2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min
= 1000;
97 static int sysctl_speed_limit_max
= 200000;
98 static inline int speed_min(struct mddev
*mddev
)
100 return mddev
->sync_speed_min
?
101 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
104 static inline int speed_max(struct mddev
*mddev
)
106 return mddev
->sync_speed_max
?
107 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
110 static struct ctl_table_header
*raid_table_header
;
112 static ctl_table raid_table
[] = {
114 .procname
= "speed_limit_min",
115 .data
= &sysctl_speed_limit_min
,
116 .maxlen
= sizeof(int),
117 .mode
= S_IRUGO
|S_IWUSR
,
118 .proc_handler
= proc_dointvec
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
134 .mode
= S_IRUGO
|S_IXUGO
,
140 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static const struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio
*bio
)
160 struct mddev
*mddev
, **mddevp
;
165 bio_free(bio
, mddev
->bio_set
);
168 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
172 struct mddev
**mddevp
;
174 if (!mddev
|| !mddev
->bio_set
)
175 return bio_alloc(gfp_mask
, nr_iovecs
);
177 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
183 b
->bi_destructor
= mddev_bio_destructor
;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
188 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
192 struct mddev
**mddevp
;
194 if (!mddev
|| !mddev
->bio_set
)
195 return bio_clone(bio
, gfp_mask
);
197 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
203 b
->bi_destructor
= mddev_bio_destructor
;
205 if (bio_integrity(bio
)) {
208 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
218 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
220 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
227 struct bio_vec
*bvec
;
231 if (offset
== 0 && size
== bio
->bi_size
)
234 bio
->bi_sector
+= offset
;
237 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
239 while (bio
->bi_idx
< bio
->bi_vcnt
&&
240 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
241 /* remove this whole bio_vec */
242 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
245 if (bio
->bi_idx
< bio
->bi_vcnt
) {
246 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
247 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
249 /* avoid any complications with bi_idx being non-zero*/
251 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
252 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
253 bio
->bi_vcnt
-= bio
->bi_idx
;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec
, bio
, i
) {
258 if (sofar
+ bvec
->bv_len
> size
)
259 bvec
->bv_len
= size
- sofar
;
260 if (bvec
->bv_len
== 0) {
264 sofar
+= bvec
->bv_len
;
267 EXPORT_SYMBOL_GPL(md_trim_bio
);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
280 static atomic_t md_event_count
;
281 void md_new_event(struct mddev
*mddev
)
283 atomic_inc(&md_event_count
);
284 wake_up(&md_event_waiters
);
286 EXPORT_SYMBOL_GPL(md_new_event
);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev
*mddev
)
293 atomic_inc(&md_event_count
);
294 wake_up(&md_event_waiters
);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs
);
302 static DEFINE_SPINLOCK(all_mddevs_lock
);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
337 const int rw
= bio_data_dir(bio
);
338 struct mddev
*mddev
= q
->queuedata
;
340 unsigned int sectors
;
342 if (mddev
== NULL
|| mddev
->pers
== NULL
347 smp_rmb(); /* Ensure implications of 'active' are visible */
349 if (mddev
->suspended
) {
352 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
353 TASK_UNINTERRUPTIBLE
);
354 if (!mddev
->suspended
)
360 finish_wait(&mddev
->sb_wait
, &__wait
);
362 atomic_inc(&mddev
->active_io
);
366 * save the sectors now since our bio can
367 * go away inside make_request
369 sectors
= bio_sectors(bio
);
370 mddev
->pers
->make_request(mddev
, bio
);
372 cpu
= part_stat_lock();
373 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
374 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
377 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
378 wake_up(&mddev
->sb_wait
);
381 /* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
387 void mddev_suspend(struct mddev
*mddev
)
389 BUG_ON(mddev
->suspended
);
390 mddev
->suspended
= 1;
392 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
393 mddev
->pers
->quiesce(mddev
, 1);
395 del_timer_sync(&mddev
->safemode_timer
);
397 EXPORT_SYMBOL_GPL(mddev_suspend
);
399 void mddev_resume(struct mddev
*mddev
)
401 mddev
->suspended
= 0;
402 wake_up(&mddev
->sb_wait
);
403 mddev
->pers
->quiesce(mddev
, 0);
405 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
406 md_wakeup_thread(mddev
->thread
);
407 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
409 EXPORT_SYMBOL_GPL(mddev_resume
);
411 int mddev_congested(struct mddev
*mddev
, int bits
)
413 return mddev
->suspended
;
415 EXPORT_SYMBOL(mddev_congested
);
418 * Generic flush handling for md
421 static void md_end_flush(struct bio
*bio
, int err
)
423 struct md_rdev
*rdev
= bio
->bi_private
;
424 struct mddev
*mddev
= rdev
->mddev
;
426 rdev_dec_pending(rdev
, mddev
);
428 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
429 /* The pre-request flush has finished */
430 queue_work(md_wq
, &mddev
->flush_work
);
435 static void md_submit_flush_data(struct work_struct
*ws
);
437 static void submit_flushes(struct work_struct
*ws
)
439 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
440 struct md_rdev
*rdev
;
442 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
443 atomic_set(&mddev
->flush_pending
, 1);
445 rdev_for_each_rcu(rdev
, mddev
)
446 if (rdev
->raid_disk
>= 0 &&
447 !test_bit(Faulty
, &rdev
->flags
)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
453 atomic_inc(&rdev
->nr_pending
);
454 atomic_inc(&rdev
->nr_pending
);
456 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
457 bi
->bi_end_io
= md_end_flush
;
458 bi
->bi_private
= rdev
;
459 bi
->bi_bdev
= rdev
->bdev
;
460 atomic_inc(&mddev
->flush_pending
);
461 submit_bio(WRITE_FLUSH
, bi
);
463 rdev_dec_pending(rdev
, mddev
);
466 if (atomic_dec_and_test(&mddev
->flush_pending
))
467 queue_work(md_wq
, &mddev
->flush_work
);
470 static void md_submit_flush_data(struct work_struct
*ws
)
472 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
473 struct bio
*bio
= mddev
->flush_bio
;
475 if (bio
->bi_size
== 0)
476 /* an empty barrier - all done */
479 bio
->bi_rw
&= ~REQ_FLUSH
;
480 mddev
->pers
->make_request(mddev
, bio
);
483 mddev
->flush_bio
= NULL
;
484 wake_up(&mddev
->sb_wait
);
487 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
489 spin_lock_irq(&mddev
->write_lock
);
490 wait_event_lock_irq(mddev
->sb_wait
,
492 mddev
->write_lock
, /*nothing*/);
493 mddev
->flush_bio
= bio
;
494 spin_unlock_irq(&mddev
->write_lock
);
496 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
497 queue_work(md_wq
, &mddev
->flush_work
);
499 EXPORT_SYMBOL(md_flush_request
);
501 /* Support for plugging.
502 * This mirrors the plugging support in request_queue, but does not
503 * require having a whole queue or request structures.
504 * We allocate an md_plug_cb for each md device and each thread it gets
505 * plugged on. This links tot the private plug_handle structure in the
506 * personality data where we keep a count of the number of outstanding
507 * plugs so other code can see if a plug is active.
510 struct blk_plug_cb cb
;
514 static void plugger_unplug(struct blk_plug_cb
*cb
)
516 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
517 md_wakeup_thread(mdcb
->mddev
->thread
);
521 /* Check that an unplug wakeup will come shortly.
522 * If not, wakeup the md thread immediately
524 int mddev_check_plugged(struct mddev
*mddev
)
526 struct blk_plug
*plug
= current
->plug
;
527 struct md_plug_cb
*mdcb
;
532 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
533 if (mdcb
->cb
.callback
== plugger_unplug
&&
534 mdcb
->mddev
== mddev
) {
535 /* Already on the list, move to top */
536 if (mdcb
!= list_first_entry(&plug
->cb_list
,
539 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
543 /* Not currently on the callback list */
544 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
549 mdcb
->cb
.callback
= plugger_unplug
;
550 list_add(&mdcb
->cb
.list
, &plug
->cb_list
);
553 EXPORT_SYMBOL_GPL(mddev_check_plugged
);
555 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
557 atomic_inc(&mddev
->active
);
561 static void mddev_delayed_delete(struct work_struct
*ws
);
563 static void mddev_put(struct mddev
*mddev
)
565 struct bio_set
*bs
= NULL
;
567 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
569 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
570 mddev
->ctime
== 0 && !mddev
->hold_active
) {
571 /* Array is not configured at all, and not held active,
573 list_del_init(&mddev
->all_mddevs
);
575 mddev
->bio_set
= NULL
;
576 if (mddev
->gendisk
) {
577 /* We did a probe so need to clean up. Call
578 * queue_work inside the spinlock so that
579 * flush_workqueue() after mddev_find will
580 * succeed in waiting for the work to be done.
582 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
583 queue_work(md_misc_wq
, &mddev
->del_work
);
587 spin_unlock(&all_mddevs_lock
);
592 void mddev_init(struct mddev
*mddev
)
594 mutex_init(&mddev
->open_mutex
);
595 mutex_init(&mddev
->reconfig_mutex
);
596 mutex_init(&mddev
->bitmap_info
.mutex
);
597 INIT_LIST_HEAD(&mddev
->disks
);
598 INIT_LIST_HEAD(&mddev
->all_mddevs
);
599 init_timer(&mddev
->safemode_timer
);
600 atomic_set(&mddev
->active
, 1);
601 atomic_set(&mddev
->openers
, 0);
602 atomic_set(&mddev
->active_io
, 0);
603 spin_lock_init(&mddev
->write_lock
);
604 atomic_set(&mddev
->flush_pending
, 0);
605 init_waitqueue_head(&mddev
->sb_wait
);
606 init_waitqueue_head(&mddev
->recovery_wait
);
607 mddev
->reshape_position
= MaxSector
;
608 mddev
->reshape_backwards
= 0;
609 mddev
->resync_min
= 0;
610 mddev
->resync_max
= MaxSector
;
611 mddev
->level
= LEVEL_NONE
;
613 EXPORT_SYMBOL_GPL(mddev_init
);
615 static struct mddev
* mddev_find(dev_t unit
)
617 struct mddev
*mddev
, *new = NULL
;
619 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
620 unit
&= ~((1<<MdpMinorShift
)-1);
623 spin_lock(&all_mddevs_lock
);
626 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
627 if (mddev
->unit
== unit
) {
629 spin_unlock(&all_mddevs_lock
);
635 list_add(&new->all_mddevs
, &all_mddevs
);
636 spin_unlock(&all_mddevs_lock
);
637 new->hold_active
= UNTIL_IOCTL
;
641 /* find an unused unit number */
642 static int next_minor
= 512;
643 int start
= next_minor
;
647 dev
= MKDEV(MD_MAJOR
, next_minor
);
649 if (next_minor
> MINORMASK
)
651 if (next_minor
== start
) {
652 /* Oh dear, all in use. */
653 spin_unlock(&all_mddevs_lock
);
659 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
660 if (mddev
->unit
== dev
) {
666 new->md_minor
= MINOR(dev
);
667 new->hold_active
= UNTIL_STOP
;
668 list_add(&new->all_mddevs
, &all_mddevs
);
669 spin_unlock(&all_mddevs_lock
);
672 spin_unlock(&all_mddevs_lock
);
674 new = kzalloc(sizeof(*new), GFP_KERNEL
);
679 if (MAJOR(unit
) == MD_MAJOR
)
680 new->md_minor
= MINOR(unit
);
682 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
689 static inline int mddev_lock(struct mddev
* mddev
)
691 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
694 static inline int mddev_is_locked(struct mddev
*mddev
)
696 return mutex_is_locked(&mddev
->reconfig_mutex
);
699 static inline int mddev_trylock(struct mddev
* mddev
)
701 return mutex_trylock(&mddev
->reconfig_mutex
);
704 static struct attribute_group md_redundancy_group
;
706 static void mddev_unlock(struct mddev
* mddev
)
708 if (mddev
->to_remove
) {
709 /* These cannot be removed under reconfig_mutex as
710 * an access to the files will try to take reconfig_mutex
711 * while holding the file unremovable, which leads to
713 * So hold set sysfs_active while the remove in happeing,
714 * and anything else which might set ->to_remove or my
715 * otherwise change the sysfs namespace will fail with
716 * -EBUSY if sysfs_active is still set.
717 * We set sysfs_active under reconfig_mutex and elsewhere
718 * test it under the same mutex to ensure its correct value
721 struct attribute_group
*to_remove
= mddev
->to_remove
;
722 mddev
->to_remove
= NULL
;
723 mddev
->sysfs_active
= 1;
724 mutex_unlock(&mddev
->reconfig_mutex
);
726 if (mddev
->kobj
.sd
) {
727 if (to_remove
!= &md_redundancy_group
)
728 sysfs_remove_group(&mddev
->kobj
, to_remove
);
729 if (mddev
->pers
== NULL
||
730 mddev
->pers
->sync_request
== NULL
) {
731 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
732 if (mddev
->sysfs_action
)
733 sysfs_put(mddev
->sysfs_action
);
734 mddev
->sysfs_action
= NULL
;
737 mddev
->sysfs_active
= 0;
739 mutex_unlock(&mddev
->reconfig_mutex
);
741 /* As we've dropped the mutex we need a spinlock to
742 * make sure the thread doesn't disappear
744 spin_lock(&pers_lock
);
745 md_wakeup_thread(mddev
->thread
);
746 spin_unlock(&pers_lock
);
749 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
751 struct md_rdev
*rdev
;
753 rdev_for_each(rdev
, mddev
)
754 if (rdev
->desc_nr
== nr
)
760 static struct md_rdev
* find_rdev(struct mddev
* mddev
, dev_t dev
)
762 struct md_rdev
*rdev
;
764 rdev_for_each(rdev
, mddev
)
765 if (rdev
->bdev
->bd_dev
== dev
)
771 static struct md_personality
*find_pers(int level
, char *clevel
)
773 struct md_personality
*pers
;
774 list_for_each_entry(pers
, &pers_list
, list
) {
775 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
777 if (strcmp(pers
->name
, clevel
)==0)
783 /* return the offset of the super block in 512byte sectors */
784 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
786 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
787 return MD_NEW_SIZE_SECTORS(num_sectors
);
790 static int alloc_disk_sb(struct md_rdev
* rdev
)
795 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
796 if (!rdev
->sb_page
) {
797 printk(KERN_ALERT
"md: out of memory.\n");
804 void md_rdev_clear(struct md_rdev
*rdev
)
807 put_page(rdev
->sb_page
);
809 rdev
->sb_page
= NULL
;
814 put_page(rdev
->bb_page
);
815 rdev
->bb_page
= NULL
;
817 kfree(rdev
->badblocks
.page
);
818 rdev
->badblocks
.page
= NULL
;
820 EXPORT_SYMBOL_GPL(md_rdev_clear
);
822 static void super_written(struct bio
*bio
, int error
)
824 struct md_rdev
*rdev
= bio
->bi_private
;
825 struct mddev
*mddev
= rdev
->mddev
;
827 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
828 printk("md: super_written gets error=%d, uptodate=%d\n",
829 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
830 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
831 md_error(mddev
, rdev
);
834 if (atomic_dec_and_test(&mddev
->pending_writes
))
835 wake_up(&mddev
->sb_wait
);
839 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
840 sector_t sector
, int size
, struct page
*page
)
842 /* write first size bytes of page to sector of rdev
843 * Increment mddev->pending_writes before returning
844 * and decrement it on completion, waking up sb_wait
845 * if zero is reached.
846 * If an error occurred, call md_error
848 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
850 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
851 bio
->bi_sector
= sector
;
852 bio_add_page(bio
, page
, size
, 0);
853 bio
->bi_private
= rdev
;
854 bio
->bi_end_io
= super_written
;
856 atomic_inc(&mddev
->pending_writes
);
857 submit_bio(WRITE_FLUSH_FUA
, bio
);
860 void md_super_wait(struct mddev
*mddev
)
862 /* wait for all superblock writes that were scheduled to complete */
865 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
866 if (atomic_read(&mddev
->pending_writes
)==0)
870 finish_wait(&mddev
->sb_wait
, &wq
);
873 static void bi_complete(struct bio
*bio
, int error
)
875 complete((struct completion
*)bio
->bi_private
);
878 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
879 struct page
*page
, int rw
, bool metadata_op
)
881 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
882 struct completion event
;
887 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
888 rdev
->meta_bdev
: rdev
->bdev
;
890 bio
->bi_sector
= sector
+ rdev
->sb_start
;
891 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
892 (rdev
->mddev
->reshape_backwards
==
893 (sector
>= rdev
->mddev
->reshape_position
)))
894 bio
->bi_sector
= sector
+ rdev
->new_data_offset
;
896 bio
->bi_sector
= sector
+ rdev
->data_offset
;
897 bio_add_page(bio
, page
, size
, 0);
898 init_completion(&event
);
899 bio
->bi_private
= &event
;
900 bio
->bi_end_io
= bi_complete
;
902 wait_for_completion(&event
);
904 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
908 EXPORT_SYMBOL_GPL(sync_page_io
);
910 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
912 char b
[BDEVNAME_SIZE
];
913 if (!rdev
->sb_page
) {
921 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
927 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
928 bdevname(rdev
->bdev
,b
));
932 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
934 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
935 sb1
->set_uuid1
== sb2
->set_uuid1
&&
936 sb1
->set_uuid2
== sb2
->set_uuid2
&&
937 sb1
->set_uuid3
== sb2
->set_uuid3
;
940 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
943 mdp_super_t
*tmp1
, *tmp2
;
945 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
946 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
948 if (!tmp1
|| !tmp2
) {
950 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
958 * nr_disks is not constant
963 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
971 static u32
md_csum_fold(u32 csum
)
973 csum
= (csum
& 0xffff) + (csum
>> 16);
974 return (csum
& 0xffff) + (csum
>> 16);
977 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
980 u32
*sb32
= (u32
*)sb
;
982 unsigned int disk_csum
, csum
;
984 disk_csum
= sb
->sb_csum
;
987 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
989 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
993 /* This used to use csum_partial, which was wrong for several
994 * reasons including that different results are returned on
995 * different architectures. It isn't critical that we get exactly
996 * the same return value as before (we always csum_fold before
997 * testing, and that removes any differences). However as we
998 * know that csum_partial always returned a 16bit value on
999 * alphas, do a fold to maximise conformity to previous behaviour.
1001 sb
->sb_csum
= md_csum_fold(disk_csum
);
1003 sb
->sb_csum
= disk_csum
;
1010 * Handle superblock details.
1011 * We want to be able to handle multiple superblock formats
1012 * so we have a common interface to them all, and an array of
1013 * different handlers.
1014 * We rely on user-space to write the initial superblock, and support
1015 * reading and updating of superblocks.
1016 * Interface methods are:
1017 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1018 * loads and validates a superblock on dev.
1019 * if refdev != NULL, compare superblocks on both devices
1021 * 0 - dev has a superblock that is compatible with refdev
1022 * 1 - dev has a superblock that is compatible and newer than refdev
1023 * so dev should be used as the refdev in future
1024 * -EINVAL superblock incompatible or invalid
1025 * -othererror e.g. -EIO
1027 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1028 * Verify that dev is acceptable into mddev.
1029 * The first time, mddev->raid_disks will be 0, and data from
1030 * dev should be merged in. Subsequent calls check that dev
1031 * is new enough. Return 0 or -EINVAL
1033 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1034 * Update the superblock for rdev with data in mddev
1035 * This does not write to disc.
1041 struct module
*owner
;
1042 int (*load_super
)(struct md_rdev
*rdev
,
1043 struct md_rdev
*refdev
,
1045 int (*validate_super
)(struct mddev
*mddev
,
1046 struct md_rdev
*rdev
);
1047 void (*sync_super
)(struct mddev
*mddev
,
1048 struct md_rdev
*rdev
);
1049 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1050 sector_t num_sectors
);
1051 int (*allow_new_offset
)(struct md_rdev
*rdev
,
1052 unsigned long long new_offset
);
1056 * Check that the given mddev has no bitmap.
1058 * This function is called from the run method of all personalities that do not
1059 * support bitmaps. It prints an error message and returns non-zero if mddev
1060 * has a bitmap. Otherwise, it returns 0.
1063 int md_check_no_bitmap(struct mddev
*mddev
)
1065 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1067 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1068 mdname(mddev
), mddev
->pers
->name
);
1071 EXPORT_SYMBOL(md_check_no_bitmap
);
1074 * load_super for 0.90.0
1076 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1078 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1083 * Calculate the position of the superblock (512byte sectors),
1084 * it's at the end of the disk.
1086 * It also happens to be a multiple of 4Kb.
1088 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1090 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1091 if (ret
) return ret
;
1095 bdevname(rdev
->bdev
, b
);
1096 sb
= page_address(rdev
->sb_page
);
1098 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1099 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1104 if (sb
->major_version
!= 0 ||
1105 sb
->minor_version
< 90 ||
1106 sb
->minor_version
> 91) {
1107 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1108 sb
->major_version
, sb
->minor_version
,
1113 if (sb
->raid_disks
<= 0)
1116 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1117 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1122 rdev
->preferred_minor
= sb
->md_minor
;
1123 rdev
->data_offset
= 0;
1124 rdev
->new_data_offset
= 0;
1125 rdev
->sb_size
= MD_SB_BYTES
;
1126 rdev
->badblocks
.shift
= -1;
1128 if (sb
->level
== LEVEL_MULTIPATH
)
1131 rdev
->desc_nr
= sb
->this_disk
.number
;
1137 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1138 if (!uuid_equal(refsb
, sb
)) {
1139 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1140 b
, bdevname(refdev
->bdev
,b2
));
1143 if (!sb_equal(refsb
, sb
)) {
1144 printk(KERN_WARNING
"md: %s has same UUID"
1145 " but different superblock to %s\n",
1146 b
, bdevname(refdev
->bdev
, b2
));
1150 ev2
= md_event(refsb
);
1156 rdev
->sectors
= rdev
->sb_start
;
1157 /* Limit to 4TB as metadata cannot record more than that */
1158 if (rdev
->sectors
>= (2ULL << 32))
1159 rdev
->sectors
= (2ULL << 32) - 2;
1161 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1162 /* "this cannot possibly happen" ... */
1170 * validate_super for 0.90.0
1172 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1175 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1176 __u64 ev1
= md_event(sb
);
1178 rdev
->raid_disk
= -1;
1179 clear_bit(Faulty
, &rdev
->flags
);
1180 clear_bit(In_sync
, &rdev
->flags
);
1181 clear_bit(WriteMostly
, &rdev
->flags
);
1183 if (mddev
->raid_disks
== 0) {
1184 mddev
->major_version
= 0;
1185 mddev
->minor_version
= sb
->minor_version
;
1186 mddev
->patch_version
= sb
->patch_version
;
1187 mddev
->external
= 0;
1188 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1189 mddev
->ctime
= sb
->ctime
;
1190 mddev
->utime
= sb
->utime
;
1191 mddev
->level
= sb
->level
;
1192 mddev
->clevel
[0] = 0;
1193 mddev
->layout
= sb
->layout
;
1194 mddev
->raid_disks
= sb
->raid_disks
;
1195 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1196 mddev
->events
= ev1
;
1197 mddev
->bitmap_info
.offset
= 0;
1198 mddev
->bitmap_info
.space
= 0;
1199 /* bitmap can use 60 K after the 4K superblocks */
1200 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1201 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1202 mddev
->reshape_backwards
= 0;
1204 if (mddev
->minor_version
>= 91) {
1205 mddev
->reshape_position
= sb
->reshape_position
;
1206 mddev
->delta_disks
= sb
->delta_disks
;
1207 mddev
->new_level
= sb
->new_level
;
1208 mddev
->new_layout
= sb
->new_layout
;
1209 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1210 if (mddev
->delta_disks
< 0)
1211 mddev
->reshape_backwards
= 1;
1213 mddev
->reshape_position
= MaxSector
;
1214 mddev
->delta_disks
= 0;
1215 mddev
->new_level
= mddev
->level
;
1216 mddev
->new_layout
= mddev
->layout
;
1217 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1220 if (sb
->state
& (1<<MD_SB_CLEAN
))
1221 mddev
->recovery_cp
= MaxSector
;
1223 if (sb
->events_hi
== sb
->cp_events_hi
&&
1224 sb
->events_lo
== sb
->cp_events_lo
) {
1225 mddev
->recovery_cp
= sb
->recovery_cp
;
1227 mddev
->recovery_cp
= 0;
1230 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1231 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1232 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1233 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1235 mddev
->max_disks
= MD_SB_DISKS
;
1237 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1238 mddev
->bitmap_info
.file
== NULL
) {
1239 mddev
->bitmap_info
.offset
=
1240 mddev
->bitmap_info
.default_offset
;
1241 mddev
->bitmap_info
.space
=
1242 mddev
->bitmap_info
.space
;
1245 } else if (mddev
->pers
== NULL
) {
1246 /* Insist on good event counter while assembling, except
1247 * for spares (which don't need an event count) */
1249 if (sb
->disks
[rdev
->desc_nr
].state
& (
1250 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1251 if (ev1
< mddev
->events
)
1253 } else if (mddev
->bitmap
) {
1254 /* if adding to array with a bitmap, then we can accept an
1255 * older device ... but not too old.
