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/poll.h>
43 #include <linux/mutex.h>
44 #include <linux/ctype.h>
45 #include <linux/freezer.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 __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
981 unsigned long long newcsum
;
982 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
983 __le32
*isuper
= (__le32
*)sb
;
986 disk_csum
= sb
->sb_csum
;
989 for (i
=0; size
>=4; size
-= 4 )
990 newcsum
+= le32_to_cpu(*isuper
++);
993 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
995 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
996 sb
->sb_csum
= disk_csum
;
997 return cpu_to_le32(csum
);
1000 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1002 struct mdp_superblock_1
*sb
;
1005 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1009 * Calculate the position of the superblock.
1010 * It is always aligned to a 4K boundary and
1011 * depeding on minor_version, it can be:
1012 * 0: At least 8K, but less than 12K, from end of device
1013 * 1: At start of device
1014 * 2: 4K from start of device.
1016 switch(minor_version
) {
1018 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1020 sb_offset
&= ~(sector_t
)(4*2-1);
1021 /* convert from sectors to K */
1033 rdev
->sb_offset
= sb_offset
;
1035 /* superblock is rarely larger than 1K, but it can be larger,
1036 * and it is safe to read 4k, so we do that
1038 ret
= read_disk_sb(rdev
, 4096);
1039 if (ret
) return ret
;
1042 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1044 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1045 sb
->major_version
!= cpu_to_le32(1) ||
1046 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1047 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1048 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1051 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1052 printk("md: invalid superblock checksum on %s\n",
1053 bdevname(rdev
->bdev
,b
));
1056 if (le64_to_cpu(sb
->data_size
) < 10) {
1057 printk("md: data_size too small on %s\n",
1058 bdevname(rdev
->bdev
,b
));
1061 rdev
->preferred_minor
= 0xffff;
1062 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1063 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1065 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1066 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1067 if (rdev
->sb_size
& bmask
)
1068 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1070 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1073 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1079 struct mdp_superblock_1
*refsb
=
1080 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1082 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1083 sb
->level
!= refsb
->level
||
1084 sb
->layout
!= refsb
->layout
||
1085 sb
->chunksize
!= refsb
->chunksize
) {
1086 printk(KERN_WARNING
"md: %s has strangely different"
1087 " superblock to %s\n",
1088 bdevname(rdev
->bdev
,b
),
1089 bdevname(refdev
->bdev
,b2
));
1092 ev1
= le64_to_cpu(sb
->events
);
1093 ev2
= le64_to_cpu(refsb
->events
);
1101 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1103 rdev
->size
= rdev
->sb_offset
;
1104 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1106 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1107 if (le32_to_cpu(sb
->chunksize
))
1108 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1110 if (le64_to_cpu(sb
->size
) > rdev
->size
*2)
1115 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1117 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1118 __u64 ev1
= le64_to_cpu(sb
->events
);
1120 rdev
->raid_disk
= -1;
1122 if (mddev
->raid_disks
== 0) {
1123 mddev
->major_version
= 1;
1124 mddev
->patch_version
= 0;
1125 mddev
->persistent
= 1;
1126 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1127 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1128 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1129 mddev
->level
= le32_to_cpu(sb
->level
);
1130 mddev
->clevel
[0] = 0;
1131 mddev
->layout
= le32_to_cpu(sb
->layout
);
1132 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1133 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1134 mddev
->events
= ev1
;
1135 mddev
->bitmap_offset
= 0;
1136 mddev
->default_bitmap_offset
= 1024 >> 9;
1138 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1139 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1141 mddev
->max_disks
= (4096-256)/2;
1143 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1144 mddev
->bitmap_file
== NULL
) {
1145 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1146 && mddev
->level
!= 10) {
1147 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1150 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1152 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1153 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1154 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1155 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1156 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1157 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1159 mddev
->reshape_position
= MaxSector
;
1160 mddev
->delta_disks
= 0;
1161 mddev
->new_level
= mddev
->level
;
1162 mddev
->new_layout
= mddev
->layout
;
1163 mddev
->new_chunk
= mddev
->chunk_size
;
1166 } else if (mddev
->pers
== NULL
) {
1167 /* Insist of good event counter while assembling */
1169 if (ev1
< mddev
->events
)
1171 } else if (mddev
->bitmap
) {
1172 /* If adding to array with a bitmap, then we can accept an
1173 * older device, but not too old.
1175 if (ev1
< mddev
->bitmap
->events_cleared
)
1178 if (ev1
< mddev
->events
)
1179 /* just a hot-add of a new device, leave raid_disk at -1 */
1182 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1184 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1186 case 0xffff: /* spare */
1188 case 0xfffe: /* faulty */
1189 set_bit(Faulty
, &rdev
->flags
);
1192 if ((le32_to_cpu(sb
->feature_map
) &
1193 MD_FEATURE_RECOVERY_OFFSET
))
1194 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1196 set_bit(In_sync
, &rdev
->flags
);
1197 rdev
->raid_disk
= role
;
1200 if (sb
->devflags
& WriteMostly1
)
1201 set_bit(WriteMostly
, &rdev
->flags
);
1202 } else /* MULTIPATH are always insync */
1203 set_bit(In_sync
, &rdev
->flags
);
1208 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1210 struct mdp_superblock_1
*sb
;
1211 struct list_head
*tmp
;
1214 /* make rdev->sb match mddev and rdev data. */
1216 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1218 sb
->feature_map
= 0;
1220 sb
->recovery_offset
= cpu_to_le64(0);
1221 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1222 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1223 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1225 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1226 sb
->events
= cpu_to_le64(mddev
->events
);
1228 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1230 sb
->resync_offset
= cpu_to_le64(0);
1232 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1234 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1235 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1237 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1238 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1239 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1242 if (rdev
->raid_disk
>= 0 &&
1243 !test_bit(In_sync
, &rdev
->flags
) &&
1244 rdev
->recovery_offset
> 0) {
1245 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1246 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1249 if (mddev
->reshape_position
!= MaxSector
) {
1250 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1251 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1252 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1253 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1254 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1255 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1259 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1260 if (rdev2
->desc_nr
+1 > max_dev
)
1261 max_dev
= rdev2
->desc_nr
+1;
1263 sb
->max_dev
= cpu_to_le32(max_dev
);
1264 for (i
=0; i
<max_dev
;i
++)
1265 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1267 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1269 if (test_bit(Faulty
, &rdev2
->flags
))
1270 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1271 else if (test_bit(In_sync
, &rdev2
->flags
))
1272 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1273 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1274 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1276 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1279 sb
->sb_csum
= calc_sb_1_csum(sb
);
1283 static struct super_type super_types
[] = {
1286 .owner
= THIS_MODULE
,
1287 .load_super
= super_90_load
,
1288 .validate_super
= super_90_validate
,
1289 .sync_super
= super_90_sync
,
1293 .owner
= THIS_MODULE
,
1294 .load_super
= super_1_load
,
1295 .validate_super
= super_1_validate
,
1296 .sync_super
= super_1_sync
,
1300 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1302 struct list_head
*tmp
, *tmp2
;
1303 mdk_rdev_t
*rdev
, *rdev2
;
1305 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1306 ITERATE_RDEV(mddev2
, rdev2
, tmp2
)
1307 if (rdev
->bdev
->bd_contains
==
1308 rdev2
->bdev
->bd_contains
)
1314 static LIST_HEAD(pending_raid_disks
);
1316 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1318 char b
[BDEVNAME_SIZE
];
1327 /* make sure rdev->size exceeds mddev->size */
1328 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1330 /* Cannot change size, so fail */
1333 mddev
->size
= rdev
->size
;
1336 /* Verify rdev->desc_nr is unique.
1337 * If it is -1, assign a free number, else
1338 * check number is not in use
1340 if (rdev
->desc_nr
< 0) {
1342 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1343 while (find_rdev_nr(mddev
, choice
))
1345 rdev
->desc_nr
= choice
;
1347 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1350 bdevname(rdev
->bdev
,b
);
1351 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1353 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1356 rdev
->mddev
= mddev
;
1357 printk(KERN_INFO
"md: bind<%s>\n", b
);
1359 rdev
->kobj
.parent
= &mddev
->kobj
;
1360 if ((err
= kobject_add(&rdev
->kobj
)))
1363 if (rdev
->bdev
->bd_part
)
1364 ko
= &rdev
->bdev
->bd_part
->kobj
;
1366 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1367 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1368 kobject_del(&rdev
->kobj
);
1371 list_add(&rdev
->same_set
, &mddev
->disks
);
1372 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1376 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1381 static void delayed_delete(struct work_struct
*ws
)
1383 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1384 kobject_del(&rdev
->kobj
);
1387 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1389 char b
[BDEVNAME_SIZE
];
1394 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1395 list_del_init(&rdev
->same_set
);
1396 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1398 sysfs_remove_link(&rdev
->kobj
, "block");
1400 /* We need to delay this, otherwise we can deadlock when
1401 * writing to 'remove' to "dev/state"
1403 INIT_WORK(&rdev
->del_work
, delayed_delete
);
1404 schedule_work(&rdev
->del_work
);
1408 * prevent the device from being mounted, repartitioned or
1409 * otherwise reused by a RAID array (or any other kernel
1410 * subsystem), by bd_claiming the device.
1412 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1415 struct block_device
*bdev
;
1416 char b
[BDEVNAME_SIZE
];
1418 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1420 printk(KERN_ERR
"md: could not open %s.\n",
1421 __bdevname(dev
, b
));
1422 return PTR_ERR(bdev
);
1424 err
= bd_claim(bdev
, rdev
);
1426 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1435 static void unlock_rdev(mdk_rdev_t
*rdev
)
1437 struct block_device
*bdev
= rdev
->bdev
;
1445 void md_autodetect_dev(dev_t dev
);
1447 static void export_rdev(mdk_rdev_t
* rdev
)
1449 char b
[BDEVNAME_SIZE
];
1450 printk(KERN_INFO
"md: export_rdev(%s)\n",
1451 bdevname(rdev
->bdev
,b
));
1455 list_del_init(&rdev
->same_set
);
1457 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1460 kobject_put(&rdev
->kobj
);
1463 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1465 unbind_rdev_from_array(rdev
);
1469 static void export_array(mddev_t
*mddev
)
1471 struct list_head
*tmp
;
1474 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1479 kick_rdev_from_array(rdev
);
1481 if (!list_empty(&mddev
->disks
))
1483 mddev
->raid_disks
= 0;
1484 mddev
->major_version
= 0;
1487 static void print_desc(mdp_disk_t
*desc
)
1489 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1490 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1493 static void print_sb(mdp_super_t
*sb
)
1498 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1499 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1500 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1502 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1503 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1504 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1505 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1506 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1507 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1508 sb
->failed_disks
, sb
->spare_disks
,
1509 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1512 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1515 desc
= sb
->disks
+ i
;
1516 if (desc
->number
|| desc
->major
|| desc
->minor
||
1517 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1518 printk(" D %2d: ", i
);
1522 printk(KERN_INFO
"md: THIS: ");
1523 print_desc(&sb
->this_disk
);
1527 static void print_rdev(mdk_rdev_t
*rdev
)
1529 char b
[BDEVNAME_SIZE
];
1530 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1531 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1532 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1534 if (rdev
->sb_loaded
) {
1535 printk(KERN_INFO
"md: rdev superblock:\n");
1536 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1538 printk(KERN_INFO
"md: no rdev superblock!\n");
1541 static void md_print_devices(void)
1543 struct list_head
*tmp
, *tmp2
;
1546 char b
[BDEVNAME_SIZE
];
1549 printk("md: **********************************\n");
1550 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1551 printk("md: **********************************\n");
1552 ITERATE_MDDEV(mddev
,tmp
) {
1555 bitmap_print_sb(mddev
->bitmap
);
1557 printk("%s: ", mdname(mddev
));
1558 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1559 printk("<%s>", bdevname(rdev
->bdev
,b
));
1562 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1565 printk("md: **********************************\n");
1570 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1572 /* Update each superblock (in-memory image), but
1573 * if we are allowed to, skip spares which already
1574 * have the right event counter, or have one earlier
1575 * (which would mean they aren't being marked as dirty
1576 * with the rest of the array)
1579 struct list_head
*tmp
;
1581 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1582 if (rdev
->sb_events
== mddev
->events
||
1584 rdev
->raid_disk
< 0 &&
1585 (rdev
->sb_events
&1)==0 &&
1586 rdev
->sb_events
+1 == mddev
->events
)) {
1587 /* Don't update this superblock */
1588 rdev
->sb_loaded
= 2;
1590 super_types
[mddev
->major_version
].
