2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part
);
71 static LIST_HEAD(pers_list
);
72 static DEFINE_SPINLOCK(pers_lock
);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
86 static int sysctl_speed_limit_min
= 1000;
87 static int sysctl_speed_limit_max
= 200000;
89 static struct ctl_table_header
*raid_table_header
;
91 static ctl_table raid_table
[] = {
93 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
94 .procname
= "speed_limit_min",
95 .data
= &sysctl_speed_limit_min
,
96 .maxlen
= sizeof(int),
98 .proc_handler
= &proc_dointvec
,
101 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
102 .procname
= "speed_limit_max",
103 .data
= &sysctl_speed_limit_max
,
104 .maxlen
= sizeof(int),
106 .proc_handler
= &proc_dointvec
,
111 static ctl_table raid_dir_table
[] = {
113 .ctl_name
= DEV_RAID
,
122 static ctl_table raid_root_table
[] = {
128 .child
= raid_dir_table
,
133 static struct block_device_operations md_fops
;
135 static int start_readonly
;
138 * We have a system wide 'event count' that is incremented
139 * on any 'interesting' event, and readers of /proc/mdstat
140 * can use 'poll' or 'select' to find out when the event
144 * start array, stop array, error, add device, remove device,
145 * start build, activate spare
147 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
148 static atomic_t md_event_count
;
149 static void md_new_event(mddev_t
*mddev
)
151 atomic_inc(&md_event_count
);
152 wake_up(&md_event_waiters
);
156 * Enables to iterate over all existing md arrays
157 * all_mddevs_lock protects this list.
159 static LIST_HEAD(all_mddevs
);
160 static DEFINE_SPINLOCK(all_mddevs_lock
);
164 * iterates through all used mddevs in the system.
165 * We take care to grab the all_mddevs_lock whenever navigating
166 * the list, and to always hold a refcount when unlocked.
167 * Any code which breaks out of this loop while own
168 * a reference to the current mddev and must mddev_put it.
170 #define ITERATE_MDDEV(mddev,tmp) \
172 for (({ spin_lock(&all_mddevs_lock); \
173 tmp = all_mddevs.next; \
175 ({ if (tmp != &all_mddevs) \
176 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177 spin_unlock(&all_mddevs_lock); \
178 if (mddev) mddev_put(mddev); \
179 mddev = list_entry(tmp, mddev_t, all_mddevs); \
180 tmp != &all_mddevs;}); \
181 ({ spin_lock(&all_mddevs_lock); \
186 static int md_fail_request (request_queue_t
*q
, struct bio
*bio
)
188 bio_io_error(bio
, bio
->bi_size
);
192 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
194 atomic_inc(&mddev
->active
);
198 static void mddev_put(mddev_t
*mddev
)
200 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
202 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
)) {
203 list_del(&mddev
->all_mddevs
);
204 blk_put_queue(mddev
->queue
);
205 kobject_unregister(&mddev
->kobj
);
207 spin_unlock(&all_mddevs_lock
);
210 static mddev_t
* mddev_find(dev_t unit
)
212 mddev_t
*mddev
, *new = NULL
;
215 spin_lock(&all_mddevs_lock
);
216 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
217 if (mddev
->unit
== unit
) {
219 spin_unlock(&all_mddevs_lock
);
225 list_add(&new->all_mddevs
, &all_mddevs
);
226 spin_unlock(&all_mddevs_lock
);
229 spin_unlock(&all_mddevs_lock
);
231 new = kzalloc(sizeof(*new), GFP_KERNEL
);
236 if (MAJOR(unit
) == MD_MAJOR
)
237 new->md_minor
= MINOR(unit
);
239 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
241 init_MUTEX(&new->reconfig_sem
);
242 INIT_LIST_HEAD(&new->disks
);
243 INIT_LIST_HEAD(&new->all_mddevs
);
244 init_timer(&new->safemode_timer
);
245 atomic_set(&new->active
, 1);
246 spin_lock_init(&new->write_lock
);
247 init_waitqueue_head(&new->sb_wait
);
249 new->queue
= blk_alloc_queue(GFP_KERNEL
);
255 blk_queue_make_request(new->queue
, md_fail_request
);
260 static inline int mddev_lock(mddev_t
* mddev
)
262 return down_interruptible(&mddev
->reconfig_sem
);
265 static inline void mddev_lock_uninterruptible(mddev_t
* mddev
)
267 down(&mddev
->reconfig_sem
);
270 static inline int mddev_trylock(mddev_t
* mddev
)
272 return down_trylock(&mddev
->reconfig_sem
);
275 static inline void mddev_unlock(mddev_t
* mddev
)
277 up(&mddev
->reconfig_sem
);
279 md_wakeup_thread(mddev
->thread
);
282 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
285 struct list_head
*tmp
;
287 ITERATE_RDEV(mddev
,rdev
,tmp
) {
288 if (rdev
->desc_nr
== nr
)
294 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
296 struct list_head
*tmp
;
299 ITERATE_RDEV(mddev
,rdev
,tmp
) {
300 if (rdev
->bdev
->bd_dev
== dev
)
306 static struct mdk_personality
*find_pers(int level
, char *clevel
)
308 struct mdk_personality
*pers
;
309 list_for_each_entry(pers
, &pers_list
, list
) {
310 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
312 if (strcmp(pers
->name
, clevel
)==0)
318 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
320 sector_t size
= bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
321 return MD_NEW_SIZE_BLOCKS(size
);
324 static sector_t
calc_dev_size(mdk_rdev_t
*rdev
, unsigned chunk_size
)
328 size
= rdev
->sb_offset
;
331 size
&= ~((sector_t
)chunk_size
/1024 - 1);
335 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
340 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
341 if (!rdev
->sb_page
) {
342 printk(KERN_ALERT
"md: out of memory.\n");
349 static void free_disk_sb(mdk_rdev_t
* rdev
)
352 put_page(rdev
->sb_page
);
354 rdev
->sb_page
= NULL
;
361 static int super_written(struct bio
*bio
, unsigned int bytes_done
, int error
)
363 mdk_rdev_t
*rdev
= bio
->bi_private
;
364 mddev_t
*mddev
= rdev
->mddev
;
368 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
369 md_error(mddev
, rdev
);
371 if (atomic_dec_and_test(&mddev
->pending_writes
))
372 wake_up(&mddev
->sb_wait
);
377 static int super_written_barrier(struct bio
*bio
, unsigned int bytes_done
, int error
)
379 struct bio
*bio2
= bio
->bi_private
;
380 mdk_rdev_t
*rdev
= bio2
->bi_private
;
381 mddev_t
*mddev
= rdev
->mddev
;
385 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
386 error
== -EOPNOTSUPP
) {
388 /* barriers don't appear to be supported :-( */
389 set_bit(BarriersNotsupp
, &rdev
->flags
);
390 mddev
->barriers_work
= 0;
391 spin_lock_irqsave(&mddev
->write_lock
, flags
);
392 bio2
->bi_next
= mddev
->biolist
;
393 mddev
->biolist
= bio2
;
394 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
395 wake_up(&mddev
->sb_wait
);
400 bio
->bi_private
= rdev
;
401 return super_written(bio
, bytes_done
, error
);
404 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
405 sector_t sector
, int size
, struct page
*page
)
407 /* write first size bytes of page to sector of rdev
408 * Increment mddev->pending_writes before returning
409 * and decrement it on completion, waking up sb_wait
410 * if zero is reached.
411 * If an error occurred, call md_error
413 * As we might need to resubmit the request if BIO_RW_BARRIER
414 * causes ENOTSUPP, we allocate a spare bio...
416 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
417 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNC
);
419 bio
->bi_bdev
= rdev
->bdev
;
420 bio
->bi_sector
= sector
;
421 bio_add_page(bio
, page
, size
, 0);
422 bio
->bi_private
= rdev
;
423 bio
->bi_end_io
= super_written
;
426 atomic_inc(&mddev
->pending_writes
);
427 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
429 rw
|= (1<<BIO_RW_BARRIER
);
430 rbio
= bio_clone(bio
, GFP_NOIO
);
431 rbio
->bi_private
= bio
;
432 rbio
->bi_end_io
= super_written_barrier
;
433 submit_bio(rw
, rbio
);
438 void md_super_wait(mddev_t
*mddev
)
440 /* wait for all superblock writes that were scheduled to complete.
441 * if any had to be retried (due to BARRIER problems), retry them
445 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
446 if (atomic_read(&mddev
->pending_writes
)==0)
448 while (mddev
->biolist
) {
450 spin_lock_irq(&mddev
->write_lock
);
451 bio
= mddev
->biolist
;
452 mddev
->biolist
= bio
->bi_next
;
454 spin_unlock_irq(&mddev
->write_lock
);
455 submit_bio(bio
->bi_rw
, bio
);
459 finish_wait(&mddev
->sb_wait
, &wq
);
462 static int bi_complete(struct bio
*bio
, unsigned int bytes_done
, int error
)
467 complete((struct completion
*)bio
->bi_private
);
471 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
472 struct page
*page
, int rw
)
474 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
475 struct completion event
;
478 rw
|= (1 << BIO_RW_SYNC
);
481 bio
->bi_sector
= sector
;
482 bio_add_page(bio
, page
, size
, 0);
483 init_completion(&event
);
484 bio
->bi_private
= &event
;
485 bio
->bi_end_io
= bi_complete
;
487 wait_for_completion(&event
);
489 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
493 EXPORT_SYMBOL_GPL(sync_page_io
);
495 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
497 char b
[BDEVNAME_SIZE
];
498 if (!rdev
->sb_page
) {
506 if (!sync_page_io(rdev
->bdev
, rdev
->sb_offset
<<1, size
, rdev
->sb_page
, READ
))
512 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
513 bdevname(rdev
->bdev
,b
));
517 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
519 if ( (sb1
->set_uuid0
== sb2
->set_uuid0
) &&
520 (sb1
->set_uuid1
== sb2
->set_uuid1
) &&
521 (sb1
->set_uuid2
== sb2
->set_uuid2
) &&
522 (sb1
->set_uuid3
== sb2
->set_uuid3
))
530 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
533 mdp_super_t
*tmp1
, *tmp2
;
535 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
536 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
538 if (!tmp1
|| !tmp2
) {
540 printk(KERN_INFO
"md.c: sb1 is not equal to sb2!\n");
548 * nr_disks is not constant
553 if (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4))
564 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
566 unsigned int disk_csum
, csum
;
568 disk_csum
= sb
->sb_csum
;
570 csum
= csum_partial((void *)sb
, MD_SB_BYTES
, 0);
571 sb
->sb_csum
= disk_csum
;
577 * Handle superblock details.
578 * We want to be able to handle multiple superblock formats
579 * so we have a common interface to them all, and an array of
580 * different handlers.
581 * We rely on user-space to write the initial superblock, and support
582 * reading and updating of superblocks.
583 * Interface methods are:
584 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
585 * loads and validates a superblock on dev.
586 * if refdev != NULL, compare superblocks on both devices
588 * 0 - dev has a superblock that is compatible with refdev
589 * 1 - dev has a superblock that is compatible and newer than refdev
590 * so dev should be used as the refdev in future
591 * -EINVAL superblock incompatible or invalid
592 * -othererror e.g. -EIO
594 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
595 * Verify that dev is acceptable into mddev.
596 * The first time, mddev->raid_disks will be 0, and data from
597 * dev should be merged in. Subsequent calls check that dev
598 * is new enough. Return 0 or -EINVAL
600 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
601 * Update the superblock for rdev with data in mddev
602 * This does not write to disc.
608 struct module
*owner
;
609 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
);
610 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
611 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
615 * load_super for 0.90.0
617 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
619 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
625 * Calculate the position of the superblock,
626 * it's at the end of the disk.
628 * It also happens to be a multiple of 4Kb.
630 sb_offset
= calc_dev_sboffset(rdev
->bdev
);
631 rdev
->sb_offset
= sb_offset
;
633 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
638 bdevname(rdev
->bdev
, b
);
639 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
641 if (sb
->md_magic
!= MD_SB_MAGIC
) {
642 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
647 if (sb
->major_version
!= 0 ||
648 sb
->minor_version
!= 90) {
649 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
650 sb
->major_version
, sb
->minor_version
,
655 if (sb
->raid_disks
<= 0)
658 if (csum_fold(calc_sb_csum(sb
)) != csum_fold(sb
->sb_csum
)) {
659 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
664 rdev
->preferred_minor
= sb
->md_minor
;
665 rdev
->data_offset
= 0;
666 rdev
->sb_size
= MD_SB_BYTES
;
668 if (sb
->level
== LEVEL_MULTIPATH
)
671 rdev
->desc_nr
= sb
->this_disk
.number
;
677 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
678 if (!uuid_equal(refsb
, sb
)) {
679 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
680 b
, bdevname(refdev
->bdev
,b2
));
683 if (!sb_equal(refsb
, sb
)) {
684 printk(KERN_WARNING
"md: %s has same UUID"
685 " but different superblock to %s\n",
686 b
, bdevname(refdev
->bdev
, b2
));
690 ev2
= md_event(refsb
);
696 rdev
->size
= calc_dev_size(rdev
, sb
->chunk_size
);
698 if (rdev
->size
< sb
->size
&& sb
->level
> 1)
699 /* "this cannot possibly happen" ... */
707 * validate_super for 0.90.0
709 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
712 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
714 rdev
->raid_disk
= -1;
716 if (mddev
->raid_disks
== 0) {
717 mddev
->major_version
= 0;
718 mddev
->minor_version
= sb
->minor_version
;
719 mddev
->patch_version
= sb
->patch_version
;
720 mddev
->persistent
= ! sb
->not_persistent
;
721 mddev
->chunk_size
= sb
->chunk_size
;
722 mddev
->ctime
= sb
->ctime
;
723 mddev
->utime
= sb
->utime
;
724 mddev
->level
= sb
->level
;
725 mddev
->clevel
[0] = 0;
726 mddev
->layout
= sb
->layout
;
727 mddev
->raid_disks
= sb
->raid_disks
;
728 mddev
->size
= sb
->size
;
729 mddev
->events
= md_event(sb
);
730 mddev
->bitmap_offset
= 0;
731 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
733 if (sb
->state
& (1<<MD_SB_CLEAN
))
734 mddev
->recovery_cp
= MaxSector
;
736 if (sb
->events_hi
== sb
->cp_events_hi
&&
737 sb
->events_lo
== sb
->cp_events_lo
) {
738 mddev
->recovery_cp
= sb
->recovery_cp
;
740 mddev
->recovery_cp
= 0;
743 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
744 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
745 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
746 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
748 mddev
->max_disks
= MD_SB_DISKS
;
750 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
751 mddev
->bitmap_file
== NULL
) {
752 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
753 && mddev
->level
!= 10) {
754 /* FIXME use a better test */
755 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
758 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
761 } else if (mddev
->pers
== NULL
) {
762 /* Insist on good event counter while assembling */
763 __u64 ev1
= md_event(sb
);
765 if (ev1
< mddev
->events
)
767 } else if (mddev
->bitmap
) {
768 /* if adding to array with a bitmap, then we can accept an
769 * older device ... but not too old.
