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>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58 static void autostart_arrays(int part
);
61 static LIST_HEAD(pers_list
);
62 static DEFINE_SPINLOCK(pers_lock
);
64 static void md_print_devices(void);
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
83 static int sysctl_speed_limit_min
= 1000;
84 static int sysctl_speed_limit_max
= 200000;
85 static inline int speed_min(mddev_t
*mddev
)
87 return mddev
->sync_speed_min
?
88 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
91 static inline int speed_max(mddev_t
*mddev
)
93 return mddev
->sync_speed_max
?
94 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
97 static struct ctl_table_header
*raid_table_header
;
99 static ctl_table raid_table
[] = {
101 .procname
= "speed_limit_min",
102 .data
= &sysctl_speed_limit_min
,
103 .maxlen
= sizeof(int),
104 .mode
= S_IRUGO
|S_IWUSR
,
105 .proc_handler
= proc_dointvec
,
108 .procname
= "speed_limit_max",
109 .data
= &sysctl_speed_limit_max
,
110 .maxlen
= sizeof(int),
111 .mode
= S_IRUGO
|S_IWUSR
,
112 .proc_handler
= proc_dointvec
,
117 static ctl_table raid_dir_table
[] = {
121 .mode
= S_IRUGO
|S_IXUGO
,
127 static ctl_table raid_root_table
[] = {
132 .child
= raid_dir_table
,
137 static const struct block_device_operations md_fops
;
139 static int start_readonly
;
142 * We have a system wide 'event count' that is incremented
143 * on any 'interesting' event, and readers of /proc/mdstat
144 * can use 'poll' or 'select' to find out when the event
148 * start array, stop array, error, add device, remove device,
149 * start build, activate spare
151 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
152 static atomic_t md_event_count
;
153 void md_new_event(mddev_t
*mddev
)
155 atomic_inc(&md_event_count
);
156 wake_up(&md_event_waiters
);
158 EXPORT_SYMBOL_GPL(md_new_event
);
160 /* Alternate version that can be called from interrupts
161 * when calling sysfs_notify isn't needed.
163 static void md_new_event_inintr(mddev_t
*mddev
)
165 atomic_inc(&md_event_count
);
166 wake_up(&md_event_waiters
);
170 * Enables to iterate over all existing md arrays
171 * all_mddevs_lock protects this list.
173 static LIST_HEAD(all_mddevs
);
174 static DEFINE_SPINLOCK(all_mddevs_lock
);
178 * iterates through all used mddevs in the system.
179 * We take care to grab the all_mddevs_lock whenever navigating
180 * the list, and to always hold a refcount when unlocked.
181 * Any code which breaks out of this loop while own
182 * a reference to the current mddev and must mddev_put it.
184 #define for_each_mddev(mddev,tmp) \
186 for (({ spin_lock(&all_mddevs_lock); \
187 tmp = all_mddevs.next; \
189 ({ if (tmp != &all_mddevs) \
190 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
191 spin_unlock(&all_mddevs_lock); \
192 if (mddev) mddev_put(mddev); \
193 mddev = list_entry(tmp, mddev_t, all_mddevs); \
194 tmp != &all_mddevs;}); \
195 ({ spin_lock(&all_mddevs_lock); \
200 /* Rather than calling directly into the personality make_request function,
201 * IO requests come here first so that we can check if the device is
202 * being suspended pending a reconfiguration.
203 * We hold a refcount over the call to ->make_request. By the time that
204 * call has finished, the bio has been linked into some internal structure
205 * and so is visible to ->quiesce(), so we don't need the refcount any more.
207 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
209 mddev_t
*mddev
= q
->queuedata
;
211 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
216 if (mddev
->suspended
|| mddev
->barrier
) {
219 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
220 TASK_UNINTERRUPTIBLE
);
221 if (!mddev
->suspended
&& !mddev
->barrier
)
227 finish_wait(&mddev
->sb_wait
, &__wait
);
229 atomic_inc(&mddev
->active_io
);
231 rv
= mddev
->pers
->make_request(q
, bio
);
232 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
233 wake_up(&mddev
->sb_wait
);
238 static void mddev_suspend(mddev_t
*mddev
)
240 BUG_ON(mddev
->suspended
);
241 mddev
->suspended
= 1;
243 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
244 mddev
->pers
->quiesce(mddev
, 1);
245 md_unregister_thread(mddev
->thread
);
246 mddev
->thread
= NULL
;
247 /* we now know that no code is executing in the personality module,
248 * except possibly the tail end of a ->bi_end_io function, but that
249 * is certain to complete before the module has a chance to get
254 static void mddev_resume(mddev_t
*mddev
)
256 mddev
->suspended
= 0;
257 wake_up(&mddev
->sb_wait
);
258 mddev
->pers
->quiesce(mddev
, 0);
261 int mddev_congested(mddev_t
*mddev
, int bits
)
265 return mddev
->suspended
;
267 EXPORT_SYMBOL(mddev_congested
);
270 * Generic barrier handling for md
273 #define POST_REQUEST_BARRIER ((void*)1)
275 static void md_end_barrier(struct bio
*bio
, int err
)
277 mdk_rdev_t
*rdev
= bio
->bi_private
;
278 mddev_t
*mddev
= rdev
->mddev
;
279 if (err
== -EOPNOTSUPP
&& mddev
->barrier
!= POST_REQUEST_BARRIER
)
280 set_bit(BIO_EOPNOTSUPP
, &mddev
->barrier
->bi_flags
);
282 rdev_dec_pending(rdev
, mddev
);
284 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
285 if (mddev
->barrier
== POST_REQUEST_BARRIER
) {
286 /* This was a post-request barrier */
287 mddev
->barrier
= NULL
;
288 wake_up(&mddev
->sb_wait
);
290 /* The pre-request barrier has finished */
291 schedule_work(&mddev
->barrier_work
);
296 static void submit_barriers(mddev_t
*mddev
)
301 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
302 if (rdev
->raid_disk
>= 0 &&
303 !test_bit(Faulty
, &rdev
->flags
)) {
304 /* Take two references, one is dropped
305 * when request finishes, one after
306 * we reclaim rcu_read_lock
309 atomic_inc(&rdev
->nr_pending
);
310 atomic_inc(&rdev
->nr_pending
);
312 bi
= bio_alloc(GFP_KERNEL
, 0);
313 bi
->bi_end_io
= md_end_barrier
;
314 bi
->bi_private
= rdev
;
315 bi
->bi_bdev
= rdev
->bdev
;
316 atomic_inc(&mddev
->flush_pending
);
317 submit_bio(WRITE_BARRIER
, bi
);
319 rdev_dec_pending(rdev
, mddev
);
324 static void md_submit_barrier(struct work_struct
*ws
)
326 mddev_t
*mddev
= container_of(ws
, mddev_t
, barrier_work
);
327 struct bio
*bio
= mddev
->barrier
;
329 atomic_set(&mddev
->flush_pending
, 1);
331 if (test_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
))
332 bio_endio(bio
, -EOPNOTSUPP
);
333 else if (bio
->bi_size
== 0)
334 /* an empty barrier - all done */
337 bio
->bi_rw
&= ~(1<<BIO_RW_BARRIER
);
338 if (mddev
->pers
->make_request(mddev
->queue
, bio
))
339 generic_make_request(bio
);
340 mddev
->barrier
= POST_REQUEST_BARRIER
;
341 submit_barriers(mddev
);
343 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
344 mddev
->barrier
= NULL
;
345 wake_up(&mddev
->sb_wait
);
349 void md_barrier_request(mddev_t
*mddev
, struct bio
*bio
)
351 spin_lock_irq(&mddev
->write_lock
);
352 wait_event_lock_irq(mddev
->sb_wait
,
354 mddev
->write_lock
, /*nothing*/);
355 mddev
->barrier
= bio
;
356 spin_unlock_irq(&mddev
->write_lock
);
358 atomic_set(&mddev
->flush_pending
, 1);
359 INIT_WORK(&mddev
->barrier_work
, md_submit_barrier
);
361 submit_barriers(mddev
);
363 if (atomic_dec_and_test(&mddev
->flush_pending
))
364 schedule_work(&mddev
->barrier_work
);
366 EXPORT_SYMBOL(md_barrier_request
);
368 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
370 atomic_inc(&mddev
->active
);
374 static void mddev_delayed_delete(struct work_struct
*ws
);
376 static void mddev_put(mddev_t
*mddev
)
378 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
380 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
381 !mddev
->hold_active
) {
382 list_del(&mddev
->all_mddevs
);
383 if (mddev
->gendisk
) {
384 /* we did a probe so need to clean up.
385 * Call schedule_work inside the spinlock
386 * so that flush_scheduled_work() after
387 * mddev_find will succeed in waiting for the
390 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
391 schedule_work(&mddev
->del_work
);
395 spin_unlock(&all_mddevs_lock
);
398 static mddev_t
* mddev_find(dev_t unit
)
400 mddev_t
*mddev
, *new = NULL
;
403 spin_lock(&all_mddevs_lock
);
406 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
407 if (mddev
->unit
== unit
) {
409 spin_unlock(&all_mddevs_lock
);
415 list_add(&new->all_mddevs
, &all_mddevs
);
416 spin_unlock(&all_mddevs_lock
);
417 new->hold_active
= UNTIL_IOCTL
;
421 /* find an unused unit number */
422 static int next_minor
= 512;
423 int start
= next_minor
;
427 dev
= MKDEV(MD_MAJOR
, next_minor
);
429 if (next_minor
> MINORMASK
)
431 if (next_minor
== start
) {
432 /* Oh dear, all in use. */
433 spin_unlock(&all_mddevs_lock
);
439 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
440 if (mddev
->unit
== dev
) {
446 new->md_minor
= MINOR(dev
);
447 new->hold_active
= UNTIL_STOP
;
448 list_add(&new->all_mddevs
, &all_mddevs
);
449 spin_unlock(&all_mddevs_lock
);
452 spin_unlock(&all_mddevs_lock
);
454 new = kzalloc(sizeof(*new), GFP_KERNEL
);
459 if (MAJOR(unit
) == MD_MAJOR
)
460 new->md_minor
= MINOR(unit
);
462 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
464 mutex_init(&new->open_mutex
);
465 mutex_init(&new->reconfig_mutex
);
466 mutex_init(&new->bitmap_info
.mutex
);
467 INIT_LIST_HEAD(&new->disks
);
468 INIT_LIST_HEAD(&new->all_mddevs
);
469 init_timer(&new->safemode_timer
);
470 atomic_set(&new->active
, 1);
471 atomic_set(&new->openers
, 0);
472 atomic_set(&new->active_io
, 0);
473 spin_lock_init(&new->write_lock
);
474 atomic_set(&new->flush_pending
, 0);
475 init_waitqueue_head(&new->sb_wait
);
476 init_waitqueue_head(&new->recovery_wait
);
477 new->reshape_position
= MaxSector
;
479 new->resync_max
= MaxSector
;
480 new->level
= LEVEL_NONE
;
485 static inline int mddev_lock(mddev_t
* mddev
)
487 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
490 static inline int mddev_is_locked(mddev_t
*mddev
)
492 return mutex_is_locked(&mddev
->reconfig_mutex
);
495 static inline int mddev_trylock(mddev_t
* mddev
)
497 return mutex_trylock(&mddev
->reconfig_mutex
);
500 static inline void mddev_unlock(mddev_t
* mddev
)
502 mutex_unlock(&mddev
->reconfig_mutex
);
504 md_wakeup_thread(mddev
->thread
);
507 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
511 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
512 if (rdev
->desc_nr
== nr
)
518 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
522 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
523 if (rdev
->bdev
->bd_dev
== dev
)
529 static struct mdk_personality
*find_pers(int level
, char *clevel
)
531 struct mdk_personality
*pers
;
532 list_for_each_entry(pers
, &pers_list
, list
) {
533 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
535 if (strcmp(pers
->name
, clevel
)==0)
541 /* return the offset of the super block in 512byte sectors */
542 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
544 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
545 return MD_NEW_SIZE_SECTORS(num_sectors
);
548 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
553 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
554 if (!rdev
->sb_page
) {
555 printk(KERN_ALERT
"md: out of memory.\n");
562 static void free_disk_sb(mdk_rdev_t
* rdev
)
565 put_page(rdev
->sb_page
);
567 rdev
->sb_page
= NULL
;
574 static void super_written(struct bio
*bio
, int error
)
576 mdk_rdev_t
*rdev
= bio
->bi_private
;
577 mddev_t
*mddev
= rdev
->mddev
;
579 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
580 printk("md: super_written gets error=%d, uptodate=%d\n",
581 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
582 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
583 md_error(mddev
, rdev
);
586 if (atomic_dec_and_test(&mddev
->pending_writes
))
587 wake_up(&mddev
->sb_wait
);
591 static void super_written_barrier(struct bio
*bio
, int error
)
593 struct bio
*bio2
= bio
->bi_private
;
594 mdk_rdev_t
*rdev
= bio2
->bi_private
;
595 mddev_t
*mddev
= rdev
->mddev
;
597 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
598 error
== -EOPNOTSUPP
) {
600 /* barriers don't appear to be supported :-( */
601 set_bit(BarriersNotsupp
, &rdev
->flags
);
602 mddev
->barriers_work
= 0;
603 spin_lock_irqsave(&mddev
->write_lock
, flags
);
604 bio2
->bi_next
= mddev
->biolist
;
605 mddev
->biolist
= bio2
;
606 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
607 wake_up(&mddev
->sb_wait
);
611 bio
->bi_private
= rdev
;
612 super_written(bio
, error
);
616 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
617 sector_t sector
, int size
, struct page
*page
)
619 /* write first size bytes of page to sector of rdev
620 * Increment mddev->pending_writes before returning
621 * and decrement it on completion, waking up sb_wait
622 * if zero is reached.
623 * If an error occurred, call md_error
625 * As we might need to resubmit the request if BIO_RW_BARRIER
626 * causes ENOTSUPP, we allocate a spare bio...
628 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
629 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
631 bio
->bi_bdev
= rdev
->bdev
;
632 bio
->bi_sector
= sector
;
633 bio_add_page(bio
, page
, size
, 0);
634 bio
->bi_private
= rdev
;
635 bio
->bi_end_io
= super_written
;
638 atomic_inc(&mddev
->pending_writes
);
639 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
641 rw
|= (1<<BIO_RW_BARRIER
);
642 rbio
= bio_clone(bio
, GFP_NOIO
);
643 rbio
->bi_private
= bio
;
644 rbio
->bi_end_io
= super_written_barrier
;
645 submit_bio(rw
, rbio
);
650 void md_super_wait(mddev_t
*mddev
)
652 /* wait for all superblock writes that were scheduled to complete.
653 * if any had to be retried (due to BARRIER problems), retry them
657 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
658 if (atomic_read(&mddev
->pending_writes
)==0)
660 while (mddev
->biolist
) {
662 spin_lock_irq(&mddev
->write_lock
);
663 bio
= mddev
->biolist
;
664 mddev
->biolist
= bio
->bi_next
;
666 spin_unlock_irq(&mddev
->write_lock
);
667 submit_bio(bio
->bi_rw
, bio
);
671 finish_wait(&mddev
->sb_wait
, &wq
);
674 static void bi_complete(struct bio
*bio
, int error
)
676 complete((struct completion
*)bio
->bi_private
);
679 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
680 struct page
*page
, int rw
)
682 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
683 struct completion event
;
686 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
689 bio
->bi_sector
= sector
;
690 bio_add_page(bio
, page
, size
, 0);
691 init_completion(&event
);
692 bio
->bi_private
= &event
;
693 bio
->bi_end_io
= bi_complete
;
695 wait_for_completion(&event
);
697 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
701 EXPORT_SYMBOL_GPL(sync_page_io
);
703 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
705 char b
[BDEVNAME_SIZE
];
706 if (!rdev
->sb_page
) {
714 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
720 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
721 bdevname(rdev
->bdev
,b
));
725 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
727 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
728 sb1
->set_uuid1
== sb2
->set_uuid1
&&
729 sb1
->set_uuid2
== sb2
->set_uuid2
&&
730 sb1
->set_uuid3
== sb2
->set_uuid3
;
733 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
736 mdp_super_t
*tmp1
, *tmp2
;
738 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
739 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
741 if (!tmp1
|| !tmp2
) {
743 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
751 * nr_disks is not constant
756 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
764 static u32
md_csum_fold(u32 csum
)
766 csum
= (csum
& 0xffff) + (csum
>> 16);
767 return (csum
& 0xffff) + (csum
>> 16);
770 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
773 u32
*sb32
= (u32
*)sb
;
775 unsigned int disk_csum
, csum
;
777 disk_csum
= sb
->sb_csum
;
780 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
782 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
786 /* This used to use csum_partial, which was wrong for several
787 * reasons including that different results are returned on
788 * different architectures. It isn't critical that we get exactly
789 * the same return value as before (we always csum_fold before
790 * testing, and that removes any differences). However as we
791 * know that csum_partial always returned a 16bit value on
792 * alphas, do a fold to maximise conformity to previous behaviour.
794 sb
->sb_csum
= md_csum_fold(disk_csum
);
796 sb
->sb_csum
= disk_csum
;
803 * Handle superblock details.
804 * We want to be able to handle multiple superblock formats
805 * so we have a common interface to them all, and an array of
806 * different handlers.
807 * We rely on user-space to write the initial superblock, and support
808 * reading and updating of superblocks.
809 * Interface methods are:
810 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
811 * loads and validates a superblock on dev.
812 * if refdev != NULL, compare superblocks on both devices
814 * 0 - dev has a superblock that is compatible with refdev
815 * 1 - dev has a superblock that is compatible and newer than refdev
816 * so dev should be used as the refdev in future
817 * -EINVAL superblock incompatible or invalid
818 * -othererror e.g. -EIO
820 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
821 * Verify that dev is acceptable into mddev.
822 * The first time, mddev->raid_disks will be 0, and data from
823 * dev should be merged in. Subsequent calls check that dev
824 * is new enough. Return 0 or -EINVAL
826 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
827 * Update the superblock for rdev with data in mddev
828 * This does not write to disc.
834 struct module
*owner
;
835 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
837 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
838 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
839 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
840 sector_t num_sectors
);
844 * Check that the given mddev has no bitmap.
846 * This function is called from the run method of all personalities that do not
847 * support bitmaps. It prints an error message and returns non-zero if mddev
848 * has a bitmap. Otherwise, it returns 0.
851 int md_check_no_bitmap(mddev_t
*mddev
)
853 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
855 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
856 mdname(mddev
), mddev
->pers
->name
);
859 EXPORT_SYMBOL(md_check_no_bitmap
);
862 * load_super for 0.90.0
864 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
866 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
871 * Calculate the position of the superblock (512byte sectors),
872 * it's at the end of the disk.
874 * It also happens to be a multiple of 4Kb.
