2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
60 static void autostart_arrays(int part
);
63 static LIST_HEAD(pers_list
);
64 static DEFINE_SPINLOCK(pers_lock
);
66 static void md_print_devices(void);
68 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
70 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
73 * Default number of read corrections we'll attempt on an rdev
74 * before ejecting it from the array. We divide the read error
75 * count by 2 for every hour elapsed between read errors.
77 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
91 static int sysctl_speed_limit_min
= 1000;
92 static int sysctl_speed_limit_max
= 200000;
93 static inline int speed_min(mddev_t
*mddev
)
95 return mddev
->sync_speed_min
?
96 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
99 static inline int speed_max(mddev_t
*mddev
)
101 return mddev
->sync_speed_max
?
102 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
105 static struct ctl_table_header
*raid_table_header
;
107 static ctl_table raid_table
[] = {
109 .procname
= "speed_limit_min",
110 .data
= &sysctl_speed_limit_min
,
111 .maxlen
= sizeof(int),
112 .mode
= S_IRUGO
|S_IWUSR
,
113 .proc_handler
= proc_dointvec
,
116 .procname
= "speed_limit_max",
117 .data
= &sysctl_speed_limit_max
,
118 .maxlen
= sizeof(int),
119 .mode
= S_IRUGO
|S_IWUSR
,
120 .proc_handler
= proc_dointvec
,
125 static ctl_table raid_dir_table
[] = {
129 .mode
= S_IRUGO
|S_IXUGO
,
135 static ctl_table raid_root_table
[] = {
140 .child
= raid_dir_table
,
145 static const struct block_device_operations md_fops
;
147 static int start_readonly
;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
160 static atomic_t md_event_count
;
161 void md_new_event(mddev_t
*mddev
)
163 atomic_inc(&md_event_count
);
164 wake_up(&md_event_waiters
);
166 EXPORT_SYMBOL_GPL(md_new_event
);
168 /* Alternate version that can be called from interrupts
169 * when calling sysfs_notify isn't needed.
171 static void md_new_event_inintr(mddev_t
*mddev
)
173 atomic_inc(&md_event_count
);
174 wake_up(&md_event_waiters
);
178 * Enables to iterate over all existing md arrays
179 * all_mddevs_lock protects this list.
181 static LIST_HEAD(all_mddevs
);
182 static DEFINE_SPINLOCK(all_mddevs_lock
);
186 * iterates through all used mddevs in the system.
187 * We take care to grab the all_mddevs_lock whenever navigating
188 * the list, and to always hold a refcount when unlocked.
189 * Any code which breaks out of this loop while own
190 * a reference to the current mddev and must mddev_put it.
192 #define for_each_mddev(mddev,tmp) \
194 for (({ spin_lock(&all_mddevs_lock); \
195 tmp = all_mddevs.next; \
197 ({ if (tmp != &all_mddevs) \
198 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
199 spin_unlock(&all_mddevs_lock); \
200 if (mddev) mddev_put(mddev); \
201 mddev = list_entry(tmp, mddev_t, all_mddevs); \
202 tmp != &all_mddevs;}); \
203 ({ spin_lock(&all_mddevs_lock); \
208 /* Rather than calling directly into the personality make_request function,
209 * IO requests come here first so that we can check if the device is
210 * being suspended pending a reconfiguration.
211 * We hold a refcount over the call to ->make_request. By the time that
212 * call has finished, the bio has been linked into some internal structure
213 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
217 const int rw
= bio_data_dir(bio
);
218 mddev_t
*mddev
= q
->queuedata
;
222 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
227 if (mddev
->suspended
|| mddev
->barrier
) {
230 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
231 TASK_UNINTERRUPTIBLE
);
232 if (!mddev
->suspended
&& !mddev
->barrier
)
238 finish_wait(&mddev
->sb_wait
, &__wait
);
240 atomic_inc(&mddev
->active_io
);
243 rv
= mddev
->pers
->make_request(mddev
, bio
);
245 cpu
= part_stat_lock();
246 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
247 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
251 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
252 wake_up(&mddev
->sb_wait
);
257 /* mddev_suspend makes sure no new requests are submitted
258 * to the device, and that any requests that have been submitted
259 * are completely handled.
260 * Once ->stop is called and completes, the module will be completely
263 static void mddev_suspend(mddev_t
*mddev
)
265 BUG_ON(mddev
->suspended
);
266 mddev
->suspended
= 1;
268 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
269 mddev
->pers
->quiesce(mddev
, 1);
272 static void mddev_resume(mddev_t
*mddev
)
274 mddev
->suspended
= 0;
275 wake_up(&mddev
->sb_wait
);
276 mddev
->pers
->quiesce(mddev
, 0);
279 int mddev_congested(mddev_t
*mddev
, int bits
)
283 return mddev
->suspended
;
285 EXPORT_SYMBOL(mddev_congested
);
288 * Generic barrier handling for md
291 #define POST_REQUEST_BARRIER ((void*)1)
293 static void md_end_barrier(struct bio
*bio
, int err
)
295 mdk_rdev_t
*rdev
= bio
->bi_private
;
296 mddev_t
*mddev
= rdev
->mddev
;
297 if (err
== -EOPNOTSUPP
&& mddev
->barrier
!= POST_REQUEST_BARRIER
)
298 set_bit(BIO_EOPNOTSUPP
, &mddev
->barrier
->bi_flags
);
300 rdev_dec_pending(rdev
, mddev
);
302 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
303 if (mddev
->barrier
== POST_REQUEST_BARRIER
) {
304 /* This was a post-request barrier */
305 mddev
->barrier
= NULL
;
306 wake_up(&mddev
->sb_wait
);
308 /* The pre-request barrier has finished */
309 schedule_work(&mddev
->barrier_work
);
314 static void submit_barriers(mddev_t
*mddev
)
319 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
320 if (rdev
->raid_disk
>= 0 &&
321 !test_bit(Faulty
, &rdev
->flags
)) {
322 /* Take two references, one is dropped
323 * when request finishes, one after
324 * we reclaim rcu_read_lock
327 atomic_inc(&rdev
->nr_pending
);
328 atomic_inc(&rdev
->nr_pending
);
330 bi
= bio_alloc(GFP_KERNEL
, 0);
331 bi
->bi_end_io
= md_end_barrier
;
332 bi
->bi_private
= rdev
;
333 bi
->bi_bdev
= rdev
->bdev
;
334 atomic_inc(&mddev
->flush_pending
);
335 submit_bio(WRITE_BARRIER
, bi
);
337 rdev_dec_pending(rdev
, mddev
);
342 static void md_submit_barrier(struct work_struct
*ws
)
344 mddev_t
*mddev
= container_of(ws
, mddev_t
, barrier_work
);
345 struct bio
*bio
= mddev
->barrier
;
347 atomic_set(&mddev
->flush_pending
, 1);
349 if (test_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
))
350 bio_endio(bio
, -EOPNOTSUPP
);
351 else if (bio
->bi_size
== 0)
352 /* an empty barrier - all done */
355 bio
->bi_rw
&= ~(1<<BIO_RW_BARRIER
);
356 if (mddev
->pers
->make_request(mddev
, bio
))
357 generic_make_request(bio
);
358 mddev
->barrier
= POST_REQUEST_BARRIER
;
359 submit_barriers(mddev
);
361 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
362 mddev
->barrier
= NULL
;
363 wake_up(&mddev
->sb_wait
);
367 void md_barrier_request(mddev_t
*mddev
, struct bio
*bio
)
369 spin_lock_irq(&mddev
->write_lock
);
370 wait_event_lock_irq(mddev
->sb_wait
,
372 mddev
->write_lock
, /*nothing*/);
373 mddev
->barrier
= bio
;
374 spin_unlock_irq(&mddev
->write_lock
);
376 atomic_set(&mddev
->flush_pending
, 1);
377 INIT_WORK(&mddev
->barrier_work
, md_submit_barrier
);
379 submit_barriers(mddev
);
381 if (atomic_dec_and_test(&mddev
->flush_pending
))
382 schedule_work(&mddev
->barrier_work
);
384 EXPORT_SYMBOL(md_barrier_request
);
386 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
388 atomic_inc(&mddev
->active
);
392 static void mddev_delayed_delete(struct work_struct
*ws
);
394 static void mddev_put(mddev_t
*mddev
)
396 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
398 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
399 mddev
->ctime
== 0 && !mddev
->hold_active
) {
400 /* Array is not configured at all, and not held active,
402 list_del(&mddev
->all_mddevs
);
403 if (mddev
->gendisk
) {
404 /* we did a probe so need to clean up.
405 * Call schedule_work inside the spinlock
406 * so that flush_scheduled_work() after
407 * mddev_find will succeed in waiting for the
410 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
411 schedule_work(&mddev
->del_work
);
415 spin_unlock(&all_mddevs_lock
);
418 static void mddev_init(mddev_t
*mddev
)
420 mutex_init(&mddev
->open_mutex
);
421 mutex_init(&mddev
->reconfig_mutex
);
422 mutex_init(&mddev
->bitmap_info
.mutex
);
423 INIT_LIST_HEAD(&mddev
->disks
);
424 INIT_LIST_HEAD(&mddev
->all_mddevs
);
425 init_timer(&mddev
->safemode_timer
);
426 atomic_set(&mddev
->active
, 1);
427 atomic_set(&mddev
->openers
, 0);
428 atomic_set(&mddev
->active_io
, 0);
429 spin_lock_init(&mddev
->write_lock
);
430 atomic_set(&mddev
->flush_pending
, 0);
431 init_waitqueue_head(&mddev
->sb_wait
);
432 init_waitqueue_head(&mddev
->recovery_wait
);
433 mddev
->reshape_position
= MaxSector
;
434 mddev
->resync_min
= 0;
435 mddev
->resync_max
= MaxSector
;
436 mddev
->level
= LEVEL_NONE
;
439 static mddev_t
* mddev_find(dev_t unit
)
441 mddev_t
*mddev
, *new = NULL
;
444 spin_lock(&all_mddevs_lock
);
447 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
448 if (mddev
->unit
== unit
) {
450 spin_unlock(&all_mddevs_lock
);
456 list_add(&new->all_mddevs
, &all_mddevs
);
457 spin_unlock(&all_mddevs_lock
);
458 new->hold_active
= UNTIL_IOCTL
;
462 /* find an unused unit number */
463 static int next_minor
= 512;
464 int start
= next_minor
;
468 dev
= MKDEV(MD_MAJOR
, next_minor
);
470 if (next_minor
> MINORMASK
)
472 if (next_minor
== start
) {
473 /* Oh dear, all in use. */
474 spin_unlock(&all_mddevs_lock
);
480 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
481 if (mddev
->unit
== dev
) {
487 new->md_minor
= MINOR(dev
);
488 new->hold_active
= UNTIL_STOP
;
489 list_add(&new->all_mddevs
, &all_mddevs
);
490 spin_unlock(&all_mddevs_lock
);
493 spin_unlock(&all_mddevs_lock
);
495 new = kzalloc(sizeof(*new), GFP_KERNEL
);
500 if (MAJOR(unit
) == MD_MAJOR
)
501 new->md_minor
= MINOR(unit
);
503 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
510 static inline int mddev_lock(mddev_t
* mddev
)
512 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
515 static inline int mddev_is_locked(mddev_t
*mddev
)
517 return mutex_is_locked(&mddev
->reconfig_mutex
);
520 static inline int mddev_trylock(mddev_t
* mddev
)
522 return mutex_trylock(&mddev
->reconfig_mutex
);
525 static struct attribute_group md_redundancy_group
;
527 static void mddev_unlock(mddev_t
* mddev
)
529 if (mddev
->to_remove
) {
530 /* These cannot be removed under reconfig_mutex as
531 * an access to the files will try to take reconfig_mutex
532 * while holding the file unremovable, which leads to
534 * So hold open_mutex instead - we are allowed to take
535 * it while holding reconfig_mutex, and md_run can
536 * use it to wait for the remove to complete.
538 struct attribute_group
*to_remove
= mddev
->to_remove
;
539 mddev
->to_remove
= NULL
;
540 mutex_lock(&mddev
->open_mutex
);
541 mutex_unlock(&mddev
->reconfig_mutex
);
543 if (to_remove
!= &md_redundancy_group
)
544 sysfs_remove_group(&mddev
->kobj
, to_remove
);
545 if (mddev
->pers
== NULL
||
546 mddev
->pers
->sync_request
== NULL
) {
547 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
548 if (mddev
->sysfs_action
)
549 sysfs_put(mddev
->sysfs_action
);
550 mddev
->sysfs_action
= NULL
;
552 mutex_unlock(&mddev
->open_mutex
);
554 mutex_unlock(&mddev
->reconfig_mutex
);
556 md_wakeup_thread(mddev
->thread
);
559 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
563 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
564 if (rdev
->desc_nr
== nr
)
570 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
574 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
575 if (rdev
->bdev
->bd_dev
== dev
)
581 static struct mdk_personality
*find_pers(int level
, char *clevel
)
583 struct mdk_personality
*pers
;
584 list_for_each_entry(pers
, &pers_list
, list
) {
585 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
587 if (strcmp(pers
->name
, clevel
)==0)
593 /* return the offset of the super block in 512byte sectors */
594 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
596 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
597 return MD_NEW_SIZE_SECTORS(num_sectors
);
600 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
605 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
606 if (!rdev
->sb_page
) {
607 printk(KERN_ALERT
"md: out of memory.\n");
614 static void free_disk_sb(mdk_rdev_t
* rdev
)
617 put_page(rdev
->sb_page
);
619 rdev
->sb_page
= NULL
;
626 static void super_written(struct bio
*bio
, int error
)
628 mdk_rdev_t
*rdev
= bio
->bi_private
;
629 mddev_t
*mddev
= rdev
->mddev
;
631 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
632 printk("md: super_written gets error=%d, uptodate=%d\n",
633 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
634 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
635 md_error(mddev
, rdev
);
638 if (atomic_dec_and_test(&mddev
->pending_writes
))
639 wake_up(&mddev
->sb_wait
);
643 static void super_written_barrier(struct bio
*bio
, int error
)
645 struct bio
*bio2
= bio
->bi_private
;
646 mdk_rdev_t
*rdev
= bio2
->bi_private
;
647 mddev_t
*mddev
= rdev
->mddev
;
649 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
650 error
== -EOPNOTSUPP
) {
652 /* barriers don't appear to be supported :-( */
653 set_bit(BarriersNotsupp
, &rdev
->flags
);
654 mddev
->barriers_work
= 0;
655 spin_lock_irqsave(&mddev
->write_lock
, flags
);
656 bio2
->bi_next
= mddev
->biolist
;
657 mddev
->biolist
= bio2
;
658 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
659 wake_up(&mddev
->sb_wait
);
663 bio
->bi_private
= rdev
;
664 super_written(bio
, error
);
668 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
669 sector_t sector
, int size
, struct page
*page
)
671 /* write first size bytes of page to sector of rdev
672 * Increment mddev->pending_writes before returning
673 * and decrement it on completion, waking up sb_wait
674 * if zero is reached.
675 * If an error occurred, call md_error
677 * As we might need to resubmit the request if BIO_RW_BARRIER
678 * causes ENOTSUPP, we allocate a spare bio...
680 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
681 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
683 bio
->bi_bdev
= rdev
->bdev
;
684 bio
->bi_sector
= sector
;
685 bio_add_page(bio
, page
, size
, 0);
686 bio
->bi_private
= rdev
;
687 bio
->bi_end_io
= super_written
;
690 atomic_inc(&mddev
->pending_writes
);
691 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
693 rw
|= (1<<BIO_RW_BARRIER
);
694 rbio
= bio_clone(bio
, GFP_NOIO
);
695 rbio
->bi_private
= bio
;
696 rbio
->bi_end_io
= super_written_barrier
;
697 submit_bio(rw
, rbio
);
702 void md_super_wait(mddev_t
*mddev
)
704 /* wait for all superblock writes that were scheduled to complete.
705 * if any had to be retried (due to BARRIER problems), retry them
709 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
710 if (atomic_read(&mddev
->pending_writes
)==0)
712 while (mddev
->biolist
) {
714 spin_lock_irq(&mddev
->write_lock
);
715 bio
= mddev
->biolist
;
716 mddev
->biolist
= bio
->bi_next
;
718 spin_unlock_irq(&mddev
->write_lock
);
719 submit_bio(bio
->bi_rw
, bio
);
723 finish_wait(&mddev
->sb_wait
, &wq
);
726 static void bi_complete(struct bio
*bio
, int error
)
728 complete((struct completion
*)bio
->bi_private
);
731 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
732 struct page
*page
, int rw
)
734 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
735 struct completion event
;
738 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
741 bio
->bi_sector
= sector
;
742 bio_add_page(bio
, page
, size
, 0);
743 init_completion(&event
);
744 bio
->bi_private
= &event
;
745 bio
->bi_end_io
= bi_complete
;
747 wait_for_completion(&event
);
749 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
753 EXPORT_SYMBOL_GPL(sync_page_io
);
755 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
757 char b
[BDEVNAME_SIZE
];
758 if (!rdev
->sb_page
) {
766 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
772 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
773 bdevname(rdev
->bdev
,b
));
777 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
779 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
780 sb1
->set_uuid1
== sb2
->set_uuid1
&&
781 sb1
->set_uuid2
== sb2
->set_uuid2
&&
782 sb1
->set_uuid3
== sb2
->set_uuid3
;
785 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
788 mdp_super_t
*tmp1
, *tmp2
;
790 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
791 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
793 if (!tmp1
|| !tmp2
) {
795 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
803 * nr_disks is not constant
808 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
816 static u32
md_csum_fold(u32 csum
)
818 csum
= (csum
& 0xffff) + (csum
>> 16);
819 return (csum
& 0xffff) + (csum
>> 16);
822 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
825 u32
*sb32
= (u32
*)sb
;
827 unsigned int disk_csum
, csum
;
829 disk_csum
= sb
->sb_csum
;
832 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
834 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
838 /* This used to use csum_partial, which was wrong for several
839 * reasons including that different results are returned on
840 * different architectures. It isn't critical that we get exactly
841 * the same return value as before (we always csum_fold before
842 * testing, and that removes any differences). However as we
843 * know that csum_partial always returned a 16bit value on
844 * alphas, do a fold to maximise conformity to previous behaviour.
846 sb
->sb_csum
= md_csum_fold(disk_csum
);
848 sb
->sb_csum
= disk_csum
;
855 * Handle superblock details.
856 * We want to be able to handle multiple superblock formats
857 * so we have a common interface to them all, and an array of
858 * different handlers.
859 * We rely on user-space to write the initial superblock, and support
860 * reading and updating of superblocks.
861 * Interface methods are:
862 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
863 * loads and validates a superblock on dev.
864 * if refdev != NULL, compare superblocks on both devices
866 * 0 - dev has a superblock that is compatible with refdev
867 * 1 - dev has a superblock that is compatible and newer than refdev
868 * so dev should be used as the refdev in future
869 * -EINVAL superblock incompatible or invalid
870 * -othererror e.g. -EIO
872 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
873 * Verify that dev is acceptable into mddev.
874 * The first time, mddev->raid_disks will be 0, and data from
875 * dev should be merged in. Subsequent calls check that dev
876 * is new enough. Return 0 or -EINVAL
878 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
879 * Update the superblock for rdev with data in mddev
880 * This does not write to disc.
886 struct module
*owner
;
887 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
889 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
890 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
891 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
892 sector_t num_sectors
);
896 * Check that the given mddev has no bitmap.
898 * This function is called from the run method of all personalities that do not
899 * support bitmaps. It prints an error message and returns non-zero if mddev
900 * has a bitmap. Otherwise, it returns 0.
903 int md_check_no_bitmap(mddev_t
*mddev
)
905 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
907 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
908 mdname(mddev
), mddev
->pers
->name
);
911 EXPORT_SYMBOL(md_check_no_bitmap
);
914 * load_super for 0.90.0
916 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
918 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
923 * Calculate the position of the superblock (512byte sectors),
924 * it's at the end of the disk.
926 * It also happens to be a multiple of 4Kb.