771 __u64 ev1
= md_event(sb
);
772 if (ev1
< mddev
->bitmap
->events_cleared
)
774 } else /* just a hot-add of a new device, leave raid_disk at -1 */
777 if (mddev
->level
!= LEVEL_MULTIPATH
) {
778 desc
= sb
->disks
+ rdev
->desc_nr
;
780 if (desc
->state
& (1<<MD_DISK_FAULTY
))
781 set_bit(Faulty
, &rdev
->flags
);
782 else if (desc
->state
& (1<<MD_DISK_SYNC
) &&
783 desc
->raid_disk
< mddev
->raid_disks
) {
784 set_bit(In_sync
, &rdev
->flags
);
785 rdev
->raid_disk
= desc
->raid_disk
;
787 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
788 set_bit(WriteMostly
, &rdev
->flags
);
789 } else /* MULTIPATH are always insync */
790 set_bit(In_sync
, &rdev
->flags
);
795 * sync_super for 0.90.0
797 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
800 struct list_head
*tmp
;
802 int next_spare
= mddev
->raid_disks
;
805 /* make rdev->sb match mddev data..
808 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
809 * 3/ any empty disks < next_spare become removed
811 * disks[0] gets initialised to REMOVED because
812 * we cannot be sure from other fields if it has
813 * been initialised or not.
816 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
818 rdev
->sb_size
= MD_SB_BYTES
;
820 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
822 memset(sb
, 0, sizeof(*sb
));
824 sb
->md_magic
= MD_SB_MAGIC
;
825 sb
->major_version
= mddev
->major_version
;
826 sb
->minor_version
= mddev
->minor_version
;
827 sb
->patch_version
= mddev
->patch_version
;
828 sb
->gvalid_words
= 0; /* ignored */
829 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
830 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
831 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
832 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
834 sb
->ctime
= mddev
->ctime
;
835 sb
->level
= mddev
->level
;
836 sb
->size
= mddev
->size
;
837 sb
->raid_disks
= mddev
->raid_disks
;
838 sb
->md_minor
= mddev
->md_minor
;
839 sb
->not_persistent
= !mddev
->persistent
;
840 sb
->utime
= mddev
->utime
;
842 sb
->events_hi
= (mddev
->events
>>32);
843 sb
->events_lo
= (u32
)mddev
->events
;
847 sb
->recovery_cp
= mddev
->recovery_cp
;
848 sb
->cp_events_hi
= (mddev
->events
>>32);
849 sb
->cp_events_lo
= (u32
)mddev
->events
;
850 if (mddev
->recovery_cp
== MaxSector
)
851 sb
->state
= (1<< MD_SB_CLEAN
);
855 sb
->layout
= mddev
->layout
;
856 sb
->chunk_size
= mddev
->chunk_size
;
858 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
859 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
861 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
862 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
865 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
866 && !test_bit(Faulty
, &rdev2
->flags
))
867 desc_nr
= rdev2
->raid_disk
;
869 desc_nr
= next_spare
++;
870 rdev2
->desc_nr
= desc_nr
;
871 d
= &sb
->disks
[rdev2
->desc_nr
];
873 d
->number
= rdev2
->desc_nr
;
874 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
875 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
876 if (rdev2
->raid_disk
>= 0 && test_bit(In_sync
, &rdev2
->flags
)
877 && !test_bit(Faulty
, &rdev2
->flags
))
878 d
->raid_disk
= rdev2
->raid_disk
;
880 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
881 if (test_bit(Faulty
, &rdev2
->flags
)) {
882 d
->state
= (1<<MD_DISK_FAULTY
);
884 } else if (test_bit(In_sync
, &rdev2
->flags
)) {
885 d
->state
= (1<<MD_DISK_ACTIVE
);
886 d
->state
|= (1<<MD_DISK_SYNC
);
894 if (test_bit(WriteMostly
, &rdev2
->flags
))
895 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
897 /* now set the "removed" and "faulty" bits on any missing devices */
898 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
899 mdp_disk_t
*d
= &sb
->disks
[i
];
900 if (d
->state
== 0 && d
->number
== 0) {
903 d
->state
= (1<<MD_DISK_REMOVED
);
904 d
->state
|= (1<<MD_DISK_FAULTY
);
908 sb
->nr_disks
= nr_disks
;
909 sb
->active_disks
= active
;
910 sb
->working_disks
= working
;
911 sb
->failed_disks
= failed
;
912 sb
->spare_disks
= spare
;
914 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
915 sb
->sb_csum
= calc_sb_csum(sb
);
919 * version 1 superblock
922 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
924 unsigned int disk_csum
, csum
;
925 unsigned long long newcsum
;
926 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
927 unsigned int *isuper
= (unsigned int*)sb
;
930 disk_csum
= sb
->sb_csum
;
933 for (i
=0; size
>=4; size
-= 4 )
934 newcsum
+= le32_to_cpu(*isuper
++);
937 newcsum
+= le16_to_cpu(*(unsigned short*) isuper
);
939 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
940 sb
->sb_csum
= disk_csum
;
941 return cpu_to_le32(csum
);
944 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
946 struct mdp_superblock_1
*sb
;
949 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
953 * Calculate the position of the superblock.
954 * It is always aligned to a 4K boundary and
955 * depeding on minor_version, it can be:
956 * 0: At least 8K, but less than 12K, from end of device
957 * 1: At start of device
958 * 2: 4K from start of device.
960 switch(minor_version
) {
962 sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> 9;
964 sb_offset
&= ~(sector_t
)(4*2-1);
965 /* convert from sectors to K */
977 rdev
->sb_offset
= sb_offset
;
979 /* superblock is rarely larger than 1K, but it can be larger,
980 * and it is safe to read 4k, so we do that
982 ret
= read_disk_sb(rdev
, 4096);
986 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
988 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
989 sb
->major_version
!= cpu_to_le32(1) ||
990 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
991 le64_to_cpu(sb
->super_offset
) != (rdev
->sb_offset
<<1) ||
992 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
995 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
996 printk("md: invalid superblock checksum on %s\n",
997 bdevname(rdev
->bdev
,b
));
1000 if (le64_to_cpu(sb
->data_size
) < 10) {
1001 printk("md: data_size too small on %s\n",
1002 bdevname(rdev
->bdev
,b
));
1005 rdev
->preferred_minor
= 0xffff;
1006 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1007 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1009 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1010 bmask
= queue_hardsect_size(rdev
->bdev
->bd_disk
->queue
)-1;
1011 if (rdev
->sb_size
& bmask
)
1012 rdev
-> sb_size
= (rdev
->sb_size
| bmask
)+1;
1018 struct mdp_superblock_1
*refsb
=
1019 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1021 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1022 sb
->level
!= refsb
->level
||
1023 sb
->layout
!= refsb
->layout
||
1024 sb
->chunksize
!= refsb
->chunksize
) {
1025 printk(KERN_WARNING
"md: %s has strangely different"
1026 " superblock to %s\n",
1027 bdevname(rdev
->bdev
,b
),
1028 bdevname(refdev
->bdev
,b2
));
1031 ev1
= le64_to_cpu(sb
->events
);
1032 ev2
= le64_to_cpu(refsb
->events
);
1038 rdev
->size
= ((rdev
->bdev
->bd_inode
->i_size
>>9) - le64_to_cpu(sb
->data_offset
)) / 2;
1040 rdev
->size
= rdev
->sb_offset
;
1041 if (rdev
->size
< le64_to_cpu(sb
->data_size
)/2)
1043 rdev
->size
= le64_to_cpu(sb
->data_size
)/2;
1044 if (le32_to_cpu(sb
->chunksize
))
1045 rdev
->size
&= ~((sector_t
)le32_to_cpu(sb
->chunksize
)/2 - 1);
1047 if (le32_to_cpu(sb
->size
) > rdev
->size
*2)
1052 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1054 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1056 rdev
->raid_disk
= -1;
1058 if (mddev
->raid_disks
== 0) {
1059 mddev
->major_version
= 1;
1060 mddev
->patch_version
= 0;
1061 mddev
->persistent
= 1;
1062 mddev
->chunk_size
= le32_to_cpu(sb
->chunksize
) << 9;
1063 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1064 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1065 mddev
->level
= le32_to_cpu(sb
->level
);
1066 mddev
->clevel
[0] = 0;
1067 mddev
->layout
= le32_to_cpu(sb
->layout
);
1068 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1069 mddev
->size
= le64_to_cpu(sb
->size
)/2;
1070 mddev
->events
= le64_to_cpu(sb
->events
);
1071 mddev
->bitmap_offset
= 0;
1072 mddev
->default_bitmap_offset
= 1024;
1074 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1075 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1077 mddev
->max_disks
= (4096-256)/2;
1079 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1080 mddev
->bitmap_file
== NULL
) {
1081 if (mddev
->level
!= 1 && mddev
->level
!= 5 && mddev
->level
!= 6
1082 && mddev
->level
!= 10) {
1083 printk(KERN_WARNING
"md: bitmaps not supported for this level.\n");
1086 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1088 } else if (mddev
->pers
== NULL
) {
1089 /* Insist of good event counter while assembling */
1090 __u64 ev1
= le64_to_cpu(sb
->events
);
1092 if (ev1
< mddev
->events
)
1094 } else if (mddev
->bitmap
) {
1095 /* If adding to array with a bitmap, then we can accept an
1096 * older device, but not too old.
1098 __u64 ev1
= le64_to_cpu(sb
->events
);
1099 if (ev1
< mddev
->bitmap
->events_cleared
)
1101 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1104 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1106 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1107 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1109 case 0xffff: /* spare */
1111 case 0xfffe: /* faulty */
1112 set_bit(Faulty
, &rdev
->flags
);
1115 set_bit(In_sync
, &rdev
->flags
);
1116 rdev
->raid_disk
= role
;
1119 if (sb
->devflags
& WriteMostly1
)
1120 set_bit(WriteMostly
, &rdev
->flags
);
1121 } else /* MULTIPATH are always insync */
1122 set_bit(In_sync
, &rdev
->flags
);
1127 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1129 struct mdp_superblock_1
*sb
;
1130 struct list_head
*tmp
;
1133 /* make rdev->sb match mddev and rdev data. */
1135 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1137 sb
->feature_map
= 0;
1139 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1140 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1141 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1143 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1144 sb
->events
= cpu_to_le64(mddev
->events
);
1146 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1148 sb
->resync_offset
= cpu_to_le64(0);
1150 sb
->cnt_corrected_read
= atomic_read(&rdev
->corrected_errors
);
1152 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1153 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1154 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1158 ITERATE_RDEV(mddev
,rdev2
,tmp
)
1159 if (rdev2
->desc_nr
+1 > max_dev
)
1160 max_dev
= rdev2
->desc_nr
+1;
1162 sb
->max_dev
= cpu_to_le32(max_dev
);
1163 for (i
=0; i
<max_dev
;i
++)
1164 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1166 ITERATE_RDEV(mddev
,rdev2
,tmp
) {
1168 if (test_bit(Faulty
, &rdev2
->flags
))
1169 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1170 else if (test_bit(In_sync
, &rdev2
->flags
))
1171 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1173 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1176 sb
->recovery_offset
= cpu_to_le64(0); /* not supported yet */
1177 sb
->sb_csum
= calc_sb_1_csum(sb
);
1181 static struct super_type super_types
[] = {
1184 .owner
= THIS_MODULE
,
1185 .load_super
= super_90_load
,
1186 .validate_super
= super_90_validate
,
1187 .sync_super
= super_90_sync
,
1191 .owner
= THIS_MODULE
,
1192 .load_super
= super_1_load
,
1193 .validate_super
= super_1_validate
,
1194 .sync_super
= super_1_sync
,
1198 static mdk_rdev_t
* match_dev_unit(mddev_t
*mddev
, mdk_rdev_t
*dev
)
1200 struct list_head
*tmp
;
1203 ITERATE_RDEV(mddev
,rdev
,tmp
)
1204 if (rdev
->bdev
->bd_contains
== dev
->bdev
->bd_contains
)
1210 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1212 struct list_head
*tmp
;
1215 ITERATE_RDEV(mddev1
,rdev
,tmp
)
1216 if (match_dev_unit(mddev2
, rdev
))
1222 static LIST_HEAD(pending_raid_disks
);
1224 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1226 mdk_rdev_t
*same_pdev
;
1227 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1234 /* make sure rdev->size exceeds mddev->size */
1235 if (rdev
->size
&& (mddev
->size
== 0 || rdev
->size
< mddev
->size
)) {
1237 /* Cannot change size, so fail */
1240 mddev
->size
= rdev
->size
;
1242 same_pdev
= match_dev_unit(mddev
, rdev
);
1245 "%s: WARNING: %s appears to be on the same physical"
1246 " disk as %s. True\n protection against single-disk"
1247 " failure might be compromised.\n",
1248 mdname(mddev
), bdevname(rdev
->bdev
,b
),
1249 bdevname(same_pdev
->bdev
,b2
));
1251 /* Verify rdev->desc_nr is unique.
