2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part
);
71 static LIST_HEAD(pers_list
);
72 static DEFINE_SPINLOCK(pers_lock
);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 * or /sys/block/mdX/md/sync_speed_{min,max}
87 static int sysctl_speed_limit_min
= 1000;
88 static int sysctl_speed_limit_max
= 200000;
89 static inline int speed_min(mddev_t
*mddev
)
91 return mddev
->sync_speed_min
?
92 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
95 static inline int speed_max(mddev_t
*mddev
)
97 return mddev
->sync_speed_max
?
98 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
101 static struct ctl_table_header
*raid_table_header
;
103 static ctl_table raid_table
[] = {
105 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
106 .procname
= "speed_limit_min",
107 .data
= &sysctl_speed_limit_min
,
108 .maxlen
= sizeof(int),
110 .proc_handler
= &proc_dointvec
,
113 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
114 .procname
= "speed_limit_max",
115 .data
= &sysctl_speed_limit_max
,
116 .maxlen
= sizeof(int),
118 .proc_handler
= &proc_dointvec
,
123 static ctl_table raid_dir_table
[] = {
125 .ctl_name
= DEV_RAID
,
134 static ctl_table raid_root_table
[] = {
140 .child
= raid_dir_table
,
145 static struct block_device_operations md_fops
;
147 static int start_readonly
;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
160 static atomic_t md_event_count
;
161 static void md_new_event(mddev_t
*mddev
)
163 atomic_inc(&md_event_count
);
164 wake_up(&md_event_waiters
);
168 * Enables to iterate over all existing md arrays
169 * all_mddevs_lock protects this list.
171 static LIST_HEAD(all_mddevs
);
172 static DEFINE_SPINLOCK(all_mddevs_lock
);
176 * iterates through all used mddevs in the system.
177 * We take care to grab the all_mddevs_lock whenever navigating
178 * the list, and to always hold a refcount when unlocked.
179 * Any code which breaks out of this loop while own
180 * a reference to the current mddev and must mddev_put it.
182 #define ITERATE_MDDEV(mddev,tmp) \
184 for (({ spin_lock(&all_mddevs_lock); \
185 tmp = all_mddevs.next; \
187 ({ if (tmp != &all_mddevs) \
188 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
189 spin_unlock(&all_mddevs_lock); \
190 if (mddev) mddev_put(mddev); \
191 mddev = list_entry(tmp, mddev_t, all_mddevs); \
192 tmp != &all_mddevs;}); \
193 ({ spin_lock(&all_mddevs_lock); \
198 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
200 bio_io_error(bio
, bio
->bi_size
);
204 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
206 atomic_inc(&mddev
->active
);
210 static void mddev_put(mddev_t
*mddev
)
212 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
214 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
215 list_del(&mddev
->all_mddevs
);
217 blk_cleanup_queue(mddev
->queue
);
218 /* that also blocks */
219 kobject_unregister(&mddev
->kobj
);
220 /* result blows... */
222 spin_unlock(&all_mddevs_lock
);
225 static mddev_t
* mddev_find(dev_t unit
)
227 mddev_t
*mddev
, *new = NULL
;
230 spin_lock(&all_mddevs_lock
);
231 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
232 if (mddev
->unit
== unit
) {
234 spin_unlock(&all_mddevs_lock
);
240 list_add(&new->all_mddevs
, &all_mddevs
);
241 spin_unlock(&all_mddevs_lock
);
244 spin_unlock(&all_mddevs_lock
);
246 new = kzalloc(sizeof(*new), GFP_KERNEL
);
251 if (MAJOR(unit
) == MD_MAJOR
)
252 new->md_minor
= MINOR(unit
);
254 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
256 init_MUTEX(&new->reconfig_sem
);
257 INIT_LIST_HEAD(&new->disks
);
258 INIT_LIST_HEAD(&new->all_mddevs
);
259 init_timer(&new->safemode_timer
);
260 atomic_set(&new->active
, 1);
261 spin_lock_init(&new->write_lock
);
262 init_waitqueue_head(&new->sb_wait
);
264 new->queue
= blk_alloc_queue(GFP_KERNEL
);
269 set_bit(QUEUE_FLAG_CLUSTER
, &new->queue
->queue_flags
);
271 blk_queue_make_request(new->queue
, md_fail_request
);
276 static inline int mddev_lock(mddev_t
* mddev
)
278 return down_interruptible(&mddev
->reconfig_sem
);
281 static inline void mddev_lock_uninterruptible(mddev_t
* mddev
)
283 down(&mddev
->reconfig_sem
);
286 static inline int mddev_trylock(mddev_t
* mddev
)
288 return down_trylock(&mddev
->reconfig_sem
);
291 static inline void mddev_unlock(mddev_t
* mddev
)
293 up(&mddev
->reconfig_sem
);
295 md_wakeup_thread(mddev
->thread
);
298 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
301 struct list_head
*tmp
;
303 ITERATE_RDEV(mddev
,rdev
,tmp
) {
304 if (rdev
->desc_nr
== nr
)
310 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
312 struct list_head
*tmp
;
315 ITERATE_RDEV(mddev
,rdev
,tmp
) {
316 if (rdev
->bdev
->bd_dev
== dev
)
322 static struct mdk_personality
*find_pers(int level
, char *clevel
)
324 struct mdk_personality
*pers
;
325 list_for_each_entry(pers
, &pers_list
, list
) {
326 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
328 if (strcmp(pers
->name
, clevel
)==0)
334 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
336 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
337 return MD_NEW_SIZE_BLOCKS(size
);
340 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
344 size
= rdev
->sb_offset
;
347 size
&= ~((sector_t
)chunk_size
/1024 - 1);
351 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
356 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
357 if (!rdev
->sb_page
) {
358 printk(KERN_ALERT
"md: out of memory.\n");
365 static void free_disk_sb(mdk_rdev_t
* rdev
)
368 put_page(rdev
->sb_page
);
370 rdev
->sb_page
= NULL
;
377 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
379 mdk_rdev_t
*rdev
= bio
->bi_private
;
380 mddev_t
*mddev
= rdev
->mddev
;
384 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
385 md_error(mddev
, rdev
);
387 if (atomic_dec_and_test(&mddev
->pending_writes
))
388 wake_up(&mddev
->sb_wait
);
393 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
395 struct bio
*bio2
= bio
->bi_private
;
396 mdk_rdev_t
*rdev
= bio2
->bi_private
;
397 mddev_t
*mddev
= rdev
->mddev
;
401 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
402 error
== -EOPNOTSUPP
) {
404 /* barriers don't appear to be supported :-( */
405 set_bit(BarriersNotsupp
, &rdev
->flags
);
406 mddev
->barriers_work
= 0;
407 spin_lock_irqsave(&mddev
->write_lock
, flags
);
408 bio2
->bi_next
= mddev
->biolist
;
409 mddev
->biolist
= bio2
;
410 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
411 wake_up(&mddev
->sb_wait
);
416 bio
->bi_private
= rdev
;
417 return super_written(bio
, bytes_done
, error
);
420 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
421 sector_t sector
, int size
, struct page
*page
)
423 /* write first size bytes of page to sector of rdev
424 * Increment mddev->pending_writes before returning
425 * and decrement it on completion, waking up sb_wait
426 * if zero is reached.
427 * If an error occurred, call md_error
429 * As we might need to resubmit the request if BIO_RW_BARRIER
430 * causes ENOTSUPP, we allocate a spare bio...
432 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
433 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
435 bio
->bi_bdev
= rdev
->bdev
;
436 bio
->bi_sector
= sector
;
437 bio_add_page(bio
, page
, size
, 0);
438 bio
->bi_private
= rdev
;
439 bio
->bi_end_io
= super_written
;
442 atomic_inc(&mddev
->pending_writes
);
443 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
445 rw
|= (1<<BIO_RW_BARRIER
);
446 rbio
= bio_clone(bio
, GFP_NOIO
);
447 rbio
->bi_private
= bio
;
448 rbio
->bi_end_io
= super_written_barrier
;
449 submit_bio(rw
, rbio
);
454 void md_super_wait(mddev_t
*mddev
)
456 /* wait for all superblock writes that were scheduled to complete.
457 * if any had to be retried (due to BARRIER problems), retry them
461 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
462 if (atomic_read(&mddev
->pending_writes
)==0)
464 while (mddev
->biolist
) {
466 spin_lock_irq(&mddev
->write_lock
);
467 bio
= mddev
->biolist
;
468 mddev
->biolist
= bio
->bi_next
;
470 spin_unlock_irq(&mddev
->write_lock
);
471 submit_bio(bio
->bi_rw
, bio
);
475 finish_wait(&mddev
->sb_wait
, &wq
);
478 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
483 complete((struct completion
*)bio
->bi_private
);
487 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
488 struct page
*page
, int rw
)
490 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
491 struct completion event
;
494 rw
|= (1 << BIO_RW_SYNC
);
497 bio
->bi_sector
= sector
;
498 bio_add_page(bio
, page
, size
, 0);
499 init_completion(&event
);
500 bio
->bi_private
= &event
;
501 bio
->bi_end_io
= bi_complete
;
503 wait_for_completion(&event
);
505 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
509 EXPORT_SYMBOL_GPL(sync_page_io
);
511 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
513 char b
[BDEVNAME_SIZE
];
514 if (!rdev
->sb_page
) {
522 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
528 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
529 bdevname(rdev
->bdev
,b
));
533 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
535 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
536 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
537 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
538 (sb1
->set_uuid3
== sb2
->set_uuid3
))
546 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
549 mdp_super_t
*tmp1
, *tmp2
;
551 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
552 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
554 if (!tmp1
|| !tmp2
) {
556 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
564 * nr_disks is not constant
569 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
580 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
582 unsigned int disk_csum
, csum
;
584 disk_csum
= sb
->sb_csum
;
586 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
587 sb
->sb_csum
= disk_csum
;
593 * Handle superblock details.
594 * We want to be able to handle multiple superblock formats
595 * so we have a common interface to them all, and an array of
596 * different handlers.
597 * We rely on user-space to write the initial superblock, and support
598 * reading and updating of superblocks.
599 * Interface methods are:
600 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
601 * loads and validates a superblock on dev.
602 * if refdev != NULL, compare superblocks on both devices
604 * 0 - dev has a superblock that is compatible with refdev
605 * 1 - dev has a superblock that is compatible and newer than refdev
606 * so dev should be used as the refdev in future
607 * -EINVAL superblock incompatible or invalid
608 * -othererror e.g. -EIO
610 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
611 * Verify that dev is acceptable into mddev.
612 * The first time, mddev->raid_disks will be 0, and data from
613 * dev should be merged in. Subsequent calls check that dev
614 * is new enough. Return 0 or -EINVAL
616 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
617 * Update the superblock for rdev with data in mddev
618 * This does not write to disc.
624 struct module
*owner
;
625 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
626 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
627 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
631 * load_super for 0.90.0
633 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
635 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
641 * Calculate the position of the superblock,
642 * it's at the end of the disk.
644 * It also happens to be a multiple of 4Kb.
646 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
647 rdev
->sb_offset
= sb_offset
;
649 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
654 bdevname(rdev
->bdev
, b
);
655 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
657 if (sb
->md_magic
!= MD_SB_MAGIC
) {
658 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
663 if (sb
->major_version
!= 0 ||
664 sb
->minor_version
!= 90) {
665 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
666 sb
->major_version
, sb
->minor_version
,
671 if (sb
->raid_disks
<= 0)
674 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
675 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
680 rdev
->preferred_minor
= sb
->md_minor
;
681 rdev
->data_offset
= 0;
682 rdev
->sb_size
= MD_SB_BYTES
;
684 if (sb
->level
== LEVEL_MULTIPATH
)
687 rdev
->desc_nr
= sb
->this_disk
.number
;
693 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
694 if (!uuid_equal(refsb
, sb
)) {
695 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
696 b
, bdevname(refdev
->bdev
,b2
));
699 if (!sb_equal(refsb
, sb
)) {
700 printk(KERN_WARNING
"md: %s has same UUID"
701 " but different superblock to %s\n",
702 b
, bdevname(refdev
->bdev
, b2
));
706 ev2
= md_event(refsb
);
712 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
714 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
715 /* "this cannot possibly happen" ... */
723 * validate_super for 0.90.0
725 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
728 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
730 rdev
->raid_disk
= -1;
732 if (mddev
->raid_disks
== 0) {
733 mddev
->major_version
= 0;
734 mddev
->minor_version
= sb
->minor_version
;
735 mddev
->patch_version
= sb
->patch_version
;
736 mddev
->persistent
= ! sb
->not_persistent
;
737 mddev
->chunk_size
= sb
->chunk_size
;
738 mddev
->ctime
= sb
->ctime
;
739 mddev
->utime
= sb
->utime
;
740 mddev
->level
= sb
->level
;
741 mddev
->clevel
[0] = 0;
742 mddev
->layout
= sb
->layout
;
743 mddev
->raid_disks
= sb
->raid_disks
;
744 mddev
->size
= sb
->size
;
745 mddev
->events
= md_event(sb
);
746 mddev
->bitmap_offset
= 0;
747 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
749 if (sb
->state
& (1<<MD_SB_CLEAN
))
750 mddev
->recovery_cp
= MaxSector
;
752 if (sb
->events_hi
== sb
->cp_events_hi
&&
753 sb
->events_lo
== sb
->cp_events_lo
) {
754 mddev
->recovery_cp
= sb
->recovery_cp
;
756 mddev
->recovery_cp
= 0;
759 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
760 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
761 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
762 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
764 mddev
->max_disks
= MD_SB_DISKS
;
766 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
767 mddev
->bitmap_file
== NULL
) {
768 if (mddev
->level
!= 1 && mddev
->level
!= 4
769 && mddev
->level
!= 5 && mddev
->level
!= 6
770 && mddev
->level
!= 10) {
771 /* FIXME use a better test */
772 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
775 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
778 } else if (mddev
->pers
== NULL
) {
779 /* Insist on good event counter while assembling */
780 __u64 ev1
= md_event(sb
);
782 if (ev1
< mddev
->events
)
784 } else if (mddev
->bitmap
) {
785 /* if adding to array with a bitmap, then we can accept an
786 * older device ... but not too old.
