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
];
2030 /* prevent duplicates */
2031 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2034 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2035 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2036 rdev
->sectors
< mddev
->dev_sectors
)) {
2038 /* Cannot change size, so fail
2039 * If mddev->level <= 0, then we don't care
2040 * about aligning sizes (e.g. linear)
2042 if (mddev
->level
> 0)
2045 mddev
->dev_sectors
= rdev
->sectors
;
2048 /* Verify rdev->desc_nr is unique.
2049 * If it is -1, assign a free number, else
2050 * check number is not in use
2053 if (rdev
->desc_nr
< 0) {
2056 choice
= mddev
->raid_disks
;
2057 while (md_find_rdev_nr_rcu(mddev
, choice
))
2059 rdev
->desc_nr
= choice
;
2061 if (md_find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2067 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2068 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2069 mdname(mddev
), mddev
->max_disks
);
2072 bdevname(rdev
->bdev
,b
);
2073 while ( (s
=strchr(b
, '/')) != NULL
)
2076 rdev
->mddev
= mddev
;
2077 printk(KERN_INFO
"md: bind<%s>\n", b
);
2079 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2082 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2083 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2084 /* failure here is OK */;
2085 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2087 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2088 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2090 /* May as well allow recovery to be retried once */
2091 mddev
->recovery_disabled
++;
2096 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2101 static void md_delayed_delete(struct work_struct
*ws
)
2103 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2104 kobject_del(&rdev
->kobj
);
2105 kobject_put(&rdev
->kobj
);
2108 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2110 char b
[BDEVNAME_SIZE
];
2112 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2113 list_del_rcu(&rdev
->same_set
);
2114 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2116 sysfs_remove_link(&rdev
->kobj
, "block");
2117 sysfs_put(rdev
->sysfs_state
);
2118 rdev
->sysfs_state
= NULL
;
2119 rdev
->badblocks
.count
= 0;
2120 /* We need to delay this, otherwise we can deadlock when
2121 * writing to 'remove' to "dev/state". We also need
2122 * to delay it due to rcu usage.
2125 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2126 kobject_get(&rdev
->kobj
);
2127 queue_work(md_misc_wq
, &rdev
->del_work
);
2131 * prevent the device from being mounted, repartitioned or
2132 * otherwise reused by a RAID array (or any other kernel
2133 * subsystem), by bd_claiming the device.
2135 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2138 struct block_device
*bdev
;
2139 char b
[BDEVNAME_SIZE
];
2141 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2142 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2144 printk(KERN_ERR
"md: could not open %s.\n",
2145 __bdevname(dev
, b
));
2146 return PTR_ERR(bdev
);
2152 static void unlock_rdev(struct md_rdev
*rdev
)
2154 struct block_device
*bdev
= rdev
->bdev
;
2156 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2159 void md_autodetect_dev(dev_t dev
);
2161 static void export_rdev(struct md_rdev
*rdev
)
2163 char b
[BDEVNAME_SIZE
];
2165 printk(KERN_INFO
"md: export_rdev(%s)\n",
2166 bdevname(rdev
->bdev
,b
));
2167 md_rdev_clear(rdev
);
2169 if (test_bit(AutoDetected
, &rdev
->flags
))
2170 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2173 kobject_put(&rdev
->kobj
);
2176 void md_kick_rdev_from_array(struct md_rdev
*rdev
)
2178 unbind_rdev_from_array(rdev
);
2181 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array
);
2183 static void export_array(struct mddev
*mddev
)
2185 struct md_rdev
*rdev
;
2187 while (!list_empty(&mddev
->disks
)) {
2188 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2190 md_kick_rdev_from_array(rdev
);
2192 mddev
->raid_disks
= 0;
2193 mddev
->major_version
= 0;
2196 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2198 /* Update each superblock (in-memory image), but
2199 * if we are allowed to, skip spares which already
2200 * have the right event counter, or have one earlier
2201 * (which would mean they aren't being marked as dirty
2202 * with the rest of the array)
2204 struct md_rdev
*rdev
;
2205 rdev_for_each(rdev
, mddev
) {
2206 if (rdev
->sb_events
== mddev
->events
||
2208 rdev
->raid_disk
< 0 &&
2209 rdev
->sb_events
+1 == mddev
->events
)) {
2210 /* Don't update this superblock */
2211 rdev
->sb_loaded
= 2;
2213 sync_super(mddev
, rdev
);
2214 rdev
->sb_loaded
= 1;
2219 void md_update_sb(struct mddev
*mddev
, int force_change
)
2221 struct md_rdev
*rdev
;
2224 int any_badblocks_changed
= 0;
2228 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2232 /* First make sure individual recovery_offsets are correct */
2233 rdev_for_each(rdev
, mddev
) {
2234 if (rdev
->raid_disk
>= 0 &&
2235 mddev
->delta_disks
>= 0 &&
2236 !test_bit(In_sync
, &rdev
->flags
) &&
2237 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2238 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2241 if (!mddev
->persistent
) {
2242 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2243 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2244 if (!mddev
->external
) {
2245 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2246 rdev_for_each(rdev
, mddev
) {
2247 if (rdev
->badblocks
.changed
) {
2248 rdev
->badblocks
.changed
= 0;
2249 md_ack_all_badblocks(&rdev
->badblocks
);
2250 md_error(mddev
, rdev
);
2252 clear_bit(Blocked
, &rdev
->flags
);
2253 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2254 wake_up(&rdev
->blocked_wait
);
2257 wake_up(&mddev
->sb_wait
);
2261 spin_lock(&mddev
->lock
);
2263 mddev
->utime
= get_seconds();
2265 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2267 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2268 /* just a clean<-> dirty transition, possibly leave spares alone,
2269 * though if events isn't the right even/odd, we will have to do
2275 if (mddev
->degraded
)
2276 /* If the array is degraded, then skipping spares is both
2277 * dangerous and fairly pointless.
2278 * Dangerous because a device that was removed from the array
2279 * might have a event_count that still looks up-to-date,
2280 * so it can be re-added without a resync.
2281 * Pointless because if there are any spares to skip,
2282 * then a recovery will happen and soon that array won't
2283 * be degraded any more and the spare can go back to sleep then.
2287 sync_req
= mddev
->in_sync
;
2289 /* If this is just a dirty<->clean transition, and the array is clean
2290 * and 'events' is odd, we can roll back to the previous clean state */
2292 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2293 && mddev
->can_decrease_events
2294 && mddev
->events
!= 1) {
2296 mddev
->can_decrease_events
= 0;
2298 /* otherwise we have to go forward and ... */
2300 mddev
->can_decrease_events
= nospares
;
2304 * This 64-bit counter should never wrap.
2305 * Either we are in around ~1 trillion A.C., assuming
2306 * 1 reboot per second, or we have a bug...
2308 WARN_ON(mddev
->events
== 0);
2310 rdev_for_each(rdev
, mddev
) {
2311 if (rdev
->badblocks
.changed
)
2312 any_badblocks_changed
++;
2313 if (test_bit(Faulty
, &rdev
->flags
))
2314 set_bit(FaultRecorded
, &rdev
->flags
);
2317 sync_sbs(mddev
, nospares
);
2318 spin_unlock(&mddev
->lock
);
2320 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2321 mdname(mddev
), mddev
->in_sync
);
2323 bitmap_update_sb(mddev
->bitmap
);
2324 rdev_for_each(rdev
, mddev
) {
2325 char b
[BDEVNAME_SIZE
];
2327 if (rdev
->sb_loaded
!= 1)
2328 continue; /* no noise on spare devices */
2330 if (!test_bit(Faulty
, &rdev
->flags
)) {
2331 md_super_write(mddev
,rdev
,
2332 rdev
->sb_start
, rdev
->sb_size
,
2334 pr_debug("md: (write) %s's sb offset: %llu\n",
2335 bdevname(rdev
->bdev
, b
),
2336 (unsigned long long)rdev
->sb_start
);
2337 rdev
->sb_events
= mddev
->events
;
2338 if (rdev
->badblocks
.size
) {
2339 md_super_write(mddev
, rdev
,
2340 rdev
->badblocks
.sector
,
2341 rdev
->badblocks
.size
<< 9,
2343 rdev
->badblocks
.size
= 0;
2347 pr_debug("md: %s (skipping faulty)\n",
2348 bdevname(rdev
->bdev
, b
));
2350 if (mddev
->level
== LEVEL_MULTIPATH
)
2351 /* only need to write one superblock... */
2354 md_super_wait(mddev
);
2355 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2357 spin_lock(&mddev
->lock
);
2358 if (mddev
->in_sync
!= sync_req
||
2359 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2360 /* have to write it out again */
2361 spin_unlock(&mddev
->lock
);
2364 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2365 spin_unlock(&mddev
->lock
);
2366 wake_up(&mddev
->sb_wait
);
2367 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2368 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2370 rdev_for_each(rdev
, mddev
) {
2371 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2372 clear_bit(Blocked
, &rdev
->flags
);
2374 if (any_badblocks_changed
)
2375 md_ack_all_badblocks(&rdev
->badblocks
);
2376 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2377 wake_up(&rdev
->blocked_wait
);
2380 EXPORT_SYMBOL(md_update_sb
);
2382 static int add_bound_rdev(struct md_rdev
*rdev
)
2384 struct mddev
*mddev
= rdev
->mddev
;
2387 if (!mddev
->pers
->hot_remove_disk
) {
2388 /* If there is hot_add_disk but no hot_remove_disk
2389 * then added disks for geometry changes,
2390 * and should be added immediately.
2392 super_types
[mddev
->major_version
].
2393 validate_super(mddev
, rdev
);
2394 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
2396 unbind_rdev_from_array(rdev
);
2401 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2403 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2404 if (mddev
->degraded
)
2405 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2406 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2407 md_new_event(mddev
);
2408 md_wakeup_thread(mddev
->thread
);
2412 /* words written to sysfs files may, or may not, be \n terminated.
2413 * We want to accept with case. For this we use cmd_match.
2415 static int cmd_match(const char *cmd
, const char *str
)
2417 /* See if cmd, written into a sysfs file, matches
2418 * str. They must either be the same, or cmd can
2419 * have a trailing newline
2421 while (*cmd
&& *str
&& *cmd
== *str
) {
2432 struct rdev_sysfs_entry
{
2433 struct attribute attr
;
2434 ssize_t (*show
)(struct md_rdev
*, char *);
2435 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2439 state_show(struct md_rdev
*rdev
, char *page
)
2443 unsigned long flags
= ACCESS_ONCE(rdev
->flags
);
2445 if (test_bit(Faulty
, &flags
) ||
2446 rdev
->badblocks
.unacked_exist
) {
2447 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2450 if (test_bit(In_sync
, &flags
)) {
2451 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2454 if (test_bit(WriteMostly
, &flags
)) {
2455 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2458 if (test_bit(Blocked
, &flags
) ||
2459 (rdev
->badblocks
.unacked_exist
2460 && !test_bit(Faulty
, &flags
))) {
2461 len
+= sprintf(page
+len
, "%sblocked", sep
);
2464 if (!test_bit(Faulty
, &flags
) &&
2465 !test_bit(In_sync
, &flags
)) {
2466 len
+= sprintf(page
+len
, "%sspare", sep
);
2469 if (test_bit(WriteErrorSeen
, &flags
)) {
2470 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2473 if (test_bit(WantReplacement
, &flags
)) {
2474 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2477 if (test_bit(Replacement
, &flags
)) {
2478 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2482 return len
+sprintf(page
+len
, "\n");
2486 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2489 * faulty - simulates an error
2490 * remove - disconnects the device
2491 * writemostly - sets write_mostly
2492 * -writemostly - clears write_mostly
2493 * blocked - sets the Blocked flags
2494 * -blocked - clears the Blocked and possibly simulates an error
2495 * insync - sets Insync providing device isn't active
2496 * -insync - clear Insync for a device with a slot assigned,
2497 * so that it gets rebuilt based on bitmap
2498 * write_error - sets WriteErrorSeen
2499 * -write_error - clears WriteErrorSeen
2502 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2503 md_error(rdev
->mddev
, rdev
);
2504 if (test_bit(Faulty
, &rdev
->flags
))
2508 } else if (cmd_match(buf
, "remove")) {
2509 if (rdev
->raid_disk
>= 0)
2512 struct mddev
*mddev
= rdev
->mddev
;
2513 if (mddev_is_clustered(mddev
))
2514 md_cluster_ops
->remove_disk(mddev
, rdev
);
2515 md_kick_rdev_from_array(rdev
);
2516 if (mddev_is_clustered(mddev
))
2517 md_cluster_ops
->metadata_update_start(mddev
);
2519 md_update_sb(mddev
, 1);
2520 md_new_event(mddev
);
2521 if (mddev_is_clustered(mddev
))
2522 md_cluster_ops
->metadata_update_finish(mddev
);
2525 } else if (cmd_match(buf
, "writemostly")) {
2526 set_bit(WriteMostly
, &rdev
->flags
);
2528 } else if (cmd_match(buf
, "-writemostly")) {
2529 clear_bit(WriteMostly
, &rdev
->flags
);
2531 } else if (cmd_match(buf
, "blocked")) {
2532 set_bit(Blocked
, &rdev
->flags
);
2534 } else if (cmd_match(buf
, "-blocked")) {
2535 if (!test_bit(Faulty
, &rdev
->flags
) &&
2536 rdev
->badblocks
.unacked_exist
) {
2537 /* metadata handler doesn't understand badblocks,
2538 * so we need to fail the device
2540 md_error(rdev
->mddev
, rdev
);
2542 clear_bit(Blocked
, &rdev
->flags
);
2543 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2544 wake_up(&rdev
->blocked_wait
);
2545 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2546 md_wakeup_thread(rdev
->mddev
->thread
);
2549 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2550 set_bit(In_sync
, &rdev
->flags
);
2552 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0) {
2553 if (rdev
->mddev
->pers
== NULL
) {
2554 clear_bit(In_sync
, &rdev
->flags
);
2555 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2556 rdev
->raid_disk
= -1;
2559 } else if (cmd_match(buf
, "write_error")) {
2560 set_bit(WriteErrorSeen
, &rdev
->flags
);
2562 } else if (cmd_match(buf
, "-write_error")) {
2563 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2565 } else if (cmd_match(buf
, "want_replacement")) {
2566 /* Any non-spare device that is not a replacement can
2567 * become want_replacement at any time, but we then need to
2568 * check if recovery is needed.
2570 if (rdev
->raid_disk
>= 0 &&
2571 !test_bit(Replacement
, &rdev
->flags
))
2572 set_bit(WantReplacement
, &rdev
->flags
);
2573 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2574 md_wakeup_thread(rdev
->mddev
->thread
);
2576 } else if (cmd_match(buf
, "-want_replacement")) {
2577 /* Clearing 'want_replacement' is always allowed.
2578 * Once replacements starts it is too late though.
2581 clear_bit(WantReplacement
, &rdev
->flags
);
2582 } else if (cmd_match(buf
, "replacement")) {
2583 /* Can only set a device as a replacement when array has not
2584 * yet been started. Once running, replacement is automatic
2585 * from spares, or by assigning 'slot'.
2587 if (rdev
->mddev
->pers
)
2590 set_bit(Replacement
, &rdev
->flags
);
2593 } else if (cmd_match(buf
, "-replacement")) {
2594 /* Similarly, can only clear Replacement before start */
2595 if (rdev
->mddev
->pers
)
2598 clear_bit(Replacement
, &rdev
->flags
);
2601 } else if (cmd_match(buf
, "re-add")) {
2602 if (test_bit(Faulty
, &rdev
->flags
) && (rdev
->raid_disk
== -1)) {
2603 /* clear_bit is performed _after_ all the devices
2604 * have their local Faulty bit cleared. If any writes
2605 * happen in the meantime in the local node, they
2606 * will land in the local bitmap, which will be synced
2607 * by this node eventually
2609 if (!mddev_is_clustered(rdev
->mddev
) ||
2610 (err
= md_cluster_ops
->gather_bitmaps(rdev
)) == 0) {
2611 clear_bit(Faulty
, &rdev
->flags
);
2612 err
= add_bound_rdev(rdev
);
2618 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2619 return err
? err
: len
;
2621 static struct rdev_sysfs_entry rdev_state
=
2622 __ATTR_PREALLOC(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2625 errors_show(struct md_rdev
*rdev
, char *page
)
2627 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2631 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2634 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2635 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2636 atomic_set(&rdev
->corrected_errors
, n
);
2641 static struct rdev_sysfs_entry rdev_errors
=
2642 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2645 slot_show(struct md_rdev
*rdev
, char *page
)
2647 if (rdev
->raid_disk
< 0)
2648 return sprintf(page
, "none\n");
2650 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2654 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2658 int slot
= simple_strtoul(buf
, &e
, 10);
2659 if (strncmp(buf
, "none", 4)==0)
2661 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2663 if (rdev
->mddev
->pers
&& slot
== -1) {
2664 /* Setting 'slot' on an active array requires also
2665 * updating the 'rd%d' link, and communicating
2666 * with the personality with ->hot_*_disk.
2667 * For now we only support removing
2668 * failed/spare devices. This normally happens automatically,
2669 * but not when the metadata is externally managed.
2671 if (rdev
->raid_disk
== -1)
2673 /* personality does all needed checks */
2674 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2676 clear_bit(Blocked
, &rdev
->flags
);
2677 remove_and_add_spares(rdev
->mddev
, rdev
);
2678 if (rdev
->raid_disk
>= 0)
2680 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2681 md_wakeup_thread(rdev
->mddev
->thread
);
2682 } else if (rdev
->mddev
->pers
) {
2683 /* Activating a spare .. or possibly reactivating
2684 * if we ever get bitmaps working here.
2687 if (rdev
->raid_disk
!= -1)
2690 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2693 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2696 if (slot
>= rdev
->mddev
->raid_disks
&&
2697 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2700 rdev
->raid_disk
= slot
;
2701 if (test_bit(In_sync
, &rdev
->flags
))
2702 rdev
->saved_raid_disk
= slot
;
2704 rdev
->saved_raid_disk
= -1;
2705 clear_bit(In_sync
, &rdev
->flags
);
2706 clear_bit(Bitmap_sync
, &rdev
->flags
);
2707 err
= rdev
->mddev
->pers
->
2708 hot_add_disk(rdev
->mddev
, rdev
);
2710 rdev
->raid_disk
= -1;
2713 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2714 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2715 /* failure here is OK */;
2716 /* don't wakeup anyone, leave that to userspace. */
2718 if (slot
>= rdev
->mddev
->raid_disks
&&
2719 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2721 rdev
->raid_disk
= slot
;
2722 /* assume it is working */
2723 clear_bit(Faulty
, &rdev
->flags
);
2724 clear_bit(WriteMostly
, &rdev
->flags
);
2725 set_bit(In_sync
, &rdev
->flags
);
2726 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2731 static struct rdev_sysfs_entry rdev_slot
=
2732 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2735 offset_show(struct md_rdev
*rdev
, char *page
)
2737 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2741 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2743 unsigned long long offset
;
2744 if (kstrtoull(buf
, 10, &offset
) < 0)
2746 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2748 if (rdev
->sectors
&& rdev
->mddev
->external
)
2749 /* Must set offset before size, so overlap checks
2752 rdev
->data_offset
= offset
;
2753 rdev
->new_data_offset
= offset
;
2757 static struct rdev_sysfs_entry rdev_offset
=
2758 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2760 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2762 return sprintf(page
, "%llu\n",
2763 (unsigned long long)rdev
->new_data_offset
);
2766 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2767 const char *buf
, size_t len
)
2769 unsigned long long new_offset
;
2770 struct mddev
*mddev
= rdev
->mddev
;
2772 if (kstrtoull(buf
, 10, &new_offset
) < 0)
2775 if (mddev
->sync_thread
||
2776 test_bit(MD_RECOVERY_RUNNING
,&mddev
->recovery
))
2778 if (new_offset
== rdev
->data_offset
)
2779 /* reset is always permitted */
2781 else if (new_offset
> rdev
->data_offset
) {
2782 /* must not push array size beyond rdev_sectors */
2783 if (new_offset
- rdev
->data_offset
2784 + mddev
->dev_sectors
> rdev
->sectors
)
2787 /* Metadata worries about other space details. */
2789 /* decreasing the offset is inconsistent with a backwards
2792 if (new_offset
< rdev
->data_offset
&&
2793 mddev
->reshape_backwards
)
2795 /* Increasing offset is inconsistent with forwards
2796 * reshape. reshape_direction should be set to
2797 * 'backwards' first.