1591 sync_super(mddev
, rdev
);
1592 rdev
->sb_loaded
= 1;
1597 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1600 struct list_head
*tmp
;
1606 spin_lock_irq(&mddev
->write_lock
);
1608 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1609 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1611 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1612 /* just a clean<-> dirty transition, possibly leave spares alone,
1613 * though if events isn't the right even/odd, we will have to do
1619 if (mddev
->degraded
)
1620 /* If the array is degraded, then skipping spares is both
1621 * dangerous and fairly pointless.
1622 * Dangerous because a device that was removed from the array
1623 * might have a event_count that still looks up-to-date,
1624 * so it can be re-added without a resync.
1625 * Pointless because if there are any spares to skip,
1626 * then a recovery will happen and soon that array won't
1627 * be degraded any more and the spare can go back to sleep then.
1631 sync_req
= mddev
->in_sync
;
1632 mddev
->utime
= get_seconds();
1634 /* If this is just a dirty<->clean transition, and the array is clean
1635 * and 'events' is odd, we can roll back to the previous clean state */
1637 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1638 && (mddev
->events
& 1)
1639 && mddev
->events
!= 1)
1642 /* otherwise we have to go forward and ... */
1644 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1645 /* .. if the array isn't clean, insist on an odd 'events' */
1646 if ((mddev
->events
&1)==0) {
1651 /* otherwise insist on an even 'events' (for clean states) */
1652 if ((mddev
->events
&1)) {
1659 if (!mddev
->events
) {
1661 * oops, this 64-bit counter should never wrap.
1662 * Either we are in around ~1 trillion A.C., assuming
1663 * 1 reboot per second, or we have a bug:
1668 sync_sbs(mddev
, nospares
);
1671 * do not write anything to disk if using
1672 * nonpersistent superblocks
1674 if (!mddev
->persistent
) {
1675 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1676 spin_unlock_irq(&mddev
->write_lock
);
1677 wake_up(&mddev
->sb_wait
);
1680 spin_unlock_irq(&mddev
->write_lock
);
1683 "md: updating %s RAID superblock on device (in sync %d)\n",
1684 mdname(mddev
),mddev
->in_sync
);
1686 err
= bitmap_update_sb(mddev
->bitmap
);
1687 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1688 char b
[BDEVNAME_SIZE
];
1689 dprintk(KERN_INFO
"md: ");
1690 if (rdev
->sb_loaded
!= 1)
1691 continue; /* no noise on spare devices */
1692 if (test_bit(Faulty
, &rdev
->flags
))
1693 dprintk("(skipping faulty ");
1695 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1696 if (!test_bit(Faulty
, &rdev
->flags
)) {
1697 md_super_write(mddev
,rdev
,
1698 rdev
->sb_offset
<<1, rdev
->sb_size
,
1700 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1701 bdevname(rdev
->bdev
,b
),
1702 (unsigned long long)rdev
->sb_offset
);
1703 rdev
->sb_events
= mddev
->events
;
1707 if (mddev
->level
== LEVEL_MULTIPATH
)
1708 /* only need to write one superblock... */
1711 md_super_wait(mddev
);
1712 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1714 spin_lock_irq(&mddev
->write_lock
);
1715 if (mddev
->in_sync
!= sync_req
||
1716 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
1717 /* have to write it out again */
1718 spin_unlock_irq(&mddev
->write_lock
);
1721 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1722 spin_unlock_irq(&mddev
->write_lock
);
1723 wake_up(&mddev
->sb_wait
);
1727 /* words written to sysfs files may, or my not, be \n terminated.
1728 * We want to accept with case. For this we use cmd_match.
1730 static int cmd_match(const char *cmd
, const char *str
)
1732 /* See if cmd, written into a sysfs file, matches
1733 * str. They must either be the same, or cmd can
1734 * have a trailing newline
1736 while (*cmd
&& *str
&& *cmd
== *str
) {
1747 struct rdev_sysfs_entry
{
1748 struct attribute attr
;
1749 ssize_t (*show
)(mdk_rdev_t
*, char *);
1750 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1754 state_show(mdk_rdev_t
*rdev
, char *page
)
1759 if (test_bit(Faulty
, &rdev
->flags
)) {
1760 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1763 if (test_bit(In_sync
, &rdev
->flags
)) {
1764 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1767 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1768 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1771 if (!test_bit(Faulty
, &rdev
->flags
) &&
1772 !test_bit(In_sync
, &rdev
->flags
)) {
1773 len
+= sprintf(page
+len
, "%sspare", sep
);
1776 return len
+sprintf(page
+len
, "\n");
1780 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1783 * faulty - simulates and error
1784 * remove - disconnects the device
1785 * writemostly - sets write_mostly
1786 * -writemostly - clears write_mostly
1789 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1790 md_error(rdev
->mddev
, rdev
);
1792 } else if (cmd_match(buf
, "remove")) {
1793 if (rdev
->raid_disk
>= 0)
1796 mddev_t
*mddev
= rdev
->mddev
;
1797 kick_rdev_from_array(rdev
);
1799 md_update_sb(mddev
, 1);
1800 md_new_event(mddev
);
1803 } else if (cmd_match(buf
, "writemostly")) {
1804 set_bit(WriteMostly
, &rdev
->flags
);
1806 } else if (cmd_match(buf
, "-writemostly")) {
1807 clear_bit(WriteMostly
, &rdev
->flags
);
1810 return err
? err
: len
;
1812 static struct rdev_sysfs_entry rdev_state
=
1813 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
1816 super_show(mdk_rdev_t
*rdev
, char *page
)
1818 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1819 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1820 return rdev
->sb_size
;
1824 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1827 errors_show(mdk_rdev_t
*rdev
, char *page
)
1829 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1833 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1836 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1837 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1838 atomic_set(&rdev
->corrected_errors
, n
);
1843 static struct rdev_sysfs_entry rdev_errors
=
1844 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
1847 slot_show(mdk_rdev_t
*rdev
, char *page
)
1849 if (rdev
->raid_disk
< 0)
1850 return sprintf(page
, "none\n");
1852 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1856 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1859 int slot
= simple_strtoul(buf
, &e
, 10);
1860 if (strncmp(buf
, "none", 4)==0)
1862 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1864 if (rdev
->mddev
->pers
)
1865 /* Cannot set slot in active array (yet) */
1867 if (slot
>= rdev
->mddev
->raid_disks
)
1869 rdev
->raid_disk
= slot
;
1870 /* assume it is working */
1872 set_bit(In_sync
, &rdev
->flags
);
1877 static struct rdev_sysfs_entry rdev_slot
=
1878 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
1881 offset_show(mdk_rdev_t
*rdev
, char *page
)
1883 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1887 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1890 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1891 if (e
==buf
|| (*e
&& *e
!= '\n'))
1893 if (rdev
->mddev
->pers
)
1895 rdev
->data_offset
= offset
;
1899 static struct rdev_sysfs_entry rdev_offset
=
1900 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
1903 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1905 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1909 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1912 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1913 if (e
==buf
|| (*e
&& *e
!= '\n'))
1915 if (rdev
->mddev
->pers
)
1918 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1919 rdev
->mddev
->size
= size
;
1923 static struct rdev_sysfs_entry rdev_size
=
1924 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
1926 static struct attribute
*rdev_default_attrs
[] = {
1936 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1938 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1939 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1943 return entry
->show(rdev
, page
);
1947 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1948 const char *page
, size_t length
)
1950 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1951 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1955 if (!capable(CAP_SYS_ADMIN
))
1957 return entry
->store(rdev
, page
, length
);
1960 static void rdev_free(struct kobject
*ko
)
1962 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1965 static struct sysfs_ops rdev_sysfs_ops
= {
1966 .show
= rdev_attr_show
,
1967 .store
= rdev_attr_store
,
1969 static struct kobj_type rdev_ktype
= {
1970 .release
= rdev_free
,
1971 .sysfs_ops
= &rdev_sysfs_ops
,
1972 .default_attrs
= rdev_default_attrs
,
1976 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1978 * mark the device faulty if:
1980 * - the device is nonexistent (zero size)
1981 * - the device has no valid superblock
1983 * a faulty rdev _never_ has rdev->sb set.
1985 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1987 char b
[BDEVNAME_SIZE
];
1992 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1994 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1995 return ERR_PTR(-ENOMEM
);
1998 if ((err
= alloc_disk_sb(rdev
)))
2001 err
= lock_rdev(rdev
, newdev
);
2005 rdev
->kobj
.parent
= NULL
;
2006 rdev
->kobj
.ktype
= &rdev_ktype
;
2007 kobject_init(&rdev
->kobj
);
2010 rdev
->saved_raid_disk
= -1;
2011 rdev
->raid_disk
= -1;
2013 rdev
->data_offset
= 0;
2014 rdev
->sb_events
= 0;
2015 atomic_set(&rdev
->nr_pending
, 0);
2016 atomic_set(&rdev
->read_errors
, 0);
2017 atomic_set(&rdev
->corrected_errors
, 0);
2019 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2022 "md: %s has zero or unknown size, marking faulty!\n",
2023 bdevname(rdev
->bdev
,b
));
2028 if (super_format
>= 0) {
2029 err
= super_types
[super_format
].
2030 load_super(rdev
, NULL
, super_minor
);
2031 if (err
== -EINVAL
) {
2033 "md: %s has invalid sb, not importing!\n",
2034 bdevname(rdev
->bdev
,b
));
2039 "md: could not read %s's sb, not importing!\n",
2040 bdevname(rdev
->bdev
,b
));
2044 INIT_LIST_HEAD(&rdev
->same_set
);
2049 if (rdev
->sb_page
) {
2055 return ERR_PTR(err
);
2059 * Check a full RAID array for plausibility
2063 static void analyze_sbs(mddev_t
* mddev
)
2066 struct list_head
*tmp
;
2067 mdk_rdev_t
*rdev
, *freshest
;
2068 char b
[BDEVNAME_SIZE
];
2071 ITERATE_RDEV(mddev
,rdev
,tmp
)
2072 switch (super_types
[mddev
->major_version
].
2073 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2081 "md: fatal superblock inconsistency in %s"
2082 " -- removing from array\n",
2083 bdevname(rdev
->bdev
,b
));
2084 kick_rdev_from_array(rdev
);
2088 super_types
[mddev
->major_version
].
2089 validate_super(mddev
, freshest
);
2092 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2093 if (rdev
!= freshest
)
2094 if (super_types
[mddev
->major_version
].
2095 validate_super(mddev
, rdev
)) {
2096 printk(KERN_WARNING
"md: kicking non-fresh %s"
2098 bdevname(rdev
->bdev
,b
));
2099 kick_rdev_from_array(rdev
);
2102 if (mddev
->level
== LEVEL_MULTIPATH
) {
2103 rdev
->desc_nr
= i
++;
2104 rdev
->raid_disk
= rdev
->desc_nr
;
2105 set_bit(In_sync
, &rdev
->flags
);
2111 if (mddev
->recovery_cp
!= MaxSector
&&
2113 printk(KERN_ERR
"md: %s: raid array is not clean"
2114 " -- starting background reconstruction\n",
2120 safe_delay_show(mddev_t
*mddev
, char *page
)
2122 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2123 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2126 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2134 /* remove a period, and count digits after it */
2135 if (len
>= sizeof(buf
))
2137 strlcpy(buf
, cbuf
, len
);
2139 for (i
=0; i
<len
; i
++) {
2141 if (isdigit(buf
[i
])) {
2146 } else if (buf
[i
] == '.') {
2151 msec
= simple_strtoul(buf
, &e
, 10);
2152 if (e
== buf
|| (*e
&& *e
!= '\n'))
2154 msec
= (msec
* 1000) / scale
;
2156 mddev
->safemode_delay
= 0;
2158 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2159 if (mddev
->safemode_delay
== 0)
2160 mddev
->safemode_delay
= 1;
2164 static struct md_sysfs_entry md_safe_delay
=
2165 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2168 level_show(mddev_t
*mddev
, char *page
)
2170 struct mdk_personality
*p
= mddev
->pers
;
2172 return sprintf(page
, "%s\n", p
->name
);
2173 else if (mddev
->clevel
[0])
2174 return sprintf(page
, "%s\n", mddev
->clevel
);
2175 else if (mddev
->level
!= LEVEL_NONE
)
2176 return sprintf(page
, "%d\n", mddev
->level
);
2182 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2189 if (len
>= sizeof(mddev
->clevel
))
2191 strncpy(mddev
->clevel
, buf
, len
);
2192 if (mddev
->clevel
[len
-1] == '\n')
2194 mddev
->clevel
[len
] = 0;
2195 mddev
->level
= LEVEL_NONE
;
2199 static struct md_sysfs_entry md_level
=
2200 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2204 layout_show(mddev_t
*mddev
, char *page
)
2206 /* just a number, not meaningful for all levels */
2207 return sprintf(page
, "%d\n", mddev
->layout
);
2211 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2214 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2218 if (!*buf
|| (*e
&& *e
!= '\n'))
2224 static struct md_sysfs_entry md_layout
=
2225 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2229 raid_disks_show(mddev_t
*mddev
, char *page
)
2231 if (mddev
->raid_disks
== 0)
2233 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2236 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2239 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2243 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2245 if (!*buf
|| (*e
&& *e
!= '\n'))
2249 rv
= update_raid_disks(mddev
, n
);
2251 mddev
->raid_disks
= n
;
2252 return rv
? rv
: len
;
2254 static struct md_sysfs_entry md_raid_disks
=
2255 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2258 chunk_size_show(mddev_t
*mddev
, char *page
)
2260 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2264 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2266 /* can only set chunk_size if array is not yet active */
2268 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2272 if (!*buf
|| (*e
&& *e
!= '\n'))
2275 mddev
->chunk_size
= n
;
2278 static struct md_sysfs_entry md_chunk_size
=
2279 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2282 resync_start_show(mddev_t
*mddev
, char *page
)
2284 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2288 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2290 /* can only set chunk_size if array is not yet active */
2292 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2296 if (!*buf
|| (*e
&& *e
!= '\n'))
2299 mddev
->recovery_cp
= n
;
2302 static struct md_sysfs_entry md_resync_start
=
2303 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2306 * The array state can be:
2309 * No devices, no size, no level
2310 * Equivalent to STOP_ARRAY ioctl
2312 * May have some settings, but array is not active
2313 * all IO results in error
2314 * When written, doesn't tear down array, but just stops it
2315 * suspended (not supported yet)
2316 * All IO requests will block. The array can be reconfigured.