928 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
930 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
935 bdevname(rdev
->bdev
, b
);
936 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
938 if (sb
->md_magic
!= MD_SB_MAGIC
) {
939 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
944 if (sb
->major_version
!= 0 ||
945 sb
->minor_version
< 90 ||
946 sb
->minor_version
> 91) {
947 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
948 sb
->major_version
, sb
->minor_version
,
953 if (sb
->raid_disks
<= 0)
956 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
957 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
962 rdev
->preferred_minor
= sb
->md_minor
;
963 rdev
->data_offset
= 0;
964 rdev
->sb_size
= MD_SB_BYTES
;
966 if (sb
->level
== LEVEL_MULTIPATH
)
969 rdev
->desc_nr
= sb
->this_disk
.number
;
975 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
976 if (!uuid_equal(refsb
, sb
)) {
977 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
978 b
, bdevname(refdev
->bdev
,b2
));
981 if (!sb_equal(refsb
, sb
)) {
982 printk(KERN_WARNING
"md: %s has same UUID"
983 " but different superblock to %s\n",
984 b
, bdevname(refdev
->bdev
, b2
));
988 ev2
= md_event(refsb
);
994 rdev
->sectors
= rdev
->sb_start
;
996 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
997 /* "this cannot possibly happen" ... */
1005 * validate_super for 0.90.0
1007 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1010 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1011 __u64 ev1
= md_event(sb
);
1013 rdev
->raid_disk
= -1;
1014 clear_bit(Faulty
, &rdev
->flags
);
1015 clear_bit(In_sync
, &rdev
->flags
);
1016 clear_bit(WriteMostly
, &rdev
->flags
);
1017 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1019 if (mddev
->raid_disks
== 0) {
1020 mddev
->major_version
= 0;
1021 mddev
->minor_version
= sb
->minor_version
;
1022 mddev
->patch_version
= sb
->patch_version
;
1023 mddev
->external
= 0;
1024 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1025 mddev
->ctime
= sb
->ctime
;
1026 mddev
->utime
= sb
->utime
;
1027 mddev
->level
= sb
->level
;
1028 mddev
->clevel
[0] = 0;
1029 mddev
->layout
= sb
->layout
;
1030 mddev
->raid_disks
= sb
->raid_disks
;
1031 mddev
->dev_sectors
= sb
->size
* 2;
1032 mddev
->events
= ev1
;
1033 mddev
->bitmap_info
.offset
= 0;
1034 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1036 if (mddev
->minor_version
>= 91) {
1037 mddev
->reshape_position
= sb
->reshape_position
;
1038 mddev
->delta_disks
= sb
->delta_disks
;
1039 mddev
->new_level
= sb
->new_level
;
1040 mddev
->new_layout
= sb
->new_layout
;
1041 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1043 mddev
->reshape_position
= MaxSector
;
1044 mddev
->delta_disks
= 0;
1045 mddev
->new_level
= mddev
->level
;
1046 mddev
->new_layout
= mddev
->layout
;
1047 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1050 if (sb
->state
& (1<<MD_SB_CLEAN
))
1051 mddev
->recovery_cp
= MaxSector
;
1053 if (sb
->events_hi
== sb
->cp_events_hi
&&
1054 sb
->events_lo
== sb
->cp_events_lo
) {
1055 mddev
->recovery_cp
= sb
->recovery_cp
;
1057 mddev
->recovery_cp
= 0;
1060 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1061 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1062 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1063 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1065 mddev
->max_disks
= MD_SB_DISKS
;
1067 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1068 mddev
->bitmap_info
.file
== NULL
)
1069 mddev
->bitmap_info
.offset
=
1070 mddev
->bitmap_info
.default_offset
;
1072 } else if (mddev
->pers
== NULL
) {
1073 /* Insist on good event counter while assembling, except
1074 * for spares (which don't need an event count) */
1076 if (sb
->disks
[rdev
->desc_nr
].state
& (
1077 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1078 if (ev1
< mddev
->events
)
1080 } else if (mddev
->bitmap
) {
1081 /* if adding to array with a bitmap, then we can accept an
1082 * older device ... but not too old.
1084 if (ev1
< mddev
->bitmap
->events_cleared
)
1087 if (ev1
< mddev
->events
)
1088 /* just a hot-add of a new device, leave raid_disk at -1 */
1092 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1093 desc
= sb
->disks
+ rdev
->desc_nr
;
1095 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1096 set_bit(Faulty
, &rdev
->flags
);
1097 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1098 desc->raid_disk < mddev->raid_disks */) {
1099 set_bit(In_sync
, &rdev
->flags
);
1100 rdev
->raid_disk
= desc
->raid_disk
;
1101 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1102 /* active but not in sync implies recovery up to
1103 * reshape position. We don't know exactly where
1104 * that is, so set to zero for now */
1105 if (mddev
->minor_version
>= 91) {
1106 rdev
->recovery_offset
= 0;
1107 rdev
->raid_disk
= desc
->raid_disk
;
1110 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1111 set_bit(WriteMostly
, &rdev
->flags
);
1112 } else /* MULTIPATH are always insync */
1113 set_bit(In_sync
, &rdev
->flags
);
1118 * sync_super for 0.90.0
1120 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1124 int next_spare
= mddev
->raid_disks
;
1127 /* make rdev->sb match mddev data..
1130 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1131 * 3/ any empty disks < next_spare become removed
1133 * disks[0] gets initialised to REMOVED because
1134 * we cannot be sure from other fields if it has
1135 * been initialised or not.
1138 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1140 rdev
->sb_size
= MD_SB_BYTES
;
1142 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1144 memset(sb
, 0, sizeof(*sb
));
1146 sb
->md_magic
= MD_SB_MAGIC
;
1147 sb
->major_version
= mddev
->major_version
;
1148 sb
->patch_version
= mddev
->patch_version
;
1149 sb
->gvalid_words
= 0; /* ignored */
1150 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1151 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1152 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1153 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1155 sb
->ctime
= mddev
->ctime
;
1156 sb
->level
= mddev
->level
;
1157 sb
->size
= mddev
->dev_sectors
/ 2;
1158 sb
->raid_disks
= mddev
->raid_disks
;
1159 sb
->md_minor
= mddev
->md_minor
;
1160 sb
->not_persistent
= 0;
1161 sb
->utime
= mddev
->utime
;
1163 sb
->events_hi
= (mddev
->events
>>32);
1164 sb
->events_lo
= (u32
)mddev
->events
;
1166 if (mddev
->reshape_position
== MaxSector
)
1167 sb
->minor_version
= 90;
1169 sb
->minor_version
= 91;
1170 sb
->reshape_position
= mddev
->reshape_position
;
1171 sb
->new_level
= mddev
->new_level
;
1172 sb
->delta_disks
= mddev
->delta_disks
;
1173 sb
->new_layout
= mddev
->new_layout
;
1174 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1176 mddev
->minor_version
= sb
->minor_version
;
1179 sb
->recovery_cp
= mddev
->recovery_cp
;
1180 sb
->cp_events_hi
= (mddev
->events
>>32);
1181 sb
->cp_events_lo
= (u32
)mddev
->events
;
1182 if (mddev
->recovery_cp
== MaxSector
)
1183 sb
->state
= (1<< MD_SB_CLEAN
);
1185 sb
->recovery_cp
= 0;
1187 sb
->layout
= mddev
->layout
;
1188 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1190 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1191 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1193 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1194 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1197 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1199 if (rdev2
->raid_disk
>= 0 &&
1200 sb
->minor_version
>= 91)
1201 /* we have nowhere to store the recovery_offset,
1202 * but if it is not below the reshape_position,
1203 * we can piggy-back on that.
1206 if (rdev2
->raid_disk
< 0 ||
1207 test_bit(Faulty
, &rdev2
->flags
))
1210 desc_nr
= rdev2
->raid_disk
;
1212 desc_nr
= next_spare
++;
1213 rdev2
->desc_nr
= desc_nr
;
1214 d
= &sb
->disks
[rdev2
->desc_nr
];
1216 d
->number
= rdev2
->desc_nr
;
1217 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1218 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1220 d
->raid_disk
= rdev2
->raid_disk
;
1222 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1223 if (test_bit(Faulty
, &rdev2
->flags
))
1224 d
->state
= (1<<MD_DISK_FAULTY
);
1225 else if (is_active
) {
1226 d
->state
= (1<<MD_DISK_ACTIVE
);
1227 if (test_bit(In_sync
, &rdev2
->flags
))
1228 d
->state
|= (1<<MD_DISK_SYNC
);
1236 if (test_bit(WriteMostly
, &rdev2
->flags
))
1237 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1239 /* now set the "removed" and "faulty" bits on any missing devices */
1240 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1241 mdp_disk_t
*d
= &sb
->disks
[i
];
1242 if (d
->state
== 0 && d
->number
== 0) {
1245 d
->state
= (1<<MD_DISK_REMOVED
);
1246 d
->state
|= (1<<MD_DISK_FAULTY
);
1250 sb
->nr_disks
= nr_disks
;
1251 sb
->active_disks
= active
;
1252 sb
->working_disks
= working
;
1253 sb
->failed_disks
= failed
;
1254 sb
->spare_disks
= spare
;
1256 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1257 sb
->sb_csum
= calc_sb_csum(sb
);
1261 * rdev_size_change for 0.90.0
1263 static unsigned long long
1264 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1266 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1267 return 0; /* component must fit device */
1268 if (rdev
->mddev
->bitmap_info
.offset
)
1269 return 0; /* can't move bitmap */
1270 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1271 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1272 num_sectors
= rdev
->sb_start
;
1273 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1275 md_super_wait(rdev
->mddev
);
1276 return num_sectors
/ 2; /* kB for sysfs */
1281 * version 1 superblock
1284 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1288 unsigned long long newcsum
;
1289 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1290 __le32
*isuper
= (__le32
*)sb
;
1293 disk_csum
= sb
->sb_csum
;
1296 for (i
=0; size
>=4; size
-= 4 )
1297 newcsum
+= le32_to_cpu(*isuper
++);
1300 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1302 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1303 sb
->sb_csum
= disk_csum
;
1304 return cpu_to_le32(csum
);
1307 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1309 struct mdp_superblock_1
*sb
;
1312 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1316 * Calculate the position of the superblock in 512byte sectors.
1317 * It is always aligned to a 4K boundary and
1318 * depeding on minor_version, it can be:
1319 * 0: At least 8K, but less than 12K, from end of device
1320 * 1: At start of device
1321 * 2: 4K from start of device.
1323 switch(minor_version
) {
1325 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1327 sb_start
&= ~(sector_t
)(4*2-1);
1338 rdev
->sb_start
= sb_start
;
1340 /* superblock is rarely larger than 1K, but it can be larger,
1341 * and it is safe to read 4k, so we do that
1343 ret
= read_disk_sb(rdev
, 4096);
1344 if (ret
) return ret
;
1347 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1349 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1350 sb
->major_version
!= cpu_to_le32(1) ||
1351 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1352 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1353 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1356 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1357 printk("md: invalid superblock checksum on %s\n",
1358 bdevname(rdev
->bdev
,b
));
1361 if (le64_to_cpu(sb
->data_size
) < 10) {
1362 printk("md: data_size too small on %s\n",
1363 bdevname(rdev
->bdev
,b
));
1367 rdev
->preferred_minor
= 0xffff;
1368 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1369 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1371 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1372 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1373 if (rdev
->sb_size
& bmask
)
1374 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1377 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1380 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1383 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1389 struct mdp_superblock_1
*refsb
=
1390 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1392 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1393 sb
->level
!= refsb
->level
||
1394 sb
->layout
!= refsb
->layout
||
1395 sb
->chunksize
!= refsb
->chunksize
) {
1396 printk(KERN_WARNING
"md: %s has strangely different"
1397 " superblock to %s\n",
1398 bdevname(rdev
->bdev
,b
),
1399 bdevname(refdev
->bdev
,b2
));
1402 ev1
= le64_to_cpu(sb
->events
);
1403 ev2
= le64_to_cpu(refsb
->events
);
1411 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1412 le64_to_cpu(sb
->data_offset
);
1414 rdev
->sectors
= rdev
->sb_start
;
1415 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1417 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1418 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1423 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1425 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1426 __u64 ev1
= le64_to_cpu(sb
->events
);
1428 rdev
->raid_disk
= -1;
1429 clear_bit(Faulty
, &rdev
->flags
);
1430 clear_bit(In_sync
, &rdev
->flags
);
1431 clear_bit(WriteMostly
, &rdev
->flags
);
1432 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1434 if (mddev
->raid_disks
== 0) {
1435 mddev
->major_version
= 1;
1436 mddev
->patch_version
= 0;
1437 mddev
->external
= 0;
1438 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1439 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1440 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1441 mddev
->level
= le32_to_cpu(sb
->level
);
1442 mddev
->clevel
[0] = 0;
1443 mddev
->layout
= le32_to_cpu(sb
->layout
);
1444 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1445 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1446 mddev
->events
= ev1
;
1447 mddev
->bitmap_info
.offset
= 0;
1448 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1450 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1451 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1453 mddev
->max_disks
= (4096-256)/2;
1455 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1456 mddev
->bitmap_info
.file
== NULL
)
1457 mddev
->bitmap_info
.offset
=
1458 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1460 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1461 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1462 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1463 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1464 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1465 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1467 mddev
->reshape_position
= MaxSector
;
1468 mddev
->delta_disks
= 0;
1469 mddev
->new_level
= mddev
->level
;
1470 mddev
->new_layout
= mddev
->layout
;
1471 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1474 } else if (mddev
->pers
== NULL
) {
1475 /* Insist of good event counter while assembling, except for
1476 * spares (which don't need an event count) */
1478 if (rdev
->desc_nr
>= 0 &&
1479 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1480 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1481 if (ev1
< mddev
->events
)
1483 } else if (mddev
->bitmap
) {
1484 /* If adding to array with a bitmap, then we can accept an
1485 * older device, but not too old.
1487 if (ev1
< mddev
->bitmap
->events_cleared
)
1490 if (ev1
< mddev
->events
)
1491 /* just a hot-add of a new device, leave raid_disk at -1 */
1494 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1496 if (rdev
->desc_nr
< 0 ||
1497 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1501 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1503 case 0xffff: /* spare */
1505 case 0xfffe: /* faulty */
1506 set_bit(Faulty
, &rdev
->flags
);
1509 if ((le32_to_cpu(sb
->feature_map
) &
1510 MD_FEATURE_RECOVERY_OFFSET
))
1511 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1513 set_bit(In_sync
, &rdev
->flags
);
1514 rdev
->raid_disk
= role
;
1517 if (sb
->devflags
& WriteMostly1
)
1518 set_bit(WriteMostly
, &rdev
->flags
);
1519 } else /* MULTIPATH are always insync */
1520 set_bit(In_sync
, &rdev
->flags
);
1525 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1527 struct mdp_superblock_1
*sb
;
1530 /* make rdev->sb match mddev and rdev data. */
1532 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1534 sb
->feature_map
= 0;
1536 sb
->recovery_offset
= cpu_to_le64(0);
1537 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1538 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1539 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1541 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1542 sb
->events
= cpu_to_le64(mddev
->events
);
1544 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1546 sb
->resync_offset
= cpu_to_le64(0);
1548 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1550 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1551 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1552 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1553 sb
->level
= cpu_to_le32(mddev
->level
);
1554 sb
->layout
= cpu_to_le32(mddev
->layout
);
1556 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1557 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1558 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1561 if (rdev
->raid_disk
>= 0 &&
1562 !test_bit(In_sync
, &rdev
->flags
)) {
1564 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1565 sb
->recovery_offset
=
1566 cpu_to_le64(rdev
->recovery_offset
);
1569 if (mddev
->reshape_position
!= MaxSector
) {
1570 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1571 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1572 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1573 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1574 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1575 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1579 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1580 if (rdev2
->desc_nr
+1 > max_dev
)
1581 max_dev
= rdev2
->desc_nr
+1;
1583 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1585 sb
->max_dev
= cpu_to_le32(max_dev
);
1586 rdev
->sb_size
= max_dev
* 2 + 256;
1587 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1588 if (rdev
->sb_size
& bmask
)
1589 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1591 for (i
=0; i
<max_dev
;i
++)
1592 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1594 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1596 if (test_bit(Faulty
, &rdev2
->flags
))
1597 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1598 else if (test_bit(In_sync
, &rdev2
->flags
))
1599 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1600 else if (rdev2
->raid_disk
>= 0)
1601 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1603 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1606 sb
->sb_csum
= calc_sb_1_csum(sb
);
1609 static unsigned long long
1610 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1612 struct mdp_superblock_1
*sb
;
1613 sector_t max_sectors
;
1614 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1615 return 0; /* component must fit device */
1616 if (rdev
->sb_start
< rdev
->data_offset
) {
1617 /* minor versions 1 and 2; superblock before data */
1618 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1619 max_sectors
-= rdev
->data_offset
;
1620 if (!num_sectors
|| num_sectors
> max_sectors
)
1621 num_sectors
= max_sectors
;
1622 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1623 /* minor version 0 with bitmap we can't move */
1626 /* minor version 0; superblock after data */
1628 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1629 sb_start
&= ~(sector_t
)(4*2 - 1);
1630 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1631 if (!num_sectors
|| num_sectors
> max_sectors
)
1632 num_sectors
= max_sectors
;
1633 rdev
->sb_start
= sb_start
;
1635 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1636 sb
->data_size
= cpu_to_le64(num_sectors
);
1637 sb
->super_offset
= rdev
->sb_start
;
1638 sb
->sb_csum
= calc_sb_1_csum(sb
);
1639 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1641 md_super_wait(rdev
->mddev
);
1642 return num_sectors
/ 2; /* kB for sysfs */
1645 static struct super_type super_types
[] = {
1648 .owner
= THIS_MODULE
,
1649 .load_super
= super_90_load
,
1650 .validate_super
= super_90_validate
,
1651 .sync_super
= super_90_sync
,
1652 .rdev_size_change
= super_90_rdev_size_change
,
1656 .owner
= THIS_MODULE
,
1657 .load_super
= super_1_load
,
1658 .validate_super
= super_1_validate
,
1659 .sync_super
= super_1_sync
,
1660 .rdev_size_change
= super_1_rdev_size_change
,
1664 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1666 mdk_rdev_t
*rdev
, *rdev2
;
1669 rdev_for_each_rcu(rdev
, mddev1
)
1670 rdev_for_each_rcu(rdev2
, mddev2
)
1671 if (rdev
->bdev
->bd_contains
==
1672 rdev2
->bdev
->bd_contains
) {
1680 static LIST_HEAD(pending_raid_disks
);
1683 * Try to register data integrity profile for an mddev
1685 * This is called when an array is started and after a disk has been kicked
1686 * from the array. It only succeeds if all working and active component devices
1687 * are integrity capable with matching profiles.
1689 int md_integrity_register(mddev_t
*mddev
)
1691 mdk_rdev_t
*rdev
, *reference
= NULL
;
1693 if (list_empty(&mddev
->disks
))
1694 return 0; /* nothing to do */
1695 if (blk_get_integrity(mddev
->gendisk
))
1696 return 0; /* already registered */
1697 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1698 /* skip spares and non-functional disks */
1699 if (test_bit(Faulty
, &rdev
->flags
))
1701 if (rdev
->raid_disk
< 0)
1704 * If at least one rdev is not integrity capable, we can not
1705 * enable data integrity for the md device.
1707 if (!bdev_get_integrity(rdev
->bdev
))
1710 /* Use the first rdev as the reference */
1714 /* does this rdev's profile match the reference profile? */
1715 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1716 rdev
->bdev
->bd_disk
) < 0)
1720 * All component devices are integrity capable and have matching
1721 * profiles, register the common profile for the md device.
1723 if (blk_integrity_register(mddev
->gendisk
,
1724 bdev_get_integrity(reference
->bdev
)) != 0) {
1725 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1729 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1733 EXPORT_SYMBOL(md_integrity_register
);
1735 /* Disable data integrity if non-capable/non-matching disk is being added */
1736 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1738 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1739 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1741 if (!bi_mddev
) /* nothing to do */
1743 if (rdev
->raid_disk
< 0) /* skip spares */
1745 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1746 rdev
->bdev
->bd_disk
) >= 0)
1748 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1749 blk_integrity_unregister(mddev
->gendisk
);
1751 EXPORT_SYMBOL(md_integrity_add_rdev
);
1753 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1755 char b
[BDEVNAME_SIZE
];
1765 /* prevent duplicates */
1766 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1769 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1770 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1771 rdev
->sectors
< mddev
->dev_sectors
)) {
1773 /* Cannot change size, so fail
1774 * If mddev->level <= 0, then we don't care
1775 * about aligning sizes (e.g. linear)
1777 if (mddev
->level
> 0)
1780 mddev
->dev_sectors
= rdev
->sectors
;
1783 /* Verify rdev->desc_nr is unique.