788 __u64 ev1
= md_event(sb
);
789 if (ev1
< mddev
->bitmap
->events_cleared
)
791 } else /* just a hot-add of a new device, leave raid_disk at -1 */
794 if (mddev
->level
!= LEVEL_MULTIPATH
) {
795 desc
= sb
->disks
+ rdev
->desc_nr
;
797 if (desc
->state
& (1<<MD_DISK_FAULTY
))
798 set_bit(Faulty
, &rdev
->flags
);
799 else if (desc
->state
& (1<<MD_DISK_SYNC
) &&
800 desc
->raid_disk
< mddev
->raid_disks
) {
801 set_bit(In_sync
, &rdev
->flags
);
802 rdev
->raid_disk
= desc
->raid_disk
;
804 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
805 set_bit(WriteMostly
, &rdev
->flags
);
806 } else /* MULTIPATH are always insync */
807 set_bit(In_sync
, &rdev
->flags
);
812 * sync_super for 0.90.0
814 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
817 struct list_head
*tmp
;
819 int next_spare
= mddev
->raid_disks
;
822 /* make rdev->sb match mddev data..
825 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
826 * 3/ any empty disks < next_spare become removed
828 * disks[0] gets initialised to REMOVED because
829 * we cannot be sure from other fields if it has
830 * been initialised or not.
833 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
835 rdev
->sb_size
= MD_SB_BYTES
;
837 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
839 memset(sb
, 0, sizeof(*sb
));
841 sb
->md_magic
= MD_SB_MAGIC
;
842 sb
->major_version
= mddev
->major_version
;
843 sb
->minor_version
= mddev
->minor_version
;
844 sb
->patch_version
= mddev
->patch_version
;
845 sb
->gvalid_words
= 0; /* ignored */
846 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
847 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
848 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
849 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
851 sb
->ctime
= mddev
->ctime
;
852 sb
->level
= mddev
->level
;
853 sb
->size
= mddev
->size
;
854 sb
->raid_disks
= mddev
->raid_disks
;
855 sb
->md_minor
= mddev
->md_minor
;
856 sb
->not_persistent
= !mddev
->persistent
;
857 sb
->utime
= mddev
->utime
;
859 sb
->events_hi
= (mddev
->events
>>32);
860 sb
->events_lo
= (u32
)mddev
->events
;
864 sb
->recovery_cp
= mddev
->recovery_cp
;
865 sb
->cp_events_hi
= (mddev
->events
>>32);
866 sb
->cp_events_lo
= (u32
)mddev
->events
;
867 if (mddev
->recovery_cp
== MaxSector
)
868 sb
->state
= (1<< MD_SB_CLEAN
);
872 sb
->layout
= mddev
->layout
;
873 sb
->chunk_size
= mddev
->chunk_size
;
875 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
876 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
878 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
879 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
882 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
883 && !test_bit(Faulty
, &rdev2
->flags
))
884 desc_nr
= rdev2
->raid_disk
;
886 desc_nr
= next_spare
++;
887 rdev2
->desc_nr
= desc_nr
;
888 d
= &sb
->disks
[rdev2
->desc_nr
];
890 d
->number
= rdev2
->desc_nr
;
891 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
892 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
893 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
894 && !test_bit(Faulty
, &rdev2
->flags
))
895 d
->raid_disk
= rdev2
->raid_disk
;
897 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
898 if (test_bit(Faulty
, &rdev2
->flags
))
899 d
->state
= (1<<MD_DISK_FAULTY
);
900 else if (test_bit(In_sync
, &rdev2
->flags
)) {
901 d
->state
= (1<<MD_DISK_ACTIVE
);
902 d
->state
|= (1<<MD_DISK_SYNC
);
910 if (test_bit(WriteMostly
, &rdev2
->flags
))
911 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
913 /* now set the "removed" and "faulty" bits on any missing devices */
914 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
915 mdp_disk_t
*d
= &sb
->disks
[i
];
916 if (d
->state
== 0 && d
->number
== 0) {
919 d
->state
= (1<<MD_DISK_REMOVED
);
920 d
->state
|= (1<<MD_DISK_FAULTY
);
924 sb
->nr_disks
= nr_disks
;
925 sb
->active_disks
= active
;
926 sb
->working_disks
= working
;
927 sb
->failed_disks
= failed
;
928 sb
->spare_disks
= spare
;
930 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
931 sb
->sb_csum
= calc_sb_csum(sb
);
935 * version 1 superblock
938 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
940 unsigned int disk_csum
, csum
;
941 unsigned long long newcsum
;
942 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
943 unsigned int *isuper
= (unsigned int*)sb
;
946 disk_csum
= sb
->sb_csum
;
949 for (i
=0; size
>=4; size
-= 4 )
950 newcsum
+= le32_to_cpu(*isuper
++);
953 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
955 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
956 sb
->sb_csum
= disk_csum
;
957 return cpu_to_le32(csum
);
960 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
962 struct mdp_superblock_1
*sb
;
965 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
969 * Calculate the position of the superblock.
970 * It is always aligned to a 4K boundary and
971 * depeding on minor_version, it can be:
972 * 0: At least 8K, but less than 12K, from end of device
973 * 1: At start of device
974 * 2: 4K from start of device.
976 switch(minor_version
) {
978 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
980 sb_offset
&= ~(sector_t
)(4*2-1);
981 /* convert from sectors to K */
993 rdev
->sb_offset
= sb_offset
;
995 /* superblock is rarely larger than 1K, but it can be larger,
996 * and it is safe to read 4k, so we do that
998 ret
= read_disk_sb(rdev
, 4096);
1002 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1004 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1005 sb
->major_version
!= cpu_to_le32(1) ||
1006 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1007 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1008 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1011 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1012 printk("md: invalid superblock checksum on %s\n",
1013 bdevname(rdev
->bdev
,b
));
1016 if (le64_to_cpu(sb
->data_size
) < 10) {
1017 printk("md: data_size too small on %s\n",
1018 bdevname(rdev
->bdev
,b
));
1021 rdev
->preferred_minor
= 0xffff;
1022 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1023 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1025 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1026 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1027 if (rdev
->sb_size
& bmask
)
1028 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1034 struct mdp_superblock_1
*refsb
=
1035 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1037 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1038 sb
->level
!= refsb
->level
||
1039 sb
->layout
!= refsb
->layout
||
1040 sb
->chunksize
!= refsb
->chunksize
) {
1041 printk(KERN_WARNING
"md: %s has strangely different"
1042 " superblock to %s\n",
1043 bdevname(rdev
->bdev
,b
),
1044 bdevname(refdev
->bdev
,b2
));
1047 ev1
= le64_to_cpu(sb
->events
);
1048 ev2
= le64_to_cpu(refsb
->events
);
1056 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1058 rdev
->size
= rdev
->sb_offset
;
1059 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1061 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1062 if (le32_to_cpu(sb
->chunksize
))
1063 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1065 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1070 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1072 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1074 rdev
->raid_disk
= -1;
1076 if (mddev
->raid_disks
== 0) {
1077 mddev
->major_version
= 1;
1078 mddev
->patch_version
= 0;
1079 mddev
->persistent
= 1;
1080 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1081 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1082 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1083 mddev
->level
= le32_to_cpu(sb
->level
);
1084 mddev
->clevel
[0] = 0;
1085 mddev
->layout
= le32_to_cpu(sb
->layout
);
1086 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1087 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1088 mddev
->events
= le64_to_cpu(sb
->events
);
1089 mddev
->bitmap_offset
= 0;
1090 mddev
->default_bitmap_offset
= 1024 >> 9;
1092 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1093 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1095 mddev
->max_disks
= (4096-256)/2;
1097 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1098 mddev
->bitmap_file
== NULL
) {
1099 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1100 && mddev
->level
!= 10) {
1101 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1104 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1106 } else if (mddev
->pers
== NULL
) {
1107 /* Insist of good event counter while assembling */
1108 __u64 ev1
= le64_to_cpu(sb
->events
);
1110 if (ev1
< mddev
->events
)
1112 } else if (mddev
->bitmap
) {
1113 /* If adding to array with a bitmap, then we can accept an
1114 * older device, but not too old.
1116 __u64 ev1
= le64_to_cpu(sb
->events
);
1117 if (ev1
< mddev
->bitmap
->events_cleared
)
1119 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1122 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1124 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1125 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1127 case 0xffff: /* spare */
1129 case 0xfffe: /* faulty */
1130 set_bit(Faulty
, &rdev
->flags
);
1133 set_bit(In_sync
, &rdev
->flags
);
1134 rdev
->raid_disk
= role
;
1137 if (sb
->devflags
& WriteMostly1
)
1138 set_bit(WriteMostly
, &rdev
->flags
);
1139 } else /* MULTIPATH are always insync */
1140 set_bit(In_sync
, &rdev
->flags
);
1145 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1147 struct mdp_superblock_1
*sb
;
1148 struct list_head
*tmp
;
1151 /* make rdev->sb match mddev and rdev data. */
1153 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1155 sb
->feature_map
= 0;
1157 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1158 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1159 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1161 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1162 sb
->events
= cpu_to_le64(mddev
->events
);
1164 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1166 sb
->resync_offset
= cpu_to_le64(0);
1168 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1170 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1171 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1173 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1174 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1175 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1179 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1180 if (rdev2
->desc_nr
+1 > max_dev
)
1181 max_dev
= rdev2
->desc_nr
+1;
1183 sb
->max_dev
= cpu_to_le32(max_dev
);
1184 for (i
=0; i
<max_dev
;i
++)
1185 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1187 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1189 if (test_bit(Faulty
, &rdev2
->flags
))
1190 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1191 else if (test_bit(In_sync
, &rdev2
->flags
))
1192 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1194 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1197 sb
->recovery_offset
= cpu_to_le64(0); /* not supported yet */
1198 sb
->sb_csum
= calc_sb_1_csum(sb
);
1202 static struct super_type super_types
[] = {
1205 .owner
= THIS_MODULE
,
1206 .load_super
= super_90_load
,
1207 .validate_super
= super_90_validate
,
1208 .sync_super
= super_90_sync
,
1212 .owner
= THIS_MODULE
,
1213 .load_super
= super_1_load
,
1214 .validate_super
= super_1_validate
,
1215 .sync_super
= super_1_sync
,
1219 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1221 struct list_head
*tmp
;
1224 ITERATE_RDEV(mddev
,rdev
,tmp
)
1225 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1231 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1233 struct list_head
*tmp
;
1236 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1237 if (match_dev_unit(mddev2
, rdev
))
1243 static LIST_HEAD(pending_raid_disks
);
1245 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1247 mdk_rdev_t
*same_pdev
;
1248 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1256 /* make sure rdev->size exceeds mddev->size */
1257 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1259 /* Cannot change size, so fail */
1262 mddev
->size
= rdev
->size
;
1264 same_pdev
= match_dev_unit(mddev
, rdev
);
1267 "%s: WARNING: %s appears to be on the same physical"
1268 " disk as %s. True\n protection against single-disk"
1269 " failure might be compromised.\n",
1270 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1271 bdevname(same_pdev
->bdev
,b2
));
1273 /* Verify rdev->desc_nr is unique.
1274 * If it is -1, assign a free number, else
1275 * check number is not in use
1277 if (rdev
->desc_nr
< 0) {
1279 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1280 while (find_rdev_nr(mddev
, choice
))
1282 rdev
->desc_nr
= choice
;
1284 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1287 bdevname(rdev
->bdev
,b
);
1288 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1290 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1293 list_add(&rdev
->same_set
, &mddev
->disks
);
1294 rdev
->mddev
= mddev
;
1295 printk(KERN_INFO
"md: bind<%s>\n", b
);
1297 rdev
->kobj
.parent
= &mddev
->kobj
;
1298 kobject_add(&rdev
->kobj
);
1300 if (rdev
->bdev
->bd_part
)
1301 ko
= &rdev
->bdev
->bd_part
->kobj
;
1303 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1304 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1305 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1309 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1311 char b
[BDEVNAME_SIZE
];
1316 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1317 list_del_init(&rdev
->same_set
);
1318 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1320 sysfs_remove_link(&rdev
->kobj
, "block");
1321 kobject_del(&rdev
->kobj
);
1325 * prevent the device from being mounted, repartitioned or
1326 * otherwise reused by a RAID array (or any other kernel
1327 * subsystem), by bd_claiming the device.
