2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/suspend.h>
44 #include <linux/poll.h>
45 #include <linux/mutex.h>
46 #include <linux/ctype.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),
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),
123 .proc_handler
= &proc_dointvec
,
128 static ctl_table raid_dir_table
[] = {
130 .ctl_name
= DEV_RAID
,
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 ITERATE_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 (request_queue_t
*q
, struct bio
*bio
)
216 bio_io_error(bio
, bio
->bi_size
);
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_unregister(&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
);
278 new->queue
= blk_alloc_queue(GFP_KERNEL
);
283 set_bit(QUEUE_FLAG_CLUSTER
, &new->queue
->queue_flags
);
285 blk_queue_make_request(new->queue
, md_fail_request
);
290 static inline int mddev_lock(mddev_t
* mddev
)
292 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
295 static inline int mddev_trylock(mddev_t
* mddev
)
297 return mutex_trylock(&mddev
->reconfig_mutex
);
300 static inline void mddev_unlock(mddev_t
* mddev
)
302 mutex_unlock(&mddev
->reconfig_mutex
);
304 md_wakeup_thread(mddev
->thread
);
307 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
310 struct list_head
*tmp
;
312 ITERATE_RDEV(mddev
,rdev
,tmp
) {
313 if (rdev
->desc_nr
== nr
)
319 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
321 struct list_head
*tmp
;
324 ITERATE_RDEV(mddev
,rdev
,tmp
) {
325 if (rdev
->bdev
->bd_dev
== dev
)
331 static struct mdk_personality
*find_pers(int level
, char *clevel
)
333 struct mdk_personality
*pers
;
334 list_for_each_entry(pers
, &pers_list
, list
) {
335 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
337 if (strcmp(pers
->name
, clevel
)==0)
343 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
345 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
346 return MD_NEW_SIZE_BLOCKS(size
);
349 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
353 size
= rdev
->sb_offset
;
356 size
&= ~((sector_t
)chunk_size
/1024 - 1);
360 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
365 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
366 if (!rdev
->sb_page
) {
367 printk(KERN_ALERT
"md: out of memory.\n");
374 static void free_disk_sb(mdk_rdev_t
* rdev
)
377 put_page(rdev
->sb_page
);
379 rdev
->sb_page
= NULL
;
386 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
388 mdk_rdev_t
*rdev
= bio
->bi_private
;
389 mddev_t
*mddev
= rdev
->mddev
;
393 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
394 md_error(mddev
, rdev
);
396 if (atomic_dec_and_test(&mddev
->pending_writes
))
397 wake_up(&mddev
->sb_wait
);
402 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
404 struct bio
*bio2
= bio
->bi_private
;
405 mdk_rdev_t
*rdev
= bio2
->bi_private
;
406 mddev_t
*mddev
= rdev
->mddev
;
410 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
411 error
== -EOPNOTSUPP
) {
413 /* barriers don't appear to be supported :-( */
414 set_bit(BarriersNotsupp
, &rdev
->flags
);
415 mddev
->barriers_work
= 0;
416 spin_lock_irqsave(&mddev
->write_lock
, flags
);
417 bio2
->bi_next
= mddev
->biolist
;
418 mddev
->biolist
= bio2
;
419 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
420 wake_up(&mddev
->sb_wait
);
425 bio
->bi_private
= rdev
;
426 return super_written(bio
, bytes_done
, error
);
429 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
430 sector_t sector
, int size
, struct page
*page
)
432 /* write first size bytes of page to sector of rdev
433 * Increment mddev->pending_writes before returning
434 * and decrement it on completion, waking up sb_wait
435 * if zero is reached.
436 * If an error occurred, call md_error
438 * As we might need to resubmit the request if BIO_RW_BARRIER
439 * causes ENOTSUPP, we allocate a spare bio...
441 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
442 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
444 bio
->bi_bdev
= rdev
->bdev
;
445 bio
->bi_sector
= sector
;
446 bio_add_page(bio
, page
, size
, 0);
447 bio
->bi_private
= rdev
;
448 bio
->bi_end_io
= super_written
;
451 atomic_inc(&mddev
->pending_writes
);
452 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
454 rw
|= (1<<BIO_RW_BARRIER
);
455 rbio
= bio_clone(bio
, GFP_NOIO
);
456 rbio
->bi_private
= bio
;
457 rbio
->bi_end_io
= super_written_barrier
;
458 submit_bio(rw
, rbio
);
463 void md_super_wait(mddev_t
*mddev
)
465 /* wait for all superblock writes that were scheduled to complete.
466 * if any had to be retried (due to BARRIER problems), retry them
470 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
471 if (atomic_read(&mddev
->pending_writes
)==0)
473 while (mddev
->biolist
) {
475 spin_lock_irq(&mddev
->write_lock
);
476 bio
= mddev
->biolist
;
477 mddev
->biolist
= bio
->bi_next
;
479 spin_unlock_irq(&mddev
->write_lock
);
480 submit_bio(bio
->bi_rw
, bio
);
484 finish_wait(&mddev
->sb_wait
, &wq
);
487 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
492 complete((struct completion
*)bio
->bi_private
);
496 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
497 struct page
*page
, int rw
)
499 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
500 struct completion event
;
503 rw
|= (1 << BIO_RW_SYNC
);
506 bio
->bi_sector
= sector
;
507 bio_add_page(bio
, page
, size
, 0);
508 init_completion(&event
);
509 bio
->bi_private
= &event
;
510 bio
->bi_end_io
= bi_complete
;
512 wait_for_completion(&event
);
514 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
518 EXPORT_SYMBOL_GPL(sync_page_io
);
520 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
522 char b
[BDEVNAME_SIZE
];
523 if (!rdev
->sb_page
) {
531 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
537 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
538 bdevname(rdev
->bdev
,b
));
542 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
544 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
545 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
546 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
547 (sb1
->set_uuid3
== sb2
->set_uuid3
))
555 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
558 mdp_super_t
*tmp1
, *tmp2
;
560 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
561 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
563 if (!tmp1
|| !tmp2
) {
565 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
573 * nr_disks is not constant
578 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
589 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
591 unsigned int disk_csum
, csum
;
593 disk_csum
= sb
->sb_csum
;
595 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
596 sb
->sb_csum
= disk_csum
;
602 * Handle superblock details.
603 * We want to be able to handle multiple superblock formats
604 * so we have a common interface to them all, and an array of
605 * different handlers.
606 * We rely on user-space to write the initial superblock, and support
607 * reading and updating of superblocks.
608 * Interface methods are:
609 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
610 * loads and validates a superblock on dev.
611 * if refdev != NULL, compare superblocks on both devices
613 * 0 - dev has a superblock that is compatible with refdev
614 * 1 - dev has a superblock that is compatible and newer than refdev
615 * so dev should be used as the refdev in future
616 * -EINVAL superblock incompatible or invalid
617 * -othererror e.g. -EIO
619 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
620 * Verify that dev is acceptable into mddev.
621 * The first time, mddev->raid_disks will be 0, and data from
622 * dev should be merged in. Subsequent calls check that dev
623 * is new enough. Return 0 or -EINVAL
625 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
626 * Update the superblock for rdev with data in mddev
627 * This does not write to disc.
633 struct module
*owner
;
634 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
635 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
636 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
640 * load_super for 0.90.0
642 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
644 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
650 * Calculate the position of the superblock,
651 * it's at the end of the disk.
653 * It also happens to be a multiple of 4Kb.
655 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
656 rdev
->sb_offset
= sb_offset
;
658 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
663 bdevname(rdev
->bdev
, b
);
664 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
666 if (sb
->md_magic
!= MD_SB_MAGIC
) {
667 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
672 if (sb
->major_version
!= 0 ||
673 sb
->minor_version
< 90 ||
674 sb
->minor_version
> 91) {
675 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
676 sb
->major_version
, sb
->minor_version
,
681 if (sb
->raid_disks
<= 0)
684 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
685 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
690 rdev
->preferred_minor
= sb
->md_minor
;
691 rdev
->data_offset
= 0;
692 rdev
->sb_size
= MD_SB_BYTES
;
694 if (sb
->level
== LEVEL_MULTIPATH
)
697 rdev
->desc_nr
= sb
->this_disk
.number
;
703 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
704 if (!uuid_equal(refsb
, sb
)) {
705 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
706 b
, bdevname(refdev
->bdev
,b2
));
709 if (!sb_equal(refsb
, sb
)) {
710 printk(KERN_WARNING
"md: %s has same UUID"
711 " but different superblock to %s\n",
712 b
, bdevname(refdev
->bdev
, b2
));
716 ev2
= md_event(refsb
);
722 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
724 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
725 /* "this cannot possibly happen" ... */
733 * validate_super for 0.90.0
735 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
738 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
739 __u64 ev1
= md_event(sb
);
741 rdev
->raid_disk
= -1;
743 if (mddev
->raid_disks
== 0) {
744 mddev
->major_version
= 0;
745 mddev
->minor_version
= sb
->minor_version
;
746 mddev
->patch_version
= sb
->patch_version
;
747 mddev
->persistent
= ! sb
->not_persistent
;
748 mddev
->chunk_size
= sb
->chunk_size
;
749 mddev
->ctime
= sb
->ctime
;
750 mddev
->utime
= sb
->utime
;
751 mddev
->level
= sb
->level
;
752 mddev
->clevel
[0] = 0;
753 mddev
->layout
= sb
->layout
;
754 mddev
->raid_disks
= sb
->raid_disks
;
755 mddev
->size
= sb
->size
;
757 mddev
->bitmap_offset
= 0;
758 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
760 if (mddev
->minor_version
>= 91) {
761 mddev
->reshape_position
= sb
->reshape_position
;
762 mddev
->delta_disks
= sb
->delta_disks
;
763 mddev
->new_level
= sb
->new_level
;
764 mddev
->new_layout
= sb
->new_layout
;
765 mddev
->new_chunk
= sb
->new_chunk
;
767 mddev
->reshape_position
= MaxSector
;
768 mddev
->delta_disks
= 0;
769 mddev
->new_level
= mddev
->level
;
770 mddev
->new_layout
= mddev
->layout
;
771 mddev
->new_chunk
= mddev
->chunk_size
;
774 if (sb
->state
& (1<<MD_SB_CLEAN
))
775 mddev
->recovery_cp
= MaxSector
;
777 if (sb
->events_hi
== sb
->cp_events_hi
&&
778 sb
->events_lo
== sb
->cp_events_lo
) {
779 mddev
->recovery_cp
= sb
->recovery_cp
;
781 mddev
->recovery_cp
= 0;
784 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
785 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
786 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
787 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
789 mddev
->max_disks
= MD_SB_DISKS
;
791 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
792 mddev
->bitmap_file
== NULL
) {
793 if (mddev
->level
!= 1 && mddev
->level
!= 4
794 && mddev
->level
!= 5 && mddev
->level
!= 6
795 && mddev
->level
!= 10) {
796 /* FIXME use a better test */
797 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
800 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
803 } else if (mddev
->pers
== NULL
) {
804 /* Insist on good event counter while assembling */
806 if (ev1
< mddev
->events
)
808 } else if (mddev
->bitmap
) {
809 /* if adding to array with a bitmap, then we can accept an
810 * older device ... but not too old.
812 if (ev1
< mddev
->bitmap
->events_cleared
)
815 if (ev1
< mddev
->events
)
816 /* just a hot-add of a new device, leave raid_disk at -1 */
820 if (mddev
->level
!= LEVEL_MULTIPATH
) {
821 desc
= sb
->disks
+ rdev
->desc_nr
;
823 if (desc
->state
& (1<<MD_DISK_FAULTY
))
824 set_bit(Faulty
, &rdev
->flags
);
825 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
826 desc->raid_disk < mddev->raid_disks */) {
827 set_bit(In_sync
, &rdev
->flags
);
828 rdev
->raid_disk
= desc
->raid_disk
;
830 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
831 set_bit(WriteMostly
, &rdev
->flags
);
832 } else /* MULTIPATH are always insync */
833 set_bit(In_sync
, &rdev
->flags
);
838 * sync_super for 0.90.0
840 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
843 struct list_head
*tmp
;
845 int next_spare
= mddev
->raid_disks
;
848 /* make rdev->sb match mddev data..
851 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
852 * 3/ any empty disks < next_spare become removed
854 * disks[0] gets initialised to REMOVED because
855 * we cannot be sure from other fields if it has
856 * been initialised or not.
859 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
861 rdev
->sb_size
= MD_SB_BYTES
;
863 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
865 memset(sb
, 0, sizeof(*sb
));
867 sb
->md_magic
= MD_SB_MAGIC
;
868 sb
->major_version
= mddev
->major_version
;
869 sb
->patch_version
= mddev
->patch_version
;
870 sb
->gvalid_words
= 0; /* ignored */
871 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
872 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
873 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
874 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
876 sb
->ctime
= mddev
->ctime
;
877 sb
->level
= mddev
->level
;
878 sb
->size
= mddev
->size
;
879 sb
->raid_disks
= mddev
->raid_disks
;
880 sb
->md_minor
= mddev
->md_minor
;
881 sb
->not_persistent
= !mddev
->persistent
;
882 sb
->utime
= mddev
->utime
;
884 sb
->events_hi
= (mddev
->events
>>32);
885 sb
->events_lo
= (u32
)mddev
->events
;
887 if (mddev
->reshape_position
== MaxSector
)
888 sb
->minor_version
= 90;
890 sb
->minor_version
= 91;
891 sb
->reshape_position
= mddev
->reshape_position
;
892 sb
->new_level
= mddev
->new_level
;
893 sb
->delta_disks
= mddev
->delta_disks
;
894 sb
->new_layout
= mddev
->new_layout
;
895 sb
->new_chunk
= mddev
->new_chunk
;
897 mddev
->minor_version
= sb
->minor_version
;
900 sb
->recovery_cp
= mddev
->recovery_cp
;
901 sb
->cp_events_hi
= (mddev
->events
>>32);
902 sb
->cp_events_lo
= (u32
)mddev
->events
;
903 if (mddev
->recovery_cp
== MaxSector
)
904 sb
->state
= (1<< MD_SB_CLEAN
);
908 sb
->layout
= mddev
->layout
;
909 sb
->chunk_size
= mddev
->chunk_size
;
911 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
912 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
914 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
915 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
918 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
919 && !test_bit(Faulty
, &rdev2
->flags
))
920 desc_nr
= rdev2
->raid_disk
;
922 desc_nr
= next_spare
++;
923 rdev2
->desc_nr
= desc_nr
;
924 d
= &sb
->disks
[rdev2
->desc_nr
];
926 d
->number
= rdev2
->desc_nr
;
927 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
928 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
929 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
930 && !test_bit(Faulty
, &rdev2
->flags
))
931 d
->raid_disk
= rdev2
->raid_disk
;
933 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
934 if (test_bit(Faulty
, &rdev2
->flags
))
935 d
->state
= (1<<MD_DISK_FAULTY
);
936 else if (test_bit(In_sync
, &rdev2
->flags
)) {
937 d
->state
= (1<<MD_DISK_ACTIVE
);
938 d
->state
|= (1<<MD_DISK_SYNC
);
946 if (test_bit(WriteMostly
, &rdev2
->flags
))
947 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
949 /* now set the "removed" and "faulty" bits on any missing devices */
950 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
951 mdp_disk_t
*d
= &sb
->disks
[i
];
952 if (d
->state
== 0 && d
->number
== 0) {
955 d
->state
= (1<<MD_DISK_REMOVED
);
956 d
->state
|= (1<<MD_DISK_FAULTY
);
960 sb
->nr_disks
= nr_disks
;
961 sb
->active_disks
= active
;
962 sb
->working_disks
= working
;
963 sb
->failed_disks
= failed
;
964 sb
->spare_disks
= spare
;
966 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
967 sb
->sb_csum
= calc_sb_csum(sb
);
971 * version 1 superblock
974 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
976 unsigned int disk_csum
, csum
;
977 unsigned long long newcsum
;
978 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
979 unsigned int *isuper
= (unsigned int*)sb
;
982 disk_csum
= sb
->sb_csum
;
985 for (i
=0; size
>=4; size
-= 4 )
986 newcsum
+= le32_to_cpu(*isuper
++);
989 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
991 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
992 sb
->sb_csum
= disk_csum
;
993 return cpu_to_le32(csum
);
996 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
998 struct mdp_superblock_1
*sb
;
1001 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1005 * Calculate the position of the superblock.
