2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
59 static void autostart_arrays(int part
);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list
);
68 static DEFINE_SPINLOCK(pers_lock
);
70 static void md_print_devices(void);
72 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
73 static struct workqueue_struct
*md_wq
;
74 static struct workqueue_struct
*md_misc_wq
;
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79 * Default number of read corrections we'll attempt on an rdev
80 * before ejecting it from the array. We divide the read error
81 * count by 2 for every hour elapsed between read errors.
83 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
86 * is 1000 KB/sec, so the extra system load does not show up that much.
87 * Increase it if you want to have more _guaranteed_ speed. Note that
88 * the RAID driver will use the maximum available bandwidth if the IO
89 * subsystem is idle. There is also an 'absolute maximum' reconstruction
90 * speed limit - in case reconstruction slows down your system despite
93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
94 * or /sys/block/mdX/md/sync_speed_{min,max}
97 static int sysctl_speed_limit_min
= 1000;
98 static int sysctl_speed_limit_max
= 200000;
99 static inline int speed_min(struct mddev
*mddev
)
101 return mddev
->sync_speed_min
?
102 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
105 static inline int speed_max(struct mddev
*mddev
)
107 return mddev
->sync_speed_max
?
108 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
111 static struct ctl_table_header
*raid_table_header
;
113 static ctl_table raid_table
[] = {
115 .procname
= "speed_limit_min",
116 .data
= &sysctl_speed_limit_min
,
117 .maxlen
= sizeof(int),
118 .mode
= S_IRUGO
|S_IWUSR
,
119 .proc_handler
= proc_dointvec
,
122 .procname
= "speed_limit_max",
123 .data
= &sysctl_speed_limit_max
,
124 .maxlen
= sizeof(int),
125 .mode
= S_IRUGO
|S_IWUSR
,
126 .proc_handler
= proc_dointvec
,
131 static ctl_table raid_dir_table
[] = {
135 .mode
= S_IRUGO
|S_IXUGO
,
141 static ctl_table raid_root_table
[] = {
146 .child
= raid_dir_table
,
151 static const struct block_device_operations md_fops
;
153 static int start_readonly
;
156 * like bio_clone, but with a local bio set
159 static void mddev_bio_destructor(struct bio
*bio
)
161 struct mddev
*mddev
, **mddevp
;
166 bio_free(bio
, mddev
->bio_set
);
169 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
173 struct mddev
**mddevp
;
175 if (!mddev
|| !mddev
->bio_set
)
176 return bio_alloc(gfp_mask
, nr_iovecs
);
178 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
184 b
->bi_destructor
= mddev_bio_destructor
;
187 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
189 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
193 struct mddev
**mddevp
;
195 if (!mddev
|| !mddev
->bio_set
)
196 return bio_clone(bio
, gfp_mask
);
198 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
204 b
->bi_destructor
= mddev_bio_destructor
;
206 if (bio_integrity(bio
)) {
209 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
219 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
221 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
223 /* 'bio' is a cloned bio which we need to trim to match
224 * the given offset and size.
225 * This requires adjusting bi_sector, bi_size, and bi_io_vec
228 struct bio_vec
*bvec
;
232 if (offset
== 0 && size
== bio
->bi_size
)
235 bio
->bi_sector
+= offset
;
238 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
240 while (bio
->bi_idx
< bio
->bi_vcnt
&&
241 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
242 /* remove this whole bio_vec */
243 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
246 if (bio
->bi_idx
< bio
->bi_vcnt
) {
247 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
248 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
250 /* avoid any complications with bi_idx being non-zero*/
252 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
253 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
254 bio
->bi_vcnt
-= bio
->bi_idx
;
257 /* Make sure vcnt and last bv are not too big */
258 bio_for_each_segment(bvec
, bio
, i
) {
259 if (sofar
+ bvec
->bv_len
> size
)
260 bvec
->bv_len
= size
- sofar
;
261 if (bvec
->bv_len
== 0) {
265 sofar
+= bvec
->bv_len
;
268 EXPORT_SYMBOL_GPL(md_trim_bio
);
271 * We have a system wide 'event count' that is incremented
272 * on any 'interesting' event, and readers of /proc/mdstat
273 * can use 'poll' or 'select' to find out when the event
277 * start array, stop array, error, add device, remove device,
278 * start build, activate spare
280 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
281 static atomic_t md_event_count
;
282 void md_new_event(struct mddev
*mddev
)
284 atomic_inc(&md_event_count
);
285 wake_up(&md_event_waiters
);
287 EXPORT_SYMBOL_GPL(md_new_event
);
289 /* Alternate version that can be called from interrupts
290 * when calling sysfs_notify isn't needed.
292 static void md_new_event_inintr(struct mddev
*mddev
)
294 atomic_inc(&md_event_count
);
295 wake_up(&md_event_waiters
);
299 * Enables to iterate over all existing md arrays
300 * all_mddevs_lock protects this list.
302 static LIST_HEAD(all_mddevs
);
303 static DEFINE_SPINLOCK(all_mddevs_lock
);
307 * iterates through all used mddevs in the system.
308 * We take care to grab the all_mddevs_lock whenever navigating
309 * the list, and to always hold a refcount when unlocked.
310 * Any code which breaks out of this loop while own
311 * a reference to the current mddev and must mddev_put it.
313 #define for_each_mddev(_mddev,_tmp) \
315 for (({ spin_lock(&all_mddevs_lock); \
316 _tmp = all_mddevs.next; \
318 ({ if (_tmp != &all_mddevs) \
319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
320 spin_unlock(&all_mddevs_lock); \
321 if (_mddev) mddev_put(_mddev); \
322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
323 _tmp != &all_mddevs;}); \
324 ({ spin_lock(&all_mddevs_lock); \
325 _tmp = _tmp->next;}) \
329 /* Rather than calling directly into the personality make_request function,
330 * IO requests come here first so that we can check if the device is
331 * being suspended pending a reconfiguration.
332 * We hold a refcount over the call to ->make_request. By the time that
333 * call has finished, the bio has been linked into some internal structure
334 * and so is visible to ->quiesce(), so we don't need the refcount any more.
336 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
338 const int rw
= bio_data_dir(bio
);
339 struct mddev
*mddev
= q
->queuedata
;
341 unsigned int sectors
;
343 if (mddev
== NULL
|| mddev
->pers
== NULL
348 smp_rmb(); /* Ensure implications of 'active' are visible */
350 if (mddev
->suspended
) {
353 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
354 TASK_UNINTERRUPTIBLE
);
355 if (!mddev
->suspended
)
361 finish_wait(&mddev
->sb_wait
, &__wait
);
363 atomic_inc(&mddev
->active_io
);
367 * save the sectors now since our bio can
368 * go away inside make_request
370 sectors
= bio_sectors(bio
);
371 mddev
->pers
->make_request(mddev
, bio
);
373 cpu
= part_stat_lock();
374 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
375 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
378 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
379 wake_up(&mddev
->sb_wait
);
382 /* mddev_suspend makes sure no new requests are submitted
383 * to the device, and that any requests that have been submitted
384 * are completely handled.
385 * Once ->stop is called and completes, the module will be completely
388 void mddev_suspend(struct mddev
*mddev
)
390 BUG_ON(mddev
->suspended
);
391 mddev
->suspended
= 1;
393 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
394 mddev
->pers
->quiesce(mddev
, 1);
396 EXPORT_SYMBOL_GPL(mddev_suspend
);
398 void mddev_resume(struct mddev
*mddev
)
400 mddev
->suspended
= 0;
401 wake_up(&mddev
->sb_wait
);
402 mddev
->pers
->quiesce(mddev
, 0);
404 md_wakeup_thread(mddev
->thread
);
405 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
407 EXPORT_SYMBOL_GPL(mddev_resume
);
409 int mddev_congested(struct mddev
*mddev
, int bits
)
411 return mddev
->suspended
;
413 EXPORT_SYMBOL(mddev_congested
);
416 * Generic flush handling for md
419 static void md_end_flush(struct bio
*bio
, int err
)
421 struct md_rdev
*rdev
= bio
->bi_private
;
422 struct mddev
*mddev
= rdev
->mddev
;
424 rdev_dec_pending(rdev
, mddev
);
426 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
427 /* The pre-request flush has finished */
428 queue_work(md_wq
, &mddev
->flush_work
);
433 static void md_submit_flush_data(struct work_struct
*ws
);
435 static void submit_flushes(struct work_struct
*ws
)
437 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
438 struct md_rdev
*rdev
;
440 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
441 atomic_set(&mddev
->flush_pending
, 1);
443 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
444 if (rdev
->raid_disk
>= 0 &&
445 !test_bit(Faulty
, &rdev
->flags
)) {
446 /* Take two references, one is dropped
447 * when request finishes, one after
448 * we reclaim rcu_read_lock
451 atomic_inc(&rdev
->nr_pending
);
452 atomic_inc(&rdev
->nr_pending
);
454 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
455 bi
->bi_end_io
= md_end_flush
;
456 bi
->bi_private
= rdev
;
457 bi
->bi_bdev
= rdev
->bdev
;
458 atomic_inc(&mddev
->flush_pending
);
459 submit_bio(WRITE_FLUSH
, bi
);
461 rdev_dec_pending(rdev
, mddev
);
464 if (atomic_dec_and_test(&mddev
->flush_pending
))
465 queue_work(md_wq
, &mddev
->flush_work
);
468 static void md_submit_flush_data(struct work_struct
*ws
)
470 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
471 struct bio
*bio
= mddev
->flush_bio
;
473 if (bio
->bi_size
== 0)
474 /* an empty barrier - all done */
477 bio
->bi_rw
&= ~REQ_FLUSH
;
478 mddev
->pers
->make_request(mddev
, bio
);
481 mddev
->flush_bio
= NULL
;
482 wake_up(&mddev
->sb_wait
);
485 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
487 spin_lock_irq(&mddev
->write_lock
);
488 wait_event_lock_irq(mddev
->sb_wait
,
490 mddev
->write_lock
, /*nothing*/);
491 mddev
->flush_bio
= bio
;
492 spin_unlock_irq(&mddev
->write_lock
);
494 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
495 queue_work(md_wq
, &mddev
->flush_work
);
497 EXPORT_SYMBOL(md_flush_request
);
499 /* Support for plugging.
500 * This mirrors the plugging support in request_queue, but does not
501 * require having a whole queue or request structures.
502 * We allocate an md_plug_cb for each md device and each thread it gets
503 * plugged on. This links tot the private plug_handle structure in the
504 * personality data where we keep a count of the number of outstanding
505 * plugs so other code can see if a plug is active.
508 struct blk_plug_cb cb
;
512 static void plugger_unplug(struct blk_plug_cb
*cb
)
514 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
515 if (atomic_dec_and_test(&mdcb
->mddev
->plug_cnt
))
516 md_wakeup_thread(mdcb
->mddev
->thread
);
520 /* Check that an unplug wakeup will come shortly.
521 * If not, wakeup the md thread immediately
523 int mddev_check_plugged(struct mddev
*mddev
)
525 struct blk_plug
*plug
= current
->plug
;
526 struct md_plug_cb
*mdcb
;
531 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
532 if (mdcb
->cb
.callback
== plugger_unplug
&&
533 mdcb
->mddev
== mddev
) {
534 /* Already on the list, move to top */
535 if (mdcb
!= list_first_entry(&plug
->cb_list
,
538 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
542 /* Not currently on the callback list */
543 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
548 mdcb
->cb
.callback
= plugger_unplug
;
549 atomic_inc(&mddev
->plug_cnt
);
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(&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 atomic_set(&mddev
->plug_cnt
, 0);
604 spin_lock_init(&mddev
->write_lock
);
605 atomic_set(&mddev
->flush_pending
, 0);
606 init_waitqueue_head(&mddev
->sb_wait
);
607 init_waitqueue_head(&mddev
->recovery_wait
);
608 mddev
->reshape_position
= MaxSector
;
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 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
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 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
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 static void free_disk_sb(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
;
820 static void super_written(struct bio
*bio
, int error
)
822 struct md_rdev
*rdev
= bio
->bi_private
;
823 struct mddev
*mddev
= rdev
->mddev
;
825 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
826 printk("md: super_written gets error=%d, uptodate=%d\n",
827 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
828 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
829 md_error(mddev
, rdev
);
832 if (atomic_dec_and_test(&mddev
->pending_writes
))
833 wake_up(&mddev
->sb_wait
);
837 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
838 sector_t sector
, int size
, struct page
*page
)
840 /* write first size bytes of page to sector of rdev
841 * Increment mddev->pending_writes before returning
842 * and decrement it on completion, waking up sb_wait
843 * if zero is reached.