1329 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1332 struct block_device
*bdev
;
1333 char b
[BDEVNAME_SIZE
];
1335 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1337 printk(KERN_ERR
"md: could not open %s.\n",
1338 __bdevname(dev
, b
));
1339 return PTR_ERR(bdev
);
1341 err
= bd_claim(bdev
, rdev
);
1343 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1352 static void unlock_rdev(mdk_rdev_t
*rdev
)
1354 struct block_device
*bdev
= rdev
->bdev
;
1362 void md_autodetect_dev(dev_t dev
);
1364 static void export_rdev(mdk_rdev_t
* rdev
)
1366 char b
[BDEVNAME_SIZE
];
1367 printk(KERN_INFO
"md: export_rdev(%s)\n",
1368 bdevname(rdev
->bdev
,b
));
1372 list_del_init(&rdev
->same_set
);
1374 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1377 kobject_put(&rdev
->kobj
);
1380 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1382 unbind_rdev_from_array(rdev
);
1386 static void export_array(mddev_t
*mddev
)
1388 struct list_head
*tmp
;
1391 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1396 kick_rdev_from_array(rdev
);
1398 if (!list_empty(&mddev
->disks
))
1400 mddev
->raid_disks
= 0;
1401 mddev
->major_version
= 0;
1404 static void print_desc(mdp_disk_t
*desc
)
1406 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1407 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1410 static void print_sb(mdp_super_t
*sb
)
1415 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1416 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1417 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1419 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1420 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1421 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1422 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1423 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1424 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1425 sb
->failed_disks
, sb
->spare_disks
,
1426 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1429 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1432 desc
= sb
->disks
+ i
;
1433 if (desc
->number
|| desc
->major
|| desc
->minor
||
1434 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1435 printk(" D %2d: ", i
);
1439 printk(KERN_INFO
"md: THIS: ");
1440 print_desc(&sb
->this_disk
);
1444 static void print_rdev(mdk_rdev_t
*rdev
)
1446 char b
[BDEVNAME_SIZE
];
1447 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1448 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1449 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1451 if (rdev
->sb_loaded
) {
1452 printk(KERN_INFO
"md: rdev superblock:\n");
1453 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1455 printk(KERN_INFO
"md: no rdev superblock!\n");
1458 void md_print_devices(void)
1460 struct list_head
*tmp
, *tmp2
;
1463 char b
[BDEVNAME_SIZE
];
1466 printk("md: **********************************\n");
1467 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1468 printk("md: **********************************\n");
1469 ITERATE_MDDEV(mddev
,tmp
) {
1472 bitmap_print_sb(mddev
->bitmap
);
1474 printk("%s: ", mdname(mddev
));
1475 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1476 printk("<%s>", bdevname(rdev
->bdev
,b
));
1479 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1482 printk("md: **********************************\n");
1487 static void sync_sbs(mddev_t
* mddev
)
1490 struct list_head
*tmp
;
1492 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1493 super_types
[mddev
->major_version
].
1494 sync_super(mddev
, rdev
);
1495 rdev
->sb_loaded
= 1;
1499 static void md_update_sb(mddev_t
* mddev
)
1502 struct list_head
*tmp
;
1507 spin_lock_irq(&mddev
->write_lock
);
1508 sync_req
= mddev
->in_sync
;
1509 mddev
->utime
= get_seconds();
1512 if (!mddev
->events
) {
1514 * oops, this 64-bit counter should never wrap.
1515 * Either we are in around ~1 trillion A.C., assuming
1516 * 1 reboot per second, or we have a bug:
1521 mddev
->sb_dirty
= 2;
1525 * do not write anything to disk if using
1526 * nonpersistent superblocks
1528 if (!mddev
->persistent
) {
1529 mddev
->sb_dirty
= 0;
1530 spin_unlock_irq(&mddev
->write_lock
);
1531 wake_up(&mddev
->sb_wait
);
1534 spin_unlock_irq(&mddev
->write_lock
);
1537 "md: updating %s RAID superblock on device (in sync %d)\n",
1538 mdname(mddev
),mddev
->in_sync
);
1540 err
= bitmap_update_sb(mddev
->bitmap
);
1541 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1542 char b
[BDEVNAME_SIZE
];
1543 dprintk(KERN_INFO
"md: ");
1544 if (test_bit(Faulty
, &rdev
->flags
))
1545 dprintk("(skipping faulty ");
1547 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1548 if (!test_bit(Faulty
, &rdev
->flags
)) {
1549 md_super_write(mddev
,rdev
,
1550 rdev
->sb_offset
<<1, rdev
->sb_size
,
1552 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1553 bdevname(rdev
->bdev
,b
),
1554 (unsigned long long)rdev
->sb_offset
);
1558 if (mddev
->level
== LEVEL_MULTIPATH
)
1559 /* only need to write one superblock... */
1562 md_super_wait(mddev
);
1563 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1565 spin_lock_irq(&mddev
->write_lock
);
1566 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1567 /* have to write it out again */
1568 spin_unlock_irq(&mddev
->write_lock
);
1571 mddev
->sb_dirty
= 0;
1572 spin_unlock_irq(&mddev
->write_lock
);
1573 wake_up(&mddev
->sb_wait
);
1577 /* words written to sysfs files may, or my not, be \n terminated.
1578 * We want to accept with case. For this we use cmd_match.
1580 static int cmd_match(const char *cmd
, const char *str
)
1582 /* See if cmd, written into a sysfs file, matches
1583 * str. They must either be the same, or cmd can
1584 * have a trailing newline
1586 while (*cmd
&& *str
&& *cmd
== *str
) {
1597 struct rdev_sysfs_entry
{
1598 struct attribute attr
;
1599 ssize_t (*show
)(mdk_rdev_t
*, char *);
1600 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1604 state_show(mdk_rdev_t
*rdev
, char *page
)
1609 if (test_bit(Faulty
, &rdev
->flags
)) {
1610 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1613 if (test_bit(In_sync
, &rdev
->flags
)) {
1614 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1617 if (!test_bit(Faulty
, &rdev
->flags
) &&
1618 !test_bit(In_sync
, &rdev
->flags
)) {
1619 len
+= sprintf(page
+len
, "%sspare", sep
);
1622 return len
+sprintf(page
+len
, "\n");
1625 static struct rdev_sysfs_entry
1626 rdev_state
= __ATTR_RO(state
);
1629 super_show(mdk_rdev_t
*rdev
, char *page
)
1631 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1632 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1633 return rdev
->sb_size
;
1637 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1640 errors_show(mdk_rdev_t
*rdev
, char *page
)
1642 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1646 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1649 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1650 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1651 atomic_set(&rdev
->corrected_errors
, n
);
1656 static struct rdev_sysfs_entry rdev_errors
=
1657 __ATTR(errors
, 0644, errors_show
, errors_store
);
1660 slot_show(mdk_rdev_t
*rdev
, char *page
)
1662 if (rdev
->raid_disk
< 0)
1663 return sprintf(page
, "none\n");
1665 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1669 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1672 int slot
= simple_strtoul(buf
, &e
, 10);
1673 if (strncmp(buf
, "none", 4)==0)
1675 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1677 if (rdev
->mddev
->pers
)
1678 /* Cannot set slot in active array (yet) */
1680 if (slot
>= rdev
->mddev
->raid_disks
)
1682 rdev
->raid_disk
= slot
;
1683 /* assume it is working */
1685 set_bit(In_sync
, &rdev
->flags
);
1690 static struct rdev_sysfs_entry rdev_slot
=
1691 __ATTR(slot
, 0644, slot_show
, slot_store
);
1694 offset_show(mdk_rdev_t
*rdev
, char *page
)
1696 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1700 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1703 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1704 if (e
==buf
|| (*e
&& *e
!= '\n'))
1706 if (rdev
->mddev
->pers
)
1708 rdev
->data_offset
= offset
;
1712 static struct rdev_sysfs_entry rdev_offset
=
1713 __ATTR(offset
, 0644, offset_show
, offset_store
);
1716 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1718 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1722 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1725 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1726 if (e
==buf
|| (*e
&& *e
!= '\n'))
1728 if (rdev
->mddev
->pers
)
1731 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1732 rdev
->mddev
->size
= size
;
1736 static struct rdev_sysfs_entry rdev_size
=
1737 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1739 static struct attribute
*rdev_default_attrs
[] = {
1749 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1751 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1752 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1756 return entry
->show(rdev
, page
);
1760 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1761 const char *page
, size_t length
)
1763 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1764 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1768 return entry
->store(rdev
, page
, length
);
1771 static void rdev_free(struct kobject
*ko
)
1773 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1776 static struct sysfs_ops rdev_sysfs_ops
= {
1777 .show
= rdev_attr_show
,
1778 .store
= rdev_attr_store
,
1780 static struct kobj_type rdev_ktype
= {
1781 .release
= rdev_free
,
1782 .sysfs_ops
= &rdev_sysfs_ops
,
1783 .default_attrs
= rdev_default_attrs
,
1787 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1789 * mark the device faulty if:
1791 * - the device is nonexistent (zero size)
1792 * - the device has no valid superblock
1794 * a faulty rdev _never_ has rdev->sb set.
1796 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1798 char b
[BDEVNAME_SIZE
];
1803 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1805 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1806 return ERR_PTR(-ENOMEM
);
1809 if ((err
= alloc_disk_sb(rdev
)))
1812 err
= lock_rdev(rdev
, newdev
);
1816 rdev
->kobj
.parent
= NULL
;
1817 rdev
->kobj
.ktype
= &rdev_ktype
;
1818 kobject_init(&rdev
->kobj
);
1822 rdev
->data_offset
= 0;
1823 atomic_set(&rdev
->nr_pending
, 0);
1824 atomic_set(&rdev
->read_errors
, 0);
1825 atomic_set(&rdev
->corrected_errors
, 0);
1827 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1830 "md: %s has zero or unknown size, marking faulty!\n",
1831 bdevname(rdev
->bdev
,b
));
1836 if (super_format
>= 0) {
1837 err
= super_types
[super_format
].
1838 load_super(rdev
, NULL
, super_minor
);
1839 if (err
== -EINVAL
) {
1841 "md: %s has invalid sb, not importing!\n",
1842 bdevname(rdev
->bdev
,b
));
1847 "md: could not read %s's sb, not importing!\n",
1848 bdevname(rdev
->bdev
,b
));
1852 INIT_LIST_HEAD(&rdev
->same_set
);
1857 if (rdev
->sb_page
) {
1863 return ERR_PTR(err
);
1867 * Check a full RAID array for plausibility
1871 static void analyze_sbs(mddev_t
* mddev
)
1874 struct list_head
*tmp
;
1875 mdk_rdev_t
*rdev
, *freshest
;
1876 char b
[BDEVNAME_SIZE
];
1879 ITERATE_RDEV(mddev
,rdev
,tmp
)
1880 switch (super_types
[mddev
->major_version
].
1881 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1889 "md: fatal superblock inconsistency in %s"
1890 " -- removing from array\n",
1891 bdevname(rdev
->bdev
,b
));
1892 kick_rdev_from_array(rdev
);
1896 super_types
[mddev
->major_version
].
1897 validate_super(mddev
, freshest
);
1900 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1901 if (rdev
!= freshest
)
1902 if (super_types
[mddev
->major_version
].
