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
->ctime
== 0 && !mddev
->hold_active
) {
390 /* Array is not configured at all, and not held active,
392 list_del(&mddev
->all_mddevs
);
393 if (mddev
->gendisk
) {
394 /* we did a probe so need to clean up.
395 * Call schedule_work inside the spinlock
396 * so that flush_scheduled_work() after
397 * mddev_find will succeed in waiting for the
400 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
401 schedule_work(&mddev
->del_work
);
405 spin_unlock(&all_mddevs_lock
);
408 static mddev_t
* mddev_find(dev_t unit
)
410 mddev_t
*mddev
, *new = NULL
;
413 spin_lock(&all_mddevs_lock
);
416 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
417 if (mddev
->unit
== unit
) {
419 spin_unlock(&all_mddevs_lock
);
425 list_add(&new->all_mddevs
, &all_mddevs
);
426 spin_unlock(&all_mddevs_lock
);
427 new->hold_active
= UNTIL_IOCTL
;
431 /* find an unused unit number */
432 static int next_minor
= 512;
433 int start
= next_minor
;
437 dev
= MKDEV(MD_MAJOR
, next_minor
);
439 if (next_minor
> MINORMASK
)
441 if (next_minor
== start
) {
442 /* Oh dear, all in use. */
443 spin_unlock(&all_mddevs_lock
);
449 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
450 if (mddev
->unit
== dev
) {
456 new->md_minor
= MINOR(dev
);
457 new->hold_active
= UNTIL_STOP
;
458 list_add(&new->all_mddevs
, &all_mddevs
);
459 spin_unlock(&all_mddevs_lock
);
462 spin_unlock(&all_mddevs_lock
);
464 new = kzalloc(sizeof(*new), GFP_KERNEL
);
469 if (MAJOR(unit
) == MD_MAJOR
)
470 new->md_minor
= MINOR(unit
);
472 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
474 mutex_init(&new->open_mutex
);
475 mutex_init(&new->reconfig_mutex
);
476 mutex_init(&new->bitmap_info
.mutex
);
477 INIT_LIST_HEAD(&new->disks
);
478 INIT_LIST_HEAD(&new->all_mddevs
);
479 init_timer(&new->safemode_timer
);
480 atomic_set(&new->active
, 1);
481 atomic_set(&new->openers
, 0);
482 atomic_set(&new->active_io
, 0);
483 spin_lock_init(&new->write_lock
);
484 atomic_set(&new->flush_pending
, 0);
485 init_waitqueue_head(&new->sb_wait
);
486 init_waitqueue_head(&new->recovery_wait
);
487 new->reshape_position
= MaxSector
;
489 new->resync_max
= MaxSector
;
490 new->level
= LEVEL_NONE
;
495 static inline int mddev_lock(mddev_t
* mddev
)
497 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
500 static inline int mddev_is_locked(mddev_t
*mddev
)
502 return mutex_is_locked(&mddev
->reconfig_mutex
);
505 static inline int mddev_trylock(mddev_t
* mddev
)
507 return mutex_trylock(&mddev
->reconfig_mutex
);
510 static struct attribute_group md_redundancy_group
;
512 static void mddev_unlock(mddev_t
* mddev
)
514 if (mddev
->to_remove
) {
515 /* These cannot be removed under reconfig_mutex as
516 * an access to the files will try to take reconfig_mutex
517 * while holding the file unremovable, which leads to
519 * So hold open_mutex instead - we are allowed to take
520 * it while holding reconfig_mutex, and md_run can
521 * use it to wait for the remove to complete.
523 struct attribute_group
*to_remove
= mddev
->to_remove
;
524 mddev
->to_remove
= NULL
;
525 mutex_lock(&mddev
->open_mutex
);
526 mutex_unlock(&mddev
->reconfig_mutex
);
528 if (to_remove
!= &md_redundancy_group
)
529 sysfs_remove_group(&mddev
->kobj
, to_remove
);
530 if (mddev
->pers
== NULL
||
531 mddev
->pers
->sync_request
== NULL
) {
532 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
533 if (mddev
->sysfs_action
)
534 sysfs_put(mddev
->sysfs_action
);
535 mddev
->sysfs_action
= NULL
;
537 mutex_unlock(&mddev
->open_mutex
);
539 mutex_unlock(&mddev
->reconfig_mutex
);
541 md_wakeup_thread(mddev
->thread
);
544 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
548 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
549 if (rdev
->desc_nr
== nr
)
555 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
559 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
560 if (rdev
->bdev
->bd_dev
== dev
)
566 static struct mdk_personality
*find_pers(int level
, char *clevel
)
568 struct mdk_personality
*pers
;
569 list_for_each_entry(pers
, &pers_list
, list
) {
570 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
572 if (strcmp(pers
->name
, clevel
)==0)
578 /* return the offset of the super block in 512byte sectors */
579 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
581 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
582 return MD_NEW_SIZE_SECTORS(num_sectors
);
585 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
590 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
591 if (!rdev
->sb_page
) {
592 printk(KERN_ALERT
"md: out of memory.\n");
599 static void free_disk_sb(mdk_rdev_t
* rdev
)
602 put_page(rdev
->sb_page
);
604 rdev
->sb_page
= NULL
;
611 static void super_written(struct bio
*bio
, int error
)
613 mdk_rdev_t
*rdev
= bio
->bi_private
;
614 mddev_t
*mddev
= rdev
->mddev
;
616 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
617 printk("md: super_written gets error=%d, uptodate=%d\n",
618 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
619 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
620 md_error(mddev
, rdev
);
623 if (atomic_dec_and_test(&mddev
->pending_writes
))
624 wake_up(&mddev
->sb_wait
);
628 static void super_written_barrier(struct bio
*bio
, int error
)
630 struct bio
*bio2
= bio
->bi_private
;
631 mdk_rdev_t
*rdev
= bio2
->bi_private
;
632 mddev_t
*mddev
= rdev
->mddev
;
634 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
635 error
== -EOPNOTSUPP
) {
637 /* barriers don't appear to be supported :-( */
638 set_bit(BarriersNotsupp
, &rdev
->flags
);
639 mddev
->barriers_work
= 0;
640 spin_lock_irqsave(&mddev
->write_lock
, flags
);
641 bio2
->bi_next
= mddev
->biolist
;
642 mddev
->biolist
= bio2
;
643 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
644 wake_up(&mddev
->sb_wait
);
648 bio
->bi_private
= rdev
;
649 super_written(bio
, error
);
653 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
654 sector_t sector
, int size
, struct page
*page
)
656 /* write first size bytes of page to sector of rdev
657 * Increment mddev->pending_writes before returning
658 * and decrement it on completion, waking up sb_wait
659 * if zero is reached.
660 * If an error occurred, call md_error
662 * As we might need to resubmit the request if BIO_RW_BARRIER
663 * causes ENOTSUPP, we allocate a spare bio...
665 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
666 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
668 bio
->bi_bdev
= rdev
->bdev
;
669 bio
->bi_sector
= sector
;
670 bio_add_page(bio
, page
, size
, 0);
671 bio
->bi_private
= rdev
;
672 bio
->bi_end_io
= super_written
;
675 atomic_inc(&mddev
->pending_writes
);
676 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
678 rw
|= (1<<BIO_RW_BARRIER
);
679 rbio
= bio_clone(bio
, GFP_NOIO
);
680 rbio
->bi_private
= bio
;
681 rbio
->bi_end_io
= super_written_barrier
;
682 submit_bio(rw
, rbio
);
687 void md_super_wait(mddev_t
*mddev
)
689 /* wait for all superblock writes that were scheduled to complete.
690 * if any had to be retried (due to BARRIER problems), retry them
694 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
695 if (atomic_read(&mddev
->pending_writes
)==0)
697 while (mddev
->biolist
) {
699 spin_lock_irq(&mddev
->write_lock
);
700 bio
= mddev
->biolist
;
701 mddev
->biolist
= bio
->bi_next
;
703 spin_unlock_irq(&mddev
->write_lock
);
704 submit_bio(bio
->bi_rw
, bio
);
708 finish_wait(&mddev
->sb_wait
, &wq
);
711 static void bi_complete(struct bio
*bio
, int error
)
713 complete((struct completion
*)bio
->bi_private
);
716 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
717 struct page
*page
, int rw
)
719 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
720 struct completion event
;
723 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
726 bio
->bi_sector
= sector
;
727 bio_add_page(bio
, page
, size
, 0);
728 init_completion(&event
);
729 bio
->bi_private
= &event
;
730 bio
->bi_end_io
= bi_complete
;
732 wait_for_completion(&event
);
734 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
738 EXPORT_SYMBOL_GPL(sync_page_io
);
740 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
742 char b
[BDEVNAME_SIZE
];
743 if (!rdev
->sb_page
) {
751 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
757 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
758 bdevname(rdev
->bdev
,b
));
762 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
764 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
765 sb1
->set_uuid1
== sb2
->set_uuid1
&&
766 sb1
->set_uuid2
== sb2
->set_uuid2
&&
767 sb1
->set_uuid3
== sb2
->set_uuid3
;
770 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
773 mdp_super_t
*tmp1
, *tmp2
;
775 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
776 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
778 if (!tmp1
|| !tmp2
) {
780 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
788 * nr_disks is not constant
793 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
801 static u32
md_csum_fold(u32 csum
)
803 csum
= (csum
& 0xffff) + (csum
>> 16);
804 return (csum
& 0xffff) + (csum
>> 16);
807 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
810 u32
*sb32
= (u32
*)sb
;
812 unsigned int disk_csum
, csum
;
814 disk_csum
= sb
->sb_csum
;
817 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
819 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
823 /* This used to use csum_partial, which was wrong for several
824 * reasons including that different results are returned on
825 * different architectures. It isn't critical that we get exactly
826 * the same return value as before (we always csum_fold before
827 * testing, and that removes any differences). However as we
828 * know that csum_partial always returned a 16bit value on
829 * alphas, do a fold to maximise conformity to previous behaviour.
831 sb
->sb_csum
= md_csum_fold(disk_csum
);
833 sb
->sb_csum
= disk_csum
;
840 * Handle superblock details.
841 * We want to be able to handle multiple superblock formats
842 * so we have a common interface to them all, and an array of
843 * different handlers.
844 * We rely on user-space to write the initial superblock, and support
845 * reading and updating of superblocks.
846 * Interface methods are:
847 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
848 * loads and validates a superblock on dev.
849 * if refdev != NULL, compare superblocks on both devices
851 * 0 - dev has a superblock that is compatible with refdev
852 * 1 - dev has a superblock that is compatible and newer than refdev
853 * so dev should be used as the refdev in future
854 * -EINVAL superblock incompatible or invalid
855 * -othererror e.g. -EIO
857 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
858 * Verify that dev is acceptable into mddev.
859 * The first time, mddev->raid_disks will be 0, and data from
860 * dev should be merged in. Subsequent calls check that dev
861 * is new enough. Return 0 or -EINVAL
863 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
864 * Update the superblock for rdev with data in mddev
865 * This does not write to disc.
871 struct module
*owner
;
872 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
874 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
875 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
876 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
877 sector_t num_sectors
);
881 * Check that the given mddev has no bitmap.
883 * This function is called from the run method of all personalities that do not
884 * support bitmaps. It prints an error message and returns non-zero if mddev
885 * has a bitmap. Otherwise, it returns 0.
888 int md_check_no_bitmap(mddev_t
*mddev
)
890 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
892 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
893 mdname(mddev
), mddev
->pers
->name
);
896 EXPORT_SYMBOL(md_check_no_bitmap
);
899 * load_super for 0.90.0
901 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
903 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
908 * Calculate the position of the superblock (512byte sectors),
909 * it's at the end of the disk.
911 * It also happens to be a multiple of 4Kb.
913 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
915 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
920 bdevname(rdev
->bdev
, b
);
921 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
923 if (sb
->md_magic
!= MD_SB_MAGIC
) {
924 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
929 if (sb
->major_version
!= 0 ||
930 sb
->minor_version
< 90 ||
931 sb
->minor_version
> 91) {
932 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
933 sb
->major_version
, sb
->minor_version
,
938 if (sb
->raid_disks
<= 0)
941 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
942 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
947 rdev
->preferred_minor
= sb
->md_minor
;
948 rdev
->data_offset
= 0;
949 rdev
->sb_size
= MD_SB_BYTES
;
951 if (sb
->level
== LEVEL_MULTIPATH
)
954 rdev
->desc_nr
= sb
->this_disk
.number
;
960 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
961 if (!uuid_equal(refsb
, sb
)) {
962 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
963 b
, bdevname(refdev
->bdev
,b2
));
966 if (!sb_equal(refsb
, sb
)) {
967 printk(KERN_WARNING
"md: %s has same UUID"
968 " but different superblock to %s\n",
969 b
, bdevname(refdev
->bdev
, b2
));
973 ev2
= md_event(refsb
);
979 rdev
->sectors
= rdev
->sb_start
;
981 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
982 /* "this cannot possibly happen" ... */
990 * validate_super for 0.90.0
992 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
995 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
996 __u64 ev1
= md_event(sb
);
998 rdev
->raid_disk
= -1;
999 clear_bit(Faulty
, &rdev
->flags
);
1000 clear_bit(In_sync
, &rdev
->flags
);
1001 clear_bit(WriteMostly
, &rdev
->flags
);
1002 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1004 if (mddev
->raid_disks
== 0) {
1005 mddev
->major_version
= 0;
1006 mddev
->minor_version
= sb
->minor_version
;
1007 mddev
->patch_version
= sb
->patch_version
;
1008 mddev
->external
= 0;
1009 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1010 mddev
->ctime
= sb
->ctime
;
1011 mddev
->utime
= sb
->utime
;
1012 mddev
->level
= sb
->level
;
1013 mddev
->clevel
[0] = 0;
1014 mddev
->layout
= sb
->layout
;
1015 mddev
->raid_disks
= sb
->raid_disks
;
1016 mddev
->dev_sectors
= sb
->size
* 2;
1017 mddev
->events
= ev1
;
1018 mddev
->bitmap_info
.offset
= 0;
1019 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1021 if (mddev
->minor_version
>= 91) {
1022 mddev
->reshape_position
= sb
->reshape_position
;
1023 mddev
->delta_disks
= sb
->delta_disks
;
1024 mddev
->new_level
= sb
->new_level
;
1025 mddev
->new_layout
= sb
->new_layout
;
1026 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1028 mddev
->reshape_position
= MaxSector
;
1029 mddev
->delta_disks
= 0;
1030 mddev
->new_level
= mddev
->level
;
1031 mddev
->new_layout
= mddev
->layout
;
1032 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1035 if (sb
->state
& (1<<MD_SB_CLEAN
))
1036 mddev
->recovery_cp
= MaxSector
;
1038 if (sb
->events_hi
== sb
->cp_events_hi
&&
1039 sb
->events_lo
== sb
->cp_events_lo
) {
1040 mddev
->recovery_cp
= sb
->recovery_cp
;
1042 mddev
->recovery_cp
= 0;
1045 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1046 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1047 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1048 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1050 mddev
->max_disks
= MD_SB_DISKS
;
1052 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1053 mddev
->bitmap_info
.file
== NULL
)
1054 mddev
->bitmap_info
.offset
=
1055 mddev
->bitmap_info
.default_offset
;
1057 } else if (mddev
->pers
== NULL
) {
1058 /* Insist on good event counter while assembling */
1060 if (ev1
< mddev
->events
)
1062 } else if (mddev
->bitmap
) {
1063 /* if adding to array with a bitmap, then we can accept an
1064 * older device ... but not too old.
1066 if (ev1
< mddev
->bitmap
->events_cleared
)
1069 if (ev1
< mddev
->events
)
1070 /* just a hot-add of a new device, leave raid_disk at -1 */
1074 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1075 desc
= sb
->disks
+ rdev
->desc_nr
;
1077 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1078 set_bit(Faulty
, &rdev
->flags
);
1079 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1080 desc->raid_disk < mddev->raid_disks */) {
1081 set_bit(In_sync
, &rdev
->flags
);
1082 rdev
->raid_disk
= desc
->raid_disk
;
1083 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1084 /* active but not in sync implies recovery up to
1085 * reshape position. We don't know exactly where
1086 * that is, so set to zero for now */
1087 if (mddev
->minor_version
>= 91) {
1088 rdev
->recovery_offset
= 0;
1089 rdev
->raid_disk
= desc
->raid_disk
;
1092 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1093 set_bit(WriteMostly
, &rdev
->flags
);
1094 } else /* MULTIPATH are always insync */
1095 set_bit(In_sync
, &rdev
->flags
);
1100 * sync_super for 0.90.0
1102 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1106 int next_spare
= mddev
->raid_disks
;
1109 /* make rdev->sb match mddev data..
1112 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1113 * 3/ any empty disks < next_spare become removed
1115 * disks[0] gets initialised to REMOVED because
1116 * we cannot be sure from other fields if it has
1117 * been initialised or not.
1120 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1122 rdev
->sb_size
= MD_SB_BYTES
;
1124 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1126 memset(sb
, 0, sizeof(*sb
));
1128 sb
->md_magic
= MD_SB_MAGIC
;
1129 sb
->major_version
= mddev
->major_version
;
1130 sb
->patch_version
= mddev
->patch_version
;
1131 sb
->gvalid_words
= 0; /* ignored */
1132 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1133 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1134 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1135 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1137 sb
->ctime
= mddev
->ctime
;
1138 sb
->level
= mddev
->level
;
1139 sb
->size
= mddev
->dev_sectors
/ 2;
1140 sb
->raid_disks
= mddev
->raid_disks
;
1141 sb
->md_minor
= mddev
->md_minor
;
1142 sb
->not_persistent
= 0;
1143 sb
->utime
= mddev
->utime
;
1145 sb
->events_hi
= (mddev
->events
>>32);
1146 sb
->events_lo
= (u32
)mddev
->events
;
1148 if (mddev
->reshape_position
== MaxSector
)
1149 sb
->minor_version
= 90;
1151 sb
->minor_version
= 91;
1152 sb
->reshape_position
= mddev
->reshape_position
;
1153 sb
->new_level
= mddev
->new_level
;
1154 sb
->delta_disks
= mddev
->delta_disks
;
1155 sb
->new_layout
= mddev
->new_layout
;
1156 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1158 mddev
->minor_version
= sb
->minor_version
;
1161 sb
->recovery_cp
= mddev
->recovery_cp
;
1162 sb
->cp_events_hi
= (mddev
->events
>>32);
1163 sb
->cp_events_lo
= (u32
)mddev
->events
;
1164 if (mddev
->recovery_cp
== MaxSector
)
1165 sb
->state
= (1<< MD_SB_CLEAN
);
1167 sb
->recovery_cp
= 0;
1169 sb
->layout
= mddev
->layout
;
1170 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1172 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1173 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1175 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1176 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1179 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1181 if (rdev2
->raid_disk
>= 0 &&
1182 sb
->minor_version
>= 91)
1183 /* we have nowhere to store the recovery_offset,
1184 * but if it is not below the reshape_position,
1185 * we can piggy-back on that.