1006 * It is always aligned to a 4K boundary and
1007 * depeding on minor_version, it can be:
1008 * 0: At least 8K, but less than 12K, from end of device
1009 * 1: At start of device
1010 * 2: 4K from start of device.
1012 switch(minor_version
) {
1014 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1016 sb_offset
&= ~(sector_t
)(4*2-1);
1017 /* convert from sectors to K */
1029 rdev
->sb_offset
= sb_offset
;
1031 /* superblock is rarely larger than 1K, but it can be larger,
1032 * and it is safe to read 4k, so we do that
1034 ret
= read_disk_sb(rdev
, 4096);
1035 if (ret
) return ret
;
1038 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1040 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1041 sb
->major_version
!= cpu_to_le32(1) ||
1042 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1043 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
1044 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1047 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1048 printk("md: invalid superblock checksum on %s\n",
1049 bdevname(rdev
->bdev
,b
));
1052 if (le64_to_cpu(sb
->data_size
) < 10) {
1053 printk("md: data_size too small on %s\n",
1054 bdevname(rdev
->bdev
,b
));
1057 rdev
->preferred_minor
= 0xffff;
1058 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1059 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1061 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1062 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1063 if (rdev
->sb_size
& bmask
)
1064 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1070 struct mdp_superblock_1
*refsb
=
1071 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1073 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1074 sb
->level
!= refsb
->level
||
1075 sb
->layout
!= refsb
->layout
||
1076 sb
->chunksize
!= refsb
->chunksize
) {
1077 printk(KERN_WARNING
"md: %s has strangely different"
1078 " superblock to %s\n",
1079 bdevname(rdev
->bdev
,b
),
1080 bdevname(refdev
->bdev
,b2
));
1083 ev1
= le64_to_cpu(sb
->events
);
1084 ev2
= le64_to_cpu(refsb
->events
);
1092 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1094 rdev
->size
= rdev
->sb_offset
;
1095 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1097 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1098 if (le32_to_cpu(sb
->chunksize
))
1099 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1101 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1106 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1108 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1109 __u64 ev1
= le64_to_cpu(sb
->events
);
1111 rdev
->raid_disk
= -1;
1113 if (mddev
->raid_disks
== 0) {
1114 mddev
->major_version
= 1;
1115 mddev
->patch_version
= 0;
1116 mddev
->persistent
= 1;
1117 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1118 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1119 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1120 mddev
->level
= le32_to_cpu(sb
->level
);
1121 mddev
->clevel
[0] = 0;
1122 mddev
->layout
= le32_to_cpu(sb
->layout
);
1123 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1124 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1125 mddev
->events
= ev1
;
1126 mddev
->bitmap_offset
= 0;
1127 mddev
->default_bitmap_offset
= 1024 >> 9;
1129 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1130 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1132 mddev
->max_disks
= (4096-256)/2;
1134 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1135 mddev
->bitmap_file
== NULL
) {
1136 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1137 && mddev
->level
!= 10) {
1138 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1141 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1143 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1144 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1145 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1146 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1147 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1148 mddev
->new_chunk
= le32_to_cpu(sb
->new_chunk
)<<9;
1150 mddev
->reshape_position
= MaxSector
;
1151 mddev
->delta_disks
= 0;
1152 mddev
->new_level
= mddev
->level
;
1153 mddev
->new_layout
= mddev
->layout
;
1154 mddev
->new_chunk
= mddev
->chunk_size
;
1157 } else if (mddev
->pers
== NULL
) {
1158 /* Insist of good event counter while assembling */
1160 if (ev1
< mddev
->events
)
1162 } else if (mddev
->bitmap
) {
1163 /* If adding to array with a bitmap, then we can accept an
1164 * older device, but not too old.
1166 if (ev1
< mddev
->bitmap
->events_cleared
)
1169 if (ev1
< mddev
->events
)
1170 /* just a hot-add of a new device, leave raid_disk at -1 */
1173 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1175 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1176 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1178 case 0xffff: /* spare */
1180 case 0xfffe: /* faulty */
1181 set_bit(Faulty
, &rdev
->flags
);
1184 if ((le32_to_cpu(sb
->feature_map
) &
1185 MD_FEATURE_RECOVERY_OFFSET
))
1186 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1188 set_bit(In_sync
, &rdev
->flags
);
1189 rdev
->raid_disk
= role
;
1192 if (sb
->devflags
& WriteMostly1
)
1193 set_bit(WriteMostly
, &rdev
->flags
);
1194 } else /* MULTIPATH are always insync */
1195 set_bit(In_sync
, &rdev
->flags
);
1200 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1202 struct mdp_superblock_1
*sb
;
1203 struct list_head
*tmp
;
1206 /* make rdev->sb match mddev and rdev data. */
1208 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1210 sb
->feature_map
= 0;
1212 sb
->recovery_offset
= cpu_to_le64(0);
1213 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1214 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1215 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1217 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1218 sb
->events
= cpu_to_le64(mddev
->events
);
1220 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1222 sb
->resync_offset
= cpu_to_le64(0);
1224 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1226 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1227 sb
->size
= cpu_to_le64(mddev
->size
<<1);
1229 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1230 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1231 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1234 if (rdev
->raid_disk
>= 0 &&
1235 !test_bit(In_sync
, &rdev
->flags
) &&
1236 rdev
->recovery_offset
> 0) {
1237 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1238 sb
->recovery_offset
= cpu_to_le64(rdev
->recovery_offset
);
1241 if (mddev
->reshape_position
!= MaxSector
) {
1242 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1243 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1244 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1245 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1246 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1247 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk
>>9);
1251 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1252 if (rdev2
->desc_nr
+1 > max_dev
)
1253 max_dev
= rdev2
->desc_nr
+1;
1255 sb
->max_dev
= cpu_to_le32(max_dev
);
1256 for (i
=0; i
<max_dev
;i
++)
1257 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1259 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1261 if (test_bit(Faulty
, &rdev2
->flags
))
1262 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1263 else if (test_bit(In_sync
, &rdev2
->flags
))
1264 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1265 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1266 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1268 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1271 sb
->sb_csum
= calc_sb_1_csum(sb
);
1275 static struct super_type super_types
[] = {
1278 .owner
= THIS_MODULE
,
1279 .load_super
= super_90_load
,
1280 .validate_super
= super_90_validate
,
1281 .sync_super
= super_90_sync
,
1285 .owner
= THIS_MODULE
,
1286 .load_super
= super_1_load
,
1287 .validate_super
= super_1_validate
,
1288 .sync_super
= super_1_sync
,
1292 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1294 struct list_head
*tmp
;
1297 ITERATE_RDEV(mddev
,rdev
,tmp
)
1298 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1304 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1306 struct list_head
*tmp
;
1309 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1310 if (match_dev_unit(mddev2
, rdev
))
1316 static LIST_HEAD(pending_raid_disks
);
1318 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1320 mdk_rdev_t
*same_pdev
;
1321 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1329 /* make sure rdev->size exceeds mddev->size */
1330 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1332 /* Cannot change size, so fail */
1335 mddev
->size
= rdev
->size
;
1337 same_pdev
= match_dev_unit(mddev
, rdev
);
1340 "%s: WARNING: %s appears to be on the same physical"
1341 " disk as %s. True\n protection against single-disk"
1342 " failure might be compromised.\n",
1343 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1344 bdevname(same_pdev
->bdev
,b2
));
1346 /* Verify rdev->desc_nr is unique.
1347 * If it is -1, assign a free number, else
1348 * check number is not in use
1350 if (rdev
->desc_nr
< 0) {
1352 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1353 while (find_rdev_nr(mddev
, choice
))
1355 rdev
->desc_nr
= choice
;
1357 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1360 bdevname(rdev
->bdev
,b
);
1361 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1363 while ( (s
=strchr(rdev
->kobj
.k_name
, '/')) != NULL
)
1366 list_add(&rdev
->same_set
, &mddev
->disks
);
1367 rdev
->mddev
= mddev
;
1368 printk(KERN_INFO
"md: bind<%s>\n", b
);
1370 rdev
->kobj
.parent
= &mddev
->kobj
;
1371 kobject_add(&rdev
->kobj
);
1373 if (rdev
->bdev
->bd_part
)
1374 ko
= &rdev
->bdev
->bd_part
->kobj
;
1376 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1377 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1378 bd_claim_by_disk(rdev
->bdev
, rdev
, mddev
->gendisk
);
1382 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1384 char b
[BDEVNAME_SIZE
];
1389 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1390 list_del_init(&rdev
->same_set
);
1391 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1393 sysfs_remove_link(&rdev
->kobj
, "block");
1394 kobject_del(&rdev
->kobj
);
1398 * prevent the device from being mounted, repartitioned or
1399 * otherwise reused by a RAID array (or any other kernel
1400 * subsystem), by bd_claiming the device.
1402 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1405 struct block_device
*bdev
;
1406 char b
[BDEVNAME_SIZE
];
1408 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1410 printk(KERN_ERR
"md: could not open %s.\n",
1411 __bdevname(dev
, b
));
1412 return PTR_ERR(bdev
);
1414 err
= bd_claim(bdev
, rdev
);
1416 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1425 static void unlock_rdev(mdk_rdev_t
*rdev
)
1427 struct block_device
*bdev
= rdev
->bdev
;
1435 void md_autodetect_dev(dev_t dev
);
1437 static void export_rdev(mdk_rdev_t
* rdev
)
1439 char b
[BDEVNAME_SIZE
];
1440 printk(KERN_INFO
"md: export_rdev(%s)\n",
1441 bdevname(rdev
->bdev
,b
));
1445 list_del_init(&rdev
->same_set
);
1447 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1450 kobject_put(&rdev
->kobj
);
1453 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1455 unbind_rdev_from_array(rdev
);
1459 static void export_array(mddev_t
*mddev
)
1461 struct list_head
*tmp
;
1464 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1469 kick_rdev_from_array(rdev
);
1471 if (!list_empty(&mddev
->disks
))
1473 mddev
->raid_disks
= 0;
1474 mddev
->major_version
= 0;
1477 static void print_desc(mdp_disk_t
*desc
)
1479 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1480 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1483 static void print_sb(mdp_super_t
*sb
)
1488 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1489 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1490 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1492 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1493 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1494 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1495 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1496 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1497 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1498 sb
->failed_disks
, sb
->spare_disks
,
1499 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1502 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1505 desc
= sb
->disks
+ i
;
1506 if (desc
->number
|| desc
->major
|| desc
->minor
||
1507 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1508 printk(" D %2d: ", i
);
1512 printk(KERN_INFO
"md: THIS: ");
1513 print_desc(&sb
->this_disk
);
1517 static void print_rdev(mdk_rdev_t
*rdev
)
1519 char b
[BDEVNAME_SIZE
];
1520 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1521 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1522 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1524 if (rdev
->sb_loaded
) {
1525 printk(KERN_INFO
"md: rdev superblock:\n");
1526 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1528 printk(KERN_INFO
"md: no rdev superblock!\n");
1531 static void md_print_devices(void)
1533 struct list_head
*tmp
, *tmp2
;
1536 char b
[BDEVNAME_SIZE
];
1539 printk("md: **********************************\n");
1540 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1541 printk("md: **********************************\n");
1542 ITERATE_MDDEV(mddev
,tmp
) {
1545 bitmap_print_sb(mddev
->bitmap
);
1547 printk("%s: ", mdname(mddev
));
1548 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1549 printk("<%s>", bdevname(rdev
->bdev
,b
));
1552 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1555 printk("md: **********************************\n");
1560 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1562 /* Update each superblock (in-memory image), but
1563 * if we are allowed to, skip spares which already
1564 * have the right event counter, or have one earlier
1565 * (which would mean they aren't being marked as dirty
1566 * with the rest of the array)
1569 struct list_head
*tmp
;
1571 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1572 if (rdev
->sb_events
== mddev
->events
||
1574 rdev
->raid_disk
< 0 &&
1575 (rdev
->sb_events
&1)==0 &&
1576 rdev
->sb_events
+1 == mddev
->events
)) {
1577 /* Don't update this superblock */
1578 rdev
->sb_loaded
= 2;
1580 super_types
[mddev
->major_version
].
1581 sync_super(mddev
, rdev
);
1582 rdev
->sb_loaded
= 1;
1587 void md_update_sb(mddev_t
* mddev
)
1590 struct list_head
*tmp
;
1596 spin_lock_irq(&mddev
->write_lock
);
1597 sync_req
= mddev
->in_sync
;
1598 mddev
->utime
= get_seconds();
1599 if (mddev
->sb_dirty
== 3)
1600 /* just a clean<-> dirty transition, possibly leave spares alone,
1601 * though if events isn't the right even/odd, we will have to do
1606 /* If this is just a dirty<->clean transition, and the array is clean
1607 * and 'events' is odd, we can roll back to the previous clean state */
1608 if (mddev
->sb_dirty
== 3
1609 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
1610 && (mddev
->events
& 1))
1613 /* otherwise we have to go forward and ... */
1615 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
1616 /* .. if the array isn't clean, insist on an odd 'events' */
1617 if ((mddev
->events
&1)==0) {
1622 /* otherwise insist on an even 'events' (for clean states) */
1623 if ((mddev
->events
&1)) {
1630 if (!mddev
->events
) {
1632 * oops, this 64-bit counter should never wrap.
