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
);
741 rdev
->raid_disk
= -1;
743 if (mddev
->raid_disks
== 0) {
744 mddev
->major_version
= 0;
745 mddev
->minor_version
= sb
->minor_version
;
746 mddev
->patch_version
= sb
->patch_version
;
747 mddev
->persistent
= ! sb
->not_persistent
;
748 mddev
->chunk_size
= sb
->chunk_size
;
749 mddev
->ctime
= sb
->ctime
;
750 mddev
->utime
= sb
->utime
;
751 mddev
->level
= sb
->level
;
752 mddev
->clevel
[0] = 0;
753 mddev
->layout
= sb
->layout
;
754 mddev
->raid_disks
= sb
->raid_disks
;
755 mddev
->size
= sb
->size
;
756 mddev
->events
= md_event(sb
);
757 mddev
->bitmap_offset
= 0;
758 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
760 if (mddev
->minor_version
>= 91) {
761 mddev
->reshape_position
= sb
->reshape_position
;
762 mddev
->delta_disks
= sb
->delta_disks
;
763 mddev
->new_level
= sb
->new_level
;
764 mddev
->new_layout
= sb
->new_layout
;
765 mddev
->new_chunk
= sb
->new_chunk
;
767 mddev
->reshape_position
= MaxSector
;
768 mddev
->delta_disks
= 0;
769 mddev
->new_level
= mddev
->level
;
770 mddev
->new_layout
= mddev
->layout
;
771 mddev
->new_chunk
= mddev
->chunk_size
;
774 if (sb
->state
& (1<<MD_SB_CLEAN
))
775 mddev
->recovery_cp
= MaxSector
;
777 if (sb
->events_hi
== sb
->cp_events_hi
&&
778 sb
->events_lo
== sb
->cp_events_lo
) {
779 mddev
->recovery_cp
= sb
->recovery_cp
;
781 mddev
->recovery_cp
= 0;
784 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
785 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
786 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
787 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
789 mddev
->max_disks
= MD_SB_DISKS
;
791 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
792 mddev
->bitmap_file
== NULL
) {
793 if (mddev
->level
!= 1 && mddev
->level
!= 4
794 && mddev
->level
!= 5 && mddev
->level
!= 6
795 && mddev
->level
!= 10) {
796 /* FIXME use a better test */
797 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
800 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
803 } else if (mddev
->pers
== NULL
) {
804 /* Insist on good event counter while assembling */
805 __u64 ev1
= md_event(sb
);
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 __u64 ev1
= md_event(sb
);
814 if (ev1
< mddev
->bitmap
->events_cleared
)
816 } else /* just a hot-add of a new device, leave raid_disk at -1 */
819 if (mddev
->level
!= LEVEL_MULTIPATH
) {
820 desc
= sb
->disks
+ rdev
->desc_nr
;
822 if (desc
->state
& (1<<MD_DISK_FAULTY
))
823 set_bit(Faulty
, &rdev
->flags
);
824 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
825 desc->raid_disk < mddev->raid_disks */) {
826 set_bit(In_sync
, &rdev
->flags
);
827 rdev
->raid_disk
= desc
->raid_disk
;
829 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
830 set_bit(WriteMostly
, &rdev
->flags
);
831 } else /* MULTIPATH are always insync */
832 set_bit(In_sync
, &rdev
->flags
);
837 * sync_super for 0.90.0
839 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
842 struct list_head
*tmp
;
844 int next_spare
= mddev
->raid_disks
;
847 /* make rdev->sb match mddev data..
850 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
851 * 3/ any empty disks < next_spare become removed
853 * disks[0] gets initialised to REMOVED because
854 * we cannot be sure from other fields if it has
855 * been initialised or not.
858 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
860 rdev
->sb_size
= MD_SB_BYTES
;
862 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
864 memset(sb
, 0, sizeof(*sb
));
866 sb
->md_magic
= MD_SB_MAGIC
;
867 sb
->major_version
= mddev
->major_version
;
868 sb
->patch_version
= mddev
->patch_version
;
869 sb
->gvalid_words
= 0; /* ignored */
870 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
871 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
872 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
873 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
875 sb
->ctime
= mddev
->ctime
;
876 sb
->level
= mddev
->level
;
877 sb
->size
= mddev
->size
;
878 sb
->raid_disks
= mddev
->raid_disks
;
879 sb
->md_minor
= mddev
->md_minor
;
880 sb
->not_persistent
= !mddev
->persistent
;
881 sb
->utime
= mddev
->utime
;
883 sb
->events_hi
= (mddev
->events
>>32);
884 sb
->events_lo
= (u32
)mddev
->events
;
886 if (mddev
->reshape_position
== MaxSector
)
887 sb
->minor_version
= 90;
889 sb
->minor_version
= 91;
890 sb
->reshape_position
= mddev
->reshape_position
;
891 sb
->new_level
= mddev
->new_level
;
892 sb
->delta_disks
= mddev
->delta_disks
;
893 sb
->new_layout
= mddev
->new_layout
;
894 sb
->new_chunk
= mddev
->new_chunk
;
896 mddev
->minor_version
= sb
->minor_version
;
899 sb
->recovery_cp
= mddev
->recovery_cp
;
900 sb
->cp_events_hi
= (mddev
->events
>>32);
901 sb
->cp_events_lo
= (u32
)mddev
->events
;
902 if (mddev
->recovery_cp
== MaxSector
)
903 sb
->state
= (1<< MD_SB_CLEAN
);
907 sb
->layout
= mddev
->layout
;
908 sb
->chunk_size
= mddev
->chunk_size
;
910 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
911 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
913 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
914 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
917 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
918 && !test_bit(Faulty
, &rdev2
->flags
))
919 desc_nr
= rdev2
->raid_disk
;
921 desc_nr
= next_spare
++;
922 rdev2
->desc_nr
= desc_nr
;
923 d
= &sb
->disks
[rdev2
->desc_nr
];
925 d
->number
= rdev2
->desc_nr
;
926 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
927 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
928 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
929 && !test_bit(Faulty
, &rdev2
->flags
))
930 d
->raid_disk
= rdev2
->raid_disk
;
932 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
933 if (test_bit(Faulty
, &rdev2
->flags
))
934 d
->state
= (1<<MD_DISK_FAULTY
);
935 else if (test_bit(In_sync
, &rdev2
->flags
)) {
936 d
->state
= (1<<MD_DISK_ACTIVE
);
937 d
->state
|= (1<<MD_DISK_SYNC
);
945 if (test_bit(WriteMostly
, &rdev2
->flags
))
946 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
948 /* now set the "removed" and "faulty" bits on any missing devices */
949 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
950 mdp_disk_t
*d
= &sb
->disks
[i
];
951 if (d
->state
== 0 && d
->number
== 0) {
954 d
->state
= (1<<MD_DISK_REMOVED
);
955 d
->state
|= (1<<MD_DISK_FAULTY
);
959 sb
->nr_disks
= nr_disks
;
960 sb
->active_disks
= active
;
961 sb
->working_disks
= working
;
962 sb
->failed_disks
= failed
;
963 sb
->spare_disks
= spare
;
965 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
966 sb
->sb_csum
= calc_sb_csum(sb
);
970 * version 1 superblock
973 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
975 unsigned int disk_csum
, csum
;
976 unsigned long long newcsum
;
977 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
978 unsigned int *isuper
= (unsigned int*)sb
;
981 disk_csum
= sb
->sb_csum
;
984 for (i
=0; size
>=4; size
-= 4 )
985 newcsum
+= le32_to_cpu(*isuper
++);
988 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
990 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
991 sb
->sb_csum
= disk_csum
;
992 return cpu_to_le32(csum
);
995 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
997 struct mdp_superblock_1
*sb
;
1000 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1004 * Calculate the position of the superblock.
1005 * It is always aligned to a 4K boundary and
1006 * depeding on minor_version, it can be:
1007 * 0: At least 8K, but less than 12K, from end of device
1008 * 1: At start of device
1009 * 2: 4K from start of device.
1011 switch(minor_version
) {
1013 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1015 sb_offset
&= ~(sector_t
)(4*2-1);
1016 /* convert from sectors to K */
1028 rdev
->sb_offset
= sb_offset
;
1030 /* superblock is rarely larger than 1K, but it can be larger,
1031 * and it is safe to read 4k, so we do that
1033 ret
= read_disk_sb(rdev
, 4096);
1034 if (ret
) return ret
;
1037 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1039 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1040 sb
->major_version
!= cpu_to_le32(1) ||
1041 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1042 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1043 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1046 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1047 printk("md: invalid superblock checksum on %s\n",
1048 bdevname(rdev
->bdev
,b
));
1051 if (le64_to_cpu(sb
->data_size
) < 10) {
1052 printk("md: data_size too small on %s\n",
1053 bdevname(rdev
->bdev
,b
));
1056 rdev
->preferred_minor
= 0xffff;
1057 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1058 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1060 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1061 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1062 if (rdev
->sb_size
& bmask
)
1063 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1069 struct mdp_superblock_1
*refsb
=
1070 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1072 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1073 sb
->level
!= refsb
->level
||
1074 sb
->layout
!= refsb
->layout
||
1075 sb
->chunksize
!= refsb
->chunksize
) {
1076 printk(KERN_WARNING
"md: %s has strangely different"
1077 " superblock to %s\n",
1078 bdevname(rdev
->bdev
,b
),
1079 bdevname(refdev
->bdev
,b2
));
1082 ev1
= le64_to_cpu(sb
->events
);
1083 ev2
= le64_to_cpu(refsb
->events
);
1091 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1093 rdev
->size
= rdev
->sb_offset
;
1094 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1096 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1097 if (le32_to_cpu(sb
->chunksize
))
1098 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1100 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1105 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1107 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1109 rdev
->raid_disk
= -1;
1111 if (mddev
->raid_disks
== 0) {
1112 mddev
->major_version
= 1;
1113 mddev
->patch_version
= 0;
1114 mddev
->persistent
= 1;
1115 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1116 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1117 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1118 mddev
->level
= le32_to_cpu(sb
->level
);
1119 mddev
->clevel
[0] = 0;
1120 mddev
->layout
= le32_to_cpu(sb
->layout
);
1121 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1122 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1123 mddev
->events
= le64_to_cpu(sb
->events
);
1124 mddev
->bitmap_offset
= 0;
1125 mddev
->default_bitmap_offset
= 1024 >> 9;
1127 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1128 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1130 mddev
->max_disks
= (4096-256)/2;
1132 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1133 mddev
->bitmap_file
== NULL
) {
1134 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1135 && mddev
->level
!= 10) {
1136 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1139 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1141 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1142 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1143 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1144 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1145 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1146 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1148 mddev
->reshape_position
= MaxSector
;
1149 mddev
->delta_disks
= 0;
1150 mddev
->new_level
= mddev
->level
;
1151 mddev
->new_layout
= mddev
->layout
;
1152 mddev
->new_chunk
= mddev
->chunk_size
;
1155 } else if (mddev
->pers
== NULL
) {
1156 /* Insist of good event counter while assembling */
1157 __u64 ev1
= le64_to_cpu(sb
->events
);
1159 if (ev1
< mddev
->events
)
1161 } else if (mddev
->bitmap
) {
1162 /* If adding to array with a bitmap, then we can accept an
1163 * older device, but not too old.
