2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/ctype.h>
49 #include <linux/init.h>
51 #include <linux/file.h>
54 #include <linux/kmod.h>
57 #include <asm/unaligned.h>
59 #define MAJOR_NR MD_MAJOR
62 /* 63 partitions with the alternate major number (mdp) */
63 #define MdpMinorShift 6
66 #define dprintk(x...) ((void)(DEBUG && printk(x)))
70 static void autostart_arrays (int part
);
73 static LIST_HEAD(pers_list
);
74 static DEFINE_SPINLOCK(pers_lock
);
76 static void md_print_devices(void);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
93 static int sysctl_speed_limit_min
= 1000;
94 static int sysctl_speed_limit_max
= 200000;
95 static inline int speed_min(mddev_t
*mddev
)
97 return mddev
->sync_speed_min
?
98 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
101 static inline int speed_max(mddev_t
*mddev
)
103 return mddev
->sync_speed_max
?
104 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
107 static struct ctl_table_header
*raid_table_header
;
109 static ctl_table raid_table
[] = {
111 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
112 .procname
= "speed_limit_min",
113 .data
= &sysctl_speed_limit_min
,
114 .maxlen
= sizeof(int),
116 .proc_handler
= &proc_dointvec
,
119 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
120 .procname
= "speed_limit_max",
121 .data
= &sysctl_speed_limit_max
,
122 .maxlen
= sizeof(int),
124 .proc_handler
= &proc_dointvec
,
129 static ctl_table raid_dir_table
[] = {
131 .ctl_name
= DEV_RAID
,
140 static ctl_table raid_root_table
[] = {
146 .child
= raid_dir_table
,
151 static struct block_device_operations md_fops
;
153 static int start_readonly
;
156 * We have a system wide 'event count' that is incremented
157 * on any 'interesting' event, and readers of /proc/mdstat
158 * can use 'poll' or 'select' to find out when the event
162 * start array, stop array, error, add device, remove device,
163 * start build, activate spare
165 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
166 static atomic_t md_event_count
;
167 void md_new_event(mddev_t
*mddev
)
169 atomic_inc(&md_event_count
);
170 wake_up(&md_event_waiters
);
171 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
173 EXPORT_SYMBOL_GPL(md_new_event
);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 void md_new_event_inintr(mddev_t
*mddev
)
180 atomic_inc(&md_event_count
);
181 wake_up(&md_event_waiters
);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs
);
189 static DEFINE_SPINLOCK(all_mddevs_lock
);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define ITERATE_MDDEV(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
217 bio_io_error(bio
, bio
->bi_size
);
221 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
223 atomic_inc(&mddev
->active
);
227 static void mddev_put(mddev_t
*mddev
)
229 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
231 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
232 list_del(&mddev
->all_mddevs
);
233 spin_unlock(&all_mddevs_lock
);
234 blk_cleanup_queue(mddev
->queue
);
235 kobject_unregister(&mddev
->kobj
);
237 spin_unlock(&all_mddevs_lock
);
240 static mddev_t
* mddev_find(dev_t unit
)
242 mddev_t
*mddev
, *new = NULL
;
245 spin_lock(&all_mddevs_lock
);
246 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
247 if (mddev
->unit
== unit
) {
249 spin_unlock(&all_mddevs_lock
);
255 list_add(&new->all_mddevs
, &all_mddevs
);
256 spin_unlock(&all_mddevs_lock
);
259 spin_unlock(&all_mddevs_lock
);
261 new = kzalloc(sizeof(*new), GFP_KERNEL
);
266 if (MAJOR(unit
) == MD_MAJOR
)
267 new->md_minor
= MINOR(unit
);
269 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
271 mutex_init(&new->reconfig_mutex
);
272 INIT_LIST_HEAD(&new->disks
);
273 INIT_LIST_HEAD(&new->all_mddevs
);
274 init_timer(&new->safemode_timer
);
275 atomic_set(&new->active
, 1);
276 spin_lock_init(&new->write_lock
);
277 init_waitqueue_head(&new->sb_wait
);
279 new->queue
= blk_alloc_queue(GFP_KERNEL
);
284 set_bit(QUEUE_FLAG_CLUSTER
, &new->queue
->queue_flags
);
286 blk_queue_make_request(new->queue
, md_fail_request
);
291 static inline int mddev_lock(mddev_t
* mddev
)
293 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
296 static inline int mddev_trylock(mddev_t
* mddev
)
298 return mutex_trylock(&mddev
->reconfig_mutex
);
301 static inline void mddev_unlock(mddev_t
* mddev
)
303 mutex_unlock(&mddev
->reconfig_mutex
);
305 md_wakeup_thread(mddev
->thread
);
308 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
311 struct list_head
*tmp
;
313 ITERATE_RDEV(mddev
,rdev
,tmp
) {
314 if (rdev
->desc_nr
== nr
)
320 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
322 struct list_head
*tmp
;
325 ITERATE_RDEV(mddev
,rdev
,tmp
) {
326 if (rdev
->bdev
->bd_dev
== dev
)
332 static struct mdk_personality
*find_pers(int level
, char *clevel
)
334 struct mdk_personality
*pers
;
335 list_for_each_entry(pers
, &pers_list
, list
) {
336 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
338 if (strcmp(pers
->name
, clevel
)==0)
344 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
346 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
347 return MD_NEW_SIZE_BLOCKS(size
);
350 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
354 size
= rdev
->sb_offset
;
357 size
&= ~((sector_t
)chunk_size
/1024 - 1);
361 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
366 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
367 if (!rdev
->sb_page
) {
368 printk(KERN_ALERT
"md: out of memory.\n");
375 static void free_disk_sb(mdk_rdev_t
* rdev
)
378 put_page(rdev
->sb_page
);
380 rdev
->sb_page
= NULL
;
387 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
389 mdk_rdev_t
*rdev
= bio
->bi_private
;
390 mddev_t
*mddev
= rdev
->mddev
;
394 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
395 md_error(mddev
, rdev
);
397 if (atomic_dec_and_test(&mddev
->pending_writes
))
398 wake_up(&mddev
->sb_wait
);
403 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
405 struct bio
*bio2
= bio
->bi_private
;
406 mdk_rdev_t
*rdev
= bio2
->bi_private
;
407 mddev_t
*mddev
= rdev
->mddev
;
411 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
412 error
== -EOPNOTSUPP
) {
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp
, &rdev
->flags
);
416 mddev
->barriers_work
= 0;
417 spin_lock_irqsave(&mddev
->write_lock
, flags
);
418 bio2
->bi_next
= mddev
->biolist
;
419 mddev
->biolist
= bio2
;
420 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
421 wake_up(&mddev
->sb_wait
);
426 bio
->bi_private
= rdev
;
427 return super_written(bio
, bytes_done
, error
);
430 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
431 sector_t sector
, int size
, struct page
*page
)
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
442 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
443 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
445 bio
->bi_bdev
= rdev
->bdev
;
446 bio
->bi_sector
= sector
;
447 bio_add_page(bio
, page
, size
, 0);
448 bio
->bi_private
= rdev
;
449 bio
->bi_end_io
= super_written
;
452 atomic_inc(&mddev
->pending_writes
);
453 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
455 rw
|= (1<<BIO_RW_BARRIER
);
456 rbio
= bio_clone(bio
, GFP_NOIO
);
457 rbio
->bi_private
= bio
;
458 rbio
->bi_end_io
= super_written_barrier
;
459 submit_bio(rw
, rbio
);
464 void md_super_wait(mddev_t
*mddev
)
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
471 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
472 if (atomic_read(&mddev
->pending_writes
)==0)
474 while (mddev
->biolist
) {
476 spin_lock_irq(&mddev
->write_lock
);
477 bio
= mddev
->biolist
;
478 mddev
->biolist
= bio
->bi_next
;
480 spin_unlock_irq(&mddev
->write_lock
);
481 submit_bio(bio
->bi_rw
, bio
);
485 finish_wait(&mddev
->sb_wait
, &wq
);
488 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
493 complete((struct completion
*)bio
->bi_private
);
497 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
498 struct page
*page
, int rw
)
500 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
501 struct completion event
;
504 rw
|= (1 << BIO_RW_SYNC
);
507 bio
->bi_sector
= sector
;
508 bio_add_page(bio
, page
, size
, 0);
509 init_completion(&event
);
510 bio
->bi_private
= &event
;
511 bio
->bi_end_io
= bi_complete
;
513 wait_for_completion(&event
);
515 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
519 EXPORT_SYMBOL_GPL(sync_page_io
);
521 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
523 char b
[BDEVNAME_SIZE
];
524 if (!rdev
->sb_page
) {
532 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
538 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
539 bdevname(rdev
->bdev
,b
));
543 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
545 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
546 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
547 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
548 (sb1
->set_uuid3
== sb2
->set_uuid3
))
556 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
559 mdp_super_t
*tmp1
, *tmp2
;
561 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
562 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
564 if (!tmp1
|| !tmp2
) {
566 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
574 * nr_disks is not constant
579 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
590 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
592 unsigned int disk_csum
, csum
;
594 disk_csum
= sb
->sb_csum
;
596 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
597 sb
->sb_csum
= disk_csum
;
603 * Handle superblock details.
604 * We want to be able to handle multiple superblock formats
605 * so we have a common interface to them all, and an array of
606 * different handlers.
607 * We rely on user-space to write the initial superblock, and support
608 * reading and updating of superblocks.
609 * Interface methods are:
610 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
611 * loads and validates a superblock on dev.
612 * if refdev != NULL, compare superblocks on both devices
614 * 0 - dev has a superblock that is compatible with refdev
615 * 1 - dev has a superblock that is compatible and newer than refdev
616 * so dev should be used as the refdev in future
617 * -EINVAL superblock incompatible or invalid
618 * -othererror e.g. -EIO
620 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
621 * Verify that dev is acceptable into mddev.
622 * The first time, mddev->raid_disks will be 0, and data from
623 * dev should be merged in. Subsequent calls check that dev
624 * is new enough. Return 0 or -EINVAL
626 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
627 * Update the superblock for rdev with data in mddev
628 * This does not write to disc.
634 struct module
*owner
;
635 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
636 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
637 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
641 * load_super for 0.90.0
643 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
645 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
651 * Calculate the position of the superblock,
652 * it's at the end of the disk.
654 * It also happens to be a multiple of 4Kb.
656 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
657 rdev
->sb_offset
= sb_offset
;
659 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
664 bdevname(rdev
->bdev
, b
);
665 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
667 if (sb
->md_magic
!= MD_SB_MAGIC
) {
668 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
673 if (sb
->major_version
!= 0 ||
674 sb
->minor_version
< 90 ||
675 sb
->minor_version
> 91) {
676 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
677 sb
->major_version
, sb
->minor_version
,
682 if (sb
->raid_disks
<= 0)
685 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
686 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
691 rdev
->preferred_minor
= sb
->md_minor
;
692 rdev
->data_offset
= 0;
693 rdev
->sb_size
= MD_SB_BYTES
;
695 if (sb
->level
== LEVEL_MULTIPATH
)
698 rdev
->desc_nr
= sb
->this_disk
.number
;
704 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
705 if (!uuid_equal(refsb
, sb
)) {
706 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
707 b
, bdevname(refdev
->bdev
,b2
));
710 if (!sb_equal(refsb
, sb
)) {
711 printk(KERN_WARNING
"md: %s has same UUID"
712 " but different superblock to %s\n",
713 b
, bdevname(refdev
->bdev
, b2
));
717 ev2
= md_event(refsb
);
723 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
725 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
726 /* "this cannot possibly happen" ... */
734 * validate_super for 0.90.0
736 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
739 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
740 __u64 ev1
= md_event(sb
);
742 rdev
->raid_disk
= -1;
744 if (mddev
->raid_disks
== 0) {
745 mddev
->major_version
= 0;
746 mddev
->minor_version
= sb
->minor_version
;
747 mddev
->patch_version
= sb
->patch_version
;
748 mddev
->persistent
= ! sb
->not_persistent
;
749 mddev
->chunk_size
= sb
->chunk_size
;
750 mddev
->ctime
= sb
->ctime
;
751 mddev
->utime
= sb
->utime
;
752 mddev
->level
= sb
->level
;
753 mddev
->clevel
[0] = 0;
754 mddev
->layout
= sb
->layout
;
755 mddev
->raid_disks
= sb
->raid_disks
;
756 mddev
->size
= sb
->size
;
758 mddev
->bitmap_offset
= 0;
759 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
761 if (mddev
->minor_version
>= 91) {
762 mddev
->reshape_position
= sb
->reshape_position
;
763 mddev
->delta_disks
= sb
->delta_disks
;
764 mddev
->new_level
= sb
->new_level
;
765 mddev
->new_layout
= sb
->new_layout
;
766 mddev
->new_chunk
= sb
->new_chunk
;
768 mddev
->reshape_position
= MaxSector
;
769 mddev
->delta_disks
= 0;
770 mddev
->new_level
= mddev
->level
;
771 mddev
->new_layout
= mddev
->layout
;
772 mddev
->new_chunk
= mddev
->chunk_size
;
775 if (sb
->state
& (1<<MD_SB_CLEAN
))
776 mddev
->recovery_cp
= MaxSector
;
778 if (sb
->events_hi
== sb
->cp_events_hi
&&
779 sb
->events_lo
== sb
->cp_events_lo
) {
780 mddev
->recovery_cp
= sb
->recovery_cp
;
782 mddev
->recovery_cp
= 0;
785 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
786 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
787 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
788 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
790 mddev
->max_disks
= MD_SB_DISKS
;
792 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
793 mddev
->bitmap_file
== NULL
) {
794 if (mddev
->level
!= 1 && mddev
->level
!= 4
795 && mddev
->level
!= 5 && mddev
->level
!= 6
796 && mddev
->level
!= 10) {
797 /* FIXME use a better test */
798 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
801 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
804 } else if (mddev
->pers
== NULL
) {
805 /* Insist on good event counter while assembling */
807 if (ev1
< mddev
->events
)
809 } else if (mddev
->bitmap
) {
810 /* if adding to array with a bitmap, then we can accept an
811 * older device ... but not too old.
