2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 static int remove_and_add_spares(struct mddev
*mddev
,
76 struct md_rdev
*this);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
90 * the RAID driver will use the maximum available bandwidth if the IO
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96 * or /sys/block/mdX/md/sync_speed_{min,max}
99 static int sysctl_speed_limit_min
= 1000;
100 static int sysctl_speed_limit_max
= 200000;
101 static inline int speed_min(struct mddev
*mddev
)
103 return mddev
->sync_speed_min
?
104 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
107 static inline int speed_max(struct mddev
*mddev
)
109 return mddev
->sync_speed_max
?
110 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
113 static struct ctl_table_header
*raid_table_header
;
115 static struct ctl_table raid_table
[] = {
117 .procname
= "speed_limit_min",
118 .data
= &sysctl_speed_limit_min
,
119 .maxlen
= sizeof(int),
120 .mode
= S_IRUGO
|S_IWUSR
,
121 .proc_handler
= proc_dointvec
,
124 .procname
= "speed_limit_max",
125 .data
= &sysctl_speed_limit_max
,
126 .maxlen
= sizeof(int),
127 .mode
= S_IRUGO
|S_IWUSR
,
128 .proc_handler
= proc_dointvec
,
133 static struct ctl_table raid_dir_table
[] = {
137 .mode
= S_IRUGO
|S_IXUGO
,
143 static struct ctl_table raid_root_table
[] = {
148 .child
= raid_dir_table
,
153 static const struct block_device_operations md_fops
;
155 static int start_readonly
;
158 * like bio_clone, but with a local bio set
161 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
166 if (!mddev
|| !mddev
->bio_set
)
167 return bio_alloc(gfp_mask
, nr_iovecs
);
169 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
174 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
176 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
179 if (!mddev
|| !mddev
->bio_set
)
180 return bio_clone(bio
, gfp_mask
);
182 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
184 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
187 * We have a system wide 'event count' that is incremented
188 * on any 'interesting' event, and readers of /proc/mdstat
189 * can use 'poll' or 'select' to find out when the event
193 * start array, stop array, error, add device, remove device,
194 * start build, activate spare
196 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
197 static atomic_t md_event_count
;
198 void md_new_event(struct mddev
*mddev
)
200 atomic_inc(&md_event_count
);
201 wake_up(&md_event_waiters
);
203 EXPORT_SYMBOL_GPL(md_new_event
);
205 /* Alternate version that can be called from interrupts
206 * when calling sysfs_notify isn't needed.
208 static void md_new_event_inintr(struct mddev
*mddev
)
210 atomic_inc(&md_event_count
);
211 wake_up(&md_event_waiters
);
215 * Enables to iterate over all existing md arrays
216 * all_mddevs_lock protects this list.
218 static LIST_HEAD(all_mddevs
);
219 static DEFINE_SPINLOCK(all_mddevs_lock
);
223 * iterates through all used mddevs in the system.
224 * We take care to grab the all_mddevs_lock whenever navigating
225 * the list, and to always hold a refcount when unlocked.
226 * Any code which breaks out of this loop while own
227 * a reference to the current mddev and must mddev_put it.
229 #define for_each_mddev(_mddev,_tmp) \
231 for (({ spin_lock(&all_mddevs_lock); \
232 _tmp = all_mddevs.next; \
234 ({ if (_tmp != &all_mddevs) \
235 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
236 spin_unlock(&all_mddevs_lock); \
237 if (_mddev) mddev_put(_mddev); \
238 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
239 _tmp != &all_mddevs;}); \
240 ({ spin_lock(&all_mddevs_lock); \
241 _tmp = _tmp->next;}) \
245 /* Rather than calling directly into the personality make_request function,
246 * IO requests come here first so that we can check if the device is
247 * being suspended pending a reconfiguration.
248 * We hold a refcount over the call to ->make_request. By the time that
249 * call has finished, the bio has been linked into some internal structure
250 * and so is visible to ->quiesce(), so we don't need the refcount any more.
252 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
254 const int rw
= bio_data_dir(bio
);
255 struct mddev
*mddev
= q
->queuedata
;
257 unsigned int sectors
;
259 if (mddev
== NULL
|| mddev
->pers
== NULL
264 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
265 bio_endio(bio
, bio_sectors(bio
) == 0 ? 0 : -EROFS
);
268 smp_rmb(); /* Ensure implications of 'active' are visible */
270 if (mddev
->suspended
) {
273 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
274 TASK_UNINTERRUPTIBLE
);
275 if (!mddev
->suspended
)
281 finish_wait(&mddev
->sb_wait
, &__wait
);
283 atomic_inc(&mddev
->active_io
);
287 * save the sectors now since our bio can
288 * go away inside make_request
290 sectors
= bio_sectors(bio
);
291 mddev
->pers
->make_request(mddev
, bio
);
293 cpu
= part_stat_lock();
294 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
295 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
298 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
299 wake_up(&mddev
->sb_wait
);
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once ->stop is called and completes, the module will be completely
308 void mddev_suspend(struct mddev
*mddev
)
310 BUG_ON(mddev
->suspended
);
311 mddev
->suspended
= 1;
313 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
314 mddev
->pers
->quiesce(mddev
, 1);
316 del_timer_sync(&mddev
->safemode_timer
);
318 EXPORT_SYMBOL_GPL(mddev_suspend
);
320 void mddev_resume(struct mddev
*mddev
)
322 mddev
->suspended
= 0;
323 wake_up(&mddev
->sb_wait
);
324 mddev
->pers
->quiesce(mddev
, 0);
326 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
327 md_wakeup_thread(mddev
->thread
);
328 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
330 EXPORT_SYMBOL_GPL(mddev_resume
);
332 int mddev_congested(struct mddev
*mddev
, int bits
)
334 return mddev
->suspended
;
336 EXPORT_SYMBOL(mddev_congested
);
339 * Generic flush handling for md
342 static void md_end_flush(struct bio
*bio
, int err
)
344 struct md_rdev
*rdev
= bio
->bi_private
;
345 struct mddev
*mddev
= rdev
->mddev
;
347 rdev_dec_pending(rdev
, mddev
);
349 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
350 /* The pre-request flush has finished */
351 queue_work(md_wq
, &mddev
->flush_work
);
356 static void md_submit_flush_data(struct work_struct
*ws
);
358 static void submit_flushes(struct work_struct
*ws
)
360 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
361 struct md_rdev
*rdev
;
363 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
364 atomic_set(&mddev
->flush_pending
, 1);
366 rdev_for_each_rcu(rdev
, mddev
)
367 if (rdev
->raid_disk
>= 0 &&
368 !test_bit(Faulty
, &rdev
->flags
)) {
369 /* Take two references, one is dropped
370 * when request finishes, one after
371 * we reclaim rcu_read_lock
374 atomic_inc(&rdev
->nr_pending
);
375 atomic_inc(&rdev
->nr_pending
);
377 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
378 bi
->bi_end_io
= md_end_flush
;
379 bi
->bi_private
= rdev
;
380 bi
->bi_bdev
= rdev
->bdev
;
381 atomic_inc(&mddev
->flush_pending
);
382 submit_bio(WRITE_FLUSH
, bi
);
384 rdev_dec_pending(rdev
, mddev
);
387 if (atomic_dec_and_test(&mddev
->flush_pending
))
388 queue_work(md_wq
, &mddev
->flush_work
);
391 static void md_submit_flush_data(struct work_struct
*ws
)
393 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
394 struct bio
*bio
= mddev
->flush_bio
;
396 if (bio
->bi_iter
.bi_size
== 0)
397 /* an empty barrier - all done */
400 bio
->bi_rw
&= ~REQ_FLUSH
;
401 mddev
->pers
->make_request(mddev
, bio
);
404 mddev
->flush_bio
= NULL
;
405 wake_up(&mddev
->sb_wait
);
408 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
410 spin_lock_irq(&mddev
->write_lock
);
411 wait_event_lock_irq(mddev
->sb_wait
,
414 mddev
->flush_bio
= bio
;
415 spin_unlock_irq(&mddev
->write_lock
);
417 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
418 queue_work(md_wq
, &mddev
->flush_work
);
420 EXPORT_SYMBOL(md_flush_request
);
422 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
424 struct mddev
*mddev
= cb
->data
;
425 md_wakeup_thread(mddev
->thread
);
428 EXPORT_SYMBOL(md_unplug
);
430 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
432 atomic_inc(&mddev
->active
);
436 static void mddev_delayed_delete(struct work_struct
*ws
);
438 static void mddev_put(struct mddev
*mddev
)
440 struct bio_set
*bs
= NULL
;
442 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
444 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
445 mddev
->ctime
== 0 && !mddev
->hold_active
) {
446 /* Array is not configured at all, and not held active,
448 list_del_init(&mddev
->all_mddevs
);
450 mddev
->bio_set
= NULL
;
451 if (mddev
->gendisk
) {
452 /* We did a probe so need to clean up. Call
453 * queue_work inside the spinlock so that
454 * flush_workqueue() after mddev_find will
455 * succeed in waiting for the work to be done.
457 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
458 queue_work(md_misc_wq
, &mddev
->del_work
);
462 spin_unlock(&all_mddevs_lock
);
467 void mddev_init(struct mddev
*mddev
)
469 mutex_init(&mddev
->open_mutex
);
470 mutex_init(&mddev
->reconfig_mutex
);
471 mutex_init(&mddev
->bitmap_info
.mutex
);
472 INIT_LIST_HEAD(&mddev
->disks
);
473 INIT_LIST_HEAD(&mddev
->all_mddevs
);
474 init_timer(&mddev
->safemode_timer
);
475 atomic_set(&mddev
->active
, 1);
476 atomic_set(&mddev
->openers
, 0);
477 atomic_set(&mddev
->active_io
, 0);
478 spin_lock_init(&mddev
->write_lock
);
479 atomic_set(&mddev
->flush_pending
, 0);
480 init_waitqueue_head(&mddev
->sb_wait
);
481 init_waitqueue_head(&mddev
->recovery_wait
);
482 mddev
->reshape_position
= MaxSector
;
483 mddev
->reshape_backwards
= 0;
484 mddev
->last_sync_action
= "none";
485 mddev
->resync_min
= 0;
486 mddev
->resync_max
= MaxSector
;
487 mddev
->level
= LEVEL_NONE
;
489 EXPORT_SYMBOL_GPL(mddev_init
);
491 static struct mddev
* mddev_find(dev_t unit
)
493 struct mddev
*mddev
, *new = NULL
;
495 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
496 unit
&= ~((1<<MdpMinorShift
)-1);
499 spin_lock(&all_mddevs_lock
);
502 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
503 if (mddev
->unit
== unit
) {
505 spin_unlock(&all_mddevs_lock
);
511 list_add(&new->all_mddevs
, &all_mddevs
);
512 spin_unlock(&all_mddevs_lock
);
513 new->hold_active
= UNTIL_IOCTL
;
517 /* find an unused unit number */
518 static int next_minor
= 512;
519 int start
= next_minor
;
523 dev
= MKDEV(MD_MAJOR
, next_minor
);
525 if (next_minor
> MINORMASK
)
527 if (next_minor
== start
) {
528 /* Oh dear, all in use. */
529 spin_unlock(&all_mddevs_lock
);
535 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
536 if (mddev
->unit
== dev
) {
542 new->md_minor
= MINOR(dev
);
543 new->hold_active
= UNTIL_STOP
;
544 list_add(&new->all_mddevs
, &all_mddevs
);
545 spin_unlock(&all_mddevs_lock
);
548 spin_unlock(&all_mddevs_lock
);
550 new = kzalloc(sizeof(*new), GFP_KERNEL
);
555 if (MAJOR(unit
) == MD_MAJOR
)
556 new->md_minor
= MINOR(unit
);
558 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
565 static inline int __must_check
mddev_lock(struct mddev
* mddev
)
567 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
570 /* Sometimes we need to take the lock in a situation where
571 * failure due to interrupts is not acceptable.
573 static inline void mddev_lock_nointr(struct mddev
* mddev
)
575 mutex_lock(&mddev
->reconfig_mutex
);
578 static inline int mddev_is_locked(struct mddev
*mddev
)
580 return mutex_is_locked(&mddev
->reconfig_mutex
);
583 static inline int mddev_trylock(struct mddev
* mddev
)
585 return mutex_trylock(&mddev
->reconfig_mutex
);
588 static struct attribute_group md_redundancy_group
;
590 static void mddev_unlock(struct mddev
* mddev
)
592 if (mddev
->to_remove
) {
593 /* These cannot be removed under reconfig_mutex as
594 * an access to the files will try to take reconfig_mutex
595 * while holding the file unremovable, which leads to
597 * So hold set sysfs_active while the remove in happeing,
598 * and anything else which might set ->to_remove or my
599 * otherwise change the sysfs namespace will fail with
600 * -EBUSY if sysfs_active is still set.
601 * We set sysfs_active under reconfig_mutex and elsewhere
602 * test it under the same mutex to ensure its correct value
605 struct attribute_group
*to_remove
= mddev
->to_remove
;
606 mddev
->to_remove
= NULL
;
607 mddev
->sysfs_active
= 1;
608 mutex_unlock(&mddev
->reconfig_mutex
);
610 if (mddev
->kobj
.sd
) {
611 if (to_remove
!= &md_redundancy_group
)
612 sysfs_remove_group(&mddev
->kobj
, to_remove
);
613 if (mddev
->pers
== NULL
||
614 mddev
->pers
->sync_request
== NULL
) {
615 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
616 if (mddev
->sysfs_action
)
617 sysfs_put(mddev
->sysfs_action
);
618 mddev
->sysfs_action
= NULL
;
621 mddev
->sysfs_active
= 0;
623 mutex_unlock(&mddev
->reconfig_mutex
);
625 /* As we've dropped the mutex we need a spinlock to
626 * make sure the thread doesn't disappear
628 spin_lock(&pers_lock
);
629 md_wakeup_thread(mddev
->thread
);
630 spin_unlock(&pers_lock
);
633 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
635 struct md_rdev
*rdev
;
637 rdev_for_each(rdev
, mddev
)
638 if (rdev
->desc_nr
== nr
)
644 static struct md_rdev
*find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
646 struct md_rdev
*rdev
;
648 rdev_for_each_rcu(rdev
, mddev
)
649 if (rdev
->desc_nr
== nr
)
655 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
657 struct md_rdev
*rdev
;
659 rdev_for_each(rdev
, mddev
)
660 if (rdev
->bdev
->bd_dev
== dev
)
666 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
668 struct md_rdev
*rdev
;
670 rdev_for_each_rcu(rdev
, mddev
)
671 if (rdev
->bdev
->bd_dev
== dev
)
677 static struct md_personality
*find_pers(int level
, char *clevel
)
679 struct md_personality
*pers
;
680 list_for_each_entry(pers
, &pers_list
, list
) {
681 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
683 if (strcmp(pers
->name
, clevel
)==0)
689 /* return the offset of the super block in 512byte sectors */
690 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
692 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
693 return MD_NEW_SIZE_SECTORS(num_sectors
);
696 static int alloc_disk_sb(struct md_rdev
* rdev
)
701 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
702 if (!rdev
->sb_page
) {
703 printk(KERN_ALERT
"md: out of memory.\n");
710 void md_rdev_clear(struct md_rdev
*rdev
)
713 put_page(rdev
->sb_page
);
715 rdev
->sb_page
= NULL
;
720 put_page(rdev
->bb_page
);
721 rdev
->bb_page
= NULL
;
723 kfree(rdev
->badblocks
.page
);
724 rdev
->badblocks
.page
= NULL
;
726 EXPORT_SYMBOL_GPL(md_rdev_clear
);
728 static void super_written(struct bio
*bio
, int error
)
730 struct md_rdev
*rdev
= bio
->bi_private
;
731 struct mddev
*mddev
= rdev
->mddev
;
733 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
734 printk("md: super_written gets error=%d, uptodate=%d\n",
735 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
736 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
737 md_error(mddev
, rdev
);
740 if (atomic_dec_and_test(&mddev
->pending_writes
))
741 wake_up(&mddev
->sb_wait
);
745 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
746 sector_t sector
, int size
, struct page
*page
)
748 /* write first size bytes of page to sector of rdev
749 * Increment mddev->pending_writes before returning
750 * and decrement it on completion, waking up sb_wait
751 * if zero is reached.
752 * If an error occurred, call md_error
754 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
756 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
757 bio
->bi_iter
.bi_sector
= sector
;
758 bio_add_page(bio
, page
, size
, 0);
759 bio
->bi_private
= rdev
;
760 bio
->bi_end_io
= super_written
;
762 atomic_inc(&mddev
->pending_writes
);
763 submit_bio(WRITE_FLUSH_FUA
, bio
);
766 void md_super_wait(struct mddev
*mddev
)
768 /* wait for all superblock writes that were scheduled to complete */
771 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
772 if (atomic_read(&mddev
->pending_writes
)==0)
776 finish_wait(&mddev
->sb_wait
, &wq
);
779 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
780 struct page
*page
, int rw
, bool metadata_op
)
782 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
785 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
786 rdev
->meta_bdev
: rdev
->bdev
;
788 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
789 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
790 (rdev
->mddev
->reshape_backwards
==
791 (sector
>= rdev
->mddev
->reshape_position
)))
792 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
794 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
795 bio_add_page(bio
, page
, size
, 0);
796 submit_bio_wait(rw
, bio
);
798 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
802 EXPORT_SYMBOL_GPL(sync_page_io
);
804 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
806 char b
[BDEVNAME_SIZE
];
807 if (!rdev
->sb_page
) {
815 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
821 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
822 bdevname(rdev
->bdev
,b
));
826 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
828 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
829 sb1
->set_uuid1
== sb2
->set_uuid1
&&
830 sb1
->set_uuid2
== sb2
->set_uuid2
&&
831 sb1
->set_uuid3
== sb2
->set_uuid3
;
834 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
837 mdp_super_t
*tmp1
, *tmp2
;
839 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
840 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
842 if (!tmp1
|| !tmp2
) {
844 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
852 * nr_disks is not constant
857 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
865 static u32
md_csum_fold(u32 csum
)
867 csum
= (csum
& 0xffff) + (csum
>> 16);
868 return (csum
& 0xffff) + (csum
>> 16);
871 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
874 u32
*sb32
= (u32
*)sb
;
876 unsigned int disk_csum
, csum
;
878 disk_csum
= sb
->sb_csum
;
881 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
883 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
887 /* This used to use csum_partial, which was wrong for several
888 * reasons including that different results are returned on
889 * different architectures. It isn't critical that we get exactly
890 * the same return value as before (we always csum_fold before
891 * testing, and that removes any differences). However as we
892 * know that csum_partial always returned a 16bit value on
893 * alphas, do a fold to maximise conformity to previous behaviour.
895 sb
->sb_csum
= md_csum_fold(disk_csum
);
897 sb
->sb_csum
= disk_csum
;
904 * Handle superblock details.
905 * We want to be able to handle multiple superblock formats
906 * so we have a common interface to them all, and an array of
907 * different handlers.
908 * We rely on user-space to write the initial superblock, and support
909 * reading and updating of superblocks.
910 * Interface methods are:
911 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
912 * loads and validates a superblock on dev.
913 * if refdev != NULL, compare superblocks on both devices
915 * 0 - dev has a superblock that is compatible with refdev
916 * 1 - dev has a superblock that is compatible and newer than refdev
917 * so dev should be used as the refdev in future
918 * -EINVAL superblock incompatible or invalid
919 * -othererror e.g. -EIO
921 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
922 * Verify that dev is acceptable into mddev.
923 * The first time, mddev->raid_disks will be 0, and data from
924 * dev should be merged in. Subsequent calls check that dev
925 * is new enough. Return 0 or -EINVAL
927 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
928 * Update the superblock for rdev with data in mddev
929 * This does not write to disc.
935 struct module
*owner
;
936 int (*load_super
)(struct md_rdev
*rdev
,
937 struct md_rdev
*refdev
,
939 int (*validate_super
)(struct mddev
*mddev
,
940 struct md_rdev
*rdev
);
941 void (*sync_super
)(struct mddev
*mddev
,
942 struct md_rdev
*rdev
);
943 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
944 sector_t num_sectors
);
945 int (*allow_new_offset
)(struct md_rdev
*rdev
,
946 unsigned long long new_offset
);
950 * Check that the given mddev has no bitmap.
952 * This function is called from the run method of all personalities that do not
953 * support bitmaps. It prints an error message and returns non-zero if mddev
954 * has a bitmap. Otherwise, it returns 0.
957 int md_check_no_bitmap(struct mddev
*mddev
)
959 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
961 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
962 mdname(mddev
), mddev
->pers
->name
);
965 EXPORT_SYMBOL(md_check_no_bitmap
);
968 * load_super for 0.90.0
970 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
972 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
977 * Calculate the position of the superblock (512byte sectors),
978 * it's at the end of the disk.
980 * It also happens to be a multiple of 4Kb.
982 rdev
->sb_start
= calc_dev_sboffset(rdev
);
984 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
989 bdevname(rdev
->bdev
, b
);
990 sb
= page_address(rdev
->sb_page
);
992 if (sb
->md_magic
!= MD_SB_MAGIC
) {
993 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
998 if (sb
->major_version
!= 0 ||
999 sb
->minor_version
< 90 ||
1000 sb
->minor_version
> 91) {
1001 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1002 sb
->major_version
, sb
->minor_version
,
1007 if (sb
->raid_disks
<= 0)
1010 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1011 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1016 rdev
->preferred_minor
= sb
->md_minor
;
1017 rdev
->data_offset
= 0;
1018 rdev
->new_data_offset
= 0;
1019 rdev
->sb_size
= MD_SB_BYTES
;
1020 rdev
->badblocks
.shift
= -1;
1022 if (sb
->level
== LEVEL_MULTIPATH
)
1025 rdev
->desc_nr
= sb
->this_disk
.number
;
1031 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1032 if (!uuid_equal(refsb
, sb
)) {
1033 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1034 b
, bdevname(refdev
->bdev
,b2
));
1037 if (!sb_equal(refsb
, sb
)) {
1038 printk(KERN_WARNING
"md: %s has same UUID"
1039 " but different superblock to %s\n",
1040 b
, bdevname(refdev
->bdev
, b2
));
1044 ev2
= md_event(refsb
);
1050 rdev
->sectors
= rdev
->sb_start
;
1051 /* Limit to 4TB as metadata cannot record more than that.
1052 * (not needed for Linear and RAID0 as metadata doesn't
1055 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1056 rdev
->sectors
= (2ULL << 32) - 2;
1058 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1059 /* "this cannot possibly happen" ... */
1067 * validate_super for 0.90.0
1069 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1072 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1073 __u64 ev1
= md_event(sb
);
1075 rdev
->raid_disk
= -1;
1076 clear_bit(Faulty
, &rdev
->flags
);
1077 clear_bit(In_sync
, &rdev
->flags
);
1078 clear_bit(Bitmap_sync
, &rdev
->flags
);
1079 clear_bit(WriteMostly
, &rdev
->flags
);
1081 if (mddev
->raid_disks
== 0) {
1082 mddev
->major_version
= 0;
1083 mddev
->minor_version
= sb
->minor_version
;
1084 mddev
->patch_version
= sb
->patch_version
;
1085 mddev
->external
= 0;
1086 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1087 mddev
->ctime
= sb
->ctime
;
1088 mddev
->utime
= sb
->utime
;
1089 mddev
->level
= sb
->level
;
1090 mddev
->clevel
[0] = 0;
1091 mddev
->layout
= sb
->layout
;
1092 mddev
->raid_disks
= sb
->raid_disks
;
1093 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1094 mddev
->events
= ev1
;
1095 mddev
->bitmap_info
.offset
= 0;
1096 mddev
->bitmap_info
.space
= 0;
1097 /* bitmap can use 60 K after the 4K superblocks */
1098 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1099 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1100 mddev
->reshape_backwards
= 0;
1102 if (mddev
->minor_version
>= 91) {
1103 mddev
->reshape_position
= sb
->reshape_position
;
1104 mddev
->delta_disks
= sb
->delta_disks
;
1105 mddev
->new_level
= sb
->new_level
;
1106 mddev
->new_layout
= sb
->new_layout
;
1107 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1108 if (mddev
->delta_disks
< 0)
1109 mddev
->reshape_backwards
= 1;
1111 mddev
->reshape_position
= MaxSector
;
1112 mddev
->delta_disks
= 0;
1113 mddev
->new_level
= mddev
->level
;
1114 mddev
->new_layout
= mddev
->layout
;
1115 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1118 if (sb
->state
& (1<<MD_SB_CLEAN
))
1119 mddev
->recovery_cp
= MaxSector
;
1121 if (sb
->events_hi
== sb
->cp_events_hi
&&
1122 sb
->events_lo
== sb
->cp_events_lo
) {
1123 mddev
->recovery_cp
= sb
->recovery_cp
;
1125 mddev
->recovery_cp
= 0;
1128 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1129 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1130 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1131 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1133 mddev
->max_disks
= MD_SB_DISKS
;
1135 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1136 mddev
->bitmap_info
.file
== NULL
) {
1137 mddev
->bitmap_info
.offset
=
1138 mddev
->bitmap_info
.default_offset
;
1139 mddev
->bitmap_info
.space
=
1140 mddev
->bitmap_info
.default_space
;
1143 } else if (mddev
->pers
== NULL
) {
1144 /* Insist on good event counter while assembling, except
1145 * for spares (which don't need an event count) */
1147 if (sb
->disks
[rdev
->desc_nr
].state
& (
1148 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1149 if (ev1
< mddev
->events
)
1151 } else if (mddev
->bitmap
) {
1152 /* if adding to array with a bitmap, then we can accept an
1153 * older device ... but not too old.