1165 __u64 ev1
= le64_to_cpu(sb
->events
);
1166 if (ev1
< mddev
->bitmap
->events_cleared
)
1168 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1171 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1173 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1174 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1176 case 0xffff: /* spare */
1178 case 0xfffe: /* faulty */
1179 set_bit(Faulty
, &rdev
->flags
);
1182 if ((le32_to_cpu(sb
->feature_map
) &
1183 MD_FEATURE_RECOVERY_OFFSET
))
1184 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1186 set_bit(In_sync
, &rdev
->flags
);
1187 rdev
->raid_disk
= role
;
1190 if (sb
->devflags
& WriteMostly1
)
1191 set_bit(WriteMostly
, &rdev
->flags
);
1192 } else /* MULTIPATH are always insync */
1193 set_bit(In_sync
, &rdev
->flags
);
1198 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1200 struct mdp_superblock_1
*sb
;
1201 struct list_head
*tmp
;
1204 /* make rdev->sb match mddev and rdev data. */
1206 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1208 sb
->feature_map
= 0;
1210 sb
->recovery_offset
= cpu_to_le64(0);
1211 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1212 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1213 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1215 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1216 sb
->events
= cpu_to_le64(mddev
->events
);
1218 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1220 sb
->resync_offset
= cpu_to_le64(0);
1222 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1224 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1225 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1227 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1228 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1229 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1232 if (rdev
->raid_disk
>= 0 &&
1233 !test_bit(In_sync
, &rdev
->flags
) &&
1234 rdev
->recovery_offset
> 0) {
1235 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1236 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1239 if (mddev
->reshape_position
!= MaxSector
) {
1240 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1241 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1242 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1243 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1244 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1245 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1249 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1250 if (rdev2
->desc_nr
+1 > max_dev
)
1251 max_dev
= rdev2
->desc_nr
+1;
1253 sb
->max_dev
= cpu_to_le32(max_dev
);
1254 for (i
=0; i
<max_dev
;i
++)
1255 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1257 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1259 if (test_bit(Faulty
, &rdev2
->flags
))
1260 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1261 else if (test_bit(In_sync
, &rdev2
->flags
))
1262 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1263 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1264 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1266 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1269 sb
->sb_csum
= calc_sb_1_csum(sb
);
1273 static struct super_type super_types
[] = {
1276 .owner
= THIS_MODULE
,
1277 .load_super
= super_90_load
,
1278 .validate_super
= super_90_validate
,
1279 .sync_super
= super_90_sync
,
1283 .owner
= THIS_MODULE
,
1284 .load_super
= super_1_load
,
1285 .validate_super
= super_1_validate
,
1286 .sync_super
= super_1_sync
,
1290 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1292 struct list_head
*tmp
;
1295 ITERATE_RDEV(mddev
,rdev
,tmp
)
1296 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1302 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1304 struct list_head
*tmp
;
1307 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1308 if (match_dev_unit(mddev2
, rdev
))
1314 static LIST_HEAD(pending_raid_disks
);
1316 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1318 mdk_rdev_t
*same_pdev
;
1319 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1327 /* make sure rdev->size exceeds mddev->size */
1328 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1330 /* Cannot change size, so fail */
1333 mddev
->size
= rdev
->size
;
1335 same_pdev
= match_dev_unit(mddev
, rdev
);
1338 "%s: WARNING: %s appears to be on the same physical"
1339 " disk as %s. True\n protection against single-disk"
1340 " failure might be compromised.\n",
1341 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1342 bdevname(same_pdev
->bdev
,b2
));
1344 /* Verify rdev->desc_nr is unique.
1345 * If it is -1, assign a free number, else
1346 * check number is not in use
1348 if (rdev
->desc_nr
< 0) {
1350 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1351 while (find_rdev_nr(mddev
, choice
))
1353 rdev
->desc_nr
= choice
;
1355 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1358 bdevname(rdev
->bdev
,b
);
1359 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1361 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1364 list_add(&rdev
->same_set
, &mddev
->disks
);
1365 rdev
->mddev
= mddev
;
1366 printk(KERN_INFO
"md: bind<%s>\n", b
);
1368 rdev
->kobj
.parent
= &mddev
->kobj
;
1369 kobject_add(&rdev
->kobj
);
1371 if (rdev
->bdev
->bd_part
)
1372 ko
= &rdev
->bdev
->bd_part
->kobj
;
1374 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1375 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1376 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1380 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1382 char b
[BDEVNAME_SIZE
];
1387 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1388 list_del_init(&rdev
->same_set
);
1389 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1391 sysfs_remove_link(&rdev
->kobj
, "block");
1392 kobject_del(&rdev
->kobj
);
1396 * prevent the device from being mounted, repartitioned or
1397 * otherwise reused by a RAID array (or any other kernel
1398 * subsystem), by bd_claiming the device.
1400 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1403 struct block_device
*bdev
;
1404 char b
[BDEVNAME_SIZE
];
1406 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1408 printk(KERN_ERR
"md: could not open %s.\n",
1409 __bdevname(dev
, b
));
1410 return PTR_ERR(bdev
);
1412 err
= bd_claim(bdev
, rdev
);
1414 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1423 static void unlock_rdev(mdk_rdev_t
*rdev
)
1425 struct block_device
*bdev
= rdev
->bdev
;
1433 void md_autodetect_dev(dev_t dev
);
1435 static void export_rdev(mdk_rdev_t
* rdev
)
1437 char b
[BDEVNAME_SIZE
];
1438 printk(KERN_INFO
"md: export_rdev(%s)\n",
1439 bdevname(rdev
->bdev
,b
));
1443 list_del_init(&rdev
->same_set
);
1445 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1448 kobject_put(&rdev
->kobj
);
1451 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1453 unbind_rdev_from_array(rdev
);
1457 static void export_array(mddev_t
*mddev
)
1459 struct list_head
*tmp
;
1462 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1467 kick_rdev_from_array(rdev
);
1469 if (!list_empty(&mddev
->disks
))
1471 mddev
->raid_disks
= 0;
1472 mddev
->major_version
= 0;
1475 static void print_desc(mdp_disk_t
*desc
)
1477 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1478 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1481 static void print_sb(mdp_super_t
*sb
)
1486 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1487 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1488 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1490 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1491 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1492 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1493 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1494 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1495 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1496 sb
->failed_disks
, sb
->spare_disks
,
1497 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1500 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1503 desc
= sb
->disks
+ i
;
1504 if (desc
->number
|| desc
->major
|| desc
->minor
||
1505 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1506 printk(" D %2d: ", i
);
1510 printk(KERN_INFO
"md: THIS: ");
1511 print_desc(&sb
->this_disk
);
1515 static void print_rdev(mdk_rdev_t
*rdev
)
1517 char b
[BDEVNAME_SIZE
];
1518 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1519 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1520 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1522 if (rdev
->sb_loaded
) {
1523 printk(KERN_INFO
"md: rdev superblock:\n");
1524 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1526 printk(KERN_INFO
"md: no rdev superblock!\n");
1529 static void md_print_devices(void)
1531 struct list_head
*tmp
, *tmp2
;
1534 char b
[BDEVNAME_SIZE
];
1537 printk("md: **********************************\n");
1538 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1539 printk("md: **********************************\n");
1540 ITERATE_MDDEV(mddev
,tmp
) {
1543 bitmap_print_sb(mddev
->bitmap
);
1545 printk("%s: ", mdname(mddev
));
1546 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1547 printk("<%s>", bdevname(rdev
->bdev
,b
));
1550 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1553 printk("md: **********************************\n");
1558 static void sync_sbs(mddev_t
* mddev
)
1561 struct list_head
*tmp
;
1563 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1564 super_types
[mddev
->major_version
].
1565 sync_super(mddev
, rdev
);
1566 rdev
->sb_loaded
= 1;
1570 void md_update_sb(mddev_t
* mddev
)
1573 struct list_head
*tmp
;
1578 spin_lock_irq(&mddev
->write_lock
);
1579 sync_req
= mddev
->in_sync
;
1580 mddev
->utime
= get_seconds();
1583 if (!mddev
->events
) {
1585 * oops, this 64-bit counter should never wrap.
1586 * Either we are in around ~1 trillion A.C., assuming
1587 * 1 reboot per second, or we have a bug:
1592 mddev
->sb_dirty
= 2;
1596 * do not write anything to disk if using
1597 * nonpersistent superblocks
1599 if (!mddev
->persistent
) {
1600 mddev
->sb_dirty
= 0;
1601 spin_unlock_irq(&mddev
->write_lock
);
1602 wake_up(&mddev
->sb_wait
);
1605 spin_unlock_irq(&mddev
->write_lock
);
1608 "md: updating %s RAID superblock on device (in sync %d)\n",
1609 mdname(mddev
),mddev
->in_sync
);
1611 err
= bitmap_update_sb(mddev
->bitmap
);
1612 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1613 char b
[BDEVNAME_SIZE
];
1614 dprintk(KERN_INFO
"md: ");
1615 if (test_bit(Faulty
, &rdev
->flags
))
1616 dprintk("(skipping faulty ");
1618 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1619 if (!test_bit(Faulty
, &rdev
->flags
)) {
1620 md_super_write(mddev
,rdev
,
1621 rdev
->sb_offset
<<1, rdev
->sb_size
,
1623 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1624 bdevname(rdev
->bdev
,b
),
1625 (unsigned long long)rdev
->sb_offset
);
1629 if (mddev
->level
== LEVEL_MULTIPATH
)
1630 /* only need to write one superblock... */
1633 md_super_wait(mddev
);
1634 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1636 spin_lock_irq(&mddev
->write_lock
);
1637 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1638 /* have to write it out again */
1639 spin_unlock_irq(&mddev
->write_lock
);
1642 mddev
->sb_dirty
= 0;
1643 spin_unlock_irq(&mddev
->write_lock
);
1644 wake_up(&mddev
->sb_wait
);
1647 EXPORT_SYMBOL_GPL(md_update_sb
);
1649 /* words written to sysfs files may, or my not, be \n terminated.
1650 * We want to accept with case. For this we use cmd_match.
1652 static int cmd_match(const char *cmd
, const char *str
)
1654 /* See if cmd, written into a sysfs file, matches
1655 * str. They must either be the same, or cmd can
1656 * have a trailing newline
1658 while (*cmd
&& *str
&& *cmd
== *str
) {
1669 struct rdev_sysfs_entry
{
1670 struct attribute attr
;
1671 ssize_t (*show
)(mdk_rdev_t
*, char *);
1672 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1676 state_show(mdk_rdev_t
*rdev
, char *page
)
1681 if (test_bit(Faulty
, &rdev
->flags
)) {
1682 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1685 if (test_bit(In_sync
, &rdev
->flags
)) {
1686 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1689 if (!test_bit(Faulty
, &rdev
->flags
) &&
1690 !test_bit(In_sync
, &rdev
->flags
)) {
1691 len
+= sprintf(page
+len
, "%sspare", sep
);
1694 return len
+sprintf(page
+len
, "\n");
1697 static struct rdev_sysfs_entry
1698 rdev_state
= __ATTR_RO(state
);
1701 super_show(mdk_rdev_t
*rdev
, char *page
)
1703 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1704 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1705 return rdev
->sb_size
;
1709 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1712 errors_show(mdk_rdev_t
*rdev
, char *page
)
1714 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1718 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1721 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1722 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1723 atomic_set(&rdev
->corrected_errors
, n
);
1728 static struct rdev_sysfs_entry rdev_errors
=
1729 __ATTR(errors
, 0644, errors_show
, errors_store
);
1732 slot_show(mdk_rdev_t
*rdev
, char *page
)
1734 if (rdev
->raid_disk
< 0)
1735 return sprintf(page
, "none\n");
1737 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1741 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1744 int slot
= simple_strtoul(buf
, &e
, 10);
1745 if (strncmp(buf
, "none", 4)==0)
1747 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1749 if (rdev
->mddev
->pers
)
1750 /* Cannot set slot in active array (yet) */
1752 if (slot
>= rdev
->mddev
->raid_disks
)
1754 rdev
->raid_disk
= slot
;
1755 /* assume it is working */
1757 set_bit(In_sync
, &rdev
->flags
);
1762 static struct rdev_sysfs_entry rdev_slot
=
1763 __ATTR(slot
, 0644, slot_show
, slot_store
);
1766 offset_show(mdk_rdev_t
*rdev
, char *page
)
1768 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1772 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1775 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1776 if (e
==buf
|| (*e
&& *e
!= '\n'))
1778 if (rdev
->mddev
->pers
)
1780 rdev
->data_offset
= offset
;
1784 static struct rdev_sysfs_entry rdev_offset
=
1785 __ATTR(offset
, 0644, offset_show
, offset_store
);
1788 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1790 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1794 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1797 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1798 if (e
==buf
|| (*e
&& *e
!= '\n'))
1800 if (rdev
->mddev
->pers
)
1803 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1804 rdev
->mddev
->size
= size
;
1808 static struct rdev_sysfs_entry rdev_size
=
1809 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1811 static struct attribute
*rdev_default_attrs
[] = {
1821 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1823 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1824 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1828 return entry
->show(rdev
, page
);
1832 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1833 const char *page
, size_t length
)
1835 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1836 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1840 return entry
->store(rdev
, page
, length
);
1843 static void rdev_free(struct kobject
*ko
)
1845 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1848 static struct sysfs_ops rdev_sysfs_ops
= {
1849 .show
= rdev_attr_show
,
1850 .store
= rdev_attr_store
,
1852 static struct kobj_type rdev_ktype
= {
1853 .release
= rdev_free
,
1854 .sysfs_ops
= &rdev_sysfs_ops
,
1855 .default_attrs
= rdev_default_attrs
,
1859 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1861 * mark the device faulty if:
1863 * - the device is nonexistent (zero size)
1864 * - the device has no valid superblock
1866 * a faulty rdev _never_ has rdev->sb set.