813 if (ev1
< mddev
->bitmap
->events_cleared
)
816 if (ev1
< mddev
->events
)
817 /* just a hot-add of a new device, leave raid_disk at -1 */
821 if (mddev
->level
!= LEVEL_MULTIPATH
) {
822 desc
= sb
->disks
+ rdev
->desc_nr
;
824 if (desc
->state
& (1<<MD_DISK_FAULTY
))
825 set_bit(Faulty
, &rdev
->flags
);
826 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
827 desc->raid_disk < mddev->raid_disks */) {
828 set_bit(In_sync
, &rdev
->flags
);
829 rdev
->raid_disk
= desc
->raid_disk
;
831 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
832 set_bit(WriteMostly
, &rdev
->flags
);
833 } else /* MULTIPATH are always insync */
834 set_bit(In_sync
, &rdev
->flags
);
839 * sync_super for 0.90.0
841 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
844 struct list_head
*tmp
;
846 int next_spare
= mddev
->raid_disks
;
849 /* make rdev->sb match mddev data..
852 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
853 * 3/ any empty disks < next_spare become removed
855 * disks[0] gets initialised to REMOVED because
856 * we cannot be sure from other fields if it has
857 * been initialised or not.
860 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
862 rdev
->sb_size
= MD_SB_BYTES
;
864 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
866 memset(sb
, 0, sizeof(*sb
));
868 sb
->md_magic
= MD_SB_MAGIC
;
869 sb
->major_version
= mddev
->major_version
;
870 sb
->patch_version
= mddev
->patch_version
;
871 sb
->gvalid_words
= 0; /* ignored */
872 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
873 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
874 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
875 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
877 sb
->ctime
= mddev
->ctime
;
878 sb
->level
= mddev
->level
;
879 sb
->size
= mddev
->size
;
880 sb
->raid_disks
= mddev
->raid_disks
;
881 sb
->md_minor
= mddev
->md_minor
;
882 sb
->not_persistent
= !mddev
->persistent
;
883 sb
->utime
= mddev
->utime
;
885 sb
->events_hi
= (mddev
->events
>>32);
886 sb
->events_lo
= (u32
)mddev
->events
;
888 if (mddev
->reshape_position
== MaxSector
)
889 sb
->minor_version
= 90;
891 sb
->minor_version
= 91;
892 sb
->reshape_position
= mddev
->reshape_position
;
893 sb
->new_level
= mddev
->new_level
;
894 sb
->delta_disks
= mddev
->delta_disks
;
895 sb
->new_layout
= mddev
->new_layout
;
896 sb
->new_chunk
= mddev
->new_chunk
;
898 mddev
->minor_version
= sb
->minor_version
;
901 sb
->recovery_cp
= mddev
->recovery_cp
;
902 sb
->cp_events_hi
= (mddev
->events
>>32);
903 sb
->cp_events_lo
= (u32
)mddev
->events
;
904 if (mddev
->recovery_cp
== MaxSector
)
905 sb
->state
= (1<< MD_SB_CLEAN
);
909 sb
->layout
= mddev
->layout
;
910 sb
->chunk_size
= mddev
->chunk_size
;
912 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
913 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
915 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
916 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
919 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
920 && !test_bit(Faulty
, &rdev2
->flags
))
921 desc_nr
= rdev2
->raid_disk
;
923 desc_nr
= next_spare
++;
924 rdev2
->desc_nr
= desc_nr
;
925 d
= &sb
->disks
[rdev2
->desc_nr
];
927 d
->number
= rdev2
->desc_nr
;
928 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
929 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
930 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
931 && !test_bit(Faulty
, &rdev2
->flags
))
932 d
->raid_disk
= rdev2
->raid_disk
;
934 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
935 if (test_bit(Faulty
, &rdev2
->flags
))
936 d
->state
= (1<<MD_DISK_FAULTY
);
937 else if (test_bit(In_sync
, &rdev2
->flags
)) {
938 d
->state
= (1<<MD_DISK_ACTIVE
);
939 d
->state
|= (1<<MD_DISK_SYNC
);
947 if (test_bit(WriteMostly
, &rdev2
->flags
))
948 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
950 /* now set the "removed" and "faulty" bits on any missing devices */
951 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
952 mdp_disk_t
*d
= &sb
->disks
[i
];
953 if (d
->state
== 0 && d
->number
== 0) {
956 d
->state
= (1<<MD_DISK_REMOVED
);
957 d
->state
|= (1<<MD_DISK_FAULTY
);
961 sb
->nr_disks
= nr_disks
;
962 sb
->active_disks
= active
;
963 sb
->working_disks
= working
;
964 sb
->failed_disks
= failed
;
965 sb
->spare_disks
= spare
;
967 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
968 sb
->sb_csum
= calc_sb_csum(sb
);
972 * version 1 superblock
975 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
977 unsigned int disk_csum
, csum
;
978 unsigned long long newcsum
;
979 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
980 unsigned int *isuper
= (unsigned int*)sb
;
983 disk_csum
= sb
->sb_csum
;
986 for (i
=0; size
>=4; size
-= 4 )
987 newcsum
+= le32_to_cpu(*isuper
++);
990 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
992 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
993 sb
->sb_csum
= disk_csum
;
994 return cpu_to_le32(csum
);
997 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
999 struct mdp_superblock_1
*sb
;
1002 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1006 * Calculate the position of the superblock.
1007 * It is always aligned to a 4K boundary and
1008 * depeding on minor_version, it can be:
1009 * 0: At least 8K, but less than 12K, from end of device
1010 * 1: At start of device
1011 * 2: 4K from start of device.
1013 switch(minor_version
) {
1015 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1017 sb_offset
&= ~(sector_t
)(4*2-1);
1018 /* convert from sectors to K */
1030 rdev
->sb_offset
= sb_offset
;
1032 /* superblock is rarely larger than 1K, but it can be larger,
1033 * and it is safe to read 4k, so we do that
1035 ret
= read_disk_sb(rdev
, 4096);
1036 if (ret
) return ret
;
1039 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1041 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1042 sb
->major_version
!= cpu_to_le32(1) ||
1043 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1044 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1045 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1048 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1049 printk("md: invalid superblock checksum on %s\n",
1050 bdevname(rdev
->bdev
,b
));
1053 if (le64_to_cpu(sb
->data_size
) < 10) {
1054 printk("md: data_size too small on %s\n",
1055 bdevname(rdev
->bdev
,b
));
1058 rdev
->preferred_minor
= 0xffff;
1059 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1060 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1062 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1063 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1064 if (rdev
->sb_size
& bmask
)
1065 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1071 struct mdp_superblock_1
*refsb
=
1072 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1074 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1075 sb
->level
!= refsb
->level
||
1076 sb
->layout
!= refsb
->layout
||
1077 sb
->chunksize
!= refsb
->chunksize
) {
1078 printk(KERN_WARNING
"md: %s has strangely different"
1079 " superblock to %s\n",
1080 bdevname(rdev
->bdev
,b
),
1081 bdevname(refdev
->bdev
,b2
));
1084 ev1
= le64_to_cpu(sb
->events
);
1085 ev2
= le64_to_cpu(refsb
->events
);
1093 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1095 rdev
->size
= rdev
->sb_offset
;
1096 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1098 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1099 if (le32_to_cpu(sb
->chunksize
))
1100 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1102 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1107 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1109 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1110 __u64 ev1
= le64_to_cpu(sb
->events
);
1112 rdev
->raid_disk
= -1;
1114 if (mddev
->raid_disks
== 0) {
1115 mddev
->major_version
= 1;
1116 mddev
->patch_version
= 0;
1117 mddev
->persistent
= 1;
1118 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1119 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1120 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1121 mddev
->level
= le32_to_cpu(sb
->level
);
1122 mddev
->clevel
[0] = 0;
1123 mddev
->layout
= le32_to_cpu(sb
->layout
);
1124 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1125 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1126 mddev
->events
= ev1
;
1127 mddev
->bitmap_offset
= 0;
1128 mddev
->default_bitmap_offset
= 1024 >> 9;
1130 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1131 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1133 mddev
->max_disks
= (4096-256)/2;
1135 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1136 mddev
->bitmap_file
== NULL
) {
1137 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1138 && mddev
->level
!= 10) {
1139 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1142 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1144 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1145 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1146 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1147 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1148 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1149 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1151 mddev
->reshape_position
= MaxSector
;
1152 mddev
->delta_disks
= 0;
1153 mddev
->new_level
= mddev
->level
;
1154 mddev
->new_layout
= mddev
->layout
;
1155 mddev
->new_chunk
= mddev
->chunk_size
;
1158 } else if (mddev
->pers
== NULL
) {
1159 /* Insist of good event counter while assembling */
1161 if (ev1
< mddev
->events
)
1163 } else if (mddev
->bitmap
) {
1164 /* If adding to array with a bitmap, then we can accept an
1165 * older device, but not too old.
1167 if (ev1
< mddev
->bitmap
->events_cleared
)
1170 if (ev1
< mddev
->events
)
1171 /* just a hot-add of a new device, leave raid_disk at -1 */
1174 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1176 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1177 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1179 case 0xffff: /* spare */
1181 case 0xfffe: /* faulty */
1182 set_bit(Faulty
, &rdev
->flags
);
1185 if ((le32_to_cpu(sb
->feature_map
) &
1186 MD_FEATURE_RECOVERY_OFFSET
))
1187 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1189 set_bit(In_sync
, &rdev
->flags
);
1190 rdev
->raid_disk
= role
;
1193 if (sb
->devflags
& WriteMostly1
)
1194 set_bit(WriteMostly
, &rdev
->flags
);
1195 } else /* MULTIPATH are always insync */
1196 set_bit(In_sync
, &rdev
->flags
);
1201 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1203 struct mdp_superblock_1
*sb
;
1204 struct list_head
*tmp
;
1207 /* make rdev->sb match mddev and rdev data. */
1209 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1211 sb
->feature_map
= 0;
1213 sb
->recovery_offset
= cpu_to_le64(0);
1214 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1215 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1216 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1218 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1219 sb
->events
= cpu_to_le64(mddev
->events
);
1221 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1223 sb
->resync_offset
= cpu_to_le64(0);
1225 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1227 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1228 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1230 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1231 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1232 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1235 if (rdev
->raid_disk
>= 0 &&
1236 !test_bit(In_sync
, &rdev
->flags
) &&
1237 rdev
->recovery_offset
> 0) {
1238 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1239 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1242 if (mddev
->reshape_position
!= MaxSector
) {
1243 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1244 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1245 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1246 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1247 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1248 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1252 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1253 if (rdev2
->desc_nr
+1 > max_dev
)
1254 max_dev
= rdev2
->desc_nr
+1;
1256 sb
->max_dev
= cpu_to_le32(max_dev
);
1257 for (i
=0; i
<max_dev
;i
++)
1258 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1260 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1262 if (test_bit(Faulty
, &rdev2
->flags
))
1263 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1264 else if (test_bit(In_sync
, &rdev2
->flags
))
1265 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1266 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1267 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1269 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1272 sb
->sb_csum
= calc_sb_1_csum(sb
);
1276 static struct super_type super_types
[] = {
1279 .owner
= THIS_MODULE
,
1280 .load_super
= super_90_load
,
1281 .validate_super
= super_90_validate
,
1282 .sync_super
= super_90_sync
,
1286 .owner
= THIS_MODULE
,
1287 .load_super
= super_1_load
,
1288 .validate_super
= super_1_validate
,
1289 .sync_super
= super_1_sync
,
1293 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1295 struct list_head
*tmp
;
1298 ITERATE_RDEV(mddev
,rdev
,tmp
)
1299 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1305 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1307 struct list_head
*tmp
;
1310 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1311 if (match_dev_unit(mddev2
, rdev
))
1317 static LIST_HEAD(pending_raid_disks
);
1319 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1321 mdk_rdev_t
*same_pdev
;
1322 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1330 /* make sure rdev->size exceeds mddev->size */
1331 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1333 /* Cannot change size, so fail */
1336 mddev
->size
= rdev
->size
;
1338 same_pdev
= match_dev_unit(mddev
, rdev
);
1341 "%s: WARNING: %s appears to be on the same physical"
1342 " disk as %s. True\n protection against single-disk"
1343 " failure might be compromised.\n",
1344 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1345 bdevname(same_pdev
->bdev
,b2
));
1347 /* Verify rdev->desc_nr is unique.
1348 * If it is -1, assign a free number, else
1349 * check number is not in use
1351 if (rdev
->desc_nr
< 0) {
1353 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1354 while (find_rdev_nr(mddev
, choice
))
1356 rdev
->desc_nr
= choice
;
1358 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1361 bdevname(rdev
->bdev
,b
);
1362 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1364 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1367 list_add(&rdev
->same_set
, &mddev
->disks
);
1368 rdev
->mddev
= mddev
;
1369 printk(KERN_INFO
"md: bind<%s>\n", b
);
1371 rdev
->kobj
.parent
= &mddev
->kobj
;
1372 kobject_add(&rdev
->kobj
);
1374 if (rdev
->bdev
->bd_part
)
1375 ko
= &rdev
->bdev
->bd_part
->kobj
;
1377 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1378 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1379 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1383 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1385 char b
[BDEVNAME_SIZE
];
1390 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1391 list_del_init(&rdev
->same_set
);
1392 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1394 sysfs_remove_link(&rdev
->kobj
, "block");
1395 kobject_del(&rdev
->kobj
);
1399 * prevent the device from being mounted, repartitioned or
1400 * otherwise reused by a RAID array (or any other kernel
1401 * subsystem), by bd_claiming the device.