1633 * Either we are in around ~1 trillion A.C., assuming
1634 * 1 reboot per second, or we have a bug:
1639 mddev
->sb_dirty
= 2;
1640 sync_sbs(mddev
, nospares
);
1643 * do not write anything to disk if using
1644 * nonpersistent superblocks
1646 if (!mddev
->persistent
) {
1647 mddev
->sb_dirty
= 0;
1648 spin_unlock_irq(&mddev
->write_lock
);
1649 wake_up(&mddev
->sb_wait
);
1652 spin_unlock_irq(&mddev
->write_lock
);
1655 "md: updating %s RAID superblock on device (in sync %d)\n",
1656 mdname(mddev
),mddev
->in_sync
);
1658 err
= bitmap_update_sb(mddev
->bitmap
);
1659 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1660 char b
[BDEVNAME_SIZE
];
1661 dprintk(KERN_INFO
"md: ");
1662 if (rdev
->sb_loaded
!= 1)
1663 continue; /* no noise on spare devices */
1664 if (test_bit(Faulty
, &rdev
->flags
))
1665 dprintk("(skipping faulty ");
1667 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1668 if (!test_bit(Faulty
, &rdev
->flags
)) {
1669 md_super_write(mddev
,rdev
,
1670 rdev
->sb_offset
<<1, rdev
->sb_size
,
1672 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1673 bdevname(rdev
->bdev
,b
),
1674 (unsigned long long)rdev
->sb_offset
);
1675 rdev
->sb_events
= mddev
->events
;
1679 if (mddev
->level
== LEVEL_MULTIPATH
)
1680 /* only need to write one superblock... */
1683 md_super_wait(mddev
);
1684 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1686 spin_lock_irq(&mddev
->write_lock
);
1687 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1688 /* have to write it out again */
1689 spin_unlock_irq(&mddev
->write_lock
);
1692 mddev
->sb_dirty
= 0;
1693 spin_unlock_irq(&mddev
->write_lock
);
1694 wake_up(&mddev
->sb_wait
);
1697 EXPORT_SYMBOL_GPL(md_update_sb
);
1699 /* words written to sysfs files may, or my not, be \n terminated.
1700 * We want to accept with case. For this we use cmd_match.
1702 static int cmd_match(const char *cmd
, const char *str
)
1704 /* See if cmd, written into a sysfs file, matches
1705 * str. They must either be the same, or cmd can
1706 * have a trailing newline
1708 while (*cmd
&& *str
&& *cmd
== *str
) {
1719 struct rdev_sysfs_entry
{
1720 struct attribute attr
;
1721 ssize_t (*show
)(mdk_rdev_t
*, char *);
1722 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1726 state_show(mdk_rdev_t
*rdev
, char *page
)
1731 if (test_bit(Faulty
, &rdev
->flags
)) {
1732 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1735 if (test_bit(In_sync
, &rdev
->flags
)) {
1736 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1739 if (test_bit(WriteMostly
, &rdev
->flags
)) {
1740 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
1743 if (!test_bit(Faulty
, &rdev
->flags
) &&
1744 !test_bit(In_sync
, &rdev
->flags
)) {
1745 len
+= sprintf(page
+len
, "%sspare", sep
);
1748 return len
+sprintf(page
+len
, "\n");
1752 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1755 * faulty - simulates and error
1756 * remove - disconnects the device
1757 * writemostly - sets write_mostly
1758 * -writemostly - clears write_mostly
1761 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
1762 md_error(rdev
->mddev
, rdev
);
1764 } else if (cmd_match(buf
, "remove")) {
1765 if (rdev
->raid_disk
>= 0)
1768 mddev_t
*mddev
= rdev
->mddev
;
1769 kick_rdev_from_array(rdev
);
1770 md_update_sb(mddev
);
1771 md_new_event(mddev
);
1774 } else if (cmd_match(buf
, "writemostly")) {
1775 set_bit(WriteMostly
, &rdev
->flags
);
1777 } else if (cmd_match(buf
, "-writemostly")) {
1778 clear_bit(WriteMostly
, &rdev
->flags
);
1781 return err
? err
: len
;
1783 static struct rdev_sysfs_entry
1784 rdev_state
= __ATTR(state
, 0644, state_show
, state_store
);
1787 super_show(mdk_rdev_t
*rdev
, char *page
)
1789 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1790 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1791 return rdev
->sb_size
;
1795 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1798 errors_show(mdk_rdev_t
*rdev
, char *page
)
1800 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1804 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1807 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1808 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1809 atomic_set(&rdev
->corrected_errors
, n
);
1814 static struct rdev_sysfs_entry rdev_errors
=
1815 __ATTR(errors
, 0644, errors_show
, errors_store
);
1818 slot_show(mdk_rdev_t
*rdev
, char *page
)
1820 if (rdev
->raid_disk
< 0)
1821 return sprintf(page
, "none\n");
1823 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1827 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1830 int slot
= simple_strtoul(buf
, &e
, 10);
1831 if (strncmp(buf
, "none", 4)==0)
1833 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1835 if (rdev
->mddev
->pers
)
1836 /* Cannot set slot in active array (yet) */
1838 if (slot
>= rdev
->mddev
->raid_disks
)
1840 rdev
->raid_disk
= slot
;
1841 /* assume it is working */
1843 set_bit(In_sync
, &rdev
->flags
);
1848 static struct rdev_sysfs_entry rdev_slot
=
1849 __ATTR(slot
, 0644, slot_show
, slot_store
);
1852 offset_show(mdk_rdev_t
*rdev
, char *page
)
1854 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1858 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1861 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1862 if (e
==buf
|| (*e
&& *e
!= '\n'))
1864 if (rdev
->mddev
->pers
)
1866 rdev
->data_offset
= offset
;
1870 static struct rdev_sysfs_entry rdev_offset
=
1871 __ATTR(offset
, 0644, offset_show
, offset_store
);
1874 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
1876 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->size
);
1880 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1883 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1884 if (e
==buf
|| (*e
&& *e
!= '\n'))
1886 if (rdev
->mddev
->pers
)
1889 if (size
< rdev
->mddev
->size
|| rdev
->mddev
->size
== 0)
1890 rdev
->mddev
->size
= size
;
1894 static struct rdev_sysfs_entry rdev_size
=
1895 __ATTR(size
, 0644, rdev_size_show
, rdev_size_store
);
1897 static struct attribute
*rdev_default_attrs
[] = {
1907 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1909 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1910 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1914 return entry
->show(rdev
, page
);
1918 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1919 const char *page
, size_t length
)
1921 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1922 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1926 return entry
->store(rdev
, page
, length
);
1929 static void rdev_free(struct kobject
*ko
)
1931 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1934 static struct sysfs_ops rdev_sysfs_ops
= {
1935 .show
= rdev_attr_show
,
1936 .store
= rdev_attr_store
,
1938 static struct kobj_type rdev_ktype
= {
1939 .release
= rdev_free
,
1940 .sysfs_ops
= &rdev_sysfs_ops
,
1941 .default_attrs
= rdev_default_attrs
,
1945 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1947 * mark the device faulty if:
1949 * - the device is nonexistent (zero size)
1950 * - the device has no valid superblock
1952 * a faulty rdev _never_ has rdev->sb set.
1954 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1956 char b
[BDEVNAME_SIZE
];
1961 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1963 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1964 return ERR_PTR(-ENOMEM
);
1967 if ((err
= alloc_disk_sb(rdev
)))
1970 err
= lock_rdev(rdev
, newdev
);
1974 rdev
->kobj
.parent
= NULL
;
1975 rdev
->kobj
.ktype
= &rdev_ktype
;
1976 kobject_init(&rdev
->kobj
);
1980 rdev
->data_offset
= 0;
1981 rdev
->sb_events
= 0;
1982 atomic_set(&rdev
->nr_pending
, 0);
1983 atomic_set(&rdev
->read_errors
, 0);
1984 atomic_set(&rdev
->corrected_errors
, 0);
1986 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1989 "md: %s has zero or unknown size, marking faulty!\n",
1990 bdevname(rdev
->bdev
,b
));
1995 if (super_format
>= 0) {
1996 err
= super_types
[super_format
].
1997 load_super(rdev
, NULL
, super_minor
);
1998 if (err
== -EINVAL
) {
2000 "md: %s has invalid sb, not importing!\n",
2001 bdevname(rdev
->bdev
,b
));
2006 "md: could not read %s's sb, not importing!\n",
2007 bdevname(rdev
->bdev
,b
));
2011 INIT_LIST_HEAD(&rdev
->same_set
);
2016 if (rdev
->sb_page
) {
2022 return ERR_PTR(err
);
2026 * Check a full RAID array for plausibility
2030 static void analyze_sbs(mddev_t
* mddev
)
2033 struct list_head
*tmp
;
2034 mdk_rdev_t
*rdev
, *freshest
;
2035 char b
[BDEVNAME_SIZE
];
2038 ITERATE_RDEV(mddev
,rdev
,tmp
)
2039 switch (super_types
[mddev
->major_version
].
2040 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2048 "md: fatal superblock inconsistency in %s"
2049 " -- removing from array\n",
2050 bdevname(rdev
->bdev
,b
));
2051 kick_rdev_from_array(rdev
);
2055 super_types
[mddev
->major_version
].
2056 validate_super(mddev
, freshest
);
2059 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2060 if (rdev
!= freshest
)
2061 if (super_types
[mddev
->major_version
].
2062 validate_super(mddev
, rdev
)) {
2063 printk(KERN_WARNING
"md: kicking non-fresh %s"
2065 bdevname(rdev
->bdev
,b
));
2066 kick_rdev_from_array(rdev
);
2069 if (mddev
->level
== LEVEL_MULTIPATH
) {
2070 rdev
->desc_nr
= i
++;
2071 rdev
->raid_disk
= rdev
->desc_nr
;
2072 set_bit(In_sync
, &rdev
->flags
);
2078 if (mddev
->recovery_cp
!= MaxSector
&&
2080 printk(KERN_ERR
"md: %s: raid array is not clean"
2081 " -- starting background reconstruction\n",
2087 safe_delay_show(mddev_t
*mddev
, char *page
)
2089 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2090 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2093 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2101 /* remove a period, and count digits after it */
2102 if (len
>= sizeof(buf
))
2104 strlcpy(buf
, cbuf
, len
);
2106 for (i
=0; i
<len
; i
++) {
2108 if (isdigit(buf
[i
])) {
2113 } else if (buf
[i
] == '.') {
2118 msec
= simple_strtoul(buf
, &e
, 10);
2119 if (e
== buf
|| (*e
&& *e
!= '\n'))
2121 msec
= (msec
* 1000) / scale
;
2123 mddev
->safemode_delay
= 0;
2125 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2126 if (mddev
->safemode_delay
== 0)
2127 mddev
->safemode_delay
= 1;
2131 static struct md_sysfs_entry md_safe_delay
=
2132 __ATTR(safe_mode_delay
, 0644,safe_delay_show
, safe_delay_store
);
2135 level_show(mddev_t
*mddev
, char *page
)
2137 struct mdk_personality
*p
= mddev
->pers
;
2139 return sprintf(page
, "%s\n", p
->name
);
2140 else if (mddev
->clevel
[0])
2141 return sprintf(page
, "%s\n", mddev
->clevel
);
2142 else if (mddev
->level
!= LEVEL_NONE
)
2143 return sprintf(page
, "%d\n", mddev
->level
);
2149 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2156 if (len
>= sizeof(mddev
->clevel
))
2158 strncpy(mddev
->clevel
, buf
, len
);
2159 if (mddev
->clevel
[len
-1] == '\n')
2161 mddev
->clevel
[len
] = 0;
2162 mddev
->level
= LEVEL_NONE
;
2166 static struct md_sysfs_entry md_level
=
2167 __ATTR(level
, 0644, level_show
, level_store
);
2171 layout_show(mddev_t
*mddev
, char *page
)
2173 /* just a number, not meaningful for all levels */
2174 return sprintf(page
, "%d\n", mddev
->layout
);
2178 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2181 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2185 if (!*buf
|| (*e
&& *e
!= '\n'))
2191 static struct md_sysfs_entry md_layout
=
2192 __ATTR(layout
, 0655, layout_show
, layout_store
);
2196 raid_disks_show(mddev_t
*mddev
, char *page
)
2198 if (mddev
->raid_disks
== 0)
2200 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2203 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2206 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2208 /* can only set raid_disks if array is not yet active */
2211 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2213 if (!*buf
|| (*e
&& *e
!= '\n'))
2217 rv
= update_raid_disks(mddev
, n
);
2219 mddev
->raid_disks
= n
;
2220 return rv
? rv
: len
;
2222 static struct md_sysfs_entry md_raid_disks
=
2223 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
2226 chunk_size_show(mddev_t
*mddev
, char *page
)
2228 return sprintf(page
, "%d\n", mddev
->chunk_size
);
2232 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2234 /* can only set chunk_size if array is not yet active */
2236 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2240 if (!*buf
|| (*e
&& *e
!= '\n'))
2243 mddev
->chunk_size
= n
;
2246 static struct md_sysfs_entry md_chunk_size
=
2247 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
2250 resync_start_show(mddev_t
*mddev
, char *page
)
2252 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2256 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2258 /* can only set chunk_size if array is not yet active */
2260 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2264 if (!*buf
|| (*e
&& *e
!= '\n'))
2267 mddev
->recovery_cp
= n
;
2270 static struct md_sysfs_entry md_resync_start
=
2271 __ATTR(resync_start
, 0644, resync_start_show
, resync_start_store
);
2274 * The array state can be:
2277 * No devices, no size, no level
2278 * Equivalent to STOP_ARRAY ioctl
2280 * May have some settings, but array is not active
2281 * all IO results in error
2282 * When written, doesn't tear down array, but just stops it
2283 * suspended (not supported yet)
2284 * All IO requests will block. The array can be reconfigured.
2285 * Writing this, if accepted, will block until array is quiessent
2287 * no resync can happen. no superblocks get written.
2288 * write requests fail
2290 * like readonly, but behaves like 'clean' on a write request.
2292 * clean - no pending writes, but otherwise active.
2293 * When written to inactive array, starts without resync
2294 * If a write request arrives then
2295 * if metadata is known, mark 'dirty' and switch to 'active'.
2296 * if not known, block and switch to write-pending
2297 * If written to an active array that has pending writes, then fails.
2299 * fully active: IO and resync can be happening.
2300 * When written to inactive array, starts with resync
2303 * clean, but writes are blocked waiting for 'active' to be written.
2306 * like active, but no writes have been seen for a while (100msec).