1252 * If it is -1, assign a free number, else
1253 * check number is not in use
1255 if (rdev
->desc_nr
< 0) {
1257 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1258 while (find_rdev_nr(mddev
, choice
))
1260 rdev
->desc_nr
= choice
;
1262 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1265 bdevname(rdev
->bdev
,b
);
1266 if (kobject_set_name(&rdev
->kobj
, "dev-%s", b
) < 0)
1269 list_add(&rdev
->same_set
, &mddev
->disks
);
1270 rdev
->mddev
= mddev
;
1271 printk(KERN_INFO
"md: bind<%s>\n", b
);
1273 rdev
->kobj
.parent
= &mddev
->kobj
;
1274 kobject_add(&rdev
->kobj
);
1276 if (rdev
->bdev
->bd_part
)
1277 ko
= &rdev
->bdev
->bd_part
->kobj
;
1279 ko
= &rdev
->bdev
->bd_disk
->kobj
;
1280 sysfs_create_link(&rdev
->kobj
, ko
, "block");
1284 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1286 char b
[BDEVNAME_SIZE
];
1291 list_del_init(&rdev
->same_set
);
1292 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1294 sysfs_remove_link(&rdev
->kobj
, "block");
1295 kobject_del(&rdev
->kobj
);
1299 * prevent the device from being mounted, repartitioned or
1300 * otherwise reused by a RAID array (or any other kernel
1301 * subsystem), by bd_claiming the device.
1303 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
)
1306 struct block_device
*bdev
;
1307 char b
[BDEVNAME_SIZE
];
1309 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1311 printk(KERN_ERR
"md: could not open %s.\n",
1312 __bdevname(dev
, b
));
1313 return PTR_ERR(bdev
);
1315 err
= bd_claim(bdev
, rdev
);
1317 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1326 static void unlock_rdev(mdk_rdev_t
*rdev
)
1328 struct block_device
*bdev
= rdev
->bdev
;
1336 void md_autodetect_dev(dev_t dev
);
1338 static void export_rdev(mdk_rdev_t
* rdev
)
1340 char b
[BDEVNAME_SIZE
];
1341 printk(KERN_INFO
"md: export_rdev(%s)\n",
1342 bdevname(rdev
->bdev
,b
));
1346 list_del_init(&rdev
->same_set
);
1348 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1351 kobject_put(&rdev
->kobj
);
1354 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1356 unbind_rdev_from_array(rdev
);
1360 static void export_array(mddev_t
*mddev
)
1362 struct list_head
*tmp
;
1365 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1370 kick_rdev_from_array(rdev
);
1372 if (!list_empty(&mddev
->disks
))
1374 mddev
->raid_disks
= 0;
1375 mddev
->major_version
= 0;
1378 static void print_desc(mdp_disk_t
*desc
)
1380 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1381 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1384 static void print_sb(mdp_super_t
*sb
)
1389 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1390 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1391 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1393 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1394 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1395 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1396 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1397 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1398 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1399 sb
->failed_disks
, sb
->spare_disks
,
1400 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1403 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1406 desc
= sb
->disks
+ i
;
1407 if (desc
->number
|| desc
->major
|| desc
->minor
||
1408 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1409 printk(" D %2d: ", i
);
1413 printk(KERN_INFO
"md: THIS: ");
1414 print_desc(&sb
->this_disk
);
1418 static void print_rdev(mdk_rdev_t
*rdev
)
1420 char b
[BDEVNAME_SIZE
];
1421 printk(KERN_INFO
"md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1422 bdevname(rdev
->bdev
,b
), (unsigned long long)rdev
->size
,
1423 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1425 if (rdev
->sb_loaded
) {
1426 printk(KERN_INFO
"md: rdev superblock:\n");
1427 print_sb((mdp_super_t
*)page_address(rdev
->sb_page
));
1429 printk(KERN_INFO
"md: no rdev superblock!\n");
1432 void md_print_devices(void)
1434 struct list_head
*tmp
, *tmp2
;
1437 char b
[BDEVNAME_SIZE
];
1440 printk("md: **********************************\n");
1441 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1442 printk("md: **********************************\n");
1443 ITERATE_MDDEV(mddev
,tmp
) {
1446 bitmap_print_sb(mddev
->bitmap
);
1448 printk("%s: ", mdname(mddev
));
1449 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1450 printk("<%s>", bdevname(rdev
->bdev
,b
));
1453 ITERATE_RDEV(mddev
,rdev
,tmp2
)
1456 printk("md: **********************************\n");
1461 static void sync_sbs(mddev_t
* mddev
)
1464 struct list_head
*tmp
;
1466 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1467 super_types
[mddev
->major_version
].
1468 sync_super(mddev
, rdev
);
1469 rdev
->sb_loaded
= 1;
1473 static void md_update_sb(mddev_t
* mddev
)
1476 struct list_head
*tmp
;
1481 spin_lock_irq(&mddev
->write_lock
);
1482 sync_req
= mddev
->in_sync
;
1483 mddev
->utime
= get_seconds();
1486 if (!mddev
->events
) {
1488 * oops, this 64-bit counter should never wrap.
1489 * Either we are in around ~1 trillion A.C., assuming
1490 * 1 reboot per second, or we have a bug:
1495 mddev
->sb_dirty
= 2;
1499 * do not write anything to disk if using
1500 * nonpersistent superblocks
1502 if (!mddev
->persistent
) {
1503 mddev
->sb_dirty
= 0;
1504 spin_unlock_irq(&mddev
->write_lock
);
1505 wake_up(&mddev
->sb_wait
);
1508 spin_unlock_irq(&mddev
->write_lock
);
1511 "md: updating %s RAID superblock on device (in sync %d)\n",
1512 mdname(mddev
),mddev
->in_sync
);
1514 err
= bitmap_update_sb(mddev
->bitmap
);
1515 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1516 char b
[BDEVNAME_SIZE
];
1517 dprintk(KERN_INFO
"md: ");
1518 if (test_bit(Faulty
, &rdev
->flags
))
1519 dprintk("(skipping faulty ");
1521 dprintk("%s ", bdevname(rdev
->bdev
,b
));
1522 if (!test_bit(Faulty
, &rdev
->flags
)) {
1523 md_super_write(mddev
,rdev
,
1524 rdev
->sb_offset
<<1, rdev
->sb_size
,
1526 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
1527 bdevname(rdev
->bdev
,b
),
1528 (unsigned long long)rdev
->sb_offset
);
1532 if (mddev
->level
== LEVEL_MULTIPATH
)
1533 /* only need to write one superblock... */
1536 md_super_wait(mddev
);
1537 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1539 spin_lock_irq(&mddev
->write_lock
);
1540 if (mddev
->in_sync
!= sync_req
|| mddev
->sb_dirty
== 1) {
1541 /* have to write it out again */
1542 spin_unlock_irq(&mddev
->write_lock
);
1545 mddev
->sb_dirty
= 0;
1546 spin_unlock_irq(&mddev
->write_lock
);
1547 wake_up(&mddev
->sb_wait
);
1551 /* words written to sysfs files may, or my not, be \n terminated.
1552 * We want to accept with case. For this we use cmd_match.
1554 static int cmd_match(const char *cmd
, const char *str
)
1556 /* See if cmd, written into a sysfs file, matches
1557 * str. They must either be the same, or cmd can
1558 * have a trailing newline
1560 while (*cmd
&& *str
&& *cmd
== *str
) {
1571 struct rdev_sysfs_entry
{
1572 struct attribute attr
;
1573 ssize_t (*show
)(mdk_rdev_t
*, char *);
1574 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
1578 state_show(mdk_rdev_t
*rdev
, char *page
)
1583 if (test_bit(Faulty
, &rdev
->flags
)) {
1584 len
+= sprintf(page
+len
, "%sfaulty",sep
);
1587 if (test_bit(In_sync
, &rdev
->flags
)) {
1588 len
+= sprintf(page
+len
, "%sin_sync",sep
);
1591 if (!test_bit(Faulty
, &rdev
->flags
) &&
1592 !test_bit(In_sync
, &rdev
->flags
)) {
1593 len
+= sprintf(page
+len
, "%sspare", sep
);
1596 return len
+sprintf(page
+len
, "\n");
1599 static struct rdev_sysfs_entry
1600 rdev_state
= __ATTR_RO(state
);
1603 super_show(mdk_rdev_t
*rdev
, char *page
)
1605 if (rdev
->sb_loaded
&& rdev
->sb_size
) {
1606 memcpy(page
, page_address(rdev
->sb_page
), rdev
->sb_size
);
1607 return rdev
->sb_size
;
1611 static struct rdev_sysfs_entry rdev_super
= __ATTR_RO(super
);
1614 errors_show(mdk_rdev_t
*rdev
, char *page
)
1616 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
1620 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1623 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1624 if (*buf
&& (*e
== 0 || *e
== '\n')) {
1625 atomic_set(&rdev
->corrected_errors
, n
);
1630 static struct rdev_sysfs_entry rdev_errors
=
1631 __ATTR(errors
, 0644, errors_show
, errors_store
);
1634 slot_show(mdk_rdev_t
*rdev
, char *page
)
1636 if (rdev
->raid_disk
< 0)
1637 return sprintf(page
, "none\n");
1639 return sprintf(page
, "%d\n", rdev
->raid_disk
);
1643 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1646 int slot
= simple_strtoul(buf
, &e
, 10);
1647 if (strncmp(buf
, "none", 4)==0)
1649 else if (e
==buf
|| (*e
&& *e
!= '\n'))
1651 if (rdev
->mddev
->pers
)
1652 /* Cannot set slot in active array (yet) */
1654 if (slot
>= rdev
->mddev
->raid_disks
)
1656 rdev
->raid_disk
= slot
;
1657 /* assume it is working */
1659 set_bit(In_sync
, &rdev
->flags
);
1664 static struct rdev_sysfs_entry rdev_slot
=
1665 __ATTR(slot
, 0644, slot_show
, slot_store
);
1668 offset_show(mdk_rdev_t
*rdev
, char *page
)
1670 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
1674 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
1677 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
1678 if (e
==buf
|| (*e
&& *e
!= '\n'))
1680 if (rdev
->mddev
->pers
)
1682 rdev
->data_offset
= offset
;
1686 static struct rdev_sysfs_entry rdev_offset
=
1687 __ATTR(offset
, 0644, offset_show
, offset_store
);
1689 static struct attribute
*rdev_default_attrs
[] = {
1698 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
1700 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1701 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1705 return entry
->show(rdev
, page
);
1709 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
1710 const char *page
, size_t length
)
1712 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
1713 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
1717 return entry
->store(rdev
, page
, length
);
1720 static void rdev_free(struct kobject
*ko
)
1722 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
1725 static struct sysfs_ops rdev_sysfs_ops
= {
1726 .show
= rdev_attr_show
,
1727 .store
= rdev_attr_store
,
1729 static struct kobj_type rdev_ktype
= {
1730 .release
= rdev_free
,
1731 .sysfs_ops
= &rdev_sysfs_ops
,
1732 .default_attrs
= rdev_default_attrs
,
1736 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1738 * mark the device faulty if:
1740 * - the device is nonexistent (zero size)
1741 * - the device has no valid superblock
1743 * a faulty rdev _never_ has rdev->sb set.
1745 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
1747 char b
[BDEVNAME_SIZE
];
1752 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
1754 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
1755 return ERR_PTR(-ENOMEM
);
1758 if ((err
= alloc_disk_sb(rdev
)))
1761 err
= lock_rdev(rdev
, newdev
);
1765 rdev
->kobj
.parent
= NULL
;
1766 rdev
->kobj
.ktype
= &rdev_ktype
;
1767 kobject_init(&rdev
->kobj
);
1771 rdev
->data_offset
= 0;
1772 atomic_set(&rdev
->nr_pending
, 0);
1773 atomic_set(&rdev
->read_errors
, 0);
1774 atomic_set(&rdev
->corrected_errors
, 0);
1776 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
1779 "md: %s has zero or unknown size, marking faulty!\n",
1780 bdevname(rdev
->bdev
,b
));
1785 if (super_format
>= 0) {
1786 err
= super_types
[super_format
].
1787 load_super(rdev
, NULL
, super_minor
);
1788 if (err
== -EINVAL
) {
1790 "md: %s has invalid sb, not importing!\n",
1791 bdevname(rdev
->bdev
,b
));
1796 "md: could not read %s's sb, not importing!\n",
1797 bdevname(rdev
->bdev
,b
));
1801 INIT_LIST_HEAD(&rdev
->same_set
);
1806 if (rdev
->sb_page
) {
1812 return ERR_PTR(err
);
1816 * Check a full RAID array for plausibility
1820 static void analyze_sbs(mddev_t
* mddev
)
1823 struct list_head
*tmp
;
1824 mdk_rdev_t
*rdev
, *freshest
;
1825 char b
[BDEVNAME_SIZE
];
1828 ITERATE_RDEV(mddev
,rdev
,tmp
)
1829 switch (super_types
[mddev
->major_version
].
1830 load_super(rdev
, freshest
, mddev
->minor_version
)) {
1838 "md: fatal superblock inconsistency in %s"
1839 " -- removing from array\n",
1840 bdevname(rdev
->bdev
,b
));
1841 kick_rdev_from_array(rdev
);
1845 super_types
[mddev
->major_version
].
1846 validate_super(mddev
, freshest
);
1849 ITERATE_RDEV(mddev
,rdev
,tmp
) {
1850 if (rdev
!= freshest
)
1851 if (super_types
[mddev
->major_version
].
