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>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
70 static struct workqueue_struct
*md_wq
;
71 static struct workqueue_struct
*md_misc_wq
;
73 static int remove_and_add_spares(struct mddev
*mddev
,
74 struct md_rdev
*this);
77 * Default number of read corrections we'll attempt on an rdev
78 * before ejecting it from the array. We divide the read error
79 * count by 2 for every hour elapsed between read errors.
81 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
83 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
84 * is 1000 KB/sec, so the extra system load does not show up that much.
85 * Increase it if you want to have more _guaranteed_ speed. Note that
86 * the RAID driver will use the maximum available bandwidth if the IO
87 * subsystem is idle. There is also an 'absolute maximum' reconstruction
88 * speed limit - in case reconstruction slows down your system despite
91 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
92 * or /sys/block/mdX/md/sync_speed_{min,max}
95 static int sysctl_speed_limit_min
= 1000;
96 static int sysctl_speed_limit_max
= 200000;
97 static inline int speed_min(struct mddev
*mddev
)
99 return mddev
->sync_speed_min
?
100 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
103 static inline int speed_max(struct mddev
*mddev
)
105 return mddev
->sync_speed_max
?
106 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
109 static struct ctl_table_header
*raid_table_header
;
111 static struct ctl_table raid_table
[] = {
113 .procname
= "speed_limit_min",
114 .data
= &sysctl_speed_limit_min
,
115 .maxlen
= sizeof(int),
116 .mode
= S_IRUGO
|S_IWUSR
,
117 .proc_handler
= proc_dointvec
,
120 .procname
= "speed_limit_max",
121 .data
= &sysctl_speed_limit_max
,
122 .maxlen
= sizeof(int),
123 .mode
= S_IRUGO
|S_IWUSR
,
124 .proc_handler
= proc_dointvec
,
129 static struct ctl_table raid_dir_table
[] = {
133 .mode
= S_IRUGO
|S_IXUGO
,
139 static struct ctl_table raid_root_table
[] = {
144 .child
= raid_dir_table
,
149 static const struct block_device_operations md_fops
;
151 static int start_readonly
;
154 * like bio_clone, but with a local bio set
157 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
162 if (!mddev
|| !mddev
->bio_set
)
163 return bio_alloc(gfp_mask
, nr_iovecs
);
165 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
170 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
172 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
175 if (!mddev
|| !mddev
->bio_set
)
176 return bio_clone(bio
, gfp_mask
);
178 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
180 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
183 * We have a system wide 'event count' that is incremented
184 * on any 'interesting' event, and readers of /proc/mdstat
185 * can use 'poll' or 'select' to find out when the event
189 * start array, stop array, error, add device, remove device,
190 * start build, activate spare
192 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
193 static atomic_t md_event_count
;
194 void md_new_event(struct mddev
*mddev
)
196 atomic_inc(&md_event_count
);
197 wake_up(&md_event_waiters
);
199 EXPORT_SYMBOL_GPL(md_new_event
);
201 /* Alternate version that can be called from interrupts
202 * when calling sysfs_notify isn't needed.
204 static void md_new_event_inintr(struct mddev
*mddev
)
206 atomic_inc(&md_event_count
);
207 wake_up(&md_event_waiters
);
211 * Enables to iterate over all existing md arrays
212 * all_mddevs_lock protects this list.
214 static LIST_HEAD(all_mddevs
);
215 static DEFINE_SPINLOCK(all_mddevs_lock
);
218 * iterates through all used mddevs in the system.
219 * We take care to grab the all_mddevs_lock whenever navigating
220 * the list, and to always hold a refcount when unlocked.
221 * Any code which breaks out of this loop while own
222 * a reference to the current mddev and must mddev_put it.
224 #define for_each_mddev(_mddev,_tmp) \
226 for (({ spin_lock(&all_mddevs_lock); \
227 _tmp = all_mddevs.next; \
229 ({ if (_tmp != &all_mddevs) \
230 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
231 spin_unlock(&all_mddevs_lock); \
232 if (_mddev) mddev_put(_mddev); \
233 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
234 _tmp != &all_mddevs;}); \
235 ({ spin_lock(&all_mddevs_lock); \
236 _tmp = _tmp->next;}) \
239 /* Rather than calling directly into the personality make_request function,
240 * IO requests come here first so that we can check if the device is
241 * being suspended pending a reconfiguration.
242 * We hold a refcount over the call to ->make_request. By the time that
243 * call has finished, the bio has been linked into some internal structure
244 * and so is visible to ->quiesce(), so we don't need the refcount any more.
246 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
248 const int rw
= bio_data_dir(bio
);
249 struct mddev
*mddev
= q
->queuedata
;
251 unsigned int sectors
;
253 if (mddev
== NULL
|| mddev
->pers
== NULL
258 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
259 bio_endio(bio
, bio_sectors(bio
) == 0 ? 0 : -EROFS
);
262 smp_rmb(); /* Ensure implications of 'active' are visible */
264 if (mddev
->suspended
) {
267 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
268 TASK_UNINTERRUPTIBLE
);
269 if (!mddev
->suspended
)
275 finish_wait(&mddev
->sb_wait
, &__wait
);
277 atomic_inc(&mddev
->active_io
);
281 * save the sectors now since our bio can
282 * go away inside make_request
284 sectors
= bio_sectors(bio
);
285 mddev
->pers
->make_request(mddev
, bio
);
287 cpu
= part_stat_lock();
288 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
289 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
292 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
293 wake_up(&mddev
->sb_wait
);
296 /* mddev_suspend makes sure no new requests are submitted
297 * to the device, and that any requests that have been submitted
298 * are completely handled.
299 * Once ->stop is called and completes, the module will be completely
302 void mddev_suspend(struct mddev
*mddev
)
304 BUG_ON(mddev
->suspended
);
305 mddev
->suspended
= 1;
307 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
308 mddev
->pers
->quiesce(mddev
, 1);
310 del_timer_sync(&mddev
->safemode_timer
);
312 EXPORT_SYMBOL_GPL(mddev_suspend
);
314 void mddev_resume(struct mddev
*mddev
)
316 mddev
->suspended
= 0;
317 wake_up(&mddev
->sb_wait
);
318 mddev
->pers
->quiesce(mddev
, 0);
320 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
321 md_wakeup_thread(mddev
->thread
);
322 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
324 EXPORT_SYMBOL_GPL(mddev_resume
);
326 int mddev_congested(struct mddev
*mddev
, int bits
)
328 return mddev
->suspended
;
330 EXPORT_SYMBOL(mddev_congested
);
333 * Generic flush handling for md
336 static void md_end_flush(struct bio
*bio
, int err
)
338 struct md_rdev
*rdev
= bio
->bi_private
;
339 struct mddev
*mddev
= rdev
->mddev
;
341 rdev_dec_pending(rdev
, mddev
);
343 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
344 /* The pre-request flush has finished */
345 queue_work(md_wq
, &mddev
->flush_work
);
350 static void md_submit_flush_data(struct work_struct
*ws
);
352 static void submit_flushes(struct work_struct
*ws
)
354 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
355 struct md_rdev
*rdev
;
357 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
358 atomic_set(&mddev
->flush_pending
, 1);
360 rdev_for_each_rcu(rdev
, mddev
)
361 if (rdev
->raid_disk
>= 0 &&
362 !test_bit(Faulty
, &rdev
->flags
)) {
363 /* Take two references, one is dropped
364 * when request finishes, one after
365 * we reclaim rcu_read_lock
368 atomic_inc(&rdev
->nr_pending
);
369 atomic_inc(&rdev
->nr_pending
);
371 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
372 bi
->bi_end_io
= md_end_flush
;
373 bi
->bi_private
= rdev
;
374 bi
->bi_bdev
= rdev
->bdev
;
375 atomic_inc(&mddev
->flush_pending
);
376 submit_bio(WRITE_FLUSH
, bi
);
378 rdev_dec_pending(rdev
, mddev
);
381 if (atomic_dec_and_test(&mddev
->flush_pending
))
382 queue_work(md_wq
, &mddev
->flush_work
);
385 static void md_submit_flush_data(struct work_struct
*ws
)
387 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
388 struct bio
*bio
= mddev
->flush_bio
;
390 if (bio
->bi_iter
.bi_size
== 0)
391 /* an empty barrier - all done */
394 bio
->bi_rw
&= ~REQ_FLUSH
;
395 mddev
->pers
->make_request(mddev
, bio
);
398 mddev
->flush_bio
= NULL
;
399 wake_up(&mddev
->sb_wait
);
402 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
404 spin_lock_irq(&mddev
->write_lock
);
405 wait_event_lock_irq(mddev
->sb_wait
,
408 mddev
->flush_bio
= bio
;
409 spin_unlock_irq(&mddev
->write_lock
);
411 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
412 queue_work(md_wq
, &mddev
->flush_work
);
414 EXPORT_SYMBOL(md_flush_request
);
416 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
418 struct mddev
*mddev
= cb
->data
;
419 md_wakeup_thread(mddev
->thread
);
422 EXPORT_SYMBOL(md_unplug
);
424 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
426 atomic_inc(&mddev
->active
);
430 static void mddev_delayed_delete(struct work_struct
*ws
);
432 static void mddev_put(struct mddev
*mddev
)
434 struct bio_set
*bs
= NULL
;
436 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
438 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
439 mddev
->ctime
== 0 && !mddev
->hold_active
) {
440 /* Array is not configured at all, and not held active,
442 list_del_init(&mddev
->all_mddevs
);
444 mddev
->bio_set
= NULL
;
445 if (mddev
->gendisk
) {
446 /* We did a probe so need to clean up. Call
447 * queue_work inside the spinlock so that
448 * flush_workqueue() after mddev_find will
449 * succeed in waiting for the work to be done.
451 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
452 queue_work(md_misc_wq
, &mddev
->del_work
);
456 spin_unlock(&all_mddevs_lock
);
461 void mddev_init(struct mddev
*mddev
)
463 mutex_init(&mddev
->open_mutex
);
464 mutex_init(&mddev
->reconfig_mutex
);
465 mutex_init(&mddev
->bitmap_info
.mutex
);
466 INIT_LIST_HEAD(&mddev
->disks
);
467 INIT_LIST_HEAD(&mddev
->all_mddevs
);
468 init_timer(&mddev
->safemode_timer
);
469 atomic_set(&mddev
->active
, 1);
470 atomic_set(&mddev
->openers
, 0);
471 atomic_set(&mddev
->active_io
, 0);
472 spin_lock_init(&mddev
->write_lock
);
473 atomic_set(&mddev
->flush_pending
, 0);
474 init_waitqueue_head(&mddev
->sb_wait
);
475 init_waitqueue_head(&mddev
->recovery_wait
);
476 mddev
->reshape_position
= MaxSector
;
477 mddev
->reshape_backwards
= 0;
478 mddev
->last_sync_action
= "none";
479 mddev
->resync_min
= 0;
480 mddev
->resync_max
= MaxSector
;
481 mddev
->level
= LEVEL_NONE
;
483 EXPORT_SYMBOL_GPL(mddev_init
);
485 static struct mddev
*mddev_find(dev_t unit
)
487 struct mddev
*mddev
, *new = NULL
;
489 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
490 unit
&= ~((1<<MdpMinorShift
)-1);
493 spin_lock(&all_mddevs_lock
);
496 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
497 if (mddev
->unit
== unit
) {
499 spin_unlock(&all_mddevs_lock
);
505 list_add(&new->all_mddevs
, &all_mddevs
);
506 spin_unlock(&all_mddevs_lock
);
507 new->hold_active
= UNTIL_IOCTL
;
511 /* find an unused unit number */
512 static int next_minor
= 512;
513 int start
= next_minor
;
517 dev
= MKDEV(MD_MAJOR
, next_minor
);
519 if (next_minor
> MINORMASK
)
521 if (next_minor
== start
) {
522 /* Oh dear, all in use. */
523 spin_unlock(&all_mddevs_lock
);
529 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
530 if (mddev
->unit
== dev
) {
536 new->md_minor
= MINOR(dev
);
537 new->hold_active
= UNTIL_STOP
;
538 list_add(&new->all_mddevs
, &all_mddevs
);
539 spin_unlock(&all_mddevs_lock
);
542 spin_unlock(&all_mddevs_lock
);
544 new = kzalloc(sizeof(*new), GFP_KERNEL
);
549 if (MAJOR(unit
) == MD_MAJOR
)
550 new->md_minor
= MINOR(unit
);
552 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
559 static inline int __must_check
mddev_lock(struct mddev
*mddev
)
561 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
564 /* Sometimes we need to take the lock in a situation where
565 * failure due to interrupts is not acceptable.
567 static inline void mddev_lock_nointr(struct mddev
*mddev
)
569 mutex_lock(&mddev
->reconfig_mutex
);
572 static inline int mddev_is_locked(struct mddev
*mddev
)
574 return mutex_is_locked(&mddev
->reconfig_mutex
);
577 static inline int mddev_trylock(struct mddev
*mddev
)
579 return mutex_trylock(&mddev
->reconfig_mutex
);
582 static struct attribute_group md_redundancy_group
;
584 static void mddev_unlock(struct mddev
*mddev
)
586 if (mddev
->to_remove
) {
587 /* These cannot be removed under reconfig_mutex as
588 * an access to the files will try to take reconfig_mutex
589 * while holding the file unremovable, which leads to
591 * So hold set sysfs_active while the remove in happeing,
592 * and anything else which might set ->to_remove or my
593 * otherwise change the sysfs namespace will fail with
594 * -EBUSY if sysfs_active is still set.
595 * We set sysfs_active under reconfig_mutex and elsewhere
596 * test it under the same mutex to ensure its correct value
599 struct attribute_group
*to_remove
= mddev
->to_remove
;
600 mddev
->to_remove
= NULL
;
601 mddev
->sysfs_active
= 1;
602 mutex_unlock(&mddev
->reconfig_mutex
);
604 if (mddev
->kobj
.sd
) {
605 if (to_remove
!= &md_redundancy_group
)
606 sysfs_remove_group(&mddev
->kobj
, to_remove
);
607 if (mddev
->pers
== NULL
||
608 mddev
->pers
->sync_request
== NULL
) {
609 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
610 if (mddev
->sysfs_action
)
611 sysfs_put(mddev
->sysfs_action
);
612 mddev
->sysfs_action
= NULL
;
615 mddev
->sysfs_active
= 0;
617 mutex_unlock(&mddev
->reconfig_mutex
);
619 /* As we've dropped the mutex we need a spinlock to
620 * make sure the thread doesn't disappear
622 spin_lock(&pers_lock
);
623 md_wakeup_thread(mddev
->thread
);
624 spin_unlock(&pers_lock
);
627 static struct md_rdev
*find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
629 struct md_rdev
*rdev
;
631 rdev_for_each_rcu(rdev
, mddev
)
632 if (rdev
->desc_nr
== nr
)
638 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
640 struct md_rdev
*rdev
;
642 rdev_for_each(rdev
, mddev
)
643 if (rdev
->bdev
->bd_dev
== dev
)
649 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
651 struct md_rdev
*rdev
;
653 rdev_for_each_rcu(rdev
, mddev
)
654 if (rdev
->bdev
->bd_dev
== dev
)
660 static struct md_personality
*find_pers(int level
, char *clevel
)
662 struct md_personality
*pers
;
663 list_for_each_entry(pers
, &pers_list
, list
) {
664 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
666 if (strcmp(pers
->name
, clevel
)==0)
672 /* return the offset of the super block in 512byte sectors */
673 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
675 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
676 return MD_NEW_SIZE_SECTORS(num_sectors
);
679 static int alloc_disk_sb(struct md_rdev
*rdev
)
681 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
682 if (!rdev
->sb_page
) {
683 printk(KERN_ALERT
"md: out of memory.\n");
690 void md_rdev_clear(struct md_rdev
*rdev
)
693 put_page(rdev
->sb_page
);
695 rdev
->sb_page
= NULL
;
700 put_page(rdev
->bb_page
);
701 rdev
->bb_page
= NULL
;
703 kfree(rdev
->badblocks
.page
);
704 rdev
->badblocks
.page
= NULL
;
706 EXPORT_SYMBOL_GPL(md_rdev_clear
);
708 static void super_written(struct bio
*bio
, int error
)
710 struct md_rdev
*rdev
= bio
->bi_private
;
711 struct mddev
*mddev
= rdev
->mddev
;
713 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
714 printk("md: super_written gets error=%d, uptodate=%d\n",
715 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
716 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
717 md_error(mddev
, rdev
);
720 if (atomic_dec_and_test(&mddev
->pending_writes
))
721 wake_up(&mddev
->sb_wait
);
725 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
726 sector_t sector
, int size
, struct page
*page
)
728 /* write first size bytes of page to sector of rdev
729 * Increment mddev->pending_writes before returning
730 * and decrement it on completion, waking up sb_wait
731 * if zero is reached.
732 * If an error occurred, call md_error
734 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
736 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
737 bio
->bi_iter
.bi_sector
= sector
;
738 bio_add_page(bio
, page
, size
, 0);
739 bio
->bi_private
= rdev
;
740 bio
->bi_end_io
= super_written
;
742 atomic_inc(&mddev
->pending_writes
);
743 submit_bio(WRITE_FLUSH_FUA
, bio
);
746 void md_super_wait(struct mddev
*mddev
)
748 /* wait for all superblock writes that were scheduled to complete */
749 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
752 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
753 struct page
*page
, int rw
, bool metadata_op
)
755 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
758 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
759 rdev
->meta_bdev
: rdev
->bdev
;
761 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
762 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
763 (rdev
->mddev
->reshape_backwards
==
764 (sector
>= rdev
->mddev
->reshape_position
)))
765 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
767 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
768 bio_add_page(bio
, page
, size
, 0);
769 submit_bio_wait(rw
, bio
);
771 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
775 EXPORT_SYMBOL_GPL(sync_page_io
);
777 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
779 char b
[BDEVNAME_SIZE
];
784 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
790 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
791 bdevname(rdev
->bdev
,b
));
795 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
797 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
798 sb1
->set_uuid1
== sb2
->set_uuid1
&&
799 sb1
->set_uuid2
== sb2
->set_uuid2
&&
800 sb1
->set_uuid3
== sb2
->set_uuid3
;
803 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
806 mdp_super_t
*tmp1
, *tmp2
;
808 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
809 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
811 if (!tmp1
|| !tmp2
) {
813 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
821 * nr_disks is not constant
826 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
833 static u32
md_csum_fold(u32 csum
)
835 csum
= (csum
& 0xffff) + (csum
>> 16);
836 return (csum
& 0xffff) + (csum
>> 16);
839 static unsigned int calc_sb_csum(mdp_super_t
*sb
)
842 u32
*sb32
= (u32
*)sb
;
844 unsigned int disk_csum
, csum
;
846 disk_csum
= sb
->sb_csum
;
849 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
851 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
854 /* This used to use csum_partial, which was wrong for several
855 * reasons including that different results are returned on
856 * different architectures. It isn't critical that we get exactly
857 * the same return value as before (we always csum_fold before
858 * testing, and that removes any differences). However as we
859 * know that csum_partial always returned a 16bit value on
860 * alphas, do a fold to maximise conformity to previous behaviour.
862 sb
->sb_csum
= md_csum_fold(disk_csum
);
864 sb
->sb_csum
= disk_csum
;
870 * Handle superblock details.
871 * We want to be able to handle multiple superblock formats
872 * so we have a common interface to them all, and an array of
873 * different handlers.
874 * We rely on user-space to write the initial superblock, and support
875 * reading and updating of superblocks.
876 * Interface methods are:
877 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
878 * loads and validates a superblock on dev.
879 * if refdev != NULL, compare superblocks on both devices
881 * 0 - dev has a superblock that is compatible with refdev
882 * 1 - dev has a superblock that is compatible and newer than refdev
883 * so dev should be used as the refdev in future
884 * -EINVAL superblock incompatible or invalid
885 * -othererror e.g. -EIO
887 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
888 * Verify that dev is acceptable into mddev.
889 * The first time, mddev->raid_disks will be 0, and data from
890 * dev should be merged in. Subsequent calls check that dev
891 * is new enough. Return 0 or -EINVAL
893 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
894 * Update the superblock for rdev with data in mddev
895 * This does not write to disc.
901 struct module
*owner
;
902 int (*load_super
)(struct md_rdev
*rdev
,
903 struct md_rdev
*refdev
,
905 int (*validate_super
)(struct mddev
*mddev
,
906 struct md_rdev
*rdev
);
907 void (*sync_super
)(struct mddev
*mddev
,
908 struct md_rdev
*rdev
);
909 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
910 sector_t num_sectors
);
911 int (*allow_new_offset
)(struct md_rdev
*rdev
,
912 unsigned long long new_offset
);
916 * Check that the given mddev has no bitmap.
918 * This function is called from the run method of all personalities that do not
919 * support bitmaps. It prints an error message and returns non-zero if mddev
920 * has a bitmap. Otherwise, it returns 0.
923 int md_check_no_bitmap(struct mddev
*mddev
)
925 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
927 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
928 mdname(mddev
), mddev
->pers
->name
);
931 EXPORT_SYMBOL(md_check_no_bitmap
);
934 * load_super for 0.90.0
936 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
938 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
943 * Calculate the position of the superblock (512byte sectors),
944 * it's at the end of the disk.
946 * It also happens to be a multiple of 4Kb.
948 rdev
->sb_start
= calc_dev_sboffset(rdev
);
950 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
955 bdevname(rdev
->bdev
, b
);
956 sb
= page_address(rdev
->sb_page
);
958 if (sb
->md_magic
!= MD_SB_MAGIC
) {
959 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
964 if (sb
->major_version
!= 0 ||
965 sb
->minor_version
< 90 ||
966 sb
->minor_version
> 91) {
967 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
968 sb
->major_version
, sb
->minor_version
,
973 if (sb
->raid_disks
<= 0)
976 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
977 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
982 rdev
->preferred_minor
= sb
->md_minor
;
983 rdev
->data_offset
= 0;
984 rdev
->new_data_offset
= 0;
985 rdev
->sb_size
= MD_SB_BYTES
;
986 rdev
->badblocks
.shift
= -1;
988 if (sb
->level
== LEVEL_MULTIPATH
)
991 rdev
->desc_nr
= sb
->this_disk
.number
;
997 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
998 if (!uuid_equal(refsb
, sb
)) {
999 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1000 b
, bdevname(refdev
->bdev
,b2
));
1003 if (!sb_equal(refsb
, sb
)) {
1004 printk(KERN_WARNING
"md: %s has same UUID"
1005 " but different superblock to %s\n",
1006 b
, bdevname(refdev
->bdev
, b2
));
1010 ev2
= md_event(refsb
);
1016 rdev
->sectors
= rdev
->sb_start
;
1017 /* Limit to 4TB as metadata cannot record more than that.
1018 * (not needed for Linear and RAID0 as metadata doesn't
1021 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1022 rdev
->sectors
= (2ULL << 32) - 2;
1024 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1025 /* "this cannot possibly happen" ... */
1033 * validate_super for 0.90.0
1035 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1038 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1039 __u64 ev1
= md_event(sb
);
1041 rdev
->raid_disk
= -1;
1042 clear_bit(Faulty
, &rdev
->flags
);
1043 clear_bit(In_sync
, &rdev
->flags
);
1044 clear_bit(Bitmap_sync
, &rdev
->flags
);
1045 clear_bit(WriteMostly
, &rdev
->flags
);
1047 if (mddev
->raid_disks
== 0) {
1048 mddev
->major_version
= 0;
1049 mddev
->minor_version
= sb
->minor_version
;
1050 mddev
->patch_version
= sb
->patch_version
;
1051 mddev
->external
= 0;
1052 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1053 mddev
->ctime
= sb
->ctime
;
1054 mddev
->utime
= sb
->utime
;
1055 mddev
->level
= sb
->level
;
1056 mddev
->clevel
[0] = 0;
1057 mddev
->layout
= sb
->layout
;
1058 mddev
->raid_disks
= sb
->raid_disks
;
1059 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1060 mddev
->events
= ev1
;
1061 mddev
->bitmap_info
.offset
= 0;
1062 mddev
->bitmap_info
.space
= 0;
1063 /* bitmap can use 60 K after the 4K superblocks */
1064 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1065 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1066 mddev
->reshape_backwards
= 0;
1068 if (mddev
->minor_version
>= 91) {
1069 mddev
->reshape_position
= sb
->reshape_position
;
1070 mddev
->delta_disks
= sb
->delta_disks
;
1071 mddev
->new_level
= sb
->new_level
;
1072 mddev
->new_layout
= sb
->new_layout
;
1073 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1074 if (mddev
->delta_disks
< 0)
1075 mddev
->reshape_backwards
= 1;
1077 mddev
->reshape_position
= MaxSector
;
1078 mddev
->delta_disks
= 0;
1079 mddev
->new_level
= mddev
->level
;
1080 mddev
->new_layout
= mddev
->layout
;
1081 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1084 if (sb
->state
& (1<<MD_SB_CLEAN
))
1085 mddev
->recovery_cp
= MaxSector
;
1087 if (sb
->events_hi
== sb
->cp_events_hi
&&
1088 sb
->events_lo
== sb
->cp_events_lo
) {
1089 mddev
->recovery_cp
= sb
->recovery_cp
;
1091 mddev
->recovery_cp
= 0;
1094 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1095 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1096 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1097 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1099 mddev
->max_disks
= MD_SB_DISKS
;
1101 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1102 mddev
->bitmap_info
.file
== NULL
) {
1103 mddev
->bitmap_info
.offset
=
1104 mddev
->bitmap_info
.default_offset
;
1105 mddev
->bitmap_info
.space
=
1106 mddev
->bitmap_info
.default_space
;
1109 } else if (mddev
->pers
== NULL
) {
1110 /* Insist on good event counter while assembling, except
1111 * for spares (which don't need an event count) */
1113 if (sb
->disks
[rdev
->desc_nr
].state
& (
1114 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1115 if (ev1
< mddev
->events
)
1117 } else if (mddev
->bitmap
) {
1118 /* if adding to array with a bitmap, then we can accept an
1119 * older device ... but not too old.