1903 validate_super(mddev
, rdev
)) {
1904 printk(KERN_WARNING
"md: kicking non-fresh %s"
1906 bdevname(rdev
->bdev
,b
));
1907 kick_rdev_from_array(rdev
);
1910 if (mddev
->level
== LEVEL_MULTIPATH
) {
1911 rdev
->desc_nr
= i
++;
1912 rdev
->raid_disk
= rdev
->desc_nr
;
1913 set_bit(In_sync
, &rdev
->flags
);
1919 if (mddev
->recovery_cp
!= MaxSector
&&
1921 printk(KERN_ERR
"md: %s: raid array is not clean"
1922 " -- starting background reconstruction\n",
1928 level_show(mddev_t
*mddev
, char *page
)
1930 struct mdk_personality
*p
= mddev
->pers
;
1932 return sprintf(page
, "%s\n", p
->name
);
1933 else if (mddev
->clevel
[0])
1934 return sprintf(page
, "%s\n", mddev
->clevel
);
1935 else if (mddev
->level
!= LEVEL_NONE
)
1936 return sprintf(page
, "%d\n", mddev
->level
);
1942 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1949 if (len
>= sizeof(mddev
->clevel
))
1951 strncpy(mddev
->clevel
, buf
, len
);
1952 if (mddev
->clevel
[len
-1] == '\n')
1954 mddev
->clevel
[len
] = 0;
1955 mddev
->level
= LEVEL_NONE
;
1959 static struct md_sysfs_entry md_level
=
1960 __ATTR(level
, 0644, level_show
, level_store
);
1963 raid_disks_show(mddev_t
*mddev
, char *page
)
1965 if (mddev
->raid_disks
== 0)
1967 return sprintf(page
, "%d\n", mddev
->raid_disks
);
1970 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
1973 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1975 /* can only set raid_disks if array is not yet active */
1978 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1980 if (!*buf
|| (*e
&& *e
!= '\n'))
1984 rv
= update_raid_disks(mddev
, n
);
1986 mddev
->raid_disks
= n
;
1987 return rv
? rv
: len
;
1989 static struct md_sysfs_entry md_raid_disks
=
1990 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
1993 chunk_size_show(mddev_t
*mddev
, char *page
)
1995 return sprintf(page
, "%d\n", mddev
->chunk_size
);
1999 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2001 /* can only set chunk_size if array is not yet active */
2003 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2007 if (!*buf
|| (*e
&& *e
!= '\n'))
2010 mddev
->chunk_size
= n
;
2013 static struct md_sysfs_entry md_chunk_size
=
2014 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2017 null_show(mddev_t
*mddev
, char *page
)
2023 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2025 /* buf must be %d:%d\n? giving major and minor numbers */
2026 /* The new device is added to the array.
2027 * If the array has a persistent superblock, we read the
2028 * superblock to initialise info and check validity.
2029 * Otherwise, only checking done is that in bind_rdev_to_array,
2030 * which mainly checks size.
2033 int major
= simple_strtoul(buf
, &e
, 10);
2039 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2041 minor
= simple_strtoul(e
+1, &e
, 10);
2042 if (*e
&& *e
!= '\n')
2044 dev
= MKDEV(major
, minor
);
2045 if (major
!= MAJOR(dev
) ||
2046 minor
!= MINOR(dev
))
2050 if (mddev
->persistent
) {
2051 rdev
= md_import_device(dev
, mddev
->major_version
,
2052 mddev
->minor_version
);
2053 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2054 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2055 mdk_rdev_t
, same_set
);
2056 err
= super_types
[mddev
->major_version
]
2057 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2062 rdev
= md_import_device(dev
, -1, -1);
2065 return PTR_ERR(rdev
);
2066 err
= bind_rdev_to_array(rdev
, mddev
);
2070 return err
? err
: len
;
2073 static struct md_sysfs_entry md_new_device
=
2074 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2077 size_show(mddev_t
*mddev
, char *page
)
2079 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2082 static int update_size(mddev_t
*mddev
, unsigned long size
);
2085 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2087 /* If array is inactive, we can reduce the component size, but
2088 * not increase it (except from 0).
2089 * If array is active, we can try an on-line resize
2093 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2094 if (!*buf
|| *buf
== '\n' ||
2099 err
= update_size(mddev
, size
);
2100 md_update_sb(mddev
);
2102 if (mddev
->size
== 0 ||
2108 return err
? err
: len
;
2111 static struct md_sysfs_entry md_size
=
2112 __ATTR(component_size
, 0644, size_show
, size_store
);
2116 * This is either 'none' for arrays with externally managed metadata,
2117 * or N.M for internally known formats
2120 metadata_show(mddev_t
*mddev
, char *page
)
2122 if (mddev
->persistent
)
2123 return sprintf(page
, "%d.%d\n",
2124 mddev
->major_version
, mddev
->minor_version
);
2126 return sprintf(page
, "none\n");
2130 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2134 if (!list_empty(&mddev
->disks
))
2137 if (cmd_match(buf
, "none")) {
2138 mddev
->persistent
= 0;
2139 mddev
->major_version
= 0;
2140 mddev
->minor_version
= 90;
2143 major
= simple_strtoul(buf
, &e
, 10);
2144 if (e
==buf
|| *e
!= '.')
2147 minor
= simple_strtoul(buf
, &e
, 10);
2148 if (e
==buf
|| *e
!= '\n')
2150 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2151 super_types
[major
].name
== NULL
)
2153 mddev
->major_version
= major
;
2154 mddev
->minor_version
= minor
;
2155 mddev
->persistent
= 1;
2159 static struct md_sysfs_entry md_metadata
=
2160 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2163 action_show(mddev_t
*mddev
, char *page
)
2165 char *type
= "idle";
2166 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2167 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2168 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2169 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2171 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2178 return sprintf(page
, "%s\n", type
);
2182 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2184 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2187 if (cmd_match(page
, "idle")) {
2188 if (mddev
->sync_thread
) {
2189 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2190 md_unregister_thread(mddev
->sync_thread
);
2191 mddev
->sync_thread
= NULL
;
2192 mddev
->recovery
= 0;
2194 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2195 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2197 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2198 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2200 if (cmd_match(page
, "check"))
2201 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2202 else if (cmd_match(page
, "repair"))
2204 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2205 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2207 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2208 md_wakeup_thread(mddev
->thread
);
2213 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2215 return sprintf(page
, "%llu\n",
2216 (unsigned long long) mddev
->resync_mismatches
);
2219 static struct md_sysfs_entry
2220 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2223 static struct md_sysfs_entry
2224 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2227 sync_min_show(mddev_t
*mddev
, char *page
)
2229 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2230 mddev
->sync_speed_min
? "local": "system");
2234 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2238 if (strncmp(buf
, "system", 6)==0) {
2239 mddev
->sync_speed_min
= 0;
2242 min
= simple_strtoul(buf
, &e
, 10);
2243 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2245 mddev
->sync_speed_min
= min
;
2249 static struct md_sysfs_entry md_sync_min
=
2250 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2253 sync_max_show(mddev_t
*mddev
, char *page
)
2255 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2256 mddev
->sync_speed_max
? "local": "system");
2260 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2264 if (strncmp(buf
, "system", 6)==0) {
2265 mddev
->sync_speed_max
= 0;
2268 max
= simple_strtoul(buf
, &e
, 10);
2269 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2271 mddev
->sync_speed_max
= max
;
2275 static struct md_sysfs_entry md_sync_max
=
2276 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2280 sync_speed_show(mddev_t
*mddev
, char *page
)
2282 unsigned long resync
, dt
, db
;
2283 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2284 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2286 db
= resync
- (mddev
->resync_mark_cnt
);
2287 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2290 static struct md_sysfs_entry
2291 md_sync_speed
= __ATTR_RO(sync_speed
);
2294 sync_completed_show(mddev_t
*mddev
, char *page
)
2296 unsigned long max_blocks
, resync
;
2298 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2299 max_blocks
= mddev
->resync_max_sectors
;
2301 max_blocks
= mddev
->size
<< 1;
2303 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2304 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2307 static struct md_sysfs_entry
2308 md_sync_completed
= __ATTR_RO(sync_completed
);
2310 static struct attribute
*md_default_attrs
[] = {
2312 &md_raid_disks
.attr
,
2313 &md_chunk_size
.attr
,
2316 &md_new_device
.attr
,
2320 static struct attribute
*md_redundancy_attrs
[] = {
2322 &md_mismatches
.attr
,
2325 &md_sync_speed
.attr
,
2326 &md_sync_completed
.attr
,
2329 static struct attribute_group md_redundancy_group
= {
2331 .attrs
= md_redundancy_attrs
,
2336 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2338 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2339 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2345 rv
= entry
->show(mddev
, page
);
2346 mddev_unlock(mddev
);
2351 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2352 const char *page
, size_t length
)
2354 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2355 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2361 rv
= entry
->store(mddev
, page
, length
);
2362 mddev_unlock(mddev
);
2366 static void md_free(struct kobject
*ko
)
2368 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2372 static struct sysfs_ops md_sysfs_ops
= {
2373 .show
= md_attr_show
,
2374 .store
= md_attr_store
,
2376 static struct kobj_type md_ktype
= {
2378 .sysfs_ops
= &md_sysfs_ops
,
2379 .default_attrs
= md_default_attrs
,
2384 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2386 static DECLARE_MUTEX(disks_sem
);
2387 mddev_t
*mddev
= mddev_find(dev
);
2388 struct gendisk
*disk
;
2389 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2390 int shift
= partitioned
? MdpMinorShift
: 0;
2391 int unit
= MINOR(dev
) >> shift
;
2397 if (mddev
->gendisk
) {
2402 disk
= alloc_disk(1 << shift
);
2408 disk
->major
= MAJOR(dev
);
2409 disk
->first_minor
= unit
<< shift
;
2411 sprintf(disk
->disk_name
, "md_d%d", unit
);
2412 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2414 sprintf(disk
->disk_name
, "md%d", unit
);
2415 sprintf(disk
->devfs_name
, "md/%d", unit
);
2417 disk
->fops
= &md_fops
;
2418 disk
->private_data
= mddev
;
2419 disk
->queue
= mddev
->queue
;
2421 mddev
->gendisk
= disk
;
2423 mddev
->kobj
.parent
= &disk
->kobj
;
2424 mddev
->kobj
.k_name
= NULL
;
2425 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2426 mddev
->kobj
.ktype
= &md_ktype
;
2427 kobject_register(&mddev
->kobj
);
2431 void md_wakeup_thread(mdk_thread_t
*thread
);
2433 static void md_safemode_timeout(unsigned long data
)
2435 mddev_t
*mddev
= (mddev_t
*) data
;
2437 mddev
->safemode
= 1;
2438 md_wakeup_thread(mddev
->thread
);
2441 static int start_dirty_degraded
;
2443 static int do_md_run(mddev_t
* mddev
)
2447 struct list_head
*tmp
;
2449 struct gendisk
*disk
;
2450 struct mdk_personality
*pers
;
2451 char b
[BDEVNAME_SIZE
];
2453 if (list_empty(&mddev
->disks
))
2454 /* cannot run an array with no devices.. */
2461 * Analyze all RAID superblock(s)
2463 if (!mddev
->raid_disks
)
2466 chunk_size
= mddev
->chunk_size
;
2469 if (chunk_size
> MAX_CHUNK_SIZE
) {
2470 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2471 chunk_size
, MAX_CHUNK_SIZE
);
2475 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2477 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2478 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2481 if (chunk_size
< PAGE_SIZE
) {
2482 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2483 chunk_size
, PAGE_SIZE
);
2487 /* devices must have minimum size of one chunk */
2488 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2489 if (test_bit(Faulty
, &rdev
->flags
))
2491 if (rdev
->size
< chunk_size
/ 1024) {
2493 "md: Dev %s smaller than chunk_size:"
2495 bdevname(rdev
->bdev
,b
),
2496 (unsigned long long)rdev
->size
,
2504 if (mddev
->level
!= LEVEL_NONE
)
2505 request_module("md-level-%d", mddev
->level
);
2506 else if (mddev
->clevel
[0])
2507 request_module("md-%s", mddev
->clevel
);
2511 * Drop all container device buffers, from now on
2512 * the only valid external interface is through the md
2514 * Also find largest hardsector size
2516 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2517 if (test_bit(Faulty
, &rdev
->flags
))
2519 sync_blockdev(rdev
->bdev
);
2520 invalidate_bdev(rdev
->bdev
, 0);
2523 md_probe(mddev
->unit
, NULL
, NULL
);
2524 disk
= mddev
->gendisk
;
2528 spin_lock(&pers_lock
);
2529 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2530 if (!pers
|| !try_module_get(pers
->owner
)) {
2531 spin_unlock(&pers_lock
);
2532 if (mddev
->level
!= LEVEL_NONE
)
2533 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2536 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2541 spin_unlock(&pers_lock
);
2542 mddev
->level
= pers
->level
;
2543 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2545 mddev
->recovery
= 0;
2546 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2547 mddev
->barriers_work
= 1;
2548 mddev
->ok_start_degraded
= start_dirty_degraded
;
2551 mddev
->ro
= 2; /* read-only, but switch on first write */
2553 err
= mddev
->pers
->run(mddev
);
2554 if (!err
&& mddev
->pers
->sync_request
) {
2555 err
= bitmap_create(mddev
);
2557 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2558 mdname(mddev
), err
);
2559 mddev
->pers
->stop(mddev
);
2563 printk(KERN_ERR
"md: pers->run() failed ...\n");
2564 module_put(mddev
->pers
->owner
);
2566 bitmap_destroy(mddev
);
2569 if (mddev
->pers
->sync_request
)
2570 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2571 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2574 atomic_set(&mddev
->writes_pending
,0);
2575 mddev
->safemode
= 0;
2576 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2577 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2578 mddev
->safemode_delay
= (20 * HZ
)/1000 +1; /* 20 msec delay */
2581 ITERATE_RDEV(mddev
,rdev
,tmp
)
2582 if (rdev
->raid_disk
>= 0) {
2584 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2585 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2588 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2589 md_wakeup_thread(mddev
->thread
);
2591 if (mddev
->sb_dirty
)
2592 md_update_sb(mddev
);
2594 set_capacity(disk
, mddev
->array_size
<<1);
2596 /* If we call blk_queue_make_request here, it will
2597 * re-initialise max_sectors etc which may have been
2598 * refined inside -> run. So just set the bits we need to set.