876 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
878 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
883 bdevname(rdev
->bdev
, b
);
884 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
886 if (sb
->md_magic
!= MD_SB_MAGIC
) {
887 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
892 if (sb
->major_version
!= 0 ||
893 sb
->minor_version
< 90 ||
894 sb
->minor_version
> 91) {
895 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
896 sb
->major_version
, sb
->minor_version
,
901 if (sb
->raid_disks
<= 0)
904 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
905 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
910 rdev
->preferred_minor
= sb
->md_minor
;
911 rdev
->data_offset
= 0;
912 rdev
->sb_size
= MD_SB_BYTES
;
914 if (sb
->level
== LEVEL_MULTIPATH
)
917 rdev
->desc_nr
= sb
->this_disk
.number
;
923 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
924 if (!uuid_equal(refsb
, sb
)) {
925 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
926 b
, bdevname(refdev
->bdev
,b2
));
929 if (!sb_equal(refsb
, sb
)) {
930 printk(KERN_WARNING
"md: %s has same UUID"
931 " but different superblock to %s\n",
932 b
, bdevname(refdev
->bdev
, b2
));
936 ev2
= md_event(refsb
);
942 rdev
->sectors
= rdev
->sb_start
;
944 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
945 /* "this cannot possibly happen" ... */
953 * validate_super for 0.90.0
955 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
958 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
959 __u64 ev1
= md_event(sb
);
961 rdev
->raid_disk
= -1;
962 clear_bit(Faulty
, &rdev
->flags
);
963 clear_bit(In_sync
, &rdev
->flags
);
964 clear_bit(WriteMostly
, &rdev
->flags
);
965 clear_bit(BarriersNotsupp
, &rdev
->flags
);
967 if (mddev
->raid_disks
== 0) {
968 mddev
->major_version
= 0;
969 mddev
->minor_version
= sb
->minor_version
;
970 mddev
->patch_version
= sb
->patch_version
;
972 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
973 mddev
->ctime
= sb
->ctime
;
974 mddev
->utime
= sb
->utime
;
975 mddev
->level
= sb
->level
;
976 mddev
->clevel
[0] = 0;
977 mddev
->layout
= sb
->layout
;
978 mddev
->raid_disks
= sb
->raid_disks
;
979 mddev
->dev_sectors
= sb
->size
* 2;
981 mddev
->bitmap_info
.offset
= 0;
982 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
984 if (mddev
->minor_version
>= 91) {
985 mddev
->reshape_position
= sb
->reshape_position
;
986 mddev
->delta_disks
= sb
->delta_disks
;
987 mddev
->new_level
= sb
->new_level
;
988 mddev
->new_layout
= sb
->new_layout
;
989 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
991 mddev
->reshape_position
= MaxSector
;
992 mddev
->delta_disks
= 0;
993 mddev
->new_level
= mddev
->level
;
994 mddev
->new_layout
= mddev
->layout
;
995 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
998 if (sb
->state
& (1<<MD_SB_CLEAN
))
999 mddev
->recovery_cp
= MaxSector
;
1001 if (sb
->events_hi
== sb
->cp_events_hi
&&
1002 sb
->events_lo
== sb
->cp_events_lo
) {
1003 mddev
->recovery_cp
= sb
->recovery_cp
;
1005 mddev
->recovery_cp
= 0;
1008 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1009 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1010 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1011 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1013 mddev
->max_disks
= MD_SB_DISKS
;
1015 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1016 mddev
->bitmap_info
.file
== NULL
)
1017 mddev
->bitmap_info
.offset
=
1018 mddev
->bitmap_info
.default_offset
;
1020 } else if (mddev
->pers
== NULL
) {
1021 /* Insist on good event counter while assembling */
1023 if (ev1
< mddev
->events
)
1025 } else if (mddev
->bitmap
) {
1026 /* if adding to array with a bitmap, then we can accept an
1027 * older device ... but not too old.
1029 if (ev1
< mddev
->bitmap
->events_cleared
)
1032 if (ev1
< mddev
->events
)
1033 /* just a hot-add of a new device, leave raid_disk at -1 */
1037 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1038 desc
= sb
->disks
+ rdev
->desc_nr
;
1040 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1041 set_bit(Faulty
, &rdev
->flags
);
1042 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1043 desc->raid_disk < mddev->raid_disks */) {
1044 set_bit(In_sync
, &rdev
->flags
);
1045 rdev
->raid_disk
= desc
->raid_disk
;
1046 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1047 /* active but not in sync implies recovery up to
1048 * reshape position. We don't know exactly where
1049 * that is, so set to zero for now */
1050 if (mddev
->minor_version
>= 91) {
1051 rdev
->recovery_offset
= 0;
1052 rdev
->raid_disk
= desc
->raid_disk
;
1055 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1056 set_bit(WriteMostly
, &rdev
->flags
);
1057 } else /* MULTIPATH are always insync */
1058 set_bit(In_sync
, &rdev
->flags
);
1063 * sync_super for 0.90.0
1065 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1069 int next_spare
= mddev
->raid_disks
;
1072 /* make rdev->sb match mddev data..
1075 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1076 * 3/ any empty disks < next_spare become removed
1078 * disks[0] gets initialised to REMOVED because
1079 * we cannot be sure from other fields if it has
1080 * been initialised or not.
1083 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1085 rdev
->sb_size
= MD_SB_BYTES
;
1087 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1089 memset(sb
, 0, sizeof(*sb
));
1091 sb
->md_magic
= MD_SB_MAGIC
;
1092 sb
->major_version
= mddev
->major_version
;
1093 sb
->patch_version
= mddev
->patch_version
;
1094 sb
->gvalid_words
= 0; /* ignored */
1095 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1096 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1097 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1098 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1100 sb
->ctime
= mddev
->ctime
;
1101 sb
->level
= mddev
->level
;
1102 sb
->size
= mddev
->dev_sectors
/ 2;
1103 sb
->raid_disks
= mddev
->raid_disks
;
1104 sb
->md_minor
= mddev
->md_minor
;
1105 sb
->not_persistent
= 0;
1106 sb
->utime
= mddev
->utime
;
1108 sb
->events_hi
= (mddev
->events
>>32);
1109 sb
->events_lo
= (u32
)mddev
->events
;
1111 if (mddev
->reshape_position
== MaxSector
)
1112 sb
->minor_version
= 90;
1114 sb
->minor_version
= 91;
1115 sb
->reshape_position
= mddev
->reshape_position
;
1116 sb
->new_level
= mddev
->new_level
;
1117 sb
->delta_disks
= mddev
->delta_disks
;
1118 sb
->new_layout
= mddev
->new_layout
;
1119 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1121 mddev
->minor_version
= sb
->minor_version
;
1124 sb
->recovery_cp
= mddev
->recovery_cp
;
1125 sb
->cp_events_hi
= (mddev
->events
>>32);
1126 sb
->cp_events_lo
= (u32
)mddev
->events
;
1127 if (mddev
->recovery_cp
== MaxSector
)
1128 sb
->state
= (1<< MD_SB_CLEAN
);
1130 sb
->recovery_cp
= 0;
1132 sb
->layout
= mddev
->layout
;
1133 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1135 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1136 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1138 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1139 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1142 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1144 if (rdev2
->raid_disk
>= 0 &&
1145 sb
->minor_version
>= 91)
1146 /* we have nowhere to store the recovery_offset,
1147 * but if it is not below the reshape_position,
1148 * we can piggy-back on that.
1151 if (rdev2
->raid_disk
< 0 ||
1152 test_bit(Faulty
, &rdev2
->flags
))
1155 desc_nr
= rdev2
->raid_disk
;
1157 desc_nr
= next_spare
++;
1158 rdev2
->desc_nr
= desc_nr
;
1159 d
= &sb
->disks
[rdev2
->desc_nr
];
1161 d
->number
= rdev2
->desc_nr
;
1162 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1163 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1165 d
->raid_disk
= rdev2
->raid_disk
;
1167 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1168 if (test_bit(Faulty
, &rdev2
->flags
))
1169 d
->state
= (1<<MD_DISK_FAULTY
);
1170 else if (is_active
) {
1171 d
->state
= (1<<MD_DISK_ACTIVE
);
1172 if (test_bit(In_sync
, &rdev2
->flags
))
1173 d
->state
|= (1<<MD_DISK_SYNC
);
1181 if (test_bit(WriteMostly
, &rdev2
->flags
))
1182 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1184 /* now set the "removed" and "faulty" bits on any missing devices */
1185 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1186 mdp_disk_t
*d
= &sb
->disks
[i
];
1187 if (d
->state
== 0 && d
->number
== 0) {
1190 d
->state
= (1<<MD_DISK_REMOVED
);
1191 d
->state
|= (1<<MD_DISK_FAULTY
);
1195 sb
->nr_disks
= nr_disks
;
1196 sb
->active_disks
= active
;
1197 sb
->working_disks
= working
;
1198 sb
->failed_disks
= failed
;
1199 sb
->spare_disks
= spare
;
1201 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1202 sb
->sb_csum
= calc_sb_csum(sb
);
1206 * rdev_size_change for 0.90.0
1208 static unsigned long long
1209 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1211 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1212 return 0; /* component must fit device */
1213 if (rdev
->mddev
->bitmap_info
.offset
)
1214 return 0; /* can't move bitmap */
1215 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1216 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1217 num_sectors
= rdev
->sb_start
;
1218 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1220 md_super_wait(rdev
->mddev
);
1221 return num_sectors
/ 2; /* kB for sysfs */
1226 * version 1 superblock
1229 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1233 unsigned long long newcsum
;
1234 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1235 __le32
*isuper
= (__le32
*)sb
;
1238 disk_csum
= sb
->sb_csum
;
1241 for (i
=0; size
>=4; size
-= 4 )
1242 newcsum
+= le32_to_cpu(*isuper
++);
1245 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1247 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1248 sb
->sb_csum
= disk_csum
;
1249 return cpu_to_le32(csum
);
1252 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1254 struct mdp_superblock_1
*sb
;
1257 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1261 * Calculate the position of the superblock in 512byte sectors.
1262 * It is always aligned to a 4K boundary and
1263 * depeding on minor_version, it can be:
1264 * 0: At least 8K, but less than 12K, from end of device
1265 * 1: At start of device
1266 * 2: 4K from start of device.
1268 switch(minor_version
) {
1270 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1272 sb_start
&= ~(sector_t
)(4*2-1);
1283 rdev
->sb_start
= sb_start
;
1285 /* superblock is rarely larger than 1K, but it can be larger,
1286 * and it is safe to read 4k, so we do that
1288 ret
= read_disk_sb(rdev
, 4096);
1289 if (ret
) return ret
;
1292 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1294 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1295 sb
->major_version
!= cpu_to_le32(1) ||
1296 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1297 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1298 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1301 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1302 printk("md: invalid superblock checksum on %s\n",
1303 bdevname(rdev
->bdev
,b
));
1306 if (le64_to_cpu(sb
->data_size
) < 10) {
1307 printk("md: data_size too small on %s\n",
1308 bdevname(rdev
->bdev
,b
));
1312 rdev
->preferred_minor
= 0xffff;
1313 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1314 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1316 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1317 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1318 if (rdev
->sb_size
& bmask
)
1319 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1322 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1325 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1328 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1334 struct mdp_superblock_1
*refsb
=
1335 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1337 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1338 sb
->level
!= refsb
->level
||
1339 sb
->layout
!= refsb
->layout
||
1340 sb
->chunksize
!= refsb
->chunksize
) {
1341 printk(KERN_WARNING
"md: %s has strangely different"
1342 " superblock to %s\n",
1343 bdevname(rdev
->bdev
,b
),
1344 bdevname(refdev
->bdev
,b2
));
1347 ev1
= le64_to_cpu(sb
->events
);
1348 ev2
= le64_to_cpu(refsb
->events
);
1356 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1357 le64_to_cpu(sb
->data_offset
);
1359 rdev
->sectors
= rdev
->sb_start
;
1360 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1362 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1363 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1368 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1370 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1371 __u64 ev1
= le64_to_cpu(sb
->events
);
1373 rdev
->raid_disk
= -1;
1374 clear_bit(Faulty
, &rdev
->flags
);
1375 clear_bit(In_sync
, &rdev
->flags
);
1376 clear_bit(WriteMostly
, &rdev
->flags
);
1377 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1379 if (mddev
->raid_disks
== 0) {
1380 mddev
->major_version
= 1;
1381 mddev
->patch_version
= 0;
1382 mddev
->external
= 0;
1383 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1384 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1385 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1386 mddev
->level
= le32_to_cpu(sb
->level
);
1387 mddev
->clevel
[0] = 0;
1388 mddev
->layout
= le32_to_cpu(sb
->layout
);
1389 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1390 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1391 mddev
->events
= ev1
;
1392 mddev
->bitmap_info
.offset
= 0;
1393 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1395 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1396 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1398 mddev
->max_disks
= (4096-256)/2;
1400 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1401 mddev
->bitmap_info
.file
== NULL
)
1402 mddev
->bitmap_info
.offset
=
1403 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1405 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1406 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1407 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1408 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1409 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1410 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1412 mddev
->reshape_position
= MaxSector
;
1413 mddev
->delta_disks
= 0;
1414 mddev
->new_level
= mddev
->level
;
1415 mddev
->new_layout
= mddev
->layout
;
1416 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1419 } else if (mddev
->pers
== NULL
) {
1420 /* Insist of good event counter while assembling */
1422 if (ev1
< mddev
->events
)
1424 } else if (mddev
->bitmap
) {
1425 /* If adding to array with a bitmap, then we can accept an
1426 * older device, but not too old.
1428 if (ev1
< mddev
->bitmap
->events_cleared
)
1431 if (ev1
< mddev
->events
)
1432 /* just a hot-add of a new device, leave raid_disk at -1 */
1435 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1437 if (rdev
->desc_nr
< 0 ||
1438 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1442 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1444 case 0xffff: /* spare */
1446 case 0xfffe: /* faulty */
1447 set_bit(Faulty
, &rdev
->flags
);
1450 if ((le32_to_cpu(sb
->feature_map
) &
1451 MD_FEATURE_RECOVERY_OFFSET
))
1452 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1454 set_bit(In_sync
, &rdev
->flags
);
1455 rdev
->raid_disk
= role
;
1458 if (sb
->devflags
& WriteMostly1
)
1459 set_bit(WriteMostly
, &rdev
->flags
);
1460 } else /* MULTIPATH are always insync */
1461 set_bit(In_sync
, &rdev
->flags
);
1466 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1468 struct mdp_superblock_1
*sb
;
1471 /* make rdev->sb match mddev and rdev data. */
1473 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1475 sb
->feature_map
= 0;
1477 sb
->recovery_offset
= cpu_to_le64(0);
1478 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1479 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1480 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1482 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1483 sb
->events
= cpu_to_le64(mddev
->events
);
1485 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1487 sb
->resync_offset
= cpu_to_le64(0);
1489 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1491 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1492 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1493 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1494 sb
->level
= cpu_to_le32(mddev
->level
);
1495 sb
->layout
= cpu_to_le32(mddev
->layout
);
1497 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1498 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1499 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1502 if (rdev
->raid_disk
>= 0 &&
1503 !test_bit(In_sync
, &rdev
->flags
)) {
1504 if (rdev
->recovery_offset
> 0) {
1506 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1507 sb
->recovery_offset
=
1508 cpu_to_le64(rdev
->recovery_offset
);
1512 if (mddev
->reshape_position
!= MaxSector
) {
1513 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1514 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1515 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1516 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1517 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1518 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1522 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1523 if (rdev2
->desc_nr
+1 > max_dev
)
1524 max_dev
= rdev2
->desc_nr
+1;
1526 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1528 sb
->max_dev
= cpu_to_le32(max_dev
);
1529 rdev
->sb_size
= max_dev
* 2 + 256;
1530 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1531 if (rdev
->sb_size
& bmask
)
1532 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1534 for (i
=0; i
<max_dev
;i
++)
1535 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1537 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1539 if (test_bit(Faulty
, &rdev2
->flags
))
1540 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1541 else if (test_bit(In_sync
, &rdev2
->flags
))
1542 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1543 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1544 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1546 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1549 sb
->sb_csum
= calc_sb_1_csum(sb
);
1552 static unsigned long long
1553 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1555 struct mdp_superblock_1
*sb
;
1556 sector_t max_sectors
;
1557 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1558 return 0; /* component must fit device */
1559 if (rdev
->sb_start
< rdev
->data_offset
) {
1560 /* minor versions 1 and 2; superblock before data */
1561 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1562 max_sectors
-= rdev
->data_offset
;
1563 if (!num_sectors
|| num_sectors
> max_sectors
)
1564 num_sectors
= max_sectors
;
1565 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1566 /* minor version 0 with bitmap we can't move */
1569 /* minor version 0; superblock after data */
1571 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1572 sb_start
&= ~(sector_t
)(4*2 - 1);
1573 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1574 if (!num_sectors
|| num_sectors
> max_sectors
)
1575 num_sectors
= max_sectors
;
1576 rdev
->sb_start
= sb_start
;
1578 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1579 sb
->data_size
= cpu_to_le64(num_sectors
);
1580 sb
->super_offset
= rdev
->sb_start
;
1581 sb
->sb_csum
= calc_sb_1_csum(sb
);
1582 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1584 md_super_wait(rdev
->mddev
);
1585 return num_sectors
/ 2; /* kB for sysfs */
1588 static struct super_type super_types
[] = {
1591 .owner
= THIS_MODULE
,
1592 .load_super
= super_90_load
,
1593 .validate_super
= super_90_validate
,
1594 .sync_super
= super_90_sync
,
1595 .rdev_size_change
= super_90_rdev_size_change
,
1599 .owner
= THIS_MODULE
,
1600 .load_super
= super_1_load
,
1601 .validate_super
= super_1_validate
,
1602 .sync_super
= super_1_sync
,
1603 .rdev_size_change
= super_1_rdev_size_change
,
1607 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1609 mdk_rdev_t
*rdev
, *rdev2
;
1612 rdev_for_each_rcu(rdev
, mddev1
)
1613 rdev_for_each_rcu(rdev2
, mddev2
)
1614 if (rdev
->bdev
->bd_contains
==
1615 rdev2
->bdev
->bd_contains
) {
1623 static LIST_HEAD(pending_raid_disks
);
1626 * Try to register data integrity profile for an mddev
1628 * This is called when an array is started and after a disk has been kicked
1629 * from the array. It only succeeds if all working and active component devices
1630 * are integrity capable with matching profiles.
1632 int md_integrity_register(mddev_t
*mddev
)
1634 mdk_rdev_t
*rdev
, *reference
= NULL
;
1636 if (list_empty(&mddev
->disks
))
1637 return 0; /* nothing to do */
1638 if (blk_get_integrity(mddev
->gendisk
))
1639 return 0; /* already registered */
1640 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1641 /* skip spares and non-functional disks */
1642 if (test_bit(Faulty
, &rdev
->flags
))
1644 if (rdev
->raid_disk
< 0)
1647 * If at least one rdev is not integrity capable, we can not
1648 * enable data integrity for the md device.
1650 if (!bdev_get_integrity(rdev
->bdev
))
1653 /* Use the first rdev as the reference */
1657 /* does this rdev's profile match the reference profile? */
1658 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1659 rdev
->bdev
->bd_disk
) < 0)
1663 * All component devices are integrity capable and have matching
1664 * profiles, register the common profile for the md device.
1666 if (blk_integrity_register(mddev
->gendisk
,
1667 bdev_get_integrity(reference
->bdev
)) != 0) {
1668 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1672 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1676 EXPORT_SYMBOL(md_integrity_register
);
1678 /* Disable data integrity if non-capable/non-matching disk is being added */
1679 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1681 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1682 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1684 if (!bi_mddev
) /* nothing to do */
1686 if (rdev
->raid_disk
< 0) /* skip spares */
1688 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1689 rdev
->bdev
->bd_disk
) >= 0)
1691 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1692 blk_integrity_unregister(mddev
->gendisk
);
1694 EXPORT_SYMBOL(md_integrity_add_rdev
);
1696 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1698 char b
[BDEVNAME_SIZE
];
1708 /* prevent duplicates */
1709 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1712 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1713 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1714 rdev
->sectors
< mddev
->dev_sectors
)) {
1716 /* Cannot change size, so fail
1717 * If mddev->level <= 0, then we don't care
1718 * about aligning sizes (e.g. linear)
1720 if (mddev
->level
> 0)
1723 mddev
->dev_sectors
= rdev
->sectors
;
1726 /* Verify rdev->desc_nr is unique.
