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/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/devfs_fs_kernel.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/suspend.h>
45 #include <linux/init.h>
47 #include <linux/file.h>
50 #include <linux/kmod.h>
53 #include <asm/unaligned.h>
55 #define MAJOR_NR MD_MAJOR
58 /* 63 partitions with the alternate major number (mdp) */
59 #define MdpMinorShift 6
62 #define dprintk(x...) ((void)(DEBUG && printk(x)))
66 static void autostart_arrays (int part
);
69 static mdk_personality_t
*pers
[MAX_PERSONALITY
];
70 static DEFINE_SPINLOCK(pers_lock
);
73 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
74 * is 1000 KB/sec, so the extra system load does not show up that much.
75 * Increase it if you want to have more _guaranteed_ speed. Note that
76 * the RAID driver will use the maximum available bandwith if the IO
77 * subsystem is idle. There is also an 'absolute maximum' reconstruction
78 * speed limit - in case reconstruction slows down your system despite
81 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 static int sysctl_speed_limit_min
= 1000;
85 static int sysctl_speed_limit_max
= 200000;
87 static struct ctl_table_header
*raid_table_header
;
89 static ctl_table raid_table
[] = {
91 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
92 .procname
= "speed_limit_min",
93 .data
= &sysctl_speed_limit_min
,
94 .maxlen
= sizeof(int),
96 .proc_handler
= &proc_dointvec
,
99 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
100 .procname
= "speed_limit_max",
101 .data
= &sysctl_speed_limit_max
,
102 .maxlen
= sizeof(int),
104 .proc_handler
= &proc_dointvec
,
109 static ctl_table raid_dir_table
[] = {
111 .ctl_name
= DEV_RAID
,
120 static ctl_table raid_root_table
[] = {
126 .child
= raid_dir_table
,
131 static struct block_device_operations md_fops
;
134 * Enables to iterate over all existing md arrays
135 * all_mddevs_lock protects this list.
137 static LIST_HEAD(all_mddevs
);
138 static DEFINE_SPINLOCK(all_mddevs_lock
);
142 * iterates through all used mddevs in the system.
143 * We take care to grab the all_mddevs_lock whenever navigating
144 * the list, and to always hold a refcount when unlocked.
145 * Any code which breaks out of this loop while own
146 * a reference to the current mddev and must mddev_put it.
148 #define ITERATE_MDDEV(mddev,tmp) \
150 for (({ spin_lock(&all_mddevs_lock); \
151 tmp = all_mddevs.next; \
153 ({ if (tmp != &all_mddevs) \
154 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
155 spin_unlock(&all_mddevs_lock); \
156 if (mddev) mddev_put(mddev); \
157 mddev = list_entry(tmp, mddev_t, all_mddevs); \
158 tmp != &all_mddevs;}); \
159 ({ spin_lock(&all_mddevs_lock); \
164 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
166 bio_io_error(bio
, bio
->bi_size
);
170 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
172 atomic_inc(&mddev
->active
);
176 static void mddev_put(mddev_t
*mddev
)
178 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
180 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
181 list_del(&mddev
->all_mddevs
);
182 blk_put_queue(mddev
->queue
);
185 spin_unlock(&all_mddevs_lock
);
188 static mddev_t
* mddev_find(dev_t unit
)
190 mddev_t
*mddev
, *new = NULL
;
193 spin_lock(&all_mddevs_lock
);
194 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
195 if (mddev
->unit
== unit
) {
197 spin_unlock(&all_mddevs_lock
);
203 list_add(&new->all_mddevs
, &all_mddevs
);
204 spin_unlock(&all_mddevs_lock
);
207 spin_unlock(&all_mddevs_lock
);
209 new = (mddev_t
*) kmalloc(sizeof(*new), GFP_KERNEL
);
213 memset(new, 0, sizeof(*new));
216 if (MAJOR(unit
) == MD_MAJOR
)
217 new->md_minor
= MINOR(unit
);
219 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
221 init_MUTEX(&new->reconfig_sem
);
222 INIT_LIST_HEAD(&new->disks
);
223 INIT_LIST_HEAD(&new->all_mddevs
);
224 init_timer(&new->safemode_timer
);
225 atomic_set(&new->active
, 1);
226 spin_lock_init(&new->write_lock
);
227 init_waitqueue_head(&new->sb_wait
);
229 new->queue
= blk_alloc_queue(GFP_KERNEL
);
235 blk_queue_make_request(new->queue
, md_fail_request
);
240 static inline int mddev_lock(mddev_t
* mddev
)
242 return down_interruptible(&mddev
->reconfig_sem
);
245 static inline void mddev_lock_uninterruptible(mddev_t
* mddev
)
247 down(&mddev
->reconfig_sem
);
250 static inline int mddev_trylock(mddev_t
* mddev
)
252 return down_trylock(&mddev
->reconfig_sem
);
255 static inline void mddev_unlock(mddev_t
* mddev
)
257 up(&mddev
->reconfig_sem
);
260 md_wakeup_thread(mddev
->thread
);
263 mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
266 struct list_head
*tmp
;
268 ITERATE_RDEV(mddev
,rdev
,tmp
) {
269 if (rdev
->desc_nr
== nr
)
275 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
277 struct list_head
*tmp
;
280 ITERATE_RDEV(mddev
,rdev
,tmp
) {
281 if (rdev
->bdev
->bd_dev
== dev
)
287 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
289 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
290 return MD_NEW_SIZE_BLOCKS(size
);
293 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
297 size
= rdev
->sb_offset
;
300 size
&= ~((sector_t
)chunk_size
/1024 - 1);
304 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
309 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
310 if (!rdev
->sb_page
) {
311 printk(KERN_ALERT
"md: out of memory.\n");
318 static void free_disk_sb(mdk_rdev_t
* rdev
)
321 page_cache_release(rdev
->sb_page
);
323 rdev
->sb_page
= NULL
;
330 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
332 mdk_rdev_t
*rdev
= bio
->bi_private
;
336 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
337 md_error(rdev
->mddev
, rdev
);
339 if (atomic_dec_and_test(&rdev
->mddev
->pending_writes
))
340 wake_up(&rdev
->mddev
->sb_wait
);
345 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
346 sector_t sector
, int size
, struct page
*page
)
348 /* write first size bytes of page to sector of rdev
349 * Increment mddev->pending_writes before returning
350 * and decrement it on completion, waking up sb_wait
351 * if zero is reached.
352 * If an error occurred, call md_error
354 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
356 bio
->bi_bdev
= rdev
->bdev
;
357 bio
->bi_sector
= sector
;
358 bio_add_page(bio
, page
, size
, 0);
359 bio
->bi_private
= rdev
;
360 bio
->bi_end_io
= super_written
;
361 atomic_inc(&mddev
->pending_writes
);
362 submit_bio((1<<BIO_RW
)|(1<<BIO_RW_SYNC
), bio
);
365 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
370 complete((struct completion
*)bio
->bi_private
);
374 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
375 struct page
*page
, int rw
)
377 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
378 struct completion event
;
381 rw
|= (1 << BIO_RW_SYNC
);
384 bio
->bi_sector
= sector
;
385 bio_add_page(bio
, page
, size
, 0);
386 init_completion(&event
);
387 bio
->bi_private
= &event
;
388 bio
->bi_end_io
= bi_complete
;
390 wait_for_completion(&event
);
392 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
397 static int read_disk_sb(mdk_rdev_t
* rdev
)
399 char b
[BDEVNAME_SIZE
];
400 if (!rdev
->sb_page
) {
408 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, MD_SB_BYTES
, rdev
->sb_page
, READ
))
414 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
415 bdevname(rdev
->bdev
,b
));
419 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
421 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
422 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
423 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
424 (sb1
->set_uuid3
== sb2
->set_uuid3
))
432 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
435 mdp_super_t
*tmp1
, *tmp2
;
437 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
438 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
440 if (!tmp1
|| !tmp2
) {
442 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
450 * nr_disks is not constant
455 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
466 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
468 unsigned int disk_csum
, csum
;
470 disk_csum
= sb
->sb_csum
;
472 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
473 sb
->sb_csum
= disk_csum
;
479 * Handle superblock details.
480 * We want to be able to handle multiple superblock formats
481 * so we have a common interface to them all, and an array of
482 * different handlers.
483 * We rely on user-space to write the initial superblock, and support
484 * reading and updating of superblocks.
485 * Interface methods are:
486 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
487 * loads and validates a superblock on dev.
488 * if refdev != NULL, compare superblocks on both devices
490 * 0 - dev has a superblock that is compatible with refdev
491 * 1 - dev has a superblock that is compatible and newer than refdev
492 * so dev should be used as the refdev in future
493 * -EINVAL superblock incompatible or invalid
494 * -othererror e.g. -EIO
496 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
497 * Verify that dev is acceptable into mddev.
498 * The first time, mddev->raid_disks will be 0, and data from
499 * dev should be merged in. Subsequent calls check that dev
500 * is new enough. Return 0 or -EINVAL
502 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
503 * Update the superblock for rdev with data in mddev
504 * This does not write to disc.
510 struct module
*owner
;
511 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
512 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
513 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
517 * load_super for 0.90.0
519 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
521 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
527 * Calculate the position of the superblock,
528 * it's at the end of the disk.
530 * It also happens to be a multiple of 4Kb.