1257 if (ev1
< mddev
->bitmap
->events_cleared
)
1260 if (ev1
< mddev
->events
)
1261 /* just a hot-add of a new device, leave raid_disk at -1 */
1265 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1266 desc
= sb
->disks
+ rdev
->desc_nr
;
1268 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1269 set_bit(Faulty
, &rdev
->flags
);
1270 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1271 desc->raid_disk < mddev->raid_disks */) {
1272 set_bit(In_sync
, &rdev
->flags
);
1273 rdev
->raid_disk
= desc
->raid_disk
;
1274 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1275 /* active but not in sync implies recovery up to
1276 * reshape position. We don't know exactly where
1277 * that is, so set to zero for now */
1278 if (mddev
->minor_version
>= 91) {
1279 rdev
->recovery_offset
= 0;
1280 rdev
->raid_disk
= desc
->raid_disk
;
1283 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1284 set_bit(WriteMostly
, &rdev
->flags
);
1285 } else /* MULTIPATH are always insync */
1286 set_bit(In_sync
, &rdev
->flags
);
1291 * sync_super for 0.90.0
1293 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1296 struct md_rdev
*rdev2
;
1297 int next_spare
= mddev
->raid_disks
;
1300 /* make rdev->sb match mddev data..
1303 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1304 * 3/ any empty disks < next_spare become removed
1306 * disks[0] gets initialised to REMOVED because
1307 * we cannot be sure from other fields if it has
1308 * been initialised or not.
1311 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1313 rdev
->sb_size
= MD_SB_BYTES
;
1315 sb
= page_address(rdev
->sb_page
);
1317 memset(sb
, 0, sizeof(*sb
));
1319 sb
->md_magic
= MD_SB_MAGIC
;
1320 sb
->major_version
= mddev
->major_version
;
1321 sb
->patch_version
= mddev
->patch_version
;
1322 sb
->gvalid_words
= 0; /* ignored */
1323 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1324 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1325 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1326 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1328 sb
->ctime
= mddev
->ctime
;
1329 sb
->level
= mddev
->level
;
1330 sb
->size
= mddev
->dev_sectors
/ 2;
1331 sb
->raid_disks
= mddev
->raid_disks
;
1332 sb
->md_minor
= mddev
->md_minor
;
1333 sb
->not_persistent
= 0;
1334 sb
->utime
= mddev
->utime
;
1336 sb
->events_hi
= (mddev
->events
>>32);
1337 sb
->events_lo
= (u32
)mddev
->events
;
1339 if (mddev
->reshape_position
== MaxSector
)
1340 sb
->minor_version
= 90;
1342 sb
->minor_version
= 91;
1343 sb
->reshape_position
= mddev
->reshape_position
;
1344 sb
->new_level
= mddev
->new_level
;
1345 sb
->delta_disks
= mddev
->delta_disks
;
1346 sb
->new_layout
= mddev
->new_layout
;
1347 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1349 mddev
->minor_version
= sb
->minor_version
;
1352 sb
->recovery_cp
= mddev
->recovery_cp
;
1353 sb
->cp_events_hi
= (mddev
->events
>>32);
1354 sb
->cp_events_lo
= (u32
)mddev
->events
;
1355 if (mddev
->recovery_cp
== MaxSector
)
1356 sb
->state
= (1<< MD_SB_CLEAN
);
1358 sb
->recovery_cp
= 0;
1360 sb
->layout
= mddev
->layout
;
1361 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1363 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1364 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1366 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1367 rdev_for_each(rdev2
, mddev
) {
1370 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1372 if (rdev2
->raid_disk
>= 0 &&
1373 sb
->minor_version
>= 91)
1374 /* we have nowhere to store the recovery_offset,
1375 * but if it is not below the reshape_position,
1376 * we can piggy-back on that.
1379 if (rdev2
->raid_disk
< 0 ||
1380 test_bit(Faulty
, &rdev2
->flags
))
1383 desc_nr
= rdev2
->raid_disk
;
1385 desc_nr
= next_spare
++;
1386 rdev2
->desc_nr
= desc_nr
;
1387 d
= &sb
->disks
[rdev2
->desc_nr
];
1389 d
->number
= rdev2
->desc_nr
;
1390 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1391 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1393 d
->raid_disk
= rdev2
->raid_disk
;
1395 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1396 if (test_bit(Faulty
, &rdev2
->flags
))
1397 d
->state
= (1<<MD_DISK_FAULTY
);
1398 else if (is_active
) {
1399 d
->state
= (1<<MD_DISK_ACTIVE
);
1400 if (test_bit(In_sync
, &rdev2
->flags
))
1401 d
->state
|= (1<<MD_DISK_SYNC
);
1409 if (test_bit(WriteMostly
, &rdev2
->flags
))
1410 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1412 /* now set the "removed" and "faulty" bits on any missing devices */
1413 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1414 mdp_disk_t
*d
= &sb
->disks
[i
];
1415 if (d
->state
== 0 && d
->number
== 0) {
1418 d
->state
= (1<<MD_DISK_REMOVED
);
1419 d
->state
|= (1<<MD_DISK_FAULTY
);
1423 sb
->nr_disks
= nr_disks
;
1424 sb
->active_disks
= active
;
1425 sb
->working_disks
= working
;
1426 sb
->failed_disks
= failed
;
1427 sb
->spare_disks
= spare
;
1429 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1430 sb
->sb_csum
= calc_sb_csum(sb
);
1434 * rdev_size_change for 0.90.0
1436 static unsigned long long
1437 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1439 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1440 return 0; /* component must fit device */
1441 if (rdev
->mddev
->bitmap_info
.offset
)
1442 return 0; /* can't move bitmap */
1443 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1444 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1445 num_sectors
= rdev
->sb_start
;
1446 /* Limit to 4TB as metadata cannot record more than that.
1447 * 4TB == 2^32 KB, or 2*2^32 sectors.
1449 if (num_sectors
>= (2ULL << 32))
1450 num_sectors
= (2ULL << 32) - 2;
1451 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1453 md_super_wait(rdev
->mddev
);
1458 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1460 /* non-zero offset changes not possible with v0.90 */
1461 return new_offset
== 0;
1465 * version 1 superblock
1468 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1472 unsigned long long newcsum
;
1473 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1474 __le32
*isuper
= (__le32
*)sb
;
1477 disk_csum
= sb
->sb_csum
;
1480 for (i
=0; size
>=4; size
-= 4 )
1481 newcsum
+= le32_to_cpu(*isuper
++);
1484 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1486 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1487 sb
->sb_csum
= disk_csum
;
1488 return cpu_to_le32(csum
);
1491 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1493 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1495 struct mdp_superblock_1
*sb
;
1499 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1503 * Calculate the position of the superblock in 512byte sectors.
1504 * It is always aligned to a 4K boundary and
1505 * depeding on minor_version, it can be:
1506 * 0: At least 8K, but less than 12K, from end of device
1507 * 1: At start of device
1508 * 2: 4K from start of device.
1510 switch(minor_version
) {
1512 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1514 sb_start
&= ~(sector_t
)(4*2-1);
1525 rdev
->sb_start
= sb_start
;
1527 /* superblock is rarely larger than 1K, but it can be larger,
1528 * and it is safe to read 4k, so we do that
1530 ret
= read_disk_sb(rdev
, 4096);
1531 if (ret
) return ret
;
1534 sb
= page_address(rdev
->sb_page
);
1536 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1537 sb
->major_version
!= cpu_to_le32(1) ||
1538 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1539 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1540 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1543 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1544 printk("md: invalid superblock checksum on %s\n",
1545 bdevname(rdev
->bdev
,b
));
1548 if (le64_to_cpu(sb
->data_size
) < 10) {
1549 printk("md: data_size too small on %s\n",
1550 bdevname(rdev
->bdev
,b
));
1555 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1556 /* Some padding is non-zero, might be a new feature */
1559 rdev
->preferred_minor
= 0xffff;
1560 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1561 rdev
->new_data_offset
= rdev
->data_offset
;
1562 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1563 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1564 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1565 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1567 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1568 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1569 if (rdev
->sb_size
& bmask
)
1570 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1573 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1576 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1579 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1582 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1584 if (!rdev
->bb_page
) {
1585 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1589 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1590 rdev
->badblocks
.count
== 0) {
1591 /* need to load the bad block list.
1592 * Currently we limit it to one page.
1598 int sectors
= le16_to_cpu(sb
->bblog_size
);
1599 if (sectors
> (PAGE_SIZE
/ 512))
1601 offset
= le32_to_cpu(sb
->bblog_offset
);
1604 bb_sector
= (long long)offset
;
1605 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1606 rdev
->bb_page
, READ
, true))
1608 bbp
= (u64
*)page_address(rdev
->bb_page
);
1609 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1610 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1611 u64 bb
= le64_to_cpu(*bbp
);
1612 int count
= bb
& (0x3ff);
1613 u64 sector
= bb
>> 10;
1614 sector
<<= sb
->bblog_shift
;
1615 count
<<= sb
->bblog_shift
;
1618 if (md_set_badblocks(&rdev
->badblocks
,
1619 sector
, count
, 1) == 0)
1622 } else if (sb
->bblog_offset
== 0)
1623 rdev
->badblocks
.shift
= -1;
1629 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1631 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1632 sb
->level
!= refsb
->level
||
1633 sb
->layout
!= refsb
->layout
||
1634 sb
->chunksize
!= refsb
->chunksize
) {
1635 printk(KERN_WARNING
"md: %s has strangely different"
1636 " superblock to %s\n",
1637 bdevname(rdev
->bdev
,b
),
1638 bdevname(refdev
->bdev
,b2
));
1641 ev1
= le64_to_cpu(sb
->events
);
1642 ev2
= le64_to_cpu(refsb
->events
);
1649 if (minor_version
) {
1650 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1651 sectors
-= rdev
->data_offset
;
1653 sectors
= rdev
->sb_start
;
1654 if (sectors
< le64_to_cpu(sb
->data_size
))
1656 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1660 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1662 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1663 __u64 ev1
= le64_to_cpu(sb
->events
);
1665 rdev
->raid_disk
= -1;
1666 clear_bit(Faulty
, &rdev
->flags
);
1667 clear_bit(In_sync
, &rdev
->flags
);
1668 clear_bit(WriteMostly
, &rdev
->flags
);
1670 if (mddev
->raid_disks
== 0) {
1671 mddev
->major_version
= 1;
1672 mddev
->patch_version
= 0;
1673 mddev
->external
= 0;
1674 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1675 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1676 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1677 mddev
->level
= le32_to_cpu(sb
->level
);
1678 mddev
->clevel
[0] = 0;
1679 mddev
->layout
= le32_to_cpu(sb
->layout
);
1680 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1681 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1682 mddev
->events
= ev1
;
1683 mddev
->bitmap_info
.offset
= 0;
1684 mddev
->bitmap_info
.space
= 0;
1685 /* Default location for bitmap is 1K after superblock
1686 * using 3K - total of 4K
1688 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1689 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1690 mddev
->reshape_backwards
= 0;
1692 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1693 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1695 mddev
->max_disks
= (4096-256)/2;
1697 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1698 mddev
->bitmap_info
.file
== NULL
) {
1699 mddev
->bitmap_info
.offset
=
1700 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1701 /* Metadata doesn't record how much space is available.
1702 * For 1.0, we assume we can use up to the superblock
1703 * if before, else to 4K beyond superblock.
1704 * For others, assume no change is possible.
1706 if (mddev
->minor_version
> 0)
1707 mddev
->bitmap_info
.space
= 0;
1708 else if (mddev
->bitmap_info
.offset
> 0)
1709 mddev
->bitmap_info
.space
=
1710 8 - mddev
->bitmap_info
.offset
;
1712 mddev
->bitmap_info
.space
=
1713 -mddev
->bitmap_info
.offset
;
1716 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1717 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1718 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1719 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1720 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1721 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1722 if (mddev
->delta_disks
< 0 ||
1723 (mddev
->delta_disks
== 0 &&
1724 (le32_to_cpu(sb
->feature_map
)
1725 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1726 mddev
->reshape_backwards
= 1;
1728 mddev
->reshape_position
= MaxSector
;
1729 mddev
->delta_disks
= 0;
1730 mddev
->new_level
= mddev
->level
;
1731 mddev
->new_layout
= mddev
->layout
;
1732 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1735 } else if (mddev
->pers
== NULL
) {
1736 /* Insist of good event counter while assembling, except for
1737 * spares (which don't need an event count) */
1739 if (rdev
->desc_nr
>= 0 &&
1740 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1741 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1742 if (ev1
< mddev
->events
)
1744 } else if (mddev
->bitmap
) {
1745 /* If adding to array with a bitmap, then we can accept an
1746 * older device, but not too old.
1748 if (ev1
< mddev
->bitmap
->events_cleared
)
1751 if (ev1
< mddev
->events
)
1752 /* just a hot-add of a new device, leave raid_disk at -1 */
1755 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1757 if (rdev
->desc_nr
< 0 ||
1758 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1762 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1764 case 0xffff: /* spare */
1766 case 0xfffe: /* faulty */
1767 set_bit(Faulty
, &rdev
->flags
);
1770 if ((le32_to_cpu(sb
->feature_map
) &
1771 MD_FEATURE_RECOVERY_OFFSET
))
1772 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1774 set_bit(In_sync
, &rdev
->flags
);
1775 rdev
->raid_disk
= role
;
1778 if (sb
->devflags
& WriteMostly1
)
1779 set_bit(WriteMostly
, &rdev
->flags
);
1780 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1781 set_bit(Replacement
, &rdev
->flags
);
1782 } else /* MULTIPATH are always insync */
1783 set_bit(In_sync
, &rdev
->flags
);
1788 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1790 struct mdp_superblock_1
*sb
;
1791 struct md_rdev
*rdev2
;
1793 /* make rdev->sb match mddev and rdev data. */
1795 sb
= page_address(rdev
->sb_page
);
1797 sb
->feature_map
= 0;
1799 sb
->recovery_offset
= cpu_to_le64(0);
1800 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1802 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1803 sb
->events
= cpu_to_le64(mddev
->events
);
1805 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1807 sb
->resync_offset
= cpu_to_le64(0);
1809 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1811 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1812 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1813 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1814 sb
->level
= cpu_to_le32(mddev
->level
);
1815 sb
->layout
= cpu_to_le32(mddev
->layout
);
1817 if (test_bit(WriteMostly
, &rdev
->flags
))
1818 sb
->devflags
|= WriteMostly1
;
1820 sb
->devflags
&= ~WriteMostly1
;
1821 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1822 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1824 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1825 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1826 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1829 if (rdev
->raid_disk
>= 0 &&
1830 !test_bit(In_sync
, &rdev
->flags
)) {
1832 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1833 sb
->recovery_offset
=
1834 cpu_to_le64(rdev
->recovery_offset
);
1836 if (test_bit(Replacement
, &rdev
->flags
))
1838 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1840 if (mddev
->reshape_position
!= MaxSector
) {
1841 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1842 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1843 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1844 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1845 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1846 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1847 if (mddev
->delta_disks
== 0 &&
1848 mddev
->reshape_backwards
)
1850 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1851 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1853 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1854 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1855 - rdev
->data_offset
));
1859 if (rdev
->badblocks
.count
== 0)
1860 /* Nothing to do for bad blocks*/ ;
1861 else if (sb
->bblog_offset
== 0)
1862 /* Cannot record bad blocks on this device */
1863 md_error(mddev
, rdev
);
1865 struct badblocks
*bb
= &rdev
->badblocks
;
1866 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1868 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1873 seq
= read_seqbegin(&bb
->lock
);
1875 memset(bbp
, 0xff, PAGE_SIZE
);
1877 for (i
= 0 ; i
< bb
->count
; i
++) {
1878 u64 internal_bb
= *p
++;
1879 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1880 | BB_LEN(internal_bb
));
1881 *bbp
++ = cpu_to_le64(store_bb
);
1884 if (read_seqretry(&bb
->lock
, seq
))
1887 bb
->sector
= (rdev
->sb_start
+
1888 (int)le32_to_cpu(sb
->bblog_offset
));
1889 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1894 rdev_for_each(rdev2
, mddev
)
1895 if (rdev2
->desc_nr
+1 > max_dev
)
1896 max_dev
= rdev2
->desc_nr
+1;
1898 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1900 sb
->max_dev
= cpu_to_le32(max_dev
);
1901 rdev
->sb_size
= max_dev
* 2 + 256;
1902 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1903 if (rdev
->sb_size
& bmask
)
1904 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1906 max_dev
= le32_to_cpu(sb
->max_dev
);
1908 for (i
=0; i
<max_dev
;i
++)
1909 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1911 rdev_for_each(rdev2
, mddev
) {
1913 if (test_bit(Faulty
, &rdev2
->flags
))
1914 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1915 else if (test_bit(In_sync
, &rdev2
->flags
))
1916 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1917 else if (rdev2
->raid_disk
>= 0)
1918 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1920 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1923 sb
->sb_csum
= calc_sb_1_csum(sb
);
1926 static unsigned long long
1927 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1929 struct mdp_superblock_1
*sb
;
1930 sector_t max_sectors
;
1931 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1932 return 0; /* component must fit device */
1933 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1934 return 0; /* too confusing */
1935 if (rdev
->sb_start
< rdev
->data_offset
) {
1936 /* minor versions 1 and 2; superblock before data */
1937 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1938 max_sectors
-= rdev
->data_offset
;
1939 if (!num_sectors
|| num_sectors
> max_sectors
)
1940 num_sectors
= max_sectors
;
1941 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1942 /* minor version 0 with bitmap we can't move */
1945 /* minor version 0; superblock after data */
1947 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1948 sb_start
&= ~(sector_t
)(4*2 - 1);
1949 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1950 if (!num_sectors
|| num_sectors
> max_sectors
)
1951 num_sectors
= max_sectors
;
1952 rdev
->sb_start
= sb_start
;
1954 sb
= page_address(rdev
->sb_page
);
1955 sb
->data_size
= cpu_to_le64(num_sectors
);
1956 sb
->super_offset
= rdev
->sb_start
;
1957 sb
->sb_csum
= calc_sb_1_csum(sb
);
1958 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1960 md_super_wait(rdev
->mddev
);
1966 super_1_allow_new_offset(struct md_rdev
*rdev
,
1967 unsigned long long new_offset
)
1969 /* All necessary checks on new >= old have been done */
1970 struct bitmap
*bitmap
;
1971 if (new_offset
>= rdev
->data_offset
)
1974 /* with 1.0 metadata, there is no metadata to tread on
1975 * so we can always move back */
1976 if (rdev
->mddev
->minor_version
== 0)
1979 /* otherwise we must be sure not to step on
1980 * any metadata, so stay:
1981 * 36K beyond start of superblock
1982 * beyond end of badblocks
1983 * beyond write-intent bitmap
1985 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1987 bitmap
= rdev
->mddev
->bitmap
;
1988 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1989 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1990 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1992 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1998 static struct super_type super_types
[] = {
2001 .owner
= THIS_MODULE
,
2002 .load_super
= super_90_load
,
2003 .validate_super
= super_90_validate
,
2004 .sync_super
= super_90_sync
,
2005 .rdev_size_change
= super_90_rdev_size_change
,
2006 .allow_new_offset
= super_90_allow_new_offset
,
2010 .owner
= THIS_MODULE
,
2011 .load_super
= super_1_load
,
2012 .validate_super
= super_1_validate
,
2013 .sync_super
= super_1_sync
,
2014 .rdev_size_change
= super_1_rdev_size_change
,
2015 .allow_new_offset
= super_1_allow_new_offset
,
2019 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
2021 if (mddev
->sync_super
) {
2022 mddev
->sync_super(mddev
, rdev
);
2026 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
2028 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
2031 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
2033 struct md_rdev
*rdev
, *rdev2
;
2036 rdev_for_each_rcu(rdev
, mddev1
)
2037 rdev_for_each_rcu(rdev2
, mddev2
)
2038 if (rdev
->bdev
->bd_contains
==
2039 rdev2
->bdev
->bd_contains
) {
2047 static LIST_HEAD(pending_raid_disks
);
2050 * Try to register data integrity profile for an mddev
2052 * This is called when an array is started and after a disk has been kicked
2053 * from the array. It only succeeds if all working and active component devices
2054 * are integrity capable with matching profiles.
2056 int md_integrity_register(struct mddev
*mddev
)
2058 struct md_rdev
*rdev
, *reference
= NULL
;
2060 if (list_empty(&mddev
->disks
))
2061 return 0; /* nothing to do */
2062 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
2063 return 0; /* shouldn't register, or already is */
2064 rdev_for_each(rdev
, mddev
) {
2065 /* skip spares and non-functional disks */
2066 if (test_bit(Faulty
, &rdev
->flags
))
2068 if (rdev
->raid_disk
< 0)
2071 /* Use the first rdev as the reference */
2075 /* does this rdev's profile match the reference profile? */
2076 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
2077 rdev
->bdev
->bd_disk
) < 0)
2080 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2083 * All component devices are integrity capable and have matching
2084 * profiles, register the common profile for the md device.
2086 if (blk_integrity_register(mddev
->gendisk
,
2087 bdev_get_integrity(reference
->bdev
)) != 0) {
2088 printk(KERN_ERR
"md: failed to register integrity for %s\n",
2092 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2093 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2094 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2100 EXPORT_SYMBOL(md_integrity_register
);
2102 /* Disable data integrity if non-capable/non-matching disk is being added */
2103 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2105 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2106 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2108 if (!bi_mddev
) /* nothing to do */
2110 if (rdev
->raid_disk
< 0) /* skip spares */
2112 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2113 rdev
->bdev
->bd_disk
) >= 0)
2115 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2116 blk_integrity_unregister(mddev
->gendisk
);
2118 EXPORT_SYMBOL(md_integrity_add_rdev
);
2120 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2122 char b
[BDEVNAME_SIZE
];
2132 /* prevent duplicates */
2133 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2136 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2137 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2138 rdev
->sectors
< mddev
->dev_sectors
)) {
2140 /* Cannot change size, so fail
2141 * If mddev->level <= 0, then we don't care
2142 * about aligning sizes (e.g. linear)
2144 if (mddev
->level
> 0)
2147 mddev
->dev_sectors
= rdev
->sectors
;
2150 /* Verify rdev->desc_nr is unique.
2151 * If it is -1, assign a free number, else
2152 * check number is not in use
2154 if (rdev
->desc_nr
< 0) {
2156 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2157 while (find_rdev_nr(mddev
, choice
))
2159 rdev
->desc_nr
= choice
;
2161 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2164 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2165 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2166 mdname(mddev
), mddev
->max_disks
);
2169 bdevname(rdev
->bdev
,b
);
2170 while ( (s
=strchr(b
, '/')) != NULL
)
2173 rdev
->mddev
= mddev
;
2174 printk(KERN_INFO
"md: bind<%s>\n", b
);
2176 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2179 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2180 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2181 /* failure here is OK */;
2182 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2184 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2185 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2187 /* May as well allow recovery to be retried once */
2188 mddev
->recovery_disabled
++;
2193 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2198 static void md_delayed_delete(struct work_struct
*ws
)
2200 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2201 kobject_del(&rdev
->kobj
);
2202 kobject_put(&rdev
->kobj
);
2205 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2207 char b
[BDEVNAME_SIZE
];
2212 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2213 list_del_rcu(&rdev
->same_set
);
2214 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2216 sysfs_remove_link(&rdev
->kobj
, "block");
2217 sysfs_put(rdev
->sysfs_state
);
2218 rdev
->sysfs_state
= NULL
;
2219 rdev
->badblocks
.count
= 0;
2220 /* We need to delay this, otherwise we can deadlock when
2221 * writing to 'remove' to "dev/state". We also need
2222 * to delay it due to rcu usage.
