2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/ctype.h>
49 #include <linux/init.h>
51 #include <linux/file.h>
54 #include <linux/kmod.h>
57 #include <asm/unaligned.h>
59 #define MAJOR_NR MD_MAJOR
62 /* 63 partitions with the alternate major number (mdp) */
63 #define MdpMinorShift 6
66 #define dprintk(x...) ((void)(DEBUG && printk(x)))
70 static void autostart_arrays (int part
);
73 static LIST_HEAD(pers_list
);
74 static DEFINE_SPINLOCK(pers_lock
);
76 static void md_print_devices(void);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
93 static int sysctl_speed_limit_min
= 1000;
94 static int sysctl_speed_limit_max
= 200000;
95 static inline int speed_min(mddev_t
*mddev
)
97 return mddev
->sync_speed_min
?
98 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
101 static inline int speed_max(mddev_t
*mddev
)
103 return mddev
->sync_speed_max
?
104 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
107 static struct ctl_table_header
*raid_table_header
;
109 static ctl_table raid_table
[] = {
111 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
112 .procname
= "speed_limit_min",
113 .data
= &sysctl_speed_limit_min
,
114 .maxlen
= sizeof(int),
116 .proc_handler
= &proc_dointvec
,
119 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
120 .procname
= "speed_limit_max",
121 .data
= &sysctl_speed_limit_max
,
122 .maxlen
= sizeof(int),
124 .proc_handler
= &proc_dointvec
,
129 static ctl_table raid_dir_table
[] = {
131 .ctl_name
= DEV_RAID
,
140 static ctl_table raid_root_table
[] = {
146 .child
= raid_dir_table
,
151 static struct block_device_operations md_fops
;
153 static int start_readonly
;
156 * We have a system wide 'event count' that is incremented
157 * on any 'interesting' event, and readers of /proc/mdstat
158 * can use 'poll' or 'select' to find out when the event
162 * start array, stop array, error, add device, remove device,
163 * start build, activate spare
165 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
166 static atomic_t md_event_count
;
167 void md_new_event(mddev_t
*mddev
)
169 atomic_inc(&md_event_count
);
170 wake_up(&md_event_waiters
);
171 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
173 EXPORT_SYMBOL_GPL(md_new_event
);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 void md_new_event_inintr(mddev_t
*mddev
)
180 atomic_inc(&md_event_count
);
181 wake_up(&md_event_waiters
);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs
);
189 static DEFINE_SPINLOCK(all_mddevs_lock
);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define ITERATE_MDDEV(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
217 bio_io_error(bio
, bio
->bi_size
);
221 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
223 atomic_inc(&mddev
->active
);
227 static void mddev_put(mddev_t
*mddev
)
229 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
231 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
232 list_del(&mddev
->all_mddevs
);
233 spin_unlock(&all_mddevs_lock
);
234 blk_cleanup_queue(mddev
->queue
);
235 kobject_unregister(&mddev
->kobj
);
237 spin_unlock(&all_mddevs_lock
);
240 static mddev_t
* mddev_find(dev_t unit
)
242 mddev_t
*mddev
, *new = NULL
;
245 spin_lock(&all_mddevs_lock
);
246 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
247 if (mddev
->unit
== unit
) {
249 spin_unlock(&all_mddevs_lock
);
255 list_add(&new->all_mddevs
, &all_mddevs
);
256 spin_unlock(&all_mddevs_lock
);
259 spin_unlock(&all_mddevs_lock
);
261 new = kzalloc(sizeof(*new), GFP_KERNEL
);
266 if (MAJOR(unit
) == MD_MAJOR
)
267 new->md_minor
= MINOR(unit
);
269 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
271 mutex_init(&new->reconfig_mutex
);
272 INIT_LIST_HEAD(&new->disks
);
273 INIT_LIST_HEAD(&new->all_mddevs
);
274 init_timer(&new->safemode_timer
);
275 atomic_set(&new->active
, 1);
276 spin_lock_init(&new->write_lock
);
277 init_waitqueue_head(&new->sb_wait
);
279 new->queue
= blk_alloc_queue(GFP_KERNEL
);
284 set_bit(QUEUE_FLAG_CLUSTER
, &new->queue
->queue_flags
);
286 blk_queue_make_request(new->queue
, md_fail_request
);
291 static inline int mddev_lock(mddev_t
* mddev
)
293 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
296 static inline int mddev_trylock(mddev_t
* mddev
)
298 return mutex_trylock(&mddev
->reconfig_mutex
);
301 static inline void mddev_unlock(mddev_t
* mddev
)
303 mutex_unlock(&mddev
->reconfig_mutex
);
305 md_wakeup_thread(mddev
->thread
);
308 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
311 struct list_head
*tmp
;
313 ITERATE_RDEV(mddev
,rdev
,tmp
) {
314 if (rdev
->desc_nr
== nr
)
320 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
322 struct list_head
*tmp
;
325 ITERATE_RDEV(mddev
,rdev
,tmp
) {
326 if (rdev
->bdev
->bd_dev
== dev
)
332 static struct mdk_personality
*find_pers(int level
, char *clevel
)
334 struct mdk_personality
*pers
;
335 list_for_each_entry(pers
, &pers_list
, list
) {
336 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
338 if (strcmp(pers
->name
, clevel
)==0)
344 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
346 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
347 return MD_NEW_SIZE_BLOCKS(size
);
350 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
354 size
= rdev
->sb_offset
;
357 size
&= ~((sector_t
)chunk_size
/1024 - 1);
361 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
366 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
367 if (!rdev
->sb_page
) {
368 printk(KERN_ALERT
"md: out of memory.\n");
375 static void free_disk_sb(mdk_rdev_t
* rdev
)
378 put_page(rdev
->sb_page
);
380 rdev
->sb_page
= NULL
;
387 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
389 mdk_rdev_t
*rdev
= bio
->bi_private
;
390 mddev_t
*mddev
= rdev
->mddev
;
394 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
395 md_error(mddev
, rdev
);
397 if (atomic_dec_and_test(&mddev
->pending_writes
))
398 wake_up(&mddev
->sb_wait
);
403 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
405 struct bio
*bio2
= bio
->bi_private
;
406 mdk_rdev_t
*rdev
= bio2
->bi_private
;
407 mddev_t
*mddev
= rdev
->mddev
;
411 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
412 error
== -EOPNOTSUPP
) {
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp
, &rdev
->flags
);
416 mddev
->barriers_work
= 0;
417 spin_lock_irqsave(&mddev
->write_lock
, flags
);
418 bio2
->bi_next
= mddev
->biolist
;
419 mddev
->biolist
= bio2
;
420 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
421 wake_up(&mddev
->sb_wait
);
426 bio
->bi_private
= rdev
;
427 return super_written(bio
, bytes_done
, error
);
430 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
431 sector_t sector
, int size
, struct page
*page
)
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
442 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
443 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
445 bio
->bi_bdev
= rdev
->bdev
;
446 bio
->bi_sector
= sector
;
447 bio_add_page(bio
, page
, size
, 0);
448 bio
->bi_private
= rdev
;
449 bio
->bi_end_io
= super_written
;
452 atomic_inc(&mddev
->pending_writes
);
453 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
455 rw
|= (1<<BIO_RW_BARRIER
);
456 rbio
= bio_clone(bio
, GFP_NOIO
);
457 rbio
->bi_private
= bio
;
458 rbio
->bi_end_io
= super_written_barrier
;
459 submit_bio(rw
, rbio
);
464 void md_super_wait(mddev_t
*mddev
)
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
471 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
472 if (atomic_read(&mddev
->pending_writes
)==0)
474 while (mddev
->biolist
) {
476 spin_lock_irq(&mddev
->write_lock
);
477 bio
= mddev
->biolist
;
478 mddev
->biolist
= bio
->bi_next
;
480 spin_unlock_irq(&mddev
->write_lock
);
481 submit_bio(bio
->bi_rw
, bio
);
485 finish_wait(&mddev
->sb_wait
, &wq
);
488 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
493 complete((struct completion
*)bio
->bi_private
);
497 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
498 struct page
*page
, int rw
)
500 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
501 struct completion event
;
504 rw
|= (1 << BIO_RW_SYNC
);
507 bio
->bi_sector
= sector
;
508 bio_add_page(bio
, page
, size
, 0);
509 init_completion(&event
);
510 bio
->bi_private
= &event
;
511 bio
->bi_end_io
= bi_complete
;
513 wait_for_completion(&event
);
515 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
519 EXPORT_SYMBOL_GPL(sync_page_io
);
521 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
523 char b
[BDEVNAME_SIZE
];
524 if (!rdev
->sb_page
) {
532 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
538 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
539 bdevname(rdev
->bdev
,b
));
543 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
545 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
546 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
547 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
548 (sb1
->set_uuid3
== sb2
->set_uuid3
))
556 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
559 mdp_super_t
*tmp1
, *tmp2
;
561 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
562 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
564 if (!tmp1
|| !tmp2
) {
566 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
574 * nr_disks is not constant
579 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
590 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
592 unsigned int disk_csum
, csum
;
594 disk_csum
= sb
->sb_csum
;
596 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
597 sb
->sb_csum
= disk_csum
;
603 * Handle superblock details.
604 * We want to be able to handle multiple superblock formats
605 * so we have a common interface to them all, and an array of
606 * different handlers.
607 * We rely on user-space to write the initial superblock, and support
608 * reading and updating of superblocks.
609 * Interface methods are:
610 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
611 * loads and validates a superblock on dev.
612 * if refdev != NULL, compare superblocks on both devices
614 * 0 - dev has a superblock that is compatible with refdev
615 * 1 - dev has a superblock that is compatible and newer than refdev
616 * so dev should be used as the refdev in future
617 * -EINVAL superblock incompatible or invalid
618 * -othererror e.g. -EIO
620 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
621 * Verify that dev is acceptable into mddev.
622 * The first time, mddev->raid_disks will be 0, and data from
623 * dev should be merged in. Subsequent calls check that dev
624 * is new enough. Return 0 or -EINVAL
626 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
627 * Update the superblock for rdev with data in mddev
628 * This does not write to disc.
634 struct module
*owner
;
635 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
636 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
637 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
641 * load_super for 0.90.0
643 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
645 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
651 * Calculate the position of the superblock,
652 * it's at the end of the disk.
654 * It also happens to be a multiple of 4Kb.
656 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
657 rdev
->sb_offset
= sb_offset
;
659 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
664 bdevname(rdev
->bdev
, b
);
665 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
667 if (sb
->md_magic
!= MD_SB_MAGIC
) {
668 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
673 if (sb
->major_version
!= 0 ||
674 sb
->minor_version
< 90 ||
675 sb
->minor_version
> 91) {
676 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
677 sb
->major_version
, sb
->minor_version
,
682 if (sb
->raid_disks
<= 0)
685 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
686 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
691 rdev
->preferred_minor
= sb
->md_minor
;
692 rdev
->data_offset
= 0;
693 rdev
->sb_size
= MD_SB_BYTES
;
695 if (sb
->level
== LEVEL_MULTIPATH
)
698 rdev
->desc_nr
= sb
->this_disk
.number
;
704 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
705 if (!uuid_equal(refsb
, sb
)) {
706 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
707 b
, bdevname(refdev
->bdev
,b2
));
710 if (!sb_equal(refsb
, sb
)) {
711 printk(KERN_WARNING
"md: %s has same UUID"
712 " but different superblock to %s\n",
713 b
, bdevname(refdev
->bdev
, b2
));
717 ev2
= md_event(refsb
);
723 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
725 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
726 /* "this cannot possibly happen" ... */
734 * validate_super for 0.90.0
736 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
739 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
740 __u64 ev1
= md_event(sb
);
742 rdev
->raid_disk
= -1;
744 if (mddev
->raid_disks
== 0) {
745 mddev
->major_version
= 0;
746 mddev
->minor_version
= sb
->minor_version
;
747 mddev
->patch_version
= sb
->patch_version
;
748 mddev
->persistent
= ! sb
->not_persistent
;
749 mddev
->chunk_size
= sb
->chunk_size
;
750 mddev
->ctime
= sb
->ctime
;
751 mddev
->utime
= sb
->utime
;
752 mddev
->level
= sb
->level
;
753 mddev
->clevel
[0] = 0;
754 mddev
->layout
= sb
->layout
;
755 mddev
->raid_disks
= sb
->raid_disks
;
756 mddev
->size
= sb
->size
;
758 mddev
->bitmap_offset
= 0;
759 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
761 if (mddev
->minor_version
>= 91) {
762 mddev
->reshape_position
= sb
->reshape_position
;
763 mddev
->delta_disks
= sb
->delta_disks
;
764 mddev
->new_level
= sb
->new_level
;
765 mddev
->new_layout
= sb
->new_layout
;
766 mddev
->new_chunk
= sb
->new_chunk
;
768 mddev
->reshape_position
= MaxSector
;
769 mddev
->delta_disks
= 0;
770 mddev
->new_level
= mddev
->level
;
771 mddev
->new_layout
= mddev
->layout
;
772 mddev
->new_chunk
= mddev
->chunk_size
;
775 if (sb
->state
& (1<<MD_SB_CLEAN
))
776 mddev
->recovery_cp
= MaxSector
;
778 if (sb
->events_hi
== sb
->cp_events_hi
&&
779 sb
->events_lo
== sb
->cp_events_lo
) {
780 mddev
->recovery_cp
= sb
->recovery_cp
;
782 mddev
->recovery_cp
= 0;
785 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
786 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
787 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
788 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
790 mddev
->max_disks
= MD_SB_DISKS
;
792 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
793 mddev
->bitmap_file
== NULL
) {
794 if (mddev
->level
!= 1 && mddev
->level
!= 4
795 && mddev
->level
!= 5 && mddev
->level
!= 6
796 && mddev
->level
!= 10) {
797 /* FIXME use a better test */
798 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
801 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
804 } else if (mddev
->pers
== NULL
) {
805 /* Insist on good event counter while assembling */
807 if (ev1
< mddev
->events
)
809 } else if (mddev
->bitmap
) {
810 /* if adding to array with a bitmap, then we can accept an
811 * older device ... but not too old.
