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
);
259 md_wakeup_thread(mddev
->thread
);
262 mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
265 struct list_head
*tmp
;
267 ITERATE_RDEV(mddev
,rdev
,tmp
) {
268 if (rdev
->desc_nr
== nr
)
274 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
276 struct list_head
*tmp
;
279 ITERATE_RDEV(mddev
,rdev
,tmp
) {
280 if (rdev
->bdev
->bd_dev
== dev
)
286 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
288 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
289 return MD_NEW_SIZE_BLOCKS(size
);
292 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
296 size
= rdev
->sb_offset
;
299 size
&= ~((sector_t
)chunk_size
/1024 - 1);
303 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
308 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
309 if (!rdev
->sb_page
) {
310 printk(KERN_ALERT
"md: out of memory.\n");
317 static void free_disk_sb(mdk_rdev_t
* rdev
)
320 page_cache_release(rdev
->sb_page
);
322 rdev
->sb_page
= NULL
;
329 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
331 mdk_rdev_t
*rdev
= bio
->bi_private
;
335 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
336 md_error(rdev
->mddev
, rdev
);
338 if (atomic_dec_and_test(&rdev
->mddev
->pending_writes
))
339 wake_up(&rdev
->mddev
->sb_wait
);
344 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
345 sector_t sector
, int size
, struct page
*page
)
347 /* write first size bytes of page to sector of rdev
348 * Increment mddev->pending_writes before returning
349 * and decrement it on completion, waking up sb_wait
350 * if zero is reached.
351 * If an error occurred, call md_error
353 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
355 bio
->bi_bdev
= rdev
->bdev
;
356 bio
->bi_sector
= sector
;
357 bio_add_page(bio
, page
, size
, 0);
358 bio
->bi_private
= rdev
;
359 bio
->bi_end_io
= super_written
;
360 atomic_inc(&mddev
->pending_writes
);
361 submit_bio((1<<BIO_RW
)|(1<<BIO_RW_SYNC
), bio
);
364 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
369 complete((struct completion
*)bio
->bi_private
);
373 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
374 struct page
*page
, int rw
)
376 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
377 struct completion event
;
380 rw
|= (1 << BIO_RW_SYNC
);
383 bio
->bi_sector
= sector
;
384 bio_add_page(bio
, page
, size
, 0);
385 init_completion(&event
);
386 bio
->bi_private
= &event
;
387 bio
->bi_end_io
= bi_complete
;
389 wait_for_completion(&event
);
391 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
396 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
398 char b
[BDEVNAME_SIZE
];
399 if (!rdev
->sb_page
) {
407 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
413 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
414 bdevname(rdev
->bdev
,b
));
418 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
420 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
421 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
422 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
423 (sb1
->set_uuid3
== sb2
->set_uuid3
))
431 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
434 mdp_super_t
*tmp1
, *tmp2
;
436 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
437 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
439 if (!tmp1
|| !tmp2
) {
441 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
449 * nr_disks is not constant
454 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
465 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
467 unsigned int disk_csum
, csum
;
469 disk_csum
= sb
->sb_csum
;
471 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
472 sb
->sb_csum
= disk_csum
;
478 * Handle superblock details.
479 * We want to be able to handle multiple superblock formats
480 * so we have a common interface to them all, and an array of
481 * different handlers.
482 * We rely on user-space to write the initial superblock, and support
483 * reading and updating of superblocks.
484 * Interface methods are:
485 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
486 * loads and validates a superblock on dev.
487 * if refdev != NULL, compare superblocks on both devices
489 * 0 - dev has a superblock that is compatible with refdev
490 * 1 - dev has a superblock that is compatible and newer than refdev
491 * so dev should be used as the refdev in future
492 * -EINVAL superblock incompatible or invalid
493 * -othererror e.g. -EIO
495 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
496 * Verify that dev is acceptable into mddev.
497 * The first time, mddev->raid_disks will be 0, and data from
498 * dev should be merged in. Subsequent calls check that dev
499 * is new enough. Return 0 or -EINVAL
501 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
502 * Update the superblock for rdev with data in mddev
503 * This does not write to disc.
509 struct module
*owner
;
510 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
511 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
512 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
516 * load_super for 0.90.0
518 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
520 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
526 * Calculate the position of the superblock,
527 * it's at the end of the disk.
529 * It also happens to be a multiple of 4Kb.
531 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
532 rdev
->sb_offset
= sb_offset
;
534 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
539 bdevname(rdev
->bdev
, b
);
540 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
542 if (sb
->md_magic
!= MD_SB_MAGIC
) {
543 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
548 if (sb
->major_version
!= 0 ||
549 sb
->minor_version
!= 90) {
550 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
551 sb
->major_version
, sb
->minor_version
,
556 if (sb
->raid_disks
<= 0)
559 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
560 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
565 rdev
->preferred_minor
= sb
->md_minor
;
566 rdev
->data_offset
= 0;
567 rdev
->sb_size
= MD_SB_BYTES
;
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
);
626 mddev
->bitmap_offset
= 0;
627 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
629 if (sb
->state
& (1<<MD_SB_CLEAN
))
630 mddev
->recovery_cp
= MaxSector
;
632 if (sb
->events_hi
== sb
->cp_events_hi
&&
633 sb
->events_lo
== sb
->cp_events_lo
) {
634 mddev
->recovery_cp
= sb
->recovery_cp
;
636 mddev
->recovery_cp
= 0;
639 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
640 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
641 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
642 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
644 mddev
->max_disks
= MD_SB_DISKS
;
646 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
647 mddev
->bitmap_file
== NULL
) {
648 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6) {
649 /* FIXME use a better test */
650 printk(KERN_WARNING
"md: bitmaps only support for raid1\n");
653 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
656 } else if (mddev
->pers
== NULL
) {
657 /* Insist on good event counter while assembling */
658 __u64 ev1
= md_event(sb
);
660 if (ev1
< mddev
->events
)
662 } else if (mddev
->bitmap
) {
663 /* if adding to array with a bitmap, then we can accept an
664 * older device ... but not too old.
666 __u64 ev1
= md_event(sb
);
667 if (ev1
< mddev
->bitmap
->events_cleared
)
669 } else /* just a hot-add of a new device, leave raid_disk at -1 */
672 if (mddev
->level
!= LEVEL_MULTIPATH
) {
675 desc
= sb
->disks
+ rdev
->desc_nr
;
677 if (desc
->state
& (1<<MD_DISK_FAULTY
))
679 else if (desc
->state
& (1<<MD_DISK_SYNC
) &&
680 desc
->raid_disk
< mddev
->raid_disks
) {
682 rdev
->raid_disk
= desc
->raid_disk
;
684 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
685 set_bit(WriteMostly
, &rdev
->flags
);
686 } else /* MULTIPATH are always insync */
692 * sync_super for 0.90.0
694 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
697 struct list_head
*tmp
;
699 int next_spare
= mddev
->raid_disks
;
701 /* make rdev->sb match mddev data..
704 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
705 * 3/ any empty disks < next_spare become removed
707 * disks[0] gets initialised to REMOVED because
708 * we cannot be sure from other fields if it has
709 * been initialised or not.
712 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
714 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
716 memset(sb
, 0, sizeof(*sb
));
718 sb
->md_magic
= MD_SB_MAGIC
;
719 sb
->major_version
= mddev
->major_version
;
720 sb
->minor_version
= mddev
->minor_version
;
721 sb
->patch_version
= mddev
->patch_version
;
722 sb
->gvalid_words
= 0; /* ignored */
723 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
724 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
725 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
726 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
728 sb
->ctime
= mddev
->ctime
;
729 sb
->level
= mddev
->level
;
730 sb
->size
= mddev
->size
;
731 sb
->raid_disks
= mddev
->raid_disks
;
732 sb
->md_minor
= mddev
->md_minor
;
733 sb
->not_persistent
= !mddev
->persistent
;
734 sb
->utime
= mddev
->utime
;
736 sb
->events_hi
= (mddev
->events
>>32);
737 sb
->events_lo
= (u32
)mddev
->events
;
741 sb
->recovery_cp
= mddev
->recovery_cp
;
742 sb
->cp_events_hi
= (mddev
->events
>>32);
743 sb
->cp_events_lo
= (u32
)mddev
->events
;
744 if (mddev
->recovery_cp
== MaxSector
)
745 sb
->state
= (1<< MD_SB_CLEAN
);
749 sb
->layout
= mddev
->layout
;
750 sb
->chunk_size
= mddev
->chunk_size
;
752 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
753 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
755 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
756 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
758 if (rdev2
->raid_disk
>= 0 && rdev2
->in_sync
&& !rdev2
->faulty
)
759 rdev2
->desc_nr
= rdev2
->raid_disk
;
761 rdev2
->desc_nr
= next_spare
++;
762 d
= &sb
->disks
[rdev2
->desc_nr
];
764 d
->number
= rdev2
->desc_nr
;
765 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
766 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
767 if (rdev2
->raid_disk
>= 0 && rdev
->in_sync
&& !rdev2
->faulty
)
768 d
->raid_disk
= rdev2
->raid_disk
;
770 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
772 d
->state
= (1<<MD_DISK_FAULTY
);
774 } else if (rdev2
->in_sync
) {
775 d
->state
= (1<<MD_DISK_ACTIVE
);
776 d
->state
|= (1<<MD_DISK_SYNC
);
784 if (test_bit(WriteMostly
, &rdev2
->flags
))
785 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
788 /* now set the "removed" and "faulty" bits on any missing devices */
789 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
790 mdp_disk_t
*d
= &sb
->disks
[i
];
791 if (d
->state
== 0 && d
->number
== 0) {
794 d
->state
= (1<<MD_DISK_REMOVED
);
795 d
->state
|= (1<<MD_DISK_FAULTY
);
799 sb
->nr_disks
= nr_disks
;
800 sb
->active_disks
= active
;
801 sb
->working_disks
= working
;
802 sb
->failed_disks
= failed
;
803 sb
->spare_disks
= spare
;
805 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
806 sb
->sb_csum
= calc_sb_csum(sb
);
810 * version 1 superblock
813 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
815 unsigned int disk_csum
, csum
;
816 unsigned long long newcsum
;
817 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
818 unsigned int *isuper
= (unsigned int*)sb
;
821 disk_csum
= sb
->sb_csum
;
824 for (i
=0; size
>=4; size
-= 4 )
825 newcsum
+= le32_to_cpu(*isuper
++);
828 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
830 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
831 sb
->sb_csum
= disk_csum
;
832 return cpu_to_le32(csum
);
835 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
837 struct mdp_superblock_1
*sb
;
840 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
844 * Calculate the position of the superblock.
