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/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58 static void autostart_arrays(int part
);
61 static LIST_HEAD(pers_list
);
62 static DEFINE_SPINLOCK(pers_lock
);
64 static void md_print_devices(void);
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
83 static int sysctl_speed_limit_min
= 1000;
84 static int sysctl_speed_limit_max
= 200000;
85 static inline int speed_min(mddev_t
*mddev
)
87 return mddev
->sync_speed_min
?
88 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
91 static inline int speed_max(mddev_t
*mddev
)
93 return mddev
->sync_speed_max
?
94 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
97 static struct ctl_table_header
*raid_table_header
;
99 static ctl_table raid_table
[] = {
101 .ctl_name
= DEV_RAID_SPEED_LIMIT_MIN
,
102 .procname
= "speed_limit_min",
103 .data
= &sysctl_speed_limit_min
,
104 .maxlen
= sizeof(int),
105 .mode
= S_IRUGO
|S_IWUSR
,
106 .proc_handler
= &proc_dointvec
,
109 .ctl_name
= DEV_RAID_SPEED_LIMIT_MAX
,
110 .procname
= "speed_limit_max",
111 .data
= &sysctl_speed_limit_max
,
112 .maxlen
= sizeof(int),
113 .mode
= S_IRUGO
|S_IWUSR
,
114 .proc_handler
= &proc_dointvec
,
119 static ctl_table raid_dir_table
[] = {
121 .ctl_name
= DEV_RAID
,
124 .mode
= S_IRUGO
|S_IXUGO
,
130 static ctl_table raid_root_table
[] = {
136 .child
= raid_dir_table
,
141 static const struct block_device_operations md_fops
;
143 static int start_readonly
;
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
156 static atomic_t md_event_count
;
157 void md_new_event(mddev_t
*mddev
)
159 atomic_inc(&md_event_count
);
160 wake_up(&md_event_waiters
);
162 EXPORT_SYMBOL_GPL(md_new_event
);
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
167 static void md_new_event_inintr(mddev_t
*mddev
)
169 atomic_inc(&md_event_count
);
170 wake_up(&md_event_waiters
);
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
177 static LIST_HEAD(all_mddevs
);
178 static DEFINE_SPINLOCK(all_mddevs_lock
);
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
188 #define for_each_mddev(mddev,tmp) \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
211 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
213 mddev_t
*mddev
= q
->queuedata
;
215 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
220 if (mddev
->suspended
) {
223 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
224 TASK_UNINTERRUPTIBLE
);
225 if (!mddev
->suspended
)
231 finish_wait(&mddev
->sb_wait
, &__wait
);
233 atomic_inc(&mddev
->active_io
);
235 rv
= mddev
->pers
->make_request(q
, bio
);
236 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
237 wake_up(&mddev
->sb_wait
);
242 static void mddev_suspend(mddev_t
*mddev
)
244 BUG_ON(mddev
->suspended
);
245 mddev
->suspended
= 1;
247 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
248 mddev
->pers
->quiesce(mddev
, 1);
249 md_unregister_thread(mddev
->thread
);
250 mddev
->thread
= NULL
;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
258 static void mddev_resume(mddev_t
*mddev
)
260 mddev
->suspended
= 0;
261 wake_up(&mddev
->sb_wait
);
262 mddev
->pers
->quiesce(mddev
, 0);
265 int mddev_congested(mddev_t
*mddev
, int bits
)
267 return mddev
->suspended
;
269 EXPORT_SYMBOL(mddev_congested
);
272 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
274 atomic_inc(&mddev
->active
);
278 static void mddev_delayed_delete(struct work_struct
*ws
);
280 static void mddev_put(mddev_t
*mddev
)
282 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
284 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
285 !mddev
->hold_active
) {
286 list_del(&mddev
->all_mddevs
);
287 if (mddev
->gendisk
) {
288 /* we did a probe so need to clean up.
289 * Call schedule_work inside the spinlock
290 * so that flush_scheduled_work() after
291 * mddev_find will succeed in waiting for the
294 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
295 schedule_work(&mddev
->del_work
);
299 spin_unlock(&all_mddevs_lock
);
302 static mddev_t
* mddev_find(dev_t unit
)
304 mddev_t
*mddev
, *new = NULL
;
307 spin_lock(&all_mddevs_lock
);
310 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
311 if (mddev
->unit
== unit
) {
313 spin_unlock(&all_mddevs_lock
);
319 list_add(&new->all_mddevs
, &all_mddevs
);
320 spin_unlock(&all_mddevs_lock
);
321 new->hold_active
= UNTIL_IOCTL
;
325 /* find an unused unit number */
326 static int next_minor
= 512;
327 int start
= next_minor
;
331 dev
= MKDEV(MD_MAJOR
, next_minor
);
333 if (next_minor
> MINORMASK
)
335 if (next_minor
== start
) {
336 /* Oh dear, all in use. */
337 spin_unlock(&all_mddevs_lock
);
343 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
344 if (mddev
->unit
== dev
) {
350 new->md_minor
= MINOR(dev
);
351 new->hold_active
= UNTIL_STOP
;
352 list_add(&new->all_mddevs
, &all_mddevs
);
353 spin_unlock(&all_mddevs_lock
);
356 spin_unlock(&all_mddevs_lock
);
358 new = kzalloc(sizeof(*new), GFP_KERNEL
);
363 if (MAJOR(unit
) == MD_MAJOR
)
364 new->md_minor
= MINOR(unit
);
366 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
368 mutex_init(&new->open_mutex
);
369 mutex_init(&new->reconfig_mutex
);
370 INIT_LIST_HEAD(&new->disks
);
371 INIT_LIST_HEAD(&new->all_mddevs
);
372 init_timer(&new->safemode_timer
);
373 atomic_set(&new->active
, 1);
374 atomic_set(&new->openers
, 0);
375 atomic_set(&new->active_io
, 0);
376 spin_lock_init(&new->write_lock
);
377 init_waitqueue_head(&new->sb_wait
);
378 init_waitqueue_head(&new->recovery_wait
);
379 new->reshape_position
= MaxSector
;
381 new->resync_max
= MaxSector
;
382 new->level
= LEVEL_NONE
;
387 static inline int mddev_lock(mddev_t
* mddev
)
389 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
392 static inline int mddev_is_locked(mddev_t
*mddev
)
394 return mutex_is_locked(&mddev
->reconfig_mutex
);
397 static inline int mddev_trylock(mddev_t
* mddev
)
399 return mutex_trylock(&mddev
->reconfig_mutex
);
402 static inline void mddev_unlock(mddev_t
* mddev
)
404 mutex_unlock(&mddev
->reconfig_mutex
);
406 md_wakeup_thread(mddev
->thread
);
409 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
413 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
414 if (rdev
->desc_nr
== nr
)
420 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
424 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
425 if (rdev
->bdev
->bd_dev
== dev
)
431 static struct mdk_personality
*find_pers(int level
, char *clevel
)
433 struct mdk_personality
*pers
;
434 list_for_each_entry(pers
, &pers_list
, list
) {
435 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
437 if (strcmp(pers
->name
, clevel
)==0)
443 /* return the offset of the super block in 512byte sectors */
444 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
446 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
447 return MD_NEW_SIZE_SECTORS(num_sectors
);
450 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
455 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
456 if (!rdev
->sb_page
) {
457 printk(KERN_ALERT
"md: out of memory.\n");
464 static void free_disk_sb(mdk_rdev_t
* rdev
)
467 put_page(rdev
->sb_page
);
469 rdev
->sb_page
= NULL
;
476 static void super_written(struct bio
*bio
, int error
)
478 mdk_rdev_t
*rdev
= bio
->bi_private
;
479 mddev_t
*mddev
= rdev
->mddev
;
481 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
482 printk("md: super_written gets error=%d, uptodate=%d\n",
483 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
484 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
485 md_error(mddev
, rdev
);
488 if (atomic_dec_and_test(&mddev
->pending_writes
))
489 wake_up(&mddev
->sb_wait
);
493 static void super_written_barrier(struct bio
*bio
, int error
)
495 struct bio
*bio2
= bio
->bi_private
;
496 mdk_rdev_t
*rdev
= bio2
->bi_private
;
497 mddev_t
*mddev
= rdev
->mddev
;
499 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
500 error
== -EOPNOTSUPP
) {
502 /* barriers don't appear to be supported :-( */
503 set_bit(BarriersNotsupp
, &rdev
->flags
);
504 mddev
->barriers_work
= 0;
505 spin_lock_irqsave(&mddev
->write_lock
, flags
);
506 bio2
->bi_next
= mddev
->biolist
;
507 mddev
->biolist
= bio2
;
508 spin_unlock_irqrestore(&mddev
->write_lock
, flags
);
509 wake_up(&mddev
->sb_wait
);
513 bio
->bi_private
= rdev
;
514 super_written(bio
, error
);
518 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
519 sector_t sector
, int size
, struct page
*page
)
521 /* write first size bytes of page to sector of rdev
522 * Increment mddev->pending_writes before returning
523 * and decrement it on completion, waking up sb_wait
524 * if zero is reached.
525 * If an error occurred, call md_error
527 * As we might need to resubmit the request if BIO_RW_BARRIER
528 * causes ENOTSUPP, we allocate a spare bio...
530 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
531 int rw
= (1<<BIO_RW
) | (1<<BIO_RW_SYNCIO
) | (1<<BIO_RW_UNPLUG
);
533 bio
->bi_bdev
= rdev
->bdev
;
534 bio
->bi_sector
= sector
;
535 bio_add_page(bio
, page
, size
, 0);
536 bio
->bi_private
= rdev
;
537 bio
->bi_end_io
= super_written
;
540 atomic_inc(&mddev
->pending_writes
);
541 if (!test_bit(BarriersNotsupp
, &rdev
->flags
)) {
543 rw
|= (1<<BIO_RW_BARRIER
);
544 rbio
= bio_clone(bio
, GFP_NOIO
);
545 rbio
->bi_private
= bio
;
546 rbio
->bi_end_io
= super_written_barrier
;
547 submit_bio(rw
, rbio
);
552 void md_super_wait(mddev_t
*mddev
)
554 /* wait for all superblock writes that were scheduled to complete.
555 * if any had to be retried (due to BARRIER problems), retry them
559 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
560 if (atomic_read(&mddev
->pending_writes
)==0)
562 while (mddev
->biolist
) {
564 spin_lock_irq(&mddev
->write_lock
);
565 bio
= mddev
->biolist
;
566 mddev
->biolist
= bio
->bi_next
;
568 spin_unlock_irq(&mddev
->write_lock
);
569 submit_bio(bio
->bi_rw
, bio
);
573 finish_wait(&mddev
->sb_wait
, &wq
);
576 static void bi_complete(struct bio
*bio
, int error
)
578 complete((struct completion
*)bio
->bi_private
);
581 int sync_page_io(struct block_device
*bdev
, sector_t sector
, int size
,
582 struct page
*page
, int rw
)
584 struct bio
*bio
= bio_alloc(GFP_NOIO
, 1);
585 struct completion event
;
588 rw
|= (1 << BIO_RW_SYNCIO
) | (1 << BIO_RW_UNPLUG
);
591 bio
->bi_sector
= sector
;
592 bio_add_page(bio
, page
, size
, 0);
593 init_completion(&event
);
594 bio
->bi_private
= &event
;
595 bio
->bi_end_io
= bi_complete
;
597 wait_for_completion(&event
);
599 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
603 EXPORT_SYMBOL_GPL(sync_page_io
);
605 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
607 char b
[BDEVNAME_SIZE
];
608 if (!rdev
->sb_page
) {
616 if (!sync_page_io(rdev
->bdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
622 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
623 bdevname(rdev
->bdev
,b
));
627 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
629 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
630 sb1
->set_uuid1
== sb2
->set_uuid1
&&
631 sb1
->set_uuid2
== sb2
->set_uuid2
&&
632 sb1
->set_uuid3
== sb2
->set_uuid3
;
635 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
638 mdp_super_t
*tmp1
, *tmp2
;
640 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
641 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
643 if (!tmp1
|| !tmp2
) {
645 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
653 * nr_disks is not constant
658 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
666 static u32
md_csum_fold(u32 csum
)
668 csum
= (csum
& 0xffff) + (csum
>> 16);
669 return (csum
& 0xffff) + (csum
>> 16);
672 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
675 u32
*sb32
= (u32
*)sb
;
677 unsigned int disk_csum
, csum
;
679 disk_csum
= sb
->sb_csum
;
682 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
684 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
688 /* This used to use csum_partial, which was wrong for several
689 * reasons including that different results are returned on
690 * different architectures. It isn't critical that we get exactly
691 * the same return value as before (we always csum_fold before
692 * testing, and that removes any differences). However as we
693 * know that csum_partial always returned a 16bit value on
694 * alphas, do a fold to maximise conformity to previous behaviour.
696 sb
->sb_csum
= md_csum_fold(disk_csum
);
698 sb
->sb_csum
= disk_csum
;
705 * Handle superblock details.
706 * We want to be able to handle multiple superblock formats
707 * so we have a common interface to them all, and an array of
708 * different handlers.
709 * We rely on user-space to write the initial superblock, and support
710 * reading and updating of superblocks.
711 * Interface methods are:
712 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
713 * loads and validates a superblock on dev.
714 * if refdev != NULL, compare superblocks on both devices
716 * 0 - dev has a superblock that is compatible with refdev
717 * 1 - dev has a superblock that is compatible and newer than refdev
718 * so dev should be used as the refdev in future
719 * -EINVAL superblock incompatible or invalid
720 * -othererror e.g. -EIO
722 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
723 * Verify that dev is acceptable into mddev.
724 * The first time, mddev->raid_disks will be 0, and data from
725 * dev should be merged in. Subsequent calls check that dev
726 * is new enough. Return 0 or -EINVAL
728 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
729 * Update the superblock for rdev with data in mddev
730 * This does not write to disc.
736 struct module
*owner
;
737 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
739 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
740 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
741 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
742 sector_t num_sectors
);
746 * Check that the given mddev has no bitmap.
748 * This function is called from the run method of all personalities that do not
749 * support bitmaps. It prints an error message and returns non-zero if mddev
750 * has a bitmap. Otherwise, it returns 0.
753 int md_check_no_bitmap(mddev_t
*mddev
)
755 if (!mddev
->bitmap_file
&& !mddev
->bitmap_offset
)
757 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
758 mdname(mddev
), mddev
->pers
->name
);
761 EXPORT_SYMBOL(md_check_no_bitmap
);
764 * load_super for 0.90.0
766 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
768 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
773 * Calculate the position of the superblock (512byte sectors),
774 * it's at the end of the disk.
776 * It also happens to be a multiple of 4Kb.
778 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
780 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
785 bdevname(rdev
->bdev
, b
);
786 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
788 if (sb
->md_magic
!= MD_SB_MAGIC
) {
789 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
794 if (sb
->major_version
!= 0 ||
795 sb
->minor_version
< 90 ||
796 sb
->minor_version
> 91) {
797 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
798 sb
->major_version
, sb
->minor_version
,
803 if (sb
->raid_disks
<= 0)
806 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
807 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
812 rdev
->preferred_minor
= sb
->md_minor
;
813 rdev
->data_offset
= 0;
814 rdev
->sb_size
= MD_SB_BYTES
;
816 if (sb
->level
== LEVEL_MULTIPATH
)
819 rdev
->desc_nr
= sb
->this_disk
.number
;
825 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
826 if (!uuid_equal(refsb
, sb
)) {
827 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
828 b
, bdevname(refdev
->bdev
,b2
));
831 if (!sb_equal(refsb
, sb
)) {
832 printk(KERN_WARNING
"md: %s has same UUID"
833 " but different superblock to %s\n",
834 b
, bdevname(refdev
->bdev
, b2
));
838 ev2
= md_event(refsb
);
844 rdev
->sectors
= rdev
->sb_start
;
846 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
847 /* "this cannot possibly happen" ... */
855 * validate_super for 0.90.0
857 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
860 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
861 __u64 ev1
= md_event(sb
);
863 rdev
->raid_disk
= -1;
864 clear_bit(Faulty
, &rdev
->flags
);
865 clear_bit(In_sync
, &rdev
->flags
);
866 clear_bit(WriteMostly
, &rdev
->flags
);
867 clear_bit(BarriersNotsupp
, &rdev
->flags
);
869 if (mddev
->raid_disks
== 0) {
870 mddev
->major_version
= 0;
871 mddev
->minor_version
= sb
->minor_version
;
872 mddev
->patch_version
= sb
->patch_version
;
874 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
875 mddev
->ctime
= sb
->ctime
;
876 mddev
->utime
= sb
->utime
;
877 mddev
->level
= sb
->level
;
878 mddev
->clevel
[0] = 0;
879 mddev
->layout
= sb
->layout
;
880 mddev
->raid_disks
= sb
->raid_disks
;
881 mddev
->dev_sectors
= sb
->size
* 2;
883 mddev
->bitmap_offset
= 0;
884 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
886 if (mddev
->minor_version
>= 91) {
887 mddev
->reshape_position
= sb
->reshape_position
;
888 mddev
->delta_disks
= sb
->delta_disks
;
889 mddev
->new_level
= sb
->new_level
;
890 mddev
->new_layout
= sb
->new_layout
;
891 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
893 mddev
->reshape_position
= MaxSector
;
894 mddev
->delta_disks
= 0;
895 mddev
->new_level
= mddev
->level
;
896 mddev
->new_layout
= mddev
->layout
;
897 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
900 if (sb
->state
& (1<<MD_SB_CLEAN
))
901 mddev
->recovery_cp
= MaxSector
;
903 if (sb
->events_hi
== sb
->cp_events_hi
&&
904 sb
->events_lo
== sb
->cp_events_lo
) {
905 mddev
->recovery_cp
= sb
->recovery_cp
;
907 mddev
->recovery_cp
= 0;
910 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
911 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
912 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
913 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
915 mddev
->max_disks
= MD_SB_DISKS
;
917 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
918 mddev
->bitmap_file
== NULL
)
919 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
921 } else if (mddev
->pers
== NULL
) {
922 /* Insist on good event counter while assembling */
924 if (ev1
< mddev
->events
)
926 } else if (mddev
->bitmap
) {
927 /* if adding to array with a bitmap, then we can accept an
928 * older device ... but not too old.
930 if (ev1
< mddev
->bitmap
->events_cleared
)
933 if (ev1
< mddev
->events
)
934 /* just a hot-add of a new device, leave raid_disk at -1 */
938 if (mddev
->level
!= LEVEL_MULTIPATH
) {
939 desc
= sb
->disks
+ rdev
->desc_nr
;
941 if (desc
->state
& (1<<MD_DISK_FAULTY
))
942 set_bit(Faulty
, &rdev
->flags
);
943 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
944 desc->raid_disk < mddev->raid_disks */) {
945 set_bit(In_sync
, &rdev
->flags
);
946 rdev
->raid_disk
= desc
->raid_disk
;
947 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
948 /* active but not in sync implies recovery up to
949 * reshape position. We don't know exactly where
950 * that is, so set to zero for now */
951 if (mddev
->minor_version
>= 91) {
952 rdev
->recovery_offset
= 0;
953 rdev
->raid_disk
= desc
->raid_disk
;
956 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
957 set_bit(WriteMostly
, &rdev
->flags
);
958 } else /* MULTIPATH are always insync */
959 set_bit(In_sync
, &rdev
->flags
);
964 * sync_super for 0.90.0
966 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
970 int next_spare
= mddev
->raid_disks
;
973 /* make rdev->sb match mddev data..