813 if (ev1
< mddev
->bitmap
->events_cleared
)
816 if (ev1
< mddev
->events
)
817 /* just a hot-add of a new device, leave raid_disk at -1 */
821 if (mddev
->level
!= LEVEL_MULTIPATH
) {
822 desc
= sb
->disks
+ rdev
->desc_nr
;
824 if (desc
->state
& (1<<MD_DISK_FAULTY
))
825 set_bit(Faulty
, &rdev
->flags
);
826 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
827 desc->raid_disk < mddev->raid_disks */) {
828 set_bit(In_sync
, &rdev
->flags
);
829 rdev
->raid_disk
= desc
->raid_disk
;
831 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
832 set_bit(WriteMostly
, &rdev
->flags
);
833 } else /* MULTIPATH are always insync */
834 set_bit(In_sync
, &rdev
->flags
);
839 * sync_super for 0.90.0
841 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
844 struct list_head
*tmp
;
846 int next_spare
= mddev
->raid_disks
;
849 /* make rdev->sb match mddev data..
852 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
853 * 3/ any empty disks < next_spare become removed
855 * disks[0] gets initialised to REMOVED because
856 * we cannot be sure from other fields if it has
857 * been initialised or not.
860 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
862 rdev
->sb_size
= MD_SB_BYTES
;
864 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
866 memset(sb
, 0, sizeof(*sb
));
868 sb
->md_magic
= MD_SB_MAGIC
;
869 sb
->major_version
= mddev
->major_version
;
870 sb
->patch_version
= mddev
->patch_version
;
871 sb
->gvalid_words
= 0; /* ignored */
872 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
873 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
874 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
875 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
877 sb
->ctime
= mddev
->ctime
;
878 sb
->level
= mddev
->level
;
879 sb
->size
= mddev
->size
;
880 sb
->raid_disks
= mddev
->raid_disks
;
881 sb
->md_minor
= mddev
->md_minor
;
882 sb
->not_persistent
= !mddev
->persistent
;
883 sb
->utime
= mddev
->utime
;
885 sb
->events_hi
= (mddev
->events
>>32);
886 sb
->events_lo
= (u32
)mddev
->events
;
888 if (mddev
->reshape_position
== MaxSector
)
889 sb
->minor_version
= 90;
891 sb
->minor_version
= 91;
892 sb
->reshape_position
= mddev
->reshape_position
;
893 sb
->new_level
= mddev
->new_level
;
894 sb
->delta_disks
= mddev
->delta_disks
;
895 sb
->new_layout
= mddev
->new_layout
;
896 sb
->new_chunk
= mddev
->new_chunk
;
898 mddev
->minor_version
= sb
->minor_version
;
901 sb
->recovery_cp
= mddev
->recovery_cp
;
902 sb
->cp_events_hi
= (mddev
->events
>>32);
903 sb
->cp_events_lo
= (u32
)mddev
->events
;
904 if (mddev
->recovery_cp
== MaxSector
)
905 sb
->state
= (1<< MD_SB_CLEAN
);
909 sb
->layout
= mddev
->layout
;
910 sb
->chunk_size
= mddev
->chunk_size
;
912 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
913 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
915 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
916 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
919 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
920 && !test_bit(Faulty
, &rdev2
->flags
))
921 desc_nr
= rdev2
->raid_disk
;
923 desc_nr
= next_spare
++;
924 rdev2
->desc_nr
= desc_nr
;
925 d
= &sb
->disks
[rdev2
->desc_nr
];
927 d
->number
= rdev2
->desc_nr
;
928 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
929 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
930 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
931 && !test_bit(Faulty
, &rdev2
->flags
))
932 d
->raid_disk
= rdev2
->raid_disk
;
934 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
935 if (test_bit(Faulty
, &rdev2
->flags
))
936 d
->state
= (1<<MD_DISK_FAULTY
);
937 else if (test_bit(In_sync
, &rdev2
->flags
)) {
938 d
->state
= (1<<MD_DISK_ACTIVE
);
939 d
->state
|= (1<<MD_DISK_SYNC
);
947 if (test_bit(WriteMostly
, &rdev2
->flags
))
948 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
950 /* now set the "removed" and "faulty" bits on any missing devices */
951 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
952 mdp_disk_t
*d
= &sb
->disks
[i
];
953 if (d
->state
== 0 && d
->number
== 0) {
956 d
->state
= (1<<MD_DISK_REMOVED
);
957 d
->state
|= (1<<MD_DISK_FAULTY
);
961 sb
->nr_disks
= nr_disks
;
962 sb
->active_disks
= active
;
963 sb
->working_disks
= working
;
964 sb
->failed_disks
= failed
;
965 sb
->spare_disks
= spare
;
967 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
968 sb
->sb_csum
= calc_sb_csum(sb
);
972 * version 1 superblock
975 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
977 unsigned int disk_csum
, csum
;
978 unsigned long long newcsum
;
979 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
980 unsigned int *isuper
= (unsigned int*)sb
;
983 disk_csum
= sb
->sb_csum
;
986 for (i
=0; size
>=4; size
-= 4 )
987 newcsum
+= le32_to_cpu(*isuper
++);
990 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
992 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
993 sb
->sb_csum
= disk_csum
;
994 return cpu_to_le32(csum
);
997 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
999 struct mdp_superblock_1
*sb
;
1002 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1006 * Calculate the position of the superblock.
1007 * It is always aligned to a 4K boundary and
1008 * depeding on minor_version, it can be:
1009 * 0: At least 8K, but less than 12K, from end of device
1010 * 1: At start of device
1011 * 2: 4K from start of device.
1013 switch(minor_version
) {
1015 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1017 sb_offset
&= ~(sector_t
)(4*2-1);
1018 /* convert from sectors to K */
1030 rdev
->sb_offset
= sb_offset
;
1032 /* superblock is rarely larger than 1K, but it can be larger,
1033 * and it is safe to read 4k, so we do that
1035 ret
= read_disk_sb(rdev
, 4096);
1036 if (ret
) return ret
;
1039 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1041 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1042 sb
->major_version
!= cpu_to_le32(1) ||
1043 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1044 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1045 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1048 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1049 printk("md: invalid superblock checksum on %s\n",
1050 bdevname(rdev
->bdev
,b
));
1053 if (le64_to_cpu(sb
->data_size
) < 10) {
1054 printk("md: data_size too small on %s\n",
1055 bdevname(rdev
->bdev
,b
));
1058 rdev
->preferred_minor
= 0xffff;
1059 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1060 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1062 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1063 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1064 if (rdev
->sb_size
& bmask
)
1065 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1071 struct mdp_superblock_1
*refsb
=
1072 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1074 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1075 sb
->level
!= refsb
->level
||
1076 sb
->layout
!= refsb
->layout
||
1077 sb
->chunksize
!= refsb
->chunksize
) {
1078 printk(KERN_WARNING
"md: %s has strangely different"
1079 " superblock to %s\n",
1080 bdevname(rdev
->bdev
,b
),
1081 bdevname(refdev
->bdev
,b2
));
1084 ev1
= le64_to_cpu(sb
->events
);
1085 ev2
= le64_to_cpu(refsb
->events
);
1093 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1095 rdev
->size
= rdev
->sb_offset
;
1096 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1098 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1099 if (le32_to_cpu(sb
->chunksize
))
1100 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1102 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1107 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1109 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1110 __u64 ev1
= le64_to_cpu(sb
->events
);
1112 rdev
->raid_disk
= -1;
1114 if (mddev
->raid_disks
== 0) {
1115 mddev
->major_version
= 1;
1116 mddev
->patch_version
= 0;
1117 mddev
->persistent
= 1;
1118 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1119 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1120 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1121 mddev
->level
= le32_to_cpu(sb
->level
);
1122 mddev
->clevel
[0] = 0;
1123 mddev
->layout
= le32_to_cpu(sb
->layout
);
1124 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1125 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1126 mddev
->events
= ev1
;
1127 mddev
->bitmap_offset
= 0;
1128 mddev
->default_bitmap_offset
= 1024 >> 9;
1130 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1131 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1133 mddev
->max_disks
= (4096-256)/2;
1135 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1136 mddev
->bitmap_file
== NULL
) {
1137 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1138 && mddev
->level
!= 10) {
1139 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1142 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1144 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1145 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1146 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1147 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1148 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1149 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1151 mddev
->reshape_position
= MaxSector
;
1152 mddev
->delta_disks
= 0;
1153 mddev
->new_level
= mddev
->level
;
1154 mddev
->new_layout
= mddev
->layout
;
1155 mddev
->new_chunk
= mddev
->chunk_size
;
1158 } else if (mddev
->pers
== NULL
) {
1159 /* Insist of good event counter while assembling */
1161 if (ev1
< mddev
->events
)
1163 } else if (mddev
->bitmap
) {
1164 /* If adding to array with a bitmap, then we can accept an
1165 * older device, but not too old.
1167 if (ev1
< mddev
->bitmap
->events_cleared
)
1170 if (ev1
< mddev
->events
)
1171 /* just a hot-add of a new device, leave raid_disk at -1 */
1174 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1176 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1177 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1179 case 0xffff: /* spare */
1181 case 0xfffe: /* faulty */
1182 set_bit(Faulty
, &rdev
->flags
);
1185 if ((le32_to_cpu(sb
->feature_map
) &
1186 MD_FEATURE_RECOVERY_OFFSET
))
1187 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1189 set_bit(In_sync
, &rdev
->flags
);
1190 rdev
->raid_disk
= role
;
1193 if (sb
->devflags
& WriteMostly1
)
1194 set_bit(WriteMostly
, &rdev
->flags
);
1195 } else /* MULTIPATH are always insync */
1196 set_bit(In_sync
, &rdev
->flags
);
1201 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1203 struct mdp_superblock_1
*sb
;
1204 struct list_head
*tmp
;
1207 /* make rdev->sb match mddev and rdev data. */
1209 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1211 sb
->feature_map
= 0;
1213 sb
->recovery_offset
= cpu_to_le64(0);
1214 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1215 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1216 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1218 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1219 sb
->events
= cpu_to_le64(mddev
->events
);
1221 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1223 sb
->resync_offset
= cpu_to_le64(0);
1225 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1227 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1228 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1230 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1231 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1232 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1235 if (rdev
->raid_disk
>= 0 &&
1236 !test_bit(In_sync
, &rdev
->flags
) &&
1237 rdev
->recovery_offset
> 0) {
1238 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1239 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1242 if (mddev
->reshape_position
!= MaxSector
) {
1243 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1244 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1245 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1246 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1247 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1248 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1252 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1253 if (rdev2
->desc_nr
+1 > max_dev
)
1254 max_dev
= rdev2
->desc_nr
+1;
1256 sb
->max_dev
= cpu_to_le32(max_dev
);
1257 for (i
=0; i
<max_dev
;i
++)
1258 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1260 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1262 if (test_bit(Faulty
, &rdev2
->flags
))
1263 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1264 else if (test_bit(In_sync
, &rdev2
->flags
))
1265 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1266 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1267 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1269 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1272 sb
->sb_csum
= calc_sb_1_csum(sb
);
1276 static struct super_type super_types
[] = {
1279 .owner
= THIS_MODULE
,
1280 .load_super
= super_90_load
,
1281 .validate_super
= super_90_validate
,
1282 .sync_super
= super_90_sync
,
1286 .owner
= THIS_MODULE
,
1287 .load_super
= super_1_load
,
1288 .validate_super
= super_1_validate
,
1289 .sync_super
= super_1_sync
,
1293 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1295 struct list_head
*tmp
;
1298 ITERATE_RDEV(mddev
,rdev
,tmp
)
1299 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1305 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1307 struct list_head
*tmp
;
1310 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1311 if (match_dev_unit(mddev2
, rdev
))
1317 static LIST_HEAD(pending_raid_disks
);
1319 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1321 mdk_rdev_t
*same_pdev
;
1322 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1330 /* make sure rdev->size exceeds mddev->size */
1331 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1333 /* Cannot change size, so fail */
1336 mddev
->size
= rdev
->size
;
1338 same_pdev
= match_dev_unit(mddev
, rdev
);
1341 "%s: WARNING: %s appears to be on the same physical"
1342 " disk as %s. True\n protection against single-disk"
1343 " failure might be compromised.\n",
1344 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1345 bdevname(same_pdev
->bdev
,b2
));
1347 /* Verify rdev->desc_nr is unique.
1348 * If it is -1, assign a free number, else
1349 * check number is not in use
1351 if (rdev
->desc_nr
< 0) {
1353 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1354 while (find_rdev_nr(mddev
, choice
))
1356 rdev
->desc_nr
= choice
;
1358 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1361 bdevname(rdev
->bdev
,b
);
1362 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1364 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1367 list_add(&rdev
->same_set
, &mddev
->disks
);
1368 rdev
->mddev
= mddev
;
1369 printk(KERN_INFO
"md: bind<%s>\n", b
);
1371 rdev
->kobj
.parent
= &mddev
->kobj
;
1372 kobject_add(&rdev
->kobj
);
1374 if (rdev
->bdev
->bd_part
)
1375 ko
= &rdev
->bdev
->bd_part
->kobj
;
1377 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1378 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1379 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1383 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1385 char b
[BDEVNAME_SIZE
];
1390 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1391 list_del_init(&rdev
->same_set
);
1392 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1394 sysfs_remove_link(&rdev
->kobj
, "block");
1395 kobject_del(&rdev
->kobj
);
1399 * prevent the device from being mounted, repartitioned or
1400 * otherwise reused by a RAID array (or any other kernel
1401 * subsystem), by bd_claiming the device.
