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/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
60 static void autostart_arrays(int part
);
63 static LIST_HEAD(pers_list
);
64 static DEFINE_SPINLOCK(pers_lock
);
66 static void md_print_devices(void);
68 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
70 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
73 * Default number of read corrections we'll attempt on an rdev
74 * before ejecting it from the array. We divide the read error
75 * count by 2 for every hour elapsed between read errors.
77 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
91 static int sysctl_speed_limit_min
= 1000;
92 static int sysctl_speed_limit_max
= 200000;
93 static inline int speed_min(mddev_t
*mddev
)
95 return mddev
->sync_speed_min
?
96 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
99 static inline int speed_max(mddev_t
*mddev
)
101 return mddev
->sync_speed_max
?
102 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
105 static struct ctl_table_header
*raid_table_header
;
107 static ctl_table raid_table
[] = {
109 .procname
= "speed_limit_min",
110 .data
= &sysctl_speed_limit_min
,
111 .maxlen
= sizeof(int),
112 .mode
= S_IRUGO
|S_IWUSR
,
113 .proc_handler
= proc_dointvec
,
116 .procname
= "speed_limit_max",
117 .data
= &sysctl_speed_limit_max
,
118 .maxlen
= sizeof(int),
119 .mode
= S_IRUGO
|S_IWUSR
,
120 .proc_handler
= proc_dointvec
,
125 static ctl_table raid_dir_table
[] = {
129 .mode
= S_IRUGO
|S_IXUGO
,
135 static ctl_table raid_root_table
[] = {
140 .child
= raid_dir_table
,
145 static const struct block_device_operations md_fops
;
147 static int start_readonly
;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
160 static atomic_t md_event_count
;
161 void md_new_event(mddev_t
*mddev
)
163 atomic_inc(&md_event_count
);
164 wake_up(&md_event_waiters
);
166 EXPORT_SYMBOL_GPL(md_new_event
);
168 /* Alternate version that can be called from interrupts
169 * when calling sysfs_notify isn't needed.
171 static void md_new_event_inintr(mddev_t
*mddev
)
173 atomic_inc(&md_event_count
);
174 wake_up(&md_event_waiters
);
178 * Enables to iterate over all existing md arrays
179 * all_mddevs_lock protects this list.
181 static LIST_HEAD(all_mddevs
);
182 static DEFINE_SPINLOCK(all_mddevs_lock
);
186 * iterates through all used mddevs in the system.
187 * We take care to grab the all_mddevs_lock whenever navigating
188 * the list, and to always hold a refcount when unlocked.
189 * Any code which breaks out of this loop while own
190 * a reference to the current mddev and must mddev_put it.
192 #define for_each_mddev(mddev,tmp) \
194 for (({ spin_lock(&all_mddevs_lock); \
195 tmp = all_mddevs.next; \
197 ({ if (tmp != &all_mddevs) \
198 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
199 spin_unlock(&all_mddevs_lock); \
200 if (mddev) mddev_put(mddev); \
201 mddev = list_entry(tmp, mddev_t, all_mddevs); \
202 tmp != &all_mddevs;}); \
203 ({ spin_lock(&all_mddevs_lock); \
208 /* Rather than calling directly into the personality make_request function,
209 * IO requests come here first so that we can check if the device is
210 * being suspended pending a reconfiguration.
211 * We hold a refcount over the call to ->make_request. By the time that
212 * call has finished, the bio has been linked into some internal structure
213 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
217 mddev_t
*mddev
= q
->queuedata
;
219 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
224 if (mddev
->suspended
|| mddev
->barrier
) {
227 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
228 TASK_UNINTERRUPTIBLE
);
229 if (!mddev
->suspended
&& !mddev
->barrier
)
235 finish_wait(&mddev
->sb_wait
, &__wait
);
237 atomic_inc(&mddev
->active_io
);
239 rv
= mddev
->pers
->make_request(q
, bio
);
240 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
241 wake_up(&mddev
->sb_wait
);
246 static void mddev_suspend(mddev_t
*mddev
)
248 BUG_ON(mddev
->suspended
);
249 mddev
->suspended
= 1;
251 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
252 mddev
->pers
->quiesce(mddev
, 1);
253 md_unregister_thread(mddev
->thread
);
254 mddev
->thread
= NULL
;
255 /* we now know that no code is executing in the personality module,
256 * except possibly the tail end of a ->bi_end_io function, but that
257 * is certain to complete before the module has a chance to get
262 static void mddev_resume(mddev_t
*mddev
)
264 mddev
->suspended
= 0;
265 wake_up(&mddev
->sb_wait
);
266 mddev
->pers
->quiesce(mddev
, 0);
269 int mddev_congested(mddev_t
*mddev
, int bits
)
273 return mddev
->suspended
;
275 EXPORT_SYMBOL(mddev_congested
);
278 * Generic barrier handling for md
281 #define POST_REQUEST_BARRIER ((void*)1)
283 static void md_end_barrier(struct bio
*bio
, int err
)
285 mdk_rdev_t
*rdev
= bio
->bi_private
;
286 mddev_t
*mddev
= rdev
->mddev
;
287 if (err
== -EOPNOTSUPP
&& mddev
->barrier
!= POST_REQUEST_BARRIER
)
288 set_bit(BIO_EOPNOTSUPP
, &mddev
->barrier
->bi_flags
);
290 rdev_dec_pending(rdev
, mddev
);
292 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
293 if (mddev
->barrier
== POST_REQUEST_BARRIER
) {
294 /* This was a post-request barrier */
295 mddev
->barrier
= NULL
;
296 wake_up(&mddev
->sb_wait
);
298 /* The pre-request barrier has finished */
299 schedule_work(&mddev
->barrier_work
);
304 static void submit_barriers(mddev_t
*mddev
)
309 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
310 if (rdev
->raid_disk
>= 0 &&
311 !test_bit(Faulty
, &rdev
->flags
)) {
312 /* Take two references, one is dropped
313 * when request finishes, one after
314 * we reclaim rcu_read_lock
317 atomic_inc(&rdev
->nr_pending
);
318 atomic_inc(&rdev
->nr_pending
);
320 bi
= bio_alloc(GFP_KERNEL
, 0);
321 bi
->bi_end_io
= md_end_barrier
;
322 bi
->bi_private
= rdev
;
323 bi
->bi_bdev
= rdev
->bdev
;
324 atomic_inc(&mddev
->flush_pending
);
325 submit_bio(WRITE_BARRIER
, bi
);
327 rdev_dec_pending(rdev
, mddev
);
332 static void md_submit_barrier(struct work_struct
*ws
)
334 mddev_t
*mddev
= container_of(ws
, mddev_t
, barrier_work
);
335 struct bio
*bio
= mddev
->barrier
;
337 atomic_set(&mddev
->flush_pending
, 1);
339 if (test_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
))
340 bio_endio(bio
, -EOPNOTSUPP
);
341 else if (bio
->bi_size
== 0)
342 /* an empty barrier - all done */
345 bio
->bi_rw
&= ~(1<<BIO_RW_BARRIER
);
346 if (mddev
->pers
->make_request(mddev
->queue
, bio
))
347 generic_make_request(bio
);
348 mddev
->barrier
= POST_REQUEST_BARRIER
;
349 submit_barriers(mddev
);
351 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
352 mddev
->barrier
= NULL
;
353 wake_up(&mddev
->sb_wait
);
357 void md_barrier_request(mddev_t
*mddev
, struct bio
*bio
)
359 spin_lock_irq(&mddev
->write_lock
);
360 wait_event_lock_irq(mddev
->sb_wait
,
362 mddev
->write_lock
, /*nothing*/);
363 mddev
->barrier
= bio
;
364 spin_unlock_irq(&mddev
->write_lock
);
366 atomic_set(&mddev
->flush_pending
, 1);
367 INIT_WORK(&mddev
->barrier_work
, md_submit_barrier
);
369 submit_barriers(mddev
);
371 if (atomic_dec_and_test(&mddev
->flush_pending
))
372 schedule_work(&mddev
->barrier_work
);
374 EXPORT_SYMBOL(md_barrier_request
);
376 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
378 atomic_inc(&mddev
->active
);
382 static void mddev_delayed_delete(struct work_struct
*ws
);
384 static void mddev_put(mddev_t
*mddev
)
386 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
388 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
389 !mddev
->hold_active
) {
390 list_del(&mddev
->all_mddevs
);
391 if (mddev
->gendisk
) {
392 /* we did a probe so need to clean up.
393 * Call schedule_work inside the spinlock
394 * so that flush_scheduled_work() after
395 * mddev_find will succeed in waiting for the
398 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
399 schedule_work(&mddev
->del_work
);
403 spin_unlock(&all_mddevs_lock
);
406 static mddev_t
* mddev_find(dev_t unit
)
408 mddev_t
*mddev
, *new = NULL
;
411 spin_lock(&all_mddevs_lock
);
414 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
415 if (mddev
->unit
== unit
) {
417 spin_unlock(&all_mddevs_lock
);
423 list_add(&new->all_mddevs
, &all_mddevs
);
424 spin_unlock(&all_mddevs_lock
);
425 new->hold_active
= UNTIL_IOCTL
;
429 /* find an unused unit number */
430 static int next_minor
= 512;
431 int start
= next_minor
;
435 dev
= MKDEV(MD_MAJOR
, next_minor
);
437 if (next_minor
> MINORMASK
)
439 if (next_minor
== start
) {
440 /* Oh dear, all in use. */
441 spin_unlock(&all_mddevs_lock
);
447 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
448 if (mddev
->unit
== dev
) {
454 new->md_minor
= MINOR(dev
);
455 new->hold_active
= UNTIL_STOP
;
456 list_add(&new->all_mddevs
, &all_mddevs
);
457 spin_unlock(&all_mddevs_lock
);
460 spin_unlock(&all_mddevs_lock
);
462 new = kzalloc(sizeof(*new), GFP_KERNEL
);
467 if (MAJOR(unit
) == MD_MAJOR
)
468 new->md_minor
= MINOR(unit
);
470 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
472 mutex_init(&new->open_mutex
);
473 mutex_init(&new->reconfig_mutex
);
474 mutex_init(&new->bitmap_info
.mutex
);
475 INIT_LIST_HEAD(&new->disks
);
476 INIT_LIST_HEAD(&new->all_mddevs
);
477 init_timer(&new->safemode_timer
);
478 atomic_set(&new->active
, 1);
479 atomic_set(&new->openers
, 0);
480 atomic_set(&new->active_io
, 0);
481 spin_lock_init(&new->write_lock
);
482 atomic_set(&new->flush_pending
, 0);
483 init_waitqueue_head(&new->sb_wait
);
484 init_waitqueue_head(&new->recovery_wait
);
485 new->reshape_position
= MaxSector
;
487 new->resync_max
= MaxSector
;
488 new->level
= LEVEL_NONE
;
493 static inline int mddev_lock(mddev_t
* mddev
)
495 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
498 static inline int mddev_is_locked(mddev_t
*mddev
)
500 return mutex_is_locked(&mddev
->reconfig_mutex
);
503 static inline int mddev_trylock(mddev_t
* mddev
)
505 return mutex_trylock(&mddev
->reconfig_mutex
);
508 static inline void mddev_unlock(mddev_t
* mddev
)
510 mutex_unlock(&mddev
->reconfig_mutex
);
512 md_wakeup_thread(mddev
->thread
);
515 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
519 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
520 if (rdev
->desc_nr
== nr
)
526 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
530 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
531 if (rdev
->bdev
->bd_dev
== dev
)
537 static struct mdk_personality
*find_pers(int level
, char *clevel
)
539 struct mdk_personality
*pers
;
540 list_for_each_entry(pers
, &pers_list
, list
) {
541 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
543 if (strcmp(pers
->name
, clevel
)==0)
549 /* return the offset of the super block in 512byte sectors */
550 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
552 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
553 return MD_NEW_SIZE_SECTORS(num_sectors
);
556 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
561 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
562 if (!rdev
->sb_page
) {
563 printk(KERN_ALERT
"md: out of memory.\n");
570 static void free_disk_sb(mdk_rdev_t
* rdev
)
573 put_page(rdev
->sb_page
);
575 rdev
->sb_page
= NULL
;
582 static void super_written(struct bio
*bio
, int error
)
584 mdk_rdev_t
*rdev
= bio
->bi_private
;
585 mddev_t
*mddev
= rdev
->mddev
;
587 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
588 printk("md: super_written gets error=%d, uptodate=%d\n",
589 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
590 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
591 md_error(mddev
, rdev
);
594 if (atomic_dec_and_test(&mddev
->pending_writes
))
595 wake_up(&mddev
->sb_wait
);
599 static void super_written_barrier(struct bio
*bio
, int error
)
601 struct bio
*bio2
= bio
->bi_private
;
602 mdk_rdev_t
*rdev
= bio2
->bi_private
;
603 mddev_t
*mddev
= rdev
->mddev
;
605 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
606 error
== -EOPNOTSUPP
) {
608 /* barriers don't appear to be supported :-( */
609 set_bit(BarriersNotsupp
, &rdev
->flags
);
610 mddev
->barriers_work
= 0;
611 spin_lock_irqsave(&mddev
->write_lock
, flags
);
612 bio2
->bi_next
= mddev
->biolist
;
613 mddev
->biolist
= bio2
;
614 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
615 wake_up(&mddev
->sb_wait
);
619 bio
->bi_private
= rdev
;
620 super_written(bio
, error
);
624 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
625 sector_t sector
, int size
, struct page
*page
)
627 /* write first size bytes of page to sector of rdev
628 * Increment mddev->pending_writes before returning
629 * and decrement it on completion, waking up sb_wait
630 * if zero is reached.
631 * If an error occurred, call md_error
633 * As we might need to resubmit the request if BIO_RW_BARRIER
634 * causes ENOTSUPP, we allocate a spare bio...
636 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
637 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
639 bio
->bi_bdev
= rdev
->bdev
;
640 bio
->bi_sector
= sector
;
641 bio_add_page(bio
, page
, size
, 0);
642 bio
->bi_private
= rdev
;
643 bio
->bi_end_io
= super_written
;
646 atomic_inc(&mddev
->pending_writes
);
647 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
649 rw
|= (1<<BIO_RW_BARRIER
);
650 rbio
= bio_clone(bio
, GFP_NOIO
);
651 rbio
->bi_private
= bio
;
652 rbio
->bi_end_io
= super_written_barrier
;
653 submit_bio(rw
, rbio
);
658 void md_super_wait(mddev_t
*mddev
)
660 /* wait for all superblock writes that were scheduled to complete.
661 * if any had to be retried (due to BARRIER problems), retry them
665 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
666 if (atomic_read(&mddev
->pending_writes
)==0)
668 while (mddev
->biolist
) {
670 spin_lock_irq(&mddev
->write_lock
);
671 bio
= mddev
->biolist
;
672 mddev
->biolist
= bio
->bi_next
;
674 spin_unlock_irq(&mddev
->write_lock
);
675 submit_bio(bio
->bi_rw
, bio
);
679 finish_wait(&mddev
->sb_wait
, &wq
);
682 static void bi_complete(struct bio
*bio
, int error
)
684 complete((struct completion
*)bio
->bi_private
);
687 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
688 struct page
*page
, int rw
)
690 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
691 struct completion event
;
694 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
697 bio
->bi_sector
= sector
;
698 bio_add_page(bio
, page
, size
, 0);
699 init_completion(&event
);
700 bio
->bi_private
= &event
;
701 bio
->bi_end_io
= bi_complete
;
703 wait_for_completion(&event
);
705 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
709 EXPORT_SYMBOL_GPL(sync_page_io
);
711 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
713 char b
[BDEVNAME_SIZE
];
714 if (!rdev
->sb_page
) {
722 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
728 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
729 bdevname(rdev
->bdev
,b
));
733 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
735 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
736 sb1
->set_uuid1
== sb2
->set_uuid1
&&
737 sb1
->set_uuid2
== sb2
->set_uuid2
&&
738 sb1
->set_uuid3
== sb2
->set_uuid3
;
741 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
744 mdp_super_t
*tmp1
, *tmp2
;
746 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
747 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
749 if (!tmp1
|| !tmp2
) {
751 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
759 * nr_disks is not constant
764 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
772 static u32
md_csum_fold(u32 csum
)
774 csum
= (csum
& 0xffff) + (csum
>> 16);
775 return (csum
& 0xffff) + (csum
>> 16);
778 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
781 u32
*sb32
= (u32
*)sb
;
783 unsigned int disk_csum
, csum
;
785 disk_csum
= sb
->sb_csum
;
788 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
790 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
794 /* This used to use csum_partial, which was wrong for several
795 * reasons including that different results are returned on
796 * different architectures. It isn't critical that we get exactly
797 * the same return value as before (we always csum_fold before
798 * testing, and that removes any differences). However as we
799 * know that csum_partial always returned a 16bit value on
800 * alphas, do a fold to maximise conformity to previous behaviour.
802 sb
->sb_csum
= md_csum_fold(disk_csum
);
804 sb
->sb_csum
= disk_csum
;
811 * Handle superblock details.
812 * We want to be able to handle multiple superblock formats
813 * so we have a common interface to them all, and an array of
814 * different handlers.
815 * We rely on user-space to write the initial superblock, and support
816 * reading and updating of superblocks.
817 * Interface methods are:
818 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
819 * loads and validates a superblock on dev.
820 * if refdev != NULL, compare superblocks on both devices
822 * 0 - dev has a superblock that is compatible with refdev
823 * 1 - dev has a superblock that is compatible and newer than refdev
824 * so dev should be used as the refdev in future
825 * -EINVAL superblock incompatible or invalid
826 * -othererror e.g. -EIO
828 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
829 * Verify that dev is acceptable into mddev.
830 * The first time, mddev->raid_disks will be 0, and data from
831 * dev should be merged in. Subsequent calls check that dev
832 * is new enough. Return 0 or -EINVAL
834 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
835 * Update the superblock for rdev with data in mddev
836 * This does not write to disc.
842 struct module
*owner
;
843 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
845 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
846 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
847 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
848 sector_t num_sectors
);
852 * Check that the given mddev has no bitmap.
854 * This function is called from the run method of all personalities that do not
855 * support bitmaps. It prints an error message and returns non-zero if mddev
856 * has a bitmap. Otherwise, it returns 0.
859 int md_check_no_bitmap(mddev_t
*mddev
)
861 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
863 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
864 mdname(mddev
), mddev
->pers
->name
);
867 EXPORT_SYMBOL(md_check_no_bitmap
);
870 * load_super for 0.90.0
872 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
874 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
879 * Calculate the position of the superblock (512byte sectors),
880 * it's at the end of the disk.
882 * It also happens to be a multiple of 4Kb.