2599 * Most initialisation happended when we called
2600 * blk_queue_make_request(..., md_fail_request)
2603 mddev
->queue
->queuedata
= mddev
;
2604 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2607 md_new_event(mddev
);
2611 static int restart_array(mddev_t
*mddev
)
2613 struct gendisk
*disk
= mddev
->gendisk
;
2617 * Complain if it has no devices
2620 if (list_empty(&mddev
->disks
))
2628 mddev
->safemode
= 0;
2630 set_disk_ro(disk
, 0);
2632 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2635 * Kick recovery or resync if necessary
2637 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2638 md_wakeup_thread(mddev
->thread
);
2641 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2650 static int do_md_stop(mddev_t
* mddev
, int ro
)
2653 struct gendisk
*disk
= mddev
->gendisk
;
2656 if (atomic_read(&mddev
->active
)>2) {
2657 printk("md: %s still in use.\n",mdname(mddev
));
2661 if (mddev
->sync_thread
) {
2662 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2663 md_unregister_thread(mddev
->sync_thread
);
2664 mddev
->sync_thread
= NULL
;
2667 del_timer_sync(&mddev
->safemode_timer
);
2669 invalidate_partition(disk
, 0);
2677 bitmap_flush(mddev
);
2678 md_super_wait(mddev
);
2680 set_disk_ro(disk
, 0);
2681 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2682 mddev
->pers
->stop(mddev
);
2683 if (mddev
->pers
->sync_request
)
2684 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2686 module_put(mddev
->pers
->owner
);
2691 if (!mddev
->in_sync
) {
2692 /* mark array as shutdown cleanly */
2694 md_update_sb(mddev
);
2697 set_disk_ro(disk
, 1);
2701 * Free resources if final stop
2705 struct list_head
*tmp
;
2706 struct gendisk
*disk
;
2707 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
2709 bitmap_destroy(mddev
);
2710 if (mddev
->bitmap_file
) {
2711 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
2712 fput(mddev
->bitmap_file
);
2713 mddev
->bitmap_file
= NULL
;
2715 mddev
->bitmap_offset
= 0;
2717 ITERATE_RDEV(mddev
,rdev
,tmp
)
2718 if (rdev
->raid_disk
>= 0) {
2720 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2721 sysfs_remove_link(&mddev
->kobj
, nm
);
2724 export_array(mddev
);
2726 mddev
->array_size
= 0;
2727 disk
= mddev
->gendisk
;
2729 set_capacity(disk
, 0);
2732 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
2735 md_new_event(mddev
);
2740 static void autorun_array(mddev_t
*mddev
)
2743 struct list_head
*tmp
;
2746 if (list_empty(&mddev
->disks
))
2749 printk(KERN_INFO
"md: running: ");
2751 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2752 char b
[BDEVNAME_SIZE
];
2753 printk("<%s>", bdevname(rdev
->bdev
,b
));
2757 err
= do_md_run (mddev
);
2759 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
2760 do_md_stop (mddev
, 0);
2765 * lets try to run arrays based on all disks that have arrived
2766 * until now. (those are in pending_raid_disks)
2768 * the method: pick the first pending disk, collect all disks with
2769 * the same UUID, remove all from the pending list and put them into
2770 * the 'same_array' list. Then order this list based on superblock
2771 * update time (freshest comes first), kick out 'old' disks and
2772 * compare superblocks. If everything's fine then run it.
2774 * If "unit" is allocated, then bump its reference count
2776 static void autorun_devices(int part
)
2778 struct list_head candidates
;
2779 struct list_head
*tmp
;
2780 mdk_rdev_t
*rdev0
, *rdev
;
2782 char b
[BDEVNAME_SIZE
];
2784 printk(KERN_INFO
"md: autorun ...\n");
2785 while (!list_empty(&pending_raid_disks
)) {
2787 rdev0
= list_entry(pending_raid_disks
.next
,
2788 mdk_rdev_t
, same_set
);
2790 printk(KERN_INFO
"md: considering %s ...\n",
2791 bdevname(rdev0
->bdev
,b
));
2792 INIT_LIST_HEAD(&candidates
);
2793 ITERATE_RDEV_PENDING(rdev
,tmp
)
2794 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
2795 printk(KERN_INFO
"md: adding %s ...\n",
2796 bdevname(rdev
->bdev
,b
));
2797 list_move(&rdev
->same_set
, &candidates
);
2800 * now we have a set of devices, with all of them having
2801 * mostly sane superblocks. It's time to allocate the
2804 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
2805 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
2806 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
2810 dev
= MKDEV(mdp_major
,
2811 rdev0
->preferred_minor
<< MdpMinorShift
);
2813 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
2815 md_probe(dev
, NULL
, NULL
);
2816 mddev
= mddev_find(dev
);
2819 "md: cannot allocate memory for md drive.\n");
2822 if (mddev_lock(mddev
))
2823 printk(KERN_WARNING
"md: %s locked, cannot run\n",
2825 else if (mddev
->raid_disks
|| mddev
->major_version
2826 || !list_empty(&mddev
->disks
)) {
2828 "md: %s already running, cannot run %s\n",
2829 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
2830 mddev_unlock(mddev
);
2832 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
2833 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
2834 list_del_init(&rdev
->same_set
);
2835 if (bind_rdev_to_array(rdev
, mddev
))
2838 autorun_array(mddev
);
2839 mddev_unlock(mddev
);
2841 /* on success, candidates will be empty, on error
2844 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
2848 printk(KERN_INFO
"md: ... autorun DONE.\n");
2852 * import RAID devices based on one partition
2853 * if possible, the array gets run as well.
2856 static int autostart_array(dev_t startdev
)
2858 char b
[BDEVNAME_SIZE
];
2859 int err
= -EINVAL
, i
;
2860 mdp_super_t
*sb
= NULL
;
2861 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
2863 start_rdev
= md_import_device(startdev
, 0, 0);
2864 if (IS_ERR(start_rdev
))
2868 /* NOTE: this can only work for 0.90.0 superblocks */
2869 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
2870 if (sb
->major_version
!= 0 ||
2871 sb
->minor_version
!= 90 ) {
2872 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
2873 export_rdev(start_rdev
);
2877 if (test_bit(Faulty
, &start_rdev
->flags
)) {
2879 "md: can not autostart based on faulty %s!\n",
2880 bdevname(start_rdev
->bdev
,b
));
2881 export_rdev(start_rdev
);
2884 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
2886 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2887 mdp_disk_t
*desc
= sb
->disks
+ i
;
2888 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
2892 if (dev
== startdev
)
2894 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
2896 rdev
= md_import_device(dev
, 0, 0);
2900 list_add(&rdev
->same_set
, &pending_raid_disks
);
2904 * possibly return codes
2912 static int get_version(void __user
* arg
)
2916 ver
.major
= MD_MAJOR_VERSION
;
2917 ver
.minor
= MD_MINOR_VERSION
;
2918 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
2920 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
2926 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
2928 mdu_array_info_t info
;
2929 int nr
,working
,active
,failed
,spare
;
2931 struct list_head
*tmp
;
2933 nr
=working
=active
=failed
=spare
=0;
2934 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2936 if (test_bit(Faulty
, &rdev
->flags
))
2940 if (test_bit(In_sync
, &rdev
->flags
))
2947 info
.major_version
= mddev
->major_version
;
2948 info
.minor_version
= mddev
->minor_version
;
2949 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
2950 info
.ctime
= mddev
->ctime
;
2951 info
.level
= mddev
->level
;
2952 info
.size
= mddev
->size
;
2953 if (info
.size
!= mddev
->size
) /* overflow */
2956 info
.raid_disks
= mddev
->raid_disks
;
2957 info
.md_minor
= mddev
->md_minor
;
2958 info
.not_persistent
= !mddev
->persistent
;
2960 info
.utime
= mddev
->utime
;
2963 info
.state
= (1<<MD_SB_CLEAN
);
2964 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
2965 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
2966 info
.active_disks
= active
;
2967 info
.working_disks
= working
;
2968 info
.failed_disks
= failed
;
2969 info
.spare_disks
= spare
;
2971 info
.layout
= mddev
->layout
;
2972 info
.chunk_size
= mddev
->chunk_size
;
2974 if (copy_to_user(arg
, &info
, sizeof(info
)))
2980 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
2982 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
2983 char *ptr
, *buf
= NULL
;
2986 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
2990 /* bitmap disabled, zero the first byte and copy out */
2991 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
2992 file
->pathname
[0] = '\0';
2996 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3000 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3004 strcpy(file
->pathname
, ptr
);
3008 if (copy_to_user(arg
, file
, sizeof(*file
)))
3016 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3018 mdu_disk_info_t info
;
3022 if (copy_from_user(&info
, arg
, sizeof(info
)))
3027 rdev
= find_rdev_nr(mddev
, nr
);
3029 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3030 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3031 info
.raid_disk
= rdev
->raid_disk
;
3033 if (test_bit(Faulty
, &rdev
->flags
))
3034 info
.state
|= (1<<MD_DISK_FAULTY
);
3035 else if (test_bit(In_sync
, &rdev
->flags
)) {
3036 info
.state
|= (1<<MD_DISK_ACTIVE
);
3037 info
.state
|= (1<<MD_DISK_SYNC
);
3039 if (test_bit(WriteMostly
, &rdev
->flags
))
3040 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3042 info
.major
= info
.minor
= 0;
3043 info
.raid_disk
= -1;
3044 info
.state
= (1<<MD_DISK_REMOVED
);
3047 if (copy_to_user(arg
, &info
, sizeof(info
)))
3053 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3055 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3057 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3059 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3062 if (!mddev
->raid_disks
) {
3064 /* expecting a device which has a superblock */
3065 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3068 "md: md_import_device returned %ld\n",
3070 return PTR_ERR(rdev
);
3072 if (!list_empty(&mddev
->disks
)) {
3073 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3074 mdk_rdev_t
, same_set
);
3075 int err
= super_types
[mddev
->major_version
]
3076 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3079 "md: %s has different UUID to %s\n",
3080 bdevname(rdev
->bdev
,b
),
3081 bdevname(rdev0
->bdev
,b2
));
3086 err
= bind_rdev_to_array(rdev
, mddev
);
3093 * add_new_disk can be used once the array is assembled
3094 * to add "hot spares". They must already have a superblock
3099 if (!mddev
->pers
->hot_add_disk
) {
3101 "%s: personality does not support diskops!\n",
3105 if (mddev
->persistent
)
3106 rdev
= md_import_device(dev
, mddev
->major_version
,
3107 mddev
->minor_version
);
3109 rdev
= md_import_device(dev
, -1, -1);
3112 "md: md_import_device returned %ld\n",
3114 return PTR_ERR(rdev
);
3116 /* set save_raid_disk if appropriate */
3117 if (!mddev
->persistent
) {
3118 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3119 info
->raid_disk
< mddev
->raid_disks
)
3120 rdev
->raid_disk
= info
->raid_disk
;
3122 rdev
->raid_disk
= -1;
3124 super_types
[mddev
->major_version
].
3125 validate_super(mddev
, rdev
);
3126 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3128 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3129 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3130 set_bit(WriteMostly
, &rdev
->flags
);
3132 rdev
->raid_disk
= -1;
3133 err
= bind_rdev_to_array(rdev
, mddev
);
3137 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3138 md_wakeup_thread(mddev
->thread
);
3142 /* otherwise, add_new_disk is only allowed
3143 * for major_version==0 superblocks
3145 if (mddev
->major_version
!= 0) {
3146 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3151 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3153 rdev
= md_import_device (dev
, -1, 0);
3156 "md: error, md_import_device() returned %ld\n",
3158 return PTR_ERR(rdev
);
3160 rdev
->desc_nr
= info
->number
;
3161 if (info
->raid_disk
< mddev
->raid_disks
)
3162 rdev
->raid_disk
= info
->raid_disk
;
3164 rdev
->raid_disk
= -1;
3168 if (rdev
->raid_disk
< mddev
->raid_disks
)
3169 if (info
->state
& (1<<MD_DISK_SYNC
))
3170 set_bit(In_sync
, &rdev
->flags
);
3172 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3173 set_bit(WriteMostly
, &rdev
->flags
);
3175 if (!mddev
->persistent
) {
3176 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3177 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3179 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3180 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3182 err
= bind_rdev_to_array(rdev
, mddev
);
3192 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3194 char b
[BDEVNAME_SIZE
];
3200 rdev
= find_rdev(mddev
, dev
);
3204 if (rdev
->raid_disk
>= 0)
3207 kick_rdev_from_array(rdev
);
3208 md_update_sb(mddev
);
3209 md_new_event(mddev
);
3213 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3214 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3218 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3220 char b
[BDEVNAME_SIZE
];
3228 if (mddev
->major_version
!= 0) {
3229 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3230 " version-0 superblocks.\n",
3234 if (!mddev
->pers
->hot_add_disk
) {
3236 "%s: personality does not support diskops!\n",
3241 rdev
= md_import_device (dev
, -1, 0);
3244 "md: error, md_import_device() returned %ld\n",
3249 if (mddev
->persistent
)
3250 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3253 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3255 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3258 if (test_bit(Faulty
, &rdev
->flags
)) {
3260 "md: can not hot-add faulty %s disk to %s!\n",
3261 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3265 clear_bit(In_sync
, &rdev
->flags
);
3267 err
= bind_rdev_to_array(rdev
, mddev
);
3272 * The rest should better be atomic, we can have disk failures
3273 * noticed in interrupt contexts ...