1727 * If it is -1, assign a free number, else
1728 * check number is not in use
1730 if (rdev
->desc_nr
< 0) {
1732 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1733 while (find_rdev_nr(mddev
, choice
))
1735 rdev
->desc_nr
= choice
;
1737 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1740 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1741 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1742 mdname(mddev
), mddev
->max_disks
);
1745 bdevname(rdev
->bdev
,b
);
1746 while ( (s
=strchr(b
, '/')) != NULL
)
1749 rdev
->mddev
= mddev
;
1750 printk(KERN_INFO
"md: bind<%s>\n", b
);
1752 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1755 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1756 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1757 kobject_del(&rdev
->kobj
);
1760 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1762 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1763 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1765 /* May as well allow recovery to be retried once */
1766 mddev
->recovery_disabled
= 0;
1771 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1776 static void md_delayed_delete(struct work_struct
*ws
)
1778 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1779 kobject_del(&rdev
->kobj
);
1780 kobject_put(&rdev
->kobj
);
1783 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1785 char b
[BDEVNAME_SIZE
];
1790 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1791 list_del_rcu(&rdev
->same_set
);
1792 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1794 sysfs_remove_link(&rdev
->kobj
, "block");
1795 sysfs_put(rdev
->sysfs_state
);
1796 rdev
->sysfs_state
= NULL
;
1797 /* We need to delay this, otherwise we can deadlock when
1798 * writing to 'remove' to "dev/state". We also need
1799 * to delay it due to rcu usage.
1802 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1803 kobject_get(&rdev
->kobj
);
1804 schedule_work(&rdev
->del_work
);
1808 * prevent the device from being mounted, repartitioned or
1809 * otherwise reused by a RAID array (or any other kernel
1810 * subsystem), by bd_claiming the device.
1812 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1815 struct block_device
*bdev
;
1816 char b
[BDEVNAME_SIZE
];
1818 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1820 printk(KERN_ERR
"md: could not open %s.\n",
1821 __bdevname(dev
, b
));
1822 return PTR_ERR(bdev
);
1824 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1826 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1828 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1832 set_bit(AllReserved
, &rdev
->flags
);
1837 static void unlock_rdev(mdk_rdev_t
*rdev
)
1839 struct block_device
*bdev
= rdev
->bdev
;
1844 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1847 void md_autodetect_dev(dev_t dev
);
1849 static void export_rdev(mdk_rdev_t
* rdev
)
1851 char b
[BDEVNAME_SIZE
];
1852 printk(KERN_INFO
"md: export_rdev(%s)\n",
1853 bdevname(rdev
->bdev
,b
));
1858 if (test_bit(AutoDetected
, &rdev
->flags
))
1859 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1862 kobject_put(&rdev
->kobj
);
1865 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1867 unbind_rdev_from_array(rdev
);
1871 static void export_array(mddev_t
*mddev
)
1873 mdk_rdev_t
*rdev
, *tmp
;
1875 rdev_for_each(rdev
, tmp
, mddev
) {
1880 kick_rdev_from_array(rdev
);
1882 if (!list_empty(&mddev
->disks
))
1884 mddev
->raid_disks
= 0;
1885 mddev
->major_version
= 0;
1888 static void print_desc(mdp_disk_t
*desc
)
1890 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1891 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1894 static void print_sb_90(mdp_super_t
*sb
)
1899 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1900 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1901 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1903 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1904 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1905 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1906 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1907 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1908 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1909 sb
->failed_disks
, sb
->spare_disks
,
1910 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1913 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1916 desc
= sb
->disks
+ i
;
1917 if (desc
->number
|| desc
->major
|| desc
->minor
||
1918 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1919 printk(" D %2d: ", i
);
1923 printk(KERN_INFO
"md: THIS: ");
1924 print_desc(&sb
->this_disk
);
1927 static void print_sb_1(struct mdp_superblock_1
*sb
)
1931 uuid
= sb
->set_uuid
;
1933 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1934 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1935 "md: Name: \"%s\" CT:%llu\n",
1936 le32_to_cpu(sb
->major_version
),
1937 le32_to_cpu(sb
->feature_map
),
1938 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1939 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1940 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1941 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1943 (unsigned long long)le64_to_cpu(sb
->ctime
)
1944 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1946 uuid
= sb
->device_uuid
;
1948 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1950 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1951 ":%02x%02x%02x%02x%02x%02x\n"
1952 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1953 "md: (MaxDev:%u) \n",
1954 le32_to_cpu(sb
->level
),
1955 (unsigned long long)le64_to_cpu(sb
->size
),
1956 le32_to_cpu(sb
->raid_disks
),
1957 le32_to_cpu(sb
->layout
),
1958 le32_to_cpu(sb
->chunksize
),
1959 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1960 (unsigned long long)le64_to_cpu(sb
->data_size
),
1961 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1962 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1963 le32_to_cpu(sb
->dev_number
),
1964 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1965 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1966 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1967 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1969 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1970 (unsigned long long)le64_to_cpu(sb
->events
),
1971 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
1972 le32_to_cpu(sb
->sb_csum
),
1973 le32_to_cpu(sb
->max_dev
)
1977 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
1979 char b
[BDEVNAME_SIZE
];
1980 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1981 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
1982 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1984 if (rdev
->sb_loaded
) {
1985 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
1986 switch (major_version
) {
1988 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
1991 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
1995 printk(KERN_INFO
"md: no rdev superblock!\n");
1998 static void md_print_devices(void)
2000 struct list_head
*tmp
;
2003 char b
[BDEVNAME_SIZE
];
2006 printk("md: **********************************\n");
2007 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2008 printk("md: **********************************\n");
2009 for_each_mddev(mddev
, tmp
) {
2012 bitmap_print_sb(mddev
->bitmap
);
2014 printk("%s: ", mdname(mddev
));
2015 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2016 printk("<%s>", bdevname(rdev
->bdev
,b
));
2019 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2020 print_rdev(rdev
, mddev
->major_version
);
2022 printk("md: **********************************\n");
2027 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2029 /* Update each superblock (in-memory image), but
2030 * if we are allowed to, skip spares which already
2031 * have the right event counter, or have one earlier
2032 * (which would mean they aren't being marked as dirty
2033 * with the rest of the array)
2037 /* First make sure individual recovery_offsets are correct */
2038 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2039 if (rdev
->raid_disk
>= 0 &&
2040 !test_bit(In_sync
, &rdev
->flags
) &&
2041 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2042 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2045 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2046 if (rdev
->sb_events
== mddev
->events
||
2048 rdev
->raid_disk
< 0 &&
2049 (rdev
->sb_events
&1)==0 &&
2050 rdev
->sb_events
+1 == mddev
->events
)) {
2051 /* Don't update this superblock */
2052 rdev
->sb_loaded
= 2;
2054 super_types
[mddev
->major_version
].
2055 sync_super(mddev
, rdev
);
2056 rdev
->sb_loaded
= 1;
2061 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2067 mddev
->utime
= get_seconds();
2068 if (mddev
->external
)
2071 spin_lock_irq(&mddev
->write_lock
);
2073 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2074 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2076 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2077 /* just a clean<-> dirty transition, possibly leave spares alone,
2078 * though if events isn't the right even/odd, we will have to do
2084 if (mddev
->degraded
)
2085 /* If the array is degraded, then skipping spares is both
2086 * dangerous and fairly pointless.
2087 * Dangerous because a device that was removed from the array
2088 * might have a event_count that still looks up-to-date,
2089 * so it can be re-added without a resync.
2090 * Pointless because if there are any spares to skip,
2091 * then a recovery will happen and soon that array won't
2092 * be degraded any more and the spare can go back to sleep then.
2096 sync_req
= mddev
->in_sync
;
2098 /* If this is just a dirty<->clean transition, and the array is clean
2099 * and 'events' is odd, we can roll back to the previous clean state */
2101 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2102 && (mddev
->events
& 1)
2103 && mddev
->events
!= 1)
2106 /* otherwise we have to go forward and ... */
2108 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
2109 /* .. if the array isn't clean, an 'even' event must also go
2111 if ((mddev
->events
&1)==0)
2114 /* otherwise an 'odd' event must go to spares */
2115 if ((mddev
->events
&1))
2120 if (!mddev
->events
) {
2122 * oops, this 64-bit counter should never wrap.
2123 * Either we are in around ~1 trillion A.C., assuming
2124 * 1 reboot per second, or we have a bug:
2131 * do not write anything to disk if using
2132 * nonpersistent superblocks
2134 if (!mddev
->persistent
) {
2135 if (!mddev
->external
)
2136 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2138 spin_unlock_irq(&mddev
->write_lock
);
2139 wake_up(&mddev
->sb_wait
);
2142 sync_sbs(mddev
, nospares
);
2143 spin_unlock_irq(&mddev
->write_lock
);
2146 "md: updating %s RAID superblock on device (in sync %d)\n",
2147 mdname(mddev
),mddev
->in_sync
);
2149 bitmap_update_sb(mddev
->bitmap
);
2150 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2151 char b
[BDEVNAME_SIZE
];
2152 dprintk(KERN_INFO
"md: ");
2153 if (rdev
->sb_loaded
!= 1)
2154 continue; /* no noise on spare devices */
2155 if (test_bit(Faulty
, &rdev
->flags
))
2156 dprintk("(skipping faulty ");
2158 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2159 if (!test_bit(Faulty
, &rdev
->flags
)) {
2160 md_super_write(mddev
,rdev
,
2161 rdev
->sb_start
, rdev
->sb_size
,
2163 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2164 bdevname(rdev
->bdev
,b
),
2165 (unsigned long long)rdev
->sb_start
);
2166 rdev
->sb_events
= mddev
->events
;
2170 if (mddev
->level
== LEVEL_MULTIPATH
)
2171 /* only need to write one superblock... */
2174 md_super_wait(mddev
);
2175 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2177 spin_lock_irq(&mddev
->write_lock
);
2178 if (mddev
->in_sync
!= sync_req
||
2179 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2180 /* have to write it out again */
2181 spin_unlock_irq(&mddev
->write_lock
);
2184 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2185 spin_unlock_irq(&mddev
->write_lock
);
2186 wake_up(&mddev
->sb_wait
);
2187 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2188 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2192 /* words written to sysfs files may, or may not, be \n terminated.
2193 * We want to accept with case. For this we use cmd_match.
2195 static int cmd_match(const char *cmd
, const char *str
)
2197 /* See if cmd, written into a sysfs file, matches
2198 * str. They must either be the same, or cmd can
2199 * have a trailing newline
2201 while (*cmd
&& *str
&& *cmd
== *str
) {
2212 struct rdev_sysfs_entry
{
2213 struct attribute attr
;
2214 ssize_t (*show
)(mdk_rdev_t
*, char *);
2215 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2219 state_show(mdk_rdev_t
*rdev
, char *page
)
2224 if (test_bit(Faulty
, &rdev
->flags
)) {
2225 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2228 if (test_bit(In_sync
, &rdev
->flags
)) {
2229 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2232 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2233 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2236 if (test_bit(Blocked
, &rdev
->flags
)) {
2237 len
+= sprintf(page
+len
, "%sblocked", sep
);
2240 if (!test_bit(Faulty
, &rdev
->flags
) &&
2241 !test_bit(In_sync
, &rdev
->flags
)) {
2242 len
+= sprintf(page
+len
, "%sspare", sep
);
2245 return len
+sprintf(page
+len
, "\n");
2249 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2252 * faulty - simulates and error
2253 * remove - disconnects the device
2254 * writemostly - sets write_mostly
2255 * -writemostly - clears write_mostly
2256 * blocked - sets the Blocked flag
2257 * -blocked - clears the Blocked flag
2258 * insync - sets Insync providing device isn't active
2261 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2262 md_error(rdev
->mddev
, rdev
);
2264 } else if (cmd_match(buf
, "remove")) {
2265 if (rdev
->raid_disk
>= 0)
2268 mddev_t
*mddev
= rdev
->mddev
;
2269 kick_rdev_from_array(rdev
);
2271 md_update_sb(mddev
, 1);
2272 md_new_event(mddev
);
2275 } else if (cmd_match(buf
, "writemostly")) {
2276 set_bit(WriteMostly
, &rdev
->flags
);
2278 } else if (cmd_match(buf
, "-writemostly")) {
2279 clear_bit(WriteMostly
, &rdev
->flags
);
2281 } else if (cmd_match(buf
, "blocked")) {
2282 set_bit(Blocked
, &rdev
->flags
);
2284 } else if (cmd_match(buf
, "-blocked")) {
2285 clear_bit(Blocked
, &rdev
->flags
);
2286 wake_up(&rdev
->blocked_wait
);
2287 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2288 md_wakeup_thread(rdev
->mddev
->thread
);
2291 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2292 set_bit(In_sync
, &rdev
->flags
);
2295 if (!err
&& rdev
->sysfs_state
)
2296 sysfs_notify_dirent(rdev
->sysfs_state
);
2297 return err
? err
: len
;
2299 static struct rdev_sysfs_entry rdev_state
=
2300 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2303 errors_show(mdk_rdev_t
*rdev
, char *page
)
2305 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2309 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2312 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2313 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2314 atomic_set(&rdev
->corrected_errors
, n
);
2319 static struct rdev_sysfs_entry rdev_errors
=
2320 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2323 slot_show(mdk_rdev_t
*rdev
, char *page
)
2325 if (rdev
->raid_disk
< 0)
2326 return sprintf(page
, "none\n");
2328 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2332 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2337 int slot
= simple_strtoul(buf
, &e
, 10);
2338 if (strncmp(buf
, "none", 4)==0)
2340 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2342 if (rdev
->mddev
->pers
&& slot
== -1) {
2343 /* Setting 'slot' on an active array requires also
2344 * updating the 'rd%d' link, and communicating
2345 * with the personality with ->hot_*_disk.
2346 * For now we only support removing
2347 * failed/spare devices. This normally happens automatically,
2348 * but not when the metadata is externally managed.
2350 if (rdev
->raid_disk
== -1)
2352 /* personality does all needed checks */
2353 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2355 err
= rdev
->mddev
->pers
->
2356 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2359 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2360 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2361 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2362 md_wakeup_thread(rdev
->mddev
->thread
);
2363 } else if (rdev
->mddev
->pers
) {
2365 /* Activating a spare .. or possibly reactivating
2366 * if we ever get bitmaps working here.
2369 if (rdev
->raid_disk
!= -1)
2372 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2375 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2376 if (rdev2
->raid_disk
== slot
)
2379 rdev
->raid_disk
= slot
;
2380 if (test_bit(In_sync
, &rdev
->flags
))
2381 rdev
->saved_raid_disk
= slot
;
2383 rdev
->saved_raid_disk
= -1;
2384 err
= rdev
->mddev
->pers
->
2385 hot_add_disk(rdev
->mddev
, rdev
);
2387 rdev
->raid_disk
= -1;
2390 sysfs_notify_dirent(rdev
->sysfs_state
);
2391 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2392 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2394 "md: cannot register "
2396 nm
, mdname(rdev
->mddev
));
2398 /* don't wakeup anyone, leave that to userspace. */
2400 if (slot
>= rdev
->mddev
->raid_disks
)
2402 rdev
->raid_disk
= slot
;
2403 /* assume it is working */
2404 clear_bit(Faulty
, &rdev
->flags
);
2405 clear_bit(WriteMostly
, &rdev
->flags
);
2406 set_bit(In_sync
, &rdev
->flags
);
2407 sysfs_notify_dirent(rdev
->sysfs_state
);
2413 static struct rdev_sysfs_entry rdev_slot
=
2414 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2417 offset_show(mdk_rdev_t
*rdev
, char *page
)
2419 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2423 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2426 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2427 if (e
==buf
|| (*e
&& *e
!= '\n'))
2429 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2431 if (rdev
->sectors
&& rdev
->mddev
->external
)
2432 /* Must set offset before size, so overlap checks
2435 rdev
->data_offset
= offset
;
2439 static struct rdev_sysfs_entry rdev_offset
=
2440 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2443 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2445 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2448 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2450 /* check if two start/length pairs overlap */
2458 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2460 unsigned long long blocks
;
2463 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2466 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2467 return -EINVAL
; /* sector conversion overflow */
2470 if (new != blocks
* 2)
2471 return -EINVAL
; /* unsigned long long to sector_t overflow */
2478 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2480 mddev_t
*my_mddev
= rdev
->mddev
;
2481 sector_t oldsectors
= rdev
->sectors
;
2484 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2486 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2487 if (my_mddev
->persistent
) {
2488 sectors
= super_types
[my_mddev
->major_version
].
2489 rdev_size_change(rdev
, sectors
);
2492 } else if (!sectors
)
2493 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2496 if (sectors
< my_mddev
->dev_sectors
)
2497 return -EINVAL
; /* component must fit device */
2499 rdev
->sectors
= sectors
;
2500 if (sectors
> oldsectors
&& my_mddev
->external
) {
2501 /* need to check that all other rdevs with the same ->bdev
2502 * do not overlap. We need to unlock the mddev to avoid
2503 * a deadlock. We have already changed rdev->sectors, and if
2504 * we have to change it back, we will have the lock again.
2508 struct list_head
*tmp
;
2510 mddev_unlock(my_mddev
);
2511 for_each_mddev(mddev
, tmp
) {
2515 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2516 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2517 (rdev
->bdev
== rdev2
->bdev
&&
2519 overlaps(rdev
->data_offset
, rdev
->sectors
,
2525 mddev_unlock(mddev
);
2531 mddev_lock(my_mddev
);
2533 /* Someone else could have slipped in a size
2534 * change here, but doing so is just silly.
2535 * We put oldsectors back because we *know* it is
2536 * safe, and trust userspace not to race with
2539 rdev
->sectors
= oldsectors
;
2546 static struct rdev_sysfs_entry rdev_size
=
2547 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2549 static struct attribute
*rdev_default_attrs
[] = {
2558 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2560 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2561 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2562 mddev_t
*mddev
= rdev
->mddev
;
2568 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2570 if (rdev
->mddev
== NULL
)
2573 rv
= entry
->show(rdev
, page
);
2574 mddev_unlock(mddev
);
2580 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2581 const char *page
, size_t length
)
2583 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2584 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2586 mddev_t
*mddev
= rdev
->mddev
;
2590 if (!capable(CAP_SYS_ADMIN
))
2592 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2594 if (rdev
->mddev
== NULL
)
2597 rv
= entry
->store(rdev
, page
, length
);
2598 mddev_unlock(mddev
);
2603 static void rdev_free(struct kobject
*ko
)
2605 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2608 static struct sysfs_ops rdev_sysfs_ops
= {
2609 .show
= rdev_attr_show
,
2610 .store
= rdev_attr_store
,
2612 static struct kobj_type rdev_ktype
= {
2613 .release
= rdev_free
,
2614 .sysfs_ops
= &rdev_sysfs_ops
,
2615 .default_attrs
= rdev_default_attrs
,
2619 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2621 * mark the device faulty if:
2623 * - the device is nonexistent (zero size)
2624 * - the device has no valid superblock
2626 * a faulty rdev _never_ has rdev->sb set.
2628 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2630 char b
[BDEVNAME_SIZE
];
2635 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2637 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2638 return ERR_PTR(-ENOMEM
);
2641 if ((err
= alloc_disk_sb(rdev
)))
2644 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2648 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2651 rdev
->saved_raid_disk
= -1;
2652 rdev
->raid_disk
= -1;
2654 rdev
->data_offset
= 0;
2655 rdev
->sb_events
= 0;
2656 atomic_set(&rdev
->nr_pending
, 0);
2657 atomic_set(&rdev
->read_errors
, 0);
2658 atomic_set(&rdev
->corrected_errors
, 0);
2660 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2663 "md: %s has zero or unknown size, marking faulty!\n",
2664 bdevname(rdev
->bdev
,b
));
2669 if (super_format
>= 0) {
2670 err
= super_types
[super_format
].