1121 if (ev1
< mddev
->bitmap
->events_cleared
)
1123 if (ev1
< mddev
->events
)
1124 set_bit(Bitmap_sync
, &rdev
->flags
);
1126 if (ev1
< mddev
->events
)
1127 /* just a hot-add of a new device, leave raid_disk at -1 */
1131 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1132 desc
= sb
->disks
+ rdev
->desc_nr
;
1134 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1135 set_bit(Faulty
, &rdev
->flags
);
1136 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1137 desc->raid_disk < mddev->raid_disks */) {
1138 set_bit(In_sync
, &rdev
->flags
);
1139 rdev
->raid_disk
= desc
->raid_disk
;
1140 rdev
->saved_raid_disk
= desc
->raid_disk
;
1141 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1142 /* active but not in sync implies recovery up to
1143 * reshape position. We don't know exactly where
1144 * that is, so set to zero for now */
1145 if (mddev
->minor_version
>= 91) {
1146 rdev
->recovery_offset
= 0;
1147 rdev
->raid_disk
= desc
->raid_disk
;
1150 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1151 set_bit(WriteMostly
, &rdev
->flags
);
1152 } else /* MULTIPATH are always insync */
1153 set_bit(In_sync
, &rdev
->flags
);
1158 * sync_super for 0.90.0
1160 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1163 struct md_rdev
*rdev2
;
1164 int next_spare
= mddev
->raid_disks
;
1166 /* make rdev->sb match mddev data..
1169 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1170 * 3/ any empty disks < next_spare become removed
1172 * disks[0] gets initialised to REMOVED because
1173 * we cannot be sure from other fields if it has
1174 * been initialised or not.
1177 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1179 rdev
->sb_size
= MD_SB_BYTES
;
1181 sb
= page_address(rdev
->sb_page
);
1183 memset(sb
, 0, sizeof(*sb
));
1185 sb
->md_magic
= MD_SB_MAGIC
;
1186 sb
->major_version
= mddev
->major_version
;
1187 sb
->patch_version
= mddev
->patch_version
;
1188 sb
->gvalid_words
= 0; /* ignored */
1189 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1190 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1191 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1192 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1194 sb
->ctime
= mddev
->ctime
;
1195 sb
->level
= mddev
->level
;
1196 sb
->size
= mddev
->dev_sectors
/ 2;
1197 sb
->raid_disks
= mddev
->raid_disks
;
1198 sb
->md_minor
= mddev
->md_minor
;
1199 sb
->not_persistent
= 0;
1200 sb
->utime
= mddev
->utime
;
1202 sb
->events_hi
= (mddev
->events
>>32);
1203 sb
->events_lo
= (u32
)mddev
->events
;
1205 if (mddev
->reshape_position
== MaxSector
)
1206 sb
->minor_version
= 90;
1208 sb
->minor_version
= 91;
1209 sb
->reshape_position
= mddev
->reshape_position
;
1210 sb
->new_level
= mddev
->new_level
;
1211 sb
->delta_disks
= mddev
->delta_disks
;
1212 sb
->new_layout
= mddev
->new_layout
;
1213 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1215 mddev
->minor_version
= sb
->minor_version
;
1218 sb
->recovery_cp
= mddev
->recovery_cp
;
1219 sb
->cp_events_hi
= (mddev
->events
>>32);
1220 sb
->cp_events_lo
= (u32
)mddev
->events
;
1221 if (mddev
->recovery_cp
== MaxSector
)
1222 sb
->state
= (1<< MD_SB_CLEAN
);
1224 sb
->recovery_cp
= 0;
1226 sb
->layout
= mddev
->layout
;
1227 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1229 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1230 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1232 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1233 rdev_for_each(rdev2
, mddev
) {
1236 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1238 if (rdev2
->raid_disk
>= 0 &&
1239 sb
->minor_version
>= 91)
1240 /* we have nowhere to store the recovery_offset,
1241 * but if it is not below the reshape_position,
1242 * we can piggy-back on that.
1245 if (rdev2
->raid_disk
< 0 ||
1246 test_bit(Faulty
, &rdev2
->flags
))
1249 desc_nr
= rdev2
->raid_disk
;
1251 desc_nr
= next_spare
++;
1252 rdev2
->desc_nr
= desc_nr
;
1253 d
= &sb
->disks
[rdev2
->desc_nr
];
1255 d
->number
= rdev2
->desc_nr
;
1256 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1257 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1259 d
->raid_disk
= rdev2
->raid_disk
;
1261 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1262 if (test_bit(Faulty
, &rdev2
->flags
))
1263 d
->state
= (1<<MD_DISK_FAULTY
);
1264 else if (is_active
) {
1265 d
->state
= (1<<MD_DISK_ACTIVE
);
1266 if (test_bit(In_sync
, &rdev2
->flags
))
1267 d
->state
|= (1<<MD_DISK_SYNC
);
1275 if (test_bit(WriteMostly
, &rdev2
->flags
))
1276 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1278 /* now set the "removed" and "faulty" bits on any missing devices */
1279 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1280 mdp_disk_t
*d
= &sb
->disks
[i
];
1281 if (d
->state
== 0 && d
->number
== 0) {
1284 d
->state
= (1<<MD_DISK_REMOVED
);
1285 d
->state
|= (1<<MD_DISK_FAULTY
);
1289 sb
->nr_disks
= nr_disks
;
1290 sb
->active_disks
= active
;
1291 sb
->working_disks
= working
;
1292 sb
->failed_disks
= failed
;
1293 sb
->spare_disks
= spare
;
1295 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1296 sb
->sb_csum
= calc_sb_csum(sb
);
1300 * rdev_size_change for 0.90.0
1302 static unsigned long long
1303 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1305 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1306 return 0; /* component must fit device */
1307 if (rdev
->mddev
->bitmap_info
.offset
)
1308 return 0; /* can't move bitmap */
1309 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1310 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1311 num_sectors
= rdev
->sb_start
;
1312 /* Limit to 4TB as metadata cannot record more than that.
1313 * 4TB == 2^32 KB, or 2*2^32 sectors.
1315 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1316 num_sectors
= (2ULL << 32) - 2;
1317 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1319 md_super_wait(rdev
->mddev
);
1324 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1326 /* non-zero offset changes not possible with v0.90 */
1327 return new_offset
== 0;
1331 * version 1 superblock
1334 static __le32
calc_sb_1_csum(struct mdp_superblock_1
*sb
)
1338 unsigned long long newcsum
;
1339 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1340 __le32
*isuper
= (__le32
*)sb
;
1342 disk_csum
= sb
->sb_csum
;
1345 for (; size
>= 4; size
-= 4)
1346 newcsum
+= le32_to_cpu(*isuper
++);
1349 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1351 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1352 sb
->sb_csum
= disk_csum
;
1353 return cpu_to_le32(csum
);
1356 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1358 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1360 struct mdp_superblock_1
*sb
;
1364 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1368 * Calculate the position of the superblock in 512byte sectors.
1369 * It is always aligned to a 4K boundary and
1370 * depeding on minor_version, it can be:
1371 * 0: At least 8K, but less than 12K, from end of device
1372 * 1: At start of device
1373 * 2: 4K from start of device.
1375 switch(minor_version
) {
1377 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1379 sb_start
&= ~(sector_t
)(4*2-1);
1390 rdev
->sb_start
= sb_start
;
1392 /* superblock is rarely larger than 1K, but it can be larger,
1393 * and it is safe to read 4k, so we do that
1395 ret
= read_disk_sb(rdev
, 4096);
1396 if (ret
) return ret
;
1398 sb
= page_address(rdev
->sb_page
);
1400 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1401 sb
->major_version
!= cpu_to_le32(1) ||
1402 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1403 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1404 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1407 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1408 printk("md: invalid superblock checksum on %s\n",
1409 bdevname(rdev
->bdev
,b
));
1412 if (le64_to_cpu(sb
->data_size
) < 10) {
1413 printk("md: data_size too small on %s\n",
1414 bdevname(rdev
->bdev
,b
));
1419 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1420 /* Some padding is non-zero, might be a new feature */
1423 rdev
->preferred_minor
= 0xffff;
1424 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1425 rdev
->new_data_offset
= rdev
->data_offset
;
1426 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1427 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1428 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1429 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1431 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1432 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1433 if (rdev
->sb_size
& bmask
)
1434 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1437 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1440 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1443 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1446 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1448 if (!rdev
->bb_page
) {
1449 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1453 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1454 rdev
->badblocks
.count
== 0) {
1455 /* need to load the bad block list.
1456 * Currently we limit it to one page.
1462 int sectors
= le16_to_cpu(sb
->bblog_size
);
1463 if (sectors
> (PAGE_SIZE
/ 512))
1465 offset
= le32_to_cpu(sb
->bblog_offset
);
1468 bb_sector
= (long long)offset
;
1469 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1470 rdev
->bb_page
, READ
, true))
1472 bbp
= (u64
*)page_address(rdev
->bb_page
);
1473 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1474 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1475 u64 bb
= le64_to_cpu(*bbp
);
1476 int count
= bb
& (0x3ff);
1477 u64 sector
= bb
>> 10;
1478 sector
<<= sb
->bblog_shift
;
1479 count
<<= sb
->bblog_shift
;
1482 if (md_set_badblocks(&rdev
->badblocks
,
1483 sector
, count
, 1) == 0)
1486 } else if (sb
->bblog_offset
!= 0)
1487 rdev
->badblocks
.shift
= 0;
1493 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1495 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1496 sb
->level
!= refsb
->level
||
1497 sb
->layout
!= refsb
->layout
||
1498 sb
->chunksize
!= refsb
->chunksize
) {
1499 printk(KERN_WARNING
"md: %s has strangely different"
1500 " superblock to %s\n",
1501 bdevname(rdev
->bdev
,b
),
1502 bdevname(refdev
->bdev
,b2
));
1505 ev1
= le64_to_cpu(sb
->events
);
1506 ev2
= le64_to_cpu(refsb
->events
);
1513 if (minor_version
) {
1514 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1515 sectors
-= rdev
->data_offset
;
1517 sectors
= rdev
->sb_start
;
1518 if (sectors
< le64_to_cpu(sb
->data_size
))
1520 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1524 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1526 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1527 __u64 ev1
= le64_to_cpu(sb
->events
);
1529 rdev
->raid_disk
= -1;
1530 clear_bit(Faulty
, &rdev
->flags
);
1531 clear_bit(In_sync
, &rdev
->flags
);
1532 clear_bit(Bitmap_sync
, &rdev
->flags
);
1533 clear_bit(WriteMostly
, &rdev
->flags
);
1535 if (mddev
->raid_disks
== 0) {
1536 mddev
->major_version
= 1;
1537 mddev
->patch_version
= 0;
1538 mddev
->external
= 0;
1539 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1540 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1541 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1542 mddev
->level
= le32_to_cpu(sb
->level
);
1543 mddev
->clevel
[0] = 0;
1544 mddev
->layout
= le32_to_cpu(sb
->layout
);
1545 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1546 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1547 mddev
->events
= ev1
;
1548 mddev
->bitmap_info
.offset
= 0;
1549 mddev
->bitmap_info
.space
= 0;
1550 /* Default location for bitmap is 1K after superblock
1551 * using 3K - total of 4K
1553 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1554 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1555 mddev
->reshape_backwards
= 0;
1557 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1558 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1560 mddev
->max_disks
= (4096-256)/2;
1562 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1563 mddev
->bitmap_info
.file
== NULL
) {
1564 mddev
->bitmap_info
.offset
=
1565 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1566 /* Metadata doesn't record how much space is available.
1567 * For 1.0, we assume we can use up to the superblock
1568 * if before, else to 4K beyond superblock.
1569 * For others, assume no change is possible.
1571 if (mddev
->minor_version
> 0)
1572 mddev
->bitmap_info
.space
= 0;
1573 else if (mddev
->bitmap_info
.offset
> 0)
1574 mddev
->bitmap_info
.space
=
1575 8 - mddev
->bitmap_info
.offset
;
1577 mddev
->bitmap_info
.space
=
1578 -mddev
->bitmap_info
.offset
;
1581 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1582 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1583 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1584 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1585 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1586 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1587 if (mddev
->delta_disks
< 0 ||
1588 (mddev
->delta_disks
== 0 &&
1589 (le32_to_cpu(sb
->feature_map
)
1590 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1591 mddev
->reshape_backwards
= 1;
1593 mddev
->reshape_position
= MaxSector
;
1594 mddev
->delta_disks
= 0;
1595 mddev
->new_level
= mddev
->level
;
1596 mddev
->new_layout
= mddev
->layout
;
1597 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1600 } else if (mddev
->pers
== NULL
) {
1601 /* Insist of good event counter while assembling, except for
1602 * spares (which don't need an event count) */
1604 if (rdev
->desc_nr
>= 0 &&
1605 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1606 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1607 if (ev1
< mddev
->events
)
1609 } else if (mddev
->bitmap
) {
1610 /* If adding to array with a bitmap, then we can accept an
1611 * older device, but not too old.
1613 if (ev1
< mddev
->bitmap
->events_cleared
)
1615 if (ev1
< mddev
->events
)
1616 set_bit(Bitmap_sync
, &rdev
->flags
);
1618 if (ev1
< mddev
->events
)
1619 /* just a hot-add of a new device, leave raid_disk at -1 */
1622 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1624 if (rdev
->desc_nr
< 0 ||
1625 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1629 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1631 case 0xffff: /* spare */
1633 case 0xfffe: /* faulty */
1634 set_bit(Faulty
, &rdev
->flags
);
1637 rdev
->saved_raid_disk
= role
;
1638 if ((le32_to_cpu(sb
->feature_map
) &
1639 MD_FEATURE_RECOVERY_OFFSET
)) {
1640 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1641 if (!(le32_to_cpu(sb
->feature_map
) &
1642 MD_FEATURE_RECOVERY_BITMAP
))
1643 rdev
->saved_raid_disk
= -1;
1645 set_bit(In_sync
, &rdev
->flags
);
1646 rdev
->raid_disk
= role
;
1649 if (sb
->devflags
& WriteMostly1
)
1650 set_bit(WriteMostly
, &rdev
->flags
);
1651 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1652 set_bit(Replacement
, &rdev
->flags
);
1653 } else /* MULTIPATH are always insync */
1654 set_bit(In_sync
, &rdev
->flags
);
1659 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1661 struct mdp_superblock_1
*sb
;
1662 struct md_rdev
*rdev2
;
1664 /* make rdev->sb match mddev and rdev data. */
1666 sb
= page_address(rdev
->sb_page
);
1668 sb
->feature_map
= 0;
1670 sb
->recovery_offset
= cpu_to_le64(0);
1671 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1673 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1674 sb
->events
= cpu_to_le64(mddev
->events
);
1676 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1678 sb
->resync_offset
= cpu_to_le64(0);
1680 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1682 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1683 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1684 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1685 sb
->level
= cpu_to_le32(mddev
->level
);
1686 sb
->layout
= cpu_to_le32(mddev
->layout
);
1688 if (test_bit(WriteMostly
, &rdev
->flags
))
1689 sb
->devflags
|= WriteMostly1
;
1691 sb
->devflags
&= ~WriteMostly1
;
1692 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1693 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1695 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1696 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1697 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1700 if (rdev
->raid_disk
>= 0 &&
1701 !test_bit(In_sync
, &rdev
->flags
)) {
1703 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1704 sb
->recovery_offset
=
1705 cpu_to_le64(rdev
->recovery_offset
);
1706 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1708 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1710 if (test_bit(Replacement
, &rdev
->flags
))
1712 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1714 if (mddev
->reshape_position
!= MaxSector
) {
1715 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1716 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1717 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1718 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1719 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1720 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1721 if (mddev
->delta_disks
== 0 &&
1722 mddev
->reshape_backwards
)
1724 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1725 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1727 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1728 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1729 - rdev
->data_offset
));
1733 if (rdev
->badblocks
.count
== 0)
1734 /* Nothing to do for bad blocks*/ ;
1735 else if (sb
->bblog_offset
== 0)
1736 /* Cannot record bad blocks on this device */
1737 md_error(mddev
, rdev
);
1739 struct badblocks
*bb
= &rdev
->badblocks
;
1740 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1742 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1747 seq
= read_seqbegin(&bb
->lock
);
1749 memset(bbp
, 0xff, PAGE_SIZE
);
1751 for (i
= 0 ; i
< bb
->count
; i
++) {
1752 u64 internal_bb
= p
[i
];
1753 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1754 | BB_LEN(internal_bb
));
1755 bbp
[i
] = cpu_to_le64(store_bb
);
1758 if (read_seqretry(&bb
->lock
, seq
))
1761 bb
->sector
= (rdev
->sb_start
+
1762 (int)le32_to_cpu(sb
->bblog_offset
));
1763 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1768 rdev_for_each(rdev2
, mddev
)
1769 if (rdev2
->desc_nr
+1 > max_dev
)
1770 max_dev
= rdev2
->desc_nr
+1;
1772 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1774 sb
->max_dev
= cpu_to_le32(max_dev
);
1775 rdev
->sb_size
= max_dev
* 2 + 256;
1776 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1777 if (rdev
->sb_size
& bmask
)
1778 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1780 max_dev
= le32_to_cpu(sb
->max_dev
);
1782 for (i
=0; i
<max_dev
;i
++)
1783 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1785 rdev_for_each(rdev2
, mddev
) {
1787 if (test_bit(Faulty
, &rdev2
->flags
))
1788 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1789 else if (test_bit(In_sync
, &rdev2
->flags
))
1790 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1791 else if (rdev2
->raid_disk
>= 0)
1792 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1794 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1797 sb
->sb_csum
= calc_sb_1_csum(sb
);
1800 static unsigned long long
1801 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1803 struct mdp_superblock_1
*sb
;
1804 sector_t max_sectors
;
1805 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1806 return 0; /* component must fit device */
1807 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1808 return 0; /* too confusing */
1809 if (rdev
->sb_start
< rdev
->data_offset
) {
1810 /* minor versions 1 and 2; superblock before data */
1811 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1812 max_sectors
-= rdev
->data_offset
;
1813 if (!num_sectors
|| num_sectors
> max_sectors
)
1814 num_sectors
= max_sectors
;
1815 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1816 /* minor version 0 with bitmap we can't move */
1819 /* minor version 0; superblock after data */
1821 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1822 sb_start
&= ~(sector_t
)(4*2 - 1);
1823 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1824 if (!num_sectors
|| num_sectors
> max_sectors
)
1825 num_sectors
= max_sectors
;
1826 rdev
->sb_start
= sb_start
;
1828 sb
= page_address(rdev
->sb_page
);
1829 sb
->data_size
= cpu_to_le64(num_sectors
);
1830 sb
->super_offset
= rdev
->sb_start
;
1831 sb
->sb_csum
= calc_sb_1_csum(sb
);
1832 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1834 md_super_wait(rdev
->mddev
);
1840 super_1_allow_new_offset(struct md_rdev
*rdev
,
1841 unsigned long long new_offset
)
1843 /* All necessary checks on new >= old have been done */
1844 struct bitmap
*bitmap
;
1845 if (new_offset
>= rdev
->data_offset
)
1848 /* with 1.0 metadata, there is no metadata to tread on
1849 * so we can always move back */
1850 if (rdev
->mddev
->minor_version
== 0)
1853 /* otherwise we must be sure not to step on
1854 * any metadata, so stay:
1855 * 36K beyond start of superblock
1856 * beyond end of badblocks
1857 * beyond write-intent bitmap
1859 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1861 bitmap
= rdev
->mddev
->bitmap
;
1862 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1863 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1864 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1866 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1872 static struct super_type super_types
[] = {
1875 .owner
= THIS_MODULE
,
1876 .load_super
= super_90_load
,
1877 .validate_super
= super_90_validate
,
1878 .sync_super
= super_90_sync
,
1879 .rdev_size_change
= super_90_rdev_size_change
,
1880 .allow_new_offset
= super_90_allow_new_offset
,
1884 .owner
= THIS_MODULE
,
1885 .load_super
= super_1_load
,
1886 .validate_super
= super_1_validate
,
1887 .sync_super
= super_1_sync
,
1888 .rdev_size_change
= super_1_rdev_size_change
,
1889 .allow_new_offset
= super_1_allow_new_offset
,
1893 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1895 if (mddev
->sync_super
) {
1896 mddev
->sync_super(mddev
, rdev
);
1900 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1902 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1905 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1907 struct md_rdev
*rdev
, *rdev2
;
1910 rdev_for_each_rcu(rdev
, mddev1
)
1911 rdev_for_each_rcu(rdev2
, mddev2
)
1912 if (rdev
->bdev
->bd_contains
==
1913 rdev2
->bdev
->bd_contains
) {
1921 static LIST_HEAD(pending_raid_disks
);
1924 * Try to register data integrity profile for an mddev
1926 * This is called when an array is started and after a disk has been kicked
1927 * from the array. It only succeeds if all working and active component devices
1928 * are integrity capable with matching profiles.
1930 int md_integrity_register(struct mddev
*mddev
)
1932 struct md_rdev
*rdev
, *reference
= NULL
;
1934 if (list_empty(&mddev
->disks
))
1935 return 0; /* nothing to do */
1936 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1937 return 0; /* shouldn't register, or already is */
1938 rdev_for_each(rdev
, mddev
) {
1939 /* skip spares and non-functional disks */
1940 if (test_bit(Faulty
, &rdev
->flags
))
1942 if (rdev
->raid_disk
< 0)
1945 /* Use the first rdev as the reference */
1949 /* does this rdev's profile match the reference profile? */
1950 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1951 rdev
->bdev
->bd_disk
) < 0)
1954 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1957 * All component devices are integrity capable and have matching
1958 * profiles, register the common profile for the md device.
1960 if (blk_integrity_register(mddev
->gendisk
,
1961 bdev_get_integrity(reference
->bdev
)) != 0) {
1962 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1966 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1967 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1968 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1974 EXPORT_SYMBOL(md_integrity_register
);
1976 /* Disable data integrity if non-capable/non-matching disk is being added */
1977 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
1979 struct blk_integrity
*bi_rdev
;
1980 struct blk_integrity
*bi_mddev
;
1982 if (!mddev
->gendisk
)
1985 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1986 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1988 if (!bi_mddev
) /* nothing to do */
1990 if (rdev
->raid_disk
< 0) /* skip spares */
1992 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1993 rdev
->bdev
->bd_disk
) >= 0)
1995 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1996 blk_integrity_unregister(mddev
->gendisk
);
1998 EXPORT_SYMBOL(md_integrity_add_rdev
);
2000 static int bind_rdev_to_array(struct md_rdev
*rdev
, struct mddev
*mddev
)
2002 char b
[BDEVNAME_SIZE
];
2007 /* prevent duplicates */
2008 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2011 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2012 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2013 rdev
->sectors
< mddev
->dev_sectors
)) {
2015 /* Cannot change size, so fail
2016 * If mddev->level <= 0, then we don't care
2017 * about aligning sizes (e.g. linear)
2019 if (mddev
->level
> 0)
2022 mddev
->dev_sectors
= rdev
->sectors
;
2025 /* Verify rdev->desc_nr is unique.
