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/kernel.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/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part
);
72 static LIST_HEAD(pers_list
);
73 static DEFINE_SPINLOCK(pers_lock
);
75 static void md_print_devices(void);
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min
= 1000;
93 static int sysctl_speed_limit_max
= 200000;
94 static inline int speed_min(mddev_t
*mddev
)
96 return mddev
->sync_speed_min
?
97 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
100 static inline int speed_max(mddev_t
*mddev
)
102 return mddev
->sync_speed_max
?
103 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
106 static struct ctl_table_header
*raid_table_header
;
108 static ctl_table raid_table
[] = {
110 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
111 .procname
= "speed_limit_min",
112 .data
= &sysctl_speed_limit_min
,
113 .maxlen
= sizeof(int),
114 .mode
= S_IRUGO
|S_IWUSR
,
115 .proc_handler
= &proc_dointvec
,
118 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
119 .procname
= "speed_limit_max",
120 .data
= &sysctl_speed_limit_max
,
121 .maxlen
= sizeof(int),
122 .mode
= S_IRUGO
|S_IWUSR
,
123 .proc_handler
= &proc_dointvec
,
128 static ctl_table raid_dir_table
[] = {
130 .ctl_name
= DEV_RAID
,
133 .mode
= S_IRUGO
|S_IXUGO
,
139 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
165 static atomic_t md_event_count
;
166 void md_new_event(mddev_t
*mddev
)
168 atomic_inc(&md_event_count
);
169 wake_up(&md_event_waiters
);
170 sysfs_notify(&mddev
->kobj
, NULL
, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event
);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t
*mddev
)
179 atomic_inc(&md_event_count
);
180 wake_up(&md_event_waiters
);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs
);
188 static DEFINE_SPINLOCK(all_mddevs_lock
);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define for_each_mddev(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (struct request_queue
*q
, struct bio
*bio
)
220 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
222 atomic_inc(&mddev
->active
);
226 static void mddev_put(mddev_t
*mddev
)
228 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
230 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
231 list_del(&mddev
->all_mddevs
);
232 spin_unlock(&all_mddevs_lock
);
233 blk_cleanup_queue(mddev
->queue
);
234 kobject_put(&mddev
->kobj
);
236 spin_unlock(&all_mddevs_lock
);
239 static mddev_t
* mddev_find(dev_t unit
)
241 mddev_t
*mddev
, *new = NULL
;
244 spin_lock(&all_mddevs_lock
);
245 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
246 if (mddev
->unit
== unit
) {
248 spin_unlock(&all_mddevs_lock
);
254 list_add(&new->all_mddevs
, &all_mddevs
);
255 spin_unlock(&all_mddevs_lock
);
258 spin_unlock(&all_mddevs_lock
);
260 new = kzalloc(sizeof(*new), GFP_KERNEL
);
265 if (MAJOR(unit
) == MD_MAJOR
)
266 new->md_minor
= MINOR(unit
);
268 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
270 mutex_init(&new->reconfig_mutex
);
271 INIT_LIST_HEAD(&new->disks
);
272 INIT_LIST_HEAD(&new->all_mddevs
);
273 init_timer(&new->safemode_timer
);
274 atomic_set(&new->active
, 1);
275 spin_lock_init(&new->write_lock
);
276 init_waitqueue_head(&new->sb_wait
);
277 new->reshape_position
= MaxSector
;
278 new->resync_max
= MaxSector
;
279 new->level
= LEVEL_NONE
;
281 new->queue
= blk_alloc_queue(GFP_KERNEL
);
286 /* Can be unlocked because the queue is new: no concurrency */
287 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, new->queue
);
289 blk_queue_make_request(new->queue
, md_fail_request
);
294 static inline int mddev_lock(mddev_t
* mddev
)
296 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
299 static inline int mddev_trylock(mddev_t
* mddev
)
301 return mutex_trylock(&mddev
->reconfig_mutex
);
304 static inline void mddev_unlock(mddev_t
* mddev
)
306 mutex_unlock(&mddev
->reconfig_mutex
);
308 md_wakeup_thread(mddev
->thread
);
311 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
314 struct list_head
*tmp
;
316 rdev_for_each(rdev
, tmp
, mddev
) {
317 if (rdev
->desc_nr
== nr
)
323 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
325 struct list_head
*tmp
;
328 rdev_for_each(rdev
, tmp
, mddev
) {
329 if (rdev
->bdev
->bd_dev
== dev
)
335 static struct mdk_personality
*find_pers(int level
, char *clevel
)
337 struct mdk_personality
*pers
;
338 list_for_each_entry(pers
, &pers_list
, list
) {
339 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
341 if (strcmp(pers
->name
, clevel
)==0)
347 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
349 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
350 return MD_NEW_SIZE_BLOCKS(size
);
353 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
357 size
= rdev
->sb_offset
;
360 size
&= ~((sector_t
)chunk_size
/1024 - 1);
364 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
369 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
370 if (!rdev
->sb_page
) {
371 printk(KERN_ALERT
"md: out of memory.\n");
378 static void free_disk_sb(mdk_rdev_t
* rdev
)
381 put_page(rdev
->sb_page
);
383 rdev
->sb_page
= NULL
;
390 static void super_written(struct bio
*bio
, int error
)
392 mdk_rdev_t
*rdev
= bio
->bi_private
;
393 mddev_t
*mddev
= rdev
->mddev
;
395 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
396 printk("md: super_written gets error=%d, uptodate=%d\n",
397 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
398 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
399 md_error(mddev
, rdev
);
402 if (atomic_dec_and_test(&mddev
->pending_writes
))
403 wake_up(&mddev
->sb_wait
);
407 static void super_written_barrier(struct bio
*bio
, int error
)
409 struct bio
*bio2
= bio
->bi_private
;
410 mdk_rdev_t
*rdev
= bio2
->bi_private
;
411 mddev_t
*mddev
= rdev
->mddev
;
413 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
414 error
== -EOPNOTSUPP
) {
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp
, &rdev
->flags
);
418 mddev
->barriers_work
= 0;
419 spin_lock_irqsave(&mddev
->write_lock
, flags
);
420 bio2
->bi_next
= mddev
->biolist
;
421 mddev
->biolist
= bio2
;
422 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
423 wake_up(&mddev
->sb_wait
);
427 bio
->bi_private
= rdev
;
428 super_written(bio
, error
);
432 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
433 sector_t sector
, int size
, struct page
*page
)
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
444 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
445 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
447 bio
->bi_bdev
= rdev
->bdev
;
448 bio
->bi_sector
= sector
;
449 bio_add_page(bio
, page
, size
, 0);
450 bio
->bi_private
= rdev
;
451 bio
->bi_end_io
= super_written
;
454 atomic_inc(&mddev
->pending_writes
);
455 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
457 rw
|= (1<<BIO_RW_BARRIER
);
458 rbio
= bio_clone(bio
, GFP_NOIO
);
459 rbio
->bi_private
= bio
;
460 rbio
->bi_end_io
= super_written_barrier
;
461 submit_bio(rw
, rbio
);
466 void md_super_wait(mddev_t
*mddev
)
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
473 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
474 if (atomic_read(&mddev
->pending_writes
)==0)
476 while (mddev
->biolist
) {
478 spin_lock_irq(&mddev
->write_lock
);
479 bio
= mddev
->biolist
;
480 mddev
->biolist
= bio
->bi_next
;
482 spin_unlock_irq(&mddev
->write_lock
);
483 submit_bio(bio
->bi_rw
, bio
);
487 finish_wait(&mddev
->sb_wait
, &wq
);
490 static void bi_complete(struct bio
*bio
, int error
)
492 complete((struct completion
*)bio
->bi_private
);
495 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
496 struct page
*page
, int rw
)
498 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
499 struct completion event
;
502 rw
|= (1 << BIO_RW_SYNC
);
505 bio
->bi_sector
= sector
;
506 bio_add_page(bio
, page
, size
, 0);
507 init_completion(&event
);
508 bio
->bi_private
= &event
;
509 bio
->bi_end_io
= bi_complete
;
511 wait_for_completion(&event
);
513 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
517 EXPORT_SYMBOL_GPL(sync_page_io
);
519 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
521 char b
[BDEVNAME_SIZE
];
522 if (!rdev
->sb_page
) {
530 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
536 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
537 bdevname(rdev
->bdev
,b
));
541 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
543 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
544 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
545 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
546 (sb1
->set_uuid3
== sb2
->set_uuid3
))
554 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
557 mdp_super_t
*tmp1
, *tmp2
;
559 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
560 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
562 if (!tmp1
|| !tmp2
) {
564 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
572 * nr_disks is not constant
577 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
589 static u32
md_csum_fold(u32 csum
)
591 csum
= (csum
& 0xffff) + (csum
>> 16);
592 return (csum
& 0xffff) + (csum
>> 16);
595 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
598 u32
*sb32
= (u32
*)sb
;
600 unsigned int disk_csum
, csum
;
602 disk_csum
= sb
->sb_csum
;
605 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
607 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
611 /* This used to use csum_partial, which was wrong for several
612 * reasons including that different results are returned on
613 * different architectures. It isn't critical that we get exactly
614 * the same return value as before (we always csum_fold before
615 * testing, and that removes any differences). However as we
616 * know that csum_partial always returned a 16bit value on
617 * alphas, do a fold to maximise conformity to previous behaviour.
619 sb
->sb_csum
= md_csum_fold(disk_csum
);
621 sb
->sb_csum
= disk_csum
;
628 * Handle superblock details.
629 * We want to be able to handle multiple superblock formats
630 * so we have a common interface to them all, and an array of
631 * different handlers.
632 * We rely on user-space to write the initial superblock, and support
633 * reading and updating of superblocks.
634 * Interface methods are:
635 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
636 * loads and validates a superblock on dev.
637 * if refdev != NULL, compare superblocks on both devices
639 * 0 - dev has a superblock that is compatible with refdev
640 * 1 - dev has a superblock that is compatible and newer than refdev
641 * so dev should be used as the refdev in future
642 * -EINVAL superblock incompatible or invalid
643 * -othererror e.g. -EIO
645 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
646 * Verify that dev is acceptable into mddev.
647 * The first time, mddev->raid_disks will be 0, and data from
648 * dev should be merged in. Subsequent calls check that dev
649 * is new enough. Return 0 or -EINVAL
651 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
652 * Update the superblock for rdev with data in mddev
653 * This does not write to disc.
659 struct module
*owner
;
660 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
661 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
662 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
666 * load_super for 0.90.0
668 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
670 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
676 * Calculate the position of the superblock,
677 * it's at the end of the disk.
679 * It also happens to be a multiple of 4Kb.
681 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
682 rdev
->sb_offset
= sb_offset
;
684 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
689 bdevname(rdev
->bdev
, b
);
690 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
692 if (sb
->md_magic
!= MD_SB_MAGIC
) {
693 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
698 if (sb
->major_version
!= 0 ||
699 sb
->minor_version
< 90 ||
700 sb
->minor_version
> 91) {
701 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
702 sb
->major_version
, sb
->minor_version
,
707 if (sb
->raid_disks
<= 0)
710 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
711 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
716 rdev
->preferred_minor
= sb
->md_minor
;
717 rdev
->data_offset
= 0;
718 rdev
->sb_size
= MD_SB_BYTES
;
720 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
721 if (sb
->level
!= 1 && sb
->level
!= 4
722 && sb
->level
!= 5 && sb
->level
!= 6
723 && sb
->level
!= 10) {
724 /* FIXME use a better test */
726 "md: bitmaps not supported for this level.\n");
731 if (sb
->level
== LEVEL_MULTIPATH
)
734 rdev
->desc_nr
= sb
->this_disk
.number
;
740 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
741 if (!uuid_equal(refsb
, sb
)) {
742 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
743 b
, bdevname(refdev
->bdev
,b2
));
746 if (!sb_equal(refsb
, sb
)) {
747 printk(KERN_WARNING
"md: %s has same UUID"
748 " but different superblock to %s\n",
749 b
, bdevname(refdev
->bdev
, b2
));
753 ev2
= md_event(refsb
);
759 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
761 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
762 /* "this cannot possibly happen" ... */
770 * validate_super for 0.90.0
772 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
775 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
776 __u64 ev1
= md_event(sb
);
778 rdev
->raid_disk
= -1;
779 clear_bit(Faulty
, &rdev
->flags
);
780 clear_bit(In_sync
, &rdev
->flags
);
781 clear_bit(WriteMostly
, &rdev
->flags
);
782 clear_bit(BarriersNotsupp
, &rdev
->flags
);
784 if (mddev
->raid_disks
== 0) {
785 mddev
->major_version
= 0;
786 mddev
->minor_version
= sb
->minor_version
;
787 mddev
->patch_version
= sb
->patch_version
;
789 mddev
->chunk_size
= sb
->chunk_size
;
790 mddev
->ctime
= sb
->ctime
;
791 mddev
->utime
= sb
->utime
;
792 mddev
->level
= sb
->level
;
793 mddev
->clevel
[0] = 0;
794 mddev
->layout
= sb
->layout
;
795 mddev
->raid_disks
= sb
->raid_disks
;
796 mddev
->size
= sb
->size
;
798 mddev
->bitmap_offset
= 0;
799 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
801 if (mddev
->minor_version
>= 91) {
802 mddev
->reshape_position
= sb
->reshape_position
;
803 mddev
->delta_disks
= sb
->delta_disks
;
804 mddev
->new_level
= sb
->new_level
;
805 mddev
->new_layout
= sb
->new_layout
;
806 mddev
->new_chunk
= sb
->new_chunk
;
808 mddev
->reshape_position
= MaxSector
;
809 mddev
->delta_disks
= 0;
810 mddev
->new_level
= mddev
->level
;
811 mddev
->new_layout
= mddev
->layout
;
812 mddev
->new_chunk
= mddev
->chunk_size
;
815 if (sb
->state
& (1<<MD_SB_CLEAN
))
816 mddev
->recovery_cp
= MaxSector
;
818 if (sb
->events_hi
== sb
->cp_events_hi
&&
819 sb
->events_lo
== sb
->cp_events_lo
) {
820 mddev
->recovery_cp
= sb
->recovery_cp
;
822 mddev
->recovery_cp
= 0;
825 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
826 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
827 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
828 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
830 mddev
->max_disks
= MD_SB_DISKS
;
832 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
833 mddev
->bitmap_file
== NULL
)
834 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
836 } else if (mddev
->pers
== NULL
) {
837 /* Insist on good event counter while assembling */
839 if (ev1
< mddev
->events
)
841 } else if (mddev
->bitmap
) {
842 /* if adding to array with a bitmap, then we can accept an
843 * older device ... but not too old.
845 if (ev1
< mddev
->bitmap
->events_cleared
)
848 if (ev1
< mddev
->events
)
849 /* just a hot-add of a new device, leave raid_disk at -1 */
853 if (mddev
->level
!= LEVEL_MULTIPATH
) {
854 desc
= sb
->disks
+ rdev
->desc_nr
;
856 if (desc
->state
& (1<<MD_DISK_FAULTY
))
857 set_bit(Faulty
, &rdev
->flags
);
858 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
859 desc->raid_disk < mddev->raid_disks */) {
860 set_bit(In_sync
, &rdev
->flags
);
861 rdev
->raid_disk
= desc
->raid_disk
;
863 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
864 set_bit(WriteMostly
, &rdev
->flags
);
865 } else /* MULTIPATH are always insync */
866 set_bit(In_sync
, &rdev
->flags
);
871 * sync_super for 0.90.0
873 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
876 struct list_head
*tmp
;
878 int next_spare
= mddev
->raid_disks
;
881 /* make rdev->sb match mddev data..
884 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
885 * 3/ any empty disks < next_spare become removed
887 * disks[0] gets initialised to REMOVED because
888 * we cannot be sure from other fields if it has
889 * been initialised or not.