2317 * Writing this, if accepted, will block until array is quiessent
2319 * no resync can happen. no superblocks get written.
2320 * write requests fail
2322 * like readonly, but behaves like 'clean' on a write request.
2324 * clean - no pending writes, but otherwise active.
2325 * When written to inactive array, starts without resync
2326 * If a write request arrives then
2327 * if metadata is known, mark 'dirty' and switch to 'active'.
2328 * if not known, block and switch to write-pending
2329 * If written to an active array that has pending writes, then fails.
2331 * fully active: IO and resync can be happening.
2332 * When written to inactive array, starts with resync
2335 * clean, but writes are blocked waiting for 'active' to be written.
2338 * like active, but no writes have been seen for a while (100msec).
2341 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2342 write_pending
, active_idle
, bad_word
};
2343 static char *array_states
[] = {
2344 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2345 "write-pending", "active-idle", NULL
};
2347 static int match_word(const char *word
, char **list
)
2350 for (n
=0; list
[n
]; n
++)
2351 if (cmd_match(word
, list
[n
]))
2357 array_state_show(mddev_t
*mddev
, char *page
)
2359 enum array_state st
= inactive
;
2372 else if (mddev
->safemode
)
2378 if (list_empty(&mddev
->disks
) &&
2379 mddev
->raid_disks
== 0 &&
2385 return sprintf(page
, "%s\n", array_states
[st
]);
2388 static int do_md_stop(mddev_t
* mddev
, int ro
);
2389 static int do_md_run(mddev_t
* mddev
);
2390 static int restart_array(mddev_t
*mddev
);
2393 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2396 enum array_state st
= match_word(buf
, array_states
);
2401 /* stopping an active array */
2403 if (atomic_read(&mddev
->active
) > 1)
2405 err
= do_md_stop(mddev
, 0);
2409 /* stopping an active array */
2411 if (atomic_read(&mddev
->active
) > 1)
2413 err
= do_md_stop(mddev
, 2);
2417 break; /* not supported yet */
2420 err
= do_md_stop(mddev
, 1);
2423 err
= do_md_run(mddev
);
2427 /* stopping an active array */
2429 err
= do_md_stop(mddev
, 1);
2431 mddev
->ro
= 2; /* FIXME mark devices writable */
2434 err
= do_md_run(mddev
);
2439 restart_array(mddev
);
2440 spin_lock_irq(&mddev
->write_lock
);
2441 if (atomic_read(&mddev
->writes_pending
) == 0) {
2443 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2445 spin_unlock_irq(&mddev
->write_lock
);
2448 mddev
->recovery_cp
= MaxSector
;
2449 err
= do_md_run(mddev
);
2454 restart_array(mddev
);
2455 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2456 wake_up(&mddev
->sb_wait
);
2460 err
= do_md_run(mddev
);
2465 /* these cannot be set */
2473 static struct md_sysfs_entry md_array_state
=
2474 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
2477 null_show(mddev_t
*mddev
, char *page
)
2483 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2485 /* buf must be %d:%d\n? giving major and minor numbers */
2486 /* The new device is added to the array.
2487 * If the array has a persistent superblock, we read the
2488 * superblock to initialise info and check validity.
2489 * Otherwise, only checking done is that in bind_rdev_to_array,
2490 * which mainly checks size.
2493 int major
= simple_strtoul(buf
, &e
, 10);
2499 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2501 minor
= simple_strtoul(e
+1, &e
, 10);
2502 if (*e
&& *e
!= '\n')
2504 dev
= MKDEV(major
, minor
);
2505 if (major
!= MAJOR(dev
) ||
2506 minor
!= MINOR(dev
))
2510 if (mddev
->persistent
) {
2511 rdev
= md_import_device(dev
, mddev
->major_version
,
2512 mddev
->minor_version
);
2513 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2514 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2515 mdk_rdev_t
, same_set
);
2516 err
= super_types
[mddev
->major_version
]
2517 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2522 rdev
= md_import_device(dev
, -1, -1);
2525 return PTR_ERR(rdev
);
2526 err
= bind_rdev_to_array(rdev
, mddev
);
2530 return err
? err
: len
;
2533 static struct md_sysfs_entry md_new_device
=
2534 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
2537 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2540 unsigned long chunk
, end_chunk
;
2544 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2546 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
2547 if (buf
== end
) break;
2548 if (*end
== '-') { /* range */
2550 end_chunk
= simple_strtoul(buf
, &end
, 0);
2551 if (buf
== end
) break;
2553 if (*end
&& !isspace(*end
)) break;
2554 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
2556 while (isspace(*buf
)) buf
++;
2558 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
2563 static struct md_sysfs_entry md_bitmap
=
2564 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
2567 size_show(mddev_t
*mddev
, char *page
)
2569 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2572 static int update_size(mddev_t
*mddev
, unsigned long size
);
2575 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2577 /* If array is inactive, we can reduce the component size, but
2578 * not increase it (except from 0).
2579 * If array is active, we can try an on-line resize
2583 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2584 if (!*buf
|| *buf
== '\n' ||
2589 err
= update_size(mddev
, size
);
2590 md_update_sb(mddev
, 1);
2592 if (mddev
->size
== 0 ||
2598 return err
? err
: len
;
2601 static struct md_sysfs_entry md_size
=
2602 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
2606 * This is either 'none' for arrays with externally managed metadata,
2607 * or N.M for internally known formats
2610 metadata_show(mddev_t
*mddev
, char *page
)
2612 if (mddev
->persistent
)
2613 return sprintf(page
, "%d.%d\n",
2614 mddev
->major_version
, mddev
->minor_version
);
2616 return sprintf(page
, "none\n");
2620 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2624 if (!list_empty(&mddev
->disks
))
2627 if (cmd_match(buf
, "none")) {
2628 mddev
->persistent
= 0;
2629 mddev
->major_version
= 0;
2630 mddev
->minor_version
= 90;
2633 major
= simple_strtoul(buf
, &e
, 10);
2634 if (e
==buf
|| *e
!= '.')
2637 minor
= simple_strtoul(buf
, &e
, 10);
2638 if (e
==buf
|| (*e
&& *e
!= '\n') )
2640 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2641 super_types
[major
].name
== NULL
)
2643 mddev
->major_version
= major
;
2644 mddev
->minor_version
= minor
;
2645 mddev
->persistent
= 1;
2649 static struct md_sysfs_entry md_metadata
=
2650 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2653 action_show(mddev_t
*mddev
, char *page
)
2655 char *type
= "idle";
2656 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2657 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2658 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2660 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2661 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2663 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2670 return sprintf(page
, "%s\n", type
);
2674 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2676 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2679 if (cmd_match(page
, "idle")) {
2680 if (mddev
->sync_thread
) {
2681 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2682 md_unregister_thread(mddev
->sync_thread
);
2683 mddev
->sync_thread
= NULL
;
2684 mddev
->recovery
= 0;
2686 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2687 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2689 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2690 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2691 else if (cmd_match(page
, "reshape")) {
2693 if (mddev
->pers
->start_reshape
== NULL
)
2695 err
= mddev
->pers
->start_reshape(mddev
);
2699 if (cmd_match(page
, "check"))
2700 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2701 else if (!cmd_match(page
, "repair"))
2703 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2704 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2706 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2707 md_wakeup_thread(mddev
->thread
);
2712 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2714 return sprintf(page
, "%llu\n",
2715 (unsigned long long) mddev
->resync_mismatches
);
2718 static struct md_sysfs_entry md_scan_mode
=
2719 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2722 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
2725 sync_min_show(mddev_t
*mddev
, char *page
)
2727 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2728 mddev
->sync_speed_min
? "local": "system");
2732 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2736 if (strncmp(buf
, "system", 6)==0) {
2737 mddev
->sync_speed_min
= 0;
2740 min
= simple_strtoul(buf
, &e
, 10);
2741 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2743 mddev
->sync_speed_min
= min
;
2747 static struct md_sysfs_entry md_sync_min
=
2748 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2751 sync_max_show(mddev_t
*mddev
, char *page
)
2753 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2754 mddev
->sync_speed_max
? "local": "system");
2758 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2762 if (strncmp(buf
, "system", 6)==0) {
2763 mddev
->sync_speed_max
= 0;
2766 max
= simple_strtoul(buf
, &e
, 10);
2767 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2769 mddev
->sync_speed_max
= max
;
2773 static struct md_sysfs_entry md_sync_max
=
2774 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2778 sync_speed_show(mddev_t
*mddev
, char *page
)
2780 unsigned long resync
, dt
, db
;
2781 resync
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
));
2782 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2784 db
= resync
- (mddev
->resync_mark_cnt
);
2785 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2788 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
2791 sync_completed_show(mddev_t
*mddev
, char *page
)
2793 unsigned long max_blocks
, resync
;
2795 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2796 max_blocks
= mddev
->resync_max_sectors
;
2798 max_blocks
= mddev
->size
<< 1;
2800 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2801 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2804 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
2807 suspend_lo_show(mddev_t
*mddev
, char *page
)
2809 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2813 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2816 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2818 if (mddev
->pers
->quiesce
== NULL
)
2820 if (buf
== e
|| (*e
&& *e
!= '\n'))
2822 if (new >= mddev
->suspend_hi
||
2823 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2824 mddev
->suspend_lo
= new;
2825 mddev
->pers
->quiesce(mddev
, 2);
2830 static struct md_sysfs_entry md_suspend_lo
=
2831 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2835 suspend_hi_show(mddev_t
*mddev
, char *page
)
2837 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2841 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2844 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2846 if (mddev
->pers
->quiesce
== NULL
)
2848 if (buf
== e
|| (*e
&& *e
!= '\n'))
2850 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2851 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2852 mddev
->suspend_hi
= new;
2853 mddev
->pers
->quiesce(mddev
, 1);
2854 mddev
->pers
->quiesce(mddev
, 0);
2859 static struct md_sysfs_entry md_suspend_hi
=
2860 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2863 static struct attribute
*md_default_attrs
[] = {
2866 &md_raid_disks
.attr
,
2867 &md_chunk_size
.attr
,
2869 &md_resync_start
.attr
,
2871 &md_new_device
.attr
,
2872 &md_safe_delay
.attr
,
2873 &md_array_state
.attr
,
2877 static struct attribute
*md_redundancy_attrs
[] = {
2879 &md_mismatches
.attr
,
2882 &md_sync_speed
.attr
,
2883 &md_sync_completed
.attr
,
2884 &md_suspend_lo
.attr
,
2885 &md_suspend_hi
.attr
,
2889 static struct attribute_group md_redundancy_group
= {
2891 .attrs
= md_redundancy_attrs
,
2896 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2898 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2899 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2904 rv