1403 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1406 struct block_device
*bdev
;
1407 char b
[BDEVNAME_SIZE
];
1409 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1411 printk(KERN_ERR
"md: could not open %s.\n",
1412 __bdevname(dev
, b
));
1413 return PTR_ERR(bdev
);
1415 err
= bd_claim(bdev
, rdev
);
1417 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1426 static void unlock_rdev(mdk_rdev_t
*rdev
)
1428 struct block_device
*bdev
= rdev
->bdev
;
1436 void md_autodetect_dev(dev_t dev
);
1438 static void export_rdev(mdk_rdev_t
* rdev
)
1440 char b
[BDEVNAME_SIZE
];
1441 printk(KERN_INFO
"md: export_rdev(%s)\n",
1442 bdevname(rdev
->bdev
,b
));
1446 list_del_init(&rdev
->same_set
);
1448 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1451 kobject_put(&rdev
->kobj
);
1454 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1456 unbind_rdev_from_array(rdev
);
1460 static void export_array(mddev_t
*mddev
)
1462 struct list_head
*tmp
;
1465 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1470 kick_rdev_from_array(rdev
);
1472 if (!list_empty(&mddev
->disks
))
1474 mddev
->raid_disks
= 0;
1475 mddev
->major_version
= 0;
1478 static void print_desc(mdp_disk_t
*desc
)
1480 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1481 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1484 static void print_sb(mdp_super_t
*sb
)
1489 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1490 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1491 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1493 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1494 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1495 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1496 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1497 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1498 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1499 sb
->failed_disks
, sb
->spare_disks
,
1500 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1503 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1506 desc
= sb
->disks
+ i
;
1507 if (desc
->number
|| desc
->major
|| desc
->minor
||
1508 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1509 printk(" D %2d: ", i
);
1513 printk(KERN_INFO
"md: THIS: ");
1514 print_desc(&sb
->this_disk
);
1518 static void print_rdev(mdk_rdev_t
*rdev
)
1520 char b
[BDEVNAME_SIZE
];
1521 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1522 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1523 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1525 if (rdev
->sb_loaded
) {
1526 printk(KERN_INFO
"md: rdev superblock:\n");
1527 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1529 printk(KERN_INFO
"md: no rdev superblock!\n");
1532 static void md_print_devices(void)
1534 struct list_head
*tmp
, *tmp2
;
1537 char b
[BDEVNAME_SIZE
];
1540 printk("md: **********************************\n");
1541 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1542 printk("md: **********************************\n");
1543 ITERATE_MDDEV(mddev
,tmp
) {
1546 bitmap_print_sb(mddev
->bitmap
);
1548 printk("%s: ", mdname(mddev
));
1549 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1550 printk("<%s>", bdevname(rdev
->bdev
,b
));
1553 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1556 printk("md: **********************************\n");
1561 static void sync_sbs(mddev_t
* mddev
)
1564 struct list_head
*tmp
;
1566 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1567 super_types
[mddev
->major_version
].
1568 sync_super(mddev
, rdev
);
1569 rdev
->sb_loaded
= 1;
1573 void md_update_sb(mddev_t
* mddev
)
1576 struct list_head
*tmp
;
1581 spin_lock_irq(&mddev
->write_lock
);
1582 sync_req
= mddev
->in_sync
;
1583 mddev
->utime
= get_seconds();
1586 if (!mddev
->events
) {
1588 * oops, this 64-bit counter should never wrap.
1589 * Either we are in around ~1 trillion A.C., assuming
1590 * 1 reboot per second, or we have a bug:
1595 mddev
->sb_dirty
= 2;
1599 * do not write anything to disk if using
1600 * nonpersistent superblocks
1602 if (!mddev
->persistent
) {
1603 mddev
->sb_dirty
= 0;
1604 spin_unlock_irq(&mddev
->write_lock
);
1605 wake_up(&mddev
->sb_wait
);
1608 spin_unlock_irq(&mddev
->write_lock
);
1611 "md: updating %s RAID superblock on device (in sync %d)\n",
1612 mdname(mddev
),mddev
->in_sync
);
1614 err
= bitmap_update_sb(mddev
->bitmap
);
1615 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1616 char b
[BDEVNAME_SIZE
];
1617 dprintk(KERN_INFO
"md: ");
1618 if (test_bit(Faulty
, &rdev
->flags
))
1619 dprintk("(skipping faulty ");
1621 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1622 if (!test_bit(Faulty
, &rdev
->flags
)) {
1623 md_super_write(mddev
,rdev
,
1624 rdev
->sb_offset
<<1, rdev
->sb_size
,
1626 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1627 bdevname(rdev
->bdev
,b
),
1628 (unsigned long long)rdev
->sb_offset
);
1632 if (mddev
->level
== LEVEL_MULTIPATH
)
1633 /* only need to write one superblock... */
1636 md_super_wait(mddev
);
1637 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1639 spin_lock_irq(&mddev
->write_lock
);
1640 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1641 /* have to write it out again */
1642 spin_unlock_irq(&mddev
->write_lock
);
1645 mddev
->sb_dirty
= 0;
1646 spin_unlock_irq(&mddev
->write_lock
);
1647 wake_up(&mddev
->sb_wait
);
1650 EXPORT_SYMBOL_GPL(md_update_sb
);
1652 /* words written to sysfs files may, or my not, be \n terminated.
1653 * We want to accept with case. For this we use cmd_match.
1655 static int cmd_match(const char *cmd
, const char *str
)
1657 /* See if cmd, written into a sysfs file, matches
1658 * str. They must either be the same, or cmd can
1659 * have a trailing newline
1661 while (*cmd
&& *str
&& *cmd
== *str
) {
1672 struct rdev_sysfs_entry
{
1673 struct attribute attr
;
1674 ssize_t (*show
)(mdk_rdev_t
*, char *);
1675 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1679 state_show(mdk_rdev_t
*rdev
, char *page
)
1684 if (test_bit(Faulty
, &rdev
->flags
)) {
1685 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1688 if (test_bit(In_sync
, &rdev
->flags
)) {
1689 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1692 if (!test_bit(Faulty
, &rdev
->flags
) &&
1693 !test_bit(In_sync
, &rdev
->flags
)) {
1694 len
+= sprintf(page
+len
, "%sspare", sep
);
1697 return len
+sprintf(page
+len
, "\n");
1700 static struct rdev_sysfs_entry
1701 rdev_state
= __ATTR_RO(state
);
1704 super_show(mdk_rdev_t
*rdev
, char *page
)
1706 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1707 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1708 return rdev
->sb_size
;
1712 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1715 errors_show(mdk_rdev_t
*rdev
, char *page
)
1717 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1721 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1724 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1725 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1726 atomic_set(&rdev
->corrected_errors
, n
);
1731 static struct rdev_sysfs_entry rdev_errors
=
1732 __ATTR(errors
, 0644, errors_show
, errors_store
);
1735 slot_show(mdk_rdev_t
*rdev
, char *page
)
1737 if (rdev
->raid_disk
< 0)
1738 return sprintf(page
, "none\n");
1740 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1744 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1747 int slot
= simple_strtoul(buf
, &e
, 10);
1748 if (strncmp(buf
, "none", 4)==0)
1750 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1752 if (rdev
->mddev
->pers
)
1753 /* Cannot set slot in active array (yet) */
1755 if (slot
>= rdev
->mddev
->raid_disks
)
1757 rdev
->raid_disk
= slot
;
1758 /* assume it is working */
1760 set_bit(In_sync
, &rdev
->flags
);
1765 static struct rdev_sysfs_entry rdev_slot
=
1766 __ATTR(slot
, 0644, slot_show
, slot_store
);
1769 offset_show(mdk_rdev_t
*rdev
, char *page
)
1771 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1775 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1778 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1779 if (e
==buf
|| (*e
&& *e
!= '\n'))
1781 if (rdev
->mddev
->pers
)
1783 rdev
->data_offset
= offset
;
1787 static struct rdev_sysfs_entry rdev_offset
=
1788 __ATTR(offset
, 0644, offset_show
, offset_store
);
1791 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1793 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1797 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1800 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1801 if (e
==buf
|| (*e
&& *e
!= '\n'))
1803 if (rdev
->mddev
->pers
)
1806 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1807 rdev
->mddev
->size
= size
;
1811 static struct rdev_sysfs_entry rdev_size
=
1812 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1814 static struct attribute
*rdev_default_attrs
[] = {
1824 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1826 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1827 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1831 return entry
->show(rdev
, page
);
1835 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1836 const char *page
, size_t length
)
1838 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1839 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1843 return entry
->store(rdev
, page
, length
);
1846 static void rdev_free(struct kobject
*ko
)
1848 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1851 static struct sysfs_ops rdev_sysfs_ops
= {
1852 .show
= rdev_attr_show
,
1853 .store
= rdev_attr_store
,
1855 static struct kobj_type rdev_ktype
= {
1856 .release
= rdev_free
,
1857 .sysfs_ops
= &rdev_sysfs_ops
,
1858 .default_attrs
= rdev_default_attrs
,
1862 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1864 * mark the device faulty if:
1866 * - the device is nonexistent (zero size)
1867 * - the device has no valid superblock
1869 * a faulty rdev _never_ has rdev->sb set.
1871 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1873 char b
[BDEVNAME_SIZE
];
1878 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1880 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1881 return ERR_PTR(-ENOMEM
);
1884 if ((err
= alloc_disk_sb(rdev
)))
1887 err
= lock_rdev(rdev
, newdev
);
1891 rdev
->kobj
.parent
= NULL
;
1892 rdev
->kobj
.ktype
= &rdev_ktype
;
1893 kobject_init(&rdev
->kobj
);
1897 rdev
->data_offset
= 0;
1898 atomic_set(&rdev
->nr_pending
, 0);
1899 atomic_set(&rdev
->read_errors
, 0);
1900 atomic_set(&rdev
->corrected_errors
, 0);
1902 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1905 "md: %s has zero or unknown size, marking faulty!\n",
1906 bdevname(rdev
->bdev
,b
));
1911 if (super_format
>= 0) {
1912 err
= super_types
[super_format
].
1913 load_super(rdev
, NULL
, super_minor
);
1914 if (err
== -EINVAL
) {
1916 "md: %s has invalid sb, not importing!\n",
1917 bdevname(rdev
->bdev
,b
));
1922 "md: could not read %s's sb, not importing!\n",
1923 bdevname(rdev
->bdev
,b
));
1927 INIT_LIST_HEAD(&rdev
->same_set
);
1932 if (rdev
->sb_page
) {
1938 return ERR_PTR(err
);
1942 * Check a full RAID array for plausibility
1946 static void analyze_sbs(mddev_t
* mddev
)
1949 struct list_head
*tmp
;
1950 mdk_rdev_t
*rdev
, *freshest
;
1951 char b
[BDEVNAME_SIZE
];
1954 ITERATE_RDEV(mddev
,rdev
,tmp
)
1955 switch (super_types
[mddev
->major_version
].
1956 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1964 "md: fatal superblock inconsistency in %s"
1965 " -- removing from array\n",
1966 bdevname(rdev
->bdev
,b
));
1967 kick_rdev_from_array(rdev
);
1971 super_types
[mddev
->major_version
].
1972 validate_super(mddev
, freshest
);
1975 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1976 if (rdev
!= freshest
)
1977 if (super_types
[mddev
->major_version
].
1978 validate_super(mddev
, rdev
)) {
1979 printk(KERN_WARNING
"md: kicking non-fresh %s"
1981 bdevname(rdev
->bdev
,b
));
1982 kick_rdev_from_array(rdev
);
1985 if (mddev
->level
== LEVEL_MULTIPATH
) {
1986 rdev
->desc_nr
= i
++;
1987 rdev
->raid_disk
= rdev
->desc_nr
;
1988 set_bit(In_sync
, &rdev
->flags
);
1994 if (mddev
->recovery_cp
!= MaxSector
&&
1996 printk(KERN_ERR
"md: %s: raid array is not clean"
1997 " -- starting background reconstruction\n",
2003 safe_delay_show(mddev_t
*mddev
, char *page
)
2005 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2006 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2009 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2017 /* remove a period, and count digits after it */
2018 if (len
>= sizeof(buf
))
2020 strlcpy(buf
, cbuf
, len
);
2022 for (i
=0; i
<len
; i
++) {
2024 if (isdigit(buf
[i
])) {
2029 } else if (buf
[i
] == '.') {
2034 msec
= simple_strtoul(buf
, &e
, 10);
2035 if (e
== buf
|| (*e
&& *e
!= '\n'))
2037 msec
= (msec
* 1000) / scale
;
2039 mddev
->safemode_delay
= 0;
2041 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2042 if (mddev
->safemode_delay
== 0)
2043 mddev
->safemode_delay
= 1;
2047 static struct md_sysfs_entry md_safe_delay
=
2048 __ATTR(safe_mode_delay
, 0644,safe_delay_show
, safe_delay_store
);
2051 level_show(mddev_t
*mddev
, char *page
)
2053 struct mdk_personality
*p
= mddev
->pers
;
2055 return sprintf(page
, "%s\n", p
->name
);
2056 else if (mddev
->clevel
[0])
2057 return sprintf(page
, "%s\n", mddev
->clevel
);
2058 else if (mddev
->level
!= LEVEL_NONE
)
2059 return sprintf(page
, "%d\n", mddev
->level
);
2065 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2072 if (len
>= sizeof(mddev
->clevel
))
2074 strncpy(mddev
->clevel
, buf
, len
);
2075 if (mddev
->clevel
[len
-1] == '\n')
2077 mddev
->clevel
[len
] = 0;
2078 mddev
->level
= LEVEL_NONE
;
2082 static struct md_sysfs_entry md_level
=
2083 __ATTR(level
, 0644, level_show
, level_store
);
2086 raid_disks_show(mddev_t
*mddev
, char *page
)
2088 if (mddev
->raid_disks
== 0)
2090 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2093 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2096 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2098 /* can only set raid_disks if array is not yet active */
2101 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2103 if (!*buf
|| (*e
&& *e
!= '\n'))
2107 rv
= update_raid_disks(mddev
, n
);
2109 mddev
->raid_disks
= n
;
2110 return rv
? rv
: len
;
2112 static struct md_sysfs_entry md_raid_disks
=
2113 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
2116 chunk_size_show(mddev_t
*mddev
, char *page
)
2118 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2122 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2124 /* can only set chunk_size if array is not yet active */
2126 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2130 if (!*buf
|| (*e
&& *e
!= '\n'))
2133 mddev
->chunk_size
= n
;
2136 static struct md_sysfs_entry md_chunk_size
=
2137 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2140 null_show(mddev_t
*mddev
, char *page
)
2146 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2148 /* buf must be %d:%d\n? giving major and minor numbers */
2149 /* The new device is added to the array.
