2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/raid/md.h>
37 #include <linux/raid/bitmap.h>
38 #include <linux/sysctl.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
49 #define MAJOR_NR MD_MAJOR
51 /* 63 partitions with the alternate major number (mdp) */
52 #define MdpMinorShift 6
55 #define dprintk(x...) ((void)(DEBUG && printk(x)))
59 static void autostart_arrays(int part
);
62 static LIST_HEAD(pers_list
);
63 static DEFINE_SPINLOCK(pers_lock
);
65 static void md_print_devices(void);
67 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
69 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
72 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
73 * is 1000 KB/sec, so the extra system load does not show up that much.
74 * Increase it if you want to have more _guaranteed_ speed. Note that
75 * the RAID driver will use the maximum available bandwidth if the IO
76 * subsystem is idle. There is also an 'absolute maximum' reconstruction
77 * speed limit - in case reconstruction slows down your system despite
80 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
81 * or /sys/block/mdX/md/sync_speed_{min,max}
84 static int sysctl_speed_limit_min
= 1000;
85 static int sysctl_speed_limit_max
= 200000;
86 static inline int speed_min(mddev_t
*mddev
)
88 return mddev
->sync_speed_min
?
89 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
92 static inline int speed_max(mddev_t
*mddev
)
94 return mddev
->sync_speed_max
?
95 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
98 static struct ctl_table_header
*raid_table_header
;
100 static ctl_table raid_table
[] = {
102 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
103 .procname
= "speed_limit_min",
104 .data
= &sysctl_speed_limit_min
,
105 .maxlen
= sizeof(int),
106 .mode
= S_IRUGO
|S_IWUSR
,
107 .proc_handler
= &proc_dointvec
,
110 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
111 .procname
= "speed_limit_max",
112 .data
= &sysctl_speed_limit_max
,
113 .maxlen
= sizeof(int),
114 .mode
= S_IRUGO
|S_IWUSR
,
115 .proc_handler
= &proc_dointvec
,
120 static ctl_table raid_dir_table
[] = {
122 .ctl_name
= DEV_RAID
,
125 .mode
= S_IRUGO
|S_IXUGO
,
131 static ctl_table raid_root_table
[] = {
137 .child
= raid_dir_table
,
142 static struct block_device_operations md_fops
;
144 static int start_readonly
;
147 * We have a system wide 'event count' that is incremented
148 * on any 'interesting' event, and readers of /proc/mdstat
149 * can use 'poll' or 'select' to find out when the event
153 * start array, stop array, error, add device, remove device,
154 * start build, activate spare
156 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
157 static atomic_t md_event_count
;
158 void md_new_event(mddev_t
*mddev
)
160 atomic_inc(&md_event_count
);
161 wake_up(&md_event_waiters
);
163 EXPORT_SYMBOL_GPL(md_new_event
);
165 /* Alternate version that can be called from interrupts
166 * when calling sysfs_notify isn't needed.
168 static void md_new_event_inintr(mddev_t
*mddev
)
170 atomic_inc(&md_event_count
);
171 wake_up(&md_event_waiters
);
175 * Enables to iterate over all existing md arrays
176 * all_mddevs_lock protects this list.
178 static LIST_HEAD(all_mddevs
);
179 static DEFINE_SPINLOCK(all_mddevs_lock
);
183 * iterates through all used mddevs in the system.
184 * We take care to grab the all_mddevs_lock whenever navigating
185 * the list, and to always hold a refcount when unlocked.
186 * Any code which breaks out of this loop while own
187 * a reference to the current mddev and must mddev_put it.
189 #define for_each_mddev(mddev,tmp) \
191 for (({ spin_lock(&all_mddevs_lock); \
192 tmp = all_mddevs.next; \
194 ({ if (tmp != &all_mddevs) \
195 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
196 spin_unlock(&all_mddevs_lock); \
197 if (mddev) mddev_put(mddev); \
198 mddev = list_entry(tmp, mddev_t, all_mddevs); \
199 tmp != &all_mddevs;}); \
200 ({ spin_lock(&all_mddevs_lock); \
205 static int md_fail_request(struct request_queue
*q
, struct bio
*bio
)
211 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
213 atomic_inc(&mddev
->active
);
217 static void mddev_put(mddev_t
*mddev
)
219 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
221 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
222 list_del(&mddev
->all_mddevs
);
223 spin_unlock(&all_mddevs_lock
);
225 blk_cleanup_queue(mddev
->queue
);
227 if (mddev
->sysfs_state
)
228 sysfs_put(mddev
->sysfs_state
);
229 mddev
->sysfs_state
= NULL
;
230 kobject_put(&mddev
->kobj
);
232 spin_unlock(&all_mddevs_lock
);
235 static mddev_t
* mddev_find(dev_t unit
)
237 mddev_t
*mddev
, *new = NULL
;
240 spin_lock(&all_mddevs_lock
);
241 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
242 if (mddev
->unit
== unit
) {
244 spin_unlock(&all_mddevs_lock
);
250 list_add(&new->all_mddevs
, &all_mddevs
);
251 spin_unlock(&all_mddevs_lock
);
254 spin_unlock(&all_mddevs_lock
);
256 new = kzalloc(sizeof(*new), GFP_KERNEL
);
261 if (MAJOR(unit
) == MD_MAJOR
)
262 new->md_minor
= MINOR(unit
);
264 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
266 mutex_init(&new->reconfig_mutex
);
267 INIT_LIST_HEAD(&new->disks
);
268 INIT_LIST_HEAD(&new->all_mddevs
);
269 init_timer(&new->safemode_timer
);
270 atomic_set(&new->active
, 1);
271 atomic_set(&new->openers
, 0);
272 spin_lock_init(&new->write_lock
);
273 init_waitqueue_head(&new->sb_wait
);
274 init_waitqueue_head(&new->recovery_wait
);
275 new->reshape_position
= MaxSector
;
277 new->resync_max
= MaxSector
;
278 new->level
= LEVEL_NONE
;
283 static inline int mddev_lock(mddev_t
* mddev
)
285 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
288 static inline int mddev_trylock(mddev_t
* mddev
)
290 return mutex_trylock(&mddev
->reconfig_mutex
);
293 static inline void mddev_unlock(mddev_t
* mddev
)
295 mutex_unlock(&mddev
->reconfig_mutex
);
297 md_wakeup_thread(mddev
->thread
);
300 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
304 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
305 if (rdev
->desc_nr
== nr
)
311 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
315 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
316 if (rdev
->bdev
->bd_dev
== dev
)
322 static struct mdk_personality
*find_pers(int level
, char *clevel
)
324 struct mdk_personality
*pers
;
325 list_for_each_entry(pers
, &pers_list
, list
) {
326 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
328 if (strcmp(pers
->name
, clevel
)==0)
334 /* return the offset of the super block in 512byte sectors */
335 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
337 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
338 return MD_NEW_SIZE_SECTORS(num_sectors
);
341 static sector_t
calc_num_sectors(mdk_rdev_t
*rdev
, unsigned chunk_size
)
343 sector_t num_sectors
= rdev
->sb_start
;
346 num_sectors
&= ~((sector_t
)chunk_size
/512 - 1);
350 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
355 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
356 if (!rdev
->sb_page
) {
357 printk(KERN_ALERT
"md: out of memory.\n");
364 static void free_disk_sb(mdk_rdev_t
* rdev
)
367 put_page(rdev
->sb_page
);
369 rdev
->sb_page
= NULL
;
376 static void super_written(struct bio
*bio
, int error
)
378 mdk_rdev_t
*rdev
= bio
->bi_private
;
379 mddev_t
*mddev
= rdev
->mddev
;
381 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
382 printk("md: super_written gets error=%d, uptodate=%d\n",
383 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
384 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
385 md_error(mddev
, rdev
);
388 if (atomic_dec_and_test(&mddev
->pending_writes
))
389 wake_up(&mddev
->sb_wait
);
393 static void super_written_barrier(struct bio
*bio
, int error
)
395 struct bio
*bio2
= bio
->bi_private
;
396 mdk_rdev_t
*rdev
= bio2
->bi_private
;
397 mddev_t
*mddev
= rdev
->mddev
;
399 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
400 error
== -EOPNOTSUPP
) {
402 /* barriers don't appear to be supported :-( */
403 set_bit(BarriersNotsupp
, &rdev
->flags
);
404 mddev
->barriers_work
= 0;
405 spin_lock_irqsave(&mddev
->write_lock
, flags
);
406 bio2
->bi_next
= mddev
->biolist
;
407 mddev
->biolist
= bio2
;
408 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
409 wake_up(&mddev
->sb_wait
);
413 bio
->bi_private
= rdev
;
414 super_written(bio
, error
);
418 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
419 sector_t sector
, int size
, struct page
*page
)
421 /* write first size bytes of page to sector of rdev
422 * Increment mddev->pending_writes before returning
423 * and decrement it on completion, waking up sb_wait
424 * if zero is reached.
425 * If an error occurred, call md_error
427 * As we might need to resubmit the request if BIO_RW_BARRIER
428 * causes ENOTSUPP, we allocate a spare bio...
430 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
431 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
433 bio
->bi_bdev
= rdev
->bdev
;
434 bio
->bi_sector
= sector
;
435 bio_add_page(bio
, page
, size
, 0);
436 bio
->bi_private
= rdev
;
437 bio
->bi_end_io
= super_written
;
440 atomic_inc(&mddev
->pending_writes
);
441 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
443 rw
|= (1<<BIO_RW_BARRIER
);
444 rbio
= bio_clone(bio
, GFP_NOIO
);
445 rbio
->bi_private
= bio
;
446 rbio
->bi_end_io
= super_written_barrier
;
447 submit_bio(rw
, rbio
);
452 void md_super_wait(mddev_t
*mddev
)
454 /* wait for all superblock writes that were scheduled to complete.
455 * if any had to be retried (due to BARRIER problems), retry them
459 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
460 if (atomic_read(&mddev
->pending_writes
)==0)
462 while (mddev
->biolist
) {
464 spin_lock_irq(&mddev
->write_lock
);
465 bio
= mddev
->biolist
;
466 mddev
->biolist
= bio
->bi_next
;
468 spin_unlock_irq(&mddev
->write_lock
);
469 submit_bio(bio
->bi_rw
, bio
);
473 finish_wait(&mddev
->sb_wait
, &wq
);
476 static void bi_complete(struct bio
*bio
, int error
)
478 complete((struct completion
*)bio
->bi_private
);
481 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
482 struct page
*page
, int rw
)
484 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
485 struct completion event
;
488 rw
|= (1 << BIO_RW_SYNC
);
491 bio
->bi_sector
= sector
;
492 bio_add_page(bio
, page
, size
, 0);
493 init_completion(&event
);
494 bio
->bi_private
= &event
;
495 bio
->bi_end_io
= bi_complete
;
497 wait_for_completion(&event
);
499 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
503 EXPORT_SYMBOL_GPL(sync_page_io
);
505 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
507 char b
[BDEVNAME_SIZE
];
508 if (!rdev
->sb_page
) {
516 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
522 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
523 bdevname(rdev
->bdev
,b
));
527 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
529 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
530 sb1
->set_uuid1
== sb2
->set_uuid1
&&
531 sb1
->set_uuid2
== sb2
->set_uuid2
&&
532 sb1
->set_uuid3
== sb2
->set_uuid3
;
535 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
538 mdp_super_t
*tmp1
, *tmp2
;
540 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
541 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
543 if (!tmp1
|| !tmp2
) {
545 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
553 * nr_disks is not constant
558 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
566 static u32
md_csum_fold(u32 csum
)
568 csum
= (csum
& 0xffff) + (csum
>> 16);
569 return (csum
& 0xffff) + (csum
>> 16);
572 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
575 u32
*sb32
= (u32
*)sb
;
577 unsigned int disk_csum
, csum
;
579 disk_csum
= sb
->sb_csum
;
582 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
584 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
588 /* This used to use csum_partial, which was wrong for several
589 * reasons including that different results are returned on
590 * different architectures. It isn't critical that we get exactly
591 * the same return value as before (we always csum_fold before
592 * testing, and that removes any differences). However as we
593 * know that csum_partial always returned a 16bit value on
594 * alphas, do a fold to maximise conformity to previous behaviour.
596 sb
->sb_csum
= md_csum_fold(disk_csum
);
598 sb
->sb_csum
= disk_csum
;
605 * Handle superblock details.
606 * We want to be able to handle multiple superblock formats
607 * so we have a common interface to them all, and an array of
608 * different handlers.
609 * We rely on user-space to write the initial superblock, and support
610 * reading and updating of superblocks.
611 * Interface methods are:
612 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
613 * loads and validates a superblock on dev.
614 * if refdev != NULL, compare superblocks on both devices
616 * 0 - dev has a superblock that is compatible with refdev
617 * 1 - dev has a superblock that is compatible and newer than refdev
618 * so dev should be used as the refdev in future
619 * -EINVAL superblock incompatible or invalid
620 * -othererror e.g. -EIO
622 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
623 * Verify that dev is acceptable into mddev.
624 * The first time, mddev->raid_disks will be 0, and data from
625 * dev should be merged in. Subsequent calls check that dev
626 * is new enough. Return 0 or -EINVAL
628 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
629 * Update the superblock for rdev with data in mddev
630 * This does not write to disc.
636 struct module
*owner
;
637 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
639 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
640 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
641 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
642 sector_t num_sectors
);
646 * load_super for 0.90.0
648 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
650 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
655 * Calculate the position of the superblock (512byte sectors),
656 * it's at the end of the disk.
658 * It also happens to be a multiple of 4Kb.
660 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
662 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
667 bdevname(rdev
->bdev
, b
);
668 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
670 if (sb
->md_magic
!= MD_SB_MAGIC
) {
671 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
676 if (sb
->major_version
!= 0 ||
677 sb
->minor_version
< 90 ||
678 sb
->minor_version
> 91) {
679 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
680 sb
->major_version
, sb
->minor_version
,
685 if (sb
->raid_disks
<= 0)
688 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
689 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
694 rdev
->preferred_minor
= sb
->md_minor
;
695 rdev
->data_offset
= 0;
696 rdev
->sb_size
= MD_SB_BYTES
;
698 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
699 if (sb
->level
!= 1 && sb
->level
!= 4
700 && sb
->level
!= 5 && sb
->level
!= 6
701 && sb
->level
!= 10) {
702 /* FIXME use a better test */
704 "md: bitmaps not supported for this level.\n");
709 if (sb
->level
== LEVEL_MULTIPATH
)
712 rdev
->desc_nr
= sb
->this_disk
.number
;
718 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
719 if (!uuid_equal(refsb
, sb
)) {
720 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
721 b
, bdevname(refdev
->bdev
,b2
));
724 if (!sb_equal(refsb
, sb
)) {
725 printk(KERN_WARNING
"md: %s has same UUID"
726 " but different superblock to %s\n",
727 b
, bdevname(refdev
->bdev
, b2
));
731 ev2
= md_event(refsb
);
737 rdev
->size
= calc_num_sectors(rdev
, sb
->chunk_size
) / 2;
739 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
740 /* "this cannot possibly happen" ... */
748 * validate_super for 0.90.0
750 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
753 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
754 __u64 ev1
= md_event(sb
);
756 rdev
->raid_disk
= -1;
757 clear_bit(Faulty
, &rdev
->flags
);
758 clear_bit(In_sync
, &rdev
->flags
);
759 clear_bit(WriteMostly
, &rdev
->flags
);
760 clear_bit(BarriersNotsupp
, &rdev
->flags
);
762 if (mddev
->raid_disks
== 0) {
763 mddev
->major_version
= 0;
764 mddev
->minor_version
= sb
->minor_version
;
765 mddev
->patch_version
= sb
->patch_version
;
767 mddev
->chunk_size
= sb
->chunk_size
;
768 mddev
->ctime
= sb
->ctime
;
769 mddev
->utime
= sb
->utime
;
770 mddev
->level
= sb
->level
;
771 mddev
->clevel
[0] = 0;
772 mddev
->layout
= sb
->layout
;
773 mddev
->raid_disks
= sb
->raid_disks
;
774 mddev
->size
= sb
->size
;
776 mddev
->bitmap_offset
= 0;
777 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
779 if (mddev
->minor_version
>= 91) {
780 mddev
->reshape_position
= sb
->reshape_position
;
781 mddev
->delta_disks
= sb
->delta_disks
;
782 mddev
->new_level
= sb
->new_level
;
783 mddev
->new_layout
= sb
->new_layout
;
784 mddev
->new_chunk
= sb
->new_chunk
;
786 mddev
->reshape_position
= MaxSector
;
787 mddev
->delta_disks
= 0;
788 mddev
->new_level
= mddev
->level
;
789 mddev
->new_layout
= mddev
->layout
;
790 mddev
->new_chunk
= mddev
->chunk_size
;
793 if (sb
->state
& (1<<MD_SB_CLEAN
))
794 mddev
->recovery_cp
= MaxSector
;
796 if (sb
->events_hi
== sb
->cp_events_hi
&&
797 sb
->events_lo
== sb
->cp_events_lo
) {
798 mddev
->recovery_cp
= sb
->recovery_cp
;
800 mddev
->recovery_cp
= 0;
803 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
804 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
805 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
806 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
808 mddev
->max_disks
= MD_SB_DISKS
;
810 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
811 mddev
->bitmap_file
== NULL
)
812 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
814 } else if (mddev
->pers
== NULL
) {
815 /* Insist on good event counter while assembling */
817 if (ev1
< mddev
->events
)
819 } else if (mddev
->bitmap
) {
820 /* if adding to array with a bitmap, then we can accept an
821 * older device ... but not too old.