532 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
533 rdev
->sb_offset
= sb_offset
;
535 ret
= read_disk_sb(rdev
);
540 bdevname(rdev
->bdev
, b
);
541 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
543 if (sb
->md_magic
!= MD_SB_MAGIC
) {
544 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
549 if (sb
->major_version
!= 0 ||
550 sb
->minor_version
!= 90) {
551 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
552 sb
->major_version
, sb
->minor_version
,
557 if (sb
->raid_disks
<= 0)
560 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
561 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
566 rdev
->preferred_minor
= sb
->md_minor
;
567 rdev
->data_offset
= 0;
569 if (sb
->level
== LEVEL_MULTIPATH
)
572 rdev
->desc_nr
= sb
->this_disk
.number
;
578 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
579 if (!uuid_equal(refsb
, sb
)) {
580 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
581 b
, bdevname(refdev
->bdev
,b2
));
584 if (!sb_equal(refsb
, sb
)) {
585 printk(KERN_WARNING
"md: %s has same UUID"
586 " but different superblock to %s\n",
587 b
, bdevname(refdev
->bdev
, b2
));
591 ev2
= md_event(refsb
);
597 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
604 * validate_super for 0.90.0
606 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
609 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
611 rdev
->raid_disk
= -1;
613 if (mddev
->raid_disks
== 0) {
614 mddev
->major_version
= 0;
615 mddev
->minor_version
= sb
->minor_version
;
616 mddev
->patch_version
= sb
->patch_version
;
617 mddev
->persistent
= ! sb
->not_persistent
;
618 mddev
->chunk_size
= sb
->chunk_size
;
619 mddev
->ctime
= sb
->ctime
;
620 mddev
->utime
= sb
->utime
;
621 mddev
->level
= sb
->level
;
622 mddev
->layout
= sb
->layout
;
623 mddev
->raid_disks
= sb
->raid_disks
;
624 mddev
->size
= sb
->size
;
625 mddev
->events
= md_event(sb
);
627 if (sb
->state
& (1<<MD_SB_CLEAN
))
628 mddev
->recovery_cp
= MaxSector
;
630 if (sb
->events_hi
== sb
->cp_events_hi
&&
631 sb
->events_lo
== sb
->cp_events_lo
) {
632 mddev
->recovery_cp
= sb
->recovery_cp
;
634 mddev
->recovery_cp
= 0;
637 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
638 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
639 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
640 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
642 mddev
->max_disks
= MD_SB_DISKS
;
644 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
645 mddev
->bitmap_file
== NULL
) {
646 if (mddev
->level
!= 1) {
647 /* FIXME use a better test */
648 printk(KERN_WARNING
"md: bitmaps only support for raid1\n");
651 mddev
->bitmap_offset
= (MD_SB_BYTES
>> 9);
654 } else if (mddev
->pers
== NULL
) {
655 /* Insist on good event counter while assembling */
656 __u64 ev1
= md_event(sb
);
658 if (ev1
< mddev
->events
)
660 } else if (mddev
->bitmap
) {
661 /* if adding to array with a bitmap, then we can accept an
662 * older device ... but not too old.
664 __u64 ev1
= md_event(sb
);
665 if (ev1
< mddev
->bitmap
->events_cleared
)
667 } else /* just a hot-add of a new device, leave raid_disk at -1 */
670 if (mddev
->level
!= LEVEL_MULTIPATH
) {
672 desc
= sb
->disks
+ rdev
->desc_nr
;
674 if (desc
->state
& (1<<MD_DISK_FAULTY
))
676 else if (desc
->state
& (1<<MD_DISK_SYNC
) &&
677 desc
->raid_disk
< mddev
->raid_disks
) {
679 rdev
->raid_disk
= desc
->raid_disk
;
681 } else /* MULTIPATH are always insync */
687 * sync_super for 0.90.0
689 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
692 struct list_head
*tmp
;
694 int next_spare
= mddev
->raid_disks
;
696 /* make rdev->sb match mddev data..
699 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
700 * 3/ any empty disks < next_spare become removed
702 * disks[0] gets initialised to REMOVED because
703 * we cannot be sure from other fields if it has
704 * been initialised or not.
707 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
709 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
711 memset(sb
, 0, sizeof(*sb
));
713 sb
->md_magic
= MD_SB_MAGIC
;
714 sb
->major_version
= mddev
->major_version
;
715 sb
->minor_version
= mddev
->minor_version
;
716 sb
->patch_version
= mddev
->patch_version
;
717 sb
->gvalid_words
= 0; /* ignored */
718 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
719 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
720 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
721 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
723 sb
->ctime
= mddev
->ctime
;
724 sb
->level
= mddev
->level
;
725 sb
->size
= mddev
->size
;
726 sb
->raid_disks
= mddev
->raid_disks
;
727 sb
->md_minor
= mddev
->md_minor
;
728 sb
->not_persistent
= !mddev
->persistent
;
729 sb
->utime
= mddev
->utime
;
731 sb
->events_hi
= (mddev
->events
>>32);
732 sb
->events_lo
= (u32
)mddev
->events
;
736 sb
->recovery_cp
= mddev
->recovery_cp
;
737 sb
->cp_events_hi
= (mddev
->events
>>32);
738 sb
->cp_events_lo
= (u32
)mddev
->events
;
739 if (mddev
->recovery_cp
== MaxSector
)
740 sb
->state
= (1<< MD_SB_CLEAN
);
744 sb
->layout
= mddev
->layout
;
745 sb
->chunk_size
= mddev
->chunk_size
;
747 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
748 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
750 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
751 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
753 if (rdev2
->raid_disk
>= 0 && rdev2
->in_sync
&& !rdev2
->faulty
)
754 rdev2
->desc_nr
= rdev2
->raid_disk
;
756 rdev2
->desc_nr
= next_spare
++;
757 d
= &sb
->disks
[rdev2
->desc_nr
];
759 d
->number
= rdev2
->desc_nr
;
760 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
761 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
762 if (rdev2
->raid_disk
>= 0 && rdev
->in_sync
&& !rdev2
->faulty
)
763 d
->raid_disk
= rdev2
->raid_disk
;
765 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
767 d
->state
= (1<<MD_DISK_FAULTY
);
769 } else if (rdev2
->in_sync
) {
770 d
->state
= (1<<MD_DISK_ACTIVE
);
771 d
->state
|= (1<<MD_DISK_SYNC
);
781 /* now set the "removed" and "faulty" bits on any missing devices */
782 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
783 mdp_disk_t
*d
= &sb
->disks
[i
];
784 if (d
->state
== 0 && d
->number
== 0) {
787 d
->state
= (1<<MD_DISK_REMOVED
);
788 d
->state
|= (1<<MD_DISK_FAULTY
);
792 sb
->nr_disks
= nr_disks
;
793 sb
->active_disks
= active
;
794 sb
->working_disks
= working
;
795 sb
->failed_disks
= failed
;
796 sb
->spare_disks
= spare
;
798 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
799 sb
->sb_csum
= calc_sb_csum(sb
);
803 * version 1 superblock
806 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
808 unsigned int disk_csum
, csum
;
809 unsigned long long newcsum
;
810 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
811 unsigned int *isuper
= (unsigned int*)sb
;
814 disk_csum
= sb
->sb_csum
;
817 for (i
=0; size
>=4; size
-= 4 )
818 newcsum
+= le32_to_cpu(*isuper
++);
821 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
823 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
824 sb
->sb_csum
= disk_csum
;
825 return cpu_to_le32(csum
);
828 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
830 struct mdp_superblock_1
*sb
;
833 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
836 * Calculate the position of the superblock.
837 * It is always aligned to a 4K boundary and
838 * depeding on minor_version, it can be:
839 * 0: At least 8K, but less than 12K, from end of device
840 * 1: At start of device
841 * 2: 4K from start of device.
843 switch(minor_version
) {
845 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
847 sb_offset
&= ~(sector_t
)(4*2-1);
848 /* convert from sectors to K */
860 rdev
->sb_offset
= sb_offset
;
862 ret
= read_disk_sb(rdev
);
866 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
868 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
869 sb
->major_version
!= cpu_to_le32(1) ||
870 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
871 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
872 sb
->feature_map
!= 0)
875 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
876 printk("md: invalid superblock checksum on %s\n",
877 bdevname(rdev
->bdev
,b
));
880 if (le64_to_cpu(sb
->data_size
) < 10) {
881 printk("md: data_size too small on %s\n",
882 bdevname(rdev
->bdev
,b
));
885 rdev
->preferred_minor
= 0xffff;
886 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
892 struct mdp_superblock_1
*refsb
=
893 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
895 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
896 sb
->level
!= refsb
->level
||
897 sb
->layout
!= refsb
->layout
||
898 sb
->chunksize
!= refsb
->chunksize
) {
899 printk(KERN_WARNING
"md: %s has strangely different"
900 " superblock to %s\n",
901 bdevname(rdev
->bdev
,b
),
902 bdevname(refdev
->bdev
,b2
));
905 ev1
= le64_to_cpu(sb
->events
);
906 ev2
= le64_to_cpu(refsb
->events
);
912 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
914 rdev
->size
= rdev
->sb_offset
;
915 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
917 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
918 if (le32_to_cpu(sb
->chunksize
))
919 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
923 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
925 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
927 rdev
->raid_disk
= -1;
929 if (mddev
->raid_disks
== 0) {
930 mddev
->major_version
= 1;
931 mddev
->patch_version
= 0;
932 mddev
->persistent
= 1;
933 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
934 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
935 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
936 mddev
->level
= le32_to_cpu(sb
->level
);
937 mddev
->layout
= le32_to_cpu(sb
->layout
);
938 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
939 mddev
->size
= le64_to_cpu(sb
->size
)/2;
940 mddev
->events
= le64_to_cpu(sb
->events
);
942 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
943 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
945 mddev
->max_disks
= (4096-256)/2;
947 if ((le32_to_cpu(sb
->feature_map
) & 1) &&
948 mddev
->bitmap_file
== NULL
) {
949 if (mddev
->level
!= 1) {
950 printk(KERN_WARNING
"md: bitmaps only supported for raid1\n");
953 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
955 } else if (mddev
->pers
== NULL
) {
956 /* Insist of good event counter while assembling */
957 __u64 ev1
= le64_to_cpu(sb
->events
);
959 if (ev1
< mddev
->events
)
961 } else if (mddev
->bitmap
) {
962 /* If adding to array with a bitmap, then we can accept an
963 * older device, but not too old.
965 __u64 ev1
= le64_to_cpu(sb
->events
);
966 if (ev1
< mddev
->bitmap
->events_cleared
)
968 } else /* just a hot-add of a new device, leave raid_disk at -1 */
971 if (mddev
->level
!= LEVEL_MULTIPATH
) {
973 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
974 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
976 case 0xffff: /* spare */
979 case 0xfffe: /* faulty */
985 rdev
->raid_disk
= role
;
988 } else /* MULTIPATH are always insync */
994 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
996 struct mdp_superblock_1
*sb
;
997 struct list_head
*tmp
;
1000 /* make rdev->sb match mddev and rdev data. */
1002 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1004 sb
->feature_map
= 0;
1006 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1007 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1008 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1010 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1011 sb
->events
= cpu_to_le64(mddev
->events
);
1013 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1015 sb
->resync_offset
= cpu_to_le64(0);
1017 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1018 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1019 sb
->feature_map
= cpu_to_le32(1);
1023 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1024 if (rdev2
->desc_nr
+1 > max_dev
)
1025 max_dev
= rdev2
->desc_nr
+1;
1027 sb
->max_dev
= cpu_to_le32(max_dev
);
1028 for (i
=0; i
<max_dev
;i
++)
1029 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1031 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1034 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1035 else if (rdev2
->in_sync
)
1036 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1038 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1041 sb
->recovery_offset
= cpu_to_le64(0); /* not supported yet */
1042 sb
->sb_csum
= calc_sb_1_csum(sb
);
1046 static struct super_type super_types
[] = {
1049 .owner
= THIS_MODULE
,
1050 .load_super
= super_90_load
,
1051 .validate_super
= super_90_validate
,
1052 .sync_super
= super_90_sync
,
1056 .owner
= THIS_MODULE
,
1057 .load_super
= super_1_load
,
1058 .validate_super
= super_1_validate
,
1059 .sync_super
= super_1_sync
,
1063 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1065 struct list_head
*tmp
;
1068 ITERATE_RDEV(mddev
,rdev
,tmp
)
1069 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1075 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1077 struct list_head
*tmp
;
1080 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1081 if (match_dev_unit(mddev2
, rdev
))
1087 static LIST_HEAD(pending_raid_disks
);
1089 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1091 mdk_rdev_t
*same_pdev
;
1092 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1098 same_pdev
= match_dev_unit(mddev
, rdev
);
1101 "%s: WARNING: %s appears to be on the same physical"
1102 " disk as %s. True\n protection against single-disk"
1103 " failure might be compromised.\n",
1104 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1105 bdevname(same_pdev
->bdev
,b2
));
1107 /* Verify rdev->desc_nr is unique.