3276 if (rdev
->desc_nr
== mddev
->max_disks
) {
3277 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3280 goto abort_unbind_export
;
3283 rdev
->raid_disk
= -1;
3285 md_update_sb(mddev
);
3288 * Kick recovery, maybe this spare has to be added to the
3289 * array immediately.
3291 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3292 md_wakeup_thread(mddev
->thread
);
3293 md_new_event(mddev
);
3296 abort_unbind_export
:
3297 unbind_rdev_from_array(rdev
);
3304 /* similar to deny_write_access, but accounts for our holding a reference
3305 * to the file ourselves */
3306 static int deny_bitmap_write_access(struct file
* file
)
3308 struct inode
*inode
= file
->f_mapping
->host
;
3310 spin_lock(&inode
->i_lock
);
3311 if (atomic_read(&inode
->i_writecount
) > 1) {
3312 spin_unlock(&inode
->i_lock
);
3315 atomic_set(&inode
->i_writecount
, -1);
3316 spin_unlock(&inode
->i_lock
);
3321 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3326 if (!mddev
->pers
->quiesce
)
3328 if (mddev
->recovery
|| mddev
->sync_thread
)
3330 /* we should be able to change the bitmap.. */
3336 return -EEXIST
; /* cannot add when bitmap is present */
3337 mddev
->bitmap_file
= fget(fd
);
3339 if (mddev
->bitmap_file
== NULL
) {
3340 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3345 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3347 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3349 fput(mddev
->bitmap_file
);
3350 mddev
->bitmap_file
= NULL
;
3353 mddev
->bitmap_offset
= 0; /* file overrides offset */
3354 } else if (mddev
->bitmap
== NULL
)
3355 return -ENOENT
; /* cannot remove what isn't there */
3358 mddev
->pers
->quiesce(mddev
, 1);
3360 err
= bitmap_create(mddev
);
3362 bitmap_destroy(mddev
);
3363 mddev
->pers
->quiesce(mddev
, 0);
3364 } else if (fd
< 0) {
3365 if (mddev
->bitmap_file
)
3366 fput(mddev
->bitmap_file
);
3367 mddev
->bitmap_file
= NULL
;
3374 * set_array_info is used two different ways
3375 * The original usage is when creating a new array.
3376 * In this usage, raid_disks is > 0 and it together with
3377 * level, size, not_persistent,layout,chunksize determine the
3378 * shape of the array.
3379 * This will always create an array with a type-0.90.0 superblock.
3380 * The newer usage is when assembling an array.
3381 * In this case raid_disks will be 0, and the major_version field is
3382 * use to determine which style super-blocks are to be found on the devices.
3383 * The minor and patch _version numbers are also kept incase the
3384 * super_block handler wishes to interpret them.
3386 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3389 if (info
->raid_disks
== 0) {
3390 /* just setting version number for superblock loading */
3391 if (info
->major_version
< 0 ||
3392 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3393 super_types
[info
->major_version
].name
== NULL
) {
3394 /* maybe try to auto-load a module? */
3396 "md: superblock version %d not known\n",
3397 info
->major_version
);
3400 mddev
->major_version
= info
->major_version
;
3401 mddev
->minor_version
= info
->minor_version
;
3402 mddev
->patch_version
= info
->patch_version
;
3405 mddev
->major_version
= MD_MAJOR_VERSION
;
3406 mddev
->minor_version
= MD_MINOR_VERSION
;
3407 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3408 mddev
->ctime
= get_seconds();
3410 mddev
->level
= info
->level
;
3411 mddev
->clevel
[0] = 0;
3412 mddev
->size
= info
->size
;
3413 mddev
->raid_disks
= info
->raid_disks
;
3414 /* don't set md_minor, it is determined by which /dev/md* was
3417 if (info
->state
& (1<<MD_SB_CLEAN
))
3418 mddev
->recovery_cp
= MaxSector
;
3420 mddev
->recovery_cp
= 0;
3421 mddev
->persistent
= ! info
->not_persistent
;
3423 mddev
->layout
= info
->layout
;
3424 mddev
->chunk_size
= info
->chunk_size
;
3426 mddev
->max_disks
= MD_SB_DISKS
;
3428 mddev
->sb_dirty
= 1;
3430 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3431 mddev
->bitmap_offset
= 0;
3434 * Generate a 128 bit UUID
3436 get_random_bytes(mddev
->uuid
, 16);
3441 static int update_size(mddev_t
*mddev
, unsigned long size
)
3445 struct list_head
*tmp
;
3447 if (mddev
->pers
->resize
== NULL
)
3449 /* The "size" is the amount of each device that is used.
3450 * This can only make sense for arrays with redundancy.
3451 * linear and raid0 always use whatever space is available
3452 * We can only consider changing the size if no resync
3453 * or reconstruction is happening, and if the new size
3454 * is acceptable. It must fit before the sb_offset or,
3455 * if that is <data_offset, it must fit before the
3456 * size of each device.
3457 * If size is zero, we find the largest size that fits.
3459 if (mddev
->sync_thread
)
3461 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3463 int fit
= (size
== 0);
3464 if (rdev
->sb_offset
> rdev
->data_offset
)
3465 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3467 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3468 - rdev
->data_offset
;
3469 if (fit
&& (size
== 0 || size
> avail
/2))
3471 if (avail
< ((sector_t
)size
<< 1))
3474 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3476 struct block_device
*bdev
;
3478 bdev
= bdget_disk(mddev
->gendisk
, 0);
3480 mutex_lock(&bdev
->bd_inode
->i_mutex
);
3481 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
3482 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
3489 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3492 /* change the number of raid disks */
3493 if (mddev
->pers
->reshape
== NULL
)
3495 if (raid_disks
<= 0 ||
3496 raid_disks
>= mddev
->max_disks
)
3498 if (mddev
->sync_thread
)
3500 rv
= mddev
->pers
->reshape(mddev
, raid_disks
);
3506 * update_array_info is used to change the configuration of an
3508 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3509 * fields in the info are checked against the array.
3510 * Any differences that cannot be handled will cause an error.
3511 * Normally, only one change can be managed at a time.
3513 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3519 /* calculate expected state,ignoring low bits */
3520 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3521 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3523 if (mddev
->major_version
!= info
->major_version
||
3524 mddev
->minor_version
!= info
->minor_version
||
3525 /* mddev->patch_version != info->patch_version || */
3526 mddev
->ctime
!= info
->ctime
||
3527 mddev
->level
!= info
->level
||
3528 /* mddev->layout != info->layout || */
3529 !mddev
->persistent
!= info
->not_persistent
||
3530 mddev
->chunk_size
!= info
->chunk_size
||
3531 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3532 ((state
^info
->state
) & 0xfffffe00)
3535 /* Check there is only one change */
3536 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
3537 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3538 if (mddev
->layout
!= info
->layout
) cnt
++;
3539 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3540 if (cnt
== 0) return 0;
3541 if (cnt
> 1) return -EINVAL
;
3543 if (mddev
->layout
!= info
->layout
) {
3545 * we don't need to do anything at the md level, the
3546 * personality will take care of it all.
3548 if (mddev
->pers
->reconfig
== NULL
)
3551 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3553 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
3554 rv
= update_size(mddev
, info
->size
);
3556 if (mddev
->raid_disks
!= info
->raid_disks
)
3557 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3559 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3560 if (mddev
->pers
->quiesce
== NULL
)
3562 if (mddev
->recovery
|| mddev
->sync_thread
)
3564 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3565 /* add the bitmap */
3568 if (mddev
->default_bitmap_offset
== 0)
3570 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3571 mddev
->pers
->quiesce(mddev
, 1);
3572 rv
= bitmap_create(mddev
);
3574 bitmap_destroy(mddev
);
3575 mddev
->pers
->quiesce(mddev
, 0);
3577 /* remove the bitmap */
3580 if (mddev
->bitmap
->file
)
3582 mddev
->pers
->quiesce(mddev
, 1);
3583 bitmap_destroy(mddev
);
3584 mddev
->pers
->quiesce(mddev
, 0);
3585 mddev
->bitmap_offset
= 0;
3588 md_update_sb(mddev
);
3592 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3596 if (mddev
->pers
== NULL
)
3599 rdev
= find_rdev(mddev
, dev
);
3603 md_error(mddev
, rdev
);
3607 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
3609 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
3613 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
3617 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3618 unsigned int cmd
, unsigned long arg
)
3621 void __user
*argp
= (void __user
*)arg
;
3622 mddev_t
*mddev
= NULL
;
3624 if (!capable(CAP_SYS_ADMIN
))
3628 * Commands dealing with the RAID driver but not any
3634 err
= get_version(argp
);
3637 case PRINT_RAID_DEBUG
:
3645 autostart_arrays(arg
);
3652 * Commands creating/starting a new array:
3655 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3663 if (cmd
== START_ARRAY
) {
3664 /* START_ARRAY doesn't need to lock the array as autostart_array
3665 * does the locking, and it could even be a different array
3670 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3671 "This will not be supported beyond July 2006\n",
3672 current
->comm
, current
->pid
);
3675 err
= autostart_array(new_decode_dev(arg
));
3677 printk(KERN_WARNING
"md: autostart failed!\n");
3683 err
= mddev_lock(mddev
);
3686 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3693 case SET_ARRAY_INFO
:
3695 mdu_array_info_t info
;
3697 memset(&info
, 0, sizeof(info
));
3698 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
3703 err
= update_array_info(mddev
, &info
);
3705 printk(KERN_WARNING
"md: couldn't update"
3706 " array info. %d\n", err
);
3711 if (!list_empty(&mddev
->disks
)) {
3713 "md: array %s already has disks!\n",
3718 if (mddev
->raid_disks
) {
3720 "md: array %s already initialised!\n",
3725 err
= set_array_info(mddev
, &info
);
3727 printk(KERN_WARNING
"md: couldn't set"
3728 " array info. %d\n", err
);
3738 * Commands querying/configuring an existing array:
3740 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3741 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3742 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
3743 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
3749 * Commands even a read-only array can execute:
3753 case GET_ARRAY_INFO
:
3754 err
= get_array_info(mddev
, argp
);
3757 case GET_BITMAP_FILE
:
3758 err
= get_bitmap_file(mddev
, argp
);
3762 err
= get_disk_info(mddev
, argp
);
3765 case RESTART_ARRAY_RW
:
3766 err
= restart_array(mddev
);
3770 err
= do_md_stop (mddev
, 0);
3774 err
= do_md_stop (mddev
, 1);
3778 * We have a problem here : there is no easy way to give a CHS
3779 * virtual geometry. We currently pretend that we have a 2 heads
3780 * 4 sectors (with a BIG number of cylinders...). This drives
3781 * dosfs just mad... ;-)
3786 * The remaining ioctls are changing the state of the
3787 * superblock, so we do not allow them on read-only arrays.
3788 * However non-MD ioctls (e.g. get-size) will still come through
3789 * here and hit the 'default' below, so only disallow
3790 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3792 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
3793 mddev
->ro
&& mddev
->pers
) {
3794 if (mddev
->ro
== 2) {
3796 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3797 md_wakeup_thread(mddev
->thread
);
3809 mdu_disk_info_t info
;
3810 if (copy_from_user(&info
, argp
, sizeof(info
)))
3813 err
= add_new_disk(mddev
, &info
);
3817 case HOT_REMOVE_DISK
:
3818 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
3822 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
3825 case SET_DISK_FAULTY
:
3826 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
3830 err
= do_md_run (mddev
);
3833 case SET_BITMAP_FILE
:
3834 err
= set_bitmap_file(mddev
, (int)arg
);
3838 if (_IOC_TYPE(cmd
) == MD_MAJOR
)
3839 printk(KERN_WARNING
"md: %s(pid %d) used"
3840 " obsolete MD ioctl, upgrade your"
3841 " software to use new ictls.\n",
3842 current
->comm
, current
->pid
);
3849 mddev_unlock(mddev
);
3859 static int md_open(struct inode
*inode
, struct file
*file
)
3862 * Succeed if we can lock the mddev, which confirms that
3863 * it isn't being stopped right now.