844 * If an error occurred, call md_error
846 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
848 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
849 bio
->bi_sector
= sector
;
850 bio_add_page(bio
, page
, size
, 0);
851 bio
->bi_private
= rdev
;
852 bio
->bi_end_io
= super_written
;
854 atomic_inc(&mddev
->pending_writes
);
855 submit_bio(WRITE_FLUSH_FUA
, bio
);
858 void md_super_wait(struct mddev
*mddev
)
860 /* wait for all superblock writes that were scheduled to complete */
863 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
864 if (atomic_read(&mddev
->pending_writes
)==0)
868 finish_wait(&mddev
->sb_wait
, &wq
);
871 static void bi_complete(struct bio
*bio
, int error
)
873 complete((struct completion
*)bio
->bi_private
);
876 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
877 struct page
*page
, int rw
, bool metadata_op
)
879 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
880 struct completion event
;
885 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
886 rdev
->meta_bdev
: rdev
->bdev
;
888 bio
->bi_sector
= sector
+ rdev
->sb_start
;
890 bio
->bi_sector
= sector
+ rdev
->data_offset
;
891 bio_add_page(bio
, page
, size
, 0);
892 init_completion(&event
);
893 bio
->bi_private
= &event
;
894 bio
->bi_end_io
= bi_complete
;
896 wait_for_completion(&event
);
898 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
902 EXPORT_SYMBOL_GPL(sync_page_io
);
904 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
906 char b
[BDEVNAME_SIZE
];
907 if (!rdev
->sb_page
) {
915 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
921 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
922 bdevname(rdev
->bdev
,b
));
926 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
928 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
929 sb1
->set_uuid1
== sb2
->set_uuid1
&&
930 sb1
->set_uuid2
== sb2
->set_uuid2
&&
931 sb1
->set_uuid3
== sb2
->set_uuid3
;
934 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
937 mdp_super_t
*tmp1
, *tmp2
;
939 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
940 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
942 if (!tmp1
|| !tmp2
) {
944 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
952 * nr_disks is not constant
957 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
965 static u32
md_csum_fold(u32 csum
)
967 csum
= (csum
& 0xffff) + (csum
>> 16);
968 return (csum
& 0xffff) + (csum
>> 16);
971 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
974 u32
*sb32
= (u32
*)sb
;
976 unsigned int disk_csum
, csum
;
978 disk_csum
= sb
->sb_csum
;
981 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
983 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
987 /* This used to use csum_partial, which was wrong for several
988 * reasons including that different results are returned on
989 * different architectures. It isn't critical that we get exactly
990 * the same return value as before (we always csum_fold before
991 * testing, and that removes any differences). However as we
992 * know that csum_partial always returned a 16bit value on
993 * alphas, do a fold to maximise conformity to previous behaviour.
995 sb
->sb_csum
= md_csum_fold(disk_csum
);
997 sb
->sb_csum
= disk_csum
;
1004 * Handle superblock details.
1005 * We want to be able to handle multiple superblock formats
1006 * so we have a common interface to them all, and an array of
1007 * different handlers.
1008 * We rely on user-space to write the initial superblock, and support
1009 * reading and updating of superblocks.
1010 * Interface methods are:
1011 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1012 * loads and validates a superblock on dev.
1013 * if refdev != NULL, compare superblocks on both devices
1015 * 0 - dev has a superblock that is compatible with refdev
1016 * 1 - dev has a superblock that is compatible and newer than refdev
1017 * so dev should be used as the refdev in future
1018 * -EINVAL superblock incompatible or invalid
1019 * -othererror e.g. -EIO
1021 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1022 * Verify that dev is acceptable into mddev.
1023 * The first time, mddev->raid_disks will be 0, and data from
1024 * dev should be merged in. Subsequent calls check that dev
1025 * is new enough. Return 0 or -EINVAL
1027 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1028 * Update the superblock for rdev with data in mddev
1029 * This does not write to disc.
1035 struct module
*owner
;
1036 int (*load_super
)(struct md_rdev
*rdev
, struct md_rdev
*refdev
,
1038 int (*validate_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1039 void (*sync_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1040 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1041 sector_t num_sectors
);
1045 * Check that the given mddev has no bitmap.
1047 * This function is called from the run method of all personalities that do not
1048 * support bitmaps. It prints an error message and returns non-zero if mddev
1049 * has a bitmap. Otherwise, it returns 0.
1052 int md_check_no_bitmap(struct mddev
*mddev
)
1054 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1056 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1057 mdname(mddev
), mddev
->pers
->name
);
1060 EXPORT_SYMBOL(md_check_no_bitmap
);
1063 * load_super for 0.90.0
1065 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1067 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1072 * Calculate the position of the superblock (512byte sectors),
1073 * it's at the end of the disk.
1075 * It also happens to be a multiple of 4Kb.
1077 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1079 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1080 if (ret
) return ret
;
1084 bdevname(rdev
->bdev
, b
);
1085 sb
= page_address(rdev
->sb_page
);
1087 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1088 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1093 if (sb
->major_version
!= 0 ||
1094 sb
->minor_version
< 90 ||
1095 sb
->minor_version
> 91) {
1096 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1097 sb
->major_version
, sb
->minor_version
,
1102 if (sb
->raid_disks
<= 0)
1105 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1106 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1111 rdev
->preferred_minor
= sb
->md_minor
;
1112 rdev
->data_offset
= 0;
1113 rdev
->sb_size
= MD_SB_BYTES
;
1114 rdev
->badblocks
.shift
= -1;
1116 if (sb
->level
== LEVEL_MULTIPATH
)
1119 rdev
->desc_nr
= sb
->this_disk
.number
;
1125 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1126 if (!uuid_equal(refsb
, sb
)) {
1127 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1128 b
, bdevname(refdev
->bdev
,b2
));
1131 if (!sb_equal(refsb
, sb
)) {
1132 printk(KERN_WARNING
"md: %s has same UUID"
1133 " but different superblock to %s\n",
1134 b
, bdevname(refdev
->bdev
, b2
));
1138 ev2
= md_event(refsb
);
1144 rdev
->sectors
= rdev
->sb_start
;
1145 /* Limit to 4TB as metadata cannot record more than that */
1146 if (rdev
->sectors
>= (2ULL << 32))
1147 rdev
->sectors
= (2ULL << 32) - 2;
1149 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1150 /* "this cannot possibly happen" ... */
1158 * validate_super for 0.90.0
1160 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1163 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1164 __u64 ev1
= md_event(sb
);
1166 rdev
->raid_disk
= -1;
1167 clear_bit(Faulty
, &rdev
->flags
);
1168 clear_bit(In_sync
, &rdev
->flags
);
1169 clear_bit(WriteMostly
, &rdev
->flags
);
1171 if (mddev
->raid_disks
== 0) {
1172 mddev
->major_version
= 0;
1173 mddev
->minor_version
= sb
->minor_version
;
1174 mddev
->patch_version
= sb
->patch_version
;
1175 mddev
->external
= 0;
1176 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1177 mddev
->ctime
= sb
->ctime
;
1178 mddev
->utime
= sb
->utime
;
1179 mddev
->level
= sb
->level
;
1180 mddev
->clevel
[0] = 0;
1181 mddev
->layout
= sb
->layout
;
1182 mddev
->raid_disks
= sb
->raid_disks
;
1183 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1184 mddev
->events
= ev1
;
1185 mddev
->bitmap_info
.offset
= 0;
1186 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1188 if (mddev
->minor_version
>= 91) {
1189 mddev
->reshape_position
= sb
->reshape_position
;
1190 mddev
->delta_disks
= sb
->delta_disks
;
1191 mddev
->new_level
= sb
->new_level
;
1192 mddev
->new_layout
= sb
->new_layout
;
1193 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1195 mddev
->reshape_position
= MaxSector
;
1196 mddev
->delta_disks
= 0;
1197 mddev
->new_level
= mddev
->level
;
1198 mddev
->new_layout
= mddev
->layout
;
1199 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1202 if (sb
->state
& (1<<MD_SB_CLEAN
))
1203 mddev
->recovery_cp
= MaxSector
;
1205 if (sb
->events_hi
== sb
->cp_events_hi
&&
1206 sb
->events_lo
== sb
->cp_events_lo
) {
1207 mddev
->recovery_cp
= sb
->recovery_cp
;
1209 mddev
->recovery_cp
= 0;
1212 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1213 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1214 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1215 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1217 mddev
->max_disks
= MD_SB_DISKS
;
1219 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1220 mddev
->bitmap_info
.file
== NULL
)
1221 mddev
->bitmap_info
.offset
=
1222 mddev
->bitmap_info
.default_offset
;
1224 } else if (mddev
->pers
== NULL
) {
1225 /* Insist on good event counter while assembling, except
1226 * for spares (which don't need an event count) */
1228 if (sb
->disks
[rdev
->desc_nr
].state
& (
1229 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1230 if (ev1
< mddev
->events
)
1232 } else if (mddev
->bitmap
) {
1233 /* if adding to array with a bitmap, then we can accept an
1234 * older device ... but not too old.
1236 if (ev1
< mddev
->bitmap
->events_cleared
)
1239 if (ev1
< mddev
->events
)
1240 /* just a hot-add of a new device, leave raid_disk at -1 */
1244 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1245 desc
= sb
->disks
+ rdev
->desc_nr
;
1247 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1248 set_bit(Faulty
, &rdev
->flags
);
1249 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1250 desc->raid_disk < mddev->raid_disks */) {
1251 set_bit(In_sync
, &rdev
->flags
);
1252 rdev
->raid_disk
= desc
->raid_disk
;
1253 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1254 /* active but not in sync implies recovery up to
1255 * reshape position. We don't know exactly where
1256 * that is, so set to zero for now */
1257 if (mddev
->minor_version
>= 91) {
1258 rdev
->recovery_offset
= 0;
1259 rdev
->raid_disk
= desc
->raid_disk
;
1262 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1263 set_bit(WriteMostly
, &rdev
->flags
);
1264 } else /* MULTIPATH are always insync */
1265 set_bit(In_sync
, &rdev
->flags
);
1270 * sync_super for 0.90.0
1272 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1275 struct md_rdev
*rdev2
;
1276 int next_spare
= mddev
->raid_disks
;
1279 /* make rdev->sb match mddev data..
1282 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1283 * 3/ any empty disks < next_spare become removed
1285 * disks[0] gets initialised to REMOVED because
1286 * we cannot be sure from other fields if it has
1287 * been initialised or not.
1290 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1292 rdev
->sb_size
= MD_SB_BYTES
;
1294 sb
= page_address(rdev
->sb_page
);
1296 memset(sb
, 0, sizeof(*sb
));
1298 sb
->md_magic
= MD_SB_MAGIC
;
1299 sb
->major_version
= mddev
->major_version
;
1300 sb
->patch_version
= mddev
->patch_version
;
1301 sb
->gvalid_words
= 0; /* ignored */
1302 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1303 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1304 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1305 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1307 sb
->ctime
= mddev
->ctime
;
1308 sb
->level
= mddev
->level
;
1309 sb
->size
= mddev
->dev_sectors
/ 2;
1310 sb
->raid_disks
= mddev
->raid_disks
;
1311 sb
->md_minor
= mddev
->md_minor
;
1312 sb
->not_persistent
= 0;
1313 sb
->utime
= mddev
->utime
;
1315 sb
->events_hi
= (mddev
->events
>>32);
1316 sb
->events_lo
= (u32
)mddev
->events
;
1318 if (mddev
->reshape_position
== MaxSector
)
1319 sb
->minor_version
= 90;
1321 sb
->minor_version
= 91;
1322 sb
->reshape_position
= mddev
->reshape_position
;
1323 sb
->new_level
= mddev
->new_level
;
1324 sb
->delta_disks
= mddev
->delta_disks
;
1325 sb
->new_layout
= mddev
->new_layout
;
1326 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1328 mddev
->minor_version
= sb
->minor_version
;
1331 sb
->recovery_cp
= mddev
->recovery_cp
;
1332 sb
->cp_events_hi
= (mddev
->events
>>32);
1333 sb
->cp_events_lo
= (u32
)mddev
->events
;
1334 if (mddev
->recovery_cp
== MaxSector
)
1335 sb
->state
= (1<< MD_SB_CLEAN
);
1337 sb
->recovery_cp
= 0;
1339 sb
->layout
= mddev
->layout
;
1340 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1342 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1343 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1345 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1346 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1349 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1351 if (rdev2
->raid_disk
>= 0 &&
1352 sb
->minor_version
>= 91)
1353 /* we have nowhere to store the recovery_offset,
1354 * but if it is not below the reshape_position,
1355 * we can piggy-back on that.
1358 if (rdev2
->raid_disk
< 0 ||
1359 test_bit(Faulty
, &rdev2
->flags
))
1362 desc_nr
= rdev2
->raid_disk
;
1364 desc_nr
= next_spare
++;
1365 rdev2
->desc_nr
= desc_nr
;
1366 d
= &sb
->disks
[rdev2
->desc_nr
];
1368 d
->number
= rdev2
->desc_nr
;
1369 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1370 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1372 d
->raid_disk
= rdev2
->raid_disk
;
1374 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1375 if (test_bit(Faulty
, &rdev2
->flags
))
1376 d
->state
= (1<<MD_DISK_FAULTY
);
1377 else if (is_active
) {
1378 d
->state
= (1<<MD_DISK_ACTIVE
);
1379 if (test_bit(In_sync
, &rdev2
->flags
))
1380 d
->state
|= (1<<MD_DISK_SYNC
);
1388 if (test_bit(WriteMostly
, &rdev2
->flags
))
1389 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1391 /* now set the "removed" and "faulty" bits on any missing devices */
1392 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1393 mdp_disk_t
*d
= &sb
->disks
[i
];
1394 if (d
->state
== 0 && d
->number
== 0) {
1397 d
->state
= (1<<MD_DISK_REMOVED
);
1398 d
->state
|= (1<<MD_DISK_FAULTY
);
1402 sb
->nr_disks
= nr_disks
;
1403 sb
->active_disks
= active
;
1404 sb
->working_disks
= working
;
1405 sb
->failed_disks
= failed
;
1406 sb
->spare_disks
= spare
;
1408 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1409 sb
->sb_csum
= calc_sb_csum(sb
);
1413 * rdev_size_change for 0.90.0
1415 static unsigned long long
1416 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1418 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1419 return 0; /* component must fit device */
1420 if (rdev
->mddev
->bitmap_info
.offset
)
1421 return 0; /* can't move bitmap */
1422 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1423 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1424 num_sectors
= rdev
->sb_start
;
1425 /* Limit to 4TB as metadata cannot record more than that.