1188 if (rdev2
->raid_disk
< 0 ||
1189 test_bit(Faulty
, &rdev2
->flags
))
1192 desc_nr
= rdev2
->raid_disk
;
1194 desc_nr
= next_spare
++;
1195 rdev2
->desc_nr
= desc_nr
;
1196 d
= &sb
->disks
[rdev2
->desc_nr
];
1198 d
->number
= rdev2
->desc_nr
;
1199 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1200 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1202 d
->raid_disk
= rdev2
->raid_disk
;
1204 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1205 if (test_bit(Faulty
, &rdev2
->flags
))
1206 d
->state
= (1<<MD_DISK_FAULTY
);
1207 else if (is_active
) {
1208 d
->state
= (1<<MD_DISK_ACTIVE
);
1209 if (test_bit(In_sync
, &rdev2
->flags
))
1210 d
->state
|= (1<<MD_DISK_SYNC
);
1218 if (test_bit(WriteMostly
, &rdev2
->flags
))
1219 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1221 /* now set the "removed" and "faulty" bits on any missing devices */
1222 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1223 mdp_disk_t
*d
= &sb
->disks
[i
];
1224 if (d
->state
== 0 && d
->number
== 0) {
1227 d
->state
= (1<<MD_DISK_REMOVED
);
1228 d
->state
|= (1<<MD_DISK_FAULTY
);
1232 sb
->nr_disks
= nr_disks
;
1233 sb
->active_disks
= active
;
1234 sb
->working_disks
= working
;
1235 sb
->failed_disks
= failed
;
1236 sb
->spare_disks
= spare
;
1238 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1239 sb
->sb_csum
= calc_sb_csum(sb
);
1243 * rdev_size_change for 0.90.0
1245 static unsigned long long
1246 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1248 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1249 return 0; /* component must fit device */
1250 if (rdev
->mddev
->bitmap_info
.offset
)
1251 return 0; /* can't move bitmap */
1252 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1253 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1254 num_sectors
= rdev
->sb_start
;
1255 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1257 md_super_wait(rdev
->mddev
);
1263 * version 1 superblock
1266 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1270 unsigned long long newcsum
;
1271 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1272 __le32
*isuper
= (__le32
*)sb
;
1275 disk_csum
= sb
->sb_csum
;
1278 for (i
=0; size
>=4; size
-= 4 )
1279 newcsum
+= le32_to_cpu(*isuper
++);
1282 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1284 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1285 sb
->sb_csum
= disk_csum
;
1286 return cpu_to_le32(csum
);
1289 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1291 struct mdp_superblock_1
*sb
;
1294 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1298 * Calculate the position of the superblock in 512byte sectors.
1299 * It is always aligned to a 4K boundary and
1300 * depeding on minor_version, it can be:
1301 * 0: At least 8K, but less than 12K, from end of device
1302 * 1: At start of device
1303 * 2: 4K from start of device.
1305 switch(minor_version
) {
1307 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1309 sb_start
&= ~(sector_t
)(4*2-1);
1320 rdev
->sb_start
= sb_start
;
1322 /* superblock is rarely larger than 1K, but it can be larger,
1323 * and it is safe to read 4k, so we do that
1325 ret
= read_disk_sb(rdev
, 4096);
1326 if (ret
) return ret
;
1329 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1331 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1332 sb
->major_version
!= cpu_to_le32(1) ||
1333 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1334 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1335 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1338 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1339 printk("md: invalid superblock checksum on %s\n",
1340 bdevname(rdev
->bdev
,b
));
1343 if (le64_to_cpu(sb
->data_size
) < 10) {
1344 printk("md: data_size too small on %s\n",
1345 bdevname(rdev
->bdev
,b
));
1349 rdev
->preferred_minor
= 0xffff;
1350 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1351 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1353 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1354 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1355 if (rdev
->sb_size
& bmask
)
1356 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1359 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1362 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1365 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1371 struct mdp_superblock_1
*refsb
=
1372 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1374 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1375 sb
->level
!= refsb
->level
||
1376 sb
->layout
!= refsb
->layout
||
1377 sb
->chunksize
!= refsb
->chunksize
) {
1378 printk(KERN_WARNING
"md: %s has strangely different"
1379 " superblock to %s\n",
1380 bdevname(rdev
->bdev
,b
),
1381 bdevname(refdev
->bdev
,b2
));
1384 ev1
= le64_to_cpu(sb
->events
);
1385 ev2
= le64_to_cpu(refsb
->events
);
1393 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1394 le64_to_cpu(sb
->data_offset
);
1396 rdev
->sectors
= rdev
->sb_start
;
1397 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1399 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1400 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1405 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1407 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1408 __u64 ev1
= le64_to_cpu(sb
->events
);
1410 rdev
->raid_disk
= -1;
1411 clear_bit(Faulty
, &rdev
->flags
);
1412 clear_bit(In_sync
, &rdev
->flags
);
1413 clear_bit(WriteMostly
, &rdev
->flags
);
1414 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1416 if (mddev
->raid_disks
== 0) {
1417 mddev
->major_version
= 1;
1418 mddev
->patch_version
= 0;
1419 mddev
->external
= 0;
1420 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1421 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1422 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1423 mddev
->level
= le32_to_cpu(sb
->level
);
1424 mddev
->clevel
[0] = 0;
1425 mddev
->layout
= le32_to_cpu(sb
->layout
);
1426 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1427 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1428 mddev
->events
= ev1
;
1429 mddev
->bitmap_info
.offset
= 0;
1430 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1432 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1433 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1435 mddev
->max_disks
= (4096-256)/2;
1437 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1438 mddev
->bitmap_info
.file
== NULL
)
1439 mddev
->bitmap_info
.offset
=
1440 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1442 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1443 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1444 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1445 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1446 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1447 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1449 mddev
->reshape_position
= MaxSector
;
1450 mddev
->delta_disks
= 0;
1451 mddev
->new_level
= mddev
->level
;
1452 mddev
->new_layout
= mddev
->layout
;
1453 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1456 } else if (mddev
->pers
== NULL
) {
1457 /* Insist of good event counter while assembling */
1459 if (ev1
< mddev
->events
)
1461 } else if (mddev
->bitmap
) {
1462 /* If adding to array with a bitmap, then we can accept an
1463 * older device, but not too old.
1465 if (ev1
< mddev
->bitmap
->events_cleared
)
1468 if (ev1
< mddev
->events
)
1469 /* just a hot-add of a new device, leave raid_disk at -1 */
1472 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1474 if (rdev
->desc_nr
< 0 ||
1475 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1479 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1481 case 0xffff: /* spare */
1483 case 0xfffe: /* faulty */
1484 set_bit(Faulty
, &rdev
->flags
);
1487 if ((le32_to_cpu(sb
->feature_map
) &
1488 MD_FEATURE_RECOVERY_OFFSET
))
1489 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1491 set_bit(In_sync
, &rdev
->flags
);
1492 rdev
->raid_disk
= role
;
1495 if (sb
->devflags
& WriteMostly1
)
1496 set_bit(WriteMostly
, &rdev
->flags
);
1497 } else /* MULTIPATH are always insync */
1498 set_bit(In_sync
, &rdev
->flags
);
1503 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1505 struct mdp_superblock_1
*sb
;
1508 /* make rdev->sb match mddev and rdev data. */
1510 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1512 sb
->feature_map
= 0;
1514 sb
->recovery_offset
= cpu_to_le64(0);
1515 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1516 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1517 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1519 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1520 sb
->events
= cpu_to_le64(mddev
->events
);
1522 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1524 sb
->resync_offset
= cpu_to_le64(0);
1526 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1528 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1529 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1530 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1531 sb
->level
= cpu_to_le32(mddev
->level
);
1532 sb
->layout
= cpu_to_le32(mddev
->layout
);
1534 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1535 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1536 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1539 if (rdev
->raid_disk
>= 0 &&
1540 !test_bit(In_sync
, &rdev
->flags
)) {
1542 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1543 sb
->recovery_offset
=
1544 cpu_to_le64(rdev
->recovery_offset
);
1547 if (mddev
->reshape_position
!= MaxSector
) {
1548 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1549 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1550 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1551 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1552 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1553 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1557 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1558 if (rdev2
->desc_nr
+1 > max_dev
)
1559 max_dev
= rdev2
->desc_nr
+1;
1561 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1563 sb
->max_dev
= cpu_to_le32(max_dev
);
1564 rdev
->sb_size
= max_dev
* 2 + 256;
1565 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1566 if (rdev
->sb_size
& bmask
)
1567 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1569 for (i
=0; i
<max_dev
;i
++)
1570 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1572 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1574 if (test_bit(Faulty
, &rdev2
->flags
))
1575 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1576 else if (test_bit(In_sync
, &rdev2
->flags
))
1577 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1578 else if (rdev2
->raid_disk
>= 0)
1579 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1581 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1584 sb
->sb_csum
= calc_sb_1_csum(sb
);
1587 static unsigned long long
1588 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1590 struct mdp_superblock_1
*sb
;
1591 sector_t max_sectors
;
1592 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1593 return 0; /* component must fit device */
1594 if (rdev
->sb_start
< rdev
->data_offset
) {
1595 /* minor versions 1 and 2; superblock before data */
1596 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1597 max_sectors
-= rdev
->data_offset
;
1598 if (!num_sectors
|| num_sectors
> max_sectors
)
1599 num_sectors
= max_sectors
;
1600 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1601 /* minor version 0 with bitmap we can't move */
1604 /* minor version 0; superblock after data */
1606 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1607 sb_start
&= ~(sector_t
)(4*2 - 1);
1608 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1609 if (!num_sectors
|| num_sectors
> max_sectors
)
1610 num_sectors
= max_sectors
;
1611 rdev
->sb_start
= sb_start
;
1613 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1614 sb
->data_size
= cpu_to_le64(num_sectors
);
1615 sb
->super_offset
= rdev
->sb_start
;
1616 sb
->sb_csum
= calc_sb_1_csum(sb
);
1617 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1619 md_super_wait(rdev
->mddev
);
1623 static struct super_type super_types
[] = {
1626 .owner
= THIS_MODULE
,
1627 .load_super
= super_90_load
,
1628 .validate_super
= super_90_validate
,
1629 .sync_super
= super_90_sync
,
1630 .rdev_size_change
= super_90_rdev_size_change
,
1634 .owner
= THIS_MODULE
,
1635 .load_super
= super_1_load
,
1636 .validate_super
= super_1_validate
,
1637 .sync_super
= super_1_sync
,
1638 .rdev_size_change
= super_1_rdev_size_change
,
1642 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1644 mdk_rdev_t
*rdev
, *rdev2
;
1647 rdev_for_each_rcu(rdev
, mddev1
)
1648 rdev_for_each_rcu(rdev2
, mddev2
)
1649 if (rdev
->bdev
->bd_contains
==
1650 rdev2
->bdev
->bd_contains
) {
1658 static LIST_HEAD(pending_raid_disks
);
1661 * Try to register data integrity profile for an mddev
1663 * This is called when an array is started and after a disk has been kicked
1664 * from the array. It only succeeds if all working and active component devices
1665 * are integrity capable with matching profiles.
1667 int md_integrity_register(mddev_t
*mddev
)
1669 mdk_rdev_t
*rdev
, *reference
= NULL
;
1671 if (list_empty(&mddev
->disks
))
1672 return 0; /* nothing to do */
1673 if (blk_get_integrity(mddev
->gendisk
))
1674 return 0; /* already registered */
1675 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1676 /* skip spares and non-functional disks */
1677 if (test_bit(Faulty
, &rdev
->flags
))
1679 if (rdev
->raid_disk
< 0)
1682 * If at least one rdev is not integrity capable, we can not
1683 * enable data integrity for the md device.
1685 if (!bdev_get_integrity(rdev
->bdev
))
1688 /* Use the first rdev as the reference */
1692 /* does this rdev's profile match the reference profile? */
1693 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1694 rdev
->bdev
->bd_disk
) < 0)
1698 * All component devices are integrity capable and have matching
1699 * profiles, register the common profile for the md device.
1701 if (blk_integrity_register(mddev
->gendisk
,
1702 bdev_get_integrity(reference
->bdev
)) != 0) {
1703 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1707 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1711 EXPORT_SYMBOL(md_integrity_register
);
1713 /* Disable data integrity if non-capable/non-matching disk is being added */
1714 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1716 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1717 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1719 if (!bi_mddev
) /* nothing to do */
1721 if (rdev
->raid_disk
< 0) /* skip spares */
1723 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1724 rdev
->bdev
->bd_disk
) >= 0)
1726 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1727 blk_integrity_unregister(mddev
->gendisk
);
1729 EXPORT_SYMBOL(md_integrity_add_rdev
);
1731 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1733 char b
[BDEVNAME_SIZE
];
1743 /* prevent duplicates */
1744 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1747 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1748 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1749 rdev
->sectors
< mddev
->dev_sectors
)) {
1751 /* Cannot change size, so fail
1752 * If mddev->level <= 0, then we don't care
1753 * about aligning sizes (e.g. linear)
1755 if (mddev
->level
> 0)
1758 mddev
->dev_sectors
= rdev
->sectors
;
1761 /* Verify rdev->desc_nr is unique.
1762 * If it is -1, assign a free number, else
1763 * check number is not in use
1765 if (rdev
->desc_nr
< 0) {
1767 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1768 while (find_rdev_nr(mddev
, choice
))
1770 rdev
->desc_nr
= choice
;
1772 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1775 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1776 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1777 mdname(mddev
), mddev
->max_disks
);
1780 bdevname(rdev
->bdev
,b
);
1781 while ( (s
=strchr(b
, '/')) != NULL
)
1784 rdev
->mddev
= mddev
;
1785 printk(KERN_INFO
"md: bind<%s>\n", b
);
1787 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1790 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1791 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1792 kobject_del(&rdev
->kobj
);
1795 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1797 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1798 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1800 /* May as well allow recovery to be retried once */
1801 mddev
->recovery_disabled
= 0;
1806 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1811 static void md_delayed_delete(struct work_struct
*ws
)
1813 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1814 kobject_del(&rdev
->kobj
);
1815 kobject_put(&rdev
->kobj
);
1818 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1820 char b
[BDEVNAME_SIZE
];
1825 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1826 list_del_rcu(&rdev
->same_set
);
1827 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1829 sysfs_remove_link(&rdev
->kobj
, "block");
1830 sysfs_put(rdev
->sysfs_state
);
1831 rdev
->sysfs_state
= NULL
;
1832 /* We need to delay this, otherwise we can deadlock when
1833 * writing to 'remove' to "dev/state". We also need
1834 * to delay it due to rcu usage.
1837 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1838 kobject_get(&rdev
->kobj
);
1839 schedule_work(&rdev
->del_work
);
1843 * prevent the device from being mounted, repartitioned or
1844 * otherwise reused by a RAID array (or any other kernel
1845 * subsystem), by bd_claiming the device.
1847 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1850 struct block_device
*bdev
;
1851 char b
[BDEVNAME_SIZE
];
1853 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1855 printk(KERN_ERR
"md: could not open %s.\n",
1856 __bdevname(dev
, b
));
1857 return PTR_ERR(bdev
);
1859 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1861 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1863 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1867 set_bit(AllReserved
, &rdev
->flags
);
1872 static void unlock_rdev(mdk_rdev_t
*rdev
)
1874 struct block_device
*bdev
= rdev
->bdev
;
1879 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1882 void md_autodetect_dev(dev_t dev
);
1884 static void export_rdev(mdk_rdev_t
* rdev
)
1886 char b
[BDEVNAME_SIZE
];
1887 printk(KERN_INFO
"md: export_rdev(%s)\n",
1888 bdevname(rdev
->bdev
,b
));
1893 if (test_bit(AutoDetected
, &rdev
->flags
))
1894 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1897 kobject_put(&rdev
->kobj
);
1900 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1902 unbind_rdev_from_array(rdev
);
1906 static void export_array(mddev_t
*mddev
)
1908 mdk_rdev_t
*rdev
, *tmp
;
1910 rdev_for_each(rdev
, tmp
, mddev
) {
1915 kick_rdev_from_array(rdev
);
1917 if (!list_empty(&mddev
->disks
))
1919 mddev
->raid_disks
= 0;
1920 mddev
->major_version
= 0;
1923 static void print_desc(mdp_disk_t
*desc
)
1925 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1926 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1929 static void print_sb_90(mdp_super_t
*sb
)
1934 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1935 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1936 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1938 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1939 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1940 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1941 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1942 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1943 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1944 sb
->failed_disks
, sb
->spare_disks
,
1945 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1948 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1951 desc
= sb
->disks
+ i
;
1952 if (desc
->number
|| desc
->major
|| desc
->minor
||
1953 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1954 printk(" D %2d: ", i
);
1958 printk(KERN_INFO
"md: THIS: ");
1959 print_desc(&sb
->this_disk
);
1962 static void print_sb_1(struct mdp_superblock_1
*sb
)
1966 uuid
= sb
->set_uuid
;
1968 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1969 "md: Name: \"%s\" CT:%llu\n",
1970 le32_to_cpu(sb
->major_version
),
1971 le32_to_cpu(sb
->feature_map
),
1974 (unsigned long long)le64_to_cpu(sb
->ctime
)
1975 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1977 uuid
= sb
->device_uuid
;
1979 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1981 "md: Dev:%08x UUID: %pU\n"
1982 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1983 "md: (MaxDev:%u) \n",
1984 le32_to_cpu(sb
->level
),
1985 (unsigned long long)le64_to_cpu(sb
->size
),
1986 le32_to_cpu(sb
->raid_disks
),
1987 le32_to_cpu(sb
->layout
),
1988 le32_to_cpu(sb
->chunksize
),
1989 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1990 (unsigned long long)le64_to_cpu(sb
->data_size
),
1991 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1992 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1993 le32_to_cpu(sb
->dev_number
),
1996 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1997 (unsigned long long)le64_to_cpu(sb
->events
),
1998 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
1999 le32_to_cpu(sb
->sb_csum
),
2000 le32_to_cpu(sb
->max_dev
)
2004 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2006 char b
[BDEVNAME_SIZE
];
2007 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2008 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2009 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2011 if (rdev
->sb_loaded
) {
2012 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2013 switch (major_version
) {
2015 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
2018 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
2022 printk(KERN_INFO
"md: no rdev superblock!\n");
2025 static void md_print_devices(void)
2027 struct list_head
*tmp
;
2030 char b
[BDEVNAME_SIZE
];
2033 printk("md: **********************************\n");
2034 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2035 printk("md: **********************************\n");
2036 for_each_mddev(mddev
, tmp
) {
2039 bitmap_print_sb(mddev
->bitmap
);
2041 printk("%s: ", mdname(mddev
));
2042 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2043 printk("<%s>", bdevname(rdev
->bdev
,b
));
2046 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2047 print_rdev(rdev
, mddev
->major_version
);
2049 printk("md: **********************************\n");
2054 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2056 /* Update each superblock (in-memory image), but
2057 * if we are allowed to, skip spares which already
2058 * have the right event counter, or have one earlier
2059 * (which would mean they aren't being marked as dirty
2060 * with the rest of the array)
2064 /* First make sure individual recovery_offsets are correct */
2065 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2066 if (rdev
->raid_disk
>= 0 &&
2067 !test_bit(In_sync
, &rdev
->flags
) &&
2068 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2069 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2072 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2073 if (rdev
->sb_events
== mddev
->events
||
2075 rdev
->raid_disk
< 0 &&
2076 (rdev
->sb_events
&1)==0 &&
2077 rdev
->sb_events
+1 == mddev
->events
)) {
2078 /* Don't update this superblock */
2079 rdev
->sb_loaded
= 2;
2081 super_types
[mddev
->major_version
].