2799 if (new_offset
> rdev
->data_offset
&&
2800 !mddev
->reshape_backwards
)
2803 if (mddev
->pers
&& mddev
->persistent
&&
2804 !super_types
[mddev
->major_version
]
2805 .allow_new_offset(rdev
, new_offset
))
2807 rdev
->new_data_offset
= new_offset
;
2808 if (new_offset
> rdev
->data_offset
)
2809 mddev
->reshape_backwards
= 1;
2810 else if (new_offset
< rdev
->data_offset
)
2811 mddev
->reshape_backwards
= 0;
2815 static struct rdev_sysfs_entry rdev_new_offset
=
2816 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2819 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2821 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2824 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2826 /* check if two start/length pairs overlap */
2834 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2836 unsigned long long blocks
;
2839 if (kstrtoull(buf
, 10, &blocks
) < 0)
2842 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2843 return -EINVAL
; /* sector conversion overflow */
2846 if (new != blocks
* 2)
2847 return -EINVAL
; /* unsigned long long to sector_t overflow */
2854 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2856 struct mddev
*my_mddev
= rdev
->mddev
;
2857 sector_t oldsectors
= rdev
->sectors
;
2860 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2862 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2863 return -EINVAL
; /* too confusing */
2864 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2865 if (my_mddev
->persistent
) {
2866 sectors
= super_types
[my_mddev
->major_version
].
2867 rdev_size_change(rdev
, sectors
);
2870 } else if (!sectors
)
2871 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2873 if (!my_mddev
->pers
->resize
)
2874 /* Cannot change size for RAID0 or Linear etc */
2877 if (sectors
< my_mddev
->dev_sectors
)
2878 return -EINVAL
; /* component must fit device */
2880 rdev
->sectors
= sectors
;
2881 if (sectors
> oldsectors
&& my_mddev
->external
) {
2882 /* Need to check that all other rdevs with the same
2883 * ->bdev do not overlap. 'rcu' is sufficient to walk
2884 * the rdev lists safely.
2885 * This check does not provide a hard guarantee, it
2886 * just helps avoid dangerous mistakes.
2888 struct mddev
*mddev
;
2890 struct list_head
*tmp
;
2893 for_each_mddev(mddev
, tmp
) {
2894 struct md_rdev
*rdev2
;
2896 rdev_for_each(rdev2
, mddev
)
2897 if (rdev
->bdev
== rdev2
->bdev
&&
2899 overlaps(rdev
->data_offset
, rdev
->sectors
,
2912 /* Someone else could have slipped in a size
2913 * change here, but doing so is just silly.
2914 * We put oldsectors back because we *know* it is
2915 * safe, and trust userspace not to race with
2918 rdev
->sectors
= oldsectors
;
2925 static struct rdev_sysfs_entry rdev_size
=
2926 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2928 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2930 unsigned long long recovery_start
= rdev
->recovery_offset
;
2932 if (test_bit(In_sync
, &rdev
->flags
) ||
2933 recovery_start
== MaxSector
)
2934 return sprintf(page
, "none\n");
2936 return sprintf(page
, "%llu\n", recovery_start
);
2939 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2941 unsigned long long recovery_start
;
2943 if (cmd_match(buf
, "none"))
2944 recovery_start
= MaxSector
;
2945 else if (kstrtoull(buf
, 10, &recovery_start
))
2948 if (rdev
->mddev
->pers
&&
2949 rdev
->raid_disk
>= 0)
2952 rdev
->recovery_offset
= recovery_start
;
2953 if (recovery_start
== MaxSector
)
2954 set_bit(In_sync
, &rdev
->flags
);
2956 clear_bit(In_sync
, &rdev
->flags
);
2960 static struct rdev_sysfs_entry rdev_recovery_start
=
2961 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2964 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2966 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2968 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2970 return badblocks_show(&rdev
->badblocks
, page
, 0);
2972 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2974 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2975 /* Maybe that ack was all we needed */
2976 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2977 wake_up(&rdev
->blocked_wait
);
2980 static struct rdev_sysfs_entry rdev_bad_blocks
=
2981 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2983 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2985 return badblocks_show(&rdev
->badblocks
, page
, 1);
2987 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2989 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2991 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2992 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2994 static struct attribute
*rdev_default_attrs
[] = {
2999 &rdev_new_offset
.attr
,
3001 &rdev_recovery_start
.attr
,
3002 &rdev_bad_blocks
.attr
,
3003 &rdev_unack_bad_blocks
.attr
,
3007 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3009 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3010 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3016 return entry
->show(rdev
, page
);
3020 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3021 const char *page
, size_t length
)
3023 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3024 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3026 struct mddev
*mddev
= rdev
->mddev
;
3030 if (!capable(CAP_SYS_ADMIN
))
3032 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3034 if (rdev
->mddev
== NULL
)
3037 rv
= entry
->store(rdev
, page
, length
);
3038 mddev_unlock(mddev
);
3043 static void rdev_free(struct kobject
*ko
)
3045 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3048 static const struct sysfs_ops rdev_sysfs_ops
= {
3049 .show
= rdev_attr_show
,
3050 .store
= rdev_attr_store
,
3052 static struct kobj_type rdev_ktype
= {
3053 .release
= rdev_free
,
3054 .sysfs_ops
= &rdev_sysfs_ops
,
3055 .default_attrs
= rdev_default_attrs
,
3058 int md_rdev_init(struct md_rdev
*rdev
)
3061 rdev
->saved_raid_disk
= -1;
3062 rdev
->raid_disk
= -1;
3064 rdev
->data_offset
= 0;
3065 rdev
->new_data_offset
= 0;
3066 rdev
->sb_events
= 0;
3067 rdev
->last_read_error
.tv_sec
= 0;
3068 rdev
->last_read_error
.tv_nsec
= 0;
3069 rdev
->sb_loaded
= 0;
3070 rdev
->bb_page
= NULL
;
3071 atomic_set(&rdev
->nr_pending
, 0);
3072 atomic_set(&rdev
->read_errors
, 0);
3073 atomic_set(&rdev
->corrected_errors
, 0);
3075 INIT_LIST_HEAD(&rdev
->same_set
);
3076 init_waitqueue_head(&rdev
->blocked_wait
);
3078 /* Add space to store bad block list.
3079 * This reserves the space even on arrays where it cannot
3080 * be used - I wonder if that matters
3082 rdev
->badblocks
.count
= 0;
3083 rdev
->badblocks
.shift
= -1; /* disabled until explicitly enabled */
3084 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3085 seqlock_init(&rdev
->badblocks
.lock
);
3086 if (rdev
->badblocks
.page
== NULL
)
3091 EXPORT_SYMBOL_GPL(md_rdev_init
);
3093 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3095 * mark the device faulty if:
3097 * - the device is nonexistent (zero size)
3098 * - the device has no valid superblock
3100 * a faulty rdev _never_ has rdev->sb set.
3102 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3104 char b
[BDEVNAME_SIZE
];
3106 struct md_rdev
*rdev
;
3109 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3111 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3112 return ERR_PTR(-ENOMEM
);
3115 err
= md_rdev_init(rdev
);
3118 err
= alloc_disk_sb(rdev
);
3122 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3126 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3128 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3131 "md: %s has zero or unknown size, marking faulty!\n",
3132 bdevname(rdev
->bdev
,b
));
3137 if (super_format
>= 0) {
3138 err
= super_types
[super_format
].
3139 load_super(rdev
, NULL
, super_minor
);
3140 if (err
== -EINVAL
) {
3142 "md: %s does not have a valid v%d.%d "
3143 "superblock, not importing!\n",
3144 bdevname(rdev
->bdev
,b
),
3145 super_format
, super_minor
);
3150 "md: could not read %s's sb, not importing!\n",
3151 bdevname(rdev
->bdev
,b
));
3161 md_rdev_clear(rdev
);
3163 return ERR_PTR(err
);
3167 * Check a full RAID array for plausibility
3170 static void analyze_sbs(struct mddev
*mddev
)
3173 struct md_rdev
*rdev
, *freshest
, *tmp
;
3174 char b
[BDEVNAME_SIZE
];
3177 rdev_for_each_safe(rdev
, tmp
, mddev
)
3178 switch (super_types
[mddev
->major_version
].
3179 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3187 "md: fatal superblock inconsistency in %s"
3188 " -- removing from array\n",
3189 bdevname(rdev
->bdev
,b
));
3190 md_kick_rdev_from_array(rdev
);
3193 super_types
[mddev
->major_version
].
3194 validate_super(mddev
, freshest
);
3197 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3198 if (mddev
->max_disks
&&
3199 (rdev
->desc_nr
>= mddev
->max_disks
||
3200 i
> mddev
->max_disks
)) {
3202 "md: %s: %s: only %d devices permitted\n",
3203 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3205 md_kick_rdev_from_array(rdev
);
3208 if (rdev
!= freshest
) {
3209 if (super_types
[mddev
->major_version
].
3210 validate_super(mddev
, rdev
)) {
3211 printk(KERN_WARNING
"md: kicking non-fresh %s"
3213 bdevname(rdev
->bdev
,b
));
3214 md_kick_rdev_from_array(rdev
);
3217 /* No device should have a Candidate flag
3218 * when reading devices
3220 if (test_bit(Candidate
, &rdev
->flags
)) {
3221 pr_info("md: kicking Cluster Candidate %s from array!\n",
3222 bdevname(rdev
->bdev
, b
));
3223 md_kick_rdev_from_array(rdev
);
3226 if (mddev
->level
== LEVEL_MULTIPATH
) {
3227 rdev
->desc_nr
= i
++;
3228 rdev
->raid_disk
= rdev
->desc_nr
;
3229 set_bit(In_sync
, &rdev
->flags
);
3230 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3231 rdev
->raid_disk
= -1;
3232 clear_bit(In_sync
, &rdev
->flags
);
3237 /* Read a fixed-point number.
3238 * Numbers in sysfs attributes should be in "standard" units where
3239 * possible, so time should be in seconds.
3240 * However we internally use a a much smaller unit such as
3241 * milliseconds or jiffies.
3242 * This function takes a decimal number with a possible fractional
3243 * component, and produces an integer which is the result of
3244 * multiplying that number by 10^'scale'.
3245 * all without any floating-point arithmetic.
3247 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3249 unsigned long result
= 0;
3251 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3254 else if (decimals
< scale
) {
3257 result
= result
* 10 + value
;
3269 while (decimals
< scale
) {
3277 static void md_safemode_timeout(unsigned long data
);
3280 safe_delay_show(struct mddev
*mddev
, char *page
)
3282 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3283 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3286 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3290 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3293 mddev
->safemode_delay
= 0;
3295 unsigned long old_delay
= mddev
->safemode_delay
;
3296 unsigned long new_delay
= (msec
*HZ
)/1000;
3300 mddev
->safemode_delay
= new_delay
;
3301 if (new_delay
< old_delay
|| old_delay
== 0)
3302 mod_timer(&mddev
->safemode_timer
, jiffies
+1);
3306 static struct md_sysfs_entry md_safe_delay
=
3307 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3310 level_show(struct mddev
*mddev
, char *page
)
3312 struct md_personality
*p
;
3314 spin_lock(&mddev
->lock
);
3317 ret
= sprintf(page
, "%s\n", p
->name
);
3318 else if (mddev
->clevel
[0])
3319 ret
= sprintf(page
, "%s\n", mddev
->clevel
);
3320 else if (mddev
->level
!= LEVEL_NONE
)
3321 ret
= sprintf(page
, "%d\n", mddev
->level
);
3324 spin_unlock(&mddev
->lock
);
3329 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3334 struct md_personality
*pers
, *oldpers
;
3336 void *priv
, *oldpriv
;
3337 struct md_rdev
*rdev
;
3339 if (slen
== 0 || slen
>= sizeof(clevel
))
3342 rv
= mddev_lock(mddev
);
3346 if (mddev
->pers
== NULL
) {
3347 strncpy(mddev
->clevel
, buf
, slen
);
3348 if (mddev
->clevel
[slen
-1] == '\n')
3350 mddev
->clevel
[slen
] = 0;
3351 mddev
->level
= LEVEL_NONE
;
3359 /* request to change the personality. Need to ensure:
3360 * - array is not engaged in resync/recovery/reshape
3361 * - old personality can be suspended
3362 * - new personality will access other array.
3366 if (mddev
->sync_thread
||
3367 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3368 mddev
->reshape_position
!= MaxSector
||
3369 mddev
->sysfs_active
)
3373 if (!mddev
->pers
->quiesce
) {
3374 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3375 mdname(mddev
), mddev
->pers
->name
);
3379 /* Now find the new personality */
3380 strncpy(clevel
, buf
, slen
);
3381 if (clevel
[slen
-1] == '\n')
3384 if (kstrtol(clevel
, 10, &level
))
3387 if (request_module("md-%s", clevel
) != 0)
3388 request_module("md-level-%s", clevel
);
3389 spin_lock(&pers_lock
);
3390 pers
= find_pers(level
, clevel
);
3391 if (!pers
|| !try_module_get(pers
->owner
)) {
3392 spin_unlock(&pers_lock
);
3393 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3397 spin_unlock(&pers_lock
);
3399 if (pers
== mddev
->pers
) {
3400 /* Nothing to do! */
3401 module_put(pers
->owner
);
3405 if (!pers
->takeover
) {
3406 module_put(pers
->owner
);
3407 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3408 mdname(mddev
), clevel
);
3413 rdev_for_each(rdev
, mddev
)
3414 rdev
->new_raid_disk
= rdev
->raid_disk
;
3416 /* ->takeover must set new_* and/or delta_disks
3417 * if it succeeds, and may set them when it fails.
3419 priv
= pers
->takeover(mddev
);
3421 mddev
->new_level
= mddev
->level
;
3422 mddev
->new_layout
= mddev
->layout
;
3423 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3424 mddev
->raid_disks
-= mddev
->delta_disks
;
3425 mddev
->delta_disks
= 0;
3426 mddev
->reshape_backwards
= 0;
3427 module_put(pers
->owner
);
3428 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3429 mdname(mddev
), clevel
);
3434 /* Looks like we have a winner */
3435 mddev_suspend(mddev
);
3436 mddev_detach(mddev
);
3438 spin_lock(&mddev
->lock
);
3439 oldpers
= mddev
->pers
;
3440 oldpriv
= mddev
->private;
3442 mddev
->private = priv
;
3443 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3444 mddev
->level
= mddev
->new_level
;
3445 mddev
->layout
= mddev
->new_layout
;
3446 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3447 mddev
->delta_disks
= 0;
3448 mddev
->reshape_backwards
= 0;
3449 mddev
->degraded
= 0;
3450 spin_unlock(&mddev
->lock
);
3452 if (oldpers
->sync_request
== NULL
&&
3454 /* We are converting from a no-redundancy array
3455 * to a redundancy array and metadata is managed
3456 * externally so we need to be sure that writes
3457 * won't block due to a need to transition
3459 * until external management is started.
3462 mddev
->safemode_delay
= 0;
3463 mddev
->safemode
= 0;
3466 oldpers
->free(mddev
, oldpriv
);
3468 if (oldpers
->sync_request
== NULL
&&
3469 pers
->sync_request
!= NULL
) {
3470 /* need to add the md_redundancy_group */
3471 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3473 "md: cannot register extra attributes for %s\n",
3475 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3477 if (oldpers
->sync_request
!= NULL
&&
3478 pers
->sync_request
== NULL
) {
3479 /* need to remove the md_redundancy_group */
3480 if (mddev
->to_remove
== NULL
)
3481 mddev
->to_remove
= &md_redundancy_group
;
3484 rdev_for_each(rdev
, mddev
) {
3485 if (rdev
->raid_disk
< 0)
3487 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3488 rdev
->new_raid_disk
= -1;
3489 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3491 sysfs_unlink_rdev(mddev
, rdev
);
3493 rdev_for_each(rdev
, mddev
) {
3494 if (rdev
->raid_disk
< 0)
3496 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3498 rdev
->raid_disk
= rdev
->new_raid_disk
;
3499 if (rdev
->raid_disk
< 0)
3500 clear_bit(In_sync
, &rdev
->flags
);
3502 if (sysfs_link_rdev(mddev
, rdev
))
3503 printk(KERN_WARNING
"md: cannot register rd%d"
3504 " for %s after level change\n",
3505 rdev
->raid_disk
, mdname(mddev
));
3509 if (pers
->sync_request
== NULL
) {
3510 /* this is now an array without redundancy, so
3511 * it must always be in_sync
3514 del_timer_sync(&mddev
->safemode_timer
);
3516 blk_set_stacking_limits(&mddev
->queue
->limits
);
3518 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3519 mddev_resume(mddev
);
3521 md_update_sb(mddev
, 1);
3522 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3523 md_new_event(mddev
);
3526 mddev_unlock(mddev
);
3530 static struct md_sysfs_entry md_level
=
3531 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3534 layout_show(struct mddev
*mddev
, char *page
)
3536 /* just a number, not meaningful for all levels */
3537 if (mddev
->reshape_position
!= MaxSector
&&
3538 mddev
->layout
!= mddev
->new_layout
)
3539 return sprintf(page
, "%d (%d)\n",
3540 mddev
->new_layout
, mddev
->layout
);
3541 return sprintf(page
, "%d\n", mddev
->layout
);
3545 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3548 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3551 if (!*buf
|| (*e
&& *e
!= '\n'))
3553 err
= mddev_lock(mddev
);
3558 if (mddev
->pers
->check_reshape
== NULL
)
3563 mddev
->new_layout
= n
;
3564 err
= mddev
->pers
->check_reshape(mddev
);
3566 mddev
->new_layout
= mddev
->layout
;
3569 mddev
->new_layout
= n
;
3570 if (mddev
->reshape_position
== MaxSector
)
3573 mddev_unlock(mddev
);
3576 static struct md_sysfs_entry md_layout
=
3577 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3580 raid_disks_show(struct mddev
*mddev
, char *page
)
3582 if (mddev
->raid_disks
== 0)
3584 if (mddev
->reshape_position
!= MaxSector
&&
3585 mddev
->delta_disks
!= 0)
3586 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3587 mddev
->raid_disks
- mddev
->delta_disks
);
3588 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3591 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3594 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3598 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3600 if (!*buf
|| (*e
&& *e
!= '\n'))
3603 err
= mddev_lock(mddev
);
3607 err
= update_raid_disks(mddev
, n
);
3608 else if (mddev
->reshape_position
!= MaxSector
) {
3609 struct md_rdev
*rdev
;
3610 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3613 rdev_for_each(rdev
, mddev
) {
3615 rdev
->data_offset
< rdev
->new_data_offset
)
3618 rdev
->data_offset
> rdev
->new_data_offset
)
3622 mddev
->delta_disks
= n
- olddisks
;
3623 mddev
->raid_disks
= n
;
3624 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3626 mddev
->raid_disks
= n
;
3628 mddev_unlock(mddev
);
3629 return err
? err
: len
;
3631 static struct md_sysfs_entry md_raid_disks
=
3632 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3635 chunk_size_show(struct mddev
*mddev
, char *page
)
3637 if (mddev
->reshape_position
!= MaxSector
&&
3638 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3639 return sprintf(page
, "%d (%d)\n",
3640 mddev
->new_chunk_sectors
<< 9,
3641 mddev
->chunk_sectors
<< 9);
3642 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3646 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3650 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3652 if (!*buf
|| (*e
&& *e
!= '\n'))
3655 err
= mddev_lock(mddev
);
3659 if (mddev
->pers
->check_reshape
== NULL
)
3664 mddev
->new_chunk_sectors
= n
>> 9;
3665 err
= mddev
->pers
->check_reshape(mddev
);
3667 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3670 mddev
->new_chunk_sectors
= n
>> 9;
3671 if (mddev
->reshape_position
== MaxSector
)
3672 mddev
->chunk_sectors
= n
>> 9;
3674 mddev_unlock(mddev
);
3677 static struct md_sysfs_entry md_chunk_size
=
3678 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3681 resync_start_show(struct mddev
*mddev
, char *page
)
3683 if (mddev
->recovery_cp
== MaxSector
)
3684 return sprintf(page
, "none\n");
3685 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3689 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3693 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3695 err
= mddev_lock(mddev
);
3698 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3700 else if (cmd_match(buf
, "none"))
3702 else if (!*buf
|| (*e
&& *e
!= '\n'))
3706 mddev
->recovery_cp
= n
;
3708 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3710 mddev_unlock(mddev
);
3713 static struct md_sysfs_entry md_resync_start
=
3714 __ATTR_PREALLOC(resync_start
, S_IRUGO
|S_IWUSR
,
3715 resync_start_show
, resync_start_store
);
3718 * The array state can be:
3721 * No devices, no size, no level
3722 * Equivalent to STOP_ARRAY ioctl
3724 * May have some settings, but array is not active
3725 * all IO results in error
3726 * When written, doesn't tear down array, but just stops it
3727 * suspended (not supported yet)
3728 * All IO requests will block. The array can be reconfigured.
