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/kthread.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/buffer_head.h> /* for invalidate_bdev */
42 #include <linux/suspend.h>
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.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
);
74 static void md_print_devices(void);
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
91 static int sysctl_speed_limit_min
= 1000;
92 static int sysctl_speed_limit_max
= 200000;
93 static inline int speed_min(mddev_t
*mddev
)
95 return mddev
->sync_speed_min
?
96 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
99 static inline int speed_max(mddev_t
*mddev
)
101 return mddev
->sync_speed_max
?
102 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
105 static struct ctl_table_header
*raid_table_header
;
107 static ctl_table raid_table
[] = {
109 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
110 .procname
= "speed_limit_min",
111 .data
= &sysctl_speed_limit_min
,
112 .maxlen
= sizeof(int),
113 .mode
= S_IRUGO
|S_IWUSR
,
114 .proc_handler
= &proc_dointvec
,
117 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
118 .procname
= "speed_limit_max",
119 .data
= &sysctl_speed_limit_max
,
120 .maxlen
= sizeof(int),
121 .mode
= S_IRUGO
|S_IWUSR
,
122 .proc_handler
= &proc_dointvec
,
127 static ctl_table raid_dir_table
[] = {
129 .ctl_name
= DEV_RAID
,
132 .mode
= S_IRUGO
|S_IXUGO
,
138 static ctl_table raid_root_table
[] = {
144 .child
= raid_dir_table
,
149 static struct block_device_operations md_fops
;
151 static int start_readonly
;
154 * We have a system wide 'event count' that is incremented
155 * on any 'interesting' event, and readers of /proc/mdstat
156 * can use 'poll' or 'select' to find out when the event
160 * start array, stop array, error, add device, remove device,
161 * start build, activate spare
163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
164 static atomic_t md_event_count
;
165 void md_new_event(mddev_t
*mddev
)
167 atomic_inc(&md_event_count
);
168 wake_up(&md_event_waiters
);
169 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
171 EXPORT_SYMBOL_GPL(md_new_event
);
173 /* Alternate version that can be called from interrupts
174 * when calling sysfs_notify isn't needed.
176 static void md_new_event_inintr(mddev_t
*mddev
)
178 atomic_inc(&md_event_count
);
179 wake_up(&md_event_waiters
);
183 * Enables to iterate over all existing md arrays
184 * all_mddevs_lock protects this list.
186 static LIST_HEAD(all_mddevs
);
187 static DEFINE_SPINLOCK(all_mddevs_lock
);
191 * iterates through all used mddevs in the system.
192 * We take care to grab the all_mddevs_lock whenever navigating
193 * the list, and to always hold a refcount when unlocked.
194 * Any code which breaks out of this loop while own
195 * a reference to the current mddev and must mddev_put it.
197 #define ITERATE_MDDEV(mddev,tmp) \
199 for (({ spin_lock(&all_mddevs_lock); \
200 tmp = all_mddevs.next; \
202 ({ if (tmp != &all_mddevs) \
203 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204 spin_unlock(&all_mddevs_lock); \
205 if (mddev) mddev_put(mddev); \
206 mddev = list_entry(tmp, mddev_t, all_mddevs); \
207 tmp != &all_mddevs;}); \
208 ({ spin_lock(&all_mddevs_lock); \
213 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
215 bio_io_error(bio
, bio
->bi_size
);
219 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
221 atomic_inc(&mddev
->active
);
225 static void mddev_put(mddev_t
*mddev
)
227 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
229 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
230 list_del(&mddev
->all_mddevs
);
231 spin_unlock(&all_mddevs_lock
);
232 blk_cleanup_queue(mddev
->queue
);
233 kobject_unregister(&mddev
->kobj
);
235 spin_unlock(&all_mddevs_lock
);
238 static mddev_t
* mddev_find(dev_t unit
)
240 mddev_t
*mddev
, *new = NULL
;
243 spin_lock(&all_mddevs_lock
);
244 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
245 if (mddev
->unit
== unit
) {
247 spin_unlock(&all_mddevs_lock
);
253 list_add(&new->all_mddevs
, &all_mddevs
);
254 spin_unlock(&all_mddevs_lock
);
257 spin_unlock(&all_mddevs_lock
);
259 new = kzalloc(sizeof(*new), GFP_KERNEL
);
264 if (MAJOR(unit
) == MD_MAJOR
)
265 new->md_minor
= MINOR(unit
);
267 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
269 mutex_init(&new->reconfig_mutex
);
270 INIT_LIST_HEAD(&new->disks
);
271 INIT_LIST_HEAD(&new->all_mddevs
);
272 init_timer(&new->safemode_timer
);
273 atomic_set(&new->active
, 1);
274 spin_lock_init(&new->write_lock
);
275 init_waitqueue_head(&new->sb_wait
);
277 new->queue
= blk_alloc_queue(GFP_KERNEL
);
282 set_bit(QUEUE_FLAG_CLUSTER
, &new->queue
->queue_flags
);
284 blk_queue_make_request(new->queue
, md_fail_request
);
289 static inline int mddev_lock(mddev_t
* mddev
)
291 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
294 static inline int mddev_trylock(mddev_t
* mddev
)
296 return mutex_trylock(&mddev
->reconfig_mutex
);
299 static inline void mddev_unlock(mddev_t
* mddev
)
301 mutex_unlock(&mddev
->reconfig_mutex
);
303 md_wakeup_thread(mddev
->thread
);
306 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
309 struct list_head
*tmp
;
311 ITERATE_RDEV(mddev
,rdev
,tmp
) {
312 if (rdev
->desc_nr
== nr
)
318 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
320 struct list_head
*tmp
;
323 ITERATE_RDEV(mddev
,rdev
,tmp
) {
324 if (rdev
->bdev
->bd_dev
== dev
)
330 static struct mdk_personality
*find_pers(int level
, char *clevel
)
332 struct mdk_personality
*pers
;
333 list_for_each_entry(pers
, &pers_list
, list
) {
334 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
336 if (strcmp(pers
->name
, clevel
)==0)
342 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
344 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
345 return MD_NEW_SIZE_BLOCKS(size
);
348 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
352 size
= rdev
->sb_offset
;
355 size
&= ~((sector_t
)chunk_size
/1024 - 1);
359 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
364 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
365 if (!rdev
->sb_page
) {
366 printk(KERN_ALERT
"md: out of memory.\n");
373 static void free_disk_sb(mdk_rdev_t
* rdev
)
376 put_page(rdev
->sb_page
);
378 rdev
->sb_page
= NULL
;
385 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
387 mdk_rdev_t
*rdev
= bio
->bi_private
;
388 mddev_t
*mddev
= rdev
->mddev
;
392 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
393 printk("md: super_written gets error=%d, uptodate=%d\n",
394 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
395 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
396 md_error(mddev
, rdev
);
399 if (atomic_dec_and_test(&mddev
->pending_writes
))
400 wake_up(&mddev
->sb_wait
);
405 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
407 struct bio
*bio2
= bio
->bi_private
;
408 mdk_rdev_t
*rdev
= bio2
->bi_private
;
409 mddev_t
*mddev
= rdev
->mddev
;
413 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
414 error
== -EOPNOTSUPP
) {
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp
, &rdev
->flags
);
418 mddev
->barriers_work
= 0;
419 spin_lock_irqsave(&mddev
->write_lock
, flags
);
420 bio2
->bi_next
= mddev
->biolist
;
421 mddev
->biolist
= bio2
;
422 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
423 wake_up(&mddev
->sb_wait
);
428 bio
->bi_private
= rdev
;
429 return super_written(bio
, bytes_done
, error
);
432 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
433 sector_t sector
, int size
, struct page
*page
)
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
444 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
445 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
447 bio
->bi_bdev
= rdev
->bdev
;
448 bio
->bi_sector
= sector
;
449 bio_add_page(bio
, page
, size
, 0);
450 bio
->bi_private
= rdev
;
451 bio
->bi_end_io
= super_written
;
454 atomic_inc(&mddev
->pending_writes
);
455 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
457 rw
|= (1<<BIO_RW_BARRIER
);
458 rbio
= bio_clone(bio
, GFP_NOIO
);
459 rbio
->bi_private
= bio
;
460 rbio
->bi_end_io
= super_written_barrier
;
461 submit_bio(rw
, rbio
);
466 void md_super_wait(mddev_t
*mddev
)
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
473 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
474 if (atomic_read(&mddev
->pending_writes
)==0)
476 while (mddev
->biolist
) {
478 spin_lock_irq(&mddev
->write_lock
);
479 bio
= mddev
->biolist
;
480 mddev
->biolist
= bio
->bi_next
;
482 spin_unlock_irq(&mddev
->write_lock
);
483 submit_bio(bio
->bi_rw
, bio
);
487 finish_wait(&mddev
->sb_wait
, &wq
);
490 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
495 complete((struct completion
*)bio
->bi_private
);
499 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
500 struct page
*page
, int rw
)
502 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
503 struct completion event
;
506 rw
|= (1 << BIO_RW_SYNC
);
509 bio
->bi_sector
= sector
;
510 bio_add_page(bio
, page
, size
, 0);
511 init_completion(&event
);
512 bio
->bi_private
= &event
;
513 bio
->bi_end_io
= bi_complete
;
515 wait_for_completion(&event
);
517 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
521 EXPORT_SYMBOL_GPL(sync_page_io
);
523 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
525 char b
[BDEVNAME_SIZE
];
526 if (!rdev
->sb_page
) {
534 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
540 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
541 bdevname(rdev
->bdev
,b
));
545 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
547 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
548 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
549 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
550 (sb1
->set_uuid3
== sb2
->set_uuid3
))
558 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
561 mdp_super_t
*tmp1
, *tmp2
;
563 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
564 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
566 if (!tmp1
|| !tmp2
) {
568 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
576 * nr_disks is not constant
581 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
592 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
594 unsigned int disk_csum
, csum
;
596 disk_csum
= sb
->sb_csum
;
598 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
599 sb
->sb_csum
= disk_csum
;
605 * Handle superblock details.
606 * We want to be able to handle multiple superblock formats
607 * so we have a common interface to them all, and an array of
608 * different handlers.
609 * We rely on user-space to write the initial superblock, and support
610 * reading and updating of superblocks.
611 * Interface methods are:
612 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
613 * loads and validates a superblock on dev.
614 * if refdev != NULL, compare superblocks on both devices
616 * 0 - dev has a superblock that is compatible with refdev
617 * 1 - dev has a superblock that is compatible and newer than refdev
618 * so dev should be used as the refdev in future
619 * -EINVAL superblock incompatible or invalid
620 * -othererror e.g. -EIO
622 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
623 * Verify that dev is acceptable into mddev.
624 * The first time, mddev->raid_disks will be 0, and data from
625 * dev should be merged in. Subsequent calls check that dev
626 * is new enough. Return 0 or -EINVAL
628 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
629 * Update the superblock for rdev with data in mddev
630 * This does not write to disc.
636 struct module
*owner
;
637 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
638 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
639 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
643 * load_super for 0.90.0
645 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
647 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
653 * Calculate the position of the superblock,
654 * it's at the end of the disk.
656 * It also happens to be a multiple of 4Kb.
658 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
659 rdev
->sb_offset
= sb_offset
;
661 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
666 bdevname(rdev
->bdev
, b
);
667 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
669 if (sb
->md_magic
!= MD_SB_MAGIC
) {
670 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
675 if (sb
->major_version
!= 0 ||
676 sb
->minor_version
< 90 ||
677 sb
->minor_version
> 91) {
678 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
679 sb
->major_version
, sb
->minor_version
,
684 if (sb
->raid_disks
<= 0)
687 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
688 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
693 rdev
->preferred_minor
= sb
->md_minor
;
694 rdev
->data_offset
= 0;
695 rdev
->sb_size
= MD_SB_BYTES
;
697 if (sb
->level
== LEVEL_MULTIPATH
)
700 rdev
->desc_nr
= sb
->this_disk
.number
;
706 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
707 if (!uuid_equal(refsb
, sb
)) {
708 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
709 b
, bdevname(refdev
->bdev
,b2
));
712 if (!sb_equal(refsb
, sb
)) {
713 printk(KERN_WARNING
"md: %s has same UUID"
714 " but different superblock to %s\n",
715 b
, bdevname(refdev
->bdev
, b2
));
719 ev2
= md_event(refsb
);
725 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
727 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
728 /* "this cannot possibly happen" ... */
736 * validate_super for 0.90.0
738 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
741 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
742 __u64 ev1
= md_event(sb
);
744 rdev
->raid_disk
= -1;
746 if (mddev
->raid_disks
== 0) {
747 mddev
->major_version
= 0;
748 mddev
->minor_version
= sb
->minor_version
;
749 mddev
->patch_version
= sb
->patch_version
;
750 mddev
->persistent
= ! sb
->not_persistent
;
751 mddev
->chunk_size
= sb
->chunk_size
;
752 mddev
->ctime
= sb
->ctime
;
753 mddev
->utime
= sb
->utime
;
754 mddev
->level
= sb
->level
;
755 mddev
->clevel
[0] = 0;
756 mddev
->layout
= sb
->layout
;
757 mddev
->raid_disks
= sb
->raid_disks
;
758 mddev
->size
= sb
->size
;
760 mddev
->bitmap_offset
= 0;
761 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
763 if (mddev
->minor_version
>= 91) {
764 mddev
->reshape_position
= sb
->reshape_position
;
765 mddev
->delta_disks
= sb
->delta_disks
;
766 mddev
->new_level
= sb
->new_level
;
767 mddev
->new_layout
= sb
->new_layout
;
768 mddev
->new_chunk
= sb
->new_chunk
;
770 mddev
->reshape_position
= MaxSector
;
771 mddev
->delta_disks
= 0;
772 mddev
->new_level
= mddev
->level
;
773 mddev
->new_layout
= mddev
->layout
;
774 mddev
->new_chunk
= mddev
->chunk_size
;
777 if (sb
->state
& (1<<MD_SB_CLEAN
))
778 mddev
->recovery_cp
= MaxSector
;
780 if (sb
->events_hi
== sb
->cp_events_hi
&&
781 sb
->events_lo
== sb
->cp_events_lo
) {
782 mddev
->recovery_cp
= sb
->recovery_cp
;
784 mddev
->recovery_cp
= 0;
787 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
788 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
789 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
790 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
792 mddev
->max_disks
= MD_SB_DISKS
;
794 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
795 mddev
->bitmap_file
== NULL
) {
796 if (mddev
->level
!= 1 && mddev
->level
!= 4
797 && mddev
->level
!= 5 && mddev
->level
!= 6
798 && mddev
->level
!= 10) {
799 /* FIXME use a better test */
800 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
803 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
806 } else if (mddev
->pers
== NULL
) {
807 /* Insist on good event counter while assembling */
809 if (ev1
< mddev
->events
)
811 } else if (mddev
->bitmap
) {
812 /* if adding to array with a bitmap, then we can accept an
813 * older device ... but not too old.