2082 sync_super(mddev
, rdev
);
2083 rdev
->sb_loaded
= 1;
2088 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2094 mddev
->utime
= get_seconds();
2095 if (mddev
->external
)
2098 spin_lock_irq(&mddev
->write_lock
);
2100 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2101 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2103 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2104 /* just a clean<-> dirty transition, possibly leave spares alone,
2105 * though if events isn't the right even/odd, we will have to do
2111 if (mddev
->degraded
)
2112 /* If the array is degraded, then skipping spares is both
2113 * dangerous and fairly pointless.
2114 * Dangerous because a device that was removed from the array
2115 * might have a event_count that still looks up-to-date,
2116 * so it can be re-added without a resync.
2117 * Pointless because if there are any spares to skip,
2118 * then a recovery will happen and soon that array won't
2119 * be degraded any more and the spare can go back to sleep then.
2123 sync_req
= mddev
->in_sync
;
2125 /* If this is just a dirty<->clean transition, and the array is clean
2126 * and 'events' is odd, we can roll back to the previous clean state */
2128 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2129 && (mddev
->events
& 1)
2130 && mddev
->events
!= 1)
2133 /* otherwise we have to go forward and ... */
2135 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
2136 /* .. if the array isn't clean, an 'even' event must also go
2138 if ((mddev
->events
&1)==0) {
2140 sync_req
= 2; /* force a second update to get the
2141 * even/odd in sync */
2144 /* otherwise an 'odd' event must go to spares */
2145 if ((mddev
->events
&1)) {
2147 sync_req
= 2; /* force a second update to get the
2148 * even/odd in sync */
2153 if (!mddev
->events
) {
2155 * oops, this 64-bit counter should never wrap.
2156 * Either we are in around ~1 trillion A.C., assuming
2157 * 1 reboot per second, or we have a bug:
2164 * do not write anything to disk if using
2165 * nonpersistent superblocks
2167 if (!mddev
->persistent
) {
2168 if (!mddev
->external
)
2169 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2171 spin_unlock_irq(&mddev
->write_lock
);
2172 wake_up(&mddev
->sb_wait
);
2175 sync_sbs(mddev
, nospares
);
2176 spin_unlock_irq(&mddev
->write_lock
);
2179 "md: updating %s RAID superblock on device (in sync %d)\n",
2180 mdname(mddev
),mddev
->in_sync
);
2182 bitmap_update_sb(mddev
->bitmap
);
2183 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2184 char b
[BDEVNAME_SIZE
];
2185 dprintk(KERN_INFO
"md: ");
2186 if (rdev
->sb_loaded
!= 1)
2187 continue; /* no noise on spare devices */
2188 if (test_bit(Faulty
, &rdev
->flags
))
2189 dprintk("(skipping faulty ");
2191 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2192 if (!test_bit(Faulty
, &rdev
->flags
)) {
2193 md_super_write(mddev
,rdev
,
2194 rdev
->sb_start
, rdev
->sb_size
,
2196 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2197 bdevname(rdev
->bdev
,b
),
2198 (unsigned long long)rdev
->sb_start
);
2199 rdev
->sb_events
= mddev
->events
;
2203 if (mddev
->level
== LEVEL_MULTIPATH
)
2204 /* only need to write one superblock... */
2207 md_super_wait(mddev
);
2208 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2210 spin_lock_irq(&mddev
->write_lock
);
2211 if (mddev
->in_sync
!= sync_req
||
2212 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2213 /* have to write it out again */
2214 spin_unlock_irq(&mddev
->write_lock
);
2217 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2218 spin_unlock_irq(&mddev
->write_lock
);
2219 wake_up(&mddev
->sb_wait
);
2220 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2221 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2225 /* words written to sysfs files may, or may not, be \n terminated.
2226 * We want to accept with case. For this we use cmd_match.
2228 static int cmd_match(const char *cmd
, const char *str
)
2230 /* See if cmd, written into a sysfs file, matches
2231 * str. They must either be the same, or cmd can
2232 * have a trailing newline
2234 while (*cmd
&& *str
&& *cmd
== *str
) {
2245 struct rdev_sysfs_entry
{
2246 struct attribute attr
;
2247 ssize_t (*show
)(mdk_rdev_t
*, char *);
2248 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2252 state_show(mdk_rdev_t
*rdev
, char *page
)
2257 if (test_bit(Faulty
, &rdev
->flags
)) {
2258 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2261 if (test_bit(In_sync
, &rdev
->flags
)) {
2262 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2265 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2266 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2269 if (test_bit(Blocked
, &rdev
->flags
)) {
2270 len
+= sprintf(page
+len
, "%sblocked", sep
);
2273 if (!test_bit(Faulty
, &rdev
->flags
) &&
2274 !test_bit(In_sync
, &rdev
->flags
)) {
2275 len
+= sprintf(page
+len
, "%sspare", sep
);
2278 return len
+sprintf(page
+len
, "\n");
2282 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2285 * faulty - simulates and error
2286 * remove - disconnects the device
2287 * writemostly - sets write_mostly
2288 * -writemostly - clears write_mostly
2289 * blocked - sets the Blocked flag
2290 * -blocked - clears the Blocked flag
2291 * insync - sets Insync providing device isn't active
2294 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2295 md_error(rdev
->mddev
, rdev
);
2297 } else if (cmd_match(buf
, "remove")) {
2298 if (rdev
->raid_disk
>= 0)
2301 mddev_t
*mddev
= rdev
->mddev
;
2302 kick_rdev_from_array(rdev
);
2304 md_update_sb(mddev
, 1);
2305 md_new_event(mddev
);
2308 } else if (cmd_match(buf
, "writemostly")) {
2309 set_bit(WriteMostly
, &rdev
->flags
);
2311 } else if (cmd_match(buf
, "-writemostly")) {
2312 clear_bit(WriteMostly
, &rdev
->flags
);
2314 } else if (cmd_match(buf
, "blocked")) {
2315 set_bit(Blocked
, &rdev
->flags
);
2317 } else if (cmd_match(buf
, "-blocked")) {
2318 clear_bit(Blocked
, &rdev
->flags
);
2319 wake_up(&rdev
->blocked_wait
);
2320 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2321 md_wakeup_thread(rdev
->mddev
->thread
);
2324 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2325 set_bit(In_sync
, &rdev
->flags
);
2328 if (!err
&& rdev
->sysfs_state
)
2329 sysfs_notify_dirent(rdev
->sysfs_state
);
2330 return err
? err
: len
;
2332 static struct rdev_sysfs_entry rdev_state
=
2333 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2336 errors_show(mdk_rdev_t
*rdev
, char *page
)
2338 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2342 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2345 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2346 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2347 atomic_set(&rdev
->corrected_errors
, n
);
2352 static struct rdev_sysfs_entry rdev_errors
=
2353 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2356 slot_show(mdk_rdev_t
*rdev
, char *page
)
2358 if (rdev
->raid_disk
< 0)
2359 return sprintf(page
, "none\n");
2361 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2365 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2370 int slot
= simple_strtoul(buf
, &e
, 10);
2371 if (strncmp(buf
, "none", 4)==0)
2373 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2375 if (rdev
->mddev
->pers
&& slot
== -1) {
2376 /* Setting 'slot' on an active array requires also
2377 * updating the 'rd%d' link, and communicating
2378 * with the personality with ->hot_*_disk.
2379 * For now we only support removing
2380 * failed/spare devices. This normally happens automatically,
2381 * but not when the metadata is externally managed.
2383 if (rdev
->raid_disk
== -1)
2385 /* personality does all needed checks */
2386 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2388 err
= rdev
->mddev
->pers
->
2389 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2392 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2393 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2394 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2395 md_wakeup_thread(rdev
->mddev
->thread
);
2396 } else if (rdev
->mddev
->pers
) {
2398 /* Activating a spare .. or possibly reactivating
2399 * if we ever get bitmaps working here.
2402 if (rdev
->raid_disk
!= -1)
2405 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2408 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2409 if (rdev2
->raid_disk
== slot
)
2412 rdev
->raid_disk
= slot
;
2413 if (test_bit(In_sync
, &rdev
->flags
))
2414 rdev
->saved_raid_disk
= slot
;
2416 rdev
->saved_raid_disk
= -1;
2417 err
= rdev
->mddev
->pers
->
2418 hot_add_disk(rdev
->mddev
, rdev
);
2420 rdev
->raid_disk
= -1;
2423 sysfs_notify_dirent(rdev
->sysfs_state
);
2424 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2425 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2427 "md: cannot register "
2429 nm
, mdname(rdev
->mddev
));
2431 /* don't wakeup anyone, leave that to userspace. */
2433 if (slot
>= rdev
->mddev
->raid_disks
)
2435 rdev
->raid_disk
= slot
;
2436 /* assume it is working */
2437 clear_bit(Faulty
, &rdev
->flags
);
2438 clear_bit(WriteMostly
, &rdev
->flags
);
2439 set_bit(In_sync
, &rdev
->flags
);
2440 sysfs_notify_dirent(rdev
->sysfs_state
);
2446 static struct rdev_sysfs_entry rdev_slot
=
2447 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2450 offset_show(mdk_rdev_t
*rdev
, char *page
)
2452 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2456 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2459 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2460 if (e
==buf
|| (*e
&& *e
!= '\n'))
2462 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2464 if (rdev
->sectors
&& rdev
->mddev
->external
)
2465 /* Must set offset before size, so overlap checks
2468 rdev
->data_offset
= offset
;
2472 static struct rdev_sysfs_entry rdev_offset
=
2473 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2476 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2478 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2481 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2483 /* check if two start/length pairs overlap */
2491 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2493 unsigned long long blocks
;
2496 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2499 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2500 return -EINVAL
; /* sector conversion overflow */
2503 if (new != blocks
* 2)
2504 return -EINVAL
; /* unsigned long long to sector_t overflow */
2511 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2513 mddev_t
*my_mddev
= rdev
->mddev
;
2514 sector_t oldsectors
= rdev
->sectors
;
2517 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2519 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2520 if (my_mddev
->persistent
) {
2521 sectors
= super_types
[my_mddev
->major_version
].
2522 rdev_size_change(rdev
, sectors
);
2525 } else if (!sectors
)
2526 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2529 if (sectors
< my_mddev
->dev_sectors
)
2530 return -EINVAL
; /* component must fit device */
2532 rdev
->sectors
= sectors
;
2533 if (sectors
> oldsectors
&& my_mddev
->external
) {
2534 /* need to check that all other rdevs with the same ->bdev
2535 * do not overlap. We need to unlock the mddev to avoid
2536 * a deadlock. We have already changed rdev->sectors, and if
2537 * we have to change it back, we will have the lock again.
2541 struct list_head
*tmp
;
2543 mddev_unlock(my_mddev
);
2544 for_each_mddev(mddev
, tmp
) {
2548 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2549 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2550 (rdev
->bdev
== rdev2
->bdev
&&
2552 overlaps(rdev
->data_offset
, rdev
->sectors
,
2558 mddev_unlock(mddev
);
2564 mddev_lock(my_mddev
);
2566 /* Someone else could have slipped in a size
2567 * change here, but doing so is just silly.
2568 * We put oldsectors back because we *know* it is
2569 * safe, and trust userspace not to race with
2572 rdev
->sectors
= oldsectors
;
2579 static struct rdev_sysfs_entry rdev_size
=
2580 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2583 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2585 unsigned long long recovery_start
= rdev
->recovery_offset
;
2587 if (test_bit(In_sync
, &rdev
->flags
) ||
2588 recovery_start
== MaxSector
)
2589 return sprintf(page
, "none\n");
2591 return sprintf(page
, "%llu\n", recovery_start
);
2594 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2596 unsigned long long recovery_start
;
2598 if (cmd_match(buf
, "none"))
2599 recovery_start
= MaxSector
;
2600 else if (strict_strtoull(buf
, 10, &recovery_start
))
2603 if (rdev
->mddev
->pers
&&
2604 rdev
->raid_disk
>= 0)
2607 rdev
->recovery_offset
= recovery_start
;
2608 if (recovery_start
== MaxSector
)
2609 set_bit(In_sync
, &rdev
->flags
);
2611 clear_bit(In_sync
, &rdev
->flags
);
2615 static struct rdev_sysfs_entry rdev_recovery_start
=
2616 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2618 static struct attribute
*rdev_default_attrs
[] = {
2624 &rdev_recovery_start
.attr
,
2628 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2630 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2631 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2632 mddev_t
*mddev
= rdev
->mddev
;
2638 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2640 if (rdev
->mddev
== NULL
)
2643 rv
= entry
->show(rdev
, page
);
2644 mddev_unlock(mddev
);
2650 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2651 const char *page
, size_t length
)
2653 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2654 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2656 mddev_t
*mddev
= rdev
->mddev
;
2660 if (!capable(CAP_SYS_ADMIN
))
2662 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2664 if (rdev
->mddev
== NULL
)
2667 rv
= entry
->store(rdev
, page
, length
);
2668 mddev_unlock(mddev
);
2673 static void rdev_free(struct kobject
*ko
)
2675 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2678 static struct sysfs_ops rdev_sysfs_ops
= {
2679 .show
= rdev_attr_show
,
2680 .store
= rdev_attr_store
,
2682 static struct kobj_type rdev_ktype
= {
2683 .release
= rdev_free
,
2684 .sysfs_ops
= &rdev_sysfs_ops
,
2685 .default_attrs
= rdev_default_attrs
,
2689 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2691 * mark the device faulty if:
2693 * - the device is nonexistent (zero size)
2694 * - the device has no valid superblock
2696 * a faulty rdev _never_ has rdev->sb set.
2698 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2700 char b
[BDEVNAME_SIZE
];
2705 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2707 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2708 return ERR_PTR(-ENOMEM
);
2711 if ((err
= alloc_disk_sb(rdev
)))
2714 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2718 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2721 rdev
->saved_raid_disk
= -1;
2722 rdev
->raid_disk
= -1;
2724 rdev
->data_offset
= 0;
2725 rdev
->sb_events
= 0;
2726 rdev
->last_read_error
.tv_sec
= 0;
2727 rdev
->last_read_error
.tv_nsec
= 0;
2728 atomic_set(&rdev
->nr_pending
, 0);
2729 atomic_set(&rdev
->read_errors
, 0);
2730 atomic_set(&rdev
->corrected_errors
, 0);
2732 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2735 "md: %s has zero or unknown size, marking faulty!\n",
2736 bdevname(rdev
->bdev
,b
));
2741 if (super_format
>= 0) {
2742 err
= super_types
[super_format
].
2743 load_super(rdev
, NULL
, super_minor
);
2744 if (err
== -EINVAL
) {
2746 "md: %s does not have a valid v%d.%d "
2747 "superblock, not importing!\n",
2748 bdevname(rdev
->bdev
,b
),
2749 super_format
, super_minor
);
2754 "md: could not read %s's sb, not importing!\n",
2755 bdevname(rdev
->bdev
,b
));
2760 INIT_LIST_HEAD(&rdev
->same_set
);
2761 init_waitqueue_head(&rdev
->blocked_wait
);
2766 if (rdev
->sb_page
) {
2772 return ERR_PTR(err
);
2776 * Check a full RAID array for plausibility
2780 static void analyze_sbs(mddev_t
* mddev
)
2783 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2784 char b
[BDEVNAME_SIZE
];
2787 rdev_for_each(rdev
, tmp
, mddev
)
2788 switch (super_types
[mddev
->major_version
].
2789 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2797 "md: fatal superblock inconsistency in %s"
2798 " -- removing from array\n",
2799 bdevname(rdev
->bdev
,b
));
2800 kick_rdev_from_array(rdev
);
2804 super_types
[mddev
->major_version
].
2805 validate_super(mddev
, freshest
);
2808 rdev_for_each(rdev
, tmp
, mddev
) {
2809 if (rdev
->desc_nr
>= mddev
->max_disks
||
2810 i
> mddev
->max_disks
) {
2812 "md: %s: %s: only %d devices permitted\n",
2813 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2815 kick_rdev_from_array(rdev
);
2818 if (rdev
!= freshest
)
2819 if (super_types
[mddev
->major_version
].
2820 validate_super(mddev
, rdev
)) {
2821 printk(KERN_WARNING
"md: kicking non-fresh %s"
2823 bdevname(rdev
->bdev
,b
));
2824 kick_rdev_from_array(rdev
);
2827 if (mddev
->level
== LEVEL_MULTIPATH
) {
2828 rdev
->desc_nr
= i
++;
2829 rdev
->raid_disk
= rdev
->desc_nr
;
2830 set_bit(In_sync
, &rdev
->flags
);
2831 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2832 rdev
->raid_disk
= -1;
2833 clear_bit(In_sync
, &rdev
->flags
);
2838 /* Read a fixed-point number.
2839 * Numbers in sysfs attributes should be in "standard" units where
2840 * possible, so time should be in seconds.
2841 * However we internally use a a much smaller unit such as
2842 * milliseconds or jiffies.
2843 * This function takes a decimal number with a possible fractional
2844 * component, and produces an integer which is the result of
2845 * multiplying that number by 10^'scale'.
2846 * all without any floating-point arithmetic.
2848 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
2850 unsigned long result
= 0;
2852 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
2855 else if (decimals
< scale
) {
2858 result
= result
* 10 + value
;
2870 while (decimals
< scale
) {
2879 static void md_safemode_timeout(unsigned long data
);
2882 safe_delay_show(mddev_t
*mddev
, char *page
)
2884 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2885 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2888 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2892 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
2895 mddev
->safemode_delay
= 0;
2897 unsigned long old_delay
= mddev
->safemode_delay
;
2898 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2899 if (mddev
->safemode_delay
== 0)
2900 mddev
->safemode_delay
= 1;
2901 if (mddev
->safemode_delay
< old_delay
)
2902 md_safemode_timeout((unsigned long)mddev
);
2906 static struct md_sysfs_entry md_safe_delay
=
2907 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2910 level_show(mddev_t
*mddev
, char *page
)
2912 struct mdk_personality
*p
= mddev
->pers
;
2914 return sprintf(page
, "%s\n", p
->name
);
2915 else if (mddev
->clevel
[0])
2916 return sprintf(page
, "%s\n", mddev
->clevel
);
2917 else if (mddev
->level
!= LEVEL_NONE
)
2918 return sprintf(page
, "%d\n", mddev
->level
);
2924 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2928 struct mdk_personality
*pers
;
2932 if (mddev
->pers
== NULL
) {
2935 if (len
>= sizeof(mddev
->clevel
))
2937 strncpy(mddev
->clevel
, buf
, len
);
2938 if (mddev
->clevel
[len
-1] == '\n')
2940 mddev
->clevel
[len
] = 0;
2941 mddev
->level
= LEVEL_NONE
;
2945 /* request to change the personality. Need to ensure:
2946 * - array is not engaged in resync/recovery/reshape
2947 * - old personality can be suspended
2948 * - new personality will access other array.