3729 * Writing this, if accepted, will block until array is quiescent
3731 * no resync can happen. no superblocks get written.
3732 * write requests fail
3734 * like readonly, but behaves like 'clean' on a write request.
3736 * clean - no pending writes, but otherwise active.
3737 * When written to inactive array, starts without resync
3738 * If a write request arrives then
3739 * if metadata is known, mark 'dirty' and switch to 'active'.
3740 * if not known, block and switch to write-pending
3741 * If written to an active array that has pending writes, then fails.
3743 * fully active: IO and resync can be happening.
3744 * When written to inactive array, starts with resync
3747 * clean, but writes are blocked waiting for 'active' to be written.
3750 * like active, but no writes have been seen for a while (100msec).
3753 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3754 write_pending
, active_idle
, bad_word
};
3755 static char *array_states
[] = {
3756 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3757 "write-pending", "active-idle", NULL
};
3759 static int match_word(const char *word
, char **list
)
3762 for (n
=0; list
[n
]; n
++)
3763 if (cmd_match(word
, list
[n
]))
3769 array_state_show(struct mddev
*mddev
, char *page
)
3771 enum array_state st
= inactive
;
3784 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3786 else if (mddev
->safemode
)
3792 if (list_empty(&mddev
->disks
) &&
3793 mddev
->raid_disks
== 0 &&
3794 mddev
->dev_sectors
== 0)
3799 return sprintf(page
, "%s\n", array_states
[st
]);
3802 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
3803 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
3804 static int do_md_run(struct mddev
*mddev
);
3805 static int restart_array(struct mddev
*mddev
);
3808 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3811 enum array_state st
= match_word(buf
, array_states
);
3813 if (mddev
->pers
&& (st
== active
|| st
== clean
) && mddev
->ro
!= 1) {
3814 /* don't take reconfig_mutex when toggling between
3817 spin_lock(&mddev
->lock
);
3819 restart_array(mddev
);
3820 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3821 wake_up(&mddev
->sb_wait
);
3823 } else /* st == clean */ {
3824 restart_array(mddev
);
3825 if (atomic_read(&mddev
->writes_pending
) == 0) {
3826 if (mddev
->in_sync
== 0) {
3828 if (mddev
->safemode
== 1)
3829 mddev
->safemode
= 0;
3830 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3836 spin_unlock(&mddev
->lock
);
3839 err
= mddev_lock(mddev
);
3847 /* stopping an active array */
3848 err
= do_md_stop(mddev
, 0, NULL
);
3851 /* stopping an active array */
3853 err
= do_md_stop(mddev
, 2, NULL
);
3855 err
= 0; /* already inactive */
3858 break; /* not supported yet */
3861 err
= md_set_readonly(mddev
, NULL
);
3864 set_disk_ro(mddev
->gendisk
, 1);
3865 err
= do_md_run(mddev
);
3871 err
= md_set_readonly(mddev
, NULL
);
3872 else if (mddev
->ro
== 1)
3873 err
= restart_array(mddev
);
3876 set_disk_ro(mddev
->gendisk
, 0);
3880 err
= do_md_run(mddev
);
3885 restart_array(mddev
);
3886 spin_lock(&mddev
->lock
);
3887 if (atomic_read(&mddev
->writes_pending
) == 0) {
3888 if (mddev
->in_sync
== 0) {
3890 if (mddev
->safemode
== 1)
3891 mddev
->safemode
= 0;
3892 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3897 spin_unlock(&mddev
->lock
);
3903 restart_array(mddev
);
3904 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3905 wake_up(&mddev
->sb_wait
);
3909 set_disk_ro(mddev
->gendisk
, 0);
3910 err
= do_md_run(mddev
);
3915 /* these cannot be set */
3920 if (mddev
->hold_active
== UNTIL_IOCTL
)
3921 mddev
->hold_active
= 0;
3922 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3924 mddev_unlock(mddev
);
3927 static struct md_sysfs_entry md_array_state
=
3928 __ATTR_PREALLOC(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3931 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3932 return sprintf(page
, "%d\n",
3933 atomic_read(&mddev
->max_corr_read_errors
));
3937 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3940 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3942 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3943 atomic_set(&mddev
->max_corr_read_errors
, n
);
3949 static struct md_sysfs_entry max_corr_read_errors
=
3950 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3951 max_corrected_read_errors_store
);
3954 null_show(struct mddev
*mddev
, char *page
)
3960 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3962 /* buf must be %d:%d\n? giving major and minor numbers */
3963 /* The new device is added to the array.
3964 * If the array has a persistent superblock, we read the
3965 * superblock to initialise info and check validity.
3966 * Otherwise, only checking done is that in bind_rdev_to_array,
3967 * which mainly checks size.
3970 int major
= simple_strtoul(buf
, &e
, 10);
3973 struct md_rdev
*rdev
;
3976 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3978 minor
= simple_strtoul(e
+1, &e
, 10);
3979 if (*e
&& *e
!= '\n')
3981 dev
= MKDEV(major
, minor
);
3982 if (major
!= MAJOR(dev
) ||
3983 minor
!= MINOR(dev
))
3986 flush_workqueue(md_misc_wq
);
3988 err
= mddev_lock(mddev
);
3991 if (mddev
->persistent
) {
3992 rdev
= md_import_device(dev
, mddev
->major_version
,
3993 mddev
->minor_version
);
3994 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3995 struct md_rdev
*rdev0
3996 = list_entry(mddev
->disks
.next
,
3997 struct md_rdev
, same_set
);
3998 err
= super_types
[mddev
->major_version
]
3999 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4003 } else if (mddev
->external
)
4004 rdev
= md_import_device(dev
, -2, -1);
4006 rdev
= md_import_device(dev
, -1, -1);
4009 return PTR_ERR(rdev
);
4010 err
= bind_rdev_to_array(rdev
, mddev
);
4014 mddev_unlock(mddev
);
4015 return err
? err
: len
;
4018 static struct md_sysfs_entry md_new_device
=
4019 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4022 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4025 unsigned long chunk
, end_chunk
;
4028 err
= mddev_lock(mddev
);
4033 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4035 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4036 if (buf
== end
) break;
4037 if (*end
== '-') { /* range */
4039 end_chunk
= simple_strtoul(buf
, &end
, 0);
4040 if (buf
== end
) break;
4042 if (*end
&& !isspace(*end
)) break;
4043 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4044 buf
= skip_spaces(end
);
4046 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4048 mddev_unlock(mddev
);
4052 static struct md_sysfs_entry md_bitmap
=
4053 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4056 size_show(struct mddev
*mddev
, char *page
)
4058 return sprintf(page
, "%llu\n",
4059 (unsigned long long)mddev
->dev_sectors
/ 2);
4062 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4065 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4067 /* If array is inactive, we can reduce the component size, but
4068 * not increase it (except from 0).
4069 * If array is active, we can try an on-line resize
4072 int err
= strict_blocks_to_sectors(buf
, §ors
);
4076 err
= mddev_lock(mddev
);
4080 if (mddev_is_clustered(mddev
))
4081 md_cluster_ops
->metadata_update_start(mddev
);
4082 err
= update_size(mddev
, sectors
);
4083 md_update_sb(mddev
, 1);
4084 if (mddev_is_clustered(mddev
))
4085 md_cluster_ops
->metadata_update_finish(mddev
);
4087 if (mddev
->dev_sectors
== 0 ||
4088 mddev
->dev_sectors
> sectors
)
4089 mddev
->dev_sectors
= sectors
;
4093 mddev_unlock(mddev
);
4094 return err
? err
: len
;
4097 static struct md_sysfs_entry md_size
=
4098 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4100 /* Metadata version.
4102 * 'none' for arrays with no metadata (good luck...)
4103 * 'external' for arrays with externally managed metadata,
4104 * or N.M for internally known formats
4107 metadata_show(struct mddev
*mddev
, char *page
)
4109 if (mddev
->persistent
)
4110 return sprintf(page
, "%d.%d\n",
4111 mddev
->major_version
, mddev
->minor_version
);
4112 else if (mddev
->external
)
4113 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4115 return sprintf(page
, "none\n");
4119 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4124 /* Changing the details of 'external' metadata is
4125 * always permitted. Otherwise there must be
4126 * no devices attached to the array.
4129 err
= mddev_lock(mddev
);
4133 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4135 else if (!list_empty(&mddev
->disks
))
4139 if (cmd_match(buf
, "none")) {
4140 mddev
->persistent
= 0;
4141 mddev
->external
= 0;
4142 mddev
->major_version
= 0;
4143 mddev
->minor_version
= 90;
4146 if (strncmp(buf
, "external:", 9) == 0) {
4147 size_t namelen
= len
-9;
4148 if (namelen
>= sizeof(mddev
->metadata_type
))
4149 namelen
= sizeof(mddev
->metadata_type
)-1;
4150 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4151 mddev
->metadata_type
[namelen
] = 0;
4152 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4153 mddev
->metadata_type
[--namelen
] = 0;
4154 mddev
->persistent
= 0;
4155 mddev
->external
= 1;
4156 mddev
->major_version
= 0;
4157 mddev
->minor_version
= 90;
4160 major
= simple_strtoul(buf
, &e
, 10);
4162 if (e
==buf
|| *e
!= '.')
4165 minor
= simple_strtoul(buf
, &e
, 10);
4166 if (e
==buf
|| (*e
&& *e
!= '\n') )
4169 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4171 mddev
->major_version
= major
;
4172 mddev
->minor_version
= minor
;
4173 mddev
->persistent
= 1;
4174 mddev
->external
= 0;
4177 mddev_unlock(mddev
);
4181 static struct md_sysfs_entry md_metadata
=
4182 __ATTR_PREALLOC(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4185 action_show(struct mddev
*mddev
, char *page
)
4187 char *type
= "idle";
4188 unsigned long recovery
= mddev
->recovery
;
4189 if (test_bit(MD_RECOVERY_FROZEN
, &recovery
))
4191 else if (test_bit(MD_RECOVERY_RUNNING
, &recovery
) ||
4192 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &recovery
))) {
4193 if (test_bit(MD_RECOVERY_RESHAPE
, &recovery
))
4195 else if (test_bit(MD_RECOVERY_SYNC
, &recovery
)) {
4196 if (!test_bit(MD_RECOVERY_REQUESTED
, &recovery
))
4198 else if (test_bit(MD_RECOVERY_CHECK
, &recovery
))
4202 } else if (test_bit(MD_RECOVERY_RECOVER
, &recovery
))
4205 return sprintf(page
, "%s\n", type
);
4209 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4211 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4214 if (cmd_match(page
, "frozen"))
4215 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4217 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4219 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4220 flush_workqueue(md_misc_wq
);
4221 if (mddev
->sync_thread
) {
4222 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4223 if (mddev_lock(mddev
) == 0) {
4224 md_reap_sync_thread(mddev
);
4225 mddev_unlock(mddev
);
4228 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4229 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4231 else if (cmd_match(page
, "resync"))
4232 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4233 else if (cmd_match(page
, "recover")) {
4234 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4235 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4236 } else if (cmd_match(page
, "reshape")) {
4238 if (mddev
->pers
->start_reshape
== NULL
)
4240 err
= mddev_lock(mddev
);
4242 err
= mddev
->pers
->start_reshape(mddev
);
4243 mddev_unlock(mddev
);
4247 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4249 if (cmd_match(page
, "check"))
4250 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4251 else if (!cmd_match(page
, "repair"))
4253 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4254 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4256 if (mddev
->ro
== 2) {
4257 /* A write to sync_action is enough to justify
4258 * canceling read-auto mode
4261 md_wakeup_thread(mddev
->sync_thread
);
4263 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4264 md_wakeup_thread(mddev
->thread
);
4265 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4269 static struct md_sysfs_entry md_scan_mode
=
4270 __ATTR_PREALLOC(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4273 last_sync_action_show(struct mddev
*mddev
, char *page
)
4275 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4278 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4281 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4283 return sprintf(page
, "%llu\n",
4284 (unsigned long long)
4285 atomic64_read(&mddev
->resync_mismatches
));
4288 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4291 sync_min_show(struct mddev
*mddev
, char *page
)
4293 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4294 mddev
->sync_speed_min
? "local": "system");
4298 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4302 if (strncmp(buf
, "system", 6)==0) {
4303 mddev
->sync_speed_min
= 0;
4306 min
= simple_strtoul(buf
, &e
, 10);
4307 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4309 mddev
->sync_speed_min
= min
;
4313 static struct md_sysfs_entry md_sync_min
=
4314 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4317 sync_max_show(struct mddev
*mddev
, char *page
)
4319 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4320 mddev
->sync_speed_max
? "local": "system");
4324 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4328 if (strncmp(buf
, "system", 6)==0) {
4329 mddev
->sync_speed_max
= 0;
4332 max
= simple_strtoul(buf
, &e
, 10);
4333 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4335 mddev
->sync_speed_max
= max
;
4339 static struct md_sysfs_entry md_sync_max
=
4340 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4343 degraded_show(struct mddev
*mddev
, char *page
)
4345 return sprintf(page
, "%d\n", mddev
->degraded
);
4347 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4350 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4352 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4356 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4360 if (kstrtol(buf
, 10, &n
))
4363 if (n
!= 0 && n
!= 1)
4366 mddev
->parallel_resync
= n
;
4368 if (mddev
->sync_thread
)
4369 wake_up(&resync_wait
);
4374 /* force parallel resync, even with shared block devices */
4375 static struct md_sysfs_entry md_sync_force_parallel
=
4376 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4377 sync_force_parallel_show
, sync_force_parallel_store
);
4380 sync_speed_show(struct mddev
*mddev
, char *page
)
4382 unsigned long resync
, dt
, db
;
4383 if (mddev
->curr_resync
== 0)
4384 return sprintf(page
, "none\n");
4385 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4386 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4388 db
= resync
- mddev
->resync_mark_cnt
;
4389 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4392 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4395 sync_completed_show(struct mddev
*mddev
, char *page
)
4397 unsigned long long max_sectors
, resync
;
4399 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4400 return sprintf(page
, "none\n");
4402 if (mddev
->curr_resync
== 1 ||
4403 mddev
->curr_resync
== 2)
4404 return sprintf(page
, "delayed\n");
4406 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4407 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4408 max_sectors
= mddev
->resync_max_sectors
;
4410 max_sectors
= mddev
->dev_sectors
;
4412 resync
= mddev
->curr_resync_completed
;
4413 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4416 static struct md_sysfs_entry md_sync_completed
=
4417 __ATTR_PREALLOC(sync_completed
, S_IRUGO
, sync_completed_show
, NULL
);
4420 min_sync_show(struct mddev
*mddev
, char *page
)
4422 return sprintf(page
, "%llu\n",
4423 (unsigned long long)mddev
->resync_min
);
4426 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4428 unsigned long long min
;
4431 if (kstrtoull(buf
, 10, &min
))
4434 spin_lock(&mddev
->lock
);
4436 if (min
> mddev
->resync_max
)
4440 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4443 /* Round down to multiple of 4K for safety */
4444 mddev
->resync_min
= round_down(min
, 8);
4448 spin_unlock(&mddev
->lock
);
4452 static struct md_sysfs_entry md_min_sync
=
4453 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4456 max_sync_show(struct mddev
*mddev
, char *page
)
4458 if (mddev
->resync_max
== MaxSector
)
4459 return sprintf(page
, "max\n");
4461 return sprintf(page
, "%llu\n",
4462 (unsigned long long)mddev
->resync_max
);
4465 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4468 spin_lock(&mddev
->lock
);
4469 if (strncmp(buf
, "max", 3) == 0)
4470 mddev
->resync_max
= MaxSector
;
4472 unsigned long long max
;
4476 if (kstrtoull(buf
, 10, &max
))
4478 if (max
< mddev
->resync_min
)
4482 if (max
< mddev
->resync_max
&&
4484 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4487 /* Must be a multiple of chunk_size */
4488 chunk
= mddev
->chunk_sectors
;
4490 sector_t temp
= max
;
4493 if (sector_div(temp
, chunk
))
4496 mddev
->resync_max
= max
;
4498 wake_up(&mddev
->recovery_wait
);
4501 spin_unlock(&mddev
->lock
);
4505 static struct md_sysfs_entry md_max_sync
=
4506 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4509 suspend_lo_show(struct mddev
*mddev
, char *page
)
4511 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4515 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4518 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4519 unsigned long long old
;
4522 if (buf
== e
|| (*e
&& *e
!= '\n'))
4525 err
= mddev_lock(mddev
);
4529 if (mddev
->pers
== NULL
||
4530 mddev
->pers
->quiesce
== NULL
)
4532 old
= mddev
->suspend_lo
;
4533 mddev
->suspend_lo
= new;
4535 /* Shrinking suspended region */
4536 mddev
->pers
->quiesce(mddev
, 2);
4538 /* Expanding suspended region - need to wait */
4539 mddev
->pers
->quiesce(mddev
, 1);
4540 mddev
->pers
->quiesce(mddev
, 0);
4544 mddev_unlock(mddev
);
4547 static struct md_sysfs_entry md_suspend_lo
=
4548 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4551 suspend_hi_show(struct mddev
*mddev
, char *page
)
4553 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4557 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4560 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4561 unsigned long long old
;
4564 if (buf
== e
|| (*e
&& *e
!= '\n'))
4567 err
= mddev_lock(mddev
);
4571 if (mddev
->pers
== NULL
||
4572 mddev
->pers
->quiesce
== NULL
)
4574 old
= mddev
->suspend_hi
;
4575 mddev
->suspend_hi
= new;
4577 /* Shrinking suspended region */
4578 mddev
->pers
->quiesce(mddev
, 2);
4580 /* Expanding suspended region - need to wait */
4581 mddev
->pers
->quiesce(mddev
, 1);
4582 mddev
->pers
->quiesce(mddev
, 0);
4586 mddev_unlock(mddev
);
4589 static struct md_sysfs_entry md_suspend_hi
=
4590 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4593 reshape_position_show(struct mddev
*mddev
, char *page
)
4595 if (mddev
->reshape_position
!= MaxSector
)
4596 return sprintf(page
, "%llu\n",
4597 (unsigned long long)mddev
->reshape_position
);
4598 strcpy(page
, "none\n");
4603 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4605 struct md_rdev
*rdev
;
4608 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4610 if (buf
== e
|| (*e
&& *e
!= '\n'))
4612 err
= mddev_lock(mddev
);
4618 mddev
->reshape_position
= new;
4619 mddev
->delta_disks
= 0;
4620 mddev
->reshape_backwards
= 0;
4621 mddev
->new_level
= mddev
->level
;
4622 mddev
->new_layout
= mddev
->layout
;
4623 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4624 rdev_for_each(rdev
, mddev
)
4625 rdev
->new_data_offset
= rdev
->data_offset
;
4628 mddev_unlock(mddev
);
4632 static struct md_sysfs_entry md_reshape_position
=
4633 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4634 reshape_position_store
);
4637 reshape_direction_show(struct mddev
*mddev
, char *page
)
4639 return sprintf(page
, "%s\n",
4640 mddev
->reshape_backwards
? "backwards" : "forwards");
4644 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4649 if (cmd_match(buf
, "forwards"))
4651 else if (cmd_match(buf
, "backwards"))
4655 if (mddev
->reshape_backwards
== backwards
)
4658 err
= mddev_lock(mddev
);
4661 /* check if we are allowed to change */
4662 if (mddev
->delta_disks
)
4664 else if (mddev
->persistent
&&
4665 mddev
->major_version
== 0)
4668 mddev
->reshape_backwards
= backwards
;
4669 mddev_unlock(mddev
);
4673 static struct md_sysfs_entry md_reshape_direction
=
4674 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4675 reshape_direction_store
);
4678 array_size_show(struct mddev
*mddev
, char *page
)
4680 if (mddev
->external_size
)
4681 return sprintf(page
, "%llu\n",
4682 (unsigned long long)mddev
->array_sectors
/2);
4684 return sprintf(page
, "default\n");
4688 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4693 err
= mddev_lock(mddev
);
4697 if (strncmp(buf
, "default", 7) == 0) {
4699 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4701 sectors
= mddev
->array_sectors
;
4703 mddev
->external_size
= 0;
4705 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4707 else if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4710 mddev
->external_size
= 1;
4714 mddev
->array_sectors
= sectors
;
4716 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4717 revalidate_disk(mddev
->gendisk
);
4720 mddev_unlock(mddev
);
4724 static struct md_sysfs_entry md_array_size
=
4725 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4728 static struct attribute
*md_default_attrs
[] = {
4731 &md_raid_disks
.attr
,
4732 &md_chunk_size
.attr
,
4734 &md_resync_start
.attr
,
4736 &md_new_device
.attr
,
4737 &md_safe_delay
.attr
,
4738 &md_array_state
.attr
,
4739 &md_reshape_position
.attr
,
4740 &md_reshape_direction
.attr
,
4741 &md_array_size
.attr
,
4742 &max_corr_read_errors
.attr
,
4746 static struct attribute
*md_redundancy_attrs
[] = {
4748 &md_last_scan_mode
.attr
,
4749 &md_mismatches
.attr
,
4752 &md_sync_speed
.attr
,
4753 &md_sync_force_parallel
.attr
,
4754 &md_sync_completed
.attr
,
4757 &md_suspend_lo
.attr
,
4758 &md_suspend_hi
.attr
,
4763 static struct attribute_group md_redundancy_group
= {
4765 .attrs
= md_redundancy_attrs
,
4769 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4771 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4772 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4777 spin_lock(&all_mddevs_lock
);
4778 if (list_empty(&mddev
->all_mddevs
)) {
4779 spin_unlock(&all_mddevs_lock
);
4783 spin_unlock(&all_mddevs_lock
);
4785 rv
= entry
->show(mddev
, page
);
4791 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4792 const char *page
, size_t length
)
4794 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4795 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4800 if (!capable(CAP_SYS_ADMIN
))
4802 spin_lock(&all_mddevs_lock
);
4803 if (list_empty(&mddev
->all_mddevs
)) {
4804 spin_unlock(&all_mddevs_lock
);
4808 spin_unlock(&all_mddevs_lock
);
4809 rv
= entry
->store(mddev
, page
, length
);
4814 static void md_free(struct kobject
*ko
)
4816 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4818 if (mddev
->sysfs_state
)
4819 sysfs_put(mddev
->sysfs_state
);
4821 if (mddev
->gendisk
) {
4822 del_gendisk(mddev
->gendisk
);
4823 put_disk(mddev
->gendisk
);
4826 blk_cleanup_queue(mddev
->queue
);
4831 static const struct sysfs_ops md_sysfs_ops
= {
4832 .show
= md_attr_show
,
4833 .store
= md_attr_store
,
4835 static struct kobj_type md_ktype
= {
4837 .sysfs_ops
= &md_sysfs_ops
,
4838 .default_attrs
= md_default_attrs
,
4843 static void mddev_delayed_delete(struct work_struct
*ws
)
4845 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4847 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4848 kobject_del(&mddev
->kobj
);
4849 kobject_put(&mddev
->kobj
);
4852 static int md_alloc(dev_t dev
, char *name
)
4854 static DEFINE_MUTEX(disks_mutex
);
4855 struct mddev
*mddev
= mddev_find(dev
);
4856 struct gendisk
*disk
;
4865 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4866 shift
= partitioned
? MdpMinorShift
: 0;
4867 unit
= MINOR(mddev
->unit
) >> shift
;
4869 /* wait for any previous instance of this device to be
4870 * completely removed (mddev_delayed_delete).