884 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
886 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
891 bdevname(rdev
->bdev
, b
);
892 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
894 if (sb
->md_magic
!= MD_SB_MAGIC
) {
895 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
900 if (sb
->major_version
!= 0 ||
901 sb
->minor_version
< 90 ||
902 sb
->minor_version
> 91) {
903 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
904 sb
->major_version
, sb
->minor_version
,
909 if (sb
->raid_disks
<= 0)
912 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
913 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
918 rdev
->preferred_minor
= sb
->md_minor
;
919 rdev
->data_offset
= 0;
920 rdev
->sb_size
= MD_SB_BYTES
;
922 if (sb
->level
== LEVEL_MULTIPATH
)
925 rdev
->desc_nr
= sb
->this_disk
.number
;
931 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
932 if (!uuid_equal(refsb
, sb
)) {
933 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
934 b
, bdevname(refdev
->bdev
,b2
));
937 if (!sb_equal(refsb
, sb
)) {
938 printk(KERN_WARNING
"md: %s has same UUID"
939 " but different superblock to %s\n",
940 b
, bdevname(refdev
->bdev
, b2
));
944 ev2
= md_event(refsb
);
950 rdev
->sectors
= rdev
->sb_start
;
952 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
953 /* "this cannot possibly happen" ... */
961 * validate_super for 0.90.0
963 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
966 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
967 __u64 ev1
= md_event(sb
);
969 rdev
->raid_disk
= -1;
970 clear_bit(Faulty
, &rdev
->flags
);
971 clear_bit(In_sync
, &rdev
->flags
);
972 clear_bit(WriteMostly
, &rdev
->flags
);
973 clear_bit(BarriersNotsupp
, &rdev
->flags
);
975 if (mddev
->raid_disks
== 0) {
976 mddev
->major_version
= 0;
977 mddev
->minor_version
= sb
->minor_version
;
978 mddev
->patch_version
= sb
->patch_version
;
980 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
981 mddev
->ctime
= sb
->ctime
;
982 mddev
->utime
= sb
->utime
;
983 mddev
->level
= sb
->level
;
984 mddev
->clevel
[0] = 0;
985 mddev
->layout
= sb
->layout
;
986 mddev
->raid_disks
= sb
->raid_disks
;
987 mddev
->dev_sectors
= sb
->size
* 2;
989 mddev
->bitmap_info
.offset
= 0;
990 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
992 if (mddev
->minor_version
>= 91) {
993 mddev
->reshape_position
= sb
->reshape_position
;
994 mddev
->delta_disks
= sb
->delta_disks
;
995 mddev
->new_level
= sb
->new_level
;
996 mddev
->new_layout
= sb
->new_layout
;
997 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
999 mddev
->reshape_position
= MaxSector
;
1000 mddev
->delta_disks
= 0;
1001 mddev
->new_level
= mddev
->level
;
1002 mddev
->new_layout
= mddev
->layout
;
1003 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1006 if (sb
->state
& (1<<MD_SB_CLEAN
))
1007 mddev
->recovery_cp
= MaxSector
;
1009 if (sb
->events_hi
== sb
->cp_events_hi
&&
1010 sb
->events_lo
== sb
->cp_events_lo
) {
1011 mddev
->recovery_cp
= sb
->recovery_cp
;
1013 mddev
->recovery_cp
= 0;
1016 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1017 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1018 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1019 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1021 mddev
->max_disks
= MD_SB_DISKS
;
1023 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1024 mddev
->bitmap_info
.file
== NULL
)
1025 mddev
->bitmap_info
.offset
=
1026 mddev
->bitmap_info
.default_offset
;
1028 } else if (mddev
->pers
== NULL
) {
1029 /* Insist on good event counter while assembling */
1031 if (ev1
< mddev
->events
)
1033 } else if (mddev
->bitmap
) {
1034 /* if adding to array with a bitmap, then we can accept an
1035 * older device ... but not too old.
1037 if (ev1
< mddev
->bitmap
->events_cleared
)
1040 if (ev1
< mddev
->events
)
1041 /* just a hot-add of a new device, leave raid_disk at -1 */
1045 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1046 desc
= sb
->disks
+ rdev
->desc_nr
;
1048 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1049 set_bit(Faulty
, &rdev
->flags
);
1050 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1051 desc->raid_disk < mddev->raid_disks */) {
1052 set_bit(In_sync
, &rdev
->flags
);
1053 rdev
->raid_disk
= desc
->raid_disk
;
1054 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1055 /* active but not in sync implies recovery up to
1056 * reshape position. We don't know exactly where
1057 * that is, so set to zero for now */
1058 if (mddev
->minor_version
>= 91) {
1059 rdev
->recovery_offset
= 0;
1060 rdev
->raid_disk
= desc
->raid_disk
;
1063 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1064 set_bit(WriteMostly
, &rdev
->flags
);
1065 } else /* MULTIPATH are always insync */
1066 set_bit(In_sync
, &rdev
->flags
);
1071 * sync_super for 0.90.0
1073 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1077 int next_spare
= mddev
->raid_disks
;
1080 /* make rdev->sb match mddev data..
1083 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1084 * 3/ any empty disks < next_spare become removed
1086 * disks[0] gets initialised to REMOVED because
1087 * we cannot be sure from other fields if it has
1088 * been initialised or not.
1091 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1093 rdev
->sb_size
= MD_SB_BYTES
;
1095 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1097 memset(sb
, 0, sizeof(*sb
));
1099 sb
->md_magic
= MD_SB_MAGIC
;
1100 sb
->major_version
= mddev
->major_version
;
1101 sb
->patch_version
= mddev
->patch_version
;
1102 sb
->gvalid_words
= 0; /* ignored */
1103 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1104 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1105 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1106 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1108 sb
->ctime
= mddev
->ctime
;
1109 sb
->level
= mddev
->level
;
1110 sb
->size
= mddev
->dev_sectors
/ 2;
1111 sb
->raid_disks
= mddev
->raid_disks
;
1112 sb
->md_minor
= mddev
->md_minor
;
1113 sb
->not_persistent
= 0;
1114 sb
->utime
= mddev
->utime
;
1116 sb
->events_hi
= (mddev
->events
>>32);
1117 sb
->events_lo
= (u32
)mddev
->events
;
1119 if (mddev
->reshape_position
== MaxSector
)
1120 sb
->minor_version
= 90;
1122 sb
->minor_version
= 91;
1123 sb
->reshape_position
= mddev
->reshape_position
;
1124 sb
->new_level
= mddev
->new_level
;
1125 sb
->delta_disks
= mddev
->delta_disks
;
1126 sb
->new_layout
= mddev
->new_layout
;
1127 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1129 mddev
->minor_version
= sb
->minor_version
;
1132 sb
->recovery_cp
= mddev
->recovery_cp
;
1133 sb
->cp_events_hi
= (mddev
->events
>>32);
1134 sb
->cp_events_lo
= (u32
)mddev
->events
;
1135 if (mddev
->recovery_cp
== MaxSector
)
1136 sb
->state
= (1<< MD_SB_CLEAN
);
1138 sb
->recovery_cp
= 0;
1140 sb
->layout
= mddev
->layout
;
1141 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1143 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1144 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1146 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1147 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1150 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1152 if (rdev2
->raid_disk
>= 0 &&
1153 sb
->minor_version
>= 91)
1154 /* we have nowhere to store the recovery_offset,
1155 * but if it is not below the reshape_position,
1156 * we can piggy-back on that.
1159 if (rdev2
->raid_disk
< 0 ||
1160 test_bit(Faulty
, &rdev2
->flags
))
1163 desc_nr
= rdev2
->raid_disk
;
1165 desc_nr
= next_spare
++;
1166 rdev2
->desc_nr
= desc_nr
;
1167 d
= &sb
->disks
[rdev2
->desc_nr
];
1169 d
->number
= rdev2
->desc_nr
;
1170 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1171 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1173 d
->raid_disk
= rdev2
->raid_disk
;
1175 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1176 if (test_bit(Faulty
, &rdev2
->flags
))
1177 d
->state
= (1<<MD_DISK_FAULTY
);
1178 else if (is_active
) {
1179 d
->state
= (1<<MD_DISK_ACTIVE
);
1180 if (test_bit(In_sync
, &rdev2
->flags
))
1181 d
->state
|= (1<<MD_DISK_SYNC
);
1189 if (test_bit(WriteMostly
, &rdev2
->flags
))
1190 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1192 /* now set the "removed" and "faulty" bits on any missing devices */
1193 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1194 mdp_disk_t
*d
= &sb
->disks
[i
];
1195 if (d
->state
== 0 && d
->number
== 0) {
1198 d
->state
= (1<<MD_DISK_REMOVED
);
1199 d
->state
|= (1<<MD_DISK_FAULTY
);
1203 sb
->nr_disks
= nr_disks
;
1204 sb
->active_disks
= active
;
1205 sb
->working_disks
= working
;
1206 sb
->failed_disks
= failed
;
1207 sb
->spare_disks
= spare
;
1209 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1210 sb
->sb_csum
= calc_sb_csum(sb
);
1214 * rdev_size_change for 0.90.0
1216 static unsigned long long
1217 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1219 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1220 return 0; /* component must fit device */
1221 if (rdev
->mddev
->bitmap_info
.offset
)
1222 return 0; /* can't move bitmap */
1223 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1224 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1225 num_sectors
= rdev
->sb_start
;
1226 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1228 md_super_wait(rdev
->mddev
);
1229 return num_sectors
/ 2; /* kB for sysfs */
1234 * version 1 superblock
1237 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1241 unsigned long long newcsum
;
1242 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1243 __le32
*isuper
= (__le32
*)sb
;
1246 disk_csum
= sb
->sb_csum
;
1249 for (i
=0; size
>=4; size
-= 4 )
1250 newcsum
+= le32_to_cpu(*isuper
++);
1253 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1255 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1256 sb
->sb_csum
= disk_csum
;
1257 return cpu_to_le32(csum
);
1260 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1262 struct mdp_superblock_1
*sb
;
1265 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1269 * Calculate the position of the superblock in 512byte sectors.
1270 * It is always aligned to a 4K boundary and
1271 * depeding on minor_version, it can be:
1272 * 0: At least 8K, but less than 12K, from end of device
1273 * 1: At start of device
1274 * 2: 4K from start of device.
1276 switch(minor_version
) {
1278 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1280 sb_start
&= ~(sector_t
)(4*2-1);
1291 rdev
->sb_start
= sb_start
;
1293 /* superblock is rarely larger than 1K, but it can be larger,
1294 * and it is safe to read 4k, so we do that
1296 ret
= read_disk_sb(rdev
, 4096);
1297 if (ret
) return ret
;
1300 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1302 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1303 sb
->major_version
!= cpu_to_le32(1) ||
1304 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1305 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1306 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1309 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1310 printk("md: invalid superblock checksum on %s\n",
1311 bdevname(rdev
->bdev
,b
));
1314 if (le64_to_cpu(sb
->data_size
) < 10) {
1315 printk("md: data_size too small on %s\n",
1316 bdevname(rdev
->bdev
,b
));
1320 rdev
->preferred_minor
= 0xffff;
1321 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1322 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1324 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1325 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1326 if (rdev
->sb_size
& bmask
)
1327 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1330 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1333 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1336 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1342 struct mdp_superblock_1
*refsb
=
1343 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1345 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1346 sb
->level
!= refsb
->level
||
1347 sb
->layout
!= refsb
->layout
||
1348 sb
->chunksize
!= refsb
->chunksize
) {
1349 printk(KERN_WARNING
"md: %s has strangely different"
1350 " superblock to %s\n",
1351 bdevname(rdev
->bdev
,b
),
1352 bdevname(refdev
->bdev
,b2
));
1355 ev1
= le64_to_cpu(sb
->events
);
1356 ev2
= le64_to_cpu(refsb
->events
);
1364 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1365 le64_to_cpu(sb
->data_offset
);
1367 rdev
->sectors
= rdev
->sb_start
;
1368 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1370 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1371 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1376 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1378 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1379 __u64 ev1
= le64_to_cpu(sb
->events
);
1381 rdev
->raid_disk
= -1;
1382 clear_bit(Faulty
, &rdev
->flags
);
1383 clear_bit(In_sync
, &rdev
->flags
);
1384 clear_bit(WriteMostly
, &rdev
->flags
);
1385 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1387 if (mddev
->raid_disks
== 0) {
1388 mddev
->major_version
= 1;
1389 mddev
->patch_version
= 0;
1390 mddev
->external
= 0;
1391 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1392 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1393 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1394 mddev
->level
= le32_to_cpu(sb
->level
);
1395 mddev
->clevel
[0] = 0;
1396 mddev
->layout
= le32_to_cpu(sb
->layout
);
1397 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1398 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1399 mddev
->events
= ev1
;
1400 mddev
->bitmap_info
.offset
= 0;
1401 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1403 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1404 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1406 mddev
->max_disks
= (4096-256)/2;
1408 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1409 mddev
->bitmap_info
.file
== NULL
)
1410 mddev
->bitmap_info
.offset
=
1411 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1413 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1414 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1415 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1416 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1417 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1418 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1420 mddev
->reshape_position
= MaxSector
;
1421 mddev
->delta_disks
= 0;
1422 mddev
->new_level
= mddev
->level
;
1423 mddev
->new_layout
= mddev
->layout
;
1424 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1427 } else if (mddev
->pers
== NULL
) {
1428 /* Insist of good event counter while assembling */
1430 if (ev1
< mddev
->events
)
1432 } else if (mddev
->bitmap
) {
1433 /* If adding to array with a bitmap, then we can accept an
1434 * older device, but not too old.
1436 if (ev1
< mddev
->bitmap
->events_cleared
)
1439 if (ev1
< mddev
->events
)
1440 /* just a hot-add of a new device, leave raid_disk at -1 */
1443 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1445 if (rdev
->desc_nr
< 0 ||
1446 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1450 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1452 case 0xffff: /* spare */
1454 case 0xfffe: /* faulty */
1455 set_bit(Faulty
, &rdev
->flags
);
1458 if ((le32_to_cpu(sb
->feature_map
) &
1459 MD_FEATURE_RECOVERY_OFFSET
))
1460 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1462 set_bit(In_sync
, &rdev
->flags
);
1463 rdev
->raid_disk
= role
;
1466 if (sb
->devflags
& WriteMostly1
)
1467 set_bit(WriteMostly
, &rdev
->flags
);
1468 } else /* MULTIPATH are always insync */
1469 set_bit(In_sync
, &rdev
->flags
);
1474 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1476 struct mdp_superblock_1
*sb
;
1479 /* make rdev->sb match mddev and rdev data. */
1481 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1483 sb
->feature_map
= 0;
1485 sb
->recovery_offset
= cpu_to_le64(0);
1486 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1487 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1488 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1490 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1491 sb
->events
= cpu_to_le64(mddev
->events
);
1493 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1495 sb
->resync_offset
= cpu_to_le64(0);
1497 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1499 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1500 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1501 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1502 sb
->level
= cpu_to_le32(mddev
->level
);
1503 sb
->layout
= cpu_to_le32(mddev
->layout
);
1505 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1506 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1507 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1510 if (rdev
->raid_disk
>= 0 &&
1511 !test_bit(In_sync
, &rdev
->flags
)) {
1513 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1514 sb
->recovery_offset
=
1515 cpu_to_le64(rdev
->recovery_offset
);
1518 if (mddev
->reshape_position
!= MaxSector
) {
1519 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1520 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1521 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1522 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1523 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1524 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1528 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1529 if (rdev2
->desc_nr
+1 > max_dev
)
1530 max_dev
= rdev2
->desc_nr
+1;
1532 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1534 sb
->max_dev
= cpu_to_le32(max_dev
);
1535 rdev
->sb_size
= max_dev
* 2 + 256;
1536 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1537 if (rdev
->sb_size
& bmask
)
1538 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1540 for (i
=0; i
<max_dev
;i
++)
1541 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1543 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1545 if (test_bit(Faulty
, &rdev2
->flags
))
1546 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1547 else if (test_bit(In_sync
, &rdev2
->flags
))
1548 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1549 else if (rdev2
->raid_disk
>= 0)
1550 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1552 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1555 sb
->sb_csum
= calc_sb_1_csum(sb
);
1558 static unsigned long long
1559 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1561 struct mdp_superblock_1
*sb
;
1562 sector_t max_sectors
;
1563 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1564 return 0; /* component must fit device */
1565 if (rdev
->sb_start
< rdev
->data_offset
) {
1566 /* minor versions 1 and 2; superblock before data */
1567 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1568 max_sectors
-= rdev
->data_offset
;
1569 if (!num_sectors
|| num_sectors
> max_sectors
)
1570 num_sectors
= max_sectors
;
1571 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1572 /* minor version 0 with bitmap we can't move */
1575 /* minor version 0; superblock after data */
1577 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1578 sb_start
&= ~(sector_t
)(4*2 - 1);
1579 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1580 if (!num_sectors
|| num_sectors
> max_sectors
)
1581 num_sectors
= max_sectors
;
1582 rdev
->sb_start
= sb_start
;
1584 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1585 sb
->data_size
= cpu_to_le64(num_sectors
);
1586 sb
->super_offset
= rdev
->sb_start
;
1587 sb
->sb_csum
= calc_sb_1_csum(sb
);
1588 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1590 md_super_wait(rdev
->mddev
);
1591 return num_sectors
/ 2; /* kB for sysfs */
1594 static struct super_type super_types
[] = {
1597 .owner
= THIS_MODULE
,
1598 .load_super
= super_90_load
,
1599 .validate_super
= super_90_validate
,
1600 .sync_super
= super_90_sync
,
1601 .rdev_size_change
= super_90_rdev_size_change
,
1605 .owner
= THIS_MODULE
,
1606 .load_super
= super_1_load
,
1607 .validate_super
= super_1_validate
,
1608 .sync_super
= super_1_sync
,
1609 .rdev_size_change
= super_1_rdev_size_change
,
1613 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1615 mdk_rdev_t
*rdev
, *rdev2
;
1618 rdev_for_each_rcu(rdev
, mddev1
)
1619 rdev_for_each_rcu(rdev2
, mddev2
)
1620 if (rdev
->bdev
->bd_contains
==
1621 rdev2
->bdev
->bd_contains
) {
1629 static LIST_HEAD(pending_raid_disks
);
1632 * Try to register data integrity profile for an mddev
1634 * This is called when an array is started and after a disk has been kicked
1635 * from the array. It only succeeds if all working and active component devices
1636 * are integrity capable with matching profiles.
1638 int md_integrity_register(mddev_t
*mddev
)
1640 mdk_rdev_t
*rdev
, *reference
= NULL
;
1642 if (list_empty(&mddev
->disks
))
1643 return 0; /* nothing to do */
1644 if (blk_get_integrity(mddev
->gendisk
))
1645 return 0; /* already registered */
1646 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1647 /* skip spares and non-functional disks */
1648 if (test_bit(Faulty
, &rdev
->flags
))
1650 if (rdev
->raid_disk
< 0)
1653 * If at least one rdev is not integrity capable, we can not
1654 * enable data integrity for the md device.
1656 if (!bdev_get_integrity(rdev
->bdev
))
1659 /* Use the first rdev as the reference */
1663 /* does this rdev's profile match the reference profile? */
1664 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1665 rdev
->bdev
->bd_disk
) < 0)
1669 * All component devices are integrity capable and have matching
1670 * profiles, register the common profile for the md device.
1672 if (blk_integrity_register(mddev
->gendisk
,
1673 bdev_get_integrity(reference
->bdev
)) != 0) {
1674 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1678 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1682 EXPORT_SYMBOL(md_integrity_register
);
1684 /* Disable data integrity if non-capable/non-matching disk is being added */
1685 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1687 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1688 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1690 if (!bi_mddev
) /* nothing to do */
1692 if (rdev
->raid_disk
< 0) /* skip spares */
1694 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1695 rdev
->bdev
->bd_disk
) >= 0)
1697 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1698 blk_integrity_unregister(mddev
->gendisk
);
1700 EXPORT_SYMBOL(md_integrity_add_rdev
);
1702 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1704 char b
[BDEVNAME_SIZE
];
1714 /* prevent duplicates */
1715 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1718 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1719 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1720 rdev
->sectors
< mddev
->dev_sectors
)) {
1722 /* Cannot change size, so fail
1723 * If mddev->level <= 0, then we don't care
1724 * about aligning sizes (e.g. linear)
1726 if (mddev
->level
> 0)
1729 mddev
->dev_sectors
= rdev
->sectors
;
1732 /* Verify rdev->desc_nr is unique.