2150 * If the array has a persistent superblock, we read the
2151 * superblock to initialise info and check validity.
2152 * Otherwise, only checking done is that in bind_rdev_to_array,
2153 * which mainly checks size.
2156 int major
= simple_strtoul(buf
, &e
, 10);
2162 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2164 minor
= simple_strtoul(e
+1, &e
, 10);
2165 if (*e
&& *e
!= '\n')
2167 dev
= MKDEV(major
, minor
);
2168 if (major
!= MAJOR(dev
) ||
2169 minor
!= MINOR(dev
))
2173 if (mddev
->persistent
) {
2174 rdev
= md_import_device(dev
, mddev
->major_version
,
2175 mddev
->minor_version
);
2176 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2177 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2178 mdk_rdev_t
, same_set
);
2179 err
= super_types
[mddev
->major_version
]
2180 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2185 rdev
= md_import_device(dev
, -1, -1);
2188 return PTR_ERR(rdev
);
2189 err
= bind_rdev_to_array(rdev
, mddev
);
2193 return err
? err
: len
;
2196 static struct md_sysfs_entry md_new_device
=
2197 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2200 size_show(mddev_t
*mddev
, char *page
)
2202 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2205 static int update_size(mddev_t
*mddev
, unsigned long size
);
2208 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2210 /* If array is inactive, we can reduce the component size, but
2211 * not increase it (except from 0).
2212 * If array is active, we can try an on-line resize
2216 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2217 if (!*buf
|| *buf
== '\n' ||
2222 err
= update_size(mddev
, size
);
2223 md_update_sb(mddev
);
2225 if (mddev
->size
== 0 ||
2231 return err
? err
: len
;
2234 static struct md_sysfs_entry md_size
=
2235 __ATTR(component_size
, 0644, size_show
, size_store
);
2239 * This is either 'none' for arrays with externally managed metadata,
2240 * or N.M for internally known formats
2243 metadata_show(mddev_t
*mddev
, char *page
)
2245 if (mddev
->persistent
)
2246 return sprintf(page
, "%d.%d\n",
2247 mddev
->major_version
, mddev
->minor_version
);
2249 return sprintf(page
, "none\n");
2253 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2257 if (!list_empty(&mddev
->disks
))
2260 if (cmd_match(buf
, "none")) {
2261 mddev
->persistent
= 0;
2262 mddev
->major_version
= 0;
2263 mddev
->minor_version
= 90;
2266 major
= simple_strtoul(buf
, &e
, 10);
2267 if (e
==buf
|| *e
!= '.')
2270 minor
= simple_strtoul(buf
, &e
, 10);
2271 if (e
==buf
|| *e
!= '\n')
2273 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2274 super_types
[major
].name
== NULL
)
2276 mddev
->major_version
= major
;
2277 mddev
->minor_version
= minor
;
2278 mddev
->persistent
= 1;
2282 static struct md_sysfs_entry md_metadata
=
2283 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2286 action_show(mddev_t
*mddev
, char *page
)
2288 char *type
= "idle";
2289 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2290 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2291 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2293 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2294 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2296 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2303 return sprintf(page
, "%s\n", type
);
2307 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2309 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2312 if (cmd_match(page
, "idle")) {
2313 if (mddev
->sync_thread
) {
2314 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2315 md_unregister_thread(mddev
->sync_thread
);
2316 mddev
->sync_thread
= NULL
;
2317 mddev
->recovery
= 0;
2319 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2320 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2322 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2323 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2324 else if (cmd_match(page
, "reshape")) {
2326 if (mddev
->pers
->start_reshape
== NULL
)
2328 err
= mddev
->pers
->start_reshape(mddev
);
2332 if (cmd_match(page
, "check"))
2333 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2334 else if (!cmd_match(page
, "repair"))
2336 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2337 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2339 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2340 md_wakeup_thread(mddev
->thread
);
2345 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2347 return sprintf(page
, "%llu\n",
2348 (unsigned long long) mddev
->resync_mismatches
);
2351 static struct md_sysfs_entry
2352 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2355 static struct md_sysfs_entry
2356 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2359 sync_min_show(mddev_t
*mddev
, char *page
)
2361 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2362 mddev
->sync_speed_min
? "local": "system");
2366 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2370 if (strncmp(buf
, "system", 6)==0) {
2371 mddev
->sync_speed_min
= 0;
2374 min
= simple_strtoul(buf
, &e
, 10);
2375 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2377 mddev
->sync_speed_min
= min
;
2381 static struct md_sysfs_entry md_sync_min
=
2382 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2385 sync_max_show(mddev_t
*mddev
, char *page
)
2387 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2388 mddev
->sync_speed_max
? "local": "system");
2392 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2396 if (strncmp(buf
, "system", 6)==0) {
2397 mddev
->sync_speed_max
= 0;
2400 max
= simple_strtoul(buf
, &e
, 10);
2401 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2403 mddev
->sync_speed_max
= max
;
2407 static struct md_sysfs_entry md_sync_max
=
2408 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2412 sync_speed_show(mddev_t
*mddev
, char *page
)
2414 unsigned long resync
, dt
, db
;
2415 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2416 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2418 db
= resync
- (mddev
->resync_mark_cnt
);
2419 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2422 static struct md_sysfs_entry
2423 md_sync_speed
= __ATTR_RO(sync_speed
);
2426 sync_completed_show(mddev_t
*mddev
, char *page
)
2428 unsigned long max_blocks
, resync
;
2430 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2431 max_blocks
= mddev
->resync_max_sectors
;
2433 max_blocks
= mddev
->size
<< 1;
2435 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2436 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2439 static struct md_sysfs_entry
2440 md_sync_completed
= __ATTR_RO(sync_completed
);
2443 suspend_lo_show(mddev_t
*mddev
, char *page
)
2445 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2449 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2452 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2454 if (mddev
->pers
->quiesce
== NULL
)
2456 if (buf
== e
|| (*e
&& *e
!= '\n'))
2458 if (new >= mddev
->suspend_hi
||
2459 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2460 mddev
->suspend_lo
= new;
2461 mddev
->pers
->quiesce(mddev
, 2);
2466 static struct md_sysfs_entry md_suspend_lo
=
2467 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2471 suspend_hi_show(mddev_t
*mddev
, char *page
)
2473 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2477 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2480 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2482 if (mddev
->pers
->quiesce
== NULL
)
2484 if (buf
== e
|| (*e
&& *e
!= '\n'))
2486 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2487 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2488 mddev
->suspend_hi
= new;
2489 mddev
->pers
->quiesce(mddev
, 1);
2490 mddev
->pers
->quiesce(mddev
, 0);
2495 static struct md_sysfs_entry md_suspend_hi
=
2496 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2499 static struct attribute
*md_default_attrs
[] = {
2501 &md_raid_disks
.attr
,
2502 &md_chunk_size
.attr
,
2505 &md_new_device
.attr
,
2506 &md_safe_delay
.attr
,
2510 static struct attribute
*md_redundancy_attrs
[] = {
2512 &md_mismatches
.attr
,
2515 &md_sync_speed
.attr
,
2516 &md_sync_completed
.attr
,
2517 &md_suspend_lo
.attr
,
2518 &md_suspend_hi
.attr
,
2521 static struct attribute_group md_redundancy_group
= {
2523 .attrs
= md_redundancy_attrs
,
2528 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2530 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2531 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2536 rv
= mddev_lock(mddev
);
2538 rv
= entry
->show(mddev
, page
);
2539 mddev_unlock(mddev
);
2545 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2546 const char *page
, size_t length
)
2548 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2549 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2554 rv
= mddev_lock(mddev
);
2556 rv
= entry
->store(mddev
, page
, length
);
2557 mddev_unlock(mddev
);
2562 static void md_free(struct kobject
*ko
)
2564 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2568 static struct sysfs_ops md_sysfs_ops
= {
2569 .show
= md_attr_show
,
2570 .store
= md_attr_store
,
2572 static struct kobj_type md_ktype
= {
2574 .sysfs_ops
= &md_sysfs_ops
,
2575 .default_attrs
= md_default_attrs
,
2580 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2582 static DEFINE_MUTEX(disks_mutex
);
2583 mddev_t
*mddev
= mddev_find(dev
);
2584 struct gendisk
*disk
;
2585 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2586 int shift
= partitioned
? MdpMinorShift
: 0;
2587 int unit
= MINOR(dev
) >> shift
;
2592 mutex_lock(&disks_mutex
);
2593 if (mddev
->gendisk
) {
2594 mutex_unlock(&disks_mutex
);
2598 disk
= alloc_disk(1 << shift
);
2600 mutex_unlock(&disks_mutex
);
2604 disk
->major
= MAJOR(dev
);
2605 disk
->first_minor
= unit
<< shift
;
2607 sprintf(disk
->disk_name
, "md_d%d", unit
);
2608 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2610 sprintf(disk
->disk_name
, "md%d", unit
);
2611 sprintf(disk
->devfs_name
, "md/%d", unit
);
2613 disk
->fops
= &md_fops
;
2614 disk
->private_data
= mddev
;
2615 disk
->queue
= mddev
->queue
;
2617 mddev
->gendisk
= disk
;
2618 mutex_unlock(&disks_mutex
);
2619 mddev
->kobj
.parent
= &disk
->kobj
;
2620 mddev
->kobj
.k_name
= NULL
;
2621 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2622 mddev
->kobj
.ktype
= &md_ktype
;
2623 kobject_register(&mddev
->kobj
);
2627 static void md_safemode_timeout(unsigned long data
)
2629 mddev_t
*mddev
= (mddev_t
*) data
;
2631 mddev
->safemode
= 1;
2632 md_wakeup_thread(mddev
->thread
);
2635 static int start_dirty_degraded
;
2637 static int do_md_run(mddev_t
* mddev
)
2641 struct list_head
*tmp
;
2643 struct gendisk
*disk
;
2644 struct mdk_personality
*pers
;
2645 char b
[BDEVNAME_SIZE
];
2647 if (list_empty(&mddev
->disks
))
2648 /* cannot run an array with no devices.. */
2655 * Analyze all RAID superblock(s)
2657 if (!mddev
->raid_disks
)
2660 chunk_size
= mddev
->chunk_size
;
2663 if (chunk_size
> MAX_CHUNK_SIZE
) {
2664 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2665 chunk_size
, MAX_CHUNK_SIZE
);
2669 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2671 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2672 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2675 if (chunk_size
< PAGE_SIZE
) {
2676 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2677 chunk_size
, PAGE_SIZE
);
2681 /* devices must have minimum size of one chunk */
2682 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2683 if (test_bit(Faulty
, &rdev
->flags
))
2685 if (rdev
->size
< chunk_size
/ 1024) {
2687 "md: Dev %s smaller than chunk_size:"
2689 bdevname(rdev
->bdev
,b
),
2690 (unsigned long long)rdev
->size
,
2698 if (mddev
->level
!= LEVEL_NONE
)
2699 request_module("md-level-%d", mddev
->level
);
2700 else if (mddev
->clevel
[0])
2701 request_module("md-%s", mddev
->clevel
);
2705 * Drop all container device buffers, from now on
2706 * the only valid external interface is through the md
2708 * Also find largest hardsector size
2710 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2711 if (test_bit(Faulty
, &rdev
->flags
))
2713 sync_blockdev(rdev
->bdev
);
2714 invalidate_bdev(rdev
->bdev
, 0);
2717 md_probe(mddev
->unit
, NULL
, NULL
);
2718 disk
= mddev
->gendisk
;
2722 spin_lock(&pers_lock
);
2723 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2724 if (!pers
|| !try_module_get(pers
->owner
)) {
2725 spin_unlock(&pers_lock
);
2726 if (mddev
->level
!= LEVEL_NONE
)
2727 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2730 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2735 spin_unlock(&pers_lock
);
2736 mddev
->level
= pers
->level
;
2737 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2739 if (mddev
->reshape_position
!= MaxSector
&&
2740 pers
->start_reshape
== NULL
) {
2741 /* This personality cannot handle reshaping... */
2743 module_put(pers
->owner
);
2747 mddev
->recovery
= 0;
2748 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2749 mddev
->barriers_work
= 1;
2750 mddev
->ok_start_degraded
= start_dirty_degraded
;
2753 mddev
->ro
= 2; /* read-only, but switch on first write */
2755 err
= mddev
->pers
->run(mddev
);
2756 if (!err
&& mddev
->pers
->sync_request
) {
2757 err
= bitmap_create(mddev
);
2759 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2760 mdname(mddev
), err
);
2761 mddev
->pers
->stop(mddev
);
2765 printk(KERN_ERR
"md: pers->run() failed ...\n");
2766 module_put(mddev
->pers
->owner
);
2768 bitmap_destroy(mddev
);
2771 if (mddev
->pers
->sync_request
)
2772 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2773 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2776 atomic_set(&mddev
->writes_pending
,0);
2777 mddev
->safemode
= 0;
2778 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2779 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2780 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
2783 ITERATE_RDEV(mddev
,rdev
,tmp
)
2784 if (rdev
->raid_disk
>= 0) {
2786 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2787 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2790 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2791 md_wakeup_thread(mddev
->thread
);
2793 if (mddev
->sb_dirty
)
2794 md_update_sb(mddev
);
2796 set_capacity(disk
, mddev
->array_size
<<1);
2798 /* If we call blk_queue_make_request here, it will
2799 * re-initialise max_sectors etc which may have been
2800 * refined inside -> run. So just set the bits we need to set.