823 if (ev1
< mddev
->bitmap
->events_cleared
)
826 if (ev1
< mddev
->events
)
827 /* just a hot-add of a new device, leave raid_disk at -1 */
831 if (mddev
->level
!= LEVEL_MULTIPATH
) {
832 desc
= sb
->disks
+ rdev
->desc_nr
;
834 if (desc
->state
& (1<<MD_DISK_FAULTY
))
835 set_bit(Faulty
, &rdev
->flags
);
836 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
837 desc->raid_disk < mddev->raid_disks */) {
838 set_bit(In_sync
, &rdev
->flags
);
839 rdev
->raid_disk
= desc
->raid_disk
;
841 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
842 set_bit(WriteMostly
, &rdev
->flags
);
843 } else /* MULTIPATH are always insync */
844 set_bit(In_sync
, &rdev
->flags
);
849 * sync_super for 0.90.0
851 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
855 int next_spare
= mddev
->raid_disks
;
858 /* make rdev->sb match mddev data..
861 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
862 * 3/ any empty disks < next_spare become removed
864 * disks[0] gets initialised to REMOVED because
865 * we cannot be sure from other fields if it has
866 * been initialised or not.
869 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
871 rdev
->sb_size
= MD_SB_BYTES
;
873 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
875 memset(sb
, 0, sizeof(*sb
));
877 sb
->md_magic
= MD_SB_MAGIC
;
878 sb
->major_version
= mddev
->major_version
;
879 sb
->patch_version
= mddev
->patch_version
;
880 sb
->gvalid_words
= 0; /* ignored */
881 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
882 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
883 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
884 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
886 sb
->ctime
= mddev
->ctime
;
887 sb
->level
= mddev
->level
;
888 sb
->size
= mddev
->size
;
889 sb
->raid_disks
= mddev
->raid_disks
;
890 sb
->md_minor
= mddev
->md_minor
;
891 sb
->not_persistent
= 0;
892 sb
->utime
= mddev
->utime
;
894 sb
->events_hi
= (mddev
->events
>>32);
895 sb
->events_lo
= (u32
)mddev
->events
;
897 if (mddev
->reshape_position
== MaxSector
)
898 sb
->minor_version
= 90;
900 sb
->minor_version
= 91;
901 sb
->reshape_position
= mddev
->reshape_position
;
902 sb
->new_level
= mddev
->new_level
;
903 sb
->delta_disks
= mddev
->delta_disks
;
904 sb
->new_layout
= mddev
->new_layout
;
905 sb
->new_chunk
= mddev
->new_chunk
;
907 mddev
->minor_version
= sb
->minor_version
;
910 sb
->recovery_cp
= mddev
->recovery_cp
;
911 sb
->cp_events_hi
= (mddev
->events
>>32);
912 sb
->cp_events_lo
= (u32
)mddev
->events
;
913 if (mddev
->recovery_cp
== MaxSector
)
914 sb
->state
= (1<< MD_SB_CLEAN
);
918 sb
->layout
= mddev
->layout
;
919 sb
->chunk_size
= mddev
->chunk_size
;
921 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
922 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
924 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
925 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
928 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
929 && !test_bit(Faulty
, &rdev2
->flags
))
930 desc_nr
= rdev2
->raid_disk
;
932 desc_nr
= next_spare
++;
933 rdev2
->desc_nr
= desc_nr
;
934 d
= &sb
->disks
[rdev2
->desc_nr
];
936 d
->number
= rdev2
->desc_nr
;
937 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
938 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
939 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
940 && !test_bit(Faulty
, &rdev2
->flags
))
941 d
->raid_disk
= rdev2
->raid_disk
;
943 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
944 if (test_bit(Faulty
, &rdev2
->flags
))
945 d
->state
= (1<<MD_DISK_FAULTY
);
946 else if (test_bit(In_sync
, &rdev2
->flags
)) {
947 d
->state
= (1<<MD_DISK_ACTIVE
);
948 d
->state
|= (1<<MD_DISK_SYNC
);
956 if (test_bit(WriteMostly
, &rdev2
->flags
))
957 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
959 /* now set the "removed" and "faulty" bits on any missing devices */
960 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
961 mdp_disk_t
*d
= &sb
->disks
[i
];
962 if (d
->state
== 0 && d
->number
== 0) {
965 d
->state
= (1<<MD_DISK_REMOVED
);
966 d
->state
|= (1<<MD_DISK_FAULTY
);
970 sb
->nr_disks
= nr_disks
;
971 sb
->active_disks
= active
;
972 sb
->working_disks
= working
;
973 sb
->failed_disks
= failed
;
974 sb
->spare_disks
= spare
;
976 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
977 sb
->sb_csum
= calc_sb_csum(sb
);
981 * rdev_size_change for 0.90.0
983 static unsigned long long
984 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
986 if (num_sectors
&& num_sectors
< rdev
->mddev
->size
* 2)
987 return 0; /* component must fit device */
988 if (rdev
->mddev
->bitmap_offset
)
989 return 0; /* can't move bitmap */
990 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
991 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
992 num_sectors
= rdev
->sb_start
;
993 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
995 md_super_wait(rdev
->mddev
);
996 return num_sectors
/ 2; /* kB for sysfs */
1001 * version 1 superblock
1004 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1008 unsigned long long newcsum
;
1009 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1010 __le32
*isuper
= (__le32
*)sb
;
1013 disk_csum
= sb
->sb_csum
;
1016 for (i
=0; size
>=4; size
-= 4 )
1017 newcsum
+= le32_to_cpu(*isuper
++);
1020 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1022 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1023 sb
->sb_csum
= disk_csum
;
1024 return cpu_to_le32(csum
);
1027 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1029 struct mdp_superblock_1
*sb
;
1032 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1036 * Calculate the position of the superblock in 512byte sectors.
1037 * It is always aligned to a 4K boundary and
1038 * depeding on minor_version, it can be:
1039 * 0: At least 8K, but less than 12K, from end of device
1040 * 1: At start of device
1041 * 2: 4K from start of device.
1043 switch(minor_version
) {
1045 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1047 sb_start
&= ~(sector_t
)(4*2-1);
1058 rdev
->sb_start
= sb_start
;
1060 /* superblock is rarely larger than 1K, but it can be larger,
1061 * and it is safe to read 4k, so we do that
1063 ret
= read_disk_sb(rdev
, 4096);
1064 if (ret
) return ret
;
1067 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1069 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1070 sb
->major_version
!= cpu_to_le32(1) ||
1071 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1072 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1073 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1076 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1077 printk("md: invalid superblock checksum on %s\n",
1078 bdevname(rdev
->bdev
,b
));
1081 if (le64_to_cpu(sb
->data_size
) < 10) {
1082 printk("md: data_size too small on %s\n",
1083 bdevname(rdev
->bdev
,b
));
1086 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)) {
1087 if (sb
->level
!= cpu_to_le32(1) &&
1088 sb
->level
!= cpu_to_le32(4) &&
1089 sb
->level
!= cpu_to_le32(5) &&
1090 sb
->level
!= cpu_to_le32(6) &&
1091 sb
->level
!= cpu_to_le32(10)) {
1093 "md: bitmaps not supported for this level.\n");
1098 rdev
->preferred_minor
= 0xffff;
1099 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1100 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1102 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1103 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1104 if (rdev
->sb_size
& bmask
)
1105 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1108 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1111 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1114 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1120 struct mdp_superblock_1
*refsb
=
1121 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1123 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1124 sb
->level
!= refsb
->level
||
1125 sb
->layout
!= refsb
->layout
||
1126 sb
->chunksize
!= refsb
->chunksize
) {
1127 printk(KERN_WARNING
"md: %s has strangely different"
1128 " superblock to %s\n",
1129 bdevname(rdev
->bdev
,b
),
1130 bdevname(refdev
->bdev
,b2
));
1133 ev1
= le64_to_cpu(sb
->events
);
1134 ev2
= le64_to_cpu(refsb
->events
);
1142 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1144 rdev
->size
= rdev
->sb_start
/ 2;
1145 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1147 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1148 if (le32_to_cpu(sb
->chunksize
))
1149 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1151 if (le64_to_cpu(sb
->size
) > rdev
->size
*2)
1156 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1158 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1159 __u64 ev1
= le64_to_cpu(sb
->events
);
1161 rdev
->raid_disk
= -1;
1162 clear_bit(Faulty
, &rdev
->flags
);
1163 clear_bit(In_sync
, &rdev
->flags
);
1164 clear_bit(WriteMostly
, &rdev
->flags
);
1165 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1167 if (mddev
->raid_disks
== 0) {
1168 mddev
->major_version
= 1;
1169 mddev
->patch_version
= 0;
1170 mddev
->external
= 0;
1171 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1172 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1173 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1174 mddev
->level
= le32_to_cpu(sb
->level
);
1175 mddev
->clevel
[0] = 0;
1176 mddev
->layout
= le32_to_cpu(sb
->layout
);
1177 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1178 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1179 mddev
->events
= ev1
;
1180 mddev
->bitmap_offset
= 0;
1181 mddev
->default_bitmap_offset
= 1024 >> 9;
1183 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1184 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1186 mddev
->max_disks
= (4096-256)/2;
1188 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1189 mddev
->bitmap_file
== NULL
)
1190 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1192 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1193 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1194 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1195 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1196 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1197 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1199 mddev
->reshape_position
= MaxSector
;
1200 mddev
->delta_disks
= 0;
1201 mddev
->new_level
= mddev
->level
;
1202 mddev
->new_layout
= mddev
->layout
;
1203 mddev
->new_chunk
= mddev
->chunk_size
;
1206 } else if (mddev
->pers
== NULL
) {
1207 /* Insist of good event counter while assembling */
1209 if (ev1
< mddev
->events
)
1211 } else if (mddev
->bitmap
) {
1212 /* If adding to array with a bitmap, then we can accept an
1213 * older device, but not too old.
1215 if (ev1
< mddev
->bitmap
->events_cleared
)
1218 if (ev1
< mddev
->events
)
1219 /* just a hot-add of a new device, leave raid_disk at -1 */
1222 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1224 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1226 case 0xffff: /* spare */
1228 case 0xfffe: /* faulty */
1229 set_bit(Faulty
, &rdev
->flags
);
1232 if ((le32_to_cpu(sb
->feature_map
) &
1233 MD_FEATURE_RECOVERY_OFFSET
))
1234 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1236 set_bit(In_sync
, &rdev
->flags
);
1237 rdev
->raid_disk
= role
;
1240 if (sb
->devflags
& WriteMostly1
)
1241 set_bit(WriteMostly
, &rdev
->flags
);
1242 } else /* MULTIPATH are always insync */
1243 set_bit(In_sync
, &rdev
->flags
);
1248 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1250 struct mdp_superblock_1
*sb
;
1253 /* make rdev->sb match mddev and rdev data. */
1255 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1257 sb
->feature_map
= 0;
1259 sb
->recovery_offset
= cpu_to_le64(0);
1260 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1261 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1262 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1264 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1265 sb
->events
= cpu_to_le64(mddev
->events
);
1267 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1269 sb
->resync_offset
= cpu_to_le64(0);
1271 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1273 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1274 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1276 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1277 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1278 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1281 if (rdev
->raid_disk
>= 0 &&
1282 !test_bit(In_sync
, &rdev
->flags
) &&
1283 rdev
->recovery_offset
> 0) {
1284 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1285 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1288 if (mddev
->reshape_position
!= MaxSector
) {
1289 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1290 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1291 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1292 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1293 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1294 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1298 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1299 if (rdev2
->desc_nr
+1 > max_dev
)
1300 max_dev
= rdev2
->desc_nr
+1;
1302 if (max_dev
> le32_to_cpu(sb
->max_dev
))
1303 sb
->max_dev
= cpu_to_le32(max_dev
);
1304 for (i
=0; i
<max_dev
;i
++)
1305 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1307 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1309 if (test_bit(Faulty
, &rdev2
->flags
))
1310 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1311 else if (test_bit(In_sync
, &rdev2
->flags
))
1312 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1313 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1314 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1316 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1319 sb
->sb_csum
= calc_sb_1_csum(sb
);
1322 static unsigned long long
1323 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1325 struct mdp_superblock_1
*sb
;
1326 sector_t max_sectors
;
1327 if (num_sectors
&& num_sectors
< rdev
->mddev
->size
* 2)
1328 return 0; /* component must fit device */
1329 if (rdev
->sb_start
< rdev
->data_offset
) {
1330 /* minor versions 1 and 2; superblock before data */
1331 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1332 max_sectors
-= rdev
->data_offset
;
1333 if (!num_sectors
|| num_sectors
> max_sectors
)
1334 num_sectors
= max_sectors
;
1335 } else if (rdev
->mddev
->bitmap_offset
) {
1336 /* minor version 0 with bitmap we can't move */
1339 /* minor version 0; superblock after data */
1341 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1342 sb_start
&= ~(sector_t
)(4*2 - 1);
1343 max_sectors
= rdev
->size
* 2 + sb_start
- rdev
->sb_start
;
1344 if (!num_sectors
|| num_sectors
> max_sectors
)
1345 num_sectors
= max_sectors
;
1346 rdev
->sb_start
= sb_start
;
1348 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1349 sb
->data_size
= cpu_to_le64(num_sectors
);
1350 sb
->super_offset
= rdev
->sb_start
;
1351 sb
->sb_csum
= calc_sb_1_csum(sb
);
1352 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1354 md_super_wait(rdev
->mddev
);
1355 return num_sectors
/ 2; /* kB for sysfs */
1358 static struct super_type super_types
[] = {
1361 .owner
= THIS_MODULE
,
1362 .load_super
= super_90_load
,
1363 .validate_super
= super_90_validate
,
1364 .sync_super
= super_90_sync
,
1365 .rdev_size_change
= super_90_rdev_size_change
,
1369 .owner
= THIS_MODULE
,
1370 .load_super
= super_1_load
,
1371 .validate_super
= super_1_validate
,
1372 .sync_super
= super_1_sync
,
1373 .rdev_size_change
= super_1_rdev_size_change
,
1377 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1379 mdk_rdev_t
*rdev
, *rdev2
;
1382 rdev_for_each_rcu(rdev
, mddev1
)
1383 rdev_for_each_rcu(rdev2
, mddev2
)
1384 if (rdev
->bdev
->bd_contains
==
1385 rdev2
->bdev
->bd_contains
) {
1393 static LIST_HEAD(pending_raid_disks
);
1395 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1397 char b
[BDEVNAME_SIZE
];
1407 /* prevent duplicates */
1408 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1411 /* make sure rdev->size exceeds mddev->size */
1412 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1414 /* Cannot change size, so fail
1415 * If mddev->level <= 0, then we don't care
1416 * about aligning sizes (e.g. linear)
1418 if (mddev
->level
> 0)
1421 mddev
->size
= rdev
->size
;
1424 /* Verify rdev->desc_nr is unique.
1425 * If it is -1, assign a free number, else
1426 * check number is not in use
1428 if (rdev
->desc_nr
< 0) {
1430 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1431 while (find_rdev_nr(mddev
, choice
))
1433 rdev
->desc_nr
= choice
;
1435 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1438 bdevname(rdev
->bdev
,b
);
1439 while ( (s
=strchr(b
, '/')) != NULL
)
1442 rdev
->mddev
= mddev
;
1443 printk(KERN_INFO
"md: bind<%s>\n", b
);
1445 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1448 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1449 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1450 kobject_del(&rdev
->kobj
);
1453 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1455 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1456 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1460 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1465 static void md_delayed_delete(struct work_struct
*ws
)
1467 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1468 kobject_del(&rdev
->kobj
);
1469 kobject_put(&rdev
->kobj
);
1472 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1474 char b
[BDEVNAME_SIZE
];
1479 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1480 list_del_rcu(&rdev
->same_set
);
1481 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1483 sysfs_remove_link(&rdev
->kobj
, "block");
1484 sysfs_put(rdev
->sysfs_state
);
1485 rdev
->sysfs_state
= NULL
;
1486 /* We need to delay this, otherwise we can deadlock when
1487 * writing to 'remove' to "dev/state". We also need
1488 * to delay it due to rcu usage.
1491 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1492 kobject_get(&rdev
->kobj
);
1493 schedule_work(&rdev
->del_work
);
1497 * prevent the device from being mounted, repartitioned or
1498 * otherwise reused by a RAID array (or any other kernel
1499 * subsystem), by bd_claiming the device.