1733 * If it is -1, assign a free number, else
1734 * check number is not in use
1736 if (rdev
->desc_nr
< 0) {
1738 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1739 while (find_rdev_nr(mddev
, choice
))
1741 rdev
->desc_nr
= choice
;
1743 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1746 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1747 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1748 mdname(mddev
), mddev
->max_disks
);
1751 bdevname(rdev
->bdev
,b
);
1752 while ( (s
=strchr(b
, '/')) != NULL
)
1755 rdev
->mddev
= mddev
;
1756 printk(KERN_INFO
"md: bind<%s>\n", b
);
1758 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1761 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1762 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1763 kobject_del(&rdev
->kobj
);
1766 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1768 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1769 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1771 /* May as well allow recovery to be retried once */
1772 mddev
->recovery_disabled
= 0;
1777 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1782 static void md_delayed_delete(struct work_struct
*ws
)
1784 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1785 kobject_del(&rdev
->kobj
);
1786 kobject_put(&rdev
->kobj
);
1789 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1791 char b
[BDEVNAME_SIZE
];
1796 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1797 list_del_rcu(&rdev
->same_set
);
1798 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1800 sysfs_remove_link(&rdev
->kobj
, "block");
1801 sysfs_put(rdev
->sysfs_state
);
1802 rdev
->sysfs_state
= NULL
;
1803 /* We need to delay this, otherwise we can deadlock when
1804 * writing to 'remove' to "dev/state". We also need
1805 * to delay it due to rcu usage.
1808 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1809 kobject_get(&rdev
->kobj
);
1810 schedule_work(&rdev
->del_work
);
1814 * prevent the device from being mounted, repartitioned or
1815 * otherwise reused by a RAID array (or any other kernel
1816 * subsystem), by bd_claiming the device.
1818 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1821 struct block_device
*bdev
;
1822 char b
[BDEVNAME_SIZE
];
1824 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1826 printk(KERN_ERR
"md: could not open %s.\n",
1827 __bdevname(dev
, b
));
1828 return PTR_ERR(bdev
);
1830 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1832 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1834 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1838 set_bit(AllReserved
, &rdev
->flags
);
1843 static void unlock_rdev(mdk_rdev_t
*rdev
)
1845 struct block_device
*bdev
= rdev
->bdev
;
1850 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1853 void md_autodetect_dev(dev_t dev
);
1855 static void export_rdev(mdk_rdev_t
* rdev
)
1857 char b
[BDEVNAME_SIZE
];
1858 printk(KERN_INFO
"md: export_rdev(%s)\n",
1859 bdevname(rdev
->bdev
,b
));
1864 if (test_bit(AutoDetected
, &rdev
->flags
))
1865 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1868 kobject_put(&rdev
->kobj
);
1871 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1873 unbind_rdev_from_array(rdev
);
1877 static void export_array(mddev_t
*mddev
)
1879 mdk_rdev_t
*rdev
, *tmp
;
1881 rdev_for_each(rdev
, tmp
, mddev
) {
1886 kick_rdev_from_array(rdev
);
1888 if (!list_empty(&mddev
->disks
))
1890 mddev
->raid_disks
= 0;
1891 mddev
->major_version
= 0;
1894 static void print_desc(mdp_disk_t
*desc
)
1896 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1897 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1900 static void print_sb_90(mdp_super_t
*sb
)
1905 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1906 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1907 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1909 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1910 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1911 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1912 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1913 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1914 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1915 sb
->failed_disks
, sb
->spare_disks
,
1916 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1919 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1922 desc
= sb
->disks
+ i
;
1923 if (desc
->number
|| desc
->major
|| desc
->minor
||
1924 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1925 printk(" D %2d: ", i
);
1929 printk(KERN_INFO
"md: THIS: ");
1930 print_desc(&sb
->this_disk
);
1933 static void print_sb_1(struct mdp_superblock_1
*sb
)
1937 uuid
= sb
->set_uuid
;
1939 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1940 "md: Name: \"%s\" CT:%llu\n",
1941 le32_to_cpu(sb
->major_version
),
1942 le32_to_cpu(sb
->feature_map
),
1945 (unsigned long long)le64_to_cpu(sb
->ctime
)
1946 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1948 uuid
= sb
->device_uuid
;
1950 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1952 "md: Dev:%08x UUID: %pU\n"
1953 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1954 "md: (MaxDev:%u) \n",
1955 le32_to_cpu(sb
->level
),
1956 (unsigned long long)le64_to_cpu(sb
->size
),
1957 le32_to_cpu(sb
->raid_disks
),
1958 le32_to_cpu(sb
->layout
),
1959 le32_to_cpu(sb
->chunksize
),
1960 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1961 (unsigned long long)le64_to_cpu(sb
->data_size
),
1962 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1963 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1964 le32_to_cpu(sb
->dev_number
),
1967 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1968 (unsigned long long)le64_to_cpu(sb
->events
),
1969 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
1970 le32_to_cpu(sb
->sb_csum
),
1971 le32_to_cpu(sb
->max_dev
)
1975 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
1977 char b
[BDEVNAME_SIZE
];
1978 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1979 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
1980 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1982 if (rdev
->sb_loaded
) {
1983 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
1984 switch (major_version
) {
1986 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
1989 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
1993 printk(KERN_INFO
"md: no rdev superblock!\n");
1996 static void md_print_devices(void)
1998 struct list_head
*tmp
;
2001 char b
[BDEVNAME_SIZE
];
2004 printk("md: **********************************\n");
2005 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2006 printk("md: **********************************\n");
2007 for_each_mddev(mddev
, tmp
) {
2010 bitmap_print_sb(mddev
->bitmap
);
2012 printk("%s: ", mdname(mddev
));
2013 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2014 printk("<%s>", bdevname(rdev
->bdev
,b
));
2017 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2018 print_rdev(rdev
, mddev
->major_version
);
2020 printk("md: **********************************\n");
2025 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2027 /* Update each superblock (in-memory image), but
2028 * if we are allowed to, skip spares which already
2029 * have the right event counter, or have one earlier
2030 * (which would mean they aren't being marked as dirty
2031 * with the rest of the array)
2035 /* First make sure individual recovery_offsets are correct */
2036 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2037 if (rdev
->raid_disk
>= 0 &&
2038 !test_bit(In_sync
, &rdev
->flags
) &&
2039 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2040 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2043 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2044 if (rdev
->sb_events
== mddev
->events
||
2046 rdev
->raid_disk
< 0 &&
2047 (rdev
->sb_events
&1)==0 &&
2048 rdev
->sb_events
+1 == mddev
->events
)) {
2049 /* Don't update this superblock */
2050 rdev
->sb_loaded
= 2;
2052 super_types
[mddev
->major_version
].
2053 sync_super(mddev
, rdev
);
2054 rdev
->sb_loaded
= 1;
2059 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2065 mddev
->utime
= get_seconds();
2066 if (mddev
->external
)
2069 spin_lock_irq(&mddev
->write_lock
);
2071 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2072 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2074 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2075 /* just a clean<-> dirty transition, possibly leave spares alone,
2076 * though if events isn't the right even/odd, we will have to do
2082 if (mddev
->degraded
)
2083 /* If the array is degraded, then skipping spares is both
2084 * dangerous and fairly pointless.
2085 * Dangerous because a device that was removed from the array
2086 * might have a event_count that still looks up-to-date,
2087 * so it can be re-added without a resync.
2088 * Pointless because if there are any spares to skip,
2089 * then a recovery will happen and soon that array won't
2090 * be degraded any more and the spare can go back to sleep then.
2094 sync_req
= mddev
->in_sync
;
2096 /* If this is just a dirty<->clean transition, and the array is clean
2097 * and 'events' is odd, we can roll back to the previous clean state */
2099 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2100 && (mddev
->events
& 1)
2101 && mddev
->events
!= 1)
2104 /* otherwise we have to go forward and ... */
2106 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
2107 /* .. if the array isn't clean, an 'even' event must also go
2109 if ((mddev
->events
&1)==0)
2112 /* otherwise an 'odd' event must go to spares */
2113 if ((mddev
->events
&1))
2118 if (!mddev
->events
) {
2120 * oops, this 64-bit counter should never wrap.
2121 * Either we are in around ~1 trillion A.C., assuming
2122 * 1 reboot per second, or we have a bug:
2129 * do not write anything to disk if using
2130 * nonpersistent superblocks
2132 if (!mddev
->persistent
) {
2133 if (!mddev
->external
)
2134 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2136 spin_unlock_irq(&mddev
->write_lock
);
2137 wake_up(&mddev
->sb_wait
);
2140 sync_sbs(mddev
, nospares
);
2141 spin_unlock_irq(&mddev
->write_lock
);
2144 "md: updating %s RAID superblock on device (in sync %d)\n",
2145 mdname(mddev
),mddev
->in_sync
);
2147 bitmap_update_sb(mddev
->bitmap
);
2148 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2149 char b
[BDEVNAME_SIZE
];
2150 dprintk(KERN_INFO
"md: ");
2151 if (rdev
->sb_loaded
!= 1)
2152 continue; /* no noise on spare devices */
2153 if (test_bit(Faulty
, &rdev
->flags
))
2154 dprintk("(skipping faulty ");
2156 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2157 if (!test_bit(Faulty
, &rdev
->flags
)) {
2158 md_super_write(mddev
,rdev
,
2159 rdev
->sb_start
, rdev
->sb_size
,
2161 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2162 bdevname(rdev
->bdev
,b
),
2163 (unsigned long long)rdev
->sb_start
);
2164 rdev
->sb_events
= mddev
->events
;
2168 if (mddev
->level
== LEVEL_MULTIPATH
)
2169 /* only need to write one superblock... */
2172 md_super_wait(mddev
);
2173 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2175 spin_lock_irq(&mddev
->write_lock
);
2176 if (mddev
->in_sync
!= sync_req
||
2177 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2178 /* have to write it out again */
2179 spin_unlock_irq(&mddev
->write_lock
);
2182 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2183 spin_unlock_irq(&mddev
->write_lock
);
2184 wake_up(&mddev
->sb_wait
);
2185 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2186 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2190 /* words written to sysfs files may, or may not, be \n terminated.
2191 * We want to accept with case. For this we use cmd_match.
2193 static int cmd_match(const char *cmd
, const char *str
)
2195 /* See if cmd, written into a sysfs file, matches
2196 * str. They must either be the same, or cmd can
2197 * have a trailing newline
2199 while (*cmd
&& *str
&& *cmd
== *str
) {
2210 struct rdev_sysfs_entry
{
2211 struct attribute attr
;
2212 ssize_t (*show
)(mdk_rdev_t
*, char *);
2213 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2217 state_show(mdk_rdev_t
*rdev
, char *page
)
2222 if (test_bit(Faulty
, &rdev
->flags
)) {
2223 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2226 if (test_bit(In_sync
, &rdev
->flags
)) {
2227 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2230 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2231 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2234 if (test_bit(Blocked
, &rdev
->flags
)) {
2235 len
+= sprintf(page
+len
, "%sblocked", sep
);
2238 if (!test_bit(Faulty
, &rdev
->flags
) &&
2239 !test_bit(In_sync
, &rdev
->flags
)) {
2240 len
+= sprintf(page
+len
, "%sspare", sep
);
2243 return len
+sprintf(page
+len
, "\n");
2247 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2250 * faulty - simulates and error
2251 * remove - disconnects the device
2252 * writemostly - sets write_mostly
2253 * -writemostly - clears write_mostly
2254 * blocked - sets the Blocked flag
2255 * -blocked - clears the Blocked flag
2256 * insync - sets Insync providing device isn't active
2259 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2260 md_error(rdev
->mddev
, rdev
);
2262 } else if (cmd_match(buf
, "remove")) {
2263 if (rdev
->raid_disk
>= 0)
2266 mddev_t
*mddev
= rdev
->mddev
;
2267 kick_rdev_from_array(rdev
);
2269 md_update_sb(mddev
, 1);
2270 md_new_event(mddev
);
2273 } else if (cmd_match(buf
, "writemostly")) {
2274 set_bit(WriteMostly
, &rdev
->flags
);
2276 } else if (cmd_match(buf
, "-writemostly")) {
2277 clear_bit(WriteMostly
, &rdev
->flags
);
2279 } else if (cmd_match(buf
, "blocked")) {
2280 set_bit(Blocked
, &rdev
->flags
);
2282 } else if (cmd_match(buf
, "-blocked")) {
2283 clear_bit(Blocked
, &rdev
->flags
);
2284 wake_up(&rdev
->blocked_wait
);
2285 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2286 md_wakeup_thread(rdev
->mddev
->thread
);
2289 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2290 set_bit(In_sync
, &rdev
->flags
);
2293 if (!err
&& rdev
->sysfs_state
)
2294 sysfs_notify_dirent(rdev
->sysfs_state
);
2295 return err
? err
: len
;
2297 static struct rdev_sysfs_entry rdev_state
=
2298 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2301 errors_show(mdk_rdev_t
*rdev
, char *page
)
2303 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2307 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2310 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2311 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2312 atomic_set(&rdev
->corrected_errors
, n
);
2317 static struct rdev_sysfs_entry rdev_errors
=
2318 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2321 slot_show(mdk_rdev_t
*rdev
, char *page
)
2323 if (rdev
->raid_disk
< 0)
2324 return sprintf(page
, "none\n");
2326 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2330 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2335 int slot
= simple_strtoul(buf
, &e
, 10);
2336 if (strncmp(buf
, "none", 4)==0)
2338 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2340 if (rdev
->mddev
->pers
&& slot
== -1) {
2341 /* Setting 'slot' on an active array requires also
2342 * updating the 'rd%d' link, and communicating
2343 * with the personality with ->hot_*_disk.
2344 * For now we only support removing
2345 * failed/spare devices. This normally happens automatically,
2346 * but not when the metadata is externally managed.
2348 if (rdev
->raid_disk
== -1)
2350 /* personality does all needed checks */
2351 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2353 err
= rdev
->mddev
->pers
->
2354 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2357 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2358 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2359 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2360 md_wakeup_thread(rdev
->mddev
->thread
);
2361 } else if (rdev
->mddev
->pers
) {
2363 /* Activating a spare .. or possibly reactivating
2364 * if we ever get bitmaps working here.
2367 if (rdev
->raid_disk
!= -1)
2370 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2373 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2374 if (rdev2
->raid_disk
== slot
)
2377 rdev
->raid_disk
= slot
;
2378 if (test_bit(In_sync
, &rdev
->flags
))
2379 rdev
->saved_raid_disk
= slot
;
2381 rdev
->saved_raid_disk
= -1;
2382 err
= rdev
->mddev
->pers
->
2383 hot_add_disk(rdev
->mddev
, rdev
);
2385 rdev
->raid_disk
= -1;
2388 sysfs_notify_dirent(rdev
->sysfs_state
);
2389 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2390 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2392 "md: cannot register "
2394 nm
, mdname(rdev
->mddev
));
2396 /* don't wakeup anyone, leave that to userspace. */
2398 if (slot
>= rdev
->mddev
->raid_disks
)
2400 rdev
->raid_disk
= slot
;
2401 /* assume it is working */
2402 clear_bit(Faulty
, &rdev
->flags
);
2403 clear_bit(WriteMostly
, &rdev
->flags
);
2404 set_bit(In_sync
, &rdev
->flags
);
2405 sysfs_notify_dirent(rdev
->sysfs_state
);
2411 static struct rdev_sysfs_entry rdev_slot
=
2412 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2415 offset_show(mdk_rdev_t
*rdev
, char *page
)
2417 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2421 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2424 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2425 if (e
==buf
|| (*e
&& *e
!= '\n'))
2427 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2429 if (rdev
->sectors
&& rdev
->mddev
->external
)
2430 /* Must set offset before size, so overlap checks
2433 rdev
->data_offset
= offset
;
2437 static struct rdev_sysfs_entry rdev_offset
=
2438 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2441 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2443 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2446 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2448 /* check if two start/length pairs overlap */
2456 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2458 unsigned long long blocks
;
2461 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2464 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2465 return -EINVAL
; /* sector conversion overflow */
2468 if (new != blocks
* 2)
2469 return -EINVAL
; /* unsigned long long to sector_t overflow */
2476 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2478 mddev_t
*my_mddev
= rdev
->mddev
;
2479 sector_t oldsectors
= rdev
->sectors
;
2482 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2484 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2485 if (my_mddev
->persistent
) {
2486 sectors
= super_types
[my_mddev
->major_version
].
2487 rdev_size_change(rdev
, sectors
);
2490 } else if (!sectors
)
2491 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2494 if (sectors
< my_mddev
->dev_sectors
)
2495 return -EINVAL
; /* component must fit device */
2497 rdev
->sectors
= sectors
;
2498 if (sectors
> oldsectors
&& my_mddev
->external
) {
2499 /* need to check that all other rdevs with the same ->bdev
2500 * do not overlap. We need to unlock the mddev to avoid
2501 * a deadlock. We have already changed rdev->sectors, and if
2502 * we have to change it back, we will have the lock again.
2506 struct list_head
*tmp
;
2508 mddev_unlock(my_mddev
);
2509 for_each_mddev(mddev
, tmp
) {
2513 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2514 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2515 (rdev
->bdev
== rdev2
->bdev
&&
2517 overlaps(rdev
->data_offset
, rdev
->sectors
,
2523 mddev_unlock(mddev
);
2529 mddev_lock(my_mddev
);
2531 /* Someone else could have slipped in a size
2532 * change here, but doing so is just silly.
2533 * We put oldsectors back because we *know* it is
2534 * safe, and trust userspace not to race with
2537 rdev
->sectors
= oldsectors
;
2544 static struct rdev_sysfs_entry rdev_size
=
2545 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2548 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2550 unsigned long long recovery_start
= rdev
->recovery_offset
;
2552 if (test_bit(In_sync
, &rdev
->flags
) ||
2553 recovery_start
== MaxSector
)
2554 return sprintf(page
, "none\n");
2556 return sprintf(page
, "%llu\n", recovery_start
);
2559 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2561 unsigned long long recovery_start
;
2563 if (cmd_match(buf
, "none"))
2564 recovery_start
= MaxSector
;
2565 else if (strict_strtoull(buf
, 10, &recovery_start
))
2568 if (rdev
->mddev
->pers
&&
2569 rdev
->raid_disk
>= 0)
2572 rdev
->recovery_offset
= recovery_start
;
2573 if (recovery_start
== MaxSector
)
2574 set_bit(In_sync
, &rdev
->flags
);
2576 clear_bit(In_sync
, &rdev
->flags
);
2580 static struct rdev_sysfs_entry rdev_recovery_start
=
2581 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2583 static struct attribute
*rdev_default_attrs
[] = {
2589 &rdev_recovery_start
.attr
,
2593 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2595 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2596 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2597 mddev_t
*mddev
= rdev
->mddev
;
2603 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2605 if (rdev
->mddev
== NULL
)
2608 rv
= entry
->show(rdev
, page
);
2609 mddev_unlock(mddev
);
2615 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2616 const char *page
, size_t length
)
2618 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2619 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2621 mddev_t
*mddev
= rdev
->mddev
;
2625 if (!capable(CAP_SYS_ADMIN
))
2627 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2629 if (rdev
->mddev
== NULL
)
2632 rv
= entry
->store(rdev
, page
, length
);
2633 mddev_unlock(mddev
);
2638 static void rdev_free(struct kobject
*ko
)
2640 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2643 static struct sysfs_ops rdev_sysfs_ops
= {
2644 .show
= rdev_attr_show
,
2645 .store
= rdev_attr_store
,
2647 static struct kobj_type rdev_ktype
= {
2648 .release
= rdev_free
,
2649 .sysfs_ops
= &rdev_sysfs_ops
,
2650 .default_attrs
= rdev_default_attrs
,
2654 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2656 * mark the device faulty if:
2658 * - the device is nonexistent (zero size)
2659 * - the device has no valid superblock
2661 * a faulty rdev _never_ has rdev->sb set.
2663 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2665 char b
[BDEVNAME_SIZE
];
2670 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2672 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2673 return ERR_PTR(-ENOMEM
);
2676 if ((err
= alloc_disk_sb(rdev
)))
2679 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2683 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2686 rdev
->saved_raid_disk
= -1;
2687 rdev
->raid_disk
= -1;
2689 rdev
->data_offset
= 0;
2690 rdev
->sb_events
= 0;
2691 rdev
->last_read_error
.tv_sec
= 0;
2692 rdev
->last_read_error
.tv_nsec
= 0;
2693 atomic_set(&rdev
->nr_pending
, 0);
2694 atomic_set(&rdev
->read_errors
, 0);
2695 atomic_set(&rdev
->corrected_errors
, 0);
2697 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2700 "md: %s has zero or unknown size, marking faulty!\n",
2701 bdevname(rdev
->bdev
,b
));
2706 if (super_format
>= 0) {
2707 err
= super_types
[super_format
].