1426 * 4TB == 2^32 KB, or 2*2^32 sectors.
1428 if (num_sectors
>= (2ULL << 32))
1429 num_sectors
= (2ULL << 32) - 2;
1430 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1432 md_super_wait(rdev
->mddev
);
1438 * version 1 superblock
1441 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1445 unsigned long long newcsum
;
1446 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1447 __le32
*isuper
= (__le32
*)sb
;
1450 disk_csum
= sb
->sb_csum
;
1453 for (i
=0; size
>=4; size
-= 4 )
1454 newcsum
+= le32_to_cpu(*isuper
++);
1457 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1459 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1460 sb
->sb_csum
= disk_csum
;
1461 return cpu_to_le32(csum
);
1464 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1466 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1468 struct mdp_superblock_1
*sb
;
1471 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1475 * Calculate the position of the superblock in 512byte sectors.
1476 * It is always aligned to a 4K boundary and
1477 * depeding on minor_version, it can be:
1478 * 0: At least 8K, but less than 12K, from end of device
1479 * 1: At start of device
1480 * 2: 4K from start of device.
1482 switch(minor_version
) {
1484 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1486 sb_start
&= ~(sector_t
)(4*2-1);
1497 rdev
->sb_start
= sb_start
;
1499 /* superblock is rarely larger than 1K, but it can be larger,
1500 * and it is safe to read 4k, so we do that
1502 ret
= read_disk_sb(rdev
, 4096);
1503 if (ret
) return ret
;
1506 sb
= page_address(rdev
->sb_page
);
1508 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1509 sb
->major_version
!= cpu_to_le32(1) ||
1510 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1511 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1512 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1515 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1516 printk("md: invalid superblock checksum on %s\n",
1517 bdevname(rdev
->bdev
,b
));
1520 if (le64_to_cpu(sb
->data_size
) < 10) {
1521 printk("md: data_size too small on %s\n",
1522 bdevname(rdev
->bdev
,b
));
1526 rdev
->preferred_minor
= 0xffff;
1527 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1528 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1530 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1531 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1532 if (rdev
->sb_size
& bmask
)
1533 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1536 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1539 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1542 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1544 if (!rdev
->bb_page
) {
1545 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1549 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1550 rdev
->badblocks
.count
== 0) {
1551 /* need to load the bad block list.
1552 * Currently we limit it to one page.
1558 int sectors
= le16_to_cpu(sb
->bblog_size
);
1559 if (sectors
> (PAGE_SIZE
/ 512))
1561 offset
= le32_to_cpu(sb
->bblog_offset
);
1564 bb_sector
= (long long)offset
;
1565 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1566 rdev
->bb_page
, READ
, true))
1568 bbp
= (u64
*)page_address(rdev
->bb_page
);
1569 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1570 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1571 u64 bb
= le64_to_cpu(*bbp
);
1572 int count
= bb
& (0x3ff);
1573 u64 sector
= bb
>> 10;
1574 sector
<<= sb
->bblog_shift
;
1575 count
<<= sb
->bblog_shift
;
1578 if (md_set_badblocks(&rdev
->badblocks
,
1579 sector
, count
, 1) == 0)
1582 } else if (sb
->bblog_offset
== 0)
1583 rdev
->badblocks
.shift
= -1;
1589 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1591 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1592 sb
->level
!= refsb
->level
||
1593 sb
->layout
!= refsb
->layout
||
1594 sb
->chunksize
!= refsb
->chunksize
) {
1595 printk(KERN_WARNING
"md: %s has strangely different"
1596 " superblock to %s\n",
1597 bdevname(rdev
->bdev
,b
),
1598 bdevname(refdev
->bdev
,b2
));
1601 ev1
= le64_to_cpu(sb
->events
);
1602 ev2
= le64_to_cpu(refsb
->events
);
1610 rdev
->sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
1611 le64_to_cpu(sb
->data_offset
);
1613 rdev
->sectors
= rdev
->sb_start
;
1614 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1616 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1617 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1622 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1624 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1625 __u64 ev1
= le64_to_cpu(sb
->events
);
1627 rdev
->raid_disk
= -1;
1628 clear_bit(Faulty
, &rdev
->flags
);
1629 clear_bit(In_sync
, &rdev
->flags
);
1630 clear_bit(WriteMostly
, &rdev
->flags
);
1632 if (mddev
->raid_disks
== 0) {
1633 mddev
->major_version
= 1;
1634 mddev
->patch_version
= 0;
1635 mddev
->external
= 0;
1636 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1637 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1638 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1639 mddev
->level
= le32_to_cpu(sb
->level
);
1640 mddev
->clevel
[0] = 0;
1641 mddev
->layout
= le32_to_cpu(sb
->layout
);
1642 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1643 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1644 mddev
->events
= ev1
;
1645 mddev
->bitmap_info
.offset
= 0;
1646 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1648 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1649 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1651 mddev
->max_disks
= (4096-256)/2;
1653 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1654 mddev
->bitmap_info
.file
== NULL
)
1655 mddev
->bitmap_info
.offset
=
1656 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1658 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1659 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1660 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1661 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1662 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1663 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1665 mddev
->reshape_position
= MaxSector
;
1666 mddev
->delta_disks
= 0;
1667 mddev
->new_level
= mddev
->level
;
1668 mddev
->new_layout
= mddev
->layout
;
1669 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1672 } else if (mddev
->pers
== NULL
) {
1673 /* Insist of good event counter while assembling, except for
1674 * spares (which don't need an event count) */
1676 if (rdev
->desc_nr
>= 0 &&
1677 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1678 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1679 if (ev1
< mddev
->events
)
1681 } else if (mddev
->bitmap
) {
1682 /* If adding to array with a bitmap, then we can accept an
1683 * older device, but not too old.
1685 if (ev1
< mddev
->bitmap
->events_cleared
)
1688 if (ev1
< mddev
->events
)
1689 /* just a hot-add of a new device, leave raid_disk at -1 */
1692 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1694 if (rdev
->desc_nr
< 0 ||
1695 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1699 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1701 case 0xffff: /* spare */
1703 case 0xfffe: /* faulty */
1704 set_bit(Faulty
, &rdev
->flags
);
1707 if ((le32_to_cpu(sb
->feature_map
) &
1708 MD_FEATURE_RECOVERY_OFFSET
))
1709 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1711 set_bit(In_sync
, &rdev
->flags
);
1712 rdev
->raid_disk
= role
;
1715 if (sb
->devflags
& WriteMostly1
)
1716 set_bit(WriteMostly
, &rdev
->flags
);
1717 } else /* MULTIPATH are always insync */
1718 set_bit(In_sync
, &rdev
->flags
);
1723 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1725 struct mdp_superblock_1
*sb
;
1726 struct md_rdev
*rdev2
;
1728 /* make rdev->sb match mddev and rdev data. */
1730 sb
= page_address(rdev
->sb_page
);
1732 sb
->feature_map
= 0;
1734 sb
->recovery_offset
= cpu_to_le64(0);
1735 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1736 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1738 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1739 sb
->events
= cpu_to_le64(mddev
->events
);
1741 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1743 sb
->resync_offset
= cpu_to_le64(0);
1745 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1747 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1748 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1749 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1750 sb
->level
= cpu_to_le32(mddev
->level
);
1751 sb
->layout
= cpu_to_le32(mddev
->layout
);
1753 if (test_bit(WriteMostly
, &rdev
->flags
))
1754 sb
->devflags
|= WriteMostly1
;
1756 sb
->devflags
&= ~WriteMostly1
;
1758 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1759 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1760 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1763 if (rdev
->raid_disk
>= 0 &&
1764 !test_bit(In_sync
, &rdev
->flags
)) {
1766 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1767 sb
->recovery_offset
=
1768 cpu_to_le64(rdev
->recovery_offset
);
1771 if (mddev
->reshape_position
!= MaxSector
) {
1772 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1773 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1774 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1775 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1776 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1777 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1780 if (rdev
->badblocks
.count
== 0)
1781 /* Nothing to do for bad blocks*/ ;
1782 else if (sb
->bblog_offset
== 0)
1783 /* Cannot record bad blocks on this device */
1784 md_error(mddev
, rdev
);
1786 struct badblocks
*bb
= &rdev
->badblocks
;
1787 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1789 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1794 seq
= read_seqbegin(&bb
->lock
);
1796 memset(bbp
, 0xff, PAGE_SIZE
);
1798 for (i
= 0 ; i
< bb
->count
; i
++) {
1799 u64 internal_bb
= *p
++;
1800 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1801 | BB_LEN(internal_bb
));
1802 *bbp
++ = cpu_to_le64(store_bb
);
1804 if (read_seqretry(&bb
->lock
, seq
))
1807 bb
->sector
= (rdev
->sb_start
+
1808 (int)le32_to_cpu(sb
->bblog_offset
));
1809 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1815 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1816 if (rdev2
->desc_nr
+1 > max_dev
)
1817 max_dev
= rdev2
->desc_nr
+1;
1819 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1821 sb
->max_dev
= cpu_to_le32(max_dev
);
1822 rdev
->sb_size
= max_dev
* 2 + 256;
1823 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1824 if (rdev
->sb_size
& bmask
)
1825 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1827 max_dev
= le32_to_cpu(sb
->max_dev
);
1829 for (i
=0; i
<max_dev
;i
++)
1830 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1832 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1834 if (test_bit(Faulty
, &rdev2
->flags
))
1835 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1836 else if (test_bit(In_sync
, &rdev2
->flags
))
1837 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1838 else if (rdev2
->raid_disk
>= 0)
1839 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1841 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1844 sb
->sb_csum
= calc_sb_1_csum(sb
);
1847 static unsigned long long
1848 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1850 struct mdp_superblock_1
*sb
;
1851 sector_t max_sectors
;
1852 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1853 return 0; /* component must fit device */
1854 if (rdev
->sb_start
< rdev
->data_offset
) {
1855 /* minor versions 1 and 2; superblock before data */
1856 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1857 max_sectors
-= rdev
->data_offset
;
1858 if (!num_sectors
|| num_sectors
> max_sectors
)
1859 num_sectors
= max_sectors
;
1860 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1861 /* minor version 0 with bitmap we can't move */
1864 /* minor version 0; superblock after data */
1866 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1867 sb_start
&= ~(sector_t
)(4*2 - 1);
1868 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1869 if (!num_sectors
|| num_sectors
> max_sectors
)
1870 num_sectors
= max_sectors
;
1871 rdev
->sb_start
= sb_start
;
1873 sb
= page_address(rdev
->sb_page
);
1874 sb
->data_size
= cpu_to_le64(num_sectors
);
1875 sb
->super_offset
= rdev
->sb_start
;
1876 sb
->sb_csum
= calc_sb_1_csum(sb
);
1877 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1879 md_super_wait(rdev
->mddev
);
1883 static struct super_type super_types
[] = {
1886 .owner
= THIS_MODULE
,
1887 .load_super
= super_90_load
,
1888 .validate_super
= super_90_validate
,
1889 .sync_super
= super_90_sync
,
1890 .rdev_size_change
= super_90_rdev_size_change
,
1894 .owner
= THIS_MODULE
,
1895 .load_super
= super_1_load
,
1896 .validate_super
= super_1_validate
,
1897 .sync_super
= super_1_sync
,
1898 .rdev_size_change
= super_1_rdev_size_change
,
1902 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1904 if (mddev
->sync_super
) {
1905 mddev
->sync_super(mddev
, rdev
);
1909 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1911 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1914 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1916 struct md_rdev
*rdev
, *rdev2
;
1919 rdev_for_each_rcu(rdev
, mddev1
)
1920 rdev_for_each_rcu(rdev2
, mddev2
)
1921 if (rdev
->bdev
->bd_contains
==
1922 rdev2
->bdev
->bd_contains
) {
1930 static LIST_HEAD(pending_raid_disks
);
1933 * Try to register data integrity profile for an mddev
1935 * This is called when an array is started and after a disk has been kicked
1936 * from the array. It only succeeds if all working and active component devices
1937 * are integrity capable with matching profiles.
1939 int md_integrity_register(struct mddev
*mddev
)
1941 struct md_rdev
*rdev
, *reference
= NULL
;
1943 if (list_empty(&mddev
->disks
))
1944 return 0; /* nothing to do */
1945 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1946 return 0; /* shouldn't register, or already is */
1947 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1948 /* skip spares and non-functional disks */
1949 if (test_bit(Faulty
, &rdev
->flags
))
1951 if (rdev
->raid_disk
< 0)
1954 /* Use the first rdev as the reference */
1958 /* does this rdev's profile match the reference profile? */
1959 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1960 rdev
->bdev
->bd_disk
) < 0)
1963 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1966 * All component devices are integrity capable and have matching
1967 * profiles, register the common profile for the md device.
1969 if (blk_integrity_register(mddev
->gendisk
,
1970 bdev_get_integrity(reference
->bdev
)) != 0) {
1971 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1975 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1976 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1977 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1983 EXPORT_SYMBOL(md_integrity_register
);
1985 /* Disable data integrity if non-capable/non-matching disk is being added */
1986 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
1988 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1989 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1991 if (!bi_mddev
) /* nothing to do */
1993 if (rdev
->raid_disk
< 0) /* skip spares */
1995 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1996 rdev
->bdev
->bd_disk
) >= 0)
1998 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1999 blk_integrity_unregister(mddev
->gendisk
);
2001 EXPORT_SYMBOL(md_integrity_add_rdev
);
2003 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2005 char b
[BDEVNAME_SIZE
];
2015 /* prevent duplicates */
2016 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2019 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2020 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2021 rdev
->sectors
< mddev
->dev_sectors
)) {
2023 /* Cannot change size, so fail
2024 * If mddev->level <= 0, then we don't care
2025 * about aligning sizes (e.g. linear)
2027 if (mddev
->level
> 0)
2030 mddev
->dev_sectors
= rdev
->sectors
;
2033 /* Verify rdev->desc_nr is unique.