2225 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2226 kobject_get(&rdev
->kobj
);
2227 queue_work(md_misc_wq
, &rdev
->del_work
);
2231 * prevent the device from being mounted, repartitioned or
2232 * otherwise reused by a RAID array (or any other kernel
2233 * subsystem), by bd_claiming the device.
2235 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2238 struct block_device
*bdev
;
2239 char b
[BDEVNAME_SIZE
];
2241 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2242 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2244 printk(KERN_ERR
"md: could not open %s.\n",
2245 __bdevname(dev
, b
));
2246 return PTR_ERR(bdev
);
2252 static void unlock_rdev(struct md_rdev
*rdev
)
2254 struct block_device
*bdev
= rdev
->bdev
;
2258 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2261 void md_autodetect_dev(dev_t dev
);
2263 static void export_rdev(struct md_rdev
* rdev
)
2265 char b
[BDEVNAME_SIZE
];
2266 printk(KERN_INFO
"md: export_rdev(%s)\n",
2267 bdevname(rdev
->bdev
,b
));
2270 md_rdev_clear(rdev
);
2272 if (test_bit(AutoDetected
, &rdev
->flags
))
2273 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2276 kobject_put(&rdev
->kobj
);
2279 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2281 unbind_rdev_from_array(rdev
);
2285 static void export_array(struct mddev
*mddev
)
2287 struct md_rdev
*rdev
, *tmp
;
2289 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2294 kick_rdev_from_array(rdev
);
2296 if (!list_empty(&mddev
->disks
))
2298 mddev
->raid_disks
= 0;
2299 mddev
->major_version
= 0;
2302 static void print_desc(mdp_disk_t
*desc
)
2304 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2305 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2308 static void print_sb_90(mdp_super_t
*sb
)
2313 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2314 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2315 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2317 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2318 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2319 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2320 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2321 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2322 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2323 sb
->failed_disks
, sb
->spare_disks
,
2324 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2327 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2330 desc
= sb
->disks
+ i
;
2331 if (desc
->number
|| desc
->major
|| desc
->minor
||
2332 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2333 printk(" D %2d: ", i
);
2337 printk(KERN_INFO
"md: THIS: ");
2338 print_desc(&sb
->this_disk
);
2341 static void print_sb_1(struct mdp_superblock_1
*sb
)
2345 uuid
= sb
->set_uuid
;
2347 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2348 "md: Name: \"%s\" CT:%llu\n",
2349 le32_to_cpu(sb
->major_version
),
2350 le32_to_cpu(sb
->feature_map
),
2353 (unsigned long long)le64_to_cpu(sb
->ctime
)
2354 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2356 uuid
= sb
->device_uuid
;
2358 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2360 "md: Dev:%08x UUID: %pU\n"
2361 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2362 "md: (MaxDev:%u) \n",
2363 le32_to_cpu(sb
->level
),
2364 (unsigned long long)le64_to_cpu(sb
->size
),
2365 le32_to_cpu(sb
->raid_disks
),
2366 le32_to_cpu(sb
->layout
),
2367 le32_to_cpu(sb
->chunksize
),
2368 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2369 (unsigned long long)le64_to_cpu(sb
->data_size
),
2370 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2371 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2372 le32_to_cpu(sb
->dev_number
),
2375 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2376 (unsigned long long)le64_to_cpu(sb
->events
),
2377 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2378 le32_to_cpu(sb
->sb_csum
),
2379 le32_to_cpu(sb
->max_dev
)
2383 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2385 char b
[BDEVNAME_SIZE
];
2386 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2387 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2388 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2390 if (rdev
->sb_loaded
) {
2391 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2392 switch (major_version
) {
2394 print_sb_90(page_address(rdev
->sb_page
));
2397 print_sb_1(page_address(rdev
->sb_page
));
2401 printk(KERN_INFO
"md: no rdev superblock!\n");
2404 static void md_print_devices(void)
2406 struct list_head
*tmp
;
2407 struct md_rdev
*rdev
;
2408 struct mddev
*mddev
;
2409 char b
[BDEVNAME_SIZE
];
2412 printk("md: **********************************\n");
2413 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2414 printk("md: **********************************\n");
2415 for_each_mddev(mddev
, tmp
) {
2418 bitmap_print_sb(mddev
->bitmap
);
2420 printk("%s: ", mdname(mddev
));
2421 rdev_for_each(rdev
, mddev
)
2422 printk("<%s>", bdevname(rdev
->bdev
,b
));
2425 rdev_for_each(rdev
, mddev
)
2426 print_rdev(rdev
, mddev
->major_version
);
2428 printk("md: **********************************\n");
2433 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2435 /* Update each superblock (in-memory image), but
2436 * if we are allowed to, skip spares which already
2437 * have the right event counter, or have one earlier
2438 * (which would mean they aren't being marked as dirty
2439 * with the rest of the array)
2441 struct md_rdev
*rdev
;
2442 rdev_for_each(rdev
, mddev
) {
2443 if (rdev
->sb_events
== mddev
->events
||
2445 rdev
->raid_disk
< 0 &&
2446 rdev
->sb_events
+1 == mddev
->events
)) {
2447 /* Don't update this superblock */
2448 rdev
->sb_loaded
= 2;
2450 sync_super(mddev
, rdev
);
2451 rdev
->sb_loaded
= 1;
2456 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2458 struct md_rdev
*rdev
;
2461 int any_badblocks_changed
= 0;
2464 /* First make sure individual recovery_offsets are correct */
2465 rdev_for_each(rdev
, mddev
) {
2466 if (rdev
->raid_disk
>= 0 &&
2467 mddev
->delta_disks
>= 0 &&
2468 !test_bit(In_sync
, &rdev
->flags
) &&
2469 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2470 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2473 if (!mddev
->persistent
) {
2474 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2475 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2476 if (!mddev
->external
) {
2477 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2478 rdev_for_each(rdev
, mddev
) {
2479 if (rdev
->badblocks
.changed
) {
2480 rdev
->badblocks
.changed
= 0;
2481 md_ack_all_badblocks(&rdev
->badblocks
);
2482 md_error(mddev
, rdev
);
2484 clear_bit(Blocked
, &rdev
->flags
);
2485 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2486 wake_up(&rdev
->blocked_wait
);
2489 wake_up(&mddev
->sb_wait
);
2493 spin_lock_irq(&mddev
->write_lock
);
2495 mddev
->utime
= get_seconds();
2497 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2499 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2500 /* just a clean<-> dirty transition, possibly leave spares alone,
2501 * though if events isn't the right even/odd, we will have to do
2507 if (mddev
->degraded
)
2508 /* If the array is degraded, then skipping spares is both
2509 * dangerous and fairly pointless.
2510 * Dangerous because a device that was removed from the array
2511 * might have a event_count that still looks up-to-date,
2512 * so it can be re-added without a resync.
2513 * Pointless because if there are any spares to skip,
2514 * then a recovery will happen and soon that array won't
2515 * be degraded any more and the spare can go back to sleep then.
2519 sync_req
= mddev
->in_sync
;
2521 /* If this is just a dirty<->clean transition, and the array is clean
2522 * and 'events' is odd, we can roll back to the previous clean state */
2524 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2525 && mddev
->can_decrease_events
2526 && mddev
->events
!= 1) {
2528 mddev
->can_decrease_events
= 0;
2530 /* otherwise we have to go forward and ... */
2532 mddev
->can_decrease_events
= nospares
;
2535 if (!mddev
->events
) {
2537 * oops, this 64-bit counter should never wrap.
2538 * Either we are in around ~1 trillion A.C., assuming
2539 * 1 reboot per second, or we have a bug:
2545 rdev_for_each(rdev
, mddev
) {
2546 if (rdev
->badblocks
.changed
)
2547 any_badblocks_changed
++;
2548 if (test_bit(Faulty
, &rdev
->flags
))
2549 set_bit(FaultRecorded
, &rdev
->flags
);
2552 sync_sbs(mddev
, nospares
);
2553 spin_unlock_irq(&mddev
->write_lock
);
2555 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2556 mdname(mddev
), mddev
->in_sync
);
2558 bitmap_update_sb(mddev
->bitmap
);
2559 rdev_for_each(rdev
, mddev
) {
2560 char b
[BDEVNAME_SIZE
];
2562 if (rdev
->sb_loaded
!= 1)
2563 continue; /* no noise on spare devices */
2565 if (!test_bit(Faulty
, &rdev
->flags
) &&
2566 rdev
->saved_raid_disk
== -1) {
2567 md_super_write(mddev
,rdev
,
2568 rdev
->sb_start
, rdev
->sb_size
,
2570 pr_debug("md: (write) %s's sb offset: %llu\n",
2571 bdevname(rdev
->bdev
, b
),
2572 (unsigned long long)rdev
->sb_start
);
2573 rdev
->sb_events
= mddev
->events
;
2574 if (rdev
->badblocks
.size
) {
2575 md_super_write(mddev
, rdev
,
2576 rdev
->badblocks
.sector
,
2577 rdev
->badblocks
.size
<< 9,
2579 rdev
->badblocks
.size
= 0;
2582 } else if (test_bit(Faulty
, &rdev
->flags
))
2583 pr_debug("md: %s (skipping faulty)\n",
2584 bdevname(rdev
->bdev
, b
));
2586 pr_debug("(skipping incremental s/r ");
2588 if (mddev
->level
== LEVEL_MULTIPATH
)
2589 /* only need to write one superblock... */
2592 md_super_wait(mddev
);
2593 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2595 spin_lock_irq(&mddev
->write_lock
);
2596 if (mddev
->in_sync
!= sync_req
||
2597 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2598 /* have to write it out again */
2599 spin_unlock_irq(&mddev
->write_lock
);
2602 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2603 spin_unlock_irq(&mddev
->write_lock
);
2604 wake_up(&mddev
->sb_wait
);
2605 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2606 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2608 rdev_for_each(rdev
, mddev
) {
2609 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2610 clear_bit(Blocked
, &rdev
->flags
);
2612 if (any_badblocks_changed
)
2613 md_ack_all_badblocks(&rdev
->badblocks
);
2614 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2615 wake_up(&rdev
->blocked_wait
);
2619 /* words written to sysfs files may, or may not, be \n terminated.
2620 * We want to accept with case. For this we use cmd_match.
2622 static int cmd_match(const char *cmd
, const char *str
)
2624 /* See if cmd, written into a sysfs file, matches
2625 * str. They must either be the same, or cmd can
2626 * have a trailing newline
2628 while (*cmd
&& *str
&& *cmd
== *str
) {
2639 struct rdev_sysfs_entry
{
2640 struct attribute attr
;
2641 ssize_t (*show
)(struct md_rdev
*, char *);
2642 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2646 state_show(struct md_rdev
*rdev
, char *page
)
2651 if (test_bit(Faulty
, &rdev
->flags
) ||
2652 rdev
->badblocks
.unacked_exist
) {
2653 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2656 if (test_bit(In_sync
, &rdev
->flags
)) {
2657 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2660 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2661 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2664 if (test_bit(Blocked
, &rdev
->flags
) ||
2665 (rdev
->badblocks
.unacked_exist
2666 && !test_bit(Faulty
, &rdev
->flags
))) {
2667 len
+= sprintf(page
+len
, "%sblocked", sep
);
2670 if (!test_bit(Faulty
, &rdev
->flags
) &&
2671 !test_bit(In_sync
, &rdev
->flags
)) {
2672 len
+= sprintf(page
+len
, "%sspare", sep
);
2675 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2676 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2679 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2680 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2683 if (test_bit(Replacement
, &rdev
->flags
)) {
2684 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2688 return len
+sprintf(page
+len
, "\n");
2692 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2695 * faulty - simulates an error
2696 * remove - disconnects the device
2697 * writemostly - sets write_mostly
2698 * -writemostly - clears write_mostly
2699 * blocked - sets the Blocked flags
2700 * -blocked - clears the Blocked and possibly simulates an error
2701 * insync - sets Insync providing device isn't active
2702 * write_error - sets WriteErrorSeen
2703 * -write_error - clears WriteErrorSeen
2706 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2707 md_error(rdev
->mddev
, rdev
);
2708 if (test_bit(Faulty
, &rdev
->flags
))
2712 } else if (cmd_match(buf
, "remove")) {
2713 if (rdev
->raid_disk
>= 0)
2716 struct mddev
*mddev
= rdev
->mddev
;
2717 kick_rdev_from_array(rdev
);
2719 md_update_sb(mddev
, 1);
2720 md_new_event(mddev
);
2723 } else if (cmd_match(buf
, "writemostly")) {
2724 set_bit(WriteMostly
, &rdev
->flags
);
2726 } else if (cmd_match(buf
, "-writemostly")) {
2727 clear_bit(WriteMostly
, &rdev
->flags
);
2729 } else if (cmd_match(buf
, "blocked")) {
2730 set_bit(Blocked
, &rdev
->flags
);
2732 } else if (cmd_match(buf
, "-blocked")) {
2733 if (!test_bit(Faulty
, &rdev
->flags
) &&
2734 rdev
->badblocks
.unacked_exist
) {
2735 /* metadata handler doesn't understand badblocks,
2736 * so we need to fail the device
2738 md_error(rdev
->mddev
, rdev
);
2740 clear_bit(Blocked
, &rdev
->flags
);
2741 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2742 wake_up(&rdev
->blocked_wait
);
2743 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2744 md_wakeup_thread(rdev
->mddev
->thread
);
2747 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2748 set_bit(In_sync
, &rdev
->flags
);
2750 } else if (cmd_match(buf
, "write_error")) {
2751 set_bit(WriteErrorSeen
, &rdev
->flags
);
2753 } else if (cmd_match(buf
, "-write_error")) {
2754 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2756 } else if (cmd_match(buf
, "want_replacement")) {
2757 /* Any non-spare device that is not a replacement can
2758 * become want_replacement at any time, but we then need to
2759 * check if recovery is needed.
2761 if (rdev
->raid_disk
>= 0 &&
2762 !test_bit(Replacement
, &rdev
->flags
))
2763 set_bit(WantReplacement
, &rdev
->flags
);
2764 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2765 md_wakeup_thread(rdev
->mddev
->thread
);
2767 } else if (cmd_match(buf
, "-want_replacement")) {
2768 /* Clearing 'want_replacement' is always allowed.
2769 * Once replacements starts it is too late though.
2772 clear_bit(WantReplacement
, &rdev
->flags
);
2773 } else if (cmd_match(buf
, "replacement")) {
2774 /* Can only set a device as a replacement when array has not
2775 * yet been started. Once running, replacement is automatic
2776 * from spares, or by assigning 'slot'.
2778 if (rdev
->mddev
->pers
)
2781 set_bit(Replacement
, &rdev
->flags
);
2784 } else if (cmd_match(buf
, "-replacement")) {
2785 /* Similarly, can only clear Replacement before start */
2786 if (rdev
->mddev
->pers
)
2789 clear_bit(Replacement
, &rdev
->flags
);
2794 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2795 return err
? err
: len
;
2797 static struct rdev_sysfs_entry rdev_state
=
2798 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2801 errors_show(struct md_rdev
*rdev
, char *page
)
2803 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2807 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2810 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2811 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2812 atomic_set(&rdev
->corrected_errors
, n
);
2817 static struct rdev_sysfs_entry rdev_errors
=
2818 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2821 slot_show(struct md_rdev
*rdev
, char *page
)
2823 if (rdev
->raid_disk
< 0)
2824 return sprintf(page
, "none\n");
2826 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2830 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2834 int slot
= simple_strtoul(buf
, &e
, 10);
2835 if (strncmp(buf
, "none", 4)==0)
2837 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2839 if (rdev
->mddev
->pers
&& slot
== -1) {
2840 /* Setting 'slot' on an active array requires also
2841 * updating the 'rd%d' link, and communicating
2842 * with the personality with ->hot_*_disk.
2843 * For now we only support removing
2844 * failed/spare devices. This normally happens automatically,
2845 * but not when the metadata is externally managed.
2847 if (rdev
->raid_disk
== -1)
2849 /* personality does all needed checks */
2850 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2852 err
= rdev
->mddev
->pers
->
2853 hot_remove_disk(rdev
->mddev
, rdev
);
2856 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2857 rdev
->raid_disk
= -1;
2858 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2859 md_wakeup_thread(rdev
->mddev
->thread
);
2860 } else if (rdev
->mddev
->pers
) {
2861 /* Activating a spare .. or possibly reactivating
2862 * if we ever get bitmaps working here.
2865 if (rdev
->raid_disk
!= -1)
2868 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2871 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2874 if (slot
>= rdev
->mddev
->raid_disks
&&
2875 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2878 rdev
->raid_disk
= slot
;
2879 if (test_bit(In_sync
, &rdev
->flags
))
2880 rdev
->saved_raid_disk
= slot
;
2882 rdev
->saved_raid_disk
= -1;
2883 clear_bit(In_sync
, &rdev
->flags
);
2884 err
= rdev
->mddev
->pers
->
2885 hot_add_disk(rdev
->mddev
, rdev
);
2887 rdev
->raid_disk
= -1;
2890 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2891 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2892 /* failure here is OK */;
2893 /* don't wakeup anyone, leave that to userspace. */
2895 if (slot
>= rdev
->mddev
->raid_disks
&&
2896 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2898 rdev
->raid_disk
= slot
;
2899 /* assume it is working */
2900 clear_bit(Faulty
, &rdev
->flags
);
2901 clear_bit(WriteMostly
, &rdev
->flags
);
2902 set_bit(In_sync
, &rdev
->flags
);
2903 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2909 static struct rdev_sysfs_entry rdev_slot
=
2910 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2913 offset_show(struct md_rdev
*rdev
, char *page
)
2915 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2919 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2921 unsigned long long offset
;
2922 if (strict_strtoull(buf
, 10, &offset
) < 0)
2924 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2926 if (rdev
->sectors
&& rdev
->mddev
->external
)
2927 /* Must set offset before size, so overlap checks
2930 rdev
->data_offset
= offset
;
2931 rdev
->new_data_offset
= offset
;
2935 static struct rdev_sysfs_entry rdev_offset
=
2936 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2938 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2940 return sprintf(page
, "%llu\n",
2941 (unsigned long long)rdev
->new_data_offset
);
2944 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2945 const char *buf
, size_t len
)
2947 unsigned long long new_offset
;
2948 struct mddev
*mddev
= rdev
->mddev
;
2950 if (strict_strtoull(buf
, 10, &new_offset
) < 0)
2953 if (mddev
->sync_thread
)
2955 if (new_offset
== rdev
->data_offset
)
2956 /* reset is always permitted */
2958 else if (new_offset
> rdev
->data_offset
) {
2959 /* must not push array size beyond rdev_sectors */
2960 if (new_offset
- rdev
->data_offset
2961 + mddev
->dev_sectors
> rdev
->sectors
)
2964 /* Metadata worries about other space details. */
2966 /* decreasing the offset is inconsistent with a backwards
2969 if (new_offset
< rdev
->data_offset
&&
2970 mddev
->reshape_backwards
)
2972 /* Increasing offset is inconsistent with forwards
2973 * reshape. reshape_direction should be set to
2974 * 'backwards' first.
2976 if (new_offset
> rdev
->data_offset
&&
2977 !mddev
->reshape_backwards
)
2980 if (mddev
->pers
&& mddev
->persistent
&&
2981 !super_types
[mddev
->major_version
]
2982 .allow_new_offset(rdev
, new_offset
))
2984 rdev
->new_data_offset
= new_offset
;
2985 if (new_offset
> rdev
->data_offset
)
2986 mddev
->reshape_backwards
= 1;
2987 else if (new_offset
< rdev
->data_offset
)
2988 mddev
->reshape_backwards
= 0;
2992 static struct rdev_sysfs_entry rdev_new_offset
=
2993 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2996 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2998 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
3001 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
3003 /* check if two start/length pairs overlap */
3011 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
3013 unsigned long long blocks
;
3016 if (strict_strtoull(buf
, 10, &blocks
) < 0)
3019 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
3020 return -EINVAL
; /* sector conversion overflow */
3023 if (new != blocks
* 2)
3024 return -EINVAL
; /* unsigned long long to sector_t overflow */
3031 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3033 struct mddev
*my_mddev
= rdev
->mddev
;
3034 sector_t oldsectors
= rdev
->sectors
;
3037 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
3039 if (rdev
->data_offset
!= rdev
->new_data_offset
)
3040 return -EINVAL
; /* too confusing */
3041 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
3042 if (my_mddev
->persistent
) {
3043 sectors
= super_types
[my_mddev
->major_version
].
3044 rdev_size_change(rdev
, sectors
);
3047 } else if (!sectors
)
3048 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
3051 if (sectors
< my_mddev
->dev_sectors
)
3052 return -EINVAL
; /* component must fit device */
3054 rdev
->sectors
= sectors
;
3055 if (sectors
> oldsectors
&& my_mddev
->external
) {
3056 /* need to check that all other rdevs with the same ->bdev
3057 * do not overlap. We need to unlock the mddev to avoid
3058 * a deadlock. We have already changed rdev->sectors, and if
3059 * we have to change it back, we will have the lock again.
3061 struct mddev
*mddev
;
3063 struct list_head
*tmp
;
3065 mddev_unlock(my_mddev
);
3066 for_each_mddev(mddev
, tmp
) {
3067 struct md_rdev
*rdev2
;
3070 rdev_for_each(rdev2
, mddev
)
3071 if (rdev
->bdev
== rdev2
->bdev
&&
3073 overlaps(rdev
->data_offset
, rdev
->sectors
,
3079 mddev_unlock(mddev
);
3085 mddev_lock(my_mddev
);
3087 /* Someone else could have slipped in a size
3088 * change here, but doing so is just silly.
3089 * We put oldsectors back because we *know* it is
3090 * safe, and trust userspace not to race with
3093 rdev
->sectors
= oldsectors
;
3100 static struct rdev_sysfs_entry rdev_size
=
3101 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3104 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3106 unsigned long long recovery_start
= rdev
->recovery_offset
;
3108 if (test_bit(In_sync
, &rdev
->flags
) ||
3109 recovery_start
== MaxSector
)
3110 return sprintf(page
, "none\n");
3112 return sprintf(page
, "%llu\n", recovery_start
);
3115 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3117 unsigned long long recovery_start
;
3119 if (cmd_match(buf
, "none"))
3120 recovery_start
= MaxSector
;
3121 else if (strict_strtoull(buf
, 10, &recovery_start
))
3124 if (rdev
->mddev
->pers
&&
3125 rdev
->raid_disk
>= 0)
3128 rdev
->recovery_offset
= recovery_start
;
3129 if (recovery_start
== MaxSector
)
3130 set_bit(In_sync
, &rdev
->flags
);
3132 clear_bit(In_sync
, &rdev
->flags
);
3136 static struct rdev_sysfs_entry rdev_recovery_start
=
3137 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3141 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
3143 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
3145 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3147 return badblocks_show(&rdev
->badblocks
, page
, 0);
3149 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3151 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3152 /* Maybe that ack was all we needed */
3153 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3154 wake_up(&rdev
->blocked_wait
);
3157 static struct rdev_sysfs_entry rdev_bad_blocks
=
3158 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3161 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3163 return badblocks_show(&rdev
->badblocks
, page
, 1);
3165 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3167 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3169 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3170 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3172 static struct attribute
*rdev_default_attrs
[] = {
3177 &rdev_new_offset
.attr
,
3179 &rdev_recovery_start
.attr
,
3180 &rdev_bad_blocks
.attr
,
3181 &rdev_unack_bad_blocks
.attr
,
3185 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3187 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3188 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3189 struct mddev
*mddev
= rdev
->mddev
;
3195 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
3197 if (rdev
->mddev
== NULL
)
3200 rv
= entry
->show(rdev
, page
);
3201 mddev_unlock(mddev
);
3207 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3208 const char *page
, size_t length
)
3210 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3211 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3213 struct mddev
*mddev
= rdev
->mddev
;
3217 if (!capable(CAP_SYS_ADMIN
))
3219 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3221 if (rdev
->mddev
== NULL
)
3224 rv
= entry
->store(rdev
, page
, length
);
3225 mddev_unlock(mddev
);
3230 static void rdev_free(struct kobject
*ko
)
3232 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3235 static const struct sysfs_ops rdev_sysfs_ops
= {
3236 .show
= rdev_attr_show
,
3237 .store
= rdev_attr_store
,
3239 static struct kobj_type rdev_ktype
= {
3240 .release
= rdev_free
,
3241 .sysfs_ops
= &rdev_sysfs_ops
,
3242 .default_attrs
= rdev_default_attrs
,
3245 int md_rdev_init(struct md_rdev
*rdev
)
3248 rdev
->saved_raid_disk
= -1;
3249 rdev
->raid_disk
= -1;
3251 rdev
->data_offset
= 0;
3252 rdev
->new_data_offset
= 0;
3253 rdev
->sb_events
= 0;
3254 rdev
->last_read_error
.tv_sec
= 0;
3255 rdev
->last_read_error
.tv_nsec
= 0;
3256 rdev
->sb_loaded
= 0;
3257 rdev
->bb_page
= NULL
;
3258 atomic_set(&rdev
->nr_pending
, 0);
3259 atomic_set(&rdev
->read_errors
, 0);
3260 atomic_set(&rdev
->corrected_errors
, 0);
3262 INIT_LIST_HEAD(&rdev
->same_set
);
3263 init_waitqueue_head(&rdev
->blocked_wait
);
3265 /* Add space to store bad block list.