1155 if (ev1
< mddev
->bitmap
->events_cleared
)
1157 if (ev1
< mddev
->events
)
1158 set_bit(Bitmap_sync
, &rdev
->flags
);
1160 if (ev1
< mddev
->events
)
1161 /* just a hot-add of a new device, leave raid_disk at -1 */
1165 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1166 desc
= sb
->disks
+ rdev
->desc_nr
;
1168 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1169 set_bit(Faulty
, &rdev
->flags
);
1170 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1171 desc->raid_disk < mddev->raid_disks */) {
1172 set_bit(In_sync
, &rdev
->flags
);
1173 rdev
->raid_disk
= desc
->raid_disk
;
1174 rdev
->saved_raid_disk
= desc
->raid_disk
;
1175 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1176 /* active but not in sync implies recovery up to
1177 * reshape position. We don't know exactly where
1178 * that is, so set to zero for now */
1179 if (mddev
->minor_version
>= 91) {
1180 rdev
->recovery_offset
= 0;
1181 rdev
->raid_disk
= desc
->raid_disk
;
1184 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1185 set_bit(WriteMostly
, &rdev
->flags
);
1186 } else /* MULTIPATH are always insync */
1187 set_bit(In_sync
, &rdev
->flags
);
1192 * sync_super for 0.90.0
1194 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1197 struct md_rdev
*rdev2
;
1198 int next_spare
= mddev
->raid_disks
;
1201 /* make rdev->sb match mddev data..
1204 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1205 * 3/ any empty disks < next_spare become removed
1207 * disks[0] gets initialised to REMOVED because
1208 * we cannot be sure from other fields if it has
1209 * been initialised or not.
1212 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1214 rdev
->sb_size
= MD_SB_BYTES
;
1216 sb
= page_address(rdev
->sb_page
);
1218 memset(sb
, 0, sizeof(*sb
));
1220 sb
->md_magic
= MD_SB_MAGIC
;
1221 sb
->major_version
= mddev
->major_version
;
1222 sb
->patch_version
= mddev
->patch_version
;
1223 sb
->gvalid_words
= 0; /* ignored */
1224 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1225 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1226 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1227 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1229 sb
->ctime
= mddev
->ctime
;
1230 sb
->level
= mddev
->level
;
1231 sb
->size
= mddev
->dev_sectors
/ 2;
1232 sb
->raid_disks
= mddev
->raid_disks
;
1233 sb
->md_minor
= mddev
->md_minor
;
1234 sb
->not_persistent
= 0;
1235 sb
->utime
= mddev
->utime
;
1237 sb
->events_hi
= (mddev
->events
>>32);
1238 sb
->events_lo
= (u32
)mddev
->events
;
1240 if (mddev
->reshape_position
== MaxSector
)
1241 sb
->minor_version
= 90;
1243 sb
->minor_version
= 91;
1244 sb
->reshape_position
= mddev
->reshape_position
;
1245 sb
->new_level
= mddev
->new_level
;
1246 sb
->delta_disks
= mddev
->delta_disks
;
1247 sb
->new_layout
= mddev
->new_layout
;
1248 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1250 mddev
->minor_version
= sb
->minor_version
;
1253 sb
->recovery_cp
= mddev
->recovery_cp
;
1254 sb
->cp_events_hi
= (mddev
->events
>>32);
1255 sb
->cp_events_lo
= (u32
)mddev
->events
;
1256 if (mddev
->recovery_cp
== MaxSector
)
1257 sb
->state
= (1<< MD_SB_CLEAN
);
1259 sb
->recovery_cp
= 0;
1261 sb
->layout
= mddev
->layout
;
1262 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1264 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1265 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1267 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1268 rdev_for_each(rdev2
, mddev
) {
1271 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1273 if (rdev2
->raid_disk
>= 0 &&
1274 sb
->minor_version
>= 91)
1275 /* we have nowhere to store the recovery_offset,
1276 * but if it is not below the reshape_position,
1277 * we can piggy-back on that.
1280 if (rdev2
->raid_disk
< 0 ||
1281 test_bit(Faulty
, &rdev2
->flags
))
1284 desc_nr
= rdev2
->raid_disk
;
1286 desc_nr
= next_spare
++;
1287 rdev2
->desc_nr
= desc_nr
;
1288 d
= &sb
->disks
[rdev2
->desc_nr
];
1290 d
->number
= rdev2
->desc_nr
;
1291 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1292 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1294 d
->raid_disk
= rdev2
->raid_disk
;
1296 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1297 if (test_bit(Faulty
, &rdev2
->flags
))
1298 d
->state
= (1<<MD_DISK_FAULTY
);
1299 else if (is_active
) {
1300 d
->state
= (1<<MD_DISK_ACTIVE
);
1301 if (test_bit(In_sync
, &rdev2
->flags
))
1302 d
->state
|= (1<<MD_DISK_SYNC
);
1310 if (test_bit(WriteMostly
, &rdev2
->flags
))
1311 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1313 /* now set the "removed" and "faulty" bits on any missing devices */
1314 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1315 mdp_disk_t
*d
= &sb
->disks
[i
];
1316 if (d
->state
== 0 && d
->number
== 0) {
1319 d
->state
= (1<<MD_DISK_REMOVED
);
1320 d
->state
|= (1<<MD_DISK_FAULTY
);
1324 sb
->nr_disks
= nr_disks
;
1325 sb
->active_disks
= active
;
1326 sb
->working_disks
= working
;
1327 sb
->failed_disks
= failed
;
1328 sb
->spare_disks
= spare
;
1330 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1331 sb
->sb_csum
= calc_sb_csum(sb
);
1335 * rdev_size_change for 0.90.0
1337 static unsigned long long
1338 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1340 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1341 return 0; /* component must fit device */
1342 if (rdev
->mddev
->bitmap_info
.offset
)
1343 return 0; /* can't move bitmap */
1344 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1345 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1346 num_sectors
= rdev
->sb_start
;
1347 /* Limit to 4TB as metadata cannot record more than that.
1348 * 4TB == 2^32 KB, or 2*2^32 sectors.
1350 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1351 num_sectors
= (2ULL << 32) - 2;
1352 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1354 md_super_wait(rdev
->mddev
);
1359 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1361 /* non-zero offset changes not possible with v0.90 */
1362 return new_offset
== 0;
1366 * version 1 superblock
1369 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1373 unsigned long long newcsum
;
1374 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1375 __le32
*isuper
= (__le32
*)sb
;
1377 disk_csum
= sb
->sb_csum
;
1380 for (; size
>= 4; size
-= 4)
1381 newcsum
+= le32_to_cpu(*isuper
++);
1384 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1386 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1387 sb
->sb_csum
= disk_csum
;
1388 return cpu_to_le32(csum
);
1391 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1393 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1395 struct mdp_superblock_1
*sb
;
1399 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1403 * Calculate the position of the superblock in 512byte sectors.
1404 * It is always aligned to a 4K boundary and
1405 * depeding on minor_version, it can be:
1406 * 0: At least 8K, but less than 12K, from end of device
1407 * 1: At start of device
1408 * 2: 4K from start of device.
1410 switch(minor_version
) {
1412 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1414 sb_start
&= ~(sector_t
)(4*2-1);
1425 rdev
->sb_start
= sb_start
;
1427 /* superblock is rarely larger than 1K, but it can be larger,
1428 * and it is safe to read 4k, so we do that
1430 ret
= read_disk_sb(rdev
, 4096);
1431 if (ret
) return ret
;
1434 sb
= page_address(rdev
->sb_page
);
1436 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1437 sb
->major_version
!= cpu_to_le32(1) ||
1438 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1439 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1440 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1443 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1444 printk("md: invalid superblock checksum on %s\n",
1445 bdevname(rdev
->bdev
,b
));
1448 if (le64_to_cpu(sb
->data_size
) < 10) {
1449 printk("md: data_size too small on %s\n",
1450 bdevname(rdev
->bdev
,b
));
1455 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1456 /* Some padding is non-zero, might be a new feature */
1459 rdev
->preferred_minor
= 0xffff;
1460 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1461 rdev
->new_data_offset
= rdev
->data_offset
;
1462 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1463 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1464 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1465 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1467 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1468 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1469 if (rdev
->sb_size
& bmask
)
1470 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1473 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1476 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1479 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1482 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1484 if (!rdev
->bb_page
) {
1485 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1489 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1490 rdev
->badblocks
.count
== 0) {
1491 /* need to load the bad block list.
1492 * Currently we limit it to one page.
1498 int sectors
= le16_to_cpu(sb
->bblog_size
);
1499 if (sectors
> (PAGE_SIZE
/ 512))
1501 offset
= le32_to_cpu(sb
->bblog_offset
);
1504 bb_sector
= (long long)offset
;
1505 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1506 rdev
->bb_page
, READ
, true))
1508 bbp
= (u64
*)page_address(rdev
->bb_page
);
1509 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1510 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1511 u64 bb
= le64_to_cpu(*bbp
);
1512 int count
= bb
& (0x3ff);
1513 u64 sector
= bb
>> 10;
1514 sector
<<= sb
->bblog_shift
;
1515 count
<<= sb
->bblog_shift
;
1518 if (md_set_badblocks(&rdev
->badblocks
,
1519 sector
, count
, 1) == 0)
1522 } else if (sb
->bblog_offset
!= 0)
1523 rdev
->badblocks
.shift
= 0;
1529 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1531 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1532 sb
->level
!= refsb
->level
||
1533 sb
->layout
!= refsb
->layout
||
1534 sb
->chunksize
!= refsb
->chunksize
) {
1535 printk(KERN_WARNING
"md: %s has strangely different"
1536 " superblock to %s\n",
1537 bdevname(rdev
->bdev
,b
),
1538 bdevname(refdev
->bdev
,b2
));
1541 ev1
= le64_to_cpu(sb
->events
);
1542 ev2
= le64_to_cpu(refsb
->events
);
1549 if (minor_version
) {
1550 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1551 sectors
-= rdev
->data_offset
;
1553 sectors
= rdev
->sb_start
;
1554 if (sectors
< le64_to_cpu(sb
->data_size
))
1556 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1560 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1562 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1563 __u64 ev1
= le64_to_cpu(sb
->events
);
1565 rdev
->raid_disk
= -1;
1566 clear_bit(Faulty
, &rdev
->flags
);
1567 clear_bit(In_sync
, &rdev
->flags
);
1568 clear_bit(Bitmap_sync
, &rdev
->flags
);
1569 clear_bit(WriteMostly
, &rdev
->flags
);
1571 if (mddev
->raid_disks
== 0) {
1572 mddev
->major_version
= 1;
1573 mddev
->patch_version
= 0;
1574 mddev
->external
= 0;
1575 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1576 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1577 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1578 mddev
->level
= le32_to_cpu(sb
->level
);
1579 mddev
->clevel
[0] = 0;
1580 mddev
->layout
= le32_to_cpu(sb
->layout
);
1581 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1582 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1583 mddev
->events
= ev1
;
1584 mddev
->bitmap_info
.offset
= 0;
1585 mddev
->bitmap_info
.space
= 0;
1586 /* Default location for bitmap is 1K after superblock
1587 * using 3K - total of 4K
1589 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1590 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1591 mddev
->reshape_backwards
= 0;
1593 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1594 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1596 mddev
->max_disks
= (4096-256)/2;
1598 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1599 mddev
->bitmap_info
.file
== NULL
) {
1600 mddev
->bitmap_info
.offset
=
1601 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1602 /* Metadata doesn't record how much space is available.
1603 * For 1.0, we assume we can use up to the superblock
1604 * if before, else to 4K beyond superblock.
1605 * For others, assume no change is possible.
1607 if (mddev
->minor_version
> 0)
1608 mddev
->bitmap_info
.space
= 0;
1609 else if (mddev
->bitmap_info
.offset
> 0)
1610 mddev
->bitmap_info
.space
=
1611 8 - mddev
->bitmap_info
.offset
;
1613 mddev
->bitmap_info
.space
=
1614 -mddev
->bitmap_info
.offset
;
1617 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1618 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1619 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1620 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1621 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1622 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1623 if (mddev
->delta_disks
< 0 ||
1624 (mddev
->delta_disks
== 0 &&
1625 (le32_to_cpu(sb
->feature_map
)
1626 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1627 mddev
->reshape_backwards
= 1;
1629 mddev
->reshape_position
= MaxSector
;
1630 mddev
->delta_disks
= 0;
1631 mddev
->new_level
= mddev
->level
;
1632 mddev
->new_layout
= mddev
->layout
;
1633 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1636 } else if (mddev
->pers
== NULL
) {
1637 /* Insist of good event counter while assembling, except for
1638 * spares (which don't need an event count) */
1640 if (rdev
->desc_nr
>= 0 &&
1641 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1642 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1643 if (ev1
< mddev
->events
)
1645 } else if (mddev
->bitmap
) {
1646 /* If adding to array with a bitmap, then we can accept an
1647 * older device, but not too old.
1649 if (ev1
< mddev
->bitmap
->events_cleared
)
1651 if (ev1
< mddev
->events
)
1652 set_bit(Bitmap_sync
, &rdev
->flags
);
1654 if (ev1
< mddev
->events
)
1655 /* just a hot-add of a new device, leave raid_disk at -1 */
1658 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1660 if (rdev
->desc_nr
< 0 ||
1661 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1665 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1667 case 0xffff: /* spare */
1669 case 0xfffe: /* faulty */
1670 set_bit(Faulty
, &rdev
->flags
);
1673 rdev
->saved_raid_disk
= role
;
1674 if ((le32_to_cpu(sb
->feature_map
) &
1675 MD_FEATURE_RECOVERY_OFFSET
)) {
1676 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1677 if (!(le32_to_cpu(sb
->feature_map
) &
1678 MD_FEATURE_RECOVERY_BITMAP
))
1679 rdev
->saved_raid_disk
= -1;
1681 set_bit(In_sync
, &rdev
->flags
);
1682 rdev
->raid_disk
= role
;
1685 if (sb
->devflags
& WriteMostly1
)
1686 set_bit(WriteMostly
, &rdev
->flags
);
1687 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1688 set_bit(Replacement
, &rdev
->flags
);
1689 } else /* MULTIPATH are always insync */
1690 set_bit(In_sync
, &rdev
->flags
);
1695 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1697 struct mdp_superblock_1
*sb
;
1698 struct md_rdev
*rdev2
;
1700 /* make rdev->sb match mddev and rdev data. */
1702 sb
= page_address(rdev
->sb_page
);
1704 sb
->feature_map
= 0;
1706 sb
->recovery_offset
= cpu_to_le64(0);
1707 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1709 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1710 sb
->events
= cpu_to_le64(mddev
->events
);
1712 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1714 sb
->resync_offset
= cpu_to_le64(0);
1716 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1718 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1719 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1720 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1721 sb
->level
= cpu_to_le32(mddev
->level
);
1722 sb
->layout
= cpu_to_le32(mddev
->layout
);
1724 if (test_bit(WriteMostly
, &rdev
->flags
))
1725 sb
->devflags
|= WriteMostly1
;
1727 sb
->devflags
&= ~WriteMostly1
;
1728 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1729 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1731 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1732 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1733 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1736 if (rdev
->raid_disk
>= 0 &&
1737 !test_bit(In_sync
, &rdev
->flags
)) {
1739 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1740 sb
->recovery_offset
=
1741 cpu_to_le64(rdev
->recovery_offset
);
1742 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1744 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1746 if (test_bit(Replacement
, &rdev
->flags
))
1748 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1750 if (mddev
->reshape_position
!= MaxSector
) {
1751 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1752 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1753 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1754 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1755 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1756 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1757 if (mddev
->delta_disks
== 0 &&
1758 mddev
->reshape_backwards
)
1760 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1761 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1763 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1764 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1765 - rdev
->data_offset
));
1769 if (rdev
->badblocks
.count
== 0)
1770 /* Nothing to do for bad blocks*/ ;
1771 else if (sb
->bblog_offset
== 0)
1772 /* Cannot record bad blocks on this device */
1773 md_error(mddev
, rdev
);
1775 struct badblocks
*bb
= &rdev
->badblocks
;
1776 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1778 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1783 seq
= read_seqbegin(&bb
->lock
);
1785 memset(bbp
, 0xff, PAGE_SIZE
);
1787 for (i
= 0 ; i
< bb
->count
; i
++) {
1788 u64 internal_bb
= p
[i
];
1789 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1790 | BB_LEN(internal_bb
));
1791 bbp
[i
] = cpu_to_le64(store_bb
);
1794 if (read_seqretry(&bb
->lock
, seq
))
1797 bb
->sector
= (rdev
->sb_start
+
1798 (int)le32_to_cpu(sb
->bblog_offset
));
1799 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1804 rdev_for_each(rdev2
, mddev
)
1805 if (rdev2
->desc_nr
+1 > max_dev
)
1806 max_dev
= rdev2
->desc_nr
+1;
1808 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1810 sb
->max_dev
= cpu_to_le32(max_dev
);
1811 rdev
->sb_size
= max_dev
* 2 + 256;
1812 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1813 if (rdev
->sb_size
& bmask
)
1814 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1816 max_dev
= le32_to_cpu(sb
->max_dev
);
1818 for (i
=0; i
<max_dev
;i
++)
1819 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1821 rdev_for_each(rdev2
, mddev
) {
1823 if (test_bit(Faulty
, &rdev2
->flags
))
1824 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1825 else if (test_bit(In_sync
, &rdev2
->flags
))
1826 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1827 else if (rdev2
->raid_disk
>= 0)
1828 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1830 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1833 sb
->sb_csum
= calc_sb_1_csum(sb
);
1836 static unsigned long long
1837 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1839 struct mdp_superblock_1
*sb
;
1840 sector_t max_sectors
;
1841 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1842 return 0; /* component must fit device */
1843 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1844 return 0; /* too confusing */
1845 if (rdev
->sb_start
< rdev
->data_offset
) {
1846 /* minor versions 1 and 2; superblock before data */
1847 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1848 max_sectors
-= rdev
->data_offset
;
1849 if (!num_sectors
|| num_sectors
> max_sectors
)
1850 num_sectors
= max_sectors
;
1851 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1852 /* minor version 0 with bitmap we can't move */
1855 /* minor version 0; superblock after data */
1857 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1858 sb_start
&= ~(sector_t
)(4*2 - 1);
1859 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1860 if (!num_sectors
|| num_sectors
> max_sectors
)
1861 num_sectors
= max_sectors
;
1862 rdev
->sb_start
= sb_start
;
1864 sb
= page_address(rdev
->sb_page
);
1865 sb
->data_size
= cpu_to_le64(num_sectors
);
1866 sb
->super_offset
= rdev
->sb_start
;
1867 sb
->sb_csum
= calc_sb_1_csum(sb
);
1868 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1870 md_super_wait(rdev
->mddev
);
1876 super_1_allow_new_offset(struct md_rdev
*rdev
,
1877 unsigned long long new_offset
)
1879 /* All necessary checks on new >= old have been done */
1880 struct bitmap
*bitmap
;
1881 if (new_offset
>= rdev
->data_offset
)
1884 /* with 1.0 metadata, there is no metadata to tread on
1885 * so we can always move back */
1886 if (rdev
->mddev
->minor_version
== 0)
1889 /* otherwise we must be sure not to step on
1890 * any metadata, so stay:
1891 * 36K beyond start of superblock
1892 * beyond end of badblocks
1893 * beyond write-intent bitmap
1895 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1897 bitmap
= rdev
->mddev
->bitmap
;
1898 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1899 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1900 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1902 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1908 static struct super_type super_types
[] = {
1911 .owner
= THIS_MODULE
,
1912 .load_super
= super_90_load
,
1913 .validate_super
= super_90_validate
,
1914 .sync_super
= super_90_sync
,
1915 .rdev_size_change
= super_90_rdev_size_change
,
1916 .allow_new_offset
= super_90_allow_new_offset
,
1920 .owner
= THIS_MODULE
,
1921 .load_super
= super_1_load
,
1922 .validate_super
= super_1_validate
,
1923 .sync_super
= super_1_sync
,
1924 .rdev_size_change
= super_1_rdev_size_change
,
1925 .allow_new_offset
= super_1_allow_new_offset
,
1929 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1931 if (mddev
->sync_super
) {
1932 mddev
->sync_super(mddev
, rdev
);
1936 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1938 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1941 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1943 struct md_rdev
*rdev
, *rdev2
;
1946 rdev_for_each_rcu(rdev
, mddev1
)
1947 rdev_for_each_rcu(rdev2
, mddev2
)
1948 if (rdev
->bdev
->bd_contains
==
1949 rdev2
->bdev
->bd_contains
) {
1957 static LIST_HEAD(pending_raid_disks
);
1960 * Try to register data integrity profile for an mddev
1962 * This is called when an array is started and after a disk has been kicked
1963 * from the array. It only succeeds if all working and active component devices
1964 * are integrity capable with matching profiles.
1966 int md_integrity_register(struct mddev
*mddev
)
1968 struct md_rdev
*rdev
, *reference
= NULL
;
1970 if (list_empty(&mddev
->disks
))
1971 return 0; /* nothing to do */
1972 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1973 return 0; /* shouldn't register, or already is */
1974 rdev_for_each(rdev
, mddev
) {
1975 /* skip spares and non-functional disks */
1976 if (test_bit(Faulty
, &rdev
->flags
))
1978 if (rdev
->raid_disk
< 0)
1981 /* Use the first rdev as the reference */
1985 /* does this rdev's profile match the reference profile? */
1986 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1987 rdev
->bdev
->bd_disk
) < 0)
1990 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1993 * All component devices are integrity capable and have matching
1994 * profiles, register the common profile for the md device.
1996 if (blk_integrity_register(mddev
->gendisk
,
1997 bdev_get_integrity(reference
->bdev
)) != 0) {
1998 printk(KERN_ERR
"md: failed to register integrity for %s\n",
2002 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2003 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2004 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2010 EXPORT_SYMBOL(md_integrity_register
);
2012 /* Disable data integrity if non-capable/non-matching disk is being added */
2013 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2015 struct blk_integrity
*bi_rdev
;
2016 struct blk_integrity
*bi_mddev
;
2018 if (!mddev
->gendisk
)
2021 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2022 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2024 if (!bi_mddev
) /* nothing to do */
2026 if (rdev
->raid_disk
< 0) /* skip spares */
2028 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2029 rdev
->bdev
->bd_disk
) >= 0)
2031 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2032 blk_integrity_unregister(mddev
->gendisk
);
2034 EXPORT_SYMBOL(md_integrity_add_rdev
);
2036 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2038 char b
[BDEVNAME_SIZE
];
2048 /* prevent duplicates */
2049 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2052 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2053 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2054 rdev
->sectors
< mddev
->dev_sectors
)) {
2056 /* Cannot change size, so fail
2057 * If mddev->level <= 0, then we don't care
2058 * about aligning sizes (e.g. linear)
2060 if (mddev
->level
> 0)
2063 mddev
->dev_sectors
= rdev
->sectors
;
2066 /* Verify rdev->desc_nr is unique.
2067 * If it is -1, assign a free number, else
2068 * check number is not in use
2070 if (rdev
->desc_nr
< 0) {
2072 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2073 while (find_rdev_nr(mddev
, choice
))
2075 rdev
->desc_nr
= choice
;
2077 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2080 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2081 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2082 mdname(mddev
), mddev
->max_disks
);
2085 bdevname(rdev
->bdev
,b
);
2086 while ( (s
=strchr(b
, '/')) != NULL
)
2089 rdev
->mddev
= mddev
;
2090 printk(KERN_INFO
"md: bind<%s>\n", b
);
2092 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2095 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2096 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2097 /* failure here is OK */;
2098 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2100 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2101 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2103 /* May as well allow recovery to be retried once */
2104 mddev
->recovery_disabled
++;
2109 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2114 static void md_delayed_delete(struct work_struct
*ws
)
2116 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2117 kobject_del(&rdev
->kobj
);
2118 kobject_put(&rdev
->kobj
);
2121 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2123 char b
[BDEVNAME_SIZE
];
2128 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2129 list_del_rcu(&rdev
->same_set
);
2130 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2132 sysfs_remove_link(&rdev
->kobj
, "block");
2133 sysfs_put(rdev
->sysfs_state
);
2134 rdev
->sysfs_state
= NULL
;
2135 rdev
->badblocks
.count
= 0;
2136 /* We need to delay this, otherwise we can deadlock when
2137 * writing to 'remove' to "dev/state". We also need
2138 * to delay it due to rcu usage.