2801 * Most initialisation happended when we called
2802 * blk_queue_make_request(..., md_fail_request)
2805 mddev
->queue
->queuedata
= mddev
;
2806 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2808 /* If there is a partially-recovered drive we need to
2809 * start recovery here. If we leave it to md_check_recovery,
2810 * it will remove the drives and not do the right thing
2812 if (mddev
->degraded
) {
2813 struct list_head
*rtmp
;
2815 ITERATE_RDEV(mddev
,rdev
,rtmp
)
2816 if (rdev
->raid_disk
>= 0 &&
2817 !test_bit(In_sync
, &rdev
->flags
) &&
2818 !test_bit(Faulty
, &rdev
->flags
))
2819 /* complete an interrupted recovery */
2821 if (spares
&& mddev
->pers
->sync_request
) {
2822 mddev
->recovery
= 0;
2823 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
2824 mddev
->sync_thread
= md_register_thread(md_do_sync
,
2827 if (!mddev
->sync_thread
) {
2828 printk(KERN_ERR
"%s: could not start resync"
2831 /* leave the spares where they are, it shouldn't hurt */
2832 mddev
->recovery
= 0;
2834 md_wakeup_thread(mddev
->sync_thread
);
2839 md_new_event(mddev
);
2843 static int restart_array(mddev_t
*mddev
)
2845 struct gendisk
*disk
= mddev
->gendisk
;
2849 * Complain if it has no devices
2852 if (list_empty(&mddev
->disks
))
2860 mddev
->safemode
= 0;
2862 set_disk_ro(disk
, 0);
2864 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2867 * Kick recovery or resync if necessary
2869 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2870 md_wakeup_thread(mddev
->thread
);
2871 md_wakeup_thread(mddev
->sync_thread
);
2874 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2883 /* similar to deny_write_access, but accounts for our holding a reference
2884 * to the file ourselves */
2885 static int deny_bitmap_write_access(struct file
* file
)
2887 struct inode
*inode
= file
->f_mapping
->host
;
2889 spin_lock(&inode
->i_lock
);
2890 if (atomic_read(&inode
->i_writecount
) > 1) {
2891 spin_unlock(&inode
->i_lock
);
2894 atomic_set(&inode
->i_writecount
, -1);
2895 spin_unlock(&inode
->i_lock
);
2900 static void restore_bitmap_write_access(struct file
*file
)
2902 struct inode
*inode
= file
->f_mapping
->host
;
2904 spin_lock(&inode
->i_lock
);
2905 atomic_set(&inode
->i_writecount
, 1);
2906 spin_unlock(&inode
->i_lock
);
2909 static int do_md_stop(mddev_t
* mddev
, int ro
)
2912 struct gendisk
*disk
= mddev
->gendisk
;
2915 if (atomic_read(&mddev
->active
)>2) {
2916 printk("md: %s still in use.\n",mdname(mddev
));
2920 if (mddev
->sync_thread
) {
2921 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
2922 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2923 md_unregister_thread(mddev
->sync_thread
);
2924 mddev
->sync_thread
= NULL
;
2927 del_timer_sync(&mddev
->safemode_timer
);
2929 invalidate_partition(disk
, 0);
2937 bitmap_flush(mddev
);
2938 md_super_wait(mddev
);
2940 set_disk_ro(disk
, 0);
2941 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2942 mddev
->pers
->stop(mddev
);
2943 if (mddev
->pers
->sync_request
)
2944 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2946 module_put(mddev
->pers
->owner
);
2951 if (!mddev
->in_sync
|| mddev
->sb_dirty
) {
2952 /* mark array as shutdown cleanly */
2954 md_update_sb(mddev
);
2957 set_disk_ro(disk
, 1);
2958 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
2962 * Free resources if final stop
2966 struct list_head
*tmp
;
2967 struct gendisk
*disk
;
2968 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
2970 bitmap_destroy(mddev
);
2971 if (mddev
->bitmap_file
) {
2972 restore_bitmap_write_access(mddev
->bitmap_file
);
2973 fput(mddev
->bitmap_file
);
2974 mddev
->bitmap_file
= NULL
;
2976 mddev
->bitmap_offset
= 0;
2978 ITERATE_RDEV(mddev
,rdev
,tmp
)
2979 if (rdev
->raid_disk
>= 0) {
2981 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2982 sysfs_remove_link(&mddev
->kobj
, nm
);
2985 export_array(mddev
);
2987 mddev
->array_size
= 0;
2988 disk
= mddev
->gendisk
;
2990 set_capacity(disk
, 0);
2992 } else if (mddev
->pers
)
2993 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
2996 md_new_event(mddev
);
3001 static void autorun_array(mddev_t
*mddev
)
3004 struct list_head
*tmp
;
3007 if (list_empty(&mddev
->disks
))
3010 printk(KERN_INFO
"md: running: ");
3012 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3013 char b
[BDEVNAME_SIZE
];
3014 printk("<%s>", bdevname(rdev
->bdev
,b
));
3018 err
= do_md_run (mddev
);
3020 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3021 do_md_stop (mddev
, 0);
3026 * lets try to run arrays based on all disks that have arrived
3027 * until now. (those are in pending_raid_disks)
3029 * the method: pick the first pending disk, collect all disks with
3030 * the same UUID, remove all from the pending list and put them into
3031 * the 'same_array' list. Then order this list based on superblock
3032 * update time (freshest comes first), kick out 'old' disks and
3033 * compare superblocks. If everything's fine then run it.
3035 * If "unit" is allocated, then bump its reference count
3037 static void autorun_devices(int part
)
3039 struct list_head
*tmp
;
3040 mdk_rdev_t
*rdev0
, *rdev
;
3042 char b
[BDEVNAME_SIZE
];
3044 printk(KERN_INFO
"md: autorun ...\n");
3045 while (!list_empty(&pending_raid_disks
)) {
3047 LIST_HEAD(candidates
);
3048 rdev0
= list_entry(pending_raid_disks
.next
,
3049 mdk_rdev_t
, same_set
);
3051 printk(KERN_INFO
"md: considering %s ...\n",
3052 bdevname(rdev0
->bdev
,b
));
3053 INIT_LIST_HEAD(&candidates
);
3054 ITERATE_RDEV_PENDING(rdev
,tmp
)
3055 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3056 printk(KERN_INFO
"md: adding %s ...\n",
3057 bdevname(rdev
->bdev
,b
));
3058 list_move(&rdev
->same_set
, &candidates
);
3061 * now we have a set of devices, with all of them having
3062 * mostly sane superblocks. It's time to allocate the
3065 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
3066 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3067 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3071 dev
= MKDEV(mdp_major
,
3072 rdev0
->preferred_minor
<< MdpMinorShift
);
3074 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3076 md_probe(dev
, NULL
, NULL
);
3077 mddev
= mddev_find(dev
);
3080 "md: cannot allocate memory for md drive.\n");
3083 if (mddev_lock(mddev
))
3084 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3086 else if (mddev
->raid_disks
|| mddev
->major_version
3087 || !list_empty(&mddev
->disks
)) {
3089 "md: %s already running, cannot run %s\n",
3090 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3091 mddev_unlock(mddev
);
3093 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3094 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3095 list_del_init(&rdev
->same_set
);
3096 if (bind_rdev_to_array(rdev
, mddev
))
3099 autorun_array(mddev
);
3100 mddev_unlock(mddev
);
3102 /* on success, candidates will be empty, on error
3105 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3109 printk(KERN_INFO
"md: ... autorun DONE.\n");
3113 * import RAID devices based on one partition
3114 * if possible, the array gets run as well.
3117 static int autostart_array(dev_t startdev
)
3119 char b
[BDEVNAME_SIZE
];
3120 int err
= -EINVAL
, i
;
3121 mdp_super_t
*sb
= NULL
;
3122 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
3124 start_rdev
= md_import_device(startdev
, 0, 0);
3125 if (IS_ERR(start_rdev
))
3129 /* NOTE: this can only work for 0.90.0 superblocks */
3130 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
3131 if (sb
->major_version
!= 0 ||
3132 sb
->minor_version
!= 90 ) {
3133 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
3134 export_rdev(start_rdev
);
3138 if (test_bit(Faulty
, &start_rdev
->flags
)) {
3140 "md: can not autostart based on faulty %s!\n",
3141 bdevname(start_rdev
->bdev
,b
));
3142 export_rdev(start_rdev
);
3145 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
3147 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
3148 mdp_disk_t
*desc
= sb
->disks
+ i
;
3149 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
3153 if (dev
== startdev
)
3155 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
3157 rdev
= md_import_device(dev
, 0, 0);
3161 list_add(&rdev
->same_set
, &pending_raid_disks
);
3165 * possibly return codes
3173 static int get_version(void __user
* arg
)
3177 ver
.major
= MD_MAJOR_VERSION
;
3178 ver
.minor
= MD_MINOR_VERSION
;
3179 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3181 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3187 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3189 mdu_array_info_t info
;
3190 int nr
,working
,active
,failed
,spare
;
3192 struct list_head
*tmp
;
3194 nr
=working
=active
=failed
=spare
=0;
3195 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3197 if (test_bit(Faulty
, &rdev
->flags
))
3201 if (test_bit(In_sync
, &rdev
->flags
))
3208 info
.major_version
= mddev
->major_version
;
3209 info
.minor_version
= mddev
->minor_version
;
3210 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3211 info
.ctime
= mddev
->ctime
;
3212 info
.level
= mddev
->level
;
3213 info
.size
= mddev
->size
;
3214 if (info
.size
!= mddev
->size
) /* overflow */
3217 info
.raid_disks
= mddev
->raid_disks
;
3218 info
.md_minor
= mddev
->md_minor
;
3219 info
.not_persistent
= !mddev
->persistent
;
3221 info
.utime
= mddev
->utime
;
3224 info
.state
= (1<<MD_SB_CLEAN
);
3225 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3226 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3227 info
.active_disks
= active
;
3228 info
.working_disks
= working
;
3229 info
.failed_disks
= failed
;
3230 info
.spare_disks
= spare
;
3232 info
.layout
= mddev
->layout
;
3233 info
.chunk_size
= mddev
->chunk_size
;
3235 if (copy_to_user(arg
, &info
, sizeof(info
)))
3241 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3243 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3244 char *ptr
, *buf
= NULL
;
3247 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3251 /* bitmap disabled, zero the first byte and copy out */
3252 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3253 file
->pathname
[0] = '\0';
3257 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3261 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3265 strcpy(file
->pathname
, ptr
);
3269 if (copy_to_user(arg
, file
, sizeof(*file
)))
3277 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3279 mdu_disk_info_t info
;
3283 if (copy_from_user(&info
, arg
, sizeof(info
)))
3288 rdev
= find_rdev_nr(mddev
, nr
);
3290 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3291 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3292 info
.raid_disk
= rdev
->raid_disk
;
3294 if (test_bit(Faulty
, &rdev
->flags
))
3295 info
.state
|= (1<<MD_DISK_FAULTY
);
3296 else if (test_bit(In_sync
, &rdev
->flags
)) {
3297 info
.state
|= (1<<MD_DISK_ACTIVE
);
3298 info
.state
|= (1<<MD_DISK_SYNC
);
3300 if (test_bit(WriteMostly
, &rdev
->flags
))
3301 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3303 info
.major
= info
.minor
= 0;
3304 info
.raid_disk
= -1;
3305 info
.state
= (1<<MD_DISK_REMOVED
);
3308 if (copy_to_user(arg
, &info
, sizeof(info
)))
3314 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3316 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3318 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3320 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3323 if (!mddev
->raid_disks
) {
3325 /* expecting a device which has a superblock */
3326 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3329 "md: md_import_device returned %ld\n",
3331 return PTR_ERR(rdev
);
3333 if (!list_empty(&mddev
->disks
)) {
3334 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3335 mdk_rdev_t
, same_set
);
3336 int err
= super_types
[mddev
->major_version
]
3337 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3340 "md: %s has different UUID to %s\n",
3341 bdevname(rdev
->bdev
,b
),
3342 bdevname(rdev0
->bdev
,b2
));
3347 err
= bind_rdev_to_array(rdev
, mddev
);
3354 * add_new_disk can be used once the array is assembled
3355 * to add "hot spares". They must already have a superblock
3360 if (!mddev
->pers
->hot_add_disk
) {
3362 "%s: personality does not support diskops!\n",
3366 if (mddev
->persistent
)
3367 rdev
= md_import_device(dev
, mddev
->major_version
,
3368 mddev
->minor_version
);
3370 rdev
= md_import_device(dev
, -1, -1);
3373 "md: md_import_device returned %ld\n",
3375 return PTR_ERR(rdev
);
3377 /* set save_raid_disk if appropriate */
3378 if (!mddev
->persistent
) {
3379 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3380 info
->raid_disk
< mddev
->raid_disks
)
3381 rdev
->raid_disk
= info
->raid_disk
;
3383 rdev
->raid_disk
= -1;
3385 super_types
[mddev
->major_version
].
3386 validate_super(mddev
, rdev
);
3387 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3389 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3390 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3391 set_bit(WriteMostly
, &rdev
->flags
);
3393 rdev
->raid_disk
= -1;
3394 err
= bind_rdev_to_array(rdev
, mddev
);
3395 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3396 /* If there is hot_add_disk but no hot_remove_disk
3397 * then added disks for geometry changes,
3398 * and should be added immediately.