815 if (ev1
< mddev
->bitmap
->events_cleared
)
818 if (ev1
< mddev
->events
)
819 /* just a hot-add of a new device, leave raid_disk at -1 */
823 if (mddev
->level
!= LEVEL_MULTIPATH
) {
824 desc
= sb
->disks
+ rdev
->desc_nr
;
826 if (desc
->state
& (1<<MD_DISK_FAULTY
))
827 set_bit(Faulty
, &rdev
->flags
);
828 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
829 desc->raid_disk < mddev->raid_disks */) {
830 set_bit(In_sync
, &rdev
->flags
);
831 rdev
->raid_disk
= desc
->raid_disk
;
833 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
834 set_bit(WriteMostly
, &rdev
->flags
);
835 } else /* MULTIPATH are always insync */
836 set_bit(In_sync
, &rdev
->flags
);
841 * sync_super for 0.90.0
843 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
846 struct list_head
*tmp
;
848 int next_spare
= mddev
->raid_disks
;
851 /* make rdev->sb match mddev data..
854 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
855 * 3/ any empty disks < next_spare become removed
857 * disks[0] gets initialised to REMOVED because
858 * we cannot be sure from other fields if it has
859 * been initialised or not.
862 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
864 rdev
->sb_size
= MD_SB_BYTES
;
866 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
868 memset(sb
, 0, sizeof(*sb
));
870 sb
->md_magic
= MD_SB_MAGIC
;
871 sb
->major_version
= mddev
->major_version
;
872 sb
->patch_version
= mddev
->patch_version
;
873 sb
->gvalid_words
= 0; /* ignored */
874 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
875 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
876 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
877 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
879 sb
->ctime
= mddev
->ctime
;
880 sb
->level
= mddev
->level
;
881 sb
->size
= mddev
->size
;
882 sb
->raid_disks
= mddev
->raid_disks
;
883 sb
->md_minor
= mddev
->md_minor
;
884 sb
->not_persistent
= !mddev
->persistent
;
885 sb
->utime
= mddev
->utime
;
887 sb
->events_hi
= (mddev
->events
>>32);
888 sb
->events_lo
= (u32
)mddev
->events
;
890 if (mddev
->reshape_position
== MaxSector
)
891 sb
->minor_version
= 90;
893 sb
->minor_version
= 91;
894 sb
->reshape_position
= mddev
->reshape_position
;
895 sb
->new_level
= mddev
->new_level
;
896 sb
->delta_disks
= mddev
->delta_disks
;
897 sb
->new_layout
= mddev
->new_layout
;
898 sb
->new_chunk
= mddev
->new_chunk
;
900 mddev
->minor_version
= sb
->minor_version
;
903 sb
->recovery_cp
= mddev
->recovery_cp
;
904 sb
->cp_events_hi
= (mddev
->events
>>32);
905 sb
->cp_events_lo
= (u32
)mddev
->events
;
906 if (mddev
->recovery_cp
== MaxSector
)
907 sb
->state
= (1<< MD_SB_CLEAN
);
911 sb
->layout
= mddev
->layout
;
912 sb
->chunk_size
= mddev
->chunk_size
;
914 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
915 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
917 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
918 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
921 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
922 && !test_bit(Faulty
, &rdev2
->flags
))
923 desc_nr
= rdev2
->raid_disk
;
925 desc_nr
= next_spare
++;
926 rdev2
->desc_nr
= desc_nr
;
927 d
= &sb
->disks
[rdev2
->desc_nr
];
929 d
->number
= rdev2
->desc_nr
;
930 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
931 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
932 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
933 && !test_bit(Faulty
, &rdev2
->flags
))
934 d
->raid_disk
= rdev2
->raid_disk
;
936 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
937 if (test_bit(Faulty
, &rdev2
->flags
))
938 d
->state
= (1<<MD_DISK_FAULTY
);
939 else if (test_bit(In_sync
, &rdev2
->flags
)) {
940 d
->state
= (1<<MD_DISK_ACTIVE
);
941 d
->state
|= (1<<MD_DISK_SYNC
);
949 if (test_bit(WriteMostly
, &rdev2
->flags
))
950 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
952 /* now set the "removed" and "faulty" bits on any missing devices */
953 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
954 mdp_disk_t
*d
= &sb
->disks
[i
];
955 if (d
->state
== 0 && d
->number
== 0) {
958 d
->state
= (1<<MD_DISK_REMOVED
);
959 d
->state
|= (1<<MD_DISK_FAULTY
);
963 sb
->nr_disks
= nr_disks
;
964 sb
->active_disks
= active
;
965 sb
->working_disks
= working
;
966 sb
->failed_disks
= failed
;
967 sb
->spare_disks
= spare
;
969 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
970 sb
->sb_csum
= calc_sb_csum(sb
);
974 * version 1 superblock
977 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
979 unsigned int disk_csum
, csum
;
980 unsigned long long newcsum
;
981 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
982 unsigned int *isuper
= (unsigned int*)sb
;
985 disk_csum
= sb
->sb_csum
;
988 for (i
=0; size
>=4; size
-= 4 )
989 newcsum
+= le32_to_cpu(*isuper
++);
992 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
994 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
995 sb
->sb_csum
= disk_csum
;
996 return cpu_to_le32(csum
);
999 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1001 struct mdp_superblock_1
*sb
;
1004 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1008 * Calculate the position of the superblock.
1009 * It is always aligned to a 4K boundary and
1010 * depeding on minor_version, it can be:
1011 * 0: At least 8K, but less than 12K, from end of device
1012 * 1: At start of device
1013 * 2: 4K from start of device.
1015 switch(minor_version
) {
1017 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1019 sb_offset
&= ~(sector_t
)(4*2-1);
1020 /* convert from sectors to K */
1032 rdev
->sb_offset
= sb_offset
;
1034 /* superblock is rarely larger than 1K, but it can be larger,
1035 * and it is safe to read 4k, so we do that
1037 ret
= read_disk_sb(rdev
, 4096);
1038 if (ret
) return ret
;
1041 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1043 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1044 sb
->major_version
!= cpu_to_le32(1) ||
1045 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1046 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1047 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1050 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1051 printk("md: invalid superblock checksum on %s\n",
1052 bdevname(rdev
->bdev
,b
));
1055 if (le64_to_cpu(sb
->data_size
) < 10) {
1056 printk("md: data_size too small on %s\n",
1057 bdevname(rdev
->bdev
,b
));
1060 rdev
->preferred_minor
= 0xffff;
1061 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1062 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1064 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1065 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1066 if (rdev
->sb_size
& bmask
)
1067 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1069 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1072 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1078 struct mdp_superblock_1
*refsb
=
1079 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1081 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1082 sb
->level
!= refsb
->level
||
1083 sb
->layout
!= refsb
->layout
||
1084 sb
->chunksize
!= refsb
->chunksize
) {
1085 printk(KERN_WARNING
"md: %s has strangely different"
1086 " superblock to %s\n",
1087 bdevname(rdev
->bdev
,b
),
1088 bdevname(refdev
->bdev
,b2
));
1091 ev1
= le64_to_cpu(sb
->events
);
1092 ev2
= le64_to_cpu(refsb
->events
);
1100 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1102 rdev
->size
= rdev
->sb_offset
;
1103 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1105 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1106 if (le32_to_cpu(sb
->chunksize
))
1107 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1109 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1114 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1116 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1117 __u64 ev1
= le64_to_cpu(sb
->events
);
1119 rdev
->raid_disk
= -1;
1121 if (mddev
->raid_disks
== 0) {
1122 mddev
->major_version
= 1;
1123 mddev
->patch_version
= 0;
1124 mddev
->persistent
= 1;
1125 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1126 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1127 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1128 mddev
->level
= le32_to_cpu(sb
->level
);
1129 mddev
->clevel
[0] = 0;
1130 mddev
->layout
= le32_to_cpu(sb
->layout
);
1131 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1132 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1133 mddev
->events
= ev1
;
1134 mddev
->bitmap_offset
= 0;
1135 mddev
->default_bitmap_offset
= 1024 >> 9;
1137 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1138 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1140 mddev
->max_disks
= (4096-256)/2;
1142 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1143 mddev
->bitmap_file
== NULL
) {
1144 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1145 && mddev
->level
!= 10) {
1146 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1149 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1151 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1152 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1153 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1154 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1155 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1156 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1158 mddev
->reshape_position
= MaxSector
;
1159 mddev
->delta_disks
= 0;
1160 mddev
->new_level
= mddev
->level
;
1161 mddev
->new_layout
= mddev
->layout
;
1162 mddev
->new_chunk
= mddev
->chunk_size
;
1165 } else if (mddev
->pers
== NULL
) {
1166 /* Insist of good event counter while assembling */
1168 if (ev1
< mddev
->events
)
1170 } else if (mddev
->bitmap
) {
1171 /* If adding to array with a bitmap, then we can accept an
1172 * older device, but not too old.
1174 if (ev1
< mddev
->bitmap
->events_cleared
)
1177 if (ev1
< mddev
->events
)
1178 /* just a hot-add of a new device, leave raid_disk at -1 */
1181 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1183 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1185 case 0xffff: /* spare */
1187 case 0xfffe: /* faulty */
1188 set_bit(Faulty
, &rdev
->flags
);
1191 if ((le32_to_cpu(sb
->feature_map
) &
1192 MD_FEATURE_RECOVERY_OFFSET
))
1193 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1195 set_bit(In_sync
, &rdev
->flags
);
1196 rdev
->raid_disk
= role
;
1199 if (sb
->devflags
& WriteMostly1
)
1200 set_bit(WriteMostly
, &rdev
->flags
);
1201 } else /* MULTIPATH are always insync */
1202 set_bit(In_sync
, &rdev
->flags
);
1207 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1209 struct mdp_superblock_1
*sb
;
1210 struct list_head
*tmp
;
1213 /* make rdev->sb match mddev and rdev data. */
1215 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1217 sb
->feature_map
= 0;
1219 sb
->recovery_offset
= cpu_to_le64(0);
1220 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1221 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1222 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1224 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1225 sb
->events
= cpu_to_le64(mddev
->events
);
1227 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1229 sb
->resync_offset
= cpu_to_le64(0);
1231 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1233 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1234 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1236 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1237 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1238 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1241 if (rdev
->raid_disk
>= 0 &&
1242 !test_bit(In_sync
, &rdev
->flags
) &&
1243 rdev
->recovery_offset
> 0) {
1244 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1245 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1248 if (mddev
->reshape_position
!= MaxSector
) {
1249 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1250 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1251 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1252 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1253 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1254 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1258 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1259 if (rdev2
->desc_nr
+1 > max_dev
)
1260 max_dev
= rdev2
->desc_nr
+1;
1262 sb
->max_dev
= cpu_to_le32(max_dev
);
1263 for (i
=0; i
<max_dev
;i
++)
1264 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1266 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1268 if (test_bit(Faulty
, &rdev2
->flags
))
1269 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1270 else if (test_bit(In_sync
, &rdev2
->flags
))
1271 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1272 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1273 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1275 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1278 sb
->sb_csum
= calc_sb_1_csum(sb
);
1282 static struct super_type super_types
[] = {
1285 .owner
= THIS_MODULE
,
1286 .load_super
= super_90_load
,
1287 .validate_super
= super_90_validate
,
1288 .sync_super
= super_90_sync
,
1292 .owner
= THIS_MODULE
,
1293 .load_super
= super_1_load
,
1294 .validate_super
= super_1_validate
,
1295 .sync_super
= super_1_sync
,
1299 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1301 struct list_head
*tmp
;
1304 ITERATE_RDEV(mddev
,rdev
,tmp
)
1305 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1311 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1313 struct list_head
*tmp
;
1316 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1317 if (match_dev_unit(mddev2
, rdev
))
1323 static LIST_HEAD(pending_raid_disks
);
1325 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1327 mdk_rdev_t
*same_pdev
;
1328 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1336 /* make sure rdev->size exceeds mddev->size */
1337 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1339 /* Cannot change size, so fail */
1342 mddev
->size
= rdev
->size
;
1344 same_pdev
= match_dev_unit(mddev
, rdev
);
1347 "%s: WARNING: %s appears to be on the same physical"
1348 " disk as %s. True\n protection against single-disk"
1349 " failure might be compromised.\n",
1350 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1351 bdevname(same_pdev
->bdev
,b2
));
1353 /* Verify rdev->desc_nr is unique.
1354 * If it is -1, assign a free number, else
1355 * check number is not in use
1357 if (rdev
->desc_nr
< 0) {
1359 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1360 while (find_rdev_nr(mddev
, choice
))
1362 rdev
->desc_nr
= choice
;
1364 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1367 bdevname(rdev
->bdev
,b
);
1368 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1370 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1373 list_add(&rdev
->same_set
, &mddev
->disks
);
1374 rdev
->mddev
= mddev
;
1375 printk(KERN_INFO
"md: bind<%s>\n", b
);
1377 rdev
->kobj
.parent
= &mddev
->kobj
;
1378 kobject_add(&rdev
->kobj
);
1380 if (rdev
->bdev
->bd_part
)
1381 ko
= &rdev
->bdev
->bd_part
->kobj
;
1383 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1384 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1385 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1389 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1391 char b
[BDEVNAME_SIZE
];
1396 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1397 list_del_init(&rdev
->same_set
);
1398 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1400 sysfs_remove_link(&rdev
->kobj
, "block");
1401 kobject_del(&rdev
->kobj
);
1405 * prevent the device from being mounted, repartitioned or
1406 * otherwise reused by a RAID array (or any other kernel
1407 * subsystem), by bd_claiming the device.