1868 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1870 char b
[BDEVNAME_SIZE
];
1875 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1877 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1878 return ERR_PTR(-ENOMEM
);
1881 if ((err
= alloc_disk_sb(rdev
)))
1884 err
= lock_rdev(rdev
, newdev
);
1888 rdev
->kobj
.parent
= NULL
;
1889 rdev
->kobj
.ktype
= &rdev_ktype
;
1890 kobject_init(&rdev
->kobj
);
1894 rdev
->data_offset
= 0;
1895 atomic_set(&rdev
->nr_pending
, 0);
1896 atomic_set(&rdev
->read_errors
, 0);
1897 atomic_set(&rdev
->corrected_errors
, 0);
1899 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1902 "md: %s has zero or unknown size, marking faulty!\n",
1903 bdevname(rdev
->bdev
,b
));
1908 if (super_format
>= 0) {
1909 err
= super_types
[super_format
].
1910 load_super(rdev
, NULL
, super_minor
);
1911 if (err
== -EINVAL
) {
1913 "md: %s has invalid sb, not importing!\n",
1914 bdevname(rdev
->bdev
,b
));
1919 "md: could not read %s's sb, not importing!\n",
1920 bdevname(rdev
->bdev
,b
));
1924 INIT_LIST_HEAD(&rdev
->same_set
);
1929 if (rdev
->sb_page
) {
1935 return ERR_PTR(err
);
1939 * Check a full RAID array for plausibility
1943 static void analyze_sbs(mddev_t
* mddev
)
1946 struct list_head
*tmp
;
1947 mdk_rdev_t
*rdev
, *freshest
;
1948 char b
[BDEVNAME_SIZE
];
1951 ITERATE_RDEV(mddev
,rdev
,tmp
)
1952 switch (super_types
[mddev
->major_version
].
1953 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1961 "md: fatal superblock inconsistency in %s"
1962 " -- removing from array\n",
1963 bdevname(rdev
->bdev
,b
));
1964 kick_rdev_from_array(rdev
);
1968 super_types
[mddev
->major_version
].
1969 validate_super(mddev
, freshest
);
1972 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1973 if (rdev
!= freshest
)
1974 if (super_types
[mddev
->major_version
].
1975 validate_super(mddev
, rdev
)) {
1976 printk(KERN_WARNING
"md: kicking non-fresh %s"
1978 bdevname(rdev
->bdev
,b
));
1979 kick_rdev_from_array(rdev
);
1982 if (mddev
->level
== LEVEL_MULTIPATH
) {
1983 rdev
->desc_nr
= i
++;
1984 rdev
->raid_disk
= rdev
->desc_nr
;
1985 set_bit(In_sync
, &rdev
->flags
);
1991 if (mddev
->recovery_cp
!= MaxSector
&&
1993 printk(KERN_ERR
"md: %s: raid array is not clean"
1994 " -- starting background reconstruction\n",
2000 safe_delay_show(mddev_t
*mddev
, char *page
)
2002 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2003 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2006 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2014 /* remove a period, and count digits after it */
2015 if (len
>= sizeof(buf
))
2017 strlcpy(buf
, cbuf
, len
);
2019 for (i
=0; i
<len
; i
++) {
2021 if (isdigit(buf
[i
])) {
2026 } else if (buf
[i
] == '.') {
2031 msec
= simple_strtoul(buf
, &e
, 10);
2032 if (e
== buf
|| (*e
&& *e
!= '\n'))
2034 msec
= (msec
* 1000) / scale
;
2036 mddev
->safemode_delay
= 0;
2038 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2039 if (mddev
->safemode_delay
== 0)
2040 mddev
->safemode_delay
= 1;
2044 static struct md_sysfs_entry md_safe_delay
=
2045 __ATTR(safe_mode_delay
, 0644,safe_delay_show
, safe_delay_store
);
2048 level_show(mddev_t
*mddev
, char *page
)
2050 struct mdk_personality
*p
= mddev
->pers
;
2052 return sprintf(page
, "%s\n", p
->name
);
2053 else if (mddev
->clevel
[0])
2054 return sprintf(page
, "%s\n", mddev
->clevel
);
2055 else if (mddev
->level
!= LEVEL_NONE
)
2056 return sprintf(page
, "%d\n", mddev
->level
);
2062 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2069 if (len
>= sizeof(mddev
->clevel
))
2071 strncpy(mddev
->clevel
, buf
, len
);
2072 if (mddev
->clevel
[len
-1] == '\n')
2074 mddev
->clevel
[len
] = 0;
2075 mddev
->level
= LEVEL_NONE
;
2079 static struct md_sysfs_entry md_level
=
2080 __ATTR(level
, 0644, level_show
, level_store
);
2083 raid_disks_show(mddev_t
*mddev
, char *page
)
2085 if (mddev
->raid_disks
== 0)
2087 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2090 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2093 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2095 /* can only set raid_disks if array is not yet active */
2098 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2100 if (!*buf
|| (*e
&& *e
!= '\n'))
2104 rv
= update_raid_disks(mddev
, n
);
2106 mddev
->raid_disks
= n
;
2107 return rv
? rv
: len
;
2109 static struct md_sysfs_entry md_raid_disks
=
2110 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
2113 chunk_size_show(mddev_t
*mddev
, char *page
)
2115 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2119 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2121 /* can only set chunk_size if array is not yet active */
2123 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2127 if (!*buf
|| (*e
&& *e
!= '\n'))
2130 mddev
->chunk_size
= n
;
2133 static struct md_sysfs_entry md_chunk_size
=
2134 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2137 null_show(mddev_t
*mddev
, char *page
)
2143 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2145 /* buf must be %d:%d\n? giving major and minor numbers */
2146 /* The new device is added to the array.
2147 * If the array has a persistent superblock, we read the
2148 * superblock to initialise info and check validity.
2149 * Otherwise, only checking done is that in bind_rdev_to_array,
2150 * which mainly checks size.
2153 int major
= simple_strtoul(buf
, &e
, 10);
2159 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2161 minor
= simple_strtoul(e
+1, &e
, 10);
2162 if (*e
&& *e
!= '\n')
2164 dev
= MKDEV(major
, minor
);
2165 if (major
!= MAJOR(dev
) ||
2166 minor
!= MINOR(dev
))
2170 if (mddev
->persistent
) {
2171 rdev
= md_import_device(dev
, mddev
->major_version
,
2172 mddev
->minor_version
);
2173 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2174 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2175 mdk_rdev_t
, same_set
);
2176 err
= super_types
[mddev
->major_version
]
2177 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2182 rdev
= md_import_device(dev
, -1, -1);
2185 return PTR_ERR(rdev
);
2186 err
= bind_rdev_to_array(rdev
, mddev
);
2190 return err
? err
: len
;
2193 static struct md_sysfs_entry md_new_device
=
2194 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2197 size_show(mddev_t
*mddev
, char *page
)
2199 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2202 static int update_size(mddev_t
*mddev
, unsigned long size
);
2205 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2207 /* If array is inactive, we can reduce the component size, but
2208 * not increase it (except from 0).
2209 * If array is active, we can try an on-line resize
2213 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2214 if (!*buf
|| *buf
== '\n' ||
2219 err
= update_size(mddev
, size
);
2220 md_update_sb(mddev
);
2222 if (mddev
->size
== 0 ||
2228 return err
? err
: len
;
2231 static struct md_sysfs_entry md_size
=
2232 __ATTR(component_size
, 0644, size_show
, size_store
);
2236 * This is either 'none' for arrays with externally managed metadata,
2237 * or N.M for internally known formats
2240 metadata_show(mddev_t
*mddev
, char *page
)
2242 if (mddev
->persistent
)
2243 return sprintf(page
, "%d.%d\n",
2244 mddev
->major_version
, mddev
->minor_version
);
2246 return sprintf(page
, "none\n");
2250 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2254 if (!list_empty(&mddev
->disks
))
2257 if (cmd_match(buf
, "none")) {
2258 mddev
->persistent
= 0;
2259 mddev
->major_version
= 0;
2260 mddev
->minor_version
= 90;
2263 major
= simple_strtoul(buf
, &e
, 10);
2264 if (e
==buf
|| *e
!= '.')
2267 minor
= simple_strtoul(buf
, &e
, 10);
2268 if (e
==buf
|| *e
!= '\n')
2270 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2271 super_types
[major
].name
== NULL
)
2273 mddev
->major_version
= major
;
2274 mddev
->minor_version
= minor
;
2275 mddev
->persistent
= 1;
2279 static struct md_sysfs_entry md_metadata
=
2280 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2283 action_show(mddev_t
*mddev
, char *page
)
2285 char *type
= "idle";
2286 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2287 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2288 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2290 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2291 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2293 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2300 return sprintf(page
, "%s\n", type
);
2304 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2306 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2309 if (cmd_match(page
, "idle")) {
2310 if (mddev
->sync_thread
) {
2311 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2312 md_unregister_thread(mddev
->sync_thread
);
2313 mddev
->sync_thread
= NULL
;
2314 mddev
->recovery
= 0;
2316 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2317 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2319 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2320 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2321 else if (cmd_match(page
, "reshape")) {
2323 if (mddev
->pers
->start_reshape
== NULL
)
2325 err
= mddev
->pers
->start_reshape(mddev
);
2329 if (cmd_match(page
, "check"))
2330 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2331 else if (!cmd_match(page
, "repair"))
2333 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2334 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2336 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2337 md_wakeup_thread(mddev
->thread
);
2342 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2344 return sprintf(page
, "%llu\n",
2345 (unsigned long long) mddev
->resync_mismatches
);
2348 static struct md_sysfs_entry
2349 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2352 static struct md_sysfs_entry
2353 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2356 sync_min_show(mddev_t
*mddev
, char *page
)
2358 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2359 mddev
->sync_speed_min
? "local": "system");
2363 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2367 if (strncmp(buf
, "system", 6)==0) {
2368 mddev
->sync_speed_min
= 0;
2371 min
= simple_strtoul(buf
, &e
, 10);
2372 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2374 mddev
->sync_speed_min
= min
;
2378 static struct md_sysfs_entry md_sync_min
=
2379 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2382 sync_max_show(mddev_t
*mddev
, char *page
)
2384 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2385 mddev
->sync_speed_max
? "local": "system");
2389 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2393 if (strncmp(buf
, "system", 6)==0) {
2394 mddev
->sync_speed_max
= 0;
2397 max
= simple_strtoul(buf
, &e
, 10);
2398 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2400 mddev
->sync_speed_max
= max
;
2404 static struct md_sysfs_entry md_sync_max
=
2405 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2409 sync_speed_show(mddev_t
*mddev
, char *page
)
2411 unsigned long resync
, dt
, db
;
2412 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2413 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2415 db
= resync
- (mddev
->resync_mark_cnt
);
2416 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2419 static struct md_sysfs_entry
2420 md_sync_speed
= __ATTR_RO(sync_speed
);
2423 sync_completed_show(mddev_t
*mddev
, char *page
)
2425 unsigned long max_blocks
, resync
;
2427 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2428 max_blocks
= mddev
->resync_max_sectors
;
2430 max_blocks
= mddev
->size
<< 1;
2432 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2433 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2436 static struct md_sysfs_entry
2437 md_sync_completed
= __ATTR_RO(sync_completed
);
2440 suspend_lo_show(mddev_t
*mddev
, char *page
)
2442 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2446 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2449 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2451 if (mddev