1501 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1504 struct block_device
*bdev
;
1505 char b
[BDEVNAME_SIZE
];
1507 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1509 printk(KERN_ERR
"md: could not open %s.\n",
1510 __bdevname(dev
, b
));
1511 return PTR_ERR(bdev
);
1513 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1515 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1517 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1521 set_bit(AllReserved
, &rdev
->flags
);
1526 static void unlock_rdev(mdk_rdev_t
*rdev
)
1528 struct block_device
*bdev
= rdev
->bdev
;
1533 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1536 void md_autodetect_dev(dev_t dev
);
1538 static void export_rdev(mdk_rdev_t
* rdev
)
1540 char b
[BDEVNAME_SIZE
];
1541 printk(KERN_INFO
"md: export_rdev(%s)\n",
1542 bdevname(rdev
->bdev
,b
));
1547 if (test_bit(AutoDetected
, &rdev
->flags
))
1548 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1551 kobject_put(&rdev
->kobj
);
1554 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1556 unbind_rdev_from_array(rdev
);
1560 static void export_array(mddev_t
*mddev
)
1562 mdk_rdev_t
*rdev
, *tmp
;
1564 rdev_for_each(rdev
, tmp
, mddev
) {
1569 kick_rdev_from_array(rdev
);
1571 if (!list_empty(&mddev
->disks
))
1573 mddev
->raid_disks
= 0;
1574 mddev
->major_version
= 0;
1577 static void print_desc(mdp_disk_t
*desc
)
1579 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1580 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1583 static void print_sb_90(mdp_super_t
*sb
)
1588 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1589 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1590 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1592 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1593 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1594 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1595 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1596 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1597 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1598 sb
->failed_disks
, sb
->spare_disks
,
1599 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1602 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1605 desc
= sb
->disks
+ i
;
1606 if (desc
->number
|| desc
->major
|| desc
->minor
||
1607 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1608 printk(" D %2d: ", i
);
1612 printk(KERN_INFO
"md: THIS: ");
1613 print_desc(&sb
->this_disk
);
1616 static void print_sb_1(struct mdp_superblock_1
*sb
)
1620 uuid
= sb
->set_uuid
;
1621 printk(KERN_INFO
"md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1622 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1623 KERN_INFO
"md: Name: \"%s\" CT:%llu\n",
1624 le32_to_cpu(sb
->major_version
),
1625 le32_to_cpu(sb
->feature_map
),
1626 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1627 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1628 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1629 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1631 (unsigned long long)le64_to_cpu(sb
->ctime
)
1632 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1634 uuid
= sb
->device_uuid
;
1635 printk(KERN_INFO
"md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1637 KERN_INFO
"md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1638 ":%02x%02x%02x%02x%02x%02x\n"
1639 KERN_INFO
"md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1640 KERN_INFO
"md: (MaxDev:%u) \n",
1641 le32_to_cpu(sb
->level
),
1642 (unsigned long long)le64_to_cpu(sb
->size
),
1643 le32_to_cpu(sb
->raid_disks
),
1644 le32_to_cpu(sb
->layout
),
1645 le32_to_cpu(sb
->chunksize
),
1646 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1647 (unsigned long long)le64_to_cpu(sb
->data_size
),
1648 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1649 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1650 le32_to_cpu(sb
->dev_number
),
1651 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1652 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1653 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1654 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1656 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1657 (unsigned long long)le64_to_cpu(sb
->events
),
1658 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
1659 le32_to_cpu(sb
->sb_csum
),
1660 le32_to_cpu(sb
->max_dev
)
1664 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
1666 char b
[BDEVNAME_SIZE
];
1667 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1668 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1669 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1671 if (rdev
->sb_loaded
) {
1672 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
1673 switch (major_version
) {
1675 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
1678 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
1682 printk(KERN_INFO
"md: no rdev superblock!\n");
1685 static void md_print_devices(void)
1687 struct list_head
*tmp
;
1690 char b
[BDEVNAME_SIZE
];
1693 printk("md: **********************************\n");
1694 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1695 printk("md: **********************************\n");
1696 for_each_mddev(mddev
, tmp
) {
1699 bitmap_print_sb(mddev
->bitmap
);
1701 printk("%s: ", mdname(mddev
));
1702 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
1703 printk("<%s>", bdevname(rdev
->bdev
,b
));
1706 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
1707 print_rdev(rdev
, mddev
->major_version
);
1709 printk("md: **********************************\n");
1714 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1716 /* Update each superblock (in-memory image), but
1717 * if we are allowed to, skip spares which already
1718 * have the right event counter, or have one earlier
1719 * (which would mean they aren't being marked as dirty
1720 * with the rest of the array)
1724 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1725 if (rdev
->sb_events
== mddev
->events
||
1727 rdev
->raid_disk
< 0 &&
1728 (rdev
->sb_events
&1)==0 &&
1729 rdev
->sb_events
+1 == mddev
->events
)) {
1730 /* Don't update this superblock */
1731 rdev
->sb_loaded
= 2;
1733 super_types
[mddev
->major_version
].
1734 sync_super(mddev
, rdev
);
1735 rdev
->sb_loaded
= 1;
1740 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1746 if (mddev
->external
)
1749 spin_lock_irq(&mddev
->write_lock
);
1751 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1752 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1754 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1755 /* just a clean<-> dirty transition, possibly leave spares alone,
1756 * though if events isn't the right even/odd, we will have to do
1762 if (mddev
->degraded
)
1763 /* If the array is degraded, then skipping spares is both
1764 * dangerous and fairly pointless.
1765 * Dangerous because a device that was removed from the array
1766 * might have a event_count that still looks up-to-date,
1767 * so it can be re-added without a resync.
1768 * Pointless because if there are any spares to skip,
1769 * then a recovery will happen and soon that array won't
1770 * be degraded any more and the spare can go back to sleep then.
1774 sync_req
= mddev
->in_sync
;
1775 mddev
->utime
= get_seconds();
1777 /* If this is just a dirty<->clean transition, and the array is clean
1778 * and 'events' is odd, we can roll back to the previous clean state */
1780 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1781 && (mddev
->events
& 1)
1782 && mddev
->events
!= 1)
1785 /* otherwise we have to go forward and ... */
1787 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1788 /* .. if the array isn't clean, insist on an odd 'events' */
1789 if ((mddev
->events
&1)==0) {
1794 /* otherwise insist on an even 'events' (for clean states) */
1795 if ((mddev
->events
&1)) {
1802 if (!mddev
->events
) {
1804 * oops, this 64-bit counter should never wrap.
1805 * Either we are in around ~1 trillion A.C., assuming
1806 * 1 reboot per second, or we have a bug:
1813 * do not write anything to disk if using
1814 * nonpersistent superblocks
1816 if (!mddev
->persistent
) {
1817 if (!mddev
->external
)
1818 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1820 spin_unlock_irq(&mddev
->write_lock
);
1821 wake_up(&mddev
->sb_wait
);
1824 sync_sbs(mddev
, nospares
);
1825 spin_unlock_irq(&mddev
->write_lock
);
1828 "md: updating %s RAID superblock on device (in sync %d)\n",
1829 mdname(mddev
),mddev
->in_sync
);
1831 bitmap_update_sb(mddev
->bitmap
);
1832 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1833 char b
[BDEVNAME_SIZE
];
1834 dprintk(KERN_INFO
"md: ");
1835 if (rdev
->sb_loaded
!= 1)
1836 continue; /* no noise on spare devices */
1837 if (test_bit(Faulty
, &rdev
->flags
))
1838 dprintk("(skipping faulty ");
1840 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1841 if (!test_bit(Faulty
, &rdev
->flags
)) {
1842 md_super_write(mddev
,rdev
,
1843 rdev
->sb_start
, rdev
->sb_size
,
1845 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1846 bdevname(rdev
->bdev
,b
),
1847 (unsigned long long)rdev
->sb_start
);
1848 rdev
->sb_events
= mddev
->events
;
1852 if (mddev
->level
== LEVEL_MULTIPATH
)
1853 /* only need to write one superblock... */
1856 md_super_wait(mddev
);
1857 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1859 spin_lock_irq(&mddev
->write_lock
);
1860 if (mddev
->in_sync
!= sync_req
||
1861 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
1862 /* have to write it out again */
1863 spin_unlock_irq(&mddev
->write_lock
);
1866 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1867 spin_unlock_irq(&mddev
->write_lock
);
1868 wake_up(&mddev
->sb_wait
);
1872 /* words written to sysfs files may, or may not, be \n terminated.
1873 * We want to accept with case. For this we use cmd_match.
1875 static int cmd_match(const char *cmd
, const char *str
)
1877 /* See if cmd, written into a sysfs file, matches
1878 * str. They must either be the same, or cmd can
1879 * have a trailing newline
1881 while (*cmd
&& *str
&& *cmd
== *str
) {
1892 struct rdev_sysfs_entry
{
1893 struct attribute attr
;
1894 ssize_t (*show
)(mdk_rdev_t
*, char *);
1895 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1899 state_show(mdk_rdev_t
*rdev
, char *page
)
1904 if (test_bit(Faulty
, &rdev
->flags
)) {
1905 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1908 if (test_bit(In_sync
, &rdev
->flags
)) {
1909 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1912 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1913 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1916 if (test_bit(Blocked
, &rdev
->flags
)) {
1917 len
+= sprintf(page
+len
, "%sblocked", sep
);
1920 if (!test_bit(Faulty
, &rdev
->flags
) &&
1921 !test_bit(In_sync
, &rdev
->flags
)) {
1922 len
+= sprintf(page
+len
, "%sspare", sep
);
1925 return len
+sprintf(page
+len
, "\n");
1929 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1932 * faulty - simulates and error
1933 * remove - disconnects the device
1934 * writemostly - sets write_mostly
1935 * -writemostly - clears write_mostly
1936 * blocked - sets the Blocked flag
1937 * -blocked - clears the Blocked flag
1940 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1941 md_error(rdev
->mddev
, rdev
);
1943 } else if (cmd_match(buf
, "remove")) {
1944 if (rdev
->raid_disk
>= 0)
1947 mddev_t
*mddev
= rdev
->mddev
;
1948 kick_rdev_from_array(rdev
);
1950 md_update_sb(mddev
, 1);
1951 md_new_event(mddev
);
1954 } else if (cmd_match(buf
, "writemostly")) {
1955 set_bit(WriteMostly
, &rdev
->flags
);
1957 } else if (cmd_match(buf
, "-writemostly")) {
1958 clear_bit(WriteMostly
, &rdev
->flags
);
1960 } else if (cmd_match(buf
, "blocked")) {
1961 set_bit(Blocked
, &rdev
->flags
);
1963 } else if (cmd_match(buf
, "-blocked")) {
1964 clear_bit(Blocked
, &rdev
->flags
);
1965 wake_up(&rdev
->blocked_wait
);
1966 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
1967 md_wakeup_thread(rdev
->mddev
->thread
);
1971 if (!err
&& rdev
->sysfs_state
)
1972 sysfs_notify_dirent(rdev
->sysfs_state
);
1973 return err
? err
: len
;
1975 static struct rdev_sysfs_entry rdev_state
=
1976 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
1979 errors_show(mdk_rdev_t
*rdev
, char *page
)
1981 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1985 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1988 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1989 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1990 atomic_set(&rdev
->corrected_errors
, n
);
1995 static struct rdev_sysfs_entry rdev_errors
=
1996 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
1999 slot_show(mdk_rdev_t
*rdev
, char *page
)
2001 if (rdev
->raid_disk
< 0)
2002 return sprintf(page
, "none\n");
2004 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2008 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2013 int slot
= simple_strtoul(buf
, &e
, 10);
2014 if (strncmp(buf
, "none", 4)==0)
2016 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2018 if (rdev
->mddev
->pers
&& slot
== -1) {
2019 /* Setting 'slot' on an active array requires also
2020 * updating the 'rd%d' link, and communicating
2021 * with the personality with ->hot_*_disk.
2022 * For now we only support removing
2023 * failed/spare devices. This normally happens automatically,
2024 * but not when the metadata is externally managed.
2026 if (rdev
->raid_disk
== -1)
2028 /* personality does all needed checks */
2029 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2031 err
= rdev
->mddev
->pers
->
2032 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2035 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2036 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2037 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2038 md_wakeup_thread(rdev
->mddev
->thread
);
2039 } else if (rdev
->mddev
->pers
) {
2041 /* Activating a spare .. or possibly reactivating
2042 * if we every get bitmaps working here.
2045 if (rdev
->raid_disk
!= -1)
2048 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2051 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2052 if (rdev2
->raid_disk
== slot
)
2055 rdev
->raid_disk
= slot
;
2056 if (test_bit(In_sync
, &rdev
->flags
))
2057 rdev
->saved_raid_disk
= slot
;
2059 rdev
->saved_raid_disk
= -1;
2060 err
= rdev
->mddev
->pers
->
2061 hot_add_disk(rdev
->mddev
, rdev
);
2063 rdev
->raid_disk
= -1;
2066 sysfs_notify_dirent(rdev
->sysfs_state
);
2067 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2068 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2070 "md: cannot register "
2072 nm
, mdname(rdev
->mddev
));
2074 /* don't wakeup anyone, leave that to userspace. */
2076 if (slot
>= rdev
->mddev
->raid_disks
)
2078 rdev
->raid_disk
= slot
;
2079 /* assume it is working */
2080 clear_bit(Faulty
, &rdev
->flags
);
2081 clear_bit(WriteMostly
, &rdev
->flags
);
2082 set_bit(In_sync
, &rdev
->flags
);
2083 sysfs_notify_dirent(rdev
->sysfs_state
);
2089 static struct rdev_sysfs_entry rdev_slot
=
2090 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2093 offset_show(mdk_rdev_t
*rdev
, char *page
)
2095 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2099 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2102 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2103 if (e
==buf
|| (*e
&& *e
!= '\n'))
2105 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2107 if (rdev
->size
&& rdev
->mddev
->external
)
2108 /* Must set offset before size, so overlap checks
2111 rdev
->data_offset
= offset
;
2115 static struct rdev_sysfs_entry rdev_offset
=
2116 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2119 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2121 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
2124 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2126 /* check if two start/length pairs overlap */
2135 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2137 unsigned long long size
;
2138 unsigned long long oldsize
= rdev
->size
;
2139 mddev_t
*my_mddev
= rdev
->mddev
;
2141 if (strict_strtoull(buf
, 10, &size
) < 0)
2143 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2144 if (my_mddev
->persistent
) {
2145 size
= super_types
[my_mddev
->major_version
].
2146 rdev_size_change(rdev
, size
* 2);
2150 size
= (rdev
->bdev
->bd_inode
->i_size
>> 10);
2151 size
-= rdev
->data_offset
/2;
2154 if (size
< my_mddev
->size
)
2155 return -EINVAL
; /* component must fit device */
2158 if (size
> oldsize
&& my_mddev
->external
) {
2159 /* need to check that all other rdevs with the same ->bdev
2160 * do not overlap. We need to unlock the mddev to avoid
2161 * a deadlock. We have already changed rdev->size, and if
2162 * we have to change it back, we will have the lock again.
2166 struct list_head
*tmp
;
2168 mddev_unlock(my_mddev
);
2169 for_each_mddev(mddev
, tmp
) {
2173 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2174 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2175 (rdev
->bdev
== rdev2
->bdev
&&
2177 overlaps(rdev
->data_offset
, rdev
->size
* 2,
2179 rdev2
->size
* 2))) {
2183 mddev_unlock(mddev
);
2189 mddev_lock(my_mddev
);
2191 /* Someone else could have slipped in a size
2192 * change here, but doing so is just silly.
2193 * We put oldsize back because we *know* it is
2194 * safe, and trust userspace not to race with
2197 rdev
->size
= oldsize
;
2204 static struct rdev_sysfs_entry rdev_size
=
2205 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2207 static struct attribute
*rdev_default_attrs
[] = {
2216 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2218 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2219 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2220 mddev_t
*mddev
= rdev
->mddev
;
2226 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2228 if (rdev
->mddev
== NULL
)
2231 rv
= entry
->show(rdev
, page
);
2232 mddev_unlock(mddev
);
2238 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2239 const char *page
, size_t length
)
2241 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2242 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2244 mddev_t
*mddev
= rdev
->mddev
;
2248 if (!capable(CAP_SYS_ADMIN
))
2250 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2252 if (rdev
->mddev
== NULL
)
2255 rv
= entry
->store(rdev
, page
, length
);
2256 mddev_unlock(mddev
);
2261 static void rdev_free(struct kobject
*ko
)
2263 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2266 static struct sysfs_ops rdev_sysfs_ops
= {
2267 .show
= rdev_attr_show
,
2268 .store
= rdev_attr_store
,
2270 static struct kobj_type rdev_ktype
= {
2271 .release
= rdev_free
,
2272 .sysfs_ops
= &rdev_sysfs_ops
,
2273 .default_attrs
= rdev_default_attrs
,
2277 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2279 * mark the device faulty if:
2281 * - the device is nonexistent (zero size)
2282 * - the device has no valid superblock
2284 * a faulty rdev _never_ has rdev->sb set.
2286 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2288 char b
[BDEVNAME_SIZE
];
2293 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2295 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2296 return ERR_PTR(-ENOMEM
);
2299 if ((err
= alloc_disk_sb(rdev
)))
2302 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2306 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2309 rdev
->saved_raid_disk
= -1;
2310 rdev
->raid_disk
= -1;
2312 rdev
->data_offset
= 0;
2313 rdev
->sb_events
= 0;
2314 atomic_set(&rdev
->nr_pending
, 0);
2315 atomic_set(&rdev
->read_errors
, 0);
2316 atomic_set(&rdev
->corrected_errors
, 0);
2318 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2321 "md: %s has zero or unknown size, marking faulty!\n",
2322 bdevname(rdev
->bdev
,b
));
2327 if (super_format
>= 0) {
2328 err
= super_types
[super_format
].
2329 load_super(rdev
, NULL
, super_minor
);
2330 if (err
== -EINVAL
) {
2332 "md: %s does not have a valid v%d.%d "
2333 "superblock, not importing!\n",
2334 bdevname(rdev
->bdev
,b
),
2335 super_format
, super_minor
);
2340 "md: could not read %s's sb, not importing!\n",
2341 bdevname(rdev
->bdev
,b
));
2346 INIT_LIST_HEAD(&rdev
->same_set
);
2347 init_waitqueue_head(&rdev
->blocked_wait
);
2352 if (rdev
->sb_page
) {
2358 return ERR_PTR(err
);
2362 * Check a full RAID array for plausibility
2366 static void analyze_sbs(mddev_t
* mddev
)
2369 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2370 char b
[BDEVNAME_SIZE
];
2373 rdev_for_each(rdev
, tmp
, mddev
)
2374 switch (super_types
[mddev
->major_version
].
2375 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2383 "md: fatal superblock inconsistency in %s"
2384 " -- removing from array\n",
2385 bdevname(rdev
->bdev
,b
));
2386 kick_rdev_from_array(rdev
);
2390 super_types
[mddev
->major_version
].
2391 validate_super(mddev
, freshest
);
2394 rdev_for_each(rdev
, tmp
, mddev
) {
2395 if (rdev
!= freshest
)
2396 if (super_types
[mddev
->major_version
].
