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.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
56 #include "md-cluster.h"
59 static void autostart_arrays(int part
);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list
);
68 static DEFINE_SPINLOCK(pers_lock
);
70 struct md_cluster_operations
*md_cluster_ops
;
71 EXPORT_SYMBOL(md_cluster_ops
);
72 struct module
*md_cluster_mod
;
73 EXPORT_SYMBOL(md_cluster_mod
);
75 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
76 static struct workqueue_struct
*md_wq
;
77 static struct workqueue_struct
*md_misc_wq
;
79 static int remove_and_add_spares(struct mddev
*mddev
,
80 struct md_rdev
*this);
81 static void mddev_detach(struct mddev
*mddev
);
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
88 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
93 * the RAID driver will use the maximum available bandwidth if the IO
94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
99 * or /sys/block/mdX/md/sync_speed_{min,max}
102 static int sysctl_speed_limit_min
= 1000;
103 static int sysctl_speed_limit_max
= 200000;
104 static inline int speed_min(struct mddev
*mddev
)
106 return mddev
->sync_speed_min
?
107 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
110 static inline int speed_max(struct mddev
*mddev
)
112 return mddev
->sync_speed_max
?
113 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
116 static struct ctl_table_header
*raid_table_header
;
118 static struct ctl_table raid_table
[] = {
120 .procname
= "speed_limit_min",
121 .data
= &sysctl_speed_limit_min
,
122 .maxlen
= sizeof(int),
123 .mode
= S_IRUGO
|S_IWUSR
,
124 .proc_handler
= proc_dointvec
,
127 .procname
= "speed_limit_max",
128 .data
= &sysctl_speed_limit_max
,
129 .maxlen
= sizeof(int),
130 .mode
= S_IRUGO
|S_IWUSR
,
131 .proc_handler
= proc_dointvec
,
136 static struct ctl_table raid_dir_table
[] = {
140 .mode
= S_IRUGO
|S_IXUGO
,
146 static struct ctl_table raid_root_table
[] = {
151 .child
= raid_dir_table
,
156 static const struct block_device_operations md_fops
;
158 static int start_readonly
;
161 * like bio_clone, but with a local bio set
164 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
169 if (!mddev
|| !mddev
->bio_set
)
170 return bio_alloc(gfp_mask
, nr_iovecs
);
172 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
177 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
179 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
182 if (!mddev
|| !mddev
->bio_set
)
183 return bio_clone(bio
, gfp_mask
);
185 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
187 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
199 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
200 static atomic_t md_event_count
;
201 void md_new_event(struct mddev
*mddev
)
203 atomic_inc(&md_event_count
);
204 wake_up(&md_event_waiters
);
206 EXPORT_SYMBOL_GPL(md_new_event
);
208 /* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
211 static void md_new_event_inintr(struct mddev
*mddev
)
213 atomic_inc(&md_event_count
);
214 wake_up(&md_event_waiters
);
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
221 static LIST_HEAD(all_mddevs
);
222 static DEFINE_SPINLOCK(all_mddevs_lock
);
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
231 #define for_each_mddev(_mddev,_tmp) \
233 for (({ spin_lock(&all_mddevs_lock); \
234 _tmp = all_mddevs.next; \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
238 spin_unlock(&all_mddevs_lock); \
239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
242 ({ spin_lock(&all_mddevs_lock); \
243 _tmp = _tmp->next;}) \
246 /* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
253 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
255 const int rw
= bio_data_dir(bio
);
256 struct mddev
*mddev
= q
->queuedata
;
257 unsigned int sectors
;
260 if (mddev
== NULL
|| mddev
->pers
== NULL
265 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
266 bio_endio(bio
, bio_sectors(bio
) == 0 ? 0 : -EROFS
);
269 smp_rmb(); /* Ensure implications of 'active' are visible */
271 if (mddev
->suspended
) {
274 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
275 TASK_UNINTERRUPTIBLE
);
276 if (!mddev
->suspended
)
282 finish_wait(&mddev
->sb_wait
, &__wait
);
284 atomic_inc(&mddev
->active_io
);
288 * save the sectors now since our bio can
289 * go away inside make_request
291 sectors
= bio_sectors(bio
);
292 mddev
->pers
->make_request(mddev
, bio
);
294 cpu
= part_stat_lock();
295 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
296 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
299 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
300 wake_up(&mddev
->sb_wait
);
303 /* mddev_suspend makes sure no new requests are submitted
304 * to the device, and that any requests that have been submitted
305 * are completely handled.
306 * Once mddev_detach() is called and completes, the module will be
309 void mddev_suspend(struct mddev
*mddev
)
311 BUG_ON(mddev
->suspended
);
312 mddev
->suspended
= 1;
314 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
315 mddev
->pers
->quiesce(mddev
, 1);
317 del_timer_sync(&mddev
->safemode_timer
);
319 EXPORT_SYMBOL_GPL(mddev_suspend
);
321 void mddev_resume(struct mddev
*mddev
)
323 mddev
->suspended
= 0;
324 wake_up(&mddev
->sb_wait
);
325 mddev
->pers
->quiesce(mddev
, 0);
327 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
328 md_wakeup_thread(mddev
->thread
);
329 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
331 EXPORT_SYMBOL_GPL(mddev_resume
);
333 int mddev_congested(struct mddev
*mddev
, int bits
)
335 struct md_personality
*pers
= mddev
->pers
;
339 if (mddev
->suspended
)
341 else if (pers
&& pers
->congested
)
342 ret
= pers
->congested(mddev
, bits
);
346 EXPORT_SYMBOL_GPL(mddev_congested
);
347 static int md_congested(void *data
, int bits
)
349 struct mddev
*mddev
= data
;
350 return mddev_congested(mddev
, bits
);
353 static int md_mergeable_bvec(struct request_queue
*q
,
354 struct bvec_merge_data
*bvm
,
355 struct bio_vec
*biovec
)
357 struct mddev
*mddev
= q
->queuedata
;
360 if (mddev
->suspended
) {
361 /* Must always allow one vec */
362 if (bvm
->bi_size
== 0)
363 ret
= biovec
->bv_len
;
367 struct md_personality
*pers
= mddev
->pers
;
368 if (pers
&& pers
->mergeable_bvec
)
369 ret
= pers
->mergeable_bvec(mddev
, bvm
, biovec
);
371 ret
= biovec
->bv_len
;
377 * Generic flush handling for md
380 static void md_end_flush(struct bio
*bio
, int err
)
382 struct md_rdev
*rdev
= bio
->bi_private
;
383 struct mddev
*mddev
= rdev
->mddev
;
385 rdev_dec_pending(rdev
, mddev
);
387 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
388 /* The pre-request flush has finished */
389 queue_work(md_wq
, &mddev
->flush_work
);
394 static void md_submit_flush_data(struct work_struct
*ws
);
396 static void submit_flushes(struct work_struct
*ws
)
398 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
399 struct md_rdev
*rdev
;
401 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
402 atomic_set(&mddev
->flush_pending
, 1);
404 rdev_for_each_rcu(rdev
, mddev
)
405 if (rdev
->raid_disk
>= 0 &&
406 !test_bit(Faulty
, &rdev
->flags
)) {
407 /* Take two references, one is dropped
408 * when request finishes, one after
409 * we reclaim rcu_read_lock
412 atomic_inc(&rdev
->nr_pending
);
413 atomic_inc(&rdev
->nr_pending
);
415 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
416 bi
->bi_end_io
= md_end_flush
;
417 bi
->bi_private
= rdev
;
418 bi
->bi_bdev
= rdev
->bdev
;
419 atomic_inc(&mddev
->flush_pending
);
420 submit_bio(WRITE_FLUSH
, bi
);
422 rdev_dec_pending(rdev
, mddev
);
425 if (atomic_dec_and_test(&mddev
->flush_pending
))
426 queue_work(md_wq
, &mddev
->flush_work
);
429 static void md_submit_flush_data(struct work_struct
*ws
)
431 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
432 struct bio
*bio
= mddev
->flush_bio
;
434 if (bio
->bi_iter
.bi_size
== 0)
435 /* an empty barrier - all done */
438 bio
->bi_rw
&= ~REQ_FLUSH
;
439 mddev
->pers
->make_request(mddev
, bio
);
442 mddev
->flush_bio
= NULL
;
443 wake_up(&mddev
->sb_wait
);
446 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
448 spin_lock_irq(&mddev
->lock
);
449 wait_event_lock_irq(mddev
->sb_wait
,
452 mddev
->flush_bio
= bio
;
453 spin_unlock_irq(&mddev
->lock
);
455 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
456 queue_work(md_wq
, &mddev
->flush_work
);
458 EXPORT_SYMBOL(md_flush_request
);
460 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
462 struct mddev
*mddev
= cb
->data
;
463 md_wakeup_thread(mddev
->thread
);
466 EXPORT_SYMBOL(md_unplug
);
468 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
470 atomic_inc(&mddev
->active
);
474 static void mddev_delayed_delete(struct work_struct
*ws
);
476 static void mddev_put(struct mddev
*mddev
)
478 struct bio_set
*bs
= NULL
;
480 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
482 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
483 mddev
->ctime
== 0 && !mddev
->hold_active
) {
484 /* Array is not configured at all, and not held active,
486 list_del_init(&mddev
->all_mddevs
);
488 mddev
->bio_set
= NULL
;
489 if (mddev
->gendisk
) {
490 /* We did a probe so need to clean up. Call
491 * queue_work inside the spinlock so that
492 * flush_workqueue() after mddev_find will
493 * succeed in waiting for the work to be done.
495 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
496 queue_work(md_misc_wq
, &mddev
->del_work
);
500 spin_unlock(&all_mddevs_lock
);
505 void mddev_init(struct mddev
*mddev
)
507 mutex_init(&mddev
->open_mutex
);
508 mutex_init(&mddev
->reconfig_mutex
);
509 mutex_init(&mddev
->bitmap_info
.mutex
);
510 INIT_LIST_HEAD(&mddev
->disks
);
511 INIT_LIST_HEAD(&mddev
->all_mddevs
);
512 init_timer(&mddev
->safemode_timer
);
513 atomic_set(&mddev
->active
, 1);
514 atomic_set(&mddev
->openers
, 0);
515 atomic_set(&mddev
->active_io
, 0);
516 spin_lock_init(&mddev
->lock
);
517 atomic_set(&mddev
->flush_pending
, 0);
518 init_waitqueue_head(&mddev
->sb_wait
);
519 init_waitqueue_head(&mddev
->recovery_wait
);
520 mddev
->reshape_position
= MaxSector
;
521 mddev
->reshape_backwards
= 0;
522 mddev
->last_sync_action
= "none";
523 mddev
->resync_min
= 0;
524 mddev
->resync_max
= MaxSector
;
525 mddev
->level
= LEVEL_NONE
;
527 EXPORT_SYMBOL_GPL(mddev_init
);
529 static struct mddev
*mddev_find(dev_t unit
)
531 struct mddev
*mddev
, *new = NULL
;
533 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
534 unit
&= ~((1<<MdpMinorShift
)-1);
537 spin_lock(&all_mddevs_lock
);
540 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
541 if (mddev
->unit
== unit
) {
543 spin_unlock(&all_mddevs_lock
);
549 list_add(&new->all_mddevs
, &all_mddevs
);
550 spin_unlock(&all_mddevs_lock
);
551 new->hold_active
= UNTIL_IOCTL
;
555 /* find an unused unit number */
556 static int next_minor
= 512;
557 int start
= next_minor
;
561 dev
= MKDEV(MD_MAJOR
, next_minor
);
563 if (next_minor
> MINORMASK
)
565 if (next_minor
== start
) {
566 /* Oh dear, all in use. */
567 spin_unlock(&all_mddevs_lock
);
573 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
574 if (mddev
->unit
== dev
) {
580 new->md_minor
= MINOR(dev
);
581 new->hold_active
= UNTIL_STOP
;
582 list_add(&new->all_mddevs
, &all_mddevs
);
583 spin_unlock(&all_mddevs_lock
);
586 spin_unlock(&all_mddevs_lock
);
588 new = kzalloc(sizeof(*new), GFP_KERNEL
);
593 if (MAJOR(unit
) == MD_MAJOR
)
594 new->md_minor
= MINOR(unit
);
596 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
603 static struct attribute_group md_redundancy_group
;
605 void mddev_unlock(struct mddev
*mddev
)
607 if (mddev
->to_remove
) {
608 /* These cannot be removed under reconfig_mutex as
609 * an access to the files will try to take reconfig_mutex
610 * while holding the file unremovable, which leads to
612 * So hold set sysfs_active while the remove in happeing,
613 * and anything else which might set ->to_remove or my
614 * otherwise change the sysfs namespace will fail with
615 * -EBUSY if sysfs_active is still set.
616 * We set sysfs_active under reconfig_mutex and elsewhere
617 * test it under the same mutex to ensure its correct value
620 struct attribute_group
*to_remove
= mddev
->to_remove
;
621 mddev
->to_remove
= NULL
;
622 mddev
->sysfs_active
= 1;
623 mutex_unlock(&mddev
->reconfig_mutex
);
625 if (mddev
->kobj
.sd
) {
626 if (to_remove
!= &md_redundancy_group
)
627 sysfs_remove_group(&mddev
->kobj
, to_remove
);
628 if (mddev
->pers
== NULL
||
629 mddev
->pers
->sync_request
== NULL
) {
630 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
631 if (mddev
->sysfs_action
)
632 sysfs_put(mddev
->sysfs_action
);
633 mddev
->sysfs_action
= NULL
;
636 mddev
->sysfs_active
= 0;
638 mutex_unlock(&mddev
->reconfig_mutex
);
640 /* As we've dropped the mutex we need a spinlock to
641 * make sure the thread doesn't disappear
643 spin_lock(&pers_lock
);
644 md_wakeup_thread(mddev
->thread
);
645 spin_unlock(&pers_lock
);
647 EXPORT_SYMBOL_GPL(mddev_unlock
);
649 struct md_rdev
*md_find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
651 struct md_rdev
*rdev
;
653 rdev_for_each_rcu(rdev
, mddev
)
654 if (rdev
->desc_nr
== nr
)
659 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu
);
661 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
663 struct md_rdev
*rdev
;
665 rdev_for_each(rdev
, mddev
)
666 if (rdev
->bdev
->bd_dev
== dev
)
672 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
674 struct md_rdev
*rdev
;
676 rdev_for_each_rcu(rdev
, mddev
)
677 if (rdev
->bdev
->bd_dev
== dev
)
683 static struct md_personality
*find_pers(int level
, char *clevel
)
685 struct md_personality
*pers
;
686 list_for_each_entry(pers
, &pers_list
, list
) {
687 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
689 if (strcmp(pers
->name
, clevel
)==0)
695 /* return the offset of the super block in 512byte sectors */
696 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
698 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
699 return MD_NEW_SIZE_SECTORS(num_sectors
);
702 static int alloc_disk_sb(struct md_rdev
*rdev
)
704 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
705 if (!rdev
->sb_page
) {
706 printk(KERN_ALERT
"md: out of memory.\n");
713 void md_rdev_clear(struct md_rdev
*rdev
)
716 put_page(rdev
->sb_page
);
718 rdev
->sb_page
= NULL
;
723 put_page(rdev
->bb_page
);
724 rdev
->bb_page
= NULL
;
726 kfree(rdev
->badblocks
.page
);
727 rdev
->badblocks
.page
= NULL
;
729 EXPORT_SYMBOL_GPL(md_rdev_clear
);
731 static void super_written(struct bio
*bio
, int error
)
733 struct md_rdev
*rdev
= bio
->bi_private
;
734 struct mddev
*mddev
= rdev
->mddev
;
736 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
737 printk("md: super_written gets error=%d, uptodate=%d\n",
738 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
739 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
740 md_error(mddev
, rdev
);
743 if (atomic_dec_and_test(&mddev
->pending_writes
))
744 wake_up(&mddev
->sb_wait
);
748 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
749 sector_t sector
, int size
, struct page
*page
)
751 /* write first size bytes of page to sector of rdev
752 * Increment mddev->pending_writes before returning
753 * and decrement it on completion, waking up sb_wait
754 * if zero is reached.
755 * If an error occurred, call md_error
757 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
759 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
760 bio
->bi_iter
.bi_sector
= sector
;
761 bio_add_page(bio
, page
, size
, 0);
762 bio
->bi_private
= rdev
;
763 bio
->bi_end_io
= super_written
;
765 atomic_inc(&mddev
->pending_writes
);
766 submit_bio(WRITE_FLUSH_FUA
, bio
);
769 void md_super_wait(struct mddev
*mddev
)
771 /* wait for all superblock writes that were scheduled to complete */
772 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
775 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
776 struct page
*page
, int rw
, bool metadata_op
)
778 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
781 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
782 rdev
->meta_bdev
: rdev
->bdev
;
784 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
785 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
786 (rdev
->mddev
->reshape_backwards
==
787 (sector
>= rdev
->mddev
->reshape_position
)))
788 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
790 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
791 bio_add_page(bio
, page
, size
, 0);
792 submit_bio_wait(rw
, bio
);
794 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
798 EXPORT_SYMBOL_GPL(sync_page_io
);
800 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
802 char b
[BDEVNAME_SIZE
];
807 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
813 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
814 bdevname(rdev
->bdev
,b
));
818 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
820 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
821 sb1
->set_uuid1
== sb2
->set_uuid1
&&
822 sb1
->set_uuid2
== sb2
->set_uuid2
&&
823 sb1
->set_uuid3
== sb2
->set_uuid3
;
826 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
829 mdp_super_t
*tmp1
, *tmp2
;
831 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
832 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
834 if (!tmp1
|| !tmp2
) {
836 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
844 * nr_disks is not constant
849 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
856 static u32
md_csum_fold(u32 csum
)
858 csum
= (csum
& 0xffff) + (csum
>> 16);
859 return (csum
& 0xffff) + (csum
>> 16);
862 static unsigned int calc_sb_csum(mdp_super_t
*sb
)
865 u32
*sb32
= (u32
*)sb
;
867 unsigned int disk_csum
, csum
;
869 disk_csum
= sb
->sb_csum
;
872 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
874 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
877 /* This used to use csum_partial, which was wrong for several
878 * reasons including that different results are returned on
879 * different architectures. It isn't critical that we get exactly
880 * the same return value as before (we always csum_fold before
881 * testing, and that removes any differences). However as we
882 * know that csum_partial always returned a 16bit value on
883 * alphas, do a fold to maximise conformity to previous behaviour.
885 sb
->sb_csum
= md_csum_fold(disk_csum
);
887 sb
->sb_csum
= disk_csum
;
893 * Handle superblock details.
894 * We want to be able to handle multiple superblock formats
895 * so we have a common interface to them all, and an array of
896 * different handlers.
897 * We rely on user-space to write the initial superblock, and support
898 * reading and updating of superblocks.
899 * Interface methods are:
900 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
901 * loads and validates a superblock on dev.
902 * if refdev != NULL, compare superblocks on both devices
904 * 0 - dev has a superblock that is compatible with refdev
905 * 1 - dev has a superblock that is compatible and newer than refdev
906 * so dev should be used as the refdev in future
907 * -EINVAL superblock incompatible or invalid
908 * -othererror e.g. -EIO
910 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
911 * Verify that dev is acceptable into mddev.
912 * The first time, mddev->raid_disks will be 0, and data from
913 * dev should be merged in. Subsequent calls check that dev
914 * is new enough. Return 0 or -EINVAL
916 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
917 * Update the superblock for rdev with data in mddev
918 * This does not write to disc.
924 struct module
*owner
;
925 int (*load_super
)(struct md_rdev
*rdev
,
926 struct md_rdev
*refdev
,
928 int (*validate_super
)(struct mddev
*mddev
,
929 struct md_rdev
*rdev
);
930 void (*sync_super
)(struct mddev
*mddev
,
931 struct md_rdev
*rdev
);
932 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
933 sector_t num_sectors
);
934 int (*allow_new_offset
)(struct md_rdev
*rdev
,
935 unsigned long long new_offset
);
939 * Check that the given mddev has no bitmap.
941 * This function is called from the run method of all personalities that do not
942 * support bitmaps. It prints an error message and returns non-zero if mddev
943 * has a bitmap. Otherwise, it returns 0.
946 int md_check_no_bitmap(struct mddev
*mddev
)
948 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
950 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
951 mdname(mddev
), mddev
->pers
->name
);
954 EXPORT_SYMBOL(md_check_no_bitmap
);
957 * load_super for 0.90.0
959 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
961 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
966 * Calculate the position of the superblock (512byte sectors),
967 * it's at the end of the disk.
969 * It also happens to be a multiple of 4Kb.
971 rdev
->sb_start
= calc_dev_sboffset(rdev
);
973 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
978 bdevname(rdev
->bdev
, b
);
979 sb
= page_address(rdev
->sb_page
);
981 if (sb
->md_magic
!= MD_SB_MAGIC
) {
982 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
987 if (sb
->major_version
!= 0 ||
988 sb
->minor_version
< 90 ||
989 sb
->minor_version
> 91) {
990 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
991 sb
->major_version
, sb
->minor_version
,
996 if (sb
->raid_disks
<= 0)
999 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1000 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1005 rdev
->preferred_minor
= sb
->md_minor
;
1006 rdev
->data_offset
= 0;
1007 rdev
->new_data_offset
= 0;
1008 rdev
->sb_size
= MD_SB_BYTES
;
1009 rdev
->badblocks
.shift
= -1;
1011 if (sb
->level
== LEVEL_MULTIPATH
)
1014 rdev
->desc_nr
= sb
->this_disk
.number
;
1020 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1021 if (!uuid_equal(refsb
, sb
)) {
1022 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1023 b
, bdevname(refdev
->bdev
,b2
));
1026 if (!sb_equal(refsb
, sb
)) {
1027 printk(KERN_WARNING
"md: %s has same UUID"
1028 " but different superblock to %s\n",
1029 b
, bdevname(refdev
->bdev
, b2
));
1033 ev2
= md_event(refsb
);
1039 rdev
->sectors
= rdev
->sb_start
;
1040 /* Limit to 4TB as metadata cannot record more than that.
1041 * (not needed for Linear and RAID0 as metadata doesn't
1044 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1045 rdev
->sectors
= (2ULL << 32) - 2;
1047 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1048 /* "this cannot possibly happen" ... */
1056 * validate_super for 0.90.0
1058 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1061 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1062 __u64 ev1
= md_event(sb
);
1064 rdev
->raid_disk
= -1;
1065 clear_bit(Faulty
, &rdev
->flags
);
1066 clear_bit(In_sync
, &rdev
->flags
);
1067 clear_bit(Bitmap_sync
, &rdev
->flags
);
1068 clear_bit(WriteMostly
, &rdev
->flags
);
1070 if (mddev
->raid_disks
== 0) {
1071 mddev
->major_version
= 0;
1072 mddev
->minor_version
= sb
->minor_version
;
1073 mddev
->patch_version
= sb
->patch_version
;
1074 mddev
->external
= 0;
1075 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1076 mddev
->ctime
= sb
->ctime
;
1077 mddev
->utime
= sb
->utime
;
1078 mddev
->level
= sb
->level
;
1079 mddev
->clevel
[0] = 0;
1080 mddev
->layout
= sb
->layout
;
1081 mddev
->raid_disks
= sb
->raid_disks
;
1082 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1083 mddev
->events
= ev1
;
1084 mddev
->bitmap_info
.offset
= 0;
1085 mddev
->bitmap_info
.space
= 0;
1086 /* bitmap can use 60 K after the 4K superblocks */
1087 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1088 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1089 mddev
->reshape_backwards
= 0;
1091 if (mddev
->minor_version
>= 91) {
1092 mddev
->reshape_position
= sb
->reshape_position
;
1093 mddev
->delta_disks
= sb
->delta_disks
;
1094 mddev
->new_level
= sb
->new_level
;
1095 mddev
->new_layout
= sb
->new_layout
;
1096 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1097 if (mddev
->delta_disks
< 0)
1098 mddev
->reshape_backwards
= 1;
1100 mddev
->reshape_position
= MaxSector
;
1101 mddev
->delta_disks
= 0;
1102 mddev
->new_level
= mddev
->level
;
1103 mddev
->new_layout
= mddev
->layout
;
1104 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1107 if (sb
->state
& (1<<MD_SB_CLEAN
))
1108 mddev
->recovery_cp
= MaxSector
;
1110 if (sb
->events_hi
== sb
->cp_events_hi
&&
1111 sb
->events_lo
== sb
->cp_events_lo
) {
1112 mddev
->recovery_cp
= sb
->recovery_cp
;
1114 mddev
->recovery_cp
= 0;
1117 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1118 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1119 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1120 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1122 mddev
->max_disks
= MD_SB_DISKS
;
1124 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1125 mddev
->bitmap_info
.file
== NULL
) {
1126 mddev
->bitmap_info
.offset
=
1127 mddev
->bitmap_info
.default_offset
;
1128 mddev
->bitmap_info
.space
=
1129 mddev
->bitmap_info
.default_space
;
1132 } else if (mddev
->pers
== NULL
) {
1133 /* Insist on good event counter while assembling, except
1134 * for spares (which don't need an event count) */
1136 if (sb
->disks
[rdev
->desc_nr
].state
& (
1137 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1138 if (ev1
< mddev
->events
)
1140 } else if (mddev
->bitmap
) {
1141 /* if adding to array with a bitmap, then we can accept an
1142 * older device ... but not too old.
1144 if (ev1
< mddev
->bitmap
->events_cleared
)
1146 if (ev1
< mddev
->events
)
1147 set_bit(Bitmap_sync
, &rdev
->flags
);
1149 if (ev1
< mddev
->events
)
1150 /* just a hot-add of a new device, leave raid_disk at -1 */
1154 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1155 desc
= sb
->disks
+ rdev
->desc_nr
;
1157 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1158 set_bit(Faulty
, &rdev
->flags
);
1159 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1160 desc->raid_disk < mddev->raid_disks */) {
1161 set_bit(In_sync
, &rdev
->flags
);
1162 rdev
->raid_disk
= desc
->raid_disk
;
1163 rdev
->saved_raid_disk
= desc
->raid_disk
;
1164 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1165 /* active but not in sync implies recovery up to
1166 * reshape position. We don't know exactly where
1167 * that is, so set to zero for now */
1168 if (mddev
->minor_version
>= 91) {
1169 rdev
->recovery_offset
= 0;
1170 rdev
->raid_disk
= desc
->raid_disk
;
1173 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1174 set_bit(WriteMostly
, &rdev
->flags
);
1175 } else /* MULTIPATH are always insync */
1176 set_bit(In_sync
, &rdev
->flags
);
1181 * sync_super for 0.90.0
1183 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1186 struct md_rdev
*rdev2
;
1187 int next_spare
= mddev
->raid_disks
;
1189 /* make rdev->sb match mddev data..
