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>
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 struct md_cluster_operations
*md_cluster_ops
;
88 EXPORT_SYMBOL(md_cluster_ops
);
89 struct module
*md_cluster_mod
;
90 EXPORT_SYMBOL(md_cluster_mod
);
92 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
93 static struct workqueue_struct
*md_wq
;
94 static struct workqueue_struct
*md_misc_wq
;
96 static int remove_and_add_spares(struct mddev
*mddev
,
97 struct md_rdev
*this);
98 static void mddev_detach(struct mddev
*mddev
);
101 * Default number of read corrections we'll attempt on an rdev
102 * before ejecting it from the array. We divide the read error
103 * count by 2 for every hour elapsed between read errors.
105 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
107 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
108 * is 1000 KB/sec, so the extra system load does not show up that much.
109 * Increase it if you want to have more _guaranteed_ speed. Note that
110 * the RAID driver will use the maximum available bandwidth if the IO
111 * subsystem is idle. There is also an 'absolute maximum' reconstruction
112 * speed limit - in case reconstruction slows down your system despite
115 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
116 * or /sys/block/mdX/md/sync_speed_{min,max}
119 static int sysctl_speed_limit_min
= 1000;
120 static int sysctl_speed_limit_max
= 200000;
121 static inline int speed_min(struct mddev
*mddev
)
123 return mddev
->sync_speed_min
?
124 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
127 static inline int speed_max(struct mddev
*mddev
)
129 return mddev
->sync_speed_max
?
130 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
133 static struct ctl_table_header
*raid_table_header
;
135 static struct ctl_table raid_table
[] = {
137 .procname
= "speed_limit_min",
138 .data
= &sysctl_speed_limit_min
,
139 .maxlen
= sizeof(int),
140 .mode
= S_IRUGO
|S_IWUSR
,
141 .proc_handler
= proc_dointvec
,
144 .procname
= "speed_limit_max",
145 .data
= &sysctl_speed_limit_max
,
146 .maxlen
= sizeof(int),
147 .mode
= S_IRUGO
|S_IWUSR
,
148 .proc_handler
= proc_dointvec
,
153 static struct ctl_table raid_dir_table
[] = {
157 .mode
= S_IRUGO
|S_IXUGO
,
163 static struct ctl_table raid_root_table
[] = {
168 .child
= raid_dir_table
,
173 static const struct block_device_operations md_fops
;
175 static int start_readonly
;
178 * The original mechanism for creating an md device is to create
179 * a device node in /dev and to open it. This causes races with device-close.
180 * The preferred method is to write to the "new_array" module parameter.
181 * This can avoid races.
182 * Setting create_on_open to false disables the original mechanism
183 * so all the races disappear.
185 static bool create_on_open
= true;
188 * like bio_clone_bioset, but with a local bio set
191 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
196 if (!mddev
|| !mddev
->bio_set
)
197 return bio_alloc(gfp_mask
, nr_iovecs
);
199 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
204 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
206 static struct bio
*md_bio_alloc_sync(struct mddev
*mddev
)
208 if (!mddev
|| !mddev
->sync_set
)
209 return bio_alloc(GFP_NOIO
, 1);
211 return bio_alloc_bioset(GFP_NOIO
, 1, mddev
->sync_set
);
215 * We have a system wide 'event count' that is incremented
216 * on any 'interesting' event, and readers of /proc/mdstat
217 * can use 'poll' or 'select' to find out when the event
221 * start array, stop array, error, add device, remove device,
222 * start build, activate spare
224 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
225 static atomic_t md_event_count
;
226 void md_new_event(struct mddev
*mddev
)
228 atomic_inc(&md_event_count
);
229 wake_up(&md_event_waiters
);
231 EXPORT_SYMBOL_GPL(md_new_event
);
234 * Enables to iterate over all existing md arrays
235 * all_mddevs_lock protects this list.
237 static LIST_HEAD(all_mddevs
);
238 static DEFINE_SPINLOCK(all_mddevs_lock
);
241 * iterates through all used mddevs in the system.
242 * We take care to grab the all_mddevs_lock whenever navigating
243 * the list, and to always hold a refcount when unlocked.
244 * Any code which breaks out of this loop while own
245 * a reference to the current mddev and must mddev_put it.
247 #define for_each_mddev(_mddev,_tmp) \
249 for (({ spin_lock(&all_mddevs_lock); \
250 _tmp = all_mddevs.next; \
252 ({ if (_tmp != &all_mddevs) \
253 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
254 spin_unlock(&all_mddevs_lock); \
255 if (_mddev) mddev_put(_mddev); \
256 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
257 _tmp != &all_mddevs;}); \
258 ({ spin_lock(&all_mddevs_lock); \
259 _tmp = _tmp->next;}) \
262 /* Rather than calling directly into the personality make_request function,
263 * IO requests come here first so that we can check if the device is
264 * being suspended pending a reconfiguration.
265 * We hold a refcount over the call to ->make_request. By the time that
266 * call has finished, the bio has been linked into some internal structure
267 * and so is visible to ->quiesce(), so we don't need the refcount any more.
269 void md_handle_request(struct mddev
*mddev
, struct bio
*bio
)
273 if (mddev
->suspended
) {
276 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
277 TASK_UNINTERRUPTIBLE
);
278 if (!mddev
->suspended
)
284 finish_wait(&mddev
->sb_wait
, &__wait
);
286 atomic_inc(&mddev
->active_io
);
289 if (!mddev
->pers
->make_request(mddev
, bio
)) {
290 atomic_dec(&mddev
->active_io
);
291 wake_up(&mddev
->sb_wait
);
292 goto check_suspended
;
295 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
296 wake_up(&mddev
->sb_wait
);
298 EXPORT_SYMBOL(md_handle_request
);
300 static blk_qc_t
md_make_request(struct request_queue
*q
, struct bio
*bio
)
302 const int rw
= bio_data_dir(bio
);
303 struct mddev
*mddev
= q
->queuedata
;
304 unsigned int sectors
;
307 blk_queue_split(q
, &bio
);
309 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
311 return BLK_QC_T_NONE
;
313 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
314 if (bio_sectors(bio
) != 0)
315 bio
->bi_status
= BLK_STS_IOERR
;
317 return BLK_QC_T_NONE
;
321 * save the sectors now since our bio can
322 * go away inside make_request
324 sectors
= bio_sectors(bio
);
325 /* bio could be mergeable after passing to underlayer */
326 bio
->bi_opf
&= ~REQ_NOMERGE
;
328 md_handle_request(mddev
, bio
);
330 cpu
= part_stat_lock();
331 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
332 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
335 return BLK_QC_T_NONE
;
338 /* mddev_suspend makes sure no new requests are submitted
339 * to the device, and that any requests that have been submitted
340 * are completely handled.
341 * Once mddev_detach() is called and completes, the module will be
344 void mddev_suspend(struct mddev
*mddev
)
346 WARN_ON_ONCE(mddev
->thread
&& current
== mddev
->thread
->tsk
);
347 if (mddev
->suspended
++)
350 wake_up(&mddev
->sb_wait
);
351 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
352 mddev
->pers
->quiesce(mddev
, 1);
354 del_timer_sync(&mddev
->safemode_timer
);
356 EXPORT_SYMBOL_GPL(mddev_suspend
);
358 void mddev_resume(struct mddev
*mddev
)
360 if (--mddev
->suspended
)
362 wake_up(&mddev
->sb_wait
);
363 mddev
->pers
->quiesce(mddev
, 0);
365 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
366 md_wakeup_thread(mddev
->thread
);
367 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
369 EXPORT_SYMBOL_GPL(mddev_resume
);
371 int mddev_congested(struct mddev
*mddev
, int bits
)
373 struct md_personality
*pers
= mddev
->pers
;
377 if (mddev
->suspended
)
379 else if (pers
&& pers
->congested
)
380 ret
= pers
->congested(mddev
, bits
);
384 EXPORT_SYMBOL_GPL(mddev_congested
);
385 static int md_congested(void *data
, int bits
)
387 struct mddev
*mddev
= data
;
388 return mddev_congested(mddev
, bits
);
392 * Generic flush handling for md
395 static void md_end_flush(struct bio
*bio
)
397 struct md_rdev
*rdev
= bio
->bi_private
;
398 struct mddev
*mddev
= rdev
->mddev
;
400 rdev_dec_pending(rdev
, mddev
);
402 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
403 /* The pre-request flush has finished */
404 queue_work(md_wq
, &mddev
->flush_work
);
409 static void md_submit_flush_data(struct work_struct
*ws
);
411 static void submit_flushes(struct work_struct
*ws
)
413 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
414 struct md_rdev
*rdev
;
416 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
417 atomic_set(&mddev
->flush_pending
, 1);
419 rdev_for_each_rcu(rdev
, mddev
)
420 if (rdev
->raid_disk
>= 0 &&
421 !test_bit(Faulty
, &rdev
->flags
)) {
422 /* Take two references, one is dropped
423 * when request finishes, one after
424 * we reclaim rcu_read_lock
427 atomic_inc(&rdev
->nr_pending
);
428 atomic_inc(&rdev
->nr_pending
);
430 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
431 bi
->bi_end_io
= md_end_flush
;
432 bi
->bi_private
= rdev
;
433 bio_set_dev(bi
, rdev
->bdev
);
434 bi
->bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
;
435 atomic_inc(&mddev
->flush_pending
);
438 rdev_dec_pending(rdev
, mddev
);
441 if (atomic_dec_and_test(&mddev
->flush_pending
))
442 queue_work(md_wq
, &mddev
->flush_work
);
445 static void md_submit_flush_data(struct work_struct
*ws
)
447 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
448 struct bio
*bio
= mddev
->flush_bio
;
451 * must reset flush_bio before calling into md_handle_request to avoid a
452 * deadlock, because other bios passed md_handle_request suspend check
453 * could wait for this and below md_handle_request could wait for those
454 * bios because of suspend check
456 mddev
->flush_bio
= NULL
;
457 wake_up(&mddev
->sb_wait
);
459 if (bio
->bi_iter
.bi_size
== 0)
460 /* an empty barrier - all done */
463 bio
->bi_opf
&= ~REQ_PREFLUSH
;
464 md_handle_request(mddev
, bio
);
468 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
470 spin_lock_irq(&mddev
->lock
);
471 wait_event_lock_irq(mddev
->sb_wait
,
474 mddev
->flush_bio
= bio
;
475 spin_unlock_irq(&mddev
->lock
);
477 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
478 queue_work(md_wq
, &mddev
->flush_work
);
480 EXPORT_SYMBOL(md_flush_request
);
482 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
484 atomic_inc(&mddev
->active
);
488 static void mddev_delayed_delete(struct work_struct
*ws
);
490 static void mddev_put(struct mddev
*mddev
)
492 struct bio_set
*bs
= NULL
, *sync_bs
= NULL
;
494 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
496 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
497 mddev
->ctime
== 0 && !mddev
->hold_active
) {
498 /* Array is not configured at all, and not held active,
500 list_del_init(&mddev
->all_mddevs
);
502 sync_bs
= mddev
->sync_set
;
503 mddev
->bio_set
= NULL
;
504 mddev
->sync_set
= NULL
;
505 if (mddev
->gendisk
) {
506 /* We did a probe so need to clean up. Call
507 * queue_work inside the spinlock so that
508 * flush_workqueue() after mddev_find will
509 * succeed in waiting for the work to be done.
511 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
512 queue_work(md_misc_wq
, &mddev
->del_work
);
516 spin_unlock(&all_mddevs_lock
);
520 bioset_free(sync_bs
);
523 static void md_safemode_timeout(unsigned long data
);
525 void mddev_init(struct mddev
*mddev
)
527 mutex_init(&mddev
->open_mutex
);
528 mutex_init(&mddev
->reconfig_mutex
);
529 mutex_init(&mddev
->bitmap_info
.mutex
);
530 INIT_LIST_HEAD(&mddev
->disks
);
531 INIT_LIST_HEAD(&mddev
->all_mddevs
);
532 setup_timer(&mddev
->safemode_timer
, md_safemode_timeout
,
533 (unsigned long) mddev
);
534 atomic_set(&mddev
->active
, 1);
535 atomic_set(&mddev
->openers
, 0);
536 atomic_set(&mddev
->active_io
, 0);
537 spin_lock_init(&mddev
->lock
);
538 atomic_set(&mddev
->flush_pending
, 0);
539 init_waitqueue_head(&mddev
->sb_wait
);
540 init_waitqueue_head(&mddev
->recovery_wait
);
541 mddev
->reshape_position
= MaxSector
;
542 mddev
->reshape_backwards
= 0;
543 mddev
->last_sync_action
= "none";
544 mddev
->resync_min
= 0;
545 mddev
->resync_max
= MaxSector
;
546 mddev
->level
= LEVEL_NONE
;
548 EXPORT_SYMBOL_GPL(mddev_init
);
550 static struct mddev
*mddev_find(dev_t unit
)
552 struct mddev
*mddev
, *new = NULL
;
554 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
555 unit
&= ~((1<<MdpMinorShift
)-1);
558 spin_lock(&all_mddevs_lock
);
561 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
562 if (mddev
->unit
== unit
) {
564 spin_unlock(&all_mddevs_lock
);
570 list_add(&new->all_mddevs
, &all_mddevs
);
571 spin_unlock(&all_mddevs_lock
);
572 new->hold_active
= UNTIL_IOCTL
;
576 /* find an unused unit number */
577 static int next_minor
= 512;
578 int start
= next_minor
;
582 dev
= MKDEV(MD_MAJOR
, next_minor
);
584 if (next_minor
> MINORMASK
)
586 if (next_minor
== start
) {
587 /* Oh dear, all in use. */
588 spin_unlock(&all_mddevs_lock
);
594 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
595 if (mddev
->unit
== dev
) {
601 new->md_minor
= MINOR(dev
);
602 new->hold_active
= UNTIL_STOP
;
603 list_add(&new->all_mddevs
, &all_mddevs
);
604 spin_unlock(&all_mddevs_lock
);
607 spin_unlock(&all_mddevs_lock
);
609 new = kzalloc(sizeof(*new), GFP_KERNEL
);
614 if (MAJOR(unit
) == MD_MAJOR
)
615 new->md_minor
= MINOR(unit
);
617 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
624 static struct attribute_group md_redundancy_group
;
626 void mddev_unlock(struct mddev
*mddev
)
628 if (mddev
->to_remove
) {
629 /* These cannot be removed under reconfig_mutex as
630 * an access to the files will try to take reconfig_mutex
631 * while holding the file unremovable, which leads to
633 * So hold set sysfs_active while the remove in happeing,
634 * and anything else which might set ->to_remove or my
635 * otherwise change the sysfs namespace will fail with
636 * -EBUSY if sysfs_active is still set.
637 * We set sysfs_active under reconfig_mutex and elsewhere
638 * test it under the same mutex to ensure its correct value
641 struct attribute_group
*to_remove
= mddev
->to_remove
;
642 mddev
->to_remove
= NULL
;
643 mddev
->sysfs_active
= 1;
644 mutex_unlock(&mddev
->reconfig_mutex
);
646 if (mddev
->kobj
.sd
) {
647 if (to_remove
!= &md_redundancy_group
)
648 sysfs_remove_group(&mddev
->kobj
, to_remove
);
649 if (mddev
->pers
== NULL
||
650 mddev
->pers
->sync_request
== NULL
) {
651 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
652 if (mddev
->sysfs_action
)
653 sysfs_put(mddev
->sysfs_action
);
654 mddev
->sysfs_action
= NULL
;
657 mddev
->sysfs_active
= 0;
659 mutex_unlock(&mddev
->reconfig_mutex
);
661 /* As we've dropped the mutex we need a spinlock to
662 * make sure the thread doesn't disappear
664 spin_lock(&pers_lock
);
665 md_wakeup_thread(mddev
->thread
);
666 spin_unlock(&pers_lock
);
668 EXPORT_SYMBOL_GPL(mddev_unlock
);
670 struct md_rdev
*md_find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
672 struct md_rdev
*rdev
;
674 rdev_for_each_rcu(rdev
, mddev
)
675 if (rdev
->desc_nr
== nr
)
680 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu
);
682 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
684 struct md_rdev
*rdev
;
686 rdev_for_each(rdev
, mddev
)
687 if (rdev
->bdev
->bd_dev
== dev
)
693 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
695 struct md_rdev
*rdev
;
697 rdev_for_each_rcu(rdev
, mddev
)
698 if (rdev
->bdev
->bd_dev
== dev
)
704 static struct md_personality
*find_pers(int level
, char *clevel
)
706 struct md_personality
*pers
;
707 list_for_each_entry(pers
, &pers_list
, list
) {
708 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
710 if (strcmp(pers
->name
, clevel
)==0)
716 /* return the offset of the super block in 512byte sectors */
717 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
719 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
720 return MD_NEW_SIZE_SECTORS(num_sectors
);
723 static int alloc_disk_sb(struct md_rdev
*rdev
)
725 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
731 void md_rdev_clear(struct md_rdev
*rdev
)
734 put_page(rdev
->sb_page
);
736 rdev
->sb_page
= NULL
;
741 put_page(rdev
->bb_page
);
742 rdev
->bb_page
= NULL
;
744 badblocks_exit(&rdev
->badblocks
);
746 EXPORT_SYMBOL_GPL(md_rdev_clear
);
748 static void super_written(struct bio
*bio
)
750 struct md_rdev
*rdev
= bio
->bi_private
;
751 struct mddev
*mddev
= rdev
->mddev
;
753 if (bio
->bi_status
) {
754 pr_err("md: super_written gets error=%d\n", bio
->bi_status
);
755 md_error(mddev
, rdev
);
756 if (!test_bit(Faulty
, &rdev
->flags
)
757 && (bio
->bi_opf
& MD_FAILFAST
)) {
758 set_bit(MD_SB_NEED_REWRITE
, &mddev
->sb_flags
);
759 set_bit(LastDev
, &rdev
->flags
);
762 clear_bit(LastDev
, &rdev
->flags
);
764 if (atomic_dec_and_test(&mddev
->pending_writes
))
765 wake_up(&mddev
->sb_wait
);
766 rdev_dec_pending(rdev
, mddev
);
770 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
771 sector_t sector
, int size
, struct page
*page
)
773 /* write first size bytes of page to sector of rdev
774 * Increment mddev->pending_writes before returning
775 * and decrement it on completion, waking up sb_wait
776 * if zero is reached.
777 * If an error occurred, call md_error
782 if (test_bit(Faulty
, &rdev
->flags
))
785 bio
= md_bio_alloc_sync(mddev
);
787 atomic_inc(&rdev
->nr_pending
);
789 bio_set_dev(bio
, rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
);
790 bio
->bi_iter
.bi_sector
= sector
;
791 bio_add_page(bio
, page
, size
, 0);
792 bio
->bi_private
= rdev
;
793 bio
->bi_end_io
= super_written
;
795 if (test_bit(MD_FAILFAST_SUPPORTED
, &mddev
->flags
) &&
796 test_bit(FailFast
, &rdev
->flags
) &&
797 !test_bit(LastDev
, &rdev
->flags
))
799 bio
->bi_opf
= REQ_OP_WRITE
| REQ_SYNC
| REQ_PREFLUSH
| REQ_FUA
| ff
;
801 atomic_inc(&mddev
->pending_writes
);
805 int md_super_wait(struct mddev
*mddev
)
807 /* wait for all superblock writes that were scheduled to complete */
808 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
809 if (test_and_clear_bit(MD_SB_NEED_REWRITE
, &mddev
->sb_flags
))
814 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
815 struct page
*page
, int op
, int op_flags
, bool metadata_op
)
817 struct bio
*bio
= md_bio_alloc_sync(rdev
->mddev
);
820 if (metadata_op
&& rdev
->meta_bdev
)
821 bio_set_dev(bio
, rdev
->meta_bdev
);
823 bio_set_dev(bio
, rdev
->bdev
);
824 bio_set_op_attrs(bio
, op
, op_flags
);
826 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
827 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
828 (rdev
->mddev
->reshape_backwards
==
829 (sector
>= rdev
->mddev
->reshape_position
)))
830 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
832 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
833 bio_add_page(bio
, page
, size
, 0);
835 submit_bio_wait(bio
);
837 ret
= !bio
->bi_status
;
841 EXPORT_SYMBOL_GPL(sync_page_io
);
843 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
845 char b
[BDEVNAME_SIZE
];
850 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, REQ_OP_READ
, 0, true))
856 pr_err("md: disabled device %s, could not read superblock.\n",
857 bdevname(rdev
->bdev
,b
));
861 static int md_uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
863 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
864 sb1
->set_uuid1
== sb2
->set_uuid1
&&
865 sb1
->set_uuid2
== sb2
->set_uuid2
&&
866 sb1
->set_uuid3
== sb2
->set_uuid3
;
869 static int md_sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
872 mdp_super_t
*tmp1
, *tmp2
;
874 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
875 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
877 if (!tmp1
|| !tmp2
) {
886 * nr_disks is not constant
891 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
898 static u32
md_csum_fold(u32 csum
)
900 csum
= (csum
& 0xffff) + (csum
>> 16);
901 return (csum
& 0xffff) + (csum
>> 16);
904 static unsigned int calc_sb_csum(mdp_super_t
*sb
)
907 u32
*sb32
= (u32
*)sb
;
909 unsigned int disk_csum
, csum
;
911 disk_csum
= sb
->sb_csum
;
914 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
916 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
919 /* This used to use csum_partial, which was wrong for several
920 * reasons including that different results are returned on
921 * different architectures. It isn't critical that we get exactly
922 * the same return value as before (we always csum_fold before
923 * testing, and that removes any differences). However as we
924 * know that csum_partial always returned a 16bit value on
925 * alphas, do a fold to maximise conformity to previous behaviour.
927 sb
->sb_csum
= md_csum_fold(disk_csum
);
929 sb
->sb_csum
= disk_csum
;
935 * Handle superblock details.
936 * We want to be able to handle multiple superblock formats
937 * so we have a common interface to them all, and an array of
938 * different handlers.
939 * We rely on user-space to write the initial superblock, and support
940 * reading and updating of superblocks.
941 * Interface methods are:
942 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
943 * loads and validates a superblock on dev.
944 * if refdev != NULL, compare superblocks on both devices
946 * 0 - dev has a superblock that is compatible with refdev
947 * 1 - dev has a superblock that is compatible and newer than refdev
948 * so dev should be used as the refdev in future
949 * -EINVAL superblock incompatible or invalid
950 * -othererror e.g. -EIO
952 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
953 * Verify that dev is acceptable into mddev.
954 * The first time, mddev->raid_disks will be 0, and data from
955 * dev should be merged in. Subsequent calls check that dev
956 * is new enough. Return 0 or -EINVAL
958 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
959 * Update the superblock for rdev with data in mddev
960 * This does not write to disc.
966 struct module
*owner
;
967 int (*load_super
)(struct md_rdev
*rdev
,
968 struct md_rdev
*refdev
,
970 int (*validate_super
)(struct mddev
*mddev
,
971 struct md_rdev
*rdev
);
972 void (*sync_super
)(struct mddev
*mddev
,
973 struct md_rdev
*rdev
);
974 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
975 sector_t num_sectors
);
976 int (*allow_new_offset
)(struct md_rdev
*rdev
,
977 unsigned long long new_offset
);
981 * Check that the given mddev has no bitmap.
983 * This function is called from the run method of all personalities that do not
984 * support bitmaps. It prints an error message and returns non-zero if mddev
985 * has a bitmap. Otherwise, it returns 0.
988 int md_check_no_bitmap(struct mddev
*mddev
)
990 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
992 pr_warn("%s: bitmaps are not supported for %s\n",
993 mdname(mddev
), mddev
->pers
->name
);
996 EXPORT_SYMBOL(md_check_no_bitmap
);
999 * load_super for 0.90.0
1001 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1003 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1008 * Calculate the position of the superblock (512byte sectors),
1009 * it's at the end of the disk.
1011 * It also happens to be a multiple of 4Kb.
1013 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1015 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1021 bdevname(rdev
->bdev
, b
);
1022 sb
= page_address(rdev
->sb_page
);
1024 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1025 pr_warn("md: invalid raid superblock magic on %s\n", b
);
1029 if (sb
->major_version
!= 0 ||
1030 sb
->minor_version
< 90 ||
1031 sb
->minor_version
> 91) {
1032 pr_warn("Bad version number %d.%d on %s\n",
1033 sb
->major_version
, sb
->minor_version
, b
);
1037 if (sb
->raid_disks
<= 0)
1040 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1041 pr_warn("md: invalid superblock checksum on %s\n", b
);
1045 rdev
->preferred_minor
= sb
->md_minor
;
1046 rdev
->data_offset
= 0;
1047 rdev
->new_data_offset
= 0;
1048 rdev
->sb_size
= MD_SB_BYTES
;
1049 rdev
->badblocks
.shift
= -1;
1051 if (sb
->level
== LEVEL_MULTIPATH
)
1054 rdev
->desc_nr
= sb
->this_disk
.number
;
1060 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1061 if (!md_uuid_equal(refsb
, sb
)) {
1062 pr_warn("md: %s has different UUID to %s\n",
1063 b
, bdevname(refdev
->bdev
,b2
));
1066 if (!md_sb_equal(refsb
, sb
)) {
1067 pr_warn("md: %s has same UUID but different superblock to %s\n",
1068 b
, bdevname(refdev
->bdev
, b2
));
1072 ev2
= md_event(refsb
);
1078 rdev
->sectors
= rdev
->sb_start
;
1079 /* Limit to 4TB as metadata cannot record more than that.
1080 * (not needed for Linear and RAID0 as metadata doesn't
1083 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)rdev
->sectors
>= (2ULL << 32) &&
1085 rdev
->sectors
= (sector_t
)(2ULL << 32) - 2;
1087 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1088 /* "this cannot possibly happen" ... */
1096 * validate_super for 0.90.0
1098 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1101 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1102 __u64 ev1
= md_event(sb
);
1104 rdev
->raid_disk
= -1;
1105 clear_bit(Faulty
, &rdev
->flags
);
1106 clear_bit(In_sync
, &rdev
->flags
);
1107 clear_bit(Bitmap_sync
, &rdev
->flags
);
1108 clear_bit(WriteMostly
, &rdev
->flags
);
1110 if (mddev
->raid_disks
== 0) {
1111 mddev
->major_version
= 0;
1112 mddev
->minor_version
= sb
->minor_version
;
1113 mddev
->patch_version
= sb
->patch_version
;
1114 mddev
->external
= 0;
1115 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1116 mddev
->ctime
= sb
->ctime
;
1117 mddev
->utime
= sb
->utime
;
1118 mddev
->level
= sb
->level
;
1119 mddev
->clevel
[0] = 0;
1120 mddev
->layout
= sb
->layout
;
1121 mddev
->raid_disks
= sb
->raid_disks
;
1122 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1123 mddev
->events
= ev1
;
1124 mddev
->bitmap_info
.offset
= 0;
1125 mddev
->bitmap_info
.space
= 0;
1126 /* bitmap can use 60 K after the 4K superblocks */
1127 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1128 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1129 mddev
->reshape_backwards
= 0;
1131 if (mddev
->minor_version
>= 91) {
1132 mddev
->reshape_position
= sb
->reshape_position
;
1133 mddev
->delta_disks
= sb
->delta_disks
;
1134 mddev
->new_level
= sb
->new_level
;
1135 mddev
->new_layout
= sb
->new_layout
;
1136 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1137 if (mddev
->delta_disks
< 0)
1138 mddev
->reshape_backwards
= 1;
1140 mddev
->reshape_position
= MaxSector
;
1141 mddev
->delta_disks
= 0;
1142 mddev
->new_level
= mddev
->level
;
1143 mddev
->new_layout
= mddev
->layout
;
1144 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1147 if (sb
->state
& (1<<MD_SB_CLEAN
))
1148 mddev
->recovery_cp
= MaxSector
;
1150 if (sb
->events_hi
== sb
->cp_events_hi
&&
1151 sb
->events_lo
== sb
->cp_events_lo
) {
1152 mddev
->recovery_cp
= sb
->recovery_cp
;
1154 mddev
->recovery_cp
= 0;
1157 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1158 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1159 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1160 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1162 mddev
->max_disks
= MD_SB_DISKS
;
1164 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1165 mddev
->bitmap_info
.file
== NULL
) {
1166 mddev
->bitmap_info
.offset
=
1167 mddev
->bitmap_info
.default_offset
;
1168 mddev
->bitmap_info
.space
=
1169 mddev
->bitmap_info
.default_space
;
1172 } else if (mddev
->pers
== NULL
) {
1173 /* Insist on good event counter while assembling, except
1174 * for spares (which don't need an event count) */
1176 if (sb
->disks
[rdev
->desc_nr
].state
& (
1177 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1178 if (ev1
< mddev
->events
)
1180 } else if (mddev
->bitmap
) {
1181 /* if adding to array with a bitmap, then we can accept an
1182 * older device ... but not too old.