1852 validate_super(mddev
, rdev
)) {
1853 printk(KERN_WARNING
"md: kicking non-fresh %s"
1855 bdevname(rdev
->bdev
,b
));
1856 kick_rdev_from_array(rdev
);
1859 if (mddev
->level
== LEVEL_MULTIPATH
) {
1860 rdev
->desc_nr
= i
++;
1861 rdev
->raid_disk
= rdev
->desc_nr
;
1862 set_bit(In_sync
, &rdev
->flags
);
1868 if (mddev
->recovery_cp
!= MaxSector
&&
1870 printk(KERN_ERR
"md: %s: raid array is not clean"
1871 " -- starting background reconstruction\n",
1877 level_show(mddev_t
*mddev
, char *page
)
1879 struct mdk_personality
*p
= mddev
->pers
;
1881 return sprintf(page
, "%s\n", p
->name
);
1882 else if (mddev
->clevel
[0])
1883 return sprintf(page
, "%s\n", mddev
->clevel
);
1884 else if (mddev
->level
!= LEVEL_NONE
)
1885 return sprintf(page
, "%d\n", mddev
->level
);
1891 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1898 if (len
>= sizeof(mddev
->clevel
))
1900 strncpy(mddev
->clevel
, buf
, len
);
1901 if (mddev
->clevel
[len
-1] == '\n')
1903 mddev
->clevel
[len
] = 0;
1904 mddev
->level
= LEVEL_NONE
;
1908 static struct md_sysfs_entry md_level
=
1909 __ATTR(level
, 0644, level_show
, level_store
);
1912 raid_disks_show(mddev_t
*mddev
, char *page
)
1914 if (mddev
->raid_disks
== 0)
1916 return sprintf(page
, "%d\n", mddev
->raid_disks
);
1919 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
1922 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1924 /* can only set raid_disks if array is not yet active */
1927 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1929 if (!*buf
|| (*e
&& *e
!= '\n'))
1933 rv
= update_raid_disks(mddev
, n
);
1935 mddev
->raid_disks
= n
;
1936 return rv
? rv
: len
;
1938 static struct md_sysfs_entry md_raid_disks
=
1939 __ATTR(raid_disks
, 0644, raid_disks_show
, raid_disks_store
);
1942 chunk_size_show(mddev_t
*mddev
, char *page
)
1944 return sprintf(page
, "%d\n", mddev
->chunk_size
);
1948 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1950 /* can only set chunk_size if array is not yet active */
1952 unsigned long n
= simple_strtoul(buf
, &e
, 10);
1956 if (!*buf
|| (*e
&& *e
!= '\n'))
1959 mddev
->chunk_size
= n
;
1962 static struct md_sysfs_entry md_chunk_size
=
1963 __ATTR(chunk_size
, 0644, chunk_size_show
, chunk_size_store
);
1967 size_show(mddev_t
*mddev
, char *page
)
1969 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->size
);
1972 static int update_size(mddev_t
*mddev
, unsigned long size
);
1975 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
1977 /* If array is inactive, we can reduce the component size, but
1978 * not increase it (except from 0).
1979 * If array is active, we can try an on-line resize
1983 unsigned long long size
= simple_strtoull(buf
, &e
, 10);
1984 if (!*buf
|| *buf
== '\n' ||
1989 err
= update_size(mddev
, size
);
1990 md_update_sb(mddev
);
1992 if (mddev
->size
== 0 ||
1998 return err
? err
: len
;
2001 static struct md_sysfs_entry md_size
=
2002 __ATTR(component_size
, 0644, size_show
, size_store
);
2006 * This is either 'none' for arrays with externally managed metadata,
2007 * or N.M for internally known formats
2010 metadata_show(mddev_t
*mddev
, char *page
)
2012 if (mddev
->persistent
)
2013 return sprintf(page
, "%d.%d\n",
2014 mddev
->major_version
, mddev
->minor_version
);
2016 return sprintf(page
, "none\n");
2020 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2024 if (!list_empty(&mddev
->disks
))
2027 if (cmd_match(buf
, "none")) {
2028 mddev
->persistent
= 0;
2029 mddev
->major_version
= 0;
2030 mddev
->minor_version
= 90;
2033 major
= simple_strtoul(buf
, &e
, 10);
2034 if (e
==buf
|| *e
!= '.')
2037 minor
= simple_strtoul(buf
, &e
, 10);
2038 if (e
==buf
|| *e
!= '\n')
2040 if (major
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
2041 super_types
[major
].name
== NULL
)
2043 mddev
->major_version
= major
;
2044 mddev
->minor_version
= minor
;
2045 mddev
->persistent
= 1;
2049 static struct md_sysfs_entry md_metadata
=
2050 __ATTR(metadata_version
, 0644, metadata_show
, metadata_store
);
2053 action_show(mddev_t
*mddev
, char *page
)
2055 char *type
= "idle";
2056 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2057 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
)) {
2058 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
2059 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2061 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
2068 return sprintf(page
, "%s\n", type
);
2072 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
2074 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
2077 if (cmd_match(page
, "idle")) {
2078 if (mddev
->sync_thread
) {
2079 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2080 md_unregister_thread(mddev
->sync_thread
);
2081 mddev
->sync_thread
= NULL
;
2082 mddev
->recovery
= 0;
2084 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
2085 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
2087 else if (cmd_match(page
, "resync") || cmd_match(page
, "recover"))
2088 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2090 if (cmd_match(page
, "check"))
2091 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
2092 else if (cmd_match(page
, "repair"))
2094 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
2095 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
2097 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2098 md_wakeup_thread(mddev
->thread
);
2103 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
2105 return sprintf(page
, "%llu\n",
2106 (unsigned long long) mddev
->resync_mismatches
);
2109 static struct md_sysfs_entry
2110 md_scan_mode
= __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
2113 static struct md_sysfs_entry
2114 md_mismatches
= __ATTR_RO(mismatch_cnt
);
2116 static struct attribute
*md_default_attrs
[] = {
2118 &md_raid_disks
.attr
,
2119 &md_chunk_size
.attr
,
2125 static struct attribute
*md_redundancy_attrs
[] = {
2127 &md_mismatches
.attr
,
2130 static struct attribute_group md_redundancy_group
= {
2132 .attrs
= md_redundancy_attrs
,
2137 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2139 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2140 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2146 rv
= entry
->show(mddev
, page
);
2147 mddev_unlock(mddev
);
2152 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2153 const char *page
, size_t length
)
2155 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
2156 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
2162 rv
= entry
->store(mddev
, page
, length
);
2163 mddev_unlock(mddev
);
2167 static void md_free(struct kobject
*ko
)
2169 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
2173 static struct sysfs_ops md_sysfs_ops
= {
2174 .show
= md_attr_show
,
2175 .store
= md_attr_store
,
2177 static struct kobj_type md_ktype
= {
2179 .sysfs_ops
= &md_sysfs_ops
,
2180 .default_attrs
= md_default_attrs
,
2185 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
2187 static DECLARE_MUTEX(disks_sem
);
2188 mddev_t
*mddev
= mddev_find(dev
);
2189 struct gendisk
*disk
;
2190 int partitioned
= (MAJOR(dev
) != MD_MAJOR
);
2191 int shift
= partitioned
? MdpMinorShift
: 0;
2192 int unit
= MINOR(dev
) >> shift
;
2198 if (mddev
->gendisk
) {
2203 disk
= alloc_disk(1 << shift
);
2209 disk
->major
= MAJOR(dev
);
2210 disk
->first_minor
= unit
<< shift
;
2212 sprintf(disk
->disk_name
, "md_d%d", unit
);
2213 sprintf(disk
->devfs_name
, "md/d%d", unit
);
2215 sprintf(disk
->disk_name
, "md%d", unit
);
2216 sprintf(disk
->devfs_name
, "md/%d", unit
);
2218 disk
->fops
= &md_fops
;
2219 disk
->private_data
= mddev
;
2220 disk
->queue
= mddev
->queue
;
2222 mddev
->gendisk
= disk
;
2224 mddev
->kobj
.parent
= &disk
->kobj
;
2225 mddev
->kobj
.k_name
= NULL
;
2226 snprintf(mddev
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "md");
2227 mddev
->kobj
.ktype
= &md_ktype
;
2228 kobject_register(&mddev
->kobj
);
2232 void md_wakeup_thread(mdk_thread_t
*thread
);
2234 static void md_safemode_timeout(unsigned long data
)
2236 mddev_t
*mddev
= (mddev_t
*) data
;
2238 mddev
->safemode
= 1;
2239 md_wakeup_thread(mddev
->thread
);
2242 static int start_dirty_degraded
;
2244 static int do_md_run(mddev_t
* mddev
)
2248 struct list_head
*tmp
;
2250 struct gendisk
*disk
;
2251 struct mdk_personality
*pers
;
2252 char b
[BDEVNAME_SIZE
];
2254 if (list_empty(&mddev
->disks
))
2255 /* cannot run an array with no devices.. */
2262 * Analyze all RAID superblock(s)
2264 if (!mddev
->raid_disks
)
2267 chunk_size
= mddev
->chunk_size
;
2270 if (chunk_size
> MAX_CHUNK_SIZE
) {
2271 printk(KERN_ERR
"too big chunk_size: %d > %d\n",
2272 chunk_size
, MAX_CHUNK_SIZE
);
2276 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2278 if ( (1 << ffz(~chunk_size
)) != chunk_size
) {
2279 printk(KERN_ERR
"chunk_size of %d not valid\n", chunk_size
);
2282 if (chunk_size
< PAGE_SIZE
) {
2283 printk(KERN_ERR
"too small chunk_size: %d < %ld\n",
2284 chunk_size
, PAGE_SIZE
);
2288 /* devices must have minimum size of one chunk */
2289 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2290 if (test_bit(Faulty
, &rdev
->flags
))
2292 if (rdev
->size
< chunk_size
/ 1024) {
2294 "md: Dev %s smaller than chunk_size:"
2296 bdevname(rdev
->bdev
,b
),
2297 (unsigned long long)rdev
->size
,
2305 if (mddev
->level
!= LEVEL_NONE
)
2306 request_module("md-level-%d", mddev
->level
);
2307 else if (mddev
->clevel
[0])
2308 request_module("md-%s", mddev
->clevel
);
2312 * Drop all container device buffers, from now on
2313 * the only valid external interface is through the md
2315 * Also find largest hardsector size
2317 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2318 if (test_bit(Faulty
, &rdev
->flags
))
2320 sync_blockdev(rdev
->bdev
);
2321 invalidate_bdev(rdev
->bdev
, 0);
2324 md_probe(mddev
->unit
, NULL
, NULL
);
2325 disk
= mddev
->gendisk
;
2329 spin_lock(&pers_lock
);
2330 pers
= find_pers(mddev
->level
, mddev
->clevel
);
2331 if (!pers
|| !try_module_get(pers
->owner
)) {
2332 spin_unlock(&pers_lock
);
2333 if (mddev
->level
!= LEVEL_NONE
)
2334 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
2337 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
2342 spin_unlock(&pers_lock
);
2343 mddev
->level
= pers
->level
;
2344 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2346 mddev
->recovery
= 0;
2347 mddev
->resync_max_sectors
= mddev
->size
<< 1; /* may be over-ridden by personality */
2348 mddev
->barriers_work
= 1;
2349 mddev
->ok_start_degraded
= start_dirty_degraded
;
2352 mddev
->ro
= 2; /* read-only, but switch on first write */
2354 err
= mddev
->pers
->run(mddev
);
2355 if (!err
&& mddev
->pers
->sync_request
) {
2356 err
= bitmap_create(mddev
);
2358 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
2359 mdname(mddev
), err
);
2360 mddev
->pers
->stop(mddev
);
2364 printk(KERN_ERR
"md: pers->run() failed ...\n");
2365 module_put(mddev
->pers
->owner
);
2367 bitmap_destroy(mddev
);
2370 if (mddev
->pers
->sync_request
)
2371 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
);
2372 else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
2375 atomic_set(&mddev
->writes_pending
,0);
2376 mddev
->safemode
= 0;
2377 mddev
->safemode_timer
.function
= md_safemode_timeout
;
2378 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
2379 mddev
->safemode_delay
= (20 * HZ
)/1000 +1; /* 20 msec delay */
2382 ITERATE_RDEV(mddev
,rdev
,tmp
)
2383 if (rdev
->raid_disk
>= 0) {
2385 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2386 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
2389 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2390 md_wakeup_thread(mddev
->thread
);
2392 if (mddev
->sb_dirty
)
2393 md_update_sb(mddev
);
2395 set_capacity(disk
, mddev
->array_size
<<1);
2397 /* If we call blk_queue_make_request here, it will
2398 * re-initialise max_sectors etc which may have been
2399 * refined inside -> run. So just set the bits we need to set.
2400 * Most initialisation happended when we called
2401 * blk_queue_make_request(..., md_fail_request)
2404 mddev
->queue
->queuedata
= mddev
;
2405 mddev
->queue
->make_request_fn
= mddev
->pers
->make_request
;
2408 md_new_event(mddev
);
2412 static int restart_array(mddev_t
*mddev
)
2414 struct gendisk
*disk
= mddev
->gendisk
;
2418 * Complain if it has no devices
2421 if (list_empty(&mddev
->disks
))
2429 mddev
->safemode
= 0;
2431 set_disk_ro(disk
, 0);
2433 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
2436 * Kick recovery or resync if necessary
2438 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2439 md_wakeup_thread(mddev
->thread
);
2442 printk(KERN_ERR
"md: %s has no personality assigned.\n",
2451 static int do_md_stop(mddev_t
* mddev
, int ro
)
2454 struct gendisk
*disk
= mddev
->gendisk
;
2457 if (atomic_read(&mddev
->active
)>2) {
2458 printk("md: %s still in use.\n",mdname(mddev
));
2462 if (mddev
->sync_thread
) {
2463 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2464 md_unregister_thread(mddev
->sync_thread
);
2465 mddev
->sync_thread
= NULL
;
2468 del_timer_sync(&mddev
->safemode_timer
);
2470 invalidate_partition(disk
, 0);
2478 bitmap_flush(mddev
);
2479 md_super_wait(mddev
);
2481 set_disk_ro(disk
, 0);
2482 blk_queue_make_request(mddev
->queue
, md_fail_request
);
2483 mddev
->pers
->stop(mddev
);
2484 if (mddev
->pers
->sync_request
)
2485 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
2487 module_put(mddev
->pers
->owner
);
2492 if (!mddev
->in_sync
) {
2493 /* mark array as shutdown cleanly */
2495 md_update_sb(mddev
);
2498 set_disk_ro(disk
, 1);
2501 bitmap_destroy(mddev
);
2502 if (mddev
->bitmap_file
) {
2503 atomic_set(&mddev
->bitmap_file
->f_dentry
->d_inode
->i_writecount
, 1);
2504 fput(mddev
->bitmap_file
);
2505 mddev
->bitmap_file
= NULL
;
2507 mddev
->bitmap_offset
= 0;
2510 * Free resources if final stop
2514 struct list_head
*tmp
;
2515 struct gendisk
*disk
;
2516 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
2518 ITERATE_RDEV(mddev
,rdev
,tmp
)
2519 if (rdev
->raid_disk
>= 0) {
2521 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2522 sysfs_remove_link(&mddev
->kobj
, nm
);
2525 export_array(mddev
);
2527 mddev
->array_size
= 0;
2528 disk
= mddev
->gendisk
;
2530 set_capacity(disk
, 0);
2533 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
2536 md_new_event(mddev
);
2541 static void autorun_array(mddev_t
*mddev
)
2544 struct list_head
*tmp
;
2547 if (list_empty(&mddev
->disks
))
2550 printk(KERN_INFO
"md: running: ");
2552 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2553 char b
[BDEVNAME_SIZE
];
2554 printk("<%s>", bdevname(rdev
->bdev
,b
));
2558 err
= do_md_run (mddev
);
2560 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
2561 do_md_stop (mddev
, 0);
2566 * lets try to run arrays based on all disks that have arrived
2567 * until now. (those are in pending_raid_disks)
2569 * the method: pick the first pending disk, collect all disks with
2570 * the same UUID, remove all from the pending list and put them into
2571 * the 'same_array' list. Then order this list based on superblock
2572 * update time (freshest comes first), kick out 'old' disks and
2573 * compare superblocks. If everything's fine then run it.