1409 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1412 struct block_device
*bdev
;
1413 char b
[BDEVNAME_SIZE
];
1415 bdev
= open_partition_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1417 printk(KERN_ERR
"md: could not open %s.\n",
1418 __bdevname(dev
, b
));
1419 return PTR_ERR(bdev
);
1421 err
= bd_claim(bdev
, rdev
);
1423 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1425 blkdev_put_partition(bdev
);
1432 static void unlock_rdev(mdk_rdev_t
*rdev
)
1434 struct block_device
*bdev
= rdev
->bdev
;
1439 blkdev_put_partition(bdev
);
1442 void md_autodetect_dev(dev_t dev
);
1444 static void export_rdev(mdk_rdev_t
* rdev
)
1446 char b
[BDEVNAME_SIZE
];
1447 printk(KERN_INFO
"md: export_rdev(%s)\n",
1448 bdevname(rdev
->bdev
,b
));
1452 list_del_init(&rdev
->same_set
);
1454 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1457 kobject_put(&rdev
->kobj
);
1460 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1462 unbind_rdev_from_array(rdev
);
1466 static void export_array(mddev_t
*mddev
)
1468 struct list_head
*tmp
;
1471 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1476 kick_rdev_from_array(rdev
);
1478 if (!list_empty(&mddev
->disks
))
1480 mddev
->raid_disks
= 0;
1481 mddev
->major_version
= 0;
1484 static void print_desc(mdp_disk_t
*desc
)
1486 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1487 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1490 static void print_sb(mdp_super_t
*sb
)
1495 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1496 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1497 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1499 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1500 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1501 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1502 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1503 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1504 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1505 sb
->failed_disks
, sb
->spare_disks
,
1506 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1509 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1512 desc
= sb
->disks
+ i
;
1513 if (desc
->number
|| desc
->major
|| desc
->minor
||
1514 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1515 printk(" D %2d: ", i
);
1519 printk(KERN_INFO
"md: THIS: ");
1520 print_desc(&sb
->this_disk
);
1524 static void print_rdev(mdk_rdev_t
*rdev
)
1526 char b
[BDEVNAME_SIZE
];
1527 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1528 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1529 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1531 if (rdev
->sb_loaded
) {
1532 printk(KERN_INFO
"md: rdev superblock:\n");
1533 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1535 printk(KERN_INFO
"md: no rdev superblock!\n");
1538 static void md_print_devices(void)
1540 struct list_head
*tmp
, *tmp2
;
1543 char b
[BDEVNAME_SIZE
];
1546 printk("md: **********************************\n");
1547 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1548 printk("md: **********************************\n");
1549 ITERATE_MDDEV(mddev
,tmp
) {
1552 bitmap_print_sb(mddev
->bitmap
);
1554 printk("%s: ", mdname(mddev
));
1555 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1556 printk("<%s>", bdevname(rdev
->bdev
,b
));
1559 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1562 printk("md: **********************************\n");
1567 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1569 /* Update each superblock (in-memory image), but
1570 * if we are allowed to, skip spares which already
1571 * have the right event counter, or have one earlier
1572 * (which would mean they aren't being marked as dirty
1573 * with the rest of the array)
1576 struct list_head
*tmp
;
1578 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1579 if (rdev
->sb_events
== mddev
->events
||
1581 rdev
->raid_disk
< 0 &&
1582 (rdev
->sb_events
&1)==0 &&
1583 rdev
->sb_events
+1 == mddev
->events
)) {
1584 /* Don't update this superblock */
1585 rdev
->sb_loaded
= 2;
1587 super_types
[mddev
->major_version
].
1588 sync_super(mddev
, rdev
);
1589 rdev
->sb_loaded
= 1;
1594 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1597 struct list_head
*tmp
;
1603 spin_lock_irq(&mddev
->write_lock
);
1605 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1606 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1608 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1609 /* just a clean<-> dirty transition, possibly leave spares alone,
1610 * though if events isn't the right even/odd, we will have to do
1616 if (mddev
->degraded
)
1617 /* If the array is degraded, then skipping spares is both
1618 * dangerous and fairly pointless.
1619 * Dangerous because a device that was removed from the array
1620 * might have a event_count that still looks up-to-date,
1621 * so it can be re-added without a resync.
1622 * Pointless because if there are any spares to skip,
1623 * then a recovery will happen and soon that array won't
1624 * be degraded any more and the spare can go back to sleep then.
1628 sync_req
= mddev
->in_sync
;
1629 mddev
->utime
= get_seconds();
1631 /* If this is just a dirty<->clean transition, and the array is clean
1632 * and 'events' is odd, we can roll back to the previous clean state */
1634 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1635 && (mddev
->events
& 1))
1638 /* otherwise we have to go forward and ... */
1640 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1641 /* .. if the array isn't clean, insist on an odd 'events' */
1642 if ((mddev
->events
&1)==0) {
1647 /* otherwise insist on an even 'events' (for clean states) */
1648 if ((mddev
->events
&1)) {
1655 if (!mddev
->events
) {
1657 * oops, this 64-bit counter should never wrap.
1658 * Either we are in around ~1 trillion A.C., assuming
1659 * 1 reboot per second, or we have a bug:
1664 sync_sbs(mddev
, nospares
);
1667 * do not write anything to disk if using
1668 * nonpersistent superblocks
1670 if (!mddev
->persistent
) {
1671 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1672 spin_unlock_irq(&mddev
->write_lock
);
1673 wake_up(&mddev
->sb_wait
);
1676 spin_unlock_irq(&mddev
->write_lock
);
1679 "md: updating %s RAID superblock on device (in sync %d)\n",
1680 mdname(mddev
),mddev
->in_sync
);
1682 err
= bitmap_update_sb(mddev
->bitmap
);
1683 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1684 char b
[BDEVNAME_SIZE
];
1685 dprintk(KERN_INFO
"md: ");
1686 if (rdev
->sb_loaded
!= 1)
1687 continue; /* no noise on spare devices */
1688 if (test_bit(Faulty
, &rdev
->flags
))
1689 dprintk("(skipping faulty ");
1691 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1692 if (!test_bit(Faulty
, &rdev
->flags
)) {
1693 md_super_write(mddev
,rdev
,
1694 rdev
->sb_offset
<<1, rdev
->sb_size
,
1696 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1697 bdevname(rdev
->bdev
,b
),
1698 (unsigned long long)rdev
->sb_offset
);
1699 rdev
->sb_events
= mddev
->events
;
1703 if (mddev
->level
== LEVEL_MULTIPATH
)
1704 /* only need to write one superblock... */
1707 md_super_wait(mddev
);
1708 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1710 spin_lock_irq(&mddev
->write_lock
);
1711 if (mddev
->in_sync
!= sync_req
||
1712 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
1713 /* have to write it out again */
1714 spin_unlock_irq(&mddev
->write_lock
);
1717 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1718 spin_unlock_irq(&mddev
->write_lock
);
1719 wake_up(&mddev
->sb_wait
);
1723 /* words written to sysfs files may, or my not, be \n terminated.
1724 * We want to accept with case. For this we use cmd_match.
1726 static int cmd_match(const char *cmd
, const char *str
)
1728 /* See if cmd, written into a sysfs file, matches
1729 * str. They must either be the same, or cmd can
1730 * have a trailing newline
1732 while (*cmd
&& *str
&& *cmd
== *str
) {
1743 struct rdev_sysfs_entry
{
1744 struct attribute attr
;
1745 ssize_t (*show
)(mdk_rdev_t
*, char *);
1746 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1750 state_show(mdk_rdev_t
*rdev
, char *page
)
1755 if (test_bit(Faulty
, &rdev
->flags
)) {
1756 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1759 if (test_bit(In_sync
, &rdev
->flags
)) {
1760 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1763 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1764 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1767 if (!test_bit(Faulty
, &rdev
->flags
) &&
1768 !test_bit(In_sync
, &rdev
->flags
)) {
1769 len
+= sprintf(page
+len
, "%sspare", sep
);
1772 return len
+sprintf(page
+len
, "\n");
1776 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1779 * faulty - simulates and error
1780 * remove - disconnects the device
1781 * writemostly - sets write_mostly
1782 * -writemostly - clears write_mostly
1785 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1786 md_error(rdev
->mddev
, rdev
);
1788 } else if (cmd_match(buf
, "remove")) {
1789 if (rdev
->raid_disk
>= 0)
1792 mddev_t
*mddev
= rdev
->mddev
;
1793 kick_rdev_from_array(rdev
);
1794 md_update_sb(mddev
, 1);
1795 md_new_event(mddev
);
1798 } else if (cmd_match(buf
, "writemostly")) {
1799 set_bit(WriteMostly
, &rdev
->flags
);
1801 } else if (cmd_match(buf
, "-writemostly")) {
1802 clear_bit(WriteMostly
, &rdev
->flags
);
1805 return err
? err
: len
;
1807 static struct rdev_sysfs_entry rdev_state
=
1808 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
1811 super_show(mdk_rdev_t
*rdev
, char *page
)
1813 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1814 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1815 return rdev
->sb_size
;
1819 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1822 errors_show(mdk_rdev_t
*rdev
, char *page
)
1824 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1828 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1831 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1832 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1833 atomic_set(&rdev
->corrected_errors
, n
);
1838 static struct rdev_sysfs_entry rdev_errors
=
1839 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
1842 slot_show(mdk_rdev_t
*rdev
, char *page
)
1844 if (rdev
->raid_disk
< 0)
1845 return sprintf(page
, "none\n");
1847 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1851 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1854 int slot
= simple_strtoul(buf
, &e
, 10);
1855 if (strncmp(buf
, "none", 4)==0)
1857 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1859 if (rdev
->mddev
->pers
)
1860 /* Cannot set slot in active array (yet) */
1862 if (slot
>= rdev
->mddev
->raid_disks
)
1864 rdev
->raid_disk
= slot
;
1865 /* assume it is working */
1867 set_bit(In_sync
, &rdev
->flags
);
1872 static struct rdev_sysfs_entry rdev_slot
=
1873 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
1876 offset_show(mdk_rdev_t
*rdev
, char *page
)
1878 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1882 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1885 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1886 if (e
==buf
|| (*e
&& *e
!= '\n'))
1888 if (rdev
->mddev
->pers
)
1890 rdev
->data_offset
= offset
;
1894 static struct rdev_sysfs_entry rdev_offset
=
1895 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
1898 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1900 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1904 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1907 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1908 if (e
==buf
|| (*e
&& *e
!= '\n'))
1910 if (rdev
->mddev
->pers
)
1913 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1914 rdev
->mddev
->size
= size
;
1918 static struct rdev_sysfs_entry rdev_size
=
1919 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
1921 static struct attribute
*rdev_default_attrs
[] = {
1931 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1933 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1934 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1938 return entry
->show(rdev
, page
);
1942 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1943 const char *page
, size_t length
)
1945 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1946 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1950 if (!capable(CAP_SYS_ADMIN
))
1952 return entry
->store(rdev
, page
, length
);
1955 static void rdev_free(struct kobject
*ko
)
1957 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1960 static struct sysfs_ops rdev_sysfs_ops
= {
1961 .show
= rdev_attr_show
,
1962 .store
= rdev_attr_store
,
1964 static struct kobj_type rdev_ktype
= {
1965 .release
= rdev_free
,
1966 .sysfs_ops
= &rdev_sysfs_ops
,
1967 .default_attrs
= rdev_default_attrs
,
1971 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1973 * mark the device faulty if:
1975 * - the device is nonexistent (zero size)
1976 * - the device has no valid superblock
1978 * a faulty rdev _never_ has rdev->sb set.
1980 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1982 char b
[BDEVNAME_SIZE
];
1987 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1989 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1990 return ERR_PTR(-ENOMEM
);
1993 if ((err
= alloc_disk_sb(rdev
)))
1996 err
= lock_rdev(rdev
, newdev
);
2000 rdev
->kobj
.parent
= NULL
;
2001 rdev
->kobj
.ktype
= &rdev_ktype
;
2002 kobject_init(&rdev
->kobj
);
2006 rdev
->data_offset
= 0;
2007 rdev
->sb_events
= 0;
2008 atomic_set(&rdev
->nr_pending
, 0);
2009 atomic_set(&rdev
->read_errors
, 0);
2010 atomic_set(&rdev
->corrected_errors
, 0);
2012 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2015 "md: %s has zero or unknown size, marking faulty!\n",
2016 bdevname(rdev
->bdev
,b
));
2021 if (super_format
>= 0) {
2022 err
= super_types
[super_format
].
2023 load_super(rdev
, NULL
, super_minor
);
2024 if (err
== -EINVAL
) {
2026 "md: %s has invalid sb, not importing!\n",
2027 bdevname(rdev
->bdev
,b
));
2032 "md: could not read %s's sb, not importing!\n",
2033 bdevname(rdev
->bdev
,b
));
2037 INIT_LIST_HEAD(&rdev
->same_set
);
2042 if (rdev
->sb_page
) {
2048 return ERR_PTR(err
);
2052 * Check a full RAID array for plausibility
2056 static void analyze_sbs(mddev_t
* mddev
)
2059 struct list_head
*tmp
;
2060 mdk_rdev_t
*rdev
, *freshest
;
2061 char b
[BDEVNAME_SIZE
];
2064 ITERATE_RDEV(mddev
,rdev
,tmp
)
2065 switch (super_types
[mddev
->major_version
].
2066 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2074 "md: fatal superblock inconsistency in %s"
2075 " -- removing from array\n",
2076 bdevname(rdev
->bdev
,b
));
2077 kick_rdev_from_array(rdev
);
2081 super_types
[mddev
->major_version
].
2082 validate_super(mddev
, freshest
);
2085 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2086 if (rdev
!= freshest
)
2087 if (super_types
[mddev
->major_version
].