= mddev_lock(mddev
);
2906 rv
= entry
->show(mddev
, page
);
2907 mddev_unlock(mddev
);
2913 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2914 const char *page
, size_t length
)
2916 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2917 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2922 if (!capable(CAP_SYS_ADMIN
))
2924 rv
= mddev_lock(mddev
);
2926 rv
= entry
->store(mddev
, page
, length
);
2927 mddev_unlock(mddev
);
2932 static void md_free(struct kobject
*ko
)
2934 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2938 static struct sysfs_ops md_sysfs_ops
= {
2939 .show
= md_attr_show
,
2940 .store
= md_attr_store
,
2942 static struct kobj_type md_ktype
= {
2944 .sysfs_ops
= &md_sysfs_ops
,
2945 .default_attrs
= md_default_attrs
,
2950 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2952 static DEFINE_MUTEX(disks_mutex
);
2953 mddev_t
*mddev
= mddev_find(dev
);
2954 struct gendisk
*disk
;
2955 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2956 int shift
= partitioned
? MdpMinorShift
: 0;
2957 int unit
= MINOR(dev
) >> shift
;
2962 mutex_lock(&disks_mutex
);
2963 if (mddev
->gendisk
) {
2964 mutex_unlock(&disks_mutex
);
2968 disk
= alloc_disk(1 << shift
);
2970 mutex_unlock(&disks_mutex
);
2974 disk
->major
= MAJOR(dev
);
2975 disk
->first_minor
= unit
<< shift
;
2977 sprintf(disk
->disk_name
, "md_d%d", unit
);
2979 sprintf(disk
->disk_name
, "md%d", unit
);
2980 disk
->fops
= &md_fops
;
2981 disk
->private_data
= mddev
;
2982 disk
->queue
= mddev
->queue
;
2984 mddev
->gendisk
= disk
;
2985 mutex_unlock(&disks_mutex
);
2986 mddev
->kobj
.parent
= &disk
->kobj
;
2987 mddev
->kobj
.k_name
= NULL
;
2988 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2989 mddev
->kobj
.ktype
= &md_ktype
;
2990 if (kobject_register(&mddev
->kobj
))
2991 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
2996 static void md_safemode_timeout(unsigned long data
)
2998 mddev_t
*mddev
= (mddev_t
*) data
;
3000 mddev
->safemode
= 1;
3001 md_wakeup_thread(mddev
->thread
);
3004 static int start_dirty_degraded
;
3006 static int do_md_run(mddev_t
* mddev
)
3010 struct list_head
*tmp
;
3012 struct gendisk
*disk
;
3013 struct mdk_personality
*pers
;
3014 char b
[BDEVNAME_SIZE
];
3016 if (list_empty(&mddev
->disks
))
3017 /* cannot run an array with no devices.. */
3024 * Analyze all RAID superblock(s)
3026 if (!mddev
->raid_disks
)
3029 chunk_size
= mddev
->chunk_size
;
3032 if (chunk_size
> MAX_CHUNK_SIZE
) {
3033 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
3034 chunk_size
, MAX_CHUNK_SIZE
);
3038 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3040 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
3041 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
3044 if (chunk_size
< PAGE_SIZE
) {
3045 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
3046 chunk_size
, PAGE_SIZE
);
3050 /* devices must have minimum size of one chunk */
3051 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3052 if (test_bit(Faulty
, &rdev
->flags
))
3054 if (rdev
->size
< chunk_size
/ 1024) {
3056 "md: Dev %s smaller than chunk_size:"
3058 bdevname(rdev
->bdev
,b
),
3059 (unsigned long long)rdev
->size
,
3067 if (mddev
->level
!= LEVEL_NONE
)
3068 request_module("md-level-%d", mddev
->level
);
3069 else if (mddev
->clevel
[0])
3070 request_module("md-%s", mddev
->clevel
);
3074 * Drop all container device buffers, from now on
3075 * the only valid external interface is through the md
3077 * Also find largest hardsector size
3079 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3080 if (test_bit(Faulty
, &rdev
->flags
))
3082 sync_blockdev(rdev
->bdev
);
3083 invalidate_bdev(rdev
->bdev
, 0);
3086 md_probe(mddev
->unit
, NULL
, NULL
);
3087 disk
= mddev
->gendisk
;
3091 spin_lock(&pers_lock
);
3092 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3093 if (!pers
|| !try_module_get(pers
->owner
)) {
3094 spin_unlock(&pers_lock
);
3095 if (mddev
->level
!= LEVEL_NONE
)
3096 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3099 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3104 spin_unlock(&pers_lock
);
3105 mddev
->level
= pers
->level
;
3106 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3108 if (mddev
->reshape_position
!= MaxSector
&&
3109 pers
->start_reshape
== NULL
) {
3110 /* This personality cannot handle reshaping... */
3112 module_put(pers
->owner
);
3116 if (pers
->sync_request
) {
3117 /* Warn if this is a potentially silly
3120 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3122 struct list_head
*tmp2
;
3124 ITERATE_RDEV(mddev
, rdev
, tmp
) {
3125 ITERATE_RDEV(mddev
, rdev2
, tmp2
) {
3127 rdev
->bdev
->bd_contains
==
3128 rdev2
->bdev
->bd_contains
) {
3130 "%s: WARNING: %s appears to be"
3131 " on the same physical disk as"
3134 bdevname(rdev
->bdev
,b
),
3135 bdevname(rdev2
->bdev
,b2
));
3142 "True protection against single-disk"
3143 " failure might be compromised.\n");
3146 mddev
->recovery
= 0;
3147 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3148 mddev
->barriers_work
= 1;
3149 mddev
->ok_start_degraded
= start_dirty_degraded
;
3152 mddev
->ro
= 2; /* read-only, but switch on first write */
3154 err
= mddev
->pers
->run(mddev
);
3155 if (!err
&& mddev
->pers
->sync_request
) {
3156 err
= bitmap_create(mddev
);
3158 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3159 mdname(mddev
), err
);
3160 mddev
->pers
->stop(mddev
);
3164 printk(KERN_ERR
"md: pers->run() failed ...\n");
3165 module_put(mddev
->pers
->owner
);
3167 bitmap_destroy(mddev
);
3170 if (mddev
->pers
->sync_request
) {
3171 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3173 "md: cannot register extra attributes for %s\n",
3175 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3178 atomic_set(&mddev
->writes_pending
,0);
3179 mddev
->safemode
= 0;
3180 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3181 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3182 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3185 ITERATE_RDEV(mddev
,rdev
,tmp
)
3186 if (rdev
->raid_disk
>= 0) {
3188 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3189 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
3190 printk("md: cannot register %s for %s\n",
3194 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3197 md_update_sb(mddev
, 0);
3199 set_capacity(disk
, mddev
->array_size
<<1);
3201 /* If we call blk_queue_make_request here, it will
3202 * re-initialise max_sectors etc which may have been
3203 * refined inside -> run. So just set the bits we need to set.
3204 * Most initialisation happended when we called
3205 * blk_queue_make_request(..., md_fail_request)
3208 mddev
->queue
->queuedata
= mddev
;
3209 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3211 /* If there is a partially-recovered drive we need to
3212 * start recovery here. If we leave it to md_check_recovery,
3213 * it will remove the drives and not do the right thing
3215 if (mddev
->degraded
&& !mddev
->sync_thread
) {
3216 struct list_head
*rtmp
;
3218 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3219 if (rdev
->raid_disk
>= 0 &&
3220 !test_bit(In_sync
, &rdev
->flags
) &&
3221 !test_bit(Faulty
, &rdev
->flags
))
3222 /* complete an interrupted recovery */
3224 if (spares
&& mddev
->pers
->sync_request
) {
3225 mddev
->recovery
= 0;
3226 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3227 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3230 if (!mddev
->sync_thread
) {
3231 printk(KERN_ERR
"%s: could not start resync"
3234 /* leave the spares where they are, it shouldn't hurt */
3235 mddev
->recovery
= 0;
3239 md_wakeup_thread(mddev
->thread
);
3240 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
3243 md_new_event(mddev
);
3244 kobject_uevent(&mddev
->gendisk
->kobj
, KOBJ_CHANGE
);
3248 static int restart_array(mddev_t
*mddev
)
3250 struct gendisk
*disk
= mddev
->gendisk
;
3254 * Complain if it has no devices
3257 if (list_empty(&mddev
->disks
))
3265 mddev
->safemode
= 0;
3267 set_disk_ro(disk
, 0);
3269 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3272 * Kick recovery or resync if necessary
3274 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3275 md_wakeup_thread(mddev
->thread
);
3276 md_wakeup_thread(mddev
->sync_thread
);
3285 /* similar to deny_write_access, but accounts for our holding a reference
3286 * to the file ourselves */
3287 static int deny_bitmap_write_access(struct file
* file
)
3289 struct inode
*inode
= file
->f_mapping
->host
;
3291 spin_lock(&inode
->i_lock
);
3292 if (atomic_read(&inode
->i_writecount
) > 1) {
3293 spin_unlock(&inode
->i_lock
);
3296 atomic_set(&inode
->i_writecount
, -1);
3297 spin_unlock(&inode
->i_lock
);
3302 static void restore_bitmap_write_access(struct file
*file
)
3304 struct inode
*inode
= file
->f_mapping
->host
;
3306 spin_lock(&inode
->i_lock
);
3307 atomic_set(&inode
->i_writecount
, 1);
3308 spin_unlock(&inode
->i_lock
);
3312 * 0 - completely stop and dis-assemble array
3313 * 1 - switch to readonly
3314 * 2 - stop but do not disassemble array
3316 static int do_md_stop(mddev_t
* mddev
, int mode
)
3319 struct gendisk
*disk
= mddev
->gendisk
;
3322 if (atomic_read(&mddev
->active
)>2) {
3323 printk("md: %s still in use.\n",mdname(mddev
));
3327 if (mddev
->sync_thread
) {
3328 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3329 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3330 md_unregister_thread(mddev
->sync_thread
);
3331 mddev
->sync_thread
= NULL
;
3334 del_timer_sync(&mddev
->safemode_timer
);
3336 invalidate_partition(disk
, 0);
3339 case 1: /* readonly */
3345 case 0: /* disassemble */
3347 bitmap_flush(mddev
);
3348 md_super_wait(mddev
);
3350 set_disk_ro(disk
, 0);
3351 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3352 mddev
->pers
->stop(mddev
);
3353 mddev
->queue
->merge_bvec_fn
= NULL
;
3354 mddev
->queue
->unplug_fn
= NULL
;
3355 mddev
->queue
->issue_flush_fn
= NULL
;
3356 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
3357 if (mddev
->pers
->sync_request
)
3358 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3360 module_put(mddev
->pers
->owner
);
3363 set_capacity(disk
, 0);
3369 if (!mddev
->in_sync
|| mddev
->flags
) {
3370 /* mark array as shutdown cleanly */
3372 md_update_sb(mddev
, 1);
3375 set_disk_ro(disk
, 1);
3376 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3380 * Free resources if final stop
3384 struct list_head
*tmp
;
3386 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3388 bitmap_destroy(mddev
);
3389 if (mddev
->bitmap_file
) {
3390 restore_bitmap_write_access(mddev
->bitmap_file
);
3391 fput(mddev
->bitmap_file
);
3392 mddev
->bitmap_file
= NULL
;
3394 mddev
->bitmap_offset
= 0;
3396 ITERATE_RDEV(mddev
,rdev
,tmp
)
3397 if (rdev
->raid_disk
>= 0) {
3399 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3400 sysfs_remove_link(&mddev
->kobj
, nm
);
3403 /* make sure all delayed_delete calls have finished */
3404 flush_scheduled_work();
3406 export_array(mddev
);
3408 mddev
->array_size
= 0;
3410 mddev
->raid_disks
= 0;
3411 mddev
->recovery_cp
= 0;
3413 } else if (mddev
->pers
)
3414 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3417 md_new_event(mddev
);
3423 static void autorun_array(mddev_t
*mddev
)
3426 struct list_head
*tmp
;
3429 if (list_empty(&mddev
->disks
))
3432 printk(KERN_INFO
"md: running: ");
3434 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3435 char b
[BDEVNAME_SIZE
];
3436 printk("<%s>", bdevname(rdev
->bdev
,b
));
3440 err
= do_md_run (mddev
);
3442 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3443 do_md_stop (mddev
, 0);
3448 * lets try to run arrays based on all disks that have arrived
3449 * until now. (those are in pending_raid_disks)
3451 * the method: pick the first pending disk, collect all disks with
3452 * the same UUID, remove all from the pending list and put them into
3453 * the 'same_array' list. Then order this list based on superblock
3454 * update time (freshest comes first), kick out 'old' disks and
3455 * compare superblocks. If everything's fine then run it.