2034 * If it is -1, assign a free number, else
2035 * check number is not in use
2037 if (rdev
->desc_nr
< 0) {
2039 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2040 while (find_rdev_nr(mddev
, choice
))
2042 rdev
->desc_nr
= choice
;
2044 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2047 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2048 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2049 mdname(mddev
), mddev
->max_disks
);
2052 bdevname(rdev
->bdev
,b
);
2053 while ( (s
=strchr(b
, '/')) != NULL
)
2056 rdev
->mddev
= mddev
;
2057 printk(KERN_INFO
"md: bind<%s>\n", b
);
2059 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2062 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2063 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2064 /* failure here is OK */;
2065 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2067 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2068 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2070 /* May as well allow recovery to be retried once */
2071 mddev
->recovery_disabled
++;
2076 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2081 static void md_delayed_delete(struct work_struct
*ws
)
2083 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2084 kobject_del(&rdev
->kobj
);
2085 kobject_put(&rdev
->kobj
);
2088 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2090 char b
[BDEVNAME_SIZE
];
2095 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2096 list_del_rcu(&rdev
->same_set
);
2097 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2099 sysfs_remove_link(&rdev
->kobj
, "block");
2100 sysfs_put(rdev
->sysfs_state
);
2101 rdev
->sysfs_state
= NULL
;
2102 kfree(rdev
->badblocks
.page
);
2103 rdev
->badblocks
.count
= 0;
2104 rdev
->badblocks
.page
= NULL
;
2105 /* We need to delay this, otherwise we can deadlock when
2106 * writing to 'remove' to "dev/state". We also need
2107 * to delay it due to rcu usage.
2110 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2111 kobject_get(&rdev
->kobj
);
2112 queue_work(md_misc_wq
, &rdev
->del_work
);
2116 * prevent the device from being mounted, repartitioned or
2117 * otherwise reused by a RAID array (or any other kernel
2118 * subsystem), by bd_claiming the device.
2120 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2123 struct block_device
*bdev
;
2124 char b
[BDEVNAME_SIZE
];
2126 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2127 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2129 printk(KERN_ERR
"md: could not open %s.\n",
2130 __bdevname(dev
, b
));
2131 return PTR_ERR(bdev
);
2137 static void unlock_rdev(struct md_rdev
*rdev
)
2139 struct block_device
*bdev
= rdev
->bdev
;
2143 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2146 void md_autodetect_dev(dev_t dev
);
2148 static void export_rdev(struct md_rdev
* rdev
)
2150 char b
[BDEVNAME_SIZE
];
2151 printk(KERN_INFO
"md: export_rdev(%s)\n",
2152 bdevname(rdev
->bdev
,b
));
2157 if (test_bit(AutoDetected
, &rdev
->flags
))
2158 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2161 kobject_put(&rdev
->kobj
);
2164 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2166 unbind_rdev_from_array(rdev
);
2170 static void export_array(struct mddev
*mddev
)
2172 struct md_rdev
*rdev
, *tmp
;
2174 rdev_for_each(rdev
, tmp
, mddev
) {
2179 kick_rdev_from_array(rdev
);
2181 if (!list_empty(&mddev
->disks
))
2183 mddev
->raid_disks
= 0;
2184 mddev
->major_version
= 0;
2187 static void print_desc(mdp_disk_t
*desc
)
2189 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2190 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2193 static void print_sb_90(mdp_super_t
*sb
)
2198 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2199 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2200 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2202 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2203 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2204 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2205 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2206 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2207 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2208 sb
->failed_disks
, sb
->spare_disks
,
2209 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2212 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2215 desc
= sb
->disks
+ i
;
2216 if (desc
->number
|| desc
->major
|| desc
->minor
||
2217 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2218 printk(" D %2d: ", i
);
2222 printk(KERN_INFO
"md: THIS: ");
2223 print_desc(&sb
->this_disk
);
2226 static void print_sb_1(struct mdp_superblock_1
*sb
)
2230 uuid
= sb
->set_uuid
;
2232 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2233 "md: Name: \"%s\" CT:%llu\n",
2234 le32_to_cpu(sb
->major_version
),
2235 le32_to_cpu(sb
->feature_map
),
2238 (unsigned long long)le64_to_cpu(sb
->ctime
)
2239 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2241 uuid
= sb
->device_uuid
;
2243 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2245 "md: Dev:%08x UUID: %pU\n"
2246 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2247 "md: (MaxDev:%u) \n",
2248 le32_to_cpu(sb
->level
),
2249 (unsigned long long)le64_to_cpu(sb
->size
),
2250 le32_to_cpu(sb
->raid_disks
),
2251 le32_to_cpu(sb
->layout
),
2252 le32_to_cpu(sb
->chunksize
),
2253 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2254 (unsigned long long)le64_to_cpu(sb
->data_size
),
2255 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2256 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2257 le32_to_cpu(sb
->dev_number
),
2260 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2261 (unsigned long long)le64_to_cpu(sb
->events
),
2262 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2263 le32_to_cpu(sb
->sb_csum
),
2264 le32_to_cpu(sb
->max_dev
)
2268 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2270 char b
[BDEVNAME_SIZE
];
2271 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2272 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2273 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2275 if (rdev
->sb_loaded
) {
2276 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2277 switch (major_version
) {
2279 print_sb_90(page_address(rdev
->sb_page
));
2282 print_sb_1(page_address(rdev
->sb_page
));
2286 printk(KERN_INFO
"md: no rdev superblock!\n");
2289 static void md_print_devices(void)
2291 struct list_head
*tmp
;
2292 struct md_rdev
*rdev
;
2293 struct mddev
*mddev
;
2294 char b
[BDEVNAME_SIZE
];
2297 printk("md: **********************************\n");
2298 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2299 printk("md: **********************************\n");
2300 for_each_mddev(mddev
, tmp
) {
2303 bitmap_print_sb(mddev
->bitmap
);
2305 printk("%s: ", mdname(mddev
));
2306 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2307 printk("<%s>", bdevname(rdev
->bdev
,b
));
2310 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2311 print_rdev(rdev
, mddev
->major_version
);
2313 printk("md: **********************************\n");
2318 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2320 /* Update each superblock (in-memory image), but
2321 * if we are allowed to, skip spares which already
2322 * have the right event counter, or have one earlier
2323 * (which would mean they aren't being marked as dirty
2324 * with the rest of the array)
2326 struct md_rdev
*rdev
;
2327 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2328 if (rdev
->sb_events
== mddev
->events
||
2330 rdev
->raid_disk
< 0 &&
2331 rdev
->sb_events
+1 == mddev
->events
)) {
2332 /* Don't update this superblock */
2333 rdev
->sb_loaded
= 2;
2335 sync_super(mddev
, rdev
);
2336 rdev
->sb_loaded
= 1;
2341 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2343 struct md_rdev
*rdev
;
2346 int any_badblocks_changed
= 0;
2349 /* First make sure individual recovery_offsets are correct */
2350 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2351 if (rdev
->raid_disk
>= 0 &&
2352 mddev
->delta_disks
>= 0 &&
2353 !test_bit(In_sync
, &rdev
->flags
) &&
2354 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2355 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2358 if (!mddev
->persistent
) {
2359 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2360 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2361 if (!mddev
->external
) {
2362 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2363 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2364 if (rdev
->badblocks
.changed
) {
2365 md_ack_all_badblocks(&rdev
->badblocks
);
2366 md_error(mddev
, rdev
);
2368 clear_bit(Blocked
, &rdev
->flags
);
2369 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2370 wake_up(&rdev
->blocked_wait
);
2373 wake_up(&mddev
->sb_wait
);
2377 spin_lock_irq(&mddev
->write_lock
);
2379 mddev
->utime
= get_seconds();
2381 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2383 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2384 /* just a clean<-> dirty transition, possibly leave spares alone,
2385 * though if events isn't the right even/odd, we will have to do
2391 if (mddev
->degraded
)
2392 /* If the array is degraded, then skipping spares is both
2393 * dangerous and fairly pointless.
2394 * Dangerous because a device that was removed from the array
2395 * might have a event_count that still looks up-to-date,
2396 * so it can be re-added without a resync.
2397 * Pointless because if there are any spares to skip,
2398 * then a recovery will happen and soon that array won't
2399 * be degraded any more and the spare can go back to sleep then.
2403 sync_req
= mddev
->in_sync
;
2405 /* If this is just a dirty<->clean transition, and the array is clean
2406 * and 'events' is odd, we can roll back to the previous clean state */
2408 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2409 && mddev
->can_decrease_events
2410 && mddev
->events
!= 1) {
2412 mddev
->can_decrease_events
= 0;
2414 /* otherwise we have to go forward and ... */
2416 mddev
->can_decrease_events
= nospares
;
2419 if (!mddev
->events
) {
2421 * oops, this 64-bit counter should never wrap.
2422 * Either we are in around ~1 trillion A.C., assuming
2423 * 1 reboot per second, or we have a bug:
2429 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2430 if (rdev
->badblocks
.changed
)
2431 any_badblocks_changed
++;
2432 if (test_bit(Faulty
, &rdev
->flags
))
2433 set_bit(FaultRecorded
, &rdev
->flags
);
2436 sync_sbs(mddev
, nospares
);
2437 spin_unlock_irq(&mddev
->write_lock
);
2439 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2440 mdname(mddev
), mddev
->in_sync
);
2442 bitmap_update_sb(mddev
->bitmap
);
2443 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2444 char b
[BDEVNAME_SIZE
];
2446 if (rdev
->sb_loaded
!= 1)
2447 continue; /* no noise on spare devices */
2449 if (!test_bit(Faulty
, &rdev
->flags
) &&
2450 rdev
->saved_raid_disk
== -1) {
2451 md_super_write(mddev
,rdev
,
2452 rdev
->sb_start
, rdev
->sb_size
,
2454 pr_debug("md: (write) %s's sb offset: %llu\n",
2455 bdevname(rdev
->bdev
, b
),
2456 (unsigned long long)rdev
->sb_start
);
2457 rdev
->sb_events
= mddev
->events
;
2458 if (rdev
->badblocks
.size
) {
2459 md_super_write(mddev
, rdev
,
2460 rdev
->badblocks
.sector
,
2461 rdev
->badblocks
.size
<< 9,
2463 rdev
->badblocks
.size
= 0;
2466 } else if (test_bit(Faulty
, &rdev
->flags
))
2467 pr_debug("md: %s (skipping faulty)\n",
2468 bdevname(rdev
->bdev
, b
));
2470 pr_debug("(skipping incremental s/r ");
2472 if (mddev
->level
== LEVEL_MULTIPATH
)
2473 /* only need to write one superblock... */
2476 md_super_wait(mddev
);
2477 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2479 spin_lock_irq(&mddev
->write_lock
);
2480 if (mddev
->in_sync
!= sync_req
||
2481 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2482 /* have to write it out again */
2483 spin_unlock_irq(&mddev
->write_lock
);
2486 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2487 spin_unlock_irq(&mddev
->write_lock
);
2488 wake_up(&mddev
->sb_wait
);
2489 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2490 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2492 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2493 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2494 clear_bit(Blocked
, &rdev
->flags
);
2496 if (any_badblocks_changed
)
2497 md_ack_all_badblocks(&rdev
->badblocks
);
2498 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2499 wake_up(&rdev
->blocked_wait
);
2503 /* words written to sysfs files may, or may not, be \n terminated.
2504 * We want to accept with case. For this we use cmd_match.