1108 * If it is -1, assign a free number, else
1109 * check number is not in use
1111 if (rdev
->desc_nr
< 0) {
1113 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1114 while (find_rdev_nr(mddev
, choice
))
1116 rdev
->desc_nr
= choice
;
1118 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1122 list_add(&rdev
->same_set
, &mddev
->disks
);
1123 rdev
->mddev
= mddev
;
1124 printk(KERN_INFO
"md: bind<%s>\n", bdevname(rdev
->bdev
,b
));
1128 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1130 char b
[BDEVNAME_SIZE
];
1135 list_del_init(&rdev
->same_set
);
1136 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1141 * prevent the device from being mounted, repartitioned or
1142 * otherwise reused by a RAID array (or any other kernel
1143 * subsystem), by bd_claiming the device.
1145 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1148 struct block_device
*bdev
;
1149 char b
[BDEVNAME_SIZE
];
1151 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1153 printk(KERN_ERR
"md: could not open %s.\n",
1154 __bdevname(dev
, b
));
1155 return PTR_ERR(bdev
);
1157 err
= bd_claim(bdev
, rdev
);
1159 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1168 static void unlock_rdev(mdk_rdev_t
*rdev
)
1170 struct block_device
*bdev
= rdev
->bdev
;
1178 void md_autodetect_dev(dev_t dev
);
1180 static void export_rdev(mdk_rdev_t
* rdev
)
1182 char b
[BDEVNAME_SIZE
];
1183 printk(KERN_INFO
"md: export_rdev(%s)\n",
1184 bdevname(rdev
->bdev
,b
));
1188 list_del_init(&rdev
->same_set
);
1190 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1196 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1198 unbind_rdev_from_array(rdev
);
1202 static void export_array(mddev_t
*mddev
)
1204 struct list_head
*tmp
;
1207 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1212 kick_rdev_from_array(rdev
);
1214 if (!list_empty(&mddev
->disks
))
1216 mddev
->raid_disks
= 0;
1217 mddev
->major_version
= 0;
1220 static void print_desc(mdp_disk_t
*desc
)
1222 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1223 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1226 static void print_sb(mdp_super_t
*sb
)
1231 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1232 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1233 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1235 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1236 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1237 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1238 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1239 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1240 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1241 sb
->failed_disks
, sb
->spare_disks
,
1242 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1245 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1248 desc
= sb
->disks
+ i
;
1249 if (desc
->number
|| desc
->major
|| desc
->minor
||
1250 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1251 printk(" D %2d: ", i
);
1255 printk(KERN_INFO
"md: THIS: ");
1256 print_desc(&sb
->this_disk
);
1260 static void print_rdev(mdk_rdev_t
*rdev
)
1262 char b
[BDEVNAME_SIZE
];
1263 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1264 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1265 rdev
->faulty
, rdev
->in_sync
, rdev
->desc_nr
);
1266 if (rdev
->sb_loaded
) {
1267 printk(KERN_INFO
"md: rdev superblock:\n");
1268 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1270 printk(KERN_INFO
"md: no rdev superblock!\n");
1273 void md_print_devices(void)
1275 struct list_head
*tmp
, *tmp2
;
1278 char b
[BDEVNAME_SIZE
];
1281 printk("md: **********************************\n");
1282 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1283 printk("md: **********************************\n");
1284 ITERATE_MDDEV(mddev
,tmp
) {
1287 bitmap_print_sb(mddev
->bitmap
);
1289 printk("%s: ", mdname(mddev
));
1290 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1291 printk("<%s>", bdevname(rdev
->bdev
,b
));
1294 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1297 printk("md: **********************************\n");
1302 static void sync_sbs(mddev_t
* mddev
)
1305 struct list_head
*tmp
;
1307 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1308 super_types
[mddev
->major_version
].
1309 sync_super(mddev
, rdev
);
1310 rdev
->sb_loaded
= 1;
1314 static void md_update_sb(mddev_t
* mddev
)
1317 struct list_head
*tmp
;
1322 spin_lock(&mddev
->write_lock
);
1323 sync_req
= mddev
->in_sync
;
1324 mddev
->utime
= get_seconds();
1327 if (!mddev
->events
) {
1329 * oops, this 64-bit counter should never wrap.
1330 * Either we are in around ~1 trillion A.C., assuming
1331 * 1 reboot per second, or we have a bug:
1336 mddev
->sb_dirty
= 2;
1340 * do not write anything to disk if using
1341 * nonpersistent superblocks
1343 if (!mddev
->persistent
) {
1344 mddev
->sb_dirty
= 0;
1345 spin_unlock(&mddev
->write_lock
);
1346 wake_up(&mddev
->sb_wait
);
1349 spin_unlock(&mddev
->write_lock
);
1352 "md: updating %s RAID superblock on device (in sync %d)\n",
1353 mdname(mddev
),mddev
->in_sync
);
1355 err
= bitmap_update_sb(mddev
->bitmap
);
1356 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1357 char b
[BDEVNAME_SIZE
];
1358 dprintk(KERN_INFO
"md: ");
1360 dprintk("(skipping faulty ");
1362 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1363 if (!rdev
->faulty
) {
1364 md_super_write(mddev
,rdev
,
1365 rdev
->sb_offset
<<1, MD_SB_BYTES
,
1367 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1368 bdevname(rdev
->bdev
,b
),
1369 (unsigned long long)rdev
->sb_offset
);
1373 if (mddev
->level
== LEVEL_MULTIPATH
)
1374 /* only need to write one superblock... */
1377 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
1378 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1380 spin_lock(&mddev
->write_lock
);
1381 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1382 /* have to write it out again */
1383 spin_unlock(&mddev
->write_lock
);
1386 mddev
->sb_dirty
= 0;
1387 spin_unlock(&mddev
->write_lock
);
1388 wake_up(&mddev
->sb_wait
);
1393 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1395 * mark the device faulty if:
1397 * - the device is nonexistent (zero size)
1398 * - the device has no valid superblock
1400 * a faulty rdev _never_ has rdev->sb set.
1402 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1404 char b
[BDEVNAME_SIZE
];
1409 rdev
= (mdk_rdev_t
*) kmalloc(sizeof(*rdev
), GFP_KERNEL
);
1411 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1412 return ERR_PTR(-ENOMEM
);
1414 memset(rdev
, 0, sizeof(*rdev
));
1416 if ((err
= alloc_disk_sb(rdev
)))
1419 err
= lock_rdev(rdev
, newdev
);
1426 rdev
->data_offset
= 0;
1427 atomic_set(&rdev
->nr_pending
, 0);
1429 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1432 "md: %s has zero or unknown size, marking faulty!\n",
1433 bdevname(rdev
->bdev
,b
));
1438 if (super_format
>= 0) {
1439 err
= super_types
[super_format
].
1440 load_super(rdev
, NULL
, super_minor
);
1441 if (err
== -EINVAL
) {
1443 "md: %s has invalid sb, not importing!\n",
1444 bdevname(rdev
->bdev
,b
));
1449 "md: could not read %s's sb, not importing!\n",
1450 bdevname(rdev
->bdev
,b
));
1454 INIT_LIST_HEAD(&rdev
->same_set
);
1459 if (rdev
->sb_page
) {
1465 return ERR_PTR(err
);
1469 * Check a full RAID array for plausibility
1473 static void analyze_sbs(mddev_t
* mddev
)
1476 struct list_head
*tmp
;
1477 mdk_rdev_t
*rdev
, *freshest
;
1478 char b
[BDEVNAME_SIZE
];
1481 ITERATE_RDEV(mddev
,rdev
,tmp
)
1482 switch (super_types
[mddev
->major_version
].
1483 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1491 "md: fatal superblock inconsistency in %s"
1492 " -- removing from array\n",
1493 bdevname(rdev
->bdev
,b
));
1494 kick_rdev_from_array(rdev
);
1498 super_types
[mddev
->major_version
].
1499 validate_super(mddev
, freshest
);
1502 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1503 if (rdev
!= freshest
)
1504 if (super_types
[mddev
->major_version
].
1505 validate_super(mddev
, rdev
)) {
1506 printk(KERN_WARNING
"md: kicking non-fresh %s"
1508 bdevname(rdev
->bdev
,b
));
1509 kick_rdev_from_array(rdev
);
1512 if (mddev
->level
== LEVEL_MULTIPATH
) {
1513 rdev
->desc_nr
= i
++;
1514 rdev
->raid_disk
= rdev
->desc_nr
;
1521 if (mddev
->recovery_cp
!= MaxSector
&&
1523 printk(KERN_ERR
"md: %s: raid array is not clean"
1524 " -- starting background reconstruction\n",
1531 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
1533 static DECLARE_MUTEX(disks_sem
);
1534 mddev_t
*mddev
= mddev_find(dev
);
1535 struct gendisk
*disk
;
1536 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
1537 int shift
= partitioned
? MdpMinorShift
: 0;
1538 int unit
= MINOR(dev
) >> shift
;
1544 if (mddev
->gendisk
) {
1549 disk
= alloc_disk(1 << shift
);
1555 disk
->major
= MAJOR(dev
);
1556 disk
->first_minor
= unit
<< shift
;
1558 sprintf(disk
->disk_name
, "md_d%d", unit
);
1559 sprintf(disk
->devfs_name
, "md/d%d", unit
);
1561 sprintf(disk
->disk_name
, "md%d", unit
);
1562 sprintf(disk
->devfs_name
, "md/%d", unit
);
1564 disk
->fops
= &md_fops
;
1565 disk
->private_data
= mddev
;
1566 disk
->queue
= mddev
->queue
;
1568 mddev
->gendisk
= disk
;
1573 void md_wakeup_thread(mdk_thread_t
*thread
);
1575 static void md_safemode_timeout(unsigned long data
)
1577 mddev_t
*mddev
= (mddev_t
*) data
;
1579 mddev
->safemode
= 1;
1580 md_wakeup_thread(mddev
->thread
);
1584 static int do_md_run(mddev_t
* mddev
)
1588 struct list_head
*tmp
;
1590 struct gendisk
*disk
;
1591 char b
[BDEVNAME_SIZE
];
1593 if (list_empty(&mddev
->disks
))
1594 /* cannot run an array with no devices.. */
1601 * Analyze all RAID superblock(s)
1603 if (!mddev
->raid_disks
)
1606 chunk_size
= mddev
->chunk_size
;
1607 pnum
= level_to_pers(mddev
->level
);
1609 if ((pnum
!= MULTIPATH
) && (pnum
!= RAID1
)) {
1612 * 'default chunksize' in the old md code used to
1613 * be PAGE_SIZE, baaad.
