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>
52 #include <linux/slab.h>
57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part
);
64 static LIST_HEAD(pers_list
);
65 static DEFINE_SPINLOCK(pers_lock
);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
74 * Default number of read corrections we'll attempt on an rdev
75 * before ejecting it from the array. We divide the read error
76 * count by 2 for every hour elapsed between read errors.
78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min
= 1000;
93 static int sysctl_speed_limit_max
= 200000;
94 static inline int speed_min(mddev_t
*mddev
)
96 return mddev
->sync_speed_min
?
97 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
100 static inline int speed_max(mddev_t
*mddev
)
102 return mddev
->sync_speed_max
?
103 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
106 static struct ctl_table_header
*raid_table_header
;
108 static ctl_table raid_table
[] = {
110 .procname
= "speed_limit_min",
111 .data
= &sysctl_speed_limit_min
,
112 .maxlen
= sizeof(int),
113 .mode
= S_IRUGO
|S_IWUSR
,
114 .proc_handler
= proc_dointvec
,
117 .procname
= "speed_limit_max",
118 .data
= &sysctl_speed_limit_max
,
119 .maxlen
= sizeof(int),
120 .mode
= S_IRUGO
|S_IWUSR
,
121 .proc_handler
= proc_dointvec
,
126 static ctl_table raid_dir_table
[] = {
130 .mode
= S_IRUGO
|S_IXUGO
,
136 static ctl_table raid_root_table
[] = {
141 .child
= raid_dir_table
,
146 static const struct block_device_operations md_fops
;
148 static int start_readonly
;
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
161 static atomic_t md_event_count
;
162 void md_new_event(mddev_t
*mddev
)
164 atomic_inc(&md_event_count
);
165 wake_up(&md_event_waiters
);
167 EXPORT_SYMBOL_GPL(md_new_event
);
169 /* Alternate version that can be called from interrupts
170 * when calling sysfs_notify isn't needed.
172 static void md_new_event_inintr(mddev_t
*mddev
)
174 atomic_inc(&md_event_count
);
175 wake_up(&md_event_waiters
);
179 * Enables to iterate over all existing md arrays
180 * all_mddevs_lock protects this list.
182 static LIST_HEAD(all_mddevs
);
183 static DEFINE_SPINLOCK(all_mddevs_lock
);
187 * iterates through all used mddevs in the system.
188 * We take care to grab the all_mddevs_lock whenever navigating
189 * the list, and to always hold a refcount when unlocked.
190 * Any code which breaks out of this loop while own
191 * a reference to the current mddev and must mddev_put it.
193 #define for_each_mddev(mddev,tmp) \
195 for (({ spin_lock(&all_mddevs_lock); \
196 tmp = all_mddevs.next; \
198 ({ if (tmp != &all_mddevs) \
199 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
200 spin_unlock(&all_mddevs_lock); \
201 if (mddev) mddev_put(mddev); \
202 mddev = list_entry(tmp, mddev_t, all_mddevs); \
203 tmp != &all_mddevs;}); \
204 ({ spin_lock(&all_mddevs_lock); \
209 /* Rather than calling directly into the personality make_request function,
210 * IO requests come here first so that we can check if the device is
211 * being suspended pending a reconfiguration.
212 * We hold a refcount over the call to ->make_request. By the time that
213 * call has finished, the bio has been linked into some internal structure
214 * and so is visible to ->quiesce(), so we don't need the refcount any more.
216 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
218 mddev_t
*mddev
= q
->queuedata
;
220 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
225 if (mddev
->suspended
|| mddev
->barrier
) {
228 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
229 TASK_UNINTERRUPTIBLE
);
230 if (!mddev
->suspended
&& !mddev
->barrier
)
236 finish_wait(&mddev
->sb_wait
, &__wait
);
238 atomic_inc(&mddev
->active_io
);
240 rv
= mddev
->pers
->make_request(q
, bio
);
241 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
242 wake_up(&mddev
->sb_wait
);
247 static void mddev_suspend(mddev_t
*mddev
)
249 BUG_ON(mddev
->suspended
);
250 mddev
->suspended
= 1;
252 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
253 mddev
->pers
->quiesce(mddev
, 1);
254 md_unregister_thread(mddev
->thread
);
255 mddev
->thread
= NULL
;
256 /* we now know that no code is executing in the personality module,
257 * except possibly the tail end of a ->bi_end_io function, but that
258 * is certain to complete before the module has a chance to get
263 static void mddev_resume(mddev_t
*mddev
)
265 mddev
->suspended
= 0;
266 wake_up(&mddev
->sb_wait
);
267 mddev
->pers
->quiesce(mddev
, 0);
270 int mddev_congested(mddev_t
*mddev
, int bits
)
274 return mddev
->suspended
;
276 EXPORT_SYMBOL(mddev_congested
);
279 * Generic barrier handling for md
282 #define POST_REQUEST_BARRIER ((void*)1)
284 static void md_end_barrier(struct bio
*bio
, int err
)
286 mdk_rdev_t
*rdev
= bio
->bi_private
;
287 mddev_t
*mddev
= rdev
->mddev
;
288 if (err
== -EOPNOTSUPP
&& mddev
->barrier
!= POST_REQUEST_BARRIER
)
289 set_bit(BIO_EOPNOTSUPP
, &mddev
->barrier
->bi_flags
);
291 rdev_dec_pending(rdev
, mddev
);
293 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
294 if (mddev
->barrier
== POST_REQUEST_BARRIER
) {
295 /* This was a post-request barrier */
296 mddev
->barrier
= NULL
;
297 wake_up(&mddev
->sb_wait
);
299 /* The pre-request barrier has finished */
300 schedule_work(&mddev
->barrier_work
);
305 static void submit_barriers(mddev_t
*mddev
)
310 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
311 if (rdev
->raid_disk
>= 0 &&
312 !test_bit(Faulty
, &rdev
->flags
)) {
313 /* Take two references, one is dropped
314 * when request finishes, one after
315 * we reclaim rcu_read_lock
318 atomic_inc(&rdev
->nr_pending
);
319 atomic_inc(&rdev
->nr_pending
);
321 bi
= bio_alloc(GFP_KERNEL
, 0);
322 bi
->bi_end_io
= md_end_barrier
;
323 bi
->bi_private
= rdev
;
324 bi
->bi_bdev
= rdev
->bdev
;
325 atomic_inc(&mddev
->flush_pending
);
326 submit_bio(WRITE_BARRIER
, bi
);
328 rdev_dec_pending(rdev
, mddev
);
333 static void md_submit_barrier(struct work_struct
*ws
)
335 mddev_t
*mddev
= container_of(ws
, mddev_t
, barrier_work
);
336 struct bio
*bio
= mddev
->barrier
;
338 atomic_set(&mddev
->flush_pending
, 1);
340 if (test_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
))
341 bio_endio(bio
, -EOPNOTSUPP
);
342 else if (bio
->bi_size
== 0)
343 /* an empty barrier - all done */
346 bio
->bi_rw
&= ~(1<<BIO_RW_BARRIER
);
347 if (mddev
->pers
->make_request(mddev
->queue
, bio
))
348 generic_make_request(bio
);
349 mddev
->barrier
= POST_REQUEST_BARRIER
;
350 submit_barriers(mddev
);
352 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
353 mddev
->barrier
= NULL
;
354 wake_up(&mddev
->sb_wait
);
358 void md_barrier_request(mddev_t
*mddev
, struct bio
*bio
)
360 spin_lock_irq(&mddev
->write_lock
);
361 wait_event_lock_irq(mddev
->sb_wait
,
363 mddev
->write_lock
, /*nothing*/);
364 mddev
->barrier
= bio
;
365 spin_unlock_irq(&mddev
->write_lock
);
367 atomic_set(&mddev
->flush_pending
, 1);
368 INIT_WORK(&mddev
->barrier_work
, md_submit_barrier
);
370 submit_barriers(mddev
);
372 if (atomic_dec_and_test(&mddev
->flush_pending
))
373 schedule_work(&mddev
->barrier_work
);
375 EXPORT_SYMBOL(md_barrier_request
);
377 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
379 atomic_inc(&mddev
->active
);
383 static void mddev_delayed_delete(struct work_struct
*ws
);
385 static void mddev_put(mddev_t
*mddev
)
387 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
389 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
390 mddev
->ctime
== 0 && !mddev
->hold_active
) {
391 /* Array is not configured at all, and not held active,
393 list_del(&mddev
->all_mddevs
);
394 if (mddev
->gendisk
) {
395 /* we did a probe so need to clean up.
396 * Call schedule_work inside the spinlock
397 * so that flush_scheduled_work() after
398 * mddev_find will succeed in waiting for the
401 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
402 schedule_work(&mddev
->del_work
);
406 spin_unlock(&all_mddevs_lock
);
409 static mddev_t
* mddev_find(dev_t unit
)
411 mddev_t
*mddev
, *new = NULL
;
414 spin_lock(&all_mddevs_lock
);
417 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
418 if (mddev
->unit
== unit
) {
420 spin_unlock(&all_mddevs_lock
);
426 list_add(&new->all_mddevs
, &all_mddevs
);
427 spin_unlock(&all_mddevs_lock
);
428 new->hold_active
= UNTIL_IOCTL
;
432 /* find an unused unit number */
433 static int next_minor
= 512;
434 int start
= next_minor
;
438 dev
= MKDEV(MD_MAJOR
, next_minor
);
440 if (next_minor
> MINORMASK
)
442 if (next_minor
== start
) {
443 /* Oh dear, all in use. */
444 spin_unlock(&all_mddevs_lock
);
450 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
451 if (mddev
->unit
== dev
) {
457 new->md_minor
= MINOR(dev
);
458 new->hold_active
= UNTIL_STOP
;
459 list_add(&new->all_mddevs
, &all_mddevs
);
460 spin_unlock(&all_mddevs_lock
);
463 spin_unlock(&all_mddevs_lock
);
465 new = kzalloc(sizeof(*new), GFP_KERNEL
);
470 if (MAJOR(unit
) == MD_MAJOR
)
471 new->md_minor
= MINOR(unit
);
473 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
475 mutex_init(&new->open_mutex
);
476 mutex_init(&new->reconfig_mutex
);
477 mutex_init(&new->bitmap_info
.mutex
);
478 INIT_LIST_HEAD(&new->disks
);
479 INIT_LIST_HEAD(&new->all_mddevs
);
480 init_timer(&new->safemode_timer
);
481 atomic_set(&new->active
, 1);
482 atomic_set(&new->openers
, 0);
483 atomic_set(&new->active_io
, 0);
484 spin_lock_init(&new->write_lock
);
485 atomic_set(&new->flush_pending
, 0);
486 init_waitqueue_head(&new->sb_wait
);
487 init_waitqueue_head(&new->recovery_wait
);
488 new->reshape_position
= MaxSector
;
490 new->resync_max
= MaxSector
;
491 new->level
= LEVEL_NONE
;
496 static inline int mddev_lock(mddev_t
* mddev
)
498 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
501 static inline int mddev_is_locked(mddev_t
*mddev
)
503 return mutex_is_locked(&mddev
->reconfig_mutex
);
506 static inline int mddev_trylock(mddev_t
* mddev
)
508 return mutex_trylock(&mddev
->reconfig_mutex
);
511 static struct attribute_group md_redundancy_group
;
513 static void mddev_unlock(mddev_t
* mddev
)
515 if (mddev
->to_remove
) {
516 /* These cannot be removed under reconfig_mutex as
517 * an access to the files will try to take reconfig_mutex
518 * while holding the file unremovable, which leads to
520 * So hold open_mutex instead - we are allowed to take
521 * it while holding reconfig_mutex, and md_run can
522 * use it to wait for the remove to complete.
524 struct attribute_group
*to_remove
= mddev
->to_remove
;
525 mddev
->to_remove
= NULL
;
526 mutex_lock(&mddev
->open_mutex
);
527 mutex_unlock(&mddev
->reconfig_mutex
);
529 if (to_remove
!= &md_redundancy_group
)
530 sysfs_remove_group(&mddev
->kobj
, to_remove
);
531 if (mddev
->pers
== NULL
||
532 mddev
->pers
->sync_request
== NULL
) {
533 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
534 if (mddev
->sysfs_action
)
535 sysfs_put(mddev
->sysfs_action
);
536 mddev
->sysfs_action
= NULL
;
538 mutex_unlock(&mddev
->open_mutex
);
540 mutex_unlock(&mddev
->reconfig_mutex
);
542 md_wakeup_thread(mddev
->thread
);
545 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
549 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
550 if (rdev
->desc_nr
== nr
)
556 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
560 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
561 if (rdev
->bdev
->bd_dev
== dev
)
567 static struct mdk_personality
*find_pers(int level
, char *clevel
)
569 struct mdk_personality
*pers
;
570 list_for_each_entry(pers
, &pers_list
, list
) {
571 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
573 if (strcmp(pers
->name
, clevel
)==0)
579 /* return the offset of the super block in 512byte sectors */
580 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
582 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
583 return MD_NEW_SIZE_SECTORS(num_sectors
);
586 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
591 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
592 if (!rdev
->sb_page
) {
593 printk(KERN_ALERT
"md: out of memory.\n");
600 static void free_disk_sb(mdk_rdev_t
* rdev
)
603 put_page(rdev
->sb_page
);
605 rdev
->sb_page
= NULL
;
612 static void super_written(struct bio
*bio
, int error
)
614 mdk_rdev_t
*rdev
= bio
->bi_private
;
615 mddev_t
*mddev
= rdev
->mddev
;
617 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
618 printk("md: super_written gets error=%d, uptodate=%d\n",
619 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
620 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
621 md_error(mddev
, rdev
);
624 if (atomic_dec_and_test(&mddev
->pending_writes
))
625 wake_up(&mddev
->sb_wait
);
629 static void super_written_barrier(struct bio
*bio
, int error
)
631 struct bio
*bio2
= bio
->bi_private
;
632 mdk_rdev_t
*rdev
= bio2
->bi_private
;
633 mddev_t
*mddev
= rdev
->mddev
;
635 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
636 error
== -EOPNOTSUPP
) {
638 /* barriers don't appear to be supported :-( */
639 set_bit(BarriersNotsupp
, &rdev
->flags
);
640 mddev
->barriers_work
= 0;
641 spin_lock_irqsave(&mddev
->write_lock
, flags
);
642 bio2
->bi_next
= mddev
->biolist
;
643 mddev
->biolist
= bio2
;
644 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
645 wake_up(&mddev
->sb_wait
);
649 bio
->bi_private
= rdev
;
650 super_written(bio
, error
);
654 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
655 sector_t sector
, int size
, struct page
*page
)
657 /* write first size bytes of page to sector of rdev
658 * Increment mddev->pending_writes before returning
659 * and decrement it on completion, waking up sb_wait
660 * if zero is reached.
661 * If an error occurred, call md_error
663 * As we might need to resubmit the request if BIO_RW_BARRIER
664 * causes ENOTSUPP, we allocate a spare bio...
666 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
667 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
669 bio
->bi_bdev
= rdev
->bdev
;
670 bio
->bi_sector
= sector
;
671 bio_add_page(bio
, page
, size
, 0);
672 bio
->bi_private
= rdev
;
673 bio
->bi_end_io
= super_written
;
676 atomic_inc(&mddev
->pending_writes
);
677 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
679 rw
|= (1<<BIO_RW_BARRIER
);
680 rbio
= bio_clone(bio
, GFP_NOIO
);
681 rbio
->bi_private
= bio
;
682 rbio
->bi_end_io
= super_written_barrier
;
683 submit_bio(rw
, rbio
);
688 void md_super_wait(mddev_t
*mddev
)
690 /* wait for all superblock writes that were scheduled to complete.
691 * if any had to be retried (due to BARRIER problems), retry them
695 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
696 if (atomic_read(&mddev
->pending_writes
)==0)
698 while (mddev
->biolist
) {
700 spin_lock_irq(&mddev
->write_lock
);
701 bio
= mddev
->biolist
;
702 mddev
->biolist
= bio
->bi_next
;
704 spin_unlock_irq(&mddev
->write_lock
);
705 submit_bio(bio
->bi_rw
, bio
);
709 finish_wait(&mddev
->sb_wait
, &wq
);
712 static void bi_complete(struct bio
*bio
, int error
)
714 complete((struct completion
*)bio
->bi_private
);
717 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
718 struct page
*page
, int rw
)
720 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
721 struct completion event
;
724 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
727 bio
->bi_sector
= sector
;
728 bio_add_page(bio
, page
, size
, 0);
729 init_completion(&event
);
730 bio
->bi_private
= &event
;
731 bio
->bi_end_io
= bi_complete
;
733 wait_for_completion(&event
);
735 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
739 EXPORT_SYMBOL_GPL(sync_page_io
);
741 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
743 char b
[BDEVNAME_SIZE
];
744 if (!rdev
->sb_page
) {
752 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
758 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
759 bdevname(rdev
->bdev
,b
));
763 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
765 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
766 sb1
->set_uuid1
== sb2
->set_uuid1
&&
767 sb1
->set_uuid2
== sb2
->set_uuid2
&&
768 sb1
->set_uuid3
== sb2
->set_uuid3
;
771 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
774 mdp_super_t
*tmp1
, *tmp2
;
776 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
777 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
779 if (!tmp1
|| !tmp2
) {
781 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
789 * nr_disks is not constant
794 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
802 static u32
md_csum_fold(u32 csum
)
804 csum
= (csum
& 0xffff) + (csum
>> 16);
805 return (csum
& 0xffff) + (csum
>> 16);
808 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
811 u32
*sb32
= (u32
*)sb
;
813 unsigned int disk_csum
, csum
;
815 disk_csum
= sb
->sb_csum
;
818 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
820 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
824 /* This used to use csum_partial, which was wrong for several
825 * reasons including that different results are returned on
826 * different architectures. It isn't critical that we get exactly
827 * the same return value as before (we always csum_fold before
828 * testing, and that removes any differences). However as we
829 * know that csum_partial always returned a 16bit value on
830 * alphas, do a fold to maximise conformity to previous behaviour.
832 sb
->sb_csum
= md_csum_fold(disk_csum
);
834 sb
->sb_csum
= disk_csum
;
841 * Handle superblock details.
842 * We want to be able to handle multiple superblock formats
843 * so we have a common interface to them all, and an array of
844 * different handlers.
845 * We rely on user-space to write the initial superblock, and support
846 * reading and updating of superblocks.
847 * Interface methods are:
848 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
849 * loads and validates a superblock on dev.
850 * if refdev != NULL, compare superblocks on both devices
852 * 0 - dev has a superblock that is compatible with refdev
853 * 1 - dev has a superblock that is compatible and newer than refdev
854 * so dev should be used as the refdev in future
855 * -EINVAL superblock incompatible or invalid
856 * -othererror e.g. -EIO
858 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
859 * Verify that dev is acceptable into mddev.
860 * The first time, mddev->raid_disks will be 0, and data from
861 * dev should be merged in. Subsequent calls check that dev
862 * is new enough. Return 0 or -EINVAL
864 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
865 * Update the superblock for rdev with data in mddev
866 * This does not write to disc.
872 struct module
*owner
;
873 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
875 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
876 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
877 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
878 sector_t num_sectors
);
882 * Check that the given mddev has no bitmap.
884 * This function is called from the run method of all personalities that do not
885 * support bitmaps. It prints an error message and returns non-zero if mddev
886 * has a bitmap. Otherwise, it returns 0.
889 int md_check_no_bitmap(mddev_t
*mddev
)
891 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
893 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
894 mdname(mddev
), mddev
->pers
->name
);
897 EXPORT_SYMBOL(md_check_no_bitmap
);
900 * load_super for 0.90.0
902 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
904 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
909 * Calculate the position of the superblock (512byte sectors),
910 * it's at the end of the disk.
912 * It also happens to be a multiple of 4Kb.