1192 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1193 * 3/ any empty disks < next_spare become removed
1195 * disks[0] gets initialised to REMOVED because
1196 * we cannot be sure from other fields if it has
1197 * been initialised or not.
1200 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1202 rdev
->sb_size
= MD_SB_BYTES
;
1204 sb
= page_address(rdev
->sb_page
);
1206 memset(sb
, 0, sizeof(*sb
));
1208 sb
->md_magic
= MD_SB_MAGIC
;
1209 sb
->major_version
= mddev
->major_version
;
1210 sb
->patch_version
= mddev
->patch_version
;
1211 sb
->gvalid_words
= 0; /* ignored */
1212 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1213 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1214 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1215 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1217 sb
->ctime
= mddev
->ctime
;
1218 sb
->level
= mddev
->level
;
1219 sb
->size
= mddev
->dev_sectors
/ 2;
1220 sb
->raid_disks
= mddev
->raid_disks
;
1221 sb
->md_minor
= mddev
->md_minor
;
1222 sb
->not_persistent
= 0;
1223 sb
->utime
= mddev
->utime
;
1225 sb
->events_hi
= (mddev
->events
>>32);
1226 sb
->events_lo
= (u32
)mddev
->events
;
1228 if (mddev
->reshape_position
== MaxSector
)
1229 sb
->minor_version
= 90;
1231 sb
->minor_version
= 91;
1232 sb
->reshape_position
= mddev
->reshape_position
;
1233 sb
->new_level
= mddev
->new_level
;
1234 sb
->delta_disks
= mddev
->delta_disks
;
1235 sb
->new_layout
= mddev
->new_layout
;
1236 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1238 mddev
->minor_version
= sb
->minor_version
;
1241 sb
->recovery_cp
= mddev
->recovery_cp
;
1242 sb
->cp_events_hi
= (mddev
->events
>>32);
1243 sb
->cp_events_lo
= (u32
)mddev
->events
;
1244 if (mddev
->recovery_cp
== MaxSector
)
1245 sb
->state
= (1<< MD_SB_CLEAN
);
1247 sb
->recovery_cp
= 0;
1249 sb
->layout
= mddev
->layout
;
1250 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1252 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1253 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1255 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1256 rdev_for_each(rdev2
, mddev
) {
1259 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1261 if (rdev2
->raid_disk
>= 0 &&
1262 sb
->minor_version
>= 91)
1263 /* we have nowhere to store the recovery_offset,
1264 * but if it is not below the reshape_position,
1265 * we can piggy-back on that.
1268 if (rdev2
->raid_disk
< 0 ||
1269 test_bit(Faulty
, &rdev2
->flags
))
1272 desc_nr
= rdev2
->raid_disk
;
1274 desc_nr
= next_spare
++;
1275 rdev2
->desc_nr
= desc_nr
;
1276 d
= &sb
->disks
[rdev2
->desc_nr
];
1278 d
->number
= rdev2
->desc_nr
;
1279 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1280 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1282 d
->raid_disk
= rdev2
->raid_disk
;
1284 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1285 if (test_bit(Faulty
, &rdev2
->flags
))
1286 d
->state
= (1<<MD_DISK_FAULTY
);
1287 else if (is_active
) {
1288 d
->state
= (1<<MD_DISK_ACTIVE
);
1289 if (test_bit(In_sync
, &rdev2
->flags
))
1290 d
->state
|= (1<<MD_DISK_SYNC
);
1298 if (test_bit(WriteMostly
, &rdev2
->flags
))
1299 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1301 /* now set the "removed" and "faulty" bits on any missing devices */
1302 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1303 mdp_disk_t
*d
= &sb
->disks
[i
];
1304 if (d
->state
== 0 && d
->number
== 0) {
1307 d
->state
= (1<<MD_DISK_REMOVED
);
1308 d
->state
|= (1<<MD_DISK_FAULTY
);
1312 sb
->nr_disks
= nr_disks
;
1313 sb
->active_disks
= active
;
1314 sb
->working_disks
= working
;
1315 sb
->failed_disks
= failed
;
1316 sb
->spare_disks
= spare
;
1318 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1319 sb
->sb_csum
= calc_sb_csum(sb
);
1323 * rdev_size_change for 0.90.0
1325 static unsigned long long
1326 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1328 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1329 return 0; /* component must fit device */
1330 if (rdev
->mddev
->bitmap_info
.offset
)
1331 return 0; /* can't move bitmap */
1332 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1333 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1334 num_sectors
= rdev
->sb_start
;
1335 /* Limit to 4TB as metadata cannot record more than that.
1336 * 4TB == 2^32 KB, or 2*2^32 sectors.
1338 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1339 num_sectors
= (2ULL << 32) - 2;
1340 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1342 md_super_wait(rdev
->mddev
);
1347 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1349 /* non-zero offset changes not possible with v0.90 */
1350 return new_offset
== 0;
1354 * version 1 superblock
1357 static __le32
calc_sb_1_csum(struct mdp_superblock_1
*sb
)
1361 unsigned long long newcsum
;
1362 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1363 __le32
*isuper
= (__le32
*)sb
;
1365 disk_csum
= sb
->sb_csum
;
1368 for (; size
>= 4; size
-= 4)
1369 newcsum
+= le32_to_cpu(*isuper
++);
1372 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1374 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1375 sb
->sb_csum
= disk_csum
;
1376 return cpu_to_le32(csum
);
1379 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1381 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1383 struct mdp_superblock_1
*sb
;
1387 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1391 * Calculate the position of the superblock in 512byte sectors.
1392 * It is always aligned to a 4K boundary and
1393 * depeding on minor_version, it can be:
1394 * 0: At least 8K, but less than 12K, from end of device
1395 * 1: At start of device
1396 * 2: 4K from start of device.
1398 switch(minor_version
) {
1400 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1402 sb_start
&= ~(sector_t
)(4*2-1);
1413 rdev
->sb_start
= sb_start
;
1415 /* superblock is rarely larger than 1K, but it can be larger,
1416 * and it is safe to read 4k, so we do that
1418 ret
= read_disk_sb(rdev
, 4096);
1419 if (ret
) return ret
;
1421 sb
= page_address(rdev
->sb_page
);
1423 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1424 sb
->major_version
!= cpu_to_le32(1) ||
1425 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1426 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1427 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1430 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1431 printk("md: invalid superblock checksum on %s\n",
1432 bdevname(rdev
->bdev
,b
));
1435 if (le64_to_cpu(sb
->data_size
) < 10) {
1436 printk("md: data_size too small on %s\n",
1437 bdevname(rdev
->bdev
,b
));
1442 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1443 /* Some padding is non-zero, might be a new feature */
1446 rdev
->preferred_minor
= 0xffff;
1447 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1448 rdev
->new_data_offset
= rdev
->data_offset
;
1449 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1450 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1451 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1452 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1454 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1455 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1456 if (rdev
->sb_size
& bmask
)
1457 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1460 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1463 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1466 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1469 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1471 if (!rdev
->bb_page
) {
1472 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1476 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1477 rdev
->badblocks
.count
== 0) {
1478 /* need to load the bad block list.
1479 * Currently we limit it to one page.
1485 int sectors
= le16_to_cpu(sb
->bblog_size
);
1486 if (sectors
> (PAGE_SIZE
/ 512))
1488 offset
= le32_to_cpu(sb
->bblog_offset
);
1491 bb_sector
= (long long)offset
;
1492 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1493 rdev
->bb_page
, READ
, true))
1495 bbp
= (u64
*)page_address(rdev
->bb_page
);
1496 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1497 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1498 u64 bb
= le64_to_cpu(*bbp
);
1499 int count
= bb
& (0x3ff);
1500 u64 sector
= bb
>> 10;
1501 sector
<<= sb
->bblog_shift
;
1502 count
<<= sb
->bblog_shift
;
1505 if (md_set_badblocks(&rdev
->badblocks
,
1506 sector
, count
, 1) == 0)
1509 } else if (sb
->bblog_offset
!= 0)
1510 rdev
->badblocks
.shift
= 0;
1516 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1518 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1519 sb
->level
!= refsb
->level
||
1520 sb
->layout
!= refsb
->layout
||
1521 sb
->chunksize
!= refsb
->chunksize
) {
1522 printk(KERN_WARNING
"md: %s has strangely different"
1523 " superblock to %s\n",
1524 bdevname(rdev
->bdev
,b
),
1525 bdevname(refdev
->bdev
,b2
));
1528 ev1
= le64_to_cpu(sb
->events
);
1529 ev2
= le64_to_cpu(refsb
->events
);
1536 if (minor_version
) {
1537 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1538 sectors
-= rdev
->data_offset
;
1540 sectors
= rdev
->sb_start
;
1541 if (sectors
< le64_to_cpu(sb
->data_size
))
1543 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1547 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1549 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1550 __u64 ev1
= le64_to_cpu(sb
->events
);
1552 rdev
->raid_disk
= -1;
1553 clear_bit(Faulty
, &rdev
->flags
);
1554 clear_bit(In_sync
, &rdev
->flags
);
1555 clear_bit(Bitmap_sync
, &rdev
->flags
);
1556 clear_bit(WriteMostly
, &rdev
->flags
);
1558 if (mddev
->raid_disks
== 0) {
1559 mddev
->major_version
= 1;
1560 mddev
->patch_version
= 0;
1561 mddev
->external
= 0;
1562 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1563 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1564 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1565 mddev
->level
= le32_to_cpu(sb
->level
);
1566 mddev
->clevel
[0] = 0;
1567 mddev
->layout
= le32_to_cpu(sb
->layout
);
1568 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1569 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1570 mddev
->events
= ev1
;
1571 mddev
->bitmap_info
.offset
= 0;
1572 mddev
->bitmap_info
.space
= 0;
1573 /* Default location for bitmap is 1K after superblock
1574 * using 3K - total of 4K
1576 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1577 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1578 mddev
->reshape_backwards
= 0;
1580 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1581 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1583 mddev
->max_disks
= (4096-256)/2;
1585 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1586 mddev
->bitmap_info
.file
== NULL
) {
1587 mddev
->bitmap_info
.offset
=
1588 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1589 /* Metadata doesn't record how much space is available.
1590 * For 1.0, we assume we can use up to the superblock
1591 * if before, else to 4K beyond superblock.
1592 * For others, assume no change is possible.
1594 if (mddev
->minor_version
> 0)
1595 mddev
->bitmap_info
.space
= 0;
1596 else if (mddev
->bitmap_info
.offset
> 0)
1597 mddev
->bitmap_info
.space
=
1598 8 - mddev
->bitmap_info
.offset
;
1600 mddev
->bitmap_info
.space
=
1601 -mddev
->bitmap_info
.offset
;
1604 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1605 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1606 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1607 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1608 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1609 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1610 if (mddev
->delta_disks
< 0 ||
1611 (mddev
->delta_disks
== 0 &&
1612 (le32_to_cpu(sb
->feature_map
)
1613 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1614 mddev
->reshape_backwards
= 1;
1616 mddev
->reshape_position
= MaxSector
;
1617 mddev
->delta_disks
= 0;
1618 mddev
->new_level
= mddev
->level
;
1619 mddev
->new_layout
= mddev
->layout
;
1620 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1623 } else if (mddev
->pers
== NULL
) {
1624 /* Insist of good event counter while assembling, except for
1625 * spares (which don't need an event count) */
1627 if (rdev
->desc_nr
>= 0 &&
1628 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1629 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1630 if (ev1
< mddev
->events
)
1632 } else if (mddev
->bitmap
) {
1633 /* If adding to array with a bitmap, then we can accept an
1634 * older device, but not too old.
1636 if (ev1
< mddev
->bitmap
->events_cleared
)
1638 if (ev1
< mddev
->events
)
1639 set_bit(Bitmap_sync
, &rdev
->flags
);
1641 if (ev1
< mddev
->events
)
1642 /* just a hot-add of a new device, leave raid_disk at -1 */
1645 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1647 if (rdev
->desc_nr
< 0 ||
1648 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1652 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1654 case 0xffff: /* spare */
1656 case 0xfffe: /* faulty */
1657 set_bit(Faulty
, &rdev
->flags
);
1660 rdev
->saved_raid_disk
= role
;
1661 if ((le32_to_cpu(sb
->feature_map
) &
1662 MD_FEATURE_RECOVERY_OFFSET
)) {
1663 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1664 if (!(le32_to_cpu(sb
->feature_map
) &
1665 MD_FEATURE_RECOVERY_BITMAP
))
1666 rdev
->saved_raid_disk
= -1;
1668 set_bit(In_sync
, &rdev
->flags
);
1669 rdev
->raid_disk
= role
;
1672 if (sb
->devflags
& WriteMostly1
)
1673 set_bit(WriteMostly
, &rdev
->flags
);
1674 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1675 set_bit(Replacement
, &rdev
->flags
);
1676 } else /* MULTIPATH are always insync */
1677 set_bit(In_sync
, &rdev
->flags
);
1682 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1684 struct mdp_superblock_1
*sb
;
1685 struct md_rdev
*rdev2
;
1687 /* make rdev->sb match mddev and rdev data. */
1689 sb
= page_address(rdev
->sb_page
);
1691 sb
->feature_map
= 0;
1693 sb
->recovery_offset
= cpu_to_le64(0);
1694 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1696 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1697 sb
->events
= cpu_to_le64(mddev
->events
);
1699 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1701 sb
->resync_offset
= cpu_to_le64(0);
1703 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1705 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1706 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1707 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1708 sb
->level
= cpu_to_le32(mddev
->level
);
1709 sb
->layout
= cpu_to_le32(mddev
->layout
);
1711 if (test_bit(WriteMostly
, &rdev
->flags
))
1712 sb
->devflags
|= WriteMostly1
;
1714 sb
->devflags
&= ~WriteMostly1
;
1715 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1716 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1718 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1719 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1720 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1723 if (rdev
->raid_disk
>= 0 &&
1724 !test_bit(In_sync
, &rdev
->flags
)) {
1726 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1727 sb
->recovery_offset
=
1728 cpu_to_le64(rdev
->recovery_offset
);
1729 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1731 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1733 if (test_bit(Replacement
, &rdev
->flags
))
1735 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1737 if (mddev
->reshape_position
!= MaxSector
) {
1738 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1739 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1740 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1741 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1742 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1743 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1744 if (mddev
->delta_disks
== 0 &&
1745 mddev
->reshape_backwards
)
1747 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1748 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1750 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1751 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1752 - rdev
->data_offset
));
1756 if (rdev
->badblocks
.count
== 0)
1757 /* Nothing to do for bad blocks*/ ;
1758 else if (sb
->bblog_offset
== 0)
1759 /* Cannot record bad blocks on this device */
1760 md_error(mddev
, rdev
);
1762 struct badblocks
*bb
= &rdev
->badblocks
;
1763 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1765 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1770 seq
= read_seqbegin(&bb
->lock
);
1772 memset(bbp
, 0xff, PAGE_SIZE
);
1774 for (i
= 0 ; i
< bb
->count
; i
++) {
1775 u64 internal_bb
= p
[i
];
1776 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1777 | BB_LEN(internal_bb
));
1778 bbp
[i
] = cpu_to_le64(store_bb
);
1781 if (read_seqretry(&bb
->lock
, seq
))
1784 bb
->sector
= (rdev
->sb_start
+
1785 (int)le32_to_cpu(sb
->bblog_offset
));
1786 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1791 rdev_for_each(rdev2
, mddev
)
1792 if (rdev2
->desc_nr
+1 > max_dev
)
1793 max_dev
= rdev2
->desc_nr
+1;
1795 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1797 sb
->max_dev
= cpu_to_le32(max_dev
);
1798 rdev
->sb_size
= max_dev
* 2 + 256;
1799 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1800 if (rdev
->sb_size
& bmask
)
1801 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1803 max_dev
= le32_to_cpu(sb
->max_dev
);
1805 for (i
=0; i
<max_dev
;i
++)
1806 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1808 rdev_for_each(rdev2
, mddev
) {
1810 if (test_bit(Faulty
, &rdev2
->flags
))
1811 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1812 else if (test_bit(In_sync
, &rdev2
->flags
))
1813 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1814 else if (rdev2
->raid_disk
>= 0)
1815 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1817 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1820 sb
->sb_csum
= calc_sb_1_csum(sb
);
1823 static unsigned long long
1824 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1826 struct mdp_superblock_1
*sb
;
1827 sector_t max_sectors
;
1828 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1829 return 0; /* component must fit device */
1830 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1831 return 0; /* too confusing */
1832 if (rdev
->sb_start
< rdev
->data_offset
) {
1833 /* minor versions 1 and 2; superblock before data */
1834 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1835 max_sectors
-= rdev
->data_offset
;
1836 if (!num_sectors
|| num_sectors
> max_sectors
)
1837 num_sectors
= max_sectors
;
1838 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1839 /* minor version 0 with bitmap we can't move */
1842 /* minor version 0; superblock after data */
1844 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1845 sb_start
&= ~(sector_t
)(4*2 - 1);
1846 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1847 if (!num_sectors
|| num_sectors
> max_sectors
)
1848 num_sectors
= max_sectors
;
1849 rdev
->sb_start
= sb_start
;
1851 sb
= page_address(rdev
->sb_page
);
1852 sb
->data_size
= cpu_to_le64(num_sectors
);
1853 sb
->super_offset
= rdev
->sb_start
;
1854 sb
->sb_csum
= calc_sb_1_csum(sb
);
1855 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1857 md_super_wait(rdev
->mddev
);
1863 super_1_allow_new_offset(struct md_rdev
*rdev
,
1864 unsigned long long new_offset
)
1866 /* All necessary checks on new >= old have been done */
1867 struct bitmap
*bitmap
;
1868 if (new_offset
>= rdev
->data_offset
)
1871 /* with 1.0 metadata, there is no metadata to tread on
1872 * so we can always move back */
1873 if (rdev
->mddev
->minor_version
== 0)
1876 /* otherwise we must be sure not to step on
1877 * any metadata, so stay:
1878 * 36K beyond start of superblock
1879 * beyond end of badblocks
1880 * beyond write-intent bitmap
1882 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1884 bitmap
= rdev
->mddev
->bitmap
;
1885 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1886 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1887 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1889 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1895 static struct super_type super_types
[] = {
1898 .owner
= THIS_MODULE
,
1899 .load_super
= super_90_load
,
1900 .validate_super
= super_90_validate
,
1901 .sync_super
= super_90_sync
,
1902 .rdev_size_change
= super_90_rdev_size_change
,
1903 .allow_new_offset
= super_90_allow_new_offset
,
1907 .owner
= THIS_MODULE
,
1908 .load_super
= super_1_load
,
1909 .validate_super
= super_1_validate
,
1910 .sync_super
= super_1_sync
,
1911 .rdev_size_change
= super_1_rdev_size_change
,
1912 .allow_new_offset
= super_1_allow_new_offset
,
1916 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1918 if (mddev
->sync_super
) {
1919 mddev
->sync_super(mddev
, rdev
);
1923 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1925 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1928 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1930 struct md_rdev
*rdev
, *rdev2
;
1933 rdev_for_each_rcu(rdev
, mddev1
)
1934 rdev_for_each_rcu(rdev2
, mddev2
)
1935 if (rdev
->bdev
->bd_contains
==
1936 rdev2
->bdev
->bd_contains
) {
1944 static LIST_HEAD(pending_raid_disks
);
1947 * Try to register data integrity profile for an mddev
1949 * This is called when an array is started and after a disk has been kicked
1950 * from the array. It only succeeds if all working and active component devices
1951 * are integrity capable with matching profiles.
1953 int md_integrity_register(struct mddev
*mddev
)
1955 struct md_rdev
*rdev
, *reference
= NULL
;
1957 if (list_empty(&mddev
->disks
))
1958 return 0; /* nothing to do */
1959 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1960 return 0; /* shouldn't register, or already is */
1961 rdev_for_each(rdev
, mddev
) {
1962 /* skip spares and non-functional disks */
1963 if (test_bit(Faulty
, &rdev
->flags
))
1965 if (rdev
->raid_disk
< 0)
1968 /* Use the first rdev as the reference */
1972 /* does this rdev's profile match the reference profile? */
1973 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1974 rdev
->bdev
->bd_disk
) < 0)
1977 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1980 * All component devices are integrity capable and have matching
1981 * profiles, register the common profile for the md device.
1983 if (blk_integrity_register(mddev
->gendisk
,
1984 bdev_get_integrity(reference
->bdev
)) != 0) {
1985 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1989 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1990 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1991 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1997 EXPORT_SYMBOL(md_integrity_register
);
1999 /* Disable data integrity if non-capable/non-matching disk is being added */
2000 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2002 struct blk_integrity
*bi_rdev
;
2003 struct blk_integrity
*bi_mddev
;
2005 if (!mddev
->gendisk
)
2008 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2009 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2011 if (!bi_mddev
) /* nothing to do */
2013 if (rdev
->raid_disk
< 0) /* skip spares */
2015 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2016 rdev
->bdev
->bd_disk
) >= 0)
2018 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2019 blk_integrity_unregister(mddev
->gendisk
);
2021 EXPORT_SYMBOL(md_integrity_add_rdev
);
2023 static int bind_rdev_to_array(struct md_rdev
*rdev
, struct mddev
*mddev
)
2025 char b
[BDEVNAME_SIZE
];
2029 /* prevent duplicates */
2030 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2033 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2034 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2035 rdev
->sectors
< mddev
->dev_sectors
)) {
2037 /* Cannot change size, so fail
2038 * If mddev->level <= 0, then we don't care
2039 * about aligning sizes (e.g. linear)
2041 if (mddev
->level
> 0)
2044 mddev
->dev_sectors
= rdev
->sectors
;
2047 /* Verify rdev->desc_nr is unique.
2048 * If it is -1, assign a free number, else
2049 * check number is not in use
2052 if (rdev
->desc_nr
< 0) {
2055 choice
= mddev
->raid_disks
;
2056 while (md_find_rdev_nr_rcu(mddev
, choice
))
2058 rdev
->desc_nr
= choice
;
2060 if (md_find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2066 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2067 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2068 mdname(mddev
), mddev
->max_disks
);
2071 bdevname(rdev
->bdev
,b
);
2072 strreplace(b
, '/', '!');
2074 rdev
->mddev
= mddev
;
2075 printk(KERN_INFO
"md: bind<%s>\n", b
);
2077 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2080 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2081 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2082 /* failure here is OK */;
2083 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2085 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2086 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2088 /* May as well allow recovery to be retried once */
2089 mddev
->recovery_disabled
++;
2094 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2099 static void md_delayed_delete(struct work_struct
*ws
)
2101 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2102 kobject_del(&rdev
->kobj
);
2103 kobject_put(&rdev
->kobj
);
2106 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2108 char b
[BDEVNAME_SIZE
];
2110 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2111 list_del_rcu(&rdev
->same_set
);
2112 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2114 sysfs_remove_link(&rdev
->kobj
, "block");
2115 sysfs_put(rdev
->sysfs_state
);
2116 rdev
->sysfs_state
= NULL
;
2117 rdev
->badblocks
.count
= 0;
2118 /* We need to delay this, otherwise we can deadlock when
2119 * writing to 'remove' to "dev/state". We also need
2120 * to delay it due to rcu usage.
2123 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2124 kobject_get(&rdev
->kobj
);
2125 queue_work(md_misc_wq
, &rdev
->del_work
);
2129 * prevent the device from being mounted, repartitioned or
2130 * otherwise reused by a RAID array (or any other kernel
2131 * subsystem), by bd_claiming the device.
2133 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2136 struct block_device
*bdev
;
2137 char b
[BDEVNAME_SIZE
];
2139 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2140 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2142 printk(KERN_ERR
"md: could not open %s.\n",
2143 __bdevname(dev
, b
));
2144 return PTR_ERR(bdev
);
2150 static void unlock_rdev(struct md_rdev
*rdev
)
2152 struct block_device
*bdev
= rdev
->bdev
;
2154 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2157 void md_autodetect_dev(dev_t dev
);
2159 static void export_rdev(struct md_rdev
*rdev
)
2161 char b
[BDEVNAME_SIZE
];
2163 printk(KERN_INFO
"md: export_rdev(%s)\n",
2164 bdevname(rdev
->bdev
,b
));
2165 md_rdev_clear(rdev
);
2167 if (test_bit(AutoDetected
, &rdev
->flags
))
2168 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2171 kobject_put(&rdev
->kobj
);
2174 void md_kick_rdev_from_array(struct md_rdev
*rdev
)
2176 unbind_rdev_from_array(rdev
);
2179 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array
);
2181 static void export_array(struct mddev
*mddev
)
2183 struct md_rdev
*rdev
;
2185 while (!list_empty(&mddev
->disks
)) {
2186 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2188 md_kick_rdev_from_array(rdev
);
2190 mddev
->raid_disks
= 0;
2191 mddev
->major_version
= 0;
2194 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2196 /* Update each superblock (in-memory image), but
2197 * if we are allowed to, skip spares which already
2198 * have the right event counter, or have one earlier
2199 * (which would mean they aren't being marked as dirty
2200 * with the rest of the array)
2202 struct md_rdev
*rdev
;
2203 rdev_for_each(rdev
, mddev
) {
2204 if (rdev
->sb_events
== mddev
->events
||
2206 rdev
->raid_disk
< 0 &&
2207 rdev
->sb_events
+1 == mddev
->events
)) {
2208 /* Don't update this superblock */
2209 rdev
->sb_loaded
= 2;
2211 sync_super(mddev
, rdev
);
2212 rdev
->sb_loaded
= 1;
2217 void md_update_sb(struct mddev
*mddev
, int force_change
)
2219 struct md_rdev
*rdev
;
2222 int any_badblocks_changed
= 0;
2226 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2230 /* First make sure individual recovery_offsets are correct */
2231 rdev_for_each(rdev
, mddev
) {
2232 if (rdev
->raid_disk
>= 0 &&
2233 mddev
->delta_disks
>= 0 &&
2234 !test_bit(In_sync
, &rdev
->flags
) &&
2235 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2236 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2239 if (!mddev
->persistent
) {
2240 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2241 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2242 if (!mddev
->external
) {
2243 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2244 rdev_for_each(rdev
, mddev
) {
2245 if (rdev
->badblocks
.changed
) {
2246 rdev
->badblocks
.changed
= 0;
2247 md_ack_all_badblocks(&rdev
->badblocks
);
2248 md_error(mddev
, rdev
);
2250 clear_bit(Blocked
, &rdev
->flags
);
2251 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2252 wake_up(&rdev
->blocked_wait
);
2255 wake_up(&mddev
->sb_wait
);
2259 spin_lock(&mddev
->lock
);
2261 mddev
->utime
= get_seconds();
2263 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2265 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2266 /* just a clean<-> dirty transition, possibly leave spares alone,
2267 * though if events isn't the right even/odd, we will have to do
2273 if (mddev
->degraded
)
2274 /* If the array is degraded, then skipping spares is both
2275 * dangerous and fairly pointless.