892 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
894 rdev
->sb_size
= MD_SB_BYTES
;
896 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
898 memset(sb
, 0, sizeof(*sb
));
900 sb
->md_magic
= MD_SB_MAGIC
;
901 sb
->major_version
= mddev
->major_version
;
902 sb
->patch_version
= mddev
->patch_version
;
903 sb
->gvalid_words
= 0; /* ignored */
904 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
905 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
906 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
907 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
909 sb
->ctime
= mddev
->ctime
;
910 sb
->level
= mddev
->level
;
911 sb
->size
= mddev
->size
;
912 sb
->raid_disks
= mddev
->raid_disks
;
913 sb
->md_minor
= mddev
->md_minor
;
914 sb
->not_persistent
= 0;
915 sb
->utime
= mddev
->utime
;
917 sb
->events_hi
= (mddev
->events
>>32);
918 sb
->events_lo
= (u32
)mddev
->events
;
920 if (mddev
->reshape_position
== MaxSector
)
921 sb
->minor_version
= 90;
923 sb
->minor_version
= 91;
924 sb
->reshape_position
= mddev
->reshape_position
;
925 sb
->new_level
= mddev
->new_level
;
926 sb
->delta_disks
= mddev
->delta_disks
;
927 sb
->new_layout
= mddev
->new_layout
;
928 sb
->new_chunk
= mddev
->new_chunk
;
930 mddev
->minor_version
= sb
->minor_version
;
933 sb
->recovery_cp
= mddev
->recovery_cp
;
934 sb
->cp_events_hi
= (mddev
->events
>>32);
935 sb
->cp_events_lo
= (u32
)mddev
->events
;
936 if (mddev
->recovery_cp
== MaxSector
)
937 sb
->state
= (1<< MD_SB_CLEAN
);
941 sb
->layout
= mddev
->layout
;
942 sb
->chunk_size
= mddev
->chunk_size
;
944 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
945 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
947 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
948 rdev_for_each(rdev2
, tmp
, mddev
) {
951 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
952 && !test_bit(Faulty
, &rdev2
->flags
))
953 desc_nr
= rdev2
->raid_disk
;
955 desc_nr
= next_spare
++;
956 rdev2
->desc_nr
= desc_nr
;
957 d
= &sb
->disks
[rdev2
->desc_nr
];
959 d
->number
= rdev2
->desc_nr
;
960 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
961 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
962 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
963 && !test_bit(Faulty
, &rdev2
->flags
))
964 d
->raid_disk
= rdev2
->raid_disk
;
966 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
967 if (test_bit(Faulty
, &rdev2
->flags
))
968 d
->state
= (1<<MD_DISK_FAULTY
);
969 else if (test_bit(In_sync
, &rdev2
->flags
)) {
970 d
->state
= (1<<MD_DISK_ACTIVE
);
971 d
->state
|= (1<<MD_DISK_SYNC
);
979 if (test_bit(WriteMostly
, &rdev2
->flags
))
980 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
982 /* now set the "removed" and "faulty" bits on any missing devices */
983 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
984 mdp_disk_t
*d
= &sb
->disks
[i
];
985 if (d
->state
== 0 && d
->number
== 0) {
988 d
->state
= (1<<MD_DISK_REMOVED
);
989 d
->state
|= (1<<MD_DISK_FAULTY
);
993 sb
->nr_disks
= nr_disks
;
994 sb
->active_disks
= active
;
995 sb
->working_disks
= working
;
996 sb
->failed_disks
= failed
;
997 sb
->spare_disks
= spare
;
999 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1000 sb
->sb_csum
= calc_sb_csum(sb
);
1004 * version 1 superblock
1007 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1011 unsigned long long newcsum
;
1012 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1013 __le32
*isuper
= (__le32
*)sb
;
1016 disk_csum
= sb
->sb_csum
;
1019 for (i
=0; size
>=4; size
-= 4 )
1020 newcsum
+= le32_to_cpu(*isuper
++);
1023 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1025 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1026 sb
->sb_csum
= disk_csum
;
1027 return cpu_to_le32(csum
);
1030 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1032 struct mdp_superblock_1
*sb
;
1035 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1039 * Calculate the position of the superblock.
1040 * It is always aligned to a 4K boundary and
1041 * depeding on minor_version, it can be:
1042 * 0: At least 8K, but less than 12K, from end of device
1043 * 1: At start of device
1044 * 2: 4K from start of device.
1046 switch(minor_version
) {
1048 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1050 sb_offset
&= ~(sector_t
)(4*2-1);
1051 /* convert from sectors to K */
1063 rdev
->sb_offset
= sb_offset
;
1065 /* superblock is rarely larger than 1K, but it can be larger,
1066 * and it is safe to read 4k, so we do that
1068 ret
= read_disk_sb(rdev
, 4096);
1069 if (ret
) return ret
;
1072 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1074 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1075 sb
->major_version
!= cpu_to_le32(1) ||
1076 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1077 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1078 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1081 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1082 printk("md: invalid superblock checksum on %s\n",
1083 bdevname(rdev
->bdev
,b
));
1086 if (le64_to_cpu(sb
->data_size
) < 10) {
1087 printk("md: data_size too small on %s\n",
1088 bdevname(rdev
->bdev
,b
));
1091 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)) {
1092 if (sb
->level
!= cpu_to_le32(1) &&
1093 sb
->level
!= cpu_to_le32(4) &&
1094 sb
->level
!= cpu_to_le32(5) &&
1095 sb
->level
!= cpu_to_le32(6) &&
1096 sb
->level
!= cpu_to_le32(10)) {
1098 "md: bitmaps not supported for this level.\n");
1103 rdev
->preferred_minor
= 0xffff;
1104 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1105 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1107 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1108 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1109 if (rdev
->sb_size
& bmask
)
1110 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1113 && rdev
->data_offset
< sb_offset
+ (rdev
->sb_size
/512))
1116 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1119 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1125 struct mdp_superblock_1
*refsb
=
1126 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1128 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1129 sb
->level
!= refsb
->level
||
1130 sb
->layout
!= refsb
->layout
||
1131 sb
->chunksize
!= refsb
->chunksize
) {
1132 printk(KERN_WARNING
"md: %s has strangely different"
1133 " superblock to %s\n",
1134 bdevname(rdev
->bdev
,b
),
1135 bdevname(refdev
->bdev
,b2
));
1138 ev1
= le64_to_cpu(sb
->events
);
1139 ev2
= le64_to_cpu(refsb
->events
);
1147 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1149 rdev
->size
= rdev
->sb_offset
;
1150 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1152 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1153 if (le32_to_cpu(sb
->chunksize
))
1154 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1156 if (le64_to_cpu(sb
->size
) > rdev
->size
*2)
1161 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1163 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1164 __u64 ev1
= le64_to_cpu(sb
->events
);
1166 rdev
->raid_disk
= -1;
1167 clear_bit(Faulty
, &rdev
->flags
);
1168 clear_bit(In_sync
, &rdev
->flags
);
1169 clear_bit(WriteMostly
, &rdev
->flags
);
1170 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1172 if (mddev
->raid_disks
== 0) {
1173 mddev
->major_version
= 1;
1174 mddev
->patch_version
= 0;
1175 mddev
->external
= 0;
1176 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1177 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1178 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1179 mddev
->level
= le32_to_cpu(sb
->level
);
1180 mddev
->clevel
[0] = 0;
1181 mddev
->layout
= le32_to_cpu(sb
->layout
);
1182 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1183 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1184 mddev
->events
= ev1
;
1185 mddev
->bitmap_offset
= 0;
1186 mddev
->default_bitmap_offset
= 1024 >> 9;
1188 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1189 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1191 mddev
->max_disks
= (4096-256)/2;
1193 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1194 mddev
->bitmap_file
== NULL
)
1195 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1197 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1198 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1199 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1200 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1201 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1202 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1204 mddev
->reshape_position
= MaxSector
;
1205 mddev
->delta_disks
= 0;
1206 mddev
->new_level
= mddev
->level
;
1207 mddev
->new_layout
= mddev
->layout
;
1208 mddev
->new_chunk
= mddev
->chunk_size
;
1211 } else if (mddev
->pers
== NULL
) {
1212 /* Insist of good event counter while assembling */
1214 if (ev1
< mddev
->events
)
1216 } else if (mddev
->bitmap
) {
1217 /* If adding to array with a bitmap, then we can accept an
1218 * older device, but not too old.
1220 if (ev1
< mddev
->bitmap
->events_cleared
)
1223 if (ev1
< mddev
->events
)
1224 /* just a hot-add of a new device, leave raid_disk at -1 */
1227 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1229 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1231 case 0xffff: /* spare */
1233 case 0xfffe: /* faulty */
1234 set_bit(Faulty
, &rdev
->flags
);
1237 if ((le32_to_cpu(sb
->feature_map
) &
1238 MD_FEATURE_RECOVERY_OFFSET
))
1239 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1241 set_bit(In_sync
, &rdev
->flags
);
1242 rdev
->raid_disk
= role
;
1245 if (sb
->devflags
& WriteMostly1
)
1246 set_bit(WriteMostly
, &rdev
->flags
);
1247 } else /* MULTIPATH are always insync */
1248 set_bit(In_sync
, &rdev
->flags
);
1253 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1255 struct mdp_superblock_1
*sb
;
1256 struct list_head
*tmp
;
1259 /* make rdev->sb match mddev and rdev data. */
1261 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1263 sb
->feature_map
= 0;
1265 sb
->recovery_offset
= cpu_to_le64(0);
1266 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1267 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1268 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1270 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1271 sb
->events
= cpu_to_le64(mddev
->events
);
1273 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1275 sb
->resync_offset
= cpu_to_le64(0);
1277 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1279 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1280 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1282 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1283 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1284 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1287 if (rdev
->raid_disk
>= 0 &&
1288 !test_bit(In_sync
, &rdev
->flags
) &&
1289 rdev
->recovery_offset
> 0) {
1290 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1291 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1294 if (mddev
->reshape_position
!= MaxSector
) {
1295 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1296 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1297 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1298 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1299 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1300 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1304 rdev_for_each(rdev2
, tmp
, mddev
)
1305 if (rdev2
->desc_nr
+1 > max_dev
)
1306 max_dev
= rdev2
->desc_nr
+1;
1308 if (max_dev
> le32_to_cpu(sb
->max_dev
))
1309 sb
->max_dev
= cpu_to_le32(max_dev
);
1310 for (i
=0; i
<max_dev
;i
++)
1311 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1313 rdev_for_each(rdev2
, tmp
, mddev
) {
1315 if (test_bit(Faulty
, &rdev2
->flags
))
1316 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1317 else if (test_bit(In_sync
, &rdev2
->flags
))
1318 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1319 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1320 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1322 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1325 sb
->sb_csum
= calc_sb_1_csum(sb
);
1329 static struct super_type super_types
[] = {
1332 .owner
= THIS_MODULE
,
1333 .load_super
= super_90_load
,
1334 .validate_super
= super_90_validate
,
1335 .sync_super
= super_90_sync
,
1339 .owner
= THIS_MODULE
,
1340 .load_super
= super_1_load
,
1341 .validate_super
= super_1_validate
,
1342 .sync_super
= super_1_sync
,
1346 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1348 struct list_head
*tmp
, *tmp2
;
1349 mdk_rdev_t
*rdev
, *rdev2
;
1351 rdev_for_each(rdev
, tmp
, mddev1
)
1352 rdev_for_each(rdev2
, tmp2
, mddev2
)
1353 if (rdev
->bdev
->bd_contains
==
1354 rdev2
->bdev
->bd_contains
)
1360 static LIST_HEAD(pending_raid_disks
);
1362 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1364 char b
[BDEVNAME_SIZE
];
1373 /* make sure rdev->size exceeds mddev->size */
1374 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1376 /* Cannot change size, so fail
1377 * If mddev->level <= 0, then we don't care
1378 * about aligning sizes (e.g. linear)
1380 if (mddev
->level
> 0)
1383 mddev
->size
= rdev
->size
;
1386 /* Verify rdev->desc_nr is unique.
1387 * If it is -1, assign a free number, else
1388 * check number is not in use
1390 if (rdev
->desc_nr
< 0) {
1392 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1393 while (find_rdev_nr(mddev
, choice
))
1395 rdev
->desc_nr
= choice
;
1397 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1400 bdevname(rdev
->bdev
,b
);
1401 while ( (s
=strchr(b
, '/')) != NULL
)
1404 rdev
->mddev
= mddev
;
1405 printk(KERN_INFO
"md: bind<%s>\n", b
);
1407 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1410 if (rdev
->bdev
->bd_part
)
1411 ko
= &rdev
->bdev
->bd_part
->dev
.kobj
;
1413 ko
= &rdev
->bdev
->bd_disk
->dev
.kobj
;
1414 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1415 kobject_del(&rdev
->kobj
);
1418 list_add(&rdev
->same_set
, &mddev
->disks
);
1419 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1423 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1428 static void md_delayed_delete(struct work_struct
*ws
)
1430 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1431 kobject_del(&rdev
->kobj
);
1432 kobject_put(&rdev
->kobj
);
1435 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1437 char b
[BDEVNAME_SIZE
];
1442 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1443 list_del_init(&rdev
->same_set
);
1444 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1446 sysfs_remove_link(&rdev
->kobj
, "block");
1448 /* We need to delay this, otherwise we can deadlock when
1449 * writing to 'remove' to "dev/state"
1451 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1452 kobject_get(&rdev
->kobj
);
1453 schedule_work(&rdev
->del_work
);
1457 * prevent the device from being mounted, repartitioned or
1458 * otherwise reused by a RAID array (or any other kernel
1459 * subsystem), by bd_claiming the device.
1461 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1464 struct block_device
*bdev
;
1465 char b
[BDEVNAME_SIZE
];
1467 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1469 printk(KERN_ERR
"md: could not open %s.\n",
1470 __bdevname(dev
, b
));
1471 return PTR_ERR(bdev
);
1473 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1475 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1481 set_bit(AllReserved
, &rdev
->flags
);
1486 static void unlock_rdev(mdk_rdev_t
*rdev
)
1488 struct block_device
*bdev
= rdev
->bdev
;
1496 void md_autodetect_dev(dev_t dev
);
1498 static void export_rdev(mdk_rdev_t
* rdev
)
1500 char b
[BDEVNAME_SIZE
];
1501 printk(KERN_INFO
"md: export_rdev(%s)\n",
1502 bdevname(rdev
->bdev
,b
));
1506 list_del_init(&rdev
->same_set
);
1508 if (test_bit(AutoDetected
, &rdev
->flags
))
1509 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1512 kobject_put(&rdev
->kobj
);
1515 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1517 unbind_rdev_from_array(rdev
);
1521 static void export_array(mddev_t
*mddev
)
1523 struct list_head
*tmp
;
1526 rdev_for_each(rdev
, tmp
, mddev
) {
1531 kick_rdev_from_array(rdev
);
1533 if (!list_empty(&mddev
->disks
))
1535 mddev
->raid_disks
= 0;
1536 mddev
->major_version
= 0;
1539 static void print_desc(mdp_disk_t
*desc
)
1541 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1542 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1545 static void print_sb(mdp_super_t
*sb
)
1550 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1551 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1552 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1554 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1555 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1556 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1557 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1558 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1559 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1560 sb
->failed_disks
, sb
->spare_disks
,
1561 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1564 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1567 desc
= sb
->disks
+ i
;
1568 if (desc
->number
|| desc
->major
|| desc
->minor
||
1569 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1570 printk(" D %2d: ", i
);
1574 printk(KERN_INFO
"md: THIS: ");
1575 print_desc(&sb
->this_disk
);
1579 static void print_rdev(mdk_rdev_t
*rdev
)
1581 char b
[BDEVNAME_SIZE
];
1582 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1583 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1584 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1586 if (rdev
->sb_loaded
) {
1587 printk(KERN_INFO
"md: rdev superblock:\n");
1588 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1590 printk(KERN_INFO
"md: no rdev superblock!\n");
1593 static void md_print_devices(void)
1595 struct list_head
*tmp
, *tmp2
;
1598 char b
[BDEVNAME_SIZE
];
1601 printk("md: **********************************\n");
1602 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1603 printk("md: **********************************\n");
1604 for_each_mddev(mddev
, tmp
) {
1607 bitmap_print_sb(mddev
->bitmap
);
1609 printk("%s: ", mdname(mddev
));
1610 rdev_for_each(rdev
, tmp2
, mddev
)
1611 printk("<%s>", bdevname(rdev
->bdev
,b
));
1614 rdev_for_each(rdev
, tmp2
, mddev
)
1617 printk("md: **********************************\n");
1622 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1624 /* Update each superblock (in-memory image), but
1625 * if we are allowed to, skip spares which already
1626 * have the right event counter, or have one earlier
1627 * (which would mean they aren't being marked as dirty
1628 * with the rest of the array)
1631 struct list_head
*tmp
;
1633 rdev_for_each(rdev
, tmp
, mddev
) {
1634 if (rdev
->sb_events
== mddev
->events
||
1636 rdev
->raid_disk
< 0 &&
1637 (rdev
->sb_events
&1)==0 &&
1638 rdev
->sb_events
+1 == mddev
->events
)) {
1639 /* Don't update this superblock */
1640 rdev
->sb_loaded
= 2;
1642 super_types
[mddev
->major_version
].