2026 * If it is -1, assign a free number, else
2027 * check number is not in use
2030 if (rdev
->desc_nr
< 0) {
2033 choice
= mddev
->raid_disks
;
2034 while (find_rdev_nr_rcu(mddev
, choice
))
2036 rdev
->desc_nr
= choice
;
2038 if (find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2044 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2045 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2046 mdname(mddev
), mddev
->max_disks
);
2049 bdevname(rdev
->bdev
,b
);
2050 while ( (s
=strchr(b
, '/')) != NULL
)
2053 rdev
->mddev
= mddev
;
2054 printk(KERN_INFO
"md: bind<%s>\n", b
);
2056 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2059 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2060 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2061 /* failure here is OK */;
2062 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2064 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2065 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2067 /* May as well allow recovery to be retried once */
2068 mddev
->recovery_disabled
++;
2073 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2078 static void md_delayed_delete(struct work_struct
*ws
)
2080 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2081 kobject_del(&rdev
->kobj
);
2082 kobject_put(&rdev
->kobj
);
2085 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2087 char b
[BDEVNAME_SIZE
];
2089 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2090 list_del_rcu(&rdev
->same_set
);
2091 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2093 sysfs_remove_link(&rdev
->kobj
, "block");
2094 sysfs_put(rdev
->sysfs_state
);
2095 rdev
->sysfs_state
= NULL
;
2096 rdev
->badblocks
.count
= 0;
2097 /* We need to delay this, otherwise we can deadlock when
2098 * writing to 'remove' to "dev/state". We also need
2099 * to delay it due to rcu usage.
2102 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2103 kobject_get(&rdev
->kobj
);
2104 queue_work(md_misc_wq
, &rdev
->del_work
);
2108 * prevent the device from being mounted, repartitioned or
2109 * otherwise reused by a RAID array (or any other kernel
2110 * subsystem), by bd_claiming the device.
2112 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2115 struct block_device
*bdev
;
2116 char b
[BDEVNAME_SIZE
];
2118 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2119 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2121 printk(KERN_ERR
"md: could not open %s.\n",
2122 __bdevname(dev
, b
));
2123 return PTR_ERR(bdev
);
2129 static void unlock_rdev(struct md_rdev
*rdev
)
2131 struct block_device
*bdev
= rdev
->bdev
;
2133 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2136 void md_autodetect_dev(dev_t dev
);
2138 static void export_rdev(struct md_rdev
*rdev
)
2140 char b
[BDEVNAME_SIZE
];
2142 printk(KERN_INFO
"md: export_rdev(%s)\n",
2143 bdevname(rdev
->bdev
,b
));
2144 md_rdev_clear(rdev
);
2146 if (test_bit(AutoDetected
, &rdev
->flags
))
2147 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2150 kobject_put(&rdev
->kobj
);
2153 static void kick_rdev_from_array(struct md_rdev
*rdev
)
2155 unbind_rdev_from_array(rdev
);
2159 static void export_array(struct mddev
*mddev
)
2161 struct md_rdev
*rdev
;
2163 while (!list_empty(&mddev
->disks
)) {
2164 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2166 kick_rdev_from_array(rdev
);
2168 mddev
->raid_disks
= 0;
2169 mddev
->major_version
= 0;
2172 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2174 /* Update each superblock (in-memory image), but
2175 * if we are allowed to, skip spares which already
2176 * have the right event counter, or have one earlier
2177 * (which would mean they aren't being marked as dirty
2178 * with the rest of the array)
2180 struct md_rdev
*rdev
;
2181 rdev_for_each(rdev
, mddev
) {
2182 if (rdev
->sb_events
== mddev
->events
||
2184 rdev
->raid_disk
< 0 &&
2185 rdev
->sb_events
+1 == mddev
->events
)) {
2186 /* Don't update this superblock */
2187 rdev
->sb_loaded
= 2;
2189 sync_super(mddev
, rdev
);
2190 rdev
->sb_loaded
= 1;
2195 static void md_update_sb(struct mddev
*mddev
, int force_change
)
2197 struct md_rdev
*rdev
;
2200 int any_badblocks_changed
= 0;
2204 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2208 /* First make sure individual recovery_offsets are correct */
2209 rdev_for_each(rdev
, mddev
) {
2210 if (rdev
->raid_disk
>= 0 &&
2211 mddev
->delta_disks
>= 0 &&
2212 !test_bit(In_sync
, &rdev
->flags
) &&
2213 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2214 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2217 if (!mddev
->persistent
) {
2218 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2219 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2220 if (!mddev
->external
) {
2221 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2222 rdev_for_each(rdev
, mddev
) {
2223 if (rdev
->badblocks
.changed
) {
2224 rdev
->badblocks
.changed
= 0;
2225 md_ack_all_badblocks(&rdev
->badblocks
);
2226 md_error(mddev
, rdev
);
2228 clear_bit(Blocked
, &rdev
->flags
);
2229 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2230 wake_up(&rdev
->blocked_wait
);
2233 wake_up(&mddev
->sb_wait
);
2237 spin_lock_irq(&mddev
->write_lock
);
2239 mddev
->utime
= get_seconds();
2241 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2243 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2244 /* just a clean<-> dirty transition, possibly leave spares alone,
2245 * though if events isn't the right even/odd, we will have to do
2251 if (mddev
->degraded
)
2252 /* If the array is degraded, then skipping spares is both
2253 * dangerous and fairly pointless.
2254 * Dangerous because a device that was removed from the array
2255 * might have a event_count that still looks up-to-date,
2256 * so it can be re-added without a resync.
2257 * Pointless because if there are any spares to skip,
2258 * then a recovery will happen and soon that array won't
2259 * be degraded any more and the spare can go back to sleep then.
2263 sync_req
= mddev
->in_sync
;
2265 /* If this is just a dirty<->clean transition, and the array is clean
2266 * and 'events' is odd, we can roll back to the previous clean state */
2268 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2269 && mddev
->can_decrease_events
2270 && mddev
->events
!= 1) {
2272 mddev
->can_decrease_events
= 0;
2274 /* otherwise we have to go forward and ... */
2276 mddev
->can_decrease_events
= nospares
;
2280 * This 64-bit counter should never wrap.
2281 * Either we are in around ~1 trillion A.C., assuming
2282 * 1 reboot per second, or we have a bug...
2284 WARN_ON(mddev
->events
== 0);
2286 rdev_for_each(rdev
, mddev
) {
2287 if (rdev
->badblocks
.changed
)
2288 any_badblocks_changed
++;
2289 if (test_bit(Faulty
, &rdev
->flags
))
2290 set_bit(FaultRecorded
, &rdev
->flags
);
2293 sync_sbs(mddev
, nospares
);
2294 spin_unlock_irq(&mddev
->write_lock
);
2296 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2297 mdname(mddev
), mddev
->in_sync
);
2299 bitmap_update_sb(mddev
->bitmap
);
2300 rdev_for_each(rdev
, mddev
) {
2301 char b
[BDEVNAME_SIZE
];
2303 if (rdev
->sb_loaded
!= 1)
2304 continue; /* no noise on spare devices */
2306 if (!test_bit(Faulty
, &rdev
->flags
)) {
2307 md_super_write(mddev
,rdev
,
2308 rdev
->sb_start
, rdev
->sb_size
,
2310 pr_debug("md: (write) %s's sb offset: %llu\n",
2311 bdevname(rdev
->bdev
, b
),
2312 (unsigned long long)rdev
->sb_start
);
2313 rdev
->sb_events
= mddev
->events
;
2314 if (rdev
->badblocks
.size
) {
2315 md_super_write(mddev
, rdev
,
2316 rdev
->badblocks
.sector
,
2317 rdev
->badblocks
.size
<< 9,
2319 rdev
->badblocks
.size
= 0;
2323 pr_debug("md: %s (skipping faulty)\n",
2324 bdevname(rdev
->bdev
, b
));
2326 if (mddev
->level
== LEVEL_MULTIPATH
)
2327 /* only need to write one superblock... */
2330 md_super_wait(mddev
);
2331 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2333 spin_lock_irq(&mddev
->write_lock
);
2334 if (mddev
->in_sync
!= sync_req
||
2335 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2336 /* have to write it out again */
2337 spin_unlock_irq(&mddev
->write_lock
);
2340 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2341 spin_unlock_irq(&mddev
->write_lock
);
2342 wake_up(&mddev
->sb_wait
);
2343 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2344 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2346 rdev_for_each(rdev
, mddev
) {
2347 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2348 clear_bit(Blocked
, &rdev
->flags
);
2350 if (any_badblocks_changed
)
2351 md_ack_all_badblocks(&rdev
->badblocks
);
2352 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2353 wake_up(&rdev
->blocked_wait
);
2357 /* words written to sysfs files may, or may not, be \n terminated.
2358 * We want to accept with case. For this we use cmd_match.
2360 static int cmd_match(const char *cmd
, const char *str
)
2362 /* See if cmd, written into a sysfs file, matches
2363 * str. They must either be the same, or cmd can
2364 * have a trailing newline
2366 while (*cmd
&& *str
&& *cmd
== *str
) {
2377 struct rdev_sysfs_entry
{
2378 struct attribute attr
;
2379 ssize_t (*show
)(struct md_rdev
*, char *);
2380 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2384 state_show(struct md_rdev
*rdev
, char *page
)
2389 if (test_bit(Faulty
, &rdev
->flags
) ||
2390 rdev
->badblocks
.unacked_exist
) {
2391 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2394 if (test_bit(In_sync
, &rdev
->flags
)) {
2395 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2398 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2399 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2402 if (test_bit(Blocked
, &rdev
->flags
) ||
2403 (rdev
->badblocks
.unacked_exist
2404 && !test_bit(Faulty
, &rdev
->flags
))) {
2405 len
+= sprintf(page
+len
, "%sblocked", sep
);
2408 if (!test_bit(Faulty
, &rdev
->flags
) &&
2409 !test_bit(In_sync
, &rdev
->flags
)) {
2410 len
+= sprintf(page
+len
, "%sspare", sep
);
2413 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2414 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2417 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2418 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2421 if (test_bit(Replacement
, &rdev
->flags
)) {
2422 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2426 return len
+sprintf(page
+len
, "\n");
2430 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2433 * faulty - simulates an error
2434 * remove - disconnects the device
2435 * writemostly - sets write_mostly
2436 * -writemostly - clears write_mostly
2437 * blocked - sets the Blocked flags
2438 * -blocked - clears the Blocked and possibly simulates an error
2439 * insync - sets Insync providing device isn't active
2440 * -insync - clear Insync for a device with a slot assigned,
2441 * so that it gets rebuilt based on bitmap
2442 * write_error - sets WriteErrorSeen
2443 * -write_error - clears WriteErrorSeen
2446 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2447 md_error(rdev
->mddev
, rdev
);
2448 if (test_bit(Faulty
, &rdev
->flags
))
2452 } else if (cmd_match(buf
, "remove")) {
2453 if (rdev
->raid_disk
>= 0)
2456 struct mddev
*mddev
= rdev
->mddev
;
2457 kick_rdev_from_array(rdev
);
2459 md_update_sb(mddev
, 1);
2460 md_new_event(mddev
);
2463 } else if (cmd_match(buf
, "writemostly")) {
2464 set_bit(WriteMostly
, &rdev
->flags
);
2466 } else if (cmd_match(buf
, "-writemostly")) {
2467 clear_bit(WriteMostly
, &rdev
->flags
);
2469 } else if (cmd_match(buf
, "blocked")) {
2470 set_bit(Blocked
, &rdev
->flags
);
2472 } else if (cmd_match(buf
, "-blocked")) {
2473 if (!test_bit(Faulty
, &rdev
->flags
) &&
2474 rdev
->badblocks
.unacked_exist
) {
2475 /* metadata handler doesn't understand badblocks,
2476 * so we need to fail the device
2478 md_error(rdev
->mddev
, rdev
);
2480 clear_bit(Blocked
, &rdev
->flags
);
2481 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2482 wake_up(&rdev
->blocked_wait
);
2483 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2484 md_wakeup_thread(rdev
->mddev
->thread
);
2487 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2488 set_bit(In_sync
, &rdev
->flags
);
2490 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0) {
2491 if (rdev
->mddev
->pers
== NULL
) {
2492 clear_bit(In_sync
, &rdev
->flags
);
2493 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2494 rdev
->raid_disk
= -1;
2497 } else if (cmd_match(buf
, "write_error")) {
2498 set_bit(WriteErrorSeen
, &rdev
->flags
);
2500 } else if (cmd_match(buf
, "-write_error")) {
2501 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2503 } else if (cmd_match(buf
, "want_replacement")) {
2504 /* Any non-spare device that is not a replacement can
2505 * become want_replacement at any time, but we then need to
2506 * check if recovery is needed.
2508 if (rdev
->raid_disk
>= 0 &&
2509 !test_bit(Replacement
, &rdev
->flags
))
2510 set_bit(WantReplacement
, &rdev
->flags
);
2511 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2512 md_wakeup_thread(rdev
->mddev
->thread
);
2514 } else if (cmd_match(buf
, "-want_replacement")) {
2515 /* Clearing 'want_replacement' is always allowed.
2516 * Once replacements starts it is too late though.
2519 clear_bit(WantReplacement
, &rdev
->flags
);
2520 } else if (cmd_match(buf
, "replacement")) {
2521 /* Can only set a device as a replacement when array has not
2522 * yet been started. Once running, replacement is automatic
2523 * from spares, or by assigning 'slot'.
2525 if (rdev
->mddev
->pers
)
2528 set_bit(Replacement
, &rdev
->flags
);
2531 } else if (cmd_match(buf
, "-replacement")) {
2532 /* Similarly, can only clear Replacement before start */
2533 if (rdev
->mddev
->pers
)
2536 clear_bit(Replacement
, &rdev
->flags
);
2541 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2542 return err
? err
: len
;
2544 static struct rdev_sysfs_entry rdev_state
=
2545 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2548 errors_show(struct md_rdev
*rdev
, char *page
)
2550 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2554 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2557 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2558 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2559 atomic_set(&rdev
->corrected_errors
, n
);
2564 static struct rdev_sysfs_entry rdev_errors
=
2565 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2568 slot_show(struct md_rdev
*rdev
, char *page
)
2570 if (rdev
->raid_disk
< 0)
2571 return sprintf(page
, "none\n");
2573 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2577 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2581 int slot
= simple_strtoul(buf
, &e
, 10);
2582 if (strncmp(buf
, "none", 4)==0)
2584 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2586 if (rdev
->mddev
->pers
&& slot
== -1) {
2587 /* Setting 'slot' on an active array requires also
2588 * updating the 'rd%d' link, and communicating
2589 * with the personality with ->hot_*_disk.
2590 * For now we only support removing
2591 * failed/spare devices. This normally happens automatically,
2592 * but not when the metadata is externally managed.
2594 if (rdev
->raid_disk
== -1)
2596 /* personality does all needed checks */
2597 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2599 clear_bit(Blocked
, &rdev
->flags
);
2600 remove_and_add_spares(rdev
->mddev
, rdev
);
2601 if (rdev
->raid_disk
>= 0)
2603 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2604 md_wakeup_thread(rdev
->mddev
->thread
);
2605 } else if (rdev
->mddev
->pers
) {
2606 /* Activating a spare .. or possibly reactivating
2607 * if we ever get bitmaps working here.
2610 if (rdev
->raid_disk
!= -1)
2613 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2616 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2619 if (slot
>= rdev
->mddev
->raid_disks
&&
2620 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2623 rdev
->raid_disk
= slot
;
2624 if (test_bit(In_sync
, &rdev
->flags
))
2625 rdev
->saved_raid_disk
= slot
;
2627 rdev
->saved_raid_disk
= -1;
2628 clear_bit(In_sync
, &rdev
->flags
);
2629 clear_bit(Bitmap_sync
, &rdev
->flags
);
2630 err
= rdev
->mddev
->pers
->
2631 hot_add_disk(rdev
->mddev
, rdev
);
2633 rdev
->raid_disk
= -1;
2636 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2637 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2638 /* failure here is OK */;
2639 /* don't wakeup anyone, leave that to userspace. */
2641 if (slot
>= rdev
->mddev
->raid_disks
&&
2642 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2644 rdev
->raid_disk
= slot
;
2645 /* assume it is working */
2646 clear_bit(Faulty
, &rdev
->flags
);
2647 clear_bit(WriteMostly
, &rdev
->flags
);
2648 set_bit(In_sync
, &rdev
->flags
);
2649 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2654 static struct rdev_sysfs_entry rdev_slot
=
2655 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2658 offset_show(struct md_rdev
*rdev
, char *page
)
2660 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2664 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2666 unsigned long long offset
;
2667 if (kstrtoull(buf
, 10, &offset
) < 0)
2669 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2671 if (rdev
->sectors
&& rdev
->mddev
->external
)
2672 /* Must set offset before size, so overlap checks
2675 rdev
->data_offset
= offset
;
2676 rdev
->new_data_offset
= offset
;
2680 static struct rdev_sysfs_entry rdev_offset
=
2681 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2683 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2685 return sprintf(page
, "%llu\n",
2686 (unsigned long long)rdev
->new_data_offset
);
2689 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2690 const char *buf
, size_t len
)
2692 unsigned long long new_offset
;
2693 struct mddev
*mddev
= rdev
->mddev
;
2695 if (kstrtoull(buf
, 10, &new_offset
) < 0)
2698 if (mddev
->sync_thread
)
2700 if (new_offset
== rdev
->data_offset
)
2701 /* reset is always permitted */
2703 else if (new_offset
> rdev
->data_offset
) {
2704 /* must not push array size beyond rdev_sectors */
2705 if (new_offset
- rdev
->data_offset
2706 + mddev
->dev_sectors
> rdev
->sectors
)
2709 /* Metadata worries about other space details. */
2711 /* decreasing the offset is inconsistent with a backwards
2714 if (new_offset
< rdev
->data_offset
&&
2715 mddev
->reshape_backwards
)
2717 /* Increasing offset is inconsistent with forwards
2718 * reshape. reshape_direction should be set to
2719 * 'backwards' first.
2721 if (new_offset
> rdev
->data_offset
&&
2722 !mddev
->reshape_backwards
)
2725 if (mddev
->pers
&& mddev
->persistent
&&
2726 !super_types
[mddev
->major_version
]
2727 .allow_new_offset(rdev
, new_offset
))
2729 rdev
->new_data_offset
= new_offset
;
2730 if (new_offset
> rdev
->data_offset
)
2731 mddev
->reshape_backwards
= 1;
2732 else if (new_offset
< rdev
->data_offset
)
2733 mddev
->reshape_backwards
= 0;
2737 static struct rdev_sysfs_entry rdev_new_offset
=
2738 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2741 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2743 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2746 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2748 /* check if two start/length pairs overlap */
2756 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2758 unsigned long long blocks
;
2761 if (kstrtoull(buf
, 10, &blocks
) < 0)
2764 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2765 return -EINVAL
; /* sector conversion overflow */
2768 if (new != blocks
* 2)
2769 return -EINVAL
; /* unsigned long long to sector_t overflow */
2776 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2778 struct mddev
*my_mddev
= rdev
->mddev
;
2779 sector_t oldsectors
= rdev
->sectors
;
2782 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2784 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2785 return -EINVAL
; /* too confusing */
2786 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2787 if (my_mddev
->persistent
) {
2788 sectors
= super_types
[my_mddev
->major_version
].
2789 rdev_size_change(rdev
, sectors
);
2792 } else if (!sectors
)
2793 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2795 if (!my_mddev
->pers
->resize
)
2796 /* Cannot change size for RAID0 or Linear etc */
2799 if (sectors
< my_mddev
->dev_sectors
)
2800 return -EINVAL
; /* component must fit device */
2802 rdev
->sectors
= sectors
;
2803 if (sectors
> oldsectors
&& my_mddev
->external
) {
2804 /* Need to check that all other rdevs with the same
2805 * ->bdev do not overlap. 'rcu' is sufficient to walk
2806 * the rdev lists safely.
2807 * This check does not provide a hard guarantee, it
2808 * just helps avoid dangerous mistakes.
2810 struct mddev
*mddev
;
2812 struct list_head
*tmp
;
2815 for_each_mddev(mddev
, tmp
) {
2816 struct md_rdev
*rdev2
;
2818 rdev_for_each(rdev2
, mddev
)
2819 if (rdev
->bdev
== rdev2
->bdev
&&
2821 overlaps(rdev
->data_offset
, rdev
->sectors
,
2834 /* Someone else could have slipped in a size
2835 * change here, but doing so is just silly.
2836 * We put oldsectors back because we *know* it is
2837 * safe, and trust userspace not to race with
2840 rdev
->sectors
= oldsectors
;
2847 static struct rdev_sysfs_entry rdev_size
=
2848 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2850 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2852 unsigned long long recovery_start
= rdev
->recovery_offset
;
2854 if (test_bit(In_sync
, &rdev
->flags
) ||
2855 recovery_start
== MaxSector
)
2856 return sprintf(page
, "none\n");
2858 return sprintf(page
, "%llu\n", recovery_start
);
2861 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2863 unsigned long long recovery_start
;
2865 if (cmd_match(buf
, "none"))
2866 recovery_start
= MaxSector
;
2867 else if (kstrtoull(buf
, 10, &recovery_start
))
2870 if (rdev
->mddev
->pers
&&
2871 rdev
->raid_disk
>= 0)
2874 rdev
->recovery_offset
= recovery_start
;
2875 if (recovery_start
== MaxSector
)
2876 set_bit(In_sync
, &rdev
->flags
);
2878 clear_bit(In_sync
, &rdev
->flags
);
2882 static struct rdev_sysfs_entry rdev_recovery_start
=
2883 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2886 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2888 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2890 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2892 return badblocks_show(&rdev
->badblocks
, page
, 0);
2894 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2896 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2897 /* Maybe that ack was all we needed */
2898 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2899 wake_up(&rdev
->blocked_wait
);
2902 static struct rdev_sysfs_entry rdev_bad_blocks
=
2903 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2905 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2907 return badblocks_show(&rdev
->badblocks
, page
, 1);
2909 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2911 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2913 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2914 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2916 static struct attribute
*rdev_default_attrs
[] = {
2921 &rdev_new_offset
.attr
,
2923 &rdev_recovery_start
.attr
,
2924 &rdev_bad_blocks
.attr
,
2925 &rdev_unack_bad_blocks
.attr
,
2929 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2931 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2932 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2933 struct mddev
*mddev
= rdev
->mddev
;
2939 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2941 if (rdev
->mddev
== NULL
)
2944 rv
= entry
->show(rdev
, page
);
2945 mddev_unlock(mddev
);
2951 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2952 const char *page
, size_t length
)
2954 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2955 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2957 struct mddev
*mddev
= rdev
->mddev
;
2961 if (!capable(CAP_SYS_ADMIN
))
2963 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2965 if (rdev
->mddev
== NULL
)
2968 rv
= entry
->store(rdev
, page
, length
);
2969 mddev_unlock(mddev
);
2974 static void rdev_free(struct kobject
*ko
)
2976 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
2979 static const struct sysfs_ops rdev_sysfs_ops
= {
2980 .show
= rdev_attr_show
,
2981 .store
= rdev_attr_store
,
2983 static struct kobj_type rdev_ktype
= {
2984 .release
= rdev_free
,
2985 .sysfs_ops
= &rdev_sysfs_ops
,
2986 .default_attrs
= rdev_default_attrs
,
2989 int md_rdev_init(struct md_rdev
*rdev
)
2992 rdev
->saved_raid_disk
= -1;
2993 rdev
->raid_disk
= -1;
2995 rdev
->data_offset
= 0;
2996 rdev
->new_data_offset
= 0;
2997 rdev
->sb_events
= 0;
2998 rdev
->last_read_error
.tv_sec
= 0;
2999 rdev
->last_read_error
.tv_nsec
= 0;
3000 rdev
->sb_loaded
= 0;
3001 rdev
->bb_page
= NULL
;
3002 atomic_set(&rdev
->nr_pending
, 0);
3003 atomic_set(&rdev
->read_errors
, 0);
3004 atomic_set(&rdev
->corrected_errors
, 0);
3006 INIT_LIST_HEAD(&rdev
->same_set
);
3007 init_waitqueue_head(&rdev
->blocked_wait
);
3009 /* Add space to store bad block list.
3010 * This reserves the space even on arrays where it cannot
3011 * be used - I wonder if that matters
3013 rdev
->badblocks
.count
= 0;
3014 rdev
->badblocks
.shift
= -1; /* disabled until explicitly enabled */
3015 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3016 seqlock_init(&rdev
->badblocks
.lock
);
3017 if (rdev
->badblocks
.page
== NULL
)
3022 EXPORT_SYMBOL_GPL(md_rdev_init
);
3024 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3026 * mark the device faulty if:
3028 * - the device is nonexistent (zero size)
3029 * - the device has no valid superblock
3031 * a faulty rdev _never_ has rdev->sb set.
