2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part
);
71 static LIST_HEAD(pers_list
);
72 static DEFINE_SPINLOCK(pers_lock
);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 * or /sys/block/mdX/md/sync_speed_{min,max}
87 static int sysctl_speed_limit_min
= 1000;
88 static int sysctl_speed_limit_max
= 200000;
89 static inline int speed_min(mddev_t
*mddev
)
91 return mddev
->sync_speed_min
?
92 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
95 static inline int speed_max(mddev_t
*mddev
)
97 return mddev
->sync_speed_max
?
98 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
101 static struct ctl_table_header
*raid_table_header
;
103 static ctl_table raid_table
[] = {
105 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
106 .procname
= "speed_limit_min",
107 .data
= &sysctl_speed_limit_min
,
108 .maxlen
= sizeof(int),
110 .proc_handler
= &proc_dointvec
,
113 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
114 .procname
= "speed_limit_max",
115 .data
= &sysctl_speed_limit_max
,
116 .maxlen
= sizeof(int),
118 .proc_handler
= &proc_dointvec
,
123 static ctl_table raid_dir_table
[] = {
125 .ctl_name
= DEV_RAID
,
134 static ctl_table raid_root_table
[] = {
140 .child
= raid_dir_table
,
145 static struct block_device_operations md_fops
;
147 static int start_readonly
;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
160 static atomic_t md_event_count
;
161 static void md_new_event(mddev_t
*mddev
)
163 atomic_inc(&md_event_count
);
164 wake_up(&md_event_waiters
);
168 * Enables to iterate over all existing md arrays
169 * all_mddevs_lock protects this list.
171 static LIST_HEAD(all_mddevs
);
172 static DEFINE_SPINLOCK(all_mddevs_lock
);
176 * iterates through all used mddevs in the system.
177 * We take care to grab the all_mddevs_lock whenever navigating
178 * the list, and to always hold a refcount when unlocked.
179 * Any code which breaks out of this loop while own
180 * a reference to the current mddev and must mddev_put it.
182 #define ITERATE_MDDEV(mddev,tmp) \
184 for (({ spin_lock(&all_mddevs_lock); \
185 tmp = all_mddevs.next; \
187 ({ if (tmp != &all_mddevs) \
188 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
189 spin_unlock(&all_mddevs_lock); \
190 if (mddev) mddev_put(mddev); \
191 mddev = list_entry(tmp, mddev_t, all_mddevs); \
192 tmp != &all_mddevs;}); \
193 ({ spin_lock(&all_mddevs_lock); \
198 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
200 bio_io_error(bio
, bio
->bi_size
);
204 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
206 atomic_inc(&mddev
->active
);
210 static void mddev_put(mddev_t
*mddev
)
212 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
214 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
215 list_del(&mddev
->all_mddevs
);
217 blk_cleanup_queue(mddev
->queue
);
218 /* that also blocks */
219 kobject_unregister(&mddev
->kobj
);
220 /* result blows... */
222 spin_unlock(&all_mddevs_lock
);
225 static mddev_t
* mddev_find(dev_t unit
)
227 mddev_t
*mddev
, *new = NULL
;
230 spin_lock(&all_mddevs_lock
);
231 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
232 if (mddev
->unit
== unit
) {
234 spin_unlock(&all_mddevs_lock
);
240 list_add(&new->all_mddevs
, &all_mddevs
);
241 spin_unlock(&all_mddevs_lock
);
244 spin_unlock(&all_mddevs_lock
);
246 new = kzalloc(sizeof(*new), GFP_KERNEL
);
251 if (MAJOR(unit
) == MD_MAJOR
)
252 new->md_minor
= MINOR(unit
);
254 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
256 init_MUTEX(&new->reconfig_sem
);
257 INIT_LIST_HEAD(&new->disks
);
258 INIT_LIST_HEAD(&new->all_mddevs
);
259 init_timer(&new->safemode_timer
);
260 atomic_set(&new->active
, 1);
261 spin_lock_init(&new->write_lock
);
262 init_waitqueue_head(&new->sb_wait
);
264 new->queue
= blk_alloc_queue(GFP_KERNEL
);
270 blk_queue_make_request(new->queue
, md_fail_request
);
275 static inline int mddev_lock(mddev_t
* mddev
)
277 return down_interruptible(&mddev
->reconfig_sem
);
280 static inline void mddev_lock_uninterruptible(mddev_t
* mddev
)
282 down(&mddev
->reconfig_sem
);
285 static inline int mddev_trylock(mddev_t
* mddev
)
287 return down_trylock(&mddev
->reconfig_sem
);
290 static inline void mddev_unlock(mddev_t
* mddev
)
292 up(&mddev
->reconfig_sem
);
294 md_wakeup_thread(mddev
->thread
);
297 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
300 struct list_head
*tmp
;
302 ITERATE_RDEV(mddev
,rdev
,tmp
) {
303 if (rdev
->desc_nr
== nr
)
309 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
311 struct list_head
*tmp
;
314 ITERATE_RDEV(mddev
,rdev
,tmp
) {
315 if (rdev
->bdev
->bd_dev
== dev
)
321 static struct mdk_personality
*find_pers(int level
, char *clevel
)
323 struct mdk_personality
*pers
;
324 list_for_each_entry(pers
, &pers_list
, list
) {
325 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
327 if (strcmp(pers
->name
, clevel
)==0)
333 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
335 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
336 return MD_NEW_SIZE_BLOCKS(size
);
339 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
343 size
= rdev
->sb_offset
;
346 size
&= ~((sector_t
)chunk_size
/1024 - 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 int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
378 mdk_rdev_t
*rdev
= bio
->bi_private
;
379 mddev_t
*mddev
= rdev
->mddev
;
383 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
384 md_error(mddev
, rdev
);
386 if (atomic_dec_and_test(&mddev
->pending_writes
))
387 wake_up(&mddev
->sb_wait
);
392 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
394 struct bio
*bio2
= bio
->bi_private
;
395 mdk_rdev_t
*rdev
= bio2
->bi_private
;
396 mddev_t
*mddev
= rdev
->mddev
;
400 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
401 error
== -EOPNOTSUPP
) {
403 /* barriers don't appear to be supported :-( */
404 set_bit(BarriersNotsupp
, &rdev
->flags
);
405 mddev
->barriers_work
= 0;
406 spin_lock_irqsave(&mddev
->write_lock
, flags
);
407 bio2
->bi_next
= mddev
->biolist
;
408 mddev
->biolist
= bio2
;
409 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
410 wake_up(&mddev
->sb_wait
);
415 bio
->bi_private
= rdev
;
416 return super_written(bio
, bytes_done
, error
);
419 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
420 sector_t sector
, int size
, struct page
*page
)
422 /* write first size bytes of page to sector of rdev
423 * Increment mddev->pending_writes before returning
424 * and decrement it on completion, waking up sb_wait
425 * if zero is reached.
426 * If an error occurred, call md_error
428 * As we might need to resubmit the request if BIO_RW_BARRIER
429 * causes ENOTSUPP, we allocate a spare bio...
431 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
432 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
434 bio
->bi_bdev
= rdev
->bdev
;
435 bio
->bi_sector
= sector
;
436 bio_add_page(bio
, page
, size
, 0);
437 bio
->bi_private
= rdev
;
438 bio
->bi_end_io
= super_written
;
441 atomic_inc(&mddev
->pending_writes
);
442 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
444 rw
|= (1<<BIO_RW_BARRIER
);
445 rbio
= bio_clone(bio
, GFP_NOIO
);
446 rbio
->bi_private
= bio
;
447 rbio
->bi_end_io
= super_written_barrier
;
448 submit_bio(rw
, rbio
);
453 void md_super_wait(mddev_t
*mddev
)
455 /* wait for all superblock writes that were scheduled to complete.
456 * if any had to be retried (due to BARRIER problems), retry them
460 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
461 if (atomic_read(&mddev
->pending_writes
)==0)
463 while (mddev
->biolist
) {
465 spin_lock_irq(&mddev
->write_lock
);
466 bio
= mddev
->biolist
;
467 mddev
->biolist
= bio
->bi_next
;
469 spin_unlock_irq(&mddev
->write_lock
);
470 submit_bio(bio
->bi_rw
, bio
);
474 finish_wait(&mddev
->sb_wait
, &wq
);
477 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
482 complete((struct completion
*)bio
->bi_private
);
486 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
487 struct page
*page
, int rw
)
489 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
490 struct completion event
;
493 rw
|= (1 << BIO_RW_SYNC
);
496 bio
->bi_sector
= sector
;
497 bio_add_page(bio
, page
, size
, 0);
498 init_completion(&event
);
499 bio
->bi_private
= &event
;
500 bio
->bi_end_io
= bi_complete
;
502 wait_for_completion(&event
);
504 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
508 EXPORT_SYMBOL_GPL(sync_page_io
);
510 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
512 char b
[BDEVNAME_SIZE
];
513 if (!rdev
->sb_page
) {
521 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
527 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
528 bdevname(rdev
->bdev
,b
));
532 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
534 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
535 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
536 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
537 (sb1
->set_uuid3
== sb2
->set_uuid3
))
545 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
548 mdp_super_t
*tmp1
, *tmp2
;
550 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
551 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
553 if (!tmp1
|| !tmp2
) {
555 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
563 * nr_disks is not constant
568 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
579 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
581 unsigned int disk_csum
, csum
;
583 disk_csum
= sb
->sb_csum
;
585 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
586 sb
->sb_csum
= disk_csum
;
592 * Handle superblock details.
593 * We want to be able to handle multiple superblock formats
594 * so we have a common interface to them all, and an array of
595 * different handlers.
596 * We rely on user-space to write the initial superblock, and support
597 * reading and updating of superblocks.
598 * Interface methods are:
599 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
600 * loads and validates a superblock on dev.
601 * if refdev != NULL, compare superblocks on both devices
603 * 0 - dev has a superblock that is compatible with refdev
604 * 1 - dev has a superblock that is compatible and newer than refdev
605 * so dev should be used as the refdev in future
606 * -EINVAL superblock incompatible or invalid
607 * -othererror e.g. -EIO
609 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
610 * Verify that dev is acceptable into mddev.
611 * The first time, mddev->raid_disks will be 0, and data from
612 * dev should be merged in. Subsequent calls check that dev
613 * is new enough. Return 0 or -EINVAL
615 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
616 * Update the superblock for rdev with data in mddev
617 * This does not write to disc.
623 struct module
*owner
;
624 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
625 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
626 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
630 * load_super for 0.90.0
632 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
634 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
640 * Calculate the position of the superblock,
641 * it's at the end of the disk.
643 * It also happens to be a multiple of 4Kb.
645 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
646 rdev
->sb_offset
= sb_offset
;
648 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
653 bdevname(rdev
->bdev
, b
);
654 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
656 if (sb
->md_magic
!= MD_SB_MAGIC
) {
657 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
662 if (sb
->major_version
!= 0 ||
663 sb
->minor_version
!= 90) {
664 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
665 sb
->major_version
, sb
->minor_version
,
670 if (sb
->raid_disks
<= 0)
673 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
674 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
679 rdev
->preferred_minor
= sb
->md_minor
;
680 rdev
->data_offset
= 0;
681 rdev
->sb_size
= MD_SB_BYTES
;
683 if (sb
->level
== LEVEL_MULTIPATH
)
686 rdev
->desc_nr
= sb
->this_disk
.number
;
692 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
693 if (!uuid_equal(refsb
, sb
)) {
694 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
695 b
, bdevname(refdev
->bdev
,b2
));
698 if (!sb_equal(refsb
, sb
)) {
699 printk(KERN_WARNING
"md: %s has same UUID"
700 " but different superblock to %s\n",
701 b
, bdevname(refdev
->bdev
, b2
));
705 ev2
= md_event(refsb
);
711 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
713 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
714 /* "this cannot possibly happen" ... */
722 * validate_super for 0.90.0
724 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
727 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
729 rdev
->raid_disk
= -1;
731 if (mddev
->raid_disks
== 0) {
732 mddev
->major_version
= 0;
733 mddev
->minor_version
= sb
->minor_version
;
734 mddev
->patch_version
= sb
->patch_version
;
735 mddev
->persistent
= ! sb
->not_persistent
;
736 mddev
->chunk_size
= sb
->chunk_size
;
737 mddev
->ctime
= sb
->ctime
;
738 mddev
->utime
= sb
->utime
;
739 mddev
->level
= sb
->level
;
740 mddev
->clevel
[0] = 0;
741 mddev
->layout
= sb
->layout
;
742 mddev
->raid_disks
= sb
->raid_disks
;
743 mddev
->size
= sb
->size
;
744 mddev
->events
= md_event(sb
);
745 mddev
->bitmap_offset
= 0;
746 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
748 if (sb
->state
& (1<<MD_SB_CLEAN
))
749 mddev
->recovery_cp
= MaxSector
;
751 if (sb
->events_hi
== sb
->cp_events_hi
&&
752 sb
->events_lo
== sb
->cp_events_lo
) {
753 mddev
->recovery_cp
= sb
->recovery_cp
;
755 mddev
->recovery_cp
= 0;
758 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
759 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
760 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
761 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
763 mddev
->max_disks
= MD_SB_DISKS
;
765 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
766 mddev
->bitmap_file
== NULL
) {
767 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
768 && mddev
->level
!= 10) {
769 /* FIXME use a better test */
770 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
773 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
776 } else if (mddev
->pers
== NULL
) {
777 /* Insist on good event counter while assembling */
778 __u64 ev1
= md_event(sb
);
780 if (ev1
< mddev
->events
)
782 } else if (mddev
->bitmap
) {
783 /* if adding to array with a bitmap, then we can accept an
784 * older device ... but not too old.