3266 * This reserves the space even on arrays where it cannot
3267 * be used - I wonder if that matters
3269 rdev
->badblocks
.count
= 0;
3270 rdev
->badblocks
.shift
= 0;
3271 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3272 seqlock_init(&rdev
->badblocks
.lock
);
3273 if (rdev
->badblocks
.page
== NULL
)
3278 EXPORT_SYMBOL_GPL(md_rdev_init
);
3280 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3282 * mark the device faulty if:
3284 * - the device is nonexistent (zero size)
3285 * - the device has no valid superblock
3287 * a faulty rdev _never_ has rdev->sb set.
3289 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3291 char b
[BDEVNAME_SIZE
];
3293 struct md_rdev
*rdev
;
3296 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3298 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3299 return ERR_PTR(-ENOMEM
);
3302 err
= md_rdev_init(rdev
);
3305 err
= alloc_disk_sb(rdev
);
3309 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3313 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3315 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3318 "md: %s has zero or unknown size, marking faulty!\n",
3319 bdevname(rdev
->bdev
,b
));
3324 if (super_format
>= 0) {
3325 err
= super_types
[super_format
].
3326 load_super(rdev
, NULL
, super_minor
);
3327 if (err
== -EINVAL
) {
3329 "md: %s does not have a valid v%d.%d "
3330 "superblock, not importing!\n",
3331 bdevname(rdev
->bdev
,b
),
3332 super_format
, super_minor
);
3337 "md: could not read %s's sb, not importing!\n",
3338 bdevname(rdev
->bdev
,b
));
3342 if (super_format
== -1)
3343 /* hot-add for 0.90, or non-persistent: so no badblocks */
3344 rdev
->badblocks
.shift
= -1;
3351 md_rdev_clear(rdev
);
3353 return ERR_PTR(err
);
3357 * Check a full RAID array for plausibility
3361 static void analyze_sbs(struct mddev
* mddev
)
3364 struct md_rdev
*rdev
, *freshest
, *tmp
;
3365 char b
[BDEVNAME_SIZE
];
3368 rdev_for_each_safe(rdev
, tmp
, mddev
)
3369 switch (super_types
[mddev
->major_version
].
3370 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3378 "md: fatal superblock inconsistency in %s"
3379 " -- removing from array\n",
3380 bdevname(rdev
->bdev
,b
));
3381 kick_rdev_from_array(rdev
);
3385 super_types
[mddev
->major_version
].
3386 validate_super(mddev
, freshest
);
3389 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3390 if (mddev
->max_disks
&&
3391 (rdev
->desc_nr
>= mddev
->max_disks
||
3392 i
> mddev
->max_disks
)) {
3394 "md: %s: %s: only %d devices permitted\n",
3395 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3397 kick_rdev_from_array(rdev
);
3400 if (rdev
!= freshest
)
3401 if (super_types
[mddev
->major_version
].
3402 validate_super(mddev
, rdev
)) {
3403 printk(KERN_WARNING
"md: kicking non-fresh %s"
3405 bdevname(rdev
->bdev
,b
));
3406 kick_rdev_from_array(rdev
);
3409 if (mddev
->level
== LEVEL_MULTIPATH
) {
3410 rdev
->desc_nr
= i
++;
3411 rdev
->raid_disk
= rdev
->desc_nr
;
3412 set_bit(In_sync
, &rdev
->flags
);
3413 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3414 rdev
->raid_disk
= -1;
3415 clear_bit(In_sync
, &rdev
->flags
);
3420 /* Read a fixed-point number.
3421 * Numbers in sysfs attributes should be in "standard" units where
3422 * possible, so time should be in seconds.
3423 * However we internally use a a much smaller unit such as
3424 * milliseconds or jiffies.
3425 * This function takes a decimal number with a possible fractional
3426 * component, and produces an integer which is the result of
3427 * multiplying that number by 10^'scale'.
3428 * all without any floating-point arithmetic.
3430 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3432 unsigned long result
= 0;
3434 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3437 else if (decimals
< scale
) {
3440 result
= result
* 10 + value
;
3452 while (decimals
< scale
) {
3461 static void md_safemode_timeout(unsigned long data
);
3464 safe_delay_show(struct mddev
*mddev
, char *page
)
3466 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3467 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3470 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3474 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3477 mddev
->safemode_delay
= 0;
3479 unsigned long old_delay
= mddev
->safemode_delay
;
3480 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3481 if (mddev
->safemode_delay
== 0)
3482 mddev
->safemode_delay
= 1;
3483 if (mddev
->safemode_delay
< old_delay
)
3484 md_safemode_timeout((unsigned long)mddev
);
3488 static struct md_sysfs_entry md_safe_delay
=
3489 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3492 level_show(struct mddev
*mddev
, char *page
)
3494 struct md_personality
*p
= mddev
->pers
;
3496 return sprintf(page
, "%s\n", p
->name
);
3497 else if (mddev
->clevel
[0])
3498 return sprintf(page
, "%s\n", mddev
->clevel
);
3499 else if (mddev
->level
!= LEVEL_NONE
)
3500 return sprintf(page
, "%d\n", mddev
->level
);
3506 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3510 struct md_personality
*pers
;
3513 struct md_rdev
*rdev
;
3515 if (mddev
->pers
== NULL
) {
3518 if (len
>= sizeof(mddev
->clevel
))
3520 strncpy(mddev
->clevel
, buf
, len
);
3521 if (mddev
->clevel
[len
-1] == '\n')
3523 mddev
->clevel
[len
] = 0;
3524 mddev
->level
= LEVEL_NONE
;
3528 /* request to change the personality. Need to ensure:
3529 * - array is not engaged in resync/recovery/reshape
3530 * - old personality can be suspended
3531 * - new personality will access other array.
3534 if (mddev
->sync_thread
||
3535 mddev
->reshape_position
!= MaxSector
||
3536 mddev
->sysfs_active
)
3539 if (!mddev
->pers
->quiesce
) {
3540 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3541 mdname(mddev
), mddev
->pers
->name
);
3545 /* Now find the new personality */
3546 if (len
== 0 || len
>= sizeof(clevel
))
3548 strncpy(clevel
, buf
, len
);
3549 if (clevel
[len
-1] == '\n')
3552 if (strict_strtol(clevel
, 10, &level
))
3555 if (request_module("md-%s", clevel
) != 0)
3556 request_module("md-level-%s", clevel
);
3557 spin_lock(&pers_lock
);
3558 pers
= find_pers(level
, clevel
);
3559 if (!pers
|| !try_module_get(pers
->owner
)) {
3560 spin_unlock(&pers_lock
);
3561 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3564 spin_unlock(&pers_lock
);
3566 if (pers
== mddev
->pers
) {
3567 /* Nothing to do! */
3568 module_put(pers
->owner
);
3571 if (!pers
->takeover
) {
3572 module_put(pers
->owner
);
3573 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3574 mdname(mddev
), clevel
);
3578 rdev_for_each(rdev
, mddev
)
3579 rdev
->new_raid_disk
= rdev
->raid_disk
;
3581 /* ->takeover must set new_* and/or delta_disks
3582 * if it succeeds, and may set them when it fails.
3584 priv
= pers
->takeover(mddev
);
3586 mddev
->new_level
= mddev
->level
;
3587 mddev
->new_layout
= mddev
->layout
;
3588 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3589 mddev
->raid_disks
-= mddev
->delta_disks
;
3590 mddev
->delta_disks
= 0;
3591 mddev
->reshape_backwards
= 0;
3592 module_put(pers
->owner
);
3593 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3594 mdname(mddev
), clevel
);
3595 return PTR_ERR(priv
);
3598 /* Looks like we have a winner */
3599 mddev_suspend(mddev
);
3600 mddev
->pers
->stop(mddev
);
3602 if (mddev
->pers
->sync_request
== NULL
&&
3603 pers
->sync_request
!= NULL
) {
3604 /* need to add the md_redundancy_group */
3605 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3607 "md: cannot register extra attributes for %s\n",
3609 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3611 if (mddev
->pers
->sync_request
!= NULL
&&
3612 pers
->sync_request
== NULL
) {
3613 /* need to remove the md_redundancy_group */
3614 if (mddev
->to_remove
== NULL
)
3615 mddev
->to_remove
= &md_redundancy_group
;
3618 if (mddev
->pers
->sync_request
== NULL
&&
3620 /* We are converting from a no-redundancy array
3621 * to a redundancy array and metadata is managed
3622 * externally so we need to be sure that writes
3623 * won't block due to a need to transition
3625 * until external management is started.
3628 mddev
->safemode_delay
= 0;
3629 mddev
->safemode
= 0;
3632 rdev_for_each(rdev
, mddev
) {
3633 if (rdev
->raid_disk
< 0)
3635 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3636 rdev
->new_raid_disk
= -1;
3637 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3639 sysfs_unlink_rdev(mddev
, rdev
);
3641 rdev_for_each(rdev
, mddev
) {
3642 if (rdev
->raid_disk
< 0)
3644 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3646 rdev
->raid_disk
= rdev
->new_raid_disk
;
3647 if (rdev
->raid_disk
< 0)
3648 clear_bit(In_sync
, &rdev
->flags
);
3650 if (sysfs_link_rdev(mddev
, rdev
))
3651 printk(KERN_WARNING
"md: cannot register rd%d"
3652 " for %s after level change\n",
3653 rdev
->raid_disk
, mdname(mddev
));
3657 module_put(mddev
->pers
->owner
);
3659 mddev
->private = priv
;
3660 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3661 mddev
->level
= mddev
->new_level
;
3662 mddev
->layout
= mddev
->new_layout
;
3663 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3664 mddev
->delta_disks
= 0;
3665 mddev
->reshape_backwards
= 0;
3666 mddev
->degraded
= 0;
3667 if (mddev
->pers
->sync_request
== NULL
) {
3668 /* this is now an array without redundancy, so
3669 * it must always be in_sync
3672 del_timer_sync(&mddev
->safemode_timer
);
3675 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3676 mddev_resume(mddev
);
3677 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3678 md_new_event(mddev
);
3682 static struct md_sysfs_entry md_level
=
3683 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3687 layout_show(struct mddev
*mddev
, char *page
)
3689 /* just a number, not meaningful for all levels */
3690 if (mddev
->reshape_position
!= MaxSector
&&
3691 mddev
->layout
!= mddev
->new_layout
)
3692 return sprintf(page
, "%d (%d)\n",
3693 mddev
->new_layout
, mddev
->layout
);
3694 return sprintf(page
, "%d\n", mddev
->layout
);
3698 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3701 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3703 if (!*buf
|| (*e
&& *e
!= '\n'))
3708 if (mddev
->pers
->check_reshape
== NULL
)
3710 mddev
->new_layout
= n
;
3711 err
= mddev
->pers
->check_reshape(mddev
);
3713 mddev
->new_layout
= mddev
->layout
;
3717 mddev
->new_layout
= n
;
3718 if (mddev
->reshape_position
== MaxSector
)
3723 static struct md_sysfs_entry md_layout
=
3724 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3728 raid_disks_show(struct mddev
*mddev
, char *page
)
3730 if (mddev
->raid_disks
== 0)
3732 if (mddev
->reshape_position
!= MaxSector
&&
3733 mddev
->delta_disks
!= 0)
3734 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3735 mddev
->raid_disks
- mddev
->delta_disks
);
3736 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3739 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3742 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3746 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3748 if (!*buf
|| (*e
&& *e
!= '\n'))
3752 rv
= update_raid_disks(mddev
, n
);
3753 else if (mddev
->reshape_position
!= MaxSector
) {
3754 struct md_rdev
*rdev
;
3755 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3757 rdev_for_each(rdev
, mddev
) {
3759 rdev
->data_offset
< rdev
->new_data_offset
)
3762 rdev
->data_offset
> rdev
->new_data_offset
)
3765 mddev
->delta_disks
= n
- olddisks
;
3766 mddev
->raid_disks
= n
;
3767 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3769 mddev
->raid_disks
= n
;
3770 return rv
? rv
: len
;
3772 static struct md_sysfs_entry md_raid_disks
=
3773 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3776 chunk_size_show(struct mddev
*mddev
, char *page
)
3778 if (mddev
->reshape_position
!= MaxSector
&&
3779 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3780 return sprintf(page
, "%d (%d)\n",
3781 mddev
->new_chunk_sectors
<< 9,
3782 mddev
->chunk_sectors
<< 9);
3783 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3787 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3790 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3792 if (!*buf
|| (*e
&& *e
!= '\n'))
3797 if (mddev
->pers
->check_reshape
== NULL
)
3799 mddev
->new_chunk_sectors
= n
>> 9;
3800 err
= mddev
->pers
->check_reshape(mddev
);
3802 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3806 mddev
->new_chunk_sectors
= n
>> 9;
3807 if (mddev
->reshape_position
== MaxSector
)
3808 mddev
->chunk_sectors
= n
>> 9;
3812 static struct md_sysfs_entry md_chunk_size
=
3813 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3816 resync_start_show(struct mddev
*mddev
, char *page
)
3818 if (mddev
->recovery_cp
== MaxSector
)
3819 return sprintf(page
, "none\n");
3820 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3824 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3827 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3829 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3831 if (cmd_match(buf
, "none"))
3833 else if (!*buf
|| (*e
&& *e
!= '\n'))
3836 mddev
->recovery_cp
= n
;
3839 static struct md_sysfs_entry md_resync_start
=
3840 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3843 * The array state can be:
3846 * No devices, no size, no level
3847 * Equivalent to STOP_ARRAY ioctl
3849 * May have some settings, but array is not active
3850 * all IO results in error
3851 * When written, doesn't tear down array, but just stops it
3852 * suspended (not supported yet)
3853 * All IO requests will block. The array can be reconfigured.
3854 * Writing this, if accepted, will block until array is quiescent
3856 * no resync can happen. no superblocks get written.
3857 * write requests fail
3859 * like readonly, but behaves like 'clean' on a write request.
3861 * clean - no pending writes, but otherwise active.
3862 * When written to inactive array, starts without resync
3863 * If a write request arrives then
3864 * if metadata is known, mark 'dirty' and switch to 'active'.
3865 * if not known, block and switch to write-pending
3866 * If written to an active array that has pending writes, then fails.
3868 * fully active: IO and resync can be happening.
3869 * When written to inactive array, starts with resync
3872 * clean, but writes are blocked waiting for 'active' to be written.
3875 * like active, but no writes have been seen for a while (100msec).
3878 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3879 write_pending
, active_idle
, bad_word
};
3880 static char *array_states
[] = {
3881 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3882 "write-pending", "active-idle", NULL
};
3884 static int match_word(const char *word
, char **list
)
3887 for (n
=0; list
[n
]; n
++)
3888 if (cmd_match(word
, list
[n
]))
3894 array_state_show(struct mddev
*mddev
, char *page
)
3896 enum array_state st
= inactive
;
3909 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3911 else if (mddev
->safemode
)
3917 if (list_empty(&mddev
->disks
) &&
3918 mddev
->raid_disks
== 0 &&
3919 mddev
->dev_sectors
== 0)
3924 return sprintf(page
, "%s\n", array_states
[st
]);
3927 static int do_md_stop(struct mddev
* mddev
, int ro
, struct block_device
*bdev
);
3928 static int md_set_readonly(struct mddev
* mddev
, struct block_device
*bdev
);
3929 static int do_md_run(struct mddev
* mddev
);
3930 static int restart_array(struct mddev
*mddev
);
3933 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3936 enum array_state st
= match_word(buf
, array_states
);
3941 /* stopping an active array */
3942 if (atomic_read(&mddev
->openers
) > 0)
3944 err
= do_md_stop(mddev
, 0, NULL
);
3947 /* stopping an active array */
3949 if (atomic_read(&mddev
->openers
) > 0)
3951 err
= do_md_stop(mddev
, 2, NULL
);
3953 err
= 0; /* already inactive */
3956 break; /* not supported yet */
3959 err
= md_set_readonly(mddev
, NULL
);
3962 set_disk_ro(mddev
->gendisk
, 1);
3963 err
= do_md_run(mddev
);
3969 err
= md_set_readonly(mddev
, NULL
);
3970 else if (mddev
->ro
== 1)
3971 err
= restart_array(mddev
);
3974 set_disk_ro(mddev
->gendisk
, 0);
3978 err
= do_md_run(mddev
);
3983 restart_array(mddev
);
3984 spin_lock_irq(&mddev
->write_lock
);
3985 if (atomic_read(&mddev
->writes_pending
) == 0) {
3986 if (mddev
->in_sync
== 0) {
3988 if (mddev
->safemode
== 1)
3989 mddev
->safemode
= 0;
3990 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3995 spin_unlock_irq(&mddev
->write_lock
);
4001 restart_array(mddev
);
4002 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
4003 wake_up(&mddev
->sb_wait
);
4007 set_disk_ro(mddev
->gendisk
, 0);
4008 err
= do_md_run(mddev
);
4013 /* these cannot be set */
4019 if (mddev
->hold_active
== UNTIL_IOCTL
)
4020 mddev
->hold_active
= 0;
4021 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4025 static struct md_sysfs_entry md_array_state
=
4026 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
4029 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
4030 return sprintf(page
, "%d\n",
4031 atomic_read(&mddev
->max_corr_read_errors
));
4035 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4038 unsigned long n
= simple_strtoul(buf
, &e
, 10);
4040 if (*buf
&& (*e
== 0 || *e
== '\n')) {
4041 atomic_set(&mddev
->max_corr_read_errors
, n
);
4047 static struct md_sysfs_entry max_corr_read_errors
=
4048 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4049 max_corrected_read_errors_store
);
4052 null_show(struct mddev
*mddev
, char *page
)
4058 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4060 /* buf must be %d:%d\n? giving major and minor numbers */
4061 /* The new device is added to the array.
4062 * If the array has a persistent superblock, we read the
4063 * superblock to initialise info and check validity.
4064 * Otherwise, only checking done is that in bind_rdev_to_array,
4065 * which mainly checks size.
4068 int major
= simple_strtoul(buf
, &e
, 10);
4071 struct md_rdev
*rdev
;
4074 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4076 minor
= simple_strtoul(e
+1, &e
, 10);
4077 if (*e
&& *e
!= '\n')
4079 dev
= MKDEV(major
, minor
);
4080 if (major
!= MAJOR(dev
) ||
4081 minor
!= MINOR(dev
))
4085 if (mddev
->persistent
) {
4086 rdev
= md_import_device(dev
, mddev
->major_version
,
4087 mddev
->minor_version
);
4088 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4089 struct md_rdev
*rdev0
4090 = list_entry(mddev
->disks
.next
,
4091 struct md_rdev
, same_set
);
4092 err
= super_types
[mddev
->major_version
]
4093 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4097 } else if (mddev
->external
)
4098 rdev
= md_import_device(dev
, -2, -1);
4100 rdev
= md_import_device(dev
, -1, -1);
4103 return PTR_ERR(rdev
);
4104 err
= bind_rdev_to_array(rdev
, mddev
);
4108 return err
? err
: len
;
4111 static struct md_sysfs_entry md_new_device
=
4112 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4115 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4118 unsigned long chunk
, end_chunk
;
4122 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4124 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4125 if (buf
== end
) break;
4126 if (*end
== '-') { /* range */
4128 end_chunk
= simple_strtoul(buf
, &end
, 0);
4129 if (buf
== end
) break;
4131 if (*end
&& !isspace(*end
)) break;
4132 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4133 buf
= skip_spaces(end
);
4135 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4140 static struct md_sysfs_entry md_bitmap
=
4141 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4144 size_show(struct mddev
*mddev
, char *page
)
4146 return sprintf(page
, "%llu\n",
4147 (unsigned long long)mddev
->dev_sectors
/ 2);
4150 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4153 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4155 /* If array is inactive, we can reduce the component size, but
4156 * not increase it (except from 0).
4157 * If array is active, we can try an on-line resize
4160 int err
= strict_blocks_to_sectors(buf
, §ors
);
4165 err
= update_size(mddev
, sectors
);
4166 md_update_sb(mddev
, 1);
4168 if (mddev
->dev_sectors
== 0 ||
4169 mddev
->dev_sectors
> sectors
)
4170 mddev
->dev_sectors
= sectors
;
4174 return err
? err
: len
;
4177 static struct md_sysfs_entry md_size
=
4178 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4183 * 'none' for arrays with no metadata (good luck...)
4184 * 'external' for arrays with externally managed metadata,
4185 * or N.M for internally known formats
4188 metadata_show(struct mddev
*mddev
, char *page
)
4190 if (mddev
->persistent
)
4191 return sprintf(page
, "%d.%d\n",
4192 mddev
->major_version
, mddev
->minor_version
);
4193 else if (mddev
->external
)
4194 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4196 return sprintf(page
, "none\n");
4200 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4204 /* Changing the details of 'external' metadata is
4205 * always permitted. Otherwise there must be
4206 * no devices attached to the array.
4208 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4210 else if (!list_empty(&mddev
->disks
))
4213 if (cmd_match(buf
, "none")) {
4214 mddev
->persistent
= 0;
4215 mddev
->external
= 0;
4216 mddev
->major_version
= 0;
4217 mddev
->minor_version
= 90;
4220 if (strncmp(buf
, "external:", 9) == 0) {
4221 size_t namelen
= len
-9;
4222 if (namelen
>= sizeof(mddev
->metadata_type
))
4223 namelen
= sizeof(mddev
->metadata_type
)-1;
4224 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4225 mddev
->metadata_type
[namelen
] = 0;
4226 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4227 mddev
->metadata_type
[--namelen
] = 0;
4228 mddev
->persistent
= 0;
4229 mddev
->external
= 1;
4230 mddev
->major_version
= 0;
4231 mddev
->minor_version
= 90;
4234 major
= simple_strtoul(buf
, &e
, 10);
4235 if (e
==buf
|| *e
!= '.')