2951 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2954 if (!mddev
->pers
->quiesce
) {
2955 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2956 mdname(mddev
), mddev
->pers
->name
);
2960 /* Now find the new personality */
2961 if (len
== 0 || len
>= sizeof(level
))
2963 strncpy(level
, buf
, len
);
2964 if (level
[len
-1] == '\n')
2968 request_module("md-%s", level
);
2969 spin_lock(&pers_lock
);
2970 pers
= find_pers(LEVEL_NONE
, level
);
2971 if (!pers
|| !try_module_get(pers
->owner
)) {
2972 spin_unlock(&pers_lock
);
2973 printk(KERN_WARNING
"md: personality %s not loaded\n", level
);
2976 spin_unlock(&pers_lock
);
2978 if (pers
== mddev
->pers
) {
2979 /* Nothing to do! */
2980 module_put(pers
->owner
);
2983 if (!pers
->takeover
) {
2984 module_put(pers
->owner
);
2985 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2986 mdname(mddev
), level
);
2990 /* ->takeover must set new_* and/or delta_disks
2991 * if it succeeds, and may set them when it fails.
2993 priv
= pers
->takeover(mddev
);
2995 mddev
->new_level
= mddev
->level
;
2996 mddev
->new_layout
= mddev
->layout
;
2997 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2998 mddev
->raid_disks
-= mddev
->delta_disks
;
2999 mddev
->delta_disks
= 0;
3000 module_put(pers
->owner
);
3001 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3002 mdname(mddev
), level
);
3003 return PTR_ERR(priv
);
3006 /* Looks like we have a winner */
3007 mddev_suspend(mddev
);
3008 mddev
->pers
->stop(mddev
);
3010 if (mddev
->pers
->sync_request
== NULL
&&
3011 pers
->sync_request
!= NULL
) {
3012 /* need to add the md_redundancy_group */
3013 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3015 "md: cannot register extra attributes for %s\n",
3017 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3019 if (mddev
->pers
->sync_request
!= NULL
&&
3020 pers
->sync_request
== NULL
) {
3021 /* need to remove the md_redundancy_group */
3022 if (mddev
->to_remove
== NULL
)
3023 mddev
->to_remove
= &md_redundancy_group
;
3026 module_put(mddev
->pers
->owner
);
3027 /* Invalidate devices that are now superfluous */
3028 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3029 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
3030 rdev
->raid_disk
= -1;
3031 clear_bit(In_sync
, &rdev
->flags
);
3034 mddev
->private = priv
;
3035 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3036 mddev
->level
= mddev
->new_level
;
3037 mddev
->layout
= mddev
->new_layout
;
3038 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3039 mddev
->delta_disks
= 0;
3041 mddev_resume(mddev
);
3042 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3043 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3044 md_wakeup_thread(mddev
->thread
);
3048 static struct md_sysfs_entry md_level
=
3049 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3053 layout_show(mddev_t
*mddev
, char *page
)
3055 /* just a number, not meaningful for all levels */
3056 if (mddev
->reshape_position
!= MaxSector
&&
3057 mddev
->layout
!= mddev
->new_layout
)
3058 return sprintf(page
, "%d (%d)\n",
3059 mddev
->new_layout
, mddev
->layout
);
3060 return sprintf(page
, "%d\n", mddev
->layout
);
3064 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3067 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3069 if (!*buf
|| (*e
&& *e
!= '\n'))
3074 if (mddev
->pers
->check_reshape
== NULL
)
3076 mddev
->new_layout
= n
;
3077 err
= mddev
->pers
->check_reshape(mddev
);
3079 mddev
->new_layout
= mddev
->layout
;
3083 mddev
->new_layout
= n
;
3084 if (mddev
->reshape_position
== MaxSector
)
3089 static struct md_sysfs_entry md_layout
=
3090 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3094 raid_disks_show(mddev_t
*mddev
, char *page
)
3096 if (mddev
->raid_disks
== 0)
3098 if (mddev
->reshape_position
!= MaxSector
&&
3099 mddev
->delta_disks
!= 0)
3100 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3101 mddev
->raid_disks
- mddev
->delta_disks
);
3102 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3105 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3108 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3112 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3114 if (!*buf
|| (*e
&& *e
!= '\n'))
3118 rv
= update_raid_disks(mddev
, n
);
3119 else if (mddev
->reshape_position
!= MaxSector
) {
3120 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3121 mddev
->delta_disks
= n
- olddisks
;
3122 mddev
->raid_disks
= n
;
3124 mddev
->raid_disks
= n
;
3125 return rv
? rv
: len
;
3127 static struct md_sysfs_entry md_raid_disks
=
3128 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3131 chunk_size_show(mddev_t
*mddev
, char *page
)
3133 if (mddev
->reshape_position
!= MaxSector
&&
3134 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3135 return sprintf(page
, "%d (%d)\n",
3136 mddev
->new_chunk_sectors
<< 9,
3137 mddev
->chunk_sectors
<< 9);
3138 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3142 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3145 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3147 if (!*buf
|| (*e
&& *e
!= '\n'))
3152 if (mddev
->pers
->check_reshape
== NULL
)
3154 mddev
->new_chunk_sectors
= n
>> 9;
3155 err
= mddev
->pers
->check_reshape(mddev
);
3157 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3161 mddev
->new_chunk_sectors
= n
>> 9;
3162 if (mddev
->reshape_position
== MaxSector
)
3163 mddev
->chunk_sectors
= n
>> 9;
3167 static struct md_sysfs_entry md_chunk_size
=
3168 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3171 resync_start_show(mddev_t
*mddev
, char *page
)
3173 if (mddev
->recovery_cp
== MaxSector
)
3174 return sprintf(page
, "none\n");
3175 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3179 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3182 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3186 if (cmd_match(buf
, "none"))
3188 else if (!*buf
|| (*e
&& *e
!= '\n'))
3191 mddev
->recovery_cp
= n
;
3194 static struct md_sysfs_entry md_resync_start
=
3195 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3198 * The array state can be:
3201 * No devices, no size, no level
3202 * Equivalent to STOP_ARRAY ioctl
3204 * May have some settings, but array is not active
3205 * all IO results in error
3206 * When written, doesn't tear down array, but just stops it
3207 * suspended (not supported yet)
3208 * All IO requests will block. The array can be reconfigured.
3209 * Writing this, if accepted, will block until array is quiescent
3211 * no resync can happen. no superblocks get written.
3212 * write requests fail
3214 * like readonly, but behaves like 'clean' on a write request.
3216 * clean - no pending writes, but otherwise active.
3217 * When written to inactive array, starts without resync
3218 * If a write request arrives then
3219 * if metadata is known, mark 'dirty' and switch to 'active'.
3220 * if not known, block and switch to write-pending
3221 * If written to an active array that has pending writes, then fails.
3223 * fully active: IO and resync can be happening.
3224 * When written to inactive array, starts with resync
3227 * clean, but writes are blocked waiting for 'active' to be written.
3230 * like active, but no writes have been seen for a while (100msec).
3233 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3234 write_pending
, active_idle
, bad_word
};
3235 static char *array_states
[] = {
3236 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3237 "write-pending", "active-idle", NULL
};
3239 static int match_word(const char *word
, char **list
)
3242 for (n
=0; list
[n
]; n
++)
3243 if (cmd_match(word
, list
[n
]))
3249 array_state_show(mddev_t
*mddev
, char *page
)
3251 enum array_state st
= inactive
;
3264 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3266 else if (mddev
->safemode
)
3272 if (list_empty(&mddev
->disks
) &&
3273 mddev
->raid_disks
== 0 &&
3274 mddev
->dev_sectors
== 0)
3279 return sprintf(page
, "%s\n", array_states
[st
]);
3282 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3283 static int do_md_run(mddev_t
* mddev
);
3284 static int restart_array(mddev_t
*mddev
);
3287 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3290 enum array_state st
= match_word(buf
, array_states
);
3295 /* stopping an active array */
3296 if (atomic_read(&mddev
->openers
) > 0)
3298 err
= do_md_stop(mddev
, 0, 0);
3301 /* stopping an active array */
3303 if (atomic_read(&mddev
->openers
) > 0)
3305 err
= do_md_stop(mddev
, 2, 0);
3307 err
= 0; /* already inactive */
3310 break; /* not supported yet */
3313 err
= do_md_stop(mddev
, 1, 0);
3316 set_disk_ro(mddev
->gendisk
, 1);
3317 err
= do_md_run(mddev
);
3323 err
= do_md_stop(mddev
, 1, 0);
3324 else if (mddev
->ro
== 1)
3325 err
= restart_array(mddev
);
3328 set_disk_ro(mddev
->gendisk
, 0);
3332 err
= do_md_run(mddev
);
3337 restart_array(mddev
);
3338 spin_lock_irq(&mddev
->write_lock
);
3339 if (atomic_read(&mddev
->writes_pending
) == 0) {
3340 if (mddev
->in_sync
== 0) {
3342 if (mddev
->safemode
== 1)
3343 mddev
->safemode
= 0;
3344 if (mddev
->persistent
)
3345 set_bit(MD_CHANGE_CLEAN
,
3351 spin_unlock_irq(&mddev
->write_lock
);
3357 restart_array(mddev
);
3358 if (mddev
->external
)
3359 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3360 wake_up(&mddev
->sb_wait
);
3364 set_disk_ro(mddev
->gendisk
, 0);
3365 err
= do_md_run(mddev
);
3370 /* these cannot be set */
3376 sysfs_notify_dirent(mddev
->sysfs_state
);
3380 static struct md_sysfs_entry md_array_state
=
3381 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3384 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3385 return sprintf(page
, "%d\n",
3386 atomic_read(&mddev
->max_corr_read_errors
));
3390 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3393 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3395 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3396 atomic_set(&mddev
->max_corr_read_errors
, n
);
3402 static struct md_sysfs_entry max_corr_read_errors
=
3403 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3404 max_corrected_read_errors_store
);
3407 null_show(mddev_t
*mddev
, char *page
)
3413 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3415 /* buf must be %d:%d\n? giving major and minor numbers */
3416 /* The new device is added to the array.
3417 * If the array has a persistent superblock, we read the
3418 * superblock to initialise info and check validity.
3419 * Otherwise, only checking done is that in bind_rdev_to_array,
3420 * which mainly checks size.
3423 int major
= simple_strtoul(buf
, &e
, 10);
3429 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3431 minor
= simple_strtoul(e
+1, &e
, 10);
3432 if (*e
&& *e
!= '\n')
3434 dev
= MKDEV(major
, minor
);
3435 if (major
!= MAJOR(dev
) ||
3436 minor
!= MINOR(dev
))
3440 if (mddev
->persistent
) {
3441 rdev
= md_import_device(dev
, mddev
->major_version
,
3442 mddev
->minor_version
);
3443 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3444 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3445 mdk_rdev_t
, same_set
);
3446 err
= super_types
[mddev
->major_version
]
3447 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3451 } else if (mddev
->external
)
3452 rdev
= md_import_device(dev
, -2, -1);
3454 rdev
= md_import_device(dev
, -1, -1);
3457 return PTR_ERR(rdev
);
3458 err
= bind_rdev_to_array(rdev
, mddev
);
3462 return err
? err
: len
;
3465 static struct md_sysfs_entry md_new_device
=
3466 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3469 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3472 unsigned long chunk
, end_chunk
;
3476 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3478 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3479 if (buf
== end
) break;
3480 if (*end
== '-') { /* range */
3482 end_chunk
= simple_strtoul(buf
, &end
, 0);
3483 if (buf
== end
) break;
3485 if (*end
&& !isspace(*end
)) break;
3486 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3487 buf
= skip_spaces(end
);
3489 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3494 static struct md_sysfs_entry md_bitmap
=
3495 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3498 size_show(mddev_t
*mddev
, char *page
)
3500 return sprintf(page
, "%llu\n",
3501 (unsigned long long)mddev
->dev_sectors
/ 2);
3504 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3507 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3509 /* If array is inactive, we can reduce the component size, but
3510 * not increase it (except from 0).
3511 * If array is active, we can try an on-line resize
3514 int err
= strict_blocks_to_sectors(buf
, §ors
);
3519 err
= update_size(mddev
, sectors
);
3520 md_update_sb(mddev
, 1);
3522 if (mddev
->dev_sectors
== 0 ||
3523 mddev
->dev_sectors
> sectors
)
3524 mddev
->dev_sectors
= sectors
;
3528 return err
? err
: len
;
3531 static struct md_sysfs_entry md_size
=
3532 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3537 * 'none' for arrays with no metadata (good luck...)
3538 * 'external' for arrays with externally managed metadata,
3539 * or N.M for internally known formats
3542 metadata_show(mddev_t
*mddev
, char *page
)
3544 if (mddev
->persistent
)
3545 return sprintf(page
, "%d.%d\n",
3546 mddev
->major_version
, mddev
->minor_version
);
3547 else if (mddev
->external
)
3548 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3550 return sprintf(page
, "none\n");
3554 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3558 /* Changing the details of 'external' metadata is
3559 * always permitted. Otherwise there must be
3560 * no devices attached to the array.
3562 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3564 else if (!list_empty(&mddev
->disks
))
3567 if (cmd_match(buf
, "none")) {
3568 mddev
->persistent
= 0;
3569 mddev
->external
= 0;
3570 mddev
->major_version
= 0;
3571 mddev
->minor_version
= 90;
3574 if (strncmp(buf
, "external:", 9) == 0) {
3575 size_t namelen
= len
-9;
3576 if (namelen
>= sizeof(mddev
->metadata_type
))
3577 namelen
= sizeof(mddev
->metadata_type
)-1;
3578 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3579 mddev
->metadata_type
[namelen
] = 0;
3580 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3581 mddev
->metadata_type
[--namelen
] = 0;
3582 mddev
->persistent
= 0;
3583 mddev
->external
= 1;
3584 mddev
->major_version
= 0;
3585 mddev
->minor_version
= 90;
3588 major
= simple_strtoul(buf
, &e
, 10);
3589 if (e
==buf
|| *e
!= '.')