2575 * If "unit" is allocated, then bump its reference count
2577 static void autorun_devices(int part
)
2579 struct list_head candidates
;
2580 struct list_head
*tmp
;
2581 mdk_rdev_t
*rdev0
, *rdev
;
2583 char b
[BDEVNAME_SIZE
];
2585 printk(KERN_INFO
"md: autorun ...\n");
2586 while (!list_empty(&pending_raid_disks
)) {
2588 rdev0
= list_entry(pending_raid_disks
.next
,
2589 mdk_rdev_t
, same_set
);
2591 printk(KERN_INFO
"md: considering %s ...\n",
2592 bdevname(rdev0
->bdev
,b
));
2593 INIT_LIST_HEAD(&candidates
);
2594 ITERATE_RDEV_PENDING(rdev
,tmp
)
2595 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
2596 printk(KERN_INFO
"md: adding %s ...\n",
2597 bdevname(rdev
->bdev
,b
));
2598 list_move(&rdev
->same_set
, &candidates
);
2601 * now we have a set of devices, with all of them having
2602 * mostly sane superblocks. It's time to allocate the
2605 if (rdev0
->preferred_minor
< 0 || rdev0
->preferred_minor
>= MAX_MD_DEVS
) {
2606 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
2607 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
2611 dev
= MKDEV(mdp_major
,
2612 rdev0
->preferred_minor
<< MdpMinorShift
);
2614 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
2616 md_probe(dev
, NULL
, NULL
);
2617 mddev
= mddev_find(dev
);
2620 "md: cannot allocate memory for md drive.\n");
2623 if (mddev_lock(mddev
))
2624 printk(KERN_WARNING
"md: %s locked, cannot run\n",
2626 else if (mddev
->raid_disks
|| mddev
->major_version
2627 || !list_empty(&mddev
->disks
)) {
2629 "md: %s already running, cannot run %s\n",
2630 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
2631 mddev_unlock(mddev
);
2633 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
2634 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
) {
2635 list_del_init(&rdev
->same_set
);
2636 if (bind_rdev_to_array(rdev
, mddev
))
2639 autorun_array(mddev
);
2640 mddev_unlock(mddev
);
2642 /* on success, candidates will be empty, on error
2645 ITERATE_RDEV_GENERIC(candidates
,rdev
,tmp
)
2649 printk(KERN_INFO
"md: ... autorun DONE.\n");
2653 * import RAID devices based on one partition
2654 * if possible, the array gets run as well.
2657 static int autostart_array(dev_t startdev
)
2659 char b
[BDEVNAME_SIZE
];
2660 int err
= -EINVAL
, i
;
2661 mdp_super_t
*sb
= NULL
;
2662 mdk_rdev_t
*start_rdev
= NULL
, *rdev
;
2664 start_rdev
= md_import_device(startdev
, 0, 0);
2665 if (IS_ERR(start_rdev
))
2669 /* NOTE: this can only work for 0.90.0 superblocks */
2670 sb
= (mdp_super_t
*)page_address(start_rdev
->sb_page
);
2671 if (sb
->major_version
!= 0 ||
2672 sb
->minor_version
!= 90 ) {
2673 printk(KERN_WARNING
"md: can only autostart 0.90.0 arrays\n");
2674 export_rdev(start_rdev
);
2678 if (test_bit(Faulty
, &start_rdev
->flags
)) {
2680 "md: can not autostart based on faulty %s!\n",
2681 bdevname(start_rdev
->bdev
,b
));
2682 export_rdev(start_rdev
);
2685 list_add(&start_rdev
->same_set
, &pending_raid_disks
);
2687 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2688 mdp_disk_t
*desc
= sb
->disks
+ i
;
2689 dev_t dev
= MKDEV(desc
->major
, desc
->minor
);
2693 if (dev
== startdev
)
2695 if (MAJOR(dev
) != desc
->major
|| MINOR(dev
) != desc
->minor
)
2697 rdev
= md_import_device(dev
, 0, 0);
2701 list_add(&rdev
->same_set
, &pending_raid_disks
);
2705 * possibly return codes
2713 static int get_version(void __user
* arg
)
2717 ver
.major
= MD_MAJOR_VERSION
;
2718 ver
.minor
= MD_MINOR_VERSION
;
2719 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
2721 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
2727 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
2729 mdu_array_info_t info
;
2730 int nr
,working
,active
,failed
,spare
;
2732 struct list_head
*tmp
;
2734 nr
=working
=active
=failed
=spare
=0;
2735 ITERATE_RDEV(mddev
,rdev
,tmp
) {
2737 if (test_bit(Faulty
, &rdev
->flags
))
2741 if (test_bit(In_sync
, &rdev
->flags
))
2748 info
.major_version
= mddev
->major_version
;
2749 info
.minor_version
= mddev
->minor_version
;
2750 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
2751 info
.ctime
= mddev
->ctime
;
2752 info
.level
= mddev
->level
;
2753 info
.size
= mddev
->size
;
2755 info
.raid_disks
= mddev
->raid_disks
;
2756 info
.md_minor
= mddev
->md_minor
;
2757 info
.not_persistent
= !mddev
->persistent
;
2759 info
.utime
= mddev
->utime
;
2762 info
.state
= (1<<MD_SB_CLEAN
);
2763 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
2764 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
2765 info
.active_disks
= active
;
2766 info
.working_disks
= working
;
2767 info
.failed_disks
= failed
;
2768 info
.spare_disks
= spare
;
2770 info
.layout
= mddev
->layout
;
2771 info
.chunk_size
= mddev
->chunk_size
;
2773 if (copy_to_user(arg
, &info
, sizeof(info
)))
2779 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
2781 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
2782 char *ptr
, *buf
= NULL
;
2785 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
2789 /* bitmap disabled, zero the first byte and copy out */
2790 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
2791 file
->pathname
[0] = '\0';
2795 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
2799 ptr
= file_path(mddev
->bitmap
->file
, buf
, sizeof(file
->pathname
));
2803 strcpy(file
->pathname
, ptr
);
2807 if (copy_to_user(arg
, file
, sizeof(*file
)))
2815 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
2817 mdu_disk_info_t info
;
2821 if (copy_from_user(&info
, arg
, sizeof(info
)))
2826 rdev
= find_rdev_nr(mddev
, nr
);
2828 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
2829 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
2830 info
.raid_disk
= rdev
->raid_disk
;
2832 if (test_bit(Faulty
, &rdev
->flags
))
2833 info
.state
|= (1<<MD_DISK_FAULTY
);
2834 else if (test_bit(In_sync
, &rdev
->flags
)) {
2835 info
.state
|= (1<<MD_DISK_ACTIVE
);
2836 info
.state
|= (1<<MD_DISK_SYNC
);
2838 if (test_bit(WriteMostly
, &rdev
->flags
))
2839 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
2841 info
.major
= info
.minor
= 0;
2842 info
.raid_disk
= -1;
2843 info
.state
= (1<<MD_DISK_REMOVED
);
2846 if (copy_to_user(arg
, &info
, sizeof(info
)))
2852 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
2854 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
2856 dev_t dev
= MKDEV(info
->major
,info
->minor
);
2858 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
2861 if (!mddev
->raid_disks
) {
2863 /* expecting a device which has a superblock */
2864 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
2867 "md: md_import_device returned %ld\n",
2869 return PTR_ERR(rdev
);
2871 if (!list_empty(&mddev
->disks
)) {
2872 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
2873 mdk_rdev_t
, same_set
);
2874 int err
= super_types
[mddev
->major_version
]
2875 .load_super(rdev
, rdev0
, mddev
->minor_version
);
2878 "md: %s has different UUID to %s\n",
2879 bdevname(rdev
->bdev
,b
),
2880 bdevname(rdev0
->bdev
,b2
));
2885 err
= bind_rdev_to_array(rdev
, mddev
);
2892 * add_new_disk can be used once the array is assembled
2893 * to add "hot spares". They must already have a superblock
2898 if (!mddev
->pers
->hot_add_disk
) {
2900 "%s: personality does not support diskops!\n",
2904 if (mddev
->persistent
)
2905 rdev
= md_import_device(dev
, mddev
->major_version
,
2906 mddev
->minor_version
);
2908 rdev
= md_import_device(dev
, -1, -1);
2911 "md: md_import_device returned %ld\n",
2913 return PTR_ERR(rdev
);
2915 /* set save_raid_disk if appropriate */
2916 if (!mddev
->persistent
) {
2917 if (info
->state
& (1<<MD_DISK_SYNC
) &&
2918 info
->raid_disk
< mddev
->raid_disks
)
2919 rdev
->raid_disk
= info
->raid_disk
;
2921 rdev
->raid_disk
= -1;
2923 super_types
[mddev
->major_version
].
2924 validate_super(mddev
, rdev
);
2925 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2927 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
2928 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
2929 set_bit(WriteMostly
, &rdev
->flags
);
2931 rdev
->raid_disk
= -1;
2932 err
= bind_rdev_to_array(rdev
, mddev
);
2936 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2937 md_wakeup_thread(mddev
->thread
);
2941 /* otherwise, add_new_disk is only allowed
2942 * for major_version==0 superblocks
2944 if (mddev
->major_version
!= 0) {
2945 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
2950 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
2952 rdev
= md_import_device (dev
, -1, 0);
2955 "md: error, md_import_device() returned %ld\n",
2957 return PTR_ERR(rdev
);
2959 rdev
->desc_nr
= info
->number
;
2960 if (info
->raid_disk
< mddev
->raid_disks
)
2961 rdev
->raid_disk
= info
->raid_disk
;
2963 rdev
->raid_disk
= -1;
2967 if (rdev
->raid_disk
< mddev
->raid_disks
)
2968 if (info
->state
& (1<<MD_DISK_SYNC
))
2969 set_bit(In_sync
, &rdev
->flags
);
2971 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
2972 set_bit(WriteMostly
, &rdev
->flags
);
2974 if (!mddev
->persistent
) {
2975 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
2976 rdev
->sb_offset
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2978 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
2979 rdev
->size
= calc_dev_size(rdev
, mddev
->chunk_size
);
2981 err
= bind_rdev_to_array(rdev
, mddev
);
2991 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
2993 char b
[BDEVNAME_SIZE
];
2999 rdev
= find_rdev(mddev
, dev
);
3003 if (rdev
->raid_disk
>= 0)
3006 kick_rdev_from_array(rdev
);
3007 md_update_sb(mddev
);
3008 md_new_event(mddev
);
3012 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ... \n",
3013 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3017 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
3019 char b
[BDEVNAME_SIZE
];
3027 if (mddev
->major_version
!= 0) {
3028 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
3029 " version-0 superblocks.\n",
3033 if (!mddev
->pers
->hot_add_disk
) {
3035 "%s: personality does not support diskops!\n",
3040 rdev
= md_import_device (dev
, -1, 0);
3043 "md: error, md_import_device() returned %ld\n",
3048 if (mddev
->persistent
)
3049 rdev
->sb_offset
= calc_dev_sboffset(rdev
->bdev
);
3052 rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
3054 size
= calc_dev_size(rdev
, mddev
->chunk_size
);
3057 if (test_bit(Faulty
, &rdev
->flags
)) {
3059 "md: can not hot-add faulty %s disk to %s!\n",
3060 bdevname(rdev
->bdev
,b
), mdname(mddev
));
3064 clear_bit(In_sync
, &rdev
->flags
);
3066 err
= bind_rdev_to_array(rdev
, mddev
);
3071 * The rest should better be atomic, we can have disk failures
3072 * noticed in interrupt contexts ...
3075 if (rdev
->desc_nr
== mddev
->max_disks
) {
3076 printk(KERN_WARNING
"%s: can not hot-add to full array!\n",
3079 goto abort_unbind_export
;
3082 rdev
->raid_disk
= -1;
3084 md_update_sb(mddev
);
3087 * Kick recovery, maybe this spare has to be added to the
3088 * array immediately.
3090 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3091 md_wakeup_thread(mddev
->thread
);
3092 md_new_event(mddev
);
3095 abort_unbind_export
:
3096 unbind_rdev_from_array(rdev
);
3103 /* similar to deny_write_access, but accounts for our holding a reference
3104 * to the file ourselves */
3105 static int deny_bitmap_write_access(struct file
* file
)
3107 struct inode
*inode
= file
->f_mapping
->host
;
3109 spin_lock(&inode
->i_lock
);
3110 if (atomic_read(&inode
->i_writecount
) > 1) {
3111 spin_unlock(&inode
->i_lock
);
3114 atomic_set(&inode
->i_writecount
, -1);
3115 spin_unlock(&inode
->i_lock
);
3120 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
3125 if (!mddev
->pers
->quiesce
)
3127 if (mddev
->recovery
|| mddev
->sync_thread
)
3129 /* we should be able to change the bitmap.. */
3135 return -EEXIST
; /* cannot add when bitmap is present */
3136 mddev
->bitmap_file
= fget(fd
);
3138 if (mddev
->bitmap_file
== NULL
) {
3139 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
3144 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
3146 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
3148 fput(mddev
->bitmap_file
);
3149 mddev
->bitmap_file
= NULL
;
3152 mddev
->bitmap_offset
= 0; /* file overrides offset */
3153 } else if (mddev
->bitmap
== NULL
)
3154 return -ENOENT
; /* cannot remove what isn't there */
3157 mddev
->pers
->quiesce(mddev
, 1);
3159 err
= bitmap_create(mddev
);
3161 bitmap_destroy(mddev
);
3162 mddev
->pers
->quiesce(mddev
, 0);
3163 } else if (fd
< 0) {
3164 if (mddev
->bitmap_file
)
3165 fput(mddev
->bitmap_file
);
3166 mddev
->bitmap_file
= NULL
;
3173 * set_array_info is used two different ways
3174 * The original usage is when creating a new array.