1184 if (ev1
< mddev
->bitmap
->events_cleared
)
1186 if (ev1
< mddev
->events
)
1187 set_bit(Bitmap_sync
, &rdev
->flags
);
1189 if (ev1
< mddev
->events
)
1190 /* just a hot-add of a new device, leave raid_disk at -1 */
1194 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1195 desc
= sb
->disks
+ rdev
->desc_nr
;
1197 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1198 set_bit(Faulty
, &rdev
->flags
);
1199 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1200 desc->raid_disk < mddev->raid_disks */) {
1201 set_bit(In_sync
, &rdev
->flags
);
1202 rdev
->raid_disk
= desc
->raid_disk
;
1203 rdev
->saved_raid_disk
= desc
->raid_disk
;
1204 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1205 /* active but not in sync implies recovery up to
1206 * reshape position. We don't know exactly where
1207 * that is, so set to zero for now */
1208 if (mddev
->minor_version
>= 91) {
1209 rdev
->recovery_offset
= 0;
1210 rdev
->raid_disk
= desc
->raid_disk
;
1213 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1214 set_bit(WriteMostly
, &rdev
->flags
);
1215 if (desc
->state
& (1<<MD_DISK_FAILFAST
))
1216 set_bit(FailFast
, &rdev
->flags
);
1217 } else /* MULTIPATH are always insync */
1218 set_bit(In_sync
, &rdev
->flags
);
1223 * sync_super for 0.90.0
1225 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1228 struct md_rdev
*rdev2
;
1229 int next_spare
= mddev
->raid_disks
;
1231 /* make rdev->sb match mddev data..
1234 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1235 * 3/ any empty disks < next_spare become removed
1237 * disks[0] gets initialised to REMOVED because
1238 * we cannot be sure from other fields if it has
1239 * been initialised or not.
1242 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1244 rdev
->sb_size
= MD_SB_BYTES
;
1246 sb
= page_address(rdev
->sb_page
);
1248 memset(sb
, 0, sizeof(*sb
));
1250 sb
->md_magic
= MD_SB_MAGIC
;
1251 sb
->major_version
= mddev
->major_version
;
1252 sb
->patch_version
= mddev
->patch_version
;
1253 sb
->gvalid_words
= 0; /* ignored */
1254 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1255 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1256 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1257 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1259 sb
->ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
1260 sb
->level
= mddev
->level
;
1261 sb
->size
= mddev
->dev_sectors
/ 2;
1262 sb
->raid_disks
= mddev
->raid_disks
;
1263 sb
->md_minor
= mddev
->md_minor
;
1264 sb
->not_persistent
= 0;
1265 sb
->utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
1267 sb
->events_hi
= (mddev
->events
>>32);
1268 sb
->events_lo
= (u32
)mddev
->events
;
1270 if (mddev
->reshape_position
== MaxSector
)
1271 sb
->minor_version
= 90;
1273 sb
->minor_version
= 91;
1274 sb
->reshape_position
= mddev
->reshape_position
;
1275 sb
->new_level
= mddev
->new_level
;
1276 sb
->delta_disks
= mddev
->delta_disks
;
1277 sb
->new_layout
= mddev
->new_layout
;
1278 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1280 mddev
->minor_version
= sb
->minor_version
;
1283 sb
->recovery_cp
= mddev
->recovery_cp
;
1284 sb
->cp_events_hi
= (mddev
->events
>>32);
1285 sb
->cp_events_lo
= (u32
)mddev
->events
;
1286 if (mddev
->recovery_cp
== MaxSector
)
1287 sb
->state
= (1<< MD_SB_CLEAN
);
1289 sb
->recovery_cp
= 0;
1291 sb
->layout
= mddev
->layout
;
1292 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1294 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1295 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1297 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1298 rdev_for_each(rdev2
, mddev
) {
1301 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1303 if (rdev2
->raid_disk
>= 0 &&
1304 sb
->minor_version
>= 91)
1305 /* we have nowhere to store the recovery_offset,
1306 * but if it is not below the reshape_position,
1307 * we can piggy-back on that.
1310 if (rdev2
->raid_disk
< 0 ||
1311 test_bit(Faulty
, &rdev2
->flags
))
1314 desc_nr
= rdev2
->raid_disk
;
1316 desc_nr
= next_spare
++;
1317 rdev2
->desc_nr
= desc_nr
;
1318 d
= &sb
->disks
[rdev2
->desc_nr
];
1320 d
->number
= rdev2
->desc_nr
;
1321 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1322 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1324 d
->raid_disk
= rdev2
->raid_disk
;
1326 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1327 if (test_bit(Faulty
, &rdev2
->flags
))
1328 d
->state
= (1<<MD_DISK_FAULTY
);
1329 else if (is_active
) {
1330 d
->state
= (1<<MD_DISK_ACTIVE
);
1331 if (test_bit(In_sync
, &rdev2
->flags
))
1332 d
->state
|= (1<<MD_DISK_SYNC
);
1340 if (test_bit(WriteMostly
, &rdev2
->flags
))
1341 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1342 if (test_bit(FailFast
, &rdev2
->flags
))
1343 d
->state
|= (1<<MD_DISK_FAILFAST
);
1345 /* now set the "removed" and "faulty" bits on any missing devices */
1346 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1347 mdp_disk_t
*d
= &sb
->disks
[i
];
1348 if (d
->state
== 0 && d
->number
== 0) {
1351 d
->state
= (1<<MD_DISK_REMOVED
);
1352 d
->state
|= (1<<MD_DISK_FAULTY
);
1356 sb
->nr_disks
= nr_disks
;
1357 sb
->active_disks
= active
;
1358 sb
->working_disks
= working
;
1359 sb
->failed_disks
= failed
;
1360 sb
->spare_disks
= spare
;
1362 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1363 sb
->sb_csum
= calc_sb_csum(sb
);
1367 * rdev_size_change for 0.90.0
1369 static unsigned long long
1370 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1372 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1373 return 0; /* component must fit device */
1374 if (rdev
->mddev
->bitmap_info
.offset
)
1375 return 0; /* can't move bitmap */
1376 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1377 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1378 num_sectors
= rdev
->sb_start
;
1379 /* Limit to 4TB as metadata cannot record more than that.
1380 * 4TB == 2^32 KB, or 2*2^32 sectors.
1382 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)num_sectors
>= (2ULL << 32) &&
1383 rdev
->mddev
->level
>= 1)
1384 num_sectors
= (sector_t
)(2ULL << 32) - 2;
1386 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1388 } while (md_super_wait(rdev
->mddev
) < 0);
1393 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1395 /* non-zero offset changes not possible with v0.90 */
1396 return new_offset
== 0;
1400 * version 1 superblock
1403 static __le32
calc_sb_1_csum(struct mdp_superblock_1
*sb
)
1407 unsigned long long newcsum
;
1408 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1409 __le32
*isuper
= (__le32
*)sb
;
1411 disk_csum
= sb
->sb_csum
;
1414 for (; size
>= 4; size
-= 4)
1415 newcsum
+= le32_to_cpu(*isuper
++);
1418 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1420 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1421 sb
->sb_csum
= disk_csum
;
1422 return cpu_to_le32(csum
);
1425 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1427 struct mdp_superblock_1
*sb
;
1431 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1435 * Calculate the position of the superblock in 512byte sectors.
1436 * It is always aligned to a 4K boundary and
1437 * depeding on minor_version, it can be:
1438 * 0: At least 8K, but less than 12K, from end of device
1439 * 1: At start of device
1440 * 2: 4K from start of device.
1442 switch(minor_version
) {
1444 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1446 sb_start
&= ~(sector_t
)(4*2-1);
1457 rdev
->sb_start
= sb_start
;
1459 /* superblock is rarely larger than 1K, but it can be larger,
1460 * and it is safe to read 4k, so we do that
1462 ret
= read_disk_sb(rdev
, 4096);
1463 if (ret
) return ret
;
1465 sb
= page_address(rdev
->sb_page
);
1467 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1468 sb
->major_version
!= cpu_to_le32(1) ||
1469 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1470 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1471 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1474 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1475 pr_warn("md: invalid superblock checksum on %s\n",
1476 bdevname(rdev
->bdev
,b
));
1479 if (le64_to_cpu(sb
->data_size
) < 10) {
1480 pr_warn("md: data_size too small on %s\n",
1481 bdevname(rdev
->bdev
,b
));
1486 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1487 /* Some padding is non-zero, might be a new feature */
1490 rdev
->preferred_minor
= 0xffff;
1491 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1492 rdev
->new_data_offset
= rdev
->data_offset
;
1493 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1494 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1495 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1496 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1498 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1499 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1500 if (rdev
->sb_size
& bmask
)
1501 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1504 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1507 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1510 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1513 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1515 if (!rdev
->bb_page
) {
1516 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1520 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1521 rdev
->badblocks
.count
== 0) {
1522 /* need to load the bad block list.
1523 * Currently we limit it to one page.
1529 int sectors
= le16_to_cpu(sb
->bblog_size
);
1530 if (sectors
> (PAGE_SIZE
/ 512))
1532 offset
= le32_to_cpu(sb
->bblog_offset
);
1535 bb_sector
= (long long)offset
;
1536 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1537 rdev
->bb_page
, REQ_OP_READ
, 0, true))
1539 bbp
= (u64
*)page_address(rdev
->bb_page
);
1540 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1541 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1542 u64 bb
= le64_to_cpu(*bbp
);
1543 int count
= bb
& (0x3ff);
1544 u64 sector
= bb
>> 10;
1545 sector
<<= sb
->bblog_shift
;
1546 count
<<= sb
->bblog_shift
;
1549 if (badblocks_set(&rdev
->badblocks
, sector
, count
, 1))
1552 } else if (sb
->bblog_offset
!= 0)
1553 rdev
->badblocks
.shift
= 0;
1555 if ((le32_to_cpu(sb
->feature_map
) &
1556 (MD_FEATURE_PPL
| MD_FEATURE_MULTIPLE_PPLS
))) {
1557 rdev
->ppl
.offset
= (__s16
)le16_to_cpu(sb
->ppl
.offset
);
1558 rdev
->ppl
.size
= le16_to_cpu(sb
->ppl
.size
);
1559 rdev
->ppl
.sector
= rdev
->sb_start
+ rdev
->ppl
.offset
;
1566 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1568 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1569 sb
->level
!= refsb
->level
||
1570 sb
->layout
!= refsb
->layout
||
1571 sb
->chunksize
!= refsb
->chunksize
) {
1572 pr_warn("md: %s has strangely different superblock to %s\n",
1573 bdevname(rdev
->bdev
,b
),
1574 bdevname(refdev
->bdev
,b2
));
1577 ev1
= le64_to_cpu(sb
->events
);
1578 ev2
= le64_to_cpu(refsb
->events
);
1585 if (minor_version
) {
1586 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1587 sectors
-= rdev
->data_offset
;
1589 sectors
= rdev
->sb_start
;
1590 if (sectors
< le64_to_cpu(sb
->data_size
))
1592 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1596 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1598 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1599 __u64 ev1
= le64_to_cpu(sb
->events
);
1601 rdev
->raid_disk
= -1;
1602 clear_bit(Faulty
, &rdev
->flags
);
1603 clear_bit(In_sync
, &rdev
->flags
);
1604 clear_bit(Bitmap_sync
, &rdev
->flags
);
1605 clear_bit(WriteMostly
, &rdev
->flags
);
1607 if (mddev
->raid_disks
== 0) {
1608 mddev
->major_version
= 1;
1609 mddev
->patch_version
= 0;
1610 mddev
->external
= 0;
1611 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1612 mddev
->ctime
= le64_to_cpu(sb
->ctime
);
1613 mddev
->utime
= le64_to_cpu(sb
->utime
);
1614 mddev
->level
= le32_to_cpu(sb
->level
);
1615 mddev
->clevel
[0] = 0;
1616 mddev
->layout
= le32_to_cpu(sb
->layout
);
1617 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1618 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1619 mddev
->events
= ev1
;
1620 mddev
->bitmap_info
.offset
= 0;
1621 mddev
->bitmap_info
.space
= 0;
1622 /* Default location for bitmap is 1K after superblock
1623 * using 3K - total of 4K
1625 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1626 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1627 mddev
->reshape_backwards
= 0;
1629 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1630 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1632 mddev
->max_disks
= (4096-256)/2;
1634 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1635 mddev
->bitmap_info
.file
== NULL
) {
1636 mddev
->bitmap_info
.offset
=
1637 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1638 /* Metadata doesn't record how much space is available.
1639 * For 1.0, we assume we can use up to the superblock
1640 * if before, else to 4K beyond superblock.
1641 * For others, assume no change is possible.
1643 if (mddev
->minor_version
> 0)
1644 mddev
->bitmap_info
.space
= 0;
1645 else if (mddev
->bitmap_info
.offset
> 0)
1646 mddev
->bitmap_info
.space
=
1647 8 - mddev
->bitmap_info
.offset
;
1649 mddev
->bitmap_info
.space
=
1650 -mddev
->bitmap_info
.offset
;
1653 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1654 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1655 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1656 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1657 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1658 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1659 if (mddev
->delta_disks
< 0 ||
1660 (mddev
->delta_disks
== 0 &&
1661 (le32_to_cpu(sb
->feature_map
)
1662 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1663 mddev
->reshape_backwards
= 1;
1665 mddev
->reshape_position
= MaxSector
;
1666 mddev
->delta_disks
= 0;
1667 mddev
->new_level
= mddev
->level
;
1668 mddev
->new_layout
= mddev
->layout
;
1669 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1672 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
)
1673 set_bit(MD_HAS_JOURNAL
, &mddev
->flags
);
1675 if (le32_to_cpu(sb
->feature_map
) &
1676 (MD_FEATURE_PPL
| MD_FEATURE_MULTIPLE_PPLS
)) {
1677 if (le32_to_cpu(sb
->feature_map
) &
1678 (MD_FEATURE_BITMAP_OFFSET
| MD_FEATURE_JOURNAL
))
1680 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_PPL
) &&
1681 (le32_to_cpu(sb
->feature_map
) &
1682 MD_FEATURE_MULTIPLE_PPLS
))
1684 set_bit(MD_HAS_PPL
, &mddev
->flags
);
1686 } else if (mddev
->pers
== NULL
) {
1687 /* Insist of good event counter while assembling, except for
1688 * spares (which don't need an event count) */
1690 if (rdev
->desc_nr
>= 0 &&
1691 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1692 (le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < MD_DISK_ROLE_MAX
||
1693 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) == MD_DISK_ROLE_JOURNAL
))
1694 if (ev1
< mddev
->events
)
1696 } else if (mddev
->bitmap
) {
1697 /* If adding to array with a bitmap, then we can accept an
1698 * older device, but not too old.
1700 if (ev1
< mddev
->bitmap
->events_cleared
)
1702 if (ev1
< mddev
->events
)
1703 set_bit(Bitmap_sync
, &rdev
->flags
);
1705 if (ev1
< mddev
->events
)
1706 /* just a hot-add of a new device, leave raid_disk at -1 */
1709 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1711 if (rdev
->desc_nr
< 0 ||
1712 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1713 role
= MD_DISK_ROLE_SPARE
;
1716 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1718 case MD_DISK_ROLE_SPARE
: /* spare */
1720 case MD_DISK_ROLE_FAULTY
: /* faulty */
1721 set_bit(Faulty
, &rdev
->flags
);
1723 case MD_DISK_ROLE_JOURNAL
: /* journal device */
1724 if (!(le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
)) {
1725 /* journal device without journal feature */
1726 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1729 set_bit(Journal
, &rdev
->flags
);
1730 rdev
->journal_tail
= le64_to_cpu(sb
->journal_tail
);
1731 rdev
->raid_disk
= 0;
1734 rdev
->saved_raid_disk
= role
;
1735 if ((le32_to_cpu(sb
->feature_map
) &
1736 MD_FEATURE_RECOVERY_OFFSET
)) {
1737 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1738 if (!(le32_to_cpu(sb
->feature_map
) &
1739 MD_FEATURE_RECOVERY_BITMAP
))
1740 rdev
->saved_raid_disk
= -1;
1742 set_bit(In_sync
, &rdev
->flags
);
1743 rdev
->raid_disk
= role
;
1746 if (sb
->devflags
& WriteMostly1
)
1747 set_bit(WriteMostly
, &rdev
->flags
);
1748 if (sb
->devflags
& FailFast1
)
1749 set_bit(FailFast
, &rdev
->flags
);
1750 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1751 set_bit(Replacement
, &rdev
->flags
);
1752 } else /* MULTIPATH are always insync */
1753 set_bit(In_sync
, &rdev
->flags
);
1758 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1760 struct mdp_superblock_1
*sb
;
1761 struct md_rdev
*rdev2
;
1763 /* make rdev->sb match mddev and rdev data. */
1765 sb
= page_address(rdev
->sb_page
);
1767 sb
->feature_map
= 0;
1769 sb
->recovery_offset
= cpu_to_le64(0);
1770 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1772 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1773 sb
->events
= cpu_to_le64(mddev
->events
);
1775 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1776 else if (test_bit(MD_JOURNAL_CLEAN
, &mddev
->flags
))
1777 sb
->resync_offset
= cpu_to_le64(MaxSector
);
1779 sb
->resync_offset
= cpu_to_le64(0);
1781 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1783 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1784 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1785 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1786 sb
->level
= cpu_to_le32(mddev
->level
);
1787 sb
->layout
= cpu_to_le32(mddev
->layout
);
1788 if (test_bit(FailFast
, &rdev
->flags
))
1789 sb
->devflags
|= FailFast1
;
1791 sb
->devflags
&= ~FailFast1
;
1793 if (test_bit(WriteMostly
, &rdev
->flags
))
1794 sb
->devflags
|= WriteMostly1
;
1796 sb
->devflags
&= ~WriteMostly1
;
1797 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1798 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1800 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1801 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1802 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1805 if (rdev
->raid_disk
>= 0 && !test_bit(Journal
, &rdev
->flags
) &&
1806 !test_bit(In_sync
, &rdev
->flags
)) {
1808 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1809 sb
->recovery_offset
=
1810 cpu_to_le64(rdev
->recovery_offset
);
1811 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1813 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1815 /* Note: recovery_offset and journal_tail share space */
1816 if (test_bit(Journal
, &rdev
->flags
))
1817 sb
->journal_tail
= cpu_to_le64(rdev
->journal_tail
);
1818 if (test_bit(Replacement
, &rdev
->flags
))
1820 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1822 if (mddev
->reshape_position
!= MaxSector
) {
1823 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1824 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1825 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1826 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1827 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1828 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1829 if (mddev
->delta_disks
== 0 &&
1830 mddev
->reshape_backwards
)
1832 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1833 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1835 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1836 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1837 - rdev
->data_offset
));
1841 if (mddev_is_clustered(mddev
))
1842 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_CLUSTERED
);
1844 if (rdev
->badblocks
.count
== 0)
1845 /* Nothing to do for bad blocks*/ ;
1846 else if (sb
->bblog_offset
== 0)
1847 /* Cannot record bad blocks on this device */
1848 md_error(mddev
, rdev
);
1850 struct badblocks
*bb
= &rdev
->badblocks
;
1851 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1853 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1858 seq
= read_seqbegin(&bb
->lock
);
1860 memset(bbp
, 0xff, PAGE_SIZE
);
1862 for (i
= 0 ; i
< bb
->count
; i
++) {
1863 u64 internal_bb
= p
[i
];
1864 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1865 | BB_LEN(internal_bb
));
1866 bbp
[i
] = cpu_to_le64(store_bb
);
1869 if (read_seqretry(&bb
->lock
, seq
))
1872 bb
->sector
= (rdev
->sb_start
+
1873 (int)le32_to_cpu(sb
->bblog_offset
));
1874 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1879 rdev_for_each(rdev2
, mddev
)
1880 if (rdev2
->desc_nr
+1 > max_dev
)
1881 max_dev
= rdev2
->desc_nr
+1;
1883 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1885 sb
->max_dev
= cpu_to_le32(max_dev
);
1886 rdev
->sb_size
= max_dev
* 2 + 256;
1887 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1888 if (rdev
->sb_size
& bmask
)
1889 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1891 max_dev
= le32_to_cpu(sb
->max_dev
);
1893 for (i
=0; i
<max_dev
;i
++)
1894 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_SPARE
);
1896 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
))
1897 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_JOURNAL
);
1899 if (test_bit(MD_HAS_PPL
, &mddev
->flags
)) {
1900 if (test_bit(MD_HAS_MULTIPLE_PPLS
, &mddev
->flags
))
1902 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS
);
1904 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_PPL
);
1905 sb
->ppl
.offset
= cpu_to_le16(rdev
->ppl
.offset
);
1906 sb
->ppl
.size
= cpu_to_le16(rdev
->ppl
.size
);
1909 rdev_for_each(rdev2
, mddev
) {
1911 if (test_bit(Faulty
, &rdev2
->flags
))
1912 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_FAULTY
);
1913 else if (test_bit(In_sync
, &rdev2
->flags
))
1914 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1915 else if (test_bit(Journal
, &rdev2
->flags
))
1916 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_JOURNAL
);
1917 else if (rdev2
->raid_disk
>= 0)
1918 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1920 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_SPARE
);
1923 sb
->sb_csum
= calc_sb_1_csum(sb
);
1926 static unsigned long long
1927 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1929 struct mdp_superblock_1
*sb
;
1930 sector_t max_sectors
;
1931 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1932 return 0; /* component must fit device */
1933 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1934 return 0; /* too confusing */
1935 if (rdev
->sb_start
< rdev
->data_offset
) {
1936 /* minor versions 1 and 2; superblock before data */
1937 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1938 max_sectors
-= rdev
->data_offset
;
1939 if (!num_sectors
|| num_sectors
> max_sectors
)
1940 num_sectors
= max_sectors
;
1941 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1942 /* minor version 0 with bitmap we can't move */
1945 /* minor version 0; superblock after data */
1947 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1948 sb_start
&= ~(sector_t
)(4*2 - 1);
1949 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1950 if (!num_sectors
|| num_sectors
> max_sectors
)
1951 num_sectors
= max_sectors
;
1952 rdev
->sb_start
= sb_start
;
1954 sb
= page_address(rdev
->sb_page
);
1955 sb
->data_size
= cpu_to_le64(num_sectors
);
1956 sb
->super_offset
= cpu_to_le64(rdev
->sb_start
);
1957 sb
->sb_csum
= calc_sb_1_csum(sb
);
1959 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1961 } while (md_super_wait(rdev
->mddev
) < 0);
1967 super_1_allow_new_offset(struct md_rdev
*rdev
,
1968 unsigned long long new_offset
)
1970 /* All necessary checks on new >= old have been done */
1971 struct bitmap
*bitmap
;
1972 if (new_offset
>= rdev
->data_offset
)
1975 /* with 1.0 metadata, there is no metadata to tread on
1976 * so we can always move back */
1977 if (rdev
->mddev
->minor_version
== 0)
1980 /* otherwise we must be sure not to step on
1981 * any metadata, so stay:
1982 * 36K beyond start of superblock
1983 * beyond end of badblocks
1984 * beyond write-intent bitmap
1986 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1988 bitmap
= rdev
->mddev
->bitmap
;
1989 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1990 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1991 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1993 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1999 static struct super_type super_types
[] = {
2002 .owner
= THIS_MODULE
,
2003 .load_super
= super_90_load
,
2004 .validate_super
= super_90_validate
,
2005 .sync_super
= super_90_sync
,
2006 .rdev_size_change
= super_90_rdev_size_change
,
2007 .allow_new_offset
= super_90_allow_new_offset
,
2011 .owner
= THIS_MODULE
,
2012 .load_super
= super_1_load
,
2013 .validate_super
= super_1_validate
,
2014 .sync_super
= super_1_sync
,
2015 .rdev_size_change
= super_1_rdev_size_change
,
2016 .allow_new_offset
= super_1_allow_new_offset
,
2020 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
2022 if (mddev
->sync_super
) {
2023 mddev
->sync_super(mddev
, rdev
);
2027 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
2029 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
2032 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
2034 struct md_rdev
*rdev
, *rdev2
;
2037 rdev_for_each_rcu(rdev
, mddev1
) {
2038 if (test_bit(Faulty
, &rdev
->flags
) ||
2039 test_bit(Journal
, &rdev
->flags
) ||
2040 rdev
->raid_disk
== -1)
2042 rdev_for_each_rcu(rdev2
, mddev2
) {
2043 if (test_bit(Faulty
, &rdev2
->flags
) ||
2044 test_bit(Journal
, &rdev2
->flags
) ||
2045 rdev2
->raid_disk
== -1)
2047 if (rdev
->bdev
->bd_contains
==
2048 rdev2
->bdev
->bd_contains
) {
2058 static LIST_HEAD(pending_raid_disks
);
2061 * Try to register data integrity profile for an mddev
2063 * This is called when an array is started and after a disk has been kicked
2064 * from the array. It only succeeds if all working and active component devices
2065 * are integrity capable with matching profiles.
2067 int md_integrity_register(struct mddev
*mddev
)
2069 struct md_rdev
*rdev
, *reference
= NULL
;
2071 if (list_empty(&mddev
->disks
))
2072 return 0; /* nothing to do */
2073 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
2074 return 0; /* shouldn't register, or already is */
2075 rdev_for_each(rdev
, mddev
) {
2076 /* skip spares and non-functional disks */
2077 if (test_bit(Faulty
, &rdev
->flags
))
2079 if (rdev
->raid_disk
< 0)
2082 /* Use the first rdev as the reference */
2086 /* does this rdev's profile match the reference profile? */
2087 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
2088 rdev
->bdev
->bd_disk
) < 0)
2091 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2094 * All component devices are integrity capable and have matching
2095 * profiles, register the common profile for the md device.
2097 blk_integrity_register(mddev
->gendisk
,
2098 bdev_get_integrity(reference
->bdev
));
2100 pr_debug("md: data integrity enabled on %s\n", mdname(mddev
));
2101 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2102 pr_err("md: failed to create integrity pool for %s\n",
2108 EXPORT_SYMBOL(md_integrity_register
);
2111 * Attempt to add an rdev, but only if it is consistent with the current
2114 int md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2116 struct blk_integrity
*bi_rdev
;
2117 struct blk_integrity
*bi_mddev
;
2118 char name
[BDEVNAME_SIZE
];
2120 if (!mddev
->gendisk
)
2123 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2124 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2126 if (!bi_mddev
) /* nothing to do */
2129 if (blk_integrity_compare(mddev
->gendisk
, rdev
->bdev
->bd_disk
) != 0) {
2130 pr_err("%s: incompatible integrity profile for %s\n",
2131 mdname(mddev
), bdevname(rdev
->bdev
, name
));
2137 EXPORT_SYMBOL(md_integrity_add_rdev
);
2139 static int bind_rdev_to_array(struct md_rdev
*rdev
, struct mddev
*mddev
)
2141 char b
[BDEVNAME_SIZE
];
2145 /* prevent duplicates */
2146 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2149 if ((bdev_read_only(rdev
->bdev
) || bdev_read_only(rdev
->meta_bdev
)) &&
2153 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2154 if (!test_bit(Journal
, &rdev
->flags
) &&
2156 (mddev
->dev_sectors
== 0 || rdev
->sectors
< mddev
->dev_sectors
)) {
2158 /* Cannot change size, so fail
2159 * If mddev->level <= 0, then we don't care
2160 * about aligning sizes (e.g. linear)
2162 if (mddev
->level
> 0)
2165 mddev
->dev_sectors
= rdev
->sectors
;
2168 /* Verify rdev->desc_nr is unique.
2169 * If it is -1, assign a free number, else
2170 * check number is not in use
2173 if (rdev
->desc_nr
< 0) {
2176 choice
= mddev
->raid_disks
;
2177 while (md_find_rdev_nr_rcu(mddev
, choice
))
2179 rdev
->desc_nr
= choice
;
2181 if (md_find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2187 if (!test_bit(Journal
, &rdev
->flags
) &&
2188 mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2189 pr_warn("md: %s: array is limited to %d devices\n",
2190 mdname(mddev
), mddev
->max_disks
);
2193 bdevname(rdev
->bdev
,b
);
2194 strreplace(b
, '/', '!');
2196 rdev
->mddev
= mddev
;
2197 pr_debug("md: bind<%s>\n", b
);
2199 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2202 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2203 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2204 /* failure here is OK */;
2205 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2207 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2208 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2210 /* May as well allow recovery to be retried once */
2211 mddev
->recovery_disabled
++;
2216 pr_warn("md: failed to register dev-%s for %s\n",
2221 static void md_delayed_delete(struct work_struct
*ws
)
2223 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2224 kobject_del(&rdev
->kobj
);
2225 kobject_put(&rdev
->kobj
);
2228 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2230 char b
[BDEVNAME_SIZE
];
2232 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2233 list_del_rcu(&rdev
->same_set
);
2234 pr_debug("md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2236 sysfs_remove_link(&rdev
->kobj
, "block");
2237 sysfs_put(rdev
->sysfs_state
);
2238 rdev
->sysfs_state
= NULL
;
2239 rdev
->badblocks
.count
= 0;
2240 /* We need to delay this, otherwise we can deadlock when
2241 * writing to 'remove' to "dev/state". We also need
2242 * to delay it due to rcu usage.
2245 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2246 kobject_get(&rdev
->kobj
);
2247 queue_work(md_misc_wq
, &rdev
->del_work
);
2251 * prevent the device from being mounted, repartitioned or
2252 * otherwise reused by a RAID array (or any other kernel
2253 * subsystem), by bd_claiming the device.