2708 load_super(rdev
, NULL
, super_minor
);
2709 if (err
== -EINVAL
) {
2711 "md: %s does not have a valid v%d.%d "
2712 "superblock, not importing!\n",
2713 bdevname(rdev
->bdev
,b
),
2714 super_format
, super_minor
);
2719 "md: could not read %s's sb, not importing!\n",
2720 bdevname(rdev
->bdev
,b
));
2725 INIT_LIST_HEAD(&rdev
->same_set
);
2726 init_waitqueue_head(&rdev
->blocked_wait
);
2731 if (rdev
->sb_page
) {
2737 return ERR_PTR(err
);
2741 * Check a full RAID array for plausibility
2745 static void analyze_sbs(mddev_t
* mddev
)
2748 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2749 char b
[BDEVNAME_SIZE
];
2752 rdev_for_each(rdev
, tmp
, mddev
)
2753 switch (super_types
[mddev
->major_version
].
2754 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2762 "md: fatal superblock inconsistency in %s"
2763 " -- removing from array\n",
2764 bdevname(rdev
->bdev
,b
));
2765 kick_rdev_from_array(rdev
);
2769 super_types
[mddev
->major_version
].
2770 validate_super(mddev
, freshest
);
2773 rdev_for_each(rdev
, tmp
, mddev
) {
2774 if (rdev
->desc_nr
>= mddev
->max_disks
||
2775 i
> mddev
->max_disks
) {
2777 "md: %s: %s: only %d devices permitted\n",
2778 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2780 kick_rdev_from_array(rdev
);
2783 if (rdev
!= freshest
)
2784 if (super_types
[mddev
->major_version
].
2785 validate_super(mddev
, rdev
)) {
2786 printk(KERN_WARNING
"md: kicking non-fresh %s"
2788 bdevname(rdev
->bdev
,b
));
2789 kick_rdev_from_array(rdev
);
2792 if (mddev
->level
== LEVEL_MULTIPATH
) {
2793 rdev
->desc_nr
= i
++;
2794 rdev
->raid_disk
= rdev
->desc_nr
;
2795 set_bit(In_sync
, &rdev
->flags
);
2796 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2797 rdev
->raid_disk
= -1;
2798 clear_bit(In_sync
, &rdev
->flags
);
2803 /* Read a fixed-point number.
2804 * Numbers in sysfs attributes should be in "standard" units where
2805 * possible, so time should be in seconds.
2806 * However we internally use a a much smaller unit such as
2807 * milliseconds or jiffies.
2808 * This function takes a decimal number with a possible fractional
2809 * component, and produces an integer which is the result of
2810 * multiplying that number by 10^'scale'.
2811 * all without any floating-point arithmetic.
2813 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
2815 unsigned long result
= 0;
2817 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
2820 else if (decimals
< scale
) {
2823 result
= result
* 10 + value
;
2835 while (decimals
< scale
) {
2844 static void md_safemode_timeout(unsigned long data
);
2847 safe_delay_show(mddev_t
*mddev
, char *page
)
2849 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2850 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2853 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2857 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
2860 mddev
->safemode_delay
= 0;
2862 unsigned long old_delay
= mddev
->safemode_delay
;
2863 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2864 if (mddev
->safemode_delay
== 0)
2865 mddev
->safemode_delay
= 1;
2866 if (mddev
->safemode_delay
< old_delay
)
2867 md_safemode_timeout((unsigned long)mddev
);
2871 static struct md_sysfs_entry md_safe_delay
=
2872 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2875 level_show(mddev_t
*mddev
, char *page
)
2877 struct mdk_personality
*p
= mddev
->pers
;
2879 return sprintf(page
, "%s\n", p
->name
);
2880 else if (mddev
->clevel
[0])
2881 return sprintf(page
, "%s\n", mddev
->clevel
);
2882 else if (mddev
->level
!= LEVEL_NONE
)
2883 return sprintf(page
, "%d\n", mddev
->level
);
2889 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2893 struct mdk_personality
*pers
;
2897 if (mddev
->pers
== NULL
) {
2900 if (len
>= sizeof(mddev
->clevel
))
2902 strncpy(mddev
->clevel
, buf
, len
);
2903 if (mddev
->clevel
[len
-1] == '\n')
2905 mddev
->clevel
[len
] = 0;
2906 mddev
->level
= LEVEL_NONE
;
2910 /* request to change the personality. Need to ensure:
2911 * - array is not engaged in resync/recovery/reshape
2912 * - old personality can be suspended
2913 * - new personality will access other array.
2916 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2919 if (!mddev
->pers
->quiesce
) {
2920 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2921 mdname(mddev
), mddev
->pers
->name
);
2925 /* Now find the new personality */
2926 if (len
== 0 || len
>= sizeof(level
))
2928 strncpy(level
, buf
, len
);
2929 if (level
[len
-1] == '\n')
2933 request_module("md-%s", level
);
2934 spin_lock(&pers_lock
);
2935 pers
= find_pers(LEVEL_NONE
, level
);
2936 if (!pers
|| !try_module_get(pers
->owner
)) {
2937 spin_unlock(&pers_lock
);
2938 printk(KERN_WARNING
"md: personality %s not loaded\n", level
);
2941 spin_unlock(&pers_lock
);
2943 if (pers
== mddev
->pers
) {
2944 /* Nothing to do! */
2945 module_put(pers
->owner
);
2948 if (!pers
->takeover
) {
2949 module_put(pers
->owner
);
2950 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2951 mdname(mddev
), level
);
2955 /* ->takeover must set new_* and/or delta_disks
2956 * if it succeeds, and may set them when it fails.
2958 priv
= pers
->takeover(mddev
);
2960 mddev
->new_level
= mddev
->level
;
2961 mddev
->new_layout
= mddev
->layout
;
2962 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2963 mddev
->raid_disks
-= mddev
->delta_disks
;
2964 mddev
->delta_disks
= 0;
2965 module_put(pers
->owner
);
2966 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
2967 mdname(mddev
), level
);
2968 return PTR_ERR(priv
);
2971 /* Looks like we have a winner */
2972 mddev_suspend(mddev
);
2973 mddev
->pers
->stop(mddev
);
2974 module_put(mddev
->pers
->owner
);
2975 /* Invalidate devices that are now superfluous */
2976 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2977 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2978 rdev
->raid_disk
= -1;
2979 clear_bit(In_sync
, &rdev
->flags
);
2982 mddev
->private = priv
;
2983 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2984 mddev
->level
= mddev
->new_level
;
2985 mddev
->layout
= mddev
->new_layout
;
2986 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
2987 mddev
->delta_disks
= 0;
2989 mddev_resume(mddev
);
2990 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2991 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2992 md_wakeup_thread(mddev
->thread
);
2996 static struct md_sysfs_entry md_level
=
2997 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3001 layout_show(mddev_t
*mddev
, char *page
)
3003 /* just a number, not meaningful for all levels */
3004 if (mddev
->reshape_position
!= MaxSector
&&
3005 mddev
->layout
!= mddev
->new_layout
)
3006 return sprintf(page
, "%d (%d)\n",
3007 mddev
->new_layout
, mddev
->layout
);
3008 return sprintf(page
, "%d\n", mddev
->layout
);
3012 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3015 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3017 if (!*buf
|| (*e
&& *e
!= '\n'))
3022 if (mddev
->pers
->check_reshape
== NULL
)
3024 mddev
->new_layout
= n
;
3025 err
= mddev
->pers
->check_reshape(mddev
);
3027 mddev
->new_layout
= mddev
->layout
;
3031 mddev
->new_layout
= n
;
3032 if (mddev
->reshape_position
== MaxSector
)
3037 static struct md_sysfs_entry md_layout
=
3038 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3042 raid_disks_show(mddev_t
*mddev
, char *page
)
3044 if (mddev
->raid_disks
== 0)
3046 if (mddev
->reshape_position
!= MaxSector
&&
3047 mddev
->delta_disks
!= 0)
3048 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3049 mddev
->raid_disks
- mddev
->delta_disks
);
3050 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3053 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3056 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3060 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3062 if (!*buf
|| (*e
&& *e
!= '\n'))
3066 rv
= update_raid_disks(mddev
, n
);
3067 else if (mddev
->reshape_position
!= MaxSector
) {
3068 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3069 mddev
->delta_disks
= n
- olddisks
;
3070 mddev
->raid_disks
= n
;
3072 mddev
->raid_disks
= n
;
3073 return rv
? rv
: len
;
3075 static struct md_sysfs_entry md_raid_disks
=
3076 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3079 chunk_size_show(mddev_t
*mddev
, char *page
)
3081 if (mddev
->reshape_position
!= MaxSector
&&
3082 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3083 return sprintf(page
, "%d (%d)\n",
3084 mddev
->new_chunk_sectors
<< 9,
3085 mddev
->chunk_sectors
<< 9);
3086 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3090 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3093 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3095 if (!*buf
|| (*e
&& *e
!= '\n'))
3100 if (mddev
->pers
->check_reshape
== NULL
)
3102 mddev
->new_chunk_sectors
= n
>> 9;
3103 err
= mddev
->pers
->check_reshape(mddev
);
3105 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3109 mddev
->new_chunk_sectors
= n
>> 9;
3110 if (mddev
->reshape_position
== MaxSector
)
3111 mddev
->chunk_sectors
= n
>> 9;
3115 static struct md_sysfs_entry md_chunk_size
=
3116 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3119 resync_start_show(mddev_t
*mddev
, char *page
)
3121 if (mddev
->recovery_cp
== MaxSector
)
3122 return sprintf(page
, "none\n");
3123 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3127 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3130 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3134 if (cmd_match(buf
, "none"))
3136 else if (!*buf
|| (*e
&& *e
!= '\n'))
3139 mddev
->recovery_cp
= n
;
3142 static struct md_sysfs_entry md_resync_start
=
3143 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3146 * The array state can be:
3149 * No devices, no size, no level
3150 * Equivalent to STOP_ARRAY ioctl
3152 * May have some settings, but array is not active
3153 * all IO results in error
3154 * When written, doesn't tear down array, but just stops it
3155 * suspended (not supported yet)
3156 * All IO requests will block. The array can be reconfigured.
3157 * Writing this, if accepted, will block until array is quiescent
3159 * no resync can happen. no superblocks get written.
3160 * write requests fail
3162 * like readonly, but behaves like 'clean' on a write request.
3164 * clean - no pending writes, but otherwise active.
3165 * When written to inactive array, starts without resync
3166 * If a write request arrives then
3167 * if metadata is known, mark 'dirty' and switch to 'active'.
3168 * if not known, block and switch to write-pending
3169 * If written to an active array that has pending writes, then fails.
3171 * fully active: IO and resync can be happening.
3172 * When written to inactive array, starts with resync
3175 * clean, but writes are blocked waiting for 'active' to be written.
3178 * like active, but no writes have been seen for a while (100msec).
3181 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3182 write_pending
, active_idle
, bad_word
};
3183 static char *array_states
[] = {
3184 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3185 "write-pending", "active-idle", NULL
};
3187 static int match_word(const char *word
, char **list
)
3190 for (n
=0; list
[n
]; n
++)
3191 if (cmd_match(word
, list
[n
]))
3197 array_state_show(mddev_t
*mddev
, char *page
)
3199 enum array_state st
= inactive
;
3212 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3214 else if (mddev
->safemode
)
3220 if (list_empty(&mddev
->disks
) &&
3221 mddev
->raid_disks
== 0 &&
3222 mddev
->dev_sectors
== 0)
3227 return sprintf(page
, "%s\n", array_states
[st
]);
3230 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3231 static int do_md_run(mddev_t
* mddev
);
3232 static int restart_array(mddev_t
*mddev
);
3235 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3238 enum array_state st
= match_word(buf
, array_states
);
3243 /* stopping an active array */
3244 if (atomic_read(&mddev
->openers
) > 0)
3246 err
= do_md_stop(mddev
, 0, 0);
3249 /* stopping an active array */
3251 if (atomic_read(&mddev
->openers
) > 0)
3253 err
= do_md_stop(mddev
, 2, 0);
3255 err
= 0; /* already inactive */
3258 break; /* not supported yet */
3261 err
= do_md_stop(mddev
, 1, 0);
3264 set_disk_ro(mddev
->gendisk
, 1);
3265 err
= do_md_run(mddev
);
3271 err
= do_md_stop(mddev
, 1, 0);
3272 else if (mddev
->ro
== 1)
3273 err
= restart_array(mddev
);
3276 set_disk_ro(mddev
->gendisk
, 0);
3280 err
= do_md_run(mddev
);
3285 restart_array(mddev
);
3286 spin_lock_irq(&mddev
->write_lock
);
3287 if (atomic_read(&mddev
->writes_pending
) == 0) {
3288 if (mddev
->in_sync
== 0) {
3290 if (mddev
->safemode
== 1)
3291 mddev
->safemode
= 0;
3292 if (mddev
->persistent
)
3293 set_bit(MD_CHANGE_CLEAN
,
3299 spin_unlock_irq(&mddev
->write_lock
);
3305 restart_array(mddev
);
3306 if (mddev
->external
)
3307 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3308 wake_up(&mddev
->sb_wait
);
3312 set_disk_ro(mddev
->gendisk
, 0);
3313 err
= do_md_run(mddev
);
3318 /* these cannot be set */
3324 sysfs_notify_dirent(mddev
->sysfs_state
);
3328 static struct md_sysfs_entry md_array_state
=
3329 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3332 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3333 return sprintf(page
, "%d\n",
3334 atomic_read(&mddev
->max_corr_read_errors
));
3338 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3341 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3343 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3344 atomic_set(&mddev
->max_corr_read_errors
, n
);
3350 static struct md_sysfs_entry max_corr_read_errors
=
3351 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3352 max_corrected_read_errors_store
);
3355 null_show(mddev_t
*mddev
, char *page
)
3361 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3363 /* buf must be %d:%d\n? giving major and minor numbers */
3364 /* The new device is added to the array.
3365 * If the array has a persistent superblock, we read the
3366 * superblock to initialise info and check validity.
3367 * Otherwise, only checking done is that in bind_rdev_to_array,
3368 * which mainly checks size.
3371 int major
= simple_strtoul(buf
, &e
, 10);
3377 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3379 minor
= simple_strtoul(e
+1, &e
, 10);
3380 if (*e
&& *e
!= '\n')
3382 dev
= MKDEV(major
, minor
);
3383 if (major
!= MAJOR(dev
) ||
3384 minor
!= MINOR(dev
))
3388 if (mddev
->persistent
) {
3389 rdev
= md_import_device(dev
, mddev
->major_version
,
3390 mddev
->minor_version
);
3391 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3392 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3393 mdk_rdev_t
, same_set
);
3394 err
= super_types
[mddev
->major_version
]
3395 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3399 } else if (mddev
->external
)
3400 rdev
= md_import_device(dev
, -2, -1);
3402 rdev
= md_import_device(dev
, -1, -1);
3405 return PTR_ERR(rdev
);
3406 err
= bind_rdev_to_array(rdev
, mddev
);
3410 return err
? err
: len
;
3413 static struct md_sysfs_entry md_new_device
=
3414 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3417 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3420 unsigned long chunk
, end_chunk
;
3424 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3426 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3427 if (buf
== end
) break;
3428 if (*end
== '-') { /* range */
3430 end_chunk
= simple_strtoul(buf
, &end
, 0);
3431 if (buf
== end
) break;
3433 if (*end
&& !isspace(*end
)) break;
3434 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3435 buf
= skip_spaces(end
);
3437 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3442 static struct md_sysfs_entry md_bitmap
=
3443 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3446 size_show(mddev_t
*mddev
, char *page
)
3448 return sprintf(page
, "%llu\n",
3449 (unsigned long long)mddev
->dev_sectors
/ 2);
3452 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3455 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3457 /* If array is inactive, we can reduce the component size, but
3458 * not increase it (except from 0).
3459 * If array is active, we can try an on-line resize
3462 int err
= strict_blocks_to_sectors(buf
, §ors
);
3467 err
= update_size(mddev
, sectors
);
3468 md_update_sb(mddev
, 1);
3470 if (mddev
->dev_sectors
== 0 ||
3471 mddev
->dev_sectors
> sectors
)
3472 mddev
->dev_sectors
= sectors
;
3476 return err
? err
: len
;
3479 static struct md_sysfs_entry md_size
=
3480 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3485 * 'none' for arrays with no metadata (good luck...)
3486 * 'external' for arrays with externally managed metadata,
3487 * or N.M for internally known formats
3490 metadata_show(mddev_t
*mddev
, char *page
)
3492 if (mddev
->persistent
)
3493 return sprintf(page
, "%d.%d\n",
3494 mddev
->major_version
, mddev
->minor_version
);
3495 else if (mddev
->external
)
3496 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3498 return sprintf(page
, "none\n");
3502 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3506 /* Changing the details of 'external' metadata is
3507 * always permitted. Otherwise there must be
3508 * no devices attached to the array.
3510 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3512 else if (!list_empty(&mddev
->disks
))
3515 if (cmd_match(buf
, "none")) {
3516 mddev
->persistent
= 0;
3517 mddev
->external
= 0;
3518 mddev
->major_version
= 0;
3519 mddev
->minor_version
= 90;
3522 if (strncmp(buf
, "external:", 9) == 0) {
3523 size_t namelen
= len
-9;
3524 if (namelen
>= sizeof(mddev
->metadata_type
))
3525 namelen
= sizeof(mddev
->metadata_type
)-1;
3526 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3527 mddev
->metadata_type
[namelen
] = 0;
3528 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3529 mddev
->metadata_type
[--namelen
] = 0;
3530 mddev
->persistent
= 0;
3531 mddev
->external
= 1;
3532 mddev
->major_version
= 0;
3533 mddev
->minor_version
= 90;
3536 major
= simple_strtoul(buf
, &e
, 10);
3537 if (e
==buf
|| *e
!= '.')