4872 flush_workqueue(md_misc_wq
);
4874 mutex_lock(&disks_mutex
);
4880 /* Need to ensure that 'name' is not a duplicate.
4882 struct mddev
*mddev2
;
4883 spin_lock(&all_mddevs_lock
);
4885 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4886 if (mddev2
->gendisk
&&
4887 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4888 spin_unlock(&all_mddevs_lock
);
4891 spin_unlock(&all_mddevs_lock
);
4895 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4898 mddev
->queue
->queuedata
= mddev
;
4900 blk_queue_make_request(mddev
->queue
, md_make_request
);
4901 blk_set_stacking_limits(&mddev
->queue
->limits
);
4903 disk
= alloc_disk(1 << shift
);
4905 blk_cleanup_queue(mddev
->queue
);
4906 mddev
->queue
= NULL
;
4909 disk
->major
= MAJOR(mddev
->unit
);
4910 disk
->first_minor
= unit
<< shift
;
4912 strcpy(disk
->disk_name
, name
);
4913 else if (partitioned
)
4914 sprintf(disk
->disk_name
, "md_d%d", unit
);
4916 sprintf(disk
->disk_name
, "md%d", unit
);
4917 disk
->fops
= &md_fops
;
4918 disk
->private_data
= mddev
;
4919 disk
->queue
= mddev
->queue
;
4920 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4921 /* Allow extended partitions. This makes the
4922 * 'mdp' device redundant, but we can't really
4925 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4926 mddev
->gendisk
= disk
;
4927 /* As soon as we call add_disk(), another thread could get
4928 * through to md_open, so make sure it doesn't get too far
4930 mutex_lock(&mddev
->open_mutex
);
4933 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4934 &disk_to_dev(disk
)->kobj
, "%s", "md");
4936 /* This isn't possible, but as kobject_init_and_add is marked
4937 * __must_check, we must do something with the result
4939 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4943 if (mddev
->kobj
.sd
&&
4944 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4945 printk(KERN_DEBUG
"pointless warning\n");
4946 mutex_unlock(&mddev
->open_mutex
);
4948 mutex_unlock(&disks_mutex
);
4949 if (!error
&& mddev
->kobj
.sd
) {
4950 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4951 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4957 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4959 md_alloc(dev
, NULL
);
4963 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4965 /* val must be "md_*" where * is not all digits.
4966 * We allocate an array with a large free minor number, and
4967 * set the name to val. val must not already be an active name.
4969 int len
= strlen(val
);
4970 char buf
[DISK_NAME_LEN
];
4972 while (len
&& val
[len
-1] == '\n')
4974 if (len
>= DISK_NAME_LEN
)
4976 strlcpy(buf
, val
, len
+1);
4977 if (strncmp(buf
, "md_", 3) != 0)
4979 return md_alloc(0, buf
);
4982 static void md_safemode_timeout(unsigned long data
)
4984 struct mddev
*mddev
= (struct mddev
*) data
;
4986 if (!atomic_read(&mddev
->writes_pending
)) {
4987 mddev
->safemode
= 1;
4988 if (mddev
->external
)
4989 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4991 md_wakeup_thread(mddev
->thread
);
4994 static int start_dirty_degraded
;
4996 int md_run(struct mddev
*mddev
)
4999 struct md_rdev
*rdev
;
5000 struct md_personality
*pers
;
5002 if (list_empty(&mddev
->disks
))
5003 /* cannot run an array with no devices.. */
5008 /* Cannot run until previous stop completes properly */
5009 if (mddev
->sysfs_active
)
5013 * Analyze all RAID superblock(s)
5015 if (!mddev
->raid_disks
) {
5016 if (!mddev
->persistent
)
5021 if (mddev
->level
!= LEVEL_NONE
)
5022 request_module("md-level-%d", mddev
->level
);
5023 else if (mddev
->clevel
[0])
5024 request_module("md-%s", mddev
->clevel
);
5027 * Drop all container device buffers, from now on
5028 * the only valid external interface is through the md
5031 rdev_for_each(rdev
, mddev
) {
5032 if (test_bit(Faulty
, &rdev
->flags
))
5034 sync_blockdev(rdev
->bdev
);
5035 invalidate_bdev(rdev
->bdev
);
5037 /* perform some consistency tests on the device.
5038 * We don't want the data to overlap the metadata,
5039 * Internal Bitmap issues have been handled elsewhere.
5041 if (rdev
->meta_bdev
) {
5042 /* Nothing to check */;
5043 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5044 if (mddev
->dev_sectors
&&
5045 rdev
->data_offset
+ mddev
->dev_sectors
5047 printk("md: %s: data overlaps metadata\n",
5052 if (rdev
->sb_start
+ rdev
->sb_size
/512
5053 > rdev
->data_offset
) {
5054 printk("md: %s: metadata overlaps data\n",
5059 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5062 if (mddev
->bio_set
== NULL
)
5063 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
5065 spin_lock(&pers_lock
);
5066 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5067 if (!pers
|| !try_module_get(pers
->owner
)) {
5068 spin_unlock(&pers_lock
);
5069 if (mddev
->level
!= LEVEL_NONE
)
5070 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5073 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5077 spin_unlock(&pers_lock
);
5078 if (mddev
->level
!= pers
->level
) {
5079 mddev
->level
= pers
->level
;
5080 mddev
->new_level
= pers
->level
;
5082 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5084 if (mddev
->reshape_position
!= MaxSector
&&
5085 pers
->start_reshape
== NULL
) {
5086 /* This personality cannot handle reshaping... */
5087 module_put(pers
->owner
);
5091 if (pers
->sync_request
) {
5092 /* Warn if this is a potentially silly
5095 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5096 struct md_rdev
*rdev2
;
5099 rdev_for_each(rdev
, mddev
)
5100 rdev_for_each(rdev2
, mddev
) {
5102 rdev
->bdev
->bd_contains
==
5103 rdev2
->bdev
->bd_contains
) {
5105 "%s: WARNING: %s appears to be"
5106 " on the same physical disk as"
5109 bdevname(rdev
->bdev
,b
),
5110 bdevname(rdev2
->bdev
,b2
));
5117 "True protection against single-disk"
5118 " failure might be compromised.\n");
5121 mddev
->recovery
= 0;
5122 /* may be over-ridden by personality */
5123 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5125 mddev
->ok_start_degraded
= start_dirty_degraded
;
5127 if (start_readonly
&& mddev
->ro
== 0)
5128 mddev
->ro
= 2; /* read-only, but switch on first write */
5130 err
= pers
->run(mddev
);
5132 printk(KERN_ERR
"md: pers->run() failed ...\n");
5133 else if (pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5134 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5135 " but 'external_size' not in effect?\n", __func__
);
5137 "md: invalid array_size %llu > default size %llu\n",
5138 (unsigned long long)mddev
->array_sectors
/ 2,
5139 (unsigned long long)pers
->size(mddev
, 0, 0) / 2);
5142 if (err
== 0 && pers
->sync_request
&&
5143 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5144 struct bitmap
*bitmap
;
5146 bitmap
= bitmap_create(mddev
, -1);
5147 if (IS_ERR(bitmap
)) {
5148 err
= PTR_ERR(bitmap
);
5149 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5150 mdname(mddev
), err
);
5152 mddev
->bitmap
= bitmap
;
5156 mddev_detach(mddev
);
5158 pers
->free(mddev
, mddev
->private);
5159 module_put(pers
->owner
);
5160 bitmap_destroy(mddev
);
5164 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
5165 mddev
->queue
->backing_dev_info
.congested_fn
= md_congested
;
5166 blk_queue_merge_bvec(mddev
->queue
, md_mergeable_bvec
);
5168 if (pers
->sync_request
) {
5169 if (mddev
->kobj
.sd
&&
5170 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5172 "md: cannot register extra attributes for %s\n",
5174 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5175 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5178 atomic_set(&mddev
->writes_pending
,0);
5179 atomic_set(&mddev
->max_corr_read_errors
,
5180 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5181 mddev
->safemode
= 0;
5182 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5183 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5184 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5187 spin_lock(&mddev
->lock
);
5190 spin_unlock(&mddev
->lock
);
5191 rdev_for_each(rdev
, mddev
)
5192 if (rdev
->raid_disk
>= 0)
5193 if (sysfs_link_rdev(mddev
, rdev
))
5194 /* failure here is OK */;
5196 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5198 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
5199 md_update_sb(mddev
, 0);
5201 md_new_event(mddev
);
5202 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5203 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5204 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5207 EXPORT_SYMBOL_GPL(md_run
);
5209 static int do_md_run(struct mddev
*mddev
)
5213 err
= md_run(mddev
);
5216 err
= bitmap_load(mddev
);
5218 bitmap_destroy(mddev
);
5222 md_wakeup_thread(mddev
->thread
);
5223 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5225 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5226 revalidate_disk(mddev
->gendisk
);
5228 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5233 static int restart_array(struct mddev
*mddev
)
5235 struct gendisk
*disk
= mddev
->gendisk
;
5237 /* Complain if it has no devices */
5238 if (list_empty(&mddev
->disks
))
5244 mddev
->safemode
= 0;
5246 set_disk_ro(disk
, 0);
5247 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5249 /* Kick recovery or resync if necessary */
5250 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5251 md_wakeup_thread(mddev
->thread
);
5252 md_wakeup_thread(mddev
->sync_thread
);
5253 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5257 static void md_clean(struct mddev
*mddev
)
5259 mddev
->array_sectors
= 0;
5260 mddev
->external_size
= 0;
5261 mddev
->dev_sectors
= 0;
5262 mddev
->raid_disks
= 0;
5263 mddev
->recovery_cp
= 0;
5264 mddev
->resync_min
= 0;
5265 mddev
->resync_max
= MaxSector
;
5266 mddev
->reshape_position
= MaxSector
;
5267 mddev
->external
= 0;
5268 mddev
->persistent
= 0;
5269 mddev
->level
= LEVEL_NONE
;
5270 mddev
->clevel
[0] = 0;
5273 mddev
->metadata_type
[0] = 0;
5274 mddev
->chunk_sectors
= 0;
5275 mddev
->ctime
= mddev
->utime
= 0;
5277 mddev
->max_disks
= 0;
5279 mddev
->can_decrease_events
= 0;
5280 mddev
->delta_disks
= 0;
5281 mddev
->reshape_backwards
= 0;
5282 mddev
->new_level
= LEVEL_NONE
;
5283 mddev
->new_layout
= 0;
5284 mddev
->new_chunk_sectors
= 0;
5285 mddev
->curr_resync
= 0;
5286 atomic64_set(&mddev
->resync_mismatches
, 0);
5287 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5288 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5289 mddev
->recovery
= 0;
5292 mddev
->degraded
= 0;
5293 mddev
->safemode
= 0;
5294 mddev
->merge_check_needed
= 0;
5295 mddev
->bitmap_info
.offset
= 0;
5296 mddev
->bitmap_info
.default_offset
= 0;
5297 mddev
->bitmap_info
.default_space
= 0;
5298 mddev
->bitmap_info
.chunksize
= 0;
5299 mddev
->bitmap_info
.daemon_sleep
= 0;
5300 mddev
->bitmap_info
.max_write_behind
= 0;
5303 static void __md_stop_writes(struct mddev
*mddev
)
5305 if (mddev_is_clustered(mddev
))
5306 md_cluster_ops
->metadata_update_start(mddev
);
5307 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5308 flush_workqueue(md_misc_wq
);
5309 if (mddev
->sync_thread
) {
5310 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5311 md_reap_sync_thread(mddev
);
5314 del_timer_sync(&mddev
->safemode_timer
);
5316 bitmap_flush(mddev
);
5317 md_super_wait(mddev
);
5319 if (mddev
->ro
== 0 &&
5320 (!mddev
->in_sync
|| (mddev
->flags
& MD_UPDATE_SB_FLAGS
))) {
5321 /* mark array as shutdown cleanly */
5323 md_update_sb(mddev
, 1);
5325 if (mddev_is_clustered(mddev
))
5326 md_cluster_ops
->metadata_update_finish(mddev
);
5329 void md_stop_writes(struct mddev
*mddev
)
5331 mddev_lock_nointr(mddev
);
5332 __md_stop_writes(mddev
);
5333 mddev_unlock(mddev
);
5335 EXPORT_SYMBOL_GPL(md_stop_writes
);
5337 static void mddev_detach(struct mddev
*mddev
)
5339 struct bitmap
*bitmap
= mddev
->bitmap
;
5340 /* wait for behind writes to complete */
5341 if (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
5342 printk(KERN_INFO
"md:%s: behind writes in progress - waiting to stop.\n",
5344 /* need to kick something here to make sure I/O goes? */
5345 wait_event(bitmap
->behind_wait
,
5346 atomic_read(&bitmap
->behind_writes
) == 0);
5348 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5349 mddev
->pers
->quiesce(mddev
, 1);
5350 mddev
->pers
->quiesce(mddev
, 0);
5352 md_unregister_thread(&mddev
->thread
);
5354 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
5357 static void __md_stop(struct mddev
*mddev
)
5359 struct md_personality
*pers
= mddev
->pers
;
5360 mddev_detach(mddev
);
5361 spin_lock(&mddev
->lock
);
5364 spin_unlock(&mddev
->lock
);
5365 pers
->free(mddev
, mddev
->private);
5366 if (pers
->sync_request
&& mddev
->to_remove
== NULL
)
5367 mddev
->to_remove
= &md_redundancy_group
;
5368 module_put(pers
->owner
);
5369 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5372 void md_stop(struct mddev
*mddev
)
5374 /* stop the array and free an attached data structures.
5375 * This is called from dm-raid
5378 bitmap_destroy(mddev
);
5380 bioset_free(mddev
->bio_set
);
5383 EXPORT_SYMBOL_GPL(md_stop
);
5385 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5390 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5392 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5393 md_wakeup_thread(mddev
->thread
);
5395 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5396 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5397 if (mddev
->sync_thread
)
5398 /* Thread might be blocked waiting for metadata update
5399 * which will now never happen */
5400 wake_up_process(mddev
->sync_thread
->tsk
);
5402 mddev_unlock(mddev
);
5403 wait_event(resync_wait
, !test_bit(MD_RECOVERY_RUNNING
,
5405 mddev_lock_nointr(mddev
);
5407 mutex_lock(&mddev
->open_mutex
);
5408 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5409 mddev
->sync_thread
||
5410 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5411 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5412 printk("md: %s still in use.\n",mdname(mddev
));
5414 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5415 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5416 md_wakeup_thread(mddev
->thread
);
5422 __md_stop_writes(mddev
);
5428 set_disk_ro(mddev
->gendisk
, 1);
5429 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5430 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5431 md_wakeup_thread(mddev
->thread
);
5432 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5436 mutex_unlock(&mddev
->open_mutex
);
5441 * 0 - completely stop and dis-assemble array
5442 * 2 - stop but do not disassemble array
5444 static int do_md_stop(struct mddev
*mddev
, int mode
,
5445 struct block_device
*bdev
)
5447 struct gendisk
*disk
= mddev
->gendisk
;
5448 struct md_rdev
*rdev
;
5451 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5453 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5454 md_wakeup_thread(mddev
->thread
);
5456 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5457 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5458 if (mddev
->sync_thread
)
5459 /* Thread might be blocked waiting for metadata update
5460 * which will now never happen */
5461 wake_up_process(mddev
->sync_thread
->tsk
);
5463 mddev_unlock(mddev
);
5464 wait_event(resync_wait
, (mddev
->sync_thread
== NULL
&&
5465 !test_bit(MD_RECOVERY_RUNNING
,
5466 &mddev
->recovery
)));
5467 mddev_lock_nointr(mddev
);
5469 mutex_lock(&mddev
->open_mutex
);
5470 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5471 mddev
->sysfs_active
||
5472 mddev
->sync_thread
||
5473 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5474 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5475 printk("md: %s still in use.\n",mdname(mddev
));
5476 mutex_unlock(&mddev
->open_mutex
);
5478 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5479 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5480 md_wakeup_thread(mddev
->thread
);
5486 set_disk_ro(disk
, 0);
5488 __md_stop_writes(mddev
);
5490 mddev
->queue
->merge_bvec_fn
= NULL
;
5491 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5493 /* tell userspace to handle 'inactive' */
5494 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5496 rdev_for_each(rdev
, mddev
)
5497 if (rdev
->raid_disk
>= 0)
5498 sysfs_unlink_rdev(mddev
, rdev
);
5500 set_capacity(disk
, 0);
5501 mutex_unlock(&mddev
->open_mutex
);
5503 revalidate_disk(disk
);
5508 mutex_unlock(&mddev
->open_mutex
);
5510 * Free resources if final stop
5513 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5515 bitmap_destroy(mddev
);
5516 if (mddev
->bitmap_info
.file
) {
5517 struct file
*f
= mddev
->bitmap_info
.file
;
5518 spin_lock(&mddev
->lock
);
5519 mddev
->bitmap_info
.file
= NULL
;
5520 spin_unlock(&mddev
->lock
);
5523 mddev
->bitmap_info
.offset
= 0;
5525 export_array(mddev
);
5528 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5529 if (mddev
->hold_active
== UNTIL_STOP
)
5530 mddev
->hold_active
= 0;
5532 blk_integrity_unregister(disk
);
5533 md_new_event(mddev
);
5534 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5539 static void autorun_array(struct mddev
*mddev
)
5541 struct md_rdev
*rdev
;
5544 if (list_empty(&mddev
->disks
))
5547 printk(KERN_INFO
"md: running: ");
5549 rdev_for_each(rdev
, mddev
) {
5550 char b
[BDEVNAME_SIZE
];
5551 printk("<%s>", bdevname(rdev
->bdev
,b
));
5555 err
= do_md_run(mddev
);
5557 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5558 do_md_stop(mddev
, 0, NULL
);
5563 * lets try to run arrays based on all disks that have arrived
5564 * until now. (those are in pending_raid_disks)
5566 * the method: pick the first pending disk, collect all disks with
5567 * the same UUID, remove all from the pending list and put them into
5568 * the 'same_array' list. Then order this list based on superblock
5569 * update time (freshest comes first), kick out 'old' disks and
5570 * compare superblocks. If everything's fine then run it.