2397 validate_super(mddev
, rdev
)) {
2398 printk(KERN_WARNING
"md: kicking non-fresh %s"
2400 bdevname(rdev
->bdev
,b
));
2401 kick_rdev_from_array(rdev
);
2404 if (mddev
->level
== LEVEL_MULTIPATH
) {
2405 rdev
->desc_nr
= i
++;
2406 rdev
->raid_disk
= rdev
->desc_nr
;
2407 set_bit(In_sync
, &rdev
->flags
);
2408 } else if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2409 rdev
->raid_disk
= -1;
2410 clear_bit(In_sync
, &rdev
->flags
);
2416 if (mddev
->recovery_cp
!= MaxSector
&&
2418 printk(KERN_ERR
"md: %s: raid array is not clean"
2419 " -- starting background reconstruction\n",
2424 static void md_safemode_timeout(unsigned long data
);
2427 safe_delay_show(mddev_t
*mddev
, char *page
)
2429 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2430 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2433 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2441 /* remove a period, and count digits after it */
2442 if (len
>= sizeof(buf
))
2444 strlcpy(buf
, cbuf
, sizeof(buf
));
2445 for (i
=0; i
<len
; i
++) {
2447 if (isdigit(buf
[i
])) {
2452 } else if (buf
[i
] == '.') {
2457 if (strict_strtoul(buf
, 10, &msec
) < 0)
2459 msec
= (msec
* 1000) / scale
;
2461 mddev
->safemode_delay
= 0;
2463 unsigned long old_delay
= mddev
->safemode_delay
;
2464 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2465 if (mddev
->safemode_delay
== 0)
2466 mddev
->safemode_delay
= 1;
2467 if (mddev
->safemode_delay
< old_delay
)
2468 md_safemode_timeout((unsigned long)mddev
);
2472 static struct md_sysfs_entry md_safe_delay
=
2473 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2476 level_show(mddev_t
*mddev
, char *page
)
2478 struct mdk_personality
*p
= mddev
->pers
;
2480 return sprintf(page
, "%s\n", p
->name
);
2481 else if (mddev
->clevel
[0])
2482 return sprintf(page
, "%s\n", mddev
->clevel
);
2483 else if (mddev
->level
!= LEVEL_NONE
)
2484 return sprintf(page
, "%d\n", mddev
->level
);
2490 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2497 if (len
>= sizeof(mddev
->clevel
))
2499 strncpy(mddev
->clevel
, buf
, len
);
2500 if (mddev
->clevel
[len
-1] == '\n')
2502 mddev
->clevel
[len
] = 0;
2503 mddev
->level
= LEVEL_NONE
;
2507 static struct md_sysfs_entry md_level
=
2508 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2512 layout_show(mddev_t
*mddev
, char *page
)
2514 /* just a number, not meaningful for all levels */
2515 if (mddev
->reshape_position
!= MaxSector
&&
2516 mddev
->layout
!= mddev
->new_layout
)
2517 return sprintf(page
, "%d (%d)\n",
2518 mddev
->new_layout
, mddev
->layout
);
2519 return sprintf(page
, "%d\n", mddev
->layout
);
2523 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2526 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2528 if (!*buf
|| (*e
&& *e
!= '\n'))
2533 if (mddev
->reshape_position
!= MaxSector
)
2534 mddev
->new_layout
= n
;
2539 static struct md_sysfs_entry md_layout
=
2540 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2544 raid_disks_show(mddev_t
*mddev
, char *page
)
2546 if (mddev
->raid_disks
== 0)
2548 if (mddev
->reshape_position
!= MaxSector
&&
2549 mddev
->delta_disks
!= 0)
2550 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
2551 mddev
->raid_disks
- mddev
->delta_disks
);
2552 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2555 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2558 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2562 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2564 if (!*buf
|| (*e
&& *e
!= '\n'))
2568 rv
= update_raid_disks(mddev
, n
);
2569 else if (mddev
->reshape_position
!= MaxSector
) {
2570 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
2571 mddev
->delta_disks
= n
- olddisks
;
2572 mddev
->raid_disks
= n
;
2574 mddev
->raid_disks
= n
;
2575 return rv
? rv
: len
;
2577 static struct md_sysfs_entry md_raid_disks
=
2578 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2581 chunk_size_show(mddev_t
*mddev
, char *page
)
2583 if (mddev
->reshape_position
!= MaxSector
&&
2584 mddev
->chunk_size
!= mddev
->new_chunk
)
2585 return sprintf(page
, "%d (%d)\n", mddev
->new_chunk
,
2587 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2591 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2593 /* can only set chunk_size if array is not yet active */
2595 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2597 if (!*buf
|| (*e
&& *e
!= '\n'))
2602 else if (mddev
->reshape_position
!= MaxSector
)
2603 mddev
->new_chunk
= n
;
2605 mddev
->chunk_size
= n
;
2608 static struct md_sysfs_entry md_chunk_size
=
2609 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2612 resync_start_show(mddev_t
*mddev
, char *page
)
2614 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2618 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2621 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2625 if (!*buf
|| (*e
&& *e
!= '\n'))
2628 mddev
->recovery_cp
= n
;
2631 static struct md_sysfs_entry md_resync_start
=
2632 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2635 * The array state can be:
2638 * No devices, no size, no level
2639 * Equivalent to STOP_ARRAY ioctl
2641 * May have some settings, but array is not active
2642 * all IO results in error
2643 * When written, doesn't tear down array, but just stops it
2644 * suspended (not supported yet)
2645 * All IO requests will block. The array can be reconfigured.
2646 * Writing this, if accepted, will block until array is quiescent
2648 * no resync can happen. no superblocks get written.
2649 * write requests fail
2651 * like readonly, but behaves like 'clean' on a write request.
2653 * clean - no pending writes, but otherwise active.
2654 * When written to inactive array, starts without resync
2655 * If a write request arrives then
2656 * if metadata is known, mark 'dirty' and switch to 'active'.
2657 * if not known, block and switch to write-pending
2658 * If written to an active array that has pending writes, then fails.
2660 * fully active: IO and resync can be happening.
2661 * When written to inactive array, starts with resync
2664 * clean, but writes are blocked waiting for 'active' to be written.
2667 * like active, but no writes have been seen for a while (100msec).
2670 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2671 write_pending
, active_idle
, bad_word
};
2672 static char *array_states
[] = {
2673 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2674 "write-pending", "active-idle", NULL
};
2676 static int match_word(const char *word
, char **list
)
2679 for (n
=0; list
[n
]; n
++)
2680 if (cmd_match(word
, list
[n
]))
2686 array_state_show(mddev_t
*mddev
, char *page
)
2688 enum array_state st
= inactive
;
2701 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2703 else if (mddev
->safemode
)
2709 if (list_empty(&mddev
->disks
) &&
2710 mddev
->raid_disks
== 0 &&
2716 return sprintf(page
, "%s\n", array_states
[st
]);
2719 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
2720 static int do_md_run(mddev_t
* mddev
);
2721 static int restart_array(mddev_t
*mddev
);
2724 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2727 enum array_state st
= match_word(buf
, array_states
);
2732 /* stopping an active array */
2733 if (atomic_read(&mddev
->openers
) > 0)
2735 err
= do_md_stop(mddev
, 0, 0);
2738 /* stopping an active array */
2740 if (atomic_read(&mddev
->openers
) > 0)
2742 err
= do_md_stop(mddev
, 2, 0);
2744 err
= 0; /* already inactive */
2747 break; /* not supported yet */
2750 err
= do_md_stop(mddev
, 1, 0);
2753 set_disk_ro(mddev
->gendisk
, 1);
2754 err
= do_md_run(mddev
);
2760 err
= do_md_stop(mddev
, 1, 0);
2761 else if (mddev
->ro
== 1)
2762 err
= restart_array(mddev
);
2765 set_disk_ro(mddev
->gendisk
, 0);
2769 err
= do_md_run(mddev
);
2774 restart_array(mddev
);
2775 spin_lock_irq(&mddev
->write_lock
);
2776 if (atomic_read(&mddev
->writes_pending
) == 0) {
2777 if (mddev
->in_sync
== 0) {
2779 if (mddev
->safemode
== 1)
2780 mddev
->safemode
= 0;
2781 if (mddev
->persistent
)
2782 set_bit(MD_CHANGE_CLEAN
,
2788 spin_unlock_irq(&mddev
->write_lock
);
2791 mddev
->recovery_cp
= MaxSector
;
2792 err
= do_md_run(mddev
);
2797 restart_array(mddev
);
2798 if (mddev
->external
)
2799 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2800 wake_up(&mddev
->sb_wait
);
2804 set_disk_ro(mddev
->gendisk
, 0);
2805 err
= do_md_run(mddev
);
2810 /* these cannot be set */
2816 sysfs_notify_dirent(mddev
->sysfs_state
);
2820 static struct md_sysfs_entry md_array_state
=
2821 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
2824 null_show(mddev_t
*mddev
, char *page
)
2830 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2832 /* buf must be %d:%d\n? giving major and minor numbers */
2833 /* The new device is added to the array.
2834 * If the array has a persistent superblock, we read the
2835 * superblock to initialise info and check validity.
2836 * Otherwise, only checking done is that in bind_rdev_to_array,
2837 * which mainly checks size.
2840 int major
= simple_strtoul(buf
, &e
, 10);
2846 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2848 minor
= simple_strtoul(e
+1, &e
, 10);
2849 if (*e
&& *e
!= '\n')
2851 dev
= MKDEV(major
, minor
);
2852 if (major
!= MAJOR(dev
) ||
2853 minor
!= MINOR(dev
))
2857 if (mddev
->persistent
) {
2858 rdev
= md_import_device(dev
, mddev
->major_version
,
2859 mddev
->minor_version
);
2860 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2861 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2862 mdk_rdev_t
, same_set
);
2863 err
= super_types
[mddev
->major_version
]
2864 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2868 } else if (mddev
->external
)
2869 rdev
= md_import_device(dev
, -2, -1);
2871 rdev
= md_import_device(dev
, -1, -1);
2874 return PTR_ERR(rdev
);
2875 err
= bind_rdev_to_array(rdev
, mddev
);
2879 return err
? err
: len
;
2882 static struct md_sysfs_entry md_new_device
=
2883 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
2886 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2889 unsigned long chunk
, end_chunk
;
2893 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2895 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
2896 if (buf
== end
) break;
2897 if (*end
== '-') { /* range */
2899 end_chunk
= simple_strtoul(buf
, &end
, 0);
2900 if (buf
== end
) break;
2902 if (*end
&& !isspace(*end
)) break;
2903 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
2905 while (isspace(*buf
)) buf
++;
2907 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
2912 static struct md_sysfs_entry md_bitmap
=
2913 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
2916 size_show(mddev_t
*mddev
, char *page
)
2918 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2921 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
2924 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2926 /* If array is inactive, we can reduce the component size, but
2927 * not increase it (except from 0).
2928 * If array is active, we can try an on-line resize
2932 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2933 if (!*buf
|| *buf
== '\n' ||
2938 err
= update_size(mddev
, size
* 2);
2939 md_update_sb(mddev
, 1);
2941 if (mddev
->size
== 0 ||
2947 return err
? err
: len
;
2950 static struct md_sysfs_entry md_size
=
2951 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
2956 * 'none' for arrays with no metadata (good luck...)
2957 * 'external' for arrays with externally managed metadata,
2958 * or N.M for internally known formats
2961 metadata_show(mddev_t
*mddev
, char *page
)
2963 if (mddev
->persistent
)
2964 return sprintf(page
, "%d.%d\n",
2965 mddev
->major_version
, mddev
->minor_version
);
2966 else if (mddev
->external
)
2967 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
2969 return sprintf(page
, "none\n");
2973 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2977 /* Changing the details of 'external' metadata is
2978 * always permitted. Otherwise there must be
2979 * no devices attached to the array.
2981 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
2983 else if (!list_empty(&mddev
->disks
))
2986 if (cmd_match(buf
, "none")) {
2987 mddev
->persistent
= 0;
2988 mddev
->external
= 0;
2989 mddev
->major_version
= 0;
2990 mddev
->minor_version
= 90;
2993 if (strncmp(buf
, "external:", 9) == 0) {
2994 size_t namelen
= len
-9;
2995 if (namelen
>= sizeof(mddev
->metadata_type
))
2996 namelen
= sizeof(mddev
->metadata_type
)-1;
2997 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
2998 mddev
->metadata_type
[namelen
] = 0;
2999 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3000 mddev
->metadata_type
[--namelen
] = 0;
3001 mddev
->persistent
= 0;
3002 mddev
->external
= 1;
3003 mddev
->major_version
= 0;
3004 mddev
->minor_version
= 90;
3007 major
= simple_strtoul(buf
, &e
, 10);
3008 if (e
==buf
|| *e
!= '.')
3011 minor
= simple_strtoul(buf
, &e
, 10);
3012 if (e
==buf
|| (*e
&& *e
!= '\n') )
3014 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3016 mddev
->major_version
= major
;
3017 mddev
->minor_version
= minor
;
3018 mddev
->persistent
= 1;
3019 mddev
->external
= 0;
3023 static struct md_sysfs_entry md_metadata
=
3024 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3027 action_show(mddev_t
*mddev
, char *page
)
3029 char *type
= "idle";
3030 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3031 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3032 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3034 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3035 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3037 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3041 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3044 return sprintf(page
, "%s\n", type
);
3048 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3050 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3053 if (cmd_match(page
, "idle")) {
3054 if (mddev
->sync_thread
) {
3055 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3056 md_unregister_thread(mddev
->sync_thread
);
3057 mddev
->sync_thread
= NULL
;
3058 mddev
->recovery
= 0;
3060 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3061 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3063 else if (cmd_match(page
, "resync"))
3064 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3065 else if (cmd_match(page
, "recover")) {
3066 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3067 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3068 } else if (cmd_match(page
, "reshape")) {
3070 if (mddev
->pers
->start_reshape
== NULL
)
3072 err
= mddev
->pers
->start_reshape(mddev
);
3075 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3077 if (cmd_match(page
, "check"))
3078 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3079 else if (!cmd_match(page
, "repair"))
3081 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3082 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3084 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3085 md_wakeup_thread(mddev
->thread
);
3086 sysfs_notify_dirent(mddev
->sysfs_action
);
3091 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3093 return sprintf(page
, "%llu\n",
3094 (unsigned long long) mddev
->resync_mismatches
);
3097 static struct md_sysfs_entry md_scan_mode
=
3098 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3101 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3104 sync_min_show(mddev_t
*mddev
, char *page
)
3106 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3107 mddev
->sync_speed_min
? "local": "system");
3111 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3115 if (strncmp(buf
, "system", 6)==0) {
3116 mddev
->sync_speed_min
= 0;
3119 min
= simple_strtoul(buf
, &e
, 10);
3120 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3122 mddev
->sync_speed_min
= min
;
3126 static struct md_sysfs_entry md_sync_min
=
3127 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3130 sync_max_show(mddev_t
*mddev
, char *page
)
3132 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3133 mddev
->sync_speed_max
? "local": "system");
3137 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3141 if (strncmp(buf
, "system", 6)==0) {
3142 mddev
->sync_speed_max
= 0;
3145 max
= simple_strtoul(buf
, &e
, 10);
3146 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3148 mddev
->sync_speed_max
= max
;
3152 static struct md_sysfs_entry md_sync_max
=
3153 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3156 degraded_show(mddev_t
*mddev
, char *page
)
3158 return sprintf(page
, "%d\n", mddev
->degraded
);
3160 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3163 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3165 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3169 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3173 if (strict_strtol(buf
, 10, &n
))
3176 if (n
!= 0 && n
!= 1)
3179 mddev
->parallel_resync
= n
;
3181 if (mddev
->sync_thread
)
3182 wake_up(&resync_wait
);
3187 /* force parallel resync, even with shared block devices */
3188 static struct md_sysfs_entry md_sync_force_parallel
=
3189 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3190 sync_force_parallel_show
, sync_force_parallel_store
);
3193 sync_speed_show(mddev_t
*mddev
, char *page
)
3195 unsigned long resync
, dt
, db
;
3196 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3197 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3199 db
= resync
- mddev
->resync_mark_cnt
;
3200 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3203 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3206 sync_completed_show(mddev_t
*mddev
, char *page
)
3208 unsigned long max_blocks
, resync
;
3210 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3211 max_blocks
= mddev
->resync_max_sectors
;
3213 max_blocks
= mddev
->size
<< 1;
3215 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
3216 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
3219 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3222 min_sync_show(mddev_t
*mddev
, char *page
)
3224 return sprintf(page
, "%llu\n",
3225 (unsigned long long)mddev
->resync_min
);
3228 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3230 unsigned long long min
;
3231 if (strict_strtoull(buf
, 10, &min
))
3233 if (min