2276 * Dangerous because a device that was removed from the array
2277 * might have a event_count that still looks up-to-date,
2278 * so it can be re-added without a resync.
2279 * Pointless because if there are any spares to skip,
2280 * then a recovery will happen and soon that array won't
2281 * be degraded any more and the spare can go back to sleep then.
2285 sync_req
= mddev
->in_sync
;
2287 /* If this is just a dirty<->clean transition, and the array is clean
2288 * and 'events' is odd, we can roll back to the previous clean state */
2290 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2291 && mddev
->can_decrease_events
2292 && mddev
->events
!= 1) {
2294 mddev
->can_decrease_events
= 0;
2296 /* otherwise we have to go forward and ... */
2298 mddev
->can_decrease_events
= nospares
;
2302 * This 64-bit counter should never wrap.
2303 * Either we are in around ~1 trillion A.C., assuming
2304 * 1 reboot per second, or we have a bug...
2306 WARN_ON(mddev
->events
== 0);
2308 rdev_for_each(rdev
, mddev
) {
2309 if (rdev
->badblocks
.changed
)
2310 any_badblocks_changed
++;
2311 if (test_bit(Faulty
, &rdev
->flags
))
2312 set_bit(FaultRecorded
, &rdev
->flags
);
2315 sync_sbs(mddev
, nospares
);
2316 spin_unlock(&mddev
->lock
);
2318 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2319 mdname(mddev
), mddev
->in_sync
);
2321 bitmap_update_sb(mddev
->bitmap
);
2322 rdev_for_each(rdev
, mddev
) {
2323 char b
[BDEVNAME_SIZE
];
2325 if (rdev
->sb_loaded
!= 1)
2326 continue; /* no noise on spare devices */
2328 if (!test_bit(Faulty
, &rdev
->flags
)) {
2329 md_super_write(mddev
,rdev
,
2330 rdev
->sb_start
, rdev
->sb_size
,
2332 pr_debug("md: (write) %s's sb offset: %llu\n",
2333 bdevname(rdev
->bdev
, b
),
2334 (unsigned long long)rdev
->sb_start
);
2335 rdev
->sb_events
= mddev
->events
;
2336 if (rdev
->badblocks
.size
) {
2337 md_super_write(mddev
, rdev
,
2338 rdev
->badblocks
.sector
,
2339 rdev
->badblocks
.size
<< 9,
2341 rdev
->badblocks
.size
= 0;
2345 pr_debug("md: %s (skipping faulty)\n",
2346 bdevname(rdev
->bdev
, b
));
2348 if (mddev
->level
== LEVEL_MULTIPATH
)
2349 /* only need to write one superblock... */
2352 md_super_wait(mddev
);
2353 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2355 spin_lock(&mddev
->lock
);
2356 if (mddev
->in_sync
!= sync_req
||
2357 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2358 /* have to write it out again */
2359 spin_unlock(&mddev
->lock
);
2362 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2363 spin_unlock(&mddev
->lock
);
2364 wake_up(&mddev
->sb_wait
);
2365 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2366 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2368 rdev_for_each(rdev
, mddev
) {
2369 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2370 clear_bit(Blocked
, &rdev
->flags
);
2372 if (any_badblocks_changed
)
2373 md_ack_all_badblocks(&rdev
->badblocks
);
2374 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2375 wake_up(&rdev
->blocked_wait
);
2378 EXPORT_SYMBOL(md_update_sb
);
2380 static int add_bound_rdev(struct md_rdev
*rdev
)
2382 struct mddev
*mddev
= rdev
->mddev
;
2385 if (!mddev
->pers
->hot_remove_disk
) {
2386 /* If there is hot_add_disk but no hot_remove_disk
2387 * then added disks for geometry changes,
2388 * and should be added immediately.
2390 super_types
[mddev
->major_version
].
2391 validate_super(mddev
, rdev
);
2392 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
2394 unbind_rdev_from_array(rdev
);
2399 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2401 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2402 if (mddev
->degraded
)
2403 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2404 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2405 md_new_event(mddev
);
2406 md_wakeup_thread(mddev
->thread
);
2410 /* words written to sysfs files may, or may not, be \n terminated.
2411 * We want to accept with case. For this we use cmd_match.
2413 static int cmd_match(const char *cmd
, const char *str
)
2415 /* See if cmd, written into a sysfs file, matches
2416 * str. They must either be the same, or cmd can
2417 * have a trailing newline
2419 while (*cmd
&& *str
&& *cmd
== *str
) {
2430 struct rdev_sysfs_entry
{
2431 struct attribute attr
;
2432 ssize_t (*show
)(struct md_rdev
*, char *);
2433 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2437 state_show(struct md_rdev
*rdev
, char *page
)
2441 unsigned long flags
= ACCESS_ONCE(rdev
->flags
);
2443 if (test_bit(Faulty
, &flags
) ||
2444 rdev
->badblocks
.unacked_exist
) {
2445 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2448 if (test_bit(In_sync
, &flags
)) {
2449 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2452 if (test_bit(WriteMostly
, &flags
)) {
2453 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2456 if (test_bit(Blocked
, &flags
) ||
2457 (rdev
->badblocks
.unacked_exist
2458 && !test_bit(Faulty
, &flags
))) {
2459 len
+= sprintf(page
+len
, "%sblocked", sep
);
2462 if (!test_bit(Faulty
, &flags
) &&
2463 !test_bit(In_sync
, &flags
)) {
2464 len
+= sprintf(page
+len
, "%sspare", sep
);
2467 if (test_bit(WriteErrorSeen
, &flags
)) {
2468 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2471 if (test_bit(WantReplacement
, &flags
)) {
2472 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2475 if (test_bit(Replacement
, &flags
)) {
2476 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2480 return len
+sprintf(page
+len
, "\n");
2484 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2487 * faulty - simulates an error
2488 * remove - disconnects the device
2489 * writemostly - sets write_mostly
2490 * -writemostly - clears write_mostly
2491 * blocked - sets the Blocked flags
2492 * -blocked - clears the Blocked and possibly simulates an error
2493 * insync - sets Insync providing device isn't active
2494 * -insync - clear Insync for a device with a slot assigned,
2495 * so that it gets rebuilt based on bitmap
2496 * write_error - sets WriteErrorSeen
2497 * -write_error - clears WriteErrorSeen
2500 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2501 md_error(rdev
->mddev
, rdev
);
2502 if (test_bit(Faulty
, &rdev
->flags
))
2506 } else if (cmd_match(buf
, "remove")) {
2507 if (rdev
->raid_disk
>= 0)
2510 struct mddev
*mddev
= rdev
->mddev
;
2511 if (mddev_is_clustered(mddev
))
2512 md_cluster_ops
->remove_disk(mddev
, rdev
);
2513 md_kick_rdev_from_array(rdev
);
2514 if (mddev_is_clustered(mddev
))
2515 md_cluster_ops
->metadata_update_start(mddev
);
2517 md_update_sb(mddev
, 1);
2518 md_new_event(mddev
);
2519 if (mddev_is_clustered(mddev
))
2520 md_cluster_ops
->metadata_update_finish(mddev
);
2523 } else if (cmd_match(buf
, "writemostly")) {
2524 set_bit(WriteMostly
, &rdev
->flags
);
2526 } else if (cmd_match(buf
, "-writemostly")) {
2527 clear_bit(WriteMostly
, &rdev
->flags
);
2529 } else if (cmd_match(buf
, "blocked")) {
2530 set_bit(Blocked
, &rdev
->flags
);
2532 } else if (cmd_match(buf
, "-blocked")) {
2533 if (!test_bit(Faulty
, &rdev
->flags
) &&
2534 rdev
->badblocks
.unacked_exist
) {
2535 /* metadata handler doesn't understand badblocks,
2536 * so we need to fail the device
2538 md_error(rdev
->mddev
, rdev
);
2540 clear_bit(Blocked
, &rdev
->flags
);
2541 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2542 wake_up(&rdev
->blocked_wait
);
2543 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2544 md_wakeup_thread(rdev
->mddev
->thread
);
2547 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2548 set_bit(In_sync
, &rdev
->flags
);
2550 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0) {
2551 if (rdev
->mddev
->pers
== NULL
) {
2552 clear_bit(In_sync
, &rdev
->flags
);
2553 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2554 rdev
->raid_disk
= -1;
2557 } else if (cmd_match(buf
, "write_error")) {
2558 set_bit(WriteErrorSeen
, &rdev
->flags
);
2560 } else if (cmd_match(buf
, "-write_error")) {
2561 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2563 } else if (cmd_match(buf
, "want_replacement")) {
2564 /* Any non-spare device that is not a replacement can
2565 * become want_replacement at any time, but we then need to
2566 * check if recovery is needed.
2568 if (rdev
->raid_disk
>= 0 &&
2569 !test_bit(Replacement
, &rdev
->flags
))
2570 set_bit(WantReplacement
, &rdev
->flags
);
2571 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2572 md_wakeup_thread(rdev
->mddev
->thread
);
2574 } else if (cmd_match(buf
, "-want_replacement")) {
2575 /* Clearing 'want_replacement' is always allowed.
2576 * Once replacements starts it is too late though.
2579 clear_bit(WantReplacement
, &rdev
->flags
);
2580 } else if (cmd_match(buf
, "replacement")) {
2581 /* Can only set a device as a replacement when array has not
2582 * yet been started. Once running, replacement is automatic
2583 * from spares, or by assigning 'slot'.
2585 if (rdev
->mddev
->pers
)
2588 set_bit(Replacement
, &rdev
->flags
);
2591 } else if (cmd_match(buf
, "-replacement")) {
2592 /* Similarly, can only clear Replacement before start */
2593 if (rdev
->mddev
->pers
)
2596 clear_bit(Replacement
, &rdev
->flags
);
2599 } else if (cmd_match(buf
, "re-add")) {
2600 if (test_bit(Faulty
, &rdev
->flags
) && (rdev
->raid_disk
== -1)) {
2601 /* clear_bit is performed _after_ all the devices
2602 * have their local Faulty bit cleared. If any writes
2603 * happen in the meantime in the local node, they
2604 * will land in the local bitmap, which will be synced
2605 * by this node eventually
2607 if (!mddev_is_clustered(rdev
->mddev
) ||
2608 (err
= md_cluster_ops
->gather_bitmaps(rdev
)) == 0) {
2609 clear_bit(Faulty
, &rdev
->flags
);
2610 err
= add_bound_rdev(rdev
);
2616 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2617 return err
? err
: len
;
2619 static struct rdev_sysfs_entry rdev_state
=
2620 __ATTR_PREALLOC(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2623 errors_show(struct md_rdev
*rdev
, char *page
)
2625 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2629 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2634 rv
= kstrtouint(buf
, 10, &n
);
2637 atomic_set(&rdev
->corrected_errors
, n
);
2640 static struct rdev_sysfs_entry rdev_errors
=
2641 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2644 slot_show(struct md_rdev
*rdev
, char *page
)
2646 if (rdev
->raid_disk
< 0)
2647 return sprintf(page
, "none\n");
2649 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2653 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2658 if (strncmp(buf
, "none", 4)==0)
2661 err
= kstrtouint(buf
, 10, (unsigned int *)&slot
);
2665 if (rdev
->mddev
->pers
&& slot
== -1) {
2666 /* Setting 'slot' on an active array requires also
2667 * updating the 'rd%d' link, and communicating
2668 * with the personality with ->hot_*_disk.
2669 * For now we only support removing
2670 * failed/spare devices. This normally happens automatically,
2671 * but not when the metadata is externally managed.
2673 if (rdev
->raid_disk
== -1)
2675 /* personality does all needed checks */
2676 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2678 clear_bit(Blocked
, &rdev
->flags
);
2679 remove_and_add_spares(rdev
->mddev
, rdev
);
2680 if (rdev
->raid_disk
>= 0)
2682 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2683 md_wakeup_thread(rdev
->mddev
->thread
);
2684 } else if (rdev
->mddev
->pers
) {
2685 /* Activating a spare .. or possibly reactivating
2686 * if we ever get bitmaps working here.
2689 if (rdev
->raid_disk
!= -1)
2692 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2695 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2698 if (slot
>= rdev
->mddev
->raid_disks
&&
2699 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2702 rdev
->raid_disk
= slot
;
2703 if (test_bit(In_sync
, &rdev
->flags
))
2704 rdev
->saved_raid_disk
= slot
;
2706 rdev
->saved_raid_disk
= -1;
2707 clear_bit(In_sync
, &rdev
->flags
);
2708 clear_bit(Bitmap_sync
, &rdev
->flags
);
2709 err
= rdev
->mddev
->pers
->
2710 hot_add_disk(rdev
->mddev
, rdev
);
2712 rdev
->raid_disk
= -1;
2715 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2716 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2717 /* failure here is OK */;
2718 /* don't wakeup anyone, leave that to userspace. */
2720 if (slot
>= rdev
->mddev
->raid_disks
&&
2721 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2723 rdev
->raid_disk
= slot
;
2724 /* assume it is working */
2725 clear_bit(Faulty
, &rdev
->flags
);
2726 clear_bit(WriteMostly
, &rdev
->flags
);
2727 set_bit(In_sync
, &rdev
->flags
);
2728 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2733 static struct rdev_sysfs_entry rdev_slot
=
2734 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2737 offset_show(struct md_rdev
*rdev
, char *page
)
2739 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2743 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2745 unsigned long long offset
;
2746 if (kstrtoull(buf
, 10, &offset
) < 0)
2748 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2750 if (rdev
->sectors
&& rdev
->mddev
->external
)
2751 /* Must set offset before size, so overlap checks
2754 rdev
->data_offset
= offset
;
2755 rdev
->new_data_offset
= offset
;
2759 static struct rdev_sysfs_entry rdev_offset
=
2760 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2762 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2764 return sprintf(page
, "%llu\n",
2765 (unsigned long long)rdev
->new_data_offset
);
2768 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2769 const char *buf
, size_t len
)
2771 unsigned long long new_offset
;
2772 struct mddev
*mddev
= rdev
->mddev
;
2774 if (kstrtoull(buf
, 10, &new_offset
) < 0)
2777 if (mddev
->sync_thread
||
2778 test_bit(MD_RECOVERY_RUNNING
,&mddev
->recovery
))
2780 if (new_offset
== rdev
->data_offset
)
2781 /* reset is always permitted */
2783 else if (new_offset
> rdev
->data_offset
) {
2784 /* must not push array size beyond rdev_sectors */
2785 if (new_offset
- rdev
->data_offset
2786 + mddev
->dev_sectors
> rdev
->sectors
)
2789 /* Metadata worries about other space details. */
2791 /* decreasing the offset is inconsistent with a backwards
2794 if (new_offset
< rdev
->data_offset
&&
2795 mddev
->reshape_backwards
)
2797 /* Increasing offset is inconsistent with forwards
2798 * reshape. reshape_direction should be set to
2799 * 'backwards' first.
2801 if (new_offset
> rdev
->data_offset
&&
2802 !mddev
->reshape_backwards
)
2805 if (mddev
->pers
&& mddev
->persistent
&&
2806 !super_types
[mddev
->major_version
]
2807 .allow_new_offset(rdev
, new_offset
))
2809 rdev
->new_data_offset
= new_offset
;
2810 if (new_offset
> rdev
->data_offset
)
2811 mddev
->reshape_backwards
= 1;
2812 else if (new_offset
< rdev
->data_offset
)
2813 mddev
->reshape_backwards
= 0;
2817 static struct rdev_sysfs_entry rdev_new_offset
=
2818 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2821 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2823 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2826 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2828 /* check if two start/length pairs overlap */
2836 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2838 unsigned long long blocks
;
2841 if (kstrtoull(buf
, 10, &blocks
) < 0)
2844 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2845 return -EINVAL
; /* sector conversion overflow */
2848 if (new != blocks
* 2)
2849 return -EINVAL
; /* unsigned long long to sector_t overflow */
2856 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2858 struct mddev
*my_mddev
= rdev
->mddev
;
2859 sector_t oldsectors
= rdev
->sectors
;
2862 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2864 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2865 return -EINVAL
; /* too confusing */
2866 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2867 if (my_mddev
->persistent
) {
2868 sectors
= super_types
[my_mddev
->major_version
].
2869 rdev_size_change(rdev
, sectors
);
2872 } else if (!sectors
)
2873 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2875 if (!my_mddev
->pers
->resize
)
2876 /* Cannot change size for RAID0 or Linear etc */
2879 if (sectors
< my_mddev
->dev_sectors
)
2880 return -EINVAL
; /* component must fit device */
2882 rdev
->sectors
= sectors
;
2883 if (sectors
> oldsectors
&& my_mddev
->external
) {
2884 /* Need to check that all other rdevs with the same
2885 * ->bdev do not overlap. 'rcu' is sufficient to walk
2886 * the rdev lists safely.
2887 * This check does not provide a hard guarantee, it
2888 * just helps avoid dangerous mistakes.
2890 struct mddev
*mddev
;
2892 struct list_head
*tmp
;
2895 for_each_mddev(mddev
, tmp
) {
2896 struct md_rdev
*rdev2
;
2898 rdev_for_each(rdev2
, mddev
)
2899 if (rdev
->bdev
== rdev2
->bdev
&&
2901 overlaps(rdev
->data_offset
, rdev
->sectors
,
2914 /* Someone else could have slipped in a size
2915 * change here, but doing so is just silly.
2916 * We put oldsectors back because we *know* it is
2917 * safe, and trust userspace not to race with
2920 rdev
->sectors
= oldsectors
;
2927 static struct rdev_sysfs_entry rdev_size
=
2928 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2930 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2932 unsigned long long recovery_start
= rdev
->recovery_offset
;
2934 if (test_bit(In_sync
, &rdev
->flags
) ||
2935 recovery_start
== MaxSector
)
2936 return sprintf(page
, "none\n");
2938 return sprintf(page
, "%llu\n", recovery_start
);
2941 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2943 unsigned long long recovery_start
;
2945 if (cmd_match(buf
, "none"))
2946 recovery_start
= MaxSector
;
2947 else if (kstrtoull(buf
, 10, &recovery_start
))
2950 if (rdev
->mddev
->pers
&&
2951 rdev
->raid_disk
>= 0)
2954 rdev
->recovery_offset
= recovery_start
;
2955 if (recovery_start
== MaxSector
)
2956 set_bit(In_sync
, &rdev
->flags
);
2958 clear_bit(In_sync
, &rdev
->flags
);
2962 static struct rdev_sysfs_entry rdev_recovery_start
=
2963 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2966 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2968 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2970 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2972 return badblocks_show(&rdev
->badblocks
, page
, 0);
2974 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2976 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2977 /* Maybe that ack was all we needed */
2978 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2979 wake_up(&rdev
->blocked_wait
);
2982 static struct rdev_sysfs_entry rdev_bad_blocks
=
2983 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2985 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2987 return badblocks_show(&rdev
->badblocks
, page
, 1);
2989 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2991 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2993 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2994 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2996 static struct attribute
*rdev_default_attrs
[] = {
3001 &rdev_new_offset
.attr
,
3003 &rdev_recovery_start
.attr
,
3004 &rdev_bad_blocks
.attr
,
3005 &rdev_unack_bad_blocks
.attr
,
3009 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3011 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3012 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3018 return entry
->show(rdev
, page
);
3022 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3023 const char *page
, size_t length
)
3025 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3026 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3028 struct mddev
*mddev
= rdev
->mddev
;
3032 if (!capable(CAP_SYS_ADMIN
))
3034 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3036 if (rdev
->mddev
== NULL
)
3039 rv
= entry
->store(rdev
, page
, length
);
3040 mddev_unlock(mddev
);
3045 static void rdev_free(struct kobject
*ko
)
3047 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3050 static const struct sysfs_ops rdev_sysfs_ops
= {
3051 .show
= rdev_attr_show
,
3052 .store
= rdev_attr_store
,
3054 static struct kobj_type rdev_ktype
= {
3055 .release
= rdev_free
,
3056 .sysfs_ops
= &rdev_sysfs_ops
,
3057 .default_attrs
= rdev_default_attrs
,
3060 int md_rdev_init(struct md_rdev
*rdev
)
3063 rdev
->saved_raid_disk
= -1;
3064 rdev
->raid_disk
= -1;
3066 rdev
->data_offset
= 0;
3067 rdev
->new_data_offset
= 0;
3068 rdev
->sb_events
= 0;
3069 rdev
->last_read_error
.tv_sec
= 0;
3070 rdev
->last_read_error
.tv_nsec
= 0;
3071 rdev
->sb_loaded
= 0;
3072 rdev
->bb_page
= NULL
;
3073 atomic_set(&rdev
->nr_pending
, 0);
3074 atomic_set(&rdev
->read_errors
, 0);
3075 atomic_set(&rdev
->corrected_errors
, 0);
3077 INIT_LIST_HEAD(&rdev
->same_set
);
3078 init_waitqueue_head(&rdev
->blocked_wait
);
3080 /* Add space to store bad block list.
3081 * This reserves the space even on arrays where it cannot
3082 * be used - I wonder if that matters
3084 rdev
->badblocks
.count
= 0;
3085 rdev
->badblocks
.shift
= -1; /* disabled until explicitly enabled */
3086 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3087 seqlock_init(&rdev
->badblocks
.lock
);
3088 if (rdev
->badblocks
.page
== NULL
)
3093 EXPORT_SYMBOL_GPL(md_rdev_init
);
3095 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3097 * mark the device faulty if:
3099 * - the device is nonexistent (zero size)
3100 * - the device has no valid superblock
3102 * a faulty rdev _never_ has rdev->sb set.
3104 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3106 char b
[BDEVNAME_SIZE
];
3108 struct md_rdev
*rdev
;
3111 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3113 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3114 return ERR_PTR(-ENOMEM
);
3117 err
= md_rdev_init(rdev
);
3120 err
= alloc_disk_sb(rdev
);
3124 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3128 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3130 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3133 "md: %s has zero or unknown size, marking faulty!\n",
3134 bdevname(rdev
->bdev
,b
));
3139 if (super_format
>= 0) {
3140 err
= super_types
[super_format
].
3141 load_super(rdev
, NULL
, super_minor
);
3142 if (err
== -EINVAL
) {
3144 "md: %s does not have a valid v%d.%d "
3145 "superblock, not importing!\n",
3146 bdevname(rdev
->bdev
,b
),
3147 super_format
, super_minor
);
3152 "md: could not read %s's sb, not importing!\n",
3153 bdevname(rdev
->bdev
,b
));
3163 md_rdev_clear(rdev
);
3165 return ERR_PTR(err
);
3169 * Check a full RAID array for plausibility
3172 static void analyze_sbs(struct mddev
*mddev
)
3175 struct md_rdev
*rdev
, *freshest
, *tmp
;
3176 char b
[BDEVNAME_SIZE
];
3179 rdev_for_each_safe(rdev
, tmp
, mddev
)
3180 switch (super_types
[mddev
->major_version
].
3181 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3189 "md: fatal superblock inconsistency in %s"
3190 " -- removing from array\n",
3191 bdevname(rdev
->bdev
,b
));
3192 md_kick_rdev_from_array(rdev
);
3195 super_types
[mddev
->major_version
].
3196 validate_super(mddev
, freshest
);
3199 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3200 if (mddev
->max_disks
&&
3201 (rdev
->desc_nr
>= mddev
->max_disks
||
3202 i
> mddev
->max_disks
)) {
3204 "md: %s: %s: only %d devices permitted\n",
3205 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3207 md_kick_rdev_from_array(rdev
);
3210 if (rdev
!= freshest
) {
3211 if (super_types
[mddev
->major_version
].
3212 validate_super(mddev
, rdev
)) {
3213 printk(KERN_WARNING
"md: kicking non-fresh %s"
3215 bdevname(rdev
->bdev
,b
));
3216 md_kick_rdev_from_array(rdev
);
3219 /* No device should have a Candidate flag
3220 * when reading devices
3222 if (test_bit(Candidate
, &rdev
->flags
)) {
3223 pr_info("md: kicking Cluster Candidate %s from array!\n",
3224 bdevname(rdev
->bdev
, b
));
3225 md_kick_rdev_from_array(rdev
);
3228 if (mddev
->level
== LEVEL_MULTIPATH
) {
3229 rdev
->desc_nr
= i
++;
3230 rdev
->raid_disk
= rdev
->desc_nr
;
3231 set_bit(In_sync
, &rdev
->flags
);
3232 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3233 rdev
->raid_disk
= -1;
3234 clear_bit(In_sync
, &rdev
->flags
);
3239 /* Read a fixed-point number.
3240 * Numbers in sysfs attributes should be in "standard" units where
3241 * possible, so time should be in seconds.
3242 * However we internally use a a much smaller unit such as
3243 * milliseconds or jiffies.
3244 * This function takes a decimal number with a possible fractional
3245 * component, and produces an integer which is the result of
3246 * multiplying that number by 10^'scale'.
3247 * all without any floating-point arithmetic.