2141 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2142 kobject_get(&rdev
->kobj
);
2143 queue_work(md_misc_wq
, &rdev
->del_work
);
2147 * prevent the device from being mounted, repartitioned or
2148 * otherwise reused by a RAID array (or any other kernel
2149 * subsystem), by bd_claiming the device.
2151 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2154 struct block_device
*bdev
;
2155 char b
[BDEVNAME_SIZE
];
2157 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2158 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2160 printk(KERN_ERR
"md: could not open %s.\n",
2161 __bdevname(dev
, b
));
2162 return PTR_ERR(bdev
);
2168 static void unlock_rdev(struct md_rdev
*rdev
)
2170 struct block_device
*bdev
= rdev
->bdev
;
2174 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2177 void md_autodetect_dev(dev_t dev
);
2179 static void export_rdev(struct md_rdev
* rdev
)
2181 char b
[BDEVNAME_SIZE
];
2182 printk(KERN_INFO
"md: export_rdev(%s)\n",
2183 bdevname(rdev
->bdev
,b
));
2186 md_rdev_clear(rdev
);
2188 if (test_bit(AutoDetected
, &rdev
->flags
))
2189 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2192 kobject_put(&rdev
->kobj
);
2195 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2197 unbind_rdev_from_array(rdev
);
2201 static void export_array(struct mddev
*mddev
)
2203 struct md_rdev
*rdev
, *tmp
;
2205 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2210 kick_rdev_from_array(rdev
);
2212 if (!list_empty(&mddev
->disks
))
2214 mddev
->raid_disks
= 0;
2215 mddev
->major_version
= 0;
2218 static void print_desc(mdp_disk_t
*desc
)
2220 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2221 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2224 static void print_sb_90(mdp_super_t
*sb
)
2229 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2230 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2231 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2233 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2234 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2235 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2236 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2237 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2238 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2239 sb
->failed_disks
, sb
->spare_disks
,
2240 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2243 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2246 desc
= sb
->disks
+ i
;
2247 if (desc
->number
|| desc
->major
|| desc
->minor
||
2248 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2249 printk(" D %2d: ", i
);
2253 printk(KERN_INFO
"md: THIS: ");
2254 print_desc(&sb
->this_disk
);
2257 static void print_sb_1(struct mdp_superblock_1
*sb
)
2261 uuid
= sb
->set_uuid
;
2263 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2264 "md: Name: \"%s\" CT:%llu\n",
2265 le32_to_cpu(sb
->major_version
),
2266 le32_to_cpu(sb
->feature_map
),
2269 (unsigned long long)le64_to_cpu(sb
->ctime
)
2270 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2272 uuid
= sb
->device_uuid
;
2274 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2276 "md: Dev:%08x UUID: %pU\n"
2277 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2278 "md: (MaxDev:%u) \n",
2279 le32_to_cpu(sb
->level
),
2280 (unsigned long long)le64_to_cpu(sb
->size
),
2281 le32_to_cpu(sb
->raid_disks
),
2282 le32_to_cpu(sb
->layout
),
2283 le32_to_cpu(sb
->chunksize
),
2284 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2285 (unsigned long long)le64_to_cpu(sb
->data_size
),
2286 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2287 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2288 le32_to_cpu(sb
->dev_number
),
2291 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2292 (unsigned long long)le64_to_cpu(sb
->events
),
2293 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2294 le32_to_cpu(sb
->sb_csum
),
2295 le32_to_cpu(sb
->max_dev
)
2299 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2301 char b
[BDEVNAME_SIZE
];
2302 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2303 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2304 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2306 if (rdev
->sb_loaded
) {
2307 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2308 switch (major_version
) {
2310 print_sb_90(page_address(rdev
->sb_page
));
2313 print_sb_1(page_address(rdev
->sb_page
));
2317 printk(KERN_INFO
"md: no rdev superblock!\n");
2320 static void md_print_devices(void)
2322 struct list_head
*tmp
;
2323 struct md_rdev
*rdev
;
2324 struct mddev
*mddev
;
2325 char b
[BDEVNAME_SIZE
];
2328 printk("md: **********************************\n");
2329 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2330 printk("md: **********************************\n");
2331 for_each_mddev(mddev
, tmp
) {
2334 bitmap_print_sb(mddev
->bitmap
);
2336 printk("%s: ", mdname(mddev
));
2337 rdev_for_each(rdev
, mddev
)
2338 printk("<%s>", bdevname(rdev
->bdev
,b
));
2341 rdev_for_each(rdev
, mddev
)
2342 print_rdev(rdev
, mddev
->major_version
);
2344 printk("md: **********************************\n");
2349 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2351 /* Update each superblock (in-memory image), but
2352 * if we are allowed to, skip spares which already
2353 * have the right event counter, or have one earlier
2354 * (which would mean they aren't being marked as dirty
2355 * with the rest of the array)
2357 struct md_rdev
*rdev
;
2358 rdev_for_each(rdev
, mddev
) {
2359 if (rdev
->sb_events
== mddev
->events
||
2361 rdev
->raid_disk
< 0 &&
2362 rdev
->sb_events
+1 == mddev
->events
)) {
2363 /* Don't update this superblock */
2364 rdev
->sb_loaded
= 2;
2366 sync_super(mddev
, rdev
);
2367 rdev
->sb_loaded
= 1;
2372 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2374 struct md_rdev
*rdev
;
2377 int any_badblocks_changed
= 0;
2381 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2385 /* First make sure individual recovery_offsets are correct */
2386 rdev_for_each(rdev
, mddev
) {
2387 if (rdev
->raid_disk
>= 0 &&
2388 mddev
->delta_disks
>= 0 &&
2389 !test_bit(In_sync
, &rdev
->flags
) &&
2390 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2391 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2394 if (!mddev
->persistent
) {
2395 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2396 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2397 if (!mddev
->external
) {
2398 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2399 rdev_for_each(rdev
, mddev
) {
2400 if (rdev
->badblocks
.changed
) {
2401 rdev
->badblocks
.changed
= 0;
2402 md_ack_all_badblocks(&rdev
->badblocks
);
2403 md_error(mddev
, rdev
);
2405 clear_bit(Blocked
, &rdev
->flags
);
2406 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2407 wake_up(&rdev
->blocked_wait
);
2410 wake_up(&mddev
->sb_wait
);
2414 spin_lock_irq(&mddev
->write_lock
);
2416 mddev
->utime
= get_seconds();
2418 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2420 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2421 /* just a clean<-> dirty transition, possibly leave spares alone,
2422 * though if events isn't the right even/odd, we will have to do
2428 if (mddev
->degraded
)
2429 /* If the array is degraded, then skipping spares is both
2430 * dangerous and fairly pointless.
2431 * Dangerous because a device that was removed from the array
2432 * might have a event_count that still looks up-to-date,
2433 * so it can be re-added without a resync.
2434 * Pointless because if there are any spares to skip,
2435 * then a recovery will happen and soon that array won't
2436 * be degraded any more and the spare can go back to sleep then.
2440 sync_req
= mddev
->in_sync
;
2442 /* If this is just a dirty<->clean transition, and the array is clean
2443 * and 'events' is odd, we can roll back to the previous clean state */
2445 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2446 && mddev
->can_decrease_events
2447 && mddev
->events
!= 1) {
2449 mddev
->can_decrease_events
= 0;
2451 /* otherwise we have to go forward and ... */
2453 mddev
->can_decrease_events
= nospares
;
2456 if (!mddev
->events
) {
2458 * oops, this 64-bit counter should never wrap.
2459 * Either we are in around ~1 trillion A.C., assuming
2460 * 1 reboot per second, or we have a bug:
2466 rdev_for_each(rdev
, mddev
) {
2467 if (rdev
->badblocks
.changed
)
2468 any_badblocks_changed
++;
2469 if (test_bit(Faulty
, &rdev
->flags
))
2470 set_bit(FaultRecorded
, &rdev
->flags
);
2473 sync_sbs(mddev
, nospares
);
2474 spin_unlock_irq(&mddev
->write_lock
);
2476 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2477 mdname(mddev
), mddev
->in_sync
);
2479 bitmap_update_sb(mddev
->bitmap
);
2480 rdev_for_each(rdev
, mddev
) {
2481 char b
[BDEVNAME_SIZE
];
2483 if (rdev
->sb_loaded
!= 1)
2484 continue; /* no noise on spare devices */
2486 if (!test_bit(Faulty
, &rdev
->flags
)) {
2487 md_super_write(mddev
,rdev
,
2488 rdev
->sb_start
, rdev
->sb_size
,
2490 pr_debug("md: (write) %s's sb offset: %llu\n",
2491 bdevname(rdev
->bdev
, b
),
2492 (unsigned long long)rdev
->sb_start
);
2493 rdev
->sb_events
= mddev
->events
;
2494 if (rdev
->badblocks
.size
) {
2495 md_super_write(mddev
, rdev
,
2496 rdev
->badblocks
.sector
,
2497 rdev
->badblocks
.size
<< 9,
2499 rdev
->badblocks
.size
= 0;
2503 pr_debug("md: %s (skipping faulty)\n",
2504 bdevname(rdev
->bdev
, b
));
2506 if (mddev
->level
== LEVEL_MULTIPATH
)
2507 /* only need to write one superblock... */
2510 md_super_wait(mddev
);
2511 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2513 spin_lock_irq(&mddev
->write_lock
);
2514 if (mddev
->in_sync
!= sync_req
||
2515 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2516 /* have to write it out again */
2517 spin_unlock_irq(&mddev
->write_lock
);
2520 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2521 spin_unlock_irq(&mddev
->write_lock
);
2522 wake_up(&mddev
->sb_wait
);
2523 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2524 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2526 rdev_for_each(rdev
, mddev
) {
2527 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2528 clear_bit(Blocked
, &rdev
->flags
);
2530 if (any_badblocks_changed
)
2531 md_ack_all_badblocks(&rdev
->badblocks
);
2532 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2533 wake_up(&rdev
->blocked_wait
);
2537 /* words written to sysfs files may, or may not, be \n terminated.
2538 * We want to accept with case. For this we use cmd_match.
2540 static int cmd_match(const char *cmd
, const char *str
)
2542 /* See if cmd, written into a sysfs file, matches
2543 * str. They must either be the same, or cmd can
2544 * have a trailing newline
2546 while (*cmd
&& *str
&& *cmd
== *str
) {
2557 struct rdev_sysfs_entry
{
2558 struct attribute attr
;
2559 ssize_t (*show
)(struct md_rdev
*, char *);
2560 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2564 state_show(struct md_rdev
*rdev
, char *page
)
2569 if (test_bit(Faulty
, &rdev
->flags
) ||
2570 rdev
->badblocks
.unacked_exist
) {
2571 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2574 if (test_bit(In_sync
, &rdev
->flags
)) {
2575 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2578 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2579 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2582 if (test_bit(Blocked
, &rdev
->flags
) ||
2583 (rdev
->badblocks
.unacked_exist
2584 && !test_bit(Faulty
, &rdev
->flags
))) {
2585 len
+= sprintf(page
+len
, "%sblocked", sep
);
2588 if (!test_bit(Faulty
, &rdev
->flags
) &&
2589 !test_bit(In_sync
, &rdev
->flags
)) {
2590 len
+= sprintf(page
+len
, "%sspare", sep
);
2593 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2594 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2597 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2598 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2601 if (test_bit(Replacement
, &rdev
->flags
)) {
2602 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2606 return len
+sprintf(page
+len
, "\n");
2610 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2613 * faulty - simulates an error
2614 * remove - disconnects the device
2615 * writemostly - sets write_mostly
2616 * -writemostly - clears write_mostly
2617 * blocked - sets the Blocked flags
2618 * -blocked - clears the Blocked and possibly simulates an error
2619 * insync - sets Insync providing device isn't active
2620 * -insync - clear Insync for a device with a slot assigned,
2621 * so that it gets rebuilt based on bitmap
2622 * write_error - sets WriteErrorSeen
2623 * -write_error - clears WriteErrorSeen
2626 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2627 md_error(rdev
->mddev
, rdev
);
2628 if (test_bit(Faulty
, &rdev
->flags
))
2632 } else if (cmd_match(buf
, "remove")) {
2633 if (rdev
->raid_disk
>= 0)
2636 struct mddev
*mddev
= rdev
->mddev
;
2637 kick_rdev_from_array(rdev
);
2639 md_update_sb(mddev
, 1);
2640 md_new_event(mddev
);
2643 } else if (cmd_match(buf
, "writemostly")) {
2644 set_bit(WriteMostly
, &rdev
->flags
);
2646 } else if (cmd_match(buf
, "-writemostly")) {
2647 clear_bit(WriteMostly
, &rdev
->flags
);
2649 } else if (cmd_match(buf
, "blocked")) {
2650 set_bit(Blocked
, &rdev
->flags
);
2652 } else if (cmd_match(buf
, "-blocked")) {
2653 if (!test_bit(Faulty
, &rdev
->flags
) &&
2654 rdev
->badblocks
.unacked_exist
) {
2655 /* metadata handler doesn't understand badblocks,
2656 * so we need to fail the device
2658 md_error(rdev
->mddev
, rdev
);
2660 clear_bit(Blocked
, &rdev
->flags
);
2661 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2662 wake_up(&rdev
->blocked_wait
);
2663 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2664 md_wakeup_thread(rdev
->mddev
->thread
);
2667 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2668 set_bit(In_sync
, &rdev
->flags
);
2670 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0) {
2671 clear_bit(In_sync
, &rdev
->flags
);
2672 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2673 rdev
->raid_disk
= -1;
2675 } else if (cmd_match(buf
, "write_error")) {
2676 set_bit(WriteErrorSeen
, &rdev
->flags
);
2678 } else if (cmd_match(buf
, "-write_error")) {
2679 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2681 } else if (cmd_match(buf
, "want_replacement")) {
2682 /* Any non-spare device that is not a replacement can
2683 * become want_replacement at any time, but we then need to
2684 * check if recovery is needed.
2686 if (rdev
->raid_disk
>= 0 &&
2687 !test_bit(Replacement
, &rdev
->flags
))
2688 set_bit(WantReplacement
, &rdev
->flags
);
2689 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2690 md_wakeup_thread(rdev
->mddev
->thread
);
2692 } else if (cmd_match(buf
, "-want_replacement")) {
2693 /* Clearing 'want_replacement' is always allowed.
2694 * Once replacements starts it is too late though.
2697 clear_bit(WantReplacement
, &rdev
->flags
);
2698 } else if (cmd_match(buf
, "replacement")) {
2699 /* Can only set a device as a replacement when array has not
2700 * yet been started. Once running, replacement is automatic
2701 * from spares, or by assigning 'slot'.
2703 if (rdev
->mddev
->pers
)
2706 set_bit(Replacement
, &rdev
->flags
);
2709 } else if (cmd_match(buf
, "-replacement")) {
2710 /* Similarly, can only clear Replacement before start */
2711 if (rdev
->mddev
->pers
)
2714 clear_bit(Replacement
, &rdev
->flags
);
2719 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2720 return err
? err
: len
;
2722 static struct rdev_sysfs_entry rdev_state
=
2723 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2726 errors_show(struct md_rdev
*rdev
, char *page
)
2728 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2732 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2735 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2736 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2737 atomic_set(&rdev
->corrected_errors
, n
);
2742 static struct rdev_sysfs_entry rdev_errors
=
2743 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2746 slot_show(struct md_rdev
*rdev
, char *page
)
2748 if (rdev
->raid_disk
< 0)
2749 return sprintf(page
, "none\n");
2751 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2755 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2759 int slot
= simple_strtoul(buf
, &e
, 10);
2760 if (strncmp(buf
, "none", 4)==0)
2762 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2764 if (rdev
->mddev
->pers
&& slot
== -1) {
2765 /* Setting 'slot' on an active array requires also
2766 * updating the 'rd%d' link, and communicating
2767 * with the personality with ->hot_*_disk.
2768 * For now we only support removing
2769 * failed/spare devices. This normally happens automatically,
2770 * but not when the metadata is externally managed.
2772 if (rdev
->raid_disk
== -1)
2774 /* personality does all needed checks */
2775 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2777 clear_bit(Blocked
, &rdev
->flags
);
2778 remove_and_add_spares(rdev
->mddev
, rdev
);
2779 if (rdev
->raid_disk
>= 0)
2781 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2782 md_wakeup_thread(rdev
->mddev
->thread
);
2783 } else if (rdev
->mddev
->pers
) {
2784 /* Activating a spare .. or possibly reactivating
2785 * if we ever get bitmaps working here.
2788 if (rdev
->raid_disk
!= -1)
2791 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2794 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2797 if (slot
>= rdev
->mddev
->raid_disks
&&
2798 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2801 rdev
->raid_disk
= slot
;
2802 if (test_bit(In_sync
, &rdev
->flags
))
2803 rdev
->saved_raid_disk
= slot
;
2805 rdev
->saved_raid_disk
= -1;
2806 clear_bit(In_sync
, &rdev
->flags
);
2807 clear_bit(Bitmap_sync
, &rdev
->flags
);
2808 err
= rdev
->mddev
->pers
->
2809 hot_add_disk(rdev
->mddev
, rdev
);
2811 rdev
->raid_disk
= -1;
2814 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2815 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2816 /* failure here is OK */;
2817 /* don't wakeup anyone, leave that to userspace. */
2819 if (slot
>= rdev
->mddev
->raid_disks
&&
2820 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2822 rdev
->raid_disk
= slot
;
2823 /* assume it is working */
2824 clear_bit(Faulty
, &rdev
->flags
);
2825 clear_bit(WriteMostly
, &rdev
->flags
);
2826 set_bit(In_sync
, &rdev
->flags
);
2827 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2833 static struct rdev_sysfs_entry rdev_slot
=
2834 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2837 offset_show(struct md_rdev
*rdev
, char *page
)
2839 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2843 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2845 unsigned long long offset
;
2846 if (kstrtoull(buf
, 10, &offset
) < 0)
2848 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2850 if (rdev
->sectors
&& rdev
->mddev
->external
)
2851 /* Must set offset before size, so overlap checks
2854 rdev
->data_offset
= offset
;
2855 rdev
->new_data_offset
= offset
;
2859 static struct rdev_sysfs_entry rdev_offset
=
2860 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2862 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2864 return sprintf(page
, "%llu\n",
2865 (unsigned long long)rdev
->new_data_offset
);
2868 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2869 const char *buf
, size_t len
)
2871 unsigned long long new_offset
;
2872 struct mddev
*mddev
= rdev
->mddev
;
2874 if (kstrtoull(buf
, 10, &new_offset
) < 0)
2877 if (mddev
->sync_thread
)
2879 if (new_offset
== rdev
->data_offset
)
2880 /* reset is always permitted */
2882 else if (new_offset
> rdev
->data_offset
) {
2883 /* must not push array size beyond rdev_sectors */
2884 if (new_offset
- rdev
->data_offset
2885 + mddev
->dev_sectors
> rdev
->sectors
)
2888 /* Metadata worries about other space details. */
2890 /* decreasing the offset is inconsistent with a backwards
2893 if (new_offset
< rdev
->data_offset
&&
2894 mddev
->reshape_backwards
)
2896 /* Increasing offset is inconsistent with forwards
2897 * reshape. reshape_direction should be set to
2898 * 'backwards' first.
2900 if (new_offset
> rdev
->data_offset
&&
2901 !mddev
->reshape_backwards
)
2904 if (mddev
->pers
&& mddev
->persistent
&&
2905 !super_types
[mddev
->major_version
]
2906 .allow_new_offset(rdev
, new_offset
))
2908 rdev
->new_data_offset
= new_offset
;
2909 if (new_offset
> rdev
->data_offset
)
2910 mddev
->reshape_backwards
= 1;
2911 else if (new_offset
< rdev
->data_offset
)
2912 mddev
->reshape_backwards
= 0;
2916 static struct rdev_sysfs_entry rdev_new_offset
=
2917 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2920 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2922 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2925 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2927 /* check if two start/length pairs overlap */
2935 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2937 unsigned long long blocks
;
2940 if (kstrtoull(buf
, 10, &blocks
) < 0)
2943 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2944 return -EINVAL
; /* sector conversion overflow */
2947 if (new != blocks
* 2)
2948 return -EINVAL
; /* unsigned long long to sector_t overflow */
2955 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2957 struct mddev
*my_mddev
= rdev
->mddev
;
2958 sector_t oldsectors
= rdev
->sectors
;
2961 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2963 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2964 return -EINVAL
; /* too confusing */
2965 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2966 if (my_mddev
->persistent
) {
2967 sectors
= super_types
[my_mddev
->major_version
].
2968 rdev_size_change(rdev
, sectors
);
2971 } else if (!sectors
)
2972 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2974 if (!my_mddev
->pers
->resize
)
2975 /* Cannot change size for RAID0 or Linear etc */
2978 if (sectors
< my_mddev
->dev_sectors
)
2979 return -EINVAL
; /* component must fit device */
2981 rdev
->sectors
= sectors
;
2982 if (sectors
> oldsectors
&& my_mddev
->external
) {
2983 /* need to check that all other rdevs with the same ->bdev
2984 * do not overlap. We need to unlock the mddev to avoid
2985 * a deadlock. We have already changed rdev->sectors, and if
2986 * we have to change it back, we will have the lock again.
2988 struct mddev
*mddev
;
2990 struct list_head
*tmp
;
2992 mddev_unlock(my_mddev
);
2993 for_each_mddev(mddev
, tmp
) {
2994 struct md_rdev
*rdev2
;
2996 mddev_lock_nointr(mddev
);
2997 rdev_for_each(rdev2
, mddev
)
2998 if (rdev
->bdev
== rdev2
->bdev
&&
3000 overlaps(rdev
->data_offset
, rdev
->sectors
,
3006 mddev_unlock(mddev
);
3012 mddev_lock_nointr(my_mddev
);
3014 /* Someone else could have slipped in a size
3015 * change here, but doing so is just silly.
3016 * We put oldsectors back because we *know* it is
3017 * safe, and trust userspace not to race with
3020 rdev
->sectors
= oldsectors
;
3027 static struct rdev_sysfs_entry rdev_size
=
3028 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3031 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3033 unsigned long long recovery_start
= rdev
->recovery_offset
;
3035 if (test_bit(In_sync
, &rdev
->flags
) ||
3036 recovery_start
== MaxSector
)
3037 return sprintf(page
, "none\n");
3039 return sprintf(page
, "%llu\n", recovery_start
);
3042 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3044 unsigned long long recovery_start
;
3046 if (cmd_match(buf
, "none"))
3047 recovery_start
= MaxSector
;
3048 else if (kstrtoull(buf
, 10, &recovery_start
))
3051 if (rdev
->mddev
->pers
&&
3052 rdev
->raid_disk
>= 0)
3055 rdev
->recovery_offset
= recovery_start
;
3056 if (recovery_start
== MaxSector
)
3057 set_bit(In_sync
, &rdev
->flags
);
3059 clear_bit(In_sync
, &rdev
->flags
);
3063 static struct rdev_sysfs_entry rdev_recovery_start
=
3064 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3068 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
3070 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
3072 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3074 return badblocks_show(&rdev
->badblocks
, page
, 0);
3076 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3078 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3079 /* Maybe that ack was all we needed */
3080 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3081 wake_up(&rdev
->blocked_wait
);
3084 static struct rdev_sysfs_entry rdev_bad_blocks
=
3085 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3088 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3090 return badblocks_show(&rdev
->badblocks
, page
, 1);
3092 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3094 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3096 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3097 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3099 static struct attribute
*rdev_default_attrs
[] = {
3104 &rdev_new_offset
.attr
,
3106 &rdev_recovery_start
.attr
,
3107 &rdev_bad_blocks
.attr
,
3108 &rdev_unack_bad_blocks
.attr
,
3112 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3114 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3115 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3116 struct mddev
*mddev
= rdev
->mddev
;
3122 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
3124 if (rdev
->mddev
== NULL
)
3127 rv
= entry
->show(rdev
, page
);
3128 mddev_unlock(mddev
);
3134 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3135 const char *page
, size_t length
)
3137 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3138 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3140 struct mddev
*mddev
= rdev
->mddev
;
3144 if (!capable(CAP_SYS_ADMIN
))
3146 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3148 if (rdev
->mddev
== NULL
)
3151 rv
= entry
->store(rdev
, page
, length
);
3152 mddev_unlock(mddev
);
3157 static void rdev_free(struct kobject
*ko
)
3159 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3162 static const struct sysfs_ops rdev_sysfs_ops
= {
3163 .show
= rdev_attr_show
,
3164 .store
= rdev_attr_store
,
3166 static struct kobj_type rdev_ktype
= {
3167 .release
= rdev_free
,
3168 .sysfs_ops
= &rdev_sysfs_ops
,
3169 .default_attrs
= rdev_default_attrs
,
3172 int md_rdev_init(struct md_rdev
*rdev
)
3175 rdev
->saved_raid_disk
= -1;
3176 rdev
->raid_disk
= -1;
3178 rdev
->data_offset
= 0;
3179 rdev
->new_data_offset
= 0;
3180 rdev
->sb_events
= 0;
3181 rdev
->last_read_error
.tv_sec
= 0;
3182 rdev
->last_read_error
.tv_nsec
= 0;
3183 rdev
->sb_loaded
= 0;
3184 rdev
->bb_page
= NULL
;
3185 atomic_set(&rdev
->nr_pending
, 0);
3186 atomic_set(&rdev
->read_errors
, 0);
3187 atomic_set(&rdev
->corrected_errors
, 0);
3189 INIT_LIST_HEAD(&rdev
->same_set
);
3190 init_waitqueue_head(&rdev
->blocked_wait
);
3192 /* Add space to store bad block list.