1403 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1406 struct block_device
*bdev
;
1407 char b
[BDEVNAME_SIZE
];
1409 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1411 printk(KERN_ERR
"md: could not open %s.\n",
1412 __bdevname(dev
, b
));
1413 return PTR_ERR(bdev
);
1415 err
= bd_claim(bdev
, rdev
);
1417 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1426 static void unlock_rdev(mdk_rdev_t
*rdev
)
1428 struct block_device
*bdev
= rdev
->bdev
;
1436 void md_autodetect_dev(dev_t dev
);
1438 static void export_rdev(mdk_rdev_t
* rdev
)
1440 char b
[BDEVNAME_SIZE
];
1441 printk(KERN_INFO
"md: export_rdev(%s)\n",
1442 bdevname(rdev
->bdev
,b
));
1446 list_del_init(&rdev
->same_set
);
1448 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1451 kobject_put(&rdev
->kobj
);
1454 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1456 unbind_rdev_from_array(rdev
);
1460 static void export_array(mddev_t
*mddev
)
1462 struct list_head
*tmp
;
1465 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1470 kick_rdev_from_array(rdev
);
1472 if (!list_empty(&mddev
->disks
))
1474 mddev
->raid_disks
= 0;
1475 mddev
->major_version
= 0;
1478 static void print_desc(mdp_disk_t
*desc
)
1480 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1481 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1484 static void print_sb(mdp_super_t
*sb
)
1489 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1490 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1491 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1493 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1494 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1495 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1496 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1497 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1498 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1499 sb
->failed_disks
, sb
->spare_disks
,
1500 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1503 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1506 desc
= sb
->disks
+ i
;
1507 if (desc
->number
|| desc
->major
|| desc
->minor
||
1508 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1509 printk(" D %2d: ", i
);
1513 printk(KERN_INFO
"md: THIS: ");
1514 print_desc(&sb
->this_disk
);
1518 static void print_rdev(mdk_rdev_t
*rdev
)
1520 char b
[BDEVNAME_SIZE
];
1521 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1522 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1523 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1525 if (rdev
->sb_loaded
) {
1526 printk(KERN_INFO
"md: rdev superblock:\n");
1527 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1529 printk(KERN_INFO
"md: no rdev superblock!\n");
1532 static void md_print_devices(void)
1534 struct list_head
*tmp
, *tmp2
;
1537 char b
[BDEVNAME_SIZE
];
1540 printk("md: **********************************\n");
1541 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1542 printk("md: **********************************\n");
1543 ITERATE_MDDEV(mddev
,tmp
) {
1546 bitmap_print_sb(mddev
->bitmap
);
1548 printk("%s: ", mdname(mddev
));
1549 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1550 printk("<%s>", bdevname(rdev
->bdev
,b
));
1553 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1556 printk("md: **********************************\n");
1561 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1563 /* Update each superblock (in-memory image), but
1564 * if we are allowed to, skip spares which already
1565 * have the right event counter, or have one earlier
1566 * (which would mean they aren't being marked as dirty
1567 * with the rest of the array)
1570 struct list_head
*tmp
;
1572 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1573 if (rdev
->sb_events
== mddev
->events
||
1575 rdev
->raid_disk
< 0 &&
1576 (rdev
->sb_events
&1)==0 &&
1577 rdev
->sb_events
+1 == mddev
->events
)) {
1578 /* Don't update this superblock */
1579 rdev
->sb_loaded
= 2;
1581 super_types
[mddev
->major_version
].
1582 sync_super(mddev
, rdev
);
1583 rdev
->sb_loaded
= 1;
1588 void md_update_sb(mddev_t
* mddev
)
1591 struct list_head
*tmp
;
1597 spin_lock_irq(&mddev
->write_lock
);
1598 sync_req
= mddev
->in_sync
;
1599 mddev
->utime
= get_seconds();
1600 if (mddev
->sb_dirty
== 3)
1601 /* just a clean<-> dirty transition, possibly leave spares alone,
1602 * though if events isn't the right even/odd, we will have to do
1607 /* If this is just a dirty<->clean transition, and the array is clean
1608 * and 'events' is odd, we can roll back to the previous clean state */
1609 if (mddev
->sb_dirty
== 3
1610 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1611 && (mddev
->events
& 1))
1614 /* otherwise we have to go forward and ... */
1616 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1617 /* .. if the array isn't clean, insist on an odd 'events' */
1618 if ((mddev
->events
&1)==0) {
1623 /* otherwise insist on an even 'events' (for clean states) */
1624 if ((mddev
->events
&1)) {
1631 if (!mddev
->events
) {
1633 * oops, this 64-bit counter should never wrap.
1634 * Either we are in around ~1 trillion A.C., assuming
1635 * 1 reboot per second, or we have a bug:
1640 mddev
->sb_dirty
= 2;
1641 sync_sbs(mddev
, nospares
);
1644 * do not write anything to disk if using
1645 * nonpersistent superblocks
1647 if (!mddev
->persistent
) {
1648 mddev
->sb_dirty
= 0;
1649 spin_unlock_irq(&mddev
->write_lock
);
1650 wake_up(&mddev
->sb_wait
);
1653 spin_unlock_irq(&mddev
->write_lock
);
1656 "md: updating %s RAID superblock on device (in sync %d)\n",
1657 mdname(mddev
),mddev
->in_sync
);
1659 err
= bitmap_update_sb(mddev
->bitmap
);
1660 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1661 char b
[BDEVNAME_SIZE
];
1662 dprintk(KERN_INFO
"md: ");
1663 if (rdev
->sb_loaded
!= 1)
1664 continue; /* no noise on spare devices */
1665 if (test_bit(Faulty
, &rdev
->flags
))
1666 dprintk("(skipping faulty ");
1668 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1669 if (!test_bit(Faulty
, &rdev
->flags
)) {
1670 md_super_write(mddev
,rdev
,
1671 rdev
->sb_offset
<<1, rdev
->sb_size
,
1673 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1674 bdevname(rdev
->bdev
,b
),
1675 (unsigned long long)rdev
->sb_offset
);
1676 rdev
->sb_events
= mddev
->events
;
1680 if (mddev
->level
== LEVEL_MULTIPATH
)
1681 /* only need to write one superblock... */
1684 md_super_wait(mddev
);
1685 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1687 spin_lock_irq(&mddev
->write_lock
);
1688 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1689 /* have to write it out again */
1690 spin_unlock_irq(&mddev
->write_lock
);
1693 mddev
->sb_dirty
= 0;
1694 spin_unlock_irq(&mddev
->write_lock
);
1695 wake_up(&mddev
->sb_wait
);
1698 EXPORT_SYMBOL_GPL(md_update_sb
);
1700 /* words written to sysfs files may, or my not, be \n terminated.
1701 * We want to accept with case. For this we use cmd_match.
1703 static int cmd_match(const char *cmd
, const char *str
)
1705 /* See if cmd, written into a sysfs file, matches
1706 * str. They must either be the same, or cmd can
1707 * have a trailing newline
1709 while (*cmd
&& *str
&& *cmd
== *str
) {
1720 struct rdev_sysfs_entry
{
1721 struct attribute attr
;
1722 ssize_t (*show
)(mdk_rdev_t
*, char *);
1723 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1727 state_show(mdk_rdev_t
*rdev
, char *page
)
1732 if (test_bit(Faulty
, &rdev
->flags
)) {
1733 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1736 if (test_bit(In_sync
, &rdev
->flags
)) {
1737 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1740 if (!test_bit(Faulty
, &rdev
->flags
) &&
1741 !test_bit(In_sync
, &rdev
->flags
)) {
1742 len
+= sprintf(page
+len
, "%sspare", sep
);
1745 return len
+sprintf(page
+len
, "\n");
1748 static struct rdev_sysfs_entry
1749 rdev_state
= __ATTR_RO(state
);
1752 super_show(mdk_rdev_t
*rdev
, char *page
)
1754 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1755 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1756 return rdev
->sb_size
;
1760 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1763 errors_show(mdk_rdev_t
*rdev
, char *page
)
1765 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1769 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1772 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1773 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1774 atomic_set(&rdev
->corrected_errors
, n
);
1779 static struct rdev_sysfs_entry rdev_errors
=
1780 __ATTR(errors
, 0644, errors_show
, errors_store
);
1783 slot_show(mdk_rdev_t
*rdev
, char *page
)
1785 if (rdev
->raid_disk
< 0)
1786 return sprintf(page
, "none\n");
1788 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1792 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1795 int slot
= simple_strtoul(buf
, &e
, 10);
1796 if (strncmp(buf
, "none", 4)==0)
1798 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1800 if (rdev
->mddev
->pers
)
1801 /* Cannot set slot in active array (yet) */
1803 if (slot
>= rdev
->mddev
->raid_disks
)
1805 rdev
->raid_disk
= slot
;
1806 /* assume it is working */
1808 set_bit(In_sync
, &rdev
->flags
);
1813 static struct rdev_sysfs_entry rdev_slot
=
1814 __ATTR(slot
, 0644, slot_show
, slot_store
);
1817 offset_show(mdk_rdev_t
*rdev
, char *page
)
1819 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1823 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1826 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1827 if (e
==buf
|| (*e
&& *e
!= '\n'))
1829 if (rdev
->mddev
->pers
)
1831 rdev
->data_offset
= offset
;
1835 static struct rdev_sysfs_entry rdev_offset
=
1836 __ATTR(offset
, 0644, offset_show
, offset_store
);
1839 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1841 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1845 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1848 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1849 if (e
==buf
|| (*e
&& *e
!= '\n'))
1851 if (rdev
->mddev
->pers
)
1854 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1855 rdev
->mddev
->size
= size
;
1859 static struct rdev_sysfs_entry rdev_size
=
1860 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1862 static struct attribute
*rdev_default_attrs
[] = {
1872 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1874 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1875 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1879 return entry
->show(rdev
, page
);
1883 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1884 const char *page
, size_t length
)
1886 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1887 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1891 return entry
->store(rdev
, page
, length
);
1894 static void rdev_free(struct kobject
*ko
)
1896 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1899 static struct sysfs_ops rdev_sysfs_ops
= {
1900 .show
= rdev_attr_show
,
1901 .store
= rdev_attr_store
,
1903 static struct kobj_type rdev_ktype
= {
1904 .release
= rdev_free
,
1905 .sysfs_ops
= &rdev_sysfs_ops
,
1906 .default_attrs
= rdev_default_attrs
,
1910 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1912 * mark the device faulty if:
1914 * - the device is nonexistent (zero size)
1915 * - the device has no valid superblock
1917 * a faulty rdev _never_ has rdev->sb set.
1919 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1921 char b
[BDEVNAME_SIZE
];
1926 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1928 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1929 return ERR_PTR(-ENOMEM
);
1932 if ((err
= alloc_disk_sb(rdev
)))
1935 err
= lock_rdev(rdev
, newdev
);
1939 rdev
->kobj
.parent
= NULL
;
1940 rdev
->kobj
.ktype
= &rdev_ktype
;
1941 kobject_init(&rdev
->kobj
);
1945 rdev
->data_offset
= 0;
1946 rdev
->sb_events
= 0;
1947 atomic_set(&rdev
->nr_pending
, 0);
1948 atomic_set(&rdev
->read_errors
, 0);
1949 atomic_set(&rdev
->corrected_errors
, 0);
1951 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1954 "md: %s has zero or unknown size, marking faulty!\n",
1955 bdevname(rdev
->bdev
,b
));
1960 if (super_format
>= 0) {
1961 err
= super_types
[super_format
].
1962 load_super(rdev
, NULL
, super_minor
);
1963 if (err
== -EINVAL
) {
1965 "md: %s has invalid sb, not importing!\n",
1966 bdevname(rdev
->bdev
,b
));
1971 "md: could not read %s's sb, not importing!\n",
1972 bdevname(rdev
->bdev
,b
));
1976 INIT_LIST_HEAD(&rdev
->same_set
);
1981 if (rdev
->sb_page
) {
1987 return ERR_PTR(err
);
1991 * Check a full RAID array for plausibility
1995 static void analyze_sbs(mddev_t
* mddev
)
1998 struct list_head
*tmp
;
1999 mdk_rdev_t
*rdev
, *freshest
;
2000 char b
[BDEVNAME_SIZE
];
2003 ITERATE_RDEV(mddev
,rdev
,tmp
)
2004 switch (super_types
[mddev
->major_version
].
2005 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2013 "md: fatal superblock inconsistency in %s"
2014 " -- removing from array\n",
2015 bdevname(rdev
->bdev
,b
));
2016 kick_rdev_from_array(rdev
);
2020 super_types
[mddev
->major_version
].
2021 validate_super(mddev
, freshest
);
2024 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2025 if (rdev
!= freshest
)
2026 if (super_types
[mddev
->major_version
].