3249 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3251 unsigned long result
= 0;
3253 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3256 else if (decimals
< scale
) {
3259 result
= result
* 10 + value
;
3271 while (decimals
< scale
) {
3279 static void md_safemode_timeout(unsigned long data
);
3282 safe_delay_show(struct mddev
*mddev
, char *page
)
3284 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3285 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3288 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3292 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3295 mddev
->safemode_delay
= 0;
3297 unsigned long old_delay
= mddev
->safemode_delay
;
3298 unsigned long new_delay
= (msec
*HZ
)/1000;
3302 mddev
->safemode_delay
= new_delay
;
3303 if (new_delay
< old_delay
|| old_delay
== 0)
3304 mod_timer(&mddev
->safemode_timer
, jiffies
+1);
3308 static struct md_sysfs_entry md_safe_delay
=
3309 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3312 level_show(struct mddev
*mddev
, char *page
)
3314 struct md_personality
*p
;
3316 spin_lock(&mddev
->lock
);
3319 ret
= sprintf(page
, "%s\n", p
->name
);
3320 else if (mddev
->clevel
[0])
3321 ret
= sprintf(page
, "%s\n", mddev
->clevel
);
3322 else if (mddev
->level
!= LEVEL_NONE
)
3323 ret
= sprintf(page
, "%d\n", mddev
->level
);
3326 spin_unlock(&mddev
->lock
);
3331 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3336 struct md_personality
*pers
, *oldpers
;
3338 void *priv
, *oldpriv
;
3339 struct md_rdev
*rdev
;
3341 if (slen
== 0 || slen
>= sizeof(clevel
))
3344 rv
= mddev_lock(mddev
);
3348 if (mddev
->pers
== NULL
) {
3349 strncpy(mddev
->clevel
, buf
, slen
);
3350 if (mddev
->clevel
[slen
-1] == '\n')
3352 mddev
->clevel
[slen
] = 0;
3353 mddev
->level
= LEVEL_NONE
;
3361 /* request to change the personality. Need to ensure:
3362 * - array is not engaged in resync/recovery/reshape
3363 * - old personality can be suspended
3364 * - new personality will access other array.
3368 if (mddev
->sync_thread
||
3369 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3370 mddev
->reshape_position
!= MaxSector
||
3371 mddev
->sysfs_active
)
3375 if (!mddev
->pers
->quiesce
) {
3376 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3377 mdname(mddev
), mddev
->pers
->name
);
3381 /* Now find the new personality */
3382 strncpy(clevel
, buf
, slen
);
3383 if (clevel
[slen
-1] == '\n')
3386 if (kstrtol(clevel
, 10, &level
))
3389 if (request_module("md-%s", clevel
) != 0)
3390 request_module("md-level-%s", clevel
);
3391 spin_lock(&pers_lock
);
3392 pers
= find_pers(level
, clevel
);
3393 if (!pers
|| !try_module_get(pers
->owner
)) {
3394 spin_unlock(&pers_lock
);
3395 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3399 spin_unlock(&pers_lock
);
3401 if (pers
== mddev
->pers
) {
3402 /* Nothing to do! */
3403 module_put(pers
->owner
);
3407 if (!pers
->takeover
) {
3408 module_put(pers
->owner
);
3409 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3410 mdname(mddev
), clevel
);
3415 rdev_for_each(rdev
, mddev
)
3416 rdev
->new_raid_disk
= rdev
->raid_disk
;
3418 /* ->takeover must set new_* and/or delta_disks
3419 * if it succeeds, and may set them when it fails.
3421 priv
= pers
->takeover(mddev
);
3423 mddev
->new_level
= mddev
->level
;
3424 mddev
->new_layout
= mddev
->layout
;
3425 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3426 mddev
->raid_disks
-= mddev
->delta_disks
;
3427 mddev
->delta_disks
= 0;
3428 mddev
->reshape_backwards
= 0;
3429 module_put(pers
->owner
);
3430 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3431 mdname(mddev
), clevel
);
3436 /* Looks like we have a winner */
3437 mddev_suspend(mddev
);
3438 mddev_detach(mddev
);
3440 spin_lock(&mddev
->lock
);
3441 oldpers
= mddev
->pers
;
3442 oldpriv
= mddev
->private;
3444 mddev
->private = priv
;
3445 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3446 mddev
->level
= mddev
->new_level
;
3447 mddev
->layout
= mddev
->new_layout
;
3448 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3449 mddev
->delta_disks
= 0;
3450 mddev
->reshape_backwards
= 0;
3451 mddev
->degraded
= 0;
3452 spin_unlock(&mddev
->lock
);
3454 if (oldpers
->sync_request
== NULL
&&
3456 /* We are converting from a no-redundancy array
3457 * to a redundancy array and metadata is managed
3458 * externally so we need to be sure that writes
3459 * won't block due to a need to transition
3461 * until external management is started.
3464 mddev
->safemode_delay
= 0;
3465 mddev
->safemode
= 0;
3468 oldpers
->free(mddev
, oldpriv
);
3470 if (oldpers
->sync_request
== NULL
&&
3471 pers
->sync_request
!= NULL
) {
3472 /* need to add the md_redundancy_group */
3473 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3475 "md: cannot register extra attributes for %s\n",
3477 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3479 if (oldpers
->sync_request
!= NULL
&&
3480 pers
->sync_request
== NULL
) {
3481 /* need to remove the md_redundancy_group */
3482 if (mddev
->to_remove
== NULL
)
3483 mddev
->to_remove
= &md_redundancy_group
;
3486 rdev_for_each(rdev
, mddev
) {
3487 if (rdev
->raid_disk
< 0)
3489 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3490 rdev
->new_raid_disk
= -1;
3491 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3493 sysfs_unlink_rdev(mddev
, rdev
);
3495 rdev_for_each(rdev
, mddev
) {
3496 if (rdev
->raid_disk
< 0)
3498 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3500 rdev
->raid_disk
= rdev
->new_raid_disk
;
3501 if (rdev
->raid_disk
< 0)
3502 clear_bit(In_sync
, &rdev
->flags
);
3504 if (sysfs_link_rdev(mddev
, rdev
))
3505 printk(KERN_WARNING
"md: cannot register rd%d"
3506 " for %s after level change\n",
3507 rdev
->raid_disk
, mdname(mddev
));
3511 if (pers
->sync_request
== NULL
) {
3512 /* this is now an array without redundancy, so
3513 * it must always be in_sync
3516 del_timer_sync(&mddev
->safemode_timer
);
3518 blk_set_stacking_limits(&mddev
->queue
->limits
);
3520 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3521 mddev_resume(mddev
);
3523 md_update_sb(mddev
, 1);
3524 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3525 md_new_event(mddev
);
3528 mddev_unlock(mddev
);
3532 static struct md_sysfs_entry md_level
=
3533 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3536 layout_show(struct mddev
*mddev
, char *page
)
3538 /* just a number, not meaningful for all levels */
3539 if (mddev
->reshape_position
!= MaxSector
&&
3540 mddev
->layout
!= mddev
->new_layout
)
3541 return sprintf(page
, "%d (%d)\n",
3542 mddev
->new_layout
, mddev
->layout
);
3543 return sprintf(page
, "%d\n", mddev
->layout
);
3547 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3552 err
= kstrtouint(buf
, 10, &n
);
3555 err
= mddev_lock(mddev
);
3560 if (mddev
->pers
->check_reshape
== NULL
)
3565 mddev
->new_layout
= n
;
3566 err
= mddev
->pers
->check_reshape(mddev
);
3568 mddev
->new_layout
= mddev
->layout
;
3571 mddev
->new_layout
= n
;
3572 if (mddev
->reshape_position
== MaxSector
)
3575 mddev_unlock(mddev
);
3578 static struct md_sysfs_entry md_layout
=
3579 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3582 raid_disks_show(struct mddev
*mddev
, char *page
)
3584 if (mddev
->raid_disks
== 0)
3586 if (mddev
->reshape_position
!= MaxSector
&&
3587 mddev
->delta_disks
!= 0)
3588 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3589 mddev
->raid_disks
- mddev
->delta_disks
);
3590 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3593 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3596 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3601 err
= kstrtouint(buf
, 10, &n
);
3605 err
= mddev_lock(mddev
);
3609 err
= update_raid_disks(mddev
, n
);
3610 else if (mddev
->reshape_position
!= MaxSector
) {
3611 struct md_rdev
*rdev
;
3612 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3615 rdev_for_each(rdev
, mddev
) {
3617 rdev
->data_offset
< rdev
->new_data_offset
)
3620 rdev
->data_offset
> rdev
->new_data_offset
)
3624 mddev
->delta_disks
= n
- olddisks
;
3625 mddev
->raid_disks
= n
;
3626 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3628 mddev
->raid_disks
= n
;
3630 mddev_unlock(mddev
);
3631 return err
? err
: len
;
3633 static struct md_sysfs_entry md_raid_disks
=
3634 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3637 chunk_size_show(struct mddev
*mddev
, char *page
)
3639 if (mddev
->reshape_position
!= MaxSector
&&
3640 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3641 return sprintf(page
, "%d (%d)\n",
3642 mddev
->new_chunk_sectors
<< 9,
3643 mddev
->chunk_sectors
<< 9);
3644 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3648 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3653 err
= kstrtoul(buf
, 10, &n
);
3657 err
= mddev_lock(mddev
);
3661 if (mddev
->pers
->check_reshape
== NULL
)
3666 mddev
->new_chunk_sectors
= n
>> 9;
3667 err
= mddev
->pers
->check_reshape(mddev
);
3669 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3672 mddev
->new_chunk_sectors
= n
>> 9;
3673 if (mddev
->reshape_position
== MaxSector
)
3674 mddev
->chunk_sectors
= n
>> 9;
3676 mddev_unlock(mddev
);
3679 static struct md_sysfs_entry md_chunk_size
=
3680 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3683 resync_start_show(struct mddev
*mddev
, char *page
)
3685 if (mddev
->recovery_cp
== MaxSector
)
3686 return sprintf(page
, "none\n");
3687 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3691 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3693 unsigned long long n
;
3696 if (cmd_match(buf
, "none"))
3699 err
= kstrtoull(buf
, 10, &n
);
3702 if (n
!= (sector_t
)n
)
3706 err
= mddev_lock(mddev
);
3709 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3713 mddev
->recovery_cp
= n
;
3715 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3717 mddev_unlock(mddev
);
3720 static struct md_sysfs_entry md_resync_start
=
3721 __ATTR_PREALLOC(resync_start
, S_IRUGO
|S_IWUSR
,
3722 resync_start_show
, resync_start_store
);
3725 * The array state can be:
3728 * No devices, no size, no level
3729 * Equivalent to STOP_ARRAY ioctl
3731 * May have some settings, but array is not active
3732 * all IO results in error
3733 * When written, doesn't tear down array, but just stops it
3734 * suspended (not supported yet)
3735 * All IO requests will block. The array can be reconfigured.
3736 * Writing this, if accepted, will block until array is quiescent
3738 * no resync can happen. no superblocks get written.
3739 * write requests fail
3741 * like readonly, but behaves like 'clean' on a write request.
3743 * clean - no pending writes, but otherwise active.
3744 * When written to inactive array, starts without resync
3745 * If a write request arrives then
3746 * if metadata is known, mark 'dirty' and switch to 'active'.
3747 * if not known, block and switch to write-pending
3748 * If written to an active array that has pending writes, then fails.
3750 * fully active: IO and resync can be happening.
3751 * When written to inactive array, starts with resync
3754 * clean, but writes are blocked waiting for 'active' to be written.
3757 * like active, but no writes have been seen for a while (100msec).
3760 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3761 write_pending
, active_idle
, bad_word
};
3762 static char *array_states
[] = {
3763 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3764 "write-pending", "active-idle", NULL
};
3766 static int match_word(const char *word
, char **list
)
3769 for (n
=0; list
[n
]; n
++)
3770 if (cmd_match(word
, list
[n
]))
3776 array_state_show(struct mddev
*mddev
, char *page
)
3778 enum array_state st
= inactive
;
3791 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3793 else if (mddev
->safemode
)
3799 if (list_empty(&mddev
->disks
) &&
3800 mddev
->raid_disks
== 0 &&
3801 mddev
->dev_sectors
== 0)
3806 return sprintf(page
, "%s\n", array_states
[st
]);
3809 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
3810 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
3811 static int do_md_run(struct mddev
*mddev
);
3812 static int restart_array(struct mddev
*mddev
);
3815 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3818 enum array_state st
= match_word(buf
, array_states
);
3820 if (mddev
->pers
&& (st
== active
|| st
== clean
) && mddev
->ro
!= 1) {
3821 /* don't take reconfig_mutex when toggling between
3824 spin_lock(&mddev
->lock
);
3826 restart_array(mddev
);
3827 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3828 wake_up(&mddev
->sb_wait
);
3830 } else /* st == clean */ {
3831 restart_array(mddev
);
3832 if (atomic_read(&mddev
->writes_pending
) == 0) {
3833 if (mddev
->in_sync
== 0) {
3835 if (mddev
->safemode
== 1)
3836 mddev
->safemode
= 0;
3837 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3843 spin_unlock(&mddev
->lock
);
3846 err
= mddev_lock(mddev
);
3854 /* stopping an active array */
3855 err
= do_md_stop(mddev
, 0, NULL
);
3858 /* stopping an active array */
3860 err
= do_md_stop(mddev
, 2, NULL
);
3862 err
= 0; /* already inactive */
3865 break; /* not supported yet */
3868 err
= md_set_readonly(mddev
, NULL
);
3871 set_disk_ro(mddev
->gendisk
, 1);
3872 err
= do_md_run(mddev
);
3878 err
= md_set_readonly(mddev
, NULL
);
3879 else if (mddev
->ro
== 1)
3880 err
= restart_array(mddev
);
3883 set_disk_ro(mddev
->gendisk
, 0);
3887 err
= do_md_run(mddev
);
3892 restart_array(mddev
);
3893 spin_lock(&mddev
->lock
);
3894 if (atomic_read(&mddev
->writes_pending
) == 0) {
3895 if (mddev
->in_sync
== 0) {
3897 if (mddev
->safemode
== 1)
3898 mddev
->safemode
= 0;
3899 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3904 spin_unlock(&mddev
->lock
);
3910 restart_array(mddev
);
3911 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3912 wake_up(&mddev
->sb_wait
);
3916 set_disk_ro(mddev
->gendisk
, 0);
3917 err
= do_md_run(mddev
);
3922 /* these cannot be set */
3927 if (mddev
->hold_active
== UNTIL_IOCTL
)
3928 mddev
->hold_active
= 0;
3929 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3931 mddev_unlock(mddev
);
3934 static struct md_sysfs_entry md_array_state
=
3935 __ATTR_PREALLOC(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3938 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3939 return sprintf(page
, "%d\n",
3940 atomic_read(&mddev
->max_corr_read_errors
));
3944 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3949 rv
= kstrtouint(buf
, 10, &n
);
3952 atomic_set(&mddev
->max_corr_read_errors
, n
);
3956 static struct md_sysfs_entry max_corr_read_errors
=
3957 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3958 max_corrected_read_errors_store
);
3961 null_show(struct mddev
*mddev
, char *page
)
3967 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3969 /* buf must be %d:%d\n? giving major and minor numbers */
3970 /* The new device is added to the array.
3971 * If the array has a persistent superblock, we read the
3972 * superblock to initialise info and check validity.
3973 * Otherwise, only checking done is that in bind_rdev_to_array,
3974 * which mainly checks size.
3977 int major
= simple_strtoul(buf
, &e
, 10);
3980 struct md_rdev
*rdev
;
3983 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3985 minor
= simple_strtoul(e
+1, &e
, 10);
3986 if (*e
&& *e
!= '\n')
3988 dev
= MKDEV(major
, minor
);
3989 if (major
!= MAJOR(dev
) ||
3990 minor
!= MINOR(dev
))
3993 flush_workqueue(md_misc_wq
);
3995 err
= mddev_lock(mddev
);
3998 if (mddev
->persistent
) {
3999 rdev
= md_import_device(dev
, mddev
->major_version
,
4000 mddev
->minor_version
);
4001 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4002 struct md_rdev
*rdev0
4003 = list_entry(mddev
->disks
.next
,
4004 struct md_rdev
, same_set
);
4005 err
= super_types
[mddev
->major_version
]
4006 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4010 } else if (mddev
->external
)
4011 rdev
= md_import_device(dev
, -2, -1);
4013 rdev
= md_import_device(dev
, -1, -1);
4016 mddev_unlock(mddev
);
4017 return PTR_ERR(rdev
);
4019 err
= bind_rdev_to_array(rdev
, mddev
);
4023 mddev_unlock(mddev
);
4024 return err
? err
: len
;
4027 static struct md_sysfs_entry md_new_device
=
4028 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4031 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4034 unsigned long chunk
, end_chunk
;
4037 err
= mddev_lock(mddev
);
4042 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4044 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4045 if (buf
== end
) break;
4046 if (*end
== '-') { /* range */
4048 end_chunk
= simple_strtoul(buf
, &end
, 0);
4049 if (buf
== end
) break;
4051 if (*end
&& !isspace(*end
)) break;
4052 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4053 buf
= skip_spaces(end
);
4055 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4057 mddev_unlock(mddev
);
4061 static struct md_sysfs_entry md_bitmap
=
4062 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4065 size_show(struct mddev
*mddev
, char *page
)
4067 return sprintf(page
, "%llu\n",
4068 (unsigned long long)mddev
->dev_sectors
/ 2);
4071 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4074 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4076 /* If array is inactive, we can reduce the component size, but
4077 * not increase it (except from 0).
4078 * If array is active, we can try an on-line resize
4081 int err
= strict_blocks_to_sectors(buf
, §ors
);
4085 err
= mddev_lock(mddev
);
4089 if (mddev_is_clustered(mddev
))
4090 md_cluster_ops
->metadata_update_start(mddev
);
4091 err
= update_size(mddev
, sectors
);
4092 md_update_sb(mddev
, 1);
4093 if (mddev_is_clustered(mddev
))
4094 md_cluster_ops
->metadata_update_finish(mddev
);
4096 if (mddev
->dev_sectors
== 0 ||
4097 mddev
->dev_sectors
> sectors
)
4098 mddev
->dev_sectors
= sectors
;
4102 mddev_unlock(mddev
);
4103 return err
? err
: len
;
4106 static struct md_sysfs_entry md_size
=
4107 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4109 /* Metadata version.
4111 * 'none' for arrays with no metadata (good luck...)
4112 * 'external' for arrays with externally managed metadata,
4113 * or N.M for internally known formats
4116 metadata_show(struct mddev
*mddev
, char *page
)
4118 if (mddev
->persistent
)
4119 return sprintf(page
, "%d.%d\n",
4120 mddev
->major_version
, mddev
->minor_version
);
4121 else if (mddev
->external
)
4122 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4124 return sprintf(page
, "none\n");
4128 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4133 /* Changing the details of 'external' metadata is
4134 * always permitted. Otherwise there must be
4135 * no devices attached to the array.
4138 err
= mddev_lock(mddev
);
4142 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4144 else if (!list_empty(&mddev
->disks
))
4148 if (cmd_match(buf
, "none")) {
4149 mddev
->persistent
= 0;
4150 mddev
->external
= 0;
4151 mddev
->major_version
= 0;
4152 mddev
->minor_version
= 90;
4155 if (strncmp(buf
, "external:", 9) == 0) {
4156 size_t namelen
= len
-9;
4157 if (namelen
>= sizeof(mddev
->metadata_type
))
4158 namelen
= sizeof(mddev
->metadata_type
)-1;
4159 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4160 mddev
->metadata_type
[namelen
] = 0;
4161 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4162 mddev
->metadata_type
[--namelen
] = 0;
4163 mddev
->persistent
= 0;
4164 mddev
->external
= 1;
4165 mddev
->major_version
= 0;
4166 mddev
->minor_version
= 90;
4169 major
= simple_strtoul(buf
, &e
, 10);
4171 if (e
==buf
|| *e
!= '.')
4174 minor
= simple_strtoul(buf
, &e
, 10);
4175 if (e
==buf
|| (*e
&& *e
!= '\n') )
4178 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4180 mddev
->major_version
= major
;
4181 mddev
->minor_version
= minor
;
4182 mddev
->persistent
= 1;
4183 mddev
->external
= 0;
4186 mddev_unlock(mddev
);
4190 static struct md_sysfs_entry md_metadata
=
4191 __ATTR_PREALLOC(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4194 action_show(struct mddev
*mddev
, char *page
)
4196 char *type
= "idle";
4197 unsigned long recovery
= mddev
->recovery
;
4198 if (test_bit(MD_RECOVERY_FROZEN
, &recovery
))
4200 else if (test_bit(MD_RECOVERY_RUNNING
, &recovery
) ||
4201 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &recovery
))) {
4202 if (test_bit(MD_RECOVERY_RESHAPE
, &recovery
))
4204 else if (test_bit(MD_RECOVERY_SYNC
, &recovery
)) {
4205 if (!test_bit(MD_RECOVERY_REQUESTED
, &recovery
))
4207 else if (test_bit(MD_RECOVERY_CHECK
, &recovery
))
4211 } else if (test_bit(MD_RECOVERY_RECOVER
, &recovery
))
4214 return sprintf(page
, "%s\n", type
);
4218 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4220 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4224 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4225 if (cmd_match(page
, "frozen"))
4226 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4228 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4229 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4230 mddev_lock(mddev
) == 0) {
4231 flush_workqueue(md_misc_wq
);
4232 if (mddev
->sync_thread
) {
4233 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4234 md_reap_sync_thread(mddev
);
4236 mddev_unlock(mddev
);
4238 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4239 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4241 else if (cmd_match(page
, "resync"))
4242 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4243 else if (cmd_match(page
, "recover")) {
4244 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4245 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4246 } else if (cmd_match(page
, "reshape")) {
4248 if (mddev
->pers
->start_reshape
== NULL
)
4250 err
= mddev_lock(mddev
);
4252 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4253 err
= mddev
->pers
->start_reshape(mddev
);
4254 mddev_unlock(mddev
);
4258 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4260 if (cmd_match(page
, "check"))
4261 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4262 else if (!cmd_match(page
, "repair"))
4264 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4265 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4266 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4268 if (mddev
->ro
== 2) {
4269 /* A write to sync_action is enough to justify
4270 * canceling read-auto mode
4273 md_wakeup_thread(mddev
->sync_thread
);
4275 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4276 md_wakeup_thread(mddev
->thread
);
4277 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4281 static struct md_sysfs_entry md_scan_mode
=
4282 __ATTR_PREALLOC(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4285 last_sync_action_show(struct mddev
*mddev
, char *page
)
4287 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4290 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4293 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4295 return sprintf(page
, "%llu\n",
4296 (unsigned long long)
4297 atomic64_read(&mddev
->resync_mismatches
));
4300 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4303 sync_min_show(struct mddev
*mddev
, char *page
)
4305 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4306 mddev
->sync_speed_min
? "local": "system");
4310 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4315 if (strncmp(buf
, "system", 6)==0) {
4318 rv
= kstrtouint(buf
, 10, &min
);
4324 mddev
->sync_speed_min
= min
;
4328 static struct md_sysfs_entry md_sync_min
=
4329 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4332 sync_max_show(struct mddev
*mddev
, char *page
)
4334 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4335 mddev
->sync_speed_max
? "local": "system");
4339 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4344 if (strncmp(buf
, "system", 6)==0) {
4347 rv
= kstrtouint(buf
, 10, &max
);
4353 mddev
->sync_speed_max
= max
;
4357 static struct md_sysfs_entry md_sync_max
=
4358 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4361 degraded_show(struct mddev
*mddev
, char *page
)
4363 return sprintf(page
, "%d\n", mddev
->degraded
);
4365 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4368 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4370 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4374 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4378 if (kstrtol(buf
, 10, &n
))
4381 if (n
!= 0 && n
!= 1)
4384 mddev
->parallel_resync
= n
;
4386 if (mddev
->sync_thread
)
4387 wake_up(&resync_wait
);
4392 /* force parallel resync, even with shared block devices */
4393 static struct md_sysfs_entry md_sync_force_parallel
=
4394 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4395 sync_force_parallel_show
, sync_force_parallel_store
);
4398 sync_speed_show(struct mddev
*mddev
, char *page
)
4400 unsigned long resync
, dt
, db
;
4401 if (mddev
->curr_resync
== 0)
4402 return sprintf(page
, "none\n");
4403 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4404 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4406 db
= resync
- mddev
->resync_mark_cnt
;
4407 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4410 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4413 sync_completed_show(struct mddev
*mddev
, char *page
)
4415 unsigned long long max_sectors
, resync
;
4417 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4418 return sprintf(page
, "none\n");
4420 if (mddev
->curr_resync
== 1 ||
4421 mddev
->curr_resync
== 2)
4422 return sprintf(page
, "delayed\n");
4424 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4425 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4426 max_sectors
= mddev
->resync_max_sectors
;
4428 max_sectors
= mddev
->dev_sectors
;
4430 resync
= mddev
->curr_resync_completed
;
4431 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4434 static struct md_sysfs_entry md_sync_completed
=
4435 __ATTR_PREALLOC(sync_completed
, S_IRUGO
, sync_completed_show
, NULL
);
4438 min_sync_show(struct mddev
*mddev
, char *page
)
4440 return sprintf(page
, "%llu\n",
4441 (unsigned long long)mddev
->resync_min
);
4444 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4446 unsigned long long min
;
4449 if (kstrtoull(buf
, 10, &min
))
4452 spin_lock(&mddev
->lock
);
4454 if (min
> mddev
->resync_max
)
4458 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4461 /* Round down to multiple of 4K for safety */
4462 mddev
->resync_min
= round_down(min
, 8);
4466 spin_unlock(&mddev
->lock
);
4470 static struct md_sysfs_entry md_min_sync
=
4471 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4474 max_sync_show(struct mddev
*mddev
, char *page
)
4476 if (mddev
->resync_max
== MaxSector
)
4477 return sprintf(page
, "max\n");
4479 return sprintf(page
, "%llu\n",
4480 (unsigned long long)mddev
->resync_max
);
4483 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4486 spin_lock(&mddev
->lock
);
4487 if (strncmp(buf
, "max", 3) == 0)
4488 mddev
->resync_max
= MaxSector
;
4490 unsigned long long max
;
4494 if (kstrtoull(buf
, 10, &max
))
4496 if (max
< mddev
->resync_min
)
4500 if (max
< mddev
->resync_max
&&
4502 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4505 /* Must be a multiple of chunk_size */
4506 chunk
= mddev
->chunk_sectors
;
4508 sector_t temp
= max
;
4511 if (sector_div(temp
, chunk
))
4514 mddev
->resync_max
= max
;
4516 wake_up(&mddev
->recovery_wait
);
4519 spin_unlock(&mddev
->lock
);
4523 static struct md_sysfs_entry md_max_sync
=
4524 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4527 suspend_lo_show(struct mddev
*mddev
, char *page
)
4529 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4533 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4535 unsigned long long old
, new;
4538 err
= kstrtoull(buf
, 10, &new);
4541 if (new != (sector_t
)new)
4544 err
= mddev_lock(mddev
);
4548 if (mddev
->pers
== NULL
||
4549 mddev
->pers
->quiesce
== NULL
)
4551 old
= mddev
->suspend_lo
;
4552 mddev
->suspend_lo
= new;
4554 /* Shrinking suspended region */
4555 mddev
->pers
->quiesce(mddev
, 2);
4557 /* Expanding suspended region - need to wait */
4558 mddev
->pers
->quiesce(mddev
, 1);
4559 mddev
->pers
->quiesce(mddev
, 0);
4563 mddev_unlock(mddev
);
4566 static struct md_sysfs_entry md_suspend_lo
=
4567 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4570 suspend_hi_show(struct mddev
*mddev
, char *page
)
4572 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4576 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4578 unsigned long long old
, new;
4581 err
= kstrtoull(buf
, 10, &new);
4584 if (new != (sector_t
)new)
4587 err
= mddev_lock(mddev
);
4591 if (mddev
->pers
== NULL
||
4592 mddev
->pers
->quiesce
== NULL
)
4594 old
= mddev
->suspend_hi
;
4595 mddev
->suspend_hi
= new;
4597 /* Shrinking suspended region */
4598 mddev
->pers
->quiesce(mddev
, 2);
4600 /* Expanding suspended region - need to wait */
4601 mddev
->pers
->quiesce(mddev
, 1);
4602 mddev
->pers
->quiesce(mddev
, 0);
4606 mddev_unlock(mddev
);
4609 static struct md_sysfs_entry md_suspend_hi
=
4610 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4613 reshape_position_show(struct mddev
*mddev
, char *page
)
4615 if (mddev
->reshape_position
!= MaxSector
)
4616 return sprintf(page
, "%llu\n",
4617 (unsigned long long)mddev
->reshape_position
);
4618 strcpy(page
, "none\n");
4623 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4625 struct md_rdev
*rdev
;
4626 unsigned long long new;
4629 err
= kstrtoull(buf
, 10, &new);
4632 if (new != (sector_t
)new)
4634 err
= mddev_lock(mddev
);
4640 mddev
->reshape_position
= new;
4641 mddev
->delta_disks
= 0;
4642 mddev
->reshape_backwards
= 0;
4643 mddev
->new_level
= mddev
->level
;
4644 mddev
->new_layout
= mddev
->layout
;
4645 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4646 rdev_for_each(rdev
, mddev
)
4647 rdev
->new_data_offset
= rdev
->data_offset
;
4650 mddev_unlock(mddev
);
4654 static struct md_sysfs_entry md_reshape_position
=
4655 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4656 reshape_position_store
);
4659 reshape_direction_show(struct mddev
*mddev
, char *page
)
4661 return sprintf(page
, "%s\n",
4662 mddev
->reshape_backwards
? "backwards" : "forwards");
4666 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4671 if (cmd_match(buf
, "forwards"))
4673 else if (cmd_match(buf
, "backwards"))
4677 if (mddev
->reshape_backwards
== backwards
)
4680 err
= mddev_lock(mddev
);
4683 /* check if we are allowed to change */
4684 if (mddev
->delta_disks
)
4686 else if (mddev
->persistent
&&
4687 mddev
->major_version
== 0)
4690 mddev
->reshape_backwards
= backwards
;
4691 mddev_unlock(mddev
);
4695 static struct md_sysfs_entry md_reshape_direction
=
4696 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4697 reshape_direction_store
);
4700 array_size_show(struct mddev
*mddev
, char *page
)
4702 if (mddev
->external_size
)
4703 return sprintf(page
, "%llu\n",
4704 (unsigned long long)mddev
->array_sectors
/2);
4706 return sprintf(page
, "default\n");
4710 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4715 err
= mddev_lock(mddev
);
4719 if (strncmp(buf
, "default", 7) == 0) {
4721 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4723 sectors
= mddev
->array_sectors
;
4725 mddev
->external_size
= 0;
4727 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4729 else if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4732 mddev
->external_size
= 1;
4736 mddev
->array_sectors
= sectors
;
4738 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4739 revalidate_disk(mddev
->gendisk
);
4742 mddev_unlock(mddev
);
4746 static struct md_sysfs_entry md_array_size
=
4747 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4750 static struct attribute
*md_default_attrs
[] = {
4753 &md_raid_disks
.attr
,
4754 &md_chunk_size
.attr
,
4756 &md_resync_start
.attr
,
4758 &md_new_device
.attr
,
4759 &md_safe_delay
.attr
,
4760 &md_array_state
.attr
,
4761 &md_reshape_position
.attr
,
4762 &md_reshape_direction
.attr
,
4763 &md_array_size
.attr
,
4764 &max_corr_read_errors
.attr
,
4768 static struct attribute
*md_redundancy_attrs
[] = {
4770 &md_last_scan_mode
.attr
,
4771 &md_mismatches
.attr
,
4774 &md_sync_speed
.attr
,
4775 &md_sync_force_parallel
.attr
,
4776 &md_sync_completed
.attr
,
4779 &md_suspend_lo
.attr
,
4780 &md_suspend_hi
.attr
,
4785 static struct attribute_group md_redundancy_group
= {
4787 .attrs
= md_redundancy_attrs
,
4791 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4793 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4794 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4799 spin_lock(&all_mddevs_lock
);
4800 if (list_empty(&mddev
->all_mddevs
)) {
4801 spin_unlock(&all_mddevs_lock
);
4805 spin_unlock(&all_mddevs_lock
);
4807 rv
= entry
->show(mddev
, page
);
4813 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4814 const char *page
, size_t length
)
4816 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4817 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4822 if (!capable(CAP_SYS_ADMIN
))
4824 spin_lock(&all_mddevs_lock
);
4825 if (list_empty(&mddev
->all_mddevs
)) {
4826 spin_unlock(&all_mddevs_lock
);
4830 spin_unlock(&all_mddevs_lock
);
4831 rv
= entry
->store(mddev
, page
, length
);
4836 static void md_free(struct kobject
*ko
)
4838 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4840 if (mddev
->sysfs_state
)
4841 sysfs_put(mddev
->sysfs_state
);
4844 blk_cleanup_queue(mddev
->queue
);
4845 if (mddev
->gendisk
) {
4846 del_gendisk(mddev
->gendisk
);
4847 put_disk(mddev
->gendisk
);
4853 static const struct sysfs_ops md_sysfs_ops
= {
4854 .show
= md_attr_show
,
4855 .store
= md_attr_store
,
4857 static struct kobj_type md_ktype
= {
4859 .sysfs_ops
= &md_sysfs_ops
,
4860 .default_attrs
= md_default_attrs
,
4865 static void mddev_delayed_delete(struct work_struct
*ws
)
4867 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4869 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4870 kobject_del(&mddev
->kobj
);
4871 kobject_put(&mddev
->kobj
);
4874 static int md_alloc(dev_t dev
, char *name
)
4876 static DEFINE_MUTEX(disks_mutex
);
4877 struct mddev
*mddev
= mddev_find(dev
);
4878 struct gendisk
*disk
;
4887 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4888 shift
= partitioned
? MdpMinorShift
: 0;
4889 unit
= MINOR(mddev
->unit
) >> shift
;
4891 /* wait for any previous instance of this device to be
4892 * completely removed (mddev_delayed_delete).