2088 validate_super(mddev
, rdev
)) {
2089 printk(KERN_WARNING
"md: kicking non-fresh %s"
2091 bdevname(rdev
->bdev
,b
));
2092 kick_rdev_from_array(rdev
);
2095 if (mddev
->level
== LEVEL_MULTIPATH
) {
2096 rdev
->desc_nr
= i
++;
2097 rdev
->raid_disk
= rdev
->desc_nr
;
2098 set_bit(In_sync
, &rdev
->flags
);
2104 if (mddev
->recovery_cp
!= MaxSector
&&
2106 printk(KERN_ERR
"md: %s: raid array is not clean"
2107 " -- starting background reconstruction\n",
2113 safe_delay_show(mddev_t
*mddev
, char *page
)
2115 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2116 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2119 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2127 /* remove a period, and count digits after it */
2128 if (len
>= sizeof(buf
))
2130 strlcpy(buf
, cbuf
, len
);
2132 for (i
=0; i
<len
; i
++) {
2134 if (isdigit(buf
[i
])) {
2139 } else if (buf
[i
] == '.') {
2144 msec
= simple_strtoul(buf
, &e
, 10);
2145 if (e
== buf
|| (*e
&& *e
!= '\n'))
2147 msec
= (msec
* 1000) / scale
;
2149 mddev
->safemode_delay
= 0;
2151 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2152 if (mddev
->safemode_delay
== 0)
2153 mddev
->safemode_delay
= 1;
2157 static struct md_sysfs_entry md_safe_delay
=
2158 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2161 level_show(mddev_t
*mddev
, char *page
)
2163 struct mdk_personality
*p
= mddev
->pers
;
2165 return sprintf(page
, "%s\n", p
->name
);
2166 else if (mddev
->clevel
[0])
2167 return sprintf(page
, "%s\n", mddev
->clevel
);
2168 else if (mddev
->level
!= LEVEL_NONE
)
2169 return sprintf(page
, "%d\n", mddev
->level
);
2175 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2182 if (len
>= sizeof(mddev
->clevel
))
2184 strncpy(mddev
->clevel
, buf
, len
);
2185 if (mddev
->clevel
[len
-1] == '\n')
2187 mddev
->clevel
[len
] = 0;
2188 mddev
->level
= LEVEL_NONE
;
2192 static struct md_sysfs_entry md_level
=
2193 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2197 layout_show(mddev_t
*mddev
, char *page
)
2199 /* just a number, not meaningful for all levels */
2200 return sprintf(page
, "%d\n", mddev
->layout
);
2204 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2207 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2211 if (!*buf
|| (*e
&& *e
!= '\n'))
2217 static struct md_sysfs_entry md_layout
=
2218 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2222 raid_disks_show(mddev_t
*mddev
, char *page
)
2224 if (mddev
->raid_disks
== 0)
2226 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2229 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2232 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2234 /* can only set raid_disks if array is not yet active */
2237 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2239 if (!*buf
|| (*e
&& *e
!= '\n'))
2243 rv
= update_raid_disks(mddev
, n
);
2245 mddev
->raid_disks
= n
;
2246 return rv
? rv
: len
;
2248 static struct md_sysfs_entry md_raid_disks
=
2249 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2252 chunk_size_show(mddev_t
*mddev
, char *page
)
2254 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2258 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2260 /* can only set chunk_size if array is not yet active */
2262 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2266 if (!*buf
|| (*e
&& *e
!= '\n'))
2269 mddev
->chunk_size
= n
;
2272 static struct md_sysfs_entry md_chunk_size
=
2273 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2276 resync_start_show(mddev_t
*mddev
, char *page
)
2278 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2282 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2284 /* can only set chunk_size if array is not yet active */
2286 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2290 if (!*buf
|| (*e
&& *e
!= '\n'))
2293 mddev
->recovery_cp
= n
;
2296 static struct md_sysfs_entry md_resync_start
=
2297 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2300 * The array state can be:
2303 * No devices, no size, no level
2304 * Equivalent to STOP_ARRAY ioctl
2306 * May have some settings, but array is not active
2307 * all IO results in error
2308 * When written, doesn't tear down array, but just stops it
2309 * suspended (not supported yet)
2310 * All IO requests will block. The array can be reconfigured.
2311 * Writing this, if accepted, will block until array is quiessent
2313 * no resync can happen. no superblocks get written.
2314 * write requests fail
2316 * like readonly, but behaves like 'clean' on a write request.
2318 * clean - no pending writes, but otherwise active.
2319 * When written to inactive array, starts without resync
2320 * If a write request arrives then
2321 * if metadata is known, mark 'dirty' and switch to 'active'.
2322 * if not known, block and switch to write-pending
2323 * If written to an active array that has pending writes, then fails.
2325 * fully active: IO and resync can be happening.
2326 * When written to inactive array, starts with resync
2329 * clean, but writes are blocked waiting for 'active' to be written.
2332 * like active, but no writes have been seen for a while (100msec).
2335 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2336 write_pending
, active_idle
, bad_word
};
2337 static char *array_states
[] = {
2338 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2339 "write-pending", "active-idle", NULL
};
2341 static int match_word(const char *word
, char **list
)
2344 for (n
=0; list
[n
]; n
++)
2345 if (cmd_match(word
, list
[n
]))
2351 array_state_show(mddev_t
*mddev
, char *page
)
2353 enum array_state st
= inactive
;
2366 else if (mddev
->safemode
)
2372 if (list_empty(&mddev
->disks
) &&
2373 mddev
->raid_disks
== 0 &&
2379 return sprintf(page
, "%s\n", array_states
[st
]);
2382 static int do_md_stop(mddev_t
* mddev
, int ro
);
2383 static int do_md_run(mddev_t
* mddev
);
2384 static int restart_array(mddev_t
*mddev
);
2387 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2390 enum array_state st
= match_word(buf
, array_states
);
2395 /* stopping an active array */
2397 if (atomic_read(&mddev
->active
) > 1)
2399 err
= do_md_stop(mddev
, 0);
2403 /* stopping an active array */
2405 if (atomic_read(&mddev
->active
) > 1)
2407 err
= do_md_stop(mddev
, 2);
2411 break; /* not supported yet */
2414 err
= do_md_stop(mddev
, 1);
2417 err
= do_md_run(mddev
);
2421 /* stopping an active array */
2423 err
= do_md_stop(mddev
, 1);
2425 mddev
->ro
= 2; /* FIXME mark devices writable */
2428 err
= do_md_run(mddev
);
2433 restart_array(mddev
);
2434 spin_lock_irq(&mddev
->write_lock
);
2435 if (atomic_read(&mddev
->writes_pending
) == 0) {
2437 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2439 spin_unlock_irq(&mddev
->write_lock
);
2442 mddev
->recovery_cp
= MaxSector
;
2443 err
= do_md_run(mddev
);
2448 restart_array(mddev
);
2449 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2450 wake_up(&mddev
->sb_wait
);
2454 err
= do_md_run(mddev
);
2459 /* these cannot be set */
2467 static struct md_sysfs_entry md_array_state
=
2468 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
2471 null_show(mddev_t
*mddev
, char *page
)
2477 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2479 /* buf must be %d:%d\n? giving major and minor numbers */
2480 /* The new device is added to the array.
2481 * If the array has a persistent superblock, we read the
2482 * superblock to initialise info and check validity.
2483 * Otherwise, only checking done is that in bind_rdev_to_array,
2484 * which mainly checks size.
2487 int major
= simple_strtoul(buf
, &e
, 10);
2493 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2495 minor
= simple_strtoul(e
+1, &e
, 10);
2496 if (*e
&& *e
!= '\n')
2498 dev
= MKDEV(major
, minor
);
2499 if (major
!= MAJOR(dev
) ||
2500 minor
!= MINOR(dev
))
2504 if (mddev
->persistent
) {
2505 rdev
= md_import_device(dev
, mddev
->major_version
,
2506 mddev
->minor_version
);
2507 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2508 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2509 mdk_rdev_t
, same_set
);
2510 err
= super_types
[mddev
->major_version
]
2511 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2516 rdev
= md_import_device(dev
, -1, -1);
2519 return PTR_ERR(rdev
);
2520 err
= bind_rdev_to_array(rdev
, mddev
);
2524 return err
? err
: len
;
2527 static struct md_sysfs_entry md_new_device
=
2528 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
2531 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2534 unsigned long chunk
, end_chunk
;
2538 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2540 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
2541 if (buf
== end
) break;
2542 if (*end
== '-') { /* range */
2544 end_chunk
= simple_strtoul(buf
, &end
, 0);
2545 if (buf
== end
) break;
2547 if (*end
&& !isspace(*end
)) break;
2548 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
2550 while (isspace(*buf
)) buf
++;
2552 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
2557 static struct md_sysfs_entry md_bitmap
=
2558 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
2561 size_show(mddev_t
*mddev
, char *page
)
2563 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2566 static int update_size(mddev_t
*mddev
, unsigned long size
);
2569 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2571 /* If array is inactive, we can reduce the component size, but
2572 * not increase it (except from 0).
2573 * If array is active, we can try an on-line resize
2577 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2578 if (!*buf
|| *buf
== '\n' ||
2583 err
= update_size(mddev
, size
);
2584 md_update_sb(mddev
, 1);
2586 if (mddev
->size
== 0 ||
2592 return err
? err
: len
;
2595 static struct md_sysfs_entry md_size
=
2596 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
2600 * This is either 'none' for arrays with externally managed metadata,
2601 * or N.M for internally known formats
2604 metadata_show(mddev_t
*mddev
, char *page
)
2606 if (mddev
->persistent
)
2607 return sprintf(page
, "%d.%d\n",
2608 mddev
->major_version
, mddev
->minor_version
);
2610 return sprintf(page
, "none\n");
2614 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2618 if (!list_empty(&mddev
->disks
))
2621 if (cmd_match(buf
, "none")) {
2622 mddev
->persistent
= 0;
2623 mddev
->major_version
= 0;
2624 mddev
->minor_version
= 90;
2627 major
= simple_strtoul(buf
, &e
, 10);
2628 if (e
==buf
|| *e
!= '.')
2631 minor
= simple_strtoul(buf
, &e
, 10);
2632 if (e
==buf
|| *e
!= '\n')
2634 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2635 super_types
[major
].name
== NULL
)
2637 mddev
->major_version
= major
;
2638 mddev
->minor_version
= minor
;
2639 mddev
->persistent
= 1;
2643 static struct md_sysfs_entry md_metadata
=
2644 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2647 action_show(mddev_t
*mddev
, char *page
)
2649 char *type
= "idle";
2650 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2651 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2652 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2654 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2655 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2657 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2664 return sprintf(page
, "%s\n", type
);
2668 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2670 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2673 if (cmd_match(page
, "idle")) {
2674 if (mddev
->sync_thread
) {
2675 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2676 md_unregister_thread(mddev
->sync_thread
);
2677 mddev
->sync_thread
= NULL
;
2678 mddev
->recovery
= 0;
2680 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2681 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2683 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2684 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2685 else if (cmd_match(page
, "reshape")) {
2687 if (mddev
->pers
->start_reshape
== NULL
)
2689 err
= mddev
->pers
->start_reshape(mddev
);
2693 if (cmd_match(page
, "check"))
2694 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2695 else if (!cmd_match(page
, "repair"))
2697 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2698 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2700 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2701 md_wakeup_thread(mddev
->thread
);
2706 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2708 return sprintf(page
, "%llu\n",
2709 (unsigned long long) mddev
->resync_mismatches
);
2712 static struct md_sysfs_entry md_scan_mode
=
2713 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2716 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
2719 sync_min_show(mddev_t
*mddev
, char *page
)
2721 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2722 mddev
->sync_speed_min
? "local": "system");
2726 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2730 if (strncmp(buf
, "system", 6)==0) {
2731 mddev
->sync_speed_min
= 0;
2734 min
= simple_strtoul(buf
, &e
, 10);
2735 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2737 mddev
->sync_speed_min
= min
;
2741 static struct md_sysfs_entry md_sync_min
=
2742 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2745 sync_max_show(mddev_t
*mddev
, char *page
)
2747 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2748 mddev
->sync_speed_max
? "local": "system");
2752 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2756 if (strncmp(buf
, "system", 6)==0) {
2757 mddev
->sync_speed_max
= 0;
2760 max
= simple_strtoul(buf
, &e
, 10);
2761 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2763 mddev
->sync_speed_max
= max
;
2767 static struct md_sysfs_entry md_sync_max
=
2768 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2772 sync_speed_show(mddev_t
*mddev
, char *page
)
2774 unsigned long resync
, dt
, db
;
2775 resync
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
));
2776 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2778 db
= resync
- (mddev
->resync_mark_cnt
);
2779 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2782 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
2785 sync_completed_show(mddev_t
*mddev
, char *page
)
2787 unsigned long max_blocks
, resync
;
2789 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2790 max_blocks
= mddev
->resync_max_sectors
;
2792 max_blocks
= mddev
->size
<< 1;
2794 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2795 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2798 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
2801 suspend_lo_show(mddev_t
*mddev
, char *page
)
2803 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2807 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2810 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2812 if (mddev
->pers
->quiesce
== NULL
)
2814 if (buf
== e
|| (*e
&& *e
!= '\n'))
2816 if (new >= mddev
->suspend_hi
||
2817 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2818 mddev
->suspend_lo
= new;
2819 mddev
->pers
->quiesce(mddev
, 2);
2824 static struct md_sysfs_entry md_suspend_lo
=
2825 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2829 suspend_hi_show(mddev_t
*mddev
, char *page
)
2831 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2835 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2838 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2840 if (mddev
->pers
->quiesce
== NULL
)
2842 if (buf
== e
|| (*e
&& *e
!= '\n'))
2844 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2845 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2846 mddev
->suspend_hi
= new;
2847 mddev
->pers
->quiesce(mddev
, 1);
2848 mddev
->pers
->quiesce(mddev
, 0);
2853 static struct md_sysfs_entry md_suspend_hi
=
2854 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2857 static struct attribute
*md_default_attrs
[] = {
2860 &md_raid_disks
.attr
,
2861 &md_chunk_size
.attr
,
2863 &md_resync_start
.attr
,
2865 &md_new_device
.attr
,
2866 &md_safe_delay
.attr
,
2867 &md_array_state
.attr
,
2871 static struct attribute
*md_redundancy_attrs
[] = {
2873 &md_mismatches
.attr
,
2876 &md_sync_speed
.attr
,
2877 &md_sync_completed
.attr
,
2878 &md_suspend_lo
.attr
,
2879 &md_suspend_hi
.attr
,
2883 static struct attribute_group md_redundancy_group
= {
2885 .attrs
= md_redundancy_attrs
,
2890 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2892 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2893 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2898 rv
= mddev_lock(mddev
);
2900 rv
= entry
->show(mddev
, page
);
2901 mddev_unlock(mddev
);
2907 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2908 const char *page
, size_t length
)
2910 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2911 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2916 if (!capable(CAP_SYS_ADMIN
))
2918 rv
= mddev_lock(mddev
);
2920 rv
= entry
->store(mddev
, page
, length
);
2921 mddev_unlock(mddev
);
2926 static void md_free(struct kobject
*ko
)
2928 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2932 static struct sysfs_ops md_sysfs_ops