845 * It is always aligned to a 4K boundary and
846 * depeding on minor_version, it can be:
847 * 0: At least 8K, but less than 12K, from end of device
848 * 1: At start of device
849 * 2: 4K from start of device.
851 switch(minor_version
) {
853 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
855 sb_offset
&= ~(sector_t
)(4*2-1);
856 /* convert from sectors to K */
868 rdev
->sb_offset
= sb_offset
;
870 /* superblock is rarely larger than 1K, but it can be larger,
871 * and it is safe to read 4k, so we do that
873 ret
= read_disk_sb(rdev
, 4096);
877 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
879 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
880 sb
->major_version
!= cpu_to_le32(1) ||
881 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
882 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
883 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
886 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
887 printk("md: invalid superblock checksum on %s\n",
888 bdevname(rdev
->bdev
,b
));
891 if (le64_to_cpu(sb
->data_size
) < 10) {
892 printk("md: data_size too small on %s\n",
893 bdevname(rdev
->bdev
,b
));
896 rdev
->preferred_minor
= 0xffff;
897 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
899 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
900 bmask
= block_size(rdev
->bdev
)-1;
901 if (rdev
->sb_size
& bmask
)
902 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
908 struct mdp_superblock_1
*refsb
=
909 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
911 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
912 sb
->level
!= refsb
->level
||
913 sb
->layout
!= refsb
->layout
||
914 sb
->chunksize
!= refsb
->chunksize
) {
915 printk(KERN_WARNING
"md: %s has strangely different"
916 " superblock to %s\n",
917 bdevname(rdev
->bdev
,b
),
918 bdevname(refdev
->bdev
,b2
));
921 ev1
= le64_to_cpu(sb
->events
);
922 ev2
= le64_to_cpu(refsb
->events
);
928 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
930 rdev
->size
= rdev
->sb_offset
;
931 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
933 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
934 if (le32_to_cpu(sb
->chunksize
))
935 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
939 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
941 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
943 rdev
->raid_disk
= -1;
945 if (mddev
->raid_disks
== 0) {
946 mddev
->major_version
= 1;
947 mddev
->patch_version
= 0;
948 mddev
->persistent
= 1;
949 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
950 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
951 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
952 mddev
->level
= le32_to_cpu(sb
->level
);
953 mddev
->layout
= le32_to_cpu(sb
->layout
);
954 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
955 mddev
->size
= le64_to_cpu(sb
->size
)/2;
956 mddev
->events
= le64_to_cpu(sb
->events
);
957 mddev
->bitmap_offset
= 0;
958 mddev
->default_bitmap_offset
= 0;
959 if (mddev
->minor_version
== 0)
960 mddev
->default_bitmap_offset
= -(64*1024)/512;
962 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
963 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
965 mddev
->max_disks
= (4096-256)/2;
967 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
968 mddev
->bitmap_file
== NULL
) {
969 if (mddev
->level
!= 1) {
970 printk(KERN_WARNING
"md: bitmaps only supported for raid1\n");
973 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
975 } else if (mddev
->pers
== NULL
) {
976 /* Insist of good event counter while assembling */
977 __u64 ev1
= le64_to_cpu(sb
->events
);
979 if (ev1
< mddev
->events
)
981 } else if (mddev
->bitmap
) {
982 /* If adding to array with a bitmap, then we can accept an
983 * older device, but not too old.
985 __u64 ev1
= le64_to_cpu(sb
->events
);
986 if (ev1
< mddev
->bitmap
->events_cleared
)
988 } else /* just a hot-add of a new device, leave raid_disk at -1 */
991 if (mddev
->level
!= LEVEL_MULTIPATH
) {
993 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
994 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
996 case 0xffff: /* spare */
999 case 0xfffe: /* faulty */
1005 rdev
->raid_disk
= role
;
1009 if (sb
->devflags
& WriteMostly1
)
1010 set_bit(WriteMostly
, &rdev
->flags
);
1011 } else /* MULTIPATH are always insync */
1017 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1019 struct mdp_superblock_1
*sb
;
1020 struct list_head
*tmp
;
1023 /* make rdev->sb match mddev and rdev data. */
1025 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1027 sb
->feature_map
= 0;
1029 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1030 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1031 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1033 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1034 sb
->events
= cpu_to_le64(mddev
->events
);
1036 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1038 sb
->resync_offset
= cpu_to_le64(0);
1040 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1041 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1042 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1046 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1047 if (rdev2
->desc_nr
+1 > max_dev
)
1048 max_dev
= rdev2
->desc_nr
+1;
1050 sb
->max_dev
= cpu_to_le32(max_dev
);
1051 for (i
=0; i
<max_dev
;i
++)
1052 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1054 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1057 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1058 else if (rdev2
->in_sync
)
1059 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1061 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1064 sb
->recovery_offset
= cpu_to_le64(0); /* not supported yet */
1065 sb
->sb_csum
= calc_sb_1_csum(sb
);
1069 static struct super_type super_types
[] = {
1072 .owner
= THIS_MODULE
,
1073 .load_super
= super_90_load
,
1074 .validate_super
= super_90_validate
,
1075 .sync_super
= super_90_sync
,
1079 .owner
= THIS_MODULE
,
1080 .load_super
= super_1_load
,
1081 .validate_super
= super_1_validate
,
1082 .sync_super
= super_1_sync
,
1086 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1088 struct list_head
*tmp
;
1091 ITERATE_RDEV(mddev
,rdev
,tmp
)
1092 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1098 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1100 struct list_head
*tmp
;
1103 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1104 if (match_dev_unit(mddev2
, rdev
))
1110 static LIST_HEAD(pending_raid_disks
);
1112 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1114 mdk_rdev_t
*same_pdev
;
1115 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1121 same_pdev
= match_dev_unit(mddev
, rdev
);
1124 "%s: WARNING: %s appears to be on the same physical"
1125 " disk as %s. True\n protection against single-disk"
1126 " failure might be compromised.\n",
1127 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1128 bdevname(same_pdev
->bdev
,b2
));
1130 /* Verify rdev->desc_nr is unique.
1131 * If it is -1, assign a free number, else
1132 * check number is not in use
1134 if (rdev
->desc_nr
< 0) {
1136 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1137 while (find_rdev_nr(mddev
, choice
))
1139 rdev
->desc_nr
= choice
;
1141 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1145 list_add(&rdev
->same_set
, &mddev
->disks
);
1146 rdev
->mddev
= mddev
;
1147 printk(KERN_INFO
"md: bind<%s>\n", bdevname(rdev
->bdev
,b
));
1151 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1153 char b
[BDEVNAME_SIZE
];
1158 list_del_init(&rdev
->same_set
);
1159 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1164 * prevent the device from being mounted, repartitioned or
1165 * otherwise reused by a RAID array (or any other kernel
1166 * subsystem), by bd_claiming the device.
1168 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1171 struct block_device
*bdev
;
1172 char b
[BDEVNAME_SIZE
];
1174 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1176 printk(KERN_ERR
"md: could not open %s.\n",
1177 __bdevname(dev
, b
));
1178 return PTR_ERR(bdev
);
1180 err
= bd_claim(bdev
, rdev
);
1182 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1191 static void unlock_rdev(mdk_rdev_t
*rdev
)
1193 struct block_device
*bdev
= rdev
->bdev
;
1201 void md_autodetect_dev(dev_t dev
);
1203 static void export_rdev(mdk_rdev_t
* rdev
)
1205 char b
[BDEVNAME_SIZE
];
1206 printk(KERN_INFO
"md: export_rdev(%s)\n",
1207 bdevname(rdev
->bdev
,b
));
1211 list_del_init(&rdev
->same_set
);
1213 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1219 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1221 unbind_rdev_from_array(rdev
);
1225 static void export_array(mddev_t
*mddev
)
1227 struct list_head
*tmp
;
1230 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1235 kick_rdev_from_array(rdev
);
1237 if (!list_empty(&mddev
->disks
))
1239 mddev
->raid_disks
= 0;
1240 mddev
->major_version
= 0;
1243 static void print_desc(mdp_disk_t
*desc
)
1245 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1246 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1249 static void print_sb(mdp_super_t
*sb
)
1254 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1255 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1256 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1258 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1259 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1260 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1261 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1262 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1263 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1264 sb
->failed_disks
, sb
->spare_disks
,
1265 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1268 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1271 desc
= sb
->disks
+ i
;
1272 if (desc
->number
|| desc
->major
|| desc
->minor
||
1273 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1274 printk(" D %2d: ", i
);
1278 printk(KERN_INFO
"md: THIS: ");
1279 print_desc(&sb
->this_disk
);
1283 static void print_rdev(mdk_rdev_t
*rdev
)
1285 char b
[BDEVNAME_SIZE
];
1286 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1287 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1288 rdev
->faulty
, rdev
->in_sync
, rdev
->desc_nr
);
1289 if (rdev
->sb_loaded
) {
1290 printk(KERN_INFO
"md: rdev superblock:\n");
1291 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1293 printk(KERN_INFO
"md: no rdev superblock!\n");
1296 void md_print_devices(void)
1298 struct list_head
*tmp
, *tmp2
;
1301 char b
[BDEVNAME_SIZE
];
1304 printk("md: **********************************\n");
1305 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1306 printk("md: **********************************\n");
1307 ITERATE_MDDEV(mddev
,tmp
) {
1310 bitmap_print_sb(mddev
->bitmap
);
1312 printk("%s: ", mdname(mddev
));
1313 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1314 printk("<%s>", bdevname(rdev
->bdev
,b
));
1317 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1320 printk("md: **********************************\n");
1325 static void sync_sbs(mddev_t
* mddev
)
1328 struct list_head
*tmp
;
1330 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1331 super_types
[mddev
->major_version
].