4894 flush_workqueue(md_misc_wq
);
4896 mutex_lock(&disks_mutex
);
4902 /* Need to ensure that 'name' is not a duplicate.
4904 struct mddev
*mddev2
;
4905 spin_lock(&all_mddevs_lock
);
4907 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4908 if (mddev2
->gendisk
&&
4909 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4910 spin_unlock(&all_mddevs_lock
);
4913 spin_unlock(&all_mddevs_lock
);
4917 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4920 mddev
->queue
->queuedata
= mddev
;
4922 blk_queue_make_request(mddev
->queue
, md_make_request
);
4923 blk_set_stacking_limits(&mddev
->queue
->limits
);
4925 disk
= alloc_disk(1 << shift
);
4927 blk_cleanup_queue(mddev
->queue
);
4928 mddev
->queue
= NULL
;
4931 disk
->major
= MAJOR(mddev
->unit
);
4932 disk
->first_minor
= unit
<< shift
;
4934 strcpy(disk
->disk_name
, name
);
4935 else if (partitioned
)
4936 sprintf(disk
->disk_name
, "md_d%d", unit
);
4938 sprintf(disk
->disk_name
, "md%d", unit
);
4939 disk
->fops
= &md_fops
;
4940 disk
->private_data
= mddev
;
4941 disk
->queue
= mddev
->queue
;
4942 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4943 /* Allow extended partitions. This makes the
4944 * 'mdp' device redundant, but we can't really
4947 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4948 mddev
->gendisk
= disk
;
4949 /* As soon as we call add_disk(), another thread could get
4950 * through to md_open, so make sure it doesn't get too far
4952 mutex_lock(&mddev
->open_mutex
);
4955 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4956 &disk_to_dev(disk
)->kobj
, "%s", "md");
4958 /* This isn't possible, but as kobject_init_and_add is marked
4959 * __must_check, we must do something with the result
4961 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4965 if (mddev
->kobj
.sd
&&
4966 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4967 printk(KERN_DEBUG
"pointless warning\n");
4968 mutex_unlock(&mddev
->open_mutex
);
4970 mutex_unlock(&disks_mutex
);
4971 if (!error
&& mddev
->kobj
.sd
) {
4972 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4973 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4979 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4981 md_alloc(dev
, NULL
);
4985 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4987 /* val must be "md_*" where * is not all digits.
4988 * We allocate an array with a large free minor number, and
4989 * set the name to val. val must not already be an active name.
4991 int len
= strlen(val
);
4992 char buf
[DISK_NAME_LEN
];
4994 while (len
&& val
[len
-1] == '\n')
4996 if (len
>= DISK_NAME_LEN
)
4998 strlcpy(buf
, val
, len
+1);
4999 if (strncmp(buf
, "md_", 3) != 0)
5001 return md_alloc(0, buf
);
5004 static void md_safemode_timeout(unsigned long data
)
5006 struct mddev
*mddev
= (struct mddev
*) data
;
5008 if (!atomic_read(&mddev
->writes_pending
)) {
5009 mddev
->safemode
= 1;
5010 if (mddev
->external
)
5011 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5013 md_wakeup_thread(mddev
->thread
);
5016 static int start_dirty_degraded
;
5018 int md_run(struct mddev
*mddev
)
5021 struct md_rdev
*rdev
;
5022 struct md_personality
*pers
;
5024 if (list_empty(&mddev
->disks
))
5025 /* cannot run an array with no devices.. */
5030 /* Cannot run until previous stop completes properly */
5031 if (mddev
->sysfs_active
)
5035 * Analyze all RAID superblock(s)
5037 if (!mddev
->raid_disks
) {
5038 if (!mddev
->persistent
)
5043 if (mddev
->level
!= LEVEL_NONE
)
5044 request_module("md-level-%d", mddev
->level
);
5045 else if (mddev
->clevel
[0])
5046 request_module("md-%s", mddev
->clevel
);
5049 * Drop all container device buffers, from now on
5050 * the only valid external interface is through the md
5053 rdev_for_each(rdev
, mddev
) {
5054 if (test_bit(Faulty
, &rdev
->flags
))
5056 sync_blockdev(rdev
->bdev
);
5057 invalidate_bdev(rdev
->bdev
);
5059 /* perform some consistency tests on the device.
5060 * We don't want the data to overlap the metadata,
5061 * Internal Bitmap issues have been handled elsewhere.
5063 if (rdev
->meta_bdev
) {
5064 /* Nothing to check */;
5065 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5066 if (mddev
->dev_sectors
&&
5067 rdev
->data_offset
+ mddev
->dev_sectors
5069 printk("md: %s: data overlaps metadata\n",
5074 if (rdev
->sb_start
+ rdev
->sb_size
/512
5075 > rdev
->data_offset
) {
5076 printk("md: %s: metadata overlaps data\n",
5081 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5084 if (mddev
->bio_set
== NULL
)
5085 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
5087 spin_lock(&pers_lock
);
5088 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5089 if (!pers
|| !try_module_get(pers
->owner
)) {
5090 spin_unlock(&pers_lock
);
5091 if (mddev
->level
!= LEVEL_NONE
)
5092 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5095 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5099 spin_unlock(&pers_lock
);
5100 if (mddev
->level
!= pers
->level
) {
5101 mddev
->level
= pers
->level
;
5102 mddev
->new_level
= pers
->level
;
5104 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5106 if (mddev
->reshape_position
!= MaxSector
&&
5107 pers
->start_reshape
== NULL
) {
5108 /* This personality cannot handle reshaping... */
5109 module_put(pers
->owner
);
5113 if (pers
->sync_request
) {
5114 /* Warn if this is a potentially silly
5117 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5118 struct md_rdev
*rdev2
;
5121 rdev_for_each(rdev
, mddev
)
5122 rdev_for_each(rdev2
, mddev
) {
5124 rdev
->bdev
->bd_contains
==
5125 rdev2
->bdev
->bd_contains
) {
5127 "%s: WARNING: %s appears to be"
5128 " on the same physical disk as"
5131 bdevname(rdev
->bdev
,b
),
5132 bdevname(rdev2
->bdev
,b2
));
5139 "True protection against single-disk"
5140 " failure might be compromised.\n");
5143 mddev
->recovery
= 0;
5144 /* may be over-ridden by personality */
5145 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5147 mddev
->ok_start_degraded
= start_dirty_degraded
;
5149 if (start_readonly
&& mddev
->ro
== 0)
5150 mddev
->ro
= 2; /* read-only, but switch on first write */
5152 err
= pers
->run(mddev
);
5154 printk(KERN_ERR
"md: pers->run() failed ...\n");
5155 else if (pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5156 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5157 " but 'external_size' not in effect?\n", __func__
);
5159 "md: invalid array_size %llu > default size %llu\n",
5160 (unsigned long long)mddev
->array_sectors
/ 2,
5161 (unsigned long long)pers
->size(mddev
, 0, 0) / 2);
5164 if (err
== 0 && pers
->sync_request
&&
5165 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5166 struct bitmap
*bitmap
;
5168 bitmap
= bitmap_create(mddev
, -1);
5169 if (IS_ERR(bitmap
)) {
5170 err
= PTR_ERR(bitmap
);
5171 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5172 mdname(mddev
), err
);
5174 mddev
->bitmap
= bitmap
;
5178 mddev_detach(mddev
);
5180 pers
->free(mddev
, mddev
->private);
5181 mddev
->private = NULL
;
5182 module_put(pers
->owner
);
5183 bitmap_destroy(mddev
);
5187 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
5188 mddev
->queue
->backing_dev_info
.congested_fn
= md_congested
;
5189 blk_queue_merge_bvec(mddev
->queue
, md_mergeable_bvec
);
5191 if (pers
->sync_request
) {
5192 if (mddev
->kobj
.sd
&&
5193 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5195 "md: cannot register extra attributes for %s\n",
5197 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5198 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5201 atomic_set(&mddev
->writes_pending
,0);
5202 atomic_set(&mddev
->max_corr_read_errors
,
5203 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5204 mddev
->safemode
= 0;
5205 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5206 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5207 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5210 spin_lock(&mddev
->lock
);
5213 spin_unlock(&mddev
->lock
);
5214 rdev_for_each(rdev
, mddev
)
5215 if (rdev
->raid_disk
>= 0)
5216 if (sysfs_link_rdev(mddev
, rdev
))
5217 /* failure here is OK */;
5219 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5221 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
5222 md_update_sb(mddev
, 0);
5224 md_new_event(mddev
);
5225 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5226 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5227 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5230 EXPORT_SYMBOL_GPL(md_run
);
5232 static int do_md_run(struct mddev
*mddev
)
5236 err
= md_run(mddev
);
5239 err
= bitmap_load(mddev
);
5241 bitmap_destroy(mddev
);
5245 md_wakeup_thread(mddev
->thread
);
5246 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5248 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5249 revalidate_disk(mddev
->gendisk
);
5251 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5256 static int restart_array(struct mddev
*mddev
)
5258 struct gendisk
*disk
= mddev
->gendisk
;
5260 /* Complain if it has no devices */
5261 if (list_empty(&mddev
->disks
))
5267 mddev
->safemode
= 0;
5269 set_disk_ro(disk
, 0);
5270 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5272 /* Kick recovery or resync if necessary */
5273 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5274 md_wakeup_thread(mddev
->thread
);
5275 md_wakeup_thread(mddev
->sync_thread
);
5276 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5280 static void md_clean(struct mddev
*mddev
)
5282 mddev
->array_sectors
= 0;
5283 mddev
->external_size
= 0;
5284 mddev
->dev_sectors
= 0;
5285 mddev
->raid_disks
= 0;
5286 mddev
->recovery_cp
= 0;
5287 mddev
->resync_min
= 0;
5288 mddev
->resync_max
= MaxSector
;
5289 mddev
->reshape_position
= MaxSector
;
5290 mddev
->external
= 0;
5291 mddev
->persistent
= 0;
5292 mddev
->level
= LEVEL_NONE
;
5293 mddev
->clevel
[0] = 0;
5296 mddev
->metadata_type
[0] = 0;
5297 mddev
->chunk_sectors
= 0;
5298 mddev
->ctime
= mddev
->utime
= 0;
5300 mddev
->max_disks
= 0;
5302 mddev
->can_decrease_events
= 0;
5303 mddev
->delta_disks
= 0;
5304 mddev
->reshape_backwards
= 0;
5305 mddev
->new_level
= LEVEL_NONE
;
5306 mddev
->new_layout
= 0;
5307 mddev
->new_chunk_sectors
= 0;
5308 mddev
->curr_resync
= 0;
5309 atomic64_set(&mddev
->resync_mismatches
, 0);
5310 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5311 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5312 mddev
->recovery
= 0;
5315 mddev
->degraded
= 0;
5316 mddev
->safemode
= 0;
5317 mddev
->private = NULL
;
5318 mddev
->merge_check_needed
= 0;
5319 mddev
->bitmap_info
.offset
= 0;
5320 mddev
->bitmap_info
.default_offset
= 0;
5321 mddev
->bitmap_info
.default_space
= 0;
5322 mddev
->bitmap_info
.chunksize
= 0;
5323 mddev
->bitmap_info
.daemon_sleep
= 0;
5324 mddev
->bitmap_info
.max_write_behind
= 0;
5327 static void __md_stop_writes(struct mddev
*mddev
)
5329 if (mddev_is_clustered(mddev
))
5330 md_cluster_ops
->metadata_update_start(mddev
);
5331 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5332 flush_workqueue(md_misc_wq
);
5333 if (mddev
->sync_thread
) {
5334 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5335 md_reap_sync_thread(mddev
);
5338 del_timer_sync(&mddev
->safemode_timer
);
5340 bitmap_flush(mddev
);
5341 md_super_wait(mddev
);
5343 if (mddev
->ro
== 0 &&
5344 (!mddev
->in_sync
|| (mddev
->flags
& MD_UPDATE_SB_FLAGS
))) {
5345 /* mark array as shutdown cleanly */
5347 md_update_sb(mddev
, 1);
5349 if (mddev_is_clustered(mddev
))
5350 md_cluster_ops
->metadata_update_finish(mddev
);
5353 void md_stop_writes(struct mddev
*mddev
)
5355 mddev_lock_nointr(mddev
);
5356 __md_stop_writes(mddev
);
5357 mddev_unlock(mddev
);
5359 EXPORT_SYMBOL_GPL(md_stop_writes
);
5361 static void mddev_detach(struct mddev
*mddev
)
5363 struct bitmap
*bitmap
= mddev
->bitmap
;
5364 /* wait for behind writes to complete */
5365 if (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
5366 printk(KERN_INFO
"md:%s: behind writes in progress - waiting to stop.\n",
5368 /* need to kick something here to make sure I/O goes? */
5369 wait_event(bitmap
->behind_wait
,
5370 atomic_read(&bitmap
->behind_writes
) == 0);
5372 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5373 mddev
->pers
->quiesce(mddev
, 1);
5374 mddev
->pers
->quiesce(mddev
, 0);
5376 md_unregister_thread(&mddev
->thread
);
5378 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
5381 static void __md_stop(struct mddev
*mddev
)
5383 struct md_personality
*pers
= mddev
->pers
;
5384 mddev_detach(mddev
);
5385 spin_lock(&mddev
->lock
);
5388 spin_unlock(&mddev
->lock
);
5389 pers
->free(mddev
, mddev
->private);
5390 mddev
->private = NULL
;
5391 if (pers
->sync_request
&& mddev
->to_remove
== NULL
)
5392 mddev
->to_remove
= &md_redundancy_group
;
5393 module_put(pers
->owner
);
5394 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5397 void md_stop(struct mddev
*mddev
)
5399 /* stop the array and free an attached data structures.
5400 * This is called from dm-raid
5403 bitmap_destroy(mddev
);
5405 bioset_free(mddev
->bio_set
);
5408 EXPORT_SYMBOL_GPL(md_stop
);
5410 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5415 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5417 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5418 md_wakeup_thread(mddev
->thread
);
5420 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5421 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5422 if (mddev
->sync_thread
)
5423 /* Thread might be blocked waiting for metadata update
5424 * which will now never happen */
5425 wake_up_process(mddev
->sync_thread
->tsk
);
5427 mddev_unlock(mddev
);
5428 wait_event(resync_wait
, !test_bit(MD_RECOVERY_RUNNING
,
5430 mddev_lock_nointr(mddev
);
5432 mutex_lock(&mddev
->open_mutex
);
5433 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5434 mddev
->sync_thread
||
5435 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5436 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5437 printk("md: %s still in use.\n",mdname(mddev
));
5439 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5440 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5441 md_wakeup_thread(mddev
->thread
);
5447 __md_stop_writes(mddev
);
5453 set_disk_ro(mddev
->gendisk
, 1);
5454 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5455 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5456 md_wakeup_thread(mddev
->thread
);
5457 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5461 mutex_unlock(&mddev
->open_mutex
);
5466 * 0 - completely stop and dis-assemble array
5467 * 2 - stop but do not disassemble array
5469 static int do_md_stop(struct mddev
*mddev
, int mode
,
5470 struct block_device
*bdev
)
5472 struct gendisk
*disk
= mddev
->gendisk
;
5473 struct md_rdev
*rdev
;
5476 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5478 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5479 md_wakeup_thread(mddev
->thread
);
5481 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5482 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5483 if (mddev
->sync_thread
)
5484 /* Thread might be blocked waiting for metadata update
5485 * which will now never happen */
5486 wake_up_process(mddev
->sync_thread
->tsk
);
5488 mddev_unlock(mddev
);
5489 wait_event(resync_wait
, (mddev
->sync_thread
== NULL
&&
5490 !test_bit(MD_RECOVERY_RUNNING
,
5491 &mddev
->recovery
)));
5492 mddev_lock_nointr(mddev
);
5494 mutex_lock(&mddev
->open_mutex
);
5495 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5496 mddev
->sysfs_active
||
5497 mddev
->sync_thread
||
5498 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5499 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5500 printk("md: %s still in use.\n",mdname(mddev
));
5501 mutex_unlock(&mddev
->open_mutex
);
5503 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5504 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5505 md_wakeup_thread(mddev
->thread
);
5511 set_disk_ro(disk
, 0);
5513 __md_stop_writes(mddev
);
5515 mddev
->queue
->merge_bvec_fn
= NULL
;
5516 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5518 /* tell userspace to handle 'inactive' */
5519 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5521 rdev_for_each(rdev
, mddev
)
5522 if (rdev
->raid_disk
>= 0)
5523 sysfs_unlink_rdev(mddev
, rdev
);
5525 set_capacity(disk
, 0);
5526 mutex_unlock(&mddev
->open_mutex
);
5528 revalidate_disk(disk
);
5533 mutex_unlock(&mddev
->open_mutex
);
5535 * Free resources if final stop
5538 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5540 bitmap_destroy(mddev
);
5541 if (mddev
->bitmap_info
.file
) {
5542 struct file
*f
= mddev
->bitmap_info
.file
;
5543 spin_lock(&mddev
->lock
);
5544 mddev
->bitmap_info
.file
= NULL
;
5545 spin_unlock(&mddev
->lock
);
5548 mddev
->bitmap_info
.offset
= 0;
5550 export_array(mddev
);
5553 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5554 if (mddev
->hold_active
== UNTIL_STOP
)
5555 mddev
->hold_active
= 0;
5557 blk_integrity_unregister(disk
);
5558 md_new_event(mddev
);
5559 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5564 static void autorun_array(struct mddev
*mddev
)
5566 struct md_rdev
*rdev
;
5569 if (list_empty(&mddev
->disks
))
5572 printk(KERN_INFO
"md: running: ");
5574 rdev_for_each(rdev
, mddev
) {
5575 char b
[BDEVNAME_SIZE
];
5576 printk("<%s>", bdevname(rdev
->bdev
,b
));
5580 err
= do_md_run(mddev
);
5582 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5583 do_md_stop(mddev
, 0, NULL
);
5588 * lets try to run arrays based on all disks that have arrived
5589 * until now. (those are in pending_raid_disks)
5591 * the method: pick the first pending disk, collect all disks with
5592 * the same UUID, remove all from the pending list and put them into
5593 * the 'same_array' list. Then order this list based on superblock
5594 * update time (freshest comes first), kick out 'old' disks and
5595 * compare superblocks. If everything's fine then run it.