786 __u64 ev1
= md_event(sb
);
787 if (ev1
< mddev
->bitmap
->events_cleared
)
789 } else /* just a hot-add of a new device, leave raid_disk at -1 */
792 if (mddev
->level
!= LEVEL_MULTIPATH
) {
793 desc
= sb
->disks
+ rdev
->desc_nr
;
795 if (desc
->state
& (1<<MD_DISK_FAULTY
))
796 set_bit(Faulty
, &rdev
->flags
);
797 else if (desc
->state
& (1<<MD_DISK_SYNC
) &&
798 desc
->raid_disk
< mddev
->raid_disks
) {
799 set_bit(In_sync
, &rdev
->flags
);
800 rdev
->raid_disk
= desc
->raid_disk
;
802 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
803 set_bit(WriteMostly
, &rdev
->flags
);
804 } else /* MULTIPATH are always insync */
805 set_bit(In_sync
, &rdev
->flags
);
810 * sync_super for 0.90.0
812 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
815 struct list_head
*tmp
;
817 int next_spare
= mddev
->raid_disks
;
820 /* make rdev->sb match mddev data..
823 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
824 * 3/ any empty disks < next_spare become removed
826 * disks[0] gets initialised to REMOVED because
827 * we cannot be sure from other fields if it has
828 * been initialised or not.
831 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
833 rdev
->sb_size
= MD_SB_BYTES
;
835 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
837 memset(sb
, 0, sizeof(*sb
));
839 sb
->md_magic
= MD_SB_MAGIC
;
840 sb
->major_version
= mddev
->major_version
;
841 sb
->minor_version
= mddev
->minor_version
;
842 sb
->patch_version
= mddev
->patch_version
;
843 sb
->gvalid_words
= 0; /* ignored */
844 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
845 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
846 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
847 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
849 sb
->ctime
= mddev
->ctime
;
850 sb
->level
= mddev
->level
;
851 sb
->size
= mddev
->size
;
852 sb
->raid_disks
= mddev
->raid_disks
;
853 sb
->md_minor
= mddev
->md_minor
;
854 sb
->not_persistent
= !mddev
->persistent
;
855 sb
->utime
= mddev
->utime
;
857 sb
->events_hi
= (mddev
->events
>>32);
858 sb
->events_lo
= (u32
)mddev
->events
;
862 sb
->recovery_cp
= mddev
->recovery_cp
;
863 sb
->cp_events_hi
= (mddev
->events
>>32);
864 sb
->cp_events_lo
= (u32
)mddev
->events
;
865 if (mddev
->recovery_cp
== MaxSector
)
866 sb
->state
= (1<< MD_SB_CLEAN
);
870 sb
->layout
= mddev
->layout
;
871 sb
->chunk_size
= mddev
->chunk_size
;
873 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
874 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
876 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
877 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
880 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
881 && !test_bit(Faulty
, &rdev2
->flags
))
882 desc_nr
= rdev2
->raid_disk
;
884 desc_nr
= next_spare
++;
885 rdev2
->desc_nr
= desc_nr
;
886 d
= &sb
->disks
[rdev2
->desc_nr
];
888 d
->number
= rdev2
->desc_nr
;
889 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
890 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
891 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
892 && !test_bit(Faulty
, &rdev2
->flags
))
893 d
->raid_disk
= rdev2
->raid_disk
;
895 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
896 if (test_bit(Faulty
, &rdev2
->flags
)) {
897 d
->state
= (1<<MD_DISK_FAULTY
);
899 } else if (test_bit(In_sync
, &rdev2
->flags
)) {
900 d
->state
= (1<<MD_DISK_ACTIVE
);
901 d
->state
|= (1<<MD_DISK_SYNC
);
909 if (test_bit(WriteMostly
, &rdev2
->flags
))
910 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
912 /* now set the "removed" and "faulty" bits on any missing devices */
913 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
914 mdp_disk_t
*d
= &sb
->disks
[i
];
915 if (d
->state
== 0 && d
->number
== 0) {
918 d
->state
= (1<<MD_DISK_REMOVED
);
919 d
->state
|= (1<<MD_DISK_FAULTY
);
923 sb
->nr_disks
= nr_disks
;
924 sb
->active_disks
= active
;
925 sb
->working_disks
= working
;
926 sb
->failed_disks
= failed
;
927 sb
->spare_disks
= spare
;
929 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
930 sb
->sb_csum
= calc_sb_csum(sb
);
934 * version 1 superblock
937 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
939 unsigned int disk_csum
, csum
;
940 unsigned long long newcsum
;
941 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
942 unsigned int *isuper
= (unsigned int*)sb
;
945 disk_csum
= sb
->sb_csum
;
948 for (i
=0; size
>=4; size
-= 4 )
949 newcsum
+= le32_to_cpu(*isuper
++);
952 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
954 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
955 sb
->sb_csum
= disk_csum
;
956 return cpu_to_le32(csum
);
959 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
961 struct mdp_superblock_1
*sb
;
964 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
968 * Calculate the position of the superblock.
969 * It is always aligned to a 4K boundary and
970 * depeding on minor_version, it can be:
971 * 0: At least 8K, but less than 12K, from end of device
972 * 1: At start of device
973 * 2: 4K from start of device.
975 switch(minor_version
) {
977 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
979 sb_offset
&= ~(sector_t
)(4*2-1);
980 /* convert from sectors to K */
992 rdev
->sb_offset
= sb_offset
;
994 /* superblock is rarely larger than 1K, but it can be larger,
995 * and it is safe to read 4k, so we do that
997 ret
= read_disk_sb(rdev
, 4096);
1001 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1003 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1004 sb
->major_version
!= cpu_to_le32(1) ||
1005 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1006 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1007 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1010 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1011 printk("md: invalid superblock checksum on %s\n",
1012 bdevname(rdev
->bdev
,b
));
1015 if (le64_to_cpu(sb
->data_size
) < 10) {
1016 printk("md: data_size too small on %s\n",
1017 bdevname(rdev
->bdev
,b
));
1020 rdev
->preferred_minor
= 0xffff;
1021 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1022 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1024 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1025 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1026 if (rdev
->sb_size
& bmask
)
1027 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1033 struct mdp_superblock_1
*refsb
=
1034 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1036 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1037 sb
->level
!= refsb
->level
||
1038 sb
->layout
!= refsb
->layout
||
1039 sb
->chunksize
!= refsb
->chunksize
) {
1040 printk(KERN_WARNING
"md: %s has strangely different"
1041 " superblock to %s\n",
1042 bdevname(rdev
->bdev
,b
),
1043 bdevname(refdev
->bdev
,b2
));
1046 ev1
= le64_to_cpu(sb
->events
);
1047 ev2
= le64_to_cpu(refsb
->events
);
1055 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1057 rdev
->size
= rdev
->sb_offset
;
1058 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1060 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1061 if (le32_to_cpu(sb
->chunksize
))
1062 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1064 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1069 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1071 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1073 rdev
->raid_disk
= -1;
1075 if (mddev
->raid_disks
== 0) {
1076 mddev
->major_version
= 1;
1077 mddev
->patch_version
= 0;
1078 mddev
->persistent
= 1;
1079 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1080 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1081 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1082 mddev
->level
= le32_to_cpu(sb
->level
);
1083 mddev
->clevel
[0] = 0;
1084 mddev
->layout
= le32_to_cpu(sb
->layout
);
1085 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1086 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1087 mddev
->events
= le64_to_cpu(sb
->events
);
1088 mddev
->bitmap_offset
= 0;
1089 mddev
->default_bitmap_offset
= 1024 >> 9;
1091 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1092 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1094 mddev
->max_disks
= (4096-256)/2;
1096 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1097 mddev
->bitmap_file
== NULL
) {
1098 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1099 && mddev
->level
!= 10) {
1100 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1103 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1105 } else if (mddev
->pers
== NULL
) {
1106 /* Insist of good event counter while assembling */
1107 __u64 ev1
= le64_to_cpu(sb
->events
);
1109 if (ev1
< mddev
->events
)
1111 } else if (mddev
->bitmap
) {
1112 /* If adding to array with a bitmap, then we can accept an
1113 * older device, but not too old.
1115 __u64 ev1
= le64_to_cpu(sb
->events
);
1116 if (ev1
< mddev
->bitmap
->events_cleared
)
1118 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1121 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1123 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1124 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1126 case 0xffff: /* spare */
1128 case 0xfffe: /* faulty */
1129 set_bit(Faulty
, &rdev
->flags
);
1132 set_bit(In_sync
, &rdev
->flags
);
1133 rdev
->raid_disk
= role
;
1136 if (sb
->devflags
& WriteMostly1
)
1137 set_bit(WriteMostly
, &rdev
->flags
);
1138 } else /* MULTIPATH are always insync */
1139 set_bit(In_sync
, &rdev
->flags
);
1144 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1146 struct mdp_superblock_1
*sb
;
1147 struct list_head
*tmp
;
1150 /* make rdev->sb match mddev and rdev data. */
1152 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1154 sb
->feature_map
= 0;
1156 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1157 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1158 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1160 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1161 sb
->events
= cpu_to_le64(mddev
->events
);
1163 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1165 sb
->resync_offset
= cpu_to_le64(0);
1167 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1169 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1170 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1172 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1173 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1174 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1178 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1179 if (rdev2
->desc_nr
+1 > max_dev
)
1180 max_dev
= rdev2
->desc_nr
+1;
1182 sb
->max_dev
= cpu_to_le32(max_dev
);
1183 for (i
=0; i
<max_dev
;i
++)
1184 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1186 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1188 if (test_bit(Faulty
, &rdev2
->flags
))
1189 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1190 else if (test_bit(In_sync
, &rdev2
->flags
))
1191 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1193 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1196 sb
->recovery_offset
= cpu_to_le64(0); /* not supported yet */
1197 sb
->sb_csum
= calc_sb_1_csum(sb
);
1201 static struct super_type super_types
[] = {
1204 .owner
= THIS_MODULE
,
1205 .load_super
= super_90_load
,
1206 .validate_super
= super_90_validate
,
1207 .sync_super
= super_90_sync
,
1211 .owner
= THIS_MODULE
,
1212 .load_super
= super_1_load
,
1213 .validate_super
= super_1_validate
,
1214 .sync_super
= super_1_sync
,
1218 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1220 struct list_head
*tmp
;
1223 ITERATE_RDEV(mddev
,rdev
,tmp
)
1224 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1230 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1232 struct list_head
*tmp
;
1235 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1236 if (match_dev_unit(mddev2
, rdev
))
1242 static LIST_HEAD(pending_raid_disks
);
1244 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1246 mdk_rdev_t
*same_pdev
;
1247 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1255 /* make sure rdev->size exceeds mddev->size */
1256 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1258 /* Cannot change size, so fail */
1261 mddev
->size
= rdev
->size
;
1263 same_pdev
= match_dev_unit(mddev
, rdev
);
1266 "%s: WARNING: %s appears to be on the same physical"
1267 " disk as %s. True\n protection against single-disk"
1268 " failure might be compromised.\n",
1269 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1270 bdevname(same_pdev
->bdev
,b2
));
1272 /* Verify rdev->desc_nr is unique.
1273 * If it is -1, assign a free number, else
1274 * check number is not in use
1276 if (rdev
->desc_nr
< 0) {
1278 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1279 while (find_rdev_nr(mddev
, choice
))
1281 rdev
->desc_nr
= choice
;
1283 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1286 bdevname(rdev
->bdev
,b
);
1287 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1289 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1292 list_add(&rdev
->same_set
, &mddev
->disks
);
1293 rdev
->mddev
= mddev
;
1294 printk(KERN_INFO
"md: bind<%s>\n", b
);
1296 rdev
->kobj
.parent
= &mddev
->kobj
;
1297 kobject_add(&rdev
->kobj
);
1299 if (rdev
->bdev
->bd_part
)
1300 ko
= &rdev
->bdev
->bd_part
->kobj
;
1302 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1303 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1304 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1308 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1310 char b
[BDEVNAME_SIZE
];
1315 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1316 list_del_init(&rdev
->same_set
);
1317 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1319 sysfs_remove_link(&rdev
->kobj
, "block");
1320 kobject_del(&rdev
->kobj
);
1324 * prevent the device from being mounted, repartitioned or
1325 * otherwise reused by a RAID array (or any other kernel
1326 * subsystem), by bd_claiming the device.