3175 * In this usage, raid_disks is > 0 and it together with
3176 * level, size, not_persistent,layout,chunksize determine the
3177 * shape of the array.
3178 * This will always create an array with a type-0.90.0 superblock.
3179 * The newer usage is when assembling an array.
3180 * In this case raid_disks will be 0, and the major_version field is
3181 * use to determine which style super-blocks are to be found on the devices.
3182 * The minor and patch _version numbers are also kept incase the
3183 * super_block handler wishes to interpret them.
3185 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
3188 if (info
->raid_disks
== 0) {
3189 /* just setting version number for superblock loading */
3190 if (info
->major_version
< 0 ||
3191 info
->major_version
>= sizeof(super_types
)/sizeof(super_types
[0]) ||
3192 super_types
[info
->major_version
].name
== NULL
) {
3193 /* maybe try to auto-load a module? */
3195 "md: superblock version %d not known\n",
3196 info
->major_version
);
3199 mddev
->major_version
= info
->major_version
;
3200 mddev
->minor_version
= info
->minor_version
;
3201 mddev
->patch_version
= info
->patch_version
;
3204 mddev
->major_version
= MD_MAJOR_VERSION
;
3205 mddev
->minor_version
= MD_MINOR_VERSION
;
3206 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
3207 mddev
->ctime
= get_seconds();
3209 mddev
->level
= info
->level
;
3210 mddev
->size
= info
->size
;
3211 mddev
->raid_disks
= info
->raid_disks
;
3212 /* don't set md_minor, it is determined by which /dev/md* was
3215 if (info
->state
& (1<<MD_SB_CLEAN
))
3216 mddev
->recovery_cp
= MaxSector
;
3218 mddev
->recovery_cp
= 0;
3219 mddev
->persistent
= ! info
->not_persistent
;
3221 mddev
->layout
= info
->layout
;
3222 mddev
->chunk_size
= info
->chunk_size
;
3224 mddev
->max_disks
= MD_SB_DISKS
;
3226 mddev
->sb_dirty
= 1;
3228 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
3229 mddev
->bitmap_offset
= 0;
3232 * Generate a 128 bit UUID
3234 get_random_bytes(mddev
->uuid
, 16);
3239 static int update_size(mddev_t
*mddev
, unsigned long size
)
3243 struct list_head
*tmp
;
3245 if (mddev
->pers
->resize
== NULL
)
3247 /* The "size" is the amount of each device that is used.
3248 * This can only make sense for arrays with redundancy.
3249 * linear and raid0 always use whatever space is available
3250 * We can only consider changing the size if no resync
3251 * or reconstruction is happening, and if the new size
3252 * is acceptable. It must fit before the sb_offset or,
3253 * if that is <data_offset, it must fit before the
3254 * size of each device.
3255 * If size is zero, we find the largest size that fits.
3257 if (mddev
->sync_thread
)
3259 ITERATE_RDEV(mddev
,rdev
,tmp
) {
3261 int fit
= (size
== 0);
3262 if (rdev
->sb_offset
> rdev
->data_offset
)
3263 avail
= (rdev
->sb_offset
*2) - rdev
->data_offset
;
3265 avail
= get_capacity(rdev
->bdev
->bd_disk
)
3266 - rdev
->data_offset
;
3267 if (fit
&& (size
== 0 || size
> avail
/2))
3269 if (avail
< ((sector_t
)size
<< 1))
3272 rv
= mddev
->pers
->resize(mddev
, (sector_t
)size
*2);
3274 struct block_device
*bdev
;
3276 bdev
= bdget_disk(mddev
->gendisk
, 0);
3278 down(&bdev
->bd_inode
->i_sem
);
3279 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
3280 up(&bdev
->bd_inode
->i_sem
);
3287 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
3290 /* change the number of raid disks */
3291 if (mddev
->pers
->reshape
== NULL
)
3293 if (raid_disks
<= 0 ||
3294 raid_disks
>= mddev
->max_disks
)
3296 if (mddev
->sync_thread
)
3298 rv
= mddev
->pers
->reshape(mddev
, raid_disks
);
3300 struct block_device
*bdev
;
3302 bdev
= bdget_disk(mddev
->gendisk
, 0);
3304 down(&bdev
->bd_inode
->i_sem
);
3305 i_size_write(bdev
->bd_inode
, mddev
->array_size
<< 10);
3306 up(&bdev
->bd_inode
->i_sem
);
3315 * update_array_info is used to change the configuration of an
3317 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3318 * fields in the info are checked against the array.
3319 * Any differences that cannot be handled will cause an error.
3320 * Normally, only one change can be managed at a time.
3322 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
3328 /* calculate expected state,ignoring low bits */
3329 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
3330 state
|= (1 << MD_SB_BITMAP_PRESENT
);
3332 if (mddev
->major_version
!= info
->major_version
||
3333 mddev
->minor_version
!= info
->minor_version
||
3334 /* mddev->patch_version != info->patch_version || */
3335 mddev
->ctime
!= info
->ctime
||
3336 mddev
->level
!= info
->level
||
3337 /* mddev->layout != info->layout || */
3338 !mddev
->persistent
!= info
->not_persistent
||
3339 mddev
->chunk_size
!= info
->chunk_size
||
3340 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3341 ((state
^info
->state
) & 0xfffffe00)
3344 /* Check there is only one change */
3345 if (mddev
->size
!= info
->size
) cnt
++;
3346 if (mddev
->raid_disks
!= info
->raid_disks
) cnt
++;
3347 if (mddev
->layout
!= info
->layout
) cnt
++;
3348 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) cnt
++;
3349 if (cnt
== 0) return 0;
3350 if (cnt
> 1) return -EINVAL
;
3352 if (mddev
->layout
!= info
->layout
) {
3354 * we don't need to do anything at the md level, the
3355 * personality will take care of it all.
3357 if (mddev
->pers
->reconfig
== NULL
)
3360 return mddev
->pers
->reconfig(mddev
, info
->layout
, -1);
3362 if (mddev
->size
!= info
->size
)
3363 rv
= update_size(mddev
, info
->size
);
3365 if (mddev
->raid_disks
!= info
->raid_disks
)
3366 rv
= update_raid_disks(mddev
, info
->raid_disks
);
3368 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
3369 if (mddev
->pers
->quiesce
== NULL
)
3371 if (mddev
->recovery
|| mddev
->sync_thread
)
3373 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
3374 /* add the bitmap */
3377 if (mddev
->default_bitmap_offset
== 0)
3379 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
3380 mddev
->pers
->quiesce(mddev
, 1);
3381 rv
= bitmap_create(mddev
);
3383 bitmap_destroy(mddev
);
3384 mddev
->pers
->quiesce(mddev
, 0);
3386 /* remove the bitmap */
3389 if (mddev
->bitmap
->file
)
3391 mddev
->pers
->quiesce(mddev
, 1);
3392 bitmap_destroy(mddev
);
3393 mddev
->pers
->quiesce(mddev
, 0);
3394 mddev
->bitmap_offset
= 0;
3397 md_update_sb(mddev
);
3401 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
3405 if (mddev
->pers
== NULL
)
3408 rdev
= find_rdev(mddev
, dev
);
3412 md_error(mddev
, rdev
);
3416 static int md_ioctl(struct inode
*inode
, struct file
*file
,
3417 unsigned int cmd
, unsigned long arg
)
3420 void __user
*argp
= (void __user
*)arg
;
3421 struct hd_geometry __user
*loc
= argp
;
3422 mddev_t
*mddev
= NULL
;
3424 if (!capable(CAP_SYS_ADMIN
))
3428 * Commands dealing with the RAID driver but not any
3434 err
= get_version(argp
);
3437 case PRINT_RAID_DEBUG
:
3445 autostart_arrays(arg
);
3452 * Commands creating/starting a new array:
3455 mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3463 if (cmd
== START_ARRAY
) {
3464 /* START_ARRAY doesn't need to lock the array as autostart_array
3465 * does the locking, and it could even be a different array
3470 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3471 "This will not be supported beyond July 2006\n",
3472 current
->comm
, current
->pid
);
3475 err
= autostart_array(new_decode_dev(arg
));
3477 printk(KERN_WARNING
"md: autostart failed!\n");
3483 err
= mddev_lock(mddev
);
3486 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3493 case SET_ARRAY_INFO
:
3495 mdu_array_info_t info
;
3497 memset(&info
, 0, sizeof(info
));
3498 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
3503 err
= update_array_info(mddev
, &info
);
3505 printk(KERN_WARNING
"md: couldn't update"
3506 " array info. %d\n", err
);
3511 if (!list_empty(&mddev
->disks
)) {
3513 "md: array %s already has disks!\n",
3518 if (mddev
->raid_disks
) {
3520 "md: array %s already initialised!\n",
3525 err
= set_array_info(mddev
, &info
);
3527 printk(KERN_WARNING
"md: couldn't set"
3528 " array info. %d\n", err
);
3538 * Commands querying/configuring an existing array:
3540 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3541 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3542 if (!mddev
->raid_disks
&& cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
3543 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
) {
3549 * Commands even a read-only array can execute:
3553 case GET_ARRAY_INFO
:
3554 err
= get_array_info(mddev
, argp
);
3557 case GET_BITMAP_FILE
:
3558 err
= get_bitmap_file(mddev
, argp
);
3562 err
= get_disk_info(mddev
, argp
);
3565 case RESTART_ARRAY_RW
:
3566 err
= restart_array(mddev
);
3570 err
= do_md_stop (mddev
, 0);
3574 err
= do_md_stop (mddev
, 1);
3578 * We have a problem here : there is no easy way to give a CHS
3579 * virtual geometry. We currently pretend that we have a 2 heads
3580 * 4 sectors (with a BIG number of cylinders...). This drives
3581 * dosfs just mad... ;-)
3588 err
= put_user (2, (char __user
*) &loc
->heads
);
3591 err
= put_user (4, (char __user
*) &loc
->sectors
);
3594 err
= put_user(get_capacity(mddev
->gendisk
)/8,
3595 (short __user
*) &loc
->cylinders
);
3598 err
= put_user (get_start_sect(inode
->i_bdev
),
3599 (long __user
*) &loc
->start
);
3604 * The remaining ioctls are changing the state of the
3605 * superblock, so we do not allow them on read-only arrays.
3606 * However non-MD ioctls (e.g. get-size) will still come through
3607 * here and hit the 'default' below, so only disallow
3608 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3610 if (_IOC_TYPE(cmd
) == MD_MAJOR
&&
3611 mddev
->ro
&& mddev
->pers
) {
3612 if (mddev
->ro
== 2) {
3614 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3615 md_wakeup_thread(mddev
->thread
);
3627 mdu_disk_info_t info
;
3628 if (copy_from_user(&info
, argp
, sizeof(info
)))
3631 err
= add_new_disk(mddev
, &info
);
3635 case HOT_REMOVE_DISK
:
3636 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
3640 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
3643 case SET_DISK_FAULTY
:
3644 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
3648 err
= do_md_run (mddev
);
3651 case SET_BITMAP_FILE
:
3652 err
= set_bitmap_file(mddev
, (int)arg
);
3656 if (_IOC_TYPE(cmd
) == MD_MAJOR
)
3657 printk(KERN_WARNING
"md: %s(pid %d) used"
3658 " obsolete MD ioctl, upgrade your"
3659 " software to use new ictls.\n",
3660 current
->comm
, current
->pid
);
3667 mddev_unlock(mddev
);
3677 static int md_open(struct inode
*inode
, struct file
*file
)
3680 * Succeed if we can lock the mddev, which confirms that
3681 * it isn't being stopped right now.
3683 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3686 if ((err
= mddev_lock(mddev
)))
3691 mddev_unlock(mddev
);
3693 check_disk_change(inode
->i_bdev
);
3698 static int md_release(struct inode
*inode
, struct file
* file
)
3700 mddev_t
*mddev
= inode
->i_bdev
->bd_disk
->private_data
;
3709 static int md_media_changed(struct gendisk
*disk
)
3711 mddev_t
*mddev
= disk
->private_data
;
3713 return mddev
->changed
;
3716 static int md_revalidate(struct gendisk
*disk
)
3718 mddev_t
*mddev
= disk
->private_data
;
3723 static struct block_device_operations md_fops
=
3725 .owner
= THIS_MODULE
,
3727 .release
= md_release
,
3729 .media_changed
= md_media_changed
,
3730 .revalidate_disk
= md_revalidate
,
3733 static int md_thread(void * arg
)
3735 mdk_thread_t
*thread
= arg
;
3738 * md_thread is a 'system-thread', it's priority should be very
3739 * high. We avoid resource deadlocks individually in each
3740 * raid personality. (RAID5 does preallocation) We also use RR and
3741 * the very same RT priority as kswapd, thus we will never get
3742 * into a priority inversion deadlock.
3744 * we definitely have to have equal or higher priority than
3745 * bdflush, otherwise bdflush will deadlock if there are too
3746 * many dirty RAID5 blocks.
3749 allow_signal(SIGKILL
);
3750 while (!kthread_should_stop()) {
3752 /* We need to wait INTERRUPTIBLE so that
3753 * we don't add to the load-average.