1784 * If it is -1, assign a free number, else
1785 * check number is not in use
1787 if (rdev
->desc_nr
< 0) {
1789 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1790 while (find_rdev_nr(mddev
, choice
))
1792 rdev
->desc_nr
= choice
;
1794 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1797 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1798 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1799 mdname(mddev
), mddev
->max_disks
);
1802 bdevname(rdev
->bdev
,b
);
1803 while ( (s
=strchr(b
, '/')) != NULL
)
1806 rdev
->mddev
= mddev
;
1807 printk(KERN_INFO
"md: bind<%s>\n", b
);
1809 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1812 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1813 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1814 kobject_del(&rdev
->kobj
);
1817 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1819 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1820 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1822 /* May as well allow recovery to be retried once */
1823 mddev
->recovery_disabled
= 0;
1828 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1833 static void md_delayed_delete(struct work_struct
*ws
)
1835 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1836 kobject_del(&rdev
->kobj
);
1837 kobject_put(&rdev
->kobj
);
1840 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1842 char b
[BDEVNAME_SIZE
];
1847 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1848 list_del_rcu(&rdev
->same_set
);
1849 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1851 sysfs_remove_link(&rdev
->kobj
, "block");
1852 sysfs_put(rdev
->sysfs_state
);
1853 rdev
->sysfs_state
= NULL
;
1854 /* We need to delay this, otherwise we can deadlock when
1855 * writing to 'remove' to "dev/state". We also need
1856 * to delay it due to rcu usage.
1859 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1860 kobject_get(&rdev
->kobj
);
1861 schedule_work(&rdev
->del_work
);
1865 * prevent the device from being mounted, repartitioned or
1866 * otherwise reused by a RAID array (or any other kernel
1867 * subsystem), by bd_claiming the device.
1869 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1872 struct block_device
*bdev
;
1873 char b
[BDEVNAME_SIZE
];
1875 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1877 printk(KERN_ERR
"md: could not open %s.\n",
1878 __bdevname(dev
, b
));
1879 return PTR_ERR(bdev
);
1881 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1883 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1885 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1889 set_bit(AllReserved
, &rdev
->flags
);
1894 static void unlock_rdev(mdk_rdev_t
*rdev
)
1896 struct block_device
*bdev
= rdev
->bdev
;
1901 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1904 void md_autodetect_dev(dev_t dev
);
1906 static void export_rdev(mdk_rdev_t
* rdev
)
1908 char b
[BDEVNAME_SIZE
];
1909 printk(KERN_INFO
"md: export_rdev(%s)\n",
1910 bdevname(rdev
->bdev
,b
));
1915 if (test_bit(AutoDetected
, &rdev
->flags
))
1916 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1919 kobject_put(&rdev
->kobj
);
1922 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1924 unbind_rdev_from_array(rdev
);
1928 static void export_array(mddev_t
*mddev
)
1930 mdk_rdev_t
*rdev
, *tmp
;
1932 rdev_for_each(rdev
, tmp
, mddev
) {
1937 kick_rdev_from_array(rdev
);
1939 if (!list_empty(&mddev
->disks
))
1941 mddev
->raid_disks
= 0;
1942 mddev
->major_version
= 0;
1945 static void print_desc(mdp_disk_t
*desc
)
1947 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1948 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1951 static void print_sb_90(mdp_super_t
*sb
)
1956 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1957 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1958 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1960 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1961 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1962 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1963 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1964 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1965 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1966 sb
->failed_disks
, sb
->spare_disks
,
1967 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1970 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1973 desc
= sb
->disks
+ i
;
1974 if (desc
->number
|| desc
->major
|| desc
->minor
||
1975 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1976 printk(" D %2d: ", i
);
1980 printk(KERN_INFO
"md: THIS: ");
1981 print_desc(&sb
->this_disk
);
1984 static void print_sb_1(struct mdp_superblock_1
*sb
)
1988 uuid
= sb
->set_uuid
;
1990 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1991 "md: Name: \"%s\" CT:%llu\n",
1992 le32_to_cpu(sb
->major_version
),
1993 le32_to_cpu(sb
->feature_map
),
1996 (unsigned long long)le64_to_cpu(sb
->ctime
)
1997 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1999 uuid
= sb
->device_uuid
;
2001 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2003 "md: Dev:%08x UUID: %pU\n"
2004 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2005 "md: (MaxDev:%u) \n",
2006 le32_to_cpu(sb
->level
),
2007 (unsigned long long)le64_to_cpu(sb
->size
),
2008 le32_to_cpu(sb
->raid_disks
),
2009 le32_to_cpu(sb
->layout
),
2010 le32_to_cpu(sb
->chunksize
),
2011 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2012 (unsigned long long)le64_to_cpu(sb
->data_size
),
2013 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2014 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2015 le32_to_cpu(sb
->dev_number
),
2018 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2019 (unsigned long long)le64_to_cpu(sb
->events
),
2020 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2021 le32_to_cpu(sb
->sb_csum
),
2022 le32_to_cpu(sb
->max_dev
)
2026 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2028 char b
[BDEVNAME_SIZE
];
2029 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2030 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2031 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2033 if (rdev
->sb_loaded
) {
2034 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2035 switch (major_version
) {
2037 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
2040 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
2044 printk(KERN_INFO
"md: no rdev superblock!\n");
2047 static void md_print_devices(void)
2049 struct list_head
*tmp
;
2052 char b
[BDEVNAME_SIZE
];
2055 printk("md: **********************************\n");
2056 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2057 printk("md: **********************************\n");
2058 for_each_mddev(mddev
, tmp
) {
2061 bitmap_print_sb(mddev
->bitmap
);
2063 printk("%s: ", mdname(mddev
));
2064 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2065 printk("<%s>", bdevname(rdev
->bdev
,b
));
2068 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2069 print_rdev(rdev
, mddev
->major_version
);
2071 printk("md: **********************************\n");
2076 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2078 /* Update each superblock (in-memory image), but
2079 * if we are allowed to, skip spares which already
2080 * have the right event counter, or have one earlier
2081 * (which would mean they aren't being marked as dirty
2082 * with the rest of the array)
2086 /* First make sure individual recovery_offsets are correct */
2087 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2088 if (rdev
->raid_disk
>= 0 &&
2089 !test_bit(In_sync
, &rdev
->flags
) &&
2090 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2091 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2094 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2095 if (rdev
->sb_events
== mddev
->events
||
2097 rdev
->raid_disk
< 0 &&
2098 rdev
->sb_events
+1 == mddev
->events
)) {
2099 /* Don't update this superblock */
2100 rdev
->sb_loaded
= 2;
2102 super_types
[mddev
->major_version
].
2103 sync_super(mddev
, rdev
);
2104 rdev
->sb_loaded
= 1;
2109 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2115 mddev
->utime
= get_seconds();
2116 if (mddev
->external
)
2119 spin_lock_irq(&mddev
->write_lock
);
2121 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2122 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2124 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2125 /* just a clean<-> dirty transition, possibly leave spares alone,
2126 * though if events isn't the right even/odd, we will have to do
2132 if (mddev
->degraded
)
2133 /* If the array is degraded, then skipping spares is both
2134 * dangerous and fairly pointless.
2135 * Dangerous because a device that was removed from the array
2136 * might have a event_count that still looks up-to-date,
2137 * so it can be re-added without a resync.
2138 * Pointless because if there are any spares to skip,
2139 * then a recovery will happen and soon that array won't
2140 * be degraded any more and the spare can go back to sleep then.
2144 sync_req
= mddev
->in_sync
;
2146 /* If this is just a dirty<->clean transition, and the array is clean
2147 * and 'events' is odd, we can roll back to the previous clean state */
2149 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2150 && mddev
->can_decrease_events
2151 && mddev
->events
!= 1) {
2153 mddev
->can_decrease_events
= 0;
2155 /* otherwise we have to go forward and ... */
2157 mddev
->can_decrease_events
= nospares
;
2160 if (!mddev
->events
) {
2162 * oops, this 64-bit counter should never wrap.
2163 * Either we are in around ~1 trillion A.C., assuming
2164 * 1 reboot per second, or we have a bug:
2171 * do not write anything to disk if using
2172 * nonpersistent superblocks
2174 if (!mddev
->persistent
) {
2175 if (!mddev
->external
)
2176 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2178 spin_unlock_irq(&mddev
->write_lock
);
2179 wake_up(&mddev
->sb_wait
);
2182 sync_sbs(mddev
, nospares
);
2183 spin_unlock_irq(&mddev
->write_lock
);
2186 "md: updating %s RAID superblock on device (in sync %d)\n",
2187 mdname(mddev
),mddev
->in_sync
);
2189 bitmap_update_sb(mddev
->bitmap
);
2190 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2191 char b
[BDEVNAME_SIZE
];
2192 dprintk(KERN_INFO
"md: ");
2193 if (rdev
->sb_loaded
!= 1)
2194 continue; /* no noise on spare devices */
2195 if (test_bit(Faulty
, &rdev
->flags
))
2196 dprintk("(skipping faulty ");
2198 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2199 if (!test_bit(Faulty
, &rdev
->flags
)) {
2200 md_super_write(mddev
,rdev
,
2201 rdev
->sb_start
, rdev
->sb_size
,
2203 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2204 bdevname(rdev
->bdev
,b
),
2205 (unsigned long long)rdev
->sb_start
);
2206 rdev
->sb_events
= mddev
->events
;
2210 if (mddev
->level
== LEVEL_MULTIPATH
)
2211 /* only need to write one superblock... */
2214 md_super_wait(mddev
);
2215 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2217 spin_lock_irq(&mddev
->write_lock
);
2218 if (mddev
->in_sync
!= sync_req
||
2219 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2220 /* have to write it out again */
2221 spin_unlock_irq(&mddev
->write_lock
);
2224 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2225 spin_unlock_irq(&mddev
->write_lock
);
2226 wake_up(&mddev
->sb_wait
);
2227 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2228 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2232 /* words written to sysfs files may, or may not, be \n terminated.
2233 * We want to accept with case. For this we use cmd_match.
2235 static int cmd_match(const char *cmd
, const char *str
)
2237 /* See if cmd, written into a sysfs file, matches
2238 * str. They must either be the same, or cmd can
2239 * have a trailing newline
2241 while (*cmd
&& *str
&& *cmd
== *str
) {
2252 struct rdev_sysfs_entry
{
2253 struct attribute attr
;
2254 ssize_t (*show
)(mdk_rdev_t
*, char *);
2255 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2259 state_show(mdk_rdev_t
*rdev
, char *page
)
2264 if (test_bit(Faulty
, &rdev
->flags
)) {
2265 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2268 if (test_bit(In_sync
, &rdev
->flags
)) {
2269 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2272 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2273 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2276 if (test_bit(Blocked
, &rdev
->flags
)) {
2277 len
+= sprintf(page
+len
, "%sblocked", sep
);
2280 if (!test_bit(Faulty
, &rdev
->flags
) &&
2281 !test_bit(In_sync
, &rdev
->flags
)) {
2282 len
+= sprintf(page
+len
, "%sspare", sep
);
2285 return len
+sprintf(page
+len
, "\n");
2289 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2292 * faulty - simulates and error
2293 * remove - disconnects the device
2294 * writemostly - sets write_mostly
2295 * -writemostly - clears write_mostly
2296 * blocked - sets the Blocked flag
2297 * -blocked - clears the Blocked flag
2298 * insync - sets Insync providing device isn't active
2301 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2302 md_error(rdev
->mddev
, rdev
);
2304 } else if (cmd_match(buf
, "remove")) {
2305 if (rdev
->raid_disk
>= 0)
2308 mddev_t
*mddev
= rdev
->mddev
;
2309 kick_rdev_from_array(rdev
);
2311 md_update_sb(mddev
, 1);
2312 md_new_event(mddev
);
2315 } else if (cmd_match(buf
, "writemostly")) {
2316 set_bit(WriteMostly
, &rdev
->flags
);
2318 } else if (cmd_match(buf
, "-writemostly")) {
2319 clear_bit(WriteMostly
, &rdev
->flags
);
2321 } else if (cmd_match(buf
, "blocked")) {
2322 set_bit(Blocked
, &rdev
->flags
);
2324 } else if (cmd_match(buf
, "-blocked")) {
2325 clear_bit(Blocked
, &rdev
->flags
);
2326 wake_up(&rdev
->blocked_wait
);
2327 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2328 md_wakeup_thread(rdev
->mddev
->thread
);
2331 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2332 set_bit(In_sync
, &rdev
->flags
);
2335 if (!err
&& rdev
->sysfs_state
)
2336 sysfs_notify_dirent(rdev
->sysfs_state
);
2337 return err
? err
: len
;
2339 static struct rdev_sysfs_entry rdev_state
=
2340 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2343 errors_show(mdk_rdev_t
*rdev
, char *page
)
2345 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2349 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2352 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2353 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2354 atomic_set(&rdev
->corrected_errors
, n
);
2359 static struct rdev_sysfs_entry rdev_errors
=
2360 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2363 slot_show(mdk_rdev_t
*rdev
, char *page
)
2365 if (rdev
->raid_disk
< 0)
2366 return sprintf(page
, "none\n");
2368 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2372 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2377 int slot
= simple_strtoul(buf
, &e
, 10);
2378 if (strncmp(buf
, "none", 4)==0)
2380 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2382 if (rdev
->mddev
->pers
&& slot
== -1) {
2383 /* Setting 'slot' on an active array requires also
2384 * updating the 'rd%d' link, and communicating
2385 * with the personality with ->hot_*_disk.
2386 * For now we only support removing
2387 * failed/spare devices. This normally happens automatically,
2388 * but not when the metadata is externally managed.
2390 if (rdev
->raid_disk
== -1)
2392 /* personality does all needed checks */
2393 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2395 err
= rdev
->mddev
->pers
->
2396 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2399 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2400 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2401 rdev
->raid_disk
= -1;
2402 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2403 md_wakeup_thread(rdev
->mddev
->thread
);
2404 } else if (rdev
->mddev
->pers
) {
2406 /* Activating a spare .. or possibly reactivating
2407 * if we ever get bitmaps working here.
2410 if (rdev
->raid_disk
!= -1)
2413 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2416 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2417 if (rdev2
->raid_disk
== slot
)
2420 rdev
->raid_disk
= slot
;
2421 if (test_bit(In_sync
, &rdev
->flags
))
2422 rdev
->saved_raid_disk
= slot
;
2424 rdev
->saved_raid_disk
= -1;
2425 err
= rdev
->mddev
->pers
->
2426 hot_add_disk(rdev
->mddev
, rdev
);
2428 rdev
->raid_disk
= -1;
2431 sysfs_notify_dirent(rdev
->sysfs_state
);
2432 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2433 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2435 "md: cannot register "
2437 nm
, mdname(rdev
->mddev
));
2439 /* don't wakeup anyone, leave that to userspace. */
2441 if (slot
>= rdev
->mddev
->raid_disks
)
2443 rdev
->raid_disk
= slot
;
2444 /* assume it is working */
2445 clear_bit(Faulty
, &rdev
->flags
);
2446 clear_bit(WriteMostly
, &rdev
->flags
);
2447 set_bit(In_sync
, &rdev
->flags
);
2448 sysfs_notify_dirent(rdev
->sysfs_state
);
2454 static struct rdev_sysfs_entry rdev_slot
=
2455 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2458 offset_show(mdk_rdev_t
*rdev
, char *page
)
2460 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2464 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2467 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2468 if (e
==buf
|| (*e
&& *e
!= '\n'))
2470 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2472 if (rdev
->sectors
&& rdev
->mddev
->external
)
2473 /* Must set offset before size, so overlap checks
2476 rdev
->data_offset
= offset
;
2480 static struct rdev_sysfs_entry rdev_offset
=
2481 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2484 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2486 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2489 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2491 /* check if two start/length pairs overlap */
2499 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2501 unsigned long long blocks
;
2504 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2507 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2508 return -EINVAL
; /* sector conversion overflow */
2511 if (new != blocks
* 2)
2512 return -EINVAL
; /* unsigned long long to sector_t overflow */
2519 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2521 mddev_t
*my_mddev
= rdev
->mddev
;
2522 sector_t oldsectors
= rdev
->sectors
;
2525 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2527 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2528 if (my_mddev
->persistent
) {
2529 sectors
= super_types
[my_mddev
->major_version
].
2530 rdev_size_change(rdev
, sectors
);
2533 } else if (!sectors
)
2534 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2537 if (sectors
< my_mddev
->dev_sectors
)
2538 return -EINVAL
; /* component must fit device */
2540 rdev
->sectors
= sectors
;
2541 if (sectors
> oldsectors
&& my_mddev
->external
) {
2542 /* need to check that all other rdevs with the same ->bdev
2543 * do not overlap. We need to unlock the mddev to avoid
2544 * a deadlock. We have already changed rdev->sectors, and if
2545 * we have to change it back, we will have the lock again.
2549 struct list_head
*tmp
;
2551 mddev_unlock(my_mddev
);
2552 for_each_mddev(mddev
, tmp
) {
2556 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2557 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2558 (rdev
->bdev
== rdev2
->bdev
&&
2560 overlaps(rdev
->data_offset
, rdev
->sectors
,
2566 mddev_unlock(mddev
);
2572 mddev_lock(my_mddev
);
2574 /* Someone else could have slipped in a size
2575 * change here, but doing so is just silly.
2576 * We put oldsectors back because we *know* it is
2577 * safe, and trust userspace not to race with
2580 rdev
->sectors
= oldsectors
;
2587 static struct rdev_sysfs_entry rdev_size
=
2588 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2591 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2593 unsigned long long recovery_start
= rdev
->recovery_offset
;
2595 if (test_bit(In_sync
, &rdev
->flags
) ||
2596 recovery_start
== MaxSector
)
2597 return sprintf(page
, "none\n");
2599 return sprintf(page
, "%llu\n", recovery_start
);
2602 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2604 unsigned long long recovery_start
;
2606 if (cmd_match(buf
, "none"))
2607 recovery_start
= MaxSector
;
2608 else if (strict_strtoull(buf
, 10, &recovery_start
))
2611 if (rdev
->mddev
->pers
&&
2612 rdev
->raid_disk
>= 0)
2615 rdev
->recovery_offset
= recovery_start
;
2616 if (recovery_start
== MaxSector
)
2617 set_bit(In_sync
, &rdev
->flags
);
2619 clear_bit(In_sync
, &rdev
->flags
);
2623 static struct rdev_sysfs_entry rdev_recovery_start
=
2624 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2626 static struct attribute
*rdev_default_attrs
[] = {
2632 &rdev_recovery_start
.attr
,
2636 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2638 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2639 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2640 mddev_t
*mddev
= rdev
->mddev
;
2646 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2648 if (rdev
->mddev
== NULL
)
2651 rv
= entry
->show(rdev
, page
);
2652 mddev_unlock(mddev
);
2658 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2659 const char *page
, size_t length
)
2661 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2662 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2664 mddev_t
*mddev
= rdev
->mddev
;
2668 if (!capable(CAP_SYS_ADMIN
))
2670 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2672 if (rdev
->mddev
== NULL
)
2675 rv
= entry
->store(rdev
, page
, length
);
2676 mddev_unlock(mddev
);
2681 static void rdev_free(struct kobject
*ko
)
2683 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2686 static struct sysfs_ops rdev_sysfs_ops
= {
2687 .show
= rdev_attr_show
,
2688 .store
= rdev_attr_store
,
2690 static struct kobj_type rdev_ktype
= {
2691 .release
= rdev_free
,
2692 .sysfs_ops
= &rdev_sysfs_ops
,
2693 .default_attrs
= rdev_default_attrs
,
2697 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2699 * mark the device faulty if:
2701 * - the device is nonexistent (zero size)
2702 * - the device has no valid superblock
2704 * a faulty rdev _never_ has rdev->sb set.