2309 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
2310 write_pending
, active_idle
, bad_word
};
2311 static char *array_states
[] = {
2312 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2313 "write-pending", "active-idle", NULL
};
2315 static int match_word(const char *word
, char **list
)
2318 for (n
=0; list
[n
]; n
++)
2319 if (cmd_match(word
, list
[n
]))
2325 array_state_show(mddev_t
*mddev
, char *page
)
2327 enum array_state st
= inactive
;
2340 else if (mddev
->safemode
)
2346 if (list_empty(&mddev
->disks
) &&
2347 mddev
->raid_disks
== 0 &&
2353 return sprintf(page
, "%s\n", array_states
[st
]);
2356 static int do_md_stop(mddev_t
* mddev
, int ro
);
2357 static int do_md_run(mddev_t
* mddev
);
2358 static int restart_array(mddev_t
*mddev
);
2361 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2364 enum array_state st
= match_word(buf
, array_states
);
2369 /* stopping an active array */
2371 if (atomic_read(&mddev
->active
) > 1)
2373 err
= do_md_stop(mddev
, 0);
2377 /* stopping an active array */
2379 if (atomic_read(&mddev
->active
) > 1)
2381 err
= do_md_stop(mddev
, 2);
2385 break; /* not supported yet */
2388 err
= do_md_stop(mddev
, 1);
2391 err
= do_md_run(mddev
);
2395 /* stopping an active array */
2397 err
= do_md_stop(mddev
, 1);
2399 mddev
->ro
= 2; /* FIXME mark devices writable */
2402 err
= do_md_run(mddev
);
2407 restart_array(mddev
);
2408 spin_lock_irq(&mddev
->write_lock
);
2409 if (atomic_read(&mddev
->writes_pending
) == 0) {
2411 mddev
->sb_dirty
= 1;
2413 spin_unlock_irq(&mddev
->write_lock
);
2416 mddev
->recovery_cp
= MaxSector
;
2417 err
= do_md_run(mddev
);
2422 restart_array(mddev
);
2423 mddev
->sb_dirty
= 0;
2424 wake_up(&mddev
->sb_wait
);
2428 err
= do_md_run(mddev
);
2433 /* these cannot be set */
2441 static struct md_sysfs_entry md_array_state
= __ATTR(array_state
, 0644, array_state_show
, array_state_store
);
2444 null_show(mddev_t
*mddev
, char *page
)
2450 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2452 /* buf must be %d:%d\n? giving major and minor numbers */
2453 /* The new device is added to the array.
2454 * If the array has a persistent superblock, we read the
2455 * superblock to initialise info and check validity.
2456 * Otherwise, only checking done is that in bind_rdev_to_array,
2457 * which mainly checks size.
2460 int major
= simple_strtoul(buf
, &e
, 10);
2466 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
2468 minor
= simple_strtoul(e
+1, &e
, 10);
2469 if (*e
&& *e
!= '\n')
2471 dev
= MKDEV(major
, minor
);
2472 if (major
!= MAJOR(dev
) ||
2473 minor
!= MINOR(dev
))
2477 if (mddev
->persistent
) {
2478 rdev
= md_import_device(dev
, mddev
->major_version
,
2479 mddev
->minor_version
);
2480 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
2481 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2482 mdk_rdev_t
, same_set
);
2483 err
= super_types
[mddev
->major_version
]
2484 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2489 rdev
= md_import_device(dev
, -1, -1);
2492 return PTR_ERR(rdev
);
2493 err
= bind_rdev_to_array(rdev
, mddev
);
2497 return err
? err
: len
;
2500 static struct md_sysfs_entry md_new_device
=
2501 __ATTR(new_dev
, 0200, null_show
, new_dev_store
);
2504 size_show(mddev_t
*mddev
, char *page
)
2506 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
2509 static int update_size(mddev_t
*mddev
, unsigned long size
);
2512 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2514 /* If array is inactive, we can reduce the component size, but
2515 * not increase it (except from 0).
2516 * If array is active, we can try an on-line resize
2520 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
2521 if (!*buf
|| *buf
== '\n' ||
2526 err
= update_size(mddev
, size
);
2527 md_update_sb(mddev
);
2529 if (mddev
->size
== 0 ||
2535 return err
? err
: len
;
2538 static struct md_sysfs_entry md_size
=
2539 __ATTR(component_size
, 0644, size_show
, size_store
);
2543 * This is either 'none' for arrays with externally managed metadata,
2544 * or N.M for internally known formats
2547 metadata_show(mddev_t
*mddev
, char *page
)
2549 if (mddev
->persistent
)
2550 return sprintf(page
, "%d.%d\n",
2551 mddev
->major_version
, mddev
->minor_version
);
2553 return sprintf(page
, "none\n");
2557 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2561 if (!list_empty(&mddev
->disks
))
2564 if (cmd_match(buf
, "none")) {
2565 mddev
->persistent
= 0;
2566 mddev
->major_version
= 0;
2567 mddev
->minor_version
= 90;
2570 major
= simple_strtoul(buf
, &e
, 10);
2571 if (e
==buf
|| *e
!= '.')
2574 minor
= simple_strtoul(buf
, &e
, 10);
2575 if (e
==buf
|| *e
!= '\n')
2577 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2578 super_types
[major
].name
== NULL
)
2580 mddev
->major_version
= major
;
2581 mddev
->minor_version
= minor
;
2582 mddev
->persistent
= 1;
2586 static struct md_sysfs_entry md_metadata
=
2587 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2590 action_show(mddev_t
*mddev
, char *page
)
2592 char *type
= "idle";
2593 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2594 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2595 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
2597 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2598 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2600 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2607 return sprintf(page
, "%s\n", type
);
2611 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2613 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2616 if (cmd_match(page
, "idle")) {
2617 if (mddev
->sync_thread
) {
2618 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2619 md_unregister_thread(mddev
->sync_thread
);
2620 mddev
->sync_thread
= NULL
;
2621 mddev
->recovery
= 0;
2623 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2624 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2626 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2627 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2628 else if (cmd_match(page
, "reshape")) {
2630 if (mddev
->pers
->start_reshape
== NULL
)
2632 err
= mddev
->pers
->start_reshape(mddev
);
2636 if (cmd_match(page
, "check"))
2637 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2638 else if (!cmd_match(page
, "repair"))
2640 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2641 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2643 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2644 md_wakeup_thread(mddev
->thread
);
2649 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2651 return sprintf(page
, "%llu\n",
2652 (unsigned long long) mddev
->resync_mismatches
);
2655 static struct md_sysfs_entry
2656 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2659 static struct md_sysfs_entry
2660 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2663 sync_min_show(mddev_t
*mddev
, char *page
)
2665 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
2666 mddev
->sync_speed_min
? "local": "system");
2670 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2674 if (strncmp(buf
, "system", 6)==0) {
2675 mddev
->sync_speed_min
= 0;
2678 min
= simple_strtoul(buf
, &e
, 10);
2679 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
2681 mddev
->sync_speed_min
= min
;
2685 static struct md_sysfs_entry md_sync_min
=
2686 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
2689 sync_max_show(mddev_t
*mddev
, char *page
)
2691 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
2692 mddev
->sync_speed_max
? "local": "system");
2696 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2700 if (strncmp(buf
, "system", 6)==0) {
2701 mddev
->sync_speed_max
= 0;
2704 max
= simple_strtoul(buf
, &e
, 10);
2705 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
2707 mddev
->sync_speed_max
= max
;
2711 static struct md_sysfs_entry md_sync_max
=
2712 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
2716 sync_speed_show(mddev_t
*mddev
, char *page
)
2718 unsigned long resync
, dt
, db
;
2719 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2720 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
2722 db
= resync
- (mddev
->resync_mark_cnt
);
2723 return sprintf(page
, "%ld\n", db
/dt
/2); /* K/sec */
2726 static struct md_sysfs_entry
2727 md_sync_speed
= __ATTR_RO(sync_speed
);
2730 sync_completed_show(mddev_t
*mddev
, char *page
)
2732 unsigned long max_blocks
, resync
;
2734 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
2735 max_blocks
= mddev
->resync_max_sectors
;
2737 max_blocks
= mddev
->size
<< 1;
2739 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
));
2740 return sprintf(page
, "%lu / %lu\n", resync
, max_blocks
);
2743 static struct md_sysfs_entry
2744 md_sync_completed
= __ATTR_RO(sync_completed
);
2747 suspend_lo_show(mddev_t
*mddev
, char *page
)
2749 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
2753 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2756 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2758 if (mddev
->pers
->quiesce
== NULL
)
2760 if (buf
== e
|| (*e
&& *e
!= '\n'))
2762 if (new >= mddev
->suspend_hi
||
2763 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
2764 mddev
->suspend_lo
= new;
2765 mddev
->pers
->quiesce(mddev
, 2);
2770 static struct md_sysfs_entry md_suspend_lo
=
2771 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
2775 suspend_hi_show(mddev_t
*mddev
, char *page
)
2777 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
2781 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2784 unsigned long long new = simple_strtoull(buf
, &e
, 10);
2786 if (mddev
->pers
->quiesce
== NULL
)
2788 if (buf
== e
|| (*e
&& *e
!= '\n'))
2790 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
2791 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
2792 mddev
->suspend_hi
= new;
2793 mddev
->pers
->quiesce(mddev
, 1);
2794 mddev
->pers
->quiesce(mddev
, 0);
2799 static struct md_sysfs_entry md_suspend_hi
=
2800 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
2803 static struct attribute
*md_default_attrs
[] = {
2806 &md_raid_disks
.attr
,
2807 &md_chunk_size
.attr
,
2809 &md_resync_start
.attr
,
2811 &md_new_device
.attr
,
2812 &md_safe_delay
.attr
,
2813 &md_array_state
.attr
,
2817 static struct attribute
*md_redundancy_attrs
[] = {
2819 &md_mismatches
.attr
,
2822 &md_sync_speed
.attr
,
2823 &md_sync_completed
.attr
,
2824 &md_suspend_lo
.attr
,
2825 &md_suspend_hi
.attr
,
2828 static struct attribute_group md_redundancy_group
= {
2830 .attrs
= md_redundancy_attrs
,
2835 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2837 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2838 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2843 rv
= mddev_lock(mddev
);
2845 rv
= entry
->show(mddev
, page
);
2846 mddev_unlock(mddev
);
2852 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2853 const char *page
, size_t length
)
2855 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2856 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2861 rv
= mddev_lock(mddev
);
2863 rv
= entry
->store(mddev
, page
, length
);
2864 mddev_unlock(mddev
);
2869 static void md_free(struct kobject
*ko
)
2871 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2875 static struct sysfs_ops md_sysfs_ops
= {
2876 .show
= md_attr_show
,
2877 .store
= md_attr_store
,
2879 static struct kobj_type md_ktype
= {
2881 .sysfs_ops
= &md_sysfs_ops
,
2882 .default_attrs
= md_default_attrs
,
2887 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2889 static DEFINE_MUTEX(disks_mutex
);
2890 mddev_t
*mddev
= mddev_find(dev
);
2891 struct gendisk
*disk
;
2892 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2893 int shift
= partitioned
? MdpMinorShift
: 0;
2894 int unit
= MINOR(dev
) >> shift
;
2899 mutex_lock(&disks_mutex
);
2900 if (mddev
->gendisk
) {
2901 mutex_unlock(&disks_mutex
);
2905 disk
= alloc_disk(1 << shift
);
2907 mutex_unlock(&disks_mutex
);
2911 disk
->major
= MAJOR(dev
);
2912 disk
->first_minor
= unit
<< shift
;
2914 sprintf(disk
->disk_name
, "md_d%d", unit
);
2916 sprintf(disk
->disk_name
, "md%d", unit
);
2917 disk
->fops
= &md_fops
;
2918 disk
->private_data
= mddev
;
2919 disk
->queue
= mddev
->queue
;
2921 mddev
->gendisk
= disk
;
2922 mutex_unlock(&disks_mutex
);
2923 mddev
->kobj
.parent
= &disk
->kobj
;
2924 mddev
->kobj
.k_name
= NULL
;
2925 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2926 mddev
->kobj
.ktype
= &md_ktype
;
2927 kobject_register(&mddev
->kobj
);
2931 static void md_safemode_timeout(unsigned long data
)
2933 mddev_t
*mddev
= (mddev_t
*) data
;
2935 mddev
->safemode
= 1;
2936 md_wakeup_thread(mddev
->thread
);
2939 static int start_dirty_degraded
;
2941 static int do_md_run(mddev_t
* mddev
)
2945 struct list_head
*tmp
;
2947 struct gendisk
*disk
;
2948 struct mdk_personality
*pers
;
2949 char b
[BDEVNAME_SIZE
];
2951 if (list_empty(&mddev
->disks
))
2952 /* cannot run an array with no devices.. */
2959 * Analyze all RAID superblock(s)
2961 if (!mddev
->raid_disks
)
2964 chunk_size
= mddev
->chunk_size
;
2967 if (chunk_size
> MAX_CHUNK_SIZE
) {
2968 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2969 chunk_size
, MAX_CHUNK_SIZE
);
2973 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2975 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2976 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2979 if (chunk_size
< PAGE_SIZE
) {
2980 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2981 chunk_size
, PAGE_SIZE
);
2985 /* devices must have minimum size of one chunk */
2986 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2987 if (test_bit(Faulty
, &rdev
->flags
))
2989 if (rdev
->size
< chunk_size
/ 1024) {
2991 "md: Dev %s smaller than chunk_size:"
2993 bdevname(rdev
->bdev
,b
),
2994 (unsigned long long)rdev
->size
,
3002 if (mddev
->level
!= LEVEL_NONE
)
3003 request_module("md-level-%d", mddev
->level
);
3004 else if (mddev
->clevel
[0])
3005 request_module("md-%s", mddev
->clevel
);
3009 * Drop all container device buffers, from now on
3010 * the only valid external interface is through the md
3012 * Also find largest hardsector size
3014 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3015 if (test_bit(Faulty
, &rdev
->flags
))
3017 sync_blockdev(rdev
->bdev
);
3018 invalidate_bdev(rdev
->bdev
, 0);
3021 md_probe(mddev
->unit
, NULL
, NULL
);
3022 disk
= mddev
->gendisk
;
3026 spin_lock(&pers_lock
);
3027 pers
= find_pers(mddev
->level
, mddev
->clevel
);
3028 if (!pers
|| !try_module_get(pers
->owner
)) {
3029 spin_unlock(&pers_lock
);
3030 if (mddev
->level
!= LEVEL_NONE
)
3031 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
3034 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
3039 spin_unlock(&pers_lock
);
3040 mddev
->level
= pers
->level
;
3041 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3043 if (mddev
->reshape_position
!= MaxSector
&&
3044 pers
->start_reshape
== NULL
) {
3045 /* This personality cannot handle reshaping... */
3047 module_put(pers
->owner
);
3051 mddev
->recovery
= 0;
3052 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
3053 mddev
->barriers_work
= 1;
3054 mddev
->ok_start_degraded
= start_dirty_degraded
;
3057 mddev
->ro
= 2; /* read-only, but switch on first write */
3059 err
= mddev
->pers
->run(mddev
);
3060 if (!err
&& mddev
->pers
->sync_request
) {
3061 err
= bitmap_create(mddev
);
3063 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
3064 mdname(mddev
), err
);
3065 mddev
->pers
->stop(mddev
);
3069 printk(KERN_ERR
"md: pers->run() failed ...\n");
3070 module_put(mddev
->pers
->owner
);
3072 bitmap_destroy(mddev
);
3075 if (mddev
->pers
->sync_request
)
3076 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
3077 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
3080 atomic_set(&mddev
->writes_pending
,0);
3081 mddev
->safemode
= 0;
3082 mddev
->safemode_timer
.function
= md_safemode_timeout
;
3083 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
3084 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
3087 ITERATE_RDEV(mddev
,rdev
,tmp
)
3088 if (rdev
->raid_disk
>= 0) {
3090 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3091 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
3094 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3095 md_wakeup_thread(mddev
->thread
);
3097 if (mddev
->sb_dirty
)
3098 md_update_sb(mddev
);
3100 set_capacity(disk
, mddev
->array_size
<<1);
3102 /* If we call blk_queue_make_request here, it will
3103 * re-initialise max_sectors etc which may have been
3104 * refined inside -> run. So just set the bits we need to set.