2255 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2258 struct block_device
*bdev
;
2259 char b
[BDEVNAME_SIZE
];
2261 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2262 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2264 pr_warn("md: could not open %s.\n", __bdevname(dev
, b
));
2265 return PTR_ERR(bdev
);
2271 static void unlock_rdev(struct md_rdev
*rdev
)
2273 struct block_device
*bdev
= rdev
->bdev
;
2275 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2278 void md_autodetect_dev(dev_t dev
);
2280 static void export_rdev(struct md_rdev
*rdev
)
2282 char b
[BDEVNAME_SIZE
];
2284 pr_debug("md: export_rdev(%s)\n", bdevname(rdev
->bdev
,b
));
2285 md_rdev_clear(rdev
);
2287 if (test_bit(AutoDetected
, &rdev
->flags
))
2288 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2291 kobject_put(&rdev
->kobj
);
2294 void md_kick_rdev_from_array(struct md_rdev
*rdev
)
2296 unbind_rdev_from_array(rdev
);
2299 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array
);
2301 static void export_array(struct mddev
*mddev
)
2303 struct md_rdev
*rdev
;
2305 while (!list_empty(&mddev
->disks
)) {
2306 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2308 md_kick_rdev_from_array(rdev
);
2310 mddev
->raid_disks
= 0;
2311 mddev
->major_version
= 0;
2314 static bool set_in_sync(struct mddev
*mddev
)
2316 WARN_ON_ONCE(NR_CPUS
!= 1 && !spin_is_locked(&mddev
->lock
));
2317 if (!mddev
->in_sync
) {
2318 mddev
->sync_checkers
++;
2319 spin_unlock(&mddev
->lock
);
2320 percpu_ref_switch_to_atomic_sync(&mddev
->writes_pending
);
2321 spin_lock(&mddev
->lock
);
2322 if (!mddev
->in_sync
&&
2323 percpu_ref_is_zero(&mddev
->writes_pending
)) {
2326 * Ensure ->in_sync is visible before we clear
2330 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
2331 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
2333 if (--mddev
->sync_checkers
== 0)
2334 percpu_ref_switch_to_percpu(&mddev
->writes_pending
);
2336 if (mddev
->safemode
== 1)
2337 mddev
->safemode
= 0;
2338 return mddev
->in_sync
;
2341 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2343 /* Update each superblock (in-memory image), but
2344 * if we are allowed to, skip spares which already
2345 * have the right event counter, or have one earlier
2346 * (which would mean they aren't being marked as dirty
2347 * with the rest of the array)
2349 struct md_rdev
*rdev
;
2350 rdev_for_each(rdev
, mddev
) {
2351 if (rdev
->sb_events
== mddev
->events
||
2353 rdev
->raid_disk
< 0 &&
2354 rdev
->sb_events
+1 == mddev
->events
)) {
2355 /* Don't update this superblock */
2356 rdev
->sb_loaded
= 2;
2358 sync_super(mddev
, rdev
);
2359 rdev
->sb_loaded
= 1;
2364 static bool does_sb_need_changing(struct mddev
*mddev
)
2366 struct md_rdev
*rdev
;
2367 struct mdp_superblock_1
*sb
;
2370 /* Find a good rdev */
2371 rdev_for_each(rdev
, mddev
)
2372 if ((rdev
->raid_disk
>= 0) && !test_bit(Faulty
, &rdev
->flags
))
2375 /* No good device found. */
2379 sb
= page_address(rdev
->sb_page
);
2380 /* Check if a device has become faulty or a spare become active */
2381 rdev_for_each(rdev
, mddev
) {
2382 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
2383 /* Device activated? */
2384 if (role
== 0xffff && rdev
->raid_disk
>=0 &&
2385 !test_bit(Faulty
, &rdev
->flags
))
2387 /* Device turned faulty? */
2388 if (test_bit(Faulty
, &rdev
->flags
) && (role
< 0xfffd))
2392 /* Check if any mddev parameters have changed */
2393 if ((mddev
->dev_sectors
!= le64_to_cpu(sb
->size
)) ||
2394 (mddev
->reshape_position
!= le64_to_cpu(sb
->reshape_position
)) ||
2395 (mddev
->layout
!= le32_to_cpu(sb
->layout
)) ||
2396 (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
)) ||
2397 (mddev
->chunk_sectors
!= le32_to_cpu(sb
->chunksize
)))
2403 void md_update_sb(struct mddev
*mddev
, int force_change
)
2405 struct md_rdev
*rdev
;
2408 int any_badblocks_changed
= 0;
2413 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2418 if (mddev_is_clustered(mddev
)) {
2419 if (test_and_clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
))
2421 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
))
2423 ret
= md_cluster_ops
->metadata_update_start(mddev
);
2424 /* Has someone else has updated the sb */
2425 if (!does_sb_need_changing(mddev
)) {
2427 md_cluster_ops
->metadata_update_cancel(mddev
);
2428 bit_clear_unless(&mddev
->sb_flags
, BIT(MD_SB_CHANGE_PENDING
),
2429 BIT(MD_SB_CHANGE_DEVS
) |
2430 BIT(MD_SB_CHANGE_CLEAN
));
2435 /* First make sure individual recovery_offsets are correct */
2436 rdev_for_each(rdev
, mddev
) {
2437 if (rdev
->raid_disk
>= 0 &&
2438 mddev
->delta_disks
>= 0 &&
2439 !test_bit(Journal
, &rdev
->flags
) &&
2440 !test_bit(In_sync
, &rdev
->flags
) &&
2441 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2442 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2445 if (!mddev
->persistent
) {
2446 clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
2447 clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2448 if (!mddev
->external
) {
2449 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
2450 rdev_for_each(rdev
, mddev
) {
2451 if (rdev
->badblocks
.changed
) {
2452 rdev
->badblocks
.changed
= 0;
2453 ack_all_badblocks(&rdev
->badblocks
);
2454 md_error(mddev
, rdev
);
2456 clear_bit(Blocked
, &rdev
->flags
);
2457 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2458 wake_up(&rdev
->blocked_wait
);
2461 wake_up(&mddev
->sb_wait
);
2465 spin_lock(&mddev
->lock
);
2467 mddev
->utime
= ktime_get_real_seconds();
2469 if (test_and_clear_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
))
2471 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
))
2472 /* just a clean<-> dirty transition, possibly leave spares alone,
2473 * though if events isn't the right even/odd, we will have to do
2479 if (mddev
->degraded
)
2480 /* If the array is degraded, then skipping spares is both
2481 * dangerous and fairly pointless.
2482 * Dangerous because a device that was removed from the array
2483 * might have a event_count that still looks up-to-date,
2484 * so it can be re-added without a resync.
2485 * Pointless because if there are any spares to skip,
2486 * then a recovery will happen and soon that array won't
2487 * be degraded any more and the spare can go back to sleep then.
2491 sync_req
= mddev
->in_sync
;
2493 /* If this is just a dirty<->clean transition, and the array is clean
2494 * and 'events' is odd, we can roll back to the previous clean state */
2496 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2497 && mddev
->can_decrease_events
2498 && mddev
->events
!= 1) {
2500 mddev
->can_decrease_events
= 0;
2502 /* otherwise we have to go forward and ... */
2504 mddev
->can_decrease_events
= nospares
;
2508 * This 64-bit counter should never wrap.
2509 * Either we are in around ~1 trillion A.C., assuming
2510 * 1 reboot per second, or we have a bug...
2512 WARN_ON(mddev
->events
== 0);
2514 rdev_for_each(rdev
, mddev
) {
2515 if (rdev
->badblocks
.changed
)
2516 any_badblocks_changed
++;
2517 if (test_bit(Faulty
, &rdev
->flags
))
2518 set_bit(FaultRecorded
, &rdev
->flags
);
2521 sync_sbs(mddev
, nospares
);
2522 spin_unlock(&mddev
->lock
);
2524 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2525 mdname(mddev
), mddev
->in_sync
);
2528 blk_add_trace_msg(mddev
->queue
, "md md_update_sb");
2530 bitmap_update_sb(mddev
->bitmap
);
2531 rdev_for_each(rdev
, mddev
) {
2532 char b
[BDEVNAME_SIZE
];
2534 if (rdev
->sb_loaded
!= 1)
2535 continue; /* no noise on spare devices */
2537 if (!test_bit(Faulty
, &rdev
->flags
)) {
2538 md_super_write(mddev
,rdev
,
2539 rdev
->sb_start
, rdev
->sb_size
,
2541 pr_debug("md: (write) %s's sb offset: %llu\n",
2542 bdevname(rdev
->bdev
, b
),
2543 (unsigned long long)rdev
->sb_start
);
2544 rdev
->sb_events
= mddev
->events
;
2545 if (rdev
->badblocks
.size
) {
2546 md_super_write(mddev
, rdev
,
2547 rdev
->badblocks
.sector
,
2548 rdev
->badblocks
.size
<< 9,
2550 rdev
->badblocks
.size
= 0;
2554 pr_debug("md: %s (skipping faulty)\n",
2555 bdevname(rdev
->bdev
, b
));
2557 if (mddev
->level
== LEVEL_MULTIPATH
)
2558 /* only need to write one superblock... */
2561 if (md_super_wait(mddev
) < 0)
2563 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2565 if (mddev_is_clustered(mddev
) && ret
== 0)
2566 md_cluster_ops
->metadata_update_finish(mddev
);
2568 if (mddev
->in_sync
!= sync_req
||
2569 !bit_clear_unless(&mddev
->sb_flags
, BIT(MD_SB_CHANGE_PENDING
),
2570 BIT(MD_SB_CHANGE_DEVS
) | BIT(MD_SB_CHANGE_CLEAN
)))
2571 /* have to write it out again */
2573 wake_up(&mddev
->sb_wait
);
2574 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2575 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2577 rdev_for_each(rdev
, mddev
) {
2578 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2579 clear_bit(Blocked
, &rdev
->flags
);
2581 if (any_badblocks_changed
)
2582 ack_all_badblocks(&rdev
->badblocks
);
2583 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2584 wake_up(&rdev
->blocked_wait
);
2587 EXPORT_SYMBOL(md_update_sb
);
2589 static int add_bound_rdev(struct md_rdev
*rdev
)
2591 struct mddev
*mddev
= rdev
->mddev
;
2593 bool add_journal
= test_bit(Journal
, &rdev
->flags
);
2595 if (!mddev
->pers
->hot_remove_disk
|| add_journal
) {
2596 /* If there is hot_add_disk but no hot_remove_disk
2597 * then added disks for geometry changes,
2598 * and should be added immediately.
2600 super_types
[mddev
->major_version
].
2601 validate_super(mddev
, rdev
);
2603 mddev_suspend(mddev
);
2604 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
2606 mddev_resume(mddev
);
2608 md_kick_rdev_from_array(rdev
);
2612 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2614 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2615 if (mddev
->degraded
)
2616 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2617 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2618 md_new_event(mddev
);
2619 md_wakeup_thread(mddev
->thread
);
2623 /* words written to sysfs files may, or may not, be \n terminated.
2624 * We want to accept with case. For this we use cmd_match.
2626 static int cmd_match(const char *cmd
, const char *str
)
2628 /* See if cmd, written into a sysfs file, matches
2629 * str. They must either be the same, or cmd can
2630 * have a trailing newline
2632 while (*cmd
&& *str
&& *cmd
== *str
) {
2643 struct rdev_sysfs_entry
{
2644 struct attribute attr
;
2645 ssize_t (*show
)(struct md_rdev
*, char *);
2646 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2650 state_show(struct md_rdev
*rdev
, char *page
)
2654 unsigned long flags
= ACCESS_ONCE(rdev
->flags
);
2656 if (test_bit(Faulty
, &flags
) ||
2657 (!test_bit(ExternalBbl
, &flags
) &&
2658 rdev
->badblocks
.unacked_exist
))
2659 len
+= sprintf(page
+len
, "faulty%s", sep
);
2660 if (test_bit(In_sync
, &flags
))
2661 len
+= sprintf(page
+len
, "in_sync%s", sep
);
2662 if (test_bit(Journal
, &flags
))
2663 len
+= sprintf(page
+len
, "journal%s", sep
);
2664 if (test_bit(WriteMostly
, &flags
))
2665 len
+= sprintf(page
+len
, "write_mostly%s", sep
);
2666 if (test_bit(Blocked
, &flags
) ||
2667 (rdev
->badblocks
.unacked_exist
2668 && !test_bit(Faulty
, &flags
)))
2669 len
+= sprintf(page
+len
, "blocked%s", sep
);
2670 if (!test_bit(Faulty
, &flags
) &&
2671 !test_bit(Journal
, &flags
) &&
2672 !test_bit(In_sync
, &flags
))
2673 len
+= sprintf(page
+len
, "spare%s", sep
);
2674 if (test_bit(WriteErrorSeen
, &flags
))
2675 len
+= sprintf(page
+len
, "write_error%s", sep
);
2676 if (test_bit(WantReplacement
, &flags
))
2677 len
+= sprintf(page
+len
, "want_replacement%s", sep
);
2678 if (test_bit(Replacement
, &flags
))
2679 len
+= sprintf(page
+len
, "replacement%s", sep
);
2680 if (test_bit(ExternalBbl
, &flags
))
2681 len
+= sprintf(page
+len
, "external_bbl%s", sep
);
2682 if (test_bit(FailFast
, &flags
))
2683 len
+= sprintf(page
+len
, "failfast%s", sep
);
2688 return len
+sprintf(page
+len
, "\n");
2692 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2695 * faulty - simulates an error
2696 * remove - disconnects the device
2697 * writemostly - sets write_mostly
2698 * -writemostly - clears write_mostly
2699 * blocked - sets the Blocked flags
2700 * -blocked - clears the Blocked and possibly simulates an error
2701 * insync - sets Insync providing device isn't active
2702 * -insync - clear Insync for a device with a slot assigned,
2703 * so that it gets rebuilt based on bitmap
2704 * write_error - sets WriteErrorSeen
2705 * -write_error - clears WriteErrorSeen
2706 * {,-}failfast - set/clear FailFast
2709 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2710 md_error(rdev
->mddev
, rdev
);
2711 if (test_bit(Faulty
, &rdev
->flags
))
2715 } else if (cmd_match(buf
, "remove")) {
2716 if (rdev
->mddev
->pers
) {
2717 clear_bit(Blocked
, &rdev
->flags
);
2718 remove_and_add_spares(rdev
->mddev
, rdev
);
2720 if (rdev
->raid_disk
>= 0)
2723 struct mddev
*mddev
= rdev
->mddev
;
2725 if (mddev_is_clustered(mddev
))
2726 err
= md_cluster_ops
->remove_disk(mddev
, rdev
);
2729 md_kick_rdev_from_array(rdev
);
2731 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
2732 md_wakeup_thread(mddev
->thread
);
2734 md_new_event(mddev
);
2737 } else if (cmd_match(buf
, "writemostly")) {
2738 set_bit(WriteMostly
, &rdev
->flags
);
2740 } else if (cmd_match(buf
, "-writemostly")) {
2741 clear_bit(WriteMostly
, &rdev
->flags
);
2743 } else if (cmd_match(buf
, "blocked")) {
2744 set_bit(Blocked
, &rdev
->flags
);
2746 } else if (cmd_match(buf
, "-blocked")) {
2747 if (!test_bit(Faulty
, &rdev
->flags
) &&
2748 !test_bit(ExternalBbl
, &rdev
->flags
) &&
2749 rdev
->badblocks
.unacked_exist
) {
2750 /* metadata handler doesn't understand badblocks,
2751 * so we need to fail the device
2753 md_error(rdev
->mddev
, rdev
);
2755 clear_bit(Blocked
, &rdev
->flags
);
2756 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2757 wake_up(&rdev
->blocked_wait
);
2758 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2759 md_wakeup_thread(rdev
->mddev
->thread
);
2762 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2763 set_bit(In_sync
, &rdev
->flags
);
2765 } else if (cmd_match(buf
, "failfast")) {
2766 set_bit(FailFast
, &rdev
->flags
);
2768 } else if (cmd_match(buf
, "-failfast")) {
2769 clear_bit(FailFast
, &rdev
->flags
);
2771 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0 &&
2772 !test_bit(Journal
, &rdev
->flags
)) {
2773 if (rdev
->mddev
->pers
== NULL
) {
2774 clear_bit(In_sync
, &rdev
->flags
);
2775 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2776 rdev
->raid_disk
= -1;
2779 } else if (cmd_match(buf
, "write_error")) {
2780 set_bit(WriteErrorSeen
, &rdev
->flags
);
2782 } else if (cmd_match(buf
, "-write_error")) {
2783 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2785 } else if (cmd_match(buf
, "want_replacement")) {
2786 /* Any non-spare device that is not a replacement can
2787 * become want_replacement at any time, but we then need to
2788 * check if recovery is needed.
2790 if (rdev
->raid_disk
>= 0 &&
2791 !test_bit(Journal
, &rdev
->flags
) &&
2792 !test_bit(Replacement
, &rdev
->flags
))
2793 set_bit(WantReplacement
, &rdev
->flags
);
2794 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2795 md_wakeup_thread(rdev
->mddev
->thread
);
2797 } else if (cmd_match(buf
, "-want_replacement")) {
2798 /* Clearing 'want_replacement' is always allowed.
2799 * Once replacements starts it is too late though.
2802 clear_bit(WantReplacement
, &rdev
->flags
);
2803 } else if (cmd_match(buf
, "replacement")) {
2804 /* Can only set a device as a replacement when array has not
2805 * yet been started. Once running, replacement is automatic
2806 * from spares, or by assigning 'slot'.
2808 if (rdev
->mddev
->pers
)
2811 set_bit(Replacement
, &rdev
->flags
);
2814 } else if (cmd_match(buf
, "-replacement")) {
2815 /* Similarly, can only clear Replacement before start */
2816 if (rdev
->mddev
->pers
)
2819 clear_bit(Replacement
, &rdev
->flags
);
2822 } else if (cmd_match(buf
, "re-add")) {
2823 if (test_bit(Faulty
, &rdev
->flags
) && (rdev
->raid_disk
== -1)) {
2824 /* clear_bit is performed _after_ all the devices
2825 * have their local Faulty bit cleared. If any writes
2826 * happen in the meantime in the local node, they
2827 * will land in the local bitmap, which will be synced
2828 * by this node eventually
2830 if (!mddev_is_clustered(rdev
->mddev
) ||
2831 (err
= md_cluster_ops
->gather_bitmaps(rdev
)) == 0) {
2832 clear_bit(Faulty
, &rdev
->flags
);
2833 err
= add_bound_rdev(rdev
);
2837 } else if (cmd_match(buf
, "external_bbl") && (rdev
->mddev
->external
)) {
2838 set_bit(ExternalBbl
, &rdev
->flags
);
2839 rdev
->badblocks
.shift
= 0;
2841 } else if (cmd_match(buf
, "-external_bbl") && (rdev
->mddev
->external
)) {
2842 clear_bit(ExternalBbl
, &rdev
->flags
);
2846 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2847 return err
? err
: len
;
2849 static struct rdev_sysfs_entry rdev_state
=
2850 __ATTR_PREALLOC(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2853 errors_show(struct md_rdev
*rdev
, char *page
)
2855 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2859 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2864 rv
= kstrtouint(buf
, 10, &n
);
2867 atomic_set(&rdev
->corrected_errors
, n
);
2870 static struct rdev_sysfs_entry rdev_errors
=
2871 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2874 slot_show(struct md_rdev
*rdev
, char *page
)
2876 if (test_bit(Journal
, &rdev
->flags
))
2877 return sprintf(page
, "journal\n");
2878 else if (rdev
->raid_disk
< 0)
2879 return sprintf(page
, "none\n");
2881 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2885 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2890 if (test_bit(Journal
, &rdev
->flags
))
2892 if (strncmp(buf
, "none", 4)==0)
2895 err
= kstrtouint(buf
, 10, (unsigned int *)&slot
);
2899 if (rdev
->mddev
->pers
&& slot
== -1) {
2900 /* Setting 'slot' on an active array requires also
2901 * updating the 'rd%d' link, and communicating
2902 * with the personality with ->hot_*_disk.
2903 * For now we only support removing
2904 * failed/spare devices. This normally happens automatically,
2905 * but not when the metadata is externally managed.
2907 if (rdev
->raid_disk
== -1)
2909 /* personality does all needed checks */
2910 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2912 clear_bit(Blocked
, &rdev
->flags
);
2913 remove_and_add_spares(rdev
->mddev
, rdev
);
2914 if (rdev
->raid_disk
>= 0)
2916 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2917 md_wakeup_thread(rdev
->mddev
->thread
);
2918 } else if (rdev
->mddev
->pers
) {
2919 /* Activating a spare .. or possibly reactivating
2920 * if we ever get bitmaps working here.
2924 if (rdev
->raid_disk
!= -1)
2927 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2930 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2933 if (slot
>= rdev
->mddev
->raid_disks
&&
2934 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2937 rdev
->raid_disk
= slot
;
2938 if (test_bit(In_sync
, &rdev
->flags
))
2939 rdev
->saved_raid_disk
= slot
;
2941 rdev
->saved_raid_disk
= -1;
2942 clear_bit(In_sync
, &rdev
->flags
);
2943 clear_bit(Bitmap_sync
, &rdev
->flags
);
2944 err
= rdev
->mddev
->pers
->
2945 hot_add_disk(rdev
->mddev
, rdev
);
2947 rdev
->raid_disk
= -1;
2950 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2951 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2952 /* failure here is OK */;
2953 /* don't wakeup anyone, leave that to userspace. */
2955 if (slot
>= rdev
->mddev
->raid_disks
&&
2956 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2958 rdev
->raid_disk
= slot
;
2959 /* assume it is working */
2960 clear_bit(Faulty
, &rdev
->flags
);
2961 clear_bit(WriteMostly
, &rdev
->flags
);
2962 set_bit(In_sync
, &rdev
->flags
);
2963 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2968 static struct rdev_sysfs_entry rdev_slot
=
2969 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2972 offset_show(struct md_rdev
*rdev
, char *page
)
2974 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2978 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2980 unsigned long long offset
;
2981 if (kstrtoull(buf
, 10, &offset
) < 0)
2983 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2985 if (rdev
->sectors
&& rdev
->mddev
->external
)
2986 /* Must set offset before size, so overlap checks
2989 rdev
->data_offset
= offset
;
2990 rdev
->new_data_offset
= offset
;
2994 static struct rdev_sysfs_entry rdev_offset
=
2995 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2997 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2999 return sprintf(page
, "%llu\n",
3000 (unsigned long long)rdev
->new_data_offset
);
3003 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
3004 const char *buf
, size_t len
)
3006 unsigned long long new_offset
;
3007 struct mddev
*mddev
= rdev
->mddev
;
3009 if (kstrtoull(buf
, 10, &new_offset
) < 0)
3012 if (mddev
->sync_thread
||
3013 test_bit(MD_RECOVERY_RUNNING
,&mddev
->recovery
))
3015 if (new_offset
== rdev
->data_offset
)
3016 /* reset is always permitted */
3018 else if (new_offset
> rdev
->data_offset
) {
3019 /* must not push array size beyond rdev_sectors */
3020 if (new_offset
- rdev
->data_offset
3021 + mddev
->dev_sectors
> rdev
->sectors
)
3024 /* Metadata worries about other space details. */
3026 /* decreasing the offset is inconsistent with a backwards
3029 if (new_offset
< rdev
->data_offset
&&
3030 mddev
->reshape_backwards
)
3032 /* Increasing offset is inconsistent with forwards
3033 * reshape. reshape_direction should be set to
3034 * 'backwards' first.
3036 if (new_offset
> rdev
->data_offset
&&
3037 !mddev
->reshape_backwards
)
3040 if (mddev
->pers
&& mddev
->persistent
&&
3041 !super_types
[mddev
->major_version
]
3042 .allow_new_offset(rdev
, new_offset
))
3044 rdev
->new_data_offset
= new_offset
;
3045 if (new_offset
> rdev
->data_offset
)
3046 mddev
->reshape_backwards
= 1;
3047 else if (new_offset
< rdev
->data_offset
)
3048 mddev
->reshape_backwards
= 0;
3052 static struct rdev_sysfs_entry rdev_new_offset
=
3053 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
3056 rdev_size_show(struct md_rdev
*rdev
, char *page
)
3058 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
3061 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
3063 /* check if two start/length pairs overlap */
3071 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
3073 unsigned long long blocks
;
3076 if (kstrtoull(buf
, 10, &blocks
) < 0)
3079 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
3080 return -EINVAL
; /* sector conversion overflow */
3083 if (new != blocks
* 2)
3084 return -EINVAL
; /* unsigned long long to sector_t overflow */
3091 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3093 struct mddev
*my_mddev
= rdev
->mddev
;
3094 sector_t oldsectors
= rdev
->sectors
;
3097 if (test_bit(Journal
, &rdev
->flags
))
3099 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3101 if (rdev
->data_offset
!= rdev
->new_data_offset
)
3102 return -EINVAL
; /* too confusing */
3103 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
3104 if (my_mddev
->persistent
) {
3105 sectors
= super_types
[my_mddev
->major_version
].
3106 rdev_size_change(rdev
, sectors
);
3109 } else if (!sectors
)
3110 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
3112 if (!my_mddev
->pers
->resize
)
3113 /* Cannot change size for RAID0 or Linear etc */
3116 if (sectors
< my_mddev
->dev_sectors
)
3117 return -EINVAL
; /* component must fit device */
3119 rdev
->sectors
= sectors
;
3120 if (sectors
> oldsectors
&& my_mddev
->external
) {
3121 /* Need to check that all other rdevs with the same
3122 * ->bdev do not overlap. 'rcu' is sufficient to walk
3123 * the rdev lists safely.
3124 * This check does not provide a hard guarantee, it
3125 * just helps avoid dangerous mistakes.
3127 struct mddev
*mddev
;
3129 struct list_head
*tmp
;
3132 for_each_mddev(mddev
, tmp
) {
3133 struct md_rdev
*rdev2
;
3135 rdev_for_each(rdev2
, mddev
)
3136 if (rdev
->bdev
== rdev2
->bdev
&&
3138 overlaps(rdev
->data_offset
, rdev
->sectors
,
3151 /* Someone else could have slipped in a size
3152 * change here, but doing so is just silly.
3153 * We put oldsectors back because we *know* it is
3154 * safe, and trust userspace not to race with
3157 rdev
->sectors
= oldsectors
;
3164 static struct rdev_sysfs_entry rdev_size
=
3165 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3167 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3169 unsigned long long recovery_start
= rdev
->recovery_offset
;
3171 if (test_bit(In_sync
, &rdev
->flags
) ||
3172 recovery_start
== MaxSector
)
3173 return sprintf(page
, "none\n");
3175 return sprintf(page
, "%llu\n", recovery_start
);
3178 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3180 unsigned long long recovery_start
;
3182 if (cmd_match(buf
, "none"))
3183 recovery_start
= MaxSector
;
3184 else if (kstrtoull(buf
, 10, &recovery_start
))
3187 if (rdev
->mddev
->pers
&&
3188 rdev
->raid_disk
>= 0)
3191 rdev
->recovery_offset
= recovery_start
;
3192 if (recovery_start
== MaxSector
)
3193 set_bit(In_sync
, &rdev
->flags
);
3195 clear_bit(In_sync
, &rdev
->flags
);
3199 static struct rdev_sysfs_entry rdev_recovery_start
=
3200 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3202 /* sysfs access to bad-blocks list.
3203 * We present two files.
3204 * 'bad-blocks' lists sector numbers and lengths of ranges that
3205 * are recorded as bad. The list is truncated to fit within
3206 * the one-page limit of sysfs.
3207 * Writing "sector length" to this file adds an acknowledged
3209 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3210 * been acknowledged. Writing to this file adds bad blocks
3211 * without acknowledging them. This is largely for testing.