1332 sync_super(mddev
, rdev
);
1333 rdev
->sb_loaded
= 1;
1337 static void md_update_sb(mddev_t
* mddev
)
1340 struct list_head
*tmp
;
1345 spin_lock(&mddev
->write_lock
);
1346 sync_req
= mddev
->in_sync
;
1347 mddev
->utime
= get_seconds();
1350 if (!mddev
->events
) {
1352 * oops, this 64-bit counter should never wrap.
1353 * Either we are in around ~1 trillion A.C., assuming
1354 * 1 reboot per second, or we have a bug:
1359 mddev
->sb_dirty
= 2;
1363 * do not write anything to disk if using
1364 * nonpersistent superblocks
1366 if (!mddev
->persistent
) {
1367 mddev
->sb_dirty
= 0;
1368 spin_unlock(&mddev
->write_lock
);
1369 wake_up(&mddev
->sb_wait
);
1372 spin_unlock(&mddev
->write_lock
);
1375 "md: updating %s RAID superblock on device (in sync %d)\n",
1376 mdname(mddev
),mddev
->in_sync
);
1378 err
= bitmap_update_sb(mddev
->bitmap
);
1379 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1380 char b
[BDEVNAME_SIZE
];
1381 dprintk(KERN_INFO
"md: ");
1383 dprintk("(skipping faulty ");
1385 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1386 if (!rdev
->faulty
) {
1387 md_super_write(mddev
,rdev
,
1388 rdev
->sb_offset
<<1, rdev
->sb_size
,
1390 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1391 bdevname(rdev
->bdev
,b
),
1392 (unsigned long long)rdev
->sb_offset
);
1396 if (mddev
->level
== LEVEL_MULTIPATH
)
1397 /* only need to write one superblock... */
1400 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
1401 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1403 spin_lock(&mddev
->write_lock
);
1404 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1405 /* have to write it out again */
1406 spin_unlock(&mddev
->write_lock
);
1409 mddev
->sb_dirty
= 0;
1410 spin_unlock(&mddev
->write_lock
);
1411 wake_up(&mddev
->sb_wait
);
1416 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1418 * mark the device faulty if:
1420 * - the device is nonexistent (zero size)
1421 * - the device has no valid superblock
1423 * a faulty rdev _never_ has rdev->sb set.
1425 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1427 char b
[BDEVNAME_SIZE
];
1432 rdev
= (mdk_rdev_t
*) kmalloc(sizeof(*rdev
), GFP_KERNEL
);
1434 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1435 return ERR_PTR(-ENOMEM
);
1437 memset(rdev
, 0, sizeof(*rdev
));
1439 if ((err
= alloc_disk_sb(rdev
)))
1442 err
= lock_rdev(rdev
, newdev
);
1449 rdev
->data_offset
= 0;
1450 atomic_set(&rdev
->nr_pending
, 0);
1452 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1455 "md: %s has zero or unknown size, marking faulty!\n",
1456 bdevname(rdev
->bdev
,b
));
1461 if (super_format
>= 0) {
1462 err
= super_types
[super_format
].
1463 load_super(rdev
, NULL
, super_minor
);
1464 if (err
== -EINVAL
) {
1466 "md: %s has invalid sb, not importing!\n",
1467 bdevname(rdev
->bdev
,b
));
1472 "md: could not read %s's sb, not importing!\n",
1473 bdevname(rdev
->bdev
,b
));
1477 INIT_LIST_HEAD(&rdev
->same_set
);
1482 if (rdev
->sb_page
) {
1488 return ERR_PTR(err
);
1492 * Check a full RAID array for plausibility
1496 static void analyze_sbs(mddev_t
* mddev
)
1499 struct list_head
*tmp
;
1500 mdk_rdev_t
*rdev
, *freshest
;
1501 char b
[BDEVNAME_SIZE
];
1504 ITERATE_RDEV(mddev
,rdev
,tmp
)
1505 switch (super_types
[mddev
->major_version
].
1506 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1514 "md: fatal superblock inconsistency in %s"
1515 " -- removing from array\n",
1516 bdevname(rdev
->bdev
,b
));
1517 kick_rdev_from_array(rdev
);
1521 super_types
[mddev
->major_version
].
1522 validate_super(mddev
, freshest
);
1525 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1526 if (rdev
!= freshest
)
1527 if (super_types
[mddev
->major_version
].
1528 validate_super(mddev
, rdev
)) {
1529 printk(KERN_WARNING
"md: kicking non-fresh %s"
1531 bdevname(rdev
->bdev
,b
));
1532 kick_rdev_from_array(rdev
);
1535 if (mddev
->level
== LEVEL_MULTIPATH
) {
1536 rdev
->desc_nr
= i
++;
1537 rdev
->raid_disk
= rdev
->desc_nr
;
1544 if (mddev
->recovery_cp
!= MaxSector
&&
1546 printk(KERN_ERR
"md: %s: raid array is not clean"
1547 " -- starting background reconstruction\n",
1554 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
1556 static DECLARE_MUTEX(disks_sem
);
1557 mddev_t
*mddev
= mddev_find(dev
);
1558 struct gendisk
*disk
;
1559 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
1560 int shift
= partitioned
? MdpMinorShift
: 0;
1561 int unit
= MINOR(dev
) >> shift
;
1567 if (mddev
->gendisk
) {
1572 disk
= alloc_disk(1 << shift
);
1578 disk
->major
= MAJOR(dev
);
1579 disk
->first_minor
= unit
<< shift
;
1581 sprintf(disk
->disk_name
, "md_d%d", unit
);
1582 sprintf(disk
->devfs_name
, "md/d%d", unit
);
1584 sprintf(disk
->disk_name
, "md%d", unit
);
1585 sprintf(disk
->devfs_name
, "md/%d", unit
);
1587 disk
->fops
= &md_fops
;
1588 disk
->private_data
= mddev
;
1589 disk
->queue
= mddev
->queue
;
1591 mddev
->gendisk
= disk
;
1596 void md_wakeup_thread(mdk_thread_t
*thread
);
1598 static void md_safemode_timeout(unsigned long data
)
1600 mddev_t
*mddev
= (mddev_t
*) data
;
1602 mddev
->safemode
= 1;
1603 md_wakeup_thread(mddev
->thread
);
1607 static int do_md_run(mddev_t
* mddev
)
1611 struct list_head
*tmp
;
1613 struct gendisk
*disk
;
1614 char b
[BDEVNAME_SIZE
];
1616 if (list_empty(&mddev
->disks
))
1617 /* cannot run an array with no devices.. */
1624 * Analyze all RAID superblock(s)
1626 if (!mddev
->raid_disks
)
1629 chunk_size
= mddev
->chunk_size
;
1630 pnum
= level_to_pers(mddev
->level
);
1632 if ((pnum
!= MULTIPATH
) && (pnum
!= RAID1
)) {
1635 * 'default chunksize' in the old md code used to
1636 * be PAGE_SIZE, baaad.
1637 * we abort here to be on the safe side. We don't
1638 * want to continue the bad practice.
1641 "no chunksize specified, see 'man raidtab'\n");
1644 if (chunk_size
> MAX_CHUNK_SIZE
) {
1645 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
1646 chunk_size
, MAX_CHUNK_SIZE
);
1650 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1652 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
1653 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
1656 if (chunk_size
< PAGE_SIZE
) {
1657 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
1658 chunk_size
, PAGE_SIZE
);
1662 /* devices must have minimum size of one chunk */
1663 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1666 if (rdev
->size
< chunk_size
/ 1024) {
1668 "md: Dev %s smaller than chunk_size:"
1670 bdevname(rdev
->bdev
,b
),
1671 (unsigned long long)rdev
->size
,
1681 request_module("md-personality-%d", pnum
);
1686 * Drop all container device buffers, from now on
1687 * the only valid external interface is through the md
1689 * Also find largest hardsector size
1691 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1694 sync_blockdev(rdev
->bdev
);
1695 invalidate_bdev(rdev
->bdev
, 0);
1698 md_probe(mddev
->unit
, NULL
, NULL
);
1699 disk
= mddev
->gendisk
;
1703 spin_lock(&pers_lock
);
1704 if (!pers
[pnum
] || !try_module_get(pers
[pnum
]->owner
)) {
1705 spin_unlock(&pers_lock
);
1706 printk(KERN_WARNING
"md: personality %d is not loaded!\n",
1711 mddev
->pers
= pers
[pnum
];
1712 spin_unlock(&pers_lock
);
1714 mddev
->recovery
= 0;
1715 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
1717 /* before we start the array running, initialise the bitmap */
1718 err
= bitmap_create(mddev
);
1720 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
1721 mdname(mddev
), err
);
1723 err
= mddev
->pers
->run(mddev
);
1725 printk(KERN_ERR
"md: pers->run() failed ...\n");
1726 module_put(mddev
->pers
->owner
);
1728 bitmap_destroy(mddev
);
1731 atomic_set(&mddev
->writes_pending
,0);
1732 mddev
->safemode
= 0;
1733 mddev
->safemode_timer
.function
= md_safemode_timeout
;
1734 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
1735 mddev
->safemode_delay
= (20 * HZ
)/1000 +1; /* 20 msec delay */
1738 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
1739 md_wakeup_thread(mddev
->thread
);
1741 if (mddev
->sb_dirty
)
1742 md_update_sb(mddev
);
1744 set_capacity(disk
, mddev
->array_size
<<1);
1746 /* If we call blk_queue_make_request here, it will
1747 * re-initialise max_sectors etc which may have been
1748 * refined inside -> run. So just set the bits we need to set.