> mddev
->resync_max
)
3235 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3238 /* Must be a multiple of chunk_size */
3239 if (mddev
->chunk_size
) {
3240 if (min
& (sector_t
)((mddev
->chunk_size
>>9)-1))
3243 mddev
->resync_min
= min
;
3248 static struct md_sysfs_entry md_min_sync
=
3249 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3252 max_sync_show(mddev_t
*mddev
, char *page
)
3254 if (mddev
->resync_max
== MaxSector
)
3255 return sprintf(page
, "max\n");
3257 return sprintf(page
, "%llu\n",
3258 (unsigned long long)mddev
->resync_max
);
3261 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3263 if (strncmp(buf
, "max", 3) == 0)
3264 mddev
->resync_max
= MaxSector
;
3266 unsigned long long max
;
3267 if (strict_strtoull(buf
, 10, &max
))
3269 if (max
< mddev
->resync_min
)
3271 if (max
< mddev
->resync_max
&&
3272 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3275 /* Must be a multiple of chunk_size */
3276 if (mddev
->chunk_size
) {
3277 if (max
& (sector_t
)((mddev
->chunk_size
>>9)-1))
3280 mddev
->resync_max
= max
;
3282 wake_up(&mddev
->recovery_wait
);
3286 static struct md_sysfs_entry md_max_sync
=
3287 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3290 suspend_lo_show(mddev_t
*mddev
, char *page
)
3292 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3296 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3299 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3301 if (mddev
->pers
->quiesce
== NULL
)
3303 if (buf
== e
|| (*e
&& *e
!= '\n'))
3305 if (new >= mddev
->suspend_hi
||
3306 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3307 mddev
->suspend_lo
= new;
3308 mddev
->pers
->quiesce(mddev
, 2);
3313 static struct md_sysfs_entry md_suspend_lo
=
3314 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3318 suspend_hi_show(mddev_t
*mddev
, char *page
)
3320 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3324 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3327 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3329 if (mddev
->pers
->quiesce
== NULL
)
3331 if (buf
== e
|| (*e
&& *e
!= '\n'))
3333 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3334 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3335 mddev
->suspend_hi
= new;
3336 mddev
->pers
->quiesce(mddev
, 1);
3337 mddev
->pers
->quiesce(mddev
, 0);
3342 static struct md_sysfs_entry md_suspend_hi
=
3343 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3346 reshape_position_show(mddev_t
*mddev
, char *page
)
3348 if (mddev
->reshape_position
!= MaxSector
)
3349 return sprintf(page
, "%llu\n",
3350 (unsigned long long)mddev
->reshape_position
);
3351 strcpy(page
, "none\n");
3356 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3359 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3362 if (buf
== e
|| (*e
&& *e
!= '\n'))
3364 mddev
->reshape_position
= new;
3365 mddev
->delta_disks
= 0;
3366 mddev
->new_level
= mddev
->level
;
3367 mddev
->new_layout
= mddev
->layout
;
3368 mddev
->new_chunk
= mddev
->chunk_size
;
3372 static struct md_sysfs_entry md_reshape_position
=
3373 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3374 reshape_position_store
);
3377 static struct attribute
*md_default_attrs
[] = {
3380 &md_raid_disks
.attr
,
3381 &md_chunk_size
.attr
,
3383 &md_resync_start
.attr
,
3385 &md_new_device
.attr
,
3386 &md_safe_delay
.attr
,
3387 &md_array_state
.attr
,
3388 &md_reshape_position
.attr
,
3392 static struct attribute
*md_redundancy_attrs
[] = {
3394 &md_mismatches
.attr
,
3397 &md_sync_speed
.attr
,
3398 &md_sync_force_parallel
.attr
,
3399 &md_sync_completed
.attr
,
3402 &md_suspend_lo
.attr
,
3403 &md_suspend_hi
.attr
,
3408 static struct attribute_group md_redundancy_group
= {
3410 .attrs
= md_redundancy_attrs
,
3415 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3417 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3418 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3423 rv
= mddev_lock(mddev
);
3425 rv
= entry
->show(mddev
, page
);
3426 mddev_unlock(mddev
);
3432 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3433 const char *page
, size_t length
)
3435 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3436 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3441 if (!capable(CAP_SYS_ADMIN
))
3443 rv
= mddev_lock(mddev
);
3445 rv
= entry
->store(mddev
, page
, length
);
3446 mddev_unlock(mddev
);
3451 static void md_free(struct kobject
*ko
)
3453 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
3457 static struct sysfs_ops md_sysfs_ops
= {
3458 .show
= md_attr_show
,
3459 .store
= md_attr_store
,
3461 static struct kobj_type md_ktype
= {
3463 .sysfs_ops
= &md_sysfs_ops
,
3464 .default_attrs
= md_default_attrs
,
3469 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
3471 static DEFINE_MUTEX(disks_mutex
);
3472 mddev_t
*mddev
= mddev_find(dev
);
3473 struct gendisk
*disk
;
3474 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
3475 int shift
= partitioned
? MdpMinorShift
: 0;
3476 int unit
= MINOR(dev
) >> shift
;
3482 mutex_lock(&disks_mutex
);
3483 if (mddev
->gendisk
) {
3484 mutex_unlock(&disks_mutex
);
3489 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
3490 if (!mddev
->queue
) {
3491 mutex_unlock(&disks_mutex
);
3495 /* Can be unlocked because the queue is new: no concurrency */
3496 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
3498 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3500 disk
= alloc_disk(1 << shift
);
3502 mutex_unlock(&disks_mutex
);
3503 blk_cleanup_queue(mddev
->queue
);
3504 mddev
->queue
= NULL
;
3508 disk
->major
= MAJOR(dev
);
3509 disk
->first_minor
= unit
<< shift
;
3511 sprintf(disk
->disk_name
, "md_d%d", unit
);
3513 sprintf(disk
->disk_name
, "md%d", unit
);
3514 disk
->fops
= &md_fops
;
3515 disk
->private_data
= mddev
;
3516 disk
->queue
= mddev
->queue
;
3517 /* Allow extended partitions. This makes the
3518 * 'mdp' device redundant, but we can really
3521 disk
->flags
|= GENHD_FL_EXT_DEVT
;
3523 mddev
->gendisk
= disk
;
3524 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
3525 &disk_to_dev(disk
)->kobj
, "%s", "md");
3526 mutex_unlock(&disks_mutex
);
3528 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
3531 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
3532 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
3537 static void md_safemode_timeout(unsigned long data
)
3539 mddev_t
*mddev
= (mddev_t
*) data
;
3541 if (!atomic_read(&mddev
->writes_pending
)) {
3542 mddev
->safemode
= 1;
3543 if (mddev
->external
)
3544 sysfs_notify_dirent(mddev
->sysfs_state
);
3546 md_wakeup_thread(mddev
->thread
);
3549 static int start_dirty_degraded
;
3551 static int do_md_run(mddev_t
* mddev
)
3556 struct gendisk
*disk
;
3557 struct mdk_personality
*pers
;
3558 char b
[BDEVNAME_SIZE
];
3560 if (list_empty(&mddev
->disks
))
3561 /* cannot run an array with no devices.. */
3568 * Analyze all RAID superblock(s)
3570 if (!mddev
->raid_disks
) {
3571 if (!mddev
->persistent
)
3576 chunk_size
= mddev
->chunk_size
;
3579 if (chunk_size
> MAX_CHUNK_SIZE
) {
3580 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
3581 chunk_size
, MAX_CHUNK_SIZE
);
3585 * chunk-size has to be a power of 2
3587 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
3588 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
3592 /* devices must have minimum size of one chunk */
3593 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3594 if (test_bit(Faulty
, &rdev
->flags
))
3596 if (rdev
->size
< chunk_size
/ 1024) {
3598 "md: Dev %s smaller than chunk_size:"
3600 bdevname(rdev
->bdev
,b
),
3601 (unsigned long long)rdev
->size
,
3608 if (mddev
->level
!= LEVEL_NONE
)
3609 request_module("md-level-%d", mddev
->level
);
3610 else if (mddev
->clevel
[0])
3611 request_module("md-%s", mddev
->clevel
);
3614 * Drop all container device buffers, from now on
3615 * the only valid external interface is through the md
3618 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3619 if (test_bit(Faulty
, &rdev
->flags
))
3621 sync_blockdev(rdev
->bdev
);
3622 invalidate_bdev(rdev
->bdev
);
3624 /* perform some consistency tests on the device.
3625 * We don't want the data to overlap the metadata,
3626 * Internal Bitmap issues has handled elsewhere.
3628 if (rdev
->data_offset
< rdev
->sb_start
) {
3630 rdev
->data_offset
+ mddev
->size
*2
3632 printk("md: %s: data overlaps metadata\n",
3637 if (rdev
->sb_start
+ rdev
->sb_size
/512
3638 > rdev
->data_offset
) {
3639 printk("md: %s: metadata overlaps data\n",
3644 sysfs_notify_dirent(rdev
->sysfs_state
);
3647 md_probe(mddev
->unit
, NULL
, NULL
);
3648 disk
= mddev
->gendisk
;
3652 spin_lock(&pers_lock
);
3653 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3654 if (!pers
|| !try_module_get(pers
->owner
)) {
3655 spin_unlock(&pers_lock
);
3656 if (mddev
->level
!= LEVEL_NONE
)
3657 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3660 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3665 spin_unlock(&pers_lock
);
3666 mddev
->level
= pers
->level
;
3667 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3669 if (mddev
->reshape_position
!= MaxSector
&&
3670 pers
->start_reshape
== NULL
) {
3671 /* This personality cannot handle reshaping... */
3673 module_put(pers
->owner
);
3677 if (pers
->sync_request
) {
3678 /* Warn if this is a potentially silly
3681 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3685 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3686 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
3688 rdev
->bdev
->bd_contains
==
3689 rdev2
->bdev
->bd_contains
) {
3691 "%s: WARNING: %s appears to be"
3692 " on the same physical disk as"
3695 bdevname(rdev
->bdev
,b
),
3696 bdevname(rdev2
->bdev
,b2
));
3703 "True protection against single-disk"
3704 " failure might be compromised.\n");
3707 mddev
->recovery
= 0;
3708 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3709 mddev
->barriers_work
= 1;
3710 mddev
->ok_start_degraded
= start_dirty_degraded
;
3713 mddev
->ro
= 2; /* read-only, but switch on first write */
3715 err
= mddev
->pers
->run(mddev
);
3717 printk(KERN_ERR
"md: pers->run() failed ...\n");
3718 else if (mddev
->pers
->sync_request
) {
3719 err
= bitmap_create(mddev
);
3721 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3722 mdname(mddev
), err
);
3723 mddev
->pers
->stop(mddev
);
3727 module_put(mddev
->pers
->owner
);
3729 bitmap_destroy(mddev
);
3732 if (mddev
->pers
->sync_request
) {
3733 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3735 "md: cannot register extra attributes for %s\n",
3737 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3738 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3741 atomic_set(&mddev
->writes_pending
,0);
3742 mddev
->safemode
= 0;
3743 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3744 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3745 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3748 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3749 if (rdev
->raid_disk
>= 0) {
3751 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3752 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
3753 printk("md: cannot register %s for %s\n",
3757 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3760 md_update_sb(mddev
, 0);
3762 set_capacity(disk
, mddev
->array_sectors
);
3764 /* If we call blk_queue_make_request here, it will
3765 * re-initialise max_sectors etc which may have been
3766 * refined inside -> run. So just set the bits we need to set.
3767 * Most initialisation happended when we called
3768 * blk_queue_make_request(..., md_fail_request)
3771 mddev
->queue
->queuedata
= mddev
;
3772 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3774 /* If there is a partially-recovered drive we need to
3775 * start recovery here. If we leave it to md_check_recovery,
3776 * it will remove the drives and not do the right thing
3778 if (mddev
->degraded
&& !mddev
->sync_thread
) {
3780 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3781 if (rdev
->raid_disk
>= 0 &&
3782 !test_bit(In_sync
, &rdev
->flags
) &&
3783 !test_bit(Faulty
, &rdev
->flags
))
3784 /* complete an interrupted recovery */
3786 if (spares
&& mddev
->pers
->sync_request
) {
3787 mddev
->recovery
= 0;
3788 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3789 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3792 if (!mddev
->sync_thread
) {
3793 printk(KERN_ERR
"%s: could not start resync"
3796 /* leave the spares where they are, it shouldn't hurt */
3797 mddev
->recovery
= 0;
3801 md_wakeup_thread(mddev
->thread
);
3802 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
3805 md_new_event(mddev
);
3806 sysfs_notify_dirent(mddev
->sysfs_state
);
3807 if (mddev
->sysfs_action
)
3808 sysfs_notify_dirent(mddev
->sysfs_action
);
3809 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3810 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
3814 static int restart_array(mddev_t
*mddev
)
3816 struct gendisk
*disk
= mddev
->gendisk
;
3818 /* Complain if it has no devices */
3819 if (list_empty(&mddev
->disks
))
3825 mddev
->safemode
= 0;
3827 set_disk_ro(disk
, 0);
3828 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3830 /* Kick recovery or resync if necessary */
3831 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3832 md_wakeup_thread(mddev
->thread
);
3833 md_wakeup_thread(mddev
->sync_thread
);
3834 sysfs_notify_dirent(mddev
->sysfs_state
);
3838 /* similar to deny_write_access, but accounts for our holding a reference
3839 * to the file ourselves */
3840 static int deny_bitmap_write_access(struct file
* file
)
3842 struct inode
*inode
= file
->f_mapping
->host
;
3844 spin_lock(&inode
->i_lock
);
3845 if (atomic_read(&inode
->i_writecount
) > 1) {
3846 spin_unlock(&inode
->i_lock
);
3849 atomic_set(&inode
->i_writecount
, -1);
3850 spin_unlock(&inode
->i_lock
);
3855 static void restore_bitmap_write_access(struct file
*file
)
3857 struct inode
*inode
= file
->f_mapping
->host
;
3859 spin_lock(&inode
->i_lock
);
3860 atomic_set(&inode
->i_writecount
, 1);
3861 spin_unlock(&inode
->i_lock
);
3865 * 0 - completely stop and dis-assemble array
3866 * 1 - switch to readonly
3867 * 2 - stop but do not disassemble array
3869 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
3872 struct gendisk
*disk
= mddev
->gendisk
;
3874 if (atomic_read(&mddev
->openers
) > is_open
) {
3875 printk("md: %s still in use.\n",mdname(mddev
));
3881 if (mddev
->sync_thread
) {
3882 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3883 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3884 md_unregister_thread(mddev
->sync_thread
);
3885 mddev
->sync_thread
= NULL
;
3888 del_timer_sync(&mddev
->safemode_timer
);
3891 case 1: /* readonly */
3897 case 0: /* disassemble */
3899 bitmap_flush(mddev
);
3900 md_super_wait(mddev
);
3902 set_disk_ro(disk
, 0);
3903 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3904 mddev
->pers
->stop(mddev
);
3905 mddev
->queue
->merge_bvec_fn
= NULL
;
3906 mddev
->queue
->unplug_fn
= NULL
;
3907 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
3908 if (mddev
->pers
->sync_request
) {
3909 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3910 if (mddev
->sysfs_action
)
3911 sysfs_put(mddev
->sysfs_action
);
3912 mddev
->sysfs_action
= NULL
;
3914 module_put(mddev
->pers
->owner
);
3916 /* tell userspace to handle 'inactive' */
3917 sysfs_notify_dirent(mddev
->sysfs_state
);
3919 set_capacity(disk
, 0);
3925 if (!mddev
->in_sync
|| mddev
->flags
) {
3926 /* mark array as shutdown cleanly */
3928 md_update_sb(mddev
, 1);
3931 set_disk_ro(disk
, 1);
3932 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3936 * Free resources if final stop
3941 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3943 bitmap_destroy(mddev
);
3944 if (mddev
->bitmap_file
) {
3945 restore_bitmap_write_access(mddev
->bitmap_file
);
3946 fput(mddev
->bitmap_file
);
3947 mddev
->bitmap_file
= NULL
;
3949 mddev
->bitmap_offset
= 0;
3951 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3952 if (rdev
->raid_disk
>= 0) {
3954 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3955 sysfs_remove_link(&mddev
->kobj
, nm
);
3958 /* make sure all md_delayed_delete calls have finished */
3959 flush_scheduled_work();
3961 export_array(mddev
);
3963 mddev
->array_sectors
= 0;
3965 mddev
->raid_disks
= 0;
3966 mddev
->recovery_cp
= 0;
3967 mddev
->resync_min
= 0;
3968 mddev
->resync_max
= MaxSector
;
3969 mddev
->reshape_position
= MaxSector
;
3970 mddev
->external
= 0;
3971 mddev
->persistent
= 0;
3972 mddev
->level
= LEVEL_NONE
;
3973 mddev
->clevel
[0] = 0;
3976 mddev
->metadata_type
[0] = 0;
3977 mddev
->chunk_size
= 0;
3978 mddev
->ctime
= mddev
->utime
= 0;
3980 mddev
->max_disks
= 0;
3982 mddev
->delta_disks
= 0;
3983 mddev
->new_level
= LEVEL_NONE
;
3984 mddev
->new_layout
= 0;
3985 mddev
->new_chunk
= 0;
3986 mddev
->curr_resync
= 0;
3987 mddev
->resync_mismatches
= 0;
3988 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
3989 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
3990 mddev
->recovery
= 0;
3993 mddev
->degraded
= 0;
3994 mddev
->barriers_work
= 0;
3995 mddev
->safemode
= 0;
3996 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
3998 } else if (mddev
->pers
)
3999 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4002 md_new_event(mddev
);
4003 sysfs_notify_dirent(mddev
->sysfs_state
);
4009 static void autorun_array(mddev_t
*mddev
)
4014 if (list_empty(&mddev
->disks
))
4017 printk(KERN_INFO
"md: running: ");
4019 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4020 char b
[BDEVNAME_SIZE
];
4021 printk("<%s>", bdevname(rdev
->bdev
,b
));
4025 err
= do_md_run(mddev
);
4027 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4028 do_md_stop(mddev
, 0, 0);
4033 * lets try to run arrays based on all disks that have arrived
4034 * until now. (those are in pending_raid_disks)
4036 * the method: pick the first pending disk, collect all disks with
4037 * the same UUID, remove all from the pending list and put them into
4038 * the 'same_array' list. Then order this list based on superblock
4039 * update time (freshest comes first), kick out 'old' disks and
4040 * compare superblocks. If everything's fine then run it.