976 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
977 * 3/ any empty disks < next_spare become removed
979 * disks[0] gets initialised to REMOVED because
980 * we cannot be sure from other fields if it has
981 * been initialised or not.
984 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
986 rdev
->sb_size
= MD_SB_BYTES
;
988 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
990 memset(sb
, 0, sizeof(*sb
));
992 sb
->md_magic
= MD_SB_MAGIC
;
993 sb
->major_version
= mddev
->major_version
;
994 sb
->patch_version
= mddev
->patch_version
;
995 sb
->gvalid_words
= 0; /* ignored */
996 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
997 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
998 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
999 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1001 sb
->ctime
= mddev
->ctime
;
1002 sb
->level
= mddev
->level
;
1003 sb
->size
= mddev
->dev_sectors
/ 2;
1004 sb
->raid_disks
= mddev
->raid_disks
;
1005 sb
->md_minor
= mddev
->md_minor
;
1006 sb
->not_persistent
= 0;
1007 sb
->utime
= mddev
->utime
;
1009 sb
->events_hi
= (mddev
->events
>>32);
1010 sb
->events_lo
= (u32
)mddev
->events
;
1012 if (mddev
->reshape_position
== MaxSector
)
1013 sb
->minor_version
= 90;
1015 sb
->minor_version
= 91;
1016 sb
->reshape_position
= mddev
->reshape_position
;
1017 sb
->new_level
= mddev
->new_level
;
1018 sb
->delta_disks
= mddev
->delta_disks
;
1019 sb
->new_layout
= mddev
->new_layout
;
1020 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1022 mddev
->minor_version
= sb
->minor_version
;
1025 sb
->recovery_cp
= mddev
->recovery_cp
;
1026 sb
->cp_events_hi
= (mddev
->events
>>32);
1027 sb
->cp_events_lo
= (u32
)mddev
->events
;
1028 if (mddev
->recovery_cp
== MaxSector
)
1029 sb
->state
= (1<< MD_SB_CLEAN
);
1031 sb
->recovery_cp
= 0;
1033 sb
->layout
= mddev
->layout
;
1034 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1036 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
)
1037 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1039 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1040 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1043 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1045 if (rdev2
->raid_disk
>= 0 &&
1046 sb
->minor_version
>= 91)
1047 /* we have nowhere to store the recovery_offset,
1048 * but if it is not below the reshape_position,
1049 * we can piggy-back on that.
1052 if (rdev2
->raid_disk
< 0 ||
1053 test_bit(Faulty
, &rdev2
->flags
))
1056 desc_nr
= rdev2
->raid_disk
;
1058 desc_nr
= next_spare
++;
1059 rdev2
->desc_nr
= desc_nr
;
1060 d
= &sb
->disks
[rdev2
->desc_nr
];
1062 d
->number
= rdev2
->desc_nr
;
1063 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1064 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1066 d
->raid_disk
= rdev2
->raid_disk
;
1068 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1069 if (test_bit(Faulty
, &rdev2
->flags
))
1070 d
->state
= (1<<MD_DISK_FAULTY
);
1071 else if (is_active
) {
1072 d
->state
= (1<<MD_DISK_ACTIVE
);
1073 if (test_bit(In_sync
, &rdev2
->flags
))
1074 d
->state
|= (1<<MD_DISK_SYNC
);
1082 if (test_bit(WriteMostly
, &rdev2
->flags
))
1083 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1085 /* now set the "removed" and "faulty" bits on any missing devices */
1086 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1087 mdp_disk_t
*d
= &sb
->disks
[i
];
1088 if (d
->state
== 0 && d
->number
== 0) {
1091 d
->state
= (1<<MD_DISK_REMOVED
);
1092 d
->state
|= (1<<MD_DISK_FAULTY
);
1096 sb
->nr_disks
= nr_disks
;
1097 sb
->active_disks
= active
;
1098 sb
->working_disks
= working
;
1099 sb
->failed_disks
= failed
;
1100 sb
->spare_disks
= spare
;
1102 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1103 sb
->sb_csum
= calc_sb_csum(sb
);
1107 * rdev_size_change for 0.90.0
1109 static unsigned long long
1110 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1112 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1113 return 0; /* component must fit device */
1114 if (rdev
->mddev
->bitmap_offset
)
1115 return 0; /* can't move bitmap */
1116 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1117 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1118 num_sectors
= rdev
->sb_start
;
1119 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1121 md_super_wait(rdev
->mddev
);
1122 return num_sectors
/ 2; /* kB for sysfs */
1127 * version 1 superblock
1130 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1134 unsigned long long newcsum
;
1135 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1136 __le32
*isuper
= (__le32
*)sb
;
1139 disk_csum
= sb
->sb_csum
;
1142 for (i
=0; size
>=4; size
-= 4 )
1143 newcsum
+= le32_to_cpu(*isuper
++);
1146 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1148 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1149 sb
->sb_csum
= disk_csum
;
1150 return cpu_to_le32(csum
);
1153 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1155 struct mdp_superblock_1
*sb
;
1158 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1162 * Calculate the position of the superblock in 512byte sectors.
1163 * It is always aligned to a 4K boundary and
1164 * depeding on minor_version, it can be:
1165 * 0: At least 8K, but less than 12K, from end of device
1166 * 1: At start of device
1167 * 2: 4K from start of device.
1169 switch(minor_version
) {
1171 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1173 sb_start
&= ~(sector_t
)(4*2-1);
1184 rdev
->sb_start
= sb_start
;
1186 /* superblock is rarely larger than 1K, but it can be larger,
1187 * and it is safe to read 4k, so we do that
1189 ret
= read_disk_sb(rdev
, 4096);
1190 if (ret
) return ret
;
1193 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1195 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1196 sb
->major_version
!= cpu_to_le32(1) ||
1197 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1198 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1199 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1202 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1203 printk("md: invalid superblock checksum on %s\n",
1204 bdevname(rdev
->bdev
,b
));
1207 if (le64_to_cpu(sb
->data_size
) < 10) {
1208 printk("md: data_size too small on %s\n",
1209 bdevname(rdev
->bdev
,b
));
1213 rdev
->preferred_minor
= 0xffff;
1214 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1215 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1217 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1218 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1219 if (rdev
->sb_size
& bmask
)
1220 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1223 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1226 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1229 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1235 struct mdp_superblock_1
*refsb
=
1236 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1238 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1239 sb
->level
!= refsb
->level
||
1240 sb
->layout
!= refsb
->layout
||
1241 sb
->chunksize
!= refsb
->chunksize
) {
1242 printk(KERN_WARNING
"md: %s has strangely different"
1243 " superblock to %s\n",
1244 bdevname(rdev
->bdev
,b
),
1245 bdevname(refdev
->bdev
,b2
));
1248 ev1
= le64_to_cpu(sb
->events
);
1249 ev2
= le64_to_cpu(refsb
->events
);
1257 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1258 le64_to_cpu(sb
->data_offset
);
1260 rdev
->sectors
= rdev
->sb_start
;
1261 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1263 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1264 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1269 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1271 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1272 __u64 ev1
= le64_to_cpu(sb
->events
);
1274 rdev
->raid_disk
= -1;
1275 clear_bit(Faulty
, &rdev
->flags
);
1276 clear_bit(In_sync
, &rdev
->flags
);
1277 clear_bit(WriteMostly
, &rdev
->flags
);
1278 clear_bit(BarriersNotsupp
, &rdev
->flags
);
1280 if (mddev
->raid_disks
== 0) {
1281 mddev
->major_version
= 1;
1282 mddev
->patch_version
= 0;
1283 mddev
->external
= 0;
1284 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1285 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1286 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1287 mddev
->level
= le32_to_cpu(sb
->level
);
1288 mddev
->clevel
[0] = 0;
1289 mddev
->layout
= le32_to_cpu(sb
->layout
);
1290 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1291 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1292 mddev
->events
= ev1
;
1293 mddev
->bitmap_offset
= 0;
1294 mddev
->default_bitmap_offset
= 1024 >> 9;
1296 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1297 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1299 mddev
->max_disks
= (4096-256)/2;
1301 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1302 mddev
->bitmap_file
== NULL
)
1303 mddev
->bitmap_offset
= (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1305 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1306 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1307 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1308 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1309 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1310 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1312 mddev
->reshape_position
= MaxSector
;
1313 mddev
->delta_disks
= 0;
1314 mddev
->new_level
= mddev
->level
;
1315 mddev
->new_layout
= mddev
->layout
;
1316 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1319 } else if (mddev
->pers
== NULL
) {
1320 /* Insist of good event counter while assembling */
1322 if (ev1
< mddev
->events
)
1324 } else if (mddev
->bitmap
) {
1325 /* If adding to array with a bitmap, then we can accept an
1326 * older device, but not too old.
1328 if (ev1
< mddev
->bitmap
->events_cleared
)
1331 if (ev1
< mddev
->events
)
1332 /* just a hot-add of a new device, leave raid_disk at -1 */
1335 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1337 if (rdev
->desc_nr
< 0 ||
1338 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1342 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1344 case 0xffff: /* spare */
1346 case 0xfffe: /* faulty */
1347 set_bit(Faulty
, &rdev
->flags
);
1350 if ((le32_to_cpu(sb
->feature_map
) &
1351 MD_FEATURE_RECOVERY_OFFSET
))
1352 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1354 set_bit(In_sync
, &rdev
->flags
);
1355 rdev
->raid_disk
= role
;
1358 if (sb
->devflags
& WriteMostly1
)
1359 set_bit(WriteMostly
, &rdev
->flags
);
1360 } else /* MULTIPATH are always insync */
1361 set_bit(In_sync
, &rdev
->flags
);
1366 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1368 struct mdp_superblock_1
*sb
;
1371 /* make rdev->sb match mddev and rdev data. */
1373 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1375 sb
->feature_map
= 0;
1377 sb
->recovery_offset
= cpu_to_le64(0);
1378 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1379 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1380 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1382 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1383 sb
->events
= cpu_to_le64(mddev
->events
);
1385 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1387 sb
->resync_offset
= cpu_to_le64(0);
1389 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1391 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1392 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1393 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1394 sb
->level
= cpu_to_le32(mddev
->level
);
1395 sb
->layout
= cpu_to_le32(mddev
->layout
);
1397 if (mddev
->bitmap
&& mddev
->bitmap_file
== NULL
) {
1398 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_offset
);
1399 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1402 if (rdev
->raid_disk
>= 0 &&
1403 !test_bit(In_sync
, &rdev
->flags
)) {
1404 if (rdev
->recovery_offset
> 0) {
1406 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1407 sb
->recovery_offset
=
1408 cpu_to_le64(rdev
->recovery_offset
);
1412 if (mddev
->reshape_position
!= MaxSector
) {
1413 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1414 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1415 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1416 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1417 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1418 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1422 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1423 if (rdev2
->desc_nr
+1 > max_dev
)
1424 max_dev
= rdev2
->desc_nr
+1;
1426 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1428 sb
->max_dev
= cpu_to_le32(max_dev
);
1429 rdev
->sb_size
= max_dev
* 2 + 256;
1430 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1431 if (rdev
->sb_size
& bmask
)
1432 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1434 for (i
=0; i
<max_dev
;i
++)
1435 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1437 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1439 if (test_bit(Faulty
, &rdev2
->flags
))
1440 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1441 else if (test_bit(In_sync
, &rdev2
->flags
))
1442 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1443 else if (rdev2
->raid_disk
>= 0 && rdev2
->recovery_offset
> 0)
1444 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1446 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1449 sb
->sb_csum
= calc_sb_1_csum(sb
);
1452 static unsigned long long
1453 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1455 struct mdp_superblock_1
*sb
;
1456 sector_t max_sectors
;
1457 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1458 return 0; /* component must fit device */
1459 if (rdev
->sb_start
< rdev
->data_offset
) {
1460 /* minor versions 1 and 2; superblock before data */
1461 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1462 max_sectors
-= rdev
->data_offset
;
1463 if (!num_sectors
|| num_sectors
> max_sectors
)
1464 num_sectors
= max_sectors
;
1465 } else if (rdev
->mddev
->bitmap_offset
) {
1466 /* minor version 0 with bitmap we can't move */
1469 /* minor version 0; superblock after data */
1471 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1472 sb_start
&= ~(sector_t
)(4*2 - 1);
1473 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1474 if (!num_sectors
|| num_sectors
> max_sectors
)
1475 num_sectors
= max_sectors
;
1476 rdev
->sb_start
= sb_start
;
1478 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1479 sb
->data_size
= cpu_to_le64(num_sectors
);
1480 sb
->super_offset
= rdev
->sb_start
;
1481 sb
->sb_csum
= calc_sb_1_csum(sb
);
1482 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1484 md_super_wait(rdev
->mddev
);
1485 return num_sectors
/ 2; /* kB for sysfs */
1488 static struct super_type super_types
[] = {
1491 .owner
= THIS_MODULE
,
1492 .load_super
= super_90_load
,
1493 .validate_super
= super_90_validate
,
1494 .sync_super
= super_90_sync
,
1495 .rdev_size_change
= super_90_rdev_size_change
,
1499 .owner
= THIS_MODULE
,
1500 .load_super
= super_1_load
,
1501 .validate_super
= super_1_validate
,
1502 .sync_super
= super_1_sync
,
1503 .rdev_size_change
= super_1_rdev_size_change
,
1507 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1509 mdk_rdev_t
*rdev
, *rdev2
;
1512 rdev_for_each_rcu(rdev
, mddev1
)
1513 rdev_for_each_rcu(rdev2
, mddev2
)
1514 if (rdev
->bdev
->bd_contains
==
1515 rdev2
->bdev
->bd_contains
) {
1523 static LIST_HEAD(pending_raid_disks
);
1526 * Try to register data integrity profile for an mddev
1528 * This is called when an array is started and after a disk has been kicked
1529 * from the array. It only succeeds if all working and active component devices
1530 * are integrity capable with matching profiles.
1532 int md_integrity_register(mddev_t
*mddev
)
1534 mdk_rdev_t
*rdev
, *reference
= NULL
;
1536 if (list_empty(&mddev
->disks
))
1537 return 0; /* nothing to do */
1538 if (blk_get_integrity(mddev
->gendisk
))
1539 return 0; /* already registered */
1540 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1541 /* skip spares and non-functional disks */
1542 if (test_bit(Faulty
, &rdev
->flags
))
1544 if (rdev
->raid_disk
< 0)
1547 * If at least one rdev is not integrity capable, we can not
1548 * enable data integrity for the md device.
1550 if (!bdev_get_integrity(rdev
->bdev
))
1553 /* Use the first rdev as the reference */
1557 /* does this rdev's profile match the reference profile? */
1558 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1559 rdev
->bdev
->bd_disk
) < 0)
1563 * All component devices are integrity capable and have matching
1564 * profiles, register the common profile for the md device.
1566 if (blk_integrity_register(mddev
->gendisk
,
1567 bdev_get_integrity(reference
->bdev
)) != 0) {
1568 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1572 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1576 EXPORT_SYMBOL(md_integrity_register
);
1578 /* Disable data integrity if non-capable/non-matching disk is being added */
1579 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1581 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1582 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1584 if (!bi_mddev
) /* nothing to do */
1586 if (rdev
->raid_disk
< 0) /* skip spares */
1588 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1589 rdev
->bdev
->bd_disk
) >= 0)
1591 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1592 blk_integrity_unregister(mddev
->gendisk
);
1594 EXPORT_SYMBOL(md_integrity_add_rdev
);
1596 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1598 char b
[BDEVNAME_SIZE
];
1608 /* prevent duplicates */
1609 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1612 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1613 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1614 rdev
->sectors
< mddev
->dev_sectors
)) {
1616 /* Cannot change size, so fail
1617 * If mddev->level <= 0, then we don't care
1618 * about aligning sizes (e.g. linear)
1620 if (mddev
->level
> 0)
1623 mddev
->dev_sectors
= rdev
->sectors
;
1626 /* Verify rdev->desc_nr is unique.
1627 * If it is -1, assign a free number, else
1628 * check number is not in use
1630 if (rdev
->desc_nr
< 0) {
1632 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1633 while (find_rdev_nr(mddev
, choice
))
1635 rdev
->desc_nr
= choice
;
1637 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1640 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1641 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1642 mdname(mddev
), mddev
->max_disks
);
1645 bdevname(rdev
->bdev
,b
);
1646 while ( (s
=strchr(b
, '/')) != NULL
)
1649 rdev
->mddev
= mddev
;
1650 printk(KERN_INFO
"md: bind<%s>\n", b
);
1652 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1655 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1656 if ((err
= sysfs_create_link(&rdev
->kobj
, ko
, "block"))) {
1657 kobject_del(&rdev
->kobj
);
1660 rdev
->sysfs_state
= sysfs_get_dirent(rdev
->kobj
.sd
, "state");
1662 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1663 bd_claim_by_disk(rdev
->bdev
, rdev
->bdev
->bd_holder
, mddev
->gendisk
);
1665 /* May as well allow recovery to be retried once */
1666 mddev
->recovery_disabled
= 0;
1671 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1676 static void md_delayed_delete(struct work_struct
*ws
)
1678 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1679 kobject_del(&rdev
->kobj
);
1680 kobject_put(&rdev
->kobj
);
1683 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1685 char b
[BDEVNAME_SIZE
];
1690 bd_release_from_disk(rdev
->bdev
, rdev
->mddev
->gendisk
);
1691 list_del_rcu(&rdev
->same_set
);
1692 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1694 sysfs_remove_link(&rdev
->kobj
, "block");
1695 sysfs_put(rdev
->sysfs_state
);
1696 rdev
->sysfs_state
= NULL
;
1697 /* We need to delay this, otherwise we can deadlock when
1698 * writing to 'remove' to "dev/state". We also need
1699 * to delay it due to rcu usage.
1702 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1703 kobject_get(&rdev
->kobj
);
1704 schedule_work(&rdev
->del_work
);
1708 * prevent the device from being mounted, repartitioned or
1709 * otherwise reused by a RAID array (or any other kernel
1710 * subsystem), by bd_claiming the device.