3754 * That means we need to be sure no signals are
3757 if (signal_pending(current
))
3758 flush_signals(current
);
3760 wait_event_interruptible_timeout
3762 test_bit(THREAD_WAKEUP
, &thread
->flags
)
3763 || kthread_should_stop(),
3767 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
3769 thread
->run(thread
->mddev
);
3775 void md_wakeup_thread(mdk_thread_t
*thread
)
3778 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
3779 set_bit(THREAD_WAKEUP
, &thread
->flags
);
3780 wake_up(&thread
->wqueue
);
3784 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
3787 mdk_thread_t
*thread
;
3789 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
3793 init_waitqueue_head(&thread
->wqueue
);
3796 thread
->mddev
= mddev
;
3797 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
3798 thread
->tsk
= kthread_run(md_thread
, thread
, name
, mdname(thread
->mddev
));
3799 if (IS_ERR(thread
->tsk
)) {
3806 void md_unregister_thread(mdk_thread_t
*thread
)
3808 dprintk("interrupting MD-thread pid %d\n", thread
->tsk
->pid
);
3810 kthread_stop(thread
->tsk
);
3814 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
3821 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
3824 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3826 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3827 __builtin_return_address(0),__builtin_return_address(1),
3828 __builtin_return_address(2),__builtin_return_address(3));
3830 if (!mddev
->pers
->error_handler
)
3832 mddev
->pers
->error_handler(mddev
,rdev
);
3833 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3834 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3835 md_wakeup_thread(mddev
->thread
);
3836 md_new_event(mddev
);
3839 /* seq_file implementation /proc/mdstat */
3841 static void status_unused(struct seq_file
*seq
)
3845 struct list_head
*tmp
;
3847 seq_printf(seq
, "unused devices: ");
3849 ITERATE_RDEV_PENDING(rdev
,tmp
) {
3850 char b
[BDEVNAME_SIZE
];
3852 seq_printf(seq
, "%s ",
3853 bdevname(rdev
->bdev
,b
));
3856 seq_printf(seq
, "<none>");
3858 seq_printf(seq
, "\n");
3862 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
3864 unsigned long max_blocks
, resync
, res
, dt
, db
, rt
;
3866 resync
= (mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
))/2;
3868 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3869 max_blocks
= mddev
->resync_max_sectors
>> 1;
3871 max_blocks
= mddev
->size
;
3874 * Should not happen.
3880 res
= (resync
/1024)*1000/(max_blocks
/1024 + 1);
3882 int i
, x
= res
/50, y
= 20-x
;
3883 seq_printf(seq
, "[");
3884 for (i
= 0; i
< x
; i
++)
3885 seq_printf(seq
, "=");
3886 seq_printf(seq
, ">");
3887 for (i
= 0; i
< y
; i
++)
3888 seq_printf(seq
, ".");
3889 seq_printf(seq
, "] ");
3891 seq_printf(seq
, " %s =%3lu.%lu%% (%lu/%lu)",
3892 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
3893 "resync" : "recovery"),
3894 res
/10, res
% 10, resync
, max_blocks
);
3897 * We do not want to overflow, so the order of operands and
3898 * the * 100 / 100 trick are important. We do a +1 to be
3899 * safe against division by zero. We only estimate anyway.
3901 * dt: time from mark until now
3902 * db: blocks written from mark until now
3903 * rt: remaining time
3905 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
3907 db
= resync
- (mddev
->resync_mark_cnt
/2);
3908 rt
= (dt
* ((max_blocks
-resync
) / (db
/100+1)))/100;
3910 seq_printf(seq
, " finish=%lu.%lumin", rt
/ 60, (rt
% 60)/6);
3912 seq_printf(seq
, " speed=%ldK/sec", db
/dt
);
3915 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
3917 struct list_head
*tmp
;
3927 spin_lock(&all_mddevs_lock
);
3928 list_for_each(tmp
,&all_mddevs
)
3930 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
3932 spin_unlock(&all_mddevs_lock
);
3935 spin_unlock(&all_mddevs_lock
);
3937 return (void*)2;/* tail */
3941 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
3943 struct list_head
*tmp
;
3944 mddev_t
*next_mddev
, *mddev
= v
;
3950 spin_lock(&all_mddevs_lock
);
3952 tmp
= all_mddevs
.next
;
3954 tmp
= mddev
->all_mddevs
.next
;
3955 if (tmp
!= &all_mddevs
)
3956 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
3958 next_mddev
= (void*)2;
3961 spin_unlock(&all_mddevs_lock
);
3969 static void md_seq_stop(struct seq_file
*seq
, void *v
)
3973 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
3977 struct mdstat_info
{
3981 static int md_seq_show(struct seq_file
*seq
, void *v
)
3985 struct list_head
*tmp2
;
3987 struct mdstat_info
*mi
= seq
->private;
3988 struct bitmap
*bitmap
;
3990 if (v
== (void*)1) {
3991 struct mdk_personality
*pers
;
3992 seq_printf(seq
, "Personalities : ");
3993 spin_lock(&pers_lock
);
3994 list_for_each_entry(pers
, &pers_list
, list
)
3995 seq_printf(seq
, "[%s] ", pers
->name
);
3997 spin_unlock(&pers_lock
);
3998 seq_printf(seq
, "\n");
3999 mi
->event
= atomic_read(&md_event_count
);
4002 if (v
== (void*)2) {
4007 if (mddev_lock(mddev
)!=0)
4009 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
4010 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
4011 mddev
->pers
? "" : "in");
4014 seq_printf(seq
, " (read-only)");
4016 seq_printf(seq
, "(auto-read-only)");
4017 seq_printf(seq
, " %s", mddev
->pers
->name
);
4021 ITERATE_RDEV(mddev
,rdev
,tmp2
) {
4022 char b
[BDEVNAME_SIZE
];
4023 seq_printf(seq
, " %s[%d]",
4024 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
4025 if (test_bit(WriteMostly
, &rdev
->flags
))
4026 seq_printf(seq
, "(W)");
4027 if (test_bit(Faulty
, &rdev
->flags
)) {
4028 seq_printf(seq
, "(F)");
4030 } else if (rdev
->raid_disk
< 0)
4031 seq_printf(seq
, "(S)"); /* spare */
4035 if (!list_empty(&mddev
->disks
)) {
4037 seq_printf(seq
, "\n %llu blocks",
4038 (unsigned long long)mddev
->array_size
);
4040 seq_printf(seq
, "\n %llu blocks",
4041 (unsigned long long)size
);
4043 if (mddev
->persistent
) {
4044 if (mddev
->major_version
!= 0 ||
4045 mddev
->minor_version
!= 90) {
4046 seq_printf(seq
," super %d.%d",
4047 mddev
->major_version
,
4048 mddev
->minor_version
);
4051 seq_printf(seq
, " super non-persistent");
4054 mddev
->pers
->status (seq
, mddev
);
4055 seq_printf(seq
, "\n ");
4056 if (mddev
->pers
->sync_request
) {
4057 if (mddev
->curr_resync
> 2) {
4058 status_resync (seq
, mddev
);
4059 seq_printf(seq
, "\n ");
4060 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
4061 seq_printf(seq
, "\tresync=DELAYED\n ");
4062 else if (mddev
->recovery_cp
< MaxSector
)
4063 seq_printf(seq
, "\tresync=PENDING\n ");
4066 seq_printf(seq
, "\n ");
4068 if ((bitmap
= mddev
->bitmap
)) {
4069 unsigned long chunk_kb
;
4070 unsigned long flags
;
4071 spin_lock_irqsave(&bitmap
->lock
, flags
);
4072 chunk_kb
= bitmap
->chunksize
>> 10;
4073 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
4075 bitmap
->pages
- bitmap
->missing_pages
,
4077 (bitmap
->pages
- bitmap
->missing_pages
)
4078 << (PAGE_SHIFT
- 10),
4079 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
4080 chunk_kb
? "KB" : "B");
4082 seq_printf(seq
, ", file: ");
4083 seq_path(seq
, bitmap
->file
->f_vfsmnt
,
4084 bitmap
->file
->f_dentry
," \t\n");
4087 seq_printf(seq
, "\n");
4088 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
4091 seq_printf(seq
, "\n");
4093 mddev_unlock(mddev
);
4098 static struct seq_operations md_seq_ops
= {
4099 .start
= md_seq_start
,
4100 .next
= md_seq_next
,
4101 .stop
= md_seq_stop
,
4102 .show
= md_seq_show
,
4105 static int md_seq_open(struct inode
*inode
, struct file
*file
)
4108 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
4112 error
= seq_open(file
, &md_seq_ops
);
4116 struct seq_file
*p
= file
->private_data
;
4118 mi
->event
= atomic_read(&md_event_count
);
4123 static int md_seq_release(struct inode
*inode
, struct file
*file
)
4125 struct seq_file
*m
= file
->private_data
;
4126 struct mdstat_info
*mi
= m
->private;
4129 return seq_release(inode
, file
);
4132 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
4134 struct seq_file
*m
= filp
->private_data
;
4135 struct mdstat_info
*mi
= m
->private;
4138 poll_wait(filp
, &md_event_waiters
, wait
);
4140 /* always allow read */
4141 mask
= POLLIN
| POLLRDNORM
;
4143 if (mi
->event
!= atomic_read(&md_event_count
))
4144 mask
|= POLLERR
| POLLPRI
;
4148 static struct file_operations md_seq_fops
= {
4149 .open
= md_seq_open
,
4151 .llseek
= seq_lseek
,
4152 .release
= md_seq_release
,
4153 .poll
= mdstat_poll
,
4156 int register_md_personality(struct mdk_personality
*p
)
4158 spin_lock(&pers_lock
);
4159 list_add_tail(&p
->list
, &pers_list
);
4160 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
4161 spin_unlock(&pers_lock
);
4165 int unregister_md_personality(struct mdk_personality
*p
)
4167 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
4168 spin_lock(&pers_lock
);
4169 list_del_init(&p
->list
);
4170 spin_unlock(&pers_lock
);
4174 static int is_mddev_idle(mddev_t
*mddev
)
4177 struct list_head
*tmp
;
4179 unsigned long curr_events
;
4182 ITERATE_RDEV(mddev
,rdev
,tmp
) {
4183 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
4184 curr_events
= disk_stat_read(disk
, sectors
[0]) +
4185 disk_stat_read(disk
, sectors
[1]) -
4186 atomic_read(&disk
->sync_io
);
4187 /* The difference between curr_events and last_events
4188 * will be affected by any new non-sync IO (making
4189 * curr_events bigger) and any difference in the amount of
4190 * in-flight syncio (making current_events bigger or smaller)
4191 * The amount in-flight is currently limited to
4192 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4193 * which is at most 4096 sectors.
4194 * These numbers are fairly fragile and should be made
4195 * more robust, probably by enforcing the
4196 * 'window size' that md_do_sync sort-of uses.
4198 * Note: the following is an unsigned comparison.
4200 if ((curr_events
- rdev
->last_events
+ 4096) > 8192) {
4201 rdev
->last_events
= curr_events
;
4208 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
4210 /* another "blocks" (512byte) blocks have been synced */
4211 atomic_sub(blocks
, &mddev
->recovery_active
);
4212 wake_up(&mddev
->recovery_wait
);
4214 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4215 md_wakeup_thread(mddev
->thread
);
4216 // stop recovery, signal do_sync ....
4221 /* md_write_start(mddev, bi)
4222 * If we need to update some array metadata (e.g. 'active' flag
4223 * in superblock) before writing, schedule a superblock update
4224 * and wait for it to complete.
4226 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
4228 if (bio_data_dir(bi
) != WRITE
)
4231 BUG_ON(mddev
->ro
== 1);
4232 if (mddev
->ro
== 2) {
4233 /* need to switch to read/write */
4235 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4236 md_wakeup_thread(mddev
->thread
);
4238 atomic_inc(&mddev
->writes_pending
);
4239 if (mddev
->in_sync
) {
4240 spin_lock_irq(&mddev
->write_lock
);
4241 if (mddev
->in_sync
) {
4243 mddev
->sb_dirty
= 1;
4244 md_wakeup_thread(mddev
->thread
);
4246 spin_unlock_irq(&mddev
->write_lock
);
4248 wait_event(mddev
->sb_wait
, mddev
->sb_dirty
==0);
4251 void md_write_end(mddev_t
*mddev
)
4253 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
4254 if (mddev
->safemode
== 2)
4255 md_wakeup_thread(mddev
->thread
);
4257 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
4261 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
4263 #define SYNC_MARKS 10
4264 #define SYNC_MARK_STEP (3*HZ)
4265 static void md_do_sync(mddev_t
*mddev
)
4268 unsigned int currspeed
= 0,
4270 sector_t max_sectors
,j
, io_sectors
;
4271 unsigned long mark
[SYNC_MARKS
];
4272 sector_t mark_cnt
[SYNC_MARKS
];
4274 struct list_head
*tmp
;
4275 sector_t last_check
;
4278 /* just incase thread restarts... */
4279 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
4282 /* we overload curr_resync somewhat here.
4283 * 0 == not engaged in resync at all
4284 * 2 == checking that there is no conflict with another sync
4285 * 1 == like 2, but have yielded to allow conflicting resync to
4287 * other == active in resync - this many blocks
4289 * Before starting a resync we must have set curr_resync to
4290 * 2, and then checked that every "conflicting" array has curr_resync
4291 * less than ours. When we find one that is the same or higher
4292 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4293 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4294 * This will mean we have to start checking from the beginning again.