3033 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3035 char b
[BDEVNAME_SIZE
];
3037 struct md_rdev
*rdev
;
3040 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3042 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3043 return ERR_PTR(-ENOMEM
);
3046 err
= md_rdev_init(rdev
);
3049 err
= alloc_disk_sb(rdev
);
3053 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3057 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3059 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3062 "md: %s has zero or unknown size, marking faulty!\n",
3063 bdevname(rdev
->bdev
,b
));
3068 if (super_format
>= 0) {
3069 err
= super_types
[super_format
].
3070 load_super(rdev
, NULL
, super_minor
);
3071 if (err
== -EINVAL
) {
3073 "md: %s does not have a valid v%d.%d "
3074 "superblock, not importing!\n",
3075 bdevname(rdev
->bdev
,b
),
3076 super_format
, super_minor
);
3081 "md: could not read %s's sb, not importing!\n",
3082 bdevname(rdev
->bdev
,b
));
3092 md_rdev_clear(rdev
);
3094 return ERR_PTR(err
);
3098 * Check a full RAID array for plausibility
3101 static void analyze_sbs(struct mddev
*mddev
)
3104 struct md_rdev
*rdev
, *freshest
, *tmp
;
3105 char b
[BDEVNAME_SIZE
];
3108 rdev_for_each_safe(rdev
, tmp
, mddev
)
3109 switch (super_types
[mddev
->major_version
].
3110 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3118 "md: fatal superblock inconsistency in %s"
3119 " -- removing from array\n",
3120 bdevname(rdev
->bdev
,b
));
3121 kick_rdev_from_array(rdev
);
3124 super_types
[mddev
->major_version
].
3125 validate_super(mddev
, freshest
);
3128 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3129 if (mddev
->max_disks
&&
3130 (rdev
->desc_nr
>= mddev
->max_disks
||
3131 i
> mddev
->max_disks
)) {
3133 "md: %s: %s: only %d devices permitted\n",
3134 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3136 kick_rdev_from_array(rdev
);
3139 if (rdev
!= freshest
)
3140 if (super_types
[mddev
->major_version
].
3141 validate_super(mddev
, rdev
)) {
3142 printk(KERN_WARNING
"md: kicking non-fresh %s"
3144 bdevname(rdev
->bdev
,b
));
3145 kick_rdev_from_array(rdev
);
3148 if (mddev
->level
== LEVEL_MULTIPATH
) {
3149 rdev
->desc_nr
= i
++;
3150 rdev
->raid_disk
= rdev
->desc_nr
;
3151 set_bit(In_sync
, &rdev
->flags
);
3152 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3153 rdev
->raid_disk
= -1;
3154 clear_bit(In_sync
, &rdev
->flags
);
3159 /* Read a fixed-point number.
3160 * Numbers in sysfs attributes should be in "standard" units where
3161 * possible, so time should be in seconds.
3162 * However we internally use a a much smaller unit such as
3163 * milliseconds or jiffies.
3164 * This function takes a decimal number with a possible fractional
3165 * component, and produces an integer which is the result of
3166 * multiplying that number by 10^'scale'.
3167 * all without any floating-point arithmetic.
3169 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3171 unsigned long result
= 0;
3173 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3176 else if (decimals
< scale
) {
3179 result
= result
* 10 + value
;
3191 while (decimals
< scale
) {
3199 static void md_safemode_timeout(unsigned long data
);
3202 safe_delay_show(struct mddev
*mddev
, char *page
)
3204 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3205 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3208 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3212 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3215 mddev
->safemode_delay
= 0;
3217 unsigned long old_delay
= mddev
->safemode_delay
;
3218 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3219 if (mddev
->safemode_delay
== 0)
3220 mddev
->safemode_delay
= 1;
3221 if (mddev
->safemode_delay
< old_delay
|| old_delay
== 0)
3222 md_safemode_timeout((unsigned long)mddev
);
3226 static struct md_sysfs_entry md_safe_delay
=
3227 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3230 level_show(struct mddev
*mddev
, char *page
)
3232 struct md_personality
*p
= mddev
->pers
;
3234 return sprintf(page
, "%s\n", p
->name
);
3235 else if (mddev
->clevel
[0])
3236 return sprintf(page
, "%s\n", mddev
->clevel
);
3237 else if (mddev
->level
!= LEVEL_NONE
)
3238 return sprintf(page
, "%d\n", mddev
->level
);
3244 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3248 struct md_personality
*pers
;
3251 struct md_rdev
*rdev
;
3253 if (mddev
->pers
== NULL
) {
3256 if (len
>= sizeof(mddev
->clevel
))
3258 strncpy(mddev
->clevel
, buf
, len
);
3259 if (mddev
->clevel
[len
-1] == '\n')
3261 mddev
->clevel
[len
] = 0;
3262 mddev
->level
= LEVEL_NONE
;
3268 /* request to change the personality. Need to ensure:
3269 * - array is not engaged in resync/recovery/reshape
3270 * - old personality can be suspended
3271 * - new personality will access other array.
3274 if (mddev
->sync_thread
||
3275 mddev
->reshape_position
!= MaxSector
||
3276 mddev
->sysfs_active
)
3279 if (!mddev
->pers
->quiesce
) {
3280 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3281 mdname(mddev
), mddev
->pers
->name
);
3285 /* Now find the new personality */
3286 if (len
== 0 || len
>= sizeof(clevel
))
3288 strncpy(clevel
, buf
, len
);
3289 if (clevel
[len
-1] == '\n')
3292 if (kstrtol(clevel
, 10, &level
))
3295 if (request_module("md-%s", clevel
) != 0)
3296 request_module("md-level-%s", clevel
);
3297 spin_lock(&pers_lock
);
3298 pers
= find_pers(level
, clevel
);
3299 if (!pers
|| !try_module_get(pers
->owner
)) {
3300 spin_unlock(&pers_lock
);
3301 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3304 spin_unlock(&pers_lock
);
3306 if (pers
== mddev
->pers
) {
3307 /* Nothing to do! */
3308 module_put(pers
->owner
);
3311 if (!pers
->takeover
) {
3312 module_put(pers
->owner
);
3313 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3314 mdname(mddev
), clevel
);
3318 rdev_for_each(rdev
, mddev
)
3319 rdev
->new_raid_disk
= rdev
->raid_disk
;
3321 /* ->takeover must set new_* and/or delta_disks
3322 * if it succeeds, and may set them when it fails.
3324 priv
= pers
->takeover(mddev
);
3326 mddev
->new_level
= mddev
->level
;
3327 mddev
->new_layout
= mddev
->layout
;
3328 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3329 mddev
->raid_disks
-= mddev
->delta_disks
;
3330 mddev
->delta_disks
= 0;
3331 mddev
->reshape_backwards
= 0;
3332 module_put(pers
->owner
);
3333 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3334 mdname(mddev
), clevel
);
3335 return PTR_ERR(priv
);
3338 /* Looks like we have a winner */
3339 mddev_suspend(mddev
);
3340 mddev
->pers
->stop(mddev
);
3342 if (mddev
->pers
->sync_request
== NULL
&&
3343 pers
->sync_request
!= NULL
) {
3344 /* need to add the md_redundancy_group */
3345 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3347 "md: cannot register extra attributes for %s\n",
3349 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3351 if (mddev
->pers
->sync_request
!= NULL
&&
3352 pers
->sync_request
== NULL
) {
3353 /* need to remove the md_redundancy_group */
3354 if (mddev
->to_remove
== NULL
)
3355 mddev
->to_remove
= &md_redundancy_group
;
3358 if (mddev
->pers
->sync_request
== NULL
&&
3360 /* We are converting from a no-redundancy array
3361 * to a redundancy array and metadata is managed
3362 * externally so we need to be sure that writes
3363 * won't block due to a need to transition
3365 * until external management is started.
3368 mddev
->safemode_delay
= 0;
3369 mddev
->safemode
= 0;
3372 rdev_for_each(rdev
, mddev
) {
3373 if (rdev
->raid_disk
< 0)
3375 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3376 rdev
->new_raid_disk
= -1;
3377 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3379 sysfs_unlink_rdev(mddev
, rdev
);
3381 rdev_for_each(rdev
, mddev
) {
3382 if (rdev
->raid_disk
< 0)
3384 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3386 rdev
->raid_disk
= rdev
->new_raid_disk
;
3387 if (rdev
->raid_disk
< 0)
3388 clear_bit(In_sync
, &rdev
->flags
);
3390 if (sysfs_link_rdev(mddev
, rdev
))
3391 printk(KERN_WARNING
"md: cannot register rd%d"
3392 " for %s after level change\n",
3393 rdev
->raid_disk
, mdname(mddev
));
3397 module_put(mddev
->pers
->owner
);
3399 mddev
->private = priv
;
3400 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3401 mddev
->level
= mddev
->new_level
;
3402 mddev
->layout
= mddev
->new_layout
;
3403 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3404 mddev
->delta_disks
= 0;
3405 mddev
->reshape_backwards
= 0;
3406 mddev
->degraded
= 0;
3407 if (mddev
->pers
->sync_request
== NULL
) {
3408 /* this is now an array without redundancy, so
3409 * it must always be in_sync
3412 del_timer_sync(&mddev
->safemode_timer
);
3414 blk_set_stacking_limits(&mddev
->queue
->limits
);
3416 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3417 mddev_resume(mddev
);
3419 md_update_sb(mddev
, 1);
3420 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3421 md_new_event(mddev
);
3425 static struct md_sysfs_entry md_level
=
3426 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3429 layout_show(struct mddev
*mddev
, char *page
)
3431 /* just a number, not meaningful for all levels */
3432 if (mddev
->reshape_position
!= MaxSector
&&
3433 mddev
->layout
!= mddev
->new_layout
)
3434 return sprintf(page
, "%d (%d)\n",
3435 mddev
->new_layout
, mddev
->layout
);
3436 return sprintf(page
, "%d\n", mddev
->layout
);
3440 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3443 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3445 if (!*buf
|| (*e
&& *e
!= '\n'))
3450 if (mddev
->pers
->check_reshape
== NULL
)
3454 mddev
->new_layout
= n
;
3455 err
= mddev
->pers
->check_reshape(mddev
);
3457 mddev
->new_layout
= mddev
->layout
;
3461 mddev
->new_layout
= n
;
3462 if (mddev
->reshape_position
== MaxSector
)
3467 static struct md_sysfs_entry md_layout
=
3468 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3471 raid_disks_show(struct mddev
*mddev
, char *page
)
3473 if (mddev
->raid_disks
== 0)
3475 if (mddev
->reshape_position
!= MaxSector
&&
3476 mddev
->delta_disks
!= 0)
3477 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3478 mddev
->raid_disks
- mddev
->delta_disks
);
3479 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3482 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3485 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3489 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3491 if (!*buf
|| (*e
&& *e
!= '\n'))
3495 rv
= update_raid_disks(mddev
, n
);
3496 else if (mddev
->reshape_position
!= MaxSector
) {
3497 struct md_rdev
*rdev
;
3498 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3500 rdev_for_each(rdev
, mddev
) {
3502 rdev
->data_offset
< rdev
->new_data_offset
)
3505 rdev
->data_offset
> rdev
->new_data_offset
)
3508 mddev
->delta_disks
= n
- olddisks
;
3509 mddev
->raid_disks
= n
;
3510 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3512 mddev
->raid_disks
= n
;
3513 return rv
? rv
: len
;
3515 static struct md_sysfs_entry md_raid_disks
=
3516 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3519 chunk_size_show(struct mddev
*mddev
, char *page
)
3521 if (mddev
->reshape_position
!= MaxSector
&&
3522 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3523 return sprintf(page
, "%d (%d)\n",
3524 mddev
->new_chunk_sectors
<< 9,
3525 mddev
->chunk_sectors
<< 9);
3526 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3530 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3533 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3535 if (!*buf
|| (*e
&& *e
!= '\n'))
3540 if (mddev
->pers
->check_reshape
== NULL
)
3544 mddev
->new_chunk_sectors
= n
>> 9;
3545 err
= mddev
->pers
->check_reshape(mddev
);
3547 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3551 mddev
->new_chunk_sectors
= n
>> 9;
3552 if (mddev
->reshape_position
== MaxSector
)
3553 mddev
->chunk_sectors
= n
>> 9;
3557 static struct md_sysfs_entry md_chunk_size
=
3558 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3561 resync_start_show(struct mddev
*mddev
, char *page
)
3563 if (mddev
->recovery_cp
== MaxSector
)
3564 return sprintf(page
, "none\n");
3565 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3569 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3572 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3574 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3576 if (cmd_match(buf
, "none"))
3578 else if (!*buf
|| (*e
&& *e
!= '\n'))
3581 mddev
->recovery_cp
= n
;
3583 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3586 static struct md_sysfs_entry md_resync_start
=
3587 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3590 * The array state can be:
3593 * No devices, no size, no level
3594 * Equivalent to STOP_ARRAY ioctl
3596 * May have some settings, but array is not active
3597 * all IO results in error
3598 * When written, doesn't tear down array, but just stops it
3599 * suspended (not supported yet)
3600 * All IO requests will block. The array can be reconfigured.
3601 * Writing this, if accepted, will block until array is quiescent
3603 * no resync can happen. no superblocks get written.
3604 * write requests fail
3606 * like readonly, but behaves like 'clean' on a write request.
3608 * clean - no pending writes, but otherwise active.
3609 * When written to inactive array, starts without resync
3610 * If a write request arrives then
3611 * if metadata is known, mark 'dirty' and switch to 'active'.
3612 * if not known, block and switch to write-pending
3613 * If written to an active array that has pending writes, then fails.
3615 * fully active: IO and resync can be happening.
3616 * When written to inactive array, starts with resync
3619 * clean, but writes are blocked waiting for 'active' to be written.
3622 * like active, but no writes have been seen for a while (100msec).
3625 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3626 write_pending
, active_idle
, bad_word
};
3627 static char *array_states
[] = {
3628 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3629 "write-pending", "active-idle", NULL
};
3631 static int match_word(const char *word
, char **list
)
3634 for (n
=0; list
[n
]; n
++)
3635 if (cmd_match(word
, list
[n
]))
3641 array_state_show(struct mddev
*mddev
, char *page
)
3643 enum array_state st
= inactive
;
3656 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3658 else if (mddev
->safemode
)
3664 if (list_empty(&mddev
->disks
) &&
3665 mddev
->raid_disks
== 0 &&
3666 mddev
->dev_sectors
== 0)
3671 return sprintf(page
, "%s\n", array_states
[st
]);
3674 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
3675 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
3676 static int do_md_run(struct mddev
*mddev
);
3677 static int restart_array(struct mddev
*mddev
);
3680 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3683 enum array_state st
= match_word(buf
, array_states
);
3688 /* stopping an active array */
3689 err
= do_md_stop(mddev
, 0, NULL
);
3692 /* stopping an active array */
3694 err
= do_md_stop(mddev
, 2, NULL
);
3696 err
= 0; /* already inactive */
3699 break; /* not supported yet */
3702 err
= md_set_readonly(mddev
, NULL
);
3705 set_disk_ro(mddev
->gendisk
, 1);
3706 err
= do_md_run(mddev
);
3712 err
= md_set_readonly(mddev
, NULL
);
3713 else if (mddev
->ro
== 1)
3714 err
= restart_array(mddev
);
3717 set_disk_ro(mddev
->gendisk
, 0);
3721 err
= do_md_run(mddev
);
3726 restart_array(mddev
);
3727 spin_lock_irq(&mddev
->write_lock
);
3728 if (atomic_read(&mddev
->writes_pending
) == 0) {
3729 if (mddev
->in_sync
== 0) {
3731 if (mddev
->safemode
== 1)
3732 mddev
->safemode
= 0;
3733 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3738 spin_unlock_irq(&mddev
->write_lock
);
3744 restart_array(mddev
);
3745 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3746 wake_up(&mddev
->sb_wait
);
3750 set_disk_ro(mddev
->gendisk
, 0);
3751 err
= do_md_run(mddev
);
3756 /* these cannot be set */
3762 if (mddev
->hold_active
== UNTIL_IOCTL
)
3763 mddev
->hold_active
= 0;
3764 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3768 static struct md_sysfs_entry md_array_state
=
3769 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3772 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3773 return sprintf(page
, "%d\n",
3774 atomic_read(&mddev
->max_corr_read_errors
));
3778 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3781 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3783 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3784 atomic_set(&mddev
->max_corr_read_errors
, n
);
3790 static struct md_sysfs_entry max_corr_read_errors
=
3791 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3792 max_corrected_read_errors_store
);
3795 null_show(struct mddev
*mddev
, char *page
)
3801 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3803 /* buf must be %d:%d\n? giving major and minor numbers */
3804 /* The new device is added to the array.
3805 * If the array has a persistent superblock, we read the
3806 * superblock to initialise info and check validity.
3807 * Otherwise, only checking done is that in bind_rdev_to_array,
3808 * which mainly checks size.
3811 int major
= simple_strtoul(buf
, &e
, 10);
3814 struct md_rdev
*rdev
;
3817 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3819 minor
= simple_strtoul(e
+1, &e
, 10);
3820 if (*e
&& *e
!= '\n')
3822 dev
= MKDEV(major
, minor
);
3823 if (major
!= MAJOR(dev
) ||
3824 minor
!= MINOR(dev
))
3827 if (mddev
->persistent
) {
3828 rdev
= md_import_device(dev
, mddev
->major_version
,
3829 mddev
->minor_version
);
3830 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3831 struct md_rdev
*rdev0
3832 = list_entry(mddev
->disks
.next
,
3833 struct md_rdev
, same_set
);
3834 err
= super_types
[mddev
->major_version
]
3835 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3839 } else if (mddev
->external
)
3840 rdev
= md_import_device(dev
, -2, -1);
3842 rdev
= md_import_device(dev
, -1, -1);
3845 return PTR_ERR(rdev
);
3846 err
= bind_rdev_to_array(rdev
, mddev
);
3850 return err
? err
: len
;
3853 static struct md_sysfs_entry md_new_device
=
3854 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3857 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3860 unsigned long chunk
, end_chunk
;
3864 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3866 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3867 if (buf
== end
) break;
3868 if (*end
== '-') { /* range */
3870 end_chunk
= simple_strtoul(buf
, &end
, 0);
3871 if (buf
== end
) break;
3873 if (*end
&& !isspace(*end
)) break;
3874 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3875 buf
= skip_spaces(end
);
3877 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3882 static struct md_sysfs_entry md_bitmap
=
3883 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3886 size_show(struct mddev
*mddev
, char *page
)
3888 return sprintf(page
, "%llu\n",
3889 (unsigned long long)mddev
->dev_sectors
/ 2);
3892 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
3895 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3897 /* If array is inactive, we can reduce the component size, but
3898 * not increase it (except from 0).
3899 * If array is active, we can try an on-line resize
3902 int err
= strict_blocks_to_sectors(buf
, §ors
);
3907 err
= update_size(mddev
, sectors
);
3908 md_update_sb(mddev
, 1);
3910 if (mddev
->dev_sectors
== 0 ||
3911 mddev
->dev_sectors
> sectors
)
3912 mddev
->dev_sectors
= sectors
;
3916 return err
? err
: len
;
3919 static struct md_sysfs_entry md_size
=
3920 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3922 /* Metadata version.
3924 * 'none' for arrays with no metadata (good luck...)
3925 * 'external' for arrays with externally managed metadata,
3926 * or N.M for internally known formats
3929 metadata_show(struct mddev
*mddev
, char *page
)
3931 if (mddev
->persistent
)
3932 return sprintf(page
, "%d.%d\n",
3933 mddev
->major_version
, mddev
->minor_version
);
3934 else if (mddev
->external
)
3935 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3937 return sprintf(page
, "none\n");
3941 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3945 /* Changing the details of 'external' metadata is
3946 * always permitted. Otherwise there must be
3947 * no devices attached to the array.
3949 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3951 else if (!list_empty(&mddev
->disks
))
3954 if (cmd_match(buf
, "none")) {
3955 mddev
->persistent
= 0;
3956 mddev
->external
= 0;
3957 mddev
->major_version
= 0;
3958 mddev
->minor_version
= 90;
3961 if (strncmp(buf
, "external:", 9) == 0) {
3962 size_t namelen
= len
-9;
3963 if (namelen
>= sizeof(mddev
->metadata_type
))
3964 namelen
= sizeof(mddev
->metadata_type
)-1;
3965 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3966 mddev
->metadata_type
[namelen
] = 0;
3967 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3968 mddev
->metadata_type
[--namelen
] = 0;
3969 mddev
->persistent
= 0;
3970 mddev
->external
= 1;
3971 mddev
->major_version
= 0;
3972 mddev
->minor_version
= 90;
3975 major
= simple_strtoul(buf
, &e
, 10);
3976 if (e
==buf
|| *e
!= '.')