3193 * This reserves the space even on arrays where it cannot
3194 * be used - I wonder if that matters
3196 rdev
->badblocks
.count
= 0;
3197 rdev
->badblocks
.shift
= -1; /* disabled until explicitly enabled */
3198 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3199 seqlock_init(&rdev
->badblocks
.lock
);
3200 if (rdev
->badblocks
.page
== NULL
)
3205 EXPORT_SYMBOL_GPL(md_rdev_init
);
3207 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3209 * mark the device faulty if:
3211 * - the device is nonexistent (zero size)
3212 * - the device has no valid superblock
3214 * a faulty rdev _never_ has rdev->sb set.
3216 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3218 char b
[BDEVNAME_SIZE
];
3220 struct md_rdev
*rdev
;
3223 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3225 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3226 return ERR_PTR(-ENOMEM
);
3229 err
= md_rdev_init(rdev
);
3232 err
= alloc_disk_sb(rdev
);
3236 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3240 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3242 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3245 "md: %s has zero or unknown size, marking faulty!\n",
3246 bdevname(rdev
->bdev
,b
));
3251 if (super_format
>= 0) {
3252 err
= super_types
[super_format
].
3253 load_super(rdev
, NULL
, super_minor
);
3254 if (err
== -EINVAL
) {
3256 "md: %s does not have a valid v%d.%d "
3257 "superblock, not importing!\n",
3258 bdevname(rdev
->bdev
,b
),
3259 super_format
, super_minor
);
3264 "md: could not read %s's sb, not importing!\n",
3265 bdevname(rdev
->bdev
,b
));
3275 md_rdev_clear(rdev
);
3277 return ERR_PTR(err
);
3281 * Check a full RAID array for plausibility
3285 static void analyze_sbs(struct mddev
* mddev
)
3288 struct md_rdev
*rdev
, *freshest
, *tmp
;
3289 char b
[BDEVNAME_SIZE
];
3292 rdev_for_each_safe(rdev
, tmp
, mddev
)
3293 switch (super_types
[mddev
->major_version
].
3294 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3302 "md: fatal superblock inconsistency in %s"
3303 " -- removing from array\n",
3304 bdevname(rdev
->bdev
,b
));
3305 kick_rdev_from_array(rdev
);
3309 super_types
[mddev
->major_version
].
3310 validate_super(mddev
, freshest
);
3313 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3314 if (mddev
->max_disks
&&
3315 (rdev
->desc_nr
>= mddev
->max_disks
||
3316 i
> mddev
->max_disks
)) {
3318 "md: %s: %s: only %d devices permitted\n",
3319 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3321 kick_rdev_from_array(rdev
);
3324 if (rdev
!= freshest
)
3325 if (super_types
[mddev
->major_version
].
3326 validate_super(mddev
, rdev
)) {
3327 printk(KERN_WARNING
"md: kicking non-fresh %s"
3329 bdevname(rdev
->bdev
,b
));
3330 kick_rdev_from_array(rdev
);
3333 if (mddev
->level
== LEVEL_MULTIPATH
) {
3334 rdev
->desc_nr
= i
++;
3335 rdev
->raid_disk
= rdev
->desc_nr
;
3336 set_bit(In_sync
, &rdev
->flags
);
3337 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3338 rdev
->raid_disk
= -1;
3339 clear_bit(In_sync
, &rdev
->flags
);
3344 /* Read a fixed-point number.
3345 * Numbers in sysfs attributes should be in "standard" units where
3346 * possible, so time should be in seconds.
3347 * However we internally use a a much smaller unit such as
3348 * milliseconds or jiffies.
3349 * This function takes a decimal number with a possible fractional
3350 * component, and produces an integer which is the result of
3351 * multiplying that number by 10^'scale'.
3352 * all without any floating-point arithmetic.
3354 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3356 unsigned long result
= 0;
3358 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3361 else if (decimals
< scale
) {
3364 result
= result
* 10 + value
;
3376 while (decimals
< scale
) {
3385 static void md_safemode_timeout(unsigned long data
);
3388 safe_delay_show(struct mddev
*mddev
, char *page
)
3390 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3391 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3394 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3398 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3401 mddev
->safemode_delay
= 0;
3403 unsigned long old_delay
= mddev
->safemode_delay
;
3404 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3405 if (mddev
->safemode_delay
== 0)
3406 mddev
->safemode_delay
= 1;
3407 if (mddev
->safemode_delay
< old_delay
|| old_delay
== 0)
3408 md_safemode_timeout((unsigned long)mddev
);
3412 static struct md_sysfs_entry md_safe_delay
=
3413 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3416 level_show(struct mddev
*mddev
, char *page
)
3418 struct md_personality
*p
= mddev
->pers
;
3420 return sprintf(page
, "%s\n", p
->name
);
3421 else if (mddev
->clevel
[0])
3422 return sprintf(page
, "%s\n", mddev
->clevel
);
3423 else if (mddev
->level
!= LEVEL_NONE
)
3424 return sprintf(page
, "%d\n", mddev
->level
);
3430 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3434 struct md_personality
*pers
;
3437 struct md_rdev
*rdev
;
3439 if (mddev
->pers
== NULL
) {
3442 if (len
>= sizeof(mddev
->clevel
))
3444 strncpy(mddev
->clevel
, buf
, len
);
3445 if (mddev
->clevel
[len
-1] == '\n')
3447 mddev
->clevel
[len
] = 0;
3448 mddev
->level
= LEVEL_NONE
;
3454 /* request to change the personality. Need to ensure:
3455 * - array is not engaged in resync/recovery/reshape
3456 * - old personality can be suspended
3457 * - new personality will access other array.
3460 if (mddev
->sync_thread
||
3461 mddev
->reshape_position
!= MaxSector
||
3462 mddev
->sysfs_active
)
3465 if (!mddev
->pers
->quiesce
) {
3466 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3467 mdname(mddev
), mddev
->pers
->name
);
3471 /* Now find the new personality */
3472 if (len
== 0 || len
>= sizeof(clevel
))
3474 strncpy(clevel
, buf
, len
);
3475 if (clevel
[len
-1] == '\n')
3478 if (kstrtol(clevel
, 10, &level
))
3481 if (request_module("md-%s", clevel
) != 0)
3482 request_module("md-level-%s", clevel
);
3483 spin_lock(&pers_lock
);
3484 pers
= find_pers(level
, clevel
);
3485 if (!pers
|| !try_module_get(pers
->owner
)) {
3486 spin_unlock(&pers_lock
);
3487 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3490 spin_unlock(&pers_lock
);
3492 if (pers
== mddev
->pers
) {
3493 /* Nothing to do! */
3494 module_put(pers
->owner
);
3497 if (!pers
->takeover
) {
3498 module_put(pers
->owner
);
3499 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3500 mdname(mddev
), clevel
);
3504 rdev_for_each(rdev
, mddev
)
3505 rdev
->new_raid_disk
= rdev
->raid_disk
;
3507 /* ->takeover must set new_* and/or delta_disks
3508 * if it succeeds, and may set them when it fails.
3510 priv
= pers
->takeover(mddev
);
3512 mddev
->new_level
= mddev
->level
;
3513 mddev
->new_layout
= mddev
->layout
;
3514 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3515 mddev
->raid_disks
-= mddev
->delta_disks
;
3516 mddev
->delta_disks
= 0;
3517 mddev
->reshape_backwards
= 0;
3518 module_put(pers
->owner
);
3519 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3520 mdname(mddev
), clevel
);
3521 return PTR_ERR(priv
);
3524 /* Looks like we have a winner */
3525 mddev_suspend(mddev
);
3526 mddev
->pers
->stop(mddev
);
3528 if (mddev
->pers
->sync_request
== NULL
&&
3529 pers
->sync_request
!= NULL
) {
3530 /* need to add the md_redundancy_group */
3531 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3533 "md: cannot register extra attributes for %s\n",
3535 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3537 if (mddev
->pers
->sync_request
!= NULL
&&
3538 pers
->sync_request
== NULL
) {
3539 /* need to remove the md_redundancy_group */
3540 if (mddev
->to_remove
== NULL
)
3541 mddev
->to_remove
= &md_redundancy_group
;
3544 if (mddev
->pers
->sync_request
== NULL
&&
3546 /* We are converting from a no-redundancy array
3547 * to a redundancy array and metadata is managed
3548 * externally so we need to be sure that writes
3549 * won't block due to a need to transition
3551 * until external management is started.
3554 mddev
->safemode_delay
= 0;
3555 mddev
->safemode
= 0;
3558 rdev_for_each(rdev
, mddev
) {
3559 if (rdev
->raid_disk
< 0)
3561 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3562 rdev
->new_raid_disk
= -1;
3563 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3565 sysfs_unlink_rdev(mddev
, rdev
);
3567 rdev_for_each(rdev
, mddev
) {
3568 if (rdev
->raid_disk
< 0)
3570 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3572 rdev
->raid_disk
= rdev
->new_raid_disk
;
3573 if (rdev
->raid_disk
< 0)
3574 clear_bit(In_sync
, &rdev
->flags
);
3576 if (sysfs_link_rdev(mddev
, rdev
))
3577 printk(KERN_WARNING
"md: cannot register rd%d"
3578 " for %s after level change\n",
3579 rdev
->raid_disk
, mdname(mddev
));
3583 module_put(mddev
->pers
->owner
);
3585 mddev
->private = priv
;
3586 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3587 mddev
->level
= mddev
->new_level
;
3588 mddev
->layout
= mddev
->new_layout
;
3589 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3590 mddev
->delta_disks
= 0;
3591 mddev
->reshape_backwards
= 0;
3592 mddev
->degraded
= 0;
3593 if (mddev
->pers
->sync_request
== NULL
) {
3594 /* this is now an array without redundancy, so
3595 * it must always be in_sync
3598 del_timer_sync(&mddev
->safemode_timer
);
3600 blk_set_stacking_limits(&mddev
->queue
->limits
);
3602 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3603 mddev_resume(mddev
);
3605 md_update_sb(mddev
, 1);
3606 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3607 md_new_event(mddev
);
3611 static struct md_sysfs_entry md_level
=
3612 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3616 layout_show(struct mddev
*mddev
, char *page
)
3618 /* just a number, not meaningful for all levels */
3619 if (mddev
->reshape_position
!= MaxSector
&&
3620 mddev
->layout
!= mddev
->new_layout
)
3621 return sprintf(page
, "%d (%d)\n",
3622 mddev
->new_layout
, mddev
->layout
);
3623 return sprintf(page
, "%d\n", mddev
->layout
);
3627 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3630 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3632 if (!*buf
|| (*e
&& *e
!= '\n'))
3637 if (mddev
->pers
->check_reshape
== NULL
)
3641 mddev
->new_layout
= n
;
3642 err
= mddev
->pers
->check_reshape(mddev
);
3644 mddev
->new_layout
= mddev
->layout
;
3648 mddev
->new_layout
= n
;
3649 if (mddev
->reshape_position
== MaxSector
)
3654 static struct md_sysfs_entry md_layout
=
3655 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3659 raid_disks_show(struct mddev
*mddev
, char *page
)
3661 if (mddev
->raid_disks
== 0)
3663 if (mddev
->reshape_position
!= MaxSector
&&
3664 mddev
->delta_disks
!= 0)
3665 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3666 mddev
->raid_disks
- mddev
->delta_disks
);
3667 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3670 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3673 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3677 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3679 if (!*buf
|| (*e
&& *e
!= '\n'))
3683 rv
= update_raid_disks(mddev
, n
);
3684 else if (mddev
->reshape_position
!= MaxSector
) {
3685 struct md_rdev
*rdev
;
3686 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3688 rdev_for_each(rdev
, mddev
) {
3690 rdev
->data_offset
< rdev
->new_data_offset
)
3693 rdev
->data_offset
> rdev
->new_data_offset
)
3696 mddev
->delta_disks
= n
- olddisks
;
3697 mddev
->raid_disks
= n
;
3698 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3700 mddev
->raid_disks
= n
;
3701 return rv
? rv
: len
;
3703 static struct md_sysfs_entry md_raid_disks
=
3704 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3707 chunk_size_show(struct mddev
*mddev
, char *page
)
3709 if (mddev
->reshape_position
!= MaxSector
&&
3710 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3711 return sprintf(page
, "%d (%d)\n",
3712 mddev
->new_chunk_sectors
<< 9,
3713 mddev
->chunk_sectors
<< 9);
3714 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3718 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3721 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3723 if (!*buf
|| (*e
&& *e
!= '\n'))
3728 if (mddev
->pers
->check_reshape
== NULL
)
3732 mddev
->new_chunk_sectors
= n
>> 9;
3733 err
= mddev
->pers
->check_reshape(mddev
);
3735 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3739 mddev
->new_chunk_sectors
= n
>> 9;
3740 if (mddev
->reshape_position
== MaxSector
)
3741 mddev
->chunk_sectors
= n
>> 9;
3745 static struct md_sysfs_entry md_chunk_size
=
3746 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3749 resync_start_show(struct mddev
*mddev
, char *page
)
3751 if (mddev
->recovery_cp
== MaxSector
)
3752 return sprintf(page
, "none\n");
3753 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3757 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3760 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3762 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3764 if (cmd_match(buf
, "none"))
3766 else if (!*buf
|| (*e
&& *e
!= '\n'))
3769 mddev
->recovery_cp
= n
;
3771 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3774 static struct md_sysfs_entry md_resync_start
=
3775 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3778 * The array state can be:
3781 * No devices, no size, no level
3782 * Equivalent to STOP_ARRAY ioctl
3784 * May have some settings, but array is not active
3785 * all IO results in error
3786 * When written, doesn't tear down array, but just stops it
3787 * suspended (not supported yet)
3788 * All IO requests will block. The array can be reconfigured.
3789 * Writing this, if accepted, will block until array is quiescent
3791 * no resync can happen. no superblocks get written.
3792 * write requests fail
3794 * like readonly, but behaves like 'clean' on a write request.
3796 * clean - no pending writes, but otherwise active.
3797 * When written to inactive array, starts without resync
3798 * If a write request arrives then
3799 * if metadata is known, mark 'dirty' and switch to 'active'.
3800 * if not known, block and switch to write-pending
3801 * If written to an active array that has pending writes, then fails.
3803 * fully active: IO and resync can be happening.
3804 * When written to inactive array, starts with resync
3807 * clean, but writes are blocked waiting for 'active' to be written.
3810 * like active, but no writes have been seen for a while (100msec).
3813 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3814 write_pending
, active_idle
, bad_word
};
3815 static char *array_states
[] = {
3816 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3817 "write-pending", "active-idle", NULL
};
3819 static int match_word(const char *word
, char **list
)
3822 for (n
=0; list
[n
]; n
++)
3823 if (cmd_match(word
, list
[n
]))
3829 array_state_show(struct mddev
*mddev
, char *page
)
3831 enum array_state st
= inactive
;
3844 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3846 else if (mddev
->safemode
)
3852 if (list_empty(&mddev
->disks
) &&
3853 mddev
->raid_disks
== 0 &&
3854 mddev
->dev_sectors
== 0)
3859 return sprintf(page
, "%s\n", array_states
[st
]);
3862 static int do_md_stop(struct mddev
* mddev
, int ro
, struct block_device
*bdev
);
3863 static int md_set_readonly(struct mddev
* mddev
, struct block_device
*bdev
);
3864 static int do_md_run(struct mddev
* mddev
);
3865 static int restart_array(struct mddev
*mddev
);
3868 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3871 enum array_state st
= match_word(buf
, array_states
);
3876 /* stopping an active array */
3877 err
= do_md_stop(mddev
, 0, NULL
);
3880 /* stopping an active array */
3882 err
= do_md_stop(mddev
, 2, NULL
);
3884 err
= 0; /* already inactive */
3887 break; /* not supported yet */
3890 err
= md_set_readonly(mddev
, NULL
);
3893 set_disk_ro(mddev
->gendisk
, 1);
3894 err
= do_md_run(mddev
);
3900 err
= md_set_readonly(mddev
, NULL
);
3901 else if (mddev
->ro
== 1)
3902 err
= restart_array(mddev
);
3905 set_disk_ro(mddev
->gendisk
, 0);
3909 err
= do_md_run(mddev
);
3914 restart_array(mddev
);
3915 spin_lock_irq(&mddev
->write_lock
);
3916 if (atomic_read(&mddev
->writes_pending
) == 0) {
3917 if (mddev
->in_sync
== 0) {
3919 if (mddev
->safemode
== 1)
3920 mddev
->safemode
= 0;
3921 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3926 spin_unlock_irq(&mddev
->write_lock
);
3932 restart_array(mddev
);
3933 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3934 wake_up(&mddev
->sb_wait
);
3938 set_disk_ro(mddev
->gendisk
, 0);
3939 err
= do_md_run(mddev
);
3944 /* these cannot be set */
3950 if (mddev
->hold_active
== UNTIL_IOCTL
)
3951 mddev
->hold_active
= 0;
3952 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3956 static struct md_sysfs_entry md_array_state
=
3957 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3960 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3961 return sprintf(page
, "%d\n",
3962 atomic_read(&mddev
->max_corr_read_errors
));
3966 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3969 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3971 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3972 atomic_set(&mddev
->max_corr_read_errors
, n
);
3978 static struct md_sysfs_entry max_corr_read_errors
=
3979 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3980 max_corrected_read_errors_store
);
3983 null_show(struct mddev
*mddev
, char *page
)
3989 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3991 /* buf must be %d:%d\n? giving major and minor numbers */
3992 /* The new device is added to the array.
3993 * If the array has a persistent superblock, we read the
3994 * superblock to initialise info and check validity.
3995 * Otherwise, only checking done is that in bind_rdev_to_array,
3996 * which mainly checks size.
3999 int major
= simple_strtoul(buf
, &e
, 10);
4002 struct md_rdev
*rdev
;
4005 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4007 minor
= simple_strtoul(e
+1, &e
, 10);
4008 if (*e
&& *e
!= '\n')
4010 dev
= MKDEV(major
, minor
);
4011 if (major
!= MAJOR(dev
) ||
4012 minor
!= MINOR(dev
))
4016 if (mddev
->persistent
) {
4017 rdev
= md_import_device(dev
, mddev
->major_version
,
4018 mddev
->minor_version
);
4019 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4020 struct md_rdev
*rdev0
4021 = list_entry(mddev
->disks
.next
,
4022 struct md_rdev
, same_set
);
4023 err
= super_types
[mddev
->major_version
]
4024 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4028 } else if (mddev
->external
)
4029 rdev
= md_import_device(dev
, -2, -1);
4031 rdev
= md_import_device(dev
, -1, -1);
4034 return PTR_ERR(rdev
);
4035 err
= bind_rdev_to_array(rdev
, mddev
);
4039 return err
? err
: len
;
4042 static struct md_sysfs_entry md_new_device
=
4043 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4046 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4049 unsigned long chunk
, end_chunk
;
4053 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4055 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4056 if (buf
== end
) break;
4057 if (*end
== '-') { /* range */
4059 end_chunk
= simple_strtoul(buf
, &end
, 0);
4060 if (buf
== end
) break;
4062 if (*end
&& !isspace(*end
)) break;
4063 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4064 buf
= skip_spaces(end
);
4066 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4071 static struct md_sysfs_entry md_bitmap
=
4072 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4075 size_show(struct mddev
*mddev
, char *page
)
4077 return sprintf(page
, "%llu\n",
4078 (unsigned long long)mddev
->dev_sectors
/ 2);
4081 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4084 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4086 /* If array is inactive, we can reduce the component size, but
4087 * not increase it (except from 0).
4088 * If array is active, we can try an on-line resize
4091 int err
= strict_blocks_to_sectors(buf
, §ors
);
4096 err
= update_size(mddev
, sectors
);
4097 md_update_sb(mddev
, 1);
4099 if (mddev
->dev_sectors
== 0 ||
4100 mddev
->dev_sectors
> sectors
)
4101 mddev
->dev_sectors
= sectors
;
4105 return err
? err
: len
;
4108 static struct md_sysfs_entry md_size
=
4109 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4112 /* Metadata version.
4114 * 'none' for arrays with no metadata (good luck...)
4115 * 'external' for arrays with externally managed metadata,
4116 * or N.M for internally known formats
4119 metadata_show(struct mddev
*mddev
, char *page
)
4121 if (mddev
->persistent
)
4122 return sprintf(page
, "%d.%d\n",
4123 mddev
->major_version
, mddev
->minor_version
);
4124 else if (mddev
->external
)
4125 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4127 return sprintf(page
, "none\n");
4131 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4135 /* Changing the details of 'external' metadata is
4136 * always permitted. Otherwise there must be
4137 * no devices attached to the array.
4139 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4141 else if (!list_empty(&mddev
->disks
))
4144 if (cmd_match(buf
, "none")) {
4145 mddev
->persistent
= 0;
4146 mddev
->external
= 0;
4147 mddev
->major_version
= 0;
4148 mddev
->minor_version
= 90;
4151 if (strncmp(buf
, "external:", 9) == 0) {
4152 size_t namelen
= len
-9;
4153 if (namelen
>= sizeof(mddev
->metadata_type
))
4154 namelen
= sizeof(mddev
->metadata_type
)-1;
4155 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4156 mddev
->metadata_type
[namelen
] = 0;
4157 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4158 mddev
->metadata_type
[--namelen
] = 0;
4159 mddev
->persistent
= 0;
4160 mddev
->external
= 1;
4161 mddev
->major_version
= 0;
4162 mddev
->minor_version
= 90;
4165 major
= simple_strtoul(buf
, &e
, 10);
4166 if (e
==buf
|| *e
!= '.')