2671 load_super(rdev
, NULL
, super_minor
);
2672 if (err
== -EINVAL
) {
2674 "md: %s does not have a valid v%d.%d "
2675 "superblock, not importing!\n",
2676 bdevname(rdev
->bdev
,b
),
2677 super_format
, super_minor
);
2682 "md: could not read %s's sb, not importing!\n",
2683 bdevname(rdev
->bdev
,b
));
2688 INIT_LIST_HEAD(&rdev
->same_set
);
2689 init_waitqueue_head(&rdev
->blocked_wait
);
2694 if (rdev
->sb_page
) {
2700 return ERR_PTR(err
);
2704 * Check a full RAID array for plausibility
2708 static void analyze_sbs(mddev_t
* mddev
)
2711 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2712 char b
[BDEVNAME_SIZE
];
2715 rdev_for_each(rdev
, tmp
, mddev
)
2716 switch (super_types
[mddev
->major_version
].
2717 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2725 "md: fatal superblock inconsistency in %s"
2726 " -- removing from array\n",
2727 bdevname(rdev
->bdev
,b
));
2728 kick_rdev_from_array(rdev
);
2732 super_types
[mddev
->major_version
].
2733 validate_super(mddev
, freshest
);
2736 rdev_for_each(rdev
, tmp
, mddev
) {
2737 if (rdev
->desc_nr
>= mddev
->max_disks
||
2738 i
> mddev
->max_disks
) {
2740 "md: %s: %s: only %d devices permitted\n",
2741 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2743 kick_rdev_from_array(rdev
);
2746 if (rdev
!= freshest
)
2747 if (super_types
[mddev
->major_version
].
2748 validate_super(mddev
, rdev
)) {
2749 printk(KERN_WARNING
"md: kicking non-fresh %s"
2751 bdevname(rdev
->bdev
,b
));
2752 kick_rdev_from_array(rdev
);
2755 if (mddev
->level
== LEVEL_MULTIPATH
) {
2756 rdev
->desc_nr
= i
++;
2757 rdev
->raid_disk
= rdev
->desc_nr
;
2758 set_bit(In_sync
, &rdev
->flags
);
2759 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2760 rdev
->raid_disk
= -1;
2761 clear_bit(In_sync
, &rdev
->flags
);
2766 static void md_safemode_timeout(unsigned long data
);
2769 safe_delay_show(mddev_t
*mddev
, char *page
)
2771 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2772 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2775 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2783 /* remove a period, and count digits after it */
2784 if (len
>= sizeof(buf
))
2786 strlcpy(buf
, cbuf
, sizeof(buf
));
2787 for (i
=0; i
<len
; i
++) {
2789 if (isdigit(buf
[i
])) {
2794 } else if (buf
[i
] == '.') {
2799 if (strict_strtoul(buf
, 10, &msec
) < 0)
2801 msec
= (msec
* 1000) / scale
;
2803 mddev
->safemode_delay
= 0;
2805 unsigned long old_delay
= mddev
->safemode_delay
;
2806 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2807 if (mddev
->safemode_delay
== 0)
2808 mddev
->safemode_delay
= 1;
2809 if (mddev
->safemode_delay
< old_delay
)
2810 md_safemode_timeout((unsigned long)mddev
);
2814 static struct md_sysfs_entry md_safe_delay
=
2815 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2818 level_show(mddev_t
*mddev
, char *page
)
2820 struct mdk_personality
*p
= mddev
->pers
;
2822 return sprintf(page
, "%s\n", p
->name
);
2823 else if (mddev
->clevel
[0])
2824 return sprintf(page
, "%s\n", mddev
->clevel
);
2825 else if (mddev
->level
!= LEVEL_NONE
)
2826 return sprintf(page
, "%d\n", mddev
->level
);
2832 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2836 struct mdk_personality
*pers
;
2840 if (mddev
->pers
== NULL
) {
2843 if (len
>= sizeof(mddev
->clevel
))
2845 strncpy(mddev
->clevel
, buf
, len
);
2846 if (mddev
->clevel
[len
-1] == '\n')
2848 mddev
->clevel
[len
] = 0;
2849 mddev
->level
= LEVEL_NONE
;
2853 /* request to change the personality. Need to ensure:
2854 * - array is not engaged in resync/recovery/reshape
2855 * - old personality can be suspended
2856 * - new personality will access other array.
2859 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2862 if (!mddev
->pers
->quiesce
) {
2863 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2864 mdname(mddev
), mddev
->pers
->name
);
2868 /* Now find the new personality */
2869 if (len
== 0 || len
>= sizeof(level
))
2871 strncpy(level
, buf
, len
);
2872 if (level
[len
-1] == '\n')
2876 request_module("md-%s", level
);
2877 spin_lock(&pers_lock
);
2878 pers
= find_pers(LEVEL_NONE
, level
);
2879 if (!pers
|| !try_module_get(pers
->owner
)) {
2880 spin_unlock(&pers_lock
);
2881 printk(KERN_WARNING
"md: personality %s not loaded\n", level
);
2884 spin_unlock(&pers_lock
);
2886 if (pers
== mddev
->pers
) {
2887 /* Nothing to do! */
2888 module_put(pers
->owner
);
2891 if (!pers
->takeover
) {
2892 module_put(pers
->owner
);
2893 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2894 mdname(mddev
), level
);
2898 /* ->takeover must set new_* and/or delta_disks
2899 * if it succeeds, and may set them when it fails.
2901 priv
= pers
->takeover(mddev
);
2903 mddev
->new_level
= mddev
->level
;
2904 mddev
->new_layout
= mddev
->layout
;
2905 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2906 mddev
->raid_disks
-= mddev
->delta_disks
;
2907 mddev
->delta_disks
= 0;
2908 module_put(pers
->owner
);
2909 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
2910 mdname(mddev
), level
);
2911 return PTR_ERR(priv
);
2914 /* Looks like we have a winner */
2915 mddev_suspend(mddev
);
2916 mddev
->pers
->stop(mddev
);
2917 module_put(mddev
->pers
->owner
);
2918 /* Invalidate devices that are now superfluous */
2919 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2920 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2921 rdev
->raid_disk
= -1;
2922 clear_bit(In_sync
, &rdev
->flags
);
2925 mddev
->private = priv
;
2926 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2927 mddev
->level
= mddev
->new_level
;
2928 mddev
->layout
= mddev
->new_layout
;
2929 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
2930 mddev
->delta_disks
= 0;
2932 mddev_resume(mddev
);
2933 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2934 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2935 md_wakeup_thread(mddev
->thread
);
2939 static struct md_sysfs_entry md_level
=
2940 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2944 layout_show(mddev_t
*mddev
, char *page
)
2946 /* just a number, not meaningful for all levels */
2947 if (mddev
->reshape_position
!= MaxSector
&&
2948 mddev
->layout
!= mddev
->new_layout
)
2949 return sprintf(page
, "%d (%d)\n",
2950 mddev
->new_layout
, mddev
->layout
);
2951 return sprintf(page
, "%d\n", mddev
->layout
);
2955 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2958 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2960 if (!*buf
|| (*e
&& *e
!= '\n'))
2965 if (mddev
->pers
->check_reshape
== NULL
)
2967 mddev
->new_layout
= n
;
2968 err
= mddev
->pers
->check_reshape(mddev
);
2970 mddev
->new_layout
= mddev
->layout
;
2974 mddev
->new_layout
= n
;
2975 if (mddev
->reshape_position
== MaxSector
)
2980 static struct md_sysfs_entry md_layout
=
2981 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2985 raid_disks_show(mddev_t
*mddev
, char *page
)
2987 if (mddev
->raid_disks
== 0)
2989 if (mddev
->reshape_position
!= MaxSector
&&
2990 mddev
->delta_disks
!= 0)
2991 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
2992 mddev
->raid_disks
- mddev
->delta_disks
);
2993 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2996 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2999 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3003 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3005 if (!*buf
|| (*e
&& *e
!= '\n'))
3009 rv
= update_raid_disks(mddev
, n
);
3010 else if (mddev
->reshape_position
!= MaxSector
) {
3011 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3012 mddev
->delta_disks
= n
- olddisks
;
3013 mddev
->raid_disks
= n
;
3015 mddev
->raid_disks
= n
;
3016 return rv
? rv
: len
;
3018 static struct md_sysfs_entry md_raid_disks
=
3019 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3022 chunk_size_show(mddev_t
*mddev
, char *page
)
3024 if (mddev
->reshape_position
!= MaxSector
&&
3025 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3026 return sprintf(page
, "%d (%d)\n",
3027 mddev
->new_chunk_sectors
<< 9,
3028 mddev
->chunk_sectors
<< 9);
3029 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3033 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3036 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3038 if (!*buf
|| (*e
&& *e
!= '\n'))
3043 if (mddev
->pers
->check_reshape
== NULL
)
3045 mddev
->new_chunk_sectors
= n
>> 9;
3046 err
= mddev
->pers
->check_reshape(mddev
);
3048 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3052 mddev
->new_chunk_sectors
= n
>> 9;
3053 if (mddev
->reshape_position
== MaxSector
)
3054 mddev
->chunk_sectors
= n
>> 9;
3058 static struct md_sysfs_entry md_chunk_size
=
3059 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3062 resync_start_show(mddev_t
*mddev
, char *page
)
3064 if (mddev
->recovery_cp
== MaxSector
)
3065 return sprintf(page
, "none\n");
3066 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3070 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3073 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3077 if (!*buf
|| (*e
&& *e
!= '\n'))
3080 mddev
->recovery_cp
= n
;
3083 static struct md_sysfs_entry md_resync_start
=
3084 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3087 * The array state can be:
3090 * No devices, no size, no level
3091 * Equivalent to STOP_ARRAY ioctl
3093 * May have some settings, but array is not active
3094 * all IO results in error
3095 * When written, doesn't tear down array, but just stops it
3096 * suspended (not supported yet)
3097 * All IO requests will block. The array can be reconfigured.
3098 * Writing this, if accepted, will block until array is quiescent
3100 * no resync can happen. no superblocks get written.
3101 * write requests fail
3103 * like readonly, but behaves like 'clean' on a write request.
3105 * clean - no pending writes, but otherwise active.
3106 * When written to inactive array, starts without resync
3107 * If a write request arrives then
3108 * if metadata is known, mark 'dirty' and switch to 'active'.
3109 * if not known, block and switch to write-pending
3110 * If written to an active array that has pending writes, then fails.
3112 * fully active: IO and resync can be happening.
3113 * When written to inactive array, starts with resync
3116 * clean, but writes are blocked waiting for 'active' to be written.
3119 * like active, but no writes have been seen for a while (100msec).
3122 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3123 write_pending
, active_idle
, bad_word
};
3124 static char *array_states
[] = {
3125 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3126 "write-pending", "active-idle", NULL
};
3128 static int match_word(const char *word
, char **list
)
3131 for (n
=0; list
[n
]; n
++)
3132 if (cmd_match(word
, list
[n
]))
3138 array_state_show(mddev_t
*mddev
, char *page
)
3140 enum array_state st
= inactive
;
3153 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3155 else if (mddev
->safemode
)
3161 if (list_empty(&mddev
->disks
) &&
3162 mddev
->raid_disks
== 0 &&
3163 mddev
->dev_sectors
== 0)
3168 return sprintf(page
, "%s\n", array_states
[st
]);
3171 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3172 static int do_md_run(mddev_t
* mddev
);
3173 static int restart_array(mddev_t
*mddev
);
3176 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3179 enum array_state st
= match_word(buf
, array_states
);
3184 /* stopping an active array */
3185 if (atomic_read(&mddev
->openers
) > 0)
3187 err
= do_md_stop(mddev
, 0, 0);
3190 /* stopping an active array */
3192 if (atomic_read(&mddev
->openers
) > 0)
3194 err
= do_md_stop(mddev
, 2, 0);
3196 err
= 0; /* already inactive */
3199 break; /* not supported yet */
3202 err
= do_md_stop(mddev
, 1, 0);
3205 set_disk_ro(mddev
->gendisk
, 1);
3206 err
= do_md_run(mddev
);
3212 err
= do_md_stop(mddev
, 1, 0);
3213 else if (mddev
->ro
== 1)
3214 err
= restart_array(mddev
);
3217 set_disk_ro(mddev
->gendisk
, 0);
3221 err
= do_md_run(mddev
);
3226 restart_array(mddev
);
3227 spin_lock_irq(&mddev
->write_lock
);
3228 if (atomic_read(&mddev
->writes_pending
) == 0) {
3229 if (mddev
->in_sync
== 0) {
3231 if (mddev
->safemode
== 1)
3232 mddev
->safemode
= 0;
3233 if (mddev
->persistent
)
3234 set_bit(MD_CHANGE_CLEAN
,
3240 spin_unlock_irq(&mddev
->write_lock
);
3246 restart_array(mddev
);
3247 if (mddev
->external
)
3248 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3249 wake_up(&mddev
->sb_wait
);
3253 set_disk_ro(mddev
->gendisk
, 0);
3254 err
= do_md_run(mddev
);
3259 /* these cannot be set */
3265 sysfs_notify_dirent(mddev
->sysfs_state
);
3269 static struct md_sysfs_entry md_array_state
=
3270 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3273 null_show(mddev_t
*mddev
, char *page
)
3279 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3281 /* buf must be %d:%d\n? giving major and minor numbers */
3282 /* The new device is added to the array.
3283 * If the array has a persistent superblock, we read the
3284 * superblock to initialise info and check validity.
3285 * Otherwise, only checking done is that in bind_rdev_to_array,
3286 * which mainly checks size.
3289 int major
= simple_strtoul(buf
, &e
, 10);
3295 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3297 minor
= simple_strtoul(e
+1, &e
, 10);
3298 if (*e
&& *e
!= '\n')
3300 dev
= MKDEV(major
, minor
);
3301 if (major
!= MAJOR(dev
) ||
3302 minor
!= MINOR(dev
))
3306 if (mddev
->persistent
) {
3307 rdev
= md_import_device(dev
, mddev
->major_version
,
3308 mddev
->minor_version
);
3309 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3310 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3311 mdk_rdev_t
, same_set
);
3312 err
= super_types
[mddev
->major_version
]
3313 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3317 } else if (mddev
->external
)
3318 rdev
= md_import_device(dev
, -2, -1);
3320 rdev
= md_import_device(dev
, -1, -1);
3323 return PTR_ERR(rdev
);
3324 err
= bind_rdev_to_array(rdev
, mddev
);
3328 return err
? err
: len
;
3331 static struct md_sysfs_entry md_new_device
=
3332 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3335 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3338 unsigned long chunk
, end_chunk
;
3342 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3344 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3345 if (buf
== end
) break;
3346 if (*end
== '-') { /* range */
3348 end_chunk
= simple_strtoul(buf
, &end
, 0);
3349 if (buf
== end
) break;
3351 if (*end
&& !isspace(*end
)) break;
3352 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3354 while (isspace(*buf
)) buf
++;
3356 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3361 static struct md_sysfs_entry md_bitmap
=
3362 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3365 size_show(mddev_t
*mddev
, char *page
)
3367 return sprintf(page
, "%llu\n",
3368 (unsigned long long)mddev
->dev_sectors
/ 2);
3371 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3374 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3376 /* If array is inactive, we can reduce the component size, but
3377 * not increase it (except from 0).
3378 * If array is active, we can try an on-line resize
3381 int err
= strict_blocks_to_sectors(buf
, §ors
);
3386 err
= update_size(mddev
, sectors
);
3387 md_update_sb(mddev
, 1);
3389 if (mddev
->dev_sectors
== 0 ||
3390 mddev
->dev_sectors
> sectors
)
3391 mddev
->dev_sectors
= sectors
;
3395 return err
? err
: len
;
3398 static struct md_sysfs_entry md_size
=
3399 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3404 * 'none' for arrays with no metadata (good luck...)
3405 * 'external' for arrays with externally managed metadata,
3406 * or N.M for internally known formats
3409 metadata_show(mddev_t
*mddev
, char *page
)
3411 if (mddev
->persistent
)
3412 return sprintf(page
, "%d.%d\n",
3413 mddev
->major_version
, mddev
->minor_version
);
3414 else if (mddev
->external
)
3415 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3417 return sprintf(page
, "none\n");
3421 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3425 /* Changing the details of 'external' metadata is
3426 * always permitted. Otherwise there must be
3427 * no devices attached to the array.
3429 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3431 else if (!list_empty(&mddev
->disks
))
3434 if (cmd_match(buf
, "none")) {
3435 mddev
->persistent
= 0;
3436 mddev
->external
= 0;
3437 mddev
->major_version
= 0;
3438 mddev
->minor_version
= 90;
3441 if (strncmp(buf
, "external:", 9) == 0) {
3442 size_t namelen
= len
-9;
3443 if (namelen
>= sizeof(mddev
->metadata_type
))
3444 namelen
= sizeof(mddev
->metadata_type
)-1;
3445 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3446 mddev
->metadata_type
[namelen
] = 0;
3447 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3448 mddev
->metadata_type
[--namelen
] = 0;
3449 mddev
->persistent
= 0;
3450 mddev
->external
= 1;
3451 mddev
->major_version
= 0;
3452 mddev
->minor_version
= 90;
3455 major
= simple_strtoul(buf
, &e
, 10);
3456 if (e
==buf
|| *e
!= '.')