1712 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1715 struct block_device
*bdev
;
1716 char b
[BDEVNAME_SIZE
];
1718 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
);
1720 printk(KERN_ERR
"md: could not open %s.\n",
1721 __bdevname(dev
, b
));
1722 return PTR_ERR(bdev
);
1724 err
= bd_claim(bdev
, shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1726 printk(KERN_ERR
"md: could not bd_claim %s.\n",
1728 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1732 set_bit(AllReserved
, &rdev
->flags
);
1737 static void unlock_rdev(mdk_rdev_t
*rdev
)
1739 struct block_device
*bdev
= rdev
->bdev
;
1744 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
);
1747 void md_autodetect_dev(dev_t dev
);
1749 static void export_rdev(mdk_rdev_t
* rdev
)
1751 char b
[BDEVNAME_SIZE
];
1752 printk(KERN_INFO
"md: export_rdev(%s)\n",
1753 bdevname(rdev
->bdev
,b
));
1758 if (test_bit(AutoDetected
, &rdev
->flags
))
1759 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1762 kobject_put(&rdev
->kobj
);
1765 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1767 unbind_rdev_from_array(rdev
);
1771 static void export_array(mddev_t
*mddev
)
1773 mdk_rdev_t
*rdev
, *tmp
;
1775 rdev_for_each(rdev
, tmp
, mddev
) {
1780 kick_rdev_from_array(rdev
);
1782 if (!list_empty(&mddev
->disks
))
1784 mddev
->raid_disks
= 0;
1785 mddev
->major_version
= 0;
1788 static void print_desc(mdp_disk_t
*desc
)
1790 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
1791 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
1794 static void print_sb_90(mdp_super_t
*sb
)
1799 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1800 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
1801 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
1803 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1804 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
1805 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
1806 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
1807 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1808 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
1809 sb
->failed_disks
, sb
->spare_disks
,
1810 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
1813 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
1816 desc
= sb
->disks
+ i
;
1817 if (desc
->number
|| desc
->major
|| desc
->minor
||
1818 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
1819 printk(" D %2d: ", i
);
1823 printk(KERN_INFO
"md: THIS: ");
1824 print_desc(&sb
->this_disk
);
1827 static void print_sb_1(struct mdp_superblock_1
*sb
)
1831 uuid
= sb
->set_uuid
;
1833 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1834 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1835 "md: Name: \"%s\" CT:%llu\n",
1836 le32_to_cpu(sb
->major_version
),
1837 le32_to_cpu(sb
->feature_map
),
1838 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1839 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1840 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1841 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1843 (unsigned long long)le64_to_cpu(sb
->ctime
)
1844 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
1846 uuid
= sb
->device_uuid
;
1848 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1850 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1851 ":%02x%02x%02x%02x%02x%02x\n"
1852 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1853 "md: (MaxDev:%u) \n",
1854 le32_to_cpu(sb
->level
),
1855 (unsigned long long)le64_to_cpu(sb
->size
),
1856 le32_to_cpu(sb
->raid_disks
),
1857 le32_to_cpu(sb
->layout
),
1858 le32_to_cpu(sb
->chunksize
),
1859 (unsigned long long)le64_to_cpu(sb
->data_offset
),
1860 (unsigned long long)le64_to_cpu(sb
->data_size
),
1861 (unsigned long long)le64_to_cpu(sb
->super_offset
),
1862 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
1863 le32_to_cpu(sb
->dev_number
),
1864 uuid
[0], uuid
[1], uuid
[2], uuid
[3],
1865 uuid
[4], uuid
[5], uuid
[6], uuid
[7],
1866 uuid
[8], uuid
[9], uuid
[10], uuid
[11],
1867 uuid
[12], uuid
[13], uuid
[14], uuid
[15],
1869 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
1870 (unsigned long long)le64_to_cpu(sb
->events
),
1871 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
1872 le32_to_cpu(sb
->sb_csum
),
1873 le32_to_cpu(sb
->max_dev
)
1877 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
1879 char b
[BDEVNAME_SIZE
];
1880 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1881 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
1882 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
1884 if (rdev
->sb_loaded
) {
1885 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
1886 switch (major_version
) {
1888 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
1891 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
1895 printk(KERN_INFO
"md: no rdev superblock!\n");
1898 static void md_print_devices(void)
1900 struct list_head
*tmp
;
1903 char b
[BDEVNAME_SIZE
];
1906 printk("md: **********************************\n");
1907 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1908 printk("md: **********************************\n");
1909 for_each_mddev(mddev
, tmp
) {
1912 bitmap_print_sb(mddev
->bitmap
);
1914 printk("%s: ", mdname(mddev
));
1915 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
1916 printk("<%s>", bdevname(rdev
->bdev
,b
));
1919 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
1920 print_rdev(rdev
, mddev
->major_version
);
1922 printk("md: **********************************\n");
1927 static void sync_sbs(mddev_t
* mddev
, int nospares
)
1929 /* Update each superblock (in-memory image), but
1930 * if we are allowed to, skip spares which already
1931 * have the right event counter, or have one earlier
1932 * (which would mean they aren't being marked as dirty
1933 * with the rest of the array)
1937 /* First make sure individual recovery_offsets are correct */
1938 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1939 if (rdev
->raid_disk
>= 0 &&
1940 !test_bit(In_sync
, &rdev
->flags
) &&
1941 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
1942 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
1945 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1946 if (rdev
->sb_events
== mddev
->events
||
1948 rdev
->raid_disk
< 0 &&
1949 (rdev
->sb_events
&1)==0 &&
1950 rdev
->sb_events
+1 == mddev
->events
)) {
1951 /* Don't update this superblock */
1952 rdev
->sb_loaded
= 2;
1954 super_types
[mddev
->major_version
].
1955 sync_super(mddev
, rdev
);
1956 rdev
->sb_loaded
= 1;
1961 static void md_update_sb(mddev_t
* mddev
, int force_change
)
1967 mddev
->utime
= get_seconds();
1968 if (mddev
->external
)
1971 spin_lock_irq(&mddev
->write_lock
);
1973 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
1974 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
1976 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
1977 /* just a clean<-> dirty transition, possibly leave spares alone,
1978 * though if events isn't the right even/odd, we will have to do
1984 if (mddev
->degraded
)
1985 /* If the array is degraded, then skipping spares is both
1986 * dangerous and fairly pointless.
1987 * Dangerous because a device that was removed from the array
1988 * might have a event_count that still looks up-to-date,
1989 * so it can be re-added without a resync.
1990 * Pointless because if there are any spares to skip,
1991 * then a recovery will happen and soon that array won't
1992 * be degraded any more and the spare can go back to sleep then.
1996 sync_req
= mddev
->in_sync
;
1998 /* If this is just a dirty<->clean transition, and the array is clean
1999 * and 'events' is odd, we can roll back to the previous clean state */
2001 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2002 && (mddev
->events
& 1)
2003 && mddev
->events
!= 1)
2006 /* otherwise we have to go forward and ... */
2008 if (!mddev
->in_sync
|| mddev
->recovery_cp
!= MaxSector
) { /* not clean */
2009 /* .. if the array isn't clean, an 'even' event must also go
2011 if ((mddev
->events
&1)==0)
2014 /* otherwise an 'odd' event must go to spares */
2015 if ((mddev
->events
&1))
2020 if (!mddev
->events
) {
2022 * oops, this 64-bit counter should never wrap.
2023 * Either we are in around ~1 trillion A.C., assuming
2024 * 1 reboot per second, or we have a bug:
2031 * do not write anything to disk if using
2032 * nonpersistent superblocks
2034 if (!mddev
->persistent
) {
2035 if (!mddev
->external
)
2036 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2038 spin_unlock_irq(&mddev
->write_lock
);
2039 wake_up(&mddev
->sb_wait
);
2042 sync_sbs(mddev
, nospares
);
2043 spin_unlock_irq(&mddev
->write_lock
);
2046 "md: updating %s RAID superblock on device (in sync %d)\n",
2047 mdname(mddev
),mddev
->in_sync
);
2049 bitmap_update_sb(mddev
->bitmap
);
2050 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2051 char b
[BDEVNAME_SIZE
];
2052 dprintk(KERN_INFO
"md: ");
2053 if (rdev
->sb_loaded
!= 1)
2054 continue; /* no noise on spare devices */
2055 if (test_bit(Faulty
, &rdev
->flags
))
2056 dprintk("(skipping faulty ");
2058 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2059 if (!test_bit(Faulty
, &rdev
->flags
)) {
2060 md_super_write(mddev
,rdev
,
2061 rdev
->sb_start
, rdev
->sb_size
,
2063 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2064 bdevname(rdev
->bdev
,b
),
2065 (unsigned long long)rdev
->sb_start
);
2066 rdev
->sb_events
= mddev
->events
;
2070 if (mddev
->level
== LEVEL_MULTIPATH
)
2071 /* only need to write one superblock... */
2074 md_super_wait(mddev
);
2075 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2077 spin_lock_irq(&mddev
->write_lock
);
2078 if (mddev
->in_sync
!= sync_req
||
2079 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2080 /* have to write it out again */
2081 spin_unlock_irq(&mddev
->write_lock
);
2084 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2085 spin_unlock_irq(&mddev
->write_lock
);
2086 wake_up(&mddev
->sb_wait
);
2087 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2088 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2092 /* words written to sysfs files may, or may not, be \n terminated.
2093 * We want to accept with case. For this we use cmd_match.
2095 static int cmd_match(const char *cmd
, const char *str
)
2097 /* See if cmd, written into a sysfs file, matches
2098 * str. They must either be the same, or cmd can
2099 * have a trailing newline
2101 while (*cmd
&& *str
&& *cmd
== *str
) {
2112 struct rdev_sysfs_entry
{
2113 struct attribute attr
;
2114 ssize_t (*show
)(mdk_rdev_t
*, char *);
2115 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2119 state_show(mdk_rdev_t
*rdev
, char *page
)
2124 if (test_bit(Faulty
, &rdev
->flags
)) {
2125 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2128 if (test_bit(In_sync
, &rdev
->flags
)) {
2129 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2132 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2133 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2136 if (test_bit(Blocked
, &rdev
->flags
)) {
2137 len
+= sprintf(page
+len
, "%sblocked", sep
);
2140 if (!test_bit(Faulty
, &rdev
->flags
) &&
2141 !test_bit(In_sync
, &rdev
->flags
)) {
2142 len
+= sprintf(page
+len
, "%sspare", sep
);
2145 return len
+sprintf(page
+len
, "\n");
2149 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2152 * faulty - simulates and error
2153 * remove - disconnects the device
2154 * writemostly - sets write_mostly
2155 * -writemostly - clears write_mostly
2156 * blocked - sets the Blocked flag
2157 * -blocked - clears the Blocked flag
2158 * insync - sets Insync providing device isn't active
2161 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2162 md_error(rdev
->mddev
, rdev
);
2164 } else if (cmd_match(buf
, "remove")) {
2165 if (rdev
->raid_disk
>= 0)
2168 mddev_t
*mddev
= rdev
->mddev
;
2169 kick_rdev_from_array(rdev
);
2171 md_update_sb(mddev
, 1);
2172 md_new_event(mddev
);
2175 } else if (cmd_match(buf
, "writemostly")) {
2176 set_bit(WriteMostly
, &rdev
->flags
);
2178 } else if (cmd_match(buf
, "-writemostly")) {
2179 clear_bit(WriteMostly
, &rdev
->flags
);
2181 } else if (cmd_match(buf
, "blocked")) {
2182 set_bit(Blocked
, &rdev
->flags
);
2184 } else if (cmd_match(buf
, "-blocked")) {
2185 clear_bit(Blocked
, &rdev
->flags
);
2186 wake_up(&rdev
->blocked_wait
);
2187 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2188 md_wakeup_thread(rdev
->mddev
->thread
);
2191 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2192 set_bit(In_sync
, &rdev
->flags
);
2195 if (!err
&& rdev
->sysfs_state
)
2196 sysfs_notify_dirent(rdev
->sysfs_state
);
2197 return err
? err
: len
;
2199 static struct rdev_sysfs_entry rdev_state
=
2200 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2203 errors_show(mdk_rdev_t
*rdev
, char *page
)
2205 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2209 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2212 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2213 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2214 atomic_set(&rdev
->corrected_errors
, n
);
2219 static struct rdev_sysfs_entry rdev_errors
=
2220 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2223 slot_show(mdk_rdev_t
*rdev
, char *page
)
2225 if (rdev
->raid_disk
< 0)
2226 return sprintf(page
, "none\n");
2228 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2232 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2237 int slot
= simple_strtoul(buf
, &e
, 10);
2238 if (strncmp(buf
, "none", 4)==0)
2240 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2242 if (rdev
->mddev
->pers
&& slot
== -1) {
2243 /* Setting 'slot' on an active array requires also
2244 * updating the 'rd%d' link, and communicating
2245 * with the personality with ->hot_*_disk.
2246 * For now we only support removing
2247 * failed/spare devices. This normally happens automatically,
2248 * but not when the metadata is externally managed.
2250 if (rdev
->raid_disk
== -1)
2252 /* personality does all needed checks */
2253 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2255 err
= rdev
->mddev
->pers
->
2256 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2259 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2260 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2261 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2262 md_wakeup_thread(rdev
->mddev
->thread
);
2263 } else if (rdev
->mddev
->pers
) {
2265 /* Activating a spare .. or possibly reactivating
2266 * if we ever get bitmaps working here.
2269 if (rdev
->raid_disk
!= -1)
2272 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2275 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2276 if (rdev2
->raid_disk
== slot
)
2279 rdev
->raid_disk
= slot
;
2280 if (test_bit(In_sync
, &rdev
->flags
))
2281 rdev
->saved_raid_disk
= slot
;
2283 rdev
->saved_raid_disk
= -1;
2284 err
= rdev
->mddev
->pers
->
2285 hot_add_disk(rdev
->mddev
, rdev
);
2287 rdev
->raid_disk
= -1;
2290 sysfs_notify_dirent(rdev
->sysfs_state
);
2291 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2292 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2294 "md: cannot register "
2296 nm
, mdname(rdev
->mddev
));
2298 /* don't wakeup anyone, leave that to userspace. */
2300 if (slot
>= rdev
->mddev
->raid_disks
)
2302 rdev
->raid_disk
= slot
;
2303 /* assume it is working */
2304 clear_bit(Faulty
, &rdev
->flags
);
2305 clear_bit(WriteMostly
, &rdev
->flags
);
2306 set_bit(In_sync
, &rdev
->flags
);
2307 sysfs_notify_dirent(rdev
->sysfs_state
);
2313 static struct rdev_sysfs_entry rdev_slot
=
2314 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2317 offset_show(mdk_rdev_t
*rdev
, char *page
)
2319 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2323 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2326 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2327 if (e
==buf
|| (*e
&& *e
!= '\n'))
2329 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2331 if (rdev
->sectors
&& rdev
->mddev
->external
)
2332 /* Must set offset before size, so overlap checks
2335 rdev
->data_offset
= offset
;
2339 static struct rdev_sysfs_entry rdev_offset
=
2340 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2343 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2345 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2348 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2350 /* check if two start/length pairs overlap */
2358 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2360 unsigned long long blocks
;
2363 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2366 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2367 return -EINVAL
; /* sector conversion overflow */
2370 if (new != blocks
* 2)
2371 return -EINVAL
; /* unsigned long long to sector_t overflow */
2378 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2380 mddev_t
*my_mddev
= rdev
->mddev
;
2381 sector_t oldsectors
= rdev
->sectors
;
2384 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2386 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2387 if (my_mddev
->persistent
) {
2388 sectors
= super_types
[my_mddev
->major_version
].
2389 rdev_size_change(rdev
, sectors
);
2392 } else if (!sectors
)
2393 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2396 if (sectors
< my_mddev
->dev_sectors
)
2397 return -EINVAL
; /* component must fit device */
2399 rdev
->sectors
= sectors
;
2400 if (sectors
> oldsectors
&& my_mddev
->external
) {
2401 /* need to check that all other rdevs with the same ->bdev
2402 * do not overlap. We need to unlock the mddev to avoid
2403 * a deadlock. We have already changed rdev->sectors, and if
2404 * we have to change it back, we will have the lock again.
2408 struct list_head
*tmp
;
2410 mddev_unlock(my_mddev
);
2411 for_each_mddev(mddev
, tmp
) {
2415 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2416 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2417 (rdev
->bdev
== rdev2
->bdev
&&
2419 overlaps(rdev
->data_offset
, rdev
->sectors
,
2425 mddev_unlock(mddev
);
2431 mddev_lock(my_mddev
);
2433 /* Someone else could have slipped in a size
2434 * change here, but doing so is just silly.
2435 * We put oldsectors back because we *know* it is
2436 * safe, and trust userspace not to race with
2439 rdev
->sectors
= oldsectors
;
2446 static struct rdev_sysfs_entry rdev_size
=
2447 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2449 static struct attribute
*rdev_default_attrs
[] = {
2458 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2460 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2461 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2462 mddev_t
*mddev
= rdev
->mddev
;
2468 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2470 if (rdev
->mddev
== NULL
)
2473 rv
= entry
->show(rdev
, page
);
2474 mddev_unlock(mddev
);
2480 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2481 const char *page
, size_t length
)
2483 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2484 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2486 mddev_t
*mddev
= rdev
->mddev
;
2490 if (!capable(CAP_SYS_ADMIN
))
2492 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2494 if (rdev
->mddev
== NULL
)
2497 rv
= entry
->store(rdev
, page
, length
);
2498 mddev_unlock(mddev
);
2503 static void rdev_free(struct kobject
*ko
)
2505 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2508 static struct sysfs_ops rdev_sysfs_ops
= {
2509 .show
= rdev_attr_show
,
2510 .store
= rdev_attr_store
,
2512 static struct kobj_type rdev_ktype
= {
2513 .release
= rdev_free
,
2514 .sysfs_ops
= &rdev_sysfs_ops
,
2515 .default_attrs
= rdev_default_attrs
,
2519 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2521 * mark the device faulty if:
2523 * - the device is nonexistent (zero size)
2524 * - the device has no valid superblock
2526 * a faulty rdev _never_ has rdev->sb set.
2528 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2530 char b
[BDEVNAME_SIZE
];
2535 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2537 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2538 return ERR_PTR(-ENOMEM
);
2541 if ((err
= alloc_disk_sb(rdev
)))
2544 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2548 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2551 rdev
->saved_raid_disk
= -1;
2552 rdev
->raid_disk
= -1;
2554 rdev
->data_offset
= 0;
2555 rdev
->sb_events
= 0;
2556 atomic_set(&rdev
->nr_pending
, 0);
2557 atomic_set(&rdev
->read_errors
, 0);
2558 atomic_set(&rdev
->corrected_errors
, 0);
2560 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2563 "md: %s has zero or unknown size, marking faulty!\n",
2564 bdevname(rdev
->bdev
,b
));
2569 if (super_format
>= 0) {
2570 err
= super_types
[super_format
].
2571 load_super(rdev
, NULL
, super_minor
);
2572 if (err
== -EINVAL
) {
2574 "md: %s does not have a valid v%d.%d "
2575 "superblock, not importing!\n",
2576 bdevname(rdev
->bdev
,b
),
2577 super_format
, super_minor
);
2582 "md: could not read %s's sb, not importing!\n",
2583 bdevname(rdev
->bdev
,b
));
2588 INIT_LIST_HEAD(&rdev
->same_set
);
2589 init_waitqueue_head(&rdev
->blocked_wait
);
2594 if (rdev
->sb_page
) {
2600 return ERR_PTR(err
);
2604 * Check a full RAID array for plausibility
2608 static void analyze_sbs(mddev_t
* mddev
)
2611 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2612 char b
[BDEVNAME_SIZE
];
2615 rdev_for_each(rdev
, tmp
, mddev
)
2616 switch (super_types
[mddev
->major_version
].
2617 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2625 "md: fatal superblock inconsistency in %s"
2626 " -- removing from array\n",
2627 bdevname(rdev
->bdev
,b
));
2628 kick_rdev_from_array(rdev
);
2632 super_types
[mddev
->major_version
].
2633 validate_super(mddev
, freshest
);
2636 rdev_for_each(rdev
, tmp
, mddev
) {
2637 if (rdev
->desc_nr
>= mddev
->max_disks
||
2638 i
> mddev
->max_disks
) {
2640 "md: %s: %s: only %d devices permitted\n",
2641 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2643 kick_rdev_from_array(rdev
);
2646 if (rdev
!= freshest
)
2647 if (super_types
[mddev
->major_version
].