2027 validate_super(mddev
, rdev
)) {
2028 printk(KERN_WARNING
"md: kicking non-fresh %s"
2030 bdevname(rdev
->bdev
,b
));
2031 kick_rdev_from_array(rdev
);
2034 if (mddev
->level
== LEVEL_MULTIPATH
) {
2035 rdev
->desc_nr
= i
++;
2036 rdev
->raid_disk
= rdev
->desc_nr
;
2037 set_bit(In_sync
, &rdev
->flags
);
2043 if (mddev
->recovery_cp
!= MaxSector
&&
2045 printk(KERN_ERR
"md: %s: raid array is not clean"
2046 " -- starting background reconstruction\n",
2052 safe_delay_show(mddev_t
*mddev
, char *page
)
2054 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2055 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2058 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2066 /* remove a period, and count digits after it */
2067 if (len
>= sizeof(buf
))
2069 strlcpy(buf
, cbuf
, len
);
2071 for (i
=0; i
<len
; i
++) {
2073 if (isdigit(buf
[i
])) {
2078 } else if (buf
[i
] == '.') {
2083 msec
= simple_strtoul(buf
, &e
, 10);
2084 if (e
== buf
|| (*e
&& *e
!= '\n'))
2086 msec
= (msec
* 1000) / scale
;
2088 mddev
->safemode_delay
= 0;
2090 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2091 if (mddev
->safemode_delay
== 0)
2092 mddev
->safemode_delay
= 1;
2096 static struct md_sysfs_entry md_safe_delay
=
2097 __ATTR(safe_mode_delay
, 0644,safe_delay_show
, safe_delay_store
);
2100 level_show(mddev_t
*mddev
, char *page
)
2102 struct mdk_personality
*p
= mddev
->pers
;
2104 return sprintf(page
, "%s\n", p
->name
);
2105 else if (mddev
->clevel
[0])
2106 return sprintf(page
, "%s\n", mddev
->clevel
);
2107 else if (mddev
->level
!= LEVEL_NONE
)
2108 return sprintf(page
, "%d\n", mddev
->level
);
2114 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2121 if (len
>= sizeof(mddev
->clevel
))
2123 strncpy(mddev
->clevel
, buf
, len
);
2124 if (mddev
->clevel
[len
-1] == '\n')
2126 mddev
->clevel
[len
] = 0;
2127 mddev
->level
= LEVEL_NONE
;
2131 static struct md_sysfs_entry md_level
=
2132 __ATTR(level
, 0644, level_show
, level_store
);
2135 raid_disks_show(mddev_t
*mddev
, char *page
)
2137 if (mddev
->raid_disks
== 0)
2139 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2142 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2145 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2147 /* can only set raid_disks if array is not yet active */
2150 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2152 if (!*buf
|| (*e
&& *e
!= '\n'))
2156 rv
= update_raid_disks(mddev
, n
);
2158 mddev
->raid_disks
= n
;
2159 return rv
? rv
: len
;
2161 static struct md_sysfs_entry md_raid_disks
=
2162 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
2165 chunk_size_show(mddev_t
*mddev
, char *page
)
2167 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2171 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2173 /* can only set chunk_size if array is not yet active */
2175 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2179 if (!*buf
|| (*e
&& *e
!= '\n'))
2182 mddev
->chunk_size
= n
;
2185 static struct md_sysfs_entry md_chunk_size
=
2186 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2189 null_show(mddev_t
*mddev
, char *page
)
2195 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2197 /* buf must be %d:%d\n? giving major and minor numbers */
2198 /* The new device is added to the array.
2199 * If the array has a persistent superblock, we read the
2200 * superblock to initialise info and check validity.
2201 * Otherwise, only checking done is that in bind_rdev_to_array,
2202 * which mainly checks size.
2205 int major
= simple_strtoul(buf
, &e
, 10);
2211 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2213 minor
= simple_strtoul(e
+1, &e
, 10);
2214 if (*e
&& *e
!= '\n')
2216 dev
= MKDEV(major
, minor
);
2217 if (major
!= MAJOR(dev
) ||
2218 minor
!= MINOR(dev
))
2222 if (mddev
->persistent
) {
2223 rdev
= md_import_device(dev
, mddev
->major_version
,
2224 mddev
->minor_version
);
2225 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2226 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2227 mdk_rdev_t
, same_set
);
2228 err
= super_types
[mddev
->major_version
]
2229 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2234 rdev
= md_import_device(dev
, -1, -1);
2237 return PTR_ERR(rdev
);
2238 err
= bind_rdev_to_array(rdev
, mddev
);
2242 return err
? err
: len
;
2245 static struct md_sysfs_entry md_new_device
=
2246 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2249 size_show(mddev_t
*mddev
, char *page
)
2251 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2254 static int update_size(mddev_t
*mddev
, unsigned long size
);
2257 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2259 /* If array is inactive, we can reduce the component size, but
2260 * not increase it (except from 0).
2261 * If array is active, we can try an on-line resize
2265 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2266 if (!*buf
|| *buf
== '\n' ||
2271 err
= update_size(mddev
, size
);
2272 md_update_sb(mddev
);
2274 if (mddev
->size
== 0 ||
2280 return err
? err
: len
;
2283 static struct md_sysfs_entry md_size
=
2284 __ATTR(component_size
, 0644, size_show
, size_store
);
2288 * This is either 'none' for arrays with externally managed metadata,
2289 * or N.M for internally known formats
2292 metadata_show(mddev_t
*mddev
, char *page
)
2294 if (mddev
->persistent
)
2295 return sprintf(page
, "%d.%d\n",
2296 mddev
->major_version
, mddev
->minor_version
);
2298 return sprintf(page
, "none\n");
2302 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2306 if (!list_empty(&mddev
->disks
))
2309 if (cmd_match(buf
, "none")) {
2310 mddev
->persistent
= 0;
2311 mddev
->major_version
= 0;
2312 mddev
->minor_version
= 90;
2315 major
= simple_strtoul(buf
, &e
, 10);
2316 if (e
==buf
|| *e
!= '.')
2319 minor
= simple_strtoul(buf
, &e
, 10);
2320 if (e
==buf
|| *e
!= '\n')
2322 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2323 super_types
[major
].name
== NULL
)
2325 mddev
->major_version
= major
;
2326 mddev
->minor_version
= minor
;
2327 mddev
->persistent
= 1;
2331 static struct md_sysfs_entry md_metadata
=
2332 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2335 action_show(mddev_t
*mddev
, char *page
)
2337 char *type
= "idle";
2338 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2339 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2340 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2342 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2343 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2345 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2352 return sprintf(page
, "%s\n", type
);
2356 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2358 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2361 if (cmd_match(page
, "idle")) {
2362 if (mddev
->sync_thread
) {
2363 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2364 md_unregister_thread(mddev
->sync_thread
);
2365 mddev
->sync_thread
= NULL
;
2366 mddev
->recovery
= 0;
2368 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2369 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2371 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2372 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2373 else if (cmd_match(page
, "reshape")) {
2375 if (mddev
->pers
->start_reshape
== NULL
)
2377 err
= mddev
->pers
->start_reshape(mddev
);
2381 if (cmd_match(page
, "check"))
2382 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2383 else if (!cmd_match(page
, "repair"))
2385 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2386 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2388 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2389 md_wakeup_thread(mddev
->thread
);
2394 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2396 return sprintf(page
, "%llu\n",
2397 (unsigned long long) mddev
->resync_mismatches
);
2400 static struct md_sysfs_entry
2401 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2404 static struct md_sysfs_entry
2405 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2408 sync_min_show(mddev_t
*mddev
, char *page
)
2410 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2411 mddev
->sync_speed_min
? "local": "system");
2415 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2419 if (strncmp(buf
, "system", 6)==0) {
2420 mddev
->sync_speed_min
= 0;
2423 min
= simple_strtoul(buf
, &e
, 10);
2424 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2426 mddev
->sync_speed_min
= min
;
2430 static struct md_sysfs_entry md_sync_min
=
2431 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2434 sync_max_show(mddev_t
*mddev
, char *page
)
2436 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2437 mddev
->sync_speed_max
? "local": "system");
2441 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2445 if (strncmp(buf
, "system", 6)==0) {
2446 mddev
->sync_speed_max
= 0;
2449 max
= simple_strtoul(buf
, &e
, 10);
2450 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2452 mddev
->sync_speed_max
= max
;
2456 static struct md_sysfs_entry md_sync_max
=
2457 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2461 sync_speed_show(mddev_t
*mddev
, char *page
)
2463 unsigned long resync
, dt
, db
;
2464 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2465 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2467 db
= resync
- (mddev
->resync_mark_cnt
);
2468 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2471 static struct md_sysfs_entry
2472 md_sync_speed
= __ATTR_RO(sync_speed
);
2475 sync_completed_show(mddev_t
*mddev
, char *page
)
2477 unsigned long max_blocks
, resync
;
2479 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2480 max_blocks
= mddev
->resync_max_sectors
;
2482 max_blocks
= mddev
->size
<< 1;
2484 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2485 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2488 static struct md_sysfs_entry
2489 md_sync_completed
= __ATTR_RO(sync_completed
);
2492 suspend_lo_show(mddev_t
*mddev
, char *page
)
2494 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2498 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2501 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2503 if (mddev
->pers
->quiesce
== NULL
)
2505 if (buf
== e
|| (*e
&& *e
!= '\n'))
2507 if (new >= mddev
->suspend_hi
||
2508 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2509 mddev
->suspend_lo
= new;
2510 mddev
->pers
->quiesce(mddev
, 2);
2515 static struct md_sysfs_entry md_suspend_lo
=
2516 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2520 suspend_hi_show(mddev_t
*mddev
, char *page
)
2522 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2526 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2529 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2531 if (mddev
->pers
->quiesce
== NULL
)
2533 if (buf
== e
|| (*e
&& *e
!= '\n'))
2535 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2536 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2537 mddev
->suspend_hi
= new;
2538 mddev
->pers
->quiesce(mddev
, 1);
2539 mddev
->pers
->quiesce(mddev
, 0);
2544 static struct md_sysfs_entry md_suspend_hi
=
2545 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2548 static struct attribute
*md_default_attrs
[] = {
2550 &md_raid_disks
.attr
,
2551 &md_chunk_size
.attr
,
2554 &md_new_device
.attr
,
2555 &md_safe_delay
.attr
,
2559 static struct attribute
*md_redundancy_attrs
[] = {
2561 &md_mismatches
.attr
,
2564 &md_sync_speed
.attr
,
2565 &md_sync_completed
.attr
,
2566 &md_suspend_lo
.attr
,
2567 &md_suspend_hi
.attr
,
2570 static struct attribute_group md_redundancy_group
= {
2572 .attrs
= md_redundancy_attrs
,
2577 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2579 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2580 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2585 rv
= mddev_lock(mddev
);
2587 rv
= entry
->show(mddev
, page
);
2588 mddev_unlock(mddev
);
2594 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2595 const char *page
, size_t length
)
2597 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2598 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2603 rv
= mddev_lock(mddev
);
2605 rv
= entry
->store(mddev
, page
, length
);
2606 mddev_unlock(mddev
);
2611 static void md_free(struct kobject
*ko
)
2613 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2617 static struct sysfs_ops md_sysfs_ops
= {
2618 .show
= md_attr_show
,
2619 .store
= md_attr_store
,
2621 static struct kobj_type md_ktype
= {
2623 .sysfs_ops
= &md_sysfs_ops
,
2624 .default_attrs
= md_default_attrs
,
2629 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2631 static DEFINE_MUTEX(disks_mutex
);
2632 mddev_t
*mddev
= mddev_find(dev
);
2633 struct gendisk
*disk
;
2634 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2635 int shift
= partitioned
? MdpMinorShift
: 0;
2636 int unit
= MINOR(dev
) >> shift
;
2641 mutex_lock(&disks_mutex
);
2642 if (mddev
->gendisk
) {
2643 mutex_unlock(&disks_mutex
);
2647 disk
= alloc_disk(1 << shift
);
2649 mutex_unlock(&disks_mutex
);
2653 disk
->major
= MAJOR(dev
);
2654 disk
->first_minor
= unit
<< shift
;
2656 sprintf(disk
->disk_name
, "md_d%d", unit
);
2657 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2659 sprintf(disk
->disk_name
, "md%d", unit
);
2660 sprintf(disk
->devfs_name
, "md/%d", unit
);
2662 disk
->fops
= &md_fops
;
2663 disk
->private_data
= mddev
;
2664 disk
->queue
= mddev
->queue
;
2666 mddev
->gendisk
= disk
;
2667 mutex_unlock(&disks_mutex
);
2668 mddev
->kobj
.parent
= &disk
->kobj
;
2669 mddev
->kobj
.k_name
= NULL
;
2670 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2671 mddev
->kobj
.ktype
= &md_ktype
;
2672 kobject_register(&mddev
->kobj
);
2676 static void md_safemode_timeout(unsigned long data
)
2678 mddev_t
*mddev
= (mddev_t
*) data
;
2680 mddev
->safemode
= 1;
2681 md_wakeup_thread(mddev
->thread
);
2684 static int start_dirty_degraded
;
2686 static int do_md_run(mddev_t
* mddev
)
2690 struct list_head
*tmp
;
2692 struct gendisk
*disk
;
2693 struct mdk_personality
*pers
;
2694 char b
[BDEVNAME_SIZE
];
2696 if (list_empty(&mddev
->disks
))
2697 /* cannot run an array with no devices.. */
2704 * Analyze all RAID superblock(s)
2706 if (!mddev
->raid_disks
)
2709 chunk_size
= mddev
->chunk_size
;
2712 if (chunk_size
> MAX_CHUNK_SIZE
) {
2713 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2714 chunk_size
, MAX_CHUNK_SIZE
);
2718 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2720 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2721 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2724 if (chunk_size
< PAGE_SIZE
) {
2725 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2726 chunk_size
, PAGE_SIZE
);
2730 /* devices must have minimum size of one chunk */
2731 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2732 if (test_bit(Faulty
, &rdev
->flags
))
2734 if (rdev
->size
< chunk_size
/ 1024) {
2736 "md: Dev %s smaller than chunk_size:"
2738 bdevname(rdev
->bdev
,b
),
2739 (unsigned long long)rdev
->size
,
2747 if (mddev
->level
!= LEVEL_NONE
)
2748 request_module("md-level-%d", mddev
->level
);
2749 else if (mddev
->clevel
[0])
2750 request_module("md-%s", mddev
->clevel
);
2754 * Drop all container device buffers, from now on
2755 * the only valid external interface is through the md
2757 * Also find largest hardsector size
2759 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2760 if (test_bit(Faulty
, &rdev
->flags
))
2762 sync_blockdev(rdev
->bdev
);
2763 invalidate_bdev(rdev
->bdev
, 0);
2766 md_probe(mddev
->unit
, NULL
, NULL
);
2767 disk
= mddev
->gendisk
;
2771 spin_lock(&pers_lock
);
2772 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2773 if (!pers
|| !try_module_get(pers
->owner
)) {
2774 spin_unlock(&pers_lock
);
2775 if (mddev
->level
!= LEVEL_NONE
)
2776 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2779 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2784 spin_unlock(&pers_lock
);
2785 mddev
->level
= pers
->level
;
2786 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2788 if (mddev
->reshape_position
!= MaxSector
&&
2789 pers
->start_reshape
== NULL
) {
2790 /* This personality cannot handle reshaping... */
2792 module_put(pers
->owner
);
2796 mddev
->recovery
= 0;
2797 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2798 mddev
->barriers_work
= 1;
2799 mddev
->ok_start_degraded
= start_dirty_degraded
;
2802 mddev
->ro
= 2; /* read-only, but switch on first write */
2804 err
= mddev
->pers
->run(mddev
);
2805 if (!err
&& mddev
->pers
->sync_request
) {
2806 err
= bitmap_create(mddev
);
2808 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2809 mdname(mddev
), err
);
2810 mddev
->pers
->stop(mddev
);
2814 printk(KERN_ERR
"md: pers->run() failed ...\n");
2815 module_put(mddev
->pers
->owner
);
2817 bitmap_destroy(mddev
);
2820 if (mddev
->pers
->sync_request
)
2821 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2822 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2825 atomic_set(&mddev
->writes_pending
,0);
2826 mddev
->safemode
= 0;
2827 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2828 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2829 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
2832 ITERATE_RDEV(mddev
,rdev
,tmp
)
2833 if (rdev
->raid_disk
>= 0) {
2835 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2836 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2839 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2840 md_wakeup_thread(mddev
->thread
);
2842 if (mddev
->sb_dirty
)
2843 md_update_sb(mddev
);
2845 set_capacity(disk
, mddev
->array_size
<<1);
2847 /* If we call blk_queue_make_request here, it will
2848 * re-initialise max_sectors etc which may have been
2849 * refined inside -> run. So just set the bits we need to set.