3400 super_types
[mddev
->major_version
].
3401 validate_super(mddev
, rdev
);
3402 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3404 unbind_rdev_from_array(rdev
);
3409 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3410 md_wakeup_thread(mddev
->thread
);
3414 /* otherwise, add_new_disk is only allowed
3415 * for major_version==0 superblocks
3417 if (mddev
->major_version
!= 0) {
3418 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3423 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3425 rdev
= md_import_device (dev
, -1, 0);
3428 "md: error, md_import_device() returned %ld\n",
3430 return PTR_ERR(rdev
);
3432 rdev
->desc_nr
= info
->number
;
3433 if (info
->raid_disk
< mddev
->raid_disks
)
3434 rdev
->raid_disk
= info
->raid_disk
;
3436 rdev
->raid_disk
= -1;
3440 if (rdev
->raid_disk
< mddev
->raid_disks
)
3441 if (info
->state
& (1<<MD_DISK_SYNC
))
3442 set_bit(In_sync
, &rdev
->flags
);
3444 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3445 set_bit(WriteMostly
, &rdev
->flags
);
3447 if (!mddev
->persistent
) {
3448 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3449 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3451 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3452 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3454 err
= bind_rdev_to_array(rdev
, mddev
);
3464 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3466 char b
[BDEVNAME_SIZE
];
3472 rdev
= find_rdev(mddev
, dev
);
3476 if (rdev
->raid_disk
>= 0)
3479 kick_rdev_from_array(rdev
);
3480 md_update_sb(mddev
);
3481 md_new_event(mddev
);
3485 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3486 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3490 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3492 char b
[BDEVNAME_SIZE
];
3500 if (mddev
->major_version
!= 0) {
3501 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3502 " version-0 superblocks.\n",
3506 if (!mddev
->pers
->hot_add_disk
) {
3508 "%s: personality does not support diskops!\n",
3513 rdev
= md_import_device (dev
, -1, 0);
3516 "md: error, md_import_device() returned %ld\n",
3521 if (mddev
->persistent
)
3522 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3525 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3527 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3530 if (test_bit(Faulty
, &rdev
->flags
)) {
3532 "md: can not hot-add faulty %s disk to %s!\n",
3533 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3537 clear_bit(In_sync
, &rdev
->flags
);
3539 err
= bind_rdev_to_array(rdev
, mddev
);
3544 * The rest should better be atomic, we can have disk failures
3545 * noticed in interrupt contexts ...
3548 if (rdev
->desc_nr
== mddev
->max_disks
) {
3549 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3552 goto abort_unbind_export
;
3555 rdev
->raid_disk
= -1;
3557 md_update_sb(mddev
);
3560 * Kick recovery, maybe this spare has to be added to the
3561 * array immediately.
3563 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3564 md_wakeup_thread(mddev
->thread
);
3565 md_new_event(mddev
);
3568 abort_unbind_export
:
3569 unbind_rdev_from_array(rdev
);
3576 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3581 if (!mddev
->pers
->quiesce
)
3583 if (mddev
->recovery
|| mddev
->sync_thread
)
3585 /* we should be able to change the bitmap.. */
3591 return -EEXIST
; /* cannot add when bitmap is present */
3592 mddev
->bitmap_file
= fget(fd
);
3594 if (mddev
->bitmap_file
== NULL
) {
3595 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3600 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3602 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3604 fput(mddev
->bitmap_file
);
3605 mddev
->bitmap_file
= NULL
;
3608 mddev
->bitmap_offset
= 0; /* file overrides offset */
3609 } else if (mddev
->bitmap
== NULL
)
3610 return -ENOENT
; /* cannot remove what isn't there */
3613 mddev
->pers
->quiesce(mddev
, 1);
3615 err
= bitmap_create(mddev
);
3616 if (fd
< 0 || err
) {
3617 bitmap_destroy(mddev
);
3618 fd
= -1; /* make sure to put the file */
3620 mddev
->pers
->quiesce(mddev
, 0);
3623 if (mddev
->bitmap_file
) {
3624 restore_bitmap_write_access(mddev
->bitmap_file
);
3625 fput(mddev
->bitmap_file
);
3627 mddev
->bitmap_file
= NULL
;
3634 * set_array_info is used two different ways
3635 * The original usage is when creating a new array.
3636 * In this usage, raid_disks is > 0 and it together with
3637 * level, size, not_persistent,layout,chunksize determine the
3638 * shape of the array.
3639 * This will always create an array with a type-0.90.0 superblock.
3640 * The newer usage is when assembling an array.
3641 * In this case raid_disks will be 0, and the major_version field is
3642 * use to determine which style super-blocks are to be found on the devices.
3643 * The minor and patch _version numbers are also kept incase the
3644 * super_block handler wishes to interpret them.
3646 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3649 if (info
->raid_disks
== 0) {
3650 /* just setting version number for superblock loading */
3651 if (info
->major_version
< 0 ||
3652 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3653 super_types
[info
->major_version
].name
== NULL
) {
3654 /* maybe try to auto-load a module? */
3656 "md: superblock version %d not known\n",
3657 info
->major_version
);
3660 mddev
->major_version
= info
->major_version
;
3661 mddev
->minor_version
= info
->minor_version
;
3662 mddev
->patch_version
= info
->patch_version
;
3665 mddev
->major_version
= MD_MAJOR_VERSION
;
3666 mddev
->minor_version
= MD_MINOR_VERSION
;
3667 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3668 mddev
->ctime
= get_seconds();
3670 mddev
->level
= info
->level
;
3671 mddev
->clevel
[0] = 0;
3672 mddev
->size
= info
->size
;
3673 mddev
->raid_disks
= info
->raid_disks
;
3674 /* don't set md_minor, it is determined by which /dev/md* was
3677 if (info
->state
& (1<<MD_SB_CLEAN
))
3678 mddev
->recovery_cp
= MaxSector
;
3680 mddev
->recovery_cp
= 0;
3681 mddev
->persistent
= ! info
->not_persistent
;
3683 mddev
->layout
= info
->layout
;
3684 mddev
->chunk_size
= info
->chunk_size
;
3686 mddev
->max_disks
= MD_SB_DISKS
;
3688 mddev
->sb_dirty
= 1;
3690 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3691 mddev
->bitmap_offset
= 0;
3693 mddev
->reshape_position
= MaxSector
;
3696 * Generate a 128 bit UUID
3698 get_random_bytes(mddev
->uuid
, 16);
3700 mddev
->new_level
= mddev
->level
;
3701 mddev
->new_chunk
= mddev
->chunk_size
;
3702 mddev
->new_layout
= mddev
->layout
;
3703 mddev
->delta_disks
= 0;
3708 static int update_size(mddev_t
*mddev
, unsigned long size
)
3712 struct list_head
*tmp
;
3713 int fit
= (size
== 0);
3715 if (mddev
->pers
->resize
== NULL
)
3717 /* The "size" is the amount of each device that is used.
3718 * This can only make sense for arrays with redundancy.
3719 * linear and raid0 always use whatever space is available
3720 * We can only consider changing the size if no resync
3721 * or reconstruction is happening, and if the new size
3722 * is acceptable. It must fit before the sb_offset or,
3723 * if that is <data_offset, it must fit before the
3724 * size of each device.
3725 * If size is zero, we find the largest size that fits.
3727 if (mddev
->sync_thread
)
3729 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3731 if (rdev
->sb_offset
> rdev
->data_offset
)
3732 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3734 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3735 - rdev
->data_offset
;
3736 if (fit
&& (size
== 0 || size
> avail
/2))
3738 if (avail
< ((sector_t
)size
<< 1))
3741 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3743 struct block_device
*bdev
;
3745 bdev
= bdget_disk(mddev
->gendisk
, 0);
3747 mutex_lock(&bdev
->bd_inode
->i_mutex
);
3748 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
3749 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
3756 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3759 /* change the number of raid disks */
3760 if (mddev
->pers
->check_reshape
== NULL
)
3762 if (raid_disks
<= 0 ||
3763 raid_disks
>= mddev
->max_disks
)
3765 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
3767 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
3769 rv
= mddev
->pers
->check_reshape(mddev
);
3775 * update_array_info is used to change the configuration of an
3777 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3778 * fields in the info are checked against the array.
3779 * Any differences that cannot be handled will cause an error.
3780 * Normally, only one change can be managed at a time.
3782 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3788 /* calculate expected state,ignoring low bits */
3789 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3790 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3792 if (mddev
->major_version
!= info
->major_version
||
3793 mddev
->minor_version
!= info
->minor_version
||
3794 /* mddev->patch_version != info->patch_version || */
3795 mddev
->ctime
!= info
->ctime
||
3796 mddev
->level
!= info
->level
||
3797 /* mddev->layout != info->layout || */
3798 !mddev
->persistent
!= info
->not_persistent
||
3799 mddev
->chunk_size
!= info
->chunk_size
||
3800 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3801 ((state
^info
->state
) & 0xfffffe00)
3804 /* Check there is only one change */
3805 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
3806 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3807 if (mddev
->layout
!= info
->layout
) cnt
++;
3808 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3809 if (cnt
== 0) return 0;
3810 if (cnt
> 1) return -EINVAL
;
3812 if (mddev
->layout
!= info
->layout
) {
3814 * we don't need to do anything at the md level, the
3815 * personality will take care of it all.
3817 if (mddev
->pers
->reconfig
== NULL
)
3820 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3822 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
3823 rv
= update_size(mddev
, info
->size
);
3825 if (mddev
->raid_disks
!= info
->raid_disks
)
3826 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3828 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3829 if (mddev
->pers
->quiesce
== NULL
)
3831 if (mddev
->recovery
|| mddev
->sync_thread
)
3833 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3834 /* add the bitmap */
3837 if (mddev
->default_bitmap_offset
== 0)
3839 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3840 mddev
->pers
->quiesce(mddev
, 1);
3841 rv
= bitmap_create(mddev
);
3843 bitmap_destroy(mddev
);
3844 mddev
->pers
->quiesce(mddev
, 0);
3846 /* remove the bitmap */
3849 if (mddev
->bitmap
->file
)
3851 mddev
->pers
->quiesce(mddev
, 1);
3852 bitmap_destroy(mddev
);
3853 mddev
->pers
->quiesce(mddev
, 0);
3854 mddev
->bitmap_offset
= 0;
3857 md_update_sb(mddev
);
3861 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3865 if (mddev
->pers
== NULL
)
3868 rdev
= find_rdev(mddev
, dev
);
3872 md_error(mddev
, rdev
);
3876 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
3878 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
3882 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
3886 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3887 unsigned int cmd
, unsigned long arg
)
3890 void __user
*argp
= (void __user
*)arg
;
3891 mddev_t
*mddev
= NULL
;
3893 if (!capable(CAP_SYS_ADMIN
))
3897 * Commands dealing with the RAID driver but not any
3903 err
= get_version(argp
);
3906 case PRINT_RAID_DEBUG
:
3914 autostart_arrays(arg
);
3921 * Commands creating/starting a new array:
3924 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3932 if (cmd
== START_ARRAY
) {
3933 /* START_ARRAY doesn't need to lock the array as autostart_array
3934 * does the locking, and it could even be a different array
3939 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3940 "This will not be supported beyond July 2006\n",
3941 current
->comm
, current
->pid
);
3944 err
= autostart_array(new_decode_dev(arg
));
3946 printk(KERN_WARNING
"md: autostart failed!\n");
3952 err
= mddev_lock(mddev
);
3955 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3962 case SET_ARRAY_INFO
:
3964 mdu_array_info_t info
;
3966 memset(&info
, 0, sizeof(info
));
3967 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
3972 err
= update_array_info(mddev
, &info
);
3974 printk(KERN_WARNING
"md: couldn't update"
3975 " array info. %d\n", err
);
3980 if (!list_empty(&mddev
->disks
)) {
3982 "md: array %s already has disks!\n",
3987 if (mddev
->raid_disks
) {
3989 "md: array %s already initialised!\n",
3994 err
= set_array_info(mddev
, &info
);
3996 printk(KERN_WARNING
"md: couldn't set"
3997 " array info. %d\n", err
);
4007 * Commands querying/configuring an existing array:
4009 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4010 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4011 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4012 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
4018 * Commands even a read-only array can execute:
4022 case GET_ARRAY_INFO
:
4023 err
= get_array_info(mddev
, argp
);
4026 case GET_BITMAP_FILE
:
4027 err
= get_bitmap_file(mddev
, argp
);
4031 err
= get_disk_info(mddev
, argp
);
4034 case RESTART_ARRAY_RW
:
4035 err
= restart_array(mddev
);
4039 err
= do_md_stop (mddev
, 0);
4043 err
= do_md_stop (mddev
, 1);
4047 * We have a problem here : there is no easy way to give a CHS
4048 * virtual geometry. We currently pretend that we have a 2 heads
4049 * 4 sectors (with a BIG number of cylinders...). This drives
4050 * dosfs just mad... ;-)
4055 * The remaining ioctls are changing the state of the
4056 * superblock, so we do not allow them on read-only arrays.
4057 * However non-MD ioctls (e.g. get-size) will still come through
4058 * here and hit the 'default' below, so only disallow
4059 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4061 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4062 mddev
->ro
&& mddev
->pers
) {
4063 if (mddev
->ro
== 2) {
4065 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4066 md_wakeup_thread(mddev
->thread
);
4078 mdu_disk_info_t info
;
4079 if (copy_from_user(&info
, argp
, sizeof(info
)))
4082 err
= add_new_disk(mddev
, &info
);
4086 case HOT_REMOVE_DISK
:
4087 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4091 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4094 case SET_DISK_FAULTY
:
4095 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4099 err
= do_md_run (mddev
);
4102 case SET_BITMAP_FILE
:
4103 err
= set_bitmap_file(mddev
, (int)arg
);
4113 mddev_unlock(mddev
);
4123 static int md_open(struct inode
*inode
, struct file
*file
)
4126 * Succeed if we can lock the mddev, which confirms that
4127 * it isn't being stopped right now.