3457 * If "unit" is allocated, then bump its reference count
3459 static void autorun_devices(int part
)
3461 struct list_head
*tmp
;
3462 mdk_rdev_t
*rdev0
, *rdev
;
3464 char b
[BDEVNAME_SIZE
];
3466 printk(KERN_INFO
"md: autorun ...\n");
3467 while (!list_empty(&pending_raid_disks
)) {
3470 LIST_HEAD(candidates
);
3471 rdev0
= list_entry(pending_raid_disks
.next
,
3472 mdk_rdev_t
, same_set
);
3474 printk(KERN_INFO
"md: considering %s ...\n",
3475 bdevname(rdev0
->bdev
,b
));
3476 INIT_LIST_HEAD(&candidates
);
3477 ITERATE_RDEV_PENDING(rdev
,tmp
)
3478 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3479 printk(KERN_INFO
"md: adding %s ...\n",
3480 bdevname(rdev
->bdev
,b
));
3481 list_move(&rdev
->same_set
, &candidates
);
3484 * now we have a set of devices, with all of them having
3485 * mostly sane superblocks. It's time to allocate the
3489 dev
= MKDEV(mdp_major
,
3490 rdev0
->preferred_minor
<< MdpMinorShift
);
3491 unit
= MINOR(dev
) >> MdpMinorShift
;
3493 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3496 if (rdev0
->preferred_minor
!= unit
) {
3497 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3498 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3502 md_probe(dev
, NULL
, NULL
);
3503 mddev
= mddev_find(dev
);
3506 "md: cannot allocate memory for md drive.\n");
3509 if (mddev_lock(mddev
))
3510 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3512 else if (mddev
->raid_disks
|| mddev
->major_version
3513 || !list_empty(&mddev
->disks
)) {
3515 "md: %s already running, cannot run %s\n",
3516 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3517 mddev_unlock(mddev
);
3519 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3520 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3521 list_del_init(&rdev
->same_set
);
3522 if (bind_rdev_to_array(rdev
, mddev
))
3525 autorun_array(mddev
);
3526 mddev_unlock(mddev
);
3528 /* on success, candidates will be empty, on error
3531 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3535 printk(KERN_INFO
"md: ... autorun DONE.\n");
3537 #endif /* !MODULE */
3539 static int get_version(void __user
* arg
)
3543 ver
.major
= MD_MAJOR_VERSION
;
3544 ver
.minor
= MD_MINOR_VERSION
;
3545 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3547 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3553 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3555 mdu_array_info_t info
;
3556 int nr
,working
,active
,failed
,spare
;
3558 struct list_head
*tmp
;
3560 nr
=working
=active
=failed
=spare
=0;
3561 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3563 if (test_bit(Faulty
, &rdev
->flags
))
3567 if (test_bit(In_sync
, &rdev
->flags
))
3574 info
.major_version
= mddev
->major_version
;
3575 info
.minor_version
= mddev
->minor_version
;
3576 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3577 info
.ctime
= mddev
->ctime
;
3578 info
.level
= mddev
->level
;
3579 info
.size
= mddev
->size
;
3580 if (info
.size
!= mddev
->size
) /* overflow */
3583 info
.raid_disks
= mddev
->raid_disks
;
3584 info
.md_minor
= mddev
->md_minor
;
3585 info
.not_persistent
= !mddev
->persistent
;
3587 info
.utime
= mddev
->utime
;
3590 info
.state
= (1<<MD_SB_CLEAN
);
3591 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3592 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3593 info
.active_disks
= active
;
3594 info
.working_disks
= working
;
3595 info
.failed_disks
= failed
;
3596 info
.spare_disks
= spare
;
3598 info
.layout
= mddev
->layout
;
3599 info
.chunk_size
= mddev
->chunk_size
;
3601 if (copy_to_user(arg
, &info
, sizeof(info
)))
3607 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3609 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3610 char *ptr
, *buf
= NULL
;
3613 md_allow_write(mddev
);
3615 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3619 /* bitmap disabled, zero the first byte and copy out */
3620 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3621 file
->pathname
[0] = '\0';
3625 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3629 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3633 strcpy(file
->pathname
, ptr
);
3637 if (copy_to_user(arg
, file
, sizeof(*file
)))
3645 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3647 mdu_disk_info_t info
;
3651 if (copy_from_user(&info
, arg
, sizeof(info
)))
3656 rdev
= find_rdev_nr(mddev
, nr
);
3658 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3659 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3660 info
.raid_disk
= rdev
->raid_disk
;
3662 if (test_bit(Faulty
, &rdev
->flags
))
3663 info
.state
|= (1<<MD_DISK_FAULTY
);
3664 else if (test_bit(In_sync
, &rdev
->flags
)) {
3665 info
.state
|= (1<<MD_DISK_ACTIVE
);
3666 info
.state
|= (1<<MD_DISK_SYNC
);
3668 if (test_bit(WriteMostly
, &rdev
->flags
))
3669 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3671 info
.major
= info
.minor
= 0;
3672 info
.raid_disk
= -1;
3673 info
.state
= (1<<MD_DISK_REMOVED
);
3676 if (copy_to_user(arg
, &info
, sizeof(info
)))
3682 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3684 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3686 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3688 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3691 if (!mddev
->raid_disks
) {
3693 /* expecting a device which has a superblock */
3694 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3697 "md: md_import_device returned %ld\n",
3699 return PTR_ERR(rdev
);
3701 if (!list_empty(&mddev
->disks
)) {
3702 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3703 mdk_rdev_t
, same_set
);
3704 int err
= super_types
[mddev
->major_version
]
3705 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3708 "md: %s has different UUID to %s\n",
3709 bdevname(rdev
->bdev
,b
),
3710 bdevname(rdev0
->bdev
,b2
));
3715 err
= bind_rdev_to_array(rdev
, mddev
);
3722 * add_new_disk can be used once the array is assembled
3723 * to add "hot spares". They must already have a superblock
3728 if (!mddev
->pers
->hot_add_disk
) {
3730 "%s: personality does not support diskops!\n",
3734 if (mddev
->persistent
)
3735 rdev
= md_import_device(dev
, mddev
->major_version
,
3736 mddev
->minor_version
);
3738 rdev
= md_import_device(dev
, -1, -1);
3741 "md: md_import_device returned %ld\n",
3743 return PTR_ERR(rdev
);
3745 /* set save_raid_disk if appropriate */
3746 if (!mddev
->persistent
) {
3747 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3748 info
->raid_disk
< mddev
->raid_disks
)
3749 rdev
->raid_disk
= info
->raid_disk
;
3751 rdev
->raid_disk
= -1;
3753 super_types
[mddev
->major_version
].
3754 validate_super(mddev
, rdev
);
3755 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3757 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3758 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3759 set_bit(WriteMostly
, &rdev
->flags
);
3761 rdev
->raid_disk
= -1;
3762 err
= bind_rdev_to_array(rdev
, mddev
);
3763 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3764 /* If there is hot_add_disk but no hot_remove_disk
3765 * then added disks for geometry changes,
3766 * and should be added immediately.
3768 super_types
[mddev
->major_version
].
3769 validate_super(mddev
, rdev
);
3770 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3772 unbind_rdev_from_array(rdev
);
3777 md_update_sb(mddev
, 1);
3778 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3779 md_wakeup_thread(mddev
->thread
);
3783 /* otherwise, add_new_disk is only allowed
3784 * for major_version==0 superblocks
3786 if (mddev
->major_version
!= 0) {
3787 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3792 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3794 rdev
= md_import_device (dev
, -1, 0);
3797 "md: error, md_import_device() returned %ld\n",
3799 return PTR_ERR(rdev
);
3801 rdev
->desc_nr
= info
->number
;
3802 if (info
->raid_disk
< mddev
->raid_disks
)
3803 rdev
->raid_disk
= info
->raid_disk
;
3805 rdev
->raid_disk
= -1;
3809 if (rdev
->raid_disk
< mddev
->raid_disks
)
3810 if (info
->state
& (1<<MD_DISK_SYNC
))
3811 set_bit(In_sync
, &rdev
->flags
);
3813 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3814 set_bit(WriteMostly
, &rdev
->flags
);
3816 if (!mddev
->persistent
) {
3817 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3818 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3820 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3821 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3823 err
= bind_rdev_to_array(rdev
, mddev
);
3833 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3835 char b
[BDEVNAME_SIZE
];
3841 rdev
= find_rdev(mddev
, dev
);
3845 if (rdev
->raid_disk
>= 0)
3848 kick_rdev_from_array(rdev
);
3849 md_update_sb(mddev
, 1);
3850 md_new_event(mddev
);
3854 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3855 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3859 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3861 char b
[BDEVNAME_SIZE
];
3869 if (mddev
->major_version
!= 0) {
3870 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3871 " version-0 superblocks.\n",
3875 if (!mddev
->pers
->hot_add_disk
) {
3877 "%s: personality does not support diskops!\n",
3882 rdev
= md_import_device (dev
, -1, 0);
3885 "md: error, md_import_device() returned %ld\n",
3890 if (mddev
->persistent
)
3891 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3894 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3896 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3899 if (test_bit(Faulty
, &rdev
->flags
)) {
3901 "md: can not hot-add faulty %s disk to %s!\n",
3902 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3906 clear_bit(In_sync
, &rdev
->flags
);
3908 rdev
->saved_raid_disk
= -1;
3909 err
= bind_rdev_to_array(rdev
, mddev
);
3914 * The rest should better be atomic, we can have disk failures
3915 * noticed in interrupt contexts ...
3918 if (rdev
->desc_nr
== mddev
->max_disks
) {
3919 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3922 goto abort_unbind_export
;
3925 rdev
->raid_disk
= -1;
3927 md_update_sb(mddev
, 1);
3930 * Kick recovery, maybe this spare has to be added to the
3931 * array immediately.
3933 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3934 md_wakeup_thread(mddev
->thread
);
3935 md_new_event(mddev
);
3938 abort_unbind_export
:
3939 unbind_rdev_from_array(rdev
);
3946 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3951 if (!mddev
->pers
->quiesce
)
3953 if (mddev
->recovery
|| mddev
->sync_thread
)
3955 /* we should be able to change the bitmap.. */
3961 return -EEXIST
; /* cannot add when bitmap is present */
3962 mddev
->bitmap_file
= fget(fd
);
3964 if (mddev
->bitmap_file
== NULL
) {
3965 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3970 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3972 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3974 fput(mddev
->bitmap_file
);
3975 mddev
->bitmap_file
= NULL
;
3978 mddev
->bitmap_offset
= 0; /* file overrides offset */
3979 } else if (mddev
->bitmap
== NULL
)
3980 return -ENOENT
; /* cannot remove what isn't there */
3983 mddev
->pers
->quiesce(mddev
, 1);
3985 err
= bitmap_create(mddev
);
3986 if (fd
< 0 || err
) {
3987 bitmap_destroy(mddev
);
3988 fd
= -1; /* make sure to put the file */
3990 mddev
->pers
->quiesce(mddev
, 0);
3993 if (mddev
->bitmap_file
) {
3994 restore_bitmap_write_access(mddev
->bitmap_file
);
3995 fput(mddev
->bitmap_file
);
3997 mddev
->bitmap_file
= NULL
;
4004 * set_array_info is used two different ways
4005 * The original usage is when creating a new array.
4006 * In this usage, raid_disks is > 0 and it together with
4007 * level, size, not_persistent,layout,chunksize determine the
4008 * shape of the array.
4009 * This will always create an array with a type-0.90.0 superblock.
4010 * The newer usage is when assembling an array.
4011 * In this case raid_disks will be 0, and the major_version field is
4012 * use to determine which style super-blocks are to be found on the devices.
4013 * The minor and patch _version numbers are also kept incase the
4014 * super_block handler wishes to interpret them.