3459 minor
= simple_strtoul(buf
, &e
, 10);
3460 if (e
==buf
|| (*e
&& *e
!= '\n') )
3462 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3464 mddev
->major_version
= major
;
3465 mddev
->minor_version
= minor
;
3466 mddev
->persistent
= 1;
3467 mddev
->external
= 0;
3471 static struct md_sysfs_entry md_metadata
=
3472 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3475 action_show(mddev_t
*mddev
, char *page
)
3477 char *type
= "idle";
3478 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3480 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3481 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3482 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3484 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3485 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3487 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3491 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3494 return sprintf(page
, "%s\n", type
);
3498 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3500 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3503 if (cmd_match(page
, "frozen"))
3504 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3506 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3508 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3509 if (mddev
->sync_thread
) {
3510 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3511 md_unregister_thread(mddev
->sync_thread
);
3512 mddev
->sync_thread
= NULL
;
3513 mddev
->recovery
= 0;
3515 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3516 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3518 else if (cmd_match(page
, "resync"))
3519 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3520 else if (cmd_match(page
, "recover")) {
3521 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3522 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3523 } else if (cmd_match(page
, "reshape")) {
3525 if (mddev
->pers
->start_reshape
== NULL
)
3527 err
= mddev
->pers
->start_reshape(mddev
);
3530 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3532 if (cmd_match(page
, "check"))
3533 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3534 else if (!cmd_match(page
, "repair"))
3536 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3537 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3539 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3540 md_wakeup_thread(mddev
->thread
);
3541 sysfs_notify_dirent(mddev
->sysfs_action
);
3546 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3548 return sprintf(page
, "%llu\n",
3549 (unsigned long long) mddev
->resync_mismatches
);
3552 static struct md_sysfs_entry md_scan_mode
=
3553 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3556 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3559 sync_min_show(mddev_t
*mddev
, char *page
)
3561 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3562 mddev
->sync_speed_min
? "local": "system");
3566 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3570 if (strncmp(buf
, "system", 6)==0) {
3571 mddev
->sync_speed_min
= 0;
3574 min
= simple_strtoul(buf
, &e
, 10);
3575 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3577 mddev
->sync_speed_min
= min
;
3581 static struct md_sysfs_entry md_sync_min
=
3582 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3585 sync_max_show(mddev_t
*mddev
, char *page
)
3587 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3588 mddev
->sync_speed_max
? "local": "system");
3592 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3596 if (strncmp(buf
, "system", 6)==0) {
3597 mddev
->sync_speed_max
= 0;
3600 max
= simple_strtoul(buf
, &e
, 10);
3601 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3603 mddev
->sync_speed_max
= max
;
3607 static struct md_sysfs_entry md_sync_max
=
3608 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3611 degraded_show(mddev_t
*mddev
, char *page
)
3613 return sprintf(page
, "%d\n", mddev
->degraded
);
3615 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3618 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3620 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3624 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3628 if (strict_strtol(buf
, 10, &n
))
3631 if (n
!= 0 && n
!= 1)
3634 mddev
->parallel_resync
= n
;
3636 if (mddev
->sync_thread
)
3637 wake_up(&resync_wait
);
3642 /* force parallel resync, even with shared block devices */
3643 static struct md_sysfs_entry md_sync_force_parallel
=
3644 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3645 sync_force_parallel_show
, sync_force_parallel_store
);
3648 sync_speed_show(mddev_t
*mddev
, char *page
)
3650 unsigned long resync
, dt
, db
;
3651 if (mddev
->curr_resync
== 0)
3652 return sprintf(page
, "none\n");
3653 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3654 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3656 db
= resync
- mddev
->resync_mark_cnt
;
3657 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3660 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3663 sync_completed_show(mddev_t
*mddev
, char *page
)
3665 unsigned long max_sectors
, resync
;
3667 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3668 return sprintf(page
, "none\n");
3670 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3671 max_sectors
= mddev
->resync_max_sectors
;
3673 max_sectors
= mddev
->dev_sectors
;
3675 resync
= mddev
->curr_resync_completed
;
3676 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3679 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3682 min_sync_show(mddev_t
*mddev
, char *page
)
3684 return sprintf(page
, "%llu\n",
3685 (unsigned long long)mddev
->resync_min
);
3688 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3690 unsigned long long min
;
3691 if (strict_strtoull(buf
, 10, &min
))
3693 if (min
> mddev
->resync_max
)
3695 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3698 /* Must be a multiple of chunk_size */
3699 if (mddev
->chunk_sectors
) {
3700 sector_t temp
= min
;
3701 if (sector_div(temp
, mddev
->chunk_sectors
))
3704 mddev
->resync_min
= min
;
3709 static struct md_sysfs_entry md_min_sync
=
3710 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3713 max_sync_show(mddev_t
*mddev
, char *page
)
3715 if (mddev
->resync_max
== MaxSector
)
3716 return sprintf(page
, "max\n");
3718 return sprintf(page
, "%llu\n",
3719 (unsigned long long)mddev
->resync_max
);
3722 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3724 if (strncmp(buf
, "max", 3) == 0)
3725 mddev
->resync_max
= MaxSector
;
3727 unsigned long long max
;
3728 if (strict_strtoull(buf
, 10, &max
))
3730 if (max
< mddev
->resync_min
)
3732 if (max
< mddev
->resync_max
&&
3734 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3737 /* Must be a multiple of chunk_size */
3738 if (mddev
->chunk_sectors
) {
3739 sector_t temp
= max
;
3740 if (sector_div(temp
, mddev
->chunk_sectors
))
3743 mddev
->resync_max
= max
;
3745 wake_up(&mddev
->recovery_wait
);
3749 static struct md_sysfs_entry md_max_sync
=
3750 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3753 suspend_lo_show(mddev_t
*mddev
, char *page
)
3755 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3759 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3762 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3764 if (mddev
->pers
== NULL
||
3765 mddev
->pers
->quiesce
== NULL
)
3767 if (buf
== e
|| (*e
&& *e
!= '\n'))
3769 if (new >= mddev
->suspend_hi
||
3770 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3771 mddev
->suspend_lo
= new;
3772 mddev
->pers
->quiesce(mddev
, 2);
3777 static struct md_sysfs_entry md_suspend_lo
=
3778 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3782 suspend_hi_show(mddev_t
*mddev
, char *page
)
3784 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3788 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3791 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3793 if (mddev
->pers
== NULL
||
3794 mddev
->pers
->quiesce
== NULL
)
3796 if (buf
== e
|| (*e
&& *e
!= '\n'))
3798 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3799 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3800 mddev
->suspend_hi
= new;
3801 mddev
->pers
->quiesce(mddev
, 1);
3802 mddev
->pers
->quiesce(mddev
, 0);
3807 static struct md_sysfs_entry md_suspend_hi
=
3808 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3811 reshape_position_show(mddev_t
*mddev
, char *page
)
3813 if (mddev
->reshape_position
!= MaxSector
)
3814 return sprintf(page
, "%llu\n",
3815 (unsigned long long)mddev
->reshape_position
);
3816 strcpy(page
, "none\n");
3821 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3824 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3827 if (buf
== e
|| (*e
&& *e
!= '\n'))
3829 mddev
->reshape_position
= new;
3830 mddev
->delta_disks
= 0;
3831 mddev
->new_level
= mddev
->level
;
3832 mddev
->new_layout
= mddev
->layout
;
3833 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3837 static struct md_sysfs_entry md_reshape_position
=
3838 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3839 reshape_position_store
);
3842 array_size_show(mddev_t
*mddev
, char *page
)
3844 if (mddev
->external_size
)
3845 return sprintf(page
, "%llu\n",
3846 (unsigned long long)mddev
->array_sectors
/2);
3848 return sprintf(page
, "default\n");
3852 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3856 if (strncmp(buf
, "default", 7) == 0) {
3858 sectors
= mddev
->pers
->size(mddev
, 0, 0);
3860 sectors
= mddev
->array_sectors
;
3862 mddev
->external_size
= 0;
3864 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3866 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
3869 mddev
->external_size
= 1;
3872 mddev
->array_sectors
= sectors
;
3873 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
3875 revalidate_disk(mddev
->gendisk
);
3880 static struct md_sysfs_entry md_array_size
=
3881 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
3884 static struct attribute
*md_default_attrs
[] = {
3887 &md_raid_disks
.attr
,
3888 &md_chunk_size
.attr
,
3890 &md_resync_start
.attr
,
3892 &md_new_device
.attr
,
3893 &md_safe_delay
.attr
,
3894 &md_array_state
.attr
,
3895 &md_reshape_position
.attr
,
3896 &md_array_size
.attr
,
3900 static struct attribute
*md_redundancy_attrs
[] = {
3902 &md_mismatches
.attr
,
3905 &md_sync_speed
.attr
,
3906 &md_sync_force_parallel
.attr
,
3907 &md_sync_completed
.attr
,
3910 &md_suspend_lo
.attr
,
3911 &md_suspend_hi
.attr
,
3916 static struct attribute_group md_redundancy_group
= {
3918 .attrs
= md_redundancy_attrs
,
3923 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3925 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3926 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3931 rv
= mddev_lock(mddev
);
3933 rv
= entry
->show(mddev
, page
);
3934 mddev_unlock(mddev
);
3940 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3941 const char *page
, size_t length
)
3943 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3944 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3949 if (!capable(CAP_SYS_ADMIN
))
3951 rv
= mddev_lock(mddev
);
3952 if (mddev
->hold_active
== UNTIL_IOCTL
)
3953 mddev
->hold_active
= 0;
3955 rv
= entry
->store(mddev
, page
, length
);
3956 mddev_unlock(mddev
);
3961 static void md_free(struct kobject
*ko
)
3963 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
3965 if (mddev
->sysfs_state
)
3966 sysfs_put(mddev
->sysfs_state
);
3968 if (mddev
->gendisk
) {
3969 del_gendisk(mddev
->gendisk
);
3970 put_disk(mddev
->gendisk
);
3973 blk_cleanup_queue(mddev
->queue
);
3978 static struct sysfs_ops md_sysfs_ops
= {
3979 .show
= md_attr_show
,
3980 .store
= md_attr_store
,
3982 static struct kobj_type md_ktype
= {
3984 .sysfs_ops
= &md_sysfs_ops
,
3985 .default_attrs
= md_default_attrs
,
3990 static void mddev_delayed_delete(struct work_struct
*ws
)
3992 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
3994 if (mddev
->private == &md_redundancy_group
) {
3995 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3996 if (mddev
->sysfs_action
)
3997 sysfs_put(mddev
->sysfs_action
);
3998 mddev
->sysfs_action
= NULL
;
3999 mddev
->private = NULL
;
4001 kobject_del(&mddev
->kobj
);
4002 kobject_put(&mddev
->kobj
);
4005 static int md_alloc(dev_t dev
, char *name
)
4007 static DEFINE_MUTEX(disks_mutex
);
4008 mddev_t
*mddev
= mddev_find(dev
);
4009 struct gendisk
*disk
;
4018 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4019 shift
= partitioned
? MdpMinorShift
: 0;
4020 unit
= MINOR(mddev
->unit
) >> shift
;
4022 /* wait for any previous instance if this device
4023 * to be completed removed (mddev_delayed_delete).
4025 flush_scheduled_work();
4027 mutex_lock(&disks_mutex
);
4033 /* Need to ensure that 'name' is not a duplicate.
4036 spin_lock(&all_mddevs_lock
);
4038 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4039 if (mddev2
->gendisk
&&
4040 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4041 spin_unlock(&all_mddevs_lock
);
4044 spin_unlock(&all_mddevs_lock
);
4048 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4051 mddev
->queue
->queuedata
= mddev
;
4053 /* Can be unlocked because the queue is new: no concurrency */
4054 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4056 blk_queue_make_request(mddev
->queue
, md_make_request
);
4058 disk
= alloc_disk(1 << shift
);
4060 blk_cleanup_queue(mddev
->queue
);
4061 mddev
->queue
= NULL
;
4064 disk
->major
= MAJOR(mddev
->unit
);
4065 disk
->first_minor
= unit
<< shift
;
4067 strcpy(disk
->disk_name
, name
);
4068 else if (partitioned
)
4069 sprintf(disk
->disk_name
, "md_d%d", unit
);
4071 sprintf(disk
->disk_name
, "md%d", unit
);
4072 disk
->fops
= &md_fops
;
4073 disk
->private_data
= mddev
;
4074 disk
->queue
= mddev
->queue
;
4075 /* Allow extended partitions. This makes the
4076 * 'mdp' device redundant, but we can't really
4079 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4081 mddev
->gendisk
= disk
;
4082 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4083 &disk_to_dev(disk
)->kobj
, "%s", "md");
4085 /* This isn't possible, but as kobject_init_and_add is marked
4086 * __must_check, we must do something with the result
4088 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4093 mutex_unlock(&disks_mutex
);
4095 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4096 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4102 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4104 md_alloc(dev
, NULL
);
4108 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4110 /* val must be "md_*" where * is not all digits.
4111 * We allocate an array with a large free minor number, and
4112 * set the name to val. val must not already be an active name.
4114 int len
= strlen(val
);
4115 char buf
[DISK_NAME_LEN
];
4117 while (len
&& val
[len
-1] == '\n')
4119 if (len
>= DISK_NAME_LEN
)
4121 strlcpy(buf
, val
, len
+1);
4122 if (strncmp(buf
, "md_", 3) != 0)
4124 return md_alloc(0, buf
);
4127 static void md_safemode_timeout(unsigned long data
)
4129 mddev_t
*mddev
= (mddev_t
*) data
;
4131 if (!atomic_read(&mddev
->writes_pending
)) {
4132 mddev
->safemode
= 1;
4133 if (mddev
->external
)
4134 sysfs_notify_dirent(mddev
->sysfs_state
);
4136 md_wakeup_thread(mddev
->thread
);
4139 static int start_dirty_degraded
;
4141 static int do_md_run(mddev_t
* mddev
)
4145 struct gendisk
*disk
;
4146 struct mdk_personality
*pers
;
4148 if (list_empty(&mddev
->disks
))
4149 /* cannot run an array with no devices.. */
4156 * Analyze all RAID superblock(s)
4158 if (!mddev
->raid_disks
) {
4159 if (!mddev
->persistent
)
4164 if (mddev
->level
!= LEVEL_NONE
)
4165 request_module("md-level-%d", mddev
->level
);
4166 else if (mddev
->clevel
[0])
4167 request_module("md-%s", mddev
->clevel
);
4170 * Drop all container device buffers, from now on
4171 * the only valid external interface is through the md
4174 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4175 if (test_bit(Faulty
, &rdev
->flags
))
4177 sync_blockdev(rdev
->bdev
);
4178 invalidate_bdev(rdev
->bdev
);
4180 /* perform some consistency tests on the device.
4181 * We don't want the data to overlap the metadata,
4182 * Internal Bitmap issues have been handled elsewhere.
4184 if (rdev
->data_offset
< rdev
->sb_start
) {
4185 if (mddev
->dev_sectors
&&
4186 rdev
->data_offset
+ mddev
->dev_sectors
4188 printk("md: %s: data overlaps metadata\n",
4193 if (rdev
->sb_start
+ rdev
->sb_size
/512
4194 > rdev
->data_offset
) {
4195 printk("md: %s: metadata overlaps data\n",
4200 sysfs_notify_dirent(rdev
->sysfs_state
);
4203 md_probe(mddev
->unit
, NULL
, NULL
);
4204 disk
= mddev
->gendisk
;
4208 spin_lock(&pers_lock
);
4209 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4210 if (!pers
|| !try_module_get(pers
->owner
)) {
4211 spin_unlock(&pers_lock
);
4212 if (mddev
->level
!= LEVEL_NONE
)
4213 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4216 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4221 spin_unlock(&pers_lock
);
4222 if (mddev
->level
!= pers
->level
) {
4223 mddev
->level
= pers
->level
;
4224 mddev
->new_level
= pers
->level
;
4226 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4228 if (mddev
->reshape_position
!= MaxSector
&&
4229 pers
->start_reshape
== NULL
) {
4230 /* This personality cannot handle reshaping... */
4232 module_put(pers
->owner
);
4236 if (pers
->sync_request
) {
4237 /* Warn if this is a potentially silly
4240 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4244 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4245 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4247 rdev
->bdev
->bd_contains
==
4248 rdev2
->bdev
->bd_contains
) {
4250 "%s: WARNING: %s appears to be"
4251 " on the same physical disk as"
4254 bdevname(rdev
->bdev
,b
),
4255 bdevname(rdev2
->bdev
,b2
));
4262 "True protection against single-disk"
4263 " failure might be compromised.\n");
4266 mddev
->recovery
= 0;
4267 /* may be over-ridden by personality */
4268 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4270 mddev
->barriers_work
= 1;
4271 mddev
->ok_start_degraded
= start_dirty_degraded
;
4274 mddev
->ro
= 2; /* read-only, but switch on first write */
4276 err
= mddev
->pers
->run(mddev
);
4278 printk(KERN_ERR
"md: pers->run() failed ...\n");
4279 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4280 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4281 " but 'external_size' not in effect?\n", __func__
);
4283 "md: invalid array_size %llu > default size %llu\n",
4284 (unsigned long long)mddev
->array_sectors
/ 2,
4285 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4287 mddev
->pers
->stop(mddev
);
4289 if (err
== 0 && mddev
->pers
->sync_request
) {
4290 err
= bitmap_create(mddev
);
4292 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4293 mdname(mddev
), err
);
4294 mddev
->pers
->stop(mddev
);
4298 module_put(mddev
->pers
->owner
);
4300 bitmap_destroy(mddev
);
4303 if (mddev
->pers
->sync_request
) {
4304 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4306 "md: cannot register extra attributes for %s\n",
4308 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4309 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4312 atomic_set(&mddev
->writes_pending
,0);
4313 mddev
->safemode
= 0;
4314 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4315 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4316 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4319 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4320 if (rdev
->raid_disk
>= 0) {
4322 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4323 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4324 printk("md: cannot register %s for %s\n",
4328 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4331 md_update_sb(mddev
, 0);
4333 set_capacity(disk
, mddev
->array_sectors
);
4335 /* If there is a partially-recovered drive we need to
4336 * start recovery here. If we leave it to md_check_recovery,
4337 * it will remove the drives and not do the right thing
4339 if (mddev
->degraded
&& !mddev
->sync_thread
) {
4341 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4342 if (rdev
->raid_disk
>= 0 &&
4343 !test_bit(In_sync
, &rdev
->flags
) &&
4344 !test_bit(Faulty
, &rdev
->flags
))
4345 /* complete an interrupted recovery */
4347 if (spares
&& mddev
->pers
->sync_request
) {
4348 mddev
->recovery
= 0;
4349 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4350 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4353 if (!mddev
->sync_thread
) {
4354 printk(KERN_ERR
"%s: could not start resync"
4357 /* leave the spares where they are, it shouldn't hurt */
4358 mddev
->recovery
= 0;
4362 md_wakeup_thread(mddev
->thread
);
4363 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4365 revalidate_disk(mddev
->gendisk
);
4367 md_new_event(mddev
);
4368 sysfs_notify_dirent(mddev
->sysfs_state
);
4369 if (mddev
->sysfs_action
)
4370 sysfs_notify_dirent(mddev
->sysfs_action
);
4371 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4372 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4376 static int restart_array(mddev_t
*mddev
)
4378 struct gendisk
*disk
= mddev
->gendisk
;
4380 /* Complain if it has no devices */
4381 if (list_empty(&mddev
->disks
))
4387 mddev
->safemode
= 0;
4389 set_disk_ro(disk
, 0);
4390 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4392 /* Kick recovery or resync if necessary */
4393 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4394 md_wakeup_thread(mddev
->thread
);
4395 md_wakeup_thread(mddev
->sync_thread
);
4396 sysfs_notify_dirent(mddev
->sysfs_state
);
4400 /* similar to deny_write_access, but accounts for our holding a reference
4401 * to the file ourselves */
4402 static int deny_bitmap_write_access(struct file
* file
)
4404 struct inode
*inode
= file
->f_mapping
->host
;
4406 spin_lock(&inode
->i_lock
);
4407 if (atomic_read(&inode
->i_writecount
) > 1) {
4408 spin_unlock(&inode
->i_lock
);
4411 atomic_set(&inode
->i_writecount
, -1);
4412 spin_unlock(&inode
->i_lock
);
4417 static void restore_bitmap_write_access(struct file
*file
)
4419 struct inode
*inode
= file
->f_mapping
->host
;
4421 spin_lock(&inode
->i_lock
);
4422 atomic_set(&inode
->i_writecount
, 1);
4423 spin_unlock(&inode
->i_lock
);
4427 * 0 - completely stop and dis-assemble array
4428 * 1 - switch to readonly
4429 * 2 - stop but do not disassemble array
4431 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4434 struct gendisk
*disk
= mddev
->gendisk
;
4437 mutex_lock(&mddev
->open_mutex
);
4438 if (atomic_read(&mddev
->openers
) > is_open
) {
4439 printk("md: %s still in use.\n",mdname(mddev
));
4441 } else if (mddev
->pers
) {
4443 if (mddev
->sync_thread
) {
4444 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4445 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4446 md_unregister_thread(mddev
->sync_thread
);
4447 mddev
->sync_thread
= NULL
;
4450 del_timer_sync(&mddev
->safemode_timer
);
4453 case 1: /* readonly */
4459 case 0: /* disassemble */
4461 bitmap_flush(mddev
);
4462 md_super_wait(mddev
);
4464 set_disk_ro(disk
, 0);
4466 mddev
->pers
->stop(mddev
);
4467 mddev
->queue
->merge_bvec_fn
= NULL
;
4468 mddev
->queue
->unplug_fn
= NULL
;
4469 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4470 module_put(mddev
->pers
->owner
);
4471 if (mddev
->pers
->sync_request
)
4472 mddev
->private = &md_redundancy_group
;
4474 /* tell userspace to handle 'inactive' */
4475 sysfs_notify_dirent(mddev
->sysfs_state
);
4477 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4478 if (rdev
->raid_disk
>= 0) {
4480 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4481 sysfs_remove_link(&mddev
->kobj
, nm
);
4484 set_capacity(disk
, 0);
4490 if (!mddev
->in_sync
|| mddev
->flags
) {
4491 /* mark array as shutdown cleanly */
4493 md_update_sb(mddev
, 1);
4496 set_disk_ro(disk
, 1);
4497 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4501 mutex_unlock(&mddev
->open_mutex
);
4505 * Free resources if final stop
4509 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4511 bitmap_destroy(mddev
);
4512 if (mddev
->bitmap_info
.file
) {
4513 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4514 fput(mddev
->bitmap_info
.file
);
4515 mddev
->bitmap_info
.file
= NULL
;
4517 mddev
->bitmap_info
.offset
= 0;
4519 /* make sure all md_delayed_delete calls have finished */
4520 flush_scheduled_work();
4522 export_array(mddev
);
4524 mddev
->array_sectors
= 0;
4525 mddev
->external_size
= 0;
4526 mddev
->dev_sectors
= 0;
4527 mddev
->raid_disks
= 0;
4528 mddev
->recovery_cp
= 0;
4529 mddev
->resync_min
= 0;
4530 mddev
->resync_max
= MaxSector
;
4531 mddev
->reshape_position
= MaxSector
;
4532 mddev
->external
= 0;
4533 mddev
->persistent
= 0;
4534 mddev
->level
= LEVEL_NONE
;
4535 mddev
->clevel
[0] = 0;
4538 mddev
->metadata_type
[0] = 0;
4539 mddev
->chunk_sectors
= 0;
4540 mddev
->ctime
= mddev
->utime
= 0;
4542 mddev
->max_disks
= 0;
4544 mddev
->delta_disks
= 0;
4545 mddev
->new_level
= LEVEL_NONE
;
4546 mddev
->new_layout
= 0;
4547 mddev
->new_chunk_sectors
= 0;
4548 mddev
->curr_resync
= 0;
4549 mddev
->resync_mismatches
= 0;
4550 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4551 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4552 mddev
->recovery
= 0;
4555 mddev
->degraded
= 0;
4556 mddev
->barriers_work
= 0;
4557 mddev
->safemode
= 0;
4558 mddev
->bitmap_info
.offset
= 0;
4559 mddev
->bitmap_info
.default_offset
= 0;
4560 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4561 if (mddev
->hold_active
== UNTIL_STOP
)
4562 mddev
->hold_active
= 0;
4564 } else if (mddev
->pers
)
4565 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4568 blk_integrity_unregister(disk
);
4569 md_new_event(mddev
);
4570 sysfs_notify_dirent(mddev
->sysfs_state
);
4575 static void autorun_array(mddev_t
*mddev
)
4580 if (list_empty(&mddev
->disks
))
4583 printk(KERN_INFO
"md: running: ");
4585 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4586 char b
[BDEVNAME_SIZE
];
4587 printk("<%s>", bdevname(rdev
->bdev
,b
));
4591 err
= do_md_run(mddev
);
4593 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4594 do_md_stop(mddev
, 0, 0);
4599 * lets try to run arrays based on all disks that have arrived
4600 * until now. (those are in pending_raid_disks)
4602 * the method: pick the first pending disk, collect all disks with
4603 * the same UUID, remove all from the pending list and put them into
4604 * the 'same_array' list. Then order this list based on superblock
4605 * update time (freshest comes first), kick out 'old' disks and
4606 * compare superblocks. If everything's fine then run it.