4129 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4132 if ((err
= mddev_lock(mddev
)))
4137 mddev_unlock(mddev
);
4139 check_disk_change(inode
->i_bdev
);
4144 static int md_release(struct inode
*inode
, struct file
* file
)
4146 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4155 static int md_media_changed(struct gendisk
*disk
)
4157 mddev_t
*mddev
= disk
->private_data
;
4159 return mddev
->changed
;
4162 static int md_revalidate(struct gendisk
*disk
)
4164 mddev_t
*mddev
= disk
->private_data
;
4169 static struct block_device_operations md_fops
=
4171 .owner
= THIS_MODULE
,
4173 .release
= md_release
,
4175 .getgeo
= md_getgeo
,
4176 .media_changed
= md_media_changed
,
4177 .revalidate_disk
= md_revalidate
,
4180 static int md_thread(void * arg
)
4182 mdk_thread_t
*thread
= arg
;
4185 * md_thread is a 'system-thread', it's priority should be very
4186 * high. We avoid resource deadlocks individually in each
4187 * raid personality. (RAID5 does preallocation) We also use RR and
4188 * the very same RT priority as kswapd, thus we will never get
4189 * into a priority inversion deadlock.
4191 * we definitely have to have equal or higher priority than
4192 * bdflush, otherwise bdflush will deadlock if there are too
4193 * many dirty RAID5 blocks.
4196 allow_signal(SIGKILL
);
4197 while (!kthread_should_stop()) {
4199 /* We need to wait INTERRUPTIBLE so that
4200 * we don't add to the load-average.
4201 * That means we need to be sure no signals are
4204 if (signal_pending(current
))
4205 flush_signals(current
);
4207 wait_event_interruptible_timeout
4209 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4210 || kthread_should_stop(),
4214 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4216 thread
->run(thread
->mddev
);
4222 void md_wakeup_thread(mdk_thread_t
*thread
)
4225 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4226 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4227 wake_up(&thread
->wqueue
);
4231 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4234 mdk_thread_t
*thread
;
4236 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4240 init_waitqueue_head(&thread
->wqueue
);
4243 thread
->mddev
= mddev
;
4244 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4245 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4246 if (IS_ERR(thread
->tsk
)) {
4253 void md_unregister_thread(mdk_thread_t
*thread
)
4255 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4257 kthread_stop(thread
->tsk
);
4261 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4268 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4271 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4273 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4274 __builtin_return_address(0),__builtin_return_address(1),
4275 __builtin_return_address(2),__builtin_return_address(3));
4277 if (!mddev
->pers
->error_handler
)
4279 mddev
->pers
->error_handler(mddev
,rdev
);
4280 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4281 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4282 md_wakeup_thread(mddev
->thread
);
4283 md_new_event_inintr(mddev
);
4286 /* seq_file implementation /proc/mdstat */
4288 static void status_unused(struct seq_file
*seq
)
4292 struct list_head
*tmp
;
4294 seq_printf(seq
, "unused devices: ");
4296 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4297 char b
[BDEVNAME_SIZE
];
4299 seq_printf(seq
, "%s ",
4300 bdevname(rdev
->bdev
,b
));
4303 seq_printf(seq
, "<none>");
4305 seq_printf(seq
, "\n");
4309 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4311 sector_t max_blocks
, resync
, res
;
4312 unsigned long dt
, db
, rt
;
4314 unsigned int per_milli
;
4316 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4318 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4319 max_blocks
= mddev
->resync_max_sectors
>> 1;
4321 max_blocks
= mddev
->size
;
4324 * Should not happen.
4330 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4331 * in a sector_t, and (max_blocks>>scale) will fit in a
4332 * u32, as those are the requirements for sector_div.
4333 * Thus 'scale' must be at least 10
4336 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4337 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4340 res
= (resync
>>scale
)*1000;
4341 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4345 int i
, x
= per_milli
/50, y
= 20-x
;
4346 seq_printf(seq
, "[");
4347 for (i
= 0; i
< x
; i
++)
4348 seq_printf(seq
, "=");
4349 seq_printf(seq
, ">");
4350 for (i
= 0; i
< y
; i
++)
4351 seq_printf(seq
, ".");
4352 seq_printf(seq
, "] ");
4354 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4355 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4357 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4358 "resync" : "recovery")),
4359 per_milli
/10, per_milli
% 10,
4360 (unsigned long long) resync
,
4361 (unsigned long long) max_blocks
);
4364 * We do not want to overflow, so the order of operands and
4365 * the * 100 / 100 trick are important. We do a +1 to be
4366 * safe against division by zero. We only estimate anyway.
4368 * dt: time from mark until now
4369 * db: blocks written from mark until now
4370 * rt: remaining time
4372 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4374 db
= resync
- (mddev
->resync_mark_cnt
/2);
4375 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/100+1)))/100;
4377 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4379 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4382 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4384 struct list_head
*tmp
;
4394 spin_lock(&all_mddevs_lock
);
4395 list_for_each(tmp
,&all_mddevs
)
4397 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4399 spin_unlock(&all_mddevs_lock
);
4402 spin_unlock(&all_mddevs_lock
);
4404 return (void*)2;/* tail */
4408 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4410 struct list_head
*tmp
;
4411 mddev_t
*next_mddev
, *mddev
= v
;
4417 spin_lock(&all_mddevs_lock
);
4419 tmp
= all_mddevs
.next
;
4421 tmp
= mddev
->all_mddevs
.next
;
4422 if (tmp
!= &all_mddevs
)
4423 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4425 next_mddev
= (void*)2;
4428 spin_unlock(&all_mddevs_lock
);
4436 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4440 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4444 struct mdstat_info
{
4448 static int md_seq_show(struct seq_file
*seq
, void *v
)
4452 struct list_head
*tmp2
;
4454 struct mdstat_info
*mi
= seq
->private;
4455 struct bitmap
*bitmap
;
4457 if (v
== (void*)1) {
4458 struct mdk_personality
*pers
;
4459 seq_printf(seq
, "Personalities : ");
4460 spin_lock(&pers_lock
);
4461 list_for_each_entry(pers
, &pers_list
, list
)
4462 seq_printf(seq
, "[%s] ", pers
->name
);
4464 spin_unlock(&pers_lock
);
4465 seq_printf(seq
, "\n");
4466 mi
->event
= atomic_read(&md_event_count
);
4469 if (v
== (void*)2) {
4474 if (mddev_lock(mddev
) < 0)
4477 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4478 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4479 mddev
->pers
? "" : "in");
4482 seq_printf(seq
, " (read-only)");
4484 seq_printf(seq
, "(auto-read-only)");
4485 seq_printf(seq
, " %s", mddev
->pers
->name
);
4489 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4490 char b
[BDEVNAME_SIZE
];
4491 seq_printf(seq
, " %s[%d]",
4492 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4493 if (test_bit(WriteMostly
, &rdev
->flags
))
4494 seq_printf(seq
, "(W)");
4495 if (test_bit(Faulty
, &rdev
->flags
)) {
4496 seq_printf(seq
, "(F)");
4498 } else if (rdev
->raid_disk
< 0)
4499 seq_printf(seq
, "(S)"); /* spare */
4503 if (!list_empty(&mddev
->disks
)) {
4505 seq_printf(seq
, "\n %llu blocks",
4506 (unsigned long long)mddev
->array_size
);
4508 seq_printf(seq
, "\n %llu blocks",
4509 (unsigned long long)size
);
4511 if (mddev
->persistent
) {
4512 if (mddev
->major_version
!= 0 ||
4513 mddev
->minor_version
!= 90) {
4514 seq_printf(seq
," super %d.%d",
4515 mddev
->major_version
,
4516 mddev
->minor_version
);
4519 seq_printf(seq
, " super non-persistent");
4522 mddev
->pers
->status (seq
, mddev
);
4523 seq_printf(seq
, "\n ");
4524 if (mddev
->pers
->sync_request
) {
4525 if (mddev
->curr_resync
> 2) {
4526 status_resync (seq
, mddev
);
4527 seq_printf(seq
, "\n ");
4528 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4529 seq_printf(seq
, "\tresync=DELAYED\n ");
4530 else if (mddev
->recovery_cp
< MaxSector
)
4531 seq_printf(seq
, "\tresync=PENDING\n ");
4534 seq_printf(seq
, "\n ");
4536 if ((bitmap
= mddev
->bitmap
)) {
4537 unsigned long chunk_kb
;
4538 unsigned long flags
;
4539 spin_lock_irqsave(&bitmap
->lock
, flags
);
4540 chunk_kb
= bitmap
->chunksize
>> 10;
4541 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4543 bitmap
->pages
- bitmap
->missing_pages
,
4545 (bitmap
->pages
- bitmap
->missing_pages
)
4546 << (PAGE_SHIFT
- 10),
4547 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4548 chunk_kb
? "KB" : "B");
4550 seq_printf(seq
, ", file: ");
4551 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4552 bitmap
->file
->f_dentry
," \t\n");
4555 seq_printf(seq
, "\n");
4556 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4559 seq_printf(seq
, "\n");
4561 mddev_unlock(mddev
);
4566 static struct seq_operations md_seq_ops
= {
4567 .start
= md_seq_start
,
4568 .next
= md_seq_next
,
4569 .stop
= md_seq_stop
,
4570 .show
= md_seq_show
,
4573 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4576 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4580 error
= seq_open(file
, &md_seq_ops
);
4584 struct seq_file
*p
= file
->private_data
;
4586 mi
->event
= atomic_read(&md_event_count
);
4591 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4593 struct seq_file
*m
= file
->private_data
;
4594 struct mdstat_info
*mi
= m
->private;
4597 return seq_release(inode
, file
);
4600 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4602 struct seq_file
*m
= filp
->private_data
;
4603 struct mdstat_info
*mi
= m
->private;
4606 poll_wait(filp
, &md_event_waiters
, wait
);
4608 /* always allow read */
4609 mask
= POLLIN
| POLLRDNORM
;
4611 if (mi
->event
!= atomic_read(&md_event_count
))
4612 mask
|= POLLERR
| POLLPRI
;
4616 static struct file_operations md_seq_fops
= {
4617 .open
= md_seq_open
,
4619 .llseek
= seq_lseek
,
4620 .release
= md_seq_release
,
4621 .poll
= mdstat_poll
,
4624 int register_md_personality(struct mdk_personality
*p
)
4626 spin_lock(&pers_lock
);
4627 list_add_tail(&p
->list
, &pers_list
);
4628 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4629 spin_unlock(&pers_lock
);
4633 int unregister_md_personality(struct mdk_personality
*p
)
4635 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4636 spin_lock(&pers_lock
);
4637 list_del_init(&p
->list
);
4638 spin_unlock(&pers_lock
);
4642 static int is_mddev_idle(mddev_t
*mddev
)
4645 struct list_head
*tmp
;
4647 unsigned long curr_events
;
4650 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4651 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4652 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4653 disk_stat_read(disk
, sectors
[1]) -
4654 atomic_read(&disk
->sync_io
);
4655 /* The difference between curr_events and last_events
4656 * will be affected by any new non-sync IO (making
4657 * curr_events bigger) and any difference in the amount of
4658 * in-flight syncio (making current_events bigger or smaller)
4659 * The amount in-flight is currently limited to
4660 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4661 * which is at most 4096 sectors.
4662 * These numbers are fairly fragile and should be made
4663 * more robust, probably by enforcing the
4664 * 'window size' that md_do_sync sort-of uses.
4666 * Note: the following is an unsigned comparison.
4668 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4669 rdev
->last_events
= curr_events
;
4676 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4678 /* another "blocks" (512byte) blocks have been synced */
4679 atomic_sub(blocks
, &mddev
->recovery_active
);
4680 wake_up(&mddev
->recovery_wait
);
4682 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4683 md_wakeup_thread(mddev
->thread
);
4684 // stop recovery, signal do_sync ....
4689 /* md_write_start(mddev, bi)
4690 * If we need to update some array metadata (e.g. 'active' flag
4691 * in superblock) before writing, schedule a superblock update
4692 * and wait for it to complete.
4694 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4696 if (bio_data_dir(bi
) != WRITE
)
4699 BUG_ON(mddev
->ro
== 1);
4700 if (mddev
->ro
== 2) {
4701 /* need to switch to read/write */
4703 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4704 md_wakeup_thread(mddev
->thread
);
4706 atomic_inc(&mddev
->writes_pending
);
4707 if (mddev
->in_sync
) {
4708 spin_lock_irq(&mddev
->write_lock
);
4709 if (mddev
->in_sync
) {
4711 mddev
->sb_dirty
= 1;
4712 md_wakeup_thread(mddev
->thread
);
4714 spin_unlock_irq(&mddev
->write_lock
);
4716 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4719 void md_write_end(mddev_t
*mddev
)
4721 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4722 if (mddev
->safemode
== 2)
4723 md_wakeup_thread(mddev
->thread
);
4724 else if (mddev
->safemode_delay
)
4725 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4729 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4731 #define SYNC_MARKS 10
4732 #define SYNC_MARK_STEP (3*HZ)
4733 void md_do_sync(mddev_t
*mddev
)
4736 unsigned int currspeed
= 0,
4738 sector_t max_sectors
,j
, io_sectors
;
4739 unsigned long mark
[SYNC_MARKS
];
4740 sector_t mark_cnt
[SYNC_MARKS
];
4742 struct list_head
*tmp
;
4743 sector_t last_check
;
4745 struct list_head
*rtmp
;
4748 /* just incase thread restarts... */
4749 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4751 if (mddev
->ro
) /* never try to sync a read-only array */
4754 /* we overload curr_resync somewhat here.