1643 sync_super(mddev
, rdev
);
1644 rdev
->sb_loaded
= 1;
1649 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1651 struct list_head
*tmp
;
1656 if (mddev
->external
)
1659 spin_lock_irq(&mddev
->write_lock
);
1661 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1662 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1664 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1665 /* just a clean<-> dirty transition, possibly leave spares alone,
1666 * though if events isn't the right even/odd, we will have to do
1672 if (mddev
->degraded
)
1673 /* If the array is degraded, then skipping spares is both
1674 * dangerous and fairly pointless.
1675 * Dangerous because a device that was removed from the array
1676 * might have a event_count that still looks up-to-date,
1677 * so it can be re-added without a resync.
1678 * Pointless because if there are any spares to skip,
1679 * then a recovery will happen and soon that array won't
1680 * be degraded any more and the spare can go back to sleep then.
1684 sync_req
= mddev
->in_sync
;
1685 mddev
->utime
= get_seconds();
1687 /* If this is just a dirty<->clean transition, and the array is clean
1688 * and 'events' is odd, we can roll back to the previous clean state */
1690 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1691 && (mddev
->events
& 1)
1692 && mddev
->events
!= 1)
1695 /* otherwise we have to go forward and ... */
1697 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1698 /* .. if the array isn't clean, insist on an odd 'events' */
1699 if ((mddev
->events
&1)==0) {
1704 /* otherwise insist on an even 'events' (for clean states) */
1705 if ((mddev
->events
&1)) {
1712 if (!mddev
->events
) {
1714 * oops, this 64-bit counter should never wrap.
1715 * Either we are in around ~1 trillion A.C., assuming
1716 * 1 reboot per second, or we have a bug:
1723 * do not write anything to disk if using
1724 * nonpersistent superblocks
1726 if (!mddev
->persistent
) {
1727 if (!mddev
->external
)
1728 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1730 spin_unlock_irq(&mddev
->write_lock
);
1731 wake_up(&mddev
->sb_wait
);
1734 sync_sbs(mddev
, nospares
);
1735 spin_unlock_irq(&mddev
->write_lock
);
1738 "md: updating %s RAID superblock on device (in sync %d)\n",
1739 mdname(mddev
),mddev
->in_sync
);
1741 bitmap_update_sb(mddev
->bitmap
);
1742 rdev_for_each(rdev
, tmp
, mddev
) {
1743 char b
[BDEVNAME_SIZE
];
1744 dprintk(KERN_INFO
"md: ");
1745 if (rdev
->sb_loaded
!= 1)
1746 continue; /* no noise on spare devices */
1747 if (test_bit(Faulty
, &rdev
->flags
))
1748 dprintk("(skipping faulty ");
1750 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1751 if (!test_bit(Faulty
, &rdev
->flags
)) {
1752 md_super_write(mddev
,rdev
,
1753 rdev
->sb_offset
<<1, rdev
->sb_size
,
1755 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1756 bdevname(rdev
->bdev
,b
),
1757 (unsigned long long)rdev
->sb_offset
);
1758 rdev
->sb_events
= mddev
->events
;
1762 if (mddev
->level
== LEVEL_MULTIPATH
)
1763 /* only need to write one superblock... */
1766 md_super_wait(mddev
);
1767 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1769 spin_lock_irq(&mddev
->write_lock
);
1770 if (mddev
->in_sync
!= sync_req
||
1771 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
1772 /* have to write it out again */
1773 spin_unlock_irq(&mddev
->write_lock
);
1776 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1777 spin_unlock_irq(&mddev
->write_lock
);
1778 wake_up(&mddev
->sb_wait
);
1782 /* words written to sysfs files may, or my not, be \n terminated.
1783 * We want to accept with case. For this we use cmd_match.
1785 static int cmd_match(const char *cmd
, const char *str
)
1787 /* See if cmd, written into a sysfs file, matches
1788 * str. They must either be the same, or cmd can
1789 * have a trailing newline
1791 while (*cmd
&& *str
&& *cmd
== *str
) {
1802 struct rdev_sysfs_entry
{
1803 struct attribute attr
;
1804 ssize_t (*show
)(mdk_rdev_t
*, char *);
1805 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1809 state_show(mdk_rdev_t
*rdev
, char *page
)
1814 if (test_bit(Faulty
, &rdev
->flags
)) {
1815 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1818 if (test_bit(In_sync
, &rdev
->flags
)) {
1819 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1822 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1823 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1826 if (!test_bit(Faulty
, &rdev
->flags
) &&
1827 !test_bit(In_sync
, &rdev
->flags
)) {
1828 len
+= sprintf(page
+len
, "%sspare", sep
);
1831 return len
+sprintf(page
+len
, "\n");
1835 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1838 * faulty - simulates and error
1839 * remove - disconnects the device
1840 * writemostly - sets write_mostly
1841 * -writemostly - clears write_mostly
1844 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1845 md_error(rdev
->mddev
, rdev
);
1847 } else if (cmd_match(buf
, "remove")) {
1848 if (rdev
->raid_disk
>= 0)
1851 mddev_t
*mddev
= rdev
->mddev
;
1852 kick_rdev_from_array(rdev
);
1854 md_update_sb(mddev
, 1);
1855 md_new_event(mddev
);
1858 } else if (cmd_match(buf
, "writemostly")) {
1859 set_bit(WriteMostly
, &rdev
->flags
);
1861 } else if (cmd_match(buf
, "-writemostly")) {
1862 clear_bit(WriteMostly
, &rdev
->flags
);
1865 return err
? err
: len
;
1867 static struct rdev_sysfs_entry rdev_state
=
1868 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
1871 errors_show(mdk_rdev_t
*rdev
, char *page
)
1873 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1877 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1880 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1881 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1882 atomic_set(&rdev
->corrected_errors
, n
);
1887 static struct rdev_sysfs_entry rdev_errors
=
1888 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
1891 slot_show(mdk_rdev_t
*rdev
, char *page
)
1893 if (rdev
->raid_disk
< 0)
1894 return sprintf(page
, "none\n");
1896 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1900 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1905 int slot
= simple_strtoul(buf
, &e
, 10);
1906 if (strncmp(buf
, "none", 4)==0)
1908 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1910 if (rdev
->mddev
->pers
) {
1911 /* Setting 'slot' on an active array requires also
1912 * updating the 'rd%d' link, and communicating
1913 * with the personality with ->hot_*_disk.
1914 * For now we only support removing
1915 * failed/spare devices. This normally happens automatically,
1916 * but not when the metadata is externally managed.
1920 if (rdev
->raid_disk
== -1)
1922 /* personality does all needed checks */
1923 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
1925 err
= rdev
->mddev
->pers
->
1926 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
1929 sprintf(nm
, "rd%d", rdev
->raid_disk
);
1930 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
1931 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
1932 md_wakeup_thread(rdev
->mddev
->thread
);
1934 if (slot
>= rdev
->mddev
->raid_disks
)
1936 rdev
->raid_disk
= slot
;
1937 /* assume it is working */
1938 clear_bit(Faulty
, &rdev
->flags
);
1939 clear_bit(WriteMostly
, &rdev
->flags
);
1940 set_bit(In_sync
, &rdev
->flags
);
1946 static struct rdev_sysfs_entry rdev_slot
=
1947 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
1950 offset_show(mdk_rdev_t
*rdev
, char *page
)
1952 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1956 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1959 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1960 if (e
==buf
|| (*e
&& *e
!= '\n'))
1962 if (rdev
->mddev
->pers
)
1964 if (rdev
->size
&& rdev
->mddev
->external
)
1965 /* Must set offset before size, so overlap checks
1968 rdev
->data_offset
= offset
;
1972 static struct rdev_sysfs_entry rdev_offset
=
1973 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
1976 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1978 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1981 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
1983 /* check if two start/length pairs overlap */
1992 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1995 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1996 unsigned long long oldsize
= rdev
->size
;
1997 mddev_t
*my_mddev
= rdev
->mddev
;
1999 if (e
==buf
|| (*e
&& *e
!= '\n'))
2004 if (size
> oldsize
&& rdev
->mddev
->external
) {
2005 /* need to check that all other rdevs with the same ->bdev
2006 * do not overlap. We need to unlock the mddev to avoid
2007 * a deadlock. We have already changed rdev->size, and if
2008 * we have to change it back, we will have the lock again.
2012 struct list_head
*tmp
, *tmp2
;
2014 mddev_unlock(my_mddev
);
2015 for_each_mddev(mddev
, tmp
) {
2019 rdev_for_each(rdev2
, tmp2
, mddev
)
2020 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2021 (rdev
->bdev
== rdev2
->bdev
&&
2023 overlaps(rdev
->data_offset
, rdev
->size
,
2024 rdev2
->data_offset
, rdev2
->size
))) {
2028 mddev_unlock(mddev
);
2034 mddev_lock(my_mddev
);
2036 /* Someone else could have slipped in a size
2037 * change here, but doing so is just silly.
2038 * We put oldsize back because we *know* it is
2039 * safe, and trust userspace not to race with
2042 rdev
->size
= oldsize
;
2046 if (size
< my_mddev
->size
|| my_mddev
->size
== 0)
2047 my_mddev
->size
= size
;
2051 static struct rdev_sysfs_entry rdev_size
=
2052 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2054 static struct attribute
*rdev_default_attrs
[] = {
2063 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2065 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2066 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2067 mddev_t
*mddev
= rdev
->mddev
;
2073 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2075 if (rdev
->mddev
== NULL
)
2078 rv
= entry
->show(rdev
, page
);
2079 mddev_unlock(mddev
);
2085 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2086 const char *page
, size_t length
)
2088 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2089 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2091 mddev_t
*mddev
= rdev
->mddev
;
2095 if (!capable(CAP_SYS_ADMIN
))
2097 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2099 if (rdev
->mddev
== NULL
)
2102 rv
= entry
->store(rdev
, page
, length
);
2103 mddev_unlock(mddev
);
2108 static void rdev_free(struct kobject
*ko
)
2110 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2113 static struct sysfs_ops rdev_sysfs_ops
= {
2114 .show
= rdev_attr_show
,
2115 .store
= rdev_attr_store
,
2117 static struct kobj_type rdev_ktype
= {
2118 .release
= rdev_free
,
2119 .sysfs_ops
= &rdev_sysfs_ops
,
2120 .default_attrs
= rdev_default_attrs
,
2124 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2126 * mark the device faulty if:
2128 * - the device is nonexistent (zero size)
2129 * - the device has no valid superblock
2131 * a faulty rdev _never_ has rdev->sb set.
2133 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2135 char b
[BDEVNAME_SIZE
];
2140 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2142 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2143 return ERR_PTR(-ENOMEM
);
2146 if ((err
= alloc_disk_sb(rdev
)))
2149 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2153 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2156 rdev
->saved_raid_disk
= -1;
2157 rdev
->raid_disk
= -1;
2159 rdev
->data_offset
= 0;
2160 rdev
->sb_events
= 0;
2161 atomic_set(&rdev
->nr_pending
, 0);
2162 atomic_set(&rdev
->read_errors
, 0);
2163 atomic_set(&rdev
->corrected_errors
, 0);
2165 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2168 "md: %s has zero or unknown size, marking faulty!\n",
2169 bdevname(rdev
->bdev
,b
));
2174 if (super_format
>= 0) {
2175 err
= super_types
[super_format
].
2176 load_super(rdev
, NULL
, super_minor
);
2177 if (err
== -EINVAL
) {
2179 "md: %s does not have a valid v%d.%d "
2180 "superblock, not importing!\n",
2181 bdevname(rdev
->bdev
,b
),
2182 super_format
, super_minor
);
2187 "md: could not read %s's sb, not importing!\n",
2188 bdevname(rdev
->bdev
,b
));
2192 INIT_LIST_HEAD(&rdev
->same_set
);
2197 if (rdev
->sb_page
) {
2203 return ERR_PTR(err
);
2207 * Check a full RAID array for plausibility
2211 static void analyze_sbs(mddev_t
* mddev
)
2214 struct list_head
*tmp
;
2215 mdk_rdev_t
*rdev
, *freshest
;
2216 char b
[BDEVNAME_SIZE
];
2219 rdev_for_each(rdev
, tmp
, mddev
)
2220 switch (super_types
[mddev
->major_version
].
2221 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2229 "md: fatal superblock inconsistency in %s"
2230 " -- removing from array\n",
2231 bdevname(rdev
->bdev
,b
));
2232 kick_rdev_from_array(rdev
);
2236 super_types
[mddev
->major_version
].
2237 validate_super(mddev
, freshest
);
2240 rdev_for_each(rdev
, tmp
, mddev
) {
2241 if (rdev
!= freshest
)
2242 if (super_types
[mddev
->major_version
].
2243 validate_super(mddev
, rdev
)) {
2244 printk(KERN_WARNING
"md: kicking non-fresh %s"
2246 bdevname(rdev
->bdev
,b
));
2247 kick_rdev_from_array(rdev
);
2250 if (mddev
->level
== LEVEL_MULTIPATH
) {
2251 rdev
->desc_nr
= i
++;
2252 rdev
->raid_disk
= rdev
->desc_nr
;
2253 set_bit(In_sync
, &rdev
->flags
);
2254 } else if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2255 rdev
->raid_disk
= -1;
2256 clear_bit(In_sync
, &rdev
->flags
);
2262 if (mddev
->recovery_cp
!= MaxSector
&&
2264 printk(KERN_ERR
"md: %s: raid array is not clean"
2265 " -- starting background reconstruction\n",
2271 safe_delay_show(mddev_t
*mddev
, char *page
)
2273 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2274 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2277 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2285 /* remove a period, and count digits after it */
2286 if (len
>= sizeof(buf
))
2288 strlcpy(buf
, cbuf
, len
);
2290 for (i
=0; i
<len
; i
++) {
2292 if (isdigit(buf
[i
])) {
2297 } else if (buf
[i
] == '.') {
2302 msec
= simple_strtoul(buf
, &e
, 10);
2303 if (e
== buf
|| (*e
&& *e
!= '\n'))
2305 msec
= (msec
* 1000) / scale
;
2307 mddev
->safemode_delay
= 0;
2309 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2310 if (mddev
->safemode_delay
== 0)
2311 mddev
->safemode_delay
= 1;
2315 static struct md_sysfs_entry md_safe_delay
=
2316 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2319 level_show(mddev_t
*mddev
, char *page
)
2321 struct mdk_personality
*p
= mddev
->pers
;
2323 return sprintf(page
, "%s\n", p
->name
);
2324 else if (mddev
->clevel
[0])
2325 return sprintf(page
, "%s\n", mddev
->clevel
);
2326 else if (mddev
->level
!= LEVEL_NONE
)
2327 return sprintf(page
, "%d\n", mddev
->level
);
2333 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2340 if (len
>= sizeof(mddev
->clevel
))
2342 strncpy(mddev
->clevel
, buf
, len
);
2343 if (mddev
->clevel
[len
-1] == '\n')
2345 mddev
->clevel
[len
] = 0;
2346 mddev
->level
= LEVEL_NONE
;
2350 static struct md_sysfs_entry md_level
=
2351 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2355 layout_show(mddev_t
*mddev
, char *page
)
2357 /* just a number, not meaningful for all levels */
2358 if (mddev
->reshape_position
!= MaxSector
&&
2359 mddev
->layout
!= mddev
->new_layout
)
2360 return sprintf(page
, "%d (%d)\n",
2361 mddev
->new_layout
, mddev
->layout
);
2362 return sprintf(page
, "%d\n", mddev
->layout
);
2366 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2369 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2371 if (!*buf
|| (*e
&& *e
!= '\n'))
2376 if (mddev
->reshape_position
!= MaxSector
)
2377 mddev
->new_layout
= n
;
2382 static struct md_sysfs_entry md_layout
=
2383 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2387 raid_disks_show(mddev_t
*mddev
, char *page
)
2389 if (mddev
->raid_disks
== 0)
2391 if (mddev
->reshape_position
!= MaxSector
&&
2392 mddev
->delta_disks
!= 0)
2393 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
2394 mddev
->raid_disks
- mddev
->delta_disks
);
2395 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2398 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2401 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2405 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2407 if (!*buf
|| (*e
&& *e
!= '\n'))
2411 rv
= update_raid_disks(mddev
, n
);
2412 else if (mddev
->reshape_position
!= MaxSector
) {
2413 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
2414 mddev
->delta_disks
= n
- olddisks
;
2415 mddev
->raid_disks
= n
;
2417 mddev
->raid_disks
= n
;
2418 return rv
? rv
: len
;
2420 static struct md_sysfs_entry md_raid_disks
=
2421 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2424 chunk_size_show(mddev_t
*mddev
, char *page
)
2426 if (mddev
->reshape_position
!= MaxSector
&&
2427 mddev
->chunk_size
!= mddev
->new_chunk
)
2428 return sprintf(page
, "%d (%d)\n", mddev
->new_chunk
,
2430 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2434 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2436 /* can only set chunk_size if array is not yet active */
2438 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2440 if (!*buf
|| (*e
&& *e
!= '\n'))
2445 else if (mddev
->reshape_position
!= MaxSector
)
2446 mddev
->new_chunk
= n
;
2448 mddev
->chunk_size
= n
;
2451 static struct md_sysfs_entry md_chunk_size
=
2452 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2455 resync_start_show(mddev_t
*mddev
, char *page
)
2457 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2461 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2464 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2468 if (!*buf
|| (*e
&& *e
!= '\n'))
2471 mddev
->recovery_cp
= n
;
2474 static struct md_sysfs_entry md_resync_start
=
2475 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2478 * The array state can be:
2481 * No devices, no size, no level
2482 * Equivalent to STOP_ARRAY ioctl
2484 * May have some settings, but array is not active
2485 * all IO results in error
2486 * When written, doesn't tear down array, but just stops it
2487 * suspended (not supported yet)
2488 * All IO requests will block. The array can be reconfigured.