4238 minor
= simple_strtoul(buf
, &e
, 10);
4239 if (e
==buf
|| (*e
&& *e
!= '\n') )
4241 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4243 mddev
->major_version
= major
;
4244 mddev
->minor_version
= minor
;
4245 mddev
->persistent
= 1;
4246 mddev
->external
= 0;
4250 static struct md_sysfs_entry md_metadata
=
4251 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4254 action_show(struct mddev
*mddev
, char *page
)
4256 char *type
= "idle";
4257 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4259 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4260 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4261 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4263 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4264 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4266 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4270 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4273 return sprintf(page
, "%s\n", type
);
4276 static void reap_sync_thread(struct mddev
*mddev
);
4279 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4281 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4284 if (cmd_match(page
, "frozen"))
4285 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4287 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4289 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4290 if (mddev
->sync_thread
) {
4291 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4292 reap_sync_thread(mddev
);
4294 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4295 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4297 else if (cmd_match(page
, "resync"))
4298 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4299 else if (cmd_match(page
, "recover")) {
4300 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4301 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4302 } else if (cmd_match(page
, "reshape")) {
4304 if (mddev
->pers
->start_reshape
== NULL
)
4306 err
= mddev
->pers
->start_reshape(mddev
);
4309 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4311 if (cmd_match(page
, "check"))
4312 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4313 else if (!cmd_match(page
, "repair"))
4315 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4316 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4318 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4319 md_wakeup_thread(mddev
->thread
);
4320 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4325 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4327 return sprintf(page
, "%llu\n",
4328 (unsigned long long) mddev
->resync_mismatches
);
4331 static struct md_sysfs_entry md_scan_mode
=
4332 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4335 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4338 sync_min_show(struct mddev
*mddev
, char *page
)
4340 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4341 mddev
->sync_speed_min
? "local": "system");
4345 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4349 if (strncmp(buf
, "system", 6)==0) {
4350 mddev
->sync_speed_min
= 0;
4353 min
= simple_strtoul(buf
, &e
, 10);
4354 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4356 mddev
->sync_speed_min
= min
;
4360 static struct md_sysfs_entry md_sync_min
=
4361 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4364 sync_max_show(struct mddev
*mddev
, char *page
)
4366 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4367 mddev
->sync_speed_max
? "local": "system");
4371 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4375 if (strncmp(buf
, "system", 6)==0) {
4376 mddev
->sync_speed_max
= 0;
4379 max
= simple_strtoul(buf
, &e
, 10);
4380 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4382 mddev
->sync_speed_max
= max
;
4386 static struct md_sysfs_entry md_sync_max
=
4387 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4390 degraded_show(struct mddev
*mddev
, char *page
)
4392 return sprintf(page
, "%d\n", mddev
->degraded
);
4394 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4397 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4399 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4403 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4407 if (strict_strtol(buf
, 10, &n
))
4410 if (n
!= 0 && n
!= 1)
4413 mddev
->parallel_resync
= n
;
4415 if (mddev
->sync_thread
)
4416 wake_up(&resync_wait
);
4421 /* force parallel resync, even with shared block devices */
4422 static struct md_sysfs_entry md_sync_force_parallel
=
4423 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4424 sync_force_parallel_show
, sync_force_parallel_store
);
4427 sync_speed_show(struct mddev
*mddev
, char *page
)
4429 unsigned long resync
, dt
, db
;
4430 if (mddev
->curr_resync
== 0)
4431 return sprintf(page
, "none\n");
4432 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4433 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4435 db
= resync
- mddev
->resync_mark_cnt
;
4436 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4439 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4442 sync_completed_show(struct mddev
*mddev
, char *page
)
4444 unsigned long long max_sectors
, resync
;
4446 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4447 return sprintf(page
, "none\n");
4449 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4450 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4451 max_sectors
= mddev
->resync_max_sectors
;
4453 max_sectors
= mddev
->dev_sectors
;
4455 resync
= mddev
->curr_resync_completed
;
4456 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4459 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4462 min_sync_show(struct mddev
*mddev
, char *page
)
4464 return sprintf(page
, "%llu\n",
4465 (unsigned long long)mddev
->resync_min
);
4468 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4470 unsigned long long min
;
4471 if (strict_strtoull(buf
, 10, &min
))
4473 if (min
> mddev
->resync_max
)
4475 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4478 /* Must be a multiple of chunk_size */
4479 if (mddev
->chunk_sectors
) {
4480 sector_t temp
= min
;
4481 if (sector_div(temp
, mddev
->chunk_sectors
))
4484 mddev
->resync_min
= min
;
4489 static struct md_sysfs_entry md_min_sync
=
4490 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4493 max_sync_show(struct mddev
*mddev
, char *page
)
4495 if (mddev
->resync_max
== MaxSector
)
4496 return sprintf(page
, "max\n");
4498 return sprintf(page
, "%llu\n",
4499 (unsigned long long)mddev
->resync_max
);
4502 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4504 if (strncmp(buf
, "max", 3) == 0)
4505 mddev
->resync_max
= MaxSector
;
4507 unsigned long long max
;
4508 if (strict_strtoull(buf
, 10, &max
))
4510 if (max
< mddev
->resync_min
)
4512 if (max
< mddev
->resync_max
&&
4514 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4517 /* Must be a multiple of chunk_size */
4518 if (mddev
->chunk_sectors
) {
4519 sector_t temp
= max
;
4520 if (sector_div(temp
, mddev
->chunk_sectors
))
4523 mddev
->resync_max
= max
;
4525 wake_up(&mddev
->recovery_wait
);
4529 static struct md_sysfs_entry md_max_sync
=
4530 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4533 suspend_lo_show(struct mddev
*mddev
, char *page
)
4535 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4539 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4542 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4543 unsigned long long old
= mddev
->suspend_lo
;
4545 if (mddev
->pers
== NULL
||
4546 mddev
->pers
->quiesce
== NULL
)
4548 if (buf
== e
|| (*e
&& *e
!= '\n'))
4551 mddev
->suspend_lo
= new;
4553 /* Shrinking suspended region */
4554 mddev
->pers
->quiesce(mddev
, 2);
4556 /* Expanding suspended region - need to wait */
4557 mddev
->pers
->quiesce(mddev
, 1);
4558 mddev
->pers
->quiesce(mddev
, 0);
4562 static struct md_sysfs_entry md_suspend_lo
=
4563 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4567 suspend_hi_show(struct mddev
*mddev
, char *page
)
4569 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4573 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4576 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4577 unsigned long long old
= mddev
->suspend_hi
;
4579 if (mddev
->pers
== NULL
||
4580 mddev
->pers
->quiesce
== NULL
)
4582 if (buf
== e
|| (*e
&& *e
!= '\n'))
4585 mddev
->suspend_hi
= new;
4587 /* Shrinking suspended region */
4588 mddev
->pers
->quiesce(mddev
, 2);
4590 /* Expanding suspended region - need to wait */
4591 mddev
->pers
->quiesce(mddev
, 1);
4592 mddev
->pers
->quiesce(mddev
, 0);
4596 static struct md_sysfs_entry md_suspend_hi
=
4597 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4600 reshape_position_show(struct mddev
*mddev
, char *page
)
4602 if (mddev
->reshape_position
!= MaxSector
)
4603 return sprintf(page
, "%llu\n",
4604 (unsigned long long)mddev
->reshape_position
);
4605 strcpy(page
, "none\n");
4610 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4612 struct md_rdev
*rdev
;
4614 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4617 if (buf
== e
|| (*e
&& *e
!= '\n'))
4619 mddev
->reshape_position
= new;
4620 mddev
->delta_disks
= 0;
4621 mddev
->reshape_backwards
= 0;
4622 mddev
->new_level
= mddev
->level
;
4623 mddev
->new_layout
= mddev
->layout
;
4624 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4625 rdev_for_each(rdev
, mddev
)
4626 rdev
->new_data_offset
= rdev
->data_offset
;
4630 static struct md_sysfs_entry md_reshape_position
=
4631 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4632 reshape_position_store
);
4635 reshape_direction_show(struct mddev
*mddev
, char *page
)
4637 return sprintf(page
, "%s\n",
4638 mddev
->reshape_backwards
? "backwards" : "forwards");
4642 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4645 if (cmd_match(buf
, "forwards"))
4647 else if (cmd_match(buf
, "backwards"))
4651 if (mddev
->reshape_backwards
== backwards
)
4654 /* check if we are allowed to change */
4655 if (mddev
->delta_disks
)
4658 if (mddev
->persistent
&&
4659 mddev
->major_version
== 0)
4662 mddev
->reshape_backwards
= backwards
;
4666 static struct md_sysfs_entry md_reshape_direction
=
4667 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4668 reshape_direction_store
);
4671 array_size_show(struct mddev
*mddev
, char *page
)
4673 if (mddev
->external_size
)
4674 return sprintf(page
, "%llu\n",
4675 (unsigned long long)mddev
->array_sectors
/2);
4677 return sprintf(page
, "default\n");
4681 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4685 if (strncmp(buf
, "default", 7) == 0) {
4687 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4689 sectors
= mddev
->array_sectors
;
4691 mddev
->external_size
= 0;
4693 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4695 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4698 mddev
->external_size
= 1;
4701 mddev
->array_sectors
= sectors
;
4703 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4704 revalidate_disk(mddev
->gendisk
);
4709 static struct md_sysfs_entry md_array_size
=
4710 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4713 static struct attribute
*md_default_attrs
[] = {
4716 &md_raid_disks
.attr
,
4717 &md_chunk_size
.attr
,
4719 &md_resync_start
.attr
,
4721 &md_new_device
.attr
,
4722 &md_safe_delay
.attr
,
4723 &md_array_state
.attr
,
4724 &md_reshape_position
.attr
,
4725 &md_reshape_direction
.attr
,
4726 &md_array_size
.attr
,
4727 &max_corr_read_errors
.attr
,
4731 static struct attribute
*md_redundancy_attrs
[] = {
4733 &md_mismatches
.attr
,
4736 &md_sync_speed
.attr
,
4737 &md_sync_force_parallel
.attr
,
4738 &md_sync_completed
.attr
,
4741 &md_suspend_lo
.attr
,
4742 &md_suspend_hi
.attr
,
4747 static struct attribute_group md_redundancy_group
= {
4749 .attrs
= md_redundancy_attrs
,
4754 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4756 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4757 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4762 spin_lock(&all_mddevs_lock
);
4763 if (list_empty(&mddev
->all_mddevs
)) {
4764 spin_unlock(&all_mddevs_lock
);
4768 spin_unlock(&all_mddevs_lock
);
4770 rv
= mddev_lock(mddev
);
4772 rv
= entry
->show(mddev
, page
);
4773 mddev_unlock(mddev
);
4780 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4781 const char *page
, size_t length
)
4783 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4784 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4789 if (!capable(CAP_SYS_ADMIN
))
4791 spin_lock(&all_mddevs_lock
);
4792 if (list_empty(&mddev
->all_mddevs
)) {
4793 spin_unlock(&all_mddevs_lock
);
4797 spin_unlock(&all_mddevs_lock
);
4798 rv
= mddev_lock(mddev
);
4800 rv
= entry
->store(mddev
, page
, length
);
4801 mddev_unlock(mddev
);
4807 static void md_free(struct kobject
*ko
)
4809 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4811 if (mddev
->sysfs_state
)
4812 sysfs_put(mddev
->sysfs_state
);
4814 if (mddev
->gendisk
) {
4815 del_gendisk(mddev
->gendisk
);
4816 put_disk(mddev
->gendisk
);
4819 blk_cleanup_queue(mddev
->queue
);
4824 static const struct sysfs_ops md_sysfs_ops
= {
4825 .show
= md_attr_show
,
4826 .store
= md_attr_store
,
4828 static struct kobj_type md_ktype
= {
4830 .sysfs_ops
= &md_sysfs_ops
,
4831 .default_attrs
= md_default_attrs
,
4836 static void mddev_delayed_delete(struct work_struct
*ws
)
4838 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4840 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4841 kobject_del(&mddev
->kobj
);
4842 kobject_put(&mddev
->kobj
);
4845 static int md_alloc(dev_t dev
, char *name
)
4847 static DEFINE_MUTEX(disks_mutex
);
4848 struct mddev
*mddev
= mddev_find(dev
);
4849 struct gendisk
*disk
;
4858 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4859 shift
= partitioned
? MdpMinorShift
: 0;
4860 unit
= MINOR(mddev
->unit
) >> shift
;
4862 /* wait for any previous instance of this device to be
4863 * completely removed (mddev_delayed_delete).
4865 flush_workqueue(md_misc_wq
);
4867 mutex_lock(&disks_mutex
);
4873 /* Need to ensure that 'name' is not a duplicate.
4875 struct mddev
*mddev2
;
4876 spin_lock(&all_mddevs_lock
);
4878 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4879 if (mddev2
->gendisk
&&
4880 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4881 spin_unlock(&all_mddevs_lock
);
4884 spin_unlock(&all_mddevs_lock
);
4888 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4891 mddev
->queue
->queuedata
= mddev
;
4893 blk_queue_make_request(mddev
->queue
, md_make_request
);
4894 blk_set_stacking_limits(&mddev
->queue
->limits
);
4896 disk
= alloc_disk(1 << shift
);
4898 blk_cleanup_queue(mddev
->queue
);
4899 mddev
->queue
= NULL
;
4902 disk
->major
= MAJOR(mddev
->unit
);
4903 disk
->first_minor
= unit
<< shift
;
4905 strcpy(disk
->disk_name
, name
);
4906 else if (partitioned
)
4907 sprintf(disk
->disk_name
, "md_d%d", unit
);
4909 sprintf(disk
->disk_name
, "md%d", unit
);
4910 disk
->fops
= &md_fops
;
4911 disk
->private_data
= mddev
;
4912 disk
->queue
= mddev
->queue
;
4913 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4914 /* Allow extended partitions. This makes the
4915 * 'mdp' device redundant, but we can't really
4918 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4919 mddev
->gendisk
= disk
;
4920 /* As soon as we call add_disk(), another thread could get
4921 * through to md_open, so make sure it doesn't get too far
4923 mutex_lock(&mddev
->open_mutex
);
4926 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4927 &disk_to_dev(disk
)->kobj
, "%s", "md");
4929 /* This isn't possible, but as kobject_init_and_add is marked
4930 * __must_check, we must do something with the result
4932 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4936 if (mddev
->kobj
.sd
&&
4937 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4938 printk(KERN_DEBUG
"pointless warning\n");
4939 mutex_unlock(&mddev
->open_mutex
);
4941 mutex_unlock(&disks_mutex
);
4942 if (!error
&& mddev
->kobj
.sd
) {
4943 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4944 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4950 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4952 md_alloc(dev
, NULL
);
4956 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4958 /* val must be "md_*" where * is not all digits.
4959 * We allocate an array with a large free minor number, and
4960 * set the name to val. val must not already be an active name.
4962 int len
= strlen(val
);
4963 char buf
[DISK_NAME_LEN
];
4965 while (len
&& val
[len
-1] == '\n')
4967 if (len
>= DISK_NAME_LEN
)
4969 strlcpy(buf
, val
, len
+1);
4970 if (strncmp(buf
, "md_", 3) != 0)
4972 return md_alloc(0, buf
);
4975 static void md_safemode_timeout(unsigned long data
)
4977 struct mddev
*mddev
= (struct mddev
*) data
;
4979 if (!atomic_read(&mddev
->writes_pending
)) {
4980 mddev
->safemode
= 1;
4981 if (mddev
->external
)
4982 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4984 md_wakeup_thread(mddev
->thread
);
4987 static int start_dirty_degraded
;
4989 int md_run(struct mddev
*mddev
)
4992 struct md_rdev
*rdev
;
4993 struct md_personality
*pers
;
4995 if (list_empty(&mddev
->disks
))
4996 /* cannot run an array with no devices.. */
5001 /* Cannot run until previous stop completes properly */
5002 if (mddev
->sysfs_active
)
5006 * Analyze all RAID superblock(s)
5008 if (!mddev
->raid_disks
) {
5009 if (!mddev
->persistent
)
5014 if (mddev
->level
!= LEVEL_NONE
)
5015 request_module("md-level-%d", mddev
->level
);
5016 else if (mddev
->clevel
[0])
5017 request_module("md-%s", mddev
->clevel
);
5020 * Drop all container device buffers, from now on
5021 * the only valid external interface is through the md
5024 rdev_for_each(rdev
, mddev
) {
5025 if (test_bit(Faulty
, &rdev
->flags
))
5027 sync_blockdev(rdev
->bdev
);
5028 invalidate_bdev(rdev
->bdev
);
5030 /* perform some consistency tests on the device.
5031 * We don't want the data to overlap the metadata,
5032 * Internal Bitmap issues have been handled elsewhere.
5034 if (rdev
->meta_bdev
) {
5035 /* Nothing to check */;
5036 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5037 if (mddev
->dev_sectors
&&
5038 rdev
->data_offset
+ mddev
->dev_sectors
5040 printk("md: %s: data overlaps metadata\n",
5045 if (rdev
->sb_start
+ rdev
->sb_size
/512
5046 > rdev
->data_offset
) {
5047 printk("md: %s: metadata overlaps data\n",
5052 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5055 if (mddev
->bio_set
== NULL
)
5056 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
5057 sizeof(struct mddev
*));
5059 spin_lock(&pers_lock
);
5060 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5061 if (!pers
|| !try_module_get(pers
->owner
)) {
5062 spin_unlock(&pers_lock
);
5063 if (mddev
->level
!= LEVEL_NONE
)
5064 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5067 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5072 spin_unlock(&pers_lock
);
5073 if (mddev
->level
!= pers
->level
) {
5074 mddev
->level
= pers
->level
;
5075 mddev
->new_level
= pers
->level
;
5077 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5079 if (mddev
->reshape_position
!= MaxSector
&&
5080 pers
->start_reshape
== NULL
) {
5081 /* This personality cannot handle reshaping... */
5083 module_put(pers
->owner
);
5087 if (pers
->sync_request
) {
5088 /* Warn if this is a potentially silly
5091 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5092 struct md_rdev
*rdev2
;
5095 rdev_for_each(rdev
, mddev
)
5096 rdev_for_each(rdev2
, mddev
) {
5098 rdev
->bdev
->bd_contains
==
5099 rdev2
->bdev
->bd_contains
) {
5101 "%s: WARNING: %s appears to be"
5102 " on the same physical disk as"
5105 bdevname(rdev
->bdev
,b
),
5106 bdevname(rdev2
->bdev
,b2
));
5113 "True protection against single-disk"
5114 " failure might be compromised.\n");
5117 mddev
->recovery
= 0;
5118 /* may be over-ridden by personality */
5119 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5121 mddev
->ok_start_degraded
= start_dirty_degraded
;
5123 if (start_readonly
&& mddev
->ro
== 0)
5124 mddev
->ro
= 2; /* read-only, but switch on first write */
5126 err
= mddev
->pers
->run(mddev
);
5128 printk(KERN_ERR
"md: pers->run() failed ...\n");
5129 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5130 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5131 " but 'external_size' not in effect?\n", __func__
);
5133 "md: invalid array_size %llu > default size %llu\n",
5134 (unsigned long long)mddev
->array_sectors
/ 2,
5135 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
5137 mddev
->pers
->stop(mddev
);
5139 if (err
== 0 && mddev
->pers
->sync_request
&&
5140 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5141 err
= bitmap_create(mddev
);
5143 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5144 mdname(mddev
), err
);
5145 mddev
->pers
->stop(mddev
);
5149 module_put(mddev
->pers
->owner
);
5151 bitmap_destroy(mddev
);
5154 if (mddev
->pers
->sync_request
) {
5155 if (mddev
->kobj
.sd
&&
5156 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5158 "md: cannot register extra attributes for %s\n",
5160 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5161 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5164 atomic_set(&mddev
->writes_pending
,0);
5165 atomic_set(&mddev
->max_corr_read_errors
,
5166 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5167 mddev
->safemode
= 0;
5168 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5169 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5170 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5174 rdev_for_each(rdev
, mddev
)
5175 if (rdev
->raid_disk
>= 0)
5176 if (sysfs_link_rdev(mddev
, rdev
))
5177 /* failure here is OK */;
5179 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5182 md_update_sb(mddev
, 0);
5184 md_new_event(mddev
);
5185 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5186 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5187 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5190 EXPORT_SYMBOL_GPL(md_run
);
5192 static int do_md_run(struct mddev
*mddev
)
5196 err
= md_run(mddev
);
5199 err
= bitmap_load(mddev
);
5201 bitmap_destroy(mddev
);
5205 md_wakeup_thread(mddev
->thread
);
5206 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5208 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5209 revalidate_disk(mddev
->gendisk
);
5211 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5216 static int restart_array(struct mddev
*mddev
)
5218 struct gendisk
*disk
= mddev
->gendisk
;
5220 /* Complain if it has no devices */
5221 if (list_empty(&mddev
->disks
))
5227 mddev
->safemode
= 0;
5229 set_disk_ro(disk
, 0);
5230 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5232 /* Kick recovery or resync if necessary */
5233 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5234 md_wakeup_thread(mddev
->thread
);
5235 md_wakeup_thread(mddev
->sync_thread
);
5236 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5240 /* similar to deny_write_access, but accounts for our holding a reference
5241 * to the file ourselves */
5242 static int deny_bitmap_write_access(struct file
* file
)
5244 struct inode
*inode
= file
->f_mapping
->host
;
5246 spin_lock(&inode
->i_lock
);
5247 if (atomic_read(&inode
->i_writecount
) > 1) {
5248 spin_unlock(&inode
->i_lock
);
5251 atomic_set(&inode
->i_writecount
, -1);
5252 spin_unlock(&inode
->i_lock
);
5257 void restore_bitmap_write_access(struct file
*file
)
5259 struct inode
*inode
= file
->f_mapping
->host
;
5261 spin_lock(&inode
->i_lock
);
5262 atomic_set(&inode
->i_writecount
, 1);
5263 spin_unlock(&inode
->i_lock
);
5266 static void md_clean(struct mddev
*mddev
)
5268 mddev
->array_sectors
= 0;
5269 mddev
->external_size
= 0;
5270 mddev
->dev_sectors
= 0;
5271 mddev
->raid_disks
= 0;
5272 mddev
->recovery_cp
= 0;
5273 mddev
->resync_min
= 0;
5274 mddev
->resync_max
= MaxSector
;
5275 mddev
->reshape_position
= MaxSector
;
5276 mddev
->external
= 0;
5277 mddev
->persistent
= 0;
5278 mddev
->level
= LEVEL_NONE
;
5279 mddev
->clevel
[0] = 0;
5282 mddev
->metadata_type
[0] = 0;
5283 mddev
->chunk_sectors
= 0;
5284 mddev
->ctime
= mddev
->utime
= 0;
5286 mddev
->max_disks
= 0;
5288 mddev
->can_decrease_events
= 0;
5289 mddev
->delta_disks
= 0;
5290 mddev
->reshape_backwards
= 0;
5291 mddev
->new_level
= LEVEL_NONE
;
5292 mddev
->new_layout
= 0;
5293 mddev
->new_chunk_sectors
= 0;
5294 mddev
->curr_resync
= 0;
5295 mddev
->resync_mismatches
= 0;
5296 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5297 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5298 mddev
->recovery
= 0;
5301 mddev
->degraded
= 0;
5302 mddev
->safemode
= 0;
5303 mddev
->merge_check_needed
= 0;
5304 mddev
->bitmap_info
.offset
= 0;
5305 mddev
->bitmap_info
.default_offset
= 0;
5306 mddev
->bitmap_info
.default_space
= 0;
5307 mddev
->bitmap_info
.chunksize
= 0;
5308 mddev
->bitmap_info
.daemon_sleep
= 0;
5309 mddev
->bitmap_info
.max_write_behind
= 0;
5312 static void __md_stop_writes(struct mddev
*mddev
)
5314 if (mddev
->sync_thread
) {
5315 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5316 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5317 reap_sync_thread(mddev
);
5320 del_timer_sync(&mddev
->safemode_timer
);
5322 bitmap_flush(mddev
);
5323 md_super_wait(mddev
);
5325 if (!mddev
->in_sync
|| mddev
->flags
) {
5326 /* mark array as shutdown cleanly */
5328 md_update_sb(mddev
, 1);
5332 void md_stop_writes(struct mddev
*mddev
)
5335 __md_stop_writes(mddev
);
5336 mddev_unlock(mddev
);
5338 EXPORT_SYMBOL_GPL(md_stop_writes
);
5340 void md_stop(struct mddev
*mddev
)
5343 mddev
->pers
->stop(mddev
);
5344 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5345 mddev
->to_remove
= &md_redundancy_group
;
5346 module_put(mddev
->pers
->owner
);
5348 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5350 EXPORT_SYMBOL_GPL(md_stop
);
5352 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5355 mutex_lock(&mddev
->open_mutex
);
5356 if (atomic_read(&mddev
->openers
) > !!bdev
) {
5357 printk("md: %s still in use.\n",mdname(mddev
));
5362 sync_blockdev(bdev
);
5364 __md_stop_writes(mddev
);
5370 set_disk_ro(mddev
->gendisk
, 1);
5371 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5372 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5376 mutex_unlock(&mddev
->open_mutex
);
5381 * 0 - completely stop and dis-assemble array
5382 * 2 - stop but do not disassemble array
5384 static int do_md_stop(struct mddev
* mddev
, int mode
,
5385 struct block_device
*bdev
)
5387 struct gendisk
*disk
= mddev
->gendisk
;
5388 struct md_rdev
*rdev
;
5390 mutex_lock(&mddev
->open_mutex
);
5391 if (atomic_read(&mddev
->openers
) > !!bdev
||
5392 mddev
->sysfs_active
) {
5393 printk("md: %s still in use.\n",mdname(mddev
));
5394 mutex_unlock(&mddev
->open_mutex
);
5398 /* It is possible IO was issued on some other
5399 * open file which was closed before we took ->open_mutex.