3540 minor
= simple_strtoul(buf
, &e
, 10);
3541 if (e
==buf
|| (*e
&& *e
!= '\n') )
3543 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3545 mddev
->major_version
= major
;
3546 mddev
->minor_version
= minor
;
3547 mddev
->persistent
= 1;
3548 mddev
->external
= 0;
3552 static struct md_sysfs_entry md_metadata
=
3553 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3556 action_show(mddev_t
*mddev
, char *page
)
3558 char *type
= "idle";
3559 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3561 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3562 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3563 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3565 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3566 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3568 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3572 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3575 return sprintf(page
, "%s\n", type
);
3579 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3581 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3584 if (cmd_match(page
, "frozen"))
3585 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3587 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3589 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3590 if (mddev
->sync_thread
) {
3591 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3592 md_unregister_thread(mddev
->sync_thread
);
3593 mddev
->sync_thread
= NULL
;
3594 mddev
->recovery
= 0;
3596 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3597 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3599 else if (cmd_match(page
, "resync"))
3600 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3601 else if (cmd_match(page
, "recover")) {
3602 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3603 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3604 } else if (cmd_match(page
, "reshape")) {
3606 if (mddev
->pers
->start_reshape
== NULL
)
3608 err
= mddev
->pers
->start_reshape(mddev
);
3611 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3613 if (cmd_match(page
, "check"))
3614 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3615 else if (!cmd_match(page
, "repair"))
3617 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3618 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3620 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3621 md_wakeup_thread(mddev
->thread
);
3622 sysfs_notify_dirent(mddev
->sysfs_action
);
3627 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3629 return sprintf(page
, "%llu\n",
3630 (unsigned long long) mddev
->resync_mismatches
);
3633 static struct md_sysfs_entry md_scan_mode
=
3634 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3637 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3640 sync_min_show(mddev_t
*mddev
, char *page
)
3642 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3643 mddev
->sync_speed_min
? "local": "system");
3647 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3651 if (strncmp(buf
, "system", 6)==0) {
3652 mddev
->sync_speed_min
= 0;
3655 min
= simple_strtoul(buf
, &e
, 10);
3656 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3658 mddev
->sync_speed_min
= min
;
3662 static struct md_sysfs_entry md_sync_min
=
3663 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3666 sync_max_show(mddev_t
*mddev
, char *page
)
3668 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3669 mddev
->sync_speed_max
? "local": "system");
3673 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3677 if (strncmp(buf
, "system", 6)==0) {
3678 mddev
->sync_speed_max
= 0;
3681 max
= simple_strtoul(buf
, &e
, 10);
3682 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3684 mddev
->sync_speed_max
= max
;
3688 static struct md_sysfs_entry md_sync_max
=
3689 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3692 degraded_show(mddev_t
*mddev
, char *page
)
3694 return sprintf(page
, "%d\n", mddev
->degraded
);
3696 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3699 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3701 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3705 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3709 if (strict_strtol(buf
, 10, &n
))
3712 if (n
!= 0 && n
!= 1)
3715 mddev
->parallel_resync
= n
;
3717 if (mddev
->sync_thread
)
3718 wake_up(&resync_wait
);
3723 /* force parallel resync, even with shared block devices */
3724 static struct md_sysfs_entry md_sync_force_parallel
=
3725 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3726 sync_force_parallel_show
, sync_force_parallel_store
);
3729 sync_speed_show(mddev_t
*mddev
, char *page
)
3731 unsigned long resync
, dt
, db
;
3732 if (mddev
->curr_resync
== 0)
3733 return sprintf(page
, "none\n");
3734 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3735 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3737 db
= resync
- mddev
->resync_mark_cnt
;
3738 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3741 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3744 sync_completed_show(mddev_t
*mddev
, char *page
)
3746 unsigned long max_sectors
, resync
;
3748 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3749 return sprintf(page
, "none\n");
3751 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3752 max_sectors
= mddev
->resync_max_sectors
;
3754 max_sectors
= mddev
->dev_sectors
;
3756 resync
= mddev
->curr_resync_completed
;
3757 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3760 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3763 min_sync_show(mddev_t
*mddev
, char *page
)
3765 return sprintf(page
, "%llu\n",
3766 (unsigned long long)mddev
->resync_min
);
3769 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3771 unsigned long long min
;
3772 if (strict_strtoull(buf
, 10, &min
))
3774 if (min
> mddev
->resync_max
)
3776 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3779 /* Must be a multiple of chunk_size */
3780 if (mddev
->chunk_sectors
) {
3781 sector_t temp
= min
;
3782 if (sector_div(temp
, mddev
->chunk_sectors
))
3785 mddev
->resync_min
= min
;
3790 static struct md_sysfs_entry md_min_sync
=
3791 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3794 max_sync_show(mddev_t
*mddev
, char *page
)
3796 if (mddev
->resync_max
== MaxSector
)
3797 return sprintf(page
, "max\n");
3799 return sprintf(page
, "%llu\n",
3800 (unsigned long long)mddev
->resync_max
);
3803 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3805 if (strncmp(buf
, "max", 3) == 0)
3806 mddev
->resync_max
= MaxSector
;
3808 unsigned long long max
;
3809 if (strict_strtoull(buf
, 10, &max
))
3811 if (max
< mddev
->resync_min
)
3813 if (max
< mddev
->resync_max
&&
3815 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3818 /* Must be a multiple of chunk_size */
3819 if (mddev
->chunk_sectors
) {
3820 sector_t temp
= max
;
3821 if (sector_div(temp
, mddev
->chunk_sectors
))
3824 mddev
->resync_max
= max
;
3826 wake_up(&mddev
->recovery_wait
);
3830 static struct md_sysfs_entry md_max_sync
=
3831 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3834 suspend_lo_show(mddev_t
*mddev
, char *page
)
3836 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3840 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3843 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3845 if (mddev
->pers
== NULL
||
3846 mddev
->pers
->quiesce
== NULL
)
3848 if (buf
== e
|| (*e
&& *e
!= '\n'))
3850 if (new >= mddev
->suspend_hi
||
3851 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3852 mddev
->suspend_lo
= new;
3853 mddev
->pers
->quiesce(mddev
, 2);
3858 static struct md_sysfs_entry md_suspend_lo
=
3859 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3863 suspend_hi_show(mddev_t
*mddev
, char *page
)
3865 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3869 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3872 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3874 if (mddev
->pers
== NULL
||
3875 mddev
->pers
->quiesce
== NULL
)
3877 if (buf
== e
|| (*e
&& *e
!= '\n'))
3879 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3880 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3881 mddev
->suspend_hi
= new;
3882 mddev
->pers
->quiesce(mddev
, 1);
3883 mddev
->pers
->quiesce(mddev
, 0);
3888 static struct md_sysfs_entry md_suspend_hi
=
3889 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3892 reshape_position_show(mddev_t
*mddev
, char *page
)
3894 if (mddev
->reshape_position
!= MaxSector
)
3895 return sprintf(page
, "%llu\n",
3896 (unsigned long long)mddev
->reshape_position
);
3897 strcpy(page
, "none\n");
3902 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3905 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3908 if (buf
== e
|| (*e
&& *e
!= '\n'))
3910 mddev
->reshape_position
= new;
3911 mddev
->delta_disks
= 0;
3912 mddev
->new_level
= mddev
->level
;
3913 mddev
->new_layout
= mddev
->layout
;
3914 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3918 static struct md_sysfs_entry md_reshape_position
=
3919 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3920 reshape_position_store
);
3923 array_size_show(mddev_t
*mddev
, char *page
)
3925 if (mddev
->external_size
)
3926 return sprintf(page
, "%llu\n",
3927 (unsigned long long)mddev
->array_sectors
/2);
3929 return sprintf(page
, "default\n");
3933 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3937 if (strncmp(buf
, "default", 7) == 0) {
3939 sectors
= mddev
->pers
->size(mddev
, 0, 0);
3941 sectors
= mddev
->array_sectors
;
3943 mddev
->external_size
= 0;
3945 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3947 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
3950 mddev
->external_size
= 1;
3953 mddev
->array_sectors
= sectors
;
3954 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
3956 revalidate_disk(mddev
->gendisk
);
3961 static struct md_sysfs_entry md_array_size
=
3962 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
3965 static struct attribute
*md_default_attrs
[] = {
3968 &md_raid_disks
.attr
,
3969 &md_chunk_size
.attr
,
3971 &md_resync_start
.attr
,
3973 &md_new_device
.attr
,
3974 &md_safe_delay
.attr
,
3975 &md_array_state
.attr
,
3976 &md_reshape_position
.attr
,
3977 &md_array_size
.attr
,
3978 &max_corr_read_errors
.attr
,
3982 static struct attribute
*md_redundancy_attrs
[] = {
3984 &md_mismatches
.attr
,
3987 &md_sync_speed
.attr
,
3988 &md_sync_force_parallel
.attr
,
3989 &md_sync_completed
.attr
,
3992 &md_suspend_lo
.attr
,
3993 &md_suspend_hi
.attr
,
3998 static struct attribute_group md_redundancy_group
= {
4000 .attrs
= md_redundancy_attrs
,
4005 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4007 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4008 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4013 rv
= mddev_lock(mddev
);
4015 rv
= entry
->show(mddev
, page
);
4016 mddev_unlock(mddev
);
4022 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4023 const char *page
, size_t length
)
4025 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4026 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4031 if (!capable(CAP_SYS_ADMIN
))
4033 rv
= mddev_lock(mddev
);
4034 if (mddev
->hold_active
== UNTIL_IOCTL
)
4035 mddev
->hold_active
= 0;
4037 rv
= entry
->store(mddev
, page
, length
);
4038 mddev_unlock(mddev
);
4043 static void md_free(struct kobject
*ko
)
4045 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4047 if (mddev
->sysfs_state
)
4048 sysfs_put(mddev
->sysfs_state
);
4050 if (mddev
->gendisk
) {
4051 del_gendisk(mddev
->gendisk
);
4052 put_disk(mddev
->gendisk
);
4055 blk_cleanup_queue(mddev
->queue
);
4060 static struct sysfs_ops md_sysfs_ops
= {
4061 .show
= md_attr_show
,
4062 .store
= md_attr_store
,
4064 static struct kobj_type md_ktype
= {
4066 .sysfs_ops
= &md_sysfs_ops
,
4067 .default_attrs
= md_default_attrs
,
4072 static void mddev_delayed_delete(struct work_struct
*ws
)
4074 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4076 if (mddev
->private == &md_redundancy_group
) {
4077 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
4078 if (mddev
->sysfs_action
)
4079 sysfs_put(mddev
->sysfs_action
);
4080 mddev
->sysfs_action
= NULL
;
4081 mddev
->private = NULL
;
4083 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4084 kobject_del(&mddev
->kobj
);
4085 kobject_put(&mddev
->kobj
);
4088 static int md_alloc(dev_t dev
, char *name
)
4090 static DEFINE_MUTEX(disks_mutex
);
4091 mddev_t
*mddev
= mddev_find(dev
);
4092 struct gendisk
*disk
;
4101 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4102 shift
= partitioned
? MdpMinorShift
: 0;
4103 unit
= MINOR(mddev
->unit
) >> shift
;
4105 /* wait for any previous instance if this device
4106 * to be completed removed (mddev_delayed_delete).
4108 flush_scheduled_work();
4110 mutex_lock(&disks_mutex
);
4116 /* Need to ensure that 'name' is not a duplicate.
4119 spin_lock(&all_mddevs_lock
);
4121 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4122 if (mddev2
->gendisk
&&
4123 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4124 spin_unlock(&all_mddevs_lock
);
4127 spin_unlock(&all_mddevs_lock
);
4131 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4134 mddev
->queue
->queuedata
= mddev
;
4136 /* Can be unlocked because the queue is new: no concurrency */
4137 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4139 blk_queue_make_request(mddev
->queue
, md_make_request
);
4141 disk
= alloc_disk(1 << shift
);
4143 blk_cleanup_queue(mddev
->queue
);
4144 mddev
->queue
= NULL
;
4147 disk
->major
= MAJOR(mddev
->unit
);
4148 disk
->first_minor
= unit
<< shift
;
4150 strcpy(disk
->disk_name
, name
);
4151 else if (partitioned
)
4152 sprintf(disk
->disk_name
, "md_d%d", unit
);
4154 sprintf(disk
->disk_name
, "md%d", unit
);
4155 disk
->fops
= &md_fops
;
4156 disk
->private_data
= mddev
;
4157 disk
->queue
= mddev
->queue
;
4158 /* Allow extended partitions. This makes the
4159 * 'mdp' device redundant, but we can't really
4162 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4164 mddev
->gendisk
= disk
;
4165 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4166 &disk_to_dev(disk
)->kobj
, "%s", "md");
4168 /* This isn't possible, but as kobject_init_and_add is marked
4169 * __must_check, we must do something with the result
4171 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4175 if (sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4176 printk(KERN_DEBUG
"pointless warning\n");
4178 mutex_unlock(&disks_mutex
);
4180 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4181 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4187 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4189 md_alloc(dev
, NULL
);
4193 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4195 /* val must be "md_*" where * is not all digits.
4196 * We allocate an array with a large free minor number, and
4197 * set the name to val. val must not already be an active name.
4199 int len
= strlen(val
);
4200 char buf
[DISK_NAME_LEN
];
4202 while (len
&& val
[len
-1] == '\n')
4204 if (len
>= DISK_NAME_LEN
)
4206 strlcpy(buf
, val
, len
+1);
4207 if (strncmp(buf
, "md_", 3) != 0)
4209 return md_alloc(0, buf
);
4212 static void md_safemode_timeout(unsigned long data
)
4214 mddev_t
*mddev
= (mddev_t
*) data
;
4216 if (!atomic_read(&mddev
->writes_pending
)) {
4217 mddev
->safemode
= 1;
4218 if (mddev
->external
)
4219 sysfs_notify_dirent(mddev
->sysfs_state
);
4221 md_wakeup_thread(mddev
->thread
);
4224 static int start_dirty_degraded
;
4226 static int do_md_run(mddev_t
* mddev
)
4230 struct gendisk
*disk
;
4231 struct mdk_personality
*pers
;
4233 if (list_empty(&mddev
->disks
))
4234 /* cannot run an array with no devices.. */
4241 * Analyze all RAID superblock(s)
4243 if (!mddev
->raid_disks
) {
4244 if (!mddev
->persistent
)
4249 if (mddev
->level
!= LEVEL_NONE
)
4250 request_module("md-level-%d", mddev
->level
);
4251 else if (mddev
->clevel
[0])
4252 request_module("md-%s", mddev
->clevel
);
4255 * Drop all container device buffers, from now on
4256 * the only valid external interface is through the md
4259 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4260 if (test_bit(Faulty
, &rdev
->flags
))
4262 sync_blockdev(rdev
->bdev
);
4263 invalidate_bdev(rdev
->bdev
);
4265 /* perform some consistency tests on the device.
4266 * We don't want the data to overlap the metadata,
4267 * Internal Bitmap issues have been handled elsewhere.
4269 if (rdev
->data_offset
< rdev
->sb_start
) {
4270 if (mddev
->dev_sectors
&&
4271 rdev
->data_offset
+ mddev
->dev_sectors
4273 printk("md: %s: data overlaps metadata\n",
4278 if (rdev
->sb_start
+ rdev
->sb_size
/512
4279 > rdev
->data_offset
) {
4280 printk("md: %s: metadata overlaps data\n",
4285 sysfs_notify_dirent(rdev
->sysfs_state
);
4288 md_probe(mddev
->unit
, NULL
, NULL
);
4289 disk
= mddev
->gendisk
;
4293 spin_lock(&pers_lock
);
4294 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4295 if (!pers
|| !try_module_get(pers
->owner
)) {
4296 spin_unlock(&pers_lock
);
4297 if (mddev
->level
!= LEVEL_NONE
)
4298 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4301 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4306 spin_unlock(&pers_lock
);
4307 if (mddev
->level
!= pers
->level
) {
4308 mddev
->level
= pers
->level
;
4309 mddev
->new_level
= pers
->level
;
4311 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4313 if (mddev
->reshape_position
!= MaxSector
&&
4314 pers
->start_reshape
== NULL
) {
4315 /* This personality cannot handle reshaping... */
4317 module_put(pers
->owner
);
4321 if (pers
->sync_request
) {
4322 /* Warn if this is a potentially silly
4325 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4329 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4330 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4332 rdev
->bdev
->bd_contains
==
4333 rdev2
->bdev
->bd_contains
) {
4335 "%s: WARNING: %s appears to be"
4336 " on the same physical disk as"
4339 bdevname(rdev
->bdev
,b
),
4340 bdevname(rdev2
->bdev
,b2
));
4347 "True protection against single-disk"
4348 " failure might be compromised.\n");
4351 mddev
->recovery
= 0;
4352 /* may be over-ridden by personality */
4353 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4355 mddev
->barriers_work
= 1;
4356 mddev
->ok_start_degraded
= start_dirty_degraded
;
4359 mddev
->ro
= 2; /* read-only, but switch on first write */
4361 err
= mddev
->pers
->run(mddev
);
4363 printk(KERN_ERR
"md: pers->run() failed ...\n");
4364 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4365 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4366 " but 'external_size' not in effect?\n", __func__
);
4368 "md: invalid array_size %llu > default size %llu\n",
4369 (unsigned long long)mddev
->array_sectors
/ 2,
4370 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4372 mddev
->pers
->stop(mddev
);
4374 if (err
== 0 && mddev
->pers
->sync_request
) {
4375 err
= bitmap_create(mddev
);
4377 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4378 mdname(mddev
), err
);
4379 mddev
->pers
->stop(mddev
);
4383 module_put(mddev
->pers
->owner
);
4385 bitmap_destroy(mddev
);
4388 if (mddev
->pers
->sync_request
) {
4389 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4391 "md: cannot register extra attributes for %s\n",
4393 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4394 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4397 atomic_set(&mddev
->writes_pending
,0);
4398 atomic_set(&mddev
->max_corr_read_errors
,
4399 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4400 mddev
->safemode
= 0;
4401 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4402 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4403 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4406 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4407 if (rdev
->raid_disk
>= 0) {
4409 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4410 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4411 printk("md: cannot register %s for %s\n",
4415 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4418 md_update_sb(mddev
, 0);
4420 set_capacity(disk
, mddev
->array_sectors
);
4422 /* If there is a partially-recovered drive we need to
4423 * start recovery here. If we leave it to md_check_recovery,
4424 * it will remove the drives and not do the right thing
4426 if (mddev
->degraded
&& !mddev
->sync_thread
) {
4428 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4429 if (rdev
->raid_disk
>= 0 &&
4430 !test_bit(In_sync
, &rdev
->flags
) &&
4431 !test_bit(Faulty
, &rdev
->flags
))
4432 /* complete an interrupted recovery */
4434 if (spares
&& mddev
->pers
->sync_request
) {
4435 mddev
->recovery
= 0;
4436 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4437 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4440 if (!mddev
->sync_thread
) {
4441 printk(KERN_ERR
"%s: could not start resync"
4444 /* leave the spares where they are, it shouldn't hurt */
4445 mddev
->recovery
= 0;
4449 md_wakeup_thread(mddev
->thread
);
4450 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4452 revalidate_disk(mddev
->gendisk
);
4454 md_new_event(mddev
);
4455 sysfs_notify_dirent(mddev
->sysfs_state
);
4456 if (mddev
->sysfs_action
)
4457 sysfs_notify_dirent(mddev
->sysfs_action
);
4458 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4459 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4463 static int restart_array(mddev_t
*mddev
)
4465 struct gendisk
*disk
= mddev
->gendisk
;
4467 /* Complain if it has no devices */
4468 if (list_empty(&mddev
->disks
))
4474 mddev
->safemode
= 0;
4476 set_disk_ro(disk
, 0);
4477 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4479 /* Kick recovery or resync if necessary */
4480 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4481 md_wakeup_thread(mddev
->thread
);
4482 md_wakeup_thread(mddev
->sync_thread
);
4483 sysfs_notify_dirent(mddev
->sysfs_state
);
4487 /* similar to deny_write_access, but accounts for our holding a reference
4488 * to the file ourselves */
4489 static int deny_bitmap_write_access(struct file
* file
)
4491 struct inode
*inode
= file
->f_mapping
->host
;
4493 spin_lock(&inode
->i_lock
);
4494 if (atomic_read(&inode
->i_writecount
) > 1) {
4495 spin_unlock(&inode
->i_lock
);
4498 atomic_set(&inode
->i_writecount
, -1);
4499 spin_unlock(&inode
->i_lock
);
4504 void restore_bitmap_write_access(struct file
*file
)
4506 struct inode
*inode
= file
->f_mapping
->host
;
4508 spin_lock(&inode
->i_lock
);
4509 atomic_set(&inode
->i_writecount
, 1);
4510 spin_unlock(&inode
->i_lock
);
4514 * 0 - completely stop and dis-assemble array
4515 * 1 - switch to readonly
4516 * 2 - stop but do not disassemble array
4518 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4521 struct gendisk
*disk
= mddev
->gendisk
;
4524 mutex_lock(&mddev
->open_mutex
);
4525 if (atomic_read(&mddev
->openers
) > is_open
) {
4526 printk("md: %s still in use.\n",mdname(mddev
));
4528 } else if (mddev
->pers
) {
4530 if (mddev
->sync_thread
) {
4531 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4532 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4533 md_unregister_thread(mddev
->sync_thread
);
4534 mddev
->sync_thread
= NULL
;
4537 del_timer_sync(&mddev
->safemode_timer
);
4540 case 1: /* readonly */
4546 case 0: /* disassemble */
4548 bitmap_flush(mddev
);
4549 md_super_wait(mddev
);
4551 set_disk_ro(disk
, 0);
4553 mddev
->pers
->stop(mddev
);
4554 mddev
->queue
->merge_bvec_fn
= NULL
;
4555 mddev
->queue
->unplug_fn
= NULL
;
4556 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4557 module_put(mddev
->pers
->owner
);
4558 if (mddev
->pers
->sync_request
)
4559 mddev
->private = &md_redundancy_group
;
4561 /* tell userspace to handle 'inactive' */
4562 sysfs_notify_dirent(mddev
->sysfs_state
);
4564 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4565 if (rdev
->raid_disk
>= 0) {
4567 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4568 sysfs_remove_link(&mddev
->kobj
, nm
);
4571 set_capacity(disk
, 0);
4577 if (!mddev
->in_sync
|| mddev
->flags
) {
4578 /* mark array as shutdown cleanly */
4580 md_update_sb(mddev
, 1);
4583 set_disk_ro(disk
, 1);
4584 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4588 mutex_unlock(&mddev
->open_mutex
);
4592 * Free resources if final stop
4596 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4598 bitmap_destroy(mddev
);
4599 if (mddev
->bitmap_info
.file
) {
4600 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4601 fput(mddev
->bitmap_info
.file
);
4602 mddev
->bitmap_info
.file
= NULL
;
4604 mddev
->bitmap_info
.offset
= 0;
4606 /* make sure all md_delayed_delete calls have finished */
4607 flush_scheduled_work();
4609 export_array(mddev
);
4611 mddev
->array_sectors
= 0;
4612 mddev
->external_size
= 0;
4613 mddev
->dev_sectors
= 0;
4614 mddev
->raid_disks
= 0;
4615 mddev
->recovery_cp
= 0;
4616 mddev
->resync_min
= 0;
4617 mddev
->resync_max
= MaxSector
;
4618 mddev
->reshape_position
= MaxSector
;
4619 mddev
->external
= 0;
4620 mddev
->persistent
= 0;
4621 mddev
->level
= LEVEL_NONE
;
4622 mddev
->clevel
[0] = 0;
4625 mddev
->metadata_type
[0] = 0;
4626 mddev
->chunk_sectors
= 0;
4627 mddev
->ctime
= mddev
->utime
= 0;
4629 mddev
->max_disks
= 0;
4631 mddev
->delta_disks
= 0;
4632 mddev
->new_level
= LEVEL_NONE
;
4633 mddev
->new_layout
= 0;
4634 mddev
->new_chunk_sectors
= 0;
4635 mddev
->curr_resync
= 0;
4636 mddev
->resync_mismatches
= 0;
4637 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4638 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4639 mddev
->recovery
= 0;
4642 mddev
->degraded
= 0;
4643 mddev
->barriers_work
= 0;
4644 mddev
->safemode
= 0;
4645 mddev
->bitmap_info
.offset
= 0;
4646 mddev
->bitmap_info
.default_offset
= 0;
4647 mddev
->bitmap_info
.chunksize
= 0;
4648 mddev
->bitmap_info
.daemon_sleep
= 0;
4649 mddev
->bitmap_info
.max_write_behind
= 0;
4650 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4651 if (mddev
->hold_active
== UNTIL_STOP
)
4652 mddev
->hold_active
= 0;
4654 } else if (mddev
->pers
)
4655 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4658 blk_integrity_unregister(disk
);
4659 md_new_event(mddev
);
4660 sysfs_notify_dirent(mddev
->sysfs_state
);
4665 static void autorun_array(mddev_t
*mddev
)
4670 if (list_empty(&mddev
->disks
))
4673 printk(KERN_INFO
"md: running: ");
4675 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4676 char b
[BDEVNAME_SIZE
];
4677 printk("<%s>", bdevname(rdev
->bdev
,b
));
4681 err
= do_md_run(mddev
);
4683 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4684 do_md_stop(mddev
, 0, 0);
4689 * lets try to run arrays based on all disks that have arrived
4690 * until now. (those are in pending_raid_disks)
4692 * the method: pick the first pending disk, collect all disks with
4693 * the same UUID, remove all from the pending list and put them into
4694 * the 'same_array' list. Then order this list based on superblock
4695 * update time (freshest comes first), kick out 'old' disks and
4696 * compare superblocks. If everything's fine then run it.
