unlock(mddev
);
8966 EXPORT_SYMBOL(md_check_recovery
);
8968 void md_reap_sync_thread(struct mddev
*mddev
)
8970 struct md_rdev
*rdev
;
8971 sector_t old_dev_sectors
= mddev
->dev_sectors
;
8972 bool is_reshaped
= false;
8974 /* resync has finished, collect result */
8975 md_unregister_thread(&mddev
->sync_thread
);
8976 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8977 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8979 /* activate any spares */
8980 if (mddev
->pers
->spare_active(mddev
)) {
8981 sysfs_notify(&mddev
->kobj
, NULL
,
8983 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8986 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8987 mddev
->pers
->finish_reshape
) {
8988 mddev
->pers
->finish_reshape(mddev
);
8989 if (mddev_is_clustered(mddev
))
8993 /* If array is no-longer degraded, then any saved_raid_disk
8994 * information must be scrapped.
8996 if (!mddev
->degraded
)
8997 rdev_for_each(rdev
, mddev
)
8998 rdev
->saved_raid_disk
= -1;
9000 md_update_sb(mddev
, 1);
9001 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9002 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9004 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED
, &mddev
->flags
))
9005 md_cluster_ops
->resync_finish(mddev
);
9006 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
9007 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
9008 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
9009 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
9010 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
9011 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
9013 * We call md_cluster_ops->update_size here because sync_size could
9014 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9015 * so it is time to update size across cluster.
9017 if (mddev_is_clustered(mddev
) && is_reshaped
9018 && !test_bit(MD_CLOSING
, &mddev
->flags
))
9019 md_cluster_ops
->update_size(mddev
, old_dev_sectors
);
9020 wake_up(&resync_wait
);
9021 /* flag recovery needed just to double check */
9022 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
9023 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
9024 md_new_event(mddev
);
9025 if (mddev
->event_work
.func
)
9026 queue_work(md_misc_wq
, &mddev
->event_work
);
9028 EXPORT_SYMBOL(md_reap_sync_thread
);
9030 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
9032 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
9033 wait_event_timeout(rdev
->blocked_wait
,
9034 !test_bit(Blocked
, &rdev
->flags
) &&
9035 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
9036 msecs_to_jiffies(5000));
9037 rdev_dec_pending(rdev
, mddev
);
9039 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
9041 void md_finish_reshape(struct mddev
*mddev
)
9043 /* called be personality module when reshape completes. */
9044 struct md_rdev
*rdev
;
9046 rdev_for_each(rdev
, mddev
) {
9047 if (rdev
->data_offset
> rdev
->new_data_offset
)
9048 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
9050 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
9051 rdev
->data_offset
= rdev
->new_data_offset
;
9054 EXPORT_SYMBOL(md_finish_reshape
);
9056 /* Bad block management */
9058 /* Returns 1 on success, 0 on failure */
9059 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
9062 struct mddev
*mddev
= rdev
->mddev
;
9065 s
+= rdev
->new_data_offset
;
9067 s
+= rdev
->data_offset
;
9068 rv
= badblocks_set(&rdev
->badblocks
, s
, sectors
, 0);
9070 /* Make sure they get written out promptly */
9071 if (test_bit(ExternalBbl
, &rdev
->flags
))
9072 sysfs_notify(&rdev
->kobj
, NULL
,
9073 "unacknowledged_bad_blocks");
9074 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
9075 set_mask_bits(&mddev
->sb_flags
, 0,
9076 BIT(MD_SB_CHANGE_CLEAN
) | BIT(MD_SB_CHANGE_PENDING
));
9077 md_wakeup_thread(rdev
->mddev
->thread
);
9082 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
9084 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
9089 s
+= rdev
->new_data_offset
;
9091 s
+= rdev
->data_offset
;
9092 rv
= badblocks_clear(&rdev
->badblocks
, s
, sectors
);
9093 if ((rv
== 0) && test_bit(ExternalBbl
, &rdev
->flags
))
9094 sysfs_notify(&rdev
->kobj
, NULL
, "bad_blocks");
9097 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
9099 static int md_notify_reboot(struct notifier_block
*this,
9100 unsigned long code
, void *x
)
9102 struct list_head
*tmp
;
9103 struct mddev
*mddev
;
9106 for_each_mddev(mddev
, tmp
) {
9107 if (mddev_trylock(mddev
)) {
9109 __md_stop_writes(mddev
);
9110 if (mddev
->persistent
)
9111 mddev
->safemode
= 2;
9112 mddev_unlock(mddev
);
9117 * certain more exotic SCSI devices are known to be
9118 * volatile wrt too early system reboots. While the
9119 * right place to handle this issue is the given
9120 * driver, we do want to have a safe RAID driver ...