5400 * As that was not the last close __blkdev_put will not
5401 * have called sync_blockdev, so we must.
5403 sync_blockdev(bdev
);
5407 set_disk_ro(disk
, 0);
5409 __md_stop_writes(mddev
);
5411 mddev
->queue
->merge_bvec_fn
= NULL
;
5412 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5414 /* tell userspace to handle 'inactive' */
5415 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5417 rdev_for_each(rdev
, mddev
)
5418 if (rdev
->raid_disk
>= 0)
5419 sysfs_unlink_rdev(mddev
, rdev
);
5421 set_capacity(disk
, 0);
5422 mutex_unlock(&mddev
->open_mutex
);
5424 revalidate_disk(disk
);
5429 mutex_unlock(&mddev
->open_mutex
);
5431 * Free resources if final stop
5434 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5436 bitmap_destroy(mddev
);
5437 if (mddev
->bitmap_info
.file
) {
5438 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5439 fput(mddev
->bitmap_info
.file
);
5440 mddev
->bitmap_info
.file
= NULL
;
5442 mddev
->bitmap_info
.offset
= 0;
5444 export_array(mddev
);
5447 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5448 if (mddev
->hold_active
== UNTIL_STOP
)
5449 mddev
->hold_active
= 0;
5451 blk_integrity_unregister(disk
);
5452 md_new_event(mddev
);
5453 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5458 static void autorun_array(struct mddev
*mddev
)
5460 struct md_rdev
*rdev
;
5463 if (list_empty(&mddev
->disks
))
5466 printk(KERN_INFO
"md: running: ");
5468 rdev_for_each(rdev
, mddev
) {
5469 char b
[BDEVNAME_SIZE
];
5470 printk("<%s>", bdevname(rdev
->bdev
,b
));
5474 err
= do_md_run(mddev
);
5476 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5477 do_md_stop(mddev
, 0, NULL
);
5482 * lets try to run arrays based on all disks that have arrived
5483 * until now. (those are in pending_raid_disks)
5485 * the method: pick the first pending disk, collect all disks with
5486 * the same UUID, remove all from the pending list and put them into
5487 * the 'same_array' list. Then order this list based on superblock
5488 * update time (freshest comes first), kick out 'old' disks and
5489 * compare superblocks. If everything's fine then run it.
5491 * If "unit" is allocated, then bump its reference count
5493 static void autorun_devices(int part
)
5495 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5496 struct mddev
*mddev
;
5497 char b
[BDEVNAME_SIZE
];
5499 printk(KERN_INFO
"md: autorun ...\n");
5500 while (!list_empty(&pending_raid_disks
)) {
5503 LIST_HEAD(candidates
);
5504 rdev0
= list_entry(pending_raid_disks
.next
,
5505 struct md_rdev
, same_set
);
5507 printk(KERN_INFO
"md: considering %s ...\n",
5508 bdevname(rdev0
->bdev
,b
));
5509 INIT_LIST_HEAD(&candidates
);
5510 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5511 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5512 printk(KERN_INFO
"md: adding %s ...\n",
5513 bdevname(rdev
->bdev
,b
));
5514 list_move(&rdev
->same_set
, &candidates
);
5517 * now we have a set of devices, with all of them having
5518 * mostly sane superblocks. It's time to allocate the
5522 dev
= MKDEV(mdp_major
,
5523 rdev0
->preferred_minor
<< MdpMinorShift
);
5524 unit
= MINOR(dev
) >> MdpMinorShift
;
5526 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5529 if (rdev0
->preferred_minor
!= unit
) {
5530 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5531 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5535 md_probe(dev
, NULL
, NULL
);
5536 mddev
= mddev_find(dev
);
5537 if (!mddev
|| !mddev
->gendisk
) {
5541 "md: cannot allocate memory for md drive.\n");
5544 if (mddev_lock(mddev
))
5545 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5547 else if (mddev
->raid_disks
|| mddev
->major_version
5548 || !list_empty(&mddev
->disks
)) {
5550 "md: %s already running, cannot run %s\n",
5551 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5552 mddev_unlock(mddev
);
5554 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5555 mddev
->persistent
= 1;
5556 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5557 list_del_init(&rdev
->same_set
);
5558 if (bind_rdev_to_array(rdev
, mddev
))
5561 autorun_array(mddev
);
5562 mddev_unlock(mddev
);
5564 /* on success, candidates will be empty, on error
5567 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5568 list_del_init(&rdev
->same_set
);
5573 printk(KERN_INFO
"md: ... autorun DONE.\n");
5575 #endif /* !MODULE */
5577 static int get_version(void __user
* arg
)
5581 ver
.major
= MD_MAJOR_VERSION
;
5582 ver
.minor
= MD_MINOR_VERSION
;
5583 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5585 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5591 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5593 mdu_array_info_t info
;
5594 int nr
,working
,insync
,failed
,spare
;
5595 struct md_rdev
*rdev
;
5597 nr
=working
=insync
=failed
=spare
=0;
5598 rdev_for_each(rdev
, mddev
) {
5600 if (test_bit(Faulty
, &rdev
->flags
))
5604 if (test_bit(In_sync
, &rdev
->flags
))
5611 info
.major_version
= mddev
->major_version
;
5612 info
.minor_version
= mddev
->minor_version
;
5613 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5614 info
.ctime
= mddev
->ctime
;
5615 info
.level
= mddev
->level
;
5616 info
.size
= mddev
->dev_sectors
/ 2;
5617 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5620 info
.raid_disks
= mddev
->raid_disks
;
5621 info
.md_minor
= mddev
->md_minor
;
5622 info
.not_persistent
= !mddev
->persistent
;
5624 info
.utime
= mddev
->utime
;
5627 info
.state
= (1<<MD_SB_CLEAN
);
5628 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5629 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5630 info
.active_disks
= insync
;
5631 info
.working_disks
= working
;
5632 info
.failed_disks
= failed
;
5633 info
.spare_disks
= spare
;
5635 info
.layout
= mddev
->layout
;
5636 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5638 if (copy_to_user(arg
, &info
, sizeof(info
)))
5644 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5646 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5647 char *ptr
, *buf
= NULL
;
5650 if (md_allow_write(mddev
))
5651 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5653 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5658 /* bitmap disabled, zero the first byte and copy out */
5659 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5660 file
->pathname
[0] = '\0';
5664 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5668 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5669 buf
, sizeof(file
->pathname
));
5673 strcpy(file
->pathname
, ptr
);
5677 if (copy_to_user(arg
, file
, sizeof(*file
)))
5685 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5687 mdu_disk_info_t info
;
5688 struct md_rdev
*rdev
;
5690 if (copy_from_user(&info
, arg
, sizeof(info
)))
5693 rdev
= find_rdev_nr(mddev
, info
.number
);
5695 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5696 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5697 info
.raid_disk
= rdev
->raid_disk
;
5699 if (test_bit(Faulty
, &rdev
->flags
))
5700 info
.state
|= (1<<MD_DISK_FAULTY
);
5701 else if (test_bit(In_sync
, &rdev
->flags
)) {
5702 info
.state
|= (1<<MD_DISK_ACTIVE
);
5703 info
.state
|= (1<<MD_DISK_SYNC
);
5705 if (test_bit(WriteMostly
, &rdev
->flags
))
5706 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5708 info
.major
= info
.minor
= 0;
5709 info
.raid_disk
= -1;
5710 info
.state
= (1<<MD_DISK_REMOVED
);
5713 if (copy_to_user(arg
, &info
, sizeof(info
)))
5719 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5721 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5722 struct md_rdev
*rdev
;
5723 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5725 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5728 if (!mddev
->raid_disks
) {
5730 /* expecting a device which has a superblock */
5731 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5734 "md: md_import_device returned %ld\n",
5736 return PTR_ERR(rdev
);
5738 if (!list_empty(&mddev
->disks
)) {
5739 struct md_rdev
*rdev0
5740 = list_entry(mddev
->disks
.next
,
5741 struct md_rdev
, same_set
);
5742 err
= super_types
[mddev
->major_version
]
5743 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5746 "md: %s has different UUID to %s\n",
5747 bdevname(rdev
->bdev
,b
),
5748 bdevname(rdev0
->bdev
,b2
));
5753 err
= bind_rdev_to_array(rdev
, mddev
);
5760 * add_new_disk can be used once the array is assembled
5761 * to add "hot spares". They must already have a superblock
5766 if (!mddev
->pers
->hot_add_disk
) {
5768 "%s: personality does not support diskops!\n",
5772 if (mddev
->persistent
)
5773 rdev
= md_import_device(dev
, mddev
->major_version
,
5774 mddev
->minor_version
);
5776 rdev
= md_import_device(dev
, -1, -1);
5779 "md: md_import_device returned %ld\n",
5781 return PTR_ERR(rdev
);
5783 /* set saved_raid_disk if appropriate */
5784 if (!mddev
->persistent
) {
5785 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5786 info
->raid_disk
< mddev
->raid_disks
) {
5787 rdev
->raid_disk
= info
->raid_disk
;
5788 set_bit(In_sync
, &rdev
->flags
);
5790 rdev
->raid_disk
= -1;
5792 super_types
[mddev
->major_version
].
5793 validate_super(mddev
, rdev
);
5794 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5795 rdev
->raid_disk
!= info
->raid_disk
) {
5796 /* This was a hot-add request, but events doesn't
5797 * match, so reject it.
5803 if (test_bit(In_sync
, &rdev
->flags
))
5804 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5806 rdev
->saved_raid_disk
= -1;
5808 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5809 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5810 set_bit(WriteMostly
, &rdev
->flags
);
5812 clear_bit(WriteMostly
, &rdev
->flags
);
5814 rdev
->raid_disk
= -1;
5815 err
= bind_rdev_to_array(rdev
, mddev
);
5816 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5817 /* If there is hot_add_disk but no hot_remove_disk
5818 * then added disks for geometry changes,
5819 * and should be added immediately.
5821 super_types
[mddev
->major_version
].
5822 validate_super(mddev
, rdev
);
5823 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5825 unbind_rdev_from_array(rdev
);
5830 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5832 md_update_sb(mddev
, 1);
5833 if (mddev
->degraded
)
5834 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5835 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5837 md_new_event(mddev
);
5838 md_wakeup_thread(mddev
->thread
);
5842 /* otherwise, add_new_disk is only allowed
5843 * for major_version==0 superblocks
5845 if (mddev
->major_version
!= 0) {
5846 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5851 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5853 rdev
= md_import_device(dev
, -1, 0);
5856 "md: error, md_import_device() returned %ld\n",
5858 return PTR_ERR(rdev
);
5860 rdev
->desc_nr
= info
->number
;
5861 if (info
->raid_disk
< mddev
->raid_disks
)
5862 rdev
->raid_disk
= info
->raid_disk
;
5864 rdev
->raid_disk
= -1;
5866 if (rdev
->raid_disk
< mddev
->raid_disks
)
5867 if (info
->state
& (1<<MD_DISK_SYNC
))
5868 set_bit(In_sync
, &rdev
->flags
);
5870 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5871 set_bit(WriteMostly
, &rdev
->flags
);
5873 if (!mddev
->persistent
) {
5874 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5875 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5877 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5878 rdev
->sectors
= rdev
->sb_start
;
5880 err
= bind_rdev_to_array(rdev
, mddev
);
5890 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5892 char b
[BDEVNAME_SIZE
];
5893 struct md_rdev
*rdev
;
5895 rdev
= find_rdev(mddev
, dev
);
5899 if (rdev
->raid_disk
>= 0)
5902 kick_rdev_from_array(rdev
);
5903 md_update_sb(mddev
, 1);
5904 md_new_event(mddev
);
5908 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5909 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5913 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5915 char b
[BDEVNAME_SIZE
];
5917 struct md_rdev
*rdev
;
5922 if (mddev
->major_version
!= 0) {
5923 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5924 " version-0 superblocks.\n",
5928 if (!mddev
->pers
->hot_add_disk
) {
5930 "%s: personality does not support diskops!\n",
5935 rdev
= md_import_device(dev
, -1, 0);
5938 "md: error, md_import_device() returned %ld\n",
5943 if (mddev
->persistent
)
5944 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5946 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5948 rdev
->sectors
= rdev
->sb_start
;
5950 if (test_bit(Faulty
, &rdev
->flags
)) {
5952 "md: can not hot-add faulty %s disk to %s!\n",
5953 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5957 clear_bit(In_sync
, &rdev
->flags
);
5959 rdev
->saved_raid_disk
= -1;
5960 err
= bind_rdev_to_array(rdev
, mddev
);
5965 * The rest should better be atomic, we can have disk failures
5966 * noticed in interrupt contexts ...
5969 rdev
->raid_disk
= -1;
5971 md_update_sb(mddev
, 1);
5974 * Kick recovery, maybe this spare has to be added to the
5975 * array immediately.
5977 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5978 md_wakeup_thread(mddev
->thread
);
5979 md_new_event(mddev
);
5987 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5992 if (!mddev
->pers
->quiesce
)
5994 if (mddev
->recovery
|| mddev
->sync_thread
)
5996 /* we should be able to change the bitmap.. */
6002 return -EEXIST
; /* cannot add when bitmap is present */
6003 mddev
->bitmap_info
.file
= fget(fd
);
6005 if (mddev
->bitmap_info
.file
== NULL
) {
6006 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
6011 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
6013 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
6015 fput(mddev
->bitmap_info
.file
);
6016 mddev
->bitmap_info
.file
= NULL
;
6019 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6020 } else if (mddev
->bitmap
== NULL
)
6021 return -ENOENT
; /* cannot remove what isn't there */
6024 mddev
->pers
->quiesce(mddev
, 1);
6026 err
= bitmap_create(mddev
);
6028 err
= bitmap_load(mddev
);
6030 if (fd
< 0 || err
) {
6031 bitmap_destroy(mddev
);
6032 fd
= -1; /* make sure to put the file */
6034 mddev
->pers
->quiesce(mddev
, 0);
6037 if (mddev
->bitmap_info
.file
) {
6038 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
6039 fput(mddev
->bitmap_info
.file
);
6041 mddev
->bitmap_info
.file
= NULL
;
6048 * set_array_info is used two different ways
6049 * The original usage is when creating a new array.
6050 * In this usage, raid_disks is > 0 and it together with
6051 * level, size, not_persistent,layout,chunksize determine the
6052 * shape of the array.
6053 * This will always create an array with a type-0.90.0 superblock.
6054 * The newer usage is when assembling an array.
6055 * In this case raid_disks will be 0, and the major_version field is
6056 * use to determine which style super-blocks are to be found on the devices.
6057 * The minor and patch _version numbers are also kept incase the
6058 * super_block handler wishes to interpret them.
6060 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
6063 if (info
->raid_disks
== 0) {
6064 /* just setting version number for superblock loading */
6065 if (info
->major_version
< 0 ||
6066 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6067 super_types
[info
->major_version
].name
== NULL
) {
6068 /* maybe try to auto-load a module? */
6070 "md: superblock version %d not known\n",
6071 info
->major_version
);
6074 mddev
->major_version
= info
->major_version
;
6075 mddev
->minor_version
= info
->minor_version
;
6076 mddev
->patch_version
= info
->patch_version
;
6077 mddev
->persistent
= !info
->not_persistent
;
6078 /* ensure mddev_put doesn't delete this now that there
6079 * is some minimal configuration.
6081 mddev
->ctime
= get_seconds();
6084 mddev
->major_version
= MD_MAJOR_VERSION
;
6085 mddev
->minor_version
= MD_MINOR_VERSION
;
6086 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6087 mddev
->ctime
= get_seconds();
6089 mddev
->level
= info
->level
;
6090 mddev
->clevel
[0] = 0;
6091 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6092 mddev
->raid_disks
= info
->raid_disks
;
6093 /* don't set md_minor, it is determined by which /dev/md* was
6096 if (info
->state
& (1<<MD_SB_CLEAN
))
6097 mddev
->recovery_cp
= MaxSector
;
6099 mddev
->recovery_cp
= 0;
6100 mddev
->persistent
= ! info
->not_persistent
;
6101 mddev
->external
= 0;
6103 mddev
->layout
= info
->layout
;
6104 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6106 mddev
->max_disks
= MD_SB_DISKS
;
6108 if (mddev
->persistent
)
6110 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6112 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6113 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6114 mddev
->bitmap_info
.offset
= 0;
6116 mddev
->reshape_position
= MaxSector
;
6119 * Generate a 128 bit UUID
6121 get_random_bytes(mddev
->uuid
, 16);
6123 mddev
->new_level
= mddev
->level
;
6124 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6125 mddev
->new_layout
= mddev
->layout
;
6126 mddev
->delta_disks
= 0;
6127 mddev
->reshape_backwards
= 0;
6132 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6134 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6136 if (mddev
->external_size
)
6139 mddev
->array_sectors
= array_sectors
;
6141 EXPORT_SYMBOL(md_set_array_sectors
);
6143 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6145 struct md_rdev
*rdev
;
6147 int fit
= (num_sectors
== 0);
6149 if (mddev
->pers
->resize
== NULL
)
6151 /* The "num_sectors" is the number of sectors of each device that
6152 * is used. This can only make sense for arrays with redundancy.
6153 * linear and raid0 always use whatever space is available. We can only
6154 * consider changing this number if no resync or reconstruction is
6155 * happening, and if the new size is acceptable. It must fit before the
6156 * sb_start or, if that is <data_offset, it must fit before the size
6157 * of each device. If num_sectors is zero, we find the largest size
6160 if (mddev
->sync_thread
)
6163 rdev_for_each(rdev
, mddev
) {
6164 sector_t avail
= rdev
->sectors
;
6166 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6167 num_sectors
= avail
;
6168 if (avail
< num_sectors
)
6171 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6173 revalidate_disk(mddev
->gendisk
);
6177 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6180 struct md_rdev
*rdev
;
6181 /* change the number of raid disks */
6182 if (mddev
->pers
->check_reshape
== NULL
)
6184 if (raid_disks
<= 0 ||
6185 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6187 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
6190 rdev_for_each(rdev
, mddev
) {
6191 if (mddev
->raid_disks
< raid_disks
&&
6192 rdev
->data_offset
< rdev
->new_data_offset
)
6194 if (mddev
->raid_disks
> raid_disks
&&
6195 rdev
->data_offset
> rdev
->new_data_offset
)
6199 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6200 if (mddev
->delta_disks
< 0)
6201 mddev
->reshape_backwards
= 1;
6202 else if (mddev
->delta_disks
> 0)
6203 mddev
->reshape_backwards
= 0;
6205 rv
= mddev
->pers
->check_reshape(mddev
);
6207 mddev
->delta_disks
= 0;
6208 mddev
->reshape_backwards
= 0;
6215 * update_array_info is used to change the configuration of an
6217 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6218 * fields in the info are checked against the array.
6219 * Any differences that cannot be handled will cause an error.
6220 * Normally, only one change can be managed at a time.
6222 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6228 /* calculate expected state,ignoring low bits */
6229 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6230 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6232 if (mddev
->major_version
!= info
->major_version
||
6233 mddev
->minor_version
!= info
->minor_version
||
6234 /* mddev->patch_version != info->patch_version || */
6235 mddev
->ctime
!= info
->ctime
||
6236 mddev
->level
!= info
->level
||
6237 /* mddev->layout != info->layout || */
6238 !mddev
->persistent
!= info
->not_persistent
||
6239 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6240 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6241 ((state
^info
->state
) & 0xfffffe00)
6244 /* Check there is only one change */
6245 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6247 if (mddev
->raid_disks
!= info
->raid_disks
)
6249 if (mddev
->layout
!= info
->layout
)
6251 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6258 if (mddev
->layout
!= info
->layout
) {
6260 * we don't need to do anything at the md level, the
6261 * personality will take care of it all.
6263 if (mddev
->pers
->check_reshape
== NULL
)
6266 mddev
->new_layout
= info
->layout
;
6267 rv
= mddev
->pers
->check_reshape(mddev
);
6269 mddev
->new_layout
= mddev
->layout
;
6273 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6274 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6276 if (mddev
->raid_disks
!= info
->raid_disks
)
6277 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6279 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6280 if (mddev
->pers
->quiesce
== NULL
)
6282 if (mddev
->recovery
|| mddev
->sync_thread
)
6284 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6285 /* add the bitmap */
6288 if (mddev
->bitmap_info
.default_offset
== 0)
6290 mddev
->bitmap_info
.offset
=
6291 mddev
->bitmap_info
.default_offset
;
6292 mddev
->bitmap_info
.space
=
6293 mddev
->bitmap_info
.default_space
;
6294 mddev
->pers
->quiesce(mddev
, 1);
6295 rv
= bitmap_create(mddev
);
6297 rv
= bitmap_load(mddev
);
6299 bitmap_destroy(mddev
);
6300 mddev
->pers
->quiesce(mddev
, 0);
6302 /* remove the bitmap */
6305 if (mddev
->bitmap
->storage
.file
)
6307 mddev
->pers
->quiesce(mddev
, 1);
6308 bitmap_destroy(mddev
);
6309 mddev
->pers
->quiesce(mddev
, 0);
6310 mddev
->bitmap_info
.offset
= 0;
6313 md_update_sb(mddev
, 1);
6317 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6319 struct md_rdev
*rdev
;
6321 if (mddev
->pers
== NULL
)
6324 rdev
= find_rdev(mddev
, dev
);
6328 md_error(mddev
, rdev
);
6329 if (!test_bit(Faulty
, &rdev
->flags
))
6335 * We have a problem here : there is no easy way to give a CHS
6336 * virtual geometry. We currently pretend that we have a 2 heads
6337 * 4 sectors (with a BIG number of cylinders...). This drives
6338 * dosfs just mad... ;-)
6340 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6342 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6346 geo
->cylinders
= mddev
->array_sectors
/ 8;
6350 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6351 unsigned int cmd
, unsigned long arg
)
6354 void __user
*argp
= (void __user
*)arg
;
6355 struct mddev
*mddev
= NULL
;
6360 case GET_ARRAY_INFO
:
6364 if (!capable(CAP_SYS_ADMIN
))
6369 * Commands dealing with the RAID driver but not any
6375 err
= get_version(argp
);
6378 case PRINT_RAID_DEBUG
:
6386 autostart_arrays(arg
);
6393 * Commands creating/starting a new array:
6396 mddev
= bdev
->bd_disk
->private_data
;
6403 err
= mddev_lock(mddev
);
6406 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6413 case SET_ARRAY_INFO
:
6415 mdu_array_info_t info
;
6417 memset(&info
, 0, sizeof(info
));
6418 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6423 err
= update_array_info(mddev
, &info
);
6425 printk(KERN_WARNING
"md: couldn't update"
6426 " array info. %d\n", err
);
6431 if (!list_empty(&mddev
->disks
)) {
6433 "md: array %s already has disks!\n",
6438 if (mddev
->raid_disks
) {
6440 "md: array %s already initialised!\n",
6445 err
= set_array_info(mddev
, &info
);
6447 printk(KERN_WARNING
"md: couldn't set"
6448 " array info. %d\n", err
);
6458 * Commands querying/configuring an existing array:
6460 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6461 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6462 if ((!mddev
->raid_disks
&& !mddev
->external
)
6463 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6464 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6465 && cmd
!= GET_BITMAP_FILE
) {
6471 * Commands even a read-only array can execute:
6475 case GET_ARRAY_INFO
:
6476 err
= get_array_info(mddev
, argp
);
6479 case GET_BITMAP_FILE
:
6480 err
= get_bitmap_file(mddev
, argp
);
6484 err
= get_disk_info(mddev
, argp
);
6487 case RESTART_ARRAY_RW
:
6488 err
= restart_array(mddev
);
6492 err
= do_md_stop(mddev
, 0, bdev
);
6496 err
= md_set_readonly(mddev
, bdev
);
6500 if (get_user(ro
, (int __user
*)(arg
))) {
6506 /* if the bdev is going readonly the value of mddev->ro
6507 * does not matter, no writes are coming
6512 /* are we are already prepared for writes? */
6516 /* transitioning to readauto need only happen for
6517 * arrays that call md_write_start
6520 err
= restart_array(mddev
);
6523 set_disk_ro(mddev
->gendisk
, 0);
6530 * The remaining ioctls are changing the state of the
6531 * superblock, so we do not allow them on read-only arrays.