4042 * If "unit" is allocated, then bump its reference count
4044 static void autorun_devices(int part
)
4046 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4048 char b
[BDEVNAME_SIZE
];
4050 printk(KERN_INFO
"md: autorun ...\n");
4051 while (!list_empty(&pending_raid_disks
)) {
4054 LIST_HEAD(candidates
);
4055 rdev0
= list_entry(pending_raid_disks
.next
,
4056 mdk_rdev_t
, same_set
);
4058 printk(KERN_INFO
"md: considering %s ...\n",
4059 bdevname(rdev0
->bdev
,b
));
4060 INIT_LIST_HEAD(&candidates
);
4061 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4062 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4063 printk(KERN_INFO
"md: adding %s ...\n",
4064 bdevname(rdev
->bdev
,b
));
4065 list_move(&rdev
->same_set
, &candidates
);
4068 * now we have a set of devices, with all of them having
4069 * mostly sane superblocks. It's time to allocate the
4073 dev
= MKDEV(mdp_major
,
4074 rdev0
->preferred_minor
<< MdpMinorShift
);
4075 unit
= MINOR(dev
) >> MdpMinorShift
;
4077 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4080 if (rdev0
->preferred_minor
!= unit
) {
4081 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4082 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4086 md_probe(dev
, NULL
, NULL
);
4087 mddev
= mddev_find(dev
);
4088 if (!mddev
|| !mddev
->gendisk
) {
4092 "md: cannot allocate memory for md drive.\n");
4095 if (mddev_lock(mddev
))
4096 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4098 else if (mddev
->raid_disks
|| mddev
->major_version
4099 || !list_empty(&mddev
->disks
)) {
4101 "md: %s already running, cannot run %s\n",
4102 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4103 mddev_unlock(mddev
);
4105 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4106 mddev
->persistent
= 1;
4107 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4108 list_del_init(&rdev
->same_set
);
4109 if (bind_rdev_to_array(rdev
, mddev
))
4112 autorun_array(mddev
);
4113 mddev_unlock(mddev
);
4115 /* on success, candidates will be empty, on error
4118 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4119 list_del_init(&rdev
->same_set
);
4124 printk(KERN_INFO
"md: ... autorun DONE.\n");
4126 #endif /* !MODULE */
4128 static int get_version(void __user
* arg
)
4132 ver
.major
= MD_MAJOR_VERSION
;
4133 ver
.minor
= MD_MINOR_VERSION
;
4134 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4136 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4142 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4144 mdu_array_info_t info
;
4145 int nr
,working
,active
,failed
,spare
;
4148 nr
=working
=active
=failed
=spare
=0;
4149 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4151 if (test_bit(Faulty
, &rdev
->flags
))
4155 if (test_bit(In_sync
, &rdev
->flags
))
4162 info
.major_version
= mddev
->major_version
;
4163 info
.minor_version
= mddev
->minor_version
;
4164 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4165 info
.ctime
= mddev
->ctime
;
4166 info
.level
= mddev
->level
;
4167 info
.size
= mddev
->size
;
4168 if (info
.size
!= mddev
->size
) /* overflow */
4171 info
.raid_disks
= mddev
->raid_disks
;
4172 info
.md_minor
= mddev
->md_minor
;
4173 info
.not_persistent
= !mddev
->persistent
;
4175 info
.utime
= mddev
->utime
;
4178 info
.state
= (1<<MD_SB_CLEAN
);
4179 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4180 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4181 info
.active_disks
= active
;
4182 info
.working_disks
= working
;
4183 info
.failed_disks
= failed
;
4184 info
.spare_disks
= spare
;
4186 info
.layout
= mddev
->layout
;
4187 info
.chunk_size
= mddev
->chunk_size
;
4189 if (copy_to_user(arg
, &info
, sizeof(info
)))
4195 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4197 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4198 char *ptr
, *buf
= NULL
;
4201 if (md_allow_write(mddev
))
4202 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4204 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4209 /* bitmap disabled, zero the first byte and copy out */
4210 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4211 file
->pathname
[0] = '\0';
4215 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4219 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4223 strcpy(file
->pathname
, ptr
);
4227 if (copy_to_user(arg
, file
, sizeof(*file
)))
4235 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4237 mdu_disk_info_t info
;
4240 if (copy_from_user(&info
, arg
, sizeof(info
)))
4243 rdev
= find_rdev_nr(mddev
, info
.number
);
4245 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4246 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4247 info
.raid_disk
= rdev
->raid_disk
;
4249 if (test_bit(Faulty
, &rdev
->flags
))
4250 info
.state
|= (1<<MD_DISK_FAULTY
);
4251 else if (test_bit(In_sync
, &rdev
->flags
)) {
4252 info
.state
|= (1<<MD_DISK_ACTIVE
);
4253 info
.state
|= (1<<MD_DISK_SYNC
);
4255 if (test_bit(WriteMostly
, &rdev
->flags
))
4256 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4258 info
.major
= info
.minor
= 0;
4259 info
.raid_disk
= -1;
4260 info
.state
= (1<<MD_DISK_REMOVED
);
4263 if (copy_to_user(arg
, &info
, sizeof(info
)))
4269 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4271 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4273 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4275 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4278 if (!mddev
->raid_disks
) {
4280 /* expecting a device which has a superblock */
4281 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4284 "md: md_import_device returned %ld\n",
4286 return PTR_ERR(rdev
);
4288 if (!list_empty(&mddev
->disks
)) {
4289 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4290 mdk_rdev_t
, same_set
);
4291 int err
= super_types
[mddev
->major_version
]
4292 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4295 "md: %s has different UUID to %s\n",
4296 bdevname(rdev
->bdev
,b
),
4297 bdevname(rdev0
->bdev
,b2
));
4302 err
= bind_rdev_to_array(rdev
, mddev
);
4309 * add_new_disk can be used once the array is assembled
4310 * to add "hot spares". They must already have a superblock
4315 if (!mddev
->pers
->hot_add_disk
) {
4317 "%s: personality does not support diskops!\n",
4321 if (mddev
->persistent
)
4322 rdev
= md_import_device(dev
, mddev
->major_version
,
4323 mddev
->minor_version
);
4325 rdev
= md_import_device(dev
, -1, -1);
4328 "md: md_import_device returned %ld\n",
4330 return PTR_ERR(rdev
);
4332 /* set save_raid_disk if appropriate */
4333 if (!mddev
->persistent
) {
4334 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4335 info
->raid_disk
< mddev
->raid_disks
)
4336 rdev
->raid_disk
= info
->raid_disk
;
4338 rdev
->raid_disk
= -1;
4340 super_types
[mddev
->major_version
].
4341 validate_super(mddev
, rdev
);
4342 rdev
->saved_raid_disk
= rdev
->raid_disk
;
4344 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
4345 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4346 set_bit(WriteMostly
, &rdev
->flags
);
4348 rdev
->raid_disk
= -1;
4349 err
= bind_rdev_to_array(rdev
, mddev
);
4350 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
4351 /* If there is hot_add_disk but no hot_remove_disk
4352 * then added disks for geometry changes,
4353 * and should be added immediately.
4355 super_types
[mddev
->major_version
].
4356 validate_super(mddev
, rdev
);
4357 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
4359 unbind_rdev_from_array(rdev
);
4364 sysfs_notify_dirent(rdev
->sysfs_state
);
4366 md_update_sb(mddev
, 1);
4367 if (mddev
->degraded
)
4368 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4369 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4370 md_wakeup_thread(mddev
->thread
);
4374 /* otherwise, add_new_disk is only allowed
4375 * for major_version==0 superblocks
4377 if (mddev
->major_version
!= 0) {
4378 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
4383 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
4385 rdev
= md_import_device(dev
, -1, 0);
4388 "md: error, md_import_device() returned %ld\n",
4390 return PTR_ERR(rdev
);
4392 rdev
->desc_nr
= info
->number
;
4393 if (info
->raid_disk
< mddev
->raid_disks
)
4394 rdev
->raid_disk
= info
->raid_disk
;
4396 rdev
->raid_disk
= -1;
4398 if (rdev
->raid_disk
< mddev
->raid_disks
)
4399 if (info
->state
& (1<<MD_DISK_SYNC
))
4400 set_bit(In_sync
, &rdev
->flags
);
4402 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4403 set_bit(WriteMostly
, &rdev
->flags
);
4405 if (!mddev
->persistent
) {
4406 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
4407 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
4409 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
4410 rdev
->size
= calc_num_sectors(rdev
, mddev
->chunk_size
) / 2;
4412 err
= bind_rdev_to_array(rdev
, mddev
);
4422 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
4424 char b
[BDEVNAME_SIZE
];
4427 rdev
= find_rdev(mddev
, dev
);
4431 if (rdev
->raid_disk
>= 0)
4434 kick_rdev_from_array(rdev
);
4435 md_update_sb(mddev
, 1);
4436 md_new_event(mddev
);
4440 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
4441 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4445 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
4447 char b
[BDEVNAME_SIZE
];
4454 if (mddev
->major_version
!= 0) {
4455 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
4456 " version-0 superblocks.\n",
4460 if (!mddev
->pers
->hot_add_disk
) {
4462 "%s: personality does not support diskops!\n",
4467 rdev
= md_import_device(dev
, -1, 0);
4470 "md: error, md_import_device() returned %ld\n",
4475 if (mddev
->persistent
)
4476 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
4478 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
4480 rdev
->size
= calc_num_sectors(rdev
, mddev
->chunk_size
) / 2;
4482 if (test_bit(Faulty
, &rdev
->flags
)) {
4484 "md: can not hot-add faulty %s disk to %s!\n",
4485 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4489 clear_bit(In_sync
, &rdev
->flags
);
4491 rdev
->saved_raid_disk
= -1;
4492 err
= bind_rdev_to_array(rdev
, mddev
);
4497 * The rest should better be atomic, we can have disk failures
4498 * noticed in interrupt contexts ...
4501 if (rdev
->desc_nr
== mddev
->max_disks
) {
4502 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
4505 goto abort_unbind_export
;
4508 rdev
->raid_disk
= -1;
4510 md_update_sb(mddev
, 1);
4513 * Kick recovery, maybe this spare has to be added to the
4514 * array immediately.
4516 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4517 md_wakeup_thread(mddev
->thread
);
4518 md_new_event(mddev
);
4521 abort_unbind_export
:
4522 unbind_rdev_from_array(rdev
);
4529 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
4534 if (!mddev
->pers
->quiesce
)
4536 if (mddev
->recovery
|| mddev
->sync_thread
)
4538 /* we should be able to change the bitmap.. */
4544 return -EEXIST
; /* cannot add when bitmap is present */
4545 mddev
->bitmap_file
= fget(fd
);
4547 if (mddev
->bitmap_file
== NULL
) {
4548 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
4553 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
4555 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
4557 fput(mddev
->bitmap_file
);
4558 mddev
->bitmap_file
= NULL
;
4561 mddev
->bitmap_offset
= 0; /* file overrides offset */
4562 } else if (mddev
->bitmap
== NULL
)
4563 return -ENOENT
; /* cannot remove what isn't there */
4566 mddev
->pers
->quiesce(mddev
, 1);
4568 err
= bitmap_create(mddev
);
4569 if (fd
< 0 || err
) {
4570 bitmap_destroy(mddev
);
4571 fd
= -1; /* make sure to put the file */
4573 mddev
->pers
->quiesce(mddev
, 0);
4576 if (mddev
->bitmap_file
) {
4577 restore_bitmap_write_access(mddev
->bitmap_file
);
4578 fput(mddev
->bitmap_file
);
4580 mddev
->bitmap_file
= NULL
;
4587 * set_array_info is used two different ways
4588 * The original usage is when creating a new array.
4589 * In this usage, raid_disks is > 0 and it together with
4590 * level, size, not_persistent,layout,chunksize determine the
4591 * shape of the array.
4592 * This will always create an array with a type-0.90.0 superblock.
4593 * The newer usage is when assembling an array.
4594 * In this case raid_disks will be 0, and the major_version field is
4595 * use to determine which style super-blocks are to be found on the devices.
4596 * The minor and patch _version numbers are also kept incase the
4597 * super_block handler wishes to interpret them.
4599 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
4602 if (info
->raid_disks
== 0) {
4603 /* just setting version number for superblock loading */
4604 if (info
->major_version
< 0 ||
4605 info
->major_version
>= ARRAY_SIZE(super_types
) ||
4606 super_types
[info
->major_version
].name
== NULL
) {
4607 /* maybe try to auto-load a module? */
4609 "md: superblock version %d not known\n",
4610 info
->major_version
);
4613 mddev
->major_version
= info
->major_version
;
4614 mddev
->minor_version
= info
->minor_version
;
4615 mddev
->patch_version
= info
->patch_version
;
4616 mddev
->persistent
= !info
->not_persistent
;
4619 mddev
->major_version
= MD_MAJOR_VERSION
;
4620 mddev
->minor_version
= MD_MINOR_VERSION
;
4621 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
4622 mddev
->ctime
= get_seconds();
4624 mddev
->level
= info
->level
;
4625 mddev
->clevel
[0] = 0;
4626 mddev
->size
= info
->size
;
4627 mddev
->raid_disks
= info
->raid_disks
;
4628 /* don't set md_minor, it is determined by which /dev/md* was
4631 if (info
->state
& (1<<MD_SB_CLEAN
))
4632 mddev
->recovery_cp
= MaxSector
;
4634 mddev
->recovery_cp
= 0;
4635 mddev
->persistent
= ! info
->not_persistent
;
4636 mddev
->external
= 0;
4638 mddev
->layout
= info
->layout
;
4639 mddev
->chunk_size
= info
->chunk_size
;
4641 mddev
->max_disks
= MD_SB_DISKS
;
4643 if (mddev
->persistent
)
4645 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4647 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4648 mddev
->bitmap_offset
= 0;
4650 mddev
->reshape_position
= MaxSector
;
4653 * Generate a 128 bit UUID
4655 get_random_bytes(mddev
->uuid
, 16);
4657 mddev
->new_level
= mddev
->level
;
4658 mddev
->new_chunk
= mddev
->chunk_size
;
4659 mddev
->new_layout
= mddev
->layout
;
4660 mddev
->delta_disks
= 0;
4665 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
4669 int fit
= (num_sectors
== 0);
4671 if (mddev
->pers
->resize
== NULL
)
4673 /* The "num_sectors" is the number of sectors of each device that
4674 * is used. This can only make sense for arrays with redundancy.
4675 * linear and raid0 always use whatever space is available. We can only
4676 * consider changing this number if no resync or reconstruction is
4677 * happening, and if the new size is acceptable. It must fit before the
4678 * sb_start or, if that is <data_offset, it must fit before the size
4679 * of each device. If num_sectors is zero, we find the largest size
4683 if (mddev
->sync_thread
)
4686 /* Sorry, cannot grow a bitmap yet, just remove it,
4690 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4692 avail
= rdev
->size
* 2;
4694 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
4695 num_sectors
= avail
;
4696 if (avail
< num_sectors
)
4699 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
4701 struct block_device
*bdev
;
4703 bdev
= bdget_disk(mddev
->gendisk
, 0);
4705 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4706 i_size_write(bdev
->bd_inode
,
4707 (loff_t
)mddev
->array_sectors
<< 9);
4708 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4715 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4718 /* change the number of raid disks */
4719 if (mddev
->pers
->check_reshape
== NULL
)
4721 if (raid_disks
<= 0 ||
4722 raid_disks
>= mddev
->max_disks
)
4724 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4726 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4728 rv
= mddev
->pers
->check_reshape(mddev
);
4734 * update_array_info is used to change the configuration of an
4736 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4737 * fields in the info are checked against the array.
4738 * Any differences that cannot be handled will cause an error.
4739 * Normally, only one change can be managed at a time.
4741 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4747 /* calculate expected state,ignoring low bits */
4748 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4749 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4751 if (mddev
->major_version
!= info
->major_version
||
4752 mddev
->minor_version
!= info
->minor_version
||
4753 /* mddev->patch_version != info->patch_version || */
4754 mddev
->ctime
!= info
->ctime
||
4755 mddev
->level
!= info
->level
||
4756 /* mddev->layout != info->layout || */
4757 !mddev
->persistent
!= info
->not_persistent
||
4758 mddev
->chunk_size
!= info
->chunk_size
||
4759 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4760 ((state
^info
->state
) & 0xfffffe00)
4763 /* Check there is only one change */
4764 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4765 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4766 if (mddev
->layout
!= info
->layout
) cnt
++;
4767 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4768 if (cnt
== 0) return 0;
4769 if (cnt
> 1) return -EINVAL
;
4771 if (mddev
->layout
!= info
->layout
) {
4773 * we don't need to do anything at the md level, the
4774 * personality will take care of it all.
4776 if (mddev
->pers
->reconfig
== NULL
)
4779 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4781 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4782 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
4784 if (mddev
->raid_disks
!= info
->raid_disks
)
4785 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4787 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4788 if (mddev
->pers
->quiesce
== NULL
)
4790 if (mddev
->recovery
|| mddev
->sync_thread
)
4792 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4793 /* add the bitmap */
4796 if (mddev
->default_bitmap_offset
== 0)
4798 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4799 mddev
->pers
->quiesce(mddev
, 1);
4800 rv
= bitmap_create(mddev
);
4802 bitmap_destroy(mddev
);
4803 mddev
->pers
->quiesce(mddev
, 0);
4805 /* remove the bitmap */
4808 if (mddev
->bitmap
->file
)
4810 mddev
->pers
->quiesce(mddev
, 1);
4811 bitmap_destroy(mddev
);
4812 mddev
->pers
->quiesce(mddev
, 0);
4813 mddev
->bitmap_offset
= 0;
4816 md_update_sb(mddev
, 1);
4820 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4824 if (mddev
->pers
== NULL
)
4827 rdev
= find_rdev(mddev
, dev
);
4831 md_error(mddev
, rdev
);
4836 * We have a problem here : there is no easy way to give a CHS
4837 * virtual geometry. We currently pretend that we have a 2 heads
4838 * 4 sectors (with a BIG number of cylinders...). This drives
4839 * dosfs just mad... ;-)
4841 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4843 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4847 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4851 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
4852 unsigned int cmd
, unsigned long arg
)
4855 void __user
*argp
= (void __user
*)arg
;
4856 mddev_t
*mddev
= NULL
;
4858 if (!capable(CAP_SYS_ADMIN
))
4862 * Commands dealing with the RAID driver but not any
4868 err
= get_version(argp
);
4871 case PRINT_RAID_DEBUG
:
4879 autostart_arrays(arg
);
4886 * Commands creating/starting a new array:
4889 mddev
= bdev
->bd_disk
->private_data
;
4896 err
= mddev_lock(mddev
);
4899 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4906 case SET_ARRAY_INFO
:
4908 mdu_array_info_t info
;
4910 memset(&info
, 0, sizeof(info
));
4911 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4916 err
= update_array_info(mddev
, &info
);
4918 printk(KERN_WARNING
"md: couldn't update"
4919 " array info. %d\n", err
);
4924 if (!list_empty(&mddev
->disks
)) {
4926 "md: array %s already has disks!\n",
4931 if (mddev
->raid_disks
) {
4933 "md: array %s already initialised!\n",
4938 err
= set_array_info(mddev
, &info
);
4940 printk(KERN_WARNING
"md: couldn't set"
4941 " array info. %d\n", err
);
4951 * Commands querying/configuring an existing array:
4953 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4954 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4955 if ((!mddev
->raid_disks
&& !mddev
->external
)
4956 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4957 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
4958 && cmd
!= GET_BITMAP_FILE
) {
4964 * Commands even a read-only array can execute:
4968 case GET_ARRAY_INFO
:
4969 err
= get_array_info(mddev
, argp
);
4972 case GET_BITMAP_FILE
:
4973 err
= get_bitmap_file(mddev
, argp
);
4977 err
= get_disk_info(mddev
, argp
);
4980 case RESTART_ARRAY_RW
:
4981 err
= restart_array(mddev
);
4985 err
= do_md_stop(mddev
, 0, 1);
4989 err
= do_md_stop(mddev
, 1, 1);
4995 * The remaining ioctls are changing the state of the
4996 * superblock, so we do not allow them on read-only arrays.