1328 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1331 struct block_device
*bdev
;
1332 char b
[BDEVNAME_SIZE
];
1334 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1336 printk(KERN_ERR
"md: could not open %s.\n",
1337 __bdevname(dev
, b
));
1338 return PTR_ERR(bdev
);
1340 err
= bd_claim(bdev
, rdev
);
1342 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1351 static void unlock_rdev(mdk_rdev_t
*rdev
)
1353 struct block_device
*bdev
= rdev
->bdev
;
1361 void md_autodetect_dev(dev_t dev
);
1363 static void export_rdev(mdk_rdev_t
* rdev
)
1365 char b
[BDEVNAME_SIZE
];
1366 printk(KERN_INFO
"md: export_rdev(%s)\n",
1367 bdevname(rdev
->bdev
,b
));
1371 list_del_init(&rdev
->same_set
);
1373 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1376 kobject_put(&rdev
->kobj
);
1379 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1381 unbind_rdev_from_array(rdev
);
1385 static void export_array(mddev_t
*mddev
)
1387 struct list_head
*tmp
;
1390 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1395 kick_rdev_from_array(rdev
);
1397 if (!list_empty(&mddev
->disks
))
1399 mddev
->raid_disks
= 0;
1400 mddev
->major_version
= 0;
1403 static void print_desc(mdp_disk_t
*desc
)
1405 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1406 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1409 static void print_sb(mdp_super_t
*sb
)
1414 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1415 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1416 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1418 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1419 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1420 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1421 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1422 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1423 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1424 sb
->failed_disks
, sb
->spare_disks
,
1425 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1428 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1431 desc
= sb
->disks
+ i
;
1432 if (desc
->number
|| desc
->major
|| desc
->minor
||
1433 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1434 printk(" D %2d: ", i
);
1438 printk(KERN_INFO
"md: THIS: ");
1439 print_desc(&sb
->this_disk
);
1443 static void print_rdev(mdk_rdev_t
*rdev
)
1445 char b
[BDEVNAME_SIZE
];
1446 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1447 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1448 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1450 if (rdev
->sb_loaded
) {
1451 printk(KERN_INFO
"md: rdev superblock:\n");
1452 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1454 printk(KERN_INFO
"md: no rdev superblock!\n");
1457 void md_print_devices(void)
1459 struct list_head
*tmp
, *tmp2
;
1462 char b
[BDEVNAME_SIZE
];
1465 printk("md: **********************************\n");
1466 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1467 printk("md: **********************************\n");
1468 ITERATE_MDDEV(mddev
,tmp
) {
1471 bitmap_print_sb(mddev
->bitmap
);
1473 printk("%s: ", mdname(mddev
));
1474 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1475 printk("<%s>", bdevname(rdev
->bdev
,b
));
1478 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1481 printk("md: **********************************\n");
1486 static void sync_sbs(mddev_t
* mddev
)
1489 struct list_head
*tmp
;
1491 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1492 super_types
[mddev
->major_version
].
1493 sync_super(mddev
, rdev
);
1494 rdev
->sb_loaded
= 1;
1498 static void md_update_sb(mddev_t
* mddev
)
1501 struct list_head
*tmp
;
1506 spin_lock_irq(&mddev
->write_lock
);
1507 sync_req
= mddev
->in_sync
;
1508 mddev
->utime
= get_seconds();
1511 if (!mddev
->events
) {
1513 * oops, this 64-bit counter should never wrap.
1514 * Either we are in around ~1 trillion A.C., assuming
1515 * 1 reboot per second, or we have a bug:
1520 mddev
->sb_dirty
= 2;
1524 * do not write anything to disk if using
1525 * nonpersistent superblocks
1527 if (!mddev
->persistent
) {
1528 mddev
->sb_dirty
= 0;
1529 spin_unlock_irq(&mddev
->write_lock
);
1530 wake_up(&mddev
->sb_wait
);
1533 spin_unlock_irq(&mddev
->write_lock
);
1536 "md: updating %s RAID superblock on device (in sync %d)\n",
1537 mdname(mddev
),mddev
->in_sync
);
1539 err
= bitmap_update_sb(mddev
->bitmap
);
1540 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1541 char b
[BDEVNAME_SIZE
];
1542 dprintk(KERN_INFO
"md: ");
1543 if (test_bit(Faulty
, &rdev
->flags
))
1544 dprintk("(skipping faulty ");
1546 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1547 if (!test_bit(Faulty
, &rdev
->flags
)) {
1548 md_super_write(mddev
,rdev
,
1549 rdev
->sb_offset
<<1, rdev
->sb_size
,
1551 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1552 bdevname(rdev
->bdev
,b
),
1553 (unsigned long long)rdev
->sb_offset
);
1557 if (mddev
->level
== LEVEL_MULTIPATH
)
1558 /* only need to write one superblock... */
1561 md_super_wait(mddev
);
1562 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1564 spin_lock_irq(&mddev
->write_lock
);
1565 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1566 /* have to write it out again */
1567 spin_unlock_irq(&mddev
->write_lock
);
1570 mddev
->sb_dirty
= 0;
1571 spin_unlock_irq(&mddev
->write_lock
);
1572 wake_up(&mddev
->sb_wait
);
1576 /* words written to sysfs files may, or my not, be \n terminated.
1577 * We want to accept with case. For this we use cmd_match.
1579 static int cmd_match(const char *cmd
, const char *str
)
1581 /* See if cmd, written into a sysfs file, matches
1582 * str. They must either be the same, or cmd can
1583 * have a trailing newline
1585 while (*cmd
&& *str
&& *cmd
== *str
) {
1596 struct rdev_sysfs_entry
{
1597 struct attribute attr
;
1598 ssize_t (*show
)(mdk_rdev_t
*, char *);
1599 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1603 state_show(mdk_rdev_t
*rdev
, char *page
)
1608 if (test_bit(Faulty
, &rdev
->flags
)) {
1609 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1612 if (test_bit(In_sync
, &rdev
->flags
)) {
1613 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1616 if (!test_bit(Faulty
, &rdev
->flags
) &&
1617 !test_bit(In_sync
, &rdev
->flags
)) {
1618 len
+= sprintf(page
+len
, "%sspare", sep
);
1621 return len
+sprintf(page
+len
, "\n");
1624 static struct rdev_sysfs_entry
1625 rdev_state
= __ATTR_RO(state
);
1628 super_show(mdk_rdev_t
*rdev
, char *page
)
1630 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1631 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1632 return rdev
->sb_size
;
1636 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1639 errors_show(mdk_rdev_t
*rdev
, char *page
)
1641 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1645 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1648 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1649 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1650 atomic_set(&rdev
->corrected_errors
, n
);
1655 static struct rdev_sysfs_entry rdev_errors
=
1656 __ATTR(errors
, 0644, errors_show
, errors_store
);
1659 slot_show(mdk_rdev_t
*rdev
, char *page
)
1661 if (rdev
->raid_disk
< 0)
1662 return sprintf(page
, "none\n");
1664 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1668 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1671 int slot
= simple_strtoul(buf
, &e
, 10);
1672 if (strncmp(buf
, "none", 4)==0)
1674 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1676 if (rdev
->mddev
->pers
)
1677 /* Cannot set slot in active array (yet) */
1679 if (slot
>= rdev
->mddev
->raid_disks
)
1681 rdev
->raid_disk
= slot
;
1682 /* assume it is working */
1684 set_bit(In_sync
, &rdev
->flags
);
1689 static struct rdev_sysfs_entry rdev_slot
=
1690 __ATTR(slot
, 0644, slot_show
, slot_store
);
1693 offset_show(mdk_rdev_t
*rdev
, char *page
)
1695 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1699 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1702 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1703 if (e
==buf
|| (*e
&& *e
!= '\n'))
1705 if (rdev
->mddev
->pers
)
1707 rdev
->data_offset
= offset
;
1711 static struct rdev_sysfs_entry rdev_offset
=
1712 __ATTR(offset
, 0644, offset_show
, offset_store
);
1715 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1717 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1721 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1724 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1725 if (e
==buf
|| (*e
&& *e
!= '\n'))
1727 if (rdev
->mddev
->pers
)
1730 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1731 rdev
->mddev
->size
= size
;
1735 static struct rdev_sysfs_entry rdev_size
=
1736 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1738 static struct attribute
*rdev_default_attrs
[] = {
1748 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1750 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1751 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1755 return entry
->show(rdev
, page
);
1759 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1760 const char *page
, size_t length
)
1762 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1763 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1767 return entry
->store(rdev
, page
, length
);
1770 static void rdev_free(struct kobject
*ko
)
1772 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1775 static struct sysfs_ops rdev_sysfs_ops
= {
1776 .show
= rdev_attr_show
,
1777 .store
= rdev_attr_store
,
1779 static struct kobj_type rdev_ktype
= {
1780 .release
= rdev_free
,
1781 .sysfs_ops
= &rdev_sysfs_ops
,
1782 .default_attrs
= rdev_default_attrs
,
1786 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1788 * mark the device faulty if:
1790 * - the device is nonexistent (zero size)
1791 * - the device has no valid superblock
1793 * a faulty rdev _never_ has rdev->sb set.
1795 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1797 char b
[BDEVNAME_SIZE
];
1802 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1804 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1805 return ERR_PTR(-ENOMEM
);
1808 if ((err
= alloc_disk_sb(rdev
)))
1811 err
= lock_rdev(rdev
, newdev
);
1815 rdev
->kobj
.parent
= NULL
;
1816 rdev
->kobj
.ktype
= &rdev_ktype
;
1817 kobject_init(&rdev
->kobj
);
1821 rdev
->data_offset
= 0;
1822 atomic_set(&rdev
->nr_pending
, 0);
1823 atomic_set(&rdev
->read_errors
, 0);
1824 atomic_set(&rdev
->corrected_errors
, 0);
1826 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1829 "md: %s has zero or unknown size, marking faulty!\n",
1830 bdevname(rdev
->bdev
,b
));
1835 if (super_format
>= 0) {
1836 err
= super_types
[super_format
].
1837 load_super(rdev
, NULL
, super_minor
);
1838 if (err
== -EINVAL
) {
1840 "md: %s has invalid sb, not importing!\n",
1841 bdevname(rdev
->bdev
,b
));
1846 "md: could not read %s's sb, not importing!\n",
1847 bdevname(rdev
->bdev
,b
));
1851 INIT_LIST_HEAD(&rdev
->same_set
);
1856 if (rdev
->sb_page
) {
1862 return ERR_PTR(err
);
1866 * Check a full RAID array for plausibility
1870 static void analyze_sbs(mddev_t
* mddev
)
1873 struct list_head
*tmp
;
1874 mdk_rdev_t
*rdev
, *freshest
;
1875 char b
[BDEVNAME_SIZE
];
1878 ITERATE_RDEV(mddev
,rdev
,tmp
)
1879 switch (super_types
[mddev
->major_version
].
1880 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1888 "md: fatal superblock inconsistency in %s"
1889 " -- removing from array\n",
1890 bdevname(rdev
->bdev
,b
));
1891 kick_rdev_from_array(rdev
);
1895 super_types
[mddev
->major_version
].
1896 validate_super(mddev
, freshest
);
1899 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1900 if (rdev
!= freshest
)
1901 if (super_types
[mddev
->major_version
].
1902 validate_super(mddev
, rdev
)) {
1903 printk(KERN_WARNING
"md: kicking non-fresh %s"
1905 bdevname(rdev
->bdev
,b
));
1906 kick_rdev_from_array(rdev
);
1909 if (mddev
->level
== LEVEL_MULTIPATH
) {
1910 rdev
->desc_nr
= i
++;
1911 rdev
->raid_disk
= rdev
->desc_nr
;
1912 set_bit(In_sync
, &rdev
->flags
);
1918 if (mddev
->recovery_cp
!= MaxSector
&&
1920 printk(KERN_ERR
"md: %s: raid array is not clean"
1921 " -- starting background reconstruction\n",
1927 level_show(mddev_t
*mddev
, char *page
)
1929 struct mdk_personality
*p
= mddev
->pers
;
1931 return sprintf(page
, "%s\n", p
->name
);
1932 else if (mddev
->clevel
[0])
1933 return sprintf(page
, "%s\n", mddev
->clevel
);
1934 else if (mddev
->level
!= LEVEL_NONE
)
1935 return sprintf(page
, "%d\n", mddev
->level
);
1941 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1948 if (len
>= sizeof(mddev
->clevel
))
1950 strncpy(mddev
->clevel
, buf
, len
);
1951 if (mddev
->clevel
[len
-1] == '\n')
1953 mddev
->clevel
[len
] = 0;
1954 mddev
->level
= LEVEL_NONE
;
1958 static struct md_sysfs_entry md_level
=
1959 __ATTR(level
, 0644, level_show
, level_store
);
1962 raid_disks_show(mddev_t
*mddev
, char *page
)
1964 if (mddev
->raid_disks
== 0)
1966 return sprintf(page
, "%d\n", mddev
->raid_disks
);
1969 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
1972 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1974 /* can only set raid_disks if array is not yet active */
1977 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1979 if (!*buf
|| (*e
&& *e
!= '\n'))
1983 rv
= update_raid_disks(mddev
, n
);
1985 mddev
->raid_disks
= n
;
1986 return rv
? rv
: len
;
1988 static struct md_sysfs_entry md_raid_disks
=
1989 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
1992 chunk_size_show(mddev_t
*mddev
, char *page
)
1994 return sprintf(page
, "%d\n", mddev
->chunk_size
);
1998 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2000 /* can only set chunk_size if array is not yet active */
2002 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2006 if (!*buf
|| (*e
&& *e
!= '\n'))
2009 mddev
->chunk_size
= n
;
2012 static struct md_sysfs_entry md_chunk_size
=
2013 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2016 null_show(mddev_t
*mddev
, char *page
)
2022 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2024 /* buf must be %d:%d\n? giving major and minor numbers */
2025 /* The new device is added to the array.