6532 * However non-MD ioctls (e.g. get-size) will still come through
6533 * here and hit the 'default' below, so only disallow
6534 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6536 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6537 if (mddev
->ro
== 2) {
6539 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6540 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6541 md_wakeup_thread(mddev
->thread
);
6552 mdu_disk_info_t info
;
6553 if (copy_from_user(&info
, argp
, sizeof(info
)))
6556 err
= add_new_disk(mddev
, &info
);
6560 case HOT_REMOVE_DISK
:
6561 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6565 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6568 case SET_DISK_FAULTY
:
6569 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6573 err
= do_md_run(mddev
);
6576 case SET_BITMAP_FILE
:
6577 err
= set_bitmap_file(mddev
, (int)arg
);
6587 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6589 mddev
->hold_active
= 0;
6590 mddev_unlock(mddev
);
6599 #ifdef CONFIG_COMPAT
6600 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6601 unsigned int cmd
, unsigned long arg
)
6604 case HOT_REMOVE_DISK
:
6606 case SET_DISK_FAULTY
:
6607 case SET_BITMAP_FILE
:
6608 /* These take in integer arg, do not convert */
6611 arg
= (unsigned long)compat_ptr(arg
);
6615 return md_ioctl(bdev
, mode
, cmd
, arg
);
6617 #endif /* CONFIG_COMPAT */
6619 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6622 * Succeed if we can lock the mddev, which confirms that
6623 * it isn't being stopped right now.
6625 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6631 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6632 /* we are racing with mddev_put which is discarding this
6636 /* Wait until bdev->bd_disk is definitely gone */
6637 flush_workqueue(md_misc_wq
);
6638 /* Then retry the open from the top */
6639 return -ERESTARTSYS
;
6641 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6643 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6647 atomic_inc(&mddev
->openers
);
6648 mutex_unlock(&mddev
->open_mutex
);
6650 check_disk_change(bdev
);
6655 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6657 struct mddev
*mddev
= disk
->private_data
;
6660 atomic_dec(&mddev
->openers
);
6666 static int md_media_changed(struct gendisk
*disk
)
6668 struct mddev
*mddev
= disk
->private_data
;
6670 return mddev
->changed
;
6673 static int md_revalidate(struct gendisk
*disk
)
6675 struct mddev
*mddev
= disk
->private_data
;
6680 static const struct block_device_operations md_fops
=
6682 .owner
= THIS_MODULE
,
6684 .release
= md_release
,
6686 #ifdef CONFIG_COMPAT
6687 .compat_ioctl
= md_compat_ioctl
,
6689 .getgeo
= md_getgeo
,
6690 .media_changed
= md_media_changed
,
6691 .revalidate_disk
= md_revalidate
,
6694 static int md_thread(void * arg
)
6696 struct md_thread
*thread
= arg
;
6699 * md_thread is a 'system-thread', it's priority should be very
6700 * high. We avoid resource deadlocks individually in each
6701 * raid personality. (RAID5 does preallocation) We also use RR and
6702 * the very same RT priority as kswapd, thus we will never get
6703 * into a priority inversion deadlock.
6705 * we definitely have to have equal or higher priority than
6706 * bdflush, otherwise bdflush will deadlock if there are too
6707 * many dirty RAID5 blocks.
6710 allow_signal(SIGKILL
);
6711 while (!kthread_should_stop()) {
6713 /* We need to wait INTERRUPTIBLE so that
6714 * we don't add to the load-average.
6715 * That means we need to be sure no signals are
6718 if (signal_pending(current
))
6719 flush_signals(current
);
6721 wait_event_interruptible_timeout
6723 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6724 || kthread_should_stop(),
6727 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6728 if (!kthread_should_stop())
6729 thread
->run(thread
->mddev
);
6735 void md_wakeup_thread(struct md_thread
*thread
)
6738 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6739 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6740 wake_up(&thread
->wqueue
);
6744 struct md_thread
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6747 struct md_thread
*thread
;
6749 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6753 init_waitqueue_head(&thread
->wqueue
);
6756 thread
->mddev
= mddev
;
6757 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6758 thread
->tsk
= kthread_run(md_thread
, thread
,
6760 mdname(thread
->mddev
),
6762 if (IS_ERR(thread
->tsk
)) {
6769 void md_unregister_thread(struct md_thread
**threadp
)
6771 struct md_thread
*thread
= *threadp
;
6774 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6775 /* Locking ensures that mddev_unlock does not wake_up a
6776 * non-existent thread
6778 spin_lock(&pers_lock
);
6780 spin_unlock(&pers_lock
);
6782 kthread_stop(thread
->tsk
);
6786 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6793 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6796 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6798 mddev
->pers
->error_handler(mddev
,rdev
);
6799 if (mddev
->degraded
)
6800 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6801 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6802 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6803 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6804 md_wakeup_thread(mddev
->thread
);
6805 if (mddev
->event_work
.func
)
6806 queue_work(md_misc_wq
, &mddev
->event_work
);
6807 md_new_event_inintr(mddev
);
6810 /* seq_file implementation /proc/mdstat */
6812 static void status_unused(struct seq_file
*seq
)
6815 struct md_rdev
*rdev
;
6817 seq_printf(seq
, "unused devices: ");
6819 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6820 char b
[BDEVNAME_SIZE
];
6822 seq_printf(seq
, "%s ",
6823 bdevname(rdev
->bdev
,b
));
6826 seq_printf(seq
, "<none>");
6828 seq_printf(seq
, "\n");
6832 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6834 sector_t max_sectors
, resync
, res
;
6835 unsigned long dt
, db
;
6838 unsigned int per_milli
;
6840 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6842 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6843 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6844 max_sectors
= mddev
->resync_max_sectors
;
6846 max_sectors
= mddev
->dev_sectors
;
6849 * Should not happen.
6855 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6856 * in a sector_t, and (max_sectors>>scale) will fit in a
6857 * u32, as those are the requirements for sector_div.
6858 * Thus 'scale' must be at least 10
6861 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6862 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6865 res
= (resync
>>scale
)*1000;
6866 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6870 int i
, x
= per_milli
/50, y
= 20-x
;
6871 seq_printf(seq
, "[");
6872 for (i
= 0; i
< x
; i
++)
6873 seq_printf(seq
, "=");
6874 seq_printf(seq
, ">");
6875 for (i
= 0; i
< y
; i
++)
6876 seq_printf(seq
, ".");
6877 seq_printf(seq
, "] ");
6879 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6880 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6882 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6884 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6885 "resync" : "recovery"))),
6886 per_milli
/10, per_milli
% 10,
6887 (unsigned long long) resync
/2,
6888 (unsigned long long) max_sectors
/2);
6891 * dt: time from mark until now
6892 * db: blocks written from mark until now
6893 * rt: remaining time
6895 * rt is a sector_t, so could be 32bit or 64bit.
6896 * So we divide before multiply in case it is 32bit and close
6898 * We scale the divisor (db) by 32 to avoid losing precision
6899 * near the end of resync when the number of remaining sectors
6901 * We then divide rt by 32 after multiplying by db to compensate.
6902 * The '+1' avoids division by zero if db is very small.
6904 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6906 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6907 - mddev
->resync_mark_cnt
;
6909 rt
= max_sectors
- resync
; /* number of remaining sectors */
6910 sector_div(rt
, db
/32+1);
6914 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6915 ((unsigned long)rt
% 60)/6);
6917 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6920 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6922 struct list_head
*tmp
;
6924 struct mddev
*mddev
;
6932 spin_lock(&all_mddevs_lock
);
6933 list_for_each(tmp
,&all_mddevs
)
6935 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6937 spin_unlock(&all_mddevs_lock
);
6940 spin_unlock(&all_mddevs_lock
);
6942 return (void*)2;/* tail */
6946 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6948 struct list_head
*tmp
;
6949 struct mddev
*next_mddev
, *mddev
= v
;
6955 spin_lock(&all_mddevs_lock
);
6957 tmp
= all_mddevs
.next
;
6959 tmp
= mddev
->all_mddevs
.next
;
6960 if (tmp
!= &all_mddevs
)
6961 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6963 next_mddev
= (void*)2;
6966 spin_unlock(&all_mddevs_lock
);
6974 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6976 struct mddev
*mddev
= v
;
6978 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6982 static int md_seq_show(struct seq_file
*seq
, void *v
)
6984 struct mddev
*mddev
= v
;
6986 struct md_rdev
*rdev
;
6988 if (v
== (void*)1) {
6989 struct md_personality
*pers
;
6990 seq_printf(seq
, "Personalities : ");
6991 spin_lock(&pers_lock
);
6992 list_for_each_entry(pers
, &pers_list
, list
)
6993 seq_printf(seq
, "[%s] ", pers
->name
);
6995 spin_unlock(&pers_lock
);
6996 seq_printf(seq
, "\n");
6997 seq
->poll_event
= atomic_read(&md_event_count
);
7000 if (v
== (void*)2) {
7005 if (mddev_lock(mddev
) < 0)
7008 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7009 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7010 mddev
->pers
? "" : "in");
7013 seq_printf(seq
, " (read-only)");
7015 seq_printf(seq
, " (auto-read-only)");
7016 seq_printf(seq
, " %s", mddev
->pers
->name
);
7020 rdev_for_each(rdev
, mddev
) {
7021 char b
[BDEVNAME_SIZE
];
7022 seq_printf(seq
, " %s[%d]",
7023 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7024 if (test_bit(WriteMostly
, &rdev
->flags
))
7025 seq_printf(seq
, "(W)");
7026 if (test_bit(Faulty
, &rdev
->flags
)) {
7027 seq_printf(seq
, "(F)");
7030 if (rdev
->raid_disk
< 0)
7031 seq_printf(seq
, "(S)"); /* spare */
7032 if (test_bit(Replacement
, &rdev
->flags
))
7033 seq_printf(seq
, "(R)");
7034 sectors
+= rdev
->sectors
;
7037 if (!list_empty(&mddev
->disks
)) {
7039 seq_printf(seq
, "\n %llu blocks",
7040 (unsigned long long)
7041 mddev
->array_sectors
/ 2);
7043 seq_printf(seq
, "\n %llu blocks",
7044 (unsigned long long)sectors
/ 2);
7046 if (mddev
->persistent
) {
7047 if (mddev
->major_version
!= 0 ||
7048 mddev
->minor_version
!= 90) {
7049 seq_printf(seq
," super %d.%d",
7050 mddev
->major_version
,
7051 mddev
->minor_version
);
7053 } else if (mddev
->external
)
7054 seq_printf(seq
, " super external:%s",
7055 mddev
->metadata_type
);
7057 seq_printf(seq
, " super non-persistent");
7060 mddev
->pers
->status(seq
, mddev
);
7061 seq_printf(seq
, "\n ");
7062 if (mddev
->pers
->sync_request
) {
7063 if (mddev
->curr_resync
> 2) {
7064 status_resync(seq
, mddev
);
7065 seq_printf(seq
, "\n ");
7066 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
7067 seq_printf(seq
, "\tresync=DELAYED\n ");
7068 else if (mddev
->recovery_cp
< MaxSector
)
7069 seq_printf(seq
, "\tresync=PENDING\n ");
7072 seq_printf(seq
, "\n ");
7074 bitmap_status(seq
, mddev
->bitmap
);
7076 seq_printf(seq
, "\n");
7078 mddev_unlock(mddev
);
7083 static const struct seq_operations md_seq_ops
= {
7084 .start
= md_seq_start
,
7085 .next
= md_seq_next
,
7086 .stop
= md_seq_stop
,
7087 .show
= md_seq_show
,
7090 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7092 struct seq_file
*seq
;
7095 error
= seq_open(file
, &md_seq_ops
);
7099 seq
= file
->private_data
;
7100 seq
->poll_event
= atomic_read(&md_event_count
);
7104 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7106 struct seq_file
*seq
= filp
->private_data
;
7109 poll_wait(filp
, &md_event_waiters
, wait
);
7111 /* always allow read */
7112 mask
= POLLIN
| POLLRDNORM
;
7114 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7115 mask
|= POLLERR
| POLLPRI
;
7119 static const struct file_operations md_seq_fops
= {
7120 .owner
= THIS_MODULE
,
7121 .open
= md_seq_open
,
7123 .llseek
= seq_lseek
,
7124 .release
= seq_release_private
,
7125 .poll
= mdstat_poll
,
7128 int register_md_personality(struct md_personality
*p
)
7130 spin_lock(&pers_lock
);
7131 list_add_tail(&p
->list
, &pers_list
);
7132 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
7133 spin_unlock(&pers_lock
);
7137 int unregister_md_personality(struct md_personality
*p
)
7139 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7140 spin_lock(&pers_lock
);
7141 list_del_init(&p
->list
);
7142 spin_unlock(&pers_lock
);
7146 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7148 struct md_rdev
* rdev
;
7154 rdev_for_each_rcu(rdev
, mddev
) {
7155 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7156 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7157 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7158 atomic_read(&disk
->sync_io
);
7159 /* sync IO will cause sync_io to increase before the disk_stats
7160 * as sync_io is counted when a request starts, and
7161 * disk_stats is counted when it completes.
7162 * So resync activity will cause curr_events to be smaller than
7163 * when there was no such activity.
7164 * non-sync IO will cause disk_stat to increase without
7165 * increasing sync_io so curr_events will (eventually)
7166 * be larger than it was before. Once it becomes
7167 * substantially larger, the test below will cause
7168 * the array to appear non-idle, and resync will slow
7170 * If there is a lot of outstanding resync activity when
7171 * we set last_event to curr_events, then all that activity
7172 * completing might cause the array to appear non-idle
7173 * and resync will be slowed down even though there might
7174 * not have been non-resync activity. This will only
7175 * happen once though. 'last_events' will soon reflect
7176 * the state where there is little or no outstanding
7177 * resync requests, and further resync activity will
7178 * always make curr_events less than last_events.
7181 if (init
|| curr_events
- rdev
->last_events
> 64) {
7182 rdev
->last_events
= curr_events
;
7190 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7192 /* another "blocks" (512byte) blocks have been synced */
7193 atomic_sub(blocks
, &mddev
->recovery_active
);
7194 wake_up(&mddev
->recovery_wait
);
7196 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7197 md_wakeup_thread(mddev
->thread
);
7198 // stop recovery, signal do_sync ....
7203 /* md_write_start(mddev, bi)
7204 * If we need to update some array metadata (e.g. 'active' flag
7205 * in superblock) before writing, schedule a superblock update
7206 * and wait for it to complete.
7208 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7211 if (bio_data_dir(bi
) != WRITE
)
7214 BUG_ON(mddev
->ro
== 1);
7215 if (mddev
->ro
== 2) {
7216 /* need to switch to read/write */
7218 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7219 md_wakeup_thread(mddev
->thread
);
7220 md_wakeup_thread(mddev
->sync_thread
);
7223 atomic_inc(&mddev
->writes_pending
);
7224 if (mddev
->safemode
== 1)
7225 mddev
->safemode
= 0;
7226 if (mddev
->in_sync
) {
7227 spin_lock_irq(&mddev
->write_lock
);
7228 if (mddev
->in_sync
) {
7230 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7231 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7232 md_wakeup_thread(mddev
->thread
);
7235 spin_unlock_irq(&mddev
->write_lock
);
7238 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7239 wait_event(mddev
->sb_wait
,
7240 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7243 void md_write_end(struct mddev
*mddev
)
7245 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7246 if (mddev
->safemode
== 2)
7247 md_wakeup_thread(mddev
->thread
);
7248 else if (mddev
->safemode_delay
)
7249 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7253 /* md_allow_write(mddev)
7254 * Calling this ensures that the array is marked 'active' so that writes
7255 * may proceed without blocking. It is important to call this before
7256 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7257 * Must be called with mddev_lock held.
7259 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7260 * is dropped, so return -EAGAIN after notifying userspace.
7262 int md_allow_write(struct mddev
*mddev
)
7268 if (!mddev
->pers
->sync_request
)
7271 spin_lock_irq(&mddev
->write_lock
);
7272 if (mddev
->in_sync
) {
7274 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7275 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7276 if (mddev
->safemode_delay
&&
7277 mddev
->safemode
== 0)
7278 mddev
->safemode
= 1;
7279 spin_unlock_irq(&mddev
->write_lock
);
7280 md_update_sb(mddev
, 0);
7281 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7283 spin_unlock_irq(&mddev
->write_lock
);
7285 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7290 EXPORT_SYMBOL_GPL(md_allow_write
);
7292 #define SYNC_MARKS 10
7293 #define SYNC_MARK_STEP (3*HZ)
7294 void md_do_sync(struct mddev
*mddev
)
7296 struct mddev
*mddev2
;
7297 unsigned int currspeed
= 0,
7299 sector_t max_sectors
,j
, io_sectors
;
7300 unsigned long mark
[SYNC_MARKS
];
7301 sector_t mark_cnt
[SYNC_MARKS
];
7303 struct list_head
*tmp
;
7304 sector_t last_check
;
7306 struct md_rdev
*rdev
;
7308 struct blk_plug plug
;
7310 /* just incase thread restarts... */
7311 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7313 if (mddev
->ro
) /* never try to sync a read-only array */
7316 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7317 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7318 desc
= "data-check";
7319 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7320 desc
= "requested-resync";
7323 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7328 /* we overload curr_resync somewhat here.
7329 * 0 == not engaged in resync at all
7330 * 2 == checking that there is no conflict with another sync
7331 * 1 == like 2, but have yielded to allow conflicting resync to
7333 * other == active in resync - this many blocks
7335 * Before starting a resync we must have set curr_resync to
7336 * 2, and then checked that every "conflicting" array has curr_resync
7337 * less than ours. When we find one that is the same or higher
7338 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7339 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7340 * This will mean we have to start checking from the beginning again.
7345 mddev
->curr_resync
= 2;
7348 if (kthread_should_stop())
7349 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7351 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7353 for_each_mddev(mddev2
, tmp
) {
7354 if (mddev2
== mddev
)
7356 if (!mddev
->parallel_resync
7357 && mddev2
->curr_resync
7358 && match_mddev_units(mddev
, mddev2
)) {
7360 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7361 /* arbitrarily yield */
7362 mddev
->curr_resync
= 1;
7363 wake_up(&resync_wait
);
7365 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7366 /* no need to wait here, we can wait the next
7367 * time 'round when curr_resync == 2
7370 /* We need to wait 'interruptible' so as not to
7371 * contribute to the load average, and not to
7372 * be caught by 'softlockup'
7374 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7375 if (!kthread_should_stop() &&
7376 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7377 printk(KERN_INFO
"md: delaying %s of %s"
7378 " until %s has finished (they"
7379 " share one or more physical units)\n",
7380 desc
, mdname(mddev
), mdname(mddev2
));
7382 if (signal_pending(current
))
7383 flush_signals(current
);
7385 finish_wait(&resync_wait
, &wq
);
7388 finish_wait(&resync_wait
, &wq
);
7391 } while (mddev
->curr_resync
< 2);
7394 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7395 /* resync follows the size requested by the personality,
7396 * which defaults to physical size, but can be virtual size
7398 max_sectors
= mddev
->resync_max_sectors
;
7399 mddev
->resync_mismatches
= 0;
7400 /* we don't use the checkpoint if there's a bitmap */
7401 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7402 j
= mddev
->resync_min
;
7403 else if (!mddev
->bitmap
)
7404 j
= mddev
->recovery_cp
;
7406 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7407 max_sectors
= mddev
->resync_max_sectors
;
7409 /* recovery follows the physical size of devices */
7410 max_sectors
= mddev
->dev_sectors
;
7413 rdev_for_each_rcu(rdev
, mddev
)
7414 if (rdev
->raid_disk
>= 0 &&
7415 !test_bit(Faulty
, &rdev
->flags
) &&
7416 !test_bit(In_sync
, &rdev
->flags
) &&
7417 rdev
->recovery_offset
< j
)
7418 j
= rdev
->recovery_offset
;
7422 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7423 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7424 " %d KB/sec/disk.\n", speed_min(mddev
));
7425 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7426 "(but not more than %d KB/sec) for %s.\n",
7427 speed_max(mddev
), desc
);
7429 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7432 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7434 mark_cnt
[m
] = io_sectors
;
7437 mddev
->resync_mark
= mark
[last_mark
];
7438 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7441 * Tune reconstruction:
7443 window
= 32*(PAGE_SIZE
/512);
7444 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7445 window
/2, (unsigned long long)max_sectors
/2);
7447 atomic_set(&mddev
->recovery_active
, 0);
7452 "md: resuming %s of %s from checkpoint.\n",
7453 desc
, mdname(mddev
));
7454 mddev
->curr_resync
= j
;
7456 mddev
->curr_resync_completed
= j
;
7458 blk_start_plug(&plug
);
7459 while (j
< max_sectors
) {
7464 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7465 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7466 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7467 > (max_sectors
>> 4)) ||
7468 (j
- mddev
->curr_resync_completed
)*2
7469 >= mddev
->resync_max
- mddev
->curr_resync_completed
7471 /* time to update curr_resync_completed */
7472 wait_event(mddev
->recovery_wait
,
7473 atomic_read(&mddev
->recovery_active
) == 0);
7474 mddev
->curr_resync_completed
= j
;
7475 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7476 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7479 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7480 /* As this condition is controlled by user-space,
7481 * we can block indefinitely, so use '_interruptible'
7482 * to avoid triggering warnings.
7484 flush_signals(current
); /* just in case */
7485 wait_event_interruptible(mddev
->recovery_wait
,
7486 mddev
->resync_max
> j
7487 || kthread_should_stop());
7490 if (kthread_should_stop())
7493 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7494 currspeed
< speed_min(mddev
));
7496 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7500 if (!skipped
) { /* actual IO requested */
7501 io_sectors
+= sectors
;
7502 atomic_add(sectors
, &mddev
->recovery_active
);
7505 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7509 if (j
>1) mddev
->curr_resync
= j
;
7510 mddev
->curr_mark_cnt
= io_sectors
;
7511 if (last_check
== 0)
7512 /* this is the earliest that rebuild will be
7513 * visible in /proc/mdstat
7515 md_new_event(mddev
);
7517 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7520 last_check
= io_sectors
;
7522 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7524 int next
= (last_mark
+1) % SYNC_MARKS
;
7526 mddev
->resync_mark
= mark
[next
];
7527 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7528 mark
[next
] = jiffies
;
7529 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7534 if (kthread_should_stop())
7539 * this loop exits only if either when we are slower than
7540 * the 'hard' speed limit, or the system was IO-idle for
7542 * the system might be non-idle CPU-wise, but we only care
7543 * about not overloading the IO subsystem. (things like an
7544 * e2fsck being done on the RAID array should execute fast)
7548 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7549 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7551 if (currspeed
> speed_min(mddev
)) {
7552 if ((currspeed
> speed_max(mddev
)) ||
7553 !is_mddev_idle(mddev
, 0)) {
7559 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7561 * this also signals 'finished resyncing' to md_stop
7564 blk_finish_plug(&plug
);
7565 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7567 /* tell personality that we are finished */
7568 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7570 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7571 mddev
->curr_resync
> 2) {
7572 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7573 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7574 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7576 "md: checkpointing %s of %s.\n",
7577 desc
, mdname(mddev
));
7578 mddev
->recovery_cp
=
7579 mddev
->curr_resync_completed
;
7582 mddev
->recovery_cp
= MaxSector
;
7584 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7585 mddev
->curr_resync
= MaxSector
;
7587 rdev_for_each_rcu(rdev
, mddev
)
7588 if (rdev
->raid_disk
>= 0 &&
7589 mddev
->delta_disks
>= 0 &&
7590 !test_bit(Faulty
, &rdev
->flags
) &&
7591 !test_bit(In_sync
, &rdev
->flags
) &&
7592 rdev
->recovery_offset
< mddev
->curr_resync
)
7593 rdev
->recovery_offset
= mddev
->curr_resync
;
7598 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7600 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7601 /* We completed so min/max setting can be forgotten if used. */
7602 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7603 mddev
->resync_min
= 0;
7604 mddev
->resync_max
= MaxSector
;
7605 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7606 mddev
->resync_min
= mddev
->curr_resync_completed
;
7607 mddev
->curr_resync
= 0;
7608 wake_up(&resync_wait
);
7609 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7610 md_wakeup_thread(mddev
->thread
);
7615 * got a signal, exit.