2850 * Most initialisation happended when we called
2851 * blk_queue_make_request(..., md_fail_request)
2854 mddev
->queue
->queuedata
= mddev
;
2855 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2857 /* If there is a partially-recovered drive we need to
2858 * start recovery here. If we leave it to md_check_recovery,
2859 * it will remove the drives and not do the right thing
2861 if (mddev
->degraded
) {
2862 struct list_head
*rtmp
;
2864 ITERATE_RDEV(mddev
,rdev
,rtmp
)
2865 if (rdev
->raid_disk
>= 0 &&
2866 !test_bit(In_sync
, &rdev
->flags
) &&
2867 !test_bit(Faulty
, &rdev
->flags
))
2868 /* complete an interrupted recovery */
2870 if (spares
&& mddev
->pers
->sync_request
) {
2871 mddev
->recovery
= 0;
2872 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
2873 mddev
->sync_thread
= md_register_thread(md_do_sync
,
2876 if (!mddev
->sync_thread
) {
2877 printk(KERN_ERR
"%s: could not start resync"
2880 /* leave the spares where they are, it shouldn't hurt */
2881 mddev
->recovery
= 0;
2883 md_wakeup_thread(mddev
->sync_thread
);
2888 md_new_event(mddev
);
2892 static int restart_array(mddev_t
*mddev
)
2894 struct gendisk
*disk
= mddev
->gendisk
;
2898 * Complain if it has no devices
2901 if (list_empty(&mddev
->disks
))
2909 mddev
->safemode
= 0;
2911 set_disk_ro(disk
, 0);
2913 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2916 * Kick recovery or resync if necessary
2918 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2919 md_wakeup_thread(mddev
->thread
);
2920 md_wakeup_thread(mddev
->sync_thread
);
2923 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2932 /* similar to deny_write_access, but accounts for our holding a reference
2933 * to the file ourselves */
2934 static int deny_bitmap_write_access(struct file
* file
)
2936 struct inode
*inode
= file
->f_mapping
->host
;
2938 spin_lock(&inode
->i_lock
);
2939 if (atomic_read(&inode
->i_writecount
) > 1) {
2940 spin_unlock(&inode
->i_lock
);
2943 atomic_set(&inode
->i_writecount
, -1);
2944 spin_unlock(&inode
->i_lock
);
2949 static void restore_bitmap_write_access(struct file
*file
)
2951 struct inode
*inode
= file
->f_mapping
->host
;
2953 spin_lock(&inode
->i_lock
);
2954 atomic_set(&inode
->i_writecount
, 1);
2955 spin_unlock(&inode
->i_lock
);
2958 static int do_md_stop(mddev_t
* mddev
, int ro
)
2961 struct gendisk
*disk
= mddev
->gendisk
;
2964 if (atomic_read(&mddev
->active
)>2) {
2965 printk("md: %s still in use.\n",mdname(mddev
));
2969 if (mddev
->sync_thread
) {
2970 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
2971 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2972 md_unregister_thread(mddev
->sync_thread
);
2973 mddev
->sync_thread
= NULL
;
2976 del_timer_sync(&mddev
->safemode_timer
);
2978 invalidate_partition(disk
, 0);
2986 bitmap_flush(mddev
);
2987 md_super_wait(mddev
);
2989 set_disk_ro(disk
, 0);
2990 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2991 mddev
->pers
->stop(mddev
);
2992 if (mddev
->pers
->sync_request
)
2993 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2995 module_put(mddev
->pers
->owner
);
3000 if (!mddev
->in_sync
|| mddev
->sb_dirty
) {
3001 /* mark array as shutdown cleanly */
3003 md_update_sb(mddev
);
3006 set_disk_ro(disk
, 1);
3007 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3011 * Free resources if final stop
3015 struct list_head
*tmp
;
3016 struct gendisk
*disk
;
3017 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3019 bitmap_destroy(mddev
);
3020 if (mddev
->bitmap_file
) {
3021 restore_bitmap_write_access(mddev
->bitmap_file
);
3022 fput(mddev
->bitmap_file
);
3023 mddev
->bitmap_file
= NULL
;
3025 mddev
->bitmap_offset
= 0;
3027 ITERATE_RDEV(mddev
,rdev
,tmp
)
3028 if (rdev
->raid_disk
>= 0) {
3030 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3031 sysfs_remove_link(&mddev
->kobj
, nm
);
3034 export_array(mddev
);
3036 mddev
->array_size
= 0;
3037 disk
= mddev
->gendisk
;
3039 set_capacity(disk
, 0);
3041 } else if (mddev
->pers
)
3042 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3045 md_new_event(mddev
);
3050 static void autorun_array(mddev_t
*mddev
)
3053 struct list_head
*tmp
;
3056 if (list_empty(&mddev
->disks
))
3059 printk(KERN_INFO
"md: running: ");
3061 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3062 char b
[BDEVNAME_SIZE
];
3063 printk("<%s>", bdevname(rdev
->bdev
,b
));
3067 err
= do_md_run (mddev
);
3069 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3070 do_md_stop (mddev
, 0);
3075 * lets try to run arrays based on all disks that have arrived
3076 * until now. (those are in pending_raid_disks)
3078 * the method: pick the first pending disk, collect all disks with
3079 * the same UUID, remove all from the pending list and put them into
3080 * the 'same_array' list. Then order this list based on superblock
3081 * update time (freshest comes first), kick out 'old' disks and
3082 * compare superblocks. If everything's fine then run it.
3084 * If "unit" is allocated, then bump its reference count
3086 static void autorun_devices(int part
)
3088 struct list_head
*tmp
;
3089 mdk_rdev_t
*rdev0
, *rdev
;
3091 char b
[BDEVNAME_SIZE
];
3093 printk(KERN_INFO
"md: autorun ...\n");
3094 while (!list_empty(&pending_raid_disks
)) {
3096 LIST_HEAD(candidates
);
3097 rdev0
= list_entry(pending_raid_disks
.next
,
3098 mdk_rdev_t
, same_set
);
3100 printk(KERN_INFO
"md: considering %s ...\n",
3101 bdevname(rdev0
->bdev
,b
));
3102 INIT_LIST_HEAD(&candidates
);
3103 ITERATE_RDEV_PENDING(rdev
,tmp
)
3104 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3105 printk(KERN_INFO
"md: adding %s ...\n",
3106 bdevname(rdev
->bdev
,b
));
3107 list_move(&rdev
->same_set
, &candidates
);
3110 * now we have a set of devices, with all of them having
3111 * mostly sane superblocks. It's time to allocate the
3114 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
3115 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3116 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3120 dev
= MKDEV(mdp_major
,
3121 rdev0
->preferred_minor
<< MdpMinorShift
);
3123 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3125 md_probe(dev
, NULL
, NULL
);
3126 mddev
= mddev_find(dev
);
3129 "md: cannot allocate memory for md drive.\n");
3132 if (mddev_lock(mddev
))
3133 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3135 else if (mddev
->raid_disks
|| mddev
->major_version
3136 || !list_empty(&mddev
->disks
)) {
3138 "md: %s already running, cannot run %s\n",
3139 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3140 mddev_unlock(mddev
);
3142 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3143 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3144 list_del_init(&rdev
->same_set
);
3145 if (bind_rdev_to_array(rdev
, mddev
))
3148 autorun_array(mddev
);
3149 mddev_unlock(mddev
);
3151 /* on success, candidates will be empty, on error
3154 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3158 printk(KERN_INFO
"md: ... autorun DONE.\n");
3162 * import RAID devices based on one partition
3163 * if possible, the array gets run as well.
3166 static int autostart_array(dev_t startdev
)
3168 char b
[BDEVNAME_SIZE
];
3169 int err
= -EINVAL
, i
;
3170 mdp_super_t
*sb
= NULL
;
3171 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
3173 start_rdev
= md_import_device(startdev
, 0, 0);
3174 if (IS_ERR(start_rdev
))
3178 /* NOTE: this can only work for 0.90.0 superblocks */
3179 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
3180 if (sb
->major_version
!= 0 ||
3181 sb
->minor_version
!= 90 ) {
3182 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
3183 export_rdev(start_rdev
);
3187 if (test_bit(Faulty
, &start_rdev
->flags
)) {
3189 "md: can not autostart based on faulty %s!\n",
3190 bdevname(start_rdev
->bdev
,b
));
3191 export_rdev(start_rdev
);
3194 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
3196 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
3197 mdp_disk_t
*desc
= sb
->disks
+ i
;
3198 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
3202 if (dev
== startdev
)
3204 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
3206 rdev
= md_import_device(dev
, 0, 0);
3210 list_add(&rdev
->same_set
, &pending_raid_disks
);
3214 * possibly return codes
3222 static int get_version(void __user
* arg
)
3226 ver
.major
= MD_MAJOR_VERSION
;
3227 ver
.minor
= MD_MINOR_VERSION
;
3228 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3230 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3236 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3238 mdu_array_info_t info
;
3239 int nr
,working
,active
,failed
,spare
;
3241 struct list_head
*tmp
;
3243 nr
=working
=active
=failed
=spare
=0;
3244 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3246 if (test_bit(Faulty
, &rdev
->flags
))
3250 if (test_bit(In_sync
, &rdev
->flags
))
3257 info
.major_version
= mddev
->major_version
;
3258 info
.minor_version
= mddev
->minor_version
;
3259 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3260 info
.ctime
= mddev
->ctime
;
3261 info
.level
= mddev
->level
;
3262 info
.size
= mddev
->size
;
3263 if (info
.size
!= mddev
->size
) /* overflow */
3266 info
.raid_disks
= mddev
->raid_disks
;
3267 info
.md_minor
= mddev
->md_minor
;
3268 info
.not_persistent
= !mddev
->persistent
;
3270 info
.utime
= mddev
->utime
;
3273 info
.state
= (1<<MD_SB_CLEAN
);
3274 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3275 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3276 info
.active_disks
= active
;
3277 info
.working_disks
= working
;
3278 info
.failed_disks
= failed
;
3279 info
.spare_disks
= spare
;
3281 info
.layout
= mddev
->layout
;
3282 info
.chunk_size
= mddev
->chunk_size
;
3284 if (copy_to_user(arg
, &info
, sizeof(info
)))
3290 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3292 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3293 char *ptr
, *buf
= NULL
;
3296 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3300 /* bitmap disabled, zero the first byte and copy out */
3301 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3302 file
->pathname
[0] = '\0';
3306 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3310 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3314 strcpy(file
->pathname
, ptr
);
3318 if (copy_to_user(arg
, file
, sizeof(*file
)))
3326 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3328 mdu_disk_info_t info
;
3332 if (copy_from_user(&info
, arg
, sizeof(info
)))
3337 rdev
= find_rdev_nr(mddev
, nr
);
3339 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3340 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3341 info
.raid_disk
= rdev
->raid_disk
;
3343 if (test_bit(Faulty
, &rdev
->flags
))
3344 info
.state
|= (1<<MD_DISK_FAULTY
);
3345 else if (test_bit(In_sync
, &rdev
->flags
)) {
3346 info
.state
|= (1<<MD_DISK_ACTIVE
);
3347 info
.state
|= (1<<MD_DISK_SYNC
);
3349 if (test_bit(WriteMostly
, &rdev
->flags
))
3350 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3352 info
.major
= info
.minor
= 0;
3353 info
.raid_disk
= -1;
3354 info
.state
= (1<<MD_DISK_REMOVED
);
3357 if (copy_to_user(arg
, &info
, sizeof(info
)))
3363 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3365 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3367 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3369 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3372 if (!mddev
->raid_disks
) {
3374 /* expecting a device which has a superblock */
3375 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3378 "md: md_import_device returned %ld\n",
3380 return PTR_ERR(rdev
);
3382 if (!list_empty(&mddev
->disks
)) {
3383 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3384 mdk_rdev_t
, same_set
);
3385 int err
= super_types
[mddev
->major_version
]
3386 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3389 "md: %s has different UUID to %s\n",
3390 bdevname(rdev
->bdev
,b
),
3391 bdevname(rdev0
->bdev
,b2
));
3396 err
= bind_rdev_to_array(rdev
, mddev
);
3403 * add_new_disk can be used once the array is assembled
3404 * to add "hot spares". They must already have a superblock
3409 if (!mddev
->pers
->hot_add_disk
) {
3411 "%s: personality does not support diskops!\n",
3415 if (mddev
->persistent
)
3416 rdev
= md_import_device(dev
, mddev
->major_version
,
3417 mddev
->minor_version
);
3419 rdev
= md_import_device(dev
, -1, -1);
3422 "md: md_import_device returned %ld\n",
3424 return PTR_ERR(rdev
);
3426 /* set save_raid_disk if appropriate */
3427 if (!mddev
->persistent
) {
3428 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3429 info
->raid_disk
< mddev
->raid_disks
)
3430 rdev
->raid_disk
= info
->raid_disk
;
3432 rdev
->raid_disk
= -1;
3434 super_types
[mddev
->major_version
].