3105 * Most initialisation happended when we called
3106 * blk_queue_make_request(..., md_fail_request)
3109 mddev
->queue
->queuedata
= mddev
;
3110 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
3112 /* If there is a partially-recovered drive we need to
3113 * start recovery here. If we leave it to md_check_recovery,
3114 * it will remove the drives and not do the right thing
3116 if (mddev
->degraded
) {
3117 struct list_head
*rtmp
;
3119 ITERATE_RDEV(mddev
,rdev
,rtmp
)
3120 if (rdev
->raid_disk
>= 0 &&
3121 !test_bit(In_sync
, &rdev
->flags
) &&
3122 !test_bit(Faulty
, &rdev
->flags
))
3123 /* complete an interrupted recovery */
3125 if (spares
&& mddev
->pers
->sync_request
) {
3126 mddev
->recovery
= 0;
3127 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
3128 mddev
->sync_thread
= md_register_thread(md_do_sync
,
3131 if (!mddev
->sync_thread
) {
3132 printk(KERN_ERR
"%s: could not start resync"
3135 /* leave the spares where they are, it shouldn't hurt */
3136 mddev
->recovery
= 0;
3138 md_wakeup_thread(mddev
->sync_thread
);
3143 md_new_event(mddev
);
3147 static int restart_array(mddev_t
*mddev
)
3149 struct gendisk
*disk
= mddev
->gendisk
;
3153 * Complain if it has no devices
3156 if (list_empty(&mddev
->disks
))
3164 mddev
->safemode
= 0;
3166 set_disk_ro(disk
, 0);
3168 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
3171 * Kick recovery or resync if necessary
3173 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3174 md_wakeup_thread(mddev
->thread
);
3175 md_wakeup_thread(mddev
->sync_thread
);
3184 /* similar to deny_write_access, but accounts for our holding a reference
3185 * to the file ourselves */
3186 static int deny_bitmap_write_access(struct file
* file
)
3188 struct inode
*inode
= file
->f_mapping
->host
;
3190 spin_lock(&inode
->i_lock
);
3191 if (atomic_read(&inode
->i_writecount
) > 1) {
3192 spin_unlock(&inode
->i_lock
);
3195 atomic_set(&inode
->i_writecount
, -1);
3196 spin_unlock(&inode
->i_lock
);
3201 static void restore_bitmap_write_access(struct file
*file
)
3203 struct inode
*inode
= file
->f_mapping
->host
;
3205 spin_lock(&inode
->i_lock
);
3206 atomic_set(&inode
->i_writecount
, 1);
3207 spin_unlock(&inode
->i_lock
);
3211 * 0 - completely stop and dis-assemble array
3212 * 1 - switch to readonly
3213 * 2 - stop but do not disassemble array
3215 static int do_md_stop(mddev_t
* mddev
, int mode
)
3218 struct gendisk
*disk
= mddev
->gendisk
;
3221 if (atomic_read(&mddev
->active
)>2) {
3222 printk("md: %s still in use.\n",mdname(mddev
));
3226 if (mddev
->sync_thread
) {
3227 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3228 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3229 md_unregister_thread(mddev
->sync_thread
);
3230 mddev
->sync_thread
= NULL
;
3233 del_timer_sync(&mddev
->safemode_timer
);
3235 invalidate_partition(disk
, 0);
3238 case 1: /* readonly */
3244 case 0: /* disassemble */
3246 bitmap_flush(mddev
);
3247 md_super_wait(mddev
);
3249 set_disk_ro(disk
, 0);
3250 blk_queue_make_request(mddev
->queue
, md_fail_request
);
3251 mddev
->pers
->stop(mddev
);
3252 if (mddev
->pers
->sync_request
)
3253 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3255 module_put(mddev
->pers
->owner
);
3260 if (!mddev
->in_sync
|| mddev
->sb_dirty
) {
3261 /* mark array as shutdown cleanly */
3263 md_update_sb(mddev
);
3266 set_disk_ro(disk
, 1);
3267 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3271 * Free resources if final stop
3275 struct list_head
*tmp
;
3276 struct gendisk
*disk
;
3277 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
3279 bitmap_destroy(mddev
);
3280 if (mddev
->bitmap_file
) {
3281 restore_bitmap_write_access(mddev
->bitmap_file
);
3282 fput(mddev
->bitmap_file
);
3283 mddev
->bitmap_file
= NULL
;
3285 mddev
->bitmap_offset
= 0;
3287 ITERATE_RDEV(mddev
,rdev
,tmp
)
3288 if (rdev
->raid_disk
>= 0) {
3290 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3291 sysfs_remove_link(&mddev
->kobj
, nm
);
3294 export_array(mddev
);
3296 mddev
->array_size
= 0;
3298 mddev
->raid_disks
= 0;
3299 mddev
->recovery_cp
= 0;
3301 disk
= mddev
->gendisk
;
3303 set_capacity(disk
, 0);
3305 } else if (mddev
->pers
)
3306 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
3309 md_new_event(mddev
);
3314 static void autorun_array(mddev_t
*mddev
)
3317 struct list_head
*tmp
;
3320 if (list_empty(&mddev
->disks
))
3323 printk(KERN_INFO
"md: running: ");
3325 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3326 char b
[BDEVNAME_SIZE
];
3327 printk("<%s>", bdevname(rdev
->bdev
,b
));
3331 err
= do_md_run (mddev
);
3333 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
3334 do_md_stop (mddev
, 0);
3339 * lets try to run arrays based on all disks that have arrived
3340 * until now. (those are in pending_raid_disks)
3342 * the method: pick the first pending disk, collect all disks with
3343 * the same UUID, remove all from the pending list and put them into
3344 * the 'same_array' list. Then order this list based on superblock
3345 * update time (freshest comes first), kick out 'old' disks and
3346 * compare superblocks. If everything's fine then run it.
3348 * If "unit" is allocated, then bump its reference count
3350 static void autorun_devices(int part
)
3352 struct list_head
*tmp
;
3353 mdk_rdev_t
*rdev0
, *rdev
;
3355 char b
[BDEVNAME_SIZE
];
3357 printk(KERN_INFO
"md: autorun ...\n");
3358 while (!list_empty(&pending_raid_disks
)) {
3360 LIST_HEAD(candidates
);
3361 rdev0
= list_entry(pending_raid_disks
.next
,
3362 mdk_rdev_t
, same_set
);
3364 printk(KERN_INFO
"md: considering %s ...\n",
3365 bdevname(rdev0
->bdev
,b
));
3366 INIT_LIST_HEAD(&candidates
);
3367 ITERATE_RDEV_PENDING(rdev
,tmp
)
3368 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
3369 printk(KERN_INFO
"md: adding %s ...\n",
3370 bdevname(rdev
->bdev
,b
));
3371 list_move(&rdev
->same_set
, &candidates
);
3374 * now we have a set of devices, with all of them having
3375 * mostly sane superblocks. It's time to allocate the
3378 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
3379 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
3380 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
3384 dev
= MKDEV(mdp_major
,
3385 rdev0
->preferred_minor
<< MdpMinorShift
);
3387 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
3389 md_probe(dev
, NULL
, NULL
);
3390 mddev
= mddev_find(dev
);
3393 "md: cannot allocate memory for md drive.\n");
3396 if (mddev_lock(mddev
))
3397 printk(KERN_WARNING
"md: %s locked, cannot run\n",
3399 else if (mddev
->raid_disks
|| mddev
->major_version
3400 || !list_empty(&mddev
->disks
)) {
3402 "md: %s already running, cannot run %s\n",
3403 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
3404 mddev_unlock(mddev
);
3406 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
3407 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
3408 list_del_init(&rdev
->same_set
);
3409 if (bind_rdev_to_array(rdev
, mddev
))
3412 autorun_array(mddev
);
3413 mddev_unlock(mddev
);
3415 /* on success, candidates will be empty, on error
3418 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
3422 printk(KERN_INFO
"md: ... autorun DONE.\n");
3426 * import RAID devices based on one partition
3427 * if possible, the array gets run as well.
3430 static int autostart_array(dev_t startdev
)
3432 char b
[BDEVNAME_SIZE
];
3433 int err
= -EINVAL
, i
;
3434 mdp_super_t
*sb
= NULL
;
3435 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
3437 start_rdev
= md_import_device(startdev
, 0, 0);
3438 if (IS_ERR(start_rdev
))
3442 /* NOTE: this can only work for 0.90.0 superblocks */
3443 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
3444 if (sb
->major_version
!= 0 ||
3445 sb
->minor_version
!= 90 ) {
3446 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
3447 export_rdev(start_rdev
);
3451 if (test_bit(Faulty
, &start_rdev
->flags
)) {
3453 "md: can not autostart based on faulty %s!\n",
3454 bdevname(start_rdev
->bdev
,b
));
3455 export_rdev(start_rdev
);
3458 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
3460 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
3461 mdp_disk_t
*desc
= sb
->disks
+ i
;
3462 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
3466 if (dev
== startdev
)
3468 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
3470 rdev
= md_import_device(dev
, 0, 0);
3474 list_add(&rdev
->same_set
, &pending_raid_disks
);
3478 * possibly return codes
3486 static int get_version(void __user
* arg
)
3490 ver
.major
= MD_MAJOR_VERSION
;
3491 ver
.minor
= MD_MINOR_VERSION
;
3492 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
3494 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
3500 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
3502 mdu_array_info_t info
;
3503 int nr
,working
,active
,failed
,spare
;
3505 struct list_head
*tmp
;
3507 nr
=working
=active
=failed
=spare
=0;
3508 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3510 if (test_bit(Faulty
, &rdev
->flags
))
3514 if (test_bit(In_sync
, &rdev
->flags
))
3521 info
.major_version
= mddev
->major_version
;
3522 info
.minor_version
= mddev
->minor_version
;
3523 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
3524 info
.ctime
= mddev
->ctime
;
3525 info
.level
= mddev
->level
;
3526 info
.size
= mddev
->size
;
3527 if (info
.size
!= mddev
->size
) /* overflow */
3530 info
.raid_disks
= mddev
->raid_disks
;
3531 info
.md_minor
= mddev
->md_minor
;
3532 info
.not_persistent
= !mddev
->persistent
;
3534 info
.utime
= mddev
->utime
;
3537 info
.state
= (1<<MD_SB_CLEAN
);
3538 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3539 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
3540 info
.active_disks
= active
;
3541 info
.working_disks
= working
;
3542 info
.failed_disks
= failed
;
3543 info
.spare_disks
= spare
;
3545 info
.layout
= mddev
->layout
;
3546 info
.chunk_size
= mddev
->chunk_size
;
3548 if (copy_to_user(arg
, &info
, sizeof(info
)))
3554 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
3556 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
3557 char *ptr
, *buf
= NULL
;
3560 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
3564 /* bitmap disabled, zero the first byte and copy out */
3565 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
3566 file
->pathname
[0] = '\0';
3570 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
3574 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
3578 strcpy(file
->pathname
, ptr
);
3582 if (copy_to_user(arg
, file
, sizeof(*file
)))
3590 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
3592 mdu_disk_info_t info
;
3596 if (copy_from_user(&info
, arg
, sizeof(info
)))
3601 rdev
= find_rdev_nr(mddev
, nr
);
3603 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
3604 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
3605 info
.raid_disk
= rdev
->raid_disk
;
3607 if (test_bit(Faulty
, &rdev
->flags
))
3608 info
.state
|= (1<<MD_DISK_FAULTY
);
3609 else if (test_bit(In_sync
, &rdev
->flags
)) {
3610 info
.state
|= (1<<MD_DISK_ACTIVE
);
3611 info
.state
|= (1<<MD_DISK_SYNC
);
3613 if (test_bit(WriteMostly
, &rdev
->flags
))
3614 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
3616 info
.major
= info
.minor
= 0;
3617 info
.raid_disk
= -1;
3618 info
.state
= (1<<MD_DISK_REMOVED
);
3621 if (copy_to_user(arg
, &info
, sizeof(info
)))
3627 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
3629 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
3631 dev_t dev
= MKDEV(info
->major
,info
->minor
);
3633 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
3636 if (!mddev
->raid_disks
) {
3638 /* expecting a device which has a superblock */
3639 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
3642 "md: md_import_device returned %ld\n",
3644 return PTR_ERR(rdev
);
3646 if (!list_empty(&mddev
->disks
)) {
3647 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3648 mdk_rdev_t
, same_set
);
3649 int err
= super_types
[mddev
->major_version
]
3650 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3653 "md: %s has different UUID to %s\n",
3654 bdevname(rdev
->bdev
,b
),
3655 bdevname(rdev0
->bdev
,b2
));
3660 err
= bind_rdev_to_array(rdev
, mddev
);
3667 * add_new_disk can be used once the array is assembled
3668 * to add "hot spares". They must already have a superblock
3673 if (!mddev
->pers
->hot_add_disk
) {
3675 "%s: personality does not support diskops!\n",
3679 if (mddev
->persistent
)
3680 rdev
= md_import_device(dev
, mddev
->major_version
,
3681 mddev
->minor_version
);
3683 rdev
= md_import_device(dev
, -1, -1);
3686 "md: md_import_device returned %ld\n",
3688 return PTR_ERR(rdev
);
3690 /* set save_raid_disk if appropriate */
3691 if (!mddev
->persistent
) {
3692 if (info
->state
& (1<<MD_DISK_SYNC
) &&
3693 info
->raid_disk
< mddev
->raid_disks
)
3694 rdev
->raid_disk
= info
->raid_disk
;
3696 rdev
->raid_disk
= -1;
3698 super_types
[mddev
->major_version
].