3592 minor
= simple_strtoul(buf
, &e
, 10);
3593 if (e
==buf
|| (*e
&& *e
!= '\n') )
3595 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3597 mddev
->major_version
= major
;
3598 mddev
->minor_version
= minor
;
3599 mddev
->persistent
= 1;
3600 mddev
->external
= 0;
3604 static struct md_sysfs_entry md_metadata
=
3605 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3608 action_show(mddev_t
*mddev
, char *page
)
3610 char *type
= "idle";
3611 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3613 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3614 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3615 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3617 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3618 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3620 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3624 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3627 return sprintf(page
, "%s\n", type
);
3631 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3633 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3636 if (cmd_match(page
, "frozen"))
3637 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3639 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3641 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3642 if (mddev
->sync_thread
) {
3643 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3644 md_unregister_thread(mddev
->sync_thread
);
3645 mddev
->sync_thread
= NULL
;
3646 mddev
->recovery
= 0;
3648 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3649 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3651 else if (cmd_match(page
, "resync"))
3652 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3653 else if (cmd_match(page
, "recover")) {
3654 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3655 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3656 } else if (cmd_match(page
, "reshape")) {
3658 if (mddev
->pers
->start_reshape
== NULL
)
3660 err
= mddev
->pers
->start_reshape(mddev
);
3663 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3665 if (cmd_match(page
, "check"))
3666 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3667 else if (!cmd_match(page
, "repair"))
3669 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3670 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3672 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3673 md_wakeup_thread(mddev
->thread
);
3674 sysfs_notify_dirent(mddev
->sysfs_action
);
3679 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3681 return sprintf(page
, "%llu\n",
3682 (unsigned long long) mddev
->resync_mismatches
);
3685 static struct md_sysfs_entry md_scan_mode
=
3686 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3689 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3692 sync_min_show(mddev_t
*mddev
, char *page
)
3694 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3695 mddev
->sync_speed_min
? "local": "system");
3699 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3703 if (strncmp(buf
, "system", 6)==0) {
3704 mddev
->sync_speed_min
= 0;
3707 min
= simple_strtoul(buf
, &e
, 10);
3708 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3710 mddev
->sync_speed_min
= min
;
3714 static struct md_sysfs_entry md_sync_min
=
3715 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3718 sync_max_show(mddev_t
*mddev
, char *page
)
3720 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3721 mddev
->sync_speed_max
? "local": "system");
3725 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3729 if (strncmp(buf
, "system", 6)==0) {
3730 mddev
->sync_speed_max
= 0;
3733 max
= simple_strtoul(buf
, &e
, 10);
3734 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3736 mddev
->sync_speed_max
= max
;
3740 static struct md_sysfs_entry md_sync_max
=
3741 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3744 degraded_show(mddev_t
*mddev
, char *page
)
3746 return sprintf(page
, "%d\n", mddev
->degraded
);
3748 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3751 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3753 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3757 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3761 if (strict_strtol(buf
, 10, &n
))
3764 if (n
!= 0 && n
!= 1)
3767 mddev
->parallel_resync
= n
;
3769 if (mddev
->sync_thread
)
3770 wake_up(&resync_wait
);
3775 /* force parallel resync, even with shared block devices */
3776 static struct md_sysfs_entry md_sync_force_parallel
=
3777 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3778 sync_force_parallel_show
, sync_force_parallel_store
);
3781 sync_speed_show(mddev_t
*mddev
, char *page
)
3783 unsigned long resync
, dt
, db
;
3784 if (mddev
->curr_resync
== 0)
3785 return sprintf(page
, "none\n");
3786 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3787 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3789 db
= resync
- mddev
->resync_mark_cnt
;
3790 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3793 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3796 sync_completed_show(mddev_t
*mddev
, char *page
)
3798 unsigned long max_sectors
, resync
;
3800 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3801 return sprintf(page
, "none\n");
3803 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3804 max_sectors
= mddev
->resync_max_sectors
;
3806 max_sectors
= mddev
->dev_sectors
;
3808 resync
= mddev
->curr_resync_completed
;
3809 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3812 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3815 min_sync_show(mddev_t
*mddev
, char *page
)
3817 return sprintf(page
, "%llu\n",
3818 (unsigned long long)mddev
->resync_min
);
3821 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3823 unsigned long long min
;
3824 if (strict_strtoull(buf
, 10, &min
))
3826 if (min
> mddev
->resync_max
)
3828 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3831 /* Must be a multiple of chunk_size */
3832 if (mddev
->chunk_sectors
) {
3833 sector_t temp
= min
;
3834 if (sector_div(temp
, mddev
->chunk_sectors
))
3837 mddev
->resync_min
= min
;
3842 static struct md_sysfs_entry md_min_sync
=
3843 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3846 max_sync_show(mddev_t
*mddev
, char *page
)
3848 if (mddev
->resync_max
== MaxSector
)
3849 return sprintf(page
, "max\n");
3851 return sprintf(page
, "%llu\n",
3852 (unsigned long long)mddev
->resync_max
);
3855 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3857 if (strncmp(buf
, "max", 3) == 0)
3858 mddev
->resync_max
= MaxSector
;
3860 unsigned long long max
;
3861 if (strict_strtoull(buf
, 10, &max
))
3863 if (max
< mddev
->resync_min
)
3865 if (max
< mddev
->resync_max
&&
3867 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3870 /* Must be a multiple of chunk_size */
3871 if (mddev
->chunk_sectors
) {
3872 sector_t temp
= max
;
3873 if (sector_div(temp
, mddev
->chunk_sectors
))
3876 mddev
->resync_max
= max
;
3878 wake_up(&mddev
->recovery_wait
);
3882 static struct md_sysfs_entry md_max_sync
=
3883 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3886 suspend_lo_show(mddev_t
*mddev
, char *page
)
3888 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3892 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3895 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3897 if (mddev
->pers
== NULL
||
3898 mddev
->pers
->quiesce
== NULL
)
3900 if (buf
== e
|| (*e
&& *e
!= '\n'))
3902 if (new >= mddev
->suspend_hi
||
3903 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3904 mddev
->suspend_lo
= new;
3905 mddev
->pers
->quiesce(mddev
, 2);
3910 static struct md_sysfs_entry md_suspend_lo
=
3911 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3915 suspend_hi_show(mddev_t
*mddev
, char *page
)
3917 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3921 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3924 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3926 if (mddev
->pers
== NULL
||
3927 mddev
->pers
->quiesce
== NULL
)
3929 if (buf
== e
|| (*e
&& *e
!= '\n'))
3931 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3932 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3933 mddev
->suspend_hi
= new;
3934 mddev
->pers
->quiesce(mddev
, 1);
3935 mddev
->pers
->quiesce(mddev
, 0);
3940 static struct md_sysfs_entry md_suspend_hi
=
3941 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3944 reshape_position_show(mddev_t
*mddev
, char *page
)
3946 if (mddev
->reshape_position
!= MaxSector
)
3947 return sprintf(page
, "%llu\n",
3948 (unsigned long long)mddev
->reshape_position
);
3949 strcpy(page
, "none\n");
3954 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3957 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3960 if (buf
== e
|| (*e
&& *e
!= '\n'))
3962 mddev
->reshape_position
= new;
3963 mddev
->delta_disks
= 0;
3964 mddev
->new_level
= mddev
->level
;
3965 mddev
->new_layout
= mddev
->layout
;
3966 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3970 static struct md_sysfs_entry md_reshape_position
=
3971 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3972 reshape_position_store
);
3975 array_size_show(mddev_t
*mddev
, char *page
)
3977 if (mddev
->external_size
)
3978 return sprintf(page
, "%llu\n",
3979 (unsigned long long)mddev
->array_sectors
/2);
3981 return sprintf(page
, "default\n");
3985 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3989 if (strncmp(buf
, "default", 7) == 0) {
3991 sectors
= mddev
->pers
->size(mddev
, 0, 0);
3993 sectors
= mddev
->array_sectors
;
3995 mddev
->external_size
= 0;
3997 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3999 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4002 mddev
->external_size
= 1;
4005 mddev
->array_sectors
= sectors
;
4006 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4008 revalidate_disk(mddev
->gendisk
);
4013 static struct md_sysfs_entry md_array_size
=
4014 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4017 static struct attribute
*md_default_attrs
[] = {
4020 &md_raid_disks
.attr
,
4021 &md_chunk_size
.attr
,
4023 &md_resync_start
.attr
,
4025 &md_new_device
.attr
,
4026 &md_safe_delay
.attr
,
4027 &md_array_state
.attr
,
4028 &md_reshape_position
.attr
,
4029 &md_array_size
.attr
,
4030 &max_corr_read_errors
.attr
,
4034 static struct attribute
*md_redundancy_attrs
[] = {
4036 &md_mismatches
.attr
,
4039 &md_sync_speed
.attr
,
4040 &md_sync_force_parallel
.attr
,
4041 &md_sync_completed
.attr
,
4044 &md_suspend_lo
.attr
,
4045 &md_suspend_hi
.attr
,
4050 static struct attribute_group md_redundancy_group
= {
4052 .attrs
= md_redundancy_attrs
,
4057 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4059 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4060 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4065 rv
= mddev_lock(mddev
);
4067 rv
= entry
->show(mddev
, page
);
4068 mddev_unlock(mddev
);
4074 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4075 const char *page
, size_t length
)
4077 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4078 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4083 if (!capable(CAP_SYS_ADMIN
))
4085 rv
= mddev_lock(mddev
);
4086 if (mddev
->hold_active
== UNTIL_IOCTL
)
4087 mddev
->hold_active
= 0;
4089 rv
= entry
->store(mddev
, page
, length
);
4090 mddev_unlock(mddev
);
4095 static void md_free(struct kobject
*ko
)
4097 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4099 if (mddev
->sysfs_state
)
4100 sysfs_put(mddev
->sysfs_state
);
4102 if (mddev
->gendisk
) {
4103 del_gendisk(mddev
->gendisk
);
4104 put_disk(mddev
->gendisk
);
4107 blk_cleanup_queue(mddev
->queue
);
4112 static struct sysfs_ops md_sysfs_ops
= {
4113 .show
= md_attr_show
,
4114 .store
= md_attr_store
,
4116 static struct kobj_type md_ktype
= {
4118 .sysfs_ops
= &md_sysfs_ops
,
4119 .default_attrs
= md_default_attrs
,
4124 static void mddev_delayed_delete(struct work_struct
*ws
)
4126 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4128 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4129 kobject_del(&mddev
->kobj
);
4130 kobject_put(&mddev
->kobj
);
4133 static int md_alloc(dev_t dev
, char *name
)
4135 static DEFINE_MUTEX(disks_mutex
);
4136 mddev_t
*mddev
= mddev_find(dev
);
4137 struct gendisk
*disk
;
4146 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4147 shift
= partitioned
? MdpMinorShift
: 0;
4148 unit
= MINOR(mddev
->unit
) >> shift
;
4150 /* wait for any previous instance if this device
4151 * to be completed removed (mddev_delayed_delete).
4153 flush_scheduled_work();
4155 mutex_lock(&disks_mutex
);
4161 /* Need to ensure that 'name' is not a duplicate.
4164 spin_lock(&all_mddevs_lock
);
4166 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4167 if (mddev2
->gendisk
&&
4168 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4169 spin_unlock(&all_mddevs_lock
);
4172 spin_unlock(&all_mddevs_lock
);
4176 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4179 mddev
->queue
->queuedata
= mddev
;
4181 /* Can be unlocked because the queue is new: no concurrency */
4182 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4184 blk_queue_make_request(mddev
->queue
, md_make_request
);
4186 disk
= alloc_disk(1 << shift
);
4188 blk_cleanup_queue(mddev
->queue
);
4189 mddev
->queue
= NULL
;
4192 disk
->major
= MAJOR(mddev
->unit
);
4193 disk
->first_minor
= unit
<< shift
;
4195 strcpy(disk
->disk_name
, name
);
4196 else if (partitioned
)
4197 sprintf(disk
->disk_name
, "md_d%d", unit
);
4199 sprintf(disk
->disk_name
, "md%d", unit
);
4200 disk
->fops
= &md_fops
;
4201 disk
->private_data
= mddev
;
4202 disk
->queue
= mddev
->queue
;
4203 /* Allow extended partitions. This makes the
4204 * 'mdp' device redundant, but we can't really
4207 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4209 mddev
->gendisk
= disk
;
4210 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4211 &disk_to_dev(disk
)->kobj
, "%s", "md");
4213 /* This isn't possible, but as kobject_init_and_add is marked
4214 * __must_check, we must do something with the result
4216 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4220 if (sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4221 printk(KERN_DEBUG
"pointless warning\n");
4223 mutex_unlock(&disks_mutex
);
4225 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4226 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4232 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4234 md_alloc(dev
, NULL
);
4238 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4240 /* val must be "md_*" where * is not all digits.
4241 * We allocate an array with a large free minor number, and
4242 * set the name to val. val must not already be an active name.
4244 int len
= strlen(val
);
4245 char buf
[DISK_NAME_LEN
];
4247 while (len
&& val
[len
-1] == '\n')
4249 if (len
>= DISK_NAME_LEN
)
4251 strlcpy(buf
, val
, len
+1);
4252 if (strncmp(buf
, "md_", 3) != 0)
4254 return md_alloc(0, buf
);
4257 static void md_safemode_timeout(unsigned long data
)
4259 mddev_t
*mddev
= (mddev_t
*) data
;
4261 if (!atomic_read(&mddev
->writes_pending
)) {
4262 mddev
->safemode
= 1;
4263 if (mddev
->external
)
4264 sysfs_notify_dirent(mddev
->sysfs_state
);
4266 md_wakeup_thread(mddev
->thread
);
4269 static int start_dirty_degraded
;
4271 static int do_md_run(mddev_t
* mddev
)
4275 struct gendisk
*disk
;
4276 struct mdk_personality
*pers
;
4278 if (list_empty(&mddev
->disks
))
4279 /* cannot run an array with no devices.. */
4285 /* These two calls synchronise us with the
4286 * sysfs_remove_group calls in mddev_unlock,
4287 * so they must have completed.
4289 mutex_lock(&mddev
->open_mutex
);
4290 mutex_unlock(&mddev
->open_mutex
);
4293 * Analyze all RAID superblock(s)
4295 if (!mddev
->raid_disks
) {
4296 if (!mddev
->persistent
)
4301 if (mddev
->level
!= LEVEL_NONE
)
4302 request_module("md-level-%d", mddev
->level
);
4303 else if (mddev
->clevel
[0])
4304 request_module("md-%s", mddev
->clevel
);
4307 * Drop all container device buffers, from now on
4308 * the only valid external interface is through the md
4311 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4312 if (test_bit(Faulty
, &rdev
->flags
))
4314 sync_blockdev(rdev
->bdev
);
4315 invalidate_bdev(rdev
->bdev
);
4317 /* perform some consistency tests on the device.
4318 * We don't want the data to overlap the metadata,
4319 * Internal Bitmap issues have been handled elsewhere.
4321 if (rdev
->data_offset
< rdev
->sb_start
) {
4322 if (mddev
->dev_sectors
&&
4323 rdev
->data_offset
+ mddev
->dev_sectors
4325 printk("md: %s: data overlaps metadata\n",
4330 if (rdev
->sb_start
+ rdev
->sb_size
/512
4331 > rdev
->data_offset
) {
4332 printk("md: %s: metadata overlaps data\n",
4337 sysfs_notify_dirent(rdev
->sysfs_state
);
4340 disk
= mddev
->gendisk
;
4342 spin_lock(&pers_lock
);
4343 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4344 if (!pers
|| !try_module_get(pers
->owner
)) {
4345 spin_unlock(&pers_lock
);
4346 if (mddev
->level
!= LEVEL_NONE
)
4347 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4350 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4355 spin_unlock(&pers_lock
);
4356 if (mddev
->level
!= pers
->level
) {
4357 mddev
->level
= pers
->level
;
4358 mddev
->new_level
= pers
->level
;
4360 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4362 if (mddev
->reshape_position
!= MaxSector
&&
4363 pers
->start_reshape
== NULL
) {
4364 /* This personality cannot handle reshaping... */
4366 module_put(pers
->owner
);
4370 if (pers
->sync_request
) {
4371 /* Warn if this is a potentially silly
4374 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4378 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4379 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4381 rdev
->bdev
->bd_contains
==
4382 rdev2
->bdev
->bd_contains
) {
4384 "%s: WARNING: %s appears to be"
4385 " on the same physical disk as"
4388 bdevname(rdev
->bdev
,b
),
4389 bdevname(rdev2
->bdev
,b2
));
4396 "True protection against single-disk"
4397 " failure might be compromised.\n");
4400 mddev
->recovery
= 0;
4401 /* may be over-ridden by personality */
4402 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4404 mddev
->barriers_work
= 1;
4405 mddev
->ok_start_degraded
= start_dirty_degraded
;
4407 if (start_readonly
&& mddev
->ro
== 0)
4408 mddev
->ro
= 2; /* read-only, but switch on first write */
4410 err
= mddev
->pers
->run(mddev
);
4412 printk(KERN_ERR
"md: pers->run() failed ...\n");
4413 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4414 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4415 " but 'external_size' not in effect?\n", __func__
);
4417 "md: invalid array_size %llu > default size %llu\n",
4418 (unsigned long long)mddev
->array_sectors
/ 2,
4419 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4421 mddev
->pers
->stop(mddev
);
4423 if (err
== 0 && mddev
->pers
->sync_request
) {
4424 err
= bitmap_create(mddev
);
4426 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4427 mdname(mddev
), err
);
4428 mddev
->pers
->stop(mddev
);
4432 module_put(mddev
->pers
->owner
);
4434 bitmap_destroy(mddev
);
4437 if (mddev
->pers
->sync_request
) {
4438 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4440 "md: cannot register extra attributes for %s\n",
4442 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4443 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4446 atomic_set(&mddev
->writes_pending
,0);
4447 atomic_set(&mddev
->max_corr_read_errors
,
4448 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4449 mddev
->safemode
= 0;
4450 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4451 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4452 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4455 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4456 if (rdev
->raid_disk
>= 0) {
4458 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4459 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4460 printk("md: cannot register %s for %s\n",
4464 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4467 md_update_sb(mddev
, 0);
4469 set_capacity(disk
, mddev
->array_sectors
);
4471 md_wakeup_thread(mddev
->thread
);
4472 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4474 revalidate_disk(mddev
->gendisk
);
4476 md_new_event(mddev
);
4477 sysfs_notify_dirent(mddev
->sysfs_state
);
4478 if (mddev
->sysfs_action
)
4479 sysfs_notify_dirent(mddev
->sysfs_action
);
4480 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4481 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4485 static int restart_array(mddev_t
*mddev
)
4487 struct gendisk
*disk
= mddev
->gendisk
;
4489 /* Complain if it has no devices */
4490 if (list_empty(&mddev
->disks
))
4496 mddev
->safemode
= 0;
4498 set_disk_ro(disk
, 0);
4499 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4501 /* Kick recovery or resync if necessary */
4502 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4503 md_wakeup_thread(mddev
->thread
);
4504 md_wakeup_thread(mddev
->sync_thread
);
4505 sysfs_notify_dirent(mddev
->sysfs_state
);
4509 /* similar to deny_write_access, but accounts for our holding a reference
4510 * to the file ourselves */
4511 static int deny_bitmap_write_access(struct file
* file
)
4513 struct inode
*inode
= file
->f_mapping
->host
;
4515 spin_lock(&inode
->i_lock
);
4516 if (atomic_read(&inode
->i_writecount
) > 1) {
4517 spin_unlock(&inode
->i_lock
);
4520 atomic_set(&inode
->i_writecount
, -1);
4521 spin_unlock(&inode
->i_lock
);
4526 void restore_bitmap_write_access(struct file
*file
)
4528 struct inode
*inode
= file
->f_mapping
->host
;
4530 spin_lock(&inode
->i_lock
);
4531 atomic_set(&inode
->i_writecount
, 1);
4532 spin_unlock(&inode
->i_lock
);
4536 * 0 - completely stop and dis-assemble array
4537 * 1 - switch to readonly
4538 * 2 - stop but do not disassemble array
4540 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4543 struct gendisk
*disk
= mddev
->gendisk
;
4546 mutex_lock(&mddev
->open_mutex
);
4547 if (atomic_read(&mddev
->openers
) > is_open
) {
4548 printk("md: %s still in use.\n",mdname(mddev
));
4550 } else if (mddev
->pers
) {
4552 if (mddev
->sync_thread
) {
4553 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4554 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4555 md_unregister_thread(mddev
->sync_thread
);
4556 mddev
->sync_thread
= NULL
;
4559 del_timer_sync(&mddev
->safemode_timer
);
4562 case 1: /* readonly */
4568 case 0: /* disassemble */
4570 bitmap_flush(mddev
);
4571 md_super_wait(mddev
);
4573 set_disk_ro(disk
, 0);
4575 mddev
->pers
->stop(mddev
);
4576 mddev
->queue
->merge_bvec_fn
= NULL
;
4577 mddev
->queue
->unplug_fn
= NULL
;
4578 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4579 module_put(mddev
->pers
->owner
);
4580 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
4581 mddev
->to_remove
= &md_redundancy_group
;
4583 /* tell userspace to handle 'inactive' */
4584 sysfs_notify_dirent(mddev
->sysfs_state
);
4586 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4587 if (rdev
->raid_disk
>= 0) {
4589 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4590 sysfs_remove_link(&mddev
->kobj
, nm
);
4593 set_capacity(disk
, 0);
4599 if (!mddev
->in_sync
|| mddev
->flags
) {
4600 /* mark array as shutdown cleanly */
4602 md_update_sb(mddev
, 1);
4605 set_disk_ro(disk
, 1);
4606 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4610 mutex_unlock(&mddev
->open_mutex
);
4614 * Free resources if final stop
4618 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4620 bitmap_destroy(mddev
);
4621 if (mddev
->bitmap_info
.file
) {
4622 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4623 fput(mddev
->bitmap_info
.file
);
4624 mddev
->bitmap_info
.file
= NULL
;
4626 mddev
->bitmap_info
.offset
= 0;
4628 export_array(mddev
);
4630 mddev
->array_sectors
= 0;
4631 mddev
->external_size
= 0;
4632 mddev
->dev_sectors
= 0;
4633 mddev
->raid_disks
= 0;
4634 mddev
->recovery_cp
= 0;
4635 mddev
->resync_min
= 0;
4636 mddev
->resync_max
= MaxSector
;
4637 mddev
->reshape_position
= MaxSector
;
4638 mddev
->external
= 0;
4639 mddev
->persistent
= 0;
4640 mddev
->level
= LEVEL_NONE
;
4641 mddev
->clevel
[0] = 0;
4644 mddev
->metadata_type
[0] = 0;
4645 mddev
->chunk_sectors
= 0;
4646 mddev
->ctime
= mddev
->utime
= 0;
4648 mddev
->max_disks
= 0;
4650 mddev
->delta_disks
= 0;
4651 mddev
->new_level
= LEVEL_NONE
;
4652 mddev
->new_layout
= 0;
4653 mddev
->new_chunk_sectors
= 0;
4654 mddev
->curr_resync
= 0;
4655 mddev
->resync_mismatches
= 0;
4656 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4657 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4658 mddev
->recovery
= 0;
4661 mddev
->degraded
= 0;
4662 mddev
->barriers_work
= 0;
4663 mddev
->safemode
= 0;
4664 mddev
->bitmap_info
.offset
= 0;
4665 mddev
->bitmap_info
.default_offset
= 0;
4666 mddev
->bitmap_info
.chunksize
= 0;
4667 mddev
->bitmap_info
.daemon_sleep
= 0;
4668 mddev
->bitmap_info
.max_write_behind
= 0;
4669 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4670 if (mddev
->hold_active
== UNTIL_STOP
)
4671 mddev
->hold_active
= 0;
4673 } else if (mddev
->pers
)
4674 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4677 blk_integrity_unregister(disk
);
4678 md_new_event(mddev
);
4679 sysfs_notify_dirent(mddev
->sysfs_state
);
4684 static void autorun_array(mddev_t
*mddev
)
4689 if (list_empty(&mddev
->disks
))
4692 printk(KERN_INFO
"md: running: ");
4694 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4695 char b
[BDEVNAME_SIZE
];
4696 printk("<%s>", bdevname(rdev
->bdev
,b
));
4700 err
= do_md_run(mddev
);
4702 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4703 do_md_stop(mddev
, 0, 0);
4708 * lets try to run arrays based on all disks that have arrived
4709 * until now. (those are in pending_raid_disks)
4711 * the method: pick the first pending disk, collect all disks with
4712 * the same UUID, remove all from the pending list and put them into
4713 * the 'same_array' list. Then order this list based on superblock
4714 * update time (freshest comes first), kick out 'old' disks and
4715 * compare superblocks. If everything's fine then run it.
