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/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part
);
72 static LIST_HEAD(pers_list
);
73 static DEFINE_SPINLOCK(pers_lock
);
75 static void md_print_devices(void);
77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min
= 1000;
95 static int sysctl_speed_limit_max
= 200000;
96 static inline int speed_min(mddev_t
*mddev
)
98 return mddev
->sync_speed_min
?
99 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
102 static inline int speed_max(mddev_t
*mddev
)
104 return mddev
->sync_speed_max
?
105 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
108 static struct ctl_table_header
*raid_table_header
;
110 static ctl_table raid_table
[] = {
112 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
113 .procname
= "speed_limit_min",
114 .data
= &sysctl_speed_limit_min
,
115 .maxlen
= sizeof(int),
116 .mode
= S_IRUGO
|S_IWUSR
,
117 .proc_handler
= &proc_dointvec
,
120 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= &proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
132 .ctl_name
= DEV_RAID
,
135 .mode
= S_IRUGO
|S_IXUGO
,
141 static ctl_table raid_root_table
[] = {
147 .child
= raid_dir_table
,
152 static struct block_device_operations md_fops
;
154 static int start_readonly
;
157 * We have a system wide 'event count' that is incremented
158 * on any 'interesting' event, and readers of /proc/mdstat
159 * can use 'poll' or 'select' to find out when the event
163 * start array, stop array, error, add device, remove device,
164 * start build, activate spare
166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
167 static atomic_t md_event_count
;
168 void md_new_event(mddev_t
*mddev
)
170 atomic_inc(&md_event_count
);
171 wake_up(&md_event_waiters
);
173 EXPORT_SYMBOL_GPL(md_new_event
);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 static void md_new_event_inintr(mddev_t
*mddev
)
180 atomic_inc(&md_event_count
);
181 wake_up(&md_event_waiters
);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs
);
189 static DEFINE_SPINLOCK(all_mddevs_lock
);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define for_each_mddev(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (struct request_queue
*q
, struct bio
*bio
)
221 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
223 atomic_inc(&mddev
->active
);
227 static void mddev_put(mddev_t
*mddev
)
229 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
231 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
232 list_del(&mddev
->all_mddevs
);
233 spin_unlock(&all_mddevs_lock
);
234 blk_cleanup_queue(mddev
->queue
);
235 kobject_put(&mddev
->kobj
);
237 spin_unlock(&all_mddevs_lock
);
240 static mddev_t
* mddev_find(dev_t unit
)
242 mddev_t
*mddev
, *new = NULL
;
245 spin_lock(&all_mddevs_lock
);
246 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
247 if (mddev
->unit
== unit
) {
249 spin_unlock(&all_mddevs_lock
);
255 list_add(&new->all_mddevs
, &all_mddevs
);
256 spin_unlock(&all_mddevs_lock
);
259 spin_unlock(&all_mddevs_lock
);
261 new = kzalloc(sizeof(*new), GFP_KERNEL
);
266 if (MAJOR(unit
) == MD_MAJOR
)
267 new->md_minor
= MINOR(unit
);
269 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
271 mutex_init(&new->reconfig_mutex
);
272 INIT_LIST_HEAD(&new->disks
);
273 INIT_LIST_HEAD(&new->all_mddevs
);
274 init_timer(&new->safemode_timer
);
275 atomic_set(&new->active
, 1);
276 spin_lock_init(&new->write_lock
);
277 init_waitqueue_head(&new->sb_wait
);
278 init_waitqueue_head(&new->recovery_wait
);
279 new->reshape_position
= MaxSector
;
281 new->resync_max
= MaxSector
;
282 new->level
= LEVEL_NONE
;
284 new->queue
= blk_alloc_queue(GFP_KERNEL
);
289 /* Can be unlocked because the queue is new: no concurrency */
290 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, new->queue
);
292 blk_queue_make_request(new->queue
, md_fail_request
);
297 static inline int mddev_lock(mddev_t
* mddev
)
299 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
302 static inline int mddev_trylock(mddev_t
* mddev
)
304 return mutex_trylock(&mddev
->reconfig_mutex
);
307 static inline void mddev_unlock(mddev_t
* mddev
)
309 mutex_unlock(&mddev
->reconfig_mutex
);
311 md_wakeup_thread(mddev
->thread
);
314 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
317 struct list_head
*tmp
;
319 rdev_for_each(rdev
, tmp
, mddev
) {
320 if (rdev
->desc_nr
== nr
)
326 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
328 struct list_head
*tmp
;
331 rdev_for_each(rdev
, tmp
, mddev
) {
332 if (rdev
->bdev
->bd_dev
== dev
)
338 static struct mdk_personality
*find_pers(int level
, char *clevel
)
340 struct mdk_personality
*pers
;
341 list_for_each_entry(pers
, &pers_list
, list
) {
342 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
344 if (strcmp(pers
->name
, clevel
)==0)
350 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
352 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
353 return MD_NEW_SIZE_BLOCKS(size
);
356 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
360 size
= rdev
->sb_offset
;
363 size
&= ~((sector_t
)chunk_size
/1024 - 1);
367 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
372 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
373 if (!rdev
->sb_page
) {
374 printk(KERN_ALERT
"md: out of memory.\n");
381 static void free_disk_sb(mdk_rdev_t
* rdev
)
384 put_page(rdev
->sb_page
);
386 rdev
->sb_page
= NULL
;
393 static void super_written(struct bio
*bio
, int error
)
395 mdk_rdev_t
*rdev
= bio
->bi_private
;
396 mddev_t
*mddev
= rdev
->mddev
;
398 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
399 printk("md: super_written gets error=%d, uptodate=%d\n",
400 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
401 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
402 md_error(mddev
, rdev
);
405 if (atomic_dec_and_test(&mddev
->pending_writes
))
406 wake_up(&mddev
->sb_wait
);
410 static void super_written_barrier(struct bio
*bio
, int error
)
412 struct bio
*bio2
= bio
->bi_private
;
413 mdk_rdev_t
*rdev
= bio2
->bi_private
;
414 mddev_t
*mddev
= rdev
->mddev
;
416 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
417 error
== -EOPNOTSUPP
) {
419 /* barriers don't appear to be supported :-( */
420 set_bit(BarriersNotsupp
, &rdev
->flags
);
421 mddev
->barriers_work
= 0;
422 spin_lock_irqsave(&mddev
->write_lock
, flags
);
423 bio2
->bi_next
= mddev
->biolist
;
424 mddev
->biolist
= bio2
;
425 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
426 wake_up(&mddev
->sb_wait
);
430 bio
->bi_private
= rdev
;
431 super_written(bio
, error
);
435 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
436 sector_t sector
, int size
, struct page
*page
)
438 /* write first size bytes of page to sector of rdev
439 * Increment mddev->pending_writes before returning
440 * and decrement it on completion, waking up sb_wait
441 * if zero is reached.
442 * If an error occurred, call md_error
444 * As we might need to resubmit the request if BIO_RW_BARRIER
445 * causes ENOTSUPP, we allocate a spare bio...
447 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
448 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
450 bio
->bi_bdev
= rdev
->bdev
;
451 bio
->bi_sector
= sector
;
452 bio_add_page(bio
, page
, size
, 0);
453 bio
->bi_private
= rdev
;
454 bio
->bi_end_io
= super_written
;
457 atomic_inc(&mddev
->pending_writes
);
458 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
460 rw
|= (1<<BIO_RW_BARRIER
);
461 rbio
= bio_clone(bio
, GFP_NOIO
);
462 rbio
->bi_private
= bio
;
463 rbio
->bi_end_io
= super_written_barrier
;
464 submit_bio(rw
, rbio
);
469 void md_super_wait(mddev_t
*mddev
)
471 /* wait for all superblock writes that were scheduled to complete.
472 * if any had to be retried (due to BARRIER problems), retry them
476 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
477 if (atomic_read(&mddev
->pending_writes
)==0)
479 while (mddev
->biolist
) {
481 spin_lock_irq(&mddev
->write_lock
);
482 bio
= mddev
->biolist
;
483 mddev
->biolist
= bio
->bi_next
;
485 spin_unlock_irq(&mddev
->write_lock
);
486 submit_bio(bio
->bi_rw
, bio
);
490 finish_wait(&mddev
->sb_wait
, &wq
);
493 static void bi_complete(struct bio
*bio
, int error
)
495 complete((struct completion
*)bio
->bi_private
);
498 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
499 struct page
*page
, int rw
)
501 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
502 struct completion event
;
505 rw
|= (1 << BIO_RW_SYNC
);
508 bio
->bi_sector
= sector
;
509 bio_add_page(bio
, page
, size
, 0);
510 init_completion(&event
);
511 bio
->bi_private
= &event
;
512 bio
->bi_end_io
= bi_complete
;
514 wait_for_completion(&event
);
516 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
520 EXPORT_SYMBOL_GPL(sync_page_io
);
522 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
524 char b
[BDEVNAME_SIZE
];
525 if (!rdev
->sb_page
) {
533 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
539 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
540 bdevname(rdev
->bdev
,b
));
544 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
546 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
547 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
548 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
549 (sb1
->set_uuid3
== sb2
->set_uuid3
))
557 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
560 mdp_super_t
*tmp1
, *tmp2
;
562 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
563 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
565 if (!tmp1
|| !tmp2
) {
567 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
575 * nr_disks is not constant
580 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
592 static u32
md_csum_fold(u32 csum
)
594 csum
= (csum
& 0xffff) + (csum
>> 16);
595 return (csum
& 0xffff) + (csum
>> 16);
598 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
601 u32
*sb32
= (u32
*)sb
;
603 unsigned int disk_csum
, csum
;
605 disk_csum
= sb
->sb_csum
;
608 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
610 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
614 /* This used to use csum_partial, which was wrong for several
615 * reasons including that different results are returned on
616 * different architectures. It isn't critical that we get exactly
617 * the same return value as before (we always csum_fold before
618 * testing, and that removes any differences). However as we
619 * know that csum_partial always returned a 16bit value on
620 * alphas, do a fold to maximise conformity to previous behaviour.
622 sb
->sb_csum
= md_csum_fold(disk_csum
);
624 sb
->sb_csum
= disk_csum
;
631 * Handle superblock details.
632 * We want to be able to handle multiple superblock formats
633 * so we have a common interface to them all, and an array of
634 * different handlers.
635 * We rely on user-space to write the initial superblock, and support
636 * reading and updating of superblocks.
637 * Interface methods are:
638 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
639 * loads and validates a superblock on dev.
640 * if refdev != NULL, compare superblocks on both devices
642 * 0 - dev has a superblock that is compatible with refdev
643 * 1 - dev has a superblock that is compatible and newer than refdev
644 * so dev should be used as the refdev in future
645 * -EINVAL superblock incompatible or invalid
646 * -othererror e.g. -EIO
648 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
649 * Verify that dev is acceptable into mddev.
650 * The first time, mddev->raid_disks will be 0, and data from
651 * dev should be merged in. Subsequent calls check that dev
652 * is new enough. Return 0 or -EINVAL
654 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
655 * Update the superblock for rdev with data in mddev
656 * This does not write to disc.
662 struct module
*owner
;
663 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
664 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
665 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
669 * load_super for 0.90.0
671 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
673 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
679 * Calculate the position of the superblock,
680 * it's at the end of the disk.
682 * It also happens to be a multiple of 4Kb.
684 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
685 rdev
->sb_offset
= sb_offset
;
687 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
692 bdevname(rdev
->bdev
, b
);
693 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
695 if (sb
->md_magic
!= MD_SB_MAGIC
) {
696 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
701 if (sb
->major_version
!= 0 ||
702 sb
->minor_version
< 90 ||
703 sb
->minor_version
> 91) {
704 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
705 sb
->major_version
, sb
->minor_version
,
710 if (sb
->raid_disks
<= 0)
713 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
714 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
719 rdev
->preferred_minor
= sb
->md_minor
;
720 rdev
->data_offset
= 0;
721 rdev
->sb_size
= MD_SB_BYTES
;
723 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
724 if (sb
->level
!= 1 && sb
->level
!= 4
725 && sb
->level
!= 5 && sb
->level
!= 6
726 && sb
->level
!= 10) {
727 /* FIXME use a better test */
729 "md: bitmaps not supported for this level.\n");
734 if (sb
->level
== LEVEL_MULTIPATH
)
737 rdev
->desc_nr
= sb
->this_disk
.number
;
743 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
744 if (!uuid_equal(refsb
, sb
)) {
745 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
746 b
, bdevname(refdev
->bdev
,b2
));
749 if (!sb_equal(refsb
, sb
)) {
750 printk(KERN_WARNING
"md: %s has same UUID"
751 " but different superblock to %s\n",
752 b
, bdevname(refdev
->bdev
, b2
));
756 ev2
= md_event(refsb
);
762 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
764 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
765 /* "this cannot possibly happen" ... */
773 * validate_super for 0.90.0
775 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
778 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
779 __u64 ev1
= md_event(sb
);
781 rdev
->raid_disk
= -1;
782 clear_bit(Faulty
, &rdev
->flags
);
783 clear_bit(In_sync
, &rdev
->flags
);
784 clear_bit(WriteMostly
, &rdev
->flags
);
785 clear_bit(BarriersNotsupp
, &rdev
->flags
);
787 if (mddev
->raid_disks
== 0) {
788 mddev
->major_version
= 0;
789 mddev
->minor_version
= sb
->minor_version
;
790 mddev
->patch_version
= sb
->patch_version
;
792 mddev
->chunk_size
= sb
->chunk_size
;
793 mddev
->ctime
= sb
->ctime
;
794 mddev
->utime
= sb
->utime
;
795 mddev
->level
= sb
->level
;
796 mddev
->clevel
[0] = 0;
797 mddev
->layout
= sb
->layout
;
798 mddev
->raid_disks
= sb
->raid_disks
;
799 mddev
->size
= sb
->size
;
801 mddev
->bitmap_offset
= 0;
802 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
804 if (mddev
->minor_version
>= 91) {
805 mddev
->reshape_position
= sb
->reshape_position
;
806 mddev
->delta_disks
= sb
->delta_disks
;
807 mddev
->new_level
= sb
->new_level
;
808 mddev
->new_layout
= sb
->new_layout
;
809 mddev
->new_chunk
= sb
->new_chunk
;
811 mddev
->reshape_position
= MaxSector
;
812 mddev
->delta_disks
= 0;
813 mddev
->new_level
= mddev
->level
;
814 mddev
->new_layout
= mddev
->layout
;
815 mddev
->new_chunk
= mddev
->chunk_size
;
818 if (sb
->state
& (1<<MD_SB_CLEAN
))
819 mddev
->recovery_cp
= MaxSector
;
821 if (sb
->events_hi
== sb
->cp_events_hi
&&
822 sb
->events_lo
== sb
->cp_events_lo
) {
823 mddev
->recovery_cp
= sb
->recovery_cp
;
825 mddev
->recovery_cp
= 0;
828 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
829 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
830 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
831 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
833 mddev
->max_disks
= MD_SB_DISKS
;
835 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
836 mddev
->bitmap_file
== NULL
)
837 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
839 } else if (mddev
->pers
== NULL
) {
840 /* Insist on good event counter while assembling */
842 if (ev1
< mddev
->events
)
844 } else if (mddev
->bitmap
) {
845 /* if adding to array with a bitmap, then we can accept an
846 * older device ... but not too old.
848 if (ev1
< mddev
->bitmap
->events_cleared
)
851 if (ev1
< mddev
->events
)
852 /* just a hot-add of a new device, leave raid_disk at -1 */
856 if (mddev
->level
!= LEVEL_MULTIPATH
) {
857 desc
= sb
->disks
+ rdev
->desc_nr
;
859 if (desc
->state
& (1<<MD_DISK_FAULTY
))
860 set_bit(Faulty
, &rdev
->flags
);
861 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
862 desc->raid_disk < mddev->raid_disks */) {
863 set_bit(In_sync
, &rdev
->flags
);
864 rdev
->raid_disk
= desc
->raid_disk
;
866 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
867 set_bit(WriteMostly
, &rdev
->flags
);
868 } else /* MULTIPATH are always insync */
869 set_bit(In_sync
, &rdev
->flags
);
874 * sync_super for 0.90.0
876 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
879 struct list_head
*tmp
;
881 int next_spare
= mddev
->raid_disks
;
884 /* make rdev->sb match mddev data..
887 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
888 * 3/ any empty disks < next_spare become removed
890 * disks[0] gets initialised to REMOVED because
891 * we cannot be sure from other fields if it has
892 * been initialised or not.