5597 * If "unit" is allocated, then bump its reference count
5599 static void autorun_devices(int part
)
5601 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5602 struct mddev
*mddev
;
5603 char b
[BDEVNAME_SIZE
];
5605 printk(KERN_INFO
"md: autorun ...\n");
5606 while (!list_empty(&pending_raid_disks
)) {
5609 LIST_HEAD(candidates
);
5610 rdev0
= list_entry(pending_raid_disks
.next
,
5611 struct md_rdev
, same_set
);
5613 printk(KERN_INFO
"md: considering %s ...\n",
5614 bdevname(rdev0
->bdev
,b
));
5615 INIT_LIST_HEAD(&candidates
);
5616 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5617 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5618 printk(KERN_INFO
"md: adding %s ...\n",
5619 bdevname(rdev
->bdev
,b
));
5620 list_move(&rdev
->same_set
, &candidates
);
5623 * now we have a set of devices, with all of them having
5624 * mostly sane superblocks. It's time to allocate the
5628 dev
= MKDEV(mdp_major
,
5629 rdev0
->preferred_minor
<< MdpMinorShift
);
5630 unit
= MINOR(dev
) >> MdpMinorShift
;
5632 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5635 if (rdev0
->preferred_minor
!= unit
) {
5636 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5637 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5641 md_probe(dev
, NULL
, NULL
);
5642 mddev
= mddev_find(dev
);
5643 if (!mddev
|| !mddev
->gendisk
) {
5647 "md: cannot allocate memory for md drive.\n");
5650 if (mddev_lock(mddev
))
5651 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5653 else if (mddev
->raid_disks
|| mddev
->major_version
5654 || !list_empty(&mddev
->disks
)) {
5656 "md: %s already running, cannot run %s\n",
5657 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5658 mddev_unlock(mddev
);
5660 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5661 mddev
->persistent
= 1;
5662 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5663 list_del_init(&rdev
->same_set
);
5664 if (bind_rdev_to_array(rdev
, mddev
))
5667 autorun_array(mddev
);
5668 mddev_unlock(mddev
);
5670 /* on success, candidates will be empty, on error
5673 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5674 list_del_init(&rdev
->same_set
);
5679 printk(KERN_INFO
"md: ... autorun DONE.\n");
5681 #endif /* !MODULE */
5683 static int get_version(void __user
*arg
)
5687 ver
.major
= MD_MAJOR_VERSION
;
5688 ver
.minor
= MD_MINOR_VERSION
;
5689 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5691 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5697 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
5699 mdu_array_info_t info
;
5700 int nr
,working
,insync
,failed
,spare
;
5701 struct md_rdev
*rdev
;
5703 nr
= working
= insync
= failed
= spare
= 0;
5705 rdev_for_each_rcu(rdev
, mddev
) {
5707 if (test_bit(Faulty
, &rdev
->flags
))
5711 if (test_bit(In_sync
, &rdev
->flags
))
5719 info
.major_version
= mddev
->major_version
;
5720 info
.minor_version
= mddev
->minor_version
;
5721 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5722 info
.ctime
= mddev
->ctime
;
5723 info
.level
= mddev
->level
;
5724 info
.size
= mddev
->dev_sectors
/ 2;
5725 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5728 info
.raid_disks
= mddev
->raid_disks
;
5729 info
.md_minor
= mddev
->md_minor
;
5730 info
.not_persistent
= !mddev
->persistent
;
5732 info
.utime
= mddev
->utime
;
5735 info
.state
= (1<<MD_SB_CLEAN
);
5736 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5737 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
5738 if (mddev_is_clustered(mddev
))
5739 info
.state
|= (1<<MD_SB_CLUSTERED
);
5740 info
.active_disks
= insync
;
5741 info
.working_disks
= working
;
5742 info
.failed_disks
= failed
;
5743 info
.spare_disks
= spare
;
5745 info
.layout
= mddev
->layout
;
5746 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5748 if (copy_to_user(arg
, &info
, sizeof(info
)))
5754 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
5756 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5760 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5765 spin_lock(&mddev
->lock
);
5766 /* bitmap disabled, zero the first byte and copy out */
5767 if (!mddev
->bitmap_info
.file
)
5768 file
->pathname
[0] = '\0';
5769 else if ((ptr
= d_path(&mddev
->bitmap_info
.file
->f_path
,
5770 file
->pathname
, sizeof(file
->pathname
))),
5774 memmove(file
->pathname
, ptr
,
5775 sizeof(file
->pathname
)-(ptr
-file
->pathname
));
5776 spin_unlock(&mddev
->lock
);
5779 copy_to_user(arg
, file
, sizeof(*file
)))
5786 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
5788 mdu_disk_info_t info
;
5789 struct md_rdev
*rdev
;
5791 if (copy_from_user(&info
, arg
, sizeof(info
)))
5795 rdev
= md_find_rdev_nr_rcu(mddev
, info
.number
);
5797 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5798 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5799 info
.raid_disk
= rdev
->raid_disk
;
5801 if (test_bit(Faulty
, &rdev
->flags
))
5802 info
.state
|= (1<<MD_DISK_FAULTY
);
5803 else if (test_bit(In_sync
, &rdev
->flags
)) {
5804 info
.state
|= (1<<MD_DISK_ACTIVE
);
5805 info
.state
|= (1<<MD_DISK_SYNC
);
5807 if (test_bit(WriteMostly
, &rdev
->flags
))
5808 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5810 info
.major
= info
.minor
= 0;
5811 info
.raid_disk
= -1;
5812 info
.state
= (1<<MD_DISK_REMOVED
);
5816 if (copy_to_user(arg
, &info
, sizeof(info
)))
5822 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
5824 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5825 struct md_rdev
*rdev
;
5826 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5828 if (mddev_is_clustered(mddev
) &&
5829 !(info
->state
& ((1 << MD_DISK_CLUSTER_ADD
) | (1 << MD_DISK_CANDIDATE
)))) {
5830 pr_err("%s: Cannot add to clustered mddev.\n",
5835 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5838 if (!mddev
->raid_disks
) {
5840 /* expecting a device which has a superblock */
5841 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5844 "md: md_import_device returned %ld\n",
5846 return PTR_ERR(rdev
);
5848 if (!list_empty(&mddev
->disks
)) {
5849 struct md_rdev
*rdev0
5850 = list_entry(mddev
->disks
.next
,
5851 struct md_rdev
, same_set
);
5852 err
= super_types
[mddev
->major_version
]
5853 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5856 "md: %s has different UUID to %s\n",
5857 bdevname(rdev
->bdev
,b
),
5858 bdevname(rdev0
->bdev
,b2
));
5863 err
= bind_rdev_to_array(rdev
, mddev
);
5870 * add_new_disk can be used once the array is assembled
5871 * to add "hot spares". They must already have a superblock
5876 if (!mddev
->pers
->hot_add_disk
) {
5878 "%s: personality does not support diskops!\n",
5882 if (mddev
->persistent
)
5883 rdev
= md_import_device(dev
, mddev
->major_version
,
5884 mddev
->minor_version
);
5886 rdev
= md_import_device(dev
, -1, -1);
5889 "md: md_import_device returned %ld\n",
5891 return PTR_ERR(rdev
);
5893 /* set saved_raid_disk if appropriate */
5894 if (!mddev
->persistent
) {
5895 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5896 info
->raid_disk
< mddev
->raid_disks
) {
5897 rdev
->raid_disk
= info
->raid_disk
;
5898 set_bit(In_sync
, &rdev
->flags
);
5899 clear_bit(Bitmap_sync
, &rdev
->flags
);
5901 rdev
->raid_disk
= -1;
5902 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5904 super_types
[mddev
->major_version
].
5905 validate_super(mddev
, rdev
);
5906 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5907 rdev
->raid_disk
!= info
->raid_disk
) {
5908 /* This was a hot-add request, but events doesn't
5909 * match, so reject it.
5915 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5916 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5917 set_bit(WriteMostly
, &rdev
->flags
);
5919 clear_bit(WriteMostly
, &rdev
->flags
);
5922 * check whether the device shows up in other nodes
5924 if (mddev_is_clustered(mddev
)) {
5925 if (info
->state
& (1 << MD_DISK_CANDIDATE
)) {
5926 /* Through --cluster-confirm */
5927 set_bit(Candidate
, &rdev
->flags
);
5928 err
= md_cluster_ops
->new_disk_ack(mddev
, true);
5933 } else if (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)) {
5934 /* --add initiated by this node */
5935 err
= md_cluster_ops
->add_new_disk_start(mddev
, rdev
);
5937 md_cluster_ops
->add_new_disk_finish(mddev
);
5944 rdev
->raid_disk
= -1;
5945 err
= bind_rdev_to_array(rdev
, mddev
);
5949 err
= add_bound_rdev(rdev
);
5950 if (mddev_is_clustered(mddev
) &&
5951 (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)))
5952 md_cluster_ops
->add_new_disk_finish(mddev
);
5956 /* otherwise, add_new_disk is only allowed
5957 * for major_version==0 superblocks
5959 if (mddev
->major_version
!= 0) {
5960 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5965 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5967 rdev
= md_import_device(dev
, -1, 0);
5970 "md: error, md_import_device() returned %ld\n",
5972 return PTR_ERR(rdev
);
5974 rdev
->desc_nr
= info
->number
;
5975 if (info
->raid_disk
< mddev
->raid_disks
)
5976 rdev
->raid_disk
= info
->raid_disk
;
5978 rdev
->raid_disk
= -1;
5980 if (rdev
->raid_disk
< mddev
->raid_disks
)
5981 if (info
->state
& (1<<MD_DISK_SYNC
))
5982 set_bit(In_sync
, &rdev
->flags
);
5984 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5985 set_bit(WriteMostly
, &rdev
->flags
);
5987 if (!mddev
->persistent
) {
5988 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5989 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5991 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5992 rdev
->sectors
= rdev
->sb_start
;
5994 err
= bind_rdev_to_array(rdev
, mddev
);
6004 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
6006 char b
[BDEVNAME_SIZE
];
6007 struct md_rdev
*rdev
;
6009 rdev
= find_rdev(mddev
, dev
);
6013 if (mddev_is_clustered(mddev
))
6014 md_cluster_ops
->metadata_update_start(mddev
);
6016 clear_bit(Blocked
, &rdev
->flags
);
6017 remove_and_add_spares(mddev
, rdev
);
6019 if (rdev
->raid_disk
>= 0)
6022 if (mddev_is_clustered(mddev
))
6023 md_cluster_ops
->remove_disk(mddev
, rdev
);
6025 md_kick_rdev_from_array(rdev
);
6026 md_update_sb(mddev
, 1);
6027 md_new_event(mddev
);
6029 if (mddev_is_clustered(mddev
))
6030 md_cluster_ops
->metadata_update_finish(mddev
);
6034 if (mddev_is_clustered(mddev
))
6035 md_cluster_ops
->metadata_update_cancel(mddev
);
6036 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
6037 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6041 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
6043 char b
[BDEVNAME_SIZE
];
6045 struct md_rdev
*rdev
;
6050 if (mddev
->major_version
!= 0) {
6051 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
6052 " version-0 superblocks.\n",
6056 if (!mddev
->pers
->hot_add_disk
) {
6058 "%s: personality does not support diskops!\n",
6063 rdev
= md_import_device(dev
, -1, 0);
6066 "md: error, md_import_device() returned %ld\n",
6071 if (mddev
->persistent
)
6072 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6074 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6076 rdev
->sectors
= rdev
->sb_start
;
6078 if (test_bit(Faulty
, &rdev
->flags
)) {
6080 "md: can not hot-add faulty %s disk to %s!\n",
6081 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6086 if (mddev_is_clustered(mddev
))
6087 md_cluster_ops
->metadata_update_start(mddev
);
6088 clear_bit(In_sync
, &rdev
->flags
);
6090 rdev
->saved_raid_disk
= -1;
6091 err
= bind_rdev_to_array(rdev
, mddev
);
6093 goto abort_clustered
;
6096 * The rest should better be atomic, we can have disk failures
6097 * noticed in interrupt contexts ...
6100 rdev
->raid_disk
= -1;
6102 md_update_sb(mddev
, 1);
6104 if (mddev_is_clustered(mddev
))
6105 md_cluster_ops
->metadata_update_finish(mddev
);
6107 * Kick recovery, maybe this spare has to be added to the
6108 * array immediately.
6110 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6111 md_wakeup_thread(mddev
->thread
);
6112 md_new_event(mddev
);
6116 if (mddev_is_clustered(mddev
))
6117 md_cluster_ops
->metadata_update_cancel(mddev
);
6123 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
6128 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
6130 if (mddev
->recovery
|| mddev
->sync_thread
)
6132 /* we should be able to change the bitmap.. */
6136 struct inode
*inode
;
6139 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
)
6140 return -EEXIST
; /* cannot add when bitmap is present */
6144 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
6149 inode
= f
->f_mapping
->host
;
6150 if (!S_ISREG(inode
->i_mode
)) {
6151 printk(KERN_ERR
"%s: error: bitmap file must be a regular file\n",
6154 } else if (!(f
->f_mode
& FMODE_WRITE
)) {
6155 printk(KERN_ERR
"%s: error: bitmap file must open for write\n",
6158 } else if (atomic_read(&inode
->i_writecount
) != 1) {
6159 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
6167 mddev
->bitmap_info
.file
= f
;
6168 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6169 } else if (mddev
->bitmap
== NULL
)
6170 return -ENOENT
; /* cannot remove what isn't there */
6173 mddev
->pers
->quiesce(mddev
, 1);
6175 struct bitmap
*bitmap
;
6177 bitmap
= bitmap_create(mddev
, -1);
6178 if (!IS_ERR(bitmap
)) {
6179 mddev
->bitmap
= bitmap
;
6180 err
= bitmap_load(mddev
);
6182 err
= PTR_ERR(bitmap
);
6184 if (fd
< 0 || err
) {
6185 bitmap_destroy(mddev
);
6186 fd
= -1; /* make sure to put the file */
6188 mddev
->pers
->quiesce(mddev
, 0);
6191 struct file
*f
= mddev
->bitmap_info
.file
;
6193 spin_lock(&mddev
->lock
);
6194 mddev
->bitmap_info
.file
= NULL
;
6195 spin_unlock(&mddev
->lock
);
6204 * set_array_info is used two different ways
6205 * The original usage is when creating a new array.
6206 * In this usage, raid_disks is > 0 and it together with
6207 * level, size, not_persistent,layout,chunksize determine the
6208 * shape of the array.
6209 * This will always create an array with a type-0.90.0 superblock.
6210 * The newer usage is when assembling an array.
6211 * In this case raid_disks will be 0, and the major_version field is
6212 * use to determine which style super-blocks are to be found on the devices.
6213 * The minor and patch _version numbers are also kept incase the
6214 * super_block handler wishes to interpret them.
6216 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6219 if (info
->raid_disks
== 0) {
6220 /* just setting version number for superblock loading */
6221 if (info
->major_version
< 0 ||
6222 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6223 super_types
[info
->major_version
].name
== NULL
) {
6224 /* maybe try to auto-load a module? */
6226 "md: superblock version %d not known\n",
6227 info
->major_version
);
6230 mddev
->major_version
= info
->major_version
;
6231 mddev
->minor_version
= info
->minor_version
;
6232 mddev
->patch_version
= info
->patch_version
;
6233 mddev
->persistent
= !info
->not_persistent
;
6234 /* ensure mddev_put doesn't delete this now that there
6235 * is some minimal configuration.
6237 mddev
->ctime
= get_seconds();
6240 mddev
->major_version
= MD_MAJOR_VERSION
;
6241 mddev
->minor_version
= MD_MINOR_VERSION
;
6242 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6243 mddev
->ctime
= get_seconds();
6245 mddev
->level
= info
->level
;
6246 mddev
->clevel
[0] = 0;
6247 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6248 mddev
->raid_disks
= info
->raid_disks
;
6249 /* don't set md_minor, it is determined by which /dev/md* was
6252 if (info
->state
& (1<<MD_SB_CLEAN
))
6253 mddev
->recovery_cp
= MaxSector
;
6255 mddev
->recovery_cp
= 0;
6256 mddev
->persistent
= ! info
->not_persistent
;
6257 mddev
->external
= 0;
6259 mddev
->layout
= info
->layout
;
6260 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6262 mddev
->max_disks
= MD_SB_DISKS
;
6264 if (mddev
->persistent
)
6266 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6268 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6269 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6270 mddev
->bitmap_info
.offset
= 0;
6272 mddev
->reshape_position
= MaxSector
;
6275 * Generate a 128 bit UUID
6277 get_random_bytes(mddev
->uuid
, 16);
6279 mddev
->new_level
= mddev
->level
;
6280 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6281 mddev
->new_layout
= mddev
->layout
;
6282 mddev
->delta_disks
= 0;
6283 mddev
->reshape_backwards
= 0;
6288 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6290 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6292 if (mddev
->external_size
)
6295 mddev
->array_sectors
= array_sectors
;
6297 EXPORT_SYMBOL(md_set_array_sectors
);
6299 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6301 struct md_rdev
*rdev
;
6303 int fit
= (num_sectors
== 0);
6305 if (mddev
->pers
->resize
== NULL
)
6307 /* The "num_sectors" is the number of sectors of each device that
6308 * is used. This can only make sense for arrays with redundancy.
6309 * linear and raid0 always use whatever space is available. We can only
6310 * consider changing this number if no resync or reconstruction is
6311 * happening, and if the new size is acceptable. It must fit before the
6312 * sb_start or, if that is <data_offset, it must fit before the size
6313 * of each device. If num_sectors is zero, we find the largest size
6316 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6322 rdev_for_each(rdev
, mddev
) {
6323 sector_t avail
= rdev
->sectors
;
6325 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6326 num_sectors
= avail
;
6327 if (avail
< num_sectors
)
6330 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6332 revalidate_disk(mddev
->gendisk
);
6336 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6339 struct md_rdev
*rdev
;
6340 /* change the number of raid disks */
6341 if (mddev
->pers
->check_reshape
== NULL
)
6345 if (raid_disks
<= 0 ||
6346 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6348 if (mddev
->sync_thread
||
6349 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6350 mddev
->reshape_position
!= MaxSector
)
6353 rdev_for_each(rdev
, mddev
) {
6354 if (mddev
->raid_disks
< raid_disks
&&
6355 rdev
->data_offset
< rdev
->new_data_offset
)
6357 if (mddev
->raid_disks
> raid_disks
&&
6358 rdev
->data_offset
> rdev
->new_data_offset
)
6362 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6363 if (mddev
->delta_disks
< 0)
6364 mddev
->reshape_backwards
= 1;
6365 else if (mddev
->delta_disks
> 0)
6366 mddev
->reshape_backwards
= 0;
6368 rv
= mddev
->pers
->check_reshape(mddev
);
6370 mddev
->delta_disks
= 0;
6371 mddev
->reshape_backwards
= 0;
6377 * update_array_info is used to change the configuration of an
6379 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6380 * fields in the info are checked against the array.
6381 * Any differences that cannot be handled will cause an error.
6382 * Normally, only one change can be managed at a time.
6384 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6390 /* calculate expected state,ignoring low bits */
6391 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6392 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6394 if (mddev
->major_version
!= info
->major_version
||
6395 mddev
->minor_version
!= info
->minor_version
||
6396 /* mddev->patch_version != info->patch_version || */
6397 mddev
->ctime
!= info
->ctime
||
6398 mddev
->level
!= info
->level
||
6399 /* mddev->layout != info->layout || */
6400 mddev
->persistent
!= !info
->not_persistent
||
6401 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6402 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6403 ((state
^info
->state
) & 0xfffffe00)
6406 /* Check there is only one change */
6407 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6409 if (mddev
->raid_disks
!= info
->raid_disks
)
6411 if (mddev
->layout
!= info
->layout
)
6413 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6420 if (mddev
->layout
!= info
->layout
) {
6422 * we don't need to do anything at the md level, the
6423 * personality will take care of it all.
6425 if (mddev
->pers
->check_reshape
== NULL
)
6428 mddev
->new_layout
= info
->layout
;
6429 rv
= mddev
->pers
->check_reshape(mddev
);
6431 mddev
->new_layout
= mddev
->layout
;
6435 if (mddev_is_clustered(mddev
))
6436 md_cluster_ops
->metadata_update_start(mddev
);
6437 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6438 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6440 if (mddev
->raid_disks
!= info
->raid_disks
)
6441 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6443 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6444 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
) {
6448 if (mddev
->recovery
|| mddev
->sync_thread
) {
6452 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6453 struct bitmap
*bitmap
;
6454 /* add the bitmap */
6455 if (mddev
->bitmap
) {
6459 if (mddev
->bitmap_info
.default_offset
== 0) {
6463 mddev
->bitmap_info
.offset
=
6464 mddev
->bitmap_info
.default_offset
;
6465 mddev
->bitmap_info
.space
=
6466 mddev
->bitmap_info
.default_space
;
6467 mddev
->pers
->quiesce(mddev
, 1);
6468 bitmap
= bitmap_create(mddev
, -1);
6469 if (!IS_ERR(bitmap
)) {
6470 mddev
->bitmap
= bitmap
;
6471 rv
= bitmap_load(mddev
);
6473 rv
= PTR_ERR(bitmap
);
6475 bitmap_destroy(mddev
);
6476 mddev
->pers
->quiesce(mddev
, 0);
6478 /* remove the bitmap */
6479 if (!mddev
->bitmap
) {
6483 if (mddev
->bitmap
->storage
.file
) {
6487 mddev
->pers
->quiesce(mddev
, 1);
6488 bitmap_destroy(mddev
);
6489 mddev
->pers
->quiesce(mddev
, 0);
6490 mddev
->bitmap_info
.offset
= 0;
6493 md_update_sb(mddev
, 1);
6494 if (mddev_is_clustered(mddev
))
6495 md_cluster_ops
->metadata_update_finish(mddev
);
6498 if (mddev_is_clustered(mddev
))
6499 md_cluster_ops
->metadata_update_cancel(mddev
);
6503 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6505 struct md_rdev
*rdev
;
6508 if (mddev
->pers
== NULL
)
6512 rdev
= find_rdev_rcu(mddev
, dev
);
6516 md_error(mddev
, rdev
);
6517 if (!test_bit(Faulty
, &rdev
->flags
))
6525 * We have a problem here : there is no easy way to give a CHS
6526 * virtual geometry. We currently pretend that we have a 2 heads
6527 * 4 sectors (with a BIG number of cylinders...). This drives
6528 * dosfs just mad... ;-)
6530 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6532 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6536 geo
->cylinders
= mddev
->array_sectors
/ 8;
6540 static inline bool md_ioctl_valid(unsigned int cmd
)
6545 case GET_ARRAY_INFO
:
6546 case GET_BITMAP_FILE
:
6549 case HOT_REMOVE_DISK
:
6552 case RESTART_ARRAY_RW
:
6554 case SET_ARRAY_INFO
:
6555 case SET_BITMAP_FILE
:
6556 case SET_DISK_FAULTY
:
6559 case CLUSTERED_DISK_NACK
:
6566 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6567 unsigned int cmd
, unsigned long arg
)
6570 void __user
*argp
= (void __user
*)arg
;
6571 struct mddev
*mddev
= NULL
;
6574 if (!md_ioctl_valid(cmd
))
6579 case GET_ARRAY_INFO
:
6583 if (!capable(CAP_SYS_ADMIN
))
6588 * Commands dealing with the RAID driver but not any
6593 err
= get_version(argp
);
6599 autostart_arrays(arg
);
6606 * Commands creating/starting a new array:
6609 mddev
= bdev
->bd_disk
->private_data
;
6616 /* Some actions do not requires the mutex */
6618 case GET_ARRAY_INFO
:
6619 if (!mddev
->raid_disks
&& !mddev
->external
)
6622 err
= get_array_info(mddev
, argp
);
6626 if (!mddev
->raid_disks
&& !mddev
->external
)
6629 err
= get_disk_info(mddev
, argp
);
6632 case SET_DISK_FAULTY
:
6633 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6636 case GET_BITMAP_FILE
:
6637 err
= get_bitmap_file(mddev
, argp
);
6642 if (cmd
== ADD_NEW_DISK
)
6643 /* need to ensure md_delayed_delete() has completed */
6644 flush_workqueue(md_misc_wq
);
6646 if (cmd
== HOT_REMOVE_DISK
)
6647 /* need to ensure recovery thread has run */
6648 wait_event_interruptible_timeout(mddev
->sb_wait
,
6649 !test_bit(MD_RECOVERY_NEEDED
,
6651 msecs_to_jiffies(5000));
6652 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
6653 /* Need to flush page cache, and ensure no-one else opens
6656 mutex_lock(&mddev
->open_mutex
);
6657 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
6658 mutex_unlock(&mddev
->open_mutex
);
6662 set_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6663 mutex_unlock(&mddev
->open_mutex
);
6664 sync_blockdev(bdev
);
6666 err
= mddev_lock(mddev
);
6669 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6674 if (cmd
== SET_ARRAY_INFO
) {
6675 mdu_array_info_t info
;
6677 memset(&info
, 0, sizeof(info
));
6678 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6683 err
= update_array_info(mddev
, &info
);
6685 printk(KERN_WARNING
"md: couldn't update"
6686 " array info. %d\n", err
);
6691 if (!list_empty(&mddev
->disks
)) {
6693 "md: array %s already has disks!\n",
6698 if (mddev
->raid_disks
) {
6700 "md: array %s already initialised!\n",
6705 err
= set_array_info(mddev
, &info
);
6707 printk(KERN_WARNING
"md: couldn't set"
6708 " array info. %d\n", err
);
6715 * Commands querying/configuring an existing array:
6717 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6718 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6719 if ((!mddev
->raid_disks
&& !mddev
->external
)
6720 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6721 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6722 && cmd
!= GET_BITMAP_FILE
) {
6728 * Commands even a read-only array can execute:
6731 case RESTART_ARRAY_RW
:
6732 err
= restart_array(mddev
);
6736 err
= do_md_stop(mddev
, 0, bdev
);
6740 err
= md_set_readonly(mddev
, bdev
);
6743 case HOT_REMOVE_DISK
:
6744 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6748 /* We can support ADD_NEW_DISK on read-only arrays
6749 * on if we are re-adding a preexisting device.
6750 * So require mddev->pers and MD_DISK_SYNC.
6753 mdu_disk_info_t info
;
6754 if (copy_from_user(&info
, argp
, sizeof(info
)))
6756 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6757 /* Need to clear read-only for this */
6760 err
= add_new_disk(mddev
, &info
);
6766 if (get_user(ro
, (int __user
*)(arg
))) {
6772 /* if the bdev is going readonly the value of mddev->ro
6773 * does not matter, no writes are coming
6778 /* are we are already prepared for writes? */
6782 /* transitioning to readauto need only happen for
6783 * arrays that call md_write_start
6786 err
= restart_array(mddev
);
6789 set_disk_ro(mddev
->gendisk
, 0);
6796 * The remaining ioctls are changing the state of the
6797 * superblock, so we do not allow them on read-only arrays.
6799 if (mddev
->ro
&& mddev
->pers
) {
6800 if (mddev
->ro
== 2) {
6802 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6803 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6804 /* mddev_unlock will wake thread */
6805 /* If a device failed while we were read-only, we
6806 * need to make sure the metadata is updated now.
