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/freezer.h>
37 #include <linux/blkdev.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
59 static void autostart_arrays(int part
);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list
);
68 static DEFINE_SPINLOCK(pers_lock
);
70 static void md_print_devices(void);
72 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
73 static struct workqueue_struct
*md_wq
;
74 static struct workqueue_struct
*md_misc_wq
;
76 static int remove_and_add_spares(struct mddev
*mddev
,
77 struct md_rdev
*this);
79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
82 * Default number of read corrections we'll attempt on an rdev
83 * before ejecting it from the array. We divide the read error
84 * count by 2 for every hour elapsed between read errors.
86 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
88 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
89 * is 1000 KB/sec, so the extra system load does not show up that much.
90 * Increase it if you want to have more _guaranteed_ speed. Note that
91 * the RAID driver will use the maximum available bandwidth if the IO
92 * subsystem is idle. There is also an 'absolute maximum' reconstruction
93 * speed limit - in case reconstruction slows down your system despite
96 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
97 * or /sys/block/mdX/md/sync_speed_{min,max}
100 static int sysctl_speed_limit_min
= 1000;
101 static int sysctl_speed_limit_max
= 200000;
102 static inline int speed_min(struct mddev
*mddev
)
104 return mddev
->sync_speed_min
?
105 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
108 static inline int speed_max(struct mddev
*mddev
)
110 return mddev
->sync_speed_max
?
111 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
114 static struct ctl_table_header
*raid_table_header
;
116 static ctl_table raid_table
[] = {
118 .procname
= "speed_limit_min",
119 .data
= &sysctl_speed_limit_min
,
120 .maxlen
= sizeof(int),
121 .mode
= S_IRUGO
|S_IWUSR
,
122 .proc_handler
= proc_dointvec
,
125 .procname
= "speed_limit_max",
126 .data
= &sysctl_speed_limit_max
,
127 .maxlen
= sizeof(int),
128 .mode
= S_IRUGO
|S_IWUSR
,
129 .proc_handler
= proc_dointvec
,
134 static ctl_table raid_dir_table
[] = {
138 .mode
= S_IRUGO
|S_IXUGO
,
144 static ctl_table raid_root_table
[] = {
149 .child
= raid_dir_table
,
154 static const struct block_device_operations md_fops
;
156 static int start_readonly
;
159 * like bio_clone, but with a local bio set
162 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
167 if (!mddev
|| !mddev
->bio_set
)
168 return bio_alloc(gfp_mask
, nr_iovecs
);
170 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
175 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
177 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
180 if (!mddev
|| !mddev
->bio_set
)
181 return bio_clone(bio
, gfp_mask
);
183 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
185 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
187 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
189 /* 'bio' is a cloned bio which we need to trim to match
190 * the given offset and size.
191 * This requires adjusting bi_sector, bi_size, and bi_io_vec
194 struct bio_vec
*bvec
;
198 if (offset
== 0 && size
== bio
->bi_size
)
201 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
203 bio_advance(bio
, offset
<< 9);
207 /* avoid any complications with bi_idx being non-zero*/
209 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
210 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
211 bio
->bi_vcnt
-= bio
->bi_idx
;
214 /* Make sure vcnt and last bv are not too big */
215 bio_for_each_segment(bvec
, bio
, i
) {
216 if (sofar
+ bvec
->bv_len
> size
)
217 bvec
->bv_len
= size
- sofar
;
218 if (bvec
->bv_len
== 0) {
222 sofar
+= bvec
->bv_len
;
225 EXPORT_SYMBOL_GPL(md_trim_bio
);
228 * We have a system wide 'event count' that is incremented
229 * on any 'interesting' event, and readers of /proc/mdstat
230 * can use 'poll' or 'select' to find out when the event
234 * start array, stop array, error, add device, remove device,
235 * start build, activate spare
237 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
238 static atomic_t md_event_count
;
239 void md_new_event(struct mddev
*mddev
)
241 atomic_inc(&md_event_count
);
242 wake_up(&md_event_waiters
);
244 EXPORT_SYMBOL_GPL(md_new_event
);
246 /* Alternate version that can be called from interrupts
247 * when calling sysfs_notify isn't needed.
249 static void md_new_event_inintr(struct mddev
*mddev
)
251 atomic_inc(&md_event_count
);
252 wake_up(&md_event_waiters
);
256 * Enables to iterate over all existing md arrays
257 * all_mddevs_lock protects this list.
259 static LIST_HEAD(all_mddevs
);
260 static DEFINE_SPINLOCK(all_mddevs_lock
);
264 * iterates through all used mddevs in the system.
265 * We take care to grab the all_mddevs_lock whenever navigating
266 * the list, and to always hold a refcount when unlocked.
267 * Any code which breaks out of this loop while own
268 * a reference to the current mddev and must mddev_put it.
270 #define for_each_mddev(_mddev,_tmp) \
272 for (({ spin_lock(&all_mddevs_lock); \
273 _tmp = all_mddevs.next; \
275 ({ if (_tmp != &all_mddevs) \
276 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
277 spin_unlock(&all_mddevs_lock); \
278 if (_mddev) mddev_put(_mddev); \
279 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
280 _tmp != &all_mddevs;}); \
281 ({ spin_lock(&all_mddevs_lock); \
282 _tmp = _tmp->next;}) \
286 /* Rather than calling directly into the personality make_request function,
287 * IO requests come here first so that we can check if the device is
288 * being suspended pending a reconfiguration.
289 * We hold a refcount over the call to ->make_request. By the time that
290 * call has finished, the bio has been linked into some internal structure
291 * and so is visible to ->quiesce(), so we don't need the refcount any more.
293 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
295 const int rw
= bio_data_dir(bio
);
296 struct mddev
*mddev
= q
->queuedata
;
298 unsigned int sectors
;
300 if (mddev
== NULL
|| mddev
->pers
== NULL
305 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
306 bio_endio(bio
, bio_sectors(bio
) == 0 ? 0 : -EROFS
);
309 smp_rmb(); /* Ensure implications of 'active' are visible */
311 if (mddev
->suspended
) {
314 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
315 TASK_UNINTERRUPTIBLE
);
316 if (!mddev
->suspended
)
322 finish_wait(&mddev
->sb_wait
, &__wait
);
324 atomic_inc(&mddev
->active_io
);
328 * save the sectors now since our bio can
329 * go away inside make_request
331 sectors
= bio_sectors(bio
);
332 mddev
->pers
->make_request(mddev
, bio
);
334 cpu
= part_stat_lock();
335 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
336 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
339 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
340 wake_up(&mddev
->sb_wait
);
343 /* mddev_suspend makes sure no new requests are submitted
344 * to the device, and that any requests that have been submitted
345 * are completely handled.
346 * Once ->stop is called and completes, the module will be completely
349 void mddev_suspend(struct mddev
*mddev
)
351 BUG_ON(mddev
->suspended
);
352 mddev
->suspended
= 1;
354 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
355 mddev
->pers
->quiesce(mddev
, 1);
357 del_timer_sync(&mddev
->safemode_timer
);
359 EXPORT_SYMBOL_GPL(mddev_suspend
);
361 void mddev_resume(struct mddev
*mddev
)
363 mddev
->suspended
= 0;
364 wake_up(&mddev
->sb_wait
);
365 mddev
->pers
->quiesce(mddev
, 0);
367 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
368 md_wakeup_thread(mddev
->thread
);
369 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
371 EXPORT_SYMBOL_GPL(mddev_resume
);
373 int mddev_congested(struct mddev
*mddev
, int bits
)
375 return mddev
->suspended
;
377 EXPORT_SYMBOL(mddev_congested
);
380 * Generic flush handling for md
383 static void md_end_flush(struct bio
*bio
, int err
)
385 struct md_rdev
*rdev
= bio
->bi_private
;
386 struct mddev
*mddev
= rdev
->mddev
;
388 rdev_dec_pending(rdev
, mddev
);
390 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
391 /* The pre-request flush has finished */
392 queue_work(md_wq
, &mddev
->flush_work
);
397 static void md_submit_flush_data(struct work_struct
*ws
);
399 static void submit_flushes(struct work_struct
*ws
)
401 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
402 struct md_rdev
*rdev
;
404 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
405 atomic_set(&mddev
->flush_pending
, 1);
407 rdev_for_each_rcu(rdev
, mddev
)
408 if (rdev
->raid_disk
>= 0 &&
409 !test_bit(Faulty
, &rdev
->flags
)) {
410 /* Take two references, one is dropped
411 * when request finishes, one after
412 * we reclaim rcu_read_lock
415 atomic_inc(&rdev
->nr_pending
);
416 atomic_inc(&rdev
->nr_pending
);
418 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
419 bi
->bi_end_io
= md_end_flush
;
420 bi
->bi_private
= rdev
;
421 bi
->bi_bdev
= rdev
->bdev
;
422 atomic_inc(&mddev
->flush_pending
);
423 submit_bio(WRITE_FLUSH
, bi
);
425 rdev_dec_pending(rdev
, mddev
);
428 if (atomic_dec_and_test(&mddev
->flush_pending
))
429 queue_work(md_wq
, &mddev
->flush_work
);
432 static void md_submit_flush_data(struct work_struct
*ws
)
434 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
435 struct bio
*bio
= mddev
->flush_bio
;
437 if (bio
->bi_size
== 0)
438 /* an empty barrier - all done */
441 bio
->bi_rw
&= ~REQ_FLUSH
;
442 mddev
->pers
->make_request(mddev
, bio
);
445 mddev
->flush_bio
= NULL
;
446 wake_up(&mddev
->sb_wait
);
449 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
451 spin_lock_irq(&mddev
->write_lock
);
452 wait_event_lock_irq(mddev
->sb_wait
,
455 mddev
->flush_bio
= bio
;
456 spin_unlock_irq(&mddev
->write_lock
);
458 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
459 queue_work(md_wq
, &mddev
->flush_work
);
461 EXPORT_SYMBOL(md_flush_request
);
463 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
465 struct mddev
*mddev
= cb
->data
;
466 md_wakeup_thread(mddev
->thread
);
469 EXPORT_SYMBOL(md_unplug
);
471 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
473 atomic_inc(&mddev
->active
);
477 static void mddev_delayed_delete(struct work_struct
*ws
);
479 static void mddev_put(struct mddev
*mddev
)
481 struct bio_set
*bs
= NULL
;
483 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
485 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
486 mddev
->ctime
== 0 && !mddev
->hold_active
) {
487 /* Array is not configured at all, and not held active,
489 list_del_init(&mddev
->all_mddevs
);
491 mddev
->bio_set
= NULL
;
492 if (mddev
->gendisk
) {
493 /* We did a probe so need to clean up. Call
494 * queue_work inside the spinlock so that
495 * flush_workqueue() after mddev_find will
496 * succeed in waiting for the work to be done.
498 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
499 queue_work(md_misc_wq
, &mddev
->del_work
);
503 spin_unlock(&all_mddevs_lock
);
508 void mddev_init(struct mddev
*mddev
)
510 mutex_init(&mddev
->open_mutex
);
511 mutex_init(&mddev
->reconfig_mutex
);
512 mutex_init(&mddev
->bitmap_info
.mutex
);
513 INIT_LIST_HEAD(&mddev
->disks
);
514 INIT_LIST_HEAD(&mddev
->all_mddevs
);
515 init_timer(&mddev
->safemode_timer
);
516 atomic_set(&mddev
->active
, 1);
517 atomic_set(&mddev
->openers
, 0);
518 atomic_set(&mddev
->active_io
, 0);
519 spin_lock_init(&mddev
->write_lock
);
520 atomic_set(&mddev
->flush_pending
, 0);
521 init_waitqueue_head(&mddev
->sb_wait
);
522 init_waitqueue_head(&mddev
->recovery_wait
);
523 mddev
->reshape_position
= MaxSector
;
524 mddev
->reshape_backwards
= 0;
525 mddev
->resync_min
= 0;
526 mddev
->resync_max
= MaxSector
;
527 mddev
->level
= LEVEL_NONE
;
529 EXPORT_SYMBOL_GPL(mddev_init
);
531 static struct mddev
* mddev_find(dev_t unit
)
533 struct mddev
*mddev
, *new = NULL
;
535 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
536 unit
&= ~((1<<MdpMinorShift
)-1);
539 spin_lock(&all_mddevs_lock
);
542 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
543 if (mddev
->unit
== unit
) {
545 spin_unlock(&all_mddevs_lock
);
551 list_add(&new->all_mddevs
, &all_mddevs
);
552 spin_unlock(&all_mddevs_lock
);
553 new->hold_active
= UNTIL_IOCTL
;
557 /* find an unused unit number */
558 static int next_minor
= 512;
559 int start
= next_minor
;
563 dev
= MKDEV(MD_MAJOR
, next_minor
);
565 if (next_minor
> MINORMASK
)
567 if (next_minor
== start
) {
568 /* Oh dear, all in use. */
569 spin_unlock(&all_mddevs_lock
);
575 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
576 if (mddev
->unit
== dev
) {
582 new->md_minor
= MINOR(dev
);
583 new->hold_active
= UNTIL_STOP
;
584 list_add(&new->all_mddevs
, &all_mddevs
);
585 spin_unlock(&all_mddevs_lock
);
588 spin_unlock(&all_mddevs_lock
);
590 new = kzalloc(sizeof(*new), GFP_KERNEL
);
595 if (MAJOR(unit
) == MD_MAJOR
)
596 new->md_minor
= MINOR(unit
);
598 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
605 static inline int mddev_lock(struct mddev
* mddev
)
607 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
610 static inline int mddev_is_locked(struct mddev
*mddev
)
612 return mutex_is_locked(&mddev
->reconfig_mutex
);
615 static inline int mddev_trylock(struct mddev
* mddev
)
617 return mutex_trylock(&mddev
->reconfig_mutex
);
620 static struct attribute_group md_redundancy_group
;
622 static void mddev_unlock(struct mddev
* mddev
)
624 if (mddev
->to_remove
) {
625 /* These cannot be removed under reconfig_mutex as
626 * an access to the files will try to take reconfig_mutex
627 * while holding the file unremovable, which leads to
629 * So hold set sysfs_active while the remove in happeing,
630 * and anything else which might set ->to_remove or my
631 * otherwise change the sysfs namespace will fail with
632 * -EBUSY if sysfs_active is still set.
633 * We set sysfs_active under reconfig_mutex and elsewhere
634 * test it under the same mutex to ensure its correct value
637 struct attribute_group
*to_remove
= mddev
->to_remove
;
638 mddev
->to_remove
= NULL
;
639 mddev
->sysfs_active
= 1;
640 mutex_unlock(&mddev
->reconfig_mutex
);
642 if (mddev
->kobj
.sd
) {
643 if (to_remove
!= &md_redundancy_group
)
644 sysfs_remove_group(&mddev
->kobj
, to_remove
);
645 if (mddev
->pers
== NULL
||
646 mddev
->pers
->sync_request
== NULL
) {
647 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
648 if (mddev
->sysfs_action
)
649 sysfs_put(mddev
->sysfs_action
);
650 mddev
->sysfs_action
= NULL
;
653 mddev
->sysfs_active
= 0;
655 mutex_unlock(&mddev
->reconfig_mutex
);
657 /* As we've dropped the mutex we need a spinlock to
658 * make sure the thread doesn't disappear
660 spin_lock(&pers_lock
);
661 md_wakeup_thread(mddev
->thread
);
662 spin_unlock(&pers_lock
);
665 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
667 struct md_rdev
*rdev
;
669 rdev_for_each(rdev
, mddev
)
670 if (rdev
->desc_nr
== nr
)
676 static struct md_rdev
*find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
678 struct md_rdev
*rdev
;
680 rdev_for_each_rcu(rdev
, mddev
)
681 if (rdev
->desc_nr
== nr
)
687 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
689 struct md_rdev
*rdev
;
691 rdev_for_each(rdev
, mddev
)
692 if (rdev
->bdev
->bd_dev
== dev
)
698 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
700 struct md_rdev
*rdev
;
702 rdev_for_each_rcu(rdev
, mddev
)
703 if (rdev
->bdev
->bd_dev
== dev
)
709 static struct md_personality
*find_pers(int level
, char *clevel
)
711 struct md_personality
*pers
;
712 list_for_each_entry(pers
, &pers_list
, list
) {
713 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
715 if (strcmp(pers
->name
, clevel
)==0)
721 /* return the offset of the super block in 512byte sectors */
722 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
724 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
725 return MD_NEW_SIZE_SECTORS(num_sectors
);
728 static int alloc_disk_sb(struct md_rdev
* rdev
)
733 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
734 if (!rdev
->sb_page
) {
735 printk(KERN_ALERT
"md: out of memory.\n");
742 void md_rdev_clear(struct md_rdev
*rdev
)
745 put_page(rdev
->sb_page
);
747 rdev
->sb_page
= NULL
;
752 put_page(rdev
->bb_page
);
753 rdev
->bb_page
= NULL
;
755 kfree(rdev
->badblocks
.page
);
756 rdev
->badblocks
.page
= NULL
;
758 EXPORT_SYMBOL_GPL(md_rdev_clear
);
760 static void super_written(struct bio
*bio
, int error
)
762 struct md_rdev
*rdev
= bio
->bi_private
;
763 struct mddev
*mddev
= rdev
->mddev
;
765 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
766 printk("md: super_written gets error=%d, uptodate=%d\n",
767 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
768 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
769 md_error(mddev
, rdev
);
772 if (atomic_dec_and_test(&mddev
->pending_writes
))
773 wake_up(&mddev
->sb_wait
);
777 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
778 sector_t sector
, int size
, struct page
*page
)
780 /* write first size bytes of page to sector of rdev
781 * Increment mddev->pending_writes before returning
782 * and decrement it on completion, waking up sb_wait
783 * if zero is reached.
784 * If an error occurred, call md_error
786 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
788 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
789 bio
->bi_sector
= sector
;
790 bio_add_page(bio
, page
, size
, 0);
791 bio
->bi_private
= rdev
;
792 bio
->bi_end_io
= super_written
;
794 atomic_inc(&mddev
->pending_writes
);
795 submit_bio(WRITE_FLUSH_FUA
, bio
);
798 void md_super_wait(struct mddev
*mddev
)
800 /* wait for all superblock writes that were scheduled to complete */
803 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
804 if (atomic_read(&mddev
->pending_writes
)==0)
808 finish_wait(&mddev
->sb_wait
, &wq
);
811 static void bi_complete(struct bio
*bio
, int error
)
813 complete((struct completion
*)bio
->bi_private
);
816 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
817 struct page
*page
, int rw
, bool metadata_op
)
819 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
820 struct completion event
;
825 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
826 rdev
->meta_bdev
: rdev
->bdev
;
828 bio
->bi_sector
= sector
+ rdev
->sb_start
;
829 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
830 (rdev
->mddev
->reshape_backwards
==
831 (sector
>= rdev
->mddev
->reshape_position
)))
832 bio
->bi_sector
= sector
+ rdev
->new_data_offset
;
834 bio
->bi_sector
= sector
+ rdev
->data_offset
;
835 bio_add_page(bio
, page
, size
, 0);
836 init_completion(&event
);
837 bio
->bi_private
= &event
;
838 bio
->bi_end_io
= bi_complete
;
840 wait_for_completion(&event
);
842 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
846 EXPORT_SYMBOL_GPL(sync_page_io
);
848 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
850 char b
[BDEVNAME_SIZE
];
851 if (!rdev
->sb_page
) {
859 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
865 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
866 bdevname(rdev
->bdev
,b
));
870 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
872 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
873 sb1
->set_uuid1
== sb2
->set_uuid1
&&
874 sb1
->set_uuid2
== sb2
->set_uuid2
&&
875 sb1
->set_uuid3
== sb2
->set_uuid3
;
878 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
881 mdp_super_t
*tmp1
, *tmp2
;
883 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
884 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
886 if (!tmp1
|| !tmp2
) {
888 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
896 * nr_disks is not constant
901 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
909 static u32
md_csum_fold(u32 csum
)
911 csum
= (csum
& 0xffff) + (csum
>> 16);
912 return (csum
& 0xffff) + (csum
>> 16);
915 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
918 u32
*sb32
= (u32
*)sb
;
920 unsigned int disk_csum
, csum
;
922 disk_csum
= sb
->sb_csum
;
925 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
927 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
931 /* This used to use csum_partial, which was wrong for several
932 * reasons including that different results are returned on
933 * different architectures. It isn't critical that we get exactly
934 * the same return value as before (we always csum_fold before
935 * testing, and that removes any differences). However as we
936 * know that csum_partial always returned a 16bit value on
937 * alphas, do a fold to maximise conformity to previous behaviour.
939 sb
->sb_csum
= md_csum_fold(disk_csum
);
941 sb
->sb_csum
= disk_csum
;
948 * Handle superblock details.
949 * We want to be able to handle multiple superblock formats
950 * so we have a common interface to them all, and an array of
951 * different handlers.
952 * We rely on user-space to write the initial superblock, and support
953 * reading and updating of superblocks.
954 * Interface methods are:
955 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
956 * loads and validates a superblock on dev.
957 * if refdev != NULL, compare superblocks on both devices
959 * 0 - dev has a superblock that is compatible with refdev
960 * 1 - dev has a superblock that is compatible and newer than refdev
961 * so dev should be used as the refdev in future
962 * -EINVAL superblock incompatible or invalid
963 * -othererror e.g. -EIO
965 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
966 * Verify that dev is acceptable into mddev.
967 * The first time, mddev->raid_disks will be 0, and data from
968 * dev should be merged in. Subsequent calls check that dev
969 * is new enough. Return 0 or -EINVAL
971 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
972 * Update the superblock for rdev with data in mddev
973 * This does not write to disc.
979 struct module
*owner
;
980 int (*load_super
)(struct md_rdev
*rdev
,
981 struct md_rdev
*refdev
,
983 int (*validate_super
)(struct mddev
*mddev
,
984 struct md_rdev
*rdev
);
985 void (*sync_super
)(struct mddev
*mddev
,
986 struct md_rdev
*rdev
);
987 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
988 sector_t num_sectors
);
989 int (*allow_new_offset
)(struct md_rdev
*rdev
,
990 unsigned long long new_offset
);
994 * Check that the given mddev has no bitmap.
996 * This function is called from the run method of all personalities that do not
997 * support bitmaps. It prints an error message and returns non-zero if mddev
998 * has a bitmap. Otherwise, it returns 0.
1001 int md_check_no_bitmap(struct mddev
*mddev
)
1003 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1005 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1006 mdname(mddev
), mddev
->pers
->name
);
1009 EXPORT_SYMBOL(md_check_no_bitmap
);
1012 * load_super for 0.90.0
1014 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1016 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1021 * Calculate the position of the superblock (512byte sectors),
1022 * it's at the end of the disk.
1024 * It also happens to be a multiple of 4Kb.
1026 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1028 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1029 if (ret
) return ret
;
1033 bdevname(rdev
->bdev
, b
);
1034 sb
= page_address(rdev
->sb_page
);
1036 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1037 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1042 if (sb
->major_version
!= 0 ||
1043 sb
->minor_version
< 90 ||
1044 sb
->minor_version
> 91) {
1045 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1046 sb
->major_version
, sb
->minor_version
,
1051 if (sb
->raid_disks
<= 0)
1054 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1055 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1060 rdev
->preferred_minor
= sb
->md_minor
;
1061 rdev
->data_offset
= 0;
1062 rdev
->new_data_offset
= 0;
1063 rdev
->sb_size
= MD_SB_BYTES
;
1064 rdev
->badblocks
.shift
= -1;
1066 if (sb
->level
== LEVEL_MULTIPATH
)
1069 rdev
->desc_nr
= sb
->this_disk
.number
;
1075 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1076 if (!uuid_equal(refsb
, sb
)) {
1077 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1078 b
, bdevname(refdev
->bdev
,b2
));
1081 if (!sb_equal(refsb
, sb
)) {
1082 printk(KERN_WARNING
"md: %s has same UUID"
1083 " but different superblock to %s\n",
1084 b
, bdevname(refdev
->bdev
, b2
));
1088 ev2
= md_event(refsb
);
1094 rdev
->sectors
= rdev
->sb_start
;
1095 /* Limit to 4TB as metadata cannot record more than that.
