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>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min
= 1000;
97 static int sysctl_speed_limit_max
= 200000;
98 static inline int speed_min(struct mddev
*mddev
)
100 return mddev
->sync_speed_min
?
101 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
104 static inline int speed_max(struct mddev
*mddev
)
106 return mddev
->sync_speed_max
?
107 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
110 static struct ctl_table_header
*raid_table_header
;
112 static ctl_table raid_table
[] = {
114 .procname
= "speed_limit_min",
115 .data
= &sysctl_speed_limit_min
,
116 .maxlen
= sizeof(int),
117 .mode
= S_IRUGO
|S_IWUSR
,
118 .proc_handler
= proc_dointvec
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
134 .mode
= S_IRUGO
|S_IXUGO
,
140 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static const struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * like bio_clone, but with a local bio set
158 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
163 if (!mddev
|| !mddev
->bio_set
)
164 return bio_alloc(gfp_mask
, nr_iovecs
);
166 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
171 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
173 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
176 if (!mddev
|| !mddev
->bio_set
)
177 return bio_clone(bio
, gfp_mask
);
179 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
181 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
183 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
185 /* 'bio' is a cloned bio which we need to trim to match
186 * the given offset and size.
187 * This requires adjusting bi_sector, bi_size, and bi_io_vec
190 struct bio_vec
*bvec
;
194 if (offset
== 0 && size
== bio
->bi_size
)
197 bio
->bi_sector
+= offset
;
200 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
202 while (bio
->bi_idx
< bio
->bi_vcnt
&&
203 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
204 /* remove this whole bio_vec */
205 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
208 if (bio
->bi_idx
< bio
->bi_vcnt
) {
209 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
210 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
212 /* avoid any complications with bi_idx being non-zero*/
214 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
215 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
216 bio
->bi_vcnt
-= bio
->bi_idx
;
219 /* Make sure vcnt and last bv are not too big */
220 bio_for_each_segment(bvec
, bio
, i
) {
221 if (sofar
+ bvec
->bv_len
> size
)
222 bvec
->bv_len
= size
- sofar
;
223 if (bvec
->bv_len
== 0) {
227 sofar
+= bvec
->bv_len
;
230 EXPORT_SYMBOL_GPL(md_trim_bio
);
233 * We have a system wide 'event count' that is incremented
234 * on any 'interesting' event, and readers of /proc/mdstat
235 * can use 'poll' or 'select' to find out when the event
239 * start array, stop array, error, add device, remove device,
240 * start build, activate spare
242 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
243 static atomic_t md_event_count
;
244 void md_new_event(struct mddev
*mddev
)
246 atomic_inc(&md_event_count
);
247 wake_up(&md_event_waiters
);
249 EXPORT_SYMBOL_GPL(md_new_event
);
251 /* Alternate version that can be called from interrupts
252 * when calling sysfs_notify isn't needed.
254 static void md_new_event_inintr(struct mddev
*mddev
)
256 atomic_inc(&md_event_count
);
257 wake_up(&md_event_waiters
);
261 * Enables to iterate over all existing md arrays
262 * all_mddevs_lock protects this list.
264 static LIST_HEAD(all_mddevs
);
265 static DEFINE_SPINLOCK(all_mddevs_lock
);
269 * iterates through all used mddevs in the system.
270 * We take care to grab the all_mddevs_lock whenever navigating
271 * the list, and to always hold a refcount when unlocked.
272 * Any code which breaks out of this loop while own
273 * a reference to the current mddev and must mddev_put it.
275 #define for_each_mddev(_mddev,_tmp) \
277 for (({ spin_lock(&all_mddevs_lock); \
278 _tmp = all_mddevs.next; \
280 ({ if (_tmp != &all_mddevs) \
281 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
282 spin_unlock(&all_mddevs_lock); \
283 if (_mddev) mddev_put(_mddev); \
284 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
285 _tmp != &all_mddevs;}); \
286 ({ spin_lock(&all_mddevs_lock); \
287 _tmp = _tmp->next;}) \
291 /* Rather than calling directly into the personality make_request function,
292 * IO requests come here first so that we can check if the device is
293 * being suspended pending a reconfiguration.
294 * We hold a refcount over the call to ->make_request. By the time that
295 * call has finished, the bio has been linked into some internal structure
296 * and so is visible to ->quiesce(), so we don't need the refcount any more.
298 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
300 const int rw
= bio_data_dir(bio
);
301 struct mddev
*mddev
= q
->queuedata
;
303 unsigned int sectors
;
305 if (mddev
== NULL
|| mddev
->pers
== NULL
310 smp_rmb(); /* Ensure implications of 'active' are visible */
312 if (mddev
->suspended
) {
315 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
316 TASK_UNINTERRUPTIBLE
);
317 if (!mddev
->suspended
)
323 finish_wait(&mddev
->sb_wait
, &__wait
);
325 atomic_inc(&mddev
->active_io
);
329 * save the sectors now since our bio can
330 * go away inside make_request
332 sectors
= bio_sectors(bio
);
333 mddev
->pers
->make_request(mddev
, bio
);
335 cpu
= part_stat_lock();
336 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
337 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
340 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
341 wake_up(&mddev
->sb_wait
);
344 /* mddev_suspend makes sure no new requests are submitted
345 * to the device, and that any requests that have been submitted
346 * are completely handled.
347 * Once ->stop is called and completes, the module will be completely
350 void mddev_suspend(struct mddev
*mddev
)
352 BUG_ON(mddev
->suspended
);
353 mddev
->suspended
= 1;
355 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
356 mddev
->pers
->quiesce(mddev
, 1);
358 del_timer_sync(&mddev
->safemode_timer
);
360 EXPORT_SYMBOL_GPL(mddev_suspend
);
362 void mddev_resume(struct mddev
*mddev
)
364 mddev
->suspended
= 0;
365 wake_up(&mddev
->sb_wait
);
366 mddev
->pers
->quiesce(mddev
, 0);
368 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
369 md_wakeup_thread(mddev
->thread
);
370 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
372 EXPORT_SYMBOL_GPL(mddev_resume
);
374 int mddev_congested(struct mddev
*mddev
, int bits
)
376 return mddev
->suspended
;
378 EXPORT_SYMBOL(mddev_congested
);
381 * Generic flush handling for md
384 static void md_end_flush(struct bio
*bio
, int err
)
386 struct md_rdev
*rdev
= bio
->bi_private
;
387 struct mddev
*mddev
= rdev
->mddev
;
389 rdev_dec_pending(rdev
, mddev
);
391 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
392 /* The pre-request flush has finished */
393 queue_work(md_wq
, &mddev
->flush_work
);
398 static void md_submit_flush_data(struct work_struct
*ws
);
400 static void submit_flushes(struct work_struct
*ws
)
402 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
403 struct md_rdev
*rdev
;
405 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
406 atomic_set(&mddev
->flush_pending
, 1);
408 rdev_for_each_rcu(rdev
, mddev
)
409 if (rdev
->raid_disk
>= 0 &&
410 !test_bit(Faulty
, &rdev
->flags
)) {
411 /* Take two references, one is dropped
412 * when request finishes, one after
413 * we reclaim rcu_read_lock
416 atomic_inc(&rdev
->nr_pending
);
417 atomic_inc(&rdev
->nr_pending
);
419 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
420 bi
->bi_end_io
= md_end_flush
;
421 bi
->bi_private
= rdev
;
422 bi
->bi_bdev
= rdev
->bdev
;
423 atomic_inc(&mddev
->flush_pending
);
424 submit_bio(WRITE_FLUSH
, bi
);
426 rdev_dec_pending(rdev
, mddev
);
429 if (atomic_dec_and_test(&mddev
->flush_pending
))
430 queue_work(md_wq
, &mddev
->flush_work
);
433 static void md_submit_flush_data(struct work_struct
*ws
)
435 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
436 struct bio
*bio
= mddev
->flush_bio
;
438 if (bio
->bi_size
== 0)
439 /* an empty barrier - all done */
442 bio
->bi_rw
&= ~REQ_FLUSH
;
443 mddev
->pers
->make_request(mddev
, bio
);
446 mddev
->flush_bio
= NULL
;
447 wake_up(&mddev
->sb_wait
);
450 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
452 spin_lock_irq(&mddev
->write_lock
);
453 wait_event_lock_irq(mddev
->sb_wait
,
455 mddev
->write_lock
, /*nothing*/);
456 mddev
->flush_bio
= bio
;
457 spin_unlock_irq(&mddev
->write_lock
);
459 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
460 queue_work(md_wq
, &mddev
->flush_work
);
462 EXPORT_SYMBOL(md_flush_request
);
464 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
466 struct mddev
*mddev
= cb
->data
;
467 md_wakeup_thread(mddev
->thread
);
470 EXPORT_SYMBOL(md_unplug
);
472 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
474 atomic_inc(&mddev
->active
);
478 static void mddev_delayed_delete(struct work_struct
*ws
);
480 static void mddev_put(struct mddev
*mddev
)
482 struct bio_set
*bs
= NULL
;
484 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
486 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
487 mddev
->ctime
== 0 && !mddev
->hold_active
) {
488 /* Array is not configured at all, and not held active,
490 list_del_init(&mddev
->all_mddevs
);
492 mddev
->bio_set
= NULL
;
493 if (mddev
->gendisk
) {
494 /* We did a probe so need to clean up. Call
495 * queue_work inside the spinlock so that
496 * flush_workqueue() after mddev_find will
497 * succeed in waiting for the work to be done.
499 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
500 queue_work(md_misc_wq
, &mddev
->del_work
);
504 spin_unlock(&all_mddevs_lock
);
509 void mddev_init(struct mddev
*mddev
)
511 mutex_init(&mddev
->open_mutex
);
512 mutex_init(&mddev
->reconfig_mutex
);
513 mutex_init(&mddev
->bitmap_info
.mutex
);
514 INIT_LIST_HEAD(&mddev
->disks
);
515 INIT_LIST_HEAD(&mddev
->all_mddevs
);
516 init_timer(&mddev
->safemode_timer
);
517 atomic_set(&mddev
->active
, 1);
518 atomic_set(&mddev
->openers
, 0);
519 atomic_set(&mddev
->active_io
, 0);
520 spin_lock_init(&mddev
->write_lock
);
521 atomic_set(&mddev
->flush_pending
, 0);
522 init_waitqueue_head(&mddev
->sb_wait
);
523 init_waitqueue_head(&mddev
->recovery_wait
);
524 mddev
->reshape_position
= MaxSector
;
525 mddev
->reshape_backwards
= 0;
526 mddev
->resync_min
= 0;
527 mddev
->resync_max
= MaxSector
;
528 mddev
->level
= LEVEL_NONE
;
530 EXPORT_SYMBOL_GPL(mddev_init
);
532 static struct mddev
* mddev_find(dev_t unit
)
534 struct mddev
*mddev
, *new = NULL
;
536 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
537 unit
&= ~((1<<MdpMinorShift
)-1);
540 spin_lock(&all_mddevs_lock
);
543 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
544 if (mddev
->unit
== unit
) {
546 spin_unlock(&all_mddevs_lock
);
552 list_add(&new->all_mddevs
, &all_mddevs
);
553 spin_unlock(&all_mddevs_lock
);
554 new->hold_active
= UNTIL_IOCTL
;
558 /* find an unused unit number */
559 static int next_minor
= 512;
560 int start
= next_minor
;
564 dev
= MKDEV(MD_MAJOR
, next_minor
);
566 if (next_minor
> MINORMASK
)
568 if (next_minor
== start
) {
569 /* Oh dear, all in use. */
570 spin_unlock(&all_mddevs_lock
);
576 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
577 if (mddev
->unit
== dev
) {
583 new->md_minor
= MINOR(dev
);
584 new->hold_active
= UNTIL_STOP
;
585 list_add(&new->all_mddevs
, &all_mddevs
);
586 spin_unlock(&all_mddevs_lock
);
589 spin_unlock(&all_mddevs_lock
);
591 new = kzalloc(sizeof(*new), GFP_KERNEL
);
596 if (MAJOR(unit
) == MD_MAJOR
)
597 new->md_minor
= MINOR(unit
);
599 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
606 static inline int mddev_lock(struct mddev
* mddev
)
608 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
611 static inline int mddev_is_locked(struct mddev
*mddev
)
613 return mutex_is_locked(&mddev
->reconfig_mutex
);
616 static inline int mddev_trylock(struct mddev
* mddev
)
618 return mutex_trylock(&mddev
->reconfig_mutex
);
621 static struct attribute_group md_redundancy_group
;
623 static void mddev_unlock(struct mddev
* mddev
)
625 if (mddev
->to_remove
) {
626 /* These cannot be removed under reconfig_mutex as
627 * an access to the files will try to take reconfig_mutex
628 * while holding the file unremovable, which leads to
630 * So hold set sysfs_active while the remove in happeing,
631 * and anything else which might set ->to_remove or my
632 * otherwise change the sysfs namespace will fail with
633 * -EBUSY if sysfs_active is still set.
634 * We set sysfs_active under reconfig_mutex and elsewhere
635 * test it under the same mutex to ensure its correct value
638 struct attribute_group
*to_remove
= mddev
->to_remove
;
639 mddev
->to_remove
= NULL
;
640 mddev
->sysfs_active
= 1;
641 mutex_unlock(&mddev
->reconfig_mutex
);
643 if (mddev
->kobj
.sd
) {
644 if (to_remove
!= &md_redundancy_group
)
645 sysfs_remove_group(&mddev
->kobj
, to_remove
);
646 if (mddev
->pers
== NULL
||
647 mddev
->pers
->sync_request
== NULL
) {
648 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
649 if (mddev
->sysfs_action
)
650 sysfs_put(mddev
->sysfs_action
);
651 mddev
->sysfs_action
= NULL
;
654 mddev
->sysfs_active
= 0;
656 mutex_unlock(&mddev
->reconfig_mutex
);
658 /* As we've dropped the mutex we need a spinlock to
659 * make sure the thread doesn't disappear
661 spin_lock(&pers_lock
);
662 md_wakeup_thread(mddev
->thread
);
663 spin_unlock(&pers_lock
);
666 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
668 struct md_rdev
*rdev
;
670 rdev_for_each(rdev
, mddev
)
671 if (rdev
->desc_nr
== nr
)
677 static struct md_rdev
* find_rdev(struct mddev
* mddev
, dev_t dev
)
679 struct md_rdev
*rdev
;
681 rdev_for_each(rdev
, mddev
)
682 if (rdev
->bdev
->bd_dev
== dev
)
688 static struct md_personality
*find_pers(int level
, char *clevel
)
690 struct md_personality
*pers
;
691 list_for_each_entry(pers
, &pers_list
, list
) {
692 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
694 if (strcmp(pers
->name
, clevel
)==0)
700 /* return the offset of the super block in 512byte sectors */
701 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
703 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
704 return MD_NEW_SIZE_SECTORS(num_sectors
);
707 static int alloc_disk_sb(struct md_rdev
* rdev
)
712 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
713 if (!rdev
->sb_page
) {
714 printk(KERN_ALERT
"md: out of memory.\n");
721 void md_rdev_clear(struct md_rdev
*rdev
)
724 put_page(rdev
->sb_page
);
726 rdev
->sb_page
= NULL
;
731 put_page(rdev
->bb_page
);
732 rdev
->bb_page
= NULL
;
734 kfree(rdev
->badblocks
.page
);
735 rdev
->badblocks
.page
= NULL
;
737 EXPORT_SYMBOL_GPL(md_rdev_clear
);
739 static void super_written(struct bio
*bio
, int error
)
741 struct md_rdev
*rdev
= bio
->bi_private
;
742 struct mddev
*mddev
= rdev
->mddev
;
744 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
745 printk("md: super_written gets error=%d, uptodate=%d\n",
746 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
747 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
748 md_error(mddev
, rdev
);
751 if (atomic_dec_and_test(&mddev
->pending_writes
))
752 wake_up(&mddev
->sb_wait
);
756 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
757 sector_t sector
, int size
, struct page
*page
)
759 /* write first size bytes of page to sector of rdev
760 * Increment mddev->pending_writes before returning
761 * and decrement it on completion, waking up sb_wait
762 * if zero is reached.
763 * If an error occurred, call md_error
765 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
767 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
768 bio
->bi_sector
= sector
;
769 bio_add_page(bio
, page
, size
, 0);
770 bio
->bi_private
= rdev
;
771 bio
->bi_end_io
= super_written
;
773 atomic_inc(&mddev
->pending_writes
);
774 submit_bio(WRITE_FLUSH_FUA
, bio
);
777 void md_super_wait(struct mddev
*mddev
)
779 /* wait for all superblock writes that were scheduled to complete */
782 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
783 if (atomic_read(&mddev
->pending_writes
)==0)
787 finish_wait(&mddev
->sb_wait
, &wq
);
790 static void bi_complete(struct bio
*bio
, int error
)
792 complete((struct completion
*)bio
->bi_private
);
795 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
796 struct page
*page
, int rw
, bool metadata_op
)
798 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
799 struct completion event
;
804 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
805 rdev
->meta_bdev
: rdev
->bdev
;
807 bio
->bi_sector
= sector
+ rdev
->sb_start
;
808 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
809 (rdev
->mddev
->reshape_backwards
==
810 (sector
>= rdev
->mddev
->reshape_position
)))
811 bio
->bi_sector
= sector
+ rdev
->new_data_offset
;
813 bio
->bi_sector
= sector
+ rdev
->data_offset
;
814 bio_add_page(bio
, page
, size
, 0);
815 init_completion(&event
);
816 bio
->bi_private
= &event
;
817 bio
->bi_end_io
= bi_complete
;
819 wait_for_completion(&event
);
821 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
825 EXPORT_SYMBOL_GPL(sync_page_io
);
827 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
829 char b
[BDEVNAME_SIZE
];
830 if (!rdev
->sb_page
) {
838 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
844 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
845 bdevname(rdev
->bdev
,b
));
849 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
851 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
852 sb1
->set_uuid1
== sb2
->set_uuid1
&&
853 sb1
->set_uuid2
== sb2
->set_uuid2
&&
854 sb1
->set_uuid3
== sb2
->set_uuid3
;
857 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
860 mdp_super_t
*tmp1
, *tmp2
;
862 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
863 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
865 if (!tmp1
|| !tmp2
) {
867 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
875 * nr_disks is not constant
880 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
888 static u32
md_csum_fold(u32 csum
)
890 csum
= (csum
& 0xffff) + (csum
>> 16);
891 return (csum
& 0xffff) + (csum
>> 16);
894 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
897 u32
*sb32
= (u32
*)sb
;
899 unsigned int disk_csum
, csum
;
901 disk_csum
= sb
->sb_csum
;
904 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
906 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
910 /* This used to use csum_partial, which was wrong for several
911 * reasons including that different results are returned on
912 * different architectures. It isn't critical that we get exactly
913 * the same return value as before (we always csum_fold before
914 * testing, and that removes any differences). However as we
915 * know that csum_partial always returned a 16bit value on
916 * alphas, do a fold to maximise conformity to previous behaviour.
918 sb
->sb_csum
= md_csum_fold(disk_csum
);
920 sb
->sb_csum
= disk_csum
;
927 * Handle superblock details.
928 * We want to be able to handle multiple superblock formats
929 * so we have a common interface to them all, and an array of
930 * different handlers.
931 * We rely on user-space to write the initial superblock, and support
932 * reading and updating of superblocks.
933 * Interface methods are:
934 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
935 * loads and validates a superblock on dev.
936 * if refdev != NULL, compare superblocks on both devices
938 * 0 - dev has a superblock that is compatible with refdev
939 * 1 - dev has a superblock that is compatible and newer than refdev
940 * so dev should be used as the refdev in future
941 * -EINVAL superblock incompatible or invalid
942 * -othererror e.g. -EIO
944 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
945 * Verify that dev is acceptable into mddev.
946 * The first time, mddev->raid_disks will be 0, and data from
947 * dev should be merged in. Subsequent calls check that dev
948 * is new enough. Return 0 or -EINVAL
950 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
951 * Update the superblock for rdev with data in mddev
952 * This does not write to disc.
958 struct module
*owner
;
959 int (*load_super
)(struct md_rdev
*rdev
,
960 struct md_rdev
*refdev
,
962 int (*validate_super
)(struct mddev
*mddev
,
963 struct md_rdev
*rdev
);
964 void (*sync_super
)(struct mddev
*mddev
,
965 struct md_rdev
*rdev
);
966 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
967 sector_t num_sectors
);
968 int (*allow_new_offset
)(struct md_rdev
*rdev
,
969 unsigned long long new_offset
);
973 * Check that the given mddev has no bitmap.
975 * This function is called from the run method of all personalities that do not
976 * support bitmaps. It prints an error message and returns non-zero if mddev
977 * has a bitmap. Otherwise, it returns 0.
980 int md_check_no_bitmap(struct mddev
*mddev
)
982 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
984 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
985 mdname(mddev
), mddev
->pers
->name
);
988 EXPORT_SYMBOL(md_check_no_bitmap
);
991 * load_super for 0.90.0
993 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
995 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1000 * Calculate the position of the superblock (512byte sectors),
1001 * it's at the end of the disk.
1003 * It also happens to be a multiple of 4Kb.
1005 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1007 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1008 if (ret
) return ret
;
1012 bdevname(rdev
->bdev
, b
);
1013 sb
= page_address(rdev
->sb_page
);
1015 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1016 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1021 if (sb
->major_version
!= 0 ||
1022 sb
->minor_version
< 90 ||
1023 sb
->minor_version
> 91) {
1024 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1025 sb
->major_version
, sb
->minor_version
,
1030 if (sb
->raid_disks
<= 0)
1033 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1034 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1039 rdev
->preferred_minor
= sb
->md_minor
;
1040 rdev
->data_offset
= 0;
1041 rdev
->new_data_offset
= 0;
1042 rdev
->sb_size
= MD_SB_BYTES
;
1043 rdev
->badblocks
.shift
= -1;
1045 if (sb
->level
== LEVEL_MULTIPATH
)
1048 rdev
->desc_nr
= sb
->this_disk
.number
;
1054 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1055 if (!uuid_equal(refsb
, sb
)) {
1056 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1057 b
, bdevname(refdev
->bdev
,b2
));
1060 if (!sb_equal(refsb
, sb
)) {
1061 printk(KERN_WARNING
"md: %s has same UUID"
1062 " but different superblock to %s\n",
1063 b
, bdevname(refdev
->bdev
, b2
));
1067 ev2
= md_event(refsb
);
1073 rdev
->sectors
= rdev
->sb_start
;
1074 /* Limit to 4TB as metadata cannot record more than that.