2026 * If the array has a persistent superblock, we read the
2027 * superblock to initialise info and check validity.
2028 * Otherwise, only checking done is that in bind_rdev_to_array,
2029 * which mainly checks size.
2032 int major
= simple_strtoul(buf
, &e
, 10);
2038 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2040 minor
= simple_strtoul(e
+1, &e
, 10);
2041 if (*e
&& *e
!= '\n')
2043 dev
= MKDEV(major
, minor
);
2044 if (major
!= MAJOR(dev
) ||
2045 minor
!= MINOR(dev
))
2049 if (mddev
->persistent
) {
2050 rdev
= md_import_device(dev
, mddev
->major_version
,
2051 mddev
->minor_version
);
2052 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2053 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2054 mdk_rdev_t
, same_set
);
2055 err
= super_types
[mddev
->major_version
]
2056 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2061 rdev
= md_import_device(dev
, -1, -1);
2064 return PTR_ERR(rdev
);
2065 err
= bind_rdev_to_array(rdev
, mddev
);
2069 return err
? err
: len
;
2072 static struct md_sysfs_entry md_new_device
=
2073 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2076 size_show(mddev_t
*mddev
, char *page
)
2078 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2081 static int update_size(mddev_t
*mddev
, unsigned long size
);
2084 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2086 /* If array is inactive, we can reduce the component size, but
2087 * not increase it (except from 0).
2088 * If array is active, we can try an on-line resize
2092 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2093 if (!*buf
|| *buf
== '\n' ||
2098 err
= update_size(mddev
, size
);
2099 md_update_sb(mddev
);
2101 if (mddev
->size
== 0 ||
2107 return err
? err
: len
;
2110 static struct md_sysfs_entry md_size
=
2111 __ATTR(component_size
, 0644, size_show
, size_store
);
2115 * This is either 'none' for arrays with externally managed metadata,
2116 * or N.M for internally known formats
2119 metadata_show(mddev_t
*mddev
, char *page
)
2121 if (mddev
->persistent
)
2122 return sprintf(page
, "%d.%d\n",
2123 mddev
->major_version
, mddev
->minor_version
);
2125 return sprintf(page
, "none\n");
2129 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2133 if (!list_empty(&mddev
->disks
))
2136 if (cmd_match(buf
, "none")) {
2137 mddev
->persistent
= 0;
2138 mddev
->major_version
= 0;
2139 mddev
->minor_version
= 90;
2142 major
= simple_strtoul(buf
, &e
, 10);
2143 if (e
==buf
|| *e
!= '.')
2146 minor
= simple_strtoul(buf
, &e
, 10);
2147 if (e
==buf
|| *e
!= '\n')
2149 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2150 super_types
[major
].name
== NULL
)
2152 mddev
->major_version
= major
;
2153 mddev
->minor_version
= minor
;
2154 mddev
->persistent
= 1;
2158 static struct md_sysfs_entry md_metadata
=
2159 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2162 action_show(mddev_t
*mddev
, char *page
)
2164 char *type
= "idle";
2165 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2166 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2167 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2168 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2170 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2177 return sprintf(page
, "%s\n", type
);
2181 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2183 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2186 if (cmd_match(page
, "idle")) {
2187 if (mddev
->sync_thread
) {
2188 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2189 md_unregister_thread(mddev
->sync_thread
);
2190 mddev
->sync_thread
= NULL
;
2191 mddev
->recovery
= 0;
2193 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2194 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2196 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2197 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2199 if (cmd_match(page
, "check"))
2200 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2201 else if (cmd_match(page
, "repair"))
2203 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2204 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2206 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2207 md_wakeup_thread(mddev
->thread
);
2212 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2214 return sprintf(page
, "%llu\n",
2215 (unsigned long long) mddev
->resync_mismatches
);
2218 static struct md_sysfs_entry
2219 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2222 static struct md_sysfs_entry
2223 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2226 sync_min_show(mddev_t
*mddev
, char *page
)
2228 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2229 mddev
->sync_speed_min
? "local": "system");
2233 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2237 if (strncmp(buf
, "system", 6)==0) {
2238 mddev
->sync_speed_min
= 0;
2241 min
= simple_strtoul(buf
, &e
, 10);
2242 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2244 mddev
->sync_speed_min
= min
;
2248 static struct md_sysfs_entry md_sync_min
=
2249 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2252 sync_max_show(mddev_t
*mddev
, char *page
)
2254 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2255 mddev
->sync_speed_max
? "local": "system");
2259 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2263 if (strncmp(buf
, "system", 6)==0) {
2264 mddev
->sync_speed_max
= 0;
2267 max
= simple_strtoul(buf
, &e
, 10);
2268 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2270 mddev
->sync_speed_max
= max
;
2274 static struct md_sysfs_entry md_sync_max
=
2275 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2279 sync_speed_show(mddev_t
*mddev
, char *page
)
2281 unsigned long resync
, dt
, db
;
2282 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2283 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2285 db
= resync
- (mddev
->resync_mark_cnt
);
2286 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2289 static struct md_sysfs_entry
2290 md_sync_speed
= __ATTR_RO(sync_speed
);
2293 sync_completed_show(mddev_t
*mddev
, char *page
)
2295 unsigned long max_blocks
, resync
;
2297 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2298 max_blocks
= mddev
->resync_max_sectors
;
2300 max_blocks
= mddev
->size
<< 1;
2302 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2303 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2306 static struct md_sysfs_entry
2307 md_sync_completed
= __ATTR_RO(sync_completed
);
2309 static struct attribute
*md_default_attrs
[] = {
2311 &md_raid_disks
.attr
,
2312 &md_chunk_size
.attr
,
2315 &md_new_device
.attr
,
2319 static struct attribute
*md_redundancy_attrs
[] = {
2321 &md_mismatches
.attr
,
2324 &md_sync_speed
.attr
,
2325 &md_sync_completed
.attr
,
2328 static struct attribute_group md_redundancy_group
= {
2330 .attrs
= md_redundancy_attrs
,
2335 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2337 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2338 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2344 rv
= entry
->show(mddev
, page
);
2345 mddev_unlock(mddev
);
2350 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2351 const char *page
, size_t length
)
2353 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2354 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2360 rv
= entry
->store(mddev
, page
, length
);
2361 mddev_unlock(mddev
);
2365 static void md_free(struct kobject
*ko
)
2367 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2371 static struct sysfs_ops md_sysfs_ops
= {
2372 .show
= md_attr_show
,
2373 .store
= md_attr_store
,
2375 static struct kobj_type md_ktype
= {
2377 .sysfs_ops
= &md_sysfs_ops
,
2378 .default_attrs
= md_default_attrs
,
2383 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2385 static DECLARE_MUTEX(disks_sem
);
2386 mddev_t
*mddev
= mddev_find(dev
);
2387 struct gendisk
*disk
;
2388 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2389 int shift
= partitioned
? MdpMinorShift
: 0;
2390 int unit
= MINOR(dev
) >> shift
;
2396 if (mddev
->gendisk
) {
2401 disk
= alloc_disk(1 << shift
);
2407 disk
->major
= MAJOR(dev
);
2408 disk
->first_minor
= unit
<< shift
;
2410 sprintf(disk
->disk_name
, "md_d%d", unit
);
2411 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2413 sprintf(disk
->disk_name
, "md%d", unit
);
2414 sprintf(disk
->devfs_name
, "md/%d", unit
);
2416 disk
->fops
= &md_fops
;
2417 disk
->private_data
= mddev
;
2418 disk
->queue
= mddev
->queue
;
2420 mddev
->gendisk
= disk
;
2422 mddev
->kobj
.parent
= &disk
->kobj
;
2423 mddev
->kobj
.k_name
= NULL
;
2424 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2425 mddev
->kobj
.ktype
= &md_ktype
;
2426 kobject_register(&mddev
->kobj
);
2430 void md_wakeup_thread(mdk_thread_t
*thread
);
2432 static void md_safemode_timeout(unsigned long data
)
2434 mddev_t
*mddev
= (mddev_t
*) data
;
2436 mddev
->safemode
= 1;
2437 md_wakeup_thread(mddev
->thread
);
2440 static int start_dirty_degraded
;
2442 static int do_md_run(mddev_t
* mddev
)
2446 struct list_head
*tmp
;
2448 struct gendisk
*disk
;
2449 struct mdk_personality
*pers
;
2450 char b
[BDEVNAME_SIZE
];
2452 if (list_empty(&mddev
->disks
))
2453 /* cannot run an array with no devices.. */
2460 * Analyze all RAID superblock(s)
2462 if (!mddev
->raid_disks
)
2465 chunk_size
= mddev
->chunk_size
;
2468 if (chunk_size
> MAX_CHUNK_SIZE
) {
2469 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2470 chunk_size
, MAX_CHUNK_SIZE
);
2474 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2476 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2477 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2480 if (chunk_size
< PAGE_SIZE
) {
2481 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2482 chunk_size
, PAGE_SIZE
);
2486 /* devices must have minimum size of one chunk */
2487 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2488 if (test_bit(Faulty
, &rdev
->flags
))
2490 if (rdev
->size
< chunk_size
/ 1024) {
2492 "md: Dev %s smaller than chunk_size:"
2494 bdevname(rdev
->bdev
,b
),
2495 (unsigned long long)rdev
->size
,
2503 if (mddev
->level
!= LEVEL_NONE
)
2504 request_module("md-level-%d", mddev
->level
);
2505 else if (mddev
->clevel
[0])
2506 request_module("md-%s", mddev
->clevel
);
2510 * Drop all container device buffers, from now on
2511 * the only valid external interface is through the md
2513 * Also find largest hardsector size
2515 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2516 if (test_bit(Faulty
, &rdev
->flags
))
2518 sync_blockdev(rdev
->bdev
);
2519 invalidate_bdev(rdev
->bdev
, 0);
2522 md_probe(mddev
->unit
, NULL
, NULL
);
2523 disk
= mddev
->gendisk
;
2527 spin_lock(&pers_lock
);
2528 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2529 if (!pers
|| !try_module_get(pers
->owner
)) {
2530 spin_unlock(&pers_lock
);
2531 if (mddev
->level
!= LEVEL_NONE
)
2532 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2535 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2540 spin_unlock(&pers_lock
);
2541 mddev
->level
= pers
->level
;
2542 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2544 mddev
->recovery
= 0;
2545 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2546 mddev
->barriers_work
= 1;
2547 mddev
->ok_start_degraded
= start_dirty_degraded
;
2550 mddev
->ro
= 2; /* read-only, but switch on first write */
2552 err
= mddev
->pers
->run(mddev
);
2553 if (!err
&& mddev
->pers
->sync_request
) {
2554 err
= bitmap_create(mddev
);
2556 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2557 mdname(mddev
), err
);
2558 mddev
->pers
->stop(mddev
);
2562 printk(KERN_ERR
"md: pers->run() failed ...\n");
2563 module_put(mddev
->pers
->owner
);
2565 bitmap_destroy(mddev
);
2568 if (mddev
->pers
->sync_request
)
2569 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2570 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2573 atomic_set(&mddev
->writes_pending
,0);
2574 mddev
->safemode
= 0;
2575 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2576 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2577 mddev
->safemode_delay
= (20 * HZ
)/1000 +1; /* 20 msec delay */
2580 ITERATE_RDEV(mddev
,rdev
,tmp
)
2581 if (rdev
->raid_disk
>= 0) {
2583 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2584 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2587 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2588 md_wakeup_thread(mddev
->thread
);
2590 if (mddev
->sb_dirty
)
2591 md_update_sb(mddev
);
2593 set_capacity(disk
, mddev
->array_size
<<1);
2595 /* If we call blk_queue_make_request here, it will
2596 * re-initialise max_sectors etc which may have been
2597 * refined inside -> run. So just set the bits we need to set.