1096 * (not needed for Linear and RAID0 as metadata doesn't
1099 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1100 rdev
->sectors
= (2ULL << 32) - 2;
1102 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1103 /* "this cannot possibly happen" ... */
1111 * validate_super for 0.90.0
1113 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1116 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1117 __u64 ev1
= md_event(sb
);
1119 rdev
->raid_disk
= -1;
1120 clear_bit(Faulty
, &rdev
->flags
);
1121 clear_bit(In_sync
, &rdev
->flags
);
1122 clear_bit(Bitmap_sync
, &rdev
->flags
);
1123 clear_bit(WriteMostly
, &rdev
->flags
);
1125 if (mddev
->raid_disks
== 0) {
1126 mddev
->major_version
= 0;
1127 mddev
->minor_version
= sb
->minor_version
;
1128 mddev
->patch_version
= sb
->patch_version
;
1129 mddev
->external
= 0;
1130 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1131 mddev
->ctime
= sb
->ctime
;
1132 mddev
->utime
= sb
->utime
;
1133 mddev
->level
= sb
->level
;
1134 mddev
->clevel
[0] = 0;
1135 mddev
->layout
= sb
->layout
;
1136 mddev
->raid_disks
= sb
->raid_disks
;
1137 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1138 mddev
->events
= ev1
;
1139 mddev
->bitmap_info
.offset
= 0;
1140 mddev
->bitmap_info
.space
= 0;
1141 /* bitmap can use 60 K after the 4K superblocks */
1142 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1143 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1144 mddev
->reshape_backwards
= 0;
1146 if (mddev
->minor_version
>= 91) {
1147 mddev
->reshape_position
= sb
->reshape_position
;
1148 mddev
->delta_disks
= sb
->delta_disks
;
1149 mddev
->new_level
= sb
->new_level
;
1150 mddev
->new_layout
= sb
->new_layout
;
1151 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1152 if (mddev
->delta_disks
< 0)
1153 mddev
->reshape_backwards
= 1;
1155 mddev
->reshape_position
= MaxSector
;
1156 mddev
->delta_disks
= 0;
1157 mddev
->new_level
= mddev
->level
;
1158 mddev
->new_layout
= mddev
->layout
;
1159 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1162 if (sb
->state
& (1<<MD_SB_CLEAN
))
1163 mddev
->recovery_cp
= MaxSector
;
1165 if (sb
->events_hi
== sb
->cp_events_hi
&&
1166 sb
->events_lo
== sb
->cp_events_lo
) {
1167 mddev
->recovery_cp
= sb
->recovery_cp
;
1169 mddev
->recovery_cp
= 0;
1172 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1173 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1174 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1175 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1177 mddev
->max_disks
= MD_SB_DISKS
;
1179 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1180 mddev
->bitmap_info
.file
== NULL
) {
1181 mddev
->bitmap_info
.offset
=
1182 mddev
->bitmap_info
.default_offset
;
1183 mddev
->bitmap_info
.space
=
1184 mddev
->bitmap_info
.space
;
1187 } else if (mddev
->pers
== NULL
) {
1188 /* Insist on good event counter while assembling, except
1189 * for spares (which don't need an event count) */
1191 if (sb
->disks
[rdev
->desc_nr
].state
& (
1192 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1193 if (ev1
< mddev
->events
)
1195 } else if (mddev
->bitmap
) {
1196 /* if adding to array with a bitmap, then we can accept an
1197 * older device ... but not too old.
1199 if (ev1
< mddev
->bitmap
->events_cleared
)
1201 if (ev1
< mddev
->events
)
1202 set_bit(Bitmap_sync
, &rdev
->flags
);
1204 if (ev1
< mddev
->events
)
1205 /* just a hot-add of a new device, leave raid_disk at -1 */
1209 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1210 desc
= sb
->disks
+ rdev
->desc_nr
;
1212 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1213 set_bit(Faulty
, &rdev
->flags
);
1214 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1215 desc->raid_disk < mddev->raid_disks */) {
1216 set_bit(In_sync
, &rdev
->flags
);
1217 rdev
->raid_disk
= desc
->raid_disk
;
1218 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1219 /* active but not in sync implies recovery up to
1220 * reshape position. We don't know exactly where
1221 * that is, so set to zero for now */
1222 if (mddev
->minor_version
>= 91) {
1223 rdev
->recovery_offset
= 0;
1224 rdev
->raid_disk
= desc
->raid_disk
;
1227 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1228 set_bit(WriteMostly
, &rdev
->flags
);
1229 } else /* MULTIPATH are always insync */
1230 set_bit(In_sync
, &rdev
->flags
);
1235 * sync_super for 0.90.0
1237 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1240 struct md_rdev
*rdev2
;
1241 int next_spare
= mddev
->raid_disks
;
1244 /* make rdev->sb match mddev data..
1247 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1248 * 3/ any empty disks < next_spare become removed
1250 * disks[0] gets initialised to REMOVED because
1251 * we cannot be sure from other fields if it has
1252 * been initialised or not.
1255 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1257 rdev
->sb_size
= MD_SB_BYTES
;
1259 sb
= page_address(rdev
->sb_page
);
1261 memset(sb
, 0, sizeof(*sb
));
1263 sb
->md_magic
= MD_SB_MAGIC
;
1264 sb
->major_version
= mddev
->major_version
;
1265 sb
->patch_version
= mddev
->patch_version
;
1266 sb
->gvalid_words
= 0; /* ignored */
1267 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1268 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1269 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1270 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1272 sb
->ctime
= mddev
->ctime
;
1273 sb
->level
= mddev
->level
;
1274 sb
->size
= mddev
->dev_sectors
/ 2;
1275 sb
->raid_disks
= mddev
->raid_disks
;
1276 sb
->md_minor
= mddev
->md_minor
;
1277 sb
->not_persistent
= 0;
1278 sb
->utime
= mddev
->utime
;
1280 sb
->events_hi
= (mddev
->events
>>32);
1281 sb
->events_lo
= (u32
)mddev
->events
;
1283 if (mddev
->reshape_position
== MaxSector
)
1284 sb
->minor_version
= 90;
1286 sb
->minor_version
= 91;
1287 sb
->reshape_position
= mddev
->reshape_position
;
1288 sb
->new_level
= mddev
->new_level
;
1289 sb
->delta_disks
= mddev
->delta_disks
;
1290 sb
->new_layout
= mddev
->new_layout
;
1291 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1293 mddev
->minor_version
= sb
->minor_version
;
1296 sb
->recovery_cp
= mddev
->recovery_cp
;
1297 sb
->cp_events_hi
= (mddev
->events
>>32);
1298 sb
->cp_events_lo
= (u32
)mddev
->events
;
1299 if (mddev
->recovery_cp
== MaxSector
)
1300 sb
->state
= (1<< MD_SB_CLEAN
);
1302 sb
->recovery_cp
= 0;
1304 sb
->layout
= mddev
->layout
;
1305 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1307 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1308 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1310 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1311 rdev_for_each(rdev2
, mddev
) {
1314 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1316 if (rdev2
->raid_disk
>= 0 &&
1317 sb
->minor_version
>= 91)
1318 /* we have nowhere to store the recovery_offset,
1319 * but if it is not below the reshape_position,
1320 * we can piggy-back on that.
1323 if (rdev2
->raid_disk
< 0 ||
1324 test_bit(Faulty
, &rdev2
->flags
))
1327 desc_nr
= rdev2
->raid_disk
;
1329 desc_nr
= next_spare
++;
1330 rdev2
->desc_nr
= desc_nr
;
1331 d
= &sb
->disks
[rdev2
->desc_nr
];
1333 d
->number
= rdev2
->desc_nr
;
1334 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1335 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1337 d
->raid_disk
= rdev2
->raid_disk
;
1339 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1340 if (test_bit(Faulty
, &rdev2
->flags
))
1341 d
->state
= (1<<MD_DISK_FAULTY
);
1342 else if (is_active
) {
1343 d
->state
= (1<<MD_DISK_ACTIVE
);
1344 if (test_bit(In_sync
, &rdev2
->flags
))
1345 d
->state
|= (1<<MD_DISK_SYNC
);
1353 if (test_bit(WriteMostly
, &rdev2
->flags
))
1354 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1356 /* now set the "removed" and "faulty" bits on any missing devices */
1357 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1358 mdp_disk_t
*d
= &sb
->disks
[i
];
1359 if (d
->state
== 0 && d
->number
== 0) {
1362 d
->state
= (1<<MD_DISK_REMOVED
);
1363 d
->state
|= (1<<MD_DISK_FAULTY
);
1367 sb
->nr_disks
= nr_disks
;
1368 sb
->active_disks
= active
;
1369 sb
->working_disks
= working
;
1370 sb
->failed_disks
= failed
;
1371 sb
->spare_disks
= spare
;
1373 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1374 sb
->sb_csum
= calc_sb_csum(sb
);
1378 * rdev_size_change for 0.90.0
1380 static unsigned long long
1381 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1383 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1384 return 0; /* component must fit device */
1385 if (rdev
->mddev
->bitmap_info
.offset
)
1386 return 0; /* can't move bitmap */
1387 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1388 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1389 num_sectors
= rdev
->sb_start
;
1390 /* Limit to 4TB as metadata cannot record more than that.
1391 * 4TB == 2^32 KB, or 2*2^32 sectors.
1393 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1394 num_sectors
= (2ULL << 32) - 2;
1395 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1397 md_super_wait(rdev
->mddev
);
1402 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1404 /* non-zero offset changes not possible with v0.90 */
1405 return new_offset
== 0;
1409 * version 1 superblock
1412 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1416 unsigned long long newcsum
;
1417 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1418 __le32
*isuper
= (__le32
*)sb
;
1420 disk_csum
= sb
->sb_csum
;
1423 for (; size
>= 4; size
-= 4)
1424 newcsum
+= le32_to_cpu(*isuper
++);
1427 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1429 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1430 sb
->sb_csum
= disk_csum
;
1431 return cpu_to_le32(csum
);
1434 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1436 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1438 struct mdp_superblock_1
*sb
;
1442 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1446 * Calculate the position of the superblock in 512byte sectors.
1447 * It is always aligned to a 4K boundary and
1448 * depeding on minor_version, it can be:
1449 * 0: At least 8K, but less than 12K, from end of device
1450 * 1: At start of device
1451 * 2: 4K from start of device.
1453 switch(minor_version
) {
1455 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1457 sb_start
&= ~(sector_t
)(4*2-1);
1468 rdev
->sb_start
= sb_start
;
1470 /* superblock is rarely larger than 1K, but it can be larger,
1471 * and it is safe to read 4k, so we do that
1473 ret
= read_disk_sb(rdev
, 4096);
1474 if (ret
) return ret
;
1477 sb
= page_address(rdev
->sb_page
);
1479 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1480 sb
->major_version
!= cpu_to_le32(1) ||
1481 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1482 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1483 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1486 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1487 printk("md: invalid superblock checksum on %s\n",
1488 bdevname(rdev
->bdev
,b
));
1491 if (le64_to_cpu(sb
->data_size
) < 10) {
1492 printk("md: data_size too small on %s\n",
1493 bdevname(rdev
->bdev
,b
));
1498 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1499 /* Some padding is non-zero, might be a new feature */
1502 rdev
->preferred_minor
= 0xffff;
1503 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1504 rdev
->new_data_offset
= rdev
->data_offset
;
1505 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1506 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1507 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1508 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1510 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1511 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1512 if (rdev
->sb_size
& bmask
)
1513 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1516 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1519 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1522 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1525 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1527 if (!rdev
->bb_page
) {
1528 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1532 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1533 rdev
->badblocks
.count
== 0) {
1534 /* need to load the bad block list.
1535 * Currently we limit it to one page.
1541 int sectors
= le16_to_cpu(sb
->bblog_size
);
1542 if (sectors
> (PAGE_SIZE
/ 512))
1544 offset
= le32_to_cpu(sb
->bblog_offset
);
1547 bb_sector
= (long long)offset
;
1548 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1549 rdev
->bb_page
, READ
, true))
1551 bbp
= (u64
*)page_address(rdev
->bb_page
);
1552 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1553 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1554 u64 bb
= le64_to_cpu(*bbp
);
1555 int count
= bb
& (0x3ff);
1556 u64 sector
= bb
>> 10;
1557 sector
<<= sb
->bblog_shift
;
1558 count
<<= sb
->bblog_shift
;
1561 if (md_set_badblocks(&rdev
->badblocks
,
1562 sector
, count
, 1) == 0)
1565 } else if (sb
->bblog_offset
!= 0)
1566 rdev
->badblocks
.shift
= 0;
1572 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1574 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1575 sb
->level
!= refsb
->level
||
1576 sb
->layout
!= refsb
->layout
||
1577 sb
->chunksize
!= refsb
->chunksize
) {
1578 printk(KERN_WARNING
"md: %s has strangely different"
1579 " superblock to %s\n",
1580 bdevname(rdev
->bdev
,b
),
1581 bdevname(refdev
->bdev
,b2
));
1584 ev1
= le64_to_cpu(sb
->events
);
1585 ev2
= le64_to_cpu(refsb
->events
);
1592 if (minor_version
) {
1593 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1594 sectors
-= rdev
->data_offset
;
1596 sectors
= rdev
->sb_start
;
1597 if (sectors
< le64_to_cpu(sb
->data_size
))
1599 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1603 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1605 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1606 __u64 ev1
= le64_to_cpu(sb
->events
);
1608 rdev
->raid_disk
= -1;
1609 clear_bit(Faulty
, &rdev
->flags
);
1610 clear_bit(In_sync
, &rdev
->flags
);
1611 clear_bit(Bitmap_sync
, &rdev
->flags
);
1612 clear_bit(WriteMostly
, &rdev
->flags
);
1614 if (mddev
->raid_disks
== 0) {
1615 mddev
->major_version
= 1;
1616 mddev
->patch_version
= 0;
1617 mddev
->external
= 0;
1618 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1619 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1620 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1621 mddev
->level
= le32_to_cpu(sb
->level
);
1622 mddev
->clevel
[0] = 0;
1623 mddev
->layout
= le32_to_cpu(sb
->layout
);
1624 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1625 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1626 mddev
->events
= ev1
;
1627 mddev
->bitmap_info
.offset
= 0;
1628 mddev
->bitmap_info
.space
= 0;
1629 /* Default location for bitmap is 1K after superblock
1630 * using 3K - total of 4K
1632 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1633 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1634 mddev
->reshape_backwards
= 0;
1636 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1637 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1639 mddev
->max_disks
= (4096-256)/2;
1641 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1642 mddev
->bitmap_info
.file
== NULL
) {
1643 mddev
->bitmap_info
.offset
=
1644 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1645 /* Metadata doesn't record how much space is available.
1646 * For 1.0, we assume we can use up to the superblock
1647 * if before, else to 4K beyond superblock.
1648 * For others, assume no change is possible.
1650 if (mddev
->minor_version
> 0)
1651 mddev
->bitmap_info
.space
= 0;
1652 else if (mddev
->bitmap_info
.offset
> 0)
1653 mddev
->bitmap_info
.space
=
1654 8 - mddev
->bitmap_info
.offset
;
1656 mddev
->bitmap_info
.space
=
1657 -mddev
->bitmap_info
.offset
;
1660 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1661 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1662 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1663 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1664 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1665 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1666 if (mddev
->delta_disks
< 0 ||
1667 (mddev
->delta_disks
== 0 &&
1668 (le32_to_cpu(sb
->feature_map
)
1669 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1670 mddev
->reshape_backwards
= 1;
1672 mddev
->reshape_position
= MaxSector
;
1673 mddev
->delta_disks
= 0;
1674 mddev
->new_level
= mddev
->level
;
1675 mddev
->new_layout
= mddev
->layout
;
1676 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1679 } else if (mddev
->pers
== NULL
) {
1680 /* Insist of good event counter while assembling, except for
1681 * spares (which don't need an event count) */
1683 if (rdev
->desc_nr
>= 0 &&
1684 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1685 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1686 if (ev1
< mddev
->events
)
1688 } else if (mddev
->bitmap
) {
1689 /* If adding to array with a bitmap, then we can accept an
1690 * older device, but not too old.
1692 if (ev1
< mddev
->bitmap
->events_cleared
)
1694 if (ev1
< mddev
->events
)
1695 set_bit(Bitmap_sync
, &rdev
->flags
);
1697 if (ev1
< mddev
->events
)
1698 /* just a hot-add of a new device, leave raid_disk at -1 */
1701 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1703 if (rdev
->desc_nr
< 0 ||
1704 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1708 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1710 case 0xffff: /* spare */
1712 case 0xfffe: /* faulty */
1713 set_bit(Faulty
, &rdev
->flags
);
1716 if ((le32_to_cpu(sb
->feature_map
) &
1717 MD_FEATURE_RECOVERY_OFFSET
))
1718 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1720 set_bit(In_sync
, &rdev
->flags
);
1721 rdev
->raid_disk
= role
;
1724 if (sb
->devflags
& WriteMostly1
)
1725 set_bit(WriteMostly
, &rdev
->flags
);
1726 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1727 set_bit(Replacement
, &rdev
->flags
);
1728 } else /* MULTIPATH are always insync */
1729 set_bit(In_sync
, &rdev
->flags
);
1734 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1736 struct mdp_superblock_1
*sb
;
1737 struct md_rdev
*rdev2
;
1739 /* make rdev->sb match mddev and rdev data. */
1741 sb
= page_address(rdev
->sb_page
);
1743 sb
->feature_map
= 0;
1745 sb
->recovery_offset
= cpu_to_le64(0);
1746 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1748 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1749 sb
->events
= cpu_to_le64(mddev
->events
);
1751 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1753 sb
->resync_offset
= cpu_to_le64(0);
1755 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1757 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1758 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1759 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1760 sb
->level
= cpu_to_le32(mddev
->level
);
1761 sb
->layout
= cpu_to_le32(mddev
->layout
);
1763 if (test_bit(WriteMostly
, &rdev
->flags
))
1764 sb
->devflags
|= WriteMostly1
;
1766 sb
->devflags
&= ~WriteMostly1
;
1767 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1768 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1770 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1771 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1772 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1775 if (rdev
->raid_disk
>= 0 &&
1776 !test_bit(In_sync
, &rdev
->flags
)) {
1778 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1779 sb
->recovery_offset
=
1780 cpu_to_le64(rdev
->recovery_offset
);
1782 if (test_bit(Replacement
, &rdev
->flags
))
1784 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1786 if (mddev
->reshape_position
!= MaxSector
) {
1787 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1788 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1789 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1790 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1791 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1792 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1793 if (mddev
->delta_disks
== 0 &&
1794 mddev
->reshape_backwards
)
1796 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1797 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1799 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1800 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1801 - rdev
->data_offset
));
1805 if (rdev
->badblocks
.count
== 0)
1806 /* Nothing to do for bad blocks*/ ;
1807 else if (sb
->bblog_offset
== 0)
1808 /* Cannot record bad blocks on this device */
1809 md_error(mddev
, rdev
);
1811 struct badblocks
*bb
= &rdev
->badblocks
;
1812 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1814 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1819 seq
= read_seqbegin(&bb
->lock
);
1821 memset(bbp
, 0xff, PAGE_SIZE
);
1823 for (i
= 0 ; i
< bb
->count
; i
++) {
1824 u64 internal_bb
= p
[i
];
1825 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1826 | BB_LEN(internal_bb
));
1827 bbp
[i
] = cpu_to_le64(store_bb
);
1830 if (read_seqretry(&bb
->lock
, seq
))
1833 bb
->sector
= (rdev
->sb_start
+
1834 (int)le32_to_cpu(sb
->bblog_offset
));
1835 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1840 rdev_for_each(rdev2
, mddev
)
1841 if (rdev2
->desc_nr
+1 > max_dev
)
1842 max_dev
= rdev2
->desc_nr
+1;
1844 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1846 sb
->max_dev
= cpu_to_le32(max_dev
);
1847 rdev
->sb_size
= max_dev
* 2 + 256;
1848 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1849 if (rdev
->sb_size
& bmask
)
1850 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1852 max_dev
= le32_to_cpu(sb
->max_dev
);
1854 for (i
=0; i
<max_dev
;i
++)
1855 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1857 rdev_for_each(rdev2
, mddev
) {
1859 if (test_bit(Faulty
, &rdev2
->flags
))
1860 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1861 else if (test_bit(In_sync
, &rdev2
->flags
))
1862 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1863 else if (rdev2
->raid_disk
>= 0)
1864 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1866 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1869 sb
->sb_csum
= calc_sb_1_csum(sb
);
1872 static unsigned long long
1873 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1875 struct mdp_superblock_1
*sb
;
1876 sector_t max_sectors
;
1877 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1878 return 0; /* component must fit device */
1879 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1880 return 0; /* too confusing */
1881 if (rdev
->sb_start
< rdev
->data_offset
) {
1882 /* minor versions 1 and 2; superblock before data */
1883 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1884 max_sectors
-= rdev
->data_offset
;
1885 if (!num_sectors
|| num_sectors
> max_sectors
)
1886 num_sectors
= max_sectors
;
1887 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1888 /* minor version 0 with bitmap we can't move */
1891 /* minor version 0; superblock after data */
1893 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1894 sb_start
&= ~(sector_t
)(4*2 - 1);
1895 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1896 if (!num_sectors
|| num_sectors
> max_sectors
)
1897 num_sectors
= max_sectors
;
1898 rdev
->sb_start
= sb_start
;
1900 sb
= page_address(rdev
->sb_page
);
1901 sb
->data_size
= cpu_to_le64(num_sectors
);
1902 sb
->super_offset
= rdev
->sb_start
;
1903 sb
->sb_csum
= calc_sb_1_csum(sb
);
1904 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1906 md_super_wait(rdev
->mddev
);
1912 super_1_allow_new_offset(struct md_rdev
*rdev
,
1913 unsigned long long new_offset
)
1915 /* All necessary checks on new >= old have been done */
1916 struct bitmap
*bitmap
;
1917 if (new_offset
>= rdev
->data_offset
)
1920 /* with 1.0 metadata, there is no metadata to tread on
1921 * so we can always move back */
1922 if (rdev
->mddev
->minor_version
== 0)
1925 /* otherwise we must be sure not to step on
1926 * any metadata, so stay:
1927 * 36K beyond start of superblock
1928 * beyond end of badblocks
1929 * beyond write-intent bitmap
1931 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1933 bitmap
= rdev
->mddev
->bitmap
;
1934 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1935 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1936 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1938 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1944 static struct super_type super_types
[] = {
1947 .owner
= THIS_MODULE
,
1948 .load_super
= super_90_load
,
1949 .validate_super
= super_90_validate
,
1950 .sync_super
= super_90_sync
,
1951 .rdev_size_change
= super_90_rdev_size_change
,
1952 .allow_new_offset
= super_90_allow_new_offset
,
1956 .owner
= THIS_MODULE
,
1957 .load_super
= super_1_load
,
1958 .validate_super
= super_1_validate
,
1959 .sync_super
= super_1_sync
,
1960 .rdev_size_change
= super_1_rdev_size_change
,
1961 .allow_new_offset
= super_1_allow_new_offset
,
1965 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1967 if (mddev
->sync_super
) {
1968 mddev
->sync_super(mddev
, rdev
);
1972 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1974 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1977 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1979 struct md_rdev
*rdev
, *rdev2
;
1982 rdev_for_each_rcu(rdev
, mddev1
)
1983 rdev_for_each_rcu(rdev2
, mddev2
)
1984 if (rdev
->bdev
->bd_contains
==
1985 rdev2
->bdev
->bd_contains
) {
1993 static LIST_HEAD(pending_raid_disks
);
1996 * Try to register data integrity profile for an mddev
1998 * This is called when an array is started and after a disk has been kicked
1999 * from the array. It only succeeds if all working and active component devices
2000 * are integrity capable with matching profiles.
2002 int md_integrity_register(struct mddev
*mddev
)
2004 struct md_rdev
*rdev
, *reference
= NULL
;
2006 if (list_empty(&mddev
->disks
))
2007 return 0; /* nothing to do */
2008 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
2009 return 0; /* shouldn't register, or already is */
2010 rdev_for_each(rdev
, mddev
) {
2011 /* skip spares and non-functional disks */
2012 if (test_bit(Faulty
, &rdev
->flags
))
2014 if (rdev
->raid_disk
< 0)
2017 /* Use the first rdev as the reference */
2021 /* does this rdev's profile match the reference profile? */
2022 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
2023 rdev
->bdev
->bd_disk
) < 0)
2026 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2029 * All component devices are integrity capable and have matching
2030 * profiles, register the common profile for the md device.
2032 if (blk_integrity_register(mddev
->gendisk
,
2033 bdev_get_integrity(reference
->bdev
)) != 0) {
2034 printk(KERN_ERR
"md: failed to register integrity for %s\n",
2038 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2039 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2040 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2046 EXPORT_SYMBOL(md_integrity_register
);
2048 /* Disable data integrity if non-capable/non-matching disk is being added */
2049 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2051 struct blk_integrity
*bi_rdev
;
2052 struct blk_integrity
*bi_mddev
;
2054 if (!mddev
->gendisk
)
2057 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2058 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2060 if (!bi_mddev
) /* nothing to do */
2062 if (rdev
->raid_disk
< 0) /* skip spares */
2064 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2065 rdev
->bdev
->bd_disk
) >= 0)
2067 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2068 blk_integrity_unregister(mddev
->gendisk
);
2070 EXPORT_SYMBOL(md_integrity_add_rdev
);
2072 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2074 char b
[BDEVNAME_SIZE
];
2084 /* prevent duplicates */
2085 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2088 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2089 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2090 rdev
->sectors
< mddev
->dev_sectors
)) {
2092 /* Cannot change size, so fail
2093 * If mddev->level <= 0, then we don't care
2094 * about aligning sizes (e.g. linear)
2096 if (mddev
->level
> 0)
2099 mddev
->dev_sectors
= rdev
->sectors
;
2102 /* Verify rdev->desc_nr is unique.