895 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
897 rdev
->sb_size
= MD_SB_BYTES
;
899 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
901 memset(sb
, 0, sizeof(*sb
));
903 sb
->md_magic
= MD_SB_MAGIC
;
904 sb
->major_version
= mddev
->major_version
;
905 sb
->patch_version
= mddev
->patch_version
;
906 sb
->gvalid_words
= 0; /* ignored */
907 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
908 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
909 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
910 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
912 sb
->ctime
= mddev
->ctime
;
913 sb
->level
= mddev
->level
;
914 sb
->size
= mddev
->size
;
915 sb
->raid_disks
= mddev
->raid_disks
;
916 sb
->md_minor
= mddev
->md_minor
;
917 sb
->not_persistent
= 0;
918 sb
->utime
= mddev
->utime
;
920 sb
->events_hi
= (mddev
->events
>>32);
921 sb
->events_lo
= (u32
)mddev
->events
;
923 if (mddev
->reshape_position
== MaxSector
)
924 sb
->minor_version
= 90;
926 sb
->minor_version
= 91;
927 sb
->reshape_position
= mddev
->reshape_position
;
928 sb
->new_level
= mddev
->new_level
;
929 sb
->delta_disks
= mddev
->delta_disks
;
930 sb
->new_layout
= mddev
->new_layout
;
931 sb
->new_chunk
= mddev
->new_chunk
;
933 mddev
->minor_version
= sb
->minor_version
;
936 sb
->recovery_cp
= mddev
->recovery_cp
;
937 sb
->cp_events_hi
= (mddev
->events
>>32);
938 sb
->cp_events_lo
= (u32
)mddev
->events
;
939 if (mddev
->recovery_cp
== MaxSector
)
940 sb
->state
= (1<< MD_SB_CLEAN
);
944 sb
->layout
= mddev
->layout
;
945 sb
->chunk_size
= mddev
->chunk_size
;
947 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
948 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
950 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
951 rdev_for_each(rdev2
, tmp
, mddev
) {
954 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
955 && !test_bit(Faulty
, &rdev2
->flags
))
956 desc_nr
= rdev2
->raid_disk
;
958 desc_nr
= next_spare
++;
959 rdev2
->desc_nr
= desc_nr
;
960 d
= &sb
->disks
[rdev2
->desc_nr
];
962 d
->number
= rdev2
->desc_nr
;
963 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
964 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
965 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
966 && !test_bit(Faulty
, &rdev2
->flags
))
967 d
->raid_disk
= rdev2
->raid_disk
;
969 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
970 if (test_bit(Faulty
, &rdev2
->flags
))
971 d
->state
= (1<<MD_DISK_FAULTY
);
972 else if (test_bit(In_sync
, &rdev2
->flags
)) {
973 d
->state
= (1<<MD_DISK_ACTIVE
);
974 d
->state
|= (1<<MD_DISK_SYNC
);
982 if (test_bit(WriteMostly
, &rdev2
->flags
))
983 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
985 /* now set the "removed" and "faulty" bits on any missing devices */
986 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
987 mdp_disk_t
*d
= &sb
->disks
[i
];
988 if (d
->state
== 0 && d
->number
== 0) {
991 d
->state
= (1<<MD_DISK_REMOVED
);
992 d
->state
|= (1<<MD_DISK_FAULTY
);
996 sb
->nr_disks
= nr_disks
;
997 sb
->active_disks
= active
;
998 sb
->working_disks
= working
;
999 sb
->failed_disks
= failed
;
1000 sb
->spare_disks
= spare
;
1002 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1003 sb
->sb_csum
= calc_sb_csum(sb
);
1007 * version 1 superblock
1010 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1014 unsigned long long newcsum
;
1015 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1016 __le32
*isuper
= (__le32
*)sb
;
1019 disk_csum
= sb
->sb_csum
;
1022 for (i
=0; size
>=4; size
-= 4 )
1023 newcsum
+= le32_to_cpu(*isuper
++);
1026 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1028 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1029 sb
->sb_csum
= disk_csum
;
1030 return cpu_to_le32(csum
);
1033 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1035 struct mdp_superblock_1
*sb
;
1038 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1042 * Calculate the position of the superblock.
1043 * It is always aligned to a 4K boundary and
1044 * depeding on minor_version, it can be:
1045 * 0: At least 8K, but less than 12K, from end of device
1046 * 1: At start of device
1047 * 2: 4K from start of device.
1049 switch(minor_version
) {
1051 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1053 sb_offset
&= ~(sector_t
)(4*2-1);
1054 /* convert from sectors to K */
1066 rdev
->sb_offset
= sb_offset
;
1068 /* superblock is rarely larger than 1K, but it can be larger,
1069 * and it is safe to read 4k, so we do that
1071 ret
= read_disk_sb(rdev
, 4096);
1072 if (ret
) return ret
;
1075 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1077 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1078 sb
->major_version
!= cpu_to_le32(1) ||
1079 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1080 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1081 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1084 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1085 printk("md: invalid superblock checksum on %s\n",
1086 bdevname(rdev
->bdev
,b
));
1089 if (le64_to_cpu(sb
->data_size
) < 10) {
1090 printk("md: data_size too small on %s\n",
1091 bdevname(rdev
->bdev
,b
));
1094 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)) {
1095 if (sb
->level
!= cpu_to_le32(1) &&
1096 sb
->level
!= cpu_to_le32(4) &&
1097 sb
->level
!= cpu_to_le32(5) &&
1098 sb
->level
!= cpu_to_le32(6) &&
1099 sb
->level
!= cpu_to_le32(10)) {
1101 "md: bitmaps not supported for this level.\n");
1106 rdev
->preferred_minor
= 0xffff;
1107 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1108 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1110 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1111 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1112 if (rdev
->sb_size
& bmask
)
1113 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1116 && rdev
->data_offset
< sb_offset
+ (rdev
->sb_size
/512))
1119 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1122 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1128 struct mdp_superblock_1
*refsb
=
1129 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1131 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1132 sb
->level
!= refsb
->level
||
1133 sb
->layout
!= refsb
->layout
||
1134 sb
->chunksize
!= refsb
->chunksize
) {
1135 printk(KERN_WARNING
"md: %s has strangely different"
1136 " superblock to %s\n",
1137 bdevname(rdev
->bdev
,b
),
1138 bdevname(refdev
->bdev
,b2
));
1141 ev1
= le64_to_cpu(sb
->events
);
1142 ev2
= le64_to_cpu(refsb
->events
);
1150 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1152 rdev
->size
= rdev
->sb_offset
;
1153 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1155 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1156 if (le32_to_cpu(sb
->chunksize
))
1157 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1159 if (le64_to_cpu(sb
->size
) > rdev
->size
*2)
1164 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1166 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1167 __u64 ev1
= le64_to_cpu(sb
->events
);
1169 rdev
->raid_disk
= -1;
1170 clear_bit(Faulty
, &rdev
->flags
);
1171 clear_bit(In_sync
, &rdev
->flags
);
1172 clear_bit(WriteMostly
, &rdev
->flags
);
1173 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1175 if (mddev
->raid_disks
== 0) {
1176 mddev
->major_version
= 1;
1177 mddev
->patch_version
= 0;
1178 mddev
->external
= 0;
1179 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1180 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1181 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1182 mddev
->level
= le32_to_cpu(sb
->level
);
1183 mddev
->clevel
[0] = 0;
1184 mddev
->layout
= le32_to_cpu(sb
->layout
);
1185 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1186 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1187 mddev
->events
= ev1
;
1188 mddev
->bitmap_offset
= 0;
1189 mddev
->default_bitmap_offset
= 1024 >> 9;
1191 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1192 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1194 mddev
->max_disks
= (4096-256)/2;
1196 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1197 mddev
->bitmap_file
== NULL
)
1198 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1200 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1201 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1202 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1203 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1204 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1205 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1207 mddev
->reshape_position
= MaxSector
;
1208 mddev
->delta_disks
= 0;
1209 mddev
->new_level
= mddev
->level
;
1210 mddev
->new_layout
= mddev
->layout
;
1211 mddev
->new_chunk
= mddev
->chunk_size
;
1214 } else if (mddev
->pers
== NULL
) {
1215 /* Insist of good event counter while assembling */
1217 if (ev1
< mddev
->events
)
1219 } else if (mddev
->bitmap
) {
1220 /* If adding to array with a bitmap, then we can accept an
1221 * older device, but not too old.
1223 if (ev1
< mddev
->bitmap
->events_cleared
)
1226 if (ev1
< mddev
->events
)
1227 /* just a hot-add of a new device, leave raid_disk at -1 */
1230 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1232 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1234 case 0xffff: /* spare */
1236 case 0xfffe: /* faulty */
1237 set_bit(Faulty
, &rdev
->flags
);
1240 if ((le32_to_cpu(sb
->feature_map
) &
1241 MD_FEATURE_RECOVERY_OFFSET
))
1242 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1244 set_bit(In_sync
, &rdev
->flags
);
1245 rdev
->raid_disk
= role
;
1248 if (sb
->devflags
& WriteMostly1
)
1249 set_bit(WriteMostly
, &rdev
->flags
);
1250 } else /* MULTIPATH are always insync */
1251 set_bit(In_sync
, &rdev
->flags
);
1256 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1258 struct mdp_superblock_1
*sb
;
1259 struct list_head
*tmp
;
1262 /* make rdev->sb match mddev and rdev data. */
1264 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1266 sb
->feature_map
= 0;
1268 sb
->recovery_offset
= cpu_to_le64(0);
1269 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1270 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1271 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1273 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1274 sb
->events
= cpu_to_le64(mddev
->events
);
1276 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1278 sb
->resync_offset
= cpu_to_le64(0);
1280 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1282 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1283 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1285 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1286 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1287 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1290 if (rdev
->raid_disk
>= 0 &&
1291 !test_bit(In_sync
, &rdev
->flags
) &&
1292 rdev
->recovery_offset
> 0) {
1293 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1294 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1297 if (mddev
->reshape_position
!= MaxSector
) {
1298 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1299 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1300 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1301 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1302 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1303 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1307 rdev_for_each(rdev2
, tmp
, mddev
)
1308 if (rdev2
->desc_nr
+1 > max_dev
)
1309 max_dev
= rdev2
->desc_nr
+1;
1311 if (max_dev
> le32_to_cpu(sb
->max_dev
))
1312 sb
->max_dev
= cpu_to_le32(max_dev
);
1313 for (i
=0; i
<max_dev
;i
++)
1314 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1316 rdev_for_each(rdev2
, tmp
, mddev
) {
1318 if (test_bit(Faulty
, &rdev2
->flags
))
1319 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1320 else if (test_bit(In_sync
, &rdev2
->flags
))
1321 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1322 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1323 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1325 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1328 sb
->sb_csum
= calc_sb_1_csum(sb
);
1332 static struct super_type super_types
[] = {
1335 .owner
= THIS_MODULE
,
1336 .load_super
= super_90_load
,
1337 .validate_super
= super_90_validate
,
1338 .sync_super
= super_90_sync
,
1342 .owner
= THIS_MODULE
,
1343 .load_super
= super_1_load
,
1344 .validate_super
= super_1_validate
,
1345 .sync_super
= super_1_sync
,
1349 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1351 struct list_head
*tmp
, *tmp2
;
1352 mdk_rdev_t
*rdev
, *rdev2
;
1354 rdev_for_each(rdev
, tmp
, mddev1
)
1355 rdev_for_each(rdev2
, tmp2
, mddev2
)
1356 if (rdev
->bdev
->bd_contains
==
1357 rdev2
->bdev
->bd_contains
)
1363 static LIST_HEAD(pending_raid_disks
);
1365 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1367 char b
[BDEVNAME_SIZE
];
1377 /* prevent duplicates */
1378 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1381 /* make sure rdev->size exceeds mddev->size */
1382 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1384 /* Cannot change size, so fail
1385 * If mddev->level <= 0, then we don't care
1386 * about aligning sizes (e.g. linear)
1388 if (mddev
->level
> 0)
1391 mddev
->size
= rdev
->size
;
1394 /* Verify rdev->desc_nr is unique.
1395 * If it is -1, assign a free number, else
1396 * check number is not in use
1398 if (rdev
->desc_nr
< 0) {
1400 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1401 while (find_rdev_nr(mddev
, choice
))
1403 rdev
->desc_nr
= choice
;
1405 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1408 bdevname(rdev
->bdev
,b
);
1409 while ( (s
=strchr(b
, '/')) != NULL
)
1412 rdev
->mddev
= mddev
;
1413 printk(KERN_INFO
"md: bind<%s>\n", b
);
1415 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1418 if (rdev
->bdev
->bd_part
)
1419 ko
= &rdev
->bdev
->bd_part
->dev
.kobj
;
1421 ko
= &rdev
->bdev
->bd_disk
->dev
.kobj
;
1422 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1423 kobject_del(&rdev
->kobj
);
1426 list_add(&rdev
->same_set
, &mddev
->disks
);
1427 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1431 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1436 static void md_delayed_delete(struct work_struct
*ws
)
1438 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1439 kobject_del(&rdev
->kobj
);
1440 kobject_put(&rdev
->kobj
);
1443 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1445 char b
[BDEVNAME_SIZE
];
1450 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1451 list_del_init(&rdev
->same_set
);
1452 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1454 sysfs_remove_link(&rdev
->kobj
, "block");
1456 /* We need to delay this, otherwise we can deadlock when
1457 * writing to 'remove' to "dev/state"
1459 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1460 kobject_get(&rdev
->kobj
);
1461 schedule_work(&rdev
->del_work
);
1465 * prevent the device from being mounted, repartitioned or
1466 * otherwise reused by a RAID array (or any other kernel
1467 * subsystem), by bd_claiming the device.
1469 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1472 struct block_device
*bdev
;
1473 char b
[BDEVNAME_SIZE
];
1475 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1477 printk(KERN_ERR
"md: could not open %s.\n",
1478 __bdevname(dev
, b
));
1479 return PTR_ERR(bdev
);
1481 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1483 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1489 set_bit(AllReserved
, &rdev
->flags
);
1494 static void unlock_rdev(mdk_rdev_t
*rdev
)
1496 struct block_device
*bdev
= rdev
->bdev
;
1504 void md_autodetect_dev(dev_t dev
);
1506 static void export_rdev(mdk_rdev_t
* rdev
)
1508 char b
[BDEVNAME_SIZE
];
1509 printk(KERN_INFO
"md: export_rdev(%s)\n",
1510 bdevname(rdev
->bdev
,b
));
1514 list_del_init(&rdev
->same_set
);
1516 if (test_bit(AutoDetected
, &rdev
->flags
))
1517 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1520 kobject_put(&rdev
->kobj
);
1523 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1525 unbind_rdev_from_array(rdev
);
1529 static void export_array(mddev_t
*mddev
)
1531 struct list_head
*tmp
;
1534 rdev_for_each(rdev
, tmp
, mddev
) {
1539 kick_rdev_from_array(rdev
);
1541 if (!list_empty(&mddev
->disks
))
1543 mddev
->raid_disks
= 0;
1544 mddev
->major_version
= 0;
1547 static void print_desc(mdp_disk_t
*desc
)
1549 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1550 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1553 static void print_sb(mdp_super_t
*sb
)
1558 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1559 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1560 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1562 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1563 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1564 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1565 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1566 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1567 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1568 sb
->failed_disks
, sb
->spare_disks
,
1569 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1572 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1575 desc
= sb
->disks
+ i
;
1576 if (desc
->number
|| desc
->major
|| desc
->minor
||
1577 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1578 printk(" D %2d: ", i
);
1582 printk(KERN_INFO
"md: THIS: ");
1583 print_desc(&sb
->this_disk
);
1587 static void print_rdev(mdk_rdev_t
*rdev
)
1589 char b
[BDEVNAME_SIZE
];
1590 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1591 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1592 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1594 if (rdev
->sb_loaded
) {
1595 printk(KERN_INFO
"md: rdev superblock:\n");
1596 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1598 printk(KERN_INFO
"md: no rdev superblock!\n");
1601 static void md_print_devices(void)
1603 struct list_head
*tmp
, *tmp2
;
1606 char b
[BDEVNAME_SIZE
];
1609 printk("md: **********************************\n");
1610 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1611 printk("md: **********************************\n");
1612 for_each_mddev(mddev
, tmp
) {
1615 bitmap_print_sb(mddev
->bitmap
);
1617 printk("%s: ", mdname(mddev
));
1618 rdev_for_each(rdev
, tmp2
, mddev
)
1619 printk("<%s>", bdevname(rdev
->bdev
,b
));
1622 rdev_for_each(rdev
, tmp2
, mddev
)
1625 printk("md: **********************************\n");
1630 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1632 /* Update each superblock (in-memory image), but
1633 * if we are allowed to, skip spares which already
1634 * have the right event counter, or have one earlier
1635 * (which would mean they aren't being marked as dirty
1636 * with the rest of the array)
1639 struct list_head
*tmp
;
1641 rdev_for_each(rdev
, tmp
, mddev
) {
1642 if (rdev
->sb_events
== mddev
->events
||
1644 rdev
->raid_disk
< 0 &&
1645 (rdev
->sb_events
&1)==0 &&
1646 rdev
->sb_events
+1 == mddev
->events
)) {
1647 /* Don't update this superblock */
1648 rdev
->sb_loaded
= 2;
1650 super_types
[mddev
->major_version
].
1651 sync_super(mddev
, rdev
);
1652 rdev
->sb_loaded
= 1;
1657 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1659 struct list_head
*tmp
;
1664 if (mddev
->external
)
1667 spin_lock_irq(&mddev
->write_lock
);
1669 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1670 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1672 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1673 /* just a clean<-> dirty transition, possibly leave spares alone,
1674 * though if events isn't the right even/odd, we will have to do
1680 if (mddev
->degraded
)
1681 /* If the array is degraded, then skipping spares is both
1682 * dangerous and fairly pointless.