2506 static int cmd_match(const char *cmd
, const char *str
)
2508 /* See if cmd, written into a sysfs file, matches
2509 * str. They must either be the same, or cmd can
2510 * have a trailing newline
2512 while (*cmd
&& *str
&& *cmd
== *str
) {
2523 struct rdev_sysfs_entry
{
2524 struct attribute attr
;
2525 ssize_t (*show
)(struct md_rdev
*, char *);
2526 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2530 state_show(struct md_rdev
*rdev
, char *page
)
2535 if (test_bit(Faulty
, &rdev
->flags
) ||
2536 rdev
->badblocks
.unacked_exist
) {
2537 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2540 if (test_bit(In_sync
, &rdev
->flags
)) {
2541 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2544 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2545 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2548 if (test_bit(Blocked
, &rdev
->flags
) ||
2549 rdev
->badblocks
.unacked_exist
) {
2550 len
+= sprintf(page
+len
, "%sblocked", sep
);
2553 if (!test_bit(Faulty
, &rdev
->flags
) &&
2554 !test_bit(In_sync
, &rdev
->flags
)) {
2555 len
+= sprintf(page
+len
, "%sspare", sep
);
2558 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2559 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2562 return len
+sprintf(page
+len
, "\n");
2566 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2569 * faulty - simulates an error
2570 * remove - disconnects the device
2571 * writemostly - sets write_mostly
2572 * -writemostly - clears write_mostly
2573 * blocked - sets the Blocked flags
2574 * -blocked - clears the Blocked and possibly simulates an error
2575 * insync - sets Insync providing device isn't active
2576 * write_error - sets WriteErrorSeen
2577 * -write_error - clears WriteErrorSeen
2580 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2581 md_error(rdev
->mddev
, rdev
);
2582 if (test_bit(Faulty
, &rdev
->flags
))
2586 } else if (cmd_match(buf
, "remove")) {
2587 if (rdev
->raid_disk
>= 0)
2590 struct mddev
*mddev
= rdev
->mddev
;
2591 kick_rdev_from_array(rdev
);
2593 md_update_sb(mddev
, 1);
2594 md_new_event(mddev
);
2597 } else if (cmd_match(buf
, "writemostly")) {
2598 set_bit(WriteMostly
, &rdev
->flags
);
2600 } else if (cmd_match(buf
, "-writemostly")) {
2601 clear_bit(WriteMostly
, &rdev
->flags
);
2603 } else if (cmd_match(buf
, "blocked")) {
2604 set_bit(Blocked
, &rdev
->flags
);
2606 } else if (cmd_match(buf
, "-blocked")) {
2607 if (!test_bit(Faulty
, &rdev
->flags
) &&
2608 rdev
->badblocks
.unacked_exist
) {
2609 /* metadata handler doesn't understand badblocks,
2610 * so we need to fail the device
2612 md_error(rdev
->mddev
, rdev
);
2614 clear_bit(Blocked
, &rdev
->flags
);
2615 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2616 wake_up(&rdev
->blocked_wait
);
2617 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2618 md_wakeup_thread(rdev
->mddev
->thread
);
2621 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2622 set_bit(In_sync
, &rdev
->flags
);
2624 } else if (cmd_match(buf
, "write_error")) {
2625 set_bit(WriteErrorSeen
, &rdev
->flags
);
2627 } else if (cmd_match(buf
, "-write_error")) {
2628 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2632 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2633 return err
? err
: len
;
2635 static struct rdev_sysfs_entry rdev_state
=
2636 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2639 errors_show(struct md_rdev
*rdev
, char *page
)
2641 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2645 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2648 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2649 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2650 atomic_set(&rdev
->corrected_errors
, n
);
2655 static struct rdev_sysfs_entry rdev_errors
=
2656 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2659 slot_show(struct md_rdev
*rdev
, char *page
)
2661 if (rdev
->raid_disk
< 0)
2662 return sprintf(page
, "none\n");
2664 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2668 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2672 int slot
= simple_strtoul(buf
, &e
, 10);
2673 if (strncmp(buf
, "none", 4)==0)
2675 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2677 if (rdev
->mddev
->pers
&& slot
== -1) {
2678 /* Setting 'slot' on an active array requires also
2679 * updating the 'rd%d' link, and communicating
2680 * with the personality with ->hot_*_disk.
2681 * For now we only support removing
2682 * failed/spare devices. This normally happens automatically,
2683 * but not when the metadata is externally managed.
2685 if (rdev
->raid_disk
== -1)
2687 /* personality does all needed checks */
2688 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2690 err
= rdev
->mddev
->pers
->
2691 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2694 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2695 rdev
->raid_disk
= -1;
2696 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2697 md_wakeup_thread(rdev
->mddev
->thread
);
2698 } else if (rdev
->mddev
->pers
) {
2699 struct md_rdev
*rdev2
;
2700 /* Activating a spare .. or possibly reactivating
2701 * if we ever get bitmaps working here.
2704 if (rdev
->raid_disk
!= -1)
2707 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2710 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2713 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2714 if (rdev2
->raid_disk
== slot
)
2717 if (slot
>= rdev
->mddev
->raid_disks
&&
2718 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2721 rdev
->raid_disk
= slot
;
2722 if (test_bit(In_sync
, &rdev
->flags
))
2723 rdev
->saved_raid_disk
= slot
;
2725 rdev
->saved_raid_disk
= -1;
2726 clear_bit(In_sync
, &rdev
->flags
);
2727 err
= rdev
->mddev
->pers
->
2728 hot_add_disk(rdev
->mddev
, rdev
);
2730 rdev
->raid_disk
= -1;
2733 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2734 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2735 /* failure here is OK */;
2736 /* don't wakeup anyone, leave that to userspace. */
2738 if (slot
>= rdev
->mddev
->raid_disks
&&
2739 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2741 rdev
->raid_disk
= slot
;
2742 /* assume it is working */
2743 clear_bit(Faulty
, &rdev
->flags
);
2744 clear_bit(WriteMostly
, &rdev
->flags
);
2745 set_bit(In_sync
, &rdev
->flags
);
2746 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2752 static struct rdev_sysfs_entry rdev_slot
=
2753 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2756 offset_show(struct md_rdev
*rdev
, char *page
)
2758 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2762 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2765 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2766 if (e
==buf
|| (*e
&& *e
!= '\n'))
2768 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2770 if (rdev
->sectors
&& rdev
->mddev
->external
)
2771 /* Must set offset before size, so overlap checks
2774 rdev
->data_offset
= offset
;
2778 static struct rdev_sysfs_entry rdev_offset
=
2779 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2782 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2784 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2787 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2789 /* check if two start/length pairs overlap */
2797 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2799 unsigned long long blocks
;
2802 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2805 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2806 return -EINVAL
; /* sector conversion overflow */
2809 if (new != blocks
* 2)
2810 return -EINVAL
; /* unsigned long long to sector_t overflow */
2817 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2819 struct mddev
*my_mddev
= rdev
->mddev
;
2820 sector_t oldsectors
= rdev
->sectors
;
2823 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2825 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2826 if (my_mddev
->persistent
) {
2827 sectors
= super_types
[my_mddev
->major_version
].
2828 rdev_size_change(rdev
, sectors
);
2831 } else if (!sectors
)
2832 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2835 if (sectors
< my_mddev
->dev_sectors
)
2836 return -EINVAL
; /* component must fit device */
2838 rdev
->sectors
= sectors
;
2839 if (sectors
> oldsectors
&& my_mddev
->external
) {
2840 /* need to check that all other rdevs with the same ->bdev
2841 * do not overlap. We need to unlock the mddev to avoid
2842 * a deadlock. We have already changed rdev->sectors, and if
2843 * we have to change it back, we will have the lock again.
2845 struct mddev
*mddev
;
2847 struct list_head
*tmp
;
2849 mddev_unlock(my_mddev
);
2850 for_each_mddev(mddev
, tmp
) {
2851 struct md_rdev
*rdev2
;
2854 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2855 if (rdev
->bdev
== rdev2
->bdev
&&
2857 overlaps(rdev
->data_offset
, rdev
->sectors
,
2863 mddev_unlock(mddev
);
2869 mddev_lock(my_mddev
);
2871 /* Someone else could have slipped in a size
2872 * change here, but doing so is just silly.
2873 * We put oldsectors back because we *know* it is
2874 * safe, and trust userspace not to race with
2877 rdev
->sectors
= oldsectors
;
2884 static struct rdev_sysfs_entry rdev_size
=
2885 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2888 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2890 unsigned long long recovery_start
= rdev
->recovery_offset
;
2892 if (test_bit(In_sync
, &rdev
->flags
) ||
2893 recovery_start
== MaxSector
)
2894 return sprintf(page
, "none\n");
2896 return sprintf(page
, "%llu\n", recovery_start
);
2899 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2901 unsigned long long recovery_start
;
2903 if (cmd_match(buf
, "none"))
2904 recovery_start
= MaxSector
;
2905 else if (strict_strtoull(buf
, 10, &recovery_start
))
2908 if (rdev
->mddev
->pers
&&
2909 rdev
->raid_disk
>= 0)
2912 rdev
->recovery_offset
= recovery_start
;
2913 if (recovery_start
== MaxSector
)
2914 set_bit(In_sync
, &rdev
->flags
);
2916 clear_bit(In_sync
, &rdev
->flags
);
2920 static struct rdev_sysfs_entry rdev_recovery_start
=
2921 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2925 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2927 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2929 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2931 return badblocks_show(&rdev
->badblocks
, page
, 0);
2933 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2935 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2936 /* Maybe that ack was all we needed */
2937 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2938 wake_up(&rdev
->blocked_wait
);
2941 static struct rdev_sysfs_entry rdev_bad_blocks
=
2942 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2945 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2947 return badblocks_show(&rdev
->badblocks
, page
, 1);
2949 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2951 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2953 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2954 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2956 static struct attribute
*rdev_default_attrs
[] = {
2962 &rdev_recovery_start
.attr
,
2963 &rdev_bad_blocks
.attr
,
2964 &rdev_unack_bad_blocks
.attr
,
2968 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2970 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2971 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2972 struct mddev
*mddev
= rdev
->mddev
;
2978 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2980 if (rdev
->mddev
== NULL
)
2983 rv
= entry
->show(rdev
, page
);
2984 mddev_unlock(mddev
);
2990 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2991 const char *page
, size_t length
)
2993 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2994 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2996 struct mddev
*mddev
= rdev
->mddev
;
3000 if (!capable(CAP_SYS_ADMIN
))
3002 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3004 if (rdev
->mddev
== NULL
)
3007 rv
= entry
->store(rdev
, page
, length
);
3008 mddev_unlock(mddev
);
3013 static void rdev_free(struct kobject
*ko
)
3015 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3018 static const struct sysfs_ops rdev_sysfs_ops
= {
3019 .show
= rdev_attr_show
,
3020 .store
= rdev_attr_store
,
3022 static struct kobj_type rdev_ktype
= {
3023 .release
= rdev_free
,
3024 .sysfs_ops
= &rdev_sysfs_ops
,
3025 .default_attrs
= rdev_default_attrs
,
3028 int md_rdev_init(struct md_rdev
*rdev
)
3031 rdev
->saved_raid_disk
= -1;
3032 rdev
->raid_disk
= -1;
3034 rdev
->data_offset
= 0;
3035 rdev
->sb_events
= 0;
3036 rdev
->last_read_error
.tv_sec
= 0;
3037 rdev
->last_read_error
.tv_nsec
= 0;
3038 rdev
->sb_loaded
= 0;
3039 rdev
->bb_page
= NULL
;
3040 atomic_set(&rdev
->nr_pending
, 0);
3041 atomic_set(&rdev
->read_errors
, 0);
3042 atomic_set(&rdev
->corrected_errors
, 0);
3044 INIT_LIST_HEAD(&rdev
->same_set
);
3045 init_waitqueue_head(&rdev
->blocked_wait
);
3047 /* Add space to store bad block list.
3048 * This reserves the space even on arrays where it cannot
3049 * be used - I wonder if that matters
3051 rdev
->badblocks
.count
= 0;
3052 rdev
->badblocks
.shift
= 0;
3053 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3054 seqlock_init(&rdev
->badblocks
.lock
);
3055 if (rdev
->badblocks
.page
== NULL
)
3060 EXPORT_SYMBOL_GPL(md_rdev_init
);
3062 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3064 * mark the device faulty if:
3066 * - the device is nonexistent (zero size)
3067 * - the device has no valid superblock
3069 * a faulty rdev _never_ has rdev->sb set.
3071 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3073 char b
[BDEVNAME_SIZE
];
3075 struct md_rdev
*rdev
;
3078 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3080 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3081 return ERR_PTR(-ENOMEM
);
3084 err
= md_rdev_init(rdev
);
3087 err
= alloc_disk_sb(rdev
);
3091 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3095 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3097 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3100 "md: %s has zero or unknown size, marking faulty!\n",
3101 bdevname(rdev
->bdev
,b
));
3106 if (super_format
>= 0) {
3107 err
= super_types
[super_format
].
3108 load_super(rdev
, NULL
, super_minor
);
3109 if (err
== -EINVAL
) {
3111 "md: %s does not have a valid v%d.%d "
3112 "superblock, not importing!\n",
3113 bdevname(rdev
->bdev
,b
),
3114 super_format
, super_minor
);
3119 "md: could not read %s's sb, not importing!\n",
3120 bdevname(rdev
->bdev
,b
));
3124 if (super_format
== -1)
3125 /* hot-add for 0.90, or non-persistent: so no badblocks */
3126 rdev
->badblocks
.shift
= -1;
3134 kfree(rdev
->badblocks
.page
);
3136 return ERR_PTR(err
);
3140 * Check a full RAID array for plausibility
3144 static void analyze_sbs(struct mddev
* mddev
)
3147 struct md_rdev
*rdev
, *freshest
, *tmp
;
3148 char b
[BDEVNAME_SIZE
];
3151 rdev_for_each(rdev
, tmp
, mddev
)
3152 switch (super_types
[mddev
->major_version
].
3153 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3161 "md: fatal superblock inconsistency in %s"
3162 " -- removing from array\n",
3163 bdevname(rdev
->bdev
,b
));
3164 kick_rdev_from_array(rdev
);
3168 super_types
[mddev
->major_version
].
3169 validate_super(mddev
, freshest
);
3172 rdev_for_each(rdev
, tmp
, mddev
) {
3173 if (mddev
->max_disks
&&
3174 (rdev
->desc_nr
>= mddev
->max_disks
||
3175 i
> mddev
->max_disks
)) {
3177 "md: %s: %s: only %d devices permitted\n",
3178 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3180 kick_rdev_from_array(rdev
);
3183 if (rdev
!= freshest
)
3184 if (super_types
[mddev
->major_version
].
3185 validate_super(mddev
, rdev
)) {
3186 printk(KERN_WARNING
"md: kicking non-fresh %s"
3188 bdevname(rdev
->bdev
,b
));
3189 kick_rdev_from_array(rdev
);
3192 if (mddev
->level
== LEVEL_MULTIPATH
) {
3193 rdev
->desc_nr
= i
++;
3194 rdev
->raid_disk
= rdev
->desc_nr
;
3195 set_bit(In_sync
, &rdev
->flags
);
3196 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3197 rdev
->raid_disk
= -1;
3198 clear_bit(In_sync
, &rdev
->flags
);
3203 /* Read a fixed-point number.