2648 validate_super(mddev
, rdev
)) {
2649 printk(KERN_WARNING
"md: kicking non-fresh %s"
2651 bdevname(rdev
->bdev
,b
));
2652 kick_rdev_from_array(rdev
);
2655 if (mddev
->level
== LEVEL_MULTIPATH
) {
2656 rdev
->desc_nr
= i
++;
2657 rdev
->raid_disk
= rdev
->desc_nr
;
2658 set_bit(In_sync
, &rdev
->flags
);
2659 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2660 rdev
->raid_disk
= -1;
2661 clear_bit(In_sync
, &rdev
->flags
);
2666 static void md_safemode_timeout(unsigned long data
);
2669 safe_delay_show(mddev_t
*mddev
, char *page
)
2671 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2672 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2675 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2683 /* remove a period, and count digits after it */
2684 if (len
>= sizeof(buf
))
2686 strlcpy(buf
, cbuf
, sizeof(buf
));
2687 for (i
=0; i
<len
; i
++) {
2689 if (isdigit(buf
[i
])) {
2694 } else if (buf
[i
] == '.') {
2699 if (strict_strtoul(buf
, 10, &msec
) < 0)
2701 msec
= (msec
* 1000) / scale
;
2703 mddev
->safemode_delay
= 0;
2705 unsigned long old_delay
= mddev
->safemode_delay
;
2706 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2707 if (mddev
->safemode_delay
== 0)
2708 mddev
->safemode_delay
= 1;
2709 if (mddev
->safemode_delay
< old_delay
)
2710 md_safemode_timeout((unsigned long)mddev
);
2714 static struct md_sysfs_entry md_safe_delay
=
2715 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2718 level_show(mddev_t
*mddev
, char *page
)
2720 struct mdk_personality
*p
= mddev
->pers
;
2722 return sprintf(page
, "%s\n", p
->name
);
2723 else if (mddev
->clevel
[0])
2724 return sprintf(page
, "%s\n", mddev
->clevel
);
2725 else if (mddev
->level
!= LEVEL_NONE
)
2726 return sprintf(page
, "%d\n", mddev
->level
);
2732 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2736 struct mdk_personality
*pers
;
2740 if (mddev
->pers
== NULL
) {
2743 if (len
>= sizeof(mddev
->clevel
))
2745 strncpy(mddev
->clevel
, buf
, len
);
2746 if (mddev
->clevel
[len
-1] == '\n')
2748 mddev
->clevel
[len
] = 0;
2749 mddev
->level
= LEVEL_NONE
;
2753 /* request to change the personality. Need to ensure:
2754 * - array is not engaged in resync/recovery/reshape
2755 * - old personality can be suspended
2756 * - new personality will access other array.
2759 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
2762 if (!mddev
->pers
->quiesce
) {
2763 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
2764 mdname(mddev
), mddev
->pers
->name
);
2768 /* Now find the new personality */
2769 if (len
== 0 || len
>= sizeof(level
))
2771 strncpy(level
, buf
, len
);
2772 if (level
[len
-1] == '\n')
2776 request_module("md-%s", level
);
2777 spin_lock(&pers_lock
);
2778 pers
= find_pers(LEVEL_NONE
, level
);
2779 if (!pers
|| !try_module_get(pers
->owner
)) {
2780 spin_unlock(&pers_lock
);
2781 printk(KERN_WARNING
"md: personality %s not loaded\n", level
);
2784 spin_unlock(&pers_lock
);
2786 if (pers
== mddev
->pers
) {
2787 /* Nothing to do! */
2788 module_put(pers
->owner
);
2791 if (!pers
->takeover
) {
2792 module_put(pers
->owner
);
2793 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
2794 mdname(mddev
), level
);
2798 /* ->takeover must set new_* and/or delta_disks
2799 * if it succeeds, and may set them when it fails.
2801 priv
= pers
->takeover(mddev
);
2803 mddev
->new_level
= mddev
->level
;
2804 mddev
->new_layout
= mddev
->layout
;
2805 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2806 mddev
->raid_disks
-= mddev
->delta_disks
;
2807 mddev
->delta_disks
= 0;
2808 module_put(pers
->owner
);
2809 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
2810 mdname(mddev
), level
);
2811 return PTR_ERR(priv
);
2814 /* Looks like we have a winner */
2815 mddev_suspend(mddev
);
2816 mddev
->pers
->stop(mddev
);
2817 module_put(mddev
->pers
->owner
);
2818 /* Invalidate devices that are now superfluous */
2819 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2820 if (rdev
->raid_disk
>= mddev
->raid_disks
) {
2821 rdev
->raid_disk
= -1;
2822 clear_bit(In_sync
, &rdev
->flags
);
2825 mddev
->private = priv
;
2826 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
2827 mddev
->level
= mddev
->new_level
;
2828 mddev
->layout
= mddev
->new_layout
;
2829 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
2830 mddev
->delta_disks
= 0;
2832 mddev_resume(mddev
);
2833 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2834 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2835 md_wakeup_thread(mddev
->thread
);
2839 static struct md_sysfs_entry md_level
=
2840 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
2844 layout_show(mddev_t
*mddev
, char *page
)
2846 /* just a number, not meaningful for all levels */
2847 if (mddev
->reshape_position
!= MaxSector
&&
2848 mddev
->layout
!= mddev
->new_layout
)
2849 return sprintf(page
, "%d (%d)\n",
2850 mddev
->new_layout
, mddev
->layout
);
2851 return sprintf(page
, "%d\n", mddev
->layout
);
2855 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2858 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2860 if (!*buf
|| (*e
&& *e
!= '\n'))
2865 if (mddev
->pers
->check_reshape
== NULL
)
2867 mddev
->new_layout
= n
;
2868 err
= mddev
->pers
->check_reshape(mddev
);
2870 mddev
->new_layout
= mddev
->layout
;
2874 mddev
->new_layout
= n
;
2875 if (mddev
->reshape_position
== MaxSector
)
2880 static struct md_sysfs_entry md_layout
=
2881 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
2885 raid_disks_show(mddev_t
*mddev
, char *page
)
2887 if (mddev
->raid_disks
== 0)
2889 if (mddev
->reshape_position
!= MaxSector
&&
2890 mddev
->delta_disks
!= 0)
2891 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
2892 mddev
->raid_disks
- mddev
->delta_disks
);
2893 return sprintf(page
, "%d\n", mddev
->raid_disks
);
2896 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
2899 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2903 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2905 if (!*buf
|| (*e
&& *e
!= '\n'))
2909 rv
= update_raid_disks(mddev
, n
);
2910 else if (mddev
->reshape_position
!= MaxSector
) {
2911 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
2912 mddev
->delta_disks
= n
- olddisks
;
2913 mddev
->raid_disks
= n
;
2915 mddev
->raid_disks
= n
;
2916 return rv
? rv
: len
;
2918 static struct md_sysfs_entry md_raid_disks
=
2919 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
2922 chunk_size_show(mddev_t
*mddev
, char *page
)
2924 if (mddev
->reshape_position
!= MaxSector
&&
2925 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
2926 return sprintf(page
, "%d (%d)\n",
2927 mddev
->new_chunk_sectors
<< 9,
2928 mddev
->chunk_sectors
<< 9);
2929 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
2933 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2936 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2938 if (!*buf
|| (*e
&& *e
!= '\n'))
2943 if (mddev
->pers
->check_reshape
== NULL
)
2945 mddev
->new_chunk_sectors
= n
>> 9;
2946 err
= mddev
->pers
->check_reshape(mddev
);
2948 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2952 mddev
->new_chunk_sectors
= n
>> 9;
2953 if (mddev
->reshape_position
== MaxSector
)
2954 mddev
->chunk_sectors
= n
>> 9;
2958 static struct md_sysfs_entry md_chunk_size
=
2959 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
2962 resync_start_show(mddev_t
*mddev
, char *page
)
2964 if (mddev
->recovery_cp
== MaxSector
)
2965 return sprintf(page
, "none\n");
2966 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
2970 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2973 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
2977 if (!*buf
|| (*e
&& *e
!= '\n'))
2980 mddev
->recovery_cp
= n
;
2983 static struct md_sysfs_entry md_resync_start
=
2984 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
2987 * The array state can be:
2990 * No devices, no size, no level
2991 * Equivalent to STOP_ARRAY ioctl
2993 * May have some settings, but array is not active
2994 * all IO results in error
2995 * When written, doesn't tear down array, but just stops it
2996 * suspended (not supported yet)
2997 * All IO requests will block. The array can be reconfigured.
2998 * Writing this, if accepted, will block until array is quiescent
3000 * no resync can happen. no superblocks get written.
3001 * write requests fail
3003 * like readonly, but behaves like 'clean' on a write request.
3005 * clean - no pending writes, but otherwise active.
3006 * When written to inactive array, starts without resync
3007 * If a write request arrives then
3008 * if metadata is known, mark 'dirty' and switch to 'active'.
3009 * if not known, block and switch to write-pending
3010 * If written to an active array that has pending writes, then fails.
3012 * fully active: IO and resync can be happening.
3013 * When written to inactive array, starts with resync
3016 * clean, but writes are blocked waiting for 'active' to be written.
3019 * like active, but no writes have been seen for a while (100msec).
3022 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3023 write_pending
, active_idle
, bad_word
};
3024 static char *array_states
[] = {
3025 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3026 "write-pending", "active-idle", NULL
};
3028 static int match_word(const char *word
, char **list
)
3031 for (n
=0; list
[n
]; n
++)
3032 if (cmd_match(word
, list
[n
]))
3038 array_state_show(mddev_t
*mddev
, char *page
)
3040 enum array_state st
= inactive
;
3053 else if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
3055 else if (mddev
->safemode
)
3061 if (list_empty(&mddev
->disks
) &&
3062 mddev
->raid_disks
== 0 &&
3063 mddev
->dev_sectors
== 0)
3068 return sprintf(page
, "%s\n", array_states
[st
]);
3071 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3072 static int do_md_run(mddev_t
* mddev
);
3073 static int restart_array(mddev_t
*mddev
);
3076 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3079 enum array_state st
= match_word(buf
, array_states
);
3084 /* stopping an active array */
3085 if (atomic_read(&mddev
->openers
) > 0)
3087 err
= do_md_stop(mddev
, 0, 0);
3090 /* stopping an active array */
3092 if (atomic_read(&mddev
->openers
) > 0)
3094 err
= do_md_stop(mddev
, 2, 0);
3096 err
= 0; /* already inactive */
3099 break; /* not supported yet */
3102 err
= do_md_stop(mddev
, 1, 0);
3105 set_disk_ro(mddev
->gendisk
, 1);
3106 err
= do_md_run(mddev
);
3112 err
= do_md_stop(mddev
, 1, 0);
3113 else if (mddev
->ro
== 1)
3114 err
= restart_array(mddev
);
3117 set_disk_ro(mddev
->gendisk
, 0);
3121 err
= do_md_run(mddev
);
3126 restart_array(mddev
);
3127 spin_lock_irq(&mddev
->write_lock
);
3128 if (atomic_read(&mddev
->writes_pending
) == 0) {
3129 if (mddev
->in_sync
== 0) {
3131 if (mddev
->safemode
== 1)
3132 mddev
->safemode
= 0;
3133 if (mddev
->persistent
)
3134 set_bit(MD_CHANGE_CLEAN
,
3140 spin_unlock_irq(&mddev
->write_lock
);
3146 restart_array(mddev
);
3147 if (mddev
->external
)
3148 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3149 wake_up(&mddev
->sb_wait
);
3153 set_disk_ro(mddev
->gendisk
, 0);
3154 err
= do_md_run(mddev
);
3159 /* these cannot be set */
3165 sysfs_notify_dirent(mddev
->sysfs_state
);
3169 static struct md_sysfs_entry md_array_state
=
3170 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3173 null_show(mddev_t
*mddev
, char *page
)
3179 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3181 /* buf must be %d:%d\n? giving major and minor numbers */
3182 /* The new device is added to the array.
3183 * If the array has a persistent superblock, we read the
3184 * superblock to initialise info and check validity.
3185 * Otherwise, only checking done is that in bind_rdev_to_array,
3186 * which mainly checks size.
3189 int major
= simple_strtoul(buf
, &e
, 10);
3195 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3197 minor
= simple_strtoul(e
+1, &e
, 10);
3198 if (*e
&& *e
!= '\n')
3200 dev
= MKDEV(major
, minor
);
3201 if (major
!= MAJOR(dev
) ||
3202 minor
!= MINOR(dev
))
3206 if (mddev
->persistent
) {
3207 rdev
= md_import_device(dev
, mddev
->major_version
,
3208 mddev
->minor_version
);
3209 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3210 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3211 mdk_rdev_t
, same_set
);
3212 err
= super_types
[mddev
->major_version
]
3213 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3217 } else if (mddev
->external
)
3218 rdev
= md_import_device(dev
, -2, -1);
3220 rdev
= md_import_device(dev
, -1, -1);
3223 return PTR_ERR(rdev
);
3224 err
= bind_rdev_to_array(rdev
, mddev
);
3228 return err
? err
: len
;
3231 static struct md_sysfs_entry md_new_device
=
3232 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3235 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3238 unsigned long chunk
, end_chunk
;
3242 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3244 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3245 if (buf
== end
) break;
3246 if (*end
== '-') { /* range */
3248 end_chunk
= simple_strtoul(buf
, &end
, 0);
3249 if (buf
== end
) break;
3251 if (*end
&& !isspace(*end
)) break;
3252 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3254 while (isspace(*buf
)) buf
++;
3256 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3261 static struct md_sysfs_entry md_bitmap
=
3262 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3265 size_show(mddev_t
*mddev
, char *page
)
3267 return sprintf(page
, "%llu\n",
3268 (unsigned long long)mddev
->dev_sectors
/ 2);
3271 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3274 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3276 /* If array is inactive, we can reduce the component size, but
3277 * not increase it (except from 0).
3278 * If array is active, we can try an on-line resize
3281 int err
= strict_blocks_to_sectors(buf
, §ors
);
3286 err
= update_size(mddev
, sectors
);
3287 md_update_sb(mddev
, 1);
3289 if (mddev
->dev_sectors
== 0 ||
3290 mddev
->dev_sectors
> sectors
)
3291 mddev
->dev_sectors
= sectors
;
3295 return err
? err
: len
;
3298 static struct md_sysfs_entry md_size
=
3299 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3304 * 'none' for arrays with no metadata (good luck...)
3305 * 'external' for arrays with externally managed metadata,
3306 * or N.M for internally known formats
3309 metadata_show(mddev_t
*mddev
, char *page
)
3311 if (mddev
->persistent
)
3312 return sprintf(page
, "%d.%d\n",
3313 mddev
->major_version
, mddev
->minor_version
);
3314 else if (mddev
->external
)
3315 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3317 return sprintf(page
, "none\n");
3321 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3325 /* Changing the details of 'external' metadata is
3326 * always permitted. Otherwise there must be
3327 * no devices attached to the array.
3329 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3331 else if (!list_empty(&mddev
->disks
))
3334 if (cmd_match(buf
, "none")) {
3335 mddev
->persistent
= 0;
3336 mddev
->external
= 0;
3337 mddev
->major_version
= 0;
3338 mddev
->minor_version
= 90;
3341 if (strncmp(buf
, "external:", 9) == 0) {
3342 size_t namelen
= len
-9;
3343 if (namelen
>= sizeof(mddev
->metadata_type
))
3344 namelen
= sizeof(mddev
->metadata_type
)-1;
3345 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3346 mddev
->metadata_type
[namelen
] = 0;
3347 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3348 mddev
->metadata_type
[--namelen
] = 0;
3349 mddev
->persistent
= 0;
3350 mddev
->external
= 1;
3351 mddev
->major_version
= 0;
3352 mddev
->minor_version
= 90;
3355 major
= simple_strtoul(buf
, &e
, 10);
3356 if (e
==buf
|| *e
!= '.')