1614 * we abort here to be on the safe side. We don't
1615 * want to continue the bad practice.
1618 "no chunksize specified, see 'man raidtab'\n");
1621 if (chunk_size
> MAX_CHUNK_SIZE
) {
1622 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
1623 chunk_size
, MAX_CHUNK_SIZE
);
1627 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1629 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
1630 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
1633 if (chunk_size
< PAGE_SIZE
) {
1634 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
1635 chunk_size
, PAGE_SIZE
);
1639 /* devices must have minimum size of one chunk */
1640 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1643 if (rdev
->size
< chunk_size
/ 1024) {
1645 "md: Dev %s smaller than chunk_size:"
1647 bdevname(rdev
->bdev
,b
),
1648 (unsigned long long)rdev
->size
,
1658 request_module("md-personality-%d", pnum
);
1663 * Drop all container device buffers, from now on
1664 * the only valid external interface is through the md
1666 * Also find largest hardsector size
1668 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1671 sync_blockdev(rdev
->bdev
);
1672 invalidate_bdev(rdev
->bdev
, 0);
1675 md_probe(mddev
->unit
, NULL
, NULL
);
1676 disk
= mddev
->gendisk
;
1680 spin_lock(&pers_lock
);
1681 if (!pers
[pnum
] || !try_module_get(pers
[pnum
]->owner
)) {
1682 spin_unlock(&pers_lock
);
1683 printk(KERN_WARNING
"md: personality %d is not loaded!\n",
1688 mddev
->pers
= pers
[pnum
];
1689 spin_unlock(&pers_lock
);
1691 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
1693 /* before we start the array running, initialise the bitmap */
1694 err
= bitmap_create(mddev
);
1696 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
1697 mdname(mddev
), err
);
1699 err
= mddev
->pers
->run(mddev
);
1701 printk(KERN_ERR
"md: pers->run() failed ...\n");
1702 module_put(mddev
->pers
->owner
);
1704 bitmap_destroy(mddev
);
1707 atomic_set(&mddev
->writes_pending
,0);
1708 mddev
->safemode
= 0;
1709 mddev
->safemode_timer
.function
= md_safemode_timeout
;
1710 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
1711 mddev
->safemode_delay
= (20 * HZ
)/1000 +1; /* 20 msec delay */
1714 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
1716 if (mddev
->sb_dirty
)
1717 md_update_sb(mddev
);
1719 set_capacity(disk
, mddev
->array_size
<<1);
1721 /* If we call blk_queue_make_request here, it will
1722 * re-initialise max_sectors etc which may have been
1723 * refined inside -> run. So just set the bits we need to set.
1724 * Most initialisation happended when we called
1725 * blk_queue_make_request(..., md_fail_request)
1728 mddev
->queue
->queuedata
= mddev
;
1729 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
1735 static int restart_array(mddev_t
*mddev
)
1737 struct gendisk
*disk
= mddev
->gendisk
;
1741 * Complain if it has no devices
1744 if (list_empty(&mddev
->disks
))
1752 mddev
->safemode
= 0;
1754 set_disk_ro(disk
, 0);
1756 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
1759 * Kick recovery or resync if necessary
1761 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
1762 md_wakeup_thread(mddev
->thread
);
1765 printk(KERN_ERR
"md: %s has no personality assigned.\n",
1774 static int do_md_stop(mddev_t
* mddev
, int ro
)
1777 struct gendisk
*disk
= mddev
->gendisk
;
1780 if (atomic_read(&mddev
->active
)>2) {
1781 printk("md: %s still in use.\n",mdname(mddev
));
1785 if (mddev
->sync_thread
) {
1786 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1787 md_unregister_thread(mddev
->sync_thread
);
1788 mddev
->sync_thread
= NULL
;
1791 del_timer_sync(&mddev
->safemode_timer
);
1793 invalidate_partition(disk
, 0);
1801 bitmap_flush(mddev
);
1802 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
1804 set_disk_ro(disk
, 0);
1805 blk_queue_make_request(mddev
->queue
, md_fail_request
);
1806 mddev
->pers
->stop(mddev
);
1807 module_put(mddev
->pers
->owner
);
1812 if (!mddev
->in_sync
) {
1813 /* mark array as shutdown cleanly */
1815 md_update_sb(mddev
);
1818 set_disk_ro(disk
, 1);
1821 bitmap_destroy(mddev
);
1822 if (mddev
->bitmap_file
) {
1823 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
1824 fput(mddev
->bitmap_file
);
1825 mddev
->bitmap_file
= NULL
;
1829 * Free resources if final stop
1832 struct gendisk
*disk
;
1833 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
1835 export_array(mddev
);
1837 mddev
->array_size
= 0;
1838 disk
= mddev
->gendisk
;
1840 set_capacity(disk
, 0);
1843 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
1850 static void autorun_array(mddev_t
*mddev
)
1853 struct list_head
*tmp
;
1856 if (list_empty(&mddev
->disks
))
1859 printk(KERN_INFO
"md: running: ");
1861 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1862 char b
[BDEVNAME_SIZE
];
1863 printk("<%s>", bdevname(rdev
->bdev
,b
));
1867 err
= do_md_run (mddev
);
1869 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
1870 do_md_stop (mddev
, 0);
1875 * lets try to run arrays based on all disks that have arrived
1876 * until now. (those are in pending_raid_disks)
1878 * the method: pick the first pending disk, collect all disks with
1879 * the same UUID, remove all from the pending list and put them into
1880 * the 'same_array' list. Then order this list based on superblock
1881 * update time (freshest comes first), kick out 'old' disks and
1882 * compare superblocks. If everything's fine then run it.
1884 * If "unit" is allocated, then bump its reference count
1886 static void autorun_devices(int part
)
1888 struct list_head candidates
;
1889 struct list_head
*tmp
;
1890 mdk_rdev_t
*rdev0
, *rdev
;
1892 char b
[BDEVNAME_SIZE
];
1894 printk(KERN_INFO
"md: autorun ...\n");
1895 while (!list_empty(&pending_raid_disks
)) {
1897 rdev0
= list_entry(pending_raid_disks
.next
,
1898 mdk_rdev_t
, same_set
);
1900 printk(KERN_INFO
"md: considering %s ...\n",
1901 bdevname(rdev0
->bdev
,b
));
1902 INIT_LIST_HEAD(&candidates
);
1903 ITERATE_RDEV_PENDING(rdev
,tmp
)
1904 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
1905 printk(KERN_INFO
"md: adding %s ...\n",
1906 bdevname(rdev
->bdev
,b
));
1907 list_move(&rdev
->same_set
, &candidates
);
1910 * now we have a set of devices, with all of them having
1911 * mostly sane superblocks. It's time to allocate the
1914 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
1915 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
1916 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
1920 dev
= MKDEV(mdp_major
,
1921 rdev0
->preferred_minor
<< MdpMinorShift
);
1923 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
1925 md_probe(dev
, NULL
, NULL
);
1926 mddev
= mddev_find(dev
);
1929 "md: cannot allocate memory for md drive.\n");
1932 if (mddev_lock(mddev
))
1933 printk(KERN_WARNING
"md: %s locked, cannot run\n",
1935 else if (mddev
->raid_disks
|| mddev
->major_version
1936 || !list_empty(&mddev
->disks
)) {
1938 "md: %s already running, cannot run %s\n",
1939 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
1940 mddev_unlock(mddev
);
1942 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
1943 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
1944 list_del_init(&rdev
->same_set
);
1945 if (bind_rdev_to_array(rdev
, mddev
))
1948 autorun_array(mddev
);
1949 mddev_unlock(mddev
);
1951 /* on success, candidates will be empty, on error
1954 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
1958 printk(KERN_INFO
"md: ... autorun DONE.\n");
1962 * import RAID devices based on one partition
1963 * if possible, the array gets run as well.