3213 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3215 return badblocks_show(&rdev
->badblocks
, page
, 0);
3217 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3219 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3220 /* Maybe that ack was all we needed */
3221 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3222 wake_up(&rdev
->blocked_wait
);
3225 static struct rdev_sysfs_entry rdev_bad_blocks
=
3226 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3228 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3230 return badblocks_show(&rdev
->badblocks
, page
, 1);
3232 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3234 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3236 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3237 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3240 ppl_sector_show(struct md_rdev
*rdev
, char *page
)
3242 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->ppl
.sector
);
3246 ppl_sector_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3248 unsigned long long sector
;
3250 if (kstrtoull(buf
, 10, §or
) < 0)
3252 if (sector
!= (sector_t
)sector
)
3255 if (rdev
->mddev
->pers
&& test_bit(MD_HAS_PPL
, &rdev
->mddev
->flags
) &&
3256 rdev
->raid_disk
>= 0)
3259 if (rdev
->mddev
->persistent
) {
3260 if (rdev
->mddev
->major_version
== 0)
3262 if ((sector
> rdev
->sb_start
&&
3263 sector
- rdev
->sb_start
> S16_MAX
) ||
3264 (sector
< rdev
->sb_start
&&
3265 rdev
->sb_start
- sector
> -S16_MIN
))
3267 rdev
->ppl
.offset
= sector
- rdev
->sb_start
;
3268 } else if (!rdev
->mddev
->external
) {
3271 rdev
->ppl
.sector
= sector
;
3275 static struct rdev_sysfs_entry rdev_ppl_sector
=
3276 __ATTR(ppl_sector
, S_IRUGO
|S_IWUSR
, ppl_sector_show
, ppl_sector_store
);
3279 ppl_size_show(struct md_rdev
*rdev
, char *page
)
3281 return sprintf(page
, "%u\n", rdev
->ppl
.size
);
3285 ppl_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3289 if (kstrtouint(buf
, 10, &size
) < 0)
3292 if (rdev
->mddev
->pers
&& test_bit(MD_HAS_PPL
, &rdev
->mddev
->flags
) &&
3293 rdev
->raid_disk
>= 0)
3296 if (rdev
->mddev
->persistent
) {
3297 if (rdev
->mddev
->major_version
== 0)
3301 } else if (!rdev
->mddev
->external
) {
3304 rdev
->ppl
.size
= size
;
3308 static struct rdev_sysfs_entry rdev_ppl_size
=
3309 __ATTR(ppl_size
, S_IRUGO
|S_IWUSR
, ppl_size_show
, ppl_size_store
);
3311 static struct attribute
*rdev_default_attrs
[] = {
3316 &rdev_new_offset
.attr
,
3318 &rdev_recovery_start
.attr
,
3319 &rdev_bad_blocks
.attr
,
3320 &rdev_unack_bad_blocks
.attr
,
3321 &rdev_ppl_sector
.attr
,
3322 &rdev_ppl_size
.attr
,
3326 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3328 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3329 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3335 return entry
->show(rdev
, page
);
3339 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3340 const char *page
, size_t length
)
3342 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3343 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3345 struct mddev
*mddev
= rdev
->mddev
;
3349 if (!capable(CAP_SYS_ADMIN
))
3351 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3353 if (rdev
->mddev
== NULL
)
3356 rv
= entry
->store(rdev
, page
, length
);
3357 mddev_unlock(mddev
);
3362 static void rdev_free(struct kobject
*ko
)
3364 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3367 static const struct sysfs_ops rdev_sysfs_ops
= {
3368 .show
= rdev_attr_show
,
3369 .store
= rdev_attr_store
,
3371 static struct kobj_type rdev_ktype
= {
3372 .release
= rdev_free
,
3373 .sysfs_ops
= &rdev_sysfs_ops
,
3374 .default_attrs
= rdev_default_attrs
,
3377 int md_rdev_init(struct md_rdev
*rdev
)
3380 rdev
->saved_raid_disk
= -1;
3381 rdev
->raid_disk
= -1;
3383 rdev
->data_offset
= 0;
3384 rdev
->new_data_offset
= 0;
3385 rdev
->sb_events
= 0;
3386 rdev
->last_read_error
= 0;
3387 rdev
->sb_loaded
= 0;
3388 rdev
->bb_page
= NULL
;
3389 atomic_set(&rdev
->nr_pending
, 0);
3390 atomic_set(&rdev
->read_errors
, 0);
3391 atomic_set(&rdev
->corrected_errors
, 0);
3393 INIT_LIST_HEAD(&rdev
->same_set
);
3394 init_waitqueue_head(&rdev
->blocked_wait
);
3396 /* Add space to store bad block list.
3397 * This reserves the space even on arrays where it cannot
3398 * be used - I wonder if that matters
3400 return badblocks_init(&rdev
->badblocks
, 0);
3402 EXPORT_SYMBOL_GPL(md_rdev_init
);
3404 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3406 * mark the device faulty if:
3408 * - the device is nonexistent (zero size)
3409 * - the device has no valid superblock
3411 * a faulty rdev _never_ has rdev->sb set.
3413 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3415 char b
[BDEVNAME_SIZE
];
3417 struct md_rdev
*rdev
;
3420 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3422 return ERR_PTR(-ENOMEM
);
3424 err
= md_rdev_init(rdev
);
3427 err
= alloc_disk_sb(rdev
);
3431 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3435 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3437 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3439 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3440 bdevname(rdev
->bdev
,b
));
3445 if (super_format
>= 0) {
3446 err
= super_types
[super_format
].
3447 load_super(rdev
, NULL
, super_minor
);
3448 if (err
== -EINVAL
) {
3449 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3450 bdevname(rdev
->bdev
,b
),
3451 super_format
, super_minor
);
3455 pr_warn("md: could not read %s's sb, not importing!\n",
3456 bdevname(rdev
->bdev
,b
));
3466 md_rdev_clear(rdev
);
3468 return ERR_PTR(err
);
3472 * Check a full RAID array for plausibility
3475 static void analyze_sbs(struct mddev
*mddev
)
3478 struct md_rdev
*rdev
, *freshest
, *tmp
;
3479 char b
[BDEVNAME_SIZE
];
3482 rdev_for_each_safe(rdev
, tmp
, mddev
)
3483 switch (super_types
[mddev
->major_version
].
3484 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3491 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3492 bdevname(rdev
->bdev
,b
));
3493 md_kick_rdev_from_array(rdev
);
3496 super_types
[mddev
->major_version
].
3497 validate_super(mddev
, freshest
);
3500 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3501 if (mddev
->max_disks
&&
3502 (rdev
->desc_nr
>= mddev
->max_disks
||
3503 i
> mddev
->max_disks
)) {
3504 pr_warn("md: %s: %s: only %d devices permitted\n",
3505 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3507 md_kick_rdev_from_array(rdev
);
3510 if (rdev
!= freshest
) {
3511 if (super_types
[mddev
->major_version
].
3512 validate_super(mddev
, rdev
)) {
3513 pr_warn("md: kicking non-fresh %s from array!\n",
3514 bdevname(rdev
->bdev
,b
));
3515 md_kick_rdev_from_array(rdev
);
3519 if (mddev
->level
== LEVEL_MULTIPATH
) {
3520 rdev
->desc_nr
= i
++;
3521 rdev
->raid_disk
= rdev
->desc_nr
;
3522 set_bit(In_sync
, &rdev
->flags
);
3523 } else if (rdev
->raid_disk
>=
3524 (mddev
->raid_disks
- min(0, mddev
->delta_disks
)) &&
3525 !test_bit(Journal
, &rdev
->flags
)) {
3526 rdev
->raid_disk
= -1;
3527 clear_bit(In_sync
, &rdev
->flags
);
3532 /* Read a fixed-point number.
3533 * Numbers in sysfs attributes should be in "standard" units where
3534 * possible, so time should be in seconds.
3535 * However we internally use a a much smaller unit such as
3536 * milliseconds or jiffies.
3537 * This function takes a decimal number with a possible fractional
3538 * component, and produces an integer which is the result of
3539 * multiplying that number by 10^'scale'.
3540 * all without any floating-point arithmetic.
3542 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3544 unsigned long result
= 0;
3546 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3549 else if (decimals
< scale
) {
3552 result
= result
* 10 + value
;
3564 while (decimals
< scale
) {
3573 safe_delay_show(struct mddev
*mddev
, char *page
)
3575 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3576 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3579 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3583 if (mddev_is_clustered(mddev
)) {
3584 pr_warn("md: Safemode is disabled for clustered mode\n");
3588 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3591 mddev
->safemode_delay
= 0;
3593 unsigned long old_delay
= mddev
->safemode_delay
;
3594 unsigned long new_delay
= (msec
*HZ
)/1000;
3598 mddev
->safemode_delay
= new_delay
;
3599 if (new_delay
< old_delay
|| old_delay
== 0)
3600 mod_timer(&mddev
->safemode_timer
, jiffies
+1);
3604 static struct md_sysfs_entry md_safe_delay
=
3605 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3608 level_show(struct mddev
*mddev
, char *page
)
3610 struct md_personality
*p
;
3612 spin_lock(&mddev
->lock
);
3615 ret
= sprintf(page
, "%s\n", p
->name
);
3616 else if (mddev
->clevel
[0])
3617 ret
= sprintf(page
, "%s\n", mddev
->clevel
);
3618 else if (mddev
->level
!= LEVEL_NONE
)
3619 ret
= sprintf(page
, "%d\n", mddev
->level
);
3622 spin_unlock(&mddev
->lock
);
3627 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3632 struct md_personality
*pers
, *oldpers
;
3634 void *priv
, *oldpriv
;
3635 struct md_rdev
*rdev
;
3637 if (slen
== 0 || slen
>= sizeof(clevel
))
3640 rv
= mddev_lock(mddev
);
3644 if (mddev
->pers
== NULL
) {
3645 strncpy(mddev
->clevel
, buf
, slen
);
3646 if (mddev
->clevel
[slen
-1] == '\n')
3648 mddev
->clevel
[slen
] = 0;
3649 mddev
->level
= LEVEL_NONE
;
3657 /* request to change the personality. Need to ensure:
3658 * - array is not engaged in resync/recovery/reshape
3659 * - old personality can be suspended
3660 * - new personality will access other array.
3664 if (mddev
->sync_thread
||
3665 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3666 mddev
->reshape_position
!= MaxSector
||
3667 mddev
->sysfs_active
)
3671 if (!mddev
->pers
->quiesce
) {
3672 pr_warn("md: %s: %s does not support online personality change\n",
3673 mdname(mddev
), mddev
->pers
->name
);
3677 /* Now find the new personality */
3678 strncpy(clevel
, buf
, slen
);
3679 if (clevel
[slen
-1] == '\n')
3682 if (kstrtol(clevel
, 10, &level
))
3685 if (request_module("md-%s", clevel
) != 0)
3686 request_module("md-level-%s", clevel
);
3687 spin_lock(&pers_lock
);
3688 pers
= find_pers(level
, clevel
);
3689 if (!pers
|| !try_module_get(pers
->owner
)) {
3690 spin_unlock(&pers_lock
);
3691 pr_warn("md: personality %s not loaded\n", clevel
);
3695 spin_unlock(&pers_lock
);
3697 if (pers
== mddev
->pers
) {
3698 /* Nothing to do! */
3699 module_put(pers
->owner
);
3703 if (!pers
->takeover
) {
3704 module_put(pers
->owner
);
3705 pr_warn("md: %s: %s does not support personality takeover\n",
3706 mdname(mddev
), clevel
);
3711 rdev_for_each(rdev
, mddev
)
3712 rdev
->new_raid_disk
= rdev
->raid_disk
;
3714 /* ->takeover must set new_* and/or delta_disks
3715 * if it succeeds, and may set them when it fails.
3717 priv
= pers
->takeover(mddev
);
3719 mddev
->new_level
= mddev
->level
;
3720 mddev
->new_layout
= mddev
->layout
;
3721 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3722 mddev
->raid_disks
-= mddev
->delta_disks
;
3723 mddev
->delta_disks
= 0;
3724 mddev
->reshape_backwards
= 0;
3725 module_put(pers
->owner
);
3726 pr_warn("md: %s: %s would not accept array\n",
3727 mdname(mddev
), clevel
);
3732 /* Looks like we have a winner */
3733 mddev_suspend(mddev
);
3734 mddev_detach(mddev
);
3736 spin_lock(&mddev
->lock
);
3737 oldpers
= mddev
->pers
;
3738 oldpriv
= mddev
->private;
3740 mddev
->private = priv
;
3741 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3742 mddev
->level
= mddev
->new_level
;
3743 mddev
->layout
= mddev
->new_layout
;
3744 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3745 mddev
->delta_disks
= 0;
3746 mddev
->reshape_backwards
= 0;
3747 mddev
->degraded
= 0;
3748 spin_unlock(&mddev
->lock
);
3750 if (oldpers
->sync_request
== NULL
&&
3752 /* We are converting from a no-redundancy array
3753 * to a redundancy array and metadata is managed
3754 * externally so we need to be sure that writes
3755 * won't block due to a need to transition
3757 * until external management is started.
3760 mddev
->safemode_delay
= 0;
3761 mddev
->safemode
= 0;
3764 oldpers
->free(mddev
, oldpriv
);
3766 if (oldpers
->sync_request
== NULL
&&
3767 pers
->sync_request
!= NULL
) {
3768 /* need to add the md_redundancy_group */
3769 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3770 pr_warn("md: cannot register extra attributes for %s\n",
3772 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3774 if (oldpers
->sync_request
!= NULL
&&
3775 pers
->sync_request
== NULL
) {
3776 /* need to remove the md_redundancy_group */
3777 if (mddev
->to_remove
== NULL
)
3778 mddev
->to_remove
= &md_redundancy_group
;
3781 module_put(oldpers
->owner
);
3783 rdev_for_each(rdev
, mddev
) {
3784 if (rdev
->raid_disk
< 0)
3786 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3787 rdev
->new_raid_disk
= -1;
3788 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3790 sysfs_unlink_rdev(mddev
, rdev
);
3792 rdev_for_each(rdev
, mddev
) {
3793 if (rdev
->raid_disk
< 0)
3795 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3797 rdev
->raid_disk
= rdev
->new_raid_disk
;
3798 if (rdev
->raid_disk
< 0)
3799 clear_bit(In_sync
, &rdev
->flags
);
3801 if (sysfs_link_rdev(mddev
, rdev
))
3802 pr_warn("md: cannot register rd%d for %s after level change\n",
3803 rdev
->raid_disk
, mdname(mddev
));
3807 if (pers
->sync_request
== NULL
) {
3808 /* this is now an array without redundancy, so
3809 * it must always be in_sync
3812 del_timer_sync(&mddev
->safemode_timer
);
3814 blk_set_stacking_limits(&mddev
->queue
->limits
);
3816 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
3817 mddev_resume(mddev
);
3819 md_update_sb(mddev
, 1);
3820 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3821 md_new_event(mddev
);
3824 mddev_unlock(mddev
);
3828 static struct md_sysfs_entry md_level
=
3829 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3832 layout_show(struct mddev
*mddev
, char *page
)
3834 /* just a number, not meaningful for all levels */
3835 if (mddev
->reshape_position
!= MaxSector
&&
3836 mddev
->layout
!= mddev
->new_layout
)
3837 return sprintf(page
, "%d (%d)\n",
3838 mddev
->new_layout
, mddev
->layout
);
3839 return sprintf(page
, "%d\n", mddev
->layout
);
3843 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3848 err
= kstrtouint(buf
, 10, &n
);
3851 err
= mddev_lock(mddev
);
3856 if (mddev
->pers
->check_reshape
== NULL
)
3861 mddev
->new_layout
= n
;
3862 err
= mddev
->pers
->check_reshape(mddev
);
3864 mddev
->new_layout
= mddev
->layout
;
3867 mddev
->new_layout
= n
;
3868 if (mddev
->reshape_position
== MaxSector
)
3871 mddev_unlock(mddev
);
3874 static struct md_sysfs_entry md_layout
=
3875 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3878 raid_disks_show(struct mddev
*mddev
, char *page
)
3880 if (mddev
->raid_disks
== 0)
3882 if (mddev
->reshape_position
!= MaxSector
&&
3883 mddev
->delta_disks
!= 0)
3884 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3885 mddev
->raid_disks
- mddev
->delta_disks
);
3886 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3889 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3892 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3897 err
= kstrtouint(buf
, 10, &n
);
3901 err
= mddev_lock(mddev
);
3905 err
= update_raid_disks(mddev
, n
);
3906 else if (mddev
->reshape_position
!= MaxSector
) {
3907 struct md_rdev
*rdev
;
3908 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3911 rdev_for_each(rdev
, mddev
) {
3913 rdev
->data_offset
< rdev
->new_data_offset
)
3916 rdev
->data_offset
> rdev
->new_data_offset
)
3920 mddev
->delta_disks
= n
- olddisks
;
3921 mddev
->raid_disks
= n
;
3922 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3924 mddev
->raid_disks
= n
;
3926 mddev_unlock(mddev
);
3927 return err
? err
: len
;
3929 static struct md_sysfs_entry md_raid_disks
=
3930 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3933 chunk_size_show(struct mddev
*mddev
, char *page
)
3935 if (mddev
->reshape_position
!= MaxSector
&&
3936 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3937 return sprintf(page
, "%d (%d)\n",
3938 mddev
->new_chunk_sectors
<< 9,
3939 mddev
->chunk_sectors
<< 9);
3940 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3944 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3949 err
= kstrtoul(buf
, 10, &n
);
3953 err
= mddev_lock(mddev
);
3957 if (mddev
->pers
->check_reshape
== NULL
)
3962 mddev
->new_chunk_sectors
= n
>> 9;
3963 err
= mddev
->pers
->check_reshape(mddev
);
3965 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3968 mddev
->new_chunk_sectors
= n
>> 9;
3969 if (mddev
->reshape_position
== MaxSector
)
3970 mddev
->chunk_sectors
= n
>> 9;
3972 mddev_unlock(mddev
);
3975 static struct md_sysfs_entry md_chunk_size
=
3976 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3979 resync_start_show(struct mddev
*mddev
, char *page
)
3981 if (mddev
->recovery_cp
== MaxSector
)
3982 return sprintf(page
, "none\n");
3983 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3987 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3989 unsigned long long n
;
3992 if (cmd_match(buf
, "none"))
3995 err
= kstrtoull(buf
, 10, &n
);
3998 if (n
!= (sector_t
)n
)
4002 err
= mddev_lock(mddev
);
4005 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4009 mddev
->recovery_cp
= n
;
4011 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
4013 mddev_unlock(mddev
);
4016 static struct md_sysfs_entry md_resync_start
=
4017 __ATTR_PREALLOC(resync_start
, S_IRUGO
|S_IWUSR
,
4018 resync_start_show
, resync_start_store
);
4021 * The array state can be:
4024 * No devices, no size, no level
4025 * Equivalent to STOP_ARRAY ioctl
4027 * May have some settings, but array is not active
4028 * all IO results in error
4029 * When written, doesn't tear down array, but just stops it
4030 * suspended (not supported yet)
4031 * All IO requests will block. The array can be reconfigured.
4032 * Writing this, if accepted, will block until array is quiescent
4034 * no resync can happen. no superblocks get written.
4035 * write requests fail
4037 * like readonly, but behaves like 'clean' on a write request.
4039 * clean - no pending writes, but otherwise active.
4040 * When written to inactive array, starts without resync
4041 * If a write request arrives then
4042 * if metadata is known, mark 'dirty' and switch to 'active'.
4043 * if not known, block and switch to write-pending
4044 * If written to an active array that has pending writes, then fails.
4046 * fully active: IO and resync can be happening.
4047 * When written to inactive array, starts with resync
4050 * clean, but writes are blocked waiting for 'active' to be written.
4053 * like active, but no writes have been seen for a while (100msec).
4056 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
4057 write_pending
, active_idle
, bad_word
};
4058 static char *array_states
[] = {
4059 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4060 "write-pending", "active-idle", NULL
};
4062 static int match_word(const char *word
, char **list
)
4065 for (n
=0; list
[n
]; n
++)
4066 if (cmd_match(word
, list
[n
]))
4072 array_state_show(struct mddev
*mddev
, char *page
)
4074 enum array_state st
= inactive
;
4085 spin_lock(&mddev
->lock
);
4086 if (test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
))
4088 else if (mddev
->in_sync
)
4090 else if (mddev
->safemode
)
4094 spin_unlock(&mddev
->lock
);
4097 if (list_empty(&mddev
->disks
) &&
4098 mddev
->raid_disks
== 0 &&
4099 mddev
->dev_sectors
== 0)
4104 return sprintf(page
, "%s\n", array_states
[st
]);
4107 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
4108 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
4109 static int do_md_run(struct mddev
*mddev
);
4110 static int restart_array(struct mddev
*mddev
);
4113 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4116 enum array_state st
= match_word(buf
, array_states
);
4118 if (mddev
->pers
&& (st
== active
|| st
== clean
) && mddev
->ro
!= 1) {
4119 /* don't take reconfig_mutex when toggling between
4122 spin_lock(&mddev
->lock
);
4124 restart_array(mddev
);
4125 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
4126 md_wakeup_thread(mddev
->thread
);
4127 wake_up(&mddev
->sb_wait
);
4128 } else /* st == clean */ {
4129 restart_array(mddev
);
4130 if (!set_in_sync(mddev
))
4134 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4135 spin_unlock(&mddev
->lock
);
4138 err
= mddev_lock(mddev
);
4146 /* stopping an active array */
4147 err
= do_md_stop(mddev
, 0, NULL
);
4150 /* stopping an active array */
4152 err
= do_md_stop(mddev
, 2, NULL
);
4154 err
= 0; /* already inactive */
4157 break; /* not supported yet */
4160 err
= md_set_readonly(mddev
, NULL
);
4163 set_disk_ro(mddev
->gendisk
, 1);
4164 err
= do_md_run(mddev
);
4170 err
= md_set_readonly(mddev
, NULL
);
4171 else if (mddev
->ro
== 1)
4172 err
= restart_array(mddev
);
4175 set_disk_ro(mddev
->gendisk
, 0);
4179 err
= do_md_run(mddev
);
4184 err
= restart_array(mddev
);
4187 spin_lock(&mddev
->lock
);
4188 if (!set_in_sync(mddev
))
4190 spin_unlock(&mddev
->lock
);
4196 err
= restart_array(mddev
);
4199 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
4200 wake_up(&mddev
->sb_wait
);
4204 set_disk_ro(mddev
->gendisk
, 0);
4205 err
= do_md_run(mddev
);
4210 /* these cannot be set */
4215 if (mddev
->hold_active
== UNTIL_IOCTL
)
4216 mddev
->hold_active
= 0;
4217 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4219 mddev_unlock(mddev
);
4222 static struct md_sysfs_entry md_array_state
=
4223 __ATTR_PREALLOC(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
4226 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
4227 return sprintf(page
, "%d\n",
4228 atomic_read(&mddev
->max_corr_read_errors
));
4232 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4237 rv
= kstrtouint(buf
, 10, &n
);
4240 atomic_set(&mddev
->max_corr_read_errors
, n
);
4244 static struct md_sysfs_entry max_corr_read_errors
=
4245 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4246 max_corrected_read_errors_store
);
4249 null_show(struct mddev
*mddev
, char *page
)
4255 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4257 /* buf must be %d:%d\n? giving major and minor numbers */
4258 /* The new device is added to the array.
4259 * If the array has a persistent superblock, we read the
4260 * superblock to initialise info and check validity.
4261 * Otherwise, only checking done is that in bind_rdev_to_array,
4262 * which mainly checks size.
4265 int major
= simple_strtoul(buf
, &e
, 10);
4268 struct md_rdev
*rdev
;
4271 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4273 minor
= simple_strtoul(e
+1, &e
, 10);
4274 if (*e
&& *e
!= '\n')
4276 dev
= MKDEV(major
, minor
);
4277 if (major
!= MAJOR(dev
) ||
4278 minor
!= MINOR(dev
))
4281 flush_workqueue(md_misc_wq
);
4283 err
= mddev_lock(mddev
);
4286 if (mddev
->persistent
) {
4287 rdev
= md_import_device(dev
, mddev
->major_version
,
4288 mddev
->minor_version
);
4289 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4290 struct md_rdev
*rdev0
4291 = list_entry(mddev
->disks
.next
,
4292 struct md_rdev
, same_set
);
4293 err
= super_types
[mddev
->major_version
]
4294 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4298 } else if (mddev
->external
)
4299 rdev
= md_import_device(dev
, -2, -1);
4301 rdev
= md_import_device(dev
, -1, -1);
4304 mddev_unlock(mddev
);
4305 return PTR_ERR(rdev
);
4307 err
= bind_rdev_to_array(rdev
, mddev
);
4311 mddev_unlock(mddev
);
4313 md_new_event(mddev
);
4314 return err
? err
: len
;
4317 static struct md_sysfs_entry md_new_device
=
4318 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4321 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4324 unsigned long chunk
, end_chunk
;
4327 err
= mddev_lock(mddev
);
4332 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4334 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4335 if (buf
== end
) break;
4336 if (*end
== '-') { /* range */
4338 end_chunk
= simple_strtoul(buf
, &end
, 0);
4339 if (buf
== end
) break;
4341 if (*end
&& !isspace(*end
)) break;
4342 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4343 buf
= skip_spaces(end
);
4345 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4347 mddev_unlock(mddev
);
4351 static struct md_sysfs_entry md_bitmap
=
4352 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4355 size_show(struct mddev
*mddev
, char *page
)
4357 return sprintf(page
, "%llu\n",
4358 (unsigned long long)mddev
->dev_sectors
/ 2);
4361 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4364 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4366 /* If array is inactive, we can reduce the component size, but
4367 * not increase it (except from 0).
4368 * If array is active, we can try an on-line resize
4371 int err
= strict_blocks_to_sectors(buf
, §ors
);
4375 err
= mddev_lock(mddev
);
4379 err
= update_size(mddev
, sectors
);
4381 md_update_sb(mddev
, 1);
4383 if (mddev
->dev_sectors
== 0 ||
4384 mddev
->dev_sectors
> sectors
)
4385 mddev
->dev_sectors
= sectors
;
4389 mddev_unlock(mddev
);
4390 return err
? err
: len
;
4393 static struct md_sysfs_entry md_size
=
4394 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4396 /* Metadata version.
4398 * 'none' for arrays with no metadata (good luck...)
4399 * 'external' for arrays with externally managed metadata,
4400 * or N.M for internally known formats
4403 metadata_show(struct mddev
*mddev
, char *page
)
4405 if (mddev
->persistent
)
4406 return sprintf(page
, "%d.%d\n",
4407 mddev
->major_version
, mddev
->minor_version
);
4408 else if (mddev
->external
)
4409 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4411 return sprintf(page
, "none\n");
4415 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4420 /* Changing the details of 'external' metadata is
4421 * always permitted. Otherwise there must be
4422 * no devices attached to the array.
4425 err
= mddev_lock(mddev
);
4429 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4431 else if (!list_empty(&mddev
->disks
))
4435 if (cmd_match(buf
, "none")) {
4436 mddev
->persistent
= 0;
4437 mddev
->external
= 0;
4438 mddev
->major_version
= 0;
4439 mddev
->minor_version
= 90;
4442 if (strncmp(buf
, "external:", 9) == 0) {
4443 size_t namelen
= len
-9;
4444 if (namelen
>= sizeof(mddev
->metadata_type
))
4445 namelen
= sizeof(mddev
->metadata_type
)-1;
4446 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4447 mddev
->metadata_type
[namelen
] = 0;
4448 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4449 mddev
->metadata_type
[--namelen
] = 0;
4450 mddev
->persistent
= 0;
4451 mddev
->external
= 1;
4452 mddev
->major_version
= 0;
4453 mddev
->minor_version
= 90;
4456 major
= simple_strtoul(buf
, &e
, 10);
4458 if (e
==buf
|| *e
!= '.')