2103 * If it is -1, assign a free number, else
2104 * check number is not in use
2106 if (rdev
->desc_nr
< 0) {
2108 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2109 while (find_rdev_nr(mddev
, choice
))
2111 rdev
->desc_nr
= choice
;
2113 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2116 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2117 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2118 mdname(mddev
), mddev
->max_disks
);
2121 bdevname(rdev
->bdev
,b
);
2122 while ( (s
=strchr(b
, '/')) != NULL
)
2125 rdev
->mddev
= mddev
;
2126 printk(KERN_INFO
"md: bind<%s>\n", b
);
2128 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2131 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2132 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2133 /* failure here is OK */;
2134 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2136 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2137 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2139 /* May as well allow recovery to be retried once */
2140 mddev
->recovery_disabled
++;
2145 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2150 static void md_delayed_delete(struct work_struct
*ws
)
2152 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2153 kobject_del(&rdev
->kobj
);
2154 kobject_put(&rdev
->kobj
);
2157 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2159 char b
[BDEVNAME_SIZE
];
2164 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2165 list_del_rcu(&rdev
->same_set
);
2166 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2168 sysfs_remove_link(&rdev
->kobj
, "block");
2169 sysfs_put(rdev
->sysfs_state
);
2170 rdev
->sysfs_state
= NULL
;
2171 rdev
->badblocks
.count
= 0;
2172 /* We need to delay this, otherwise we can deadlock when
2173 * writing to 'remove' to "dev/state". We also need
2174 * to delay it due to rcu usage.
2177 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2178 kobject_get(&rdev
->kobj
);
2179 queue_work(md_misc_wq
, &rdev
->del_work
);
2183 * prevent the device from being mounted, repartitioned or
2184 * otherwise reused by a RAID array (or any other kernel
2185 * subsystem), by bd_claiming the device.
2187 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2190 struct block_device
*bdev
;
2191 char b
[BDEVNAME_SIZE
];
2193 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2194 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2196 printk(KERN_ERR
"md: could not open %s.\n",
2197 __bdevname(dev
, b
));
2198 return PTR_ERR(bdev
);
2204 static void unlock_rdev(struct md_rdev
*rdev
)
2206 struct block_device
*bdev
= rdev
->bdev
;
2210 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2213 void md_autodetect_dev(dev_t dev
);
2215 static void export_rdev(struct md_rdev
* rdev
)
2217 char b
[BDEVNAME_SIZE
];
2218 printk(KERN_INFO
"md: export_rdev(%s)\n",
2219 bdevname(rdev
->bdev
,b
));
2222 md_rdev_clear(rdev
);
2224 if (test_bit(AutoDetected
, &rdev
->flags
))
2225 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2228 kobject_put(&rdev
->kobj
);
2231 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2233 unbind_rdev_from_array(rdev
);
2237 static void export_array(struct mddev
*mddev
)
2239 struct md_rdev
*rdev
, *tmp
;
2241 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2246 kick_rdev_from_array(rdev
);
2248 if (!list_empty(&mddev
->disks
))
2250 mddev
->raid_disks
= 0;
2251 mddev
->major_version
= 0;
2254 static void print_desc(mdp_disk_t
*desc
)
2256 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2257 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2260 static void print_sb_90(mdp_super_t
*sb
)
2265 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2266 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2267 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2269 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2270 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2271 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2272 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2273 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2274 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2275 sb
->failed_disks
, sb
->spare_disks
,
2276 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2279 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2282 desc
= sb
->disks
+ i
;
2283 if (desc
->number
|| desc
->major
|| desc
->minor
||
2284 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2285 printk(" D %2d: ", i
);
2289 printk(KERN_INFO
"md: THIS: ");
2290 print_desc(&sb
->this_disk
);
2293 static void print_sb_1(struct mdp_superblock_1
*sb
)
2297 uuid
= sb
->set_uuid
;
2299 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2300 "md: Name: \"%s\" CT:%llu\n",
2301 le32_to_cpu(sb
->major_version
),
2302 le32_to_cpu(sb
->feature_map
),
2305 (unsigned long long)le64_to_cpu(sb
->ctime
)
2306 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2308 uuid
= sb
->device_uuid
;
2310 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2312 "md: Dev:%08x UUID: %pU\n"
2313 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2314 "md: (MaxDev:%u) \n",
2315 le32_to_cpu(sb
->level
),
2316 (unsigned long long)le64_to_cpu(sb
->size
),
2317 le32_to_cpu(sb
->raid_disks
),
2318 le32_to_cpu(sb
->layout
),
2319 le32_to_cpu(sb
->chunksize
),
2320 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2321 (unsigned long long)le64_to_cpu(sb
->data_size
),
2322 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2323 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2324 le32_to_cpu(sb
->dev_number
),
2327 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2328 (unsigned long long)le64_to_cpu(sb
->events
),
2329 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2330 le32_to_cpu(sb
->sb_csum
),
2331 le32_to_cpu(sb
->max_dev
)
2335 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2337 char b
[BDEVNAME_SIZE
];
2338 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2339 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2340 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2342 if (rdev
->sb_loaded
) {
2343 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2344 switch (major_version
) {
2346 print_sb_90(page_address(rdev
->sb_page
));
2349 print_sb_1(page_address(rdev
->sb_page
));
2353 printk(KERN_INFO
"md: no rdev superblock!\n");
2356 static void md_print_devices(void)
2358 struct list_head
*tmp
;
2359 struct md_rdev
*rdev
;
2360 struct mddev
*mddev
;
2361 char b
[BDEVNAME_SIZE
];
2364 printk("md: **********************************\n");
2365 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2366 printk("md: **********************************\n");
2367 for_each_mddev(mddev
, tmp
) {
2370 bitmap_print_sb(mddev
->bitmap
);
2372 printk("%s: ", mdname(mddev
));
2373 rdev_for_each(rdev
, mddev
)
2374 printk("<%s>", bdevname(rdev
->bdev
,b
));
2377 rdev_for_each(rdev
, mddev
)
2378 print_rdev(rdev
, mddev
->major_version
);
2380 printk("md: **********************************\n");
2385 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2387 /* Update each superblock (in-memory image), but
2388 * if we are allowed to, skip spares which already
2389 * have the right event counter, or have one earlier
2390 * (which would mean they aren't being marked as dirty
2391 * with the rest of the array)
2393 struct md_rdev
*rdev
;
2394 rdev_for_each(rdev
, mddev
) {
2395 if (rdev
->sb_events
== mddev
->events
||
2397 rdev
->raid_disk
< 0 &&
2398 rdev
->sb_events
+1 == mddev
->events
)) {
2399 /* Don't update this superblock */
2400 rdev
->sb_loaded
= 2;
2402 sync_super(mddev
, rdev
);
2403 rdev
->sb_loaded
= 1;
2408 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2410 struct md_rdev
*rdev
;
2413 int any_badblocks_changed
= 0;
2417 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2421 /* First make sure individual recovery_offsets are correct */
2422 rdev_for_each(rdev
, mddev
) {
2423 if (rdev
->raid_disk
>= 0 &&
2424 mddev
->delta_disks
>= 0 &&
2425 !test_bit(In_sync
, &rdev
->flags
) &&
2426 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2427 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2430 if (!mddev
->persistent
) {
2431 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2432 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2433 if (!mddev
->external
) {
2434 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2435 rdev_for_each(rdev
, mddev
) {
2436 if (rdev
->badblocks
.changed
) {
2437 rdev
->badblocks
.changed
= 0;
2438 md_ack_all_badblocks(&rdev
->badblocks
);
2439 md_error(mddev
, rdev
);
2441 clear_bit(Blocked
, &rdev
->flags
);
2442 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2443 wake_up(&rdev
->blocked_wait
);
2446 wake_up(&mddev
->sb_wait
);
2450 spin_lock_irq(&mddev
->write_lock
);
2452 mddev
->utime
= get_seconds();
2454 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2456 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2457 /* just a clean<-> dirty transition, possibly leave spares alone,
2458 * though if events isn't the right even/odd, we will have to do
2464 if (mddev
->degraded
)
2465 /* If the array is degraded, then skipping spares is both
2466 * dangerous and fairly pointless.
2467 * Dangerous because a device that was removed from the array
2468 * might have a event_count that still looks up-to-date,
2469 * so it can be re-added without a resync.
2470 * Pointless because if there are any spares to skip,
2471 * then a recovery will happen and soon that array won't
2472 * be degraded any more and the spare can go back to sleep then.
2476 sync_req
= mddev
->in_sync
;
2478 /* If this is just a dirty<->clean transition, and the array is clean
2479 * and 'events' is odd, we can roll back to the previous clean state */
2481 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2482 && mddev
->can_decrease_events
2483 && mddev
->events
!= 1) {
2485 mddev
->can_decrease_events
= 0;
2487 /* otherwise we have to go forward and ... */
2489 mddev
->can_decrease_events
= nospares
;
2492 if (!mddev
->events
) {
2494 * oops, this 64-bit counter should never wrap.
2495 * Either we are in around ~1 trillion A.C., assuming
2496 * 1 reboot per second, or we have a bug:
2502 rdev_for_each(rdev
, mddev
) {
2503 if (rdev
->badblocks
.changed
)
2504 any_badblocks_changed
++;
2505 if (test_bit(Faulty
, &rdev
->flags
))
2506 set_bit(FaultRecorded
, &rdev
->flags
);
2509 sync_sbs(mddev
, nospares
);
2510 spin_unlock_irq(&mddev
->write_lock
);
2512 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2513 mdname(mddev
), mddev
->in_sync
);
2515 bitmap_update_sb(mddev
->bitmap
);
2516 rdev_for_each(rdev
, mddev
) {
2517 char b
[BDEVNAME_SIZE
];
2519 if (rdev
->sb_loaded
!= 1)
2520 continue; /* no noise on spare devices */
2522 if (!test_bit(Faulty
, &rdev
->flags
) &&
2523 rdev
->saved_raid_disk
== -1) {
2524 md_super_write(mddev
,rdev
,
2525 rdev
->sb_start
, rdev
->sb_size
,
2527 pr_debug("md: (write) %s's sb offset: %llu\n",
2528 bdevname(rdev
->bdev
, b
),
2529 (unsigned long long)rdev
->sb_start
);
2530 rdev
->sb_events
= mddev
->events
;
2531 if (rdev
->badblocks
.size
) {
2532 md_super_write(mddev
, rdev
,
2533 rdev
->badblocks
.sector
,
2534 rdev
->badblocks
.size
<< 9,
2536 rdev
->badblocks
.size
= 0;
2539 } else if (test_bit(Faulty
, &rdev
->flags
))
2540 pr_debug("md: %s (skipping faulty)\n",
2541 bdevname(rdev
->bdev
, b
));
2543 pr_debug("(skipping incremental s/r ");
2545 if (mddev
->level
== LEVEL_MULTIPATH
)
2546 /* only need to write one superblock... */
2549 md_super_wait(mddev
);
2550 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2552 spin_lock_irq(&mddev
->write_lock
);
2553 if (mddev
->in_sync
!= sync_req
||
2554 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2555 /* have to write it out again */
2556 spin_unlock_irq(&mddev
->write_lock
);
2559 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2560 spin_unlock_irq(&mddev
->write_lock
);
2561 wake_up(&mddev
->sb_wait
);
2562 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2563 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2565 rdev_for_each(rdev
, mddev
) {
2566 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2567 clear_bit(Blocked
, &rdev
->flags
);
2569 if (any_badblocks_changed
)
2570 md_ack_all_badblocks(&rdev
->badblocks
);
2571 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2572 wake_up(&rdev
->blocked_wait
);
2576 /* words written to sysfs files may, or may not, be \n terminated.
2577 * We want to accept with case. For this we use cmd_match.
2579 static int cmd_match(const char *cmd
, const char *str
)
2581 /* See if cmd, written into a sysfs file, matches
2582 * str. They must either be the same, or cmd can
2583 * have a trailing newline
2585 while (*cmd
&& *str
&& *cmd
== *str
) {
2596 struct rdev_sysfs_entry
{
2597 struct attribute attr
;
2598 ssize_t (*show
)(struct md_rdev
*, char *);
2599 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2603 state_show(struct md_rdev
*rdev
, char *page
)
2608 if (test_bit(Faulty
, &rdev
->flags
) ||
2609 rdev
->badblocks
.unacked_exist
) {
2610 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2613 if (test_bit(In_sync
, &rdev
->flags
)) {
2614 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2617 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2618 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2621 if (test_bit(Blocked
, &rdev
->flags
) ||
2622 (rdev
->badblocks
.unacked_exist
2623 && !test_bit(Faulty
, &rdev
->flags
))) {
2624 len
+= sprintf(page
+len
, "%sblocked", sep
);
2627 if (!test_bit(Faulty
, &rdev
->flags
) &&
2628 !test_bit(In_sync
, &rdev
->flags
)) {
2629 len
+= sprintf(page
+len
, "%sspare", sep
);
2632 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2633 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2636 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2637 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2640 if (test_bit(Replacement
, &rdev
->flags
)) {
2641 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2645 return len
+sprintf(page
+len
, "\n");
2649 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2652 * faulty - simulates an error
2653 * remove - disconnects the device
2654 * writemostly - sets write_mostly
2655 * -writemostly - clears write_mostly
2656 * blocked - sets the Blocked flags
2657 * -blocked - clears the Blocked and possibly simulates an error
2658 * insync - sets Insync providing device isn't active
2659 * write_error - sets WriteErrorSeen
2660 * -write_error - clears WriteErrorSeen
2663 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2664 md_error(rdev
->mddev
, rdev
);
2665 if (test_bit(Faulty
, &rdev
->flags
))
2669 } else if (cmd_match(buf
, "remove")) {
2670 if (rdev
->raid_disk
>= 0)
2673 struct mddev
*mddev
= rdev
->mddev
;
2674 kick_rdev_from_array(rdev
);
2676 md_update_sb(mddev
, 1);
2677 md_new_event(mddev
);
2680 } else if (cmd_match(buf
, "writemostly")) {
2681 set_bit(WriteMostly
, &rdev
->flags
);
2683 } else if (cmd_match(buf
, "-writemostly")) {
2684 clear_bit(WriteMostly
, &rdev
->flags
);
2686 } else if (cmd_match(buf
, "blocked")) {
2687 set_bit(Blocked
, &rdev
->flags
);
2689 } else if (cmd_match(buf
, "-blocked")) {
2690 if (!test_bit(Faulty
, &rdev
->flags
) &&
2691 rdev
->badblocks
.unacked_exist
) {
2692 /* metadata handler doesn't understand badblocks,
2693 * so we need to fail the device
2695 md_error(rdev
->mddev
, rdev
);
2697 clear_bit(Blocked
, &rdev
->flags
);
2698 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2699 wake_up(&rdev
->blocked_wait
);
2700 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2701 md_wakeup_thread(rdev
->mddev
->thread
);
2704 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2705 set_bit(In_sync
, &rdev
->flags
);
2707 } else if (cmd_match(buf
, "write_error")) {
2708 set_bit(WriteErrorSeen
, &rdev
->flags
);
2710 } else if (cmd_match(buf
, "-write_error")) {
2711 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2713 } else if (cmd_match(buf
, "want_replacement")) {
2714 /* Any non-spare device that is not a replacement can
2715 * become want_replacement at any time, but we then need to
2716 * check if recovery is needed.
2718 if (rdev
->raid_disk
>= 0 &&
2719 !test_bit(Replacement
, &rdev
->flags
))
2720 set_bit(WantReplacement
, &rdev
->flags
);
2721 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2722 md_wakeup_thread(rdev
->mddev
->thread
);
2724 } else if (cmd_match(buf
, "-want_replacement")) {
2725 /* Clearing 'want_replacement' is always allowed.
2726 * Once replacements starts it is too late though.
2729 clear_bit(WantReplacement
, &rdev
->flags
);
2730 } else if (cmd_match(buf
, "replacement")) {
2731 /* Can only set a device as a replacement when array has not
2732 * yet been started. Once running, replacement is automatic
2733 * from spares, or by assigning 'slot'.
2735 if (rdev
->mddev
->pers
)
2738 set_bit(Replacement
, &rdev
->flags
);
2741 } else if (cmd_match(buf
, "-replacement")) {
2742 /* Similarly, can only clear Replacement before start */
2743 if (rdev
->mddev
->pers
)
2746 clear_bit(Replacement
, &rdev
->flags
);
2751 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2752 return err
? err
: len
;
2754 static struct rdev_sysfs_entry rdev_state
=
2755 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2758 errors_show(struct md_rdev
*rdev
, char *page
)
2760 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2764 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2767 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2768 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2769 atomic_set(&rdev
->corrected_errors
, n
);
2774 static struct rdev_sysfs_entry rdev_errors
=
2775 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2778 slot_show(struct md_rdev
*rdev
, char *page
)
2780 if (rdev
->raid_disk
< 0)
2781 return sprintf(page
, "none\n");
2783 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2787 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2791 int slot
= simple_strtoul(buf
, &e
, 10);
2792 if (strncmp(buf
, "none", 4)==0)
2794 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2796 if (rdev
->mddev
->pers
&& slot
== -1) {
2797 /* Setting 'slot' on an active array requires also
2798 * updating the 'rd%d' link, and communicating
2799 * with the personality with ->hot_*_disk.
2800 * For now we only support removing
2801 * failed/spare devices. This normally happens automatically,
2802 * but not when the metadata is externally managed.
2804 if (rdev
->raid_disk
== -1)
2806 /* personality does all needed checks */
2807 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2809 clear_bit(Blocked
, &rdev
->flags
);
2810 remove_and_add_spares(rdev
->mddev
, rdev
);
2811 if (rdev
->raid_disk
>= 0)
2813 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2814 md_wakeup_thread(rdev
->mddev
->thread
);
2815 } else if (rdev
->mddev
->pers
) {
2816 /* Activating a spare .. or possibly reactivating
2817 * if we ever get bitmaps working here.
2820 if (rdev
->raid_disk
!= -1)
2823 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2826 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2829 if (slot
>= rdev
->mddev
->raid_disks
&&
2830 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2833 rdev
->raid_disk
= slot
;
2834 if (test_bit(In_sync
, &rdev
->flags
))
2835 rdev
->saved_raid_disk
= slot
;
2837 rdev
->saved_raid_disk
= -1;
2838 clear_bit(In_sync
, &rdev
->flags
);
2839 clear_bit(Bitmap_sync
, &rdev
->flags
);
2840 err
= rdev
->mddev
->pers
->
2841 hot_add_disk(rdev
->mddev
, rdev
);
2843 rdev
->raid_disk
= -1;
2846 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2847 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2848 /* failure here is OK */;
2849 /* don't wakeup anyone, leave that to userspace. */
2851 if (slot
>= rdev
->mddev
->raid_disks
&&
2852 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2854 rdev
->raid_disk
= slot
;
2855 /* assume it is working */
2856 clear_bit(Faulty
, &rdev
->flags
);
2857 clear_bit(WriteMostly
, &rdev
->flags
);
2858 set_bit(In_sync
, &rdev
->flags
);
2859 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2865 static struct rdev_sysfs_entry rdev_slot
=
2866 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2869 offset_show(struct md_rdev
*rdev
, char *page
)
2871 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2875 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2877 unsigned long long offset
;
2878 if (strict_strtoull(buf
, 10, &offset
) < 0)
2880 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2882 if (rdev
->sectors
&& rdev
->mddev
->external
)
2883 /* Must set offset before size, so overlap checks
2886 rdev
->data_offset
= offset
;
2887 rdev
->new_data_offset
= offset
;
2891 static struct rdev_sysfs_entry rdev_offset
=
2892 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2894 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2896 return sprintf(page
, "%llu\n",
2897 (unsigned long long)rdev
->new_data_offset
);
2900 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2901 const char *buf
, size_t len
)
2903 unsigned long long new_offset
;
2904 struct mddev
*mddev
= rdev
->mddev
;
2906 if (strict_strtoull(buf
, 10, &new_offset
) < 0)
2909 if (mddev
->sync_thread
)
2911 if (new_offset
== rdev
->data_offset
)
2912 /* reset is always permitted */
2914 else if (new_offset
> rdev
->data_offset
) {
2915 /* must not push array size beyond rdev_sectors */
2916 if (new_offset
- rdev
->data_offset
2917 + mddev
->dev_sectors
> rdev
->sectors
)
2920 /* Metadata worries about other space details. */
2922 /* decreasing the offset is inconsistent with a backwards
2925 if (new_offset
< rdev
->data_offset
&&
2926 mddev
->reshape_backwards
)
2928 /* Increasing offset is inconsistent with forwards
2929 * reshape. reshape_direction should be set to
2930 * 'backwards' first.
2932 if (new_offset
> rdev
->data_offset
&&
2933 !mddev
->reshape_backwards
)
2936 if (mddev
->pers
&& mddev
->persistent
&&
2937 !super_types
[mddev
->major_version
]
2938 .allow_new_offset(rdev
, new_offset
))
2940 rdev
->new_data_offset
= new_offset
;
2941 if (new_offset
> rdev
->data_offset
)
2942 mddev
->reshape_backwards
= 1;
2943 else if (new_offset
< rdev
->data_offset
)
2944 mddev
->reshape_backwards
= 0;
2948 static struct rdev_sysfs_entry rdev_new_offset
=
2949 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2952 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2954 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2957 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2959 /* check if two start/length pairs overlap */
2967 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2969 unsigned long long blocks
;
2972 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2975 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2976 return -EINVAL
; /* sector conversion overflow */
2979 if (new != blocks
* 2)
2980 return -EINVAL
; /* unsigned long long to sector_t overflow */
2987 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2989 struct mddev
*my_mddev
= rdev
->mddev
;
2990 sector_t oldsectors
= rdev
->sectors
;
2993 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2995 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2996 return -EINVAL
; /* too confusing */
2997 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2998 if (my_mddev
->persistent
) {
2999 sectors
= super_types
[my_mddev
->major_version
].
3000 rdev_size_change(rdev
, sectors
);
3003 } else if (!sectors
)
3004 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
3006 if (!my_mddev
->pers
->resize
)
3007 /* Cannot change size for RAID0 or Linear etc */
3010 if (sectors
< my_mddev
->dev_sectors
)
3011 return -EINVAL
; /* component must fit device */
3013 rdev
->sectors
= sectors
;
3014 if (sectors
> oldsectors
&& my_mddev
->external
) {
3015 /* need to check that all other rdevs with the same ->bdev
3016 * do not overlap. We need to unlock the mddev to avoid
3017 * a deadlock. We have already changed rdev->sectors, and if
3018 * we have to change it back, we will have the lock again.
3020 struct mddev
*mddev
;
3022 struct list_head
*tmp
;
3024 mddev_unlock(my_mddev
);
3025 for_each_mddev(mddev
, tmp
) {
3026 struct md_rdev
*rdev2
;
3029 rdev_for_each(rdev2
, mddev
)
3030 if (rdev
->bdev
== rdev2
->bdev
&&
3032 overlaps(rdev
->data_offset
, rdev
->sectors
,
3038 mddev_unlock(mddev
);
3044 mddev_lock(my_mddev
);
3046 /* Someone else could have slipped in a size
3047 * change here, but doing so is just silly.