->pers
->quiesce
== NULL
)
2453 if (buf
== e
|| (*e
&& *e
!= '\n'))
2455 if (new >= mddev
->suspend_hi
||
2456 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2457 mddev
->suspend_lo
= new;
2458 mddev
->pers
->quiesce(mddev
, 2);
2463 static struct md_sysfs_entry md_suspend_lo
=
2464 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2468 suspend_hi_show(mddev_t
*mddev
, char *page
)
2470 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2474 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2477 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2479 if (mddev
->pers
->quiesce
== NULL
)
2481 if (buf
== e
|| (*e
&& *e
!= '\n'))
2483 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2484 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2485 mddev
->suspend_hi
= new;
2486 mddev
->pers
->quiesce(mddev
, 1);
2487 mddev
->pers
->quiesce(mddev
, 0);
2492 static struct md_sysfs_entry md_suspend_hi
=
2493 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2496 static struct attribute
*md_default_attrs
[] = {
2498 &md_raid_disks
.attr
,
2499 &md_chunk_size
.attr
,
2502 &md_new_device
.attr
,
2503 &md_safe_delay
.attr
,
2507 static struct attribute
*md_redundancy_attrs
[] = {
2509 &md_mismatches
.attr
,
2512 &md_sync_speed
.attr
,
2513 &md_sync_completed
.attr
,
2514 &md_suspend_lo
.attr
,
2515 &md_suspend_hi
.attr
,
2518 static struct attribute_group md_redundancy_group
= {
2520 .attrs
= md_redundancy_attrs
,
2525 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2527 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2528 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2533 rv
= mddev_lock(mddev
);
2535 rv
= entry
->show(mddev
, page
);
2536 mddev_unlock(mddev
);
2542 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2543 const char *page
, size_t length
)
2545 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2546 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2551 rv
= mddev_lock(mddev
);
2553 rv
= entry
->store(mddev
, page
, length
);
2554 mddev_unlock(mddev
);
2559 static void md_free(struct kobject
*ko
)
2561 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2565 static struct sysfs_ops md_sysfs_ops
= {
2566 .show
= md_attr_show
,
2567 .store
= md_attr_store
,
2569 static struct kobj_type md_ktype
= {
2571 .sysfs_ops
= &md_sysfs_ops
,
2572 .default_attrs
= md_default_attrs
,
2577 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2579 static DEFINE_MUTEX(disks_mutex
);
2580 mddev_t
*mddev
= mddev_find(dev
);
2581 struct gendisk
*disk
;
2582 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2583 int shift
= partitioned
? MdpMinorShift
: 0;
2584 int unit
= MINOR(dev
) >> shift
;
2589 mutex_lock(&disks_mutex
);
2590 if (mddev
->gendisk
) {
2591 mutex_unlock(&disks_mutex
);
2595 disk
= alloc_disk(1 << shift
);
2597 mutex_unlock(&disks_mutex
);
2601 disk
->major
= MAJOR(dev
);
2602 disk
->first_minor
= unit
<< shift
;
2604 sprintf(disk
->disk_name
, "md_d%d", unit
);
2605 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2607 sprintf(disk
->disk_name
, "md%d", unit
);
2608 sprintf(disk
->devfs_name
, "md/%d", unit
);
2610 disk
->fops
= &md_fops
;
2611 disk
->private_data
= mddev
;
2612 disk
->queue
= mddev
->queue
;
2614 mddev
->gendisk
= disk
;
2615 mutex_unlock(&disks_mutex
);
2616 mddev
->kobj
.parent
= &disk
->kobj
;
2617 mddev
->kobj
.k_name
= NULL
;
2618 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2619 mddev
->kobj
.ktype
= &md_ktype
;
2620 kobject_register(&mddev
->kobj
);
2624 static void md_safemode_timeout(unsigned long data
)
2626 mddev_t
*mddev
= (mddev_t
*) data
;
2628 mddev
->safemode
= 1;
2629 md_wakeup_thread(mddev
->thread
);
2632 static int start_dirty_degraded
;
2634 static int do_md_run(mddev_t
* mddev
)
2638 struct list_head
*tmp
;
2640 struct gendisk
*disk
;
2641 struct mdk_personality
*pers
;
2642 char b
[BDEVNAME_SIZE
];
2644 if (list_empty(&mddev
->disks
))
2645 /* cannot run an array with no devices.. */
2652 * Analyze all RAID superblock(s)
2654 if (!mddev
->raid_disks
)
2657 chunk_size
= mddev
->chunk_size
;
2660 if (chunk_size
> MAX_CHUNK_SIZE
) {
2661 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2662 chunk_size
, MAX_CHUNK_SIZE
);
2666 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2668 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2669 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2672 if (chunk_size
< PAGE_SIZE
) {
2673 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2674 chunk_size
, PAGE_SIZE
);
2678 /* devices must have minimum size of one chunk */
2679 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2680 if (test_bit(Faulty
, &rdev
->flags
))
2682 if (rdev
->size
< chunk_size
/ 1024) {
2684 "md: Dev %s smaller than chunk_size:"
2686 bdevname(rdev
->bdev
,b
),
2687 (unsigned long long)rdev
->size
,
2695 if (mddev
->level
!= LEVEL_NONE
)
2696 request_module("md-level-%d", mddev
->level
);
2697 else if (mddev
->clevel
[0])
2698 request_module("md-%s", mddev
->clevel
);
2702 * Drop all container device buffers, from now on
2703 * the only valid external interface is through the md
2705 * Also find largest hardsector size
2707 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2708 if (test_bit(Faulty
, &rdev
->flags
))
2710 sync_blockdev(rdev
->bdev
);
2711 invalidate_bdev(rdev
->bdev
, 0);
2714 md_probe(mddev
->unit
, NULL
, NULL
);
2715 disk
= mddev
->gendisk
;
2719 spin_lock(&pers_lock
);
2720 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2721 if (!pers
|| !try_module_get(pers
->owner
)) {
2722 spin_unlock(&pers_lock
);
2723 if (mddev
->level
!= LEVEL_NONE
)
2724 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2727 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2732 spin_unlock(&pers_lock
);
2733 mddev
->level
= pers
->level
;
2734 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2736 if (mddev
->reshape_position
!= MaxSector
&&
2737 pers
->start_reshape
== NULL
) {
2738 /* This personality cannot handle reshaping... */
2740 module_put(pers
->owner
);
2744 mddev
->recovery
= 0;
2745 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2746 mddev
->barriers_work
= 1;
2747 mddev
->ok_start_degraded
= start_dirty_degraded
;
2750 mddev
->ro
= 2; /* read-only, but switch on first write */
2752 err
= mddev
->pers
->run(mddev
);
2753 if (!err
&& mddev
->pers
->sync_request
) {
2754 err
= bitmap_create(mddev
);
2756 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2757 mdname(mddev
), err
);
2758 mddev
->pers
->stop(mddev
);
2762 printk(KERN_ERR
"md: pers->run() failed ...\n");
2763 module_put(mddev
->pers
->owner
);
2765 bitmap_destroy(mddev
);
2768 if (mddev
->pers
->sync_request
)
2769 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2770 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2773 atomic_set(&mddev
->writes_pending
,0);
2774 mddev
->safemode
= 0;
2775 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2776 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2777 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
2780 ITERATE_RDEV(mddev
,rdev
,tmp
)
2781 if (rdev
->raid_disk
>= 0) {
2783 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2784 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2787 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2788 md_wakeup_thread(mddev
->thread
);
2790 if (mddev
->sb_dirty
)
2791 md_update_sb(mddev
);
2793 set_capacity(disk
, mddev
->array_size
<<1);
2795 /* If we call blk_queue_make_request here, it will
2796 * re-initialise max_sectors etc which may have been
2797 * refined inside -> run. So just set the bits we need to set.
2798 * Most initialisation happended when we called
2799 * blk_queue_make_request(..., md_fail_request)
2802 mddev
->queue
->queuedata
= mddev
;
2803 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2805 /* If there is a partially-recovered drive we need to
2806 * start recovery here. If we leave it to md_check_recovery,
2807 * it will remove the drives and not do the right thing
2809 if (mddev
->degraded
) {
2810 struct list_head
*rtmp
;
2812 ITERATE_RDEV(mddev
,rdev
,rtmp
)
2813 if (rdev
->raid_disk
>= 0 &&
2814 !test_bit(In_sync
, &rdev
->flags
) &&
2815 !test_bit(Faulty
, &rdev
->flags
))
2816 /* complete an interrupted recovery */
2818 if (spares
&& mddev
->pers
->sync_request
) {
2819 mddev
->recovery
= 0;
2820 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
2821 mddev
->sync_thread
= md_register_thread(md_do_sync
,
2824 if (!mddev
->sync_thread
) {
2825 printk(KERN_ERR
"%s: could not start resync"
2828 /* leave the spares where they are, it shouldn't hurt */
2829 mddev
->recovery
= 0;
2831 md_wakeup_thread(mddev
->sync_thread
);
2836 md_new_event(mddev
);
2840 static int restart_array(mddev_t
*mddev
)
2842 struct gendisk
*disk
= mddev
->gendisk
;
2846 * Complain if it has no devices
2849 if (list_empty(&mddev
->disks
))
2857 mddev
->safemode
= 0;
2859 set_disk_ro(disk
, 0);
2861 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2864 * Kick recovery or resync if necessary
2866 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2867 md_wakeup_thread(mddev
->thread
);
2868 md_wakeup_thread(mddev
->sync_thread
);
2871 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2880 static int do_md_stop(mddev_t
* mddev
, int ro
)
2883 struct gendisk
*disk
= mddev
->gendisk
;
2886 if (atomic_read(&mddev
->active
)>2) {
2887 printk("md: %s still in use.\n",mdname(mddev
));
2891 if (mddev
->sync_thread
) {
2892 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
2893 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2894 md_unregister_thread(mddev
->sync_thread
);
2895 mddev
->sync_thread
= NULL
;
2898 del_timer_sync(&mddev
->safemode_timer
);
2900 invalidate_partition(disk
, 0);
2908 bitmap_flush(mddev
);
2909 md_super_wait(mddev
);
2911 set_disk_ro(disk
, 0);
2912 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2913 mddev
->pers
->stop(mddev
);
2914 if (mddev
->pers
->sync_request
)
2915 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2917 module_put(mddev
->pers
->owner
);
2922 if (!mddev
->in_sync
|| mddev
->sb_dirty
) {
2923 /* mark array as shutdown cleanly */
2925 md_update_sb(mddev
);
2928 set_disk_ro(disk
, 1);
2929 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
2933 * Free resources if final stop
2937 struct list_head
*tmp
;
2938 struct gendisk
*disk
;
2939 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
2941 bitmap_destroy(mddev
);
2942 if (mddev
->bitmap_file
) {
2943 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
2944 fput(mddev
->bitmap_file
);
2945 mddev
->bitmap_file
= NULL
;
2947 mddev
->bitmap_offset
= 0;
2949 ITERATE_RDEV(mddev
,rdev
,tmp
)
2950 if (rdev
->raid_disk
>= 0) {
2952 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2953 sysfs_remove_link(&mddev
->kobj
, nm
);
2956 export_array(mddev
);
2958 mddev
->array_size
= 0;
2959 disk
= mddev
->gendisk
;
2961 set_capacity(disk
, 0);
2963 } else if (mddev
->pers
)
2964 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
2967 md_new_event(mddev
);
2972 static void autorun_array(mddev_t
*mddev
)
2975 struct list_head
*tmp
;
2978 if (list_empty(&mddev
->disks
))
2981 printk(KERN_INFO
"md: running: ");
2983 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2984 char b
[BDEVNAME_SIZE
];
2985 printk("<%s>", bdevname(rdev
->bdev
,b
));
2989 err
= do_md_run (mddev
);
2991 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
2992 do_md_stop (mddev
, 0);
2997 * lets try to run arrays based on all disks that have arrived
2998 * until now. (those are in pending_raid_disks)
3000 * the method: pick the first pending disk, collect all disks with
3001 * the same UUID, remove all from the pending list and put them into
3002 * the 'same_array' list. Then order this list based on superblock
3003 * update time (freshest comes first), kick out 'old' disks and
3004 * compare superblocks. If everything's fine then run it.