4461 minor
= simple_strtoul(buf
, &e
, 10);
4462 if (e
==buf
|| (*e
&& *e
!= '\n') )
4465 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4467 mddev
->major_version
= major
;
4468 mddev
->minor_version
= minor
;
4469 mddev
->persistent
= 1;
4470 mddev
->external
= 0;
4473 mddev_unlock(mddev
);
4477 static struct md_sysfs_entry md_metadata
=
4478 __ATTR_PREALLOC(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4481 action_show(struct mddev
*mddev
, char *page
)
4483 char *type
= "idle";
4484 unsigned long recovery
= mddev
->recovery
;
4485 if (test_bit(MD_RECOVERY_FROZEN
, &recovery
))
4487 else if (test_bit(MD_RECOVERY_RUNNING
, &recovery
) ||
4488 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &recovery
))) {
4489 if (test_bit(MD_RECOVERY_RESHAPE
, &recovery
))
4491 else if (test_bit(MD_RECOVERY_SYNC
, &recovery
)) {
4492 if (!test_bit(MD_RECOVERY_REQUESTED
, &recovery
))
4494 else if (test_bit(MD_RECOVERY_CHECK
, &recovery
))
4498 } else if (test_bit(MD_RECOVERY_RECOVER
, &recovery
))
4500 else if (mddev
->reshape_position
!= MaxSector
)
4503 return sprintf(page
, "%s\n", type
);
4507 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4509 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4513 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4514 if (cmd_match(page
, "frozen"))
4515 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4517 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4518 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4519 mddev_lock(mddev
) == 0) {
4520 flush_workqueue(md_misc_wq
);
4521 if (mddev
->sync_thread
) {
4522 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4523 md_reap_sync_thread(mddev
);
4525 mddev_unlock(mddev
);
4527 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4529 else if (cmd_match(page
, "resync"))
4530 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4531 else if (cmd_match(page
, "recover")) {
4532 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4533 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4534 } else if (cmd_match(page
, "reshape")) {
4536 if (mddev
->pers
->start_reshape
== NULL
)
4538 err
= mddev_lock(mddev
);
4540 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4543 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4544 err
= mddev
->pers
->start_reshape(mddev
);
4546 mddev_unlock(mddev
);
4550 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4552 if (cmd_match(page
, "check"))
4553 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4554 else if (!cmd_match(page
, "repair"))
4556 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4557 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4558 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4560 if (mddev
->ro
== 2) {
4561 /* A write to sync_action is enough to justify
4562 * canceling read-auto mode
4565 md_wakeup_thread(mddev
->sync_thread
);
4567 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4568 md_wakeup_thread(mddev
->thread
);
4569 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4573 static struct md_sysfs_entry md_scan_mode
=
4574 __ATTR_PREALLOC(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4577 last_sync_action_show(struct mddev
*mddev
, char *page
)
4579 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4582 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4585 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4587 return sprintf(page
, "%llu\n",
4588 (unsigned long long)
4589 atomic64_read(&mddev
->resync_mismatches
));
4592 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4595 sync_min_show(struct mddev
*mddev
, char *page
)
4597 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4598 mddev
->sync_speed_min
? "local": "system");
4602 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4607 if (strncmp(buf
, "system", 6)==0) {
4610 rv
= kstrtouint(buf
, 10, &min
);
4616 mddev
->sync_speed_min
= min
;
4620 static struct md_sysfs_entry md_sync_min
=
4621 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4624 sync_max_show(struct mddev
*mddev
, char *page
)
4626 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4627 mddev
->sync_speed_max
? "local": "system");
4631 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4636 if (strncmp(buf
, "system", 6)==0) {
4639 rv
= kstrtouint(buf
, 10, &max
);
4645 mddev
->sync_speed_max
= max
;
4649 static struct md_sysfs_entry md_sync_max
=
4650 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4653 degraded_show(struct mddev
*mddev
, char *page
)
4655 return sprintf(page
, "%d\n", mddev
->degraded
);
4657 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4660 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4662 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4666 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4670 if (kstrtol(buf
, 10, &n
))
4673 if (n
!= 0 && n
!= 1)
4676 mddev
->parallel_resync
= n
;
4678 if (mddev
->sync_thread
)
4679 wake_up(&resync_wait
);
4684 /* force parallel resync, even with shared block devices */
4685 static struct md_sysfs_entry md_sync_force_parallel
=
4686 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4687 sync_force_parallel_show
, sync_force_parallel_store
);
4690 sync_speed_show(struct mddev
*mddev
, char *page
)
4692 unsigned long resync
, dt
, db
;
4693 if (mddev
->curr_resync
== 0)
4694 return sprintf(page
, "none\n");
4695 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4696 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4698 db
= resync
- mddev
->resync_mark_cnt
;
4699 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4702 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4705 sync_completed_show(struct mddev
*mddev
, char *page
)
4707 unsigned long long max_sectors
, resync
;
4709 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4710 return sprintf(page
, "none\n");
4712 if (mddev
->curr_resync
== 1 ||
4713 mddev
->curr_resync
== 2)
4714 return sprintf(page
, "delayed\n");
4716 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4717 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4718 max_sectors
= mddev
->resync_max_sectors
;
4720 max_sectors
= mddev
->dev_sectors
;
4722 resync
= mddev
->curr_resync_completed
;
4723 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4726 static struct md_sysfs_entry md_sync_completed
=
4727 __ATTR_PREALLOC(sync_completed
, S_IRUGO
, sync_completed_show
, NULL
);
4730 min_sync_show(struct mddev
*mddev
, char *page
)
4732 return sprintf(page
, "%llu\n",
4733 (unsigned long long)mddev
->resync_min
);
4736 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4738 unsigned long long min
;
4741 if (kstrtoull(buf
, 10, &min
))
4744 spin_lock(&mddev
->lock
);
4746 if (min
> mddev
->resync_max
)
4750 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4753 /* Round down to multiple of 4K for safety */
4754 mddev
->resync_min
= round_down(min
, 8);
4758 spin_unlock(&mddev
->lock
);
4762 static struct md_sysfs_entry md_min_sync
=
4763 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4766 max_sync_show(struct mddev
*mddev
, char *page
)
4768 if (mddev
->resync_max
== MaxSector
)
4769 return sprintf(page
, "max\n");
4771 return sprintf(page
, "%llu\n",
4772 (unsigned long long)mddev
->resync_max
);
4775 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4778 spin_lock(&mddev
->lock
);
4779 if (strncmp(buf
, "max", 3) == 0)
4780 mddev
->resync_max
= MaxSector
;
4782 unsigned long long max
;
4786 if (kstrtoull(buf
, 10, &max
))
4788 if (max
< mddev
->resync_min
)
4792 if (max
< mddev
->resync_max
&&
4794 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4797 /* Must be a multiple of chunk_size */
4798 chunk
= mddev
->chunk_sectors
;
4800 sector_t temp
= max
;
4803 if (sector_div(temp
, chunk
))
4806 mddev
->resync_max
= max
;
4808 wake_up(&mddev
->recovery_wait
);
4811 spin_unlock(&mddev
->lock
);
4815 static struct md_sysfs_entry md_max_sync
=
4816 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4819 suspend_lo_show(struct mddev
*mddev
, char *page
)
4821 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4825 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4827 unsigned long long old
, new;
4830 err
= kstrtoull(buf
, 10, &new);
4833 if (new != (sector_t
)new)
4836 err
= mddev_lock(mddev
);
4840 if (mddev
->pers
== NULL
||
4841 mddev
->pers
->quiesce
== NULL
)
4843 old
= mddev
->suspend_lo
;
4844 mddev
->suspend_lo
= new;
4846 /* Shrinking suspended region */
4847 mddev
->pers
->quiesce(mddev
, 2);
4849 /* Expanding suspended region - need to wait */
4850 mddev
->pers
->quiesce(mddev
, 1);
4851 mddev
->pers
->quiesce(mddev
, 0);
4855 mddev_unlock(mddev
);
4858 static struct md_sysfs_entry md_suspend_lo
=
4859 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4862 suspend_hi_show(struct mddev
*mddev
, char *page
)
4864 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4868 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4870 unsigned long long old
, new;
4873 err
= kstrtoull(buf
, 10, &new);
4876 if (new != (sector_t
)new)
4879 err
= mddev_lock(mddev
);
4883 if (mddev
->pers
== NULL
||
4884 mddev
->pers
->quiesce
== NULL
)
4886 old
= mddev
->suspend_hi
;
4887 mddev
->suspend_hi
= new;
4889 /* Shrinking suspended region */
4890 mddev
->pers
->quiesce(mddev
, 2);
4892 /* Expanding suspended region - need to wait */
4893 mddev
->pers
->quiesce(mddev
, 1);
4894 mddev
->pers
->quiesce(mddev
, 0);
4898 mddev_unlock(mddev
);
4901 static struct md_sysfs_entry md_suspend_hi
=
4902 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4905 reshape_position_show(struct mddev
*mddev
, char *page
)
4907 if (mddev
->reshape_position
!= MaxSector
)
4908 return sprintf(page
, "%llu\n",
4909 (unsigned long long)mddev
->reshape_position
);
4910 strcpy(page
, "none\n");
4915 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4917 struct md_rdev
*rdev
;
4918 unsigned long long new;
4921 err
= kstrtoull(buf
, 10, &new);
4924 if (new != (sector_t
)new)
4926 err
= mddev_lock(mddev
);
4932 mddev
->reshape_position
= new;
4933 mddev
->delta_disks
= 0;
4934 mddev
->reshape_backwards
= 0;
4935 mddev
->new_level
= mddev
->level
;
4936 mddev
->new_layout
= mddev
->layout
;
4937 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4938 rdev_for_each(rdev
, mddev
)
4939 rdev
->new_data_offset
= rdev
->data_offset
;
4942 mddev_unlock(mddev
);
4946 static struct md_sysfs_entry md_reshape_position
=
4947 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4948 reshape_position_store
);
4951 reshape_direction_show(struct mddev
*mddev
, char *page
)
4953 return sprintf(page
, "%s\n",
4954 mddev
->reshape_backwards
? "backwards" : "forwards");
4958 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4963 if (cmd_match(buf
, "forwards"))
4965 else if (cmd_match(buf
, "backwards"))
4969 if (mddev
->reshape_backwards
== backwards
)
4972 err
= mddev_lock(mddev
);
4975 /* check if we are allowed to change */
4976 if (mddev
->delta_disks
)
4978 else if (mddev
->persistent
&&
4979 mddev
->major_version
== 0)
4982 mddev
->reshape_backwards
= backwards
;
4983 mddev_unlock(mddev
);
4987 static struct md_sysfs_entry md_reshape_direction
=
4988 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4989 reshape_direction_store
);
4992 array_size_show(struct mddev
*mddev
, char *page
)
4994 if (mddev
->external_size
)
4995 return sprintf(page
, "%llu\n",
4996 (unsigned long long)mddev
->array_sectors
/2);
4998 return sprintf(page
, "default\n");
5002 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
5007 err
= mddev_lock(mddev
);
5011 /* cluster raid doesn't support change array_sectors */
5012 if (mddev_is_clustered(mddev
)) {
5013 mddev_unlock(mddev
);
5017 if (strncmp(buf
, "default", 7) == 0) {
5019 sectors
= mddev
->pers
->size(mddev
, 0, 0);
5021 sectors
= mddev
->array_sectors
;
5023 mddev
->external_size
= 0;
5025 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
5027 else if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
5030 mddev
->external_size
= 1;
5034 mddev
->array_sectors
= sectors
;
5036 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5037 revalidate_disk(mddev
->gendisk
);
5040 mddev_unlock(mddev
);
5044 static struct md_sysfs_entry md_array_size
=
5045 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
5049 consistency_policy_show(struct mddev
*mddev
, char *page
)
5053 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
)) {
5054 ret
= sprintf(page
, "journal\n");
5055 } else if (test_bit(MD_HAS_PPL
, &mddev
->flags
)) {
5056 ret
= sprintf(page
, "ppl\n");
5057 } else if (mddev
->bitmap
) {
5058 ret
= sprintf(page
, "bitmap\n");
5059 } else if (mddev
->pers
) {
5060 if (mddev
->pers
->sync_request
)
5061 ret
= sprintf(page
, "resync\n");
5063 ret
= sprintf(page
, "none\n");
5065 ret
= sprintf(page
, "unknown\n");
5072 consistency_policy_store(struct mddev
*mddev
, const char *buf
, size_t len
)
5077 if (mddev
->pers
->change_consistency_policy
)
5078 err
= mddev
->pers
->change_consistency_policy(mddev
, buf
);
5081 } else if (mddev
->external
&& strncmp(buf
, "ppl", 3) == 0) {
5082 set_bit(MD_HAS_PPL
, &mddev
->flags
);
5087 return err
? err
: len
;
5090 static struct md_sysfs_entry md_consistency_policy
=
5091 __ATTR(consistency_policy
, S_IRUGO
| S_IWUSR
, consistency_policy_show
,
5092 consistency_policy_store
);
5094 static struct attribute
*md_default_attrs
[] = {
5097 &md_raid_disks
.attr
,
5098 &md_chunk_size
.attr
,
5100 &md_resync_start
.attr
,
5102 &md_new_device
.attr
,
5103 &md_safe_delay
.attr
,
5104 &md_array_state
.attr
,
5105 &md_reshape_position
.attr
,
5106 &md_reshape_direction
.attr
,
5107 &md_array_size
.attr
,
5108 &max_corr_read_errors
.attr
,
5109 &md_consistency_policy
.attr
,
5113 static struct attribute
*md_redundancy_attrs
[] = {
5115 &md_last_scan_mode
.attr
,
5116 &md_mismatches
.attr
,
5119 &md_sync_speed
.attr
,
5120 &md_sync_force_parallel
.attr
,
5121 &md_sync_completed
.attr
,
5124 &md_suspend_lo
.attr
,
5125 &md_suspend_hi
.attr
,
5130 static struct attribute_group md_redundancy_group
= {
5132 .attrs
= md_redundancy_attrs
,
5136 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
5138 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
5139 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
5144 spin_lock(&all_mddevs_lock
);
5145 if (list_empty(&mddev
->all_mddevs
)) {
5146 spin_unlock(&all_mddevs_lock
);
5150 spin_unlock(&all_mddevs_lock
);
5152 rv
= entry
->show(mddev
, page
);
5158 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
5159 const char *page
, size_t length
)
5161 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
5162 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
5167 if (!capable(CAP_SYS_ADMIN
))
5169 spin_lock(&all_mddevs_lock
);
5170 if (list_empty(&mddev
->all_mddevs
)) {
5171 spin_unlock(&all_mddevs_lock
);
5175 spin_unlock(&all_mddevs_lock
);
5176 rv
= entry
->store(mddev
, page
, length
);
5181 static void md_free(struct kobject
*ko
)
5183 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
5185 if (mddev
->sysfs_state
)
5186 sysfs_put(mddev
->sysfs_state
);
5189 blk_cleanup_queue(mddev
->queue
);
5190 if (mddev
->gendisk
) {
5191 del_gendisk(mddev
->gendisk
);
5192 put_disk(mddev
->gendisk
);
5194 percpu_ref_exit(&mddev
->writes_pending
);
5199 static const struct sysfs_ops md_sysfs_ops
= {
5200 .show
= md_attr_show
,
5201 .store
= md_attr_store
,
5203 static struct kobj_type md_ktype
= {
5205 .sysfs_ops
= &md_sysfs_ops
,
5206 .default_attrs
= md_default_attrs
,
5211 static void mddev_delayed_delete(struct work_struct
*ws
)
5213 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
5215 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
5216 kobject_del(&mddev
->kobj
);
5217 kobject_put(&mddev
->kobj
);
5220 static void no_op(struct percpu_ref
*r
) {}
5222 int mddev_init_writes_pending(struct mddev
*mddev
)
5224 if (mddev
->writes_pending
.percpu_count_ptr
)
5226 if (percpu_ref_init(&mddev
->writes_pending
, no_op
, 0, GFP_KERNEL
) < 0)
5228 /* We want to start with the refcount at zero */
5229 percpu_ref_put(&mddev
->writes_pending
);
5232 EXPORT_SYMBOL_GPL(mddev_init_writes_pending
);
5234 static int md_alloc(dev_t dev
, char *name
)
5237 * If dev is zero, name is the name of a device to allocate with
5238 * an arbitrary minor number. It will be "md_???"
5239 * If dev is non-zero it must be a device number with a MAJOR of
5240 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5241 * the device is being created by opening a node in /dev.
5242 * If "name" is not NULL, the device is being created by
5243 * writing to /sys/module/md_mod/parameters/new_array.
5245 static DEFINE_MUTEX(disks_mutex
);
5246 struct mddev
*mddev
= mddev_find(dev
);
5247 struct gendisk
*disk
;
5256 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
5257 shift
= partitioned
? MdpMinorShift
: 0;
5258 unit
= MINOR(mddev
->unit
) >> shift
;
5260 /* wait for any previous instance of this device to be
5261 * completely removed (mddev_delayed_delete).
5263 flush_workqueue(md_misc_wq
);
5265 mutex_lock(&disks_mutex
);
5271 /* Need to ensure that 'name' is not a duplicate.
5273 struct mddev
*mddev2
;
5274 spin_lock(&all_mddevs_lock
);
5276 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
5277 if (mddev2
->gendisk
&&
5278 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
5279 spin_unlock(&all_mddevs_lock
);
5282 spin_unlock(&all_mddevs_lock
);
5286 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5288 mddev
->hold_active
= UNTIL_STOP
;
5291 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
5294 mddev
->queue
->queuedata
= mddev
;
5296 blk_queue_make_request(mddev
->queue
, md_make_request
);
5297 blk_set_stacking_limits(&mddev
->queue
->limits
);
5299 disk
= alloc_disk(1 << shift
);
5301 blk_cleanup_queue(mddev
->queue
);
5302 mddev
->queue
= NULL
;
5305 disk
->major
= MAJOR(mddev
->unit
);
5306 disk
->first_minor
= unit
<< shift
;
5308 strcpy(disk
->disk_name
, name
);
5309 else if (partitioned
)
5310 sprintf(disk
->disk_name
, "md_d%d", unit
);
5312 sprintf(disk
->disk_name
, "md%d", unit
);
5313 disk
->fops
= &md_fops
;
5314 disk
->private_data
= mddev
;
5315 disk
->queue
= mddev
->queue
;
5316 blk_queue_write_cache(mddev
->queue
, true, true);
5317 /* Allow extended partitions. This makes the
5318 * 'mdp' device redundant, but we can't really
5321 disk
->flags
|= GENHD_FL_EXT_DEVT
;
5322 mddev
->gendisk
= disk
;
5323 /* As soon as we call add_disk(), another thread could get
5324 * through to md_open, so make sure it doesn't get too far
5326 mutex_lock(&mddev
->open_mutex
);
5329 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
5330 &disk_to_dev(disk
)->kobj
, "%s", "md");
5332 /* This isn't possible, but as kobject_init_and_add is marked
5333 * __must_check, we must do something with the result
5335 pr_debug("md: cannot register %s/md - name in use\n",
5339 if (mddev
->kobj
.sd
&&
5340 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
5341 pr_debug("pointless warning\n");
5342 mutex_unlock(&mddev
->open_mutex
);
5344 mutex_unlock(&disks_mutex
);
5345 if (!error
&& mddev
->kobj
.sd
) {
5346 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
5347 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
5353 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
5356 md_alloc(dev
, NULL
);
5360 static int add_named_array(const char *val
, struct kernel_param
*kp
)
5363 * val must be "md_*" or "mdNNN".
5364 * For "md_*" we allocate an array with a large free minor number, and
5365 * set the name to val. val must not already be an active name.
5366 * For "mdNNN" we allocate an array with the minor number NNN
5367 * which must not already be in use.
5369 int len
= strlen(val
);
5370 char buf
[DISK_NAME_LEN
];
5371 unsigned long devnum
;
5373 while (len
&& val
[len
-1] == '\n')
5375 if (len
>= DISK_NAME_LEN
)
5377 strlcpy(buf
, val
, len
+1);
5378 if (strncmp(buf
, "md_", 3) == 0)
5379 return md_alloc(0, buf
);
5380 if (strncmp(buf
, "md", 2) == 0 &&
5382 kstrtoul(buf
+2, 10, &devnum
) == 0 &&
5383 devnum
<= MINORMASK
)
5384 return md_alloc(MKDEV(MD_MAJOR
, devnum
), NULL
);
5389 static void md_safemode_timeout(unsigned long data
)
5391 struct mddev
*mddev
= (struct mddev
*) data
;
5393 mddev
->safemode
= 1;
5394 if (mddev
->external
)
5395 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5397 md_wakeup_thread(mddev
->thread
);
5400 static int start_dirty_degraded
;
5402 int md_run(struct mddev
*mddev
)
5405 struct md_rdev
*rdev
;
5406 struct md_personality
*pers
;
5408 if (list_empty(&mddev
->disks
))
5409 /* cannot run an array with no devices.. */
5414 /* Cannot run until previous stop completes properly */
5415 if (mddev
->sysfs_active
)
5419 * Analyze all RAID superblock(s)
5421 if (!mddev
->raid_disks
) {
5422 if (!mddev
->persistent
)
5427 if (mddev
->level
!= LEVEL_NONE
)
5428 request_module("md-level-%d", mddev
->level
);
5429 else if (mddev
->clevel
[0])
5430 request_module("md-%s", mddev
->clevel
);
5433 * Drop all container device buffers, from now on
5434 * the only valid external interface is through the md
5437 rdev_for_each(rdev
, mddev
) {
5438 if (test_bit(Faulty
, &rdev
->flags
))
5440 sync_blockdev(rdev
->bdev
);
5441 invalidate_bdev(rdev
->bdev
);
5442 if (mddev
->ro
!= 1 &&
5443 (bdev_read_only(rdev
->bdev
) ||
5444 bdev_read_only(rdev
->meta_bdev
))) {
5447 set_disk_ro(mddev
->gendisk
, 1);
5450 /* perform some consistency tests on the device.
5451 * We don't want the data to overlap the metadata,
5452 * Internal Bitmap issues have been handled elsewhere.
5454 if (rdev
->meta_bdev
) {
5455 /* Nothing to check */;
5456 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5457 if (mddev
->dev_sectors
&&
5458 rdev
->data_offset
+ mddev
->dev_sectors
5460 pr_warn("md: %s: data overlaps metadata\n",
5465 if (rdev
->sb_start
+ rdev
->sb_size
/512
5466 > rdev
->data_offset
) {
5467 pr_warn("md: %s: metadata overlaps data\n",
5472 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5475 if (mddev
->bio_set
== NULL
) {
5476 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
5477 if (!mddev
->bio_set
)
5480 if (mddev
->sync_set
== NULL
) {
5481 mddev
->sync_set
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
5482 if (!mddev
->sync_set
)
5486 spin_lock(&pers_lock
);
5487 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5488 if (!pers
|| !try_module_get(pers
->owner
)) {
5489 spin_unlock(&pers_lock
);
5490 if (mddev
->level
!= LEVEL_NONE
)
5491 pr_warn("md: personality for level %d is not loaded!\n",
5494 pr_warn("md: personality for level %s is not loaded!\n",
5498 spin_unlock(&pers_lock
);
5499 if (mddev
->level
!= pers
->level
) {
5500 mddev
->level
= pers
->level
;
5501 mddev
->new_level
= pers
->level
;
5503 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5505 if (mddev
->reshape_position
!= MaxSector
&&
5506 pers
->start_reshape
== NULL
) {
5507 /* This personality cannot handle reshaping... */
5508 module_put(pers
->owner
);
5512 if (pers
->sync_request
) {
5513 /* Warn if this is a potentially silly
5516 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5517 struct md_rdev
*rdev2
;
5520 rdev_for_each(rdev
, mddev
)
5521 rdev_for_each(rdev2
, mddev
) {
5523 rdev
->bdev
->bd_contains
==
5524 rdev2
->bdev
->bd_contains
) {
5525 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5527 bdevname(rdev
->bdev
,b
),
5528 bdevname(rdev2
->bdev
,b2
));
5534 pr_warn("True protection against single-disk failure might be compromised.\n");
5537 mddev
->recovery
= 0;
5538 /* may be over-ridden by personality */
5539 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5541 mddev
->ok_start_degraded
= start_dirty_degraded
;
5543 if (start_readonly
&& mddev
->ro
== 0)
5544 mddev
->ro
= 2; /* read-only, but switch on first write */
5547 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5548 * up mddev->thread. It is important to initialize critical
5549 * resources for mddev->thread BEFORE calling pers->run().
5551 err
= pers
->run(mddev
);
5553 pr_warn("md: pers->run() failed ...\n");
5554 else if (pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5555 WARN_ONCE(!mddev
->external_size
,
5556 "%s: default size too small, but 'external_size' not in effect?\n",
5558 pr_warn("md: invalid array_size %llu > default size %llu\n",
5559 (unsigned long long)mddev
->array_sectors
/ 2,
5560 (unsigned long long)pers
->size(mddev
, 0, 0) / 2);
5563 if (err
== 0 && pers
->sync_request
&&
5564 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5565 struct bitmap
*bitmap
;
5567 bitmap
= bitmap_create(mddev
, -1);
5568 if (IS_ERR(bitmap
)) {
5569 err
= PTR_ERR(bitmap
);
5570 pr_warn("%s: failed to create bitmap (%d)\n",
5571 mdname(mddev
), err
);
5573 mddev
->bitmap
= bitmap
;
5577 mddev_detach(mddev
);
5579 pers
->free(mddev
, mddev
->private);
5580 mddev
->private = NULL
;
5581 module_put(pers
->owner
);
5582 bitmap_destroy(mddev
);
5588 rdev_for_each(rdev
, mddev
) {
5589 if (rdev
->raid_disk
>= 0 &&
5590 !blk_queue_nonrot(bdev_get_queue(rdev
->bdev
))) {
5595 if (mddev
->degraded
)
5598 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, mddev
->queue
);
5600 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT
, mddev
->queue
);
5601 mddev
->queue
->backing_dev_info
->congested_data
= mddev
;
5602 mddev
->queue
->backing_dev_info
->congested_fn
= md_congested
;
5604 if (pers
->sync_request
) {
5605 if (mddev
->kobj
.sd
&&
5606 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5607 pr_warn("md: cannot register extra attributes for %s\n",
5609 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5610 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5613 atomic_set(&mddev
->max_corr_read_errors
,
5614 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5615 mddev
->safemode
= 0;
5616 if (mddev_is_clustered(mddev
))
5617 mddev
->safemode_delay
= 0;
5619 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5622 spin_lock(&mddev
->lock
);
5624 spin_unlock(&mddev
->lock
);
5625 rdev_for_each(rdev
, mddev
)
5626 if (rdev
->raid_disk
>= 0)
5627 if (sysfs_link_rdev(mddev
, rdev
))
5628 /* failure here is OK */;
5630 if (mddev
->degraded
&& !mddev
->ro
)
5631 /* This ensures that recovering status is reported immediately
5632 * via sysfs - until a lack of spares is confirmed.
5634 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5635 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5637 if (mddev
->sb_flags
)
5638 md_update_sb(mddev
, 0);
5640 md_new_event(mddev
);
5641 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5642 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5643 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5646 EXPORT_SYMBOL_GPL(md_run
);
5648 static int do_md_run(struct mddev
*mddev
)
5652 err
= md_run(mddev
);
5655 err
= bitmap_load(mddev
);
5657 bitmap_destroy(mddev
);
5661 if (mddev_is_clustered(mddev
))
5662 md_allow_write(mddev
);
5664 md_wakeup_thread(mddev
->thread
);
5665 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5667 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5668 revalidate_disk(mddev
->gendisk
);
5670 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5675 static int restart_array(struct mddev
*mddev
)
5677 struct gendisk
*disk
= mddev
->gendisk
;
5678 struct md_rdev
*rdev
;
5679 bool has_journal
= false;
5680 bool has_readonly
= false;
5682 /* Complain if it has no devices */
5683 if (list_empty(&mddev
->disks
))
5691 rdev_for_each_rcu(rdev
, mddev
) {
5692 if (test_bit(Journal
, &rdev
->flags
) &&
5693 !test_bit(Faulty
, &rdev
->flags
))
5695 if (bdev_read_only(rdev
->bdev
))
5696 has_readonly
= true;
5699 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
) && !has_journal
)
5700 /* Don't restart rw with journal missing/faulty */
5705 mddev
->safemode
= 0;
5707 set_disk_ro(disk
, 0);
5708 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev
));
5709 /* Kick recovery or resync if necessary */
5710 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5711 md_wakeup_thread(mddev
->thread
);
5712 md_wakeup_thread(mddev
->sync_thread
);
5713 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5717 static void md_clean(struct mddev
*mddev
)
5719 mddev
->array_sectors
= 0;
5720 mddev
->external_size
= 0;
5721 mddev
->dev_sectors
= 0;
5722 mddev
->raid_disks
= 0;
5723 mddev
->recovery_cp
= 0;
5724 mddev
->resync_min
= 0;
5725 mddev
->resync_max
= MaxSector
;
5726 mddev
->reshape_position
= MaxSector
;
5727 mddev
->external
= 0;
5728 mddev
->persistent
= 0;
5729 mddev
->level
= LEVEL_NONE
;
5730 mddev
->clevel
[0] = 0;
5732 mddev
->sb_flags
= 0;
5734 mddev
->metadata_type
[0] = 0;
5735 mddev
->chunk_sectors
= 0;
5736 mddev
->ctime
= mddev
->utime
= 0;
5738 mddev
->max_disks
= 0;
5740 mddev
->can_decrease_events
= 0;
5741 mddev
->delta_disks
= 0;
5742 mddev
->reshape_backwards
= 0;
5743 mddev
->new_level
= LEVEL_NONE
;
5744 mddev
->new_layout
= 0;
5745 mddev
->new_chunk_sectors
= 0;
5746 mddev
->curr_resync
= 0;
5747 atomic64_set(&mddev
->resync_mismatches
, 0);
5748 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5749 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5750 mddev
->recovery
= 0;
5753 mddev
->degraded
= 0;
5754 mddev
->safemode
= 0;
5755 mddev
->private = NULL
;
5756 mddev
->cluster_info
= NULL
;
5757 mddev
->bitmap_info
.offset
= 0;
5758 mddev
->bitmap_info
.default_offset
= 0;
5759 mddev
->bitmap_info
.default_space
= 0;
5760 mddev
->bitmap_info
.chunksize
= 0;
5761 mddev
->bitmap_info
.daemon_sleep
= 0;
5762 mddev
->bitmap_info
.max_write_behind
= 0;
5763 mddev
->bitmap_info
.nodes
= 0;
5766 static void __md_stop_writes(struct mddev
*mddev
)
5768 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5769 flush_workqueue(md_misc_wq
);
5770 if (mddev
->sync_thread
) {
5771 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5772 md_reap_sync_thread(mddev
);
5775 del_timer_sync(&mddev
->safemode_timer
);
5777 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5778 mddev
->pers
->quiesce(mddev
, 1);
5779 mddev
->pers
->quiesce(mddev
, 0);
5781 bitmap_flush(mddev
);
5783 if (mddev
->ro
== 0 &&
5784 ((!mddev
->in_sync
&& !mddev_is_clustered(mddev
)) ||
5786 /* mark array as shutdown cleanly */
5787 if (!mddev_is_clustered(mddev
))
5789 md_update_sb(mddev
, 1);
5793 void md_stop_writes(struct mddev
*mddev
)
5795 mddev_lock_nointr(mddev
);
5796 __md_stop_writes(mddev
);
5797 mddev_unlock(mddev
);
5799 EXPORT_SYMBOL_GPL(md_stop_writes
);
5801 static void mddev_detach(struct mddev
*mddev
)
5803 bitmap_wait_behind_writes(mddev
);
5804 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5805 mddev
->pers
->quiesce(mddev
, 1);
5806 mddev
->pers
->quiesce(mddev
, 0);
5808 md_unregister_thread(&mddev
->thread
);
5810 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
5813 static void __md_stop(struct mddev
*mddev
)
5815 struct md_personality
*pers
= mddev
->pers
;
5816 bitmap_destroy(mddev
);
5817 mddev_detach(mddev
);
5818 /* Ensure ->event_work is done */
5819 flush_workqueue(md_misc_wq
);
5820 spin_lock(&mddev
->lock
);
5822 spin_unlock(&mddev
->lock
);
5823 pers
->free(mddev
, mddev
->private);
5824 mddev
->private = NULL
;
5825 if (pers
->sync_request
&& mddev
->to_remove
== NULL
)
5826 mddev
->to_remove
= &md_redundancy_group
;
5827 module_put(pers
->owner
);
5828 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5831 void md_stop(struct mddev
*mddev
)
5833 /* stop the array and free an attached data structures.