4299 mddev
->curr_resync
= 2;
4302 if (kthread_should_stop()) {
4303 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4306 ITERATE_MDDEV(mddev2
,tmp
) {
4307 if (mddev2
== mddev
)
4309 if (mddev2
->curr_resync
&&
4310 match_mddev_units(mddev
,mddev2
)) {
4312 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
4313 /* arbitrarily yield */
4314 mddev
->curr_resync
= 1;
4315 wake_up(&resync_wait
);
4317 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
4318 /* no need to wait here, we can wait the next
4319 * time 'round when curr_resync == 2
4322 prepare_to_wait(&resync_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
4323 if (!kthread_should_stop() &&
4324 mddev2
->curr_resync
>= mddev
->curr_resync
) {
4325 printk(KERN_INFO
"md: delaying resync of %s"
4326 " until %s has finished resync (they"
4327 " share one or more physical units)\n",
4328 mdname(mddev
), mdname(mddev2
));
4331 finish_wait(&resync_wait
, &wq
);
4334 finish_wait(&resync_wait
, &wq
);
4337 } while (mddev
->curr_resync
< 2);
4339 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4340 /* resync follows the size requested by the personality,
4341 * which defaults to physical size, but can be virtual size
4343 max_sectors
= mddev
->resync_max_sectors
;
4344 mddev
->resync_mismatches
= 0;
4346 /* recovery follows the physical size of devices */
4347 max_sectors
= mddev
->size
<< 1;
4349 printk(KERN_INFO
"md: syncing RAID array %s\n", mdname(mddev
));
4350 printk(KERN_INFO
"md: minimum _guaranteed_ reconstruction speed:"
4351 " %d KB/sec/disc.\n", sysctl_speed_limit_min
);
4352 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
4353 "(but not more than %d KB/sec) for reconstruction.\n",
4354 sysctl_speed_limit_max
);
4356 is_mddev_idle(mddev
); /* this also initializes IO event counters */
4357 /* we don't use the checkpoint if there's a bitmap */
4358 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && !mddev
->bitmap
4359 && ! test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4360 j
= mddev
->recovery_cp
;
4364 for (m
= 0; m
< SYNC_MARKS
; m
++) {
4366 mark_cnt
[m
] = io_sectors
;
4369 mddev
->resync_mark
= mark
[last_mark
];
4370 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
4373 * Tune reconstruction:
4375 window
= 32*(PAGE_SIZE
/512);
4376 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
4377 window
/2,(unsigned long long) max_sectors
/2);
4379 atomic_set(&mddev
->recovery_active
, 0);
4380 init_waitqueue_head(&mddev
->recovery_wait
);
4385 "md: resuming recovery of %s from checkpoint.\n",
4387 mddev
->curr_resync
= j
;
4390 while (j
< max_sectors
) {
4394 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
4395 currspeed
< sysctl_speed_limit_min
);
4397 set_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4401 if (!skipped
) { /* actual IO requested */
4402 io_sectors
+= sectors
;
4403 atomic_add(sectors
, &mddev
->recovery_active
);
4407 if (j
>1) mddev
->curr_resync
= j
;
4408 if (last_check
== 0)
4409 /* this is the earliers that rebuilt will be
4410 * visible in /proc/mdstat
4412 md_new_event(mddev
);
4414 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
4417 last_check
= io_sectors
;
4419 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) ||
4420 test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
))
4424 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
4426 int next
= (last_mark
+1) % SYNC_MARKS
;
4428 mddev
->resync_mark
= mark
[next
];
4429 mddev
->resync_mark_cnt
= mark_cnt
[next
];
4430 mark
[next
] = jiffies
;
4431 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
4436 if (kthread_should_stop()) {
4438 * got a signal, exit.
4441 "md: md_do_sync() got signal ... exiting\n");
4442 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4447 * this loop exits only if either when we are slower than
4448 * the 'hard' speed limit, or the system was IO-idle for
4450 * the system might be non-idle CPU-wise, but we only care
4451 * about not overloading the IO subsystem. (things like an
4452 * e2fsck being done on the RAID array should execute fast)
4454 mddev
->queue
->unplug_fn(mddev
->queue
);
4457 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
4458 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
4460 if (currspeed
> sysctl_speed_limit_min
) {
4461 if ((currspeed
> sysctl_speed_limit_max
) ||
4462 !is_mddev_idle(mddev
)) {
4468 printk(KERN_INFO
"md: %s: sync done.\n",mdname(mddev
));
4470 * this also signals 'finished resyncing' to md_stop
4473 mddev
->queue
->unplug_fn(mddev
->queue
);
4475 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
4477 /* tell personality that we are finished */
4478 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
4480 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4481 mddev
->curr_resync
> 2 &&
4482 mddev
->curr_resync
>= mddev
->recovery_cp
) {
4483 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4485 "md: checkpointing recovery of %s.\n",
4487 mddev
->recovery_cp
= mddev
->curr_resync
;
4489 mddev
->recovery_cp
= MaxSector
;
4493 mddev
->curr_resync
= 0;
4494 wake_up(&resync_wait
);
4495 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4496 md_wakeup_thread(mddev
->thread
);
4501 * This routine is regularly called by all per-raid-array threads to
4502 * deal with generic issues like resync and super-block update.
4503 * Raid personalities that don't have a thread (linear/raid0) do not
4504 * need this as they never do any recovery or update the superblock.
4506 * It does not do any resync itself, but rather "forks" off other threads
4507 * to do that as needed.
4508 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4509 * "->recovery" and create a thread at ->sync_thread.
4510 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4511 * and wakeups up this thread which will reap the thread and finish up.
4512 * This thread also removes any faulty devices (with nr_pending == 0).
4514 * The overall approach is:
4515 * 1/ if the superblock needs updating, update it.
4516 * 2/ If a recovery thread is running, don't do anything else.
4517 * 3/ If recovery has finished, clean up, possibly marking spares active.
4518 * 4/ If there are any faulty devices, remove them.
4519 * 5/ If array is degraded, try to add spares devices
4520 * 6/ If array has spares or is not in-sync, start a resync thread.
4522 void md_check_recovery(mddev_t
*mddev
)
4525 struct list_head
*rtmp
;
4529 bitmap_daemon_work(mddev
->bitmap
);
4534 if (signal_pending(current
)) {
4535 if (mddev
->pers
->sync_request
) {
4536 printk(KERN_INFO
"md: %s in immediate safe mode\n",
4538 mddev
->safemode
= 2;
4540 flush_signals(current
);
4545 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
4546 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
4547 (mddev
->safemode
== 1) ||
4548 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
4549 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
4553 if (mddev_trylock(mddev
)==0) {
4556 spin_lock_irq(&mddev
->write_lock
);
4557 if (mddev
->safemode
&& !atomic_read(&mddev
->writes_pending
) &&
4558 !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
) {
4560 mddev
->sb_dirty
= 1;
4562 if (mddev
->safemode
== 1)
4563 mddev
->safemode
= 0;
4564 spin_unlock_irq(&mddev
->write_lock
);
4566 if (mddev
->sb_dirty
)
4567 md_update_sb(mddev
);
4570 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4571 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
4572 /* resync/recovery still happening */
4573 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4576 if (mddev
->sync_thread
) {
4577 /* resync has finished, collect result */
4578 md_unregister_thread(mddev
->sync_thread
);
4579 mddev
->sync_thread
= NULL
;
4580 if (!test_bit(MD_RECOVERY_ERR
, &mddev
->recovery
) &&
4581 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
4583 /* activate any spares */
4584 mddev
->pers
->spare_active(mddev
);
4586 md_update_sb(mddev
);
4588 /* if array is no-longer degraded, then any saved_raid_disk
4589 * information must be scrapped
4591 if (!mddev
->degraded
)
4592 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4593 rdev
->saved_raid_disk
= -1;
4595 mddev
->recovery
= 0;
4596 /* flag recovery needed just to double check */
4597 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4598 md_new_event(mddev
);
4601 /* Clear some bits that don't mean anything, but
4604 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4605 clear_bit(MD_RECOVERY_ERR
, &mddev
->recovery
);
4606 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4607 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
4609 /* no recovery is running.
4610 * remove any failed drives, then
4611 * add spares if possible.
4612 * Spare are also removed and re-added, to allow
4613 * the personality to fail the re-add.
4615 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4616 if (rdev
->raid_disk
>= 0 &&
4617 (test_bit(Faulty
, &rdev
->flags
) || ! test_bit(In_sync
, &rdev
->flags
)) &&
4618 atomic_read(&rdev
->nr_pending
)==0) {
4619 if (mddev
->pers
->hot_remove_disk(mddev
, rdev
->raid_disk
)==0) {
4621 sprintf(nm
,"rd%d", rdev
->raid_disk
);
4622 sysfs_remove_link(&mddev
->kobj
, nm
);
4623 rdev
->raid_disk
= -1;
4627 if (mddev
->degraded
) {
4628 ITERATE_RDEV(mddev
,rdev
,rtmp
)
4629 if (rdev
->raid_disk
< 0
4630 && !test_bit(Faulty
, &rdev
->flags
)) {
4631 if (mddev
->pers
->hot_add_disk(mddev
,rdev
)) {
4633 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4634 sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
);
4636 md_new_event(mddev
);
4643 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4644 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4645 } else if (mddev
->recovery_cp
< MaxSector
) {
4646 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4647 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4648 /* nothing to be done ... */
4651 if (mddev
->pers
->sync_request
) {
4652 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4653 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
4654 /* We are adding a device or devices to an array
4655 * which has the bitmap stored on all devices.
4656 * So make sure all bitmap pages get written
4658 bitmap_write_all(mddev
->bitmap
);
4660 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4663 if (!mddev
->sync_thread
) {
4664 printk(KERN_ERR
"%s: could not start resync"
4667 /* leave the spares where they are, it shouldn't hurt */
4668 mddev
->recovery
= 0;
4670 md_wakeup_thread(mddev
->sync_thread
);
4671 md_new_event(mddev
);
4674 mddev_unlock(mddev
);
4678 static int md_notify_reboot(struct notifier_block
*this,
4679 unsigned long code
, void *x
)
4681 struct list_head
*tmp
;
4684 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
4686 printk(KERN_INFO
"md: stopping all md devices.\n");
4688 ITERATE_MDDEV(mddev
,tmp
)
4689 if (mddev_trylock(mddev
)==0)
4690 do_md_stop (mddev
, 1);
4692 * certain more exotic SCSI devices are known to be
4693 * volatile wrt too early system reboots. While the
4694 * right place to handle this issue is the given
4695 * driver, we do want to have a safe RAID driver ...
4702 static struct notifier_block md_notifier
= {
4703 .notifier_call
= md_notify_reboot
,
4705 .priority
= INT_MAX
, /* before any real devices */
4708 static void md_geninit(void)
4710 struct proc_dir_entry
*p
;
4712 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
4714 p
= create_proc_entry("mdstat", S_IRUGO
, NULL
);
4716 p
->proc_fops
= &md_seq_fops
;
4719 static int __init
md_init(void)
4723 printk(KERN_INFO
"md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4724 " MD_SB_DISKS=%d\n",
4725 MD_MAJOR_VERSION
, MD_MINOR_VERSION
,
4726 MD_PATCHLEVEL_VERSION
, MAX_MD_DEVS
, MD_SB_DISKS
);
4727 printk(KERN_INFO
"md: bitmap version %d.%d\n", BITMAP_MAJOR_HI
,
4730 if (register_blkdev(MAJOR_NR
, "md"))
4732 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
4733 unregister_blkdev(MAJOR_NR
, "md");
4737 blk_register_region(MKDEV(MAJOR_NR
, 0), MAX_MD_DEVS
, THIS_MODULE
,
4738 md_probe
, NULL
, NULL
);
4739 blk_register_region(MKDEV(mdp_major
, 0), MAX_MD_DEVS
<<MdpMinorShift
, THIS_MODULE
,
4740 md_probe
, NULL
, NULL
);
4742 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4743 devfs_mk_bdev(MKDEV(MAJOR_NR
, minor
),
4744 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4747 for (minor
=0; minor
< MAX_MD_DEVS
; ++minor
)
4748 devfs_mk_bdev(MKDEV(mdp_major
, minor
<<MdpMinorShift
),
4749 S_IFBLK
|S_IRUSR
|S_IWUSR
,
4753 register_reboot_notifier(&md_notifier
);
4754 raid_table_header
= register_sysctl_table(raid_root_table
, 1);
4764 * Searches all registered partitions for autorun RAID arrays
4767 static dev_t detected_devices
[128];
4770 void md_autodetect_dev(dev_t dev
)
4772 if (dev_cnt
>= 0 && dev_cnt
< 127)
4773 detected_devices
[dev_cnt
++] = dev
;
4777 static void autostart_arrays(int part
)
4782 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
4784 for (i
= 0; i
< dev_cnt
; i
++) {
4785 dev_t dev
= detected_devices
[i
];
4787 rdev
= md_import_device(dev
,0, 0);
4791 if (test_bit(Faulty
, &rdev
->flags
)) {
4795 list_add(&rdev
->same_set
, &pending_raid_disks
);
4799 autorun_devices(part
);
4804 static __exit
void md_exit(void)
4807 struct list_head
*tmp
;
4809 blk_unregister_region(MKDEV(MAJOR_NR
,0), MAX_MD_DEVS
);
4810 blk_unregister_region(MKDEV(mdp_major
,0), MAX_MD_DEVS
<< MdpMinorShift
);
4811 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4812 devfs_remove("md/%d", i
);
4813 for (i
=0; i
< MAX_MD_DEVS
; i
++)
4814 devfs_remove("md/d%d", i
);
4818 unregister_blkdev(MAJOR_NR
,"md");
4819 unregister_blkdev(mdp_major
, "mdp");
4820 unregister_reboot_notifier(&md_notifier
);
4821 unregister_sysctl_table(raid_table_header
);
4822 remove_proc_entry("mdstat", NULL
);
4823 ITERATE_MDDEV(mddev
,tmp
) {
4824 struct gendisk
*disk
= mddev
->gendisk
;
4827 export_array(mddev
);
4830 mddev
->gendisk
= NULL
;
4835 module_init(md_init
)
4836 module_exit(md_exit
)
4838 static int get_ro(char *buffer
, struct kernel_param
*kp
)
4840 return sprintf(buffer
, "%d", start_readonly
);
4842 static int set_ro(const char *val
, struct kernel_param
*kp
)
4845 int num
= simple_strtoul(val
, &e
, 10);
4846 if (*val
&& (*e
== '\0' || *e
== '\n')) {
4847 start_readonly
= num
;
4853 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, 0600);
4854 module_param(start_dirty_degraded
, int, 0644);
4857 EXPORT_SYMBOL(register_md_personality
);
4858 EXPORT_SYMBOL(unregister_md_personality
);
4859 EXPORT_SYMBOL(md_error
);
4860 EXPORT_SYMBOL(md_done_sync
);
4861 EXPORT_SYMBOL(md_write_start
);
4862 EXPORT_SYMBOL(md_write_end
);
4863 EXPORT_SYMBOL(md_register_thread
);
4864 EXPORT_SYMBOL(md_unregister_thread
);
4865 EXPORT_SYMBOL(md_wakeup_thread
);
4866 EXPORT_SYMBOL(md_print_devices
);
4867 EXPORT_SYMBOL(md_check_recovery
);
4868 MODULE_LICENSE("GPL");
4870 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);