3006 * If "unit" is allocated, then bump its reference count
3008 static void autorun_devices(int part
)
3010 struct list_head
*tmp
;
3011 mdk_rdev_t
*rdev0
, *rdev
;
3013 char b
[BDEVNAME_SIZE
];
3015 printk(KERN_INFO
"md: autorun ...\n");
3016 while (!list_empty(&pending_raid_disks
)) {
3018 LIST_HEAD(candidates
);
3019 rdev0
= list_entry(pending_raid_disks
.next
,
3020 mdk_rdev_t
, same_set
);
3022 printk(KERN_INFO
"md: considering %s ...\n",
3023 bdevname(rdev0
->bdev
,b
));
3024 INIT_LIST_HEAD(&candidates
);
3025 ITERATE_RDEV_PENDING(rdev
,tmp
)
3026 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3027 printk(KERN_INFO
"md: adding %s ...\n",
3028 bdevname(rdev
->bdev
,b
));
3029 list_move(&rdev
->same_set
, &candidates
);
3032 * now we have a set of devices, with all of them having
3033 * mostly sane superblocks. It's time to allocate the
3036 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
3037 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3038 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3042 dev
= MKDEV(mdp_major
,
3043 rdev0
->preferred_minor
<< MdpMinorShift
);
3045 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3047 md_probe(dev
, NULL
, NULL
);
3048 mddev
= mddev_find(dev
);
3051 "md: cannot allocate memory for md drive.\n");
3054 if (mddev_lock(mddev
))
3055 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3057 else if (mddev
->raid_disks
|| mddev
->major_version
3058 || !list_empty(&mddev
->disks
)) {
3060 "md: %s already running, cannot run %s\n",
3061 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3062 mddev_unlock(mddev
);
3064 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3065 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3066 list_del_init(&rdev
->same_set
);
3067 if (bind_rdev_to_array(rdev
, mddev
))
3070 autorun_array(mddev
);
3071 mddev_unlock(mddev
);
3073 /* on success, candidates will be empty, on error
3076 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3080 printk(KERN_INFO
"md: ... autorun DONE.\n");
3084 * import RAID devices based on one partition
3085 * if possible, the array gets run as well.
3088 static int autostart_array(dev_t startdev
)
3090 char b
[BDEVNAME_SIZE
];
3091 int err
= -EINVAL
, i
;
3092 mdp_super_t
*sb
= NULL
;
3093 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
3095 start_rdev
= md_import_device(startdev
, 0, 0);
3096 if (IS_ERR(start_rdev
))
3100 /* NOTE: this can only work for 0.90.0 superblocks */
3101 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
3102 if (sb
->major_version
!= 0 ||
3103 sb
->minor_version
!= 90 ) {
3104 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
3105 export_rdev(start_rdev
);
3109 if (test_bit(Faulty
, &start_rdev
->flags
)) {
3111 "md: can not autostart based on faulty %s!\n",
3112 bdevname(start_rdev
->bdev
,b
));
3113 export_rdev(start_rdev
);
3116 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
3118 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
3119 mdp_disk_t
*desc
= sb
->disks
+ i
;
3120 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
3124 if (dev
== startdev
)
3126 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
3128 rdev
= md_import_device(dev
, 0, 0);
3132 list_add(&rdev
->same_set
, &pending_raid_disks
);
3136 * possibly return codes
3144 static int get_version(void __user
* arg
)
3148 ver
.major
= MD_MAJOR_VERSION
;
3149 ver
.minor
= MD_MINOR_VERSION
;
3150 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3152 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3158 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3160 mdu_array_info_t info
;
3161 int nr
,working
,active
,failed
,spare
;
3163 struct list_head
*tmp
;
3165 nr
=working
=active
=failed
=spare
=0;
3166 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3168 if (test_bit(Faulty
, &rdev
->flags
))
3172 if (test_bit(In_sync
, &rdev
->flags
))
3179 info
.major_version
= mddev
->major_version
;
3180 info
.minor_version
= mddev
->minor_version
;
3181 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3182 info
.ctime
= mddev
->ctime
;
3183 info
.level
= mddev
->level
;
3184 info
.size
= mddev
->size
;
3185 if (info
.size
!= mddev
->size
) /* overflow */
3188 info
.raid_disks
= mddev
->raid_disks
;
3189 info
.md_minor
= mddev
->md_minor
;
3190 info
.not_persistent
= !mddev
->persistent
;
3192 info
.utime
= mddev
->utime
;
3195 info
.state
= (1<<MD_SB_CLEAN
);
3196 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3197 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3198 info
.active_disks
= active
;
3199 info
.working_disks
= working
;
3200 info
.failed_disks
= failed
;
3201 info
.spare_disks
= spare
;
3203 info
.layout
= mddev
->layout
;
3204 info
.chunk_size
= mddev
->chunk_size
;
3206 if (copy_to_user(arg
, &info
, sizeof(info
)))
3212 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3214 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3215 char *ptr
, *buf
= NULL
;
3218 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3222 /* bitmap disabled, zero the first byte and copy out */
3223 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3224 file
->pathname
[0] = '\0';
3228 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3232 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3236 strcpy(file
->pathname
, ptr
);
3240 if (copy_to_user(arg
, file
, sizeof(*file
)))
3248 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3250 mdu_disk_info_t info
;
3254 if (copy_from_user(&info
, arg
, sizeof(info
)))
3259 rdev
= find_rdev_nr(mddev
, nr
);
3261 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3262 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3263 info
.raid_disk
= rdev
->raid_disk
;
3265 if (test_bit(Faulty
, &rdev
->flags
))
3266 info
.state
|= (1<<MD_DISK_FAULTY
);
3267 else if (test_bit(In_sync
, &rdev
->flags
)) {
3268 info
.state
|= (1<<MD_DISK_ACTIVE
);
3269 info
.state
|= (1<<MD_DISK_SYNC
);
3271 if (test_bit(WriteMostly
, &rdev
->flags
))
3272 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3274 info
.major
= info
.minor
= 0;
3275 info
.raid_disk
= -1;
3276 info
.state
= (1<<MD_DISK_REMOVED
);
3279 if (copy_to_user(arg
, &info
, sizeof(info
)))
3285 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3287 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3289 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3291 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3294 if (!mddev
->raid_disks
) {
3296 /* expecting a device which has a superblock */
3297 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3300 "md: md_import_device returned %ld\n",
3302 return PTR_ERR(rdev
);
3304 if (!list_empty(&mddev
->disks
)) {
3305 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3306 mdk_rdev_t
, same_set
);
3307 int err
= super_types
[mddev
->major_version
]
3308 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3311 "md: %s has different UUID to %s\n",
3312 bdevname(rdev
->bdev
,b
),
3313 bdevname(rdev0
->bdev
,b2
));
3318 err
= bind_rdev_to_array(rdev
, mddev
);
3325 * add_new_disk can be used once the array is assembled
3326 * to add "hot spares". They must already have a superblock
3331 if (!mddev
->pers
->hot_add_disk
) {
3333 "%s: personality does not support diskops!\n",
3337 if (mddev
->persistent
)
3338 rdev
= md_import_device(dev
, mddev
->major_version
,
3339 mddev
->minor_version
);
3341 rdev
= md_import_device(dev
, -1, -1);
3344 "md: md_import_device returned %ld\n",
3346 return PTR_ERR(rdev
);
3348 /* set save_raid_disk if appropriate */
3349 if (!mddev
->persistent
) {
3350 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3351 info
->raid_disk
< mddev
->raid_disks
)
3352 rdev
->raid_disk
= info
->raid_disk
;
3354 rdev
->raid_disk
= -1;
3356 super_types
[mddev
->major_version
].
3357 validate_super(mddev
, rdev
);
3358 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3360 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3361 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3362 set_bit(WriteMostly
, &rdev
->flags
);
3364 rdev
->raid_disk
= -1;
3365 err
= bind_rdev_to_array(rdev
, mddev
);
3366 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3367 /* If there is hot_add_disk but no hot_remove_disk
3368 * then added disks for geometry changes,
3369 * and should be added immediately.
3371 super_types
[mddev
->major_version
].
3372 validate_super(mddev
, rdev
);
3373 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3375 unbind_rdev_from_array(rdev
);
3380 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3381 md_wakeup_thread(mddev
->thread
);
3385 /* otherwise, add_new_disk is only allowed
3386 * for major_version==0 superblocks
3388 if (mddev
->major_version
!= 0) {
3389 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3394 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3396 rdev
= md_import_device (dev
, -1, 0);
3399 "md: error, md_import_device() returned %ld\n",
3401 return PTR_ERR(rdev
);
3403 rdev
->desc_nr
= info
->number
;
3404 if (info
->raid_disk
< mddev
->raid_disks
)
3405 rdev
->raid_disk
= info
->raid_disk
;
3407 rdev
->raid_disk
= -1;
3411 if (rdev
->raid_disk
< mddev
->raid_disks
)
3412 if (info
->state
& (1<<MD_DISK_SYNC
))
3413 set_bit(In_sync
, &rdev
->flags
);
3415 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3416 set_bit(WriteMostly
, &rdev
->flags
);
3418 if (!mddev
->persistent
) {
3419 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3420 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3422 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3423 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3425 err
= bind_rdev_to_array(rdev
, mddev
);
3435 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3437 char b
[BDEVNAME_SIZE
];
3443 rdev
= find_rdev(mddev
, dev
);
3447 if (rdev
->raid_disk
>= 0)
3450 kick_rdev_from_array(rdev
);
3451 md_update_sb(mddev
);
3452 md_new_event(mddev
);
3456 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3457 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3461 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3463 char b
[BDEVNAME_SIZE
];
3471 if (mddev
->major_version
!= 0) {
3472 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3473 " version-0 superblocks.\n",
3477 if (!mddev
->pers
->hot_add_disk
) {
3479 "%s: personality does not support diskops!\n",
3484 rdev
= md_import_device (dev
, -1, 0);
3487 "md: error, md_import_device() returned %ld\n",
3492 if (mddev
->persistent
)
3493 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3496 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3498 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3501 if (test_bit(Faulty
, &rdev
->flags
)) {
3503 "md: can not hot-add faulty %s disk to %s!\n",
3504 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3508 clear_bit(In_sync
, &rdev
->flags
);
3510 err
= bind_rdev_to_array(rdev
, mddev
);
3515 * The rest should better be atomic, we can have disk failures
3516 * noticed in interrupt contexts ...
3519 if (rdev
->desc_nr
== mddev
->max_disks
) {
3520 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3523 goto abort_unbind_export
;
3526 rdev
->raid_disk
= -1;
3528 md_update_sb(mddev
);
3531 * Kick recovery, maybe this spare has to be added to the
3532 * array immediately.
3534 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3535 md_wakeup_thread(mddev
->thread
);
3536 md_new_event(mddev
);
3539 abort_unbind_export
:
3540 unbind_rdev_from_array(rdev
);
3547 /* similar to deny_write_access, but accounts for our holding a reference
3548 * to the file ourselves */
3549 static int deny_bitmap_write_access(struct file
* file
)
3551 struct inode
*inode
= file
->f_mapping
->host
;
3553 spin_lock(&inode
->i_lock
);
3554 if (atomic_read(&inode
->i_writecount
) > 1) {
3555 spin_unlock(&inode
->i_lock
);
3558 atomic_set(&inode
->i_writecount
, -1);
3559 spin_unlock(&inode
->i_lock
);
3564 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3569 if (!mddev
->pers
->quiesce
)
3571 if (mddev
->recovery
|| mddev
->sync_thread
)
3573 /* we should be able to change the bitmap.. */
3579 return -EEXIST
; /* cannot add when bitmap is present */
3580 mddev
->bitmap_file
= fget(fd
);
3582 if (mddev
->bitmap_file
== NULL
) {
3583 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3588 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3590 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3592 fput(mddev
->bitmap_file
);
3593 mddev
->bitmap_file
= NULL
;
3596 mddev
->bitmap_offset
= 0; /* file overrides offset */
3597 } else if (mddev
->bitmap
== NULL
)
3598 return -ENOENT
; /* cannot remove what isn't there */
3601 mddev
->pers
->quiesce(mddev
, 1);
3603 err
= bitmap_create(mddev
);
3605 bitmap_destroy(mddev
);
3606 mddev
->pers
->quiesce(mddev
, 0);
3607 } else if (fd
< 0) {
3608 if (mddev
->bitmap_file
)
3609 fput(mddev
->bitmap_file
);
3610 mddev
->bitmap_file
= NULL
;
3617 * set_array_info is used two different ways
3618 * The original usage is when creating a new array.
3619 * In this usage, raid_disks is > 0 and it together with
3620 * level, size, not_persistent,layout,chunksize determine the
3621 * shape of the array.
3622 * This will always create an array with a type-0.90.0 superblock.
3623 * The newer usage is when assembling an array.
3624 * In this case raid_disks will be 0, and the major_version field is
3625 * use to determine which style super-blocks are to be found on the devices.