3048 * We put oldsectors back because we *know* it is
3049 * safe, and trust userspace not to race with
3052 rdev
->sectors
= oldsectors
;
3059 static struct rdev_sysfs_entry rdev_size
=
3060 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3063 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3065 unsigned long long recovery_start
= rdev
->recovery_offset
;
3067 if (test_bit(In_sync
, &rdev
->flags
) ||
3068 recovery_start
== MaxSector
)
3069 return sprintf(page
, "none\n");
3071 return sprintf(page
, "%llu\n", recovery_start
);
3074 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3076 unsigned long long recovery_start
;
3078 if (cmd_match(buf
, "none"))
3079 recovery_start
= MaxSector
;
3080 else if (strict_strtoull(buf
, 10, &recovery_start
))
3083 if (rdev
->mddev
->pers
&&
3084 rdev
->raid_disk
>= 0)
3087 rdev
->recovery_offset
= recovery_start
;
3088 if (recovery_start
== MaxSector
)
3089 set_bit(In_sync
, &rdev
->flags
);
3091 clear_bit(In_sync
, &rdev
->flags
);
3095 static struct rdev_sysfs_entry rdev_recovery_start
=
3096 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3100 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
3102 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
3104 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3106 return badblocks_show(&rdev
->badblocks
, page
, 0);
3108 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3110 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3111 /* Maybe that ack was all we needed */
3112 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3113 wake_up(&rdev
->blocked_wait
);
3116 static struct rdev_sysfs_entry rdev_bad_blocks
=
3117 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3120 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3122 return badblocks_show(&rdev
->badblocks
, page
, 1);
3124 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3126 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3128 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3129 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3131 static struct attribute
*rdev_default_attrs
[] = {
3136 &rdev_new_offset
.attr
,
3138 &rdev_recovery_start
.attr
,
3139 &rdev_bad_blocks
.attr
,
3140 &rdev_unack_bad_blocks
.attr
,
3144 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3146 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3147 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3148 struct mddev
*mddev
= rdev
->mddev
;
3154 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
3156 if (rdev
->mddev
== NULL
)
3159 rv
= entry
->show(rdev
, page
);
3160 mddev_unlock(mddev
);
3166 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3167 const char *page
, size_t length
)
3169 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3170 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3172 struct mddev
*mddev
= rdev
->mddev
;
3176 if (!capable(CAP_SYS_ADMIN
))
3178 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3180 if (rdev
->mddev
== NULL
)
3183 rv
= entry
->store(rdev
, page
, length
);
3184 mddev_unlock(mddev
);
3189 static void rdev_free(struct kobject
*ko
)
3191 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3194 static const struct sysfs_ops rdev_sysfs_ops
= {
3195 .show
= rdev_attr_show
,
3196 .store
= rdev_attr_store
,
3198 static struct kobj_type rdev_ktype
= {
3199 .release
= rdev_free
,
3200 .sysfs_ops
= &rdev_sysfs_ops
,
3201 .default_attrs
= rdev_default_attrs
,
3204 int md_rdev_init(struct md_rdev
*rdev
)
3207 rdev
->saved_raid_disk
= -1;
3208 rdev
->raid_disk
= -1;
3210 rdev
->data_offset
= 0;
3211 rdev
->new_data_offset
= 0;
3212 rdev
->sb_events
= 0;
3213 rdev
->last_read_error
.tv_sec
= 0;
3214 rdev
->last_read_error
.tv_nsec
= 0;
3215 rdev
->sb_loaded
= 0;
3216 rdev
->bb_page
= NULL
;
3217 atomic_set(&rdev
->nr_pending
, 0);
3218 atomic_set(&rdev
->read_errors
, 0);
3219 atomic_set(&rdev
->corrected_errors
, 0);
3221 INIT_LIST_HEAD(&rdev
->same_set
);
3222 init_waitqueue_head(&rdev
->blocked_wait
);
3224 /* Add space to store bad block list.
3225 * This reserves the space even on arrays where it cannot
3226 * be used - I wonder if that matters
3228 rdev
->badblocks
.count
= 0;
3229 rdev
->badblocks
.shift
= -1; /* disabled until explicitly enabled */
3230 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3231 seqlock_init(&rdev
->badblocks
.lock
);
3232 if (rdev
->badblocks
.page
== NULL
)
3237 EXPORT_SYMBOL_GPL(md_rdev_init
);
3239 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3241 * mark the device faulty if:
3243 * - the device is nonexistent (zero size)
3244 * - the device has no valid superblock
3246 * a faulty rdev _never_ has rdev->sb set.
3248 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3250 char b
[BDEVNAME_SIZE
];
3252 struct md_rdev
*rdev
;
3255 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3257 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3258 return ERR_PTR(-ENOMEM
);
3261 err
= md_rdev_init(rdev
);
3264 err
= alloc_disk_sb(rdev
);
3268 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3272 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3274 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3277 "md: %s has zero or unknown size, marking faulty!\n",
3278 bdevname(rdev
->bdev
,b
));
3283 if (super_format
>= 0) {
3284 err
= super_types
[super_format
].
3285 load_super(rdev
, NULL
, super_minor
);
3286 if (err
== -EINVAL
) {
3288 "md: %s does not have a valid v%d.%d "
3289 "superblock, not importing!\n",
3290 bdevname(rdev
->bdev
,b
),
3291 super_format
, super_minor
);
3296 "md: could not read %s's sb, not importing!\n",
3297 bdevname(rdev
->bdev
,b
));
3307 md_rdev_clear(rdev
);
3309 return ERR_PTR(err
);
3313 * Check a full RAID array for plausibility
3317 static void analyze_sbs(struct mddev
* mddev
)
3320 struct md_rdev
*rdev
, *freshest
, *tmp
;
3321 char b
[BDEVNAME_SIZE
];
3324 rdev_for_each_safe(rdev
, tmp
, mddev
)
3325 switch (super_types
[mddev
->major_version
].
3326 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3334 "md: fatal superblock inconsistency in %s"
3335 " -- removing from array\n",
3336 bdevname(rdev
->bdev
,b
));
3337 kick_rdev_from_array(rdev
);
3341 super_types
[mddev
->major_version
].
3342 validate_super(mddev
, freshest
);
3345 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3346 if (mddev
->max_disks
&&
3347 (rdev
->desc_nr
>= mddev
->max_disks
||
3348 i
> mddev
->max_disks
)) {
3350 "md: %s: %s: only %d devices permitted\n",
3351 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3353 kick_rdev_from_array(rdev
);
3356 if (rdev
!= freshest
)
3357 if (super_types
[mddev
->major_version
].
3358 validate_super(mddev
, rdev
)) {
3359 printk(KERN_WARNING
"md: kicking non-fresh %s"
3361 bdevname(rdev
->bdev
,b
));
3362 kick_rdev_from_array(rdev
);
3365 if (mddev
->level
== LEVEL_MULTIPATH
) {
3366 rdev
->desc_nr
= i
++;
3367 rdev
->raid_disk
= rdev
->desc_nr
;
3368 set_bit(In_sync
, &rdev
->flags
);
3369 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3370 rdev
->raid_disk
= -1;
3371 clear_bit(In_sync
, &rdev
->flags
);
3376 /* Read a fixed-point number.
3377 * Numbers in sysfs attributes should be in "standard" units where
3378 * possible, so time should be in seconds.
3379 * However we internally use a a much smaller unit such as
3380 * milliseconds or jiffies.
3381 * This function takes a decimal number with a possible fractional
3382 * component, and produces an integer which is the result of
3383 * multiplying that number by 10^'scale'.
3384 * all without any floating-point arithmetic.
3386 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3388 unsigned long result
= 0;
3390 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3393 else if (decimals
< scale
) {
3396 result
= result
* 10 + value
;
3408 while (decimals
< scale
) {
3417 static void md_safemode_timeout(unsigned long data
);
3420 safe_delay_show(struct mddev
*mddev
, char *page
)
3422 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3423 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3426 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3430 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3433 mddev
->safemode_delay
= 0;
3435 unsigned long old_delay
= mddev
->safemode_delay
;
3436 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3437 if (mddev
->safemode_delay
== 0)
3438 mddev
->safemode_delay
= 1;
3439 if (mddev
->safemode_delay
< old_delay
)
3440 md_safemode_timeout((unsigned long)mddev
);
3444 static struct md_sysfs_entry md_safe_delay
=
3445 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3448 level_show(struct mddev
*mddev
, char *page
)
3450 struct md_personality
*p
= mddev
->pers
;
3452 return sprintf(page
, "%s\n", p
->name
);
3453 else if (mddev
->clevel
[0])
3454 return sprintf(page
, "%s\n", mddev
->clevel
);
3455 else if (mddev
->level
!= LEVEL_NONE
)
3456 return sprintf(page
, "%d\n", mddev
->level
);
3462 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3466 struct md_personality
*pers
;
3469 struct md_rdev
*rdev
;
3471 if (mddev
->pers
== NULL
) {
3474 if (len
>= sizeof(mddev
->clevel
))
3476 strncpy(mddev
->clevel
, buf
, len
);
3477 if (mddev
->clevel
[len
-1] == '\n')
3479 mddev
->clevel
[len
] = 0;
3480 mddev
->level
= LEVEL_NONE
;
3484 /* request to change the personality. Need to ensure:
3485 * - array is not engaged in resync/recovery/reshape
3486 * - old personality can be suspended
3487 * - new personality will access other array.
3490 if (mddev
->sync_thread
||
3491 mddev
->reshape_position
!= MaxSector
||
3492 mddev
->sysfs_active
)
3495 if (!mddev
->pers
->quiesce
) {
3496 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3497 mdname(mddev
), mddev
->pers
->name
);
3501 /* Now find the new personality */
3502 if (len
== 0 || len
>= sizeof(clevel
))
3504 strncpy(clevel
, buf
, len
);
3505 if (clevel
[len
-1] == '\n')
3508 if (strict_strtol(clevel
, 10, &level
))
3511 if (request_module("md-%s", clevel
) != 0)
3512 request_module("md-level-%s", clevel
);
3513 spin_lock(&pers_lock
);
3514 pers
= find_pers(level
, clevel
);
3515 if (!pers
|| !try_module_get(pers
->owner
)) {
3516 spin_unlock(&pers_lock
);
3517 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3520 spin_unlock(&pers_lock
);
3522 if (pers
== mddev
->pers
) {
3523 /* Nothing to do! */
3524 module_put(pers
->owner
);
3527 if (!pers
->takeover
) {
3528 module_put(pers
->owner
);
3529 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3530 mdname(mddev
), clevel
);
3534 rdev_for_each(rdev
, mddev
)
3535 rdev
->new_raid_disk
= rdev
->raid_disk
;
3537 /* ->takeover must set new_* and/or delta_disks
3538 * if it succeeds, and may set them when it fails.
3540 priv
= pers
->takeover(mddev
);
3542 mddev
->new_level
= mddev
->level
;
3543 mddev
->new_layout
= mddev
->layout
;
3544 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3545 mddev
->raid_disks
-= mddev
->delta_disks
;
3546 mddev
->delta_disks
= 0;
3547 mddev
->reshape_backwards
= 0;
3548 module_put(pers
->owner
);
3549 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3550 mdname(mddev
), clevel
);
3551 return PTR_ERR(priv
);
3554 /* Looks like we have a winner */
3555 mddev_suspend(mddev
);
3556 mddev
->pers
->stop(mddev
);
3558 if (mddev
->pers
->sync_request
== NULL
&&
3559 pers
->sync_request
!= NULL
) {
3560 /* need to add the md_redundancy_group */
3561 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3563 "md: cannot register extra attributes for %s\n",
3565 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3567 if (mddev
->pers
->sync_request
!= NULL
&&
3568 pers
->sync_request
== NULL
) {
3569 /* need to remove the md_redundancy_group */
3570 if (mddev
->to_remove
== NULL
)
3571 mddev
->to_remove
= &md_redundancy_group
;
3574 if (mddev
->pers
->sync_request
== NULL
&&
3576 /* We are converting from a no-redundancy array
3577 * to a redundancy array and metadata is managed
3578 * externally so we need to be sure that writes
3579 * won't block due to a need to transition
3581 * until external management is started.
3584 mddev
->safemode_delay
= 0;
3585 mddev
->safemode
= 0;
3588 rdev_for_each(rdev
, mddev
) {
3589 if (rdev
->raid_disk
< 0)
3591 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3592 rdev
->new_raid_disk
= -1;
3593 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3595 sysfs_unlink_rdev(mddev
, rdev
);
3597 rdev_for_each(rdev
, mddev
) {
3598 if (rdev
->raid_disk
< 0)
3600 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3602 rdev
->raid_disk
= rdev
->new_raid_disk
;
3603 if (rdev
->raid_disk
< 0)
3604 clear_bit(In_sync
, &rdev
->flags
);
3606 if (sysfs_link_rdev(mddev
, rdev
))
3607 printk(KERN_WARNING
"md: cannot register rd%d"
3608 " for %s after level change\n",
3609 rdev
->raid_disk
, mdname(mddev
));
3613 module_put(mddev
->pers
->owner
);
3615 mddev
->private = priv
;
3616 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3617 mddev
->level
= mddev
->new_level
;
3618 mddev
->layout
= mddev
->new_layout
;
3619 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3620 mddev
->delta_disks
= 0;
3621 mddev
->reshape_backwards
= 0;
3622 mddev
->degraded
= 0;
3623 if (mddev
->pers
->sync_request
== NULL
) {
3624 /* this is now an array without redundancy, so
3625 * it must always be in_sync
3628 del_timer_sync(&mddev
->safemode_timer
);
3630 blk_set_stacking_limits(&mddev
->queue
->limits
);
3632 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3633 mddev_resume(mddev
);
3634 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3635 md_new_event(mddev
);
3639 static struct md_sysfs_entry md_level
=
3640 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3644 layout_show(struct mddev
*mddev
, char *page
)
3646 /* just a number, not meaningful for all levels */
3647 if (mddev
->reshape_position
!= MaxSector
&&
3648 mddev
->layout
!= mddev
->new_layout
)
3649 return sprintf(page
, "%d (%d)\n",
3650 mddev
->new_layout
, mddev
->layout
);
3651 return sprintf(page
, "%d\n", mddev
->layout
);
3655 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3658 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3660 if (!*buf
|| (*e
&& *e
!= '\n'))
3665 if (mddev
->pers
->check_reshape
== NULL
)
3667 mddev
->new_layout
= n
;
3668 err
= mddev
->pers
->check_reshape(mddev
);
3670 mddev
->new_layout
= mddev
->layout
;
3674 mddev
->new_layout
= n
;
3675 if (mddev
->reshape_position
== MaxSector
)
3680 static struct md_sysfs_entry md_layout
=
3681 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3685 raid_disks_show(struct mddev
*mddev
, char *page
)
3687 if (mddev
->raid_disks
== 0)
3689 if (mddev
->reshape_position
!= MaxSector
&&
3690 mddev
->delta_disks
!= 0)
3691 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3692 mddev
->raid_disks
- mddev
->delta_disks
);
3693 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3696 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3699 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3703 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3705 if (!*buf
|| (*e
&& *e
!= '\n'))
3709 rv
= update_raid_disks(mddev
, n
);
3710 else if (mddev
->reshape_position
!= MaxSector
) {
3711 struct md_rdev
*rdev
;
3712 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3714 rdev_for_each(rdev
, mddev
) {
3716 rdev
->data_offset
< rdev
->new_data_offset
)
3719 rdev
->data_offset
> rdev
->new_data_offset
)
3722 mddev
->delta_disks
= n
- olddisks
;
3723 mddev
->raid_disks
= n
;
3724 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3726 mddev
->raid_disks
= n
;
3727 return rv
? rv
: len
;
3729 static struct md_sysfs_entry md_raid_disks
=
3730 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3733 chunk_size_show(struct mddev
*mddev
, char *page
)
3735 if (mddev
->reshape_position
!= MaxSector
&&
3736 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3737 return sprintf(page
, "%d (%d)\n",
3738 mddev
->new_chunk_sectors
<< 9,
3739 mddev
->chunk_sectors
<< 9);
3740 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3744 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3747 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3749 if (!*buf
|| (*e
&& *e
!= '\n'))
3754 if (mddev
->pers
->check_reshape
== NULL
)
3756 mddev
->new_chunk_sectors
= n
>> 9;
3757 err
= mddev
->pers
->check_reshape(mddev
);
3759 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3763 mddev
->new_chunk_sectors
= n
>> 9;
3764 if (mddev
->reshape_position
== MaxSector
)
3765 mddev
->chunk_sectors
= n
>> 9;
3769 static struct md_sysfs_entry md_chunk_size
=
3770 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3773 resync_start_show(struct mddev
*mddev
, char *page
)
3775 if (mddev
->recovery_cp
== MaxSector
)
3776 return sprintf(page
, "none\n");
3777 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3781 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3784 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3786 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3788 if (cmd_match(buf
, "none"))
3790 else if (!*buf
|| (*e
&& *e
!= '\n'))
3793 mddev
->recovery_cp
= n
;
3795 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3798 static struct md_sysfs_entry md_resync_start
=
3799 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3802 * The array state can be:
3805 * No devices, no size, no level
3806 * Equivalent to STOP_ARRAY ioctl
3808 * May have some settings, but array is not active
3809 * all IO results in error
3810 * When written, doesn't tear down array, but just stops it
3811 * suspended (not supported yet)
3812 * All IO requests will block. The array can be reconfigured.
3813 * Writing this, if accepted, will block until array is quiescent
3815 * no resync can happen. no superblocks get written.
3816 * write requests fail
3818 * like readonly, but behaves like 'clean' on a write request.
3820 * clean - no pending writes, but otherwise active.
3821 * When written to inactive array, starts without resync
3822 * If a write request arrives then
3823 * if metadata is known, mark 'dirty' and switch to 'active'.
3824 * if not known, block and switch to write-pending
3825 * If written to an active array that has pending writes, then fails.
3827 * fully active: IO and resync can be happening.
3828 * When written to inactive array, starts with resync
3831 * clean, but writes are blocked waiting for 'active' to be written.
3834 * like active, but no writes have been seen for a while (100msec).
3837 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3838 write_pending
, active_idle
, bad_word
};
3839 static char *array_states
[] = {
3840 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3841 "write-pending", "active-idle", NULL
};
3843 static int match_word(const char *word
, char **list
)
3846 for (n
=0; list
[n
]; n
++)
3847 if (cmd_match(word
, list
[n
]))
3853 array_state_show(struct mddev
*mddev
, char *page
)
3855 enum array_state st
= inactive
;
3868 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3870 else if (mddev
->safemode
)
3876 if (list_empty(&mddev
->disks
) &&
3877 mddev
->raid_disks
== 0 &&
3878 mddev
->dev_sectors
== 0)
3883 return sprintf(page
, "%s\n", array_states
[st
]);
3886 static int do_md_stop(struct mddev
* mddev
, int ro
, struct block_device
*bdev
);
3887 static int md_set_readonly(struct mddev
* mddev
, struct block_device
*bdev
);
3888 static int do_md_run(struct mddev
* mddev
);
3889 static int restart_array(struct mddev
*mddev
);
3892 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3895 enum array_state st
= match_word(buf
, array_states
);
3900 /* stopping an active array */
3901 err
= do_md_stop(mddev
, 0, NULL
);
3904 /* stopping an active array */
3906 err
= do_md_stop(mddev
, 2, NULL
);
3908 err
= 0; /* already inactive */
3911 break; /* not supported yet */
3914 err
= md_set_readonly(mddev
, NULL
);
3917 set_disk_ro(mddev
->gendisk
, 1);
3918 err
= do_md_run(mddev
);
3924 err
= md_set_readonly(mddev
, NULL
);
3925 else if (mddev
->ro
== 1)
3926 err
= restart_array(mddev
);
3929 set_disk_ro(mddev
->gendisk
, 0);
3933 err
= do_md_run(mddev
);
3938 restart_array(mddev
);
3939 spin_lock_irq(&mddev
->write_lock
);
3940 if (atomic_read(&mddev
->writes_pending
) == 0) {
3941 if (mddev
->in_sync
== 0) {
3943 if (mddev
->safemode
== 1)
3944 mddev
->safemode
= 0;
3945 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3950 spin_unlock_irq(&mddev
->write_lock
);
3956 restart_array(mddev
);
3957 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3958 wake_up(&mddev
->sb_wait
);
3962 set_disk_ro(mddev
->gendisk
, 0);
3963 err
= do_md_run(mddev
);
3968 /* these cannot be set */
3974 if (mddev
->hold_active
== UNTIL_IOCTL
)
3975 mddev
->hold_active
= 0;
3976 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3980 static struct md_sysfs_entry md_array_state
=
3981 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3984 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3985 return sprintf(page
, "%d\n",
3986 atomic_read(&mddev
->max_corr_read_errors
));
3990 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3993 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3995 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3996 atomic_set(&mddev
->max_corr_read_errors
, n
);
4002 static struct md_sysfs_entry max_corr_read_errors
=
4003 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4004 max_corrected_read_errors_store
);
4007 null_show(struct mddev
*mddev
, char *page
)
4013 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4015 /* buf must be %d:%d\n? giving major and minor numbers */
4016 /* The new device is added to the array.
4017 * If the array has a persistent superblock, we read the
4018 * superblock to initialise info and check validity.
4019 * Otherwise, only checking done is that in bind_rdev_to_array,
4020 * which mainly checks size.
4023 int major
= simple_strtoul(buf
, &e
, 10);
4026 struct md_rdev
*rdev
;
4029 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4031 minor
= simple_strtoul(e
+1, &e
, 10);
4032 if (*e
&& *e
!= '\n')
4034 dev
= MKDEV(major
, minor
);
4035 if (major
!= MAJOR(dev
) ||
4036 minor
!= MINOR(dev
))
4040 if (mddev
->persistent
) {
4041 rdev
= md_import_device(dev
, mddev
->major_version
,
4042 mddev
->minor_version
);
4043 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4044 struct md_rdev
*rdev0
4045 = list_entry(mddev
->disks
.next
,
4046 struct md_rdev
, same_set
);
4047 err
= super_types
[mddev
->major_version
]
4048 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4052 } else if (mddev
->external
)
4053 rdev
= md_import_device(dev
, -2, -1);
4055 rdev
= md_import_device(dev
, -1, -1);
4058 return PTR_ERR(rdev
);
4059 err
= bind_rdev_to_array(rdev
, mddev
);
4063 return err
? err
: len
;
4066 static struct md_sysfs_entry md_new_device
=
4067 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4070 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4073 unsigned long chunk
, end_chunk
;
4077 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4079 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4080 if (buf
== end
) break;
4081 if (*end
== '-') { /* range */
4083 end_chunk
= simple_strtoul(buf
, &end
, 0);
4084 if (buf
== end
) break;
4086 if (*end
&& !isspace(*end
)) break;
4087 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4088 buf
= skip_spaces(end
);
4090 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4095 static struct md_sysfs_entry md_bitmap
=
4096 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4099 size_show(struct mddev
*mddev
, char *page
)
4101 return sprintf(page
, "%llu\n",
4102 (unsigned long long)mddev
->dev_sectors
/ 2);
4105 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4108 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4110 /* If array is inactive, we can reduce the component size, but
4111 * not increase it (except from 0).
4112 * If array is active, we can try an on-line resize
4115 int err
= strict_blocks_to_sectors(buf
, §ors
);
4120 err
= update_size(mddev
, sectors
);
4121 md_update_sb(mddev
, 1);
4123 if (mddev
->dev_sectors
== 0 ||
4124 mddev
->dev_sectors
> sectors
)
4125 mddev
->dev_sectors
= sectors
;
4129 return err
? err
: len
;
4132 static struct md_sysfs_entry md_size
=
4133 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4136 /* Metadata version.
4138 * 'none' for arrays with no metadata (good luck...)
4139 * 'external' for arrays with externally managed metadata,
4140 * or N.M for internally known formats
4143 metadata_show(struct mddev
*mddev
, char *page
)
4145 if (mddev
->persistent
)
4146 return sprintf(page
, "%d.%d\n",
4147 mddev
->major_version
, mddev
->minor_version
);
4148 else if (mddev
->external
)
4149 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4151 return sprintf(page
, "none\n");
4155 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4159 /* Changing the details of 'external' metadata is
4160 * always permitted. Otherwise there must be
4161 * no devices attached to the array.
4163 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4165 else if (!list_empty(&mddev
->disks
))
4168 if (cmd_match(buf
, "none")) {
4169 mddev
->persistent
= 0;
4170 mddev
->external
= 0;
4171 mddev
->major_version
= 0;
4172 mddev
->minor_version
= 90;
4175 if (strncmp(buf
, "external:", 9) == 0) {
4176 size_t namelen
= len
-9;
4177 if (namelen
>= sizeof(mddev
->metadata_type
))
4178 namelen
= sizeof(mddev
->metadata_type
)-1;
4179 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4180 mddev
->metadata_type
[namelen
] = 0;
4181 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4182 mddev
->metadata_type
[--namelen
] = 0;
4183 mddev
->persistent
= 0;
4184 mddev
->external
= 1;
4185 mddev
->major_version
= 0;
4186 mddev
->minor_version
= 90;
4189 major
= simple_strtoul(buf
, &e
, 10);
4190 if (e
==buf
|| *e
!= '.')