2598 * Most initialisation happended when we called
2599 * blk_queue_make_request(..., md_fail_request)
2602 mddev
->queue
->queuedata
= mddev
;
2603 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2606 md_new_event(mddev
);
2610 static int restart_array(mddev_t
*mddev
)
2612 struct gendisk
*disk
= mddev
->gendisk
;
2616 * Complain if it has no devices
2619 if (list_empty(&mddev
->disks
))
2627 mddev
->safemode
= 0;
2629 set_disk_ro(disk
, 0);
2631 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2634 * Kick recovery or resync if necessary
2636 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2637 md_wakeup_thread(mddev
->thread
);
2640 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2649 static int do_md_stop(mddev_t
* mddev
, int ro
)
2652 struct gendisk
*disk
= mddev
->gendisk
;
2655 if (atomic_read(&mddev
->active
)>2) {
2656 printk("md: %s still in use.\n",mdname(mddev
));
2660 if (mddev
->sync_thread
) {
2661 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2662 md_unregister_thread(mddev
->sync_thread
);
2663 mddev
->sync_thread
= NULL
;
2666 del_timer_sync(&mddev
->safemode_timer
);
2668 invalidate_partition(disk
, 0);
2676 bitmap_flush(mddev
);
2677 md_super_wait(mddev
);
2679 set_disk_ro(disk
, 0);
2680 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2681 mddev
->pers
->stop(mddev
);
2682 if (mddev
->pers
->sync_request
)
2683 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2685 module_put(mddev
->pers
->owner
);
2690 if (!mddev
->in_sync
) {
2691 /* mark array as shutdown cleanly */
2693 md_update_sb(mddev
);
2696 set_disk_ro(disk
, 1);
2700 * Free resources if final stop
2704 struct list_head
*tmp
;
2705 struct gendisk
*disk
;
2706 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
2708 bitmap_destroy(mddev
);
2709 if (mddev
->bitmap_file
) {
2710 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
2711 fput(mddev
->bitmap_file
);
2712 mddev
->bitmap_file
= NULL
;
2714 mddev
->bitmap_offset
= 0;
2716 ITERATE_RDEV(mddev
,rdev
,tmp
)
2717 if (rdev
->raid_disk
>= 0) {
2719 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2720 sysfs_remove_link(&mddev
->kobj
, nm
);
2723 export_array(mddev
);
2725 mddev
->array_size
= 0;
2726 disk
= mddev
->gendisk
;
2728 set_capacity(disk
, 0);
2731 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
2734 md_new_event(mddev
);
2739 static void autorun_array(mddev_t
*mddev
)
2742 struct list_head
*tmp
;
2745 if (list_empty(&mddev
->disks
))
2748 printk(KERN_INFO
"md: running: ");
2750 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2751 char b
[BDEVNAME_SIZE
];
2752 printk("<%s>", bdevname(rdev
->bdev
,b
));
2756 err
= do_md_run (mddev
);
2758 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
2759 do_md_stop (mddev
, 0);
2764 * lets try to run arrays based on all disks that have arrived
2765 * until now. (those are in pending_raid_disks)
2767 * the method: pick the first pending disk, collect all disks with
2768 * the same UUID, remove all from the pending list and put them into
2769 * the 'same_array' list. Then order this list based on superblock
2770 * update time (freshest comes first), kick out 'old' disks and
2771 * compare superblocks. If everything's fine then run it.
2773 * If "unit" is allocated, then bump its reference count
2775 static void autorun_devices(int part
)
2777 struct list_head candidates
;
2778 struct list_head
*tmp
;
2779 mdk_rdev_t
*rdev0
, *rdev
;
2781 char b
[BDEVNAME_SIZE
];
2783 printk(KERN_INFO
"md: autorun ...\n");
2784 while (!list_empty(&pending_raid_disks
)) {
2786 rdev0
= list_entry(pending_raid_disks
.next
,
2787 mdk_rdev_t
, same_set
);
2789 printk(KERN_INFO
"md: considering %s ...\n",
2790 bdevname(rdev0
->bdev
,b
));
2791 INIT_LIST_HEAD(&candidates
);
2792 ITERATE_RDEV_PENDING(rdev
,tmp
)
2793 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
2794 printk(KERN_INFO
"md: adding %s ...\n",
2795 bdevname(rdev
->bdev
,b
));
2796 list_move(&rdev
->same_set
, &candidates
);
2799 * now we have a set of devices, with all of them having
2800 * mostly sane superblocks. It's time to allocate the
2803 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
2804 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
2805 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
2809 dev
= MKDEV(mdp_major
,
2810 rdev0
->preferred_minor
<< MdpMinorShift
);
2812 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
2814 md_probe(dev
, NULL
, NULL
);
2815 mddev
= mddev_find(dev
);
2818 "md: cannot allocate memory for md drive.\n");
2821 if (mddev_lock(mddev
))
2822 printk(KERN_WARNING
"md: %s locked, cannot run\n",
2824 else if (mddev
->raid_disks
|| mddev
->major_version
2825 || !list_empty(&mddev
->disks
)) {
2827 "md: %s already running, cannot run %s\n",
2828 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
2829 mddev_unlock(mddev
);
2831 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
2832 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
2833 list_del_init(&rdev
->same_set
);
2834 if (bind_rdev_to_array(rdev
, mddev
))
2837 autorun_array(mddev
);
2838 mddev_unlock(mddev
);
2840 /* on success, candidates will be empty, on error
2843 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
2847 printk(KERN_INFO
"md: ... autorun DONE.\n");
2851 * import RAID devices based on one partition
2852 * if possible, the array gets run as well.
2855 static int autostart_array(dev_t startdev
)
2857 char b
[BDEVNAME_SIZE
];
2858 int err
= -EINVAL
, i
;
2859 mdp_super_t
*sb
= NULL
;
2860 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
2862 start_rdev
= md_import_device(startdev
, 0, 0);
2863 if (IS_ERR(start_rdev
))
2867 /* NOTE: this can only work for 0.90.0 superblocks */
2868 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
2869 if (sb
->major_version
!= 0 ||
2870 sb
->minor_version
!= 90 ) {
2871 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
2872 export_rdev(start_rdev
);
2876 if (test_bit(Faulty
, &start_rdev
->flags
)) {
2878 "md: can not autostart based on faulty %s!\n",
2879 bdevname(start_rdev
->bdev
,b
));
2880 export_rdev(start_rdev
);
2883 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
2885 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2886 mdp_disk_t
*desc
= sb
->disks
+ i
;
2887 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
2891 if (dev
== startdev
)
2893 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
2895 rdev
= md_import_device(dev
, 0, 0);
2899 list_add(&rdev
->same_set
, &pending_raid_disks
);
2903 * possibly return codes
2911 static int get_version(void __user
* arg
)
2915 ver
.major
= MD_MAJOR_VERSION
;
2916 ver
.minor
= MD_MINOR_VERSION
;
2917 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
2919 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
2925 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
2927 mdu_array_info_t info
;
2928 int nr
,working
,active
,failed
,spare
;
2930 struct list_head
*tmp
;
2932 nr
=working
=active
=failed
=spare
=0;
2933 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2935 if (test_bit(Faulty
, &rdev
->flags
))
2939 if (test_bit(In_sync
, &rdev
->flags
))
2946 info
.major_version
= mddev
->major_version
;
2947 info
.minor_version
= mddev
->minor_version
;
2948 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
2949 info
.ctime
= mddev
->ctime
;
2950 info
.level
= mddev
->level
;
2951 info
.size
= mddev
->size
;
2952 if (info
.size
!= mddev
->size
) /* overflow */
2955 info
.raid_disks
= mddev
->raid_disks
;
2956 info
.md_minor
= mddev
->md_minor
;
2957 info
.not_persistent
= !mddev
->persistent
;
2959 info
.utime
= mddev
->utime
;
2962 info
.state
= (1<<MD_SB_CLEAN
);
2963 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
2964 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
2965 info
.active_disks
= active
;
2966 info
.working_disks
= working
;
2967 info
.failed_disks
= failed
;
2968 info
.spare_disks
= spare
;
2970 info
.layout
= mddev
->layout
;
2971 info
.chunk_size
= mddev
->chunk_size
;
2973 if (copy_to_user(arg
, &info
, sizeof(info
)))
2979 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
2981 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
2982 char *ptr
, *buf
= NULL
;
2985 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
2989 /* bitmap disabled, zero the first byte and copy out */
2990 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
2991 file
->pathname
[0] = '\0';
2995 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
2999 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3003 strcpy(file
->pathname
, ptr
);
3007 if (copy_to_user(arg
, file
, sizeof(*file
)))
3015 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3017 mdu_disk_info_t info
;
3021 if (copy_from_user(&info
, arg
, sizeof(info
)))
3026 rdev
= find_rdev_nr(mddev
, nr
);
3028 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3029 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3030 info
.raid_disk
= rdev
->raid_disk
;
3032 if (test_bit(Faulty
, &rdev
->flags
))
3033 info
.state
|= (1<<MD_DISK_FAULTY
);
3034 else if (test_bit(In_sync
, &rdev
->flags
)) {
3035 info
.state
|= (1<<MD_DISK_ACTIVE
);
3036 info
.state
|= (1<<MD_DISK_SYNC
);
3038 if (test_bit(WriteMostly
, &rdev
->flags
))
3039 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3041 info
.major
= info
.minor
= 0;
3042 info
.raid_disk
= -1;
3043 info
.state
= (1<<MD_DISK_REMOVED
);
3046 if (copy_to_user(arg
, &info
, sizeof(info
)))
3052 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3054 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3056 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3058 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3061 if (!mddev
->raid_disks
) {
3063 /* expecting a device which has a superblock */
3064 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3067 "md: md_import_device returned %ld\n",
3069 return PTR_ERR(rdev
);
3071 if (!list_empty(&mddev
->disks
)) {
3072 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3073 mdk_rdev_t
, same_set
);
3074 int err
= super_types
[mddev
->major_version
]
3075 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3078 "md: %s has different UUID to %s\n",
3079 bdevname(rdev
->bdev
,b
),
3080 bdevname(rdev0
->bdev
,b2
));
3085 err
= bind_rdev_to_array(rdev
, mddev
);
3092 * add_new_disk can be used once the array is assembled
3093 * to add "hot spares". They must already have a superblock
3098 if (!mddev
->pers
->hot_add_disk
) {
3100 "%s: personality does not support diskops!\n",
3104 if (mddev
->persistent
)
3105 rdev
= md_import_device(dev
, mddev
->major_version
,
3106 mddev
->minor_version
);
3108 rdev
= md_import_device(dev
, -1, -1);
3111 "md: md_import_device returned %ld\n",
3113 return PTR_ERR(rdev
);
3115 /* set save_raid_disk if appropriate */
3116 if (!mddev
->persistent
) {
3117 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3118 info
->raid_disk
< mddev
->raid_disks
)
3119 rdev
->raid_disk
= info
->raid_disk
;
3121 rdev
->raid_disk
= -1;
3123 super_types
[mddev
->major_version
].
3124 validate_super(mddev
, rdev
);
3125 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3127 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3128 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3129 set_bit(WriteMostly
, &rdev
->flags
);
3131 rdev
->raid_disk
= -1;
3132 err
= bind_rdev_to_array(rdev
, mddev
);
3136 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3137 md_wakeup_thread(mddev
->thread
);
3141 /* otherwise, add_new_disk is only allowed
3142 * for major_version==0 superblocks
3144 if (mddev
->major_version
!= 0) {
3145 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3150 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3152 rdev
= md_import_device (dev
, -1, 0);
3155 "md: error, md_import_device() returned %ld\n",
3157 return PTR_ERR(rdev
);
3159 rdev
->desc_nr
= info
->number
;
3160 if (info
->raid_disk
< mddev
->raid_disks
)
3161 rdev
->raid_disk
= info
->raid_disk
;
3163 rdev
->raid_disk
= -1;
3167 if (rdev
->raid_disk
< mddev
->raid_disks
)
3168 if (info
->state
& (1<<MD_DISK_SYNC
))
3169 set_bit(In_sync
, &rdev
->flags
);
3171 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3172 set_bit(WriteMostly
, &rdev
->flags
);
3174 if (!mddev
->persistent
) {
3175 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3176 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3178 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3179 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3181 err
= bind_rdev_to_array(rdev
, mddev
);
3191 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3193 char b
[BDEVNAME_SIZE
];
3199 rdev
= find_rdev(mddev
, dev
);
3203 if (rdev
->raid_disk
>= 0)
3206 kick_rdev_from_array(rdev
);
3207 md_update_sb(mddev
);
3208 md_new_event(mddev
);
3212 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3213 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3217 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3219 char b
[BDEVNAME_SIZE
];
3227 if (mddev
->major_version
!= 0) {
3228 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3229 " version-0 superblocks.\n",
3233 if (!mddev
->pers
->hot_add_disk
) {
3235 "%s: personality does not support diskops!\n",
3240 rdev
= md_import_device (dev
, -1, 0);
3243 "md: error, md_import_device() returned %ld\n",
3248 if (mddev
->persistent
)
3249 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3252 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3254 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3257 if (test_bit(Faulty
, &rdev
->flags
)) {
3259 "md: can not hot-add faulty %s disk to %s!\n",
3260 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3264 clear_bit(In_sync
, &rdev
->flags
);
3266 err
= bind_rdev_to_array(rdev
, mddev
);
3271 * The rest should better be atomic, we can have disk failures
3272 * noticed in interrupt contexts ...