3435 validate_super(mddev
, rdev
);
3436 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3438 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3439 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3440 set_bit(WriteMostly
, &rdev
->flags
);
3442 rdev
->raid_disk
= -1;
3443 err
= bind_rdev_to_array(rdev
, mddev
);
3444 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3445 /* If there is hot_add_disk but no hot_remove_disk
3446 * then added disks for geometry changes,
3447 * and should be added immediately.
3449 super_types
[mddev
->major_version
].
3450 validate_super(mddev
, rdev
);
3451 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3453 unbind_rdev_from_array(rdev
);
3458 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3459 md_wakeup_thread(mddev
->thread
);
3463 /* otherwise, add_new_disk is only allowed
3464 * for major_version==0 superblocks
3466 if (mddev
->major_version
!= 0) {
3467 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3472 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3474 rdev
= md_import_device (dev
, -1, 0);
3477 "md: error, md_import_device() returned %ld\n",
3479 return PTR_ERR(rdev
);
3481 rdev
->desc_nr
= info
->number
;
3482 if (info
->raid_disk
< mddev
->raid_disks
)
3483 rdev
->raid_disk
= info
->raid_disk
;
3485 rdev
->raid_disk
= -1;
3489 if (rdev
->raid_disk
< mddev
->raid_disks
)
3490 if (info
->state
& (1<<MD_DISK_SYNC
))
3491 set_bit(In_sync
, &rdev
->flags
);
3493 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3494 set_bit(WriteMostly
, &rdev
->flags
);
3496 if (!mddev
->persistent
) {
3497 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3498 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3500 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3501 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3503 err
= bind_rdev_to_array(rdev
, mddev
);
3513 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3515 char b
[BDEVNAME_SIZE
];
3521 rdev
= find_rdev(mddev
, dev
);
3525 if (rdev
->raid_disk
>= 0)
3528 kick_rdev_from_array(rdev
);
3529 md_update_sb(mddev
);
3530 md_new_event(mddev
);
3534 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3535 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3539 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3541 char b
[BDEVNAME_SIZE
];
3549 if (mddev
->major_version
!= 0) {
3550 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3551 " version-0 superblocks.\n",
3555 if (!mddev
->pers
->hot_add_disk
) {
3557 "%s: personality does not support diskops!\n",
3562 rdev
= md_import_device (dev
, -1, 0);
3565 "md: error, md_import_device() returned %ld\n",
3570 if (mddev
->persistent
)
3571 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3574 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3576 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3579 if (test_bit(Faulty
, &rdev
->flags
)) {
3581 "md: can not hot-add faulty %s disk to %s!\n",
3582 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3586 clear_bit(In_sync
, &rdev
->flags
);
3588 err
= bind_rdev_to_array(rdev
, mddev
);
3593 * The rest should better be atomic, we can have disk failures
3594 * noticed in interrupt contexts ...
3597 if (rdev
->desc_nr
== mddev
->max_disks
) {
3598 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3601 goto abort_unbind_export
;
3604 rdev
->raid_disk
= -1;
3606 md_update_sb(mddev
);
3609 * Kick recovery, maybe this spare has to be added to the
3610 * array immediately.
3612 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3613 md_wakeup_thread(mddev
->thread
);
3614 md_new_event(mddev
);
3617 abort_unbind_export
:
3618 unbind_rdev_from_array(rdev
);
3625 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3630 if (!mddev
->pers
->quiesce
)
3632 if (mddev
->recovery
|| mddev
->sync_thread
)
3634 /* we should be able to change the bitmap.. */
3640 return -EEXIST
; /* cannot add when bitmap is present */
3641 mddev
->bitmap_file
= fget(fd
);
3643 if (mddev
->bitmap_file
== NULL
) {
3644 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3649 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3651 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3653 fput(mddev
->bitmap_file
);
3654 mddev
->bitmap_file
= NULL
;
3657 mddev
->bitmap_offset
= 0; /* file overrides offset */
3658 } else if (mddev
->bitmap
== NULL
)
3659 return -ENOENT
; /* cannot remove what isn't there */
3662 mddev
->pers
->quiesce(mddev
, 1);
3664 err
= bitmap_create(mddev
);
3665 if (fd
< 0 || err
) {
3666 bitmap_destroy(mddev
);
3667 fd
= -1; /* make sure to put the file */
3669 mddev
->pers
->quiesce(mddev
, 0);
3672 if (mddev
->bitmap_file
) {
3673 restore_bitmap_write_access(mddev
->bitmap_file
);
3674 fput(mddev
->bitmap_file
);
3676 mddev
->bitmap_file
= NULL
;
3683 * set_array_info is used two different ways
3684 * The original usage is when creating a new array.
3685 * In this usage, raid_disks is > 0 and it together with
3686 * level, size, not_persistent,layout,chunksize determine the
3687 * shape of the array.
3688 * This will always create an array with a type-0.90.0 superblock.
3689 * The newer usage is when assembling an array.
3690 * In this case raid_disks will be 0, and the major_version field is
3691 * use to determine which style super-blocks are to be found on the devices.
3692 * The minor and patch _version numbers are also kept incase the
3693 * super_block handler wishes to interpret them.
3695 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3698 if (info
->raid_disks
== 0) {
3699 /* just setting version number for superblock loading */
3700 if (info
->major_version
< 0 ||
3701 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3702 super_types
[info
->major_version
].name
== NULL
) {
3703 /* maybe try to auto-load a module? */
3705 "md: superblock version %d not known\n",
3706 info
->major_version
);
3709 mddev
->major_version
= info
->major_version
;
3710 mddev
->minor_version
= info
->minor_version
;
3711 mddev
->patch_version
= info
->patch_version
;
3714 mddev
->major_version
= MD_MAJOR_VERSION
;
3715 mddev
->minor_version
= MD_MINOR_VERSION
;
3716 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3717 mddev
->ctime
= get_seconds();
3719 mddev
->level
= info
->level
;
3720 mddev
->clevel
[0] = 0;
3721 mddev
->size
= info
->size
;
3722 mddev
->raid_disks
= info
->raid_disks
;
3723 /* don't set md_minor, it is determined by which /dev/md* was
3726 if (info
->state
& (1<<MD_SB_CLEAN
))
3727 mddev
->recovery_cp
= MaxSector
;
3729 mddev
->recovery_cp
= 0;
3730 mddev
->persistent
= ! info
->not_persistent
;
3732 mddev
->layout
= info
->layout
;
3733 mddev
->chunk_size
= info
->chunk_size
;
3735 mddev
->max_disks
= MD_SB_DISKS
;
3737 mddev
->sb_dirty
= 1;
3739 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3740 mddev
->bitmap_offset
= 0;
3742 mddev
->reshape_position
= MaxSector
;
3745 * Generate a 128 bit UUID
3747 get_random_bytes(mddev
->uuid
, 16);
3749 mddev
->new_level
= mddev
->level
;
3750 mddev
->new_chunk
= mddev
->chunk_size
;
3751 mddev
->new_layout
= mddev
->layout
;
3752 mddev
->delta_disks
= 0;
3757 static int update_size(mddev_t
*mddev
, unsigned long size
)
3761 struct list_head
*tmp
;
3762 int fit
= (size
== 0);
3764 if (mddev
->pers
->resize
== NULL
)
3766 /* The "size" is the amount of each device that is used.
3767 * This can only make sense for arrays with redundancy.
3768 * linear and raid0 always use whatever space is available
3769 * We can only consider changing the size if no resync
3770 * or reconstruction is happening, and if the new size
3771 * is acceptable. It must fit before the sb_offset or,
3772 * if that is <data_offset, it must fit before the
3773 * size of each device.
3774 * If size is zero, we find the largest size that fits.
3776 if (mddev
->sync_thread
)
3778 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3780 if (rdev
->sb_offset
> rdev
->data_offset
)
3781 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3783 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3784 - rdev
->data_offset
;
3785 if (fit
&& (size
== 0 || size
> avail
/2))
3787 if (avail
< ((sector_t
)size
<< 1))
3790 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3792 struct block_device
*bdev
;
3794 bdev
= bdget_disk(mddev
->gendisk
, 0);
3796 mutex_lock(&bdev
->bd_inode
->i_mutex
);
3797 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
3798 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
3805 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3808 /* change the number of raid disks */
3809 if (mddev
->pers
->check_reshape
== NULL
)
3811 if (raid_disks
<= 0 ||
3812 raid_disks
>= mddev
->max_disks
)
3814 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
3816 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
3818 rv
= mddev
->pers
->check_reshape(mddev
);
3824 * update_array_info is used to change the configuration of an
3826 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3827 * fields in the info are checked against the array.
3828 * Any differences that cannot be handled will cause an error.
3829 * Normally, only one change can be managed at a time.
3831 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3837 /* calculate expected state,ignoring low bits */
3838 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3839 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3841 if (mddev
->major_version
!= info
->major_version
||
3842 mddev
->minor_version
!= info
->minor_version
||
3843 /* mddev->patch_version != info->patch_version || */
3844 mddev
->ctime
!= info
->ctime
||
3845 mddev
->level
!= info
->level
||
3846 /* mddev->layout != info->layout || */
3847 !mddev
->persistent
!= info
->not_persistent
||
3848 mddev
->chunk_size
!= info
->chunk_size
||
3849 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3850 ((state
^info
->state
) & 0xfffffe00)
3853 /* Check there is only one change */
3854 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
3855 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3856 if (mddev
->layout
!= info
->layout
) cnt
++;
3857 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3858 if (cnt
== 0) return 0;
3859 if (cnt
> 1) return -EINVAL
;
3861 if (mddev
->layout
!= info
->layout
) {
3863 * we don't need to do anything at the md level, the
3864 * personality will take care of it all.
3866 if (mddev
->pers
->reconfig
== NULL
)
3869 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3871 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
3872 rv
= update_size(mddev
, info
->size
);
3874 if (mddev
->raid_disks
!= info
->raid_disks
)
3875 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3877 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3878 if (mddev
->pers
->quiesce
== NULL
)
3880 if (mddev
->recovery
|| mddev
->sync_thread
)
3882 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3883 /* add the bitmap */
3886 if (mddev
->default_bitmap_offset
== 0)
3888 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3889 mddev
->pers
->quiesce(mddev
, 1);
3890 rv
= bitmap_create(mddev
);
3892 bitmap_destroy(mddev
);
3893 mddev
->pers
->quiesce(mddev
, 0);
3895 /* remove the bitmap */
3898 if (mddev
->bitmap
->file
)
3900 mddev
->pers
->quiesce(mddev
, 1);
3901 bitmap_destroy(mddev
);
3902 mddev
->pers
->quiesce(mddev
, 0);
3903 mddev
->bitmap_offset
= 0;
3906 md_update_sb(mddev
);
3910 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3914 if (mddev
->pers
== NULL
)
3917 rdev
= find_rdev(mddev
, dev
);
3921 md_error(mddev
, rdev
);
3925 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
3927 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
3931 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
3935 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3936 unsigned int cmd
, unsigned long arg
)
3939 void __user
*argp
= (void __user
*)arg
;
3940 mddev_t
*mddev
= NULL
;
3942 if (!capable(CAP_SYS_ADMIN
))
3946 * Commands dealing with the RAID driver but not any
3952 err
= get_version(argp
);
3955 case PRINT_RAID_DEBUG
:
3963 autostart_arrays(arg
);
3970 * Commands creating/starting a new array:
3973 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3981 if (cmd
== START_ARRAY
) {
3982 /* START_ARRAY doesn't need to lock the array as autostart_array
3983 * does the locking, and it could even be a different array
3988 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3989 "This will not be supported beyond July 2006\n",
3990 current
->comm
, current
->pid
);
3993 err
= autostart_array(new_decode_dev(arg
));
3995 printk(KERN_WARNING
"md: autostart failed!\n");
4001 err
= mddev_lock(mddev
);
4004 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4011 case SET_ARRAY_INFO
:
4013 mdu_array_info_t info
;
4015 memset(&info
, 0, sizeof(info
));
4016 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4021 err
= update_array_info(mddev
, &info
);
4023 printk(KERN_WARNING
"md: couldn't update"
4024 " array info. %d\n", err
);
4029 if (!list_empty(&mddev
->disks
)) {
4031 "md: array %s already has disks!\n",
4036 if (mddev
->raid_disks
) {
4038 "md: array %s already initialised!\n",
4043 err
= set_array_info(mddev
, &info
);
4045 printk(KERN_WARNING
"md: couldn't set"
4046 " array info. %d\n", err
);
4056 * Commands querying/configuring an existing array:
4058 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4059 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4060 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4061 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
4067 * Commands even a read-only array can execute:
4071 case GET_ARRAY_INFO
:
4072 err
= get_array_info(mddev
, argp
);
4075 case GET_BITMAP_FILE
:
4076 err
= get_bitmap_file(mddev
, argp
);
4080 err
= get_disk_info(mddev
, argp
);
4083 case RESTART_ARRAY_RW
:
4084 err
= restart_array(mddev
);
4088 err
= do_md_stop (mddev
, 0);
4092 err
= do_md_stop (mddev
, 1);
4096 * We have a problem here : there is no easy way to give a CHS
4097 * virtual geometry. We currently pretend that we have a 2 heads
4098 * 4 sectors (with a BIG number of cylinders...). This drives
4099 * dosfs just mad... ;-)
4104 * The remaining ioctls are changing the state of the
4105 * superblock, so we do not allow them on read-only arrays.