4608 * If "unit" is allocated, then bump its reference count
4610 static void autorun_devices(int part
)
4612 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4614 char b
[BDEVNAME_SIZE
];
4616 printk(KERN_INFO
"md: autorun ...\n");
4617 while (!list_empty(&pending_raid_disks
)) {
4620 LIST_HEAD(candidates
);
4621 rdev0
= list_entry(pending_raid_disks
.next
,
4622 mdk_rdev_t
, same_set
);
4624 printk(KERN_INFO
"md: considering %s ...\n",
4625 bdevname(rdev0
->bdev
,b
));
4626 INIT_LIST_HEAD(&candidates
);
4627 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4628 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4629 printk(KERN_INFO
"md: adding %s ...\n",
4630 bdevname(rdev
->bdev
,b
));
4631 list_move(&rdev
->same_set
, &candidates
);
4634 * now we have a set of devices, with all of them having
4635 * mostly sane superblocks. It's time to allocate the
4639 dev
= MKDEV(mdp_major
,
4640 rdev0
->preferred_minor
<< MdpMinorShift
);
4641 unit
= MINOR(dev
) >> MdpMinorShift
;
4643 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4646 if (rdev0
->preferred_minor
!= unit
) {
4647 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4648 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4652 md_probe(dev
, NULL
, NULL
);
4653 mddev
= mddev_find(dev
);
4654 if (!mddev
|| !mddev
->gendisk
) {
4658 "md: cannot allocate memory for md drive.\n");
4661 if (mddev_lock(mddev
))
4662 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4664 else if (mddev
->raid_disks
|| mddev
->major_version
4665 || !list_empty(&mddev
->disks
)) {
4667 "md: %s already running, cannot run %s\n",
4668 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4669 mddev_unlock(mddev
);
4671 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4672 mddev
->persistent
= 1;
4673 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4674 list_del_init(&rdev
->same_set
);
4675 if (bind_rdev_to_array(rdev
, mddev
))
4678 autorun_array(mddev
);
4679 mddev_unlock(mddev
);
4681 /* on success, candidates will be empty, on error
4684 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4685 list_del_init(&rdev
->same_set
);
4690 printk(KERN_INFO
"md: ... autorun DONE.\n");
4692 #endif /* !MODULE */
4694 static int get_version(void __user
* arg
)
4698 ver
.major
= MD_MAJOR_VERSION
;
4699 ver
.minor
= MD_MINOR_VERSION
;
4700 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4702 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4708 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4710 mdu_array_info_t info
;
4711 int nr
,working
,insync
,failed
,spare
;
4714 nr
=working
=insync
=failed
=spare
=0;
4715 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4717 if (test_bit(Faulty
, &rdev
->flags
))
4721 if (test_bit(In_sync
, &rdev
->flags
))
4728 info
.major_version
= mddev
->major_version
;
4729 info
.minor_version
= mddev
->minor_version
;
4730 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4731 info
.ctime
= mddev
->ctime
;
4732 info
.level
= mddev
->level
;
4733 info
.size
= mddev
->dev_sectors
/ 2;
4734 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4737 info
.raid_disks
= mddev
->raid_disks
;
4738 info
.md_minor
= mddev
->md_minor
;
4739 info
.not_persistent
= !mddev
->persistent
;
4741 info
.utime
= mddev
->utime
;
4744 info
.state
= (1<<MD_SB_CLEAN
);
4745 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
4746 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4747 info
.active_disks
= insync
;
4748 info
.working_disks
= working
;
4749 info
.failed_disks
= failed
;
4750 info
.spare_disks
= spare
;
4752 info
.layout
= mddev
->layout
;
4753 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4755 if (copy_to_user(arg
, &info
, sizeof(info
)))
4761 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4763 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4764 char *ptr
, *buf
= NULL
;
4767 if (md_allow_write(mddev
))
4768 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4770 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4775 /* bitmap disabled, zero the first byte and copy out */
4776 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4777 file
->pathname
[0] = '\0';
4781 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4785 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4789 strcpy(file
->pathname
, ptr
);
4793 if (copy_to_user(arg
, file
, sizeof(*file
)))
4801 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4803 mdu_disk_info_t info
;
4806 if (copy_from_user(&info
, arg
, sizeof(info
)))
4809 rdev
= find_rdev_nr(mddev
, info
.number
);
4811 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4812 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4813 info
.raid_disk
= rdev
->raid_disk
;
4815 if (test_bit(Faulty
, &rdev
->flags
))
4816 info
.state
|= (1<<MD_DISK_FAULTY
);
4817 else if (test_bit(In_sync
, &rdev
->flags
)) {
4818 info
.state
|= (1<<MD_DISK_ACTIVE
);
4819 info
.state
|= (1<<MD_DISK_SYNC
);
4821 if (test_bit(WriteMostly
, &rdev
->flags
))
4822 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4824 info
.major
= info
.minor
= 0;
4825 info
.raid_disk
= -1;
4826 info
.state
= (1<<MD_DISK_REMOVED
);
4829 if (copy_to_user(arg
, &info
, sizeof(info
)))
4835 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4837 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4839 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4841 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4844 if (!mddev
->raid_disks
) {
4846 /* expecting a device which has a superblock */
4847 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4850 "md: md_import_device returned %ld\n",
4852 return PTR_ERR(rdev
);
4854 if (!list_empty(&mddev
->disks
)) {
4855 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4856 mdk_rdev_t
, same_set
);
4857 err
= super_types
[mddev
->major_version
]
4858 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4861 "md: %s has different UUID to %s\n",
4862 bdevname(rdev
->bdev
,b
),
4863 bdevname(rdev0
->bdev
,b2
));
4868 err
= bind_rdev_to_array(rdev
, mddev
);
4875 * add_new_disk can be used once the array is assembled
4876 * to add "hot spares". They must already have a superblock
4881 if (!mddev
->pers
->hot_add_disk
) {
4883 "%s: personality does not support diskops!\n",
4887 if (mddev
->persistent
)
4888 rdev
= md_import_device(dev
, mddev
->major_version
,
4889 mddev
->minor_version
);
4891 rdev
= md_import_device(dev
, -1, -1);
4894 "md: md_import_device returned %ld\n",
4896 return PTR_ERR(rdev
);
4898 /* set save_raid_disk if appropriate */
4899 if (!mddev
->persistent
) {
4900 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4901 info
->raid_disk
< mddev
->raid_disks
)
4902 rdev
->raid_disk
= info
->raid_disk
;
4904 rdev
->raid_disk
= -1;
4906 super_types
[mddev
->major_version
].
4907 validate_super(mddev
, rdev
);
4908 rdev
->saved_raid_disk
= rdev
->raid_disk
;
4910 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
4911 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4912 set_bit(WriteMostly
, &rdev
->flags
);
4914 clear_bit(WriteMostly
, &rdev
->flags
);
4916 rdev
->raid_disk
= -1;
4917 err
= bind_rdev_to_array(rdev
, mddev
);
4918 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
4919 /* If there is hot_add_disk but no hot_remove_disk
4920 * then added disks for geometry changes,
4921 * and should be added immediately.
4923 super_types
[mddev
->major_version
].
4924 validate_super(mddev
, rdev
);
4925 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
4927 unbind_rdev_from_array(rdev
);
4932 sysfs_notify_dirent(rdev
->sysfs_state
);
4934 md_update_sb(mddev
, 1);
4935 if (mddev
->degraded
)
4936 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4937 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4938 md_wakeup_thread(mddev
->thread
);
4942 /* otherwise, add_new_disk is only allowed
4943 * for major_version==0 superblocks
4945 if (mddev
->major_version
!= 0) {
4946 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
4951 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
4953 rdev
= md_import_device(dev
, -1, 0);
4956 "md: error, md_import_device() returned %ld\n",
4958 return PTR_ERR(rdev
);
4960 rdev
->desc_nr
= info
->number
;
4961 if (info
->raid_disk
< mddev
->raid_disks
)
4962 rdev
->raid_disk
= info
->raid_disk
;
4964 rdev
->raid_disk
= -1;
4966 if (rdev
->raid_disk
< mddev
->raid_disks
)
4967 if (info
->state
& (1<<MD_DISK_SYNC
))
4968 set_bit(In_sync
, &rdev
->flags
);
4970 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4971 set_bit(WriteMostly
, &rdev
->flags
);
4973 if (!mddev
->persistent
) {
4974 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
4975 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
4977 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
4978 rdev
->sectors
= rdev
->sb_start
;
4980 err
= bind_rdev_to_array(rdev
, mddev
);
4990 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
4992 char b
[BDEVNAME_SIZE
];
4995 rdev
= find_rdev(mddev
, dev
);
4999 if (rdev
->raid_disk
>= 0)
5002 kick_rdev_from_array(rdev
);
5003 md_update_sb(mddev
, 1);
5004 md_new_event(mddev
);
5008 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5009 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5013 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5015 char b
[BDEVNAME_SIZE
];
5022 if (mddev
->major_version
!= 0) {
5023 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5024 " version-0 superblocks.\n",
5028 if (!mddev
->pers
->hot_add_disk
) {
5030 "%s: personality does not support diskops!\n",
5035 rdev
= md_import_device(dev
, -1, 0);
5038 "md: error, md_import_device() returned %ld\n",
5043 if (mddev
->persistent
)
5044 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5046 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5048 rdev
->sectors
= rdev
->sb_start
;
5050 if (test_bit(Faulty
, &rdev
->flags
)) {
5052 "md: can not hot-add faulty %s disk to %s!\n",
5053 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5057 clear_bit(In_sync
, &rdev
->flags
);
5059 rdev
->saved_raid_disk
= -1;
5060 err
= bind_rdev_to_array(rdev
, mddev
);
5065 * The rest should better be atomic, we can have disk failures
5066 * noticed in interrupt contexts ...
5069 rdev
->raid_disk
= -1;
5071 md_update_sb(mddev
, 1);
5074 * Kick recovery, maybe this spare has to be added to the
5075 * array immediately.
5077 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5078 md_wakeup_thread(mddev
->thread
);
5079 md_new_event(mddev
);
5087 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5092 if (!mddev
->pers
->quiesce
)
5094 if (mddev
->recovery
|| mddev
->sync_thread
)
5096 /* we should be able to change the bitmap.. */
5102 return -EEXIST
; /* cannot add when bitmap is present */
5103 mddev
->bitmap_info
.file
= fget(fd
);
5105 if (mddev
->bitmap_info
.file
== NULL
) {
5106 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5111 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5113 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5115 fput(mddev
->bitmap_info
.file
);
5116 mddev
->bitmap_info
.file
= NULL
;
5119 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5120 } else if (mddev
->bitmap
== NULL
)
5121 return -ENOENT
; /* cannot remove what isn't there */
5124 mddev
->pers
->quiesce(mddev
, 1);
5126 err
= bitmap_create(mddev
);
5127 if (fd
< 0 || err
) {
5128 bitmap_destroy(mddev
);
5129 fd
= -1; /* make sure to put the file */
5131 mddev
->pers
->quiesce(mddev
, 0);
5134 if (mddev
->bitmap_info
.file
) {
5135 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5136 fput(mddev
->bitmap_info
.file
);
5138 mddev
->bitmap_info
.file
= NULL
;
5145 * set_array_info is used two different ways
5146 * The original usage is when creating a new array.
5147 * In this usage, raid_disks is > 0 and it together with
5148 * level, size, not_persistent,layout,chunksize determine the
5149 * shape of the array.
5150 * This will always create an array with a type-0.90.0 superblock.
5151 * The newer usage is when assembling an array.
5152 * In this case raid_disks will be 0, and the major_version field is
5153 * use to determine which style super-blocks are to be found on the devices.
5154 * The minor and patch _version numbers are also kept incase the
5155 * super_block handler wishes to interpret them.
5157 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5160 if (info
->raid_disks
== 0) {
5161 /* just setting version number for superblock loading */
5162 if (info
->major_version
< 0 ||
5163 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5164 super_types
[info
->major_version
].name
== NULL
) {
5165 /* maybe try to auto-load a module? */
5167 "md: superblock version %d not known\n",
5168 info
->major_version
);
5171 mddev
->major_version
= info
->major_version
;
5172 mddev
->minor_version
= info
->minor_version
;
5173 mddev
->patch_version
= info
->patch_version
;
5174 mddev
->persistent
= !info
->not_persistent
;
5177 mddev
->major_version
= MD_MAJOR_VERSION
;
5178 mddev
->minor_version
= MD_MINOR_VERSION
;
5179 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5180 mddev
->ctime
= get_seconds();
5182 mddev
->level
= info
->level
;
5183 mddev
->clevel
[0] = 0;
5184 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5185 mddev
->raid_disks
= info
->raid_disks
;
5186 /* don't set md_minor, it is determined by which /dev/md* was
5189 if (info
->state
& (1<<MD_SB_CLEAN
))
5190 mddev
->recovery_cp
= MaxSector
;
5192 mddev
->recovery_cp
= 0;
5193 mddev
->persistent
= ! info
->not_persistent
;
5194 mddev
->external
= 0;
5196 mddev
->layout
= info
->layout
;
5197 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5199 mddev
->max_disks
= MD_SB_DISKS
;
5201 if (mddev
->persistent
)
5203 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5205 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5206 mddev
->bitmap_info
.offset
= 0;
5208 mddev
->reshape_position
= MaxSector
;
5211 * Generate a 128 bit UUID
5213 get_random_bytes(mddev
->uuid
, 16);
5215 mddev
->new_level
= mddev
->level
;
5216 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5217 mddev
->new_layout
= mddev
->layout
;
5218 mddev
->delta_disks
= 0;
5223 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5225 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5227 if (mddev
->external_size
)
5230 mddev
->array_sectors
= array_sectors
;
5232 EXPORT_SYMBOL(md_set_array_sectors
);
5234 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5238 int fit
= (num_sectors
== 0);
5240 if (mddev
->pers
->resize
== NULL
)
5242 /* The "num_sectors" is the number of sectors of each device that
5243 * is used. This can only make sense for arrays with redundancy.
5244 * linear and raid0 always use whatever space is available. We can only
5245 * consider changing this number if no resync or reconstruction is
5246 * happening, and if the new size is acceptable. It must fit before the
5247 * sb_start or, if that is <data_offset, it must fit before the size
5248 * of each device. If num_sectors is zero, we find the largest size
5252 if (mddev
->sync_thread
)
5255 /* Sorry, cannot grow a bitmap yet, just remove it,
5259 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5260 sector_t avail
= rdev
->sectors
;
5262 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5263 num_sectors
= avail
;
5264 if (avail
< num_sectors
)
5267 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5269 revalidate_disk(mddev
->gendisk
);
5273 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5276 /* change the number of raid disks */
5277 if (mddev
->pers
->check_reshape
== NULL
)
5279 if (raid_disks
<= 0 ||
5280 raid_disks
>= mddev
->max_disks
)
5282 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5284 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5286 rv
= mddev
->pers
->check_reshape(mddev
);
5292 * update_array_info is used to change the configuration of an
5294 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5295 * fields in the info are checked against the array.
5296 * Any differences that cannot be handled will cause an error.
5297 * Normally, only one change can be managed at a time.
5299 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5305 /* calculate expected state,ignoring low bits */
5306 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5307 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5309 if (mddev
->major_version
!= info
->major_version
||
5310 mddev
->minor_version
!= info
->minor_version
||
5311 /* mddev->patch_version != info->patch_version || */
5312 mddev
->ctime
!= info
->ctime
||
5313 mddev
->level
!= info
->level
||
5314 /* mddev->layout != info->layout || */
5315 !mddev
->persistent
!= info
->not_persistent
||
5316 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5317 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5318 ((state
^info
->state
) & 0xfffffe00)
5321 /* Check there is only one change */
5322 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5324 if (mddev
->raid_disks
!= info
->raid_disks
)
5326 if (mddev
->layout
!= info
->layout
)
5328 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5335 if (mddev
->layout
!= info
->layout
) {
5337 * we don't need to do anything at the md level, the
5338 * personality will take care of it all.