4698 * If "unit" is allocated, then bump its reference count
4700 static void autorun_devices(int part
)
4702 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4704 char b
[BDEVNAME_SIZE
];
4706 printk(KERN_INFO
"md: autorun ...\n");
4707 while (!list_empty(&pending_raid_disks
)) {
4710 LIST_HEAD(candidates
);
4711 rdev0
= list_entry(pending_raid_disks
.next
,
4712 mdk_rdev_t
, same_set
);
4714 printk(KERN_INFO
"md: considering %s ...\n",
4715 bdevname(rdev0
->bdev
,b
));
4716 INIT_LIST_HEAD(&candidates
);
4717 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4718 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4719 printk(KERN_INFO
"md: adding %s ...\n",
4720 bdevname(rdev
->bdev
,b
));
4721 list_move(&rdev
->same_set
, &candidates
);
4724 * now we have a set of devices, with all of them having
4725 * mostly sane superblocks. It's time to allocate the
4729 dev
= MKDEV(mdp_major
,
4730 rdev0
->preferred_minor
<< MdpMinorShift
);
4731 unit
= MINOR(dev
) >> MdpMinorShift
;
4733 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4736 if (rdev0
->preferred_minor
!= unit
) {
4737 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4738 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4742 md_probe(dev
, NULL
, NULL
);
4743 mddev
= mddev_find(dev
);
4744 if (!mddev
|| !mddev
->gendisk
) {
4748 "md: cannot allocate memory for md drive.\n");
4751 if (mddev_lock(mddev
))
4752 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4754 else if (mddev
->raid_disks
|| mddev
->major_version
4755 || !list_empty(&mddev
->disks
)) {
4757 "md: %s already running, cannot run %s\n",
4758 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4759 mddev_unlock(mddev
);
4761 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4762 mddev
->persistent
= 1;
4763 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4764 list_del_init(&rdev
->same_set
);
4765 if (bind_rdev_to_array(rdev
, mddev
))
4768 autorun_array(mddev
);
4769 mddev_unlock(mddev
);
4771 /* on success, candidates will be empty, on error
4774 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4775 list_del_init(&rdev
->same_set
);
4780 printk(KERN_INFO
"md: ... autorun DONE.\n");
4782 #endif /* !MODULE */
4784 static int get_version(void __user
* arg
)
4788 ver
.major
= MD_MAJOR_VERSION
;
4789 ver
.minor
= MD_MINOR_VERSION
;
4790 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4792 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4798 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4800 mdu_array_info_t info
;
4801 int nr
,working
,insync
,failed
,spare
;
4804 nr
=working
=insync
=failed
=spare
=0;
4805 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4807 if (test_bit(Faulty
, &rdev
->flags
))
4811 if (test_bit(In_sync
, &rdev
->flags
))
4818 info
.major_version
= mddev
->major_version
;
4819 info
.minor_version
= mddev
->minor_version
;
4820 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4821 info
.ctime
= mddev
->ctime
;
4822 info
.level
= mddev
->level
;
4823 info
.size
= mddev
->dev_sectors
/ 2;
4824 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4827 info
.raid_disks
= mddev
->raid_disks
;
4828 info
.md_minor
= mddev
->md_minor
;
4829 info
.not_persistent
= !mddev
->persistent
;
4831 info
.utime
= mddev
->utime
;
4834 info
.state
= (1<<MD_SB_CLEAN
);
4835 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
4836 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4837 info
.active_disks
= insync
;
4838 info
.working_disks
= working
;
4839 info
.failed_disks
= failed
;
4840 info
.spare_disks
= spare
;
4842 info
.layout
= mddev
->layout
;
4843 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4845 if (copy_to_user(arg
, &info
, sizeof(info
)))
4851 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4853 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4854 char *ptr
, *buf
= NULL
;
4857 if (md_allow_write(mddev
))
4858 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4860 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4865 /* bitmap disabled, zero the first byte and copy out */
4866 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4867 file
->pathname
[0] = '\0';
4871 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4875 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4879 strcpy(file
->pathname
, ptr
);
4883 if (copy_to_user(arg
, file
, sizeof(*file
)))
4891 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4893 mdu_disk_info_t info
;
4896 if (copy_from_user(&info
, arg
, sizeof(info
)))
4899 rdev
= find_rdev_nr(mddev
, info
.number
);
4901 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4902 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4903 info
.raid_disk
= rdev
->raid_disk
;
4905 if (test_bit(Faulty
, &rdev
->flags
))
4906 info
.state
|= (1<<MD_DISK_FAULTY
);
4907 else if (test_bit(In_sync
, &rdev
->flags
)) {
4908 info
.state
|= (1<<MD_DISK_ACTIVE
);
4909 info
.state
|= (1<<MD_DISK_SYNC
);
4911 if (test_bit(WriteMostly
, &rdev
->flags
))
4912 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4914 info
.major
= info
.minor
= 0;
4915 info
.raid_disk
= -1;
4916 info
.state
= (1<<MD_DISK_REMOVED
);
4919 if (copy_to_user(arg
, &info
, sizeof(info
)))
4925 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4927 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4929 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4931 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4934 if (!mddev
->raid_disks
) {
4936 /* expecting a device which has a superblock */
4937 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4940 "md: md_import_device returned %ld\n",
4942 return PTR_ERR(rdev
);
4944 if (!list_empty(&mddev
->disks
)) {
4945 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4946 mdk_rdev_t
, same_set
);
4947 err
= super_types
[mddev
->major_version
]
4948 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4951 "md: %s has different UUID to %s\n",
4952 bdevname(rdev
->bdev
,b
),
4953 bdevname(rdev0
->bdev
,b2
));
4958 err
= bind_rdev_to_array(rdev
, mddev
);
4965 * add_new_disk can be used once the array is assembled
4966 * to add "hot spares". They must already have a superblock
4971 if (!mddev
->pers
->hot_add_disk
) {
4973 "%s: personality does not support diskops!\n",
4977 if (mddev
->persistent
)
4978 rdev
= md_import_device(dev
, mddev
->major_version
,
4979 mddev
->minor_version
);
4981 rdev
= md_import_device(dev
, -1, -1);
4984 "md: md_import_device returned %ld\n",
4986 return PTR_ERR(rdev
);
4988 /* set save_raid_disk if appropriate */
4989 if (!mddev
->persistent
) {
4990 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4991 info
->raid_disk
< mddev
->raid_disks
)
4992 rdev
->raid_disk
= info
->raid_disk
;
4994 rdev
->raid_disk
= -1;
4996 super_types
[mddev
->major_version
].
4997 validate_super(mddev
, rdev
);
4998 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5000 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5001 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5002 set_bit(WriteMostly
, &rdev
->flags
);
5004 clear_bit(WriteMostly
, &rdev
->flags
);
5006 rdev
->raid_disk
= -1;
5007 err
= bind_rdev_to_array(rdev
, mddev
);
5008 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5009 /* If there is hot_add_disk but no hot_remove_disk
5010 * then added disks for geometry changes,
5011 * and should be added immediately.
5013 super_types
[mddev
->major_version
].
5014 validate_super(mddev
, rdev
);
5015 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5017 unbind_rdev_from_array(rdev
);
5022 sysfs_notify_dirent(rdev
->sysfs_state
);
5024 md_update_sb(mddev
, 1);
5025 if (mddev
->degraded
)
5026 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5027 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5028 md_wakeup_thread(mddev
->thread
);
5032 /* otherwise, add_new_disk is only allowed
5033 * for major_version==0 superblocks
5035 if (mddev
->major_version
!= 0) {
5036 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5041 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5043 rdev
= md_import_device(dev
, -1, 0);
5046 "md: error, md_import_device() returned %ld\n",
5048 return PTR_ERR(rdev
);
5050 rdev
->desc_nr
= info
->number
;
5051 if (info
->raid_disk
< mddev
->raid_disks
)
5052 rdev
->raid_disk
= info
->raid_disk
;
5054 rdev
->raid_disk
= -1;
5056 if (rdev
->raid_disk
< mddev
->raid_disks
)
5057 if (info
->state
& (1<<MD_DISK_SYNC
))
5058 set_bit(In_sync
, &rdev
->flags
);
5060 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5061 set_bit(WriteMostly
, &rdev
->flags
);
5063 if (!mddev
->persistent
) {
5064 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5065 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5067 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5068 rdev
->sectors
= rdev
->sb_start
;
5070 err
= bind_rdev_to_array(rdev
, mddev
);
5080 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5082 char b
[BDEVNAME_SIZE
];
5085 rdev
= find_rdev(mddev
, dev
);
5089 if (rdev
->raid_disk
>= 0)
5092 kick_rdev_from_array(rdev
);
5093 md_update_sb(mddev
, 1);
5094 md_new_event(mddev
);
5098 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5099 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5103 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5105 char b
[BDEVNAME_SIZE
];
5112 if (mddev
->major_version
!= 0) {
5113 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5114 " version-0 superblocks.\n",
5118 if (!mddev
->pers
->hot_add_disk
) {
5120 "%s: personality does not support diskops!\n",
5125 rdev
= md_import_device(dev
, -1, 0);
5128 "md: error, md_import_device() returned %ld\n",
5133 if (mddev
->persistent
)
5134 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5136 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5138 rdev
->sectors
= rdev
->sb_start
;
5140 if (test_bit(Faulty
, &rdev
->flags
)) {
5142 "md: can not hot-add faulty %s disk to %s!\n",
5143 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5147 clear_bit(In_sync
, &rdev
->flags
);
5149 rdev
->saved_raid_disk
= -1;
5150 err
= bind_rdev_to_array(rdev
, mddev
);
5155 * The rest should better be atomic, we can have disk failures
5156 * noticed in interrupt contexts ...
5159 rdev
->raid_disk
= -1;
5161 md_update_sb(mddev
, 1);
5164 * Kick recovery, maybe this spare has to be added to the
5165 * array immediately.
5167 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5168 md_wakeup_thread(mddev
->thread
);
5169 md_new_event(mddev
);
5177 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5182 if (!mddev
->pers
->quiesce
)
5184 if (mddev
->recovery
|| mddev
->sync_thread
)
5186 /* we should be able to change the bitmap.. */
5192 return -EEXIST
; /* cannot add when bitmap is present */
5193 mddev
->bitmap_info
.file
= fget(fd
);
5195 if (mddev
->bitmap_info
.file
== NULL
) {
5196 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5201 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5203 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5205 fput(mddev
->bitmap_info
.file
);
5206 mddev
->bitmap_info
.file
= NULL
;
5209 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5210 } else if (mddev
->bitmap
== NULL
)
5211 return -ENOENT
; /* cannot remove what isn't there */
5214 mddev
->pers
->quiesce(mddev
, 1);
5216 err
= bitmap_create(mddev
);
5217 if (fd
< 0 || err
) {
5218 bitmap_destroy(mddev
);
5219 fd
= -1; /* make sure to put the file */
5221 mddev
->pers
->quiesce(mddev
, 0);
5224 if (mddev
->bitmap_info
.file
) {
5225 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5226 fput(mddev
->bitmap_info
.file
);
5228 mddev
->bitmap_info
.file
= NULL
;
5235 * set_array_info is used two different ways
5236 * The original usage is when creating a new array.
5237 * In this usage, raid_disks is > 0 and it together with
5238 * level, size, not_persistent,layout,chunksize determine the
5239 * shape of the array.
5240 * This will always create an array with a type-0.90.0 superblock.
5241 * The newer usage is when assembling an array.
5242 * In this case raid_disks will be 0, and the major_version field is
5243 * use to determine which style super-blocks are to be found on the devices.
5244 * The minor and patch _version numbers are also kept incase the
5245 * super_block handler wishes to interpret them.
5247 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5250 if (info
->raid_disks
== 0) {
5251 /* just setting version number for superblock loading */
5252 if (info
->major_version
< 0 ||
5253 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5254 super_types
[info
->major_version
].name
== NULL
) {
5255 /* maybe try to auto-load a module? */
5257 "md: superblock version %d not known\n",
5258 info
->major_version
);
5261 mddev
->major_version
= info
->major_version
;
5262 mddev
->minor_version
= info
->minor_version
;
5263 mddev
->patch_version
= info
->patch_version
;
5264 mddev
->persistent
= !info
->not_persistent
;
5267 mddev
->major_version
= MD_MAJOR_VERSION
;
5268 mddev
->minor_version
= MD_MINOR_VERSION
;
5269 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5270 mddev
->ctime
= get_seconds();
5272 mddev
->level
= info
->level
;
5273 mddev
->clevel
[0] = 0;
5274 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5275 mddev
->raid_disks
= info
->raid_disks
;
5276 /* don't set md_minor, it is determined by which /dev/md* was
5279 if (info
->state
& (1<<MD_SB_CLEAN
))
5280 mddev
->recovery_cp
= MaxSector
;
5282 mddev
->recovery_cp
= 0;
5283 mddev
->persistent
= ! info
->not_persistent
;
5284 mddev
->external
= 0;
5286 mddev
->layout
= info
->layout
;
5287 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5289 mddev
->max_disks
= MD_SB_DISKS
;
5291 if (mddev
->persistent
)
5293 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5295 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5296 mddev
->bitmap_info
.offset
= 0;
5298 mddev
->reshape_position
= MaxSector
;
5301 * Generate a 128 bit UUID
5303 get_random_bytes(mddev
->uuid
, 16);
5305 mddev
->new_level
= mddev
->level
;
5306 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5307 mddev
->new_layout
= mddev
->layout
;
5308 mddev
->delta_disks
= 0;
5313 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5315 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5317 if (mddev
->external_size
)
5320 mddev
->array_sectors
= array_sectors
;
5322 EXPORT_SYMBOL(md_set_array_sectors
);
5324 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5328 int fit
= (num_sectors
== 0);
5330 if (mddev
->pers
->resize
== NULL
)
5332 /* The "num_sectors" is the number of sectors of each device that
5333 * is used. This can only make sense for arrays with redundancy.
5334 * linear and raid0 always use whatever space is available. We can only
5335 * consider changing this number if no resync or reconstruction is
5336 * happening, and if the new size is acceptable. It must fit before the
5337 * sb_start or, if that is <data_offset, it must fit before the size
5338 * of each device. If num_sectors is zero, we find the largest size
5342 if (mddev
->sync_thread
)
5345 /* Sorry, cannot grow a bitmap yet, just remove it,
5349 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5350 sector_t avail
= rdev
->sectors
;
5352 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5353 num_sectors
= avail
;
5354 if (avail
< num_sectors
)
5357 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5359 revalidate_disk(mddev
->gendisk
);
5363 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5366 /* change the number of raid disks */
5367 if (mddev
->pers
->check_reshape
== NULL
)
5369 if (raid_disks
<= 0 ||
5370 raid_disks
>= mddev
->max_disks
)
5372 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5374 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5376 rv
= mddev
->pers
->check_reshape(mddev
);
5382 * update_array_info is used to change the configuration of an
5384 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5385 * fields in the info are checked against the array.
5386 * Any differences that cannot be handled will cause an error.
5387 * Normally, only one change can be managed at a time.
5389 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5395 /* calculate expected state,ignoring low bits */
5396 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5397 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5399 if (mddev
->major_version
!= info
->major_version
||
5400 mddev
->minor_version
!= info
->minor_version
||
5401 /* mddev->patch_version != info->patch_version || */
5402 mddev
->ctime
!= info
->ctime
||
5403 mddev
->level
!= info
->level
||
5404 /* mddev->layout != info->layout || */
5405 !mddev
->persistent
!= info
->not_persistent
||
5406 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5407 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5408 ((state
^info
->state
) & 0xfffffe00)
5411 /* Check there is only one change */
5412 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5414 if (mddev
->raid_disks
!= info
->raid_disks
)
5416 if (mddev
->layout
!= info
->layout
)
5418 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5425 if (mddev
->layout
!= info
->layout
) {
5427 * we don't need to do anything at the md level, the
5428 * personality will take care of it all.