3699 validate_super(mddev
, rdev
);
3700 rdev
->saved_raid_disk
= rdev
->raid_disk
;
3702 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
3703 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3704 set_bit(WriteMostly
, &rdev
->flags
);
3706 rdev
->raid_disk
= -1;
3707 err
= bind_rdev_to_array(rdev
, mddev
);
3708 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
3709 /* If there is hot_add_disk but no hot_remove_disk
3710 * then added disks for geometry changes,
3711 * and should be added immediately.
3713 super_types
[mddev
->major_version
].
3714 validate_super(mddev
, rdev
);
3715 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
3717 unbind_rdev_from_array(rdev
);
3722 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3723 md_wakeup_thread(mddev
->thread
);
3727 /* otherwise, add_new_disk is only allowed
3728 * for major_version==0 superblocks
3730 if (mddev
->major_version
!= 0) {
3731 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
3736 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
3738 rdev
= md_import_device (dev
, -1, 0);
3741 "md: error, md_import_device() returned %ld\n",
3743 return PTR_ERR(rdev
);
3745 rdev
->desc_nr
= info
->number
;
3746 if (info
->raid_disk
< mddev
->raid_disks
)
3747 rdev
->raid_disk
= info
->raid_disk
;
3749 rdev
->raid_disk
= -1;
3753 if (rdev
->raid_disk
< mddev
->raid_disks
)
3754 if (info
->state
& (1<<MD_DISK_SYNC
))
3755 set_bit(In_sync
, &rdev
->flags
);
3757 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
3758 set_bit(WriteMostly
, &rdev
->flags
);
3760 if (!mddev
->persistent
) {
3761 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
3762 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3764 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3765 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3767 err
= bind_rdev_to_array(rdev
, mddev
);
3777 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
3779 char b
[BDEVNAME_SIZE
];
3785 rdev
= find_rdev(mddev
, dev
);
3789 if (rdev
->raid_disk
>= 0)
3792 kick_rdev_from_array(rdev
);
3793 md_update_sb(mddev
);
3794 md_new_event(mddev
);
3798 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3799 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3803 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3805 char b
[BDEVNAME_SIZE
];
3813 if (mddev
->major_version
!= 0) {
3814 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3815 " version-0 superblocks.\n",
3819 if (!mddev
->pers
->hot_add_disk
) {
3821 "%s: personality does not support diskops!\n",
3826 rdev
= md_import_device (dev
, -1, 0);
3829 "md: error, md_import_device() returned %ld\n",
3834 if (mddev
->persistent
)
3835 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3838 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3840 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3843 if (test_bit(Faulty
, &rdev
->flags
)) {
3845 "md: can not hot-add faulty %s disk to %s!\n",
3846 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3850 clear_bit(In_sync
, &rdev
->flags
);
3852 err
= bind_rdev_to_array(rdev
, mddev
);
3857 * The rest should better be atomic, we can have disk failures
3858 * noticed in interrupt contexts ...
3861 if (rdev
->desc_nr
== mddev
->max_disks
) {
3862 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3865 goto abort_unbind_export
;
3868 rdev
->raid_disk
= -1;
3870 md_update_sb(mddev
);
3873 * Kick recovery, maybe this spare has to be added to the
3874 * array immediately.
3876 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3877 md_wakeup_thread(mddev
->thread
);
3878 md_new_event(mddev
);
3881 abort_unbind_export
:
3882 unbind_rdev_from_array(rdev
);
3889 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3894 if (!mddev
->pers
->quiesce
)
3896 if (mddev
->recovery
|| mddev
->sync_thread
)
3898 /* we should be able to change the bitmap.. */
3904 return -EEXIST
; /* cannot add when bitmap is present */
3905 mddev
->bitmap_file
= fget(fd
);
3907 if (mddev
->bitmap_file
== NULL
) {
3908 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3913 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3915 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3917 fput(mddev
->bitmap_file
);
3918 mddev
->bitmap_file
= NULL
;
3921 mddev
->bitmap_offset
= 0; /* file overrides offset */
3922 } else if (mddev
->bitmap
== NULL
)
3923 return -ENOENT
; /* cannot remove what isn't there */
3926 mddev
->pers
->quiesce(mddev
, 1);
3928 err
= bitmap_create(mddev
);
3929 if (fd
< 0 || err
) {
3930 bitmap_destroy(mddev
);
3931 fd
= -1; /* make sure to put the file */
3933 mddev
->pers
->quiesce(mddev
, 0);
3936 if (mddev
->bitmap_file
) {
3937 restore_bitmap_write_access(mddev
->bitmap_file
);
3938 fput(mddev
->bitmap_file
);
3940 mddev
->bitmap_file
= NULL
;
3947 * set_array_info is used two different ways
3948 * The original usage is when creating a new array.
3949 * In this usage, raid_disks is > 0 and it together with
3950 * level, size, not_persistent,layout,chunksize determine the
3951 * shape of the array.
3952 * This will always create an array with a type-0.90.0 superblock.
3953 * The newer usage is when assembling an array.
3954 * In this case raid_disks will be 0, and the major_version field is
3955 * use to determine which style super-blocks are to be found on the devices.
3956 * The minor and patch _version numbers are also kept incase the
3957 * super_block handler wishes to interpret them.
3959 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3962 if (info
->raid_disks
== 0) {
3963 /* just setting version number for superblock loading */
3964 if (info
->major_version
< 0 ||
3965 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3966 super_types
[info
->major_version
].name
== NULL
) {
3967 /* maybe try to auto-load a module? */
3969 "md: superblock version %d not known\n",
3970 info
->major_version
);
3973 mddev
->major_version
= info
->major_version
;
3974 mddev
->minor_version
= info
->minor_version
;
3975 mddev
->patch_version
= info
->patch_version
;
3978 mddev
->major_version
= MD_MAJOR_VERSION
;
3979 mddev
->minor_version
= MD_MINOR_VERSION
;
3980 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3981 mddev
->ctime
= get_seconds();
3983 mddev
->level
= info
->level
;
3984 mddev
->clevel
[0] = 0;
3985 mddev
->size
= info
->size
;
3986 mddev
->raid_disks
= info
->raid_disks
;
3987 /* don't set md_minor, it is determined by which /dev/md* was
3990 if (info
->state
& (1<<MD_SB_CLEAN
))
3991 mddev
->recovery_cp
= MaxSector
;
3993 mddev
->recovery_cp
= 0;
3994 mddev
->persistent
= ! info
->not_persistent
;
3996 mddev
->layout
= info
->layout
;
3997 mddev
->chunk_size
= info
->chunk_size
;
3999 mddev
->max_disks
= MD_SB_DISKS
;
4001 mddev
->sb_dirty
= 1;
4003 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
4004 mddev
->bitmap_offset
= 0;
4006 mddev
->reshape_position
= MaxSector
;
4009 * Generate a 128 bit UUID
4011 get_random_bytes(mddev
->uuid
, 16);
4013 mddev
->new_level
= mddev
->level
;
4014 mddev
->new_chunk
= mddev
->chunk_size
;
4015 mddev
->new_layout
= mddev
->layout
;
4016 mddev
->delta_disks
= 0;
4021 static int update_size(mddev_t
*mddev
, unsigned long size
)
4025 struct list_head
*tmp
;
4026 int fit
= (size
== 0);
4028 if (mddev
->pers
->resize
== NULL
)
4030 /* The "size" is the amount of each device that is used.
4031 * This can only make sense for arrays with redundancy.
4032 * linear and raid0 always use whatever space is available
4033 * We can only consider changing the size if no resync
4034 * or reconstruction is happening, and if the new size
4035 * is acceptable. It must fit before the sb_offset or,
4036 * if that is <data_offset, it must fit before the
4037 * size of each device.
4038 * If size is zero, we find the largest size that fits.
4040 if (mddev
->sync_thread
)
4042 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4044 if (rdev
->sb_offset
> rdev
->data_offset
)
4045 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
4047 avail
= get_capacity(rdev
->bdev
->bd_disk
)
4048 - rdev
->data_offset
;
4049 if (fit
&& (size
== 0 || size
> avail
/2))
4051 if (avail
< ((sector_t
)size
<< 1))
4054 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
4056 struct block_device
*bdev
;
4058 bdev
= bdget_disk(mddev
->gendisk
, 0);
4060 mutex_lock(&bdev
->bd_inode
->i_mutex
);
4061 i_size_write(bdev
->bd_inode
, (loff_t
)mddev
->array_size
<< 10);
4062 mutex_unlock(&bdev
->bd_inode
->i_mutex
);
4069 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
4072 /* change the number of raid disks */
4073 if (mddev
->pers
->check_reshape
== NULL
)
4075 if (raid_disks
<= 0 ||
4076 raid_disks
>= mddev
->max_disks
)
4078 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
4080 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
4082 rv
= mddev
->pers
->check_reshape(mddev
);
4088 * update_array_info is used to change the configuration of an
4090 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4091 * fields in the info are checked against the array.
4092 * Any differences that cannot be handled will cause an error.
4093 * Normally, only one change can be managed at a time.
4095 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
4101 /* calculate expected state,ignoring low bits */
4102 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4103 state
|= (1 << MD_SB_BITMAP_PRESENT
);
4105 if (mddev
->major_version
!= info
->major_version
||
4106 mddev
->minor_version
!= info
->minor_version
||
4107 /* mddev->patch_version != info->patch_version || */
4108 mddev
->ctime
!= info
->ctime
||
4109 mddev
->level
!= info
->level
||
4110 /* mddev->layout != info->layout || */
4111 !mddev
->persistent
!= info
->not_persistent
||
4112 mddev
->chunk_size
!= info
->chunk_size
||
4113 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4114 ((state
^info
->state
) & 0xfffffe00)
4117 /* Check there is only one change */
4118 if (info
->size
>= 0 && mddev
->size
!= info
->size
) cnt
++;
4119 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
4120 if (mddev
->layout
!= info
->layout
) cnt
++;
4121 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
4122 if (cnt
== 0) return 0;
4123 if (cnt
> 1) return -EINVAL
;
4125 if (mddev
->layout
!= info
->layout
) {
4127 * we don't need to do anything at the md level, the
4128 * personality will take care of it all.
4130 if (mddev
->pers
->reconfig
== NULL
)
4133 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
4135 if (info
->size
>= 0 && mddev
->size
!= info
->size
)
4136 rv
= update_size(mddev
, info
->size
);
4138 if (mddev
->raid_disks
!= info
->raid_disks
)
4139 rv
= update_raid_disks(mddev
, info
->raid_disks
);
4141 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
4142 if (mddev
->pers
->quiesce
== NULL
)
4144 if (mddev
->recovery
|| mddev
->sync_thread
)
4146 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
4147 /* add the bitmap */
4150 if (mddev
->default_bitmap_offset
== 0)
4152 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
4153 mddev
->pers
->quiesce(mddev
, 1);
4154 rv
= bitmap_create(mddev
);
4156 bitmap_destroy(mddev
);
4157 mddev
->pers
->quiesce(mddev
, 0);
4159 /* remove the bitmap */
4162 if (mddev
->bitmap
->file
)
4164 mddev
->pers
->quiesce(mddev
, 1);
4165 bitmap_destroy(mddev
);
4166 mddev
->pers
->quiesce(mddev
, 0);
4167 mddev
->bitmap_offset
= 0;
4170 md_update_sb(mddev
);
4174 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
4178 if (mddev
->pers
== NULL
)
4181 rdev
= find_rdev(mddev
, dev
);
4185 md_error(mddev
, rdev
);
4189 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
4191 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
4195 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
4199 static int md_ioctl(struct inode
*inode
, struct file
*file
,
4200 unsigned int cmd
, unsigned long arg
)
4203 void __user
*argp
= (void __user
*)arg
;
4204 mddev_t
*mddev
= NULL
;
4206 if (!capable(CAP_SYS_ADMIN
))
4210 * Commands dealing with the RAID driver but not any
4216 err
= get_version(argp
);
4219 case PRINT_RAID_DEBUG
:
4227 autostart_arrays(arg
);
4234 * Commands creating/starting a new array:
4237 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4245 if (cmd
== START_ARRAY
) {
4246 /* START_ARRAY doesn't need to lock the array as autostart_array
4247 * does the locking, and it could even be a different array
4252 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
4253 "This will not be supported beyond July 2006\n",
4254 current
->comm
, current
->pid
);
4257 err
= autostart_array(new_decode_dev(arg
));
4259 printk(KERN_WARNING
"md: autostart failed!\n");
4265 err
= mddev_lock(mddev
);
4268 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4275 case SET_ARRAY_INFO
:
4277 mdu_array_info_t info
;
4279 memset(&info
, 0, sizeof(info
));
4280 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
4285 err
= update_array_info(mddev
, &info
);
4287 printk(KERN_WARNING
"md: couldn't update"
4288 " array info. %d\n", err
);
4293 if (!list_empty(&mddev
->disks
)) {
4295 "md: array %s already has disks!\n",
4300 if (mddev
->raid_disks
) {
4302 "md: array %s already initialised!\n",
4307 err
= set_array_info(mddev
, &info
);
4309 printk(KERN_WARNING
"md: couldn't set"
4310 " array info. %d\n", err
);
4320 * Commands querying/configuring an existing array:
4322 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4323 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4324 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
4325 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
4331 * Commands even a read-only array can execute:
4335 case GET_ARRAY_INFO
:
4336 err
= get_array_info(mddev
, argp
);
4339 case GET_BITMAP_FILE
:
4340 err
= get_bitmap_file(mddev
, argp
);
4344 err
= get_disk_info(mddev
, argp
);
4347 case RESTART_ARRAY_RW
:
4348 err
= restart_array(mddev
);
4352 err
= do_md_stop (mddev
, 0);
4356 err
= do_md_stop (mddev
, 1);
4360 * We have a problem here : there is no easy way to give a CHS
4361 * virtual geometry. We currently pretend that we have a 2 heads
4362 * 4 sectors (with a BIG number of cylinders...). This drives
4363 * dosfs just mad... ;-)
4368 * The remaining ioctls are changing the state of the
4369 * superblock, so we do not allow them on read-only arrays.