5340 if (mddev
->pers
->check_reshape
== NULL
)
5343 mddev
->new_layout
= info
->layout
;
5344 rv
= mddev
->pers
->check_reshape(mddev
);
5346 mddev
->new_layout
= mddev
->layout
;
5350 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5351 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5353 if (mddev
->raid_disks
!= info
->raid_disks
)
5354 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5356 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5357 if (mddev
->pers
->quiesce
== NULL
)
5359 if (mddev
->recovery
|| mddev
->sync_thread
)
5361 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5362 /* add the bitmap */
5365 if (mddev
->bitmap_info
.default_offset
== 0)
5367 mddev
->bitmap_info
.offset
=
5368 mddev
->bitmap_info
.default_offset
;
5369 mddev
->pers
->quiesce(mddev
, 1);
5370 rv
= bitmap_create(mddev
);
5372 bitmap_destroy(mddev
);
5373 mddev
->pers
->quiesce(mddev
, 0);
5375 /* remove the bitmap */
5378 if (mddev
->bitmap
->file
)
5380 mddev
->pers
->quiesce(mddev
, 1);
5381 bitmap_destroy(mddev
);
5382 mddev
->pers
->quiesce(mddev
, 0);
5383 mddev
->bitmap_info
.offset
= 0;
5386 md_update_sb(mddev
, 1);
5390 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5394 if (mddev
->pers
== NULL
)
5397 rdev
= find_rdev(mddev
, dev
);
5401 md_error(mddev
, rdev
);
5406 * We have a problem here : there is no easy way to give a CHS
5407 * virtual geometry. We currently pretend that we have a 2 heads
5408 * 4 sectors (with a BIG number of cylinders...). This drives
5409 * dosfs just mad... ;-)
5411 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5413 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5417 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
5421 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5422 unsigned int cmd
, unsigned long arg
)
5425 void __user
*argp
= (void __user
*)arg
;
5426 mddev_t
*mddev
= NULL
;
5428 if (!capable(CAP_SYS_ADMIN
))
5432 * Commands dealing with the RAID driver but not any
5438 err
= get_version(argp
);
5441 case PRINT_RAID_DEBUG
:
5449 autostart_arrays(arg
);
5456 * Commands creating/starting a new array:
5459 mddev
= bdev
->bd_disk
->private_data
;
5466 err
= mddev_lock(mddev
);
5469 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5476 case SET_ARRAY_INFO
:
5478 mdu_array_info_t info
;
5480 memset(&info
, 0, sizeof(info
));
5481 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5486 err
= update_array_info(mddev
, &info
);
5488 printk(KERN_WARNING
"md: couldn't update"
5489 " array info. %d\n", err
);
5494 if (!list_empty(&mddev
->disks
)) {
5496 "md: array %s already has disks!\n",
5501 if (mddev
->raid_disks
) {
5503 "md: array %s already initialised!\n",
5508 err
= set_array_info(mddev
, &info
);
5510 printk(KERN_WARNING
"md: couldn't set"
5511 " array info. %d\n", err
);
5521 * Commands querying/configuring an existing array:
5523 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5524 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5525 if ((!mddev
->raid_disks
&& !mddev
->external
)
5526 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5527 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5528 && cmd
!= GET_BITMAP_FILE
) {
5534 * Commands even a read-only array can execute:
5538 case GET_ARRAY_INFO
:
5539 err
= get_array_info(mddev
, argp
);
5542 case GET_BITMAP_FILE
:
5543 err
= get_bitmap_file(mddev
, argp
);
5547 err
= get_disk_info(mddev
, argp
);
5550 case RESTART_ARRAY_RW
:
5551 err
= restart_array(mddev
);
5555 err
= do_md_stop(mddev
, 0, 1);
5559 err
= do_md_stop(mddev
, 1, 1);
5565 * The remaining ioctls are changing the state of the
5566 * superblock, so we do not allow them on read-only arrays.
5567 * However non-MD ioctls (e.g. get-size) will still come through
5568 * here and hit the 'default' below, so only disallow
5569 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5571 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5572 if (mddev
->ro
== 2) {
5574 sysfs_notify_dirent(mddev
->sysfs_state
);
5575 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5576 md_wakeup_thread(mddev
->thread
);
5587 mdu_disk_info_t info
;
5588 if (copy_from_user(&info
, argp
, sizeof(info
)))
5591 err
= add_new_disk(mddev
, &info
);
5595 case HOT_REMOVE_DISK
:
5596 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5600 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5603 case SET_DISK_FAULTY
:
5604 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5608 err
= do_md_run(mddev
);
5611 case SET_BITMAP_FILE
:
5612 err
= set_bitmap_file(mddev
, (int)arg
);
5622 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5624 mddev
->hold_active
= 0;
5625 mddev_unlock(mddev
);
5635 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5638 * Succeed if we can lock the mddev, which confirms that
5639 * it isn't being stopped right now.
5641 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5644 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5645 /* we are racing with mddev_put which is discarding this
5649 /* Wait until bdev->bd_disk is definitely gone */
5650 flush_scheduled_work();
5651 /* Then retry the open from the top */
5652 return -ERESTARTSYS
;
5654 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5656 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5660 atomic_inc(&mddev
->openers
);
5661 mutex_unlock(&mddev
->open_mutex
);
5663 check_disk_change(bdev
);
5668 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5670 mddev_t
*mddev
= disk
->private_data
;
5673 atomic_dec(&mddev
->openers
);
5679 static int md_media_changed(struct gendisk
*disk
)
5681 mddev_t
*mddev
= disk
->private_data
;
5683 return mddev
->changed
;
5686 static int md_revalidate(struct gendisk
*disk
)
5688 mddev_t
*mddev
= disk
->private_data
;
5693 static const struct block_device_operations md_fops
=
5695 .owner
= THIS_MODULE
,
5697 .release
= md_release
,
5699 .getgeo
= md_getgeo
,
5700 .media_changed
= md_media_changed
,
5701 .revalidate_disk
= md_revalidate
,
5704 static int md_thread(void * arg
)
5706 mdk_thread_t
*thread
= arg
;
5709 * md_thread is a 'system-thread', it's priority should be very
5710 * high. We avoid resource deadlocks individually in each
5711 * raid personality. (RAID5 does preallocation) We also use RR and
5712 * the very same RT priority as kswapd, thus we will never get
5713 * into a priority inversion deadlock.
5715 * we definitely have to have equal or higher priority than
5716 * bdflush, otherwise bdflush will deadlock if there are too
5717 * many dirty RAID5 blocks.
5720 allow_signal(SIGKILL
);
5721 while (!kthread_should_stop()) {
5723 /* We need to wait INTERRUPTIBLE so that
5724 * we don't add to the load-average.
5725 * That means we need to be sure no signals are
5728 if (signal_pending(current
))
5729 flush_signals(current
);
5731 wait_event_interruptible_timeout
5733 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5734 || kthread_should_stop(),
5737 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5739 thread
->run(thread
->mddev
);
5745 void md_wakeup_thread(mdk_thread_t
*thread
)
5748 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5749 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5750 wake_up(&thread
->wqueue
);
5754 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5757 mdk_thread_t
*thread
;
5759 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5763 init_waitqueue_head(&thread
->wqueue
);
5766 thread
->mddev
= mddev
;
5767 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5768 thread
->tsk
= kthread_run(md_thread
, thread
,
5770 mdname(thread
->mddev
),
5771 name
?: mddev
->pers
->name
);
5772 if (IS_ERR(thread
->tsk
)) {
5779 void md_unregister_thread(mdk_thread_t
*thread
)
5783 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5785 kthread_stop(thread
->tsk
);
5789 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5796 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5799 if (mddev
->external
)
5800 set_bit(Blocked
, &rdev
->flags
);
5802 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5804 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5805 __builtin_return_address(0),__builtin_return_address(1),
5806 __builtin_return_address(2),__builtin_return_address(3));
5810 if (!mddev
->pers
->error_handler
)
5812 mddev
->pers
->error_handler(mddev
,rdev
);
5813 if (mddev
->degraded
)
5814 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5815 set_bit(StateChanged
, &rdev
->flags
);
5816 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5817 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5818 md_wakeup_thread(mddev
->thread
);
5819 md_new_event_inintr(mddev
);
5822 /* seq_file implementation /proc/mdstat */
5824 static void status_unused(struct seq_file
*seq
)
5829 seq_printf(seq
, "unused devices: ");
5831 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
5832 char b
[BDEVNAME_SIZE
];
5834 seq_printf(seq
, "%s ",
5835 bdevname(rdev
->bdev
,b
));
5838 seq_printf(seq
, "<none>");
5840 seq_printf(seq
, "\n");
5844 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
5846 sector_t max_sectors
, resync
, res
;
5847 unsigned long dt
, db
;
5850 unsigned int per_milli
;
5852 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
5854 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5855 max_sectors
= mddev
->resync_max_sectors
;
5857 max_sectors
= mddev
->dev_sectors
;
5860 * Should not happen.
5866 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5867 * in a sector_t, and (max_sectors>>scale) will fit in a
5868 * u32, as those are the requirements for sector_div.
5869 * Thus 'scale' must be at least 10
5872 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5873 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
5876 res
= (resync
>>scale
)*1000;
5877 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
5881 int i
, x
= per_milli
/50, y
= 20-x
;
5882 seq_printf(seq
, "[");
5883 for (i
= 0; i
< x
; i
++)
5884 seq_printf(seq
, "=");
5885 seq_printf(seq
, ">");
5886 for (i
= 0; i
< y
; i
++)
5887 seq_printf(seq
, ".");
5888 seq_printf(seq
, "] ");
5890 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
5891 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
5893 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
5895 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
5896 "resync" : "recovery"))),
5897 per_milli
/10, per_milli
% 10,
5898 (unsigned long long) resync
/2,
5899 (unsigned long long) max_sectors
/2);
5902 * dt: time from mark until now
5903 * db: blocks written from mark until now
5904 * rt: remaining time
5906 * rt is a sector_t, so could be 32bit or 64bit.
5907 * So we divide before multiply in case it is 32bit and close
5909 * We scale the divisor (db) by 32 to avoid loosing precision
5910 * near the end of resync when the number of remaining sectors
5912 * We then divide rt by 32 after multiplying by db to compensate.
5913 * The '+1' avoids division by zero if db is very small.
5915 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
5917 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
5918 - mddev
->resync_mark_cnt
;
5920 rt
= max_sectors
- resync
; /* number of remaining sectors */
5921 sector_div(rt
, db
/32+1);
5925 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
5926 ((unsigned long)rt
% 60)/6);
5928 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
5931 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
5933 struct list_head
*tmp
;
5943 spin_lock(&all_mddevs_lock
);
5944 list_for_each(tmp
,&all_mddevs
)
5946 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
5948 spin_unlock(&all_mddevs_lock
);
5951 spin_unlock(&all_mddevs_lock
);
5953 return (void*)2;/* tail */
5957 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
5959 struct list_head
*tmp
;
5960 mddev_t
*next_mddev
, *mddev
= v
;
5966 spin_lock(&all_mddevs_lock
);
5968 tmp
= all_mddevs
.next
;
5970 tmp
= mddev
->all_mddevs
.next
;
5971 if (tmp
!= &all_mddevs
)
5972 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
5974 next_mddev
= (void*)2;
5977 spin_unlock(&all_mddevs_lock
);
5985 static void md_seq_stop(struct seq_file
*seq
, void *v
)
5989 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
5993 struct mdstat_info
{
5997 static int md_seq_show(struct seq_file
*seq
, void *v
)
6002 struct mdstat_info
*mi
= seq
->private;
6003 struct bitmap
*bitmap
;
6005 if (v
== (void*)1) {
6006 struct mdk_personality
*pers
;
6007 seq_printf(seq
, "Personalities : ");
6008 spin_lock(&pers_lock
);
6009 list_for_each_entry(pers
, &pers_list
, list
)
6010 seq_printf(seq
, "[%s] ", pers
->name
);
6012 spin_unlock(&pers_lock
);
6013 seq_printf(seq
, "\n");
6014 mi
->event
= atomic_read(&md_event_count
);
6017 if (v
== (void*)2) {
6022 if (mddev_lock(mddev
) < 0)
6025 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6026 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6027 mddev
->pers
? "" : "in");
6030 seq_printf(seq
, " (read-only)");
6032 seq_printf(seq
, " (auto-read-only)");
6033 seq_printf(seq
, " %s", mddev
->pers
->name
);
6037 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6038 char b
[BDEVNAME_SIZE
];
6039 seq_printf(seq
, " %s[%d]",
6040 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6041 if (test_bit(WriteMostly
, &rdev
->flags
))
6042 seq_printf(seq
, "(W)");
6043 if (test_bit(Faulty
, &rdev
->flags
)) {
6044 seq_printf(seq
, "(F)");
6046 } else if (rdev
->raid_disk
< 0)
6047 seq_printf(seq
, "(S)"); /* spare */
6048 sectors
+= rdev
->sectors
;
6051 if (!list_empty(&mddev
->disks
)) {
6053 seq_printf(seq
, "\n %llu blocks",
6054 (unsigned long long)
6055 mddev
->array_sectors
/ 2);
6057 seq_printf(seq
, "\n %llu blocks",
6058 (unsigned long long)sectors
/ 2);
6060 if (mddev
->persistent
) {
6061 if (mddev
->major_version
!= 0 ||
6062 mddev
->minor_version
!= 90) {
6063 seq_printf(seq
," super %d.%d",
6064 mddev
->major_version
,
6065 mddev
->minor_version
);
6067 } else if (mddev
->external
)
6068 seq_printf(seq
, " super external:%s",
6069 mddev
->metadata_type
);
6071 seq_printf(seq
, " super non-persistent");
6074 mddev
->pers
->status(seq
, mddev
);
6075 seq_printf(seq
, "\n ");
6076 if (mddev
->pers
->sync_request
) {
6077 if (mddev
->curr_resync
> 2) {
6078 status_resync(seq
, mddev
);
6079 seq_printf(seq
, "\n ");
6080 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6081 seq_printf(seq
, "\tresync=DELAYED\n ");
6082 else if (mddev
->recovery_cp
< MaxSector
)
6083 seq_printf(seq
, "\tresync=PENDING\n ");
6086 seq_printf(seq
, "\n ");
6088 if ((bitmap
= mddev
->bitmap
)) {
6089 unsigned long chunk_kb
;
6090 unsigned long flags
;
6091 spin_lock_irqsave(&bitmap
->lock
, flags
);
6092 chunk_kb
= bitmap
->chunksize
>> 10;
6093 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6095 bitmap
->pages
- bitmap
->missing_pages
,
6097 (bitmap
->pages
- bitmap
->missing_pages
)
6098 << (PAGE_SHIFT
- 10),
6099 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
6100 chunk_kb
? "KB" : "B");
6102 seq_printf(seq
, ", file: ");
6103 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6106 seq_printf(seq
, "\n");
6107 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6110 seq_printf(seq
, "\n");
6112 mddev_unlock(mddev
);
6117 static const struct seq_operations md_seq_ops
= {
6118 .start
= md_seq_start
,
6119 .next
= md_seq_next
,
6120 .stop
= md_seq_stop
,
6121 .show
= md_seq_show
,
6124 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6127 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6131 error
= seq_open(file
, &md_seq_ops
);
6135 struct seq_file
*p
= file
->private_data
;
6137 mi
->event
= atomic_read(&md_event_count
);
6142 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6144 struct seq_file
*m
= filp
->private_data
;
6145 struct mdstat_info
*mi
= m
->private;
6148 poll_wait(filp
, &md_event_waiters
, wait
);
6150 /* always allow read */
6151 mask
= POLLIN
| POLLRDNORM
;
6153 if (mi
->event
!= atomic_read(&md_event_count
))
6154 mask
|= POLLERR
| POLLPRI
;
6158 static const struct file_operations md_seq_fops
= {
6159 .owner
= THIS_MODULE
,
6160 .open
= md_seq_open
,
6162 .llseek
= seq_lseek
,
6163 .release
= seq_release_private
,
6164 .poll
= mdstat_poll
,
6167 int register_md_personality(struct mdk_personality
*p
)
6169 spin_lock(&pers_lock
);
6170 list_add_tail(&p
->list
, &pers_list
);
6171 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6172 spin_unlock(&pers_lock
);
6176 int unregister_md_personality(struct mdk_personality
*p
)
6178 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6179 spin_lock(&pers_lock
);
6180 list_del_init(&p
->list
);
6181 spin_unlock(&pers_lock
);
6185 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6193 rdev_for_each_rcu(rdev
, mddev
) {
6194 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6195 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6196 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6197 atomic_read(&disk
->sync_io
);
6198 /* sync IO will cause sync_io to increase before the disk_stats
6199 * as sync_io is counted when a request starts, and
6200 * disk_stats is counted when it completes.
6201 * So resync activity will cause curr_events to be smaller than
6202 * when there was no such activity.
6203 * non-sync IO will cause disk_stat to increase without
6204 * increasing sync_io so curr_events will (eventually)
6205 * be larger than it was before. Once it becomes
6206 * substantially larger, the test below will cause
6207 * the array to appear non-idle, and resync will slow
6209 * If there is a lot of outstanding resync activity when
6210 * we set last_event to curr_events, then all that activity
6211 * completing might cause the array to appear non-idle
6212 * and resync will be slowed down even though there might
6213 * not have been non-resync activity. This will only
6214 * happen once though. 'last_events' will soon reflect
6215 * the state where there is little or no outstanding
6216 * resync requests, and further resync activity will
6217 * always make curr_events less than last_events.
6220 if (init
|| curr_events
- rdev
->last_events
> 64) {
6221 rdev
->last_events
= curr_events
;
6229 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6231 /* another "blocks" (512byte) blocks have been synced */
6232 atomic_sub(blocks
, &mddev
->recovery_active
);
6233 wake_up(&mddev
->recovery_wait
);
6235 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6236 md_wakeup_thread(mddev
->thread
);
6237 // stop recovery, signal do_sync ....
6242 /* md_write_start(mddev, bi)
6243 * If we need to update some array metadata (e.g. 'active' flag
6244 * in superblock) before writing, schedule a superblock update
6245 * and wait for it to complete.
6247 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6250 if (bio_data_dir(bi
) != WRITE
)
6253 BUG_ON(mddev
->ro
== 1);
6254 if (mddev
->ro
== 2) {
6255 /* need to switch to read/write */
6257 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6258 md_wakeup_thread(mddev
->thread
);
6259 md_wakeup_thread(mddev
->sync_thread
);
6262 atomic_inc(&mddev
->writes_pending
);
6263 if (mddev
->safemode
== 1)
6264 mddev
->safemode
= 0;
6265 if (mddev
->in_sync
) {
6266 spin_lock_irq(&mddev
->write_lock
);
6267 if (mddev
->in_sync
) {
6269 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6270 md_wakeup_thread(mddev
->thread
);
6273 spin_unlock_irq(&mddev
->write_lock
);
6276 sysfs_notify_dirent(mddev
->sysfs_state
);
6277 wait_event(mddev
->sb_wait
,
6278 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6279 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6282 void md_write_end(mddev_t
*mddev
)
6284 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6285 if (mddev
->safemode
== 2)
6286 md_wakeup_thread(mddev
->thread
);
6287 else if (mddev
->safemode_delay
)
6288 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6292 /* md_allow_write(mddev)
6293 * Calling this ensures that the array is marked 'active' so that writes
6294 * may proceed without blocking. It is important to call this before
6295 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6296 * Must be called with mddev_lock held.