3626 * The minor and patch _version numbers are also kept incase the
3627 * super_block handler wishes to interpret them.
3629 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3632 if (info
->raid_disks
== 0) {
3633 /* just setting version number for superblock loading */
3634 if (info
->major_version
< 0 ||
3635 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3636 super_types
[info
->major_version
].name
== NULL
) {
3637 /* maybe try to auto-load a module? */
3639 "md: superblock version %d not known\n",
3640 info
->major_version
);
3643 mddev
->major_version
= info
->major_version
;
3644 mddev
->minor_version
= info
->minor_version
;
3645 mddev
->patch_version
= info
->patch_version
;
3648 mddev
->major_version
= MD_MAJOR_VERSION
;
3649 mddev
->minor_version
= MD_MINOR_VERSION
;
3650 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3651 mddev
->ctime
= get_seconds();
3653 mddev
->level
= info
->level
;
3654 mddev
->clevel
[0] = 0;
3655 mddev
->size
= info
->size
;
3656 mddev
->raid_disks
= info
->raid_disks
;
3657 /* don't set md_minor, it is determined by which /dev/md* was
3660 if (info
->state
& (1<<MD_SB_CLEAN
))
3661 mddev
->recovery_cp
= MaxSector
;
3663 mddev
->recovery_cp
= 0;
3664 mddev
->persistent
= ! info
->not_persistent
;
3666 mddev
->layout
= info
->layout
;
3667 mddev
->chunk_size
= info
->chunk_size
;
3669 mddev
->max_disks
= MD_SB_DISKS
;
3671 mddev
->sb_dirty
= 1;
3673 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3674 mddev
->bitmap_offset
= 0;
3676 mddev
->reshape_position
= MaxSector
;
3679 * Generate a 128 bit UUID
3681 get_random_bytes(mddev
->uuid
, 16);
3683 mddev
->new_level
= mddev
->level
;
3684 mddev
->new_chunk
= mddev
->chunk_size
;
3685 mddev
->new_layout
= mddev
->layout
;
3686 mddev
->delta_disks
= 0;
3691 static int update_size(mddev_t
*mddev
, unsigned long size
)
3695 struct list_head
*tmp
;
3696 int fit
= (size
== 0);
3698 if (mddev
->pers
->resize
== NULL
)
3700 /* The "size" is the amount of each device that is used.
3701 * This can only make sense for arrays with redundancy.
3702 * linear and raid0 always use whatever space is available
3703 * We can only consider changing the size if no resync
3704 * or reconstruction is happening, and if the new size
3705 * is acceptable. It must fit before the sb_offset or,
3706 * if that is <data_offset, it must fit before the
3707 * size of each device.
3708 * If size is zero, we find the largest size that fits.
3710 if (mddev
->sync_thread
)
3712 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3714 if (rdev
->sb_offset
> rdev
->data_offset
)
3715 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3717 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3718 - rdev
->data_offset
;
3719 if (fit
&& (size
== 0 || size
> avail
/2))
3721 if (avail
< ((sector_t
)size
<< 1))
3724 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3726 struct block_device
*bdev
;
3728 bdev
= bdget_disk(mddev
->gendisk
, 0);
3730 mutex_lock(&bdev
->bd_inode
->i_mutex
);
3731 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
3732 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
3739 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3742 /* change the number of raid disks */
3743 if (mddev
->pers
->check_reshape
== NULL
)
3745 if (raid_disks
<= 0 ||
3746 raid_disks
>= mddev
->max_disks
)
3748 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
3750 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
3752 rv
= mddev
->pers
->check_reshape(mddev
);
3758 * update_array_info is used to change the configuration of an
3760 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3761 * fields in the info are checked against the array.
3762 * Any differences that cannot be handled will cause an error.
3763 * Normally, only one change can be managed at a time.
3765 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3771 /* calculate expected state,ignoring low bits */
3772 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3773 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3775 if (mddev
->major_version
!= info
->major_version
||
3776 mddev
->minor_version
!= info
->minor_version
||
3777 /* mddev->patch_version != info->patch_version || */
3778 mddev
->ctime
!= info
->ctime
||
3779 mddev
->level
!= info
->level
||
3780 /* mddev->layout != info->layout || */
3781 !mddev
->persistent
!= info
->not_persistent
||
3782 mddev
->chunk_size
!= info
->chunk_size
||
3783 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3784 ((state
^info
->state
) & 0xfffffe00)
3787 /* Check there is only one change */
3788 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
3789 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3790 if (mddev
->layout
!= info
->layout
) cnt
++;
3791 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3792 if (cnt
== 0) return 0;
3793 if (cnt
> 1) return -EINVAL
;
3795 if (mddev
->layout
!= info
->layout
) {
3797 * we don't need to do anything at the md level, the
3798 * personality will take care of it all.
3800 if (mddev
->pers
->reconfig
== NULL
)
3803 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3805 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
3806 rv
= update_size(mddev
, info
->size
);
3808 if (mddev
->raid_disks
!= info
->raid_disks
)
3809 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3811 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3812 if (mddev
->pers
->quiesce
== NULL
)
3814 if (mddev
->recovery
|| mddev
->sync_thread
)
3816 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3817 /* add the bitmap */
3820 if (mddev
->default_bitmap_offset
== 0)
3822 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3823 mddev
->pers
->quiesce(mddev
, 1);
3824 rv
= bitmap_create(mddev
);
3826 bitmap_destroy(mddev
);
3827 mddev
->pers
->quiesce(mddev
, 0);
3829 /* remove the bitmap */
3832 if (mddev
->bitmap
->file
)
3834 mddev
->pers
->quiesce(mddev
, 1);
3835 bitmap_destroy(mddev
);
3836 mddev
->pers
->quiesce(mddev
, 0);
3837 mddev
->bitmap_offset
= 0;
3840 md_update_sb(mddev
);
3844 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3848 if (mddev
->pers
== NULL
)
3851 rdev
= find_rdev(mddev
, dev
);
3855 md_error(mddev
, rdev
);
3859 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
3861 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
3865 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
3869 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3870 unsigned int cmd
, unsigned long arg
)
3873 void __user
*argp
= (void __user
*)arg
;
3874 mddev_t
*mddev
= NULL
;
3876 if (!capable(CAP_SYS_ADMIN
))
3880 * Commands dealing with the RAID driver but not any
3886 err
= get_version(argp
);
3889 case PRINT_RAID_DEBUG
:
3897 autostart_arrays(arg
);
3904 * Commands creating/starting a new array:
3907 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3915 if (cmd
== START_ARRAY
) {
3916 /* START_ARRAY doesn't need to lock the array as autostart_array
3917 * does the locking, and it could even be a different array
3922 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3923 "This will not be supported beyond July 2006\n",
3924 current
->comm
, current
->pid
);
3927 err
= autostart_array(new_decode_dev(arg
));
3929 printk(KERN_WARNING
"md: autostart failed!\n");
3935 err
= mddev_lock(mddev
);
3938 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3945 case SET_ARRAY_INFO
:
3947 mdu_array_info_t info
;
3949 memset(&info
, 0, sizeof(info
));
3950 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
3955 err
= update_array_info(mddev
, &info
);
3957 printk(KERN_WARNING
"md: couldn't update"
3958 " array info. %d\n", err
);
3963 if (!list_empty(&mddev
->disks
)) {
3965 "md: array %s already has disks!\n",
3970 if (mddev
->raid_disks
) {
3972 "md: array %s already initialised!\n",
3977 err
= set_array_info(mddev
, &info
);
3979 printk(KERN_WARNING
"md: couldn't set"
3980 " array info. %d\n", err
);
3990 * Commands querying/configuring an existing array:
3992 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3993 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3994 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
3995 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
4001 * Commands even a read-only array can execute:
4005 case GET_ARRAY_INFO
:
4006 err
= get_array_info(mddev
, argp
);
4009 case GET_BITMAP_FILE
:
4010 err
= get_bitmap_file(mddev
, argp
);
4014 err
= get_disk_info(mddev
, argp
);
4017 case RESTART_ARRAY_RW
:
4018 err
= restart_array(mddev
);
4022 err
= do_md_stop (mddev
, 0);
4026 err
= do_md_stop (mddev
, 1);
4030 * We have a problem here : there is no easy way to give a CHS
4031 * virtual geometry. We currently pretend that we have a 2 heads
4032 * 4 sectors (with a BIG number of cylinders...). This drives
4033 * dosfs just mad... ;-)
4038 * The remaining ioctls are changing the state of the
4039 * superblock, so we do not allow them on read-only arrays.
4040 * However non-MD ioctls (e.g. get-size) will still come through
4041 * here and hit the 'default' below, so only disallow
4042 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4044 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4045 mddev
->ro
&& mddev
->pers
) {
4046 if (mddev
->ro
== 2) {
4048 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4049 md_wakeup_thread(mddev
->thread
);
4061 mdu_disk_info_t info
;
4062 if (copy_from_user(&info
, argp
, sizeof(info
)))
4065 err
= add_new_disk(mddev
, &info
);
4069 case HOT_REMOVE_DISK
:
4070 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4074 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4077 case SET_DISK_FAULTY
:
4078 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4082 err
= do_md_run (mddev
);
4085 case SET_BITMAP_FILE
:
4086 err
= set_bitmap_file(mddev
, (int)arg
);
4096 mddev_unlock(mddev
);
4106 static int md_open(struct inode
*inode
, struct file
*file
)
4109 * Succeed if we can lock the mddev, which confirms that
4110 * it isn't being stopped right now.
4112 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4115 if ((err
= mddev_lock(mddev
)))
4120 mddev_unlock(mddev
);
4122 check_disk_change(inode
->i_bdev
);
4127 static int md_release(struct inode
*inode
, struct file
* file
)
4129 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4138 static int md_media_changed(struct gendisk
*disk
)
4140 mddev_t
*mddev
= disk
->private_data
;
4142 return mddev
->changed
;
4145 static int md_revalidate(struct gendisk
*disk
)
4147 mddev_t
*mddev
= disk
->private_data
;
4152 static struct block_device_operations md_fops
=
4154 .owner
= THIS_MODULE
,
4156 .release
= md_release
,
4158 .getgeo
= md_getgeo
,
4159 .media_changed
= md_media_changed
,
4160 .revalidate_disk
= md_revalidate
,
4163 static int md_thread(void * arg
)
4165 mdk_thread_t
*thread
= arg
;
4168 * md_thread is a 'system-thread', it's priority should be very
4169 * high. We avoid resource deadlocks individually in each
4170 * raid personality. (RAID5 does preallocation) We also use RR and
4171 * the very same RT priority as kswapd, thus we will never get
4172 * into a priority inversion deadlock.
4174 * we definitely have to have equal or higher priority than
4175 * bdflush, otherwise bdflush will deadlock if there are too
4176 * many dirty RAID5 blocks.
4179 allow_signal(SIGKILL
);
4180 while (!kthread_should_stop()) {
4182 /* We need to wait INTERRUPTIBLE so that
4183 * we don't add to the load-average.
4184 * That means we need to be sure no signals are
4187 if (signal_pending(current
))
4188 flush_signals(current
);
4190 wait_event_interruptible_timeout
4192 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4193 || kthread_should_stop(),
4197 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4199 thread
->run(thread
->mddev
);
4205 void md_wakeup_thread(mdk_thread_t
*thread
)
4208 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4209 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4210 wake_up(&thread
->wqueue
);
4214 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4217 mdk_thread_t
*thread
;
4219 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4223 init_waitqueue_head(&thread
->wqueue
);
4226 thread
->mddev
= mddev
;
4227 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4228 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4229 if (IS_ERR(thread
->tsk
)) {
4236 void md_unregister_thread(mdk_thread_t
*thread
)
4238 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4240 kthread_stop(thread
->tsk
);
4244 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4251 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4254 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4256 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4257 __builtin_return_address(0),__builtin_return_address(1),
4258 __builtin_return_address(2),__builtin_return_address(3));
4260 if (!mddev
->pers
->error_handler
)
4262 mddev
->pers
->error_handler(mddev
,rdev
);
4263 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4264 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4265 md_wakeup_thread(mddev
->thread
);
4266 md_new_event_inintr(mddev
);
4269 /* seq_file implementation /proc/mdstat */
4271 static void status_unused(struct seq_file
*seq
)
4275 struct list_head
*tmp
;
4277 seq_printf(seq
, "unused devices: ");
4279 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4280 char b
[BDEVNAME_SIZE
];
4282 seq_printf(seq
, "%s ",
4283 bdevname(rdev
->bdev
,b
));
4286 seq_printf(seq
, "<none>");
4288 seq_printf(seq
, "\n");
4292 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4294 sector_t max_blocks
, resync
, res
;
4295 unsigned long dt
, db
, rt
;
4297 unsigned int per_milli
;
4299 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4301 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4302 max_blocks
= mddev
->resync_max_sectors
>> 1;
4304 max_blocks
= mddev
->size
;
4307 * Should not happen.