3275 if (rdev
->desc_nr
== mddev
->max_disks
) {
3276 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3279 goto abort_unbind_export
;
3282 rdev
->raid_disk
= -1;
3284 md_update_sb(mddev
);
3287 * Kick recovery, maybe this spare has to be added to the
3288 * array immediately.
3290 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3291 md_wakeup_thread(mddev
->thread
);
3292 md_new_event(mddev
);
3295 abort_unbind_export
:
3296 unbind_rdev_from_array(rdev
);
3303 /* similar to deny_write_access, but accounts for our holding a reference
3304 * to the file ourselves */
3305 static int deny_bitmap_write_access(struct file
* file
)
3307 struct inode
*inode
= file
->f_mapping
->host
;
3309 spin_lock(&inode
->i_lock
);
3310 if (atomic_read(&inode
->i_writecount
) > 1) {
3311 spin_unlock(&inode
->i_lock
);
3314 atomic_set(&inode
->i_writecount
, -1);
3315 spin_unlock(&inode
->i_lock
);
3320 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3325 if (!mddev
->pers
->quiesce
)
3327 if (mddev
->recovery
|| mddev
->sync_thread
)
3329 /* we should be able to change the bitmap.. */
3335 return -EEXIST
; /* cannot add when bitmap is present */
3336 mddev
->bitmap_file
= fget(fd
);
3338 if (mddev
->bitmap_file
== NULL
) {
3339 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3344 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3346 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3348 fput(mddev
->bitmap_file
);
3349 mddev
->bitmap_file
= NULL
;
3352 mddev
->bitmap_offset
= 0; /* file overrides offset */
3353 } else if (mddev
->bitmap
== NULL
)
3354 return -ENOENT
; /* cannot remove what isn't there */
3357 mddev
->pers
->quiesce(mddev
, 1);
3359 err
= bitmap_create(mddev
);
3361 bitmap_destroy(mddev
);
3362 mddev
->pers
->quiesce(mddev
, 0);
3363 } else if (fd
< 0) {
3364 if (mddev
->bitmap_file
)
3365 fput(mddev
->bitmap_file
);
3366 mddev
->bitmap_file
= NULL
;
3373 * set_array_info is used two different ways
3374 * The original usage is when creating a new array.
3375 * In this usage, raid_disks is > 0 and it together with
3376 * level, size, not_persistent,layout,chunksize determine the
3377 * shape of the array.
3378 * This will always create an array with a type-0.90.0 superblock.
3379 * The newer usage is when assembling an array.
3380 * In this case raid_disks will be 0, and the major_version field is
3381 * use to determine which style super-blocks are to be found on the devices.
3382 * The minor and patch _version numbers are also kept incase the
3383 * super_block handler wishes to interpret them.
3385 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3388 if (info
->raid_disks
== 0) {
3389 /* just setting version number for superblock loading */
3390 if (info
->major_version
< 0 ||
3391 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3392 super_types
[info
->major_version
].name
== NULL
) {
3393 /* maybe try to auto-load a module? */
3395 "md: superblock version %d not known\n",
3396 info
->major_version
);
3399 mddev
->major_version
= info
->major_version
;
3400 mddev
->minor_version
= info
->minor_version
;
3401 mddev
->patch_version
= info
->patch_version
;
3404 mddev
->major_version
= MD_MAJOR_VERSION
;
3405 mddev
->minor_version
= MD_MINOR_VERSION
;
3406 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3407 mddev
->ctime
= get_seconds();
3409 mddev
->level
= info
->level
;
3410 mddev
->clevel
[0] = 0;
3411 mddev
->size
= info
->size
;
3412 mddev
->raid_disks
= info
->raid_disks
;
3413 /* don't set md_minor, it is determined by which /dev/md* was
3416 if (info
->state
& (1<<MD_SB_CLEAN
))
3417 mddev
->recovery_cp
= MaxSector
;
3419 mddev
->recovery_cp
= 0;
3420 mddev
->persistent
= ! info
->not_persistent
;
3422 mddev
->layout
= info
->layout
;
3423 mddev
->chunk_size
= info
->chunk_size
;
3425 mddev
->max_disks
= MD_SB_DISKS
;
3427 mddev
->sb_dirty
= 1;
3429 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3430 mddev
->bitmap_offset
= 0;
3433 * Generate a 128 bit UUID
3435 get_random_bytes(mddev
->uuid
, 16);
3440 static int update_size(mddev_t
*mddev
, unsigned long size
)
3444 struct list_head
*tmp
;
3446 if (mddev
->pers
->resize
== NULL
)
3448 /* The "size" is the amount of each device that is used.
3449 * This can only make sense for arrays with redundancy.
3450 * linear and raid0 always use whatever space is available
3451 * We can only consider changing the size if no resync
3452 * or reconstruction is happening, and if the new size
3453 * is acceptable. It must fit before the sb_offset or,
3454 * if that is <data_offset, it must fit before the
3455 * size of each device.
3456 * If size is zero, we find the largest size that fits.
3458 if (mddev
->sync_thread
)
3460 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3462 int fit
= (size
== 0);
3463 if (rdev
->sb_offset
> rdev
->data_offset
)
3464 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3466 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3467 - rdev
->data_offset
;
3468 if (fit
&& (size
== 0 || size
> avail
/2))
3470 if (avail
< ((sector_t
)size
<< 1))
3473 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3475 struct block_device
*bdev
;
3477 bdev
= bdget_disk(mddev
->gendisk
, 0);
3479 mutex_lock(&bdev
->bd_inode
->i_mutex
);
3480 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
3481 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
3488 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3491 /* change the number of raid disks */
3492 if (mddev
->pers
->reshape
== NULL
)
3494 if (raid_disks
<= 0 ||
3495 raid_disks
>= mddev
->max_disks
)
3497 if (mddev
->sync_thread
)
3499 rv
= mddev
->pers
->reshape(mddev
, raid_disks
);
3505 * update_array_info is used to change the configuration of an
3507 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3508 * fields in the info are checked against the array.
3509 * Any differences that cannot be handled will cause an error.
3510 * Normally, only one change can be managed at a time.
3512 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3518 /* calculate expected state,ignoring low bits */
3519 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3520 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3522 if (mddev
->major_version
!= info
->major_version
||
3523 mddev
->minor_version
!= info
->minor_version
||
3524 /* mddev->patch_version != info->patch_version || */
3525 mddev
->ctime
!= info
->ctime
||
3526 mddev
->level
!= info
->level
||
3527 /* mddev->layout != info->layout || */
3528 !mddev
->persistent
!= info
->not_persistent
||
3529 mddev
->chunk_size
!= info
->chunk_size
||
3530 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3531 ((state
^info
->state
) & 0xfffffe00)
3534 /* Check there is only one change */
3535 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
3536 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3537 if (mddev
->layout
!= info
->layout
) cnt
++;
3538 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3539 if (cnt
== 0) return 0;
3540 if (cnt
> 1) return -EINVAL
;
3542 if (mddev
->layout
!= info
->layout
) {
3544 * we don't need to do anything at the md level, the
3545 * personality will take care of it all.
3547 if (mddev
->pers
->reconfig
== NULL
)
3550 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3552 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
3553 rv
= update_size(mddev
, info
->size
);
3555 if (mddev
->raid_disks
!= info
->raid_disks
)
3556 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3558 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3559 if (mddev
->pers
->quiesce
== NULL
)
3561 if (mddev
->recovery
|| mddev
->sync_thread
)
3563 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3564 /* add the bitmap */
3567 if (mddev
->default_bitmap_offset
== 0)
3569 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3570 mddev
->pers
->quiesce(mddev
, 1);
3571 rv
= bitmap_create(mddev
);
3573 bitmap_destroy(mddev
);
3574 mddev
->pers
->quiesce(mddev
, 0);
3576 /* remove the bitmap */
3579 if (mddev
->bitmap
->file
)
3581 mddev
->pers
->quiesce(mddev
, 1);
3582 bitmap_destroy(mddev
);
3583 mddev
->pers
->quiesce(mddev
, 0);
3584 mddev
->bitmap_offset
= 0;
3587 md_update_sb(mddev
);
3591 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3595 if (mddev
->pers
== NULL
)
3598 rdev
= find_rdev(mddev
, dev
);
3602 md_error(mddev
, rdev
);
3606 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
3608 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
3612 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
3616 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3617 unsigned int cmd
, unsigned long arg
)
3620 void __user
*argp
= (void __user
*)arg
;
3621 mddev_t
*mddev
= NULL
;
3623 if (!capable(CAP_SYS_ADMIN
))
3627 * Commands dealing with the RAID driver but not any
3633 err
= get_version(argp
);
3636 case PRINT_RAID_DEBUG
:
3644 autostart_arrays(arg
);
3651 * Commands creating/starting a new array:
3654 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3662 if (cmd
== START_ARRAY
) {
3663 /* START_ARRAY doesn't need to lock the array as autostart_array
3664 * does the locking, and it could even be a different array
3669 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3670 "This will not be supported beyond July 2006\n",
3671 current
->comm
, current
->pid
);
3674 err
= autostart_array(new_decode_dev(arg
));
3676 printk(KERN_WARNING
"md: autostart failed!\n");
3682 err
= mddev_lock(mddev
);
3685 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3692 case SET_ARRAY_INFO
:
3694 mdu_array_info_t info
;
3696 memset(&info
, 0, sizeof(info
));
3697 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
3702 err
= update_array_info(mddev
, &info
);
3704 printk(KERN_WARNING
"md: couldn't update"
3705 " array info. %d\n", err
);
3710 if (!list_empty(&mddev
->disks
)) {
3712 "md: array %s already has disks!\n",
3717 if (mddev
->raid_disks
) {
3719 "md: array %s already initialised!\n",
3724 err
= set_array_info(mddev
, &info
);
3726 printk(KERN_WARNING
"md: couldn't set"
3727 " array info. %d\n", err
);
3737 * Commands querying/configuring an existing array:
3739 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3740 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3741 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
3742 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
3748 * Commands even a read-only array can execute:
3752 case GET_ARRAY_INFO
:
3753 err
= get_array_info(mddev
, argp
);
3756 case GET_BITMAP_FILE
:
3757 err
= get_bitmap_file(mddev
, argp
);
3761 err
= get_disk_info(mddev
, argp
);
3764 case RESTART_ARRAY_RW
:
3765 err
= restart_array(mddev
);
3769 err
= do_md_stop (mddev
, 0);
3773 err
= do_md_stop (mddev
, 1);
3777 * We have a problem here : there is no easy way to give a CHS
3778 * virtual geometry. We currently pretend that we have a 2 heads
3779 * 4 sectors (with a BIG number of cylinders...). This drives
3780 * dosfs just mad... ;-)
3785 * The remaining ioctls are changing the state of the
3786 * superblock, so we do not allow them on read-only arrays.
3787 * However non-MD ioctls (e.g. get-size) will still come through
3788 * here and hit the 'default' below, so only disallow
3789 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3791 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
3792 mddev
->ro
&& mddev
->pers
) {
3793 if (mddev
->ro
== 2) {
3795 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3796 md_wakeup_thread(mddev
->thread
);
3808 mdu_disk_info_t info
;
3809 if (copy_from_user(&info
, argp
, sizeof(info
)))
3812 err
= add_new_disk(mddev
, &info
);
3816 case HOT_REMOVE_DISK
:
3817 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
3821 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
3824 case SET_DISK_FAULTY
:
3825 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
3829 err
= do_md_run (mddev
);
3832 case SET_BITMAP_FILE
:
3833 err
= set_bitmap_file(mddev
, (int)arg
);
3837 if (_IOC_TYPE(cmd
) == MD_MAJOR
)
3838 printk(KERN_WARNING
"md: %s(pid %d) used"
3839 " obsolete MD ioctl, upgrade your"
3840 " software to use new ictls.\n",
3841 current
->comm
, current
->pid
);
3848 mddev_unlock(mddev
);
3858 static int md_open(struct inode
*inode
, struct file
*file
)
3861 * Succeed if we can lock the mddev, which confirms that
3862 * it isn't being stopped right now.
3864 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3867 if ((err
= mddev_lock(mddev
)))
3872 mddev_unlock(mddev
);
3874 check_disk_change(inode
->i_bdev
);
3879 static int md_release(struct inode
*inode
, struct file
* file
)
3881 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3890 static int md_media_changed(struct gendisk
*disk
)
3892 mddev_t
*mddev
= disk
->private_data
;
3894 return mddev
->changed
;
3897 static int md_revalidate(struct gendisk
*disk
)
3899 mddev_t
*mddev
= disk
->private_data
;
3904 static struct block_device_operations md_fops
=
3906 .owner
= THIS_MODULE
,
3908 .release
= md_release
,
3910 .getgeo
= md_getgeo
,
3911 .media_changed
= md_media_changed
,
3912 .revalidate_disk
= md_revalidate
,
3915 static int md_thread(void * arg
)
3917 mdk_thread_t
*thread
= arg
;
3920 * md_thread is a 'system-thread', it's priority should be very
3921 * high. We avoid resource deadlocks individually in each
3922 * raid personality. (RAID5 does preallocation) We also use RR and
3923 * the very same RT priority as kswapd, thus we will never get
3924 * into a priority inversion deadlock.