1749 * Most initialisation happended when we called
1750 * blk_queue_make_request(..., md_fail_request)
1753 mddev
->queue
->queuedata
= mddev
;
1754 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
1760 static int restart_array(mddev_t
*mddev
)
1762 struct gendisk
*disk
= mddev
->gendisk
;
1766 * Complain if it has no devices
1769 if (list_empty(&mddev
->disks
))
1777 mddev
->safemode
= 0;
1779 set_disk_ro(disk
, 0);
1781 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
1784 * Kick recovery or resync if necessary
1786 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
1787 md_wakeup_thread(mddev
->thread
);
1790 printk(KERN_ERR
"md: %s has no personality assigned.\n",
1799 static int do_md_stop(mddev_t
* mddev
, int ro
)
1802 struct gendisk
*disk
= mddev
->gendisk
;
1805 if (atomic_read(&mddev
->active
)>2) {
1806 printk("md: %s still in use.\n",mdname(mddev
));
1810 if (mddev
->sync_thread
) {
1811 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1812 md_unregister_thread(mddev
->sync_thread
);
1813 mddev
->sync_thread
= NULL
;
1816 del_timer_sync(&mddev
->safemode_timer
);
1818 invalidate_partition(disk
, 0);
1826 bitmap_flush(mddev
);
1827 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
1829 set_disk_ro(disk
, 0);
1830 blk_queue_make_request(mddev
->queue
, md_fail_request
);
1831 mddev
->pers
->stop(mddev
);
1832 module_put(mddev
->pers
->owner
);
1837 if (!mddev
->in_sync
) {
1838 /* mark array as shutdown cleanly */
1840 md_update_sb(mddev
);
1843 set_disk_ro(disk
, 1);
1846 bitmap_destroy(mddev
);
1847 if (mddev
->bitmap_file
) {
1848 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
1849 fput(mddev
->bitmap_file
);
1850 mddev
->bitmap_file
= NULL
;
1852 mddev
->bitmap_offset
= 0;
1855 * Free resources if final stop
1858 struct gendisk
*disk
;
1859 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
1861 export_array(mddev
);
1863 mddev
->array_size
= 0;
1864 disk
= mddev
->gendisk
;
1866 set_capacity(disk
, 0);
1869 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
1876 static void autorun_array(mddev_t
*mddev
)
1879 struct list_head
*tmp
;
1882 if (list_empty(&mddev
->disks
))
1885 printk(KERN_INFO
"md: running: ");
1887 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1888 char b
[BDEVNAME_SIZE
];
1889 printk("<%s>", bdevname(rdev
->bdev
,b
));
1893 err
= do_md_run (mddev
);
1895 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
1896 do_md_stop (mddev
, 0);
1901 * lets try to run arrays based on all disks that have arrived
1902 * until now. (those are in pending_raid_disks)
1904 * the method: pick the first pending disk, collect all disks with
1905 * the same UUID, remove all from the pending list and put them into
1906 * the 'same_array' list. Then order this list based on superblock
1907 * update time (freshest comes first), kick out 'old' disks and
1908 * compare superblocks. If everything's fine then run it.
1910 * If "unit" is allocated, then bump its reference count
1912 static void autorun_devices(int part
)
1914 struct list_head candidates
;
1915 struct list_head
*tmp
;
1916 mdk_rdev_t
*rdev0
, *rdev
;
1918 char b
[BDEVNAME_SIZE
];
1920 printk(KERN_INFO
"md: autorun ...\n");
1921 while (!list_empty(&pending_raid_disks
)) {
1923 rdev0
= list_entry(pending_raid_disks
.next
,
1924 mdk_rdev_t
, same_set
);
1926 printk(KERN_INFO
"md: considering %s ...\n",
1927 bdevname(rdev0
->bdev
,b
));
1928 INIT_LIST_HEAD(&candidates
);
1929 ITERATE_RDEV_PENDING(rdev
,tmp
)
1930 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
1931 printk(KERN_INFO
"md: adding %s ...\n",
1932 bdevname(rdev
->bdev
,b
));
1933 list_move(&rdev
->same_set
, &candidates
);
1936 * now we have a set of devices, with all of them having
1937 * mostly sane superblocks. It's time to allocate the
1940 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
1941 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
1942 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
1946 dev
= MKDEV(mdp_major
,
1947 rdev0
->preferred_minor
<< MdpMinorShift
);
1949 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
1951 md_probe(dev
, NULL
, NULL
);
1952 mddev
= mddev_find(dev
);
1955 "md: cannot allocate memory for md drive.\n");
1958 if (mddev_lock(mddev
))
1959 printk(KERN_WARNING
"md: %s locked, cannot run\n",
1961 else if (mddev
->raid_disks
|| mddev
->major_version
1962 || !list_empty(&mddev
->disks
)) {
1964 "md: %s already running, cannot run %s\n",
1965 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
1966 mddev_unlock(mddev
);
1968 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
1969 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
1970 list_del_init(&rdev
->same_set
);
1971 if (bind_rdev_to_array(rdev
, mddev
))
1974 autorun_array(mddev
);
1975 mddev_unlock(mddev
);
1977 /* on success, candidates will be empty, on error
1980 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
1984 printk(KERN_INFO
"md: ... autorun DONE.\n");
1988 * import RAID devices based on one partition
1989 * if possible, the array gets run as well.
1992 static int autostart_array(dev_t startdev
)
1994 char b
[BDEVNAME_SIZE
];
1995 int err
= -EINVAL
, i
;
1996 mdp_super_t
*sb
= NULL
;
1997 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
1999 start_rdev
= md_import_device(startdev
, 0, 0);
2000 if (IS_ERR(start_rdev
))
2004 /* NOTE: this can only work for 0.90.0 superblocks */
2005 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
2006 if (sb
->major_version
!= 0 ||
2007 sb
->minor_version
!= 90 ) {
2008 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
2009 export_rdev(start_rdev
);
2013 if (start_rdev
->faulty
) {
2015 "md: can not autostart based on faulty %s!\n",
2016 bdevname(start_rdev
->bdev
,b
));
2017 export_rdev(start_rdev
);
2020 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
2022 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2023 mdp_disk_t
*desc
= sb
->disks
+ i
;
2024 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
2028 if (dev
== startdev
)
2030 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
2032 rdev
= md_import_device(dev
, 0, 0);
2036 list_add(&rdev
->same_set
, &pending_raid_disks
);
2040 * possibly return codes
2048 static int get_version(void __user
* arg
)
2052 ver
.major
= MD_MAJOR_VERSION
;
2053 ver
.minor
= MD_MINOR_VERSION
;
2054 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
2056 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
2062 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
2064 mdu_array_info_t info
;
2065 int nr
,working
,active
,failed
,spare
;
2067 struct list_head
*tmp
;
2069 nr
=working
=active
=failed
=spare
=0;
2070 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2083 info
.major_version
= mddev
->major_version
;
2084 info
.minor_version
= mddev
->minor_version
;
2085 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
2086 info
.ctime
= mddev
->ctime
;
2087 info
.level
= mddev
->level
;
2088 info
.size
= mddev
->size
;
2090 info
.raid_disks
= mddev
->raid_disks
;
2091 info
.md_minor
= mddev
->md_minor
;
2092 info
.not_persistent
= !mddev
->persistent
;
2094 info
.utime
= mddev
->utime
;
2097 info
.state
= (1<<MD_SB_CLEAN
);
2098 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
2099 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
2100 info
.active_disks
= active
;
2101 info
.working_disks
= working
;
2102 info
.failed_disks
= failed
;
2103 info
.spare_disks
= spare
;
2105 info
.layout
= mddev
->layout
;
2106 info
.chunk_size
= mddev
->chunk_size
;
2108 if (copy_to_user(arg
, &info
, sizeof(info
)))
2114 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
2116 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
2117 char *ptr
, *buf
= NULL
;
2120 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
2124 /* bitmap disabled, zero the first byte and copy out */
2125 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
2126 file
->pathname
[0] = '\0';
2130 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
2134 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
2138 strcpy(file
->pathname
, ptr
);
2142 if (copy_to_user(arg
, file
, sizeof(*file
)))
2150 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
2152 mdu_disk_info_t info
;
2156 if (copy_from_user(&info
, arg
, sizeof(info
)))
2161 rdev
= find_rdev_nr(mddev
, nr
);
2163 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
2164 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
2165 info
.raid_disk
= rdev
->raid_disk
;
2168 info
.state
|= (1<<MD_DISK_FAULTY
);
2169 else if (rdev
->in_sync
) {
2170 info
.state
|= (1<<MD_DISK_ACTIVE
);
2171 info
.state
|= (1<<MD_DISK_SYNC
);
2173 if (test_bit(WriteMostly
, &rdev
->flags
))
2174 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
2176 info
.major
= info
.minor
= 0;
2177 info
.raid_disk
= -1;
2178 info
.state
= (1<<MD_DISK_REMOVED
);
2181 if (copy_to_user(arg
, &info
, sizeof(info
)))
2187 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
2189 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
2191 dev_t dev
= MKDEV(info
->major
,info
->minor
);
2193 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
2196 if (!mddev
->raid_disks
) {
2198 /* expecting a device which has a superblock */
2199 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
2202 "md: md_import_device returned %ld\n",
2204 return PTR_ERR(rdev
);
2206 if (!list_empty(&mddev
->disks
)) {
2207 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2208 mdk_rdev_t
, same_set
);
2209 int err
= super_types
[mddev
->major_version
]
2210 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2213 "md: %s has different UUID to %s\n",
2214 bdevname(rdev
->bdev
,b
),
2215 bdevname(rdev0
->bdev
,b2
));
2220 err
= bind_rdev_to_array(rdev
, mddev
);
2227 * add_new_disk can be used once the array is assembled
2228 * to add "hot spares". They must already have a superblock
2233 if (!mddev
->pers
->hot_add_disk
) {
2235 "%s: personality does not support diskops!\n",
2239 if (mddev
->persistent
)
2240 rdev
= md_import_device(dev
, mddev
->major_version
,
2241 mddev
->minor_version
);
2243 rdev
= md_import_device(dev
, -1, -1);
2246 "md: md_import_device returned %ld\n",
2248 return PTR_ERR(rdev
);
2250 /* set save_raid_disk if appropriate */
2251 if (!mddev
->persistent
) {
2252 if (info
->state
& (1<<MD_DISK_SYNC
) &&
2253 info
->raid_disk
< mddev
->raid_disks
)
2254 rdev
->raid_disk
= info
->raid_disk
;
2256 rdev
->raid_disk
= -1;
2258 super_types
[mddev
->major_version
].