914 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
916 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
921 bdevname(rdev
->bdev
, b
);
922 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
924 if (sb
->md_magic
!= MD_SB_MAGIC
) {
925 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
930 if (sb
->major_version
!= 0 ||
931 sb
->minor_version
< 90 ||
932 sb
->minor_version
> 91) {
933 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
934 sb
->major_version
, sb
->minor_version
,
939 if (sb
->raid_disks
<= 0)
942 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
943 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
948 rdev
->preferred_minor
= sb
->md_minor
;
949 rdev
->data_offset
= 0;
950 rdev
->sb_size
= MD_SB_BYTES
;
952 if (sb
->level
== LEVEL_MULTIPATH
)
955 rdev
->desc_nr
= sb
->this_disk
.number
;
961 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
962 if (!uuid_equal(refsb
, sb
)) {
963 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
964 b
, bdevname(refdev
->bdev
,b2
));
967 if (!sb_equal(refsb
, sb
)) {
968 printk(KERN_WARNING
"md: %s has same UUID"
969 " but different superblock to %s\n",
970 b
, bdevname(refdev
->bdev
, b2
));
974 ev2
= md_event(refsb
);
980 rdev
->sectors
= rdev
->sb_start
;
982 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
983 /* "this cannot possibly happen" ... */
991 * validate_super for 0.90.0
993 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
996 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
997 __u64 ev1
= md_event(sb
);
999 rdev
->raid_disk
= -1;
1000 clear_bit(Faulty
, &rdev
->flags
);
1001 clear_bit(In_sync
, &rdev
->flags
);
1002 clear_bit(WriteMostly
, &rdev
->flags
);
1003 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1005 if (mddev
->raid_disks
== 0) {
1006 mddev
->major_version
= 0;
1007 mddev
->minor_version
= sb
->minor_version
;
1008 mddev
->patch_version
= sb
->patch_version
;
1009 mddev
->external
= 0;
1010 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1011 mddev
->ctime
= sb
->ctime
;
1012 mddev
->utime
= sb
->utime
;
1013 mddev
->level
= sb
->level
;
1014 mddev
->clevel
[0] = 0;
1015 mddev
->layout
= sb
->layout
;
1016 mddev
->raid_disks
= sb
->raid_disks
;
1017 mddev
->dev_sectors
= sb
->size
* 2;
1018 mddev
->events
= ev1
;
1019 mddev
->bitmap_info
.offset
= 0;
1020 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1022 if (mddev
->minor_version
>= 91) {
1023 mddev
->reshape_position
= sb
->reshape_position
;
1024 mddev
->delta_disks
= sb
->delta_disks
;
1025 mddev
->new_level
= sb
->new_level
;
1026 mddev
->new_layout
= sb
->new_layout
;
1027 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1029 mddev
->reshape_position
= MaxSector
;
1030 mddev
->delta_disks
= 0;
1031 mddev
->new_level
= mddev
->level
;
1032 mddev
->new_layout
= mddev
->layout
;
1033 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1036 if (sb
->state
& (1<<MD_SB_CLEAN
))
1037 mddev
->recovery_cp
= MaxSector
;
1039 if (sb
->events_hi
== sb
->cp_events_hi
&&
1040 sb
->events_lo
== sb
->cp_events_lo
) {
1041 mddev
->recovery_cp
= sb
->recovery_cp
;
1043 mddev
->recovery_cp
= 0;
1046 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1047 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1048 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1049 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1051 mddev
->max_disks
= MD_SB_DISKS
;
1053 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1054 mddev
->bitmap_info
.file
== NULL
)
1055 mddev
->bitmap_info
.offset
=
1056 mddev
->bitmap_info
.default_offset
;
1058 } else if (mddev
->pers
== NULL
) {
1059 /* Insist on good event counter while assembling */
1061 if (ev1
< mddev
->events
)
1063 } else if (mddev
->bitmap
) {
1064 /* if adding to array with a bitmap, then we can accept an
1065 * older device ... but not too old.
1067 if (ev1
< mddev
->bitmap
->events_cleared
)
1070 if (ev1
< mddev
->events
)
1071 /* just a hot-add of a new device, leave raid_disk at -1 */
1075 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1076 desc
= sb
->disks
+ rdev
->desc_nr
;
1078 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1079 set_bit(Faulty
, &rdev
->flags
);
1080 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1081 desc->raid_disk < mddev->raid_disks */) {
1082 set_bit(In_sync
, &rdev
->flags
);
1083 rdev
->raid_disk
= desc
->raid_disk
;
1084 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1085 /* active but not in sync implies recovery up to
1086 * reshape position. We don't know exactly where
1087 * that is, so set to zero for now */
1088 if (mddev
->minor_version
>= 91) {
1089 rdev
->recovery_offset
= 0;
1090 rdev
->raid_disk
= desc
->raid_disk
;
1093 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1094 set_bit(WriteMostly
, &rdev
->flags
);
1095 } else /* MULTIPATH are always insync */
1096 set_bit(In_sync
, &rdev
->flags
);
1101 * sync_super for 0.90.0
1103 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1107 int next_spare
= mddev
->raid_disks
;
1110 /* make rdev->sb match mddev data..
1113 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1114 * 3/ any empty disks < next_spare become removed
1116 * disks[0] gets initialised to REMOVED because
1117 * we cannot be sure from other fields if it has
1118 * been initialised or not.
1121 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1123 rdev
->sb_size
= MD_SB_BYTES
;
1125 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1127 memset(sb
, 0, sizeof(*sb
));
1129 sb
->md_magic
= MD_SB_MAGIC
;
1130 sb
->major_version
= mddev
->major_version
;
1131 sb
->patch_version
= mddev
->patch_version
;
1132 sb
->gvalid_words
= 0; /* ignored */
1133 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1134 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1135 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1136 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1138 sb
->ctime
= mddev
->ctime
;
1139 sb
->level
= mddev
->level
;
1140 sb
->size
= mddev
->dev_sectors
/ 2;
1141 sb
->raid_disks
= mddev
->raid_disks
;
1142 sb
->md_minor
= mddev
->md_minor
;
1143 sb
->not_persistent
= 0;
1144 sb
->utime
= mddev
->utime
;
1146 sb
->events_hi
= (mddev
->events
>>32);
1147 sb
->events_lo
= (u32
)mddev
->events
;
1149 if (mddev
->reshape_position
== MaxSector
)
1150 sb
->minor_version
= 90;
1152 sb
->minor_version
= 91;
1153 sb
->reshape_position
= mddev
->reshape_position
;
1154 sb
->new_level
= mddev
->new_level
;
1155 sb
->delta_disks
= mddev
->delta_disks
;
1156 sb
->new_layout
= mddev
->new_layout
;
1157 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1159 mddev
->minor_version
= sb
->minor_version
;
1162 sb
->recovery_cp
= mddev
->recovery_cp
;
1163 sb
->cp_events_hi
= (mddev
->events
>>32);
1164 sb
->cp_events_lo
= (u32
)mddev
->events
;
1165 if (mddev
->recovery_cp
== MaxSector
)
1166 sb
->state
= (1<< MD_SB_CLEAN
);
1168 sb
->recovery_cp
= 0;
1170 sb
->layout
= mddev
->layout
;
1171 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1173 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1174 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1176 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1177 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1180 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1182 if (rdev2
->raid_disk
>= 0 &&
1183 sb
->minor_version
>= 91)
1184 /* we have nowhere to store the recovery_offset,
1185 * but if it is not below the reshape_position,
1186 * we can piggy-back on that.
1189 if (rdev2
->raid_disk
< 0 ||
1190 test_bit(Faulty
, &rdev2
->flags
))
1193 desc_nr
= rdev2
->raid_disk
;
1195 desc_nr
= next_spare
++;
1196 rdev2
->desc_nr
= desc_nr
;
1197 d
= &sb
->disks
[rdev2
->desc_nr
];
1199 d
->number
= rdev2
->desc_nr
;
1200 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1201 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1203 d
->raid_disk
= rdev2
->raid_disk
;
1205 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1206 if (test_bit(Faulty
, &rdev2
->flags
))
1207 d
->state
= (1<<MD_DISK_FAULTY
);
1208 else if (is_active
) {
1209 d
->state
= (1<<MD_DISK_ACTIVE
);
1210 if (test_bit(In_sync
, &rdev2
->flags
))
1211 d
->state
|= (1<<MD_DISK_SYNC
);
1219 if (test_bit(WriteMostly
, &rdev2
->flags
))
1220 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1222 /* now set the "removed" and "faulty" bits on any missing devices */
1223 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1224 mdp_disk_t
*d
= &sb
->disks
[i
];
1225 if (d
->state
== 0 && d
->number
== 0) {
1228 d
->state
= (1<<MD_DISK_REMOVED
);
1229 d
->state
|= (1<<MD_DISK_FAULTY
);
1233 sb
->nr_disks
= nr_disks
;
1234 sb
->active_disks
= active
;
1235 sb
->working_disks
= working
;
1236 sb
->failed_disks
= failed
;
1237 sb
->spare_disks
= spare
;
1239 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1240 sb
->sb_csum
= calc_sb_csum(sb
);
1244 * rdev_size_change for 0.90.0
1246 static unsigned long long
1247 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1249 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1250 return 0; /* component must fit device */
1251 if (rdev
->mddev
->bitmap_info
.offset
)
1252 return 0; /* can't move bitmap */
1253 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1254 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1255 num_sectors
= rdev
->sb_start
;
1256 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1258 md_super_wait(rdev
->mddev
);
1259 return num_sectors
/ 2; /* kB for sysfs */
1264 * version 1 superblock
1267 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1271 unsigned long long newcsum
;
1272 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1273 __le32
*isuper
= (__le32
*)sb
;
1276 disk_csum
= sb
->sb_csum
;
1279 for (i
=0; size
>=4; size
-= 4 )
1280 newcsum
+= le32_to_cpu(*isuper
++);
1283 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1285 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1286 sb
->sb_csum
= disk_csum
;
1287 return cpu_to_le32(csum
);
1290 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1292 struct mdp_superblock_1
*sb
;
1295 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1299 * Calculate the position of the superblock in 512byte sectors.
1300 * It is always aligned to a 4K boundary and
1301 * depeding on minor_version, it can be:
1302 * 0: At least 8K, but less than 12K, from end of device
1303 * 1: At start of device
1304 * 2: 4K from start of device.
1306 switch(minor_version
) {
1308 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1310 sb_start
&= ~(sector_t
)(4*2-1);
1321 rdev
->sb_start
= sb_start
;
1323 /* superblock is rarely larger than 1K, but it can be larger,
1324 * and it is safe to read 4k, so we do that
1326 ret
= read_disk_sb(rdev
, 4096);
1327 if (ret
) return ret
;
1330 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1332 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1333 sb
->major_version
!= cpu_to_le32(1) ||
1334 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1335 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1336 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1339 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1340 printk("md: invalid superblock checksum on %s\n",
1341 bdevname(rdev
->bdev
,b
));
1344 if (le64_to_cpu(sb
->data_size
) < 10) {
1345 printk("md: data_size too small on %s\n",
1346 bdevname(rdev
->bdev
,b
));
1350 rdev
->preferred_minor
= 0xffff;
1351 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1352 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1354 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1355 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1356 if (rdev
->sb_size
& bmask
)
1357 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1360 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1363 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1366 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1372 struct mdp_superblock_1
*refsb
=
1373 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1375 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1376 sb
->level
!= refsb
->level
||
1377 sb
->layout
!= refsb
->layout
||
1378 sb
->chunksize
!= refsb
->chunksize
) {
1379 printk(KERN_WARNING
"md: %s has strangely different"
1380 " superblock to %s\n",
1381 bdevname(rdev
->bdev
,b
),
1382 bdevname(refdev
->bdev
,b2
));
1385 ev1
= le64_to_cpu(sb
->events
);
1386 ev2
= le64_to_cpu(refsb
->events
);
1394 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1395 le64_to_cpu(sb
->data_offset
);
1397 rdev
->sectors
= rdev
->sb_start
;
1398 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1400 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1401 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1406 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1408 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1409 __u64 ev1
= le64_to_cpu(sb
->events
);
1411 rdev
->raid_disk
= -1;
1412 clear_bit(Faulty
, &rdev
->flags
);
1413 clear_bit(In_sync
, &rdev
->flags
);
1414 clear_bit(WriteMostly
, &rdev
->flags
);
1415 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1417 if (mddev
->raid_disks
== 0) {
1418 mddev
->major_version
= 1;
1419 mddev
->patch_version
= 0;
1420 mddev
->external
= 0;
1421 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1422 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1423 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1424 mddev
->level
= le32_to_cpu(sb
->level
);
1425 mddev
->clevel
[0] = 0;
1426 mddev
->layout
= le32_to_cpu(sb
->layout
);
1427 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1428 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1429 mddev
->events
= ev1
;
1430 mddev
->bitmap_info
.offset
= 0;
1431 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1433 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1434 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1436 mddev
->max_disks
= (4096-256)/2;
1438 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1439 mddev
->bitmap_info
.file
== NULL
)
1440 mddev
->bitmap_info
.offset
=
1441 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1443 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1444 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1445 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1446 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1447 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1448 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1450 mddev
->reshape_position
= MaxSector
;
1451 mddev
->delta_disks
= 0;
1452 mddev
->new_level
= mddev
->level
;
1453 mddev
->new_layout
= mddev
->layout
;
1454 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1457 } else if (mddev
->pers
== NULL
) {
1458 /* Insist of good event counter while assembling */
1460 if (ev1
< mddev
->events
)
1462 } else if (mddev
->bitmap
) {
1463 /* If adding to array with a bitmap, then we can accept an
1464 * older device, but not too old.
1466 if (ev1
< mddev
->bitmap
->events_cleared
)
1469 if (ev1
< mddev
->events
)
1470 /* just a hot-add of a new device, leave raid_disk at -1 */
1473 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1475 if (rdev
->desc_nr
< 0 ||
1476 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1480 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1482 case 0xffff: /* spare */
1484 case 0xfffe: /* faulty */
1485 set_bit(Faulty
, &rdev
->flags
);
1488 if ((le32_to_cpu(sb
->feature_map
) &
1489 MD_FEATURE_RECOVERY_OFFSET
))
1490 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1492 set_bit(In_sync
, &rdev
->flags
);
1493 rdev
->raid_disk
= role
;
1496 if (sb
->devflags
& WriteMostly1
)
1497 set_bit(WriteMostly
, &rdev
->flags
);
1498 } else /* MULTIPATH are always insync */
1499 set_bit(In_sync
, &rdev
->flags
);
1504 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1506 struct mdp_superblock_1
*sb
;
1509 /* make rdev->sb match mddev and rdev data. */
1511 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1513 sb
->feature_map
= 0;
1515 sb
->recovery_offset
= cpu_to_le64(0);
1516 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1517 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1518 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1520 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1521 sb
->events
= cpu_to_le64(mddev
->events
);
1523 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1525 sb
->resync_offset
= cpu_to_le64(0);
1527 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1529 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1530 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1531 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1532 sb
->level
= cpu_to_le32(mddev
->level
);
1533 sb
->layout
= cpu_to_le32(mddev
->layout
);
1535 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1536 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1537 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1540 if (rdev
->raid_disk
>= 0 &&
1541 !test_bit(In_sync
, &rdev
->flags
)) {
1543 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1544 sb
->recovery_offset
=
1545 cpu_to_le64(rdev
->recovery_offset
);
1548 if (mddev
->reshape_position
!= MaxSector
) {
1549 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1550 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1551 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1552 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1553 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1554 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1558 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1559 if (rdev2
->desc_nr
+1 > max_dev
)
1560 max_dev
= rdev2
->desc_nr
+1;
1562 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1564 sb
->max_dev
= cpu_to_le32(max_dev
);
1565 rdev
->sb_size
= max_dev
* 2 + 256;
1566 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1567 if (rdev
->sb_size
& bmask
)
1568 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1570 for (i
=0; i
<max_dev
;i
++)
1571 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1573 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1575 if (test_bit(Faulty
, &rdev2
->flags
))
1576 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1577 else if (test_bit(In_sync
, &rdev2
->flags
))
1578 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1579 else if (rdev2
->raid_disk
>= 0)
1580 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1582 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1585 sb
->sb_csum
= calc_sb_1_csum(sb
);
1588 static unsigned long long
1589 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1591 struct mdp_superblock_1
*sb
;
1592 sector_t max_sectors
;
1593 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1594 return 0; /* component must fit device */
1595 if (rdev
->sb_start
< rdev
->data_offset
) {
1596 /* minor versions 1 and 2; superblock before data */
1597 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1598 max_sectors
-= rdev
->data_offset
;
1599 if (!num_sectors
|| num_sectors
> max_sectors
)
1600 num_sectors
= max_sectors
;
1601 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1602 /* minor version 0 with bitmap we can't move */
1605 /* minor version 0; superblock after data */
1607 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1608 sb_start
&= ~(sector_t
)(4*2 - 1);
1609 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1610 if (!num_sectors
|| num_sectors
> max_sectors
)
1611 num_sectors
= max_sectors
;
1612 rdev
->sb_start
= sb_start
;
1614 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1615 sb
->data_size
= cpu_to_le64(num_sectors
);
1616 sb
->super_offset
= rdev
->sb_start
;
1617 sb
->sb_csum
= calc_sb_1_csum(sb
);
1618 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1620 md_super_wait(rdev
->mddev
);
1621 return num_sectors
/ 2; /* kB for sysfs */
1624 static struct super_type super_types
[] = {
1627 .owner
= THIS_MODULE
,
1628 .load_super
= super_90_load
,
1629 .validate_super
= super_90_validate
,
1630 .sync_super
= super_90_sync
,
1631 .rdev_size_change
= super_90_rdev_size_change
,
1635 .owner
= THIS_MODULE
,
1636 .load_super
= super_1_load
,
1637 .validate_super
= super_1_validate
,
1638 .sync_super
= super_1_sync
,
1639 .rdev_size_change
= super_1_rdev_size_change
,
1643 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1645 mdk_rdev_t
*rdev
, *rdev2
;
1648 rdev_for_each_rcu(rdev
, mddev1
)
1649 rdev_for_each_rcu(rdev2
, mddev2
)
1650 if (rdev
->bdev
->bd_contains
==
1651 rdev2
->bdev
->bd_contains
) {
1659 static LIST_HEAD(pending_raid_disks
);
1662 * Try to register data integrity profile for an mddev
1664 * This is called when an array is started and after a disk has been kicked
1665 * from the array. It only succeeds if all working and active component devices
1666 * are integrity capable with matching profiles.
1668 int md_integrity_register(mddev_t
*mddev
)
1670 mdk_rdev_t
*rdev
, *reference
= NULL
;
1672 if (list_empty(&mddev
->disks
))
1673 return 0; /* nothing to do */
1674 if (blk_get_integrity(mddev
->gendisk
))
1675 return 0; /* already registered */
1676 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1677 /* skip spares and non-functional disks */
1678 if (test_bit(Faulty
, &rdev
->flags
))
1680 if (rdev
->raid_disk
< 0)
1683 * If at least one rdev is not integrity capable, we can not
1684 * enable data integrity for the md device.
1686 if (!bdev_get_integrity(rdev
->bdev
))
1689 /* Use the first rdev as the reference */
1693 /* does this rdev's profile match the reference profile? */
1694 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1695 rdev
->bdev
->bd_disk
) < 0)
1699 * All component devices are integrity capable and have matching
1700 * profiles, register the common profile for the md device.
1702 if (blk_integrity_register(mddev
->gendisk
,
1703 bdev_get_integrity(reference
->bdev
)) != 0) {
1704 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1708 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1712 EXPORT_SYMBOL(md_integrity_register
);
1714 /* Disable data integrity if non-capable/non-matching disk is being added */
1715 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1717 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1718 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1720 if (!bi_mddev
) /* nothing to do */
1722 if (rdev
->raid_disk
< 0) /* skip spares */
1724 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1725 rdev
->bdev
->bd_disk
) >= 0)
1727 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1728 blk_integrity_unregister(mddev
->gendisk
);
1730 EXPORT_SYMBOL(md_integrity_add_rdev
);
1732 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1734 char b
[BDEVNAME_SIZE
];
1744 /* prevent duplicates */
1745 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1748 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1749 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1750 rdev
->sectors
< mddev
->dev_sectors
)) {
1752 /* Cannot change size, so fail
1753 * If mddev->level <= 0, then we don't care
1754 * about aligning sizes (e.g. linear)
1756 if (mddev
->level
> 0)
1759 mddev
->dev_sectors
= rdev
->sectors
;
1762 /* Verify rdev->desc_nr is unique.