1683 * Dangerous because a device that was removed from the array
1684 * might have a event_count that still looks up-to-date,
1685 * so it can be re-added without a resync.
1686 * Pointless because if there are any spares to skip,
1687 * then a recovery will happen and soon that array won't
1688 * be degraded any more and the spare can go back to sleep then.
1692 sync_req
= mddev
->in_sync
;
1693 mddev
->utime
= get_seconds();
1695 /* If this is just a dirty<->clean transition, and the array is clean
1696 * and 'events' is odd, we can roll back to the previous clean state */
1698 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1699 && (mddev
->events
& 1)
1700 && mddev
->events
!= 1)
1703 /* otherwise we have to go forward and ... */
1705 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1706 /* .. if the array isn't clean, insist on an odd 'events' */
1707 if ((mddev
->events
&1)==0) {
1712 /* otherwise insist on an even 'events' (for clean states) */
1713 if ((mddev
->events
&1)) {
1720 if (!mddev
->events
) {
1722 * oops, this 64-bit counter should never wrap.
1723 * Either we are in around ~1 trillion A.C., assuming
1724 * 1 reboot per second, or we have a bug:
1731 * do not write anything to disk if using
1732 * nonpersistent superblocks
1734 if (!mddev
->persistent
) {
1735 if (!mddev
->external
)
1736 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1738 spin_unlock_irq(&mddev
->write_lock
);
1739 wake_up(&mddev
->sb_wait
);
1742 sync_sbs(mddev
, nospares
);
1743 spin_unlock_irq(&mddev
->write_lock
);
1746 "md: updating %s RAID superblock on device (in sync %d)\n",
1747 mdname(mddev
),mddev
->in_sync
);
1749 bitmap_update_sb(mddev
->bitmap
);
1750 rdev_for_each(rdev
, tmp
, mddev
) {
1751 char b
[BDEVNAME_SIZE
];
1752 dprintk(KERN_INFO
"md: ");
1753 if (rdev
->sb_loaded
!= 1)
1754 continue; /* no noise on spare devices */
1755 if (test_bit(Faulty
, &rdev
->flags
))
1756 dprintk("(skipping faulty ");
1758 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1759 if (!test_bit(Faulty
, &rdev
->flags
)) {
1760 md_super_write(mddev
,rdev
,
1761 rdev
->sb_offset
<<1, rdev
->sb_size
,
1763 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1764 bdevname(rdev
->bdev
,b
),
1765 (unsigned long long)rdev
->sb_offset
);
1766 rdev
->sb_events
= mddev
->events
;
1770 if (mddev
->level
== LEVEL_MULTIPATH
)
1771 /* only need to write one superblock... */
1774 md_super_wait(mddev
);
1775 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1777 spin_lock_irq(&mddev
->write_lock
);
1778 if (mddev
->in_sync
!= sync_req
||
1779 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
1780 /* have to write it out again */
1781 spin_unlock_irq(&mddev
->write_lock
);
1784 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1785 spin_unlock_irq(&mddev
->write_lock
);
1786 wake_up(&mddev
->sb_wait
);
1790 /* words written to sysfs files may, or my not, be \n terminated.
1791 * We want to accept with case. For this we use cmd_match.
1793 static int cmd_match(const char *cmd
, const char *str
)
1795 /* See if cmd, written into a sysfs file, matches
1796 * str. They must either be the same, or cmd can
1797 * have a trailing newline
1799 while (*cmd
&& *str
&& *cmd
== *str
) {
1810 struct rdev_sysfs_entry
{
1811 struct attribute attr
;
1812 ssize_t (*show
)(mdk_rdev_t
*, char *);
1813 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1817 state_show(mdk_rdev_t
*rdev
, char *page
)
1822 if (test_bit(Faulty
, &rdev
->flags
)) {
1823 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1826 if (test_bit(In_sync
, &rdev
->flags
)) {
1827 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1830 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1831 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1834 if (test_bit(Blocked
, &rdev
->flags
)) {
1835 len
+= sprintf(page
+len
, "%sblocked", sep
);
1838 if (!test_bit(Faulty
, &rdev
->flags
) &&
1839 !test_bit(In_sync
, &rdev
->flags
)) {
1840 len
+= sprintf(page
+len
, "%sspare", sep
);
1843 return len
+sprintf(page
+len
, "\n");
1847 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1850 * faulty - simulates and error
1851 * remove - disconnects the device
1852 * writemostly - sets write_mostly
1853 * -writemostly - clears write_mostly
1854 * blocked - sets the Blocked flag
1855 * -blocked - clears the Blocked flag
1858 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1859 md_error(rdev
->mddev
, rdev
);
1861 } else if (cmd_match(buf
, "remove")) {
1862 if (rdev
->raid_disk
>= 0)
1865 mddev_t
*mddev
= rdev
->mddev
;
1866 kick_rdev_from_array(rdev
);
1868 md_update_sb(mddev
, 1);
1869 md_new_event(mddev
);
1872 } else if (cmd_match(buf
, "writemostly")) {
1873 set_bit(WriteMostly
, &rdev
->flags
);
1875 } else if (cmd_match(buf
, "-writemostly")) {
1876 clear_bit(WriteMostly
, &rdev
->flags
);
1878 } else if (cmd_match(buf
, "blocked")) {
1879 set_bit(Blocked
, &rdev
->flags
);
1881 } else if (cmd_match(buf
, "-blocked")) {
1882 clear_bit(Blocked
, &rdev
->flags
);
1883 wake_up(&rdev
->blocked_wait
);
1884 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
1885 md_wakeup_thread(rdev
->mddev
->thread
);
1889 return err
? err
: len
;
1891 static struct rdev_sysfs_entry rdev_state
=
1892 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
1895 errors_show(mdk_rdev_t
*rdev
, char *page
)
1897 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1901 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1904 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1905 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1906 atomic_set(&rdev
->corrected_errors
, n
);
1911 static struct rdev_sysfs_entry rdev_errors
=
1912 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
1915 slot_show(mdk_rdev_t
*rdev
, char *page
)
1917 if (rdev
->raid_disk
< 0)
1918 return sprintf(page
, "none\n");
1920 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1924 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1929 int slot
= simple_strtoul(buf
, &e
, 10);
1930 if (strncmp(buf
, "none", 4)==0)
1932 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1934 if (rdev
->mddev
->pers
&& slot
== -1) {
1935 /* Setting 'slot' on an active array requires also
1936 * updating the 'rd%d' link, and communicating
1937 * with the personality with ->hot_*_disk.
1938 * For now we only support removing
1939 * failed/spare devices. This normally happens automatically,
1940 * but not when the metadata is externally managed.
1942 if (rdev
->raid_disk
== -1)
1944 /* personality does all needed checks */
1945 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
1947 err
= rdev
->mddev
->pers
->
1948 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
1951 sprintf(nm
, "rd%d", rdev
->raid_disk
);
1952 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
1953 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
1954 md_wakeup_thread(rdev
->mddev
->thread
);
1955 } else if (rdev
->mddev
->pers
) {
1957 struct list_head
*tmp
;
1958 /* Activating a spare .. or possibly reactivating
1959 * if we every get bitmaps working here.
1962 if (rdev
->raid_disk
!= -1)
1965 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
1968 rdev_for_each(rdev2
, tmp
, rdev
->mddev
)
1969 if (rdev2
->raid_disk
== slot
)
1972 rdev
->raid_disk
= slot
;
1973 if (test_bit(In_sync
, &rdev
->flags
))
1974 rdev
->saved_raid_disk
= slot
;
1976 rdev
->saved_raid_disk
= -1;
1977 err
= rdev
->mddev
->pers
->
1978 hot_add_disk(rdev
->mddev
, rdev
);
1980 rdev
->raid_disk
= -1;
1983 sprintf(nm
, "rd%d", rdev
->raid_disk
);
1984 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
1986 "md: cannot register "
1988 nm
, mdname(rdev
->mddev
));
1990 /* don't wakeup anyone, leave that to userspace. */
1992 if (slot
>= rdev
->mddev
->raid_disks
)
1994 rdev
->raid_disk
= slot
;
1995 /* assume it is working */
1996 clear_bit(Faulty
, &rdev
->flags
);
1997 clear_bit(WriteMostly
, &rdev
->flags
);
1998 set_bit(In_sync
, &rdev
->flags
);
2004 static struct rdev_sysfs_entry rdev_slot
=
2005 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2008 offset_show(mdk_rdev_t
*rdev
, char *page
)
2010 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2014 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2017 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2018 if (e
==buf
|| (*e
&& *e
!= '\n'))
2020 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2022 if (rdev
->size
&& rdev
->mddev
->external
)
2023 /* Must set offset before size, so overlap checks
2026 rdev
->data_offset
= offset
;
2030 static struct rdev_sysfs_entry rdev_offset
=
2031 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2034 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2036 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
2039 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2041 /* check if two start/length pairs overlap */
2050 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2053 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2054 unsigned long long oldsize
= rdev
->size
;
2055 mddev_t
*my_mddev
= rdev
->mddev
;
2057 if (e
==buf
|| (*e
&& *e
!= '\n'))
2059 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0)
2062 if (size
> oldsize
&& rdev
->mddev
->external
) {
2063 /* need to check that all other rdevs with the same ->bdev
2064 * do not overlap. We need to unlock the mddev to avoid
2065 * a deadlock. We have already changed rdev->size, and if
2066 * we have to change it back, we will have the lock again.
2070 struct list_head
*tmp
, *tmp2
;
2072 mddev_unlock(my_mddev
);
2073 for_each_mddev(mddev
, tmp
) {
2077 rdev_for_each(rdev2
, tmp2
, mddev
)
2078 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2079 (rdev
->bdev
== rdev2
->bdev
&&
2081 overlaps(rdev
->data_offset
, rdev
->size
,
2082 rdev2
->data_offset
, rdev2
->size
))) {
2086 mddev_unlock(mddev
);
2092 mddev_lock(my_mddev
);
2094 /* Someone else could have slipped in a size
2095 * change here, but doing so is just silly.
2096 * We put oldsize back because we *know* it is
2097 * safe, and trust userspace not to race with
2100 rdev
->size
= oldsize
;
2104 if (size
< my_mddev
->size
|| my_mddev
->size
== 0)
2105 my_mddev
->size
= size
;
2109 static struct rdev_sysfs_entry rdev_size
=
2110 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2112 static struct attribute
*rdev_default_attrs
[] = {
2121 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2123 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2124 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2125 mddev_t
*mddev
= rdev
->mddev
;
2131 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2133 if (rdev
->mddev
== NULL
)
2136 rv
= entry
->show(rdev
, page
);
2137 mddev_unlock(mddev
);
2143 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2144 const char *page
, size_t length
)
2146 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2147 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2149 mddev_t
*mddev
= rdev
->mddev
;
2153 if (!capable(CAP_SYS_ADMIN
))
2155 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2157 if (rdev
->mddev
== NULL
)
2160 rv
= entry
->store(rdev
, page
, length
);
2161 mddev_unlock(mddev
);
2166 static void rdev_free(struct kobject
*ko
)
2168 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2171 static struct sysfs_ops rdev_sysfs_ops
= {
2172 .show
= rdev_attr_show
,
2173 .store
= rdev_attr_store
,
2175 static struct kobj_type rdev_ktype
= {
2176 .release
= rdev_free
,
2177 .sysfs_ops
= &rdev_sysfs_ops
,
2178 .default_attrs
= rdev_default_attrs
,
2182 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2184 * mark the device faulty if:
2186 * - the device is nonexistent (zero size)
2187 * - the device has no valid superblock
2189 * a faulty rdev _never_ has rdev->sb set.
2191 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2193 char b
[BDEVNAME_SIZE
];
2198 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2200 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2201 return ERR_PTR(-ENOMEM
);
2204 if ((err
= alloc_disk_sb(rdev
)))
2207 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2211 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2214 rdev
->saved_raid_disk
= -1;
2215 rdev
->raid_disk
= -1;
2217 rdev
->data_offset
= 0;
2218 rdev
->sb_events
= 0;
2219 atomic_set(&rdev
->nr_pending
, 0);
2220 atomic_set(&rdev
->read_errors
, 0);
2221 atomic_set(&rdev
->corrected_errors
, 0);
2223 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2226 "md: %s has zero or unknown size, marking faulty!\n",
2227 bdevname(rdev
->bdev
,b
));
2232 if (super_format
>= 0) {
2233 err
= super_types
[super_format
].
2234 load_super(rdev
, NULL
, super_minor
);
2235 if (err
== -EINVAL
) {
2237 "md: %s does not have a valid v%d.%d "
2238 "superblock, not importing!\n",
2239 bdevname(rdev
->bdev
,b
),
2240 super_format
, super_minor
);
2245 "md: could not read %s's sb, not importing!\n",
2246 bdevname(rdev
->bdev
,b
));
2251 INIT_LIST_HEAD(&rdev
->same_set
);
2252 init_waitqueue_head(&rdev
->blocked_wait
);
2257 if (rdev
->sb_page
) {
2263 return ERR_PTR(err
);
2267 * Check a full RAID array for plausibility
2271 static void analyze_sbs(mddev_t
* mddev
)
2274 struct list_head
*tmp
;
2275 mdk_rdev_t
*rdev
, *freshest
;
2276 char b
[BDEVNAME_SIZE
];
2279 rdev_for_each(rdev
, tmp
, mddev
)
2280 switch (super_types
[mddev
->major_version
].
2281 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2289 "md: fatal superblock inconsistency in %s"
2290 " -- removing from array\n",
2291 bdevname(rdev
->bdev
,b
));
2292 kick_rdev_from_array(rdev
);
2296 super_types
[mddev
->major_version
].
2297 validate_super(mddev
, freshest
);
2300 rdev_for_each(rdev
, tmp
, mddev
) {
2301 if (rdev
!= freshest
)
2302 if (super_types
[mddev
->major_version
].
2303 validate_super(mddev
, rdev
)) {
2304 printk(KERN_WARNING
"md: kicking non-fresh %s"
2306 bdevname(rdev
->bdev
,b
));
2307 kick_rdev_from_array(rdev
);
2310 if (mddev
->level
== LEVEL_MULTIPATH
) {
2311 rdev
->desc_nr
= i
++;
2312 rdev
->raid_disk
= rdev
->desc_nr
;
2313 set_bit(In_sync
, &rdev
->flags
);
2314 } else if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2315 rdev
->raid_disk
= -1;
2316 clear_bit(In_sync
, &rdev
->flags
);
2322 if (mddev
->recovery_cp
!= MaxSector
&&
2324 printk(KERN_ERR
"md: %s: raid array is not clean"
2325 " -- starting background reconstruction\n",
2331 safe_delay_show(mddev_t
*mddev
, char *page
)
2333 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2334 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2337 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2345 /* remove a period, and count digits after it */
2346 if (len
>= sizeof(buf
))
2348 strlcpy(buf
, cbuf
, len
);
2350 for (i
=0; i
<len
; i
++) {
2352 if (isdigit(buf
[i
])) {
2357 } else if (buf
[i
] == '.') {
2362 msec
= simple_strtoul(buf
, &e
, 10);
2363 if (e
== buf
|| (*e
&& *e
!= '\n'))
2365 msec
= (msec
* 1000) / scale
;
2367 mddev
->safemode_delay
= 0;
2369 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2370 if (mddev
->safemode_delay
== 0)
2371 mddev
->safemode_delay
= 1;
2375 static struct md_sysfs_entry md_safe_delay
=
2376 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2379 level_show(mddev_t
*mddev
, char *page
)
2381 struct mdk_personality
*p
= mddev
->pers
;
2383 return sprintf(page
, "%s\n", p
->name
);
2384 else if (mddev
->clevel
[0])
2385 return sprintf(page
, "%s\n", mddev
->clevel
);
2386 else if (mddev
->level
!= LEVEL_NONE
)
2387 return sprintf(page
, "%d\n", mddev
->level
);
2393 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2400 if (len
>= sizeof(mddev
->clevel
))
2402 strncpy(mddev
->clevel
, buf
, len
);
2403 if (mddev
->clevel
[len
-1] == '\n')
2405 mddev
->clevel
[len
] = 0;
2406 mddev
->level
= LEVEL_NONE
;
2410 static struct md_sysfs_entry md_level
=
2411 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2415 layout_show(mddev_t
*mddev
, char *page
)
2417 /* just a number, not meaningful for all levels */
2418 if (mddev
->reshape_position
!= MaxSector
&&
2419 mddev
->layout
!= mddev
->new_layout
)
2420 return sprintf(page
, "%d (%d)\n",
2421 mddev
->new_layout
, mddev
->layout
);
2422 return sprintf(page
, "%d\n", mddev
->layout
);
2426 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2429 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2431 if (!*buf
|| (*e
&& *e
!= '\n'))
2436 if (mddev
->reshape_position
!= MaxSector
)
2437 mddev
->new_layout
= n
;
2442 static struct md_sysfs_entry md_layout
=
2443 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2447 raid_disks_show(mddev_t
*mddev
, char *page
)
2449 if (mddev
->raid_disks
== 0)
2451 if (mddev
->reshape_position
!= MaxSector
&&
2452 mddev
->delta_disks
!= 0)
2453 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
2454 mddev
->raid_disks
- mddev
->delta_disks
);
2455 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2458 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2461 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2465 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2467 if (!*buf
|| (*e
&& *e
!= '\n'))
2471 rv
= update_raid_disks(mddev
, n
);
2472 else if (mddev
->reshape_position
!= MaxSector
) {
2473 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
2474 mddev
->delta_disks
= n
- olddisks
;
2475 mddev
->raid_disks
= n
;
2477 mddev
->raid_disks
= n
;
2478 return rv
? rv
: len
;
2480 static struct md_sysfs_entry md_raid_disks
=
2481 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2484 chunk_size_show(mddev_t
*mddev
, char *page
)
2486 if (mddev
->reshape_position
!= MaxSector
&&
2487 mddev
->chunk_size
!= mddev
->new_chunk
)
2488 return sprintf(page
, "%d (%d)\n", mddev
->new_chunk
,
2490 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2494 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2496 /* can only set chunk_size if array is not yet active */
2498 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2500 if (!*buf
|| (*e
&& *e
!= '\n'))
2505 else if (mddev
->reshape_position
!= MaxSector
)
2506 mddev
->new_chunk
= n
;
2508 mddev
->chunk_size
= n
;
2511 static struct md_sysfs_entry md_chunk_size
=
2512 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2515 resync_start_show(mddev_t
*mddev
, char *page
)
2517 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2521 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2524 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2528 if (!*buf
|| (*e
&& *e
!= '\n'))
2531 mddev
->recovery_cp
= n
;
2534 static struct md_sysfs_entry md_resync_start
=
2535 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2538 * The array state can be:
2541 * No devices, no size, no level
2542 * Equivalent to STOP_ARRAY ioctl
2544 * May have some settings, but array is not active
2545 * all IO results in error
2546 * When written, doesn't tear down array, but just stops it
2547 * suspended (not supported yet)
2548 * All IO requests will block. The array can be reconfigured.