4106 * However non-MD ioctls (e.g. get-size) will still come through
4107 * here and hit the 'default' below, so only disallow
4108 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4110 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4111 mddev
->ro
&& mddev
->pers
) {
4112 if (mddev
->ro
== 2) {
4114 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4115 md_wakeup_thread(mddev
->thread
);
4127 mdu_disk_info_t info
;
4128 if (copy_from_user(&info
, argp
, sizeof(info
)))
4131 err
= add_new_disk(mddev
, &info
);
4135 case HOT_REMOVE_DISK
:
4136 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4140 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4143 case SET_DISK_FAULTY
:
4144 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4148 err
= do_md_run (mddev
);
4151 case SET_BITMAP_FILE
:
4152 err
= set_bitmap_file(mddev
, (int)arg
);
4162 mddev_unlock(mddev
);
4172 static int md_open(struct inode
*inode
, struct file
*file
)
4175 * Succeed if we can lock the mddev, which confirms that
4176 * it isn't being stopped right now.
4178 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4181 if ((err
= mddev_lock(mddev
)))
4186 mddev_unlock(mddev
);
4188 check_disk_change(inode
->i_bdev
);
4193 static int md_release(struct inode
*inode
, struct file
* file
)
4195 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4204 static int md_media_changed(struct gendisk
*disk
)
4206 mddev_t
*mddev
= disk
->private_data
;
4208 return mddev
->changed
;
4211 static int md_revalidate(struct gendisk
*disk
)
4213 mddev_t
*mddev
= disk
->private_data
;
4218 static struct block_device_operations md_fops
=
4220 .owner
= THIS_MODULE
,
4222 .release
= md_release
,
4224 .getgeo
= md_getgeo
,
4225 .media_changed
= md_media_changed
,
4226 .revalidate_disk
= md_revalidate
,
4229 static int md_thread(void * arg
)
4231 mdk_thread_t
*thread
= arg
;
4234 * md_thread is a 'system-thread', it's priority should be very
4235 * high. We avoid resource deadlocks individually in each
4236 * raid personality. (RAID5 does preallocation) We also use RR and
4237 * the very same RT priority as kswapd, thus we will never get
4238 * into a priority inversion deadlock.
4240 * we definitely have to have equal or higher priority than
4241 * bdflush, otherwise bdflush will deadlock if there are too
4242 * many dirty RAID5 blocks.
4245 allow_signal(SIGKILL
);
4246 while (!kthread_should_stop()) {
4248 /* We need to wait INTERRUPTIBLE so that
4249 * we don't add to the load-average.
4250 * That means we need to be sure no signals are
4253 if (signal_pending(current
))
4254 flush_signals(current
);
4256 wait_event_interruptible_timeout
4258 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4259 || kthread_should_stop(),
4263 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4265 thread
->run(thread
->mddev
);
4271 void md_wakeup_thread(mdk_thread_t
*thread
)
4274 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4275 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4276 wake_up(&thread
->wqueue
);
4280 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4283 mdk_thread_t
*thread
;
4285 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4289 init_waitqueue_head(&thread
->wqueue
);
4292 thread
->mddev
= mddev
;
4293 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4294 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4295 if (IS_ERR(thread
->tsk
)) {
4302 void md_unregister_thread(mdk_thread_t
*thread
)
4304 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4306 kthread_stop(thread
->tsk
);
4310 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4317 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4320 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4322 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4323 __builtin_return_address(0),__builtin_return_address(1),
4324 __builtin_return_address(2),__builtin_return_address(3));
4326 if (!mddev
->pers
->error_handler
)
4328 mddev
->pers
->error_handler(mddev
,rdev
);
4329 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4330 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4331 md_wakeup_thread(mddev
->thread
);
4332 md_new_event_inintr(mddev
);
4335 /* seq_file implementation /proc/mdstat */
4337 static void status_unused(struct seq_file
*seq
)
4341 struct list_head
*tmp
;
4343 seq_printf(seq
, "unused devices: ");
4345 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4346 char b
[BDEVNAME_SIZE
];
4348 seq_printf(seq
, "%s ",
4349 bdevname(rdev
->bdev
,b
));
4352 seq_printf(seq
, "<none>");
4354 seq_printf(seq
, "\n");
4358 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4360 sector_t max_blocks
, resync
, res
;
4361 unsigned long dt
, db
, rt
;
4363 unsigned int per_milli
;
4365 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4367 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4368 max_blocks
= mddev
->resync_max_sectors
>> 1;
4370 max_blocks
= mddev
->size
;
4373 * Should not happen.
4379 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4380 * in a sector_t, and (max_blocks>>scale) will fit in a
4381 * u32, as those are the requirements for sector_div.
4382 * Thus 'scale' must be at least 10
4385 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4386 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4389 res
= (resync
>>scale
)*1000;
4390 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4394 int i
, x
= per_milli
/50, y
= 20-x
;
4395 seq_printf(seq
, "[");
4396 for (i
= 0; i
< x
; i
++)
4397 seq_printf(seq
, "=");
4398 seq_printf(seq
, ">");
4399 for (i
= 0; i
< y
; i
++)
4400 seq_printf(seq
, ".");
4401 seq_printf(seq
, "] ");
4403 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4404 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4406 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4407 "resync" : "recovery")),
4408 per_milli
/10, per_milli
% 10,
4409 (unsigned long long) resync
,
4410 (unsigned long long) max_blocks
);
4413 * We do not want to overflow, so the order of operands and
4414 * the * 100 / 100 trick are important. We do a +1 to be
4415 * safe against division by zero. We only estimate anyway.
4417 * dt: time from mark until now
4418 * db: blocks written from mark until now
4419 * rt: remaining time
4421 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4423 db
= resync
- (mddev
->resync_mark_cnt
/2);
4424 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/100+1)))/100;
4426 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4428 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4431 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4433 struct list_head
*tmp
;
4443 spin_lock(&all_mddevs_lock
);
4444 list_for_each(tmp
,&all_mddevs
)
4446 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4448 spin_unlock(&all_mddevs_lock
);
4451 spin_unlock(&all_mddevs_lock
);
4453 return (void*)2;/* tail */
4457 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4459 struct list_head
*tmp
;
4460 mddev_t
*next_mddev
, *mddev
= v
;
4466 spin_lock(&all_mddevs_lock
);
4468 tmp
= all_mddevs
.next
;
4470 tmp
= mddev
->all_mddevs
.next
;
4471 if (tmp
!= &all_mddevs
)
4472 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4474 next_mddev
= (void*)2;
4477 spin_unlock(&all_mddevs_lock
);
4485 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4489 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4493 struct mdstat_info
{
4497 static int md_seq_show(struct seq_file
*seq
, void *v
)
4501 struct list_head
*tmp2
;
4503 struct mdstat_info
*mi
= seq
->private;
4504 struct bitmap
*bitmap
;
4506 if (v
== (void*)1) {
4507 struct mdk_personality
*pers
;
4508 seq_printf(seq
, "Personalities : ");
4509 spin_lock(&pers_lock
);
4510 list_for_each_entry(pers
, &pers_list
, list
)
4511 seq_printf(seq
, "[%s] ", pers
->name
);
4513 spin_unlock(&pers_lock
);
4514 seq_printf(seq
, "\n");
4515 mi
->event
= atomic_read(&md_event_count
);
4518 if (v
== (void*)2) {
4523 if (mddev_lock(mddev
) < 0)
4526 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4527 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4528 mddev
->pers
? "" : "in");
4531 seq_printf(seq
, " (read-only)");
4533 seq_printf(seq
, "(auto-read-only)");
4534 seq_printf(seq
, " %s", mddev
->pers
->name
);
4538 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4539 char b
[BDEVNAME_SIZE
];
4540 seq_printf(seq
, " %s[%d]",
4541 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4542 if (test_bit(WriteMostly
, &rdev
->flags
))
4543 seq_printf(seq
, "(W)");
4544 if (test_bit(Faulty
, &rdev
->flags
)) {
4545 seq_printf(seq
, "(F)");
4547 } else if (rdev
->raid_disk
< 0)
4548 seq_printf(seq
, "(S)"); /* spare */
4552 if (!list_empty(&mddev
->disks
)) {
4554 seq_printf(seq
, "\n %llu blocks",
4555 (unsigned long long)mddev
->array_size
);
4557 seq_printf(seq
, "\n %llu blocks",
4558 (unsigned long long)size
);
4560 if (mddev
->persistent
) {
4561 if (mddev
->major_version
!= 0 ||
4562 mddev
->minor_version
!= 90) {
4563 seq_printf(seq
," super %d.%d",
4564 mddev
->major_version
,
4565 mddev
->minor_version
);
4568 seq_printf(seq
, " super non-persistent");
4571 mddev
->pers
->status (seq
, mddev
);
4572 seq_printf(seq
, "\n ");
4573 if (mddev
->pers
->sync_request
) {
4574 if (mddev
->curr_resync
> 2) {
4575 status_resync (seq
, mddev
);
4576 seq_printf(seq
, "\n ");
4577 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4578 seq_printf(seq
, "\tresync=DELAYED\n ");
4579 else if (mddev
->recovery_cp
< MaxSector
)
4580 seq_printf(seq
, "\tresync=PENDING\n ");
4583 seq_printf(seq
, "\n ");
4585 if ((bitmap
= mddev
->bitmap
)) {
4586 unsigned long chunk_kb
;
4587 unsigned long flags
;
4588 spin_lock_irqsave(&bitmap
->lock
, flags
);
4589 chunk_kb
= bitmap
->chunksize
>> 10;
4590 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4592 bitmap
->pages
- bitmap
->missing_pages
,
4594 (bitmap
->pages
- bitmap
->missing_pages
)
4595 << (PAGE_SHIFT
- 10),
4596 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4597 chunk_kb
? "KB" : "B");
4599 seq_printf(seq
, ", file: ");
4600 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4601 bitmap
->file
->f_dentry
," \t\n");
4604 seq_printf(seq
, "\n");
4605 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4608 seq_printf(seq
, "\n");
4610 mddev_unlock(mddev
);
4615 static struct seq_operations md_seq_ops
= {
4616 .start
= md_seq_start
,
4617 .next
= md_seq_next
,
4618 .stop
= md_seq_stop
,
4619 .show
= md_seq_show
,
4622 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4625 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4629 error
= seq_open(file
, &md_seq_ops
);
4633 struct seq_file
*p
= file
->private_data
;
4635 mi
->event
= atomic_read(&md_event_count
);
4640 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4642 struct seq_file
*m
= file
->private_data
;
4643 struct mdstat_info
*mi
= m
->private;
4646 return seq_release(inode
, file
);
4649 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4651 struct seq_file
*m
= filp
->private_data
;
4652 struct mdstat_info
*mi
= m
->private;
4655 poll_wait(filp
, &md_event_waiters
, wait
);
4657 /* always allow read */
4658 mask
= POLLIN
| POLLRDNORM
;
4660 if (mi
->event
!= atomic_read(&md_event_count
))
4661 mask
|= POLLERR
| POLLPRI
;
4665 static struct file_operations md_seq_fops
= {
4666 .open
= md_seq_open
,
4668 .llseek
= seq_lseek
,
4669 .release
= md_seq_release
,
4670 .poll
= mdstat_poll
,
4673 int register_md_personality(struct mdk_personality
*p
)
4675 spin_lock(&pers_lock
);
4676 list_add_tail(&p
->list
, &pers_list
);
4677 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4678 spin_unlock(&pers_lock
);
4682 int unregister_md_personality(struct mdk_personality
*p
)
4684 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4685 spin_lock(&pers_lock
);
4686 list_del_init(&p
->list
);
4687 spin_unlock(&pers_lock
);
4691 static int is_mddev_idle(mddev_t
*mddev
)
4694 struct list_head
*tmp
;
4696 unsigned long curr_events
;
4699 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4700 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4701 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4702 disk_stat_read(disk
, sectors
[1]) -
4703 atomic_read(&disk
->sync_io
);
4704 /* The difference between curr_events and last_events
4705 * will be affected by any new non-sync IO (making
4706 * curr_events bigger) and any difference in the amount of
4707 * in-flight syncio (making current_events bigger or smaller)
4708 * The amount in-flight is currently limited to
4709 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4710 * which is at most 4096 sectors.
4711 * These numbers are fairly fragile and should be made
4712 * more robust, probably by enforcing the
4713 * 'window size' that md_do_sync sort-of uses.
4715 * Note: the following is an unsigned comparison.
4717 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4718 rdev
->last_events
= curr_events
;
4725 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4727 /* another "blocks" (512byte) blocks have been synced */
4728 atomic_sub(blocks
, &mddev
->recovery_active
);
4729 wake_up(&mddev
->recovery_wait
);
4731 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4732 md_wakeup_thread(mddev
->thread
);
4733 // stop recovery, signal do_sync ....
4738 /* md_write_start(mddev, bi)
4739 * If we need to update some array metadata (e.g. 'active' flag
4740 * in superblock) before writing, schedule a superblock update
4741 * and wait for it to complete.
4743 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4745 if (bio_data_dir(bi
) != WRITE
)
4748 BUG_ON(mddev
->ro
== 1);
4749 if (mddev
->ro
== 2) {
4750 /* need to switch to read/write */
4752 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4753 md_wakeup_thread(mddev
->thread
);
4755 atomic_inc(&mddev
->writes_pending
);
4756 if (mddev
->in_sync
) {
4757 spin_lock_irq(&mddev
->write_lock
);
4758 if (mddev
->in_sync
) {
4760 mddev
->sb_dirty
= 3;
4761 md_wakeup_thread(mddev
->thread
);
4763 spin_unlock_irq(&mddev
->write_lock
);
4765 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4768 void md_write_end(mddev_t
*mddev
)
4770 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4771 if (mddev
->safemode
== 2)
4772 md_wakeup_thread(mddev
->thread
);
4773 else if (mddev
->safemode_delay
)
4774 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4778 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4780 #define SYNC_MARKS 10
4781 #define SYNC_MARK_STEP (3*HZ)
4782 void md_do_sync(mddev_t
*mddev
)
4785 unsigned int currspeed
= 0,
4787 sector_t max_sectors
,j
, io_sectors
;
4788 unsigned long mark
[SYNC_MARKS
];
4789 sector_t mark_cnt
[SYNC_MARKS
];
4791 struct list_head
*tmp
;
4792 sector_t last_check
;
4794 struct list_head
*rtmp
;
4797 /* just incase thread restarts... */
4798 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4800 if (mddev
->ro
) /* never try to sync a read-only array */
4803 /* we overload curr_resync somewhat here.