1075 * (not needed for Linear and RAID0 as metadata doesn't
1078 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1079 rdev
->sectors
= (2ULL << 32) - 2;
1081 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1082 /* "this cannot possibly happen" ... */
1090 * validate_super for 0.90.0
1092 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1095 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1096 __u64 ev1
= md_event(sb
);
1098 rdev
->raid_disk
= -1;
1099 clear_bit(Faulty
, &rdev
->flags
);
1100 clear_bit(In_sync
, &rdev
->flags
);
1101 clear_bit(WriteMostly
, &rdev
->flags
);
1103 if (mddev
->raid_disks
== 0) {
1104 mddev
->major_version
= 0;
1105 mddev
->minor_version
= sb
->minor_version
;
1106 mddev
->patch_version
= sb
->patch_version
;
1107 mddev
->external
= 0;
1108 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1109 mddev
->ctime
= sb
->ctime
;
1110 mddev
->utime
= sb
->utime
;
1111 mddev
->level
= sb
->level
;
1112 mddev
->clevel
[0] = 0;
1113 mddev
->layout
= sb
->layout
;
1114 mddev
->raid_disks
= sb
->raid_disks
;
1115 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1116 mddev
->events
= ev1
;
1117 mddev
->bitmap_info
.offset
= 0;
1118 mddev
->bitmap_info
.space
= 0;
1119 /* bitmap can use 60 K after the 4K superblocks */
1120 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1121 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1122 mddev
->reshape_backwards
= 0;
1124 if (mddev
->minor_version
>= 91) {
1125 mddev
->reshape_position
= sb
->reshape_position
;
1126 mddev
->delta_disks
= sb
->delta_disks
;
1127 mddev
->new_level
= sb
->new_level
;
1128 mddev
->new_layout
= sb
->new_layout
;
1129 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1130 if (mddev
->delta_disks
< 0)
1131 mddev
->reshape_backwards
= 1;
1133 mddev
->reshape_position
= MaxSector
;
1134 mddev
->delta_disks
= 0;
1135 mddev
->new_level
= mddev
->level
;
1136 mddev
->new_layout
= mddev
->layout
;
1137 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1140 if (sb
->state
& (1<<MD_SB_CLEAN
))
1141 mddev
->recovery_cp
= MaxSector
;
1143 if (sb
->events_hi
== sb
->cp_events_hi
&&
1144 sb
->events_lo
== sb
->cp_events_lo
) {
1145 mddev
->recovery_cp
= sb
->recovery_cp
;
1147 mddev
->recovery_cp
= 0;
1150 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1151 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1152 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1153 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1155 mddev
->max_disks
= MD_SB_DISKS
;
1157 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1158 mddev
->bitmap_info
.file
== NULL
) {
1159 mddev
->bitmap_info
.offset
=
1160 mddev
->bitmap_info
.default_offset
;
1161 mddev
->bitmap_info
.space
=
1162 mddev
->bitmap_info
.space
;
1165 } else if (mddev
->pers
== NULL
) {
1166 /* Insist on good event counter while assembling, except
1167 * for spares (which don't need an event count) */
1169 if (sb
->disks
[rdev
->desc_nr
].state
& (
1170 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1171 if (ev1
< mddev
->events
)
1173 } else if (mddev
->bitmap
) {
1174 /* if adding to array with a bitmap, then we can accept an
1175 * older device ... but not too old.
1177 if (ev1
< mddev
->bitmap
->events_cleared
)
1180 if (ev1
< mddev
->events
)
1181 /* just a hot-add of a new device, leave raid_disk at -1 */
1185 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1186 desc
= sb
->disks
+ rdev
->desc_nr
;
1188 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1189 set_bit(Faulty
, &rdev
->flags
);
1190 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1191 desc->raid_disk < mddev->raid_disks */) {
1192 set_bit(In_sync
, &rdev
->flags
);
1193 rdev
->raid_disk
= desc
->raid_disk
;
1194 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1195 /* active but not in sync implies recovery up to
1196 * reshape position. We don't know exactly where
1197 * that is, so set to zero for now */
1198 if (mddev
->minor_version
>= 91) {
1199 rdev
->recovery_offset
= 0;
1200 rdev
->raid_disk
= desc
->raid_disk
;
1203 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1204 set_bit(WriteMostly
, &rdev
->flags
);
1205 } else /* MULTIPATH are always insync */
1206 set_bit(In_sync
, &rdev
->flags
);
1211 * sync_super for 0.90.0
1213 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1216 struct md_rdev
*rdev2
;
1217 int next_spare
= mddev
->raid_disks
;
1220 /* make rdev->sb match mddev data..
1223 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1224 * 3/ any empty disks < next_spare become removed
1226 * disks[0] gets initialised to REMOVED because
1227 * we cannot be sure from other fields if it has
1228 * been initialised or not.
1231 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1233 rdev
->sb_size
= MD_SB_BYTES
;
1235 sb
= page_address(rdev
->sb_page
);
1237 memset(sb
, 0, sizeof(*sb
));
1239 sb
->md_magic
= MD_SB_MAGIC
;
1240 sb
->major_version
= mddev
->major_version
;
1241 sb
->patch_version
= mddev
->patch_version
;
1242 sb
->gvalid_words
= 0; /* ignored */
1243 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1244 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1245 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1246 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1248 sb
->ctime
= mddev
->ctime
;
1249 sb
->level
= mddev
->level
;
1250 sb
->size
= mddev
->dev_sectors
/ 2;
1251 sb
->raid_disks
= mddev
->raid_disks
;
1252 sb
->md_minor
= mddev
->md_minor
;
1253 sb
->not_persistent
= 0;
1254 sb
->utime
= mddev
->utime
;
1256 sb
->events_hi
= (mddev
->events
>>32);
1257 sb
->events_lo
= (u32
)mddev
->events
;
1259 if (mddev
->reshape_position
== MaxSector
)
1260 sb
->minor_version
= 90;
1262 sb
->minor_version
= 91;
1263 sb
->reshape_position
= mddev
->reshape_position
;
1264 sb
->new_level
= mddev
->new_level
;
1265 sb
->delta_disks
= mddev
->delta_disks
;
1266 sb
->new_layout
= mddev
->new_layout
;
1267 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1269 mddev
->minor_version
= sb
->minor_version
;
1272 sb
->recovery_cp
= mddev
->recovery_cp
;
1273 sb
->cp_events_hi
= (mddev
->events
>>32);
1274 sb
->cp_events_lo
= (u32
)mddev
->events
;
1275 if (mddev
->recovery_cp
== MaxSector
)
1276 sb
->state
= (1<< MD_SB_CLEAN
);
1278 sb
->recovery_cp
= 0;
1280 sb
->layout
= mddev
->layout
;
1281 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1283 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1284 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1286 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1287 rdev_for_each(rdev2
, mddev
) {
1290 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1292 if (rdev2
->raid_disk
>= 0 &&
1293 sb
->minor_version
>= 91)
1294 /* we have nowhere to store the recovery_offset,
1295 * but if it is not below the reshape_position,
1296 * we can piggy-back on that.
1299 if (rdev2
->raid_disk
< 0 ||
1300 test_bit(Faulty
, &rdev2
->flags
))
1303 desc_nr
= rdev2
->raid_disk
;
1305 desc_nr
= next_spare
++;
1306 rdev2
->desc_nr
= desc_nr
;
1307 d
= &sb
->disks
[rdev2
->desc_nr
];
1309 d
->number
= rdev2
->desc_nr
;
1310 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1311 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1313 d
->raid_disk
= rdev2
->raid_disk
;
1315 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1316 if (test_bit(Faulty
, &rdev2
->flags
))
1317 d
->state
= (1<<MD_DISK_FAULTY
);
1318 else if (is_active
) {
1319 d
->state
= (1<<MD_DISK_ACTIVE
);
1320 if (test_bit(In_sync
, &rdev2
->flags
))
1321 d
->state
|= (1<<MD_DISK_SYNC
);
1329 if (test_bit(WriteMostly
, &rdev2
->flags
))
1330 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1332 /* now set the "removed" and "faulty" bits on any missing devices */
1333 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1334 mdp_disk_t
*d
= &sb
->disks
[i
];
1335 if (d
->state
== 0 && d
->number
== 0) {
1338 d
->state
= (1<<MD_DISK_REMOVED
);
1339 d
->state
|= (1<<MD_DISK_FAULTY
);
1343 sb
->nr_disks
= nr_disks
;
1344 sb
->active_disks
= active
;
1345 sb
->working_disks
= working
;
1346 sb
->failed_disks
= failed
;
1347 sb
->spare_disks
= spare
;
1349 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1350 sb
->sb_csum
= calc_sb_csum(sb
);
1354 * rdev_size_change for 0.90.0
1356 static unsigned long long
1357 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1359 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1360 return 0; /* component must fit device */
1361 if (rdev
->mddev
->bitmap_info
.offset
)
1362 return 0; /* can't move bitmap */
1363 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1364 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1365 num_sectors
= rdev
->sb_start
;
1366 /* Limit to 4TB as metadata cannot record more than that.
1367 * 4TB == 2^32 KB, or 2*2^32 sectors.
1369 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1370 num_sectors
= (2ULL << 32) - 2;
1371 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1373 md_super_wait(rdev
->mddev
);
1378 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1380 /* non-zero offset changes not possible with v0.90 */
1381 return new_offset
== 0;
1385 * version 1 superblock
1388 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1392 unsigned long long newcsum
;
1393 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1394 __le32
*isuper
= (__le32
*)sb
;
1397 disk_csum
= sb
->sb_csum
;
1400 for (i
=0; size
>=4; size
-= 4 )
1401 newcsum
+= le32_to_cpu(*isuper
++);
1404 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1406 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1407 sb
->sb_csum
= disk_csum
;
1408 return cpu_to_le32(csum
);
1411 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1413 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1415 struct mdp_superblock_1
*sb
;
1419 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1423 * Calculate the position of the superblock in 512byte sectors.
1424 * It is always aligned to a 4K boundary and
1425 * depeding on minor_version, it can be:
1426 * 0: At least 8K, but less than 12K, from end of device
1427 * 1: At start of device
1428 * 2: 4K from start of device.
1430 switch(minor_version
) {
1432 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1434 sb_start
&= ~(sector_t
)(4*2-1);
1445 rdev
->sb_start
= sb_start
;
1447 /* superblock is rarely larger than 1K, but it can be larger,
1448 * and it is safe to read 4k, so we do that
1450 ret
= read_disk_sb(rdev
, 4096);
1451 if (ret
) return ret
;
1454 sb
= page_address(rdev
->sb_page
);
1456 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1457 sb
->major_version
!= cpu_to_le32(1) ||
1458 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1459 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1460 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1463 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1464 printk("md: invalid superblock checksum on %s\n",
1465 bdevname(rdev
->bdev
,b
));
1468 if (le64_to_cpu(sb
->data_size
) < 10) {
1469 printk("md: data_size too small on %s\n",
1470 bdevname(rdev
->bdev
,b
));
1475 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1476 /* Some padding is non-zero, might be a new feature */
1479 rdev
->preferred_minor
= 0xffff;
1480 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1481 rdev
->new_data_offset
= rdev
->data_offset
;
1482 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1483 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1484 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1485 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1487 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1488 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1489 if (rdev
->sb_size
& bmask
)
1490 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1493 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1496 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1499 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1502 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1504 if (!rdev
->bb_page
) {
1505 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1509 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1510 rdev
->badblocks
.count
== 0) {
1511 /* need to load the bad block list.
1512 * Currently we limit it to one page.
1518 int sectors
= le16_to_cpu(sb
->bblog_size
);
1519 if (sectors
> (PAGE_SIZE
/ 512))
1521 offset
= le32_to_cpu(sb
->bblog_offset
);
1524 bb_sector
= (long long)offset
;
1525 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1526 rdev
->bb_page
, READ
, true))
1528 bbp
= (u64
*)page_address(rdev
->bb_page
);
1529 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1530 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1531 u64 bb
= le64_to_cpu(*bbp
);
1532 int count
= bb
& (0x3ff);
1533 u64 sector
= bb
>> 10;
1534 sector
<<= sb
->bblog_shift
;
1535 count
<<= sb
->bblog_shift
;
1538 if (md_set_badblocks(&rdev
->badblocks
,
1539 sector
, count
, 1) == 0)
1542 } else if (sb
->bblog_offset
== 0)
1543 rdev
->badblocks
.shift
= -1;
1549 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1551 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1552 sb
->level
!= refsb
->level
||
1553 sb
->layout
!= refsb
->layout
||
1554 sb
->chunksize
!= refsb
->chunksize
) {
1555 printk(KERN_WARNING
"md: %s has strangely different"
1556 " superblock to %s\n",
1557 bdevname(rdev
->bdev
,b
),
1558 bdevname(refdev
->bdev
,b2
));
1561 ev1
= le64_to_cpu(sb
->events
);
1562 ev2
= le64_to_cpu(refsb
->events
);
1569 if (minor_version
) {
1570 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1571 sectors
-= rdev
->data_offset
;
1573 sectors
= rdev
->sb_start
;
1574 if (sectors
< le64_to_cpu(sb
->data_size
))
1576 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1580 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1582 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1583 __u64 ev1
= le64_to_cpu(sb
->events
);
1585 rdev
->raid_disk
= -1;
1586 clear_bit(Faulty
, &rdev
->flags
);
1587 clear_bit(In_sync
, &rdev
->flags
);
1588 clear_bit(WriteMostly
, &rdev
->flags
);
1590 if (mddev
->raid_disks
== 0) {
1591 mddev
->major_version
= 1;
1592 mddev
->patch_version
= 0;
1593 mddev
->external
= 0;
1594 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1595 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1596 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1597 mddev
->level
= le32_to_cpu(sb
->level
);
1598 mddev
->clevel
[0] = 0;
1599 mddev
->layout
= le32_to_cpu(sb
->layout
);
1600 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1601 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1602 mddev
->events
= ev1
;
1603 mddev
->bitmap_info
.offset
= 0;
1604 mddev
->bitmap_info
.space
= 0;
1605 /* Default location for bitmap is 1K after superblock
1606 * using 3K - total of 4K
1608 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1609 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1610 mddev
->reshape_backwards
= 0;
1612 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1613 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1615 mddev
->max_disks
= (4096-256)/2;
1617 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1618 mddev
->bitmap_info
.file
== NULL
) {
1619 mddev
->bitmap_info
.offset
=
1620 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1621 /* Metadata doesn't record how much space is available.
1622 * For 1.0, we assume we can use up to the superblock
1623 * if before, else to 4K beyond superblock.
1624 * For others, assume no change is possible.
1626 if (mddev
->minor_version
> 0)
1627 mddev
->bitmap_info
.space
= 0;
1628 else if (mddev
->bitmap_info
.offset
> 0)
1629 mddev
->bitmap_info
.space
=
1630 8 - mddev
->bitmap_info
.offset
;
1632 mddev
->bitmap_info
.space
=
1633 -mddev
->bitmap_info
.offset
;
1636 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1637 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1638 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1639 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1640 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1641 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1642 if (mddev
->delta_disks
< 0 ||
1643 (mddev
->delta_disks
== 0 &&
1644 (le32_to_cpu(sb
->feature_map
)
1645 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1646 mddev
->reshape_backwards
= 1;
1648 mddev
->reshape_position
= MaxSector
;
1649 mddev
->delta_disks
= 0;
1650 mddev
->new_level
= mddev
->level
;
1651 mddev
->new_layout
= mddev
->layout
;
1652 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1655 } else if (mddev
->pers
== NULL
) {
1656 /* Insist of good event counter while assembling, except for
1657 * spares (which don't need an event count) */
1659 if (rdev
->desc_nr
>= 0 &&
1660 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1661 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1662 if (ev1
< mddev
->events
)
1664 } else if (mddev
->bitmap
) {
1665 /* If adding to array with a bitmap, then we can accept an
1666 * older device, but not too old.
1668 if (ev1
< mddev
->bitmap
->events_cleared
)
1671 if (ev1
< mddev
->events
)
1672 /* just a hot-add of a new device, leave raid_disk at -1 */
1675 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1677 if (rdev
->desc_nr
< 0 ||
1678 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1682 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1684 case 0xffff: /* spare */
1686 case 0xfffe: /* faulty */
1687 set_bit(Faulty
, &rdev
->flags
);
1690 if ((le32_to_cpu(sb
->feature_map
) &
1691 MD_FEATURE_RECOVERY_OFFSET
))
1692 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1694 set_bit(In_sync
, &rdev
->flags
);
1695 rdev
->raid_disk
= role
;
1698 if (sb
->devflags
& WriteMostly1
)
1699 set_bit(WriteMostly
, &rdev
->flags
);
1700 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1701 set_bit(Replacement
, &rdev
->flags
);
1702 } else /* MULTIPATH are always insync */
1703 set_bit(In_sync
, &rdev
->flags
);
1708 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1710 struct mdp_superblock_1
*sb
;
1711 struct md_rdev
*rdev2
;
1713 /* make rdev->sb match mddev and rdev data. */
1715 sb
= page_address(rdev
->sb_page
);
1717 sb
->feature_map
= 0;
1719 sb
->recovery_offset
= cpu_to_le64(0);
1720 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1722 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1723 sb
->events
= cpu_to_le64(mddev
->events
);
1725 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1727 sb
->resync_offset
= cpu_to_le64(0);
1729 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1731 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1732 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1733 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1734 sb
->level
= cpu_to_le32(mddev
->level
);
1735 sb
->layout
= cpu_to_le32(mddev
->layout
);
1737 if (test_bit(WriteMostly
, &rdev
->flags
))
1738 sb
->devflags
|= WriteMostly1
;
1740 sb
->devflags
&= ~WriteMostly1
;
1741 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1742 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1744 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1745 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1746 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1749 if (rdev
->raid_disk
>= 0 &&
1750 !test_bit(In_sync
, &rdev
->flags
)) {
1752 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1753 sb
->recovery_offset
=
1754 cpu_to_le64(rdev
->recovery_offset
);
1756 if (test_bit(Replacement
, &rdev
->flags
))
1758 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1760 if (mddev
->reshape_position
!= MaxSector
) {
1761 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1762 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1763 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1764 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1765 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1766 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1767 if (mddev
->delta_disks
== 0 &&
1768 mddev
->reshape_backwards
)
1770 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1771 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1773 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1774 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1775 - rdev
->data_offset
));
1779 if (rdev
->badblocks
.count
== 0)
1780 /* Nothing to do for bad blocks*/ ;
1781 else if (sb
->bblog_offset
== 0)
1782 /* Cannot record bad blocks on this device */
1783 md_error(mddev
, rdev
);
1785 struct badblocks
*bb
= &rdev
->badblocks
;
1786 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1788 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1793 seq
= read_seqbegin(&bb
->lock
);
1795 memset(bbp
, 0xff, PAGE_SIZE
);
1797 for (i
= 0 ; i
< bb
->count
; i
++) {
1798 u64 internal_bb
= *p
++;
1799 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1800 | BB_LEN(internal_bb
));
1801 *bbp
++ = cpu_to_le64(store_bb
);
1804 if (read_seqretry(&bb
->lock
, seq
))
1807 bb
->sector
= (rdev
->sb_start
+
1808 (int)le32_to_cpu(sb
->bblog_offset
));
1809 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1814 rdev_for_each(rdev2
, mddev
)
1815 if (rdev2
->desc_nr
+1 > max_dev
)
1816 max_dev
= rdev2
->desc_nr
+1;
1818 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1820 sb
->max_dev
= cpu_to_le32(max_dev
);
1821 rdev
->sb_size
= max_dev
* 2 + 256;
1822 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1823 if (rdev
->sb_size
& bmask
)
1824 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1826 max_dev
= le32_to_cpu(sb
->max_dev
);
1828 for (i
=0; i
<max_dev
;i
++)
1829 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1831 rdev_for_each(rdev2
, mddev
) {
1833 if (test_bit(Faulty
, &rdev2
->flags
))
1834 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1835 else if (test_bit(In_sync
, &rdev2
->flags
))
1836 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1837 else if (rdev2
->raid_disk
>= 0)
1838 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1840 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1843 sb
->sb_csum
= calc_sb_1_csum(sb
);
1846 static unsigned long long
1847 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1849 struct mdp_superblock_1
*sb
;
1850 sector_t max_sectors
;
1851 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1852 return 0; /* component must fit device */
1853 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1854 return 0; /* too confusing */
1855 if (rdev
->sb_start
< rdev
->data_offset
) {
1856 /* minor versions 1 and 2; superblock before data */
1857 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1858 max_sectors
-= rdev
->data_offset
;
1859 if (!num_sectors
|| num_sectors
> max_sectors
)
1860 num_sectors
= max_sectors
;
1861 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1862 /* minor version 0 with bitmap we can't move */
1865 /* minor version 0; superblock after data */
1867 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1868 sb_start
&= ~(sector_t
)(4*2 - 1);
1869 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1870 if (!num_sectors
|| num_sectors
> max_sectors
)
1871 num_sectors
= max_sectors
;
1872 rdev
->sb_start
= sb_start
;
1874 sb
= page_address(rdev
->sb_page
);
1875 sb
->data_size
= cpu_to_le64(num_sectors
);
1876 sb
->super_offset
= rdev
->sb_start
;
1877 sb
->sb_csum
= calc_sb_1_csum(sb
);
1878 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1880 md_super_wait(rdev
->mddev
);
1886 super_1_allow_new_offset(struct md_rdev
*rdev
,
1887 unsigned long long new_offset
)
1889 /* All necessary checks on new >= old have been done */
1890 struct bitmap
*bitmap
;
1891 if (new_offset
>= rdev
->data_offset
)
1894 /* with 1.0 metadata, there is no metadata to tread on
1895 * so we can always move back */
1896 if (rdev
->mddev
->minor_version
== 0)
1899 /* otherwise we must be sure not to step on
1900 * any metadata, so stay:
1901 * 36K beyond start of superblock
1902 * beyond end of badblocks
1903 * beyond write-intent bitmap
1905 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1907 bitmap
= rdev
->mddev
->bitmap
;
1908 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1909 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1910 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1912 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1918 static struct super_type super_types
[] = {
1921 .owner
= THIS_MODULE
,
1922 .load_super
= super_90_load
,
1923 .validate_super
= super_90_validate
,
1924 .sync_super
= super_90_sync
,
1925 .rdev_size_change
= super_90_rdev_size_change
,
1926 .allow_new_offset
= super_90_allow_new_offset
,
1930 .owner
= THIS_MODULE
,
1931 .load_super
= super_1_load
,
1932 .validate_super
= super_1_validate
,
1933 .sync_super
= super_1_sync
,
1934 .rdev_size_change
= super_1_rdev_size_change
,
1935 .allow_new_offset
= super_1_allow_new_offset
,
1939 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1941 if (mddev
->sync_super
) {
1942 mddev
->sync_super(mddev
, rdev
);
1946 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1948 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1951 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1953 struct md_rdev
*rdev
, *rdev2
;
1956 rdev_for_each_rcu(rdev
, mddev1
)
1957 rdev_for_each_rcu(rdev2
, mddev2
)
1958 if (rdev
->bdev
->bd_contains
==
1959 rdev2
->bdev
->bd_contains
) {
1967 static LIST_HEAD(pending_raid_disks
);
1970 * Try to register data integrity profile for an mddev
1972 * This is called when an array is started and after a disk has been kicked
1973 * from the array. It only succeeds if all working and active component devices
1974 * are integrity capable with matching profiles.
1976 int md_integrity_register(struct mddev
*mddev
)
1978 struct md_rdev
*rdev
, *reference
= NULL
;
1980 if (list_empty(&mddev
->disks
))
1981 return 0; /* nothing to do */
1982 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1983 return 0; /* shouldn't register, or already is */
1984 rdev_for_each(rdev
, mddev
) {
1985 /* skip spares and non-functional disks */
1986 if (test_bit(Faulty
, &rdev
->flags
))
1988 if (rdev
->raid_disk
< 0)
1991 /* Use the first rdev as the reference */
1995 /* does this rdev's profile match the reference profile? */
1996 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1997 rdev
->bdev
->bd_disk
) < 0)
2000 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2003 * All component devices are integrity capable and have matching
2004 * profiles, register the common profile for the md device.
2006 if (blk_integrity_register(mddev
->gendisk
,
2007 bdev_get_integrity(reference
->bdev
)) != 0) {
2008 printk(KERN_ERR
"md: failed to register integrity for %s\n",
2012 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2013 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2014 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2020 EXPORT_SYMBOL(md_integrity_register
);
2022 /* Disable data integrity if non-capable/non-matching disk is being added */
2023 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2025 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2026 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2028 if (!bi_mddev
) /* nothing to do */
2030 if (rdev
->raid_disk
< 0) /* skip spares */
2032 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2033 rdev
->bdev
->bd_disk
) >= 0)
2035 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2036 blk_integrity_unregister(mddev
->gendisk
);
2038 EXPORT_SYMBOL(md_integrity_add_rdev
);
2040 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2042 char b
[BDEVNAME_SIZE
];
2052 /* prevent duplicates */
2053 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2056 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2057 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2058 rdev
->sectors
< mddev
->dev_sectors
)) {
2060 /* Cannot change size, so fail
2061 * If mddev->level <= 0, then we don't care
2062 * about aligning sizes (e.g. linear)
2064 if (mddev
->level
> 0)
2067 mddev
->dev_sectors
= rdev
->sectors
;
2070 /* Verify rdev->desc_nr is unique.