3979 minor
= simple_strtoul(buf
, &e
, 10);
3980 if (e
==buf
|| (*e
&& *e
!= '\n') )
3982 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3984 mddev
->major_version
= major
;
3985 mddev
->minor_version
= minor
;
3986 mddev
->persistent
= 1;
3987 mddev
->external
= 0;
3991 static struct md_sysfs_entry md_metadata
=
3992 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3995 action_show(struct mddev
*mddev
, char *page
)
3997 char *type
= "idle";
3998 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4000 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4001 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4002 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4004 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4005 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4007 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4011 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4014 return sprintf(page
, "%s\n", type
);
4018 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4020 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4023 if (cmd_match(page
, "frozen"))
4024 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4026 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4028 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4029 if (mddev
->sync_thread
) {
4030 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4031 md_reap_sync_thread(mddev
);
4033 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4034 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4036 else if (cmd_match(page
, "resync"))
4037 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4038 else if (cmd_match(page
, "recover")) {
4039 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4040 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4041 } else if (cmd_match(page
, "reshape")) {
4043 if (mddev
->pers
->start_reshape
== NULL
)
4045 err
= mddev
->pers
->start_reshape(mddev
);
4048 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4050 if (cmd_match(page
, "check"))
4051 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4052 else if (!cmd_match(page
, "repair"))
4054 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4055 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4057 if (mddev
->ro
== 2) {
4058 /* A write to sync_action is enough to justify
4059 * canceling read-auto mode
4062 md_wakeup_thread(mddev
->sync_thread
);
4064 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4065 md_wakeup_thread(mddev
->thread
);
4066 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4070 static struct md_sysfs_entry md_scan_mode
=
4071 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4074 last_sync_action_show(struct mddev
*mddev
, char *page
)
4076 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4079 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4082 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4084 return sprintf(page
, "%llu\n",
4085 (unsigned long long)
4086 atomic64_read(&mddev
->resync_mismatches
));
4089 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4092 sync_min_show(struct mddev
*mddev
, char *page
)
4094 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4095 mddev
->sync_speed_min
? "local": "system");
4099 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4103 if (strncmp(buf
, "system", 6)==0) {
4104 mddev
->sync_speed_min
= 0;
4107 min
= simple_strtoul(buf
, &e
, 10);
4108 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4110 mddev
->sync_speed_min
= min
;
4114 static struct md_sysfs_entry md_sync_min
=
4115 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4118 sync_max_show(struct mddev
*mddev
, char *page
)
4120 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4121 mddev
->sync_speed_max
? "local": "system");
4125 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4129 if (strncmp(buf
, "system", 6)==0) {
4130 mddev
->sync_speed_max
= 0;
4133 max
= simple_strtoul(buf
, &e
, 10);
4134 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4136 mddev
->sync_speed_max
= max
;
4140 static struct md_sysfs_entry md_sync_max
=
4141 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4144 degraded_show(struct mddev
*mddev
, char *page
)
4146 return sprintf(page
, "%d\n", mddev
->degraded
);
4148 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4151 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4153 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4157 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4161 if (kstrtol(buf
, 10, &n
))
4164 if (n
!= 0 && n
!= 1)
4167 mddev
->parallel_resync
= n
;
4169 if (mddev
->sync_thread
)
4170 wake_up(&resync_wait
);
4175 /* force parallel resync, even with shared block devices */
4176 static struct md_sysfs_entry md_sync_force_parallel
=
4177 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4178 sync_force_parallel_show
, sync_force_parallel_store
);
4181 sync_speed_show(struct mddev
*mddev
, char *page
)
4183 unsigned long resync
, dt
, db
;
4184 if (mddev
->curr_resync
== 0)
4185 return sprintf(page
, "none\n");
4186 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4187 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4189 db
= resync
- mddev
->resync_mark_cnt
;
4190 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4193 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4196 sync_completed_show(struct mddev
*mddev
, char *page
)
4198 unsigned long long max_sectors
, resync
;
4200 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4201 return sprintf(page
, "none\n");
4203 if (mddev
->curr_resync
== 1 ||
4204 mddev
->curr_resync
== 2)
4205 return sprintf(page
, "delayed\n");
4207 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4208 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4209 max_sectors
= mddev
->resync_max_sectors
;
4211 max_sectors
= mddev
->dev_sectors
;
4213 resync
= mddev
->curr_resync_completed
;
4214 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4217 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4220 min_sync_show(struct mddev
*mddev
, char *page
)
4222 return sprintf(page
, "%llu\n",
4223 (unsigned long long)mddev
->resync_min
);
4226 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4228 unsigned long long min
;
4229 if (kstrtoull(buf
, 10, &min
))
4231 if (min
> mddev
->resync_max
)
4233 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4236 /* Must be a multiple of chunk_size */
4237 if (mddev
->chunk_sectors
) {
4238 sector_t temp
= min
;
4239 if (sector_div(temp
, mddev
->chunk_sectors
))
4242 mddev
->resync_min
= min
;
4247 static struct md_sysfs_entry md_min_sync
=
4248 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4251 max_sync_show(struct mddev
*mddev
, char *page
)
4253 if (mddev
->resync_max
== MaxSector
)
4254 return sprintf(page
, "max\n");
4256 return sprintf(page
, "%llu\n",
4257 (unsigned long long)mddev
->resync_max
);
4260 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4262 if (strncmp(buf
, "max", 3) == 0)
4263 mddev
->resync_max
= MaxSector
;
4265 unsigned long long max
;
4266 if (kstrtoull(buf
, 10, &max
))
4268 if (max
< mddev
->resync_min
)
4270 if (max
< mddev
->resync_max
&&
4272 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4275 /* Must be a multiple of chunk_size */
4276 if (mddev
->chunk_sectors
) {
4277 sector_t temp
= max
;
4278 if (sector_div(temp
, mddev
->chunk_sectors
))
4281 mddev
->resync_max
= max
;
4283 wake_up(&mddev
->recovery_wait
);
4287 static struct md_sysfs_entry md_max_sync
=
4288 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4291 suspend_lo_show(struct mddev
*mddev
, char *page
)
4293 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4297 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4300 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4301 unsigned long long old
= mddev
->suspend_lo
;
4303 if (mddev
->pers
== NULL
||
4304 mddev
->pers
->quiesce
== NULL
)
4306 if (buf
== e
|| (*e
&& *e
!= '\n'))
4309 mddev
->suspend_lo
= new;
4311 /* Shrinking suspended region */
4312 mddev
->pers
->quiesce(mddev
, 2);
4314 /* Expanding suspended region - need to wait */
4315 mddev
->pers
->quiesce(mddev
, 1);
4316 mddev
->pers
->quiesce(mddev
, 0);
4320 static struct md_sysfs_entry md_suspend_lo
=
4321 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4324 suspend_hi_show(struct mddev
*mddev
, char *page
)
4326 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4330 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4333 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4334 unsigned long long old
= mddev
->suspend_hi
;
4336 if (mddev
->pers
== NULL
||
4337 mddev
->pers
->quiesce
== NULL
)
4339 if (buf
== e
|| (*e
&& *e
!= '\n'))
4342 mddev
->suspend_hi
= new;
4344 /* Shrinking suspended region */
4345 mddev
->pers
->quiesce(mddev
, 2);
4347 /* Expanding suspended region - need to wait */
4348 mddev
->pers
->quiesce(mddev
, 1);
4349 mddev
->pers
->quiesce(mddev
, 0);
4353 static struct md_sysfs_entry md_suspend_hi
=
4354 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4357 reshape_position_show(struct mddev
*mddev
, char *page
)
4359 if (mddev
->reshape_position
!= MaxSector
)
4360 return sprintf(page
, "%llu\n",
4361 (unsigned long long)mddev
->reshape_position
);
4362 strcpy(page
, "none\n");
4367 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4369 struct md_rdev
*rdev
;
4371 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4374 if (buf
== e
|| (*e
&& *e
!= '\n'))
4376 mddev
->reshape_position
= new;
4377 mddev
->delta_disks
= 0;
4378 mddev
->reshape_backwards
= 0;
4379 mddev
->new_level
= mddev
->level
;
4380 mddev
->new_layout
= mddev
->layout
;
4381 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4382 rdev_for_each(rdev
, mddev
)
4383 rdev
->new_data_offset
= rdev
->data_offset
;
4387 static struct md_sysfs_entry md_reshape_position
=
4388 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4389 reshape_position_store
);
4392 reshape_direction_show(struct mddev
*mddev
, char *page
)
4394 return sprintf(page
, "%s\n",
4395 mddev
->reshape_backwards
? "backwards" : "forwards");
4399 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4402 if (cmd_match(buf
, "forwards"))
4404 else if (cmd_match(buf
, "backwards"))
4408 if (mddev
->reshape_backwards
== backwards
)
4411 /* check if we are allowed to change */
4412 if (mddev
->delta_disks
)
4415 if (mddev
->persistent
&&
4416 mddev
->major_version
== 0)
4419 mddev
->reshape_backwards
= backwards
;
4423 static struct md_sysfs_entry md_reshape_direction
=
4424 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4425 reshape_direction_store
);
4428 array_size_show(struct mddev
*mddev
, char *page
)
4430 if (mddev
->external_size
)
4431 return sprintf(page
, "%llu\n",
4432 (unsigned long long)mddev
->array_sectors
/2);
4434 return sprintf(page
, "default\n");
4438 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4442 if (strncmp(buf
, "default", 7) == 0) {
4444 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4446 sectors
= mddev
->array_sectors
;
4448 mddev
->external_size
= 0;
4450 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4452 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4455 mddev
->external_size
= 1;
4458 mddev
->array_sectors
= sectors
;
4460 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4461 revalidate_disk(mddev
->gendisk
);
4466 static struct md_sysfs_entry md_array_size
=
4467 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4470 static struct attribute
*md_default_attrs
[] = {
4473 &md_raid_disks
.attr
,
4474 &md_chunk_size
.attr
,
4476 &md_resync_start
.attr
,
4478 &md_new_device
.attr
,
4479 &md_safe_delay
.attr
,
4480 &md_array_state
.attr
,
4481 &md_reshape_position
.attr
,
4482 &md_reshape_direction
.attr
,
4483 &md_array_size
.attr
,
4484 &max_corr_read_errors
.attr
,
4488 static struct attribute
*md_redundancy_attrs
[] = {
4490 &md_last_scan_mode
.attr
,
4491 &md_mismatches
.attr
,
4494 &md_sync_speed
.attr
,
4495 &md_sync_force_parallel
.attr
,
4496 &md_sync_completed
.attr
,
4499 &md_suspend_lo
.attr
,
4500 &md_suspend_hi
.attr
,
4505 static struct attribute_group md_redundancy_group
= {
4507 .attrs
= md_redundancy_attrs
,
4511 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4513 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4514 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4519 spin_lock(&all_mddevs_lock
);
4520 if (list_empty(&mddev
->all_mddevs
)) {
4521 spin_unlock(&all_mddevs_lock
);
4525 spin_unlock(&all_mddevs_lock
);
4527 rv
= mddev_lock(mddev
);
4529 rv
= entry
->show(mddev
, page
);
4530 mddev_unlock(mddev
);
4537 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4538 const char *page
, size_t length
)
4540 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4541 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4546 if (!capable(CAP_SYS_ADMIN
))
4548 spin_lock(&all_mddevs_lock
);
4549 if (list_empty(&mddev
->all_mddevs
)) {
4550 spin_unlock(&all_mddevs_lock
);
4554 spin_unlock(&all_mddevs_lock
);
4555 if (entry
->store
== new_dev_store
)
4556 flush_workqueue(md_misc_wq
);
4557 rv
= mddev_lock(mddev
);
4559 rv
= entry
->store(mddev
, page
, length
);
4560 mddev_unlock(mddev
);
4566 static void md_free(struct kobject
*ko
)
4568 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4570 if (mddev
->sysfs_state
)
4571 sysfs_put(mddev
->sysfs_state
);
4573 if (mddev
->gendisk
) {
4574 del_gendisk(mddev
->gendisk
);
4575 put_disk(mddev
->gendisk
);
4578 blk_cleanup_queue(mddev
->queue
);
4583 static const struct sysfs_ops md_sysfs_ops
= {
4584 .show
= md_attr_show
,
4585 .store
= md_attr_store
,
4587 static struct kobj_type md_ktype
= {
4589 .sysfs_ops
= &md_sysfs_ops
,
4590 .default_attrs
= md_default_attrs
,
4595 static void mddev_delayed_delete(struct work_struct
*ws
)
4597 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4599 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4600 kobject_del(&mddev
->kobj
);
4601 kobject_put(&mddev
->kobj
);
4604 static int md_alloc(dev_t dev
, char *name
)
4606 static DEFINE_MUTEX(disks_mutex
);
4607 struct mddev
*mddev
= mddev_find(dev
);
4608 struct gendisk
*disk
;
4617 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4618 shift
= partitioned
? MdpMinorShift
: 0;
4619 unit
= MINOR(mddev
->unit
) >> shift
;
4621 /* wait for any previous instance of this device to be
4622 * completely removed (mddev_delayed_delete).
4624 flush_workqueue(md_misc_wq
);
4626 mutex_lock(&disks_mutex
);
4632 /* Need to ensure that 'name' is not a duplicate.
4634 struct mddev
*mddev2
;
4635 spin_lock(&all_mddevs_lock
);
4637 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4638 if (mddev2
->gendisk
&&
4639 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4640 spin_unlock(&all_mddevs_lock
);
4643 spin_unlock(&all_mddevs_lock
);
4647 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4650 mddev
->queue
->queuedata
= mddev
;
4652 blk_queue_make_request(mddev
->queue
, md_make_request
);
4653 blk_set_stacking_limits(&mddev
->queue
->limits
);
4655 disk
= alloc_disk(1 << shift
);
4657 blk_cleanup_queue(mddev
->queue
);
4658 mddev
->queue
= NULL
;
4661 disk
->major
= MAJOR(mddev
->unit
);
4662 disk
->first_minor
= unit
<< shift
;
4664 strcpy(disk
->disk_name
, name
);
4665 else if (partitioned
)
4666 sprintf(disk
->disk_name
, "md_d%d", unit
);
4668 sprintf(disk
->disk_name
, "md%d", unit
);
4669 disk
->fops
= &md_fops
;
4670 disk
->private_data
= mddev
;
4671 disk
->queue
= mddev
->queue
;
4672 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4673 /* Allow extended partitions. This makes the
4674 * 'mdp' device redundant, but we can't really
4677 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4678 mddev
->gendisk
= disk
;
4679 /* As soon as we call add_disk(), another thread could get
4680 * through to md_open, so make sure it doesn't get too far
4682 mutex_lock(&mddev
->open_mutex
);
4685 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4686 &disk_to_dev(disk
)->kobj
, "%s", "md");
4688 /* This isn't possible, but as kobject_init_and_add is marked
4689 * __must_check, we must do something with the result
4691 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4695 if (mddev
->kobj
.sd
&&
4696 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4697 printk(KERN_DEBUG
"pointless warning\n");
4698 mutex_unlock(&mddev
->open_mutex
);
4700 mutex_unlock(&disks_mutex
);
4701 if (!error
&& mddev
->kobj
.sd
) {
4702 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4703 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4709 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4711 md_alloc(dev
, NULL
);
4715 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4717 /* val must be "md_*" where * is not all digits.
4718 * We allocate an array with a large free minor number, and
4719 * set the name to val. val must not already be an active name.
4721 int len
= strlen(val
);
4722 char buf
[DISK_NAME_LEN
];
4724 while (len
&& val
[len
-1] == '\n')
4726 if (len
>= DISK_NAME_LEN
)
4728 strlcpy(buf
, val
, len
+1);
4729 if (strncmp(buf
, "md_", 3) != 0)
4731 return md_alloc(0, buf
);
4734 static void md_safemode_timeout(unsigned long data
)
4736 struct mddev
*mddev
= (struct mddev
*) data
;
4738 if (!atomic_read(&mddev
->writes_pending
)) {
4739 mddev
->safemode
= 1;
4740 if (mddev
->external
)
4741 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4743 md_wakeup_thread(mddev
->thread
);
4746 static int start_dirty_degraded
;
4748 int md_run(struct mddev
*mddev
)
4751 struct md_rdev
*rdev
;
4752 struct md_personality
*pers
;
4754 if (list_empty(&mddev
->disks
))
4755 /* cannot run an array with no devices.. */
4760 /* Cannot run until previous stop completes properly */
4761 if (mddev
->sysfs_active
)
4765 * Analyze all RAID superblock(s)
4767 if (!mddev
->raid_disks
) {
4768 if (!mddev
->persistent
)
4773 if (mddev
->level
!= LEVEL_NONE
)
4774 request_module("md-level-%d", mddev
->level
);
4775 else if (mddev
->clevel
[0])
4776 request_module("md-%s", mddev
->clevel
);
4779 * Drop all container device buffers, from now on
4780 * the only valid external interface is through the md
4783 rdev_for_each(rdev
, mddev
) {
4784 if (test_bit(Faulty
, &rdev
->flags
))
4786 sync_blockdev(rdev
->bdev
);
4787 invalidate_bdev(rdev
->bdev
);
4789 /* perform some consistency tests on the device.
4790 * We don't want the data to overlap the metadata,
4791 * Internal Bitmap issues have been handled elsewhere.
4793 if (rdev
->meta_bdev
) {
4794 /* Nothing to check */;
4795 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4796 if (mddev
->dev_sectors
&&
4797 rdev
->data_offset
+ mddev
->dev_sectors
4799 printk("md: %s: data overlaps metadata\n",
4804 if (rdev
->sb_start
+ rdev
->sb_size
/512
4805 > rdev
->data_offset
) {
4806 printk("md: %s: metadata overlaps data\n",
4811 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4814 if (mddev
->bio_set
== NULL
)
4815 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
4817 spin_lock(&pers_lock
);
4818 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4819 if (!pers
|| !try_module_get(pers
->owner
)) {
4820 spin_unlock(&pers_lock
);
4821 if (mddev
->level
!= LEVEL_NONE
)
4822 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4825 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4830 spin_unlock(&pers_lock
);
4831 if (mddev
->level
!= pers
->level
) {
4832 mddev
->level
= pers
->level
;
4833 mddev
->new_level
= pers
->level
;
4835 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4837 if (mddev
->reshape_position
!= MaxSector
&&
4838 pers
->start_reshape
== NULL
) {
4839 /* This personality cannot handle reshaping... */
4841 module_put(pers
->owner
);
4845 if (pers
->sync_request
) {
4846 /* Warn if this is a potentially silly
4849 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4850 struct md_rdev
*rdev2
;
4853 rdev_for_each(rdev
, mddev
)
4854 rdev_for_each(rdev2
, mddev
) {
4856 rdev
->bdev
->bd_contains
==
4857 rdev2
->bdev
->bd_contains
) {
4859 "%s: WARNING: %s appears to be"
4860 " on the same physical disk as"
4863 bdevname(rdev
->bdev
,b
),
4864 bdevname(rdev2
->bdev
,b2
));
4871 "True protection against single-disk"
4872 " failure might be compromised.\n");
4875 mddev
->recovery
= 0;
4876 /* may be over-ridden by personality */
4877 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4879 mddev
->ok_start_degraded
= start_dirty_degraded
;
4881 if (start_readonly
&& mddev
->ro
== 0)
4882 mddev
->ro
= 2; /* read-only, but switch on first write */
4884 err
= mddev
->pers
->run(mddev
);
4886 printk(KERN_ERR
"md: pers->run() failed ...\n");
4887 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4888 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4889 " but 'external_size' not in effect?\n", __func__
);
4891 "md: invalid array_size %llu > default size %llu\n",
4892 (unsigned long long)mddev
->array_sectors
/ 2,
4893 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4895 mddev
->pers
->stop(mddev
);
4897 if (err
== 0 && mddev
->pers
->sync_request
&&
4898 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
4899 err
= bitmap_create(mddev
);
4901 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4902 mdname(mddev
), err
);
4903 mddev
->pers
->stop(mddev
);
4907 module_put(mddev
->pers
->owner
);
4909 bitmap_destroy(mddev
);
4912 if (mddev
->pers
->sync_request
) {
4913 if (mddev
->kobj
.sd
&&
4914 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4916 "md: cannot register extra attributes for %s\n",
4918 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4919 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4922 atomic_set(&mddev
->writes_pending
,0);
4923 atomic_set(&mddev
->max_corr_read_errors
,
4924 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4925 mddev
->safemode
= 0;
4926 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4927 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4928 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4932 rdev_for_each(rdev
, mddev
)
4933 if (rdev
->raid_disk
>= 0)
4934 if (sysfs_link_rdev(mddev
, rdev
))
4935 /* failure here is OK */;
4937 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4939 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
4940 md_update_sb(mddev
, 0);
4942 md_new_event(mddev
);
4943 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4944 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4945 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4948 EXPORT_SYMBOL_GPL(md_run
);
4950 static int do_md_run(struct mddev
*mddev
)
4954 err
= md_run(mddev
);
4957 err
= bitmap_load(mddev
);
4959 bitmap_destroy(mddev
);
4963 md_wakeup_thread(mddev
->thread
);
4964 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4966 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4967 revalidate_disk(mddev
->gendisk
);
4969 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4974 static int restart_array(struct mddev
*mddev
)
4976 struct gendisk
*disk
= mddev
->gendisk
;
4978 /* Complain if it has no devices */
4979 if (list_empty(&mddev
->disks
))
4985 mddev
->safemode
= 0;
4987 set_disk_ro(disk
, 0);
4988 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4990 /* Kick recovery or resync if necessary */
4991 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4992 md_wakeup_thread(mddev
->thread
);
4993 md_wakeup_thread(mddev
->sync_thread
);
4994 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4998 static void md_clean(struct mddev
*mddev
)
5000 mddev
->array_sectors
= 0;
5001 mddev
->external_size
= 0;
5002 mddev
->dev_sectors
= 0;
5003 mddev
->raid_disks
= 0;
5004 mddev
->recovery_cp
= 0;
5005 mddev
->resync_min
= 0;
5006 mddev
->resync_max
= MaxSector
;
5007 mddev
->reshape_position
= MaxSector
;
5008 mddev
->external
= 0;
5009 mddev
->persistent
= 0;
5010 mddev
->level
= LEVEL_NONE
;
5011 mddev
->clevel
[0] = 0;
5014 mddev
->metadata_type
[0] = 0;
5015 mddev
->chunk_sectors
= 0;
5016 mddev
->ctime
= mddev
->utime
= 0;
5018 mddev
->max_disks
= 0;
5020 mddev
->can_decrease_events
= 0;
5021 mddev
->delta_disks
= 0;
5022 mddev
->reshape_backwards
= 0;
5023 mddev
->new_level
= LEVEL_NONE
;
5024 mddev
->new_layout
= 0;
5025 mddev
->new_chunk_sectors
= 0;
5026 mddev
->curr_resync
= 0;
5027 atomic64_set(&mddev
->resync_mismatches
, 0);
5028 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5029 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5030 mddev
->recovery
= 0;
5033 mddev
->degraded
= 0;
5034 mddev
->safemode
= 0;
5035 mddev
->merge_check_needed
= 0;
5036 mddev
->bitmap_info
.offset
= 0;
5037 mddev
->bitmap_info
.default_offset
= 0;
5038 mddev
->bitmap_info
.default_space
= 0;
5039 mddev
->bitmap_info
.chunksize
= 0;
5040 mddev
->bitmap_info
.daemon_sleep
= 0;
5041 mddev
->bitmap_info
.max_write_behind
= 0;
5044 static void __md_stop_writes(struct mddev
*mddev
)
5046 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5047 if (mddev
->sync_thread
) {
5048 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5049 md_reap_sync_thread(mddev
);
5052 del_timer_sync(&mddev
->safemode_timer
);
5054 bitmap_flush(mddev
);
5055 md_super_wait(mddev
);
5057 if (mddev
->ro
== 0 &&
5058 (!mddev
->in_sync
|| (mddev
->flags
& MD_UPDATE_SB_FLAGS
))) {
5059 /* mark array as shutdown cleanly */
5061 md_update_sb(mddev
, 1);
5065 void md_stop_writes(struct mddev
*mddev
)
5067 mddev_lock_nointr(mddev
);
5068 __md_stop_writes(mddev
);
5069 mddev_unlock(mddev
);
5071 EXPORT_SYMBOL_GPL(md_stop_writes
);
5073 static void __md_stop(struct mddev
*mddev
)
5076 mddev
->pers
->stop(mddev
);
5077 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5078 mddev
->to_remove
= &md_redundancy_group
;
5079 module_put(mddev
->pers
->owner
);
5081 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5084 void md_stop(struct mddev
*mddev
)
5086 /* stop the array and free an attached data structures.
5087 * This is called from dm-raid
5090 bitmap_destroy(mddev
);
5092 bioset_free(mddev
->bio_set
);
5095 EXPORT_SYMBOL_GPL(md_stop
);
5097 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5102 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5104 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5105 md_wakeup_thread(mddev
->thread
);
5107 if (mddev
->sync_thread
) {
5108 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5109 /* Thread might be blocked waiting for metadata update
5110 * which will now never happen */
5111 wake_up_process(mddev
->sync_thread
->tsk
);
5113 mddev_unlock(mddev
);
5114 wait_event(resync_wait
, mddev
->sync_thread
== NULL
);
5115 mddev_lock_nointr(mddev
);
5117 mutex_lock(&mddev
->open_mutex
);
5118 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5119 mddev
->sync_thread
||
5120 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5121 printk("md: %s still in use.\n",mdname(mddev
));
5123 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5124 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5125 md_wakeup_thread(mddev
->thread
);
5131 __md_stop_writes(mddev
);
5137 set_disk_ro(mddev
->gendisk
, 1);
5138 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5139 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5140 md_wakeup_thread(mddev
->thread
);
5141 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5145 mutex_unlock(&mddev
->open_mutex
);
5150 * 0 - completely stop and dis-assemble array
5151 * 2 - stop but do not disassemble array
5153 static int do_md_stop(struct mddev
*mddev
, int mode
,
5154 struct block_device
*bdev
)
5156 struct gendisk
*disk
= mddev
->gendisk
;
5157 struct md_rdev
*rdev
;
5160 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5162 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5163 md_wakeup_thread(mddev
->thread
);
5165 if (mddev
->sync_thread
) {
5166 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5167 /* Thread might be blocked waiting for metadata update
5168 * which will now never happen */
5169 wake_up_process(mddev
->sync_thread
->tsk
);
5171 mddev_unlock(mddev
);
5172 wait_event(resync_wait
, mddev
->sync_thread
== NULL
);
5173 mddev_lock_nointr(mddev
);
5175 mutex_lock(&mddev
->open_mutex
);
5176 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5177 mddev
->sysfs_active
||
5178 mddev
->sync_thread
||
5179 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5180 printk("md: %s still in use.\n",mdname(mddev
));
5181 mutex_unlock(&mddev
->open_mutex
);
5183 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5184 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5185 md_wakeup_thread(mddev
->thread
);
5191 set_disk_ro(disk
, 0);
5193 __md_stop_writes(mddev
);
5195 mddev
->queue
->merge_bvec_fn
= NULL
;
5196 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5198 /* tell userspace to handle 'inactive' */
5199 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5201 rdev_for_each(rdev
, mddev
)
5202 if (rdev
->raid_disk
>= 0)
5203 sysfs_unlink_rdev(mddev
, rdev
);
5205 set_capacity(disk
, 0);
5206 mutex_unlock(&mddev
->open_mutex
);
5208 revalidate_disk(disk
);
5213 mutex_unlock(&mddev
->open_mutex
);
5215 * Free resources if final stop
5218 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5220 bitmap_destroy(mddev
);
5221 if (mddev
->bitmap_info
.file
) {
5222 fput(mddev
->bitmap_info
.file
);
5223 mddev
->bitmap_info
.file
= NULL
;
5225 mddev
->bitmap_info
.offset
= 0;
5227 export_array(mddev
);
5230 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5231 if (mddev
->hold_active
== UNTIL_STOP
)
5232 mddev
->hold_active
= 0;
5234 blk_integrity_unregister(disk
);
5235 md_new_event(mddev
);
5236 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5241 static void autorun_array(struct mddev
*mddev
)
5243 struct md_rdev
*rdev
;
5246 if (list_empty(&mddev
->disks
))
5249 printk(KERN_INFO
"md: running: ");
5251 rdev_for_each(rdev
, mddev
) {
5252 char b
[BDEVNAME_SIZE
];
5253 printk("<%s>", bdevname(rdev
->bdev
,b
));
5257 err
= do_md_run(mddev
);
5259 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5260 do_md_stop(mddev
, 0, NULL
);
5265 * lets try to run arrays based on all disks that have arrived
5266 * until now. (those are in pending_raid_disks)
5268 * the method: pick the first pending disk, collect all disks with
5269 * the same UUID, remove all from the pending list and put them into
5270 * the 'same_array' list. Then order this list based on superblock
5271 * update time (freshest comes first), kick out 'old' disks and
5272 * compare superblocks. If everything's fine then run it.