5834 * This is called from dm-raid
5838 bioset_free(mddev
->bio_set
);
5841 EXPORT_SYMBOL_GPL(md_stop
);
5843 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5848 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5850 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5851 md_wakeup_thread(mddev
->thread
);
5853 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5854 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5855 if (mddev
->sync_thread
)
5856 /* Thread might be blocked waiting for metadata update
5857 * which will now never happen */
5858 wake_up_process(mddev
->sync_thread
->tsk
);
5860 if (mddev
->external
&& test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
))
5862 mddev_unlock(mddev
);
5863 wait_event(resync_wait
, !test_bit(MD_RECOVERY_RUNNING
,
5865 wait_event(mddev
->sb_wait
,
5866 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
5867 mddev_lock_nointr(mddev
);
5869 mutex_lock(&mddev
->open_mutex
);
5870 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5871 mddev
->sync_thread
||
5872 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
)) {
5873 pr_warn("md: %s still in use.\n",mdname(mddev
));
5875 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5876 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5877 md_wakeup_thread(mddev
->thread
);
5883 __md_stop_writes(mddev
);
5889 set_disk_ro(mddev
->gendisk
, 1);
5890 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5891 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5892 md_wakeup_thread(mddev
->thread
);
5893 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5897 mutex_unlock(&mddev
->open_mutex
);
5902 * 0 - completely stop and dis-assemble array
5903 * 2 - stop but do not disassemble array
5905 static int do_md_stop(struct mddev
*mddev
, int mode
,
5906 struct block_device
*bdev
)
5908 struct gendisk
*disk
= mddev
->gendisk
;
5909 struct md_rdev
*rdev
;
5912 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5914 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5915 md_wakeup_thread(mddev
->thread
);
5917 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5918 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5919 if (mddev
->sync_thread
)
5920 /* Thread might be blocked waiting for metadata update
5921 * which will now never happen */
5922 wake_up_process(mddev
->sync_thread
->tsk
);
5924 mddev_unlock(mddev
);
5925 wait_event(resync_wait
, (mddev
->sync_thread
== NULL
&&
5926 !test_bit(MD_RECOVERY_RUNNING
,
5927 &mddev
->recovery
)));
5928 mddev_lock_nointr(mddev
);
5930 mutex_lock(&mddev
->open_mutex
);
5931 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5932 mddev
->sysfs_active
||
5933 mddev
->sync_thread
||
5934 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
)) {
5935 pr_warn("md: %s still in use.\n",mdname(mddev
));
5936 mutex_unlock(&mddev
->open_mutex
);
5938 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5939 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5940 md_wakeup_thread(mddev
->thread
);
5946 set_disk_ro(disk
, 0);
5948 __md_stop_writes(mddev
);
5950 mddev
->queue
->backing_dev_info
->congested_fn
= NULL
;
5952 /* tell userspace to handle 'inactive' */
5953 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5955 rdev_for_each(rdev
, mddev
)
5956 if (rdev
->raid_disk
>= 0)
5957 sysfs_unlink_rdev(mddev
, rdev
);
5959 set_capacity(disk
, 0);
5960 mutex_unlock(&mddev
->open_mutex
);
5962 revalidate_disk(disk
);
5967 mutex_unlock(&mddev
->open_mutex
);
5969 * Free resources if final stop
5972 pr_info("md: %s stopped.\n", mdname(mddev
));
5974 if (mddev
->bitmap_info
.file
) {
5975 struct file
*f
= mddev
->bitmap_info
.file
;
5976 spin_lock(&mddev
->lock
);
5977 mddev
->bitmap_info
.file
= NULL
;
5978 spin_unlock(&mddev
->lock
);
5981 mddev
->bitmap_info
.offset
= 0;
5983 export_array(mddev
);
5986 if (mddev
->hold_active
== UNTIL_STOP
)
5987 mddev
->hold_active
= 0;
5989 md_new_event(mddev
);
5990 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5995 static void autorun_array(struct mddev
*mddev
)
5997 struct md_rdev
*rdev
;
6000 if (list_empty(&mddev
->disks
))
6003 pr_info("md: running: ");
6005 rdev_for_each(rdev
, mddev
) {
6006 char b
[BDEVNAME_SIZE
];
6007 pr_cont("<%s>", bdevname(rdev
->bdev
,b
));
6011 err
= do_md_run(mddev
);
6013 pr_warn("md: do_md_run() returned %d\n", err
);
6014 do_md_stop(mddev
, 0, NULL
);
6019 * lets try to run arrays based on all disks that have arrived
6020 * until now. (those are in pending_raid_disks)
6022 * the method: pick the first pending disk, collect all disks with
6023 * the same UUID, remove all from the pending list and put them into
6024 * the 'same_array' list. Then order this list based on superblock
6025 * update time (freshest comes first), kick out 'old' disks and
6026 * compare superblocks. If everything's fine then run it.
6028 * If "unit" is allocated, then bump its reference count
6030 static void autorun_devices(int part
)
6032 struct md_rdev
*rdev0
, *rdev
, *tmp
;
6033 struct mddev
*mddev
;
6034 char b
[BDEVNAME_SIZE
];
6036 pr_info("md: autorun ...\n");
6037 while (!list_empty(&pending_raid_disks
)) {
6040 LIST_HEAD(candidates
);
6041 rdev0
= list_entry(pending_raid_disks
.next
,
6042 struct md_rdev
, same_set
);
6044 pr_debug("md: considering %s ...\n", bdevname(rdev0
->bdev
,b
));
6045 INIT_LIST_HEAD(&candidates
);
6046 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
6047 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
6048 pr_debug("md: adding %s ...\n",
6049 bdevname(rdev
->bdev
,b
));
6050 list_move(&rdev
->same_set
, &candidates
);
6053 * now we have a set of devices, with all of them having
6054 * mostly sane superblocks. It's time to allocate the
6058 dev
= MKDEV(mdp_major
,
6059 rdev0
->preferred_minor
<< MdpMinorShift
);
6060 unit
= MINOR(dev
) >> MdpMinorShift
;
6062 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
6065 if (rdev0
->preferred_minor
!= unit
) {
6066 pr_warn("md: unit number in %s is bad: %d\n",
6067 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
6071 md_probe(dev
, NULL
, NULL
);
6072 mddev
= mddev_find(dev
);
6073 if (!mddev
|| !mddev
->gendisk
) {
6078 if (mddev_lock(mddev
))
6079 pr_warn("md: %s locked, cannot run\n", mdname(mddev
));
6080 else if (mddev
->raid_disks
|| mddev
->major_version
6081 || !list_empty(&mddev
->disks
)) {
6082 pr_warn("md: %s already running, cannot run %s\n",
6083 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
6084 mddev_unlock(mddev
);
6086 pr_debug("md: created %s\n", mdname(mddev
));
6087 mddev
->persistent
= 1;
6088 rdev_for_each_list(rdev
, tmp
, &candidates
) {
6089 list_del_init(&rdev
->same_set
);
6090 if (bind_rdev_to_array(rdev
, mddev
))
6093 autorun_array(mddev
);
6094 mddev_unlock(mddev
);
6096 /* on success, candidates will be empty, on error
6099 rdev_for_each_list(rdev
, tmp
, &candidates
) {
6100 list_del_init(&rdev
->same_set
);
6105 pr_info("md: ... autorun DONE.\n");
6107 #endif /* !MODULE */
6109 static int get_version(void __user
*arg
)
6113 ver
.major
= MD_MAJOR_VERSION
;
6114 ver
.minor
= MD_MINOR_VERSION
;
6115 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
6117 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
6123 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
6125 mdu_array_info_t info
;
6126 int nr
,working
,insync
,failed
,spare
;
6127 struct md_rdev
*rdev
;
6129 nr
= working
= insync
= failed
= spare
= 0;
6131 rdev_for_each_rcu(rdev
, mddev
) {
6133 if (test_bit(Faulty
, &rdev
->flags
))
6137 if (test_bit(In_sync
, &rdev
->flags
))
6139 else if (test_bit(Journal
, &rdev
->flags
))
6140 /* TODO: add journal count to md_u.h */
6148 info
.major_version
= mddev
->major_version
;
6149 info
.minor_version
= mddev
->minor_version
;
6150 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
6151 info
.ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
6152 info
.level
= mddev
->level
;
6153 info
.size
= mddev
->dev_sectors
/ 2;
6154 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
6157 info
.raid_disks
= mddev
->raid_disks
;
6158 info
.md_minor
= mddev
->md_minor
;
6159 info
.not_persistent
= !mddev
->persistent
;
6161 info
.utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
6164 info
.state
= (1<<MD_SB_CLEAN
);
6165 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6166 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
6167 if (mddev_is_clustered(mddev
))
6168 info
.state
|= (1<<MD_SB_CLUSTERED
);
6169 info
.active_disks
= insync
;
6170 info
.working_disks
= working
;
6171 info
.failed_disks
= failed
;
6172 info
.spare_disks
= spare
;
6174 info
.layout
= mddev
->layout
;
6175 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
6177 if (copy_to_user(arg
, &info
, sizeof(info
)))
6183 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
6185 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
6189 file
= kzalloc(sizeof(*file
), GFP_NOIO
);
6194 spin_lock(&mddev
->lock
);
6195 /* bitmap enabled */
6196 if (mddev
->bitmap_info
.file
) {
6197 ptr
= file_path(mddev
->bitmap_info
.file
, file
->pathname
,
6198 sizeof(file
->pathname
));
6202 memmove(file
->pathname
, ptr
,
6203 sizeof(file
->pathname
)-(ptr
-file
->pathname
));
6205 spin_unlock(&mddev
->lock
);
6208 copy_to_user(arg
, file
, sizeof(*file
)))
6215 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
6217 mdu_disk_info_t info
;
6218 struct md_rdev
*rdev
;
6220 if (copy_from_user(&info
, arg
, sizeof(info
)))
6224 rdev
= md_find_rdev_nr_rcu(mddev
, info
.number
);
6226 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
6227 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
6228 info
.raid_disk
= rdev
->raid_disk
;
6230 if (test_bit(Faulty
, &rdev
->flags
))
6231 info
.state
|= (1<<MD_DISK_FAULTY
);
6232 else if (test_bit(In_sync
, &rdev
->flags
)) {
6233 info
.state
|= (1<<MD_DISK_ACTIVE
);
6234 info
.state
|= (1<<MD_DISK_SYNC
);
6236 if (test_bit(Journal
, &rdev
->flags
))
6237 info
.state
|= (1<<MD_DISK_JOURNAL
);
6238 if (test_bit(WriteMostly
, &rdev
->flags
))
6239 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
6240 if (test_bit(FailFast
, &rdev
->flags
))
6241 info
.state
|= (1<<MD_DISK_FAILFAST
);
6243 info
.major
= info
.minor
= 0;
6244 info
.raid_disk
= -1;
6245 info
.state
= (1<<MD_DISK_REMOVED
);
6249 if (copy_to_user(arg
, &info
, sizeof(info
)))
6255 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
6257 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
6258 struct md_rdev
*rdev
;
6259 dev_t dev
= MKDEV(info
->major
,info
->minor
);
6261 if (mddev_is_clustered(mddev
) &&
6262 !(info
->state
& ((1 << MD_DISK_CLUSTER_ADD
) | (1 << MD_DISK_CANDIDATE
)))) {
6263 pr_warn("%s: Cannot add to clustered mddev.\n",
6268 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
6271 if (!mddev
->raid_disks
) {
6273 /* expecting a device which has a superblock */
6274 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
6276 pr_warn("md: md_import_device returned %ld\n",
6278 return PTR_ERR(rdev
);
6280 if (!list_empty(&mddev
->disks
)) {
6281 struct md_rdev
*rdev0
6282 = list_entry(mddev
->disks
.next
,
6283 struct md_rdev
, same_set
);
6284 err
= super_types
[mddev
->major_version
]
6285 .load_super(rdev
, rdev0
, mddev
->minor_version
);
6287 pr_warn("md: %s has different UUID to %s\n",
6288 bdevname(rdev
->bdev
,b
),
6289 bdevname(rdev0
->bdev
,b2
));
6294 err
= bind_rdev_to_array(rdev
, mddev
);
6301 * add_new_disk can be used once the array is assembled
6302 * to add "hot spares". They must already have a superblock
6307 if (!mddev
->pers
->hot_add_disk
) {
6308 pr_warn("%s: personality does not support diskops!\n",
6312 if (mddev
->persistent
)
6313 rdev
= md_import_device(dev
, mddev
->major_version
,
6314 mddev
->minor_version
);
6316 rdev
= md_import_device(dev
, -1, -1);
6318 pr_warn("md: md_import_device returned %ld\n",
6320 return PTR_ERR(rdev
);
6322 /* set saved_raid_disk if appropriate */
6323 if (!mddev
->persistent
) {
6324 if (info
->state
& (1<<MD_DISK_SYNC
) &&
6325 info
->raid_disk
< mddev
->raid_disks
) {
6326 rdev
->raid_disk
= info
->raid_disk
;
6327 set_bit(In_sync
, &rdev
->flags
);
6328 clear_bit(Bitmap_sync
, &rdev
->flags
);
6330 rdev
->raid_disk
= -1;
6331 rdev
->saved_raid_disk
= rdev
->raid_disk
;
6333 super_types
[mddev
->major_version
].
6334 validate_super(mddev
, rdev
);
6335 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
6336 rdev
->raid_disk
!= info
->raid_disk
) {
6337 /* This was a hot-add request, but events doesn't
6338 * match, so reject it.
6344 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
6345 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6346 set_bit(WriteMostly
, &rdev
->flags
);
6348 clear_bit(WriteMostly
, &rdev
->flags
);
6349 if (info
->state
& (1<<MD_DISK_FAILFAST
))
6350 set_bit(FailFast
, &rdev
->flags
);
6352 clear_bit(FailFast
, &rdev
->flags
);
6354 if (info
->state
& (1<<MD_DISK_JOURNAL
)) {
6355 struct md_rdev
*rdev2
;
6356 bool has_journal
= false;
6358 /* make sure no existing journal disk */
6359 rdev_for_each(rdev2
, mddev
) {
6360 if (test_bit(Journal
, &rdev2
->flags
)) {
6369 set_bit(Journal
, &rdev
->flags
);
6372 * check whether the device shows up in other nodes
6374 if (mddev_is_clustered(mddev
)) {
6375 if (info
->state
& (1 << MD_DISK_CANDIDATE
))
6376 set_bit(Candidate
, &rdev
->flags
);
6377 else if (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)) {
6378 /* --add initiated by this node */
6379 err
= md_cluster_ops
->add_new_disk(mddev
, rdev
);
6387 rdev
->raid_disk
= -1;
6388 err
= bind_rdev_to_array(rdev
, mddev
);
6393 if (mddev_is_clustered(mddev
)) {
6394 if (info
->state
& (1 << MD_DISK_CANDIDATE
)) {
6396 err
= md_cluster_ops
->new_disk_ack(mddev
,
6399 md_kick_rdev_from_array(rdev
);
6403 md_cluster_ops
->add_new_disk_cancel(mddev
);
6405 err
= add_bound_rdev(rdev
);
6409 err
= add_bound_rdev(rdev
);
6414 /* otherwise, add_new_disk is only allowed
6415 * for major_version==0 superblocks
6417 if (mddev
->major_version
!= 0) {
6418 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev
));
6422 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
6424 rdev
= md_import_device(dev
, -1, 0);
6426 pr_warn("md: error, md_import_device() returned %ld\n",
6428 return PTR_ERR(rdev
);
6430 rdev
->desc_nr
= info
->number
;
6431 if (info
->raid_disk
< mddev
->raid_disks
)
6432 rdev
->raid_disk
= info
->raid_disk
;
6434 rdev
->raid_disk
= -1;
6436 if (rdev
->raid_disk
< mddev
->raid_disks
)
6437 if (info
->state
& (1<<MD_DISK_SYNC
))
6438 set_bit(In_sync
, &rdev
->flags
);
6440 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6441 set_bit(WriteMostly
, &rdev
->flags
);
6442 if (info
->state
& (1<<MD_DISK_FAILFAST
))
6443 set_bit(FailFast
, &rdev
->flags
);
6445 if (!mddev
->persistent
) {
6446 pr_debug("md: nonpersistent superblock ...\n");
6447 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6449 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6450 rdev
->sectors
= rdev
->sb_start
;
6452 err
= bind_rdev_to_array(rdev
, mddev
);
6462 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
6464 char b
[BDEVNAME_SIZE
];
6465 struct md_rdev
*rdev
;
6467 rdev
= find_rdev(mddev
, dev
);
6471 if (rdev
->raid_disk
< 0)
6474 clear_bit(Blocked
, &rdev
->flags
);
6475 remove_and_add_spares(mddev
, rdev
);
6477 if (rdev
->raid_disk
>= 0)
6481 if (mddev_is_clustered(mddev
))
6482 md_cluster_ops
->remove_disk(mddev
, rdev
);
6484 md_kick_rdev_from_array(rdev
);
6485 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6487 md_wakeup_thread(mddev
->thread
);
6489 md_update_sb(mddev
, 1);
6490 md_new_event(mddev
);
6494 pr_debug("md: cannot remove active disk %s from %s ...\n",
6495 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6499 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
6501 char b
[BDEVNAME_SIZE
];
6503 struct md_rdev
*rdev
;
6508 if (mddev
->major_version
!= 0) {
6509 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6513 if (!mddev
->pers
->hot_add_disk
) {
6514 pr_warn("%s: personality does not support diskops!\n",
6519 rdev
= md_import_device(dev
, -1, 0);
6521 pr_warn("md: error, md_import_device() returned %ld\n",
6526 if (mddev
->persistent
)
6527 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6529 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6531 rdev
->sectors
= rdev
->sb_start
;
6533 if (test_bit(Faulty
, &rdev
->flags
)) {
6534 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6535 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6540 clear_bit(In_sync
, &rdev
->flags
);
6542 rdev
->saved_raid_disk
= -1;
6543 err
= bind_rdev_to_array(rdev
, mddev
);
6548 * The rest should better be atomic, we can have disk failures
6549 * noticed in interrupt contexts ...
6552 rdev
->raid_disk
= -1;
6554 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6556 md_update_sb(mddev
, 1);
6558 * Kick recovery, maybe this spare has to be added to the
6559 * array immediately.
6561 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6562 md_wakeup_thread(mddev
->thread
);
6563 md_new_event(mddev
);
6571 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
6576 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
6578 if (mddev
->recovery
|| mddev
->sync_thread
)
6580 /* we should be able to change the bitmap.. */
6584 struct inode
*inode
;
6587 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
)
6588 return -EEXIST
; /* cannot add when bitmap is present */
6592 pr_warn("%s: error: failed to get bitmap file\n",
6597 inode
= f
->f_mapping
->host
;
6598 if (!S_ISREG(inode
->i_mode
)) {
6599 pr_warn("%s: error: bitmap file must be a regular file\n",
6602 } else if (!(f
->f_mode
& FMODE_WRITE
)) {
6603 pr_warn("%s: error: bitmap file must open for write\n",
6606 } else if (atomic_read(&inode
->i_writecount
) != 1) {
6607 pr_warn("%s: error: bitmap file is already in use\n",
6615 mddev
->bitmap_info
.file
= f
;
6616 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6617 } else if (mddev
->bitmap
== NULL
)
6618 return -ENOENT
; /* cannot remove what isn't there */
6621 mddev
->pers
->quiesce(mddev
, 1);
6623 struct bitmap
*bitmap
;
6625 bitmap
= bitmap_create(mddev
, -1);
6626 if (!IS_ERR(bitmap
)) {
6627 mddev
->bitmap
= bitmap
;
6628 err
= bitmap_load(mddev
);
6630 err
= PTR_ERR(bitmap
);
6632 if (fd
< 0 || err
) {
6633 bitmap_destroy(mddev
);
6634 fd
= -1; /* make sure to put the file */
6636 mddev
->pers
->quiesce(mddev
, 0);
6639 struct file
*f
= mddev
->bitmap_info
.file
;
6641 spin_lock(&mddev
->lock
);
6642 mddev
->bitmap_info
.file
= NULL
;
6643 spin_unlock(&mddev
->lock
);
6652 * set_array_info is used two different ways
6653 * The original usage is when creating a new array.
6654 * In this usage, raid_disks is > 0 and it together with
6655 * level, size, not_persistent,layout,chunksize determine the
6656 * shape of the array.
6657 * This will always create an array with a type-0.90.0 superblock.
6658 * The newer usage is when assembling an array.
6659 * In this case raid_disks will be 0, and the major_version field is
6660 * use to determine which style super-blocks are to be found on the devices.
6661 * The minor and patch _version numbers are also kept incase the
6662 * super_block handler wishes to interpret them.
6664 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6667 if (info
->raid_disks
== 0) {
6668 /* just setting version number for superblock loading */
6669 if (info
->major_version
< 0 ||
6670 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6671 super_types
[info
->major_version
].name
== NULL
) {
6672 /* maybe try to auto-load a module? */
6673 pr_warn("md: superblock version %d not known\n",
6674 info
->major_version
);
6677 mddev
->major_version
= info
->major_version
;
6678 mddev
->minor_version
= info
->minor_version
;
6679 mddev
->patch_version
= info
->patch_version
;
6680 mddev
->persistent
= !info
->not_persistent
;
6681 /* ensure mddev_put doesn't delete this now that there
6682 * is some minimal configuration.
6684 mddev
->ctime
= ktime_get_real_seconds();
6687 mddev
->major_version
= MD_MAJOR_VERSION
;
6688 mddev
->minor_version
= MD_MINOR_VERSION
;
6689 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6690 mddev
->ctime
= ktime_get_real_seconds();
6692 mddev
->level
= info
->level
;
6693 mddev
->clevel
[0] = 0;
6694 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6695 mddev
->raid_disks
= info
->raid_disks
;
6696 /* don't set md_minor, it is determined by which /dev/md* was
6699 if (info
->state
& (1<<MD_SB_CLEAN
))
6700 mddev
->recovery_cp
= MaxSector
;
6702 mddev
->recovery_cp
= 0;
6703 mddev
->persistent
= ! info
->not_persistent
;
6704 mddev
->external
= 0;
6706 mddev
->layout
= info
->layout
;
6707 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6709 if (mddev
->persistent
) {
6710 mddev
->max_disks
= MD_SB_DISKS
;
6712 mddev
->sb_flags
= 0;
6714 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
6716 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6717 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6718 mddev
->bitmap_info
.offset
= 0;
6720 mddev
->reshape_position
= MaxSector
;
6723 * Generate a 128 bit UUID
6725 get_random_bytes(mddev
->uuid
, 16);
6727 mddev
->new_level
= mddev
->level
;
6728 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6729 mddev
->new_layout
= mddev
->layout
;
6730 mddev
->delta_disks
= 0;
6731 mddev
->reshape_backwards
= 0;
6736 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6738 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6740 if (mddev
->external_size
)
6743 mddev
->array_sectors
= array_sectors
;
6745 EXPORT_SYMBOL(md_set_array_sectors
);
6747 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6749 struct md_rdev
*rdev
;
6751 int fit
= (num_sectors
== 0);
6752 sector_t old_dev_sectors
= mddev
->dev_sectors
;
6754 if (mddev
->pers
->resize
== NULL
)
6756 /* The "num_sectors" is the number of sectors of each device that
6757 * is used. This can only make sense for arrays with redundancy.
6758 * linear and raid0 always use whatever space is available. We can only
6759 * consider changing this number if no resync or reconstruction is
6760 * happening, and if the new size is acceptable. It must fit before the
6761 * sb_start or, if that is <data_offset, it must fit before the size
6762 * of each device. If num_sectors is zero, we find the largest size
6765 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6771 rdev_for_each(rdev
, mddev
) {
6772 sector_t avail
= rdev
->sectors
;
6774 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6775 num_sectors
= avail
;
6776 if (avail
< num_sectors
)
6779 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6781 if (mddev_is_clustered(mddev
))
6782 md_cluster_ops
->update_size(mddev
, old_dev_sectors
);
6783 else if (mddev
->queue
) {
6784 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
6785 revalidate_disk(mddev
->gendisk
);
6791 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6794 struct md_rdev
*rdev
;
6795 /* change the number of raid disks */
6796 if (mddev
->pers
->check_reshape
== NULL
)
6800 if (raid_disks
<= 0 ||
6801 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6803 if (mddev
->sync_thread
||
6804 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6805 mddev
->reshape_position
!= MaxSector
)
6808 rdev_for_each(rdev
, mddev
) {
6809 if (mddev
->raid_disks
< raid_disks
&&
6810 rdev
->data_offset
< rdev
->new_data_offset
)
6812 if (mddev
->raid_disks
> raid_disks
&&
6813 rdev
->data_offset
> rdev
->new_data_offset
)
6817 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6818 if (mddev
->delta_disks
< 0)
6819 mddev
->reshape_backwards
= 1;
6820 else if (mddev
->delta_disks
> 0)
6821 mddev
->reshape_backwards
= 0;
6823 rv
= mddev
->pers
->check_reshape(mddev
);
6825 mddev
->delta_disks
= 0;
6826 mddev
->reshape_backwards
= 0;
6832 * update_array_info is used to change the configuration of an
6834 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6835 * fields in the info are checked against the array.
6836 * Any differences that cannot be handled will cause an error.
6837 * Normally, only one change can be managed at a time.
6839 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6845 /* calculate expected state,ignoring low bits */
6846 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6847 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6849 if (mddev
->major_version
!= info
->major_version
||
6850 mddev
->minor_version
!= info
->minor_version
||
6851 /* mddev->patch_version != info->patch_version || */
6852 mddev
->ctime
!= info
->ctime
||
6853 mddev
->level
!= info
->level
||
6854 /* mddev->layout != info->layout || */
6855 mddev
->persistent
!= !info
->not_persistent
||
6856 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6857 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6858 ((state
^info
->state
) & 0xfffffe00)
6861 /* Check there is only one change */
6862 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6864 if (mddev
->raid_disks
!= info
->raid_disks
)
6866 if (mddev
->layout
!= info
->layout
)
6868 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6875 if (mddev
->layout
!= info
->layout
) {
6877 * we don't need to do anything at the md level, the
6878 * personality will take care of it all.