1763 * If it is -1, assign a free number, else
1764 * check number is not in use
1766 if (rdev
->desc_nr
< 0) {
1768 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1769 while (find_rdev_nr(mddev
, choice
))
1771 rdev
->desc_nr
= choice
;
1773 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1776 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1777 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1778 mdname(mddev
), mddev
->max_disks
);
1781 bdevname(rdev
->bdev
,b
);
1782 while ( (s
=strchr(b
, '/')) != NULL
)
1785 rdev
->mddev
= mddev
;
1786 printk(KERN_INFO
"md: bind<%s>\n", b
);
1788 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1791 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1792 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1793 kobject_del(&rdev
->kobj
);
1796 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1798 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1799 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1801 /* May as well allow recovery to be retried once */
1802 mddev
->recovery_disabled
= 0;
1807 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1812 static void md_delayed_delete(struct work_struct
*ws
)
1814 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1815 kobject_del(&rdev
->kobj
);
1816 kobject_put(&rdev
->kobj
);
1819 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1821 char b
[BDEVNAME_SIZE
];
1826 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1827 list_del_rcu(&rdev
->same_set
);
1828 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1830 sysfs_remove_link(&rdev
->kobj
, "block");
1831 sysfs_put(rdev
->sysfs_state
);
1832 rdev
->sysfs_state
= NULL
;
1833 /* We need to delay this, otherwise we can deadlock when
1834 * writing to 'remove' to "dev/state". We also need
1835 * to delay it due to rcu usage.
1838 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1839 kobject_get(&rdev
->kobj
);
1840 schedule_work(&rdev
->del_work
);
1844 * prevent the device from being mounted, repartitioned or
1845 * otherwise reused by a RAID array (or any other kernel
1846 * subsystem), by bd_claiming the device.
1848 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1851 struct block_device
*bdev
;
1852 char b
[BDEVNAME_SIZE
];
1854 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1856 printk(KERN_ERR
"md: could not open %s.\n",
1857 __bdevname(dev
, b
));
1858 return PTR_ERR(bdev
);
1860 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1862 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1864 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1868 set_bit(AllReserved
, &rdev
->flags
);
1873 static void unlock_rdev(mdk_rdev_t
*rdev
)
1875 struct block_device
*bdev
= rdev
->bdev
;
1880 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1883 void md_autodetect_dev(dev_t dev
);
1885 static void export_rdev(mdk_rdev_t
* rdev
)
1887 char b
[BDEVNAME_SIZE
];
1888 printk(KERN_INFO
"md: export_rdev(%s)\n",
1889 bdevname(rdev
->bdev
,b
));
1894 if (test_bit(AutoDetected
, &rdev
->flags
))
1895 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1898 kobject_put(&rdev
->kobj
);
1901 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1903 unbind_rdev_from_array(rdev
);
1907 static void export_array(mddev_t
*mddev
)
1909 mdk_rdev_t
*rdev
, *tmp
;
1911 rdev_for_each(rdev
, tmp
, mddev
) {
1916 kick_rdev_from_array(rdev
);
1918 if (!list_empty(&mddev
->disks
))
1920 mddev
->raid_disks
= 0;
1921 mddev
->major_version
= 0;
1924 static void print_desc(mdp_disk_t
*desc
)
1926 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1927 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1930 static void print_sb_90(mdp_super_t
*sb
)
1935 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1936 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1937 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1939 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1940 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1941 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1942 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1943 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1944 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1945 sb
->failed_disks
, sb
->spare_disks
,
1946 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1949 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1952 desc
= sb
->disks
+ i
;
1953 if (desc
->number
|| desc
->major
|| desc
->minor
||
1954 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1955 printk(" D %2d: ", i
);
1959 printk(KERN_INFO
"md: THIS: ");
1960 print_desc(&sb
->this_disk
);
1963 static void print_sb_1(struct mdp_superblock_1
*sb
)
1967 uuid
= sb
->set_uuid
;
1969 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1970 "md: Name: \"%s\" CT:%llu\n",
1971 le32_to_cpu(sb
->major_version
),
1972 le32_to_cpu(sb
->feature_map
),
1975 (unsigned long long)le64_to_cpu(sb
->ctime
)
1976 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1978 uuid
= sb
->device_uuid
;
1980 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1982 "md: Dev:%08x UUID: %pU\n"
1983 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1984 "md: (MaxDev:%u) \n",
1985 le32_to_cpu(sb
->level
),
1986 (unsigned long long)le64_to_cpu(sb
->size
),
1987 le32_to_cpu(sb
->raid_disks
),
1988 le32_to_cpu(sb
->layout
),
1989 le32_to_cpu(sb
->chunksize
),
1990 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1991 (unsigned long long)le64_to_cpu(sb
->data_size
),
1992 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1993 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1994 le32_to_cpu(sb
->dev_number
),
1997 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1998 (unsigned long long)le64_to_cpu(sb
->events
),
1999 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2000 le32_to_cpu(sb
->sb_csum
),
2001 le32_to_cpu(sb
->max_dev
)
2005 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2007 char b
[BDEVNAME_SIZE
];
2008 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2009 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2010 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2012 if (rdev
->sb_loaded
) {
2013 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2014 switch (major_version
) {
2016 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
2019 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
2023 printk(KERN_INFO
"md: no rdev superblock!\n");
2026 static void md_print_devices(void)
2028 struct list_head
*tmp
;
2031 char b
[BDEVNAME_SIZE
];
2034 printk("md: **********************************\n");
2035 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2036 printk("md: **********************************\n");
2037 for_each_mddev(mddev
, tmp
) {
2040 bitmap_print_sb(mddev
->bitmap
);
2042 printk("%s: ", mdname(mddev
));
2043 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2044 printk("<%s>", bdevname(rdev
->bdev
,b
));
2047 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2048 print_rdev(rdev
, mddev
->major_version
);
2050 printk("md: **********************************\n");
2055 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2057 /* Update each superblock (in-memory image), but
2058 * if we are allowed to, skip spares which already
2059 * have the right event counter, or have one earlier
2060 * (which would mean they aren't being marked as dirty
2061 * with the rest of the array)
2065 /* First make sure individual recovery_offsets are correct */
2066 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2067 if (rdev
->raid_disk
>= 0 &&
2068 !test_bit(In_sync
, &rdev
->flags
) &&
2069 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2070 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2073 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2074 if (rdev
->sb_events
== mddev
->events
||
2076 rdev
->raid_disk
< 0 &&
2077 (rdev
->sb_events
&1)==0 &&
2078 rdev
->sb_events
+1 == mddev
->events
)) {
2079 /* Don't update this superblock */
2080 rdev
->sb_loaded
= 2;
2082 super_types
[mddev
->major_version
].
2083 sync_super(mddev
, rdev
);
2084 rdev
->sb_loaded
= 1;
2089 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2095 mddev
->utime
= get_seconds();
2096 if (mddev
->external
)
2099 spin_lock_irq(&mddev
->write_lock
);
2101 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2102 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2104 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2105 /* just a clean<-> dirty transition, possibly leave spares alone,
2106 * though if events isn't the right even/odd, we will have to do
2112 if (mddev
->degraded
)
2113 /* If the array is degraded, then skipping spares is both
2114 * dangerous and fairly pointless.
2115 * Dangerous because a device that was removed from the array
2116 * might have a event_count that still looks up-to-date,
2117 * so it can be re-added without a resync.
2118 * Pointless because if there are any spares to skip,
2119 * then a recovery will happen and soon that array won't
2120 * be degraded any more and the spare can go back to sleep then.
2124 sync_req
= mddev
->in_sync
;
2126 /* If this is just a dirty<->clean transition, and the array is clean
2127 * and 'events' is odd, we can roll back to the previous clean state */
2129 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2130 && (mddev
->events
& 1)
2131 && mddev
->events
!= 1)
2134 /* otherwise we have to go forward and ... */
2136 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
2137 /* .. if the array isn't clean, an 'even' event must also go
2139 if ((mddev
->events
&1)==0) {
2141 sync_req
= 2; /* force a second update to get the
2142 * even/odd in sync */
2145 /* otherwise an 'odd' event must go to spares */
2146 if ((mddev
->events
&1)) {
2148 sync_req
= 2; /* force a second update to get the
2149 * even/odd in sync */
2154 if (!mddev
->events
) {
2156 * oops, this 64-bit counter should never wrap.
2157 * Either we are in around ~1 trillion A.C., assuming
2158 * 1 reboot per second, or we have a bug:
2165 * do not write anything to disk if using
2166 * nonpersistent superblocks
2168 if (!mddev
->persistent
) {
2169 if (!mddev
->external
)
2170 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2172 spin_unlock_irq(&mddev
->write_lock
);
2173 wake_up(&mddev
->sb_wait
);
2176 sync_sbs(mddev
, nospares
);
2177 spin_unlock_irq(&mddev
->write_lock
);
2180 "md: updating %s RAID superblock on device (in sync %d)\n",
2181 mdname(mddev
),mddev
->in_sync
);
2183 bitmap_update_sb(mddev
->bitmap
);
2184 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2185 char b
[BDEVNAME_SIZE
];
2186 dprintk(KERN_INFO
"md: ");
2187 if (rdev
->sb_loaded
!= 1)
2188 continue; /* no noise on spare devices */
2189 if (test_bit(Faulty
, &rdev
->flags
))
2190 dprintk("(skipping faulty ");
2192 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2193 if (!test_bit(Faulty
, &rdev
->flags
)) {
2194 md_super_write(mddev
,rdev
,
2195 rdev
->sb_start
, rdev
->sb_size
,
2197 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2198 bdevname(rdev
->bdev
,b
),
2199 (unsigned long long)rdev
->sb_start
);
2200 rdev
->sb_events
= mddev
->events
;
2204 if (mddev
->level
== LEVEL_MULTIPATH
)
2205 /* only need to write one superblock... */
2208 md_super_wait(mddev
);
2209 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2211 spin_lock_irq(&mddev
->write_lock
);
2212 if (mddev
->in_sync
!= sync_req
||
2213 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2214 /* have to write it out again */
2215 spin_unlock_irq(&mddev
->write_lock
);
2218 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2219 spin_unlock_irq(&mddev
->write_lock
);
2220 wake_up(&mddev
->sb_wait
);
2221 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2222 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2226 /* words written to sysfs files may, or may not, be \n terminated.
2227 * We want to accept with case. For this we use cmd_match.
2229 static int cmd_match(const char *cmd
, const char *str
)
2231 /* See if cmd, written into a sysfs file, matches
2232 * str. They must either be the same, or cmd can
2233 * have a trailing newline
2235 while (*cmd
&& *str
&& *cmd
== *str
) {
2246 struct rdev_sysfs_entry
{
2247 struct attribute attr
;
2248 ssize_t (*show
)(mdk_rdev_t
*, char *);
2249 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2253 state_show(mdk_rdev_t
*rdev
, char *page
)
2258 if (test_bit(Faulty
, &rdev
->flags
)) {
2259 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2262 if (test_bit(In_sync
, &rdev
->flags
)) {
2263 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2266 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2267 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2270 if (test_bit(Blocked
, &rdev
->flags
)) {
2271 len
+= sprintf(page
+len
, "%sblocked", sep
);
2274 if (!test_bit(Faulty
, &rdev
->flags
) &&
2275 !test_bit(In_sync
, &rdev
->flags
)) {
2276 len
+= sprintf(page
+len
, "%sspare", sep
);
2279 return len
+sprintf(page
+len
, "\n");
2283 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2286 * faulty - simulates and error
2287 * remove - disconnects the device
2288 * writemostly - sets write_mostly
2289 * -writemostly - clears write_mostly
2290 * blocked - sets the Blocked flag
2291 * -blocked - clears the Blocked flag
2292 * insync - sets Insync providing device isn't active
2295 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2296 md_error(rdev
->mddev
, rdev
);
2298 } else if (cmd_match(buf
, "remove")) {
2299 if (rdev
->raid_disk
>= 0)
2302 mddev_t
*mddev
= rdev
->mddev
;
2303 kick_rdev_from_array(rdev
);
2305 md_update_sb(mddev
, 1);
2306 md_new_event(mddev
);
2309 } else if (cmd_match(buf
, "writemostly")) {
2310 set_bit(WriteMostly
, &rdev
->flags
);
2312 } else if (cmd_match(buf
, "-writemostly")) {
2313 clear_bit(WriteMostly
, &rdev
->flags
);
2315 } else if (cmd_match(buf
, "blocked")) {
2316 set_bit(Blocked
, &rdev
->flags
);
2318 } else if (cmd_match(buf
, "-blocked")) {
2319 clear_bit(Blocked
, &rdev
->flags
);
2320 wake_up(&rdev
->blocked_wait
);
2321 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2322 md_wakeup_thread(rdev
->mddev
->thread
);
2325 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2326 set_bit(In_sync
, &rdev
->flags
);
2329 if (!err
&& rdev
->sysfs_state
)
2330 sysfs_notify_dirent(rdev
->sysfs_state
);
2331 return err
? err
: len
;
2333 static struct rdev_sysfs_entry rdev_state
=
2334 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2337 errors_show(mdk_rdev_t
*rdev
, char *page
)
2339 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2343 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2346 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2347 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2348 atomic_set(&rdev
->corrected_errors
, n
);
2353 static struct rdev_sysfs_entry rdev_errors
=
2354 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2357 slot_show(mdk_rdev_t
*rdev
, char *page
)
2359 if (rdev
->raid_disk
< 0)
2360 return sprintf(page
, "none\n");
2362 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2366 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2371 int slot
= simple_strtoul(buf
, &e
, 10);
2372 if (strncmp(buf
, "none", 4)==0)
2374 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2376 if (rdev
->mddev
->pers
&& slot
== -1) {
2377 /* Setting 'slot' on an active array requires also
2378 * updating the 'rd%d' link, and communicating
2379 * with the personality with ->hot_*_disk.
2380 * For now we only support removing
2381 * failed/spare devices. This normally happens automatically,
2382 * but not when the metadata is externally managed.
2384 if (rdev
->raid_disk
== -1)
2386 /* personality does all needed checks */
2387 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2389 err
= rdev
->mddev
->pers
->
2390 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2393 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2394 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2395 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2396 md_wakeup_thread(rdev
->mddev
->thread
);
2397 } else if (rdev
->mddev
->pers
) {
2399 /* Activating a spare .. or possibly reactivating
2400 * if we ever get bitmaps working here.
2403 if (rdev
->raid_disk
!= -1)
2406 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2409 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2410 if (rdev2
->raid_disk
== slot
)
2413 rdev
->raid_disk
= slot
;
2414 if (test_bit(In_sync
, &rdev
->flags
))
2415 rdev
->saved_raid_disk
= slot
;
2417 rdev
->saved_raid_disk
= -1;
2418 err
= rdev
->mddev
->pers
->
2419 hot_add_disk(rdev
->mddev
, rdev
);
2421 rdev
->raid_disk
= -1;
2424 sysfs_notify_dirent(rdev
->sysfs_state
);
2425 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2426 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2428 "md: cannot register "
2430 nm
, mdname(rdev
->mddev
));
2432 /* don't wakeup anyone, leave that to userspace. */
2434 if (slot
>= rdev
->mddev
->raid_disks
)
2436 rdev
->raid_disk
= slot
;
2437 /* assume it is working */
2438 clear_bit(Faulty
, &rdev
->flags
);
2439 clear_bit(WriteMostly
, &rdev
->flags
);
2440 set_bit(In_sync
, &rdev
->flags
);
2441 sysfs_notify_dirent(rdev
->sysfs_state
);
2447 static struct rdev_sysfs_entry rdev_slot
=
2448 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2451 offset_show(mdk_rdev_t
*rdev
, char *page
)
2453 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2457 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2460 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2461 if (e
==buf
|| (*e
&& *e
!= '\n'))
2463 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2465 if (rdev
->sectors
&& rdev
->mddev
->external
)
2466 /* Must set offset before size, so overlap checks
2469 rdev
->data_offset
= offset
;
2473 static struct rdev_sysfs_entry rdev_offset
=
2474 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2477 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2479 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2482 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2484 /* check if two start/length pairs overlap */
2492 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2494 unsigned long long blocks
;
2497 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2500 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2501 return -EINVAL
; /* sector conversion overflow */
2504 if (new != blocks
* 2)
2505 return -EINVAL
; /* unsigned long long to sector_t overflow */
2512 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2514 mddev_t
*my_mddev
= rdev
->mddev
;
2515 sector_t oldsectors
= rdev
->sectors
;
2518 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2520 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2521 if (my_mddev
->persistent
) {
2522 sectors
= super_types
[my_mddev
->major_version
].
2523 rdev_size_change(rdev
, sectors
);
2526 } else if (!sectors
)
2527 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2530 if (sectors
< my_mddev
->dev_sectors
)
2531 return -EINVAL
; /* component must fit device */
2533 rdev
->sectors
= sectors
;
2534 if (sectors
> oldsectors
&& my_mddev
->external
) {
2535 /* need to check that all other rdevs with the same ->bdev
2536 * do not overlap. We need to unlock the mddev to avoid
2537 * a deadlock. We have already changed rdev->sectors, and if
2538 * we have to change it back, we will have the lock again.
2542 struct list_head
*tmp
;
2544 mddev_unlock(my_mddev
);
2545 for_each_mddev(mddev
, tmp
) {
2549 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2550 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2551 (rdev
->bdev
== rdev2
->bdev
&&
2553 overlaps(rdev
->data_offset
, rdev
->sectors
,
2559 mddev_unlock(mddev
);
2565 mddev_lock(my_mddev
);
2567 /* Someone else could have slipped in a size
2568 * change here, but doing so is just silly.
2569 * We put oldsectors back because we *know* it is
2570 * safe, and trust userspace not to race with
2573 rdev
->sectors
= oldsectors
;
2580 static struct rdev_sysfs_entry rdev_size
=
2581 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2584 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2586 unsigned long long recovery_start
= rdev
->recovery_offset
;
2588 if (test_bit(In_sync
, &rdev
->flags
) ||
2589 recovery_start
== MaxSector
)
2590 return sprintf(page
, "none\n");
2592 return sprintf(page
, "%llu\n", recovery_start
);
2595 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2597 unsigned long long recovery_start
;
2599 if (cmd_match(buf
, "none"))
2600 recovery_start
= MaxSector
;
2601 else if (strict_strtoull(buf
, 10, &recovery_start
))
2604 if (rdev
->mddev
->pers
&&
2605 rdev
->raid_disk
>= 0)
2608 rdev
->recovery_offset
= recovery_start
;
2609 if (recovery_start
== MaxSector
)
2610 set_bit(In_sync
, &rdev
->flags
);
2612 clear_bit(In_sync
, &rdev
->flags
);
2616 static struct rdev_sysfs_entry rdev_recovery_start
=
2617 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2619 static struct attribute
*rdev_default_attrs
[] = {
2625 &rdev_recovery_start
.attr
,
2629 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2631 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2632 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2633 mddev_t
*mddev
= rdev
->mddev
;
2639 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2641 if (rdev
->mddev
== NULL
)
2644 rv
= entry
->show(rdev
, page
);
2645 mddev_unlock(mddev
);
2651 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2652 const char *page
, size_t length
)
2654 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2655 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2657 mddev_t
*mddev
= rdev
->mddev
;
2661 if (!capable(CAP_SYS_ADMIN
))
2663 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2665 if (rdev
->mddev
== NULL
)
2668 rv
= entry
->store(rdev
, page
, length
);
2669 mddev_unlock(mddev
);
2674 static void rdev_free(struct kobject
*ko
)
2676 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2679 static const struct sysfs_ops rdev_sysfs_ops
= {
2680 .show
= rdev_attr_show
,
2681 .store
= rdev_attr_store
,
2683 static struct kobj_type rdev_ktype
= {
2684 .release
= rdev_free
,
2685 .sysfs_ops
= &rdev_sysfs_ops
,
2686 .default_attrs
= rdev_default_attrs
,
2690 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2692 * mark the device faulty if:
2694 * - the device is nonexistent (zero size)
2695 * - the device has no valid superblock
2697 * a faulty rdev _never_ has rdev->sb set.