1966 static int autostart_array(dev_t startdev
)
1968 char b
[BDEVNAME_SIZE
];
1969 int err
= -EINVAL
, i
;
1970 mdp_super_t
*sb
= NULL
;
1971 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
1973 start_rdev
= md_import_device(startdev
, 0, 0);
1974 if (IS_ERR(start_rdev
))
1978 /* NOTE: this can only work for 0.90.0 superblocks */
1979 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
1980 if (sb
->major_version
!= 0 ||
1981 sb
->minor_version
!= 90 ) {
1982 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
1983 export_rdev(start_rdev
);
1987 if (start_rdev
->faulty
) {
1989 "md: can not autostart based on faulty %s!\n",
1990 bdevname(start_rdev
->bdev
,b
));
1991 export_rdev(start_rdev
);
1994 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
1996 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1997 mdp_disk_t
*desc
= sb
->disks
+ i
;
1998 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
2002 if (dev
== startdev
)
2004 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
2006 rdev
= md_import_device(dev
, 0, 0);
2010 list_add(&rdev
->same_set
, &pending_raid_disks
);
2014 * possibly return codes
2022 static int get_version(void __user
* arg
)
2026 ver
.major
= MD_MAJOR_VERSION
;
2027 ver
.minor
= MD_MINOR_VERSION
;
2028 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
2030 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
2036 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
2038 mdu_array_info_t info
;
2039 int nr
,working
,active
,failed
,spare
;
2041 struct list_head
*tmp
;
2043 nr
=working
=active
=failed
=spare
=0;
2044 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2057 info
.major_version
= mddev
->major_version
;
2058 info
.minor_version
= mddev
->minor_version
;
2059 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
2060 info
.ctime
= mddev
->ctime
;
2061 info
.level
= mddev
->level
;
2062 info
.size
= mddev
->size
;
2064 info
.raid_disks
= mddev
->raid_disks
;
2065 info
.md_minor
= mddev
->md_minor
;
2066 info
.not_persistent
= !mddev
->persistent
;
2068 info
.utime
= mddev
->utime
;
2071 info
.state
= (1<<MD_SB_CLEAN
);
2072 info
.active_disks
= active
;
2073 info
.working_disks
= working
;
2074 info
.failed_disks
= failed
;
2075 info
.spare_disks
= spare
;
2077 info
.layout
= mddev
->layout
;
2078 info
.chunk_size
= mddev
->chunk_size
;
2080 if (copy_to_user(arg
, &info
, sizeof(info
)))
2086 static int get_bitmap_file(mddev_t
* mddev
, void * arg
)
2088 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
2089 char *ptr
, *buf
= NULL
;
2092 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
2096 /* bitmap disabled, zero the first byte and copy out */
2097 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
2098 file
->pathname
[0] = '\0';
2102 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
2106 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
2110 strcpy(file
->pathname
, ptr
);
2114 if (copy_to_user(arg
, file
, sizeof(*file
)))
2122 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
2124 mdu_disk_info_t info
;
2128 if (copy_from_user(&info
, arg
, sizeof(info
)))
2133 rdev
= find_rdev_nr(mddev
, nr
);
2135 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
2136 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
2137 info
.raid_disk
= rdev
->raid_disk
;
2140 info
.state
|= (1<<MD_DISK_FAULTY
);
2141 else if (rdev
->in_sync
) {
2142 info
.state
|= (1<<MD_DISK_ACTIVE
);
2143 info
.state
|= (1<<MD_DISK_SYNC
);
2146 info
.major
= info
.minor
= 0;
2147 info
.raid_disk
= -1;
2148 info
.state
= (1<<MD_DISK_REMOVED
);
2151 if (copy_to_user(arg
, &info
, sizeof(info
)))
2157 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
2159 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
2161 dev_t dev
= MKDEV(info
->major
,info
->minor
);
2163 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
2166 if (!mddev
->raid_disks
) {
2168 /* expecting a device which has a superblock */
2169 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
2172 "md: md_import_device returned %ld\n",
2174 return PTR_ERR(rdev
);
2176 if (!list_empty(&mddev
->disks
)) {
2177 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2178 mdk_rdev_t
, same_set
);
2179 int err
= super_types
[mddev
->major_version
]
2180 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2183 "md: %s has different UUID to %s\n",
2184 bdevname(rdev
->bdev
,b
),
2185 bdevname(rdev0
->bdev
,b2
));
2190 err
= bind_rdev_to_array(rdev
, mddev
);
2197 * add_new_disk can be used once the array is assembled
2198 * to add "hot spares". They must already have a superblock
2203 if (!mddev
->pers
->hot_add_disk
) {
2205 "%s: personality does not support diskops!\n",
2209 rdev
= md_import_device(dev
, mddev
->major_version
,
2210 mddev
->minor_version
);
2213 "md: md_import_device returned %ld\n",
2215 return PTR_ERR(rdev
);
2217 /* set save_raid_disk if appropriate */
2218 if (!mddev
->persistent
) {
2219 if (info
->state
& (1<<MD_DISK_SYNC
) &&
2220 info
->raid_disk
< mddev
->raid_disks
)
2221 rdev
->raid_disk
= info
->raid_disk
;
2223 rdev
->raid_disk
= -1;
2225 super_types
[mddev
->major_version
].
2226 validate_super(mddev
, rdev
);
2227 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2229 rdev
->in_sync
= 0; /* just to be sure */
2230 rdev
->raid_disk
= -1;
2231 err
= bind_rdev_to_array(rdev
, mddev
);
2235 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2237 md_wakeup_thread(mddev
->thread
);
2241 /* otherwise, add_new_disk is only allowed
2242 * for major_version==0 superblocks
2244 if (mddev
->major_version
!= 0) {
2245 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
2250 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
2252 rdev
= md_import_device (dev
, -1, 0);
2255 "md: error, md_import_device() returned %ld\n",
2257 return PTR_ERR(rdev
);
2259 rdev
->desc_nr
= info
->number
;
2260 if (info
->raid_disk
< mddev
->raid_disks
)
2261 rdev
->raid_disk
= info
->raid_disk
;
2263 rdev
->raid_disk
= -1;
2266 if (rdev
->raid_disk
< mddev
->raid_disks
)
2267 rdev
->in_sync
= (info
->state
& (1<<MD_DISK_SYNC
));
2271 err
= bind_rdev_to_array(rdev
, mddev
);
2277 if (!mddev
->persistent
) {
2278 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
2279 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2281 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
2282 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
2284 if (!mddev
->size
|| (mddev
->size
> rdev
->size
))
2285 mddev
->size
= rdev
->size
;
2291 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
2293 char b
[BDEVNAME_SIZE
];
2299 rdev
= find_rdev(mddev
, dev
);
2303 if (rdev
->raid_disk
>= 0)
2306 kick_rdev_from_array(rdev
);
2307 md_update_sb(mddev
);
2311 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
2312 bdevname(rdev
->bdev
,b
), mdname(mddev
));
2316 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
2318 char b
[BDEVNAME_SIZE
];
2326 if (mddev
->major_version
!= 0) {
2327 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
2328 " version-0 superblocks.\n",
2332 if (!mddev
->pers
->hot_add_disk
) {
2334 "%s: personality does not support diskops!\n",
2339 rdev
= md_import_device (dev
, -1, 0);
2342 "md: error, md_import_device() returned %ld\n",
2347 if (mddev
->persistent
)
2348 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
2351 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2353 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
2356 if (size
< mddev
->size
) {
2358 "%s: disk size %llu blocks < array size %llu\n",
2359 mdname(mddev
), (unsigned long long)size
,
2360 (unsigned long long)mddev
->size
);
2367 "md: can not hot-add faulty %s disk to %s!\n",
2368 bdevname(rdev
->bdev
,b
), mdname(mddev
));
2374 bind_rdev_to_array(rdev
, mddev
);
2377 * The rest should better be atomic, we can have disk failures
2378 * noticed in interrupt contexts ...
2381 if (rdev
->desc_nr
== mddev
->max_disks
) {
2382 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
2385 goto abort_unbind_export
;
2388 rdev
->raid_disk
= -1;
2390 md_update_sb(mddev
);
2393 * Kick recovery, maybe this spare has to be added to the
2394 * array immediately.
2396 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2397 md_wakeup_thread(mddev
->thread
);
2401 abort_unbind_export
:
2402 unbind_rdev_from_array(rdev
);
2409 /* similar to deny_write_access, but accounts for our holding a reference
2410 * to the file ourselves */
2411 static int deny_bitmap_write_access(struct file
* file
)
2413 struct inode
*inode
= file
->f_mapping
->host
;
2415 spin_lock(&inode
->i_lock
);
2416 if (atomic_read(&inode
->i_writecount
) > 1) {
2417 spin_unlock(&inode
->i_lock
);
2420 atomic_set(&inode
->i_writecount
, -1);
2421 spin_unlock(&inode
->i_lock
);
2426 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
2433 mddev
->bitmap_file
= fget(fd
);
2435 if (mddev
->bitmap_file
== NULL
) {
2436 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
2441 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
2443 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
2445 fput(mddev
->bitmap_file
);
2446 mddev
->bitmap_file
= NULL
;
2448 mddev
->bitmap_offset
= 0; /* file overrides offset */
2453 * set_array_info is used two different ways
2454 * The original usage is when creating a new array.
2455 * In this usage, raid_disks is > 0 and it together with
2456 * level, size, not_persistent,layout,chunksize determine the
2457 * shape of the array.
2458 * This will always create an array with a type-0.90.0 superblock.
2459 * The newer usage is when assembling an array.
2460 * In this case raid_disks will be 0, and the major_version field is
2461 * use to determine which style super-blocks are to be found on the devices.
2462 * The minor and patch _version numbers are also kept incase the
2463 * super_block handler wishes to interpret them.
2465 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
2468 if (info
->raid_disks
== 0) {
2469 /* just setting version number for superblock loading */
2470 if (info
->major_version
< 0 ||
2471 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2472 super_types
[info
->major_version
].name
== NULL
) {
2473 /* maybe try to auto-load a module? */
2475 "md: superblock version %d not known\n",
2476 info
->major_version
);
2479 mddev
->major_version
= info
->major_version
;
2480 mddev
->minor_version
= info
->minor_version
;
2481 mddev
->patch_version
= info
->patch_version
;
2484 mddev
->major_version
= MD_MAJOR_VERSION
;
2485 mddev
->minor_version
= MD_MINOR_VERSION
;
2486 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
2487 mddev
->ctime
= get_seconds();
2489 mddev
->level
= info
->level
;
2490 mddev
->size
= info
->size
;
2491 mddev
->raid_disks
= info
->raid_disks
;
2492 /* don't set md_minor, it is determined by which /dev/md* was
2495 if (info
->state
& (1<<MD_SB_CLEAN
))
2496 mddev
->recovery_cp
= MaxSector
;
2498 mddev
->recovery_cp
= 0;
2499 mddev
->persistent
= ! info
->not_persistent
;
2501 mddev
->layout
= info
->layout
;
2502 mddev
->chunk_size
= info
->chunk_size
;
2504 mddev
->max_disks
= MD_SB_DISKS
;
2506 mddev
->sb_dirty
= 1;
2509 * Generate a 128 bit UUID
2511 get_random_bytes(mddev
->uuid
, 16);
2517 * update_array_info is used to change the configuration of an
2519 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2520 * fields in the info are checked against the array.
2521 * Any differences that cannot be handled will cause an error.
2522 * Normally, only one change can be managed at a time.
2524 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
2529 if (mddev
->major_version
!= info
->major_version
||
2530 mddev
->minor_version
!= info
->minor_version
||
2531 /* mddev->patch_version != info->patch_version || */
2532 mddev
->ctime
!= info
->ctime
||
2533 mddev
->level
!= info
->level
||
2534 /* mddev->layout != info->layout || */
2535 !mddev
->persistent
!= info
->not_persistent
||
2536 mddev
->chunk_size
!= info
->chunk_size
)
2538 /* Check there is only one change */
2539 if (mddev
->size
!= info
->size
) cnt
++;
2540 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
2541 if (mddev
->layout
!= info
->layout
) cnt
++;
2542 if (cnt
== 0) return 0;
2543 if (cnt
> 1) return -EINVAL
;
2545 if (mddev
->layout
!= info
->layout
) {
2547 * we don't need to do anything at the md level, the
2548 * personality will take care of it all.
2550 if (mddev
->pers
->reconfig
== NULL
)
2553 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
2555 if (mddev
->size
!= info
->size
) {
2557 struct list_head
*tmp
;
2558 if (mddev
->pers
->resize
== NULL
)
2560 /* The "size" is the amount of each device that is used.