2706 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2708 char b
[BDEVNAME_SIZE
];
2713 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2715 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2716 return ERR_PTR(-ENOMEM
);
2719 if ((err
= alloc_disk_sb(rdev
)))
2722 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2726 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2729 rdev
->saved_raid_disk
= -1;
2730 rdev
->raid_disk
= -1;
2732 rdev
->data_offset
= 0;
2733 rdev
->sb_events
= 0;
2734 rdev
->last_read_error
.tv_sec
= 0;
2735 rdev
->last_read_error
.tv_nsec
= 0;
2736 atomic_set(&rdev
->nr_pending
, 0);
2737 atomic_set(&rdev
->read_errors
, 0);
2738 atomic_set(&rdev
->corrected_errors
, 0);
2740 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2743 "md: %s has zero or unknown size, marking faulty!\n",
2744 bdevname(rdev
->bdev
,b
));
2749 if (super_format
>= 0) {
2750 err
= super_types
[super_format
].
2751 load_super(rdev
, NULL
, super_minor
);
2752 if (err
== -EINVAL
) {
2754 "md: %s does not have a valid v%d.%d "
2755 "superblock, not importing!\n",
2756 bdevname(rdev
->bdev
,b
),
2757 super_format
, super_minor
);
2762 "md: could not read %s's sb, not importing!\n",
2763 bdevname(rdev
->bdev
,b
));
2768 INIT_LIST_HEAD(&rdev
->same_set
);
2769 init_waitqueue_head(&rdev
->blocked_wait
);
2774 if (rdev
->sb_page
) {
2780 return ERR_PTR(err
);
2784 * Check a full RAID array for plausibility
2788 static void analyze_sbs(mddev_t
* mddev
)
2791 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2792 char b
[BDEVNAME_SIZE
];
2795 rdev_for_each(rdev
, tmp
, mddev
)
2796 switch (super_types
[mddev
->major_version
].
2797 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2805 "md: fatal superblock inconsistency in %s"
2806 " -- removing from array\n",
2807 bdevname(rdev
->bdev
,b
));
2808 kick_rdev_from_array(rdev
);
2812 super_types
[mddev
->major_version
].
2813 validate_super(mddev
, freshest
);
2816 rdev_for_each(rdev
, tmp
, mddev
) {
2817 if (mddev
->max_disks
&&
2818 (rdev
->desc_nr
>= mddev
->max_disks
||
2819 i
> mddev
->max_disks
)) {
2821 "md: %s: %s: only %d devices permitted\n",
2822 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2824 kick_rdev_from_array(rdev
);
2827 if (rdev
!= freshest
)
2828 if (super_types
[mddev
->major_version
].
2829 validate_super(mddev
, rdev
)) {
2830 printk(KERN_WARNING
"md: kicking non-fresh %s"
2832 bdevname(rdev
->bdev
,b
));
2833 kick_rdev_from_array(rdev
);
2836 if (mddev
->level
== LEVEL_MULTIPATH
) {
2837 rdev
->desc_nr
= i
++;
2838 rdev
->raid_disk
= rdev
->desc_nr
;
2839 set_bit(In_sync
, &rdev
->flags
);
2840 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2841 rdev
->raid_disk
= -1;
2842 clear_bit(In_sync
, &rdev
->flags
);
2847 /* Read a fixed-point number.
2848 * Numbers in sysfs attributes should be in "standard" units where
2849 * possible, so time should be in seconds.
2850 * However we internally use a a much smaller unit such as
2851 * milliseconds or jiffies.
2852 * This function takes a decimal number with a possible fractional
2853 * component, and produces an integer which is the result of
2854 * multiplying that number by 10^'scale'.
2855 * all without any floating-point arithmetic.
2857 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
2859 unsigned long result
= 0;
2861 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
2864 else if (decimals
< scale
) {
2867 result
= result
* 10 + value
;
2879 while (decimals
< scale
) {
2888 static void md_safemode_timeout(unsigned long data
);
2891 safe_delay_show(mddev_t
*mddev
, char *page
)
2893 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2894 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2897 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2901 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
2904 mddev
->safemode_delay
= 0;
2906 unsigned long old_delay
= mddev
->safemode_delay
;
2907 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2908 if (mddev
->safemode_delay
== 0)
2909 mddev
->safemode_delay
= 1;
2910 if (mddev
->safemode_delay
< old_delay
)
2911 md_safemode_timeout((unsigned long)mddev
);
2915 static struct md_sysfs_entry md_safe_delay
=
2916 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2919 level_show(mddev_t
*mddev
, char *page
)
2921 struct mdk_personality
*p
= mddev
->pers
;
2923 return sprintf(page
, "%s\n", p
->name
);
2924 else if (mddev
->clevel
[0])
2925 return sprintf(page
, "%s\n", mddev
->clevel
);
2926 else if (mddev
->level
!= LEVEL_NONE
)
2927 return sprintf(page
, "%d\n", mddev
->level
);
2933 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2937 struct mdk_personality
*pers
;
2942 if (mddev
->pers
== NULL
) {
2945 if (len
>= sizeof(mddev
->clevel
))
2947 strncpy(mddev
->clevel
, buf
, len
);
2948 if (mddev
->clevel
[len
-1] == '\n')
2950 mddev
->clevel
[len
] = 0;
2951 mddev
->level
= LEVEL_NONE
;
2955 /* request to change the personality. Need to ensure:
2956 * - array is not engaged in resync/recovery/reshape
2957 * - old personality can be suspended
2958 * - new personality will access other array.
2961 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2964 if (!mddev
->pers
->quiesce
) {
2965 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2966 mdname(mddev
), mddev
->pers
->name
);
2970 /* Now find the new personality */
2971 if (len
== 0 || len
>= sizeof(clevel
))
2973 strncpy(clevel
, buf
, len
);
2974 if (clevel
[len
-1] == '\n')
2977 if (strict_strtol(clevel
, 10, &level
))
2980 if (request_module("md-%s", clevel
) != 0)
2981 request_module("md-level-%s", clevel
);
2982 spin_lock(&pers_lock
);
2983 pers
= find_pers(level
, clevel
);
2984 if (!pers
|| !try_module_get(pers
->owner
)) {
2985 spin_unlock(&pers_lock
);
2986 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
2989 spin_unlock(&pers_lock
);
2991 if (pers
== mddev
->pers
) {
2992 /* Nothing to do! */
2993 module_put(pers
->owner
);
2996 if (!pers
->takeover
) {
2997 module_put(pers
->owner
);
2998 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2999 mdname(mddev
), clevel
);
3003 /* ->takeover must set new_* and/or delta_disks
3004 * if it succeeds, and may set them when it fails.
3006 priv
= pers
->takeover(mddev
);
3008 mddev
->new_level
= mddev
->level
;
3009 mddev
->new_layout
= mddev
->layout
;
3010 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3011 mddev
->raid_disks
-= mddev
->delta_disks
;
3012 mddev
->delta_disks
= 0;
3013 module_put(pers
->owner
);
3014 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3015 mdname(mddev
), clevel
);
3016 return PTR_ERR(priv
);
3019 /* Looks like we have a winner */
3020 mddev_suspend(mddev
);
3021 mddev
->pers
->stop(mddev
);
3023 if (mddev
->pers
->sync_request
== NULL
&&
3024 pers
->sync_request
!= NULL
) {
3025 /* need to add the md_redundancy_group */
3026 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3028 "md: cannot register extra attributes for %s\n",
3030 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3032 if (mddev
->pers
->sync_request
!= NULL
&&
3033 pers
->sync_request
== NULL
) {
3034 /* need to remove the md_redundancy_group */
3035 if (mddev
->to_remove
== NULL
)
3036 mddev
->to_remove
= &md_redundancy_group
;
3039 if (mddev
->pers
->sync_request
== NULL
&&
3041 /* We are converting from a no-redundancy array
3042 * to a redundancy array and metadata is managed
3043 * externally so we need to be sure that writes
3044 * won't block due to a need to transition
3046 * until external management is started.
3049 mddev
->safemode_delay
= 0;
3050 mddev
->safemode
= 0;
3053 module_put(mddev
->pers
->owner
);
3054 /* Invalidate devices that are now superfluous */
3055 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3056 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
3057 rdev
->raid_disk
= -1;
3058 clear_bit(In_sync
, &rdev
->flags
);
3061 mddev
->private = priv
;
3062 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3063 mddev
->level
= mddev
->new_level
;
3064 mddev
->layout
= mddev
->new_layout
;
3065 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3066 mddev
->delta_disks
= 0;
3067 if (mddev
->pers
->sync_request
== NULL
) {
3068 /* this is now an array without redundancy, so
3069 * it must always be in_sync
3072 del_timer_sync(&mddev
->safemode_timer
);
3075 mddev_resume(mddev
);
3076 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3077 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3078 md_wakeup_thread(mddev
->thread
);
3079 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3080 md_new_event(mddev
);
3084 static struct md_sysfs_entry md_level
=
3085 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3089 layout_show(mddev_t
*mddev
, char *page
)
3091 /* just a number, not meaningful for all levels */
3092 if (mddev
->reshape_position
!= MaxSector
&&
3093 mddev
->layout
!= mddev
->new_layout
)
3094 return sprintf(page
, "%d (%d)\n",
3095 mddev
->new_layout
, mddev
->layout
);
3096 return sprintf(page
, "%d\n", mddev
->layout
);
3100 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3103 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3105 if (!*buf
|| (*e
&& *e
!= '\n'))
3110 if (mddev
->pers
->check_reshape
== NULL
)
3112 mddev
->new_layout
= n
;
3113 err
= mddev
->pers
->check_reshape(mddev
);
3115 mddev
->new_layout
= mddev
->layout
;
3119 mddev
->new_layout
= n
;
3120 if (mddev
->reshape_position
== MaxSector
)
3125 static struct md_sysfs_entry md_layout
=
3126 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3130 raid_disks_show(mddev_t
*mddev
, char *page
)
3132 if (mddev
->raid_disks
== 0)
3134 if (mddev
->reshape_position
!= MaxSector
&&
3135 mddev
->delta_disks
!= 0)
3136 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3137 mddev
->raid_disks
- mddev
->delta_disks
);
3138 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3141 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3144 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3148 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3150 if (!*buf
|| (*e
&& *e
!= '\n'))
3154 rv
= update_raid_disks(mddev
, n
);
3155 else if (mddev
->reshape_position
!= MaxSector
) {
3156 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3157 mddev
->delta_disks
= n
- olddisks
;
3158 mddev
->raid_disks
= n
;
3160 mddev
->raid_disks
= n
;
3161 return rv
? rv
: len
;
3163 static struct md_sysfs_entry md_raid_disks
=
3164 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3167 chunk_size_show(mddev_t
*mddev
, char *page
)
3169 if (mddev
->reshape_position
!= MaxSector
&&
3170 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3171 return sprintf(page
, "%d (%d)\n",
3172 mddev
->new_chunk_sectors
<< 9,
3173 mddev
->chunk_sectors
<< 9);
3174 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3178 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3181 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3183 if (!*buf
|| (*e
&& *e
!= '\n'))
3188 if (mddev
->pers
->check_reshape
== NULL
)
3190 mddev
->new_chunk_sectors
= n
>> 9;
3191 err
= mddev
->pers
->check_reshape(mddev
);
3193 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3197 mddev
->new_chunk_sectors
= n
>> 9;
3198 if (mddev
->reshape_position
== MaxSector
)
3199 mddev
->chunk_sectors
= n
>> 9;
3203 static struct md_sysfs_entry md_chunk_size
=
3204 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3207 resync_start_show(mddev_t
*mddev
, char *page
)
3209 if (mddev
->recovery_cp
== MaxSector
)
3210 return sprintf(page
, "none\n");
3211 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3215 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3218 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3222 if (cmd_match(buf
, "none"))
3224 else if (!*buf
|| (*e
&& *e
!= '\n'))
3227 mddev
->recovery_cp
= n
;
3230 static struct md_sysfs_entry md_resync_start
=
3231 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3234 * The array state can be:
3237 * No devices, no size, no level
3238 * Equivalent to STOP_ARRAY ioctl
3240 * May have some settings, but array is not active
3241 * all IO results in error
3242 * When written, doesn't tear down array, but just stops it
3243 * suspended (not supported yet)
3244 * All IO requests will block. The array can be reconfigured.
3245 * Writing this, if accepted, will block until array is quiescent
3247 * no resync can happen. no superblocks get written.
3248 * write requests fail
3250 * like readonly, but behaves like 'clean' on a write request.
3252 * clean - no pending writes, but otherwise active.
3253 * When written to inactive array, starts without resync
3254 * If a write request arrives then
3255 * if metadata is known, mark 'dirty' and switch to 'active'.
3256 * if not known, block and switch to write-pending
3257 * If written to an active array that has pending writes, then fails.
3259 * fully active: IO and resync can be happening.
3260 * When written to inactive array, starts with resync
3263 * clean, but writes are blocked waiting for 'active' to be written.
3266 * like active, but no writes have been seen for a while (100msec).
3269 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3270 write_pending
, active_idle
, bad_word
};
3271 static char *array_states
[] = {
3272 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3273 "write-pending", "active-idle", NULL
};
3275 static int match_word(const char *word
, char **list
)
3278 for (n
=0; list
[n
]; n
++)
3279 if (cmd_match(word
, list
[n
]))
3285 array_state_show(mddev_t
*mddev
, char *page
)
3287 enum array_state st
= inactive
;
3300 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3302 else if (mddev
->safemode
)
3308 if (list_empty(&mddev
->disks
) &&
3309 mddev
->raid_disks
== 0 &&
3310 mddev
->dev_sectors
== 0)
3315 return sprintf(page
, "%s\n", array_states
[st
]);
3318 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3319 static int md_set_readonly(mddev_t
* mddev
, int is_open
);
3320 static int do_md_run(mddev_t
* mddev
);
3321 static int restart_array(mddev_t
*mddev
);
3324 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3327 enum array_state st
= match_word(buf
, array_states
);
3332 /* stopping an active array */
3333 if (atomic_read(&mddev
->openers
) > 0)
3335 err
= do_md_stop(mddev
, 0, 0);
3338 /* stopping an active array */
3340 if (atomic_read(&mddev
->openers
) > 0)
3342 err
= do_md_stop(mddev
, 2, 0);
3344 err
= 0; /* already inactive */
3347 break; /* not supported yet */
3350 err
= md_set_readonly(mddev
, 0);
3353 set_disk_ro(mddev
->gendisk
, 1);
3354 err
= do_md_run(mddev
);
3360 err
= md_set_readonly(mddev
, 0);
3361 else if (mddev
->ro
== 1)
3362 err
= restart_array(mddev
);
3365 set_disk_ro(mddev
->gendisk
, 0);
3369 err
= do_md_run(mddev
);
3374 restart_array(mddev
);
3375 spin_lock_irq(&mddev
->write_lock
);
3376 if (atomic_read(&mddev
->writes_pending
) == 0) {
3377 if (mddev
->in_sync
== 0) {
3379 if (mddev
->safemode
== 1)
3380 mddev
->safemode
= 0;
3381 if (mddev
->persistent
)
3382 set_bit(MD_CHANGE_CLEAN
,
3388 spin_unlock_irq(&mddev
->write_lock
);
3394 restart_array(mddev
);
3395 if (mddev
->external
)
3396 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3397 wake_up(&mddev
->sb_wait
);
3401 set_disk_ro(mddev
->gendisk
, 0);
3402 err
= do_md_run(mddev
);
3407 /* these cannot be set */
3413 sysfs_notify_dirent(mddev
->sysfs_state
);
3417 static struct md_sysfs_entry md_array_state
=
3418 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3421 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3422 return sprintf(page
, "%d\n",
3423 atomic_read(&mddev
->max_corr_read_errors
));
3427 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3430 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3432 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3433 atomic_set(&mddev
->max_corr_read_errors
, n
);
3439 static struct md_sysfs_entry max_corr_read_errors
=
3440 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3441 max_corrected_read_errors_store
);
3444 null_show(mddev_t
*mddev
, char *page
)
3450 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3452 /* buf must be %d:%d\n? giving major and minor numbers */
3453 /* The new device is added to the array.
3454 * If the array has a persistent superblock, we read the
3455 * superblock to initialise info and check validity.
3456 * Otherwise, only checking done is that in bind_rdev_to_array,
3457 * which mainly checks size.
3460 int major
= simple_strtoul(buf
, &e
, 10);
3466 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3468 minor
= simple_strtoul(e
+1, &e
, 10);
3469 if (*e
&& *e
!= '\n')
3471 dev
= MKDEV(major
, minor
);
3472 if (major
!= MAJOR(dev
) ||
3473 minor
!= MINOR(dev
))
3477 if (mddev
->persistent
) {
3478 rdev
= md_import_device(dev
, mddev
->major_version
,
3479 mddev
->minor_version
);
3480 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3481 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3482 mdk_rdev_t
, same_set
);
3483 err
= super_types
[mddev
->major_version
]
3484 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3488 } else if (mddev
->external
)
3489 rdev
= md_import_device(dev
, -2, -1);
3491 rdev
= md_import_device(dev
, -1, -1);
3494 return PTR_ERR(rdev
);
3495 err
= bind_rdev_to_array(rdev
, mddev
);
3499 return err
? err
: len
;
3502 static struct md_sysfs_entry md_new_device
=
3503 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3506 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3509 unsigned long chunk
, end_chunk
;
3513 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3515 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3516 if (buf
== end
) break;
3517 if (*end
== '-') { /* range */
3519 end_chunk
= simple_strtoul(buf
, &end
, 0);
3520 if (buf
== end
) break;
3522 if (*end
&& !isspace(*end
)) break;
3523 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3524 buf
= skip_spaces(end
);
3526 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3531 static struct md_sysfs_entry md_bitmap
=
3532 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3535 size_show(mddev_t
*mddev
, char *page
)
3537 return sprintf(page
, "%llu\n",
3538 (unsigned long long)mddev
->dev_sectors
/ 2);
3541 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3544 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3546 /* If array is inactive, we can reduce the component size, but
3547 * not increase it (except from 0).
3548 * If array is active, we can try an on-line resize
3551 int err
= strict_blocks_to_sectors(buf
, §ors
);
3556 err
= update_size(mddev
, sectors
);
3557 md_update_sb(mddev
, 1);
3559 if (mddev
->dev_sectors
== 0 ||
3560 mddev
->dev_sectors
> sectors
)
3561 mddev
->dev_sectors
= sectors
;
3565 return err
? err
: len
;
3568 static struct md_sysfs_entry md_size
=
3569 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3574 * 'none' for arrays with no metadata (good luck...)
3575 * 'external' for arrays with externally managed metadata,
3576 * or N.M for internally known formats
3579 metadata_show(mddev_t
*mddev
, char *page
)
3581 if (mddev
->persistent
)
3582 return sprintf(page
, "%d.%d\n",
3583 mddev
->major_version
, mddev
->minor_version
);
3584 else if (mddev
->external
)
3585 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3587 return sprintf(page
, "none\n");
3591 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3595 /* Changing the details of 'external' metadata is
3596 * always permitted. Otherwise there must be
3597 * no devices attached to the array.
3599 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3601 else if (!list_empty(&mddev
->disks
))
3604 if (cmd_match(buf
, "none")) {
3605 mddev
->persistent
= 0;
3606 mddev
->external
= 0;
3607 mddev
->major_version
= 0;
3608 mddev
->minor_version
= 90;
3611 if (strncmp(buf
, "external:", 9) == 0) {
3612 size_t namelen
= len
-9;
3613 if (namelen
>= sizeof(mddev
->metadata_type
))
3614 namelen
= sizeof(mddev
->metadata_type
)-1;
3615 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3616 mddev
->metadata_type
[namelen
] = 0;
3617 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3618 mddev
->metadata_type
[--namelen
] = 0;
3619 mddev
->persistent
= 0;
3620 mddev
->external
= 1;
3621 mddev
->major_version
= 0;
3622 mddev
->minor_version
= 90;
3625 major
= simple_strtoul(buf
, &e
, 10);
3626 if (e
==buf
|| *e
!= '.')