2699 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2701 char b
[BDEVNAME_SIZE
];
2706 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2708 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2709 return ERR_PTR(-ENOMEM
);
2712 if ((err
= alloc_disk_sb(rdev
)))
2715 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2719 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2722 rdev
->saved_raid_disk
= -1;
2723 rdev
->raid_disk
= -1;
2725 rdev
->data_offset
= 0;
2726 rdev
->sb_events
= 0;
2727 rdev
->last_read_error
.tv_sec
= 0;
2728 rdev
->last_read_error
.tv_nsec
= 0;
2729 atomic_set(&rdev
->nr_pending
, 0);
2730 atomic_set(&rdev
->read_errors
, 0);
2731 atomic_set(&rdev
->corrected_errors
, 0);
2733 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2736 "md: %s has zero or unknown size, marking faulty!\n",
2737 bdevname(rdev
->bdev
,b
));
2742 if (super_format
>= 0) {
2743 err
= super_types
[super_format
].
2744 load_super(rdev
, NULL
, super_minor
);
2745 if (err
== -EINVAL
) {
2747 "md: %s does not have a valid v%d.%d "
2748 "superblock, not importing!\n",
2749 bdevname(rdev
->bdev
,b
),
2750 super_format
, super_minor
);
2755 "md: could not read %s's sb, not importing!\n",
2756 bdevname(rdev
->bdev
,b
));
2761 INIT_LIST_HEAD(&rdev
->same_set
);
2762 init_waitqueue_head(&rdev
->blocked_wait
);
2767 if (rdev
->sb_page
) {
2773 return ERR_PTR(err
);
2777 * Check a full RAID array for plausibility
2781 static void analyze_sbs(mddev_t
* mddev
)
2784 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2785 char b
[BDEVNAME_SIZE
];
2788 rdev_for_each(rdev
, tmp
, mddev
)
2789 switch (super_types
[mddev
->major_version
].
2790 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2798 "md: fatal superblock inconsistency in %s"
2799 " -- removing from array\n",
2800 bdevname(rdev
->bdev
,b
));
2801 kick_rdev_from_array(rdev
);
2805 super_types
[mddev
->major_version
].
2806 validate_super(mddev
, freshest
);
2809 rdev_for_each(rdev
, tmp
, mddev
) {
2810 if (rdev
->desc_nr
>= mddev
->max_disks
||
2811 i
> mddev
->max_disks
) {
2813 "md: %s: %s: only %d devices permitted\n",
2814 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2816 kick_rdev_from_array(rdev
);
2819 if (rdev
!= freshest
)
2820 if (super_types
[mddev
->major_version
].
2821 validate_super(mddev
, rdev
)) {
2822 printk(KERN_WARNING
"md: kicking non-fresh %s"
2824 bdevname(rdev
->bdev
,b
));
2825 kick_rdev_from_array(rdev
);
2828 if (mddev
->level
== LEVEL_MULTIPATH
) {
2829 rdev
->desc_nr
= i
++;
2830 rdev
->raid_disk
= rdev
->desc_nr
;
2831 set_bit(In_sync
, &rdev
->flags
);
2832 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2833 rdev
->raid_disk
= -1;
2834 clear_bit(In_sync
, &rdev
->flags
);
2839 /* Read a fixed-point number.
2840 * Numbers in sysfs attributes should be in "standard" units where
2841 * possible, so time should be in seconds.
2842 * However we internally use a a much smaller unit such as
2843 * milliseconds or jiffies.
2844 * This function takes a decimal number with a possible fractional
2845 * component, and produces an integer which is the result of
2846 * multiplying that number by 10^'scale'.
2847 * all without any floating-point arithmetic.
2849 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
2851 unsigned long result
= 0;
2853 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
2856 else if (decimals
< scale
) {
2859 result
= result
* 10 + value
;
2871 while (decimals
< scale
) {
2880 static void md_safemode_timeout(unsigned long data
);
2883 safe_delay_show(mddev_t
*mddev
, char *page
)
2885 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2886 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2889 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2893 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
2896 mddev
->safemode_delay
= 0;
2898 unsigned long old_delay
= mddev
->safemode_delay
;
2899 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2900 if (mddev
->safemode_delay
== 0)
2901 mddev
->safemode_delay
= 1;
2902 if (mddev
->safemode_delay
< old_delay
)
2903 md_safemode_timeout((unsigned long)mddev
);
2907 static struct md_sysfs_entry md_safe_delay
=
2908 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2911 level_show(mddev_t
*mddev
, char *page
)
2913 struct mdk_personality
*p
= mddev
->pers
;
2915 return sprintf(page
, "%s\n", p
->name
);
2916 else if (mddev
->clevel
[0])
2917 return sprintf(page
, "%s\n", mddev
->clevel
);
2918 else if (mddev
->level
!= LEVEL_NONE
)
2919 return sprintf(page
, "%d\n", mddev
->level
);
2925 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2929 struct mdk_personality
*pers
;
2933 if (mddev
->pers
== NULL
) {
2936 if (len
>= sizeof(mddev
->clevel
))
2938 strncpy(mddev
->clevel
, buf
, len
);
2939 if (mddev
->clevel
[len
-1] == '\n')
2941 mddev
->clevel
[len
] = 0;
2942 mddev
->level
= LEVEL_NONE
;
2946 /* request to change the personality. Need to ensure:
2947 * - array is not engaged in resync/recovery/reshape
2948 * - old personality can be suspended
2949 * - new personality will access other array.
2952 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2955 if (!mddev
->pers
->quiesce
) {
2956 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2957 mdname(mddev
), mddev
->pers
->name
);
2961 /* Now find the new personality */
2962 if (len
== 0 || len
>= sizeof(level
))
2964 strncpy(level
, buf
, len
);
2965 if (level
[len
-1] == '\n')
2969 request_module("md-%s", level
);
2970 spin_lock(&pers_lock
);
2971 pers
= find_pers(LEVEL_NONE
, level
);
2972 if (!pers
|| !try_module_get(pers
->owner
)) {
2973 spin_unlock(&pers_lock
);
2974 printk(KERN_WARNING
"md: personality %s not loaded\n", level
);
2977 spin_unlock(&pers_lock
);
2979 if (pers
== mddev
->pers
) {
2980 /* Nothing to do! */
2981 module_put(pers
->owner
);
2984 if (!pers
->takeover
) {
2985 module_put(pers
->owner
);
2986 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2987 mdname(mddev
), level
);
2991 /* ->takeover must set new_* and/or delta_disks
2992 * if it succeeds, and may set them when it fails.
2994 priv
= pers
->takeover(mddev
);
2996 mddev
->new_level
= mddev
->level
;
2997 mddev
->new_layout
= mddev
->layout
;
2998 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2999 mddev
->raid_disks
-= mddev
->delta_disks
;
3000 mddev
->delta_disks
= 0;
3001 module_put(pers
->owner
);
3002 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3003 mdname(mddev
), level
);
3004 return PTR_ERR(priv
);
3007 /* Looks like we have a winner */
3008 mddev_suspend(mddev
);
3009 mddev
->pers
->stop(mddev
);
3011 if (mddev
->pers
->sync_request
== NULL
&&
3012 pers
->sync_request
!= NULL
) {
3013 /* need to add the md_redundancy_group */
3014 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3016 "md: cannot register extra attributes for %s\n",
3018 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3020 if (mddev
->pers
->sync_request
!= NULL
&&
3021 pers
->sync_request
== NULL
) {
3022 /* need to remove the md_redundancy_group */
3023 if (mddev
->to_remove
== NULL
)
3024 mddev
->to_remove
= &md_redundancy_group
;
3027 module_put(mddev
->pers
->owner
);
3028 /* Invalidate devices that are now superfluous */
3029 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3030 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
3031 rdev
->raid_disk
= -1;
3032 clear_bit(In_sync
, &rdev
->flags
);
3035 mddev
->private = priv
;
3036 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3037 mddev
->level
= mddev
->new_level
;
3038 mddev
->layout
= mddev
->new_layout
;
3039 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3040 mddev
->delta_disks
= 0;
3042 mddev_resume(mddev
);
3043 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3044 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3045 md_wakeup_thread(mddev
->thread
);
3049 static struct md_sysfs_entry md_level
=
3050 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3054 layout_show(mddev_t
*mddev
, char *page
)
3056 /* just a number, not meaningful for all levels */
3057 if (mddev
->reshape_position
!= MaxSector
&&
3058 mddev
->layout
!= mddev
->new_layout
)
3059 return sprintf(page
, "%d (%d)\n",
3060 mddev
->new_layout
, mddev
->layout
);
3061 return sprintf(page
, "%d\n", mddev
->layout
);
3065 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3068 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3070 if (!*buf
|| (*e
&& *e
!= '\n'))
3075 if (mddev
->pers
->check_reshape
== NULL
)
3077 mddev
->new_layout
= n
;
3078 err
= mddev
->pers
->check_reshape(mddev
);
3080 mddev
->new_layout
= mddev
->layout
;
3084 mddev
->new_layout
= n
;
3085 if (mddev
->reshape_position
== MaxSector
)
3090 static struct md_sysfs_entry md_layout
=
3091 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3095 raid_disks_show(mddev_t
*mddev
, char *page
)
3097 if (mddev
->raid_disks
== 0)
3099 if (mddev
->reshape_position
!= MaxSector
&&
3100 mddev
->delta_disks
!= 0)
3101 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3102 mddev
->raid_disks
- mddev
->delta_disks
);
3103 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3106 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3109 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3113 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3115 if (!*buf
|| (*e
&& *e
!= '\n'))
3119 rv
= update_raid_disks(mddev
, n
);
3120 else if (mddev
->reshape_position
!= MaxSector
) {
3121 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3122 mddev
->delta_disks
= n
- olddisks
;
3123 mddev
->raid_disks
= n
;
3125 mddev
->raid_disks
= n
;
3126 return rv
? rv
: len
;
3128 static struct md_sysfs_entry md_raid_disks
=
3129 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3132 chunk_size_show(mddev_t
*mddev
, char *page
)
3134 if (mddev
->reshape_position
!= MaxSector
&&
3135 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3136 return sprintf(page
, "%d (%d)\n",
3137 mddev
->new_chunk_sectors
<< 9,
3138 mddev
->chunk_sectors
<< 9);
3139 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3143 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3146 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3148 if (!*buf
|| (*e
&& *e
!= '\n'))
3153 if (mddev
->pers
->check_reshape
== NULL
)
3155 mddev
->new_chunk_sectors
= n
>> 9;
3156 err
= mddev
->pers
->check_reshape(mddev
);
3158 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3162 mddev
->new_chunk_sectors
= n
>> 9;
3163 if (mddev
->reshape_position
== MaxSector
)
3164 mddev
->chunk_sectors
= n
>> 9;
3168 static struct md_sysfs_entry md_chunk_size
=
3169 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3172 resync_start_show(mddev_t
*mddev
, char *page
)
3174 if (mddev
->recovery_cp
== MaxSector
)
3175 return sprintf(page
, "none\n");
3176 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3180 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3183 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3187 if (cmd_match(buf
, "none"))
3189 else if (!*buf
|| (*e
&& *e
!= '\n'))
3192 mddev
->recovery_cp
= n
;
3195 static struct md_sysfs_entry md_resync_start
=
3196 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3199 * The array state can be:
3202 * No devices, no size, no level
3203 * Equivalent to STOP_ARRAY ioctl
3205 * May have some settings, but array is not active
3206 * all IO results in error
3207 * When written, doesn't tear down array, but just stops it
3208 * suspended (not supported yet)
3209 * All IO requests will block. The array can be reconfigured.
3210 * Writing this, if accepted, will block until array is quiescent
3212 * no resync can happen. no superblocks get written.
3213 * write requests fail
3215 * like readonly, but behaves like 'clean' on a write request.
3217 * clean - no pending writes, but otherwise active.
3218 * When written to inactive array, starts without resync
3219 * If a write request arrives then
3220 * if metadata is known, mark 'dirty' and switch to 'active'.
3221 * if not known, block and switch to write-pending
3222 * If written to an active array that has pending writes, then fails.
3224 * fully active: IO and resync can be happening.
3225 * When written to inactive array, starts with resync
3228 * clean, but writes are blocked waiting for 'active' to be written.
3231 * like active, but no writes have been seen for a while (100msec).
3234 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3235 write_pending
, active_idle
, bad_word
};
3236 static char *array_states
[] = {
3237 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3238 "write-pending", "active-idle", NULL
};
3240 static int match_word(const char *word
, char **list
)
3243 for (n
=0; list
[n
]; n
++)
3244 if (cmd_match(word
, list
[n
]))
3250 array_state_show(mddev_t
*mddev
, char *page
)
3252 enum array_state st
= inactive
;
3265 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3267 else if (mddev
->safemode
)
3273 if (list_empty(&mddev
->disks
) &&
3274 mddev
->raid_disks
== 0 &&
3275 mddev
->dev_sectors
== 0)
3280 return sprintf(page
, "%s\n", array_states
[st
]);
3283 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3284 static int do_md_run(mddev_t
* mddev
);
3285 static int restart_array(mddev_t
*mddev
);
3288 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3291 enum array_state st
= match_word(buf
, array_states
);
3296 /* stopping an active array */
3297 if (atomic_read(&mddev
->openers
) > 0)
3299 err
= do_md_stop(mddev
, 0, 0);
3302 /* stopping an active array */
3304 if (atomic_read(&mddev
->openers
) > 0)
3306 err
= do_md_stop(mddev
, 2, 0);
3308 err
= 0; /* already inactive */
3311 break; /* not supported yet */
3314 err
= do_md_stop(mddev
, 1, 0);
3317 set_disk_ro(mddev
->gendisk
, 1);
3318 err
= do_md_run(mddev
);
3324 err
= do_md_stop(mddev
, 1, 0);
3325 else if (mddev
->ro
== 1)
3326 err
= restart_array(mddev
);
3329 set_disk_ro(mddev
->gendisk
, 0);
3333 err
= do_md_run(mddev
);
3338 restart_array(mddev
);
3339 spin_lock_irq(&mddev
->write_lock
);
3340 if (atomic_read(&mddev
->writes_pending
) == 0) {
3341 if (mddev
->in_sync
== 0) {
3343 if (mddev
->safemode
== 1)
3344 mddev
->safemode
= 0;
3345 if (mddev
->persistent
)
3346 set_bit(MD_CHANGE_CLEAN
,
3352 spin_unlock_irq(&mddev
->write_lock
);
3358 restart_array(mddev
);
3359 if (mddev
->external
)
3360 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3361 wake_up(&mddev
->sb_wait
);
3365 set_disk_ro(mddev
->gendisk
, 0);
3366 err
= do_md_run(mddev
);
3371 /* these cannot be set */
3377 sysfs_notify_dirent(mddev
->sysfs_state
);
3381 static struct md_sysfs_entry md_array_state
=
3382 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3385 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3386 return sprintf(page
, "%d\n",
3387 atomic_read(&mddev
->max_corr_read_errors
));
3391 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3394 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3396 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3397 atomic_set(&mddev
->max_corr_read_errors
, n
);
3403 static struct md_sysfs_entry max_corr_read_errors
=
3404 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3405 max_corrected_read_errors_store
);
3408 null_show(mddev_t
*mddev
, char *page
)
3414 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3416 /* buf must be %d:%d\n? giving major and minor numbers */
3417 /* The new device is added to the array.
3418 * If the array has a persistent superblock, we read the
3419 * superblock to initialise info and check validity.
3420 * Otherwise, only checking done is that in bind_rdev_to_array,
3421 * which mainly checks size.
3424 int major
= simple_strtoul(buf
, &e
, 10);
3430 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3432 minor
= simple_strtoul(e
+1, &e
, 10);
3433 if (*e
&& *e
!= '\n')
3435 dev
= MKDEV(major
, minor
);
3436 if (major
!= MAJOR(dev
) ||
3437 minor
!= MINOR(dev
))
3441 if (mddev
->persistent
) {
3442 rdev
= md_import_device(dev
, mddev
->major_version
,
3443 mddev
->minor_version
);
3444 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3445 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3446 mdk_rdev_t
, same_set
);
3447 err
= super_types
[mddev
->major_version
]
3448 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3452 } else if (mddev
->external
)
3453 rdev
= md_import_device(dev
, -2, -1);
3455 rdev
= md_import_device(dev
, -1, -1);
3458 return PTR_ERR(rdev
);
3459 err
= bind_rdev_to_array(rdev
, mddev
);
3463 return err
? err
: len
;
3466 static struct md_sysfs_entry md_new_device
=
3467 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3470 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3473 unsigned long chunk
, end_chunk
;
3477 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3479 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3480 if (buf
== end
) break;
3481 if (*end
== '-') { /* range */
3483 end_chunk
= simple_strtoul(buf
, &end
, 0);
3484 if (buf
== end
) break;
3486 if (*end
&& !isspace(*end
)) break;
3487 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3488 buf
= skip_spaces(end
);
3490 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3495 static struct md_sysfs_entry md_bitmap
=
3496 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3499 size_show(mddev_t
*mddev
, char *page
)
3501 return sprintf(page
, "%llu\n",
3502 (unsigned long long)mddev
->dev_sectors
/ 2);
3505 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3508 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3510 /* If array is inactive, we can reduce the component size, but
3511 * not increase it (except from 0).
3512 * If array is active, we can try an on-line resize
3515 int err
= strict_blocks_to_sectors(buf
, §ors
);
3520 err
= update_size(mddev
, sectors
);
3521 md_update_sb(mddev
, 1);
3523 if (mddev
->dev_sectors
== 0 ||
3524 mddev
->dev_sectors
> sectors
)
3525 mddev
->dev_sectors
= sectors
;
3529 return err
? err
: len
;
3532 static struct md_sysfs_entry md_size
=
3533 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3538 * 'none' for arrays with no metadata (good luck...)
3539 * 'external' for arrays with externally managed metadata,
3540 * or N.M for internally known formats
3543 metadata_show(mddev_t
*mddev
, char *page
)
3545 if (mddev
->persistent
)
3546 return sprintf(page
, "%d.%d\n",
3547 mddev
->major_version
, mddev
->minor_version
);
3548 else if (mddev
->external
)
3549 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3551 return sprintf(page
, "none\n");
3555 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3559 /* Changing the details of 'external' metadata is
3560 * always permitted. Otherwise there must be
3561 * no devices attached to the array.
3563 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3565 else if (!list_empty(&mddev
->disks
))
3568 if (cmd_match(buf
, "none")) {
3569 mddev
->persistent
= 0;
3570 mddev
->external
= 0;
3571 mddev
->major_version
= 0;
3572 mddev
->minor_version
= 90;
3575 if (strncmp(buf
, "external:", 9) == 0) {
3576 size_t namelen
= len
-9;
3577 if (namelen
>= sizeof(mddev
->metadata_type
))
3578 namelen
= sizeof(mddev
->metadata_type
)-1;
3579 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3580 mddev
->metadata_type
[namelen
] = 0;
3581 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3582 mddev
->metadata_type
[--namelen
] = 0;
3583 mddev
->persistent
= 0;
3584 mddev
->external
= 1;
3585 mddev
->major_version
= 0;
3586 mddev
->minor_version
= 90;
3589 major
= simple_strtoul(buf
, &e
, 10);
3590 if (e
==buf
|| *e
!= '.')