4313 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4314 * in a sector_t, and (max_blocks>>scale) will fit in a
4315 * u32, as those are the requirements for sector_div.
4316 * Thus 'scale' must be at least 10
4319 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4320 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4323 res
= (resync
>>scale
)*1000;
4324 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4328 int i
, x
= per_milli
/50, y
= 20-x
;
4329 seq_printf(seq
, "[");
4330 for (i
= 0; i
< x
; i
++)
4331 seq_printf(seq
, "=");
4332 seq_printf(seq
, ">");
4333 for (i
= 0; i
< y
; i
++)
4334 seq_printf(seq
, ".");
4335 seq_printf(seq
, "] ");
4337 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4338 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4340 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4341 "resync" : "recovery")),
4342 per_milli
/10, per_milli
% 10,
4343 (unsigned long long) resync
,
4344 (unsigned long long) max_blocks
);
4347 * We do not want to overflow, so the order of operands and
4348 * the * 100 / 100 trick are important. We do a +1 to be
4349 * safe against division by zero. We only estimate anyway.
4351 * dt: time from mark until now
4352 * db: blocks written from mark until now
4353 * rt: remaining time
4355 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4357 db
= resync
- (mddev
->resync_mark_cnt
/2);
4358 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/100+1)))/100;
4360 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4362 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4365 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4367 struct list_head
*tmp
;
4377 spin_lock(&all_mddevs_lock
);
4378 list_for_each(tmp
,&all_mddevs
)
4380 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4382 spin_unlock(&all_mddevs_lock
);
4385 spin_unlock(&all_mddevs_lock
);
4387 return (void*)2;/* tail */
4391 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4393 struct list_head
*tmp
;
4394 mddev_t
*next_mddev
, *mddev
= v
;
4400 spin_lock(&all_mddevs_lock
);
4402 tmp
= all_mddevs
.next
;
4404 tmp
= mddev
->all_mddevs
.next
;
4405 if (tmp
!= &all_mddevs
)
4406 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4408 next_mddev
= (void*)2;
4411 spin_unlock(&all_mddevs_lock
);
4419 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4423 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4427 struct mdstat_info
{
4431 static int md_seq_show(struct seq_file
*seq
, void *v
)
4435 struct list_head
*tmp2
;
4437 struct mdstat_info
*mi
= seq
->private;
4438 struct bitmap
*bitmap
;
4440 if (v
== (void*)1) {
4441 struct mdk_personality
*pers
;
4442 seq_printf(seq
, "Personalities : ");
4443 spin_lock(&pers_lock
);
4444 list_for_each_entry(pers
, &pers_list
, list
)
4445 seq_printf(seq
, "[%s] ", pers
->name
);
4447 spin_unlock(&pers_lock
);
4448 seq_printf(seq
, "\n");
4449 mi
->event
= atomic_read(&md_event_count
);
4452 if (v
== (void*)2) {
4457 if (mddev_lock(mddev
) < 0)
4460 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4461 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4462 mddev
->pers
? "" : "in");
4465 seq_printf(seq
, " (read-only)");
4467 seq_printf(seq
, "(auto-read-only)");
4468 seq_printf(seq
, " %s", mddev
->pers
->name
);
4472 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4473 char b
[BDEVNAME_SIZE
];
4474 seq_printf(seq
, " %s[%d]",
4475 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4476 if (test_bit(WriteMostly
, &rdev
->flags
))
4477 seq_printf(seq
, "(W)");
4478 if (test_bit(Faulty
, &rdev
->flags
)) {
4479 seq_printf(seq
, "(F)");
4481 } else if (rdev
->raid_disk
< 0)
4482 seq_printf(seq
, "(S)"); /* spare */
4486 if (!list_empty(&mddev
->disks
)) {
4488 seq_printf(seq
, "\n %llu blocks",
4489 (unsigned long long)mddev
->array_size
);
4491 seq_printf(seq
, "\n %llu blocks",
4492 (unsigned long long)size
);
4494 if (mddev
->persistent
) {
4495 if (mddev
->major_version
!= 0 ||
4496 mddev
->minor_version
!= 90) {
4497 seq_printf(seq
," super %d.%d",
4498 mddev
->major_version
,
4499 mddev
->minor_version
);
4502 seq_printf(seq
, " super non-persistent");
4505 mddev
->pers
->status (seq
, mddev
);
4506 seq_printf(seq
, "\n ");
4507 if (mddev
->pers
->sync_request
) {
4508 if (mddev
->curr_resync
> 2) {
4509 status_resync (seq
, mddev
);
4510 seq_printf(seq
, "\n ");
4511 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4512 seq_printf(seq
, "\tresync=DELAYED\n ");
4513 else if (mddev
->recovery_cp
< MaxSector
)
4514 seq_printf(seq
, "\tresync=PENDING\n ");
4517 seq_printf(seq
, "\n ");
4519 if ((bitmap
= mddev
->bitmap
)) {
4520 unsigned long chunk_kb
;
4521 unsigned long flags
;
4522 spin_lock_irqsave(&bitmap
->lock
, flags
);
4523 chunk_kb
= bitmap
->chunksize
>> 10;
4524 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4526 bitmap
->pages
- bitmap
->missing_pages
,
4528 (bitmap
->pages
- bitmap
->missing_pages
)
4529 << (PAGE_SHIFT
- 10),
4530 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4531 chunk_kb
? "KB" : "B");
4533 seq_printf(seq
, ", file: ");
4534 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4535 bitmap
->file
->f_dentry
," \t\n");
4538 seq_printf(seq
, "\n");
4539 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4542 seq_printf(seq
, "\n");
4544 mddev_unlock(mddev
);
4549 static struct seq_operations md_seq_ops
= {
4550 .start
= md_seq_start
,
4551 .next
= md_seq_next
,
4552 .stop
= md_seq_stop
,
4553 .show
= md_seq_show
,
4556 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4559 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4563 error
= seq_open(file
, &md_seq_ops
);
4567 struct seq_file
*p
= file
->private_data
;
4569 mi
->event
= atomic_read(&md_event_count
);
4574 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4576 struct seq_file
*m
= file
->private_data
;
4577 struct mdstat_info
*mi
= m
->private;
4580 return seq_release(inode
, file
);
4583 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4585 struct seq_file
*m
= filp
->private_data
;
4586 struct mdstat_info
*mi
= m
->private;
4589 poll_wait(filp
, &md_event_waiters
, wait
);
4591 /* always allow read */
4592 mask
= POLLIN
| POLLRDNORM
;
4594 if (mi
->event
!= atomic_read(&md_event_count
))
4595 mask
|= POLLERR
| POLLPRI
;
4599 static struct file_operations md_seq_fops
= {
4600 .open
= md_seq_open
,
4602 .llseek
= seq_lseek
,
4603 .release
= md_seq_release
,
4604 .poll
= mdstat_poll
,
4607 int register_md_personality(struct mdk_personality
*p
)
4609 spin_lock(&pers_lock
);
4610 list_add_tail(&p
->list
, &pers_list
);
4611 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4612 spin_unlock(&pers_lock
);
4616 int unregister_md_personality(struct mdk_personality
*p
)
4618 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4619 spin_lock(&pers_lock
);
4620 list_del_init(&p
->list
);
4621 spin_unlock(&pers_lock
);
4625 static int is_mddev_idle(mddev_t
*mddev
)
4628 struct list_head
*tmp
;
4630 unsigned long curr_events
;
4633 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4634 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4635 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4636 disk_stat_read(disk
, sectors
[1]) -
4637 atomic_read(&disk
->sync_io
);
4638 /* The difference between curr_events and last_events
4639 * will be affected by any new non-sync IO (making
4640 * curr_events bigger) and any difference in the amount of
4641 * in-flight syncio (making current_events bigger or smaller)
4642 * The amount in-flight is currently limited to
4643 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4644 * which is at most 4096 sectors.
4645 * These numbers are fairly fragile and should be made
4646 * more robust, probably by enforcing the
4647 * 'window size' that md_do_sync sort-of uses.
4649 * Note: the following is an unsigned comparison.
4651 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4652 rdev
->last_events
= curr_events
;
4659 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4661 /* another "blocks" (512byte) blocks have been synced */
4662 atomic_sub(blocks
, &mddev
->recovery_active
);
4663 wake_up(&mddev
->recovery_wait
);
4665 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4666 md_wakeup_thread(mddev
->thread
);
4667 // stop recovery, signal do_sync ....
4672 /* md_write_start(mddev, bi)
4673 * If we need to update some array metadata (e.g. 'active' flag
4674 * in superblock) before writing, schedule a superblock update
4675 * and wait for it to complete.
4677 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4679 if (bio_data_dir(bi
) != WRITE
)
4682 BUG_ON(mddev
->ro
== 1);
4683 if (mddev
->ro
== 2) {
4684 /* need to switch to read/write */
4686 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4687 md_wakeup_thread(mddev
->thread
);
4689 atomic_inc(&mddev
->writes_pending
);
4690 if (mddev
->in_sync
) {
4691 spin_lock_irq(&mddev
->write_lock
);
4692 if (mddev
->in_sync
) {
4694 mddev
->sb_dirty
= 1;
4695 md_wakeup_thread(mddev
->thread
);
4697 spin_unlock_irq(&mddev
->write_lock
);
4699 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4702 void md_write_end(mddev_t
*mddev
)
4704 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4705 if (mddev
->safemode
== 2)
4706 md_wakeup_thread(mddev
->thread
);
4707 else if (mddev
->safemode_delay
)
4708 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4712 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4714 #define SYNC_MARKS 10
4715 #define SYNC_MARK_STEP (3*HZ)
4716 void md_do_sync(mddev_t
*mddev
)
4719 unsigned int currspeed
= 0,
4721 sector_t max_sectors
,j
, io_sectors
;
4722 unsigned long mark
[SYNC_MARKS
];
4723 sector_t mark_cnt
[SYNC_MARKS
];
4725 struct list_head
*tmp
;
4726 sector_t last_check
;
4728 struct list_head
*rtmp
;
4731 /* just incase thread restarts... */
4732 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4734 if (mddev
->ro
) /* never try to sync a read-only array */
4737 /* we overload curr_resync somewhat here.
4738 * 0 == not engaged in resync at all
4739 * 2 == checking that there is no conflict with another sync
4740 * 1 == like 2, but have yielded to allow conflicting resync to
4742 * other == active in resync - this many blocks
4744 * Before starting a resync we must have set curr_resync to
4745 * 2, and then checked that every "conflicting" array has curr_resync
4746 * less than ours. When we find one that is the same or higher
4747 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4748 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4749 * This will mean we have to start checking from the beginning again.