3629 minor
= simple_strtoul(buf
, &e
, 10);
3630 if (e
==buf
|| (*e
&& *e
!= '\n') )
3632 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3634 mddev
->major_version
= major
;
3635 mddev
->minor_version
= minor
;
3636 mddev
->persistent
= 1;
3637 mddev
->external
= 0;
3641 static struct md_sysfs_entry md_metadata
=
3642 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3645 action_show(mddev_t
*mddev
, char *page
)
3647 char *type
= "idle";
3648 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3650 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3651 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3652 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3654 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3655 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3657 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3661 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3664 return sprintf(page
, "%s\n", type
);
3668 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3670 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3673 if (cmd_match(page
, "frozen"))
3674 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3676 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3678 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3679 if (mddev
->sync_thread
) {
3680 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3681 md_unregister_thread(mddev
->sync_thread
);
3682 mddev
->sync_thread
= NULL
;
3683 mddev
->recovery
= 0;
3685 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3686 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3688 else if (cmd_match(page
, "resync"))
3689 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3690 else if (cmd_match(page
, "recover")) {
3691 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3692 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3693 } else if (cmd_match(page
, "reshape")) {
3695 if (mddev
->pers
->start_reshape
== NULL
)
3697 err
= mddev
->pers
->start_reshape(mddev
);
3700 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3702 if (cmd_match(page
, "check"))
3703 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3704 else if (!cmd_match(page
, "repair"))
3706 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3707 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3709 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3710 md_wakeup_thread(mddev
->thread
);
3711 sysfs_notify_dirent(mddev
->sysfs_action
);
3716 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3718 return sprintf(page
, "%llu\n",
3719 (unsigned long long) mddev
->resync_mismatches
);
3722 static struct md_sysfs_entry md_scan_mode
=
3723 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3726 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3729 sync_min_show(mddev_t
*mddev
, char *page
)
3731 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3732 mddev
->sync_speed_min
? "local": "system");
3736 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3740 if (strncmp(buf
, "system", 6)==0) {
3741 mddev
->sync_speed_min
= 0;
3744 min
= simple_strtoul(buf
, &e
, 10);
3745 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3747 mddev
->sync_speed_min
= min
;
3751 static struct md_sysfs_entry md_sync_min
=
3752 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3755 sync_max_show(mddev_t
*mddev
, char *page
)
3757 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3758 mddev
->sync_speed_max
? "local": "system");
3762 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3766 if (strncmp(buf
, "system", 6)==0) {
3767 mddev
->sync_speed_max
= 0;
3770 max
= simple_strtoul(buf
, &e
, 10);
3771 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3773 mddev
->sync_speed_max
= max
;
3777 static struct md_sysfs_entry md_sync_max
=
3778 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3781 degraded_show(mddev_t
*mddev
, char *page
)
3783 return sprintf(page
, "%d\n", mddev
->degraded
);
3785 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3788 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3790 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3794 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3798 if (strict_strtol(buf
, 10, &n
))
3801 if (n
!= 0 && n
!= 1)
3804 mddev
->parallel_resync
= n
;
3806 if (mddev
->sync_thread
)
3807 wake_up(&resync_wait
);
3812 /* force parallel resync, even with shared block devices */
3813 static struct md_sysfs_entry md_sync_force_parallel
=
3814 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3815 sync_force_parallel_show
, sync_force_parallel_store
);
3818 sync_speed_show(mddev_t
*mddev
, char *page
)
3820 unsigned long resync
, dt
, db
;
3821 if (mddev
->curr_resync
== 0)
3822 return sprintf(page
, "none\n");
3823 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3824 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3826 db
= resync
- mddev
->resync_mark_cnt
;
3827 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3830 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3833 sync_completed_show(mddev_t
*mddev
, char *page
)
3835 unsigned long max_sectors
, resync
;
3837 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3838 return sprintf(page
, "none\n");
3840 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3841 max_sectors
= mddev
->resync_max_sectors
;
3843 max_sectors
= mddev
->dev_sectors
;
3845 resync
= mddev
->curr_resync_completed
;
3846 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3849 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3852 min_sync_show(mddev_t
*mddev
, char *page
)
3854 return sprintf(page
, "%llu\n",
3855 (unsigned long long)mddev
->resync_min
);
3858 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3860 unsigned long long min
;
3861 if (strict_strtoull(buf
, 10, &min
))
3863 if (min
> mddev
->resync_max
)
3865 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3868 /* Must be a multiple of chunk_size */
3869 if (mddev
->chunk_sectors
) {
3870 sector_t temp
= min
;
3871 if (sector_div(temp
, mddev
->chunk_sectors
))
3874 mddev
->resync_min
= min
;
3879 static struct md_sysfs_entry md_min_sync
=
3880 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3883 max_sync_show(mddev_t
*mddev
, char *page
)
3885 if (mddev
->resync_max
== MaxSector
)
3886 return sprintf(page
, "max\n");
3888 return sprintf(page
, "%llu\n",
3889 (unsigned long long)mddev
->resync_max
);
3892 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3894 if (strncmp(buf
, "max", 3) == 0)
3895 mddev
->resync_max
= MaxSector
;
3897 unsigned long long max
;
3898 if (strict_strtoull(buf
, 10, &max
))
3900 if (max
< mddev
->resync_min
)
3902 if (max
< mddev
->resync_max
&&
3904 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3907 /* Must be a multiple of chunk_size */
3908 if (mddev
->chunk_sectors
) {
3909 sector_t temp
= max
;
3910 if (sector_div(temp
, mddev
->chunk_sectors
))
3913 mddev
->resync_max
= max
;
3915 wake_up(&mddev
->recovery_wait
);
3919 static struct md_sysfs_entry md_max_sync
=
3920 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3923 suspend_lo_show(mddev_t
*mddev
, char *page
)
3925 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3929 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3932 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3934 if (mddev
->pers
== NULL
||
3935 mddev
->pers
->quiesce
== NULL
)
3937 if (buf
== e
|| (*e
&& *e
!= '\n'))
3939 if (new >= mddev
->suspend_hi
||
3940 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3941 mddev
->suspend_lo
= new;
3942 mddev
->pers
->quiesce(mddev
, 2);
3947 static struct md_sysfs_entry md_suspend_lo
=
3948 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3952 suspend_hi_show(mddev_t
*mddev
, char *page
)
3954 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3958 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3961 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3963 if (mddev
->pers
== NULL
||
3964 mddev
->pers
->quiesce
== NULL
)
3966 if (buf
== e
|| (*e
&& *e
!= '\n'))
3968 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3969 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3970 mddev
->suspend_hi
= new;
3971 mddev
->pers
->quiesce(mddev
, 1);
3972 mddev
->pers
->quiesce(mddev
, 0);
3977 static struct md_sysfs_entry md_suspend_hi
=
3978 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3981 reshape_position_show(mddev_t
*mddev
, char *page
)
3983 if (mddev
->reshape_position
!= MaxSector
)
3984 return sprintf(page
, "%llu\n",
3985 (unsigned long long)mddev
->reshape_position
);
3986 strcpy(page
, "none\n");
3991 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3994 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3997 if (buf
== e
|| (*e
&& *e
!= '\n'))
3999 mddev
->reshape_position
= new;
4000 mddev
->delta_disks
= 0;
4001 mddev
->new_level
= mddev
->level
;
4002 mddev
->new_layout
= mddev
->layout
;
4003 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4007 static struct md_sysfs_entry md_reshape_position
=
4008 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4009 reshape_position_store
);
4012 array_size_show(mddev_t
*mddev
, char *page
)
4014 if (mddev
->external_size
)
4015 return sprintf(page
, "%llu\n",
4016 (unsigned long long)mddev
->array_sectors
/2);
4018 return sprintf(page
, "default\n");
4022 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4026 if (strncmp(buf
, "default", 7) == 0) {
4028 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4030 sectors
= mddev
->array_sectors
;
4032 mddev
->external_size
= 0;
4034 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4036 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4039 mddev
->external_size
= 1;
4042 mddev
->array_sectors
= sectors
;
4043 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4045 revalidate_disk(mddev
->gendisk
);
4050 static struct md_sysfs_entry md_array_size
=
4051 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4054 static struct attribute
*md_default_attrs
[] = {
4057 &md_raid_disks
.attr
,
4058 &md_chunk_size
.attr
,
4060 &md_resync_start
.attr
,
4062 &md_new_device
.attr
,
4063 &md_safe_delay
.attr
,
4064 &md_array_state
.attr
,
4065 &md_reshape_position
.attr
,
4066 &md_array_size
.attr
,
4067 &max_corr_read_errors
.attr
,
4071 static struct attribute
*md_redundancy_attrs
[] = {
4073 &md_mismatches
.attr
,
4076 &md_sync_speed
.attr
,
4077 &md_sync_force_parallel
.attr
,
4078 &md_sync_completed
.attr
,
4081 &md_suspend_lo
.attr
,
4082 &md_suspend_hi
.attr
,
4087 static struct attribute_group md_redundancy_group
= {
4089 .attrs
= md_redundancy_attrs
,
4094 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4096 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4097 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4102 rv
= mddev_lock(mddev
);
4104 rv
= entry
->show(mddev
, page
);
4105 mddev_unlock(mddev
);
4111 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4112 const char *page
, size_t length
)
4114 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4115 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4120 if (!capable(CAP_SYS_ADMIN
))
4122 rv
= mddev_lock(mddev
);
4123 if (mddev
->hold_active
== UNTIL_IOCTL
)
4124 mddev
->hold_active
= 0;
4126 rv
= entry
->store(mddev
, page
, length
);
4127 mddev_unlock(mddev
);
4132 static void md_free(struct kobject
*ko
)
4134 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4136 if (mddev
->sysfs_state
)
4137 sysfs_put(mddev
->sysfs_state
);
4139 if (mddev
->gendisk
) {
4140 del_gendisk(mddev
->gendisk
);
4141 put_disk(mddev
->gendisk
);
4144 blk_cleanup_queue(mddev
->queue
);
4149 static struct sysfs_ops md_sysfs_ops
= {
4150 .show
= md_attr_show
,
4151 .store
= md_attr_store
,
4153 static struct kobj_type md_ktype
= {
4155 .sysfs_ops
= &md_sysfs_ops
,
4156 .default_attrs
= md_default_attrs
,
4161 static void mddev_delayed_delete(struct work_struct
*ws
)
4163 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4165 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4166 kobject_del(&mddev
->kobj
);
4167 kobject_put(&mddev
->kobj
);
4170 static int md_alloc(dev_t dev
, char *name
)
4172 static DEFINE_MUTEX(disks_mutex
);
4173 mddev_t
*mddev
= mddev_find(dev
);
4174 struct gendisk
*disk
;
4183 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4184 shift
= partitioned
? MdpMinorShift
: 0;
4185 unit
= MINOR(mddev
->unit
) >> shift
;
4187 /* wait for any previous instance if this device
4188 * to be completed removed (mddev_delayed_delete).
4190 flush_scheduled_work();
4192 mutex_lock(&disks_mutex
);
4198 /* Need to ensure that 'name' is not a duplicate.
4201 spin_lock(&all_mddevs_lock
);
4203 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4204 if (mddev2
->gendisk
&&
4205 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4206 spin_unlock(&all_mddevs_lock
);
4209 spin_unlock(&all_mddevs_lock
);
4213 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4216 mddev
->queue
->queuedata
= mddev
;
4218 /* Can be unlocked because the queue is new: no concurrency */
4219 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4221 blk_queue_make_request(mddev
->queue
, md_make_request
);
4223 disk
= alloc_disk(1 << shift
);
4225 blk_cleanup_queue(mddev
->queue
);
4226 mddev
->queue
= NULL
;
4229 disk
->major
= MAJOR(mddev
->unit
);
4230 disk
->first_minor
= unit
<< shift
;
4232 strcpy(disk
->disk_name
, name
);
4233 else if (partitioned
)
4234 sprintf(disk
->disk_name
, "md_d%d", unit
);
4236 sprintf(disk
->disk_name
, "md%d", unit
);
4237 disk
->fops
= &md_fops
;
4238 disk
->private_data
= mddev
;
4239 disk
->queue
= mddev
->queue
;
4240 /* Allow extended partitions. This makes the
4241 * 'mdp' device redundant, but we can't really
4244 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4246 mddev
->gendisk
= disk
;
4247 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4248 &disk_to_dev(disk
)->kobj
, "%s", "md");
4250 /* This isn't possible, but as kobject_init_and_add is marked
4251 * __must_check, we must do something with the result
4253 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4257 if (sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4258 printk(KERN_DEBUG
"pointless warning\n");
4260 mutex_unlock(&disks_mutex
);
4262 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4263 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4269 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4271 md_alloc(dev
, NULL
);
4275 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4277 /* val must be "md_*" where * is not all digits.
4278 * We allocate an array with a large free minor number, and
4279 * set the name to val. val must not already be an active name.
4281 int len
= strlen(val
);
4282 char buf
[DISK_NAME_LEN
];
4284 while (len
&& val
[len
-1] == '\n')
4286 if (len
>= DISK_NAME_LEN
)
4288 strlcpy(buf
, val
, len
+1);
4289 if (strncmp(buf
, "md_", 3) != 0)
4291 return md_alloc(0, buf
);
4294 static void md_safemode_timeout(unsigned long data
)
4296 mddev_t
*mddev
= (mddev_t
*) data
;
4298 if (!atomic_read(&mddev
->writes_pending
)) {
4299 mddev
->safemode
= 1;
4300 if (mddev
->external
)
4301 sysfs_notify_dirent(mddev
->sysfs_state
);
4303 md_wakeup_thread(mddev
->thread
);
4306 static int start_dirty_degraded
;
4308 static int md_run(mddev_t
*mddev
)
4312 struct mdk_personality
*pers
;
4314 if (list_empty(&mddev
->disks
))
4315 /* cannot run an array with no devices.. */
4321 /* These two calls synchronise us with the
4322 * sysfs_remove_group calls in mddev_unlock,
4323 * so they must have completed.
4325 mutex_lock(&mddev
->open_mutex
);
4326 mutex_unlock(&mddev
->open_mutex
);
4329 * Analyze all RAID superblock(s)
4331 if (!mddev
->raid_disks
) {
4332 if (!mddev
->persistent
)
4337 if (mddev
->level
!= LEVEL_NONE
)
4338 request_module("md-level-%d", mddev
->level
);
4339 else if (mddev
->clevel
[0])
4340 request_module("md-%s", mddev
->clevel
);
4343 * Drop all container device buffers, from now on
4344 * the only valid external interface is through the md
4347 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4348 if (test_bit(Faulty
, &rdev
->flags
))
4350 sync_blockdev(rdev
->bdev
);
4351 invalidate_bdev(rdev
->bdev
);
4353 /* perform some consistency tests on the device.
4354 * We don't want the data to overlap the metadata,
4355 * Internal Bitmap issues have been handled elsewhere.