5572 * If "unit" is allocated, then bump its reference count
5574 static void autorun_devices(int part
)
5576 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5577 struct mddev
*mddev
;
5578 char b
[BDEVNAME_SIZE
];
5580 printk(KERN_INFO
"md: autorun ...\n");
5581 while (!list_empty(&pending_raid_disks
)) {
5584 LIST_HEAD(candidates
);
5585 rdev0
= list_entry(pending_raid_disks
.next
,
5586 struct md_rdev
, same_set
);
5588 printk(KERN_INFO
"md: considering %s ...\n",
5589 bdevname(rdev0
->bdev
,b
));
5590 INIT_LIST_HEAD(&candidates
);
5591 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5592 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5593 printk(KERN_INFO
"md: adding %s ...\n",
5594 bdevname(rdev
->bdev
,b
));
5595 list_move(&rdev
->same_set
, &candidates
);
5598 * now we have a set of devices, with all of them having
5599 * mostly sane superblocks. It's time to allocate the
5603 dev
= MKDEV(mdp_major
,
5604 rdev0
->preferred_minor
<< MdpMinorShift
);
5605 unit
= MINOR(dev
) >> MdpMinorShift
;
5607 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5610 if (rdev0
->preferred_minor
!= unit
) {
5611 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5612 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5616 md_probe(dev
, NULL
, NULL
);
5617 mddev
= mddev_find(dev
);
5618 if (!mddev
|| !mddev
->gendisk
) {
5622 "md: cannot allocate memory for md drive.\n");
5625 if (mddev_lock(mddev
))
5626 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5628 else if (mddev
->raid_disks
|| mddev
->major_version
5629 || !list_empty(&mddev
->disks
)) {
5631 "md: %s already running, cannot run %s\n",
5632 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5633 mddev_unlock(mddev
);
5635 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5636 mddev
->persistent
= 1;
5637 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5638 list_del_init(&rdev
->same_set
);
5639 if (bind_rdev_to_array(rdev
, mddev
))
5642 autorun_array(mddev
);
5643 mddev_unlock(mddev
);
5645 /* on success, candidates will be empty, on error
5648 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5649 list_del_init(&rdev
->same_set
);
5654 printk(KERN_INFO
"md: ... autorun DONE.\n");
5656 #endif /* !MODULE */
5658 static int get_version(void __user
*arg
)
5662 ver
.major
= MD_MAJOR_VERSION
;
5663 ver
.minor
= MD_MINOR_VERSION
;
5664 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5666 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5672 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
5674 mdu_array_info_t info
;
5675 int nr
,working
,insync
,failed
,spare
;
5676 struct md_rdev
*rdev
;
5678 nr
= working
= insync
= failed
= spare
= 0;
5680 rdev_for_each_rcu(rdev
, mddev
) {
5682 if (test_bit(Faulty
, &rdev
->flags
))
5686 if (test_bit(In_sync
, &rdev
->flags
))
5694 info
.major_version
= mddev
->major_version
;
5695 info
.minor_version
= mddev
->minor_version
;
5696 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5697 info
.ctime
= mddev
->ctime
;
5698 info
.level
= mddev
->level
;
5699 info
.size
= mddev
->dev_sectors
/ 2;
5700 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5703 info
.raid_disks
= mddev
->raid_disks
;
5704 info
.md_minor
= mddev
->md_minor
;
5705 info
.not_persistent
= !mddev
->persistent
;
5707 info
.utime
= mddev
->utime
;
5710 info
.state
= (1<<MD_SB_CLEAN
);
5711 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5712 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
5713 if (mddev_is_clustered(mddev
))
5714 info
.state
|= (1<<MD_SB_CLUSTERED
);
5715 info
.active_disks
= insync
;
5716 info
.working_disks
= working
;
5717 info
.failed_disks
= failed
;
5718 info
.spare_disks
= spare
;
5720 info
.layout
= mddev
->layout
;
5721 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5723 if (copy_to_user(arg
, &info
, sizeof(info
)))
5729 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
5731 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5735 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5740 spin_lock(&mddev
->lock
);
5741 /* bitmap disabled, zero the first byte and copy out */
5742 if (!mddev
->bitmap_info
.file
)
5743 file
->pathname
[0] = '\0';
5744 else if ((ptr
= d_path(&mddev
->bitmap_info
.file
->f_path
,
5745 file
->pathname
, sizeof(file
->pathname
))),
5749 memmove(file
->pathname
, ptr
,
5750 sizeof(file
->pathname
)-(ptr
-file
->pathname
));
5751 spin_unlock(&mddev
->lock
);
5754 copy_to_user(arg
, file
, sizeof(*file
)))
5761 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
5763 mdu_disk_info_t info
;
5764 struct md_rdev
*rdev
;
5766 if (copy_from_user(&info
, arg
, sizeof(info
)))
5770 rdev
= md_find_rdev_nr_rcu(mddev
, info
.number
);
5772 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5773 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5774 info
.raid_disk
= rdev
->raid_disk
;
5776 if (test_bit(Faulty
, &rdev
->flags
))
5777 info
.state
|= (1<<MD_DISK_FAULTY
);
5778 else if (test_bit(In_sync
, &rdev
->flags
)) {
5779 info
.state
|= (1<<MD_DISK_ACTIVE
);
5780 info
.state
|= (1<<MD_DISK_SYNC
);
5782 if (test_bit(WriteMostly
, &rdev
->flags
))
5783 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5785 info
.major
= info
.minor
= 0;
5786 info
.raid_disk
= -1;
5787 info
.state
= (1<<MD_DISK_REMOVED
);
5791 if (copy_to_user(arg
, &info
, sizeof(info
)))
5797 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
5799 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5800 struct md_rdev
*rdev
;
5801 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5803 if (mddev_is_clustered(mddev
) &&
5804 !(info
->state
& ((1 << MD_DISK_CLUSTER_ADD
) | (1 << MD_DISK_CANDIDATE
)))) {
5805 pr_err("%s: Cannot add to clustered mddev.\n",
5810 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5813 if (!mddev
->raid_disks
) {
5815 /* expecting a device which has a superblock */
5816 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5819 "md: md_import_device returned %ld\n",
5821 return PTR_ERR(rdev
);
5823 if (!list_empty(&mddev
->disks
)) {
5824 struct md_rdev
*rdev0
5825 = list_entry(mddev
->disks
.next
,
5826 struct md_rdev
, same_set
);
5827 err
= super_types
[mddev
->major_version
]
5828 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5831 "md: %s has different UUID to %s\n",
5832 bdevname(rdev
->bdev
,b
),
5833 bdevname(rdev0
->bdev
,b2
));
5838 err
= bind_rdev_to_array(rdev
, mddev
);
5845 * add_new_disk can be used once the array is assembled
5846 * to add "hot spares". They must already have a superblock
5851 if (!mddev
->pers
->hot_add_disk
) {
5853 "%s: personality does not support diskops!\n",
5857 if (mddev
->persistent
)
5858 rdev
= md_import_device(dev
, mddev
->major_version
,
5859 mddev
->minor_version
);
5861 rdev
= md_import_device(dev
, -1, -1);
5864 "md: md_import_device returned %ld\n",
5866 return PTR_ERR(rdev
);
5868 /* set saved_raid_disk if appropriate */
5869 if (!mddev
->persistent
) {
5870 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5871 info
->raid_disk
< mddev
->raid_disks
) {
5872 rdev
->raid_disk
= info
->raid_disk
;
5873 set_bit(In_sync
, &rdev
->flags
);
5874 clear_bit(Bitmap_sync
, &rdev
->flags
);
5876 rdev
->raid_disk
= -1;
5877 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5879 super_types
[mddev
->major_version
].
5880 validate_super(mddev
, rdev
);
5881 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5882 rdev
->raid_disk
!= info
->raid_disk
) {
5883 /* This was a hot-add request, but events doesn't
5884 * match, so reject it.
5890 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5891 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5892 set_bit(WriteMostly
, &rdev
->flags
);
5894 clear_bit(WriteMostly
, &rdev
->flags
);
5897 * check whether the device shows up in other nodes
5899 if (mddev_is_clustered(mddev
)) {
5900 if (info
->state
& (1 << MD_DISK_CANDIDATE
)) {
5901 /* Through --cluster-confirm */
5902 set_bit(Candidate
, &rdev
->flags
);
5903 err
= md_cluster_ops
->new_disk_ack(mddev
, true);
5908 } else if (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)) {
5909 /* --add initiated by this node */
5910 err
= md_cluster_ops
->add_new_disk_start(mddev
, rdev
);
5912 md_cluster_ops
->add_new_disk_finish(mddev
);
5919 rdev
->raid_disk
= -1;
5920 err
= bind_rdev_to_array(rdev
, mddev
);
5924 err
= add_bound_rdev(rdev
);
5925 if (mddev_is_clustered(mddev
) &&
5926 (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)))
5927 md_cluster_ops
->add_new_disk_finish(mddev
);
5931 /* otherwise, add_new_disk is only allowed
5932 * for major_version==0 superblocks
5934 if (mddev
->major_version
!= 0) {
5935 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5940 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5942 rdev
= md_import_device(dev
, -1, 0);
5945 "md: error, md_import_device() returned %ld\n",
5947 return PTR_ERR(rdev
);
5949 rdev
->desc_nr
= info
->number
;
5950 if (info
->raid_disk
< mddev
->raid_disks
)
5951 rdev
->raid_disk
= info
->raid_disk
;
5953 rdev
->raid_disk
= -1;
5955 if (rdev
->raid_disk
< mddev
->raid_disks
)
5956 if (info
->state
& (1<<MD_DISK_SYNC
))
5957 set_bit(In_sync
, &rdev
->flags
);
5959 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5960 set_bit(WriteMostly
, &rdev
->flags
);
5962 if (!mddev
->persistent
) {
5963 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5964 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5966 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5967 rdev
->sectors
= rdev
->sb_start
;
5969 err
= bind_rdev_to_array(rdev
, mddev
);
5979 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
5981 char b
[BDEVNAME_SIZE
];
5982 struct md_rdev
*rdev
;
5984 rdev
= find_rdev(mddev
, dev
);
5988 if (mddev_is_clustered(mddev
))
5989 md_cluster_ops
->metadata_update_start(mddev
);
5991 clear_bit(Blocked
, &rdev
->flags
);
5992 remove_and_add_spares(mddev
, rdev
);
5994 if (rdev
->raid_disk
>= 0)
5997 if (mddev_is_clustered(mddev
))
5998 md_cluster_ops
->remove_disk(mddev
, rdev
);
6000 md_kick_rdev_from_array(rdev
);
6001 md_update_sb(mddev
, 1);
6002 md_new_event(mddev
);
6004 if (mddev_is_clustered(mddev
))
6005 md_cluster_ops
->metadata_update_finish(mddev
);
6009 if (mddev_is_clustered(mddev
))
6010 md_cluster_ops
->metadata_update_cancel(mddev
);
6011 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
6012 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6016 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
6018 char b
[BDEVNAME_SIZE
];
6020 struct md_rdev
*rdev
;
6025 if (mddev
->major_version
!= 0) {
6026 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
6027 " version-0 superblocks.\n",
6031 if (!mddev
->pers
->hot_add_disk
) {
6033 "%s: personality does not support diskops!\n",
6038 rdev
= md_import_device(dev
, -1, 0);
6041 "md: error, md_import_device() returned %ld\n",
6046 if (mddev
->persistent
)
6047 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6049 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6051 rdev
->sectors
= rdev
->sb_start
;
6053 if (test_bit(Faulty
, &rdev
->flags
)) {
6055 "md: can not hot-add faulty %s disk to %s!\n",
6056 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6061 if (mddev_is_clustered(mddev
))
6062 md_cluster_ops
->metadata_update_start(mddev
);
6063 clear_bit(In_sync
, &rdev
->flags
);
6065 rdev
->saved_raid_disk
= -1;
6066 err
= bind_rdev_to_array(rdev
, mddev
);
6068 goto abort_clustered
;
6071 * The rest should better be atomic, we can have disk failures
6072 * noticed in interrupt contexts ...
6075 rdev
->raid_disk
= -1;
6077 md_update_sb(mddev
, 1);
6079 if (mddev_is_clustered(mddev
))
6080 md_cluster_ops
->metadata_update_finish(mddev
);
6082 * Kick recovery, maybe this spare has to be added to the
6083 * array immediately.
6085 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6086 md_wakeup_thread(mddev
->thread
);
6087 md_new_event(mddev
);
6091 if (mddev_is_clustered(mddev
))
6092 md_cluster_ops
->metadata_update_cancel(mddev
);
6098 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
6103 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
6105 if (mddev
->recovery
|| mddev
->sync_thread
)
6107 /* we should be able to change the bitmap.. */
6111 struct inode
*inode
;
6114 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
)
6115 return -EEXIST
; /* cannot add when bitmap is present */
6119 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
6124 inode
= f
->f_mapping
->host
;
6125 if (!S_ISREG(inode
->i_mode
)) {
6126 printk(KERN_ERR
"%s: error: bitmap file must be a regular file\n",
6129 } else if (!(f
->f_mode
& FMODE_WRITE
)) {
6130 printk(KERN_ERR
"%s: error: bitmap file must open for write\n",
6133 } else if (atomic_read(&inode
->i_writecount
) != 1) {
6134 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
6142 mddev
->bitmap_info
.file
= f
;
6143 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6144 } else if (mddev
->bitmap
== NULL
)
6145 return -ENOENT
; /* cannot remove what isn't there */
6148 mddev
->pers
->quiesce(mddev
, 1);
6150 struct bitmap
*bitmap
;
6152 bitmap
= bitmap_create(mddev
, -1);
6153 if (!IS_ERR(bitmap
)) {
6154 mddev
->bitmap
= bitmap
;
6155 err
= bitmap_load(mddev
);
6157 err
= PTR_ERR(bitmap
);
6159 if (fd
< 0 || err
) {
6160 bitmap_destroy(mddev
);
6161 fd
= -1; /* make sure to put the file */
6163 mddev
->pers
->quiesce(mddev
, 0);
6166 struct file
*f
= mddev
->bitmap_info
.file
;
6168 spin_lock(&mddev
->lock
);
6169 mddev
->bitmap_info
.file
= NULL
;
6170 spin_unlock(&mddev
->lock
);
6179 * set_array_info is used two different ways
6180 * The original usage is when creating a new array.
6181 * In this usage, raid_disks is > 0 and it together with
6182 * level, size, not_persistent,layout,chunksize determine the
6183 * shape of the array.
6184 * This will always create an array with a type-0.90.0 superblock.
6185 * The newer usage is when assembling an array.
6186 * In this case raid_disks will be 0, and the major_version field is
6187 * use to determine which style super-blocks are to be found on the devices.
6188 * The minor and patch _version numbers are also kept incase the
6189 * super_block handler wishes to interpret them.
6191 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6194 if (info
->raid_disks
== 0) {
6195 /* just setting version number for superblock loading */
6196 if (info
->major_version
< 0 ||
6197 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6198 super_types
[info
->major_version
].name
== NULL
) {
6199 /* maybe try to auto-load a module? */
6201 "md: superblock version %d not known\n",
6202 info
->major_version
);
6205 mddev
->major_version
= info
->major_version
;
6206 mddev
->minor_version
= info
->minor_version
;
6207 mddev
->patch_version
= info
->patch_version
;
6208 mddev
->persistent
= !info
->not_persistent
;
6209 /* ensure mddev_put doesn't delete this now that there
6210 * is some minimal configuration.
6212 mddev
->ctime
= get_seconds();
6215 mddev
->major_version
= MD_MAJOR_VERSION
;
6216 mddev
->minor_version
= MD_MINOR_VERSION
;
6217 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6218 mddev
->ctime
= get_seconds();
6220 mddev
->level
= info
->level
;
6221 mddev
->clevel
[0] = 0;
6222 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6223 mddev
->raid_disks
= info
->raid_disks
;
6224 /* don't set md_minor, it is determined by which /dev/md* was
6227 if (info
->state
& (1<<MD_SB_CLEAN
))
6228 mddev
->recovery_cp
= MaxSector
;
6230 mddev
->recovery_cp
= 0;
6231 mddev
->persistent
= ! info
->not_persistent
;
6232 mddev
->external
= 0;
6234 mddev
->layout
= info
->layout
;
6235 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6237 mddev
->max_disks
= MD_SB_DISKS
;
6239 if (mddev
->persistent
)
6241 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6243 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6244 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6245 mddev
->bitmap_info
.offset
= 0;
6247 mddev
->reshape_position
= MaxSector
;
6250 * Generate a 128 bit UUID
6252 get_random_bytes(mddev
->uuid
, 16);
6254 mddev
->new_level
= mddev
->level
;
6255 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6256 mddev
->new_layout
= mddev
->layout
;
6257 mddev
->delta_disks
= 0;
6258 mddev
->reshape_backwards
= 0;
6263 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6265 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6267 if (mddev
->external_size
)
6270 mddev
->array_sectors
= array_sectors
;
6272 EXPORT_SYMBOL(md_set_array_sectors
);
6274 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6276 struct md_rdev
*rdev
;
6278 int fit
= (num_sectors
== 0);
6280 if (mddev
->pers
->resize
== NULL
)
6282 /* The "num_sectors" is the number of sectors of each device that
6283 * is used. This can only make sense for arrays with redundancy.
6284 * linear and raid0 always use whatever space is available. We can only
6285 * consider changing this number if no resync or reconstruction is
6286 * happening, and if the new size is acceptable. It must fit before the
6287 * sb_start or, if that is <data_offset, it must fit before the size
6288 * of each device. If num_sectors is zero, we find the largest size
6291 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6297 rdev_for_each(rdev
, mddev
) {
6298 sector_t avail
= rdev
->sectors
;
6300 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6301 num_sectors
= avail
;
6302 if (avail
< num_sectors
)
6305 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6307 revalidate_disk(mddev
->gendisk
);
6311 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6314 struct md_rdev
*rdev
;
6315 /* change the number of raid disks */
6316 if (mddev
->pers
->check_reshape
== NULL
)
6320 if (raid_disks
<= 0 ||
6321 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6323 if (mddev
->sync_thread
||
6324 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6325 mddev
->reshape_position
!= MaxSector
)
6328 rdev_for_each(rdev
, mddev
) {
6329 if (mddev
->raid_disks
< raid_disks
&&
6330 rdev
->data_offset
< rdev
->new_data_offset
)
6332 if (mddev
->raid_disks
> raid_disks
&&
6333 rdev
->data_offset
> rdev
->new_data_offset
)
6337 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6338 if (mddev
->delta_disks
< 0)
6339 mddev
->reshape_backwards
= 1;
6340 else if (mddev
->delta_disks
> 0)
6341 mddev
->reshape_backwards
= 0;
6343 rv
= mddev
->pers
->check_reshape(mddev
);
6345 mddev
->delta_disks
= 0;
6346 mddev
->reshape_backwards
= 0;
6352 * update_array_info is used to change the configuration of an
6354 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6355 * fields in the info are checked against the array.
6356 * Any differences that cannot be handled will cause an error.
6357 * Normally, only one change can be managed at a time.