5274 * If "unit" is allocated, then bump its reference count
5276 static void autorun_devices(int part
)
5278 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5279 struct mddev
*mddev
;
5280 char b
[BDEVNAME_SIZE
];
5282 printk(KERN_INFO
"md: autorun ...\n");
5283 while (!list_empty(&pending_raid_disks
)) {
5286 LIST_HEAD(candidates
);
5287 rdev0
= list_entry(pending_raid_disks
.next
,
5288 struct md_rdev
, same_set
);
5290 printk(KERN_INFO
"md: considering %s ...\n",
5291 bdevname(rdev0
->bdev
,b
));
5292 INIT_LIST_HEAD(&candidates
);
5293 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5294 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5295 printk(KERN_INFO
"md: adding %s ...\n",
5296 bdevname(rdev
->bdev
,b
));
5297 list_move(&rdev
->same_set
, &candidates
);
5300 * now we have a set of devices, with all of them having
5301 * mostly sane superblocks. It's time to allocate the
5305 dev
= MKDEV(mdp_major
,
5306 rdev0
->preferred_minor
<< MdpMinorShift
);
5307 unit
= MINOR(dev
) >> MdpMinorShift
;
5309 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5312 if (rdev0
->preferred_minor
!= unit
) {
5313 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5314 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5318 md_probe(dev
, NULL
, NULL
);
5319 mddev
= mddev_find(dev
);
5320 if (!mddev
|| !mddev
->gendisk
) {
5324 "md: cannot allocate memory for md drive.\n");
5327 if (mddev_lock(mddev
))
5328 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5330 else if (mddev
->raid_disks
|| mddev
->major_version
5331 || !list_empty(&mddev
->disks
)) {
5333 "md: %s already running, cannot run %s\n",
5334 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5335 mddev_unlock(mddev
);
5337 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5338 mddev
->persistent
= 1;
5339 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5340 list_del_init(&rdev
->same_set
);
5341 if (bind_rdev_to_array(rdev
, mddev
))
5344 autorun_array(mddev
);
5345 mddev_unlock(mddev
);
5347 /* on success, candidates will be empty, on error
5350 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5351 list_del_init(&rdev
->same_set
);
5356 printk(KERN_INFO
"md: ... autorun DONE.\n");
5358 #endif /* !MODULE */
5360 static int get_version(void __user
*arg
)
5364 ver
.major
= MD_MAJOR_VERSION
;
5365 ver
.minor
= MD_MINOR_VERSION
;
5366 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5368 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5374 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
5376 mdu_array_info_t info
;
5377 int nr
,working
,insync
,failed
,spare
;
5378 struct md_rdev
*rdev
;
5380 nr
= working
= insync
= failed
= spare
= 0;
5382 rdev_for_each_rcu(rdev
, mddev
) {
5384 if (test_bit(Faulty
, &rdev
->flags
))
5388 if (test_bit(In_sync
, &rdev
->flags
))
5396 info
.major_version
= mddev
->major_version
;
5397 info
.minor_version
= mddev
->minor_version
;
5398 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5399 info
.ctime
= mddev
->ctime
;
5400 info
.level
= mddev
->level
;
5401 info
.size
= mddev
->dev_sectors
/ 2;
5402 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5405 info
.raid_disks
= mddev
->raid_disks
;
5406 info
.md_minor
= mddev
->md_minor
;
5407 info
.not_persistent
= !mddev
->persistent
;
5409 info
.utime
= mddev
->utime
;
5412 info
.state
= (1<<MD_SB_CLEAN
);
5413 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5414 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
5415 info
.active_disks
= insync
;
5416 info
.working_disks
= working
;
5417 info
.failed_disks
= failed
;
5418 info
.spare_disks
= spare
;
5420 info
.layout
= mddev
->layout
;
5421 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5423 if (copy_to_user(arg
, &info
, sizeof(info
)))
5429 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
5431 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5432 char *ptr
, *buf
= NULL
;
5435 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5440 /* bitmap disabled, zero the first byte and copy out */
5441 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5442 file
->pathname
[0] = '\0';
5446 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5450 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5451 buf
, sizeof(file
->pathname
));
5455 strcpy(file
->pathname
, ptr
);
5459 if (copy_to_user(arg
, file
, sizeof(*file
)))
5467 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
5469 mdu_disk_info_t info
;
5470 struct md_rdev
*rdev
;
5472 if (copy_from_user(&info
, arg
, sizeof(info
)))
5476 rdev
= find_rdev_nr_rcu(mddev
, info
.number
);
5478 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5479 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5480 info
.raid_disk
= rdev
->raid_disk
;
5482 if (test_bit(Faulty
, &rdev
->flags
))
5483 info
.state
|= (1<<MD_DISK_FAULTY
);
5484 else if (test_bit(In_sync
, &rdev
->flags
)) {
5485 info
.state
|= (1<<MD_DISK_ACTIVE
);
5486 info
.state
|= (1<<MD_DISK_SYNC
);
5488 if (test_bit(WriteMostly
, &rdev
->flags
))
5489 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5491 info
.major
= info
.minor
= 0;
5492 info
.raid_disk
= -1;
5493 info
.state
= (1<<MD_DISK_REMOVED
);
5497 if (copy_to_user(arg
, &info
, sizeof(info
)))
5503 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
5505 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5506 struct md_rdev
*rdev
;
5507 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5509 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5512 if (!mddev
->raid_disks
) {
5514 /* expecting a device which has a superblock */
5515 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5518 "md: md_import_device returned %ld\n",
5520 return PTR_ERR(rdev
);
5522 if (!list_empty(&mddev
->disks
)) {
5523 struct md_rdev
*rdev0
5524 = list_entry(mddev
->disks
.next
,
5525 struct md_rdev
, same_set
);
5526 err
= super_types
[mddev
->major_version
]
5527 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5530 "md: %s has different UUID to %s\n",
5531 bdevname(rdev
->bdev
,b
),
5532 bdevname(rdev0
->bdev
,b2
));
5537 err
= bind_rdev_to_array(rdev
, mddev
);
5544 * add_new_disk can be used once the array is assembled
5545 * to add "hot spares". They must already have a superblock
5550 if (!mddev
->pers
->hot_add_disk
) {
5552 "%s: personality does not support diskops!\n",
5556 if (mddev
->persistent
)
5557 rdev
= md_import_device(dev
, mddev
->major_version
,
5558 mddev
->minor_version
);
5560 rdev
= md_import_device(dev
, -1, -1);
5563 "md: md_import_device returned %ld\n",
5565 return PTR_ERR(rdev
);
5567 /* set saved_raid_disk if appropriate */
5568 if (!mddev
->persistent
) {
5569 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5570 info
->raid_disk
< mddev
->raid_disks
) {
5571 rdev
->raid_disk
= info
->raid_disk
;
5572 set_bit(In_sync
, &rdev
->flags
);
5573 clear_bit(Bitmap_sync
, &rdev
->flags
);
5575 rdev
->raid_disk
= -1;
5576 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5578 super_types
[mddev
->major_version
].
5579 validate_super(mddev
, rdev
);
5580 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5581 rdev
->raid_disk
!= info
->raid_disk
) {
5582 /* This was a hot-add request, but events doesn't
5583 * match, so reject it.
5589 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5590 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5591 set_bit(WriteMostly
, &rdev
->flags
);
5593 clear_bit(WriteMostly
, &rdev
->flags
);
5595 rdev
->raid_disk
= -1;
5596 err
= bind_rdev_to_array(rdev
, mddev
);
5597 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5598 /* If there is hot_add_disk but no hot_remove_disk
5599 * then added disks for geometry changes,
5600 * and should be added immediately.
5602 super_types
[mddev
->major_version
].
5603 validate_super(mddev
, rdev
);
5604 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5606 unbind_rdev_from_array(rdev
);
5611 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5613 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5614 if (mddev
->degraded
)
5615 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5616 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5618 md_new_event(mddev
);
5619 md_wakeup_thread(mddev
->thread
);
5623 /* otherwise, add_new_disk is only allowed
5624 * for major_version==0 superblocks
5626 if (mddev
->major_version
!= 0) {
5627 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5632 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5634 rdev
= md_import_device(dev
, -1, 0);
5637 "md: error, md_import_device() returned %ld\n",
5639 return PTR_ERR(rdev
);
5641 rdev
->desc_nr
= info
->number
;
5642 if (info
->raid_disk
< mddev
->raid_disks
)
5643 rdev
->raid_disk
= info
->raid_disk
;
5645 rdev
->raid_disk
= -1;
5647 if (rdev
->raid_disk
< mddev
->raid_disks
)
5648 if (info
->state
& (1<<MD_DISK_SYNC
))
5649 set_bit(In_sync
, &rdev
->flags
);
5651 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5652 set_bit(WriteMostly
, &rdev
->flags
);
5654 if (!mddev
->persistent
) {
5655 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5656 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5658 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5659 rdev
->sectors
= rdev
->sb_start
;
5661 err
= bind_rdev_to_array(rdev
, mddev
);
5671 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
5673 char b
[BDEVNAME_SIZE
];
5674 struct md_rdev
*rdev
;
5676 rdev
= find_rdev(mddev
, dev
);
5680 clear_bit(Blocked
, &rdev
->flags
);
5681 remove_and_add_spares(mddev
, rdev
);
5683 if (rdev
->raid_disk
>= 0)
5686 kick_rdev_from_array(rdev
);
5687 md_update_sb(mddev
, 1);
5688 md_new_event(mddev
);
5692 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5693 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5697 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
5699 char b
[BDEVNAME_SIZE
];
5701 struct md_rdev
*rdev
;
5706 if (mddev
->major_version
!= 0) {
5707 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5708 " version-0 superblocks.\n",
5712 if (!mddev
->pers
->hot_add_disk
) {
5714 "%s: personality does not support diskops!\n",
5719 rdev
= md_import_device(dev
, -1, 0);
5722 "md: error, md_import_device() returned %ld\n",
5727 if (mddev
->persistent
)
5728 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5730 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5732 rdev
->sectors
= rdev
->sb_start
;
5734 if (test_bit(Faulty
, &rdev
->flags
)) {
5736 "md: can not hot-add faulty %s disk to %s!\n",
5737 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5741 clear_bit(In_sync
, &rdev
->flags
);
5743 rdev
->saved_raid_disk
= -1;
5744 err
= bind_rdev_to_array(rdev
, mddev
);
5749 * The rest should better be atomic, we can have disk failures
5750 * noticed in interrupt contexts ...
5753 rdev
->raid_disk
= -1;
5755 md_update_sb(mddev
, 1);
5758 * Kick recovery, maybe this spare has to be added to the
5759 * array immediately.
5761 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5762 md_wakeup_thread(mddev
->thread
);
5763 md_new_event(mddev
);
5771 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5776 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
5778 if (mddev
->recovery
|| mddev
->sync_thread
)
5780 /* we should be able to change the bitmap.. */
5784 struct inode
*inode
;
5786 return -EEXIST
; /* cannot add when bitmap is present */
5787 mddev
->bitmap_info
.file
= fget(fd
);
5789 if (mddev
->bitmap_info
.file
== NULL
) {
5790 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5795 inode
= mddev
->bitmap_info
.file
->f_mapping
->host
;
5796 if (!S_ISREG(inode
->i_mode
)) {
5797 printk(KERN_ERR
"%s: error: bitmap file must be a regular file\n",
5800 } else if (!(mddev
->bitmap_info
.file
->f_mode
& FMODE_WRITE
)) {
5801 printk(KERN_ERR
"%s: error: bitmap file must open for write\n",
5804 } else if (atomic_read(&inode
->i_writecount
) != 1) {
5805 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5810 fput(mddev
->bitmap_info
.file
);
5811 mddev
->bitmap_info
.file
= NULL
;
5814 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5815 } else if (mddev
->bitmap
== NULL
)
5816 return -ENOENT
; /* cannot remove what isn't there */
5819 mddev
->pers
->quiesce(mddev
, 1);
5821 err
= bitmap_create(mddev
);
5823 err
= bitmap_load(mddev
);
5825 if (fd
< 0 || err
) {
5826 bitmap_destroy(mddev
);
5827 fd
= -1; /* make sure to put the file */
5829 mddev
->pers
->quiesce(mddev
, 0);
5832 if (mddev
->bitmap_info
.file
)
5833 fput(mddev
->bitmap_info
.file
);
5834 mddev
->bitmap_info
.file
= NULL
;
5841 * set_array_info is used two different ways
5842 * The original usage is when creating a new array.
5843 * In this usage, raid_disks is > 0 and it together with
5844 * level, size, not_persistent,layout,chunksize determine the
5845 * shape of the array.
5846 * This will always create an array with a type-0.90.0 superblock.
5847 * The newer usage is when assembling an array.
5848 * In this case raid_disks will be 0, and the major_version field is
5849 * use to determine which style super-blocks are to be found on the devices.
5850 * The minor and patch _version numbers are also kept incase the
5851 * super_block handler wishes to interpret them.
5853 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
5856 if (info
->raid_disks
== 0) {
5857 /* just setting version number for superblock loading */
5858 if (info
->major_version
< 0 ||
5859 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5860 super_types
[info
->major_version
].name
== NULL
) {
5861 /* maybe try to auto-load a module? */
5863 "md: superblock version %d not known\n",
5864 info
->major_version
);
5867 mddev
->major_version
= info
->major_version
;
5868 mddev
->minor_version
= info
->minor_version
;
5869 mddev
->patch_version
= info
->patch_version
;
5870 mddev
->persistent
= !info
->not_persistent
;
5871 /* ensure mddev_put doesn't delete this now that there
5872 * is some minimal configuration.
5874 mddev
->ctime
= get_seconds();
5877 mddev
->major_version
= MD_MAJOR_VERSION
;
5878 mddev
->minor_version
= MD_MINOR_VERSION
;
5879 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5880 mddev
->ctime
= get_seconds();
5882 mddev
->level
= info
->level
;
5883 mddev
->clevel
[0] = 0;
5884 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5885 mddev
->raid_disks
= info
->raid_disks
;
5886 /* don't set md_minor, it is determined by which /dev/md* was
5889 if (info
->state
& (1<<MD_SB_CLEAN
))
5890 mddev
->recovery_cp
= MaxSector
;
5892 mddev
->recovery_cp
= 0;
5893 mddev
->persistent
= ! info
->not_persistent
;
5894 mddev
->external
= 0;
5896 mddev
->layout
= info
->layout
;
5897 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5899 mddev
->max_disks
= MD_SB_DISKS
;
5901 if (mddev
->persistent
)
5903 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5905 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5906 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
5907 mddev
->bitmap_info
.offset
= 0;
5909 mddev
->reshape_position
= MaxSector
;
5912 * Generate a 128 bit UUID
5914 get_random_bytes(mddev
->uuid
, 16);
5916 mddev
->new_level
= mddev
->level
;
5917 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5918 mddev
->new_layout
= mddev
->layout
;
5919 mddev
->delta_disks
= 0;
5920 mddev
->reshape_backwards
= 0;
5925 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
5927 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5929 if (mddev
->external_size
)
5932 mddev
->array_sectors
= array_sectors
;
5934 EXPORT_SYMBOL(md_set_array_sectors
);
5936 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
5938 struct md_rdev
*rdev
;
5940 int fit
= (num_sectors
== 0);
5942 if (mddev
->pers
->resize
== NULL
)
5944 /* The "num_sectors" is the number of sectors of each device that
5945 * is used. This can only make sense for arrays with redundancy.
5946 * linear and raid0 always use whatever space is available. We can only
5947 * consider changing this number if no resync or reconstruction is
5948 * happening, and if the new size is acceptable. It must fit before the
5949 * sb_start or, if that is <data_offset, it must fit before the size
5950 * of each device. If num_sectors is zero, we find the largest size
5953 if (mddev
->sync_thread
)
5958 rdev_for_each(rdev
, mddev
) {
5959 sector_t avail
= rdev
->sectors
;
5961 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5962 num_sectors
= avail
;
5963 if (avail
< num_sectors
)
5966 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5968 revalidate_disk(mddev
->gendisk
);
5972 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
5975 struct md_rdev
*rdev
;
5976 /* change the number of raid disks */
5977 if (mddev
->pers
->check_reshape
== NULL
)
5981 if (raid_disks
<= 0 ||
5982 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5984 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5987 rdev_for_each(rdev
, mddev
) {
5988 if (mddev
->raid_disks
< raid_disks
&&
5989 rdev
->data_offset
< rdev
->new_data_offset
)
5991 if (mddev
->raid_disks
> raid_disks
&&
5992 rdev
->data_offset
> rdev
->new_data_offset
)
5996 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5997 if (mddev
->delta_disks
< 0)
5998 mddev
->reshape_backwards
= 1;
5999 else if (mddev
->delta_disks
> 0)
6000 mddev
->reshape_backwards
= 0;
6002 rv
= mddev
->pers
->check_reshape(mddev
);
6004 mddev
->delta_disks
= 0;
6005 mddev
->reshape_backwards
= 0;
6011 * update_array_info is used to change the configuration of an
6013 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6014 * fields in the info are checked against the array.
6015 * Any differences that cannot be handled will cause an error.
6016 * Normally, only one change can be managed at a time.
6018 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6024 /* calculate expected state,ignoring low bits */
6025 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6026 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6028 if (mddev
->major_version
!= info
->major_version
||
6029 mddev
->minor_version
!= info
->minor_version
||
6030 /* mddev->patch_version != info->patch_version || */
6031 mddev
->ctime
!= info
->ctime
||
6032 mddev
->level
!= info
->level
||
6033 /* mddev->layout != info->layout || */
6034 !mddev
->persistent
!= info
->not_persistent
||
6035 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6036 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6037 ((state
^info
->state
) & 0xfffffe00)
6040 /* Check there is only one change */
6041 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6043 if (mddev
->raid_disks
!= info
->raid_disks
)
6045 if (mddev
->layout
!= info
->layout
)
6047 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6054 if (mddev
->layout
!= info
->layout
) {
6056 * we don't need to do anything at the md level, the
6057 * personality will take care of it all.
6059 if (mddev
->pers
->check_reshape
== NULL
)
6062 mddev
->new_layout
= info
->layout
;
6063 rv
= mddev
->pers
->check_reshape(mddev
);
6065 mddev
->new_layout
= mddev
->layout
;
6069 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6070 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6072 if (mddev
->raid_disks
!= info
->raid_disks
)
6073 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6075 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6076 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
)
6078 if (mddev
->recovery
|| mddev
->sync_thread
)
6080 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6081 /* add the bitmap */
6084 if (mddev
->bitmap_info
.default_offset
== 0)
6086 mddev
->bitmap_info
.offset
=
6087 mddev
->bitmap_info
.default_offset
;
6088 mddev
->bitmap_info
.space
=
6089 mddev
->bitmap_info
.default_space
;
6090 mddev
->pers
->quiesce(mddev
, 1);
6091 rv
= bitmap_create(mddev
);
6093 rv
= bitmap_load(mddev
);
6095 bitmap_destroy(mddev
);
6096 mddev
->pers
->quiesce(mddev
, 0);
6098 /* remove the bitmap */
6101 if (mddev
->bitmap
->storage
.file
)
6103 mddev
->pers
->quiesce(mddev
, 1);
6104 bitmap_destroy(mddev
);
6105 mddev
->pers
->quiesce(mddev
, 0);
6106 mddev
->bitmap_info
.offset
= 0;
6109 md_update_sb(mddev
, 1);
6113 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6115 struct md_rdev
*rdev
;
6118 if (mddev
->pers
== NULL
)
6122 rdev
= find_rdev_rcu(mddev
, dev
);
6126 md_error(mddev
, rdev
);
6127 if (!test_bit(Faulty
, &rdev
->flags
))
6135 * We have a problem here : there is no easy way to give a CHS
6136 * virtual geometry. We currently pretend that we have a 2 heads
6137 * 4 sectors (with a BIG number of cylinders...). This drives
6138 * dosfs just mad... ;-)
6140 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6142 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6146 geo
->cylinders
= mddev
->array_sectors
/ 8;
6150 static inline bool md_ioctl_valid(unsigned int cmd
)
6155 case GET_ARRAY_INFO
:
6156 case GET_BITMAP_FILE
:
6159 case HOT_REMOVE_DISK
:
6162 case RESTART_ARRAY_RW
:
6164 case SET_ARRAY_INFO
:
6165 case SET_BITMAP_FILE
:
6166 case SET_DISK_FAULTY
:
6175 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6176 unsigned int cmd
, unsigned long arg
)
6179 void __user
*argp
= (void __user
*)arg
;
6180 struct mddev
*mddev
= NULL
;
6183 if (!md_ioctl_valid(cmd
))
6188 case GET_ARRAY_INFO
:
6192 if (!capable(CAP_SYS_ADMIN
))
6197 * Commands dealing with the RAID driver but not any
6202 err
= get_version(argp
);
6208 autostart_arrays(arg
);
6215 * Commands creating/starting a new array:
6218 mddev
= bdev
->bd_disk
->private_data
;
6225 /* Some actions do not requires the mutex */
6227 case GET_ARRAY_INFO
:
6228 if (!mddev
->raid_disks
&& !mddev
->external
)
6231 err
= get_array_info(mddev
, argp
);
6235 if (!mddev
->raid_disks
&& !mddev
->external
)
6238 err
= get_disk_info(mddev
, argp
);
6241 case SET_DISK_FAULTY
:
6242 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6246 if (cmd
== ADD_NEW_DISK
)
6247 /* need to ensure md_delayed_delete() has completed */
6248 flush_workqueue(md_misc_wq
);
6250 if (cmd
== HOT_REMOVE_DISK
)
6251 /* need to ensure recovery thread has run */
6252 wait_event_interruptible_timeout(mddev
->sb_wait
,
6253 !test_bit(MD_RECOVERY_NEEDED
,
6255 msecs_to_jiffies(5000));
6256 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
6257 /* Need to flush page cache, and ensure no-one else opens
6260 mutex_lock(&mddev
->open_mutex
);
6261 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
6262 mutex_unlock(&mddev
->open_mutex
);
6266 set_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6267 mutex_unlock(&mddev
->open_mutex
);
6268 sync_blockdev(bdev
);
6270 err
= mddev_lock(mddev
);
6273 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6278 if (cmd
== SET_ARRAY_INFO
) {
6279 mdu_array_info_t info
;
6281 memset(&info
, 0, sizeof(info
));
6282 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6287 err
= update_array_info(mddev
, &info
);
6289 printk(KERN_WARNING
"md: couldn't update"
6290 " array info. %d\n", err
);
6295 if (!list_empty(&mddev
->disks
)) {
6297 "md: array %s already has disks!\n",
6302 if (mddev
->raid_disks
) {
6304 "md: array %s already initialised!\n",
6309 err
= set_array_info(mddev
, &info
);
6311 printk(KERN_WARNING
"md: couldn't set"
6312 " array info. %d\n", err
);
6319 * Commands querying/configuring an existing array:
6321 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6322 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6323 if ((!mddev
->raid_disks
&& !mddev
->external
)
6324 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6325 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6326 && cmd
!= GET_BITMAP_FILE
) {
6332 * Commands even a read-only array can execute:
6335 case GET_BITMAP_FILE
:
6336 err
= get_bitmap_file(mddev
, argp
);
6339 case RESTART_ARRAY_RW
:
6340 err
= restart_array(mddev
);
6344 err
= do_md_stop(mddev
, 0, bdev
);
6348 err
= md_set_readonly(mddev
, bdev
);
6351 case HOT_REMOVE_DISK
:
6352 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6356 /* We can support ADD_NEW_DISK on read-only arrays
6357 * on if we are re-adding a preexisting device.
6358 * So require mddev->pers and MD_DISK_SYNC.
6361 mdu_disk_info_t info
;
6362 if (copy_from_user(&info
, argp
, sizeof(info
)))
6364 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6365 /* Need to clear read-only for this */
6368 err
= add_new_disk(mddev
, &info
);
6374 if (get_user(ro
, (int __user
*)(arg
))) {
6380 /* if the bdev is going readonly the value of mddev->ro
6381 * does not matter, no writes are coming
6386 /* are we are already prepared for writes? */
6390 /* transitioning to readauto need only happen for
6391 * arrays that call md_write_start
6394 err
= restart_array(mddev
);
6397 set_disk_ro(mddev
->gendisk
, 0);
6404 * The remaining ioctls are changing the state of the
6405 * superblock, so we do not allow them on read-only arrays.