3593 minor
= simple_strtoul(buf
, &e
, 10);
3594 if (e
==buf
|| (*e
&& *e
!= '\n') )
3596 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3598 mddev
->major_version
= major
;
3599 mddev
->minor_version
= minor
;
3600 mddev
->persistent
= 1;
3601 mddev
->external
= 0;
3605 static struct md_sysfs_entry md_metadata
=
3606 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3609 action_show(mddev_t
*mddev
, char *page
)
3611 char *type
= "idle";
3612 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3614 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3615 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3616 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3618 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3619 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3621 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3625 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3628 return sprintf(page
, "%s\n", type
);
3632 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3634 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3637 if (cmd_match(page
, "frozen"))
3638 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3640 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3642 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3643 if (mddev
->sync_thread
) {
3644 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3645 md_unregister_thread(mddev
->sync_thread
);
3646 mddev
->sync_thread
= NULL
;
3647 mddev
->recovery
= 0;
3649 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3650 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3652 else if (cmd_match(page
, "resync"))
3653 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3654 else if (cmd_match(page
, "recover")) {
3655 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3656 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3657 } else if (cmd_match(page
, "reshape")) {
3659 if (mddev
->pers
->start_reshape
== NULL
)
3661 err
= mddev
->pers
->start_reshape(mddev
);
3664 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3666 if (cmd_match(page
, "check"))
3667 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3668 else if (!cmd_match(page
, "repair"))
3670 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3671 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3673 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3674 md_wakeup_thread(mddev
->thread
);
3675 sysfs_notify_dirent(mddev
->sysfs_action
);
3680 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3682 return sprintf(page
, "%llu\n",
3683 (unsigned long long) mddev
->resync_mismatches
);
3686 static struct md_sysfs_entry md_scan_mode
=
3687 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3690 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3693 sync_min_show(mddev_t
*mddev
, char *page
)
3695 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3696 mddev
->sync_speed_min
? "local": "system");
3700 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3704 if (strncmp(buf
, "system", 6)==0) {
3705 mddev
->sync_speed_min
= 0;
3708 min
= simple_strtoul(buf
, &e
, 10);
3709 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3711 mddev
->sync_speed_min
= min
;
3715 static struct md_sysfs_entry md_sync_min
=
3716 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3719 sync_max_show(mddev_t
*mddev
, char *page
)
3721 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3722 mddev
->sync_speed_max
? "local": "system");
3726 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3730 if (strncmp(buf
, "system", 6)==0) {
3731 mddev
->sync_speed_max
= 0;
3734 max
= simple_strtoul(buf
, &e
, 10);
3735 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3737 mddev
->sync_speed_max
= max
;
3741 static struct md_sysfs_entry md_sync_max
=
3742 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3745 degraded_show(mddev_t
*mddev
, char *page
)
3747 return sprintf(page
, "%d\n", mddev
->degraded
);
3749 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3752 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3754 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3758 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3762 if (strict_strtol(buf
, 10, &n
))
3765 if (n
!= 0 && n
!= 1)
3768 mddev
->parallel_resync
= n
;
3770 if (mddev
->sync_thread
)
3771 wake_up(&resync_wait
);
3776 /* force parallel resync, even with shared block devices */
3777 static struct md_sysfs_entry md_sync_force_parallel
=
3778 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3779 sync_force_parallel_show
, sync_force_parallel_store
);
3782 sync_speed_show(mddev_t
*mddev
, char *page
)
3784 unsigned long resync
, dt
, db
;
3785 if (mddev
->curr_resync
== 0)
3786 return sprintf(page
, "none\n");
3787 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3788 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3790 db
= resync
- mddev
->resync_mark_cnt
;
3791 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3794 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3797 sync_completed_show(mddev_t
*mddev
, char *page
)
3799 unsigned long max_sectors
, resync
;
3801 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3802 return sprintf(page
, "none\n");
3804 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3805 max_sectors
= mddev
->resync_max_sectors
;
3807 max_sectors
= mddev
->dev_sectors
;
3809 resync
= mddev
->curr_resync_completed
;
3810 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3813 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3816 min_sync_show(mddev_t
*mddev
, char *page
)
3818 return sprintf(page
, "%llu\n",
3819 (unsigned long long)mddev
->resync_min
);
3822 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3824 unsigned long long min
;
3825 if (strict_strtoull(buf
, 10, &min
))
3827 if (min
> mddev
->resync_max
)
3829 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3832 /* Must be a multiple of chunk_size */
3833 if (mddev
->chunk_sectors
) {
3834 sector_t temp
= min
;
3835 if (sector_div(temp
, mddev
->chunk_sectors
))
3838 mddev
->resync_min
= min
;
3843 static struct md_sysfs_entry md_min_sync
=
3844 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3847 max_sync_show(mddev_t
*mddev
, char *page
)
3849 if (mddev
->resync_max
== MaxSector
)
3850 return sprintf(page
, "max\n");
3852 return sprintf(page
, "%llu\n",
3853 (unsigned long long)mddev
->resync_max
);
3856 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3858 if (strncmp(buf
, "max", 3) == 0)
3859 mddev
->resync_max
= MaxSector
;
3861 unsigned long long max
;
3862 if (strict_strtoull(buf
, 10, &max
))
3864 if (max
< mddev
->resync_min
)
3866 if (max
< mddev
->resync_max
&&
3868 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3871 /* Must be a multiple of chunk_size */
3872 if (mddev
->chunk_sectors
) {
3873 sector_t temp
= max
;
3874 if (sector_div(temp
, mddev
->chunk_sectors
))
3877 mddev
->resync_max
= max
;
3879 wake_up(&mddev
->recovery_wait
);
3883 static struct md_sysfs_entry md_max_sync
=
3884 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3887 suspend_lo_show(mddev_t
*mddev
, char *page
)
3889 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3893 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3896 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3898 if (mddev
->pers
== NULL
||
3899 mddev
->pers
->quiesce
== NULL
)
3901 if (buf
== e
|| (*e
&& *e
!= '\n'))
3903 if (new >= mddev
->suspend_hi
||
3904 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3905 mddev
->suspend_lo
= new;
3906 mddev
->pers
->quiesce(mddev
, 2);
3911 static struct md_sysfs_entry md_suspend_lo
=
3912 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3916 suspend_hi_show(mddev_t
*mddev
, char *page
)
3918 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3922 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3925 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3927 if (mddev
->pers
== NULL
||
3928 mddev
->pers
->quiesce
== NULL
)
3930 if (buf
== e
|| (*e
&& *e
!= '\n'))
3932 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3933 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3934 mddev
->suspend_hi
= new;
3935 mddev
->pers
->quiesce(mddev
, 1);
3936 mddev
->pers
->quiesce(mddev
, 0);
3941 static struct md_sysfs_entry md_suspend_hi
=
3942 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3945 reshape_position_show(mddev_t
*mddev
, char *page
)
3947 if (mddev
->reshape_position
!= MaxSector
)
3948 return sprintf(page
, "%llu\n",
3949 (unsigned long long)mddev
->reshape_position
);
3950 strcpy(page
, "none\n");
3955 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3958 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3961 if (buf
== e
|| (*e
&& *e
!= '\n'))
3963 mddev
->reshape_position
= new;
3964 mddev
->delta_disks
= 0;
3965 mddev
->new_level
= mddev
->level
;
3966 mddev
->new_layout
= mddev
->layout
;
3967 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3971 static struct md_sysfs_entry md_reshape_position
=
3972 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3973 reshape_position_store
);
3976 array_size_show(mddev_t
*mddev
, char *page
)
3978 if (mddev
->external_size
)
3979 return sprintf(page
, "%llu\n",
3980 (unsigned long long)mddev
->array_sectors
/2);
3982 return sprintf(page
, "default\n");
3986 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3990 if (strncmp(buf
, "default", 7) == 0) {
3992 sectors
= mddev
->pers
->size(mddev
, 0, 0);
3994 sectors
= mddev
->array_sectors
;
3996 mddev
->external_size
= 0;
3998 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4000 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4003 mddev
->external_size
= 1;
4006 mddev
->array_sectors
= sectors
;
4007 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4009 revalidate_disk(mddev
->gendisk
);
4014 static struct md_sysfs_entry md_array_size
=
4015 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4018 static struct attribute
*md_default_attrs
[] = {
4021 &md_raid_disks
.attr
,
4022 &md_chunk_size
.attr
,
4024 &md_resync_start
.attr
,
4026 &md_new_device
.attr
,
4027 &md_safe_delay
.attr
,
4028 &md_array_state
.attr
,
4029 &md_reshape_position
.attr
,
4030 &md_array_size
.attr
,
4031 &max_corr_read_errors
.attr
,
4035 static struct attribute
*md_redundancy_attrs
[] = {
4037 &md_mismatches
.attr
,
4040 &md_sync_speed
.attr
,
4041 &md_sync_force_parallel
.attr
,
4042 &md_sync_completed
.attr
,
4045 &md_suspend_lo
.attr
,
4046 &md_suspend_hi
.attr
,
4051 static struct attribute_group md_redundancy_group
= {
4053 .attrs
= md_redundancy_attrs
,
4058 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4060 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4061 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4066 rv
= mddev_lock(mddev
);
4068 rv
= entry
->show(mddev
, page
);
4069 mddev_unlock(mddev
);
4075 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4076 const char *page
, size_t length
)
4078 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4079 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4084 if (!capable(CAP_SYS_ADMIN
))
4086 rv
= mddev_lock(mddev
);
4087 if (mddev
->hold_active
== UNTIL_IOCTL
)
4088 mddev
->hold_active
= 0;
4090 rv
= entry
->store(mddev
, page
, length
);
4091 mddev_unlock(mddev
);
4096 static void md_free(struct kobject
*ko
)
4098 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4100 if (mddev
->sysfs_state
)
4101 sysfs_put(mddev
->sysfs_state
);
4103 if (mddev
->gendisk
) {
4104 del_gendisk(mddev
->gendisk
);
4105 put_disk(mddev
->gendisk
);
4108 blk_cleanup_queue(mddev
->queue
);
4113 static const struct sysfs_ops md_sysfs_ops
= {
4114 .show
= md_attr_show
,
4115 .store
= md_attr_store
,
4117 static struct kobj_type md_ktype
= {
4119 .sysfs_ops
= &md_sysfs_ops
,
4120 .default_attrs
= md_default_attrs
,
4125 static void mddev_delayed_delete(struct work_struct
*ws
)
4127 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4129 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4130 kobject_del(&mddev
->kobj
);
4131 kobject_put(&mddev
->kobj
);
4134 static int md_alloc(dev_t dev
, char *name
)
4136 static DEFINE_MUTEX(disks_mutex
);
4137 mddev_t
*mddev
= mddev_find(dev
);
4138 struct gendisk
*disk
;
4147 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4148 shift
= partitioned
? MdpMinorShift
: 0;
4149 unit
= MINOR(mddev
->unit
) >> shift
;
4151 /* wait for any previous instance if this device
4152 * to be completed removed (mddev_delayed_delete).
4154 flush_scheduled_work();
4156 mutex_lock(&disks_mutex
);
4162 /* Need to ensure that 'name' is not a duplicate.
4165 spin_lock(&all_mddevs_lock
);
4167 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4168 if (mddev2
->gendisk
&&
4169 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4170 spin_unlock(&all_mddevs_lock
);
4173 spin_unlock(&all_mddevs_lock
);
4177 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4180 mddev
->queue
->queuedata
= mddev
;
4182 /* Can be unlocked because the queue is new: no concurrency */
4183 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4185 blk_queue_make_request(mddev
->queue
, md_make_request
);
4187 disk
= alloc_disk(1 << shift
);
4189 blk_cleanup_queue(mddev
->queue
);
4190 mddev
->queue
= NULL
;
4193 disk
->major
= MAJOR(mddev
->unit
);
4194 disk
->first_minor
= unit
<< shift
;
4196 strcpy(disk
->disk_name
, name
);
4197 else if (partitioned
)
4198 sprintf(disk
->disk_name
, "md_d%d", unit
);
4200 sprintf(disk
->disk_name
, "md%d", unit
);
4201 disk
->fops
= &md_fops
;
4202 disk
->private_data
= mddev
;
4203 disk
->queue
= mddev
->queue
;
4204 /* Allow extended partitions. This makes the
4205 * 'mdp' device redundant, but we can't really
4208 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4210 mddev
->gendisk
= disk
;
4211 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4212 &disk_to_dev(disk
)->kobj
, "%s", "md");
4214 /* This isn't possible, but as kobject_init_and_add is marked
4215 * __must_check, we must do something with the result
4217 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4221 if (sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4222 printk(KERN_DEBUG
"pointless warning\n");
4224 mutex_unlock(&disks_mutex
);
4226 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4227 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4233 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4235 md_alloc(dev
, NULL
);
4239 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4241 /* val must be "md_*" where * is not all digits.
4242 * We allocate an array with a large free minor number, and
4243 * set the name to val. val must not already be an active name.
4245 int len
= strlen(val
);
4246 char buf
[DISK_NAME_LEN
];
4248 while (len
&& val
[len
-1] == '\n')
4250 if (len
>= DISK_NAME_LEN
)
4252 strlcpy(buf
, val
, len
+1);
4253 if (strncmp(buf
, "md_", 3) != 0)
4255 return md_alloc(0, buf
);
4258 static void md_safemode_timeout(unsigned long data
)
4260 mddev_t
*mddev
= (mddev_t
*) data
;
4262 if (!atomic_read(&mddev
->writes_pending
)) {
4263 mddev
->safemode
= 1;
4264 if (mddev
->external
)
4265 sysfs_notify_dirent(mddev
->sysfs_state
);
4267 md_wakeup_thread(mddev
->thread
);
4270 static int start_dirty_degraded
;
4272 static int do_md_run(mddev_t
* mddev
)
4276 struct gendisk
*disk
;
4277 struct mdk_personality
*pers
;
4279 if (list_empty(&mddev
->disks
))
4280 /* cannot run an array with no devices.. */
4286 /* These two calls synchronise us with the
4287 * sysfs_remove_group calls in mddev_unlock,
4288 * so they must have completed.
4290 mutex_lock(&mddev
->open_mutex
);
4291 mutex_unlock(&mddev
->open_mutex
);
4294 * Analyze all RAID superblock(s)
4296 if (!mddev
->raid_disks
) {
4297 if (!mddev
->persistent
)
4302 if (mddev
->level
!= LEVEL_NONE
)
4303 request_module("md-level-%d", mddev
->level
);
4304 else if (mddev
->clevel
[0])
4305 request_module("md-%s", mddev
->clevel
);
4308 * Drop all container device buffers, from now on
4309 * the only valid external interface is through the md
4312 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4313 if (test_bit(Faulty
, &rdev
->flags
))
4315 sync_blockdev(rdev
->bdev
);
4316 invalidate_bdev(rdev
->bdev
);
4318 /* perform some consistency tests on the device.
4319 * We don't want the data to overlap the metadata,
4320 * Internal Bitmap issues have been handled elsewhere.
4322 if (rdev
->data_offset
< rdev
->sb_start
) {
4323 if (mddev
->dev_sectors
&&
4324 rdev
->data_offset
+ mddev
->dev_sectors
4326 printk("md: %s: data overlaps metadata\n",
4331 if (rdev
->sb_start
+ rdev
->sb_size
/512
4332 > rdev
->data_offset
) {
4333 printk("md: %s: metadata overlaps data\n",
4338 sysfs_notify_dirent(rdev
->sysfs_state
);
4341 disk
= mddev
->gendisk
;
4343 spin_lock(&pers_lock
);
4344 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4345 if (!pers
|| !try_module_get(pers
->owner
)) {
4346 spin_unlock(&pers_lock
);
4347 if (mddev
->level
!= LEVEL_NONE
)
4348 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4351 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4356 spin_unlock(&pers_lock
);
4357 if (mddev
->level
!= pers
->level
) {
4358 mddev
->level
= pers
->level
;
4359 mddev
->new_level
= pers
->level
;
4361 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4363 if (mddev
->reshape_position
!= MaxSector
&&
4364 pers
->start_reshape
== NULL
) {
4365 /* This personality cannot handle reshaping... */
4367 module_put(pers
->owner
);
4371 if (pers
->sync_request
) {
4372 /* Warn if this is a potentially silly
4375 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4379 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4380 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4382 rdev
->bdev
->bd_contains
==
4383 rdev2
->bdev
->bd_contains
) {
4385 "%s: WARNING: %s appears to be"
4386 " on the same physical disk as"
4389 bdevname(rdev
->bdev
,b
),
4390 bdevname(rdev2
->bdev
,b2
));
4397 "True protection against single-disk"
4398 " failure might be compromised.\n");
4401 mddev
->recovery
= 0;
4402 /* may be over-ridden by personality */
4403 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4405 mddev
->barriers_work
= 1;
4406 mddev
->ok_start_degraded
= start_dirty_degraded
;
4408 if (start_readonly
&& mddev
->ro
== 0)
4409 mddev
->ro
= 2; /* read-only, but switch on first write */
4411 err
= mddev
->pers
->run(mddev
);
4413 printk(KERN_ERR
"md: pers->run() failed ...\n");
4414 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4415 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4416 " but 'external_size' not in effect?\n", __func__
);
4418 "md: invalid array_size %llu > default size %llu\n",
4419 (unsigned long long)mddev
->array_sectors
/ 2,
4420 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4422 mddev
->pers
->stop(mddev
);
4424 if (err
== 0 && mddev
->pers
->sync_request
) {
4425 err
= bitmap_create(mddev
);
4427 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4428 mdname(mddev
), err
);
4429 mddev
->pers
->stop(mddev
);
4433 module_put(mddev
->pers
->owner
);
4435 bitmap_destroy(mddev
);
4438 if (mddev
->pers
->sync_request
) {
4439 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4441 "md: cannot register extra attributes for %s\n",
4443 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4444 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4447 atomic_set(&mddev
->writes_pending
,0);
4448 atomic_set(&mddev
->max_corr_read_errors
,
4449 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4450 mddev
->safemode
= 0;
4451 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4452 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4453 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4456 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4457 if (rdev
->raid_disk
>= 0) {
4459 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4460 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4461 printk("md: cannot register %s for %s\n",
4465 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4468 md_update_sb(mddev
, 0);
4470 set_capacity(disk
, mddev
->array_sectors
);
4472 md_wakeup_thread(mddev
->thread
);
4473 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4475 revalidate_disk(mddev
->gendisk
);
4477 md_new_event(mddev
);
4478 sysfs_notify_dirent(mddev
->sysfs_state
);
4479 if (mddev
->sysfs_action
)
4480 sysfs_notify_dirent(mddev
->sysfs_action
);
4481 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4482 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4486 static int restart_array(mddev_t
*mddev
)
4488 struct gendisk
*disk
= mddev
->gendisk
;
4490 /* Complain if it has no devices */
4491 if (list_empty(&mddev
->disks
))
4497 mddev
->safemode
= 0;
4499 set_disk_ro(disk
, 0);
4500 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4502 /* Kick recovery or resync if necessary */
4503 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4504 md_wakeup_thread(mddev
->thread
);
4505 md_wakeup_thread(mddev
->sync_thread
);
4506 sysfs_notify_dirent(mddev
->sysfs_state
);
4510 /* similar to deny_write_access, but accounts for our holding a reference
4511 * to the file ourselves */
4512 static int deny_bitmap_write_access(struct file
* file
)
4514 struct inode
*inode
= file
->f_mapping
->host
;
4516 spin_lock(&inode
->i_lock
);
4517 if (atomic_read(&inode
->i_writecount
) > 1) {
4518 spin_unlock(&inode
->i_lock
);
4521 atomic_set(&inode
->i_writecount
, -1);
4522 spin_unlock(&inode
->i_lock
);
4527 void restore_bitmap_write_access(struct file
*file
)
4529 struct inode
*inode
= file
->f_mapping
->host
;
4531 spin_lock(&inode
->i_lock
);
4532 atomic_set(&inode
->i_writecount
, 1);
4533 spin_unlock(&inode
->i_lock
);
4537 * 0 - completely stop and dis-assemble array
4538 * 1 - switch to readonly
4539 * 2 - stop but do not disassemble array
4541 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4544 struct gendisk
*disk
= mddev
->gendisk
;
4547 mutex_lock(&mddev
->open_mutex
);
4548 if (atomic_read(&mddev
->openers
) > is_open
) {
4549 printk("md: %s still in use.\n",mdname(mddev
));
4551 } else if (mddev
->pers
) {
4553 if (mddev
->sync_thread
) {
4554 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4555 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4556 md_unregister_thread(mddev
->sync_thread
);
4557 mddev
->sync_thread
= NULL
;
4560 del_timer_sync(&mddev
->safemode_timer
);
4563 case 1: /* readonly */
4569 case 0: /* disassemble */
4571 bitmap_flush(mddev
);
4572 md_super_wait(mddev
);
4574 set_disk_ro(disk
, 0);
4576 mddev
->pers
->stop(mddev
);
4577 mddev
->queue
->merge_bvec_fn
= NULL
;
4578 mddev
->queue
->unplug_fn
= NULL
;
4579 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4580 module_put(mddev
->pers
->owner
);
4581 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
4582 mddev
->to_remove
= &md_redundancy_group
;
4584 /* tell userspace to handle 'inactive' */
4585 sysfs_notify_dirent(mddev
->sysfs_state
);
4587 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4588 if (rdev
->raid_disk
>= 0) {
4590 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4591 sysfs_remove_link(&mddev
->kobj
, nm
);
4594 set_capacity(disk
, 0);
4600 if (!mddev
->in_sync
|| mddev
->flags
) {
4601 /* mark array as shutdown cleanly */
4603 md_update_sb(mddev
, 1);
4606 set_disk_ro(disk
, 1);
4607 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4611 mutex_unlock(&mddev
->open_mutex
);
4615 * Free resources if final stop
4619 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4621 bitmap_destroy(mddev
);
4622 if (mddev
->bitmap_info
.file
) {
4623 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4624 fput(mddev
->bitmap_info
.file
);
4625 mddev
->bitmap_info
.file
= NULL
;
4627 mddev
->bitmap_info
.offset
= 0;
4629 export_array(mddev
);
4631 mddev
->array_sectors
= 0;
4632 mddev
->external_size
= 0;
4633 mddev
->dev_sectors
= 0;
4634 mddev
->raid_disks
= 0;
4635 mddev
->recovery_cp
= 0;
4636 mddev
->resync_min
= 0;
4637 mddev
->resync_max
= MaxSector
;
4638 mddev
->reshape_position
= MaxSector
;
4639 mddev
->external
= 0;
4640 mddev
->persistent
= 0;
4641 mddev
->level
= LEVEL_NONE
;
4642 mddev
->clevel
[0] = 0;
4645 mddev
->metadata_type
[0] = 0;
4646 mddev
->chunk_sectors
= 0;
4647 mddev
->ctime
= mddev
->utime
= 0;
4649 mddev
->max_disks
= 0;
4651 mddev
->delta_disks
= 0;
4652 mddev
->new_level
= LEVEL_NONE
;
4653 mddev
->new_layout
= 0;
4654 mddev
->new_chunk_sectors
= 0;
4655 mddev
->curr_resync
= 0;
4656 mddev
->resync_mismatches
= 0;
4657 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4658 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4659 mddev
->recovery
= 0;
4662 mddev
->degraded
= 0;
4663 mddev
->barriers_work
= 0;
4664 mddev
->safemode
= 0;
4665 mddev
->bitmap_info
.offset
= 0;
4666 mddev
->bitmap_info
.default_offset
= 0;
4667 mddev
->bitmap_info
.chunksize
= 0;
4668 mddev
->bitmap_info
.daemon_sleep
= 0;
4669 mddev
->bitmap_info
.max_write_behind
= 0;
4670 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4671 if (mddev
->hold_active
== UNTIL_STOP
)
4672 mddev
->hold_active
= 0;
4674 } else if (mddev
->pers
)
4675 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4678 blk_integrity_unregister(disk
);
4679 md_new_event(mddev
);
4680 sysfs_notify_dirent(mddev
->sysfs_state
);
4685 static void autorun_array(mddev_t
*mddev
)
4690 if (list_empty(&mddev
->disks
))
4693 printk(KERN_INFO
"md: running: ");
4695 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4696 char b
[BDEVNAME_SIZE
];
4697 printk("<%s>", bdevname(rdev
->bdev
,b
));
4701 err
= do_md_run(mddev
);
4703 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4704 do_md_stop(mddev
, 0, 0);
4709 * lets try to run arrays based on all disks that have arrived
4710 * until now. (those are in pending_raid_disks)
4712 * the method: pick the first pending disk, collect all disks with
4713 * the same UUID, remove all from the pending list and put them into
4714 * the 'same_array' list. Then order this list based on superblock
4715 * update time (freshest comes first), kick out 'old' disks and
4716 * compare superblocks. If everything's fine then run it.