4193 minor
= simple_strtoul(buf
, &e
, 10);
4194 if (e
==buf
|| (*e
&& *e
!= '\n') )
4196 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4198 mddev
->major_version
= major
;
4199 mddev
->minor_version
= minor
;
4200 mddev
->persistent
= 1;
4201 mddev
->external
= 0;
4205 static struct md_sysfs_entry md_metadata
=
4206 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4209 action_show(struct mddev
*mddev
, char *page
)
4211 char *type
= "idle";
4212 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4214 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4215 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4216 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4218 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4219 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4221 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4225 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4228 return sprintf(page
, "%s\n", type
);
4232 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4234 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4237 if (cmd_match(page
, "frozen"))
4238 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4240 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4242 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4243 if (mddev
->sync_thread
) {
4244 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4245 md_reap_sync_thread(mddev
);
4247 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4248 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4250 else if (cmd_match(page
, "resync"))
4251 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4252 else if (cmd_match(page
, "recover")) {
4253 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4254 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4255 } else if (cmd_match(page
, "reshape")) {
4257 if (mddev
->pers
->start_reshape
== NULL
)
4259 err
= mddev
->pers
->start_reshape(mddev
);
4262 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4264 if (cmd_match(page
, "check"))
4265 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4266 else if (!cmd_match(page
, "repair"))
4268 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4269 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4271 if (mddev
->ro
== 2) {
4272 /* A write to sync_action is enough to justify
4273 * canceling read-auto mode
4276 md_wakeup_thread(mddev
->sync_thread
);
4278 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4279 md_wakeup_thread(mddev
->thread
);
4280 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4285 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4287 return sprintf(page
, "%llu\n",
4288 (unsigned long long)
4289 atomic64_read(&mddev
->resync_mismatches
));
4292 static struct md_sysfs_entry md_scan_mode
=
4293 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4296 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4299 sync_min_show(struct mddev
*mddev
, char *page
)
4301 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4302 mddev
->sync_speed_min
? "local": "system");
4306 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4310 if (strncmp(buf
, "system", 6)==0) {
4311 mddev
->sync_speed_min
= 0;
4314 min
= simple_strtoul(buf
, &e
, 10);
4315 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4317 mddev
->sync_speed_min
= min
;
4321 static struct md_sysfs_entry md_sync_min
=
4322 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4325 sync_max_show(struct mddev
*mddev
, char *page
)
4327 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4328 mddev
->sync_speed_max
? "local": "system");
4332 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4336 if (strncmp(buf
, "system", 6)==0) {
4337 mddev
->sync_speed_max
= 0;
4340 max
= simple_strtoul(buf
, &e
, 10);
4341 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4343 mddev
->sync_speed_max
= max
;
4347 static struct md_sysfs_entry md_sync_max
=
4348 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4351 degraded_show(struct mddev
*mddev
, char *page
)
4353 return sprintf(page
, "%d\n", mddev
->degraded
);
4355 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4358 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4360 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4364 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4368 if (strict_strtol(buf
, 10, &n
))
4371 if (n
!= 0 && n
!= 1)
4374 mddev
->parallel_resync
= n
;
4376 if (mddev
->sync_thread
)
4377 wake_up(&resync_wait
);
4382 /* force parallel resync, even with shared block devices */
4383 static struct md_sysfs_entry md_sync_force_parallel
=
4384 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4385 sync_force_parallel_show
, sync_force_parallel_store
);
4388 sync_speed_show(struct mddev
*mddev
, char *page
)
4390 unsigned long resync
, dt
, db
;
4391 if (mddev
->curr_resync
== 0)
4392 return sprintf(page
, "none\n");
4393 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4394 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4396 db
= resync
- mddev
->resync_mark_cnt
;
4397 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4400 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4403 sync_completed_show(struct mddev
*mddev
, char *page
)
4405 unsigned long long max_sectors
, resync
;
4407 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4408 return sprintf(page
, "none\n");
4410 if (mddev
->curr_resync
== 1 ||
4411 mddev
->curr_resync
== 2)
4412 return sprintf(page
, "delayed\n");
4414 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4415 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4416 max_sectors
= mddev
->resync_max_sectors
;
4418 max_sectors
= mddev
->dev_sectors
;
4420 resync
= mddev
->curr_resync_completed
;
4421 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4424 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4427 min_sync_show(struct mddev
*mddev
, char *page
)
4429 return sprintf(page
, "%llu\n",
4430 (unsigned long long)mddev
->resync_min
);
4433 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4435 unsigned long long min
;
4436 if (strict_strtoull(buf
, 10, &min
))
4438 if (min
> mddev
->resync_max
)
4440 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4443 /* Must be a multiple of chunk_size */
4444 if (mddev
->chunk_sectors
) {
4445 sector_t temp
= min
;
4446 if (sector_div(temp
, mddev
->chunk_sectors
))
4449 mddev
->resync_min
= min
;
4454 static struct md_sysfs_entry md_min_sync
=
4455 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4458 max_sync_show(struct mddev
*mddev
, char *page
)
4460 if (mddev
->resync_max
== MaxSector
)
4461 return sprintf(page
, "max\n");
4463 return sprintf(page
, "%llu\n",
4464 (unsigned long long)mddev
->resync_max
);
4467 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4469 if (strncmp(buf
, "max", 3) == 0)
4470 mddev
->resync_max
= MaxSector
;
4472 unsigned long long max
;
4473 if (strict_strtoull(buf
, 10, &max
))
4475 if (max
< mddev
->resync_min
)
4477 if (max
< mddev
->resync_max
&&
4479 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4482 /* Must be a multiple of chunk_size */
4483 if (mddev
->chunk_sectors
) {
4484 sector_t temp
= max
;
4485 if (sector_div(temp
, mddev
->chunk_sectors
))
4488 mddev
->resync_max
= max
;
4490 wake_up(&mddev
->recovery_wait
);
4494 static struct md_sysfs_entry md_max_sync
=
4495 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4498 suspend_lo_show(struct mddev
*mddev
, char *page
)
4500 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4504 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4507 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4508 unsigned long long old
= mddev
->suspend_lo
;
4510 if (mddev
->pers
== NULL
||
4511 mddev
->pers
->quiesce
== NULL
)
4513 if (buf
== e
|| (*e
&& *e
!= '\n'))
4516 mddev
->suspend_lo
= new;
4518 /* Shrinking suspended region */
4519 mddev
->pers
->quiesce(mddev
, 2);
4521 /* Expanding suspended region - need to wait */
4522 mddev
->pers
->quiesce(mddev
, 1);
4523 mddev
->pers
->quiesce(mddev
, 0);
4527 static struct md_sysfs_entry md_suspend_lo
=
4528 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4532 suspend_hi_show(struct mddev
*mddev
, char *page
)
4534 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4538 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4541 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4542 unsigned long long old
= mddev
->suspend_hi
;
4544 if (mddev
->pers
== NULL
||
4545 mddev
->pers
->quiesce
== NULL
)
4547 if (buf
== e
|| (*e
&& *e
!= '\n'))
4550 mddev
->suspend_hi
= new;
4552 /* Shrinking suspended region */
4553 mddev
->pers
->quiesce(mddev
, 2);
4555 /* Expanding suspended region - need to wait */
4556 mddev
->pers
->quiesce(mddev
, 1);
4557 mddev
->pers
->quiesce(mddev
, 0);
4561 static struct md_sysfs_entry md_suspend_hi
=
4562 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4565 reshape_position_show(struct mddev
*mddev
, char *page
)
4567 if (mddev
->reshape_position
!= MaxSector
)
4568 return sprintf(page
, "%llu\n",
4569 (unsigned long long)mddev
->reshape_position
);
4570 strcpy(page
, "none\n");
4575 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4577 struct md_rdev
*rdev
;
4579 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4582 if (buf
== e
|| (*e
&& *e
!= '\n'))
4584 mddev
->reshape_position
= new;
4585 mddev
->delta_disks
= 0;
4586 mddev
->reshape_backwards
= 0;
4587 mddev
->new_level
= mddev
->level
;
4588 mddev
->new_layout
= mddev
->layout
;
4589 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4590 rdev_for_each(rdev
, mddev
)
4591 rdev
->new_data_offset
= rdev
->data_offset
;
4595 static struct md_sysfs_entry md_reshape_position
=
4596 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4597 reshape_position_store
);
4600 reshape_direction_show(struct mddev
*mddev
, char *page
)
4602 return sprintf(page
, "%s\n",
4603 mddev
->reshape_backwards
? "backwards" : "forwards");
4607 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4610 if (cmd_match(buf
, "forwards"))
4612 else if (cmd_match(buf
, "backwards"))
4616 if (mddev
->reshape_backwards
== backwards
)
4619 /* check if we are allowed to change */
4620 if (mddev
->delta_disks
)
4623 if (mddev
->persistent
&&
4624 mddev
->major_version
== 0)
4627 mddev
->reshape_backwards
= backwards
;
4631 static struct md_sysfs_entry md_reshape_direction
=
4632 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4633 reshape_direction_store
);
4636 array_size_show(struct mddev
*mddev
, char *page
)
4638 if (mddev
->external_size
)
4639 return sprintf(page
, "%llu\n",
4640 (unsigned long long)mddev
->array_sectors
/2);
4642 return sprintf(page
, "default\n");
4646 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4650 if (strncmp(buf
, "default", 7) == 0) {
4652 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4654 sectors
= mddev
->array_sectors
;
4656 mddev
->external_size
= 0;
4658 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4660 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4663 mddev
->external_size
= 1;
4666 mddev
->array_sectors
= sectors
;
4668 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4669 revalidate_disk(mddev
->gendisk
);
4674 static struct md_sysfs_entry md_array_size
=
4675 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4678 static struct attribute
*md_default_attrs
[] = {
4681 &md_raid_disks
.attr
,
4682 &md_chunk_size
.attr
,
4684 &md_resync_start
.attr
,
4686 &md_new_device
.attr
,
4687 &md_safe_delay
.attr
,
4688 &md_array_state
.attr
,
4689 &md_reshape_position
.attr
,
4690 &md_reshape_direction
.attr
,
4691 &md_array_size
.attr
,
4692 &max_corr_read_errors
.attr
,
4696 static struct attribute
*md_redundancy_attrs
[] = {
4698 &md_mismatches
.attr
,
4701 &md_sync_speed
.attr
,
4702 &md_sync_force_parallel
.attr
,
4703 &md_sync_completed
.attr
,
4706 &md_suspend_lo
.attr
,
4707 &md_suspend_hi
.attr
,
4712 static struct attribute_group md_redundancy_group
= {
4714 .attrs
= md_redundancy_attrs
,
4719 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4721 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4722 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4727 spin_lock(&all_mddevs_lock
);
4728 if (list_empty(&mddev
->all_mddevs
)) {
4729 spin_unlock(&all_mddevs_lock
);
4733 spin_unlock(&all_mddevs_lock
);
4735 rv
= mddev_lock(mddev
);
4737 rv
= entry
->show(mddev
, page
);
4738 mddev_unlock(mddev
);
4745 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4746 const char *page
, size_t length
)
4748 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4749 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4754 if (!capable(CAP_SYS_ADMIN
))
4756 spin_lock(&all_mddevs_lock
);
4757 if (list_empty(&mddev
->all_mddevs
)) {
4758 spin_unlock(&all_mddevs_lock
);
4762 spin_unlock(&all_mddevs_lock
);
4763 if (entry
->store
== new_dev_store
)
4764 flush_workqueue(md_misc_wq
);
4765 rv
= mddev_lock(mddev
);
4767 rv
= entry
->store(mddev
, page
, length
);
4768 mddev_unlock(mddev
);
4774 static void md_free(struct kobject
*ko
)
4776 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4778 if (mddev
->sysfs_state
)
4779 sysfs_put(mddev
->sysfs_state
);
4781 if (mddev
->gendisk
) {
4782 del_gendisk(mddev
->gendisk
);
4783 put_disk(mddev
->gendisk
);
4786 blk_cleanup_queue(mddev
->queue
);
4791 static const struct sysfs_ops md_sysfs_ops
= {
4792 .show
= md_attr_show
,
4793 .store
= md_attr_store
,
4795 static struct kobj_type md_ktype
= {
4797 .sysfs_ops
= &md_sysfs_ops
,
4798 .default_attrs
= md_default_attrs
,
4803 static void mddev_delayed_delete(struct work_struct
*ws
)
4805 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4807 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4808 kobject_del(&mddev
->kobj
);
4809 kobject_put(&mddev
->kobj
);
4812 static int md_alloc(dev_t dev
, char *name
)
4814 static DEFINE_MUTEX(disks_mutex
);
4815 struct mddev
*mddev
= mddev_find(dev
);
4816 struct gendisk
*disk
;
4825 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4826 shift
= partitioned
? MdpMinorShift
: 0;
4827 unit
= MINOR(mddev
->unit
) >> shift
;
4829 /* wait for any previous instance of this device to be
4830 * completely removed (mddev_delayed_delete).
4832 flush_workqueue(md_misc_wq
);
4834 mutex_lock(&disks_mutex
);
4840 /* Need to ensure that 'name' is not a duplicate.
4842 struct mddev
*mddev2
;
4843 spin_lock(&all_mddevs_lock
);
4845 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4846 if (mddev2
->gendisk
&&
4847 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4848 spin_unlock(&all_mddevs_lock
);
4851 spin_unlock(&all_mddevs_lock
);
4855 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4858 mddev
->queue
->queuedata
= mddev
;
4860 blk_queue_make_request(mddev
->queue
, md_make_request
);
4861 blk_set_stacking_limits(&mddev
->queue
->limits
);
4863 disk
= alloc_disk(1 << shift
);
4865 blk_cleanup_queue(mddev
->queue
);
4866 mddev
->queue
= NULL
;
4869 disk
->major
= MAJOR(mddev
->unit
);
4870 disk
->first_minor
= unit
<< shift
;
4872 strcpy(disk
->disk_name
, name
);
4873 else if (partitioned
)
4874 sprintf(disk
->disk_name
, "md_d%d", unit
);
4876 sprintf(disk
->disk_name
, "md%d", unit
);
4877 disk
->fops
= &md_fops
;
4878 disk
->private_data
= mddev
;
4879 disk
->queue
= mddev
->queue
;
4880 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4881 /* Allow extended partitions. This makes the
4882 * 'mdp' device redundant, but we can't really
4885 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4886 mddev
->gendisk
= disk
;
4887 /* As soon as we call add_disk(), another thread could get
4888 * through to md_open, so make sure it doesn't get too far
4890 mutex_lock(&mddev
->open_mutex
);
4893 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4894 &disk_to_dev(disk
)->kobj
, "%s", "md");
4896 /* This isn't possible, but as kobject_init_and_add is marked
4897 * __must_check, we must do something with the result
4899 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4903 if (mddev
->kobj
.sd
&&
4904 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4905 printk(KERN_DEBUG
"pointless warning\n");
4906 mutex_unlock(&mddev
->open_mutex
);
4908 mutex_unlock(&disks_mutex
);
4909 if (!error
&& mddev
->kobj
.sd
) {
4910 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4911 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4917 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4919 md_alloc(dev
, NULL
);
4923 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4925 /* val must be "md_*" where * is not all digits.
4926 * We allocate an array with a large free minor number, and
4927 * set the name to val. val must not already be an active name.
4929 int len
= strlen(val
);
4930 char buf
[DISK_NAME_LEN
];
4932 while (len
&& val
[len
-1] == '\n')
4934 if (len
>= DISK_NAME_LEN
)
4936 strlcpy(buf
, val
, len
+1);
4937 if (strncmp(buf
, "md_", 3) != 0)
4939 return md_alloc(0, buf
);
4942 static void md_safemode_timeout(unsigned long data
)
4944 struct mddev
*mddev
= (struct mddev
*) data
;
4946 if (!atomic_read(&mddev
->writes_pending
)) {
4947 mddev
->safemode
= 1;
4948 if (mddev
->external
)
4949 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4951 md_wakeup_thread(mddev
->thread
);
4954 static int start_dirty_degraded
;
4956 int md_run(struct mddev
*mddev
)
4959 struct md_rdev
*rdev
;
4960 struct md_personality
*pers
;
4962 if (list_empty(&mddev
->disks
))
4963 /* cannot run an array with no devices.. */
4968 /* Cannot run until previous stop completes properly */
4969 if (mddev
->sysfs_active
)
4973 * Analyze all RAID superblock(s)
4975 if (!mddev
->raid_disks
) {
4976 if (!mddev
->persistent
)
4981 if (mddev
->level
!= LEVEL_NONE
)
4982 request_module("md-level-%d", mddev
->level
);
4983 else if (mddev
->clevel
[0])
4984 request_module("md-%s", mddev
->clevel
);
4987 * Drop all container device buffers, from now on
4988 * the only valid external interface is through the md
4991 rdev_for_each(rdev
, mddev
) {
4992 if (test_bit(Faulty
, &rdev
->flags
))
4994 sync_blockdev(rdev
->bdev
);
4995 invalidate_bdev(rdev
->bdev
);
4997 /* perform some consistency tests on the device.
4998 * We don't want the data to overlap the metadata,
4999 * Internal Bitmap issues have been handled elsewhere.
5001 if (rdev
->meta_bdev
) {
5002 /* Nothing to check */;
5003 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5004 if (mddev
->dev_sectors
&&
5005 rdev
->data_offset
+ mddev
->dev_sectors
5007 printk("md: %s: data overlaps metadata\n",
5012 if (rdev
->sb_start
+ rdev
->sb_size
/512
5013 > rdev
->data_offset
) {
5014 printk("md: %s: metadata overlaps data\n",
5019 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5022 if (mddev
->bio_set
== NULL
)
5023 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
5025 spin_lock(&pers_lock
);
5026 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5027 if (!pers
|| !try_module_get(pers
->owner
)) {
5028 spin_unlock(&pers_lock
);
5029 if (mddev
->level
!= LEVEL_NONE
)
5030 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5033 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5038 spin_unlock(&pers_lock
);
5039 if (mddev
->level
!= pers
->level
) {
5040 mddev
->level
= pers
->level
;
5041 mddev
->new_level
= pers
->level
;
5043 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5045 if (mddev
->reshape_position
!= MaxSector
&&
5046 pers
->start_reshape
== NULL
) {
5047 /* This personality cannot handle reshaping... */
5049 module_put(pers
->owner
);
5053 if (pers
->sync_request
) {
5054 /* Warn if this is a potentially silly
5057 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5058 struct md_rdev
*rdev2
;
5061 rdev_for_each(rdev
, mddev
)
5062 rdev_for_each(rdev2
, mddev
) {
5064 rdev
->bdev
->bd_contains
==
5065 rdev2
->bdev
->bd_contains
) {
5067 "%s: WARNING: %s appears to be"
5068 " on the same physical disk as"
5071 bdevname(rdev
->bdev
,b
),
5072 bdevname(rdev2
->bdev
,b2
));
5079 "True protection against single-disk"
5080 " failure might be compromised.\n");
5083 mddev
->recovery
= 0;
5084 /* may be over-ridden by personality */
5085 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5087 mddev
->ok_start_degraded
= start_dirty_degraded
;
5089 if (start_readonly
&& mddev
->ro
== 0)
5090 mddev
->ro
= 2; /* read-only, but switch on first write */
5092 err
= mddev
->pers
->run(mddev
);
5094 printk(KERN_ERR
"md: pers->run() failed ...\n");
5095 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5096 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5097 " but 'external_size' not in effect?\n", __func__
);
5099 "md: invalid array_size %llu > default size %llu\n",
5100 (unsigned long long)mddev
->array_sectors
/ 2,
5101 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
5103 mddev
->pers
->stop(mddev
);
5105 if (err
== 0 && mddev
->pers
->sync_request
&&
5106 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5107 err
= bitmap_create(mddev
);
5109 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5110 mdname(mddev
), err
);
5111 mddev
->pers
->stop(mddev
);
5115 module_put(mddev
->pers
->owner
);
5117 bitmap_destroy(mddev
);
5120 if (mddev
->pers
->sync_request
) {
5121 if (mddev
->kobj
.sd
&&
5122 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5124 "md: cannot register extra attributes for %s\n",
5126 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5127 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5130 atomic_set(&mddev
->writes_pending
,0);
5131 atomic_set(&mddev
->max_corr_read_errors
,
5132 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5133 mddev
->safemode
= 0;
5134 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5135 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5136 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5140 rdev_for_each(rdev
, mddev
)
5141 if (rdev
->raid_disk
>= 0)
5142 if (sysfs_link_rdev(mddev
, rdev
))
5143 /* failure here is OK */;
5145 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5148 md_update_sb(mddev
, 0);
5150 md_new_event(mddev
);
5151 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5152 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5153 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5156 EXPORT_SYMBOL_GPL(md_run
);
5158 static int do_md_run(struct mddev
*mddev
)
5162 err
= md_run(mddev
);
5165 err
= bitmap_load(mddev
);
5167 bitmap_destroy(mddev
);
5171 md_wakeup_thread(mddev
->thread
);
5172 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5174 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5175 revalidate_disk(mddev
->gendisk
);
5177 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5182 static int restart_array(struct mddev
*mddev
)
5184 struct gendisk
*disk
= mddev
->gendisk
;
5186 /* Complain if it has no devices */
5187 if (list_empty(&mddev
->disks
))
5193 mddev
->safemode
= 0;
5195 set_disk_ro(disk
, 0);
5196 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5198 /* Kick recovery or resync if necessary */
5199 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5200 md_wakeup_thread(mddev
->thread
);
5201 md_wakeup_thread(mddev
->sync_thread
);
5202 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5206 /* similar to deny_write_access, but accounts for our holding a reference
5207 * to the file ourselves */
5208 static int deny_bitmap_write_access(struct file
* file
)
5210 struct inode
*inode
= file
->f_mapping
->host
;
5212 spin_lock(&inode
->i_lock
);
5213 if (atomic_read(&inode
->i_writecount
) > 1) {
5214 spin_unlock(&inode
->i_lock
);
5217 atomic_set(&inode
->i_writecount
, -1);
5218 spin_unlock(&inode
->i_lock
);
5223 void restore_bitmap_write_access(struct file
*file
)
5225 struct inode
*inode
= file
->f_mapping
->host
;
5227 spin_lock(&inode
->i_lock
);
5228 atomic_set(&inode
->i_writecount
, 1);
5229 spin_unlock(&inode
->i_lock
);
5232 static void md_clean(struct mddev
*mddev
)
5234 mddev
->array_sectors
= 0;
5235 mddev
->external_size
= 0;
5236 mddev
->dev_sectors
= 0;
5237 mddev
->raid_disks
= 0;
5238 mddev
->recovery_cp
= 0;
5239 mddev
->resync_min
= 0;
5240 mddev
->resync_max
= MaxSector
;
5241 mddev
->reshape_position
= MaxSector
;
5242 mddev
->external
= 0;
5243 mddev
->persistent
= 0;
5244 mddev
->level
= LEVEL_NONE
;
5245 mddev
->clevel
[0] = 0;
5248 mddev
->metadata_type
[0] = 0;
5249 mddev
->chunk_sectors
= 0;
5250 mddev
->ctime
= mddev
->utime
= 0;
5252 mddev
->max_disks
= 0;
5254 mddev
->can_decrease_events
= 0;
5255 mddev
->delta_disks
= 0;
5256 mddev
->reshape_backwards
= 0;
5257 mddev
->new_level
= LEVEL_NONE
;
5258 mddev
->new_layout
= 0;
5259 mddev
->new_chunk_sectors
= 0;
5260 mddev
->curr_resync
= 0;
5261 atomic64_set(&mddev
->resync_mismatches
, 0);
5262 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5263 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5264 mddev
->recovery
= 0;
5267 mddev
->degraded
= 0;
5268 mddev
->safemode
= 0;
5269 mddev
->merge_check_needed
= 0;
5270 mddev
->bitmap_info
.offset
= 0;
5271 mddev
->bitmap_info
.default_offset
= 0;
5272 mddev
->bitmap_info
.default_space
= 0;
5273 mddev
->bitmap_info
.chunksize
= 0;
5274 mddev
->bitmap_info
.daemon_sleep
= 0;
5275 mddev
->bitmap_info
.max_write_behind
= 0;
5278 static void __md_stop_writes(struct mddev
*mddev
)
5280 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5281 if (mddev
->sync_thread
) {
5282 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5283 md_reap_sync_thread(mddev
);
5286 del_timer_sync(&mddev
->safemode_timer
);
5288 bitmap_flush(mddev
);
5289 md_super_wait(mddev
);
5291 if (mddev
->ro
== 0 &&
5292 (!mddev
->in_sync
|| mddev
->flags
)) {
5293 /* mark array as shutdown cleanly */
5295 md_update_sb(mddev
, 1);
5299 void md_stop_writes(struct mddev
*mddev
)
5302 __md_stop_writes(mddev
);
5303 mddev_unlock(mddev
);
5305 EXPORT_SYMBOL_GPL(md_stop_writes
);
5307 static void __md_stop(struct mddev
*mddev
)
5310 mddev
->pers
->stop(mddev
);
5311 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5312 mddev
->to_remove
= &md_redundancy_group
;
5313 module_put(mddev
->pers
->owner
);
5315 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5318 void md_stop(struct mddev
*mddev
)
5320 /* stop the array and free an attached data structures.
5321 * This is called from dm-raid
5324 bitmap_destroy(mddev
);
5326 bioset_free(mddev
->bio_set
);
5329 EXPORT_SYMBOL_GPL(md_stop
);
5331 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5334 mutex_lock(&mddev
->open_mutex
);
5335 if (atomic_read(&mddev
->openers
) > !!bdev
) {
5336 printk("md: %s still in use.\n",mdname(mddev
));
5341 sync_blockdev(bdev
);
5343 __md_stop_writes(mddev
);
5349 set_disk_ro(mddev
->gendisk
, 1);
5350 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5351 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5355 mutex_unlock(&mddev
->open_mutex
);
5360 * 0 - completely stop and dis-assemble array
5361 * 2 - stop but do not disassemble array
5363 static int do_md_stop(struct mddev
* mddev
, int mode
,
5364 struct block_device
*bdev
)
5366 struct gendisk
*disk
= mddev
->gendisk
;
5367 struct md_rdev
*rdev
;
5369 mutex_lock(&mddev
->open_mutex
);
5370 if (atomic_read(&mddev
->openers
) > !!bdev
||
5371 mddev
->sysfs_active
) {
5372 printk("md: %s still in use.\n",mdname(mddev
));
5373 mutex_unlock(&mddev
->open_mutex
);
5377 /* It is possible IO was issued on some other
5378 * open file which was closed before we took ->open_mutex.