2489 * Writing this, if accepted, will block until array is quiessent
2491 * no resync can happen. no superblocks get written.
2492 * write requests fail
2494 * like readonly, but behaves like 'clean' on a write request.
2496 * clean - no pending writes, but otherwise active.
2497 * When written to inactive array, starts without resync
2498 * If a write request arrives then
2499 * if metadata is known, mark 'dirty' and switch to 'active'.
2500 * if not known, block and switch to write-pending
2501 * If written to an active array that has pending writes, then fails.
2503 * fully active: IO and resync can be happening.
2504 * When written to inactive array, starts with resync
2507 * clean, but writes are blocked waiting for 'active' to be written.
2510 * like active, but no writes have been seen for a while (100msec).
2513 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2514 write_pending
, active_idle
, bad_word
};
2515 static char *array_states
[] = {
2516 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2517 "write-pending", "active-idle", NULL
};
2519 static int match_word(const char *word
, char **list
)
2522 for (n
=0; list
[n
]; n
++)
2523 if (cmd_match(word
, list
[n
]))
2529 array_state_show(mddev_t
*mddev
, char *page
)
2531 enum array_state st
= inactive
;
2544 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2546 else if (mddev
->safemode
)
2552 if (list_empty(&mddev
->disks
) &&
2553 mddev
->raid_disks
== 0 &&
2559 return sprintf(page
, "%s\n", array_states
[st
]);
2562 static int do_md_stop(mddev_t
* mddev
, int ro
);
2563 static int do_md_run(mddev_t
* mddev
);
2564 static int restart_array(mddev_t
*mddev
);
2567 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2570 enum array_state st
= match_word(buf
, array_states
);
2575 /* stopping an active array */
2576 if (atomic_read(&mddev
->active
) > 1)
2578 err
= do_md_stop(mddev
, 0);
2581 /* stopping an active array */
2583 if (atomic_read(&mddev
->active
) > 1)
2585 err
= do_md_stop(mddev
, 2);
2587 err
= 0; /* already inactive */
2590 break; /* not supported yet */
2593 err
= do_md_stop(mddev
, 1);
2596 set_disk_ro(mddev
->gendisk
, 1);
2597 err
= do_md_run(mddev
);
2603 err
= do_md_stop(mddev
, 1);
2605 err
= restart_array(mddev
);
2608 set_disk_ro(mddev
->gendisk
, 0);
2612 err
= do_md_run(mddev
);
2617 restart_array(mddev
);
2618 spin_lock_irq(&mddev
->write_lock
);
2619 if (atomic_read(&mddev
->writes_pending
) == 0) {
2620 if (mddev
->in_sync
== 0) {
2622 if (mddev
->safemode
== 1)
2623 mddev
->safemode
= 0;
2624 if (mddev
->persistent
)
2625 set_bit(MD_CHANGE_CLEAN
,
2631 spin_unlock_irq(&mddev
->write_lock
);
2634 mddev
->recovery_cp
= MaxSector
;
2635 err
= do_md_run(mddev
);
2640 restart_array(mddev
);
2641 if (mddev
->external
)
2642 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2643 wake_up(&mddev
->sb_wait
);
2647 set_disk_ro(mddev
->gendisk
, 0);
2648 err
= do_md_run(mddev
);
2653 /* these cannot be set */
2661 static struct md_sysfs_entry md_array_state
=
2662 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
2665 null_show(mddev_t
*mddev
, char *page
)
2671 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2673 /* buf must be %d:%d\n? giving major and minor numbers */
2674 /* The new device is added to the array.
2675 * If the array has a persistent superblock, we read the
2676 * superblock to initialise info and check validity.
2677 * Otherwise, only checking done is that in bind_rdev_to_array,
2678 * which mainly checks size.
2681 int major
= simple_strtoul(buf
, &e
, 10);
2687 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2689 minor
= simple_strtoul(e
+1, &e
, 10);
2690 if (*e
&& *e
!= '\n')
2692 dev
= MKDEV(major
, minor
);
2693 if (major
!= MAJOR(dev
) ||
2694 minor
!= MINOR(dev
))
2698 if (mddev
->persistent
) {
2699 rdev
= md_import_device(dev
, mddev
->major_version
,
2700 mddev
->minor_version
);
2701 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2702 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2703 mdk_rdev_t
, same_set
);
2704 err
= super_types
[mddev
->major_version
]
2705 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2709 } else if (mddev
->external
)
2710 rdev
= md_import_device(dev
, -2, -1);
2712 rdev
= md_import_device(dev
, -1, -1);
2715 return PTR_ERR(rdev
);
2716 err
= bind_rdev_to_array(rdev
, mddev
);
2720 return err
? err
: len
;
2723 static struct md_sysfs_entry md_new_device
=
2724 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
2727 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2730 unsigned long chunk
, end_chunk
;
2734 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2736 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
2737 if (buf
== end
) break;
2738 if (*end
== '-') { /* range */
2740 end_chunk
= simple_strtoul(buf
, &end
, 0);
2741 if (buf
== end
) break;
2743 if (*end
&& !isspace(*end
)) break;
2744 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
2746 while (isspace(*buf
)) buf
++;
2748 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
2753 static struct md_sysfs_entry md_bitmap
=
2754 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
2757 size_show(mddev_t
*mddev
, char *page
)
2759 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2762 static int update_size(mddev_t
*mddev
, unsigned long size
);
2765 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2767 /* If array is inactive, we can reduce the component size, but
2768 * not increase it (except from 0).
2769 * If array is active, we can try an on-line resize
2773 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2774 if (!*buf
|| *buf
== '\n' ||
2779 err
= update_size(mddev
, size
);
2780 md_update_sb(mddev
, 1);
2782 if (mddev
->size
== 0 ||
2788 return err
? err
: len
;
2791 static struct md_sysfs_entry md_size
=
2792 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
2797 * 'none' for arrays with no metadata (good luck...)
2798 * 'external' for arrays with externally managed metadata,
2799 * or N.M for internally known formats
2802 metadata_show(mddev_t
*mddev
, char *page
)
2804 if (mddev
->persistent
)
2805 return sprintf(page
, "%d.%d\n",
2806 mddev
->major_version
, mddev
->minor_version
);
2807 else if (mddev
->external
)
2808 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
2810 return sprintf(page
, "none\n");
2814 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2818 if (!list_empty(&mddev
->disks
))
2821 if (cmd_match(buf
, "none")) {
2822 mddev
->persistent
= 0;
2823 mddev
->external
= 0;
2824 mddev
->major_version
= 0;
2825 mddev
->minor_version
= 90;
2828 if (strncmp(buf
, "external:", 9) == 0) {
2829 size_t namelen
= len
-9;
2830 if (namelen
>= sizeof(mddev
->metadata_type
))
2831 namelen
= sizeof(mddev
->metadata_type
)-1;
2832 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
2833 mddev
->metadata_type
[namelen
] = 0;
2834 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
2835 mddev
->metadata_type
[--namelen
] = 0;
2836 mddev
->persistent
= 0;
2837 mddev
->external
= 1;
2838 mddev
->major_version
= 0;
2839 mddev
->minor_version
= 90;
2842 major
= simple_strtoul(buf
, &e
, 10);
2843 if (e
==buf
|| *e
!= '.')
2846 minor
= simple_strtoul(buf
, &e
, 10);
2847 if (e
==buf
|| (*e
&& *e
!= '\n') )
2849 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
2851 mddev
->major_version
= major
;
2852 mddev
->minor_version
= minor
;
2853 mddev
->persistent
= 1;
2854 mddev
->external
= 0;
2858 static struct md_sysfs_entry md_metadata
=
2859 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
2862 action_show(mddev_t
*mddev
, char *page
)
2864 char *type
= "idle";
2865 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2866 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
2867 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2869 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2870 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2872 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2879 return sprintf(page
, "%s\n", type
);
2883 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2885 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2888 if (cmd_match(page
, "idle")) {
2889 if (mddev
->sync_thread
) {
2890 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2891 md_unregister_thread(mddev
->sync_thread
);
2892 mddev
->sync_thread
= NULL
;
2893 mddev
->recovery
= 0;
2895 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2896 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2898 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2899 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2900 else if (cmd_match(page
, "reshape")) {
2902 if (mddev
->pers
->start_reshape
== NULL
)
2904 err
= mddev
->pers
->start_reshape(mddev
);
2908 if (cmd_match(page
, "check"))
2909 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2910 else if (!cmd_match(page
, "repair"))
2912 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2913 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2915 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2916 md_wakeup_thread(mddev
->thread
);
2921 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2923 return sprintf(page
, "%llu\n",
2924 (unsigned long long) mddev
->resync_mismatches
);
2927 static struct md_sysfs_entry md_scan_mode
=
2928 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2931 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
2934 sync_min_show(mddev_t
*mddev
, char *page
)
2936 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2937 mddev
->sync_speed_min
? "local": "system");
2941 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2945 if (strncmp(buf
, "system", 6)==0) {
2946 mddev
->sync_speed_min
= 0;
2949 min
= simple_strtoul(buf
, &e
, 10);
2950 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2952 mddev
->sync_speed_min
= min
;
2956 static struct md_sysfs_entry md_sync_min
=
2957 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2960 sync_max_show(mddev_t
*mddev
, char *page
)
2962 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2963 mddev
->sync_speed_max
? "local": "system");
2967 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2971 if (strncmp(buf
, "system", 6)==0) {
2972 mddev
->sync_speed_max
= 0;
2975 max
= simple_strtoul(buf
, &e
, 10);
2976 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2978 mddev
->sync_speed_max
= max
;
2982 static struct md_sysfs_entry md_sync_max
=
2983 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2986 degraded_show(mddev_t
*mddev
, char *page
)
2988 return sprintf(page
, "%d\n", mddev
->degraded
);
2990 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
2993 sync_speed_show(mddev_t
*mddev
, char *page
)
2995 unsigned long resync
, dt
, db
;
2996 resync
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
));
2997 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2999 db
= resync
- (mddev
->resync_mark_cnt
);
3000 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
3003 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3006 sync_completed_show(mddev_t
*mddev
, char *page
)
3008 unsigned long max_blocks
, resync
;
3010 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3011 max_blocks
= mddev
->resync_max_sectors
;
3013 max_blocks
= mddev
->size
<< 1;
3015 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
3016 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
3019 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3022 max_sync_show(mddev_t
*mddev
, char *page
)
3024 if (mddev
->resync_max
== MaxSector
)
3025 return sprintf(page
, "max\n");
3027 return sprintf(page
, "%llu\n",
3028 (unsigned long long)mddev
->resync_max
);
3031 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3033 if (strncmp(buf
, "max", 3) == 0)
3034 mddev
->resync_max
= MaxSector
;
3037 unsigned long long max
= simple_strtoull(buf
, &ep
, 10);
3038 if (ep
== buf
|| (*ep
!= 0 && *ep
!= '\n'))
3040 if (max
< mddev
->resync_max
&&
3041 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3044 /* Must be a multiple of chunk_size */
3045 if (mddev
->chunk_size
) {
3046 if (max
& (sector_t
)((mddev
->chunk_size
>>9)-1))
3049 mddev
->resync_max
= max
;
3051 wake_up(&mddev
->recovery_wait
);
3055 static struct md_sysfs_entry md_max_sync
=
3056 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3059 suspend_lo_show(mddev_t
*mddev
, char *page
)
3061 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3065 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3068 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3070 if (mddev
->pers
->quiesce
== NULL
)
3072 if (buf
== e
|| (*e
&& *e
!= '\n'))
3074 if (new >= mddev
->suspend_hi
||
3075 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3076 mddev
->suspend_lo
= new;
3077 mddev
->pers
->quiesce(mddev
, 2);
3082 static struct md_sysfs_entry md_suspend_lo
=
3083 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3087 suspend_hi_show(mddev_t
*mddev
, char *page
)
3089 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3093 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3096 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3098 if (mddev
->pers
->quiesce
== NULL
)
3100 if (buf
== e
|| (*e
&& *e
!= '\n'))
3102 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3103 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3104 mddev
->suspend_hi
= new;
3105 mddev
->pers
->quiesce(mddev
, 1);
3106 mddev
->pers
->quiesce(mddev
, 0);
3111 static struct md_sysfs_entry md_suspend_hi
=
3112 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3115 reshape_position_show(mddev_t
*mddev
, char *page
)
3117 if (mddev
->reshape_position
!= MaxSector
)
3118 return sprintf(page
, "%llu\n",
3119 (unsigned long long)mddev
->reshape_position
);
3120 strcpy(page
, "none\n");
3125 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3128 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3131 if (buf
== e
|| (*e
&& *e
!= '\n'))
3133 mddev
->reshape_position
= new;
3134 mddev
->delta_disks
= 0;
3135 mddev
->new_level
= mddev
->level
;
3136 mddev
->new_layout
= mddev
->layout
;
3137 mddev
->new_chunk
= mddev
->chunk_size
;
3141 static struct md_sysfs_entry md_reshape_position
=
3142 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3143 reshape_position_store
);
3146 static struct attribute
*md_default_attrs
[] = {
3149 &md_raid_disks
.attr
,
3150 &md_chunk_size
.attr
,
3152 &md_resync_start
.attr
,
3154 &md_new_device
.attr
,
3155 &md_safe_delay
.attr
,
3156 &md_array_state
.attr
,
3157 &md_reshape_position
.attr
,
3161 static struct attribute
*md_redundancy_attrs
[] = {
3163 &md_mismatches
.attr
,
3166 &md_sync_speed
.attr
,
3167 &md_sync_completed
.attr
,
3169 &md_suspend_lo
.attr
,
3170 &md_suspend_hi
.attr
,
3175 static struct attribute_group md_redundancy_group
= {
3177 .attrs
= md_redundancy_attrs
,
3182 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3184 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3185 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3190 rv
= mddev_lock(mddev
);
3192 rv
= entry
->show(mddev
, page
);
3193 mddev_unlock(mddev
);
3199 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3200 const char *page
, size_t length
)
3202 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3203 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3208 if (!capable(CAP_SYS_ADMIN
))
3210 rv
= mddev_lock(mddev
);
3212 rv
= entry
->store(mddev
, page
, length
);
3213 mddev_unlock(mddev
);
3218 static void md_free(struct kobject
*ko
)
3220 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
3224 static struct sysfs_ops md_sysfs_ops
= {
3225 .show
= md_attr_show
,
3226 .store
= md_attr_store
,
3228 static struct kobj_type md_ktype
= {
3230 .sysfs_ops
= &md_sysfs_ops
,
3231 .default_attrs
= md_default_attrs
,
3236 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
3238 static DEFINE_MUTEX(disks_mutex
);
3239 mddev_t
*mddev
= mddev_find(dev
);
3240 struct gendisk
*disk
;
3241 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
3242 int shift
= partitioned
? MdpMinorShift
: 0;
3243 int unit
= MINOR(dev
) >> shift
;
3249 mutex_lock(&disks_mutex
);
3250 if (mddev
->gendisk
) {
3251 mutex_unlock(&disks_mutex
);
3255 disk
= alloc_disk(1 << shift
);
3257 mutex_unlock(&disks_mutex
);
3261 disk
->major
= MAJOR(dev
);
3262 disk
->first_minor
= unit
<< shift
;
3264 sprintf(disk
->disk_name
, "md_d%d", unit
);
3266 sprintf(disk
->disk_name
, "md%d", unit
);
3267 disk
->fops
= &md_fops
;
3268 disk
->private_data
= mddev
;
3269 disk
->queue
= mddev
->queue
;
3271 mddev
->gendisk
= disk
;
3272 mutex_unlock(&disks_mutex
);
3273 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
, &disk
->dev
.kobj
,
3276 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
3279 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
3283 static void md_safemode_timeout(unsigned long data
)
3285 mddev_t
*mddev
= (mddev_t
*) data
;
3287 mddev
->safemode
= 1;
3288 md_wakeup_thread(mddev
->thread
);
3291 static int start_dirty_degraded
;
3293 static int do_md_run(mddev_t
* mddev
)
3297 struct list_head
*tmp
;
3299 struct gendisk
*disk
;
3300 struct mdk_personality
*pers
;
3301 char b
[BDEVNAME_SIZE
];
3303 if (list_empty(&mddev
->disks
))
3304 /* cannot run an array with no devices.. */
3311 * Analyze all RAID superblock(s)
3313 if (!mddev
->raid_disks
) {
3314 if (!mddev
->persistent
)
3319 chunk_size
= mddev
->chunk_size
;
3322 if (chunk_size
> MAX_CHUNK_SIZE
) {
3323 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
3324 chunk_size
, MAX_CHUNK_SIZE
);
3328 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3330 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
3331 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
3334 if (chunk_size
< PAGE_SIZE
) {
3335 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
3336 chunk_size
, PAGE_SIZE
);
3340 /* devices must have minimum size of one chunk */
3341 rdev_for_each(rdev
, tmp
, mddev
) {
3342 if (test_bit(Faulty
, &rdev
->flags
))
3344 if (rdev
->size
< chunk_size
/ 1024) {
3346 "md: Dev %s smaller than chunk_size:"
3348 bdevname(rdev
->bdev
,b
),
3349 (unsigned long long)rdev
->size
,
3357 if (mddev
->level
!= LEVEL_NONE
)
3358 request_module("md-level-%d", mddev
->level
);
3359 else if (mddev
->clevel
[0])
3360 request_module("md-%s", mddev
->clevel
);
3364 * Drop all container device buffers, from now on
3365 * the only valid external interface is through the md
3368 rdev_for_each(rdev
, tmp
, mddev
) {
3369 if (test_bit(Faulty
, &rdev
->flags
))
3371 sync_blockdev(rdev
->bdev
);
3372 invalidate_bdev(rdev
->bdev
);
3374 /* perform some consistency tests on the device.