4370 * However non-MD ioctls (e.g. get-size) will still come through
4371 * here and hit the 'default' below, so only disallow
4372 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4374 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
4375 mddev
->ro
&& mddev
->pers
) {
4376 if (mddev
->ro
== 2) {
4378 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4379 md_wakeup_thread(mddev
->thread
);
4391 mdu_disk_info_t info
;
4392 if (copy_from_user(&info
, argp
, sizeof(info
)))
4395 err
= add_new_disk(mddev
, &info
);
4399 case HOT_REMOVE_DISK
:
4400 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
4404 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
4407 case SET_DISK_FAULTY
:
4408 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
4412 err
= do_md_run (mddev
);
4415 case SET_BITMAP_FILE
:
4416 err
= set_bitmap_file(mddev
, (int)arg
);
4426 mddev_unlock(mddev
);
4436 static int md_open(struct inode
*inode
, struct file
*file
)
4439 * Succeed if we can lock the mddev, which confirms that
4440 * it isn't being stopped right now.
4442 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4445 if ((err
= mddev_lock(mddev
)))
4450 mddev_unlock(mddev
);
4452 check_disk_change(inode
->i_bdev
);
4457 static int md_release(struct inode
*inode
, struct file
* file
)
4459 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
4468 static int md_media_changed(struct gendisk
*disk
)
4470 mddev_t
*mddev
= disk
->private_data
;
4472 return mddev
->changed
;
4475 static int md_revalidate(struct gendisk
*disk
)
4477 mddev_t
*mddev
= disk
->private_data
;
4482 static struct block_device_operations md_fops
=
4484 .owner
= THIS_MODULE
,
4486 .release
= md_release
,
4488 .getgeo
= md_getgeo
,
4489 .media_changed
= md_media_changed
,
4490 .revalidate_disk
= md_revalidate
,
4493 static int md_thread(void * arg
)
4495 mdk_thread_t
*thread
= arg
;
4498 * md_thread is a 'system-thread', it's priority should be very
4499 * high. We avoid resource deadlocks individually in each
4500 * raid personality. (RAID5 does preallocation) We also use RR and
4501 * the very same RT priority as kswapd, thus we will never get
4502 * into a priority inversion deadlock.
4504 * we definitely have to have equal or higher priority than
4505 * bdflush, otherwise bdflush will deadlock if there are too
4506 * many dirty RAID5 blocks.
4509 allow_signal(SIGKILL
);
4510 while (!kthread_should_stop()) {
4512 /* We need to wait INTERRUPTIBLE so that
4513 * we don't add to the load-average.
4514 * That means we need to be sure no signals are
4517 if (signal_pending(current
))
4518 flush_signals(current
);
4520 wait_event_interruptible_timeout
4522 test_bit(THREAD_WAKEUP
, &thread
->flags
)
4523 || kthread_should_stop(),
4527 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
4529 thread
->run(thread
->mddev
);
4535 void md_wakeup_thread(mdk_thread_t
*thread
)
4538 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
4539 set_bit(THREAD_WAKEUP
, &thread
->flags
);
4540 wake_up(&thread
->wqueue
);
4544 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
4547 mdk_thread_t
*thread
;
4549 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
4553 init_waitqueue_head(&thread
->wqueue
);
4556 thread
->mddev
= mddev
;
4557 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
4558 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
4559 if (IS_ERR(thread
->tsk
)) {
4566 void md_unregister_thread(mdk_thread_t
*thread
)
4568 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
4570 kthread_stop(thread
->tsk
);
4574 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
4581 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
4584 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4586 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4587 __builtin_return_address(0),__builtin_return_address(1),
4588 __builtin_return_address(2),__builtin_return_address(3));
4590 if (!mddev
->pers
->error_handler
)
4592 mddev
->pers
->error_handler(mddev
,rdev
);
4593 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4594 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4595 md_wakeup_thread(mddev
->thread
);
4596 md_new_event_inintr(mddev
);
4599 /* seq_file implementation /proc/mdstat */
4601 static void status_unused(struct seq_file
*seq
)
4605 struct list_head
*tmp
;
4607 seq_printf(seq
, "unused devices: ");
4609 ITERATE_RDEV_PENDING(rdev
,tmp
) {
4610 char b
[BDEVNAME_SIZE
];
4612 seq_printf(seq
, "%s ",
4613 bdevname(rdev
->bdev
,b
));
4616 seq_printf(seq
, "<none>");
4618 seq_printf(seq
, "\n");
4622 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
4624 sector_t max_blocks
, resync
, res
;
4625 unsigned long dt
, db
, rt
;
4627 unsigned int per_milli
;
4629 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
4631 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4632 max_blocks
= mddev
->resync_max_sectors
>> 1;
4634 max_blocks
= mddev
->size
;
4637 * Should not happen.
4643 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4644 * in a sector_t, and (max_blocks>>scale) will fit in a
4645 * u32, as those are the requirements for sector_div.
4646 * Thus 'scale' must be at least 10
4649 if (sizeof(sector_t
) > sizeof(unsigned long)) {
4650 while ( max_blocks
/2 > (1ULL<<(scale
+32)))
4653 res
= (resync
>>scale
)*1000;
4654 sector_div(res
, (u32
)((max_blocks
>>scale
)+1));
4658 int i
, x
= per_milli
/50, y
= 20-x
;
4659 seq_printf(seq
, "[");
4660 for (i
= 0; i
< x
; i
++)
4661 seq_printf(seq
, "=");
4662 seq_printf(seq
, ">");
4663 for (i
= 0; i
< y
; i
++)
4664 seq_printf(seq
, ".");
4665 seq_printf(seq
, "] ");
4667 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
4668 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
4670 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
4671 "resync" : "recovery")),
4672 per_milli
/10, per_milli
% 10,
4673 (unsigned long long) resync
,
4674 (unsigned long long) max_blocks
);
4677 * We do not want to overflow, so the order of operands and
4678 * the * 100 / 100 trick are important. We do a +1 to be
4679 * safe against division by zero. We only estimate anyway.
4681 * dt: time from mark until now
4682 * db: blocks written from mark until now
4683 * rt: remaining time
4685 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
4687 db
= resync
- (mddev
->resync_mark_cnt
/2);
4688 rt
= (dt
* ((unsigned long)(max_blocks
-resync
) / (db
/100+1)))/100;
4690 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
4692 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
4695 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4697 struct list_head
*tmp
;
4707 spin_lock(&all_mddevs_lock
);
4708 list_for_each(tmp
,&all_mddevs
)
4710 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
4712 spin_unlock(&all_mddevs_lock
);
4715 spin_unlock(&all_mddevs_lock
);
4717 return (void*)2;/* tail */
4721 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4723 struct list_head
*tmp
;
4724 mddev_t
*next_mddev
, *mddev
= v
;
4730 spin_lock(&all_mddevs_lock
);
4732 tmp
= all_mddevs
.next
;
4734 tmp
= mddev
->all_mddevs
.next
;
4735 if (tmp
!= &all_mddevs
)
4736 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
4738 next_mddev
= (void*)2;
4741 spin_unlock(&all_mddevs_lock
);
4749 static void md_seq_stop(struct seq_file
*seq
, void *v
)
4753 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
4757 struct mdstat_info
{
4761 static int md_seq_show(struct seq_file
*seq
, void *v
)
4765 struct list_head
*tmp2
;
4767 struct mdstat_info
*mi
= seq
->private;
4768 struct bitmap
*bitmap
;
4770 if (v
== (void*)1) {
4771 struct mdk_personality
*pers
;
4772 seq_printf(seq
, "Personalities : ");
4773 spin_lock(&pers_lock
);
4774 list_for_each_entry(pers
, &pers_list
, list
)
4775 seq_printf(seq
, "[%s] ", pers
->name
);
4777 spin_unlock(&pers_lock
);
4778 seq_printf(seq
, "\n");
4779 mi
->event
= atomic_read(&md_event_count
);
4782 if (v
== (void*)2) {
4787 if (mddev_lock(mddev
) < 0)
4790 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4791 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4792 mddev
->pers
? "" : "in");
4795 seq_printf(seq
, " (read-only)");
4797 seq_printf(seq
, "(auto-read-only)");
4798 seq_printf(seq
, " %s", mddev
->pers
->name
);
4802 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4803 char b
[BDEVNAME_SIZE
];
4804 seq_printf(seq
, " %s[%d]",
4805 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4806 if (test_bit(WriteMostly
, &rdev
->flags
))
4807 seq_printf(seq
, "(W)");
4808 if (test_bit(Faulty
, &rdev
->flags
)) {
4809 seq_printf(seq
, "(F)");
4811 } else if (rdev
->raid_disk
< 0)
4812 seq_printf(seq
, "(S)"); /* spare */
4816 if (!list_empty(&mddev
->disks
)) {
4818 seq_printf(seq
, "\n %llu blocks",
4819 (unsigned long long)mddev
->array_size
);
4821 seq_printf(seq
, "\n %llu blocks",
4822 (unsigned long long)size
);
4824 if (mddev
->persistent
) {
4825 if (mddev
->major_version
!= 0 ||
4826 mddev
->minor_version
!= 90) {
4827 seq_printf(seq
," super %d.%d",
4828 mddev
->major_version
,
4829 mddev
->minor_version
);
4832 seq_printf(seq
, " super non-persistent");
4835 mddev
->pers
->status (seq
, mddev
);
4836 seq_printf(seq
, "\n ");
4837 if (mddev
->pers
->sync_request
) {
4838 if (mddev
->curr_resync
> 2) {
4839 status_resync (seq
, mddev
);
4840 seq_printf(seq
, "\n ");
4841 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4842 seq_printf(seq
, "\tresync=DELAYED\n ");
4843 else if (mddev
->recovery_cp
< MaxSector
)
4844 seq_printf(seq
, "\tresync=PENDING\n ");
4847 seq_printf(seq
, "\n ");
4849 if ((bitmap
= mddev
->bitmap
)) {
4850 unsigned long chunk_kb
;
4851 unsigned long flags
;
4852 spin_lock_irqsave(&bitmap
->lock
, flags
);
4853 chunk_kb
= bitmap
->chunksize
>> 10;
4854 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4856 bitmap
->pages
- bitmap
->missing_pages
,
4858 (bitmap
->pages
- bitmap
->missing_pages
)
4859 << (PAGE_SHIFT
- 10),
4860 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4861 chunk_kb
? "KB" : "B");
4863 seq_printf(seq
, ", file: ");
4864 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4865 bitmap
->file
->f_dentry
," \t\n");
4868 seq_printf(seq
, "\n");
4869 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4872 seq_printf(seq
, "\n");
4874 mddev_unlock(mddev
);
4879 static struct seq_operations md_seq_ops
= {
4880 .start
= md_seq_start
,
4881 .next
= md_seq_next
,
4882 .stop
= md_seq_stop
,
4883 .show
= md_seq_show
,
4886 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4889 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4893 error
= seq_open(file
, &md_seq_ops
);
4897 struct seq_file
*p
= file
->private_data
;
4899 mi
->event
= atomic_read(&md_event_count
);
4904 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4906 struct seq_file
*m
= file
->private_data
;
4907 struct mdstat_info
*mi
= m
->private;
4910 return seq_release(inode
, file
);
4913 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4915 struct seq_file
*m
= filp
->private_data
;
4916 struct mdstat_info
*mi
= m
->private;
4919 poll_wait(filp
, &md_event_waiters
, wait
);
4921 /* always allow read */
4922 mask
= POLLIN
| POLLRDNORM
;
4924 if (mi
->event
!= atomic_read(&md_event_count
))
4925 mask
|= POLLERR
| POLLPRI
;
4929 static struct file_operations md_seq_fops
= {
4930 .open
= md_seq_open
,
4932 .llseek
= seq_lseek
,
4933 .release
= md_seq_release
,
4934 .poll
= mdstat_poll
,
4937 int register_md_personality(struct mdk_personality
*p
)
4939 spin_lock(&pers_lock
);
4940 list_add_tail(&p
->list
, &pers_list
);
4941 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4942 spin_unlock(&pers_lock
);
4946 int unregister_md_personality(struct mdk_personality
*p
)
4948 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4949 spin_lock(&pers_lock
);
4950 list_del_init(&p
->list
);
4951 spin_unlock(&pers_lock
);
4955 static int is_mddev_idle(mddev_t
*mddev
)
4958 struct list_head
*tmp
;
4960 unsigned long curr_events
;
4963 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4964 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4965 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4966 disk_stat_read(disk
, sectors
[1]) -
4967 atomic_read(&disk
->sync_io
);
4968 /* The difference between curr_events and last_events
4969 * will be affected by any new non-sync IO (making
4970 * curr_events bigger) and any difference in the amount of
4971 * in-flight syncio (making current_events bigger or smaller)
4972 * The amount in-flight is currently limited to
4973 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4974 * which is at most 4096 sectors.
4975 * These numbers are fairly fragile and should be made
4976 * more robust, probably by enforcing the
4977 * 'window size' that md_do_sync sort-of uses.
4979 * Note: the following is an unsigned comparison.
4981 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4982 rdev
->last_events
= curr_events
;
4989 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4991 /* another "blocks" (512byte) blocks have been synced */
4992 atomic_sub(blocks
, &mddev
->recovery_active
);
4993 wake_up(&mddev
->recovery_wait
);
4995 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4996 md_wakeup_thread(mddev
->thread
);
4997 // stop recovery, signal do_sync ....
5002 /* md_write_start(mddev, bi)
5003 * If we need to update some array metadata (e.g. 'active' flag
5004 * in superblock) before writing, schedule a superblock update
5005 * and wait for it to complete.
5007 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
5009 if (bio_data_dir(bi
) != WRITE
)
5012 BUG_ON(mddev
->ro
== 1);
5013 if (mddev
->ro
== 2) {
5014 /* need to switch to read/write */
5016 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5017 md_wakeup_thread(mddev
->thread
);
5019 atomic_inc(&mddev
->writes_pending
);
5020 if (mddev
->in_sync
) {
5021 spin_lock_irq(&mddev
->write_lock
);
5022 if (mddev
->in_sync
) {
5024 mddev
->sb_dirty
= 3;
5025 md_wakeup_thread(mddev
->thread
);
5027 spin_unlock_irq(&mddev
->write_lock
);
5029 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
5032 void md_write_end(mddev_t
*mddev
)
5034 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
5035 if (mddev
->safemode
== 2)
5036 md_wakeup_thread(mddev
->thread
);
5037 else if (mddev
->safemode_delay
)
5038 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
5042 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
5044 #define SYNC_MARKS 10
5045 #define SYNC_MARK_STEP (3*HZ)
5046 void md_do_sync(mddev_t
*mddev
)
5049 unsigned int currspeed
= 0,
5051 sector_t max_sectors
,j
, io_sectors
;
5052 unsigned long mark
[SYNC_MARKS
];
5053 sector_t mark_cnt
[SYNC_MARKS
];
5055 struct list_head
*tmp
;
5056 sector_t last_check
;
5058 struct list_head
*rtmp
;
5061 /* just incase thread restarts... */
5062 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
5064 if (mddev
->ro
) /* never try to sync a read-only array */
5067 /* we overload curr_resync somewhat here.