4016 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
4019 if (info
->raid_disks
== 0) {
4020 /* just setting version number for superblock loading */
4021 if (info
->major_version
< 0 ||
4022 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
4023 super_types
[info
->major_version
].name
== NULL
) {
4024 /* maybe try to auto-load a module? */
4026 "md: superblock version %d not known\n",
4027 info
->major_version
);
4030 mddev
->major_version
= info
->major_version
;
4031 mddev
->minor_version
= info
->minor_version
;
4032 mddev
->patch_version
= info
->patch_version
;
4033 mddev
->persistent
= !info
->not_persistent
;
4036 mddev
->major_version
= MD_MAJOR_VERSION
;
4037 mddev
->minor_version
= MD_MINOR_VERSION
;
4038 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
4039 mddev
->ctime
= get_seconds();
4041 mddev
->level
= info
->level
;
4042 mddev
->clevel
[0] = 0;
4043 mddev
->size
= info
->size
;
4044 mddev
->raid_disks
= info
->raid_disks
;
4045 /* don't set md_minor, it is determined by which /dev/md* was
4048 if (info
->state
& (1<<MD_SB_CLEAN
))
4049 mddev
->recovery_cp
= MaxSector
;
4051 mddev
->recovery_cp
= 0;
4052 mddev
->persistent
= ! info
->not_persistent
;
4054 mddev
->layout
= info
->layout
;
4055 mddev
->chunk_size
= info
->chunk_size
;
4057 mddev
->max_disks
= MD_SB_DISKS
;
4060 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4062 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4063 mddev
->bitmap_offset
= 0;
4065 mddev
->reshape_position
= MaxSector
;
4068 * Generate a 128 bit UUID
4070 get_random_bytes(mddev
->uuid
, 16);
4072 mddev
->new_level
= mddev
->level
;
4073 mddev
->new_chunk
= mddev
->chunk_size
;
4074 mddev
->new_layout
= mddev
->layout
;
4075 mddev
->delta_disks
= 0;
4080 static int update_size(mddev_t
*mddev
, unsigned long size
)
4084 struct list_head
*tmp
;
4085 int fit
= (size
== 0);
4087 if (mddev
->pers
->resize
== NULL
)
4089 /* The "size" is the amount of each device that is used.
4090 * This can only make sense for arrays with redundancy.
4091 * linear and raid0 always use whatever space is available
4092 * We can only consider changing the size if no resync
4093 * or reconstruction is happening, and if the new size
4094 * is acceptable. It must fit before the sb_offset or,
4095 * if that is <data_offset, it must fit before the
4096 * size of each device.
4097 * If size is zero, we find the largest size that fits.
4099 if (mddev
->sync_thread
)
4101 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4103 avail
= rdev
->size
* 2;
4105 if (fit
&& (size
== 0 || size
> avail
/2))
4107 if (avail
< ((sector_t
)size
<< 1))
4110 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
4112 struct block_device
*bdev
;
4114 bdev
= bdget_disk(mddev
->gendisk
, 0);
4116 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4117 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
4118 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4125 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4128 /* change the number of raid disks */
4129 if (mddev
->pers
->check_reshape
== NULL
)
4131 if (raid_disks
<= 0 ||
4132 raid_disks
>= mddev
->max_disks
)
4134 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4136 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4138 rv
= mddev
->pers
->check_reshape(mddev
);
4144 * update_array_info is used to change the configuration of an
4146 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4147 * fields in the info are checked against the array.
4148 * Any differences that cannot be handled will cause an error.
4149 * Normally, only one change can be managed at a time.
4151 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4157 /* calculate expected state,ignoring low bits */
4158 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4159 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4161 if (mddev
->major_version
!= info
->major_version
||
4162 mddev
->minor_version
!= info
->minor_version
||
4163 /* mddev->patch_version != info->patch_version || */
4164 mddev
->ctime
!= info
->ctime
||
4165 mddev
->level
!= info
->level
||
4166 /* mddev->layout != info->layout || */
4167 !mddev
->persistent
!= info
->not_persistent
||
4168 mddev
->chunk_size
!= info
->chunk_size
||
4169 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4170 ((state
^info
->state
) & 0xfffffe00)
4173 /* Check there is only one change */
4174 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4175 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4176 if (mddev
->layout
!= info
->layout
) cnt
++;
4177 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4178 if (cnt
== 0) return 0;
4179 if (cnt
> 1) return -EINVAL
;
4181 if (mddev
->layout
!= info
->layout
) {
4183 * we don't need to do anything at the md level, the
4184 * personality will take care of it all.
4186 if (mddev
->pers
->reconfig
== NULL
)
4189 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4191 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4192 rv
= update_size(mddev
, info
->size
);
4194 if (mddev
->raid_disks
!= info
->raid_disks
)
4195 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4197 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4198 if (mddev
->pers
->quiesce
== NULL
)
4200 if (mddev
->recovery
|| mddev
->sync_thread
)
4202 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4203 /* add the bitmap */
4206 if (mddev
->default_bitmap_offset
== 0)
4208 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4209 mddev
->pers
->quiesce(mddev
, 1);
4210 rv
= bitmap_create(mddev
);
4212 bitmap_destroy(mddev
);
4213 mddev
->pers
->quiesce(mddev
, 0);
4215 /* remove the bitmap */
4218 if (mddev
->bitmap
->file
)
4220 mddev
->pers
->quiesce(mddev
, 1);
4221 bitmap_destroy(mddev
);
4222 mddev
->pers
->quiesce(mddev
, 0);
4223 mddev
->bitmap_offset
= 0;
4226 md_update_sb(mddev
, 1);
4230 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4234 if (mddev
->pers
== NULL
)
4237 rdev
= find_rdev(mddev
, dev
);
4241 md_error(mddev
, rdev
);
4245 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4247 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4251 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4255 static int md_ioctl(struct inode
*inode
, struct file
*file
,
4256 unsigned int cmd
, unsigned long arg
)
4259 void __user
*argp
= (void __user
*)arg
;
4260 mddev_t
*mddev
= NULL
;
4262 if (!capable(CAP_SYS_ADMIN
))
4266 * Commands dealing with the RAID driver but not any
4272 err
= get_version(argp
);
4275 case PRINT_RAID_DEBUG
:
4283 autostart_arrays(arg
);
4290 * Commands creating/starting a new array:
4293 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4300 err
= mddev_lock(mddev
);
4303 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4310 case SET_ARRAY_INFO
:
4312 mdu_array_info_t info
;
4314 memset(&info
, 0, sizeof(info
));
4315 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4320 err
= update_array_info(mddev
, &info
);
4322 printk(KERN_WARNING
"md: couldn't update"
4323 " array info. %d\n", err
);
4328 if (!list_empty(&mddev
->disks
)) {
4330 "md: array %s already has disks!\n",
4335 if (mddev
->raid_disks
) {
4337 "md: array %s already initialised!\n",
4342 err
= set_array_info(mddev
, &info
);
4344 printk(KERN_WARNING
"md: couldn't set"
4345 " array info. %d\n", err
);
4355 * Commands querying/configuring an existing array:
4357 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4358 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4359 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4360 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
4361 && cmd
!= GET_BITMAP_FILE
) {
4367 * Commands even a read-only array can execute:
4371 case GET_ARRAY_INFO
:
4372 err
= get_array_info(mddev
, argp
);
4375 case GET_BITMAP_FILE
:
4376 err
= get_bitmap_file(mddev
, argp
);
4380 err
= get_disk_info(mddev
, argp
);
4383 case RESTART_ARRAY_RW
:
4384 err
= restart_array(mddev
);
4388 err
= do_md_stop (mddev
, 0);
4392 err
= do_md_stop (mddev
, 1);
4396 * We have a problem here : there is no easy way to give a CHS
4397 * virtual geometry. We currently pretend that we have a 2 heads
4398 * 4 sectors (with a BIG number of cylinders...). This drives
4399 * dosfs just mad... ;-)
4404 * The remaining ioctls are changing the state of the
4405 * superblock, so we do not allow them on read-only arrays.
4406 * However non-MD ioctls (e.g. get-size) will still come through
4407 * here and hit the 'default' below, so only disallow
4408 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4410 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4411 mddev
->ro
&& mddev
->pers
) {
4412 if (mddev
->ro
== 2) {
4414 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4415 md_wakeup_thread(mddev
->thread
);
4427 mdu_disk_info_t info
;
4428 if (copy_from_user(&info
, argp
, sizeof(info
)))
4431 err
= add_new_disk(mddev
, &info
);
4435 case HOT_REMOVE_DISK
:
4436 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4440 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4443 case SET_DISK_FAULTY
:
4444 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4448 err
= do_md_run (mddev
);
4451 case SET_BITMAP_FILE
:
4452 err
= set_bitmap_file(mddev
, (int)arg
);
4462 mddev_unlock(mddev
);
4472 static int md_open(struct inode
*inode
, struct file
*file
)
4475 * Succeed if we can lock the mddev, which confirms that
4476 * it isn't being stopped right now.
4478 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4481 if ((err
= mutex_lock_interruptible_nested(&mddev
->reconfig_mutex
, 1)))
4486 mddev_unlock(mddev
);
4488 check_disk_change(inode
->i_bdev
);
4493 static int md_release(struct inode
*inode
, struct file
* file
)
4495 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4503 static int md_media_changed(struct gendisk
*disk
)
4505 mddev_t
*mddev
= disk
->private_data
;
4507 return mddev
->changed
;
4510 static int md_revalidate(struct gendisk
*disk
)
4512 mddev_t
*mddev
= disk
->private_data
;
4517 static struct block_device_operations md_fops
=
4519 .owner
= THIS_MODULE
,
4521 .release
= md_release
,
4523 .getgeo
= md_getgeo
,
4524 .media_changed
= md_media_changed
,
4525 .revalidate_disk
= md_revalidate
,
4528 static int md_thread(void * arg
)
4530 mdk_thread_t
*thread
= arg
;
4533 * md_thread is a 'system-thread', it's priority should be very
4534 * high. We avoid resource deadlocks individually in each
4535 * raid personality. (RAID5 does preallocation) We also use RR and
4536 * the very same RT priority as kswapd, thus we will never get
4537 * into a priority inversion deadlock.
4539 * we definitely have to have equal or higher priority than
4540 * bdflush, otherwise bdflush will deadlock if there are too
4541 * many dirty RAID5 blocks.
4544 current
->flags
|= PF_NOFREEZE
;
4545 allow_signal(SIGKILL
);
4546 while (!kthread_should_stop()) {
4548 /* We need to wait INTERRUPTIBLE so that
4549 * we don't add to the load-average.
4550 * That means we need to be sure no signals are
4553 if (signal_pending(current
))
4554 flush_signals(current
);
4556 wait_event_interruptible_timeout
4558 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4559 || kthread_should_stop(),
4562 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4564 thread
->run(thread
->mddev
);
4570 void md_wakeup_thread(mdk_thread_t
*thread
)
4573 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4574 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4575 wake_up(&thread
->wqueue
);
4579 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4582 mdk_thread_t
*thread
;
4584 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4588 init_waitqueue_head(&thread
->wqueue
);
4591 thread
->mddev
= mddev
;
4592 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4593 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4594 if (IS_ERR(thread
->tsk
)) {
4601 void md_unregister_thread(mdk_thread_t
*thread
)
4603 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4605 kthread_stop(thread
->tsk
);
4609 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4616 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4619 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4621 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4622 __builtin_return_address(0),__builtin_return_address(1),
4623 __builtin_return_address(2),__builtin_return_address(3));
4627 if (!mddev
->pers
->error_handler
)
4629 mddev
->pers
->error_handler(mddev
,rdev
);
4630 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4631 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4632 md_wakeup_thread(mddev
->thread
);
4633 md_new_event_inintr(mddev
);
4636 /* seq_file implementation /proc/mdstat */
4638 static void status_unused(struct seq_file
*seq
)
4642 struct list_head
*tmp
;
4644 seq_printf(seq
, "unused devices: ");
4646 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4647 char b
[BDEVNAME_SIZE
];
4649 seq_printf(seq
, "%s ",
4650 bdevname(rdev
->bdev
,b
));
4653 seq_printf(seq
, "<none>");
4655 seq_printf(seq
, "\n");
4659 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4661 sector_t max_blocks
, resync
, res
;
4662 unsigned long dt
, db
, rt
;
4664 unsigned int per_milli
;
4666 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4668 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4669 max_blocks
= mddev
->resync_max_sectors
>> 1;
4671 max_blocks
= mddev
->size
;
4674 * Should not happen.
4680 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4681 * in a sector_t, and (max_blocks>>scale) will fit in a
4682 * u32, as those are the requirements for sector_div.
4683 * Thus 'scale' must be at least 10
4686 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4687 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4690 res
= (resync
>>scale
)*1000;
4691 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4695 int i
, x
= per_milli
/50, y
= 20-x
;
4696 seq_printf(seq
, "[");
4697 for (i
= 0; i
< x
; i
++)
4698 seq_printf(seq
, "=");
4699 seq_printf(seq
, ">");
4700 for (i
= 0; i
< y
; i
++)
4701 seq_printf(seq
, ".");
4702 seq_printf(seq
, "] ");
4704 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4705 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4707 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
4709 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4710 "resync" : "recovery"))),
4711 per_milli
/10, per_milli
% 10,
4712 (unsigned long long) resync
,
4713 (unsigned long long) max_blocks
);
4716 * We do not want to overflow, so the order of operands and
4717 * the * 100 / 100 trick are important. We do a +1 to be
4718 * safe against division by zero. We only estimate anyway.