3375 * We don't want the data to overlap the metadata,
3376 * Internal Bitmap issues has handled elsewhere.
3378 if (rdev
->data_offset
< rdev
->sb_offset
) {
3380 rdev
->data_offset
+ mddev
->size
*2
3381 > rdev
->sb_offset
*2) {
3382 printk("md: %s: data overlaps metadata\n",
3387 if (rdev
->sb_offset
*2 + rdev
->sb_size
/512
3388 > rdev
->data_offset
) {
3389 printk("md: %s: metadata overlaps data\n",
3396 md_probe(mddev
->unit
, NULL
, NULL
);
3397 disk
= mddev
->gendisk
;
3401 spin_lock(&pers_lock
);
3402 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3403 if (!pers
|| !try_module_get(pers
->owner
)) {
3404 spin_unlock(&pers_lock
);
3405 if (mddev
->level
!= LEVEL_NONE
)
3406 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3409 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3414 spin_unlock(&pers_lock
);
3415 mddev
->level
= pers
->level
;
3416 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3418 if (mddev
->reshape_position
!= MaxSector
&&
3419 pers
->start_reshape
== NULL
) {
3420 /* This personality cannot handle reshaping... */
3422 module_put(pers
->owner
);
3426 if (pers
->sync_request
) {
3427 /* Warn if this is a potentially silly
3430 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3432 struct list_head
*tmp2
;
3434 rdev_for_each(rdev
, tmp
, mddev
) {
3435 rdev_for_each(rdev2
, tmp2
, mddev
) {
3437 rdev
->bdev
->bd_contains
==
3438 rdev2
->bdev
->bd_contains
) {
3440 "%s: WARNING: %s appears to be"
3441 " on the same physical disk as"
3444 bdevname(rdev
->bdev
,b
),
3445 bdevname(rdev2
->bdev
,b2
));
3452 "True protection against single-disk"
3453 " failure might be compromised.\n");
3456 mddev
->recovery
= 0;
3457 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3458 mddev
->barriers_work
= 1;
3459 mddev
->ok_start_degraded
= start_dirty_degraded
;
3462 mddev
->ro
= 2; /* read-only, but switch on first write */
3464 err
= mddev
->pers
->run(mddev
);
3465 if (!err
&& mddev
->pers
->sync_request
) {
3466 err
= bitmap_create(mddev
);
3468 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3469 mdname(mddev
), err
);
3470 mddev
->pers
->stop(mddev
);
3474 printk(KERN_ERR
"md: pers->run() failed ...\n");
3475 module_put(mddev
->pers
->owner
);
3477 bitmap_destroy(mddev
);
3480 if (mddev
->pers
->sync_request
) {
3481 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3483 "md: cannot register extra attributes for %s\n",
3485 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3488 atomic_set(&mddev
->writes_pending
,0);
3489 mddev
->safemode
= 0;
3490 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3491 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3492 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3495 rdev_for_each(rdev
, tmp
, mddev
)
3496 if (rdev
->raid_disk
>= 0) {
3498 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3499 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
3500 printk("md: cannot register %s for %s\n",
3504 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3507 md_update_sb(mddev
, 0);
3509 set_capacity(disk
, mddev
->array_size
<<1);
3511 /* If we call blk_queue_make_request here, it will
3512 * re-initialise max_sectors etc which may have been
3513 * refined inside -> run. So just set the bits we need to set.
3514 * Most initialisation happended when we called
3515 * blk_queue_make_request(..., md_fail_request)
3518 mddev
->queue
->queuedata
= mddev
;
3519 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3521 /* If there is a partially-recovered drive we need to
3522 * start recovery here. If we leave it to md_check_recovery,
3523 * it will remove the drives and not do the right thing
3525 if (mddev
->degraded
&& !mddev
->sync_thread
) {
3526 struct list_head
*rtmp
;
3528 rdev_for_each(rdev
, rtmp
, mddev
)
3529 if (rdev
->raid_disk
>= 0 &&
3530 !test_bit(In_sync
, &rdev
->flags
) &&
3531 !test_bit(Faulty
, &rdev
->flags
))
3532 /* complete an interrupted recovery */
3534 if (spares
&& mddev
->pers
->sync_request
) {
3535 mddev
->recovery
= 0;
3536 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3537 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3540 if (!mddev
->sync_thread
) {
3541 printk(KERN_ERR
"%s: could not start resync"
3544 /* leave the spares where they are, it shouldn't hurt */
3545 mddev
->recovery
= 0;
3549 md_wakeup_thread(mddev
->thread
);
3550 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
3553 md_new_event(mddev
);
3554 kobject_uevent(&mddev
->gendisk
->dev
.kobj
, KOBJ_CHANGE
);
3558 static int restart_array(mddev_t
*mddev
)
3560 struct gendisk
*disk
= mddev
->gendisk
;
3564 * Complain if it has no devices
3567 if (list_empty(&mddev
->disks
))
3575 mddev
->safemode
= 0;
3577 set_disk_ro(disk
, 0);
3579 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3582 * Kick recovery or resync if necessary
3584 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3585 md_wakeup_thread(mddev
->thread
);
3586 md_wakeup_thread(mddev
->sync_thread
);
3595 /* similar to deny_write_access, but accounts for our holding a reference
3596 * to the file ourselves */
3597 static int deny_bitmap_write_access(struct file
* file
)
3599 struct inode
*inode
= file
->f_mapping
->host
;
3601 spin_lock(&inode
->i_lock
);
3602 if (atomic_read(&inode
->i_writecount
) > 1) {
3603 spin_unlock(&inode
->i_lock
);
3606 atomic_set(&inode
->i_writecount
, -1);
3607 spin_unlock(&inode
->i_lock
);
3612 static void restore_bitmap_write_access(struct file
*file
)
3614 struct inode
*inode
= file
->f_mapping
->host
;
3616 spin_lock(&inode
->i_lock
);
3617 atomic_set(&inode
->i_writecount
, 1);
3618 spin_unlock(&inode
->i_lock
);
3622 * 0 - completely stop and dis-assemble array
3623 * 1 - switch to readonly
3624 * 2 - stop but do not disassemble array
3626 static int do_md_stop(mddev_t
* mddev
, int mode
)
3629 struct gendisk
*disk
= mddev
->gendisk
;
3632 if (atomic_read(&mddev
->active
)>2) {
3633 printk("md: %s still in use.\n",mdname(mddev
));
3637 if (mddev
->sync_thread
) {
3638 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3639 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3640 md_unregister_thread(mddev
->sync_thread
);
3641 mddev
->sync_thread
= NULL
;
3644 del_timer_sync(&mddev
->safemode_timer
);
3646 invalidate_partition(disk
, 0);
3649 case 1: /* readonly */
3655 case 0: /* disassemble */
3657 bitmap_flush(mddev
);
3658 md_super_wait(mddev
);
3660 set_disk_ro(disk
, 0);
3661 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3662 mddev
->pers
->stop(mddev
);
3663 mddev
->queue
->merge_bvec_fn
= NULL
;
3664 mddev
->queue
->unplug_fn
= NULL
;
3665 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
3666 if (mddev
->pers
->sync_request
)
3667 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3669 module_put(mddev
->pers
->owner
);
3672 set_capacity(disk
, 0);
3678 if (!mddev
->in_sync
|| mddev
->flags
) {
3679 /* mark array as shutdown cleanly */
3681 md_update_sb(mddev
, 1);
3684 set_disk_ro(disk
, 1);
3685 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3689 * Free resources if final stop
3693 struct list_head
*tmp
;
3695 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3697 bitmap_destroy(mddev
);
3698 if (mddev
->bitmap_file
) {
3699 restore_bitmap_write_access(mddev
->bitmap_file
);
3700 fput(mddev
->bitmap_file
);
3701 mddev
->bitmap_file
= NULL
;
3703 mddev
->bitmap_offset
= 0;
3705 rdev_for_each(rdev
, tmp
, mddev
)
3706 if (rdev
->raid_disk
>= 0) {
3708 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3709 sysfs_remove_link(&mddev
->kobj
, nm
);
3712 /* make sure all md_delayed_delete calls have finished */
3713 flush_scheduled_work();
3715 export_array(mddev
);
3717 mddev
->array_size
= 0;
3719 mddev
->raid_disks
= 0;
3720 mddev
->recovery_cp
= 0;
3721 mddev
->resync_max
= MaxSector
;
3722 mddev
->reshape_position
= MaxSector
;
3723 mddev
->external
= 0;
3724 mddev
->persistent
= 0;
3725 mddev
->level
= LEVEL_NONE
;
3726 mddev
->clevel
[0] = 0;
3729 mddev
->metadata_type
[0] = 0;
3730 mddev
->chunk_size
= 0;
3731 mddev
->ctime
= mddev
->utime
= 0;
3733 mddev
->max_disks
= 0;
3735 mddev
->delta_disks
= 0;
3736 mddev
->new_level
= LEVEL_NONE
;
3737 mddev
->new_layout
= 0;
3738 mddev
->new_chunk
= 0;
3739 mddev
->curr_resync
= 0;
3740 mddev
->resync_mismatches
= 0;
3741 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
3742 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
3743 mddev
->recovery
= 0;
3746 mddev
->degraded
= 0;
3747 mddev
->barriers_work
= 0;
3748 mddev
->safemode
= 0;
3750 } else if (mddev
->pers
)
3751 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3754 md_new_event(mddev
);
3760 static void autorun_array(mddev_t
*mddev
)
3763 struct list_head
*tmp
;
3766 if (list_empty(&mddev
->disks
))
3769 printk(KERN_INFO
"md: running: ");
3771 rdev_for_each(rdev
, tmp
, mddev
) {
3772 char b
[BDEVNAME_SIZE
];
3773 printk("<%s>", bdevname(rdev
->bdev
,b
));
3777 err
= do_md_run (mddev
);
3779 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3780 do_md_stop (mddev
, 0);
3785 * lets try to run arrays based on all disks that have arrived
3786 * until now. (those are in pending_raid_disks)
3788 * the method: pick the first pending disk, collect all disks with
3789 * the same UUID, remove all from the pending list and put them into
3790 * the 'same_array' list. Then order this list based on superblock
3791 * update time (freshest comes first), kick out 'old' disks and
3792 * compare superblocks. If everything's fine then run it.