3359 minor
= simple_strtoul(buf
, &e
, 10);
3360 if (e
==buf
|| (*e
&& *e
!= '\n') )
3362 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3364 mddev
->major_version
= major
;
3365 mddev
->minor_version
= minor
;
3366 mddev
->persistent
= 1;
3367 mddev
->external
= 0;
3371 static struct md_sysfs_entry md_metadata
=
3372 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3375 action_show(mddev_t
*mddev
, char *page
)
3377 char *type
= "idle";
3378 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3380 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3381 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3382 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3384 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3385 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3387 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3391 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3394 return sprintf(page
, "%s\n", type
);
3398 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3400 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3403 if (cmd_match(page
, "frozen"))
3404 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3406 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3408 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3409 if (mddev
->sync_thread
) {
3410 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3411 md_unregister_thread(mddev
->sync_thread
);
3412 mddev
->sync_thread
= NULL
;
3413 mddev
->recovery
= 0;
3415 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3416 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3418 else if (cmd_match(page
, "resync"))
3419 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3420 else if (cmd_match(page
, "recover")) {
3421 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3422 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3423 } else if (cmd_match(page
, "reshape")) {
3425 if (mddev
->pers
->start_reshape
== NULL
)
3427 err
= mddev
->pers
->start_reshape(mddev
);
3430 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3432 if (cmd_match(page
, "check"))
3433 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3434 else if (!cmd_match(page
, "repair"))
3436 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3437 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3439 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3440 md_wakeup_thread(mddev
->thread
);
3441 sysfs_notify_dirent(mddev
->sysfs_action
);
3446 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3448 return sprintf(page
, "%llu\n",
3449 (unsigned long long) mddev
->resync_mismatches
);
3452 static struct md_sysfs_entry md_scan_mode
=
3453 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3456 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3459 sync_min_show(mddev_t
*mddev
, char *page
)
3461 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3462 mddev
->sync_speed_min
? "local": "system");
3466 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3470 if (strncmp(buf
, "system", 6)==0) {
3471 mddev
->sync_speed_min
= 0;
3474 min
= simple_strtoul(buf
, &e
, 10);
3475 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3477 mddev
->sync_speed_min
= min
;
3481 static struct md_sysfs_entry md_sync_min
=
3482 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3485 sync_max_show(mddev_t
*mddev
, char *page
)
3487 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3488 mddev
->sync_speed_max
? "local": "system");
3492 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3496 if (strncmp(buf
, "system", 6)==0) {
3497 mddev
->sync_speed_max
= 0;
3500 max
= simple_strtoul(buf
, &e
, 10);
3501 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3503 mddev
->sync_speed_max
= max
;
3507 static struct md_sysfs_entry md_sync_max
=
3508 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3511 degraded_show(mddev_t
*mddev
, char *page
)
3513 return sprintf(page
, "%d\n", mddev
->degraded
);
3515 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3518 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3520 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3524 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3528 if (strict_strtol(buf
, 10, &n
))
3531 if (n
!= 0 && n
!= 1)
3534 mddev
->parallel_resync
= n
;
3536 if (mddev
->sync_thread
)
3537 wake_up(&resync_wait
);
3542 /* force parallel resync, even with shared block devices */
3543 static struct md_sysfs_entry md_sync_force_parallel
=
3544 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3545 sync_force_parallel_show
, sync_force_parallel_store
);
3548 sync_speed_show(mddev_t
*mddev
, char *page
)
3550 unsigned long resync
, dt
, db
;
3551 if (mddev
->curr_resync
== 0)
3552 return sprintf(page
, "none\n");
3553 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3554 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3556 db
= resync
- mddev
->resync_mark_cnt
;
3557 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3560 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3563 sync_completed_show(mddev_t
*mddev
, char *page
)
3565 unsigned long max_sectors
, resync
;
3567 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3568 return sprintf(page
, "none\n");
3570 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3571 max_sectors
= mddev
->resync_max_sectors
;
3573 max_sectors
= mddev
->dev_sectors
;
3575 resync
= mddev
->curr_resync_completed
;
3576 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3579 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3582 min_sync_show(mddev_t
*mddev
, char *page
)
3584 return sprintf(page
, "%llu\n",
3585 (unsigned long long)mddev
->resync_min
);
3588 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3590 unsigned long long min
;
3591 if (strict_strtoull(buf
, 10, &min
))
3593 if (min
> mddev
->resync_max
)
3595 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3598 /* Must be a multiple of chunk_size */
3599 if (mddev
->chunk_sectors
) {
3600 sector_t temp
= min
;
3601 if (sector_div(temp
, mddev
->chunk_sectors
))
3604 mddev
->resync_min
= min
;
3609 static struct md_sysfs_entry md_min_sync
=
3610 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3613 max_sync_show(mddev_t
*mddev
, char *page
)
3615 if (mddev
->resync_max
== MaxSector
)
3616 return sprintf(page
, "max\n");
3618 return sprintf(page
, "%llu\n",
3619 (unsigned long long)mddev
->resync_max
);
3622 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3624 if (strncmp(buf
, "max", 3) == 0)
3625 mddev
->resync_max
= MaxSector
;
3627 unsigned long long max
;
3628 if (strict_strtoull(buf
, 10, &max
))
3630 if (max
< mddev
->resync_min
)
3632 if (max
< mddev
->resync_max
&&
3634 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3637 /* Must be a multiple of chunk_size */
3638 if (mddev
->chunk_sectors
) {
3639 sector_t temp
= max
;
3640 if (sector_div(temp
, mddev
->chunk_sectors
))
3643 mddev
->resync_max
= max
;
3645 wake_up(&mddev
->recovery_wait
);
3649 static struct md_sysfs_entry md_max_sync
=
3650 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3653 suspend_lo_show(mddev_t
*mddev
, char *page
)
3655 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
3659 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3662 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3664 if (mddev
->pers
== NULL
||
3665 mddev
->pers
->quiesce
== NULL
)
3667 if (buf
== e
|| (*e
&& *e
!= '\n'))
3669 if (new >= mddev
->suspend_hi
||
3670 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
3671 mddev
->suspend_lo
= new;
3672 mddev
->pers
->quiesce(mddev
, 2);
3677 static struct md_sysfs_entry md_suspend_lo
=
3678 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
3682 suspend_hi_show(mddev_t
*mddev
, char *page
)
3684 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
3688 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3691 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3693 if (mddev
->pers
== NULL
||
3694 mddev
->pers
->quiesce
== NULL
)
3696 if (buf
== e
|| (*e
&& *e
!= '\n'))
3698 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
3699 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
3700 mddev
->suspend_hi
= new;
3701 mddev
->pers
->quiesce(mddev
, 1);
3702 mddev
->pers
->quiesce(mddev
, 0);
3707 static struct md_sysfs_entry md_suspend_hi
=
3708 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
3711 reshape_position_show(mddev_t
*mddev
, char *page
)
3713 if (mddev
->reshape_position
!= MaxSector
)
3714 return sprintf(page
, "%llu\n",
3715 (unsigned long long)mddev
->reshape_position
);
3716 strcpy(page
, "none\n");
3721 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3724 unsigned long long new = simple_strtoull(buf
, &e
, 10);
3727 if (buf
== e
|| (*e
&& *e
!= '\n'))
3729 mddev
->reshape_position
= new;
3730 mddev
->delta_disks
= 0;
3731 mddev
->new_level
= mddev
->level
;
3732 mddev
->new_layout
= mddev
->layout
;
3733 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3737 static struct md_sysfs_entry md_reshape_position
=
3738 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
3739 reshape_position_store
);
3742 array_size_show(mddev_t
*mddev
, char *page
)
3744 if (mddev
->external_size
)
3745 return sprintf(page
, "%llu\n",
3746 (unsigned long long)mddev
->array_sectors
/2);
3748 return sprintf(page
, "default\n");
3752 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3756 if (strncmp(buf
, "default", 7) == 0) {
3758 sectors
= mddev
->pers
->size(mddev
, 0, 0);
3760 sectors
= mddev
->array_sectors
;
3762 mddev
->external_size
= 0;
3764 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3766 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
3769 mddev
->external_size
= 1;
3772 mddev
->array_sectors
= sectors
;
3773 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
3775 revalidate_disk(mddev
->gendisk
);
3780 static struct md_sysfs_entry md_array_size
=
3781 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
3784 static struct attribute
*md_default_attrs
[] = {
3787 &md_raid_disks
.attr
,
3788 &md_chunk_size
.attr
,
3790 &md_resync_start
.attr
,
3792 &md_new_device
.attr
,
3793 &md_safe_delay
.attr
,
3794 &md_array_state
.attr
,
3795 &md_reshape_position
.attr
,
3796 &md_array_size
.attr
,
3800 static struct attribute
*md_redundancy_attrs
[] = {
3802 &md_mismatches
.attr
,
3805 &md_sync_speed
.attr
,
3806 &md_sync_force_parallel
.attr
,
3807 &md_sync_completed
.attr
,
3810 &md_suspend_lo
.attr
,
3811 &md_suspend_hi
.attr
,
3816 static struct attribute_group md_redundancy_group
= {
3818 .attrs
= md_redundancy_attrs
,
3823 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3825 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3826 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3831 rv
= mddev_lock(mddev
);
3833 rv
= entry
->show(mddev
, page
);
3834 mddev_unlock(mddev
);
3840 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3841 const char *page
, size_t length
)
3843 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
3844 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
3849 if (!capable(CAP_SYS_ADMIN
))
3851 rv
= mddev_lock(mddev
);
3852 if (mddev
->hold_active
== UNTIL_IOCTL
)
3853 mddev
->hold_active
= 0;
3855 rv
= entry
->store(mddev
, page
, length
);
3856 mddev_unlock(mddev
);
3861 static void md_free(struct kobject
*ko
)
3863 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
3865 if (mddev
->sysfs_state
)
3866 sysfs_put(mddev
->sysfs_state
);
3868 if (mddev
->gendisk
) {
3869 del_gendisk(mddev
->gendisk
);
3870 put_disk(mddev
->gendisk
);
3873 blk_cleanup_queue(mddev
->queue
);
3878 static struct sysfs_ops md_sysfs_ops
= {
3879 .show
= md_attr_show
,
3880 .store
= md_attr_store
,
3882 static struct kobj_type md_ktype
= {
3884 .sysfs_ops
= &md_sysfs_ops
,
3885 .default_attrs
= md_default_attrs
,
3890 static void mddev_delayed_delete(struct work_struct
*ws
)
3892 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
3894 if (mddev
->private == &md_redundancy_group
) {
3895 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
3896 if (mddev
->sysfs_action
)
3897 sysfs_put(mddev
->sysfs_action
);
3898 mddev
->sysfs_action
= NULL
;
3899 mddev
->private = NULL
;
3901 kobject_del(&mddev
->kobj
);
3902 kobject_put(&mddev
->kobj
);
3905 static int md_alloc(dev_t dev
, char *name
)
3907 static DEFINE_MUTEX(disks_mutex
);
3908 mddev_t
*mddev
= mddev_find(dev
);
3909 struct gendisk
*disk
;
3918 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
3919 shift
= partitioned
? MdpMinorShift
: 0;
3920 unit
= MINOR(mddev
->unit
) >> shift
;
3922 /* wait for any previous instance if this device
3923 * to be completed removed (mddev_delayed_delete).
3925 flush_scheduled_work();
3927 mutex_lock(&disks_mutex
);
3933 /* Need to ensure that 'name' is not a duplicate.
3936 spin_lock(&all_mddevs_lock
);
3938 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
3939 if (mddev2
->gendisk
&&
3940 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
3941 spin_unlock(&all_mddevs_lock
);
3944 spin_unlock(&all_mddevs_lock
);
3948 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
3951 mddev
->queue
->queuedata
= mddev
;
3953 /* Can be unlocked because the queue is new: no concurrency */
3954 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
3956 blk_queue_make_request(mddev
->queue
, md_make_request
);
3958 disk
= alloc_disk(1 << shift
);
3960 blk_cleanup_queue(mddev
->queue
);
3961 mddev
->queue
= NULL
;
3964 disk
->major
= MAJOR(mddev
->unit
);
3965 disk
->first_minor
= unit
<< shift
;
3967 strcpy(disk
->disk_name
, name
);
3968 else if (partitioned
)
3969 sprintf(disk
->disk_name
, "md_d%d", unit
);
3971 sprintf(disk
->disk_name
, "md%d", unit
);
3972 disk
->fops
= &md_fops
;
3973 disk
->private_data
= mddev
;
3974 disk
->queue
= mddev
->queue
;
3975 /* Allow extended partitions. This makes the
3976 * 'mdp' device redundant, but we can't really
3979 disk
->flags
|= GENHD_FL_EXT_DEVT
;
3981 mddev
->gendisk
= disk
;
3982 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
3983 &disk_to_dev(disk
)->kobj
, "%s", "md");
3985 /* This isn't possible, but as kobject_init_and_add is marked
3986 * __must_check, we must do something with the result
3988 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
3993 mutex_unlock(&disks_mutex
);
3995 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
3996 mddev
->sysfs_state
= sysfs_get_dirent(mddev
->kobj
.sd
, "array_state");
4002 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4004 md_alloc(dev
, NULL
);
4008 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4010 /* val must be "md_*" where * is not all digits.
4011 * We allocate an array with a large free minor number, and
4012 * set the name to val. val must not already be an active name.
4014 int len
= strlen(val
);
4015 char buf
[DISK_NAME_LEN
];
4017 while (len
&& val
[len
-1] == '\n')
4019 if (len
>= DISK_NAME_LEN
)
4021 strlcpy(buf
, val
, len
+1);
4022 if (strncmp(buf
, "md_", 3) != 0)
4024 return md_alloc(0, buf
);
4027 static void md_safemode_timeout(unsigned long data
)
4029 mddev_t
*mddev
= (mddev_t
*) data
;
4031 if (!atomic_read(&mddev
->writes_pending
)) {
4032 mddev
->safemode
= 1;
4033 if (mddev
->external
)
4034 sysfs_notify_dirent(mddev
->sysfs_state
);
4036 md_wakeup_thread(mddev
->thread
);
4039 static int start_dirty_degraded
;
4041 static int do_md_run(mddev_t
* mddev
)
4045 struct gendisk
*disk
;
4046 struct mdk_personality
*pers
;
4048 if (list_empty(&mddev
->disks
))
4049 /* cannot run an array with no devices.. */
4056 * Analyze all RAID superblock(s)
4058 if (!mddev
->raid_disks
) {
4059 if (!mddev
->persistent
)
4064 if (mddev
->level
!= LEVEL_NONE
)
4065 request_module("md-level-%d", mddev
->level
);
4066 else if (mddev
->clevel
[0])
4067 request_module("md-%s", mddev
->clevel
);
4070 * Drop all container device buffers, from now on
4071 * the only valid external interface is through the md
4074 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4075 if (test_bit(Faulty
, &rdev
->flags
))
4077 sync_blockdev(rdev
->bdev
);
4078 invalidate_bdev(rdev
->bdev
);
4080 /* perform some consistency tests on the device.
4081 * We don't want the data to overlap the metadata,
4082 * Internal Bitmap issues have been handled elsewhere.
4084 if (rdev
->data_offset
< rdev
->sb_start
) {
4085 if (mddev
->dev_sectors
&&
4086 rdev
->data_offset
+ mddev
->dev_sectors
4088 printk("md: %s: data overlaps metadata\n",
4093 if (rdev
->sb_start
+ rdev
->sb_size
/512
4094 > rdev
->data_offset
) {
4095 printk("md: %s: metadata overlaps data\n",
4100 sysfs_notify_dirent(rdev
->sysfs_state
);
4103 md_probe(mddev
->unit
, NULL
, NULL
);
4104 disk
= mddev
->gendisk
;
4108 spin_lock(&pers_lock
);
4109 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4110 if (!pers
|| !try_module_get(pers
->owner
)) {
4111 spin_unlock(&pers_lock
);
4112 if (mddev
->level
!= LEVEL_NONE
)
4113 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4116 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4121 spin_unlock(&pers_lock
);
4122 if (mddev
->level
!= pers
->level
) {
4123 mddev
->level
= pers
->level
;
4124 mddev
->new_level
= pers
->level
;
4126 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4128 if (mddev
->reshape_position
!= MaxSector
&&
4129 pers
->start_reshape
== NULL
) {
4130 /* This personality cannot handle reshaping... */
4132 module_put(pers
->owner
);
4136 if (pers
->sync_request
) {
4137 /* Warn if this is a potentially silly
4140 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4144 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4145 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4147 rdev
->bdev
->bd_contains
==
4148 rdev2
->bdev
->bd_contains
) {
4150 "%s: WARNING: %s appears to be"
4151 " on the same physical disk as"
4154 bdevname(rdev
->bdev
,b
),
4155 bdevname(rdev2
->bdev
,b2
));
4162 "True protection against single-disk"
4163 " failure might be compromised.\n");
4166 mddev
->recovery
= 0;
4167 /* may be over-ridden by personality */
4168 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4170 mddev
->barriers_work
= 1;
4171 mddev
->ok_start_degraded
= start_dirty_degraded
;
4174 mddev
->ro
= 2; /* read-only, but switch on first write */
4176 err
= mddev
->pers
->run(mddev
);
4178 printk(KERN_ERR
"md: pers->run() failed ...\n");
4179 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4180 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4181 " but 'external_size' not in effect?\n", __func__
);
4183 "md: invalid array_size %llu > default size %llu\n",
4184 (unsigned long long)mddev
->array_sectors
/ 2,
4185 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4187 mddev
->pers
->stop(mddev
);
4189 if (err
== 0 && mddev
->pers
->sync_request
) {
4190 err
= bitmap_create(mddev
);
4192 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4193 mdname(mddev
), err
);
4194 mddev
->pers
->stop(mddev
);
4198 module_put(mddev
->pers
->owner
);
4200 bitmap_destroy(mddev
);
4203 if (mddev
->pers
->sync_request
) {
4204 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4206 "md: cannot register extra attributes for %s\n",
4208 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
4209 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4212 atomic_set(&mddev
->writes_pending
,0);
4213 mddev
->safemode
= 0;
4214 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4215 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4216 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4219 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4220 if (rdev
->raid_disk
>= 0) {
4222 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4223 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4224 printk("md: cannot register %s for %s\n",
4228 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4231 md_update_sb(mddev
, 0);
4233 set_capacity(disk
, mddev
->array_sectors
);
4235 /* If there is a partially-recovered drive we need to
4236 * start recovery here. If we leave it to md_check_recovery,
4237 * it will remove the drives and not do the right thing
4239 if (mddev
->degraded
&& !mddev
->sync_thread
) {
4241 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4242 if (rdev
->raid_disk
>= 0 &&
4243 !test_bit(In_sync
, &rdev
->flags
) &&
4244 !test_bit(Faulty
, &rdev
->flags
))
4245 /* complete an interrupted recovery */
4247 if (spares
&& mddev
->pers
->sync_request
) {
4248 mddev
->recovery
= 0;
4249 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
4250 mddev
->sync_thread
= md_register_thread(md_do_sync
,
4253 if (!mddev
->sync_thread
) {
4254 printk(KERN_ERR
"%s: could not start resync"
4257 /* leave the spares where they are, it shouldn't hurt */
4258 mddev
->recovery
= 0;
4262 md_wakeup_thread(mddev
->thread
);
4263 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4265 revalidate_disk(mddev
->gendisk
);
4267 md_new_event(mddev
);
4268 sysfs_notify_dirent(mddev
->sysfs_state
);
4269 if (mddev
->sysfs_action
)
4270 sysfs_notify_dirent(mddev
->sysfs_action
);
4271 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4272 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4276 static int restart_array(mddev_t
*mddev
)
4278 struct gendisk
*disk
= mddev
->gendisk
;
4280 /* Complain if it has no devices */
4281 if (list_empty(&mddev
->disks
))
4287 mddev
->safemode
= 0;
4289 set_disk_ro(disk
, 0);
4290 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4292 /* Kick recovery or resync if necessary */
4293 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4294 md_wakeup_thread(mddev
->thread
);
4295 md_wakeup_thread(mddev
->sync_thread
);
4296 sysfs_notify_dirent(mddev
->sysfs_state
);
4300 /* similar to deny_write_access, but accounts for our holding a reference
4301 * to the file ourselves */
4302 static int deny_bitmap_write_access(struct file
* file
)
4304 struct inode
*inode
= file
->f_mapping
->host
;
4306 spin_lock(&inode
->i_lock
);
4307 if (atomic_read(&inode
->i_writecount
) > 1) {
4308 spin_unlock(&inode
->i_lock
);
4311 atomic_set(&inode
->i_writecount
, -1);
4312 spin_unlock(&inode
->i_lock
);
4317 static void restore_bitmap_write_access(struct file
*file
)
4319 struct inode
*inode
= file
->f_mapping
->host
;
4321 spin_lock(&inode
->i_lock
);
4322 atomic_set(&inode
->i_writecount
, 1);
4323 spin_unlock(&inode
->i_lock
);
4327 * 0 - completely stop and dis-assemble array
4328 * 1 - switch to readonly
4329 * 2 - stop but do not disassemble array
4331 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4334 struct gendisk
*disk
= mddev
->gendisk
;
4337 mutex_lock(&mddev
->open_mutex
);
4338 if (atomic_read(&mddev
->openers
) > is_open
) {
4339 printk("md: %s still in use.\n",mdname(mddev
));
4341 } else if (mddev
->pers
) {
4343 if (mddev
->sync_thread
) {
4344 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4345 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4346 md_unregister_thread(mddev
->sync_thread
);
4347 mddev
->sync_thread
= NULL
;
4350 del_timer_sync(&mddev
->safemode_timer
);
4353 case 1: /* readonly */
4359 case 0: /* disassemble */
4361 bitmap_flush(mddev
);
4362 md_super_wait(mddev
);
4364 set_disk_ro(disk
, 0);
4366 mddev
->pers
->stop(mddev
);
4367 mddev
->queue
->merge_bvec_fn
= NULL
;
4368 mddev
->queue
->unplug_fn
= NULL
;
4369 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4370 module_put(mddev
->pers
->owner
);
4371 if (mddev
->pers
->sync_request
)
4372 mddev
->private = &md_redundancy_group
;
4374 /* tell userspace to handle 'inactive' */
4375 sysfs_notify_dirent(mddev
->sysfs_state
);
4377 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4378 if (rdev
->raid_disk
>= 0) {
4380 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4381 sysfs_remove_link(&mddev
->kobj
, nm
);
4384 set_capacity(disk
, 0);
4390 if (!mddev
->in_sync
|| mddev
->flags
) {
4391 /* mark array as shutdown cleanly */
4393 md_update_sb(mddev
, 1);
4396 set_disk_ro(disk
, 1);
4397 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4401 mutex_unlock(&mddev
->open_mutex
);
4405 * Free resources if final stop
4409 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4411 bitmap_destroy(mddev
);
4412 if (mddev
->bitmap_file
) {
4413 restore_bitmap_write_access(mddev
->bitmap_file
);
4414 fput(mddev
->bitmap_file
);
4415 mddev
->bitmap_file
= NULL
;
4417 mddev
->bitmap_offset
= 0;
4419 /* make sure all md_delayed_delete calls have finished */
4420 flush_scheduled_work();
4422 export_array(mddev
);
4424 mddev
->array_sectors
= 0;
4425 mddev
->external_size
= 0;
4426 mddev
->dev_sectors
= 0;
4427 mddev
->raid_disks
= 0;
4428 mddev
->recovery_cp
= 0;
4429 mddev
->resync_min
= 0;
4430 mddev
->resync_max
= MaxSector
;
4431 mddev
->reshape_position
= MaxSector
;
4432 mddev
->external
= 0;
4433 mddev
->persistent
= 0;
4434 mddev
->level
= LEVEL_NONE
;
4435 mddev
->clevel
[0] = 0;
4438 mddev
->metadata_type
[0] = 0;
4439 mddev
->chunk_sectors
= 0;
4440 mddev
->ctime
= mddev
->utime
= 0;
4442 mddev
->max_disks
= 0;
4444 mddev
->delta_disks
= 0;
4445 mddev
->new_level
= LEVEL_NONE
;
4446 mddev
->new_layout
= 0;
4447 mddev
->new_chunk_sectors
= 0;
4448 mddev
->curr_resync
= 0;
4449 mddev
->resync_mismatches
= 0;
4450 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4451 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4452 mddev
->recovery
= 0;
4455 mddev
->degraded
= 0;
4456 mddev
->barriers_work
= 0;
4457 mddev
->safemode
= 0;
4458 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4459 if (mddev
->hold_active
== UNTIL_STOP
)
4460 mddev
->hold_active
= 0;
4462 } else if (mddev
->pers
)
4463 printk(KERN_INFO
"md: %s switched to read-only mode.\n",
4466 blk_integrity_unregister(disk
);
4467 md_new_event(mddev
);
4468 sysfs_notify_dirent(mddev
->sysfs_state
);
4473 static void autorun_array(mddev_t
*mddev
)
4478 if (list_empty(&mddev
->disks
))
4481 printk(KERN_INFO
"md: running: ");
4483 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4484 char b
[BDEVNAME_SIZE
];
4485 printk("<%s>", bdevname(rdev
->bdev
,b
));
4489 err
= do_md_run(mddev
);
4491 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4492 do_md_stop(mddev
, 0, 0);
4497 * lets try to run arrays based on all disks that have arrived
4498 * until now. (those are in pending_raid_disks)
4500 * the method: pick the first pending disk, collect all disks with
4501 * the same UUID, remove all from the pending list and put them into
4502 * the 'same_array' list. Then order this list based on superblock
4503 * update time (freshest comes first), kick out 'old' disks and
4504 * compare superblocks. If everything's fine then run it.