3204 * Numbers in sysfs attributes should be in "standard" units where
3205 * possible, so time should be in seconds.
3206 * However we internally use a a much smaller unit such as
3207 * milliseconds or jiffies.
3208 * This function takes a decimal number with a possible fractional
3209 * component, and produces an integer which is the result of
3210 * multiplying that number by 10^'scale'.
3211 * all without any floating-point arithmetic.
3213 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3215 unsigned long result
= 0;
3217 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3220 else if (decimals
< scale
) {
3223 result
= result
* 10 + value
;
3235 while (decimals
< scale
) {
3244 static void md_safemode_timeout(unsigned long data
);
3247 safe_delay_show(struct mddev
*mddev
, char *page
)
3249 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3250 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3253 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3257 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3260 mddev
->safemode_delay
= 0;
3262 unsigned long old_delay
= mddev
->safemode_delay
;
3263 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3264 if (mddev
->safemode_delay
== 0)
3265 mddev
->safemode_delay
= 1;
3266 if (mddev
->safemode_delay
< old_delay
)
3267 md_safemode_timeout((unsigned long)mddev
);
3271 static struct md_sysfs_entry md_safe_delay
=
3272 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3275 level_show(struct mddev
*mddev
, char *page
)
3277 struct md_personality
*p
= mddev
->pers
;
3279 return sprintf(page
, "%s\n", p
->name
);
3280 else if (mddev
->clevel
[0])
3281 return sprintf(page
, "%s\n", mddev
->clevel
);
3282 else if (mddev
->level
!= LEVEL_NONE
)
3283 return sprintf(page
, "%d\n", mddev
->level
);
3289 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3293 struct md_personality
*pers
;
3296 struct md_rdev
*rdev
;
3298 if (mddev
->pers
== NULL
) {
3301 if (len
>= sizeof(mddev
->clevel
))
3303 strncpy(mddev
->clevel
, buf
, len
);
3304 if (mddev
->clevel
[len
-1] == '\n')
3306 mddev
->clevel
[len
] = 0;
3307 mddev
->level
= LEVEL_NONE
;
3311 /* request to change the personality. Need to ensure:
3312 * - array is not engaged in resync/recovery/reshape
3313 * - old personality can be suspended
3314 * - new personality will access other array.
3317 if (mddev
->sync_thread
||
3318 mddev
->reshape_position
!= MaxSector
||
3319 mddev
->sysfs_active
)
3322 if (!mddev
->pers
->quiesce
) {
3323 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3324 mdname(mddev
), mddev
->pers
->name
);
3328 /* Now find the new personality */
3329 if (len
== 0 || len
>= sizeof(clevel
))
3331 strncpy(clevel
, buf
, len
);
3332 if (clevel
[len
-1] == '\n')
3335 if (strict_strtol(clevel
, 10, &level
))
3338 if (request_module("md-%s", clevel
) != 0)
3339 request_module("md-level-%s", clevel
);
3340 spin_lock(&pers_lock
);
3341 pers
= find_pers(level
, clevel
);
3342 if (!pers
|| !try_module_get(pers
->owner
)) {
3343 spin_unlock(&pers_lock
);
3344 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3347 spin_unlock(&pers_lock
);
3349 if (pers
== mddev
->pers
) {
3350 /* Nothing to do! */
3351 module_put(pers
->owner
);
3354 if (!pers
->takeover
) {
3355 module_put(pers
->owner
);
3356 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3357 mdname(mddev
), clevel
);
3361 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3362 rdev
->new_raid_disk
= rdev
->raid_disk
;
3364 /* ->takeover must set new_* and/or delta_disks
3365 * if it succeeds, and may set them when it fails.
3367 priv
= pers
->takeover(mddev
);
3369 mddev
->new_level
= mddev
->level
;
3370 mddev
->new_layout
= mddev
->layout
;
3371 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3372 mddev
->raid_disks
-= mddev
->delta_disks
;
3373 mddev
->delta_disks
= 0;
3374 module_put(pers
->owner
);
3375 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3376 mdname(mddev
), clevel
);
3377 return PTR_ERR(priv
);
3380 /* Looks like we have a winner */
3381 mddev_suspend(mddev
);
3382 mddev
->pers
->stop(mddev
);
3384 if (mddev
->pers
->sync_request
== NULL
&&
3385 pers
->sync_request
!= NULL
) {
3386 /* need to add the md_redundancy_group */
3387 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3389 "md: cannot register extra attributes for %s\n",
3391 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3393 if (mddev
->pers
->sync_request
!= NULL
&&
3394 pers
->sync_request
== NULL
) {
3395 /* need to remove the md_redundancy_group */
3396 if (mddev
->to_remove
== NULL
)
3397 mddev
->to_remove
= &md_redundancy_group
;
3400 if (mddev
->pers
->sync_request
== NULL
&&
3402 /* We are converting from a no-redundancy array
3403 * to a redundancy array and metadata is managed
3404 * externally so we need to be sure that writes
3405 * won't block due to a need to transition
3407 * until external management is started.
3410 mddev
->safemode_delay
= 0;
3411 mddev
->safemode
= 0;
3414 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3415 if (rdev
->raid_disk
< 0)
3417 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3418 rdev
->new_raid_disk
= -1;
3419 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3421 sysfs_unlink_rdev(mddev
, rdev
);
3423 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3424 if (rdev
->raid_disk
< 0)
3426 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3428 rdev
->raid_disk
= rdev
->new_raid_disk
;
3429 if (rdev
->raid_disk
< 0)
3430 clear_bit(In_sync
, &rdev
->flags
);
3432 if (sysfs_link_rdev(mddev
, rdev
))
3433 printk(KERN_WARNING
"md: cannot register rd%d"
3434 " for %s after level change\n",
3435 rdev
->raid_disk
, mdname(mddev
));
3439 module_put(mddev
->pers
->owner
);
3441 mddev
->private = priv
;
3442 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3443 mddev
->level
= mddev
->new_level
;
3444 mddev
->layout
= mddev
->new_layout
;
3445 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3446 mddev
->delta_disks
= 0;
3447 mddev
->degraded
= 0;
3448 if (mddev
->pers
->sync_request
== NULL
) {
3449 /* this is now an array without redundancy, so
3450 * it must always be in_sync
3453 del_timer_sync(&mddev
->safemode_timer
);
3456 mddev_resume(mddev
);
3457 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3458 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3459 md_wakeup_thread(mddev
->thread
);
3460 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3461 md_new_event(mddev
);
3465 static struct md_sysfs_entry md_level
=
3466 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3470 layout_show(struct mddev
*mddev
, char *page
)
3472 /* just a number, not meaningful for all levels */
3473 if (mddev
->reshape_position
!= MaxSector
&&
3474 mddev
->layout
!= mddev
->new_layout
)
3475 return sprintf(page
, "%d (%d)\n",
3476 mddev
->new_layout
, mddev
->layout
);
3477 return sprintf(page
, "%d\n", mddev
->layout
);
3481 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3484 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3486 if (!*buf
|| (*e
&& *e
!= '\n'))
3491 if (mddev
->pers
->check_reshape
== NULL
)
3493 mddev
->new_layout
= n
;
3494 err
= mddev
->pers
->check_reshape(mddev
);
3496 mddev
->new_layout
= mddev
->layout
;
3500 mddev
->new_layout
= n
;
3501 if (mddev
->reshape_position
== MaxSector
)
3506 static struct md_sysfs_entry md_layout
=
3507 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3511 raid_disks_show(struct mddev
*mddev
, char *page
)
3513 if (mddev
->raid_disks
== 0)
3515 if (mddev
->reshape_position
!= MaxSector
&&
3516 mddev
->delta_disks
!= 0)
3517 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3518 mddev
->raid_disks
- mddev
->delta_disks
);
3519 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3522 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3525 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3529 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3531 if (!*buf
|| (*e
&& *e
!= '\n'))
3535 rv
= update_raid_disks(mddev
, n
);
3536 else if (mddev
->reshape_position
!= MaxSector
) {
3537 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3538 mddev
->delta_disks
= n
- olddisks
;
3539 mddev
->raid_disks
= n
;
3541 mddev
->raid_disks
= n
;
3542 return rv
? rv
: len
;
3544 static struct md_sysfs_entry md_raid_disks
=
3545 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3548 chunk_size_show(struct mddev
*mddev
, char *page
)
3550 if (mddev
->reshape_position
!= MaxSector
&&
3551 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3552 return sprintf(page
, "%d (%d)\n",
3553 mddev
->new_chunk_sectors
<< 9,
3554 mddev
->chunk_sectors
<< 9);
3555 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3559 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3562 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3564 if (!*buf
|| (*e
&& *e
!= '\n'))
3569 if (mddev
->pers
->check_reshape
== NULL
)
3571 mddev
->new_chunk_sectors
= n
>> 9;
3572 err
= mddev
->pers
->check_reshape(mddev
);
3574 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3578 mddev
->new_chunk_sectors
= n
>> 9;
3579 if (mddev
->reshape_position
== MaxSector
)
3580 mddev
->chunk_sectors
= n
>> 9;
3584 static struct md_sysfs_entry md_chunk_size
=
3585 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3588 resync_start_show(struct mddev
*mddev
, char *page
)
3590 if (mddev
->recovery_cp
== MaxSector
)
3591 return sprintf(page
, "none\n");
3592 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3596 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3599 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3601 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3603 if (cmd_match(buf
, "none"))
3605 else if (!*buf
|| (*e
&& *e
!= '\n'))
3608 mddev
->recovery_cp
= n
;
3611 static struct md_sysfs_entry md_resync_start
=
3612 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3615 * The array state can be:
3618 * No devices, no size, no level
3619 * Equivalent to STOP_ARRAY ioctl
3621 * May have some settings, but array is not active
3622 * all IO results in error
3623 * When written, doesn't tear down array, but just stops it
3624 * suspended (not supported yet)
3625 * All IO requests will block. The array can be reconfigured.
3626 * Writing this, if accepted, will block until array is quiescent
3628 * no resync can happen. no superblocks get written.
3629 * write requests fail
3631 * like readonly, but behaves like 'clean' on a write request.
3633 * clean - no pending writes, but otherwise active.
3634 * When written to inactive array, starts without resync
3635 * If a write request arrives then
3636 * if metadata is known, mark 'dirty' and switch to 'active'.
3637 * if not known, block and switch to write-pending
3638 * If written to an active array that has pending writes, then fails.
3640 * fully active: IO and resync can be happening.
3641 * When written to inactive array, starts with resync
3644 * clean, but writes are blocked waiting for 'active' to be written.
3647 * like active, but no writes have been seen for a while (100msec).
3650 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3651 write_pending
, active_idle
, bad_word
};
3652 static char *array_states
[] = {
3653 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3654 "write-pending", "active-idle", NULL
};
3656 static int match_word(const char *word
, char **list
)
3659 for (n
=0; list
[n
]; n
++)
3660 if (cmd_match(word
, list
[n
]))
3666 array_state_show(struct mddev
*mddev
, char *page
)
3668 enum array_state st
= inactive
;
3681 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3683 else if (mddev
->safemode
)
3689 if (list_empty(&mddev
->disks
) &&
3690 mddev
->raid_disks
== 0 &&
3691 mddev
->dev_sectors
== 0)
3696 return sprintf(page
, "%s\n", array_states
[st
]);
3699 static int do_md_stop(struct mddev
* mddev
, int ro
, int is_open
);
3700 static int md_set_readonly(struct mddev
* mddev
, int is_open
);
3701 static int do_md_run(struct mddev
* mddev
);
3702 static int restart_array(struct mddev
*mddev
);
3705 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3708 enum array_state st
= match_word(buf
, array_states
);
3713 /* stopping an active array */
3714 if (atomic_read(&mddev
->openers
) > 0)
3716 err
= do_md_stop(mddev
, 0, 0);
3719 /* stopping an active array */
3721 if (atomic_read(&mddev
->openers
) > 0)
3723 err
= do_md_stop(mddev
, 2, 0);
3725 err
= 0; /* already inactive */
3728 break; /* not supported yet */
3731 err
= md_set_readonly(mddev
, 0);
3734 set_disk_ro(mddev
->gendisk
, 1);
3735 err
= do_md_run(mddev
);
3741 err
= md_set_readonly(mddev
, 0);
3742 else if (mddev
->ro
== 1)
3743 err
= restart_array(mddev
);
3746 set_disk_ro(mddev
->gendisk
, 0);
3750 err
= do_md_run(mddev
);
3755 restart_array(mddev
);
3756 spin_lock_irq(&mddev
->write_lock
);
3757 if (atomic_read(&mddev
->writes_pending
) == 0) {
3758 if (mddev
->in_sync
== 0) {
3760 if (mddev
->safemode
== 1)
3761 mddev
->safemode
= 0;
3762 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3767 spin_unlock_irq(&mddev
->write_lock
);
3773 restart_array(mddev
);
3774 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3775 wake_up(&mddev
->sb_wait
);
3779 set_disk_ro(mddev
->gendisk
, 0);
3780 err
= do_md_run(mddev
);
3785 /* these cannot be set */
3791 if (mddev
->hold_active
== UNTIL_IOCTL
)
3792 mddev
->hold_active
= 0;
3793 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3797 static struct md_sysfs_entry md_array_state
=
3798 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3801 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3802 return sprintf(page
, "%d\n",
3803 atomic_read(&mddev
->max_corr_read_errors
));
3807 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3810 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3812 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3813 atomic_set(&mddev
->max_corr_read_errors
, n
);
3819 static struct md_sysfs_entry max_corr_read_errors
=
3820 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3821 max_corrected_read_errors_store
);
3824 null_show(struct mddev
*mddev
, char *page
)
3830 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3832 /* buf must be %d:%d\n? giving major and minor numbers */
3833 /* The new device is added to the array.