3794 * If "unit" is allocated, then bump its reference count
3796 static void autorun_devices(int part
)
3798 struct list_head
*tmp
;
3799 mdk_rdev_t
*rdev0
, *rdev
;
3801 char b
[BDEVNAME_SIZE
];
3803 printk(KERN_INFO
"md: autorun ...\n");
3804 while (!list_empty(&pending_raid_disks
)) {
3807 LIST_HEAD(candidates
);
3808 rdev0
= list_entry(pending_raid_disks
.next
,
3809 mdk_rdev_t
, same_set
);
3811 printk(KERN_INFO
"md: considering %s ...\n",
3812 bdevname(rdev0
->bdev
,b
));
3813 INIT_LIST_HEAD(&candidates
);
3814 rdev_for_each_list(rdev
, tmp
, pending_raid_disks
)
3815 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3816 printk(KERN_INFO
"md: adding %s ...\n",
3817 bdevname(rdev
->bdev
,b
));
3818 list_move(&rdev
->same_set
, &candidates
);
3821 * now we have a set of devices, with all of them having
3822 * mostly sane superblocks. It's time to allocate the
3826 dev
= MKDEV(mdp_major
,
3827 rdev0
->preferred_minor
<< MdpMinorShift
);
3828 unit
= MINOR(dev
) >> MdpMinorShift
;
3830 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3833 if (rdev0
->preferred_minor
!= unit
) {
3834 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3835 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3839 md_probe(dev
, NULL
, NULL
);
3840 mddev
= mddev_find(dev
);
3843 "md: cannot allocate memory for md drive.\n");
3846 if (mddev_lock(mddev
))
3847 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3849 else if (mddev
->raid_disks
|| mddev
->major_version
3850 || !list_empty(&mddev
->disks
)) {
3852 "md: %s already running, cannot run %s\n",
3853 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3854 mddev_unlock(mddev
);
3856 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3857 mddev
->persistent
= 1;
3858 rdev_for_each_list(rdev
, tmp
, candidates
) {
3859 list_del_init(&rdev
->same_set
);
3860 if (bind_rdev_to_array(rdev
, mddev
))
3863 autorun_array(mddev
);
3864 mddev_unlock(mddev
);
3866 /* on success, candidates will be empty, on error
3869 rdev_for_each_list(rdev
, tmp
, candidates
)
3873 printk(KERN_INFO
"md: ... autorun DONE.\n");
3875 #endif /* !MODULE */
3877 static int get_version(void __user
* arg
)
3881 ver
.major
= MD_MAJOR_VERSION
;
3882 ver
.minor
= MD_MINOR_VERSION
;
3883 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3885 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3891 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3893 mdu_array_info_t info
;
3894 int nr
,working
,active
,failed
,spare
;
3896 struct list_head
*tmp
;
3898 nr
=working
=active
=failed
=spare
=0;
3899 rdev_for_each(rdev
, tmp
, mddev
) {
3901 if (test_bit(Faulty
, &rdev
->flags
))
3905 if (test_bit(In_sync
, &rdev
->flags
))
3912 info
.major_version
= mddev
->major_version
;
3913 info
.minor_version
= mddev
->minor_version
;
3914 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3915 info
.ctime
= mddev
->ctime
;
3916 info
.level
= mddev
->level
;
3917 info
.size
= mddev
->size
;
3918 if (info
.size
!= mddev
->size
) /* overflow */
3921 info
.raid_disks
= mddev
->raid_disks
;
3922 info
.md_minor
= mddev
->md_minor
;
3923 info
.not_persistent
= !mddev
->persistent
;
3925 info
.utime
= mddev
->utime
;
3928 info
.state
= (1<<MD_SB_CLEAN
);
3929 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3930 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3931 info
.active_disks
= active
;
3932 info
.working_disks
= working
;
3933 info
.failed_disks
= failed
;
3934 info
.spare_disks
= spare
;
3936 info
.layout
= mddev
->layout
;
3937 info
.chunk_size
= mddev
->chunk_size
;
3939 if (copy_to_user(arg
, &info
, sizeof(info
)))
3945 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3947 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3948 char *ptr
, *buf
= NULL
;
3951 md_allow_write(mddev
);
3953 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3957 /* bitmap disabled, zero the first byte and copy out */
3958 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3959 file
->pathname
[0] = '\0';
3963 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3967 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3971 strcpy(file
->pathname
, ptr
);
3975 if (copy_to_user(arg
, file
, sizeof(*file
)))
3983 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3985 mdu_disk_info_t info
;
3989 if (copy_from_user(&info
, arg
, sizeof(info
)))
3994 rdev
= find_rdev_nr(mddev
, nr
);
3996 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3997 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3998 info
.raid_disk
= rdev
->raid_disk
;
4000 if (test_bit(Faulty
, &rdev
->flags
))
4001 info
.state
|= (1<<MD_DISK_FAULTY
);
4002 else if (test_bit(In_sync
, &rdev
->flags
)) {
4003 info
.state
|= (1<<MD_DISK_ACTIVE
);
4004 info
.state
|= (1<<MD_DISK_SYNC
);
4006 if (test_bit(WriteMostly
, &rdev
->flags
))
4007 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4009 info
.major
= info
.minor
= 0;
4010 info
.raid_disk
= -1;
4011 info
.state
= (1<<MD_DISK_REMOVED
);
4014 if (copy_to_user(arg
, &info
, sizeof(info
)))
4020 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4022 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4024 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4026 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4029 if (!mddev
->raid_disks
) {
4031 /* expecting a device which has a superblock */
4032 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4035 "md: md_import_device returned %ld\n",
4037 return PTR_ERR(rdev
);
4039 if (!list_empty(&mddev
->disks
)) {
4040 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4041 mdk_rdev_t
, same_set
);
4042 int err
= super_types
[mddev
->major_version
]
4043 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4046 "md: %s has different UUID to %s\n",
4047 bdevname(rdev
->bdev
,b
),
4048 bdevname(rdev0
->bdev
,b2
));
4053 err
= bind_rdev_to_array(rdev
, mddev
);
4060 * add_new_disk can be used once the array is assembled
4061 * to add "hot spares". They must already have a superblock
4066 if (!mddev
->pers
->hot_add_disk
) {
4068 "%s: personality does not support diskops!\n",
4072 if (mddev
->persistent
)
4073 rdev
= md_import_device(dev
, mddev
->major_version
,
4074 mddev
->minor_version
);
4076 rdev
= md_import_device(dev
, -1, -1);
4079 "md: md_import_device returned %ld\n",
4081 return PTR_ERR(rdev
);
4083 /* set save_raid_disk if appropriate */
4084 if (!mddev
->persistent
) {
4085 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4086 info
->raid_disk
< mddev
->raid_disks
)
4087 rdev
->raid_disk
= info
->raid_disk
;
4089 rdev
->raid_disk
= -1;
4091 super_types
[mddev
->major_version
].
4092 validate_super(mddev
, rdev
);
4093 rdev
->saved_raid_disk
= rdev
->raid_disk
;
4095 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
4096 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4097 set_bit(WriteMostly
, &rdev
->flags
);
4099 rdev
->raid_disk
= -1;
4100 err
= bind_rdev_to_array(rdev
, mddev
);
4101 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
4102 /* If there is hot_add_disk but no hot_remove_disk
4103 * then added disks for geometry changes,
4104 * and should be added immediately.
4106 super_types
[mddev
->major_version
].
4107 validate_super(mddev
, rdev
);
4108 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
4110 unbind_rdev_from_array(rdev
);
4115 md_update_sb(mddev
, 1);
4116 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4117 md_wakeup_thread(mddev
->thread
);
4121 /* otherwise, add_new_disk is only allowed
4122 * for major_version==0 superblocks
4124 if (mddev
->major_version
!= 0) {
4125 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
4130 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
4132 rdev
= md_import_device (dev
, -1, 0);
4135 "md: error, md_import_device() returned %ld\n",
4137 return PTR_ERR(rdev
);
4139 rdev
->desc_nr
= info
->number
;
4140 if (info
->raid_disk
< mddev
->raid_disks
)
4141 rdev
->raid_disk
= info
->raid_disk
;
4143 rdev
->raid_disk
= -1;
4145 if (rdev
->raid_disk
< mddev
->raid_disks
)
4146 if (info
->state
& (1<<MD_DISK_SYNC
))
4147 set_bit(In_sync
, &rdev
->flags
);
4149 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4150 set_bit(WriteMostly
, &rdev
->flags
);
4152 if (!mddev
->persistent
) {
4153 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
4154 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
4156 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
4157 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
4159 err
= bind_rdev_to_array(rdev
, mddev
);
4169 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
4171 char b
[BDEVNAME_SIZE
];
4177 rdev
= find_rdev(mddev
, dev
);
4181 if (rdev
->raid_disk
>= 0)
4184 kick_rdev_from_array(rdev
);
4185 md_update_sb(mddev
, 1);
4186 md_new_event(mddev
);
4190 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
4191 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4195 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
4197 char b
[BDEVNAME_SIZE
];
4205 if (mddev
->major_version
!= 0) {
4206 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
4207 " version-0 superblocks.\n",
4211 if (!mddev
->pers
->hot_add_disk
) {
4213 "%s: personality does not support diskops!\n",
4218 rdev
= md_import_device (dev
, -1, 0);
4221 "md: error, md_import_device() returned %ld\n",
4226 if (mddev
->persistent
)
4227 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
4230 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
4232 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
4235 if (test_bit(Faulty
, &rdev
->flags
)) {
4237 "md: can not hot-add faulty %s disk to %s!\n",
4238 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4242 clear_bit(In_sync
, &rdev
->flags
);
4244 rdev
->saved_raid_disk
= -1;
4245 err
= bind_rdev_to_array(rdev
, mddev
);
4250 * The rest should better be atomic, we can have disk failures
4251 * noticed in interrupt contexts ...
4254 if (rdev
->desc_nr
== mddev
->max_disks
) {
4255 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
4258 goto abort_unbind_export
;
4261 rdev
->raid_disk
= -1;
4263 md_update_sb(mddev
, 1);
4266 * Kick recovery, maybe this spare has to be added to the
4267 * array immediately.
4269 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4270 md_wakeup_thread(mddev
->thread
);
4271 md_new_event(mddev
);
4274 abort_unbind_export
:
4275 unbind_rdev_from_array(rdev
);
4282 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
4287 if (!mddev
->pers
->quiesce
)
4289 if (mddev
->recovery
|| mddev
->sync_thread
)
4291 /* we should be able to change the bitmap.. */
4297 return -EEXIST
; /* cannot add when bitmap is present */
4298 mddev
->bitmap_file
= fget(fd
);
4300 if (mddev
->bitmap_file
== NULL
) {
4301 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
4306 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
4308 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
4310 fput(mddev
->bitmap_file
);
4311 mddev
->bitmap_file
= NULL
;
4314 mddev
->bitmap_offset
= 0; /* file overrides offset */
4315 } else if (mddev
->bitmap
== NULL
)
4316 return -ENOENT
; /* cannot remove what isn't there */
4319 mddev
->pers
->quiesce(mddev
, 1);
4321 err
= bitmap_create(mddev
);
4322 if (fd
< 0 || err
) {
4323 bitmap_destroy(mddev
);
4324 fd
= -1; /* make sure to put the file */
4326 mddev
->pers
->quiesce(mddev
, 0);
4329 if (mddev
->bitmap_file
) {
4330 restore_bitmap_write_access(mddev
->bitmap_file
);
4331 fput(mddev
->bitmap_file
);
4333 mddev
->bitmap_file
= NULL
;
4340 * set_array_info is used two different ways
4341 * The original usage is when creating a new array.
4342 * In this usage, raid_disks is > 0 and it together with
4343 * level, size, not_persistent,layout,chunksize determine the
4344 * shape of the array.
4345 * This will always create an array with a type-0.90.0 superblock.
4346 * The newer usage is when assembling an array.
4347 * In this case raid_disks will be 0, and the major_version field is
4348 * use to determine which style super-blocks are to be found on the devices.
4349 * The minor and patch _version numbers are also kept incase the
4350 * super_block handler wishes to interpret them.
4352 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
4355 if (info
->raid_disks
== 0) {
4356 /* just setting version number for superblock loading */
4357 if (info
->major_version
< 0 ||
4358 info
->major_version
>= ARRAY_SIZE(super_types
) ||
4359 super_types
[info
->major_version
].name
== NULL
) {
4360 /* maybe try to auto-load a module? */
4362 "md: superblock version %d not known\n",
4363 info
->major_version
);
4366 mddev
->major_version
= info
->major_version
;
4367 mddev
->minor_version
= info
->minor_version
;
4368 mddev
->patch_version
= info
->patch_version
;
4369 mddev
->persistent
= !info
->not_persistent
;
4372 mddev
->major_version
= MD_MAJOR_VERSION
;
4373 mddev
->minor_version
= MD_MINOR_VERSION
;
4374 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
4375 mddev
->ctime
= get_seconds();
4377 mddev
->level
= info
->level
;
4378 mddev
->clevel
[0] = 0;
4379 mddev
->size
= info
->size
;
4380 mddev
->raid_disks
= info
->raid_disks
;
4381 /* don't set md_minor, it is determined by which /dev/md* was
4384 if (info
->state
& (1<<MD_SB_CLEAN
))
4385 mddev
->recovery_cp
= MaxSector
;
4387 mddev
->recovery_cp
= 0;
4388 mddev
->persistent
= ! info
->not_persistent
;
4389 mddev
->external
= 0;
4391 mddev
->layout
= info
->layout
;
4392 mddev
->chunk_size
= info
->chunk_size
;
4394 mddev
->max_disks
= MD_SB_DISKS
;
4396 if (mddev
->persistent
)
4398 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
4400 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4401 mddev
->bitmap_offset
= 0;
4403 mddev
->reshape_position
= MaxSector
;
4406 * Generate a 128 bit UUID
4408 get_random_bytes(mddev
->uuid
, 16);
4410 mddev
->new_level
= mddev
->level
;
4411 mddev
->new_chunk
= mddev
->chunk_size
;
4412 mddev
->new_layout
= mddev
->layout
;
4413 mddev
->delta_disks
= 0;
4418 static int update_size(mddev_t
*mddev
, unsigned long size
)
4422 struct list_head
*tmp
;
4423 int fit
= (size
== 0);
4425 if (mddev
->pers
->resize
== NULL
)
4427 /* The "size" is the amount of each device that is used.
4428 * This can only make sense for arrays with redundancy.
4429 * linear and raid0 always use whatever space is available
4430 * We can only consider changing the size if no resync
4431 * or reconstruction is happening, and if the new size
4432 * is acceptable. It must fit before the sb_offset or,
4433 * if that is <data_offset, it must fit before the
4434 * size of each device.
4435 * If size is zero, we find the largest size that fits.
4437 if (mddev
->sync_thread
)
4439 rdev_for_each(rdev
, tmp
, mddev
) {
4441 avail
= rdev
->size
* 2;
4443 if (fit
&& (size
== 0 || size
> avail
/2))
4445 if (avail
< ((sector_t
)size
<< 1))
4448 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
4450 struct block_device
*bdev
;
4452 bdev
= bdget_disk(mddev
->gendisk
, 0);
4454 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4455 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
4456 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4463 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4466 /* change the number of raid disks */
4467 if (mddev
->pers
->check_reshape
== NULL
)
4469 if (raid_disks
<= 0 ||
4470 raid_disks
>= mddev
->max_disks
)
4472 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4474 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4476 rv
= mddev
->pers
->check_reshape(mddev
);
4482 * update_array_info is used to change the configuration of an
4484 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4485 * fields in the info are checked against the array.
4486 * Any differences that cannot be handled will cause an error.
4487 * Normally, only one change can be managed at a time.
4489 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4495 /* calculate expected state,ignoring low bits */
4496 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4497 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4499 if (mddev
->major_version
!= info
->major_version
||
4500 mddev
->minor_version
!= info
->minor_version
||
4501 /* mddev->patch_version != info->patch_version || */
4502 mddev
->ctime
!= info
->ctime
||
4503 mddev
->level
!= info
->level
||
4504 /* mddev->layout != info->layout || */
4505 !mddev
->persistent
!= info
->not_persistent
||
4506 mddev
->chunk_size
!= info
->chunk_size
||
4507 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4508 ((state
^info
->state
) & 0xfffffe00)
4511 /* Check there is only one change */
4512 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4513 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4514 if (mddev
->layout
!= info
->layout
) cnt
++;
4515 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4516 if (cnt
== 0) return 0;
4517 if (cnt
> 1) return -EINVAL
;
4519 if (mddev
->layout
!= info
->layout
) {
4521 * we don't need to do anything at the md level, the
4522 * personality will take care of it all.
4524 if (mddev
->pers
->reconfig
== NULL
)
4527 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4529 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4530 rv
= update_size(mddev
, info
->size
);
4532 if (mddev
->raid_disks
!= info
->raid_disks
)
4533 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4535 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4536 if (mddev
->pers
->quiesce
== NULL
)
4538 if (mddev
->recovery
|| mddev
->sync_thread
)
4540 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4541 /* add the bitmap */
4544 if (mddev
->default_bitmap_offset
== 0)
4546 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4547 mddev
->pers
->quiesce(mddev
, 1);
4548 rv
= bitmap_create(mddev
);
4550 bitmap_destroy(mddev
);
4551 mddev
->pers
->quiesce(mddev
, 0);
4553 /* remove the bitmap */
4556 if (mddev
->bitmap
->file
)
4558 mddev
->pers
->quiesce(mddev
, 1);
4559 bitmap_destroy(mddev
);
4560 mddev
->pers
->quiesce(mddev
, 0);
4561 mddev
->bitmap_offset
= 0;
4564 md_update_sb(mddev
, 1);
4568 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4572 if (mddev
->pers
== NULL
)
4575 rdev
= find_rdev(mddev
, dev
);
4579 md_error(mddev
, rdev
);
4583 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4585 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4589 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4593 static int md_ioctl(struct inode
*inode
, struct file
*file
,
4594 unsigned int cmd
, unsigned long arg
)
4597 void __user
*argp
= (void __user
*)arg
;
4598 mddev_t
*mddev
= NULL
;
4600 if (!capable(CAP_SYS_ADMIN
))
4604 * Commands dealing with the RAID driver but not any
4610 err
= get_version(argp
);
4613 case PRINT_RAID_DEBUG
:
4621 autostart_arrays(arg
);
4628 * Commands creating/starting a new array:
4631 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4638 err
= mddev_lock(mddev
);
4641 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4648 case SET_ARRAY_INFO
:
4650 mdu_array_info_t info
;
4652 memset(&info
, 0, sizeof(info
));
4653 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4658 err
= update_array_info(mddev
, &info
);
4660 printk(KERN_WARNING
"md: couldn't update"
4661 " array info. %d\n", err
);
4666 if (!list_empty(&mddev
->disks
)) {
4668 "md: array %s already has disks!\n",
4673 if (mddev
->raid_disks
) {
4675 "md: array %s already initialised!\n",
4680 err
= set_array_info(mddev
, &info
);
4682 printk(KERN_WARNING
"md: couldn't set"
4683 " array info. %d\n", err
);
4693 * Commands querying/configuring an existing array:
4695 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4696 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4697 if ((!mddev
->raid_disks
&& !mddev
->external
)
4698 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4699 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
4700 && cmd
!= GET_BITMAP_FILE
) {
4706 * Commands even a read-only array can execute:
4710 case GET_ARRAY_INFO
:
4711 err
= get_array_info(mddev
, argp
);
4714 case GET_BITMAP_FILE
:
4715 err
= get_bitmap_file(mddev
, argp
);
4719 err
= get_disk_info(mddev
, argp
);
4722 case RESTART_ARRAY_RW
:
4723 err
= restart_array(mddev
);
4727 err
= do_md_stop (mddev
, 0);
4731 err
= do_md_stop (mddev
, 1);
4735 * We have a problem here : there is no easy way to give a CHS
4736 * virtual geometry. We currently pretend that we have a 2 heads
4737 * 4 sectors (with a BIG number of cylinders...). This drives
4738 * dosfs just mad... ;-)
4743 * The remaining ioctls are changing the state of the
4744 * superblock, so we do not allow them on read-only arrays.