2071 * If it is -1, assign a free number, else
2072 * check number is not in use
2074 if (rdev
->desc_nr
< 0) {
2076 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2077 while (find_rdev_nr(mddev
, choice
))
2079 rdev
->desc_nr
= choice
;
2081 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2084 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2085 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2086 mdname(mddev
), mddev
->max_disks
);
2089 bdevname(rdev
->bdev
,b
);
2090 while ( (s
=strchr(b
, '/')) != NULL
)
2093 rdev
->mddev
= mddev
;
2094 printk(KERN_INFO
"md: bind<%s>\n", b
);
2096 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2099 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2100 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2101 /* failure here is OK */;
2102 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2104 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2105 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2107 /* May as well allow recovery to be retried once */
2108 mddev
->recovery_disabled
++;
2113 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2118 static void md_delayed_delete(struct work_struct
*ws
)
2120 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2121 kobject_del(&rdev
->kobj
);
2122 kobject_put(&rdev
->kobj
);
2125 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2127 char b
[BDEVNAME_SIZE
];
2132 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2133 list_del_rcu(&rdev
->same_set
);
2134 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2136 sysfs_remove_link(&rdev
->kobj
, "block");
2137 sysfs_put(rdev
->sysfs_state
);
2138 rdev
->sysfs_state
= NULL
;
2139 rdev
->badblocks
.count
= 0;
2140 /* We need to delay this, otherwise we can deadlock when
2141 * writing to 'remove' to "dev/state". We also need
2142 * to delay it due to rcu usage.
2145 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2146 kobject_get(&rdev
->kobj
);
2147 queue_work(md_misc_wq
, &rdev
->del_work
);
2151 * prevent the device from being mounted, repartitioned or
2152 * otherwise reused by a RAID array (or any other kernel
2153 * subsystem), by bd_claiming the device.
2155 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2158 struct block_device
*bdev
;
2159 char b
[BDEVNAME_SIZE
];
2161 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2162 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2164 printk(KERN_ERR
"md: could not open %s.\n",
2165 __bdevname(dev
, b
));
2166 return PTR_ERR(bdev
);
2172 static void unlock_rdev(struct md_rdev
*rdev
)
2174 struct block_device
*bdev
= rdev
->bdev
;
2178 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2181 void md_autodetect_dev(dev_t dev
);
2183 static void export_rdev(struct md_rdev
* rdev
)
2185 char b
[BDEVNAME_SIZE
];
2186 printk(KERN_INFO
"md: export_rdev(%s)\n",
2187 bdevname(rdev
->bdev
,b
));
2190 md_rdev_clear(rdev
);
2192 if (test_bit(AutoDetected
, &rdev
->flags
))
2193 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2196 kobject_put(&rdev
->kobj
);
2199 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2201 unbind_rdev_from_array(rdev
);
2205 static void export_array(struct mddev
*mddev
)
2207 struct md_rdev
*rdev
, *tmp
;
2209 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2214 kick_rdev_from_array(rdev
);
2216 if (!list_empty(&mddev
->disks
))
2218 mddev
->raid_disks
= 0;
2219 mddev
->major_version
= 0;
2222 static void print_desc(mdp_disk_t
*desc
)
2224 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2225 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2228 static void print_sb_90(mdp_super_t
*sb
)
2233 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2234 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2235 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2237 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2238 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2239 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2240 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2241 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2242 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2243 sb
->failed_disks
, sb
->spare_disks
,
2244 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2247 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2250 desc
= sb
->disks
+ i
;
2251 if (desc
->number
|| desc
->major
|| desc
->minor
||
2252 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2253 printk(" D %2d: ", i
);
2257 printk(KERN_INFO
"md: THIS: ");
2258 print_desc(&sb
->this_disk
);
2261 static void print_sb_1(struct mdp_superblock_1
*sb
)
2265 uuid
= sb
->set_uuid
;
2267 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2268 "md: Name: \"%s\" CT:%llu\n",
2269 le32_to_cpu(sb
->major_version
),
2270 le32_to_cpu(sb
->feature_map
),
2273 (unsigned long long)le64_to_cpu(sb
->ctime
)
2274 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2276 uuid
= sb
->device_uuid
;
2278 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2280 "md: Dev:%08x UUID: %pU\n"
2281 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2282 "md: (MaxDev:%u) \n",
2283 le32_to_cpu(sb
->level
),
2284 (unsigned long long)le64_to_cpu(sb
->size
),
2285 le32_to_cpu(sb
->raid_disks
),
2286 le32_to_cpu(sb
->layout
),
2287 le32_to_cpu(sb
->chunksize
),
2288 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2289 (unsigned long long)le64_to_cpu(sb
->data_size
),
2290 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2291 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2292 le32_to_cpu(sb
->dev_number
),
2295 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2296 (unsigned long long)le64_to_cpu(sb
->events
),
2297 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2298 le32_to_cpu(sb
->sb_csum
),
2299 le32_to_cpu(sb
->max_dev
)
2303 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2305 char b
[BDEVNAME_SIZE
];
2306 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2307 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2308 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2310 if (rdev
->sb_loaded
) {
2311 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2312 switch (major_version
) {
2314 print_sb_90(page_address(rdev
->sb_page
));
2317 print_sb_1(page_address(rdev
->sb_page
));
2321 printk(KERN_INFO
"md: no rdev superblock!\n");
2324 static void md_print_devices(void)
2326 struct list_head
*tmp
;
2327 struct md_rdev
*rdev
;
2328 struct mddev
*mddev
;
2329 char b
[BDEVNAME_SIZE
];
2332 printk("md: **********************************\n");
2333 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2334 printk("md: **********************************\n");
2335 for_each_mddev(mddev
, tmp
) {
2338 bitmap_print_sb(mddev
->bitmap
);
2340 printk("%s: ", mdname(mddev
));
2341 rdev_for_each(rdev
, mddev
)
2342 printk("<%s>", bdevname(rdev
->bdev
,b
));
2345 rdev_for_each(rdev
, mddev
)
2346 print_rdev(rdev
, mddev
->major_version
);
2348 printk("md: **********************************\n");
2353 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2355 /* Update each superblock (in-memory image), but
2356 * if we are allowed to, skip spares which already
2357 * have the right event counter, or have one earlier
2358 * (which would mean they aren't being marked as dirty
2359 * with the rest of the array)
2361 struct md_rdev
*rdev
;
2362 rdev_for_each(rdev
, mddev
) {
2363 if (rdev
->sb_events
== mddev
->events
||
2365 rdev
->raid_disk
< 0 &&
2366 rdev
->sb_events
+1 == mddev
->events
)) {
2367 /* Don't update this superblock */
2368 rdev
->sb_loaded
= 2;
2370 sync_super(mddev
, rdev
);
2371 rdev
->sb_loaded
= 1;
2376 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2378 struct md_rdev
*rdev
;
2381 int any_badblocks_changed
= 0;
2384 /* First make sure individual recovery_offsets are correct */
2385 rdev_for_each(rdev
, mddev
) {
2386 if (rdev
->raid_disk
>= 0 &&
2387 mddev
->delta_disks
>= 0 &&
2388 !test_bit(In_sync
, &rdev
->flags
) &&
2389 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2390 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2393 if (!mddev
->persistent
) {
2394 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2395 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2396 if (!mddev
->external
) {
2397 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2398 rdev_for_each(rdev
, mddev
) {
2399 if (rdev
->badblocks
.changed
) {
2400 rdev
->badblocks
.changed
= 0;
2401 md_ack_all_badblocks(&rdev
->badblocks
);
2402 md_error(mddev
, rdev
);
2404 clear_bit(Blocked
, &rdev
->flags
);
2405 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2406 wake_up(&rdev
->blocked_wait
);
2409 wake_up(&mddev
->sb_wait
);
2413 spin_lock_irq(&mddev
->write_lock
);
2415 mddev
->utime
= get_seconds();
2417 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2419 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2420 /* just a clean<-> dirty transition, possibly leave spares alone,
2421 * though if events isn't the right even/odd, we will have to do
2427 if (mddev
->degraded
)
2428 /* If the array is degraded, then skipping spares is both
2429 * dangerous and fairly pointless.
2430 * Dangerous because a device that was removed from the array
2431 * might have a event_count that still looks up-to-date,
2432 * so it can be re-added without a resync.
2433 * Pointless because if there are any spares to skip,
2434 * then a recovery will happen and soon that array won't
2435 * be degraded any more and the spare can go back to sleep then.
2439 sync_req
= mddev
->in_sync
;
2441 /* If this is just a dirty<->clean transition, and the array is clean
2442 * and 'events' is odd, we can roll back to the previous clean state */
2444 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2445 && mddev
->can_decrease_events
2446 && mddev
->events
!= 1) {
2448 mddev
->can_decrease_events
= 0;
2450 /* otherwise we have to go forward and ... */
2452 mddev
->can_decrease_events
= nospares
;
2455 if (!mddev
->events
) {
2457 * oops, this 64-bit counter should never wrap.
2458 * Either we are in around ~1 trillion A.C., assuming
2459 * 1 reboot per second, or we have a bug:
2465 rdev_for_each(rdev
, mddev
) {
2466 if (rdev
->badblocks
.changed
)
2467 any_badblocks_changed
++;
2468 if (test_bit(Faulty
, &rdev
->flags
))
2469 set_bit(FaultRecorded
, &rdev
->flags
);
2472 sync_sbs(mddev
, nospares
);
2473 spin_unlock_irq(&mddev
->write_lock
);
2475 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2476 mdname(mddev
), mddev
->in_sync
);
2478 bitmap_update_sb(mddev
->bitmap
);
2479 rdev_for_each(rdev
, mddev
) {
2480 char b
[BDEVNAME_SIZE
];
2482 if (rdev
->sb_loaded
!= 1)
2483 continue; /* no noise on spare devices */
2485 if (!test_bit(Faulty
, &rdev
->flags
) &&
2486 rdev
->saved_raid_disk
== -1) {
2487 md_super_write(mddev
,rdev
,
2488 rdev
->sb_start
, rdev
->sb_size
,
2490 pr_debug("md: (write) %s's sb offset: %llu\n",
2491 bdevname(rdev
->bdev
, b
),
2492 (unsigned long long)rdev
->sb_start
);
2493 rdev
->sb_events
= mddev
->events
;
2494 if (rdev
->badblocks
.size
) {
2495 md_super_write(mddev
, rdev
,
2496 rdev
->badblocks
.sector
,
2497 rdev
->badblocks
.size
<< 9,
2499 rdev
->badblocks
.size
= 0;
2502 } else if (test_bit(Faulty
, &rdev
->flags
))
2503 pr_debug("md: %s (skipping faulty)\n",
2504 bdevname(rdev
->bdev
, b
));
2506 pr_debug("(skipping incremental s/r ");
2508 if (mddev
->level
== LEVEL_MULTIPATH
)
2509 /* only need to write one superblock... */
2512 md_super_wait(mddev
);
2513 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2515 spin_lock_irq(&mddev
->write_lock
);
2516 if (mddev
->in_sync
!= sync_req
||
2517 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2518 /* have to write it out again */
2519 spin_unlock_irq(&mddev
->write_lock
);
2522 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2523 spin_unlock_irq(&mddev
->write_lock
);
2524 wake_up(&mddev
->sb_wait
);
2525 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2526 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2528 rdev_for_each(rdev
, mddev
) {
2529 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2530 clear_bit(Blocked
, &rdev
->flags
);
2532 if (any_badblocks_changed
)
2533 md_ack_all_badblocks(&rdev
->badblocks
);
2534 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2535 wake_up(&rdev
->blocked_wait
);
2539 /* words written to sysfs files may, or may not, be \n terminated.
2540 * We want to accept with case. For this we use cmd_match.
2542 static int cmd_match(const char *cmd
, const char *str
)
2544 /* See if cmd, written into a sysfs file, matches
2545 * str. They must either be the same, or cmd can
2546 * have a trailing newline
2548 while (*cmd
&& *str
&& *cmd
== *str
) {
2559 struct rdev_sysfs_entry
{
2560 struct attribute attr
;
2561 ssize_t (*show
)(struct md_rdev
*, char *);
2562 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2566 state_show(struct md_rdev
*rdev
, char *page
)
2571 if (test_bit(Faulty
, &rdev
->flags
) ||
2572 rdev
->badblocks
.unacked_exist
) {
2573 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2576 if (test_bit(In_sync
, &rdev
->flags
)) {
2577 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2580 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2581 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2584 if (test_bit(Blocked
, &rdev
->flags
) ||
2585 (rdev
->badblocks
.unacked_exist
2586 && !test_bit(Faulty
, &rdev
->flags
))) {
2587 len
+= sprintf(page
+len
, "%sblocked", sep
);
2590 if (!test_bit(Faulty
, &rdev
->flags
) &&
2591 !test_bit(In_sync
, &rdev
->flags
)) {
2592 len
+= sprintf(page
+len
, "%sspare", sep
);
2595 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2596 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2599 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2600 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2603 if (test_bit(Replacement
, &rdev
->flags
)) {
2604 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2608 return len
+sprintf(page
+len
, "\n");
2612 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2615 * faulty - simulates an error
2616 * remove - disconnects the device
2617 * writemostly - sets write_mostly
2618 * -writemostly - clears write_mostly
2619 * blocked - sets the Blocked flags
2620 * -blocked - clears the Blocked and possibly simulates an error
2621 * insync - sets Insync providing device isn't active
2622 * write_error - sets WriteErrorSeen
2623 * -write_error - clears WriteErrorSeen
2626 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2627 md_error(rdev
->mddev
, rdev
);
2628 if (test_bit(Faulty
, &rdev
->flags
))
2632 } else if (cmd_match(buf
, "remove")) {
2633 if (rdev
->raid_disk
>= 0)
2636 struct mddev
*mddev
= rdev
->mddev
;
2637 kick_rdev_from_array(rdev
);
2639 md_update_sb(mddev
, 1);
2640 md_new_event(mddev
);
2643 } else if (cmd_match(buf
, "writemostly")) {
2644 set_bit(WriteMostly
, &rdev
->flags
);
2646 } else if (cmd_match(buf
, "-writemostly")) {
2647 clear_bit(WriteMostly
, &rdev
->flags
);
2649 } else if (cmd_match(buf
, "blocked")) {
2650 set_bit(Blocked
, &rdev
->flags
);
2652 } else if (cmd_match(buf
, "-blocked")) {
2653 if (!test_bit(Faulty
, &rdev
->flags
) &&
2654 rdev
->badblocks
.unacked_exist
) {
2655 /* metadata handler doesn't understand badblocks,
2656 * so we need to fail the device
2658 md_error(rdev
->mddev
, rdev
);
2660 clear_bit(Blocked
, &rdev
->flags
);
2661 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2662 wake_up(&rdev
->blocked_wait
);
2663 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2664 md_wakeup_thread(rdev
->mddev
->thread
);
2667 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2668 set_bit(In_sync
, &rdev
->flags
);
2670 } else if (cmd_match(buf
, "write_error")) {
2671 set_bit(WriteErrorSeen
, &rdev
->flags
);
2673 } else if (cmd_match(buf
, "-write_error")) {
2674 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2676 } else if (cmd_match(buf
, "want_replacement")) {
2677 /* Any non-spare device that is not a replacement can
2678 * become want_replacement at any time, but we then need to
2679 * check if recovery is needed.
2681 if (rdev
->raid_disk
>= 0 &&
2682 !test_bit(Replacement
, &rdev
->flags
))
2683 set_bit(WantReplacement
, &rdev
->flags
);
2684 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2685 md_wakeup_thread(rdev
->mddev
->thread
);
2687 } else if (cmd_match(buf
, "-want_replacement")) {
2688 /* Clearing 'want_replacement' is always allowed.
2689 * Once replacements starts it is too late though.
2692 clear_bit(WantReplacement
, &rdev
->flags
);
2693 } else if (cmd_match(buf
, "replacement")) {
2694 /* Can only set a device as a replacement when array has not
2695 * yet been started. Once running, replacement is automatic
2696 * from spares, or by assigning 'slot'.
2698 if (rdev
->mddev
->pers
)
2701 set_bit(Replacement
, &rdev
->flags
);
2704 } else if (cmd_match(buf
, "-replacement")) {
2705 /* Similarly, can only clear Replacement before start */
2706 if (rdev
->mddev
->pers
)
2709 clear_bit(Replacement
, &rdev
->flags
);
2714 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2715 return err
? err
: len
;
2717 static struct rdev_sysfs_entry rdev_state
=
2718 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2721 errors_show(struct md_rdev
*rdev
, char *page
)
2723 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2727 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2730 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2731 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2732 atomic_set(&rdev
->corrected_errors
, n
);
2737 static struct rdev_sysfs_entry rdev_errors
=
2738 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2741 slot_show(struct md_rdev
*rdev
, char *page
)
2743 if (rdev
->raid_disk
< 0)
2744 return sprintf(page
, "none\n");
2746 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2750 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2754 int slot
= simple_strtoul(buf
, &e
, 10);
2755 if (strncmp(buf
, "none", 4)==0)
2757 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2759 if (rdev
->mddev
->pers
&& slot
== -1) {
2760 /* Setting 'slot' on an active array requires also
2761 * updating the 'rd%d' link, and communicating
2762 * with the personality with ->hot_*_disk.
2763 * For now we only support removing
2764 * failed/spare devices. This normally happens automatically,
2765 * but not when the metadata is externally managed.
2767 if (rdev
->raid_disk
== -1)
2769 /* personality does all needed checks */
2770 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2772 err
= rdev
->mddev
->pers
->
2773 hot_remove_disk(rdev
->mddev
, rdev
);
2776 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2777 rdev
->raid_disk
= -1;
2778 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2779 md_wakeup_thread(rdev
->mddev
->thread
);
2780 } else if (rdev
->mddev
->pers
) {
2781 /* Activating a spare .. or possibly reactivating
2782 * if we ever get bitmaps working here.
2785 if (rdev
->raid_disk
!= -1)
2788 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2791 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2794 if (slot
>= rdev
->mddev
->raid_disks
&&
2795 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2798 rdev
->raid_disk
= slot
;
2799 if (test_bit(In_sync
, &rdev
->flags
))
2800 rdev
->saved_raid_disk
= slot
;
2802 rdev
->saved_raid_disk
= -1;
2803 clear_bit(In_sync
, &rdev
->flags
);
2804 err
= rdev
->mddev
->pers
->
2805 hot_add_disk(rdev
->mddev
, rdev
);
2807 rdev
->raid_disk
= -1;
2810 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2811 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2812 /* failure here is OK */;
2813 /* don't wakeup anyone, leave that to userspace. */
2815 if (slot
>= rdev
->mddev
->raid_disks
&&
2816 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2818 rdev
->raid_disk
= slot
;
2819 /* assume it is working */
2820 clear_bit(Faulty
, &rdev
->flags
);
2821 clear_bit(WriteMostly
, &rdev
->flags
);
2822 set_bit(In_sync
, &rdev
->flags
);
2823 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2829 static struct rdev_sysfs_entry rdev_slot
=
2830 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2833 offset_show(struct md_rdev
*rdev
, char *page
)
2835 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2839 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2841 unsigned long long offset
;
2842 if (strict_strtoull(buf
, 10, &offset
) < 0)
2844 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2846 if (rdev
->sectors
&& rdev
->mddev
->external
)
2847 /* Must set offset before size, so overlap checks
2850 rdev
->data_offset
= offset
;
2851 rdev
->new_data_offset
= offset
;
2855 static struct rdev_sysfs_entry rdev_offset
=
2856 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2858 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2860 return sprintf(page
, "%llu\n",
2861 (unsigned long long)rdev
->new_data_offset
);
2864 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2865 const char *buf
, size_t len
)
2867 unsigned long long new_offset
;
2868 struct mddev
*mddev
= rdev
->mddev
;
2870 if (strict_strtoull(buf
, 10, &new_offset
) < 0)
2873 if (mddev
->sync_thread
)
2875 if (new_offset
== rdev
->data_offset
)
2876 /* reset is always permitted */
2878 else if (new_offset
> rdev
->data_offset
) {
2879 /* must not push array size beyond rdev_sectors */
2880 if (new_offset
- rdev
->data_offset
2881 + mddev
->dev_sectors
> rdev
->sectors
)
2884 /* Metadata worries about other space details. */
2886 /* decreasing the offset is inconsistent with a backwards
2889 if (new_offset
< rdev
->data_offset
&&
2890 mddev
->reshape_backwards
)
2892 /* Increasing offset is inconsistent with forwards
2893 * reshape. reshape_direction should be set to
2894 * 'backwards' first.
2896 if (new_offset
> rdev
->data_offset
&&
2897 !mddev
->reshape_backwards
)
2900 if (mddev
->pers
&& mddev
->persistent
&&
2901 !super_types
[mddev
->major_version
]
2902 .allow_new_offset(rdev
, new_offset
))
2904 rdev
->new_data_offset
= new_offset
;
2905 if (new_offset
> rdev
->data_offset
)
2906 mddev
->reshape_backwards
= 1;
2907 else if (new_offset
< rdev
->data_offset
)
2908 mddev
->reshape_backwards
= 0;
2912 static struct rdev_sysfs_entry rdev_new_offset
=
2913 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2916 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2918 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2921 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2923 /* check if two start/length pairs overlap */
2931 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2933 unsigned long long blocks
;
2936 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2939 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2940 return -EINVAL
; /* sector conversion overflow */
2943 if (new != blocks
* 2)
2944 return -EINVAL
; /* unsigned long long to sector_t overflow */
2951 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2953 struct mddev
*my_mddev
= rdev
->mddev
;
2954 sector_t oldsectors
= rdev
->sectors
;
2957 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2959 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2960 return -EINVAL
; /* too confusing */
2961 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2962 if (my_mddev
->persistent
) {
2963 sectors
= super_types
[my_mddev
->major_version
].
2964 rdev_size_change(rdev
, sectors
);
2967 } else if (!sectors
)
2968 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2971 if (sectors
< my_mddev
->dev_sectors
)
2972 return -EINVAL
; /* component must fit device */
2974 rdev
->sectors
= sectors
;
2975 if (sectors
> oldsectors
&& my_mddev
->external
) {
2976 /* need to check that all other rdevs with the same ->bdev
2977 * do not overlap. We need to unlock the mddev to avoid
2978 * a deadlock. We have already changed rdev->sectors, and if
2979 * we have to change it back, we will have the lock again.
2981 struct mddev
*mddev
;
2983 struct list_head
*tmp
;
2985 mddev_unlock(my_mddev
);
2986 for_each_mddev(mddev
, tmp
) {
2987 struct md_rdev
*rdev2
;
2990 rdev_for_each(rdev2
, mddev
)
2991 if (rdev
->bdev
== rdev2
->bdev
&&
2993 overlaps(rdev
->data_offset
, rdev
->sectors
,
2999 mddev_unlock(mddev
);
3005 mddev_lock(my_mddev
);
3007 /* Someone else could have slipped in a size
3008 * change here, but doing so is just silly.