5379 * As that was not the last close __blkdev_put will not
5380 * have called sync_blockdev, so we must.
5382 sync_blockdev(bdev
);
5386 set_disk_ro(disk
, 0);
5388 __md_stop_writes(mddev
);
5390 mddev
->queue
->merge_bvec_fn
= NULL
;
5391 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5393 /* tell userspace to handle 'inactive' */
5394 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5396 rdev_for_each(rdev
, mddev
)
5397 if (rdev
->raid_disk
>= 0)
5398 sysfs_unlink_rdev(mddev
, rdev
);
5400 set_capacity(disk
, 0);
5401 mutex_unlock(&mddev
->open_mutex
);
5403 revalidate_disk(disk
);
5408 mutex_unlock(&mddev
->open_mutex
);
5410 * Free resources if final stop
5413 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5415 bitmap_destroy(mddev
);
5416 if (mddev
->bitmap_info
.file
) {
5417 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5418 fput(mddev
->bitmap_info
.file
);
5419 mddev
->bitmap_info
.file
= NULL
;
5421 mddev
->bitmap_info
.offset
= 0;
5423 export_array(mddev
);
5426 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5427 if (mddev
->hold_active
== UNTIL_STOP
)
5428 mddev
->hold_active
= 0;
5430 blk_integrity_unregister(disk
);
5431 md_new_event(mddev
);
5432 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5437 static void autorun_array(struct mddev
*mddev
)
5439 struct md_rdev
*rdev
;
5442 if (list_empty(&mddev
->disks
))
5445 printk(KERN_INFO
"md: running: ");
5447 rdev_for_each(rdev
, mddev
) {
5448 char b
[BDEVNAME_SIZE
];
5449 printk("<%s>", bdevname(rdev
->bdev
,b
));
5453 err
= do_md_run(mddev
);
5455 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5456 do_md_stop(mddev
, 0, NULL
);
5461 * lets try to run arrays based on all disks that have arrived
5462 * until now. (those are in pending_raid_disks)
5464 * the method: pick the first pending disk, collect all disks with
5465 * the same UUID, remove all from the pending list and put them into
5466 * the 'same_array' list. Then order this list based on superblock
5467 * update time (freshest comes first), kick out 'old' disks and
5468 * compare superblocks. If everything's fine then run it.
5470 * If "unit" is allocated, then bump its reference count
5472 static void autorun_devices(int part
)
5474 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5475 struct mddev
*mddev
;
5476 char b
[BDEVNAME_SIZE
];
5478 printk(KERN_INFO
"md: autorun ...\n");
5479 while (!list_empty(&pending_raid_disks
)) {
5482 LIST_HEAD(candidates
);
5483 rdev0
= list_entry(pending_raid_disks
.next
,
5484 struct md_rdev
, same_set
);
5486 printk(KERN_INFO
"md: considering %s ...\n",
5487 bdevname(rdev0
->bdev
,b
));
5488 INIT_LIST_HEAD(&candidates
);
5489 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5490 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5491 printk(KERN_INFO
"md: adding %s ...\n",
5492 bdevname(rdev
->bdev
,b
));
5493 list_move(&rdev
->same_set
, &candidates
);
5496 * now we have a set of devices, with all of them having
5497 * mostly sane superblocks. It's time to allocate the
5501 dev
= MKDEV(mdp_major
,
5502 rdev0
->preferred_minor
<< MdpMinorShift
);
5503 unit
= MINOR(dev
) >> MdpMinorShift
;
5505 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5508 if (rdev0
->preferred_minor
!= unit
) {
5509 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5510 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5514 md_probe(dev
, NULL
, NULL
);
5515 mddev
= mddev_find(dev
);
5516 if (!mddev
|| !mddev
->gendisk
) {
5520 "md: cannot allocate memory for md drive.\n");
5523 if (mddev_lock(mddev
))
5524 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5526 else if (mddev
->raid_disks
|| mddev
->major_version
5527 || !list_empty(&mddev
->disks
)) {
5529 "md: %s already running, cannot run %s\n",
5530 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5531 mddev_unlock(mddev
);
5533 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5534 mddev
->persistent
= 1;
5535 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5536 list_del_init(&rdev
->same_set
);
5537 if (bind_rdev_to_array(rdev
, mddev
))
5540 autorun_array(mddev
);
5541 mddev_unlock(mddev
);
5543 /* on success, candidates will be empty, on error
5546 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5547 list_del_init(&rdev
->same_set
);
5552 printk(KERN_INFO
"md: ... autorun DONE.\n");
5554 #endif /* !MODULE */
5556 static int get_version(void __user
* arg
)
5560 ver
.major
= MD_MAJOR_VERSION
;
5561 ver
.minor
= MD_MINOR_VERSION
;
5562 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5564 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5570 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5572 mdu_array_info_t info
;
5573 int nr
,working
,insync
,failed
,spare
;
5574 struct md_rdev
*rdev
;
5576 nr
= working
= insync
= failed
= spare
= 0;
5578 rdev_for_each_rcu(rdev
, mddev
) {
5580 if (test_bit(Faulty
, &rdev
->flags
))
5584 if (test_bit(In_sync
, &rdev
->flags
))
5592 info
.major_version
= mddev
->major_version
;
5593 info
.minor_version
= mddev
->minor_version
;
5594 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5595 info
.ctime
= mddev
->ctime
;
5596 info
.level
= mddev
->level
;
5597 info
.size
= mddev
->dev_sectors
/ 2;
5598 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5601 info
.raid_disks
= mddev
->raid_disks
;
5602 info
.md_minor
= mddev
->md_minor
;
5603 info
.not_persistent
= !mddev
->persistent
;
5605 info
.utime
= mddev
->utime
;
5608 info
.state
= (1<<MD_SB_CLEAN
);
5609 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5610 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5611 info
.active_disks
= insync
;
5612 info
.working_disks
= working
;
5613 info
.failed_disks
= failed
;
5614 info
.spare_disks
= spare
;
5616 info
.layout
= mddev
->layout
;
5617 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5619 if (copy_to_user(arg
, &info
, sizeof(info
)))
5625 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5627 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5628 char *ptr
, *buf
= NULL
;
5631 if (md_allow_write(mddev
))
5632 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5634 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5639 /* bitmap disabled, zero the first byte and copy out */
5640 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5641 file
->pathname
[0] = '\0';
5645 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5649 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5650 buf
, sizeof(file
->pathname
));
5654 strcpy(file
->pathname
, ptr
);
5658 if (copy_to_user(arg
, file
, sizeof(*file
)))
5666 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5668 mdu_disk_info_t info
;
5669 struct md_rdev
*rdev
;
5671 if (copy_from_user(&info
, arg
, sizeof(info
)))
5675 rdev
= find_rdev_nr_rcu(mddev
, info
.number
);
5677 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5678 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5679 info
.raid_disk
= rdev
->raid_disk
;
5681 if (test_bit(Faulty
, &rdev
->flags
))
5682 info
.state
|= (1<<MD_DISK_FAULTY
);
5683 else if (test_bit(In_sync
, &rdev
->flags
)) {
5684 info
.state
|= (1<<MD_DISK_ACTIVE
);
5685 info
.state
|= (1<<MD_DISK_SYNC
);
5687 if (test_bit(WriteMostly
, &rdev
->flags
))
5688 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5690 info
.major
= info
.minor
= 0;
5691 info
.raid_disk
= -1;
5692 info
.state
= (1<<MD_DISK_REMOVED
);
5696 if (copy_to_user(arg
, &info
, sizeof(info
)))
5702 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5704 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5705 struct md_rdev
*rdev
;
5706 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5708 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5711 if (!mddev
->raid_disks
) {
5713 /* expecting a device which has a superblock */
5714 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5717 "md: md_import_device returned %ld\n",
5719 return PTR_ERR(rdev
);
5721 if (!list_empty(&mddev
->disks
)) {
5722 struct md_rdev
*rdev0
5723 = list_entry(mddev
->disks
.next
,
5724 struct md_rdev
, same_set
);
5725 err
= super_types
[mddev
->major_version
]
5726 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5729 "md: %s has different UUID to %s\n",
5730 bdevname(rdev
->bdev
,b
),
5731 bdevname(rdev0
->bdev
,b2
));
5736 err
= bind_rdev_to_array(rdev
, mddev
);
5743 * add_new_disk can be used once the array is assembled
5744 * to add "hot spares". They must already have a superblock
5749 if (!mddev
->pers
->hot_add_disk
) {
5751 "%s: personality does not support diskops!\n",
5755 if (mddev
->persistent
)
5756 rdev
= md_import_device(dev
, mddev
->major_version
,
5757 mddev
->minor_version
);
5759 rdev
= md_import_device(dev
, -1, -1);
5762 "md: md_import_device returned %ld\n",
5764 return PTR_ERR(rdev
);
5766 /* set saved_raid_disk if appropriate */
5767 if (!mddev
->persistent
) {
5768 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5769 info
->raid_disk
< mddev
->raid_disks
) {
5770 rdev
->raid_disk
= info
->raid_disk
;
5771 set_bit(In_sync
, &rdev
->flags
);
5772 clear_bit(Bitmap_sync
, &rdev
->flags
);
5774 rdev
->raid_disk
= -1;
5776 super_types
[mddev
->major_version
].
5777 validate_super(mddev
, rdev
);
5778 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5779 rdev
->raid_disk
!= info
->raid_disk
) {
5780 /* This was a hot-add request, but events doesn't
5781 * match, so reject it.
5787 if (test_bit(In_sync
, &rdev
->flags
))
5788 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5790 rdev
->saved_raid_disk
= -1;
5792 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5793 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5794 set_bit(WriteMostly
, &rdev
->flags
);
5796 clear_bit(WriteMostly
, &rdev
->flags
);
5798 rdev
->raid_disk
= -1;
5799 err
= bind_rdev_to_array(rdev
, mddev
);
5800 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5801 /* If there is hot_add_disk but no hot_remove_disk
5802 * then added disks for geometry changes,
5803 * and should be added immediately.
5805 super_types
[mddev
->major_version
].
5806 validate_super(mddev
, rdev
);
5807 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5809 unbind_rdev_from_array(rdev
);
5814 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5816 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5817 if (mddev
->degraded
)
5818 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5819 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5821 md_new_event(mddev
);
5822 md_wakeup_thread(mddev
->thread
);
5826 /* otherwise, add_new_disk is only allowed
5827 * for major_version==0 superblocks
5829 if (mddev
->major_version
!= 0) {
5830 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5835 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5837 rdev
= md_import_device(dev
, -1, 0);
5840 "md: error, md_import_device() returned %ld\n",
5842 return PTR_ERR(rdev
);
5844 rdev
->desc_nr
= info
->number
;
5845 if (info
->raid_disk
< mddev
->raid_disks
)
5846 rdev
->raid_disk
= info
->raid_disk
;
5848 rdev
->raid_disk
= -1;
5850 if (rdev
->raid_disk
< mddev
->raid_disks
)
5851 if (info
->state
& (1<<MD_DISK_SYNC
))
5852 set_bit(In_sync
, &rdev
->flags
);
5854 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5855 set_bit(WriteMostly
, &rdev
->flags
);
5857 if (!mddev
->persistent
) {
5858 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5859 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5861 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5862 rdev
->sectors
= rdev
->sb_start
;
5864 err
= bind_rdev_to_array(rdev
, mddev
);
5874 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5876 char b
[BDEVNAME_SIZE
];
5877 struct md_rdev
*rdev
;
5879 rdev
= find_rdev(mddev
, dev
);
5883 clear_bit(Blocked
, &rdev
->flags
);
5884 remove_and_add_spares(mddev
, rdev
);
5886 if (rdev
->raid_disk
>= 0)
5889 kick_rdev_from_array(rdev
);
5890 md_update_sb(mddev
, 1);
5891 md_new_event(mddev
);
5895 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5896 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5900 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5902 char b
[BDEVNAME_SIZE
];
5904 struct md_rdev
*rdev
;
5909 if (mddev
->major_version
!= 0) {
5910 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5911 " version-0 superblocks.\n",
5915 if (!mddev
->pers
->hot_add_disk
) {
5917 "%s: personality does not support diskops!\n",
5922 rdev
= md_import_device(dev
, -1, 0);
5925 "md: error, md_import_device() returned %ld\n",
5930 if (mddev
->persistent
)
5931 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5933 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5935 rdev
->sectors
= rdev
->sb_start
;
5937 if (test_bit(Faulty
, &rdev
->flags
)) {
5939 "md: can not hot-add faulty %s disk to %s!\n",
5940 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5944 clear_bit(In_sync
, &rdev
->flags
);
5946 rdev
->saved_raid_disk
= -1;
5947 err
= bind_rdev_to_array(rdev
, mddev
);
5952 * The rest should better be atomic, we can have disk failures
5953 * noticed in interrupt contexts ...
5956 rdev
->raid_disk
= -1;
5958 md_update_sb(mddev
, 1);
5961 * Kick recovery, maybe this spare has to be added to the
5962 * array immediately.
5964 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5965 md_wakeup_thread(mddev
->thread
);
5966 md_new_event(mddev
);
5974 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5979 if (!mddev
->pers
->quiesce
)
5981 if (mddev
->recovery
|| mddev
->sync_thread
)
5983 /* we should be able to change the bitmap.. */
5989 return -EEXIST
; /* cannot add when bitmap is present */
5990 mddev
->bitmap_info
.file
= fget(fd
);
5992 if (mddev
->bitmap_info
.file
== NULL
) {
5993 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5998 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
6000 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
6002 fput(mddev
->bitmap_info
.file
);
6003 mddev
->bitmap_info
.file
= NULL
;
6006 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6007 } else if (mddev
->bitmap
== NULL
)
6008 return -ENOENT
; /* cannot remove what isn't there */
6011 mddev
->pers
->quiesce(mddev
, 1);
6013 err
= bitmap_create(mddev
);
6015 err
= bitmap_load(mddev
);
6017 if (fd
< 0 || err
) {
6018 bitmap_destroy(mddev
);
6019 fd
= -1; /* make sure to put the file */
6021 mddev
->pers
->quiesce(mddev
, 0);
6024 if (mddev
->bitmap_info
.file
) {
6025 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
6026 fput(mddev
->bitmap_info
.file
);
6028 mddev
->bitmap_info
.file
= NULL
;
6035 * set_array_info is used two different ways
6036 * The original usage is when creating a new array.
6037 * In this usage, raid_disks is > 0 and it together with
6038 * level, size, not_persistent,layout,chunksize determine the
6039 * shape of the array.
6040 * This will always create an array with a type-0.90.0 superblock.
6041 * The newer usage is when assembling an array.
6042 * In this case raid_disks will be 0, and the major_version field is
6043 * use to determine which style super-blocks are to be found on the devices.
6044 * The minor and patch _version numbers are also kept incase the
6045 * super_block handler wishes to interpret them.
6047 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
6050 if (info
->raid_disks
== 0) {
6051 /* just setting version number for superblock loading */
6052 if (info
->major_version
< 0 ||
6053 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6054 super_types
[info
->major_version
].name
== NULL
) {
6055 /* maybe try to auto-load a module? */
6057 "md: superblock version %d not known\n",
6058 info
->major_version
);
6061 mddev
->major_version
= info
->major_version
;
6062 mddev
->minor_version
= info
->minor_version
;
6063 mddev
->patch_version
= info
->patch_version
;
6064 mddev
->persistent
= !info
->not_persistent
;
6065 /* ensure mddev_put doesn't delete this now that there
6066 * is some minimal configuration.
6068 mddev
->ctime
= get_seconds();
6071 mddev
->major_version
= MD_MAJOR_VERSION
;
6072 mddev
->minor_version
= MD_MINOR_VERSION
;
6073 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6074 mddev
->ctime
= get_seconds();
6076 mddev
->level
= info
->level
;
6077 mddev
->clevel
[0] = 0;
6078 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6079 mddev
->raid_disks
= info
->raid_disks
;
6080 /* don't set md_minor, it is determined by which /dev/md* was
6083 if (info
->state
& (1<<MD_SB_CLEAN
))
6084 mddev
->recovery_cp
= MaxSector
;
6086 mddev
->recovery_cp
= 0;
6087 mddev
->persistent
= ! info
->not_persistent
;
6088 mddev
->external
= 0;
6090 mddev
->layout
= info
->layout
;
6091 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6093 mddev
->max_disks
= MD_SB_DISKS
;
6095 if (mddev
->persistent
)
6097 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6099 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6100 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6101 mddev
->bitmap_info
.offset
= 0;
6103 mddev
->reshape_position
= MaxSector
;
6106 * Generate a 128 bit UUID
6108 get_random_bytes(mddev
->uuid
, 16);
6110 mddev
->new_level
= mddev
->level
;
6111 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6112 mddev
->new_layout
= mddev
->layout
;
6113 mddev
->delta_disks
= 0;
6114 mddev
->reshape_backwards
= 0;
6119 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6121 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6123 if (mddev
->external_size
)
6126 mddev
->array_sectors
= array_sectors
;
6128 EXPORT_SYMBOL(md_set_array_sectors
);
6130 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6132 struct md_rdev
*rdev
;
6134 int fit
= (num_sectors
== 0);
6136 if (mddev
->pers
->resize
== NULL
)
6138 /* The "num_sectors" is the number of sectors of each device that
6139 * is used. This can only make sense for arrays with redundancy.
6140 * linear and raid0 always use whatever space is available. We can only
6141 * consider changing this number if no resync or reconstruction is
6142 * happening, and if the new size is acceptable. It must fit before the
6143 * sb_start or, if that is <data_offset, it must fit before the size
6144 * of each device. If num_sectors is zero, we find the largest size
6147 if (mddev
->sync_thread
)
6150 rdev_for_each(rdev
, mddev
) {
6151 sector_t avail
= rdev
->sectors
;
6153 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6154 num_sectors
= avail
;
6155 if (avail
< num_sectors
)
6158 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6160 revalidate_disk(mddev
->gendisk
);
6164 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6167 struct md_rdev
*rdev
;
6168 /* change the number of raid disks */
6169 if (mddev
->pers
->check_reshape
== NULL
)
6171 if (raid_disks
<= 0 ||
6172 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6174 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
6177 rdev_for_each(rdev
, mddev
) {
6178 if (mddev
->raid_disks
< raid_disks
&&
6179 rdev
->data_offset
< rdev
->new_data_offset
)
6181 if (mddev
->raid_disks
> raid_disks
&&
6182 rdev
->data_offset
> rdev
->new_data_offset
)
6186 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6187 if (mddev
->delta_disks
< 0)
6188 mddev
->reshape_backwards
= 1;
6189 else if (mddev
->delta_disks
> 0)
6190 mddev
->reshape_backwards
= 0;
6192 rv
= mddev
->pers
->check_reshape(mddev
);
6194 mddev
->delta_disks
= 0;
6195 mddev
->reshape_backwards
= 0;
6202 * update_array_info is used to change the configuration of an
6204 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6205 * fields in the info are checked against the array.
6206 * Any differences that cannot be handled will cause an error.
6207 * Normally, only one change can be managed at a time.
6209 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6215 /* calculate expected state,ignoring low bits */
6216 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6217 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6219 if (mddev
->major_version
!= info
->major_version
||
6220 mddev
->minor_version
!= info
->minor_version
||
6221 /* mddev->patch_version != info->patch_version || */
6222 mddev
->ctime
!= info
->ctime
||
6223 mddev
->level
!= info
->level
||
6224 /* mddev->layout != info->layout || */
6225 !mddev
->persistent
!= info
->not_persistent
||
6226 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6227 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6228 ((state
^info
->state
) & 0xfffffe00)
6231 /* Check there is only one change */
6232 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6234 if (mddev
->raid_disks
!= info
->raid_disks
)
6236 if (mddev
->layout
!= info
->layout
)
6238 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6245 if (mddev
->layout
!= info
->layout
) {
6247 * we don't need to do anything at the md level, the
6248 * personality will take care of it all.
6250 if (mddev
->pers
->check_reshape
== NULL
)
6253 mddev
->new_layout
= info
->layout
;
6254 rv
= mddev
->pers
->check_reshape(mddev
);
6256 mddev
->new_layout
= mddev
->layout
;
6260 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6261 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6263 if (mddev
->raid_disks
!= info
->raid_disks
)
6264 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6266 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6267 if (mddev
->pers
->quiesce
== NULL
)
6269 if (mddev
->recovery
|| mddev
->sync_thread
)
6271 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6272 /* add the bitmap */
6275 if (mddev
->bitmap_info
.default_offset
== 0)
6277 mddev
->bitmap_info
.offset
=
6278 mddev
->bitmap_info
.default_offset
;
6279 mddev
->bitmap_info
.space
=
6280 mddev
->bitmap_info
.default_space
;
6281 mddev
->pers
->quiesce(mddev
, 1);
6282 rv
= bitmap_create(mddev
);
6284 rv
= bitmap_load(mddev
);
6286 bitmap_destroy(mddev
);
6287 mddev
->pers
->quiesce(mddev
, 0);
6289 /* remove the bitmap */
6292 if (mddev
->bitmap
->storage
.file
)
6294 mddev
->pers
->quiesce(mddev
, 1);
6295 bitmap_destroy(mddev
);
6296 mddev
->pers
->quiesce(mddev
, 0);
6297 mddev
->bitmap_info
.offset
= 0;
6300 md_update_sb(mddev
, 1);
6304 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6306 struct md_rdev
*rdev
;
6309 if (mddev
->pers
== NULL
)
6313 rdev
= find_rdev_rcu(mddev
, dev
);
6317 md_error(mddev
, rdev
);
6318 if (!test_bit(Faulty
, &rdev
->flags
))
6326 * We have a problem here : there is no easy way to give a CHS
6327 * virtual geometry. We currently pretend that we have a 2 heads
6328 * 4 sectors (with a BIG number of cylinders...). This drives
6329 * dosfs just mad... ;-)
6331 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6333 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6337 geo
->cylinders
= mddev
->array_sectors
/ 8;
6341 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6342 unsigned int cmd
, unsigned long arg
)
6345 void __user
*argp
= (void __user
*)arg
;
6346 struct mddev
*mddev
= NULL
;
6351 case GET_ARRAY_INFO
:
6355 if (!capable(CAP_SYS_ADMIN
))
6360 * Commands dealing with the RAID driver but not any
6365 err
= get_version(argp
);
6368 case PRINT_RAID_DEBUG
:
6376 autostart_arrays(arg
);
6383 * Commands creating/starting a new array:
6386 mddev
= bdev
->bd_disk
->private_data
;
6393 /* Some actions do not requires the mutex */
6395 case GET_ARRAY_INFO
:
6396 if (!mddev
->raid_disks
&& !mddev
->external
)
6399 err
= get_array_info(mddev
, argp
);
6403 if (!mddev
->raid_disks
&& !mddev
->external
)
6406 err
= get_disk_info(mddev
, argp
);
6409 case SET_DISK_FAULTY
:
6410 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6414 if (cmd
== ADD_NEW_DISK
)
6415 /* need to ensure md_delayed_delete() has completed */
6416 flush_workqueue(md_misc_wq
);
6418 err
= mddev_lock(mddev
);
6421 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6426 if (cmd
== SET_ARRAY_INFO
) {
6427 mdu_array_info_t info
;
6429 memset(&info
, 0, sizeof(info
));
6430 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6435 err
= update_array_info(mddev
, &info
);
6437 printk(KERN_WARNING
"md: couldn't update"
6438 " array info. %d\n", err
);
6443 if (!list_empty(&mddev
->disks
)) {
6445 "md: array %s already has disks!\n",
6450 if (mddev
->raid_disks
) {
6452 "md: array %s already initialised!\n",
6457 err
= set_array_info(mddev
, &info
);
6459 printk(KERN_WARNING
"md: couldn't set"
6460 " array info. %d\n", err
);
6467 * Commands querying/configuring an existing array:
6469 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6470 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6471 if ((!mddev
->raid_disks
&& !mddev
->external
)
6472 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6473 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6474 && cmd
!= GET_BITMAP_FILE
) {
6480 * Commands even a read-only array can execute:
6483 case GET_BITMAP_FILE
:
6484 err
= get_bitmap_file(mddev
, argp
);
6487 case RESTART_ARRAY_RW
:
6488 err
= restart_array(mddev
);
6492 err
= do_md_stop(mddev
, 0, bdev
);
6496 err
= md_set_readonly(mddev
, bdev
);
6499 case HOT_REMOVE_DISK
:
6500 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6504 /* We can support ADD_NEW_DISK on read-only arrays
6505 * on if we are re-adding a preexisting device.