2549 * Writing this, if accepted, will block until array is quiessent
2551 * no resync can happen. no superblocks get written.
2552 * write requests fail
2554 * like readonly, but behaves like 'clean' on a write request.
2556 * clean - no pending writes, but otherwise active.
2557 * When written to inactive array, starts without resync
2558 * If a write request arrives then
2559 * if metadata is known, mark 'dirty' and switch to 'active'.
2560 * if not known, block and switch to write-pending
2561 * If written to an active array that has pending writes, then fails.
2563 * fully active: IO and resync can be happening.
2564 * When written to inactive array, starts with resync
2567 * clean, but writes are blocked waiting for 'active' to be written.
2570 * like active, but no writes have been seen for a while (100msec).
2573 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2574 write_pending
, active_idle
, bad_word
};
2575 static char *array_states
[] = {
2576 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2577 "write-pending", "active-idle", NULL
};
2579 static int match_word(const char *word
, char **list
)
2582 for (n
=0; list
[n
]; n
++)
2583 if (cmd_match(word
, list
[n
]))
2589 array_state_show(mddev_t
*mddev
, char *page
)
2591 enum array_state st
= inactive
;
2604 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2606 else if (mddev
->safemode
)
2612 if (list_empty(&mddev
->disks
) &&
2613 mddev
->raid_disks
== 0 &&
2619 return sprintf(page
, "%s\n", array_states
[st
]);
2622 static int do_md_stop(mddev_t
* mddev
, int ro
);
2623 static int do_md_run(mddev_t
* mddev
);
2624 static int restart_array(mddev_t
*mddev
);
2627 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2630 enum array_state st
= match_word(buf
, array_states
);
2635 /* stopping an active array */
2636 if (atomic_read(&mddev
->active
) > 1)
2638 err
= do_md_stop(mddev
, 0);
2641 /* stopping an active array */
2643 if (atomic_read(&mddev
->active
) > 1)
2645 err
= do_md_stop(mddev
, 2);
2647 err
= 0; /* already inactive */
2650 break; /* not supported yet */
2653 err
= do_md_stop(mddev
, 1);
2656 set_disk_ro(mddev
->gendisk
, 1);
2657 err
= do_md_run(mddev
);
2663 err
= do_md_stop(mddev
, 1);
2665 err
= restart_array(mddev
);
2668 set_disk_ro(mddev
->gendisk
, 0);
2672 err
= do_md_run(mddev
);
2677 restart_array(mddev
);
2678 spin_lock_irq(&mddev
->write_lock
);
2679 if (atomic_read(&mddev
->writes_pending
) == 0) {
2680 if (mddev
->in_sync
== 0) {
2682 if (mddev
->safemode
== 1)
2683 mddev
->safemode
= 0;
2684 if (mddev
->persistent
)
2685 set_bit(MD_CHANGE_CLEAN
,
2691 spin_unlock_irq(&mddev
->write_lock
);
2694 mddev
->recovery_cp
= MaxSector
;
2695 err
= do_md_run(mddev
);
2700 restart_array(mddev
);
2701 if (mddev
->external
)
2702 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2703 wake_up(&mddev
->sb_wait
);
2707 set_disk_ro(mddev
->gendisk
, 0);
2708 err
= do_md_run(mddev
);
2713 /* these cannot be set */
2719 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
2723 static struct md_sysfs_entry md_array_state
=
2724 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
2727 null_show(mddev_t
*mddev
, char *page
)
2733 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2735 /* buf must be %d:%d\n? giving major and minor numbers */
2736 /* The new device is added to the array.
2737 * If the array has a persistent superblock, we read the
2738 * superblock to initialise info and check validity.
2739 * Otherwise, only checking done is that in bind_rdev_to_array,
2740 * which mainly checks size.
2743 int major
= simple_strtoul(buf
, &e
, 10);
2749 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2751 minor
= simple_strtoul(e
+1, &e
, 10);
2752 if (*e
&& *e
!= '\n')
2754 dev
= MKDEV(major
, minor
);
2755 if (major
!= MAJOR(dev
) ||
2756 minor
!= MINOR(dev
))
2760 if (mddev
->persistent
) {
2761 rdev
= md_import_device(dev
, mddev
->major_version
,
2762 mddev
->minor_version
);
2763 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2764 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2765 mdk_rdev_t
, same_set
);
2766 err
= super_types
[mddev
->major_version
]
2767 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2771 } else if (mddev
->external
)
2772 rdev
= md_import_device(dev
, -2, -1);
2774 rdev
= md_import_device(dev
, -1, -1);
2777 return PTR_ERR(rdev
);
2778 err
= bind_rdev_to_array(rdev
, mddev
);
2782 return err
? err
: len
;
2785 static struct md_sysfs_entry md_new_device
=
2786 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
2789 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2792 unsigned long chunk
, end_chunk
;
2796 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2798 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
2799 if (buf
== end
) break;
2800 if (*end
== '-') { /* range */
2802 end_chunk
= simple_strtoul(buf
, &end
, 0);
2803 if (buf
== end
) break;
2805 if (*end
&& !isspace(*end
)) break;
2806 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
2808 while (isspace(*buf
)) buf
++;
2810 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
2815 static struct md_sysfs_entry md_bitmap
=
2816 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
2819 size_show(mddev_t
*mddev
, char *page
)
2821 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2824 static int update_size(mddev_t
*mddev
, unsigned long size
);
2827 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2829 /* If array is inactive, we can reduce the component size, but
2830 * not increase it (except from 0).
2831 * If array is active, we can try an on-line resize
2835 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2836 if (!*buf
|| *buf
== '\n' ||
2841 err
= update_size(mddev
, size
);
2842 md_update_sb(mddev
, 1);
2844 if (mddev
->size
== 0 ||
2850 return err
? err
: len
;
2853 static struct md_sysfs_entry md_size
=
2854 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
2859 * 'none' for arrays with no metadata (good luck...)
2860 * 'external' for arrays with externally managed metadata,
2861 * or N.M for internally known formats
2864 metadata_show(mddev_t
*mddev
, char *page
)
2866 if (mddev
->persistent
)
2867 return sprintf(page
, "%d.%d\n",
2868 mddev
->major_version
, mddev
->minor_version
);
2869 else if (mddev
->external
)
2870 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
2872 return sprintf(page
, "none\n");
2876 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2880 if (!list_empty(&mddev
->disks
))
2883 if (cmd_match(buf
, "none")) {
2884 mddev
->persistent
= 0;
2885 mddev
->external
= 0;
2886 mddev
->major_version
= 0;
2887 mddev
->minor_version
= 90;
2890 if (strncmp(buf
, "external:", 9) == 0) {
2891 size_t namelen
= len
-9;
2892 if (namelen
>= sizeof(mddev
->metadata_type
))
2893 namelen
= sizeof(mddev
->metadata_type
)-1;
2894 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
2895 mddev
->metadata_type
[namelen
] = 0;
2896 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
2897 mddev
->metadata_type
[--namelen
] = 0;
2898 mddev
->persistent
= 0;
2899 mddev
->external
= 1;
2900 mddev
->major_version
= 0;
2901 mddev
->minor_version
= 90;
2904 major
= simple_strtoul(buf
, &e
, 10);
2905 if (e
==buf
|| *e
!= '.')
2908 minor
= simple_strtoul(buf
, &e
, 10);
2909 if (e
==buf
|| (*e
&& *e
!= '\n') )
2911 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
2913 mddev
->major_version
= major
;
2914 mddev
->minor_version
= minor
;
2915 mddev
->persistent
= 1;
2916 mddev
->external
= 0;
2920 static struct md_sysfs_entry md_metadata
=
2921 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2924 action_show(mddev_t
*mddev
, char *page
)
2926 char *type
= "idle";
2927 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2928 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
2929 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2931 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2932 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2934 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2938 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
2941 return sprintf(page
, "%s\n", type
);
2945 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2947 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2950 if (cmd_match(page
, "idle")) {
2951 if (mddev
->sync_thread
) {
2952 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2953 md_unregister_thread(mddev
->sync_thread
);
2954 mddev
->sync_thread
= NULL
;
2955 mddev
->recovery
= 0;
2957 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2958 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2960 else if (cmd_match(page
, "resync"))
2961 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2962 else if (cmd_match(page
, "recover")) {
2963 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2964 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2965 } else if (cmd_match(page
, "reshape")) {
2967 if (mddev
->pers
->start_reshape
== NULL
)
2969 err
= mddev
->pers
->start_reshape(mddev
);
2973 if (cmd_match(page
, "check"))
2974 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2975 else if (!cmd_match(page
, "repair"))
2977 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2978 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2980 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2981 md_wakeup_thread(mddev
->thread
);
2982 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
2987 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2989 return sprintf(page
, "%llu\n",
2990 (unsigned long long) mddev
->resync_mismatches
);
2993 static struct md_sysfs_entry md_scan_mode
=
2994 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2997 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3000 sync_min_show(mddev_t
*mddev
, char *page
)
3002 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3003 mddev
->sync_speed_min
? "local": "system");
3007 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3011 if (strncmp(buf
, "system", 6)==0) {
3012 mddev
->sync_speed_min
= 0;
3015 min
= simple_strtoul(buf
, &e
, 10);
3016 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3018 mddev
->sync_speed_min
= min
;
3022 static struct md_sysfs_entry md_sync_min
=
3023 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3026 sync_max_show(mddev_t
*mddev
, char *page
)
3028 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3029 mddev
->sync_speed_max
? "local": "system");
3033 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3037 if (strncmp(buf
, "system", 6)==0) {
3038 mddev
->sync_speed_max
= 0;
3041 max
= simple_strtoul(buf
, &e
, 10);
3042 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3044 mddev
->sync_speed_max
= max
;
3048 static struct md_sysfs_entry md_sync_max
=
3049 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3052 degraded_show(mddev_t
*mddev
, char *page
)
3054 return sprintf(page
, "%d\n", mddev
->degraded
);
3056 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3059 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3061 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3065 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3069 if (strict_strtol(buf
, 10, &n
))
3072 if (n
!= 0 && n
!= 1)
3075 mddev
->parallel_resync
= n
;
3077 if (mddev
->sync_thread
)
3078 wake_up(&resync_wait
);
3083 /* force parallel resync, even with shared block devices */
3084 static struct md_sysfs_entry md_sync_force_parallel
=
3085 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3086 sync_force_parallel_show
, sync_force_parallel_store
);
3089 sync_speed_show(mddev_t
*mddev
, char *page
)
3091 unsigned long resync
, dt
, db
;
3092 resync
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
));
3093 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
3095 db
= resync
- (mddev
->resync_mark_cnt
);
3096 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
3099 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3102 sync_completed_show(mddev_t
*mddev
, char *page
)
3104 unsigned long max_blocks
, resync
;
3106 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3107 max_blocks
= mddev
->resync_max_sectors
;
3109 max_blocks
= mddev
->size
<< 1;
3111 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
3112 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
3115 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3118 min_sync_show(mddev_t
*mddev
, char *page
)
3120 return sprintf(page
, "%llu\n",
3121 (unsigned long long)mddev
->resync_min
);
3124 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3126 unsigned long long min
;
3127 if (strict_strtoull(buf
, 10, &min
))
3129 if (min
> mddev
->resync_max
)
3131 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3134 /* Must be a multiple of chunk_size */
3135 if (mddev
->chunk_size
) {
3136 if (min
& (sector_t
)((mddev
->chunk_size
>>9)-1))
3139 mddev
->resync_min
= min
;
3144 static struct md_sysfs_entry md_min_sync
=
3145 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3148 max_sync_show(mddev_t
*mddev
, char *page
)
3150 if (mddev
->resync_max
== MaxSector
)
3151 return sprintf(page
, "max\n");
3153 return sprintf(page
, "%llu\n",
3154 (unsigned long long)mddev
->resync_max
);
3157 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3159 if (strncmp(buf
, "max", 3) == 0)
3160 mddev
->resync_max
= MaxSector
;
3162 unsigned long long max
;
3163 if (strict_strtoull(buf
, 10, &max
))
3165 if (max
< mddev
->resync_min
)
3167 if (max
< mddev
->resync_max
&&
3168 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3171 /* Must be a multiple of chunk_size */
3172 if (mddev
->chunk_size
) {
3173 if (max
& (sector_t
)((mddev
->chunk_size
>>9)-1))
3176 mddev
->resync_max
= max
;
3178 wake_up(&mddev
->recovery_wait
);
3182 static struct md_sysfs_entry md_max_sync
=
3183 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3186 suspend_lo_show(mddev_t
*mddev
, char *page
)
3188 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3192 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3195 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3197 if (mddev
->pers
->quiesce
== NULL
)
3199 if (buf
== e
|| (*e
&& *e
!= '\n'))
3201 if (new >= mddev
->suspend_hi
||
3202 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3203 mddev
->suspend_lo
= new;
3204 mddev
->pers
->quiesce(mddev
, 2);
3209 static struct md_sysfs_entry md_suspend_lo
=
3210 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3214 suspend_hi_show(mddev_t
*mddev
, char *page
)
3216 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3220 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3223 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3225 if (mddev
->pers
->quiesce
== NULL
)
3227 if (buf
== e
|| (*e
&& *e
!= '\n'))
3229 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3230 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3231 mddev
->suspend_hi
= new;
3232 mddev
->pers
->quiesce(mddev
, 1);
3233 mddev
->pers
->quiesce(mddev
, 0);
3238 static struct md_sysfs_entry md_suspend_hi
=
3239 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3242 reshape_position_show(mddev_t
*mddev
, char *page
)
3244 if (mddev
->reshape_position
!= MaxSector
)
3245 return sprintf(page
, "%llu\n",
3246 (unsigned long long)mddev
->reshape_position
);
3247 strcpy(page
, "none\n");
3252 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3255 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3258 if (buf
== e
|| (*e
&& *e
!= '\n'))
3260 mddev
->reshape_position
= new;
3261 mddev
->delta_disks
= 0;
3262 mddev
->new_level
= mddev
->level
;
3263 mddev
->new_layout
= mddev
->layout
;
3264 mddev
->new_chunk
= mddev
->chunk_size
;
3268 static struct md_sysfs_entry md_reshape_position
=
3269 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3270 reshape_position_store
);
3273 static struct attribute
*md_default_attrs
[] = {
3276 &md_raid_disks
.attr
,
3277 &md_chunk_size
.attr
,
3279 &md_resync_start
.attr
,
3281 &md_new_device
.attr
,
3282 &md_safe_delay
.attr
,
3283 &md_array_state
.attr
,
3284 &md_reshape_position
.attr
,
3288 static struct attribute
*md_redundancy_attrs
[] = {
3290 &md_mismatches
.attr
,
3293 &md_sync_speed
.attr
,
3294 &md_sync_force_parallel
.attr
,
3295 &md_sync_completed
.attr
,
3298 &md_suspend_lo
.attr
,
3299 &md_suspend_hi
.attr
,
3304 static struct attribute_group md_redundancy_group
= {
3306 .attrs
= md_redundancy_attrs
,
3311 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3313 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3314 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3319 rv
= mddev_lock(mddev
);
3321 rv
= entry
->show(mddev
, page
);
3322 mddev_unlock(mddev
);
3328 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3329 const char *page
, size_t length
)
3331 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3332 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3337 if (!capable(CAP_SYS_ADMIN
))
3339 rv
= mddev_lock(mddev
);
3341 rv
= entry
->store(mddev
, page
, length
);
3342 mddev_unlock(mddev
);
3347 static void md_free(struct kobject
*ko
)
3349 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
3353 static struct sysfs_ops md_sysfs_ops
= {
3354 .show
= md_attr_show
,
3355 .store
= md_attr_store
,
3357 static struct kobj_type md_ktype
= {
3359 .sysfs_ops
= &md_sysfs_ops
,
3360 .default_attrs
= md_default_attrs
,
3365 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
3367 static DEFINE_MUTEX(disks_mutex
);
3368 mddev_t
*mddev
= mddev_find(dev
);
3369 struct gendisk
*disk
;
3370 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
3371 int shift
= partitioned
? MdpMinorShift
: 0;
3372 int unit
= MINOR(dev
) >> shift
;
3378 mutex_lock(&disks_mutex
);
3379 if (mddev
->gendisk
) {
3380 mutex_unlock(&disks_mutex
);
3384 disk
= alloc_disk(1 << shift
);
3386 mutex_unlock(&disks_mutex
);
3390 disk
->major
= MAJOR(dev
);
3391 disk
->first_minor
= unit
<< shift
;
3393 sprintf(disk
->disk_name
, "md_d%d", unit
);
3395 sprintf(disk
->disk_name
, "md%d", unit
);
3396 disk
->fops
= &md_fops
;
3397 disk
->private_data
= mddev
;
3398 disk
->queue
= mddev
->queue
;
3400 mddev
->gendisk
= disk
;
3401 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
, &disk
->dev
.kobj
,
3403 mutex_unlock(&disks_mutex
);
3405 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
3408 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
3412 static void md_safemode_timeout(unsigned long data
)
3414 mddev_t
*mddev
= (mddev_t
*) data
;
3416 if (!atomic_read(&mddev
->writes_pending
)) {
3417 mddev
->safemode
= 1;
3418 if (mddev
->external
)
3419 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
3421 md_wakeup_thread(mddev
->thread
);
3424 static int start_dirty_degraded
;
3426 static int do_md_run(mddev_t
* mddev
)
3430 struct list_head
*tmp
;
3432 struct gendisk
*disk
;
3433 struct mdk_personality
*pers
;
3434 char b
[BDEVNAME_SIZE
];
3436 if (list_empty(&mddev
->disks
))
3437 /* cannot run an array with no devices.. */
3444 * Analyze all RAID superblock(s)
3446 if (!mddev
->raid_disks
) {
3447 if (!mddev
->persistent
)
3452 chunk_size
= mddev
->chunk_size
;
3455 if (chunk_size
> MAX_CHUNK_SIZE
) {
3456 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
3457 chunk_size
, MAX_CHUNK_SIZE
);
3461 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3463 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
3464 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
3467 if (chunk_size
< PAGE_SIZE
) {
3468 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
3469 chunk_size
, PAGE_SIZE
);
3473 /* devices must have minimum size of one chunk */
3474 rdev_for_each(rdev
, tmp
, mddev
) {
3475 if (test_bit(Faulty
, &rdev
->flags
))
3477 if (rdev
->size
< chunk_size
/ 1024) {
3479 "md: Dev %s smaller than chunk_size:"
3481 bdevname(rdev
->bdev
,b
),
3482 (unsigned long long)rdev
->size
,
3490 if (mddev
->level
!= LEVEL_NONE
)
3491 request_module("md-level-%d", mddev
->level
);
3492 else if (mddev
->clevel
[0])
3493 request_module("md-%s", mddev
->clevel
);
3497 * Drop all container device buffers, from now on
3498 * the only valid external interface is through the md
3501 rdev_for_each(rdev
, tmp
, mddev
) {
3502 if (test_bit(Faulty
, &rdev
->flags
))
3504 sync_blockdev(rdev
->bdev
);
3505 invalidate_bdev(rdev
->bdev
);
3507 /* perform some consistency tests on the device.