9128 static struct notifier_block md_notifier
= {
9129 .notifier_call
= md_notify_reboot
,
9131 .priority
= INT_MAX
, /* before any real devices */
9134 static void md_geninit(void)
9136 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
9138 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
9141 static int __init
md_init(void)
9145 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
9149 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
9153 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
9156 if ((ret
= register_blkdev(0, "mdp")) < 0)
9160 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
9161 md_probe
, NULL
, NULL
);
9162 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
9163 md_probe
, NULL
, NULL
);
9165 register_reboot_notifier(&md_notifier
);
9166 raid_table_header
= register_sysctl_table(raid_root_table
);
9172 unregister_blkdev(MD_MAJOR
, "md");
9174 destroy_workqueue(md_misc_wq
);
9176 destroy_workqueue(md_wq
);
9181 static void check_sb_changes(struct mddev
*mddev
, struct md_rdev
*rdev
)
9183 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
9184 struct md_rdev
*rdev2
;
9186 char b
[BDEVNAME_SIZE
];
9189 * If size is changed in another node then we need to
9190 * do resize as well.
9192 if (mddev
->dev_sectors
!= le64_to_cpu(sb
->size
)) {
9193 ret
= mddev
->pers
->resize(mddev
, le64_to_cpu(sb
->size
));
9195 pr_info("md-cluster: resize failed\n");
9197 md_bitmap_update_sb(mddev
->bitmap
);
9200 /* Check for change of roles in the active devices */
9201 rdev_for_each(rdev2
, mddev
) {
9202 if (test_bit(Faulty
, &rdev2
->flags
))
9205 /* Check if the roles changed */
9206 role
= le16_to_cpu(sb
->dev_roles
[rdev2
->desc_nr
]);
9208 if (test_bit(Candidate
, &rdev2
->flags
)) {
9209 if (role
== 0xfffe) {
9210 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2
->bdev
,b
));
9211 md_kick_rdev_from_array(rdev2
);
9215 clear_bit(Candidate
, &rdev2
->flags
);
9218 if (role
!= rdev2
->raid_disk
) {
9220 * got activated except reshape is happening.
9222 if (rdev2
->raid_disk
== -1 && role
!= 0xffff &&
9223 !(le32_to_cpu(sb
->feature_map
) &
9224 MD_FEATURE_RESHAPE_ACTIVE
)) {
9225 rdev2
->saved_raid_disk
= role
;
9226 ret
= remove_and_add_spares(mddev
, rdev2
);
9227 pr_info("Activated spare: %s\n",
9228 bdevname(rdev2
->bdev
,b
));
9229 /* wakeup mddev->thread here, so array could
9230 * perform resync with the new activated disk */
9231 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
9232 md_wakeup_thread(mddev
->thread
);
9236 * We just want to do the minimum to mark the disk
9237 * as faulty. The recovery is performed by the
9238 * one who initiated the error.
9240 if ((role
== 0xfffe) || (role
== 0xfffd)) {
9241 md_error(mddev
, rdev2
);
9242 clear_bit(Blocked
, &rdev2
->flags
);
9247 if (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
))
9248 update_raid_disks(mddev
, le32_to_cpu(sb
->raid_disks
));
9251 * Since mddev->delta_disks has already updated in update_raid_disks,
9252 * so it is time to check reshape.
9254 if (test_bit(MD_RESYNCING_REMOTE
, &mddev
->recovery
) &&
9255 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
9257 * reshape is happening in the remote node, we need to
9258 * update reshape_position and call start_reshape.
9260 mddev
->reshape_position
= sb
->reshape_position
;
9261 if (mddev
->pers
->update_reshape_pos
)
9262 mddev
->pers
->update_reshape_pos(mddev
);
9263 if (mddev
->pers
->start_reshape
)
9264 mddev
->pers
->start_reshape(mddev
);
9265 } else if (test_bit(MD_RESYNCING_REMOTE
, &mddev
->recovery
) &&
9266 mddev
->reshape_position
!= MaxSector
&&
9267 !(le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
9268 /* reshape is just done in another node. */
9269 mddev
->reshape_position
= MaxSector
;
9270 if (mddev
->pers
->update_reshape_pos
)
9271 mddev
->pers
->update_reshape_pos(mddev
);
9274 /* Finally set the event to be up to date */
9275 mddev
->events
= le64_to_cpu(sb
->events
);
9278 static int read_rdev(struct mddev
*mddev
, struct md_rdev
*rdev
)
9281 struct page
*swapout
= rdev
->sb_page
;
9282 struct mdp_superblock_1
*sb
;
9284 /* Store the sb page of the rdev in the swapout temporary
9285 * variable in case we err in the future
9287 rdev
->sb_page
= NULL
;
9288 err
= alloc_disk_sb(rdev
);
9290 ClearPageUptodate(rdev
->sb_page
);
9291 rdev
->sb_loaded
= 0;
9292 err
= super_types
[mddev
->major_version
].