3009 * We put oldsectors back because we *know* it is
3010 * safe, and trust userspace not to race with
3013 rdev
->sectors
= oldsectors
;
3020 static struct rdev_sysfs_entry rdev_size
=
3021 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3024 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3026 unsigned long long recovery_start
= rdev
->recovery_offset
;
3028 if (test_bit(In_sync
, &rdev
->flags
) ||
3029 recovery_start
== MaxSector
)
3030 return sprintf(page
, "none\n");
3032 return sprintf(page
, "%llu\n", recovery_start
);
3035 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3037 unsigned long long recovery_start
;
3039 if (cmd_match(buf
, "none"))
3040 recovery_start
= MaxSector
;
3041 else if (strict_strtoull(buf
, 10, &recovery_start
))
3044 if (rdev
->mddev
->pers
&&
3045 rdev
->raid_disk
>= 0)
3048 rdev
->recovery_offset
= recovery_start
;
3049 if (recovery_start
== MaxSector
)
3050 set_bit(In_sync
, &rdev
->flags
);
3052 clear_bit(In_sync
, &rdev
->flags
);
3056 static struct rdev_sysfs_entry rdev_recovery_start
=
3057 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3061 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
3063 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
3065 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3067 return badblocks_show(&rdev
->badblocks
, page
, 0);
3069 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3071 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3072 /* Maybe that ack was all we needed */
3073 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3074 wake_up(&rdev
->blocked_wait
);
3077 static struct rdev_sysfs_entry rdev_bad_blocks
=
3078 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3081 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3083 return badblocks_show(&rdev
->badblocks
, page
, 1);
3085 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3087 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3089 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3090 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3092 static struct attribute
*rdev_default_attrs
[] = {
3097 &rdev_new_offset
.attr
,
3099 &rdev_recovery_start
.attr
,
3100 &rdev_bad_blocks
.attr
,
3101 &rdev_unack_bad_blocks
.attr
,
3105 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3107 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3108 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3109 struct mddev
*mddev
= rdev
->mddev
;
3115 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
3117 if (rdev
->mddev
== NULL
)
3120 rv
= entry
->show(rdev
, page
);
3121 mddev_unlock(mddev
);
3127 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3128 const char *page
, size_t length
)
3130 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3131 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3133 struct mddev
*mddev
= rdev
->mddev
;
3137 if (!capable(CAP_SYS_ADMIN
))
3139 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3141 if (rdev
->mddev
== NULL
)
3144 rv
= entry
->store(rdev
, page
, length
);
3145 mddev_unlock(mddev
);
3150 static void rdev_free(struct kobject
*ko
)
3152 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3155 static const struct sysfs_ops rdev_sysfs_ops
= {
3156 .show
= rdev_attr_show
,
3157 .store
= rdev_attr_store
,
3159 static struct kobj_type rdev_ktype
= {
3160 .release
= rdev_free
,
3161 .sysfs_ops
= &rdev_sysfs_ops
,
3162 .default_attrs
= rdev_default_attrs
,
3165 int md_rdev_init(struct md_rdev
*rdev
)
3168 rdev
->saved_raid_disk
= -1;
3169 rdev
->raid_disk
= -1;
3171 rdev
->data_offset
= 0;
3172 rdev
->new_data_offset
= 0;
3173 rdev
->sb_events
= 0;
3174 rdev
->last_read_error
.tv_sec
= 0;
3175 rdev
->last_read_error
.tv_nsec
= 0;
3176 rdev
->sb_loaded
= 0;
3177 rdev
->bb_page
= NULL
;
3178 atomic_set(&rdev
->nr_pending
, 0);
3179 atomic_set(&rdev
->read_errors
, 0);
3180 atomic_set(&rdev
->corrected_errors
, 0);
3182 INIT_LIST_HEAD(&rdev
->same_set
);
3183 init_waitqueue_head(&rdev
->blocked_wait
);
3185 /* Add space to store bad block list.
3186 * This reserves the space even on arrays where it cannot
3187 * be used - I wonder if that matters
3189 rdev
->badblocks
.count
= 0;
3190 rdev
->badblocks
.shift
= 0;
3191 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3192 seqlock_init(&rdev
->badblocks
.lock
);
3193 if (rdev
->badblocks
.page
== NULL
)
3198 EXPORT_SYMBOL_GPL(md_rdev_init
);
3200 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3202 * mark the device faulty if:
3204 * - the device is nonexistent (zero size)
3205 * - the device has no valid superblock
3207 * a faulty rdev _never_ has rdev->sb set.
3209 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3211 char b
[BDEVNAME_SIZE
];
3213 struct md_rdev
*rdev
;
3216 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3218 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3219 return ERR_PTR(-ENOMEM
);
3222 err
= md_rdev_init(rdev
);
3225 err
= alloc_disk_sb(rdev
);
3229 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3233 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3235 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3238 "md: %s has zero or unknown size, marking faulty!\n",
3239 bdevname(rdev
->bdev
,b
));
3244 if (super_format
>= 0) {
3245 err
= super_types
[super_format
].
3246 load_super(rdev
, NULL
, super_minor
);
3247 if (err
== -EINVAL
) {
3249 "md: %s does not have a valid v%d.%d "
3250 "superblock, not importing!\n",
3251 bdevname(rdev
->bdev
,b
),
3252 super_format
, super_minor
);
3257 "md: could not read %s's sb, not importing!\n",
3258 bdevname(rdev
->bdev
,b
));
3262 if (super_format
== -1)
3263 /* hot-add for 0.90, or non-persistent: so no badblocks */
3264 rdev
->badblocks
.shift
= -1;
3271 md_rdev_clear(rdev
);
3273 return ERR_PTR(err
);
3277 * Check a full RAID array for plausibility
3281 static void analyze_sbs(struct mddev
* mddev
)
3284 struct md_rdev
*rdev
, *freshest
, *tmp
;
3285 char b
[BDEVNAME_SIZE
];
3288 rdev_for_each_safe(rdev
, tmp
, mddev
)
3289 switch (super_types
[mddev
->major_version
].
3290 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3298 "md: fatal superblock inconsistency in %s"
3299 " -- removing from array\n",
3300 bdevname(rdev
->bdev
,b
));
3301 kick_rdev_from_array(rdev
);
3305 super_types
[mddev
->major_version
].
3306 validate_super(mddev
, freshest
);
3309 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3310 if (mddev
->max_disks
&&
3311 (rdev
->desc_nr
>= mddev
->max_disks
||
3312 i
> mddev
->max_disks
)) {
3314 "md: %s: %s: only %d devices permitted\n",
3315 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3317 kick_rdev_from_array(rdev
);
3320 if (rdev
!= freshest
)
3321 if (super_types
[mddev
->major_version
].
3322 validate_super(mddev
, rdev
)) {
3323 printk(KERN_WARNING
"md: kicking non-fresh %s"
3325 bdevname(rdev
->bdev
,b
));
3326 kick_rdev_from_array(rdev
);
3329 if (mddev
->level
== LEVEL_MULTIPATH
) {
3330 rdev
->desc_nr
= i
++;
3331 rdev
->raid_disk
= rdev
->desc_nr
;
3332 set_bit(In_sync
, &rdev
->flags
);
3333 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3334 rdev
->raid_disk
= -1;
3335 clear_bit(In_sync
, &rdev
->flags
);
3340 /* Read a fixed-point number.
3341 * Numbers in sysfs attributes should be in "standard" units where
3342 * possible, so time should be in seconds.
3343 * However we internally use a a much smaller unit such as
3344 * milliseconds or jiffies.
3345 * This function takes a decimal number with a possible fractional
3346 * component, and produces an integer which is the result of
3347 * multiplying that number by 10^'scale'.
3348 * all without any floating-point arithmetic.
3350 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3352 unsigned long result
= 0;
3354 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3357 else if (decimals
< scale
) {
3360 result
= result
* 10 + value
;
3372 while (decimals
< scale
) {
3381 static void md_safemode_timeout(unsigned long data
);
3384 safe_delay_show(struct mddev
*mddev
, char *page
)
3386 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3387 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3390 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3394 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3397 mddev
->safemode_delay
= 0;
3399 unsigned long old_delay
= mddev
->safemode_delay
;
3400 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3401 if (mddev
->safemode_delay
== 0)
3402 mddev
->safemode_delay
= 1;
3403 if (mddev
->safemode_delay
< old_delay
)
3404 md_safemode_timeout((unsigned long)mddev
);
3408 static struct md_sysfs_entry md_safe_delay
=
3409 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3412 level_show(struct mddev
*mddev
, char *page
)
3414 struct md_personality
*p
= mddev
->pers
;
3416 return sprintf(page
, "%s\n", p
->name
);
3417 else if (mddev
->clevel
[0])
3418 return sprintf(page
, "%s\n", mddev
->clevel
);
3419 else if (mddev
->level
!= LEVEL_NONE
)
3420 return sprintf(page
, "%d\n", mddev
->level
);
3426 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3430 struct md_personality
*pers
;
3433 struct md_rdev
*rdev
;
3435 if (mddev
->pers
== NULL
) {
3438 if (len
>= sizeof(mddev
->clevel
))
3440 strncpy(mddev
->clevel
, buf
, len
);
3441 if (mddev
->clevel
[len
-1] == '\n')
3443 mddev
->clevel
[len
] = 0;
3444 mddev
->level
= LEVEL_NONE
;
3448 /* request to change the personality. Need to ensure:
3449 * - array is not engaged in resync/recovery/reshape
3450 * - old personality can be suspended
3451 * - new personality will access other array.
3454 if (mddev
->sync_thread
||
3455 mddev
->reshape_position
!= MaxSector
||
3456 mddev
->sysfs_active
)
3459 if (!mddev
->pers
->quiesce
) {
3460 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3461 mdname(mddev
), mddev
->pers
->name
);
3465 /* Now find the new personality */
3466 if (len
== 0 || len
>= sizeof(clevel
))
3468 strncpy(clevel
, buf
, len
);
3469 if (clevel
[len
-1] == '\n')
3472 if (strict_strtol(clevel
, 10, &level
))
3475 if (request_module("md-%s", clevel
) != 0)
3476 request_module("md-level-%s", clevel
);
3477 spin_lock(&pers_lock
);
3478 pers
= find_pers(level
, clevel
);
3479 if (!pers
|| !try_module_get(pers
->owner
)) {
3480 spin_unlock(&pers_lock
);
3481 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3484 spin_unlock(&pers_lock
);
3486 if (pers
== mddev
->pers
) {
3487 /* Nothing to do! */
3488 module_put(pers
->owner
);
3491 if (!pers
->takeover
) {
3492 module_put(pers
->owner
);
3493 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3494 mdname(mddev
), clevel
);
3498 rdev_for_each(rdev
, mddev
)
3499 rdev
->new_raid_disk
= rdev
->raid_disk
;
3501 /* ->takeover must set new_* and/or delta_disks
3502 * if it succeeds, and may set them when it fails.
3504 priv
= pers
->takeover(mddev
);
3506 mddev
->new_level
= mddev
->level
;
3507 mddev
->new_layout
= mddev
->layout
;
3508 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3509 mddev
->raid_disks
-= mddev
->delta_disks
;
3510 mddev
->delta_disks
= 0;
3511 mddev
->reshape_backwards
= 0;
3512 module_put(pers
->owner
);
3513 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3514 mdname(mddev
), clevel
);
3515 return PTR_ERR(priv
);
3518 /* Looks like we have a winner */
3519 mddev_suspend(mddev
);
3520 mddev
->pers
->stop(mddev
);
3522 if (mddev
->pers
->sync_request
== NULL
&&
3523 pers
->sync_request
!= NULL
) {
3524 /* need to add the md_redundancy_group */
3525 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3527 "md: cannot register extra attributes for %s\n",
3529 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3531 if (mddev
->pers
->sync_request
!= NULL
&&
3532 pers
->sync_request
== NULL
) {
3533 /* need to remove the md_redundancy_group */
3534 if (mddev
->to_remove
== NULL
)
3535 mddev
->to_remove
= &md_redundancy_group
;
3538 if (mddev
->pers
->sync_request
== NULL
&&
3540 /* We are converting from a no-redundancy array
3541 * to a redundancy array and metadata is managed
3542 * externally so we need to be sure that writes
3543 * won't block due to a need to transition
3545 * until external management is started.
3548 mddev
->safemode_delay
= 0;
3549 mddev
->safemode
= 0;
3552 rdev_for_each(rdev
, mddev
) {
3553 if (rdev
->raid_disk
< 0)
3555 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3556 rdev
->new_raid_disk
= -1;
3557 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3559 sysfs_unlink_rdev(mddev
, rdev
);
3561 rdev_for_each(rdev
, mddev
) {
3562 if (rdev
->raid_disk
< 0)
3564 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3566 rdev
->raid_disk
= rdev
->new_raid_disk
;
3567 if (rdev
->raid_disk
< 0)
3568 clear_bit(In_sync
, &rdev
->flags
);
3570 if (sysfs_link_rdev(mddev
, rdev
))
3571 printk(KERN_WARNING
"md: cannot register rd%d"
3572 " for %s after level change\n",
3573 rdev
->raid_disk
, mdname(mddev
));
3577 module_put(mddev
->pers
->owner
);
3579 mddev
->private = priv
;
3580 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3581 mddev
->level
= mddev
->new_level
;
3582 mddev
->layout
= mddev
->new_layout
;
3583 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3584 mddev
->delta_disks
= 0;
3585 mddev
->reshape_backwards
= 0;
3586 mddev
->degraded
= 0;
3587 if (mddev
->pers
->sync_request
== NULL
) {
3588 /* this is now an array without redundancy, so
3589 * it must always be in_sync
3592 del_timer_sync(&mddev
->safemode_timer
);
3595 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3596 mddev_resume(mddev
);
3597 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3598 md_new_event(mddev
);
3602 static struct md_sysfs_entry md_level
=
3603 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3607 layout_show(struct mddev
*mddev
, char *page
)
3609 /* just a number, not meaningful for all levels */
3610 if (mddev
->reshape_position
!= MaxSector
&&
3611 mddev
->layout
!= mddev
->new_layout
)
3612 return sprintf(page
, "%d (%d)\n",
3613 mddev
->new_layout
, mddev
->layout
);
3614 return sprintf(page
, "%d\n", mddev
->layout
);
3618 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3621 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3623 if (!*buf
|| (*e
&& *e
!= '\n'))
3628 if (mddev
->pers
->check_reshape
== NULL
)
3630 mddev
->new_layout
= n
;
3631 err
= mddev
->pers
->check_reshape(mddev
);
3633 mddev
->new_layout
= mddev
->layout
;
3637 mddev
->new_layout
= n
;
3638 if (mddev
->reshape_position
== MaxSector
)
3643 static struct md_sysfs_entry md_layout
=
3644 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3648 raid_disks_show(struct mddev
*mddev
, char *page
)
3650 if (mddev
->raid_disks
== 0)
3652 if (mddev
->reshape_position
!= MaxSector
&&
3653 mddev
->delta_disks
!= 0)
3654 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3655 mddev
->raid_disks
- mddev
->delta_disks
);
3656 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3659 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3662 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3666 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3668 if (!*buf
|| (*e
&& *e
!= '\n'))
3672 rv
= update_raid_disks(mddev
, n
);
3673 else if (mddev
->reshape_position
!= MaxSector
) {
3674 struct md_rdev
*rdev
;
3675 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3677 rdev_for_each(rdev
, mddev
) {
3679 rdev
->data_offset
< rdev
->new_data_offset
)
3682 rdev
->data_offset
> rdev
->new_data_offset
)
3685 mddev
->delta_disks
= n
- olddisks
;
3686 mddev
->raid_disks
= n
;
3687 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3689 mddev
->raid_disks
= n
;
3690 return rv
? rv
: len
;
3692 static struct md_sysfs_entry md_raid_disks
=
3693 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3696 chunk_size_show(struct mddev
*mddev
, char *page
)
3698 if (mddev
->reshape_position
!= MaxSector
&&
3699 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3700 return sprintf(page
, "%d (%d)\n",
3701 mddev
->new_chunk_sectors
<< 9,
3702 mddev
->chunk_sectors
<< 9);
3703 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3707 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3710 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3712 if (!*buf
|| (*e
&& *e
!= '\n'))
3717 if (mddev
->pers
->check_reshape
== NULL
)
3719 mddev
->new_chunk_sectors
= n
>> 9;
3720 err
= mddev
->pers
->check_reshape(mddev
);
3722 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3726 mddev
->new_chunk_sectors
= n
>> 9;
3727 if (mddev
->reshape_position
== MaxSector
)
3728 mddev
->chunk_sectors
= n
>> 9;
3732 static struct md_sysfs_entry md_chunk_size
=
3733 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3736 resync_start_show(struct mddev
*mddev
, char *page
)
3738 if (mddev
->recovery_cp
== MaxSector
)
3739 return sprintf(page
, "none\n");
3740 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3744 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3747 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3749 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3751 if (cmd_match(buf
, "none"))
3753 else if (!*buf
|| (*e
&& *e
!= '\n'))
3756 mddev
->recovery_cp
= n
;
3759 static struct md_sysfs_entry md_resync_start
=
3760 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3763 * The array state can be:
3766 * No devices, no size, no level
3767 * Equivalent to STOP_ARRAY ioctl
3769 * May have some settings, but array is not active
3770 * all IO results in error
3771 * When written, doesn't tear down array, but just stops it
3772 * suspended (not supported yet)
3773 * All IO requests will block. The array can be reconfigured.
3774 * Writing this, if accepted, will block until array is quiescent
3776 * no resync can happen. no superblocks get written.
3777 * write requests fail
3779 * like readonly, but behaves like 'clean' on a write request.
3781 * clean - no pending writes, but otherwise active.
3782 * When written to inactive array, starts without resync
3783 * If a write request arrives then
3784 * if metadata is known, mark 'dirty' and switch to 'active'.
3785 * if not known, block and switch to write-pending
3786 * If written to an active array that has pending writes, then fails.
3788 * fully active: IO and resync can be happening.
3789 * When written to inactive array, starts with resync
3792 * clean, but writes are blocked waiting for 'active' to be written.
3795 * like active, but no writes have been seen for a while (100msec).
3798 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3799 write_pending
, active_idle
, bad_word
};
3800 static char *array_states
[] = {
3801 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3802 "write-pending", "active-idle", NULL
};
3804 static int match_word(const char *word
, char **list
)
3807 for (n
=0; list
[n
]; n
++)
3808 if (cmd_match(word
, list
[n
]))
3814 array_state_show(struct mddev
*mddev
, char *page
)
3816 enum array_state st
= inactive
;
3829 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3831 else if (mddev
->safemode
)
3837 if (list_empty(&mddev
->disks
) &&
3838 mddev
->raid_disks
== 0 &&
3839 mddev
->dev_sectors
== 0)
3844 return sprintf(page
, "%s\n", array_states
[st
]);
3847 static int do_md_stop(struct mddev
* mddev
, int ro
, struct block_device
*bdev
);
3848 static int md_set_readonly(struct mddev
* mddev
, struct block_device
*bdev
);
3849 static int do_md_run(struct mddev
* mddev
);
3850 static int restart_array(struct mddev
*mddev
);
3853 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3856 enum array_state st
= match_word(buf
, array_states
);
3861 /* stopping an active array */
3862 err
= do_md_stop(mddev
, 0, NULL
);
3865 /* stopping an active array */
3867 err
= do_md_stop(mddev
, 2, NULL
);
3869 err
= 0; /* already inactive */
3872 break; /* not supported yet */
3875 err
= md_set_readonly(mddev
, NULL
);
3878 set_disk_ro(mddev
->gendisk
, 1);
3879 err
= do_md_run(mddev
);
3885 err
= md_set_readonly(mddev
, NULL
);
3886 else if (mddev
->ro
== 1)
3887 err
= restart_array(mddev
);
3890 set_disk_ro(mddev
->gendisk
, 0);
3894 err
= do_md_run(mddev
);
3899 restart_array(mddev
);
3900 spin_lock_irq(&mddev
->write_lock
);
3901 if (atomic_read(&mddev
->writes_pending
) == 0) {
3902 if (mddev
->in_sync
== 0) {
3904 if (mddev
->safemode
== 1)
3905 mddev
->safemode
= 0;
3906 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3911 spin_unlock_irq(&mddev
->write_lock
);
3917 restart_array(mddev
);
3918 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3919 wake_up(&mddev
->sb_wait
);
3923 set_disk_ro(mddev
->gendisk
, 0);
3924 err
= do_md_run(mddev
);
3929 /* these cannot be set */
3935 if (mddev
->hold_active
== UNTIL_IOCTL
)
3936 mddev
->hold_active
= 0;
3937 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3941 static struct md_sysfs_entry md_array_state
=
3942 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3945 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3946 return sprintf(page
, "%d\n",
3947 atomic_read(&mddev
->max_corr_read_errors
));
3951 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3954 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3956 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3957 atomic_set(&mddev
->max_corr_read_errors
, n
);
3963 static struct md_sysfs_entry max_corr_read_errors
=
3964 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3965 max_corrected_read_errors_store
);
3968 null_show(struct mddev
*mddev
, char *page
)
3974 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3976 /* buf must be %d:%d\n? giving major and minor numbers */
3977 /* The new device is added to the array.
3978 * If the array has a persistent superblock, we read the
3979 * superblock to initialise info and check validity.
3980 * Otherwise, only checking done is that in bind_rdev_to_array,
3981 * which mainly checks size.
3984 int major
= simple_strtoul(buf
, &e
, 10);
3987 struct md_rdev
*rdev
;
3990 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3992 minor
= simple_strtoul(e
+1, &e
, 10);
3993 if (*e
&& *e
!= '\n')
3995 dev
= MKDEV(major
, minor
);
3996 if (major
!= MAJOR(dev
) ||
3997 minor
!= MINOR(dev
))
4001 if (mddev
->persistent
) {
4002 rdev
= md_import_device(dev
, mddev
->major_version
,
4003 mddev
->minor_version
);
4004 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4005 struct md_rdev
*rdev0
4006 = list_entry(mddev
->disks
.next
,
4007 struct md_rdev
, same_set
);
4008 err
= super_types
[mddev
->major_version
]
4009 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4013 } else if (mddev
->external
)
4014 rdev
= md_import_device(dev
, -2, -1);
4016 rdev
= md_import_device(dev
, -1, -1);
4019 return PTR_ERR(rdev
);
4020 err
= bind_rdev_to_array(rdev
, mddev
);
4024 return err
? err
: len
;
4027 static struct md_sysfs_entry md_new_device
=
4028 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4031 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4034 unsigned long chunk
, end_chunk
;
4038 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4040 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4041 if (buf
== end
) break;
4042 if (*end
== '-') { /* range */
4044 end_chunk
= simple_strtoul(buf
, &end
, 0);
4045 if (buf
== end
) break;
4047 if (*end
&& !isspace(*end
)) break;
4048 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4049 buf
= skip_spaces(end
);
4051 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4056 static struct md_sysfs_entry md_bitmap
=
4057 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4060 size_show(struct mddev
*mddev
, char *page
)
4062 return sprintf(page
, "%llu\n",
4063 (unsigned long long)mddev
->dev_sectors
/ 2);
4066 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4069 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4071 /* If array is inactive, we can reduce the component size, but
4072 * not increase it (except from 0).
4073 * If array is active, we can try an on-line resize
4076 int err
= strict_blocks_to_sectors(buf
, §ors
);
4081 err
= update_size(mddev
, sectors
);
4082 md_update_sb(mddev
, 1);
4084 if (mddev
->dev_sectors
== 0 ||
4085 mddev
->dev_sectors
> sectors
)
4086 mddev
->dev_sectors
= sectors
;
4090 return err
? err
: len
;
4093 static struct md_sysfs_entry md_size
=
4094 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4099 * 'none' for arrays with no metadata (good luck...)
4100 * 'external' for arrays with externally managed metadata,
4101 * or N.M for internally known formats
4104 metadata_show(struct mddev
*mddev
, char *page
)
4106 if (mddev
->persistent
)
4107 return sprintf(page
, "%d.%d\n",
4108 mddev
->major_version
, mddev
->minor_version
);
4109 else if (mddev
->external
)
4110 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4112 return sprintf(page
, "none\n");
4116 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4120 /* Changing the details of 'external' metadata is
4121 * always permitted. Otherwise there must be
4122 * no devices attached to the array.
4124 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4126 else if (!list_empty(&mddev
->disks
))
4129 if (cmd_match(buf
, "none")) {
4130 mddev
->persistent
= 0;
4131 mddev
->external
= 0;
4132 mddev
->major_version
= 0;
4133 mddev
->minor_version
= 90;
4136 if (strncmp(buf
, "external:", 9) == 0) {
4137 size_t namelen
= len
-9;
4138 if (namelen
>= sizeof(mddev
->metadata_type
))
4139 namelen
= sizeof(mddev
->metadata_type
)-1;
4140 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4141 mddev
->metadata_type
[namelen
] = 0;
4142 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4143 mddev
->metadata_type
[--namelen
] = 0;
4144 mddev
->persistent
= 0;
4145 mddev
->external
= 1;
4146 mddev
->major_version
= 0;
4147 mddev
->minor_version
= 90;
4150 major
= simple_strtoul(buf
, &e
, 10);
4151 if (e
==buf
|| *e
!= '.')