5430 if (mddev
->pers
->check_reshape
== NULL
)
5433 mddev
->new_layout
= info
->layout
;
5434 rv
= mddev
->pers
->check_reshape(mddev
);
5436 mddev
->new_layout
= mddev
->layout
;
5440 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5441 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5443 if (mddev
->raid_disks
!= info
->raid_disks
)
5444 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5446 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5447 if (mddev
->pers
->quiesce
== NULL
)
5449 if (mddev
->recovery
|| mddev
->sync_thread
)
5451 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5452 /* add the bitmap */
5455 if (mddev
->bitmap_info
.default_offset
== 0)
5457 mddev
->bitmap_info
.offset
=
5458 mddev
->bitmap_info
.default_offset
;
5459 mddev
->pers
->quiesce(mddev
, 1);
5460 rv
= bitmap_create(mddev
);
5462 bitmap_destroy(mddev
);
5463 mddev
->pers
->quiesce(mddev
, 0);
5465 /* remove the bitmap */
5468 if (mddev
->bitmap
->file
)
5470 mddev
->pers
->quiesce(mddev
, 1);
5471 bitmap_destroy(mddev
);
5472 mddev
->pers
->quiesce(mddev
, 0);
5473 mddev
->bitmap_info
.offset
= 0;
5476 md_update_sb(mddev
, 1);
5480 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5484 if (mddev
->pers
== NULL
)
5487 rdev
= find_rdev(mddev
, dev
);
5491 md_error(mddev
, rdev
);
5496 * We have a problem here : there is no easy way to give a CHS
5497 * virtual geometry. We currently pretend that we have a 2 heads
5498 * 4 sectors (with a BIG number of cylinders...). This drives
5499 * dosfs just mad... ;-)
5501 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5503 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5507 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
5511 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5512 unsigned int cmd
, unsigned long arg
)
5515 void __user
*argp
= (void __user
*)arg
;
5516 mddev_t
*mddev
= NULL
;
5518 if (!capable(CAP_SYS_ADMIN
))
5522 * Commands dealing with the RAID driver but not any
5528 err
= get_version(argp
);
5531 case PRINT_RAID_DEBUG
:
5539 autostart_arrays(arg
);
5546 * Commands creating/starting a new array:
5549 mddev
= bdev
->bd_disk
->private_data
;
5556 err
= mddev_lock(mddev
);
5559 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5566 case SET_ARRAY_INFO
:
5568 mdu_array_info_t info
;
5570 memset(&info
, 0, sizeof(info
));
5571 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5576 err
= update_array_info(mddev
, &info
);
5578 printk(KERN_WARNING
"md: couldn't update"
5579 " array info. %d\n", err
);
5584 if (!list_empty(&mddev
->disks
)) {
5586 "md: array %s already has disks!\n",
5591 if (mddev
->raid_disks
) {
5593 "md: array %s already initialised!\n",
5598 err
= set_array_info(mddev
, &info
);
5600 printk(KERN_WARNING
"md: couldn't set"
5601 " array info. %d\n", err
);
5611 * Commands querying/configuring an existing array:
5613 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5614 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5615 if ((!mddev
->raid_disks
&& !mddev
->external
)
5616 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5617 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5618 && cmd
!= GET_BITMAP_FILE
) {
5624 * Commands even a read-only array can execute:
5628 case GET_ARRAY_INFO
:
5629 err
= get_array_info(mddev
, argp
);
5632 case GET_BITMAP_FILE
:
5633 err
= get_bitmap_file(mddev
, argp
);
5637 err
= get_disk_info(mddev
, argp
);
5640 case RESTART_ARRAY_RW
:
5641 err
= restart_array(mddev
);
5645 err
= do_md_stop(mddev
, 0, 1);
5649 err
= do_md_stop(mddev
, 1, 1);
5655 * The remaining ioctls are changing the state of the
5656 * superblock, so we do not allow them on read-only arrays.
5657 * However non-MD ioctls (e.g. get-size) will still come through
5658 * here and hit the 'default' below, so only disallow
5659 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5661 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5662 if (mddev
->ro
== 2) {
5664 sysfs_notify_dirent(mddev
->sysfs_state
);
5665 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5666 md_wakeup_thread(mddev
->thread
);
5677 mdu_disk_info_t info
;
5678 if (copy_from_user(&info
, argp
, sizeof(info
)))
5681 err
= add_new_disk(mddev
, &info
);
5685 case HOT_REMOVE_DISK
:
5686 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5690 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5693 case SET_DISK_FAULTY
:
5694 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5698 err
= do_md_run(mddev
);
5701 case SET_BITMAP_FILE
:
5702 err
= set_bitmap_file(mddev
, (int)arg
);
5712 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5714 mddev
->hold_active
= 0;
5715 mddev_unlock(mddev
);
5724 #ifdef CONFIG_COMPAT
5725 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
5726 unsigned int cmd
, unsigned long arg
)
5729 case HOT_REMOVE_DISK
:
5731 case SET_DISK_FAULTY
:
5732 case SET_BITMAP_FILE
:
5733 /* These take in integer arg, do not convert */
5736 arg
= (unsigned long)compat_ptr(arg
);
5740 return md_ioctl(bdev
, mode
, cmd
, arg
);
5742 #endif /* CONFIG_COMPAT */
5744 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5747 * Succeed if we can lock the mddev, which confirms that
5748 * it isn't being stopped right now.
5750 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5753 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5754 /* we are racing with mddev_put which is discarding this
5758 /* Wait until bdev->bd_disk is definitely gone */
5759 flush_scheduled_work();
5760 /* Then retry the open from the top */
5761 return -ERESTARTSYS
;
5763 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5765 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5769 atomic_inc(&mddev
->openers
);
5770 mutex_unlock(&mddev
->open_mutex
);
5772 check_disk_change(bdev
);
5777 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5779 mddev_t
*mddev
= disk
->private_data
;
5782 atomic_dec(&mddev
->openers
);
5788 static int md_media_changed(struct gendisk
*disk
)
5790 mddev_t
*mddev
= disk
->private_data
;
5792 return mddev
->changed
;
5795 static int md_revalidate(struct gendisk
*disk
)
5797 mddev_t
*mddev
= disk
->private_data
;
5802 static const struct block_device_operations md_fops
=
5804 .owner
= THIS_MODULE
,
5806 .release
= md_release
,
5808 #ifdef CONFIG_COMPAT
5809 .compat_ioctl
= md_compat_ioctl
,
5811 .getgeo
= md_getgeo
,
5812 .media_changed
= md_media_changed
,
5813 .revalidate_disk
= md_revalidate
,
5816 static int md_thread(void * arg
)
5818 mdk_thread_t
*thread
= arg
;
5821 * md_thread is a 'system-thread', it's priority should be very
5822 * high. We avoid resource deadlocks individually in each
5823 * raid personality. (RAID5 does preallocation) We also use RR and
5824 * the very same RT priority as kswapd, thus we will never get
5825 * into a priority inversion deadlock.
5827 * we definitely have to have equal or higher priority than
5828 * bdflush, otherwise bdflush will deadlock if there are too
5829 * many dirty RAID5 blocks.
5832 allow_signal(SIGKILL
);
5833 while (!kthread_should_stop()) {
5835 /* We need to wait INTERRUPTIBLE so that
5836 * we don't add to the load-average.
5837 * That means we need to be sure no signals are
5840 if (signal_pending(current
))
5841 flush_signals(current
);
5843 wait_event_interruptible_timeout
5845 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5846 || kthread_should_stop(),
5849 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5851 thread
->run(thread
->mddev
);
5857 void md_wakeup_thread(mdk_thread_t
*thread
)
5860 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5861 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5862 wake_up(&thread
->wqueue
);
5866 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5869 mdk_thread_t
*thread
;
5871 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5875 init_waitqueue_head(&thread
->wqueue
);
5878 thread
->mddev
= mddev
;
5879 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5880 thread
->tsk
= kthread_run(md_thread
, thread
,
5882 mdname(thread
->mddev
),
5883 name
?: mddev
->pers
->name
);
5884 if (IS_ERR(thread
->tsk
)) {
5891 void md_unregister_thread(mdk_thread_t
*thread
)
5895 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5897 kthread_stop(thread
->tsk
);
5901 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5908 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5911 if (mddev
->external
)
5912 set_bit(Blocked
, &rdev
->flags
);
5914 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5916 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5917 __builtin_return_address(0),__builtin_return_address(1),
5918 __builtin_return_address(2),__builtin_return_address(3));
5922 if (!mddev
->pers
->error_handler
)
5924 mddev
->pers
->error_handler(mddev
,rdev
);
5925 if (mddev
->degraded
)
5926 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5927 set_bit(StateChanged
, &rdev
->flags
);
5928 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5929 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5930 md_wakeup_thread(mddev
->thread
);
5931 md_new_event_inintr(mddev
);
5934 /* seq_file implementation /proc/mdstat */
5936 static void status_unused(struct seq_file
*seq
)
5941 seq_printf(seq
, "unused devices: ");
5943 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
5944 char b
[BDEVNAME_SIZE
];
5946 seq_printf(seq
, "%s ",
5947 bdevname(rdev
->bdev
,b
));
5950 seq_printf(seq
, "<none>");
5952 seq_printf(seq
, "\n");
5956 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
5958 sector_t max_sectors
, resync
, res
;
5959 unsigned long dt
, db
;
5962 unsigned int per_milli
;
5964 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
5966 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5967 max_sectors
= mddev
->resync_max_sectors
;
5969 max_sectors
= mddev
->dev_sectors
;
5972 * Should not happen.
5978 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5979 * in a sector_t, and (max_sectors>>scale) will fit in a
5980 * u32, as those are the requirements for sector_div.
5981 * Thus 'scale' must be at least 10
5984 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5985 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
5988 res
= (resync
>>scale
)*1000;
5989 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
5993 int i
, x
= per_milli
/50, y
= 20-x
;
5994 seq_printf(seq
, "[");
5995 for (i
= 0; i
< x
; i
++)
5996 seq_printf(seq
, "=");
5997 seq_printf(seq
, ">");
5998 for (i
= 0; i
< y
; i
++)
5999 seq_printf(seq
, ".");
6000 seq_printf(seq
, "] ");
6002 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6003 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6005 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6007 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6008 "resync" : "recovery"))),
6009 per_milli
/10, per_milli
% 10,
6010 (unsigned long long) resync
/2,
6011 (unsigned long long) max_sectors
/2);
6014 * dt: time from mark until now
6015 * db: blocks written from mark until now
6016 * rt: remaining time
6018 * rt is a sector_t, so could be 32bit or 64bit.
6019 * So we divide before multiply in case it is 32bit and close
6021 * We scale the divisor (db) by 32 to avoid loosing precision
6022 * near the end of resync when the number of remaining sectors
6024 * We then divide rt by 32 after multiplying by db to compensate.
6025 * The '+1' avoids division by zero if db is very small.
6027 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6029 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6030 - mddev
->resync_mark_cnt
;
6032 rt
= max_sectors
- resync
; /* number of remaining sectors */
6033 sector_div(rt
, db
/32+1);
6037 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6038 ((unsigned long)rt
% 60)/6);
6040 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6043 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6045 struct list_head
*tmp
;
6055 spin_lock(&all_mddevs_lock
);
6056 list_for_each(tmp
,&all_mddevs
)
6058 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6060 spin_unlock(&all_mddevs_lock
);
6063 spin_unlock(&all_mddevs_lock
);
6065 return (void*)2;/* tail */
6069 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6071 struct list_head
*tmp
;
6072 mddev_t
*next_mddev
, *mddev
= v
;
6078 spin_lock(&all_mddevs_lock
);
6080 tmp
= all_mddevs
.next
;
6082 tmp
= mddev
->all_mddevs
.next
;
6083 if (tmp
!= &all_mddevs
)
6084 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6086 next_mddev
= (void*)2;
6089 spin_unlock(&all_mddevs_lock
);
6097 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6101 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6105 struct mdstat_info
{
6109 static int md_seq_show(struct seq_file
*seq
, void *v
)
6114 struct mdstat_info
*mi
= seq
->private;
6115 struct bitmap
*bitmap
;
6117 if (v
== (void*)1) {
6118 struct mdk_personality
*pers
;
6119 seq_printf(seq
, "Personalities : ");
6120 spin_lock(&pers_lock
);
6121 list_for_each_entry(pers
, &pers_list
, list
)
6122 seq_printf(seq
, "[%s] ", pers
->name
);
6124 spin_unlock(&pers_lock
);
6125 seq_printf(seq
, "\n");
6126 mi
->event
= atomic_read(&md_event_count
);
6129 if (v
== (void*)2) {
6134 if (mddev_lock(mddev
) < 0)
6137 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6138 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6139 mddev
->pers
? "" : "in");
6142 seq_printf(seq
, " (read-only)");
6144 seq_printf(seq
, " (auto-read-only)");
6145 seq_printf(seq
, " %s", mddev
->pers
->name
);
6149 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6150 char b
[BDEVNAME_SIZE
];
6151 seq_printf(seq
, " %s[%d]",
6152 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6153 if (test_bit(WriteMostly
, &rdev
->flags
))
6154 seq_printf(seq
, "(W)");
6155 if (test_bit(Faulty
, &rdev
->flags
)) {
6156 seq_printf(seq
, "(F)");
6158 } else if (rdev
->raid_disk
< 0)
6159 seq_printf(seq
, "(S)"); /* spare */
6160 sectors
+= rdev
->sectors
;
6163 if (!list_empty(&mddev
->disks
)) {
6165 seq_printf(seq
, "\n %llu blocks",
6166 (unsigned long long)
6167 mddev
->array_sectors
/ 2);
6169 seq_printf(seq
, "\n %llu blocks",
6170 (unsigned long long)sectors
/ 2);
6172 if (mddev
->persistent
) {
6173 if (mddev
->major_version
!= 0 ||
6174 mddev
->minor_version
!= 90) {
6175 seq_printf(seq
," super %d.%d",
6176 mddev
->major_version
,
6177 mddev
->minor_version
);
6179 } else if (mddev
->external
)
6180 seq_printf(seq
, " super external:%s",
6181 mddev
->metadata_type
);
6183 seq_printf(seq
, " super non-persistent");
6186 mddev
->pers
->status(seq
, mddev
);
6187 seq_printf(seq
, "\n ");
6188 if (mddev
->pers
->sync_request
) {
6189 if (mddev
->curr_resync
> 2) {
6190 status_resync(seq
, mddev
);
6191 seq_printf(seq
, "\n ");
6192 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6193 seq_printf(seq
, "\tresync=DELAYED\n ");
6194 else if (mddev
->recovery_cp
< MaxSector
)
6195 seq_printf(seq
, "\tresync=PENDING\n ");
6198 seq_printf(seq
, "\n ");
6200 if ((bitmap
= mddev
->bitmap
)) {
6201 unsigned long chunk_kb
;
6202 unsigned long flags
;
6203 spin_lock_irqsave(&bitmap
->lock
, flags
);
6204 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6205 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6207 bitmap
->pages
- bitmap
->missing_pages
,
6209 (bitmap
->pages
- bitmap
->missing_pages
)
6210 << (PAGE_SHIFT
- 10),
6211 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6212 chunk_kb
? "KB" : "B");
6214 seq_printf(seq
, ", file: ");
6215 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6218 seq_printf(seq
, "\n");
6219 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6222 seq_printf(seq
, "\n");
6224 mddev_unlock(mddev
);
6229 static const struct seq_operations md_seq_ops
= {
6230 .start
= md_seq_start
,
6231 .next
= md_seq_next
,
6232 .stop
= md_seq_stop
,
6233 .show
= md_seq_show
,
6236 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6239 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6243 error
= seq_open(file
, &md_seq_ops
);
6247 struct seq_file
*p
= file
->private_data
;
6249 mi
->event
= atomic_read(&md_event_count
);
6254 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6256 struct seq_file
*m
= filp
->private_data
;
6257 struct mdstat_info
*mi
= m
->private;
6260 poll_wait(filp
, &md_event_waiters
, wait
);
6262 /* always allow read */
6263 mask
= POLLIN
| POLLRDNORM
;
6265 if (mi
->event
!= atomic_read(&md_event_count
))
6266 mask
|= POLLERR
| POLLPRI
;
6270 static const struct file_operations md_seq_fops
= {
6271 .owner
= THIS_MODULE
,
6272 .open
= md_seq_open
,
6274 .llseek
= seq_lseek
,
6275 .release
= seq_release_private
,
6276 .poll
= mdstat_poll
,
6279 int register_md_personality(struct mdk_personality
*p
)
6281 spin_lock(&pers_lock
);
6282 list_add_tail(&p
->list
, &pers_list
);
6283 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6284 spin_unlock(&pers_lock
);
6288 int unregister_md_personality(struct mdk_personality
*p
)
6290 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6291 spin_lock(&pers_lock
);
6292 list_del_init(&p
->list
);
6293 spin_unlock(&pers_lock
);
6297 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6305 rdev_for_each_rcu(rdev
, mddev
) {
6306 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6307 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6308 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6309 atomic_read(&disk
->sync_io
);
6310 /* sync IO will cause sync_io to increase before the disk_stats
6311 * as sync_io is counted when a request starts, and
6312 * disk_stats is counted when it completes.
6313 * So resync activity will cause curr_events to be smaller than
6314 * when there was no such activity.
6315 * non-sync IO will cause disk_stat to increase without
6316 * increasing sync_io so curr_events will (eventually)
6317 * be larger than it was before. Once it becomes
6318 * substantially larger, the test below will cause
6319 * the array to appear non-idle, and resync will slow
6321 * If there is a lot of outstanding resync activity when
6322 * we set last_event to curr_events, then all that activity
6323 * completing might cause the array to appear non-idle
6324 * and resync will be slowed down even though there might
6325 * not have been non-resync activity. This will only
6326 * happen once though. 'last_events' will soon reflect
6327 * the state where there is little or no outstanding
6328 * resync requests, and further resync activity will
6329 * always make curr_events less than last_events.
6332 if (init
|| curr_events
- rdev
->last_events
> 64) {
6333 rdev
->last_events
= curr_events
;
6341 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6343 /* another "blocks" (512byte) blocks have been synced */
6344 atomic_sub(blocks
, &mddev
->recovery_active
);
6345 wake_up(&mddev
->recovery_wait
);
6347 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6348 md_wakeup_thread(mddev
->thread
);
6349 // stop recovery, signal do_sync ....
6354 /* md_write_start(mddev, bi)
6355 * If we need to update some array metadata (e.g. 'active' flag
6356 * in superblock) before writing, schedule a superblock update
6357 * and wait for it to complete.
6359 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6362 if (bio_data_dir(bi
) != WRITE
)
6365 BUG_ON(mddev
->ro
== 1);
6366 if (mddev
->ro
== 2) {
6367 /* need to switch to read/write */
6369 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6370 md_wakeup_thread(mddev
->thread
);
6371 md_wakeup_thread(mddev
->sync_thread
);
6374 atomic_inc(&mddev
->writes_pending
);
6375 if (mddev
->safemode
== 1)
6376 mddev
->safemode
= 0;
6377 if (mddev
->in_sync
) {
6378 spin_lock_irq(&mddev
->write_lock
);
6379 if (mddev
->in_sync
) {
6381 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6382 md_wakeup_thread(mddev
->thread
);
6385 spin_unlock_irq(&mddev
->write_lock
);
6388 sysfs_notify_dirent(mddev
->sysfs_state
);
6389 wait_event(mddev
->sb_wait
,
6390 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6391 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6394 void md_write_end(mddev_t
*mddev
)
6396 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6397 if (mddev
->safemode
== 2)
6398 md_wakeup_thread(mddev
->thread
);
6399 else if (mddev
->safemode_delay
)
6400 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6404 /* md_allow_write(mddev)
6405 * Calling this ensures that the array is marked 'active' so that writes
6406 * may proceed without blocking. It is important to call this before
6407 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6408 * Must be called with mddev_lock held.