2561 * This can only make sense for arrays with redundancy.
2562 * linear and raid0 always use whatever space is available
2563 * We can only consider changing the size if no resync
2564 * or reconstruction is happening, and if the new size
2565 * is acceptable. It must fit before the sb_offset or,
2566 * if that is <data_offset, it must fit before the
2567 * size of each device.
2568 * If size is zero, we find the largest size that fits.
2570 if (mddev
->sync_thread
)
2572 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2574 int fit
= (info
->size
== 0);
2575 if (rdev
->sb_offset
> rdev
->data_offset
)
2576 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
2578 avail
= get_capacity(rdev
->bdev
->bd_disk
)
2579 - rdev
->data_offset
;
2580 if (fit
&& (info
->size
== 0 || info
->size
> avail
/2))
2581 info
->size
= avail
/2;
2582 if (avail
< ((sector_t
)info
->size
<< 1))
2585 rv
= mddev
->pers
->resize(mddev
, (sector_t
)info
->size
*2);
2587 struct block_device
*bdev
;
2589 bdev
= bdget_disk(mddev
->gendisk
, 0);
2591 down(&bdev
->bd_inode
->i_sem
);
2592 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
2593 up(&bdev
->bd_inode
->i_sem
);
2598 if (mddev
->raid_disks
!= info
->raid_disks
) {
2599 /* change the number of raid disks */
2600 if (mddev
->pers
->reshape
== NULL
)
2602 if (info
->raid_disks
<= 0 ||
2603 info
->raid_disks
>= mddev
->max_disks
)
2605 if (mddev
->sync_thread
)
2607 rv
= mddev
->pers
->reshape(mddev
, info
->raid_disks
);
2609 struct block_device
*bdev
;
2611 bdev
= bdget_disk(mddev
->gendisk
, 0);
2613 down(&bdev
->bd_inode
->i_sem
);
2614 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
2615 up(&bdev
->bd_inode
->i_sem
);
2620 md_update_sb(mddev
);
2624 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
2628 if (mddev
->pers
== NULL
)
2631 rdev
= find_rdev(mddev
, dev
);
2635 md_error(mddev
, rdev
);
2639 static int md_ioctl(struct inode
*inode
, struct file
*file
,
2640 unsigned int cmd
, unsigned long arg
)
2643 void __user
*argp
= (void __user
*)arg
;
2644 struct hd_geometry __user
*loc
= argp
;
2645 mddev_t
*mddev
= NULL
;
2647 if (!capable(CAP_SYS_ADMIN
))
2651 * Commands dealing with the RAID driver but not any
2657 err
= get_version(argp
);
2660 case PRINT_RAID_DEBUG
:
2668 autostart_arrays(arg
);
2675 * Commands creating/starting a new array:
2678 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
2686 if (cmd
== START_ARRAY
) {
2687 /* START_ARRAY doesn't need to lock the array as autostart_array
2688 * does the locking, and it could even be a different array
2693 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2694 "This will not be supported beyond 2.6\n",
2695 current
->comm
, current
->pid
);
2698 err
= autostart_array(new_decode_dev(arg
));
2700 printk(KERN_WARNING
"md: autostart failed!\n");
2706 err
= mddev_lock(mddev
);
2709 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2716 case SET_ARRAY_INFO
:
2718 mdu_array_info_t info
;
2720 memset(&info
, 0, sizeof(info
));
2721 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
2726 err
= update_array_info(mddev
, &info
);
2728 printk(KERN_WARNING
"md: couldn't update"
2729 " array info. %d\n", err
);
2734 if (!list_empty(&mddev
->disks
)) {
2736 "md: array %s already has disks!\n",
2741 if (mddev
->raid_disks
) {
2743 "md: array %s already initialised!\n",
2748 err
= set_array_info(mddev
, &info
);
2750 printk(KERN_WARNING
"md: couldn't set"
2751 " array info. %d\n", err
);
2761 * Commands querying/configuring an existing array:
2763 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2764 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2765 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
2766 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
2772 * Commands even a read-only array can execute:
2776 case GET_ARRAY_INFO
:
2777 err
= get_array_info(mddev
, argp
);
2780 case GET_BITMAP_FILE
:
2781 err
= get_bitmap_file(mddev
, (void *)arg
);
2785 err
= get_disk_info(mddev
, argp
);
2788 case RESTART_ARRAY_RW
:
2789 err
= restart_array(mddev
);
2793 err
= do_md_stop (mddev
, 0);
2797 err
= do_md_stop (mddev
, 1);
2801 * We have a problem here : there is no easy way to give a CHS
2802 * virtual geometry. We currently pretend that we have a 2 heads
2803 * 4 sectors (with a BIG number of cylinders...). This drives
2804 * dosfs just mad... ;-)
2811 err
= put_user (2, (char __user
*) &loc
->heads
);
2814 err
= put_user (4, (char __user
*) &loc
->sectors
);
2817 err
= put_user(get_capacity(mddev
->gendisk
)/8,
2818 (short __user
*) &loc
->cylinders
);
2821 err
= put_user (get_start_sect(inode
->i_bdev
),
2822 (long __user
*) &loc
->start
);
2827 * The remaining ioctls are changing the state of the
2828 * superblock, so we do not allow read-only arrays
2840 mdu_disk_info_t info
;
2841 if (copy_from_user(&info
, argp
, sizeof(info
)))
2844 err
= add_new_disk(mddev
, &info
);
2848 case HOT_REMOVE_DISK
:
2849 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
2853 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
2856 case SET_DISK_FAULTY
:
2857 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
2861 err
= do_md_run (mddev
);
2864 case SET_BITMAP_FILE
:
2865 err
= set_bitmap_file(mddev
, (int)arg
);
2869 if (_IOC_TYPE(cmd
) == MD_MAJOR
)
2870 printk(KERN_WARNING
"md: %s(pid %d) used"
2871 " obsolete MD ioctl, upgrade your"
2872 " software to use new ictls.\n",
2873 current
->comm
, current
->pid
);
2880 mddev_unlock(mddev
);
2890 static int md_open(struct inode
*inode
, struct file
*file
)
2893 * Succeed if we can lock the mddev, which confirms that
2894 * it isn't being stopped right now.
2896 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
2899 if ((err
= mddev_lock(mddev
)))
2904 mddev_unlock(mddev
);
2906 check_disk_change(inode
->i_bdev
);
2911 static int md_release(struct inode
*inode
, struct file
* file
)
2913 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
2922 static int md_media_changed(struct gendisk
*disk
)
2924 mddev_t
*mddev
= disk
->private_data
;
2926 return mddev
->changed
;
2929 static int md_revalidate(struct gendisk
*disk
)
2931 mddev_t
*mddev
= disk
->private_data
;
2936 static struct block_device_operations md_fops
=
2938 .owner
= THIS_MODULE
,
2940 .release
= md_release
,
2942 .media_changed
= md_media_changed
,
2943 .revalidate_disk
= md_revalidate
,
2946 static int md_thread(void * arg
)
2948 mdk_thread_t
*thread
= arg
;
2956 daemonize(thread
->name
, mdname(thread
->mddev
));
2958 current
->exit_signal
= SIGCHLD
;
2959 allow_signal(SIGKILL
);
2960 thread
->tsk
= current
;
2963 * md_thread is a 'system-thread', it's priority should be very
2964 * high. We avoid resource deadlocks individually in each
2965 * raid personality. (RAID5 does preallocation) We also use RR and
2966 * the very same RT priority as kswapd, thus we will never get
2967 * into a priority inversion deadlock.
2969 * we definitely have to have equal or higher priority than
2970 * bdflush, otherwise bdflush will deadlock if there are too
2971 * many dirty RAID5 blocks.
2975 complete(thread
->event
);
2976 while (thread
->run
) {
2977 void (*run
)(mddev_t
*);
2979 wait_event_interruptible_timeout(thread
->wqueue
,
2980 test_bit(THREAD_WAKEUP
, &thread
->flags
),
2984 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
2990 if (signal_pending(current
))
2991 flush_signals(current
);
2993 complete(thread
->event
);
2997 void md_wakeup_thread(mdk_thread_t
*thread
)
3000 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
3001 set_bit(THREAD_WAKEUP
, &thread
->flags
);
3002 wake_up(&thread
->wqueue
);
3006 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
3009 mdk_thread_t
*thread
;
3011 struct completion event
;
3013 thread
= (mdk_thread_t
*) kmalloc
3014 (sizeof(mdk_thread_t
), GFP_KERNEL
);
3018 memset(thread
, 0, sizeof(mdk_thread_t
));
3019 init_waitqueue_head(&thread
->wqueue
);
3021 init_completion(&event
);
3022 thread
->event
= &event
;
3024 thread
->mddev
= mddev
;
3025 thread
->name
= name
;
3026 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
3027 ret
= kernel_thread(md_thread
, thread
, 0);
3032 wait_for_completion(&event
);
3036 void md_unregister_thread(mdk_thread_t
*thread
)
3038 struct completion event
;
3040 init_completion(&event
);
3042 thread
->event
= &event
;
3044 /* As soon as ->run is set to NULL, the task could disappear,
3045 * so we need to hold tasklist_lock until we have sent the signal
3047 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
3048 read_lock(&tasklist_lock
);
3050 send_sig(SIGKILL
, thread
->tsk
, 1);
3051 read_unlock(&tasklist_lock
);
3052 wait_for_completion(&event
);
3056 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
3063 if (!rdev
|| rdev
->faulty
)
3066 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3068 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3069 __builtin_return_address(0),__builtin_return_address(1),
3070 __builtin_return_address(2),__builtin_return_address(3));
3072 if (!mddev
->pers
->error_handler
)
3074 mddev
->pers
->error_handler(mddev
,rdev
);
3075 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3076 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3077 md_wakeup_thread(mddev
->thread
);
3080 /* seq_file implementation /proc/mdstat */
3082 static void status_unused(struct seq_file
*seq
)
3086 struct list_head
*tmp
;
3088 seq_printf(seq
, "unused devices: ");
3090 ITERATE_RDEV_PENDING(rdev
,tmp
) {
3091 char b
[BDEVNAME_SIZE
];
3093 seq_printf(seq
, "%s ",
3094 bdevname(rdev
->bdev
,b
));
3097 seq_printf(seq
, "<none>");
3099 seq_printf(seq
, "\n");
3103 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
3105 unsigned long max_blocks
, resync
, res
, dt
, db
, rt
;
3107 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
3109 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3110 max_blocks
= mddev
->resync_max_sectors
>> 1;
3112 max_blocks
= mddev
->size
;
3115 * Should not happen.