6407 if (mddev
->ro
&& mddev
->pers
) {
6408 if (mddev
->ro
== 2) {
6410 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6411 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6412 /* mddev_unlock will wake thread */
6413 /* If a device failed while we were read-only, we
6414 * need to make sure the metadata is updated now.
6416 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6417 mddev_unlock(mddev
);
6418 wait_event(mddev
->sb_wait
,
6419 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6420 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6421 mddev_lock_nointr(mddev
);
6432 mdu_disk_info_t info
;
6433 if (copy_from_user(&info
, argp
, sizeof(info
)))
6436 err
= add_new_disk(mddev
, &info
);
6441 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6445 err
= do_md_run(mddev
);
6448 case SET_BITMAP_FILE
:
6449 err
= set_bitmap_file(mddev
, (int)arg
);
6458 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6460 mddev
->hold_active
= 0;
6461 mddev_unlock(mddev
);
6465 #ifdef CONFIG_COMPAT
6466 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6467 unsigned int cmd
, unsigned long arg
)
6470 case HOT_REMOVE_DISK
:
6472 case SET_DISK_FAULTY
:
6473 case SET_BITMAP_FILE
:
6474 /* These take in integer arg, do not convert */
6477 arg
= (unsigned long)compat_ptr(arg
);
6481 return md_ioctl(bdev
, mode
, cmd
, arg
);
6483 #endif /* CONFIG_COMPAT */
6485 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6488 * Succeed if we can lock the mddev, which confirms that
6489 * it isn't being stopped right now.
6491 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6497 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6498 /* we are racing with mddev_put which is discarding this
6502 /* Wait until bdev->bd_disk is definitely gone */
6503 flush_workqueue(md_misc_wq
);
6504 /* Then retry the open from the top */
6505 return -ERESTARTSYS
;
6507 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6509 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6513 atomic_inc(&mddev
->openers
);
6514 clear_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6515 mutex_unlock(&mddev
->open_mutex
);
6517 check_disk_change(bdev
);
6522 static void md_release(struct gendisk
*disk
, fmode_t mode
)
6524 struct mddev
*mddev
= disk
->private_data
;
6527 atomic_dec(&mddev
->openers
);
6531 static int md_media_changed(struct gendisk
*disk
)
6533 struct mddev
*mddev
= disk
->private_data
;
6535 return mddev
->changed
;
6538 static int md_revalidate(struct gendisk
*disk
)
6540 struct mddev
*mddev
= disk
->private_data
;
6545 static const struct block_device_operations md_fops
=
6547 .owner
= THIS_MODULE
,
6549 .release
= md_release
,
6551 #ifdef CONFIG_COMPAT
6552 .compat_ioctl
= md_compat_ioctl
,
6554 .getgeo
= md_getgeo
,
6555 .media_changed
= md_media_changed
,
6556 .revalidate_disk
= md_revalidate
,
6559 static int md_thread(void *arg
)
6561 struct md_thread
*thread
= arg
;
6564 * md_thread is a 'system-thread', it's priority should be very
6565 * high. We avoid resource deadlocks individually in each
6566 * raid personality. (RAID5 does preallocation) We also use RR and
6567 * the very same RT priority as kswapd, thus we will never get
6568 * into a priority inversion deadlock.
6570 * we definitely have to have equal or higher priority than
6571 * bdflush, otherwise bdflush will deadlock if there are too
6572 * many dirty RAID5 blocks.
6575 allow_signal(SIGKILL
);
6576 while (!kthread_should_stop()) {
6578 /* We need to wait INTERRUPTIBLE so that
6579 * we don't add to the load-average.
6580 * That means we need to be sure no signals are
6583 if (signal_pending(current
))
6584 flush_signals(current
);
6586 wait_event_interruptible_timeout
6588 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6589 || kthread_should_stop(),
6592 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6593 if (!kthread_should_stop())
6594 thread
->run(thread
);
6600 void md_wakeup_thread(struct md_thread
*thread
)
6603 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6604 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6605 wake_up(&thread
->wqueue
);
6608 EXPORT_SYMBOL(md_wakeup_thread
);
6610 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
6611 struct mddev
*mddev
, const char *name
)
6613 struct md_thread
*thread
;
6615 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6619 init_waitqueue_head(&thread
->wqueue
);
6622 thread
->mddev
= mddev
;
6623 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6624 thread
->tsk
= kthread_run(md_thread
, thread
,
6626 mdname(thread
->mddev
),
6628 if (IS_ERR(thread
->tsk
)) {
6634 EXPORT_SYMBOL(md_register_thread
);
6636 void md_unregister_thread(struct md_thread
**threadp
)
6638 struct md_thread
*thread
= *threadp
;
6641 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6642 /* Locking ensures that mddev_unlock does not wake_up a
6643 * non-existent thread
6645 spin_lock(&pers_lock
);
6647 spin_unlock(&pers_lock
);
6649 kthread_stop(thread
->tsk
);
6652 EXPORT_SYMBOL(md_unregister_thread
);
6654 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6656 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6659 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6661 mddev
->pers
->error_handler(mddev
,rdev
);
6662 if (mddev
->degraded
)
6663 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6664 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6665 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6666 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6667 md_wakeup_thread(mddev
->thread
);
6668 if (mddev
->event_work
.func
)
6669 queue_work(md_misc_wq
, &mddev
->event_work
);
6670 md_new_event_inintr(mddev
);
6672 EXPORT_SYMBOL(md_error
);
6674 /* seq_file implementation /proc/mdstat */
6676 static void status_unused(struct seq_file
*seq
)
6679 struct md_rdev
*rdev
;
6681 seq_printf(seq
, "unused devices: ");
6683 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6684 char b
[BDEVNAME_SIZE
];
6686 seq_printf(seq
, "%s ",
6687 bdevname(rdev
->bdev
,b
));
6690 seq_printf(seq
, "<none>");
6692 seq_printf(seq
, "\n");
6695 static void status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
6697 sector_t max_sectors
, resync
, res
;
6698 unsigned long dt
, db
;
6701 unsigned int per_milli
;
6703 if (mddev
->curr_resync
<= 3)
6706 resync
= mddev
->curr_resync
6707 - atomic_read(&mddev
->recovery_active
);
6709 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6710 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6711 max_sectors
= mddev
->resync_max_sectors
;
6713 max_sectors
= mddev
->dev_sectors
;
6715 WARN_ON(max_sectors
== 0);
6716 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6717 * in a sector_t, and (max_sectors>>scale) will fit in a
6718 * u32, as those are the requirements for sector_div.
6719 * Thus 'scale' must be at least 10
6722 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6723 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6726 res
= (resync
>>scale
)*1000;
6727 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6731 int i
, x
= per_milli
/50, y
= 20-x
;
6732 seq_printf(seq
, "[");
6733 for (i
= 0; i
< x
; i
++)
6734 seq_printf(seq
, "=");
6735 seq_printf(seq
, ">");
6736 for (i
= 0; i
< y
; i
++)
6737 seq_printf(seq
, ".");
6738 seq_printf(seq
, "] ");
6740 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6741 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6743 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6745 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6746 "resync" : "recovery"))),
6747 per_milli
/10, per_milli
% 10,
6748 (unsigned long long) resync
/2,
6749 (unsigned long long) max_sectors
/2);
6752 * dt: time from mark until now
6753 * db: blocks written from mark until now
6754 * rt: remaining time
6756 * rt is a sector_t, so could be 32bit or 64bit.
6757 * So we divide before multiply in case it is 32bit and close
6759 * We scale the divisor (db) by 32 to avoid losing precision
6760 * near the end of resync when the number of remaining sectors
6762 * We then divide rt by 32 after multiplying by db to compensate.
6763 * The '+1' avoids division by zero if db is very small.
6765 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6767 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6768 - mddev
->resync_mark_cnt
;
6770 rt
= max_sectors
- resync
; /* number of remaining sectors */
6771 sector_div(rt
, db
/32+1);
6775 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6776 ((unsigned long)rt
% 60)/6);
6778 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6781 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6783 struct list_head
*tmp
;
6785 struct mddev
*mddev
;
6793 spin_lock(&all_mddevs_lock
);
6794 list_for_each(tmp
,&all_mddevs
)
6796 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6798 spin_unlock(&all_mddevs_lock
);
6801 spin_unlock(&all_mddevs_lock
);
6803 return (void*)2;/* tail */
6807 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6809 struct list_head
*tmp
;
6810 struct mddev
*next_mddev
, *mddev
= v
;
6816 spin_lock(&all_mddevs_lock
);
6818 tmp
= all_mddevs
.next
;
6820 tmp
= mddev
->all_mddevs
.next
;
6821 if (tmp
!= &all_mddevs
)
6822 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6824 next_mddev
= (void*)2;
6827 spin_unlock(&all_mddevs_lock
);
6835 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6837 struct mddev
*mddev
= v
;
6839 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6843 static int md_seq_show(struct seq_file
*seq
, void *v
)
6845 struct mddev
*mddev
= v
;
6847 struct md_rdev
*rdev
;
6849 if (v
== (void*)1) {
6850 struct md_personality
*pers
;
6851 seq_printf(seq
, "Personalities : ");
6852 spin_lock(&pers_lock
);
6853 list_for_each_entry(pers
, &pers_list
, list
)
6854 seq_printf(seq
, "[%s] ", pers
->name
);
6856 spin_unlock(&pers_lock
);
6857 seq_printf(seq
, "\n");
6858 seq
->poll_event
= atomic_read(&md_event_count
);
6861 if (v
== (void*)2) {
6866 if (mddev_lock(mddev
) < 0)
6869 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6870 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6871 mddev
->pers
? "" : "in");
6874 seq_printf(seq
, " (read-only)");
6876 seq_printf(seq
, " (auto-read-only)");
6877 seq_printf(seq
, " %s", mddev
->pers
->name
);
6881 rdev_for_each(rdev
, mddev
) {
6882 char b
[BDEVNAME_SIZE
];
6883 seq_printf(seq
, " %s[%d]",
6884 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6885 if (test_bit(WriteMostly
, &rdev
->flags
))
6886 seq_printf(seq
, "(W)");
6887 if (test_bit(Faulty
, &rdev
->flags
)) {
6888 seq_printf(seq
, "(F)");
6891 if (rdev
->raid_disk
< 0)
6892 seq_printf(seq
, "(S)"); /* spare */
6893 if (test_bit(Replacement
, &rdev
->flags
))
6894 seq_printf(seq
, "(R)");
6895 sectors
+= rdev
->sectors
;
6898 if (!list_empty(&mddev
->disks
)) {
6900 seq_printf(seq
, "\n %llu blocks",
6901 (unsigned long long)
6902 mddev
->array_sectors
/ 2);
6904 seq_printf(seq
, "\n %llu blocks",
6905 (unsigned long long)sectors
/ 2);
6907 if (mddev
->persistent
) {
6908 if (mddev
->major_version
!= 0 ||
6909 mddev
->minor_version
!= 90) {
6910 seq_printf(seq
," super %d.%d",
6911 mddev
->major_version
,
6912 mddev
->minor_version
);
6914 } else if (mddev
->external
)
6915 seq_printf(seq
, " super external:%s",
6916 mddev
->metadata_type
);
6918 seq_printf(seq
, " super non-persistent");
6921 mddev
->pers
->status(seq
, mddev
);
6922 seq_printf(seq
, "\n ");
6923 if (mddev
->pers
->sync_request
) {
6924 if (mddev
->curr_resync
> 2) {
6925 status_resync(seq
, mddev
);
6926 seq_printf(seq
, "\n ");
6927 } else if (mddev
->curr_resync
>= 1)
6928 seq_printf(seq
, "\tresync=DELAYED\n ");
6929 else if (mddev
->recovery_cp
< MaxSector
)
6930 seq_printf(seq
, "\tresync=PENDING\n ");
6933 seq_printf(seq
, "\n ");
6935 bitmap_status(seq
, mddev
->bitmap
);
6937 seq_printf(seq
, "\n");
6939 mddev_unlock(mddev
);
6944 static const struct seq_operations md_seq_ops
= {
6945 .start
= md_seq_start
,
6946 .next
= md_seq_next
,
6947 .stop
= md_seq_stop
,
6948 .show
= md_seq_show
,
6951 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6953 struct seq_file
*seq
;
6956 error
= seq_open(file
, &md_seq_ops
);
6960 seq
= file
->private_data
;
6961 seq
->poll_event
= atomic_read(&md_event_count
);
6965 static int md_unloading
;
6966 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6968 struct seq_file
*seq
= filp
->private_data
;
6972 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;;
6973 poll_wait(filp
, &md_event_waiters
, wait
);
6975 /* always allow read */
6976 mask
= POLLIN
| POLLRDNORM
;
6978 if (seq
->poll_event
!= atomic_read(&md_event_count
))
6979 mask
|= POLLERR
| POLLPRI
;
6983 static const struct file_operations md_seq_fops
= {
6984 .owner
= THIS_MODULE
,
6985 .open
= md_seq_open
,
6987 .llseek
= seq_lseek
,
6988 .release
= seq_release_private
,
6989 .poll
= mdstat_poll
,
6992 int register_md_personality(struct md_personality
*p
)
6994 printk(KERN_INFO
"md: %s personality registered for level %d\n",
6996 spin_lock(&pers_lock
);
6997 list_add_tail(&p
->list
, &pers_list
);
6998 spin_unlock(&pers_lock
);
7001 EXPORT_SYMBOL(register_md_personality
);
7003 int unregister_md_personality(struct md_personality
*p
)
7005 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7006 spin_lock(&pers_lock
);
7007 list_del_init(&p
->list
);
7008 spin_unlock(&pers_lock
);
7011 EXPORT_SYMBOL(unregister_md_personality
);
7013 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7015 struct md_rdev
*rdev
;
7021 rdev_for_each_rcu(rdev
, mddev
) {
7022 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7023 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7024 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7025 atomic_read(&disk
->sync_io
);
7026 /* sync IO will cause sync_io to increase before the disk_stats
7027 * as sync_io is counted when a request starts, and
7028 * disk_stats is counted when it completes.
7029 * So resync activity will cause curr_events to be smaller than
7030 * when there was no such activity.
7031 * non-sync IO will cause disk_stat to increase without
7032 * increasing sync_io so curr_events will (eventually)
7033 * be larger than it was before. Once it becomes
7034 * substantially larger, the test below will cause
7035 * the array to appear non-idle, and resync will slow
7037 * If there is a lot of outstanding resync activity when
7038 * we set last_event to curr_events, then all that activity
7039 * completing might cause the array to appear non-idle
7040 * and resync will be slowed down even though there might
7041 * not have been non-resync activity. This will only
7042 * happen once though. 'last_events' will soon reflect
7043 * the state where there is little or no outstanding
7044 * resync requests, and further resync activity will
7045 * always make curr_events less than last_events.
7048 if (init
|| curr_events
- rdev
->last_events
> 64) {
7049 rdev
->last_events
= curr_events
;
7057 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7059 /* another "blocks" (512byte) blocks have been synced */
7060 atomic_sub(blocks
, &mddev
->recovery_active
);
7061 wake_up(&mddev
->recovery_wait
);
7063 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7064 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7065 md_wakeup_thread(mddev
->thread
);
7066 // stop recovery, signal do_sync ....
7069 EXPORT_SYMBOL(md_done_sync
);
7071 /* md_write_start(mddev, bi)
7072 * If we need to update some array metadata (e.g. 'active' flag
7073 * in superblock) before writing, schedule a superblock update
7074 * and wait for it to complete.
7076 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7079 if (bio_data_dir(bi
) != WRITE
)
7082 BUG_ON(mddev
->ro
== 1);
7083 if (mddev
->ro
== 2) {
7084 /* need to switch to read/write */
7086 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7087 md_wakeup_thread(mddev
->thread
);
7088 md_wakeup_thread(mddev
->sync_thread
);
7091 atomic_inc(&mddev
->writes_pending
);
7092 if (mddev
->safemode
== 1)
7093 mddev
->safemode
= 0;
7094 if (mddev
->in_sync
) {
7095 spin_lock_irq(&mddev
->write_lock
);
7096 if (mddev
->in_sync
) {
7098 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7099 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7100 md_wakeup_thread(mddev
->thread
);
7103 spin_unlock_irq(&mddev
->write_lock
);
7106 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7107 wait_event(mddev
->sb_wait
,
7108 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7110 EXPORT_SYMBOL(md_write_start
);
7112 void md_write_end(struct mddev
*mddev
)
7114 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7115 if (mddev
->safemode
== 2)
7116 md_wakeup_thread(mddev
->thread
);
7117 else if (mddev
->safemode_delay
)
7118 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7121 EXPORT_SYMBOL(md_write_end
);
7123 /* md_allow_write(mddev)
7124 * Calling this ensures that the array is marked 'active' so that writes
7125 * may proceed without blocking. It is important to call this before
7126 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7127 * Must be called with mddev_lock held.
7129 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7130 * is dropped, so return -EAGAIN after notifying userspace.
7132 int md_allow_write(struct mddev
*mddev
)
7138 if (!mddev
->pers
->sync_request
)
7141 spin_lock_irq(&mddev
->write_lock
);
7142 if (mddev
->in_sync
) {
7144 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7145 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7146 if (mddev
->safemode_delay
&&
7147 mddev
->safemode
== 0)
7148 mddev
->safemode
= 1;
7149 spin_unlock_irq(&mddev
->write_lock
);
7150 md_update_sb(mddev
, 0);
7151 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7153 spin_unlock_irq(&mddev
->write_lock
);
7155 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7160 EXPORT_SYMBOL_GPL(md_allow_write
);
7162 #define SYNC_MARKS 10
7163 #define SYNC_MARK_STEP (3*HZ)
7164 #define UPDATE_FREQUENCY (5*60*HZ)
7165 void md_do_sync(struct md_thread
*thread
)
7167 struct mddev
*mddev
= thread
->mddev
;
7168 struct mddev
*mddev2
;
7169 unsigned int currspeed
= 0,
7171 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
7172 unsigned long mark
[SYNC_MARKS
];
7173 unsigned long update_time
;
7174 sector_t mark_cnt
[SYNC_MARKS
];
7176 struct list_head
*tmp
;
7177 sector_t last_check
;
7179 struct md_rdev
*rdev
;
7180 char *desc
, *action
= NULL
;
7181 struct blk_plug plug
;
7183 /* just incase thread restarts... */
7184 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7186 if (mddev
->ro
) {/* never try to sync a read-only array */
7187 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7191 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7192 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
7193 desc
= "data-check";
7195 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7196 desc
= "requested-resync";
7200 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7205 mddev
->last_sync_action
= action
?: desc
;
7207 /* we overload curr_resync somewhat here.
7208 * 0 == not engaged in resync at all
7209 * 2 == checking that there is no conflict with another sync
7210 * 1 == like 2, but have yielded to allow conflicting resync to
7212 * other == active in resync - this many blocks
7214 * Before starting a resync we must have set curr_resync to
7215 * 2, and then checked that every "conflicting" array has curr_resync
7216 * less than ours. When we find one that is the same or higher
7217 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7218 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7219 * This will mean we have to start checking from the beginning again.
7224 mddev
->curr_resync
= 2;
7227 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7229 for_each_mddev(mddev2
, tmp
) {
7230 if (mddev2
== mddev
)
7232 if (!mddev
->parallel_resync
7233 && mddev2
->curr_resync
7234 && match_mddev_units(mddev
, mddev2
)) {
7236 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7237 /* arbitrarily yield */
7238 mddev
->curr_resync
= 1;
7239 wake_up(&resync_wait
);
7241 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7242 /* no need to wait here, we can wait the next
7243 * time 'round when curr_resync == 2
7246 /* We need to wait 'interruptible' so as not to
7247 * contribute to the load average, and not to
7248 * be caught by 'softlockup'
7250 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7251 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7252 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7253 printk(KERN_INFO
"md: delaying %s of %s"
7254 " until %s has finished (they"
7255 " share one or more physical units)\n",
7256 desc
, mdname(mddev
), mdname(mddev2
));
7258 if (signal_pending(current
))
7259 flush_signals(current
);
7261 finish_wait(&resync_wait
, &wq
);
7264 finish_wait(&resync_wait
, &wq
);
7267 } while (mddev
->curr_resync
< 2);
7270 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7271 /* resync follows the size requested by the personality,
7272 * which defaults to physical size, but can be virtual size
7274 max_sectors
= mddev
->resync_max_sectors
;
7275 atomic64_set(&mddev
->resync_mismatches
, 0);
7276 /* we don't use the checkpoint if there's a bitmap */
7277 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7278 j
= mddev
->resync_min
;
7279 else if (!mddev
->bitmap
)
7280 j
= mddev
->recovery_cp
;
7282 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7283 max_sectors
= mddev
->resync_max_sectors
;
7285 /* recovery follows the physical size of devices */
7286 max_sectors
= mddev
->dev_sectors
;
7289 rdev_for_each_rcu(rdev
, mddev
)
7290 if (rdev
->raid_disk
>= 0 &&
7291 !test_bit(Faulty
, &rdev
->flags
) &&
7292 !test_bit(In_sync
, &rdev
->flags
) &&
7293 rdev
->recovery_offset
< j
)
7294 j
= rdev
->recovery_offset
;
7297 /* If there is a bitmap, we need to make sure all
7298 * writes that started before we added a spare
7299 * complete before we start doing a recovery.
7300 * Otherwise the write might complete and (via
7301 * bitmap_endwrite) set a bit in the bitmap after the
7302 * recovery has checked that bit and skipped that
7305 if (mddev
->bitmap
) {
7306 mddev
->pers
->quiesce(mddev
, 1);
7307 mddev
->pers
->quiesce(mddev
, 0);
7311 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7312 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7313 " %d KB/sec/disk.\n", speed_min(mddev
));
7314 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7315 "(but not more than %d KB/sec) for %s.\n",
7316 speed_max(mddev
), desc
);
7318 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7321 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7323 mark_cnt
[m
] = io_sectors
;
7326 mddev
->resync_mark
= mark
[last_mark
];
7327 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7330 * Tune reconstruction:
7332 window
= 32*(PAGE_SIZE
/512);
7333 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7334 window
/2, (unsigned long long)max_sectors
/2);
7336 atomic_set(&mddev
->recovery_active
, 0);
7341 "md: resuming %s of %s from checkpoint.\n",
7342 desc
, mdname(mddev
));
7343 mddev
->curr_resync
= j
;
7345 mddev
->curr_resync
= 3; /* no longer delayed */
7346 mddev
->curr_resync_completed
= j
;
7347 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7348 md_new_event(mddev
);
7349 update_time
= jiffies
;
7351 blk_start_plug(&plug
);
7352 while (j
< max_sectors
) {
7357 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7358 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7359 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7360 > (max_sectors
>> 4)) ||
7361 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7362 (j
- mddev
->curr_resync_completed
)*2
7363 >= mddev
->resync_max
- mddev
->curr_resync_completed
7365 /* time to update curr_resync_completed */
7366 wait_event(mddev
->recovery_wait
,
7367 atomic_read(&mddev
->recovery_active
) == 0);
7368 mddev
->curr_resync_completed
= j
;
7369 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
7370 j
> mddev
->recovery_cp
)
7371 mddev
->recovery_cp
= j
;
7372 update_time
= jiffies
;
7373 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7374 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7377 while (j
>= mddev
->resync_max
&&
7378 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7379 /* As this condition is controlled by user-space,
7380 * we can block indefinitely, so use '_interruptible'
7381 * to avoid triggering warnings.