3865 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3868 if ((err
= mddev_lock(mddev
)))
3873 mddev_unlock(mddev
);
3875 check_disk_change(inode
->i_bdev
);
3880 static int md_release(struct inode
*inode
, struct file
* file
)
3882 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3891 static int md_media_changed(struct gendisk
*disk
)
3893 mddev_t
*mddev
= disk
->private_data
;
3895 return mddev
->changed
;
3898 static int md_revalidate(struct gendisk
*disk
)
3900 mddev_t
*mddev
= disk
->private_data
;
3905 static struct block_device_operations md_fops
=
3907 .owner
= THIS_MODULE
,
3909 .release
= md_release
,
3911 .getgeo
= md_getgeo
,
3912 .media_changed
= md_media_changed
,
3913 .revalidate_disk
= md_revalidate
,
3916 static int md_thread(void * arg
)
3918 mdk_thread_t
*thread
= arg
;
3921 * md_thread is a 'system-thread', it's priority should be very
3922 * high. We avoid resource deadlocks individually in each
3923 * raid personality. (RAID5 does preallocation) We also use RR and
3924 * the very same RT priority as kswapd, thus we will never get
3925 * into a priority inversion deadlock.
3927 * we definitely have to have equal or higher priority than
3928 * bdflush, otherwise bdflush will deadlock if there are too
3929 * many dirty RAID5 blocks.
3932 allow_signal(SIGKILL
);
3933 while (!kthread_should_stop()) {
3935 /* We need to wait INTERRUPTIBLE so that
3936 * we don't add to the load-average.
3937 * That means we need to be sure no signals are
3940 if (signal_pending(current
))
3941 flush_signals(current
);
3943 wait_event_interruptible_timeout
3945 test_bit(THREAD_WAKEUP
, &thread
->flags
)
3946 || kthread_should_stop(),
3950 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
3952 thread
->run(thread
->mddev
);
3958 void md_wakeup_thread(mdk_thread_t
*thread
)
3961 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
3962 set_bit(THREAD_WAKEUP
, &thread
->flags
);
3963 wake_up(&thread
->wqueue
);
3967 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
3970 mdk_thread_t
*thread
;
3972 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
3976 init_waitqueue_head(&thread
->wqueue
);
3979 thread
->mddev
= mddev
;
3980 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
3981 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
3982 if (IS_ERR(thread
->tsk
)) {
3989 void md_unregister_thread(mdk_thread_t
*thread
)
3991 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
3993 kthread_stop(thread
->tsk
);
3997 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4004 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4007 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4009 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4010 __builtin_return_address(0),__builtin_return_address(1),
4011 __builtin_return_address(2),__builtin_return_address(3));
4013 if (!mddev
->pers
->error_handler
)
4015 mddev
->pers
->error_handler(mddev
,rdev
);
4016 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4017 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4018 md_wakeup_thread(mddev
->thread
);
4019 md_new_event(mddev
);
4022 /* seq_file implementation /proc/mdstat */
4024 static void status_unused(struct seq_file
*seq
)
4028 struct list_head
*tmp
;
4030 seq_printf(seq
, "unused devices: ");
4032 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4033 char b
[BDEVNAME_SIZE
];
4035 seq_printf(seq
, "%s ",
4036 bdevname(rdev
->bdev
,b
));
4039 seq_printf(seq
, "<none>");
4041 seq_printf(seq
, "\n");
4045 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4047 unsigned long max_blocks
, resync
, res
, dt
, db
, rt
;
4049 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4051 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4052 max_blocks
= mddev
->resync_max_sectors
>> 1;
4054 max_blocks
= mddev
->size
;
4057 * Should not happen.
4063 res
= (resync
/1024)*1000/(max_blocks
/1024 + 1);
4065 int i
, x
= res
/50, y
= 20-x
;
4066 seq_printf(seq
, "[");
4067 for (i
= 0; i
< x
; i
++)
4068 seq_printf(seq
, "=");
4069 seq_printf(seq
, ">");
4070 for (i
= 0; i
< y
; i
++)
4071 seq_printf(seq
, ".");
4072 seq_printf(seq
, "] ");
4074 seq_printf(seq
, " %s =%3lu.%lu%% (%lu/%lu)",
4075 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4076 "resync" : "recovery"),
4077 res
/10, res
% 10, resync
, max_blocks
);
4080 * We do not want to overflow, so the order of operands and
4081 * the * 100 / 100 trick are important. We do a +1 to be
4082 * safe against division by zero. We only estimate anyway.
4084 * dt: time from mark until now
4085 * db: blocks written from mark until now
4086 * rt: remaining time
4088 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4090 db
= resync
- (mddev
->resync_mark_cnt
/2);
4091 rt
= (dt
* ((max_blocks
-resync
) / (db
/100+1)))/100;
4093 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4095 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4098 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4100 struct list_head
*tmp
;
4110 spin_lock(&all_mddevs_lock
);
4111 list_for_each(tmp
,&all_mddevs
)
4113 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4115 spin_unlock(&all_mddevs_lock
);
4118 spin_unlock(&all_mddevs_lock
);
4120 return (void*)2;/* tail */
4124 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4126 struct list_head
*tmp
;
4127 mddev_t
*next_mddev
, *mddev
= v
;
4133 spin_lock(&all_mddevs_lock
);
4135 tmp
= all_mddevs
.next
;
4137 tmp
= mddev
->all_mddevs
.next
;
4138 if (tmp
!= &all_mddevs
)
4139 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4141 next_mddev
= (void*)2;
4144 spin_unlock(&all_mddevs_lock
);
4152 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4156 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4160 struct mdstat_info
{
4164 static int md_seq_show(struct seq_file
*seq
, void *v
)
4168 struct list_head
*tmp2
;
4170 struct mdstat_info
*mi
= seq
->private;
4171 struct bitmap
*bitmap
;
4173 if (v
== (void*)1) {
4174 struct mdk_personality
*pers
;
4175 seq_printf(seq
, "Personalities : ");
4176 spin_lock(&pers_lock
);
4177 list_for_each_entry(pers
, &pers_list
, list
)
4178 seq_printf(seq
, "[%s] ", pers
->name
);
4180 spin_unlock(&pers_lock
);
4181 seq_printf(seq
, "\n");
4182 mi
->event
= atomic_read(&md_event_count
);
4185 if (v
== (void*)2) {
4190 if (mddev_lock(mddev
)!=0)
4192 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4193 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4194 mddev
->pers
? "" : "in");
4197 seq_printf(seq
, " (read-only)");
4199 seq_printf(seq
, "(auto-read-only)");
4200 seq_printf(seq
, " %s", mddev
->pers
->name
);
4204 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4205 char b
[BDEVNAME_SIZE
];
4206 seq_printf(seq
, " %s[%d]",
4207 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4208 if (test_bit(WriteMostly
, &rdev
->flags
))
4209 seq_printf(seq
, "(W)");
4210 if (test_bit(Faulty
, &rdev
->flags
)) {
4211 seq_printf(seq
, "(F)");
4213 } else if (rdev
->raid_disk
< 0)
4214 seq_printf(seq
, "(S)"); /* spare */
4218 if (!list_empty(&mddev
->disks
)) {
4220 seq_printf(seq
, "\n %llu blocks",
4221 (unsigned long long)mddev
->array_size
);
4223 seq_printf(seq
, "\n %llu blocks",
4224 (unsigned long long)size
);
4226 if (mddev
->persistent
) {
4227 if (mddev
->major_version
!= 0 ||
4228 mddev
->minor_version
!= 90) {
4229 seq_printf(seq
," super %d.%d",
4230 mddev
->major_version
,
4231 mddev
->minor_version
);
4234 seq_printf(seq
, " super non-persistent");
4237 mddev
->pers
->status (seq
, mddev
);
4238 seq_printf(seq
, "\n ");
4239 if (mddev
->pers
->sync_request
) {
4240 if (mddev
->curr_resync
> 2) {
4241 status_resync (seq
, mddev
);
4242 seq_printf(seq
, "\n ");
4243 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4244 seq_printf(seq
, "\tresync=DELAYED\n ");
4245 else if (mddev
->recovery_cp
< MaxSector
)
4246 seq_printf(seq
, "\tresync=PENDING\n ");
4249 seq_printf(seq
, "\n ");
4251 if ((bitmap
= mddev
->bitmap
)) {
4252 unsigned long chunk_kb
;
4253 unsigned long flags
;
4254 spin_lock_irqsave(&bitmap
->lock
, flags
);
4255 chunk_kb
= bitmap
->chunksize
>> 10;
4256 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4258 bitmap
->pages
- bitmap
->missing_pages
,
4260 (bitmap
->pages
- bitmap
->missing_pages
)
4261 << (PAGE_SHIFT
- 10),
4262 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4263 chunk_kb
? "KB" : "B");
4265 seq_printf(seq
, ", file: ");
4266 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4267 bitmap
->file
->f_dentry
," \t\n");
4270 seq_printf(seq
, "\n");
4271 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4274 seq_printf(seq
, "\n");
4276 mddev_unlock(mddev
);
4281 static struct seq_operations md_seq_ops
= {
4282 .start
= md_seq_start
,
4283 .next
= md_seq_next
,
4284 .stop
= md_seq_stop
,
4285 .show
= md_seq_show
,
4288 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4291 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4295 error
= seq_open(file
, &md_seq_ops
);
4299 struct seq_file
*p
= file
->private_data
;
4301 mi
->event
= atomic_read(&md_event_count
);
4306 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4308 struct seq_file
*m
= file
->private_data
;
4309 struct mdstat_info
*mi
= m
->private;
4312 return seq_release(inode
, file
);
4315 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4317 struct seq_file
*m
= filp
->private_data
;
4318 struct mdstat_info
*mi
= m
->private;
4321 poll_wait(filp
, &md_event_waiters
, wait
);
4323 /* always allow read */
4324 mask
= POLLIN
| POLLRDNORM
;
4326 if (mi
->event
!= atomic_read(&md_event_count
))
4327 mask
|= POLLERR
| POLLPRI
;
4331 static struct file_operations md_seq_fops
= {
4332 .open
= md_seq_open
,
4334 .llseek
= seq_lseek
,
4335 .release
= md_seq_release
,
4336 .poll
= mdstat_poll
,
4339 int register_md_personality(struct mdk_personality
*p
)
4341 spin_lock(&pers_lock
);
4342 list_add_tail(&p
->list
, &pers_list
);
4343 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4344 spin_unlock(&pers_lock
);
4348 int unregister_md_personality(struct mdk_personality
*p
)
4350 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4351 spin_lock(&pers_lock
);
4352 list_del_init(&p
->list
);
4353 spin_unlock(&pers_lock
);
4357 static int is_mddev_idle(mddev_t
*mddev
)
4360 struct list_head
*tmp
;
4362 unsigned long curr_events
;
4365 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4366 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4367 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4368 disk_stat_read(disk
, sectors
[1]) -
4369 atomic_read(&disk
->sync_io
);
4370 /* The difference between curr_events and last_events
4371 * will be affected by any new non-sync IO (making
4372 * curr_events bigger) and any difference in the amount of
4373 * in-flight syncio (making current_events bigger or smaller)
4374 * The amount in-flight is currently limited to
4375 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4376 * which is at most 4096 sectors.
4377 * These numbers are fairly fragile and should be made
4378 * more robust, probably by enforcing the
4379 * 'window size' that md_do_sync sort-of uses.
4381 * Note: the following is an unsigned comparison.
4383 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4384 rdev
->last_events
= curr_events
;
4391 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4393 /* another "blocks" (512byte) blocks have been synced */
4394 atomic_sub(blocks
, &mddev
->recovery_active
);
4395 wake_up(&mddev
->recovery_wait
);
4397 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4398 md_wakeup_thread(mddev
->thread
);
4399 // stop recovery, signal do_sync ....
4404 /* md_write_start(mddev, bi)
4405 * If we need to update some array metadata (e.g. 'active' flag
4406 * in superblock) before writing, schedule a superblock update
4407 * and wait for it to complete.
4409 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4411 if (bio_data_dir(bi
) != WRITE
)
4414 BUG_ON(mddev
->ro
== 1);
4415 if (mddev
->ro
== 2) {
4416 /* need to switch to read/write */
4418 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4419 md_wakeup_thread(mddev
->thread
);
4421 atomic_inc(&mddev
->writes_pending
);
4422 if (mddev
->in_sync
) {
4423 spin_lock_irq(&mddev
->write_lock
);
4424 if (mddev
->in_sync
) {
4426 mddev
->sb_dirty
= 1;
4427 md_wakeup_thread(mddev
->thread
);
4429 spin_unlock_irq(&mddev
->write_lock
);
4431 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4434 void md_write_end(mddev_t
*mddev
)
4436 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4437 if (mddev
->safemode
== 2)
4438 md_wakeup_thread(mddev
->thread
);
4440 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4444 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4446 #define SYNC_MARKS 10
4447 #define SYNC_MARK_STEP (3*HZ)
4448 static void md_do_sync(mddev_t
*mddev
)
4451 unsigned int currspeed
= 0,
4453 sector_t max_sectors
,j
, io_sectors
;
4454 unsigned long mark
[SYNC_MARKS
];
4455 sector_t mark_cnt
[SYNC_MARKS
];
4457 struct list_head
*tmp
;
4458 sector_t last_check
;
4461 /* just incase thread restarts... */
4462 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4465 /* we overload curr_resync somewhat here.