4154 minor
= simple_strtoul(buf
, &e
, 10);
4155 if (e
==buf
|| (*e
&& *e
!= '\n') )
4157 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4159 mddev
->major_version
= major
;
4160 mddev
->minor_version
= minor
;
4161 mddev
->persistent
= 1;
4162 mddev
->external
= 0;
4166 static struct md_sysfs_entry md_metadata
=
4167 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4170 action_show(struct mddev
*mddev
, char *page
)
4172 char *type
= "idle";
4173 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4175 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4176 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4177 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4179 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4180 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4182 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4186 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4189 return sprintf(page
, "%s\n", type
);
4192 static void reap_sync_thread(struct mddev
*mddev
);
4195 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4197 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4200 if (cmd_match(page
, "frozen"))
4201 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4203 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4205 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4206 if (mddev
->sync_thread
) {
4207 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4208 reap_sync_thread(mddev
);
4210 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4211 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4213 else if (cmd_match(page
, "resync"))
4214 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4215 else if (cmd_match(page
, "recover")) {
4216 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4217 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4218 } else if (cmd_match(page
, "reshape")) {
4220 if (mddev
->pers
->start_reshape
== NULL
)
4222 err
= mddev
->pers
->start_reshape(mddev
);
4225 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4227 if (cmd_match(page
, "check"))
4228 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4229 else if (!cmd_match(page
, "repair"))
4231 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4232 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4234 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4235 md_wakeup_thread(mddev
->thread
);
4236 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4241 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4243 return sprintf(page
, "%llu\n",
4244 (unsigned long long) mddev
->resync_mismatches
);
4247 static struct md_sysfs_entry md_scan_mode
=
4248 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4251 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4254 sync_min_show(struct mddev
*mddev
, char *page
)
4256 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4257 mddev
->sync_speed_min
? "local": "system");
4261 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4265 if (strncmp(buf
, "system", 6)==0) {
4266 mddev
->sync_speed_min
= 0;
4269 min
= simple_strtoul(buf
, &e
, 10);
4270 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4272 mddev
->sync_speed_min
= min
;
4276 static struct md_sysfs_entry md_sync_min
=
4277 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4280 sync_max_show(struct mddev
*mddev
, char *page
)
4282 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4283 mddev
->sync_speed_max
? "local": "system");
4287 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4291 if (strncmp(buf
, "system", 6)==0) {
4292 mddev
->sync_speed_max
= 0;
4295 max
= simple_strtoul(buf
, &e
, 10);
4296 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4298 mddev
->sync_speed_max
= max
;
4302 static struct md_sysfs_entry md_sync_max
=
4303 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4306 degraded_show(struct mddev
*mddev
, char *page
)
4308 return sprintf(page
, "%d\n", mddev
->degraded
);
4310 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4313 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4315 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4319 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4323 if (strict_strtol(buf
, 10, &n
))
4326 if (n
!= 0 && n
!= 1)
4329 mddev
->parallel_resync
= n
;
4331 if (mddev
->sync_thread
)
4332 wake_up(&resync_wait
);
4337 /* force parallel resync, even with shared block devices */
4338 static struct md_sysfs_entry md_sync_force_parallel
=
4339 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4340 sync_force_parallel_show
, sync_force_parallel_store
);
4343 sync_speed_show(struct mddev
*mddev
, char *page
)
4345 unsigned long resync
, dt
, db
;
4346 if (mddev
->curr_resync
== 0)
4347 return sprintf(page
, "none\n");
4348 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4349 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4351 db
= resync
- mddev
->resync_mark_cnt
;
4352 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4355 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4358 sync_completed_show(struct mddev
*mddev
, char *page
)
4360 unsigned long long max_sectors
, resync
;
4362 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4363 return sprintf(page
, "none\n");
4365 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4366 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4367 max_sectors
= mddev
->resync_max_sectors
;
4369 max_sectors
= mddev
->dev_sectors
;
4371 resync
= mddev
->curr_resync_completed
;
4372 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4375 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4378 min_sync_show(struct mddev
*mddev
, char *page
)
4380 return sprintf(page
, "%llu\n",
4381 (unsigned long long)mddev
->resync_min
);
4384 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4386 unsigned long long min
;
4387 if (strict_strtoull(buf
, 10, &min
))
4389 if (min
> mddev
->resync_max
)
4391 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4394 /* Must be a multiple of chunk_size */
4395 if (mddev
->chunk_sectors
) {
4396 sector_t temp
= min
;
4397 if (sector_div(temp
, mddev
->chunk_sectors
))
4400 mddev
->resync_min
= min
;
4405 static struct md_sysfs_entry md_min_sync
=
4406 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4409 max_sync_show(struct mddev
*mddev
, char *page
)
4411 if (mddev
->resync_max
== MaxSector
)
4412 return sprintf(page
, "max\n");
4414 return sprintf(page
, "%llu\n",
4415 (unsigned long long)mddev
->resync_max
);
4418 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4420 if (strncmp(buf
, "max", 3) == 0)
4421 mddev
->resync_max
= MaxSector
;
4423 unsigned long long max
;
4424 if (strict_strtoull(buf
, 10, &max
))
4426 if (max
< mddev
->resync_min
)
4428 if (max
< mddev
->resync_max
&&
4430 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4433 /* Must be a multiple of chunk_size */
4434 if (mddev
->chunk_sectors
) {
4435 sector_t temp
= max
;
4436 if (sector_div(temp
, mddev
->chunk_sectors
))
4439 mddev
->resync_max
= max
;
4441 wake_up(&mddev
->recovery_wait
);
4445 static struct md_sysfs_entry md_max_sync
=
4446 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4449 suspend_lo_show(struct mddev
*mddev
, char *page
)
4451 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4455 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4458 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4459 unsigned long long old
= mddev
->suspend_lo
;
4461 if (mddev
->pers
== NULL
||
4462 mddev
->pers
->quiesce
== NULL
)
4464 if (buf
== e
|| (*e
&& *e
!= '\n'))
4467 mddev
->suspend_lo
= new;
4469 /* Shrinking suspended region */
4470 mddev
->pers
->quiesce(mddev
, 2);
4472 /* Expanding suspended region - need to wait */
4473 mddev
->pers
->quiesce(mddev
, 1);
4474 mddev
->pers
->quiesce(mddev
, 0);
4478 static struct md_sysfs_entry md_suspend_lo
=
4479 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4483 suspend_hi_show(struct mddev
*mddev
, char *page
)
4485 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4489 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4492 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4493 unsigned long long old
= mddev
->suspend_hi
;
4495 if (mddev
->pers
== NULL
||
4496 mddev
->pers
->quiesce
== NULL
)
4498 if (buf
== e
|| (*e
&& *e
!= '\n'))
4501 mddev
->suspend_hi
= new;
4503 /* Shrinking suspended region */
4504 mddev
->pers
->quiesce(mddev
, 2);
4506 /* Expanding suspended region - need to wait */
4507 mddev
->pers
->quiesce(mddev
, 1);
4508 mddev
->pers
->quiesce(mddev
, 0);
4512 static struct md_sysfs_entry md_suspend_hi
=
4513 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4516 reshape_position_show(struct mddev
*mddev
, char *page
)
4518 if (mddev
->reshape_position
!= MaxSector
)
4519 return sprintf(page
, "%llu\n",
4520 (unsigned long long)mddev
->reshape_position
);
4521 strcpy(page
, "none\n");
4526 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4528 struct md_rdev
*rdev
;
4530 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4533 if (buf
== e
|| (*e
&& *e
!= '\n'))
4535 mddev
->reshape_position
= new;
4536 mddev
->delta_disks
= 0;
4537 mddev
->reshape_backwards
= 0;
4538 mddev
->new_level
= mddev
->level
;
4539 mddev
->new_layout
= mddev
->layout
;
4540 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4541 rdev_for_each(rdev
, mddev
)
4542 rdev
->new_data_offset
= rdev
->data_offset
;
4546 static struct md_sysfs_entry md_reshape_position
=
4547 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4548 reshape_position_store
);
4551 reshape_direction_show(struct mddev
*mddev
, char *page
)
4553 return sprintf(page
, "%s\n",
4554 mddev
->reshape_backwards
? "backwards" : "forwards");
4558 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4561 if (cmd_match(buf
, "forwards"))
4563 else if (cmd_match(buf
, "backwards"))
4567 if (mddev
->reshape_backwards
== backwards
)
4570 /* check if we are allowed to change */
4571 if (mddev
->delta_disks
)
4574 if (mddev
->persistent
&&
4575 mddev
->major_version
== 0)
4578 mddev
->reshape_backwards
= backwards
;
4582 static struct md_sysfs_entry md_reshape_direction
=
4583 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4584 reshape_direction_store
);
4587 array_size_show(struct mddev
*mddev
, char *page
)
4589 if (mddev
->external_size
)
4590 return sprintf(page
, "%llu\n",
4591 (unsigned long long)mddev
->array_sectors
/2);
4593 return sprintf(page
, "default\n");
4597 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4601 if (strncmp(buf
, "default", 7) == 0) {
4603 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4605 sectors
= mddev
->array_sectors
;
4607 mddev
->external_size
= 0;
4609 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4611 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4614 mddev
->external_size
= 1;
4617 mddev
->array_sectors
= sectors
;
4619 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4620 revalidate_disk(mddev
->gendisk
);
4625 static struct md_sysfs_entry md_array_size
=
4626 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4629 static struct attribute
*md_default_attrs
[] = {
4632 &md_raid_disks
.attr
,
4633 &md_chunk_size
.attr
,
4635 &md_resync_start
.attr
,
4637 &md_new_device
.attr
,
4638 &md_safe_delay
.attr
,
4639 &md_array_state
.attr
,
4640 &md_reshape_position
.attr
,
4641 &md_reshape_direction
.attr
,
4642 &md_array_size
.attr
,
4643 &max_corr_read_errors
.attr
,
4647 static struct attribute
*md_redundancy_attrs
[] = {
4649 &md_mismatches
.attr
,
4652 &md_sync_speed
.attr
,
4653 &md_sync_force_parallel
.attr
,
4654 &md_sync_completed
.attr
,
4657 &md_suspend_lo
.attr
,
4658 &md_suspend_hi
.attr
,
4663 static struct attribute_group md_redundancy_group
= {
4665 .attrs
= md_redundancy_attrs
,
4670 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4672 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4673 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4678 spin_lock(&all_mddevs_lock
);
4679 if (list_empty(&mddev
->all_mddevs
)) {
4680 spin_unlock(&all_mddevs_lock
);
4684 spin_unlock(&all_mddevs_lock
);
4686 rv
= mddev_lock(mddev
);
4688 rv
= entry
->show(mddev
, page
);
4689 mddev_unlock(mddev
);
4696 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4697 const char *page
, size_t length
)
4699 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4700 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4705 if (!capable(CAP_SYS_ADMIN
))
4707 spin_lock(&all_mddevs_lock
);
4708 if (list_empty(&mddev
->all_mddevs
)) {
4709 spin_unlock(&all_mddevs_lock
);
4713 spin_unlock(&all_mddevs_lock
);
4714 rv
= mddev_lock(mddev
);
4716 rv
= entry
->store(mddev
, page
, length
);
4717 mddev_unlock(mddev
);
4723 static void md_free(struct kobject
*ko
)
4725 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4727 if (mddev
->sysfs_state
)
4728 sysfs_put(mddev
->sysfs_state
);
4730 if (mddev
->gendisk
) {
4731 del_gendisk(mddev
->gendisk
);
4732 put_disk(mddev
->gendisk
);
4735 blk_cleanup_queue(mddev
->queue
);
4740 static const struct sysfs_ops md_sysfs_ops
= {
4741 .show
= md_attr_show
,
4742 .store
= md_attr_store
,
4744 static struct kobj_type md_ktype
= {
4746 .sysfs_ops
= &md_sysfs_ops
,
4747 .default_attrs
= md_default_attrs
,
4752 static void mddev_delayed_delete(struct work_struct
*ws
)
4754 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4756 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4757 kobject_del(&mddev
->kobj
);
4758 kobject_put(&mddev
->kobj
);
4761 static int md_alloc(dev_t dev
, char *name
)
4763 static DEFINE_MUTEX(disks_mutex
);
4764 struct mddev
*mddev
= mddev_find(dev
);
4765 struct gendisk
*disk
;
4774 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4775 shift
= partitioned
? MdpMinorShift
: 0;
4776 unit
= MINOR(mddev
->unit
) >> shift
;
4778 /* wait for any previous instance of this device to be
4779 * completely removed (mddev_delayed_delete).
4781 flush_workqueue(md_misc_wq
);
4783 mutex_lock(&disks_mutex
);
4789 /* Need to ensure that 'name' is not a duplicate.
4791 struct mddev
*mddev2
;
4792 spin_lock(&all_mddevs_lock
);
4794 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4795 if (mddev2
->gendisk
&&
4796 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4797 spin_unlock(&all_mddevs_lock
);
4800 spin_unlock(&all_mddevs_lock
);
4804 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4807 mddev
->queue
->queuedata
= mddev
;
4809 blk_queue_make_request(mddev
->queue
, md_make_request
);
4810 blk_set_stacking_limits(&mddev
->queue
->limits
);
4812 disk
= alloc_disk(1 << shift
);
4814 blk_cleanup_queue(mddev
->queue
);
4815 mddev
->queue
= NULL
;
4818 disk
->major
= MAJOR(mddev
->unit
);
4819 disk
->first_minor
= unit
<< shift
;
4821 strcpy(disk
->disk_name
, name
);
4822 else if (partitioned
)
4823 sprintf(disk
->disk_name
, "md_d%d", unit
);
4825 sprintf(disk
->disk_name
, "md%d", unit
);
4826 disk
->fops
= &md_fops
;
4827 disk
->private_data
= mddev
;
4828 disk
->queue
= mddev
->queue
;
4829 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4830 /* Allow extended partitions. This makes the
4831 * 'mdp' device redundant, but we can't really
4834 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4835 mddev
->gendisk
= disk
;
4836 /* As soon as we call add_disk(), another thread could get
4837 * through to md_open, so make sure it doesn't get too far
4839 mutex_lock(&mddev
->open_mutex
);
4842 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4843 &disk_to_dev(disk
)->kobj
, "%s", "md");
4845 /* This isn't possible, but as kobject_init_and_add is marked
4846 * __must_check, we must do something with the result
4848 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4852 if (mddev
->kobj
.sd
&&
4853 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4854 printk(KERN_DEBUG
"pointless warning\n");
4855 mutex_unlock(&mddev
->open_mutex
);
4857 mutex_unlock(&disks_mutex
);
4858 if (!error
&& mddev
->kobj
.sd
) {
4859 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4860 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4866 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4868 md_alloc(dev
, NULL
);
4872 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4874 /* val must be "md_*" where * is not all digits.
4875 * We allocate an array with a large free minor number, and
4876 * set the name to val. val must not already be an active name.
4878 int len
= strlen(val
);
4879 char buf
[DISK_NAME_LEN
];
4881 while (len
&& val
[len
-1] == '\n')
4883 if (len
>= DISK_NAME_LEN
)
4885 strlcpy(buf
, val
, len
+1);
4886 if (strncmp(buf
, "md_", 3) != 0)
4888 return md_alloc(0, buf
);
4891 static void md_safemode_timeout(unsigned long data
)
4893 struct mddev
*mddev
= (struct mddev
*) data
;
4895 if (!atomic_read(&mddev
->writes_pending
)) {
4896 mddev
->safemode
= 1;
4897 if (mddev
->external
)
4898 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4900 md_wakeup_thread(mddev
->thread
);
4903 static int start_dirty_degraded
;
4905 int md_run(struct mddev
*mddev
)
4908 struct md_rdev
*rdev
;
4909 struct md_personality
*pers
;
4911 if (list_empty(&mddev
->disks
))
4912 /* cannot run an array with no devices.. */
4917 /* Cannot run until previous stop completes properly */
4918 if (mddev
->sysfs_active
)
4922 * Analyze all RAID superblock(s)
4924 if (!mddev
->raid_disks
) {
4925 if (!mddev
->persistent
)
4930 if (mddev
->level
!= LEVEL_NONE
)
4931 request_module("md-level-%d", mddev
->level
);
4932 else if (mddev
->clevel
[0])
4933 request_module("md-%s", mddev
->clevel
);
4936 * Drop all container device buffers, from now on
4937 * the only valid external interface is through the md
4940 rdev_for_each(rdev
, mddev
) {
4941 if (test_bit(Faulty
, &rdev
->flags
))
4943 sync_blockdev(rdev
->bdev
);
4944 invalidate_bdev(rdev
->bdev
);
4946 /* perform some consistency tests on the device.
4947 * We don't want the data to overlap the metadata,
4948 * Internal Bitmap issues have been handled elsewhere.
4950 if (rdev
->meta_bdev
) {
4951 /* Nothing to check */;
4952 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4953 if (mddev
->dev_sectors
&&
4954 rdev
->data_offset
+ mddev
->dev_sectors
4956 printk("md: %s: data overlaps metadata\n",
4961 if (rdev
->sb_start
+ rdev
->sb_size
/512
4962 > rdev
->data_offset
) {
4963 printk("md: %s: metadata overlaps data\n",
4968 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4971 if (mddev
->bio_set
== NULL
)
4972 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
4974 spin_lock(&pers_lock
);
4975 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4976 if (!pers
|| !try_module_get(pers
->owner
)) {
4977 spin_unlock(&pers_lock
);
4978 if (mddev
->level
!= LEVEL_NONE
)
4979 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4982 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4987 spin_unlock(&pers_lock
);
4988 if (mddev
->level
!= pers
->level
) {
4989 mddev
->level
= pers
->level
;
4990 mddev
->new_level
= pers
->level
;
4992 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4994 if (mddev
->reshape_position
!= MaxSector
&&
4995 pers
->start_reshape
== NULL
) {
4996 /* This personality cannot handle reshaping... */
4998 module_put(pers
->owner
);
5002 if (pers
->sync_request
) {
5003 /* Warn if this is a potentially silly
5006 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5007 struct md_rdev
*rdev2
;
5010 rdev_for_each(rdev
, mddev
)
5011 rdev_for_each(rdev2
, mddev
) {
5013 rdev
->bdev
->bd_contains
==
5014 rdev2
->bdev
->bd_contains
) {
5016 "%s: WARNING: %s appears to be"
5017 " on the same physical disk as"
5020 bdevname(rdev
->bdev
,b
),
5021 bdevname(rdev2
->bdev
,b2
));
5028 "True protection against single-disk"
5029 " failure might be compromised.\n");
5032 mddev
->recovery
= 0;
5033 /* may be over-ridden by personality */
5034 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5036 mddev
->ok_start_degraded
= start_dirty_degraded
;
5038 if (start_readonly
&& mddev
->ro
== 0)
5039 mddev
->ro
= 2; /* read-only, but switch on first write */
5041 err
= mddev
->pers
->run(mddev
);
5043 printk(KERN_ERR
"md: pers->run() failed ...\n");
5044 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5045 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5046 " but 'external_size' not in effect?\n", __func__
);
5048 "md: invalid array_size %llu > default size %llu\n",
5049 (unsigned long long)mddev
->array_sectors
/ 2,
5050 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
5052 mddev
->pers
->stop(mddev
);
5054 if (err
== 0 && mddev
->pers
->sync_request
&&
5055 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5056 err
= bitmap_create(mddev
);
5058 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5059 mdname(mddev
), err
);
5060 mddev
->pers
->stop(mddev
);
5064 module_put(mddev
->pers
->owner
);
5066 bitmap_destroy(mddev
);
5069 if (mddev
->pers
->sync_request
) {
5070 if (mddev
->kobj
.sd
&&
5071 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5073 "md: cannot register extra attributes for %s\n",
5075 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5076 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5079 atomic_set(&mddev
->writes_pending
,0);
5080 atomic_set(&mddev
->max_corr_read_errors
,
5081 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5082 mddev
->safemode
= 0;
5083 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5084 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5085 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5089 rdev_for_each(rdev
, mddev
)
5090 if (rdev
->raid_disk
>= 0)
5091 if (sysfs_link_rdev(mddev
, rdev
))
5092 /* failure here is OK */;
5094 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5097 md_update_sb(mddev
, 0);
5099 md_new_event(mddev
);
5100 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5101 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5102 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5105 EXPORT_SYMBOL_GPL(md_run
);
5107 static int do_md_run(struct mddev
*mddev
)
5111 err
= md_run(mddev
);
5114 err
= bitmap_load(mddev
);
5116 bitmap_destroy(mddev
);
5120 md_wakeup_thread(mddev
->thread
);
5121 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5123 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5124 revalidate_disk(mddev
->gendisk
);
5126 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5131 static int restart_array(struct mddev
*mddev
)
5133 struct gendisk
*disk
= mddev
->gendisk
;
5135 /* Complain if it has no devices */
5136 if (list_empty(&mddev
->disks
))
5142 mddev
->safemode
= 0;
5144 set_disk_ro(disk
, 0);
5145 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5147 /* Kick recovery or resync if necessary */
5148 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5149 md_wakeup_thread(mddev
->thread
);
5150 md_wakeup_thread(mddev
->sync_thread
);
5151 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5155 /* similar to deny_write_access, but accounts for our holding a reference
5156 * to the file ourselves */
5157 static int deny_bitmap_write_access(struct file
* file
)
5159 struct inode
*inode
= file
->f_mapping
->host
;
5161 spin_lock(&inode
->i_lock
);
5162 if (atomic_read(&inode
->i_writecount
) > 1) {
5163 spin_unlock(&inode
->i_lock
);
5166 atomic_set(&inode
->i_writecount
, -1);
5167 spin_unlock(&inode
->i_lock
);
5172 void restore_bitmap_write_access(struct file
*file
)
5174 struct inode
*inode
= file
->f_mapping
->host
;
5176 spin_lock(&inode
->i_lock
);
5177 atomic_set(&inode
->i_writecount
, 1);
5178 spin_unlock(&inode
->i_lock
);
5181 static void md_clean(struct mddev
*mddev
)
5183 mddev
->array_sectors
= 0;
5184 mddev
->external_size
= 0;
5185 mddev
->dev_sectors
= 0;
5186 mddev
->raid_disks
= 0;
5187 mddev
->recovery_cp
= 0;
5188 mddev
->resync_min
= 0;
5189 mddev
->resync_max
= MaxSector
;
5190 mddev
->reshape_position
= MaxSector
;
5191 mddev
->external
= 0;
5192 mddev
->persistent
= 0;
5193 mddev
->level
= LEVEL_NONE
;
5194 mddev
->clevel
[0] = 0;
5197 mddev
->metadata_type
[0] = 0;
5198 mddev
->chunk_sectors
= 0;
5199 mddev
->ctime
= mddev
->utime
= 0;
5201 mddev
->max_disks
= 0;
5203 mddev
->can_decrease_events
= 0;
5204 mddev
->delta_disks
= 0;
5205 mddev
->reshape_backwards
= 0;
5206 mddev
->new_level
= LEVEL_NONE
;
5207 mddev
->new_layout
= 0;
5208 mddev
->new_chunk_sectors
= 0;
5209 mddev
->curr_resync
= 0;
5210 mddev
->resync_mismatches
= 0;
5211 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5212 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5213 mddev
->recovery
= 0;
5216 mddev
->degraded
= 0;
5217 mddev
->safemode
= 0;
5218 mddev
->merge_check_needed
= 0;
5219 mddev
->bitmap_info
.offset
= 0;
5220 mddev
->bitmap_info
.default_offset
= 0;
5221 mddev
->bitmap_info
.default_space
= 0;
5222 mddev
->bitmap_info
.chunksize
= 0;
5223 mddev
->bitmap_info
.daemon_sleep
= 0;
5224 mddev
->bitmap_info
.max_write_behind
= 0;
5227 static void __md_stop_writes(struct mddev
*mddev
)
5229 if (mddev
->sync_thread
) {
5230 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5231 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5232 reap_sync_thread(mddev
);
5235 del_timer_sync(&mddev
->safemode_timer
);
5237 bitmap_flush(mddev
);
5238 md_super_wait(mddev
);
5240 if (!mddev
->in_sync
|| mddev
->flags
) {
5241 /* mark array as shutdown cleanly */
5243 md_update_sb(mddev
, 1);
5247 void md_stop_writes(struct mddev
*mddev
)
5250 __md_stop_writes(mddev
);
5251 mddev_unlock(mddev
);
5253 EXPORT_SYMBOL_GPL(md_stop_writes
);
5255 void md_stop(struct mddev
*mddev
)
5258 mddev
->pers
->stop(mddev
);
5259 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5260 mddev
->to_remove
= &md_redundancy_group
;
5261 module_put(mddev
->pers
->owner
);
5263 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5265 EXPORT_SYMBOL_GPL(md_stop
);
5267 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5270 mutex_lock(&mddev
->open_mutex
);
5271 if (atomic_read(&mddev
->openers
) > !!bdev
) {
5272 printk("md: %s still in use.\n",mdname(mddev
));
5277 sync_blockdev(bdev
);
5279 __md_stop_writes(mddev
);
5285 set_disk_ro(mddev
->gendisk
, 1);
5286 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5287 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5291 mutex_unlock(&mddev
->open_mutex
);
5296 * 0 - completely stop and dis-assemble array
5297 * 2 - stop but do not disassemble array
5299 static int do_md_stop(struct mddev
* mddev
, int mode
,
5300 struct block_device
*bdev
)
5302 struct gendisk
*disk
= mddev
->gendisk
;
5303 struct md_rdev
*rdev
;
5305 mutex_lock(&mddev
->open_mutex
);
5306 if (atomic_read(&mddev
->openers
) > !!bdev
||
5307 mddev
->sysfs_active
) {
5308 printk("md: %s still in use.\n",mdname(mddev
));
5309 mutex_unlock(&mddev
->open_mutex
);
5313 /* It is possible IO was issued on some other
5314 * open file which was closed before we took ->open_mutex.