2259 validate_super(mddev
, rdev
);
2260 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2262 rdev
->in_sync
= 0; /* just to be sure */
2263 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
2264 set_bit(WriteMostly
, &rdev
->flags
);
2266 rdev
->raid_disk
= -1;
2267 err
= bind_rdev_to_array(rdev
, mddev
);
2271 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2272 md_wakeup_thread(mddev
->thread
);
2276 /* otherwise, add_new_disk is only allowed
2277 * for major_version==0 superblocks
2279 if (mddev
->major_version
!= 0) {
2280 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
2285 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
2287 rdev
= md_import_device (dev
, -1, 0);
2290 "md: error, md_import_device() returned %ld\n",
2292 return PTR_ERR(rdev
);
2294 rdev
->desc_nr
= info
->number
;
2295 if (info
->raid_disk
< mddev
->raid_disks
)
2296 rdev
->raid_disk
= info
->raid_disk
;
2298 rdev
->raid_disk
= -1;
2301 if (rdev
->raid_disk
< mddev
->raid_disks
)
2302 rdev
->in_sync
= (info
->state
& (1<<MD_DISK_SYNC
));
2306 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
2307 set_bit(WriteMostly
, &rdev
->flags
);
2309 err
= bind_rdev_to_array(rdev
, mddev
);
2315 if (!mddev
->persistent
) {
2316 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
2317 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2319 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
2320 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
2322 if (!mddev
->size
|| (mddev
->size
> rdev
->size
))
2323 mddev
->size
= rdev
->size
;
2329 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
2331 char b
[BDEVNAME_SIZE
];
2337 rdev
= find_rdev(mddev
, dev
);
2341 if (rdev
->raid_disk
>= 0)
2344 kick_rdev_from_array(rdev
);
2345 md_update_sb(mddev
);
2349 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
2350 bdevname(rdev
->bdev
,b
), mdname(mddev
));
2354 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
2356 char b
[BDEVNAME_SIZE
];
2364 if (mddev
->major_version
!= 0) {
2365 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
2366 " version-0 superblocks.\n",
2370 if (!mddev
->pers
->hot_add_disk
) {
2372 "%s: personality does not support diskops!\n",
2377 rdev
= md_import_device (dev
, -1, 0);
2380 "md: error, md_import_device() returned %ld\n",
2385 if (mddev
->persistent
)
2386 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
2389 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2391 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
2394 if (size
< mddev
->size
) {
2396 "%s: disk size %llu blocks < array size %llu\n",
2397 mdname(mddev
), (unsigned long long)size
,
2398 (unsigned long long)mddev
->size
);
2405 "md: can not hot-add faulty %s disk to %s!\n",
2406 bdevname(rdev
->bdev
,b
), mdname(mddev
));
2412 bind_rdev_to_array(rdev
, mddev
);
2415 * The rest should better be atomic, we can have disk failures
2416 * noticed in interrupt contexts ...
2419 if (rdev
->desc_nr
== mddev
->max_disks
) {
2420 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
2423 goto abort_unbind_export
;
2426 rdev
->raid_disk
= -1;
2428 md_update_sb(mddev
);
2431 * Kick recovery, maybe this spare has to be added to the
2432 * array immediately.
2434 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2435 md_wakeup_thread(mddev
->thread
);
2439 abort_unbind_export
:
2440 unbind_rdev_from_array(rdev
);
2447 /* similar to deny_write_access, but accounts for our holding a reference
2448 * to the file ourselves */
2449 static int deny_bitmap_write_access(struct file
* file
)
2451 struct inode
*inode
= file
->f_mapping
->host
;
2453 spin_lock(&inode
->i_lock
);
2454 if (atomic_read(&inode
->i_writecount
) > 1) {
2455 spin_unlock(&inode
->i_lock
);
2458 atomic_set(&inode
->i_writecount
, -1);
2459 spin_unlock(&inode
->i_lock
);
2464 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
2469 if (!mddev
->pers
->quiesce
)
2471 if (mddev
->recovery
|| mddev
->sync_thread
)
2473 /* we should be able to change the bitmap.. */
2479 return -EEXIST
; /* cannot add when bitmap is present */
2480 mddev
->bitmap_file
= fget(fd
);
2482 if (mddev
->bitmap_file
== NULL
) {
2483 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
2488 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
2490 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
2492 fput(mddev
->bitmap_file
);
2493 mddev
->bitmap_file
= NULL
;
2496 mddev
->bitmap_offset
= 0; /* file overrides offset */
2497 } else if (mddev
->bitmap
== NULL
)
2498 return -ENOENT
; /* cannot remove what isn't there */
2501 mddev
->pers
->quiesce(mddev
, 1);
2503 err
= bitmap_create(mddev
);
2505 bitmap_destroy(mddev
);
2506 mddev
->pers
->quiesce(mddev
, 0);
2507 } else if (fd
< 0) {
2508 if (mddev
->bitmap_file
)
2509 fput(mddev
->bitmap_file
);
2510 mddev
->bitmap_file
= NULL
;
2517 * set_array_info is used two different ways
2518 * The original usage is when creating a new array.
2519 * In this usage, raid_disks is > 0 and it together with
2520 * level, size, not_persistent,layout,chunksize determine the
2521 * shape of the array.
2522 * This will always create an array with a type-0.90.0 superblock.
2523 * The newer usage is when assembling an array.
2524 * In this case raid_disks will be 0, and the major_version field is
2525 * use to determine which style super-blocks are to be found on the devices.
2526 * The minor and patch _version numbers are also kept incase the
2527 * super_block handler wishes to interpret them.
2529 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
2532 if (info
->raid_disks
== 0) {
2533 /* just setting version number for superblock loading */
2534 if (info
->major_version
< 0 ||
2535 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2536 super_types
[info
->major_version
].name
== NULL
) {
2537 /* maybe try to auto-load a module? */
2539 "md: superblock version %d not known\n",
2540 info
->major_version
);
2543 mddev
->major_version
= info
->major_version
;
2544 mddev
->minor_version
= info
->minor_version
;
2545 mddev
->patch_version
= info
->patch_version
;
2548 mddev
->major_version
= MD_MAJOR_VERSION
;
2549 mddev
->minor_version
= MD_MINOR_VERSION
;
2550 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
2551 mddev
->ctime
= get_seconds();
2553 mddev
->level
= info
->level
;
2554 mddev
->size
= info
->size
;
2555 mddev
->raid_disks
= info
->raid_disks
;
2556 /* don't set md_minor, it is determined by which /dev/md* was
2559 if (info
->state
& (1<<MD_SB_CLEAN
))
2560 mddev
->recovery_cp
= MaxSector
;
2562 mddev
->recovery_cp
= 0;
2563 mddev
->persistent
= ! info
->not_persistent
;
2565 mddev
->layout
= info
->layout
;
2566 mddev
->chunk_size
= info
->chunk_size
;
2568 mddev
->max_disks
= MD_SB_DISKS
;
2570 mddev
->sb_dirty
= 1;
2573 * Generate a 128 bit UUID
2575 get_random_bytes(mddev
->uuid
, 16);
2581 * update_array_info is used to change the configuration of an
2583 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2584 * fields in the info are checked against the array.
2585 * Any differences that cannot be handled will cause an error.
2586 * Normally, only one change can be managed at a time.
2588 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
2594 /* calculate expected state,ignoring low bits */
2595 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
2596 state
|= (1 << MD_SB_BITMAP_PRESENT
);
2598 if (mddev
->major_version
!= info
->major_version
||
2599 mddev
->minor_version
!= info
->minor_version
||
2600 /* mddev->patch_version != info->patch_version || */
2601 mddev
->ctime
!= info
->ctime
||
2602 mddev
->level
!= info
->level
||
2603 /* mddev->layout != info->layout || */
2604 !mddev
->persistent
!= info
->not_persistent
||
2605 mddev
->chunk_size
!= info
->chunk_size
||
2606 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2607 ((state
^info
->state
) & 0xfffffe00)
2610 /* Check there is only one change */
2611 if (mddev
->size
!= info
->size
) cnt
++;
2612 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
2613 if (mddev
->layout
!= info
->layout
) cnt
++;
2614 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
2615 if (cnt
== 0) return 0;
2616 if (cnt
> 1) return -EINVAL
;
2618 if (mddev
->layout
!= info
->layout
) {
2620 * we don't need to do anything at the md level, the
2621 * personality will take care of it all.
2623 if (mddev
->pers
->reconfig
== NULL
)
2626 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
2628 if (mddev
->size
!= info
->size
) {
2630 struct list_head
*tmp
;
2631 if (mddev
->pers
->resize
== NULL
)
2633 /* The "size" is the amount of each device that is used.
2634 * This can only make sense for arrays with redundancy.
2635 * linear and raid0 always use whatever space is available
2636 * We can only consider changing the size if no resync
2637 * or reconstruction is happening, and if the new size
2638 * is acceptable. It must fit before the sb_offset or,
2639 * if that is <data_offset, it must fit before the
2640 * size of each device.
2641 * If size is zero, we find the largest size that fits.