6359 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6365 /* calculate expected state,ignoring low bits */
6366 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6367 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6369 if (mddev
->major_version
!= info
->major_version
||
6370 mddev
->minor_version
!= info
->minor_version
||
6371 /* mddev->patch_version != info->patch_version || */
6372 mddev
->ctime
!= info
->ctime
||
6373 mddev
->level
!= info
->level
||
6374 /* mddev->layout != info->layout || */
6375 !mddev
->persistent
!= info
->not_persistent
||
6376 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6377 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6378 ((state
^info
->state
) & 0xfffffe00)
6381 /* Check there is only one change */
6382 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6384 if (mddev
->raid_disks
!= info
->raid_disks
)
6386 if (mddev
->layout
!= info
->layout
)
6388 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6395 if (mddev
->layout
!= info
->layout
) {
6397 * we don't need to do anything at the md level, the
6398 * personality will take care of it all.
6400 if (mddev
->pers
->check_reshape
== NULL
)
6403 mddev
->new_layout
= info
->layout
;
6404 rv
= mddev
->pers
->check_reshape(mddev
);
6406 mddev
->new_layout
= mddev
->layout
;
6410 if (mddev_is_clustered(mddev
))
6411 md_cluster_ops
->metadata_update_start(mddev
);
6412 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6413 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6415 if (mddev
->raid_disks
!= info
->raid_disks
)
6416 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6418 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6419 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
) {
6423 if (mddev
->recovery
|| mddev
->sync_thread
) {
6427 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6428 struct bitmap
*bitmap
;
6429 /* add the bitmap */
6430 if (mddev
->bitmap
) {
6434 if (mddev
->bitmap_info
.default_offset
== 0) {
6438 mddev
->bitmap_info
.offset
=
6439 mddev
->bitmap_info
.default_offset
;
6440 mddev
->bitmap_info
.space
=
6441 mddev
->bitmap_info
.default_space
;
6442 mddev
->pers
->quiesce(mddev
, 1);
6443 bitmap
= bitmap_create(mddev
, -1);
6444 if (!IS_ERR(bitmap
)) {
6445 mddev
->bitmap
= bitmap
;
6446 rv
= bitmap_load(mddev
);
6448 rv
= PTR_ERR(bitmap
);
6450 bitmap_destroy(mddev
);
6451 mddev
->pers
->quiesce(mddev
, 0);
6453 /* remove the bitmap */
6454 if (!mddev
->bitmap
) {
6458 if (mddev
->bitmap
->storage
.file
) {
6462 mddev
->pers
->quiesce(mddev
, 1);
6463 bitmap_destroy(mddev
);
6464 mddev
->pers
->quiesce(mddev
, 0);
6465 mddev
->bitmap_info
.offset
= 0;
6468 md_update_sb(mddev
, 1);
6469 if (mddev_is_clustered(mddev
))
6470 md_cluster_ops
->metadata_update_finish(mddev
);
6473 if (mddev_is_clustered(mddev
))
6474 md_cluster_ops
->metadata_update_cancel(mddev
);
6478 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6480 struct md_rdev
*rdev
;
6483 if (mddev
->pers
== NULL
)
6487 rdev
= find_rdev_rcu(mddev
, dev
);
6491 md_error(mddev
, rdev
);
6492 if (!test_bit(Faulty
, &rdev
->flags
))
6500 * We have a problem here : there is no easy way to give a CHS
6501 * virtual geometry. We currently pretend that we have a 2 heads
6502 * 4 sectors (with a BIG number of cylinders...). This drives
6503 * dosfs just mad... ;-)
6505 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6507 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6511 geo
->cylinders
= mddev
->array_sectors
/ 8;
6515 static inline bool md_ioctl_valid(unsigned int cmd
)
6520 case GET_ARRAY_INFO
:
6521 case GET_BITMAP_FILE
:
6524 case HOT_REMOVE_DISK
:
6527 case RESTART_ARRAY_RW
:
6529 case SET_ARRAY_INFO
:
6530 case SET_BITMAP_FILE
:
6531 case SET_DISK_FAULTY
:
6534 case CLUSTERED_DISK_NACK
:
6541 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6542 unsigned int cmd
, unsigned long arg
)
6545 void __user
*argp
= (void __user
*)arg
;
6546 struct mddev
*mddev
= NULL
;
6549 if (!md_ioctl_valid(cmd
))
6554 case GET_ARRAY_INFO
:
6558 if (!capable(CAP_SYS_ADMIN
))
6563 * Commands dealing with the RAID driver but not any
6568 err
= get_version(argp
);
6574 autostart_arrays(arg
);
6581 * Commands creating/starting a new array:
6584 mddev
= bdev
->bd_disk
->private_data
;
6591 /* Some actions do not requires the mutex */
6593 case GET_ARRAY_INFO
:
6594 if (!mddev
->raid_disks
&& !mddev
->external
)
6597 err
= get_array_info(mddev
, argp
);
6601 if (!mddev
->raid_disks
&& !mddev
->external
)
6604 err
= get_disk_info(mddev
, argp
);
6607 case SET_DISK_FAULTY
:
6608 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6611 case GET_BITMAP_FILE
:
6612 err
= get_bitmap_file(mddev
, argp
);
6617 if (cmd
== ADD_NEW_DISK
)
6618 /* need to ensure md_delayed_delete() has completed */
6619 flush_workqueue(md_misc_wq
);
6621 if (cmd
== HOT_REMOVE_DISK
)
6622 /* need to ensure recovery thread has run */
6623 wait_event_interruptible_timeout(mddev
->sb_wait
,
6624 !test_bit(MD_RECOVERY_NEEDED
,
6626 msecs_to_jiffies(5000));
6627 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
6628 /* Need to flush page cache, and ensure no-one else opens
6631 mutex_lock(&mddev
->open_mutex
);
6632 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
6633 mutex_unlock(&mddev
->open_mutex
);
6637 set_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6638 mutex_unlock(&mddev
->open_mutex
);
6639 sync_blockdev(bdev
);
6641 err
= mddev_lock(mddev
);
6644 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6649 if (cmd
== SET_ARRAY_INFO
) {
6650 mdu_array_info_t info
;
6652 memset(&info
, 0, sizeof(info
));
6653 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6658 err
= update_array_info(mddev
, &info
);
6660 printk(KERN_WARNING
"md: couldn't update"
6661 " array info. %d\n", err
);
6666 if (!list_empty(&mddev
->disks
)) {
6668 "md: array %s already has disks!\n",
6673 if (mddev
->raid_disks
) {
6675 "md: array %s already initialised!\n",
6680 err
= set_array_info(mddev
, &info
);
6682 printk(KERN_WARNING
"md: couldn't set"
6683 " array info. %d\n", err
);
6690 * Commands querying/configuring an existing array:
6692 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6693 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6694 if ((!mddev
->raid_disks
&& !mddev
->external
)
6695 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6696 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6697 && cmd
!= GET_BITMAP_FILE
) {
6703 * Commands even a read-only array can execute:
6706 case RESTART_ARRAY_RW
:
6707 err
= restart_array(mddev
);
6711 err
= do_md_stop(mddev
, 0, bdev
);
6715 err
= md_set_readonly(mddev
, bdev
);
6718 case HOT_REMOVE_DISK
:
6719 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6723 /* We can support ADD_NEW_DISK on read-only arrays
6724 * on if we are re-adding a preexisting device.
6725 * So require mddev->pers and MD_DISK_SYNC.
6728 mdu_disk_info_t info
;
6729 if (copy_from_user(&info
, argp
, sizeof(info
)))
6731 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6732 /* Need to clear read-only for this */
6735 err
= add_new_disk(mddev
, &info
);
6741 if (get_user(ro
, (int __user
*)(arg
))) {
6747 /* if the bdev is going readonly the value of mddev->ro
6748 * does not matter, no writes are coming
6753 /* are we are already prepared for writes? */
6757 /* transitioning to readauto need only happen for
6758 * arrays that call md_write_start
6761 err
= restart_array(mddev
);
6764 set_disk_ro(mddev
->gendisk
, 0);
6771 * The remaining ioctls are changing the state of the
6772 * superblock, so we do not allow them on read-only arrays.
6774 if (mddev
->ro
&& mddev
->pers
) {
6775 if (mddev
->ro
== 2) {
6777 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6778 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6779 /* mddev_unlock will wake thread */
6780 /* If a device failed while we were read-only, we
6781 * need to make sure the metadata is updated now.
6783 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6784 mddev_unlock(mddev
);
6785 wait_event(mddev
->sb_wait
,
6786 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6787 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6788 mddev_lock_nointr(mddev
);
6799 mdu_disk_info_t info
;
6800 if (copy_from_user(&info
, argp
, sizeof(info
)))
6803 err
= add_new_disk(mddev
, &info
);
6807 case CLUSTERED_DISK_NACK
:
6808 if (mddev_is_clustered(mddev
))
6809 md_cluster_ops
->new_disk_ack(mddev
, false);
6815 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6819 err
= do_md_run(mddev
);
6822 case SET_BITMAP_FILE
:
6823 err
= set_bitmap_file(mddev
, (int)arg
);
6832 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6834 mddev
->hold_active
= 0;
6835 mddev_unlock(mddev
);
6839 #ifdef CONFIG_COMPAT
6840 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6841 unsigned int cmd
, unsigned long arg
)
6844 case HOT_REMOVE_DISK
:
6846 case SET_DISK_FAULTY
:
6847 case SET_BITMAP_FILE
:
6848 /* These take in integer arg, do not convert */
6851 arg
= (unsigned long)compat_ptr(arg
);
6855 return md_ioctl(bdev
, mode
, cmd
, arg
);
6857 #endif /* CONFIG_COMPAT */
6859 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6862 * Succeed if we can lock the mddev, which confirms that
6863 * it isn't being stopped right now.
6865 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6871 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6872 /* we are racing with mddev_put which is discarding this
6876 /* Wait until bdev->bd_disk is definitely gone */
6877 flush_workqueue(md_misc_wq
);
6878 /* Then retry the open from the top */
6879 return -ERESTARTSYS
;
6881 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6883 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6887 atomic_inc(&mddev
->openers
);
6888 clear_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6889 mutex_unlock(&mddev
->open_mutex
);
6891 check_disk_change(bdev
);
6896 static void md_release(struct gendisk
*disk
, fmode_t mode
)
6898 struct mddev
*mddev
= disk
->private_data
;
6901 atomic_dec(&mddev
->openers
);
6905 static int md_media_changed(struct gendisk
*disk
)
6907 struct mddev
*mddev
= disk
->private_data
;
6909 return mddev
->changed
;
6912 static int md_revalidate(struct gendisk
*disk
)
6914 struct mddev
*mddev
= disk
->private_data
;
6919 static const struct block_device_operations md_fops
=
6921 .owner
= THIS_MODULE
,
6923 .release
= md_release
,
6925 #ifdef CONFIG_COMPAT
6926 .compat_ioctl
= md_compat_ioctl
,
6928 .getgeo
= md_getgeo
,
6929 .media_changed
= md_media_changed
,
6930 .revalidate_disk
= md_revalidate
,
6933 static int md_thread(void *arg
)
6935 struct md_thread
*thread
= arg
;
6938 * md_thread is a 'system-thread', it's priority should be very
6939 * high. We avoid resource deadlocks individually in each
6940 * raid personality. (RAID5 does preallocation) We also use RR and
6941 * the very same RT priority as kswapd, thus we will never get
6942 * into a priority inversion deadlock.
6944 * we definitely have to have equal or higher priority than
6945 * bdflush, otherwise bdflush will deadlock if there are too
6946 * many dirty RAID5 blocks.
6949 allow_signal(SIGKILL
);
6950 while (!kthread_should_stop()) {
6952 /* We need to wait INTERRUPTIBLE so that
6953 * we don't add to the load-average.
6954 * That means we need to be sure no signals are
6957 if (signal_pending(current
))
6958 flush_signals(current
);
6960 wait_event_interruptible_timeout
6962 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6963 || kthread_should_stop(),
6966 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6967 if (!kthread_should_stop())
6968 thread
->run(thread
);
6974 void md_wakeup_thread(struct md_thread
*thread
)
6977 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6978 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6979 wake_up(&thread
->wqueue
);
6982 EXPORT_SYMBOL(md_wakeup_thread
);
6984 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
6985 struct mddev
*mddev
, const char *name
)
6987 struct md_thread
*thread
;
6989 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6993 init_waitqueue_head(&thread
->wqueue
);
6996 thread
->mddev
= mddev
;
6997 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6998 thread
->tsk
= kthread_run(md_thread
, thread
,
7000 mdname(thread
->mddev
),
7002 if (IS_ERR(thread
->tsk
)) {
7008 EXPORT_SYMBOL(md_register_thread
);
7010 void md_unregister_thread(struct md_thread
**threadp
)
7012 struct md_thread
*thread
= *threadp
;
7015 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
7016 /* Locking ensures that mddev_unlock does not wake_up a
7017 * non-existent thread
7019 spin_lock(&pers_lock
);
7021 spin_unlock(&pers_lock
);
7023 kthread_stop(thread
->tsk
);
7026 EXPORT_SYMBOL(md_unregister_thread
);
7028 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
7030 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
7033 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
7035 mddev
->pers
->error_handler(mddev
,rdev
);
7036 if (mddev
->degraded
)
7037 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7038 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7039 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7040 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7041 md_wakeup_thread(mddev
->thread
);
7042 if (mddev
->event_work
.func
)
7043 queue_work(md_misc_wq
, &mddev
->event_work
);
7044 md_new_event_inintr(mddev
);
7046 EXPORT_SYMBOL(md_error
);
7048 /* seq_file implementation /proc/mdstat */
7050 static void status_unused(struct seq_file
*seq
)
7053 struct md_rdev
*rdev
;
7055 seq_printf(seq
, "unused devices: ");
7057 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
7058 char b
[BDEVNAME_SIZE
];
7060 seq_printf(seq
, "%s ",
7061 bdevname(rdev
->bdev
,b
));
7064 seq_printf(seq
, "<none>");
7066 seq_printf(seq
, "\n");
7069 static void status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
7071 sector_t max_sectors
, resync
, res
;
7072 unsigned long dt
, db
;
7075 unsigned int per_milli
;
7077 if (mddev
->curr_resync
<= 3)
7080 resync
= mddev
->curr_resync
7081 - atomic_read(&mddev
->recovery_active
);
7083 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7084 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7085 max_sectors
= mddev
->resync_max_sectors
;
7087 max_sectors
= mddev
->dev_sectors
;
7089 WARN_ON(max_sectors
== 0);
7090 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7091 * in a sector_t, and (max_sectors>>scale) will fit in a
7092 * u32, as those are the requirements for sector_div.
7093 * Thus 'scale' must be at least 10
7096 if (sizeof(sector_t
) > sizeof(unsigned long)) {
7097 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
7100 res
= (resync
>>scale
)*1000;
7101 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
7105 int i
, x
= per_milli
/50, y
= 20-x
;
7106 seq_printf(seq
, "[");
7107 for (i
= 0; i
< x
; i
++)
7108 seq_printf(seq
, "=");
7109 seq_printf(seq
, ">");
7110 for (i
= 0; i
< y
; i
++)
7111 seq_printf(seq
, ".");
7112 seq_printf(seq
, "] ");
7114 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
7115 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
7117 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
7119 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
7120 "resync" : "recovery"))),
7121 per_milli
/10, per_milli
% 10,
7122 (unsigned long long) resync
/2,
7123 (unsigned long long) max_sectors
/2);
7126 * dt: time from mark until now
7127 * db: blocks written from mark until now
7128 * rt: remaining time
7130 * rt is a sector_t, so could be 32bit or 64bit.
7131 * So we divide before multiply in case it is 32bit and close
7133 * We scale the divisor (db) by 32 to avoid losing precision
7134 * near the end of resync when the number of remaining sectors
7136 * We then divide rt by 32 after multiplying by db to compensate.
7137 * The '+1' avoids division by zero if db is very small.
7139 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
7141 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
7142 - mddev
->resync_mark_cnt
;
7144 rt
= max_sectors
- resync
; /* number of remaining sectors */
7145 sector_div(rt
, db
/32+1);
7149 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
7150 ((unsigned long)rt
% 60)/6);
7152 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
7155 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
7157 struct list_head
*tmp
;
7159 struct mddev
*mddev
;
7167 spin_lock(&all_mddevs_lock
);
7168 list_for_each(tmp
,&all_mddevs
)
7170 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7172 spin_unlock(&all_mddevs_lock
);
7175 spin_unlock(&all_mddevs_lock
);
7177 return (void*)2;/* tail */
7181 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7183 struct list_head
*tmp
;
7184 struct mddev
*next_mddev
, *mddev
= v
;
7190 spin_lock(&all_mddevs_lock
);
7192 tmp
= all_mddevs
.next
;
7194 tmp
= mddev
->all_mddevs
.next
;
7195 if (tmp
!= &all_mddevs
)
7196 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7198 next_mddev
= (void*)2;
7201 spin_unlock(&all_mddevs_lock
);
7209 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7211 struct mddev
*mddev
= v
;
7213 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7217 static int md_seq_show(struct seq_file
*seq
, void *v
)
7219 struct mddev
*mddev
= v
;
7221 struct md_rdev
*rdev
;
7223 if (v
== (void*)1) {
7224 struct md_personality
*pers
;
7225 seq_printf(seq
, "Personalities : ");
7226 spin_lock(&pers_lock
);
7227 list_for_each_entry(pers
, &pers_list
, list
)
7228 seq_printf(seq
, "[%s] ", pers
->name
);
7230 spin_unlock(&pers_lock
);
7231 seq_printf(seq
, "\n");
7232 seq
->poll_event
= atomic_read(&md_event_count
);
7235 if (v
== (void*)2) {
7240 spin_lock(&mddev
->lock
);
7241 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7242 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7243 mddev
->pers
? "" : "in");
7246 seq_printf(seq
, " (read-only)");
7248 seq_printf(seq
, " (auto-read-only)");
7249 seq_printf(seq
, " %s", mddev
->pers
->name
);
7254 rdev_for_each_rcu(rdev
, mddev
) {
7255 char b
[BDEVNAME_SIZE
];
7256 seq_printf(seq
, " %s[%d]",
7257 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7258 if (test_bit(WriteMostly
, &rdev
->flags
))
7259 seq_printf(seq
, "(W)");
7260 if (test_bit(Faulty
, &rdev
->flags
)) {
7261 seq_printf(seq
, "(F)");
7264 if (rdev
->raid_disk
< 0)
7265 seq_printf(seq
, "(S)"); /* spare */
7266 if (test_bit(Replacement
, &rdev
->flags
))
7267 seq_printf(seq
, "(R)");
7268 sectors
+= rdev
->sectors
;
7272 if (!list_empty(&mddev
->disks
)) {
7274 seq_printf(seq
, "\n %llu blocks",
7275 (unsigned long long)
7276 mddev
->array_sectors
/ 2);
7278 seq_printf(seq
, "\n %llu blocks",
7279 (unsigned long long)sectors
/ 2);
7281 if (mddev
->persistent
) {
7282 if (mddev
->major_version
!= 0 ||
7283 mddev
->minor_version
!= 90) {
7284 seq_printf(seq
," super %d.%d",
7285 mddev
->major_version
,
7286 mddev
->minor_version
);
7288 } else if (mddev
->external
)
7289 seq_printf(seq
, " super external:%s",
7290 mddev
->metadata_type
);
7292 seq_printf(seq
, " super non-persistent");
7295 mddev
->pers
->status(seq
, mddev
);
7296 seq_printf(seq
, "\n ");
7297 if (mddev
->pers
->sync_request
) {
7298 if (mddev
->curr_resync
> 2) {
7299 status_resync(seq
, mddev
);
7300 seq_printf(seq
, "\n ");
7301 } else if (mddev
->curr_resync
>= 1)
7302 seq_printf(seq
, "\tresync=DELAYED\n ");
7303 else if (mddev
->recovery_cp
< MaxSector
)
7304 seq_printf(seq
, "\tresync=PENDING\n ");
7307 seq_printf(seq
, "\n ");
7309 bitmap_status(seq
, mddev
->bitmap
);
7311 seq_printf(seq
, "\n");
7313 spin_unlock(&mddev
->lock
);
7318 static const struct seq_operations md_seq_ops
= {
7319 .start
= md_seq_start
,
7320 .next
= md_seq_next
,
7321 .stop
= md_seq_stop
,
7322 .show
= md_seq_show
,
7325 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7327 struct seq_file
*seq
;
7330 error
= seq_open(file
, &md_seq_ops
);
7334 seq
= file
->private_data
;
7335 seq
->poll_event
= atomic_read(&md_event_count
);
7339 static int md_unloading
;
7340 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7342 struct seq_file
*seq
= filp
->private_data
;
7346 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;
7347 poll_wait(filp
, &md_event_waiters
, wait
);
7349 /* always allow read */
7350 mask
= POLLIN
| POLLRDNORM
;
7352 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7353 mask
|= POLLERR
| POLLPRI
;
7357 static const struct file_operations md_seq_fops
= {
7358 .owner
= THIS_MODULE
,
7359 .open
= md_seq_open
,
7361 .llseek
= seq_lseek
,
7362 .release
= seq_release_private
,
7363 .poll
= mdstat_poll
,
7366 int register_md_personality(struct md_personality
*p
)
7368 printk(KERN_INFO
"md: %s personality registered for level %d\n",
7370 spin_lock(&pers_lock
);
7371 list_add_tail(&p
->list
, &pers_list
);
7372 spin_unlock(&pers_lock
);
7375 EXPORT_SYMBOL(register_md_personality
);
7377 int unregister_md_personality(struct md_personality
*p
)
7379 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7380 spin_lock(&pers_lock
);
7381 list_del_init(&p
->list
);
7382 spin_unlock(&pers_lock
);
7385 EXPORT_SYMBOL(unregister_md_personality
);
7387 int register_md_cluster_operations(struct md_cluster_operations
*ops
, struct module
*module
)
7389 if (md_cluster_ops
!= NULL
)
7391 spin_lock(&pers_lock
);
7392 md_cluster_ops
= ops
;
7393 md_cluster_mod
= module
;
7394 spin_unlock(&pers_lock
);
7397 EXPORT_SYMBOL(register_md_cluster_operations
);
7399 int unregister_md_cluster_operations(void)
7401 spin_lock(&pers_lock
);
7402 md_cluster_ops
= NULL
;
7403 spin_unlock(&pers_lock
);
7406 EXPORT_SYMBOL(unregister_md_cluster_operations
);
7408 int md_setup_cluster(struct mddev
*mddev
, int nodes
)
7412 err
= request_module("md-cluster");
7414 pr_err("md-cluster module not found.\n");
7418 spin_lock(&pers_lock
);
7419 if (!md_cluster_ops
|| !try_module_get(md_cluster_mod
)) {
7420 spin_unlock(&pers_lock
);
7423 spin_unlock(&pers_lock
);
7425 return md_cluster_ops
->join(mddev
, nodes
);
7428 void md_cluster_stop(struct mddev
*mddev
)
7430 if (!md_cluster_ops
)
7432 md_cluster_ops
->leave(mddev
);
7433 module_put(md_cluster_mod
);
7436 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7438 struct md_rdev
*rdev
;
7444 rdev_for_each_rcu(rdev
, mddev
) {
7445 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7446 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7447 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7448 atomic_read(&disk
->sync_io
);
7449 /* sync IO will cause sync_io to increase before the disk_stats
7450 * as sync_io is counted when a request starts, and
7451 * disk_stats is counted when it completes.