7618 "md: md_do_sync() got signal ... exiting\n");
7619 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7623 EXPORT_SYMBOL_GPL(md_do_sync
);
7625 static int remove_and_add_spares(struct mddev
*mddev
)
7627 struct md_rdev
*rdev
;
7631 mddev
->curr_resync_completed
= 0;
7633 rdev_for_each(rdev
, mddev
)
7634 if (rdev
->raid_disk
>= 0 &&
7635 !test_bit(Blocked
, &rdev
->flags
) &&
7636 (test_bit(Faulty
, &rdev
->flags
) ||
7637 ! test_bit(In_sync
, &rdev
->flags
)) &&
7638 atomic_read(&rdev
->nr_pending
)==0) {
7639 if (mddev
->pers
->hot_remove_disk(
7640 mddev
, rdev
) == 0) {
7641 sysfs_unlink_rdev(mddev
, rdev
);
7642 rdev
->raid_disk
= -1;
7647 sysfs_notify(&mddev
->kobj
, NULL
,
7651 rdev_for_each(rdev
, mddev
) {
7652 if (rdev
->raid_disk
>= 0 &&
7653 !test_bit(In_sync
, &rdev
->flags
) &&
7654 !test_bit(Faulty
, &rdev
->flags
))
7656 if (rdev
->raid_disk
< 0
7657 && !test_bit(Faulty
, &rdev
->flags
)) {
7658 rdev
->recovery_offset
= 0;
7660 hot_add_disk(mddev
, rdev
) == 0) {
7661 if (sysfs_link_rdev(mddev
, rdev
))
7662 /* failure here is OK */;
7664 md_new_event(mddev
);
7665 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7672 static void reap_sync_thread(struct mddev
*mddev
)
7674 struct md_rdev
*rdev
;
7676 /* resync has finished, collect result */
7677 md_unregister_thread(&mddev
->sync_thread
);
7678 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7679 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7681 /* activate any spares */
7682 if (mddev
->pers
->spare_active(mddev
))
7683 sysfs_notify(&mddev
->kobj
, NULL
,
7686 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7687 mddev
->pers
->finish_reshape
)
7688 mddev
->pers
->finish_reshape(mddev
);
7690 /* If array is no-longer degraded, then any saved_raid_disk
7691 * information must be scrapped. Also if any device is now
7692 * In_sync we must scrape the saved_raid_disk for that device
7693 * do the superblock for an incrementally recovered device
7696 rdev_for_each(rdev
, mddev
)
7697 if (!mddev
->degraded
||
7698 test_bit(In_sync
, &rdev
->flags
))
7699 rdev
->saved_raid_disk
= -1;
7701 md_update_sb(mddev
, 1);
7702 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7703 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7704 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7705 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7706 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7707 /* flag recovery needed just to double check */
7708 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7709 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7710 md_new_event(mddev
);
7711 if (mddev
->event_work
.func
)
7712 queue_work(md_misc_wq
, &mddev
->event_work
);
7716 * This routine is regularly called by all per-raid-array threads to
7717 * deal with generic issues like resync and super-block update.
7718 * Raid personalities that don't have a thread (linear/raid0) do not
7719 * need this as they never do any recovery or update the superblock.
7721 * It does not do any resync itself, but rather "forks" off other threads
7722 * to do that as needed.
7723 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7724 * "->recovery" and create a thread at ->sync_thread.
7725 * When the thread finishes it sets MD_RECOVERY_DONE
7726 * and wakeups up this thread which will reap the thread and finish up.
7727 * This thread also removes any faulty devices (with nr_pending == 0).
7729 * The overall approach is:
7730 * 1/ if the superblock needs updating, update it.
7731 * 2/ If a recovery thread is running, don't do anything else.
7732 * 3/ If recovery has finished, clean up, possibly marking spares active.
7733 * 4/ If there are any faulty devices, remove them.
7734 * 5/ If array is degraded, try to add spares devices
7735 * 6/ If array has spares or is not in-sync, start a resync thread.
7737 void md_check_recovery(struct mddev
*mddev
)
7739 if (mddev
->suspended
)
7743 bitmap_daemon_work(mddev
);
7745 if (signal_pending(current
)) {
7746 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7747 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7749 mddev
->safemode
= 2;
7751 flush_signals(current
);
7754 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7757 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7758 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7759 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7760 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7761 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7762 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7766 if (mddev_trylock(mddev
)) {
7770 /* Only thing we do on a ro array is remove
7773 struct md_rdev
*rdev
;
7774 rdev_for_each(rdev
, mddev
)
7775 if (rdev
->raid_disk
>= 0 &&
7776 !test_bit(Blocked
, &rdev
->flags
) &&
7777 test_bit(Faulty
, &rdev
->flags
) &&
7778 atomic_read(&rdev
->nr_pending
)==0) {
7779 if (mddev
->pers
->hot_remove_disk(
7780 mddev
, rdev
) == 0) {
7781 sysfs_unlink_rdev(mddev
, rdev
);
7782 rdev
->raid_disk
= -1;
7785 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7789 if (!mddev
->external
) {
7791 spin_lock_irq(&mddev
->write_lock
);
7792 if (mddev
->safemode
&&
7793 !atomic_read(&mddev
->writes_pending
) &&
7795 mddev
->recovery_cp
== MaxSector
) {
7798 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7800 if (mddev
->safemode
== 1)
7801 mddev
->safemode
= 0;
7802 spin_unlock_irq(&mddev
->write_lock
);
7804 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7808 md_update_sb(mddev
, 0);
7810 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7811 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7812 /* resync/recovery still happening */
7813 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7816 if (mddev
->sync_thread
) {
7817 reap_sync_thread(mddev
);
7820 /* Set RUNNING before clearing NEEDED to avoid
7821 * any transients in the value of "sync_action".
7823 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7824 /* Clear some bits that don't mean anything, but
7827 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7828 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7830 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7831 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7833 /* no recovery is running.
7834 * remove any failed drives, then
7835 * add spares if possible.
7836 * Spare are also removed and re-added, to allow
7837 * the personality to fail the re-add.
7840 if (mddev
->reshape_position
!= MaxSector
) {
7841 if (mddev
->pers
->check_reshape
== NULL
||
7842 mddev
->pers
->check_reshape(mddev
) != 0)
7843 /* Cannot proceed */
7845 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7846 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7847 } else if ((spares
= remove_and_add_spares(mddev
))) {
7848 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7849 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7850 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7851 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7852 } else if (mddev
->recovery_cp
< MaxSector
) {
7853 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7854 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7855 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7856 /* nothing to be done ... */
7859 if (mddev
->pers
->sync_request
) {
7861 /* We are adding a device or devices to an array
7862 * which has the bitmap stored on all devices.
7863 * So make sure all bitmap pages get written
7865 bitmap_write_all(mddev
->bitmap
);
7867 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7870 if (!mddev
->sync_thread
) {
7871 printk(KERN_ERR
"%s: could not start resync"
7874 /* leave the spares where they are, it shouldn't hurt */
7875 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7876 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7877 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7878 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7879 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7881 md_wakeup_thread(mddev
->sync_thread
);
7882 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7883 md_new_event(mddev
);
7886 if (!mddev
->sync_thread
) {
7887 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7888 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7890 if (mddev
->sysfs_action
)
7891 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7893 mddev_unlock(mddev
);
7897 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7899 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7900 wait_event_timeout(rdev
->blocked_wait
,
7901 !test_bit(Blocked
, &rdev
->flags
) &&
7902 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7903 msecs_to_jiffies(5000));
7904 rdev_dec_pending(rdev
, mddev
);
7906 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7908 void md_finish_reshape(struct mddev
*mddev
)
7910 /* called be personality module when reshape completes. */
7911 struct md_rdev
*rdev
;
7913 rdev_for_each(rdev
, mddev
) {
7914 if (rdev
->data_offset
> rdev
->new_data_offset
)
7915 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
7917 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
7918 rdev
->data_offset
= rdev
->new_data_offset
;
7921 EXPORT_SYMBOL(md_finish_reshape
);
7923 /* Bad block management.
7924 * We can record which blocks on each device are 'bad' and so just
7925 * fail those blocks, or that stripe, rather than the whole device.
7926 * Entries in the bad-block table are 64bits wide. This comprises:
7927 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7928 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7929 * A 'shift' can be set so that larger blocks are tracked and
7930 * consequently larger devices can be covered.
7931 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7933 * Locking of the bad-block table uses a seqlock so md_is_badblock
7934 * might need to retry if it is very unlucky.
7935 * We will sometimes want to check for bad blocks in a bi_end_io function,
7936 * so we use the write_seqlock_irq variant.
7938 * When looking for a bad block we specify a range and want to
7939 * know if any block in the range is bad. So we binary-search
7940 * to the last range that starts at-or-before the given endpoint,
7941 * (or "before the sector after the target range")
7942 * then see if it ends after the given start.
7944 * 0 if there are no known bad blocks in the range
7945 * 1 if there are known bad block which are all acknowledged
7946 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7947 * plus the start/length of the first bad section we overlap.
7949 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7950 sector_t
*first_bad
, int *bad_sectors
)
7956 sector_t target
= s
+ sectors
;
7959 if (bb
->shift
> 0) {
7960 /* round the start down, and the end up */
7962 target
+= (1<<bb
->shift
) - 1;
7963 target
>>= bb
->shift
;
7964 sectors
= target
- s
;
7966 /* 'target' is now the first block after the bad range */
7969 seq
= read_seqbegin(&bb
->lock
);
7973 /* Binary search between lo and hi for 'target'
7974 * i.e. for the last range that starts before 'target'
7976 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7977 * are known not to be the last range before target.
7978 * VARIANT: hi-lo is the number of possible
7979 * ranges, and decreases until it reaches 1
7981 while (hi
- lo
> 1) {
7982 int mid
= (lo
+ hi
) / 2;
7983 sector_t a
= BB_OFFSET(p
[mid
]);
7985 /* This could still be the one, earlier ranges
7989 /* This and later ranges are definitely out. */
7992 /* 'lo' might be the last that started before target, but 'hi' isn't */
7994 /* need to check all range that end after 's' to see if
7995 * any are unacknowledged.
7998 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7999 if (BB_OFFSET(p
[lo
]) < target
) {
8000 /* starts before the end, and finishes after
8001 * the start, so they must overlap
8003 if (rv
!= -1 && BB_ACK(p
[lo
]))
8007 *first_bad
= BB_OFFSET(p
[lo
]);
8008 *bad_sectors
= BB_LEN(p
[lo
]);
8014 if (read_seqretry(&bb
->lock
, seq
))
8019 EXPORT_SYMBOL_GPL(md_is_badblock
);
8022 * Add a range of bad blocks to the table.
8023 * This might extend the table, or might contract it
8024 * if two adjacent ranges can be merged.
8025 * We binary-search to find the 'insertion' point, then
8026 * decide how best to handle it.
8028 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
8036 /* badblocks are disabled */
8040 /* round the start down, and the end up */
8041 sector_t next
= s
+ sectors
;
8043 next
+= (1<<bb
->shift
) - 1;
8048 write_seqlock_irq(&bb
->lock
);
8053 /* Find the last range that starts at-or-before 's' */
8054 while (hi
- lo
> 1) {
8055 int mid
= (lo
+ hi
) / 2;
8056 sector_t a
= BB_OFFSET(p
[mid
]);
8062 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8066 /* we found a range that might merge with the start
8069 sector_t a
= BB_OFFSET(p
[lo
]);
8070 sector_t e
= a
+ BB_LEN(p
[lo
]);
8071 int ack
= BB_ACK(p
[lo
]);
8073 /* Yes, we can merge with a previous range */
8074 if (s
== a
&& s
+ sectors
>= e
)
8075 /* new range covers old */
8078 ack
= ack
&& acknowledged
;
8080 if (e
< s
+ sectors
)
8082 if (e
- a
<= BB_MAX_LEN
) {
8083 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8086 /* does not all fit in one range,
8087 * make p[lo] maximal
8089 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8090 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8096 if (sectors
&& hi
< bb
->count
) {
8097 /* 'hi' points to the first range that starts after 's'.
8098 * Maybe we can merge with the start of that range */
8099 sector_t a
= BB_OFFSET(p
[hi
]);
8100 sector_t e
= a
+ BB_LEN(p
[hi
]);
8101 int ack
= BB_ACK(p
[hi
]);
8102 if (a
<= s
+ sectors
) {
8103 /* merging is possible */
8104 if (e
<= s
+ sectors
) {
8109 ack
= ack
&& acknowledged
;
8112 if (e
- a
<= BB_MAX_LEN
) {
8113 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8116 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8124 if (sectors
== 0 && hi
< bb
->count
) {
8125 /* we might be able to combine lo and hi */
8126 /* Note: 's' is at the end of 'lo' */
8127 sector_t a
= BB_OFFSET(p
[hi
]);
8128 int lolen
= BB_LEN(p
[lo
]);
8129 int hilen
= BB_LEN(p
[hi
]);
8130 int newlen
= lolen
+ hilen
- (s
- a
);
8131 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8132 /* yes, we can combine them */
8133 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8134 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8135 memmove(p
+ hi
, p
+ hi
+ 1,
8136 (bb
->count
- hi
- 1) * 8);
8141 /* didn't merge (it all).
8142 * Need to add a range just before 'hi' */
8143 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8144 /* No room for more */
8148 int this_sectors
= sectors
;
8149 memmove(p
+ hi
+ 1, p
+ hi
,
8150 (bb
->count
- hi
) * 8);
8153 if (this_sectors
> BB_MAX_LEN
)
8154 this_sectors
= BB_MAX_LEN
;
8155 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8156 sectors
-= this_sectors
;
8163 bb
->unacked_exist
= 1;
8164 write_sequnlock_irq(&bb
->lock
);
8169 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8174 s
+= rdev
->new_data_offset
;
8176 s
+= rdev
->data_offset
;
8177 rv
= md_set_badblocks(&rdev
->badblocks
,
8180 /* Make sure they get written out promptly */
8181 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8182 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8183 md_wakeup_thread(rdev
->mddev
->thread
);
8187 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8190 * Remove a range of bad blocks from the table.
8191 * This may involve extending the table if we spilt a region,
8192 * but it must not fail. So if the table becomes full, we just
8193 * drop the remove request.
8195 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8199 sector_t target
= s
+ sectors
;
8202 if (bb
->shift
> 0) {
8203 /* When clearing we round the start up and the end down.
8204 * This should not matter as the shift should align with
8205 * the block size and no rounding should ever be needed.
8206 * However it is better the think a block is bad when it
8207 * isn't than to think a block is not bad when it is.
8209 s
+= (1<<bb
->shift
) - 1;
8211 target
>>= bb
->shift
;
8212 sectors
= target
- s
;
8215 write_seqlock_irq(&bb
->lock
);
8220 /* Find the last range that starts before 'target' */
8221 while (hi
- lo
> 1) {
8222 int mid
= (lo
+ hi
) / 2;
8223 sector_t a
= BB_OFFSET(p
[mid
]);
8230 /* p[lo] is the last range that could overlap the
8231 * current range. Earlier ranges could also overlap,
8232 * but only this one can overlap the end of the range.
8234 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8235 /* Partial overlap, leave the tail of this range */
8236 int ack
= BB_ACK(p
[lo
]);
8237 sector_t a
= BB_OFFSET(p
[lo
]);
8238 sector_t end
= a
+ BB_LEN(p
[lo
]);
8241 /* we need to split this range */
8242 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8246 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8248 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8251 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8252 /* there is no longer an overlap */
8257 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8258 /* This range does overlap */
8259 if (BB_OFFSET(p
[lo
]) < s
) {
8260 /* Keep the early parts of this range. */
8261 int ack
= BB_ACK(p
[lo
]);
8262 sector_t start
= BB_OFFSET(p
[lo
]);
8263 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8264 /* now low doesn't overlap, so.. */
8269 /* 'lo' is strictly before, 'hi' is strictly after,
8270 * anything between needs to be discarded
8273 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8274 bb
->count
-= (hi
- lo
- 1);
8280 write_sequnlock_irq(&bb
->lock
);
8284 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8288 s
+= rdev
->new_data_offset
;
8290 s
+= rdev
->data_offset
;
8291 return md_clear_badblocks(&rdev
->badblocks
,
8294 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8297 * Acknowledge all bad blocks in a list.
8298 * This only succeeds if ->changed is clear. It is used by
8299 * in-kernel metadata updates
8301 void md_ack_all_badblocks(struct badblocks
*bb
)
8303 if (bb
->page
== NULL
|| bb
->changed
)
8304 /* no point even trying */
8306 write_seqlock_irq(&bb
->lock
);
8308 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8311 for (i
= 0; i
< bb
->count
; i
++) {
8312 if (!BB_ACK(p
[i
])) {
8313 sector_t start
= BB_OFFSET(p
[i
]);
8314 int len
= BB_LEN(p
[i
]);
8315 p
[i
] = BB_MAKE(start
, len
, 1);
8318 bb
->unacked_exist
= 0;
8320 write_sequnlock_irq(&bb
->lock
);
8322 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8324 /* sysfs access to bad-blocks list.
8325 * We present two files.
8326 * 'bad-blocks' lists sector numbers and lengths of ranges that
8327 * are recorded as bad. The list is truncated to fit within
8328 * the one-page limit of sysfs.
8329 * Writing "sector length" to this file adds an acknowledged
8331 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8332 * been acknowledged. Writing to this file adds bad blocks
8333 * without acknowledging them. This is largely for testing.
8337 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8348 seq
= read_seqbegin(&bb
->lock
);
8353 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8354 sector_t s
= BB_OFFSET(p
[i
]);
8355 unsigned int length
= BB_LEN(p
[i
]);
8356 int ack
= BB_ACK(p
[i
]);
8362 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8363 (unsigned long long)s
<< bb
->shift
,
8364 length
<< bb
->shift
);
8366 if (unack
&& len
== 0)
8367 bb
->unacked_exist
= 0;
8369 if (read_seqretry(&bb
->lock
, seq
))
8378 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8380 unsigned long long sector
;
8384 /* Allow clearing via sysfs *only* for testing/debugging.
8385 * Normally only a successful write may clear a badblock
8388 if (page
[0] == '-') {
8392 #endif /* DO_DEBUG */
8394 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8396 if (newline
!= '\n')
8408 md_clear_badblocks(bb
, sector
, length
);
8411 #endif /* DO_DEBUG */
8412 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8418 static int md_notify_reboot(struct notifier_block
*this,
8419 unsigned long code
, void *x
)
8421 struct list_head
*tmp
;
8422 struct mddev
*mddev
;
8425 for_each_mddev(mddev
, tmp
) {
8426 if (mddev_trylock(mddev
)) {
8428 __md_stop_writes(mddev
);
8429 mddev
->safemode
= 2;
8430 mddev_unlock(mddev
);
8435 * certain more exotic SCSI devices are known to be
8436 * volatile wrt too early system reboots. While the
8437 * right place to handle this issue is the given
8438 * driver, we do want to have a safe RAID driver ...
8446 static struct notifier_block md_notifier
= {
8447 .notifier_call
= md_notify_reboot
,
8449 .priority
= INT_MAX
, /* before any real devices */
8452 static void md_geninit(void)
8454 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8456 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8459 static int __init
md_init(void)
8463 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8467 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8471 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8474 if ((ret
= register_blkdev(0, "mdp")) < 0)
8478 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8479 md_probe
, NULL
, NULL
);
8480 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8481 md_probe
, NULL
, NULL
);
8483 register_reboot_notifier(&md_notifier
);
8484 raid_table_header
= register_sysctl_table(raid_root_table
);
8490 unregister_blkdev(MD_MAJOR
, "md");
8492 destroy_workqueue(md_misc_wq
);
8494 destroy_workqueue(md_wq
);
8502 * Searches all registered partitions for autorun RAID arrays
8506 static LIST_HEAD(all_detected_devices
);
8507 struct detected_devices_node
{
8508 struct list_head list
;
8512 void md_autodetect_dev(dev_t dev
)
8514 struct detected_devices_node
*node_detected_dev
;
8516 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8517 if (node_detected_dev
) {
8518 node_detected_dev
->dev
= dev
;
8519 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8521 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8522 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8527 static void autostart_arrays(int part
)
8529 struct md_rdev
*rdev
;
8530 struct detected_devices_node
*node_detected_dev
;
8532 int i_scanned
, i_passed
;
8537 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8539 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8541 node_detected_dev
= list_entry(all_detected_devices
.next
,
8542 struct detected_devices_node
, list
);
8543 list_del(&node_detected_dev
->list
);
8544 dev
= node_detected_dev
->dev
;
8545 kfree(node_detected_dev
);
8546 rdev
= md_import_device(dev
,0, 90);
8550 if (test_bit(Faulty
, &rdev
->flags
)) {
8554 set_bit(AutoDetected
, &rdev
->flags
);
8555 list_add(&rdev
->same_set
, &pending_raid_disks
);
8559 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8560 i_scanned
, i_passed
);
8562 autorun_devices(part
);
8565 #endif /* !MODULE */
8567 static __exit
void md_exit(void)
8569 struct mddev
*mddev
;
8570 struct list_head
*tmp
;
8572 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8573 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8575 unregister_blkdev(MD_MAJOR
,"md");
8576 unregister_blkdev(mdp_major
, "mdp");
8577 unregister_reboot_notifier(&md_notifier
);
8578 unregister_sysctl_table(raid_table_header
);
8579 remove_proc_entry("mdstat", NULL
);
8580 for_each_mddev(mddev
, tmp
) {
8581 export_array(mddev
);
8582 mddev
->hold_active
= 0;
8584 destroy_workqueue(md_misc_wq
);
8585 destroy_workqueue(md_wq
);
8588 subsys_initcall(md_init
);
8589 module_exit(md_exit
)
8591 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8593 return sprintf(buffer
, "%d", start_readonly
);
8595 static int set_ro(const char *val
, struct kernel_param
*kp
)
8598 int num
= simple_strtoul(val
, &e
, 10);
8599 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8600 start_readonly
= num
;
8606 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8607 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8609 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8611 EXPORT_SYMBOL(register_md_personality
);
8612 EXPORT_SYMBOL(unregister_md_personality
);
8613 EXPORT_SYMBOL(md_error
);
8614 EXPORT_SYMBOL(md_done_sync
);
8615 EXPORT_SYMBOL(md_write_start
);
8616 EXPORT_SYMBOL(md_write_end
);
8617 EXPORT_SYMBOL(md_register_thread
);
8618 EXPORT_SYMBOL(md_unregister_thread
);
8619 EXPORT_SYMBOL(md_wakeup_thread
);
8620 EXPORT_SYMBOL(md_check_recovery
);
8621 MODULE_LICENSE("GPL");
8622 MODULE_DESCRIPTION("MD RAID framework");
8624 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);