4718 * If "unit" is allocated, then bump its reference count
4720 static void autorun_devices(int part
)
4722 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4724 char b
[BDEVNAME_SIZE
];
4726 printk(KERN_INFO
"md: autorun ...\n");
4727 while (!list_empty(&pending_raid_disks
)) {
4730 LIST_HEAD(candidates
);
4731 rdev0
= list_entry(pending_raid_disks
.next
,
4732 mdk_rdev_t
, same_set
);
4734 printk(KERN_INFO
"md: considering %s ...\n",
4735 bdevname(rdev0
->bdev
,b
));
4736 INIT_LIST_HEAD(&candidates
);
4737 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4738 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4739 printk(KERN_INFO
"md: adding %s ...\n",
4740 bdevname(rdev
->bdev
,b
));
4741 list_move(&rdev
->same_set
, &candidates
);
4744 * now we have a set of devices, with all of them having
4745 * mostly sane superblocks. It's time to allocate the
4749 dev
= MKDEV(mdp_major
,
4750 rdev0
->preferred_minor
<< MdpMinorShift
);
4751 unit
= MINOR(dev
) >> MdpMinorShift
;
4753 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4756 if (rdev0
->preferred_minor
!= unit
) {
4757 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4758 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4762 md_probe(dev
, NULL
, NULL
);
4763 mddev
= mddev_find(dev
);
4764 if (!mddev
|| !mddev
->gendisk
) {
4768 "md: cannot allocate memory for md drive.\n");
4771 if (mddev_lock(mddev
))
4772 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4774 else if (mddev
->raid_disks
|| mddev
->major_version
4775 || !list_empty(&mddev
->disks
)) {
4777 "md: %s already running, cannot run %s\n",
4778 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4779 mddev_unlock(mddev
);
4781 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4782 mddev
->persistent
= 1;
4783 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4784 list_del_init(&rdev
->same_set
);
4785 if (bind_rdev_to_array(rdev
, mddev
))
4788 autorun_array(mddev
);
4789 mddev_unlock(mddev
);
4791 /* on success, candidates will be empty, on error
4794 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4795 list_del_init(&rdev
->same_set
);
4800 printk(KERN_INFO
"md: ... autorun DONE.\n");
4802 #endif /* !MODULE */
4804 static int get_version(void __user
* arg
)
4808 ver
.major
= MD_MAJOR_VERSION
;
4809 ver
.minor
= MD_MINOR_VERSION
;
4810 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4812 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4818 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4820 mdu_array_info_t info
;
4821 int nr
,working
,insync
,failed
,spare
;
4824 nr
=working
=insync
=failed
=spare
=0;
4825 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4827 if (test_bit(Faulty
, &rdev
->flags
))
4831 if (test_bit(In_sync
, &rdev
->flags
))
4838 info
.major_version
= mddev
->major_version
;
4839 info
.minor_version
= mddev
->minor_version
;
4840 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4841 info
.ctime
= mddev
->ctime
;
4842 info
.level
= mddev
->level
;
4843 info
.size
= mddev
->dev_sectors
/ 2;
4844 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4847 info
.raid_disks
= mddev
->raid_disks
;
4848 info
.md_minor
= mddev
->md_minor
;
4849 info
.not_persistent
= !mddev
->persistent
;
4851 info
.utime
= mddev
->utime
;
4854 info
.state
= (1<<MD_SB_CLEAN
);
4855 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
4856 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4857 info
.active_disks
= insync
;
4858 info
.working_disks
= working
;
4859 info
.failed_disks
= failed
;
4860 info
.spare_disks
= spare
;
4862 info
.layout
= mddev
->layout
;
4863 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4865 if (copy_to_user(arg
, &info
, sizeof(info
)))
4871 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4873 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4874 char *ptr
, *buf
= NULL
;
4877 if (md_allow_write(mddev
))
4878 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4880 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4885 /* bitmap disabled, zero the first byte and copy out */
4886 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4887 file
->pathname
[0] = '\0';
4891 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4895 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4899 strcpy(file
->pathname
, ptr
);
4903 if (copy_to_user(arg
, file
, sizeof(*file
)))
4911 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4913 mdu_disk_info_t info
;
4916 if (copy_from_user(&info
, arg
, sizeof(info
)))
4919 rdev
= find_rdev_nr(mddev
, info
.number
);
4921 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4922 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4923 info
.raid_disk
= rdev
->raid_disk
;
4925 if (test_bit(Faulty
, &rdev
->flags
))
4926 info
.state
|= (1<<MD_DISK_FAULTY
);
4927 else if (test_bit(In_sync
, &rdev
->flags
)) {
4928 info
.state
|= (1<<MD_DISK_ACTIVE
);
4929 info
.state
|= (1<<MD_DISK_SYNC
);
4931 if (test_bit(WriteMostly
, &rdev
->flags
))
4932 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4934 info
.major
= info
.minor
= 0;
4935 info
.raid_disk
= -1;
4936 info
.state
= (1<<MD_DISK_REMOVED
);
4939 if (copy_to_user(arg
, &info
, sizeof(info
)))
4945 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4947 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4949 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4951 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4954 if (!mddev
->raid_disks
) {
4956 /* expecting a device which has a superblock */
4957 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4960 "md: md_import_device returned %ld\n",
4962 return PTR_ERR(rdev
);
4964 if (!list_empty(&mddev
->disks
)) {
4965 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4966 mdk_rdev_t
, same_set
);
4967 err
= super_types
[mddev
->major_version
]
4968 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4971 "md: %s has different UUID to %s\n",
4972 bdevname(rdev
->bdev
,b
),
4973 bdevname(rdev0
->bdev
,b2
));
4978 err
= bind_rdev_to_array(rdev
, mddev
);
4985 * add_new_disk can be used once the array is assembled
4986 * to add "hot spares". They must already have a superblock
4991 if (!mddev
->pers
->hot_add_disk
) {
4993 "%s: personality does not support diskops!\n",
4997 if (mddev
->persistent
)
4998 rdev
= md_import_device(dev
, mddev
->major_version
,
4999 mddev
->minor_version
);
5001 rdev
= md_import_device(dev
, -1, -1);
5004 "md: md_import_device returned %ld\n",
5006 return PTR_ERR(rdev
);
5008 /* set save_raid_disk if appropriate */
5009 if (!mddev
->persistent
) {
5010 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5011 info
->raid_disk
< mddev
->raid_disks
)
5012 rdev
->raid_disk
= info
->raid_disk
;
5014 rdev
->raid_disk
= -1;
5016 super_types
[mddev
->major_version
].
5017 validate_super(mddev
, rdev
);
5018 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5020 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5021 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5022 set_bit(WriteMostly
, &rdev
->flags
);
5024 clear_bit(WriteMostly
, &rdev
->flags
);
5026 rdev
->raid_disk
= -1;
5027 err
= bind_rdev_to_array(rdev
, mddev
);
5028 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5029 /* If there is hot_add_disk but no hot_remove_disk
5030 * then added disks for geometry changes,
5031 * and should be added immediately.
5033 super_types
[mddev
->major_version
].
5034 validate_super(mddev
, rdev
);
5035 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5037 unbind_rdev_from_array(rdev
);
5042 sysfs_notify_dirent(rdev
->sysfs_state
);
5044 md_update_sb(mddev
, 1);
5045 if (mddev
->degraded
)
5046 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5047 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5048 md_wakeup_thread(mddev
->thread
);
5052 /* otherwise, add_new_disk is only allowed
5053 * for major_version==0 superblocks
5055 if (mddev
->major_version
!= 0) {
5056 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5061 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5063 rdev
= md_import_device(dev
, -1, 0);
5066 "md: error, md_import_device() returned %ld\n",
5068 return PTR_ERR(rdev
);
5070 rdev
->desc_nr
= info
->number
;
5071 if (info
->raid_disk
< mddev
->raid_disks
)
5072 rdev
->raid_disk
= info
->raid_disk
;
5074 rdev
->raid_disk
= -1;
5076 if (rdev
->raid_disk
< mddev
->raid_disks
)
5077 if (info
->state
& (1<<MD_DISK_SYNC
))
5078 set_bit(In_sync
, &rdev
->flags
);
5080 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5081 set_bit(WriteMostly
, &rdev
->flags
);
5083 if (!mddev
->persistent
) {
5084 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5085 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5087 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5088 rdev
->sectors
= rdev
->sb_start
;
5090 err
= bind_rdev_to_array(rdev
, mddev
);
5100 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5102 char b
[BDEVNAME_SIZE
];
5105 rdev
= find_rdev(mddev
, dev
);
5109 if (rdev
->raid_disk
>= 0)
5112 kick_rdev_from_array(rdev
);
5113 md_update_sb(mddev
, 1);
5114 md_new_event(mddev
);
5118 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5119 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5123 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5125 char b
[BDEVNAME_SIZE
];
5132 if (mddev
->major_version
!= 0) {
5133 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5134 " version-0 superblocks.\n",
5138 if (!mddev
->pers
->hot_add_disk
) {
5140 "%s: personality does not support diskops!\n",
5145 rdev
= md_import_device(dev
, -1, 0);
5148 "md: error, md_import_device() returned %ld\n",
5153 if (mddev
->persistent
)
5154 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5156 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5158 rdev
->sectors
= rdev
->sb_start
;
5160 if (test_bit(Faulty
, &rdev
->flags
)) {
5162 "md: can not hot-add faulty %s disk to %s!\n",
5163 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5167 clear_bit(In_sync
, &rdev
->flags
);
5169 rdev
->saved_raid_disk
= -1;
5170 err
= bind_rdev_to_array(rdev
, mddev
);
5175 * The rest should better be atomic, we can have disk failures
5176 * noticed in interrupt contexts ...
5179 rdev
->raid_disk
= -1;
5181 md_update_sb(mddev
, 1);
5184 * Kick recovery, maybe this spare has to be added to the
5185 * array immediately.
5187 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5188 md_wakeup_thread(mddev
->thread
);
5189 md_new_event(mddev
);
5197 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5202 if (!mddev
->pers
->quiesce
)
5204 if (mddev
->recovery
|| mddev
->sync_thread
)
5206 /* we should be able to change the bitmap.. */
5212 return -EEXIST
; /* cannot add when bitmap is present */
5213 mddev
->bitmap_info
.file
= fget(fd
);
5215 if (mddev
->bitmap_info
.file
== NULL
) {
5216 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5221 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5223 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5225 fput(mddev
->bitmap_info
.file
);
5226 mddev
->bitmap_info
.file
= NULL
;
5229 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5230 } else if (mddev
->bitmap
== NULL
)
5231 return -ENOENT
; /* cannot remove what isn't there */
5234 mddev
->pers
->quiesce(mddev
, 1);
5236 err
= bitmap_create(mddev
);
5237 if (fd
< 0 || err
) {
5238 bitmap_destroy(mddev
);
5239 fd
= -1; /* make sure to put the file */
5241 mddev
->pers
->quiesce(mddev
, 0);
5244 if (mddev
->bitmap_info
.file
) {
5245 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5246 fput(mddev
->bitmap_info
.file
);
5248 mddev
->bitmap_info
.file
= NULL
;
5255 * set_array_info is used two different ways
5256 * The original usage is when creating a new array.
5257 * In this usage, raid_disks is > 0 and it together with
5258 * level, size, not_persistent,layout,chunksize determine the
5259 * shape of the array.
5260 * This will always create an array with a type-0.90.0 superblock.
5261 * The newer usage is when assembling an array.
5262 * In this case raid_disks will be 0, and the major_version field is
5263 * use to determine which style super-blocks are to be found on the devices.
5264 * The minor and patch _version numbers are also kept incase the
5265 * super_block handler wishes to interpret them.
5267 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5270 if (info
->raid_disks
== 0) {
5271 /* just setting version number for superblock loading */
5272 if (info
->major_version
< 0 ||
5273 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5274 super_types
[info
->major_version
].name
== NULL
) {
5275 /* maybe try to auto-load a module? */
5277 "md: superblock version %d not known\n",
5278 info
->major_version
);
5281 mddev
->major_version
= info
->major_version
;
5282 mddev
->minor_version
= info
->minor_version
;
5283 mddev
->patch_version
= info
->patch_version
;
5284 mddev
->persistent
= !info
->not_persistent
;
5285 /* ensure mddev_put doesn't delete this now that there
5286 * is some minimal configuration.
5288 mddev
->ctime
= get_seconds();
5291 mddev
->major_version
= MD_MAJOR_VERSION
;
5292 mddev
->minor_version
= MD_MINOR_VERSION
;
5293 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5294 mddev
->ctime
= get_seconds();
5296 mddev
->level
= info
->level
;
5297 mddev
->clevel
[0] = 0;
5298 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5299 mddev
->raid_disks
= info
->raid_disks
;
5300 /* don't set md_minor, it is determined by which /dev/md* was
5303 if (info
->state
& (1<<MD_SB_CLEAN
))
5304 mddev
->recovery_cp
= MaxSector
;
5306 mddev
->recovery_cp
= 0;
5307 mddev
->persistent
= ! info
->not_persistent
;
5308 mddev
->external
= 0;
5310 mddev
->layout
= info
->layout
;
5311 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5313 mddev
->max_disks
= MD_SB_DISKS
;
5315 if (mddev
->persistent
)
5317 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5319 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5320 mddev
->bitmap_info
.offset
= 0;
5322 mddev
->reshape_position
= MaxSector
;
5325 * Generate a 128 bit UUID
5327 get_random_bytes(mddev
->uuid
, 16);
5329 mddev
->new_level
= mddev
->level
;
5330 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5331 mddev
->new_layout
= mddev
->layout
;
5332 mddev
->delta_disks
= 0;
5337 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5339 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5341 if (mddev
->external_size
)
5344 mddev
->array_sectors
= array_sectors
;
5346 EXPORT_SYMBOL(md_set_array_sectors
);
5348 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5352 int fit
= (num_sectors
== 0);
5354 if (mddev
->pers
->resize
== NULL
)
5356 /* The "num_sectors" is the number of sectors of each device that
5357 * is used. This can only make sense for arrays with redundancy.
5358 * linear and raid0 always use whatever space is available. We can only
5359 * consider changing this number if no resync or reconstruction is
5360 * happening, and if the new size is acceptable. It must fit before the
5361 * sb_start or, if that is <data_offset, it must fit before the size
5362 * of each device. If num_sectors is zero, we find the largest size
5366 if (mddev
->sync_thread
)
5369 /* Sorry, cannot grow a bitmap yet, just remove it,
5373 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5374 sector_t avail
= rdev
->sectors
;
5376 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5377 num_sectors
= avail
;
5378 if (avail
< num_sectors
)
5381 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5383 revalidate_disk(mddev
->gendisk
);
5387 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5390 /* change the number of raid disks */
5391 if (mddev
->pers
->check_reshape
== NULL
)
5393 if (raid_disks
<= 0 ||
5394 raid_disks
>= mddev
->max_disks
)
5396 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5398 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5400 rv
= mddev
->pers
->check_reshape(mddev
);
5406 * update_array_info is used to change the configuration of an
5408 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5409 * fields in the info are checked against the array.
5410 * Any differences that cannot be handled will cause an error.
5411 * Normally, only one change can be managed at a time.
5413 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5419 /* calculate expected state,ignoring low bits */
5420 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5421 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5423 if (mddev
->major_version
!= info
->major_version
||
5424 mddev
->minor_version
!= info
->minor_version
||
5425 /* mddev->patch_version != info->patch_version || */
5426 mddev
->ctime
!= info
->ctime
||
5427 mddev
->level
!= info
->level
||
5428 /* mddev->layout != info->layout || */
5429 !mddev
->persistent
!= info
->not_persistent
||
5430 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5431 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5432 ((state
^info
->state
) & 0xfffffe00)
5435 /* Check there is only one change */
5436 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5438 if (mddev
->raid_disks
!= info
->raid_disks
)
5440 if (mddev
->layout
!= info
->layout
)
5442 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5449 if (mddev
->layout
!= info
->layout
) {
5451 * we don't need to do anything at the md level, the
5452 * personality will take care of it all.