3508 * We don't want the data to overlap the metadata,
3509 * Internal Bitmap issues has handled elsewhere.
3511 if (rdev
->data_offset
< rdev
->sb_offset
) {
3513 rdev
->data_offset
+ mddev
->size
*2
3514 > rdev
->sb_offset
*2) {
3515 printk("md: %s: data overlaps metadata\n",
3520 if (rdev
->sb_offset
*2 + rdev
->sb_size
/512
3521 > rdev
->data_offset
) {
3522 printk("md: %s: metadata overlaps data\n",
3529 md_probe(mddev
->unit
, NULL
, NULL
);
3530 disk
= mddev
->gendisk
;
3534 spin_lock(&pers_lock
);
3535 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3536 if (!pers
|| !try_module_get(pers
->owner
)) {
3537 spin_unlock(&pers_lock
);
3538 if (mddev
->level
!= LEVEL_NONE
)
3539 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3542 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3547 spin_unlock(&pers_lock
);
3548 mddev
->level
= pers
->level
;
3549 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3551 if (mddev
->reshape_position
!= MaxSector
&&
3552 pers
->start_reshape
== NULL
) {
3553 /* This personality cannot handle reshaping... */
3555 module_put(pers
->owner
);
3559 if (pers
->sync_request
) {
3560 /* Warn if this is a potentially silly
3563 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3565 struct list_head
*tmp2
;
3567 rdev_for_each(rdev
, tmp
, mddev
) {
3568 rdev_for_each(rdev2
, tmp2
, mddev
) {
3570 rdev
->bdev
->bd_contains
==
3571 rdev2
->bdev
->bd_contains
) {
3573 "%s: WARNING: %s appears to be"
3574 " on the same physical disk as"
3577 bdevname(rdev
->bdev
,b
),
3578 bdevname(rdev2
->bdev
,b2
));
3585 "True protection against single-disk"
3586 " failure might be compromised.\n");
3589 mddev
->recovery
= 0;
3590 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3591 mddev
->barriers_work
= 1;
3592 mddev
->ok_start_degraded
= start_dirty_degraded
;
3595 mddev
->ro
= 2; /* read-only, but switch on first write */
3597 err
= mddev
->pers
->run(mddev
);
3598 if (!err
&& mddev
->pers
->sync_request
) {
3599 err
= bitmap_create(mddev
);
3601 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3602 mdname(mddev
), err
);
3603 mddev
->pers
->stop(mddev
);
3607 printk(KERN_ERR
"md: pers->run() failed ...\n");
3608 module_put(mddev
->pers
->owner
);
3610 bitmap_destroy(mddev
);
3613 if (mddev
->pers
->sync_request
) {
3614 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3616 "md: cannot register extra attributes for %s\n",
3618 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3621 atomic_set(&mddev
->writes_pending
,0);
3622 mddev
->safemode
= 0;
3623 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3624 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3625 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3628 rdev_for_each(rdev
, tmp
, mddev
)
3629 if (rdev
->raid_disk
>= 0) {
3631 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3632 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
3633 printk("md: cannot register %s for %s\n",
3637 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3640 md_update_sb(mddev
, 0);
3642 set_capacity(disk
, mddev
->array_size
<<1);
3644 /* If we call blk_queue_make_request here, it will
3645 * re-initialise max_sectors etc which may have been
3646 * refined inside -> run. So just set the bits we need to set.
3647 * Most initialisation happended when we called
3648 * blk_queue_make_request(..., md_fail_request)
3651 mddev
->queue
->queuedata
= mddev
;
3652 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3654 /* If there is a partially-recovered drive we need to
3655 * start recovery here. If we leave it to md_check_recovery,
3656 * it will remove the drives and not do the right thing
3658 if (mddev
->degraded
&& !mddev
->sync_thread
) {
3659 struct list_head
*rtmp
;
3661 rdev_for_each(rdev
, rtmp
, mddev
)
3662 if (rdev
->raid_disk
>= 0 &&
3663 !test_bit(In_sync
, &rdev
->flags
) &&
3664 !test_bit(Faulty
, &rdev
->flags
))
3665 /* complete an interrupted recovery */
3667 if (spares
&& mddev
->pers
->sync_request
) {
3668 mddev
->recovery
= 0;
3669 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3670 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3673 if (!mddev
->sync_thread
) {
3674 printk(KERN_ERR
"%s: could not start resync"
3677 /* leave the spares where they are, it shouldn't hurt */
3678 mddev
->recovery
= 0;
3682 md_wakeup_thread(mddev
->thread
);
3683 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
3686 md_new_event(mddev
);
3687 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
3688 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
3689 kobject_uevent(&mddev
->gendisk
->dev
.kobj
, KOBJ_CHANGE
);
3693 static int restart_array(mddev_t
*mddev
)
3695 struct gendisk
*disk
= mddev
->gendisk
;
3699 * Complain if it has no devices
3702 if (list_empty(&mddev
->disks
))
3710 mddev
->safemode
= 0;
3712 set_disk_ro(disk
, 0);
3714 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3717 * Kick recovery or resync if necessary
3719 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3720 md_wakeup_thread(mddev
->thread
);
3721 md_wakeup_thread(mddev
->sync_thread
);
3723 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
3732 /* similar to deny_write_access, but accounts for our holding a reference
3733 * to the file ourselves */
3734 static int deny_bitmap_write_access(struct file
* file
)
3736 struct inode
*inode
= file
->f_mapping
->host
;
3738 spin_lock(&inode
->i_lock
);
3739 if (atomic_read(&inode
->i_writecount
) > 1) {
3740 spin_unlock(&inode
->i_lock
);
3743 atomic_set(&inode
->i_writecount
, -1);
3744 spin_unlock(&inode
->i_lock
);
3749 static void restore_bitmap_write_access(struct file
*file
)
3751 struct inode
*inode
= file
->f_mapping
->host
;
3753 spin_lock(&inode
->i_lock
);
3754 atomic_set(&inode
->i_writecount
, 1);
3755 spin_unlock(&inode
->i_lock
);
3759 * 0 - completely stop and dis-assemble array
3760 * 1 - switch to readonly
3761 * 2 - stop but do not disassemble array
3763 static int do_md_stop(mddev_t
* mddev
, int mode
)
3766 struct gendisk
*disk
= mddev
->gendisk
;
3769 if (atomic_read(&mddev
->active
)>2) {
3770 printk("md: %s still in use.\n",mdname(mddev
));
3774 if (mddev
->sync_thread
) {
3775 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3776 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3777 md_unregister_thread(mddev
->sync_thread
);
3778 mddev
->sync_thread
= NULL
;
3781 del_timer_sync(&mddev
->safemode_timer
);
3783 invalidate_partition(disk
, 0);
3786 case 1: /* readonly */
3792 case 0: /* disassemble */
3794 bitmap_flush(mddev
);
3795 md_super_wait(mddev
);
3797 set_disk_ro(disk
, 0);
3798 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3799 mddev
->pers
->stop(mddev
);
3800 mddev
->queue
->merge_bvec_fn
= NULL
;
3801 mddev
->queue
->unplug_fn
= NULL
;
3802 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
3803 if (mddev
->pers
->sync_request
)
3804 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3806 module_put(mddev
->pers
->owner
);
3808 /* tell userspace to handle 'inactive' */
3809 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
3811 set_capacity(disk
, 0);
3817 if (!mddev
->in_sync
|| mddev
->flags
) {
3818 /* mark array as shutdown cleanly */
3820 md_update_sb(mddev
, 1);
3823 set_disk_ro(disk
, 1);
3824 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3828 * Free resources if final stop
3832 struct list_head
*tmp
;
3834 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3836 bitmap_destroy(mddev
);
3837 if (mddev
->bitmap_file
) {
3838 restore_bitmap_write_access(mddev
->bitmap_file
);
3839 fput(mddev
->bitmap_file
);
3840 mddev
->bitmap_file
= NULL
;
3842 mddev
->bitmap_offset
= 0;
3844 rdev_for_each(rdev
, tmp
, mddev
)
3845 if (rdev
->raid_disk
>= 0) {
3847 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3848 sysfs_remove_link(&mddev
->kobj
, nm
);
3851 /* make sure all md_delayed_delete calls have finished */
3852 flush_scheduled_work();
3854 export_array(mddev
);
3856 mddev
->array_size
= 0;
3858 mddev
->raid_disks
= 0;
3859 mddev
->recovery_cp
= 0;
3860 mddev
->resync_min
= 0;
3861 mddev
->resync_max
= MaxSector
;
3862 mddev
->reshape_position
= MaxSector
;
3863 mddev
->external
= 0;
3864 mddev
->persistent
= 0;
3865 mddev
->level
= LEVEL_NONE
;
3866 mddev
->clevel
[0] = 0;
3869 mddev
->metadata_type
[0] = 0;
3870 mddev
->chunk_size
= 0;
3871 mddev
->ctime
= mddev
->utime
= 0;
3873 mddev
->max_disks
= 0;
3875 mddev
->delta_disks
= 0;
3876 mddev
->new_level
= LEVEL_NONE
;
3877 mddev
->new_layout
= 0;
3878 mddev
->new_chunk
= 0;
3879 mddev
->curr_resync
= 0;
3880 mddev
->resync_mismatches
= 0;
3881 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
3882 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
3883 mddev
->recovery
= 0;
3886 mddev
->degraded
= 0;
3887 mddev
->barriers_work
= 0;
3888 mddev
->safemode
= 0;
3890 } else if (mddev
->pers
)
3891 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3894 md_new_event(mddev
);
3895 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
3901 static void autorun_array(mddev_t
*mddev
)
3904 struct list_head
*tmp
;
3907 if (list_empty(&mddev
->disks
))
3910 printk(KERN_INFO
"md: running: ");
3912 rdev_for_each(rdev
, tmp
, mddev
) {
3913 char b
[BDEVNAME_SIZE
];
3914 printk("<%s>", bdevname(rdev
->bdev
,b
));
3918 err
= do_md_run (mddev
);
3920 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3921 do_md_stop (mddev
, 0);
3926 * lets try to run arrays based on all disks that have arrived
3927 * until now. (those are in pending_raid_disks)
3929 * the method: pick the first pending disk, collect all disks with
3930 * the same UUID, remove all from the pending list and put them into
3931 * the 'same_array' list. Then order this list based on superblock
3932 * update time (freshest comes first), kick out 'old' disks and
3933 * compare superblocks. If everything's fine then run it.