= {
2933 .show
= md_attr_show
,
2934 .store
= md_attr_store
,
2936 static struct kobj_type md_ktype
= {
2938 .sysfs_ops
= &md_sysfs_ops
,
2939 .default_attrs
= md_default_attrs
,
2944 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2946 static DEFINE_MUTEX(disks_mutex
);
2947 mddev_t
*mddev
= mddev_find(dev
);
2948 struct gendisk
*disk
;
2949 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2950 int shift
= partitioned
? MdpMinorShift
: 0;
2951 int unit
= MINOR(dev
) >> shift
;
2956 mutex_lock(&disks_mutex
);
2957 if (mddev
->gendisk
) {
2958 mutex_unlock(&disks_mutex
);
2962 disk
= alloc_disk(1 << shift
);
2964 mutex_unlock(&disks_mutex
);
2968 disk
->major
= MAJOR(dev
);
2969 disk
->first_minor
= unit
<< shift
;
2971 sprintf(disk
->disk_name
, "md_d%d", unit
);
2973 sprintf(disk
->disk_name
, "md%d", unit
);
2974 disk
->fops
= &md_fops
;
2975 disk
->private_data
= mddev
;
2976 disk
->queue
= mddev
->queue
;
2978 mddev
->gendisk
= disk
;
2979 mutex_unlock(&disks_mutex
);
2980 mddev
->kobj
.parent
= &disk
->kobj
;
2981 mddev
->kobj
.k_name
= NULL
;
2982 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2983 mddev
->kobj
.ktype
= &md_ktype
;
2984 kobject_register(&mddev
->kobj
);
2988 static void md_safemode_timeout(unsigned long data
)
2990 mddev_t
*mddev
= (mddev_t
*) data
;
2992 mddev
->safemode
= 1;
2993 md_wakeup_thread(mddev
->thread
);
2996 static int start_dirty_degraded
;
2998 static int do_md_run(mddev_t
* mddev
)
3002 struct list_head
*tmp
;
3004 struct gendisk
*disk
;
3005 struct mdk_personality
*pers
;
3006 char b
[BDEVNAME_SIZE
];
3008 if (list_empty(&mddev
->disks
))
3009 /* cannot run an array with no devices.. */
3016 * Analyze all RAID superblock(s)
3018 if (!mddev
->raid_disks
)
3021 chunk_size
= mddev
->chunk_size
;
3024 if (chunk_size
> MAX_CHUNK_SIZE
) {
3025 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
3026 chunk_size
, MAX_CHUNK_SIZE
);
3030 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3032 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
3033 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
3036 if (chunk_size
< PAGE_SIZE
) {
3037 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
3038 chunk_size
, PAGE_SIZE
);
3042 /* devices must have minimum size of one chunk */
3043 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3044 if (test_bit(Faulty
, &rdev
->flags
))
3046 if (rdev
->size
< chunk_size
/ 1024) {
3048 "md: Dev %s smaller than chunk_size:"
3050 bdevname(rdev
->bdev
,b
),
3051 (unsigned long long)rdev
->size
,
3059 if (mddev
->level
!= LEVEL_NONE
)
3060 request_module("md-level-%d", mddev
->level
);
3061 else if (mddev
->clevel
[0])
3062 request_module("md-%s", mddev
->clevel
);
3066 * Drop all container device buffers, from now on
3067 * the only valid external interface is through the md
3069 * Also find largest hardsector size
3071 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3072 if (test_bit(Faulty
, &rdev
->flags
))
3074 sync_blockdev(rdev
->bdev
);
3075 invalidate_bdev(rdev
->bdev
, 0);
3078 md_probe(mddev
->unit
, NULL
, NULL
);
3079 disk
= mddev
->gendisk
;
3083 spin_lock(&pers_lock
);
3084 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3085 if (!pers
|| !try_module_get(pers
->owner
)) {
3086 spin_unlock(&pers_lock
);
3087 if (mddev
->level
!= LEVEL_NONE
)
3088 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3091 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3096 spin_unlock(&pers_lock
);
3097 mddev
->level
= pers
->level
;
3098 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3100 if (mddev
->reshape_position
!= MaxSector
&&
3101 pers
->start_reshape
== NULL
) {
3102 /* This personality cannot handle reshaping... */
3104 module_put(pers
->owner
);
3108 mddev
->recovery
= 0;
3109 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3110 mddev
->barriers_work
= 1;
3111 mddev
->ok_start_degraded
= start_dirty_degraded
;
3114 mddev
->ro
= 2; /* read-only, but switch on first write */
3116 err
= mddev
->pers
->run(mddev
);
3117 if (!err
&& mddev
->pers
->sync_request
) {
3118 err
= bitmap_create(mddev
);
3120 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3121 mdname(mddev
), err
);
3122 mddev
->pers
->stop(mddev
);
3126 printk(KERN_ERR
"md: pers->run() failed ...\n");
3127 module_put(mddev
->pers
->owner
);
3129 bitmap_destroy(mddev
);
3132 if (mddev
->pers
->sync_request
)
3133 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
3134 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3137 atomic_set(&mddev
->writes_pending
,0);
3138 mddev
->safemode
= 0;
3139 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3140 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3141 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3144 ITERATE_RDEV(mddev
,rdev
,tmp
)
3145 if (rdev
->raid_disk
>= 0) {
3147 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3148 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
3151 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3154 md_update_sb(mddev
, 0);
3156 set_capacity(disk
, mddev
->array_size
<<1);
3158 /* If we call blk_queue_make_request here, it will
3159 * re-initialise max_sectors etc which may have been
3160 * refined inside -> run. So just set the bits we need to set.
3161 * Most initialisation happended when we called
3162 * blk_queue_make_request(..., md_fail_request)
3165 mddev
->queue
->queuedata
= mddev
;
3166 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3168 /* If there is a partially-recovered drive we need to
3169 * start recovery here. If we leave it to md_check_recovery,
3170 * it will remove the drives and not do the right thing
3172 if (mddev
->degraded
&& !mddev
->sync_thread
) {
3173 struct list_head
*rtmp
;
3175 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3176 if (rdev
->raid_disk
>= 0 &&
3177 !test_bit(In_sync
, &rdev
->flags
) &&
3178 !test_bit(Faulty
, &rdev
->flags
))
3179 /* complete an interrupted recovery */
3181 if (spares
&& mddev
->pers
->sync_request
) {
3182 mddev
->recovery
= 0;
3183 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3184 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3187 if (!mddev
->sync_thread
) {
3188 printk(KERN_ERR
"%s: could not start resync"
3191 /* leave the spares where they are, it shouldn't hurt */
3192 mddev
->recovery
= 0;
3196 md_wakeup_thread(mddev
->thread
);
3197 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
3200 md_new_event(mddev
);
3204 static int restart_array(mddev_t
*mddev
)
3206 struct gendisk
*disk
= mddev
->gendisk
;
3210 * Complain if it has no devices
3213 if (list_empty(&mddev
->disks
))
3221 mddev
->safemode
= 0;
3223 set_disk_ro(disk
, 0);
3225 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3228 * Kick recovery or resync if necessary
3230 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3231 md_wakeup_thread(mddev
->thread
);
3232 md_wakeup_thread(mddev
->sync_thread
);
3241 /* similar to deny_write_access, but accounts for our holding a reference
3242 * to the file ourselves */
3243 static int deny_bitmap_write_access(struct file
* file
)
3245 struct inode
*inode
= file
->f_mapping
->host
;
3247 spin_lock(&inode
->i_lock
);
3248 if (atomic_read(&inode
->i_writecount
) > 1) {
3249 spin_unlock(&inode
->i_lock
);
3252 atomic_set(&inode
->i_writecount
, -1);
3253 spin_unlock(&inode
->i_lock
);
3258 static void restore_bitmap_write_access(struct file
*file
)
3260 struct inode
*inode
= file
->f_mapping
->host
;
3262 spin_lock(&inode
->i_lock
);
3263 atomic_set(&inode
->i_writecount
, 1);
3264 spin_unlock(&inode
->i_lock
);
3268 * 0 - completely stop and dis-assemble array
3269 * 1 - switch to readonly
3270 * 2 - stop but do not disassemble array
3272 static int do_md_stop(mddev_t
* mddev
, int mode
)
3275 struct gendisk
*disk
= mddev
->gendisk
;
3278 if (atomic_read(&mddev
->active
)>2) {
3279 printk("md: %s still in use.\n",mdname(mddev
));
3283 if (mddev
->sync_thread
) {
3284 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3285 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3286 md_unregister_thread(mddev
->sync_thread
);
3287 mddev
->sync_thread
= NULL
;
3290 del_timer_sync(&mddev
->safemode_timer
);
3292 invalidate_partition(disk
, 0);
3295 case 1: /* readonly */
3301 case 0: /* disassemble */
3303 bitmap_flush(mddev
);
3304 md_super_wait(mddev
);
3306 set_disk_ro(disk
, 0);
3307 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3308 mddev
->pers
->stop(mddev
);
3309 if (mddev
->pers
->sync_request
)
3310 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3312 module_put(mddev
->pers
->owner
);
3317 if (!mddev
->in_sync
|| mddev
->flags
) {
3318 /* mark array as shutdown cleanly */
3320 md_update_sb(mddev
, 1);
3323 set_disk_ro(disk
, 1);
3324 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3328 * Free resources if final stop
3332 struct list_head
*tmp
;
3333 struct gendisk
*disk
;
3334 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3336 bitmap_destroy(mddev
);
3337 if (mddev
->bitmap_file
) {
3338 restore_bitmap_write_access(mddev
->bitmap_file
);
3339 fput(mddev
->bitmap_file
);
3340 mddev
->bitmap_file
= NULL
;
3342 mddev
->bitmap_offset
= 0;
3344 ITERATE_RDEV(mddev
,rdev
,tmp
)
3345 if (rdev
->raid_disk
>= 0) {
3347 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3348 sysfs_remove_link(&mddev
->kobj
, nm
);
3351 export_array(mddev
);
3353 mddev
->array_size
= 0;
3355 mddev
->raid_disks
= 0;
3356 mddev
->recovery_cp
= 0;
3358 disk
= mddev
->gendisk
;
3360 set_capacity(disk
, 0);
3362 } else if (mddev
->pers
)
3363 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3366 md_new_event(mddev
);
3371 static void autorun_array(mddev_t
*mddev
)
3374 struct list_head
*tmp
;
3377 if (list_empty(&mddev
->disks
))
3380 printk(KERN_INFO
"md: running: ");
3382 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3383 char b
[BDEVNAME_SIZE
];
3384 printk("<%s>", bdevname(rdev
->bdev
,b
));
3388 err
= do_md_run (mddev
);
3390 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3391 do_md_stop (mddev
, 0);
3396 * lets try to run arrays based on all disks that have arrived
3397 * until now. (those are in pending_raid_disks)
3399 * the method: pick the first pending disk, collect all disks with
3400 * the same UUID, remove all from the pending list and put them into
3401 * the 'same_array' list. Then order this list based on superblock
3402 * update time (freshest comes first), kick out 'old' disks and
3403 * compare superblocks. If everything's fine then run it.
3405 * If "unit" is allocated, then bump its reference count
3407 static void autorun_devices(int part
)
3409 struct list_head
*tmp
;
3410 mdk_rdev_t
*rdev0
, *rdev
;
3412 char b
[BDEVNAME_SIZE
];
3414 printk(KERN_INFO
"md: autorun ...\n");
3415 while (!list_empty(&pending_raid_disks
)) {
3418 LIST_HEAD(candidates
);
3419 rdev0
= list_entry(pending_raid_disks
.next
,
3420 mdk_rdev_t
, same_set
);
3422 printk(KERN_INFO
"md: considering %s ...\n",
3423 bdevname(rdev0
->bdev
,b
));
3424 INIT_LIST_HEAD(&candidates
);
3425 ITERATE_RDEV_PENDING(rdev
,tmp
)
3426 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3427 printk(KERN_INFO
"md: adding %s ...\n",
3428 bdevname(rdev
->bdev
,b
));
3429 list_move(&rdev
->same_set
, &candidates
);
3432 * now we have a set of devices, with all of them having
3433 * mostly sane superblocks. It's time to allocate the
3437 dev
= MKDEV(mdp_major
,
3438 rdev0
->preferred_minor
<< MdpMinorShift
);
3439 unit
= MINOR(dev
) >> MdpMinorShift
;
3441 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3444 if (rdev0
->preferred_minor
!= unit
) {
3445 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3446 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3450 md_probe(dev
, NULL
, NULL
);
3451 mddev
= mddev_find(dev
);
3454 "md: cannot allocate memory for md drive.\n");
3457 if (mddev_lock(mddev
))
3458 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3460 else if (mddev
->raid_disks
|| mddev
->major_version
3461 || !list_empty(&mddev
->disks
)) {
3463 "md: %s already running, cannot run %s\n",
3464 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3465 mddev_unlock(mddev
);
3467 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3468 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3469 list_del_init(&rdev
->same_set
);
3470 if (bind_rdev_to_array(rdev
, mddev
))
3473 autorun_array(mddev
);
3474 mddev_unlock(mddev
);
3476 /* on success, candidates will be empty, on error
3479 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3483 printk(KERN_INFO
"md: ... autorun DONE.\n");
3486 static int get_version(void __user
* arg
)
3490 ver
.major
= MD_MAJOR_VERSION
;
3491 ver
.minor
= MD_MINOR_VERSION
;
3492 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3494 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3500 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3502 mdu_array_info_t info
;
3503 int nr
,working
,active
,failed
,spare
;
3505 struct list_head
*tmp
;
3507 nr
=working
=active
=failed
=spare
=0;
3508 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3510 if (test_bit(Faulty
, &rdev
->flags
))
3514 if (test_bit(In_sync
, &rdev
->flags
))
3521 info
.major_version
= mddev
->major_version
;
3522 info
.minor_version
= mddev
->minor_version
;
3523 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3524 info
.ctime
= mddev
->ctime
;
3525 info
.level
= mddev
->level
;
3526 info
.size
= mddev
->size
;
3527 if (info
.size
!= mddev
->size
) /* overflow */
3530 info
.raid_disks
= mddev
->raid_disks
;
3531 info
.md_minor
= mddev
->md_minor
;
3532 info
.not_persistent
= !mddev
->persistent
;
3534 info
.utime
= mddev
->utime
;
3537 info
.state
= (1<<MD_SB_CLEAN
);
3538 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3539 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3540 info
.active_disks
= active
;
3541 info
.working_disks
= working
;
3542 info
.failed_disks
= failed
;
3543 info
.spare_disks
= spare
;
3545 info
.layout
= mddev
->layout
;
3546 info
.chunk_size
= mddev
->chunk_size
;
3548 if (copy_to_user(arg
, &info
, sizeof(info
)))
3554 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3556 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3557 char *ptr
, *buf
= NULL
;
3560 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3564 /* bitmap disabled, zero the first byte and copy out */
3565 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3566 file
->pathname
[0] = '\0';
3570 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3574 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3578 strcpy(file
->pathname
, ptr
);
3582 if (copy_to_user(arg
, file
, sizeof(*file
)))
3590 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3592 mdu_disk_info_t info
;
3596 if (copy_from_user(&info
, arg
, sizeof(info
)))
3601 rdev
= find_rdev_nr(mddev
, nr
);
3603 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3604 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3605 info
.raid_disk
= rdev
->raid_disk
;
3607 if (test_bit(Faulty
, &rdev
->flags
))
3608 info
.state
|= (1<<MD_DISK_FAULTY
);
3609 else if (test_bit(In_sync
, &rdev
->flags
)) {
3610 info
.state
|= (1<<MD_DISK_ACTIVE
);
3611 info
.state
|= (1<<MD_DISK_SYNC
);
3613 if (test_bit(WriteMostly
, &rdev
->flags
))
3614 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3616 info
.major
= info
.minor
= 0;
3617 info
.raid_disk
= -1;
3618 info
.state
= (1<<MD_DISK_REMOVED
);
3621 if (copy_to_user(arg
, &info
, sizeof(info
)))
3627 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3629 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3631 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3633 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3636 if (!mddev
->raid_disks
) {
3638 /* expecting a device which has a superblock */
3639 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3642 "md: md_import_device returned %ld\n",
3644 return PTR_ERR(rdev
);
3646 if (!list_empty(&mddev
->disks
)) {
3647 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3648 mdk_rdev_t
, same_set
);
3649 int err
= super_types
[mddev
->major_version
]
3650 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3653 "md: %s has different UUID to %s\n",
3654 bdevname(rdev
->bdev
,b
),
3655 bdevname(rdev0
->bdev
,b2
));
3660 err
= bind_rdev_to_array(rdev
, mddev
);
3667 * add_new_disk can be used once the array is assembled
3668 * to add "hot spares". They must already have a superblock
3673 if (!mddev
->pers
->hot_add_disk
) {
3675 "%s: personality does not support diskops!\n",
3679 if (mddev
->persistent
)
3680 rdev
= md_import_device(dev
, mddev
->major_version
,
3681 mddev
->minor_version
);
3683 rdev
= md_import_device(dev
, -1, -1);
3686 "md: md_import_device returned %ld\n",
3688 return PTR_ERR(rdev
);
3690 /* set save_raid_disk if appropriate */
3691 if (!mddev
->persistent
) {
3692 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3693 info
->raid_disk
< mddev
->raid_disks
)
3694 rdev
->raid_disk
= info
->raid_disk
;
3696 rdev
->raid_disk
= -1;
3698 super_types
[mddev
->major_version
].