3121 res
= (resync
/1024)*1000/(max_blocks
/1024 + 1);
3123 int i
, x
= res
/50, y
= 20-x
;
3124 seq_printf(seq
, "[");
3125 for (i
= 0; i
< x
; i
++)
3126 seq_printf(seq
, "=");
3127 seq_printf(seq
, ">");
3128 for (i
= 0; i
< y
; i
++)
3129 seq_printf(seq
, ".");
3130 seq_printf(seq
, "] ");
3132 seq_printf(seq
, " %s =%3lu.%lu%% (%lu/%lu)",
3133 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
3134 "resync" : "recovery"),
3135 res
/10, res
% 10, resync
, max_blocks
);
3138 * We do not want to overflow, so the order of operands and
3139 * the * 100 / 100 trick are important. We do a +1 to be
3140 * safe against division by zero. We only estimate anyway.
3142 * dt: time from mark until now
3143 * db: blocks written from mark until now
3144 * rt: remaining time
3146 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
3148 db
= resync
- (mddev
->resync_mark_cnt
/2);
3149 rt
= (dt
* ((max_blocks
-resync
) / (db
/100+1)))/100;
3151 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
3153 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
3156 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3158 struct list_head
*tmp
;
3168 spin_lock(&all_mddevs_lock
);
3169 list_for_each(tmp
,&all_mddevs
)
3171 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
3173 spin_unlock(&all_mddevs_lock
);
3176 spin_unlock(&all_mddevs_lock
);
3178 return (void*)2;/* tail */
3182 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3184 struct list_head
*tmp
;
3185 mddev_t
*next_mddev
, *mddev
= v
;
3191 spin_lock(&all_mddevs_lock
);
3193 tmp
= all_mddevs
.next
;
3195 tmp
= mddev
->all_mddevs
.next
;
3196 if (tmp
!= &all_mddevs
)
3197 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
3199 next_mddev
= (void*)2;
3202 spin_unlock(&all_mddevs_lock
);
3210 static void md_seq_stop(struct seq_file
*seq
, void *v
)
3214 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
3218 static int md_seq_show(struct seq_file
*seq
, void *v
)
3222 struct list_head
*tmp2
;
3225 struct bitmap
*bitmap
;
3227 if (v
== (void*)1) {
3228 seq_printf(seq
, "Personalities : ");
3229 spin_lock(&pers_lock
);
3230 for (i
= 0; i
< MAX_PERSONALITY
; i
++)
3232 seq_printf(seq
, "[%s] ", pers
[i
]->name
);
3234 spin_unlock(&pers_lock
);
3235 seq_printf(seq
, "\n");
3238 if (v
== (void*)2) {
3243 if (mddev_lock(mddev
)!=0)
3245 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
3246 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
3247 mddev
->pers
? "" : "in");
3250 seq_printf(seq
, " (read-only)");
3251 seq_printf(seq
, " %s", mddev
->pers
->name
);
3255 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
3256 char b
[BDEVNAME_SIZE
];
3257 seq_printf(seq
, " %s[%d]",
3258 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
3260 seq_printf(seq
, "(F)");
3266 if (!list_empty(&mddev
->disks
)) {
3268 seq_printf(seq
, "\n %llu blocks",
3269 (unsigned long long)mddev
->array_size
);
3271 seq_printf(seq
, "\n %llu blocks",
3272 (unsigned long long)size
);
3276 mddev
->pers
->status (seq
, mddev
);
3277 seq_printf(seq
, "\n ");
3278 if (mddev
->curr_resync
> 2) {
3279 status_resync (seq
, mddev
);
3280 seq_printf(seq
, "\n ");
3281 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
3282 seq_printf(seq
, " resync=DELAYED\n ");
3284 seq_printf(seq
, "\n ");
3286 if ((bitmap
= mddev
->bitmap
)) {
3287 unsigned long chunk_kb
;
3288 unsigned long flags
;
3289 spin_lock_irqsave(&bitmap
->lock
, flags
);
3290 chunk_kb
= bitmap
->chunksize
>> 10;
3291 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
3293 bitmap
->pages
- bitmap
->missing_pages
,
3295 (bitmap
->pages
- bitmap
->missing_pages
)
3296 << (PAGE_SHIFT
- 10),
3297 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
3298 chunk_kb
? "KB" : "B");
3300 seq_printf(seq
, ", file: ");
3301 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
3302 bitmap
->file
->f_dentry
," \t\n");
3305 seq_printf(seq
, "\n");
3306 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
3309 seq_printf(seq
, "\n");
3311 mddev_unlock(mddev
);
3316 static struct seq_operations md_seq_ops
= {
3317 .start
= md_seq_start
,
3318 .next
= md_seq_next
,
3319 .stop
= md_seq_stop
,
3320 .show
= md_seq_show
,
3323 static int md_seq_open(struct inode
*inode
, struct file
*file
)
3327 error
= seq_open(file
, &md_seq_ops
);
3331 static struct file_operations md_seq_fops
= {
3332 .open
= md_seq_open
,
3334 .llseek
= seq_lseek
,
3335 .release
= seq_release
,
3338 int register_md_personality(int pnum
, mdk_personality_t
*p
)
3340 if (pnum
>= MAX_PERSONALITY
) {
3342 "md: tried to install personality %s as nr %d, but max is %lu\n",
3343 p
->name
, pnum
, MAX_PERSONALITY
-1);
3347 spin_lock(&pers_lock
);
3349 spin_unlock(&pers_lock
);
3354 printk(KERN_INFO
"md: %s personality registered as nr %d\n", p
->name
, pnum
);
3355 spin_unlock(&pers_lock
);
3359 int unregister_md_personality(int pnum
)
3361 if (pnum
>= MAX_PERSONALITY
)
3364 printk(KERN_INFO
"md: %s personality unregistered\n", pers
[pnum
]->name
);
3365 spin_lock(&pers_lock
);
3367 spin_unlock(&pers_lock
);
3371 static int is_mddev_idle(mddev_t
*mddev
)
3374 struct list_head
*tmp
;
3376 unsigned long curr_events
;
3379 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3380 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
3381 curr_events
= disk_stat_read(disk
, read_sectors
) +
3382 disk_stat_read(disk
, write_sectors
) -
3383 atomic_read(&disk
->sync_io
);
3384 /* Allow some slack between valud of curr_events and last_events,
3385 * as there are some uninteresting races.
3386 * Note: the following is an unsigned comparison.
3388 if ((curr_events
- rdev
->last_events
+ 32) > 64) {
3389 rdev
->last_events
= curr_events
;
3396 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
3398 /* another "blocks" (512byte) blocks have been synced */
3399 atomic_sub(blocks
, &mddev
->recovery_active
);
3400 wake_up(&mddev
->recovery_wait
);
3402 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
3403 md_wakeup_thread(mddev
->thread
);
3404 // stop recovery, signal do_sync ....
3409 /* md_write_start(mddev, bi)
3410 * If we need to update some array metadata (e.g. 'active' flag
3411 * in superblock) before writing, schedule a superblock update
3412 * and wait for it to complete.
3414 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
3417 if (bio_data_dir(bi
) != WRITE
)
3420 atomic_inc(&mddev
->writes_pending
);
3421 if (mddev
->in_sync
) {
3422 spin_lock(&mddev
->write_lock
);
3423 if (mddev
->in_sync
) {
3425 mddev
->sb_dirty
= 1;
3426 md_wakeup_thread(mddev
->thread
);
3428 spin_unlock(&mddev
->write_lock
);
3430 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
3433 void md_write_end(mddev_t
*mddev
)
3435 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
3436 if (mddev
->safemode
== 2)
3437 md_wakeup_thread(mddev
->thread
);
3439 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
3443 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
3445 #define SYNC_MARKS 10
3446 #define SYNC_MARK_STEP (3*HZ)
3447 static void md_do_sync(mddev_t
*mddev
)
3450 unsigned int currspeed
= 0,
3452 sector_t max_sectors
,j
, io_sectors
;
3453 unsigned long mark
[SYNC_MARKS
];
3454 sector_t mark_cnt
[SYNC_MARKS
];
3456 struct list_head
*tmp
;
3457 sector_t last_check
;
3460 /* just incase thread restarts... */
3461 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
3464 /* we overload curr_resync somewhat here.
3465 * 0 == not engaged in resync at all
3466 * 2 == checking that there is no conflict with another sync
3467 * 1 == like 2, but have yielded to allow conflicting resync to
3469 * other == active in resync - this many blocks
3471 * Before starting a resync we must have set curr_resync to
3472 * 2, and then checked that every "conflicting" array has curr_resync
3473 * less than ours. When we find one that is the same or higher
3474 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3475 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3476 * This will mean we have to start checking from the beginning again.
3481 mddev
->curr_resync
= 2;
3484 if (signal_pending(current
)) {
3485 flush_signals(current
);
3488 ITERATE_MDDEV(mddev2
,tmp
) {
3489 if (mddev2
== mddev
)
3491 if (mddev2
->curr_resync
&&
3492 match_mddev_units(mddev
,mddev2
)) {
3494 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
3495 /* arbitrarily yield */
3496 mddev
->curr_resync
= 1;
3497 wake_up(&resync_wait
);
3499 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
3500 /* no need to wait here, we can wait the next
3501 * time 'round when curr_resync == 2
3504 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
3505 if (!signal_pending(current
)
3506 && mddev2
->curr_resync
>= mddev
->curr_resync
) {
3507 printk(KERN_INFO
"md: delaying resync of %s"
3508 " until %s has finished resync (they"
3509 " share one or more physical units)\n",
3510 mdname(mddev
), mdname(mddev2
));
3513 finish_wait(&resync_wait
, &wq
);
3516 finish_wait(&resync_wait
, &wq
);
3519 } while (mddev
->curr_resync
< 2);
3521 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3522 /* resync follows the size requested by the personality,
3523 * which defaults to physical size, but can be virtual size
3525 max_sectors
= mddev
->resync_max_sectors
;
3527 /* recovery follows the physical size of devices */
3528 max_sectors
= mddev
->size
<< 1;
3530 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
3531 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
3532 " %d KB/sec/disc.\n", sysctl_speed_limit_min
);
3533 printk(KERN_INFO
"md: using maximum available idle IO bandwith "
3534 "(but not more than %d KB/sec) for reconstruction.\n",
3535 sysctl_speed_limit_max
);
3537 is_mddev_idle(mddev
); /* this also initializes IO event counters */
3538 /* we don't use the checkpoint if there's a bitmap */
3539 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && !mddev
->bitmap
)
3540 j
= mddev
->recovery_cp
;
3544 for (m
= 0; m
< SYNC_MARKS
; m
++) {
3546 mark_cnt
[m
] = io_sectors
;
3549 mddev
->resync_mark
= mark
[last_mark
];
3550 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
3553 * Tune reconstruction:
3555 window
= 32*(PAGE_SIZE
/512);
3556 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
3557 window
/2,(unsigned long long) max_sectors
/2);
3559 atomic_set(&mddev
->recovery_active
, 0);
3560 init_waitqueue_head(&mddev
->recovery_wait
);
3565 "md: resuming recovery of %s from checkpoint.\n",
3567 mddev
->curr_resync
= j
;
3570 while (j
< max_sectors
) {
3574 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
3575 currspeed
< sysctl_speed_limit_min
);
3577 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
3581 if (!skipped
) { /* actual IO requested */
3582 io_sectors
+= sectors
;
3583 atomic_add(sectors
, &mddev
->recovery_active
);
3587 if (j
>1) mddev
->curr_resync
= j
;
3590 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
3593 last_check
= io_sectors
;
3595 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
3596 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
3600 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
3602 int next
= (last_mark
+1) % SYNC_MARKS
;
3604 mddev
->resync_mark
= mark
[next
];
3605 mddev
->resync_mark_cnt
= mark_cnt
[next
];
3606 mark
[next
] = jiffies
;
3607 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
3612 if (signal_pending(current
)) {
3614 * got a signal, exit.