4506 * If "unit" is allocated, then bump its reference count
4508 static void autorun_devices(int part
)
4510 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4512 char b
[BDEVNAME_SIZE
];
4514 printk(KERN_INFO
"md: autorun ...\n");
4515 while (!list_empty(&pending_raid_disks
)) {
4518 LIST_HEAD(candidates
);
4519 rdev0
= list_entry(pending_raid_disks
.next
,
4520 mdk_rdev_t
, same_set
);
4522 printk(KERN_INFO
"md: considering %s ...\n",
4523 bdevname(rdev0
->bdev
,b
));
4524 INIT_LIST_HEAD(&candidates
);
4525 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4526 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4527 printk(KERN_INFO
"md: adding %s ...\n",
4528 bdevname(rdev
->bdev
,b
));
4529 list_move(&rdev
->same_set
, &candidates
);
4532 * now we have a set of devices, with all of them having
4533 * mostly sane superblocks. It's time to allocate the
4537 dev
= MKDEV(mdp_major
,
4538 rdev0
->preferred_minor
<< MdpMinorShift
);
4539 unit
= MINOR(dev
) >> MdpMinorShift
;
4541 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4544 if (rdev0
->preferred_minor
!= unit
) {
4545 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4546 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4550 md_probe(dev
, NULL
, NULL
);
4551 mddev
= mddev_find(dev
);
4552 if (!mddev
|| !mddev
->gendisk
) {
4556 "md: cannot allocate memory for md drive.\n");
4559 if (mddev_lock(mddev
))
4560 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4562 else if (mddev
->raid_disks
|| mddev
->major_version
4563 || !list_empty(&mddev
->disks
)) {
4565 "md: %s already running, cannot run %s\n",
4566 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4567 mddev_unlock(mddev
);
4569 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4570 mddev
->persistent
= 1;
4571 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4572 list_del_init(&rdev
->same_set
);
4573 if (bind_rdev_to_array(rdev
, mddev
))
4576 autorun_array(mddev
);
4577 mddev_unlock(mddev
);
4579 /* on success, candidates will be empty, on error
4582 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4583 list_del_init(&rdev
->same_set
);
4588 printk(KERN_INFO
"md: ... autorun DONE.\n");
4590 #endif /* !MODULE */
4592 static int get_version(void __user
* arg
)
4596 ver
.major
= MD_MAJOR_VERSION
;
4597 ver
.minor
= MD_MINOR_VERSION
;
4598 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4600 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4606 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4608 mdu_array_info_t info
;
4609 int nr
,working
,insync
,failed
,spare
;
4612 nr
=working
=insync
=failed
=spare
=0;
4613 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4615 if (test_bit(Faulty
, &rdev
->flags
))
4619 if (test_bit(In_sync
, &rdev
->flags
))
4626 info
.major_version
= mddev
->major_version
;
4627 info
.minor_version
= mddev
->minor_version
;
4628 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4629 info
.ctime
= mddev
->ctime
;
4630 info
.level
= mddev
->level
;
4631 info
.size
= mddev
->dev_sectors
/ 2;
4632 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4635 info
.raid_disks
= mddev
->raid_disks
;
4636 info
.md_minor
= mddev
->md_minor
;
4637 info
.not_persistent
= !mddev
->persistent
;
4639 info
.utime
= mddev
->utime
;
4642 info
.state
= (1<<MD_SB_CLEAN
);
4643 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
4644 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
4645 info
.active_disks
= insync
;
4646 info
.working_disks
= working
;
4647 info
.failed_disks
= failed
;
4648 info
.spare_disks
= spare
;
4650 info
.layout
= mddev
->layout
;
4651 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
4653 if (copy_to_user(arg
, &info
, sizeof(info
)))
4659 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
4661 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
4662 char *ptr
, *buf
= NULL
;
4665 if (md_allow_write(mddev
))
4666 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
4668 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
4673 /* bitmap disabled, zero the first byte and copy out */
4674 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
4675 file
->pathname
[0] = '\0';
4679 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
4683 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
4687 strcpy(file
->pathname
, ptr
);
4691 if (copy_to_user(arg
, file
, sizeof(*file
)))
4699 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
4701 mdu_disk_info_t info
;
4704 if (copy_from_user(&info
, arg
, sizeof(info
)))
4707 rdev
= find_rdev_nr(mddev
, info
.number
);
4709 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
4710 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
4711 info
.raid_disk
= rdev
->raid_disk
;
4713 if (test_bit(Faulty
, &rdev
->flags
))
4714 info
.state
|= (1<<MD_DISK_FAULTY
);
4715 else if (test_bit(In_sync
, &rdev
->flags
)) {
4716 info
.state
|= (1<<MD_DISK_ACTIVE
);
4717 info
.state
|= (1<<MD_DISK_SYNC
);
4719 if (test_bit(WriteMostly
, &rdev
->flags
))
4720 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
4722 info
.major
= info
.minor
= 0;
4723 info
.raid_disk
= -1;
4724 info
.state
= (1<<MD_DISK_REMOVED
);
4727 if (copy_to_user(arg
, &info
, sizeof(info
)))
4733 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
4735 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4737 dev_t dev
= MKDEV(info
->major
,info
->minor
);
4739 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
4742 if (!mddev
->raid_disks
) {
4744 /* expecting a device which has a superblock */
4745 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
4748 "md: md_import_device returned %ld\n",
4750 return PTR_ERR(rdev
);
4752 if (!list_empty(&mddev
->disks
)) {
4753 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
4754 mdk_rdev_t
, same_set
);
4755 err
= super_types
[mddev
->major_version
]
4756 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4759 "md: %s has different UUID to %s\n",
4760 bdevname(rdev
->bdev
,b
),
4761 bdevname(rdev0
->bdev
,b2
));
4766 err
= bind_rdev_to_array(rdev
, mddev
);
4773 * add_new_disk can be used once the array is assembled
4774 * to add "hot spares". They must already have a superblock
4779 if (!mddev
->pers
->hot_add_disk
) {
4781 "%s: personality does not support diskops!\n",
4785 if (mddev
->persistent
)
4786 rdev
= md_import_device(dev
, mddev
->major_version
,
4787 mddev
->minor_version
);
4789 rdev
= md_import_device(dev
, -1, -1);
4792 "md: md_import_device returned %ld\n",
4794 return PTR_ERR(rdev
);
4796 /* set save_raid_disk if appropriate */
4797 if (!mddev
->persistent
) {
4798 if (info
->state
& (1<<MD_DISK_SYNC
) &&
4799 info
->raid_disk
< mddev
->raid_disks
)
4800 rdev
->raid_disk
= info
->raid_disk
;
4802 rdev
->raid_disk
= -1;
4804 super_types
[mddev
->major_version
].
4805 validate_super(mddev
, rdev
);
4806 rdev
->saved_raid_disk
= rdev
->raid_disk
;
4808 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
4809 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4810 set_bit(WriteMostly
, &rdev
->flags
);
4812 clear_bit(WriteMostly
, &rdev
->flags
);
4814 rdev
->raid_disk
= -1;
4815 err
= bind_rdev_to_array(rdev
, mddev
);
4816 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
4817 /* If there is hot_add_disk but no hot_remove_disk
4818 * then added disks for geometry changes,
4819 * and should be added immediately.
4821 super_types
[mddev
->major_version
].
4822 validate_super(mddev
, rdev
);
4823 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
4825 unbind_rdev_from_array(rdev
);
4830 sysfs_notify_dirent(rdev
->sysfs_state
);
4832 md_update_sb(mddev
, 1);
4833 if (mddev
->degraded
)
4834 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4835 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4836 md_wakeup_thread(mddev
->thread
);
4840 /* otherwise, add_new_disk is only allowed
4841 * for major_version==0 superblocks
4843 if (mddev
->major_version
!= 0) {
4844 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
4849 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
4851 rdev
= md_import_device(dev
, -1, 0);
4854 "md: error, md_import_device() returned %ld\n",
4856 return PTR_ERR(rdev
);
4858 rdev
->desc_nr
= info
->number
;
4859 if (info
->raid_disk
< mddev
->raid_disks
)
4860 rdev
->raid_disk
= info
->raid_disk
;
4862 rdev
->raid_disk
= -1;
4864 if (rdev
->raid_disk
< mddev
->raid_disks
)
4865 if (info
->state
& (1<<MD_DISK_SYNC
))
4866 set_bit(In_sync
, &rdev
->flags
);
4868 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
4869 set_bit(WriteMostly
, &rdev
->flags
);
4871 if (!mddev
->persistent
) {
4872 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
4873 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
4875 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
4876 rdev
->sectors
= rdev
->sb_start
;
4878 err
= bind_rdev_to_array(rdev
, mddev
);
4888 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
4890 char b
[BDEVNAME_SIZE
];
4893 rdev
= find_rdev(mddev
, dev
);
4897 if (rdev
->raid_disk
>= 0)
4900 kick_rdev_from_array(rdev
);
4901 md_update_sb(mddev
, 1);
4902 md_new_event(mddev
);
4906 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
4907 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4911 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
4913 char b
[BDEVNAME_SIZE
];
4920 if (mddev
->major_version
!= 0) {
4921 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
4922 " version-0 superblocks.\n",
4926 if (!mddev
->pers
->hot_add_disk
) {
4928 "%s: personality does not support diskops!\n",
4933 rdev
= md_import_device(dev
, -1, 0);
4936 "md: error, md_import_device() returned %ld\n",
4941 if (mddev
->persistent
)
4942 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
4944 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
4946 rdev
->sectors
= rdev
->sb_start
;
4948 if (test_bit(Faulty
, &rdev
->flags
)) {
4950 "md: can not hot-add faulty %s disk to %s!\n",
4951 bdevname(rdev
->bdev
,b
), mdname(mddev
));
4955 clear_bit(In_sync
, &rdev
->flags
);
4957 rdev
->saved_raid_disk
= -1;
4958 err
= bind_rdev_to_array(rdev
, mddev
);
4963 * The rest should better be atomic, we can have disk failures
4964 * noticed in interrupt contexts ...
4967 rdev
->raid_disk
= -1;
4969 md_update_sb(mddev
, 1);
4972 * Kick recovery, maybe this spare has to be added to the
4973 * array immediately.
4975 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4976 md_wakeup_thread(mddev
->thread
);
4977 md_new_event(mddev
);
4985 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
4990 if (!mddev
->pers
->quiesce
)
4992 if (mddev
->recovery
|| mddev
->sync_thread
)
4994 /* we should be able to change the bitmap.. */
5000 return -EEXIST
; /* cannot add when bitmap is present */
5001 mddev
->bitmap_file
= fget(fd
);
5003 if (mddev
->bitmap_file
== NULL
) {
5004 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5009 err
= deny_bitmap_write_access(mddev
->bitmap_file
);
5011 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5013 fput(mddev
->bitmap_file
);
5014 mddev
->bitmap_file
= NULL
;
5017 mddev
->bitmap_offset
= 0; /* file overrides offset */
5018 } else if (mddev
->bitmap
== NULL
)
5019 return -ENOENT
; /* cannot remove what isn't there */
5022 mddev
->pers
->quiesce(mddev
, 1);
5024 err
= bitmap_create(mddev
);
5025 if (fd
< 0 || err
) {
5026 bitmap_destroy(mddev
);
5027 fd
= -1; /* make sure to put the file */
5029 mddev
->pers
->quiesce(mddev
, 0);
5032 if (mddev
->bitmap_file
) {
5033 restore_bitmap_write_access(mddev
->bitmap_file
);
5034 fput(mddev
->bitmap_file
);
5036 mddev
->bitmap_file
= NULL
;
5043 * set_array_info is used two different ways
5044 * The original usage is when creating a new array.
5045 * In this usage, raid_disks is > 0 and it together with
5046 * level, size, not_persistent,layout,chunksize determine the
5047 * shape of the array.
5048 * This will always create an array with a type-0.90.0 superblock.
5049 * The newer usage is when assembling an array.
5050 * In this case raid_disks will be 0, and the major_version field is
5051 * use to determine which style super-blocks are to be found on the devices.
5052 * The minor and patch _version numbers are also kept incase the
5053 * super_block handler wishes to interpret them.
5055 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5058 if (info
->raid_disks
== 0) {
5059 /* just setting version number for superblock loading */
5060 if (info
->major_version
< 0 ||
5061 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5062 super_types
[info
->major_version
].name
== NULL
) {
5063 /* maybe try to auto-load a module? */
5065 "md: superblock version %d not known\n",
5066 info
->major_version
);
5069 mddev
->major_version
= info
->major_version
;
5070 mddev
->minor_version
= info
->minor_version
;
5071 mddev
->patch_version
= info
->patch_version
;
5072 mddev
->persistent
= !info
->not_persistent
;
5075 mddev
->major_version
= MD_MAJOR_VERSION
;
5076 mddev
->minor_version
= MD_MINOR_VERSION
;
5077 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5078 mddev
->ctime
= get_seconds();
5080 mddev
->level
= info
->level
;
5081 mddev
->clevel
[0] = 0;
5082 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5083 mddev
->raid_disks
= info
->raid_disks
;
5084 /* don't set md_minor, it is determined by which /dev/md* was
5087 if (info
->state
& (1<<MD_SB_CLEAN
))
5088 mddev
->recovery_cp
= MaxSector
;
5090 mddev
->recovery_cp
= 0;
5091 mddev
->persistent
= ! info
->not_persistent
;
5092 mddev
->external
= 0;
5094 mddev
->layout
= info
->layout
;
5095 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5097 mddev
->max_disks
= MD_SB_DISKS
;
5099 if (mddev
->persistent
)
5101 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5103 mddev
->default_bitmap_offset
= MD_SB_BYTES
>> 9;
5104 mddev
->bitmap_offset
= 0;
5106 mddev
->reshape_position
= MaxSector
;
5109 * Generate a 128 bit UUID
5111 get_random_bytes(mddev
->uuid
, 16);
5113 mddev
->new_level
= mddev
->level
;
5114 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5115 mddev
->new_layout
= mddev
->layout
;
5116 mddev
->delta_disks
= 0;
5121 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5123 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5125 if (mddev
->external_size
)
5128 mddev
->array_sectors
= array_sectors
;
5130 EXPORT_SYMBOL(md_set_array_sectors
);
5132 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5136 int fit
= (num_sectors
== 0);
5138 if (mddev
->pers
->resize
== NULL
)
5140 /* The "num_sectors" is the number of sectors of each device that
5141 * is used. This can only make sense for arrays with redundancy.
5142 * linear and raid0 always use whatever space is available. We can only
5143 * consider changing this number if no resync or reconstruction is
5144 * happening, and if the new size is acceptable. It must fit before the
5145 * sb_start or, if that is <data_offset, it must fit before the size
5146 * of each device. If num_sectors is zero, we find the largest size
5150 if (mddev
->sync_thread
)
5153 /* Sorry, cannot grow a bitmap yet, just remove it,
5157 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5158 sector_t avail
= rdev
->sectors
;
5160 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5161 num_sectors
= avail
;
5162 if (avail
< num_sectors
)
5165 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5167 revalidate_disk(mddev
->gendisk
);
5171 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5174 /* change the number of raid disks */
5175 if (mddev
->pers
->check_reshape
== NULL
)
5177 if (raid_disks
<= 0 ||
5178 raid_disks
>= mddev
->max_disks
)
5180 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5182 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5184 rv
= mddev
->pers
->check_reshape(mddev
);
5190 * update_array_info is used to change the configuration of an
5192 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5193 * fields in the info are checked against the array.
5194 * Any differences that cannot be handled will cause an error.
5195 * Normally, only one change can be managed at a time.
5197 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5203 /* calculate expected state,ignoring low bits */
5204 if (mddev
->bitmap
&& mddev
->bitmap_offset
)
5205 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5207 if (mddev
->major_version
!= info
->major_version
||
5208 mddev
->minor_version
!= info
->minor_version
||
5209 /* mddev->patch_version != info->patch_version || */
5210 mddev
->ctime
!= info
->ctime
||
5211 mddev
->level
!= info
->level
||
5212 /* mddev->layout != info->layout || */
5213 !mddev
->persistent
!= info
->not_persistent
||
5214 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5215 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5216 ((state
^info
->state
) & 0xfffffe00)
5219 /* Check there is only one change */
5220 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5222 if (mddev
->raid_disks
!= info
->raid_disks
)
5224 if (mddev
->layout
!= info
->layout
)
5226 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5233 if (mddev
->layout
!= info
->layout
) {
5235 * we don't need to do anything at the md level, the
5236 * personality will take care of it all.