4466 * 0 == not engaged in resync at all
4467 * 2 == checking that there is no conflict with another sync
4468 * 1 == like 2, but have yielded to allow conflicting resync to
4470 * other == active in resync - this many blocks
4472 * Before starting a resync we must have set curr_resync to
4473 * 2, and then checked that every "conflicting" array has curr_resync
4474 * less than ours. When we find one that is the same or higher
4475 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4476 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4477 * This will mean we have to start checking from the beginning again.
4482 mddev
->curr_resync
= 2;
4485 if (kthread_should_stop()) {
4486 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4489 ITERATE_MDDEV(mddev2
,tmp
) {
4490 if (mddev2
== mddev
)
4492 if (mddev2
->curr_resync
&&
4493 match_mddev_units(mddev
,mddev2
)) {
4495 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4496 /* arbitrarily yield */
4497 mddev
->curr_resync
= 1;
4498 wake_up(&resync_wait
);
4500 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4501 /* no need to wait here, we can wait the next
4502 * time 'round when curr_resync == 2
4505 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4506 if (!kthread_should_stop() &&
4507 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4508 printk(KERN_INFO
"md: delaying resync of %s"
4509 " until %s has finished resync (they"
4510 " share one or more physical units)\n",
4511 mdname(mddev
), mdname(mddev2
));
4514 finish_wait(&resync_wait
, &wq
);
4517 finish_wait(&resync_wait
, &wq
);
4520 } while (mddev
->curr_resync
< 2);
4522 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4523 /* resync follows the size requested by the personality,
4524 * which defaults to physical size, but can be virtual size
4526 max_sectors
= mddev
->resync_max_sectors
;
4527 mddev
->resync_mismatches
= 0;
4529 /* recovery follows the physical size of devices */
4530 max_sectors
= mddev
->size
<< 1;
4532 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4533 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4534 " %d KB/sec/disc.\n", speed_min(mddev
));
4535 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4536 "(but not more than %d KB/sec) for reconstruction.\n",
4539 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4540 /* we don't use the checkpoint if there's a bitmap */
4541 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && !mddev
->bitmap
4542 && ! test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4543 j
= mddev
->recovery_cp
;
4547 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4549 mark_cnt
[m
] = io_sectors
;
4552 mddev
->resync_mark
= mark
[last_mark
];
4553 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4556 * Tune reconstruction:
4558 window
= 32*(PAGE_SIZE
/512);
4559 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4560 window
/2,(unsigned long long) max_sectors
/2);
4562 atomic_set(&mddev
->recovery_active
, 0);
4563 init_waitqueue_head(&mddev
->recovery_wait
);
4568 "md: resuming recovery of %s from checkpoint.\n",
4570 mddev
->curr_resync
= j
;
4573 while (j
< max_sectors
) {
4577 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4578 currspeed
< speed_min(mddev
));
4580 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4584 if (!skipped
) { /* actual IO requested */
4585 io_sectors
+= sectors
;
4586 atomic_add(sectors
, &mddev
->recovery_active
);
4590 if (j
>1) mddev
->curr_resync
= j
;
4591 if (last_check
== 0)
4592 /* this is the earliers that rebuilt will be
4593 * visible in /proc/mdstat
4595 md_new_event(mddev
);
4597 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4600 last_check
= io_sectors
;
4602 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4603 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4607 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4609 int next
= (last_mark
+1) % SYNC_MARKS
;
4611 mddev
->resync_mark
= mark
[next
];
4612 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4613 mark
[next
] = jiffies
;
4614 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4619 if (kthread_should_stop()) {
4621 * got a signal, exit.
4624 "md: md_do_sync() got signal ... exiting\n");
4625 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4630 * this loop exits only if either when we are slower than
4631 * the 'hard' speed limit, or the system was IO-idle for
4633 * the system might be non-idle CPU-wise, but we only care
4634 * about not overloading the IO subsystem. (things like an
4635 * e2fsck being done on the RAID array should execute fast)
4637 mddev
->queue
->unplug_fn(mddev
->queue
);
4640 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4641 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4643 if (currspeed
> speed_min(mddev
)) {
4644 if ((currspeed
> speed_max(mddev
)) ||
4645 !is_mddev_idle(mddev
)) {
4651 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
4653 * this also signals 'finished resyncing' to md_stop
4656 mddev
->queue
->unplug_fn(mddev
->queue
);
4658 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
4660 /* tell personality that we are finished */
4661 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
4663 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4664 mddev
->curr_resync
> 2 &&
4665 mddev
->curr_resync
>= mddev
->recovery_cp
) {
4666 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4668 "md: checkpointing recovery of %s.\n",
4670 mddev
->recovery_cp
= mddev
->curr_resync
;
4672 mddev
->recovery_cp
= MaxSector
;
4676 mddev
->curr_resync
= 0;
4677 wake_up(&resync_wait
);
4678 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4679 md_wakeup_thread(mddev
->thread
);
4684 * This routine is regularly called by all per-raid-array threads to
4685 * deal with generic issues like resync and super-block update.
4686 * Raid personalities that don't have a thread (linear/raid0) do not
4687 * need this as they never do any recovery or update the superblock.
4689 * It does not do any resync itself, but rather "forks" off other threads
4690 * to do that as needed.
4691 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4692 * "->recovery" and create a thread at ->sync_thread.
4693 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4694 * and wakeups up this thread which will reap the thread and finish up.
4695 * This thread also removes any faulty devices (with nr_pending == 0).
4697 * The overall approach is:
4698 * 1/ if the superblock needs updating, update it.
4699 * 2/ If a recovery thread is running, don't do anything else.
4700 * 3/ If recovery has finished, clean up, possibly marking spares active.
4701 * 4/ If there are any faulty devices, remove them.
4702 * 5/ If array is degraded, try to add spares devices
4703 * 6/ If array has spares or is not in-sync, start a resync thread.
4705 void md_check_recovery(mddev_t
*mddev
)
4708 struct list_head
*rtmp
;
4712 bitmap_daemon_work(mddev
->bitmap
);
4717 if (signal_pending(current
)) {
4718 if (mddev
->pers
->sync_request
) {
4719 printk(KERN_INFO
"md: %s in immediate safe mode\n",
4721 mddev
->safemode
= 2;
4723 flush_signals(current
);
4728 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
4729 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
4730 (mddev
->safemode
== 1) ||
4731 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
4732 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
4736 if (mddev_trylock(mddev
)==0) {
4739 spin_lock_irq(&mddev
->write_lock
);
4740 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
4741 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
4743 mddev
->sb_dirty
= 1;
4745 if (mddev
->safemode
== 1)
4746 mddev
->safemode
= 0;
4747 spin_unlock_irq(&mddev
->write_lock
);
4749 if (mddev
->sb_dirty
)
4750 md_update_sb(mddev
);
4753 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4754 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
4755 /* resync/recovery still happening */
4756 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4759 if (mddev
->sync_thread
) {
4760 /* resync has finished, collect result */
4761 md_unregister_thread(mddev
->sync_thread
);
4762 mddev
->sync_thread
= NULL
;
4763 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4764 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4766 /* activate any spares */
4767 mddev
->pers
->spare_active(mddev
);
4769 md_update_sb(mddev
);
4771 /* if array is no-longer degraded, then any saved_raid_disk
4772 * information must be scrapped
4774 if (!mddev
->degraded
)
4775 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4776 rdev
->saved_raid_disk
= -1;
4778 mddev
->recovery
= 0;
4779 /* flag recovery needed just to double check */
4780 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4781 md_new_event(mddev
);
4784 /* Clear some bits that don't mean anything, but
4787 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4788 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4789 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4790 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4792 /* no recovery is running.
4793 * remove any failed drives, then
4794 * add spares if possible.
4795 * Spare are also removed and re-added, to allow
4796 * the personality to fail the re-add.
4798 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4799 if (rdev
->raid_disk
>= 0 &&
4800 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
4801 atomic_read(&rdev
->nr_pending
)==0) {
4802 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
4804 sprintf(nm
,"rd%d", rdev
->raid_disk
);
4805 sysfs_remove_link(&mddev
->kobj
, nm
);
4806 rdev
->raid_disk
= -1;
4810 if (mddev
->degraded
) {
4811 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4812 if (rdev
->raid_disk
< 0
4813 && !test_bit(Faulty
, &rdev
->flags
)) {
4814 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
4816 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4817 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
4819 md_new_event(mddev
);
4826 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4827 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4828 } else if (mddev
->recovery_cp
< MaxSector
) {
4829 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4830 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4831 /* nothing to be done ... */
4834 if (mddev
->pers
->sync_request
) {
4835 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4836 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
4837 /* We are adding a device or devices to an array
4838 * which has the bitmap stored on all devices.
4839 * So make sure all bitmap pages get written
4841 bitmap_write_all(mddev
->bitmap
);
4843 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4846 if (!mddev
->sync_thread
) {
4847 printk(KERN_ERR
"%s: could not start resync"
4850 /* leave the spares where they are, it shouldn't hurt */
4851 mddev
->recovery
= 0;
4853 md_wakeup_thread(mddev
->sync_thread
);
4854 md_new_event(mddev
);
4857 mddev_unlock(mddev
);
4861 static int md_notify_reboot(struct notifier_block
*this,
4862 unsigned long code
, void *x
)
4864 struct list_head
*tmp
;
4867 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
4869 printk(KERN_INFO
"md: stopping all md devices.\n");
4871 ITERATE_MDDEV(mddev
,tmp
)
4872 if (mddev_trylock(mddev
)==0)
4873 do_md_stop (mddev
, 1);
4875 * certain more exotic SCSI devices are known to be
4876 * volatile wrt too early system reboots. While the
4877 * right place to handle this issue is the given
4878 * driver, we do want to have a safe RAID driver ...
4885 static struct notifier_block md_notifier
= {
4886 .notifier_call
= md_notify_reboot
,
4888 .priority
= INT_MAX
, /* before any real devices */
4891 static void md_geninit(void)
4893 struct proc_dir_entry
*p
;
4895 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
4897 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
4899 p
->proc_fops
= &md_seq_fops
;
4902 static int __init
md_init(void)
4906 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4907 " MD_SB_DISKS=%d\n",
4908 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
4909 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
4910 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
4913 if (register_blkdev(MAJOR_NR
, "md"))
4915 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
4916 unregister_blkdev(MAJOR_NR
, "md");
4920 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
4921 md_probe
, NULL
, NULL
);
4922 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
4923 md_probe
, NULL
, NULL
);
4925 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4926 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
4927 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4930 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4931 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
4932 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4936 register_reboot_notifier(&md_notifier
);
4937 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
4947 * Searches all registered partitions for autorun RAID arrays
4950 static dev_t detected_devices
[128];
4953 void md_autodetect_dev(dev_t dev
)
4955 if (dev_cnt
>= 0 && dev_cnt
< 127)
4956 detected_devices
[dev_cnt
++] = dev
;
4960 static void autostart_arrays(int part
)
4965 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
4967 for (i
= 0; i
< dev_cnt
; i
++) {
4968 dev_t dev
= detected_devices
[i
];
4970 rdev
= md_import_device(dev
,0, 0);
4974 if (test_bit(Faulty
, &rdev
->flags
)) {
4978 list_add(&rdev
->same_set
, &pending_raid_disks
);
4982 autorun_devices(part
);
4987 static __exit
void md_exit(void)
4990 struct list_head
*tmp
;
4992 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
4993 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
4994 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4995 devfs_remove("md/%d", i
);
4996 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4997 devfs_remove("md/d%d", i
);
5001 unregister_blkdev(MAJOR_NR
,"md");
5002 unregister_blkdev(mdp_major
, "mdp");
5003 unregister_reboot_notifier(&md_notifier
);
5004 unregister_sysctl_table(raid_table_header
);
5005 remove_proc_entry("mdstat", NULL
);
5006 ITERATE_MDDEV(mddev
,tmp
) {
5007 struct gendisk
*disk
= mddev
->gendisk
;
5010 export_array(mddev
);
5013 mddev
->gendisk
= NULL
;
5018 module_init(md_init
)
5019 module_exit(md_exit
)
5021 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5023 return sprintf(buffer
, "%d", start_readonly
);
5025 static int set_ro(const char *val
, struct kernel_param
*kp
)
5028 int num
= simple_strtoul(val
, &e
, 10);
5029 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5030 start_readonly
= num
;
5036 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5037 module_param(start_dirty_degraded
, int, 0644);
5040 EXPORT_SYMBOL(register_md_personality
);
5041 EXPORT_SYMBOL(unregister_md_personality
);
5042 EXPORT_SYMBOL(md_error
);
5043 EXPORT_SYMBOL(md_done_sync
);
5044 EXPORT_SYMBOL(md_write_start
);
5045 EXPORT_SYMBOL(md_write_end
);
5046 EXPORT_SYMBOL(md_register_thread
);
5047 EXPORT_SYMBOL(md_unregister_thread
);
5048 EXPORT_SYMBOL(md_wakeup_thread
);
5049 EXPORT_SYMBOL(md_print_devices
);
5050 EXPORT_SYMBOL(md_check_recovery
);
5051 MODULE_LICENSE("GPL");
5053 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);