4357 if (rdev
->data_offset
< rdev
->sb_start
) {
4358 if (mddev
->dev_sectors
&&
4359 rdev
->data_offset
+ mddev
->dev_sectors
4361 printk("md: %s: data overlaps metadata\n",
4366 if (rdev
->sb_start
+ rdev
->sb_size
/512
4367 > rdev
->data_offset
) {
4368 printk("md: %s: metadata overlaps data\n",
4373 sysfs_notify_dirent(rdev
->sysfs_state
);
4376 spin_lock(&pers_lock
);
4377 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4378 if (!pers
|| !try_module_get(pers
->owner
)) {
4379 spin_unlock(&pers_lock
);
4380 if (mddev
->level
!= LEVEL_NONE
)
4381 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4384 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4389 spin_unlock(&pers_lock
);
4390 if (mddev
->level
!= pers
->level
) {
4391 mddev
->level
= pers
->level
;
4392 mddev
->new_level
= pers
->level
;
4394 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4396 if (mddev
->reshape_position
!= MaxSector
&&
4397 pers
->start_reshape
== NULL
) {
4398 /* This personality cannot handle reshaping... */
4400 module_put(pers
->owner
);
4404 if (pers
->sync_request
) {
4405 /* Warn if this is a potentially silly
4408 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4412 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4413 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4415 rdev
->bdev
->bd_contains
==
4416 rdev2
->bdev
->bd_contains
) {
4418 "%s: WARNING: %s appears to be"
4419 " on the same physical disk as"
4422 bdevname(rdev
->bdev
,b
),
4423 bdevname(rdev2
->bdev
,b2
));
4430 "True protection against single-disk"
4431 " failure might be compromised.\n");
4434 mddev
->recovery
= 0;
4435 /* may be over-ridden by personality */
4436 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4438 mddev
->barriers_work
= 1;
4439 mddev
->ok_start_degraded
= start_dirty_degraded
;
4441 if (start_readonly
&& mddev
->ro
== 0)
4442 mddev
->ro
= 2; /* read-only, but switch on first write */
4444 err
= mddev
->pers
->run(mddev
);
4446 printk(KERN_ERR
"md: pers->run() failed ...\n");
4447 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4448 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4449 " but 'external_size' not in effect?\n", __func__
);
4451 "md: invalid array_size %llu > default size %llu\n",
4452 (unsigned long long)mddev
->array_sectors
/ 2,
4453 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4455 mddev
->pers
->stop(mddev
);
4457 if (err
== 0 && mddev
->pers
->sync_request
) {
4458 err
= bitmap_create(mddev
);
4460 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4461 mdname(mddev
), err
);
4462 mddev
->pers
->stop(mddev
);
4466 module_put(mddev
->pers
->owner
);
4468 bitmap_destroy(mddev
);
4471 if (mddev
->pers
->sync_request
) {
4472 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4474 "md: cannot register extra attributes for %s\n",
4476 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4477 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4480 atomic_set(&mddev
->writes_pending
,0);
4481 atomic_set(&mddev
->max_corr_read_errors
,
4482 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4483 mddev
->safemode
= 0;
4484 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4485 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4486 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4489 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4490 if (rdev
->raid_disk
>= 0) {
4492 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4493 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4494 printk("md: cannot register %s for %s\n",
4498 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4501 md_update_sb(mddev
, 0);
4503 md_wakeup_thread(mddev
->thread
);
4504 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4506 md_new_event(mddev
);
4507 sysfs_notify_dirent(mddev
->sysfs_state
);
4508 if (mddev
->sysfs_action
)
4509 sysfs_notify_dirent(mddev
->sysfs_action
);
4510 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4514 static int do_md_run(mddev_t
*mddev
)
4518 err
= md_run(mddev
);
4522 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4523 revalidate_disk(mddev
->gendisk
);
4524 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4529 static int restart_array(mddev_t
*mddev
)
4531 struct gendisk
*disk
= mddev
->gendisk
;
4533 /* Complain if it has no devices */
4534 if (list_empty(&mddev
->disks
))
4540 mddev
->safemode
= 0;
4542 set_disk_ro(disk
, 0);
4543 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4545 /* Kick recovery or resync if necessary */
4546 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4547 md_wakeup_thread(mddev
->thread
);
4548 md_wakeup_thread(mddev
->sync_thread
);
4549 sysfs_notify_dirent(mddev
->sysfs_state
);
4553 /* similar to deny_write_access, but accounts for our holding a reference
4554 * to the file ourselves */
4555 static int deny_bitmap_write_access(struct file
* file
)
4557 struct inode
*inode
= file
->f_mapping
->host
;
4559 spin_lock(&inode
->i_lock
);
4560 if (atomic_read(&inode
->i_writecount
) > 1) {
4561 spin_unlock(&inode
->i_lock
);
4564 atomic_set(&inode
->i_writecount
, -1);
4565 spin_unlock(&inode
->i_lock
);
4570 void restore_bitmap_write_access(struct file
*file
)
4572 struct inode
*inode
= file
->f_mapping
->host
;
4574 spin_lock(&inode
->i_lock
);
4575 atomic_set(&inode
->i_writecount
, 1);
4576 spin_unlock(&inode
->i_lock
);
4579 static void md_clean(mddev_t
*mddev
)
4581 mddev
->array_sectors
= 0;
4582 mddev
->external_size
= 0;
4583 mddev
->dev_sectors
= 0;
4584 mddev
->raid_disks
= 0;
4585 mddev
->recovery_cp
= 0;
4586 mddev
->resync_min
= 0;
4587 mddev
->resync_max
= MaxSector
;
4588 mddev
->reshape_position
= MaxSector
;
4589 mddev
->external
= 0;
4590 mddev
->persistent
= 0;
4591 mddev
->level
= LEVEL_NONE
;
4592 mddev
->clevel
[0] = 0;
4595 mddev
->metadata_type
[0] = 0;
4596 mddev
->chunk_sectors
= 0;
4597 mddev
->ctime
= mddev
->utime
= 0;
4599 mddev
->max_disks
= 0;
4601 mddev
->can_decrease_events
= 0;
4602 mddev
->delta_disks
= 0;
4603 mddev
->new_level
= LEVEL_NONE
;
4604 mddev
->new_layout
= 0;
4605 mddev
->new_chunk_sectors
= 0;
4606 mddev
->curr_resync
= 0;
4607 mddev
->resync_mismatches
= 0;
4608 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4609 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4610 mddev
->recovery
= 0;
4612 mddev
->degraded
= 0;
4613 mddev
->barriers_work
= 0;
4614 mddev
->safemode
= 0;
4615 mddev
->bitmap_info
.offset
= 0;
4616 mddev
->bitmap_info
.default_offset
= 0;
4617 mddev
->bitmap_info
.chunksize
= 0;
4618 mddev
->bitmap_info
.daemon_sleep
= 0;
4619 mddev
->bitmap_info
.max_write_behind
= 0;
4622 static void md_stop_writes(mddev_t
*mddev
)
4624 if (mddev
->sync_thread
) {
4625 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4626 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4627 md_unregister_thread(mddev
->sync_thread
);
4628 mddev
->sync_thread
= NULL
;
4631 del_timer_sync(&mddev
->safemode_timer
);
4633 bitmap_flush(mddev
);
4634 md_super_wait(mddev
);
4636 if (!mddev
->in_sync
|| mddev
->flags
) {
4637 /* mark array as shutdown cleanly */
4639 md_update_sb(mddev
, 1);
4643 static void md_stop(mddev_t
*mddev
)
4645 md_stop_writes(mddev
);
4647 mddev
->pers
->stop(mddev
);
4648 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
4649 mddev
->to_remove
= &md_redundancy_group
;
4650 module_put(mddev
->pers
->owner
);
4652 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4655 static int md_set_readonly(mddev_t
*mddev
, int is_open
)
4658 mutex_lock(&mddev
->open_mutex
);
4659 if (atomic_read(&mddev
->openers
) > is_open
) {
4660 printk("md: %s still in use.\n",mdname(mddev
));
4665 md_stop_writes(mddev
);
4671 set_disk_ro(mddev
->gendisk
, 1);
4672 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4673 sysfs_notify_dirent(mddev
->sysfs_state
);
4677 mutex_unlock(&mddev
->open_mutex
);
4682 * 0 - completely stop and dis-assemble array
4683 * 2 - stop but do not disassemble array
4685 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4688 struct gendisk
*disk
= mddev
->gendisk
;
4691 mutex_lock(&mddev
->open_mutex
);
4692 if (atomic_read(&mddev
->openers
) > is_open
) {
4693 printk("md: %s still in use.\n",mdname(mddev
));
4695 } else if (mddev
->pers
) {
4698 set_disk_ro(disk
, 0);
4701 mddev
->queue
->merge_bvec_fn
= NULL
;
4702 mddev
->queue
->unplug_fn
= NULL
;
4703 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4705 /* tell userspace to handle 'inactive' */
4706 sysfs_notify_dirent(mddev
->sysfs_state
);
4708 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4709 if (rdev
->raid_disk
>= 0) {
4711 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4712 sysfs_remove_link(&mddev
->kobj
, nm
);
4715 set_capacity(disk
, 0);
4716 revalidate_disk(disk
);
4723 mutex_unlock(&mddev
->open_mutex
);
4727 * Free resources if final stop
4731 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4733 bitmap_destroy(mddev
);
4734 if (mddev
->bitmap_info
.file
) {
4735 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4736 fput(mddev
->bitmap_info
.file
);
4737 mddev
->bitmap_info
.file
= NULL
;
4739 mddev
->bitmap_info
.offset
= 0;
4741 export_array(mddev
);
4744 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4745 if (mddev
->hold_active
== UNTIL_STOP
)
4746 mddev
->hold_active
= 0;
4750 blk_integrity_unregister(disk
);
4751 md_new_event(mddev
);
4752 sysfs_notify_dirent(mddev
->sysfs_state
);
4757 static void autorun_array(mddev_t
*mddev
)
4762 if (list_empty(&mddev
->disks
))
4765 printk(KERN_INFO
"md: running: ");
4767 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4768 char b
[BDEVNAME_SIZE
];
4769 printk("<%s>", bdevname(rdev
->bdev
,b
));
4773 err
= do_md_run(mddev
);
4775 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4776 do_md_stop(mddev
, 0, 0);
4781 * lets try to run arrays based on all disks that have arrived
4782 * until now. (those are in pending_raid_disks)
4784 * the method: pick the first pending disk, collect all disks with
4785 * the same UUID, remove all from the pending list and put them into
4786 * the 'same_array' list. Then order this list based on superblock
4787 * update time (freshest comes first), kick out 'old' disks and
4788 * compare superblocks. If everything's fine then run it.
4790 * If "unit" is allocated, then bump its reference count
4792 static void autorun_devices(int part
)
4794 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4796 char b
[BDEVNAME_SIZE
];
4798 printk(KERN_INFO
"md: autorun ...\n");
4799 while (!list_empty(&pending_raid_disks
)) {
4802 LIST_HEAD(candidates
);
4803 rdev0
= list_entry(pending_raid_disks
.next
,
4804 mdk_rdev_t
, same_set
);
4806 printk(KERN_INFO
"md: considering %s ...\n",
4807 bdevname(rdev0
->bdev
,b
));
4808 INIT_LIST_HEAD(&candidates
);
4809 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4810 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4811 printk(KERN_INFO
"md: adding %s ...\n",
4812 bdevname(rdev
->bdev
,b
));
4813 list_move(&rdev
->same_set
, &candidates
);
4816 * now we have a set of devices, with all of them having
4817 * mostly sane superblocks. It's time to allocate the
4821 dev
= MKDEV(mdp_major
,
4822 rdev0
->preferred_minor
<< MdpMinorShift
);
4823 unit
= MINOR(dev
) >> MdpMinorShift
;
4825 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4828 if (rdev0
->preferred_minor
!= unit
) {
4829 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4830 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4834 md_probe(dev
, NULL
, NULL
);
4835 mddev
= mddev_find(dev
);
4836 if (!mddev
|| !mddev
->gendisk
) {
4840 "md: cannot allocate memory for md drive.\n");
4843 if (mddev_lock(mddev
))
4844 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4846 else if (mddev
->raid_disks
|| mddev
->major_version
4847 || !list_empty(&mddev
->disks
)) {
4849 "md: %s already running, cannot run %s\n",
4850 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4851 mddev_unlock(mddev
);
4853 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4854 mddev
->persistent
= 1;
4855 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4856 list_del_init(&rdev
->same_set
);
4857 if (bind_rdev_to_array(rdev
, mddev
))
4860 autorun_array(mddev
);
4861 mddev_unlock(mddev
);
4863 /* on success, candidates will be empty, on error
4866 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4867 list_del_init(&rdev
->same_set
);
4872 printk(KERN_INFO
"md: ... autorun DONE.\n");
4874 #endif /* !MODULE */
4876 static int get_version(void __user
* arg
)
4880 ver
.major
= MD_MAJOR_VERSION
;
4881 ver
.minor
= MD_MINOR_VERSION
;
4882 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4884 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4890 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4892 mdu_array_info_t info
;
4893 int nr
,working
,insync
,failed
,spare
;
4896 nr
=working
=insync
=failed
=spare
=0;
4897 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4899 if (test_bit(Faulty
, &rdev
->flags
))
4903 if (test_bit(In_sync
, &rdev
->flags
))
4910 info
.major_version
= mddev
->major_version
;
4911 info
.minor_version
= mddev
->minor_version
;
4912 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4913 info
.ctime
= mddev
->ctime
;
4914 info
.level
= mddev
->level
;
4915 info
.size
= mddev
->dev_sectors
/ 2;
4916 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4919 info
.raid_disks
= mddev
->raid_disks
;
4920 info
.md_minor
= mddev
->md_minor
;
4921 info
.not_persistent
= !mddev
->persistent
;
4923 info
.utime
= mddev
->utime
;
4926 info
.state
= (1<<MD_SB_CLEAN
);
4927 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
4928 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4929 info
.active_disks
= insync
;
4930 info
.working_disks
= working
;
4931 info
.failed_disks
= failed
;
4932 info
.spare_disks
= spare
;
4934 info
.layout
= mddev
->layout
;
4935 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4937 if (copy_to_user(arg
, &info
, sizeof(info
)))
4943 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4945 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4946 char *ptr
, *buf
= NULL
;
4949 if (md_allow_write(mddev
))
4950 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4952 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4957 /* bitmap disabled, zero the first byte and copy out */
4958 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4959 file
->pathname
[0] = '\0';
4963 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4967 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4971 strcpy(file
->pathname
, ptr
);
4975 if (copy_to_user(arg
, file
, sizeof(*file
)))
4983 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4985 mdu_disk_info_t info
;
4988 if (copy_from_user(&info
, arg
, sizeof(info
)))
4991 rdev
= find_rdev_nr(mddev
, info
.number
);
4993 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4994 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4995 info
.raid_disk
= rdev
->raid_disk
;
4997 if (test_bit(Faulty
, &rdev
->flags
))
4998 info
.state
|= (1<<MD_DISK_FAULTY
);
4999 else if (test_bit(In_sync
, &rdev
->flags
)) {
5000 info
.state
|= (1<<MD_DISK_ACTIVE
);
5001 info
.state
|= (1<<MD_DISK_SYNC
);
5003 if (test_bit(WriteMostly
, &rdev
->flags
))
5004 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5006 info
.major
= info
.minor
= 0;
5007 info
.raid_disk
= -1;
5008 info
.state
= (1<<MD_DISK_REMOVED
);
5011 if (copy_to_user(arg
, &info
, sizeof(info
)))
5017 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
5019 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5021 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5023 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5026 if (!mddev
->raid_disks
) {
5028 /* expecting a device which has a superblock */
5029 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5032 "md: md_import_device returned %ld\n",
5034 return PTR_ERR(rdev
);
5036 if (!list_empty(&mddev
->disks
)) {
5037 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
5038 mdk_rdev_t
, same_set
);
5039 err
= super_types
[mddev
->major_version
]
5040 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5043 "md: %s has different UUID to %s\n",
5044 bdevname(rdev
->bdev
,b
),
5045 bdevname(rdev0
->bdev
,b2
));
5050 err
= bind_rdev_to_array(rdev
, mddev
);
5057 * add_new_disk can be used once the array is assembled
5058 * to add "hot spares". They must already have a superblock
5063 if (!mddev
->pers
->hot_add_disk
) {
5065 "%s: personality does not support diskops!\n",
5069 if (mddev
->persistent
)
5070 rdev
= md_import_device(dev
, mddev
->major_version
,
5071 mddev
->minor_version
);
5073 rdev
= md_import_device(dev
, -1, -1);
5076 "md: md_import_device returned %ld\n",
5078 return PTR_ERR(rdev
);
5080 /* set save_raid_disk if appropriate */
5081 if (!mddev
->persistent
) {
5082 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5083 info
->raid_disk
< mddev
->raid_disks
)
5084 rdev
->raid_disk
= info
->raid_disk
;
5086 rdev
->raid_disk
= -1;
5088 super_types
[mddev
->major_version
].
5089 validate_super(mddev
, rdev
);
5090 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5092 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5093 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5094 set_bit(WriteMostly
, &rdev
->flags
);
5096 clear_bit(WriteMostly
, &rdev
->flags
);
5098 rdev
->raid_disk
= -1;
5099 err
= bind_rdev_to_array(rdev
, mddev
);
5100 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5101 /* If there is hot_add_disk but no hot_remove_disk
5102 * then added disks for geometry changes,
5103 * and should be added immediately.
5105 super_types
[mddev
->major_version
].
5106 validate_super(mddev
, rdev
);
5107 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5109 unbind_rdev_from_array(rdev
);
5114 sysfs_notify_dirent(rdev
->sysfs_state
);
5116 md_update_sb(mddev
, 1);
5117 if (mddev
->degraded
)
5118 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5119 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5120 md_wakeup_thread(mddev
->thread
);
5124 /* otherwise, add_new_disk is only allowed
5125 * for major_version==0 superblocks
5127 if (mddev
->major_version
!= 0) {
5128 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5133 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5135 rdev
= md_import_device(dev
, -1, 0);
5138 "md: error, md_import_device() returned %ld\n",
5140 return PTR_ERR(rdev
);
5142 rdev
->desc_nr
= info
->number
;
5143 if (info
->raid_disk
< mddev
->raid_disks
)
5144 rdev
->raid_disk
= info
->raid_disk
;
5146 rdev
->raid_disk
= -1;
5148 if (rdev
->raid_disk
< mddev
->raid_disks
)
5149 if (info
->state
& (1<<MD_DISK_SYNC
))
5150 set_bit(In_sync
, &rdev
->flags
);
5152 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5153 set_bit(WriteMostly
, &rdev
->flags
);
5155 if (!mddev
->persistent
) {
5156 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5157 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5159 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5160 rdev
->sectors
= rdev
->sb_start
;
5162 err
= bind_rdev_to_array(rdev
, mddev
);
5172 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5174 char b
[BDEVNAME_SIZE
];
5177 rdev
= find_rdev(mddev
, dev
);
5181 if (rdev
->raid_disk
>= 0)
5184 kick_rdev_from_array(rdev
);
5185 md_update_sb(mddev
, 1);
5186 md_new_event(mddev
);
5190 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5191 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5195 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5197 char b
[BDEVNAME_SIZE
];
5204 if (mddev
->major_version
!= 0) {
5205 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5206 " version-0 superblocks.\n",
5210 if (!mddev
->pers
->hot_add_disk
) {
5212 "%s: personality does not support diskops!\n",
5217 rdev
= md_import_device(dev
, -1, 0);
5220 "md: error, md_import_device() returned %ld\n",
5225 if (mddev
->persistent
)
5226 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5228 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5230 rdev
->sectors
= rdev
->sb_start
;
5232 if (test_bit(Faulty
, &rdev
->flags
)) {
5234 "md: can not hot-add faulty %s disk to %s!\n",
5235 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5239 clear_bit(In_sync
, &rdev
->flags
);
5241 rdev
->saved_raid_disk
= -1;
5242 err
= bind_rdev_to_array(rdev
, mddev
);
5247 * The rest should better be atomic, we can have disk failures
5248 * noticed in interrupt contexts ...
5251 rdev
->raid_disk
= -1;
5253 md_update_sb(mddev
, 1);
5256 * Kick recovery, maybe this spare has to be added to the
5257 * array immediately.
5259 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5260 md_wakeup_thread(mddev
->thread
);
5261 md_new_event(mddev
);
5269 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5274 if (!mddev
->pers
->quiesce
)
5276 if (mddev
->recovery
|| mddev
->sync_thread
)
5278 /* we should be able to change the bitmap.. */
5284 return -EEXIST
; /* cannot add when bitmap is present */
5285 mddev
->bitmap_info
.file
= fget(fd
);
5287 if (mddev
->bitmap_info
.file
== NULL
) {
5288 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5293 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5295 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5297 fput(mddev
->bitmap_info
.file
);
5298 mddev
->bitmap_info
.file
= NULL
;
5301 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5302 } else if (mddev
->bitmap
== NULL
)
5303 return -ENOENT
; /* cannot remove what isn't there */
5306 mddev
->pers
->quiesce(mddev
, 1);
5308 err
= bitmap_create(mddev
);
5309 if (fd
< 0 || err
) {
5310 bitmap_destroy(mddev
);
5311 fd
= -1; /* make sure to put the file */
5313 mddev
->pers
->quiesce(mddev
, 0);
5316 if (mddev
->bitmap_info
.file
) {
5317 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5318 fput(mddev
->bitmap_info
.file
);
5320 mddev
->bitmap_info
.file
= NULL
;
5327 * set_array_info is used two different ways
5328 * The original usage is when creating a new array.
5329 * In this usage, raid_disks is > 0 and it together with
5330 * level, size, not_persistent,layout,chunksize determine the
5331 * shape of the array.
5332 * This will always create an array with a type-0.90.0 superblock.
5333 * The newer usage is when assembling an array.
5334 * In this case raid_disks will be 0, and the major_version field is
5335 * use to determine which style super-blocks are to be found on the devices.
5336 * The minor and patch _version numbers are also kept incase the
5337 * super_block handler wishes to interpret them.
5339 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5342 if (info
->raid_disks
== 0) {
5343 /* just setting version number for superblock loading */
5344 if (info
->major_version
< 0 ||
5345 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5346 super_types
[info
->major_version
].name
== NULL
) {
5347 /* maybe try to auto-load a module? */
5349 "md: superblock version %d not known\n",
5350 info
->major_version
);
5353 mddev
->major_version
= info
->major_version
;
5354 mddev
->minor_version
= info
->minor_version
;
5355 mddev
->patch_version
= info
->patch_version
;
5356 mddev
->persistent
= !info
->not_persistent
;
5357 /* ensure mddev_put doesn't delete this now that there
5358 * is some minimal configuration.
5360 mddev
->ctime
= get_seconds();
5363 mddev
->major_version
= MD_MAJOR_VERSION
;
5364 mddev
->minor_version
= MD_MINOR_VERSION
;
5365 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5366 mddev
->ctime
= get_seconds();
5368 mddev
->level
= info
->level
;
5369 mddev
->clevel
[0] = 0;
5370 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5371 mddev
->raid_disks
= info
->raid_disks
;
5372 /* don't set md_minor, it is determined by which /dev/md* was
5375 if (info
->state
& (1<<MD_SB_CLEAN
))
5376 mddev
->recovery_cp
= MaxSector
;
5378 mddev
->recovery_cp
= 0;
5379 mddev
->persistent
= ! info
->not_persistent
;
5380 mddev
->external
= 0;
5382 mddev
->layout
= info
->layout
;
5383 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5385 mddev
->max_disks
= MD_SB_DISKS
;
5387 if (mddev
->persistent
)
5389 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5391 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5392 mddev
->bitmap_info
.offset
= 0;
5394 mddev
->reshape_position
= MaxSector
;
5397 * Generate a 128 bit UUID
5399 get_random_bytes(mddev
->uuid
, 16);
5401 mddev
->new_level
= mddev
->level
;
5402 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5403 mddev
->new_layout
= mddev
->layout
;
5404 mddev
->delta_disks
= 0;
5409 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5411 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5413 if (mddev
->external_size
)
5416 mddev
->array_sectors
= array_sectors
;
5418 EXPORT_SYMBOL(md_set_array_sectors
);
5420 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5424 int fit
= (num_sectors
== 0);
5426 if (mddev
->pers
->resize
== NULL
)
5428 /* The "num_sectors" is the number of sectors of each device that
5429 * is used. This can only make sense for arrays with redundancy.
5430 * linear and raid0 always use whatever space is available. We can only
5431 * consider changing this number if no resync or reconstruction is
5432 * happening, and if the new size is acceptable. It must fit before the
5433 * sb_start or, if that is <data_offset, it must fit before the size
5434 * of each device. If num_sectors is zero, we find the largest size
5438 if (mddev
->sync_thread
)
5441 /* Sorry, cannot grow a bitmap yet, just remove it,
5445 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5446 sector_t avail
= rdev
->sectors
;
5448 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5449 num_sectors
= avail
;
5450 if (avail
< num_sectors
)
5453 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5455 revalidate_disk(mddev
->gendisk
);
5459 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5462 /* change the number of raid disks */
5463 if (mddev
->pers
->check_reshape
== NULL
)
5465 if (raid_disks
<= 0 ||
5466 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5468 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5470 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5472 rv
= mddev
->pers
->check_reshape(mddev
);
5478 * update_array_info is used to change the configuration of an
5480 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5481 * fields in the info are checked against the array.