4720 * dt: time from mark until now
4721 * db: blocks written from mark until now
4722 * rt: remaining time
4724 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4726 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
4727 - mddev
->resync_mark_cnt
;
4728 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/2/100+1)))/100;
4730 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4732 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
4735 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4737 struct list_head
*tmp
;
4747 spin_lock(&all_mddevs_lock
);
4748 list_for_each(tmp
,&all_mddevs
)
4750 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4752 spin_unlock(&all_mddevs_lock
);
4755 spin_unlock(&all_mddevs_lock
);
4757 return (void*)2;/* tail */
4761 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4763 struct list_head
*tmp
;
4764 mddev_t
*next_mddev
, *mddev
= v
;
4770 spin_lock(&all_mddevs_lock
);
4772 tmp
= all_mddevs
.next
;
4774 tmp
= mddev
->all_mddevs
.next
;
4775 if (tmp
!= &all_mddevs
)
4776 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4778 next_mddev
= (void*)2;
4781 spin_unlock(&all_mddevs_lock
);
4789 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4793 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4797 struct mdstat_info
{
4801 static int md_seq_show(struct seq_file
*seq
, void *v
)
4805 struct list_head
*tmp2
;
4807 struct mdstat_info
*mi
= seq
->private;
4808 struct bitmap
*bitmap
;
4810 if (v
== (void*)1) {
4811 struct mdk_personality
*pers
;
4812 seq_printf(seq
, "Personalities : ");
4813 spin_lock(&pers_lock
);
4814 list_for_each_entry(pers
, &pers_list
, list
)
4815 seq_printf(seq
, "[%s] ", pers
->name
);
4817 spin_unlock(&pers_lock
);
4818 seq_printf(seq
, "\n");
4819 mi
->event
= atomic_read(&md_event_count
);
4822 if (v
== (void*)2) {
4827 if (mddev_lock(mddev
) < 0)
4830 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4831 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4832 mddev
->pers
? "" : "in");
4835 seq_printf(seq
, " (read-only)");
4837 seq_printf(seq
, "(auto-read-only)");
4838 seq_printf(seq
, " %s", mddev
->pers
->name
);
4842 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4843 char b
[BDEVNAME_SIZE
];
4844 seq_printf(seq
, " %s[%d]",
4845 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4846 if (test_bit(WriteMostly
, &rdev
->flags
))
4847 seq_printf(seq
, "(W)");
4848 if (test_bit(Faulty
, &rdev
->flags
)) {
4849 seq_printf(seq
, "(F)");
4851 } else if (rdev
->raid_disk
< 0)
4852 seq_printf(seq
, "(S)"); /* spare */
4856 if (!list_empty(&mddev
->disks
)) {
4858 seq_printf(seq
, "\n %llu blocks",
4859 (unsigned long long)mddev
->array_size
);
4861 seq_printf(seq
, "\n %llu blocks",
4862 (unsigned long long)size
);
4864 if (mddev
->persistent
) {
4865 if (mddev
->major_version
!= 0 ||
4866 mddev
->minor_version
!= 90) {
4867 seq_printf(seq
," super %d.%d",
4868 mddev
->major_version
,
4869 mddev
->minor_version
);
4872 seq_printf(seq
, " super non-persistent");
4875 mddev
->pers
->status (seq
, mddev
);
4876 seq_printf(seq
, "\n ");
4877 if (mddev
->pers
->sync_request
) {
4878 if (mddev
->curr_resync
> 2) {
4879 status_resync (seq
, mddev
);
4880 seq_printf(seq
, "\n ");
4881 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4882 seq_printf(seq
, "\tresync=DELAYED\n ");
4883 else if (mddev
->recovery_cp
< MaxSector
)
4884 seq_printf(seq
, "\tresync=PENDING\n ");
4887 seq_printf(seq
, "\n ");
4889 if ((bitmap
= mddev
->bitmap
)) {
4890 unsigned long chunk_kb
;
4891 unsigned long flags
;
4892 spin_lock_irqsave(&bitmap
->lock
, flags
);
4893 chunk_kb
= bitmap
->chunksize
>> 10;
4894 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4896 bitmap
->pages
- bitmap
->missing_pages
,
4898 (bitmap
->pages
- bitmap
->missing_pages
)
4899 << (PAGE_SHIFT
- 10),
4900 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4901 chunk_kb
? "KB" : "B");
4903 seq_printf(seq
, ", file: ");
4904 seq_path(seq
, bitmap
->file
->f_path
.mnt
,
4905 bitmap
->file
->f_path
.dentry
," \t\n");
4908 seq_printf(seq
, "\n");
4909 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4912 seq_printf(seq
, "\n");
4914 mddev_unlock(mddev
);
4919 static struct seq_operations md_seq_ops
= {
4920 .start
= md_seq_start
,
4921 .next
= md_seq_next
,
4922 .stop
= md_seq_stop
,
4923 .show
= md_seq_show
,
4926 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4929 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4933 error
= seq_open(file
, &md_seq_ops
);
4937 struct seq_file
*p
= file
->private_data
;
4939 mi
->event
= atomic_read(&md_event_count
);
4944 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4946 struct seq_file
*m
= file
->private_data
;
4947 struct mdstat_info
*mi
= m
->private;
4950 return seq_release(inode
, file
);
4953 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4955 struct seq_file
*m
= filp
->private_data
;
4956 struct mdstat_info
*mi
= m
->private;
4959 poll_wait(filp
, &md_event_waiters
, wait
);
4961 /* always allow read */
4962 mask
= POLLIN
| POLLRDNORM
;
4964 if (mi
->event
!= atomic_read(&md_event_count
))
4965 mask
|= POLLERR
| POLLPRI
;
4969 static const struct file_operations md_seq_fops
= {
4970 .owner
= THIS_MODULE
,
4971 .open
= md_seq_open
,
4973 .llseek
= seq_lseek
,
4974 .release
= md_seq_release
,
4975 .poll
= mdstat_poll
,
4978 int register_md_personality(struct mdk_personality
*p
)
4980 spin_lock(&pers_lock
);
4981 list_add_tail(&p
->list
, &pers_list
);
4982 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4983 spin_unlock(&pers_lock
);
4987 int unregister_md_personality(struct mdk_personality
*p
)
4989 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4990 spin_lock(&pers_lock
);
4991 list_del_init(&p
->list
);
4992 spin_unlock(&pers_lock
);
4996 static int is_mddev_idle(mddev_t
*mddev
)
4999 struct list_head
*tmp
;
5001 unsigned long curr_events
;
5004 ITERATE_RDEV(mddev
,rdev
,tmp
) {
5005 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
5006 curr_events
= disk_stat_read(disk
, sectors
[0]) +
5007 disk_stat_read(disk
, sectors
[1]) -
5008 atomic_read(&disk
->sync_io
);
5009 /* The difference between curr_events and last_events
5010 * will be affected by any new non-sync IO (making
5011 * curr_events bigger) and any difference in the amount of
5012 * in-flight syncio (making current_events bigger or smaller)
5013 * The amount in-flight is currently limited to
5014 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5015 * which is at most 4096 sectors.
5016 * These numbers are fairly fragile and should be made
5017 * more robust, probably by enforcing the
5018 * 'window size' that md_do_sync sort-of uses.
5020 * Note: the following is an unsigned comparison.
5022 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
5023 rdev
->last_events
= curr_events
;
5030 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
5032 /* another "blocks" (512byte) blocks have been synced */
5033 atomic_sub(blocks
, &mddev
->recovery_active
);
5034 wake_up(&mddev
->recovery_wait
);
5036 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5037 md_wakeup_thread(mddev
->thread
);
5038 // stop recovery, signal do_sync ....
5043 /* md_write_start(mddev, bi)
5044 * If we need to update some array metadata (e.g. 'active' flag
5045 * in superblock) before writing, schedule a superblock update
5046 * and wait for it to complete.
5048 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
5050 if (bio_data_dir(bi
) != WRITE
)
5053 BUG_ON(mddev
->ro
== 1);
5054 if (mddev
->ro
== 2) {
5055 /* need to switch to read/write */
5057 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5058 md_wakeup_thread(mddev
->thread
);
5060 atomic_inc(&mddev
->writes_pending
);
5061 if (mddev
->in_sync
) {
5062 spin_lock_irq(&mddev
->write_lock
);
5063 if (mddev
->in_sync
) {
5065 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5066 md_wakeup_thread(mddev
->thread
);
5068 spin_unlock_irq(&mddev
->write_lock
);
5070 wait_event(mddev
->sb_wait
, mddev
->flags
==0);
5073 void md_write_end(mddev_t
*mddev
)
5075 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5076 if (mddev
->safemode
== 2)
5077 md_wakeup_thread(mddev
->thread
);
5078 else if (mddev
->safemode_delay
)
5079 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5083 /* md_allow_write(mddev)
5084 * Calling this ensures that the array is marked 'active' so that writes
5085 * may proceed without blocking. It is important to call this before
5086 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5087 * Must be called with mddev_lock held.
5089 void md_allow_write(mddev_t
*mddev
)
5096 spin_lock_irq(&mddev
->write_lock
);
5097 if (mddev
->in_sync
) {
5099 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5100 if (mddev
->safemode_delay
&&
5101 mddev
->safemode
== 0)
5102 mddev
->safemode
= 1;
5103 spin_unlock_irq(&mddev
->write_lock
);
5104 md_update_sb(mddev
, 0);
5106 spin_unlock_irq(&mddev
->write_lock
);
5108 EXPORT_SYMBOL_GPL(md_allow_write
);
5110 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
5112 #define SYNC_MARKS 10
5113 #define SYNC_MARK_STEP (3*HZ)
5114 void md_do_sync(mddev_t
*mddev
)
5117 unsigned int currspeed
= 0,
5119 sector_t max_sectors
,j
, io_sectors
;
5120 unsigned long mark
[SYNC_MARKS
];
5121 sector_t mark_cnt
[SYNC_MARKS
];
5123 struct list_head
*tmp
;
5124 sector_t last_check
;
5126 struct list_head
*rtmp
;
5130 /* just incase thread restarts... */
5131 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5133 if (mddev
->ro
) /* never try to sync a read-only array */
5136 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5137 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
5138 desc
= "data-check";
5139 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5140 desc
= "requested-resync";
5143 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5148 /* we overload curr_resync somewhat here.
5149 * 0 == not engaged in resync at all
5150 * 2 == checking that there is no conflict with another sync
5151 * 1 == like 2, but have yielded to allow conflicting resync to
5153 * other == active in resync - this many blocks
5155 * Before starting a resync we must have set curr_resync to
5156 * 2, and then checked that every "conflicting" array has curr_resync
5157 * less than ours. When we find one that is the same or higher
5158 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5159 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5160 * This will mean we have to start checking from the beginning again.