5068 * 0 == not engaged in resync at all
5069 * 2 == checking that there is no conflict with another sync
5070 * 1 == like 2, but have yielded to allow conflicting resync to
5072 * other == active in resync - this many blocks
5074 * Before starting a resync we must have set curr_resync to
5075 * 2, and then checked that every "conflicting" array has curr_resync
5076 * less than ours. When we find one that is the same or higher
5077 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5078 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5079 * This will mean we have to start checking from the beginning again.
5084 mddev
->curr_resync
= 2;
5087 if (kthread_should_stop()) {
5088 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5091 ITERATE_MDDEV(mddev2
,tmp
) {
5092 if (mddev2
== mddev
)
5094 if (mddev2
->curr_resync
&&
5095 match_mddev_units(mddev
,mddev2
)) {
5097 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
5098 /* arbitrarily yield */
5099 mddev
->curr_resync
= 1;
5100 wake_up(&resync_wait
);
5102 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
5103 /* no need to wait here, we can wait the next
5104 * time 'round when curr_resync == 2
5107 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
5108 if (!kthread_should_stop() &&
5109 mddev2
->curr_resync
>= mddev
->curr_resync
) {
5110 printk(KERN_INFO
"md: delaying resync of %s"
5111 " until %s has finished resync (they"
5112 " share one or more physical units)\n",
5113 mdname(mddev
), mdname(mddev2
));
5116 finish_wait(&resync_wait
, &wq
);
5119 finish_wait(&resync_wait
, &wq
);
5122 } while (mddev
->curr_resync
< 2);
5125 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5126 /* resync follows the size requested by the personality,
5127 * which defaults to physical size, but can be virtual size
5129 max_sectors
= mddev
->resync_max_sectors
;
5130 mddev
->resync_mismatches
= 0;
5131 /* we don't use the checkpoint if there's a bitmap */
5132 if (!mddev
->bitmap
&&
5133 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
5134 j
= mddev
->recovery_cp
;
5135 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
5136 max_sectors
= mddev
->size
<< 1;
5138 /* recovery follows the physical size of devices */
5139 max_sectors
= mddev
->size
<< 1;
5141 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5142 if (rdev
->raid_disk
>= 0 &&
5143 !test_bit(Faulty
, &rdev
->flags
) &&
5144 !test_bit(In_sync
, &rdev
->flags
) &&
5145 rdev
->recovery_offset
< j
)
5146 j
= rdev
->recovery_offset
;
5149 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
5150 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
5151 " %d KB/sec/disc.\n", speed_min(mddev
));
5152 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
5153 "(but not more than %d KB/sec) for reconstruction.\n",
5156 is_mddev_idle(mddev
); /* this also initializes IO event counters */
5159 for (m
= 0; m
< SYNC_MARKS
; m
++) {
5161 mark_cnt
[m
] = io_sectors
;
5164 mddev
->resync_mark
= mark
[last_mark
];
5165 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
5168 * Tune reconstruction:
5170 window
= 32*(PAGE_SIZE
/512);
5171 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
5172 window
/2,(unsigned long long) max_sectors
/2);
5174 atomic_set(&mddev
->recovery_active
, 0);
5175 init_waitqueue_head(&mddev
->recovery_wait
);
5180 "md: resuming recovery of %s from checkpoint.\n",
5182 mddev
->curr_resync
= j
;
5185 while (j
< max_sectors
) {
5189 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
5190 currspeed
< speed_min(mddev
));
5192 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5196 if (!skipped
) { /* actual IO requested */
5197 io_sectors
+= sectors
;
5198 atomic_add(sectors
, &mddev
->recovery_active
);
5202 if (j
>1) mddev
->curr_resync
= j
;
5203 if (last_check
== 0)
5204 /* this is the earliers that rebuilt will be
5205 * visible in /proc/mdstat
5207 md_new_event(mddev
);
5209 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
5212 last_check
= io_sectors
;
5214 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
5215 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
5219 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
5221 int next
= (last_mark
+1) % SYNC_MARKS
;
5223 mddev
->resync_mark
= mark
[next
];
5224 mddev
->resync_mark_cnt
= mark_cnt
[next
];
5225 mark
[next
] = jiffies
;
5226 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
5231 if (kthread_should_stop()) {
5233 * got a signal, exit.
5236 "md: md_do_sync() got signal ... exiting\n");
5237 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5242 * this loop exits only if either when we are slower than
5243 * the 'hard' speed limit, or the system was IO-idle for
5245 * the system might be non-idle CPU-wise, but we only care
5246 * about not overloading the IO subsystem. (things like an
5247 * e2fsck being done on the RAID array should execute fast)
5249 mddev
->queue
->unplug_fn(mddev
->queue
);
5252 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
5253 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
5255 if (currspeed
> speed_min(mddev
)) {
5256 if ((currspeed
> speed_max(mddev
)) ||
5257 !is_mddev_idle(mddev
)) {
5263 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
5265 * this also signals 'finished resyncing' to md_stop
5268 mddev
->queue
->unplug_fn(mddev
->queue
);
5270 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
5272 /* tell personality that we are finished */
5273 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
5275 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5276 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
5277 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
5278 mddev
->curr_resync
> 2) {
5279 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
5280 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5281 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
5283 "md: checkpointing recovery of %s.\n",
5285 mddev
->recovery_cp
= mddev
->curr_resync
;
5288 mddev
->recovery_cp
= MaxSector
;
5290 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
5291 mddev
->curr_resync
= MaxSector
;
5292 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5293 if (rdev
->raid_disk
>= 0 &&
5294 !test_bit(Faulty
, &rdev
->flags
) &&
5295 !test_bit(In_sync
, &rdev
->flags
) &&
5296 rdev
->recovery_offset
< mddev
->curr_resync
)
5297 rdev
->recovery_offset
= mddev
->curr_resync
;
5298 mddev
->sb_dirty
= 1;
5303 mddev
->curr_resync
= 0;
5304 wake_up(&resync_wait
);
5305 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5306 md_wakeup_thread(mddev
->thread
);
5308 EXPORT_SYMBOL_GPL(md_do_sync
);
5312 * This routine is regularly called by all per-raid-array threads to
5313 * deal with generic issues like resync and super-block update.
5314 * Raid personalities that don't have a thread (linear/raid0) do not
5315 * need this as they never do any recovery or update the superblock.
5317 * It does not do any resync itself, but rather "forks" off other threads
5318 * to do that as needed.
5319 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5320 * "->recovery" and create a thread at ->sync_thread.
5321 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5322 * and wakeups up this thread which will reap the thread and finish up.
5323 * This thread also removes any faulty devices (with nr_pending == 0).
5325 * The overall approach is:
5326 * 1/ if the superblock needs updating, update it.
5327 * 2/ If a recovery thread is running, don't do anything else.
5328 * 3/ If recovery has finished, clean up, possibly marking spares active.
5329 * 4/ If there are any faulty devices, remove them.
5330 * 5/ If array is degraded, try to add spares devices
5331 * 6/ If array has spares or is not in-sync, start a resync thread.
5333 void md_check_recovery(mddev_t
*mddev
)
5336 struct list_head
*rtmp
;
5340 bitmap_daemon_work(mddev
->bitmap
);
5345 if (signal_pending(current
)) {
5346 if (mddev
->pers
->sync_request
) {
5347 printk(KERN_INFO
"md: %s in immediate safe mode\n",
5349 mddev
->safemode
= 2;
5351 flush_signals(current
);
5356 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
5357 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
5358 (mddev
->safemode
== 1) ||
5359 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
5360 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
5364 if (mddev_trylock(mddev
)) {
5367 spin_lock_irq(&mddev
->write_lock
);
5368 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
5369 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
5371 mddev
->sb_dirty
= 3;
5373 if (mddev
->safemode
== 1)
5374 mddev
->safemode
= 0;
5375 spin_unlock_irq(&mddev
->write_lock
);
5377 if (mddev
->sb_dirty
)
5378 md_update_sb(mddev
);
5381 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
5382 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
5383 /* resync/recovery still happening */
5384 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5387 if (mddev
->sync_thread
) {
5388 /* resync has finished, collect result */
5389 md_unregister_thread(mddev
->sync_thread
);
5390 mddev
->sync_thread
= NULL
;
5391 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
5392 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
5394 /* activate any spares */
5395 mddev
->pers
->spare_active(mddev
);
5397 md_update_sb(mddev
);
5399 /* if array is no-longer degraded, then any saved_raid_disk
5400 * information must be scrapped
5402 if (!mddev
->degraded
)
5403 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5404 rdev
->saved_raid_disk
= -1;
5406 mddev
->recovery
= 0;
5407 /* flag recovery needed just to double check */
5408 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5409 md_new_event(mddev
);
5412 /* Clear some bits that don't mean anything, but
5415 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5416 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
5417 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5418 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
5420 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
5422 /* no recovery is running.
5423 * remove any failed drives, then
5424 * add spares if possible.
5425 * Spare are also removed and re-added, to allow
5426 * the personality to fail the re-add.
5428 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5429 if (rdev
->raid_disk
>= 0 &&
5430 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
5431 atomic_read(&rdev
->nr_pending
)==0) {
5432 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
5434 sprintf(nm
,"rd%d", rdev
->raid_disk
);
5435 sysfs_remove_link(&mddev
->kobj
, nm
);
5436 rdev
->raid_disk
= -1;
5440 if (mddev
->degraded
) {
5441 ITERATE_RDEV(mddev
,rdev
,rtmp
)
5442 if (rdev
->raid_disk
< 0
5443 && !test_bit(Faulty
, &rdev
->flags
)) {
5444 rdev
->recovery_offset
= 0;
5445 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
5447 sprintf(nm
, "rd%d", rdev
->raid_disk
);
5448 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
5450 md_new_event(mddev
);
5457 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5458 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
5459 } else if (mddev
->recovery_cp
< MaxSector
) {
5460 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
5461 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5462 /* nothing to be done ... */
5465 if (mddev
->pers
->sync_request
) {
5466 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
5467 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
5468 /* We are adding a device or devices to an array
5469 * which has the bitmap stored on all devices.
5470 * So make sure all bitmap pages get written
5472 bitmap_write_all(mddev
->bitmap
);
5474 mddev
->sync_thread
= md_register_thread(md_do_sync
,
5477 if (!mddev
->sync_thread
) {
5478 printk(KERN_ERR
"%s: could not start resync"
5481 /* leave the spares where they are, it shouldn't hurt */
5482 mddev
->recovery
= 0;
5484 md_wakeup_thread(mddev
->sync_thread
);
5485 md_new_event(mddev
);
5488 mddev_unlock(mddev
);
5492 static int md_notify_reboot(struct notifier_block
*this,
5493 unsigned long code
, void *x
)
5495 struct list_head
*tmp
;
5498 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
5500 printk(KERN_INFO
"md: stopping all md devices.\n");
5502 ITERATE_MDDEV(mddev
,tmp
)
5503 if (mddev_trylock(mddev
)) {
5504 do_md_stop (mddev
, 1);
5505 mddev_unlock(mddev
);
5508 * certain more exotic SCSI devices are known to be
5509 * volatile wrt too early system reboots. While the
5510 * right place to handle this issue is the given
5511 * driver, we do want to have a safe RAID driver ...
5518 static struct notifier_block md_notifier
= {
5519 .notifier_call
= md_notify_reboot
,
5521 .priority
= INT_MAX
, /* before any real devices */
5524 static void md_geninit(void)
5526 struct proc_dir_entry
*p
;
5528 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
5530 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
5532 p
->proc_fops
= &md_seq_fops
;
5535 static int __init
md_init(void)
5537 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5538 " MD_SB_DISKS=%d\n",
5539 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
5540 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
5541 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
5544 if (register_blkdev(MAJOR_NR
, "md"))
5546 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
5547 unregister_blkdev(MAJOR_NR
, "md");
5550 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
5551 md_probe
, NULL
, NULL
);
5552 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
5553 md_probe
, NULL
, NULL
);
5555 register_reboot_notifier(&md_notifier
);
5556 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
5566 * Searches all registered partitions for autorun RAID arrays
5569 static dev_t detected_devices
[128];
5572 void md_autodetect_dev(dev_t dev
)
5574 if (dev_cnt
>= 0 && dev_cnt
< 127)
5575 detected_devices
[dev_cnt
++] = dev
;
5579 static void autostart_arrays(int part
)
5584 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
5586 for (i
= 0; i
< dev_cnt
; i
++) {
5587 dev_t dev
= detected_devices
[i
];
5589 rdev
= md_import_device(dev
,0, 0);
5593 if (test_bit(Faulty
, &rdev
->flags
)) {
5597 list_add(&rdev
->same_set
, &pending_raid_disks
);
5601 autorun_devices(part
);
5606 static __exit
void md_exit(void)
5609 struct list_head
*tmp
;
5611 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
5612 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
5614 unregister_blkdev(MAJOR_NR
,"md");
5615 unregister_blkdev(mdp_major
, "mdp");
5616 unregister_reboot_notifier(&md_notifier
);
5617 unregister_sysctl_table(raid_table_header
);
5618 remove_proc_entry("mdstat", NULL
);
5619 ITERATE_MDDEV(mddev
,tmp
) {
5620 struct gendisk
*disk
= mddev
->gendisk
;
5623 export_array(mddev
);
5626 mddev
->gendisk
= NULL
;
5631 module_init(md_init
)
5632 module_exit(md_exit
)
5634 static int get_ro(char *buffer
, struct kernel_param
*kp
)
5636 return sprintf(buffer
, "%d", start_readonly
);
5638 static int set_ro(const char *val
, struct kernel_param
*kp
)
5641 int num
= simple_strtoul(val
, &e
, 10);
5642 if (*val
&& (*e
== '\0' || *e
== '\n')) {
5643 start_readonly
= num
;
5649 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
5650 module_param(start_dirty_degraded
, int, 0644);
5653 EXPORT_SYMBOL(register_md_personality
);
5654 EXPORT_SYMBOL(unregister_md_personality
);
5655 EXPORT_SYMBOL(md_error
);
5656 EXPORT_SYMBOL(md_done_sync
);
5657 EXPORT_SYMBOL(md_write_start
);
5658 EXPORT_SYMBOL(md_write_end
);
5659 EXPORT_SYMBOL(md_register_thread
);
5660 EXPORT_SYMBOL(md_unregister_thread
);
5661 EXPORT_SYMBOL(md_wakeup_thread
);
5662 EXPORT_SYMBOL(md_check_recovery
);
5663 MODULE_LICENSE("GPL");
5665 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);