6808 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6809 mddev_unlock(mddev
);
6810 wait_event(mddev
->sb_wait
,
6811 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6812 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6813 mddev_lock_nointr(mddev
);
6824 mdu_disk_info_t info
;
6825 if (copy_from_user(&info
, argp
, sizeof(info
)))
6828 err
= add_new_disk(mddev
, &info
);
6832 case CLUSTERED_DISK_NACK
:
6833 if (mddev_is_clustered(mddev
))
6834 md_cluster_ops
->new_disk_ack(mddev
, false);
6840 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6844 err
= do_md_run(mddev
);
6847 case SET_BITMAP_FILE
:
6848 err
= set_bitmap_file(mddev
, (int)arg
);
6857 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6859 mddev
->hold_active
= 0;
6860 mddev_unlock(mddev
);
6864 #ifdef CONFIG_COMPAT
6865 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6866 unsigned int cmd
, unsigned long arg
)
6869 case HOT_REMOVE_DISK
:
6871 case SET_DISK_FAULTY
:
6872 case SET_BITMAP_FILE
:
6873 /* These take in integer arg, do not convert */
6876 arg
= (unsigned long)compat_ptr(arg
);
6880 return md_ioctl(bdev
, mode
, cmd
, arg
);
6882 #endif /* CONFIG_COMPAT */
6884 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6887 * Succeed if we can lock the mddev, which confirms that
6888 * it isn't being stopped right now.
6890 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6896 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6897 /* we are racing with mddev_put which is discarding this
6901 /* Wait until bdev->bd_disk is definitely gone */
6902 flush_workqueue(md_misc_wq
);
6903 /* Then retry the open from the top */
6904 return -ERESTARTSYS
;
6906 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6908 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6912 atomic_inc(&mddev
->openers
);
6913 clear_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6914 mutex_unlock(&mddev
->open_mutex
);
6916 check_disk_change(bdev
);
6921 static void md_release(struct gendisk
*disk
, fmode_t mode
)
6923 struct mddev
*mddev
= disk
->private_data
;
6926 atomic_dec(&mddev
->openers
);
6930 static int md_media_changed(struct gendisk
*disk
)
6932 struct mddev
*mddev
= disk
->private_data
;
6934 return mddev
->changed
;
6937 static int md_revalidate(struct gendisk
*disk
)
6939 struct mddev
*mddev
= disk
->private_data
;
6944 static const struct block_device_operations md_fops
=
6946 .owner
= THIS_MODULE
,
6948 .release
= md_release
,
6950 #ifdef CONFIG_COMPAT
6951 .compat_ioctl
= md_compat_ioctl
,
6953 .getgeo
= md_getgeo
,
6954 .media_changed
= md_media_changed
,
6955 .revalidate_disk
= md_revalidate
,
6958 static int md_thread(void *arg
)
6960 struct md_thread
*thread
= arg
;
6963 * md_thread is a 'system-thread', it's priority should be very
6964 * high. We avoid resource deadlocks individually in each
6965 * raid personality. (RAID5 does preallocation) We also use RR and
6966 * the very same RT priority as kswapd, thus we will never get
6967 * into a priority inversion deadlock.
6969 * we definitely have to have equal or higher priority than
6970 * bdflush, otherwise bdflush will deadlock if there are too
6971 * many dirty RAID5 blocks.
6974 allow_signal(SIGKILL
);
6975 while (!kthread_should_stop()) {
6977 /* We need to wait INTERRUPTIBLE so that
6978 * we don't add to the load-average.
6979 * That means we need to be sure no signals are
6982 if (signal_pending(current
))
6983 flush_signals(current
);
6985 wait_event_interruptible_timeout
6987 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6988 || kthread_should_stop(),
6991 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6992 if (!kthread_should_stop())
6993 thread
->run(thread
);
6999 void md_wakeup_thread(struct md_thread
*thread
)
7002 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
7003 set_bit(THREAD_WAKEUP
, &thread
->flags
);
7004 wake_up(&thread
->wqueue
);
7007 EXPORT_SYMBOL(md_wakeup_thread
);
7009 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
7010 struct mddev
*mddev
, const char *name
)
7012 struct md_thread
*thread
;
7014 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
7018 init_waitqueue_head(&thread
->wqueue
);
7021 thread
->mddev
= mddev
;
7022 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
7023 thread
->tsk
= kthread_run(md_thread
, thread
,
7025 mdname(thread
->mddev
),
7027 if (IS_ERR(thread
->tsk
)) {
7033 EXPORT_SYMBOL(md_register_thread
);
7035 void md_unregister_thread(struct md_thread
**threadp
)
7037 struct md_thread
*thread
= *threadp
;
7040 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
7041 /* Locking ensures that mddev_unlock does not wake_up a
7042 * non-existent thread
7044 spin_lock(&pers_lock
);
7046 spin_unlock(&pers_lock
);
7048 kthread_stop(thread
->tsk
);
7051 EXPORT_SYMBOL(md_unregister_thread
);
7053 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
7055 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
7058 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
7060 mddev
->pers
->error_handler(mddev
,rdev
);
7061 if (mddev
->degraded
)
7062 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7063 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7064 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7065 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7066 md_wakeup_thread(mddev
->thread
);
7067 if (mddev
->event_work
.func
)
7068 queue_work(md_misc_wq
, &mddev
->event_work
);
7069 md_new_event_inintr(mddev
);
7071 EXPORT_SYMBOL(md_error
);
7073 /* seq_file implementation /proc/mdstat */
7075 static void status_unused(struct seq_file
*seq
)
7078 struct md_rdev
*rdev
;
7080 seq_printf(seq
, "unused devices: ");
7082 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
7083 char b
[BDEVNAME_SIZE
];
7085 seq_printf(seq
, "%s ",
7086 bdevname(rdev
->bdev
,b
));
7089 seq_printf(seq
, "<none>");
7091 seq_printf(seq
, "\n");
7094 static void status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
7096 sector_t max_sectors
, resync
, res
;
7097 unsigned long dt
, db
;
7100 unsigned int per_milli
;
7102 if (mddev
->curr_resync
<= 3)
7105 resync
= mddev
->curr_resync
7106 - atomic_read(&mddev
->recovery_active
);
7108 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7109 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7110 max_sectors
= mddev
->resync_max_sectors
;
7112 max_sectors
= mddev
->dev_sectors
;
7114 WARN_ON(max_sectors
== 0);
7115 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7116 * in a sector_t, and (max_sectors>>scale) will fit in a
7117 * u32, as those are the requirements for sector_div.
7118 * Thus 'scale' must be at least 10
7121 if (sizeof(sector_t
) > sizeof(unsigned long)) {
7122 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
7125 res
= (resync
>>scale
)*1000;
7126 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
7130 int i
, x
= per_milli
/50, y
= 20-x
;
7131 seq_printf(seq
, "[");
7132 for (i
= 0; i
< x
; i
++)
7133 seq_printf(seq
, "=");
7134 seq_printf(seq
, ">");
7135 for (i
= 0; i
< y
; i
++)
7136 seq_printf(seq
, ".");
7137 seq_printf(seq
, "] ");
7139 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
7140 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
7142 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
7144 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
7145 "resync" : "recovery"))),
7146 per_milli
/10, per_milli
% 10,
7147 (unsigned long long) resync
/2,
7148 (unsigned long long) max_sectors
/2);
7151 * dt: time from mark until now
7152 * db: blocks written from mark until now
7153 * rt: remaining time
7155 * rt is a sector_t, so could be 32bit or 64bit.
7156 * So we divide before multiply in case it is 32bit and close
7158 * We scale the divisor (db) by 32 to avoid losing precision
7159 * near the end of resync when the number of remaining sectors
7161 * We then divide rt by 32 after multiplying by db to compensate.
7162 * The '+1' avoids division by zero if db is very small.
7164 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
7166 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
7167 - mddev
->resync_mark_cnt
;
7169 rt
= max_sectors
- resync
; /* number of remaining sectors */
7170 sector_div(rt
, db
/32+1);
7174 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
7175 ((unsigned long)rt
% 60)/6);
7177 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
7180 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
7182 struct list_head
*tmp
;
7184 struct mddev
*mddev
;
7192 spin_lock(&all_mddevs_lock
);
7193 list_for_each(tmp
,&all_mddevs
)
7195 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7197 spin_unlock(&all_mddevs_lock
);
7200 spin_unlock(&all_mddevs_lock
);
7202 return (void*)2;/* tail */
7206 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7208 struct list_head
*tmp
;
7209 struct mddev
*next_mddev
, *mddev
= v
;
7215 spin_lock(&all_mddevs_lock
);
7217 tmp
= all_mddevs
.next
;
7219 tmp
= mddev
->all_mddevs
.next
;
7220 if (tmp
!= &all_mddevs
)
7221 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7223 next_mddev
= (void*)2;
7226 spin_unlock(&all_mddevs_lock
);
7234 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7236 struct mddev
*mddev
= v
;
7238 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7242 static int md_seq_show(struct seq_file
*seq
, void *v
)
7244 struct mddev
*mddev
= v
;
7246 struct md_rdev
*rdev
;
7248 if (v
== (void*)1) {
7249 struct md_personality
*pers
;
7250 seq_printf(seq
, "Personalities : ");
7251 spin_lock(&pers_lock
);
7252 list_for_each_entry(pers
, &pers_list
, list
)
7253 seq_printf(seq
, "[%s] ", pers
->name
);
7255 spin_unlock(&pers_lock
);
7256 seq_printf(seq
, "\n");
7257 seq
->poll_event
= atomic_read(&md_event_count
);
7260 if (v
== (void*)2) {
7265 spin_lock(&mddev
->lock
);
7266 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7267 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7268 mddev
->pers
? "" : "in");
7271 seq_printf(seq
, " (read-only)");
7273 seq_printf(seq
, " (auto-read-only)");
7274 seq_printf(seq
, " %s", mddev
->pers
->name
);
7279 rdev_for_each_rcu(rdev
, mddev
) {
7280 char b
[BDEVNAME_SIZE
];
7281 seq_printf(seq
, " %s[%d]",
7282 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7283 if (test_bit(WriteMostly
, &rdev
->flags
))
7284 seq_printf(seq
, "(W)");
7285 if (test_bit(Faulty
, &rdev
->flags
)) {
7286 seq_printf(seq
, "(F)");
7289 if (rdev
->raid_disk
< 0)
7290 seq_printf(seq
, "(S)"); /* spare */
7291 if (test_bit(Replacement
, &rdev
->flags
))
7292 seq_printf(seq
, "(R)");
7293 sectors
+= rdev
->sectors
;
7297 if (!list_empty(&mddev
->disks
)) {
7299 seq_printf(seq
, "\n %llu blocks",
7300 (unsigned long long)
7301 mddev
->array_sectors
/ 2);
7303 seq_printf(seq
, "\n %llu blocks",
7304 (unsigned long long)sectors
/ 2);
7306 if (mddev
->persistent
) {
7307 if (mddev
->major_version
!= 0 ||
7308 mddev
->minor_version
!= 90) {
7309 seq_printf(seq
," super %d.%d",
7310 mddev
->major_version
,
7311 mddev
->minor_version
);
7313 } else if (mddev
->external
)
7314 seq_printf(seq
, " super external:%s",
7315 mddev
->metadata_type
);
7317 seq_printf(seq
, " super non-persistent");
7320 mddev
->pers
->status(seq
, mddev
);
7321 seq_printf(seq
, "\n ");
7322 if (mddev
->pers
->sync_request
) {
7323 if (mddev
->curr_resync
> 2) {
7324 status_resync(seq
, mddev
);
7325 seq_printf(seq
, "\n ");
7326 } else if (mddev
->curr_resync
>= 1)
7327 seq_printf(seq
, "\tresync=DELAYED\n ");
7328 else if (mddev
->recovery_cp
< MaxSector
)
7329 seq_printf(seq
, "\tresync=PENDING\n ");
7332 seq_printf(seq
, "\n ");
7334 bitmap_status(seq
, mddev
->bitmap
);
7336 seq_printf(seq
, "\n");
7338 spin_unlock(&mddev
->lock
);
7343 static const struct seq_operations md_seq_ops
= {
7344 .start
= md_seq_start
,
7345 .next
= md_seq_next
,
7346 .stop
= md_seq_stop
,
7347 .show
= md_seq_show
,
7350 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7352 struct seq_file
*seq
;
7355 error
= seq_open(file
, &md_seq_ops
);
7359 seq
= file
->private_data
;
7360 seq
->poll_event
= atomic_read(&md_event_count
);
7364 static int md_unloading
;
7365 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7367 struct seq_file
*seq
= filp
->private_data
;
7371 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;
7372 poll_wait(filp
, &md_event_waiters
, wait
);
7374 /* always allow read */
7375 mask
= POLLIN
| POLLRDNORM
;
7377 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7378 mask
|= POLLERR
| POLLPRI
;
7382 static const struct file_operations md_seq_fops
= {
7383 .owner
= THIS_MODULE
,
7384 .open
= md_seq_open
,
7386 .llseek
= seq_lseek
,
7387 .release
= seq_release_private
,
7388 .poll
= mdstat_poll
,
7391 int register_md_personality(struct md_personality
*p
)
7393 printk(KERN_INFO
"md: %s personality registered for level %d\n",
7395 spin_lock(&pers_lock
);
7396 list_add_tail(&p
->list
, &pers_list
);
7397 spin_unlock(&pers_lock
);
7400 EXPORT_SYMBOL(register_md_personality
);
7402 int unregister_md_personality(struct md_personality
*p
)
7404 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7405 spin_lock(&pers_lock
);
7406 list_del_init(&p
->list
);
7407 spin_unlock(&pers_lock
);
7410 EXPORT_SYMBOL(unregister_md_personality
);
7412 int register_md_cluster_operations(struct md_cluster_operations
*ops
, struct module
*module
)
7414 if (md_cluster_ops
!= NULL
)
7416 spin_lock(&pers_lock
);
7417 md_cluster_ops
= ops
;
7418 md_cluster_mod
= module
;
7419 spin_unlock(&pers_lock
);
7422 EXPORT_SYMBOL(register_md_cluster_operations
);
7424 int unregister_md_cluster_operations(void)
7426 spin_lock(&pers_lock
);
7427 md_cluster_ops
= NULL
;
7428 spin_unlock(&pers_lock
);
7431 EXPORT_SYMBOL(unregister_md_cluster_operations
);
7433 int md_setup_cluster(struct mddev
*mddev
, int nodes
)
7437 err
= request_module("md-cluster");
7439 pr_err("md-cluster module not found.\n");
7443 spin_lock(&pers_lock
);
7444 if (!md_cluster_ops
|| !try_module_get(md_cluster_mod
)) {
7445 spin_unlock(&pers_lock
);
7448 spin_unlock(&pers_lock
);
7450 return md_cluster_ops
->join(mddev
, nodes
);
7453 void md_cluster_stop(struct mddev
*mddev
)
7455 if (!md_cluster_ops
)
7457 md_cluster_ops
->leave(mddev
);
7458 module_put(md_cluster_mod
);
7461 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7463 struct md_rdev
*rdev
;
7469 rdev_for_each_rcu(rdev
, mddev
) {
7470 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7471 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7472 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7473 atomic_read(&disk
->sync_io
);
7474 /* sync IO will cause sync_io to increase before the disk_stats
7475 * as sync_io is counted when a request starts, and
7476 * disk_stats is counted when it completes.
7477 * So resync activity will cause curr_events to be smaller than
7478 * when there was no such activity.
7479 * non-sync IO will cause disk_stat to increase without
7480 * increasing sync_io so curr_events will (eventually)
7481 * be larger than it was before. Once it becomes
7482 * substantially larger, the test below will cause
7483 * the array to appear non-idle, and resync will slow
7485 * If there is a lot of outstanding resync activity when
7486 * we set last_event to curr_events, then all that activity
7487 * completing might cause the array to appear non-idle
7488 * and resync will be slowed down even though there might
7489 * not have been non-resync activity. This will only
7490 * happen once though. 'last_events' will soon reflect
7491 * the state where there is little or no outstanding
7492 * resync requests, and further resync activity will
7493 * always make curr_events less than last_events.
7496 if (init
|| curr_events
- rdev
->last_events
> 64) {
7497 rdev
->last_events
= curr_events
;
7505 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7507 /* another "blocks" (512byte) blocks have been synced */
7508 atomic_sub(blocks
, &mddev
->recovery_active
);
7509 wake_up(&mddev
->recovery_wait
);
7511 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7512 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7513 md_wakeup_thread(mddev
->thread
);
7514 // stop recovery, signal do_sync ....
7517 EXPORT_SYMBOL(md_done_sync
);
7519 /* md_write_start(mddev, bi)
7520 * If we need to update some array metadata (e.g. 'active' flag
7521 * in superblock) before writing, schedule a superblock update
7522 * and wait for it to complete.
7524 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7527 if (bio_data_dir(bi
) != WRITE
)
7530 BUG_ON(mddev
->ro
== 1);
7531 if (mddev
->ro
== 2) {
7532 /* need to switch to read/write */
7534 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7535 md_wakeup_thread(mddev
->thread
);
7536 md_wakeup_thread(mddev
->sync_thread
);
7539 atomic_inc(&mddev
->writes_pending
);
7540 if (mddev
->safemode
== 1)
7541 mddev
->safemode
= 0;
7542 if (mddev
->in_sync
) {
7543 spin_lock(&mddev
->lock
);
7544 if (mddev
->in_sync
) {
7546 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7547 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7548 md_wakeup_thread(mddev
->thread
);
7551 spin_unlock(&mddev
->lock
);
7554 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7555 wait_event(mddev
->sb_wait
,
7556 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7558 EXPORT_SYMBOL(md_write_start
);
7560 void md_write_end(struct mddev
*mddev
)
7562 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7563 if (mddev
->safemode
== 2)
7564 md_wakeup_thread(mddev
->thread
);
7565 else if (mddev
->safemode_delay
)
7566 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7569 EXPORT_SYMBOL(md_write_end
);
7571 /* md_allow_write(mddev)
7572 * Calling this ensures that the array is marked 'active' so that writes
7573 * may proceed without blocking. It is important to call this before
7574 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7575 * Must be called with mddev_lock held.
7577 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7578 * is dropped, so return -EAGAIN after notifying userspace.
7580 int md_allow_write(struct mddev
*mddev
)
7586 if (!mddev
->pers
->sync_request
)
7589 spin_lock(&mddev
->lock
);
7590 if (mddev
->in_sync
) {
7592 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7593 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7594 if (mddev
->safemode_delay
&&
7595 mddev
->safemode
== 0)
7596 mddev
->safemode
= 1;
7597 spin_unlock(&mddev
->lock
);
7598 if (mddev_is_clustered(mddev
))
7599 md_cluster_ops
->metadata_update_start(mddev
);
7600 md_update_sb(mddev
, 0);
7601 if (mddev_is_clustered(mddev
))
7602 md_cluster_ops
->metadata_update_finish(mddev
);
7603 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7605 spin_unlock(&mddev
->lock
);
7607 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7612 EXPORT_SYMBOL_GPL(md_allow_write
);
7614 #define SYNC_MARKS 10
7615 #define SYNC_MARK_STEP (3*HZ)
7616 #define UPDATE_FREQUENCY (5*60*HZ)
7617 void md_do_sync(struct md_thread
*thread
)
7619 struct mddev
*mddev
= thread
->mddev
;
7620 struct mddev
*mddev2
;
7621 unsigned int currspeed
= 0,
7623 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
7624 unsigned long mark
[SYNC_MARKS
];
7625 unsigned long update_time
;
7626 sector_t mark_cnt
[SYNC_MARKS
];
7628 struct list_head
*tmp
;
7629 sector_t last_check
;
7631 struct md_rdev
*rdev
;
7632 char *desc
, *action
= NULL
;
7633 struct blk_plug plug
;
7635 /* just incase thread restarts... */
7636 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7638 if (mddev
->ro
) {/* never try to sync a read-only array */
7639 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7643 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7644 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
7645 desc
= "data-check";
7647 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7648 desc
= "requested-resync";
7652 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7657 mddev
->last_sync_action
= action
?: desc
;
7659 /* we overload curr_resync somewhat here.
7660 * 0 == not engaged in resync at all
7661 * 2 == checking that there is no conflict with another sync
7662 * 1 == like 2, but have yielded to allow conflicting resync to
7664 * other == active in resync - this many blocks
7666 * Before starting a resync we must have set curr_resync to
7667 * 2, and then checked that every "conflicting" array has curr_resync
7668 * less than ours. When we find one that is the same or higher
7669 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7670 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7671 * This will mean we have to start checking from the beginning again.
7676 mddev
->curr_resync
= 2;
7679 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7681 for_each_mddev(mddev2
, tmp
) {
7682 if (mddev2
== mddev
)
7684 if (!mddev
->parallel_resync
7685 && mddev2
->curr_resync
7686 && match_mddev_units(mddev
, mddev2
)) {
7688 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7689 /* arbitrarily yield */
7690 mddev
->curr_resync
= 1;
7691 wake_up(&resync_wait
);
7693 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7694 /* no need to wait here, we can wait the next
7695 * time 'round when curr_resync == 2
7698 /* We need to wait 'interruptible' so as not to
7699 * contribute to the load average, and not to
7700 * be caught by 'softlockup'
7702 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7703 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7704 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7705 printk(KERN_INFO
"md: delaying %s of %s"
7706 " until %s has finished (they"
7707 " share one or more physical units)\n",
7708 desc
, mdname(mddev
), mdname(mddev2
));
7710 if (signal_pending(current
))
7711 flush_signals(current
);
7713 finish_wait(&resync_wait
, &wq
);
7716 finish_wait(&resync_wait
, &wq
);
7719 } while (mddev
->curr_resync
< 2);
7722 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7723 /* resync follows the size requested by the personality,
7724 * which defaults to physical size, but can be virtual size
7726 max_sectors
= mddev
->resync_max_sectors
;
7727 atomic64_set(&mddev
->resync_mismatches
, 0);
7728 /* we don't use the checkpoint if there's a bitmap */
7729 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7730 j
= mddev
->resync_min
;
7731 else if (!mddev
->bitmap
)
7732 j
= mddev
->recovery_cp
;
7734 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7735 max_sectors
= mddev
->resync_max_sectors
;
7737 /* recovery follows the physical size of devices */
7738 max_sectors
= mddev
->dev_sectors
;
7741 rdev_for_each_rcu(rdev
, mddev
)
7742 if (rdev
->raid_disk
>= 0 &&
7743 !test_bit(Faulty
, &rdev
->flags
) &&
7744 !test_bit(In_sync
, &rdev
->flags
) &&
7745 rdev
->recovery_offset
< j
)
7746 j
= rdev
->recovery_offset
;
7749 /* If there is a bitmap, we need to make sure all
7750 * writes that started before we added a spare
7751 * complete before we start doing a recovery.
7752 * Otherwise the write might complete and (via
7753 * bitmap_endwrite) set a bit in the bitmap after the
7754 * recovery has checked that bit and skipped that
7757 if (mddev
->bitmap
) {
7758 mddev
->pers
->quiesce(mddev
, 1);
7759 mddev
->pers
->quiesce(mddev
, 0);
7763 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7764 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7765 " %d KB/sec/disk.\n", speed_min(mddev
));
7766 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7767 "(but not more than %d KB/sec) for %s.\n",
7768 speed_max(mddev
), desc
);
7770 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7773 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7775 mark_cnt
[m
] = io_sectors
;
7778 mddev
->resync_mark
= mark
[last_mark
];
7779 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7782 * Tune reconstruction:
7784 window
= 32*(PAGE_SIZE
/512);
7785 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7786 window
/2, (unsigned long long)max_sectors
/2);
7788 atomic_set(&mddev
->recovery_active
, 0);
7793 "md: resuming %s of %s from checkpoint.\n",
7794 desc
, mdname(mddev
));
7795 mddev
->curr_resync
= j
;
7797 mddev
->curr_resync
= 3; /* no longer delayed */
7798 mddev
->curr_resync_completed
= j
;
7799 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7800 md_new_event(mddev
);
7801 update_time
= jiffies
;
7803 if (mddev_is_clustered(mddev
))
7804 md_cluster_ops
->resync_start(mddev
, j
, max_sectors
);
7806 blk_start_plug(&plug
);
7807 while (j
< max_sectors
) {
7812 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7813 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7814 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7815 > (max_sectors
>> 4)) ||
7816 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7817 (j
- mddev
->curr_resync_completed
)*2
7818 >= mddev
->resync_max
- mddev
->curr_resync_completed
7820 /* time to update curr_resync_completed */
7821 wait_event(mddev
->recovery_wait
,
7822 atomic_read(&mddev
->recovery_active
) == 0);
7823 mddev
->curr_resync_completed
= j
;
7824 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
7825 j
> mddev
->recovery_cp
)
7826 mddev
->recovery_cp
= j
;
7827 update_time
= jiffies
;
7828 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7829 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7832 while (j
>= mddev
->resync_max
&&
7833 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7834 /* As this condition is controlled by user-space,
7835 * we can block indefinitely, so use '_interruptible'
7836 * to avoid triggering warnings.
7838 flush_signals(current
); /* just in case */
7839 wait_event_interruptible(mddev
->recovery_wait
,
7840 mddev
->resync_max
> j
7841 || test_bit(MD_RECOVERY_INTR
,
7845 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7848 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
);
7850 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7854 if (!skipped
) { /* actual IO requested */
7855 io_sectors
+= sectors
;
7856 atomic_add(sectors
, &mddev
->recovery_active
);
7859 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7864 mddev
->curr_resync
= j
;
7865 if (mddev_is_clustered(mddev
))
7866 md_cluster_ops
->resync_info_update(mddev
, j
, max_sectors
);
7867 mddev
->curr_mark_cnt
= io_sectors
;
7868 if (last_check
== 0)
7869 /* this is the earliest that rebuild will be
7870 * visible in /proc/mdstat
7872 md_new_event(mddev
);
7874 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7877 last_check
= io_sectors
;
7879 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7881 int next
= (last_mark
+1) % SYNC_MARKS
;
7883 mddev
->resync_mark
= mark
[next
];
7884 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7885 mark
[next
] = jiffies
;
7886 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7890 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7894 * this loop exits only if either when we are slower than
7895 * the 'hard' speed limit, or the system was IO-idle for
7897 * the system might be non-idle CPU-wise, but we only care
7898 * about not overloading the IO subsystem. (things like an
7899 * e2fsck being done on the RAID array should execute fast)
7903 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
7904 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
7905 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7907 if (currspeed
> speed_min(mddev
)) {
7908 if (currspeed
> speed_max(mddev
)) {
7912 if (!is_mddev_idle(mddev
, 0)) {
7914 * Give other IO more of a chance.
7915 * The faster the devices, the less we wait.