4717 * If "unit" is allocated, then bump its reference count
4719 static void autorun_devices(int part
)
4721 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4723 char b
[BDEVNAME_SIZE
];
4725 printk(KERN_INFO
"md: autorun ...\n");
4726 while (!list_empty(&pending_raid_disks
)) {
4729 LIST_HEAD(candidates
);
4730 rdev0
= list_entry(pending_raid_disks
.next
,
4731 mdk_rdev_t
, same_set
);
4733 printk(KERN_INFO
"md: considering %s ...\n",
4734 bdevname(rdev0
->bdev
,b
));
4735 INIT_LIST_HEAD(&candidates
);
4736 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4737 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4738 printk(KERN_INFO
"md: adding %s ...\n",
4739 bdevname(rdev
->bdev
,b
));
4740 list_move(&rdev
->same_set
, &candidates
);
4743 * now we have a set of devices, with all of them having
4744 * mostly sane superblocks. It's time to allocate the
4748 dev
= MKDEV(mdp_major
,
4749 rdev0
->preferred_minor
<< MdpMinorShift
);
4750 unit
= MINOR(dev
) >> MdpMinorShift
;
4752 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4755 if (rdev0
->preferred_minor
!= unit
) {
4756 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4757 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4761 md_probe(dev
, NULL
, NULL
);
4762 mddev
= mddev_find(dev
);
4763 if (!mddev
|| !mddev
->gendisk
) {
4767 "md: cannot allocate memory for md drive.\n");
4770 if (mddev_lock(mddev
))
4771 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4773 else if (mddev
->raid_disks
|| mddev
->major_version
4774 || !list_empty(&mddev
->disks
)) {
4776 "md: %s already running, cannot run %s\n",
4777 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4778 mddev_unlock(mddev
);
4780 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4781 mddev
->persistent
= 1;
4782 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4783 list_del_init(&rdev
->same_set
);
4784 if (bind_rdev_to_array(rdev
, mddev
))
4787 autorun_array(mddev
);
4788 mddev_unlock(mddev
);
4790 /* on success, candidates will be empty, on error
4793 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4794 list_del_init(&rdev
->same_set
);
4799 printk(KERN_INFO
"md: ... autorun DONE.\n");
4801 #endif /* !MODULE */
4803 static int get_version(void __user
* arg
)
4807 ver
.major
= MD_MAJOR_VERSION
;
4808 ver
.minor
= MD_MINOR_VERSION
;
4809 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4811 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4817 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4819 mdu_array_info_t info
;
4820 int nr
,working
,insync
,failed
,spare
;
4823 nr
=working
=insync
=failed
=spare
=0;
4824 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4826 if (test_bit(Faulty
, &rdev
->flags
))
4830 if (test_bit(In_sync
, &rdev
->flags
))
4837 info
.major_version
= mddev
->major_version
;
4838 info
.minor_version
= mddev
->minor_version
;
4839 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4840 info
.ctime
= mddev
->ctime
;
4841 info
.level
= mddev
->level
;
4842 info
.size
= mddev
->dev_sectors
/ 2;
4843 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4846 info
.raid_disks
= mddev
->raid_disks
;
4847 info
.md_minor
= mddev
->md_minor
;
4848 info
.not_persistent
= !mddev
->persistent
;
4850 info
.utime
= mddev
->utime
;
4853 info
.state
= (1<<MD_SB_CLEAN
);
4854 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
4855 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4856 info
.active_disks
= insync
;
4857 info
.working_disks
= working
;
4858 info
.failed_disks
= failed
;
4859 info
.spare_disks
= spare
;
4861 info
.layout
= mddev
->layout
;
4862 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4864 if (copy_to_user(arg
, &info
, sizeof(info
)))
4870 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4872 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4873 char *ptr
, *buf
= NULL
;
4876 if (md_allow_write(mddev
))
4877 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4879 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4884 /* bitmap disabled, zero the first byte and copy out */
4885 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4886 file
->pathname
[0] = '\0';
4890 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4894 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4898 strcpy(file
->pathname
, ptr
);
4902 if (copy_to_user(arg
, file
, sizeof(*file
)))
4910 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4912 mdu_disk_info_t info
;
4915 if (copy_from_user(&info
, arg
, sizeof(info
)))
4918 rdev
= find_rdev_nr(mddev
, info
.number
);
4920 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4921 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4922 info
.raid_disk
= rdev
->raid_disk
;
4924 if (test_bit(Faulty
, &rdev
->flags
))
4925 info
.state
|= (1<<MD_DISK_FAULTY
);
4926 else if (test_bit(In_sync
, &rdev
->flags
)) {
4927 info
.state
|= (1<<MD_DISK_ACTIVE
);
4928 info
.state
|= (1<<MD_DISK_SYNC
);
4930 if (test_bit(WriteMostly
, &rdev
->flags
))
4931 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4933 info
.major
= info
.minor
= 0;
4934 info
.raid_disk
= -1;
4935 info
.state
= (1<<MD_DISK_REMOVED
);
4938 if (copy_to_user(arg
, &info
, sizeof(info
)))
4944 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4946 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4948 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4950 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4953 if (!mddev
->raid_disks
) {
4955 /* expecting a device which has a superblock */
4956 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4959 "md: md_import_device returned %ld\n",
4961 return PTR_ERR(rdev
);
4963 if (!list_empty(&mddev
->disks
)) {
4964 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4965 mdk_rdev_t
, same_set
);
4966 err
= super_types
[mddev
->major_version
]
4967 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4970 "md: %s has different UUID to %s\n",
4971 bdevname(rdev
->bdev
,b
),
4972 bdevname(rdev0
->bdev
,b2
));
4977 err
= bind_rdev_to_array(rdev
, mddev
);
4984 * add_new_disk can be used once the array is assembled
4985 * to add "hot spares". They must already have a superblock
4990 if (!mddev
->pers
->hot_add_disk
) {
4992 "%s: personality does not support diskops!\n",
4996 if (mddev
->persistent
)
4997 rdev
= md_import_device(dev
, mddev
->major_version
,
4998 mddev
->minor_version
);
5000 rdev
= md_import_device(dev
, -1, -1);
5003 "md: md_import_device returned %ld\n",
5005 return PTR_ERR(rdev
);
5007 /* set saved_raid_disk if appropriate */
5008 if (!mddev
->persistent
) {
5009 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5010 info
->raid_disk
< mddev
->raid_disks
) {
5011 rdev
->raid_disk
= info
->raid_disk
;
5012 set_bit(In_sync
, &rdev
->flags
);
5014 rdev
->raid_disk
= -1;
5016 super_types
[mddev
->major_version
].
5017 validate_super(mddev
, rdev
);
5018 if (test_bit(In_sync
, &rdev
->flags
))
5019 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5021 rdev
->saved_raid_disk
= -1;
5023 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5024 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5025 set_bit(WriteMostly
, &rdev
->flags
);
5027 clear_bit(WriteMostly
, &rdev
->flags
);
5029 rdev
->raid_disk
= -1;
5030 err
= bind_rdev_to_array(rdev
, mddev
);
5031 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5032 /* If there is hot_add_disk but no hot_remove_disk
5033 * then added disks for geometry changes,
5034 * and should be added immediately.
5036 super_types
[mddev
->major_version
].
5037 validate_super(mddev
, rdev
);
5038 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5040 unbind_rdev_from_array(rdev
);
5045 sysfs_notify_dirent(rdev
->sysfs_state
);
5047 md_update_sb(mddev
, 1);
5048 if (mddev
->degraded
)
5049 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5050 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5051 md_wakeup_thread(mddev
->thread
);
5055 /* otherwise, add_new_disk is only allowed
5056 * for major_version==0 superblocks
5058 if (mddev
->major_version
!= 0) {
5059 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5064 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5066 rdev
= md_import_device(dev
, -1, 0);
5069 "md: error, md_import_device() returned %ld\n",
5071 return PTR_ERR(rdev
);
5073 rdev
->desc_nr
= info
->number
;
5074 if (info
->raid_disk
< mddev
->raid_disks
)
5075 rdev
->raid_disk
= info
->raid_disk
;
5077 rdev
->raid_disk
= -1;
5079 if (rdev
->raid_disk
< mddev
->raid_disks
)
5080 if (info
->state
& (1<<MD_DISK_SYNC
))
5081 set_bit(In_sync
, &rdev
->flags
);
5083 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5084 set_bit(WriteMostly
, &rdev
->flags
);
5086 if (!mddev
->persistent
) {
5087 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5088 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5090 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5091 rdev
->sectors
= rdev
->sb_start
;
5093 err
= bind_rdev_to_array(rdev
, mddev
);
5103 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5105 char b
[BDEVNAME_SIZE
];
5108 rdev
= find_rdev(mddev
, dev
);
5112 if (rdev
->raid_disk
>= 0)
5115 kick_rdev_from_array(rdev
);
5116 md_update_sb(mddev
, 1);
5117 md_new_event(mddev
);
5121 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5122 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5126 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5128 char b
[BDEVNAME_SIZE
];
5135 if (mddev
->major_version
!= 0) {
5136 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5137 " version-0 superblocks.\n",
5141 if (!mddev
->pers
->hot_add_disk
) {
5143 "%s: personality does not support diskops!\n",
5148 rdev
= md_import_device(dev
, -1, 0);
5151 "md: error, md_import_device() returned %ld\n",
5156 if (mddev
->persistent
)
5157 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5159 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5161 rdev
->sectors
= rdev
->sb_start
;
5163 if (test_bit(Faulty
, &rdev
->flags
)) {
5165 "md: can not hot-add faulty %s disk to %s!\n",
5166 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5170 clear_bit(In_sync
, &rdev
->flags
);
5172 rdev
->saved_raid_disk
= -1;
5173 err
= bind_rdev_to_array(rdev
, mddev
);
5178 * The rest should better be atomic, we can have disk failures
5179 * noticed in interrupt contexts ...
5182 rdev
->raid_disk
= -1;
5184 md_update_sb(mddev
, 1);
5187 * Kick recovery, maybe this spare has to be added to the
5188 * array immediately.
5190 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5191 md_wakeup_thread(mddev
->thread
);
5192 md_new_event(mddev
);
5200 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5205 if (!mddev
->pers
->quiesce
)
5207 if (mddev
->recovery
|| mddev
->sync_thread
)
5209 /* we should be able to change the bitmap.. */
5215 return -EEXIST
; /* cannot add when bitmap is present */
5216 mddev
->bitmap_info
.file
= fget(fd
);
5218 if (mddev
->bitmap_info
.file
== NULL
) {
5219 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5224 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5226 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5228 fput(mddev
->bitmap_info
.file
);
5229 mddev
->bitmap_info
.file
= NULL
;
5232 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5233 } else if (mddev
->bitmap
== NULL
)
5234 return -ENOENT
; /* cannot remove what isn't there */
5237 mddev
->pers
->quiesce(mddev
, 1);
5239 err
= bitmap_create(mddev
);
5240 if (fd
< 0 || err
) {
5241 bitmap_destroy(mddev
);
5242 fd
= -1; /* make sure to put the file */
5244 mddev
->pers
->quiesce(mddev
, 0);
5247 if (mddev
->bitmap_info
.file
) {
5248 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5249 fput(mddev
->bitmap_info
.file
);
5251 mddev
->bitmap_info
.file
= NULL
;
5258 * set_array_info is used two different ways
5259 * The original usage is when creating a new array.
5260 * In this usage, raid_disks is > 0 and it together with
5261 * level, size, not_persistent,layout,chunksize determine the
5262 * shape of the array.
5263 * This will always create an array with a type-0.90.0 superblock.
5264 * The newer usage is when assembling an array.
5265 * In this case raid_disks will be 0, and the major_version field is
5266 * use to determine which style super-blocks are to be found on the devices.
5267 * The minor and patch _version numbers are also kept incase the
5268 * super_block handler wishes to interpret them.
5270 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5273 if (info
->raid_disks
== 0) {
5274 /* just setting version number for superblock loading */
5275 if (info
->major_version
< 0 ||
5276 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5277 super_types
[info
->major_version
].name
== NULL
) {
5278 /* maybe try to auto-load a module? */
5280 "md: superblock version %d not known\n",
5281 info
->major_version
);
5284 mddev
->major_version
= info
->major_version
;
5285 mddev
->minor_version
= info
->minor_version
;
5286 mddev
->patch_version
= info
->patch_version
;
5287 mddev
->persistent
= !info
->not_persistent
;
5288 /* ensure mddev_put doesn't delete this now that there
5289 * is some minimal configuration.
5291 mddev
->ctime
= get_seconds();
5294 mddev
->major_version
= MD_MAJOR_VERSION
;
5295 mddev
->minor_version
= MD_MINOR_VERSION
;
5296 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5297 mddev
->ctime
= get_seconds();
5299 mddev
->level
= info
->level
;
5300 mddev
->clevel
[0] = 0;
5301 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5302 mddev
->raid_disks
= info
->raid_disks
;
5303 /* don't set md_minor, it is determined by which /dev/md* was
5306 if (info
->state
& (1<<MD_SB_CLEAN
))
5307 mddev
->recovery_cp
= MaxSector
;
5309 mddev
->recovery_cp
= 0;
5310 mddev
->persistent
= ! info
->not_persistent
;
5311 mddev
->external
= 0;
5313 mddev
->layout
= info
->layout
;
5314 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5316 mddev
->max_disks
= MD_SB_DISKS
;
5318 if (mddev
->persistent
)
5320 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5322 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5323 mddev
->bitmap_info
.offset
= 0;
5325 mddev
->reshape_position
= MaxSector
;
5328 * Generate a 128 bit UUID
5330 get_random_bytes(mddev
->uuid
, 16);
5332 mddev
->new_level
= mddev
->level
;
5333 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5334 mddev
->new_layout
= mddev
->layout
;
5335 mddev
->delta_disks
= 0;
5340 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5342 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5344 if (mddev
->external_size
)
5347 mddev
->array_sectors
= array_sectors
;
5349 EXPORT_SYMBOL(md_set_array_sectors
);
5351 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5355 int fit
= (num_sectors
== 0);
5357 if (mddev
->pers
->resize
== NULL
)
5359 /* The "num_sectors" is the number of sectors of each device that
5360 * is used. This can only make sense for arrays with redundancy.
5361 * linear and raid0 always use whatever space is available. We can only
5362 * consider changing this number if no resync or reconstruction is
5363 * happening, and if the new size is acceptable. It must fit before the
5364 * sb_start or, if that is <data_offset, it must fit before the size
5365 * of each device. If num_sectors is zero, we find the largest size
5369 if (mddev
->sync_thread
)
5372 /* Sorry, cannot grow a bitmap yet, just remove it,
5376 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5377 sector_t avail
= rdev
->sectors
;
5379 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5380 num_sectors
= avail
;
5381 if (avail
< num_sectors
)
5384 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5386 revalidate_disk(mddev
->gendisk
);
5390 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5393 /* change the number of raid disks */
5394 if (mddev
->pers
->check_reshape
== NULL
)
5396 if (raid_disks
<= 0 ||
5397 raid_disks
>= mddev
->max_disks
)
5399 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5401 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5403 rv
= mddev
->pers
->check_reshape(mddev
);
5409 * update_array_info is used to change the configuration of an
5411 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5412 * fields in the info are checked against the array.
5413 * Any differences that cannot be handled will cause an error.
5414 * Normally, only one change can be managed at a time.
5416 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5422 /* calculate expected state,ignoring low bits */
5423 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5424 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5426 if (mddev
->major_version
!= info
->major_version
||
5427 mddev
->minor_version
!= info
->minor_version
||
5428 /* mddev->patch_version != info->patch_version || */
5429 mddev
->ctime
!= info
->ctime
||
5430 mddev
->level
!= info
->level
||
5431 /* mddev->layout != info->layout || */
5432 !mddev
->persistent
!= info
->not_persistent
||
5433 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5434 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5435 ((state
^info
->state
) & 0xfffffe00)
5438 /* Check there is only one change */
5439 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5441 if (mddev
->raid_disks
!= info
->raid_disks
)
5443 if (mddev
->layout
!= info
->layout
)
5445 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5452 if (mddev
->layout
!= info
->layout
) {
5454 * we don't need to do anything at the md level, the
5455 * personality will take care of it all.