3935 * If "unit" is allocated, then bump its reference count
3937 static void autorun_devices(int part
)
3939 struct list_head
*tmp
;
3940 mdk_rdev_t
*rdev0
, *rdev
;
3942 char b
[BDEVNAME_SIZE
];
3944 printk(KERN_INFO
"md: autorun ...\n");
3945 while (!list_empty(&pending_raid_disks
)) {
3948 LIST_HEAD(candidates
);
3949 rdev0
= list_entry(pending_raid_disks
.next
,
3950 mdk_rdev_t
, same_set
);
3952 printk(KERN_INFO
"md: considering %s ...\n",
3953 bdevname(rdev0
->bdev
,b
));
3954 INIT_LIST_HEAD(&candidates
);
3955 rdev_for_each_list(rdev
, tmp
, pending_raid_disks
)
3956 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3957 printk(KERN_INFO
"md: adding %s ...\n",
3958 bdevname(rdev
->bdev
,b
));
3959 list_move(&rdev
->same_set
, &candidates
);
3962 * now we have a set of devices, with all of them having
3963 * mostly sane superblocks. It's time to allocate the
3967 dev
= MKDEV(mdp_major
,
3968 rdev0
->preferred_minor
<< MdpMinorShift
);
3969 unit
= MINOR(dev
) >> MdpMinorShift
;
3971 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3974 if (rdev0
->preferred_minor
!= unit
) {
3975 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3976 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3980 md_probe(dev
, NULL
, NULL
);
3981 mddev
= mddev_find(dev
);
3982 if (!mddev
|| !mddev
->gendisk
) {
3986 "md: cannot allocate memory for md drive.\n");
3989 if (mddev_lock(mddev
))
3990 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3992 else if (mddev
->raid_disks
|| mddev
->major_version
3993 || !list_empty(&mddev
->disks
)) {
3995 "md: %s already running, cannot run %s\n",
3996 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3997 mddev_unlock(mddev
);
3999 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4000 mddev
->persistent
= 1;
4001 rdev_for_each_list(rdev
, tmp
, candidates
) {
4002 list_del_init(&rdev
->same_set
);
4003 if (bind_rdev_to_array(rdev
, mddev
))
4006 autorun_array(mddev
);
4007 mddev_unlock(mddev
);
4009 /* on success, candidates will be empty, on error
4012 rdev_for_each_list(rdev
, tmp
, candidates
)
4016 printk(KERN_INFO
"md: ... autorun DONE.\n");
4018 #endif /* !MODULE */
4020 static int get_version(void __user
* arg
)
4024 ver
.major
= MD_MAJOR_VERSION
;
4025 ver
.minor
= MD_MINOR_VERSION
;
4026 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4028 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4034 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4036 mdu_array_info_t info
;
4037 int nr
,working
,active
,failed
,spare
;
4039 struct list_head
*tmp
;
4041 nr
=working
=active
=failed
=spare
=0;
4042 rdev_for_each(rdev
, tmp
, mddev
) {
4044 if (test_bit(Faulty
, &rdev
->flags
))
4048 if (test_bit(In_sync
, &rdev
->flags
))
4055 info
.major_version
= mddev
->major_version
;
4056 info
.minor_version
= mddev
->minor_version
;
4057 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4058 info
.ctime
= mddev
->ctime
;
4059 info
.level
= mddev
->level
;
4060 info
.size
= mddev
->size
;
4061 if (info
.size
!= mddev
->size
) /* overflow */
4064 info
.raid_disks
= mddev
->raid_disks
;
4065 info
.md_minor
= mddev
->md_minor
;
4066 info
.not_persistent
= !mddev
->persistent
;
4068 info
.utime
= mddev
->utime
;
4071 info
.state
= (1<<MD_SB_CLEAN
);
4072 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4073 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4074 info
.active_disks
= active
;
4075 info
.working_disks
= working
;
4076 info
.failed_disks
= failed
;
4077 info
.spare_disks
= spare
;
4079 info
.layout
= mddev
->layout
;
4080 info
.chunk_size
= mddev
->chunk_size
;
4082 if (copy_to_user(arg
, &info
, sizeof(info
)))
4088 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4090 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4091 char *ptr
, *buf
= NULL
;
4094 md_allow_write(mddev
);
4096 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4100 /* bitmap disabled, zero the first byte and copy out */
4101 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4102 file
->pathname
[0] = '\0';
4106 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4110 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4114 strcpy(file
->pathname
, ptr
);
4118 if (copy_to_user(arg
, file
, sizeof(*file
)))
4126 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4128 mdu_disk_info_t info
;
4132 if (copy_from_user(&info
, arg
, sizeof(info
)))
4137 rdev
= find_rdev_nr(mddev
, nr
);
4139 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4140 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4141 info
.raid_disk
= rdev
->raid_disk
;
4143 if (test_bit(Faulty
, &rdev
->flags
))
4144 info
.state
|= (1<<MD_DISK_FAULTY
);
4145 else if (test_bit(In_sync
, &rdev
->flags
)) {
4146 info
.state
|= (1<<MD_DISK_ACTIVE
);
4147 info
.state
|= (1<<MD_DISK_SYNC
);
4149 if (test_bit(WriteMostly
, &rdev
->flags
))
4150 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4152 info
.major
= info
.minor
= 0;
4153 info
.raid_disk
= -1;
4154 info
.state
= (1<<MD_DISK_REMOVED
);
4157 if (copy_to_user(arg
, &info
, sizeof(info
)))
4163 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4165 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4167 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4169 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4172 if (!mddev
->raid_disks
) {
4174 /* expecting a device which has a superblock */
4175 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4178 "md: md_import_device returned %ld\n",
4180 return PTR_ERR(rdev
);
4182 if (!list_empty(&mddev
->disks
)) {
4183 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4184 mdk_rdev_t
, same_set
);
4185 int err
= super_types
[mddev
->major_version
]
4186 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4189 "md: %s has different UUID to %s\n",
4190 bdevname(rdev
->bdev
,b
),
4191 bdevname(rdev0
->bdev
,b2
));
4196 err
= bind_rdev_to_array(rdev
, mddev
);
4203 * add_new_disk can be used once the array is assembled
4204 * to add "hot spares". They must already have a superblock
4209 if (!mddev
->pers
->hot_add_disk
) {
4211 "%s: personality does not support diskops!\n",
4215 if (mddev
->persistent
)
4216 rdev
= md_import_device(dev
, mddev
->major_version
,
4217 mddev
->minor_version
);
4219 rdev
= md_import_device(dev
, -1, -1);
4222 "md: md_import_device returned %ld\n",
4224 return PTR_ERR(rdev
);
4226 /* set save_raid_disk if appropriate */
4227 if (!mddev
->persistent
) {
4228 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4229 info
->raid_disk
< mddev
->raid_disks
)
4230 rdev
->raid_disk
= info
->raid_disk
;
4232 rdev
->raid_disk
= -1;
4234 super_types
[mddev
->major_version
].
4235 validate_super(mddev
, rdev
);
4236 rdev
->saved_raid_disk
= rdev
->raid_disk
;
4238 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
4239 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4240 set_bit(WriteMostly
, &rdev
->flags
);
4242 rdev
->raid_disk
= -1;
4243 err
= bind_rdev_to_array(rdev
, mddev
);
4244 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
4245 /* If there is hot_add_disk but no hot_remove_disk
4246 * then added disks for geometry changes,
4247 * and should be added immediately.
4249 super_types
[mddev
->major_version
].
4250 validate_super(mddev
, rdev
);
4251 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
4253 unbind_rdev_from_array(rdev
);
4258 md_update_sb(mddev
, 1);
4259 if (mddev
->degraded
)
4260 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4261 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4262 md_wakeup_thread(mddev
->thread
);
4266 /* otherwise, add_new_disk is only allowed
4267 * for major_version==0 superblocks
4269 if (mddev
->major_version
!= 0) {
4270 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
4275 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
4277 rdev
= md_import_device (dev
, -1, 0);
4280 "md: error, md_import_device() returned %ld\n",
4282 return PTR_ERR(rdev
);
4284 rdev
->desc_nr
= info
->number
;
4285 if (info
->raid_disk
< mddev
->raid_disks
)
4286 rdev
->raid_disk
= info
->raid_disk
;
4288 rdev
->raid_disk
= -1;
4290 if (rdev
->raid_disk
< mddev
->raid_disks
)
4291 if (info
->state
& (1<<MD_DISK_SYNC
))
4292 set_bit(In_sync
, &rdev
->flags
);
4294 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4295 set_bit(WriteMostly
, &rdev
->flags
);
4297 if (!mddev
->persistent
) {
4298 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
4299 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
4301 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
4302 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
4304 err
= bind_rdev_to_array(rdev
, mddev
);
4314 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
4316 char b
[BDEVNAME_SIZE
];
4319 rdev
= find_rdev(mddev
, dev
);
4323 if (rdev
->raid_disk
>= 0)
4326 kick_rdev_from_array(rdev
);
4327 md_update_sb(mddev
, 1);
4328 md_new_event(mddev
);
4332 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
4333 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4337 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
4339 char b
[BDEVNAME_SIZE
];
4347 if (mddev
->major_version
!= 0) {
4348 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
4349 " version-0 superblocks.\n",
4353 if (!mddev
->pers
->hot_add_disk
) {
4355 "%s: personality does not support diskops!\n",
4360 rdev
= md_import_device (dev
, -1, 0);
4363 "md: error, md_import_device() returned %ld\n",
4368 if (mddev
->persistent
)
4369 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
4372 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
4374 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
4377 if (test_bit(Faulty
, &rdev
->flags
)) {
4379 "md: can not hot-add faulty %s disk to %s!\n",
4380 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4384 clear_bit(In_sync
, &rdev
->flags
);
4386 rdev
->saved_raid_disk
= -1;
4387 err
= bind_rdev_to_array(rdev
, mddev
);
4392 * The rest should better be atomic, we can have disk failures
4393 * noticed in interrupt contexts ...
4396 if (rdev
->desc_nr
== mddev
->max_disks
) {
4397 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
4400 goto abort_unbind_export
;
4403 rdev
->raid_disk
= -1;
4405 md_update_sb(mddev
, 1);
4408 * Kick recovery, maybe this spare has to be added to the
4409 * array immediately.
4411 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4412 md_wakeup_thread(mddev
->thread
);
4413 md_new_event(mddev
);
4416 abort_unbind_export
:
4417 unbind_rdev_from_array(rdev
);
4424 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
4429 if (!mddev
->pers
->quiesce
)
4431 if (mddev
->recovery
|| mddev
->sync_thread
)
4433 /* we should be able to change the bitmap.. */
4439 return -EEXIST
; /* cannot add when bitmap is present */
4440 mddev
->bitmap_file
= fget(fd
);
4442 if (mddev
->bitmap_file
== NULL
) {
4443 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
4448 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
4450 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
4452 fput(mddev
->bitmap_file
);
4453 mddev
->bitmap_file
= NULL
;
4456 mddev
->bitmap_offset
= 0; /* file overrides offset */
4457 } else if (mddev
->bitmap
== NULL
)
4458 return -ENOENT
; /* cannot remove what isn't there */
4461 mddev
->pers
->quiesce(mddev
, 1);
4463 err
= bitmap_create(mddev
);
4464 if (fd
< 0 || err
) {
4465 bitmap_destroy(mddev
);
4466 fd
= -1; /* make sure to put the file */
4468 mddev
->pers
->quiesce(mddev
, 0);
4471 if (mddev
->bitmap_file
) {
4472 restore_bitmap_write_access(mddev
->bitmap_file
);
4473 fput(mddev
->bitmap_file
);
4475 mddev
->bitmap_file
= NULL
;
4482 * set_array_info is used two different ways
4483 * The original usage is when creating a new array.
4484 * In this usage, raid_disks is > 0 and it together with
4485 * level, size, not_persistent,layout,chunksize determine the
4486 * shape of the array.
4487 * This will always create an array with a type-0.90.0 superblock.
4488 * The newer usage is when assembling an array.
4489 * In this case raid_disks will be 0, and the major_version field is
4490 * use to determine which style super-blocks are to be found on the devices.
4491 * The minor and patch _version numbers are also kept incase the
4492 * super_block handler wishes to interpret them.
4494 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
4497 if (info
->raid_disks
== 0) {
4498 /* just setting version number for superblock loading */
4499 if (info
->major_version
< 0 ||
4500 info
->major_version
>= ARRAY_SIZE(super_types
) ||
4501 super_types
[info
->major_version
].name
== NULL
) {
4502 /* maybe try to auto-load a module? */
4504 "md: superblock version %d not known\n",
4505 info
->major_version
);
4508 mddev
->major_version
= info
->major_version
;
4509 mddev
->minor_version
= info
->minor_version
;
4510 mddev
->patch_version
= info
->patch_version
;
4511 mddev
->persistent
= !info
->not_persistent
;
4514 mddev
->major_version
= MD_MAJOR_VERSION
;
4515 mddev
->minor_version
= MD_MINOR_VERSION
;
4516 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
4517 mddev
->ctime
= get_seconds();
4519 mddev
->level
= info
->level
;
4520 mddev
->clevel
[0] = 0;
4521 mddev
->size
= info
->size
;
4522 mddev
->raid_disks
= info
->raid_disks
;
4523 /* don't set md_minor, it is determined by which /dev/md* was
4526 if (info
->state
& (1<<MD_SB_CLEAN
))
4527 mddev
->recovery_cp
= MaxSector
;
4529 mddev
->recovery_cp
= 0;
4530 mddev
->persistent
= ! info
->not_persistent
;
4531 mddev
->external
= 0;
4533 mddev
->layout
= info
->layout
;
4534 mddev
->chunk_size
= info
->chunk_size
;
4536 mddev
->max_disks
= MD_SB_DISKS
;
4538 if (mddev
->persistent
)
4540 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4542 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4543 mddev
->bitmap_offset
= 0;
4545 mddev
->reshape_position
= MaxSector
;
4548 * Generate a 128 bit UUID
4550 get_random_bytes(mddev
->uuid
, 16);
4552 mddev
->new_level
= mddev
->level
;
4553 mddev
->new_chunk
= mddev
->chunk_size
;
4554 mddev
->new_layout
= mddev
->layout
;
4555 mddev
->delta_disks
= 0;
4560 static int update_size(mddev_t
*mddev
, unsigned long size
)
4564 struct list_head
*tmp
;
4565 int fit
= (size
== 0);
4567 if (mddev
->pers
->resize
== NULL
)
4569 /* The "size" is the amount of each device that is used.
4570 * This can only make sense for arrays with redundancy.
4571 * linear and raid0 always use whatever space is available
4572 * We can only consider changing the size if no resync
4573 * or reconstruction is happening, and if the new size
4574 * is acceptable. It must fit before the sb_offset or,
4575 * if that is <data_offset, it must fit before the
4576 * size of each device.
4577 * If size is zero, we find the largest size that fits.
4579 if (mddev
->sync_thread
)
4581 rdev_for_each(rdev
, tmp
, mddev
) {
4583 avail
= rdev
->size
* 2;
4585 if (fit
&& (size
== 0 || size
> avail
/2))
4587 if (avail
< ((sector_t
)size
<< 1))
4590 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
4592 struct block_device
*bdev
;
4594 bdev
= bdget_disk(mddev
->gendisk
, 0);
4596 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4597 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
4598 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4605 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4608 /* change the number of raid disks */
4609 if (mddev
->pers
->check_reshape
== NULL
)
4611 if (raid_disks
<= 0 ||
4612 raid_disks
>= mddev
->max_disks
)
4614 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4616 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4618 rv
= mddev
->pers
->check_reshape(mddev
);
4624 * update_array_info is used to change the configuration of an
4626 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4627 * fields in the info are checked against the array.
4628 * Any differences that cannot be handled will cause an error.
4629 * Normally, only one change can be managed at a time.
4631 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4637 /* calculate expected state,ignoring low bits */
4638 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4639 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4641 if (mddev
->major_version
!= info
->major_version
||
4642 mddev
->minor_version
!= info
->minor_version
||
4643 /* mddev->patch_version != info->patch_version || */
4644 mddev
->ctime
!= info
->ctime
||
4645 mddev
->level
!= info
->level
||
4646 /* mddev->layout != info->layout || */
4647 !mddev
->persistent
!= info
->not_persistent
||
4648 mddev
->chunk_size
!= info
->chunk_size
||
4649 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4650 ((state
^info
->state
) & 0xfffffe00)
4653 /* Check there is only one change */
4654 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4655 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4656 if (mddev
->layout
!= info
->layout
) cnt
++;
4657 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4658 if (cnt
== 0) return 0;
4659 if (cnt
> 1) return -EINVAL
;
4661 if (mddev
->layout
!= info
->layout
) {
4663 * we don't need to do anything at the md level, the
4664 * personality will take care of it all.
4666 if (mddev
->pers
->reconfig
== NULL
)
4669 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4671 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4672 rv
= update_size(mddev
, info
->size
);
4674 if (mddev
->raid_disks
!= info
->raid_disks
)
4675 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4677 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4678 if (mddev
->pers
->quiesce
== NULL
)
4680 if (mddev
->recovery
|| mddev
->sync_thread
)
4682 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4683 /* add the bitmap */
4686 if (mddev
->default_bitmap_offset
== 0)
4688 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4689 mddev
->pers
->quiesce(mddev
, 1);
4690 rv
= bitmap_create(mddev
);
4692 bitmap_destroy(mddev
);
4693 mddev
->pers
->quiesce(mddev
, 0);
4695 /* remove the bitmap */
4698 if (mddev
->bitmap
->file
)
4700 mddev
->pers
->quiesce(mddev
, 1);
4701 bitmap_destroy(mddev
);
4702 mddev
->pers
->quiesce(mddev
, 0);
4703 mddev
->bitmap_offset
= 0;
4706 md_update_sb(mddev
, 1);
4710 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4714 if (mddev
->pers
== NULL
)
4717 rdev
= find_rdev(mddev
, dev
);
4721 md_error(mddev
, rdev
);
4725 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4727 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4731 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4735 static int md_ioctl(struct inode
*inode
, struct file
*file
,
4736 unsigned int cmd
, unsigned long arg
)
4739 void __user
*argp
= (void __user
*)arg
;
4740 mddev_t
*mddev
= NULL
;
4742 if (!capable(CAP_SYS_ADMIN
))
4746 * Commands dealing with the RAID driver but not any
4752 err
= get_version(argp
);
4755 case PRINT_RAID_DEBUG
:
4763 autostart_arrays(arg
);
4770 * Commands creating/starting a new array:
4773 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4780 err
= mddev_lock(mddev
);
4783 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4790 case SET_ARRAY_INFO
:
4792 mdu_array_info_t info
;
4794 memset(&info
, 0, sizeof(info
));
4795 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4800 err
= update_array_info(mddev
, &info
);
4802 printk(KERN_WARNING
"md: couldn't update"
4803 " array info. %d\n", err
);
4808 if (!list_empty(&mddev
->disks
)) {
4810 "md: array %s already has disks!\n",
4815 if (mddev
->raid_disks
) {
4817 "md: array %s already initialised!\n",
4822 err
= set_array_info(mddev
, &info
);
4824 printk(KERN_WARNING
"md: couldn't set"
4825 " array info. %d\n", err
);
4835 * Commands querying/configuring an existing array:
4837 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4838 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4839 if ((!mddev
->raid_disks
&& !mddev
->external
)
4840 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4841 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
4842 && cmd
!= GET_BITMAP_FILE
) {
4848 * Commands even a read-only array can execute:
4852 case GET_ARRAY_INFO
:
4853 err
= get_array_info(mddev
, argp
);
4856 case GET_BITMAP_FILE
:
4857 err
= get_bitmap_file(mddev
, argp
);
4861 err
= get_disk_info(mddev
, argp
);
4864 case RESTART_ARRAY_RW
:
4865 err
= restart_array(mddev
);
4869 err
= do_md_stop (mddev
, 0);
4873 err
= do_md_stop (mddev
, 1);
4877 * We have a problem here : there is no easy way to give a CHS
4878 * virtual geometry. We currently pretend that we have a 2 heads
4879 * 4 sectors (with a BIG number of cylinders...). This drives
4880 * dosfs just mad... ;-)
4885 * The remaining ioctls are changing the state of the
4886 * superblock, so we do not allow them on read-only arrays.