3834 * If the array has a persistent superblock, we read the
3835 * superblock to initialise info and check validity.
3836 * Otherwise, only checking done is that in bind_rdev_to_array,
3837 * which mainly checks size.
3840 int major
= simple_strtoul(buf
, &e
, 10);
3843 struct md_rdev
*rdev
;
3846 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3848 minor
= simple_strtoul(e
+1, &e
, 10);
3849 if (*e
&& *e
!= '\n')
3851 dev
= MKDEV(major
, minor
);
3852 if (major
!= MAJOR(dev
) ||
3853 minor
!= MINOR(dev
))
3857 if (mddev
->persistent
) {
3858 rdev
= md_import_device(dev
, mddev
->major_version
,
3859 mddev
->minor_version
);
3860 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3861 struct md_rdev
*rdev0
3862 = list_entry(mddev
->disks
.next
,
3863 struct md_rdev
, same_set
);
3864 err
= super_types
[mddev
->major_version
]
3865 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3869 } else if (mddev
->external
)
3870 rdev
= md_import_device(dev
, -2, -1);
3872 rdev
= md_import_device(dev
, -1, -1);
3875 return PTR_ERR(rdev
);
3876 err
= bind_rdev_to_array(rdev
, mddev
);
3880 return err
? err
: len
;
3883 static struct md_sysfs_entry md_new_device
=
3884 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3887 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3890 unsigned long chunk
, end_chunk
;
3894 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3896 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3897 if (buf
== end
) break;
3898 if (*end
== '-') { /* range */
3900 end_chunk
= simple_strtoul(buf
, &end
, 0);
3901 if (buf
== end
) break;
3903 if (*end
&& !isspace(*end
)) break;
3904 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3905 buf
= skip_spaces(end
);
3907 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3912 static struct md_sysfs_entry md_bitmap
=
3913 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3916 size_show(struct mddev
*mddev
, char *page
)
3918 return sprintf(page
, "%llu\n",
3919 (unsigned long long)mddev
->dev_sectors
/ 2);
3922 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
3925 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3927 /* If array is inactive, we can reduce the component size, but
3928 * not increase it (except from 0).
3929 * If array is active, we can try an on-line resize
3932 int err
= strict_blocks_to_sectors(buf
, §ors
);
3937 err
= update_size(mddev
, sectors
);
3938 md_update_sb(mddev
, 1);
3940 if (mddev
->dev_sectors
== 0 ||
3941 mddev
->dev_sectors
> sectors
)
3942 mddev
->dev_sectors
= sectors
;
3946 return err
? err
: len
;
3949 static struct md_sysfs_entry md_size
=
3950 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3955 * 'none' for arrays with no metadata (good luck...)
3956 * 'external' for arrays with externally managed metadata,
3957 * or N.M for internally known formats
3960 metadata_show(struct mddev
*mddev
, char *page
)
3962 if (mddev
->persistent
)
3963 return sprintf(page
, "%d.%d\n",
3964 mddev
->major_version
, mddev
->minor_version
);
3965 else if (mddev
->external
)
3966 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3968 return sprintf(page
, "none\n");
3972 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3976 /* Changing the details of 'external' metadata is
3977 * always permitted. Otherwise there must be
3978 * no devices attached to the array.
3980 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3982 else if (!list_empty(&mddev
->disks
))
3985 if (cmd_match(buf
, "none")) {
3986 mddev
->persistent
= 0;
3987 mddev
->external
= 0;
3988 mddev
->major_version
= 0;
3989 mddev
->minor_version
= 90;
3992 if (strncmp(buf
, "external:", 9) == 0) {
3993 size_t namelen
= len
-9;
3994 if (namelen
>= sizeof(mddev
->metadata_type
))
3995 namelen
= sizeof(mddev
->metadata_type
)-1;
3996 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3997 mddev
->metadata_type
[namelen
] = 0;
3998 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3999 mddev
->metadata_type
[--namelen
] = 0;
4000 mddev
->persistent
= 0;
4001 mddev
->external
= 1;
4002 mddev
->major_version
= 0;
4003 mddev
->minor_version
= 90;
4006 major
= simple_strtoul(buf
, &e
, 10);
4007 if (e
==buf
|| *e
!= '.')
4010 minor
= simple_strtoul(buf
, &e
, 10);
4011 if (e
==buf
|| (*e
&& *e
!= '\n') )
4013 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4015 mddev
->major_version
= major
;
4016 mddev
->minor_version
= minor
;
4017 mddev
->persistent
= 1;
4018 mddev
->external
= 0;
4022 static struct md_sysfs_entry md_metadata
=
4023 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4026 action_show(struct mddev
*mddev
, char *page
)
4028 char *type
= "idle";
4029 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4031 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4032 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4033 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4035 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4036 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4038 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4042 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4045 return sprintf(page
, "%s\n", type
);
4048 static void reap_sync_thread(struct mddev
*mddev
);
4051 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4053 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4056 if (cmd_match(page
, "frozen"))
4057 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4059 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4061 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4062 if (mddev
->sync_thread
) {
4063 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4064 reap_sync_thread(mddev
);
4066 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4067 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4069 else if (cmd_match(page
, "resync"))
4070 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4071 else if (cmd_match(page
, "recover")) {
4072 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4073 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4074 } else if (cmd_match(page
, "reshape")) {
4076 if (mddev
->pers
->start_reshape
== NULL
)
4078 err
= mddev
->pers
->start_reshape(mddev
);
4081 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4083 if (cmd_match(page
, "check"))
4084 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4085 else if (!cmd_match(page
, "repair"))
4087 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4088 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4090 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4091 md_wakeup_thread(mddev
->thread
);
4092 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4097 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4099 return sprintf(page
, "%llu\n",
4100 (unsigned long long) mddev
->resync_mismatches
);
4103 static struct md_sysfs_entry md_scan_mode
=
4104 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4107 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4110 sync_min_show(struct mddev
*mddev
, char *page
)
4112 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4113 mddev
->sync_speed_min
? "local": "system");
4117 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4121 if (strncmp(buf
, "system", 6)==0) {
4122 mddev
->sync_speed_min
= 0;
4125 min
= simple_strtoul(buf
, &e
, 10);
4126 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4128 mddev
->sync_speed_min
= min
;
4132 static struct md_sysfs_entry md_sync_min
=
4133 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4136 sync_max_show(struct mddev
*mddev
, char *page
)
4138 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4139 mddev
->sync_speed_max
? "local": "system");
4143 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4147 if (strncmp(buf
, "system", 6)==0) {
4148 mddev
->sync_speed_max
= 0;
4151 max
= simple_strtoul(buf
, &e
, 10);
4152 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4154 mddev
->sync_speed_max
= max
;
4158 static struct md_sysfs_entry md_sync_max
=
4159 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4162 degraded_show(struct mddev
*mddev
, char *page
)
4164 return sprintf(page
, "%d\n", mddev
->degraded
);
4166 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4169 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4171 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4175 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4179 if (strict_strtol(buf
, 10, &n
))
4182 if (n
!= 0 && n
!= 1)
4185 mddev
->parallel_resync
= n
;
4187 if (mddev
->sync_thread
)
4188 wake_up(&resync_wait
);
4193 /* force parallel resync, even with shared block devices */
4194 static struct md_sysfs_entry md_sync_force_parallel
=
4195 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4196 sync_force_parallel_show
, sync_force_parallel_store
);
4199 sync_speed_show(struct mddev
*mddev
, char *page
)
4201 unsigned long resync
, dt
, db
;
4202 if (mddev
->curr_resync
== 0)
4203 return sprintf(page
, "none\n");
4204 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4205 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4207 db
= resync
- mddev
->resync_mark_cnt
;
4208 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4211 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4214 sync_completed_show(struct mddev
*mddev
, char *page
)
4216 unsigned long long max_sectors
, resync
;
4218 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4219 return sprintf(page
, "none\n");
4221 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4222 max_sectors
= mddev
->resync_max_sectors
;
4224 max_sectors
= mddev
->dev_sectors
;
4226 resync
= mddev
->curr_resync_completed
;
4227 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4230 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4233 min_sync_show(struct mddev
*mddev
, char *page
)
4235 return sprintf(page
, "%llu\n",
4236 (unsigned long long)mddev
->resync_min
);
4239 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4241 unsigned long long min
;
4242 if (strict_strtoull(buf
, 10, &min
))
4244 if (min
> mddev
->resync_max
)
4246 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4249 /* Must be a multiple of chunk_size */
4250 if (mddev
->chunk_sectors
) {
4251 sector_t temp
= min
;
4252 if (sector_div(temp
, mddev
->chunk_sectors
))
4255 mddev
->resync_min
= min
;
4260 static struct md_sysfs_entry md_min_sync
=
4261 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4264 max_sync_show(struct mddev
*mddev
, char *page
)
4266 if (mddev
->resync_max
== MaxSector
)
4267 return sprintf(page
, "max\n");
4269 return sprintf(page
, "%llu\n",
4270 (unsigned long long)mddev
->resync_max
);
4273 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4275 if (strncmp(buf
, "max", 3) == 0)
4276 mddev
->resync_max
= MaxSector
;
4278 unsigned long long max
;
4279 if (strict_strtoull(buf
, 10, &max
))
4281 if (max
< mddev
->resync_min
)
4283 if (max
< mddev
->resync_max
&&
4285 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4288 /* Must be a multiple of chunk_size */
4289 if (mddev
->chunk_sectors
) {
4290 sector_t temp
= max
;
4291 if (sector_div(temp
, mddev
->chunk_sectors
))
4294 mddev
->resync_max
= max
;
4296 wake_up(&mddev
->recovery_wait
);
4300 static struct md_sysfs_entry md_max_sync
=
4301 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4304 suspend_lo_show(struct mddev
*mddev
, char *page
)
4306 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4310 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4313 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4314 unsigned long long old
= mddev
->suspend_lo
;
4316 if (mddev
->pers
== NULL
||
4317 mddev
->pers
->quiesce
== NULL
)
4319 if (buf
== e
|| (*e
&& *e
!= '\n'))
4322 mddev
->suspend_lo
= new;
4324 /* Shrinking suspended region */
4325 mddev
->pers
->quiesce(mddev
, 2);
4327 /* Expanding suspended region - need to wait */
4328 mddev
->pers
->quiesce(mddev
, 1);
4329 mddev
->pers
->quiesce(mddev
, 0);
4333 static struct md_sysfs_entry md_suspend_lo
=
4334 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4338 suspend_hi_show(struct mddev
*mddev
, char *page
)
4340 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4344 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4347 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4348 unsigned long long old
= mddev
->suspend_hi
;
4350 if (mddev
->pers
== NULL
||
4351 mddev
->pers
->quiesce
== NULL
)
4353 if (buf
== e
|| (*e
&& *e
!= '\n'))
4356 mddev
->suspend_hi
= new;
4358 /* Shrinking suspended region */
4359 mddev
->pers
->quiesce(mddev
, 2);
4361 /* Expanding suspended region - need to wait */
4362 mddev
->pers
->quiesce(mddev
, 1);
4363 mddev
->pers
->quiesce(mddev
, 0);
4367 static struct md_sysfs_entry md_suspend_hi
=
4368 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4371 reshape_position_show(struct mddev
*mddev
, char *page
)
4373 if (mddev
->reshape_position
!= MaxSector
)
4374 return sprintf(page
, "%llu\n",
4375 (unsigned long long)mddev
->reshape_position
);
4376 strcpy(page
, "none\n");
4381 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4384 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4387 if (buf
== e
|| (*e
&& *e
!= '\n'))
4389 mddev
->reshape_position
= new;
4390 mddev
->delta_disks
= 0;
4391 mddev
->new_level
= mddev
->level
;
4392 mddev
->new_layout
= mddev
->layout
;
4393 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4397 static struct md_sysfs_entry md_reshape_position
=
4398 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4399 reshape_position_store
);
4402 array_size_show(struct mddev
*mddev
, char *page
)
4404 if (mddev
->external_size
)
4405 return sprintf(page
, "%llu\n",
4406 (unsigned long long)mddev
->array_sectors
/2);
4408 return sprintf(page
, "default\n");
4412 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4416 if (strncmp(buf
, "default", 7) == 0) {
4418 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4420 sectors
= mddev
->array_sectors
;
4422 mddev
->external_size
= 0;
4424 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4426 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4429 mddev
->external_size
= 1;
4432 mddev
->array_sectors
= sectors
;
4434 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4435 revalidate_disk(mddev
->gendisk
);
4440 static struct md_sysfs_entry md_array_size
=
4441 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4444 static struct attribute
*md_default_attrs
[] = {
4447 &md_raid_disks
.attr
,
4448 &md_chunk_size
.attr
,
4450 &md_resync_start
.attr
,
4452 &md_new_device
.attr
,
4453 &md_safe_delay
.attr
,
4454 &md_array_state
.attr
,
4455 &md_reshape_position
.attr
,
4456 &md_array_size
.attr
,
4457 &max_corr_read_errors
.attr
,
4461 static struct attribute
*md_redundancy_attrs
[] = {
4463 &md_mismatches
.attr
,
4466 &md_sync_speed
.attr
,
4467 &md_sync_force_parallel
.attr
,
4468 &md_sync_completed
.attr
,
4471 &md_suspend_lo
.attr
,
4472 &md_suspend_hi
.attr
,
4477 static struct attribute_group md_redundancy_group
= {
4479 .attrs
= md_redundancy_attrs
,
4484 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4486 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4487 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4492 rv
= mddev_lock(mddev
);
4494 rv
= entry
->show(mddev
, page
);
4495 mddev_unlock(mddev
);
4501 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4502 const char *page
, size_t length
)
4504 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4505 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4510 if (!capable(CAP_SYS_ADMIN
))
4512 rv
= mddev_lock(mddev
);
4514 rv
= entry
->store(mddev
, page
, length
);
4515 mddev_unlock(mddev
);
4520 static void md_free(struct kobject
*ko
)
4522 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4524 if (mddev
->sysfs_state
)
4525 sysfs_put(mddev
->sysfs_state
);
4527 if (mddev
->gendisk
) {
4528 del_gendisk(mddev
->gendisk
);
4529 put_disk(mddev
->gendisk
);
4532 blk_cleanup_queue(mddev
->queue
);
4537 static const struct sysfs_ops md_sysfs_ops
= {
4538 .show
= md_attr_show
,
4539 .store
= md_attr_store
,
4541 static struct kobj_type md_ktype
= {
4543 .sysfs_ops
= &md_sysfs_ops
,
4544 .default_attrs
= md_default_attrs
,
4549 static void mddev_delayed_delete(struct work_struct
*ws
)
4551 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4553 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4554 kobject_del(&mddev
->kobj
);
4555 kobject_put(&mddev
->kobj
);
4558 static int md_alloc(dev_t dev
, char *name
)
4560 static DEFINE_MUTEX(disks_mutex
);
4561 struct mddev
*mddev
= mddev_find(dev
);
4562 struct gendisk
*disk
;
4571 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4572 shift
= partitioned
? MdpMinorShift
: 0;
4573 unit
= MINOR(mddev
->unit
) >> shift
;
4575 /* wait for any previous instance of this device to be
4576 * completely removed (mddev_delayed_delete).