5315 * As that was not the last close __blkdev_put will not
5316 * have called sync_blockdev, so we must.
5318 sync_blockdev(bdev
);
5322 set_disk_ro(disk
, 0);
5324 __md_stop_writes(mddev
);
5326 mddev
->queue
->merge_bvec_fn
= NULL
;
5327 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5329 /* tell userspace to handle 'inactive' */
5330 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5332 rdev_for_each(rdev
, mddev
)
5333 if (rdev
->raid_disk
>= 0)
5334 sysfs_unlink_rdev(mddev
, rdev
);
5336 set_capacity(disk
, 0);
5337 mutex_unlock(&mddev
->open_mutex
);
5339 revalidate_disk(disk
);
5344 mutex_unlock(&mddev
->open_mutex
);
5346 * Free resources if final stop
5349 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5351 bitmap_destroy(mddev
);
5352 if (mddev
->bitmap_info
.file
) {
5353 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5354 fput(mddev
->bitmap_info
.file
);
5355 mddev
->bitmap_info
.file
= NULL
;
5357 mddev
->bitmap_info
.offset
= 0;
5359 export_array(mddev
);
5362 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5363 if (mddev
->hold_active
== UNTIL_STOP
)
5364 mddev
->hold_active
= 0;
5366 blk_integrity_unregister(disk
);
5367 md_new_event(mddev
);
5368 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5373 static void autorun_array(struct mddev
*mddev
)
5375 struct md_rdev
*rdev
;
5378 if (list_empty(&mddev
->disks
))
5381 printk(KERN_INFO
"md: running: ");
5383 rdev_for_each(rdev
, mddev
) {
5384 char b
[BDEVNAME_SIZE
];
5385 printk("<%s>", bdevname(rdev
->bdev
,b
));
5389 err
= do_md_run(mddev
);
5391 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5392 do_md_stop(mddev
, 0, NULL
);
5397 * lets try to run arrays based on all disks that have arrived
5398 * until now. (those are in pending_raid_disks)
5400 * the method: pick the first pending disk, collect all disks with
5401 * the same UUID, remove all from the pending list and put them into
5402 * the 'same_array' list. Then order this list based on superblock
5403 * update time (freshest comes first), kick out 'old' disks and
5404 * compare superblocks. If everything's fine then run it.
5406 * If "unit" is allocated, then bump its reference count
5408 static void autorun_devices(int part
)
5410 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5411 struct mddev
*mddev
;
5412 char b
[BDEVNAME_SIZE
];
5414 printk(KERN_INFO
"md: autorun ...\n");
5415 while (!list_empty(&pending_raid_disks
)) {
5418 LIST_HEAD(candidates
);
5419 rdev0
= list_entry(pending_raid_disks
.next
,
5420 struct md_rdev
, same_set
);
5422 printk(KERN_INFO
"md: considering %s ...\n",
5423 bdevname(rdev0
->bdev
,b
));
5424 INIT_LIST_HEAD(&candidates
);
5425 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5426 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5427 printk(KERN_INFO
"md: adding %s ...\n",
5428 bdevname(rdev
->bdev
,b
));
5429 list_move(&rdev
->same_set
, &candidates
);
5432 * now we have a set of devices, with all of them having
5433 * mostly sane superblocks. It's time to allocate the
5437 dev
= MKDEV(mdp_major
,
5438 rdev0
->preferred_minor
<< MdpMinorShift
);
5439 unit
= MINOR(dev
) >> MdpMinorShift
;
5441 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5444 if (rdev0
->preferred_minor
!= unit
) {
5445 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5446 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5450 md_probe(dev
, NULL
, NULL
);
5451 mddev
= mddev_find(dev
);
5452 if (!mddev
|| !mddev
->gendisk
) {
5456 "md: cannot allocate memory for md drive.\n");
5459 if (mddev_lock(mddev
))
5460 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5462 else if (mddev
->raid_disks
|| mddev
->major_version
5463 || !list_empty(&mddev
->disks
)) {
5465 "md: %s already running, cannot run %s\n",
5466 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5467 mddev_unlock(mddev
);
5469 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5470 mddev
->persistent
= 1;
5471 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5472 list_del_init(&rdev
->same_set
);
5473 if (bind_rdev_to_array(rdev
, mddev
))
5476 autorun_array(mddev
);
5477 mddev_unlock(mddev
);
5479 /* on success, candidates will be empty, on error
5482 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5483 list_del_init(&rdev
->same_set
);
5488 printk(KERN_INFO
"md: ... autorun DONE.\n");
5490 #endif /* !MODULE */
5492 static int get_version(void __user
* arg
)
5496 ver
.major
= MD_MAJOR_VERSION
;
5497 ver
.minor
= MD_MINOR_VERSION
;
5498 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5500 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5506 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5508 mdu_array_info_t info
;
5509 int nr
,working
,insync
,failed
,spare
;
5510 struct md_rdev
*rdev
;
5512 nr
=working
=insync
=failed
=spare
=0;
5513 rdev_for_each(rdev
, mddev
) {
5515 if (test_bit(Faulty
, &rdev
->flags
))
5519 if (test_bit(In_sync
, &rdev
->flags
))
5526 info
.major_version
= mddev
->major_version
;
5527 info
.minor_version
= mddev
->minor_version
;
5528 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5529 info
.ctime
= mddev
->ctime
;
5530 info
.level
= mddev
->level
;
5531 info
.size
= mddev
->dev_sectors
/ 2;
5532 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5535 info
.raid_disks
= mddev
->raid_disks
;
5536 info
.md_minor
= mddev
->md_minor
;
5537 info
.not_persistent
= !mddev
->persistent
;
5539 info
.utime
= mddev
->utime
;
5542 info
.state
= (1<<MD_SB_CLEAN
);
5543 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5544 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5545 info
.active_disks
= insync
;
5546 info
.working_disks
= working
;
5547 info
.failed_disks
= failed
;
5548 info
.spare_disks
= spare
;
5550 info
.layout
= mddev
->layout
;
5551 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5553 if (copy_to_user(arg
, &info
, sizeof(info
)))
5559 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5561 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5562 char *ptr
, *buf
= NULL
;
5565 if (md_allow_write(mddev
))
5566 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5568 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5573 /* bitmap disabled, zero the first byte and copy out */
5574 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5575 file
->pathname
[0] = '\0';
5579 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5583 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5584 buf
, sizeof(file
->pathname
));
5588 strcpy(file
->pathname
, ptr
);
5592 if (copy_to_user(arg
, file
, sizeof(*file
)))
5600 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5602 mdu_disk_info_t info
;
5603 struct md_rdev
*rdev
;
5605 if (copy_from_user(&info
, arg
, sizeof(info
)))
5608 rdev
= find_rdev_nr(mddev
, info
.number
);
5610 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5611 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5612 info
.raid_disk
= rdev
->raid_disk
;
5614 if (test_bit(Faulty
, &rdev
->flags
))
5615 info
.state
|= (1<<MD_DISK_FAULTY
);
5616 else if (test_bit(In_sync
, &rdev
->flags
)) {
5617 info
.state
|= (1<<MD_DISK_ACTIVE
);
5618 info
.state
|= (1<<MD_DISK_SYNC
);
5620 if (test_bit(WriteMostly
, &rdev
->flags
))
5621 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5623 info
.major
= info
.minor
= 0;
5624 info
.raid_disk
= -1;
5625 info
.state
= (1<<MD_DISK_REMOVED
);
5628 if (copy_to_user(arg
, &info
, sizeof(info
)))
5634 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5636 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5637 struct md_rdev
*rdev
;
5638 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5640 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5643 if (!mddev
->raid_disks
) {
5645 /* expecting a device which has a superblock */
5646 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5649 "md: md_import_device returned %ld\n",
5651 return PTR_ERR(rdev
);
5653 if (!list_empty(&mddev
->disks
)) {
5654 struct md_rdev
*rdev0
5655 = list_entry(mddev
->disks
.next
,
5656 struct md_rdev
, same_set
);
5657 err
= super_types
[mddev
->major_version
]
5658 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5661 "md: %s has different UUID to %s\n",
5662 bdevname(rdev
->bdev
,b
),
5663 bdevname(rdev0
->bdev
,b2
));
5668 err
= bind_rdev_to_array(rdev
, mddev
);
5675 * add_new_disk can be used once the array is assembled
5676 * to add "hot spares". They must already have a superblock
5681 if (!mddev
->pers
->hot_add_disk
) {
5683 "%s: personality does not support diskops!\n",
5687 if (mddev
->persistent
)
5688 rdev
= md_import_device(dev
, mddev
->major_version
,
5689 mddev
->minor_version
);
5691 rdev
= md_import_device(dev
, -1, -1);
5694 "md: md_import_device returned %ld\n",
5696 return PTR_ERR(rdev
);
5698 /* set saved_raid_disk if appropriate */
5699 if (!mddev
->persistent
) {
5700 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5701 info
->raid_disk
< mddev
->raid_disks
) {
5702 rdev
->raid_disk
= info
->raid_disk
;
5703 set_bit(In_sync
, &rdev
->flags
);
5705 rdev
->raid_disk
= -1;
5707 super_types
[mddev
->major_version
].
5708 validate_super(mddev
, rdev
);
5709 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5710 rdev
->raid_disk
!= info
->raid_disk
) {
5711 /* This was a hot-add request, but events doesn't
5712 * match, so reject it.
5718 if (test_bit(In_sync
, &rdev
->flags
))
5719 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5721 rdev
->saved_raid_disk
= -1;
5723 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5724 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5725 set_bit(WriteMostly
, &rdev
->flags
);
5727 clear_bit(WriteMostly
, &rdev
->flags
);
5729 rdev
->raid_disk
= -1;
5730 err
= bind_rdev_to_array(rdev
, mddev
);
5731 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5732 /* If there is hot_add_disk but no hot_remove_disk
5733 * then added disks for geometry changes,
5734 * and should be added immediately.
5736 super_types
[mddev
->major_version
].
5737 validate_super(mddev
, rdev
);
5738 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5740 unbind_rdev_from_array(rdev
);
5745 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5747 md_update_sb(mddev
, 1);
5748 if (mddev
->degraded
)
5749 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5750 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5752 md_new_event(mddev
);
5753 md_wakeup_thread(mddev
->thread
);
5757 /* otherwise, add_new_disk is only allowed
5758 * for major_version==0 superblocks
5760 if (mddev
->major_version
!= 0) {
5761 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5766 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5768 rdev
= md_import_device(dev
, -1, 0);
5771 "md: error, md_import_device() returned %ld\n",
5773 return PTR_ERR(rdev
);
5775 rdev
->desc_nr
= info
->number
;
5776 if (info
->raid_disk
< mddev
->raid_disks
)
5777 rdev
->raid_disk
= info
->raid_disk
;
5779 rdev
->raid_disk
= -1;
5781 if (rdev
->raid_disk
< mddev
->raid_disks
)
5782 if (info
->state
& (1<<MD_DISK_SYNC
))
5783 set_bit(In_sync
, &rdev
->flags
);
5785 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5786 set_bit(WriteMostly
, &rdev
->flags
);
5788 if (!mddev
->persistent
) {
5789 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5790 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5792 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5793 rdev
->sectors
= rdev
->sb_start
;
5795 err
= bind_rdev_to_array(rdev
, mddev
);
5805 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5807 char b
[BDEVNAME_SIZE
];
5808 struct md_rdev
*rdev
;
5810 rdev
= find_rdev(mddev
, dev
);
5814 if (rdev
->raid_disk
>= 0)
5817 kick_rdev_from_array(rdev
);
5818 md_update_sb(mddev
, 1);
5819 md_new_event(mddev
);
5823 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5824 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5828 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5830 char b
[BDEVNAME_SIZE
];
5832 struct md_rdev
*rdev
;
5837 if (mddev
->major_version
!= 0) {
5838 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5839 " version-0 superblocks.\n",
5843 if (!mddev
->pers
->hot_add_disk
) {
5845 "%s: personality does not support diskops!\n",
5850 rdev
= md_import_device(dev
, -1, 0);
5853 "md: error, md_import_device() returned %ld\n",
5858 if (mddev
->persistent
)
5859 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5861 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5863 rdev
->sectors
= rdev
->sb_start
;
5865 if (test_bit(Faulty
, &rdev
->flags
)) {
5867 "md: can not hot-add faulty %s disk to %s!\n",
5868 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5872 clear_bit(In_sync
, &rdev
->flags
);
5874 rdev
->saved_raid_disk
= -1;
5875 err
= bind_rdev_to_array(rdev
, mddev
);
5880 * The rest should better be atomic, we can have disk failures
5881 * noticed in interrupt contexts ...
5884 rdev
->raid_disk
= -1;
5886 md_update_sb(mddev
, 1);
5889 * Kick recovery, maybe this spare has to be added to the
5890 * array immediately.
5892 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5893 md_wakeup_thread(mddev
->thread
);
5894 md_new_event(mddev
);
5902 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5907 if (!mddev
->pers
->quiesce
)
5909 if (mddev
->recovery
|| mddev
->sync_thread
)
5911 /* we should be able to change the bitmap.. */
5917 return -EEXIST
; /* cannot add when bitmap is present */
5918 mddev
->bitmap_info
.file
= fget(fd
);
5920 if (mddev
->bitmap_info
.file
== NULL
) {
5921 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5926 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5928 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5930 fput(mddev
->bitmap_info
.file
);
5931 mddev
->bitmap_info
.file
= NULL
;
5934 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5935 } else if (mddev
->bitmap
== NULL
)
5936 return -ENOENT
; /* cannot remove what isn't there */
5939 mddev
->pers
->quiesce(mddev
, 1);
5941 err
= bitmap_create(mddev
);
5943 err
= bitmap_load(mddev
);
5945 if (fd
< 0 || err
) {
5946 bitmap_destroy(mddev
);
5947 fd
= -1; /* make sure to put the file */
5949 mddev
->pers
->quiesce(mddev
, 0);
5952 if (mddev
->bitmap_info
.file
) {
5953 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5954 fput(mddev
->bitmap_info
.file
);
5956 mddev
->bitmap_info
.file
= NULL
;
5963 * set_array_info is used two different ways
5964 * The original usage is when creating a new array.
5965 * In this usage, raid_disks is > 0 and it together with
5966 * level, size, not_persistent,layout,chunksize determine the
5967 * shape of the array.
5968 * This will always create an array with a type-0.90.0 superblock.
5969 * The newer usage is when assembling an array.
5970 * In this case raid_disks will be 0, and the major_version field is
5971 * use to determine which style super-blocks are to be found on the devices.
5972 * The minor and patch _version numbers are also kept incase the
5973 * super_block handler wishes to interpret them.
5975 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
5978 if (info
->raid_disks
== 0) {
5979 /* just setting version number for superblock loading */
5980 if (info
->major_version
< 0 ||
5981 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5982 super_types
[info
->major_version
].name
== NULL
) {
5983 /* maybe try to auto-load a module? */
5985 "md: superblock version %d not known\n",
5986 info
->major_version
);
5989 mddev
->major_version
= info
->major_version
;
5990 mddev
->minor_version
= info
->minor_version
;
5991 mddev
->patch_version
= info
->patch_version
;
5992 mddev
->persistent
= !info
->not_persistent
;
5993 /* ensure mddev_put doesn't delete this now that there
5994 * is some minimal configuration.
5996 mddev
->ctime
= get_seconds();
5999 mddev
->major_version
= MD_MAJOR_VERSION
;
6000 mddev
->minor_version
= MD_MINOR_VERSION
;
6001 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6002 mddev
->ctime
= get_seconds();
6004 mddev
->level
= info
->level
;
6005 mddev
->clevel
[0] = 0;
6006 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6007 mddev
->raid_disks
= info
->raid_disks
;
6008 /* don't set md_minor, it is determined by which /dev/md* was
6011 if (info
->state
& (1<<MD_SB_CLEAN
))
6012 mddev
->recovery_cp
= MaxSector
;
6014 mddev
->recovery_cp
= 0;
6015 mddev
->persistent
= ! info
->not_persistent
;
6016 mddev
->external
= 0;
6018 mddev
->layout
= info
->layout
;
6019 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6021 mddev
->max_disks
= MD_SB_DISKS
;
6023 if (mddev
->persistent
)
6025 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6027 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6028 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6029 mddev
->bitmap_info
.offset
= 0;
6031 mddev
->reshape_position
= MaxSector
;
6034 * Generate a 128 bit UUID
6036 get_random_bytes(mddev
->uuid
, 16);
6038 mddev
->new_level
= mddev
->level
;
6039 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6040 mddev
->new_layout
= mddev
->layout
;
6041 mddev
->delta_disks
= 0;
6042 mddev
->reshape_backwards
= 0;
6047 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6049 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6051 if (mddev
->external_size
)
6054 mddev
->array_sectors
= array_sectors
;
6056 EXPORT_SYMBOL(md_set_array_sectors
);
6058 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6060 struct md_rdev
*rdev
;
6062 int fit
= (num_sectors
== 0);
6064 if (mddev
->pers
->resize
== NULL
)
6066 /* The "num_sectors" is the number of sectors of each device that
6067 * is used. This can only make sense for arrays with redundancy.
6068 * linear and raid0 always use whatever space is available. We can only
6069 * consider changing this number if no resync or reconstruction is
6070 * happening, and if the new size is acceptable. It must fit before the
6071 * sb_start or, if that is <data_offset, it must fit before the size
6072 * of each device. If num_sectors is zero, we find the largest size
6075 if (mddev
->sync_thread
)
6078 rdev_for_each(rdev
, mddev
) {
6079 sector_t avail
= rdev
->sectors
;
6081 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6082 num_sectors
= avail
;
6083 if (avail
< num_sectors
)
6086 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6088 revalidate_disk(mddev
->gendisk
);
6092 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6095 struct md_rdev
*rdev
;
6096 /* change the number of raid disks */
6097 if (mddev
->pers
->check_reshape
== NULL
)
6099 if (raid_disks
<= 0 ||
6100 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6102 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
6105 rdev_for_each(rdev
, mddev
) {
6106 if (mddev
->raid_disks
< raid_disks
&&
6107 rdev
->data_offset
< rdev
->new_data_offset
)
6109 if (mddev
->raid_disks
> raid_disks
&&
6110 rdev
->data_offset
> rdev
->new_data_offset
)
6114 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6115 if (mddev
->delta_disks
< 0)
6116 mddev
->reshape_backwards
= 1;
6117 else if (mddev
->delta_disks
> 0)
6118 mddev
->reshape_backwards
= 0;
6120 rv
= mddev
->pers
->check_reshape(mddev
);
6122 mddev
->delta_disks
= 0;
6123 mddev
->reshape_backwards
= 0;
6130 * update_array_info is used to change the configuration of an
6132 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6133 * fields in the info are checked against the array.
6134 * Any differences that cannot be handled will cause an error.
6135 * Normally, only one change can be managed at a time.
6137 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6143 /* calculate expected state,ignoring low bits */
6144 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6145 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6147 if (mddev
->major_version
!= info
->major_version
||
6148 mddev
->minor_version
!= info
->minor_version
||
6149 /* mddev->patch_version != info->patch_version || */
6150 mddev
->ctime
!= info
->ctime
||
6151 mddev
->level
!= info
->level
||
6152 /* mddev->layout != info->layout || */
6153 !mddev
->persistent
!= info
->not_persistent
||
6154 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6155 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6156 ((state
^info
->state
) & 0xfffffe00)
6159 /* Check there is only one change */
6160 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6162 if (mddev
->raid_disks
!= info
->raid_disks
)
6164 if (mddev
->layout
!= info
->layout
)
6166 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6173 if (mddev
->layout
!= info
->layout
) {
6175 * we don't need to do anything at the md level, the
6176 * personality will take care of it all.
6178 if (mddev
->pers
->check_reshape
== NULL
)
6181 mddev
->new_layout
= info
->layout
;
6182 rv
= mddev
->pers
->check_reshape(mddev
);
6184 mddev
->new_layout
= mddev
->layout
;
6188 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6189 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6191 if (mddev
->raid_disks
!= info
->raid_disks
)
6192 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6194 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6195 if (mddev
->pers
->quiesce
== NULL
)
6197 if (mddev
->recovery
|| mddev
->sync_thread
)
6199 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6200 /* add the bitmap */
6203 if (mddev
->bitmap_info
.default_offset
== 0)
6205 mddev
->bitmap_info
.offset
=
6206 mddev
->bitmap_info
.default_offset
;
6207 mddev
->bitmap_info
.space
=
6208 mddev
->bitmap_info
.default_space
;
6209 mddev
->pers
->quiesce(mddev
, 1);
6210 rv
= bitmap_create(mddev
);
6212 rv
= bitmap_load(mddev
);
6214 bitmap_destroy(mddev
);
6215 mddev
->pers
->quiesce(mddev
, 0);
6217 /* remove the bitmap */
6220 if (mddev
->bitmap
->storage
.file
)
6222 mddev
->pers
->quiesce(mddev
, 1);
6223 bitmap_destroy(mddev
);
6224 mddev
->pers
->quiesce(mddev
, 0);
6225 mddev
->bitmap_info
.offset
= 0;
6228 md_update_sb(mddev
, 1);
6232 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6234 struct md_rdev
*rdev
;
6236 if (mddev
->pers
== NULL
)
6239 rdev
= find_rdev(mddev
, dev
);
6243 md_error(mddev
, rdev
);
6244 if (!test_bit(Faulty
, &rdev
->flags
))
6250 * We have a problem here : there is no easy way to give a CHS
6251 * virtual geometry. We currently pretend that we have a 2 heads
6252 * 4 sectors (with a BIG number of cylinders...). This drives
6253 * dosfs just mad... ;-)
6255 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6257 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6261 geo
->cylinders
= mddev
->array_sectors
/ 8;
6265 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6266 unsigned int cmd
, unsigned long arg
)
6269 void __user
*argp
= (void __user
*)arg
;
6270 struct mddev
*mddev
= NULL
;
6275 case GET_ARRAY_INFO
:
6279 if (!capable(CAP_SYS_ADMIN
))
6284 * Commands dealing with the RAID driver but not any
6290 err
= get_version(argp
);
6293 case PRINT_RAID_DEBUG
:
6301 autostart_arrays(arg
);
6308 * Commands creating/starting a new array:
6311 mddev
= bdev
->bd_disk
->private_data
;
6318 err
= mddev_lock(mddev
);
6321 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6328 case SET_ARRAY_INFO
:
6330 mdu_array_info_t info
;
6332 memset(&info
, 0, sizeof(info
));
6333 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6338 err
= update_array_info(mddev
, &info
);
6340 printk(KERN_WARNING
"md: couldn't update"
6341 " array info. %d\n", err
);
6346 if (!list_empty(&mddev
->disks
)) {
6348 "md: array %s already has disks!\n",
6353 if (mddev
->raid_disks
) {
6355 "md: array %s already initialised!\n",
6360 err
= set_array_info(mddev
, &info
);
6362 printk(KERN_WARNING
"md: couldn't set"
6363 " array info. %d\n", err
);
6373 * Commands querying/configuring an existing array:
6375 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6376 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6377 if ((!mddev
->raid_disks
&& !mddev
->external
)
6378 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6379 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6380 && cmd
!= GET_BITMAP_FILE
) {
6386 * Commands even a read-only array can execute:
6390 case GET_ARRAY_INFO
:
6391 err
= get_array_info(mddev
, argp
);
6394 case GET_BITMAP_FILE
:
6395 err
= get_bitmap_file(mddev
, argp
);
6399 err
= get_disk_info(mddev
, argp
);
6402 case RESTART_ARRAY_RW
:
6403 err
= restart_array(mddev
);
6407 err
= do_md_stop(mddev
, 0, bdev
);
6411 err
= md_set_readonly(mddev
, bdev
);
6415 if (get_user(ro
, (int __user
*)(arg
))) {
6421 /* if the bdev is going readonly the value of mddev->ro
6422 * does not matter, no writes are coming
6427 /* are we are already prepared for writes? */
6431 /* transitioning to readauto need only happen for
6432 * arrays that call md_write_start
6435 err
= restart_array(mddev
);
6438 set_disk_ro(mddev
->gendisk
, 0);
6445 * The remaining ioctls are changing the state of the
6446 * superblock, so we do not allow them on read-only arrays.