5238 if (mddev
->pers
->check_reshape
== NULL
)
5241 mddev
->new_layout
= info
->layout
;
5242 rv
= mddev
->pers
->check_reshape(mddev
);
5244 mddev
->new_layout
= mddev
->layout
;
5248 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5249 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5251 if (mddev
->raid_disks
!= info
->raid_disks
)
5252 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5254 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5255 if (mddev
->pers
->quiesce
== NULL
)
5257 if (mddev
->recovery
|| mddev
->sync_thread
)
5259 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5260 /* add the bitmap */
5263 if (mddev
->default_bitmap_offset
== 0)
5265 mddev
->bitmap_offset
= mddev
->default_bitmap_offset
;
5266 mddev
->pers
->quiesce(mddev
, 1);
5267 rv
= bitmap_create(mddev
);
5269 bitmap_destroy(mddev
);
5270 mddev
->pers
->quiesce(mddev
, 0);
5272 /* remove the bitmap */
5275 if (mddev
->bitmap
->file
)
5277 mddev
->pers
->quiesce(mddev
, 1);
5278 bitmap_destroy(mddev
);
5279 mddev
->pers
->quiesce(mddev
, 0);
5280 mddev
->bitmap_offset
= 0;
5283 md_update_sb(mddev
, 1);
5287 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5291 if (mddev
->pers
== NULL
)
5294 rdev
= find_rdev(mddev
, dev
);
5298 md_error(mddev
, rdev
);
5303 * We have a problem here : there is no easy way to give a CHS
5304 * virtual geometry. We currently pretend that we have a 2 heads
5305 * 4 sectors (with a BIG number of cylinders...). This drives
5306 * dosfs just mad... ;-)
5308 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5310 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5314 geo
->cylinders
= get_capacity(mddev
->gendisk
) / 8;
5318 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5319 unsigned int cmd
, unsigned long arg
)
5322 void __user
*argp
= (void __user
*)arg
;
5323 mddev_t
*mddev
= NULL
;
5325 if (!capable(CAP_SYS_ADMIN
))
5329 * Commands dealing with the RAID driver but not any
5335 err
= get_version(argp
);
5338 case PRINT_RAID_DEBUG
:
5346 autostart_arrays(arg
);
5353 * Commands creating/starting a new array:
5356 mddev
= bdev
->bd_disk
->private_data
;
5363 err
= mddev_lock(mddev
);
5366 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5373 case SET_ARRAY_INFO
:
5375 mdu_array_info_t info
;
5377 memset(&info
, 0, sizeof(info
));
5378 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5383 err
= update_array_info(mddev
, &info
);
5385 printk(KERN_WARNING
"md: couldn't update"
5386 " array info. %d\n", err
);
5391 if (!list_empty(&mddev
->disks
)) {
5393 "md: array %s already has disks!\n",
5398 if (mddev
->raid_disks
) {
5400 "md: array %s already initialised!\n",
5405 err
= set_array_info(mddev
, &info
);
5407 printk(KERN_WARNING
"md: couldn't set"
5408 " array info. %d\n", err
);
5418 * Commands querying/configuring an existing array:
5420 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5421 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5422 if ((!mddev
->raid_disks
&& !mddev
->external
)
5423 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5424 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5425 && cmd
!= GET_BITMAP_FILE
) {
5431 * Commands even a read-only array can execute:
5435 case GET_ARRAY_INFO
:
5436 err
= get_array_info(mddev
, argp
);
5439 case GET_BITMAP_FILE
:
5440 err
= get_bitmap_file(mddev
, argp
);
5444 err
= get_disk_info(mddev
, argp
);
5447 case RESTART_ARRAY_RW
:
5448 err
= restart_array(mddev
);
5452 err
= do_md_stop(mddev
, 0, 1);
5456 err
= do_md_stop(mddev
, 1, 1);
5462 * The remaining ioctls are changing the state of the
5463 * superblock, so we do not allow them on read-only arrays.
5464 * However non-MD ioctls (e.g. get-size) will still come through
5465 * here and hit the 'default' below, so only disallow
5466 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5468 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5469 if (mddev
->ro
== 2) {
5471 sysfs_notify_dirent(mddev
->sysfs_state
);
5472 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5473 md_wakeup_thread(mddev
->thread
);
5484 mdu_disk_info_t info
;
5485 if (copy_from_user(&info
, argp
, sizeof(info
)))
5488 err
= add_new_disk(mddev
, &info
);
5492 case HOT_REMOVE_DISK
:
5493 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5497 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5500 case SET_DISK_FAULTY
:
5501 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5505 err
= do_md_run(mddev
);
5508 case SET_BITMAP_FILE
:
5509 err
= set_bitmap_file(mddev
, (int)arg
);
5519 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5521 mddev
->hold_active
= 0;
5522 mddev_unlock(mddev
);
5532 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5535 * Succeed if we can lock the mddev, which confirms that
5536 * it isn't being stopped right now.
5538 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5541 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5542 /* we are racing with mddev_put which is discarding this
5546 /* Wait until bdev->bd_disk is definitely gone */
5547 flush_scheduled_work();
5548 /* Then retry the open from the top */
5549 return -ERESTARTSYS
;
5551 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5553 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5557 atomic_inc(&mddev
->openers
);
5558 mutex_unlock(&mddev
->open_mutex
);
5560 check_disk_change(bdev
);
5565 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5567 mddev_t
*mddev
= disk
->private_data
;
5570 atomic_dec(&mddev
->openers
);
5576 static int md_media_changed(struct gendisk
*disk
)
5578 mddev_t
*mddev
= disk
->private_data
;
5580 return mddev
->changed
;
5583 static int md_revalidate(struct gendisk
*disk
)
5585 mddev_t
*mddev
= disk
->private_data
;
5590 static const struct block_device_operations md_fops
=
5592 .owner
= THIS_MODULE
,
5594 .release
= md_release
,
5596 .getgeo
= md_getgeo
,
5597 .media_changed
= md_media_changed
,
5598 .revalidate_disk
= md_revalidate
,
5601 static int md_thread(void * arg
)
5603 mdk_thread_t
*thread
= arg
;
5606 * md_thread is a 'system-thread', it's priority should be very
5607 * high. We avoid resource deadlocks individually in each
5608 * raid personality. (RAID5 does preallocation) We also use RR and
5609 * the very same RT priority as kswapd, thus we will never get
5610 * into a priority inversion deadlock.
5612 * we definitely have to have equal or higher priority than
5613 * bdflush, otherwise bdflush will deadlock if there are too
5614 * many dirty RAID5 blocks.
5617 allow_signal(SIGKILL
);
5618 while (!kthread_should_stop()) {
5620 /* We need to wait INTERRUPTIBLE so that
5621 * we don't add to the load-average.
5622 * That means we need to be sure no signals are
5625 if (signal_pending(current
))
5626 flush_signals(current
);
5628 wait_event_interruptible_timeout
5630 test_bit(THREAD_WAKEUP
, &thread
->flags
)
5631 || kthread_should_stop(),
5634 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
5636 thread
->run(thread
->mddev
);
5642 void md_wakeup_thread(mdk_thread_t
*thread
)
5645 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
5646 set_bit(THREAD_WAKEUP
, &thread
->flags
);
5647 wake_up(&thread
->wqueue
);
5651 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
5654 mdk_thread_t
*thread
;
5656 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
5660 init_waitqueue_head(&thread
->wqueue
);
5663 thread
->mddev
= mddev
;
5664 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
5665 thread
->tsk
= kthread_run(md_thread
, thread
,
5667 mdname(thread
->mddev
),
5668 name
?: mddev
->pers
->name
);
5669 if (IS_ERR(thread
->tsk
)) {
5676 void md_unregister_thread(mdk_thread_t
*thread
)
5680 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
5682 kthread_stop(thread
->tsk
);
5686 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
5693 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
5696 if (mddev
->external
)
5697 set_bit(Blocked
, &rdev
->flags
);
5699 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5701 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5702 __builtin_return_address(0),__builtin_return_address(1),
5703 __builtin_return_address(2),__builtin_return_address(3));
5707 if (!mddev
->pers
->error_handler
)
5709 mddev
->pers
->error_handler(mddev
,rdev
);
5710 if (mddev
->degraded
)
5711 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5712 set_bit(StateChanged
, &rdev
->flags
);
5713 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5714 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5715 md_wakeup_thread(mddev
->thread
);
5716 md_new_event_inintr(mddev
);
5719 /* seq_file implementation /proc/mdstat */
5721 static void status_unused(struct seq_file
*seq
)
5726 seq_printf(seq
, "unused devices: ");
5728 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
5729 char b
[BDEVNAME_SIZE
];
5731 seq_printf(seq
, "%s ",
5732 bdevname(rdev
->bdev
,b
));
5735 seq_printf(seq
, "<none>");
5737 seq_printf(seq
, "\n");
5741 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
5743 sector_t max_sectors
, resync
, res
;
5744 unsigned long dt
, db
;
5747 unsigned int per_milli
;
5749 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
5751 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
5752 max_sectors
= mddev
->resync_max_sectors
;
5754 max_sectors
= mddev
->dev_sectors
;
5757 * Should not happen.
5763 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5764 * in a sector_t, and (max_sectors>>scale) will fit in a
5765 * u32, as those are the requirements for sector_div.
5766 * Thus 'scale' must be at least 10
5769 if (sizeof(sector_t
) > sizeof(unsigned long)) {
5770 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
5773 res
= (resync
>>scale
)*1000;
5774 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
5778 int i
, x
= per_milli
/50, y
= 20-x
;
5779 seq_printf(seq
, "[");
5780 for (i
= 0; i
< x
; i
++)
5781 seq_printf(seq
, "=");
5782 seq_printf(seq
, ">");
5783 for (i
= 0; i
< y
; i
++)
5784 seq_printf(seq
, ".");
5785 seq_printf(seq
, "] ");
5787 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
5788 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
5790 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
5792 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
5793 "resync" : "recovery"))),
5794 per_milli
/10, per_milli
% 10,
5795 (unsigned long long) resync
/2,
5796 (unsigned long long) max_sectors
/2);
5799 * dt: time from mark until now
5800 * db: blocks written from mark until now
5801 * rt: remaining time
5803 * rt is a sector_t, so could be 32bit or 64bit.
5804 * So we divide before multiply in case it is 32bit and close
5806 * We scale the divisor (db) by 32 to avoid loosing precision
5807 * near the end of resync when the number of remaining sectors
5809 * We then divide rt by 32 after multiplying by db to compensate.
5810 * The '+1' avoids division by zero if db is very small.
5812 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
5814 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
5815 - mddev
->resync_mark_cnt
;
5817 rt
= max_sectors
- resync
; /* number of remaining sectors */
5818 sector_div(rt
, db
/32+1);
5822 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
5823 ((unsigned long)rt
% 60)/6);
5825 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
5828 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
5830 struct list_head
*tmp
;
5840 spin_lock(&all_mddevs_lock
);
5841 list_for_each(tmp
,&all_mddevs
)
5843 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
5845 spin_unlock(&all_mddevs_lock
);
5848 spin_unlock(&all_mddevs_lock
);
5850 return (void*)2;/* tail */
5854 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
5856 struct list_head
*tmp
;
5857 mddev_t
*next_mddev
, *mddev
= v
;
5863 spin_lock(&all_mddevs_lock
);
5865 tmp
= all_mddevs
.next
;
5867 tmp
= mddev
->all_mddevs
.next
;
5868 if (tmp
!= &all_mddevs
)
5869 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
5871 next_mddev
= (void*)2;
5874 spin_unlock(&all_mddevs_lock
);
5882 static void md_seq_stop(struct seq_file
*seq
, void *v
)
5886 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
5890 struct mdstat_info
{
5894 static int md_seq_show(struct seq_file
*seq
, void *v
)
5899 struct mdstat_info
*mi
= seq
->private;
5900 struct bitmap
*bitmap
;
5902 if (v
== (void*)1) {
5903 struct mdk_personality
*pers
;
5904 seq_printf(seq
, "Personalities : ");
5905 spin_lock(&pers_lock
);
5906 list_for_each_entry(pers
, &pers_list
, list
)
5907 seq_printf(seq
, "[%s] ", pers
->name
);
5909 spin_unlock(&pers_lock
);
5910 seq_printf(seq
, "\n");
5911 mi
->event
= atomic_read(&md_event_count
);
5914 if (v
== (void*)2) {
5919 if (mddev_lock(mddev
) < 0)
5922 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
5923 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
5924 mddev
->pers
? "" : "in");
5927 seq_printf(seq
, " (read-only)");
5929 seq_printf(seq
, " (auto-read-only)");
5930 seq_printf(seq
, " %s", mddev
->pers
->name
);
5934 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5935 char b
[BDEVNAME_SIZE
];
5936 seq_printf(seq
, " %s[%d]",
5937 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
5938 if (test_bit(WriteMostly
, &rdev
->flags
))
5939 seq_printf(seq
, "(W)");
5940 if (test_bit(Faulty
, &rdev
->flags
)) {
5941 seq_printf(seq
, "(F)");
5943 } else if (rdev
->raid_disk
< 0)
5944 seq_printf(seq
, "(S)"); /* spare */
5945 sectors
+= rdev
->sectors
;
5948 if (!list_empty(&mddev
->disks
)) {
5950 seq_printf(seq
, "\n %llu blocks",
5951 (unsigned long long)
5952 mddev
->array_sectors
/ 2);
5954 seq_printf(seq
, "\n %llu blocks",
5955 (unsigned long long)sectors
/ 2);
5957 if (mddev
->persistent
) {
5958 if (mddev
->major_version
!= 0 ||
5959 mddev
->minor_version
!= 90) {
5960 seq_printf(seq
," super %d.%d",
5961 mddev
->major_version
,
5962 mddev
->minor_version
);
5964 } else if (mddev
->external
)
5965 seq_printf(seq
, " super external:%s",
5966 mddev
->metadata_type
);
5968 seq_printf(seq
, " super non-persistent");
5971 mddev
->pers
->status(seq
, mddev
);
5972 seq_printf(seq
, "\n ");
5973 if (mddev
->pers
->sync_request
) {
5974 if (mddev
->curr_resync
> 2) {
5975 status_resync(seq
, mddev
);
5976 seq_printf(seq
, "\n ");
5977 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
5978 seq_printf(seq
, "\tresync=DELAYED\n ");
5979 else if (mddev
->recovery_cp
< MaxSector
)
5980 seq_printf(seq
, "\tresync=PENDING\n ");
5983 seq_printf(seq
, "\n ");
5985 if ((bitmap
= mddev
->bitmap
)) {
5986 unsigned long chunk_kb
;
5987 unsigned long flags
;
5988 spin_lock_irqsave(&bitmap
->lock
, flags
);
5989 chunk_kb
= bitmap
->chunksize
>> 10;
5990 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
5992 bitmap
->pages
- bitmap
->missing_pages
,
5994 (bitmap
->pages
- bitmap
->missing_pages
)
5995 << (PAGE_SHIFT
- 10),
5996 chunk_kb
? chunk_kb
: bitmap
->chunksize
,
5997 chunk_kb
? "KB" : "B");
5999 seq_printf(seq
, ", file: ");
6000 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6003 seq_printf(seq
, "\n");
6004 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6007 seq_printf(seq
, "\n");
6009 mddev_unlock(mddev
);
6014 static const struct seq_operations md_seq_ops
= {
6015 .start
= md_seq_start
,
6016 .next
= md_seq_next
,
6017 .stop
= md_seq_stop
,
6018 .show
= md_seq_show
,
6021 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6024 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6028 error
= seq_open(file
, &md_seq_ops
);
6032 struct seq_file
*p
= file
->private_data
;
6034 mi
->event
= atomic_read(&md_event_count
);
6039 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6041 struct seq_file
*m
= filp
->private_data
;
6042 struct mdstat_info
*mi
= m
->private;
6045 poll_wait(filp
, &md_event_waiters
, wait
);
6047 /* always allow read */
6048 mask
= POLLIN
| POLLRDNORM
;
6050 if (mi
->event
!= atomic_read(&md_event_count
))
6051 mask
|= POLLERR
| POLLPRI
;
6055 static const struct file_operations md_seq_fops
= {
6056 .owner
= THIS_MODULE
,
6057 .open
= md_seq_open
,
6059 .llseek
= seq_lseek
,
6060 .release
= seq_release_private
,
6061 .poll
= mdstat_poll
,
6064 int register_md_personality(struct mdk_personality
*p
)
6066 spin_lock(&pers_lock
);
6067 list_add_tail(&p
->list
, &pers_list
);
6068 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6069 spin_unlock(&pers_lock
);
6073 int unregister_md_personality(struct mdk_personality
*p
)
6075 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6076 spin_lock(&pers_lock
);
6077 list_del_init(&p
->list
);
6078 spin_unlock(&pers_lock
);
6082 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6090 rdev_for_each_rcu(rdev
, mddev
) {
6091 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6092 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6093 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6094 atomic_read(&disk
->sync_io
);
6095 /* sync IO will cause sync_io to increase before the disk_stats
6096 * as sync_io is counted when a request starts, and
6097 * disk_stats is counted when it completes.
6098 * So resync activity will cause curr_events to be smaller than
6099 * when there was no such activity.
6100 * non-sync IO will cause disk_stat to increase without
6101 * increasing sync_io so curr_events will (eventually)
6102 * be larger than it was before. Once it becomes
6103 * substantially larger, the test below will cause
6104 * the array to appear non-idle, and resync will slow
6106 * If there is a lot of outstanding resync activity when
6107 * we set last_event to curr_events, then all that activity
6108 * completing might cause the array to appear non-idle
6109 * and resync will be slowed down even though there might
6110 * not have been non-resync activity. This will only
6111 * happen once though. 'last_events' will soon reflect
6112 * the state where there is little or no outstanding
6113 * resync requests, and further resync activity will
6114 * always make curr_events less than last_events.
6117 if (init
|| curr_events
- rdev
->last_events
> 64) {
6118 rdev
->last_events
= curr_events
;
6126 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6128 /* another "blocks" (512byte) blocks have been synced */
6129 atomic_sub(blocks
, &mddev
->recovery_active
);
6130 wake_up(&mddev
->recovery_wait
);
6132 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6133 md_wakeup_thread(mddev
->thread
);
6134 // stop recovery, signal do_sync ....
6139 /* md_write_start(mddev, bi)
6140 * If we need to update some array metadata (e.g. 'active' flag
6141 * in superblock) before writing, schedule a superblock update
6142 * and wait for it to complete.
6144 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6147 if (bio_data_dir(bi
) != WRITE
)
6150 BUG_ON(mddev
->ro
== 1);
6151 if (mddev
->ro
== 2) {
6152 /* need to switch to read/write */
6154 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6155 md_wakeup_thread(mddev
->thread
);
6156 md_wakeup_thread(mddev
->sync_thread
);
6159 atomic_inc(&mddev
->writes_pending
);
6160 if (mddev
->safemode
== 1)
6161 mddev
->safemode
= 0;
6162 if (mddev
->in_sync
) {
6163 spin_lock_irq(&mddev
->write_lock
);
6164 if (mddev
->in_sync
) {
6166 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6167 md_wakeup_thread(mddev
->thread
);
6170 spin_unlock_irq(&mddev
->write_lock
);
6173 sysfs_notify_dirent(mddev
->sysfs_state
);
6174 wait_event(mddev
->sb_wait
,
6175 !test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
) &&
6176 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6179 void md_write_end(mddev_t
*mddev
)
6181 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6182 if (mddev
->safemode
== 2)
6183 md_wakeup_thread(mddev
->thread
);
6184 else if (mddev
->safemode_delay
)
6185 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6189 /* md_allow_write(mddev)
6190 * Calling this ensures that the array is marked 'active' so that writes
6191 * may proceed without blocking. It is important to call this before
6192 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6193 * Must be called with mddev_lock held.
6195 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6196 * is dropped, so return -EAGAIN after notifying userspace.