4754 mddev
->curr_resync
= 2;
4757 if (kthread_should_stop()) {
4758 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4761 ITERATE_MDDEV(mddev2
,tmp
) {
4762 if (mddev2
== mddev
)
4764 if (mddev2
->curr_resync
&&
4765 match_mddev_units(mddev
,mddev2
)) {
4767 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4768 /* arbitrarily yield */
4769 mddev
->curr_resync
= 1;
4770 wake_up(&resync_wait
);
4772 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4773 /* no need to wait here, we can wait the next
4774 * time 'round when curr_resync == 2
4777 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4778 if (!kthread_should_stop() &&
4779 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4780 printk(KERN_INFO
"md: delaying resync of %s"
4781 " until %s has finished resync (they"
4782 " share one or more physical units)\n",
4783 mdname(mddev
), mdname(mddev2
));
4786 finish_wait(&resync_wait
, &wq
);
4789 finish_wait(&resync_wait
, &wq
);
4792 } while (mddev
->curr_resync
< 2);
4795 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4796 /* resync follows the size requested by the personality,
4797 * which defaults to physical size, but can be virtual size
4799 max_sectors
= mddev
->resync_max_sectors
;
4800 mddev
->resync_mismatches
= 0;
4801 /* we don't use the checkpoint if there's a bitmap */
4802 if (!mddev
->bitmap
&&
4803 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4804 j
= mddev
->recovery_cp
;
4805 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4806 max_sectors
= mddev
->size
<< 1;
4808 /* recovery follows the physical size of devices */
4809 max_sectors
= mddev
->size
<< 1;
4811 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4812 if (rdev
->raid_disk
>= 0 &&
4813 !test_bit(Faulty
, &rdev
->flags
) &&
4814 !test_bit(In_sync
, &rdev
->flags
) &&
4815 rdev
->recovery_offset
< j
)
4816 j
= rdev
->recovery_offset
;
4819 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4820 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4821 " %d KB/sec/disc.\n", speed_min(mddev
));
4822 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4823 "(but not more than %d KB/sec) for reconstruction.\n",
4826 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4829 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4831 mark_cnt
[m
] = io_sectors
;
4834 mddev
->resync_mark
= mark
[last_mark
];
4835 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4838 * Tune reconstruction:
4840 window
= 32*(PAGE_SIZE
/512);
4841 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4842 window
/2,(unsigned long long) max_sectors
/2);
4844 atomic_set(&mddev
->recovery_active
, 0);
4845 init_waitqueue_head(&mddev
->recovery_wait
);
4850 "md: resuming recovery of %s from checkpoint.\n",
4852 mddev
->curr_resync
= j
;
4855 while (j
< max_sectors
) {
4859 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4860 currspeed
< speed_min(mddev
));
4862 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4866 if (!skipped
) { /* actual IO requested */
4867 io_sectors
+= sectors
;
4868 atomic_add(sectors
, &mddev
->recovery_active
);
4872 if (j
>1) mddev
->curr_resync
= j
;
4873 if (last_check
== 0)
4874 /* this is the earliers that rebuilt will be
4875 * visible in /proc/mdstat
4877 md_new_event(mddev
);
4879 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4882 last_check
= io_sectors
;
4884 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4885 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4889 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4891 int next
= (last_mark
+1) % SYNC_MARKS
;
4893 mddev
->resync_mark
= mark
[next
];
4894 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4895 mark
[next
] = jiffies
;
4896 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4901 if (kthread_should_stop()) {
4903 * got a signal, exit.
4906 "md: md_do_sync() got signal ... exiting\n");
4907 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4912 * this loop exits only if either when we are slower than
4913 * the 'hard' speed limit, or the system was IO-idle for
4915 * the system might be non-idle CPU-wise, but we only care
4916 * about not overloading the IO subsystem. (things like an
4917 * e2fsck being done on the RAID array should execute fast)
4919 mddev
->queue
->unplug_fn(mddev
->queue
);
4922 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4923 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4925 if (currspeed
> speed_min(mddev
)) {
4926 if ((currspeed
> speed_max(mddev
)) ||
4927 !is_mddev_idle(mddev
)) {
4933 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
4935 * this also signals 'finished resyncing' to md_stop
4938 mddev
->queue
->unplug_fn(mddev
->queue
);
4940 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
4942 /* tell personality that we are finished */
4943 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
4945 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4946 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
4947 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
4948 mddev
->curr_resync
> 2) {
4949 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4950 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4951 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
4953 "md: checkpointing recovery of %s.\n",
4955 mddev
->recovery_cp
= mddev
->curr_resync
;
4958 mddev
->recovery_cp
= MaxSector
;
4960 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
4961 mddev
->curr_resync
= MaxSector
;
4962 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4963 if (rdev
->raid_disk
>= 0 &&
4964 !test_bit(Faulty
, &rdev
->flags
) &&
4965 !test_bit(In_sync
, &rdev
->flags
) &&
4966 rdev
->recovery_offset
< mddev
->curr_resync
)
4967 rdev
->recovery_offset
= mddev
->curr_resync
;
4968 mddev
->sb_dirty
= 1;
4973 mddev
->curr_resync
= 0;
4974 wake_up(&resync_wait
);
4975 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4976 md_wakeup_thread(mddev
->thread
);
4978 EXPORT_SYMBOL_GPL(md_do_sync
);
4982 * This routine is regularly called by all per-raid-array threads to
4983 * deal with generic issues like resync and super-block update.
4984 * Raid personalities that don't have a thread (linear/raid0) do not
4985 * need this as they never do any recovery or update the superblock.
4987 * It does not do any resync itself, but rather "forks" off other threads
4988 * to do that as needed.
4989 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4990 * "->recovery" and create a thread at ->sync_thread.
4991 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4992 * and wakeups up this thread which will reap the thread and finish up.
4993 * This thread also removes any faulty devices (with nr_pending == 0).
4995 * The overall approach is:
4996 * 1/ if the superblock needs updating, update it.
4997 * 2/ If a recovery thread is running, don't do anything else.
4998 * 3/ If recovery has finished, clean up, possibly marking spares active.
4999 * 4/ If there are any faulty devices, remove them.
5000 * 5/ If array is degraded, try to add spares devices
5001 * 6/ If array has spares or is not in-sync, start a resync thread.
5003 void md_check_recovery(mddev_t
*mddev
)
5006 struct list_head
*rtmp
;
5010 bitmap_daemon_work(mddev
->bitmap
);
5015 if (signal_pending(current
)) {
5016 if (mddev
->pers
->sync_request
) {
5017 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5019 mddev
->safemode
= 2;
5021 flush_signals(current
);
5026 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5027 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5028 (mddev
->safemode
== 1) ||
5029 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5030 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5034 if (mddev_trylock(mddev
)) {
5037 spin_lock_irq(&mddev
->write_lock
);
5038 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5039 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5041 mddev
->sb_dirty
= 1;
5043 if (mddev
->safemode
== 1)
5044 mddev
->safemode
= 0;
5045 spin_unlock_irq(&mddev
->write_lock
);
5047 if (mddev
->sb_dirty
)
5048 md_update_sb(mddev
);
5051 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5052 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5053 /* resync/recovery still happening */
5054 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5057 if (mddev
->sync_thread
) {
5058 /* resync has finished, collect result */
5059 md_unregister_thread(mddev
->sync_thread
);
5060 mddev
->sync_thread
= NULL
;
5061 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5062 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5064 /* activate any spares */
5065 mddev
->pers
->spare_active(mddev
);
5067 md_update_sb(mddev
);
5069 /* if array is no-longer degraded, then any saved_raid_disk
5070 * information must be scrapped
5072 if (!mddev
->degraded
)
5073 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5074 rdev
->saved_raid_disk
= -1;
5076 mddev
->recovery
= 0;
5077 /* flag recovery needed just to double check */
5078 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5079 md_new_event(mddev
);
5082 /* Clear some bits that don't mean anything, but
5085 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5086 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5087 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5088 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5090 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5092 /* no recovery is running.
5093 * remove any failed drives, then
5094 * add spares if possible.
5095 * Spare are also removed and re-added, to allow
5096 * the personality to fail the re-add.
5098 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5099 if (rdev
->raid_disk
>= 0 &&
5100 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
5101 atomic_read(&rdev
->nr_pending
)==0) {
5102 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
5104 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5105 sysfs_remove_link(&mddev
->kobj
, nm
);
5106 rdev
->raid_disk
= -1;
5110 if (mddev
->degraded
) {
5111 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5112 if (rdev
->raid_disk
< 0
5113 && !test_bit(Faulty
, &rdev
->flags
)) {
5114 rdev
->recovery_offset
= 0;
5115 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5117 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5118 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
5120 md_new_event(mddev
);
5127 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5128 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5129 } else if (mddev
->recovery_cp
< MaxSector
) {
5130 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5131 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5132 /* nothing to be done ... */
5135 if (mddev
->pers
->sync_request
) {
5136 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5137 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5138 /* We are adding a device or devices to an array
5139 * which has the bitmap stored on all devices.
5140 * So make sure all bitmap pages get written
5142 bitmap_write_all(mddev
->bitmap
);
5144 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5147 if (!mddev
->sync_thread
) {
5148 printk(KERN_ERR
"%s: could not start resync"
5151 /* leave the spares where they are, it shouldn't hurt */
5152 mddev
->recovery
= 0;
5154 md_wakeup_thread(mddev
->sync_thread
);
5155 md_new_event(mddev
);
5158 mddev_unlock(mddev
);
5162 static int md_notify_reboot(struct notifier_block
*this,
5163 unsigned long code
, void *x
)
5165 struct list_head
*tmp
;
5168 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5170 printk(KERN_INFO
"md: stopping all md devices.\n");
5172 ITERATE_MDDEV(mddev
,tmp
)
5173 if (mddev_trylock(mddev
)) {
5174 do_md_stop (mddev
, 1);
5175 mddev_unlock(mddev
);
5178 * certain more exotic SCSI devices are known to be
5179 * volatile wrt too early system reboots. While the
5180 * right place to handle this issue is the given
5181 * driver, we do want to have a safe RAID driver ...
5188 static struct notifier_block md_notifier
= {
5189 .notifier_call
= md_notify_reboot
,
5191 .priority
= INT_MAX
, /* before any real devices */
5194 static void md_geninit(void)
5196 struct proc_dir_entry
*p
;
5198 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5200 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5202 p
->proc_fops
= &md_seq_fops
;
5205 static int __init
md_init(void)
5209 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5210 " MD_SB_DISKS=%d\n",
5211 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
5212 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
5213 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
5216 if (register_blkdev(MAJOR_NR
, "md"))
5218 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5219 unregister_blkdev(MAJOR_NR
, "md");
5223 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
5224 md_probe
, NULL
, NULL
);
5225 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
5226 md_probe
, NULL
, NULL
);
5228 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5229 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
5230 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5233 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5234 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
5235 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5239 register_reboot_notifier(&md_notifier
);
5240 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
5250 * Searches all registered partitions for autorun RAID arrays
5253 static dev_t detected_devices
[128];
5256 void md_autodetect_dev(dev_t dev
)
5258 if (dev_cnt
>= 0 && dev_cnt
< 127)
5259 detected_devices
[dev_cnt
++] = dev
;
5263 static void autostart_arrays(int part
)
5268 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5270 for (i
= 0; i
< dev_cnt
; i
++) {
5271 dev_t dev
= detected_devices
[i
];
5273 rdev
= md_import_device(dev
,0, 0);
5277 if (test_bit(Faulty
, &rdev
->flags
)) {
5281 list_add(&rdev
->same_set
, &pending_raid_disks
);
5285 autorun_devices(part
);
5290 static __exit
void md_exit(void)
5293 struct list_head
*tmp
;
5295 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
5296 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
5297 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5298 devfs_remove("md/%d", i
);
5299 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5300 devfs_remove("md/d%d", i
);
5304 unregister_blkdev(MAJOR_NR
,"md");
5305 unregister_blkdev(mdp_major
, "mdp");
5306 unregister_reboot_notifier(&md_notifier
);
5307 unregister_sysctl_table(raid_table_header
);
5308 remove_proc_entry("mdstat", NULL
);
5309 ITERATE_MDDEV(mddev
,tmp
) {
5310 struct gendisk
*disk
= mddev
->gendisk
;
5313 export_array(mddev
);
5316 mddev
->gendisk
= NULL
;
5321 module_init(md_init
)
5322 module_exit(md_exit
)
5324 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5326 return sprintf(buffer
, "%d", start_readonly
);
5328 static int set_ro(const char *val
, struct kernel_param
*kp
)
5331 int num
= simple_strtoul(val
, &e
, 10);
5332 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5333 start_readonly
= num
;
5339 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5340 module_param(start_dirty_degraded
, int, 0644);
5343 EXPORT_SYMBOL(register_md_personality
);
5344 EXPORT_SYMBOL(unregister_md_personality
);
5345 EXPORT_SYMBOL(md_error
);
5346 EXPORT_SYMBOL(md_done_sync
);
5347 EXPORT_SYMBOL(md_write_start
);
5348 EXPORT_SYMBOL(md_write_end
);
5349 EXPORT_SYMBOL(md_register_thread
);
5350 EXPORT_SYMBOL(md_unregister_thread
);
5351 EXPORT_SYMBOL(md_wakeup_thread
);
5352 EXPORT_SYMBOL(md_check_recovery
);
5353 MODULE_LICENSE("GPL");
5355 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);