4169 minor
= simple_strtoul(buf
, &e
, 10);
4170 if (e
==buf
|| (*e
&& *e
!= '\n') )
4172 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4174 mddev
->major_version
= major
;
4175 mddev
->minor_version
= minor
;
4176 mddev
->persistent
= 1;
4177 mddev
->external
= 0;
4181 static struct md_sysfs_entry md_metadata
=
4182 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4185 action_show(struct mddev
*mddev
, char *page
)
4187 char *type
= "idle";
4188 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4190 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4191 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4192 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4194 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4195 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4197 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4201 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4204 return sprintf(page
, "%s\n", type
);
4208 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4210 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4213 if (cmd_match(page
, "frozen"))
4214 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4216 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4218 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4219 if (mddev
->sync_thread
) {
4220 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4221 md_reap_sync_thread(mddev
);
4223 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4224 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4226 else if (cmd_match(page
, "resync"))
4227 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4228 else if (cmd_match(page
, "recover")) {
4229 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4230 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4231 } else if (cmd_match(page
, "reshape")) {
4233 if (mddev
->pers
->start_reshape
== NULL
)
4235 err
= mddev
->pers
->start_reshape(mddev
);
4238 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4240 if (cmd_match(page
, "check"))
4241 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4242 else if (!cmd_match(page
, "repair"))
4244 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4245 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4247 if (mddev
->ro
== 2) {
4248 /* A write to sync_action is enough to justify
4249 * canceling read-auto mode
4252 md_wakeup_thread(mddev
->sync_thread
);
4254 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4255 md_wakeup_thread(mddev
->thread
);
4256 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4260 static struct md_sysfs_entry md_scan_mode
=
4261 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4264 last_sync_action_show(struct mddev
*mddev
, char *page
)
4266 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4269 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4272 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4274 return sprintf(page
, "%llu\n",
4275 (unsigned long long)
4276 atomic64_read(&mddev
->resync_mismatches
));
4279 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4282 sync_min_show(struct mddev
*mddev
, char *page
)
4284 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4285 mddev
->sync_speed_min
? "local": "system");
4289 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4293 if (strncmp(buf
, "system", 6)==0) {
4294 mddev
->sync_speed_min
= 0;
4297 min
= simple_strtoul(buf
, &e
, 10);
4298 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4300 mddev
->sync_speed_min
= min
;
4304 static struct md_sysfs_entry md_sync_min
=
4305 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4308 sync_max_show(struct mddev
*mddev
, char *page
)
4310 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4311 mddev
->sync_speed_max
? "local": "system");
4315 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4319 if (strncmp(buf
, "system", 6)==0) {
4320 mddev
->sync_speed_max
= 0;
4323 max
= simple_strtoul(buf
, &e
, 10);
4324 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4326 mddev
->sync_speed_max
= max
;
4330 static struct md_sysfs_entry md_sync_max
=
4331 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4334 degraded_show(struct mddev
*mddev
, char *page
)
4336 return sprintf(page
, "%d\n", mddev
->degraded
);
4338 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4341 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4343 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4347 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4351 if (kstrtol(buf
, 10, &n
))
4354 if (n
!= 0 && n
!= 1)
4357 mddev
->parallel_resync
= n
;
4359 if (mddev
->sync_thread
)
4360 wake_up(&resync_wait
);
4365 /* force parallel resync, even with shared block devices */
4366 static struct md_sysfs_entry md_sync_force_parallel
=
4367 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4368 sync_force_parallel_show
, sync_force_parallel_store
);
4371 sync_speed_show(struct mddev
*mddev
, char *page
)
4373 unsigned long resync
, dt
, db
;
4374 if (mddev
->curr_resync
== 0)
4375 return sprintf(page
, "none\n");
4376 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4377 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4379 db
= resync
- mddev
->resync_mark_cnt
;
4380 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4383 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4386 sync_completed_show(struct mddev
*mddev
, char *page
)
4388 unsigned long long max_sectors
, resync
;
4390 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4391 return sprintf(page
, "none\n");
4393 if (mddev
->curr_resync
== 1 ||
4394 mddev
->curr_resync
== 2)
4395 return sprintf(page
, "delayed\n");
4397 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4398 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4399 max_sectors
= mddev
->resync_max_sectors
;
4401 max_sectors
= mddev
->dev_sectors
;
4403 resync
= mddev
->curr_resync_completed
;
4404 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4407 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4410 min_sync_show(struct mddev
*mddev
, char *page
)
4412 return sprintf(page
, "%llu\n",
4413 (unsigned long long)mddev
->resync_min
);
4416 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4418 unsigned long long min
;
4419 if (kstrtoull(buf
, 10, &min
))
4421 if (min
> mddev
->resync_max
)
4423 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4426 /* Must be a multiple of chunk_size */
4427 if (mddev
->chunk_sectors
) {
4428 sector_t temp
= min
;
4429 if (sector_div(temp
, mddev
->chunk_sectors
))
4432 mddev
->resync_min
= min
;
4437 static struct md_sysfs_entry md_min_sync
=
4438 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4441 max_sync_show(struct mddev
*mddev
, char *page
)
4443 if (mddev
->resync_max
== MaxSector
)
4444 return sprintf(page
, "max\n");
4446 return sprintf(page
, "%llu\n",
4447 (unsigned long long)mddev
->resync_max
);
4450 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4452 if (strncmp(buf
, "max", 3) == 0)
4453 mddev
->resync_max
= MaxSector
;
4455 unsigned long long max
;
4456 if (kstrtoull(buf
, 10, &max
))
4458 if (max
< mddev
->resync_min
)
4460 if (max
< mddev
->resync_max
&&
4462 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4465 /* Must be a multiple of chunk_size */
4466 if (mddev
->chunk_sectors
) {
4467 sector_t temp
= max
;
4468 if (sector_div(temp
, mddev
->chunk_sectors
))
4471 mddev
->resync_max
= max
;
4473 wake_up(&mddev
->recovery_wait
);
4477 static struct md_sysfs_entry md_max_sync
=
4478 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4481 suspend_lo_show(struct mddev
*mddev
, char *page
)
4483 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4487 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4490 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4491 unsigned long long old
= mddev
->suspend_lo
;
4493 if (mddev
->pers
== NULL
||
4494 mddev
->pers
->quiesce
== NULL
)
4496 if (buf
== e
|| (*e
&& *e
!= '\n'))
4499 mddev
->suspend_lo
= new;
4501 /* Shrinking suspended region */
4502 mddev
->pers
->quiesce(mddev
, 2);
4504 /* Expanding suspended region - need to wait */
4505 mddev
->pers
->quiesce(mddev
, 1);
4506 mddev
->pers
->quiesce(mddev
, 0);
4510 static struct md_sysfs_entry md_suspend_lo
=
4511 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4515 suspend_hi_show(struct mddev
*mddev
, char *page
)
4517 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4521 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4524 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4525 unsigned long long old
= mddev
->suspend_hi
;
4527 if (mddev
->pers
== NULL
||
4528 mddev
->pers
->quiesce
== NULL
)
4530 if (buf
== e
|| (*e
&& *e
!= '\n'))
4533 mddev
->suspend_hi
= new;
4535 /* Shrinking suspended region */
4536 mddev
->pers
->quiesce(mddev
, 2);
4538 /* Expanding suspended region - need to wait */
4539 mddev
->pers
->quiesce(mddev
, 1);
4540 mddev
->pers
->quiesce(mddev
, 0);
4544 static struct md_sysfs_entry md_suspend_hi
=
4545 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4548 reshape_position_show(struct mddev
*mddev
, char *page
)
4550 if (mddev
->reshape_position
!= MaxSector
)
4551 return sprintf(page
, "%llu\n",
4552 (unsigned long long)mddev
->reshape_position
);
4553 strcpy(page
, "none\n");
4558 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4560 struct md_rdev
*rdev
;
4562 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4565 if (buf
== e
|| (*e
&& *e
!= '\n'))
4567 mddev
->reshape_position
= new;
4568 mddev
->delta_disks
= 0;
4569 mddev
->reshape_backwards
= 0;
4570 mddev
->new_level
= mddev
->level
;
4571 mddev
->new_layout
= mddev
->layout
;
4572 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4573 rdev_for_each(rdev
, mddev
)
4574 rdev
->new_data_offset
= rdev
->data_offset
;
4578 static struct md_sysfs_entry md_reshape_position
=
4579 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4580 reshape_position_store
);
4583 reshape_direction_show(struct mddev
*mddev
, char *page
)
4585 return sprintf(page
, "%s\n",
4586 mddev
->reshape_backwards
? "backwards" : "forwards");
4590 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4593 if (cmd_match(buf
, "forwards"))
4595 else if (cmd_match(buf
, "backwards"))
4599 if (mddev
->reshape_backwards
== backwards
)
4602 /* check if we are allowed to change */
4603 if (mddev
->delta_disks
)
4606 if (mddev
->persistent
&&
4607 mddev
->major_version
== 0)
4610 mddev
->reshape_backwards
= backwards
;
4614 static struct md_sysfs_entry md_reshape_direction
=
4615 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4616 reshape_direction_store
);
4619 array_size_show(struct mddev
*mddev
, char *page
)
4621 if (mddev
->external_size
)
4622 return sprintf(page
, "%llu\n",
4623 (unsigned long long)mddev
->array_sectors
/2);
4625 return sprintf(page
, "default\n");
4629 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4633 if (strncmp(buf
, "default", 7) == 0) {
4635 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4637 sectors
= mddev
->array_sectors
;
4639 mddev
->external_size
= 0;
4641 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4643 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4646 mddev
->external_size
= 1;
4649 mddev
->array_sectors
= sectors
;
4651 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4652 revalidate_disk(mddev
->gendisk
);
4657 static struct md_sysfs_entry md_array_size
=
4658 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4661 static struct attribute
*md_default_attrs
[] = {
4664 &md_raid_disks
.attr
,
4665 &md_chunk_size
.attr
,
4667 &md_resync_start
.attr
,
4669 &md_new_device
.attr
,
4670 &md_safe_delay
.attr
,
4671 &md_array_state
.attr
,
4672 &md_reshape_position
.attr
,
4673 &md_reshape_direction
.attr
,
4674 &md_array_size
.attr
,
4675 &max_corr_read_errors
.attr
,
4679 static struct attribute
*md_redundancy_attrs
[] = {
4681 &md_last_scan_mode
.attr
,
4682 &md_mismatches
.attr
,
4685 &md_sync_speed
.attr
,
4686 &md_sync_force_parallel
.attr
,
4687 &md_sync_completed
.attr
,
4690 &md_suspend_lo
.attr
,
4691 &md_suspend_hi
.attr
,
4696 static struct attribute_group md_redundancy_group
= {
4698 .attrs
= md_redundancy_attrs
,
4703 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4705 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4706 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4711 spin_lock(&all_mddevs_lock
);
4712 if (list_empty(&mddev
->all_mddevs
)) {
4713 spin_unlock(&all_mddevs_lock
);
4717 spin_unlock(&all_mddevs_lock
);
4719 rv
= mddev_lock(mddev
);
4721 rv
= entry
->show(mddev
, page
);
4722 mddev_unlock(mddev
);
4729 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4730 const char *page
, size_t length
)
4732 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4733 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4738 if (!capable(CAP_SYS_ADMIN
))
4740 spin_lock(&all_mddevs_lock
);
4741 if (list_empty(&mddev
->all_mddevs
)) {
4742 spin_unlock(&all_mddevs_lock
);
4746 spin_unlock(&all_mddevs_lock
);
4747 if (entry
->store
== new_dev_store
)
4748 flush_workqueue(md_misc_wq
);
4749 rv
= mddev_lock(mddev
);
4751 rv
= entry
->store(mddev
, page
, length
);
4752 mddev_unlock(mddev
);
4758 static void md_free(struct kobject
*ko
)
4760 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4762 if (mddev
->sysfs_state
)
4763 sysfs_put(mddev
->sysfs_state
);
4765 if (mddev
->gendisk
) {
4766 del_gendisk(mddev
->gendisk
);
4767 put_disk(mddev
->gendisk
);
4770 blk_cleanup_queue(mddev
->queue
);
4775 static const struct sysfs_ops md_sysfs_ops
= {
4776 .show
= md_attr_show
,
4777 .store
= md_attr_store
,
4779 static struct kobj_type md_ktype
= {
4781 .sysfs_ops
= &md_sysfs_ops
,
4782 .default_attrs
= md_default_attrs
,
4787 static void mddev_delayed_delete(struct work_struct
*ws
)
4789 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4791 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4792 kobject_del(&mddev
->kobj
);
4793 kobject_put(&mddev
->kobj
);
4796 static int md_alloc(dev_t dev
, char *name
)
4798 static DEFINE_MUTEX(disks_mutex
);
4799 struct mddev
*mddev
= mddev_find(dev
);
4800 struct gendisk
*disk
;
4809 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4810 shift
= partitioned
? MdpMinorShift
: 0;
4811 unit
= MINOR(mddev
->unit
) >> shift
;
4813 /* wait for any previous instance of this device to be
4814 * completely removed (mddev_delayed_delete).
4816 flush_workqueue(md_misc_wq
);
4818 mutex_lock(&disks_mutex
);
4824 /* Need to ensure that 'name' is not a duplicate.
4826 struct mddev
*mddev2
;
4827 spin_lock(&all_mddevs_lock
);
4829 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4830 if (mddev2
->gendisk
&&
4831 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4832 spin_unlock(&all_mddevs_lock
);
4835 spin_unlock(&all_mddevs_lock
);
4839 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4842 mddev
->queue
->queuedata
= mddev
;
4844 blk_queue_make_request(mddev
->queue
, md_make_request
);
4845 blk_set_stacking_limits(&mddev
->queue
->limits
);
4847 disk
= alloc_disk(1 << shift
);
4849 blk_cleanup_queue(mddev
->queue
);
4850 mddev
->queue
= NULL
;
4853 disk
->major
= MAJOR(mddev
->unit
);
4854 disk
->first_minor
= unit
<< shift
;
4856 strcpy(disk
->disk_name
, name
);
4857 else if (partitioned
)
4858 sprintf(disk
->disk_name
, "md_d%d", unit
);
4860 sprintf(disk
->disk_name
, "md%d", unit
);
4861 disk
->fops
= &md_fops
;
4862 disk
->private_data
= mddev
;
4863 disk
->queue
= mddev
->queue
;
4864 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4865 /* Allow extended partitions. This makes the
4866 * 'mdp' device redundant, but we can't really
4869 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4870 mddev
->gendisk
= disk
;
4871 /* As soon as we call add_disk(), another thread could get
4872 * through to md_open, so make sure it doesn't get too far
4874 mutex_lock(&mddev
->open_mutex
);
4877 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4878 &disk_to_dev(disk
)->kobj
, "%s", "md");
4880 /* This isn't possible, but as kobject_init_and_add is marked
4881 * __must_check, we must do something with the result
4883 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4887 if (mddev
->kobj
.sd
&&
4888 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4889 printk(KERN_DEBUG
"pointless warning\n");
4890 mutex_unlock(&mddev
->open_mutex
);
4892 mutex_unlock(&disks_mutex
);
4893 if (!error
&& mddev
->kobj
.sd
) {
4894 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4895 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4901 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4903 md_alloc(dev
, NULL
);
4907 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4909 /* val must be "md_*" where * is not all digits.
4910 * We allocate an array with a large free minor number, and
4911 * set the name to val. val must not already be an active name.
4913 int len
= strlen(val
);
4914 char buf
[DISK_NAME_LEN
];
4916 while (len
&& val
[len
-1] == '\n')
4918 if (len
>= DISK_NAME_LEN
)
4920 strlcpy(buf
, val
, len
+1);
4921 if (strncmp(buf
, "md_", 3) != 0)
4923 return md_alloc(0, buf
);
4926 static void md_safemode_timeout(unsigned long data
)
4928 struct mddev
*mddev
= (struct mddev
*) data
;
4930 if (!atomic_read(&mddev
->writes_pending
)) {
4931 mddev
->safemode
= 1;
4932 if (mddev
->external
)
4933 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4935 md_wakeup_thread(mddev
->thread
);
4938 static int start_dirty_degraded
;
4940 int md_run(struct mddev
*mddev
)
4943 struct md_rdev
*rdev
;
4944 struct md_personality
*pers
;
4946 if (list_empty(&mddev
->disks
))
4947 /* cannot run an array with no devices.. */
4952 /* Cannot run until previous stop completes properly */
4953 if (mddev
->sysfs_active
)
4957 * Analyze all RAID superblock(s)
4959 if (!mddev
->raid_disks
) {
4960 if (!mddev
->persistent
)
4965 if (mddev
->level
!= LEVEL_NONE
)
4966 request_module("md-level-%d", mddev
->level
);
4967 else if (mddev
->clevel
[0])
4968 request_module("md-%s", mddev
->clevel
);
4971 * Drop all container device buffers, from now on
4972 * the only valid external interface is through the md
4975 rdev_for_each(rdev
, mddev
) {
4976 if (test_bit(Faulty
, &rdev
->flags
))
4978 sync_blockdev(rdev
->bdev
);
4979 invalidate_bdev(rdev
->bdev
);
4981 /* perform some consistency tests on the device.
4982 * We don't want the data to overlap the metadata,
4983 * Internal Bitmap issues have been handled elsewhere.
4985 if (rdev
->meta_bdev
) {
4986 /* Nothing to check */;
4987 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4988 if (mddev
->dev_sectors
&&
4989 rdev
->data_offset
+ mddev
->dev_sectors
4991 printk("md: %s: data overlaps metadata\n",
4996 if (rdev
->sb_start
+ rdev
->sb_size
/512
4997 > rdev
->data_offset
) {
4998 printk("md: %s: metadata overlaps data\n",
5003 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5006 if (mddev
->bio_set
== NULL
)
5007 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
5009 spin_lock(&pers_lock
);
5010 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5011 if (!pers
|| !try_module_get(pers
->owner
)) {
5012 spin_unlock(&pers_lock
);
5013 if (mddev
->level
!= LEVEL_NONE
)
5014 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5017 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5022 spin_unlock(&pers_lock
);
5023 if (mddev
->level
!= pers
->level
) {
5024 mddev
->level
= pers
->level
;
5025 mddev
->new_level
= pers
->level
;
5027 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5029 if (mddev
->reshape_position
!= MaxSector
&&
5030 pers
->start_reshape
== NULL
) {
5031 /* This personality cannot handle reshaping... */
5033 module_put(pers
->owner
);
5037 if (pers
->sync_request
) {
5038 /* Warn if this is a potentially silly
5041 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5042 struct md_rdev
*rdev2
;
5045 rdev_for_each(rdev
, mddev
)
5046 rdev_for_each(rdev2
, mddev
) {
5048 rdev
->bdev
->bd_contains
==
5049 rdev2
->bdev
->bd_contains
) {
5051 "%s: WARNING: %s appears to be"
5052 " on the same physical disk as"
5055 bdevname(rdev
->bdev
,b
),
5056 bdevname(rdev2
->bdev
,b2
));
5063 "True protection against single-disk"
5064 " failure might be compromised.\n");
5067 mddev
->recovery
= 0;
5068 /* may be over-ridden by personality */
5069 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5071 mddev
->ok_start_degraded
= start_dirty_degraded
;
5073 if (start_readonly
&& mddev
->ro
== 0)
5074 mddev
->ro
= 2; /* read-only, but switch on first write */
5076 err
= mddev
->pers
->run(mddev
);
5078 printk(KERN_ERR
"md: pers->run() failed ...\n");
5079 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5080 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5081 " but 'external_size' not in effect?\n", __func__
);
5083 "md: invalid array_size %llu > default size %llu\n",
5084 (unsigned long long)mddev
->array_sectors
/ 2,
5085 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
5087 mddev
->pers
->stop(mddev
);
5089 if (err
== 0 && mddev
->pers
->sync_request
&&
5090 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5091 err
= bitmap_create(mddev
);
5093 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5094 mdname(mddev
), err
);
5095 mddev
->pers
->stop(mddev
);
5099 module_put(mddev
->pers
->owner
);
5101 bitmap_destroy(mddev
);
5104 if (mddev
->pers
->sync_request
) {
5105 if (mddev
->kobj
.sd
&&
5106 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5108 "md: cannot register extra attributes for %s\n",
5110 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5111 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5114 atomic_set(&mddev
->writes_pending
,0);
5115 atomic_set(&mddev
->max_corr_read_errors
,
5116 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5117 mddev
->safemode
= 0;
5118 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5119 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5120 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5124 rdev_for_each(rdev
, mddev
)
5125 if (rdev
->raid_disk
>= 0)
5126 if (sysfs_link_rdev(mddev
, rdev
))
5127 /* failure here is OK */;
5129 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5131 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
5132 md_update_sb(mddev
, 0);
5134 md_new_event(mddev
);
5135 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5136 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5137 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5140 EXPORT_SYMBOL_GPL(md_run
);
5142 static int do_md_run(struct mddev
*mddev
)
5146 err
= md_run(mddev
);
5149 err
= bitmap_load(mddev
);
5151 bitmap_destroy(mddev
);
5155 md_wakeup_thread(mddev
->thread
);
5156 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5158 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5159 revalidate_disk(mddev
->gendisk
);
5161 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5166 static int restart_array(struct mddev
*mddev
)
5168 struct gendisk
*disk
= mddev
->gendisk
;
5170 /* Complain if it has no devices */
5171 if (list_empty(&mddev
->disks
))
5177 mddev
->safemode
= 0;
5179 set_disk_ro(disk
, 0);
5180 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5182 /* Kick recovery or resync if necessary */
5183 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5184 md_wakeup_thread(mddev
->thread
);
5185 md_wakeup_thread(mddev
->sync_thread
);
5186 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5190 static void md_clean(struct mddev
*mddev
)
5192 mddev
->array_sectors
= 0;
5193 mddev
->external_size
= 0;
5194 mddev
->dev_sectors
= 0;
5195 mddev
->raid_disks
= 0;
5196 mddev
->recovery_cp
= 0;
5197 mddev
->resync_min
= 0;
5198 mddev
->resync_max
= MaxSector
;
5199 mddev
->reshape_position
= MaxSector
;
5200 mddev
->external
= 0;
5201 mddev
->persistent
= 0;
5202 mddev
->level
= LEVEL_NONE
;
5203 mddev
->clevel
[0] = 0;
5206 mddev
->metadata_type
[0] = 0;
5207 mddev
->chunk_sectors
= 0;
5208 mddev
->ctime
= mddev
->utime
= 0;
5210 mddev
->max_disks
= 0;
5212 mddev
->can_decrease_events
= 0;
5213 mddev
->delta_disks
= 0;
5214 mddev
->reshape_backwards
= 0;
5215 mddev
->new_level
= LEVEL_NONE
;
5216 mddev
->new_layout
= 0;
5217 mddev
->new_chunk_sectors
= 0;
5218 mddev
->curr_resync
= 0;
5219 atomic64_set(&mddev
->resync_mismatches
, 0);
5220 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5221 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5222 mddev
->recovery
= 0;
5225 mddev
->degraded
= 0;
5226 mddev
->safemode
= 0;
5227 mddev
->merge_check_needed
= 0;
5228 mddev
->bitmap_info
.offset
= 0;
5229 mddev
->bitmap_info
.default_offset
= 0;
5230 mddev
->bitmap_info
.default_space
= 0;
5231 mddev
->bitmap_info
.chunksize
= 0;
5232 mddev
->bitmap_info
.daemon_sleep
= 0;
5233 mddev
->bitmap_info
.max_write_behind
= 0;
5236 static void __md_stop_writes(struct mddev
*mddev
)
5238 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5239 if (mddev
->sync_thread
) {
5240 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5241 md_reap_sync_thread(mddev
);
5244 del_timer_sync(&mddev
->safemode_timer
);
5246 bitmap_flush(mddev
);
5247 md_super_wait(mddev
);
5249 if (mddev
->ro
== 0 &&
5250 (!mddev
->in_sync
|| (mddev
->flags
& MD_UPDATE_SB_FLAGS
))) {
5251 /* mark array as shutdown cleanly */
5253 md_update_sb(mddev
, 1);
5257 void md_stop_writes(struct mddev
*mddev
)
5259 mddev_lock_nointr(mddev
);
5260 __md_stop_writes(mddev
);
5261 mddev_unlock(mddev
);
5263 EXPORT_SYMBOL_GPL(md_stop_writes
);
5265 static void __md_stop(struct mddev
*mddev
)
5268 mddev
->pers
->stop(mddev
);
5269 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5270 mddev
->to_remove
= &md_redundancy_group
;
5271 module_put(mddev
->pers
->owner
);
5273 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5276 void md_stop(struct mddev
*mddev
)
5278 /* stop the array and free an attached data structures.
5279 * This is called from dm-raid
5282 bitmap_destroy(mddev
);
5284 bioset_free(mddev
->bio_set
);
5287 EXPORT_SYMBOL_GPL(md_stop
);
5289 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5294 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5296 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5297 md_wakeup_thread(mddev
->thread
);
5299 if (mddev
->sync_thread
) {
5300 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5301 /* Thread might be blocked waiting for metadata update
5302 * which will now never happen */
5303 wake_up_process(mddev
->sync_thread
->tsk
);
5305 mddev_unlock(mddev
);
5306 wait_event(resync_wait
, mddev
->sync_thread
== NULL
);
5307 mddev_lock_nointr(mddev
);
5309 mutex_lock(&mddev
->open_mutex
);
5310 if (atomic_read(&mddev
->openers
) > !!bdev
||
5311 mddev
->sync_thread
||
5312 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5313 printk("md: %s still in use.\n",mdname(mddev
));
5315 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5316 md_wakeup_thread(mddev
->thread
);
5322 __md_stop_writes(mddev
);
5328 set_disk_ro(mddev
->gendisk
, 1);
5329 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5330 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5334 mutex_unlock(&mddev
->open_mutex
);
5339 * 0 - completely stop and dis-assemble array
5340 * 2 - stop but do not disassemble array
5342 static int do_md_stop(struct mddev
* mddev
, int mode
,
5343 struct block_device
*bdev
)
5345 struct gendisk
*disk
= mddev
->gendisk
;
5346 struct md_rdev
*rdev
;
5349 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5351 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5352 md_wakeup_thread(mddev
->thread
);
5354 if (mddev
->sync_thread
) {
5355 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5356 /* Thread might be blocked waiting for metadata update
5357 * which will now never happen */
5358 wake_up_process(mddev
->sync_thread
->tsk
);
5360 mddev_unlock(mddev
);
5361 wait_event(resync_wait
, mddev
->sync_thread
== NULL
);
5362 mddev_lock_nointr(mddev
);
5364 mutex_lock(&mddev
->open_mutex
);
5365 if (atomic_read(&mddev
->openers
) > !!bdev
||
5366 mddev
->sysfs_active
||
5367 mddev
->sync_thread
||
5368 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5369 printk("md: %s still in use.\n",mdname(mddev
));
5370 mutex_unlock(&mddev
->open_mutex
);
5372 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5373 md_wakeup_thread(mddev
->thread
);
5379 set_disk_ro(disk
, 0);
5381 __md_stop_writes(mddev
);
5383 mddev
->queue
->merge_bvec_fn
= NULL
;
5384 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5386 /* tell userspace to handle 'inactive' */
5387 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5389 rdev_for_each(rdev
, mddev
)
5390 if (rdev
->raid_disk
>= 0)
5391 sysfs_unlink_rdev(mddev
, rdev
);
5393 set_capacity(disk
, 0);
5394 mutex_unlock(&mddev
->open_mutex
);
5396 revalidate_disk(disk
);
5401 mutex_unlock(&mddev
->open_mutex
);
5403 * Free resources if final stop
5406 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5408 bitmap_destroy(mddev
);
5409 if (mddev
->bitmap_info
.file
) {
5410 fput(mddev
->bitmap_info
.file
);
5411 mddev
->bitmap_info
.file
= NULL
;
5413 mddev
->bitmap_info
.offset
= 0;
5415 export_array(mddev
);
5418 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5419 if (mddev
->hold_active
== UNTIL_STOP
)
5420 mddev
->hold_active
= 0;
5422 blk_integrity_unregister(disk
);
5423 md_new_event(mddev
);
5424 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5429 static void autorun_array(struct mddev
*mddev
)
5431 struct md_rdev
*rdev
;
5434 if (list_empty(&mddev
->disks
))
5437 printk(KERN_INFO
"md: running: ");
5439 rdev_for_each(rdev
, mddev
) {
5440 char b
[BDEVNAME_SIZE
];
5441 printk("<%s>", bdevname(rdev
->bdev
,b
));
5445 err
= do_md_run(mddev
);
5447 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5448 do_md_stop(mddev
, 0, NULL
);
5453 * lets try to run arrays based on all disks that have arrived
5454 * until now. (those are in pending_raid_disks)
5456 * the method: pick the first pending disk, collect all disks with
5457 * the same UUID, remove all from the pending list and put them into
5458 * the 'same_array' list. Then order this list based on superblock
5459 * update time (freshest comes first), kick out 'old' disks and
5460 * compare superblocks. If everything's fine then run it.