3699 validate_super(mddev
, rdev
);
3700 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3702 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3703 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3704 set_bit(WriteMostly
, &rdev
->flags
);
3706 rdev
->raid_disk
= -1;
3707 err
= bind_rdev_to_array(rdev
, mddev
);
3708 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3709 /* If there is hot_add_disk but no hot_remove_disk
3710 * then added disks for geometry changes,
3711 * and should be added immediately.
3713 super_types
[mddev
->major_version
].
3714 validate_super(mddev
, rdev
);
3715 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3717 unbind_rdev_from_array(rdev
);
3722 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3723 md_wakeup_thread(mddev
->thread
);
3727 /* otherwise, add_new_disk is only allowed
3728 * for major_version==0 superblocks
3730 if (mddev
->major_version
!= 0) {
3731 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3736 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3738 rdev
= md_import_device (dev
, -1, 0);
3741 "md: error, md_import_device() returned %ld\n",
3743 return PTR_ERR(rdev
);
3745 rdev
->desc_nr
= info
->number
;
3746 if (info
->raid_disk
< mddev
->raid_disks
)
3747 rdev
->raid_disk
= info
->raid_disk
;
3749 rdev
->raid_disk
= -1;
3753 if (rdev
->raid_disk
< mddev
->raid_disks
)
3754 if (info
->state
& (1<<MD_DISK_SYNC
))
3755 set_bit(In_sync
, &rdev
->flags
);
3757 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3758 set_bit(WriteMostly
, &rdev
->flags
);
3760 if (!mddev
->persistent
) {
3761 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3762 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3764 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3765 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3767 err
= bind_rdev_to_array(rdev
, mddev
);
3777 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3779 char b
[BDEVNAME_SIZE
];
3785 rdev
= find_rdev(mddev
, dev
);
3789 if (rdev
->raid_disk
>= 0)
3792 kick_rdev_from_array(rdev
);
3793 md_update_sb(mddev
, 1);
3794 md_new_event(mddev
);
3798 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3799 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3803 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3805 char b
[BDEVNAME_SIZE
];
3813 if (mddev
->major_version
!= 0) {
3814 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3815 " version-0 superblocks.\n",
3819 if (!mddev
->pers
->hot_add_disk
) {
3821 "%s: personality does not support diskops!\n",
3826 rdev
= md_import_device (dev
, -1, 0);
3829 "md: error, md_import_device() returned %ld\n",
3834 if (mddev
->persistent
)
3835 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3838 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3840 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3843 if (test_bit(Faulty
, &rdev
->flags
)) {
3845 "md: can not hot-add faulty %s disk to %s!\n",
3846 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3850 clear_bit(In_sync
, &rdev
->flags
);
3852 rdev
->saved_raid_disk
= -1;
3853 err
= bind_rdev_to_array(rdev
, mddev
);
3858 * The rest should better be atomic, we can have disk failures
3859 * noticed in interrupt contexts ...
3862 if (rdev
->desc_nr
== mddev
->max_disks
) {
3863 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3866 goto abort_unbind_export
;
3869 rdev
->raid_disk
= -1;
3871 md_update_sb(mddev
, 1);
3874 * Kick recovery, maybe this spare has to be added to the
3875 * array immediately.
3877 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3878 md_wakeup_thread(mddev
->thread
);
3879 md_new_event(mddev
);
3882 abort_unbind_export
:
3883 unbind_rdev_from_array(rdev
);
3890 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3895 if (!mddev
->pers
->quiesce
)
3897 if (mddev
->recovery
|| mddev
->sync_thread
)
3899 /* we should be able to change the bitmap.. */
3905 return -EEXIST
; /* cannot add when bitmap is present */
3906 mddev
->bitmap_file
= fget(fd
);
3908 if (mddev
->bitmap_file
== NULL
) {
3909 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3914 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3916 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3918 fput(mddev
->bitmap_file
);
3919 mddev
->bitmap_file
= NULL
;
3922 mddev
->bitmap_offset
= 0; /* file overrides offset */
3923 } else if (mddev
->bitmap
== NULL
)
3924 return -ENOENT
; /* cannot remove what isn't there */
3927 mddev
->pers
->quiesce(mddev
, 1);
3929 err
= bitmap_create(mddev
);
3930 if (fd
< 0 || err
) {
3931 bitmap_destroy(mddev
);
3932 fd
= -1; /* make sure to put the file */
3934 mddev
->pers
->quiesce(mddev
, 0);
3937 if (mddev
->bitmap_file
) {
3938 restore_bitmap_write_access(mddev
->bitmap_file
);
3939 fput(mddev
->bitmap_file
);
3941 mddev
->bitmap_file
= NULL
;
3948 * set_array_info is used two different ways
3949 * The original usage is when creating a new array.
3950 * In this usage, raid_disks is > 0 and it together with
3951 * level, size, not_persistent,layout,chunksize determine the
3952 * shape of the array.
3953 * This will always create an array with a type-0.90.0 superblock.
3954 * The newer usage is when assembling an array.
3955 * In this case raid_disks will be 0, and the major_version field is
3956 * use to determine which style super-blocks are to be found on the devices.
3957 * The minor and patch _version numbers are also kept incase the
3958 * super_block handler wishes to interpret them.
3960 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3963 if (info
->raid_disks
== 0) {
3964 /* just setting version number for superblock loading */
3965 if (info
->major_version
< 0 ||
3966 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3967 super_types
[info
->major_version
].name
== NULL
) {
3968 /* maybe try to auto-load a module? */
3970 "md: superblock version %d not known\n",
3971 info
->major_version
);
3974 mddev
->major_version
= info
->major_version
;
3975 mddev
->minor_version
= info
->minor_version
;
3976 mddev
->patch_version
= info
->patch_version
;
3979 mddev
->major_version
= MD_MAJOR_VERSION
;
3980 mddev
->minor_version
= MD_MINOR_VERSION
;
3981 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3982 mddev
->ctime
= get_seconds();
3984 mddev
->level
= info
->level
;
3985 mddev
->clevel
[0] = 0;
3986 mddev
->size
= info
->size
;
3987 mddev
->raid_disks
= info
->raid_disks
;
3988 /* don't set md_minor, it is determined by which /dev/md* was
3991 if (info
->state
& (1<<MD_SB_CLEAN
))
3992 mddev
->recovery_cp
= MaxSector
;
3994 mddev
->recovery_cp
= 0;
3995 mddev
->persistent
= ! info
->not_persistent
;
3997 mddev
->layout
= info
->layout
;
3998 mddev
->chunk_size
= info
->chunk_size
;
4000 mddev
->max_disks
= MD_SB_DISKS
;
4003 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4005 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4006 mddev
->bitmap_offset
= 0;
4008 mddev
->reshape_position
= MaxSector
;
4011 * Generate a 128 bit UUID
4013 get_random_bytes(mddev
->uuid
, 16);
4015 mddev
->new_level
= mddev
->level
;
4016 mddev
->new_chunk
= mddev
->chunk_size
;
4017 mddev
->new_layout
= mddev
->layout
;
4018 mddev
->delta_disks
= 0;
4023 static int update_size(mddev_t
*mddev
, unsigned long size
)
4027 struct list_head
*tmp
;
4028 int fit
= (size
== 0);
4030 if (mddev
->pers
->resize
== NULL
)
4032 /* The "size" is the amount of each device that is used.
4033 * This can only make sense for arrays with redundancy.
4034 * linear and raid0 always use whatever space is available
4035 * We can only consider changing the size if no resync
4036 * or reconstruction is happening, and if the new size
4037 * is acceptable. It must fit before the sb_offset or,
4038 * if that is <data_offset, it must fit before the
4039 * size of each device.
4040 * If size is zero, we find the largest size that fits.
4042 if (mddev
->sync_thread
)
4044 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4046 if (rdev
->sb_offset
> rdev
->data_offset
)
4047 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
4049 avail
= get_capacity(rdev
->bdev
->bd_disk
)
4050 - rdev
->data_offset
;
4051 if (fit
&& (size
== 0 || size
> avail
/2))
4053 if (avail
< ((sector_t
)size
<< 1))
4056 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
4058 struct block_device
*bdev
;
4060 bdev
= bdget_disk(mddev
->gendisk
, 0);
4062 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4063 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
4064 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4071 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4074 /* change the number of raid disks */
4075 if (mddev
->pers
->check_reshape
== NULL
)
4077 if (raid_disks
<= 0 ||
4078 raid_disks
>= mddev
->max_disks
)
4080 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4082 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4084 rv
= mddev
->pers
->check_reshape(mddev
);
4090 * update_array_info is used to change the configuration of an
4092 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4093 * fields in the info are checked against the array.
4094 * Any differences that cannot be handled will cause an error.
4095 * Normally, only one change can be managed at a time.
4097 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4103 /* calculate expected state,ignoring low bits */
4104 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4105 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4107 if (mddev
->major_version
!= info
->major_version
||
4108 mddev
->minor_version
!= info
->minor_version
||
4109 /* mddev->patch_version != info->patch_version || */
4110 mddev
->ctime
!= info
->ctime
||
4111 mddev
->level
!= info
->level
||
4112 /* mddev->layout != info->layout || */
4113 !mddev
->persistent
!= info
->not_persistent
||
4114 mddev
->chunk_size
!= info
->chunk_size
||
4115 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4116 ((state
^info
->state
) & 0xfffffe00)
4119 /* Check there is only one change */
4120 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4121 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4122 if (mddev
->layout
!= info
->layout
) cnt
++;
4123 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4124 if (cnt
== 0) return 0;
4125 if (cnt
> 1) return -EINVAL
;
4127 if (mddev
->layout
!= info
->layout
) {
4129 * we don't need to do anything at the md level, the
4130 * personality will take care of it all.
4132 if (mddev
->pers
->reconfig
== NULL
)
4135 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4137 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4138 rv
= update_size(mddev
, info
->size
);
4140 if (mddev
->raid_disks
!= info
->raid_disks
)
4141 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4143 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4144 if (mddev
->pers
->quiesce
== NULL
)
4146 if (mddev
->recovery
|| mddev
->sync_thread
)
4148 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4149 /* add the bitmap */
4152 if (mddev
->default_bitmap_offset
== 0)
4154 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4155 mddev
->pers
->quiesce(mddev
, 1);
4156 rv
= bitmap_create(mddev
);
4158 bitmap_destroy(mddev
);
4159 mddev
->pers
->quiesce(mddev
, 0);
4161 /* remove the bitmap */
4164 if (mddev
->bitmap
->file
)
4166 mddev
->pers
->quiesce(mddev
, 1);
4167 bitmap_destroy(mddev
);
4168 mddev
->pers
->quiesce(mddev
, 0);
4169 mddev
->bitmap_offset
= 0;
4172 md_update_sb(mddev
, 1);
4176 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4180 if (mddev
->pers
== NULL
)
4183 rdev
= find_rdev(mddev
, dev
);
4187 md_error(mddev
, rdev
);
4191 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4193 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4197 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4201 static int md_ioctl(struct inode
*inode
, struct file
*file
,
4202 unsigned int cmd
, unsigned long arg
)
4205 void __user
*argp
= (void __user
*)arg
;
4206 mddev_t
*mddev
= NULL
;
4208 if (!capable(CAP_SYS_ADMIN
))
4212 * Commands dealing with the RAID driver but not any
4218 err
= get_version(argp
);
4221 case PRINT_RAID_DEBUG
:
4229 autostart_arrays(arg
);
4236 * Commands creating/starting a new array:
4239 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4246 err
= mddev_lock(mddev
);
4249 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4256 case SET_ARRAY_INFO
:
4258 mdu_array_info_t info
;
4260 memset(&info
, 0, sizeof(info
));
4261 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4266 err
= update_array_info(mddev
, &info
);
4268 printk(KERN_WARNING
"md: couldn't update"
4269 " array info. %d\n", err
);
4274 if (!list_empty(&mddev
->disks
)) {
4276 "md: array %s already has disks!\n",
4281 if (mddev
->raid_disks
) {
4283 "md: array %s already initialised!\n",
4288 err
= set_array_info(mddev
, &info
);
4290 printk(KERN_WARNING
"md: couldn't set"
4291 " array info. %d\n", err
);
4301 * Commands querying/configuring an existing array:
4303 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4304 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4305 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4306 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
4312 * Commands even a read-only array can execute:
4316 case GET_ARRAY_INFO
:
4317 err
= get_array_info(mddev
, argp
);
4320 case GET_BITMAP_FILE
:
4321 err
= get_bitmap_file(mddev
, argp
);
4325 err
= get_disk_info(mddev
, argp
);
4328 case RESTART_ARRAY_RW
:
4329 err
= restart_array(mddev
);
4333 err
= do_md_stop (mddev
, 0);
4337 err
= do_md_stop (mddev
, 1);
4341 * We have a problem here : there is no easy way to give a CHS
4342 * virtual geometry. We currently pretend that we have a 2 heads
4343 * 4 sectors (with a BIG number of cylinders...). This drives
4344 * dosfs just mad... ;-)
4349 * The remaining ioctls are changing the state of the
4350 * superblock, so we do not allow them on read-only arrays.