4887 * However non-MD ioctls (e.g. get-size) will still come through
4888 * here and hit the 'default' below, so only disallow
4889 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4891 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4892 mddev
->ro
&& mddev
->pers
) {
4893 if (mddev
->ro
== 2) {
4895 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
4896 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4897 md_wakeup_thread(mddev
->thread
);
4909 mdu_disk_info_t info
;
4910 if (copy_from_user(&info
, argp
, sizeof(info
)))
4913 err
= add_new_disk(mddev
, &info
);
4917 case HOT_REMOVE_DISK
:
4918 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4922 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4925 case SET_DISK_FAULTY
:
4926 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4930 err
= do_md_run (mddev
);
4933 case SET_BITMAP_FILE
:
4934 err
= set_bitmap_file(mddev
, (int)arg
);
4944 mddev_unlock(mddev
);
4954 static int md_open(struct inode
*inode
, struct file
*file
)
4957 * Succeed if we can lock the mddev, which confirms that
4958 * it isn't being stopped right now.
4960 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4963 if ((err
= mutex_lock_interruptible_nested(&mddev
->reconfig_mutex
, 1)))
4968 mddev_unlock(mddev
);
4970 check_disk_change(inode
->i_bdev
);
4975 static int md_release(struct inode
*inode
, struct file
* file
)
4977 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4985 static int md_media_changed(struct gendisk
*disk
)
4987 mddev_t
*mddev
= disk
->private_data
;
4989 return mddev
->changed
;
4992 static int md_revalidate(struct gendisk
*disk
)
4994 mddev_t
*mddev
= disk
->private_data
;
4999 static struct block_device_operations md_fops
=
5001 .owner
= THIS_MODULE
,
5003 .release
= md_release
,
5005 .getgeo
= md_getgeo
,
5006 .media_changed
= md_media_changed
,
5007 .revalidate_disk
= md_revalidate
,
5010 static int md_thread(void * arg
)
5012 mdk_thread_t
*thread
= arg
;
5015 * md_thread is a 'system-thread', it's priority should be very
5016 * high. We avoid resource deadlocks individually in each
5017 * raid personality. (RAID5 does preallocation) We also use RR and
5018 * the very same RT priority as kswapd, thus we will never get
5019 * into a priority inversion deadlock.
5021 * we definitely have to have equal or higher priority than
5022 * bdflush, otherwise bdflush will deadlock if there are too
5023 * many dirty RAID5 blocks.
5026 allow_signal(SIGKILL
);
5027 while (!kthread_should_stop()) {
5029 /* We need to wait INTERRUPTIBLE so that
5030 * we don't add to the load-average.
5031 * That means we need to be sure no signals are
5034 if (signal_pending(current
))
5035 flush_signals(current
);
5037 wait_event_interruptible_timeout
5039 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5040 || kthread_should_stop(),
5043 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5045 thread
->run(thread
->mddev
);
5051 void md_wakeup_thread(mdk_thread_t
*thread
)
5054 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5055 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5056 wake_up(&thread
->wqueue
);
5060 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5063 mdk_thread_t
*thread
;
5065 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5069 init_waitqueue_head(&thread
->wqueue
);
5072 thread
->mddev
= mddev
;
5073 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5074 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
5075 if (IS_ERR(thread
->tsk
)) {
5082 void md_unregister_thread(mdk_thread_t
*thread
)
5084 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5086 kthread_stop(thread
->tsk
);
5090 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5097 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5100 if (mddev
->external
)
5101 set_bit(Blocked
, &rdev
->flags
);
5103 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5105 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5106 __builtin_return_address(0),__builtin_return_address(1),
5107 __builtin_return_address(2),__builtin_return_address(3));
5111 if (!mddev
->pers
->error_handler
)
5113 mddev
->pers
->error_handler(mddev
,rdev
);
5114 if (mddev
->degraded
)
5115 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5116 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5117 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5118 md_wakeup_thread(mddev
->thread
);
5119 md_new_event_inintr(mddev
);
5122 /* seq_file implementation /proc/mdstat */
5124 static void status_unused(struct seq_file
*seq
)
5128 struct list_head
*tmp
;
5130 seq_printf(seq
, "unused devices: ");
5132 rdev_for_each_list(rdev
, tmp
, pending_raid_disks
) {
5133 char b
[BDEVNAME_SIZE
];
5135 seq_printf(seq
, "%s ",
5136 bdevname(rdev
->bdev
,b
));
5139 seq_printf(seq
, "<none>");
5141 seq_printf(seq
, "\n");
5145 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
5147 sector_t max_blocks
, resync
, res
;
5148 unsigned long dt
, db
, rt
;
5150 unsigned int per_milli
;
5152 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
5154 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5155 max_blocks
= mddev
->resync_max_sectors
>> 1;
5157 max_blocks
= mddev
->size
;
5160 * Should not happen.
5166 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5167 * in a sector_t, and (max_blocks>>scale) will fit in a
5168 * u32, as those are the requirements for sector_div.
5169 * Thus 'scale' must be at least 10
5172 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5173 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
5176 res
= (resync
>>scale
)*1000;
5177 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
5181 int i
, x
= per_milli
/50, y
= 20-x
;
5182 seq_printf(seq
, "[");
5183 for (i
= 0; i
< x
; i
++)
5184 seq_printf(seq
, "=");
5185 seq_printf(seq
, ">");
5186 for (i
= 0; i
< y
; i
++)
5187 seq_printf(seq
, ".");
5188 seq_printf(seq
, "] ");
5190 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
5191 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
5193 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
5195 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
5196 "resync" : "recovery"))),
5197 per_milli
/10, per_milli
% 10,
5198 (unsigned long long) resync
,
5199 (unsigned long long) max_blocks
);
5202 * We do not want to overflow, so the order of operands and
5203 * the * 100 / 100 trick are important. We do a +1 to be
5204 * safe against division by zero. We only estimate anyway.
5206 * dt: time from mark until now
5207 * db: blocks written from mark until now
5208 * rt: remaining time
5210 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
5212 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
5213 - mddev
->resync_mark_cnt
;
5214 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/2/100+1)))/100;
5216 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
5218 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
5221 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
5223 struct list_head
*tmp
;
5233 spin_lock(&all_mddevs_lock
);
5234 list_for_each(tmp
,&all_mddevs
)
5236 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
5238 spin_unlock(&all_mddevs_lock
);
5241 spin_unlock(&all_mddevs_lock
);
5243 return (void*)2;/* tail */
5247 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
5249 struct list_head
*tmp
;
5250 mddev_t
*next_mddev
, *mddev
= v
;
5256 spin_lock(&all_mddevs_lock
);
5258 tmp
= all_mddevs
.next
;
5260 tmp
= mddev
->all_mddevs
.next
;
5261 if (tmp
!= &all_mddevs
)
5262 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
5264 next_mddev
= (void*)2;
5267 spin_unlock(&all_mddevs_lock
);
5275 static void md_seq_stop(struct seq_file
*seq
, void *v
)
5279 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
5283 struct mdstat_info
{
5287 static int md_seq_show(struct seq_file
*seq
, void *v
)
5291 struct list_head
*tmp2
;
5293 struct mdstat_info
*mi
= seq
->private;
5294 struct bitmap
*bitmap
;
5296 if (v
== (void*)1) {
5297 struct mdk_personality
*pers
;
5298 seq_printf(seq
, "Personalities : ");
5299 spin_lock(&pers_lock
);
5300 list_for_each_entry(pers
, &pers_list
, list
)
5301 seq_printf(seq
, "[%s] ", pers
->name
);
5303 spin_unlock(&pers_lock
);
5304 seq_printf(seq
, "\n");
5305 mi
->event
= atomic_read(&md_event_count
);
5308 if (v
== (void*)2) {
5313 if (mddev_lock(mddev
) < 0)
5316 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
5317 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
5318 mddev
->pers
? "" : "in");
5321 seq_printf(seq
, " (read-only)");
5323 seq_printf(seq
, " (auto-read-only)");
5324 seq_printf(seq
, " %s", mddev
->pers
->name
);
5328 rdev_for_each(rdev
, tmp2
, mddev
) {
5329 char b
[BDEVNAME_SIZE
];
5330 seq_printf(seq
, " %s[%d]",
5331 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
5332 if (test_bit(WriteMostly
, &rdev
->flags
))
5333 seq_printf(seq
, "(W)");
5334 if (test_bit(Faulty
, &rdev
->flags
)) {
5335 seq_printf(seq
, "(F)");
5337 } else if (rdev
->raid_disk
< 0)
5338 seq_printf(seq
, "(S)"); /* spare */
5342 if (!list_empty(&mddev
->disks
)) {
5344 seq_printf(seq
, "\n %llu blocks",
5345 (unsigned long long)mddev
->array_size
);
5347 seq_printf(seq
, "\n %llu blocks",
5348 (unsigned long long)size
);
5350 if (mddev
->persistent
) {
5351 if (mddev
->major_version
!= 0 ||
5352 mddev
->minor_version
!= 90) {
5353 seq_printf(seq
," super %d.%d",
5354 mddev
->major_version
,
5355 mddev
->minor_version
);
5357 } else if (mddev
->external
)
5358 seq_printf(seq
, " super external:%s",
5359 mddev
->metadata_type
);
5361 seq_printf(seq
, " super non-persistent");
5364 mddev
->pers
->status (seq
, mddev
);
5365 seq_printf(seq
, "\n ");
5366 if (mddev
->pers
->sync_request
) {
5367 if (mddev
->curr_resync
> 2) {
5368 status_resync (seq
, mddev
);
5369 seq_printf(seq
, "\n ");
5370 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
5371 seq_printf(seq
, "\tresync=DELAYED\n ");
5372 else if (mddev
->recovery_cp
< MaxSector
)
5373 seq_printf(seq
, "\tresync=PENDING\n ");
5376 seq_printf(seq
, "\n ");
5378 if ((bitmap
= mddev
->bitmap
)) {
5379 unsigned long chunk_kb
;
5380 unsigned long flags
;
5381 spin_lock_irqsave(&bitmap
->lock
, flags
);
5382 chunk_kb
= bitmap
->chunksize
>> 10;
5383 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
5385 bitmap
->pages
- bitmap
->missing_pages
,
5387 (bitmap
->pages
- bitmap
->missing_pages
)
5388 << (PAGE_SHIFT
- 10),
5389 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
5390 chunk_kb
? "KB" : "B");
5392 seq_printf(seq
, ", file: ");
5393 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
5396 seq_printf(seq
, "\n");
5397 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
5400 seq_printf(seq
, "\n");
5402 mddev_unlock(mddev
);
5407 static struct seq_operations md_seq_ops
= {
5408 .start
= md_seq_start
,
5409 .next
= md_seq_next
,
5410 .stop
= md_seq_stop
,
5411 .show
= md_seq_show
,
5414 static int md_seq_open(struct inode
*inode
, struct file
*file
)
5417 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
5421 error
= seq_open(file
, &md_seq_ops
);
5425 struct seq_file
*p
= file
->private_data
;
5427 mi
->event
= atomic_read(&md_event_count
);
5432 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
5434 struct seq_file
*m
= filp
->private_data
;
5435 struct mdstat_info
*mi
= m
->private;
5438 poll_wait(filp
, &md_event_waiters
, wait
);
5440 /* always allow read */
5441 mask
= POLLIN
| POLLRDNORM
;
5443 if (mi
->event
!= atomic_read(&md_event_count
))
5444 mask
|= POLLERR
| POLLPRI
;
5448 static const struct file_operations md_seq_fops
= {
5449 .owner
= THIS_MODULE
,
5450 .open
= md_seq_open
,
5452 .llseek
= seq_lseek
,
5453 .release
= seq_release_private
,
5454 .poll
= mdstat_poll
,
5457 int register_md_personality(struct mdk_personality
*p
)
5459 spin_lock(&pers_lock
);
5460 list_add_tail(&p
->list
, &pers_list
);
5461 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
5462 spin_unlock(&pers_lock
);
5466 int unregister_md_personality(struct mdk_personality
*p
)
5468 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
5469 spin_lock(&pers_lock
);
5470 list_del_init(&p
->list
);
5471 spin_unlock(&pers_lock
);
5475 static int is_mddev_idle(mddev_t
*mddev
)
5478 struct list_head
*tmp
;
5483 rdev_for_each(rdev
, tmp
, mddev
) {
5484 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
5485 curr_events
= disk_stat_read(disk
, sectors
[0]) +
5486 disk_stat_read(disk
, sectors
[1]) -
5487 atomic_read(&disk
->sync_io
);
5488 /* sync IO will cause sync_io to increase before the disk_stats
5489 * as sync_io is counted when a request starts, and
5490 * disk_stats is counted when it completes.
5491 * So resync activity will cause curr_events to be smaller than
5492 * when there was no such activity.
5493 * non-sync IO will cause disk_stat to increase without
5494 * increasing sync_io so curr_events will (eventually)
5495 * be larger than it was before. Once it becomes
5496 * substantially larger, the test below will cause
5497 * the array to appear non-idle, and resync will slow
5499 * If there is a lot of outstanding resync activity when
5500 * we set last_event to curr_events, then all that activity
5501 * completing might cause the array to appear non-idle
5502 * and resync will be slowed down even though there might
5503 * not have been non-resync activity. This will only
5504 * happen once though. 'last_events' will soon reflect
5505 * the state where there is little or no outstanding
5506 * resync requests, and further resync activity will
5507 * always make curr_events less than last_events.
5510 if (curr_events
- rdev
->last_events
> 4096) {
5511 rdev
->last_events
= curr_events
;
5518 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
5520 /* another "blocks" (512byte) blocks have been synced */
5521 atomic_sub(blocks
, &mddev
->recovery_active
);
5522 wake_up(&mddev
->recovery_wait
);
5524 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5525 md_wakeup_thread(mddev
->thread
);
5526 // stop recovery, signal do_sync ....
5531 /* md_write_start(mddev, bi)
5532 * If we need to update some array metadata (e.g. 'active' flag
5533 * in superblock) before writing, schedule a superblock update
5534 * and wait for it to complete.
5536 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
5539 if (bio_data_dir(bi
) != WRITE
)
5542 BUG_ON(mddev
->ro
== 1);
5543 if (mddev
->ro
== 2) {
5544 /* need to switch to read/write */
5546 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5547 md_wakeup_thread(mddev
->thread
);
5548 md_wakeup_thread(mddev
->sync_thread
);
5551 atomic_inc(&mddev
->writes_pending
);
5552 if (mddev
->safemode
== 1)
5553 mddev
->safemode
= 0;
5554 if (mddev
->in_sync
) {
5555 spin_lock_irq(&mddev
->write_lock
);
5556 if (mddev
->in_sync
) {
5558 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5559 md_wakeup_thread(mddev
->thread
);
5562 spin_unlock_irq(&mddev
->write_lock
);
5565 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
5566 wait_event(mddev
->sb_wait
,
5567 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
5568 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
5571 void md_write_end(mddev_t
*mddev
)
5573 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5574 if (mddev
->safemode
== 2)
5575 md_wakeup_thread(mddev
->thread
);
5576 else if (mddev
->safemode_delay
)
5577 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5581 /* md_allow_write(mddev)
5582 * Calling this ensures that the array is marked 'active' so that writes
5583 * may proceed without blocking. It is important to call this before
5584 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5585 * Must be called with mddev_lock held.
5587 void md_allow_write(mddev_t
*mddev
)
5593 if (!mddev
->pers
->sync_request
)
5596 spin_lock_irq(&mddev
->write_lock
);
5597 if (mddev
->in_sync
) {
5599 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5600 if (mddev
->safemode_delay
&&
5601 mddev
->safemode
== 0)
5602 mddev
->safemode
= 1;
5603 spin_unlock_irq(&mddev
->write_lock
);
5604 md_update_sb(mddev
, 0);
5606 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
5607 /* wait for the dirty state to be recorded in the metadata */
5608 wait_event(mddev
->sb_wait
,
5609 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
5610 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
5612 spin_unlock_irq(&mddev
->write_lock
);
5614 EXPORT_SYMBOL_GPL(md_allow_write
);
5616 #define SYNC_MARKS 10
5617 #define SYNC_MARK_STEP (3*HZ)
5618 void md_do_sync(mddev_t
*mddev
)
5621 unsigned int currspeed
= 0,
5623 sector_t max_sectors
,j
, io_sectors
;
5624 unsigned long mark
[SYNC_MARKS
];
5625 sector_t mark_cnt
[SYNC_MARKS
];
5627 struct list_head
*tmp
;
5628 sector_t last_check
;
5630 struct list_head
*rtmp
;
5634 /* just incase thread restarts... */
5635 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5637 if (mddev
->ro
) /* never try to sync a read-only array */
5640 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5641 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
5642 desc
= "data-check";
5643 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5644 desc
= "requested-resync";
5647 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5652 /* we overload curr_resync somewhat here.