5457 if (mddev
->pers
->check_reshape
== NULL
)
5460 mddev
->new_layout
= info
->layout
;
5461 rv
= mddev
->pers
->check_reshape(mddev
);
5463 mddev
->new_layout
= mddev
->layout
;
5467 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5468 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5470 if (mddev
->raid_disks
!= info
->raid_disks
)
5471 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5473 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5474 if (mddev
->pers
->quiesce
== NULL
)
5476 if (mddev
->recovery
|| mddev
->sync_thread
)
5478 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5479 /* add the bitmap */
5482 if (mddev
->bitmap_info
.default_offset
== 0)
5484 mddev
->bitmap_info
.offset
=
5485 mddev
->bitmap_info
.default_offset
;
5486 mddev
->pers
->quiesce(mddev
, 1);
5487 rv
= bitmap_create(mddev
);
5489 bitmap_destroy(mddev
);
5490 mddev
->pers
->quiesce(mddev
, 0);
5492 /* remove the bitmap */
5495 if (mddev
->bitmap
->file
)
5497 mddev
->pers
->quiesce(mddev
, 1);
5498 bitmap_destroy(mddev
);
5499 mddev
->pers
->quiesce(mddev
, 0);
5500 mddev
->bitmap_info
.offset
= 0;
5503 md_update_sb(mddev
, 1);
5507 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5511 if (mddev
->pers
== NULL
)
5514 rdev
= find_rdev(mddev
, dev
);
5518 md_error(mddev
, rdev
);
5523 * We have a problem here : there is no easy way to give a CHS
5524 * virtual geometry. We currently pretend that we have a 2 heads
5525 * 4 sectors (with a BIG number of cylinders...). This drives
5526 * dosfs just mad... ;-)
5528 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5530 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5534 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
5538 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5539 unsigned int cmd
, unsigned long arg
)
5542 void __user
*argp
= (void __user
*)arg
;
5543 mddev_t
*mddev
= NULL
;
5546 if (!capable(CAP_SYS_ADMIN
))
5550 * Commands dealing with the RAID driver but not any
5556 err
= get_version(argp
);
5559 case PRINT_RAID_DEBUG
:
5567 autostart_arrays(arg
);
5574 * Commands creating/starting a new array:
5577 mddev
= bdev
->bd_disk
->private_data
;
5584 err
= mddev_lock(mddev
);
5587 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5594 case SET_ARRAY_INFO
:
5596 mdu_array_info_t info
;
5598 memset(&info
, 0, sizeof(info
));
5599 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5604 err
= update_array_info(mddev
, &info
);
5606 printk(KERN_WARNING
"md: couldn't update"
5607 " array info. %d\n", err
);
5612 if (!list_empty(&mddev
->disks
)) {
5614 "md: array %s already has disks!\n",
5619 if (mddev
->raid_disks
) {
5621 "md: array %s already initialised!\n",
5626 err
= set_array_info(mddev
, &info
);
5628 printk(KERN_WARNING
"md: couldn't set"
5629 " array info. %d\n", err
);
5639 * Commands querying/configuring an existing array:
5641 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5642 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5643 if ((!mddev
->raid_disks
&& !mddev
->external
)
5644 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5645 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5646 && cmd
!= GET_BITMAP_FILE
) {
5652 * Commands even a read-only array can execute:
5656 case GET_ARRAY_INFO
:
5657 err
= get_array_info(mddev
, argp
);
5660 case GET_BITMAP_FILE
:
5661 err
= get_bitmap_file(mddev
, argp
);
5665 err
= get_disk_info(mddev
, argp
);
5668 case RESTART_ARRAY_RW
:
5669 err
= restart_array(mddev
);
5673 err
= do_md_stop(mddev
, 0, 1);
5677 err
= do_md_stop(mddev
, 1, 1);
5681 if (get_user(ro
, (int __user
*)(arg
))) {
5687 /* if the bdev is going readonly the value of mddev->ro
5688 * does not matter, no writes are coming
5693 /* are we are already prepared for writes? */
5697 /* transitioning to readauto need only happen for
5698 * arrays that call md_write_start
5701 err
= restart_array(mddev
);
5704 set_disk_ro(mddev
->gendisk
, 0);
5711 * The remaining ioctls are changing the state of the
5712 * superblock, so we do not allow them on read-only arrays.
5713 * However non-MD ioctls (e.g. get-size) will still come through
5714 * here and hit the 'default' below, so only disallow
5715 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5717 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5718 if (mddev
->ro
== 2) {
5720 sysfs_notify_dirent(mddev
->sysfs_state
);
5721 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5722 md_wakeup_thread(mddev
->thread
);
5733 mdu_disk_info_t info
;
5734 if (copy_from_user(&info
, argp
, sizeof(info
)))
5737 err
= add_new_disk(mddev
, &info
);
5741 case HOT_REMOVE_DISK
:
5742 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5746 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5749 case SET_DISK_FAULTY
:
5750 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5754 err
= do_md_run(mddev
);
5757 case SET_BITMAP_FILE
:
5758 err
= set_bitmap_file(mddev
, (int)arg
);
5768 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5770 mddev
->hold_active
= 0;
5771 mddev_unlock(mddev
);
5780 #ifdef CONFIG_COMPAT
5781 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
5782 unsigned int cmd
, unsigned long arg
)
5785 case HOT_REMOVE_DISK
:
5787 case SET_DISK_FAULTY
:
5788 case SET_BITMAP_FILE
:
5789 /* These take in integer arg, do not convert */
5792 arg
= (unsigned long)compat_ptr(arg
);
5796 return md_ioctl(bdev
, mode
, cmd
, arg
);
5798 #endif /* CONFIG_COMPAT */
5800 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5803 * Succeed if we can lock the mddev, which confirms that
5804 * it isn't being stopped right now.
5806 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5809 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5810 /* we are racing with mddev_put which is discarding this
5814 /* Wait until bdev->bd_disk is definitely gone */
5815 flush_scheduled_work();
5816 /* Then retry the open from the top */
5817 return -ERESTARTSYS
;
5819 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5821 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5825 atomic_inc(&mddev
->openers
);
5826 mutex_unlock(&mddev
->open_mutex
);
5828 check_disk_change(bdev
);
5833 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5835 mddev_t
*mddev
= disk
->private_data
;
5838 atomic_dec(&mddev
->openers
);
5844 static int md_media_changed(struct gendisk
*disk
)
5846 mddev_t
*mddev
= disk
->private_data
;
5848 return mddev
->changed
;
5851 static int md_revalidate(struct gendisk
*disk
)
5853 mddev_t
*mddev
= disk
->private_data
;
5858 static const struct block_device_operations md_fops
=
5860 .owner
= THIS_MODULE
,
5862 .release
= md_release
,
5864 #ifdef CONFIG_COMPAT
5865 .compat_ioctl
= md_compat_ioctl
,
5867 .getgeo
= md_getgeo
,
5868 .media_changed
= md_media_changed
,
5869 .revalidate_disk
= md_revalidate
,
5872 static int md_thread(void * arg
)
5874 mdk_thread_t
*thread
= arg
;
5877 * md_thread is a 'system-thread', it's priority should be very
5878 * high. We avoid resource deadlocks individually in each
5879 * raid personality. (RAID5 does preallocation) We also use RR and
5880 * the very same RT priority as kswapd, thus we will never get
5881 * into a priority inversion deadlock.
5883 * we definitely have to have equal or higher priority than
5884 * bdflush, otherwise bdflush will deadlock if there are too
5885 * many dirty RAID5 blocks.
5888 allow_signal(SIGKILL
);
5889 while (!kthread_should_stop()) {
5891 /* We need to wait INTERRUPTIBLE so that
5892 * we don't add to the load-average.
5893 * That means we need to be sure no signals are
5896 if (signal_pending(current
))
5897 flush_signals(current
);
5899 wait_event_interruptible_timeout
5901 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5902 || kthread_should_stop(),
5905 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5907 thread
->run(thread
->mddev
);
5913 void md_wakeup_thread(mdk_thread_t
*thread
)
5916 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5917 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5918 wake_up(&thread
->wqueue
);
5922 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5925 mdk_thread_t
*thread
;
5927 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5931 init_waitqueue_head(&thread
->wqueue
);
5934 thread
->mddev
= mddev
;
5935 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5936 thread
->tsk
= kthread_run(md_thread
, thread
,
5938 mdname(thread
->mddev
),
5939 name
?: mddev
->pers
->name
);
5940 if (IS_ERR(thread
->tsk
)) {
5947 void md_unregister_thread(mdk_thread_t
*thread
)
5951 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5953 kthread_stop(thread
->tsk
);
5957 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5964 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5967 if (mddev
->external
)
5968 set_bit(Blocked
, &rdev
->flags
);
5970 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5972 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5973 __builtin_return_address(0),__builtin_return_address(1),
5974 __builtin_return_address(2),__builtin_return_address(3));
5978 if (!mddev
->pers
->error_handler
)
5980 mddev
->pers
->error_handler(mddev
,rdev
);
5981 if (mddev
->degraded
)
5982 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5983 set_bit(StateChanged
, &rdev
->flags
);
5984 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5985 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5986 md_wakeup_thread(mddev
->thread
);
5987 md_new_event_inintr(mddev
);
5990 /* seq_file implementation /proc/mdstat */
5992 static void status_unused(struct seq_file
*seq
)
5997 seq_printf(seq
, "unused devices: ");
5999 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6000 char b
[BDEVNAME_SIZE
];
6002 seq_printf(seq
, "%s ",
6003 bdevname(rdev
->bdev
,b
));
6006 seq_printf(seq
, "<none>");
6008 seq_printf(seq
, "\n");
6012 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6014 sector_t max_sectors
, resync
, res
;
6015 unsigned long dt
, db
;
6018 unsigned int per_milli
;
6020 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6022 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6023 max_sectors
= mddev
->resync_max_sectors
;
6025 max_sectors
= mddev
->dev_sectors
;
6028 * Should not happen.
6034 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6035 * in a sector_t, and (max_sectors>>scale) will fit in a
6036 * u32, as those are the requirements for sector_div.
6037 * Thus 'scale' must be at least 10
6040 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6041 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6044 res
= (resync
>>scale
)*1000;
6045 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6049 int i
, x
= per_milli
/50, y
= 20-x
;
6050 seq_printf(seq
, "[");
6051 for (i
= 0; i
< x
; i
++)
6052 seq_printf(seq
, "=");
6053 seq_printf(seq
, ">");
6054 for (i
= 0; i
< y
; i
++)
6055 seq_printf(seq
, ".");
6056 seq_printf(seq
, "] ");
6058 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6059 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6061 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6063 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6064 "resync" : "recovery"))),
6065 per_milli
/10, per_milli
% 10,
6066 (unsigned long long) resync
/2,
6067 (unsigned long long) max_sectors
/2);
6070 * dt: time from mark until now
6071 * db: blocks written from mark until now
6072 * rt: remaining time
6074 * rt is a sector_t, so could be 32bit or 64bit.
6075 * So we divide before multiply in case it is 32bit and close
6077 * We scale the divisor (db) by 32 to avoid loosing precision
6078 * near the end of resync when the number of remaining sectors
6080 * We then divide rt by 32 after multiplying by db to compensate.
6081 * The '+1' avoids division by zero if db is very small.
6083 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6085 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6086 - mddev
->resync_mark_cnt
;
6088 rt
= max_sectors
- resync
; /* number of remaining sectors */
6089 sector_div(rt
, db
/32+1);
6093 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6094 ((unsigned long)rt
% 60)/6);
6096 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6099 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6101 struct list_head
*tmp
;
6111 spin_lock(&all_mddevs_lock
);
6112 list_for_each(tmp
,&all_mddevs
)
6114 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6116 spin_unlock(&all_mddevs_lock
);
6119 spin_unlock(&all_mddevs_lock
);
6121 return (void*)2;/* tail */
6125 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6127 struct list_head
*tmp
;
6128 mddev_t
*next_mddev
, *mddev
= v
;
6134 spin_lock(&all_mddevs_lock
);
6136 tmp
= all_mddevs
.next
;
6138 tmp
= mddev
->all_mddevs
.next
;
6139 if (tmp
!= &all_mddevs
)
6140 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6142 next_mddev
= (void*)2;
6145 spin_unlock(&all_mddevs_lock
);
6153 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6157 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6161 struct mdstat_info
{
6165 static int md_seq_show(struct seq_file
*seq
, void *v
)
6170 struct mdstat_info
*mi
= seq
->private;
6171 struct bitmap
*bitmap
;
6173 if (v
== (void*)1) {
6174 struct mdk_personality
*pers
;
6175 seq_printf(seq
, "Personalities : ");
6176 spin_lock(&pers_lock
);
6177 list_for_each_entry(pers
, &pers_list
, list
)
6178 seq_printf(seq
, "[%s] ", pers
->name
);
6180 spin_unlock(&pers_lock
);
6181 seq_printf(seq
, "\n");
6182 mi
->event
= atomic_read(&md_event_count
);
6185 if (v
== (void*)2) {
6190 if (mddev_lock(mddev
) < 0)
6193 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6194 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6195 mddev
->pers
? "" : "in");
6198 seq_printf(seq
, " (read-only)");
6200 seq_printf(seq
, " (auto-read-only)");
6201 seq_printf(seq
, " %s", mddev
->pers
->name
);
6205 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6206 char b
[BDEVNAME_SIZE
];
6207 seq_printf(seq
, " %s[%d]",
6208 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6209 if (test_bit(WriteMostly
, &rdev
->flags
))
6210 seq_printf(seq
, "(W)");
6211 if (test_bit(Faulty
, &rdev
->flags
)) {
6212 seq_printf(seq
, "(F)");
6214 } else if (rdev
->raid_disk
< 0)
6215 seq_printf(seq
, "(S)"); /* spare */
6216 sectors
+= rdev
->sectors
;
6219 if (!list_empty(&mddev
->disks
)) {
6221 seq_printf(seq
, "\n %llu blocks",
6222 (unsigned long long)
6223 mddev
->array_sectors
/ 2);
6225 seq_printf(seq
, "\n %llu blocks",
6226 (unsigned long long)sectors
/ 2);
6228 if (mddev
->persistent
) {
6229 if (mddev
->major_version
!= 0 ||
6230 mddev
->minor_version
!= 90) {
6231 seq_printf(seq
," super %d.%d",
6232 mddev
->major_version
,
6233 mddev
->minor_version
);
6235 } else if (mddev
->external
)
6236 seq_printf(seq
, " super external:%s",
6237 mddev
->metadata_type
);
6239 seq_printf(seq
, " super non-persistent");
6242 mddev
->pers
->status(seq
, mddev
);
6243 seq_printf(seq
, "\n ");
6244 if (mddev
->pers
->sync_request
) {
6245 if (mddev
->curr_resync
> 2) {
6246 status_resync(seq
, mddev
);
6247 seq_printf(seq
, "\n ");
6248 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6249 seq_printf(seq
, "\tresync=DELAYED\n ");
6250 else if (mddev
->recovery_cp
< MaxSector
)
6251 seq_printf(seq
, "\tresync=PENDING\n ");
6254 seq_printf(seq
, "\n ");
6256 if ((bitmap
= mddev
->bitmap
)) {
6257 unsigned long chunk_kb
;
6258 unsigned long flags
;
6259 spin_lock_irqsave(&bitmap
->lock
, flags
);
6260 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6261 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6263 bitmap
->pages
- bitmap
->missing_pages
,
6265 (bitmap
->pages
- bitmap
->missing_pages
)
6266 << (PAGE_SHIFT
- 10),
6267 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6268 chunk_kb
? "KB" : "B");
6270 seq_printf(seq
, ", file: ");
6271 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6274 seq_printf(seq
, "\n");
6275 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6278 seq_printf(seq
, "\n");
6280 mddev_unlock(mddev
);
6285 static const struct seq_operations md_seq_ops
= {
6286 .start
= md_seq_start
,
6287 .next
= md_seq_next
,
6288 .stop
= md_seq_stop
,
6289 .show
= md_seq_show
,
6292 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6295 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6299 error
= seq_open(file
, &md_seq_ops
);
6303 struct seq_file
*p
= file
->private_data
;
6305 mi
->event
= atomic_read(&md_event_count
);
6310 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6312 struct seq_file
*m
= filp
->private_data
;
6313 struct mdstat_info
*mi
= m
->private;
6316 poll_wait(filp
, &md_event_waiters
, wait
);
6318 /* always allow read */
6319 mask
= POLLIN
| POLLRDNORM
;
6321 if (mi
->event
!= atomic_read(&md_event_count
))
6322 mask
|= POLLERR
| POLLPRI
;
6326 static const struct file_operations md_seq_fops
= {
6327 .owner
= THIS_MODULE
,
6328 .open
= md_seq_open
,
6330 .llseek
= seq_lseek
,
6331 .release
= seq_release_private
,
6332 .poll
= mdstat_poll
,
6335 int register_md_personality(struct mdk_personality
*p
)
6337 spin_lock(&pers_lock
);
6338 list_add_tail(&p
->list
, &pers_list
);
6339 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6340 spin_unlock(&pers_lock
);
6344 int unregister_md_personality(struct mdk_personality
*p
)
6346 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6347 spin_lock(&pers_lock
);
6348 list_del_init(&p
->list
);
6349 spin_unlock(&pers_lock
);
6353 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6361 rdev_for_each_rcu(rdev
, mddev
) {
6362 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6363 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6364 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6365 atomic_read(&disk
->sync_io
);
6366 /* sync IO will cause sync_io to increase before the disk_stats
6367 * as sync_io is counted when a request starts, and
6368 * disk_stats is counted when it completes.
6369 * So resync activity will cause curr_events to be smaller than
6370 * when there was no such activity.
6371 * non-sync IO will cause disk_stat to increase without
6372 * increasing sync_io so curr_events will (eventually)
6373 * be larger than it was before. Once it becomes
6374 * substantially larger, the test below will cause
6375 * the array to appear non-idle, and resync will slow
6377 * If there is a lot of outstanding resync activity when
6378 * we set last_event to curr_events, then all that activity
6379 * completing might cause the array to appear non-idle
6380 * and resync will be slowed down even though there might
6381 * not have been non-resync activity. This will only
6382 * happen once though. 'last_events' will soon reflect
6383 * the state where there is little or no outstanding
6384 * resync requests, and further resync activity will
6385 * always make curr_events less than last_events.
6388 if (init
|| curr_events
- rdev
->last_events
> 64) {
6389 rdev
->last_events
= curr_events
;
6397 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6399 /* another "blocks" (512byte) blocks have been synced */
6400 atomic_sub(blocks
, &mddev
->recovery_active
);
6401 wake_up(&mddev
->recovery_wait
);
6403 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6404 md_wakeup_thread(mddev
->thread
);
6405 // stop recovery, signal do_sync ....
6410 /* md_write_start(mddev, bi)
6411 * If we need to update some array metadata (e.g. 'active' flag
6412 * in superblock) before writing, schedule a superblock update
6413 * and wait for it to complete.
6415 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6418 if (bio_data_dir(bi
) != WRITE
)
6421 BUG_ON(mddev
->ro
== 1);
6422 if (mddev
->ro
== 2) {
6423 /* need to switch to read/write */
6425 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6426 md_wakeup_thread(mddev
->thread
);
6427 md_wakeup_thread(mddev
->sync_thread
);
6430 atomic_inc(&mddev
->writes_pending
);
6431 if (mddev
->safemode
== 1)
6432 mddev
->safemode
= 0;
6433 if (mddev
->in_sync
) {
6434 spin_lock_irq(&mddev
->write_lock
);
6435 if (mddev
->in_sync
) {
6437 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6438 md_wakeup_thread(mddev
->thread
);
6441 spin_unlock_irq(&mddev
->write_lock
);
6444 sysfs_notify_dirent(mddev
->sysfs_state
);
6445 wait_event(mddev
->sb_wait
,
6446 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6447 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6450 void md_write_end(mddev_t
*mddev
)
6452 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6453 if (mddev
->safemode
== 2)
6454 md_wakeup_thread(mddev
->thread
);
6455 else if (mddev
->safemode_delay
)
6456 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6460 /* md_allow_write(mddev)
6461 * Calling this ensures that the array is marked 'active' so that writes
6462 * may proceed without blocking. It is important to call this before
6463 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6464 * Must be called with mddev_lock held.
6466 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6467 * is dropped, so return -EAGAIN after notifying userspace.