7452 * So resync activity will cause curr_events to be smaller than
7453 * when there was no such activity.
7454 * non-sync IO will cause disk_stat to increase without
7455 * increasing sync_io so curr_events will (eventually)
7456 * be larger than it was before. Once it becomes
7457 * substantially larger, the test below will cause
7458 * the array to appear non-idle, and resync will slow
7460 * If there is a lot of outstanding resync activity when
7461 * we set last_event to curr_events, then all that activity
7462 * completing might cause the array to appear non-idle
7463 * and resync will be slowed down even though there might
7464 * not have been non-resync activity. This will only
7465 * happen once though. 'last_events' will soon reflect
7466 * the state where there is little or no outstanding
7467 * resync requests, and further resync activity will
7468 * always make curr_events less than last_events.
7471 if (init
|| curr_events
- rdev
->last_events
> 64) {
7472 rdev
->last_events
= curr_events
;
7480 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7482 /* another "blocks" (512byte) blocks have been synced */
7483 atomic_sub(blocks
, &mddev
->recovery_active
);
7484 wake_up(&mddev
->recovery_wait
);
7486 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7487 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7488 md_wakeup_thread(mddev
->thread
);
7489 // stop recovery, signal do_sync ....
7492 EXPORT_SYMBOL(md_done_sync
);
7494 /* md_write_start(mddev, bi)
7495 * If we need to update some array metadata (e.g. 'active' flag
7496 * in superblock) before writing, schedule a superblock update
7497 * and wait for it to complete.
7499 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7502 if (bio_data_dir(bi
) != WRITE
)
7505 BUG_ON(mddev
->ro
== 1);
7506 if (mddev
->ro
== 2) {
7507 /* need to switch to read/write */
7509 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7510 md_wakeup_thread(mddev
->thread
);
7511 md_wakeup_thread(mddev
->sync_thread
);
7514 atomic_inc(&mddev
->writes_pending
);
7515 if (mddev
->safemode
== 1)
7516 mddev
->safemode
= 0;
7517 if (mddev
->in_sync
) {
7518 spin_lock(&mddev
->lock
);
7519 if (mddev
->in_sync
) {
7521 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7522 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7523 md_wakeup_thread(mddev
->thread
);
7526 spin_unlock(&mddev
->lock
);
7529 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7530 wait_event(mddev
->sb_wait
,
7531 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7533 EXPORT_SYMBOL(md_write_start
);
7535 void md_write_end(struct mddev
*mddev
)
7537 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7538 if (mddev
->safemode
== 2)
7539 md_wakeup_thread(mddev
->thread
);
7540 else if (mddev
->safemode_delay
)
7541 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7544 EXPORT_SYMBOL(md_write_end
);
7546 /* md_allow_write(mddev)
7547 * Calling this ensures that the array is marked 'active' so that writes
7548 * may proceed without blocking. It is important to call this before
7549 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7550 * Must be called with mddev_lock held.
7552 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7553 * is dropped, so return -EAGAIN after notifying userspace.
7555 int md_allow_write(struct mddev
*mddev
)
7561 if (!mddev
->pers
->sync_request
)
7564 spin_lock(&mddev
->lock
);
7565 if (mddev
->in_sync
) {
7567 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7568 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7569 if (mddev
->safemode_delay
&&
7570 mddev
->safemode
== 0)
7571 mddev
->safemode
= 1;
7572 spin_unlock(&mddev
->lock
);
7573 if (mddev_is_clustered(mddev
))
7574 md_cluster_ops
->metadata_update_start(mddev
);
7575 md_update_sb(mddev
, 0);
7576 if (mddev_is_clustered(mddev
))
7577 md_cluster_ops
->metadata_update_finish(mddev
);
7578 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7580 spin_unlock(&mddev
->lock
);
7582 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7587 EXPORT_SYMBOL_GPL(md_allow_write
);
7589 #define SYNC_MARKS 10
7590 #define SYNC_MARK_STEP (3*HZ)
7591 #define UPDATE_FREQUENCY (5*60*HZ)
7592 void md_do_sync(struct md_thread
*thread
)
7594 struct mddev
*mddev
= thread
->mddev
;
7595 struct mddev
*mddev2
;
7596 unsigned int currspeed
= 0,
7598 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
7599 unsigned long mark
[SYNC_MARKS
];
7600 unsigned long update_time
;
7601 sector_t mark_cnt
[SYNC_MARKS
];
7603 struct list_head
*tmp
;
7604 sector_t last_check
;
7606 struct md_rdev
*rdev
;
7607 char *desc
, *action
= NULL
;
7608 struct blk_plug plug
;
7610 /* just incase thread restarts... */
7611 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7613 if (mddev
->ro
) {/* never try to sync a read-only array */
7614 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7618 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7619 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
7620 desc
= "data-check";
7622 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7623 desc
= "requested-resync";
7627 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7632 mddev
->last_sync_action
= action
?: desc
;
7634 /* we overload curr_resync somewhat here.
7635 * 0 == not engaged in resync at all
7636 * 2 == checking that there is no conflict with another sync
7637 * 1 == like 2, but have yielded to allow conflicting resync to
7639 * other == active in resync - this many blocks
7641 * Before starting a resync we must have set curr_resync to
7642 * 2, and then checked that every "conflicting" array has curr_resync
7643 * less than ours. When we find one that is the same or higher
7644 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7645 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7646 * This will mean we have to start checking from the beginning again.
7651 mddev
->curr_resync
= 2;
7654 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7656 for_each_mddev(mddev2
, tmp
) {
7657 if (mddev2
== mddev
)
7659 if (!mddev
->parallel_resync
7660 && mddev2
->curr_resync
7661 && match_mddev_units(mddev
, mddev2
)) {
7663 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7664 /* arbitrarily yield */
7665 mddev
->curr_resync
= 1;
7666 wake_up(&resync_wait
);
7668 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7669 /* no need to wait here, we can wait the next
7670 * time 'round when curr_resync == 2
7673 /* We need to wait 'interruptible' so as not to
7674 * contribute to the load average, and not to
7675 * be caught by 'softlockup'
7677 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7678 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7679 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7680 printk(KERN_INFO
"md: delaying %s of %s"
7681 " until %s has finished (they"
7682 " share one or more physical units)\n",
7683 desc
, mdname(mddev
), mdname(mddev2
));
7685 if (signal_pending(current
))
7686 flush_signals(current
);
7688 finish_wait(&resync_wait
, &wq
);
7691 finish_wait(&resync_wait
, &wq
);
7694 } while (mddev
->curr_resync
< 2);
7697 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7698 /* resync follows the size requested by the personality,
7699 * which defaults to physical size, but can be virtual size
7701 max_sectors
= mddev
->resync_max_sectors
;
7702 atomic64_set(&mddev
->resync_mismatches
, 0);
7703 /* we don't use the checkpoint if there's a bitmap */
7704 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7705 j
= mddev
->resync_min
;
7706 else if (!mddev
->bitmap
)
7707 j
= mddev
->recovery_cp
;
7709 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7710 max_sectors
= mddev
->resync_max_sectors
;
7712 /* recovery follows the physical size of devices */
7713 max_sectors
= mddev
->dev_sectors
;
7716 rdev_for_each_rcu(rdev
, mddev
)
7717 if (rdev
->raid_disk
>= 0 &&
7718 !test_bit(Faulty
, &rdev
->flags
) &&
7719 !test_bit(In_sync
, &rdev
->flags
) &&
7720 rdev
->recovery_offset
< j
)
7721 j
= rdev
->recovery_offset
;
7724 /* If there is a bitmap, we need to make sure all
7725 * writes that started before we added a spare
7726 * complete before we start doing a recovery.
7727 * Otherwise the write might complete and (via
7728 * bitmap_endwrite) set a bit in the bitmap after the
7729 * recovery has checked that bit and skipped that
7732 if (mddev
->bitmap
) {
7733 mddev
->pers
->quiesce(mddev
, 1);
7734 mddev
->pers
->quiesce(mddev
, 0);
7738 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7739 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7740 " %d KB/sec/disk.\n", speed_min(mddev
));
7741 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7742 "(but not more than %d KB/sec) for %s.\n",
7743 speed_max(mddev
), desc
);
7745 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7748 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7750 mark_cnt
[m
] = io_sectors
;
7753 mddev
->resync_mark
= mark
[last_mark
];
7754 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7757 * Tune reconstruction:
7759 window
= 32*(PAGE_SIZE
/512);
7760 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7761 window
/2, (unsigned long long)max_sectors
/2);
7763 atomic_set(&mddev
->recovery_active
, 0);
7768 "md: resuming %s of %s from checkpoint.\n",
7769 desc
, mdname(mddev
));
7770 mddev
->curr_resync
= j
;
7772 mddev
->curr_resync
= 3; /* no longer delayed */
7773 mddev
->curr_resync_completed
= j
;
7774 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7775 md_new_event(mddev
);
7776 update_time
= jiffies
;
7778 if (mddev_is_clustered(mddev
))
7779 md_cluster_ops
->resync_start(mddev
, j
, max_sectors
);
7781 blk_start_plug(&plug
);
7782 while (j
< max_sectors
) {
7787 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7788 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7789 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7790 > (max_sectors
>> 4)) ||
7791 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7792 (j
- mddev
->curr_resync_completed
)*2
7793 >= mddev
->resync_max
- mddev
->curr_resync_completed
7795 /* time to update curr_resync_completed */
7796 wait_event(mddev
->recovery_wait
,
7797 atomic_read(&mddev
->recovery_active
) == 0);
7798 mddev
->curr_resync_completed
= j
;
7799 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
7800 j
> mddev
->recovery_cp
)
7801 mddev
->recovery_cp
= j
;
7802 update_time
= jiffies
;
7803 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7804 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7807 while (j
>= mddev
->resync_max
&&
7808 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7809 /* As this condition is controlled by user-space,
7810 * we can block indefinitely, so use '_interruptible'
7811 * to avoid triggering warnings.
7813 flush_signals(current
); /* just in case */
7814 wait_event_interruptible(mddev
->recovery_wait
,
7815 mddev
->resync_max
> j
7816 || test_bit(MD_RECOVERY_INTR
,
7820 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7823 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
);
7825 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7829 if (!skipped
) { /* actual IO requested */
7830 io_sectors
+= sectors
;
7831 atomic_add(sectors
, &mddev
->recovery_active
);
7834 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7839 mddev
->curr_resync
= j
;
7840 if (mddev_is_clustered(mddev
))
7841 md_cluster_ops
->resync_info_update(mddev
, j
, max_sectors
);
7842 mddev
->curr_mark_cnt
= io_sectors
;
7843 if (last_check
== 0)
7844 /* this is the earliest that rebuild will be
7845 * visible in /proc/mdstat
7847 md_new_event(mddev
);
7849 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7852 last_check
= io_sectors
;
7854 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7856 int next
= (last_mark
+1) % SYNC_MARKS
;
7858 mddev
->resync_mark
= mark
[next
];
7859 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7860 mark
[next
] = jiffies
;
7861 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7865 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7869 * this loop exits only if either when we are slower than
7870 * the 'hard' speed limit, or the system was IO-idle for
7872 * the system might be non-idle CPU-wise, but we only care
7873 * about not overloading the IO subsystem. (things like an
7874 * e2fsck being done on the RAID array should execute fast)
7878 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
7879 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
7880 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7882 if (currspeed
> speed_min(mddev
)) {
7883 if (currspeed
> speed_max(mddev
)) {
7887 if (!is_mddev_idle(mddev
, 0)) {
7889 * Give other IO more of a chance.
7890 * The faster the devices, the less we wait.
7892 wait_event(mddev
->recovery_wait
,
7893 !atomic_read(&mddev
->recovery_active
));
7897 printk(KERN_INFO
"md: %s: %s %s.\n",mdname(mddev
), desc
,
7898 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
7899 ? "interrupted" : "done");
7901 * this also signals 'finished resyncing' to md_stop
7903 blk_finish_plug(&plug
);
7904 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7906 /* tell personality that we are finished */
7907 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
);
7909 if (mddev_is_clustered(mddev
))
7910 md_cluster_ops
->resync_finish(mddev
);
7912 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7913 mddev
->curr_resync
> 2) {
7914 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7915 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7916 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7918 "md: checkpointing %s of %s.\n",
7919 desc
, mdname(mddev
));
7920 if (test_bit(MD_RECOVERY_ERROR
,
7922 mddev
->recovery_cp
=
7923 mddev
->curr_resync_completed
;
7925 mddev
->recovery_cp
=
7929 mddev
->recovery_cp
= MaxSector
;
7931 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7932 mddev
->curr_resync
= MaxSector
;
7934 rdev_for_each_rcu(rdev
, mddev
)
7935 if (rdev
->raid_disk
>= 0 &&
7936 mddev
->delta_disks
>= 0 &&
7937 !test_bit(Faulty
, &rdev
->flags
) &&
7938 !test_bit(In_sync
, &rdev
->flags
) &&
7939 rdev
->recovery_offset
< mddev
->curr_resync
)
7940 rdev
->recovery_offset
= mddev
->curr_resync
;
7945 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7947 spin_lock(&mddev
->lock
);
7948 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7949 /* We completed so min/max setting can be forgotten if used. */
7950 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7951 mddev
->resync_min
= 0;
7952 mddev
->resync_max
= MaxSector
;
7953 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7954 mddev
->resync_min
= mddev
->curr_resync_completed
;
7955 mddev
->curr_resync
= 0;
7956 spin_unlock(&mddev
->lock
);
7958 wake_up(&resync_wait
);
7959 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7960 md_wakeup_thread(mddev
->thread
);
7963 EXPORT_SYMBOL_GPL(md_do_sync
);
7965 static int remove_and_add_spares(struct mddev
*mddev
,
7966 struct md_rdev
*this)
7968 struct md_rdev
*rdev
;
7972 rdev_for_each(rdev
, mddev
)
7973 if ((this == NULL
|| rdev
== this) &&
7974 rdev
->raid_disk
>= 0 &&
7975 !test_bit(Blocked
, &rdev
->flags
) &&
7976 (test_bit(Faulty
, &rdev
->flags
) ||
7977 ! test_bit(In_sync
, &rdev
->flags
)) &&
7978 atomic_read(&rdev
->nr_pending
)==0) {
7979 if (mddev
->pers
->hot_remove_disk(
7980 mddev
, rdev
) == 0) {
7981 sysfs_unlink_rdev(mddev
, rdev
);
7982 rdev
->raid_disk
= -1;
7986 if (removed
&& mddev
->kobj
.sd
)
7987 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
7992 rdev_for_each(rdev
, mddev
) {
7993 if (rdev
->raid_disk
>= 0 &&
7994 !test_bit(In_sync
, &rdev
->flags
) &&
7995 !test_bit(Faulty
, &rdev
->flags
))
7997 if (rdev
->raid_disk
>= 0)
7999 if (test_bit(Faulty
, &rdev
->flags
))
8002 ! (rdev
->saved_raid_disk
>= 0 &&
8003 !test_bit(Bitmap_sync
, &rdev
->flags
)))
8006 if (rdev
->saved_raid_disk
< 0)
8007 rdev
->recovery_offset
= 0;
8009 hot_add_disk(mddev
, rdev
) == 0) {
8010 if (sysfs_link_rdev(mddev
, rdev
))
8011 /* failure here is OK */;
8013 md_new_event(mddev
);
8014 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8019 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8023 static void md_start_sync(struct work_struct
*ws
)
8025 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
8027 mddev
->sync_thread
= md_register_thread(md_do_sync
,
8030 if (!mddev
->sync_thread
) {
8031 printk(KERN_ERR
"%s: could not start resync"
8034 /* leave the spares where they are, it shouldn't hurt */
8035 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8036 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8037 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8038 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8039 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8040 wake_up(&resync_wait
);
8041 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8043 if (mddev
->sysfs_action
)
8044 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8046 md_wakeup_thread(mddev
->sync_thread
);
8047 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8048 md_new_event(mddev
);
8052 * This routine is regularly called by all per-raid-array threads to
8053 * deal with generic issues like resync and super-block update.
8054 * Raid personalities that don't have a thread (linear/raid0) do not
8055 * need this as they never do any recovery or update the superblock.
8057 * It does not do any resync itself, but rather "forks" off other threads
8058 * to do that as needed.
8059 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8060 * "->recovery" and create a thread at ->sync_thread.
8061 * When the thread finishes it sets MD_RECOVERY_DONE
8062 * and wakeups up this thread which will reap the thread and finish up.
8063 * This thread also removes any faulty devices (with nr_pending == 0).
8065 * The overall approach is:
8066 * 1/ if the superblock needs updating, update it.
8067 * 2/ If a recovery thread is running, don't do anything else.
8068 * 3/ If recovery has finished, clean up, possibly marking spares active.
8069 * 4/ If there are any faulty devices, remove them.
8070 * 5/ If array is degraded, try to add spares devices
8071 * 6/ If array has spares or is not in-sync, start a resync thread.
8073 void md_check_recovery(struct mddev
*mddev
)
8075 if (mddev
->suspended
)
8079 bitmap_daemon_work(mddev
);
8081 if (signal_pending(current
)) {
8082 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
8083 printk(KERN_INFO
"md: %s in immediate safe mode\n",
8085 mddev
->safemode
= 2;
8087 flush_signals(current
);
8090 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
8093 (mddev
->flags
& MD_UPDATE_SB_FLAGS
& ~ (1<<MD_CHANGE_PENDING
)) ||
8094 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8095 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8096 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
8097 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
8098 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
8102 if (mddev_trylock(mddev
)) {
8106 /* On a read-only array we can:
8107 * - remove failed devices
8108 * - add already-in_sync devices if the array itself
8110 * As we only add devices that are already in-sync,
8111 * we can activate the spares immediately.