5454 if (mddev
->pers
->check_reshape
== NULL
)
5457 mddev
->new_layout
= info
->layout
;
5458 rv
= mddev
->pers
->check_reshape(mddev
);
5460 mddev
->new_layout
= mddev
->layout
;
5464 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5465 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5467 if (mddev
->raid_disks
!= info
->raid_disks
)
5468 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5470 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5471 if (mddev
->pers
->quiesce
== NULL
)
5473 if (mddev
->recovery
|| mddev
->sync_thread
)
5475 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5476 /* add the bitmap */
5479 if (mddev
->bitmap_info
.default_offset
== 0)
5481 mddev
->bitmap_info
.offset
=
5482 mddev
->bitmap_info
.default_offset
;
5483 mddev
->pers
->quiesce(mddev
, 1);
5484 rv
= bitmap_create(mddev
);
5486 bitmap_destroy(mddev
);
5487 mddev
->pers
->quiesce(mddev
, 0);
5489 /* remove the bitmap */
5492 if (mddev
->bitmap
->file
)
5494 mddev
->pers
->quiesce(mddev
, 1);
5495 bitmap_destroy(mddev
);
5496 mddev
->pers
->quiesce(mddev
, 0);
5497 mddev
->bitmap_info
.offset
= 0;
5500 md_update_sb(mddev
, 1);
5504 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5508 if (mddev
->pers
== NULL
)
5511 rdev
= find_rdev(mddev
, dev
);
5515 md_error(mddev
, rdev
);
5520 * We have a problem here : there is no easy way to give a CHS
5521 * virtual geometry. We currently pretend that we have a 2 heads
5522 * 4 sectors (with a BIG number of cylinders...). This drives
5523 * dosfs just mad... ;-)
5525 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5527 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5531 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
5535 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5536 unsigned int cmd
, unsigned long arg
)
5539 void __user
*argp
= (void __user
*)arg
;
5540 mddev_t
*mddev
= NULL
;
5543 if (!capable(CAP_SYS_ADMIN
))
5547 * Commands dealing with the RAID driver but not any
5553 err
= get_version(argp
);
5556 case PRINT_RAID_DEBUG
:
5564 autostart_arrays(arg
);
5571 * Commands creating/starting a new array:
5574 mddev
= bdev
->bd_disk
->private_data
;
5581 err
= mddev_lock(mddev
);
5584 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5591 case SET_ARRAY_INFO
:
5593 mdu_array_info_t info
;
5595 memset(&info
, 0, sizeof(info
));
5596 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5601 err
= update_array_info(mddev
, &info
);
5603 printk(KERN_WARNING
"md: couldn't update"
5604 " array info. %d\n", err
);
5609 if (!list_empty(&mddev
->disks
)) {
5611 "md: array %s already has disks!\n",
5616 if (mddev
->raid_disks
) {
5618 "md: array %s already initialised!\n",
5623 err
= set_array_info(mddev
, &info
);
5625 printk(KERN_WARNING
"md: couldn't set"
5626 " array info. %d\n", err
);
5636 * Commands querying/configuring an existing array:
5638 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5639 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5640 if ((!mddev
->raid_disks
&& !mddev
->external
)
5641 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5642 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5643 && cmd
!= GET_BITMAP_FILE
) {
5649 * Commands even a read-only array can execute:
5653 case GET_ARRAY_INFO
:
5654 err
= get_array_info(mddev
, argp
);
5657 case GET_BITMAP_FILE
:
5658 err
= get_bitmap_file(mddev
, argp
);
5662 err
= get_disk_info(mddev
, argp
);
5665 case RESTART_ARRAY_RW
:
5666 err
= restart_array(mddev
);
5670 err
= do_md_stop(mddev
, 0, 1);
5674 err
= do_md_stop(mddev
, 1, 1);
5678 if (get_user(ro
, (int __user
*)(arg
))) {
5684 /* if the bdev is going readonly the value of mddev->ro
5685 * does not matter, no writes are coming
5690 /* are we are already prepared for writes? */
5694 /* transitioning to readauto need only happen for
5695 * arrays that call md_write_start
5698 err
= restart_array(mddev
);
5701 set_disk_ro(mddev
->gendisk
, 0);
5708 * The remaining ioctls are changing the state of the
5709 * superblock, so we do not allow them on read-only arrays.
5710 * However non-MD ioctls (e.g. get-size) will still come through
5711 * here and hit the 'default' below, so only disallow
5712 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5714 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5715 if (mddev
->ro
== 2) {
5717 sysfs_notify_dirent(mddev
->sysfs_state
);
5718 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5719 md_wakeup_thread(mddev
->thread
);
5730 mdu_disk_info_t info
;
5731 if (copy_from_user(&info
, argp
, sizeof(info
)))
5734 err
= add_new_disk(mddev
, &info
);
5738 case HOT_REMOVE_DISK
:
5739 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5743 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5746 case SET_DISK_FAULTY
:
5747 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5751 err
= do_md_run(mddev
);
5754 case SET_BITMAP_FILE
:
5755 err
= set_bitmap_file(mddev
, (int)arg
);
5765 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5767 mddev
->hold_active
= 0;
5768 mddev_unlock(mddev
);
5777 #ifdef CONFIG_COMPAT
5778 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
5779 unsigned int cmd
, unsigned long arg
)
5782 case HOT_REMOVE_DISK
:
5784 case SET_DISK_FAULTY
:
5785 case SET_BITMAP_FILE
:
5786 /* These take in integer arg, do not convert */
5789 arg
= (unsigned long)compat_ptr(arg
);
5793 return md_ioctl(bdev
, mode
, cmd
, arg
);
5795 #endif /* CONFIG_COMPAT */
5797 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5800 * Succeed if we can lock the mddev, which confirms that
5801 * it isn't being stopped right now.
5803 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5806 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5807 /* we are racing with mddev_put which is discarding this
5811 /* Wait until bdev->bd_disk is definitely gone */
5812 flush_scheduled_work();
5813 /* Then retry the open from the top */
5814 return -ERESTARTSYS
;
5816 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5818 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5822 atomic_inc(&mddev
->openers
);
5823 mutex_unlock(&mddev
->open_mutex
);
5825 check_disk_change(bdev
);
5830 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5832 mddev_t
*mddev
= disk
->private_data
;
5835 atomic_dec(&mddev
->openers
);
5841 static int md_media_changed(struct gendisk
*disk
)
5843 mddev_t
*mddev
= disk
->private_data
;
5845 return mddev
->changed
;
5848 static int md_revalidate(struct gendisk
*disk
)
5850 mddev_t
*mddev
= disk
->private_data
;
5855 static const struct block_device_operations md_fops
=
5857 .owner
= THIS_MODULE
,
5859 .release
= md_release
,
5861 #ifdef CONFIG_COMPAT
5862 .compat_ioctl
= md_compat_ioctl
,
5864 .getgeo
= md_getgeo
,
5865 .media_changed
= md_media_changed
,
5866 .revalidate_disk
= md_revalidate
,
5869 static int md_thread(void * arg
)
5871 mdk_thread_t
*thread
= arg
;
5874 * md_thread is a 'system-thread', it's priority should be very
5875 * high. We avoid resource deadlocks individually in each
5876 * raid personality. (RAID5 does preallocation) We also use RR and
5877 * the very same RT priority as kswapd, thus we will never get
5878 * into a priority inversion deadlock.
5880 * we definitely have to have equal or higher priority than
5881 * bdflush, otherwise bdflush will deadlock if there are too
5882 * many dirty RAID5 blocks.
5885 allow_signal(SIGKILL
);
5886 while (!kthread_should_stop()) {
5888 /* We need to wait INTERRUPTIBLE so that
5889 * we don't add to the load-average.
5890 * That means we need to be sure no signals are
5893 if (signal_pending(current
))
5894 flush_signals(current
);
5896 wait_event_interruptible_timeout
5898 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5899 || kthread_should_stop(),
5902 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5904 thread
->run(thread
->mddev
);
5910 void md_wakeup_thread(mdk_thread_t
*thread
)
5913 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5914 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5915 wake_up(&thread
->wqueue
);
5919 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5922 mdk_thread_t
*thread
;
5924 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5928 init_waitqueue_head(&thread
->wqueue
);
5931 thread
->mddev
= mddev
;
5932 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5933 thread
->tsk
= kthread_run(md_thread
, thread
,
5935 mdname(thread
->mddev
),
5936 name
?: mddev
->pers
->name
);
5937 if (IS_ERR(thread
->tsk
)) {
5944 void md_unregister_thread(mdk_thread_t
*thread
)
5948 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5950 kthread_stop(thread
->tsk
);
5954 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5961 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5964 if (mddev
->external
)
5965 set_bit(Blocked
, &rdev
->flags
);
5967 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5969 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5970 __builtin_return_address(0),__builtin_return_address(1),
5971 __builtin_return_address(2),__builtin_return_address(3));
5975 if (!mddev
->pers
->error_handler
)
5977 mddev
->pers
->error_handler(mddev
,rdev
);
5978 if (mddev
->degraded
)
5979 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5980 set_bit(StateChanged
, &rdev
->flags
);
5981 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5982 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5983 md_wakeup_thread(mddev
->thread
);
5984 md_new_event_inintr(mddev
);
5987 /* seq_file implementation /proc/mdstat */
5989 static void status_unused(struct seq_file
*seq
)
5994 seq_printf(seq
, "unused devices: ");
5996 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
5997 char b
[BDEVNAME_SIZE
];
5999 seq_printf(seq
, "%s ",
6000 bdevname(rdev
->bdev
,b
));
6003 seq_printf(seq
, "<none>");
6005 seq_printf(seq
, "\n");
6009 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6011 sector_t max_sectors
, resync
, res
;
6012 unsigned long dt
, db
;
6015 unsigned int per_milli
;
6017 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6019 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6020 max_sectors
= mddev
->resync_max_sectors
;
6022 max_sectors
= mddev
->dev_sectors
;
6025 * Should not happen.
6031 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6032 * in a sector_t, and (max_sectors>>scale) will fit in a
6033 * u32, as those are the requirements for sector_div.
6034 * Thus 'scale' must be at least 10
6037 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6038 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6041 res
= (resync
>>scale
)*1000;
6042 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6046 int i
, x
= per_milli
/50, y
= 20-x
;
6047 seq_printf(seq
, "[");
6048 for (i
= 0; i
< x
; i
++)
6049 seq_printf(seq
, "=");
6050 seq_printf(seq
, ">");
6051 for (i
= 0; i
< y
; i
++)
6052 seq_printf(seq
, ".");
6053 seq_printf(seq
, "] ");
6055 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6056 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6058 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6060 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6061 "resync" : "recovery"))),
6062 per_milli
/10, per_milli
% 10,
6063 (unsigned long long) resync
/2,
6064 (unsigned long long) max_sectors
/2);
6067 * dt: time from mark until now
6068 * db: blocks written from mark until now
6069 * rt: remaining time
6071 * rt is a sector_t, so could be 32bit or 64bit.
6072 * So we divide before multiply in case it is 32bit and close
6074 * We scale the divisor (db) by 32 to avoid loosing precision
6075 * near the end of resync when the number of remaining sectors
6077 * We then divide rt by 32 after multiplying by db to compensate.
6078 * The '+1' avoids division by zero if db is very small.
6080 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6082 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6083 - mddev
->resync_mark_cnt
;
6085 rt
= max_sectors
- resync
; /* number of remaining sectors */
6086 sector_div(rt
, db
/32+1);
6090 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6091 ((unsigned long)rt
% 60)/6);
6093 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6096 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6098 struct list_head
*tmp
;
6108 spin_lock(&all_mddevs_lock
);
6109 list_for_each(tmp
,&all_mddevs
)
6111 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6113 spin_unlock(&all_mddevs_lock
);
6116 spin_unlock(&all_mddevs_lock
);
6118 return (void*)2;/* tail */
6122 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6124 struct list_head
*tmp
;
6125 mddev_t
*next_mddev
, *mddev
= v
;
6131 spin_lock(&all_mddevs_lock
);
6133 tmp
= all_mddevs
.next
;
6135 tmp
= mddev
->all_mddevs
.next
;
6136 if (tmp
!= &all_mddevs
)
6137 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6139 next_mddev
= (void*)2;
6142 spin_unlock(&all_mddevs_lock
);
6150 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6154 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6158 struct mdstat_info
{
6162 static int md_seq_show(struct seq_file
*seq
, void *v
)
6167 struct mdstat_info
*mi
= seq
->private;
6168 struct bitmap
*bitmap
;
6170 if (v
== (void*)1) {
6171 struct mdk_personality
*pers
;
6172 seq_printf(seq
, "Personalities : ");
6173 spin_lock(&pers_lock
);
6174 list_for_each_entry(pers
, &pers_list
, list
)
6175 seq_printf(seq
, "[%s] ", pers
->name
);
6177 spin_unlock(&pers_lock
);
6178 seq_printf(seq
, "\n");
6179 mi
->event
= atomic_read(&md_event_count
);
6182 if (v
== (void*)2) {
6187 if (mddev_lock(mddev
) < 0)
6190 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6191 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6192 mddev
->pers
? "" : "in");
6195 seq_printf(seq
, " (read-only)");
6197 seq_printf(seq
, " (auto-read-only)");
6198 seq_printf(seq
, " %s", mddev
->pers
->name
);
6202 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6203 char b
[BDEVNAME_SIZE
];
6204 seq_printf(seq
, " %s[%d]",
6205 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6206 if (test_bit(WriteMostly
, &rdev
->flags
))
6207 seq_printf(seq
, "(W)");
6208 if (test_bit(Faulty
, &rdev
->flags
)) {
6209 seq_printf(seq
, "(F)");
6211 } else if (rdev
->raid_disk
< 0)
6212 seq_printf(seq
, "(S)"); /* spare */
6213 sectors
+= rdev
->sectors
;
6216 if (!list_empty(&mddev
->disks
)) {
6218 seq_printf(seq
, "\n %llu blocks",
6219 (unsigned long long)
6220 mddev
->array_sectors
/ 2);
6222 seq_printf(seq
, "\n %llu blocks",
6223 (unsigned long long)sectors
/ 2);
6225 if (mddev
->persistent
) {
6226 if (mddev
->major_version
!= 0 ||
6227 mddev
->minor_version
!= 90) {
6228 seq_printf(seq
," super %d.%d",
6229 mddev
->major_version
,
6230 mddev
->minor_version
);
6232 } else if (mddev
->external
)
6233 seq_printf(seq
, " super external:%s",
6234 mddev
->metadata_type
);
6236 seq_printf(seq
, " super non-persistent");
6239 mddev
->pers
->status(seq
, mddev
);
6240 seq_printf(seq
, "\n ");
6241 if (mddev
->pers
->sync_request
) {
6242 if (mddev
->curr_resync
> 2) {
6243 status_resync(seq
, mddev
);
6244 seq_printf(seq
, "\n ");
6245 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6246 seq_printf(seq
, "\tresync=DELAYED\n ");
6247 else if (mddev
->recovery_cp
< MaxSector
)
6248 seq_printf(seq
, "\tresync=PENDING\n ");
6251 seq_printf(seq
, "\n ");
6253 if ((bitmap
= mddev
->bitmap
)) {
6254 unsigned long chunk_kb
;
6255 unsigned long flags
;
6256 spin_lock_irqsave(&bitmap
->lock
, flags
);
6257 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6258 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6260 bitmap
->pages
- bitmap
->missing_pages
,
6262 (bitmap
->pages
- bitmap
->missing_pages
)
6263 << (PAGE_SHIFT
- 10),
6264 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6265 chunk_kb
? "KB" : "B");
6267 seq_printf(seq
, ", file: ");
6268 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6271 seq_printf(seq
, "\n");
6272 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6275 seq_printf(seq
, "\n");
6277 mddev_unlock(mddev
);
6282 static const struct seq_operations md_seq_ops
= {
6283 .start
= md_seq_start
,
6284 .next
= md_seq_next
,
6285 .stop
= md_seq_stop
,
6286 .show
= md_seq_show
,
6289 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6292 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6296 error
= seq_open(file
, &md_seq_ops
);
6300 struct seq_file
*p
= file
->private_data
;
6302 mi
->event
= atomic_read(&md_event_count
);
6307 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6309 struct seq_file
*m
= filp
->private_data
;
6310 struct mdstat_info
*mi
= m
->private;
6313 poll_wait(filp
, &md_event_waiters
, wait
);
6315 /* always allow read */
6316 mask
= POLLIN
| POLLRDNORM
;
6318 if (mi
->event
!= atomic_read(&md_event_count
))
6319 mask
|= POLLERR
| POLLPRI
;
6323 static const struct file_operations md_seq_fops
= {
6324 .owner
= THIS_MODULE
,
6325 .open
= md_seq_open
,
6327 .llseek
= seq_lseek
,
6328 .release
= seq_release_private
,
6329 .poll
= mdstat_poll
,
6332 int register_md_personality(struct mdk_personality
*p
)
6334 spin_lock(&pers_lock
);
6335 list_add_tail(&p
->list
, &pers_list
);
6336 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6337 spin_unlock(&pers_lock
);
6341 int unregister_md_personality(struct mdk_personality
*p
)
6343 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6344 spin_lock(&pers_lock
);
6345 list_del_init(&p
->list
);
6346 spin_unlock(&pers_lock
);
6350 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6358 rdev_for_each_rcu(rdev
, mddev
) {
6359 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6360 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6361 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6362 atomic_read(&disk
->sync_io
);
6363 /* sync IO will cause sync_io to increase before the disk_stats
6364 * as sync_io is counted when a request starts, and
6365 * disk_stats is counted when it completes.
6366 * So resync activity will cause curr_events to be smaller than
6367 * when there was no such activity.
6368 * non-sync IO will cause disk_stat to increase without
6369 * increasing sync_io so curr_events will (eventually)
6370 * be larger than it was before. Once it becomes
6371 * substantially larger, the test below will cause
6372 * the array to appear non-idle, and resync will slow
6374 * If there is a lot of outstanding resync activity when
6375 * we set last_event to curr_events, then all that activity
6376 * completing might cause the array to appear non-idle
6377 * and resync will be slowed down even though there might
6378 * not have been non-resync activity. This will only
6379 * happen once though. 'last_events' will soon reflect
6380 * the state where there is little or no outstanding
6381 * resync requests, and further resync activity will
6382 * always make curr_events less than last_events.
6385 if (init
|| curr_events
- rdev
->last_events
> 64) {
6386 rdev
->last_events
= curr_events
;
6394 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6396 /* another "blocks" (512byte) blocks have been synced */
6397 atomic_sub(blocks
, &mddev
->recovery_active
);
6398 wake_up(&mddev
->recovery_wait
);
6400 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6401 md_wakeup_thread(mddev
->thread
);
6402 // stop recovery, signal do_sync ....
6407 /* md_write_start(mddev, bi)
6408 * If we need to update some array metadata (e.g. 'active' flag
6409 * in superblock) before writing, schedule a superblock update
6410 * and wait for it to complete.
6412 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6415 if (bio_data_dir(bi
) != WRITE
)
6418 BUG_ON(mddev
->ro
== 1);
6419 if (mddev
->ro
== 2) {
6420 /* need to switch to read/write */
6422 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6423 md_wakeup_thread(mddev
->thread
);
6424 md_wakeup_thread(mddev
->sync_thread
);
6427 atomic_inc(&mddev
->writes_pending
);
6428 if (mddev
->safemode
== 1)
6429 mddev
->safemode
= 0;
6430 if (mddev
->in_sync
) {
6431 spin_lock_irq(&mddev
->write_lock
);
6432 if (mddev
->in_sync
) {
6434 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6435 md_wakeup_thread(mddev
->thread
);
6438 spin_unlock_irq(&mddev
->write_lock
);
6441 sysfs_notify_dirent(mddev
->sysfs_state
);
6442 wait_event(mddev
->sb_wait
,
6443 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6444 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6447 void md_write_end(mddev_t
*mddev
)
6449 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6450 if (mddev
->safemode
== 2)
6451 md_wakeup_thread(mddev
->thread
);
6452 else if (mddev
->safemode_delay
)
6453 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6457 /* md_allow_write(mddev)
6458 * Calling this ensures that the array is marked 'active' so that writes
6459 * may proceed without blocking. It is important to call this before
6460 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6461 * Must be called with mddev_lock held.