4755 * 0 == not engaged in resync at all
4756 * 2 == checking that there is no conflict with another sync
4757 * 1 == like 2, but have yielded to allow conflicting resync to
4759 * other == active in resync - this many blocks
4761 * Before starting a resync we must have set curr_resync to
4762 * 2, and then checked that every "conflicting" array has curr_resync
4763 * less than ours. When we find one that is the same or higher
4764 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4765 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4766 * This will mean we have to start checking from the beginning again.
4771 mddev
->curr_resync
= 2;
4774 if (kthread_should_stop()) {
4775 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4778 ITERATE_MDDEV(mddev2
,tmp
) {
4779 if (mddev2
== mddev
)
4781 if (mddev2
->curr_resync
&&
4782 match_mddev_units(mddev
,mddev2
)) {
4784 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4785 /* arbitrarily yield */
4786 mddev
->curr_resync
= 1;
4787 wake_up(&resync_wait
);
4789 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4790 /* no need to wait here, we can wait the next
4791 * time 'round when curr_resync == 2
4794 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4795 if (!kthread_should_stop() &&
4796 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4797 printk(KERN_INFO
"md: delaying resync of %s"
4798 " until %s has finished resync (they"
4799 " share one or more physical units)\n",
4800 mdname(mddev
), mdname(mddev2
));
4803 finish_wait(&resync_wait
, &wq
);
4806 finish_wait(&resync_wait
, &wq
);
4809 } while (mddev
->curr_resync
< 2);
4812 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4813 /* resync follows the size requested by the personality,
4814 * which defaults to physical size, but can be virtual size
4816 max_sectors
= mddev
->resync_max_sectors
;
4817 mddev
->resync_mismatches
= 0;
4818 /* we don't use the checkpoint if there's a bitmap */
4819 if (!mddev
->bitmap
&&
4820 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4821 j
= mddev
->recovery_cp
;
4822 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4823 max_sectors
= mddev
->size
<< 1;
4825 /* recovery follows the physical size of devices */
4826 max_sectors
= mddev
->size
<< 1;
4828 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4829 if (rdev
->raid_disk
>= 0 &&
4830 !test_bit(Faulty
, &rdev
->flags
) &&
4831 !test_bit(In_sync
, &rdev
->flags
) &&
4832 rdev
->recovery_offset
< j
)
4833 j
= rdev
->recovery_offset
;
4836 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4837 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4838 " %d KB/sec/disc.\n", speed_min(mddev
));
4839 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4840 "(but not more than %d KB/sec) for reconstruction.\n",
4843 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4846 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4848 mark_cnt
[m
] = io_sectors
;
4851 mddev
->resync_mark
= mark
[last_mark
];
4852 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4855 * Tune reconstruction:
4857 window
= 32*(PAGE_SIZE
/512);
4858 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4859 window
/2,(unsigned long long) max_sectors
/2);
4861 atomic_set(&mddev
->recovery_active
, 0);
4862 init_waitqueue_head(&mddev
->recovery_wait
);
4867 "md: resuming recovery of %s from checkpoint.\n",
4869 mddev
->curr_resync
= j
;
4872 while (j
< max_sectors
) {
4876 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4877 currspeed
< speed_min(mddev
));
4879 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4883 if (!skipped
) { /* actual IO requested */
4884 io_sectors
+= sectors
;
4885 atomic_add(sectors
, &mddev
->recovery_active
);
4889 if (j
>1) mddev
->curr_resync
= j
;
4890 if (last_check
== 0)
4891 /* this is the earliers that rebuilt will be
4892 * visible in /proc/mdstat
4894 md_new_event(mddev
);
4896 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4899 last_check
= io_sectors
;
4901 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4902 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4906 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4908 int next
= (last_mark
+1) % SYNC_MARKS
;
4910 mddev
->resync_mark
= mark
[next
];
4911 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4912 mark
[next
] = jiffies
;
4913 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4918 if (kthread_should_stop()) {
4920 * got a signal, exit.
4923 "md: md_do_sync() got signal ... exiting\n");
4924 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4929 * this loop exits only if either when we are slower than
4930 * the 'hard' speed limit, or the system was IO-idle for
4932 * the system might be non-idle CPU-wise, but we only care
4933 * about not overloading the IO subsystem. (things like an
4934 * e2fsck being done on the RAID array should execute fast)
4936 mddev
->queue
->unplug_fn(mddev
->queue
);
4939 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4940 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4942 if (currspeed
> speed_min(mddev
)) {
4943 if ((currspeed
> speed_max(mddev
)) ||
4944 !is_mddev_idle(mddev
)) {
4950 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
4952 * this also signals 'finished resyncing' to md_stop
4955 mddev
->queue
->unplug_fn(mddev
->queue
);
4957 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
4959 /* tell personality that we are finished */
4960 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
4962 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4963 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
4964 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
4965 mddev
->curr_resync
> 2) {
4966 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4967 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4968 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
4970 "md: checkpointing recovery of %s.\n",
4972 mddev
->recovery_cp
= mddev
->curr_resync
;
4975 mddev
->recovery_cp
= MaxSector
;
4977 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
4978 mddev
->curr_resync
= MaxSector
;
4979 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4980 if (rdev
->raid_disk
>= 0 &&
4981 !test_bit(Faulty
, &rdev
->flags
) &&
4982 !test_bit(In_sync
, &rdev
->flags
) &&
4983 rdev
->recovery_offset
< mddev
->curr_resync
)
4984 rdev
->recovery_offset
= mddev
->curr_resync
;
4985 mddev
->sb_dirty
= 1;
4990 mddev
->curr_resync
= 0;
4991 wake_up(&resync_wait
);
4992 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4993 md_wakeup_thread(mddev
->thread
);
4995 EXPORT_SYMBOL_GPL(md_do_sync
);
4999 * This routine is regularly called by all per-raid-array threads to
5000 * deal with generic issues like resync and super-block update.
5001 * Raid personalities that don't have a thread (linear/raid0) do not
5002 * need this as they never do any recovery or update the superblock.
5004 * It does not do any resync itself, but rather "forks" off other threads
5005 * to do that as needed.
5006 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5007 * "->recovery" and create a thread at ->sync_thread.
5008 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5009 * and wakeups up this thread which will reap the thread and finish up.
5010 * This thread also removes any faulty devices (with nr_pending == 0).
5012 * The overall approach is:
5013 * 1/ if the superblock needs updating, update it.
5014 * 2/ If a recovery thread is running, don't do anything else.
5015 * 3/ If recovery has finished, clean up, possibly marking spares active.
5016 * 4/ If there are any faulty devices, remove them.
5017 * 5/ If array is degraded, try to add spares devices
5018 * 6/ If array has spares or is not in-sync, start a resync thread.
5020 void md_check_recovery(mddev_t
*mddev
)
5023 struct list_head
*rtmp
;
5027 bitmap_daemon_work(mddev
->bitmap
);
5032 if (signal_pending(current
)) {
5033 if (mddev
->pers
->sync_request
) {
5034 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5036 mddev
->safemode
= 2;
5038 flush_signals(current
);
5043 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5044 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5045 (mddev
->safemode
== 1) ||
5046 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5047 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5051 if (mddev_trylock(mddev
)) {
5054 spin_lock_irq(&mddev
->write_lock
);
5055 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5056 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5058 mddev
->sb_dirty
= 1;
5060 if (mddev
->safemode
== 1)
5061 mddev
->safemode
= 0;
5062 spin_unlock_irq(&mddev
->write_lock
);
5064 if (mddev
->sb_dirty
)
5065 md_update_sb(mddev
);
5068 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5069 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5070 /* resync/recovery still happening */
5071 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5074 if (mddev
->sync_thread
) {
5075 /* resync has finished, collect result */
5076 md_unregister_thread(mddev
->sync_thread
);
5077 mddev
->sync_thread
= NULL
;
5078 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5079 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5081 /* activate any spares */
5082 mddev
->pers
->spare_active(mddev
);
5084 md_update_sb(mddev
);
5086 /* if array is no-longer degraded, then any saved_raid_disk
5087 * information must be scrapped
5089 if (!mddev
->degraded
)
5090 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5091 rdev
->saved_raid_disk
= -1;
5093 mddev
->recovery
= 0;
5094 /* flag recovery needed just to double check */
5095 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5096 md_new_event(mddev
);
5099 /* Clear some bits that don't mean anything, but
5102 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5103 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5104 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5105 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5107 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5109 /* no recovery is running.
5110 * remove any failed drives, then
5111 * add spares if possible.
5112 * Spare are also removed and re-added, to allow
5113 * the personality to fail the re-add.
5115 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5116 if (rdev
->raid_disk
>= 0 &&
5117 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
5118 atomic_read(&rdev
->nr_pending
)==0) {
5119 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
5121 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5122 sysfs_remove_link(&mddev
->kobj
, nm
);
5123 rdev
->raid_disk
= -1;
5127 if (mddev
->degraded
) {
5128 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5129 if (rdev
->raid_disk
< 0
5130 && !test_bit(Faulty
, &rdev
->flags
)) {
5131 rdev
->recovery_offset
= 0;
5132 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5134 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5135 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
5137 md_new_event(mddev
);
5144 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5145 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5146 } else if (mddev
->recovery_cp
< MaxSector
) {
5147 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5148 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5149 /* nothing to be done ... */
5152 if (mddev
->pers
->sync_request
) {
5153 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5154 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5155 /* We are adding a device or devices to an array
5156 * which has the bitmap stored on all devices.
5157 * So make sure all bitmap pages get written
5159 bitmap_write_all(mddev
->bitmap
);
5161 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5164 if (!mddev
->sync_thread
) {
5165 printk(KERN_ERR
"%s: could not start resync"
5168 /* leave the spares where they are, it shouldn't hurt */
5169 mddev
->recovery
= 0;
5171 md_wakeup_thread(mddev
->sync_thread
);
5172 md_new_event(mddev
);
5175 mddev_unlock(mddev
);
5179 static int md_notify_reboot(struct notifier_block
*this,
5180 unsigned long code
, void *x
)
5182 struct list_head
*tmp
;
5185 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5187 printk(KERN_INFO
"md: stopping all md devices.\n");
5189 ITERATE_MDDEV(mddev
,tmp
)
5190 if (mddev_trylock(mddev
)) {
5191 do_md_stop (mddev
, 1);
5192 mddev_unlock(mddev
);
5195 * certain more exotic SCSI devices are known to be
5196 * volatile wrt too early system reboots. While the
5197 * right place to handle this issue is the given
5198 * driver, we do want to have a safe RAID driver ...
5205 static struct notifier_block md_notifier
= {
5206 .notifier_call
= md_notify_reboot
,
5208 .priority
= INT_MAX
, /* before any real devices */
5211 static void md_geninit(void)
5213 struct proc_dir_entry
*p
;
5215 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5217 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5219 p
->proc_fops
= &md_seq_fops
;
5222 static int __init
md_init(void)
5226 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5227 " MD_SB_DISKS=%d\n",
5228 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
5229 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
5230 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
5233 if (register_blkdev(MAJOR_NR
, "md"))
5235 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5236 unregister_blkdev(MAJOR_NR
, "md");
5240 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
5241 md_probe
, NULL
, NULL
);
5242 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
5243 md_probe
, NULL
, NULL
);
5245 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5246 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
5247 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5250 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
5251 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
5252 S_IFBLK
|S_IRUSR
|S_IWUSR
,
5256 register_reboot_notifier(&md_notifier
);
5257 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
5267 * Searches all registered partitions for autorun RAID arrays
5270 static dev_t detected_devices
[128];
5273 void md_autodetect_dev(dev_t dev
)
5275 if (dev_cnt
>= 0 && dev_cnt
< 127)
5276 detected_devices
[dev_cnt
++] = dev
;
5280 static void autostart_arrays(int part
)
5285 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5287 for (i
= 0; i
< dev_cnt
; i
++) {
5288 dev_t dev
= detected_devices
[i
];
5290 rdev
= md_import_device(dev
,0, 0);
5294 if (test_bit(Faulty
, &rdev
->flags
)) {
5298 list_add(&rdev
->same_set
, &pending_raid_disks
);
5302 autorun_devices(part
);
5307 static __exit
void md_exit(void)
5310 struct list_head
*tmp
;
5312 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
5313 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
5314 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5315 devfs_remove("md/%d", i
);
5316 for (i
=0; i
< MAX_MD_DEVS
; i
++)
5317 devfs_remove("md/d%d", i
);
5321 unregister_blkdev(MAJOR_NR
,"md");
5322 unregister_blkdev(mdp_major
, "mdp");
5323 unregister_reboot_notifier(&md_notifier
);
5324 unregister_sysctl_table(raid_table_header
);
5325 remove_proc_entry("mdstat", NULL
);
5326 ITERATE_MDDEV(mddev
,tmp
) {
5327 struct gendisk
*disk
= mddev
->gendisk
;
5330 export_array(mddev
);
5333 mddev
->gendisk
= NULL
;
5338 module_init(md_init
)
5339 module_exit(md_exit
)
5341 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5343 return sprintf(buffer
, "%d", start_readonly
);
5345 static int set_ro(const char *val
, struct kernel_param
*kp
)
5348 int num
= simple_strtoul(val
, &e
, 10);
5349 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5350 start_readonly
= num
;
5356 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5357 module_param(start_dirty_degraded
, int, 0644);
5360 EXPORT_SYMBOL(register_md_personality
);
5361 EXPORT_SYMBOL(unregister_md_personality
);
5362 EXPORT_SYMBOL(md_error
);
5363 EXPORT_SYMBOL(md_done_sync
);
5364 EXPORT_SYMBOL(md_write_start
);
5365 EXPORT_SYMBOL(md_write_end
);
5366 EXPORT_SYMBOL(md_register_thread
);
5367 EXPORT_SYMBOL(md_unregister_thread
);
5368 EXPORT_SYMBOL(md_wakeup_thread
);
5369 EXPORT_SYMBOL(md_check_recovery
);
5370 MODULE_LICENSE("GPL");
5372 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);