2643 if (mddev
->sync_thread
)
2645 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2647 int fit
= (info
->size
== 0);
2648 if (rdev
->sb_offset
> rdev
->data_offset
)
2649 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
2651 avail
= get_capacity(rdev
->bdev
->bd_disk
)
2652 - rdev
->data_offset
;
2653 if (fit
&& (info
->size
== 0 || info
->size
> avail
/2))
2654 info
->size
= avail
/2;
2655 if (avail
< ((sector_t
)info
->size
<< 1))
2658 rv
= mddev
->pers
->resize(mddev
, (sector_t
)info
->size
*2);
2660 struct block_device
*bdev
;
2662 bdev
= bdget_disk(mddev
->gendisk
, 0);
2664 down(&bdev
->bd_inode
->i_sem
);
2665 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
2666 up(&bdev
->bd_inode
->i_sem
);
2671 if (mddev
->raid_disks
!= info
->raid_disks
) {
2672 /* change the number of raid disks */
2673 if (mddev
->pers
->reshape
== NULL
)
2675 if (info
->raid_disks
<= 0 ||
2676 info
->raid_disks
>= mddev
->max_disks
)
2678 if (mddev
->sync_thread
)
2680 rv
= mddev
->pers
->reshape(mddev
, info
->raid_disks
);
2682 struct block_device
*bdev
;
2684 bdev
= bdget_disk(mddev
->gendisk
, 0);
2686 down(&bdev
->bd_inode
->i_sem
);
2687 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
2688 up(&bdev
->bd_inode
->i_sem
);
2693 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
2694 if (mddev
->pers
->quiesce
== NULL
)
2696 if (mddev
->recovery
|| mddev
->sync_thread
)
2698 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
2699 /* add the bitmap */
2702 if (mddev
->default_bitmap_offset
== 0)
2704 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
2705 mddev
->pers
->quiesce(mddev
, 1);
2706 rv
= bitmap_create(mddev
);
2708 bitmap_destroy(mddev
);
2709 mddev
->pers
->quiesce(mddev
, 0);
2711 /* remove the bitmap */
2714 if (mddev
->bitmap
->file
)
2716 mddev
->pers
->quiesce(mddev
, 1);
2717 bitmap_destroy(mddev
);
2718 mddev
->pers
->quiesce(mddev
, 0);
2719 mddev
->bitmap_offset
= 0;
2722 md_update_sb(mddev
);
2726 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
2730 if (mddev
->pers
== NULL
)
2733 rdev
= find_rdev(mddev
, dev
);
2737 md_error(mddev
, rdev
);
2741 static int md_ioctl(struct inode
*inode
, struct file
*file
,
2742 unsigned int cmd
, unsigned long arg
)
2745 void __user
*argp
= (void __user
*)arg
;
2746 struct hd_geometry __user
*loc
= argp
;
2747 mddev_t
*mddev
= NULL
;
2749 if (!capable(CAP_SYS_ADMIN
))
2753 * Commands dealing with the RAID driver but not any
2759 err
= get_version(argp
);
2762 case PRINT_RAID_DEBUG
:
2770 autostart_arrays(arg
);
2777 * Commands creating/starting a new array:
2780 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
2788 if (cmd
== START_ARRAY
) {
2789 /* START_ARRAY doesn't need to lock the array as autostart_array
2790 * does the locking, and it could even be a different array
2795 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2796 "This will not be supported beyond 2.6\n",
2797 current
->comm
, current
->pid
);
2800 err
= autostart_array(new_decode_dev(arg
));
2802 printk(KERN_WARNING
"md: autostart failed!\n");
2808 err
= mddev_lock(mddev
);
2811 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2818 case SET_ARRAY_INFO
:
2820 mdu_array_info_t info
;
2822 memset(&info
, 0, sizeof(info
));
2823 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
2828 err
= update_array_info(mddev
, &info
);
2830 printk(KERN_WARNING
"md: couldn't update"
2831 " array info. %d\n", err
);
2836 if (!list_empty(&mddev
->disks
)) {
2838 "md: array %s already has disks!\n",
2843 if (mddev
->raid_disks
) {
2845 "md: array %s already initialised!\n",
2850 err
= set_array_info(mddev
, &info
);
2852 printk(KERN_WARNING
"md: couldn't set"
2853 " array info. %d\n", err
);
2863 * Commands querying/configuring an existing array:
2865 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2866 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2867 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
2868 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
2874 * Commands even a read-only array can execute:
2878 case GET_ARRAY_INFO
:
2879 err
= get_array_info(mddev
, argp
);
2882 case GET_BITMAP_FILE
:
2883 err
= get_bitmap_file(mddev
, argp
);
2887 err
= get_disk_info(mddev
, argp
);
2890 case RESTART_ARRAY_RW
:
2891 err
= restart_array(mddev
);
2895 err
= do_md_stop (mddev
, 0);
2899 err
= do_md_stop (mddev
, 1);
2903 * We have a problem here : there is no easy way to give a CHS
2904 * virtual geometry. We currently pretend that we have a 2 heads
2905 * 4 sectors (with a BIG number of cylinders...). This drives
2906 * dosfs just mad... ;-)
2913 err
= put_user (2, (char __user
*) &loc
->heads
);
2916 err
= put_user (4, (char __user
*) &loc
->sectors
);
2919 err
= put_user(get_capacity(mddev
->gendisk
)/8,
2920 (short __user
*) &loc
->cylinders
);
2923 err
= put_user (get_start_sect(inode
->i_bdev
),
2924 (long __user
*) &loc
->start
);
2929 * The remaining ioctls are changing the state of the
2930 * superblock, so we do not allow read-only arrays
2942 mdu_disk_info_t info
;
2943 if (copy_from_user(&info
, argp
, sizeof(info
)))
2946 err
= add_new_disk(mddev
, &info
);
2950 case HOT_REMOVE_DISK
:
2951 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
2955 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
2958 case SET_DISK_FAULTY
:
2959 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
2963 err
= do_md_run (mddev
);
2966 case SET_BITMAP_FILE
:
2967 err
= set_bitmap_file(mddev
, (int)arg
);
2971 if (_IOC_TYPE(cmd
) == MD_MAJOR
)
2972 printk(KERN_WARNING
"md: %s(pid %d) used"
2973 " obsolete MD ioctl, upgrade your"
2974 " software to use new ictls.\n",
2975 current
->comm
, current
->pid
);
2982 mddev_unlock(mddev
);
2992 static int md_open(struct inode
*inode
, struct file
*file
)
2995 * Succeed if we can lock the mddev, which confirms that
2996 * it isn't being stopped right now.
2998 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3001 if ((err
= mddev_lock(mddev
)))
3006 mddev_unlock(mddev
);
3008 check_disk_change(inode
->i_bdev
);
3013 static int md_release(struct inode
*inode
, struct file
* file
)
3015 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3024 static int md_media_changed(struct gendisk
*disk
)
3026 mddev_t
*mddev
= disk
->private_data
;
3028 return mddev
->changed
;
3031 static int md_revalidate(struct gendisk
*disk
)
3033 mddev_t
*mddev
= disk
->private_data
;
3038 static struct block_device_operations md_fops
=
3040 .owner
= THIS_MODULE
,
3042 .release
= md_release
,
3044 .media_changed
= md_media_changed
,
3045 .revalidate_disk
= md_revalidate
,
3048 static int md_thread(void * arg
)
3050 mdk_thread_t
*thread
= arg
;
3058 daemonize(thread
->name
, mdname(thread
->mddev
));
3060 current
->exit_signal
= SIGCHLD
;
3061 allow_signal(SIGKILL
);
3062 thread
->tsk
= current
;
3065 * md_thread is a 'system-thread', it's priority should be very
3066 * high. We avoid resource deadlocks individually in each
3067 * raid personality. (RAID5 does preallocation) We also use RR and
3068 * the very same RT priority as kswapd, thus we will never get
3069 * into a priority inversion deadlock.
3071 * we definitely have to have equal or higher priority than
3072 * bdflush, otherwise bdflush will deadlock if there are too
3073 * many dirty RAID5 blocks.
3077 complete(thread
->event
);
3078 while (thread
->run
) {
3079 void (*run
)(mddev_t
*);
3081 wait_event_interruptible_timeout(thread
->wqueue
,
3082 test_bit(THREAD_WAKEUP
, &thread
->flags
),
3086 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
3092 if (signal_pending(current
))
3093 flush_signals(current
);
3095 complete(thread
->event
);
3099 void md_wakeup_thread(mdk_thread_t
*thread
)
3102 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
3103 set_bit(THREAD_WAKEUP
, &thread
->flags
);
3104 wake_up(&thread
->wqueue
);
3108 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
3111 mdk_thread_t
*thread
;
3113 struct completion event
;
3115 thread
= (mdk_thread_t
*) kmalloc
3116 (sizeof(mdk_thread_t
), GFP_KERNEL
);
3120 memset(thread
, 0, sizeof(mdk_thread_t
));
3121 init_waitqueue_head(&thread
->wqueue
);
3123 init_completion(&event
);
3124 thread
->event
= &event
;
3126 thread
->mddev
= mddev
;
3127 thread
->name
= name
;
3128 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
3129 ret
= kernel_thread(md_thread
, thread
, 0);
3134 wait_for_completion(&event
);
3138 void md_unregister_thread(mdk_thread_t
*thread
)
3140 struct completion event
;
3142 init_completion(&event
);
3144 thread
->event
= &event
;
3146 /* As soon as ->run is set to NULL, the task could disappear,
3147 * so we need to hold tasklist_lock until we have sent the signal
3149 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
3150 read_lock(&tasklist_lock
);
3152 send_sig(SIGKILL
, thread
->tsk
, 1);
3153 read_unlock(&tasklist_lock
);
3154 wait_for_completion(&event
);
3158 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
3165 if (!rdev
|| rdev
->faulty
)
3168 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3170 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3171 __builtin_return_address(0),__builtin_return_address(1),
3172 __builtin_return_address(2),__builtin_return_address(3));
3174 if (!mddev
->pers
->error_handler
)
3176 mddev
->pers
->error_handler(mddev
,rdev
);
3177 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3178 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3179 md_wakeup_thread(mddev
->thread
);
3182 /* seq_file implementation /proc/mdstat */
3184 static void status_unused(struct seq_file
*seq
)
3188 struct list_head
*tmp
;
3190 seq_printf(seq
, "unused devices: ");
3192 ITERATE_RDEV_PENDING(rdev
,tmp
) {
3193 char b
[BDEVNAME_SIZE
];
3195 seq_printf(seq
, "%s ",
3196 bdevname(rdev
->bdev
,b
));
3199 seq_printf(seq
, "<none>");
3201 seq_printf(seq
, "\n");
3205 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
3207 unsigned long max_blocks
, resync
, res
, dt
, db
, rt
;
3209 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
3211 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3212 max_blocks
= mddev
->resync_max_sectors
>> 1;
3214 max_blocks
= mddev
->size
;
3217 * Should not happen.
3223 res
= (resync
/1024)*1000/(max_blocks
/1024 + 1);
3225 int i
, x
= res
/50, y
= 20-x
;
3226 seq_printf(seq
, "[");
3227 for (i
= 0; i
< x
; i
++)
3228 seq_printf(seq
, "=");
3229 seq_printf(seq
, ">");
3230 for (i
= 0; i
< y
; i
++)
3231 seq_printf(seq
, ".");
3232 seq_printf(seq
, "] ");
3234 seq_printf(seq
, " %s =%3lu.%lu%% (%lu/%lu)",
3235 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
3236 "resync" : "recovery"),
3237 res
/10, res
% 10, resync
, max_blocks
);
3240 * We do not want to overflow, so the order of operands and
3241 * the * 100 / 100 trick are important. We do a +1 to be
3242 * safe against division by zero. We only estimate anyway.