5482 * Any differences that cannot be handled will cause an error.
5483 * Normally, only one change can be managed at a time.
5485 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5491 /* calculate expected state,ignoring low bits */
5492 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5493 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5495 if (mddev
->major_version
!= info
->major_version
||
5496 mddev
->minor_version
!= info
->minor_version
||
5497 /* mddev->patch_version != info->patch_version || */
5498 mddev
->ctime
!= info
->ctime
||
5499 mddev
->level
!= info
->level
||
5500 /* mddev->layout != info->layout || */
5501 !mddev
->persistent
!= info
->not_persistent
||
5502 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5503 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5504 ((state
^info
->state
) & 0xfffffe00)
5507 /* Check there is only one change */
5508 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5510 if (mddev
->raid_disks
!= info
->raid_disks
)
5512 if (mddev
->layout
!= info
->layout
)
5514 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5521 if (mddev
->layout
!= info
->layout
) {
5523 * we don't need to do anything at the md level, the
5524 * personality will take care of it all.
5526 if (mddev
->pers
->check_reshape
== NULL
)
5529 mddev
->new_layout
= info
->layout
;
5530 rv
= mddev
->pers
->check_reshape(mddev
);
5532 mddev
->new_layout
= mddev
->layout
;
5536 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5537 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5539 if (mddev
->raid_disks
!= info
->raid_disks
)
5540 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5542 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5543 if (mddev
->pers
->quiesce
== NULL
)
5545 if (mddev
->recovery
|| mddev
->sync_thread
)
5547 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5548 /* add the bitmap */
5551 if (mddev
->bitmap_info
.default_offset
== 0)
5553 mddev
->bitmap_info
.offset
=
5554 mddev
->bitmap_info
.default_offset
;
5555 mddev
->pers
->quiesce(mddev
, 1);
5556 rv
= bitmap_create(mddev
);
5558 bitmap_destroy(mddev
);
5559 mddev
->pers
->quiesce(mddev
, 0);
5561 /* remove the bitmap */
5564 if (mddev
->bitmap
->file
)
5566 mddev
->pers
->quiesce(mddev
, 1);
5567 bitmap_destroy(mddev
);
5568 mddev
->pers
->quiesce(mddev
, 0);
5569 mddev
->bitmap_info
.offset
= 0;
5572 md_update_sb(mddev
, 1);
5576 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5580 if (mddev
->pers
== NULL
)
5583 rdev
= find_rdev(mddev
, dev
);
5587 md_error(mddev
, rdev
);
5592 * We have a problem here : there is no easy way to give a CHS
5593 * virtual geometry. We currently pretend that we have a 2 heads
5594 * 4 sectors (with a BIG number of cylinders...). This drives
5595 * dosfs just mad... ;-)
5597 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5599 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5603 geo
->cylinders
= mddev
->array_sectors
/ 8;
5607 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5608 unsigned int cmd
, unsigned long arg
)
5611 void __user
*argp
= (void __user
*)arg
;
5612 mddev_t
*mddev
= NULL
;
5615 if (!capable(CAP_SYS_ADMIN
))
5619 * Commands dealing with the RAID driver but not any
5625 err
= get_version(argp
);
5628 case PRINT_RAID_DEBUG
:
5636 autostart_arrays(arg
);
5643 * Commands creating/starting a new array:
5646 mddev
= bdev
->bd_disk
->private_data
;
5653 err
= mddev_lock(mddev
);
5656 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5663 case SET_ARRAY_INFO
:
5665 mdu_array_info_t info
;
5667 memset(&info
, 0, sizeof(info
));
5668 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5673 err
= update_array_info(mddev
, &info
);
5675 printk(KERN_WARNING
"md: couldn't update"
5676 " array info. %d\n", err
);
5681 if (!list_empty(&mddev
->disks
)) {
5683 "md: array %s already has disks!\n",
5688 if (mddev
->raid_disks
) {
5690 "md: array %s already initialised!\n",
5695 err
= set_array_info(mddev
, &info
);
5697 printk(KERN_WARNING
"md: couldn't set"
5698 " array info. %d\n", err
);
5708 * Commands querying/configuring an existing array:
5710 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5711 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5712 if ((!mddev
->raid_disks
&& !mddev
->external
)
5713 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5714 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5715 && cmd
!= GET_BITMAP_FILE
) {
5721 * Commands even a read-only array can execute:
5725 case GET_ARRAY_INFO
:
5726 err
= get_array_info(mddev
, argp
);
5729 case GET_BITMAP_FILE
:
5730 err
= get_bitmap_file(mddev
, argp
);
5734 err
= get_disk_info(mddev
, argp
);
5737 case RESTART_ARRAY_RW
:
5738 err
= restart_array(mddev
);
5742 err
= do_md_stop(mddev
, 0, 1);
5746 err
= md_set_readonly(mddev
, 1);
5750 if (get_user(ro
, (int __user
*)(arg
))) {
5756 /* if the bdev is going readonly the value of mddev->ro
5757 * does not matter, no writes are coming
5762 /* are we are already prepared for writes? */
5766 /* transitioning to readauto need only happen for
5767 * arrays that call md_write_start
5770 err
= restart_array(mddev
);
5773 set_disk_ro(mddev
->gendisk
, 0);
5780 * The remaining ioctls are changing the state of the
5781 * superblock, so we do not allow them on read-only arrays.
5782 * However non-MD ioctls (e.g. get-size) will still come through
5783 * here and hit the 'default' below, so only disallow
5784 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5786 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5787 if (mddev
->ro
== 2) {
5789 sysfs_notify_dirent(mddev
->sysfs_state
);
5790 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5791 md_wakeup_thread(mddev
->thread
);
5802 mdu_disk_info_t info
;
5803 if (copy_from_user(&info
, argp
, sizeof(info
)))
5806 err
= add_new_disk(mddev
, &info
);
5810 case HOT_REMOVE_DISK
:
5811 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5815 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5818 case SET_DISK_FAULTY
:
5819 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5823 err
= do_md_run(mddev
);
5826 case SET_BITMAP_FILE
:
5827 err
= set_bitmap_file(mddev
, (int)arg
);
5837 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5839 mddev
->hold_active
= 0;
5840 mddev_unlock(mddev
);
5849 #ifdef CONFIG_COMPAT
5850 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
5851 unsigned int cmd
, unsigned long arg
)
5854 case HOT_REMOVE_DISK
:
5856 case SET_DISK_FAULTY
:
5857 case SET_BITMAP_FILE
:
5858 /* These take in integer arg, do not convert */
5861 arg
= (unsigned long)compat_ptr(arg
);
5865 return md_ioctl(bdev
, mode
, cmd
, arg
);
5867 #endif /* CONFIG_COMPAT */
5869 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5872 * Succeed if we can lock the mddev, which confirms that
5873 * it isn't being stopped right now.
5875 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5878 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5879 /* we are racing with mddev_put which is discarding this
5883 /* Wait until bdev->bd_disk is definitely gone */
5884 flush_scheduled_work();
5885 /* Then retry the open from the top */
5886 return -ERESTARTSYS
;
5888 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5890 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5894 atomic_inc(&mddev
->openers
);
5895 mutex_unlock(&mddev
->open_mutex
);
5901 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5903 mddev_t
*mddev
= disk
->private_data
;
5906 atomic_dec(&mddev
->openers
);
5911 static const struct block_device_operations md_fops
=
5913 .owner
= THIS_MODULE
,
5915 .release
= md_release
,
5917 #ifdef CONFIG_COMPAT
5918 .compat_ioctl
= md_compat_ioctl
,
5920 .getgeo
= md_getgeo
,
5923 static int md_thread(void * arg
)
5925 mdk_thread_t
*thread
= arg
;
5928 * md_thread is a 'system-thread', it's priority should be very
5929 * high. We avoid resource deadlocks individually in each
5930 * raid personality. (RAID5 does preallocation) We also use RR and
5931 * the very same RT priority as kswapd, thus we will never get
5932 * into a priority inversion deadlock.
5934 * we definitely have to have equal or higher priority than
5935 * bdflush, otherwise bdflush will deadlock if there are too
5936 * many dirty RAID5 blocks.
5939 allow_signal(SIGKILL
);
5940 while (!kthread_should_stop()) {
5942 /* We need to wait INTERRUPTIBLE so that
5943 * we don't add to the load-average.
5944 * That means we need to be sure no signals are
5947 if (signal_pending(current
))
5948 flush_signals(current
);
5950 wait_event_interruptible_timeout
5952 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5953 || kthread_should_stop(),
5956 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5958 thread
->run(thread
->mddev
);
5964 void md_wakeup_thread(mdk_thread_t
*thread
)
5967 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5968 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5969 wake_up(&thread
->wqueue
);
5973 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5976 mdk_thread_t
*thread
;
5978 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5982 init_waitqueue_head(&thread
->wqueue
);
5985 thread
->mddev
= mddev
;
5986 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5987 thread
->tsk
= kthread_run(md_thread
, thread
,
5989 mdname(thread
->mddev
),
5990 name
?: mddev
->pers
->name
);
5991 if (IS_ERR(thread
->tsk
)) {
5998 void md_unregister_thread(mdk_thread_t
*thread
)
6002 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6004 kthread_stop(thread
->tsk
);
6008 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
6015 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6018 if (mddev
->external
)
6019 set_bit(Blocked
, &rdev
->flags
);
6021 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6023 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6024 __builtin_return_address(0),__builtin_return_address(1),
6025 __builtin_return_address(2),__builtin_return_address(3));
6029 if (!mddev
->pers
->error_handler
)
6031 mddev
->pers
->error_handler(mddev
,rdev
);
6032 if (mddev
->degraded
)
6033 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6034 sysfs_notify_dirent(rdev
->sysfs_state
);
6035 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6036 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6037 md_wakeup_thread(mddev
->thread
);
6038 md_new_event_inintr(mddev
);
6041 /* seq_file implementation /proc/mdstat */
6043 static void status_unused(struct seq_file
*seq
)
6048 seq_printf(seq
, "unused devices: ");
6050 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6051 char b
[BDEVNAME_SIZE
];
6053 seq_printf(seq
, "%s ",
6054 bdevname(rdev
->bdev
,b
));
6057 seq_printf(seq
, "<none>");
6059 seq_printf(seq
, "\n");
6063 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6065 sector_t max_sectors
, resync
, res
;
6066 unsigned long dt
, db
;
6069 unsigned int per_milli
;
6071 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6073 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6074 max_sectors
= mddev
->resync_max_sectors
;
6076 max_sectors
= mddev
->dev_sectors
;
6079 * Should not happen.
6085 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6086 * in a sector_t, and (max_sectors>>scale) will fit in a
6087 * u32, as those are the requirements for sector_div.
6088 * Thus 'scale' must be at least 10
6091 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6092 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6095 res
= (resync
>>scale
)*1000;
6096 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6100 int i
, x
= per_milli
/50, y
= 20-x
;
6101 seq_printf(seq
, "[");
6102 for (i
= 0; i
< x
; i
++)
6103 seq_printf(seq
, "=");
6104 seq_printf(seq
, ">");
6105 for (i
= 0; i
< y
; i
++)
6106 seq_printf(seq
, ".");
6107 seq_printf(seq
, "] ");
6109 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6110 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6112 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6114 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6115 "resync" : "recovery"))),
6116 per_milli
/10, per_milli
% 10,
6117 (unsigned long long) resync
/2,
6118 (unsigned long long) max_sectors
/2);
6121 * dt: time from mark until now
6122 * db: blocks written from mark until now
6123 * rt: remaining time
6125 * rt is a sector_t, so could be 32bit or 64bit.
6126 * So we divide before multiply in case it is 32bit and close
6128 * We scale the divisor (db) by 32 to avoid loosing precision
6129 * near the end of resync when the number of remaining sectors
6131 * We then divide rt by 32 after multiplying by db to compensate.
6132 * The '+1' avoids division by zero if db is very small.
6134 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6136 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6137 - mddev
->resync_mark_cnt
;
6139 rt
= max_sectors
- resync
; /* number of remaining sectors */
6140 sector_div(rt
, db
/32+1);
6144 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6145 ((unsigned long)rt
% 60)/6);
6147 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6150 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6152 struct list_head
*tmp
;
6162 spin_lock(&all_mddevs_lock
);
6163 list_for_each(tmp
,&all_mddevs
)
6165 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6167 spin_unlock(&all_mddevs_lock
);
6170 spin_unlock(&all_mddevs_lock
);
6172 return (void*)2;/* tail */
6176 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6178 struct list_head
*tmp
;
6179 mddev_t
*next_mddev
, *mddev
= v
;
6185 spin_lock(&all_mddevs_lock
);
6187 tmp
= all_mddevs
.next
;
6189 tmp
= mddev
->all_mddevs
.next
;
6190 if (tmp
!= &all_mddevs
)
6191 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6193 next_mddev
= (void*)2;
6196 spin_unlock(&all_mddevs_lock
);
6204 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6208 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6212 struct mdstat_info
{
6216 static int md_seq_show(struct seq_file
*seq
, void *v
)
6221 struct mdstat_info
*mi
= seq
->private;
6222 struct bitmap
*bitmap
;
6224 if (v
== (void*)1) {
6225 struct mdk_personality
*pers
;
6226 seq_printf(seq
, "Personalities : ");
6227 spin_lock(&pers_lock
);
6228 list_for_each_entry(pers
, &pers_list
, list
)
6229 seq_printf(seq
, "[%s] ", pers
->name
);
6231 spin_unlock(&pers_lock
);
6232 seq_printf(seq
, "\n");
6233 mi
->event
= atomic_read(&md_event_count
);
6236 if (v
== (void*)2) {
6241 if (mddev_lock(mddev
) < 0)
6244 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6245 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6246 mddev
->pers
? "" : "in");
6249 seq_printf(seq
, " (read-only)");
6251 seq_printf(seq
, " (auto-read-only)");
6252 seq_printf(seq
, " %s", mddev
->pers
->name
);
6256 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6257 char b
[BDEVNAME_SIZE
];
6258 seq_printf(seq
, " %s[%d]",
6259 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6260 if (test_bit(WriteMostly
, &rdev
->flags
))
6261 seq_printf(seq
, "(W)");
6262 if (test_bit(Faulty
, &rdev
->flags
)) {
6263 seq_printf(seq
, "(F)");
6265 } else if (rdev
->raid_disk
< 0)
6266 seq_printf(seq
, "(S)"); /* spare */
6267 sectors
+= rdev
->sectors
;
6270 if (!list_empty(&mddev
->disks
)) {
6272 seq_printf(seq
, "\n %llu blocks",
6273 (unsigned long long)
6274 mddev
->array_sectors
/ 2);
6276 seq_printf(seq
, "\n %llu blocks",
6277 (unsigned long long)sectors
/ 2);
6279 if (mddev
->persistent
) {
6280 if (mddev
->major_version
!= 0 ||
6281 mddev
->minor_version
!= 90) {
6282 seq_printf(seq
," super %d.%d",
6283 mddev
->major_version
,
6284 mddev
->minor_version
);
6286 } else if (mddev
->external
)
6287 seq_printf(seq
, " super external:%s",
6288 mddev
->metadata_type
);
6290 seq_printf(seq
, " super non-persistent");
6293 mddev
->pers
->status(seq
, mddev
);
6294 seq_printf(seq
, "\n ");
6295 if (mddev
->pers
->sync_request
) {
6296 if (mddev
->curr_resync
> 2) {
6297 status_resync(seq
, mddev
);
6298 seq_printf(seq
, "\n ");
6299 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6300 seq_printf(seq
, "\tresync=DELAYED\n ");
6301 else if (mddev
->recovery_cp
< MaxSector
)
6302 seq_printf(seq
, "\tresync=PENDING\n ");
6305 seq_printf(seq
, "\n ");
6307 if ((bitmap
= mddev
->bitmap
)) {
6308 unsigned long chunk_kb
;
6309 unsigned long flags
;
6310 spin_lock_irqsave(&bitmap
->lock
, flags
);
6311 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6312 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6314 bitmap
->pages
- bitmap
->missing_pages
,
6316 (bitmap
->pages
- bitmap
->missing_pages
)
6317 << (PAGE_SHIFT
- 10),
6318 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6319 chunk_kb
? "KB" : "B");
6321 seq_printf(seq
, ", file: ");
6322 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6325 seq_printf(seq
, "\n");
6326 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6329 seq_printf(seq
, "\n");
6331 mddev_unlock(mddev
);
6336 static const struct seq_operations md_seq_ops
= {
6337 .start
= md_seq_start
,
6338 .next
= md_seq_next
,
6339 .stop
= md_seq_stop
,
6340 .show
= md_seq_show
,
6343 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6346 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6350 error
= seq_open(file
, &md_seq_ops
);
6354 struct seq_file
*p
= file
->private_data
;
6356 mi
->event
= atomic_read(&md_event_count
);
6361 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6363 struct seq_file
*m
= filp
->private_data
;
6364 struct mdstat_info
*mi
= m
->private;
6367 poll_wait(filp
, &md_event_waiters
, wait
);
6369 /* always allow read */
6370 mask
= POLLIN
| POLLRDNORM
;
6372 if (mi
->event
!= atomic_read(&md_event_count
))
6373 mask
|= POLLERR
| POLLPRI
;
6377 static const struct file_operations md_seq_fops
= {
6378 .owner
= THIS_MODULE
,
6379 .open
= md_seq_open
,
6381 .llseek
= seq_lseek
,
6382 .release
= seq_release_private
,
6383 .poll
= mdstat_poll
,
6386 int register_md_personality(struct mdk_personality
*p
)
6388 spin_lock(&pers_lock
);
6389 list_add_tail(&p
->list
, &pers_list
);
6390 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6391 spin_unlock(&pers_lock
);
6395 int unregister_md_personality(struct mdk_personality
*p
)
6397 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6398 spin_lock(&pers_lock
);
6399 list_del_init(&p
->list
);
6400 spin_unlock(&pers_lock
);
6404 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6412 rdev_for_each_rcu(rdev
, mddev
) {
6413 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6414 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6415 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6416 atomic_read(&disk
->sync_io
);
6417 /* sync IO will cause sync_io to increase before the disk_stats
6418 * as sync_io is counted when a request starts, and
6419 * disk_stats is counted when it completes.
6420 * So resync activity will cause curr_events to be smaller than
6421 * when there was no such activity.
6422 * non-sync IO will cause disk_stat to increase without
6423 * increasing sync_io so curr_events will (eventually)
6424 * be larger than it was before. Once it becomes
6425 * substantially larger, the test below will cause
6426 * the array to appear non-idle, and resync will slow
6428 * If there is a lot of outstanding resync activity when
6429 * we set last_event to curr_events, then all that activity
6430 * completing might cause the array to appear non-idle
6431 * and resync will be slowed down even though there might
6432 * not have been non-resync activity. This will only
6433 * happen once though. 'last_events' will soon reflect
6434 * the state where there is little or no outstanding
6435 * resync requests, and further resync activity will
6436 * always make curr_events less than last_events.
6439 if (init
|| curr_events
- rdev
->last_events
> 64) {
6440 rdev
->last_events
= curr_events
;
6448 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6450 /* another "blocks" (512byte) blocks have been synced */
6451 atomic_sub(blocks
, &mddev
->recovery_active
);
6452 wake_up(&mddev
->recovery_wait
);
6454 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6455 md_wakeup_thread(mddev
->thread
);
6456 // stop recovery, signal do_sync ....
6461 /* md_write_start(mddev, bi)
6462 * If we need to update some array metadata (e.g. 'active' flag
6463 * in superblock) before writing, schedule a superblock update
6464 * and wait for it to complete.