3617 "md: md_do_sync() got signal ... exiting\n");
3618 flush_signals(current
);
3619 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3624 * this loop exits only if either when we are slower than
3625 * the 'hard' speed limit, or the system was IO-idle for
3627 * the system might be non-idle CPU-wise, but we only care
3628 * about not overloading the IO subsystem. (things like an
3629 * e2fsck being done on the RAID array should execute fast)
3631 mddev
->queue
->unplug_fn(mddev
->queue
);
3634 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
3635 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
3637 if (currspeed
> sysctl_speed_limit_min
) {
3638 if ((currspeed
> sysctl_speed_limit_max
) ||
3639 !is_mddev_idle(mddev
)) {
3640 msleep_interruptible(250);
3645 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
3647 * this also signals 'finished resyncing' to md_stop
3650 mddev
->queue
->unplug_fn(mddev
->queue
);
3652 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
3654 /* tell personality that we are finished */
3655 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
3657 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
3658 mddev
->curr_resync
> 2 &&
3659 mddev
->curr_resync
>= mddev
->recovery_cp
) {
3660 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
3662 "md: checkpointing recovery of %s.\n",
3664 mddev
->recovery_cp
= mddev
->curr_resync
;
3666 mddev
->recovery_cp
= MaxSector
;
3670 mddev
->curr_resync
= 0;
3671 wake_up(&resync_wait
);
3672 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
3673 md_wakeup_thread(mddev
->thread
);
3678 * This routine is regularly called by all per-raid-array threads to
3679 * deal with generic issues like resync and super-block update.
3680 * Raid personalities that don't have a thread (linear/raid0) do not
3681 * need this as they never do any recovery or update the superblock.
3683 * It does not do any resync itself, but rather "forks" off other threads
3684 * to do that as needed.
3685 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3686 * "->recovery" and create a thread at ->sync_thread.
3687 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3688 * and wakeups up this thread which will reap the thread and finish up.
3689 * This thread also removes any faulty devices (with nr_pending == 0).
3691 * The overall approach is:
3692 * 1/ if the superblock needs updating, update it.
3693 * 2/ If a recovery thread is running, don't do anything else.
3694 * 3/ If recovery has finished, clean up, possibly marking spares active.
3695 * 4/ If there are any faulty devices, remove them.
3696 * 5/ If array is degraded, try to add spares devices
3697 * 6/ If array has spares or is not in-sync, start a resync thread.
3699 void md_check_recovery(mddev_t
*mddev
)
3702 struct list_head
*rtmp
;
3706 bitmap_daemon_work(mddev
->bitmap
);
3711 if (signal_pending(current
)) {
3712 if (mddev
->pers
->sync_request
) {
3713 printk(KERN_INFO
"md: %s in immediate safe mode\n",
3715 mddev
->safemode
= 2;
3717 flush_signals(current
);
3722 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
3723 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
3724 (mddev
->safemode
== 1) ||
3725 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
3726 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
3730 if (mddev_trylock(mddev
)==0) {
3733 spin_lock(&mddev
->write_lock
);
3734 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
3735 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
3737 mddev
->sb_dirty
= 1;
3739 if (mddev
->safemode
== 1)
3740 mddev
->safemode
= 0;
3741 spin_unlock(&mddev
->write_lock
);
3743 if (mddev
->sb_dirty
)
3744 md_update_sb(mddev
);
3747 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
3748 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
3749 /* resync/recovery still happening */
3750 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3753 if (mddev
->sync_thread
) {
3754 /* resync has finished, collect result */
3755 md_unregister_thread(mddev
->sync_thread
);
3756 mddev
->sync_thread
= NULL
;
3757 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
3758 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
3760 /* activate any spares */
3761 mddev
->pers
->spare_active(mddev
);
3763 md_update_sb(mddev
);
3765 /* if array is no-longer degraded, then any saved_raid_disk
3766 * information must be scrapped
3768 if (!mddev
->degraded
)
3769 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3770 rdev
->saved_raid_disk
= -1;
3772 mddev
->recovery
= 0;
3773 /* flag recovery needed just to double check */
3774 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3777 if (mddev
->recovery
)
3778 /* probably just the RECOVERY_NEEDED flag */
3779 mddev
->recovery
= 0;
3781 /* no recovery is running.
3782 * remove any failed drives, then
3783 * add spares if possible.
3784 * Spare are also removed and re-added, to allow
3785 * the personality to fail the re-add.
3787 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3788 if (rdev
->raid_disk
>= 0 &&
3789 (rdev
->faulty
|| ! rdev
->in_sync
) &&
3790 atomic_read(&rdev
->nr_pending
)==0) {
3791 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0)
3792 rdev
->raid_disk
= -1;
3795 if (mddev
->degraded
) {
3796 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3797 if (rdev
->raid_disk
< 0
3799 if (mddev
->pers
->hot_add_disk(mddev
,rdev
))
3806 if (!spares
&& (mddev
->recovery_cp
== MaxSector
)) {
3807 /* nothing we can do ... */
3810 if (mddev
->pers
->sync_request
) {
3811 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3813 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3814 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
3815 /* We are adding a device or devices to an array
3816 * which has the bitmap stored on all devices.
3817 * So make sure all bitmap pages get written
3819 bitmap_write_all(mddev
->bitmap
);
3821 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3824 if (!mddev
->sync_thread
) {
3825 printk(KERN_ERR
"%s: could not start resync"
3828 /* leave the spares where they are, it shouldn't hurt */
3829 mddev
->recovery
= 0;
3831 md_wakeup_thread(mddev
->sync_thread
);
3835 mddev_unlock(mddev
);
3839 static int md_notify_reboot(struct notifier_block
*this,
3840 unsigned long code
, void *x
)
3842 struct list_head
*tmp
;
3845 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
3847 printk(KERN_INFO
"md: stopping all md devices.\n");
3849 ITERATE_MDDEV(mddev
,tmp
)
3850 if (mddev_trylock(mddev
)==0)
3851 do_md_stop (mddev
, 1);
3853 * certain more exotic SCSI devices are known to be
3854 * volatile wrt too early system reboots. While the
3855 * right place to handle this issue is the given
3856 * driver, we do want to have a safe RAID driver ...
3863 static struct notifier_block md_notifier
= {
3864 .notifier_call
= md_notify_reboot
,
3866 .priority
= INT_MAX
, /* before any real devices */
3869 static void md_geninit(void)
3871 struct proc_dir_entry
*p
;
3873 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
3875 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
3877 p
->proc_fops
= &md_seq_fops
;
3880 static int __init
md_init(void)
3884 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3885 " MD_SB_DISKS=%d\n",
3886 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
3887 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
3888 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR
,
3891 if (register_blkdev(MAJOR_NR
, "md"))
3893 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
3894 unregister_blkdev(MAJOR_NR
, "md");
3898 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
3899 md_probe
, NULL
, NULL
);
3900 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
3901 md_probe
, NULL
, NULL
);
3903 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
3904 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
3905 S_IFBLK
|S_IRUSR
|S_IWUSR
,
3908 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
3909 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
3910 S_IFBLK
|S_IRUSR
|S_IWUSR
,
3914 register_reboot_notifier(&md_notifier
);
3915 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
3925 * Searches all registered partitions for autorun RAID arrays
3928 static dev_t detected_devices
[128];
3931 void md_autodetect_dev(dev_t dev
)
3933 if (dev_cnt
>= 0 && dev_cnt
< 127)
3934 detected_devices
[dev_cnt
++] = dev
;
3938 static void autostart_arrays(int part
)
3943 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
3945 for (i
= 0; i
< dev_cnt
; i
++) {
3946 dev_t dev
= detected_devices
[i
];
3948 rdev
= md_import_device(dev
,0, 0);
3956 list_add(&rdev
->same_set
, &pending_raid_disks
);
3960 autorun_devices(part
);
3965 static __exit
void md_exit(void)
3968 struct list_head
*tmp
;
3970 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
3971 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
3972 for (i
=0; i
< MAX_MD_DEVS
; i
++)
3973 devfs_remove("md/%d", i
);
3974 for (i
=0; i
< MAX_MD_DEVS
; i
++)
3975 devfs_remove("md/d%d", i
);
3979 unregister_blkdev(MAJOR_NR
,"md");
3980 unregister_blkdev(mdp_major
, "mdp");
3981 unregister_reboot_notifier(&md_notifier
);
3982 unregister_sysctl_table(raid_table_header
);
3983 remove_proc_entry("mdstat", NULL
);
3984 ITERATE_MDDEV(mddev
,tmp
) {
3985 struct gendisk
*disk
= mddev
->gendisk
;
3988 export_array(mddev
);
3991 mddev
->gendisk
= NULL
;
3996 module_init(md_init
)
3997 module_exit(md_exit
)
3999 EXPORT_SYMBOL(register_md_personality
);
4000 EXPORT_SYMBOL(unregister_md_personality
);
4001 EXPORT_SYMBOL(md_error
);
4002 EXPORT_SYMBOL(md_done_sync
);
4003 EXPORT_SYMBOL(md_write_start
);
4004 EXPORT_SYMBOL(md_write_end
);
4005 EXPORT_SYMBOL(md_register_thread
);
4006 EXPORT_SYMBOL(md_unregister_thread
);
4007 EXPORT_SYMBOL(md_wakeup_thread
);
4008 EXPORT_SYMBOL(md_print_devices
);
4009 EXPORT_SYMBOL(md_check_recovery
);
4010 MODULE_LICENSE("GPL");