5462 * If "unit" is allocated, then bump its reference count
5464 static void autorun_devices(int part
)
5466 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5467 struct mddev
*mddev
;
5468 char b
[BDEVNAME_SIZE
];
5470 printk(KERN_INFO
"md: autorun ...\n");
5471 while (!list_empty(&pending_raid_disks
)) {
5474 LIST_HEAD(candidates
);
5475 rdev0
= list_entry(pending_raid_disks
.next
,
5476 struct md_rdev
, same_set
);
5478 printk(KERN_INFO
"md: considering %s ...\n",
5479 bdevname(rdev0
->bdev
,b
));
5480 INIT_LIST_HEAD(&candidates
);
5481 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5482 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5483 printk(KERN_INFO
"md: adding %s ...\n",
5484 bdevname(rdev
->bdev
,b
));
5485 list_move(&rdev
->same_set
, &candidates
);
5488 * now we have a set of devices, with all of them having
5489 * mostly sane superblocks. It's time to allocate the
5493 dev
= MKDEV(mdp_major
,
5494 rdev0
->preferred_minor
<< MdpMinorShift
);
5495 unit
= MINOR(dev
) >> MdpMinorShift
;
5497 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5500 if (rdev0
->preferred_minor
!= unit
) {
5501 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5502 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5506 md_probe(dev
, NULL
, NULL
);
5507 mddev
= mddev_find(dev
);
5508 if (!mddev
|| !mddev
->gendisk
) {
5512 "md: cannot allocate memory for md drive.\n");
5515 if (mddev_lock(mddev
))
5516 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5518 else if (mddev
->raid_disks
|| mddev
->major_version
5519 || !list_empty(&mddev
->disks
)) {
5521 "md: %s already running, cannot run %s\n",
5522 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5523 mddev_unlock(mddev
);
5525 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5526 mddev
->persistent
= 1;
5527 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5528 list_del_init(&rdev
->same_set
);
5529 if (bind_rdev_to_array(rdev
, mddev
))
5532 autorun_array(mddev
);
5533 mddev_unlock(mddev
);
5535 /* on success, candidates will be empty, on error
5538 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5539 list_del_init(&rdev
->same_set
);
5544 printk(KERN_INFO
"md: ... autorun DONE.\n");
5546 #endif /* !MODULE */
5548 static int get_version(void __user
* arg
)
5552 ver
.major
= MD_MAJOR_VERSION
;
5553 ver
.minor
= MD_MINOR_VERSION
;
5554 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5556 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5562 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5564 mdu_array_info_t info
;
5565 int nr
,working
,insync
,failed
,spare
;
5566 struct md_rdev
*rdev
;
5568 nr
= working
= insync
= failed
= spare
= 0;
5570 rdev_for_each_rcu(rdev
, mddev
) {
5572 if (test_bit(Faulty
, &rdev
->flags
))
5576 if (test_bit(In_sync
, &rdev
->flags
))
5584 info
.major_version
= mddev
->major_version
;
5585 info
.minor_version
= mddev
->minor_version
;
5586 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5587 info
.ctime
= mddev
->ctime
;
5588 info
.level
= mddev
->level
;
5589 info
.size
= mddev
->dev_sectors
/ 2;
5590 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5593 info
.raid_disks
= mddev
->raid_disks
;
5594 info
.md_minor
= mddev
->md_minor
;
5595 info
.not_persistent
= !mddev
->persistent
;
5597 info
.utime
= mddev
->utime
;
5600 info
.state
= (1<<MD_SB_CLEAN
);
5601 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5602 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
5603 info
.active_disks
= insync
;
5604 info
.working_disks
= working
;
5605 info
.failed_disks
= failed
;
5606 info
.spare_disks
= spare
;
5608 info
.layout
= mddev
->layout
;
5609 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5611 if (copy_to_user(arg
, &info
, sizeof(info
)))
5617 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5619 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5620 char *ptr
, *buf
= NULL
;
5623 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5628 /* bitmap disabled, zero the first byte and copy out */
5629 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5630 file
->pathname
[0] = '\0';
5634 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5638 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5639 buf
, sizeof(file
->pathname
));
5643 strcpy(file
->pathname
, ptr
);
5647 if (copy_to_user(arg
, file
, sizeof(*file
)))
5655 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5657 mdu_disk_info_t info
;
5658 struct md_rdev
*rdev
;
5660 if (copy_from_user(&info
, arg
, sizeof(info
)))
5664 rdev
= find_rdev_nr_rcu(mddev
, info
.number
);
5666 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5667 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5668 info
.raid_disk
= rdev
->raid_disk
;
5670 if (test_bit(Faulty
, &rdev
->flags
))
5671 info
.state
|= (1<<MD_DISK_FAULTY
);
5672 else if (test_bit(In_sync
, &rdev
->flags
)) {
5673 info
.state
|= (1<<MD_DISK_ACTIVE
);
5674 info
.state
|= (1<<MD_DISK_SYNC
);
5676 if (test_bit(WriteMostly
, &rdev
->flags
))
5677 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5679 info
.major
= info
.minor
= 0;
5680 info
.raid_disk
= -1;
5681 info
.state
= (1<<MD_DISK_REMOVED
);
5685 if (copy_to_user(arg
, &info
, sizeof(info
)))
5691 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5693 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5694 struct md_rdev
*rdev
;
5695 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5697 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5700 if (!mddev
->raid_disks
) {
5702 /* expecting a device which has a superblock */
5703 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5706 "md: md_import_device returned %ld\n",
5708 return PTR_ERR(rdev
);
5710 if (!list_empty(&mddev
->disks
)) {
5711 struct md_rdev
*rdev0
5712 = list_entry(mddev
->disks
.next
,
5713 struct md_rdev
, same_set
);
5714 err
= super_types
[mddev
->major_version
]
5715 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5718 "md: %s has different UUID to %s\n",
5719 bdevname(rdev
->bdev
,b
),
5720 bdevname(rdev0
->bdev
,b2
));
5725 err
= bind_rdev_to_array(rdev
, mddev
);
5732 * add_new_disk can be used once the array is assembled
5733 * to add "hot spares". They must already have a superblock
5738 if (!mddev
->pers
->hot_add_disk
) {
5740 "%s: personality does not support diskops!\n",
5744 if (mddev
->persistent
)
5745 rdev
= md_import_device(dev
, mddev
->major_version
,
5746 mddev
->minor_version
);
5748 rdev
= md_import_device(dev
, -1, -1);
5751 "md: md_import_device returned %ld\n",
5753 return PTR_ERR(rdev
);
5755 /* set saved_raid_disk if appropriate */
5756 if (!mddev
->persistent
) {
5757 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5758 info
->raid_disk
< mddev
->raid_disks
) {
5759 rdev
->raid_disk
= info
->raid_disk
;
5760 set_bit(In_sync
, &rdev
->flags
);
5761 clear_bit(Bitmap_sync
, &rdev
->flags
);
5763 rdev
->raid_disk
= -1;
5764 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5766 super_types
[mddev
->major_version
].
5767 validate_super(mddev
, rdev
);
5768 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5769 rdev
->raid_disk
!= info
->raid_disk
) {
5770 /* This was a hot-add request, but events doesn't
5771 * match, so reject it.
5777 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5778 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5779 set_bit(WriteMostly
, &rdev
->flags
);
5781 clear_bit(WriteMostly
, &rdev
->flags
);
5783 rdev
->raid_disk
= -1;
5784 err
= bind_rdev_to_array(rdev
, mddev
);
5785 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5786 /* If there is hot_add_disk but no hot_remove_disk
5787 * then added disks for geometry changes,
5788 * and should be added immediately.
5790 super_types
[mddev
->major_version
].
5791 validate_super(mddev
, rdev
);
5792 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5794 unbind_rdev_from_array(rdev
);
5799 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5801 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5802 if (mddev
->degraded
)
5803 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5804 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5806 md_new_event(mddev
);
5807 md_wakeup_thread(mddev
->thread
);
5811 /* otherwise, add_new_disk is only allowed
5812 * for major_version==0 superblocks
5814 if (mddev
->major_version
!= 0) {
5815 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5820 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5822 rdev
= md_import_device(dev
, -1, 0);
5825 "md: error, md_import_device() returned %ld\n",
5827 return PTR_ERR(rdev
);
5829 rdev
->desc_nr
= info
->number
;
5830 if (info
->raid_disk
< mddev
->raid_disks
)
5831 rdev
->raid_disk
= info
->raid_disk
;
5833 rdev
->raid_disk
= -1;
5835 if (rdev
->raid_disk
< mddev
->raid_disks
)
5836 if (info
->state
& (1<<MD_DISK_SYNC
))
5837 set_bit(In_sync
, &rdev
->flags
);
5839 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5840 set_bit(WriteMostly
, &rdev
->flags
);
5842 if (!mddev
->persistent
) {
5843 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5844 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5846 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5847 rdev
->sectors
= rdev
->sb_start
;
5849 err
= bind_rdev_to_array(rdev
, mddev
);
5859 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5861 char b
[BDEVNAME_SIZE
];
5862 struct md_rdev
*rdev
;
5864 rdev
= find_rdev(mddev
, dev
);
5868 clear_bit(Blocked
, &rdev
->flags
);
5869 remove_and_add_spares(mddev
, rdev
);
5871 if (rdev
->raid_disk
>= 0)
5874 kick_rdev_from_array(rdev
);
5875 md_update_sb(mddev
, 1);
5876 md_new_event(mddev
);
5880 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5881 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5885 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5887 char b
[BDEVNAME_SIZE
];
5889 struct md_rdev
*rdev
;
5894 if (mddev
->major_version
!= 0) {
5895 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5896 " version-0 superblocks.\n",
5900 if (!mddev
->pers
->hot_add_disk
) {
5902 "%s: personality does not support diskops!\n",
5907 rdev
= md_import_device(dev
, -1, 0);
5910 "md: error, md_import_device() returned %ld\n",
5915 if (mddev
->persistent
)
5916 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5918 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5920 rdev
->sectors
= rdev
->sb_start
;
5922 if (test_bit(Faulty
, &rdev
->flags
)) {
5924 "md: can not hot-add faulty %s disk to %s!\n",
5925 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5929 clear_bit(In_sync
, &rdev
->flags
);
5931 rdev
->saved_raid_disk
= -1;
5932 err
= bind_rdev_to_array(rdev
, mddev
);
5937 * The rest should better be atomic, we can have disk failures
5938 * noticed in interrupt contexts ...
5941 rdev
->raid_disk
= -1;
5943 md_update_sb(mddev
, 1);
5946 * Kick recovery, maybe this spare has to be added to the
5947 * array immediately.
5949 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5950 md_wakeup_thread(mddev
->thread
);
5951 md_new_event(mddev
);
5959 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5964 if (!mddev
->pers
->quiesce
)
5966 if (mddev
->recovery
|| mddev
->sync_thread
)
5968 /* we should be able to change the bitmap.. */
5973 struct inode
*inode
;
5975 return -EEXIST
; /* cannot add when bitmap is present */
5976 mddev
->bitmap_info
.file
= fget(fd
);
5978 if (mddev
->bitmap_info
.file
== NULL
) {
5979 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5984 inode
= mddev
->bitmap_info
.file
->f_mapping
->host
;
5985 if (!S_ISREG(inode
->i_mode
)) {
5986 printk(KERN_ERR
"%s: error: bitmap file must be a regular file\n",
5989 } else if (!(mddev
->bitmap_info
.file
->f_mode
& FMODE_WRITE
)) {
5990 printk(KERN_ERR
"%s: error: bitmap file must open for write\n",
5993 } else if (atomic_read(&inode
->i_writecount
) != 1) {
5994 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5999 fput(mddev
->bitmap_info
.file
);
6000 mddev
->bitmap_info
.file
= NULL
;
6003 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6004 } else if (mddev
->bitmap
== NULL
)
6005 return -ENOENT
; /* cannot remove what isn't there */
6008 mddev
->pers
->quiesce(mddev
, 1);
6010 err
= bitmap_create(mddev
);
6012 err
= bitmap_load(mddev
);
6014 if (fd
< 0 || err
) {
6015 bitmap_destroy(mddev
);
6016 fd
= -1; /* make sure to put the file */
6018 mddev
->pers
->quiesce(mddev
, 0);
6021 if (mddev
->bitmap_info
.file
)
6022 fput(mddev
->bitmap_info
.file
);
6023 mddev
->bitmap_info
.file
= NULL
;
6030 * set_array_info is used two different ways
6031 * The original usage is when creating a new array.
6032 * In this usage, raid_disks is > 0 and it together with
6033 * level, size, not_persistent,layout,chunksize determine the
6034 * shape of the array.
6035 * This will always create an array with a type-0.90.0 superblock.
6036 * The newer usage is when assembling an array.
6037 * In this case raid_disks will be 0, and the major_version field is
6038 * use to determine which style super-blocks are to be found on the devices.
6039 * The minor and patch _version numbers are also kept incase the
6040 * super_block handler wishes to interpret them.
6042 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
6045 if (info
->raid_disks
== 0) {
6046 /* just setting version number for superblock loading */
6047 if (info
->major_version
< 0 ||
6048 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6049 super_types
[info
->major_version
].name
== NULL
) {
6050 /* maybe try to auto-load a module? */
6052 "md: superblock version %d not known\n",
6053 info
->major_version
);
6056 mddev
->major_version
= info
->major_version
;
6057 mddev
->minor_version
= info
->minor_version
;
6058 mddev
->patch_version
= info
->patch_version
;
6059 mddev
->persistent
= !info
->not_persistent
;
6060 /* ensure mddev_put doesn't delete this now that there
6061 * is some minimal configuration.
6063 mddev
->ctime
= get_seconds();
6066 mddev
->major_version
= MD_MAJOR_VERSION
;
6067 mddev
->minor_version
= MD_MINOR_VERSION
;
6068 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6069 mddev
->ctime
= get_seconds();
6071 mddev
->level
= info
->level
;
6072 mddev
->clevel
[0] = 0;
6073 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6074 mddev
->raid_disks
= info
->raid_disks
;
6075 /* don't set md_minor, it is determined by which /dev/md* was
6078 if (info
->state
& (1<<MD_SB_CLEAN
))
6079 mddev
->recovery_cp
= MaxSector
;
6081 mddev
->recovery_cp
= 0;
6082 mddev
->persistent
= ! info
->not_persistent
;
6083 mddev
->external
= 0;
6085 mddev
->layout
= info
->layout
;
6086 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6088 mddev
->max_disks
= MD_SB_DISKS
;
6090 if (mddev
->persistent
)
6092 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6094 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6095 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6096 mddev
->bitmap_info
.offset
= 0;
6098 mddev
->reshape_position
= MaxSector
;
6101 * Generate a 128 bit UUID
6103 get_random_bytes(mddev
->uuid
, 16);
6105 mddev
->new_level
= mddev
->level
;
6106 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6107 mddev
->new_layout
= mddev
->layout
;
6108 mddev
->delta_disks
= 0;
6109 mddev
->reshape_backwards
= 0;
6114 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6116 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6118 if (mddev
->external_size
)
6121 mddev
->array_sectors
= array_sectors
;
6123 EXPORT_SYMBOL(md_set_array_sectors
);
6125 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6127 struct md_rdev
*rdev
;
6129 int fit
= (num_sectors
== 0);
6131 if (mddev
->pers
->resize
== NULL
)
6133 /* The "num_sectors" is the number of sectors of each device that
6134 * is used. This can only make sense for arrays with redundancy.
6135 * linear and raid0 always use whatever space is available. We can only
6136 * consider changing this number if no resync or reconstruction is
6137 * happening, and if the new size is acceptable. It must fit before the
6138 * sb_start or, if that is <data_offset, it must fit before the size
6139 * of each device. If num_sectors is zero, we find the largest size
6142 if (mddev
->sync_thread
)
6147 rdev_for_each(rdev
, mddev
) {
6148 sector_t avail
= rdev
->sectors
;
6150 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6151 num_sectors
= avail
;
6152 if (avail
< num_sectors
)
6155 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6157 revalidate_disk(mddev
->gendisk
);
6161 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6164 struct md_rdev
*rdev
;
6165 /* change the number of raid disks */
6166 if (mddev
->pers
->check_reshape
== NULL
)
6170 if (raid_disks
<= 0 ||
6171 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6173 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
6176 rdev_for_each(rdev
, mddev
) {
6177 if (mddev
->raid_disks
< raid_disks
&&
6178 rdev
->data_offset
< rdev
->new_data_offset
)
6180 if (mddev
->raid_disks
> raid_disks
&&
6181 rdev
->data_offset
> rdev
->new_data_offset
)
6185 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6186 if (mddev
->delta_disks
< 0)
6187 mddev
->reshape_backwards
= 1;
6188 else if (mddev
->delta_disks
> 0)
6189 mddev
->reshape_backwards
= 0;
6191 rv
= mddev
->pers
->check_reshape(mddev
);
6193 mddev
->delta_disks
= 0;
6194 mddev
->reshape_backwards
= 0;
6201 * update_array_info is used to change the configuration of an
6203 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6204 * fields in the info are checked against the array.
6205 * Any differences that cannot be handled will cause an error.
6206 * Normally, only one change can be managed at a time.
6208 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6214 /* calculate expected state,ignoring low bits */
6215 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6216 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6218 if (mddev
->major_version
!= info
->major_version
||
6219 mddev
->minor_version
!= info
->minor_version
||
6220 /* mddev->patch_version != info->patch_version || */
6221 mddev
->ctime
!= info
->ctime
||
6222 mddev
->level
!= info
->level
||
6223 /* mddev->layout != info->layout || */
6224 !mddev
->persistent
!= info
->not_persistent
||
6225 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6226 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6227 ((state
^info
->state
) & 0xfffffe00)
6230 /* Check there is only one change */
6231 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6233 if (mddev
->raid_disks
!= info
->raid_disks
)
6235 if (mddev
->layout
!= info
->layout
)
6237 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6244 if (mddev
->layout
!= info
->layout
) {
6246 * we don't need to do anything at the md level, the
6247 * personality will take care of it all.
6249 if (mddev
->pers
->check_reshape
== NULL
)
6252 mddev
->new_layout
= info
->layout
;
6253 rv
= mddev
->pers
->check_reshape(mddev
);
6255 mddev
->new_layout
= mddev
->layout
;
6259 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6260 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6262 if (mddev
->raid_disks
!= info
->raid_disks
)
6263 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6265 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6266 if (mddev
->pers
->quiesce
== NULL
)
6268 if (mddev
->recovery
|| mddev
->sync_thread
)
6270 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6271 /* add the bitmap */
6274 if (mddev
->bitmap_info
.default_offset
== 0)
6276 mddev
->bitmap_info
.offset
=
6277 mddev
->bitmap_info
.default_offset
;
6278 mddev
->bitmap_info
.space
=
6279 mddev
->bitmap_info
.default_space
;
6280 mddev
->pers
->quiesce(mddev
, 1);
6281 rv
= bitmap_create(mddev
);
6283 rv
= bitmap_load(mddev
);
6285 bitmap_destroy(mddev
);
6286 mddev
->pers
->quiesce(mddev
, 0);
6288 /* remove the bitmap */
6291 if (mddev
->bitmap
->storage
.file
)
6293 mddev
->pers
->quiesce(mddev
, 1);
6294 bitmap_destroy(mddev
);
6295 mddev
->pers
->quiesce(mddev
, 0);
6296 mddev
->bitmap_info
.offset
= 0;
6299 md_update_sb(mddev
, 1);
6303 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6305 struct md_rdev
*rdev
;
6308 if (mddev
->pers
== NULL
)
6312 rdev
= find_rdev_rcu(mddev
, dev
);
6316 md_error(mddev
, rdev
);
6317 if (!test_bit(Faulty
, &rdev
->flags
))
6325 * We have a problem here : there is no easy way to give a CHS
6326 * virtual geometry. We currently pretend that we have a 2 heads
6327 * 4 sectors (with a BIG number of cylinders...). This drives
6328 * dosfs just mad... ;-)
6330 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6332 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6336 geo
->cylinders
= mddev
->array_sectors
/ 8;
6340 static inline bool md_ioctl_valid(unsigned int cmd
)
6345 case GET_ARRAY_INFO
:
6346 case GET_BITMAP_FILE
:
6349 case HOT_REMOVE_DISK
:
6350 case PRINT_RAID_DEBUG
:
6353 case RESTART_ARRAY_RW
:
6355 case SET_ARRAY_INFO
:
6356 case SET_BITMAP_FILE
:
6357 case SET_DISK_FAULTY
:
6366 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6367 unsigned int cmd
, unsigned long arg
)
6370 void __user
*argp
= (void __user
*)arg
;
6371 struct mddev
*mddev
= NULL
;
6374 if (!md_ioctl_valid(cmd
))
6379 case GET_ARRAY_INFO
:
6383 if (!capable(CAP_SYS_ADMIN
))
6388 * Commands dealing with the RAID driver but not any
6393 err
= get_version(argp
);
6396 case PRINT_RAID_DEBUG
:
6404 autostart_arrays(arg
);
6411 * Commands creating/starting a new array:
6414 mddev
= bdev
->bd_disk
->private_data
;
6421 /* Some actions do not requires the mutex */
6423 case GET_ARRAY_INFO
:
6424 if (!mddev
->raid_disks
&& !mddev
->external
)
6427 err
= get_array_info(mddev
, argp
);
6431 if (!mddev
->raid_disks
&& !mddev
->external
)
6434 err
= get_disk_info(mddev
, argp
);
6437 case SET_DISK_FAULTY
:
6438 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6442 if (cmd
== ADD_NEW_DISK
)
6443 /* need to ensure md_delayed_delete() has completed */
6444 flush_workqueue(md_misc_wq
);
6446 if (cmd
== HOT_REMOVE_DISK
)
6447 /* need to ensure recovery thread has run */
6448 wait_event_interruptible_timeout(mddev
->sb_wait
,
6449 !test_bit(MD_RECOVERY_NEEDED
,
6451 msecs_to_jiffies(5000));
6452 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
6453 /* Need to flush page cache, and ensure no-one else opens
6456 mutex_lock(&mddev
->open_mutex
);
6457 if (atomic_read(&mddev
->openers
) > 1) {
6458 mutex_unlock(&mddev
->open_mutex
);
6462 set_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6463 mutex_unlock(&mddev
->open_mutex
);
6464 sync_blockdev(bdev
);
6466 err
= mddev_lock(mddev
);
6469 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6474 if (cmd
== SET_ARRAY_INFO
) {
6475 mdu_array_info_t info
;
6477 memset(&info
, 0, sizeof(info
));
6478 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6483 err
= update_array_info(mddev
, &info
);
6485 printk(KERN_WARNING
"md: couldn't update"
6486 " array info. %d\n", err
);
6491 if (!list_empty(&mddev
->disks
)) {
6493 "md: array %s already has disks!\n",
6498 if (mddev
->raid_disks
) {
6500 "md: array %s already initialised!\n",
6505 err
= set_array_info(mddev
, &info
);
6507 printk(KERN_WARNING
"md: couldn't set"
6508 " array info. %d\n", err
);
6515 * Commands querying/configuring an existing array:
6517 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6518 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6519 if ((!mddev
->raid_disks
&& !mddev
->external
)
6520 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6521 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6522 && cmd
!= GET_BITMAP_FILE
) {
6528 * Commands even a read-only array can execute:
6531 case GET_BITMAP_FILE
:
6532 err
= get_bitmap_file(mddev
, argp
);
6535 case RESTART_ARRAY_RW
:
6536 err
= restart_array(mddev
);
6540 err
= do_md_stop(mddev
, 0, bdev
);
6544 err
= md_set_readonly(mddev
, bdev
);
6547 case HOT_REMOVE_DISK
:
6548 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6552 /* We can support ADD_NEW_DISK on read-only arrays
6553 * on if we are re-adding a preexisting device.