6466 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6469 if (bio_data_dir(bi
) != WRITE
)
6472 BUG_ON(mddev
->ro
== 1);
6473 if (mddev
->ro
== 2) {
6474 /* need to switch to read/write */
6476 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6477 md_wakeup_thread(mddev
->thread
);
6478 md_wakeup_thread(mddev
->sync_thread
);
6481 atomic_inc(&mddev
->writes_pending
);
6482 if (mddev
->safemode
== 1)
6483 mddev
->safemode
= 0;
6484 if (mddev
->in_sync
) {
6485 spin_lock_irq(&mddev
->write_lock
);
6486 if (mddev
->in_sync
) {
6488 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6489 md_wakeup_thread(mddev
->thread
);
6492 spin_unlock_irq(&mddev
->write_lock
);
6495 sysfs_notify_dirent(mddev
->sysfs_state
);
6496 wait_event(mddev
->sb_wait
,
6497 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6498 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6501 void md_write_end(mddev_t
*mddev
)
6503 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6504 if (mddev
->safemode
== 2)
6505 md_wakeup_thread(mddev
->thread
);
6506 else if (mddev
->safemode_delay
)
6507 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6511 /* md_allow_write(mddev)
6512 * Calling this ensures that the array is marked 'active' so that writes
6513 * may proceed without blocking. It is important to call this before
6514 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6515 * Must be called with mddev_lock held.
6517 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6518 * is dropped, so return -EAGAIN after notifying userspace.
6520 int md_allow_write(mddev_t
*mddev
)
6526 if (!mddev
->pers
->sync_request
)
6529 spin_lock_irq(&mddev
->write_lock
);
6530 if (mddev
->in_sync
) {
6532 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6533 if (mddev
->safemode_delay
&&
6534 mddev
->safemode
== 0)
6535 mddev
->safemode
= 1;
6536 spin_unlock_irq(&mddev
->write_lock
);
6537 md_update_sb(mddev
, 0);
6538 sysfs_notify_dirent(mddev
->sysfs_state
);
6540 spin_unlock_irq(&mddev
->write_lock
);
6542 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6547 EXPORT_SYMBOL_GPL(md_allow_write
);
6549 #define SYNC_MARKS 10
6550 #define SYNC_MARK_STEP (3*HZ)
6551 void md_do_sync(mddev_t
*mddev
)
6554 unsigned int currspeed
= 0,
6556 sector_t max_sectors
,j
, io_sectors
;
6557 unsigned long mark
[SYNC_MARKS
];
6558 sector_t mark_cnt
[SYNC_MARKS
];
6560 struct list_head
*tmp
;
6561 sector_t last_check
;
6566 /* just incase thread restarts... */
6567 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6569 if (mddev
->ro
) /* never try to sync a read-only array */
6572 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6573 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6574 desc
= "data-check";
6575 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6576 desc
= "requested-resync";
6579 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6584 /* we overload curr_resync somewhat here.
6585 * 0 == not engaged in resync at all
6586 * 2 == checking that there is no conflict with another sync
6587 * 1 == like 2, but have yielded to allow conflicting resync to
6589 * other == active in resync - this many blocks
6591 * Before starting a resync we must have set curr_resync to
6592 * 2, and then checked that every "conflicting" array has curr_resync
6593 * less than ours. When we find one that is the same or higher
6594 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6595 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6596 * This will mean we have to start checking from the beginning again.
6601 mddev
->curr_resync
= 2;
6604 if (kthread_should_stop())
6605 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6607 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6609 for_each_mddev(mddev2
, tmp
) {
6610 if (mddev2
== mddev
)
6612 if (!mddev
->parallel_resync
6613 && mddev2
->curr_resync
6614 && match_mddev_units(mddev
, mddev2
)) {
6616 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6617 /* arbitrarily yield */
6618 mddev
->curr_resync
= 1;
6619 wake_up(&resync_wait
);
6621 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6622 /* no need to wait here, we can wait the next
6623 * time 'round when curr_resync == 2
6626 /* We need to wait 'interruptible' so as not to
6627 * contribute to the load average, and not to
6628 * be caught by 'softlockup'
6630 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6631 if (!kthread_should_stop() &&
6632 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6633 printk(KERN_INFO
"md: delaying %s of %s"
6634 " until %s has finished (they"
6635 " share one or more physical units)\n",
6636 desc
, mdname(mddev
), mdname(mddev2
));
6638 if (signal_pending(current
))
6639 flush_signals(current
);
6641 finish_wait(&resync_wait
, &wq
);
6644 finish_wait(&resync_wait
, &wq
);
6647 } while (mddev
->curr_resync
< 2);
6650 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6651 /* resync follows the size requested by the personality,
6652 * which defaults to physical size, but can be virtual size
6654 max_sectors
= mddev
->resync_max_sectors
;
6655 mddev
->resync_mismatches
= 0;
6656 /* we don't use the checkpoint if there's a bitmap */
6657 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6658 j
= mddev
->resync_min
;
6659 else if (!mddev
->bitmap
)
6660 j
= mddev
->recovery_cp
;
6662 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6663 max_sectors
= mddev
->dev_sectors
;
6665 /* recovery follows the physical size of devices */
6666 max_sectors
= mddev
->dev_sectors
;
6669 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6670 if (rdev
->raid_disk
>= 0 &&
6671 !test_bit(Faulty
, &rdev
->flags
) &&
6672 !test_bit(In_sync
, &rdev
->flags
) &&
6673 rdev
->recovery_offset
< j
)
6674 j
= rdev
->recovery_offset
;
6678 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6679 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6680 " %d KB/sec/disk.\n", speed_min(mddev
));
6681 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6682 "(but not more than %d KB/sec) for %s.\n",
6683 speed_max(mddev
), desc
);
6685 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6688 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6690 mark_cnt
[m
] = io_sectors
;
6693 mddev
->resync_mark
= mark
[last_mark
];
6694 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6697 * Tune reconstruction:
6699 window
= 32*(PAGE_SIZE
/512);
6700 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6701 window
/2,(unsigned long long) max_sectors
/2);
6703 atomic_set(&mddev
->recovery_active
, 0);
6708 "md: resuming %s of %s from checkpoint.\n",
6709 desc
, mdname(mddev
));
6710 mddev
->curr_resync
= j
;
6712 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6714 while (j
< max_sectors
) {
6719 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6720 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6721 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6722 > (max_sectors
>> 4)) ||
6723 (j
- mddev
->curr_resync_completed
)*2
6724 >= mddev
->resync_max
- mddev
->curr_resync_completed
6726 /* time to update curr_resync_completed */
6727 blk_unplug(mddev
->queue
);
6728 wait_event(mddev
->recovery_wait
,
6729 atomic_read(&mddev
->recovery_active
) == 0);
6730 mddev
->curr_resync_completed
=
6732 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6733 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6736 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6737 /* As this condition is controlled by user-space,
6738 * we can block indefinitely, so use '_interruptible'
6739 * to avoid triggering warnings.
6741 flush_signals(current
); /* just in case */
6742 wait_event_interruptible(mddev
->recovery_wait
,
6743 mddev
->resync_max
> j
6744 || kthread_should_stop());
6747 if (kthread_should_stop())
6750 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6751 currspeed
< speed_min(mddev
));
6753 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6757 if (!skipped
) { /* actual IO requested */
6758 io_sectors
+= sectors
;
6759 atomic_add(sectors
, &mddev
->recovery_active
);
6763 if (j
>1) mddev
->curr_resync
= j
;
6764 mddev
->curr_mark_cnt
= io_sectors
;
6765 if (last_check
== 0)
6766 /* this is the earliers that rebuilt will be
6767 * visible in /proc/mdstat
6769 md_new_event(mddev
);
6771 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6774 last_check
= io_sectors
;
6776 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6780 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6782 int next
= (last_mark
+1) % SYNC_MARKS
;
6784 mddev
->resync_mark
= mark
[next
];
6785 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6786 mark
[next
] = jiffies
;
6787 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6792 if (kthread_should_stop())
6797 * this loop exits only if either when we are slower than
6798 * the 'hard' speed limit, or the system was IO-idle for
6800 * the system might be non-idle CPU-wise, but we only care
6801 * about not overloading the IO subsystem. (things like an
6802 * e2fsck being done on the RAID array should execute fast)
6804 blk_unplug(mddev
->queue
);
6807 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6808 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6810 if (currspeed
> speed_min(mddev
)) {
6811 if ((currspeed
> speed_max(mddev
)) ||
6812 !is_mddev_idle(mddev
, 0)) {
6818 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6820 * this also signals 'finished resyncing' to md_stop
6823 blk_unplug(mddev
->queue
);
6825 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6827 /* tell personality that we are finished */
6828 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6830 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6831 mddev
->curr_resync
> 2) {
6832 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6833 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6834 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6836 "md: checkpointing %s of %s.\n",
6837 desc
, mdname(mddev
));
6838 mddev
->recovery_cp
= mddev
->curr_resync
;
6841 mddev
->recovery_cp
= MaxSector
;
6843 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6844 mddev
->curr_resync
= MaxSector
;
6846 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6847 if (rdev
->raid_disk
>= 0 &&
6848 !test_bit(Faulty
, &rdev
->flags
) &&
6849 !test_bit(In_sync
, &rdev
->flags
) &&
6850 rdev
->recovery_offset
< mddev
->curr_resync
)
6851 rdev
->recovery_offset
= mddev
->curr_resync
;
6855 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6858 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6859 /* We completed so min/max setting can be forgotten if used. */
6860 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6861 mddev
->resync_min
= 0;
6862 mddev
->resync_max
= MaxSector
;
6863 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6864 mddev
->resync_min
= mddev
->curr_resync_completed
;
6865 mddev
->curr_resync
= 0;
6866 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6867 mddev
->curr_resync_completed
= 0;
6868 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6869 wake_up(&resync_wait
);
6870 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6871 md_wakeup_thread(mddev
->thread
);
6876 * got a signal, exit.
6879 "md: md_do_sync() got signal ... exiting\n");
6880 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6884 EXPORT_SYMBOL_GPL(md_do_sync
);
6887 static int remove_and_add_spares(mddev_t
*mddev
)
6892 mddev
->curr_resync_completed
= 0;
6894 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6895 if (rdev
->raid_disk
>= 0 &&
6896 !test_bit(Blocked
, &rdev
->flags
) &&
6897 (test_bit(Faulty
, &rdev
->flags
) ||
6898 ! test_bit(In_sync
, &rdev
->flags
)) &&
6899 atomic_read(&rdev
->nr_pending
)==0) {
6900 if (mddev
->pers
->hot_remove_disk(
6901 mddev
, rdev
->raid_disk
)==0) {
6903 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6904 sysfs_remove_link(&mddev
->kobj
, nm
);
6905 rdev
->raid_disk
= -1;
6909 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6910 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6911 if (rdev
->raid_disk
>= 0 &&
6912 !test_bit(In_sync
, &rdev
->flags
) &&
6913 !test_bit(Blocked
, &rdev
->flags
))
6915 if (rdev
->raid_disk
< 0
6916 && !test_bit(Faulty
, &rdev
->flags
)) {
6917 rdev
->recovery_offset
= 0;
6919 hot_add_disk(mddev
, rdev
) == 0) {
6921 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6922 if (sysfs_create_link(&mddev
->kobj
,
6925 "md: cannot register "
6929 md_new_event(mddev
);
6930 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6939 * This routine is regularly called by all per-raid-array threads to
6940 * deal with generic issues like resync and super-block update.
6941 * Raid personalities that don't have a thread (linear/raid0) do not
6942 * need this as they never do any recovery or update the superblock.
6944 * It does not do any resync itself, but rather "forks" off other threads
6945 * to do that as needed.
6946 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6947 * "->recovery" and create a thread at ->sync_thread.
6948 * When the thread finishes it sets MD_RECOVERY_DONE
6949 * and wakeups up this thread which will reap the thread and finish up.
6950 * This thread also removes any faulty devices (with nr_pending == 0).
6952 * The overall approach is:
6953 * 1/ if the superblock needs updating, update it.
6954 * 2/ If a recovery thread is running, don't do anything else.
6955 * 3/ If recovery has finished, clean up, possibly marking spares active.
6956 * 4/ If there are any faulty devices, remove them.
6957 * 5/ If array is degraded, try to add spares devices
6958 * 6/ If array has spares or is not in-sync, start a resync thread.
6960 void md_check_recovery(mddev_t
*mddev
)
6966 bitmap_daemon_work(mddev
);
6971 if (signal_pending(current
)) {
6972 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6973 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6975 mddev
->safemode
= 2;
6977 flush_signals(current
);
6980 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6983 (mddev
->flags
&& !mddev
->external
) ||
6984 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6985 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6986 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6987 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6988 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6992 if (mddev_trylock(mddev
)) {
6996 /* Only thing we do on a ro array is remove
6999 remove_and_add_spares(mddev
);
7000 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7004 if (!mddev
->external
) {
7006 spin_lock_irq(&mddev
->write_lock
);
7007 if (mddev
->safemode
&&
7008 !atomic_read(&mddev
->writes_pending
) &&
7010 mddev
->recovery_cp
== MaxSector
) {
7013 if (mddev
->persistent
)
7014 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7016 if (mddev
->safemode
== 1)
7017 mddev
->safemode
= 0;
7018 spin_unlock_irq(&mddev
->write_lock
);
7020 sysfs_notify_dirent(mddev
->sysfs_state
);
7024 md_update_sb(mddev
, 0);
7026 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7027 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7028 /* resync/recovery still happening */
7029 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7032 if (mddev
->sync_thread
) {
7033 /* resync has finished, collect result */
7034 md_unregister_thread(mddev
->sync_thread
);
7035 mddev
->sync_thread
= NULL
;
7036 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7037 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7039 /* activate any spares */
7040 if (mddev
->pers
->spare_active(mddev
))
7041 sysfs_notify(&mddev
->kobj
, NULL
,
7044 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7045 mddev
->pers
->finish_reshape
)
7046 mddev
->pers
->finish_reshape(mddev
);
7047 md_update_sb(mddev
, 1);
7049 /* if array is no-longer degraded, then any saved_raid_disk
7050 * information must be scrapped
7052 if (!mddev
->degraded
)
7053 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7054 rdev
->saved_raid_disk
= -1;
7056 mddev
->recovery
= 0;
7057 /* flag recovery needed just to double check */
7058 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7059 sysfs_notify_dirent(mddev
->sysfs_action
);
7060 md_new_event(mddev
);
7063 /* Set RUNNING before clearing NEEDED to avoid
7064 * any transients in the value of "sync_action".
7066 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7067 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7068 /* Clear some bits that don't mean anything, but
7071 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7072 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7074 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7076 /* no recovery is running.
7077 * remove any failed drives, then
7078 * add spares if possible.
7079 * Spare are also removed and re-added, to allow
7080 * the personality to fail the re-add.
7083 if (mddev
->reshape_position
!= MaxSector
) {
7084 if (mddev
->pers
->check_reshape
== NULL
||
7085 mddev
->pers
->check_reshape(mddev
) != 0)
7086 /* Cannot proceed */
7088 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7089 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7090 } else if ((spares
= remove_and_add_spares(mddev
))) {
7091 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7092 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7093 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7094 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7095 } else if (mddev
->recovery_cp
< MaxSector
) {
7096 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7097 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7098 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7099 /* nothing to be done ... */
7102 if (mddev
->pers
->sync_request
) {
7103 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7104 /* We are adding a device or devices to an array
7105 * which has the bitmap stored on all devices.
7106 * So make sure all bitmap pages get written
7108 bitmap_write_all(mddev
->bitmap
);
7110 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7113 if (!mddev
->sync_thread
) {
7114 printk(KERN_ERR
"%s: could not start resync"
7117 /* leave the spares where they are, it shouldn't hurt */
7118 mddev
->recovery
= 0;
7120 md_wakeup_thread(mddev
->sync_thread
);
7121 sysfs_notify_dirent(mddev
->sysfs_action
);
7122 md_new_event(mddev
);
7125 if (!mddev
->sync_thread
) {
7126 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7127 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7129 if (mddev
->sysfs_action
)
7130 sysfs_notify_dirent(mddev
->sysfs_action
);
7132 mddev_unlock(mddev
);
7136 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7138 sysfs_notify_dirent(rdev
->sysfs_state
);
7139 wait_event_timeout(rdev
->blocked_wait
,
7140 !test_bit(Blocked
, &rdev
->flags
),
7141 msecs_to_jiffies(5000));
7142 rdev_dec_pending(rdev
, mddev
);
7144 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7146 static int md_notify_reboot(struct notifier_block
*this,
7147 unsigned long code
, void *x
)
7149 struct list_head
*tmp
;
7152 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7154 printk(KERN_INFO
"md: stopping all md devices.\n");
7156 for_each_mddev(mddev
, tmp
)
7157 if (mddev_trylock(mddev
)) {
7158 /* Force a switch to readonly even array
7159 * appears to still be in use. Hence
7162 md_set_readonly(mddev
, 100);
7163 mddev_unlock(mddev
);
7166 * certain more exotic SCSI devices are known to be
7167 * volatile wrt too early system reboots. While the
7168 * right place to handle this issue is the given
7169 * driver, we do want to have a safe RAID driver ...
7176 static struct notifier_block md_notifier
= {
7177 .notifier_call
= md_notify_reboot
,
7179 .priority
= INT_MAX
, /* before any real devices */
7182 static void md_geninit(void)
7184 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
7186 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
7189 static int __init
md_init(void)
7191 if (register_blkdev(MD_MAJOR
, "md"))
7193 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
7194 unregister_blkdev(MD_MAJOR
, "md");
7197 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7198 md_probe
, NULL
, NULL
);
7199 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7200 md_probe
, NULL
, NULL
);
7202 register_reboot_notifier(&md_notifier
);
7203 raid_table_header
= register_sysctl_table(raid_root_table
);
7213 * Searches all registered partitions for autorun RAID arrays
7217 static LIST_HEAD(all_detected_devices
);
7218 struct detected_devices_node
{
7219 struct list_head list
;
7223 void md_autodetect_dev(dev_t dev
)
7225 struct detected_devices_node
*node_detected_dev
;
7227 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7228 if (node_detected_dev
) {
7229 node_detected_dev
->dev
= dev
;
7230 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7232 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7233 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7238 static void autostart_arrays(int part
)
7241 struct detected_devices_node
*node_detected_dev
;
7243 int i_scanned
, i_passed
;
7248 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7250 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7252 node_detected_dev
= list_entry(all_detected_devices
.next
,
7253 struct detected_devices_node
, list
);
7254 list_del(&node_detected_dev
->list
);
7255 dev
= node_detected_dev
->dev
;
7256 kfree(node_detected_dev
);
7257 rdev
= md_import_device(dev
,0, 90);
7261 if (test_bit(Faulty
, &rdev
->flags
)) {
7265 set_bit(AutoDetected
, &rdev
->flags
);
7266 list_add(&rdev
->same_set
, &pending_raid_disks
);
7270 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7271 i_scanned
, i_passed
);
7273 autorun_devices(part
);
7276 #endif /* !MODULE */
7278 static __exit
void md_exit(void)
7281 struct list_head
*tmp
;
7283 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7284 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7286 unregister_blkdev(MD_MAJOR
,"md");
7287 unregister_blkdev(mdp_major
, "mdp");
7288 unregister_reboot_notifier(&md_notifier
);
7289 unregister_sysctl_table(raid_table_header
);
7290 remove_proc_entry("mdstat", NULL
);
7291 for_each_mddev(mddev
, tmp
) {
7292 export_array(mddev
);
7293 mddev
->hold_active
= 0;
7297 subsys_initcall(md_init
);
7298 module_exit(md_exit
)
7300 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7302 return sprintf(buffer
, "%d", start_readonly
);
7304 static int set_ro(const char *val
, struct kernel_param
*kp
)
7307 int num
= simple_strtoul(val
, &e
, 10);
7308 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7309 start_readonly
= num
;
7315 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7316 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7318 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7320 EXPORT_SYMBOL(register_md_personality
);
7321 EXPORT_SYMBOL(unregister_md_personality
);
7322 EXPORT_SYMBOL(md_error
);
7323 EXPORT_SYMBOL(md_done_sync
);
7324 EXPORT_SYMBOL(md_write_start
);
7325 EXPORT_SYMBOL(md_write_end
);
7326 EXPORT_SYMBOL(md_register_thread
);
7327 EXPORT_SYMBOL(md_unregister_thread
);
7328 EXPORT_SYMBOL(md_wakeup_thread
);
7329 EXPORT_SYMBOL(md_check_recovery
);
7330 MODULE_LICENSE("GPL");
7331 MODULE_DESCRIPTION("MD RAID framework");
7333 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);