9293 load_super(rdev
, NULL
, mddev
->minor_version
);
9296 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9297 __func__
, __LINE__
, rdev
->desc_nr
, err
);
9299 put_page(rdev
->sb_page
);
9300 rdev
->sb_page
= swapout
;
9301 rdev
->sb_loaded
= 1;
9305 sb
= page_address(rdev
->sb_page
);
9306 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9310 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
))
9311 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
9313 /* The other node finished recovery, call spare_active to set
9314 * device In_sync and mddev->degraded
9316 if (rdev
->recovery_offset
== MaxSector
&&
9317 !test_bit(In_sync
, &rdev
->flags
) &&
9318 mddev
->pers
->spare_active(mddev
))
9319 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
9325 void md_reload_sb(struct mddev
*mddev
, int nr
)
9327 struct md_rdev
*rdev
;
9331 rdev_for_each_rcu(rdev
, mddev
) {
9332 if (rdev
->desc_nr
== nr
)
9336 if (!rdev
|| rdev
->desc_nr
!= nr
) {
9337 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__
, __LINE__
, nr
);
9341 err
= read_rdev(mddev
, rdev
);
9345 check_sb_changes(mddev
, rdev
);
9347 /* Read all rdev's to update recovery_offset */
9348 rdev_for_each_rcu(rdev
, mddev
) {
9349 if (!test_bit(Faulty
, &rdev
->flags
))
9350 read_rdev(mddev
, rdev
);
9353 EXPORT_SYMBOL(md_reload_sb
);
9358 * Searches all registered partitions for autorun RAID arrays
9362 static DEFINE_MUTEX(detected_devices_mutex
);
9363 static LIST_HEAD(all_detected_devices
);
9364 struct detected_devices_node
{
9365 struct list_head list
;
9369 void md_autodetect_dev(dev_t dev
)
9371 struct detected_devices_node
*node_detected_dev
;
9373 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
9374 if (node_detected_dev
) {
9375 node_detected_dev
->dev
= dev
;
9376 mutex_lock(&detected_devices_mutex
);
9377 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
9378 mutex_unlock(&detected_devices_mutex
);
9382 static void autostart_arrays(int part
)
9384 struct md_rdev
*rdev
;
9385 struct detected_devices_node
*node_detected_dev
;
9387 int i_scanned
, i_passed
;
9392 pr_info("md: Autodetecting RAID arrays.\n");
9394 mutex_lock(&detected_devices_mutex
);
9395 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
9397 node_detected_dev
= list_entry(all_detected_devices
.next
,
9398 struct detected_devices_node
, list
);
9399 list_del(&node_detected_dev
->list
);
9400 dev
= node_detected_dev
->dev
;
9401 kfree(node_detected_dev
);
9402 mutex_unlock(&detected_devices_mutex
);
9403 rdev
= md_import_device(dev
,0, 90);
9404 mutex_lock(&detected_devices_mutex
);
9408 if (test_bit(Faulty
, &rdev
->flags
))
9411 set_bit(AutoDetected
, &rdev
->flags
);
9412 list_add(&rdev
->same_set
, &pending_raid_disks
);
9415 mutex_unlock(&detected_devices_mutex
);
9417 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned
, i_passed
);
9419 autorun_devices(part
);
9422 #endif /* !MODULE */
9424 static __exit
void md_exit(void)
9426 struct mddev
*mddev
;
9427 struct list_head
*tmp
;
9430 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
9431 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
9433 unregister_blkdev(MD_MAJOR
,"md");
9434 unregister_blkdev(mdp_major
, "mdp");
9435 unregister_reboot_notifier(&md_notifier
);
9436 unregister_sysctl_table(raid_table_header
);
9438 /* We cannot unload the modules while some process is
9439 * waiting for us in select() or poll() - wake them up
9442 while (waitqueue_active(&md_event_waiters
)) {
9443 /* not safe to leave yet */
9444 wake_up(&md_event_waiters
);
9448 remove_proc_entry("mdstat", NULL
);
9450 for_each_mddev(mddev
, tmp
) {
9451 export_array(mddev
);
9453 mddev
->hold_active
= 0;
9455 * for_each_mddev() will call mddev_put() at the end of each
9456 * iteration. As the mddev is now fully clear, this will
9457 * schedule the mddev for destruction by a workqueue, and the
9458 * destroy_workqueue() below will wait for that to complete.
9461 destroy_workqueue(md_misc_wq
);
9462 destroy_workqueue(md_wq
);
9465 subsys_initcall(md_init
);
9466 module_exit(md_exit
)
9468 static int get_ro(char *buffer
, const struct kernel_param
*kp
)
9470 return sprintf(buffer
, "%d", start_readonly
);
9472 static int set_ro(const char *val
, const struct kernel_param
*kp
)
9474 return kstrtouint(val
, 10, (unsigned int *)&start_readonly
);
9477 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
9478 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
9479 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
9480 module_param(create_on_open
, bool, S_IRUSR
|S_IWUSR
);
9482 MODULE_LICENSE("GPL");
9483 MODULE_DESCRIPTION("MD RAID framework");
9485 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);