8113 remove_and_add_spares(mddev
, NULL
);
8114 /* There is no thread, but we need to call
8115 * ->spare_active and clear saved_raid_disk
8117 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8118 md_reap_sync_thread(mddev
);
8119 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8123 if (!mddev
->external
) {
8125 spin_lock(&mddev
->lock
);
8126 if (mddev
->safemode
&&
8127 !atomic_read(&mddev
->writes_pending
) &&
8129 mddev
->recovery_cp
== MaxSector
) {
8132 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
8134 if (mddev
->safemode
== 1)
8135 mddev
->safemode
= 0;
8136 spin_unlock(&mddev
->lock
);
8138 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8141 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
) {
8142 if (mddev_is_clustered(mddev
))
8143 md_cluster_ops
->metadata_update_start(mddev
);
8144 md_update_sb(mddev
, 0);
8145 if (mddev_is_clustered(mddev
))
8146 md_cluster_ops
->metadata_update_finish(mddev
);
8149 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
8150 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
8151 /* resync/recovery still happening */
8152 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8155 if (mddev
->sync_thread
) {
8156 md_reap_sync_thread(mddev
);
8159 /* Set RUNNING before clearing NEEDED to avoid
8160 * any transients in the value of "sync_action".
8162 mddev
->curr_resync_completed
= 0;
8163 spin_lock(&mddev
->lock
);
8164 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8165 spin_unlock(&mddev
->lock
);
8166 /* Clear some bits that don't mean anything, but
8169 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8170 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8172 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8173 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
8175 /* no recovery is running.
8176 * remove any failed drives, then
8177 * add spares if possible.
8178 * Spares are also removed and re-added, to allow
8179 * the personality to fail the re-add.
8182 if (mddev
->reshape_position
!= MaxSector
) {
8183 if (mddev
->pers
->check_reshape
== NULL
||
8184 mddev
->pers
->check_reshape(mddev
) != 0)
8185 /* Cannot proceed */
8187 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8188 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8189 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
8190 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8191 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8192 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8193 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8194 } else if (mddev
->recovery_cp
< MaxSector
) {
8195 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8196 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8197 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
8198 /* nothing to be done ... */
8201 if (mddev
->pers
->sync_request
) {
8203 /* We are adding a device or devices to an array
8204 * which has the bitmap stored on all devices.
8205 * So make sure all bitmap pages get written
8207 bitmap_write_all(mddev
->bitmap
);
8209 INIT_WORK(&mddev
->del_work
, md_start_sync
);
8210 queue_work(md_misc_wq
, &mddev
->del_work
);
8214 if (!mddev
->sync_thread
) {
8215 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8216 wake_up(&resync_wait
);
8217 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8219 if (mddev
->sysfs_action
)
8220 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8223 wake_up(&mddev
->sb_wait
);
8224 mddev_unlock(mddev
);
8227 EXPORT_SYMBOL(md_check_recovery
);
8229 void md_reap_sync_thread(struct mddev
*mddev
)
8231 struct md_rdev
*rdev
;
8233 /* resync has finished, collect result */
8234 md_unregister_thread(&mddev
->sync_thread
);
8235 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8236 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8238 /* activate any spares */
8239 if (mddev
->pers
->spare_active(mddev
)) {
8240 sysfs_notify(&mddev
->kobj
, NULL
,
8242 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8245 if (mddev_is_clustered(mddev
))
8246 md_cluster_ops
->metadata_update_start(mddev
);
8247 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8248 mddev
->pers
->finish_reshape
)
8249 mddev
->pers
->finish_reshape(mddev
);
8251 /* If array is no-longer degraded, then any saved_raid_disk
8252 * information must be scrapped.
8254 if (!mddev
->degraded
)
8255 rdev_for_each(rdev
, mddev
)
8256 rdev
->saved_raid_disk
= -1;
8258 md_update_sb(mddev
, 1);
8259 if (mddev_is_clustered(mddev
))
8260 md_cluster_ops
->metadata_update_finish(mddev
);
8261 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8262 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8263 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8264 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8265 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8266 wake_up(&resync_wait
);
8267 /* flag recovery needed just to double check */
8268 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8269 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8270 md_new_event(mddev
);
8271 if (mddev
->event_work
.func
)
8272 queue_work(md_misc_wq
, &mddev
->event_work
);
8274 EXPORT_SYMBOL(md_reap_sync_thread
);
8276 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
8278 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8279 wait_event_timeout(rdev
->blocked_wait
,
8280 !test_bit(Blocked
, &rdev
->flags
) &&
8281 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
8282 msecs_to_jiffies(5000));
8283 rdev_dec_pending(rdev
, mddev
);
8285 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
8287 void md_finish_reshape(struct mddev
*mddev
)
8289 /* called be personality module when reshape completes. */
8290 struct md_rdev
*rdev
;
8292 rdev_for_each(rdev
, mddev
) {
8293 if (rdev
->data_offset
> rdev
->new_data_offset
)
8294 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
8296 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
8297 rdev
->data_offset
= rdev
->new_data_offset
;
8300 EXPORT_SYMBOL(md_finish_reshape
);
8302 /* Bad block management.
8303 * We can record which blocks on each device are 'bad' and so just
8304 * fail those blocks, or that stripe, rather than the whole device.
8305 * Entries in the bad-block table are 64bits wide. This comprises:
8306 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8307 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8308 * A 'shift' can be set so that larger blocks are tracked and
8309 * consequently larger devices can be covered.
8310 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8312 * Locking of the bad-block table uses a seqlock so md_is_badblock
8313 * might need to retry if it is very unlucky.
8314 * We will sometimes want to check for bad blocks in a bi_end_io function,
8315 * so we use the write_seqlock_irq variant.
8317 * When looking for a bad block we specify a range and want to
8318 * know if any block in the range is bad. So we binary-search
8319 * to the last range that starts at-or-before the given endpoint,
8320 * (or "before the sector after the target range")
8321 * then see if it ends after the given start.
8323 * 0 if there are no known bad blocks in the range
8324 * 1 if there are known bad block which are all acknowledged
8325 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8326 * plus the start/length of the first bad section we overlap.
8328 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
8329 sector_t
*first_bad
, int *bad_sectors
)
8335 sector_t target
= s
+ sectors
;
8338 if (bb
->shift
> 0) {
8339 /* round the start down, and the end up */
8341 target
+= (1<<bb
->shift
) - 1;
8342 target
>>= bb
->shift
;
8343 sectors
= target
- s
;
8345 /* 'target' is now the first block after the bad range */
8348 seq
= read_seqbegin(&bb
->lock
);
8353 /* Binary search between lo and hi for 'target'
8354 * i.e. for the last range that starts before 'target'
8356 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8357 * are known not to be the last range before target.
8358 * VARIANT: hi-lo is the number of possible
8359 * ranges, and decreases until it reaches 1
8361 while (hi
- lo
> 1) {
8362 int mid
= (lo
+ hi
) / 2;
8363 sector_t a
= BB_OFFSET(p
[mid
]);
8365 /* This could still be the one, earlier ranges
8369 /* This and later ranges are definitely out. */
8372 /* 'lo' might be the last that started before target, but 'hi' isn't */
8374 /* need to check all range that end after 's' to see if
8375 * any are unacknowledged.
8378 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8379 if (BB_OFFSET(p
[lo
]) < target
) {
8380 /* starts before the end, and finishes after
8381 * the start, so they must overlap
8383 if (rv
!= -1 && BB_ACK(p
[lo
]))
8387 *first_bad
= BB_OFFSET(p
[lo
]);
8388 *bad_sectors
= BB_LEN(p
[lo
]);
8394 if (read_seqretry(&bb
->lock
, seq
))
8399 EXPORT_SYMBOL_GPL(md_is_badblock
);
8402 * Add a range of bad blocks to the table.
8403 * This might extend the table, or might contract it
8404 * if two adjacent ranges can be merged.
8405 * We binary-search to find the 'insertion' point, then
8406 * decide how best to handle it.
8408 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
8414 unsigned long flags
;
8417 /* badblocks are disabled */
8421 /* round the start down, and the end up */
8422 sector_t next
= s
+ sectors
;
8424 next
+= (1<<bb
->shift
) - 1;
8429 write_seqlock_irqsave(&bb
->lock
, flags
);
8434 /* Find the last range that starts at-or-before 's' */
8435 while (hi
- lo
> 1) {
8436 int mid
= (lo
+ hi
) / 2;
8437 sector_t a
= BB_OFFSET(p
[mid
]);
8443 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8447 /* we found a range that might merge with the start
8450 sector_t a
= BB_OFFSET(p
[lo
]);
8451 sector_t e
= a
+ BB_LEN(p
[lo
]);
8452 int ack
= BB_ACK(p
[lo
]);
8454 /* Yes, we can merge with a previous range */
8455 if (s
== a
&& s
+ sectors
>= e
)
8456 /* new range covers old */
8459 ack
= ack
&& acknowledged
;
8461 if (e
< s
+ sectors
)
8463 if (e
- a
<= BB_MAX_LEN
) {
8464 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8467 /* does not all fit in one range,
8468 * make p[lo] maximal
8470 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8471 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8477 if (sectors
&& hi
< bb
->count
) {
8478 /* 'hi' points to the first range that starts after 's'.
8479 * Maybe we can merge with the start of that range */
8480 sector_t a
= BB_OFFSET(p
[hi
]);
8481 sector_t e
= a
+ BB_LEN(p
[hi
]);
8482 int ack
= BB_ACK(p
[hi
]);
8483 if (a
<= s
+ sectors
) {
8484 /* merging is possible */
8485 if (e
<= s
+ sectors
) {
8490 ack
= ack
&& acknowledged
;
8493 if (e
- a
<= BB_MAX_LEN
) {
8494 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8497 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8505 if (sectors
== 0 && hi
< bb
->count
) {
8506 /* we might be able to combine lo and hi */
8507 /* Note: 's' is at the end of 'lo' */
8508 sector_t a
= BB_OFFSET(p
[hi
]);
8509 int lolen
= BB_LEN(p
[lo
]);
8510 int hilen
= BB_LEN(p
[hi
]);
8511 int newlen
= lolen
+ hilen
- (s
- a
);
8512 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8513 /* yes, we can combine them */
8514 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8515 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8516 memmove(p
+ hi
, p
+ hi
+ 1,
8517 (bb
->count
- hi
- 1) * 8);
8522 /* didn't merge (it all).
8523 * Need to add a range just before 'hi' */
8524 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8525 /* No room for more */
8529 int this_sectors
= sectors
;
8530 memmove(p
+ hi
+ 1, p
+ hi
,
8531 (bb
->count
- hi
) * 8);
8534 if (this_sectors
> BB_MAX_LEN
)
8535 this_sectors
= BB_MAX_LEN
;
8536 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8537 sectors
-= this_sectors
;
8544 bb
->unacked_exist
= 1;
8545 write_sequnlock_irqrestore(&bb
->lock
, flags
);
8550 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8555 s
+= rdev
->new_data_offset
;
8557 s
+= rdev
->data_offset
;
8558 rv
= md_set_badblocks(&rdev
->badblocks
,
8561 /* Make sure they get written out promptly */
8562 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8563 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8564 md_wakeup_thread(rdev
->mddev
->thread
);
8568 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8571 * Remove a range of bad blocks from the table.
8572 * This may involve extending the table if we spilt a region,
8573 * but it must not fail. So if the table becomes full, we just
8574 * drop the remove request.
8576 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8580 sector_t target
= s
+ sectors
;
8583 if (bb
->shift
> 0) {
8584 /* When clearing we round the start up and the end down.
8585 * This should not matter as the shift should align with
8586 * the block size and no rounding should ever be needed.
8587 * However it is better the think a block is bad when it
8588 * isn't than to think a block is not bad when it is.
8590 s
+= (1<<bb
->shift
) - 1;
8592 target
>>= bb
->shift
;
8593 sectors
= target
- s
;
8596 write_seqlock_irq(&bb
->lock
);
8601 /* Find the last range that starts before 'target' */
8602 while (hi
- lo
> 1) {
8603 int mid
= (lo
+ hi
) / 2;
8604 sector_t a
= BB_OFFSET(p
[mid
]);
8611 /* p[lo] is the last range that could overlap the
8612 * current range. Earlier ranges could also overlap,
8613 * but only this one can overlap the end of the range.
8615 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8616 /* Partial overlap, leave the tail of this range */
8617 int ack
= BB_ACK(p
[lo
]);
8618 sector_t a
= BB_OFFSET(p
[lo
]);
8619 sector_t end
= a
+ BB_LEN(p
[lo
]);
8622 /* we need to split this range */
8623 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8627 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8629 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8632 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8633 /* there is no longer an overlap */
8638 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8639 /* This range does overlap */
8640 if (BB_OFFSET(p
[lo
]) < s
) {
8641 /* Keep the early parts of this range. */
8642 int ack
= BB_ACK(p
[lo
]);
8643 sector_t start
= BB_OFFSET(p
[lo
]);
8644 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8645 /* now low doesn't overlap, so.. */
8650 /* 'lo' is strictly before, 'hi' is strictly after,
8651 * anything between needs to be discarded
8654 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8655 bb
->count
-= (hi
- lo
- 1);
8661 write_sequnlock_irq(&bb
->lock
);
8665 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8669 s
+= rdev
->new_data_offset
;
8671 s
+= rdev
->data_offset
;
8672 return md_clear_badblocks(&rdev
->badblocks
,
8675 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8678 * Acknowledge all bad blocks in a list.
8679 * This only succeeds if ->changed is clear. It is used by
8680 * in-kernel metadata updates
8682 void md_ack_all_badblocks(struct badblocks
*bb
)
8684 if (bb
->page
== NULL
|| bb
->changed
)
8685 /* no point even trying */
8687 write_seqlock_irq(&bb
->lock
);
8689 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8692 for (i
= 0; i
< bb
->count
; i
++) {
8693 if (!BB_ACK(p
[i
])) {
8694 sector_t start
= BB_OFFSET(p
[i
]);
8695 int len
= BB_LEN(p
[i
]);
8696 p
[i
] = BB_MAKE(start
, len
, 1);
8699 bb
->unacked_exist
= 0;
8701 write_sequnlock_irq(&bb
->lock
);
8703 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8705 /* sysfs access to bad-blocks list.
8706 * We present two files.
8707 * 'bad-blocks' lists sector numbers and lengths of ranges that
8708 * are recorded as bad. The list is truncated to fit within
8709 * the one-page limit of sysfs.
8710 * Writing "sector length" to this file adds an acknowledged
8712 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8713 * been acknowledged. Writing to this file adds bad blocks
8714 * without acknowledging them. This is largely for testing.
8718 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8729 seq
= read_seqbegin(&bb
->lock
);
8734 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8735 sector_t s
= BB_OFFSET(p
[i
]);
8736 unsigned int length
= BB_LEN(p
[i
]);
8737 int ack
= BB_ACK(p
[i
]);
8743 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8744 (unsigned long long)s
<< bb
->shift
,
8745 length
<< bb
->shift
);
8747 if (unack
&& len
== 0)
8748 bb
->unacked_exist
= 0;
8750 if (read_seqretry(&bb
->lock
, seq
))
8759 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8761 unsigned long long sector
;
8765 /* Allow clearing via sysfs *only* for testing/debugging.
8766 * Normally only a successful write may clear a badblock
8769 if (page
[0] == '-') {
8773 #endif /* DO_DEBUG */
8775 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8777 if (newline
!= '\n')
8789 md_clear_badblocks(bb
, sector
, length
);
8792 #endif /* DO_DEBUG */
8793 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8799 static int md_notify_reboot(struct notifier_block
*this,
8800 unsigned long code
, void *x
)
8802 struct list_head
*tmp
;
8803 struct mddev
*mddev
;
8806 for_each_mddev(mddev
, tmp
) {
8807 if (mddev_trylock(mddev
)) {
8809 __md_stop_writes(mddev
);
8810 if (mddev
->persistent
)
8811 mddev
->safemode
= 2;
8812 mddev_unlock(mddev
);
8817 * certain more exotic SCSI devices are known to be
8818 * volatile wrt too early system reboots. While the
8819 * right place to handle this issue is the given
8820 * driver, we do want to have a safe RAID driver ...
8828 static struct notifier_block md_notifier
= {
8829 .notifier_call
= md_notify_reboot
,
8831 .priority
= INT_MAX
, /* before any real devices */
8834 static void md_geninit(void)
8836 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8838 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8841 static int __init
md_init(void)
8845 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8849 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8853 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8856 if ((ret
= register_blkdev(0, "mdp")) < 0)
8860 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
8861 md_probe
, NULL
, NULL
);
8862 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8863 md_probe
, NULL
, NULL
);
8865 register_reboot_notifier(&md_notifier
);
8866 raid_table_header
= register_sysctl_table(raid_root_table
);
8872 unregister_blkdev(MD_MAJOR
, "md");
8874 destroy_workqueue(md_misc_wq
);
8876 destroy_workqueue(md_wq
);
8881 void md_reload_sb(struct mddev
*mddev
)
8883 struct md_rdev
*rdev
, *tmp
;
8885 rdev_for_each_safe(rdev
, tmp
, mddev
) {
8886 rdev
->sb_loaded
= 0;
8887 ClearPageUptodate(rdev
->sb_page
);
8889 mddev
->raid_disks
= 0;
8891 rdev_for_each_safe(rdev
, tmp
, mddev
) {
8892 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
8893 /* since we don't write to faulty devices, we figure out if the
8894 * disk is faulty by comparing events
8896 if (mddev
->events
> sb
->events
)
8897 set_bit(Faulty
, &rdev
->flags
);
8901 EXPORT_SYMBOL(md_reload_sb
);
8906 * Searches all registered partitions for autorun RAID arrays
8910 static LIST_HEAD(all_detected_devices
);
8911 struct detected_devices_node
{
8912 struct list_head list
;
8916 void md_autodetect_dev(dev_t dev
)
8918 struct detected_devices_node
*node_detected_dev
;
8920 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8921 if (node_detected_dev
) {
8922 node_detected_dev
->dev
= dev
;
8923 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8925 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8926 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8930 static void autostart_arrays(int part
)
8932 struct md_rdev
*rdev
;
8933 struct detected_devices_node
*node_detected_dev
;
8935 int i_scanned
, i_passed
;
8940 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8942 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8944 node_detected_dev
= list_entry(all_detected_devices
.next
,
8945 struct detected_devices_node
, list
);
8946 list_del(&node_detected_dev
->list
);
8947 dev
= node_detected_dev
->dev
;
8948 kfree(node_detected_dev
);
8949 rdev
= md_import_device(dev
,0, 90);
8953 if (test_bit(Faulty
, &rdev
->flags
))
8956 set_bit(AutoDetected
, &rdev
->flags
);
8957 list_add(&rdev
->same_set
, &pending_raid_disks
);
8961 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8962 i_scanned
, i_passed
);
8964 autorun_devices(part
);
8967 #endif /* !MODULE */
8969 static __exit
void md_exit(void)
8971 struct mddev
*mddev
;
8972 struct list_head
*tmp
;
8975 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
8976 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8978 unregister_blkdev(MD_MAJOR
,"md");
8979 unregister_blkdev(mdp_major
, "mdp");
8980 unregister_reboot_notifier(&md_notifier
);
8981 unregister_sysctl_table(raid_table_header
);
8983 /* We cannot unload the modules while some process is
8984 * waiting for us in select() or poll() - wake them up
8987 while (waitqueue_active(&md_event_waiters
)) {
8988 /* not safe to leave yet */
8989 wake_up(&md_event_waiters
);
8993 remove_proc_entry("mdstat", NULL
);
8995 for_each_mddev(mddev
, tmp
) {
8996 export_array(mddev
);
8997 mddev
->hold_active
= 0;
8999 destroy_workqueue(md_misc_wq
);
9000 destroy_workqueue(md_wq
);
9003 subsys_initcall(md_init
);
9004 module_exit(md_exit
)
9006 static int get_ro(char *buffer
, struct kernel_param
*kp
)
9008 return sprintf(buffer
, "%d", start_readonly
);
9010 static int set_ro(const char *val
, struct kernel_param
*kp
)
9013 int num
= simple_strtoul(val
, &e
, 10);
9014 if (*val
&& (*e
== '\0' || *e
== '\n')) {
9015 start_readonly
= num
;
9021 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
9022 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
9023 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
9025 MODULE_LICENSE("GPL");
9026 MODULE_DESCRIPTION("MD RAID framework");
9028 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);