6880 if (mddev
->pers
->check_reshape
== NULL
)
6883 mddev
->new_layout
= info
->layout
;
6884 rv
= mddev
->pers
->check_reshape(mddev
);
6886 mddev
->new_layout
= mddev
->layout
;
6890 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6891 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6893 if (mddev
->raid_disks
!= info
->raid_disks
)
6894 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6896 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6897 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
) {
6901 if (mddev
->recovery
|| mddev
->sync_thread
) {
6905 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6906 struct bitmap
*bitmap
;
6907 /* add the bitmap */
6908 if (mddev
->bitmap
) {
6912 if (mddev
->bitmap_info
.default_offset
== 0) {
6916 mddev
->bitmap_info
.offset
=
6917 mddev
->bitmap_info
.default_offset
;
6918 mddev
->bitmap_info
.space
=
6919 mddev
->bitmap_info
.default_space
;
6920 mddev
->pers
->quiesce(mddev
, 1);
6921 bitmap
= bitmap_create(mddev
, -1);
6922 if (!IS_ERR(bitmap
)) {
6923 mddev
->bitmap
= bitmap
;
6924 rv
= bitmap_load(mddev
);
6926 rv
= PTR_ERR(bitmap
);
6928 bitmap_destroy(mddev
);
6929 mddev
->pers
->quiesce(mddev
, 0);
6931 /* remove the bitmap */
6932 if (!mddev
->bitmap
) {
6936 if (mddev
->bitmap
->storage
.file
) {
6940 if (mddev
->bitmap_info
.nodes
) {
6941 /* hold PW on all the bitmap lock */
6942 if (md_cluster_ops
->lock_all_bitmaps(mddev
) <= 0) {
6943 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6945 md_cluster_ops
->unlock_all_bitmaps(mddev
);
6949 mddev
->bitmap_info
.nodes
= 0;
6950 md_cluster_ops
->leave(mddev
);
6952 mddev
->pers
->quiesce(mddev
, 1);
6953 bitmap_destroy(mddev
);
6954 mddev
->pers
->quiesce(mddev
, 0);
6955 mddev
->bitmap_info
.offset
= 0;
6958 md_update_sb(mddev
, 1);
6964 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6966 struct md_rdev
*rdev
;
6969 if (mddev
->pers
== NULL
)
6973 rdev
= find_rdev_rcu(mddev
, dev
);
6977 md_error(mddev
, rdev
);
6978 if (!test_bit(Faulty
, &rdev
->flags
))
6986 * We have a problem here : there is no easy way to give a CHS
6987 * virtual geometry. We currently pretend that we have a 2 heads
6988 * 4 sectors (with a BIG number of cylinders...). This drives
6989 * dosfs just mad... ;-)
6991 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6993 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6997 geo
->cylinders
= mddev
->array_sectors
/ 8;
7001 static inline bool md_ioctl_valid(unsigned int cmd
)
7006 case GET_ARRAY_INFO
:
7007 case GET_BITMAP_FILE
:
7010 case HOT_REMOVE_DISK
:
7013 case RESTART_ARRAY_RW
:
7015 case SET_ARRAY_INFO
:
7016 case SET_BITMAP_FILE
:
7017 case SET_DISK_FAULTY
:
7020 case CLUSTERED_DISK_NACK
:
7027 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
7028 unsigned int cmd
, unsigned long arg
)
7031 void __user
*argp
= (void __user
*)arg
;
7032 struct mddev
*mddev
= NULL
;
7034 bool did_set_md_closing
= false;
7036 if (!md_ioctl_valid(cmd
))
7041 case GET_ARRAY_INFO
:
7045 if (!capable(CAP_SYS_ADMIN
))
7050 * Commands dealing with the RAID driver but not any
7055 err
= get_version(argp
);
7061 autostart_arrays(arg
);
7068 * Commands creating/starting a new array:
7071 mddev
= bdev
->bd_disk
->private_data
;
7078 /* Some actions do not requires the mutex */
7080 case GET_ARRAY_INFO
:
7081 if (!mddev
->raid_disks
&& !mddev
->external
)
7084 err
= get_array_info(mddev
, argp
);
7088 if (!mddev
->raid_disks
&& !mddev
->external
)
7091 err
= get_disk_info(mddev
, argp
);
7094 case SET_DISK_FAULTY
:
7095 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
7098 case GET_BITMAP_FILE
:
7099 err
= get_bitmap_file(mddev
, argp
);
7104 if (cmd
== ADD_NEW_DISK
)
7105 /* need to ensure md_delayed_delete() has completed */
7106 flush_workqueue(md_misc_wq
);
7108 if (cmd
== HOT_REMOVE_DISK
)
7109 /* need to ensure recovery thread has run */
7110 wait_event_interruptible_timeout(mddev
->sb_wait
,
7111 !test_bit(MD_RECOVERY_NEEDED
,
7113 msecs_to_jiffies(5000));
7114 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
7115 /* Need to flush page cache, and ensure no-one else opens
7118 mutex_lock(&mddev
->open_mutex
);
7119 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
7120 mutex_unlock(&mddev
->open_mutex
);
7124 WARN_ON_ONCE(test_bit(MD_CLOSING
, &mddev
->flags
));
7125 set_bit(MD_CLOSING
, &mddev
->flags
);
7126 did_set_md_closing
= true;
7127 mutex_unlock(&mddev
->open_mutex
);
7128 sync_blockdev(bdev
);
7130 err
= mddev_lock(mddev
);
7132 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7137 if (cmd
== SET_ARRAY_INFO
) {
7138 mdu_array_info_t info
;
7140 memset(&info
, 0, sizeof(info
));
7141 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
7146 err
= update_array_info(mddev
, &info
);
7148 pr_warn("md: couldn't update array info. %d\n", err
);
7153 if (!list_empty(&mddev
->disks
)) {
7154 pr_warn("md: array %s already has disks!\n", mdname(mddev
));
7158 if (mddev
->raid_disks
) {
7159 pr_warn("md: array %s already initialised!\n", mdname(mddev
));
7163 err
= set_array_info(mddev
, &info
);
7165 pr_warn("md: couldn't set array info. %d\n", err
);
7172 * Commands querying/configuring an existing array:
7174 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7175 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7176 if ((!mddev
->raid_disks
&& !mddev
->external
)
7177 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
7178 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
7179 && cmd
!= GET_BITMAP_FILE
) {
7185 * Commands even a read-only array can execute:
7188 case RESTART_ARRAY_RW
:
7189 err
= restart_array(mddev
);
7193 err
= do_md_stop(mddev
, 0, bdev
);
7197 err
= md_set_readonly(mddev
, bdev
);
7200 case HOT_REMOVE_DISK
:
7201 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
7205 /* We can support ADD_NEW_DISK on read-only arrays
7206 * only if we are re-adding a preexisting device.
7207 * So require mddev->pers and MD_DISK_SYNC.
7210 mdu_disk_info_t info
;
7211 if (copy_from_user(&info
, argp
, sizeof(info
)))
7213 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
7214 /* Need to clear read-only for this */
7217 err
= add_new_disk(mddev
, &info
);
7223 if (get_user(ro
, (int __user
*)(arg
))) {
7229 /* if the bdev is going readonly the value of mddev->ro
7230 * does not matter, no writes are coming
7235 /* are we are already prepared for writes? */
7239 /* transitioning to readauto need only happen for
7240 * arrays that call md_write_start
7243 err
= restart_array(mddev
);
7246 set_disk_ro(mddev
->gendisk
, 0);
7253 * The remaining ioctls are changing the state of the
7254 * superblock, so we do not allow them on read-only arrays.
7256 if (mddev
->ro
&& mddev
->pers
) {
7257 if (mddev
->ro
== 2) {
7259 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7260 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7261 /* mddev_unlock will wake thread */
7262 /* If a device failed while we were read-only, we
7263 * need to make sure the metadata is updated now.
7265 if (test_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
)) {
7266 mddev_unlock(mddev
);
7267 wait_event(mddev
->sb_wait
,
7268 !test_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
) &&
7269 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
7270 mddev_lock_nointr(mddev
);
7281 mdu_disk_info_t info
;
7282 if (copy_from_user(&info
, argp
, sizeof(info
)))
7285 err
= add_new_disk(mddev
, &info
);
7289 case CLUSTERED_DISK_NACK
:
7290 if (mddev_is_clustered(mddev
))
7291 md_cluster_ops
->new_disk_ack(mddev
, false);
7297 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
7301 err
= do_md_run(mddev
);
7304 case SET_BITMAP_FILE
:
7305 err
= set_bitmap_file(mddev
, (int)arg
);
7314 if (mddev
->hold_active
== UNTIL_IOCTL
&&
7316 mddev
->hold_active
= 0;
7317 mddev_unlock(mddev
);
7319 if(did_set_md_closing
)
7320 clear_bit(MD_CLOSING
, &mddev
->flags
);
7323 #ifdef CONFIG_COMPAT
7324 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
7325 unsigned int cmd
, unsigned long arg
)
7328 case HOT_REMOVE_DISK
:
7330 case SET_DISK_FAULTY
:
7331 case SET_BITMAP_FILE
:
7332 /* These take in integer arg, do not convert */
7335 arg
= (unsigned long)compat_ptr(arg
);
7339 return md_ioctl(bdev
, mode
, cmd
, arg
);
7341 #endif /* CONFIG_COMPAT */
7343 static int md_open(struct block_device
*bdev
, fmode_t mode
)
7346 * Succeed if we can lock the mddev, which confirms that
7347 * it isn't being stopped right now.
7349 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
7355 if (mddev
->gendisk
!= bdev
->bd_disk
) {
7356 /* we are racing with mddev_put which is discarding this
7360 /* Wait until bdev->bd_disk is definitely gone */
7361 flush_workqueue(md_misc_wq
);
7362 /* Then retry the open from the top */
7363 return -ERESTARTSYS
;
7365 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
7367 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
7370 if (test_bit(MD_CLOSING
, &mddev
->flags
)) {
7371 mutex_unlock(&mddev
->open_mutex
);
7377 atomic_inc(&mddev
->openers
);
7378 mutex_unlock(&mddev
->open_mutex
);
7380 check_disk_change(bdev
);
7387 static void md_release(struct gendisk
*disk
, fmode_t mode
)
7389 struct mddev
*mddev
= disk
->private_data
;
7392 atomic_dec(&mddev
->openers
);
7396 static int md_media_changed(struct gendisk
*disk
)
7398 struct mddev
*mddev
= disk
->private_data
;
7400 return mddev
->changed
;
7403 static int md_revalidate(struct gendisk
*disk
)
7405 struct mddev
*mddev
= disk
->private_data
;
7410 static const struct block_device_operations md_fops
=
7412 .owner
= THIS_MODULE
,
7414 .release
= md_release
,
7416 #ifdef CONFIG_COMPAT
7417 .compat_ioctl
= md_compat_ioctl
,
7419 .getgeo
= md_getgeo
,
7420 .media_changed
= md_media_changed
,
7421 .revalidate_disk
= md_revalidate
,
7424 static int md_thread(void *arg
)
7426 struct md_thread
*thread
= arg
;
7429 * md_thread is a 'system-thread', it's priority should be very
7430 * high. We avoid resource deadlocks individually in each
7431 * raid personality. (RAID5 does preallocation) We also use RR and
7432 * the very same RT priority as kswapd, thus we will never get
7433 * into a priority inversion deadlock.
7435 * we definitely have to have equal or higher priority than
7436 * bdflush, otherwise bdflush will deadlock if there are too
7437 * many dirty RAID5 blocks.
7440 allow_signal(SIGKILL
);
7441 while (!kthread_should_stop()) {
7443 /* We need to wait INTERRUPTIBLE so that
7444 * we don't add to the load-average.
7445 * That means we need to be sure no signals are
7448 if (signal_pending(current
))
7449 flush_signals(current
);
7451 wait_event_interruptible_timeout
7453 test_bit(THREAD_WAKEUP
, &thread
->flags
)
7454 || kthread_should_stop() || kthread_should_park(),
7457 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
7458 if (kthread_should_park())
7460 if (!kthread_should_stop())
7461 thread
->run(thread
);
7467 void md_wakeup_thread(struct md_thread
*thread
)
7470 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
7471 if (!test_and_set_bit(THREAD_WAKEUP
, &thread
->flags
))
7472 wake_up(&thread
->wqueue
);
7475 EXPORT_SYMBOL(md_wakeup_thread
);
7477 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
7478 struct mddev
*mddev
, const char *name
)
7480 struct md_thread
*thread
;
7482 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
7486 init_waitqueue_head(&thread
->wqueue
);
7489 thread
->mddev
= mddev
;
7490 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
7491 thread
->tsk
= kthread_run(md_thread
, thread
,
7493 mdname(thread
->mddev
),
7495 if (IS_ERR(thread
->tsk
)) {
7501 EXPORT_SYMBOL(md_register_thread
);
7503 void md_unregister_thread(struct md_thread
**threadp
)
7505 struct md_thread
*thread
= *threadp
;
7508 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
7509 /* Locking ensures that mddev_unlock does not wake_up a
7510 * non-existent thread
7512 spin_lock(&pers_lock
);
7514 spin_unlock(&pers_lock
);
7516 kthread_stop(thread
->tsk
);
7519 EXPORT_SYMBOL(md_unregister_thread
);
7521 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
7523 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
7526 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
7528 mddev
->pers
->error_handler(mddev
,rdev
);
7529 if (mddev
->degraded
)
7530 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7531 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7532 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7533 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7534 md_wakeup_thread(mddev
->thread
);
7535 if (mddev
->event_work
.func
)
7536 queue_work(md_misc_wq
, &mddev
->event_work
);
7537 md_new_event(mddev
);
7539 EXPORT_SYMBOL(md_error
);
7541 /* seq_file implementation /proc/mdstat */
7543 static void status_unused(struct seq_file
*seq
)
7546 struct md_rdev
*rdev
;
7548 seq_printf(seq
, "unused devices: ");
7550 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
7551 char b
[BDEVNAME_SIZE
];
7553 seq_printf(seq
, "%s ",
7554 bdevname(rdev
->bdev
,b
));
7557 seq_printf(seq
, "<none>");
7559 seq_printf(seq
, "\n");
7562 static int status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
7564 sector_t max_sectors
, resync
, res
;
7565 unsigned long dt
, db
;
7568 unsigned int per_milli
;
7570 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7571 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7572 max_sectors
= mddev
->resync_max_sectors
;
7574 max_sectors
= mddev
->dev_sectors
;
7576 resync
= mddev
->curr_resync
;
7578 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7579 /* Still cleaning up */
7580 resync
= max_sectors
;
7582 resync
-= atomic_read(&mddev
->recovery_active
);
7585 if (mddev
->recovery_cp
< MaxSector
) {
7586 seq_printf(seq
, "\tresync=PENDING");
7592 seq_printf(seq
, "\tresync=DELAYED");
7596 WARN_ON(max_sectors
== 0);
7597 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7598 * in a sector_t, and (max_sectors>>scale) will fit in a
7599 * u32, as those are the requirements for sector_div.
7600 * Thus 'scale' must be at least 10
7603 if (sizeof(sector_t
) > sizeof(unsigned long)) {
7604 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
7607 res
= (resync
>>scale
)*1000;
7608 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
7612 int i
, x
= per_milli
/50, y
= 20-x
;
7613 seq_printf(seq
, "[");
7614 for (i
= 0; i
< x
; i
++)
7615 seq_printf(seq
, "=");
7616 seq_printf(seq
, ">");
7617 for (i
= 0; i
< y
; i
++)
7618 seq_printf(seq
, ".");
7619 seq_printf(seq
, "] ");
7621 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
7622 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
7624 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
7626 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
7627 "resync" : "recovery"))),
7628 per_milli
/10, per_milli
% 10,
7629 (unsigned long long) resync
/2,
7630 (unsigned long long) max_sectors
/2);
7633 * dt: time from mark until now
7634 * db: blocks written from mark until now
7635 * rt: remaining time
7637 * rt is a sector_t, so could be 32bit or 64bit.
7638 * So we divide before multiply in case it is 32bit and close
7640 * We scale the divisor (db) by 32 to avoid losing precision
7641 * near the end of resync when the number of remaining sectors
7643 * We then divide rt by 32 after multiplying by db to compensate.
7644 * The '+1' avoids division by zero if db is very small.
7646 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
7648 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
7649 - mddev
->resync_mark_cnt
;
7651 rt
= max_sectors
- resync
; /* number of remaining sectors */
7652 sector_div(rt
, db
/32+1);
7656 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
7657 ((unsigned long)rt
% 60)/6);
7659 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
7663 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
7665 struct list_head
*tmp
;
7667 struct mddev
*mddev
;
7675 spin_lock(&all_mddevs_lock
);
7676 list_for_each(tmp
,&all_mddevs
)
7678 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7680 spin_unlock(&all_mddevs_lock
);
7683 spin_unlock(&all_mddevs_lock
);
7685 return (void*)2;/* tail */
7689 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7691 struct list_head
*tmp
;
7692 struct mddev
*next_mddev
, *mddev
= v
;
7698 spin_lock(&all_mddevs_lock
);
7700 tmp
= all_mddevs
.next
;
7702 tmp
= mddev
->all_mddevs
.next
;
7703 if (tmp
!= &all_mddevs
)
7704 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7706 next_mddev
= (void*)2;
7709 spin_unlock(&all_mddevs_lock
);
7717 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7719 struct mddev
*mddev
= v
;
7721 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7725 static int md_seq_show(struct seq_file
*seq
, void *v
)
7727 struct mddev
*mddev
= v
;
7729 struct md_rdev
*rdev
;
7731 if (v
== (void*)1) {
7732 struct md_personality
*pers
;
7733 seq_printf(seq
, "Personalities : ");
7734 spin_lock(&pers_lock
);
7735 list_for_each_entry(pers
, &pers_list
, list
)
7736 seq_printf(seq
, "[%s] ", pers
->name
);
7738 spin_unlock(&pers_lock
);
7739 seq_printf(seq
, "\n");
7740 seq
->poll_event
= atomic_read(&md_event_count
);
7743 if (v
== (void*)2) {
7748 spin_lock(&mddev
->lock
);
7749 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7750 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7751 mddev
->pers
? "" : "in");
7754 seq_printf(seq
, " (read-only)");
7756 seq_printf(seq
, " (auto-read-only)");
7757 seq_printf(seq
, " %s", mddev
->pers
->name
);
7762 rdev_for_each_rcu(rdev
, mddev
) {
7763 char b
[BDEVNAME_SIZE
];
7764 seq_printf(seq
, " %s[%d]",
7765 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7766 if (test_bit(WriteMostly
, &rdev
->flags
))
7767 seq_printf(seq
, "(W)");
7768 if (test_bit(Journal
, &rdev
->flags
))
7769 seq_printf(seq
, "(J)");
7770 if (test_bit(Faulty
, &rdev
->flags
)) {
7771 seq_printf(seq
, "(F)");
7774 if (rdev
->raid_disk
< 0)
7775 seq_printf(seq
, "(S)"); /* spare */
7776 if (test_bit(Replacement
, &rdev
->flags
))
7777 seq_printf(seq
, "(R)");
7778 sectors
+= rdev
->sectors
;
7782 if (!list_empty(&mddev
->disks
)) {
7784 seq_printf(seq
, "\n %llu blocks",
7785 (unsigned long long)
7786 mddev
->array_sectors
/ 2);
7788 seq_printf(seq
, "\n %llu blocks",
7789 (unsigned long long)sectors
/ 2);
7791 if (mddev
->persistent
) {
7792 if (mddev
->major_version
!= 0 ||
7793 mddev
->minor_version
!= 90) {
7794 seq_printf(seq
," super %d.%d",
7795 mddev
->major_version
,
7796 mddev
->minor_version
);
7798 } else if (mddev
->external
)
7799 seq_printf(seq
, " super external:%s",
7800 mddev
->metadata_type
);
7802 seq_printf(seq
, " super non-persistent");
7805 mddev
->pers
->status(seq
, mddev
);
7806 seq_printf(seq
, "\n ");
7807 if (mddev
->pers
->sync_request
) {
7808 if (status_resync(seq
, mddev
))
7809 seq_printf(seq
, "\n ");
7812 seq_printf(seq
, "\n ");
7814 bitmap_status(seq
, mddev
->bitmap
);
7816 seq_printf(seq
, "\n");
7818 spin_unlock(&mddev
->lock
);
7823 static const struct seq_operations md_seq_ops
= {
7824 .start
= md_seq_start
,
7825 .next
= md_seq_next
,
7826 .stop
= md_seq_stop
,
7827 .show
= md_seq_show
,
7830 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7832 struct seq_file
*seq
;
7835 error
= seq_open(file
, &md_seq_ops
);
7839 seq
= file
->private_data
;
7840 seq
->poll_event
= atomic_read(&md_event_count
);
7844 static int md_unloading
;
7845 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7847 struct seq_file
*seq
= filp
->private_data
;
7851 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;
7852 poll_wait(filp
, &md_event_waiters
, wait
);
7854 /* always allow read */
7855 mask
= POLLIN
| POLLRDNORM
;
7857 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7858 mask
|= POLLERR
| POLLPRI
;
7862 static const struct file_operations md_seq_fops
= {
7863 .owner
= THIS_MODULE
,
7864 .open
= md_seq_open
,
7866 .llseek
= seq_lseek
,
7867 .release
= seq_release
,
7868 .poll
= mdstat_poll
,
7871 int register_md_personality(struct md_personality
*p
)
7873 pr_debug("md: %s personality registered for level %d\n",
7875 spin_lock(&pers_lock
);
7876 list_add_tail(&p
->list
, &pers_list
);
7877 spin_unlock(&pers_lock
);
7880 EXPORT_SYMBOL(register_md_personality
);
7882 int unregister_md_personality(struct md_personality
*p
)
7884 pr_debug("md: %s personality unregistered\n", p
->name
);
7885 spin_lock(&pers_lock
);
7886 list_del_init(&p
->list
);
7887 spin_unlock(&pers_lock
);
7890 EXPORT_SYMBOL(unregister_md_personality
);
7892 int register_md_cluster_operations(struct md_cluster_operations
*ops
,
7893 struct module
*module
)
7896 spin_lock(&pers_lock
);
7897 if (md_cluster_ops
!= NULL
)
7900 md_cluster_ops
= ops
;
7901 md_cluster_mod
= module
;
7903 spin_unlock(&pers_lock
);
7906 EXPORT_SYMBOL(register_md_cluster_operations
);
7908 int unregister_md_cluster_operations(void)
7910 spin_lock(&pers_lock
);
7911 md_cluster_ops
= NULL
;
7912 spin_unlock(&pers_lock
);
7915 EXPORT_SYMBOL(unregister_md_cluster_operations
);
7917 int md_setup_cluster(struct mddev
*mddev
, int nodes
)
7919 if (!md_cluster_ops
)
7920 request_module("md-cluster");
7921 spin_lock(&pers_lock
);
7922 /* ensure module won't be unloaded */
7923 if (!md_cluster_ops
|| !try_module_get(md_cluster_mod
)) {
7924 pr_warn("can't find md-cluster module or get it's reference.\n");
7925 spin_unlock(&pers_lock
);
7928 spin_unlock(&pers_lock
);
7930 return md_cluster_ops
->join(mddev
, nodes
);
7933 void md_cluster_stop(struct mddev
*mddev
)
7935 if (!md_cluster_ops
)
7937 md_cluster_ops
->leave(mddev
);
7938 module_put(md_cluster_mod
);
7941 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7943 struct md_rdev
*rdev
;
7949 rdev_for_each_rcu(rdev
, mddev
) {
7950 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7951 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7952 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7953 atomic_read(&disk
->sync_io
);
7954 /* sync IO will cause sync_io to increase before the disk_stats
7955 * as sync_io is counted when a request starts, and
7956 * disk_stats is counted when it completes.
7957 * So resync activity will cause curr_events to be smaller than
7958 * when there was no such activity.
7959 * non-sync IO will cause disk_stat to increase without
7960 * increasing sync_io so curr_events will (eventually)
7961 * be larger than it was before. Once it becomes
7962 * substantially larger, the test below will cause
7963 * the array to appear non-idle, and resync will slow
7965 * If there is a lot of outstanding resync activity when
7966 * we set last_event to curr_events, then all that activity
7967 * completing might cause the array to appear non-idle
7968 * and resync will be slowed down even though there might
7969 * not have been non-resync activity. This will only
7970 * happen once though. 'last_events' will soon reflect
7971 * the state where there is little or no outstanding
7972 * resync requests, and further resync activity will
7973 * always make curr_events less than last_events.
7976 if (init
|| curr_events
- rdev
->last_events
> 64) {
7977 rdev
->last_events
= curr_events
;
7985 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7987 /* another "blocks" (512byte) blocks have been synced */
7988 atomic_sub(blocks
, &mddev
->recovery_active
);
7989 wake_up(&mddev
->recovery_wait
);
7991 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7992 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7993 md_wakeup_thread(mddev
->thread
);
7994 // stop recovery, signal do_sync ....
7997 EXPORT_SYMBOL(md_done_sync
);
7999 /* md_write_start(mddev, bi)
8000 * If we need to update some array metadata (e.g. 'active' flag
8001 * in superblock) before writing, schedule a superblock update
8002 * and wait for it to complete.
8003 * A return value of 'false' means that the write wasn't recorded
8004 * and cannot proceed as the array is being suspend.
8006 bool md_write_start(struct mddev
*mddev
, struct bio
*bi
)
8009 if (bio_data_dir(bi
) != WRITE
)
8012 BUG_ON(mddev
->ro
== 1);
8013 if (mddev
->ro
== 2) {
8014 /* need to switch to read/write */
8016 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8017 md_wakeup_thread(mddev
->thread
);
8018 md_wakeup_thread(mddev
->sync_thread
);
8022 percpu_ref_get(&mddev
->writes_pending
);
8023 smp_mb(); /* Match smp_mb in set_in_sync() */
8024 if (mddev
->safemode
== 1)
8025 mddev
->safemode
= 0;
8026 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8027 if (mddev
->in_sync
|| mddev
->sync_checkers
) {
8028 spin_lock(&mddev
->lock
);
8029 if (mddev
->in_sync
) {
8031 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8032 set_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8033 md_wakeup_thread(mddev
->thread
);
8036 spin_unlock(&mddev
->lock
);
8040 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8041 wait_event(mddev
->sb_wait
,
8042 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
) && !mddev
->suspended
);
8043 if (test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
)) {
8044 percpu_ref_put(&mddev
->writes_pending
);
8049 EXPORT_SYMBOL(md_write_start
);
8051 /* md_write_inc can only be called when md_write_start() has
8052 * already been called at least once of the current request.
8053 * It increments the counter and is useful when a single request
8054 * is split into several parts. Each part causes an increment and
8055 * so needs a matching md_write_end().
8056 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8057 * a spinlocked region.
8059 void md_write_inc(struct mddev
*mddev
, struct bio
*bi
)
8061 if (bio_data_dir(bi
) != WRITE
)
8063 WARN_ON_ONCE(mddev
->in_sync
|| mddev
->ro
);
8064 percpu_ref_get(&mddev
->writes_pending
);
8066 EXPORT_SYMBOL(md_write_inc
);
8068 void md_write_end(struct mddev
*mddev
)
8070 percpu_ref_put(&mddev
->writes_pending
);
8072 if (mddev
->safemode
== 2)
8073 md_wakeup_thread(mddev
->thread
);
8074 else if (mddev
->safemode_delay
)
8075 /* The roundup() ensures this only performs locking once
8076 * every ->safemode_delay jiffies
8078 mod_timer(&mddev
->safemode_timer
,
8079 roundup(jiffies
, mddev
->safemode_delay
) +
8080 mddev
->safemode_delay
);
8083 EXPORT_SYMBOL(md_write_end
);
8085 /* md_allow_write(mddev)
8086 * Calling this ensures that the array is marked 'active' so that writes
8087 * may proceed without blocking. It is important to call this before
8088 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8089 * Must be called with mddev_lock held.
8091 void md_allow_write(struct mddev
*mddev
)
8097 if (!mddev
->pers
->sync_request
)
8100 spin_lock(&mddev
->lock
);
8101 if (mddev
->in_sync
) {
8103 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8104 set_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8105 if (mddev
->safemode_delay
&&
8106 mddev
->safemode
== 0)
8107 mddev
->safemode
= 1;
8108 spin_unlock(&mddev
->lock
);
8109 md_update_sb(mddev
, 0);
8110 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8111 /* wait for the dirty state to be recorded in the metadata */
8112 wait_event(mddev
->sb_wait
,
8113 !test_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
) &&
8114 !test_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
));
8116 spin_unlock(&mddev
->lock
);
8118 EXPORT_SYMBOL_GPL(md_allow_write
);
8120 #define SYNC_MARKS 10
8121 #define SYNC_MARK_STEP (3*HZ)
8122 #define UPDATE_FREQUENCY (5*60*HZ)
8123 void md_do_sync(struct md_thread
*thread
)
8125 struct mddev
*mddev
= thread
->mddev
;
8126 struct mddev
*mddev2
;
8127 unsigned int currspeed
= 0,
8129 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
8130 unsigned long mark
[SYNC_MARKS
];
8131 unsigned long update_time
;
8132 sector_t mark_cnt
[SYNC_MARKS
];
8134 struct list_head
*tmp
;
8135 sector_t last_check
;
8137 struct md_rdev
*rdev
;
8138 char *desc
, *action
= NULL
;
8139 struct blk_plug plug
;
8142 /* just incase thread restarts... */
8143 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
8145 if (mddev
->ro
) {/* never try to sync a read-only array */
8146 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8150 if (mddev_is_clustered(mddev
)) {
8151 ret
= md_cluster_ops
->resync_start(mddev
);
8155 set_bit(MD_CLUSTER_RESYNC_LOCKED
, &mddev
->flags
);
8156 if (!(test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
8157 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) ||
8158 test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
8159 && ((unsigned long long)mddev
->curr_resync_completed
8160 < (unsigned long long)mddev
->resync_max_sectors
))
8164 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8165 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
8166 desc
= "data-check";
8168 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8169 desc
= "requested-resync";
8173 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
8178 mddev
->last_sync_action
= action
?: desc
;
8180 /* we overload curr_resync somewhat here.
8181 * 0 == not engaged in resync at all
8182 * 2 == checking that there is no conflict with another sync
8183 * 1 == like 2, but have yielded to allow conflicting resync to
8185 * other == active in resync - this many blocks
8187 * Before starting a resync we must have set curr_resync to
8188 * 2, and then checked that every "conflicting" array has curr_resync
8189 * less than ours. When we find one that is the same or higher
8190 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8191 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8192 * This will mean we have to start checking from the beginning again.