5653 * 0 == not engaged in resync at all
5654 * 2 == checking that there is no conflict with another sync
5655 * 1 == like 2, but have yielded to allow conflicting resync to
5657 * other == active in resync - this many blocks
5659 * Before starting a resync we must have set curr_resync to
5660 * 2, and then checked that every "conflicting" array has curr_resync
5661 * less than ours. When we find one that is the same or higher
5662 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5663 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5664 * This will mean we have to start checking from the beginning again.
5669 mddev
->curr_resync
= 2;
5672 if (kthread_should_stop()) {
5673 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5676 for_each_mddev(mddev2
, tmp
) {
5677 if (mddev2
== mddev
)
5679 if (!mddev
->parallel_resync
5680 && mddev2
->curr_resync
5681 && match_mddev_units(mddev
, mddev2
)) {
5683 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5684 /* arbitrarily yield */
5685 mddev
->curr_resync
= 1;
5686 wake_up(&resync_wait
);
5688 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5689 /* no need to wait here, we can wait the next
5690 * time 'round when curr_resync == 2
5693 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
5694 if (!kthread_should_stop() &&
5695 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5696 printk(KERN_INFO
"md: delaying %s of %s"
5697 " until %s has finished (they"
5698 " share one or more physical units)\n",
5699 desc
, mdname(mddev
), mdname(mddev2
));
5702 finish_wait(&resync_wait
, &wq
);
5705 finish_wait(&resync_wait
, &wq
);
5708 } while (mddev
->curr_resync
< 2);
5711 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5712 /* resync follows the size requested by the personality,
5713 * which defaults to physical size, but can be virtual size
5715 max_sectors
= mddev
->resync_max_sectors
;
5716 mddev
->resync_mismatches
= 0;
5717 /* we don't use the checkpoint if there's a bitmap */
5718 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5719 j
= mddev
->resync_min
;
5720 else if (!mddev
->bitmap
)
5721 j
= mddev
->recovery_cp
;
5723 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5724 max_sectors
= mddev
->size
<< 1;
5726 /* recovery follows the physical size of devices */
5727 max_sectors
= mddev
->size
<< 1;
5729 rdev_for_each(rdev
, rtmp
, mddev
)
5730 if (rdev
->raid_disk
>= 0 &&
5731 !test_bit(Faulty
, &rdev
->flags
) &&
5732 !test_bit(In_sync
, &rdev
->flags
) &&
5733 rdev
->recovery_offset
< j
)
5734 j
= rdev
->recovery_offset
;
5737 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
5738 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
5739 " %d KB/sec/disk.\n", speed_min(mddev
));
5740 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5741 "(but not more than %d KB/sec) for %s.\n",
5742 speed_max(mddev
), desc
);
5744 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5747 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5749 mark_cnt
[m
] = io_sectors
;
5752 mddev
->resync_mark
= mark
[last_mark
];
5753 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5756 * Tune reconstruction:
5758 window
= 32*(PAGE_SIZE
/512);
5759 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5760 window
/2,(unsigned long long) max_sectors
/2);
5762 atomic_set(&mddev
->recovery_active
, 0);
5767 "md: resuming %s of %s from checkpoint.\n",
5768 desc
, mdname(mddev
));
5769 mddev
->curr_resync
= j
;
5772 while (j
< max_sectors
) {
5776 if (j
>= mddev
->resync_max
) {
5777 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
5778 wait_event(mddev
->recovery_wait
,
5779 mddev
->resync_max
> j
5780 || kthread_should_stop());
5782 if (kthread_should_stop())
5784 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5785 currspeed
< speed_min(mddev
));
5787 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5791 if (!skipped
) { /* actual IO requested */
5792 io_sectors
+= sectors
;
5793 atomic_add(sectors
, &mddev
->recovery_active
);
5797 if (j
>1) mddev
->curr_resync
= j
;
5798 mddev
->curr_mark_cnt
= io_sectors
;
5799 if (last_check
== 0)
5800 /* this is the earliers that rebuilt will be
5801 * visible in /proc/mdstat
5803 md_new_event(mddev
);
5805 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5808 last_check
= io_sectors
;
5810 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5814 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5816 int next
= (last_mark
+1) % SYNC_MARKS
;
5818 mddev
->resync_mark
= mark
[next
];
5819 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5820 mark
[next
] = jiffies
;
5821 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5826 if (kthread_should_stop())
5831 * this loop exits only if either when we are slower than
5832 * the 'hard' speed limit, or the system was IO-idle for
5834 * the system might be non-idle CPU-wise, but we only care
5835 * about not overloading the IO subsystem. (things like an
5836 * e2fsck being done on the RAID array should execute fast)
5838 blk_unplug(mddev
->queue
);
5841 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5842 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5844 if (currspeed
> speed_min(mddev
)) {
5845 if ((currspeed
> speed_max(mddev
)) ||
5846 !is_mddev_idle(mddev
)) {
5852 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
5854 * this also signals 'finished resyncing' to md_stop
5857 blk_unplug(mddev
->queue
);
5859 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5861 /* tell personality that we are finished */
5862 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5864 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5865 mddev
->curr_resync
> 2) {
5866 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5867 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5868 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5870 "md: checkpointing %s of %s.\n",
5871 desc
, mdname(mddev
));
5872 mddev
->recovery_cp
= mddev
->curr_resync
;
5875 mddev
->recovery_cp
= MaxSector
;
5877 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5878 mddev
->curr_resync
= MaxSector
;
5879 rdev_for_each(rdev
, rtmp
, mddev
)
5880 if (rdev
->raid_disk
>= 0 &&
5881 !test_bit(Faulty
, &rdev
->flags
) &&
5882 !test_bit(In_sync
, &rdev
->flags
) &&
5883 rdev
->recovery_offset
< mddev
->curr_resync
)
5884 rdev
->recovery_offset
= mddev
->curr_resync
;
5887 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5890 mddev
->curr_resync
= 0;
5891 mddev
->resync_min
= 0;
5892 mddev
->resync_max
= MaxSector
;
5893 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
5894 wake_up(&resync_wait
);
5895 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5896 md_wakeup_thread(mddev
->thread
);
5901 * got a signal, exit.
5904 "md: md_do_sync() got signal ... exiting\n");
5905 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5909 EXPORT_SYMBOL_GPL(md_do_sync
);
5912 static int remove_and_add_spares(mddev_t
*mddev
)
5915 struct list_head
*rtmp
;
5918 rdev_for_each(rdev
, rtmp
, mddev
)
5919 if (rdev
->raid_disk
>= 0 &&
5920 !test_bit(Blocked
, &rdev
->flags
) &&
5921 (test_bit(Faulty
, &rdev
->flags
) ||
5922 ! test_bit(In_sync
, &rdev
->flags
)) &&
5923 atomic_read(&rdev
->nr_pending
)==0) {
5924 if (mddev
->pers
->hot_remove_disk(
5925 mddev
, rdev
->raid_disk
)==0) {
5927 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5928 sysfs_remove_link(&mddev
->kobj
, nm
);
5929 rdev
->raid_disk
= -1;
5933 if (mddev
->degraded
) {
5934 rdev_for_each(rdev
, rtmp
, mddev
) {
5935 if (rdev
->raid_disk
>= 0 &&
5936 !test_bit(In_sync
, &rdev
->flags
))
5938 if (rdev
->raid_disk
< 0
5939 && !test_bit(Faulty
, &rdev
->flags
)) {
5940 rdev
->recovery_offset
= 0;
5942 hot_add_disk(mddev
, rdev
) == 0) {
5944 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5945 if (sysfs_create_link(&mddev
->kobj
,
5948 "md: cannot register "
5952 md_new_event(mddev
);
5961 * This routine is regularly called by all per-raid-array threads to
5962 * deal with generic issues like resync and super-block update.
5963 * Raid personalities that don't have a thread (linear/raid0) do not
5964 * need this as they never do any recovery or update the superblock.
5966 * It does not do any resync itself, but rather "forks" off other threads
5967 * to do that as needed.
5968 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5969 * "->recovery" and create a thread at ->sync_thread.
5970 * When the thread finishes it sets MD_RECOVERY_DONE
5971 * and wakeups up this thread which will reap the thread and finish up.
5972 * This thread also removes any faulty devices (with nr_pending == 0).
5974 * The overall approach is:
5975 * 1/ if the superblock needs updating, update it.
5976 * 2/ If a recovery thread is running, don't do anything else.
5977 * 3/ If recovery has finished, clean up, possibly marking spares active.
5978 * 4/ If there are any faulty devices, remove them.
5979 * 5/ If array is degraded, try to add spares devices
5980 * 6/ If array has spares or is not in-sync, start a resync thread.
5982 void md_check_recovery(mddev_t
*mddev
)
5985 struct list_head
*rtmp
;
5989 bitmap_daemon_work(mddev
->bitmap
);
5994 if (signal_pending(current
)) {
5995 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
5996 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5998 mddev
->safemode
= 2;
6000 flush_signals(current
);
6004 (mddev
->flags
&& !mddev
->external
) ||
6005 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6006 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6007 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6008 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6009 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6013 if (mddev_trylock(mddev
)) {
6016 if (!mddev
->external
) {
6018 spin_lock_irq(&mddev
->write_lock
);
6019 if (mddev
->safemode
&&
6020 !atomic_read(&mddev
->writes_pending
) &&
6022 mddev
->recovery_cp
== MaxSector
) {
6025 if (mddev
->persistent
)
6026 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6028 if (mddev
->safemode
== 1)
6029 mddev
->safemode
= 0;
6030 spin_unlock_irq(&mddev
->write_lock
);
6032 sysfs_notify(&mddev
->kobj
, NULL
, "array_state");
6036 md_update_sb(mddev
, 0);
6039 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6040 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6041 /* resync/recovery still happening */
6042 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6045 if (mddev
->sync_thread
) {
6046 /* resync has finished, collect result */
6047 md_unregister_thread(mddev
->sync_thread
);
6048 mddev
->sync_thread
= NULL
;
6049 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6051 /* activate any spares */
6052 mddev
->pers
->spare_active(mddev
);
6054 md_update_sb(mddev
, 1);
6056 /* if array is no-longer degraded, then any saved_raid_disk
6057 * information must be scrapped
6059 if (!mddev
->degraded
)
6060 rdev_for_each(rdev
, rtmp
, mddev
)
6061 rdev
->saved_raid_disk
= -1;
6063 mddev
->recovery
= 0;
6064 /* flag recovery needed just to double check */
6065 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6066 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
6067 md_new_event(mddev
);
6070 /* Set RUNNING before clearing NEEDED to avoid
6071 * any transients in the value of "sync_action".
6073 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6074 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6075 /* Clear some bits that don't mean anything, but
6078 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6079 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6081 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
6083 /* no recovery is running.
6084 * remove any failed drives, then
6085 * add spares if possible.
6086 * Spare are also removed and re-added, to allow
6087 * the personality to fail the re-add.
6090 if (mddev
->reshape_position
!= MaxSector
) {
6091 if (mddev
->pers
->check_reshape(mddev
) != 0)
6092 /* Cannot proceed */
6094 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
6095 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6096 } else if ((spares
= remove_and_add_spares(mddev
))) {
6097 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6098 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
6099 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6100 } else if (mddev
->recovery_cp
< MaxSector
) {
6101 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6102 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6103 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6104 /* nothing to be done ... */
6107 if (mddev
->pers
->sync_request
) {
6108 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
6109 /* We are adding a device or devices to an array
6110 * which has the bitmap stored on all devices.
6111 * So make sure all bitmap pages get written
6113 bitmap_write_all(mddev
->bitmap
);
6115 mddev
->sync_thread
= md_register_thread(md_do_sync
,
6118 if (!mddev
->sync_thread
) {
6119 printk(KERN_ERR
"%s: could not start resync"
6122 /* leave the spares where they are, it shouldn't hurt */
6123 mddev
->recovery
= 0;
6125 md_wakeup_thread(mddev
->sync_thread
);
6126 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
6127 md_new_event(mddev
);
6130 if (!mddev
->sync_thread
) {
6131 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6132 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
6134 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
6136 mddev_unlock(mddev
);
6140 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
6142 sysfs_notify(&rdev
->kobj
, NULL
, "state");
6143 wait_event_timeout(rdev
->blocked_wait
,
6144 !test_bit(Blocked
, &rdev
->flags
),
6145 msecs_to_jiffies(5000));
6146 rdev_dec_pending(rdev
, mddev
);
6148 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
6150 static int md_notify_reboot(struct notifier_block
*this,
6151 unsigned long code
, void *x
)
6153 struct list_head
*tmp
;
6156 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
6158 printk(KERN_INFO
"md: stopping all md devices.\n");
6160 for_each_mddev(mddev
, tmp
)
6161 if (mddev_trylock(mddev
)) {
6162 do_md_stop (mddev
, 1);
6163 mddev_unlock(mddev
);
6166 * certain more exotic SCSI devices are known to be
6167 * volatile wrt too early system reboots. While the
6168 * right place to handle this issue is the given
6169 * driver, we do want to have a safe RAID driver ...
6176 static struct notifier_block md_notifier
= {
6177 .notifier_call
= md_notify_reboot
,
6179 .priority
= INT_MAX
, /* before any real devices */
6182 static void md_geninit(void)
6184 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
6186 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
6189 static int __init
md_init(void)
6191 if (register_blkdev(MAJOR_NR
, "md"))
6193 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
6194 unregister_blkdev(MAJOR_NR
, "md");
6197 blk_register_region(MKDEV(MAJOR_NR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6198 md_probe
, NULL
, NULL
);
6199 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6200 md_probe
, NULL
, NULL
);
6202 register_reboot_notifier(&md_notifier
);
6203 raid_table_header
= register_sysctl_table(raid_root_table
);
6213 * Searches all registered partitions for autorun RAID arrays
6217 static LIST_HEAD(all_detected_devices
);
6218 struct detected_devices_node
{
6219 struct list_head list
;
6223 void md_autodetect_dev(dev_t dev
)
6225 struct detected_devices_node
*node_detected_dev
;
6227 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
6228 if (node_detected_dev
) {
6229 node_detected_dev
->dev
= dev
;
6230 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
6232 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
6233 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
6238 static void autostart_arrays(int part
)
6241 struct detected_devices_node
*node_detected_dev
;
6243 int i_scanned
, i_passed
;
6248 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
6250 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
6252 node_detected_dev
= list_entry(all_detected_devices
.next
,
6253 struct detected_devices_node
, list
);
6254 list_del(&node_detected_dev
->list
);
6255 dev
= node_detected_dev
->dev
;
6256 kfree(node_detected_dev
);
6257 rdev
= md_import_device(dev
,0, 90);
6261 if (test_bit(Faulty
, &rdev
->flags
)) {
6265 set_bit(AutoDetected
, &rdev
->flags
);
6266 list_add(&rdev
->same_set
, &pending_raid_disks
);
6270 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
6271 i_scanned
, i_passed
);
6273 autorun_devices(part
);
6276 #endif /* !MODULE */
6278 static __exit
void md_exit(void)
6281 struct list_head
*tmp
;
6283 blk_unregister_region(MKDEV(MAJOR_NR
,0), 1U << MINORBITS
);
6284 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
6286 unregister_blkdev(MAJOR_NR
,"md");
6287 unregister_blkdev(mdp_major
, "mdp");
6288 unregister_reboot_notifier(&md_notifier
);
6289 unregister_sysctl_table(raid_table_header
);
6290 remove_proc_entry("mdstat", NULL
);
6291 for_each_mddev(mddev
, tmp
) {
6292 struct gendisk
*disk
= mddev
->gendisk
;
6295 export_array(mddev
);
6298 mddev
->gendisk
= NULL
;
6303 subsys_initcall(md_init
);
6304 module_exit(md_exit
)
6306 static int get_ro(char *buffer
, struct kernel_param
*kp
)
6308 return sprintf(buffer
, "%d", start_readonly
);
6310 static int set_ro(const char *val
, struct kernel_param
*kp
)
6313 int num
= simple_strtoul(val
, &e
, 10);
6314 if (*val
&& (*e
== '\0' || *e
== '\n')) {
6315 start_readonly
= num
;
6321 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
6322 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
6325 EXPORT_SYMBOL(register_md_personality
);
6326 EXPORT_SYMBOL(unregister_md_personality
);
6327 EXPORT_SYMBOL(md_error
);
6328 EXPORT_SYMBOL(md_done_sync
);
6329 EXPORT_SYMBOL(md_write_start
);
6330 EXPORT_SYMBOL(md_write_end
);
6331 EXPORT_SYMBOL(md_register_thread
);
6332 EXPORT_SYMBOL(md_unregister_thread
);
6333 EXPORT_SYMBOL(md_wakeup_thread
);
6334 EXPORT_SYMBOL(md_check_recovery
);
6335 MODULE_LICENSE("GPL");
6337 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);