4351 * However non-MD ioctls (e.g. get-size) will still come through
4352 * here and hit the 'default' below, so only disallow
4353 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4355 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4356 mddev
->ro
&& mddev
->pers
) {
4357 if (mddev
->ro
== 2) {
4359 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4360 md_wakeup_thread(mddev
->thread
);
4372 mdu_disk_info_t info
;
4373 if (copy_from_user(&info
, argp
, sizeof(info
)))
4376 err
= add_new_disk(mddev
, &info
);
4380 case HOT_REMOVE_DISK
:
4381 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4385 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4388 case SET_DISK_FAULTY
:
4389 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4393 err
= do_md_run (mddev
);
4396 case SET_BITMAP_FILE
:
4397 err
= set_bitmap_file(mddev
, (int)arg
);
4407 mddev_unlock(mddev
);
4417 static int md_open(struct inode
*inode
, struct file
*file
)
4420 * Succeed if we can lock the mddev, which confirms that
4421 * it isn't being stopped right now.
4423 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4426 if ((err
= mddev_lock(mddev
)))
4431 mddev_unlock(mddev
);
4433 check_disk_change(inode
->i_bdev
);
4438 static int md_release(struct inode
*inode
, struct file
* file
)
4440 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4448 static int md_media_changed(struct gendisk
*disk
)
4450 mddev_t
*mddev
= disk
->private_data
;
4452 return mddev
->changed
;
4455 static int md_revalidate(struct gendisk
*disk
)
4457 mddev_t
*mddev
= disk
->private_data
;
4462 static struct block_device_operations md_fops
=
4464 .owner
= THIS_MODULE
,
4466 .release
= md_release
,
4468 .getgeo
= md_getgeo
,
4469 .media_changed
= md_media_changed
,
4470 .revalidate_disk
= md_revalidate
,
4473 static int md_thread(void * arg
)
4475 mdk_thread_t
*thread
= arg
;
4478 * md_thread is a 'system-thread', it's priority should be very
4479 * high. We avoid resource deadlocks individually in each
4480 * raid personality. (RAID5 does preallocation) We also use RR and
4481 * the very same RT priority as kswapd, thus we will never get
4482 * into a priority inversion deadlock.
4484 * we definitely have to have equal or higher priority than
4485 * bdflush, otherwise bdflush will deadlock if there are too
4486 * many dirty RAID5 blocks.
4489 allow_signal(SIGKILL
);
4490 while (!kthread_should_stop()) {
4492 /* We need to wait INTERRUPTIBLE so that
4493 * we don't add to the load-average.
4494 * That means we need to be sure no signals are
4497 if (signal_pending(current
))
4498 flush_signals(current
);
4500 wait_event_interruptible_timeout
4502 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4503 || kthread_should_stop(),
4507 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4509 thread
->run(thread
->mddev
);
4515 void md_wakeup_thread(mdk_thread_t
*thread
)
4518 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4519 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4520 wake_up(&thread
->wqueue
);
4524 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4527 mdk_thread_t
*thread
;
4529 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4533 init_waitqueue_head(&thread
->wqueue
);
4536 thread
->mddev
= mddev
;
4537 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4538 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4539 if (IS_ERR(thread
->tsk
)) {
4546 void md_unregister_thread(mdk_thread_t
*thread
)
4548 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4550 kthread_stop(thread
->tsk
);
4554 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4561 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4564 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4566 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4567 __builtin_return_address(0),__builtin_return_address(1),
4568 __builtin_return_address(2),__builtin_return_address(3));
4572 if (!mddev
->pers
->error_handler
)
4574 mddev
->pers
->error_handler(mddev
,rdev
);
4575 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4576 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4577 md_wakeup_thread(mddev
->thread
);
4578 md_new_event_inintr(mddev
);
4581 /* seq_file implementation /proc/mdstat */
4583 static void status_unused(struct seq_file
*seq
)
4587 struct list_head
*tmp
;
4589 seq_printf(seq
, "unused devices: ");
4591 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4592 char b
[BDEVNAME_SIZE
];
4594 seq_printf(seq
, "%s ",
4595 bdevname(rdev
->bdev
,b
));
4598 seq_printf(seq
, "<none>");
4600 seq_printf(seq
, "\n");
4604 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4606 sector_t max_blocks
, resync
, res
;
4607 unsigned long dt
, db
, rt
;
4609 unsigned int per_milli
;
4611 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4613 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4614 max_blocks
= mddev
->resync_max_sectors
>> 1;
4616 max_blocks
= mddev
->size
;
4619 * Should not happen.
4625 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4626 * in a sector_t, and (max_blocks>>scale) will fit in a
4627 * u32, as those are the requirements for sector_div.
4628 * Thus 'scale' must be at least 10
4631 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4632 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4635 res
= (resync
>>scale
)*1000;
4636 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4640 int i
, x
= per_milli
/50, y
= 20-x
;
4641 seq_printf(seq
, "[");
4642 for (i
= 0; i
< x
; i
++)
4643 seq_printf(seq
, "=");
4644 seq_printf(seq
, ">");
4645 for (i
= 0; i
< y
; i
++)
4646 seq_printf(seq
, ".");
4647 seq_printf(seq
, "] ");
4649 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4650 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4652 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
4654 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4655 "resync" : "recovery"))),
4656 per_milli
/10, per_milli
% 10,
4657 (unsigned long long) resync
,
4658 (unsigned long long) max_blocks
);
4661 * We do not want to overflow, so the order of operands and
4662 * the * 100 / 100 trick are important. We do a +1 to be
4663 * safe against division by zero. We only estimate anyway.
4665 * dt: time from mark until now
4666 * db: blocks written from mark until now
4667 * rt: remaining time
4669 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4671 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
4672 - mddev
->resync_mark_cnt
;
4673 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/2/100+1)))/100;
4675 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4677 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
4680 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4682 struct list_head
*tmp
;
4692 spin_lock(&all_mddevs_lock
);
4693 list_for_each(tmp
,&all_mddevs
)
4695 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4697 spin_unlock(&all_mddevs_lock
);
4700 spin_unlock(&all_mddevs_lock
);
4702 return (void*)2;/* tail */
4706 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4708 struct list_head
*tmp
;
4709 mddev_t
*next_mddev
, *mddev
= v
;
4715 spin_lock(&all_mddevs_lock
);
4717 tmp
= all_mddevs
.next
;
4719 tmp
= mddev
->all_mddevs
.next
;
4720 if (tmp
!= &all_mddevs
)
4721 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4723 next_mddev
= (void*)2;
4726 spin_unlock(&all_mddevs_lock
);
4734 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4738 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4742 struct mdstat_info
{
4746 static int md_seq_show(struct seq_file
*seq
, void *v
)
4750 struct list_head
*tmp2
;
4752 struct mdstat_info
*mi
= seq
->private;
4753 struct bitmap
*bitmap
;
4755 if (v
== (void*)1) {
4756 struct mdk_personality
*pers
;
4757 seq_printf(seq
, "Personalities : ");
4758 spin_lock(&pers_lock
);
4759 list_for_each_entry(pers
, &pers_list
, list
)
4760 seq_printf(seq
, "[%s] ", pers
->name
);
4762 spin_unlock(&pers_lock
);
4763 seq_printf(seq
, "\n");
4764 mi
->event
= atomic_read(&md_event_count
);
4767 if (v
== (void*)2) {
4772 if (mddev_lock(mddev
) < 0)
4775 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4776 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4777 mddev
->pers
? "" : "in");
4780 seq_printf(seq
, " (read-only)");
4782 seq_printf(seq
, "(auto-read-only)");
4783 seq_printf(seq
, " %s", mddev
->pers
->name
);
4787 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4788 char b
[BDEVNAME_SIZE
];
4789 seq_printf(seq
, " %s[%d]",
4790 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4791 if (test_bit(WriteMostly
, &rdev
->flags
))
4792 seq_printf(seq
, "(W)");
4793 if (test_bit(Faulty
, &rdev
->flags
)) {
4794 seq_printf(seq
, "(F)");
4796 } else if (rdev
->raid_disk
< 0)
4797 seq_printf(seq
, "(S)"); /* spare */
4801 if (!list_empty(&mddev
->disks
)) {
4803 seq_printf(seq
, "\n %llu blocks",
4804 (unsigned long long)mddev
->array_size
);
4806 seq_printf(seq
, "\n %llu blocks",
4807 (unsigned long long)size
);
4809 if (mddev
->persistent
) {
4810 if (mddev
->major_version
!= 0 ||
4811 mddev
->minor_version
!= 90) {
4812 seq_printf(seq
," super %d.%d",
4813 mddev
->major_version
,
4814 mddev
->minor_version
);
4817 seq_printf(seq
, " super non-persistent");
4820 mddev
->pers
->status (seq
, mddev
);
4821 seq_printf(seq
, "\n ");
4822 if (mddev
->pers
->sync_request
) {
4823 if (mddev
->curr_resync
> 2) {
4824 status_resync (seq
, mddev
);
4825 seq_printf(seq
, "\n ");
4826 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4827 seq_printf(seq
, "\tresync=DELAYED\n ");
4828 else if (mddev
->recovery_cp
< MaxSector
)
4829 seq_printf(seq
, "\tresync=PENDING\n ");
4832 seq_printf(seq
, "\n ");
4834 if ((bitmap
= mddev
->bitmap
)) {
4835 unsigned long chunk_kb
;
4836 unsigned long flags
;
4837 spin_lock_irqsave(&bitmap
->lock
, flags
);
4838 chunk_kb
= bitmap
->chunksize
>> 10;
4839 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4841 bitmap
->pages
- bitmap
->missing_pages
,
4843 (bitmap
->pages
- bitmap
->missing_pages
)
4844 << (PAGE_SHIFT
- 10),
4845 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4846 chunk_kb
? "KB" : "B");
4848 seq_printf(seq
, ", file: ");
4849 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4850 bitmap
->file
->f_dentry
," \t\n");
4853 seq_printf(seq
, "\n");
4854 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4857 seq_printf(seq
, "\n");
4859 mddev_unlock(mddev
);
4864 static struct seq_operations md_seq_ops
= {
4865 .start
= md_seq_start
,
4866 .next
= md_seq_next
,
4867 .stop
= md_seq_stop
,
4868 .show
= md_seq_show
,
4871 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4874 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4878 error
= seq_open(file
, &md_seq_ops
);
4882 struct seq_file
*p
= file
->private_data
;
4884 mi
->event
= atomic_read(&md_event_count
);
4889 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4891 struct seq_file
*m
= file
->private_data
;
4892 struct mdstat_info
*mi
= m
->private;
4895 return seq_release(inode
, file
);
4898 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4900 struct seq_file
*m
= filp
->private_data
;
4901 struct mdstat_info
*mi
= m
->private;
4904 poll_wait(filp
, &md_event_waiters
, wait
);
4906 /* always allow read */
4907 mask
= POLLIN
| POLLRDNORM
;
4909 if (mi
->event
!= atomic_read(&md_event_count
))
4910 mask
|= POLLERR
| POLLPRI
;
4914 static struct file_operations md_seq_fops
= {
4915 .owner
= THIS_MODULE
,
4916 .open
= md_seq_open
,
4918 .llseek
= seq_lseek
,
4919 .release
= md_seq_release
,
4920 .poll
= mdstat_poll
,
4923 int register_md_personality(struct mdk_personality
*p
)
4925 spin_lock(&pers_lock
);
4926 list_add_tail(&p
->list
, &pers_list
);
4927 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4928 spin_unlock(&pers_lock
);
4932 int unregister_md_personality(struct mdk_personality
*p
)
4934 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4935 spin_lock(&pers_lock
);
4936 list_del_init(&p
->list
);
4937 spin_unlock(&pers_lock
);
4941 static int is_mddev_idle(mddev_t
*mddev
)
4944 struct list_head
*tmp
;
4946 unsigned long curr_events
;
4949 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4950 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4951 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4952 disk_stat_read(disk
, sectors
[1]) -
4953 atomic_read(&disk
->sync_io
);
4954 /* The difference between curr_events and last_events
4955 * will be affected by any new non-sync IO (making
4956 * curr_events bigger) and any difference in the amount of
4957 * in-flight syncio (making current_events bigger or smaller)
4958 * The amount in-flight is currently limited to
4959 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4960 * which is at most 4096 sectors.
4961 * These numbers are fairly fragile and should be made
4962 * more robust, probably by enforcing the
4963 * 'window size' that md_do_sync sort-of uses.
4965 * Note: the following is an unsigned comparison.
4967 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4968 rdev
->last_events
= curr_events
;
4975 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4977 /* another "blocks" (512byte) blocks have been synced */
4978 atomic_sub(blocks
, &mddev
->recovery_active
);
4979 wake_up(&mddev
->recovery_wait
);
4981 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4982 md_wakeup_thread(mddev
->thread
);
4983 // stop recovery, signal do_sync ....
4988 /* md_write_start(mddev, bi)
4989 * If we need to update some array metadata (e.g. 'active' flag
4990 * in superblock) before writing, schedule a superblock update
4991 * and wait for it to complete.
4993 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4995 if (bio_data_dir(bi
) != WRITE
)
4998 BUG_ON(mddev
->ro
== 1);
4999 if (mddev
->ro
== 2) {
5000 /* need to switch to read/write */
5002 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5003 md_wakeup_thread(mddev
->thread
);
5005 atomic_inc(&mddev
->writes_pending
);
5006 if (mddev
->in_sync
) {
5007 spin_lock_irq(&mddev
->write_lock
);
5008 if (mddev
->in_sync
) {
5010 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5011 md_wakeup_thread(mddev
->thread
);
5013 spin_unlock_irq(&mddev
->write_lock
);
5015 wait_event(mddev
->sb_wait
, mddev
->flags
==0);
5018 void md_write_end(mddev_t
*mddev
)
5020 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5021 if (mddev
->safemode
== 2)
5022 md_wakeup_thread(mddev
->thread
);
5023 else if (mddev
->safemode_delay
)
5024 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5028 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
5030 #define SYNC_MARKS 10
5031 #define SYNC_MARK_STEP (3*HZ)
5032 void md_do_sync(mddev_t
*mddev
)
5035 unsigned int currspeed
= 0,
5037 sector_t max_sectors
,j
, io_sectors
;
5038 unsigned long mark
[SYNC_MARKS
];
5039 sector_t mark_cnt
[SYNC_MARKS
];
5041 struct list_head
*tmp
;
5042 sector_t last_check
;
5044 struct list_head
*rtmp
;
5048 /* just incase thread restarts... */
5049 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5051 if (mddev
->ro
) /* never try to sync a read-only array */
5054 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5055 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
5056 desc
= "data-check";
5057 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5058 desc
= "requested-resync";
5061 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5066 /* we overload curr_resync somewhat here.