6198 int md_allow_write(mddev_t
*mddev
)
6204 if (!mddev
->pers
->sync_request
)
6207 spin_lock_irq(&mddev
->write_lock
);
6208 if (mddev
->in_sync
) {
6210 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6211 if (mddev
->safemode_delay
&&
6212 mddev
->safemode
== 0)
6213 mddev
->safemode
= 1;
6214 spin_unlock_irq(&mddev
->write_lock
);
6215 md_update_sb(mddev
, 0);
6216 sysfs_notify_dirent(mddev
->sysfs_state
);
6218 spin_unlock_irq(&mddev
->write_lock
);
6220 if (test_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
6225 EXPORT_SYMBOL_GPL(md_allow_write
);
6227 #define SYNC_MARKS 10
6228 #define SYNC_MARK_STEP (3*HZ)
6229 void md_do_sync(mddev_t
*mddev
)
6232 unsigned int currspeed
= 0,
6234 sector_t max_sectors
,j
, io_sectors
;
6235 unsigned long mark
[SYNC_MARKS
];
6236 sector_t mark_cnt
[SYNC_MARKS
];
6238 struct list_head
*tmp
;
6239 sector_t last_check
;
6244 /* just incase thread restarts... */
6245 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6247 if (mddev
->ro
) /* never try to sync a read-only array */
6250 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6251 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6252 desc
= "data-check";
6253 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6254 desc
= "requested-resync";
6257 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6262 /* we overload curr_resync somewhat here.
6263 * 0 == not engaged in resync at all
6264 * 2 == checking that there is no conflict with another sync
6265 * 1 == like 2, but have yielded to allow conflicting resync to
6267 * other == active in resync - this many blocks
6269 * Before starting a resync we must have set curr_resync to
6270 * 2, and then checked that every "conflicting" array has curr_resync
6271 * less than ours. When we find one that is the same or higher
6272 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6273 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6274 * This will mean we have to start checking from the beginning again.
6279 mddev
->curr_resync
= 2;
6282 if (kthread_should_stop()) {
6283 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6286 for_each_mddev(mddev2
, tmp
) {
6287 if (mddev2
== mddev
)
6289 if (!mddev
->parallel_resync
6290 && mddev2
->curr_resync
6291 && match_mddev_units(mddev
, mddev2
)) {
6293 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6294 /* arbitrarily yield */
6295 mddev
->curr_resync
= 1;
6296 wake_up(&resync_wait
);
6298 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6299 /* no need to wait here, we can wait the next
6300 * time 'round when curr_resync == 2
6303 /* We need to wait 'interruptible' so as not to
6304 * contribute to the load average, and not to
6305 * be caught by 'softlockup'
6307 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6308 if (!kthread_should_stop() &&
6309 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6310 printk(KERN_INFO
"md: delaying %s of %s"
6311 " until %s has finished (they"
6312 " share one or more physical units)\n",
6313 desc
, mdname(mddev
), mdname(mddev2
));
6315 if (signal_pending(current
))
6316 flush_signals(current
);
6318 finish_wait(&resync_wait
, &wq
);
6321 finish_wait(&resync_wait
, &wq
);
6324 } while (mddev
->curr_resync
< 2);
6327 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6328 /* resync follows the size requested by the personality,
6329 * which defaults to physical size, but can be virtual size
6331 max_sectors
= mddev
->resync_max_sectors
;
6332 mddev
->resync_mismatches
= 0;
6333 /* we don't use the checkpoint if there's a bitmap */
6334 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6335 j
= mddev
->resync_min
;
6336 else if (!mddev
->bitmap
)
6337 j
= mddev
->recovery_cp
;
6339 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6340 max_sectors
= mddev
->dev_sectors
;
6342 /* recovery follows the physical size of devices */
6343 max_sectors
= mddev
->dev_sectors
;
6345 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6346 if (rdev
->raid_disk
>= 0 &&
6347 !test_bit(Faulty
, &rdev
->flags
) &&
6348 !test_bit(In_sync
, &rdev
->flags
) &&
6349 rdev
->recovery_offset
< j
)
6350 j
= rdev
->recovery_offset
;
6353 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6354 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6355 " %d KB/sec/disk.\n", speed_min(mddev
));
6356 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6357 "(but not more than %d KB/sec) for %s.\n",
6358 speed_max(mddev
), desc
);
6360 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6363 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6365 mark_cnt
[m
] = io_sectors
;
6368 mddev
->resync_mark
= mark
[last_mark
];
6369 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6372 * Tune reconstruction:
6374 window
= 32*(PAGE_SIZE
/512);
6375 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6376 window
/2,(unsigned long long) max_sectors
/2);
6378 atomic_set(&mddev
->recovery_active
, 0);
6383 "md: resuming %s of %s from checkpoint.\n",
6384 desc
, mdname(mddev
));
6385 mddev
->curr_resync
= j
;
6388 while (j
< max_sectors
) {
6393 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6394 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6395 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6396 > (max_sectors
>> 4)) ||
6397 (j
- mddev
->curr_resync_completed
)*2
6398 >= mddev
->resync_max
- mddev
->curr_resync_completed
6400 /* time to update curr_resync_completed */
6401 blk_unplug(mddev
->queue
);
6402 wait_event(mddev
->recovery_wait
,
6403 atomic_read(&mddev
->recovery_active
) == 0);
6404 mddev
->curr_resync_completed
=
6406 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6407 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6410 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6411 /* As this condition is controlled by user-space,
6412 * we can block indefinitely, so use '_interruptible'
6413 * to avoid triggering warnings.
6415 flush_signals(current
); /* just in case */
6416 wait_event_interruptible(mddev
->recovery_wait
,
6417 mddev
->resync_max
> j
6418 || kthread_should_stop());
6421 if (kthread_should_stop())
6424 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6425 currspeed
< speed_min(mddev
));
6427 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6431 if (!skipped
) { /* actual IO requested */
6432 io_sectors
+= sectors
;
6433 atomic_add(sectors
, &mddev
->recovery_active
);
6437 if (j
>1) mddev
->curr_resync
= j
;
6438 mddev
->curr_mark_cnt
= io_sectors
;
6439 if (last_check
== 0)
6440 /* this is the earliers that rebuilt will be
6441 * visible in /proc/mdstat
6443 md_new_event(mddev
);
6445 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6448 last_check
= io_sectors
;
6450 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6454 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6456 int next
= (last_mark
+1) % SYNC_MARKS
;
6458 mddev
->resync_mark
= mark
[next
];
6459 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6460 mark
[next
] = jiffies
;
6461 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6466 if (kthread_should_stop())
6471 * this loop exits only if either when we are slower than
6472 * the 'hard' speed limit, or the system was IO-idle for
6474 * the system might be non-idle CPU-wise, but we only care
6475 * about not overloading the IO subsystem. (things like an
6476 * e2fsck being done on the RAID array should execute fast)
6478 blk_unplug(mddev
->queue
);
6481 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6482 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6484 if (currspeed
> speed_min(mddev
)) {
6485 if ((currspeed
> speed_max(mddev
)) ||
6486 !is_mddev_idle(mddev
, 0)) {
6492 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6494 * this also signals 'finished resyncing' to md_stop
6497 blk_unplug(mddev
->queue
);
6499 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6501 /* tell personality that we are finished */
6502 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6504 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6505 mddev
->curr_resync
> 2) {
6506 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6507 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6508 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6510 "md: checkpointing %s of %s.\n",
6511 desc
, mdname(mddev
));
6512 mddev
->recovery_cp
= mddev
->curr_resync
;
6515 mddev
->recovery_cp
= MaxSector
;
6517 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6518 mddev
->curr_resync
= MaxSector
;
6519 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6520 if (rdev
->raid_disk
>= 0 &&
6521 !test_bit(Faulty
, &rdev
->flags
) &&
6522 !test_bit(In_sync
, &rdev
->flags
) &&
6523 rdev
->recovery_offset
< mddev
->curr_resync
)
6524 rdev
->recovery_offset
= mddev
->curr_resync
;
6527 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6530 mddev
->curr_resync
= 0;
6531 mddev
->curr_resync_completed
= 0;
6532 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6533 /* We completed so max setting can be forgotten. */
6534 mddev
->resync_max
= MaxSector
;
6535 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6536 wake_up(&resync_wait
);
6537 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6538 md_wakeup_thread(mddev
->thread
);
6543 * got a signal, exit.
6546 "md: md_do_sync() got signal ... exiting\n");
6547 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6551 EXPORT_SYMBOL_GPL(md_do_sync
);
6554 static int remove_and_add_spares(mddev_t
*mddev
)
6559 mddev
->curr_resync_completed
= 0;
6561 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6562 if (rdev
->raid_disk
>= 0 &&
6563 !test_bit(Blocked
, &rdev
->flags
) &&
6564 (test_bit(Faulty
, &rdev
->flags
) ||
6565 ! test_bit(In_sync
, &rdev
->flags
)) &&
6566 atomic_read(&rdev
->nr_pending
)==0) {
6567 if (mddev
->pers
->hot_remove_disk(
6568 mddev
, rdev
->raid_disk
)==0) {
6570 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6571 sysfs_remove_link(&mddev
->kobj
, nm
);
6572 rdev
->raid_disk
= -1;
6576 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
6577 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6578 if (rdev
->raid_disk
>= 0 &&
6579 !test_bit(In_sync
, &rdev
->flags
) &&
6580 !test_bit(Blocked
, &rdev
->flags
))
6582 if (rdev
->raid_disk
< 0
6583 && !test_bit(Faulty
, &rdev
->flags
)) {
6584 rdev
->recovery_offset
= 0;
6586 hot_add_disk(mddev
, rdev
) == 0) {
6588 sprintf(nm
, "rd%d", rdev
->raid_disk
);
6589 if (sysfs_create_link(&mddev
->kobj
,
6592 "md: cannot register "
6596 md_new_event(mddev
);
6605 * This routine is regularly called by all per-raid-array threads to
6606 * deal with generic issues like resync and super-block update.
6607 * Raid personalities that don't have a thread (linear/raid0) do not
6608 * need this as they never do any recovery or update the superblock.
6610 * It does not do any resync itself, but rather "forks" off other threads
6611 * to do that as needed.
6612 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6613 * "->recovery" and create a thread at ->sync_thread.
6614 * When the thread finishes it sets MD_RECOVERY_DONE
6615 * and wakeups up this thread which will reap the thread and finish up.
6616 * This thread also removes any faulty devices (with nr_pending == 0).
6618 * The overall approach is:
6619 * 1/ if the superblock needs updating, update it.
6620 * 2/ If a recovery thread is running, don't do anything else.
6621 * 3/ If recovery has finished, clean up, possibly marking spares active.
6622 * 4/ If there are any faulty devices, remove them.
6623 * 5/ If array is degraded, try to add spares devices
6624 * 6/ If array has spares or is not in-sync, start a resync thread.
6626 void md_check_recovery(mddev_t
*mddev
)
6632 bitmap_daemon_work(mddev
->bitmap
);
6637 if (signal_pending(current
)) {
6638 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
6639 printk(KERN_INFO
"md: %s in immediate safe mode\n",
6641 mddev
->safemode
= 2;
6643 flush_signals(current
);
6646 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
6649 (mddev
->flags
&& !mddev
->external
) ||
6650 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
6651 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
6652 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
6653 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
6654 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
6658 if (mddev_trylock(mddev
)) {
6662 /* Only thing we do on a ro array is remove
6665 remove_and_add_spares(mddev
);
6666 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6670 if (!mddev
->external
) {
6672 spin_lock_irq(&mddev
->write_lock
);
6673 if (mddev
->safemode
&&
6674 !atomic_read(&mddev
->writes_pending
) &&
6676 mddev
->recovery_cp
== MaxSector
) {
6679 if (mddev
->persistent
)
6680 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6682 if (mddev
->safemode
== 1)
6683 mddev
->safemode
= 0;
6684 spin_unlock_irq(&mddev
->write_lock
);
6686 sysfs_notify_dirent(mddev
->sysfs_state
);
6690 md_update_sb(mddev
, 0);
6692 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6693 if (test_and_clear_bit(StateChanged
, &rdev
->flags
))
6694 sysfs_notify_dirent(rdev
->sysfs_state
);
6697 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
6698 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
6699 /* resync/recovery still happening */
6700 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6703 if (mddev
->sync_thread
) {
6704 /* resync has finished, collect result */
6705 md_unregister_thread(mddev
->sync_thread
);
6706 mddev
->sync_thread
= NULL
;
6707 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
6708 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
6710 /* activate any spares */
6711 if (mddev
->pers
->spare_active(mddev
))
6712 sysfs_notify(&mddev
->kobj
, NULL
,
6715 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6716 mddev
->pers
->finish_reshape
)
6717 mddev
->pers
->finish_reshape(mddev
);
6718 md_update_sb(mddev
, 1);
6720 /* if array is no-longer degraded, then any saved_raid_disk
6721 * information must be scrapped
6723 if (!mddev
->degraded
)
6724 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6725 rdev
->saved_raid_disk
= -1;
6727 mddev
->recovery
= 0;
6728 /* flag recovery needed just to double check */
6729 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6730 sysfs_notify_dirent(mddev
->sysfs_action
);
6731 md_new_event(mddev
);
6734 /* Set RUNNING before clearing NEEDED to avoid
6735 * any transients in the value of "sync_action".
6737 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6738 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6739 /* Clear some bits that don't mean anything, but
6742 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6743 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6745 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
6747 /* no recovery is running.
6748 * remove any failed drives, then
6749 * add spares if possible.
6750 * Spare are also removed and re-added, to allow
6751 * the personality to fail the re-add.
6754 if (mddev
->reshape_position
!= MaxSector
) {
6755 if (mddev
->pers
->check_reshape
== NULL
||
6756 mddev
->pers
->check_reshape(mddev
) != 0)
6757 /* Cannot proceed */
6759 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
6760 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6761 } else if ((spares
= remove_and_add_spares(mddev
))) {
6762 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6763 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
6764 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
6765 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6766 } else if (mddev
->recovery_cp
< MaxSector
) {
6767 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
6768 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6769 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6770 /* nothing to be done ... */
6773 if (mddev
->pers
->sync_request
) {
6774 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
6775 /* We are adding a device or devices to an array
6776 * which has the bitmap stored on all devices.
6777 * So make sure all bitmap pages get written
6779 bitmap_write_all(mddev
->bitmap
);
6781 mddev
->sync_thread
= md_register_thread(md_do_sync
,
6784 if (!mddev
->sync_thread
) {
6785 printk(KERN_ERR
"%s: could not start resync"
6788 /* leave the spares where they are, it shouldn't hurt */
6789 mddev
->recovery
= 0;
6791 md_wakeup_thread(mddev
->sync_thread
);
6792 sysfs_notify_dirent(mddev
->sysfs_action
);
6793 md_new_event(mddev
);
6796 if (!mddev
->sync_thread
) {
6797 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
6798 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
6800 if (mddev
->sysfs_action
)
6801 sysfs_notify_dirent(mddev
->sysfs_action
);
6803 mddev_unlock(mddev
);
6807 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
6809 sysfs_notify_dirent(rdev
->sysfs_state
);
6810 wait_event_timeout(rdev
->blocked_wait
,
6811 !test_bit(Blocked
, &rdev
->flags
),
6812 msecs_to_jiffies(5000));
6813 rdev_dec_pending(rdev
, mddev
);
6815 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
6817 static int md_notify_reboot(struct notifier_block
*this,
6818 unsigned long code
, void *x
)
6820 struct list_head
*tmp
;
6823 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
6825 printk(KERN_INFO
"md: stopping all md devices.\n");
6827 for_each_mddev(mddev
, tmp
)
6828 if (mddev_trylock(mddev
)) {
6829 /* Force a switch to readonly even array
6830 * appears to still be in use. Hence
6833 do_md_stop(mddev
, 1, 100);
6834 mddev_unlock(mddev
);
6837 * certain more exotic SCSI devices are known to be
6838 * volatile wrt too early system reboots. While the
6839 * right place to handle this issue is the given
6840 * driver, we do want to have a safe RAID driver ...
6847 static struct notifier_block md_notifier
= {
6848 .notifier_call
= md_notify_reboot
,
6850 .priority
= INT_MAX
, /* before any real devices */
6853 static void md_geninit(void)
6855 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
6857 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
6860 static int __init
md_init(void)
6862 if (register_blkdev(MD_MAJOR
, "md"))
6864 if ((mdp_major
=register_blkdev(0, "mdp"))<=0) {
6865 unregister_blkdev(MD_MAJOR
, "md");
6868 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6869 md_probe
, NULL
, NULL
);
6870 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
6871 md_probe
, NULL
, NULL
);
6873 register_reboot_notifier(&md_notifier
);
6874 raid_table_header
= register_sysctl_table(raid_root_table
);
6884 * Searches all registered partitions for autorun RAID arrays
6888 static LIST_HEAD(all_detected_devices
);
6889 struct detected_devices_node
{
6890 struct list_head list
;
6894 void md_autodetect_dev(dev_t dev
)
6896 struct detected_devices_node
*node_detected_dev
;
6898 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
6899 if (node_detected_dev
) {
6900 node_detected_dev
->dev
= dev
;
6901 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
6903 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
6904 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
6909 static void autostart_arrays(int part
)
6912 struct detected_devices_node
*node_detected_dev
;
6914 int i_scanned
, i_passed
;
6919 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
6921 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
6923 node_detected_dev
= list_entry(all_detected_devices
.next
,
6924 struct detected_devices_node
, list
);
6925 list_del(&node_detected_dev
->list
);
6926 dev
= node_detected_dev
->dev
;
6927 kfree(node_detected_dev
);
6928 rdev
= md_import_device(dev
,0, 90);
6932 if (test_bit(Faulty
, &rdev
->flags
)) {
6936 set_bit(AutoDetected
, &rdev
->flags
);
6937 list_add(&rdev
->same_set
, &pending_raid_disks
);
6941 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
6942 i_scanned
, i_passed
);
6944 autorun_devices(part
);
6947 #endif /* !MODULE */
6949 static __exit
void md_exit(void)
6952 struct list_head
*tmp
;
6954 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
6955 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
6957 unregister_blkdev(MD_MAJOR
,"md");
6958 unregister_blkdev(mdp_major
, "mdp");
6959 unregister_reboot_notifier(&md_notifier
);
6960 unregister_sysctl_table(raid_table_header
);
6961 remove_proc_entry("mdstat", NULL
);
6962 for_each_mddev(mddev
, tmp
) {
6963 export_array(mddev
);
6964 mddev
->hold_active
= 0;
6968 subsys_initcall(md_init
);
6969 module_exit(md_exit
)
6971 static int get_ro(char *buffer
, struct kernel_param
*kp
)
6973 return sprintf(buffer
, "%d", start_readonly
);
6975 static int set_ro(const char *val
, struct kernel_param
*kp
)
6978 int num
= simple_strtoul(val
, &e
, 10);
6979 if (*val
&& (*e
== '\0' || *e
== '\n')) {
6980 start_readonly
= num
;
6986 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
6987 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
6989 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
6991 EXPORT_SYMBOL(register_md_personality
);
6992 EXPORT_SYMBOL(unregister_md_personality
);
6993 EXPORT_SYMBOL(md_error
);
6994 EXPORT_SYMBOL(md_done_sync
);
6995 EXPORT_SYMBOL(md_write_start
);
6996 EXPORT_SYMBOL(md_write_end
);
6997 EXPORT_SYMBOL(md_register_thread
);
6998 EXPORT_SYMBOL(md_unregister_thread
);
6999 EXPORT_SYMBOL(md_wakeup_thread
);
7000 EXPORT_SYMBOL(md_check_recovery
);
7001 MODULE_LICENSE("GPL");
7003 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);