7383 flush_signals(current
); /* just in case */
7384 wait_event_interruptible(mddev
->recovery_wait
,
7385 mddev
->resync_max
> j
7386 || test_bit(MD_RECOVERY_INTR
,
7390 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7393 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7394 currspeed
< speed_min(mddev
));
7396 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7400 if (!skipped
) { /* actual IO requested */
7401 io_sectors
+= sectors
;
7402 atomic_add(sectors
, &mddev
->recovery_active
);
7405 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7410 mddev
->curr_resync
= j
;
7411 mddev
->curr_mark_cnt
= io_sectors
;
7412 if (last_check
== 0)
7413 /* this is the earliest that rebuild will be
7414 * visible in /proc/mdstat
7416 md_new_event(mddev
);
7418 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7421 last_check
= io_sectors
;
7423 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7425 int next
= (last_mark
+1) % SYNC_MARKS
;
7427 mddev
->resync_mark
= mark
[next
];
7428 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7429 mark
[next
] = jiffies
;
7430 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7434 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7438 * this loop exits only if either when we are slower than
7439 * the 'hard' speed limit, or the system was IO-idle for
7441 * the system might be non-idle CPU-wise, but we only care
7442 * about not overloading the IO subsystem. (things like an
7443 * e2fsck being done on the RAID array should execute fast)
7447 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
7448 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
7449 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7451 if (currspeed
> speed_min(mddev
)) {
7452 if ((currspeed
> speed_max(mddev
)) ||
7453 !is_mddev_idle(mddev
, 0)) {
7459 printk(KERN_INFO
"md: %s: %s %s.\n",mdname(mddev
), desc
,
7460 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
7461 ? "interrupted" : "done");
7463 * this also signals 'finished resyncing' to md_stop
7465 blk_finish_plug(&plug
);
7466 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7468 /* tell personality that we are finished */
7469 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7471 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7472 mddev
->curr_resync
> 2) {
7473 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7474 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7475 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7477 "md: checkpointing %s of %s.\n",
7478 desc
, mdname(mddev
));
7479 if (test_bit(MD_RECOVERY_ERROR
,
7481 mddev
->recovery_cp
=
7482 mddev
->curr_resync_completed
;
7484 mddev
->recovery_cp
=
7488 mddev
->recovery_cp
= MaxSector
;
7490 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7491 mddev
->curr_resync
= MaxSector
;
7493 rdev_for_each_rcu(rdev
, mddev
)
7494 if (rdev
->raid_disk
>= 0 &&
7495 mddev
->delta_disks
>= 0 &&
7496 !test_bit(Faulty
, &rdev
->flags
) &&
7497 !test_bit(In_sync
, &rdev
->flags
) &&
7498 rdev
->recovery_offset
< mddev
->curr_resync
)
7499 rdev
->recovery_offset
= mddev
->curr_resync
;
7504 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7506 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7507 /* We completed so min/max setting can be forgotten if used. */
7508 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7509 mddev
->resync_min
= 0;
7510 mddev
->resync_max
= MaxSector
;
7511 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7512 mddev
->resync_min
= mddev
->curr_resync_completed
;
7513 mddev
->curr_resync
= 0;
7514 wake_up(&resync_wait
);
7515 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7516 md_wakeup_thread(mddev
->thread
);
7519 EXPORT_SYMBOL_GPL(md_do_sync
);
7521 static int remove_and_add_spares(struct mddev
*mddev
,
7522 struct md_rdev
*this)
7524 struct md_rdev
*rdev
;
7528 rdev_for_each(rdev
, mddev
)
7529 if ((this == NULL
|| rdev
== this) &&
7530 rdev
->raid_disk
>= 0 &&
7531 !test_bit(Blocked
, &rdev
->flags
) &&
7532 (test_bit(Faulty
, &rdev
->flags
) ||
7533 ! test_bit(In_sync
, &rdev
->flags
)) &&
7534 atomic_read(&rdev
->nr_pending
)==0) {
7535 if (mddev
->pers
->hot_remove_disk(
7536 mddev
, rdev
) == 0) {
7537 sysfs_unlink_rdev(mddev
, rdev
);
7538 rdev
->raid_disk
= -1;
7542 if (removed
&& mddev
->kobj
.sd
)
7543 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
7548 rdev_for_each(rdev
, mddev
) {
7549 if (rdev
->raid_disk
>= 0 &&
7550 !test_bit(In_sync
, &rdev
->flags
) &&
7551 !test_bit(Faulty
, &rdev
->flags
))
7553 if (rdev
->raid_disk
>= 0)
7555 if (test_bit(Faulty
, &rdev
->flags
))
7558 ! (rdev
->saved_raid_disk
>= 0 &&
7559 !test_bit(Bitmap_sync
, &rdev
->flags
)))
7562 if (rdev
->saved_raid_disk
< 0)
7563 rdev
->recovery_offset
= 0;
7565 hot_add_disk(mddev
, rdev
) == 0) {
7566 if (sysfs_link_rdev(mddev
, rdev
))
7567 /* failure here is OK */;
7569 md_new_event(mddev
);
7570 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7575 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7579 static void md_start_sync(struct work_struct
*ws
)
7581 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
7583 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7586 if (!mddev
->sync_thread
) {
7587 printk(KERN_ERR
"%s: could not start resync"
7590 /* leave the spares where they are, it shouldn't hurt */
7591 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7592 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7593 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7594 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7595 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7596 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7598 if (mddev
->sysfs_action
)
7599 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7601 md_wakeup_thread(mddev
->sync_thread
);
7602 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7603 md_new_event(mddev
);
7607 * This routine is regularly called by all per-raid-array threads to
7608 * deal with generic issues like resync and super-block update.
7609 * Raid personalities that don't have a thread (linear/raid0) do not
7610 * need this as they never do any recovery or update the superblock.
7612 * It does not do any resync itself, but rather "forks" off other threads
7613 * to do that as needed.
7614 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7615 * "->recovery" and create a thread at ->sync_thread.
7616 * When the thread finishes it sets MD_RECOVERY_DONE
7617 * and wakeups up this thread which will reap the thread and finish up.
7618 * This thread also removes any faulty devices (with nr_pending == 0).
7620 * The overall approach is:
7621 * 1/ if the superblock needs updating, update it.
7622 * 2/ If a recovery thread is running, don't do anything else.
7623 * 3/ If recovery has finished, clean up, possibly marking spares active.
7624 * 4/ If there are any faulty devices, remove them.
7625 * 5/ If array is degraded, try to add spares devices
7626 * 6/ If array has spares or is not in-sync, start a resync thread.
7628 void md_check_recovery(struct mddev
*mddev
)
7630 if (mddev
->suspended
)
7634 bitmap_daemon_work(mddev
);
7636 if (signal_pending(current
)) {
7637 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7638 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7640 mddev
->safemode
= 2;
7642 flush_signals(current
);
7645 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7648 (mddev
->flags
& MD_UPDATE_SB_FLAGS
& ~ (1<<MD_CHANGE_PENDING
)) ||
7649 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7650 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7651 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7652 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7653 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7657 if (mddev_trylock(mddev
)) {
7661 /* On a read-only array we can:
7662 * - remove failed devices
7663 * - add already-in_sync devices if the array itself
7665 * As we only add devices that are already in-sync,
7666 * we can activate the spares immediately.
7668 remove_and_add_spares(mddev
, NULL
);
7669 /* There is no thread, but we need to call
7670 * ->spare_active and clear saved_raid_disk
7672 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7673 md_reap_sync_thread(mddev
);
7674 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7678 if (!mddev
->external
) {
7680 spin_lock_irq(&mddev
->write_lock
);
7681 if (mddev
->safemode
&&
7682 !atomic_read(&mddev
->writes_pending
) &&
7684 mddev
->recovery_cp
== MaxSector
) {
7687 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7689 if (mddev
->safemode
== 1)
7690 mddev
->safemode
= 0;
7691 spin_unlock_irq(&mddev
->write_lock
);
7693 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7696 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
7697 md_update_sb(mddev
, 0);
7699 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7700 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7701 /* resync/recovery still happening */
7702 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7705 if (mddev
->sync_thread
) {
7706 md_reap_sync_thread(mddev
);
7709 /* Set RUNNING before clearing NEEDED to avoid
7710 * any transients in the value of "sync_action".
7712 mddev
->curr_resync_completed
= 0;
7713 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7714 /* Clear some bits that don't mean anything, but
7717 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7718 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7720 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7721 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7723 /* no recovery is running.
7724 * remove any failed drives, then
7725 * add spares if possible.
7726 * Spares are also removed and re-added, to allow
7727 * the personality to fail the re-add.
7730 if (mddev
->reshape_position
!= MaxSector
) {
7731 if (mddev
->pers
->check_reshape
== NULL
||
7732 mddev
->pers
->check_reshape(mddev
) != 0)
7733 /* Cannot proceed */
7735 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7736 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7737 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
7738 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7739 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7740 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7741 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7742 } else if (mddev
->recovery_cp
< MaxSector
) {
7743 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7744 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7745 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7746 /* nothing to be done ... */
7749 if (mddev
->pers
->sync_request
) {
7751 /* We are adding a device or devices to an array
7752 * which has the bitmap stored on all devices.
7753 * So make sure all bitmap pages get written
7755 bitmap_write_all(mddev
->bitmap
);
7757 INIT_WORK(&mddev
->del_work
, md_start_sync
);
7758 queue_work(md_misc_wq
, &mddev
->del_work
);
7762 if (!mddev
->sync_thread
) {
7763 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7764 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7766 if (mddev
->sysfs_action
)
7767 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7770 wake_up(&mddev
->sb_wait
);
7771 mddev_unlock(mddev
);
7774 EXPORT_SYMBOL(md_check_recovery
);
7776 void md_reap_sync_thread(struct mddev
*mddev
)
7778 struct md_rdev
*rdev
;
7780 /* resync has finished, collect result */
7781 md_unregister_thread(&mddev
->sync_thread
);
7782 wake_up(&resync_wait
);
7783 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7784 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7786 /* activate any spares */
7787 if (mddev
->pers
->spare_active(mddev
)) {
7788 sysfs_notify(&mddev
->kobj
, NULL
,
7790 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7793 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7794 mddev
->pers
->finish_reshape
)
7795 mddev
->pers
->finish_reshape(mddev
);
7797 /* If array is no-longer degraded, then any saved_raid_disk
7798 * information must be scrapped.
7800 if (!mddev
->degraded
)
7801 rdev_for_each(rdev
, mddev
)
7802 rdev
->saved_raid_disk
= -1;
7804 md_update_sb(mddev
, 1);
7805 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7806 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7807 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7808 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7809 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7810 /* flag recovery needed just to double check */
7811 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7812 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7813 md_new_event(mddev
);
7814 if (mddev
->event_work
.func
)
7815 queue_work(md_misc_wq
, &mddev
->event_work
);
7817 EXPORT_SYMBOL(md_reap_sync_thread
);
7819 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7821 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7822 wait_event_timeout(rdev
->blocked_wait
,
7823 !test_bit(Blocked
, &rdev
->flags
) &&
7824 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7825 msecs_to_jiffies(5000));
7826 rdev_dec_pending(rdev
, mddev
);
7828 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7830 void md_finish_reshape(struct mddev
*mddev
)
7832 /* called be personality module when reshape completes. */
7833 struct md_rdev
*rdev
;
7835 rdev_for_each(rdev
, mddev
) {
7836 if (rdev
->data_offset
> rdev
->new_data_offset
)
7837 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
7839 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
7840 rdev
->data_offset
= rdev
->new_data_offset
;
7843 EXPORT_SYMBOL(md_finish_reshape
);
7845 /* Bad block management.
7846 * We can record which blocks on each device are 'bad' and so just
7847 * fail those blocks, or that stripe, rather than the whole device.
7848 * Entries in the bad-block table are 64bits wide. This comprises:
7849 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7850 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7851 * A 'shift' can be set so that larger blocks are tracked and
7852 * consequently larger devices can be covered.
7853 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7855 * Locking of the bad-block table uses a seqlock so md_is_badblock
7856 * might need to retry if it is very unlucky.
7857 * We will sometimes want to check for bad blocks in a bi_end_io function,
7858 * so we use the write_seqlock_irq variant.
7860 * When looking for a bad block we specify a range and want to
7861 * know if any block in the range is bad. So we binary-search
7862 * to the last range that starts at-or-before the given endpoint,
7863 * (or "before the sector after the target range")
7864 * then see if it ends after the given start.
7866 * 0 if there are no known bad blocks in the range
7867 * 1 if there are known bad block which are all acknowledged
7868 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7869 * plus the start/length of the first bad section we overlap.
7871 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7872 sector_t
*first_bad
, int *bad_sectors
)
7878 sector_t target
= s
+ sectors
;
7881 if (bb
->shift
> 0) {
7882 /* round the start down, and the end up */
7884 target
+= (1<<bb
->shift
) - 1;
7885 target
>>= bb
->shift
;
7886 sectors
= target
- s
;
7888 /* 'target' is now the first block after the bad range */
7891 seq
= read_seqbegin(&bb
->lock
);
7896 /* Binary search between lo and hi for 'target'
7897 * i.e. for the last range that starts before 'target'
7899 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7900 * are known not to be the last range before target.
7901 * VARIANT: hi-lo is the number of possible
7902 * ranges, and decreases until it reaches 1
7904 while (hi
- lo
> 1) {
7905 int mid
= (lo
+ hi
) / 2;
7906 sector_t a
= BB_OFFSET(p
[mid
]);
7908 /* This could still be the one, earlier ranges
7912 /* This and later ranges are definitely out. */
7915 /* 'lo' might be the last that started before target, but 'hi' isn't */
7917 /* need to check all range that end after 's' to see if
7918 * any are unacknowledged.
7921 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7922 if (BB_OFFSET(p
[lo
]) < target
) {
7923 /* starts before the end, and finishes after
7924 * the start, so they must overlap
7926 if (rv
!= -1 && BB_ACK(p
[lo
]))
7930 *first_bad
= BB_OFFSET(p
[lo
]);
7931 *bad_sectors
= BB_LEN(p
[lo
]);
7937 if (read_seqretry(&bb
->lock
, seq
))
7942 EXPORT_SYMBOL_GPL(md_is_badblock
);
7945 * Add a range of bad blocks to the table.
7946 * This might extend the table, or might contract it
7947 * if two adjacent ranges can be merged.
7948 * We binary-search to find the 'insertion' point, then
7949 * decide how best to handle it.
7951 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7957 unsigned long flags
;
7960 /* badblocks are disabled */
7964 /* round the start down, and the end up */
7965 sector_t next
= s
+ sectors
;
7967 next
+= (1<<bb
->shift
) - 1;
7972 write_seqlock_irqsave(&bb
->lock
, flags
);
7977 /* Find the last range that starts at-or-before 's' */
7978 while (hi
- lo
> 1) {
7979 int mid
= (lo
+ hi
) / 2;
7980 sector_t a
= BB_OFFSET(p
[mid
]);
7986 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7990 /* we found a range that might merge with the start
7993 sector_t a
= BB_OFFSET(p
[lo
]);
7994 sector_t e
= a
+ BB_LEN(p
[lo
]);
7995 int ack
= BB_ACK(p
[lo
]);
7997 /* Yes, we can merge with a previous range */
7998 if (s
== a
&& s
+ sectors
>= e
)
7999 /* new range covers old */
8002 ack
= ack
&& acknowledged
;
8004 if (e
< s
+ sectors
)
8006 if (e
- a
<= BB_MAX_LEN
) {
8007 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8010 /* does not all fit in one range,
8011 * make p[lo] maximal
8013 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8014 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8020 if (sectors
&& hi
< bb
->count
) {
8021 /* 'hi' points to the first range that starts after 's'.
8022 * Maybe we can merge with the start of that range */
8023 sector_t a
= BB_OFFSET(p
[hi
]);
8024 sector_t e
= a
+ BB_LEN(p
[hi
]);
8025 int ack
= BB_ACK(p
[hi
]);
8026 if (a
<= s
+ sectors
) {
8027 /* merging is possible */
8028 if (e
<= s
+ sectors
) {
8033 ack
= ack
&& acknowledged
;
8036 if (e
- a
<= BB_MAX_LEN
) {
8037 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8040 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8048 if (sectors
== 0 && hi
< bb
->count
) {
8049 /* we might be able to combine lo and hi */
8050 /* Note: 's' is at the end of 'lo' */
8051 sector_t a
= BB_OFFSET(p
[hi
]);
8052 int lolen
= BB_LEN(p
[lo
]);
8053 int hilen
= BB_LEN(p
[hi
]);
8054 int newlen
= lolen
+ hilen
- (s
- a
);
8055 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8056 /* yes, we can combine them */
8057 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8058 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8059 memmove(p
+ hi
, p
+ hi
+ 1,
8060 (bb
->count
- hi
- 1) * 8);
8065 /* didn't merge (it all).
8066 * Need to add a range just before 'hi' */
8067 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8068 /* No room for more */
8072 int this_sectors
= sectors
;
8073 memmove(p
+ hi
+ 1, p
+ hi
,
8074 (bb
->count
- hi
) * 8);
8077 if (this_sectors
> BB_MAX_LEN
)
8078 this_sectors
= BB_MAX_LEN
;
8079 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8080 sectors
-= this_sectors
;
8087 bb
->unacked_exist
= 1;
8088 write_sequnlock_irqrestore(&bb
->lock
, flags
);
8093 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8098 s
+= rdev
->new_data_offset
;
8100 s
+= rdev
->data_offset
;
8101 rv
= md_set_badblocks(&rdev
->badblocks
,
8104 /* Make sure they get written out promptly */
8105 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8106 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8107 md_wakeup_thread(rdev
->mddev
->thread
);
8111 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8114 * Remove a range of bad blocks from the table.
8115 * This may involve extending the table if we spilt a region,
8116 * but it must not fail. So if the table becomes full, we just
8117 * drop the remove request.
8119 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8123 sector_t target
= s
+ sectors
;
8126 if (bb
->shift
> 0) {
8127 /* When clearing we round the start up and the end down.
8128 * This should not matter as the shift should align with
8129 * the block size and no rounding should ever be needed.
8130 * However it is better the think a block is bad when it
8131 * isn't than to think a block is not bad when it is.
8133 s
+= (1<<bb
->shift
) - 1;
8135 target
>>= bb
->shift
;
8136 sectors
= target
- s
;
8139 write_seqlock_irq(&bb
->lock
);
8144 /* Find the last range that starts before 'target' */
8145 while (hi
- lo
> 1) {
8146 int mid
= (lo
+ hi
) / 2;
8147 sector_t a
= BB_OFFSET(p
[mid
]);
8154 /* p[lo] is the last range that could overlap the
8155 * current range. Earlier ranges could also overlap,
8156 * but only this one can overlap the end of the range.
8158 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8159 /* Partial overlap, leave the tail of this range */
8160 int ack
= BB_ACK(p
[lo
]);
8161 sector_t a
= BB_OFFSET(p
[lo
]);
8162 sector_t end
= a
+ BB_LEN(p
[lo
]);
8165 /* we need to split this range */
8166 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8170 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8172 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8175 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8176 /* there is no longer an overlap */
8181 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8182 /* This range does overlap */
8183 if (BB_OFFSET(p
[lo
]) < s
) {
8184 /* Keep the early parts of this range. */
8185 int ack
= BB_ACK(p
[lo
]);
8186 sector_t start
= BB_OFFSET(p
[lo
]);
8187 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8188 /* now low doesn't overlap, so.. */
8193 /* 'lo' is strictly before, 'hi' is strictly after,
8194 * anything between needs to be discarded
8197 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8198 bb
->count
-= (hi
- lo
- 1);
8204 write_sequnlock_irq(&bb
->lock
);
8208 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8212 s
+= rdev
->new_data_offset
;
8214 s
+= rdev
->data_offset
;
8215 return md_clear_badblocks(&rdev
->badblocks
,
8218 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8221 * Acknowledge all bad blocks in a list.
8222 * This only succeeds if ->changed is clear. It is used by
8223 * in-kernel metadata updates
8225 void md_ack_all_badblocks(struct badblocks
*bb
)
8227 if (bb
->page
== NULL
|| bb
->changed
)
8228 /* no point even trying */
8230 write_seqlock_irq(&bb
->lock
);
8232 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8235 for (i
= 0; i
< bb
->count
; i
++) {
8236 if (!BB_ACK(p
[i
])) {
8237 sector_t start
= BB_OFFSET(p
[i
]);
8238 int len
= BB_LEN(p
[i
]);
8239 p
[i
] = BB_MAKE(start
, len
, 1);
8242 bb
->unacked_exist
= 0;
8244 write_sequnlock_irq(&bb
->lock
);
8246 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8248 /* sysfs access to bad-blocks list.
8249 * We present two files.
8250 * 'bad-blocks' lists sector numbers and lengths of ranges that
8251 * are recorded as bad. The list is truncated to fit within
8252 * the one-page limit of sysfs.
8253 * Writing "sector length" to this file adds an acknowledged
8255 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8256 * been acknowledged. Writing to this file adds bad blocks
8257 * without acknowledging them. This is largely for testing.
8261 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8272 seq
= read_seqbegin(&bb
->lock
);
8277 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8278 sector_t s
= BB_OFFSET(p
[i
]);
8279 unsigned int length
= BB_LEN(p
[i
]);
8280 int ack
= BB_ACK(p
[i
]);
8286 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8287 (unsigned long long)s
<< bb
->shift
,
8288 length
<< bb
->shift
);
8290 if (unack
&& len
== 0)
8291 bb
->unacked_exist
= 0;
8293 if (read_seqretry(&bb
->lock
, seq
))
8302 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8304 unsigned long long sector
;
8308 /* Allow clearing via sysfs *only* for testing/debugging.
8309 * Normally only a successful write may clear a badblock
8312 if (page
[0] == '-') {
8316 #endif /* DO_DEBUG */
8318 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8320 if (newline
!= '\n')
8332 md_clear_badblocks(bb
, sector
, length
);
8335 #endif /* DO_DEBUG */
8336 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8342 static int md_notify_reboot(struct notifier_block
*this,
8343 unsigned long code
, void *x
)
8345 struct list_head
*tmp
;
8346 struct mddev
*mddev
;
8349 for_each_mddev(mddev
, tmp
) {
8350 if (mddev_trylock(mddev
)) {
8352 __md_stop_writes(mddev
);
8353 if (mddev
->persistent
)
8354 mddev
->safemode
= 2;
8355 mddev_unlock(mddev
);
8360 * certain more exotic SCSI devices are known to be
8361 * volatile wrt too early system reboots. While the
8362 * right place to handle this issue is the given
8363 * driver, we do want to have a safe RAID driver ...
8371 static struct notifier_block md_notifier
= {
8372 .notifier_call
= md_notify_reboot
,
8374 .priority
= INT_MAX
, /* before any real devices */
8377 static void md_geninit(void)
8379 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8381 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8384 static int __init
md_init(void)
8388 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8392 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8396 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8399 if ((ret
= register_blkdev(0, "mdp")) < 0)
8403 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
8404 md_probe
, NULL
, NULL
);
8405 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8406 md_probe
, NULL
, NULL
);
8408 register_reboot_notifier(&md_notifier
);
8409 raid_table_header
= register_sysctl_table(raid_root_table
);
8415 unregister_blkdev(MD_MAJOR
, "md");
8417 destroy_workqueue(md_misc_wq
);
8419 destroy_workqueue(md_wq
);
8427 * Searches all registered partitions for autorun RAID arrays
8431 static LIST_HEAD(all_detected_devices
);
8432 struct detected_devices_node
{
8433 struct list_head list
;
8437 void md_autodetect_dev(dev_t dev
)
8439 struct detected_devices_node
*node_detected_dev
;
8441 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8442 if (node_detected_dev
) {
8443 node_detected_dev
->dev
= dev
;
8444 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8446 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8447 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8451 static void autostart_arrays(int part
)
8453 struct md_rdev
*rdev
;
8454 struct detected_devices_node
*node_detected_dev
;
8456 int i_scanned
, i_passed
;
8461 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8463 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8465 node_detected_dev
= list_entry(all_detected_devices
.next
,
8466 struct detected_devices_node
, list
);
8467 list_del(&node_detected_dev
->list
);
8468 dev
= node_detected_dev
->dev
;
8469 kfree(node_detected_dev
);
8470 rdev
= md_import_device(dev
,0, 90);
8474 if (test_bit(Faulty
, &rdev
->flags
))
8477 set_bit(AutoDetected
, &rdev
->flags
);
8478 list_add(&rdev
->same_set
, &pending_raid_disks
);
8482 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8483 i_scanned
, i_passed
);
8485 autorun_devices(part
);
8488 #endif /* !MODULE */
8490 static __exit
void md_exit(void)
8492 struct mddev
*mddev
;
8493 struct list_head
*tmp
;
8496 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
8497 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8499 unregister_blkdev(MD_MAJOR
,"md");
8500 unregister_blkdev(mdp_major
, "mdp");
8501 unregister_reboot_notifier(&md_notifier
);
8502 unregister_sysctl_table(raid_table_header
);
8504 /* We cannot unload the modules while some process is
8505 * waiting for us in select() or poll() - wake them up
8508 while (waitqueue_active(&md_event_waiters
)) {
8509 /* not safe to leave yet */
8510 wake_up(&md_event_waiters
);
8514 remove_proc_entry("mdstat", NULL
);
8516 for_each_mddev(mddev
, tmp
) {
8517 export_array(mddev
);
8518 mddev
->hold_active
= 0;
8520 destroy_workqueue(md_misc_wq
);
8521 destroy_workqueue(md_wq
);
8524 subsys_initcall(md_init
);
8525 module_exit(md_exit
)
8527 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8529 return sprintf(buffer
, "%d", start_readonly
);
8531 static int set_ro(const char *val
, struct kernel_param
*kp
)
8534 int num
= simple_strtoul(val
, &e
, 10);
8535 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8536 start_readonly
= num
;
8542 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8543 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8544 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8546 MODULE_LICENSE("GPL");
8547 MODULE_DESCRIPTION("MD RAID framework");
8549 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);