4804 * 0 == not engaged in resync at all
4805 * 2 == checking that there is no conflict with another sync
4806 * 1 == like 2, but have yielded to allow conflicting resync to
4808 * other == active in resync - this many blocks
4810 * Before starting a resync we must have set curr_resync to
4811 * 2, and then checked that every "conflicting" array has curr_resync
4812 * less than ours. When we find one that is the same or higher
4813 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4814 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4815 * This will mean we have to start checking from the beginning again.
4820 mddev
->curr_resync
= 2;
4823 if (kthread_should_stop()) {
4824 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4827 ITERATE_MDDEV(mddev2
,tmp
) {
4828 if (mddev2
== mddev
)
4830 if (mddev2
->curr_resync
&&
4831 match_mddev_units(mddev
,mddev2
)) {
4833 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4834 /* arbitrarily yield */
4835 mddev
->curr_resync
= 1;
4836 wake_up(&resync_wait
);
4838 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4839 /* no need to wait here, we can wait the next
4840 * time 'round when curr_resync == 2
4843 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4844 if (!kthread_should_stop() &&
4845 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4846 printk(KERN_INFO
"md: delaying resync of %s"
4847 " until %s has finished resync (they"
4848 " share one or more physical units)\n",
4849 mdname(mddev
), mdname(mddev2
));
4852 finish_wait(&resync_wait
, &wq
);
4855 finish_wait(&resync_wait
, &wq
);
4858 } while (mddev
->curr_resync
< 2);
4861 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4862 /* resync follows the size requested by the personality,
4863 * which defaults to physical size, but can be virtual size
4865 max_sectors
= mddev
->resync_max_sectors
;
4866 mddev
->resync_mismatches
= 0;
4867 /* we don't use the checkpoint if there's a bitmap */
4868 if (!mddev
->bitmap
&&
4869 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4870 j
= mddev
->recovery_cp
;
4871 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4872 max_sectors
= mddev
->size
<< 1;
4874 /* recovery follows the physical size of devices */
4875 max_sectors
= mddev
->size
<< 1;
4877 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4878 if (rdev
->raid_disk
>= 0 &&
4879 !test_bit(Faulty
, &rdev
->flags
) &&
4880 !test_bit(In_sync
, &rdev
->flags
) &&
4881 rdev
->recovery_offset
< j
)
4882 j
= rdev
->recovery_offset
;
4885 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4886 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4887 " %d KB/sec/disc.\n", speed_min(mddev
));
4888 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4889 "(but not more than %d KB/sec) for reconstruction.\n",
4892 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4895 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4897 mark_cnt
[m
] = io_sectors
;
4900 mddev
->resync_mark
= mark
[last_mark
];
4901 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4904 * Tune reconstruction:
4906 window
= 32*(PAGE_SIZE
/512);
4907 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4908 window
/2,(unsigned long long) max_sectors
/2);
4910 atomic_set(&mddev
->recovery_active
, 0);
4911 init_waitqueue_head(&mddev
->recovery_wait
);
4916 "md: resuming recovery of %s from checkpoint.\n",
4918 mddev
->curr_resync
= j
;
4921 while (j
< max_sectors
) {
4925 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4926 currspeed
< speed_min(mddev
));
4928 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4932 if (!skipped
) { /* actual IO requested */
4933 io_sectors
+= sectors
;
4934 atomic_add(sectors
, &mddev
->recovery_active
);
4938 if (j
>1) mddev
->curr_resync
= j
;
4939 if (last_check
== 0)
4940 /* this is the earliers that rebuilt will be
4941 * visible in /proc/mdstat
4943 md_new_event(mddev
);
4945 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4948 last_check
= io_sectors
;
4950 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4951 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4955 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4957 int next
= (last_mark
+1) % SYNC_MARKS
;
4959 mddev
->resync_mark
= mark
[next
];
4960 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4961 mark
[next
] = jiffies
;
4962 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4967 if (kthread_should_stop()) {
4969 * got a signal, exit.
4972 "md: md_do_sync() got signal ... exiting\n");
4973 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4978 * this loop exits only if either when we are slower than
4979 * the 'hard' speed limit, or the system was IO-idle for
4981 * the system might be non-idle CPU-wise, but we only care
4982 * about not overloading the IO subsystem. (things like an
4983 * e2fsck being done on the RAID array should execute fast)
4985 mddev
->queue
->unplug_fn(mddev
->queue
);
4988 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4989 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4991 if (currspeed
> speed_min(mddev
)) {
4992 if ((currspeed
> speed_max(mddev
)) ||
4993 !is_mddev_idle(mddev
)) {
4999 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
5001 * this also signals 'finished resyncing' to md_stop
5004 mddev
->queue
->unplug_fn(mddev
->queue
);
5006 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5008 /* tell personality that we are finished */
5009 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5011 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5012 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
5013 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5014 mddev
->curr_resync
> 2) {
5015 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5016 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5017 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5019 "md: checkpointing recovery of %s.\n",
5021 mddev
->recovery_cp
= mddev
->curr_resync
;
5024 mddev
->recovery_cp
= MaxSector
;
5026 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5027 mddev
->curr_resync
= MaxSector
;
5028 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5029 if (rdev
->raid_disk
>= 0 &&
5030 !test_bit(Faulty
, &rdev
->flags
) &&
5031 !test_bit(In_sync
, &rdev
->flags
) &&
5032 rdev
->recovery_offset
< mddev
->curr_resync
)
5033 rdev
->recovery_offset
= mddev
->curr_resync
;
5034 mddev
->sb_dirty
= 1;
5039 mddev
->curr_resync
= 0;
5040 wake_up(&resync_wait
);
5041 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5042 md_wakeup_thread(mddev
->thread
);
5044 EXPORT_SYMBOL_GPL(md_do_sync
);
5048 * This routine is regularly called by all per-raid-array threads to
5049 * deal with generic issues like resync and super-block update.
5050 * Raid personalities that don't have a thread (linear/raid0) do not
5051 * need this as they never do any recovery or update the superblock.
5053 * It does not do any resync itself, but rather "forks" off other threads
5054 * to do that as needed.
5055 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5056 * "->recovery" and create a thread at ->sync_thread.
5057 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5058 * and wakeups up this thread which will reap the thread and finish up.
5059 * This thread also removes any faulty devices (with nr_pending == 0).
5061 * The overall approach is:
5062 * 1/ if the superblock needs updating, update it.
5063 * 2/ If a recovery thread is running, don't do anything else.
5064 * 3/ If recovery has finished, clean up, possibly marking spares active.
5065 * 4/ If there are any faulty devices, remove them.
5066 * 5/ If array is degraded, try to add spares devices
5067 * 6/ If array has spares or is not in-sync, start a resync thread.
5069 void md_check_recovery(mddev_t
*mddev
)
5072 struct list_head
*rtmp
;
5076 bitmap_daemon_work(mddev
->bitmap
);
5081 if (signal_pending(current
)) {
5082 if (mddev
->pers
->sync_request
) {
5083 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5085 mddev
->safemode
= 2;
5087 flush_signals(current
);
5092 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5093 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5094 (mddev
->safemode
== 1) ||
5095 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5096 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5100 if (mddev_trylock(mddev
)) {
5103 spin_lock_irq(&mddev
->write_lock
);
5104 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5105 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5107 mddev
->sb_dirty
= 3;
5109 if (mddev
->safemode
== 1)
5110 mddev
->safemode
= 0;
5111 spin_unlock_irq(&mddev
->write_lock
);
5113 if (mddev
->sb_dirty
)
5114 md_update_sb(mddev
);
5117 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5118 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5119 /* resync/recovery still happening */
5120 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5123 if (mddev
->sync_thread
) {
5124 /* resync has finished, collect result */
5125 md_unregister_thread(mddev
->sync_thread
);
5126 mddev
->sync_thread
= NULL
;
5127 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5128 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5130 /* activate any spares */
5131 mddev
->pers
->spare_active(mddev
);
5133 md_update_sb(mddev
);
5135 /* if array is no-longer degraded, then any saved_raid_disk
5136 * information must be scrapped
5138 if (!mddev
->degraded
)
5139 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5140 rdev
->saved_raid_disk
= -1;
5142 mddev
->recovery
= 0;
5143 /* flag recovery needed just to double check */
5144 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5145 md_new_event(mddev
);
5148 /* Clear some bits that don't mean anything, but
5151 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5152 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5153 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5154 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5156 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5158 /* no recovery is running.
5159 * remove any failed drives, then
5160 * add spares if possible.
5161 * Spare are also removed and re-added, to allow
5162 * the personality to fail the re-add.
5164 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5165 if (rdev
->raid_disk
>= 0 &&
5166 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
5167 atomic_read(&rdev
->nr_pending
)==0) {
5168 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
5170 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5171 sysfs_remove_link(&mddev
->kobj
, nm
);
5172 rdev
->raid_disk
= -1;
5176 if (mddev
->degraded
) {
5177 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5178 if (rdev
->raid_disk
< 0
5179 && !test_bit(Faulty
, &rdev
->flags
)) {
5180 rdev
->recovery_offset
= 0;
5181 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5183 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5184 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
5186 md_new_event(mddev
);
5193 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5194 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5195 } else if (mddev
->recovery_cp
< MaxSector
) {
5196 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5197 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5198 /* nothing to be done ... */
5201 if (mddev
->pers
->sync_request
) {
5202 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5203 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5204 /* We are adding a device or devices to an array
5205 * which has the bitmap stored on all devices.
5206 * So make sure all bitmap pages get written
5208 bitmap_write_all(mddev
->bitmap
);
5210 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5213 if (!mddev
->sync_thread
) {
5214 printk(KERN_ERR
"%s: could not start resync"
5217 /* leave the spares where they are, it shouldn't hurt */
5218 mddev
->recovery
= 0;
5220 md_wakeup_thread(mddev
->sync_thread
);
5221 md_new_event(mddev
);
5224 mddev_unlock(mddev
);
5228 static int md_notify_reboot(struct notifier_block
*this,
5229 unsigned long code
, void *x
)
5231 struct list_head
*tmp
;
5234 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5236 printk(KERN_INFO
"md: stopping all md devices.\n");
5238 ITERATE_MDDEV(mddev
,tmp
)
5239 if (mddev_trylock(mddev
)) {
5240 do_md_stop (mddev
, 1);
5241 mddev_unlock(mddev
);
5244 * certain more exotic SCSI devices are known to be
5245 * volatile wrt too early system reboots. While the
5246 * right place to handle this issue is the given
5247 * driver, we do want to have a safe RAID driver ...
5254 static struct notifier_block md_notifier
= {
5255 .notifier_call
= md_notify_reboot
,
5257 .priority
= INT_MAX
, /* before any real devices */
5260 static void md_geninit(void)
5262 struct proc_dir_entry
*p
;
5264 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5266 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5268 p
->proc_fops
= &md_seq_fops
;
5271 static int __init
md_init(void)
5275 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5276 " MD_SB_DISKS=%d\n",
5277 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
5278 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
5279 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
5282 if (register_blkdev(MAJOR_NR
, "md"))
5284 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5285 unregister_blkdev(MAJOR_NR
, "md");
5289 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
5290 md_probe
, NULL
, NULL
);
5291 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
5292 md_probe
, NULL
, NULL
);
5294 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5295 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
5296 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5299 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5300 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
5301 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5305 register_reboot_notifier(&md_notifier
);
5306 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
5316 * Searches all registered partitions for autorun RAID arrays
5319 static dev_t detected_devices
[128];
5322 void md_autodetect_dev(dev_t dev
)
5324 if (dev_cnt
>= 0 && dev_cnt
< 127)
5325 detected_devices
[dev_cnt
++] = dev
;
5329 static void autostart_arrays(int part
)
5334 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5336 for (i
= 0; i
< dev_cnt
; i
++) {
5337 dev_t dev
= detected_devices
[i
];
5339 rdev
= md_import_device(dev
,0, 0);
5343 if (test_bit(Faulty
, &rdev
->flags
)) {
5347 list_add(&rdev
->same_set
, &pending_raid_disks
);
5351 autorun_devices(part
);
5356 static __exit
void md_exit(void)
5359 struct list_head
*tmp
;
5361 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
5362 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
5363 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5364 devfs_remove("md/%d", i
);
5365 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5366 devfs_remove("md/d%d", i
);
5370 unregister_blkdev(MAJOR_NR
,"md");
5371 unregister_blkdev(mdp_major
, "mdp");
5372 unregister_reboot_notifier(&md_notifier
);
5373 unregister_sysctl_table(raid_table_header
);
5374 remove_proc_entry("mdstat", NULL
);
5375 ITERATE_MDDEV(mddev
,tmp
) {
5376 struct gendisk
*disk
= mddev
->gendisk
;
5379 export_array(mddev
);
5382 mddev
->gendisk
= NULL
;
5387 module_init(md_init
)
5388 module_exit(md_exit
)
5390 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5392 return sprintf(buffer
, "%d", start_readonly
);
5394 static int set_ro(const char *val
, struct kernel_param
*kp
)
5397 int num
= simple_strtoul(val
, &e
, 10);
5398 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5399 start_readonly
= num
;
5405 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5406 module_param(start_dirty_degraded
, int, 0644);
5409 EXPORT_SYMBOL(register_md_personality
);
5410 EXPORT_SYMBOL(unregister_md_personality
);
5411 EXPORT_SYMBOL(md_error
);
5412 EXPORT_SYMBOL(md_done_sync
);
5413 EXPORT_SYMBOL(md_write_start
);
5414 EXPORT_SYMBOL(md_write_end
);
5415 EXPORT_SYMBOL(md_register_thread
);
5416 EXPORT_SYMBOL(md_unregister_thread
);
5417 EXPORT_SYMBOL(md_wakeup_thread
);
5418 EXPORT_SYMBOL(md_check_recovery
);
5419 MODULE_LICENSE("GPL");
5421 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);