7917 wait_event(mddev
->recovery_wait
,
7918 !atomic_read(&mddev
->recovery_active
));
7922 printk(KERN_INFO
"md: %s: %s %s.\n",mdname(mddev
), desc
,
7923 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
7924 ? "interrupted" : "done");
7926 * this also signals 'finished resyncing' to md_stop
7928 blk_finish_plug(&plug
);
7929 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7931 /* tell personality that we are finished */
7932 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
);
7934 if (mddev_is_clustered(mddev
))
7935 md_cluster_ops
->resync_finish(mddev
);
7937 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7938 mddev
->curr_resync
> 2) {
7939 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7940 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7941 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7943 "md: checkpointing %s of %s.\n",
7944 desc
, mdname(mddev
));
7945 if (test_bit(MD_RECOVERY_ERROR
,
7947 mddev
->recovery_cp
=
7948 mddev
->curr_resync_completed
;
7950 mddev
->recovery_cp
=
7954 mddev
->recovery_cp
= MaxSector
;
7956 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7957 mddev
->curr_resync
= MaxSector
;
7959 rdev_for_each_rcu(rdev
, mddev
)
7960 if (rdev
->raid_disk
>= 0 &&
7961 mddev
->delta_disks
>= 0 &&
7962 !test_bit(Faulty
, &rdev
->flags
) &&
7963 !test_bit(In_sync
, &rdev
->flags
) &&
7964 rdev
->recovery_offset
< mddev
->curr_resync
)
7965 rdev
->recovery_offset
= mddev
->curr_resync
;
7970 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7972 spin_lock(&mddev
->lock
);
7973 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7974 /* We completed so min/max setting can be forgotten if used. */
7975 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7976 mddev
->resync_min
= 0;
7977 mddev
->resync_max
= MaxSector
;
7978 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7979 mddev
->resync_min
= mddev
->curr_resync_completed
;
7980 mddev
->curr_resync
= 0;
7981 spin_unlock(&mddev
->lock
);
7983 wake_up(&resync_wait
);
7984 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7985 md_wakeup_thread(mddev
->thread
);
7988 EXPORT_SYMBOL_GPL(md_do_sync
);
7990 static int remove_and_add_spares(struct mddev
*mddev
,
7991 struct md_rdev
*this)
7993 struct md_rdev
*rdev
;
7997 rdev_for_each(rdev
, mddev
)
7998 if ((this == NULL
|| rdev
== this) &&
7999 rdev
->raid_disk
>= 0 &&
8000 !test_bit(Blocked
, &rdev
->flags
) &&
8001 (test_bit(Faulty
, &rdev
->flags
) ||
8002 ! test_bit(In_sync
, &rdev
->flags
)) &&
8003 atomic_read(&rdev
->nr_pending
)==0) {
8004 if (mddev
->pers
->hot_remove_disk(
8005 mddev
, rdev
) == 0) {
8006 sysfs_unlink_rdev(mddev
, rdev
);
8007 rdev
->raid_disk
= -1;
8011 if (removed
&& mddev
->kobj
.sd
)
8012 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
8017 rdev_for_each(rdev
, mddev
) {
8018 if (rdev
->raid_disk
>= 0 &&
8019 !test_bit(In_sync
, &rdev
->flags
) &&
8020 !test_bit(Faulty
, &rdev
->flags
))
8022 if (rdev
->raid_disk
>= 0)
8024 if (test_bit(Faulty
, &rdev
->flags
))
8027 ! (rdev
->saved_raid_disk
>= 0 &&
8028 !test_bit(Bitmap_sync
, &rdev
->flags
)))
8031 if (rdev
->saved_raid_disk
< 0)
8032 rdev
->recovery_offset
= 0;
8034 hot_add_disk(mddev
, rdev
) == 0) {
8035 if (sysfs_link_rdev(mddev
, rdev
))
8036 /* failure here is OK */;
8038 md_new_event(mddev
);
8039 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8044 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8048 static void md_start_sync(struct work_struct
*ws
)
8050 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
8052 mddev
->sync_thread
= md_register_thread(md_do_sync
,
8055 if (!mddev
->sync_thread
) {
8056 printk(KERN_ERR
"%s: could not start resync"
8059 /* leave the spares where they are, it shouldn't hurt */
8060 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8061 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8062 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8063 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8064 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8065 wake_up(&resync_wait
);
8066 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8068 if (mddev
->sysfs_action
)
8069 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8071 md_wakeup_thread(mddev
->sync_thread
);
8072 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8073 md_new_event(mddev
);
8077 * This routine is regularly called by all per-raid-array threads to
8078 * deal with generic issues like resync and super-block update.
8079 * Raid personalities that don't have a thread (linear/raid0) do not
8080 * need this as they never do any recovery or update the superblock.
8082 * It does not do any resync itself, but rather "forks" off other threads
8083 * to do that as needed.
8084 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8085 * "->recovery" and create a thread at ->sync_thread.
8086 * When the thread finishes it sets MD_RECOVERY_DONE
8087 * and wakeups up this thread which will reap the thread and finish up.
8088 * This thread also removes any faulty devices (with nr_pending == 0).
8090 * The overall approach is:
8091 * 1/ if the superblock needs updating, update it.
8092 * 2/ If a recovery thread is running, don't do anything else.
8093 * 3/ If recovery has finished, clean up, possibly marking spares active.
8094 * 4/ If there are any faulty devices, remove them.
8095 * 5/ If array is degraded, try to add spares devices
8096 * 6/ If array has spares or is not in-sync, start a resync thread.
8098 void md_check_recovery(struct mddev
*mddev
)
8100 if (mddev
->suspended
)
8104 bitmap_daemon_work(mddev
);
8106 if (signal_pending(current
)) {
8107 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
8108 printk(KERN_INFO
"md: %s in immediate safe mode\n",
8110 mddev
->safemode
= 2;
8112 flush_signals(current
);
8115 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
8118 (mddev
->flags
& MD_UPDATE_SB_FLAGS
& ~ (1<<MD_CHANGE_PENDING
)) ||
8119 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8120 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8121 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
8122 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
8123 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
8127 if (mddev_trylock(mddev
)) {
8131 struct md_rdev
*rdev
;
8132 if (!mddev
->external
&& mddev
->in_sync
)
8133 /* 'Blocked' flag not needed as failed devices
8134 * will be recorded if array switched to read/write.
8135 * Leaving it set will prevent the device
8136 * from being removed.
8138 rdev_for_each(rdev
, mddev
)
8139 clear_bit(Blocked
, &rdev
->flags
);
8140 /* On a read-only array we can:
8141 * - remove failed devices
8142 * - add already-in_sync devices if the array itself
8144 * As we only add devices that are already in-sync,
8145 * we can activate the spares immediately.
8147 remove_and_add_spares(mddev
, NULL
);
8148 /* There is no thread, but we need to call
8149 * ->spare_active and clear saved_raid_disk
8151 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8152 md_reap_sync_thread(mddev
);
8153 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8157 if (!mddev
->external
) {
8159 spin_lock(&mddev
->lock
);
8160 if (mddev
->safemode
&&
8161 !atomic_read(&mddev
->writes_pending
) &&
8163 mddev
->recovery_cp
== MaxSector
) {
8166 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
8168 if (mddev
->safemode
== 1)
8169 mddev
->safemode
= 0;
8170 spin_unlock(&mddev
->lock
);
8172 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8175 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
) {
8176 if (mddev_is_clustered(mddev
))
8177 md_cluster_ops
->metadata_update_start(mddev
);
8178 md_update_sb(mddev
, 0);
8179 if (mddev_is_clustered(mddev
))
8180 md_cluster_ops
->metadata_update_finish(mddev
);
8183 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
8184 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
8185 /* resync/recovery still happening */
8186 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8189 if (mddev
->sync_thread
) {
8190 md_reap_sync_thread(mddev
);
8193 /* Set RUNNING before clearing NEEDED to avoid
8194 * any transients in the value of "sync_action".
8196 mddev
->curr_resync_completed
= 0;
8197 spin_lock(&mddev
->lock
);
8198 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8199 spin_unlock(&mddev
->lock
);
8200 /* Clear some bits that don't mean anything, but
8203 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8204 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8206 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8207 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
8209 /* no recovery is running.
8210 * remove any failed drives, then
8211 * add spares if possible.
8212 * Spares are also removed and re-added, to allow
8213 * the personality to fail the re-add.
8216 if (mddev
->reshape_position
!= MaxSector
) {
8217 if (mddev
->pers
->check_reshape
== NULL
||
8218 mddev
->pers
->check_reshape(mddev
) != 0)
8219 /* Cannot proceed */
8221 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8222 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8223 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
8224 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8225 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8226 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8227 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8228 } else if (mddev
->recovery_cp
< MaxSector
) {
8229 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8230 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8231 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
8232 /* nothing to be done ... */
8235 if (mddev
->pers
->sync_request
) {
8237 /* We are adding a device or devices to an array
8238 * which has the bitmap stored on all devices.
8239 * So make sure all bitmap pages get written
8241 bitmap_write_all(mddev
->bitmap
);
8243 INIT_WORK(&mddev
->del_work
, md_start_sync
);
8244 queue_work(md_misc_wq
, &mddev
->del_work
);
8248 if (!mddev
->sync_thread
) {
8249 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8250 wake_up(&resync_wait
);
8251 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8253 if (mddev
->sysfs_action
)
8254 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8257 wake_up(&mddev
->sb_wait
);
8258 mddev_unlock(mddev
);
8261 EXPORT_SYMBOL(md_check_recovery
);
8263 void md_reap_sync_thread(struct mddev
*mddev
)
8265 struct md_rdev
*rdev
;
8267 /* resync has finished, collect result */
8268 md_unregister_thread(&mddev
->sync_thread
);
8269 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8270 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8272 /* activate any spares */
8273 if (mddev
->pers
->spare_active(mddev
)) {
8274 sysfs_notify(&mddev
->kobj
, NULL
,
8276 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8279 if (mddev_is_clustered(mddev
))
8280 md_cluster_ops
->metadata_update_start(mddev
);
8281 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8282 mddev
->pers
->finish_reshape
)
8283 mddev
->pers
->finish_reshape(mddev
);
8285 /* If array is no-longer degraded, then any saved_raid_disk
8286 * information must be scrapped.
8288 if (!mddev
->degraded
)
8289 rdev_for_each(rdev
, mddev
)
8290 rdev
->saved_raid_disk
= -1;
8292 md_update_sb(mddev
, 1);
8293 if (mddev_is_clustered(mddev
))
8294 md_cluster_ops
->metadata_update_finish(mddev
);
8295 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8296 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8297 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8298 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8299 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8300 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8301 wake_up(&resync_wait
);
8302 /* flag recovery needed just to double check */
8303 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8304 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8305 md_new_event(mddev
);
8306 if (mddev
->event_work
.func
)
8307 queue_work(md_misc_wq
, &mddev
->event_work
);
8309 EXPORT_SYMBOL(md_reap_sync_thread
);
8311 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
8313 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8314 wait_event_timeout(rdev
->blocked_wait
,
8315 !test_bit(Blocked
, &rdev
->flags
) &&
8316 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
8317 msecs_to_jiffies(5000));
8318 rdev_dec_pending(rdev
, mddev
);
8320 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
8322 void md_finish_reshape(struct mddev
*mddev
)
8324 /* called be personality module when reshape completes. */
8325 struct md_rdev
*rdev
;
8327 rdev_for_each(rdev
, mddev
) {
8328 if (rdev
->data_offset
> rdev
->new_data_offset
)
8329 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
8331 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
8332 rdev
->data_offset
= rdev
->new_data_offset
;
8335 EXPORT_SYMBOL(md_finish_reshape
);
8337 /* Bad block management.
8338 * We can record which blocks on each device are 'bad' and so just
8339 * fail those blocks, or that stripe, rather than the whole device.
8340 * Entries in the bad-block table are 64bits wide. This comprises:
8341 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8342 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8343 * A 'shift' can be set so that larger blocks are tracked and
8344 * consequently larger devices can be covered.
8345 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8347 * Locking of the bad-block table uses a seqlock so md_is_badblock
8348 * might need to retry if it is very unlucky.
8349 * We will sometimes want to check for bad blocks in a bi_end_io function,
8350 * so we use the write_seqlock_irq variant.
8352 * When looking for a bad block we specify a range and want to
8353 * know if any block in the range is bad. So we binary-search
8354 * to the last range that starts at-or-before the given endpoint,
8355 * (or "before the sector after the target range")
8356 * then see if it ends after the given start.
8358 * 0 if there are no known bad blocks in the range
8359 * 1 if there are known bad block which are all acknowledged
8360 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8361 * plus the start/length of the first bad section we overlap.
8363 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
8364 sector_t
*first_bad
, int *bad_sectors
)
8370 sector_t target
= s
+ sectors
;
8373 if (bb
->shift
> 0) {
8374 /* round the start down, and the end up */
8376 target
+= (1<<bb
->shift
) - 1;
8377 target
>>= bb
->shift
;
8378 sectors
= target
- s
;
8380 /* 'target' is now the first block after the bad range */
8383 seq
= read_seqbegin(&bb
->lock
);
8388 /* Binary search between lo and hi for 'target'
8389 * i.e. for the last range that starts before 'target'
8391 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8392 * are known not to be the last range before target.
8393 * VARIANT: hi-lo is the number of possible
8394 * ranges, and decreases until it reaches 1
8396 while (hi
- lo
> 1) {
8397 int mid
= (lo
+ hi
) / 2;
8398 sector_t a
= BB_OFFSET(p
[mid
]);
8400 /* This could still be the one, earlier ranges
8404 /* This and later ranges are definitely out. */
8407 /* 'lo' might be the last that started before target, but 'hi' isn't */
8409 /* need to check all range that end after 's' to see if
8410 * any are unacknowledged.
8413 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8414 if (BB_OFFSET(p
[lo
]) < target
) {
8415 /* starts before the end, and finishes after
8416 * the start, so they must overlap
8418 if (rv
!= -1 && BB_ACK(p
[lo
]))
8422 *first_bad
= BB_OFFSET(p
[lo
]);
8423 *bad_sectors
= BB_LEN(p
[lo
]);
8429 if (read_seqretry(&bb
->lock
, seq
))
8434 EXPORT_SYMBOL_GPL(md_is_badblock
);
8437 * Add a range of bad blocks to the table.
8438 * This might extend the table, or might contract it
8439 * if two adjacent ranges can be merged.
8440 * We binary-search to find the 'insertion' point, then
8441 * decide how best to handle it.
8443 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
8449 unsigned long flags
;
8452 /* badblocks are disabled */
8456 /* round the start down, and the end up */
8457 sector_t next
= s
+ sectors
;
8459 next
+= (1<<bb
->shift
) - 1;
8464 write_seqlock_irqsave(&bb
->lock
, flags
);
8469 /* Find the last range that starts at-or-before 's' */
8470 while (hi
- lo
> 1) {
8471 int mid
= (lo
+ hi
) / 2;
8472 sector_t a
= BB_OFFSET(p
[mid
]);
8478 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8482 /* we found a range that might merge with the start
8485 sector_t a
= BB_OFFSET(p
[lo
]);
8486 sector_t e
= a
+ BB_LEN(p
[lo
]);
8487 int ack
= BB_ACK(p
[lo
]);
8489 /* Yes, we can merge with a previous range */
8490 if (s
== a
&& s
+ sectors
>= e
)
8491 /* new range covers old */
8494 ack
= ack
&& acknowledged
;
8496 if (e
< s
+ sectors
)
8498 if (e
- a
<= BB_MAX_LEN
) {
8499 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8502 /* does not all fit in one range,
8503 * make p[lo] maximal
8505 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8506 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8512 if (sectors
&& hi
< bb
->count
) {
8513 /* 'hi' points to the first range that starts after 's'.
8514 * Maybe we can merge with the start of that range */
8515 sector_t a
= BB_OFFSET(p
[hi
]);
8516 sector_t e
= a
+ BB_LEN(p
[hi
]);
8517 int ack
= BB_ACK(p
[hi
]);
8518 if (a
<= s
+ sectors
) {
8519 /* merging is possible */
8520 if (e
<= s
+ sectors
) {
8525 ack
= ack
&& acknowledged
;
8528 if (e
- a
<= BB_MAX_LEN
) {
8529 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8532 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8540 if (sectors
== 0 && hi
< bb
->count
) {
8541 /* we might be able to combine lo and hi */
8542 /* Note: 's' is at the end of 'lo' */
8543 sector_t a
= BB_OFFSET(p
[hi
]);
8544 int lolen
= BB_LEN(p
[lo
]);
8545 int hilen
= BB_LEN(p
[hi
]);
8546 int newlen
= lolen
+ hilen
- (s
- a
);
8547 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8548 /* yes, we can combine them */
8549 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8550 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8551 memmove(p
+ hi
, p
+ hi
+ 1,
8552 (bb
->count
- hi
- 1) * 8);
8557 /* didn't merge (it all).
8558 * Need to add a range just before 'hi' */
8559 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8560 /* No room for more */
8564 int this_sectors
= sectors
;
8565 memmove(p
+ hi
+ 1, p
+ hi
,
8566 (bb
->count
- hi
) * 8);
8569 if (this_sectors
> BB_MAX_LEN
)
8570 this_sectors
= BB_MAX_LEN
;
8571 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8572 sectors
-= this_sectors
;
8579 bb
->unacked_exist
= 1;
8580 write_sequnlock_irqrestore(&bb
->lock
, flags
);
8585 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8590 s
+= rdev
->new_data_offset
;
8592 s
+= rdev
->data_offset
;
8593 rv
= md_set_badblocks(&rdev
->badblocks
,
8596 /* Make sure they get written out promptly */
8597 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8598 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8599 md_wakeup_thread(rdev
->mddev
->thread
);
8603 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8606 * Remove a range of bad blocks from the table.
8607 * This may involve extending the table if we spilt a region,
8608 * but it must not fail. So if the table becomes full, we just
8609 * drop the remove request.
8611 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8615 sector_t target
= s
+ sectors
;
8618 if (bb
->shift
> 0) {
8619 /* When clearing we round the start up and the end down.
8620 * This should not matter as the shift should align with
8621 * the block size and no rounding should ever be needed.
8622 * However it is better the think a block is bad when it
8623 * isn't than to think a block is not bad when it is.
8625 s
+= (1<<bb
->shift
) - 1;
8627 target
>>= bb
->shift
;
8628 sectors
= target
- s
;
8631 write_seqlock_irq(&bb
->lock
);
8636 /* Find the last range that starts before 'target' */
8637 while (hi
- lo
> 1) {
8638 int mid
= (lo
+ hi
) / 2;
8639 sector_t a
= BB_OFFSET(p
[mid
]);
8646 /* p[lo] is the last range that could overlap the
8647 * current range. Earlier ranges could also overlap,
8648 * but only this one can overlap the end of the range.
8650 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8651 /* Partial overlap, leave the tail of this range */
8652 int ack
= BB_ACK(p
[lo
]);
8653 sector_t a
= BB_OFFSET(p
[lo
]);
8654 sector_t end
= a
+ BB_LEN(p
[lo
]);
8657 /* we need to split this range */
8658 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8662 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8664 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8667 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8668 /* there is no longer an overlap */
8673 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8674 /* This range does overlap */
8675 if (BB_OFFSET(p
[lo
]) < s
) {
8676 /* Keep the early parts of this range. */
8677 int ack
= BB_ACK(p
[lo
]);
8678 sector_t start
= BB_OFFSET(p
[lo
]);
8679 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8680 /* now low doesn't overlap, so.. */
8685 /* 'lo' is strictly before, 'hi' is strictly after,
8686 * anything between needs to be discarded
8689 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8690 bb
->count
-= (hi
- lo
- 1);
8696 write_sequnlock_irq(&bb
->lock
);
8700 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8704 s
+= rdev
->new_data_offset
;
8706 s
+= rdev
->data_offset
;
8707 return md_clear_badblocks(&rdev
->badblocks
,
8710 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8713 * Acknowledge all bad blocks in a list.
8714 * This only succeeds if ->changed is clear. It is used by
8715 * in-kernel metadata updates
8717 void md_ack_all_badblocks(struct badblocks
*bb
)
8719 if (bb
->page
== NULL
|| bb
->changed
)
8720 /* no point even trying */
8722 write_seqlock_irq(&bb
->lock
);
8724 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8727 for (i
= 0; i
< bb
->count
; i
++) {
8728 if (!BB_ACK(p
[i
])) {
8729 sector_t start
= BB_OFFSET(p
[i
]);
8730 int len
= BB_LEN(p
[i
]);
8731 p
[i
] = BB_MAKE(start
, len
, 1);
8734 bb
->unacked_exist
= 0;
8736 write_sequnlock_irq(&bb
->lock
);
8738 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8740 /* sysfs access to bad-blocks list.
8741 * We present two files.
8742 * 'bad-blocks' lists sector numbers and lengths of ranges that
8743 * are recorded as bad. The list is truncated to fit within
8744 * the one-page limit of sysfs.
8745 * Writing "sector length" to this file adds an acknowledged
8747 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8748 * been acknowledged. Writing to this file adds bad blocks
8749 * without acknowledging them. This is largely for testing.
8753 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8764 seq
= read_seqbegin(&bb
->lock
);
8769 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8770 sector_t s
= BB_OFFSET(p
[i
]);
8771 unsigned int length
= BB_LEN(p
[i
]);
8772 int ack
= BB_ACK(p
[i
]);
8778 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8779 (unsigned long long)s
<< bb
->shift
,
8780 length
<< bb
->shift
);
8782 if (unack
&& len
== 0)
8783 bb
->unacked_exist
= 0;
8785 if (read_seqretry(&bb
->lock
, seq
))
8794 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8796 unsigned long long sector
;
8800 /* Allow clearing via sysfs *only* for testing/debugging.
8801 * Normally only a successful write may clear a badblock
8804 if (page
[0] == '-') {
8808 #endif /* DO_DEBUG */
8810 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8812 if (newline
!= '\n')
8824 md_clear_badblocks(bb
, sector
, length
);
8827 #endif /* DO_DEBUG */
8828 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8834 static int md_notify_reboot(struct notifier_block
*this,
8835 unsigned long code
, void *x
)
8837 struct list_head
*tmp
;
8838 struct mddev
*mddev
;
8841 for_each_mddev(mddev
, tmp
) {
8842 if (mddev_trylock(mddev
)) {
8844 __md_stop_writes(mddev
);
8845 if (mddev
->persistent
)
8846 mddev
->safemode
= 2;
8847 mddev_unlock(mddev
);
8852 * certain more exotic SCSI devices are known to be
8853 * volatile wrt too early system reboots. While the
8854 * right place to handle this issue is the given
8855 * driver, we do want to have a safe RAID driver ...
8863 static struct notifier_block md_notifier
= {
8864 .notifier_call
= md_notify_reboot
,
8866 .priority
= INT_MAX
, /* before any real devices */
8869 static void md_geninit(void)
8871 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8873 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8876 static int __init
md_init(void)
8880 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8884 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8888 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8891 if ((ret
= register_blkdev(0, "mdp")) < 0)
8895 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
8896 md_probe
, NULL
, NULL
);
8897 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8898 md_probe
, NULL
, NULL
);
8900 register_reboot_notifier(&md_notifier
);
8901 raid_table_header
= register_sysctl_table(raid_root_table
);
8907 unregister_blkdev(MD_MAJOR
, "md");
8909 destroy_workqueue(md_misc_wq
);
8911 destroy_workqueue(md_wq
);
8916 void md_reload_sb(struct mddev
*mddev
)
8918 struct md_rdev
*rdev
, *tmp
;
8920 rdev_for_each_safe(rdev
, tmp
, mddev
) {
8921 rdev
->sb_loaded
= 0;
8922 ClearPageUptodate(rdev
->sb_page
);
8924 mddev
->raid_disks
= 0;
8926 rdev_for_each_safe(rdev
, tmp
, mddev
) {
8927 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
8928 /* since we don't write to faulty devices, we figure out if the
8929 * disk is faulty by comparing events
8931 if (mddev
->events
> sb
->events
)
8932 set_bit(Faulty
, &rdev
->flags
);
8936 EXPORT_SYMBOL(md_reload_sb
);
8941 * Searches all registered partitions for autorun RAID arrays
8945 static LIST_HEAD(all_detected_devices
);
8946 struct detected_devices_node
{
8947 struct list_head list
;
8951 void md_autodetect_dev(dev_t dev
)
8953 struct detected_devices_node
*node_detected_dev
;
8955 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8956 if (node_detected_dev
) {
8957 node_detected_dev
->dev
= dev
;
8958 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8960 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8961 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8965 static void autostart_arrays(int part
)
8967 struct md_rdev
*rdev
;
8968 struct detected_devices_node
*node_detected_dev
;
8970 int i_scanned
, i_passed
;
8975 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8977 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8979 node_detected_dev
= list_entry(all_detected_devices
.next
,
8980 struct detected_devices_node
, list
);
8981 list_del(&node_detected_dev
->list
);
8982 dev
= node_detected_dev
->dev
;
8983 kfree(node_detected_dev
);
8984 rdev
= md_import_device(dev
,0, 90);
8988 if (test_bit(Faulty
, &rdev
->flags
))
8991 set_bit(AutoDetected
, &rdev
->flags
);
8992 list_add(&rdev
->same_set
, &pending_raid_disks
);
8996 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8997 i_scanned
, i_passed
);
8999 autorun_devices(part
);
9002 #endif /* !MODULE */
9004 static __exit
void md_exit(void)
9006 struct mddev
*mddev
;
9007 struct list_head
*tmp
;
9010 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
9011 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
9013 unregister_blkdev(MD_MAJOR
,"md");
9014 unregister_blkdev(mdp_major
, "mdp");
9015 unregister_reboot_notifier(&md_notifier
);
9016 unregister_sysctl_table(raid_table_header
);
9018 /* We cannot unload the modules while some process is
9019 * waiting for us in select() or poll() - wake them up
9022 while (waitqueue_active(&md_event_waiters
)) {
9023 /* not safe to leave yet */
9024 wake_up(&md_event_waiters
);
9028 remove_proc_entry("mdstat", NULL
);
9030 for_each_mddev(mddev
, tmp
) {
9031 export_array(mddev
);
9032 mddev
->hold_active
= 0;
9034 destroy_workqueue(md_misc_wq
);
9035 destroy_workqueue(md_wq
);
9038 subsys_initcall(md_init
);
9039 module_exit(md_exit
)
9041 static int get_ro(char *buffer
, struct kernel_param
*kp
)
9043 return sprintf(buffer
, "%d", start_readonly
);
9045 static int set_ro(const char *val
, struct kernel_param
*kp
)
9047 return kstrtouint(val
, 10, (unsigned int *)&start_readonly
);
9050 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
9051 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
9052 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
9054 MODULE_LICENSE("GPL");
9055 MODULE_DESCRIPTION("MD RAID framework");
9057 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);