3244 * dt: time from mark until now
3245 * db: blocks written from mark until now
3246 * rt: remaining time
3248 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
3250 db
= resync
- (mddev
->resync_mark_cnt
/2);
3251 rt
= (dt
* ((max_blocks
-resync
) / (db
/100+1)))/100;
3253 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
3255 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
3258 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3260 struct list_head
*tmp
;
3270 spin_lock(&all_mddevs_lock
);
3271 list_for_each(tmp
,&all_mddevs
)
3273 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
3275 spin_unlock(&all_mddevs_lock
);
3278 spin_unlock(&all_mddevs_lock
);
3280 return (void*)2;/* tail */
3284 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3286 struct list_head
*tmp
;
3287 mddev_t
*next_mddev
, *mddev
= v
;
3293 spin_lock(&all_mddevs_lock
);
3295 tmp
= all_mddevs
.next
;
3297 tmp
= mddev
->all_mddevs
.next
;
3298 if (tmp
!= &all_mddevs
)
3299 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
3301 next_mddev
= (void*)2;
3304 spin_unlock(&all_mddevs_lock
);
3312 static void md_seq_stop(struct seq_file
*seq
, void *v
)
3316 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
3320 static int md_seq_show(struct seq_file
*seq
, void *v
)
3324 struct list_head
*tmp2
;
3327 struct bitmap
*bitmap
;
3329 if (v
== (void*)1) {
3330 seq_printf(seq
, "Personalities : ");
3331 spin_lock(&pers_lock
);
3332 for (i
= 0; i
< MAX_PERSONALITY
; i
++)
3334 seq_printf(seq
, "[%s] ", pers
[i
]->name
);
3336 spin_unlock(&pers_lock
);
3337 seq_printf(seq
, "\n");
3340 if (v
== (void*)2) {
3345 if (mddev_lock(mddev
)!=0)
3347 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
3348 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
3349 mddev
->pers
? "" : "in");
3352 seq_printf(seq
, " (read-only)");
3353 seq_printf(seq
, " %s", mddev
->pers
->name
);
3357 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
3358 char b
[BDEVNAME_SIZE
];
3359 seq_printf(seq
, " %s[%d]",
3360 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
3361 if (test_bit(WriteMostly
, &rdev
->flags
))
3362 seq_printf(seq
, "(W)");
3364 seq_printf(seq
, "(F)");
3370 if (!list_empty(&mddev
->disks
)) {
3372 seq_printf(seq
, "\n %llu blocks",
3373 (unsigned long long)mddev
->array_size
);
3375 seq_printf(seq
, "\n %llu blocks",
3376 (unsigned long long)size
);
3380 mddev
->pers
->status (seq
, mddev
);
3381 seq_printf(seq
, "\n ");
3382 if (mddev
->curr_resync
> 2) {
3383 status_resync (seq
, mddev
);
3384 seq_printf(seq
, "\n ");
3385 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
3386 seq_printf(seq
, " resync=DELAYED\n ");
3388 seq_printf(seq
, "\n ");
3390 if ((bitmap
= mddev
->bitmap
)) {
3391 unsigned long chunk_kb
;
3392 unsigned long flags
;
3393 spin_lock_irqsave(&bitmap
->lock
, flags
);
3394 chunk_kb
= bitmap
->chunksize
>> 10;
3395 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
3397 bitmap
->pages
- bitmap
->missing_pages
,
3399 (bitmap
->pages
- bitmap
->missing_pages
)
3400 << (PAGE_SHIFT
- 10),
3401 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
3402 chunk_kb
? "KB" : "B");
3404 seq_printf(seq
, ", file: ");
3405 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
3406 bitmap
->file
->f_dentry
," \t\n");
3409 seq_printf(seq
, "\n");
3410 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
3413 seq_printf(seq
, "\n");
3415 mddev_unlock(mddev
);
3420 static struct seq_operations md_seq_ops
= {
3421 .start
= md_seq_start
,
3422 .next
= md_seq_next
,
3423 .stop
= md_seq_stop
,
3424 .show
= md_seq_show
,
3427 static int md_seq_open(struct inode
*inode
, struct file
*file
)
3431 error
= seq_open(file
, &md_seq_ops
);
3435 static struct file_operations md_seq_fops
= {
3436 .open
= md_seq_open
,
3438 .llseek
= seq_lseek
,
3439 .release
= seq_release
,
3442 int register_md_personality(int pnum
, mdk_personality_t
*p
)
3444 if (pnum
>= MAX_PERSONALITY
) {
3446 "md: tried to install personality %s as nr %d, but max is %lu\n",
3447 p
->name
, pnum
, MAX_PERSONALITY
-1);
3451 spin_lock(&pers_lock
);
3453 spin_unlock(&pers_lock
);
3458 printk(KERN_INFO
"md: %s personality registered as nr %d\n", p
->name
, pnum
);
3459 spin_unlock(&pers_lock
);
3463 int unregister_md_personality(int pnum
)
3465 if (pnum
>= MAX_PERSONALITY
)
3468 printk(KERN_INFO
"md: %s personality unregistered\n", pers
[pnum
]->name
);
3469 spin_lock(&pers_lock
);
3471 spin_unlock(&pers_lock
);
3475 static int is_mddev_idle(mddev_t
*mddev
)
3478 struct list_head
*tmp
;
3480 unsigned long curr_events
;
3483 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3484 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
3485 curr_events
= disk_stat_read(disk
, read_sectors
) +
3486 disk_stat_read(disk
, write_sectors
) -
3487 atomic_read(&disk
->sync_io
);
3488 /* Allow some slack between valud of curr_events and last_events,
3489 * as there are some uninteresting races.
3490 * Note: the following is an unsigned comparison.
3492 if ((curr_events
- rdev
->last_events
+ 32) > 64) {
3493 rdev
->last_events
= curr_events
;
3500 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
3502 /* another "blocks" (512byte) blocks have been synced */
3503 atomic_sub(blocks
, &mddev
->recovery_active
);
3504 wake_up(&mddev
->recovery_wait
);
3506 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
3507 md_wakeup_thread(mddev
->thread
);
3508 // stop recovery, signal do_sync ....
3513 /* md_write_start(mddev, bi)
3514 * If we need to update some array metadata (e.g. 'active' flag
3515 * in superblock) before writing, schedule a superblock update
3516 * and wait for it to complete.
3518 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
3520 if (bio_data_dir(bi
) != WRITE
)
3523 atomic_inc(&mddev
->writes_pending
);
3524 if (mddev
->in_sync
) {
3525 spin_lock(&mddev
->write_lock
);
3526 if (mddev
->in_sync
) {
3528 mddev
->sb_dirty
= 1;
3529 md_wakeup_thread(mddev
->thread
);
3531 spin_unlock(&mddev
->write_lock
);
3533 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
3536 void md_write_end(mddev_t
*mddev
)
3538 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
3539 if (mddev
->safemode
== 2)
3540 md_wakeup_thread(mddev
->thread
);
3542 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
3546 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
3548 #define SYNC_MARKS 10
3549 #define SYNC_MARK_STEP (3*HZ)
3550 static void md_do_sync(mddev_t
*mddev
)
3553 unsigned int currspeed
= 0,
3555 sector_t max_sectors
,j
, io_sectors
;
3556 unsigned long mark
[SYNC_MARKS
];
3557 sector_t mark_cnt
[SYNC_MARKS
];
3559 struct list_head
*tmp
;
3560 sector_t last_check
;
3563 /* just incase thread restarts... */
3564 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
3567 /* we overload curr_resync somewhat here.
3568 * 0 == not engaged in resync at all
3569 * 2 == checking that there is no conflict with another sync
3570 * 1 == like 2, but have yielded to allow conflicting resync to
3572 * other == active in resync - this many blocks
3574 * Before starting a resync we must have set curr_resync to
3575 * 2, and then checked that every "conflicting" array has curr_resync
3576 * less than ours. When we find one that is the same or higher
3577 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3578 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3579 * This will mean we have to start checking from the beginning again.
3584 mddev
->curr_resync
= 2;
3587 if (signal_pending(current
)) {
3588 flush_signals(current
);
3591 ITERATE_MDDEV(mddev2
,tmp
) {
3592 if (mddev2
== mddev
)
3594 if (mddev2
->curr_resync
&&
3595 match_mddev_units(mddev
,mddev2
)) {
3597 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
3598 /* arbitrarily yield */
3599 mddev
->curr_resync
= 1;
3600 wake_up(&resync_wait
);
3602 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
3603 /* no need to wait here, we can wait the next
3604 * time 'round when curr_resync == 2
3607 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
3608 if (!signal_pending(current
)
3609 && mddev2
->curr_resync
>= mddev
->curr_resync
) {
3610 printk(KERN_INFO
"md: delaying resync of %s"
3611 " until %s has finished resync (they"
3612 " share one or more physical units)\n",
3613 mdname(mddev
), mdname(mddev2
));
3616 finish_wait(&resync_wait
, &wq
);
3619 finish_wait(&resync_wait
, &wq
);
3622 } while (mddev
->curr_resync
< 2);
3624 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3625 /* resync follows the size requested by the personality,
3626 * which defaults to physical size, but can be virtual size
3628 max_sectors
= mddev
->resync_max_sectors
;
3630 /* recovery follows the physical size of devices */
3631 max_sectors
= mddev
->size
<< 1;
3633 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
3634 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
3635 " %d KB/sec/disc.\n", sysctl_speed_limit_min
);
3636 printk(KERN_INFO
"md: using maximum available idle IO bandwith "
3637 "(but not more than %d KB/sec) for reconstruction.\n",
3638 sysctl_speed_limit_max
);
3640 is_mddev_idle(mddev
); /* this also initializes IO event counters */
3641 /* we don't use the checkpoint if there's a bitmap */
3642 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && !mddev
->bitmap
)
3643 j
= mddev
->recovery_cp
;
3647 for (m
= 0; m
< SYNC_MARKS
; m
++) {
3649 mark_cnt
[m
] = io_sectors
;
3652 mddev
->resync_mark
= mark
[last_mark
];
3653 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
3656 * Tune reconstruction:
3658 window
= 32*(PAGE_SIZE
/512);
3659 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
3660 window
/2,(unsigned long long) max_sectors
/2);
3662 atomic_set(&mddev
->recovery_active
, 0);
3663 init_waitqueue_head(&mddev
->recovery_wait
);
3668 "md: resuming recovery of %s from checkpoint.\n",
3670 mddev
->curr_resync
= j
;
3673 while (j
< max_sectors
) {
3677 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
3678 currspeed
< sysctl_speed_limit_min
);
3680 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
3684 if (!skipped
) { /* actual IO requested */
3685 io_sectors
+= sectors
;
3686 atomic_add(sectors
, &mddev
->recovery_active
);
3690 if (j
>1) mddev
->curr_resync
= j
;
3693 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
3696 last_check
= io_sectors
;
3698 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
3699 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
3703 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
3705 int next
= (last_mark
+1) % SYNC_MARKS
;
3707 mddev
->resync_mark
= mark
[next
];
3708 mddev
->resync_mark_cnt
= mark_cnt
[next
];
3709 mark
[next
] = jiffies
;
3710 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
3715 if (signal_pending(current
)) {
3717 * got a signal, exit.