5165 mddev
->curr_resync
= 2;
5168 if (kthread_should_stop()) {
5169 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5172 ITERATE_MDDEV(mddev2
,tmp
) {
5173 if (mddev2
== mddev
)
5175 if (mddev2
->curr_resync
&&
5176 match_mddev_units(mddev
,mddev2
)) {
5178 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5179 /* arbitrarily yield */
5180 mddev
->curr_resync
= 1;
5181 wake_up(&resync_wait
);
5183 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5184 /* no need to wait here, we can wait the next
5185 * time 'round when curr_resync == 2
5188 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
5189 if (!kthread_should_stop() &&
5190 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5191 printk(KERN_INFO
"md: delaying %s of %s"
5192 " until %s has finished (they"
5193 " share one or more physical units)\n",
5194 desc
, mdname(mddev
), mdname(mddev2
));
5197 finish_wait(&resync_wait
, &wq
);
5200 finish_wait(&resync_wait
, &wq
);
5203 } while (mddev
->curr_resync
< 2);
5206 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5207 /* resync follows the size requested by the personality,
5208 * which defaults to physical size, but can be virtual size
5210 max_sectors
= mddev
->resync_max_sectors
;
5211 mddev
->resync_mismatches
= 0;
5212 /* we don't use the checkpoint if there's a bitmap */
5213 if (!mddev
->bitmap
&&
5214 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5215 j
= mddev
->recovery_cp
;
5216 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5217 max_sectors
= mddev
->size
<< 1;
5219 /* recovery follows the physical size of devices */
5220 max_sectors
= mddev
->size
<< 1;
5222 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5223 if (rdev
->raid_disk
>= 0 &&
5224 !test_bit(Faulty
, &rdev
->flags
) &&
5225 !test_bit(In_sync
, &rdev
->flags
) &&
5226 rdev
->recovery_offset
< j
)
5227 j
= rdev
->recovery_offset
;
5230 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
5231 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
5232 " %d KB/sec/disk.\n", speed_min(mddev
));
5233 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5234 "(but not more than %d KB/sec) for %s.\n",
5235 speed_max(mddev
), desc
);
5237 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5240 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5242 mark_cnt
[m
] = io_sectors
;
5245 mddev
->resync_mark
= mark
[last_mark
];
5246 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5249 * Tune reconstruction:
5251 window
= 32*(PAGE_SIZE
/512);
5252 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5253 window
/2,(unsigned long long) max_sectors
/2);
5255 atomic_set(&mddev
->recovery_active
, 0);
5256 init_waitqueue_head(&mddev
->recovery_wait
);
5261 "md: resuming %s of %s from checkpoint.\n",
5262 desc
, mdname(mddev
));
5263 mddev
->curr_resync
= j
;
5266 while (j
< max_sectors
) {
5270 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5271 currspeed
< speed_min(mddev
));
5273 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5277 if (!skipped
) { /* actual IO requested */
5278 io_sectors
+= sectors
;
5279 atomic_add(sectors
, &mddev
->recovery_active
);
5283 if (j
>1) mddev
->curr_resync
= j
;
5284 mddev
->curr_mark_cnt
= io_sectors
;
5285 if (last_check
== 0)
5286 /* this is the earliers that rebuilt will be
5287 * visible in /proc/mdstat
5289 md_new_event(mddev
);
5291 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5294 last_check
= io_sectors
;
5296 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
5297 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
5301 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5303 int next
= (last_mark
+1) % SYNC_MARKS
;
5305 mddev
->resync_mark
= mark
[next
];
5306 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5307 mark
[next
] = jiffies
;
5308 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5313 if (kthread_should_stop()) {
5315 * got a signal, exit.
5318 "md: md_do_sync() got signal ... exiting\n");
5319 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5324 * this loop exits only if either when we are slower than
5325 * the 'hard' speed limit, or the system was IO-idle for
5327 * the system might be non-idle CPU-wise, but we only care
5328 * about not overloading the IO subsystem. (things like an
5329 * e2fsck being done on the RAID array should execute fast)
5331 mddev
->queue
->unplug_fn(mddev
->queue
);
5334 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5335 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5337 if (currspeed
> speed_min(mddev
)) {
5338 if ((currspeed
> speed_max(mddev
)) ||
5339 !is_mddev_idle(mddev
)) {
5345 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
5347 * this also signals 'finished resyncing' to md_stop
5350 mddev
->queue
->unplug_fn(mddev
->queue
);
5352 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5354 /* tell personality that we are finished */
5355 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5357 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5358 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5359 mddev
->curr_resync
> 2) {
5360 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5361 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5362 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5364 "md: checkpointing %s of %s.\n",
5365 desc
, mdname(mddev
));
5366 mddev
->recovery_cp
= mddev
->curr_resync
;
5369 mddev
->recovery_cp
= MaxSector
;
5371 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5372 mddev
->curr_resync
= MaxSector
;
5373 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5374 if (rdev
->raid_disk
>= 0 &&
5375 !test_bit(Faulty
, &rdev
->flags
) &&
5376 !test_bit(In_sync
, &rdev
->flags
) &&
5377 rdev
->recovery_offset
< mddev
->curr_resync
)
5378 rdev
->recovery_offset
= mddev
->curr_resync
;
5381 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5384 mddev
->curr_resync
= 0;
5385 wake_up(&resync_wait
);
5386 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5387 md_wakeup_thread(mddev
->thread
);
5389 EXPORT_SYMBOL_GPL(md_do_sync
);
5392 static int remove_and_add_spares(mddev_t
*mddev
)
5395 struct list_head
*rtmp
;
5398 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5399 if (rdev
->raid_disk
>= 0 &&
5400 (test_bit(Faulty
, &rdev
->flags
) ||
5401 ! test_bit(In_sync
, &rdev
->flags
)) &&
5402 atomic_read(&rdev
->nr_pending
)==0) {
5403 if (mddev
->pers
->hot_remove_disk(
5404 mddev
, rdev
->raid_disk
)==0) {
5406 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5407 sysfs_remove_link(&mddev
->kobj
, nm
);
5408 rdev
->raid_disk
= -1;
5412 if (mddev
->degraded
) {
5413 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5414 if (rdev
->raid_disk
< 0
5415 && !test_bit(Faulty
, &rdev
->flags
)) {
5416 rdev
->recovery_offset
= 0;
5417 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5419 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5420 if (sysfs_create_link(&mddev
->kobj
,
5423 "md: cannot register "
5427 md_new_event(mddev
);
5435 * This routine is regularly called by all per-raid-array threads to
5436 * deal with generic issues like resync and super-block update.
5437 * Raid personalities that don't have a thread (linear/raid0) do not
5438 * need this as they never do any recovery or update the superblock.
5440 * It does not do any resync itself, but rather "forks" off other threads
5441 * to do that as needed.
5442 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5443 * "->recovery" and create a thread at ->sync_thread.
5444 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5445 * and wakeups up this thread which will reap the thread and finish up.
5446 * This thread also removes any faulty devices (with nr_pending == 0).
5448 * The overall approach is:
5449 * 1/ if the superblock needs updating, update it.
5450 * 2/ If a recovery thread is running, don't do anything else.
5451 * 3/ If recovery has finished, clean up, possibly marking spares active.
5452 * 4/ If there are any faulty devices, remove them.
5453 * 5/ If array is degraded, try to add spares devices
5454 * 6/ If array has spares or is not in-sync, start a resync thread.
5456 void md_check_recovery(mddev_t
*mddev
)
5459 struct list_head
*rtmp
;
5463 bitmap_daemon_work(mddev
->bitmap
);
5468 if (signal_pending(current
)) {
5469 if (mddev
->pers
->sync_request
) {
5470 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5472 mddev
->safemode
= 2;
5474 flush_signals(current
);
5479 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5480 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5481 (mddev
->safemode
== 1) ||
5482 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5483 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5487 if (mddev_trylock(mddev
)) {
5490 spin_lock_irq(&mddev
->write_lock
);
5491 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5492 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5494 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5496 if (mddev
->safemode
== 1)
5497 mddev
->safemode
= 0;
5498 spin_unlock_irq(&mddev
->write_lock
);
5501 md_update_sb(mddev
, 0);
5504 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5505 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5506 /* resync/recovery still happening */
5507 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5510 if (mddev
->sync_thread
) {
5511 /* resync has finished, collect result */
5512 md_unregister_thread(mddev
->sync_thread
);
5513 mddev
->sync_thread
= NULL
;
5514 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5515 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5517 /* activate any spares */
5518 mddev
->pers
->spare_active(mddev
);
5520 md_update_sb(mddev
, 1);
5522 /* if array is no-longer degraded, then any saved_raid_disk
5523 * information must be scrapped
5525 if (!mddev
->degraded
)
5526 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5527 rdev
->saved_raid_disk
= -1;
5529 mddev
->recovery
= 0;
5530 /* flag recovery needed just to double check */
5531 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5532 md_new_event(mddev
);
5535 /* Clear some bits that don't mean anything, but
5538 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5539 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5540 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5541 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5543 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5545 /* no recovery is running.
5546 * remove any failed drives, then
5547 * add spares if possible.
5548 * Spare are also removed and re-added, to allow
5549 * the personality to fail the re-add.
5552 if (mddev
->reshape_position
!= MaxSector
) {
5553 if (mddev
->pers
->check_reshape(mddev
) != 0)
5554 /* Cannot proceed */
5556 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
5557 } else if ((spares
= remove_and_add_spares(mddev
))) {
5558 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5559 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5560 } else if (mddev
->recovery_cp
< MaxSector
) {
5561 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5562 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5563 /* nothing to be done ... */
5566 if (mddev
->pers
->sync_request
) {
5567 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5568 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5569 /* We are adding a device or devices to an array
5570 * which has the bitmap stored on all devices.
5571 * So make sure all bitmap pages get written
5573 bitmap_write_all(mddev
->bitmap
);
5575 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5578 if (!mddev
->sync_thread
) {
5579 printk(KERN_ERR
"%s: could not start resync"
5582 /* leave the spares where they are, it shouldn't hurt */
5583 mddev
->recovery
= 0;
5585 md_wakeup_thread(mddev
->sync_thread
);
5586 md_new_event(mddev
);
5589 mddev_unlock(mddev
);
5593 static int md_notify_reboot(struct notifier_block
*this,
5594 unsigned long code
, void *x
)
5596 struct list_head
*tmp
;
5599 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5601 printk(KERN_INFO
"md: stopping all md devices.\n");
5603 ITERATE_MDDEV(mddev
,tmp
)
5604 if (mddev_trylock(mddev
)) {
5605 do_md_stop (mddev
, 1);
5606 mddev_unlock(mddev
);
5609 * certain more exotic SCSI devices are known to be
5610 * volatile wrt too early system reboots. While the
5611 * right place to handle this issue is the given
5612 * driver, we do want to have a safe RAID driver ...
5619 static struct notifier_block md_notifier
= {
5620 .notifier_call
= md_notify_reboot
,
5622 .priority
= INT_MAX
, /* before any real devices */
5625 static void md_geninit(void)
5627 struct proc_dir_entry
*p
;
5629 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5631 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5633 p
->proc_fops
= &md_seq_fops
;
5636 static int __init
md_init(void)
5638 if (register_blkdev(MAJOR_NR
, "md"))
5640 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5641 unregister_blkdev(MAJOR_NR
, "md");
5644 blk_register_region(MKDEV(MAJOR_NR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
5645 md_probe
, NULL
, NULL
);
5646 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
5647 md_probe
, NULL
, NULL
);
5649 register_reboot_notifier(&md_notifier
);
5650 raid_table_header
= register_sysctl_table(raid_root_table
);
5660 * Searches all registered partitions for autorun RAID arrays
5663 static dev_t detected_devices
[128];
5666 void md_autodetect_dev(dev_t dev
)
5668 if (dev_cnt
>= 0 && dev_cnt
< 127)
5669 detected_devices
[dev_cnt
++] = dev
;
5673 static void autostart_arrays(int part
)
5678 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5680 for (i
= 0; i
< dev_cnt
; i
++) {
5681 dev_t dev
= detected_devices
[i
];
5683 rdev
= md_import_device(dev
,0, 0);
5687 if (test_bit(Faulty
, &rdev
->flags
)) {
5691 list_add(&rdev
->same_set
, &pending_raid_disks
);
5695 autorun_devices(part
);
5698 #endif /* !MODULE */
5700 static __exit
void md_exit(void)
5703 struct list_head
*tmp
;
5705 blk_unregister_region(MKDEV(MAJOR_NR
,0), 1U << MINORBITS
);
5706 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
5708 unregister_blkdev(MAJOR_NR
,"md");
5709 unregister_blkdev(mdp_major
, "mdp");
5710 unregister_reboot_notifier(&md_notifier
);
5711 unregister_sysctl_table(raid_table_header
);
5712 remove_proc_entry("mdstat", NULL
);
5713 ITERATE_MDDEV(mddev
,tmp
) {
5714 struct gendisk
*disk
= mddev
->gendisk
;
5717 export_array(mddev
);
5720 mddev
->gendisk
= NULL
;
5725 module_init(md_init
)
5726 module_exit(md_exit
)
5728 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5730 return sprintf(buffer
, "%d", start_readonly
);
5732 static int set_ro(const char *val
, struct kernel_param
*kp
)
5735 int num
= simple_strtoul(val
, &e
, 10);
5736 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5737 start_readonly
= num
;
5743 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
5744 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
5747 EXPORT_SYMBOL(register_md_personality
);
5748 EXPORT_SYMBOL(unregister_md_personality
);
5749 EXPORT_SYMBOL(md_error
);
5750 EXPORT_SYMBOL(md_done_sync
);
5751 EXPORT_SYMBOL(md_write_start
);
5752 EXPORT_SYMBOL(md_write_end
);
5753 EXPORT_SYMBOL(md_register_thread
);
5754 EXPORT_SYMBOL(md_unregister_thread
);
5755 EXPORT_SYMBOL(md_wakeup_thread
);
5756 EXPORT_SYMBOL(md_check_recovery
);
5757 MODULE_LICENSE("GPL");
5759 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);