6298 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6299 * is dropped, so return -EAGAIN after notifying userspace.
6301 int md_allow_write(mddev_t
*mddev
)
6307 if (!mddev
->pers
->sync_request
)
6310 spin_lock_irq(&mddev
->write_lock
);
6311 if (mddev
->in_sync
) {
6313 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6314 if (mddev
->safemode_delay
&&
6315 mddev
->safemode
== 0)
6316 mddev
->safemode
= 1;
6317 spin_unlock_irq(&mddev
->write_lock
);
6318 md_update_sb(mddev
, 0);
6319 sysfs_notify_dirent(mddev
->sysfs_state
);
6321 spin_unlock_irq(&mddev
->write_lock
);
6323 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6328 EXPORT_SYMBOL_GPL(md_allow_write
);
6330 #define SYNC_MARKS 10
6331 #define SYNC_MARK_STEP (3*HZ)
6332 void md_do_sync(mddev_t
*mddev
)
6335 unsigned int currspeed
= 0,
6337 sector_t max_sectors
,j
, io_sectors
;
6338 unsigned long mark
[SYNC_MARKS
];
6339 sector_t mark_cnt
[SYNC_MARKS
];
6341 struct list_head
*tmp
;
6342 sector_t last_check
;
6347 /* just incase thread restarts... */
6348 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6350 if (mddev
->ro
) /* never try to sync a read-only array */
6353 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6354 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6355 desc
= "data-check";
6356 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6357 desc
= "requested-resync";
6360 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6365 /* we overload curr_resync somewhat here.
6366 * 0 == not engaged in resync at all
6367 * 2 == checking that there is no conflict with another sync
6368 * 1 == like 2, but have yielded to allow conflicting resync to
6370 * other == active in resync - this many blocks
6372 * Before starting a resync we must have set curr_resync to
6373 * 2, and then checked that every "conflicting" array has curr_resync
6374 * less than ours. When we find one that is the same or higher
6375 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6376 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6377 * This will mean we have to start checking from the beginning again.
6382 mddev
->curr_resync
= 2;
6385 if (kthread_should_stop()) {
6386 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6389 for_each_mddev(mddev2
, tmp
) {
6390 if (mddev2
== mddev
)
6392 if (!mddev
->parallel_resync
6393 && mddev2
->curr_resync
6394 && match_mddev_units(mddev
, mddev2
)) {
6396 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6397 /* arbitrarily yield */
6398 mddev
->curr_resync
= 1;
6399 wake_up(&resync_wait
);
6401 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6402 /* no need to wait here, we can wait the next
6403 * time 'round when curr_resync == 2
6406 /* We need to wait 'interruptible' so as not to
6407 * contribute to the load average, and not to
6408 * be caught by 'softlockup'
6410 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6411 if (!kthread_should_stop() &&
6412 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6413 printk(KERN_INFO
"md: delaying %s of %s"
6414 " until %s has finished (they"
6415 " share one or more physical units)\n",
6416 desc
, mdname(mddev
), mdname(mddev2
));
6418 if (signal_pending(current
))
6419 flush_signals(current
);
6421 finish_wait(&resync_wait
, &wq
);
6424 finish_wait(&resync_wait
, &wq
);
6427 } while (mddev
->curr_resync
< 2);
6430 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6431 /* resync follows the size requested by the personality,
6432 * which defaults to physical size, but can be virtual size
6434 max_sectors
= mddev
->resync_max_sectors
;
6435 mddev
->resync_mismatches
= 0;
6436 /* we don't use the checkpoint if there's a bitmap */
6437 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6438 j
= mddev
->resync_min
;
6439 else if (!mddev
->bitmap
)
6440 j
= mddev
->recovery_cp
;
6442 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6443 max_sectors
= mddev
->dev_sectors
;
6445 /* recovery follows the physical size of devices */
6446 max_sectors
= mddev
->dev_sectors
;
6448 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6449 if (rdev
->raid_disk
>= 0 &&
6450 !test_bit(Faulty
, &rdev
->flags
) &&
6451 !test_bit(In_sync
, &rdev
->flags
) &&
6452 rdev
->recovery_offset
< j
)
6453 j
= rdev
->recovery_offset
;
6456 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6457 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6458 " %d KB/sec/disk.\n", speed_min(mddev
));
6459 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6460 "(but not more than %d KB/sec) for %s.\n",
6461 speed_max(mddev
), desc
);
6463 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6466 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6468 mark_cnt
[m
] = io_sectors
;
6471 mddev
->resync_mark
= mark
[last_mark
];
6472 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6475 * Tune reconstruction:
6477 window
= 32*(PAGE_SIZE
/512);
6478 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6479 window
/2,(unsigned long long) max_sectors
/2);
6481 atomic_set(&mddev
->recovery_active
, 0);
6486 "md: resuming %s of %s from checkpoint.\n",
6487 desc
, mdname(mddev
));
6488 mddev
->curr_resync
= j
;
6490 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6492 while (j
< max_sectors
) {
6497 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6498 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6499 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6500 > (max_sectors
>> 4)) ||
6501 (j
- mddev
->curr_resync_completed
)*2
6502 >= mddev
->resync_max
- mddev
->curr_resync_completed
6504 /* time to update curr_resync_completed */
6505 blk_unplug(mddev
->queue
);
6506 wait_event(mddev
->recovery_wait
,
6507 atomic_read(&mddev
->recovery_active
) == 0);
6508 mddev
->curr_resync_completed
=
6510 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6511 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6514 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6515 /* As this condition is controlled by user-space,
6516 * we can block indefinitely, so use '_interruptible'
6517 * to avoid triggering warnings.
6519 flush_signals(current
); /* just in case */
6520 wait_event_interruptible(mddev
->recovery_wait
,
6521 mddev
->resync_max
> j
6522 || kthread_should_stop());
6525 if (kthread_should_stop())
6528 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6529 currspeed
< speed_min(mddev
));
6531 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6535 if (!skipped
) { /* actual IO requested */
6536 io_sectors
+= sectors
;
6537 atomic_add(sectors
, &mddev
->recovery_active
);
6541 if (j
>1) mddev
->curr_resync
= j
;
6542 mddev
->curr_mark_cnt
= io_sectors
;
6543 if (last_check
== 0)
6544 /* this is the earliers that rebuilt will be
6545 * visible in /proc/mdstat
6547 md_new_event(mddev
);
6549 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6552 last_check
= io_sectors
;
6554 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6558 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6560 int next
= (last_mark
+1) % SYNC_MARKS
;
6562 mddev
->resync_mark
= mark
[next
];
6563 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6564 mark
[next
] = jiffies
;
6565 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6570 if (kthread_should_stop())
6575 * this loop exits only if either when we are slower than
6576 * the 'hard' speed limit, or the system was IO-idle for
6578 * the system might be non-idle CPU-wise, but we only care
6579 * about not overloading the IO subsystem. (things like an
6580 * e2fsck being done on the RAID array should execute fast)
6582 blk_unplug(mddev
->queue
);
6585 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6586 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6588 if (currspeed
> speed_min(mddev
)) {
6589 if ((currspeed
> speed_max(mddev
)) ||
6590 !is_mddev_idle(mddev
, 0)) {
6596 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6598 * this also signals 'finished resyncing' to md_stop
6601 blk_unplug(mddev
->queue
);
6603 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6605 /* tell personality that we are finished */
6606 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6608 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6609 mddev
->curr_resync
> 2) {
6610 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6611 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6612 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6614 "md: checkpointing %s of %s.\n",
6615 desc
, mdname(mddev
));
6616 mddev
->recovery_cp
= mddev
->curr_resync
;
6619 mddev
->recovery_cp
= MaxSector
;
6621 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6622 mddev
->curr_resync
= MaxSector
;
6623 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6624 if (rdev
->raid_disk
>= 0 &&
6625 !test_bit(Faulty
, &rdev
->flags
) &&
6626 !test_bit(In_sync
, &rdev
->flags
) &&
6627 rdev
->recovery_offset
< mddev
->curr_resync
)
6628 rdev
->recovery_offset
= mddev
->curr_resync
;
6631 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6634 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6635 /* We completed so min/max setting can be forgotten if used. */
6636 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6637 mddev
->resync_min
= 0;
6638 mddev
->resync_max
= MaxSector
;
6639 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6640 mddev
->resync_min
= mddev
->curr_resync_completed
;
6641 mddev
->curr_resync
= 0;
6642 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6643 mddev
->curr_resync_completed
= 0;
6644 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6645 wake_up(&resync_wait
);
6646 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6647 md_wakeup_thread(mddev
->thread
);
6652 * got a signal, exit.
6655 "md: md_do_sync() got signal ... exiting\n");
6656 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6660 EXPORT_SYMBOL_GPL(md_do_sync
);
6663 static int remove_and_add_spares(mddev_t
*mddev
)
6668 mddev
->curr_resync_completed
= 0;
6670 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6671 if (rdev
->raid_disk
>= 0 &&
6672 !test_bit(Blocked
, &rdev
->flags
) &&
6673 (test_bit(Faulty
, &rdev
->flags
) ||
6674 ! test_bit(In_sync
, &rdev
->flags
)) &&
6675 atomic_read(&rdev
->nr_pending
)==0) {
6676 if (mddev
->pers
->hot_remove_disk(
6677 mddev
, rdev
->raid_disk
)==0) {
6679 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6680 sysfs_remove_link(&mddev
->kobj
, nm
);
6681 rdev
->raid_disk
= -1;
6685 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6686 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6687 if (rdev
->raid_disk
>= 0 &&
6688 !test_bit(In_sync
, &rdev
->flags
) &&
6689 !test_bit(Blocked
, &rdev
->flags
))
6691 if (rdev
->raid_disk
< 0
6692 && !test_bit(Faulty
, &rdev
->flags
)) {
6693 rdev
->recovery_offset
= 0;
6695 hot_add_disk(mddev
, rdev
) == 0) {
6697 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6698 if (sysfs_create_link(&mddev
->kobj
,
6701 "md: cannot register "
6705 md_new_event(mddev
);
6714 * This routine is regularly called by all per-raid-array threads to
6715 * deal with generic issues like resync and super-block update.
6716 * Raid personalities that don't have a thread (linear/raid0) do not
6717 * need this as they never do any recovery or update the superblock.
6719 * It does not do any resync itself, but rather "forks" off other threads
6720 * to do that as needed.
6721 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6722 * "->recovery" and create a thread at ->sync_thread.
6723 * When the thread finishes it sets MD_RECOVERY_DONE
6724 * and wakeups up this thread which will reap the thread and finish up.
6725 * This thread also removes any faulty devices (with nr_pending == 0).
6727 * The overall approach is:
6728 * 1/ if the superblock needs updating, update it.
6729 * 2/ If a recovery thread is running, don't do anything else.
6730 * 3/ If recovery has finished, clean up, possibly marking spares active.
6731 * 4/ If there are any faulty devices, remove them.
6732 * 5/ If array is degraded, try to add spares devices
6733 * 6/ If array has spares or is not in-sync, start a resync thread.
6735 void md_check_recovery(mddev_t
*mddev
)
6741 bitmap_daemon_work(mddev
);
6746 if (signal_pending(current
)) {
6747 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6748 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6750 mddev
->safemode
= 2;
6752 flush_signals(current
);
6755 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6758 (mddev
->flags
&& !mddev
->external
) ||
6759 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6760 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6761 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6762 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6763 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6767 if (mddev_trylock(mddev
)) {
6771 /* Only thing we do on a ro array is remove
6774 remove_and_add_spares(mddev
);
6775 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6779 if (!mddev
->external
) {
6781 spin_lock_irq(&mddev
->write_lock
);
6782 if (mddev
->safemode
&&
6783 !atomic_read(&mddev
->writes_pending
) &&
6785 mddev
->recovery_cp
== MaxSector
) {
6788 if (mddev
->persistent
)
6789 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6791 if (mddev
->safemode
== 1)
6792 mddev
->safemode
= 0;
6793 spin_unlock_irq(&mddev
->write_lock
);
6795 sysfs_notify_dirent(mddev
->sysfs_state
);
6799 md_update_sb(mddev
, 0);
6801 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6802 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6803 sysfs_notify_dirent(rdev
->sysfs_state
);
6806 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6807 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6808 /* resync/recovery still happening */
6809 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6812 if (mddev
->sync_thread
) {
6813 /* resync has finished, collect result */
6814 md_unregister_thread(mddev
->sync_thread
);
6815 mddev
->sync_thread
= NULL
;
6816 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6817 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6819 /* activate any spares */
6820 if (mddev
->pers
->spare_active(mddev
))
6821 sysfs_notify(&mddev
->kobj
, NULL
,
6824 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6825 mddev
->pers
->finish_reshape
)
6826 mddev
->pers
->finish_reshape(mddev
);
6827 md_update_sb(mddev
, 1);
6829 /* if array is no-longer degraded, then any saved_raid_disk
6830 * information must be scrapped
6832 if (!mddev
->degraded
)
6833 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6834 rdev
->saved_raid_disk
= -1;
6836 mddev
->recovery
= 0;
6837 /* flag recovery needed just to double check */
6838 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6839 sysfs_notify_dirent(mddev
->sysfs_action
);
6840 md_new_event(mddev
);
6843 /* Set RUNNING before clearing NEEDED to avoid
6844 * any transients in the value of "sync_action".
6846 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6847 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6848 /* Clear some bits that don't mean anything, but
6851 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6852 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6854 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
6856 /* no recovery is running.
6857 * remove any failed drives, then
6858 * add spares if possible.
6859 * Spare are also removed and re-added, to allow
6860 * the personality to fail the re-add.
6863 if (mddev
->reshape_position
!= MaxSector
) {
6864 if (mddev
->pers
->check_reshape
== NULL
||
6865 mddev
->pers
->check_reshape(mddev
) != 0)
6866 /* Cannot proceed */
6868 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
6869 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6870 } else if ((spares
= remove_and_add_spares(mddev
))) {
6871 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6872 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
6873 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
6874 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6875 } else if (mddev
->recovery_cp
< MaxSector
) {
6876 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6877 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6878 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6879 /* nothing to be done ... */
6882 if (mddev
->pers
->sync_request
) {
6883 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
6884 /* We are adding a device or devices to an array
6885 * which has the bitmap stored on all devices.
6886 * So make sure all bitmap pages get written
6888 bitmap_write_all(mddev
->bitmap
);
6890 mddev
->sync_thread
= md_register_thread(md_do_sync
,
6893 if (!mddev
->sync_thread
) {
6894 printk(KERN_ERR
"%s: could not start resync"
6897 /* leave the spares where they are, it shouldn't hurt */
6898 mddev
->recovery
= 0;
6900 md_wakeup_thread(mddev
->sync_thread
);
6901 sysfs_notify_dirent(mddev
->sysfs_action
);
6902 md_new_event(mddev
);
6905 if (!mddev
->sync_thread
) {
6906 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6907 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
6909 if (mddev
->sysfs_action
)
6910 sysfs_notify_dirent(mddev
->sysfs_action
);
6912 mddev_unlock(mddev
);
6916 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
6918 sysfs_notify_dirent(rdev
->sysfs_state
);
6919 wait_event_timeout(rdev
->blocked_wait
,
6920 !test_bit(Blocked
, &rdev
->flags
),
6921 msecs_to_jiffies(5000));
6922 rdev_dec_pending(rdev
, mddev
);
6924 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
6926 static int md_notify_reboot(struct notifier_block
*this,
6927 unsigned long code
, void *x
)
6929 struct list_head
*tmp
;
6932 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
6934 printk(KERN_INFO
"md: stopping all md devices.\n");
6936 for_each_mddev(mddev
, tmp
)
6937 if (mddev_trylock(mddev
)) {
6938 /* Force a switch to readonly even array
6939 * appears to still be in use. Hence
6942 do_md_stop(mddev
, 1, 100);
6943 mddev_unlock(mddev
);
6946 * certain more exotic SCSI devices are known to be
6947 * volatile wrt too early system reboots. While the
6948 * right place to handle this issue is the given
6949 * driver, we do want to have a safe RAID driver ...
6956 static struct notifier_block md_notifier
= {
6957 .notifier_call
= md_notify_reboot
,
6959 .priority
= INT_MAX
, /* before any real devices */
6962 static void md_geninit(void)
6964 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
6966 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
6969 static int __init
md_init(void)
6971 if (register_blkdev(MD_MAJOR
, "md"))
6973 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
6974 unregister_blkdev(MD_MAJOR
, "md");
6977 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6978 md_probe
, NULL
, NULL
);
6979 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6980 md_probe
, NULL
, NULL
);
6982 register_reboot_notifier(&md_notifier
);
6983 raid_table_header
= register_sysctl_table(raid_root_table
);
6993 * Searches all registered partitions for autorun RAID arrays
6997 static LIST_HEAD(all_detected_devices
);
6998 struct detected_devices_node
{
6999 struct list_head list
;
7003 void md_autodetect_dev(dev_t dev
)
7005 struct detected_devices_node
*node_detected_dev
;
7007 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7008 if (node_detected_dev
) {
7009 node_detected_dev
->dev
= dev
;
7010 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7012 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7013 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7018 static void autostart_arrays(int part
)
7021 struct detected_devices_node
*node_detected_dev
;
7023 int i_scanned
, i_passed
;
7028 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7030 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7032 node_detected_dev
= list_entry(all_detected_devices
.next
,
7033 struct detected_devices_node
, list
);
7034 list_del(&node_detected_dev
->list
);
7035 dev
= node_detected_dev
->dev
;
7036 kfree(node_detected_dev
);
7037 rdev
= md_import_device(dev
,0, 90);
7041 if (test_bit(Faulty
, &rdev
->flags
)) {
7045 set_bit(AutoDetected
, &rdev
->flags
);
7046 list_add(&rdev
->same_set
, &pending_raid_disks
);
7050 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7051 i_scanned
, i_passed
);
7053 autorun_devices(part
);
7056 #endif /* !MODULE */
7058 static __exit
void md_exit(void)
7061 struct list_head
*tmp
;
7063 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7064 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7066 unregister_blkdev(MD_MAJOR
,"md");
7067 unregister_blkdev(mdp_major
, "mdp");
7068 unregister_reboot_notifier(&md_notifier
);
7069 unregister_sysctl_table(raid_table_header
);
7070 remove_proc_entry("mdstat", NULL
);
7071 for_each_mddev(mddev
, tmp
) {
7072 export_array(mddev
);
7073 mddev
->hold_active
= 0;
7077 subsys_initcall(md_init
);
7078 module_exit(md_exit
)
7080 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7082 return sprintf(buffer
, "%d", start_readonly
);
7084 static int set_ro(const char *val
, struct kernel_param
*kp
)
7087 int num
= simple_strtoul(val
, &e
, 10);
7088 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7089 start_readonly
= num
;
7095 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7096 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7098 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7100 EXPORT_SYMBOL(register_md_personality
);
7101 EXPORT_SYMBOL(unregister_md_personality
);
7102 EXPORT_SYMBOL(md_error
);
7103 EXPORT_SYMBOL(md_done_sync
);
7104 EXPORT_SYMBOL(md_write_start
);
7105 EXPORT_SYMBOL(md_write_end
);
7106 EXPORT_SYMBOL(md_register_thread
);
7107 EXPORT_SYMBOL(md_unregister_thread
);
7108 EXPORT_SYMBOL(md_wakeup_thread
);
7109 EXPORT_SYMBOL(md_check_recovery
);
7110 MODULE_LICENSE("GPL");
7112 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);