4997 * However non-MD ioctls (e.g. get-size) will still come through
4998 * here and hit the 'default' below, so only disallow
4999 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5001 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5002 if (mddev
->ro
== 2) {
5004 sysfs_notify_dirent(mddev
->sysfs_state
);
5005 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5006 md_wakeup_thread(mddev
->thread
);
5017 mdu_disk_info_t info
;
5018 if (copy_from_user(&info
, argp
, sizeof(info
)))
5021 err
= add_new_disk(mddev
, &info
);
5025 case HOT_REMOVE_DISK
:
5026 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5030 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5033 case SET_DISK_FAULTY
:
5034 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5038 err
= do_md_run(mddev
);
5041 case SET_BITMAP_FILE
:
5042 err
= set_bitmap_file(mddev
, (int)arg
);
5052 mddev_unlock(mddev
);
5062 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5065 * Succeed if we can lock the mddev, which confirms that
5066 * it isn't being stopped right now.
5068 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5071 if ((err
= mutex_lock_interruptible_nested(&mddev
->reconfig_mutex
, 1)))
5076 atomic_inc(&mddev
->openers
);
5077 mddev_unlock(mddev
);
5079 check_disk_change(bdev
);
5084 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5086 mddev_t
*mddev
= disk
->private_data
;
5089 atomic_dec(&mddev
->openers
);
5095 static int md_media_changed(struct gendisk
*disk
)
5097 mddev_t
*mddev
= disk
->private_data
;
5099 return mddev
->changed
;
5102 static int md_revalidate(struct gendisk
*disk
)
5104 mddev_t
*mddev
= disk
->private_data
;
5109 static struct block_device_operations md_fops
=
5111 .owner
= THIS_MODULE
,
5113 .release
= md_release
,
5114 .locked_ioctl
= md_ioctl
,
5115 .getgeo
= md_getgeo
,
5116 .media_changed
= md_media_changed
,
5117 .revalidate_disk
= md_revalidate
,
5120 static int md_thread(void * arg
)
5122 mdk_thread_t
*thread
= arg
;
5125 * md_thread is a 'system-thread', it's priority should be very
5126 * high. We avoid resource deadlocks individually in each
5127 * raid personality. (RAID5 does preallocation) We also use RR and
5128 * the very same RT priority as kswapd, thus we will never get
5129 * into a priority inversion deadlock.
5131 * we definitely have to have equal or higher priority than
5132 * bdflush, otherwise bdflush will deadlock if there are too
5133 * many dirty RAID5 blocks.
5136 allow_signal(SIGKILL
);
5137 while (!kthread_should_stop()) {
5139 /* We need to wait INTERRUPTIBLE so that
5140 * we don't add to the load-average.
5141 * That means we need to be sure no signals are
5144 if (signal_pending(current
))
5145 flush_signals(current
);
5147 wait_event_interruptible_timeout
5149 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5150 || kthread_should_stop(),
5153 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5155 thread
->run(thread
->mddev
);
5161 void md_wakeup_thread(mdk_thread_t
*thread
)
5164 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5165 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5166 wake_up(&thread
->wqueue
);
5170 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5173 mdk_thread_t
*thread
;
5175 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5179 init_waitqueue_head(&thread
->wqueue
);
5182 thread
->mddev
= mddev
;
5183 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5184 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
5185 if (IS_ERR(thread
->tsk
)) {
5192 void md_unregister_thread(mdk_thread_t
*thread
)
5194 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5196 kthread_stop(thread
->tsk
);
5200 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5207 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5210 if (mddev
->external
)
5211 set_bit(Blocked
, &rdev
->flags
);
5213 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5215 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5216 __builtin_return_address(0),__builtin_return_address(1),
5217 __builtin_return_address(2),__builtin_return_address(3));
5221 if (!mddev
->pers
->error_handler
)
5223 mddev
->pers
->error_handler(mddev
,rdev
);
5224 if (mddev
->degraded
)
5225 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5226 set_bit(StateChanged
, &rdev
->flags
);
5227 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5228 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5229 md_wakeup_thread(mddev
->thread
);
5230 md_new_event_inintr(mddev
);
5233 /* seq_file implementation /proc/mdstat */
5235 static void status_unused(struct seq_file
*seq
)
5240 seq_printf(seq
, "unused devices: ");
5242 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
5243 char b
[BDEVNAME_SIZE
];
5245 seq_printf(seq
, "%s ",
5246 bdevname(rdev
->bdev
,b
));
5249 seq_printf(seq
, "<none>");
5251 seq_printf(seq
, "\n");
5255 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
5257 sector_t max_blocks
, resync
, res
;
5258 unsigned long dt
, db
, rt
;
5260 unsigned int per_milli
;
5262 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
5264 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5265 max_blocks
= mddev
->resync_max_sectors
>> 1;
5267 max_blocks
= mddev
->size
;
5270 * Should not happen.
5276 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5277 * in a sector_t, and (max_blocks>>scale) will fit in a
5278 * u32, as those are the requirements for sector_div.
5279 * Thus 'scale' must be at least 10
5282 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5283 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
5286 res
= (resync
>>scale
)*1000;
5287 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
5291 int i
, x
= per_milli
/50, y
= 20-x
;
5292 seq_printf(seq
, "[");
5293 for (i
= 0; i
< x
; i
++)
5294 seq_printf(seq
, "=");
5295 seq_printf(seq
, ">");
5296 for (i
= 0; i
< y
; i
++)
5297 seq_printf(seq
, ".");
5298 seq_printf(seq
, "] ");
5300 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
5301 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
5303 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
5305 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
5306 "resync" : "recovery"))),
5307 per_milli
/10, per_milli
% 10,
5308 (unsigned long long) resync
,
5309 (unsigned long long) max_blocks
);
5312 * We do not want to overflow, so the order of operands and
5313 * the * 100 / 100 trick are important. We do a +1 to be
5314 * safe against division by zero. We only estimate anyway.
5316 * dt: time from mark until now
5317 * db: blocks written from mark until now
5318 * rt: remaining time
5320 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
5322 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
5323 - mddev
->resync_mark_cnt
;
5324 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/2/100+1)))/100;
5326 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
5328 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
5331 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
5333 struct list_head
*tmp
;
5343 spin_lock(&all_mddevs_lock
);
5344 list_for_each(tmp
,&all_mddevs
)
5346 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
5348 spin_unlock(&all_mddevs_lock
);
5351 spin_unlock(&all_mddevs_lock
);
5353 return (void*)2;/* tail */
5357 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
5359 struct list_head
*tmp
;
5360 mddev_t
*next_mddev
, *mddev
= v
;
5366 spin_lock(&all_mddevs_lock
);
5368 tmp
= all_mddevs
.next
;
5370 tmp
= mddev
->all_mddevs
.next
;
5371 if (tmp
!= &all_mddevs
)
5372 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
5374 next_mddev
= (void*)2;
5377 spin_unlock(&all_mddevs_lock
);
5385 static void md_seq_stop(struct seq_file
*seq
, void *v
)
5389 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
5393 struct mdstat_info
{
5397 static int md_seq_show(struct seq_file
*seq
, void *v
)
5402 struct mdstat_info
*mi
= seq
->private;
5403 struct bitmap
*bitmap
;
5405 if (v
== (void*)1) {
5406 struct mdk_personality
*pers
;
5407 seq_printf(seq
, "Personalities : ");
5408 spin_lock(&pers_lock
);
5409 list_for_each_entry(pers
, &pers_list
, list
)
5410 seq_printf(seq
, "[%s] ", pers
->name
);
5412 spin_unlock(&pers_lock
);
5413 seq_printf(seq
, "\n");
5414 mi
->event
= atomic_read(&md_event_count
);
5417 if (v
== (void*)2) {
5422 if (mddev_lock(mddev
) < 0)
5425 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
5426 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
5427 mddev
->pers
? "" : "in");
5430 seq_printf(seq
, " (read-only)");
5432 seq_printf(seq
, " (auto-read-only)");
5433 seq_printf(seq
, " %s", mddev
->pers
->name
);
5437 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5438 char b
[BDEVNAME_SIZE
];
5439 seq_printf(seq
, " %s[%d]",
5440 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
5441 if (test_bit(WriteMostly
, &rdev
->flags
))
5442 seq_printf(seq
, "(W)");
5443 if (test_bit(Faulty
, &rdev
->flags
)) {
5444 seq_printf(seq
, "(F)");
5446 } else if (rdev
->raid_disk
< 0)
5447 seq_printf(seq
, "(S)"); /* spare */
5451 if (!list_empty(&mddev
->disks
)) {
5453 seq_printf(seq
, "\n %llu blocks",
5454 (unsigned long long)
5455 mddev
->array_sectors
/ 2);
5457 seq_printf(seq
, "\n %llu blocks",
5458 (unsigned long long)size
);
5460 if (mddev
->persistent
) {
5461 if (mddev
->major_version
!= 0 ||
5462 mddev
->minor_version
!= 90) {
5463 seq_printf(seq
," super %d.%d",
5464 mddev
->major_version
,
5465 mddev
->minor_version
);
5467 } else if (mddev
->external
)
5468 seq_printf(seq
, " super external:%s",
5469 mddev
->metadata_type
);
5471 seq_printf(seq
, " super non-persistent");
5474 mddev
->pers
->status(seq
, mddev
);
5475 seq_printf(seq
, "\n ");
5476 if (mddev
->pers
->sync_request
) {
5477 if (mddev
->curr_resync
> 2) {
5478 status_resync(seq
, mddev
);
5479 seq_printf(seq
, "\n ");
5480 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
5481 seq_printf(seq
, "\tresync=DELAYED\n ");
5482 else if (mddev
->recovery_cp
< MaxSector
)
5483 seq_printf(seq
, "\tresync=PENDING\n ");
5486 seq_printf(seq
, "\n ");
5488 if ((bitmap
= mddev
->bitmap
)) {
5489 unsigned long chunk_kb
;
5490 unsigned long flags
;
5491 spin_lock_irqsave(&bitmap
->lock
, flags
);
5492 chunk_kb
= bitmap
->chunksize
>> 10;
5493 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
5495 bitmap
->pages
- bitmap
->missing_pages
,
5497 (bitmap
->pages
- bitmap
->missing_pages
)
5498 << (PAGE_SHIFT
- 10),
5499 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
5500 chunk_kb
? "KB" : "B");
5502 seq_printf(seq
, ", file: ");
5503 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
5506 seq_printf(seq
, "\n");
5507 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
5510 seq_printf(seq
, "\n");
5512 mddev_unlock(mddev
);
5517 static struct seq_operations md_seq_ops
= {
5518 .start
= md_seq_start
,
5519 .next
= md_seq_next
,
5520 .stop
= md_seq_stop
,
5521 .show
= md_seq_show
,
5524 static int md_seq_open(struct inode
*inode
, struct file
*file
)
5527 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
5531 error
= seq_open(file
, &md_seq_ops
);
5535 struct seq_file
*p
= file
->private_data
;
5537 mi
->event
= atomic_read(&md_event_count
);
5542 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
5544 struct seq_file
*m
= filp
->private_data
;
5545 struct mdstat_info
*mi
= m
->private;
5548 poll_wait(filp
, &md_event_waiters
, wait
);
5550 /* always allow read */
5551 mask
= POLLIN
| POLLRDNORM
;
5553 if (mi
->event
!= atomic_read(&md_event_count
))
5554 mask
|= POLLERR
| POLLPRI
;
5558 static const struct file_operations md_seq_fops
= {
5559 .owner
= THIS_MODULE
,
5560 .open
= md_seq_open
,
5562 .llseek
= seq_lseek
,
5563 .release
= seq_release_private
,
5564 .poll
= mdstat_poll
,
5567 int register_md_personality(struct mdk_personality
*p
)
5569 spin_lock(&pers_lock
);
5570 list_add_tail(&p
->list
, &pers_list
);
5571 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
5572 spin_unlock(&pers_lock
);
5576 int unregister_md_personality(struct mdk_personality
*p
)
5578 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
5579 spin_lock(&pers_lock
);
5580 list_del_init(&p
->list
);
5581 spin_unlock(&pers_lock
);
5585 static int is_mddev_idle(mddev_t
*mddev
)
5593 rdev_for_each_rcu(rdev
, mddev
) {
5594 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
5595 curr_events
= part_stat_read(&disk
->part0
, sectors
[0]) +
5596 part_stat_read(&disk
->part0
, sectors
[1]) -
5597 atomic_read(&disk
->sync_io
);
5598 /* sync IO will cause sync_io to increase before the disk_stats
5599 * as sync_io is counted when a request starts, and
5600 * disk_stats is counted when it completes.
5601 * So resync activity will cause curr_events to be smaller than
5602 * when there was no such activity.
5603 * non-sync IO will cause disk_stat to increase without
5604 * increasing sync_io so curr_events will (eventually)
5605 * be larger than it was before. Once it becomes
5606 * substantially larger, the test below will cause
5607 * the array to appear non-idle, and resync will slow
5609 * If there is a lot of outstanding resync activity when
5610 * we set last_event to curr_events, then all that activity
5611 * completing might cause the array to appear non-idle
5612 * and resync will be slowed down even though there might
5613 * not have been non-resync activity. This will only
5614 * happen once though. 'last_events' will soon reflect
5615 * the state where there is little or no outstanding
5616 * resync requests, and further resync activity will
5617 * always make curr_events less than last_events.
5620 if (curr_events
- rdev
->last_events
> 4096) {
5621 rdev
->last_events
= curr_events
;
5629 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
5631 /* another "blocks" (512byte) blocks have been synced */
5632 atomic_sub(blocks
, &mddev
->recovery_active
);
5633 wake_up(&mddev
->recovery_wait
);
5635 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5636 md_wakeup_thread(mddev
->thread
);
5637 // stop recovery, signal do_sync ....
5642 /* md_write_start(mddev, bi)
5643 * If we need to update some array metadata (e.g. 'active' flag
5644 * in superblock) before writing, schedule a superblock update
5645 * and wait for it to complete.
5647 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
5650 if (bio_data_dir(bi
) != WRITE
)
5653 BUG_ON(mddev
->ro
== 1);
5654 if (mddev
->ro
== 2) {
5655 /* need to switch to read/write */
5657 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5658 md_wakeup_thread(mddev
->thread
);
5659 md_wakeup_thread(mddev
->sync_thread
);
5662 atomic_inc(&mddev
->writes_pending
);
5663 if (mddev
->safemode
== 1)
5664 mddev
->safemode
= 0;
5665 if (mddev
->in_sync
) {
5666 spin_lock_irq(&mddev
->write_lock
);
5667 if (mddev
->in_sync
) {
5669 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5670 md_wakeup_thread(mddev
->thread
);
5673 spin_unlock_irq(&mddev
->write_lock
);
5676 sysfs_notify_dirent(mddev
->sysfs_state
);
5677 wait_event(mddev
->sb_wait
,
5678 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
5679 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
5682 void md_write_end(mddev_t
*mddev
)
5684 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5685 if (mddev
->safemode
== 2)
5686 md_wakeup_thread(mddev
->thread
);
5687 else if (mddev
->safemode_delay
)
5688 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5692 /* md_allow_write(mddev)
5693 * Calling this ensures that the array is marked 'active' so that writes
5694 * may proceed without blocking. It is important to call this before
5695 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5696 * Must be called with mddev_lock held.
5698 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5699 * is dropped, so return -EAGAIN after notifying userspace.
5701 int md_allow_write(mddev_t
*mddev
)
5707 if (!mddev
->pers
->sync_request
)
5710 spin_lock_irq(&mddev
->write_lock
);
5711 if (mddev
->in_sync
) {
5713 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5714 if (mddev
->safemode_delay
&&
5715 mddev
->safemode
== 0)
5716 mddev
->safemode
= 1;
5717 spin_unlock_irq(&mddev
->write_lock
);
5718 md_update_sb(mddev
, 0);
5719 sysfs_notify_dirent(mddev
->sysfs_state
);
5721 spin_unlock_irq(&mddev
->write_lock
);
5723 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
5728 EXPORT_SYMBOL_GPL(md_allow_write
);
5730 #define SYNC_MARKS 10
5731 #define SYNC_MARK_STEP (3*HZ)
5732 void md_do_sync(mddev_t
*mddev
)
5735 unsigned int currspeed
= 0,
5737 sector_t max_sectors
,j
, io_sectors
;
5738 unsigned long mark
[SYNC_MARKS
];
5739 sector_t mark_cnt
[SYNC_MARKS
];
5741 struct list_head
*tmp
;
5742 sector_t last_check
;
5747 /* just incase thread restarts... */
5748 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5750 if (mddev
->ro
) /* never try to sync a read-only array */
5753 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5754 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
5755 desc
= "data-check";
5756 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5757 desc
= "requested-resync";
5760 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5765 /* we overload curr_resync somewhat here.