3926 * we definitely have to have equal or higher priority than
3927 * bdflush, otherwise bdflush will deadlock if there are too
3928 * many dirty RAID5 blocks.
3931 allow_signal(SIGKILL
);
3932 while (!kthread_should_stop()) {
3934 /* We need to wait INTERRUPTIBLE so that
3935 * we don't add to the load-average.
3936 * That means we need to be sure no signals are
3939 if (signal_pending(current
))
3940 flush_signals(current
);
3942 wait_event_interruptible_timeout
3944 test_bit(THREAD_WAKEUP
, &thread
->flags
)
3945 || kthread_should_stop(),
3949 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
3951 thread
->run(thread
->mddev
);
3957 void md_wakeup_thread(mdk_thread_t
*thread
)
3960 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
3961 set_bit(THREAD_WAKEUP
, &thread
->flags
);
3962 wake_up(&thread
->wqueue
);
3966 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
3969 mdk_thread_t
*thread
;
3971 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
3975 init_waitqueue_head(&thread
->wqueue
);
3978 thread
->mddev
= mddev
;
3979 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
3980 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
3981 if (IS_ERR(thread
->tsk
)) {
3988 void md_unregister_thread(mdk_thread_t
*thread
)
3990 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
3992 kthread_stop(thread
->tsk
);
3996 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4003 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4006 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4008 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4009 __builtin_return_address(0),__builtin_return_address(1),
4010 __builtin_return_address(2),__builtin_return_address(3));
4012 if (!mddev
->pers
->error_handler
)
4014 mddev
->pers
->error_handler(mddev
,rdev
);
4015 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4016 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4017 md_wakeup_thread(mddev
->thread
);
4018 md_new_event(mddev
);
4021 /* seq_file implementation /proc/mdstat */
4023 static void status_unused(struct seq_file
*seq
)
4027 struct list_head
*tmp
;
4029 seq_printf(seq
, "unused devices: ");
4031 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4032 char b
[BDEVNAME_SIZE
];
4034 seq_printf(seq
, "%s ",
4035 bdevname(rdev
->bdev
,b
));
4038 seq_printf(seq
, "<none>");
4040 seq_printf(seq
, "\n");
4044 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4046 unsigned long max_blocks
, resync
, res
, dt
, db
, rt
;
4048 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4050 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4051 max_blocks
= mddev
->resync_max_sectors
>> 1;
4053 max_blocks
= mddev
->size
;
4056 * Should not happen.
4062 res
= (resync
/1024)*1000/(max_blocks
/1024 + 1);
4064 int i
, x
= res
/50, y
= 20-x
;
4065 seq_printf(seq
, "[");
4066 for (i
= 0; i
< x
; i
++)
4067 seq_printf(seq
, "=");
4068 seq_printf(seq
, ">");
4069 for (i
= 0; i
< y
; i
++)
4070 seq_printf(seq
, ".");
4071 seq_printf(seq
, "] ");
4073 seq_printf(seq
, " %s =%3lu.%lu%% (%lu/%lu)",
4074 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4075 "resync" : "recovery"),
4076 res
/10, res
% 10, resync
, max_blocks
);
4079 * We do not want to overflow, so the order of operands and
4080 * the * 100 / 100 trick are important. We do a +1 to be
4081 * safe against division by zero. We only estimate anyway.
4083 * dt: time from mark until now
4084 * db: blocks written from mark until now
4085 * rt: remaining time
4087 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4089 db
= resync
- (mddev
->resync_mark_cnt
/2);
4090 rt
= (dt
* ((max_blocks
-resync
) / (db
/100+1)))/100;
4092 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4094 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4097 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4099 struct list_head
*tmp
;
4109 spin_lock(&all_mddevs_lock
);
4110 list_for_each(tmp
,&all_mddevs
)
4112 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4114 spin_unlock(&all_mddevs_lock
);
4117 spin_unlock(&all_mddevs_lock
);
4119 return (void*)2;/* tail */
4123 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4125 struct list_head
*tmp
;
4126 mddev_t
*next_mddev
, *mddev
= v
;
4132 spin_lock(&all_mddevs_lock
);
4134 tmp
= all_mddevs
.next
;
4136 tmp
= mddev
->all_mddevs
.next
;
4137 if (tmp
!= &all_mddevs
)
4138 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4140 next_mddev
= (void*)2;
4143 spin_unlock(&all_mddevs_lock
);
4151 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4155 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4159 struct mdstat_info
{
4163 static int md_seq_show(struct seq_file
*seq
, void *v
)
4167 struct list_head
*tmp2
;
4169 struct mdstat_info
*mi
= seq
->private;
4170 struct bitmap
*bitmap
;
4172 if (v
== (void*)1) {
4173 struct mdk_personality
*pers
;
4174 seq_printf(seq
, "Personalities : ");
4175 spin_lock(&pers_lock
);
4176 list_for_each_entry(pers
, &pers_list
, list
)
4177 seq_printf(seq
, "[%s] ", pers
->name
);
4179 spin_unlock(&pers_lock
);
4180 seq_printf(seq
, "\n");
4181 mi
->event
= atomic_read(&md_event_count
);
4184 if (v
== (void*)2) {
4189 if (mddev_lock(mddev
)!=0)
4191 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4192 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4193 mddev
->pers
? "" : "in");
4196 seq_printf(seq
, " (read-only)");
4198 seq_printf(seq
, "(auto-read-only)");
4199 seq_printf(seq
, " %s", mddev
->pers
->name
);
4203 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4204 char b
[BDEVNAME_SIZE
];
4205 seq_printf(seq
, " %s[%d]",
4206 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4207 if (test_bit(WriteMostly
, &rdev
->flags
))
4208 seq_printf(seq
, "(W)");
4209 if (test_bit(Faulty
, &rdev
->flags
)) {
4210 seq_printf(seq
, "(F)");
4212 } else if (rdev
->raid_disk
< 0)
4213 seq_printf(seq
, "(S)"); /* spare */
4217 if (!list_empty(&mddev
->disks
)) {
4219 seq_printf(seq
, "\n %llu blocks",
4220 (unsigned long long)mddev
->array_size
);
4222 seq_printf(seq
, "\n %llu blocks",
4223 (unsigned long long)size
);
4225 if (mddev
->persistent
) {
4226 if (mddev
->major_version
!= 0 ||
4227 mddev
->minor_version
!= 90) {
4228 seq_printf(seq
," super %d.%d",
4229 mddev
->major_version
,
4230 mddev
->minor_version
);
4233 seq_printf(seq
, " super non-persistent");
4236 mddev
->pers
->status (seq
, mddev
);
4237 seq_printf(seq
, "\n ");
4238 if (mddev
->pers
->sync_request
) {
4239 if (mddev
->curr_resync
> 2) {
4240 status_resync (seq
, mddev
);
4241 seq_printf(seq
, "\n ");
4242 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4243 seq_printf(seq
, "\tresync=DELAYED\n ");
4244 else if (mddev
->recovery_cp
< MaxSector
)
4245 seq_printf(seq
, "\tresync=PENDING\n ");
4248 seq_printf(seq
, "\n ");
4250 if ((bitmap
= mddev
->bitmap
)) {
4251 unsigned long chunk_kb
;
4252 unsigned long flags
;
4253 spin_lock_irqsave(&bitmap
->lock
, flags
);
4254 chunk_kb
= bitmap
->chunksize
>> 10;
4255 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4257 bitmap
->pages
- bitmap
->missing_pages
,
4259 (bitmap
->pages
- bitmap
->missing_pages
)
4260 << (PAGE_SHIFT
- 10),
4261 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4262 chunk_kb
? "KB" : "B");
4264 seq_printf(seq
, ", file: ");
4265 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4266 bitmap
->file
->f_dentry
," \t\n");
4269 seq_printf(seq
, "\n");
4270 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4273 seq_printf(seq
, "\n");
4275 mddev_unlock(mddev
);
4280 static struct seq_operations md_seq_ops
= {
4281 .start
= md_seq_start
,
4282 .next
= md_seq_next
,
4283 .stop
= md_seq_stop
,
4284 .show
= md_seq_show
,
4287 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4290 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4294 error
= seq_open(file
, &md_seq_ops
);
4298 struct seq_file
*p
= file
->private_data
;
4300 mi
->event
= atomic_read(&md_event_count
);
4305 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4307 struct seq_file
*m
= file
->private_data
;
4308 struct mdstat_info
*mi
= m
->private;
4311 return seq_release(inode
, file
);
4314 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4316 struct seq_file
*m
= filp
->private_data
;
4317 struct mdstat_info
*mi
= m
->private;
4320 poll_wait(filp
, &md_event_waiters
, wait
);
4322 /* always allow read */
4323 mask
= POLLIN
| POLLRDNORM
;
4325 if (mi
->event
!= atomic_read(&md_event_count
))
4326 mask
|= POLLERR
| POLLPRI
;
4330 static struct file_operations md_seq_fops
= {
4331 .open
= md_seq_open
,
4333 .llseek
= seq_lseek
,
4334 .release
= md_seq_release
,
4335 .poll
= mdstat_poll
,
4338 int register_md_personality(struct mdk_personality
*p
)
4340 spin_lock(&pers_lock
);
4341 list_add_tail(&p
->list
, &pers_list
);
4342 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4343 spin_unlock(&pers_lock
);
4347 int unregister_md_personality(struct mdk_personality
*p
)
4349 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4350 spin_lock(&pers_lock
);
4351 list_del_init(&p
->list
);
4352 spin_unlock(&pers_lock
);
4356 static int is_mddev_idle(mddev_t
*mddev
)
4359 struct list_head
*tmp
;
4361 unsigned long curr_events
;
4364 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4365 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4366 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4367 disk_stat_read(disk
, sectors
[1]) -
4368 atomic_read(&disk
->sync_io
);
4369 /* The difference between curr_events and last_events
4370 * will be affected by any new non-sync IO (making
4371 * curr_events bigger) and any difference in the amount of
4372 * in-flight syncio (making current_events bigger or smaller)
4373 * The amount in-flight is currently limited to
4374 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4375 * which is at most 4096 sectors.
4376 * These numbers are fairly fragile and should be made
4377 * more robust, probably by enforcing the
4378 * 'window size' that md_do_sync sort-of uses.
4380 * Note: the following is an unsigned comparison.
4382 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4383 rdev
->last_events
= curr_events
;
4390 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4392 /* another "blocks" (512byte) blocks have been synced */
4393 atomic_sub(blocks
, &mddev
->recovery_active
);
4394 wake_up(&mddev
->recovery_wait
);
4396 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4397 md_wakeup_thread(mddev
->thread
);
4398 // stop recovery, signal do_sync ....
4403 /* md_write_start(mddev, bi)
4404 * If we need to update some array metadata (e.g. 'active' flag
4405 * in superblock) before writing, schedule a superblock update
4406 * and wait for it to complete.
4408 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4410 if (bio_data_dir(bi
) != WRITE
)
4413 BUG_ON(mddev
->ro
== 1);
4414 if (mddev
->ro
== 2) {
4415 /* need to switch to read/write */
4417 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4418 md_wakeup_thread(mddev
->thread
);
4420 atomic_inc(&mddev
->writes_pending
);
4421 if (mddev
->in_sync
) {
4422 spin_lock_irq(&mddev
->write_lock
);
4423 if (mddev
->in_sync
) {
4425 mddev
->sb_dirty
= 1;
4426 md_wakeup_thread(mddev
->thread
);
4428 spin_unlock_irq(&mddev
->write_lock
);
4430 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4433 void md_write_end(mddev_t
*mddev
)
4435 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4436 if (mddev
->safemode
== 2)
4437 md_wakeup_thread(mddev
->thread
);
4439 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4443 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4445 #define SYNC_MARKS 10
4446 #define SYNC_MARK_STEP (3*HZ)
4447 static void md_do_sync(mddev_t
*mddev
)
4450 unsigned int currspeed
= 0,
4452 sector_t max_sectors
,j
, io_sectors
;
4453 unsigned long mark
[SYNC_MARKS
];
4454 sector_t mark_cnt
[SYNC_MARKS
];
4456 struct list_head
*tmp
;
4457 sector_t last_check
;
4460 /* just incase thread restarts... */
4461 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4464 /* we overload curr_resync somewhat here.