4745 * However non-MD ioctls (e.g. get-size) will still come through
4746 * here and hit the 'default' below, so only disallow
4747 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4749 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4750 mddev
->ro
&& mddev
->pers
) {
4751 if (mddev
->ro
== 2) {
4753 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4754 md_wakeup_thread(mddev
->thread
);
4766 mdu_disk_info_t info
;
4767 if (copy_from_user(&info
, argp
, sizeof(info
)))
4770 err
= add_new_disk(mddev
, &info
);
4774 case HOT_REMOVE_DISK
:
4775 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4779 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4782 case SET_DISK_FAULTY
:
4783 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4787 err
= do_md_run (mddev
);
4790 case SET_BITMAP_FILE
:
4791 err
= set_bitmap_file(mddev
, (int)arg
);
4801 mddev_unlock(mddev
);
4811 static int md_open(struct inode
*inode
, struct file
*file
)
4814 * Succeed if we can lock the mddev, which confirms that
4815 * it isn't being stopped right now.
4817 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4820 if ((err
= mutex_lock_interruptible_nested(&mddev
->reconfig_mutex
, 1)))
4825 mddev_unlock(mddev
);
4827 check_disk_change(inode
->i_bdev
);
4832 static int md_release(struct inode
*inode
, struct file
* file
)
4834 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4842 static int md_media_changed(struct gendisk
*disk
)
4844 mddev_t
*mddev
= disk
->private_data
;
4846 return mddev
->changed
;
4849 static int md_revalidate(struct gendisk
*disk
)
4851 mddev_t
*mddev
= disk
->private_data
;
4856 static struct block_device_operations md_fops
=
4858 .owner
= THIS_MODULE
,
4860 .release
= md_release
,
4862 .getgeo
= md_getgeo
,
4863 .media_changed
= md_media_changed
,
4864 .revalidate_disk
= md_revalidate
,
4867 static int md_thread(void * arg
)
4869 mdk_thread_t
*thread
= arg
;
4872 * md_thread is a 'system-thread', it's priority should be very
4873 * high. We avoid resource deadlocks individually in each
4874 * raid personality. (RAID5 does preallocation) We also use RR and
4875 * the very same RT priority as kswapd, thus we will never get
4876 * into a priority inversion deadlock.
4878 * we definitely have to have equal or higher priority than
4879 * bdflush, otherwise bdflush will deadlock if there are too
4880 * many dirty RAID5 blocks.
4883 allow_signal(SIGKILL
);
4884 while (!kthread_should_stop()) {
4886 /* We need to wait INTERRUPTIBLE so that
4887 * we don't add to the load-average.
4888 * That means we need to be sure no signals are
4891 if (signal_pending(current
))
4892 flush_signals(current
);
4894 wait_event_interruptible_timeout
4896 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4897 || kthread_should_stop(),
4900 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4902 thread
->run(thread
->mddev
);
4908 void md_wakeup_thread(mdk_thread_t
*thread
)
4911 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4912 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4913 wake_up(&thread
->wqueue
);
4917 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4920 mdk_thread_t
*thread
;
4922 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4926 init_waitqueue_head(&thread
->wqueue
);
4929 thread
->mddev
= mddev
;
4930 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4931 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4932 if (IS_ERR(thread
->tsk
)) {
4939 void md_unregister_thread(mdk_thread_t
*thread
)
4941 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
4943 kthread_stop(thread
->tsk
);
4947 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4954 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4957 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4959 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4960 __builtin_return_address(0),__builtin_return_address(1),
4961 __builtin_return_address(2),__builtin_return_address(3));
4965 if (!mddev
->pers
->error_handler
)
4967 mddev
->pers
->error_handler(mddev
,rdev
);
4968 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4969 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4970 md_wakeup_thread(mddev
->thread
);
4971 md_new_event_inintr(mddev
);
4974 /* seq_file implementation /proc/mdstat */
4976 static void status_unused(struct seq_file
*seq
)
4980 struct list_head
*tmp
;
4982 seq_printf(seq
, "unused devices: ");
4984 rdev_for_each_list(rdev
, tmp
, pending_raid_disks
) {
4985 char b
[BDEVNAME_SIZE
];
4987 seq_printf(seq
, "%s ",
4988 bdevname(rdev
->bdev
,b
));
4991 seq_printf(seq
, "<none>");
4993 seq_printf(seq
, "\n");
4997 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4999 sector_t max_blocks
, resync
, res
;
5000 unsigned long dt
, db
, rt
;
5002 unsigned int per_milli
;
5004 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
5006 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5007 max_blocks
= mddev
->resync_max_sectors
>> 1;
5009 max_blocks
= mddev
->size
;
5012 * Should not happen.
5018 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5019 * in a sector_t, and (max_blocks>>scale) will fit in a
5020 * u32, as those are the requirements for sector_div.
5021 * Thus 'scale' must be at least 10
5024 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5025 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
5028 res
= (resync
>>scale
)*1000;
5029 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
5033 int i
, x
= per_milli
/50, y
= 20-x
;
5034 seq_printf(seq
, "[");
5035 for (i
= 0; i
< x
; i
++)
5036 seq_printf(seq
, "=");
5037 seq_printf(seq
, ">");
5038 for (i
= 0; i
< y
; i
++)
5039 seq_printf(seq
, ".");
5040 seq_printf(seq
, "] ");
5042 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
5043 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
5045 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
5047 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
5048 "resync" : "recovery"))),
5049 per_milli
/10, per_milli
% 10,
5050 (unsigned long long) resync
,
5051 (unsigned long long) max_blocks
);
5054 * We do not want to overflow, so the order of operands and
5055 * the * 100 / 100 trick are important. We do a +1 to be
5056 * safe against division by zero. We only estimate anyway.
5058 * dt: time from mark until now
5059 * db: blocks written from mark until now
5060 * rt: remaining time
5062 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
5064 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
5065 - mddev
->resync_mark_cnt
;
5066 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/2/100+1)))/100;
5068 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
5070 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
5073 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
5075 struct list_head
*tmp
;
5085 spin_lock(&all_mddevs_lock
);
5086 list_for_each(tmp
,&all_mddevs
)
5088 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
5090 spin_unlock(&all_mddevs_lock
);
5093 spin_unlock(&all_mddevs_lock
);
5095 return (void*)2;/* tail */
5099 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
5101 struct list_head
*tmp
;
5102 mddev_t
*next_mddev
, *mddev
= v
;
5108 spin_lock(&all_mddevs_lock
);
5110 tmp
= all_mddevs
.next
;
5112 tmp
= mddev
->all_mddevs
.next
;
5113 if (tmp
!= &all_mddevs
)
5114 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
5116 next_mddev
= (void*)2;
5119 spin_unlock(&all_mddevs_lock
);
5127 static void md_seq_stop(struct seq_file
*seq
, void *v
)
5131 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
5135 struct mdstat_info
{
5139 static int md_seq_show(struct seq_file
*seq
, void *v
)
5143 struct list_head
*tmp2
;
5145 struct mdstat_info
*mi
= seq
->private;
5146 struct bitmap
*bitmap
;
5148 if (v
== (void*)1) {
5149 struct mdk_personality
*pers
;
5150 seq_printf(seq
, "Personalities : ");
5151 spin_lock(&pers_lock
);
5152 list_for_each_entry(pers
, &pers_list
, list
)
5153 seq_printf(seq
, "[%s] ", pers
->name
);
5155 spin_unlock(&pers_lock
);
5156 seq_printf(seq
, "\n");
5157 mi
->event
= atomic_read(&md_event_count
);
5160 if (v
== (void*)2) {
5165 if (mddev_lock(mddev
) < 0)
5168 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
5169 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
5170 mddev
->pers
? "" : "in");
5173 seq_printf(seq
, " (read-only)");
5175 seq_printf(seq
, " (auto-read-only)");
5176 seq_printf(seq
, " %s", mddev
->pers
->name
);
5180 rdev_for_each(rdev
, tmp2
, mddev
) {
5181 char b
[BDEVNAME_SIZE
];
5182 seq_printf(seq
, " %s[%d]",
5183 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
5184 if (test_bit(WriteMostly
, &rdev
->flags
))
5185 seq_printf(seq
, "(W)");
5186 if (test_bit(Faulty
, &rdev
->flags
)) {
5187 seq_printf(seq
, "(F)");
5189 } else if (rdev
->raid_disk
< 0)
5190 seq_printf(seq
, "(S)"); /* spare */
5194 if (!list_empty(&mddev
->disks
)) {
5196 seq_printf(seq
, "\n %llu blocks",
5197 (unsigned long long)mddev
->array_size
);
5199 seq_printf(seq
, "\n %llu blocks",
5200 (unsigned long long)size
);
5202 if (mddev
->persistent
) {
5203 if (mddev
->major_version
!= 0 ||
5204 mddev
->minor_version
!= 90) {
5205 seq_printf(seq
," super %d.%d",
5206 mddev
->major_version
,
5207 mddev
->minor_version
);
5209 } else if (mddev
->external
)
5210 seq_printf(seq
, " super external:%s",
5211 mddev
->metadata_type
);
5213 seq_printf(seq
, " super non-persistent");
5216 mddev
->pers
->status (seq
, mddev
);
5217 seq_printf(seq
, "\n ");
5218 if (mddev
->pers
->sync_request
) {
5219 if (mddev
->curr_resync
> 2) {
5220 status_resync (seq
, mddev
);
5221 seq_printf(seq
, "\n ");
5222 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
5223 seq_printf(seq
, "\tresync=DELAYED\n ");
5224 else if (mddev
->recovery_cp
< MaxSector
)
5225 seq_printf(seq
, "\tresync=PENDING\n ");
5228 seq_printf(seq
, "\n ");
5230 if ((bitmap
= mddev
->bitmap
)) {
5231 unsigned long chunk_kb
;
5232 unsigned long flags
;
5233 spin_lock_irqsave(&bitmap
->lock
, flags
);
5234 chunk_kb
= bitmap
->chunksize
>> 10;
5235 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
5237 bitmap
->pages
- bitmap
->missing_pages
,
5239 (bitmap
->pages
- bitmap
->missing_pages
)
5240 << (PAGE_SHIFT
- 10),
5241 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
5242 chunk_kb
? "KB" : "B");
5244 seq_printf(seq
, ", file: ");
5245 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
5248 seq_printf(seq
, "\n");
5249 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
5252 seq_printf(seq
, "\n");
5254 mddev_unlock(mddev
);
5259 static struct seq_operations md_seq_ops
= {
5260 .start
= md_seq_start
,
5261 .next
= md_seq_next
,
5262 .stop
= md_seq_stop
,
5263 .show
= md_seq_show
,
5266 static int md_seq_open(struct inode
*inode
, struct file
*file
)
5269 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
5273 error
= seq_open(file
, &md_seq_ops
);
5277 struct seq_file
*p
= file
->private_data
;
5279 mi
->event
= atomic_read(&md_event_count
);
5284 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
5286 struct seq_file
*m
= filp
->private_data
;
5287 struct mdstat_info
*mi
= m
->private;
5290 poll_wait(filp
, &md_event_waiters
, wait
);
5292 /* always allow read */
5293 mask
= POLLIN
| POLLRDNORM
;
5295 if (mi
->event
!= atomic_read(&md_event_count
))
5296 mask
|= POLLERR
| POLLPRI
;
5300 static const struct file_operations md_seq_fops
= {
5301 .owner
= THIS_MODULE
,
5302 .open
= md_seq_open
,
5304 .llseek
= seq_lseek
,
5305 .release
= seq_release_private
,
5306 .poll
= mdstat_poll
,
5309 int register_md_personality(struct mdk_personality
*p
)
5311 spin_lock(&pers_lock
);
5312 list_add_tail(&p
->list
, &pers_list
);
5313 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
5314 spin_unlock(&pers_lock
);
5318 int unregister_md_personality(struct mdk_personality
*p
)
5320 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
5321 spin_lock(&pers_lock
);
5322 list_del_init(&p
->list
);
5323 spin_unlock(&pers_lock
);
5327 static int is_mddev_idle(mddev_t
*mddev
)
5330 struct list_head
*tmp
;
5335 rdev_for_each(rdev
, tmp
, mddev
) {
5336 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
5337 curr_events
= disk_stat_read(disk
, sectors
[0]) +
5338 disk_stat_read(disk
, sectors
[1]) -
5339 atomic_read(&disk
->sync_io
);
5340 /* sync IO will cause sync_io to increase before the disk_stats
5341 * as sync_io is counted when a request starts, and
5342 * disk_stats is counted when it completes.
5343 * So resync activity will cause curr_events to be smaller than
5344 * when there was no such activity.
5345 * non-sync IO will cause disk_stat to increase without
5346 * increasing sync_io so curr_events will (eventually)
5347 * be larger than it was before. Once it becomes
5348 * substantially larger, the test below will cause
5349 * the array to appear non-idle, and resync will slow
5351 * If there is a lot of outstanding resync activity when
5352 * we set last_event to curr_events, then all that activity
5353 * completing might cause the array to appear non-idle
5354 * and resync will be slowed down even though there might
5355 * not have been non-resync activity. This will only
5356 * happen once though. 'last_events' will soon reflect
5357 * the state where there is little or no outstanding
5358 * resync requests, and further resync activity will
5359 * always make curr_events less than last_events.
5362 if (curr_events
- rdev
->last_events
> 4096) {
5363 rdev
->last_events
= curr_events
;
5370 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
5372 /* another "blocks" (512byte) blocks have been synced */
5373 atomic_sub(blocks
, &mddev
->recovery_active
);
5374 wake_up(&mddev
->recovery_wait
);
5376 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5377 md_wakeup_thread(mddev
->thread
);
5378 // stop recovery, signal do_sync ....
5383 /* md_write_start(mddev, bi)
5384 * If we need to update some array metadata (e.g. 'active' flag
5385 * in superblock) before writing, schedule a superblock update
5386 * and wait for it to complete.
5388 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
5390 if (bio_data_dir(bi
) != WRITE
)
5393 BUG_ON(mddev
->ro
== 1);
5394 if (mddev
->ro
== 2) {
5395 /* need to switch to read/write */
5397 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5398 md_wakeup_thread(mddev
->thread
);
5399 md_wakeup_thread(mddev
->sync_thread
);
5401 atomic_inc(&mddev
->writes_pending
);
5402 if (mddev
->safemode
== 1)
5403 mddev
->safemode
= 0;
5404 if (mddev
->in_sync
) {
5405 spin_lock_irq(&mddev
->write_lock
);
5406 if (mddev
->in_sync
) {
5408 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5409 md_wakeup_thread(mddev
->thread
);
5411 spin_unlock_irq(&mddev
->write_lock
);
5413 wait_event(mddev
->sb_wait
, mddev
->flags
==0);
5416 void md_write_end(mddev_t
*mddev
)
5418 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5419 if (mddev
->safemode
== 2)
5420 md_wakeup_thread(mddev
->thread
);
5421 else if (mddev
->safemode_delay
)
5422 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5426 /* md_allow_write(mddev)
5427 * Calling this ensures that the array is marked 'active' so that writes
5428 * may proceed without blocking. It is important to call this before
5429 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5430 * Must be called with mddev_lock held.
5432 void md_allow_write(mddev_t
*mddev
)
5439 spin_lock_irq(&mddev
->write_lock
);
5440 if (mddev
->in_sync
) {
5442 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5443 if (mddev
->safemode_delay
&&
5444 mddev
->safemode
== 0)
5445 mddev
->safemode
= 1;
5446 spin_unlock_irq(&mddev
->write_lock
);
5447 md_update_sb(mddev
, 0);
5449 spin_unlock_irq(&mddev
->write_lock
);
5451 EXPORT_SYMBOL_GPL(md_allow_write
);
5453 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
5455 #define SYNC_MARKS 10
5456 #define SYNC_MARK_STEP (3*HZ)
5457 void md_do_sync(mddev_t
*mddev
)
5460 unsigned int currspeed
= 0,
5462 sector_t max_sectors
,j
, io_sectors
;
5463 unsigned long mark
[SYNC_MARKS
];
5464 sector_t mark_cnt
[SYNC_MARKS
];
5466 struct list_head
*tmp
;
5467 sector_t last_check
;
5469 struct list_head
*rtmp
;
5473 /* just incase thread restarts... */
5474 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5476 if (mddev
->ro
) /* never try to sync a read-only array */
5479 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5480 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
5481 desc
= "data-check";
5482 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5483 desc
= "requested-resync";
5486 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5491 /* we overload curr_resync somewhat here.