6506 * So require mddev->pers and MD_DISK_SYNC.
6509 mdu_disk_info_t info
;
6510 if (copy_from_user(&info
, argp
, sizeof(info
)))
6512 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6513 /* Need to clear read-only for this */
6516 err
= add_new_disk(mddev
, &info
);
6522 if (get_user(ro
, (int __user
*)(arg
))) {
6528 /* if the bdev is going readonly the value of mddev->ro
6529 * does not matter, no writes are coming
6534 /* are we are already prepared for writes? */
6538 /* transitioning to readauto need only happen for
6539 * arrays that call md_write_start
6542 err
= restart_array(mddev
);
6545 set_disk_ro(mddev
->gendisk
, 0);
6552 * The remaining ioctls are changing the state of the
6553 * superblock, so we do not allow them on read-only arrays.
6554 * However non-MD ioctls (e.g. get-size) will still come through
6555 * here and hit the 'default' below, so only disallow
6556 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6558 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6559 if (mddev
->ro
== 2) {
6561 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6562 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6563 /* mddev_unlock will wake thread */
6564 /* If a device failed while we were read-only, we
6565 * need to make sure the metadata is updated now.
6567 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6568 mddev_unlock(mddev
);
6569 wait_event(mddev
->sb_wait
,
6570 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6571 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6583 mdu_disk_info_t info
;
6584 if (copy_from_user(&info
, argp
, sizeof(info
)))
6587 err
= add_new_disk(mddev
, &info
);
6592 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6596 err
= do_md_run(mddev
);
6599 case SET_BITMAP_FILE
:
6600 err
= set_bitmap_file(mddev
, (int)arg
);
6610 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6612 mddev
->hold_active
= 0;
6613 mddev_unlock(mddev
);
6622 #ifdef CONFIG_COMPAT
6623 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6624 unsigned int cmd
, unsigned long arg
)
6627 case HOT_REMOVE_DISK
:
6629 case SET_DISK_FAULTY
:
6630 case SET_BITMAP_FILE
:
6631 /* These take in integer arg, do not convert */
6634 arg
= (unsigned long)compat_ptr(arg
);
6638 return md_ioctl(bdev
, mode
, cmd
, arg
);
6640 #endif /* CONFIG_COMPAT */
6642 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6645 * Succeed if we can lock the mddev, which confirms that
6646 * it isn't being stopped right now.
6648 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6654 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6655 /* we are racing with mddev_put which is discarding this
6659 /* Wait until bdev->bd_disk is definitely gone */
6660 flush_workqueue(md_misc_wq
);
6661 /* Then retry the open from the top */
6662 return -ERESTARTSYS
;
6664 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6666 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6670 atomic_inc(&mddev
->openers
);
6671 mutex_unlock(&mddev
->open_mutex
);
6673 check_disk_change(bdev
);
6678 static void md_release(struct gendisk
*disk
, fmode_t mode
)
6680 struct mddev
*mddev
= disk
->private_data
;
6683 atomic_dec(&mddev
->openers
);
6687 static int md_media_changed(struct gendisk
*disk
)
6689 struct mddev
*mddev
= disk
->private_data
;
6691 return mddev
->changed
;
6694 static int md_revalidate(struct gendisk
*disk
)
6696 struct mddev
*mddev
= disk
->private_data
;
6701 static const struct block_device_operations md_fops
=
6703 .owner
= THIS_MODULE
,
6705 .release
= md_release
,
6707 #ifdef CONFIG_COMPAT
6708 .compat_ioctl
= md_compat_ioctl
,
6710 .getgeo
= md_getgeo
,
6711 .media_changed
= md_media_changed
,
6712 .revalidate_disk
= md_revalidate
,
6715 static int md_thread(void * arg
)
6717 struct md_thread
*thread
= arg
;
6720 * md_thread is a 'system-thread', it's priority should be very
6721 * high. We avoid resource deadlocks individually in each
6722 * raid personality. (RAID5 does preallocation) We also use RR and
6723 * the very same RT priority as kswapd, thus we will never get
6724 * into a priority inversion deadlock.
6726 * we definitely have to have equal or higher priority than
6727 * bdflush, otherwise bdflush will deadlock if there are too
6728 * many dirty RAID5 blocks.
6731 allow_signal(SIGKILL
);
6732 while (!kthread_should_stop()) {
6734 /* We need to wait INTERRUPTIBLE so that
6735 * we don't add to the load-average.
6736 * That means we need to be sure no signals are
6739 if (signal_pending(current
))
6740 flush_signals(current
);
6742 wait_event_interruptible_timeout
6744 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6745 || kthread_should_stop(),
6748 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6749 if (!kthread_should_stop())
6750 thread
->run(thread
);
6756 void md_wakeup_thread(struct md_thread
*thread
)
6759 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6760 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6761 wake_up(&thread
->wqueue
);
6765 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
6766 struct mddev
*mddev
, const char *name
)
6768 struct md_thread
*thread
;
6770 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6774 init_waitqueue_head(&thread
->wqueue
);
6777 thread
->mddev
= mddev
;
6778 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6779 thread
->tsk
= kthread_run(md_thread
, thread
,
6781 mdname(thread
->mddev
),
6783 if (IS_ERR(thread
->tsk
)) {
6790 void md_unregister_thread(struct md_thread
**threadp
)
6792 struct md_thread
*thread
= *threadp
;
6795 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6796 /* Locking ensures that mddev_unlock does not wake_up a
6797 * non-existent thread
6799 spin_lock(&pers_lock
);
6801 spin_unlock(&pers_lock
);
6803 kthread_stop(thread
->tsk
);
6807 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6814 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6817 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6819 mddev
->pers
->error_handler(mddev
,rdev
);
6820 if (mddev
->degraded
)
6821 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6822 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6823 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6824 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6825 md_wakeup_thread(mddev
->thread
);
6826 if (mddev
->event_work
.func
)
6827 queue_work(md_misc_wq
, &mddev
->event_work
);
6828 md_new_event_inintr(mddev
);
6831 /* seq_file implementation /proc/mdstat */
6833 static void status_unused(struct seq_file
*seq
)
6836 struct md_rdev
*rdev
;
6838 seq_printf(seq
, "unused devices: ");
6840 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6841 char b
[BDEVNAME_SIZE
];
6843 seq_printf(seq
, "%s ",
6844 bdevname(rdev
->bdev
,b
));
6847 seq_printf(seq
, "<none>");
6849 seq_printf(seq
, "\n");
6853 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6855 sector_t max_sectors
, resync
, res
;
6856 unsigned long dt
, db
;
6859 unsigned int per_milli
;
6861 if (mddev
->curr_resync
<= 3)
6864 resync
= mddev
->curr_resync
6865 - atomic_read(&mddev
->recovery_active
);
6867 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6868 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6869 max_sectors
= mddev
->resync_max_sectors
;
6871 max_sectors
= mddev
->dev_sectors
;
6874 * Should not happen.
6880 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6881 * in a sector_t, and (max_sectors>>scale) will fit in a
6882 * u32, as those are the requirements for sector_div.
6883 * Thus 'scale' must be at least 10
6886 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6887 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6890 res
= (resync
>>scale
)*1000;
6891 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6895 int i
, x
= per_milli
/50, y
= 20-x
;
6896 seq_printf(seq
, "[");
6897 for (i
= 0; i
< x
; i
++)
6898 seq_printf(seq
, "=");
6899 seq_printf(seq
, ">");
6900 for (i
= 0; i
< y
; i
++)
6901 seq_printf(seq
, ".");
6902 seq_printf(seq
, "] ");
6904 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6905 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6907 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6909 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6910 "resync" : "recovery"))),
6911 per_milli
/10, per_milli
% 10,
6912 (unsigned long long) resync
/2,
6913 (unsigned long long) max_sectors
/2);
6916 * dt: time from mark until now
6917 * db: blocks written from mark until now
6918 * rt: remaining time
6920 * rt is a sector_t, so could be 32bit or 64bit.
6921 * So we divide before multiply in case it is 32bit and close
6923 * We scale the divisor (db) by 32 to avoid losing precision
6924 * near the end of resync when the number of remaining sectors
6926 * We then divide rt by 32 after multiplying by db to compensate.
6927 * The '+1' avoids division by zero if db is very small.
6929 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6931 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6932 - mddev
->resync_mark_cnt
;
6934 rt
= max_sectors
- resync
; /* number of remaining sectors */
6935 sector_div(rt
, db
/32+1);
6939 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6940 ((unsigned long)rt
% 60)/6);
6942 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6945 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6947 struct list_head
*tmp
;
6949 struct mddev
*mddev
;
6957 spin_lock(&all_mddevs_lock
);
6958 list_for_each(tmp
,&all_mddevs
)
6960 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6962 spin_unlock(&all_mddevs_lock
);
6965 spin_unlock(&all_mddevs_lock
);
6967 return (void*)2;/* tail */
6971 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6973 struct list_head
*tmp
;
6974 struct mddev
*next_mddev
, *mddev
= v
;
6980 spin_lock(&all_mddevs_lock
);
6982 tmp
= all_mddevs
.next
;
6984 tmp
= mddev
->all_mddevs
.next
;
6985 if (tmp
!= &all_mddevs
)
6986 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6988 next_mddev
= (void*)2;
6991 spin_unlock(&all_mddevs_lock
);
6999 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7001 struct mddev
*mddev
= v
;
7003 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7007 static int md_seq_show(struct seq_file
*seq
, void *v
)
7009 struct mddev
*mddev
= v
;
7011 struct md_rdev
*rdev
;
7013 if (v
== (void*)1) {
7014 struct md_personality
*pers
;
7015 seq_printf(seq
, "Personalities : ");
7016 spin_lock(&pers_lock
);
7017 list_for_each_entry(pers
, &pers_list
, list
)
7018 seq_printf(seq
, "[%s] ", pers
->name
);
7020 spin_unlock(&pers_lock
);
7021 seq_printf(seq
, "\n");
7022 seq
->poll_event
= atomic_read(&md_event_count
);
7025 if (v
== (void*)2) {
7030 if (mddev_lock(mddev
) < 0)
7033 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7034 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7035 mddev
->pers
? "" : "in");
7038 seq_printf(seq
, " (read-only)");
7040 seq_printf(seq
, " (auto-read-only)");
7041 seq_printf(seq
, " %s", mddev
->pers
->name
);
7045 rdev_for_each(rdev
, mddev
) {
7046 char b
[BDEVNAME_SIZE
];
7047 seq_printf(seq
, " %s[%d]",
7048 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7049 if (test_bit(WriteMostly
, &rdev
->flags
))
7050 seq_printf(seq
, "(W)");
7051 if (test_bit(Faulty
, &rdev
->flags
)) {
7052 seq_printf(seq
, "(F)");
7055 if (rdev
->raid_disk
< 0)
7056 seq_printf(seq
, "(S)"); /* spare */
7057 if (test_bit(Replacement
, &rdev
->flags
))
7058 seq_printf(seq
, "(R)");
7059 sectors
+= rdev
->sectors
;
7062 if (!list_empty(&mddev
->disks
)) {
7064 seq_printf(seq
, "\n %llu blocks",
7065 (unsigned long long)
7066 mddev
->array_sectors
/ 2);
7068 seq_printf(seq
, "\n %llu blocks",
7069 (unsigned long long)sectors
/ 2);
7071 if (mddev
->persistent
) {
7072 if (mddev
->major_version
!= 0 ||
7073 mddev
->minor_version
!= 90) {
7074 seq_printf(seq
," super %d.%d",
7075 mddev
->major_version
,
7076 mddev
->minor_version
);
7078 } else if (mddev
->external
)
7079 seq_printf(seq
, " super external:%s",
7080 mddev
->metadata_type
);
7082 seq_printf(seq
, " super non-persistent");
7085 mddev
->pers
->status(seq
, mddev
);
7086 seq_printf(seq
, "\n ");
7087 if (mddev
->pers
->sync_request
) {
7088 if (mddev
->curr_resync
> 2) {
7089 status_resync(seq
, mddev
);
7090 seq_printf(seq
, "\n ");
7091 } else if (mddev
->curr_resync
>= 1)
7092 seq_printf(seq
, "\tresync=DELAYED\n ");
7093 else if (mddev
->recovery_cp
< MaxSector
)
7094 seq_printf(seq
, "\tresync=PENDING\n ");
7097 seq_printf(seq
, "\n ");
7099 bitmap_status(seq
, mddev
->bitmap
);
7101 seq_printf(seq
, "\n");
7103 mddev_unlock(mddev
);
7108 static const struct seq_operations md_seq_ops
= {
7109 .start
= md_seq_start
,
7110 .next
= md_seq_next
,
7111 .stop
= md_seq_stop
,
7112 .show
= md_seq_show
,
7115 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7117 struct seq_file
*seq
;
7120 error
= seq_open(file
, &md_seq_ops
);
7124 seq
= file
->private_data
;
7125 seq
->poll_event
= atomic_read(&md_event_count
);
7129 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7131 struct seq_file
*seq
= filp
->private_data
;
7134 poll_wait(filp
, &md_event_waiters
, wait
);
7136 /* always allow read */
7137 mask
= POLLIN
| POLLRDNORM
;
7139 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7140 mask
|= POLLERR
| POLLPRI
;
7144 static const struct file_operations md_seq_fops
= {
7145 .owner
= THIS_MODULE
,
7146 .open
= md_seq_open
,
7148 .llseek
= seq_lseek
,
7149 .release
= seq_release_private
,
7150 .poll
= mdstat_poll
,
7153 int register_md_personality(struct md_personality
*p
)
7155 spin_lock(&pers_lock
);
7156 list_add_tail(&p
->list
, &pers_list
);
7157 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
7158 spin_unlock(&pers_lock
);
7162 int unregister_md_personality(struct md_personality
*p
)
7164 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7165 spin_lock(&pers_lock
);
7166 list_del_init(&p
->list
);
7167 spin_unlock(&pers_lock
);
7171 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7173 struct md_rdev
* rdev
;
7179 rdev_for_each_rcu(rdev
, mddev
) {
7180 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7181 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7182 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7183 atomic_read(&disk
->sync_io
);
7184 /* sync IO will cause sync_io to increase before the disk_stats
7185 * as sync_io is counted when a request starts, and
7186 * disk_stats is counted when it completes.
7187 * So resync activity will cause curr_events to be smaller than
7188 * when there was no such activity.
7189 * non-sync IO will cause disk_stat to increase without
7190 * increasing sync_io so curr_events will (eventually)
7191 * be larger than it was before. Once it becomes
7192 * substantially larger, the test below will cause
7193 * the array to appear non-idle, and resync will slow
7195 * If there is a lot of outstanding resync activity when
7196 * we set last_event to curr_events, then all that activity
7197 * completing might cause the array to appear non-idle
7198 * and resync will be slowed down even though there might
7199 * not have been non-resync activity. This will only
7200 * happen once though. 'last_events' will soon reflect
7201 * the state where there is little or no outstanding
7202 * resync requests, and further resync activity will
7203 * always make curr_events less than last_events.
7206 if (init
|| curr_events
- rdev
->last_events
> 64) {
7207 rdev
->last_events
= curr_events
;
7215 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7217 /* another "blocks" (512byte) blocks have been synced */
7218 atomic_sub(blocks
, &mddev
->recovery_active
);
7219 wake_up(&mddev
->recovery_wait
);
7221 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7222 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7223 md_wakeup_thread(mddev
->thread
);
7224 // stop recovery, signal do_sync ....
7229 /* md_write_start(mddev, bi)
7230 * If we need to update some array metadata (e.g. 'active' flag
7231 * in superblock) before writing, schedule a superblock update
7232 * and wait for it to complete.
7234 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7237 if (bio_data_dir(bi
) != WRITE
)
7240 BUG_ON(mddev
->ro
== 1);
7241 if (mddev
->ro
== 2) {
7242 /* need to switch to read/write */
7244 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7245 md_wakeup_thread(mddev
->thread
);
7246 md_wakeup_thread(mddev
->sync_thread
);
7249 atomic_inc(&mddev
->writes_pending
);
7250 if (mddev
->safemode
== 1)
7251 mddev
->safemode
= 0;
7252 if (mddev
->in_sync
) {
7253 spin_lock_irq(&mddev
->write_lock
);
7254 if (mddev
->in_sync
) {
7256 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7257 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7258 md_wakeup_thread(mddev
->thread
);
7261 spin_unlock_irq(&mddev
->write_lock
);
7264 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7265 wait_event(mddev
->sb_wait
,
7266 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7269 void md_write_end(struct mddev
*mddev
)
7271 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7272 if (mddev
->safemode
== 2)
7273 md_wakeup_thread(mddev
->thread
);
7274 else if (mddev
->safemode_delay
)
7275 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7279 /* md_allow_write(mddev)
7280 * Calling this ensures that the array is marked 'active' so that writes
7281 * may proceed without blocking. It is important to call this before
7282 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7283 * Must be called with mddev_lock held.
7285 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7286 * is dropped, so return -EAGAIN after notifying userspace.
7288 int md_allow_write(struct mddev
*mddev
)
7294 if (!mddev
->pers
->sync_request
)
7297 spin_lock_irq(&mddev
->write_lock
);
7298 if (mddev
->in_sync
) {
7300 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7301 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7302 if (mddev
->safemode_delay
&&
7303 mddev
->safemode
== 0)
7304 mddev
->safemode
= 1;
7305 spin_unlock_irq(&mddev
->write_lock
);
7306 md_update_sb(mddev
, 0);
7307 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7309 spin_unlock_irq(&mddev
->write_lock
);
7311 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7316 EXPORT_SYMBOL_GPL(md_allow_write
);
7318 #define SYNC_MARKS 10
7319 #define SYNC_MARK_STEP (3*HZ)
7320 #define UPDATE_FREQUENCY (5*60*HZ)
7321 void md_do_sync(struct md_thread
*thread
)
7323 struct mddev
*mddev
= thread
->mddev
;
7324 struct mddev
*mddev2
;
7325 unsigned int currspeed
= 0,
7327 sector_t max_sectors
,j
, io_sectors
;
7328 unsigned long mark
[SYNC_MARKS
];
7329 unsigned long update_time
;
7330 sector_t mark_cnt
[SYNC_MARKS
];
7332 struct list_head
*tmp
;
7333 sector_t last_check
;
7335 struct md_rdev
*rdev
;
7337 struct blk_plug plug
;
7339 /* just incase thread restarts... */
7340 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7342 if (mddev
->ro
) {/* never try to sync a read-only array */
7343 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7347 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7348 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7349 desc
= "data-check";
7350 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7351 desc
= "requested-resync";
7354 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7359 /* we overload curr_resync somewhat here.
7360 * 0 == not engaged in resync at all
7361 * 2 == checking that there is no conflict with another sync
7362 * 1 == like 2, but have yielded to allow conflicting resync to
7364 * other == active in resync - this many blocks
7366 * Before starting a resync we must have set curr_resync to
7367 * 2, and then checked that every "conflicting" array has curr_resync
7368 * less than ours. When we find one that is the same or higher
7369 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7370 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7371 * This will mean we have to start checking from the beginning again.
7378 mddev
->curr_resync
= 2;
7382 if (kthread_freezable_should_stop(NULL
))
7383 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7385 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7387 for_each_mddev(mddev2
, tmp
) {
7388 if (mddev2
== mddev
)
7390 if (!mddev
->parallel_resync
7391 && mddev2
->curr_resync
7392 && match_mddev_units(mddev
, mddev2
)) {
7394 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7395 /* arbitrarily yield */
7396 mddev
->curr_resync
= 1;
7397 wake_up(&resync_wait
);
7399 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7400 /* no need to wait here, we can wait the next
7401 * time 'round when curr_resync == 2
7407 /* We need to wait 'interruptible' so as not to
7408 * contribute to the load average, and not to
7409 * be caught by 'softlockup'
7411 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7412 if (!kthread_should_stop() &&
7413 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7414 printk(KERN_INFO
"md: delaying %s of %s"
7415 " until %s has finished (they"
7416 " share one or more physical units)\n",
7417 desc
, mdname(mddev
), mdname(mddev2
));
7420 if (signal_pending(current
))
7421 flush_signals(current
);
7423 finish_wait(&resync_wait
, &wq
);
7426 finish_wait(&resync_wait
, &wq
);
7429 } while (mddev
->curr_resync
< 2);
7432 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7433 /* resync follows the size requested by the personality,
7434 * which defaults to physical size, but can be virtual size
7436 max_sectors
= mddev
->resync_max_sectors
;
7437 atomic64_set(&mddev
->resync_mismatches
, 0);
7438 /* we don't use the checkpoint if there's a bitmap */
7439 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7440 j
= mddev
->resync_min
;
7441 else if (!mddev
->bitmap
)
7442 j
= mddev
->recovery_cp
;
7444 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7445 max_sectors
= mddev
->resync_max_sectors
;
7447 /* recovery follows the physical size of devices */
7448 max_sectors
= mddev
->dev_sectors
;
7451 rdev_for_each_rcu(rdev
, mddev
)
7452 if (rdev
->raid_disk
>= 0 &&
7453 !test_bit(Faulty
, &rdev
->flags
) &&
7454 !test_bit(In_sync
, &rdev
->flags
) &&
7455 rdev
->recovery_offset
< j
)
7456 j
= rdev
->recovery_offset
;
7459 /* If there is a bitmap, we need to make sure all
7460 * writes that started before we added a spare
7461 * complete before we start doing a recovery.
7462 * Otherwise the write might complete and (via
7463 * bitmap_endwrite) set a bit in the bitmap after the
7464 * recovery has checked that bit and skipped that
7467 if (mddev
->bitmap
) {
7468 mddev
->pers
->quiesce(mddev
, 1);
7469 mddev
->pers
->quiesce(mddev
, 0);
7473 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7474 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7475 " %d KB/sec/disk.\n", speed_min(mddev
));
7476 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7477 "(but not more than %d KB/sec) for %s.\n",
7478 speed_max(mddev
), desc
);
7480 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7483 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7485 mark_cnt
[m
] = io_sectors
;
7488 mddev
->resync_mark
= mark
[last_mark
];
7489 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7492 * Tune reconstruction:
7494 window
= 32*(PAGE_SIZE
/512);
7495 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7496 window
/2, (unsigned long long)max_sectors
/2);
7498 atomic_set(&mddev
->recovery_active
, 0);
7503 "md: resuming %s of %s from checkpoint.\n",
7504 desc
, mdname(mddev
));
7505 mddev
->curr_resync
= j
;
7507 mddev
->curr_resync
= 3; /* no longer delayed */
7508 mddev
->curr_resync_completed
= j
;
7509 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7510 md_new_event(mddev
);
7511 update_time
= jiffies
;
7513 blk_start_plug(&plug
);
7514 while (j
< max_sectors
) {
7519 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7520 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7521 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7522 > (max_sectors
>> 4)) ||
7523 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7524 (j
- mddev
->curr_resync_completed
)*2
7525 >= mddev
->resync_max
- mddev
->curr_resync_completed
7527 /* time to update curr_resync_completed */
7528 wait_event(mddev
->recovery_wait
,
7529 atomic_read(&mddev
->recovery_active
) == 0);
7530 mddev
->curr_resync_completed
= j
;
7531 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
7532 j
> mddev
->recovery_cp
)
7533 mddev
->recovery_cp
= j
;
7534 update_time
= jiffies
;
7535 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7536 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7539 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7540 /* As this condition is controlled by user-space,
7541 * we can block indefinitely, so use '_interruptible'
7542 * to avoid triggering warnings.