5067 * 0 == not engaged in resync at all
5068 * 2 == checking that there is no conflict with another sync
5069 * 1 == like 2, but have yielded to allow conflicting resync to
5071 * other == active in resync - this many blocks
5073 * Before starting a resync we must have set curr_resync to
5074 * 2, and then checked that every "conflicting" array has curr_resync
5075 * less than ours. When we find one that is the same or higher
5076 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5077 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5078 * This will mean we have to start checking from the beginning again.
5083 mddev
->curr_resync
= 2;
5086 if (kthread_should_stop()) {
5087 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5090 ITERATE_MDDEV(mddev2
,tmp
) {
5091 if (mddev2
== mddev
)
5093 if (mddev2
->curr_resync
&&
5094 match_mddev_units(mddev
,mddev2
)) {
5096 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5097 /* arbitrarily yield */
5098 mddev
->curr_resync
= 1;
5099 wake_up(&resync_wait
);
5101 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5102 /* no need to wait here, we can wait the next
5103 * time 'round when curr_resync == 2
5106 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
5107 if (!kthread_should_stop() &&
5108 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5109 printk(KERN_INFO
"md: delaying %s of %s"
5110 " until %s has finished (they"
5111 " share one or more physical units)\n",
5112 desc
, mdname(mddev
), mdname(mddev2
));
5115 finish_wait(&resync_wait
, &wq
);
5118 finish_wait(&resync_wait
, &wq
);
5121 } while (mddev
->curr_resync
< 2);
5124 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5125 /* resync follows the size requested by the personality,
5126 * which defaults to physical size, but can be virtual size
5128 max_sectors
= mddev
->resync_max_sectors
;
5129 mddev
->resync_mismatches
= 0;
5130 /* we don't use the checkpoint if there's a bitmap */
5131 if (!mddev
->bitmap
&&
5132 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5133 j
= mddev
->recovery_cp
;
5134 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5135 max_sectors
= mddev
->size
<< 1;
5137 /* recovery follows the physical size of devices */
5138 max_sectors
= mddev
->size
<< 1;
5140 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5141 if (rdev
->raid_disk
>= 0 &&
5142 !test_bit(Faulty
, &rdev
->flags
) &&
5143 !test_bit(In_sync
, &rdev
->flags
) &&
5144 rdev
->recovery_offset
< j
)
5145 j
= rdev
->recovery_offset
;
5148 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
5149 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
5150 " %d KB/sec/disk.\n", speed_min(mddev
));
5151 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5152 "(but not more than %d KB/sec) for %s.\n",
5153 speed_max(mddev
), desc
);
5155 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5158 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5160 mark_cnt
[m
] = io_sectors
;
5163 mddev
->resync_mark
= mark
[last_mark
];
5164 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5167 * Tune reconstruction:
5169 window
= 32*(PAGE_SIZE
/512);
5170 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5171 window
/2,(unsigned long long) max_sectors
/2);
5173 atomic_set(&mddev
->recovery_active
, 0);
5174 init_waitqueue_head(&mddev
->recovery_wait
);
5179 "md: resuming %s of %s from checkpoint.\n",
5180 desc
, mdname(mddev
));
5181 mddev
->curr_resync
= j
;
5184 while (j
< max_sectors
) {
5188 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5189 currspeed
< speed_min(mddev
));
5191 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5195 if (!skipped
) { /* actual IO requested */
5196 io_sectors
+= sectors
;
5197 atomic_add(sectors
, &mddev
->recovery_active
);
5201 if (j
>1) mddev
->curr_resync
= j
;
5202 mddev
->curr_mark_cnt
= io_sectors
;
5203 if (last_check
== 0)
5204 /* this is the earliers that rebuilt will be
5205 * visible in /proc/mdstat
5207 md_new_event(mddev
);
5209 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5212 last_check
= io_sectors
;
5214 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
5215 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
5219 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5221 int next
= (last_mark
+1) % SYNC_MARKS
;
5223 mddev
->resync_mark
= mark
[next
];
5224 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5225 mark
[next
] = jiffies
;
5226 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5231 if (kthread_should_stop()) {
5233 * got a signal, exit.
5236 "md: md_do_sync() got signal ... exiting\n");
5237 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5242 * this loop exits only if either when we are slower than
5243 * the 'hard' speed limit, or the system was IO-idle for
5245 * the system might be non-idle CPU-wise, but we only care
5246 * about not overloading the IO subsystem. (things like an
5247 * e2fsck being done on the RAID array should execute fast)
5249 mddev
->queue
->unplug_fn(mddev
->queue
);
5252 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5253 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5255 if (currspeed
> speed_min(mddev
)) {
5256 if ((currspeed
> speed_max(mddev
)) ||
5257 !is_mddev_idle(mddev
)) {
5263 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
5265 * this also signals 'finished resyncing' to md_stop
5268 mddev
->queue
->unplug_fn(mddev
->queue
);
5270 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5272 /* tell personality that we are finished */
5273 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5275 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5276 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
5277 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5278 mddev
->curr_resync
> 2) {
5279 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5280 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5281 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5283 "md: checkpointing %s of %s.\n",
5284 desc
, mdname(mddev
));
5285 mddev
->recovery_cp
= mddev
->curr_resync
;
5288 mddev
->recovery_cp
= MaxSector
;
5290 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5291 mddev
->curr_resync
= MaxSector
;
5292 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5293 if (rdev
->raid_disk
>= 0 &&
5294 !test_bit(Faulty
, &rdev
->flags
) &&
5295 !test_bit(In_sync
, &rdev
->flags
) &&
5296 rdev
->recovery_offset
< mddev
->curr_resync
)
5297 rdev
->recovery_offset
= mddev
->curr_resync
;
5302 mddev
->curr_resync
= 0;
5303 wake_up(&resync_wait
);
5304 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5305 md_wakeup_thread(mddev
->thread
);
5307 EXPORT_SYMBOL_GPL(md_do_sync
);
5311 * This routine is regularly called by all per-raid-array threads to
5312 * deal with generic issues like resync and super-block update.
5313 * Raid personalities that don't have a thread (linear/raid0) do not
5314 * need this as they never do any recovery or update the superblock.
5316 * It does not do any resync itself, but rather "forks" off other threads
5317 * to do that as needed.
5318 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5319 * "->recovery" and create a thread at ->sync_thread.
5320 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5321 * and wakeups up this thread which will reap the thread and finish up.
5322 * This thread also removes any faulty devices (with nr_pending == 0).
5324 * The overall approach is:
5325 * 1/ if the superblock needs updating, update it.
5326 * 2/ If a recovery thread is running, don't do anything else.
5327 * 3/ If recovery has finished, clean up, possibly marking spares active.
5328 * 4/ If there are any faulty devices, remove them.
5329 * 5/ If array is degraded, try to add spares devices
5330 * 6/ If array has spares or is not in-sync, start a resync thread.
5332 void md_check_recovery(mddev_t
*mddev
)
5335 struct list_head
*rtmp
;
5339 bitmap_daemon_work(mddev
->bitmap
);
5344 if (signal_pending(current
)) {
5345 if (mddev
->pers
->sync_request
) {
5346 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5348 mddev
->safemode
= 2;
5350 flush_signals(current
);
5355 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5356 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5357 (mddev
->safemode
== 1) ||
5358 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5359 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5363 if (mddev_trylock(mddev
)) {
5366 spin_lock_irq(&mddev
->write_lock
);
5367 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5368 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5370 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5372 if (mddev
->safemode
== 1)
5373 mddev
->safemode
= 0;
5374 spin_unlock_irq(&mddev
->write_lock
);
5377 md_update_sb(mddev
, 0);
5380 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5381 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5382 /* resync/recovery still happening */
5383 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5386 if (mddev
->sync_thread
) {
5387 /* resync has finished, collect result */
5388 md_unregister_thread(mddev
->sync_thread
);
5389 mddev
->sync_thread
= NULL
;
5390 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5391 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5393 /* activate any spares */
5394 mddev
->pers
->spare_active(mddev
);
5396 md_update_sb(mddev
, 1);
5398 /* if array is no-longer degraded, then any saved_raid_disk
5399 * information must be scrapped
5401 if (!mddev
->degraded
)
5402 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5403 rdev
->saved_raid_disk
= -1;
5405 mddev
->recovery
= 0;
5406 /* flag recovery needed just to double check */
5407 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5408 md_new_event(mddev
);
5411 /* Clear some bits that don't mean anything, but
5414 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5415 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5416 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5417 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5419 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5421 /* no recovery is running.
5422 * remove any failed drives, then
5423 * add spares if possible.
5424 * Spare are also removed and re-added, to allow
5425 * the personality to fail the re-add.
5427 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5428 if (rdev
->raid_disk
>= 0 &&
5429 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
5430 atomic_read(&rdev
->nr_pending
)==0) {
5431 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
5433 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5434 sysfs_remove_link(&mddev
->kobj
, nm
);
5435 rdev
->raid_disk
= -1;
5439 if (mddev
->degraded
) {
5440 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5441 if (rdev
->raid_disk
< 0
5442 && !test_bit(Faulty
, &rdev
->flags
)) {
5443 rdev
->recovery_offset
= 0;
5444 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5446 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5447 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
5449 md_new_event(mddev
);
5456 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5457 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5458 } else if (mddev
->recovery_cp
< MaxSector
) {
5459 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5460 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5461 /* nothing to be done ... */
5464 if (mddev
->pers
->sync_request
) {
5465 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5466 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5467 /* We are adding a device or devices to an array
5468 * which has the bitmap stored on all devices.
5469 * So make sure all bitmap pages get written
5471 bitmap_write_all(mddev
->bitmap
);
5473 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5476 if (!mddev
->sync_thread
) {
5477 printk(KERN_ERR
"%s: could not start resync"
5480 /* leave the spares where they are, it shouldn't hurt */
5481 mddev
->recovery
= 0;
5483 md_wakeup_thread(mddev
->sync_thread
);
5484 md_new_event(mddev
);
5487 mddev_unlock(mddev
);
5491 static int md_notify_reboot(struct notifier_block
*this,
5492 unsigned long code
, void *x
)
5494 struct list_head
*tmp
;
5497 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5499 printk(KERN_INFO
"md: stopping all md devices.\n");
5501 ITERATE_MDDEV(mddev
,tmp
)
5502 if (mddev_trylock(mddev
)) {
5503 do_md_stop (mddev
, 1);
5504 mddev_unlock(mddev
);
5507 * certain more exotic SCSI devices are known to be
5508 * volatile wrt too early system reboots. While the
5509 * right place to handle this issue is the given
5510 * driver, we do want to have a safe RAID driver ...
5517 static struct notifier_block md_notifier
= {
5518 .notifier_call
= md_notify_reboot
,
5520 .priority
= INT_MAX
, /* before any real devices */
5523 static void md_geninit(void)
5525 struct proc_dir_entry
*p
;
5527 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5529 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5531 p
->proc_fops
= &md_seq_fops
;
5534 static int __init
md_init(void)
5536 if (register_blkdev(MAJOR_NR
, "md"))
5538 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5539 unregister_blkdev(MAJOR_NR
, "md");
5542 blk_register_region(MKDEV(MAJOR_NR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
5543 md_probe
, NULL
, NULL
);
5544 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
5545 md_probe
, NULL
, NULL
);
5547 register_reboot_notifier(&md_notifier
);
5548 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
5558 * Searches all registered partitions for autorun RAID arrays
5561 static dev_t detected_devices
[128];
5564 void md_autodetect_dev(dev_t dev
)
5566 if (dev_cnt
>= 0 && dev_cnt
< 127)
5567 detected_devices
[dev_cnt
++] = dev
;
5571 static void autostart_arrays(int part
)
5576 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5578 for (i
= 0; i
< dev_cnt
; i
++) {
5579 dev_t dev
= detected_devices
[i
];
5581 rdev
= md_import_device(dev
,0, 0);
5585 if (test_bit(Faulty
, &rdev
->flags
)) {
5589 list_add(&rdev
->same_set
, &pending_raid_disks
);
5593 autorun_devices(part
);
5598 static __exit
void md_exit(void)
5601 struct list_head
*tmp
;
5603 blk_unregister_region(MKDEV(MAJOR_NR
,0), 1U << MINORBITS
);
5604 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
5606 unregister_blkdev(MAJOR_NR
,"md");
5607 unregister_blkdev(mdp_major
, "mdp");
5608 unregister_reboot_notifier(&md_notifier
);
5609 unregister_sysctl_table(raid_table_header
);
5610 remove_proc_entry("mdstat", NULL
);
5611 ITERATE_MDDEV(mddev
,tmp
) {
5612 struct gendisk
*disk
= mddev
->gendisk
;
5615 export_array(mddev
);
5618 mddev
->gendisk
= NULL
;
5623 module_init(md_init
)
5624 module_exit(md_exit
)
5626 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5628 return sprintf(buffer
, "%d", start_readonly
);
5630 static int set_ro(const char *val
, struct kernel_param
*kp
)
5633 int num
= simple_strtoul(val
, &e
, 10);
5634 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5635 start_readonly
= num
;
5641 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
5642 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
5645 EXPORT_SYMBOL(register_md_personality
);
5646 EXPORT_SYMBOL(unregister_md_personality
);
5647 EXPORT_SYMBOL(md_error
);
5648 EXPORT_SYMBOL(md_done_sync
);
5649 EXPORT_SYMBOL(md_write_start
);
5650 EXPORT_SYMBOL(md_write_end
);
5651 EXPORT_SYMBOL(md_register_thread
);
5652 EXPORT_SYMBOL(md_unregister_thread
);
5653 EXPORT_SYMBOL(md_wakeup_thread
);
5654 EXPORT_SYMBOL(md_check_recovery
);
5655 MODULE_LICENSE("GPL");
5657 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);