8197 int mddev2_minor
= -1;
8198 mddev
->curr_resync
= 2;
8201 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8203 for_each_mddev(mddev2
, tmp
) {
8204 if (mddev2
== mddev
)
8206 if (!mddev
->parallel_resync
8207 && mddev2
->curr_resync
8208 && match_mddev_units(mddev
, mddev2
)) {
8210 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
8211 /* arbitrarily yield */
8212 mddev
->curr_resync
= 1;
8213 wake_up(&resync_wait
);
8215 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
8216 /* no need to wait here, we can wait the next
8217 * time 'round when curr_resync == 2
8220 /* We need to wait 'interruptible' so as not to
8221 * contribute to the load average, and not to
8222 * be caught by 'softlockup'
8224 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
8225 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8226 mddev2
->curr_resync
>= mddev
->curr_resync
) {
8227 if (mddev2_minor
!= mddev2
->md_minor
) {
8228 mddev2_minor
= mddev2
->md_minor
;
8229 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8230 desc
, mdname(mddev
),
8234 if (signal_pending(current
))
8235 flush_signals(current
);
8237 finish_wait(&resync_wait
, &wq
);
8240 finish_wait(&resync_wait
, &wq
);
8243 } while (mddev
->curr_resync
< 2);
8246 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8247 /* resync follows the size requested by the personality,
8248 * which defaults to physical size, but can be virtual size
8250 max_sectors
= mddev
->resync_max_sectors
;
8251 atomic64_set(&mddev
->resync_mismatches
, 0);
8252 /* we don't use the checkpoint if there's a bitmap */
8253 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8254 j
= mddev
->resync_min
;
8255 else if (!mddev
->bitmap
)
8256 j
= mddev
->recovery_cp
;
8258 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
8259 max_sectors
= mddev
->resync_max_sectors
;
8261 /* recovery follows the physical size of devices */
8262 max_sectors
= mddev
->dev_sectors
;
8265 rdev_for_each_rcu(rdev
, mddev
)
8266 if (rdev
->raid_disk
>= 0 &&
8267 !test_bit(Journal
, &rdev
->flags
) &&
8268 !test_bit(Faulty
, &rdev
->flags
) &&
8269 !test_bit(In_sync
, &rdev
->flags
) &&
8270 rdev
->recovery_offset
< j
)
8271 j
= rdev
->recovery_offset
;
8274 /* If there is a bitmap, we need to make sure all
8275 * writes that started before we added a spare
8276 * complete before we start doing a recovery.
8277 * Otherwise the write might complete and (via
8278 * bitmap_endwrite) set a bit in the bitmap after the
8279 * recovery has checked that bit and skipped that
8282 if (mddev
->bitmap
) {
8283 mddev
->pers
->quiesce(mddev
, 1);
8284 mddev
->pers
->quiesce(mddev
, 0);
8288 pr_info("md: %s of RAID array %s\n", desc
, mdname(mddev
));
8289 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev
));
8290 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8291 speed_max(mddev
), desc
);
8293 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
8296 for (m
= 0; m
< SYNC_MARKS
; m
++) {
8298 mark_cnt
[m
] = io_sectors
;
8301 mddev
->resync_mark
= mark
[last_mark
];
8302 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
8305 * Tune reconstruction:
8307 window
= 32*(PAGE_SIZE
/512);
8308 pr_debug("md: using %dk window, over a total of %lluk.\n",
8309 window
/2, (unsigned long long)max_sectors
/2);
8311 atomic_set(&mddev
->recovery_active
, 0);
8315 pr_debug("md: resuming %s of %s from checkpoint.\n",
8316 desc
, mdname(mddev
));
8317 mddev
->curr_resync
= j
;
8319 mddev
->curr_resync
= 3; /* no longer delayed */
8320 mddev
->curr_resync_completed
= j
;
8321 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8322 md_new_event(mddev
);
8323 update_time
= jiffies
;
8325 blk_start_plug(&plug
);
8326 while (j
< max_sectors
) {
8331 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8332 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
8333 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
8334 > (max_sectors
>> 4)) ||
8335 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
8336 (j
- mddev
->curr_resync_completed
)*2
8337 >= mddev
->resync_max
- mddev
->curr_resync_completed
||
8338 mddev
->curr_resync_completed
> mddev
->resync_max
8340 /* time to update curr_resync_completed */
8341 wait_event(mddev
->recovery_wait
,
8342 atomic_read(&mddev
->recovery_active
) == 0);
8343 mddev
->curr_resync_completed
= j
;
8344 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
8345 j
> mddev
->recovery_cp
)
8346 mddev
->recovery_cp
= j
;
8347 update_time
= jiffies
;
8348 set_bit(MD_SB_CHANGE_CLEAN
, &mddev
->sb_flags
);
8349 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8352 while (j
>= mddev
->resync_max
&&
8353 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8354 /* As this condition is controlled by user-space,
8355 * we can block indefinitely, so use '_interruptible'
8356 * to avoid triggering warnings.
8358 flush_signals(current
); /* just in case */
8359 wait_event_interruptible(mddev
->recovery_wait
,
8360 mddev
->resync_max
> j
8361 || test_bit(MD_RECOVERY_INTR
,
8365 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8368 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
);
8370 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8374 if (!skipped
) { /* actual IO requested */
8375 io_sectors
+= sectors
;
8376 atomic_add(sectors
, &mddev
->recovery_active
);
8379 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8383 if (j
> max_sectors
)
8384 /* when skipping, extra large numbers can be returned. */
8387 mddev
->curr_resync
= j
;
8388 mddev
->curr_mark_cnt
= io_sectors
;
8389 if (last_check
== 0)
8390 /* this is the earliest that rebuild will be
8391 * visible in /proc/mdstat
8393 md_new_event(mddev
);
8395 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
8398 last_check
= io_sectors
;
8400 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
8402 int next
= (last_mark
+1) % SYNC_MARKS
;
8404 mddev
->resync_mark
= mark
[next
];
8405 mddev
->resync_mark_cnt
= mark_cnt
[next
];
8406 mark
[next
] = jiffies
;
8407 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
8411 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8415 * this loop exits only if either when we are slower than
8416 * the 'hard' speed limit, or the system was IO-idle for
8418 * the system might be non-idle CPU-wise, but we only care
8419 * about not overloading the IO subsystem. (things like an
8420 * e2fsck being done on the RAID array should execute fast)
8424 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
8425 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
8426 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
8428 if (currspeed
> speed_min(mddev
)) {
8429 if (currspeed
> speed_max(mddev
)) {
8433 if (!is_mddev_idle(mddev
, 0)) {
8435 * Give other IO more of a chance.
8436 * The faster the devices, the less we wait.
8438 wait_event(mddev
->recovery_wait
,
8439 !atomic_read(&mddev
->recovery_active
));
8443 pr_info("md: %s: %s %s.\n",mdname(mddev
), desc
,
8444 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
8445 ? "interrupted" : "done");
8447 * this also signals 'finished resyncing' to md_stop
8449 blk_finish_plug(&plug
);
8450 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
8452 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8453 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8454 mddev
->curr_resync
> 3) {
8455 mddev
->curr_resync_completed
= mddev
->curr_resync
;
8456 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8458 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
);
8460 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
8461 mddev
->curr_resync
> 3) {
8462 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8463 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8464 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
8465 pr_debug("md: checkpointing %s of %s.\n",
8466 desc
, mdname(mddev
));
8467 if (test_bit(MD_RECOVERY_ERROR
,
8469 mddev
->recovery_cp
=
8470 mddev
->curr_resync_completed
;
8472 mddev
->recovery_cp
=
8476 mddev
->recovery_cp
= MaxSector
;
8478 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8479 mddev
->curr_resync
= MaxSector
;
8481 rdev_for_each_rcu(rdev
, mddev
)
8482 if (rdev
->raid_disk
>= 0 &&
8483 mddev
->delta_disks
>= 0 &&
8484 !test_bit(Journal
, &rdev
->flags
) &&
8485 !test_bit(Faulty
, &rdev
->flags
) &&
8486 !test_bit(In_sync
, &rdev
->flags
) &&
8487 rdev
->recovery_offset
< mddev
->curr_resync
)
8488 rdev
->recovery_offset
= mddev
->curr_resync
;
8493 /* set CHANGE_PENDING here since maybe another update is needed,
8494 * so other nodes are informed. It should be harmless for normal
8496 set_mask_bits(&mddev
->sb_flags
, 0,
8497 BIT(MD_SB_CHANGE_PENDING
) | BIT(MD_SB_CHANGE_DEVS
));
8499 spin_lock(&mddev
->lock
);
8500 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8501 /* We completed so min/max setting can be forgotten if used. */
8502 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8503 mddev
->resync_min
= 0;
8504 mddev
->resync_max
= MaxSector
;
8505 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8506 mddev
->resync_min
= mddev
->curr_resync_completed
;
8507 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8508 mddev
->curr_resync
= 0;
8509 spin_unlock(&mddev
->lock
);
8511 wake_up(&resync_wait
);
8512 md_wakeup_thread(mddev
->thread
);
8515 EXPORT_SYMBOL_GPL(md_do_sync
);
8517 static int remove_and_add_spares(struct mddev
*mddev
,
8518 struct md_rdev
*this)
8520 struct md_rdev
*rdev
;
8523 bool remove_some
= false;
8525 rdev_for_each(rdev
, mddev
) {
8526 if ((this == NULL
|| rdev
== this) &&
8527 rdev
->raid_disk
>= 0 &&
8528 !test_bit(Blocked
, &rdev
->flags
) &&
8529 test_bit(Faulty
, &rdev
->flags
) &&
8530 atomic_read(&rdev
->nr_pending
)==0) {
8531 /* Faulty non-Blocked devices with nr_pending == 0
8532 * never get nr_pending incremented,
8533 * never get Faulty cleared, and never get Blocked set.
8534 * So we can synchronize_rcu now rather than once per device
8537 set_bit(RemoveSynchronized
, &rdev
->flags
);
8543 rdev_for_each(rdev
, mddev
) {
8544 if ((this == NULL
|| rdev
== this) &&
8545 rdev
->raid_disk
>= 0 &&
8546 !test_bit(Blocked
, &rdev
->flags
) &&
8547 ((test_bit(RemoveSynchronized
, &rdev
->flags
) ||
8548 (!test_bit(In_sync
, &rdev
->flags
) &&
8549 !test_bit(Journal
, &rdev
->flags
))) &&
8550 atomic_read(&rdev
->nr_pending
)==0)) {
8551 if (mddev
->pers
->hot_remove_disk(
8552 mddev
, rdev
) == 0) {
8553 sysfs_unlink_rdev(mddev
, rdev
);
8554 rdev
->raid_disk
= -1;
8558 if (remove_some
&& test_bit(RemoveSynchronized
, &rdev
->flags
))
8559 clear_bit(RemoveSynchronized
, &rdev
->flags
);
8562 if (removed
&& mddev
->kobj
.sd
)
8563 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
8565 if (this && removed
)
8568 rdev_for_each(rdev
, mddev
) {
8569 if (this && this != rdev
)
8571 if (test_bit(Candidate
, &rdev
->flags
))
8573 if (rdev
->raid_disk
>= 0 &&
8574 !test_bit(In_sync
, &rdev
->flags
) &&
8575 !test_bit(Journal
, &rdev
->flags
) &&
8576 !test_bit(Faulty
, &rdev
->flags
))
8578 if (rdev
->raid_disk
>= 0)
8580 if (test_bit(Faulty
, &rdev
->flags
))
8582 if (!test_bit(Journal
, &rdev
->flags
)) {
8584 ! (rdev
->saved_raid_disk
>= 0 &&
8585 !test_bit(Bitmap_sync
, &rdev
->flags
)))
8588 rdev
->recovery_offset
= 0;
8591 hot_add_disk(mddev
, rdev
) == 0) {
8592 if (sysfs_link_rdev(mddev
, rdev
))
8593 /* failure here is OK */;
8594 if (!test_bit(Journal
, &rdev
->flags
))
8596 md_new_event(mddev
);
8597 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8602 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8606 static void md_start_sync(struct work_struct
*ws
)
8608 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
8610 mddev
->sync_thread
= md_register_thread(md_do_sync
,
8613 if (!mddev
->sync_thread
) {
8614 pr_warn("%s: could not start resync thread...\n",
8616 /* leave the spares where they are, it shouldn't hurt */
8617 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8618 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8619 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8620 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8621 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8622 wake_up(&resync_wait
);
8623 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8625 if (mddev
->sysfs_action
)
8626 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8628 md_wakeup_thread(mddev
->sync_thread
);
8629 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8630 md_new_event(mddev
);
8634 * This routine is regularly called by all per-raid-array threads to
8635 * deal with generic issues like resync and super-block update.
8636 * Raid personalities that don't have a thread (linear/raid0) do not
8637 * need this as they never do any recovery or update the superblock.
8639 * It does not do any resync itself, but rather "forks" off other threads
8640 * to do that as needed.
8641 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8642 * "->recovery" and create a thread at ->sync_thread.
8643 * When the thread finishes it sets MD_RECOVERY_DONE
8644 * and wakeups up this thread which will reap the thread and finish up.
8645 * This thread also removes any faulty devices (with nr_pending == 0).
8647 * The overall approach is:
8648 * 1/ if the superblock needs updating, update it.
8649 * 2/ If a recovery thread is running, don't do anything else.
8650 * 3/ If recovery has finished, clean up, possibly marking spares active.
8651 * 4/ If there are any faulty devices, remove them.
8652 * 5/ If array is degraded, try to add spares devices
8653 * 6/ If array has spares or is not in-sync, start a resync thread.
8655 void md_check_recovery(struct mddev
*mddev
)
8657 if (mddev
->suspended
)
8661 bitmap_daemon_work(mddev
);
8663 if (signal_pending(current
)) {
8664 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
8665 pr_debug("md: %s in immediate safe mode\n",
8667 mddev
->safemode
= 2;
8669 flush_signals(current
);
8672 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
8675 (mddev
->sb_flags
& ~ (1<<MD_SB_CHANGE_PENDING
)) ||
8676 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8677 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8678 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
8679 (mddev
->safemode
== 2
8680 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
8684 if (mddev_trylock(mddev
)) {
8687 if (!mddev
->external
&& mddev
->safemode
== 1)
8688 mddev
->safemode
= 0;
8691 struct md_rdev
*rdev
;
8692 if (!mddev
->external
&& mddev
->in_sync
)
8693 /* 'Blocked' flag not needed as failed devices
8694 * will be recorded if array switched to read/write.
8695 * Leaving it set will prevent the device
8696 * from being removed.
8698 rdev_for_each(rdev
, mddev
)
8699 clear_bit(Blocked
, &rdev
->flags
);
8700 /* On a read-only array we can:
8701 * - remove failed devices
8702 * - add already-in_sync devices if the array itself
8704 * As we only add devices that are already in-sync,
8705 * we can activate the spares immediately.
8707 remove_and_add_spares(mddev
, NULL
);
8708 /* There is no thread, but we need to call
8709 * ->spare_active and clear saved_raid_disk
8711 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8712 md_reap_sync_thread(mddev
);
8713 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8714 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8715 clear_bit(MD_SB_CHANGE_PENDING
, &mddev
->sb_flags
);
8719 if (mddev_is_clustered(mddev
)) {
8720 struct md_rdev
*rdev
;
8721 /* kick the device if another node issued a
8724 rdev_for_each(rdev
, mddev
) {
8725 if (test_and_clear_bit(ClusterRemove
, &rdev
->flags
) &&
8726 rdev
->raid_disk
< 0)
8727 md_kick_rdev_from_array(rdev
);
8731 if (!mddev
->external
&& !mddev
->in_sync
) {
8732 spin_lock(&mddev
->lock
);
8734 spin_unlock(&mddev
->lock
);
8737 if (mddev
->sb_flags
)
8738 md_update_sb(mddev
, 0);
8740 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
8741 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
8742 /* resync/recovery still happening */
8743 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8746 if (mddev
->sync_thread
) {
8747 md_reap_sync_thread(mddev
);
8750 /* Set RUNNING before clearing NEEDED to avoid
8751 * any transients in the value of "sync_action".
8753 mddev
->curr_resync_completed
= 0;
8754 spin_lock(&mddev
->lock
);
8755 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8756 spin_unlock(&mddev
->lock
);
8757 /* Clear some bits that don't mean anything, but
8760 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8761 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8763 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8764 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
8766 /* no recovery is running.
8767 * remove any failed drives, then
8768 * add spares if possible.
8769 * Spares are also removed and re-added, to allow
8770 * the personality to fail the re-add.
8773 if (mddev
->reshape_position
!= MaxSector
) {
8774 if (mddev
->pers
->check_reshape
== NULL
||
8775 mddev
->pers
->check_reshape(mddev
) != 0)
8776 /* Cannot proceed */
8778 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8779 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8780 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
8781 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8782 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8783 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8784 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8785 } else if (mddev
->recovery_cp
< MaxSector
) {
8786 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8787 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8788 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
8789 /* nothing to be done ... */
8792 if (mddev
->pers
->sync_request
) {
8794 /* We are adding a device or devices to an array
8795 * which has the bitmap stored on all devices.
8796 * So make sure all bitmap pages get written
8798 bitmap_write_all(mddev
->bitmap
);
8800 INIT_WORK(&mddev
->del_work
, md_start_sync
);
8801 queue_work(md_misc_wq
, &mddev
->del_work
);
8805 if (!mddev
->sync_thread
) {
8806 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8807 wake_up(&resync_wait
);
8808 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8810 if (mddev
->sysfs_action
)
8811 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8814 wake_up(&mddev
->sb_wait
);
8815 mddev_unlock(mddev
);
8818 EXPORT_SYMBOL(md_check_recovery
);
8820 void md_reap_sync_thread(struct mddev
*mddev
)
8822 struct md_rdev
*rdev
;
8824 /* resync has finished, collect result */
8825 md_unregister_thread(&mddev
->sync_thread
);
8826 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8827 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8829 /* activate any spares */
8830 if (mddev
->pers
->spare_active(mddev
)) {
8831 sysfs_notify(&mddev
->kobj
, NULL
,
8833 set_bit(MD_SB_CHANGE_DEVS
, &mddev
->sb_flags
);
8836 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8837 mddev
->pers
->finish_reshape
)
8838 mddev
->pers
->finish_reshape(mddev
);
8840 /* If array is no-longer degraded, then any saved_raid_disk
8841 * information must be scrapped.
8843 if (!mddev
->degraded
)
8844 rdev_for_each(rdev
, mddev
)
8845 rdev
->saved_raid_disk
= -1;
8847 md_update_sb(mddev
, 1);
8848 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8849 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8851 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED
, &mddev
->flags
))
8852 md_cluster_ops
->resync_finish(mddev
);
8853 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8854 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8855 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8856 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8857 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8858 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8859 wake_up(&resync_wait
);
8860 /* flag recovery needed just to double check */
8861 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8862 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8863 md_new_event(mddev
);
8864 if (mddev
->event_work
.func
)
8865 queue_work(md_misc_wq
, &mddev
->event_work
);
8867 EXPORT_SYMBOL(md_reap_sync_thread
);
8869 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
8871 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8872 wait_event_timeout(rdev
->blocked_wait
,
8873 !test_bit(Blocked
, &rdev
->flags
) &&
8874 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
8875 msecs_to_jiffies(5000));
8876 rdev_dec_pending(rdev
, mddev
);
8878 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
8880 void md_finish_reshape(struct mddev
*mddev
)
8882 /* called be personality module when reshape completes. */
8883 struct md_rdev
*rdev
;
8885 rdev_for_each(rdev
, mddev
) {
8886 if (rdev
->data_offset
> rdev
->new_data_offset
)
8887 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
8889 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
8890 rdev
->data_offset
= rdev
->new_data_offset
;
8893 EXPORT_SYMBOL(md_finish_reshape
);
8895 /* Bad block management */
8897 /* Returns 1 on success, 0 on failure */
8898 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8901 struct mddev
*mddev
= rdev
->mddev
;
8904 s
+= rdev
->new_data_offset
;
8906 s
+= rdev
->data_offset
;
8907 rv
= badblocks_set(&rdev
->badblocks
, s
, sectors
, 0);
8909 /* Make sure they get written out promptly */
8910 if (test_bit(ExternalBbl
, &rdev
->flags
))
8911 sysfs_notify(&rdev
->kobj
, NULL
,
8912 "unacknowledged_bad_blocks");
8913 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8914 set_mask_bits(&mddev
->sb_flags
, 0,
8915 BIT(MD_SB_CHANGE_CLEAN
) | BIT(MD_SB_CHANGE_PENDING
));
8916 md_wakeup_thread(rdev
->mddev
->thread
);
8921 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8923 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8928 s
+= rdev
->new_data_offset
;
8930 s
+= rdev
->data_offset
;
8931 rv
= badblocks_clear(&rdev
->badblocks
, s
, sectors
);
8932 if ((rv
== 0) && test_bit(ExternalBbl
, &rdev
->flags
))
8933 sysfs_notify(&rdev
->kobj
, NULL
, "bad_blocks");
8936 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8938 static int md_notify_reboot(struct notifier_block
*this,
8939 unsigned long code
, void *x
)
8941 struct list_head
*tmp
;
8942 struct mddev
*mddev
;
8945 for_each_mddev(mddev
, tmp
) {
8946 if (mddev_trylock(mddev
)) {
8948 __md_stop_writes(mddev
);
8949 if (mddev
->persistent
)
8950 mddev
->safemode
= 2;
8951 mddev_unlock(mddev
);
8956 * certain more exotic SCSI devices are known to be
8957 * volatile wrt too early system reboots. While the
8958 * right place to handle this issue is the given
8959 * driver, we do want to have a safe RAID driver ...
8967 static struct notifier_block md_notifier
= {
8968 .notifier_call
= md_notify_reboot
,
8970 .priority
= INT_MAX
, /* before any real devices */
8973 static void md_geninit(void)
8975 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8977 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8980 static int __init
md_init(void)
8984 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8988 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8992 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8995 if ((ret
= register_blkdev(0, "mdp")) < 0)
8999 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
9000 md_probe
, NULL
, NULL
);
9001 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
9002 md_probe
, NULL
, NULL
);
9004 register_reboot_notifier(&md_notifier
);
9005 raid_table_header
= register_sysctl_table(raid_root_table
);
9011 unregister_blkdev(MD_MAJOR
, "md");
9013 destroy_workqueue(md_misc_wq
);
9015 destroy_workqueue(md_wq
);
9020 static void check_sb_changes(struct mddev
*mddev
, struct md_rdev
*rdev
)
9022 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
9023 struct md_rdev
*rdev2
;
9025 char b
[BDEVNAME_SIZE
];
9028 * If size is changed in another node then we need to
9029 * do resize as well.
9031 if (mddev
->dev_sectors
!= le64_to_cpu(sb
->size
)) {
9032 ret
= mddev
->pers
->resize(mddev
, le64_to_cpu(sb
->size
));
9034 pr_info("md-cluster: resize failed\n");
9036 bitmap_update_sb(mddev
->bitmap
);
9039 /* Check for change of roles in the active devices */
9040 rdev_for_each(rdev2
, mddev
) {
9041 if (test_bit(Faulty
, &rdev2
->flags
))
9044 /* Check if the roles changed */
9045 role
= le16_to_cpu(sb
->dev_roles
[rdev2
->desc_nr
]);
9047 if (test_bit(Candidate
, &rdev2
->flags
)) {
9048 if (role
== 0xfffe) {
9049 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2
->bdev
,b
));
9050 md_kick_rdev_from_array(rdev2
);
9054 clear_bit(Candidate
, &rdev2
->flags
);
9057 if (role
!= rdev2
->raid_disk
) {
9059 if (rdev2
->raid_disk
== -1 && role
!= 0xffff) {
9060 rdev2
->saved_raid_disk
= role
;
9061 ret
= remove_and_add_spares(mddev
, rdev2
);
9062 pr_info("Activated spare: %s\n",
9063 bdevname(rdev2
->bdev
,b
));
9064 /* wakeup mddev->thread here, so array could
9065 * perform resync with the new activated disk */
9066 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
9067 md_wakeup_thread(mddev
->thread
);
9071 * We just want to do the minimum to mark the disk
9072 * as faulty. The recovery is performed by the
9073 * one who initiated the error.
9075 if ((role
== 0xfffe) || (role
== 0xfffd)) {
9076 md_error(mddev
, rdev2
);
9077 clear_bit(Blocked
, &rdev2
->flags
);
9082 if (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
))
9083 update_raid_disks(mddev
, le32_to_cpu(sb
->raid_disks
));
9085 /* Finally set the event to be up to date */
9086 mddev
->events
= le64_to_cpu(sb
->events
);
9089 static int read_rdev(struct mddev
*mddev
, struct md_rdev
*rdev
)
9092 struct page
*swapout
= rdev
->sb_page
;
9093 struct mdp_superblock_1
*sb
;
9095 /* Store the sb page of the rdev in the swapout temporary
9096 * variable in case we err in the future
9098 rdev
->sb_page
= NULL
;
9099 err
= alloc_disk_sb(rdev
);
9101 ClearPageUptodate(rdev
->sb_page
);
9102 rdev
->sb_loaded
= 0;
9103 err
= super_types
[mddev
->major_version
].
9104 load_super(rdev
, NULL
, mddev
->minor_version
);
9107 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9108 __func__
, __LINE__
, rdev
->desc_nr
, err
);
9110 put_page(rdev
->sb_page
);
9111 rdev
->sb_page
= swapout
;
9112 rdev
->sb_loaded
= 1;
9116 sb
= page_address(rdev
->sb_page
);
9117 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9121 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
))
9122 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
9124 /* The other node finished recovery, call spare_active to set
9125 * device In_sync and mddev->degraded
9127 if (rdev
->recovery_offset
== MaxSector
&&
9128 !test_bit(In_sync
, &rdev
->flags
) &&
9129 mddev
->pers
->spare_active(mddev
))
9130 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
9136 void md_reload_sb(struct mddev
*mddev
, int nr
)
9138 struct md_rdev
*rdev
;
9142 rdev_for_each_rcu(rdev
, mddev
) {
9143 if (rdev
->desc_nr
== nr
)
9147 if (!rdev
|| rdev
->desc_nr
!= nr
) {
9148 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__
, __LINE__
, nr
);
9152 err
= read_rdev(mddev
, rdev
);
9156 check_sb_changes(mddev
, rdev
);
9158 /* Read all rdev's to update recovery_offset */
9159 rdev_for_each_rcu(rdev
, mddev
)
9160 read_rdev(mddev
, rdev
);
9162 EXPORT_SYMBOL(md_reload_sb
);
9167 * Searches all registered partitions for autorun RAID arrays
9171 static DEFINE_MUTEX(detected_devices_mutex
);
9172 static LIST_HEAD(all_detected_devices
);
9173 struct detected_devices_node
{
9174 struct list_head list
;
9178 void md_autodetect_dev(dev_t dev
)
9180 struct detected_devices_node
*node_detected_dev
;
9182 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
9183 if (node_detected_dev
) {
9184 node_detected_dev
->dev
= dev
;
9185 mutex_lock(&detected_devices_mutex
);
9186 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
9187 mutex_unlock(&detected_devices_mutex
);
9191 static void autostart_arrays(int part
)
9193 struct md_rdev
*rdev
;
9194 struct detected_devices_node
*node_detected_dev
;
9196 int i_scanned
, i_passed
;
9201 pr_info("md: Autodetecting RAID arrays.\n");
9203 mutex_lock(&detected_devices_mutex
);
9204 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
9206 node_detected_dev
= list_entry(all_detected_devices
.next
,
9207 struct detected_devices_node
, list
);
9208 list_del(&node_detected_dev
->list
);
9209 dev
= node_detected_dev
->dev
;
9210 kfree(node_detected_dev
);
9211 mutex_unlock(&detected_devices_mutex
);
9212 rdev
= md_import_device(dev
,0, 90);
9213 mutex_lock(&detected_devices_mutex
);
9217 if (test_bit(Faulty
, &rdev
->flags
))
9220 set_bit(AutoDetected
, &rdev
->flags
);
9221 list_add(&rdev
->same_set
, &pending_raid_disks
);
9224 mutex_unlock(&detected_devices_mutex
);
9226 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned
, i_passed
);
9228 autorun_devices(part
);
9231 #endif /* !MODULE */
9233 static __exit
void md_exit(void)
9235 struct mddev
*mddev
;
9236 struct list_head
*tmp
;
9239 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
9240 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
9242 unregister_blkdev(MD_MAJOR
,"md");
9243 unregister_blkdev(mdp_major
, "mdp");
9244 unregister_reboot_notifier(&md_notifier
);
9245 unregister_sysctl_table(raid_table_header
);
9247 /* We cannot unload the modules while some process is
9248 * waiting for us in select() or poll() - wake them up
9251 while (waitqueue_active(&md_event_waiters
)) {
9252 /* not safe to leave yet */
9253 wake_up(&md_event_waiters
);
9257 remove_proc_entry("mdstat", NULL
);
9259 for_each_mddev(mddev
, tmp
) {
9260 export_array(mddev
);
9262 mddev
->hold_active
= 0;
9264 * for_each_mddev() will call mddev_put() at the end of each
9265 * iteration. As the mddev is now fully clear, this will
9266 * schedule the mddev for destruction by a workqueue, and the
9267 * destroy_workqueue() below will wait for that to complete.
9270 destroy_workqueue(md_misc_wq
);
9271 destroy_workqueue(md_wq
);
9274 subsys_initcall(md_init
);
9275 module_exit(md_exit
)
9277 static int get_ro(char *buffer
, struct kernel_param
*kp
)
9279 return sprintf(buffer
, "%d", start_readonly
);
9281 static int set_ro(const char *val
, struct kernel_param
*kp
)
9283 return kstrtouint(val
, 10, (unsigned int *)&start_readonly
);
9286 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
9287 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
9288 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
9289 module_param(create_on_open
, bool, S_IRUSR
|S_IWUSR
);
9291 MODULE_LICENSE("GPL");
9292 MODULE_DESCRIPTION("MD RAID framework");
9294 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);