7544 flush_signals(current
); /* just in case */
7545 wait_event_interruptible(mddev
->recovery_wait
,
7546 mddev
->resync_max
> j
7547 || kthread_should_stop());
7550 if (kthread_freezable_should_stop(NULL
))
7553 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7554 currspeed
< speed_min(mddev
));
7556 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7560 if (!skipped
) { /* actual IO requested */
7561 io_sectors
+= sectors
;
7562 atomic_add(sectors
, &mddev
->recovery_active
);
7565 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7570 mddev
->curr_resync
= j
;
7571 mddev
->curr_mark_cnt
= io_sectors
;
7572 if (last_check
== 0)
7573 /* this is the earliest that rebuild will be
7574 * visible in /proc/mdstat
7576 md_new_event(mddev
);
7578 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7581 last_check
= io_sectors
;
7583 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7585 int next
= (last_mark
+1) % SYNC_MARKS
;
7587 mddev
->resync_mark
= mark
[next
];
7588 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7589 mark
[next
] = jiffies
;
7590 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7594 if (kthread_freezable_should_stop(NULL
))
7599 * this loop exits only if either when we are slower than
7600 * the 'hard' speed limit, or the system was IO-idle for
7602 * the system might be non-idle CPU-wise, but we only care
7603 * about not overloading the IO subsystem. (things like an
7604 * e2fsck being done on the RAID array should execute fast)
7608 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7609 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7611 if (currspeed
> speed_min(mddev
)) {
7612 if ((currspeed
> speed_max(mddev
)) ||
7613 !is_mddev_idle(mddev
, 0)) {
7619 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7621 * this also signals 'finished resyncing' to md_stop
7624 blk_finish_plug(&plug
);
7625 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7627 /* tell personality that we are finished */
7628 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7630 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7631 mddev
->curr_resync
> 2) {
7632 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7633 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7634 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7636 "md: checkpointing %s of %s.\n",
7637 desc
, mdname(mddev
));
7638 if (test_bit(MD_RECOVERY_ERROR
,
7640 mddev
->recovery_cp
=
7641 mddev
->curr_resync_completed
;
7643 mddev
->recovery_cp
=
7647 mddev
->recovery_cp
= MaxSector
;
7649 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7650 mddev
->curr_resync
= MaxSector
;
7652 rdev_for_each_rcu(rdev
, mddev
)
7653 if (rdev
->raid_disk
>= 0 &&
7654 mddev
->delta_disks
>= 0 &&
7655 !test_bit(Faulty
, &rdev
->flags
) &&
7656 !test_bit(In_sync
, &rdev
->flags
) &&
7657 rdev
->recovery_offset
< mddev
->curr_resync
)
7658 rdev
->recovery_offset
= mddev
->curr_resync
;
7663 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7665 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7666 /* We completed so min/max setting can be forgotten if used. */
7667 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7668 mddev
->resync_min
= 0;
7669 mddev
->resync_max
= MaxSector
;
7670 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7671 mddev
->resync_min
= mddev
->curr_resync_completed
;
7672 mddev
->curr_resync
= 0;
7673 wake_up(&resync_wait
);
7674 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7675 md_wakeup_thread(mddev
->thread
);
7680 * got a signal, exit.
7683 "md: md_do_sync() got signal ... exiting\n");
7684 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7688 EXPORT_SYMBOL_GPL(md_do_sync
);
7690 static int remove_and_add_spares(struct mddev
*mddev
,
7691 struct md_rdev
*this)
7693 struct md_rdev
*rdev
;
7697 rdev_for_each(rdev
, mddev
)
7698 if ((this == NULL
|| rdev
== this) &&
7699 rdev
->raid_disk
>= 0 &&
7700 !test_bit(Blocked
, &rdev
->flags
) &&
7701 (test_bit(Faulty
, &rdev
->flags
) ||
7702 ! test_bit(In_sync
, &rdev
->flags
)) &&
7703 atomic_read(&rdev
->nr_pending
)==0) {
7704 if (mddev
->pers
->hot_remove_disk(
7705 mddev
, rdev
) == 0) {
7706 sysfs_unlink_rdev(mddev
, rdev
);
7707 rdev
->raid_disk
= -1;
7711 if (removed
&& mddev
->kobj
.sd
)
7712 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
7717 rdev_for_each(rdev
, mddev
) {
7718 if (rdev
->raid_disk
>= 0 &&
7719 !test_bit(In_sync
, &rdev
->flags
) &&
7720 !test_bit(Faulty
, &rdev
->flags
))
7722 if (rdev
->raid_disk
>= 0)
7724 if (test_bit(Faulty
, &rdev
->flags
))
7727 ! (rdev
->saved_raid_disk
>= 0 &&
7728 !test_bit(Bitmap_sync
, &rdev
->flags
)))
7731 rdev
->recovery_offset
= 0;
7733 hot_add_disk(mddev
, rdev
) == 0) {
7734 if (sysfs_link_rdev(mddev
, rdev
))
7735 /* failure here is OK */;
7737 md_new_event(mddev
);
7738 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7743 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7748 * This routine is regularly called by all per-raid-array threads to
7749 * deal with generic issues like resync and super-block update.
7750 * Raid personalities that don't have a thread (linear/raid0) do not
7751 * need this as they never do any recovery or update the superblock.
7753 * It does not do any resync itself, but rather "forks" off other threads
7754 * to do that as needed.
7755 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7756 * "->recovery" and create a thread at ->sync_thread.
7757 * When the thread finishes it sets MD_RECOVERY_DONE
7758 * and wakeups up this thread which will reap the thread and finish up.
7759 * This thread also removes any faulty devices (with nr_pending == 0).
7761 * The overall approach is:
7762 * 1/ if the superblock needs updating, update it.
7763 * 2/ If a recovery thread is running, don't do anything else.
7764 * 3/ If recovery has finished, clean up, possibly marking spares active.
7765 * 4/ If there are any faulty devices, remove them.
7766 * 5/ If array is degraded, try to add spares devices
7767 * 6/ If array has spares or is not in-sync, start a resync thread.
7769 void md_check_recovery(struct mddev
*mddev
)
7771 #ifdef CONFIG_FREEZER
7772 if (mddev
->suspended
|| unlikely(atomic_read(&system_freezing_cnt
)))
7777 bitmap_daemon_work(mddev
);
7779 if (signal_pending(current
)) {
7780 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7781 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7783 mddev
->safemode
= 2;
7785 flush_signals(current
);
7788 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7791 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7792 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7793 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7794 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7795 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7796 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7800 if (mddev_trylock(mddev
)) {
7804 /* On a read-only array we can:
7805 * - remove failed devices
7806 * - add already-in_sync devices if the array itself
7808 * As we only add devices that are already in-sync,
7809 * we can activate the spares immediately.
7811 remove_and_add_spares(mddev
, NULL
);
7812 /* There is no thread, but we need to call
7813 * ->spare_active and clear saved_raid_disk
7815 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7816 md_reap_sync_thread(mddev
);
7817 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7821 if (!mddev
->external
) {
7823 spin_lock_irq(&mddev
->write_lock
);
7824 if (mddev
->safemode
&&
7825 !atomic_read(&mddev
->writes_pending
) &&
7827 mddev
->recovery_cp
== MaxSector
) {
7830 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7832 if (mddev
->safemode
== 1)
7833 mddev
->safemode
= 0;
7834 spin_unlock_irq(&mddev
->write_lock
);
7836 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7840 md_update_sb(mddev
, 0);
7842 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7843 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7844 /* resync/recovery still happening */
7845 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7848 if (mddev
->sync_thread
) {
7849 md_reap_sync_thread(mddev
);
7852 /* Set RUNNING before clearing NEEDED to avoid
7853 * any transients in the value of "sync_action".
7855 mddev
->curr_resync_completed
= 0;
7856 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7857 /* Clear some bits that don't mean anything, but
7860 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7861 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7863 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7864 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7866 /* no recovery is running.
7867 * remove any failed drives, then
7868 * add spares if possible.
7869 * Spares are also removed and re-added, to allow
7870 * the personality to fail the re-add.
7873 if (mddev
->reshape_position
!= MaxSector
) {
7874 if (mddev
->pers
->check_reshape
== NULL
||
7875 mddev
->pers
->check_reshape(mddev
) != 0)
7876 /* Cannot proceed */
7878 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7879 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7880 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
7881 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7882 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7883 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7884 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7885 } else if (mddev
->recovery_cp
< MaxSector
) {
7886 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7887 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7888 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7889 /* nothing to be done ... */
7892 if (mddev
->pers
->sync_request
) {
7894 /* We are adding a device or devices to an array
7895 * which has the bitmap stored on all devices.
7896 * So make sure all bitmap pages get written
7898 bitmap_write_all(mddev
->bitmap
);
7900 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7903 if (!mddev
->sync_thread
) {
7904 printk(KERN_ERR
"%s: could not start resync"
7907 /* leave the spares where they are, it shouldn't hurt */
7908 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7909 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7910 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7911 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7912 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7914 md_wakeup_thread(mddev
->sync_thread
);
7915 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7916 md_new_event(mddev
);
7919 if (!mddev
->sync_thread
) {
7920 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7921 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7923 if (mddev
->sysfs_action
)
7924 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7926 mddev_unlock(mddev
);
7930 void md_reap_sync_thread(struct mddev
*mddev
)
7932 struct md_rdev
*rdev
;
7934 /* resync has finished, collect result */
7935 md_unregister_thread(&mddev
->sync_thread
);
7936 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7937 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7939 /* activate any spares */
7940 if (mddev
->pers
->spare_active(mddev
)) {
7941 sysfs_notify(&mddev
->kobj
, NULL
,
7943 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7946 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7947 mddev
->pers
->finish_reshape
)
7948 mddev
->pers
->finish_reshape(mddev
);
7950 /* If array is no-longer degraded, then any saved_raid_disk
7951 * information must be scrapped. Also if any device is now
7952 * In_sync we must scrape the saved_raid_disk for that device
7953 * do the superblock for an incrementally recovered device
7956 rdev_for_each(rdev
, mddev
)
7957 if (!mddev
->degraded
||
7958 test_bit(In_sync
, &rdev
->flags
))
7959 rdev
->saved_raid_disk
= -1;
7961 md_update_sb(mddev
, 1);
7962 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7963 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7964 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7965 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7966 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7967 /* flag recovery needed just to double check */
7968 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7969 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7970 md_new_event(mddev
);
7971 if (mddev
->event_work
.func
)
7972 queue_work(md_misc_wq
, &mddev
->event_work
);
7975 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7977 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7978 wait_event_timeout(rdev
->blocked_wait
,
7979 !test_bit(Blocked
, &rdev
->flags
) &&
7980 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7981 msecs_to_jiffies(5000));
7982 rdev_dec_pending(rdev
, mddev
);
7984 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7986 void md_finish_reshape(struct mddev
*mddev
)
7988 /* called be personality module when reshape completes. */
7989 struct md_rdev
*rdev
;
7991 rdev_for_each(rdev
, mddev
) {
7992 if (rdev
->data_offset
> rdev
->new_data_offset
)
7993 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
7995 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
7996 rdev
->data_offset
= rdev
->new_data_offset
;
7999 EXPORT_SYMBOL(md_finish_reshape
);
8001 /* Bad block management.
8002 * We can record which blocks on each device are 'bad' and so just
8003 * fail those blocks, or that stripe, rather than the whole device.
8004 * Entries in the bad-block table are 64bits wide. This comprises:
8005 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8006 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8007 * A 'shift' can be set so that larger blocks are tracked and
8008 * consequently larger devices can be covered.
8009 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8011 * Locking of the bad-block table uses a seqlock so md_is_badblock
8012 * might need to retry if it is very unlucky.
8013 * We will sometimes want to check for bad blocks in a bi_end_io function,
8014 * so we use the write_seqlock_irq variant.
8016 * When looking for a bad block we specify a range and want to
8017 * know if any block in the range is bad. So we binary-search
8018 * to the last range that starts at-or-before the given endpoint,
8019 * (or "before the sector after the target range")
8020 * then see if it ends after the given start.
8022 * 0 if there are no known bad blocks in the range
8023 * 1 if there are known bad block which are all acknowledged
8024 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8025 * plus the start/length of the first bad section we overlap.
8027 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
8028 sector_t
*first_bad
, int *bad_sectors
)
8034 sector_t target
= s
+ sectors
;
8037 if (bb
->shift
> 0) {
8038 /* round the start down, and the end up */
8040 target
+= (1<<bb
->shift
) - 1;
8041 target
>>= bb
->shift
;
8042 sectors
= target
- s
;
8044 /* 'target' is now the first block after the bad range */
8047 seq
= read_seqbegin(&bb
->lock
);
8052 /* Binary search between lo and hi for 'target'
8053 * i.e. for the last range that starts before 'target'
8055 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8056 * are known not to be the last range before target.
8057 * VARIANT: hi-lo is the number of possible
8058 * ranges, and decreases until it reaches 1
8060 while (hi
- lo
> 1) {
8061 int mid
= (lo
+ hi
) / 2;
8062 sector_t a
= BB_OFFSET(p
[mid
]);
8064 /* This could still be the one, earlier ranges
8068 /* This and later ranges are definitely out. */
8071 /* 'lo' might be the last that started before target, but 'hi' isn't */
8073 /* need to check all range that end after 's' to see if
8074 * any are unacknowledged.
8077 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8078 if (BB_OFFSET(p
[lo
]) < target
) {
8079 /* starts before the end, and finishes after
8080 * the start, so they must overlap
8082 if (rv
!= -1 && BB_ACK(p
[lo
]))
8086 *first_bad
= BB_OFFSET(p
[lo
]);
8087 *bad_sectors
= BB_LEN(p
[lo
]);
8093 if (read_seqretry(&bb
->lock
, seq
))
8098 EXPORT_SYMBOL_GPL(md_is_badblock
);
8101 * Add a range of bad blocks to the table.
8102 * This might extend the table, or might contract it
8103 * if two adjacent ranges can be merged.
8104 * We binary-search to find the 'insertion' point, then
8105 * decide how best to handle it.
8107 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
8113 unsigned long flags
;
8116 /* badblocks are disabled */
8120 /* round the start down, and the end up */
8121 sector_t next
= s
+ sectors
;
8123 next
+= (1<<bb
->shift
) - 1;
8128 write_seqlock_irqsave(&bb
->lock
, flags
);
8133 /* Find the last range that starts at-or-before 's' */
8134 while (hi
- lo
> 1) {
8135 int mid
= (lo
+ hi
) / 2;
8136 sector_t a
= BB_OFFSET(p
[mid
]);
8142 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8146 /* we found a range that might merge with the start
8149 sector_t a
= BB_OFFSET(p
[lo
]);
8150 sector_t e
= a
+ BB_LEN(p
[lo
]);
8151 int ack
= BB_ACK(p
[lo
]);
8153 /* Yes, we can merge with a previous range */
8154 if (s
== a
&& s
+ sectors
>= e
)
8155 /* new range covers old */
8158 ack
= ack
&& acknowledged
;
8160 if (e
< s
+ sectors
)
8162 if (e
- a
<= BB_MAX_LEN
) {
8163 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8166 /* does not all fit in one range,
8167 * make p[lo] maximal
8169 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8170 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8176 if (sectors
&& hi
< bb
->count
) {
8177 /* 'hi' points to the first range that starts after 's'.
8178 * Maybe we can merge with the start of that range */
8179 sector_t a
= BB_OFFSET(p
[hi
]);
8180 sector_t e
= a
+ BB_LEN(p
[hi
]);
8181 int ack
= BB_ACK(p
[hi
]);
8182 if (a
<= s
+ sectors
) {
8183 /* merging is possible */
8184 if (e
<= s
+ sectors
) {
8189 ack
= ack
&& acknowledged
;
8192 if (e
- a
<= BB_MAX_LEN
) {
8193 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8196 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8204 if (sectors
== 0 && hi
< bb
->count
) {
8205 /* we might be able to combine lo and hi */
8206 /* Note: 's' is at the end of 'lo' */
8207 sector_t a
= BB_OFFSET(p
[hi
]);
8208 int lolen
= BB_LEN(p
[lo
]);
8209 int hilen
= BB_LEN(p
[hi
]);
8210 int newlen
= lolen
+ hilen
- (s
- a
);
8211 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8212 /* yes, we can combine them */
8213 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8214 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8215 memmove(p
+ hi
, p
+ hi
+ 1,
8216 (bb
->count
- hi
- 1) * 8);
8221 /* didn't merge (it all).
8222 * Need to add a range just before 'hi' */
8223 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8224 /* No room for more */
8228 int this_sectors
= sectors
;
8229 memmove(p
+ hi
+ 1, p
+ hi
,
8230 (bb
->count
- hi
) * 8);
8233 if (this_sectors
> BB_MAX_LEN
)
8234 this_sectors
= BB_MAX_LEN
;
8235 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8236 sectors
-= this_sectors
;
8243 bb
->unacked_exist
= 1;
8244 write_sequnlock_irqrestore(&bb
->lock
, flags
);
8249 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8254 s
+= rdev
->new_data_offset
;
8256 s
+= rdev
->data_offset
;
8257 rv
= md_set_badblocks(&rdev
->badblocks
,
8260 /* Make sure they get written out promptly */
8261 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8262 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8263 md_wakeup_thread(rdev
->mddev
->thread
);
8267 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8270 * Remove a range of bad blocks from the table.
8271 * This may involve extending the table if we spilt a region,
8272 * but it must not fail. So if the table becomes full, we just
8273 * drop the remove request.
8275 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8279 sector_t target
= s
+ sectors
;
8282 if (bb
->shift
> 0) {
8283 /* When clearing we round the start up and the end down.
8284 * This should not matter as the shift should align with
8285 * the block size and no rounding should ever be needed.
8286 * However it is better the think a block is bad when it
8287 * isn't than to think a block is not bad when it is.
8289 s
+= (1<<bb
->shift
) - 1;
8291 target
>>= bb
->shift
;
8292 sectors
= target
- s
;
8295 write_seqlock_irq(&bb
->lock
);
8300 /* Find the last range that starts before 'target' */
8301 while (hi
- lo
> 1) {
8302 int mid
= (lo
+ hi
) / 2;
8303 sector_t a
= BB_OFFSET(p
[mid
]);
8310 /* p[lo] is the last range that could overlap the
8311 * current range. Earlier ranges could also overlap,
8312 * but only this one can overlap the end of the range.
8314 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8315 /* Partial overlap, leave the tail of this range */
8316 int ack
= BB_ACK(p
[lo
]);
8317 sector_t a
= BB_OFFSET(p
[lo
]);
8318 sector_t end
= a
+ BB_LEN(p
[lo
]);
8321 /* we need to split this range */
8322 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8326 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8328 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8331 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8332 /* there is no longer an overlap */
8337 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8338 /* This range does overlap */
8339 if (BB_OFFSET(p
[lo
]) < s
) {
8340 /* Keep the early parts of this range. */
8341 int ack
= BB_ACK(p
[lo
]);
8342 sector_t start
= BB_OFFSET(p
[lo
]);
8343 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8344 /* now low doesn't overlap, so.. */
8349 /* 'lo' is strictly before, 'hi' is strictly after,
8350 * anything between needs to be discarded
8353 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8354 bb
->count
-= (hi
- lo
- 1);
8360 write_sequnlock_irq(&bb
->lock
);
8364 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8368 s
+= rdev
->new_data_offset
;
8370 s
+= rdev
->data_offset
;
8371 return md_clear_badblocks(&rdev
->badblocks
,
8374 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8377 * Acknowledge all bad blocks in a list.
8378 * This only succeeds if ->changed is clear. It is used by
8379 * in-kernel metadata updates
8381 void md_ack_all_badblocks(struct badblocks
*bb
)
8383 if (bb
->page
== NULL
|| bb
->changed
)
8384 /* no point even trying */
8386 write_seqlock_irq(&bb
->lock
);
8388 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8391 for (i
= 0; i
< bb
->count
; i
++) {
8392 if (!BB_ACK(p
[i
])) {
8393 sector_t start
= BB_OFFSET(p
[i
]);
8394 int len
= BB_LEN(p
[i
]);
8395 p
[i
] = BB_MAKE(start
, len
, 1);
8398 bb
->unacked_exist
= 0;
8400 write_sequnlock_irq(&bb
->lock
);
8402 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8404 /* sysfs access to bad-blocks list.
8405 * We present two files.
8406 * 'bad-blocks' lists sector numbers and lengths of ranges that
8407 * are recorded as bad. The list is truncated to fit within
8408 * the one-page limit of sysfs.
8409 * Writing "sector length" to this file adds an acknowledged
8411 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8412 * been acknowledged. Writing to this file adds bad blocks
8413 * without acknowledging them. This is largely for testing.
8417 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8428 seq
= read_seqbegin(&bb
->lock
);
8433 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8434 sector_t s
= BB_OFFSET(p
[i
]);
8435 unsigned int length
= BB_LEN(p
[i
]);
8436 int ack
= BB_ACK(p
[i
]);
8442 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8443 (unsigned long long)s
<< bb
->shift
,
8444 length
<< bb
->shift
);
8446 if (unack
&& len
== 0)
8447 bb
->unacked_exist
= 0;
8449 if (read_seqretry(&bb
->lock
, seq
))
8458 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8460 unsigned long long sector
;
8464 /* Allow clearing via sysfs *only* for testing/debugging.
8465 * Normally only a successful write may clear a badblock
8468 if (page
[0] == '-') {
8472 #endif /* DO_DEBUG */
8474 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8476 if (newline
!= '\n')
8488 md_clear_badblocks(bb
, sector
, length
);
8491 #endif /* DO_DEBUG */
8492 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8498 static int md_notify_reboot(struct notifier_block
*this,
8499 unsigned long code
, void *x
)
8501 struct list_head
*tmp
;
8502 struct mddev
*mddev
;
8505 for_each_mddev(mddev
, tmp
) {
8506 if (mddev_trylock(mddev
)) {
8508 __md_stop_writes(mddev
);
8509 if (mddev
->persistent
)
8510 mddev
->safemode
= 2;
8511 mddev_unlock(mddev
);
8516 * certain more exotic SCSI devices are known to be
8517 * volatile wrt too early system reboots. While the
8518 * right place to handle this issue is the given
8519 * driver, we do want to have a safe RAID driver ...
8527 static struct notifier_block md_notifier
= {
8528 .notifier_call
= md_notify_reboot
,
8530 .priority
= INT_MAX
, /* before any real devices */
8533 static void md_geninit(void)
8535 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8537 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8540 static int __init
md_init(void)
8544 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8548 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8552 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8555 if ((ret
= register_blkdev(0, "mdp")) < 0)
8559 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8560 md_probe
, NULL
, NULL
);
8561 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8562 md_probe
, NULL
, NULL
);
8564 register_reboot_notifier(&md_notifier
);
8565 raid_table_header
= register_sysctl_table(raid_root_table
);
8571 unregister_blkdev(MD_MAJOR
, "md");
8573 destroy_workqueue(md_misc_wq
);
8575 destroy_workqueue(md_wq
);
8583 * Searches all registered partitions for autorun RAID arrays
8587 static LIST_HEAD(all_detected_devices
);
8588 struct detected_devices_node
{
8589 struct list_head list
;
8593 void md_autodetect_dev(dev_t dev
)
8595 struct detected_devices_node
*node_detected_dev
;
8597 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8598 if (node_detected_dev
) {
8599 node_detected_dev
->dev
= dev
;
8600 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8602 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8603 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8608 static void autostart_arrays(int part
)
8610 struct md_rdev
*rdev
;
8611 struct detected_devices_node
*node_detected_dev
;
8613 int i_scanned
, i_passed
;
8618 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8620 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8622 node_detected_dev
= list_entry(all_detected_devices
.next
,
8623 struct detected_devices_node
, list
);
8624 list_del(&node_detected_dev
->list
);
8625 dev
= node_detected_dev
->dev
;
8626 kfree(node_detected_dev
);
8627 rdev
= md_import_device(dev
,0, 90);
8631 if (test_bit(Faulty
, &rdev
->flags
)) {
8635 set_bit(AutoDetected
, &rdev
->flags
);
8636 list_add(&rdev
->same_set
, &pending_raid_disks
);
8640 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8641 i_scanned
, i_passed
);
8643 autorun_devices(part
);
8646 #endif /* !MODULE */
8648 static __exit
void md_exit(void)
8650 struct mddev
*mddev
;
8651 struct list_head
*tmp
;
8653 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8654 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8656 unregister_blkdev(MD_MAJOR
,"md");
8657 unregister_blkdev(mdp_major
, "mdp");
8658 unregister_reboot_notifier(&md_notifier
);
8659 unregister_sysctl_table(raid_table_header
);
8660 remove_proc_entry("mdstat", NULL
);
8661 for_each_mddev(mddev
, tmp
) {
8662 export_array(mddev
);
8663 mddev
->hold_active
= 0;
8665 destroy_workqueue(md_misc_wq
);
8666 destroy_workqueue(md_wq
);
8669 subsys_initcall(md_init
);
8670 module_exit(md_exit
)
8672 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8674 return sprintf(buffer
, "%d", start_readonly
);
8676 static int set_ro(const char *val
, struct kernel_param
*kp
)
8679 int num
= simple_strtoul(val
, &e
, 10);
8680 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8681 start_readonly
= num
;
8687 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8688 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8690 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8692 EXPORT_SYMBOL(register_md_personality
);
8693 EXPORT_SYMBOL(unregister_md_personality
);
8694 EXPORT_SYMBOL(md_error
);
8695 EXPORT_SYMBOL(md_done_sync
);
8696 EXPORT_SYMBOL(md_write_start
);
8697 EXPORT_SYMBOL(md_write_end
);
8698 EXPORT_SYMBOL(md_register_thread
);
8699 EXPORT_SYMBOL(md_unregister_thread
);
8700 EXPORT_SYMBOL(md_wakeup_thread
);
8701 EXPORT_SYMBOL(md_check_recovery
);
8702 EXPORT_SYMBOL(md_reap_sync_thread
);
8703 MODULE_LICENSE("GPL");
8704 MODULE_DESCRIPTION("MD RAID framework");
8706 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);