6554 * So require mddev->pers and MD_DISK_SYNC.
6557 mdu_disk_info_t info
;
6558 if (copy_from_user(&info
, argp
, sizeof(info
)))
6560 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6561 /* Need to clear read-only for this */
6564 err
= add_new_disk(mddev
, &info
);
6570 if (get_user(ro
, (int __user
*)(arg
))) {
6576 /* if the bdev is going readonly the value of mddev->ro
6577 * does not matter, no writes are coming
6582 /* are we are already prepared for writes? */
6586 /* transitioning to readauto need only happen for
6587 * arrays that call md_write_start
6590 err
= restart_array(mddev
);
6593 set_disk_ro(mddev
->gendisk
, 0);
6600 * The remaining ioctls are changing the state of the
6601 * superblock, so we do not allow them on read-only arrays.
6602 * However non-MD ioctls (e.g. get-size) will still come through
6603 * here and hit the 'default' below, so only disallow
6604 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6606 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6607 if (mddev
->ro
== 2) {
6609 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6610 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6611 /* mddev_unlock will wake thread */
6612 /* If a device failed while we were read-only, we
6613 * need to make sure the metadata is updated now.
6615 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6616 mddev_unlock(mddev
);
6617 wait_event(mddev
->sb_wait
,
6618 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6619 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6620 mddev_lock_nointr(mddev
);
6631 mdu_disk_info_t info
;
6632 if (copy_from_user(&info
, argp
, sizeof(info
)))
6635 err
= add_new_disk(mddev
, &info
);
6640 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6644 err
= do_md_run(mddev
);
6647 case SET_BITMAP_FILE
:
6648 err
= set_bitmap_file(mddev
, (int)arg
);
6658 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6660 mddev
->hold_active
= 0;
6661 mddev_unlock(mddev
);
6670 #ifdef CONFIG_COMPAT
6671 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6672 unsigned int cmd
, unsigned long arg
)
6675 case HOT_REMOVE_DISK
:
6677 case SET_DISK_FAULTY
:
6678 case SET_BITMAP_FILE
:
6679 /* These take in integer arg, do not convert */
6682 arg
= (unsigned long)compat_ptr(arg
);
6686 return md_ioctl(bdev
, mode
, cmd
, arg
);
6688 #endif /* CONFIG_COMPAT */
6690 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6693 * Succeed if we can lock the mddev, which confirms that
6694 * it isn't being stopped right now.
6696 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6702 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6703 /* we are racing with mddev_put which is discarding this
6707 /* Wait until bdev->bd_disk is definitely gone */
6708 flush_workqueue(md_misc_wq
);
6709 /* Then retry the open from the top */
6710 return -ERESTARTSYS
;
6712 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6714 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6718 atomic_inc(&mddev
->openers
);
6719 clear_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6720 mutex_unlock(&mddev
->open_mutex
);
6722 check_disk_change(bdev
);
6727 static void md_release(struct gendisk
*disk
, fmode_t mode
)
6729 struct mddev
*mddev
= disk
->private_data
;
6732 atomic_dec(&mddev
->openers
);
6736 static int md_media_changed(struct gendisk
*disk
)
6738 struct mddev
*mddev
= disk
->private_data
;
6740 return mddev
->changed
;
6743 static int md_revalidate(struct gendisk
*disk
)
6745 struct mddev
*mddev
= disk
->private_data
;
6750 static const struct block_device_operations md_fops
=
6752 .owner
= THIS_MODULE
,
6754 .release
= md_release
,
6756 #ifdef CONFIG_COMPAT
6757 .compat_ioctl
= md_compat_ioctl
,
6759 .getgeo
= md_getgeo
,
6760 .media_changed
= md_media_changed
,
6761 .revalidate_disk
= md_revalidate
,
6764 static int md_thread(void * arg
)
6766 struct md_thread
*thread
= arg
;
6769 * md_thread is a 'system-thread', it's priority should be very
6770 * high. We avoid resource deadlocks individually in each
6771 * raid personality. (RAID5 does preallocation) We also use RR and
6772 * the very same RT priority as kswapd, thus we will never get
6773 * into a priority inversion deadlock.
6775 * we definitely have to have equal or higher priority than
6776 * bdflush, otherwise bdflush will deadlock if there are too
6777 * many dirty RAID5 blocks.
6780 allow_signal(SIGKILL
);
6781 while (!kthread_should_stop()) {
6783 /* We need to wait INTERRUPTIBLE so that
6784 * we don't add to the load-average.
6785 * That means we need to be sure no signals are
6788 if (signal_pending(current
))
6789 flush_signals(current
);
6791 wait_event_interruptible_timeout
6793 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6794 || kthread_should_stop(),
6797 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6798 if (!kthread_should_stop())
6799 thread
->run(thread
);
6805 void md_wakeup_thread(struct md_thread
*thread
)
6808 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6809 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6810 wake_up(&thread
->wqueue
);
6814 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
6815 struct mddev
*mddev
, const char *name
)
6817 struct md_thread
*thread
;
6819 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6823 init_waitqueue_head(&thread
->wqueue
);
6826 thread
->mddev
= mddev
;
6827 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6828 thread
->tsk
= kthread_run(md_thread
, thread
,
6830 mdname(thread
->mddev
),
6832 if (IS_ERR(thread
->tsk
)) {
6839 void md_unregister_thread(struct md_thread
**threadp
)
6841 struct md_thread
*thread
= *threadp
;
6844 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6845 /* Locking ensures that mddev_unlock does not wake_up a
6846 * non-existent thread
6848 spin_lock(&pers_lock
);
6850 spin_unlock(&pers_lock
);
6852 kthread_stop(thread
->tsk
);
6856 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6863 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6866 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6868 mddev
->pers
->error_handler(mddev
,rdev
);
6869 if (mddev
->degraded
)
6870 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6871 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6872 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6873 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6874 md_wakeup_thread(mddev
->thread
);
6875 if (mddev
->event_work
.func
)
6876 queue_work(md_misc_wq
, &mddev
->event_work
);
6877 md_new_event_inintr(mddev
);
6880 /* seq_file implementation /proc/mdstat */
6882 static void status_unused(struct seq_file
*seq
)
6885 struct md_rdev
*rdev
;
6887 seq_printf(seq
, "unused devices: ");
6889 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6890 char b
[BDEVNAME_SIZE
];
6892 seq_printf(seq
, "%s ",
6893 bdevname(rdev
->bdev
,b
));
6896 seq_printf(seq
, "<none>");
6898 seq_printf(seq
, "\n");
6902 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6904 sector_t max_sectors
, resync
, res
;
6905 unsigned long dt
, db
;
6908 unsigned int per_milli
;
6910 if (mddev
->curr_resync
<= 3)
6913 resync
= mddev
->curr_resync
6914 - atomic_read(&mddev
->recovery_active
);
6916 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6917 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6918 max_sectors
= mddev
->resync_max_sectors
;
6920 max_sectors
= mddev
->dev_sectors
;
6923 * Should not happen.
6929 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6930 * in a sector_t, and (max_sectors>>scale) will fit in a
6931 * u32, as those are the requirements for sector_div.
6932 * Thus 'scale' must be at least 10
6935 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6936 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6939 res
= (resync
>>scale
)*1000;
6940 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6944 int i
, x
= per_milli
/50, y
= 20-x
;
6945 seq_printf(seq
, "[");
6946 for (i
= 0; i
< x
; i
++)
6947 seq_printf(seq
, "=");
6948 seq_printf(seq
, ">");
6949 for (i
= 0; i
< y
; i
++)
6950 seq_printf(seq
, ".");
6951 seq_printf(seq
, "] ");
6953 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6954 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6956 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6958 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6959 "resync" : "recovery"))),
6960 per_milli
/10, per_milli
% 10,
6961 (unsigned long long) resync
/2,
6962 (unsigned long long) max_sectors
/2);
6965 * dt: time from mark until now
6966 * db: blocks written from mark until now
6967 * rt: remaining time
6969 * rt is a sector_t, so could be 32bit or 64bit.
6970 * So we divide before multiply in case it is 32bit and close
6972 * We scale the divisor (db) by 32 to avoid losing precision
6973 * near the end of resync when the number of remaining sectors
6975 * We then divide rt by 32 after multiplying by db to compensate.
6976 * The '+1' avoids division by zero if db is very small.
6978 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6980 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6981 - mddev
->resync_mark_cnt
;
6983 rt
= max_sectors
- resync
; /* number of remaining sectors */
6984 sector_div(rt
, db
/32+1);
6988 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6989 ((unsigned long)rt
% 60)/6);
6991 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6994 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6996 struct list_head
*tmp
;
6998 struct mddev
*mddev
;
7006 spin_lock(&all_mddevs_lock
);
7007 list_for_each(tmp
,&all_mddevs
)
7009 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7011 spin_unlock(&all_mddevs_lock
);
7014 spin_unlock(&all_mddevs_lock
);
7016 return (void*)2;/* tail */
7020 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7022 struct list_head
*tmp
;
7023 struct mddev
*next_mddev
, *mddev
= v
;
7029 spin_lock(&all_mddevs_lock
);
7031 tmp
= all_mddevs
.next
;
7033 tmp
= mddev
->all_mddevs
.next
;
7034 if (tmp
!= &all_mddevs
)
7035 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7037 next_mddev
= (void*)2;
7040 spin_unlock(&all_mddevs_lock
);
7048 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7050 struct mddev
*mddev
= v
;
7052 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7056 static int md_seq_show(struct seq_file
*seq
, void *v
)
7058 struct mddev
*mddev
= v
;
7060 struct md_rdev
*rdev
;
7062 if (v
== (void*)1) {
7063 struct md_personality
*pers
;
7064 seq_printf(seq
, "Personalities : ");
7065 spin_lock(&pers_lock
);
7066 list_for_each_entry(pers
, &pers_list
, list
)
7067 seq_printf(seq
, "[%s] ", pers
->name
);
7069 spin_unlock(&pers_lock
);
7070 seq_printf(seq
, "\n");
7071 seq
->poll_event
= atomic_read(&md_event_count
);
7074 if (v
== (void*)2) {
7079 if (mddev_lock(mddev
) < 0)
7082 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7083 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7084 mddev
->pers
? "" : "in");
7087 seq_printf(seq
, " (read-only)");
7089 seq_printf(seq
, " (auto-read-only)");
7090 seq_printf(seq
, " %s", mddev
->pers
->name
);
7094 rdev_for_each(rdev
, mddev
) {
7095 char b
[BDEVNAME_SIZE
];
7096 seq_printf(seq
, " %s[%d]",
7097 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7098 if (test_bit(WriteMostly
, &rdev
->flags
))
7099 seq_printf(seq
, "(W)");
7100 if (test_bit(Faulty
, &rdev
->flags
)) {
7101 seq_printf(seq
, "(F)");
7104 if (rdev
->raid_disk
< 0)
7105 seq_printf(seq
, "(S)"); /* spare */
7106 if (test_bit(Replacement
, &rdev
->flags
))
7107 seq_printf(seq
, "(R)");
7108 sectors
+= rdev
->sectors
;
7111 if (!list_empty(&mddev
->disks
)) {
7113 seq_printf(seq
, "\n %llu blocks",
7114 (unsigned long long)
7115 mddev
->array_sectors
/ 2);
7117 seq_printf(seq
, "\n %llu blocks",
7118 (unsigned long long)sectors
/ 2);
7120 if (mddev
->persistent
) {
7121 if (mddev
->major_version
!= 0 ||
7122 mddev
->minor_version
!= 90) {
7123 seq_printf(seq
," super %d.%d",
7124 mddev
->major_version
,
7125 mddev
->minor_version
);
7127 } else if (mddev
->external
)
7128 seq_printf(seq
, " super external:%s",
7129 mddev
->metadata_type
);
7131 seq_printf(seq
, " super non-persistent");
7134 mddev
->pers
->status(seq
, mddev
);
7135 seq_printf(seq
, "\n ");
7136 if (mddev
->pers
->sync_request
) {
7137 if (mddev
->curr_resync
> 2) {
7138 status_resync(seq
, mddev
);
7139 seq_printf(seq
, "\n ");
7140 } else if (mddev
->curr_resync
>= 1)
7141 seq_printf(seq
, "\tresync=DELAYED\n ");
7142 else if (mddev
->recovery_cp
< MaxSector
)
7143 seq_printf(seq
, "\tresync=PENDING\n ");
7146 seq_printf(seq
, "\n ");
7148 bitmap_status(seq
, mddev
->bitmap
);
7150 seq_printf(seq
, "\n");
7152 mddev_unlock(mddev
);
7157 static const struct seq_operations md_seq_ops
= {
7158 .start
= md_seq_start
,
7159 .next
= md_seq_next
,
7160 .stop
= md_seq_stop
,
7161 .show
= md_seq_show
,
7164 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7166 struct seq_file
*seq
;
7169 error
= seq_open(file
, &md_seq_ops
);
7173 seq
= file
->private_data
;
7174 seq
->poll_event
= atomic_read(&md_event_count
);
7178 static int md_unloading
;
7179 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7181 struct seq_file
*seq
= filp
->private_data
;
7185 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;;
7186 poll_wait(filp
, &md_event_waiters
, wait
);
7188 /* always allow read */
7189 mask
= POLLIN
| POLLRDNORM
;
7191 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7192 mask
|= POLLERR
| POLLPRI
;
7196 static const struct file_operations md_seq_fops
= {
7197 .owner
= THIS_MODULE
,
7198 .open
= md_seq_open
,
7200 .llseek
= seq_lseek
,
7201 .release
= seq_release_private
,
7202 .poll
= mdstat_poll
,
7205 int register_md_personality(struct md_personality
*p
)
7207 spin_lock(&pers_lock
);
7208 list_add_tail(&p
->list
, &pers_list
);
7209 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
7210 spin_unlock(&pers_lock
);
7214 int unregister_md_personality(struct md_personality
*p
)
7216 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7217 spin_lock(&pers_lock
);
7218 list_del_init(&p
->list
);
7219 spin_unlock(&pers_lock
);
7223 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7225 struct md_rdev
* rdev
;
7231 rdev_for_each_rcu(rdev
, mddev
) {
7232 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7233 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7234 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7235 atomic_read(&disk
->sync_io
);
7236 /* sync IO will cause sync_io to increase before the disk_stats
7237 * as sync_io is counted when a request starts, and
7238 * disk_stats is counted when it completes.
7239 * So resync activity will cause curr_events to be smaller than
7240 * when there was no such activity.
7241 * non-sync IO will cause disk_stat to increase without
7242 * increasing sync_io so curr_events will (eventually)
7243 * be larger than it was before. Once it becomes
7244 * substantially larger, the test below will cause
7245 * the array to appear non-idle, and resync will slow
7247 * If there is a lot of outstanding resync activity when
7248 * we set last_event to curr_events, then all that activity
7249 * completing might cause the array to appear non-idle
7250 * and resync will be slowed down even though there might
7251 * not have been non-resync activity. This will only
7252 * happen once though. 'last_events' will soon reflect
7253 * the state where there is little or no outstanding
7254 * resync requests, and further resync activity will
7255 * always make curr_events less than last_events.
7258 if (init
|| curr_events
- rdev
->last_events
> 64) {
7259 rdev
->last_events
= curr_events
;
7267 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7269 /* another "blocks" (512byte) blocks have been synced */
7270 atomic_sub(blocks
, &mddev
->recovery_active
);
7271 wake_up(&mddev
->recovery_wait
);
7273 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7274 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7275 md_wakeup_thread(mddev
->thread
);
7276 // stop recovery, signal do_sync ....
7281 /* md_write_start(mddev, bi)
7282 * If we need to update some array metadata (e.g. 'active' flag
7283 * in superblock) before writing, schedule a superblock update
7284 * and wait for it to complete.
7286 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7289 if (bio_data_dir(bi
) != WRITE
)
7292 BUG_ON(mddev
->ro
== 1);
7293 if (mddev
->ro
== 2) {
7294 /* need to switch to read/write */
7296 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7297 md_wakeup_thread(mddev
->thread
);
7298 md_wakeup_thread(mddev
->sync_thread
);
7301 atomic_inc(&mddev
->writes_pending
);
7302 if (mddev
->safemode
== 1)
7303 mddev
->safemode
= 0;
7304 if (mddev
->in_sync
) {
7305 spin_lock_irq(&mddev
->write_lock
);
7306 if (mddev
->in_sync
) {
7308 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7309 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7310 md_wakeup_thread(mddev
->thread
);
7313 spin_unlock_irq(&mddev
->write_lock
);
7316 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7317 wait_event(mddev
->sb_wait
,
7318 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7321 void md_write_end(struct mddev
*mddev
)
7323 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7324 if (mddev
->safemode
== 2)
7325 md_wakeup_thread(mddev
->thread
);
7326 else if (mddev
->safemode_delay
)
7327 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7331 /* md_allow_write(mddev)
7332 * Calling this ensures that the array is marked 'active' so that writes
7333 * may proceed without blocking. It is important to call this before
7334 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7335 * Must be called with mddev_lock held.
7337 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7338 * is dropped, so return -EAGAIN after notifying userspace.
7340 int md_allow_write(struct mddev
*mddev
)
7346 if (!mddev
->pers
->sync_request
)
7349 spin_lock_irq(&mddev
->write_lock
);
7350 if (mddev
->in_sync
) {
7352 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7353 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7354 if (mddev
->safemode_delay
&&
7355 mddev
->safemode
== 0)
7356 mddev
->safemode
= 1;
7357 spin_unlock_irq(&mddev
->write_lock
);
7358 md_update_sb(mddev
, 0);
7359 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7361 spin_unlock_irq(&mddev
->write_lock
);
7363 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7368 EXPORT_SYMBOL_GPL(md_allow_write
);
7370 #define SYNC_MARKS 10
7371 #define SYNC_MARK_STEP (3*HZ)
7372 #define UPDATE_FREQUENCY (5*60*HZ)
7373 void md_do_sync(struct md_thread
*thread
)
7375 struct mddev
*mddev
= thread
->mddev
;
7376 struct mddev
*mddev2
;
7377 unsigned int currspeed
= 0,
7379 sector_t max_sectors
,j
, io_sectors
;
7380 unsigned long mark
[SYNC_MARKS
];
7381 unsigned long update_time
;
7382 sector_t mark_cnt
[SYNC_MARKS
];
7384 struct list_head
*tmp
;
7385 sector_t last_check
;
7387 struct md_rdev
*rdev
;
7388 char *desc
, *action
= NULL
;
7389 struct blk_plug plug
;
7391 /* just incase thread restarts... */
7392 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7394 if (mddev
->ro
) {/* never try to sync a read-only array */
7395 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7399 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7400 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
7401 desc
= "data-check";
7403 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7404 desc
= "requested-resync";
7408 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7413 mddev
->last_sync_action
= action
?: desc
;
7415 /* we overload curr_resync somewhat here.
7416 * 0 == not engaged in resync at all
7417 * 2 == checking that there is no conflict with another sync
7418 * 1 == like 2, but have yielded to allow conflicting resync to
7420 * other == active in resync - this many blocks
7422 * Before starting a resync we must have set curr_resync to
7423 * 2, and then checked that every "conflicting" array has curr_resync
7424 * less than ours. When we find one that is the same or higher
7425 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7426 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7427 * This will mean we have to start checking from the beginning again.
7432 mddev
->curr_resync
= 2;
7435 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7437 for_each_mddev(mddev2
, tmp
) {
7438 if (mddev2
== mddev
)
7440 if (!mddev
->parallel_resync
7441 && mddev2
->curr_resync
7442 && match_mddev_units(mddev
, mddev2
)) {
7444 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7445 /* arbitrarily yield */
7446 mddev
->curr_resync
= 1;
7447 wake_up(&resync_wait
);
7449 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7450 /* no need to wait here, we can wait the next
7451 * time 'round when curr_resync == 2
7454 /* We need to wait 'interruptible' so as not to
7455 * contribute to the load average, and not to
7456 * be caught by 'softlockup'
7458 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7459 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7460 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7461 printk(KERN_INFO
"md: delaying %s of %s"
7462 " until %s has finished (they"
7463 " share one or more physical units)\n",
7464 desc
, mdname(mddev
), mdname(mddev2
));
7466 if (signal_pending(current
))
7467 flush_signals(current
);
7469 finish_wait(&resync_wait
, &wq
);
7472 finish_wait(&resync_wait
, &wq
);
7475 } while (mddev
->curr_resync
< 2);
7478 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7479 /* resync follows the size requested by the personality,
7480 * which defaults to physical size, but can be virtual size
7482 max_sectors
= mddev
->resync_max_sectors
;
7483 atomic64_set(&mddev
->resync_mismatches
, 0);
7484 /* we don't use the checkpoint if there's a bitmap */
7485 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7486 j
= mddev
->resync_min
;
7487 else if (!mddev
->bitmap
)
7488 j
= mddev
->recovery_cp
;
7490 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7491 max_sectors
= mddev
->resync_max_sectors
;
7493 /* recovery follows the physical size of devices */
7494 max_sectors
= mddev
->dev_sectors
;
7497 rdev_for_each_rcu(rdev
, mddev
)
7498 if (rdev
->raid_disk
>= 0 &&
7499 !test_bit(Faulty
, &rdev
->flags
) &&
7500 !test_bit(In_sync
, &rdev
->flags
) &&
7501 rdev
->recovery_offset
< j
)
7502 j
= rdev
->recovery_offset
;
7505 /* If there is a bitmap, we need to make sure all
7506 * writes that started before we added a spare
7507 * complete before we start doing a recovery.
7508 * Otherwise the write might complete and (via
7509 * bitmap_endwrite) set a bit in the bitmap after the
7510 * recovery has checked that bit and skipped that
7513 if (mddev
->bitmap
) {
7514 mddev
->pers
->quiesce(mddev
, 1);
7515 mddev
->pers
->quiesce(mddev
, 0);
7519 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7520 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7521 " %d KB/sec/disk.\n", speed_min(mddev
));
7522 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7523 "(but not more than %d KB/sec) for %s.\n",
7524 speed_max(mddev
), desc
);
7526 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7529 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7531 mark_cnt
[m
] = io_sectors
;
7534 mddev
->resync_mark
= mark
[last_mark
];
7535 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7538 * Tune reconstruction:
7540 window
= 32*(PAGE_SIZE
/512);
7541 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7542 window
/2, (unsigned long long)max_sectors
/2);
7544 atomic_set(&mddev
->recovery_active
, 0);
7549 "md: resuming %s of %s from checkpoint.\n",
7550 desc
, mdname(mddev
));
7551 mddev
->curr_resync
= j
;
7553 mddev
->curr_resync
= 3; /* no longer delayed */
7554 mddev
->curr_resync_completed
= j
;
7555 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7556 md_new_event(mddev
);
7557 update_time
= jiffies
;
7559 blk_start_plug(&plug
);
7560 while (j
< max_sectors
) {
7565 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7566 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7567 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7568 > (max_sectors
>> 4)) ||
7569 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7570 (j
- mddev
->curr_resync_completed
)*2
7571 >= mddev
->resync_max
- mddev
->curr_resync_completed
7573 /* time to update curr_resync_completed */
7574 wait_event(mddev
->recovery_wait
,
7575 atomic_read(&mddev
->recovery_active
) == 0);
7576 mddev
->curr_resync_completed
= j
;
7577 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
7578 j
> mddev
->recovery_cp
)
7579 mddev
->recovery_cp
= j
;
7580 update_time
= jiffies
;
7581 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7582 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7585 while (j
>= mddev
->resync_max
&&
7586 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7587 /* As this condition is controlled by user-space,
7588 * we can block indefinitely, so use '_interruptible'
7589 * to avoid triggering warnings.