5766 * 0 == not engaged in resync at all
5767 * 2 == checking that there is no conflict with another sync
5768 * 1 == like 2, but have yielded to allow conflicting resync to
5770 * other == active in resync - this many blocks
5772 * Before starting a resync we must have set curr_resync to
5773 * 2, and then checked that every "conflicting" array has curr_resync
5774 * less than ours. When we find one that is the same or higher
5775 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5776 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5777 * This will mean we have to start checking from the beginning again.
5782 mddev
->curr_resync
= 2;
5785 if (kthread_should_stop()) {
5786 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5789 for_each_mddev(mddev2
, tmp
) {
5790 if (mddev2
== mddev
)
5792 if (!mddev
->parallel_resync
5793 && mddev2
->curr_resync
5794 && match_mddev_units(mddev
, mddev2
)) {
5796 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5797 /* arbitrarily yield */
5798 mddev
->curr_resync
= 1;
5799 wake_up(&resync_wait
);
5801 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5802 /* no need to wait here, we can wait the next
5803 * time 'round when curr_resync == 2
5806 /* We need to wait 'interruptible' so as not to
5807 * contribute to the load average, and not to
5808 * be caught by 'softlockup'
5810 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
5811 if (!kthread_should_stop() &&
5812 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5813 printk(KERN_INFO
"md: delaying %s of %s"
5814 " until %s has finished (they"
5815 " share one or more physical units)\n",
5816 desc
, mdname(mddev
), mdname(mddev2
));
5818 if (signal_pending(current
))
5819 flush_signals(current
);
5821 finish_wait(&resync_wait
, &wq
);
5824 finish_wait(&resync_wait
, &wq
);
5827 } while (mddev
->curr_resync
< 2);
5830 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5831 /* resync follows the size requested by the personality,
5832 * which defaults to physical size, but can be virtual size
5834 max_sectors
= mddev
->resync_max_sectors
;
5835 mddev
->resync_mismatches
= 0;
5836 /* we don't use the checkpoint if there's a bitmap */
5837 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5838 j
= mddev
->resync_min
;
5839 else if (!mddev
->bitmap
)
5840 j
= mddev
->recovery_cp
;
5842 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5843 max_sectors
= mddev
->size
<< 1;
5845 /* recovery follows the physical size of devices */
5846 max_sectors
= mddev
->size
<< 1;
5848 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5849 if (rdev
->raid_disk
>= 0 &&
5850 !test_bit(Faulty
, &rdev
->flags
) &&
5851 !test_bit(In_sync
, &rdev
->flags
) &&
5852 rdev
->recovery_offset
< j
)
5853 j
= rdev
->recovery_offset
;
5856 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
5857 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
5858 " %d KB/sec/disk.\n", speed_min(mddev
));
5859 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5860 "(but not more than %d KB/sec) for %s.\n",
5861 speed_max(mddev
), desc
);
5863 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5866 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5868 mark_cnt
[m
] = io_sectors
;
5871 mddev
->resync_mark
= mark
[last_mark
];
5872 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5875 * Tune reconstruction:
5877 window
= 32*(PAGE_SIZE
/512);
5878 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5879 window
/2,(unsigned long long) max_sectors
/2);
5881 atomic_set(&mddev
->recovery_active
, 0);
5886 "md: resuming %s of %s from checkpoint.\n",
5887 desc
, mdname(mddev
));
5888 mddev
->curr_resync
= j
;
5891 while (j
< max_sectors
) {
5895 if (j
>= mddev
->resync_max
) {
5896 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
5897 wait_event(mddev
->recovery_wait
,
5898 mddev
->resync_max
> j
5899 || kthread_should_stop());
5901 if (kthread_should_stop())
5903 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5904 currspeed
< speed_min(mddev
));
5906 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5910 if (!skipped
) { /* actual IO requested */
5911 io_sectors
+= sectors
;
5912 atomic_add(sectors
, &mddev
->recovery_active
);
5916 if (j
>1) mddev
->curr_resync
= j
;
5917 mddev
->curr_mark_cnt
= io_sectors
;
5918 if (last_check
== 0)
5919 /* this is the earliers that rebuilt will be
5920 * visible in /proc/mdstat
5922 md_new_event(mddev
);
5924 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5927 last_check
= io_sectors
;
5929 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5933 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5935 int next
= (last_mark
+1) % SYNC_MARKS
;
5937 mddev
->resync_mark
= mark
[next
];
5938 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5939 mark
[next
] = jiffies
;
5940 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5945 if (kthread_should_stop())
5950 * this loop exits only if either when we are slower than
5951 * the 'hard' speed limit, or the system was IO-idle for
5953 * the system might be non-idle CPU-wise, but we only care
5954 * about not overloading the IO subsystem. (things like an
5955 * e2fsck being done on the RAID array should execute fast)
5957 blk_unplug(mddev
->queue
);
5960 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5961 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5963 if (currspeed
> speed_min(mddev
)) {
5964 if ((currspeed
> speed_max(mddev
)) ||
5965 !is_mddev_idle(mddev
)) {
5971 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
5973 * this also signals 'finished resyncing' to md_stop
5976 blk_unplug(mddev
->queue
);
5978 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5980 /* tell personality that we are finished */
5981 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5983 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5984 mddev
->curr_resync
> 2) {
5985 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5986 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5987 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5989 "md: checkpointing %s of %s.\n",
5990 desc
, mdname(mddev
));
5991 mddev
->recovery_cp
= mddev
->curr_resync
;
5994 mddev
->recovery_cp
= MaxSector
;
5996 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5997 mddev
->curr_resync
= MaxSector
;
5998 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5999 if (rdev
->raid_disk
>= 0 &&
6000 !test_bit(Faulty
, &rdev
->flags
) &&
6001 !test_bit(In_sync
, &rdev
->flags
) &&
6002 rdev
->recovery_offset
< mddev
->curr_resync
)
6003 rdev
->recovery_offset
= mddev
->curr_resync
;
6006 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6009 mddev
->curr_resync
= 0;
6010 mddev
->resync_min
= 0;
6011 mddev
->resync_max
= MaxSector
;
6012 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6013 wake_up(&resync_wait
);
6014 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6015 md_wakeup_thread(mddev
->thread
);
6020 * got a signal, exit.
6023 "md: md_do_sync() got signal ... exiting\n");
6024 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6028 EXPORT_SYMBOL_GPL(md_do_sync
);
6031 static int remove_and_add_spares(mddev_t
*mddev
)
6036 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6037 if (rdev
->raid_disk
>= 0 &&
6038 !test_bit(Blocked
, &rdev
->flags
) &&
6039 (test_bit(Faulty
, &rdev
->flags
) ||
6040 ! test_bit(In_sync
, &rdev
->flags
)) &&
6041 atomic_read(&rdev
->nr_pending
)==0) {
6042 if (mddev
->pers
->hot_remove_disk(
6043 mddev
, rdev
->raid_disk
)==0) {
6045 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6046 sysfs_remove_link(&mddev
->kobj
, nm
);
6047 rdev
->raid_disk
= -1;
6051 if (mddev
->degraded
&& ! mddev
->ro
) {
6052 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6053 if (rdev
->raid_disk
>= 0 &&
6054 !test_bit(In_sync
, &rdev
->flags
) &&
6055 !test_bit(Blocked
, &rdev
->flags
))
6057 if (rdev
->raid_disk
< 0
6058 && !test_bit(Faulty
, &rdev
->flags
)) {
6059 rdev
->recovery_offset
= 0;
6061 hot_add_disk(mddev
, rdev
) == 0) {
6063 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6064 if (sysfs_create_link(&mddev
->kobj
,
6067 "md: cannot register "
6071 md_new_event(mddev
);
6080 * This routine is regularly called by all per-raid-array threads to
6081 * deal with generic issues like resync and super-block update.
6082 * Raid personalities that don't have a thread (linear/raid0) do not
6083 * need this as they never do any recovery or update the superblock.
6085 * It does not do any resync itself, but rather "forks" off other threads
6086 * to do that as needed.
6087 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6088 * "->recovery" and create a thread at ->sync_thread.
6089 * When the thread finishes it sets MD_RECOVERY_DONE
6090 * and wakeups up this thread which will reap the thread and finish up.
6091 * This thread also removes any faulty devices (with nr_pending == 0).
6093 * The overall approach is:
6094 * 1/ if the superblock needs updating, update it.
6095 * 2/ If a recovery thread is running, don't do anything else.
6096 * 3/ If recovery has finished, clean up, possibly marking spares active.
6097 * 4/ If there are any faulty devices, remove them.
6098 * 5/ If array is degraded, try to add spares devices
6099 * 6/ If array has spares or is not in-sync, start a resync thread.
6101 void md_check_recovery(mddev_t
*mddev
)
6107 bitmap_daemon_work(mddev
->bitmap
);
6112 if (signal_pending(current
)) {
6113 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6114 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6116 mddev
->safemode
= 2;
6118 flush_signals(current
);
6121 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6124 (mddev
->flags
&& !mddev
->external
) ||
6125 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6126 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6127 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6128 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6129 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6133 if (mddev_trylock(mddev
)) {
6137 /* Only thing we do on a ro array is remove
6140 remove_and_add_spares(mddev
);
6141 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6145 if (!mddev
->external
) {
6147 spin_lock_irq(&mddev
->write_lock
);
6148 if (mddev
->safemode
&&
6149 !atomic_read(&mddev
->writes_pending
) &&
6151 mddev
->recovery_cp
== MaxSector
) {
6154 if (mddev
->persistent
)
6155 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6157 if (mddev
->safemode
== 1)
6158 mddev
->safemode
= 0;
6159 spin_unlock_irq(&mddev
->write_lock
);
6161 sysfs_notify_dirent(mddev
->sysfs_state
);
6165 md_update_sb(mddev
, 0);
6167 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6168 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6169 sysfs_notify_dirent(rdev
->sysfs_state
);
6172 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6173 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6174 /* resync/recovery still happening */
6175 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6178 if (mddev
->sync_thread
) {
6179 /* resync has finished, collect result */
6180 md_unregister_thread(mddev
->sync_thread
);
6181 mddev
->sync_thread
= NULL
;
6182 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6183 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6185 /* activate any spares */
6186 if (mddev
->pers
->spare_active(mddev
))
6187 sysfs_notify(&mddev
->kobj
, NULL
,
6190 md_update_sb(mddev
, 1);
6192 /* if array is no-longer degraded, then any saved_raid_disk
6193 * information must be scrapped
6195 if (!mddev
->degraded
)
6196 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6197 rdev
->saved_raid_disk
= -1;
6199 mddev
->recovery
= 0;
6200 /* flag recovery needed just to double check */
6201 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6202 sysfs_notify_dirent(mddev
->sysfs_action
);
6203 md_new_event(mddev
);
6206 /* Set RUNNING before clearing NEEDED to avoid
6207 * any transients in the value of "sync_action".
6209 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6210 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6211 /* Clear some bits that don't mean anything, but
6214 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6215 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6217 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
6219 /* no recovery is running.
6220 * remove any failed drives, then
6221 * add spares if possible.
6222 * Spare are also removed and re-added, to allow
6223 * the personality to fail the re-add.
6226 if (mddev
->reshape_position
!= MaxSector
) {
6227 if (mddev
->pers
->check_reshape(mddev
) != 0)
6228 /* Cannot proceed */
6230 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
6231 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6232 } else if ((spares
= remove_and_add_spares(mddev
))) {
6233 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6234 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
6235 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
6236 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6237 } else if (mddev
->recovery_cp
< MaxSector
) {
6238 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6239 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6240 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6241 /* nothing to be done ... */
6244 if (mddev
->pers
->sync_request
) {
6245 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
6246 /* We are adding a device or devices to an array
6247 * which has the bitmap stored on all devices.
6248 * So make sure all bitmap pages get written
6250 bitmap_write_all(mddev
->bitmap
);
6252 mddev
->sync_thread
= md_register_thread(md_do_sync
,
6255 if (!mddev
->sync_thread
) {
6256 printk(KERN_ERR
"%s: could not start resync"
6259 /* leave the spares where they are, it shouldn't hurt */
6260 mddev
->recovery
= 0;
6262 md_wakeup_thread(mddev
->sync_thread
);
6263 sysfs_notify_dirent(mddev
->sysfs_action
);
6264 md_new_event(mddev
);
6267 if (!mddev
->sync_thread
) {
6268 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6269 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
6271 if (mddev
->sysfs_action
)
6272 sysfs_notify_dirent(mddev
->sysfs_action
);
6274 mddev_unlock(mddev
);
6278 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
6280 sysfs_notify_dirent(rdev
->sysfs_state
);
6281 wait_event_timeout(rdev
->blocked_wait
,
6282 !test_bit(Blocked
, &rdev
->flags
),
6283 msecs_to_jiffies(5000));
6284 rdev_dec_pending(rdev
, mddev
);
6286 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
6288 static int md_notify_reboot(struct notifier_block
*this,
6289 unsigned long code
, void *x
)
6291 struct list_head
*tmp
;
6294 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
6296 printk(KERN_INFO
"md: stopping all md devices.\n");
6298 for_each_mddev(mddev
, tmp
)
6299 if (mddev_trylock(mddev
)) {
6300 /* Force a switch to readonly even array
6301 * appears to still be in use. Hence
6304 do_md_stop(mddev
, 1, 100);
6305 mddev_unlock(mddev
);
6308 * certain more exotic SCSI devices are known to be
6309 * volatile wrt too early system reboots. While the
6310 * right place to handle this issue is the given
6311 * driver, we do want to have a safe RAID driver ...
6318 static struct notifier_block md_notifier
= {
6319 .notifier_call
= md_notify_reboot
,
6321 .priority
= INT_MAX
, /* before any real devices */
6324 static void md_geninit(void)
6326 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
6328 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
6331 static int __init
md_init(void)
6333 if (register_blkdev(MAJOR_NR
, "md"))
6335 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
6336 unregister_blkdev(MAJOR_NR
, "md");
6339 blk_register_region(MKDEV(MAJOR_NR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6340 md_probe
, NULL
, NULL
);
6341 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6342 md_probe
, NULL
, NULL
);
6344 register_reboot_notifier(&md_notifier
);
6345 raid_table_header
= register_sysctl_table(raid_root_table
);
6355 * Searches all registered partitions for autorun RAID arrays
6359 static LIST_HEAD(all_detected_devices
);
6360 struct detected_devices_node
{
6361 struct list_head list
;
6365 void md_autodetect_dev(dev_t dev
)
6367 struct detected_devices_node
*node_detected_dev
;
6369 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
6370 if (node_detected_dev
) {
6371 node_detected_dev
->dev
= dev
;
6372 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
6374 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
6375 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
6380 static void autostart_arrays(int part
)
6383 struct detected_devices_node
*node_detected_dev
;
6385 int i_scanned
, i_passed
;
6390 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
6392 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
6394 node_detected_dev
= list_entry(all_detected_devices
.next
,
6395 struct detected_devices_node
, list
);
6396 list_del(&node_detected_dev
->list
);
6397 dev
= node_detected_dev
->dev
;
6398 kfree(node_detected_dev
);
6399 rdev
= md_import_device(dev
,0, 90);
6403 if (test_bit(Faulty
, &rdev
->flags
)) {
6407 set_bit(AutoDetected
, &rdev
->flags
);
6408 list_add(&rdev
->same_set
, &pending_raid_disks
);
6412 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
6413 i_scanned
, i_passed
);
6415 autorun_devices(part
);
6418 #endif /* !MODULE */
6420 static __exit
void md_exit(void)
6423 struct list_head
*tmp
;
6425 blk_unregister_region(MKDEV(MAJOR_NR
,0), 1U << MINORBITS
);
6426 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
6428 unregister_blkdev(MAJOR_NR
,"md");
6429 unregister_blkdev(mdp_major
, "mdp");
6430 unregister_reboot_notifier(&md_notifier
);
6431 unregister_sysctl_table(raid_table_header
);
6432 remove_proc_entry("mdstat", NULL
);
6433 for_each_mddev(mddev
, tmp
) {
6434 struct gendisk
*disk
= mddev
->gendisk
;
6437 export_array(mddev
);
6440 mddev
->gendisk
= NULL
;
6445 subsys_initcall(md_init
);
6446 module_exit(md_exit
)
6448 static int get_ro(char *buffer
, struct kernel_param
*kp
)
6450 return sprintf(buffer
, "%d", start_readonly
);
6452 static int set_ro(const char *val
, struct kernel_param
*kp
)
6455 int num
= simple_strtoul(val
, &e
, 10);
6456 if (*val
&& (*e
== '\0' || *e
== '\n')) {
6457 start_readonly
= num
;
6463 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
6464 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
6467 EXPORT_SYMBOL(register_md_personality
);
6468 EXPORT_SYMBOL(unregister_md_personality
);
6469 EXPORT_SYMBOL(md_error
);
6470 EXPORT_SYMBOL(md_done_sync
);
6471 EXPORT_SYMBOL(md_write_start
);
6472 EXPORT_SYMBOL(md_write_end
);
6473 EXPORT_SYMBOL(md_register_thread
);
6474 EXPORT_SYMBOL(md_unregister_thread
);
6475 EXPORT_SYMBOL(md_wakeup_thread
);
6476 EXPORT_SYMBOL(md_check_recovery
);
6477 MODULE_LICENSE("GPL");
6479 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);