4465 * 0 == not engaged in resync at all
4466 * 2 == checking that there is no conflict with another sync
4467 * 1 == like 2, but have yielded to allow conflicting resync to
4469 * other == active in resync - this many blocks
4471 * Before starting a resync we must have set curr_resync to
4472 * 2, and then checked that every "conflicting" array has curr_resync
4473 * less than ours. When we find one that is the same or higher
4474 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4475 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4476 * This will mean we have to start checking from the beginning again.
4481 mddev
->curr_resync
= 2;
4484 if (kthread_should_stop()) {
4485 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4488 ITERATE_MDDEV(mddev2
,tmp
) {
4489 if (mddev2
== mddev
)
4491 if (mddev2
->curr_resync
&&
4492 match_mddev_units(mddev
,mddev2
)) {
4494 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4495 /* arbitrarily yield */
4496 mddev
->curr_resync
= 1;
4497 wake_up(&resync_wait
);
4499 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4500 /* no need to wait here, we can wait the next
4501 * time 'round when curr_resync == 2
4504 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4505 if (!kthread_should_stop() &&
4506 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4507 printk(KERN_INFO
"md: delaying resync of %s"
4508 " until %s has finished resync (they"
4509 " share one or more physical units)\n",
4510 mdname(mddev
), mdname(mddev2
));
4513 finish_wait(&resync_wait
, &wq
);
4516 finish_wait(&resync_wait
, &wq
);
4519 } while (mddev
->curr_resync
< 2);
4521 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4522 /* resync follows the size requested by the personality,
4523 * which defaults to physical size, but can be virtual size
4525 max_sectors
= mddev
->resync_max_sectors
;
4526 mddev
->resync_mismatches
= 0;
4528 /* recovery follows the physical size of devices */
4529 max_sectors
= mddev
->size
<< 1;
4531 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4532 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4533 " %d KB/sec/disc.\n", speed_min(mddev
));
4534 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4535 "(but not more than %d KB/sec) for reconstruction.\n",
4538 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4539 /* we don't use the checkpoint if there's a bitmap */
4540 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && !mddev
->bitmap
4541 && ! test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4542 j
= mddev
->recovery_cp
;
4546 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4548 mark_cnt
[m
] = io_sectors
;
4551 mddev
->resync_mark
= mark
[last_mark
];
4552 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4555 * Tune reconstruction:
4557 window
= 32*(PAGE_SIZE
/512);
4558 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4559 window
/2,(unsigned long long) max_sectors
/2);
4561 atomic_set(&mddev
->recovery_active
, 0);
4562 init_waitqueue_head(&mddev
->recovery_wait
);
4567 "md: resuming recovery of %s from checkpoint.\n",
4569 mddev
->curr_resync
= j
;
4572 while (j
< max_sectors
) {
4576 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4577 currspeed
< speed_min(mddev
));
4579 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4583 if (!skipped
) { /* actual IO requested */
4584 io_sectors
+= sectors
;
4585 atomic_add(sectors
, &mddev
->recovery_active
);
4589 if (j
>1) mddev
->curr_resync
= j
;
4590 if (last_check
== 0)
4591 /* this is the earliers that rebuilt will be
4592 * visible in /proc/mdstat
4594 md_new_event(mddev
);
4596 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4599 last_check
= io_sectors
;
4601 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4602 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4606 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4608 int next
= (last_mark
+1) % SYNC_MARKS
;
4610 mddev
->resync_mark
= mark
[next
];
4611 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4612 mark
[next
] = jiffies
;
4613 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4618 if (kthread_should_stop()) {
4620 * got a signal, exit.
4623 "md: md_do_sync() got signal ... exiting\n");
4624 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4629 * this loop exits only if either when we are slower than
4630 * the 'hard' speed limit, or the system was IO-idle for
4632 * the system might be non-idle CPU-wise, but we only care
4633 * about not overloading the IO subsystem. (things like an
4634 * e2fsck being done on the RAID array should execute fast)
4636 mddev
->queue
->unplug_fn(mddev
->queue
);
4639 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4640 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4642 if (currspeed
> speed_min(mddev
)) {
4643 if ((currspeed
> speed_max(mddev
)) ||
4644 !is_mddev_idle(mddev
)) {
4650 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
4652 * this also signals 'finished resyncing' to md_stop
4655 mddev
->queue
->unplug_fn(mddev
->queue
);
4657 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
4659 /* tell personality that we are finished */
4660 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
4662 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4663 mddev
->curr_resync
> 2 &&
4664 mddev
->curr_resync
>= mddev
->recovery_cp
) {
4665 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4667 "md: checkpointing recovery of %s.\n",
4669 mddev
->recovery_cp
= mddev
->curr_resync
;
4671 mddev
->recovery_cp
= MaxSector
;
4675 mddev
->curr_resync
= 0;
4676 wake_up(&resync_wait
);
4677 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4678 md_wakeup_thread(mddev
->thread
);
4683 * This routine is regularly called by all per-raid-array threads to
4684 * deal with generic issues like resync and super-block update.
4685 * Raid personalities that don't have a thread (linear/raid0) do not
4686 * need this as they never do any recovery or update the superblock.
4688 * It does not do any resync itself, but rather "forks" off other threads
4689 * to do that as needed.
4690 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4691 * "->recovery" and create a thread at ->sync_thread.
4692 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4693 * and wakeups up this thread which will reap the thread and finish up.
4694 * This thread also removes any faulty devices (with nr_pending == 0).
4696 * The overall approach is:
4697 * 1/ if the superblock needs updating, update it.
4698 * 2/ If a recovery thread is running, don't do anything else.
4699 * 3/ If recovery has finished, clean up, possibly marking spares active.
4700 * 4/ If there are any faulty devices, remove them.
4701 * 5/ If array is degraded, try to add spares devices
4702 * 6/ If array has spares or is not in-sync, start a resync thread.
4704 void md_check_recovery(mddev_t
*mddev
)
4707 struct list_head
*rtmp
;
4711 bitmap_daemon_work(mddev
->bitmap
);
4716 if (signal_pending(current
)) {
4717 if (mddev
->pers
->sync_request
) {
4718 printk(KERN_INFO
"md: %s in immediate safe mode\n",
4720 mddev
->safemode
= 2;
4722 flush_signals(current
);
4727 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
4728 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
4729 (mddev
->safemode
== 1) ||
4730 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
4731 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
4735 if (mddev_trylock(mddev
)==0) {
4738 spin_lock_irq(&mddev
->write_lock
);
4739 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
4740 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
4742 mddev
->sb_dirty
= 1;
4744 if (mddev
->safemode
== 1)
4745 mddev
->safemode
= 0;
4746 spin_unlock_irq(&mddev
->write_lock
);
4748 if (mddev
->sb_dirty
)
4749 md_update_sb(mddev
);
4752 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4753 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
4754 /* resync/recovery still happening */
4755 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4758 if (mddev
->sync_thread
) {
4759 /* resync has finished, collect result */
4760 md_unregister_thread(mddev
->sync_thread
);
4761 mddev
->sync_thread
= NULL
;
4762 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4763 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4765 /* activate any spares */
4766 mddev
->pers
->spare_active(mddev
);
4768 md_update_sb(mddev
);
4770 /* if array is no-longer degraded, then any saved_raid_disk
4771 * information must be scrapped
4773 if (!mddev
->degraded
)
4774 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4775 rdev
->saved_raid_disk
= -1;
4777 mddev
->recovery
= 0;
4778 /* flag recovery needed just to double check */
4779 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4780 md_new_event(mddev
);
4783 /* Clear some bits that don't mean anything, but
4786 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4787 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4788 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4789 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4791 /* no recovery is running.
4792 * remove any failed drives, then
4793 * add spares if possible.
4794 * Spare are also removed and re-added, to allow
4795 * the personality to fail the re-add.
4797 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4798 if (rdev
->raid_disk
>= 0 &&
4799 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
4800 atomic_read(&rdev
->nr_pending
)==0) {
4801 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
4803 sprintf(nm
,"rd%d", rdev
->raid_disk
);
4804 sysfs_remove_link(&mddev
->kobj
, nm
);
4805 rdev
->raid_disk
= -1;
4809 if (mddev
->degraded
) {
4810 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4811 if (rdev
->raid_disk
< 0
4812 && !test_bit(Faulty
, &rdev
->flags
)) {
4813 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
4815 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4816 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
4818 md_new_event(mddev
);
4825 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4826 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4827 } else if (mddev
->recovery_cp
< MaxSector
) {
4828 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4829 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4830 /* nothing to be done ... */
4833 if (mddev
->pers
->sync_request
) {
4834 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4835 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
4836 /* We are adding a device or devices to an array
4837 * which has the bitmap stored on all devices.
4838 * So make sure all bitmap pages get written
4840 bitmap_write_all(mddev
->bitmap
);
4842 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4845 if (!mddev
->sync_thread
) {
4846 printk(KERN_ERR
"%s: could not start resync"
4849 /* leave the spares where they are, it shouldn't hurt */
4850 mddev
->recovery
= 0;
4852 md_wakeup_thread(mddev
->sync_thread
);
4853 md_new_event(mddev
);
4856 mddev_unlock(mddev
);
4860 static int md_notify_reboot(struct notifier_block
*this,
4861 unsigned long code
, void *x
)
4863 struct list_head
*tmp
;
4866 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
4868 printk(KERN_INFO
"md: stopping all md devices.\n");
4870 ITERATE_MDDEV(mddev
,tmp
)
4871 if (mddev_trylock(mddev
)==0)
4872 do_md_stop (mddev
, 1);
4874 * certain more exotic SCSI devices are known to be
4875 * volatile wrt too early system reboots. While the
4876 * right place to handle this issue is the given
4877 * driver, we do want to have a safe RAID driver ...
4884 static struct notifier_block md_notifier
= {
4885 .notifier_call
= md_notify_reboot
,
4887 .priority
= INT_MAX
, /* before any real devices */
4890 static void md_geninit(void)
4892 struct proc_dir_entry
*p
;
4894 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
4896 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
4898 p
->proc_fops
= &md_seq_fops
;
4901 static int __init
md_init(void)
4905 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4906 " MD_SB_DISKS=%d\n",
4907 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
4908 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
4909 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
4912 if (register_blkdev(MAJOR_NR
, "md"))
4914 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
4915 unregister_blkdev(MAJOR_NR
, "md");
4919 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
4920 md_probe
, NULL
, NULL
);
4921 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
4922 md_probe
, NULL
, NULL
);
4924 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4925 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
4926 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4929 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4930 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
4931 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4935 register_reboot_notifier(&md_notifier
);
4936 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
4946 * Searches all registered partitions for autorun RAID arrays
4949 static dev_t detected_devices
[128];
4952 void md_autodetect_dev(dev_t dev
)
4954 if (dev_cnt
>= 0 && dev_cnt
< 127)
4955 detected_devices
[dev_cnt
++] = dev
;
4959 static void autostart_arrays(int part
)
4964 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
4966 for (i
= 0; i
< dev_cnt
; i
++) {
4967 dev_t dev
= detected_devices
[i
];
4969 rdev
= md_import_device(dev
,0, 0);
4973 if (test_bit(Faulty
, &rdev
->flags
)) {
4977 list_add(&rdev
->same_set
, &pending_raid_disks
);
4981 autorun_devices(part
);
4986 static __exit
void md_exit(void)
4989 struct list_head
*tmp
;
4991 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
4992 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
4993 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4994 devfs_remove("md/%d", i
);
4995 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4996 devfs_remove("md/d%d", i
);
5000 unregister_blkdev(MAJOR_NR
,"md");
5001 unregister_blkdev(mdp_major
, "mdp");
5002 unregister_reboot_notifier(&md_notifier
);
5003 unregister_sysctl_table(raid_table_header
);
5004 remove_proc_entry("mdstat", NULL
);
5005 ITERATE_MDDEV(mddev
,tmp
) {
5006 struct gendisk
*disk
= mddev
->gendisk
;
5009 export_array(mddev
);
5012 mddev
->gendisk
= NULL
;
5017 module_init(md_init
)
5018 module_exit(md_exit
)
5020 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5022 return sprintf(buffer
, "%d", start_readonly
);
5024 static int set_ro(const char *val
, struct kernel_param
*kp
)
5027 int num
= simple_strtoul(val
, &e
, 10);
5028 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5029 start_readonly
= num
;
5035 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5036 module_param(start_dirty_degraded
, int, 0644);
5039 EXPORT_SYMBOL(register_md_personality
);
5040 EXPORT_SYMBOL(unregister_md_personality
);
5041 EXPORT_SYMBOL(md_error
);
5042 EXPORT_SYMBOL(md_done_sync
);
5043 EXPORT_SYMBOL(md_write_start
);
5044 EXPORT_SYMBOL(md_write_end
);
5045 EXPORT_SYMBOL(md_register_thread
);
5046 EXPORT_SYMBOL(md_unregister_thread
);
5047 EXPORT_SYMBOL(md_wakeup_thread
);
5048 EXPORT_SYMBOL(md_print_devices
);
5049 EXPORT_SYMBOL(md_check_recovery
);
5050 MODULE_LICENSE("GPL");
5052 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);