7591 flush_signals(current
); /* just in case */
7592 wait_event_interruptible(mddev
->recovery_wait
,
7593 mddev
->resync_max
> j
7594 || test_bit(MD_RECOVERY_INTR
,
7598 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7601 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7602 currspeed
< speed_min(mddev
));
7604 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7608 if (!skipped
) { /* actual IO requested */
7609 io_sectors
+= sectors
;
7610 atomic_add(sectors
, &mddev
->recovery_active
);
7613 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7618 mddev
->curr_resync
= j
;
7619 mddev
->curr_mark_cnt
= io_sectors
;
7620 if (last_check
== 0)
7621 /* this is the earliest that rebuild will be
7622 * visible in /proc/mdstat
7624 md_new_event(mddev
);
7626 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7629 last_check
= io_sectors
;
7631 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7633 int next
= (last_mark
+1) % SYNC_MARKS
;
7635 mddev
->resync_mark
= mark
[next
];
7636 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7637 mark
[next
] = jiffies
;
7638 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7642 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7646 * this loop exits only if either when we are slower than
7647 * the 'hard' speed limit, or the system was IO-idle for
7649 * the system might be non-idle CPU-wise, but we only care
7650 * about not overloading the IO subsystem. (things like an
7651 * e2fsck being done on the RAID array should execute fast)
7655 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7656 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7658 if (currspeed
> speed_min(mddev
)) {
7659 if ((currspeed
> speed_max(mddev
)) ||
7660 !is_mddev_idle(mddev
, 0)) {
7666 printk(KERN_INFO
"md: %s: %s %s.\n",mdname(mddev
), desc
,
7667 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
7668 ? "interrupted" : "done");
7670 * this also signals 'finished resyncing' to md_stop
7672 blk_finish_plug(&plug
);
7673 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7675 /* tell personality that we are finished */
7676 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7678 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7679 mddev
->curr_resync
> 2) {
7680 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7681 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7682 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7684 "md: checkpointing %s of %s.\n",
7685 desc
, mdname(mddev
));
7686 if (test_bit(MD_RECOVERY_ERROR
,
7688 mddev
->recovery_cp
=
7689 mddev
->curr_resync_completed
;
7691 mddev
->recovery_cp
=
7695 mddev
->recovery_cp
= MaxSector
;
7697 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7698 mddev
->curr_resync
= MaxSector
;
7700 rdev_for_each_rcu(rdev
, mddev
)
7701 if (rdev
->raid_disk
>= 0 &&
7702 mddev
->delta_disks
>= 0 &&
7703 !test_bit(Faulty
, &rdev
->flags
) &&
7704 !test_bit(In_sync
, &rdev
->flags
) &&
7705 rdev
->recovery_offset
< mddev
->curr_resync
)
7706 rdev
->recovery_offset
= mddev
->curr_resync
;
7711 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7713 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7714 /* We completed so min/max setting can be forgotten if used. */
7715 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7716 mddev
->resync_min
= 0;
7717 mddev
->resync_max
= MaxSector
;
7718 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7719 mddev
->resync_min
= mddev
->curr_resync_completed
;
7720 mddev
->curr_resync
= 0;
7721 wake_up(&resync_wait
);
7722 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7723 md_wakeup_thread(mddev
->thread
);
7726 EXPORT_SYMBOL_GPL(md_do_sync
);
7728 static int remove_and_add_spares(struct mddev
*mddev
,
7729 struct md_rdev
*this)
7731 struct md_rdev
*rdev
;
7735 rdev_for_each(rdev
, mddev
)
7736 if ((this == NULL
|| rdev
== this) &&
7737 rdev
->raid_disk
>= 0 &&
7738 !test_bit(Blocked
, &rdev
->flags
) &&
7739 (test_bit(Faulty
, &rdev
->flags
) ||
7740 ! test_bit(In_sync
, &rdev
->flags
)) &&
7741 atomic_read(&rdev
->nr_pending
)==0) {
7742 if (mddev
->pers
->hot_remove_disk(
7743 mddev
, rdev
) == 0) {
7744 sysfs_unlink_rdev(mddev
, rdev
);
7745 rdev
->raid_disk
= -1;
7749 if (removed
&& mddev
->kobj
.sd
)
7750 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
7755 rdev_for_each(rdev
, mddev
) {
7756 if (rdev
->raid_disk
>= 0 &&
7757 !test_bit(In_sync
, &rdev
->flags
) &&
7758 !test_bit(Faulty
, &rdev
->flags
))
7760 if (rdev
->raid_disk
>= 0)
7762 if (test_bit(Faulty
, &rdev
->flags
))
7765 ! (rdev
->saved_raid_disk
>= 0 &&
7766 !test_bit(Bitmap_sync
, &rdev
->flags
)))
7769 if (rdev
->saved_raid_disk
< 0)
7770 rdev
->recovery_offset
= 0;
7772 hot_add_disk(mddev
, rdev
) == 0) {
7773 if (sysfs_link_rdev(mddev
, rdev
))
7774 /* failure here is OK */;
7776 md_new_event(mddev
);
7777 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7782 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7787 * This routine is regularly called by all per-raid-array threads to
7788 * deal with generic issues like resync and super-block update.
7789 * Raid personalities that don't have a thread (linear/raid0) do not
7790 * need this as they never do any recovery or update the superblock.
7792 * It does not do any resync itself, but rather "forks" off other threads
7793 * to do that as needed.
7794 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7795 * "->recovery" and create a thread at ->sync_thread.
7796 * When the thread finishes it sets MD_RECOVERY_DONE
7797 * and wakeups up this thread which will reap the thread and finish up.
7798 * This thread also removes any faulty devices (with nr_pending == 0).
7800 * The overall approach is:
7801 * 1/ if the superblock needs updating, update it.
7802 * 2/ If a recovery thread is running, don't do anything else.
7803 * 3/ If recovery has finished, clean up, possibly marking spares active.
7804 * 4/ If there are any faulty devices, remove them.
7805 * 5/ If array is degraded, try to add spares devices
7806 * 6/ If array has spares or is not in-sync, start a resync thread.
7808 void md_check_recovery(struct mddev
*mddev
)
7810 if (mddev
->suspended
)
7814 bitmap_daemon_work(mddev
);
7816 if (signal_pending(current
)) {
7817 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7818 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7820 mddev
->safemode
= 2;
7822 flush_signals(current
);
7825 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7828 (mddev
->flags
& MD_UPDATE_SB_FLAGS
& ~ (1<<MD_CHANGE_PENDING
)) ||
7829 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7830 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7831 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7832 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7833 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7837 if (mddev_trylock(mddev
)) {
7841 /* On a read-only array we can:
7842 * - remove failed devices
7843 * - add already-in_sync devices if the array itself
7845 * As we only add devices that are already in-sync,
7846 * we can activate the spares immediately.
7848 remove_and_add_spares(mddev
, NULL
);
7849 /* There is no thread, but we need to call
7850 * ->spare_active and clear saved_raid_disk
7852 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7853 md_reap_sync_thread(mddev
);
7854 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7858 if (!mddev
->external
) {
7860 spin_lock_irq(&mddev
->write_lock
);
7861 if (mddev
->safemode
&&
7862 !atomic_read(&mddev
->writes_pending
) &&
7864 mddev
->recovery_cp
== MaxSector
) {
7867 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7869 if (mddev
->safemode
== 1)
7870 mddev
->safemode
= 0;
7871 spin_unlock_irq(&mddev
->write_lock
);
7873 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7876 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
7877 md_update_sb(mddev
, 0);
7879 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7880 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7881 /* resync/recovery still happening */
7882 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7885 if (mddev
->sync_thread
) {
7886 md_reap_sync_thread(mddev
);
7889 /* Set RUNNING before clearing NEEDED to avoid
7890 * any transients in the value of "sync_action".
7892 mddev
->curr_resync_completed
= 0;
7893 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7894 /* Clear some bits that don't mean anything, but
7897 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7898 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7900 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7901 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7903 /* no recovery is running.
7904 * remove any failed drives, then
7905 * add spares if possible.
7906 * Spares are also removed and re-added, to allow
7907 * the personality to fail the re-add.
7910 if (mddev
->reshape_position
!= MaxSector
) {
7911 if (mddev
->pers
->check_reshape
== NULL
||
7912 mddev
->pers
->check_reshape(mddev
) != 0)
7913 /* Cannot proceed */
7915 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7916 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7917 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
7918 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7919 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7920 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7921 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7922 } else if (mddev
->recovery_cp
< MaxSector
) {
7923 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7924 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7925 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7926 /* nothing to be done ... */
7929 if (mddev
->pers
->sync_request
) {
7931 /* We are adding a device or devices to an array
7932 * which has the bitmap stored on all devices.
7933 * So make sure all bitmap pages get written
7935 bitmap_write_all(mddev
->bitmap
);
7937 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7940 if (!mddev
->sync_thread
) {
7941 printk(KERN_ERR
"%s: could not start resync"
7944 /* leave the spares where they are, it shouldn't hurt */
7945 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7946 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7947 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7948 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7949 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7951 md_wakeup_thread(mddev
->sync_thread
);
7952 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7953 md_new_event(mddev
);
7956 wake_up(&mddev
->sb_wait
);
7958 if (!mddev
->sync_thread
) {
7959 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7960 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7962 if (mddev
->sysfs_action
)
7963 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7965 mddev_unlock(mddev
);
7969 void md_reap_sync_thread(struct mddev
*mddev
)
7971 struct md_rdev
*rdev
;
7973 /* resync has finished, collect result */
7974 md_unregister_thread(&mddev
->sync_thread
);
7975 wake_up(&resync_wait
);
7976 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7977 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7979 /* activate any spares */
7980 if (mddev
->pers
->spare_active(mddev
)) {
7981 sysfs_notify(&mddev
->kobj
, NULL
,
7983 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7986 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7987 mddev
->pers
->finish_reshape
)
7988 mddev
->pers
->finish_reshape(mddev
);
7990 /* If array is no-longer degraded, then any saved_raid_disk
7991 * information must be scrapped.
7993 if (!mddev
->degraded
)
7994 rdev_for_each(rdev
, mddev
)
7995 rdev
->saved_raid_disk
= -1;
7997 md_update_sb(mddev
, 1);
7998 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7999 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8000 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8001 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8002 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8003 /* flag recovery needed just to double check */
8004 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8005 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8006 md_new_event(mddev
);
8007 if (mddev
->event_work
.func
)
8008 queue_work(md_misc_wq
, &mddev
->event_work
);
8011 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
8013 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8014 wait_event_timeout(rdev
->blocked_wait
,
8015 !test_bit(Blocked
, &rdev
->flags
) &&
8016 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
8017 msecs_to_jiffies(5000));
8018 rdev_dec_pending(rdev
, mddev
);
8020 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
8022 void md_finish_reshape(struct mddev
*mddev
)
8024 /* called be personality module when reshape completes. */
8025 struct md_rdev
*rdev
;
8027 rdev_for_each(rdev
, mddev
) {
8028 if (rdev
->data_offset
> rdev
->new_data_offset
)
8029 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
8031 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
8032 rdev
->data_offset
= rdev
->new_data_offset
;
8035 EXPORT_SYMBOL(md_finish_reshape
);
8037 /* Bad block management.
8038 * We can record which blocks on each device are 'bad' and so just
8039 * fail those blocks, or that stripe, rather than the whole device.
8040 * Entries in the bad-block table are 64bits wide. This comprises:
8041 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8042 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8043 * A 'shift' can be set so that larger blocks are tracked and
8044 * consequently larger devices can be covered.
8045 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8047 * Locking of the bad-block table uses a seqlock so md_is_badblock
8048 * might need to retry if it is very unlucky.
8049 * We will sometimes want to check for bad blocks in a bi_end_io function,
8050 * so we use the write_seqlock_irq variant.
8052 * When looking for a bad block we specify a range and want to
8053 * know if any block in the range is bad. So we binary-search
8054 * to the last range that starts at-or-before the given endpoint,
8055 * (or "before the sector after the target range")
8056 * then see if it ends after the given start.
8058 * 0 if there are no known bad blocks in the range
8059 * 1 if there are known bad block which are all acknowledged
8060 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8061 * plus the start/length of the first bad section we overlap.
8063 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
8064 sector_t
*first_bad
, int *bad_sectors
)
8070 sector_t target
= s
+ sectors
;
8073 if (bb
->shift
> 0) {
8074 /* round the start down, and the end up */
8076 target
+= (1<<bb
->shift
) - 1;
8077 target
>>= bb
->shift
;
8078 sectors
= target
- s
;
8080 /* 'target' is now the first block after the bad range */
8083 seq
= read_seqbegin(&bb
->lock
);
8088 /* Binary search between lo and hi for 'target'
8089 * i.e. for the last range that starts before 'target'
8091 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8092 * are known not to be the last range before target.
8093 * VARIANT: hi-lo is the number of possible
8094 * ranges, and decreases until it reaches 1
8096 while (hi
- lo
> 1) {
8097 int mid
= (lo
+ hi
) / 2;
8098 sector_t a
= BB_OFFSET(p
[mid
]);
8100 /* This could still be the one, earlier ranges
8104 /* This and later ranges are definitely out. */
8107 /* 'lo' might be the last that started before target, but 'hi' isn't */
8109 /* need to check all range that end after 's' to see if
8110 * any are unacknowledged.
8113 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8114 if (BB_OFFSET(p
[lo
]) < target
) {
8115 /* starts before the end, and finishes after
8116 * the start, so they must overlap
8118 if (rv
!= -1 && BB_ACK(p
[lo
]))
8122 *first_bad
= BB_OFFSET(p
[lo
]);
8123 *bad_sectors
= BB_LEN(p
[lo
]);
8129 if (read_seqretry(&bb
->lock
, seq
))
8134 EXPORT_SYMBOL_GPL(md_is_badblock
);
8137 * Add a range of bad blocks to the table.
8138 * This might extend the table, or might contract it
8139 * if two adjacent ranges can be merged.
8140 * We binary-search to find the 'insertion' point, then
8141 * decide how best to handle it.
8143 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
8149 unsigned long flags
;
8152 /* badblocks are disabled */
8156 /* round the start down, and the end up */
8157 sector_t next
= s
+ sectors
;
8159 next
+= (1<<bb
->shift
) - 1;
8164 write_seqlock_irqsave(&bb
->lock
, flags
);
8169 /* Find the last range that starts at-or-before 's' */
8170 while (hi
- lo
> 1) {
8171 int mid
= (lo
+ hi
) / 2;
8172 sector_t a
= BB_OFFSET(p
[mid
]);
8178 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8182 /* we found a range that might merge with the start
8185 sector_t a
= BB_OFFSET(p
[lo
]);
8186 sector_t e
= a
+ BB_LEN(p
[lo
]);
8187 int ack
= BB_ACK(p
[lo
]);
8189 /* Yes, we can merge with a previous range */
8190 if (s
== a
&& s
+ sectors
>= e
)
8191 /* new range covers old */
8194 ack
= ack
&& acknowledged
;
8196 if (e
< s
+ sectors
)
8198 if (e
- a
<= BB_MAX_LEN
) {
8199 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8202 /* does not all fit in one range,
8203 * make p[lo] maximal
8205 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8206 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8212 if (sectors
&& hi
< bb
->count
) {
8213 /* 'hi' points to the first range that starts after 's'.
8214 * Maybe we can merge with the start of that range */
8215 sector_t a
= BB_OFFSET(p
[hi
]);
8216 sector_t e
= a
+ BB_LEN(p
[hi
]);
8217 int ack
= BB_ACK(p
[hi
]);
8218 if (a
<= s
+ sectors
) {
8219 /* merging is possible */
8220 if (e
<= s
+ sectors
) {
8225 ack
= ack
&& acknowledged
;
8228 if (e
- a
<= BB_MAX_LEN
) {
8229 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8232 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8240 if (sectors
== 0 && hi
< bb
->count
) {
8241 /* we might be able to combine lo and hi */
8242 /* Note: 's' is at the end of 'lo' */
8243 sector_t a
= BB_OFFSET(p
[hi
]);
8244 int lolen
= BB_LEN(p
[lo
]);
8245 int hilen
= BB_LEN(p
[hi
]);
8246 int newlen
= lolen
+ hilen
- (s
- a
);
8247 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8248 /* yes, we can combine them */
8249 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8250 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8251 memmove(p
+ hi
, p
+ hi
+ 1,
8252 (bb
->count
- hi
- 1) * 8);
8257 /* didn't merge (it all).
8258 * Need to add a range just before 'hi' */
8259 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8260 /* No room for more */
8264 int this_sectors
= sectors
;
8265 memmove(p
+ hi
+ 1, p
+ hi
,
8266 (bb
->count
- hi
) * 8);
8269 if (this_sectors
> BB_MAX_LEN
)
8270 this_sectors
= BB_MAX_LEN
;
8271 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8272 sectors
-= this_sectors
;
8279 bb
->unacked_exist
= 1;
8280 write_sequnlock_irqrestore(&bb
->lock
, flags
);
8285 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8290 s
+= rdev
->new_data_offset
;
8292 s
+= rdev
->data_offset
;
8293 rv
= md_set_badblocks(&rdev
->badblocks
,
8296 /* Make sure they get written out promptly */
8297 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8298 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8299 md_wakeup_thread(rdev
->mddev
->thread
);
8303 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8306 * Remove a range of bad blocks from the table.
8307 * This may involve extending the table if we spilt a region,
8308 * but it must not fail. So if the table becomes full, we just
8309 * drop the remove request.
8311 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8315 sector_t target
= s
+ sectors
;
8318 if (bb
->shift
> 0) {
8319 /* When clearing we round the start up and the end down.
8320 * This should not matter as the shift should align with
8321 * the block size and no rounding should ever be needed.
8322 * However it is better the think a block is bad when it
8323 * isn't than to think a block is not bad when it is.
8325 s
+= (1<<bb
->shift
) - 1;
8327 target
>>= bb
->shift
;
8328 sectors
= target
- s
;
8331 write_seqlock_irq(&bb
->lock
);
8336 /* Find the last range that starts before 'target' */
8337 while (hi
- lo
> 1) {
8338 int mid
= (lo
+ hi
) / 2;
8339 sector_t a
= BB_OFFSET(p
[mid
]);
8346 /* p[lo] is the last range that could overlap the
8347 * current range. Earlier ranges could also overlap,
8348 * but only this one can overlap the end of the range.
8350 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8351 /* Partial overlap, leave the tail of this range */
8352 int ack
= BB_ACK(p
[lo
]);
8353 sector_t a
= BB_OFFSET(p
[lo
]);
8354 sector_t end
= a
+ BB_LEN(p
[lo
]);
8357 /* we need to split this range */
8358 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8362 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8364 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8367 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8368 /* there is no longer an overlap */
8373 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8374 /* This range does overlap */
8375 if (BB_OFFSET(p
[lo
]) < s
) {
8376 /* Keep the early parts of this range. */
8377 int ack
= BB_ACK(p
[lo
]);
8378 sector_t start
= BB_OFFSET(p
[lo
]);
8379 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8380 /* now low doesn't overlap, so.. */
8385 /* 'lo' is strictly before, 'hi' is strictly after,
8386 * anything between needs to be discarded
8389 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8390 bb
->count
-= (hi
- lo
- 1);
8396 write_sequnlock_irq(&bb
->lock
);
8400 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8404 s
+= rdev
->new_data_offset
;
8406 s
+= rdev
->data_offset
;
8407 return md_clear_badblocks(&rdev
->badblocks
,
8410 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8413 * Acknowledge all bad blocks in a list.
8414 * This only succeeds if ->changed is clear. It is used by
8415 * in-kernel metadata updates
8417 void md_ack_all_badblocks(struct badblocks
*bb
)
8419 if (bb
->page
== NULL
|| bb
->changed
)
8420 /* no point even trying */
8422 write_seqlock_irq(&bb
->lock
);
8424 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8427 for (i
= 0; i
< bb
->count
; i
++) {
8428 if (!BB_ACK(p
[i
])) {
8429 sector_t start
= BB_OFFSET(p
[i
]);
8430 int len
= BB_LEN(p
[i
]);
8431 p
[i
] = BB_MAKE(start
, len
, 1);
8434 bb
->unacked_exist
= 0;
8436 write_sequnlock_irq(&bb
->lock
);
8438 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8440 /* sysfs access to bad-blocks list.
8441 * We present two files.
8442 * 'bad-blocks' lists sector numbers and lengths of ranges that
8443 * are recorded as bad. The list is truncated to fit within
8444 * the one-page limit of sysfs.
8445 * Writing "sector length" to this file adds an acknowledged
8447 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8448 * been acknowledged. Writing to this file adds bad blocks
8449 * without acknowledging them. This is largely for testing.
8453 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8464 seq
= read_seqbegin(&bb
->lock
);
8469 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8470 sector_t s
= BB_OFFSET(p
[i
]);
8471 unsigned int length
= BB_LEN(p
[i
]);
8472 int ack
= BB_ACK(p
[i
]);
8478 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8479 (unsigned long long)s
<< bb
->shift
,
8480 length
<< bb
->shift
);
8482 if (unack
&& len
== 0)
8483 bb
->unacked_exist
= 0;
8485 if (read_seqretry(&bb
->lock
, seq
))
8494 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8496 unsigned long long sector
;
8500 /* Allow clearing via sysfs *only* for testing/debugging.
8501 * Normally only a successful write may clear a badblock
8504 if (page
[0] == '-') {
8508 #endif /* DO_DEBUG */
8510 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8512 if (newline
!= '\n')
8524 md_clear_badblocks(bb
, sector
, length
);
8527 #endif /* DO_DEBUG */
8528 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8534 static int md_notify_reboot(struct notifier_block
*this,
8535 unsigned long code
, void *x
)
8537 struct list_head
*tmp
;
8538 struct mddev
*mddev
;
8541 for_each_mddev(mddev
, tmp
) {
8542 if (mddev_trylock(mddev
)) {
8544 __md_stop_writes(mddev
);
8545 if (mddev
->persistent
)
8546 mddev
->safemode
= 2;
8547 mddev_unlock(mddev
);
8552 * certain more exotic SCSI devices are known to be
8553 * volatile wrt too early system reboots. While the
8554 * right place to handle this issue is the given
8555 * driver, we do want to have a safe RAID driver ...
8563 static struct notifier_block md_notifier
= {
8564 .notifier_call
= md_notify_reboot
,
8566 .priority
= INT_MAX
, /* before any real devices */
8569 static void md_geninit(void)
8571 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8573 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8576 static int __init
md_init(void)
8580 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8584 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8588 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8591 if ((ret
= register_blkdev(0, "mdp")) < 0)
8595 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8596 md_probe
, NULL
, NULL
);
8597 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8598 md_probe
, NULL
, NULL
);
8600 register_reboot_notifier(&md_notifier
);
8601 raid_table_header
= register_sysctl_table(raid_root_table
);
8607 unregister_blkdev(MD_MAJOR
, "md");
8609 destroy_workqueue(md_misc_wq
);
8611 destroy_workqueue(md_wq
);
8619 * Searches all registered partitions for autorun RAID arrays
8623 static LIST_HEAD(all_detected_devices
);
8624 struct detected_devices_node
{
8625 struct list_head list
;
8629 void md_autodetect_dev(dev_t dev
)
8631 struct detected_devices_node
*node_detected_dev
;
8633 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8634 if (node_detected_dev
) {
8635 node_detected_dev
->dev
= dev
;
8636 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8638 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8639 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8644 static void autostart_arrays(int part
)
8646 struct md_rdev
*rdev
;
8647 struct detected_devices_node
*node_detected_dev
;
8649 int i_scanned
, i_passed
;
8654 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8656 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8658 node_detected_dev
= list_entry(all_detected_devices
.next
,
8659 struct detected_devices_node
, list
);
8660 list_del(&node_detected_dev
->list
);
8661 dev
= node_detected_dev
->dev
;
8662 kfree(node_detected_dev
);
8663 rdev
= md_import_device(dev
,0, 90);
8667 if (test_bit(Faulty
, &rdev
->flags
)) {
8671 set_bit(AutoDetected
, &rdev
->flags
);
8672 list_add(&rdev
->same_set
, &pending_raid_disks
);
8676 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8677 i_scanned
, i_passed
);
8679 autorun_devices(part
);
8682 #endif /* !MODULE */
8684 static __exit
void md_exit(void)
8686 struct mddev
*mddev
;
8687 struct list_head
*tmp
;
8690 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8691 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8693 unregister_blkdev(MD_MAJOR
,"md");
8694 unregister_blkdev(mdp_major
, "mdp");
8695 unregister_reboot_notifier(&md_notifier
);
8696 unregister_sysctl_table(raid_table_header
);
8698 /* We cannot unload the modules while some process is
8699 * waiting for us in select() or poll() - wake them up
8702 while (waitqueue_active(&md_event_waiters
)) {
8703 /* not safe to leave yet */
8704 wake_up(&md_event_waiters
);
8708 remove_proc_entry("mdstat", NULL
);
8710 for_each_mddev(mddev
, tmp
) {
8711 export_array(mddev
);
8712 mddev
->hold_active
= 0;
8714 destroy_workqueue(md_misc_wq
);
8715 destroy_workqueue(md_wq
);
8718 subsys_initcall(md_init
);
8719 module_exit(md_exit
)
8721 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8723 return sprintf(buffer
, "%d", start_readonly
);
8725 static int set_ro(const char *val
, struct kernel_param
*kp
)
8728 int num
= simple_strtoul(val
, &e
, 10);
8729 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8730 start_readonly
= num
;
8736 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8737 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8739 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8741 EXPORT_SYMBOL(register_md_personality
);
8742 EXPORT_SYMBOL(unregister_md_personality
);
8743 EXPORT_SYMBOL(md_error
);
8744 EXPORT_SYMBOL(md_done_sync
);
8745 EXPORT_SYMBOL(md_write_start
);
8746 EXPORT_SYMBOL(md_write_end
);
8747 EXPORT_SYMBOL(md_register_thread
);
8748 EXPORT_SYMBOL(md_unregister_thread
);
8749 EXPORT_SYMBOL(md_wakeup_thread
);
8750 EXPORT_SYMBOL(md_check_recovery
);
8751 EXPORT_SYMBOL(md_reap_sync_thread
);
8752 MODULE_LICENSE("GPL");
8753 MODULE_DESCRIPTION("MD RAID framework");
8755 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);