6447 * However non-MD ioctls (e.g. get-size) will still come through
6448 * here and hit the 'default' below, so only disallow
6449 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6451 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6452 if (mddev
->ro
== 2) {
6454 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6455 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6456 md_wakeup_thread(mddev
->thread
);
6467 mdu_disk_info_t info
;
6468 if (copy_from_user(&info
, argp
, sizeof(info
)))
6471 err
= add_new_disk(mddev
, &info
);
6475 case HOT_REMOVE_DISK
:
6476 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6480 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6483 case SET_DISK_FAULTY
:
6484 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6488 err
= do_md_run(mddev
);
6491 case SET_BITMAP_FILE
:
6492 err
= set_bitmap_file(mddev
, (int)arg
);
6502 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6504 mddev
->hold_active
= 0;
6505 mddev_unlock(mddev
);
6514 #ifdef CONFIG_COMPAT
6515 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6516 unsigned int cmd
, unsigned long arg
)
6519 case HOT_REMOVE_DISK
:
6521 case SET_DISK_FAULTY
:
6522 case SET_BITMAP_FILE
:
6523 /* These take in integer arg, do not convert */
6526 arg
= (unsigned long)compat_ptr(arg
);
6530 return md_ioctl(bdev
, mode
, cmd
, arg
);
6532 #endif /* CONFIG_COMPAT */
6534 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6537 * Succeed if we can lock the mddev, which confirms that
6538 * it isn't being stopped right now.
6540 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6546 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6547 /* we are racing with mddev_put which is discarding this
6551 /* Wait until bdev->bd_disk is definitely gone */
6552 flush_workqueue(md_misc_wq
);
6553 /* Then retry the open from the top */
6554 return -ERESTARTSYS
;
6556 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6558 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6562 atomic_inc(&mddev
->openers
);
6563 mutex_unlock(&mddev
->open_mutex
);
6565 check_disk_change(bdev
);
6570 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6572 struct mddev
*mddev
= disk
->private_data
;
6575 atomic_dec(&mddev
->openers
);
6581 static int md_media_changed(struct gendisk
*disk
)
6583 struct mddev
*mddev
= disk
->private_data
;
6585 return mddev
->changed
;
6588 static int md_revalidate(struct gendisk
*disk
)
6590 struct mddev
*mddev
= disk
->private_data
;
6595 static const struct block_device_operations md_fops
=
6597 .owner
= THIS_MODULE
,
6599 .release
= md_release
,
6601 #ifdef CONFIG_COMPAT
6602 .compat_ioctl
= md_compat_ioctl
,
6604 .getgeo
= md_getgeo
,
6605 .media_changed
= md_media_changed
,
6606 .revalidate_disk
= md_revalidate
,
6609 static int md_thread(void * arg
)
6611 struct md_thread
*thread
= arg
;
6614 * md_thread is a 'system-thread', it's priority should be very
6615 * high. We avoid resource deadlocks individually in each
6616 * raid personality. (RAID5 does preallocation) We also use RR and
6617 * the very same RT priority as kswapd, thus we will never get
6618 * into a priority inversion deadlock.
6620 * we definitely have to have equal or higher priority than
6621 * bdflush, otherwise bdflush will deadlock if there are too
6622 * many dirty RAID5 blocks.
6625 allow_signal(SIGKILL
);
6626 while (!kthread_should_stop()) {
6628 /* We need to wait INTERRUPTIBLE so that
6629 * we don't add to the load-average.
6630 * That means we need to be sure no signals are
6633 if (signal_pending(current
))
6634 flush_signals(current
);
6636 wait_event_interruptible_timeout
6638 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6639 || kthread_should_stop(),
6642 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6643 if (!kthread_should_stop())
6644 thread
->run(thread
->mddev
);
6650 void md_wakeup_thread(struct md_thread
*thread
)
6653 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6654 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6655 wake_up(&thread
->wqueue
);
6659 struct md_thread
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6662 struct md_thread
*thread
;
6664 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6668 init_waitqueue_head(&thread
->wqueue
);
6671 thread
->mddev
= mddev
;
6672 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6673 thread
->tsk
= kthread_run(md_thread
, thread
,
6675 mdname(thread
->mddev
),
6677 if (IS_ERR(thread
->tsk
)) {
6684 void md_unregister_thread(struct md_thread
**threadp
)
6686 struct md_thread
*thread
= *threadp
;
6689 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6690 /* Locking ensures that mddev_unlock does not wake_up a
6691 * non-existent thread
6693 spin_lock(&pers_lock
);
6695 spin_unlock(&pers_lock
);
6697 kthread_stop(thread
->tsk
);
6701 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6708 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6711 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6713 mddev
->pers
->error_handler(mddev
,rdev
);
6714 if (mddev
->degraded
)
6715 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6716 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6717 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6718 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6719 md_wakeup_thread(mddev
->thread
);
6720 if (mddev
->event_work
.func
)
6721 queue_work(md_misc_wq
, &mddev
->event_work
);
6722 md_new_event_inintr(mddev
);
6725 /* seq_file implementation /proc/mdstat */
6727 static void status_unused(struct seq_file
*seq
)
6730 struct md_rdev
*rdev
;
6732 seq_printf(seq
, "unused devices: ");
6734 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6735 char b
[BDEVNAME_SIZE
];
6737 seq_printf(seq
, "%s ",
6738 bdevname(rdev
->bdev
,b
));
6741 seq_printf(seq
, "<none>");
6743 seq_printf(seq
, "\n");
6747 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6749 sector_t max_sectors
, resync
, res
;
6750 unsigned long dt
, db
;
6753 unsigned int per_milli
;
6755 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6757 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6758 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6759 max_sectors
= mddev
->resync_max_sectors
;
6761 max_sectors
= mddev
->dev_sectors
;
6764 * Should not happen.
6770 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6771 * in a sector_t, and (max_sectors>>scale) will fit in a
6772 * u32, as those are the requirements for sector_div.
6773 * Thus 'scale' must be at least 10
6776 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6777 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6780 res
= (resync
>>scale
)*1000;
6781 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6785 int i
, x
= per_milli
/50, y
= 20-x
;
6786 seq_printf(seq
, "[");
6787 for (i
= 0; i
< x
; i
++)
6788 seq_printf(seq
, "=");
6789 seq_printf(seq
, ">");
6790 for (i
= 0; i
< y
; i
++)
6791 seq_printf(seq
, ".");
6792 seq_printf(seq
, "] ");
6794 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6795 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6797 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6799 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6800 "resync" : "recovery"))),
6801 per_milli
/10, per_milli
% 10,
6802 (unsigned long long) resync
/2,
6803 (unsigned long long) max_sectors
/2);
6806 * dt: time from mark until now
6807 * db: blocks written from mark until now
6808 * rt: remaining time
6810 * rt is a sector_t, so could be 32bit or 64bit.
6811 * So we divide before multiply in case it is 32bit and close
6813 * We scale the divisor (db) by 32 to avoid losing precision
6814 * near the end of resync when the number of remaining sectors
6816 * We then divide rt by 32 after multiplying by db to compensate.
6817 * The '+1' avoids division by zero if db is very small.
6819 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6821 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6822 - mddev
->resync_mark_cnt
;
6824 rt
= max_sectors
- resync
; /* number of remaining sectors */
6825 sector_div(rt
, db
/32+1);
6829 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6830 ((unsigned long)rt
% 60)/6);
6832 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6835 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6837 struct list_head
*tmp
;
6839 struct mddev
*mddev
;
6847 spin_lock(&all_mddevs_lock
);
6848 list_for_each(tmp
,&all_mddevs
)
6850 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6852 spin_unlock(&all_mddevs_lock
);
6855 spin_unlock(&all_mddevs_lock
);
6857 return (void*)2;/* tail */
6861 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6863 struct list_head
*tmp
;
6864 struct mddev
*next_mddev
, *mddev
= v
;
6870 spin_lock(&all_mddevs_lock
);
6872 tmp
= all_mddevs
.next
;
6874 tmp
= mddev
->all_mddevs
.next
;
6875 if (tmp
!= &all_mddevs
)
6876 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6878 next_mddev
= (void*)2;
6881 spin_unlock(&all_mddevs_lock
);
6889 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6891 struct mddev
*mddev
= v
;
6893 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6897 static int md_seq_show(struct seq_file
*seq
, void *v
)
6899 struct mddev
*mddev
= v
;
6901 struct md_rdev
*rdev
;
6903 if (v
== (void*)1) {
6904 struct md_personality
*pers
;
6905 seq_printf(seq
, "Personalities : ");
6906 spin_lock(&pers_lock
);
6907 list_for_each_entry(pers
, &pers_list
, list
)
6908 seq_printf(seq
, "[%s] ", pers
->name
);
6910 spin_unlock(&pers_lock
);
6911 seq_printf(seq
, "\n");
6912 seq
->poll_event
= atomic_read(&md_event_count
);
6915 if (v
== (void*)2) {
6920 if (mddev_lock(mddev
) < 0)
6923 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6924 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6925 mddev
->pers
? "" : "in");
6928 seq_printf(seq
, " (read-only)");
6930 seq_printf(seq
, " (auto-read-only)");
6931 seq_printf(seq
, " %s", mddev
->pers
->name
);
6935 rdev_for_each(rdev
, mddev
) {
6936 char b
[BDEVNAME_SIZE
];
6937 seq_printf(seq
, " %s[%d]",
6938 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6939 if (test_bit(WriteMostly
, &rdev
->flags
))
6940 seq_printf(seq
, "(W)");
6941 if (test_bit(Faulty
, &rdev
->flags
)) {
6942 seq_printf(seq
, "(F)");
6945 if (rdev
->raid_disk
< 0)
6946 seq_printf(seq
, "(S)"); /* spare */
6947 if (test_bit(Replacement
, &rdev
->flags
))
6948 seq_printf(seq
, "(R)");
6949 sectors
+= rdev
->sectors
;
6952 if (!list_empty(&mddev
->disks
)) {
6954 seq_printf(seq
, "\n %llu blocks",
6955 (unsigned long long)
6956 mddev
->array_sectors
/ 2);
6958 seq_printf(seq
, "\n %llu blocks",
6959 (unsigned long long)sectors
/ 2);
6961 if (mddev
->persistent
) {
6962 if (mddev
->major_version
!= 0 ||
6963 mddev
->minor_version
!= 90) {
6964 seq_printf(seq
," super %d.%d",
6965 mddev
->major_version
,
6966 mddev
->minor_version
);
6968 } else if (mddev
->external
)
6969 seq_printf(seq
, " super external:%s",
6970 mddev
->metadata_type
);
6972 seq_printf(seq
, " super non-persistent");
6975 mddev
->pers
->status(seq
, mddev
);
6976 seq_printf(seq
, "\n ");
6977 if (mddev
->pers
->sync_request
) {
6978 if (mddev
->curr_resync
> 2) {
6979 status_resync(seq
, mddev
);
6980 seq_printf(seq
, "\n ");
6981 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6982 seq_printf(seq
, "\tresync=DELAYED\n ");
6983 else if (mddev
->recovery_cp
< MaxSector
)
6984 seq_printf(seq
, "\tresync=PENDING\n ");
6987 seq_printf(seq
, "\n ");
6989 bitmap_status(seq
, mddev
->bitmap
);
6991 seq_printf(seq
, "\n");
6993 mddev_unlock(mddev
);
6998 static const struct seq_operations md_seq_ops
= {
6999 .start
= md_seq_start
,
7000 .next
= md_seq_next
,
7001 .stop
= md_seq_stop
,
7002 .show
= md_seq_show
,
7005 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7007 struct seq_file
*seq
;
7010 error
= seq_open(file
, &md_seq_ops
);
7014 seq
= file
->private_data
;
7015 seq
->poll_event
= atomic_read(&md_event_count
);
7019 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7021 struct seq_file
*seq
= filp
->private_data
;
7024 poll_wait(filp
, &md_event_waiters
, wait
);
7026 /* always allow read */
7027 mask
= POLLIN
| POLLRDNORM
;
7029 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7030 mask
|= POLLERR
| POLLPRI
;
7034 static const struct file_operations md_seq_fops
= {
7035 .owner
= THIS_MODULE
,
7036 .open
= md_seq_open
,
7038 .llseek
= seq_lseek
,
7039 .release
= seq_release_private
,
7040 .poll
= mdstat_poll
,
7043 int register_md_personality(struct md_personality
*p
)
7045 spin_lock(&pers_lock
);
7046 list_add_tail(&p
->list
, &pers_list
);
7047 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
7048 spin_unlock(&pers_lock
);
7052 int unregister_md_personality(struct md_personality
*p
)
7054 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7055 spin_lock(&pers_lock
);
7056 list_del_init(&p
->list
);
7057 spin_unlock(&pers_lock
);
7061 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7063 struct md_rdev
* rdev
;
7069 rdev_for_each_rcu(rdev
, mddev
) {
7070 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7071 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7072 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7073 atomic_read(&disk
->sync_io
);
7074 /* sync IO will cause sync_io to increase before the disk_stats
7075 * as sync_io is counted when a request starts, and
7076 * disk_stats is counted when it completes.
7077 * So resync activity will cause curr_events to be smaller than
7078 * when there was no such activity.
7079 * non-sync IO will cause disk_stat to increase without
7080 * increasing sync_io so curr_events will (eventually)
7081 * be larger than it was before. Once it becomes
7082 * substantially larger, the test below will cause
7083 * the array to appear non-idle, and resync will slow
7085 * If there is a lot of outstanding resync activity when
7086 * we set last_event to curr_events, then all that activity
7087 * completing might cause the array to appear non-idle
7088 * and resync will be slowed down even though there might
7089 * not have been non-resync activity. This will only
7090 * happen once though. 'last_events' will soon reflect
7091 * the state where there is little or no outstanding
7092 * resync requests, and further resync activity will
7093 * always make curr_events less than last_events.
7096 if (init
|| curr_events
- rdev
->last_events
> 64) {
7097 rdev
->last_events
= curr_events
;
7105 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7107 /* another "blocks" (512byte) blocks have been synced */
7108 atomic_sub(blocks
, &mddev
->recovery_active
);
7109 wake_up(&mddev
->recovery_wait
);
7111 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7112 md_wakeup_thread(mddev
->thread
);
7113 // stop recovery, signal do_sync ....
7118 /* md_write_start(mddev, bi)
7119 * If we need to update some array metadata (e.g. 'active' flag
7120 * in superblock) before writing, schedule a superblock update
7121 * and wait for it to complete.
7123 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7126 if (bio_data_dir(bi
) != WRITE
)
7129 BUG_ON(mddev
->ro
== 1);
7130 if (mddev
->ro
== 2) {
7131 /* need to switch to read/write */
7133 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7134 md_wakeup_thread(mddev
->thread
);
7135 md_wakeup_thread(mddev
->sync_thread
);
7138 atomic_inc(&mddev
->writes_pending
);
7139 if (mddev
->safemode
== 1)
7140 mddev
->safemode
= 0;
7141 if (mddev
->in_sync
) {
7142 spin_lock_irq(&mddev
->write_lock
);
7143 if (mddev
->in_sync
) {
7145 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7146 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7147 md_wakeup_thread(mddev
->thread
);
7150 spin_unlock_irq(&mddev
->write_lock
);
7153 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7154 wait_event(mddev
->sb_wait
,
7155 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7158 void md_write_end(struct mddev
*mddev
)
7160 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7161 if (mddev
->safemode
== 2)
7162 md_wakeup_thread(mddev
->thread
);
7163 else if (mddev
->safemode_delay
)
7164 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7168 /* md_allow_write(mddev)
7169 * Calling this ensures that the array is marked 'active' so that writes
7170 * may proceed without blocking. It is important to call this before
7171 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7172 * Must be called with mddev_lock held.
7174 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7175 * is dropped, so return -EAGAIN after notifying userspace.
7177 int md_allow_write(struct mddev
*mddev
)
7183 if (!mddev
->pers
->sync_request
)
7186 spin_lock_irq(&mddev
->write_lock
);
7187 if (mddev
->in_sync
) {
7189 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7190 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7191 if (mddev
->safemode_delay
&&
7192 mddev
->safemode
== 0)
7193 mddev
->safemode
= 1;
7194 spin_unlock_irq(&mddev
->write_lock
);
7195 md_update_sb(mddev
, 0);
7196 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7198 spin_unlock_irq(&mddev
->write_lock
);
7200 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7205 EXPORT_SYMBOL_GPL(md_allow_write
);
7207 #define SYNC_MARKS 10
7208 #define SYNC_MARK_STEP (3*HZ)
7209 void md_do_sync(struct mddev
*mddev
)
7211 struct mddev
*mddev2
;
7212 unsigned int currspeed
= 0,
7214 sector_t max_sectors
,j
, io_sectors
;
7215 unsigned long mark
[SYNC_MARKS
];
7216 sector_t mark_cnt
[SYNC_MARKS
];
7218 struct list_head
*tmp
;
7219 sector_t last_check
;
7221 struct md_rdev
*rdev
;
7223 struct blk_plug plug
;
7225 /* just incase thread restarts... */
7226 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7228 if (mddev
->ro
) /* never try to sync a read-only array */
7231 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7232 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7233 desc
= "data-check";
7234 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7235 desc
= "requested-resync";
7238 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7243 /* we overload curr_resync somewhat here.
7244 * 0 == not engaged in resync at all
7245 * 2 == checking that there is no conflict with another sync
7246 * 1 == like 2, but have yielded to allow conflicting resync to
7248 * other == active in resync - this many blocks
7250 * Before starting a resync we must have set curr_resync to
7251 * 2, and then checked that every "conflicting" array has curr_resync
7252 * less than ours. When we find one that is the same or higher
7253 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7254 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7255 * This will mean we have to start checking from the beginning again.
7260 mddev
->curr_resync
= 2;
7263 if (kthread_should_stop())
7264 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7266 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7268 for_each_mddev(mddev2
, tmp
) {
7269 if (mddev2
== mddev
)
7271 if (!mddev
->parallel_resync
7272 && mddev2
->curr_resync
7273 && match_mddev_units(mddev
, mddev2
)) {
7275 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7276 /* arbitrarily yield */
7277 mddev
->curr_resync
= 1;
7278 wake_up(&resync_wait
);
7280 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7281 /* no need to wait here, we can wait the next
7282 * time 'round when curr_resync == 2
7285 /* We need to wait 'interruptible' so as not to
7286 * contribute to the load average, and not to
7287 * be caught by 'softlockup'
7289 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7290 if (!kthread_should_stop() &&
7291 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7292 printk(KERN_INFO
"md: delaying %s of %s"
7293 " until %s has finished (they"
7294 " share one or more physical units)\n",
7295 desc
, mdname(mddev
), mdname(mddev2
));
7297 if (signal_pending(current
))
7298 flush_signals(current
);
7300 finish_wait(&resync_wait
, &wq
);
7303 finish_wait(&resync_wait
, &wq
);
7306 } while (mddev
->curr_resync
< 2);
7309 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7310 /* resync follows the size requested by the personality,
7311 * which defaults to physical size, but can be virtual size
7313 max_sectors
= mddev
->resync_max_sectors
;
7314 mddev
->resync_mismatches
= 0;
7315 /* we don't use the checkpoint if there's a bitmap */
7316 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7317 j
= mddev
->resync_min
;
7318 else if (!mddev
->bitmap
)
7319 j
= mddev
->recovery_cp
;
7321 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7322 max_sectors
= mddev
->resync_max_sectors
;
7324 /* recovery follows the physical size of devices */
7325 max_sectors
= mddev
->dev_sectors
;
7328 rdev_for_each_rcu(rdev
, mddev
)
7329 if (rdev
->raid_disk
>= 0 &&
7330 !test_bit(Faulty
, &rdev
->flags
) &&
7331 !test_bit(In_sync
, &rdev
->flags
) &&
7332 rdev
->recovery_offset
< j
)
7333 j
= rdev
->recovery_offset
;
7337 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7338 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7339 " %d KB/sec/disk.\n", speed_min(mddev
));
7340 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7341 "(but not more than %d KB/sec) for %s.\n",
7342 speed_max(mddev
), desc
);
7344 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7347 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7349 mark_cnt
[m
] = io_sectors
;
7352 mddev
->resync_mark
= mark
[last_mark
];
7353 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7356 * Tune reconstruction:
7358 window
= 32*(PAGE_SIZE
/512);
7359 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7360 window
/2, (unsigned long long)max_sectors
/2);
7362 atomic_set(&mddev
->recovery_active
, 0);
7367 "md: resuming %s of %s from checkpoint.\n",
7368 desc
, mdname(mddev
));
7369 mddev
->curr_resync
= j
;
7371 mddev
->curr_resync_completed
= j
;
7373 blk_start_plug(&plug
);
7374 while (j
< max_sectors
) {
7379 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7380 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7381 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7382 > (max_sectors
>> 4)) ||
7383 (j
- mddev
->curr_resync_completed
)*2
7384 >= mddev
->resync_max
- mddev
->curr_resync_completed
7386 /* time to update curr_resync_completed */
7387 wait_event(mddev
->recovery_wait
,
7388 atomic_read(&mddev
->recovery_active
) == 0);
7389 mddev
->curr_resync_completed
= j
;
7390 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7391 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7394 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7395 /* As this condition is controlled by user-space,
7396 * we can block indefinitely, so use '_interruptible'
7397 * to avoid triggering warnings.
7399 flush_signals(current
); /* just in case */
7400 wait_event_interruptible(mddev
->recovery_wait
,
7401 mddev
->resync_max
> j
7402 || kthread_should_stop());
7405 if (kthread_should_stop())
7408 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7409 currspeed
< speed_min(mddev
));
7411 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7415 if (!skipped
) { /* actual IO requested */
7416 io_sectors
+= sectors
;
7417 atomic_add(sectors
, &mddev
->recovery_active
);
7420 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7424 if (j
>1) mddev
->curr_resync
= j
;
7425 mddev
->curr_mark_cnt
= io_sectors
;
7426 if (last_check
== 0)
7427 /* this is the earliest that rebuild will be
7428 * visible in /proc/mdstat
7430 md_new_event(mddev
);
7432 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7435 last_check
= io_sectors
;
7437 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7439 int next
= (last_mark
+1) % SYNC_MARKS
;
7441 mddev
->resync_mark
= mark
[next
];
7442 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7443 mark
[next
] = jiffies
;
7444 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7449 if (kthread_should_stop())
7454 * this loop exits only if either when we are slower than
7455 * the 'hard' speed limit, or the system was IO-idle for
7457 * the system might be non-idle CPU-wise, but we only care
7458 * about not overloading the IO subsystem. (things like an
7459 * e2fsck being done on the RAID array should execute fast)
7463 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7464 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7466 if (currspeed
> speed_min(mddev
)) {
7467 if ((currspeed
> speed_max(mddev
)) ||
7468 !is_mddev_idle(mddev
, 0)) {
7474 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7476 * this also signals 'finished resyncing' to md_stop
7479 blk_finish_plug(&plug
);
7480 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7482 /* tell personality that we are finished */
7483 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7485 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7486 mddev
->curr_resync
> 2) {
7487 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7488 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7489 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7491 "md: checkpointing %s of %s.\n",
7492 desc
, mdname(mddev
));
7493 mddev
->recovery_cp
=
7494 mddev
->curr_resync_completed
;
7497 mddev
->recovery_cp
= MaxSector
;
7499 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7500 mddev
->curr_resync
= MaxSector
;
7502 rdev_for_each_rcu(rdev
, mddev
)
7503 if (rdev
->raid_disk
>= 0 &&
7504 mddev
->delta_disks
>= 0 &&
7505 !test_bit(Faulty
, &rdev
->flags
) &&
7506 !test_bit(In_sync
, &rdev
->flags
) &&
7507 rdev
->recovery_offset
< mddev
->curr_resync
)
7508 rdev
->recovery_offset
= mddev
->curr_resync
;
7513 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7515 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7516 /* We completed so min/max setting can be forgotten if used. */
7517 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7518 mddev
->resync_min
= 0;
7519 mddev
->resync_max
= MaxSector
;
7520 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7521 mddev
->resync_min
= mddev
->curr_resync_completed
;
7522 mddev
->curr_resync
= 0;
7523 wake_up(&resync_wait
);
7524 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7525 md_wakeup_thread(mddev
->thread
);
7530 * got a signal, exit.