6469 int md_allow_write(mddev_t
*mddev
)
6475 if (!mddev
->pers
->sync_request
)
6478 spin_lock_irq(&mddev
->write_lock
);
6479 if (mddev
->in_sync
) {
6481 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6482 if (mddev
->safemode_delay
&&
6483 mddev
->safemode
== 0)
6484 mddev
->safemode
= 1;
6485 spin_unlock_irq(&mddev
->write_lock
);
6486 md_update_sb(mddev
, 0);
6487 sysfs_notify_dirent(mddev
->sysfs_state
);
6489 spin_unlock_irq(&mddev
->write_lock
);
6491 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6496 EXPORT_SYMBOL_GPL(md_allow_write
);
6498 #define SYNC_MARKS 10
6499 #define SYNC_MARK_STEP (3*HZ)
6500 void md_do_sync(mddev_t
*mddev
)
6503 unsigned int currspeed
= 0,
6505 sector_t max_sectors
,j
, io_sectors
;
6506 unsigned long mark
[SYNC_MARKS
];
6507 sector_t mark_cnt
[SYNC_MARKS
];
6509 struct list_head
*tmp
;
6510 sector_t last_check
;
6515 /* just incase thread restarts... */
6516 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6518 if (mddev
->ro
) /* never try to sync a read-only array */
6521 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6522 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6523 desc
= "data-check";
6524 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6525 desc
= "requested-resync";
6528 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6533 /* we overload curr_resync somewhat here.
6534 * 0 == not engaged in resync at all
6535 * 2 == checking that there is no conflict with another sync
6536 * 1 == like 2, but have yielded to allow conflicting resync to
6538 * other == active in resync - this many blocks
6540 * Before starting a resync we must have set curr_resync to
6541 * 2, and then checked that every "conflicting" array has curr_resync
6542 * less than ours. When we find one that is the same or higher
6543 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6544 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6545 * This will mean we have to start checking from the beginning again.
6550 mddev
->curr_resync
= 2;
6553 if (kthread_should_stop())
6554 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6556 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6558 for_each_mddev(mddev2
, tmp
) {
6559 if (mddev2
== mddev
)
6561 if (!mddev
->parallel_resync
6562 && mddev2
->curr_resync
6563 && match_mddev_units(mddev
, mddev2
)) {
6565 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6566 /* arbitrarily yield */
6567 mddev
->curr_resync
= 1;
6568 wake_up(&resync_wait
);
6570 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6571 /* no need to wait here, we can wait the next
6572 * time 'round when curr_resync == 2
6575 /* We need to wait 'interruptible' so as not to
6576 * contribute to the load average, and not to
6577 * be caught by 'softlockup'
6579 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6580 if (!kthread_should_stop() &&
6581 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6582 printk(KERN_INFO
"md: delaying %s of %s"
6583 " until %s has finished (they"
6584 " share one or more physical units)\n",
6585 desc
, mdname(mddev
), mdname(mddev2
));
6587 if (signal_pending(current
))
6588 flush_signals(current
);
6590 finish_wait(&resync_wait
, &wq
);
6593 finish_wait(&resync_wait
, &wq
);
6596 } while (mddev
->curr_resync
< 2);
6599 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6600 /* resync follows the size requested by the personality,
6601 * which defaults to physical size, but can be virtual size
6603 max_sectors
= mddev
->resync_max_sectors
;
6604 mddev
->resync_mismatches
= 0;
6605 /* we don't use the checkpoint if there's a bitmap */
6606 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6607 j
= mddev
->resync_min
;
6608 else if (!mddev
->bitmap
)
6609 j
= mddev
->recovery_cp
;
6611 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6612 max_sectors
= mddev
->dev_sectors
;
6614 /* recovery follows the physical size of devices */
6615 max_sectors
= mddev
->dev_sectors
;
6618 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6619 if (rdev
->raid_disk
>= 0 &&
6620 !test_bit(Faulty
, &rdev
->flags
) &&
6621 !test_bit(In_sync
, &rdev
->flags
) &&
6622 rdev
->recovery_offset
< j
)
6623 j
= rdev
->recovery_offset
;
6627 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6628 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6629 " %d KB/sec/disk.\n", speed_min(mddev
));
6630 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6631 "(but not more than %d KB/sec) for %s.\n",
6632 speed_max(mddev
), desc
);
6634 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6637 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6639 mark_cnt
[m
] = io_sectors
;
6642 mddev
->resync_mark
= mark
[last_mark
];
6643 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6646 * Tune reconstruction:
6648 window
= 32*(PAGE_SIZE
/512);
6649 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6650 window
/2,(unsigned long long) max_sectors
/2);
6652 atomic_set(&mddev
->recovery_active
, 0);
6657 "md: resuming %s of %s from checkpoint.\n",
6658 desc
, mdname(mddev
));
6659 mddev
->curr_resync
= j
;
6661 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6663 while (j
< max_sectors
) {
6668 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6669 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6670 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6671 > (max_sectors
>> 4)) ||
6672 (j
- mddev
->curr_resync_completed
)*2
6673 >= mddev
->resync_max
- mddev
->curr_resync_completed
6675 /* time to update curr_resync_completed */
6676 blk_unplug(mddev
->queue
);
6677 wait_event(mddev
->recovery_wait
,
6678 atomic_read(&mddev
->recovery_active
) == 0);
6679 mddev
->curr_resync_completed
=
6681 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6682 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6685 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6686 /* As this condition is controlled by user-space,
6687 * we can block indefinitely, so use '_interruptible'
6688 * to avoid triggering warnings.
6690 flush_signals(current
); /* just in case */
6691 wait_event_interruptible(mddev
->recovery_wait
,
6692 mddev
->resync_max
> j
6693 || kthread_should_stop());
6696 if (kthread_should_stop())
6699 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6700 currspeed
< speed_min(mddev
));
6702 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6706 if (!skipped
) { /* actual IO requested */
6707 io_sectors
+= sectors
;
6708 atomic_add(sectors
, &mddev
->recovery_active
);
6712 if (j
>1) mddev
->curr_resync
= j
;
6713 mddev
->curr_mark_cnt
= io_sectors
;
6714 if (last_check
== 0)
6715 /* this is the earliers that rebuilt will be
6716 * visible in /proc/mdstat
6718 md_new_event(mddev
);
6720 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6723 last_check
= io_sectors
;
6725 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6729 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6731 int next
= (last_mark
+1) % SYNC_MARKS
;
6733 mddev
->resync_mark
= mark
[next
];
6734 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6735 mark
[next
] = jiffies
;
6736 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6741 if (kthread_should_stop())
6746 * this loop exits only if either when we are slower than
6747 * the 'hard' speed limit, or the system was IO-idle for
6749 * the system might be non-idle CPU-wise, but we only care
6750 * about not overloading the IO subsystem. (things like an
6751 * e2fsck being done on the RAID array should execute fast)
6753 blk_unplug(mddev
->queue
);
6756 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6757 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6759 if (currspeed
> speed_min(mddev
)) {
6760 if ((currspeed
> speed_max(mddev
)) ||
6761 !is_mddev_idle(mddev
, 0)) {
6767 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6769 * this also signals 'finished resyncing' to md_stop
6772 blk_unplug(mddev
->queue
);
6774 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6776 /* tell personality that we are finished */
6777 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6779 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6780 mddev
->curr_resync
> 2) {
6781 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6782 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6783 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6785 "md: checkpointing %s of %s.\n",
6786 desc
, mdname(mddev
));
6787 mddev
->recovery_cp
= mddev
->curr_resync
;
6790 mddev
->recovery_cp
= MaxSector
;
6792 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6793 mddev
->curr_resync
= MaxSector
;
6795 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6796 if (rdev
->raid_disk
>= 0 &&
6797 !test_bit(Faulty
, &rdev
->flags
) &&
6798 !test_bit(In_sync
, &rdev
->flags
) &&
6799 rdev
->recovery_offset
< mddev
->curr_resync
)
6800 rdev
->recovery_offset
= mddev
->curr_resync
;
6804 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6807 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6808 /* We completed so min/max setting can be forgotten if used. */
6809 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6810 mddev
->resync_min
= 0;
6811 mddev
->resync_max
= MaxSector
;
6812 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6813 mddev
->resync_min
= mddev
->curr_resync_completed
;
6814 mddev
->curr_resync
= 0;
6815 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6816 mddev
->curr_resync_completed
= 0;
6817 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6818 wake_up(&resync_wait
);
6819 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6820 md_wakeup_thread(mddev
->thread
);
6825 * got a signal, exit.
6828 "md: md_do_sync() got signal ... exiting\n");
6829 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6833 EXPORT_SYMBOL_GPL(md_do_sync
);
6836 static int remove_and_add_spares(mddev_t
*mddev
)
6841 mddev
->curr_resync_completed
= 0;
6843 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6844 if (rdev
->raid_disk
>= 0 &&
6845 !test_bit(Blocked
, &rdev
->flags
) &&
6846 (test_bit(Faulty
, &rdev
->flags
) ||
6847 ! test_bit(In_sync
, &rdev
->flags
)) &&
6848 atomic_read(&rdev
->nr_pending
)==0) {
6849 if (mddev
->pers
->hot_remove_disk(
6850 mddev
, rdev
->raid_disk
)==0) {
6852 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6853 sysfs_remove_link(&mddev
->kobj
, nm
);
6854 rdev
->raid_disk
= -1;
6858 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6859 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6860 if (rdev
->raid_disk
>= 0 &&
6861 !test_bit(In_sync
, &rdev
->flags
) &&
6862 !test_bit(Blocked
, &rdev
->flags
))
6864 if (rdev
->raid_disk
< 0
6865 && !test_bit(Faulty
, &rdev
->flags
)) {
6866 rdev
->recovery_offset
= 0;
6868 hot_add_disk(mddev
, rdev
) == 0) {
6870 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6871 if (sysfs_create_link(&mddev
->kobj
,
6874 "md: cannot register "
6878 md_new_event(mddev
);
6879 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6888 * This routine is regularly called by all per-raid-array threads to
6889 * deal with generic issues like resync and super-block update.
6890 * Raid personalities that don't have a thread (linear/raid0) do not
6891 * need this as they never do any recovery or update the superblock.
6893 * It does not do any resync itself, but rather "forks" off other threads
6894 * to do that as needed.
6895 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6896 * "->recovery" and create a thread at ->sync_thread.
6897 * When the thread finishes it sets MD_RECOVERY_DONE
6898 * and wakeups up this thread which will reap the thread and finish up.
6899 * This thread also removes any faulty devices (with nr_pending == 0).
6901 * The overall approach is:
6902 * 1/ if the superblock needs updating, update it.
6903 * 2/ If a recovery thread is running, don't do anything else.
6904 * 3/ If recovery has finished, clean up, possibly marking spares active.
6905 * 4/ If there are any faulty devices, remove them.
6906 * 5/ If array is degraded, try to add spares devices
6907 * 6/ If array has spares or is not in-sync, start a resync thread.
6909 void md_check_recovery(mddev_t
*mddev
)
6915 bitmap_daemon_work(mddev
);
6920 if (signal_pending(current
)) {
6921 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6922 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6924 mddev
->safemode
= 2;
6926 flush_signals(current
);
6929 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6932 (mddev
->flags
&& !mddev
->external
) ||
6933 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6934 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6935 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6936 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6937 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6941 if (mddev_trylock(mddev
)) {
6945 /* Only thing we do on a ro array is remove
6948 remove_and_add_spares(mddev
);
6949 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6953 if (!mddev
->external
) {
6955 spin_lock_irq(&mddev
->write_lock
);
6956 if (mddev
->safemode
&&
6957 !atomic_read(&mddev
->writes_pending
) &&
6959 mddev
->recovery_cp
== MaxSector
) {
6962 if (mddev
->persistent
)
6963 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6965 if (mddev
->safemode
== 1)
6966 mddev
->safemode
= 0;
6967 spin_unlock_irq(&mddev
->write_lock
);
6969 sysfs_notify_dirent(mddev
->sysfs_state
);
6973 md_update_sb(mddev
, 0);
6975 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6976 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6977 sysfs_notify_dirent(rdev
->sysfs_state
);
6980 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6981 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6982 /* resync/recovery still happening */
6983 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6986 if (mddev
->sync_thread
) {
6987 /* resync has finished, collect result */
6988 md_unregister_thread(mddev
->sync_thread
);
6989 mddev
->sync_thread
= NULL
;
6990 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6991 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6993 /* activate any spares */
6994 if (mddev
->pers
->spare_active(mddev
))
6995 sysfs_notify(&mddev
->kobj
, NULL
,
6998 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6999 mddev
->pers
->finish_reshape
)
7000 mddev
->pers
->finish_reshape(mddev
);
7001 md_update_sb(mddev
, 1);
7003 /* if array is no-longer degraded, then any saved_raid_disk
7004 * information must be scrapped
7006 if (!mddev
->degraded
)
7007 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7008 rdev
->saved_raid_disk
= -1;
7010 mddev
->recovery
= 0;
7011 /* flag recovery needed just to double check */
7012 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7013 sysfs_notify_dirent(mddev
->sysfs_action
);
7014 md_new_event(mddev
);
7017 /* Set RUNNING before clearing NEEDED to avoid
7018 * any transients in the value of "sync_action".
7020 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7021 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7022 /* Clear some bits that don't mean anything, but
7025 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7026 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7028 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7030 /* no recovery is running.
7031 * remove any failed drives, then
7032 * add spares if possible.
7033 * Spare are also removed and re-added, to allow
7034 * the personality to fail the re-add.
7037 if (mddev
->reshape_position
!= MaxSector
) {
7038 if (mddev
->pers
->check_reshape
== NULL
||
7039 mddev
->pers
->check_reshape(mddev
) != 0)
7040 /* Cannot proceed */
7042 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7043 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7044 } else if ((spares
= remove_and_add_spares(mddev
))) {
7045 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7046 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7047 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7048 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7049 } else if (mddev
->recovery_cp
< MaxSector
) {
7050 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7051 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7052 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7053 /* nothing to be done ... */
7056 if (mddev
->pers
->sync_request
) {
7057 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7058 /* We are adding a device or devices to an array
7059 * which has the bitmap stored on all devices.
7060 * So make sure all bitmap pages get written
7062 bitmap_write_all(mddev
->bitmap
);
7064 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7067 if (!mddev
->sync_thread
) {
7068 printk(KERN_ERR
"%s: could not start resync"
7071 /* leave the spares where they are, it shouldn't hurt */
7072 mddev
->recovery
= 0;
7074 md_wakeup_thread(mddev
->sync_thread
);
7075 sysfs_notify_dirent(mddev
->sysfs_action
);
7076 md_new_event(mddev
);
7079 if (!mddev
->sync_thread
) {
7080 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7081 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7083 if (mddev
->sysfs_action
)
7084 sysfs_notify_dirent(mddev
->sysfs_action
);
7086 mddev_unlock(mddev
);
7090 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7092 sysfs_notify_dirent(rdev
->sysfs_state
);
7093 wait_event_timeout(rdev
->blocked_wait
,
7094 !test_bit(Blocked
, &rdev
->flags
),
7095 msecs_to_jiffies(5000));
7096 rdev_dec_pending(rdev
, mddev
);
7098 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7100 static int md_notify_reboot(struct notifier_block
*this,
7101 unsigned long code
, void *x
)
7103 struct list_head
*tmp
;
7106 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7108 printk(KERN_INFO
"md: stopping all md devices.\n");
7110 for_each_mddev(mddev
, tmp
)
7111 if (mddev_trylock(mddev
)) {
7112 /* Force a switch to readonly even array
7113 * appears to still be in use. Hence
7116 do_md_stop(mddev
, 1, 100);
7117 mddev_unlock(mddev
);
7120 * certain more exotic SCSI devices are known to be
7121 * volatile wrt too early system reboots. While the
7122 * right place to handle this issue is the given
7123 * driver, we do want to have a safe RAID driver ...
7130 static struct notifier_block md_notifier
= {
7131 .notifier_call
= md_notify_reboot
,
7133 .priority
= INT_MAX
, /* before any real devices */
7136 static void md_geninit(void)
7138 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
7140 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
7143 static int __init
md_init(void)
7145 if (register_blkdev(MD_MAJOR
, "md"))
7147 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
7148 unregister_blkdev(MD_MAJOR
, "md");
7151 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7152 md_probe
, NULL
, NULL
);
7153 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7154 md_probe
, NULL
, NULL
);
7156 register_reboot_notifier(&md_notifier
);
7157 raid_table_header
= register_sysctl_table(raid_root_table
);
7167 * Searches all registered partitions for autorun RAID arrays
7171 static LIST_HEAD(all_detected_devices
);
7172 struct detected_devices_node
{
7173 struct list_head list
;
7177 void md_autodetect_dev(dev_t dev
)
7179 struct detected_devices_node
*node_detected_dev
;
7181 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7182 if (node_detected_dev
) {
7183 node_detected_dev
->dev
= dev
;
7184 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7186 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7187 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7192 static void autostart_arrays(int part
)
7195 struct detected_devices_node
*node_detected_dev
;
7197 int i_scanned
, i_passed
;
7202 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7204 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7206 node_detected_dev
= list_entry(all_detected_devices
.next
,
7207 struct detected_devices_node
, list
);
7208 list_del(&node_detected_dev
->list
);
7209 dev
= node_detected_dev
->dev
;
7210 kfree(node_detected_dev
);
7211 rdev
= md_import_device(dev
,0, 90);
7215 if (test_bit(Faulty
, &rdev
->flags
)) {
7219 set_bit(AutoDetected
, &rdev
->flags
);
7220 list_add(&rdev
->same_set
, &pending_raid_disks
);
7224 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7225 i_scanned
, i_passed
);
7227 autorun_devices(part
);
7230 #endif /* !MODULE */
7232 static __exit
void md_exit(void)
7235 struct list_head
*tmp
;
7237 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7238 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7240 unregister_blkdev(MD_MAJOR
,"md");
7241 unregister_blkdev(mdp_major
, "mdp");
7242 unregister_reboot_notifier(&md_notifier
);
7243 unregister_sysctl_table(raid_table_header
);
7244 remove_proc_entry("mdstat", NULL
);
7245 for_each_mddev(mddev
, tmp
) {
7246 export_array(mddev
);
7247 mddev
->hold_active
= 0;
7251 subsys_initcall(md_init
);
7252 module_exit(md_exit
)
7254 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7256 return sprintf(buffer
, "%d", start_readonly
);
7258 static int set_ro(const char *val
, struct kernel_param
*kp
)
7261 int num
= simple_strtoul(val
, &e
, 10);
7262 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7263 start_readonly
= num
;
7269 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7270 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7272 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7274 EXPORT_SYMBOL(register_md_personality
);
7275 EXPORT_SYMBOL(unregister_md_personality
);
7276 EXPORT_SYMBOL(md_error
);
7277 EXPORT_SYMBOL(md_done_sync
);
7278 EXPORT_SYMBOL(md_write_start
);
7279 EXPORT_SYMBOL(md_write_end
);
7280 EXPORT_SYMBOL(md_register_thread
);
7281 EXPORT_SYMBOL(md_unregister_thread
);
7282 EXPORT_SYMBOL(md_wakeup_thread
);
7283 EXPORT_SYMBOL(md_check_recovery
);
7284 MODULE_LICENSE("GPL");
7285 MODULE_DESCRIPTION("MD RAID framework");
7287 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);