6463 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6464 * is dropped, so return -EAGAIN after notifying userspace.
6466 int md_allow_write(mddev_t
*mddev
)
6472 if (!mddev
->pers
->sync_request
)
6475 spin_lock_irq(&mddev
->write_lock
);
6476 if (mddev
->in_sync
) {
6478 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6479 if (mddev
->safemode_delay
&&
6480 mddev
->safemode
== 0)
6481 mddev
->safemode
= 1;
6482 spin_unlock_irq(&mddev
->write_lock
);
6483 md_update_sb(mddev
, 0);
6484 sysfs_notify_dirent(mddev
->sysfs_state
);
6486 spin_unlock_irq(&mddev
->write_lock
);
6488 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6493 EXPORT_SYMBOL_GPL(md_allow_write
);
6495 #define SYNC_MARKS 10
6496 #define SYNC_MARK_STEP (3*HZ)
6497 void md_do_sync(mddev_t
*mddev
)
6500 unsigned int currspeed
= 0,
6502 sector_t max_sectors
,j
, io_sectors
;
6503 unsigned long mark
[SYNC_MARKS
];
6504 sector_t mark_cnt
[SYNC_MARKS
];
6506 struct list_head
*tmp
;
6507 sector_t last_check
;
6512 /* just incase thread restarts... */
6513 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6515 if (mddev
->ro
) /* never try to sync a read-only array */
6518 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6519 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6520 desc
= "data-check";
6521 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6522 desc
= "requested-resync";
6525 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6530 /* we overload curr_resync somewhat here.
6531 * 0 == not engaged in resync at all
6532 * 2 == checking that there is no conflict with another sync
6533 * 1 == like 2, but have yielded to allow conflicting resync to
6535 * other == active in resync - this many blocks
6537 * Before starting a resync we must have set curr_resync to
6538 * 2, and then checked that every "conflicting" array has curr_resync
6539 * less than ours. When we find one that is the same or higher
6540 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6541 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6542 * This will mean we have to start checking from the beginning again.
6547 mddev
->curr_resync
= 2;
6550 if (kthread_should_stop())
6551 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6553 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6555 for_each_mddev(mddev2
, tmp
) {
6556 if (mddev2
== mddev
)
6558 if (!mddev
->parallel_resync
6559 && mddev2
->curr_resync
6560 && match_mddev_units(mddev
, mddev2
)) {
6562 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6563 /* arbitrarily yield */
6564 mddev
->curr_resync
= 1;
6565 wake_up(&resync_wait
);
6567 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6568 /* no need to wait here, we can wait the next
6569 * time 'round when curr_resync == 2
6572 /* We need to wait 'interruptible' so as not to
6573 * contribute to the load average, and not to
6574 * be caught by 'softlockup'
6576 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6577 if (!kthread_should_stop() &&
6578 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6579 printk(KERN_INFO
"md: delaying %s of %s"
6580 " until %s has finished (they"
6581 " share one or more physical units)\n",
6582 desc
, mdname(mddev
), mdname(mddev2
));
6584 if (signal_pending(current
))
6585 flush_signals(current
);
6587 finish_wait(&resync_wait
, &wq
);
6590 finish_wait(&resync_wait
, &wq
);
6593 } while (mddev
->curr_resync
< 2);
6596 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6597 /* resync follows the size requested by the personality,
6598 * which defaults to physical size, but can be virtual size
6600 max_sectors
= mddev
->resync_max_sectors
;
6601 mddev
->resync_mismatches
= 0;
6602 /* we don't use the checkpoint if there's a bitmap */
6603 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6604 j
= mddev
->resync_min
;
6605 else if (!mddev
->bitmap
)
6606 j
= mddev
->recovery_cp
;
6608 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6609 max_sectors
= mddev
->dev_sectors
;
6611 /* recovery follows the physical size of devices */
6612 max_sectors
= mddev
->dev_sectors
;
6615 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6616 if (rdev
->raid_disk
>= 0 &&
6617 !test_bit(Faulty
, &rdev
->flags
) &&
6618 !test_bit(In_sync
, &rdev
->flags
) &&
6619 rdev
->recovery_offset
< j
)
6620 j
= rdev
->recovery_offset
;
6624 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6625 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6626 " %d KB/sec/disk.\n", speed_min(mddev
));
6627 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6628 "(but not more than %d KB/sec) for %s.\n",
6629 speed_max(mddev
), desc
);
6631 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6634 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6636 mark_cnt
[m
] = io_sectors
;
6639 mddev
->resync_mark
= mark
[last_mark
];
6640 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6643 * Tune reconstruction:
6645 window
= 32*(PAGE_SIZE
/512);
6646 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6647 window
/2,(unsigned long long) max_sectors
/2);
6649 atomic_set(&mddev
->recovery_active
, 0);
6654 "md: resuming %s of %s from checkpoint.\n",
6655 desc
, mdname(mddev
));
6656 mddev
->curr_resync
= j
;
6658 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6660 while (j
< max_sectors
) {
6665 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6666 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6667 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6668 > (max_sectors
>> 4)) ||
6669 (j
- mddev
->curr_resync_completed
)*2
6670 >= mddev
->resync_max
- mddev
->curr_resync_completed
6672 /* time to update curr_resync_completed */
6673 blk_unplug(mddev
->queue
);
6674 wait_event(mddev
->recovery_wait
,
6675 atomic_read(&mddev
->recovery_active
) == 0);
6676 mddev
->curr_resync_completed
=
6678 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6679 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6682 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6683 /* As this condition is controlled by user-space,
6684 * we can block indefinitely, so use '_interruptible'
6685 * to avoid triggering warnings.
6687 flush_signals(current
); /* just in case */
6688 wait_event_interruptible(mddev
->recovery_wait
,
6689 mddev
->resync_max
> j
6690 || kthread_should_stop());
6693 if (kthread_should_stop())
6696 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6697 currspeed
< speed_min(mddev
));
6699 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6703 if (!skipped
) { /* actual IO requested */
6704 io_sectors
+= sectors
;
6705 atomic_add(sectors
, &mddev
->recovery_active
);
6709 if (j
>1) mddev
->curr_resync
= j
;
6710 mddev
->curr_mark_cnt
= io_sectors
;
6711 if (last_check
== 0)
6712 /* this is the earliers that rebuilt will be
6713 * visible in /proc/mdstat
6715 md_new_event(mddev
);
6717 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6720 last_check
= io_sectors
;
6722 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6726 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6728 int next
= (last_mark
+1) % SYNC_MARKS
;
6730 mddev
->resync_mark
= mark
[next
];
6731 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6732 mark
[next
] = jiffies
;
6733 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6738 if (kthread_should_stop())
6743 * this loop exits only if either when we are slower than
6744 * the 'hard' speed limit, or the system was IO-idle for
6746 * the system might be non-idle CPU-wise, but we only care
6747 * about not overloading the IO subsystem. (things like an
6748 * e2fsck being done on the RAID array should execute fast)
6750 blk_unplug(mddev
->queue
);
6753 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6754 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6756 if (currspeed
> speed_min(mddev
)) {
6757 if ((currspeed
> speed_max(mddev
)) ||
6758 !is_mddev_idle(mddev
, 0)) {
6764 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6766 * this also signals 'finished resyncing' to md_stop
6769 blk_unplug(mddev
->queue
);
6771 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6773 /* tell personality that we are finished */
6774 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6776 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6777 mddev
->curr_resync
> 2) {
6778 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6779 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6780 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6782 "md: checkpointing %s of %s.\n",
6783 desc
, mdname(mddev
));
6784 mddev
->recovery_cp
= mddev
->curr_resync
;
6787 mddev
->recovery_cp
= MaxSector
;
6789 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6790 mddev
->curr_resync
= MaxSector
;
6792 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6793 if (rdev
->raid_disk
>= 0 &&
6794 !test_bit(Faulty
, &rdev
->flags
) &&
6795 !test_bit(In_sync
, &rdev
->flags
) &&
6796 rdev
->recovery_offset
< mddev
->curr_resync
)
6797 rdev
->recovery_offset
= mddev
->curr_resync
;
6801 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6804 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6805 /* We completed so min/max setting can be forgotten if used. */
6806 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6807 mddev
->resync_min
= 0;
6808 mddev
->resync_max
= MaxSector
;
6809 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6810 mddev
->resync_min
= mddev
->curr_resync_completed
;
6811 mddev
->curr_resync
= 0;
6812 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6813 mddev
->curr_resync_completed
= 0;
6814 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6815 wake_up(&resync_wait
);
6816 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6817 md_wakeup_thread(mddev
->thread
);
6822 * got a signal, exit.
6825 "md: md_do_sync() got signal ... exiting\n");
6826 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6830 EXPORT_SYMBOL_GPL(md_do_sync
);
6833 static int remove_and_add_spares(mddev_t
*mddev
)
6838 mddev
->curr_resync_completed
= 0;
6840 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6841 if (rdev
->raid_disk
>= 0 &&
6842 !test_bit(Blocked
, &rdev
->flags
) &&
6843 (test_bit(Faulty
, &rdev
->flags
) ||
6844 ! test_bit(In_sync
, &rdev
->flags
)) &&
6845 atomic_read(&rdev
->nr_pending
)==0) {
6846 if (mddev
->pers
->hot_remove_disk(
6847 mddev
, rdev
->raid_disk
)==0) {
6849 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6850 sysfs_remove_link(&mddev
->kobj
, nm
);
6851 rdev
->raid_disk
= -1;
6855 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6856 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6857 if (rdev
->raid_disk
>= 0 &&
6858 !test_bit(In_sync
, &rdev
->flags
) &&
6859 !test_bit(Blocked
, &rdev
->flags
))
6861 if (rdev
->raid_disk
< 0
6862 && !test_bit(Faulty
, &rdev
->flags
)) {
6863 rdev
->recovery_offset
= 0;
6865 hot_add_disk(mddev
, rdev
) == 0) {
6867 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6868 if (sysfs_create_link(&mddev
->kobj
,
6871 "md: cannot register "
6875 md_new_event(mddev
);
6876 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6885 * This routine is regularly called by all per-raid-array threads to
6886 * deal with generic issues like resync and super-block update.
6887 * Raid personalities that don't have a thread (linear/raid0) do not
6888 * need this as they never do any recovery or update the superblock.
6890 * It does not do any resync itself, but rather "forks" off other threads
6891 * to do that as needed.
6892 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6893 * "->recovery" and create a thread at ->sync_thread.
6894 * When the thread finishes it sets MD_RECOVERY_DONE
6895 * and wakeups up this thread which will reap the thread and finish up.
6896 * This thread also removes any faulty devices (with nr_pending == 0).
6898 * The overall approach is:
6899 * 1/ if the superblock needs updating, update it.
6900 * 2/ If a recovery thread is running, don't do anything else.
6901 * 3/ If recovery has finished, clean up, possibly marking spares active.
6902 * 4/ If there are any faulty devices, remove them.
6903 * 5/ If array is degraded, try to add spares devices
6904 * 6/ If array has spares or is not in-sync, start a resync thread.
6906 void md_check_recovery(mddev_t
*mddev
)
6912 bitmap_daemon_work(mddev
);
6917 if (signal_pending(current
)) {
6918 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6919 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6921 mddev
->safemode
= 2;
6923 flush_signals(current
);
6926 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6929 (mddev
->flags
&& !mddev
->external
) ||
6930 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6931 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6932 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6933 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6934 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6938 if (mddev_trylock(mddev
)) {
6942 /* Only thing we do on a ro array is remove
6945 remove_and_add_spares(mddev
);
6946 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6950 if (!mddev
->external
) {
6952 spin_lock_irq(&mddev
->write_lock
);
6953 if (mddev
->safemode
&&
6954 !atomic_read(&mddev
->writes_pending
) &&
6956 mddev
->recovery_cp
== MaxSector
) {
6959 if (mddev
->persistent
)
6960 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6962 if (mddev
->safemode
== 1)
6963 mddev
->safemode
= 0;
6964 spin_unlock_irq(&mddev
->write_lock
);
6966 sysfs_notify_dirent(mddev
->sysfs_state
);
6970 md_update_sb(mddev
, 0);
6972 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6973 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6974 sysfs_notify_dirent(rdev
->sysfs_state
);
6977 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6978 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6979 /* resync/recovery still happening */
6980 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6983 if (mddev
->sync_thread
) {
6984 /* resync has finished, collect result */
6985 md_unregister_thread(mddev
->sync_thread
);
6986 mddev
->sync_thread
= NULL
;
6987 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6988 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6990 /* activate any spares */
6991 if (mddev
->pers
->spare_active(mddev
))
6992 sysfs_notify(&mddev
->kobj
, NULL
,
6995 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6996 mddev
->pers
->finish_reshape
)
6997 mddev
->pers
->finish_reshape(mddev
);
6998 md_update_sb(mddev
, 1);
7000 /* if array is no-longer degraded, then any saved_raid_disk
7001 * information must be scrapped
7003 if (!mddev
->degraded
)
7004 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7005 rdev
->saved_raid_disk
= -1;
7007 mddev
->recovery
= 0;
7008 /* flag recovery needed just to double check */
7009 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7010 sysfs_notify_dirent(mddev
->sysfs_action
);
7011 md_new_event(mddev
);
7014 /* Set RUNNING before clearing NEEDED to avoid
7015 * any transients in the value of "sync_action".
7017 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7018 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7019 /* Clear some bits that don't mean anything, but
7022 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7023 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7025 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7027 /* no recovery is running.
7028 * remove any failed drives, then
7029 * add spares if possible.
7030 * Spare are also removed and re-added, to allow
7031 * the personality to fail the re-add.
7034 if (mddev
->reshape_position
!= MaxSector
) {
7035 if (mddev
->pers
->check_reshape
== NULL
||
7036 mddev
->pers
->check_reshape(mddev
) != 0)
7037 /* Cannot proceed */
7039 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7040 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7041 } else if ((spares
= remove_and_add_spares(mddev
))) {
7042 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7043 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7044 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7045 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7046 } else if (mddev
->recovery_cp
< MaxSector
) {
7047 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7048 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7049 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7050 /* nothing to be done ... */
7053 if (mddev
->pers
->sync_request
) {
7054 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7055 /* We are adding a device or devices to an array
7056 * which has the bitmap stored on all devices.
7057 * So make sure all bitmap pages get written
7059 bitmap_write_all(mddev
->bitmap
);
7061 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7064 if (!mddev
->sync_thread
) {
7065 printk(KERN_ERR
"%s: could not start resync"
7068 /* leave the spares where they are, it shouldn't hurt */
7069 mddev
->recovery
= 0;
7071 md_wakeup_thread(mddev
->sync_thread
);
7072 sysfs_notify_dirent(mddev
->sysfs_action
);
7073 md_new_event(mddev
);
7076 if (!mddev
->sync_thread
) {
7077 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7078 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7080 if (mddev
->sysfs_action
)
7081 sysfs_notify_dirent(mddev
->sysfs_action
);
7083 mddev_unlock(mddev
);
7087 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7089 sysfs_notify_dirent(rdev
->sysfs_state
);
7090 wait_event_timeout(rdev
->blocked_wait
,
7091 !test_bit(Blocked
, &rdev
->flags
),
7092 msecs_to_jiffies(5000));
7093 rdev_dec_pending(rdev
, mddev
);
7095 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7097 static int md_notify_reboot(struct notifier_block
*this,
7098 unsigned long code
, void *x
)
7100 struct list_head
*tmp
;
7103 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7105 printk(KERN_INFO
"md: stopping all md devices.\n");
7107 for_each_mddev(mddev
, tmp
)
7108 if (mddev_trylock(mddev
)) {
7109 /* Force a switch to readonly even array
7110 * appears to still be in use. Hence
7113 do_md_stop(mddev
, 1, 100);
7114 mddev_unlock(mddev
);
7117 * certain more exotic SCSI devices are known to be
7118 * volatile wrt too early system reboots. While the
7119 * right place to handle this issue is the given
7120 * driver, we do want to have a safe RAID driver ...
7127 static struct notifier_block md_notifier
= {
7128 .notifier_call
= md_notify_reboot
,
7130 .priority
= INT_MAX
, /* before any real devices */
7133 static void md_geninit(void)
7135 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
7137 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
7140 static int __init
md_init(void)
7142 if (register_blkdev(MD_MAJOR
, "md"))
7144 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
7145 unregister_blkdev(MD_MAJOR
, "md");
7148 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7149 md_probe
, NULL
, NULL
);
7150 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7151 md_probe
, NULL
, NULL
);
7153 register_reboot_notifier(&md_notifier
);
7154 raid_table_header
= register_sysctl_table(raid_root_table
);
7164 * Searches all registered partitions for autorun RAID arrays
7168 static LIST_HEAD(all_detected_devices
);
7169 struct detected_devices_node
{
7170 struct list_head list
;
7174 void md_autodetect_dev(dev_t dev
)
7176 struct detected_devices_node
*node_detected_dev
;
7178 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7179 if (node_detected_dev
) {
7180 node_detected_dev
->dev
= dev
;
7181 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7183 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7184 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7189 static void autostart_arrays(int part
)
7192 struct detected_devices_node
*node_detected_dev
;
7194 int i_scanned
, i_passed
;
7199 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7201 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7203 node_detected_dev
= list_entry(all_detected_devices
.next
,
7204 struct detected_devices_node
, list
);
7205 list_del(&node_detected_dev
->list
);
7206 dev
= node_detected_dev
->dev
;
7207 kfree(node_detected_dev
);
7208 rdev
= md_import_device(dev
,0, 90);
7212 if (test_bit(Faulty
, &rdev
->flags
)) {
7216 set_bit(AutoDetected
, &rdev
->flags
);
7217 list_add(&rdev
->same_set
, &pending_raid_disks
);
7221 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7222 i_scanned
, i_passed
);
7224 autorun_devices(part
);
7227 #endif /* !MODULE */
7229 static __exit
void md_exit(void)
7232 struct list_head
*tmp
;
7234 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7235 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7237 unregister_blkdev(MD_MAJOR
,"md");
7238 unregister_blkdev(mdp_major
, "mdp");
7239 unregister_reboot_notifier(&md_notifier
);
7240 unregister_sysctl_table(raid_table_header
);
7241 remove_proc_entry("mdstat", NULL
);
7242 for_each_mddev(mddev
, tmp
) {
7243 export_array(mddev
);
7244 mddev
->hold_active
= 0;
7248 subsys_initcall(md_init
);
7249 module_exit(md_exit
)
7251 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7253 return sprintf(buffer
, "%d", start_readonly
);
7255 static int set_ro(const char *val
, struct kernel_param
*kp
)
7258 int num
= simple_strtoul(val
, &e
, 10);
7259 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7260 start_readonly
= num
;
7266 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7267 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7269 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7271 EXPORT_SYMBOL(register_md_personality
);
7272 EXPORT_SYMBOL(unregister_md_personality
);
7273 EXPORT_SYMBOL(md_error
);
7274 EXPORT_SYMBOL(md_done_sync
);
7275 EXPORT_SYMBOL(md_write_start
);
7276 EXPORT_SYMBOL(md_write_end
);
7277 EXPORT_SYMBOL(md_register_thread
);
7278 EXPORT_SYMBOL(md_unregister_thread
);
7279 EXPORT_SYMBOL(md_wakeup_thread
);
7280 EXPORT_SYMBOL(md_check_recovery
);
7281 MODULE_LICENSE("GPL");
7282 MODULE_DESCRIPTION("MD RAID framework");
7284 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);