6410 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6411 * is dropped, so return -EAGAIN after notifying userspace.
6413 int md_allow_write(mddev_t
*mddev
)
6419 if (!mddev
->pers
->sync_request
)
6422 spin_lock_irq(&mddev
->write_lock
);
6423 if (mddev
->in_sync
) {
6425 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6426 if (mddev
->safemode_delay
&&
6427 mddev
->safemode
== 0)
6428 mddev
->safemode
= 1;
6429 spin_unlock_irq(&mddev
->write_lock
);
6430 md_update_sb(mddev
, 0);
6431 sysfs_notify_dirent(mddev
->sysfs_state
);
6433 spin_unlock_irq(&mddev
->write_lock
);
6435 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6440 EXPORT_SYMBOL_GPL(md_allow_write
);
6442 #define SYNC_MARKS 10
6443 #define SYNC_MARK_STEP (3*HZ)
6444 void md_do_sync(mddev_t
*mddev
)
6447 unsigned int currspeed
= 0,
6449 sector_t max_sectors
,j
, io_sectors
;
6450 unsigned long mark
[SYNC_MARKS
];
6451 sector_t mark_cnt
[SYNC_MARKS
];
6453 struct list_head
*tmp
;
6454 sector_t last_check
;
6459 /* just incase thread restarts... */
6460 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6462 if (mddev
->ro
) /* never try to sync a read-only array */
6465 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6466 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6467 desc
= "data-check";
6468 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6469 desc
= "requested-resync";
6472 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6477 /* we overload curr_resync somewhat here.
6478 * 0 == not engaged in resync at all
6479 * 2 == checking that there is no conflict with another sync
6480 * 1 == like 2, but have yielded to allow conflicting resync to
6482 * other == active in resync - this many blocks
6484 * Before starting a resync we must have set curr_resync to
6485 * 2, and then checked that every "conflicting" array has curr_resync
6486 * less than ours. When we find one that is the same or higher
6487 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6488 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6489 * This will mean we have to start checking from the beginning again.
6494 mddev
->curr_resync
= 2;
6497 if (kthread_should_stop()) {
6498 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6501 for_each_mddev(mddev2
, tmp
) {
6502 if (mddev2
== mddev
)
6504 if (!mddev
->parallel_resync
6505 && mddev2
->curr_resync
6506 && match_mddev_units(mddev
, mddev2
)) {
6508 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6509 /* arbitrarily yield */
6510 mddev
->curr_resync
= 1;
6511 wake_up(&resync_wait
);
6513 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6514 /* no need to wait here, we can wait the next
6515 * time 'round when curr_resync == 2
6518 /* We need to wait 'interruptible' so as not to
6519 * contribute to the load average, and not to
6520 * be caught by 'softlockup'
6522 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6523 if (!kthread_should_stop() &&
6524 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6525 printk(KERN_INFO
"md: delaying %s of %s"
6526 " until %s has finished (they"
6527 " share one or more physical units)\n",
6528 desc
, mdname(mddev
), mdname(mddev2
));
6530 if (signal_pending(current
))
6531 flush_signals(current
);
6533 finish_wait(&resync_wait
, &wq
);
6536 finish_wait(&resync_wait
, &wq
);
6539 } while (mddev
->curr_resync
< 2);
6542 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6543 /* resync follows the size requested by the personality,
6544 * which defaults to physical size, but can be virtual size
6546 max_sectors
= mddev
->resync_max_sectors
;
6547 mddev
->resync_mismatches
= 0;
6548 /* we don't use the checkpoint if there's a bitmap */
6549 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6550 j
= mddev
->resync_min
;
6551 else if (!mddev
->bitmap
)
6552 j
= mddev
->recovery_cp
;
6554 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6555 max_sectors
= mddev
->dev_sectors
;
6557 /* recovery follows the physical size of devices */
6558 max_sectors
= mddev
->dev_sectors
;
6561 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6562 if (rdev
->raid_disk
>= 0 &&
6563 !test_bit(Faulty
, &rdev
->flags
) &&
6564 !test_bit(In_sync
, &rdev
->flags
) &&
6565 rdev
->recovery_offset
< j
)
6566 j
= rdev
->recovery_offset
;
6570 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6571 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6572 " %d KB/sec/disk.\n", speed_min(mddev
));
6573 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6574 "(but not more than %d KB/sec) for %s.\n",
6575 speed_max(mddev
), desc
);
6577 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6580 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6582 mark_cnt
[m
] = io_sectors
;
6585 mddev
->resync_mark
= mark
[last_mark
];
6586 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6589 * Tune reconstruction:
6591 window
= 32*(PAGE_SIZE
/512);
6592 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6593 window
/2,(unsigned long long) max_sectors
/2);
6595 atomic_set(&mddev
->recovery_active
, 0);
6600 "md: resuming %s of %s from checkpoint.\n",
6601 desc
, mdname(mddev
));
6602 mddev
->curr_resync
= j
;
6604 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6606 while (j
< max_sectors
) {
6611 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6612 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6613 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6614 > (max_sectors
>> 4)) ||
6615 (j
- mddev
->curr_resync_completed
)*2
6616 >= mddev
->resync_max
- mddev
->curr_resync_completed
6618 /* time to update curr_resync_completed */
6619 blk_unplug(mddev
->queue
);
6620 wait_event(mddev
->recovery_wait
,
6621 atomic_read(&mddev
->recovery_active
) == 0);
6622 mddev
->curr_resync_completed
=
6624 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6625 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6628 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6629 /* As this condition is controlled by user-space,
6630 * we can block indefinitely, so use '_interruptible'
6631 * to avoid triggering warnings.
6633 flush_signals(current
); /* just in case */
6634 wait_event_interruptible(mddev
->recovery_wait
,
6635 mddev
->resync_max
> j
6636 || kthread_should_stop());
6639 if (kthread_should_stop())
6642 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6643 currspeed
< speed_min(mddev
));
6645 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6649 if (!skipped
) { /* actual IO requested */
6650 io_sectors
+= sectors
;
6651 atomic_add(sectors
, &mddev
->recovery_active
);
6655 if (j
>1) mddev
->curr_resync
= j
;
6656 mddev
->curr_mark_cnt
= io_sectors
;
6657 if (last_check
== 0)
6658 /* this is the earliers that rebuilt will be
6659 * visible in /proc/mdstat
6661 md_new_event(mddev
);
6663 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6666 last_check
= io_sectors
;
6668 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6672 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6674 int next
= (last_mark
+1) % SYNC_MARKS
;
6676 mddev
->resync_mark
= mark
[next
];
6677 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6678 mark
[next
] = jiffies
;
6679 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6684 if (kthread_should_stop())
6689 * this loop exits only if either when we are slower than
6690 * the 'hard' speed limit, or the system was IO-idle for
6692 * the system might be non-idle CPU-wise, but we only care
6693 * about not overloading the IO subsystem. (things like an
6694 * e2fsck being done on the RAID array should execute fast)
6696 blk_unplug(mddev
->queue
);
6699 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6700 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6702 if (currspeed
> speed_min(mddev
)) {
6703 if ((currspeed
> speed_max(mddev
)) ||
6704 !is_mddev_idle(mddev
, 0)) {
6710 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6712 * this also signals 'finished resyncing' to md_stop
6715 blk_unplug(mddev
->queue
);
6717 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6719 /* tell personality that we are finished */
6720 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6722 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6723 mddev
->curr_resync
> 2) {
6724 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6725 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6726 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6728 "md: checkpointing %s of %s.\n",
6729 desc
, mdname(mddev
));
6730 mddev
->recovery_cp
= mddev
->curr_resync
;
6733 mddev
->recovery_cp
= MaxSector
;
6735 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6736 mddev
->curr_resync
= MaxSector
;
6738 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6739 if (rdev
->raid_disk
>= 0 &&
6740 !test_bit(Faulty
, &rdev
->flags
) &&
6741 !test_bit(In_sync
, &rdev
->flags
) &&
6742 rdev
->recovery_offset
< mddev
->curr_resync
)
6743 rdev
->recovery_offset
= mddev
->curr_resync
;
6747 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6750 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6751 /* We completed so min/max setting can be forgotten if used. */
6752 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6753 mddev
->resync_min
= 0;
6754 mddev
->resync_max
= MaxSector
;
6755 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6756 mddev
->resync_min
= mddev
->curr_resync_completed
;
6757 mddev
->curr_resync
= 0;
6758 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6759 mddev
->curr_resync_completed
= 0;
6760 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6761 wake_up(&resync_wait
);
6762 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6763 md_wakeup_thread(mddev
->thread
);
6768 * got a signal, exit.
6771 "md: md_do_sync() got signal ... exiting\n");
6772 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6776 EXPORT_SYMBOL_GPL(md_do_sync
);
6779 static int remove_and_add_spares(mddev_t
*mddev
)
6784 mddev
->curr_resync_completed
= 0;
6786 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6787 if (rdev
->raid_disk
>= 0 &&
6788 !test_bit(Blocked
, &rdev
->flags
) &&
6789 (test_bit(Faulty
, &rdev
->flags
) ||
6790 ! test_bit(In_sync
, &rdev
->flags
)) &&
6791 atomic_read(&rdev
->nr_pending
)==0) {
6792 if (mddev
->pers
->hot_remove_disk(
6793 mddev
, rdev
->raid_disk
)==0) {
6795 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6796 sysfs_remove_link(&mddev
->kobj
, nm
);
6797 rdev
->raid_disk
= -1;
6801 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6802 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6803 if (rdev
->raid_disk
>= 0 &&
6804 !test_bit(In_sync
, &rdev
->flags
) &&
6805 !test_bit(Blocked
, &rdev
->flags
))
6807 if (rdev
->raid_disk
< 0
6808 && !test_bit(Faulty
, &rdev
->flags
)) {
6809 rdev
->recovery_offset
= 0;
6811 hot_add_disk(mddev
, rdev
) == 0) {
6813 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6814 if (sysfs_create_link(&mddev
->kobj
,
6817 "md: cannot register "
6821 md_new_event(mddev
);
6822 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6831 * This routine is regularly called by all per-raid-array threads to
6832 * deal with generic issues like resync and super-block update.
6833 * Raid personalities that don't have a thread (linear/raid0) do not
6834 * need this as they never do any recovery or update the superblock.
6836 * It does not do any resync itself, but rather "forks" off other threads
6837 * to do that as needed.
6838 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6839 * "->recovery" and create a thread at ->sync_thread.
6840 * When the thread finishes it sets MD_RECOVERY_DONE
6841 * and wakeups up this thread which will reap the thread and finish up.
6842 * This thread also removes any faulty devices (with nr_pending == 0).
6844 * The overall approach is:
6845 * 1/ if the superblock needs updating, update it.
6846 * 2/ If a recovery thread is running, don't do anything else.
6847 * 3/ If recovery has finished, clean up, possibly marking spares active.
6848 * 4/ If there are any faulty devices, remove them.
6849 * 5/ If array is degraded, try to add spares devices
6850 * 6/ If array has spares or is not in-sync, start a resync thread.
6852 void md_check_recovery(mddev_t
*mddev
)
6858 bitmap_daemon_work(mddev
);
6863 if (signal_pending(current
)) {
6864 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6865 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6867 mddev
->safemode
= 2;
6869 flush_signals(current
);
6872 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6875 (mddev
->flags
&& !mddev
->external
) ||
6876 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6877 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6878 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6879 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6880 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6884 if (mddev_trylock(mddev
)) {
6888 /* Only thing we do on a ro array is remove
6891 remove_and_add_spares(mddev
);
6892 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6896 if (!mddev
->external
) {
6898 spin_lock_irq(&mddev
->write_lock
);
6899 if (mddev
->safemode
&&
6900 !atomic_read(&mddev
->writes_pending
) &&
6902 mddev
->recovery_cp
== MaxSector
) {
6905 if (mddev
->persistent
)
6906 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6908 if (mddev
->safemode
== 1)
6909 mddev
->safemode
= 0;
6910 spin_unlock_irq(&mddev
->write_lock
);
6912 sysfs_notify_dirent(mddev
->sysfs_state
);
6916 md_update_sb(mddev
, 0);
6918 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6919 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6920 sysfs_notify_dirent(rdev
->sysfs_state
);
6923 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6924 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6925 /* resync/recovery still happening */
6926 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6929 if (mddev
->sync_thread
) {
6930 /* resync has finished, collect result */
6931 md_unregister_thread(mddev
->sync_thread
);
6932 mddev
->sync_thread
= NULL
;
6933 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6934 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6936 /* activate any spares */
6937 if (mddev
->pers
->spare_active(mddev
))
6938 sysfs_notify(&mddev
->kobj
, NULL
,
6941 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6942 mddev
->pers
->finish_reshape
)
6943 mddev
->pers
->finish_reshape(mddev
);
6944 md_update_sb(mddev
, 1);
6946 /* if array is no-longer degraded, then any saved_raid_disk
6947 * information must be scrapped
6949 if (!mddev
->degraded
)
6950 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6951 rdev
->saved_raid_disk
= -1;
6953 mddev
->recovery
= 0;
6954 /* flag recovery needed just to double check */
6955 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6956 sysfs_notify_dirent(mddev
->sysfs_action
);
6957 md_new_event(mddev
);
6960 /* Set RUNNING before clearing NEEDED to avoid
6961 * any transients in the value of "sync_action".
6963 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6964 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6965 /* Clear some bits that don't mean anything, but
6968 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6969 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6971 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
6973 /* no recovery is running.
6974 * remove any failed drives, then
6975 * add spares if possible.
6976 * Spare are also removed and re-added, to allow
6977 * the personality to fail the re-add.
6980 if (mddev
->reshape_position
!= MaxSector
) {
6981 if (mddev
->pers
->check_reshape
== NULL
||
6982 mddev
->pers
->check_reshape(mddev
) != 0)
6983 /* Cannot proceed */
6985 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
6986 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6987 } else if ((spares
= remove_and_add_spares(mddev
))) {
6988 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6989 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
6990 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
6991 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6992 } else if (mddev
->recovery_cp
< MaxSector
) {
6993 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6994 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6995 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6996 /* nothing to be done ... */
6999 if (mddev
->pers
->sync_request
) {
7000 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7001 /* We are adding a device or devices to an array
7002 * which has the bitmap stored on all devices.
7003 * So make sure all bitmap pages get written
7005 bitmap_write_all(mddev
->bitmap
);
7007 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7010 if (!mddev
->sync_thread
) {
7011 printk(KERN_ERR
"%s: could not start resync"
7014 /* leave the spares where they are, it shouldn't hurt */
7015 mddev
->recovery
= 0;
7017 md_wakeup_thread(mddev
->sync_thread
);
7018 sysfs_notify_dirent(mddev
->sysfs_action
);
7019 md_new_event(mddev
);
7022 if (!mddev
->sync_thread
) {
7023 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7024 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7026 if (mddev
->sysfs_action
)
7027 sysfs_notify_dirent(mddev
->sysfs_action
);
7029 mddev_unlock(mddev
);
7033 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7035 sysfs_notify_dirent(rdev
->sysfs_state
);
7036 wait_event_timeout(rdev
->blocked_wait
,
7037 !test_bit(Blocked
, &rdev
->flags
),
7038 msecs_to_jiffies(5000));
7039 rdev_dec_pending(rdev
, mddev
);
7041 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7043 static int md_notify_reboot(struct notifier_block
*this,
7044 unsigned long code
, void *x
)
7046 struct list_head
*tmp
;
7049 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7051 printk(KERN_INFO
"md: stopping all md devices.\n");
7053 for_each_mddev(mddev
, tmp
)
7054 if (mddev_trylock(mddev
)) {
7055 /* Force a switch to readonly even array
7056 * appears to still be in use. Hence
7059 do_md_stop(mddev
, 1, 100);
7060 mddev_unlock(mddev
);
7063 * certain more exotic SCSI devices are known to be
7064 * volatile wrt too early system reboots. While the
7065 * right place to handle this issue is the given
7066 * driver, we do want to have a safe RAID driver ...
7073 static struct notifier_block md_notifier
= {
7074 .notifier_call
= md_notify_reboot
,
7076 .priority
= INT_MAX
, /* before any real devices */
7079 static void md_geninit(void)
7081 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
7083 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
7086 static int __init
md_init(void)
7088 if (register_blkdev(MD_MAJOR
, "md"))
7090 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
7091 unregister_blkdev(MD_MAJOR
, "md");
7094 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7095 md_probe
, NULL
, NULL
);
7096 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7097 md_probe
, NULL
, NULL
);
7099 register_reboot_notifier(&md_notifier
);
7100 raid_table_header
= register_sysctl_table(raid_root_table
);
7110 * Searches all registered partitions for autorun RAID arrays
7114 static LIST_HEAD(all_detected_devices
);
7115 struct detected_devices_node
{
7116 struct list_head list
;
7120 void md_autodetect_dev(dev_t dev
)
7122 struct detected_devices_node
*node_detected_dev
;
7124 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7125 if (node_detected_dev
) {
7126 node_detected_dev
->dev
= dev
;
7127 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7129 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7130 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7135 static void autostart_arrays(int part
)
7138 struct detected_devices_node
*node_detected_dev
;
7140 int i_scanned
, i_passed
;
7145 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7147 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7149 node_detected_dev
= list_entry(all_detected_devices
.next
,
7150 struct detected_devices_node
, list
);
7151 list_del(&node_detected_dev
->list
);
7152 dev
= node_detected_dev
->dev
;
7153 kfree(node_detected_dev
);
7154 rdev
= md_import_device(dev
,0, 90);
7158 if (test_bit(Faulty
, &rdev
->flags
)) {
7162 set_bit(AutoDetected
, &rdev
->flags
);
7163 list_add(&rdev
->same_set
, &pending_raid_disks
);
7167 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7168 i_scanned
, i_passed
);
7170 autorun_devices(part
);
7173 #endif /* !MODULE */
7175 static __exit
void md_exit(void)
7178 struct list_head
*tmp
;
7180 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7181 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7183 unregister_blkdev(MD_MAJOR
,"md");
7184 unregister_blkdev(mdp_major
, "mdp");
7185 unregister_reboot_notifier(&md_notifier
);
7186 unregister_sysctl_table(raid_table_header
);
7187 remove_proc_entry("mdstat", NULL
);
7188 for_each_mddev(mddev
, tmp
) {
7189 export_array(mddev
);
7190 mddev
->hold_active
= 0;
7194 subsys_initcall(md_init
);
7195 module_exit(md_exit
)
7197 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7199 return sprintf(buffer
, "%d", start_readonly
);
7201 static int set_ro(const char *val
, struct kernel_param
*kp
)
7204 int num
= simple_strtoul(val
, &e
, 10);
7205 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7206 start_readonly
= num
;
7212 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7213 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7215 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7217 EXPORT_SYMBOL(register_md_personality
);
7218 EXPORT_SYMBOL(unregister_md_personality
);
7219 EXPORT_SYMBOL(md_error
);
7220 EXPORT_SYMBOL(md_done_sync
);
7221 EXPORT_SYMBOL(md_write_start
);
7222 EXPORT_SYMBOL(md_write_end
);
7223 EXPORT_SYMBOL(md_register_thread
);
7224 EXPORT_SYMBOL(md_unregister_thread
);
7225 EXPORT_SYMBOL(md_wakeup_thread
);
7226 EXPORT_SYMBOL(md_check_recovery
);
7227 MODULE_LICENSE("GPL");
7228 MODULE_DESCRIPTION("MD RAID framework");
7230 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);