5492 * 0 == not engaged in resync at all
5493 * 2 == checking that there is no conflict with another sync
5494 * 1 == like 2, but have yielded to allow conflicting resync to
5496 * other == active in resync - this many blocks
5498 * Before starting a resync we must have set curr_resync to
5499 * 2, and then checked that every "conflicting" array has curr_resync
5500 * less than ours. When we find one that is the same or higher
5501 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5502 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5503 * This will mean we have to start checking from the beginning again.
5508 mddev
->curr_resync
= 2;
5511 if (kthread_should_stop()) {
5512 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5515 for_each_mddev(mddev2
, tmp
) {
5516 if (mddev2
== mddev
)
5518 if (mddev2
->curr_resync
&&
5519 match_mddev_units(mddev
,mddev2
)) {
5521 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5522 /* arbitrarily yield */
5523 mddev
->curr_resync
= 1;
5524 wake_up(&resync_wait
);
5526 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5527 /* no need to wait here, we can wait the next
5528 * time 'round when curr_resync == 2
5531 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
5532 if (!kthread_should_stop() &&
5533 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5534 printk(KERN_INFO
"md: delaying %s of %s"
5535 " until %s has finished (they"
5536 " share one or more physical units)\n",
5537 desc
, mdname(mddev
), mdname(mddev2
));
5540 finish_wait(&resync_wait
, &wq
);
5543 finish_wait(&resync_wait
, &wq
);
5546 } while (mddev
->curr_resync
< 2);
5549 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5550 /* resync follows the size requested by the personality,
5551 * which defaults to physical size, but can be virtual size
5553 max_sectors
= mddev
->resync_max_sectors
;
5554 mddev
->resync_mismatches
= 0;
5555 /* we don't use the checkpoint if there's a bitmap */
5556 if (!mddev
->bitmap
&&
5557 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5558 j
= mddev
->recovery_cp
;
5559 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5560 max_sectors
= mddev
->size
<< 1;
5562 /* recovery follows the physical size of devices */
5563 max_sectors
= mddev
->size
<< 1;
5565 rdev_for_each(rdev
, rtmp
, mddev
)
5566 if (rdev
->raid_disk
>= 0 &&
5567 !test_bit(Faulty
, &rdev
->flags
) &&
5568 !test_bit(In_sync
, &rdev
->flags
) &&
5569 rdev
->recovery_offset
< j
)
5570 j
= rdev
->recovery_offset
;
5573 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
5574 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
5575 " %d KB/sec/disk.\n", speed_min(mddev
));
5576 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5577 "(but not more than %d KB/sec) for %s.\n",
5578 speed_max(mddev
), desc
);
5580 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5583 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5585 mark_cnt
[m
] = io_sectors
;
5588 mddev
->resync_mark
= mark
[last_mark
];
5589 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5592 * Tune reconstruction:
5594 window
= 32*(PAGE_SIZE
/512);
5595 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5596 window
/2,(unsigned long long) max_sectors
/2);
5598 atomic_set(&mddev
->recovery_active
, 0);
5599 init_waitqueue_head(&mddev
->recovery_wait
);
5604 "md: resuming %s of %s from checkpoint.\n",
5605 desc
, mdname(mddev
));
5606 mddev
->curr_resync
= j
;
5609 while (j
< max_sectors
) {
5613 if (j
>= mddev
->resync_max
) {
5614 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
5615 wait_event(mddev
->recovery_wait
,
5616 mddev
->resync_max
> j
5617 || kthread_should_stop());
5619 if (kthread_should_stop())
5621 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5622 currspeed
< speed_min(mddev
));
5624 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5628 if (!skipped
) { /* actual IO requested */
5629 io_sectors
+= sectors
;
5630 atomic_add(sectors
, &mddev
->recovery_active
);
5634 if (j
>1) mddev
->curr_resync
= j
;
5635 mddev
->curr_mark_cnt
= io_sectors
;
5636 if (last_check
== 0)
5637 /* this is the earliers that rebuilt will be
5638 * visible in /proc/mdstat
5640 md_new_event(mddev
);
5642 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5645 last_check
= io_sectors
;
5647 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
5648 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
5652 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5654 int next
= (last_mark
+1) % SYNC_MARKS
;
5656 mddev
->resync_mark
= mark
[next
];
5657 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5658 mark
[next
] = jiffies
;
5659 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5664 if (kthread_should_stop())
5669 * this loop exits only if either when we are slower than
5670 * the 'hard' speed limit, or the system was IO-idle for
5672 * the system might be non-idle CPU-wise, but we only care
5673 * about not overloading the IO subsystem. (things like an
5674 * e2fsck being done on the RAID array should execute fast)
5676 blk_unplug(mddev
->queue
);
5679 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5680 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5682 if (currspeed
> speed_min(mddev
)) {
5683 if ((currspeed
> speed_max(mddev
)) ||
5684 !is_mddev_idle(mddev
)) {
5690 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
5692 * this also signals 'finished resyncing' to md_stop
5695 blk_unplug(mddev
->queue
);
5697 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5699 /* tell personality that we are finished */
5700 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5702 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5703 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5704 mddev
->curr_resync
> 2) {
5705 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5706 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5707 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5709 "md: checkpointing %s of %s.\n",
5710 desc
, mdname(mddev
));
5711 mddev
->recovery_cp
= mddev
->curr_resync
;
5714 mddev
->recovery_cp
= MaxSector
;
5716 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5717 mddev
->curr_resync
= MaxSector
;
5718 rdev_for_each(rdev
, rtmp
, mddev
)
5719 if (rdev
->raid_disk
>= 0 &&
5720 !test_bit(Faulty
, &rdev
->flags
) &&
5721 !test_bit(In_sync
, &rdev
->flags
) &&
5722 rdev
->recovery_offset
< mddev
->curr_resync
)
5723 rdev
->recovery_offset
= mddev
->curr_resync
;
5726 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5729 mddev
->curr_resync
= 0;
5730 mddev
->resync_max
= MaxSector
;
5731 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
5732 wake_up(&resync_wait
);
5733 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5734 md_wakeup_thread(mddev
->thread
);
5739 * got a signal, exit.
5742 "md: md_do_sync() got signal ... exiting\n");
5743 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5747 EXPORT_SYMBOL_GPL(md_do_sync
);
5750 static int remove_and_add_spares(mddev_t
*mddev
)
5753 struct list_head
*rtmp
;
5756 rdev_for_each(rdev
, rtmp
, mddev
)
5757 if (rdev
->raid_disk
>= 0 &&
5759 (test_bit(Faulty
, &rdev
->flags
) ||
5760 ! test_bit(In_sync
, &rdev
->flags
)) &&
5761 atomic_read(&rdev
->nr_pending
)==0) {
5762 if (mddev
->pers
->hot_remove_disk(
5763 mddev
, rdev
->raid_disk
)==0) {
5765 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5766 sysfs_remove_link(&mddev
->kobj
, nm
);
5767 rdev
->raid_disk
= -1;
5771 if (mddev
->degraded
) {
5772 rdev_for_each(rdev
, rtmp
, mddev
)
5773 if (rdev
->raid_disk
< 0
5774 && !test_bit(Faulty
, &rdev
->flags
)) {
5775 rdev
->recovery_offset
= 0;
5776 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5778 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5779 if (sysfs_create_link(&mddev
->kobj
,
5782 "md: cannot register "
5786 md_new_event(mddev
);
5794 * This routine is regularly called by all per-raid-array threads to
5795 * deal with generic issues like resync and super-block update.
5796 * Raid personalities that don't have a thread (linear/raid0) do not
5797 * need this as they never do any recovery or update the superblock.
5799 * It does not do any resync itself, but rather "forks" off other threads
5800 * to do that as needed.
5801 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5802 * "->recovery" and create a thread at ->sync_thread.
5803 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5804 * and wakeups up this thread which will reap the thread and finish up.
5805 * This thread also removes any faulty devices (with nr_pending == 0).
5807 * The overall approach is:
5808 * 1/ if the superblock needs updating, update it.
5809 * 2/ If a recovery thread is running, don't do anything else.
5810 * 3/ If recovery has finished, clean up, possibly marking spares active.
5811 * 4/ If there are any faulty devices, remove them.
5812 * 5/ If array is degraded, try to add spares devices
5813 * 6/ If array has spares or is not in-sync, start a resync thread.
5815 void md_check_recovery(mddev_t
*mddev
)
5818 struct list_head
*rtmp
;
5822 bitmap_daemon_work(mddev
->bitmap
);
5827 if (signal_pending(current
)) {
5828 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
5829 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5831 mddev
->safemode
= 2;
5833 flush_signals(current
);
5837 (mddev
->flags
&& !mddev
->external
) ||
5838 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5839 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5840 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
5841 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5842 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5846 if (mddev_trylock(mddev
)) {
5849 if (!mddev
->external
) {
5850 spin_lock_irq(&mddev
->write_lock
);
5851 if (mddev
->safemode
&&
5852 !atomic_read(&mddev
->writes_pending
) &&
5854 mddev
->recovery_cp
== MaxSector
) {
5856 if (mddev
->persistent
)
5857 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
5859 if (mddev
->safemode
== 1)
5860 mddev
->safemode
= 0;
5861 spin_unlock_irq(&mddev
->write_lock
);
5865 md_update_sb(mddev
, 0);
5868 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5869 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5870 /* resync/recovery still happening */
5871 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5874 if (mddev
->sync_thread
) {
5875 /* resync has finished, collect result */
5876 md_unregister_thread(mddev
->sync_thread
);
5877 mddev
->sync_thread
= NULL
;
5878 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5879 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5881 /* activate any spares */
5882 mddev
->pers
->spare_active(mddev
);
5884 md_update_sb(mddev
, 1);
5886 /* if array is no-longer degraded, then any saved_raid_disk
5887 * information must be scrapped
5889 if (!mddev
->degraded
)
5890 rdev_for_each(rdev
, rtmp
, mddev
)
5891 rdev
->saved_raid_disk
= -1;
5893 mddev
->recovery
= 0;
5894 /* flag recovery needed just to double check */
5895 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5896 md_new_event(mddev
);
5899 /* Clear some bits that don't mean anything, but
5902 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5903 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5904 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5905 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5907 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5909 /* no recovery is running.
5910 * remove any failed drives, then
5911 * add spares if possible.
5912 * Spare are also removed and re-added, to allow
5913 * the personality to fail the re-add.
5916 if (mddev
->reshape_position
!= MaxSector
) {
5917 if (mddev
->pers
->check_reshape(mddev
) != 0)
5918 /* Cannot proceed */
5920 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
5921 } else if ((spares
= remove_and_add_spares(mddev
))) {
5922 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5923 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5924 } else if (mddev
->recovery_cp
< MaxSector
) {
5925 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5926 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5927 /* nothing to be done ... */
5930 if (mddev
->pers
->sync_request
) {
5931 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5932 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5933 /* We are adding a device or devices to an array
5934 * which has the bitmap stored on all devices.
5935 * So make sure all bitmap pages get written
5937 bitmap_write_all(mddev
->bitmap
);
5939 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5942 if (!mddev
->sync_thread
) {
5943 printk(KERN_ERR
"%s: could not start resync"
5946 /* leave the spares where they are, it shouldn't hurt */
5947 mddev
->recovery
= 0;
5949 md_wakeup_thread(mddev
->sync_thread
);
5950 md_new_event(mddev
);
5953 mddev_unlock(mddev
);
5957 static int md_notify_reboot(struct notifier_block
*this,
5958 unsigned long code
, void *x
)
5960 struct list_head
*tmp
;
5963 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5965 printk(KERN_INFO
"md: stopping all md devices.\n");
5967 for_each_mddev(mddev
, tmp
)
5968 if (mddev_trylock(mddev
)) {
5969 do_md_stop (mddev
, 1);
5970 mddev_unlock(mddev
);
5973 * certain more exotic SCSI devices are known to be
5974 * volatile wrt too early system reboots. While the
5975 * right place to handle this issue is the given
5976 * driver, we do want to have a safe RAID driver ...
5983 static struct notifier_block md_notifier
= {
5984 .notifier_call
= md_notify_reboot
,
5986 .priority
= INT_MAX
, /* before any real devices */
5989 static void md_geninit(void)
5991 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5993 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
5996 static int __init
md_init(void)
5998 if (register_blkdev(MAJOR_NR
, "md"))
6000 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
6001 unregister_blkdev(MAJOR_NR
, "md");
6004 blk_register_region(MKDEV(MAJOR_NR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6005 md_probe
, NULL
, NULL
);
6006 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6007 md_probe
, NULL
, NULL
);
6009 register_reboot_notifier(&md_notifier
);
6010 raid_table_header
= register_sysctl_table(raid_root_table
);
6020 * Searches all registered partitions for autorun RAID arrays
6024 static LIST_HEAD(all_detected_devices
);
6025 struct detected_devices_node
{
6026 struct list_head list
;
6030 void md_autodetect_dev(dev_t dev
)
6032 struct detected_devices_node
*node_detected_dev
;
6034 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
6035 if (node_detected_dev
) {
6036 node_detected_dev
->dev
= dev
;
6037 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
6039 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
6040 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
6045 static void autostart_arrays(int part
)
6048 struct detected_devices_node
*node_detected_dev
;
6050 int i_scanned
, i_passed
;
6055 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
6057 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
6059 node_detected_dev
= list_entry(all_detected_devices
.next
,
6060 struct detected_devices_node
, list
);
6061 list_del(&node_detected_dev
->list
);
6062 dev
= node_detected_dev
->dev
;
6063 kfree(node_detected_dev
);
6064 rdev
= md_import_device(dev
,0, 90);
6068 if (test_bit(Faulty
, &rdev
->flags
)) {
6072 set_bit(AutoDetected
, &rdev
->flags
);
6073 list_add(&rdev
->same_set
, &pending_raid_disks
);
6077 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
6078 i_scanned
, i_passed
);
6080 autorun_devices(part
);
6083 #endif /* !MODULE */
6085 static __exit
void md_exit(void)
6088 struct list_head
*tmp
;
6090 blk_unregister_region(MKDEV(MAJOR_NR
,0), 1U << MINORBITS
);
6091 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
6093 unregister_blkdev(MAJOR_NR
,"md");
6094 unregister_blkdev(mdp_major
, "mdp");
6095 unregister_reboot_notifier(&md_notifier
);
6096 unregister_sysctl_table(raid_table_header
);
6097 remove_proc_entry("mdstat", NULL
);
6098 for_each_mddev(mddev
, tmp
) {
6099 struct gendisk
*disk
= mddev
->gendisk
;
6102 export_array(mddev
);
6105 mddev
->gendisk
= NULL
;
6110 subsys_initcall(md_init
);
6111 module_exit(md_exit
)
6113 static int get_ro(char *buffer
, struct kernel_param
*kp
)
6115 return sprintf(buffer
, "%d", start_readonly
);
6117 static int set_ro(const char *val
, struct kernel_param
*kp
)
6120 int num
= simple_strtoul(val
, &e
, 10);
6121 if (*val
&& (*e
== '\0' || *e
== '\n')) {
6122 start_readonly
= num
;
6128 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
6129 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
6132 EXPORT_SYMBOL(register_md_personality
);
6133 EXPORT_SYMBOL(unregister_md_personality
);
6134 EXPORT_SYMBOL(md_error
);
6135 EXPORT_SYMBOL(md_done_sync
);
6136 EXPORT_SYMBOL(md_write_start
);
6137 EXPORT_SYMBOL(md_write_end
);
6138 EXPORT_SYMBOL(md_register_thread
);
6139 EXPORT_SYMBOL(md_unregister_thread
);
6140 EXPORT_SYMBOL(md_wakeup_thread
);
6141 EXPORT_SYMBOL(md_check_recovery
);
6142 MODULE_LICENSE("GPL");
6144 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);