7533 "md: md_do_sync() got signal ... exiting\n");
7534 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7538 EXPORT_SYMBOL_GPL(md_do_sync
);
7540 static int remove_and_add_spares(struct mddev
*mddev
)
7542 struct md_rdev
*rdev
;
7546 mddev
->curr_resync_completed
= 0;
7548 rdev_for_each(rdev
, mddev
)
7549 if (rdev
->raid_disk
>= 0 &&
7550 !test_bit(Blocked
, &rdev
->flags
) &&
7551 (test_bit(Faulty
, &rdev
->flags
) ||
7552 ! test_bit(In_sync
, &rdev
->flags
)) &&
7553 atomic_read(&rdev
->nr_pending
)==0) {
7554 if (mddev
->pers
->hot_remove_disk(
7555 mddev
, rdev
) == 0) {
7556 sysfs_unlink_rdev(mddev
, rdev
);
7557 rdev
->raid_disk
= -1;
7562 sysfs_notify(&mddev
->kobj
, NULL
,
7566 rdev_for_each(rdev
, mddev
) {
7567 if (rdev
->raid_disk
>= 0 &&
7568 !test_bit(In_sync
, &rdev
->flags
) &&
7569 !test_bit(Faulty
, &rdev
->flags
))
7571 if (rdev
->raid_disk
< 0
7572 && !test_bit(Faulty
, &rdev
->flags
)) {
7573 rdev
->recovery_offset
= 0;
7575 hot_add_disk(mddev
, rdev
) == 0) {
7576 if (sysfs_link_rdev(mddev
, rdev
))
7577 /* failure here is OK */;
7579 md_new_event(mddev
);
7580 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7585 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7589 static void reap_sync_thread(struct mddev
*mddev
)
7591 struct md_rdev
*rdev
;
7593 /* resync has finished, collect result */
7594 md_unregister_thread(&mddev
->sync_thread
);
7595 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7596 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7598 /* activate any spares */
7599 if (mddev
->pers
->spare_active(mddev
)) {
7600 sysfs_notify(&mddev
->kobj
, NULL
,
7602 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7605 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7606 mddev
->pers
->finish_reshape
)
7607 mddev
->pers
->finish_reshape(mddev
);
7609 /* If array is no-longer degraded, then any saved_raid_disk
7610 * information must be scrapped. Also if any device is now
7611 * In_sync we must scrape the saved_raid_disk for that device
7612 * do the superblock for an incrementally recovered device
7615 rdev_for_each(rdev
, mddev
)
7616 if (!mddev
->degraded
||
7617 test_bit(In_sync
, &rdev
->flags
))
7618 rdev
->saved_raid_disk
= -1;
7620 md_update_sb(mddev
, 1);
7621 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7622 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7623 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7624 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7625 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7626 /* flag recovery needed just to double check */
7627 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7628 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7629 md_new_event(mddev
);
7630 if (mddev
->event_work
.func
)
7631 queue_work(md_misc_wq
, &mddev
->event_work
);
7635 * This routine is regularly called by all per-raid-array threads to
7636 * deal with generic issues like resync and super-block update.
7637 * Raid personalities that don't have a thread (linear/raid0) do not
7638 * need this as they never do any recovery or update the superblock.
7640 * It does not do any resync itself, but rather "forks" off other threads
7641 * to do that as needed.
7642 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7643 * "->recovery" and create a thread at ->sync_thread.
7644 * When the thread finishes it sets MD_RECOVERY_DONE
7645 * and wakeups up this thread which will reap the thread and finish up.
7646 * This thread also removes any faulty devices (with nr_pending == 0).
7648 * The overall approach is:
7649 * 1/ if the superblock needs updating, update it.
7650 * 2/ If a recovery thread is running, don't do anything else.
7651 * 3/ If recovery has finished, clean up, possibly marking spares active.
7652 * 4/ If there are any faulty devices, remove them.
7653 * 5/ If array is degraded, try to add spares devices
7654 * 6/ If array has spares or is not in-sync, start a resync thread.
7656 void md_check_recovery(struct mddev
*mddev
)
7658 if (mddev
->suspended
)
7662 bitmap_daemon_work(mddev
);
7664 if (signal_pending(current
)) {
7665 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7666 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7668 mddev
->safemode
= 2;
7670 flush_signals(current
);
7673 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7676 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7677 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7678 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7679 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7680 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7681 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7685 if (mddev_trylock(mddev
)) {
7689 /* Only thing we do on a ro array is remove
7692 struct md_rdev
*rdev
;
7693 rdev_for_each(rdev
, mddev
)
7694 if (rdev
->raid_disk
>= 0 &&
7695 !test_bit(Blocked
, &rdev
->flags
) &&
7696 test_bit(Faulty
, &rdev
->flags
) &&
7697 atomic_read(&rdev
->nr_pending
)==0) {
7698 if (mddev
->pers
->hot_remove_disk(
7699 mddev
, rdev
) == 0) {
7700 sysfs_unlink_rdev(mddev
, rdev
);
7701 rdev
->raid_disk
= -1;
7704 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7708 if (!mddev
->external
) {
7710 spin_lock_irq(&mddev
->write_lock
);
7711 if (mddev
->safemode
&&
7712 !atomic_read(&mddev
->writes_pending
) &&
7714 mddev
->recovery_cp
== MaxSector
) {
7717 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7719 if (mddev
->safemode
== 1)
7720 mddev
->safemode
= 0;
7721 spin_unlock_irq(&mddev
->write_lock
);
7723 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7727 md_update_sb(mddev
, 0);
7729 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7730 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7731 /* resync/recovery still happening */
7732 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7735 if (mddev
->sync_thread
) {
7736 reap_sync_thread(mddev
);
7739 /* Set RUNNING before clearing NEEDED to avoid
7740 * any transients in the value of "sync_action".
7742 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7743 /* Clear some bits that don't mean anything, but
7746 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7747 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7749 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7750 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7752 /* no recovery is running.
7753 * remove any failed drives, then
7754 * add spares if possible.
7755 * Spare are also removed and re-added, to allow
7756 * the personality to fail the re-add.
7759 if (mddev
->reshape_position
!= MaxSector
) {
7760 if (mddev
->pers
->check_reshape
== NULL
||
7761 mddev
->pers
->check_reshape(mddev
) != 0)
7762 /* Cannot proceed */
7764 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7765 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7766 } else if ((spares
= remove_and_add_spares(mddev
))) {
7767 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7768 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7769 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7770 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7771 } else if (mddev
->recovery_cp
< MaxSector
) {
7772 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7773 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7774 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7775 /* nothing to be done ... */
7778 if (mddev
->pers
->sync_request
) {
7780 /* We are adding a device or devices to an array
7781 * which has the bitmap stored on all devices.
7782 * So make sure all bitmap pages get written
7784 bitmap_write_all(mddev
->bitmap
);
7786 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7789 if (!mddev
->sync_thread
) {
7790 printk(KERN_ERR
"%s: could not start resync"
7793 /* leave the spares where they are, it shouldn't hurt */
7794 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7795 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7796 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7797 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7798 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7800 md_wakeup_thread(mddev
->sync_thread
);
7801 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7802 md_new_event(mddev
);
7805 if (!mddev
->sync_thread
) {
7806 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7807 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7809 if (mddev
->sysfs_action
)
7810 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7812 mddev_unlock(mddev
);
7816 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7818 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7819 wait_event_timeout(rdev
->blocked_wait
,
7820 !test_bit(Blocked
, &rdev
->flags
) &&
7821 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7822 msecs_to_jiffies(5000));
7823 rdev_dec_pending(rdev
, mddev
);
7825 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7827 void md_finish_reshape(struct mddev
*mddev
)
7829 /* called be personality module when reshape completes. */
7830 struct md_rdev
*rdev
;
7832 rdev_for_each(rdev
, mddev
) {
7833 if (rdev
->data_offset
> rdev
->new_data_offset
)
7834 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
7836 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
7837 rdev
->data_offset
= rdev
->new_data_offset
;
7840 EXPORT_SYMBOL(md_finish_reshape
);
7842 /* Bad block management.
7843 * We can record which blocks on each device are 'bad' and so just
7844 * fail those blocks, or that stripe, rather than the whole device.
7845 * Entries in the bad-block table are 64bits wide. This comprises:
7846 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7847 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7848 * A 'shift' can be set so that larger blocks are tracked and
7849 * consequently larger devices can be covered.
7850 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7852 * Locking of the bad-block table uses a seqlock so md_is_badblock
7853 * might need to retry if it is very unlucky.
7854 * We will sometimes want to check for bad blocks in a bi_end_io function,
7855 * so we use the write_seqlock_irq variant.
7857 * When looking for a bad block we specify a range and want to
7858 * know if any block in the range is bad. So we binary-search
7859 * to the last range that starts at-or-before the given endpoint,
7860 * (or "before the sector after the target range")
7861 * then see if it ends after the given start.
7863 * 0 if there are no known bad blocks in the range
7864 * 1 if there are known bad block which are all acknowledged
7865 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7866 * plus the start/length of the first bad section we overlap.
7868 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7869 sector_t
*first_bad
, int *bad_sectors
)
7875 sector_t target
= s
+ sectors
;
7878 if (bb
->shift
> 0) {
7879 /* round the start down, and the end up */
7881 target
+= (1<<bb
->shift
) - 1;
7882 target
>>= bb
->shift
;
7883 sectors
= target
- s
;
7885 /* 'target' is now the first block after the bad range */
7888 seq
= read_seqbegin(&bb
->lock
);
7892 /* Binary search between lo and hi for 'target'
7893 * i.e. for the last range that starts before 'target'
7895 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7896 * are known not to be the last range before target.
7897 * VARIANT: hi-lo is the number of possible
7898 * ranges, and decreases until it reaches 1
7900 while (hi
- lo
> 1) {
7901 int mid
= (lo
+ hi
) / 2;
7902 sector_t a
= BB_OFFSET(p
[mid
]);
7904 /* This could still be the one, earlier ranges
7908 /* This and later ranges are definitely out. */
7911 /* 'lo' might be the last that started before target, but 'hi' isn't */
7913 /* need to check all range that end after 's' to see if
7914 * any are unacknowledged.
7917 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7918 if (BB_OFFSET(p
[lo
]) < target
) {
7919 /* starts before the end, and finishes after
7920 * the start, so they must overlap
7922 if (rv
!= -1 && BB_ACK(p
[lo
]))
7926 *first_bad
= BB_OFFSET(p
[lo
]);
7927 *bad_sectors
= BB_LEN(p
[lo
]);
7933 if (read_seqretry(&bb
->lock
, seq
))
7938 EXPORT_SYMBOL_GPL(md_is_badblock
);
7941 * Add a range of bad blocks to the table.
7942 * This might extend the table, or might contract it
7943 * if two adjacent ranges can be merged.
7944 * We binary-search to find the 'insertion' point, then
7945 * decide how best to handle it.
7947 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7955 /* badblocks are disabled */
7959 /* round the start down, and the end up */
7960 sector_t next
= s
+ sectors
;
7962 next
+= (1<<bb
->shift
) - 1;
7967 write_seqlock_irq(&bb
->lock
);
7972 /* Find the last range that starts at-or-before 's' */
7973 while (hi
- lo
> 1) {
7974 int mid
= (lo
+ hi
) / 2;
7975 sector_t a
= BB_OFFSET(p
[mid
]);
7981 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7985 /* we found a range that might merge with the start
7988 sector_t a
= BB_OFFSET(p
[lo
]);
7989 sector_t e
= a
+ BB_LEN(p
[lo
]);
7990 int ack
= BB_ACK(p
[lo
]);
7992 /* Yes, we can merge with a previous range */
7993 if (s
== a
&& s
+ sectors
>= e
)
7994 /* new range covers old */
7997 ack
= ack
&& acknowledged
;
7999 if (e
< s
+ sectors
)
8001 if (e
- a
<= BB_MAX_LEN
) {
8002 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8005 /* does not all fit in one range,
8006 * make p[lo] maximal
8008 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8009 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8015 if (sectors
&& hi
< bb
->count
) {
8016 /* 'hi' points to the first range that starts after 's'.
8017 * Maybe we can merge with the start of that range */
8018 sector_t a
= BB_OFFSET(p
[hi
]);
8019 sector_t e
= a
+ BB_LEN(p
[hi
]);
8020 int ack
= BB_ACK(p
[hi
]);
8021 if (a
<= s
+ sectors
) {
8022 /* merging is possible */
8023 if (e
<= s
+ sectors
) {
8028 ack
= ack
&& acknowledged
;
8031 if (e
- a
<= BB_MAX_LEN
) {
8032 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8035 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8043 if (sectors
== 0 && hi
< bb
->count
) {
8044 /* we might be able to combine lo and hi */
8045 /* Note: 's' is at the end of 'lo' */
8046 sector_t a
= BB_OFFSET(p
[hi
]);
8047 int lolen
= BB_LEN(p
[lo
]);
8048 int hilen
= BB_LEN(p
[hi
]);
8049 int newlen
= lolen
+ hilen
- (s
- a
);
8050 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8051 /* yes, we can combine them */
8052 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8053 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8054 memmove(p
+ hi
, p
+ hi
+ 1,
8055 (bb
->count
- hi
- 1) * 8);
8060 /* didn't merge (it all).
8061 * Need to add a range just before 'hi' */
8062 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8063 /* No room for more */
8067 int this_sectors
= sectors
;
8068 memmove(p
+ hi
+ 1, p
+ hi
,
8069 (bb
->count
- hi
) * 8);
8072 if (this_sectors
> BB_MAX_LEN
)
8073 this_sectors
= BB_MAX_LEN
;
8074 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8075 sectors
-= this_sectors
;
8082 bb
->unacked_exist
= 1;
8083 write_sequnlock_irq(&bb
->lock
);
8088 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8093 s
+= rdev
->new_data_offset
;
8095 s
+= rdev
->data_offset
;
8096 rv
= md_set_badblocks(&rdev
->badblocks
,
8099 /* Make sure they get written out promptly */
8100 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8101 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8102 md_wakeup_thread(rdev
->mddev
->thread
);
8106 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8109 * Remove a range of bad blocks from the table.
8110 * This may involve extending the table if we spilt a region,
8111 * but it must not fail. So if the table becomes full, we just
8112 * drop the remove request.
8114 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8118 sector_t target
= s
+ sectors
;
8121 if (bb
->shift
> 0) {
8122 /* When clearing we round the start up and the end down.
8123 * This should not matter as the shift should align with
8124 * the block size and no rounding should ever be needed.
8125 * However it is better the think a block is bad when it
8126 * isn't than to think a block is not bad when it is.
8128 s
+= (1<<bb
->shift
) - 1;
8130 target
>>= bb
->shift
;
8131 sectors
= target
- s
;
8134 write_seqlock_irq(&bb
->lock
);
8139 /* Find the last range that starts before 'target' */
8140 while (hi
- lo
> 1) {
8141 int mid
= (lo
+ hi
) / 2;
8142 sector_t a
= BB_OFFSET(p
[mid
]);
8149 /* p[lo] is the last range that could overlap the
8150 * current range. Earlier ranges could also overlap,
8151 * but only this one can overlap the end of the range.
8153 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8154 /* Partial overlap, leave the tail of this range */
8155 int ack
= BB_ACK(p
[lo
]);
8156 sector_t a
= BB_OFFSET(p
[lo
]);
8157 sector_t end
= a
+ BB_LEN(p
[lo
]);
8160 /* we need to split this range */
8161 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8165 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8167 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8170 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8171 /* there is no longer an overlap */
8176 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8177 /* This range does overlap */
8178 if (BB_OFFSET(p
[lo
]) < s
) {
8179 /* Keep the early parts of this range. */
8180 int ack
= BB_ACK(p
[lo
]);
8181 sector_t start
= BB_OFFSET(p
[lo
]);
8182 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8183 /* now low doesn't overlap, so.. */
8188 /* 'lo' is strictly before, 'hi' is strictly after,
8189 * anything between needs to be discarded
8192 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8193 bb
->count
-= (hi
- lo
- 1);
8199 write_sequnlock_irq(&bb
->lock
);
8203 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8207 s
+= rdev
->new_data_offset
;
8209 s
+= rdev
->data_offset
;
8210 return md_clear_badblocks(&rdev
->badblocks
,
8213 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8216 * Acknowledge all bad blocks in a list.
8217 * This only succeeds if ->changed is clear. It is used by
8218 * in-kernel metadata updates
8220 void md_ack_all_badblocks(struct badblocks
*bb
)
8222 if (bb
->page
== NULL
|| bb
->changed
)
8223 /* no point even trying */
8225 write_seqlock_irq(&bb
->lock
);
8227 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8230 for (i
= 0; i
< bb
->count
; i
++) {
8231 if (!BB_ACK(p
[i
])) {
8232 sector_t start
= BB_OFFSET(p
[i
]);
8233 int len
= BB_LEN(p
[i
]);
8234 p
[i
] = BB_MAKE(start
, len
, 1);
8237 bb
->unacked_exist
= 0;
8239 write_sequnlock_irq(&bb
->lock
);
8241 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8243 /* sysfs access to bad-blocks list.
8244 * We present two files.
8245 * 'bad-blocks' lists sector numbers and lengths of ranges that
8246 * are recorded as bad. The list is truncated to fit within
8247 * the one-page limit of sysfs.
8248 * Writing "sector length" to this file adds an acknowledged
8250 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8251 * been acknowledged. Writing to this file adds bad blocks
8252 * without acknowledging them. This is largely for testing.
8256 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8267 seq
= read_seqbegin(&bb
->lock
);
8272 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8273 sector_t s
= BB_OFFSET(p
[i
]);
8274 unsigned int length
= BB_LEN(p
[i
]);
8275 int ack
= BB_ACK(p
[i
]);
8281 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8282 (unsigned long long)s
<< bb
->shift
,
8283 length
<< bb
->shift
);
8285 if (unack
&& len
== 0)
8286 bb
->unacked_exist
= 0;
8288 if (read_seqretry(&bb
->lock
, seq
))
8297 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8299 unsigned long long sector
;
8303 /* Allow clearing via sysfs *only* for testing/debugging.
8304 * Normally only a successful write may clear a badblock
8307 if (page
[0] == '-') {
8311 #endif /* DO_DEBUG */
8313 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8315 if (newline
!= '\n')
8327 md_clear_badblocks(bb
, sector
, length
);
8330 #endif /* DO_DEBUG */
8331 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8337 static int md_notify_reboot(struct notifier_block
*this,
8338 unsigned long code
, void *x
)
8340 struct list_head
*tmp
;
8341 struct mddev
*mddev
;
8344 for_each_mddev(mddev
, tmp
) {
8345 if (mddev_trylock(mddev
)) {
8347 __md_stop_writes(mddev
);
8348 mddev
->safemode
= 2;
8349 mddev_unlock(mddev
);
8354 * certain more exotic SCSI devices are known to be
8355 * volatile wrt too early system reboots. While the
8356 * right place to handle this issue is the given
8357 * driver, we do want to have a safe RAID driver ...
8365 static struct notifier_block md_notifier
= {
8366 .notifier_call
= md_notify_reboot
,
8368 .priority
= INT_MAX
, /* before any real devices */
8371 static void md_geninit(void)
8373 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8375 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8378 static int __init
md_init(void)
8382 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8386 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8390 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8393 if ((ret
= register_blkdev(0, "mdp")) < 0)
8397 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8398 md_probe
, NULL
, NULL
);
8399 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8400 md_probe
, NULL
, NULL
);
8402 register_reboot_notifier(&md_notifier
);
8403 raid_table_header
= register_sysctl_table(raid_root_table
);
8409 unregister_blkdev(MD_MAJOR
, "md");
8411 destroy_workqueue(md_misc_wq
);
8413 destroy_workqueue(md_wq
);
8421 * Searches all registered partitions for autorun RAID arrays
8425 static LIST_HEAD(all_detected_devices
);
8426 struct detected_devices_node
{
8427 struct list_head list
;
8431 void md_autodetect_dev(dev_t dev
)
8433 struct detected_devices_node
*node_detected_dev
;
8435 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8436 if (node_detected_dev
) {
8437 node_detected_dev
->dev
= dev
;
8438 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8440 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8441 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8446 static void autostart_arrays(int part
)
8448 struct md_rdev
*rdev
;
8449 struct detected_devices_node
*node_detected_dev
;
8451 int i_scanned
, i_passed
;
8456 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8458 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8460 node_detected_dev
= list_entry(all_detected_devices
.next
,
8461 struct detected_devices_node
, list
);
8462 list_del(&node_detected_dev
->list
);
8463 dev
= node_detected_dev
->dev
;
8464 kfree(node_detected_dev
);
8465 rdev
= md_import_device(dev
,0, 90);
8469 if (test_bit(Faulty
, &rdev
->flags
)) {
8473 set_bit(AutoDetected
, &rdev
->flags
);
8474 list_add(&rdev
->same_set
, &pending_raid_disks
);
8478 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8479 i_scanned
, i_passed
);
8481 autorun_devices(part
);
8484 #endif /* !MODULE */
8486 static __exit
void md_exit(void)
8488 struct mddev
*mddev
;
8489 struct list_head
*tmp
;
8491 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8492 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8494 unregister_blkdev(MD_MAJOR
,"md");
8495 unregister_blkdev(mdp_major
, "mdp");
8496 unregister_reboot_notifier(&md_notifier
);
8497 unregister_sysctl_table(raid_table_header
);
8498 remove_proc_entry("mdstat", NULL
);
8499 for_each_mddev(mddev
, tmp
) {
8500 export_array(mddev
);
8501 mddev
->hold_active
= 0;
8503 destroy_workqueue(md_misc_wq
);
8504 destroy_workqueue(md_wq
);
8507 subsys_initcall(md_init
);
8508 module_exit(md_exit
)
8510 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8512 return sprintf(buffer
, "%d", start_readonly
);
8514 static int set_ro(const char *val
, struct kernel_param
*kp
)
8517 int num
= simple_strtoul(val
, &e
, 10);
8518 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8519 start_readonly
= num
;
8525 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8526 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8528 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8530 EXPORT_SYMBOL(register_md_personality
);
8531 EXPORT_SYMBOL(unregister_md_personality
);
8532 EXPORT_SYMBOL(md_error
);
8533 EXPORT_SYMBOL(md_done_sync
);
8534 EXPORT_SYMBOL(md_write_start
);
8535 EXPORT_SYMBOL(md_write_end
);
8536 EXPORT_SYMBOL(md_register_thread
);
8537 EXPORT_SYMBOL(md_unregister_thread
);
8538 EXPORT_SYMBOL(md_wakeup_thread
);
8539 EXPORT_SYMBOL(md_check_recovery
);
8540 MODULE_LICENSE("GPL");
8541 MODULE_DESCRIPTION("MD RAID framework");
8543 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);