3720 "md: md_do_sync() got signal ... exiting\n");
3721 flush_signals(current
);
3722 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3727 * this loop exits only if either when we are slower than
3728 * the 'hard' speed limit, or the system was IO-idle for
3730 * the system might be non-idle CPU-wise, but we only care
3731 * about not overloading the IO subsystem. (things like an
3732 * e2fsck being done on the RAID array should execute fast)
3734 mddev
->queue
->unplug_fn(mddev
->queue
);
3737 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
3738 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
3740 if (currspeed
> sysctl_speed_limit_min
) {
3741 if ((currspeed
> sysctl_speed_limit_max
) ||
3742 !is_mddev_idle(mddev
)) {
3743 msleep_interruptible(250);
3748 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
3750 * this also signals 'finished resyncing' to md_stop
3753 mddev
->queue
->unplug_fn(mddev
->queue
);
3755 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
3757 /* tell personality that we are finished */
3758 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
3760 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
3761 mddev
->curr_resync
> 2 &&
3762 mddev
->curr_resync
>= mddev
->recovery_cp
) {
3763 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
3765 "md: checkpointing recovery of %s.\n",
3767 mddev
->recovery_cp
= mddev
->curr_resync
;
3769 mddev
->recovery_cp
= MaxSector
;
3773 mddev
->curr_resync
= 0;
3774 wake_up(&resync_wait
);
3775 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
3776 md_wakeup_thread(mddev
->thread
);
3781 * This routine is regularly called by all per-raid-array threads to
3782 * deal with generic issues like resync and super-block update.
3783 * Raid personalities that don't have a thread (linear/raid0) do not
3784 * need this as they never do any recovery or update the superblock.
3786 * It does not do any resync itself, but rather "forks" off other threads
3787 * to do that as needed.
3788 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3789 * "->recovery" and create a thread at ->sync_thread.
3790 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3791 * and wakeups up this thread which will reap the thread and finish up.
3792 * This thread also removes any faulty devices (with nr_pending == 0).
3794 * The overall approach is:
3795 * 1/ if the superblock needs updating, update it.
3796 * 2/ If a recovery thread is running, don't do anything else.
3797 * 3/ If recovery has finished, clean up, possibly marking spares active.
3798 * 4/ If there are any faulty devices, remove them.
3799 * 5/ If array is degraded, try to add spares devices
3800 * 6/ If array has spares or is not in-sync, start a resync thread.
3802 void md_check_recovery(mddev_t
*mddev
)
3805 struct list_head
*rtmp
;
3809 bitmap_daemon_work(mddev
->bitmap
);
3814 if (signal_pending(current
)) {
3815 if (mddev
->pers
->sync_request
) {
3816 printk(KERN_INFO
"md: %s in immediate safe mode\n",
3818 mddev
->safemode
= 2;
3820 flush_signals(current
);
3825 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
3826 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
3827 (mddev
->safemode
== 1) ||
3828 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
3829 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
3833 if (mddev_trylock(mddev
)==0) {
3836 spin_lock(&mddev
->write_lock
);
3837 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
3838 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
3840 mddev
->sb_dirty
= 1;
3842 if (mddev
->safemode
== 1)
3843 mddev
->safemode
= 0;
3844 spin_unlock(&mddev
->write_lock
);
3846 if (mddev
->sb_dirty
)
3847 md_update_sb(mddev
);
3850 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
3851 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
3852 /* resync/recovery still happening */
3853 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3856 if (mddev
->sync_thread
) {
3857 /* resync has finished, collect result */
3858 md_unregister_thread(mddev
->sync_thread
);
3859 mddev
->sync_thread
= NULL
;
3860 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
3861 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
3863 /* activate any spares */
3864 mddev
->pers
->spare_active(mddev
);
3866 md_update_sb(mddev
);
3868 /* if array is no-longer degraded, then any saved_raid_disk
3869 * information must be scrapped
3871 if (!mddev
->degraded
)
3872 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3873 rdev
->saved_raid_disk
= -1;
3875 mddev
->recovery
= 0;
3876 /* flag recovery needed just to double check */
3877 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3880 if (mddev
->recovery
)
3881 /* probably just the RECOVERY_NEEDED flag */
3882 mddev
->recovery
= 0;
3884 /* no recovery is running.
3885 * remove any failed drives, then
3886 * add spares if possible.
3887 * Spare are also removed and re-added, to allow
3888 * the personality to fail the re-add.
3890 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3891 if (rdev
->raid_disk
>= 0 &&
3892 (rdev
->faulty
|| ! rdev
->in_sync
) &&
3893 atomic_read(&rdev
->nr_pending
)==0) {
3894 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0)
3895 rdev
->raid_disk
= -1;
3898 if (mddev
->degraded
) {
3899 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3900 if (rdev
->raid_disk
< 0
3902 if (mddev
->pers
->hot_add_disk(mddev
,rdev
))
3909 if (!spares
&& (mddev
->recovery_cp
== MaxSector
)) {
3910 /* nothing we can do ... */
3913 if (mddev
->pers
->sync_request
) {
3914 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3916 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3917 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
3918 /* We are adding a device or devices to an array
3919 * which has the bitmap stored on all devices.
3920 * So make sure all bitmap pages get written
3922 bitmap_write_all(mddev
->bitmap
);
3924 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3927 if (!mddev
->sync_thread
) {
3928 printk(KERN_ERR
"%s: could not start resync"
3931 /* leave the spares where they are, it shouldn't hurt */
3932 mddev
->recovery
= 0;
3934 md_wakeup_thread(mddev
->sync_thread
);
3938 mddev_unlock(mddev
);
3942 static int md_notify_reboot(struct notifier_block
*this,
3943 unsigned long code
, void *x
)
3945 struct list_head
*tmp
;
3948 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
3950 printk(KERN_INFO
"md: stopping all md devices.\n");
3952 ITERATE_MDDEV(mddev
,tmp
)
3953 if (mddev_trylock(mddev
)==0)
3954 do_md_stop (mddev
, 1);
3956 * certain more exotic SCSI devices are known to be
3957 * volatile wrt too early system reboots. While the
3958 * right place to handle this issue is the given
3959 * driver, we do want to have a safe RAID driver ...
3966 static struct notifier_block md_notifier
= {
3967 .notifier_call
= md_notify_reboot
,
3969 .priority
= INT_MAX
, /* before any real devices */
3972 static void md_geninit(void)
3974 struct proc_dir_entry
*p
;
3976 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
3978 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
3980 p
->proc_fops
= &md_seq_fops
;
3983 static int __init
md_init(void)
3987 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3988 " MD_SB_DISKS=%d\n",
3989 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
3990 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
3991 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR
,
3994 if (register_blkdev(MAJOR_NR
, "md"))
3996 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
3997 unregister_blkdev(MAJOR_NR
, "md");
4001 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
4002 md_probe
, NULL
, NULL
);
4003 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
4004 md_probe
, NULL
, NULL
);
4006 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4007 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
4008 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4011 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4012 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
4013 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4017 register_reboot_notifier(&md_notifier
);
4018 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
4028 * Searches all registered partitions for autorun RAID arrays
4031 static dev_t detected_devices
[128];
4034 void md_autodetect_dev(dev_t dev
)
4036 if (dev_cnt
>= 0 && dev_cnt
< 127)
4037 detected_devices
[dev_cnt
++] = dev
;
4041 static void autostart_arrays(int part
)
4046 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
4048 for (i
= 0; i
< dev_cnt
; i
++) {
4049 dev_t dev
= detected_devices
[i
];
4051 rdev
= md_import_device(dev
,0, 0);
4059 list_add(&rdev
->same_set
, &pending_raid_disks
);
4063 autorun_devices(part
);
4068 static __exit
void md_exit(void)
4071 struct list_head
*tmp
;
4073 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
4074 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
4075 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4076 devfs_remove("md/%d", i
);
4077 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4078 devfs_remove("md/d%d", i
);
4082 unregister_blkdev(MAJOR_NR
,"md");
4083 unregister_blkdev(mdp_major
, "mdp");
4084 unregister_reboot_notifier(&md_notifier
);
4085 unregister_sysctl_table(raid_table_header
);
4086 remove_proc_entry("mdstat", NULL
);
4087 ITERATE_MDDEV(mddev
,tmp
) {
4088 struct gendisk
*disk
= mddev
->gendisk
;
4091 export_array(mddev
);
4094 mddev
->gendisk
= NULL
;
4099 module_init(md_init
)
4100 module_exit(md_exit
)
4102 EXPORT_SYMBOL(register_md_personality
);
4103 EXPORT_SYMBOL(unregister_md_personality
);
4104 EXPORT_SYMBOL(md_error
);
4105 EXPORT_SYMBOL(md_done_sync
);
4106 EXPORT_SYMBOL(md_write_start
);
4107 EXPORT_SYMBOL(md_write_end
);
4108 EXPORT_SYMBOL(md_register_thread
);
4109 EXPORT_SYMBOL(md_unregister_thread
);
4110 EXPORT_SYMBOL(md_wakeup_thread
);
4111 EXPORT_SYMBOL(md_print_devices
);
4112 EXPORT_SYMBOL(md_check_recovery
);
4113 MODULE_LICENSE("GPL");
4115 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);