4578 flush_workqueue(md_misc_wq
);
4580 mutex_lock(&disks_mutex
);
4586 /* Need to ensure that 'name' is not a duplicate.
4588 struct mddev
*mddev2
;
4589 spin_lock(&all_mddevs_lock
);
4591 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4592 if (mddev2
->gendisk
&&
4593 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4594 spin_unlock(&all_mddevs_lock
);
4597 spin_unlock(&all_mddevs_lock
);
4601 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4604 mddev
->queue
->queuedata
= mddev
;
4606 blk_queue_make_request(mddev
->queue
, md_make_request
);
4608 disk
= alloc_disk(1 << shift
);
4610 blk_cleanup_queue(mddev
->queue
);
4611 mddev
->queue
= NULL
;
4614 disk
->major
= MAJOR(mddev
->unit
);
4615 disk
->first_minor
= unit
<< shift
;
4617 strcpy(disk
->disk_name
, name
);
4618 else if (partitioned
)
4619 sprintf(disk
->disk_name
, "md_d%d", unit
);
4621 sprintf(disk
->disk_name
, "md%d", unit
);
4622 disk
->fops
= &md_fops
;
4623 disk
->private_data
= mddev
;
4624 disk
->queue
= mddev
->queue
;
4625 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4626 /* Allow extended partitions. This makes the
4627 * 'mdp' device redundant, but we can't really
4630 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4631 mddev
->gendisk
= disk
;
4632 /* As soon as we call add_disk(), another thread could get
4633 * through to md_open, so make sure it doesn't get too far
4635 mutex_lock(&mddev
->open_mutex
);
4638 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4639 &disk_to_dev(disk
)->kobj
, "%s", "md");
4641 /* This isn't possible, but as kobject_init_and_add is marked
4642 * __must_check, we must do something with the result
4644 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4648 if (mddev
->kobj
.sd
&&
4649 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4650 printk(KERN_DEBUG
"pointless warning\n");
4651 mutex_unlock(&mddev
->open_mutex
);
4653 mutex_unlock(&disks_mutex
);
4654 if (!error
&& mddev
->kobj
.sd
) {
4655 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4656 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4662 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4664 md_alloc(dev
, NULL
);
4668 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4670 /* val must be "md_*" where * is not all digits.
4671 * We allocate an array with a large free minor number, and
4672 * set the name to val. val must not already be an active name.
4674 int len
= strlen(val
);
4675 char buf
[DISK_NAME_LEN
];
4677 while (len
&& val
[len
-1] == '\n')
4679 if (len
>= DISK_NAME_LEN
)
4681 strlcpy(buf
, val
, len
+1);
4682 if (strncmp(buf
, "md_", 3) != 0)
4684 return md_alloc(0, buf
);
4687 static void md_safemode_timeout(unsigned long data
)
4689 struct mddev
*mddev
= (struct mddev
*) data
;
4691 if (!atomic_read(&mddev
->writes_pending
)) {
4692 mddev
->safemode
= 1;
4693 if (mddev
->external
)
4694 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4696 md_wakeup_thread(mddev
->thread
);
4699 static int start_dirty_degraded
;
4701 int md_run(struct mddev
*mddev
)
4704 struct md_rdev
*rdev
;
4705 struct md_personality
*pers
;
4707 if (list_empty(&mddev
->disks
))
4708 /* cannot run an array with no devices.. */
4713 /* Cannot run until previous stop completes properly */
4714 if (mddev
->sysfs_active
)
4718 * Analyze all RAID superblock(s)
4720 if (!mddev
->raid_disks
) {
4721 if (!mddev
->persistent
)
4726 if (mddev
->level
!= LEVEL_NONE
)
4727 request_module("md-level-%d", mddev
->level
);
4728 else if (mddev
->clevel
[0])
4729 request_module("md-%s", mddev
->clevel
);
4732 * Drop all container device buffers, from now on
4733 * the only valid external interface is through the md
4736 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4737 if (test_bit(Faulty
, &rdev
->flags
))
4739 sync_blockdev(rdev
->bdev
);
4740 invalidate_bdev(rdev
->bdev
);
4742 /* perform some consistency tests on the device.
4743 * We don't want the data to overlap the metadata,
4744 * Internal Bitmap issues have been handled elsewhere.
4746 if (rdev
->meta_bdev
) {
4747 /* Nothing to check */;
4748 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4749 if (mddev
->dev_sectors
&&
4750 rdev
->data_offset
+ mddev
->dev_sectors
4752 printk("md: %s: data overlaps metadata\n",
4757 if (rdev
->sb_start
+ rdev
->sb_size
/512
4758 > rdev
->data_offset
) {
4759 printk("md: %s: metadata overlaps data\n",
4764 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4767 if (mddev
->bio_set
== NULL
)
4768 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
4769 sizeof(struct mddev
*));
4771 spin_lock(&pers_lock
);
4772 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4773 if (!pers
|| !try_module_get(pers
->owner
)) {
4774 spin_unlock(&pers_lock
);
4775 if (mddev
->level
!= LEVEL_NONE
)
4776 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4779 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4784 spin_unlock(&pers_lock
);
4785 if (mddev
->level
!= pers
->level
) {
4786 mddev
->level
= pers
->level
;
4787 mddev
->new_level
= pers
->level
;
4789 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4791 if (mddev
->reshape_position
!= MaxSector
&&
4792 pers
->start_reshape
== NULL
) {
4793 /* This personality cannot handle reshaping... */
4795 module_put(pers
->owner
);
4799 if (pers
->sync_request
) {
4800 /* Warn if this is a potentially silly
4803 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4804 struct md_rdev
*rdev2
;
4807 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4808 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4810 rdev
->bdev
->bd_contains
==
4811 rdev2
->bdev
->bd_contains
) {
4813 "%s: WARNING: %s appears to be"
4814 " on the same physical disk as"
4817 bdevname(rdev
->bdev
,b
),
4818 bdevname(rdev2
->bdev
,b2
));
4825 "True protection against single-disk"
4826 " failure might be compromised.\n");
4829 mddev
->recovery
= 0;
4830 /* may be over-ridden by personality */
4831 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4833 mddev
->ok_start_degraded
= start_dirty_degraded
;
4835 if (start_readonly
&& mddev
->ro
== 0)
4836 mddev
->ro
= 2; /* read-only, but switch on first write */
4838 err
= mddev
->pers
->run(mddev
);
4840 printk(KERN_ERR
"md: pers->run() failed ...\n");
4841 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4842 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4843 " but 'external_size' not in effect?\n", __func__
);
4845 "md: invalid array_size %llu > default size %llu\n",
4846 (unsigned long long)mddev
->array_sectors
/ 2,
4847 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4849 mddev
->pers
->stop(mddev
);
4851 if (err
== 0 && mddev
->pers
->sync_request
) {
4852 err
= bitmap_create(mddev
);
4854 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4855 mdname(mddev
), err
);
4856 mddev
->pers
->stop(mddev
);
4860 module_put(mddev
->pers
->owner
);
4862 bitmap_destroy(mddev
);
4865 if (mddev
->pers
->sync_request
) {
4866 if (mddev
->kobj
.sd
&&
4867 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4869 "md: cannot register extra attributes for %s\n",
4871 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4872 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4875 atomic_set(&mddev
->writes_pending
,0);
4876 atomic_set(&mddev
->max_corr_read_errors
,
4877 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4878 mddev
->safemode
= 0;
4879 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4880 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4881 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4885 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4886 if (rdev
->raid_disk
>= 0)
4887 if (sysfs_link_rdev(mddev
, rdev
))
4888 /* failure here is OK */;
4890 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4893 md_update_sb(mddev
, 0);
4895 md_new_event(mddev
);
4896 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4897 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4898 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4901 EXPORT_SYMBOL_GPL(md_run
);
4903 static int do_md_run(struct mddev
*mddev
)
4907 err
= md_run(mddev
);
4910 err
= bitmap_load(mddev
);
4912 bitmap_destroy(mddev
);
4916 md_wakeup_thread(mddev
->thread
);
4917 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4919 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4920 revalidate_disk(mddev
->gendisk
);
4922 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4927 static int restart_array(struct mddev
*mddev
)
4929 struct gendisk
*disk
= mddev
->gendisk
;
4931 /* Complain if it has no devices */
4932 if (list_empty(&mddev
->disks
))
4938 mddev
->safemode
= 0;
4940 set_disk_ro(disk
, 0);
4941 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4943 /* Kick recovery or resync if necessary */
4944 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4945 md_wakeup_thread(mddev
->thread
);
4946 md_wakeup_thread(mddev
->sync_thread
);
4947 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4951 /* similar to deny_write_access, but accounts for our holding a reference
4952 * to the file ourselves */
4953 static int deny_bitmap_write_access(struct file
* file
)
4955 struct inode
*inode
= file
->f_mapping
->host
;
4957 spin_lock(&inode
->i_lock
);
4958 if (atomic_read(&inode
->i_writecount
) > 1) {
4959 spin_unlock(&inode
->i_lock
);
4962 atomic_set(&inode
->i_writecount
, -1);
4963 spin_unlock(&inode
->i_lock
);
4968 void restore_bitmap_write_access(struct file
*file
)
4970 struct inode
*inode
= file
->f_mapping
->host
;
4972 spin_lock(&inode
->i_lock
);
4973 atomic_set(&inode
->i_writecount
, 1);
4974 spin_unlock(&inode
->i_lock
);
4977 static void md_clean(struct mddev
*mddev
)
4979 mddev
->array_sectors
= 0;
4980 mddev
->external_size
= 0;
4981 mddev
->dev_sectors
= 0;
4982 mddev
->raid_disks
= 0;
4983 mddev
->recovery_cp
= 0;
4984 mddev
->resync_min
= 0;
4985 mddev
->resync_max
= MaxSector
;
4986 mddev
->reshape_position
= MaxSector
;
4987 mddev
->external
= 0;
4988 mddev
->persistent
= 0;
4989 mddev
->level
= LEVEL_NONE
;
4990 mddev
->clevel
[0] = 0;
4993 mddev
->metadata_type
[0] = 0;
4994 mddev
->chunk_sectors
= 0;
4995 mddev
->ctime
= mddev
->utime
= 0;
4997 mddev
->max_disks
= 0;
4999 mddev
->can_decrease_events
= 0;
5000 mddev
->delta_disks
= 0;
5001 mddev
->new_level
= LEVEL_NONE
;
5002 mddev
->new_layout
= 0;
5003 mddev
->new_chunk_sectors
= 0;
5004 mddev
->curr_resync
= 0;
5005 mddev
->resync_mismatches
= 0;
5006 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5007 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5008 mddev
->recovery
= 0;
5011 mddev
->degraded
= 0;
5012 mddev
->safemode
= 0;
5013 mddev
->bitmap_info
.offset
= 0;
5014 mddev
->bitmap_info
.default_offset
= 0;
5015 mddev
->bitmap_info
.chunksize
= 0;
5016 mddev
->bitmap_info
.daemon_sleep
= 0;
5017 mddev
->bitmap_info
.max_write_behind
= 0;
5020 static void __md_stop_writes(struct mddev
*mddev
)
5022 if (mddev
->sync_thread
) {
5023 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5024 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5025 reap_sync_thread(mddev
);
5028 del_timer_sync(&mddev
->safemode_timer
);
5030 bitmap_flush(mddev
);
5031 md_super_wait(mddev
);
5033 if (!mddev
->in_sync
|| mddev
->flags
) {
5034 /* mark array as shutdown cleanly */
5036 md_update_sb(mddev
, 1);
5040 void md_stop_writes(struct mddev
*mddev
)
5043 __md_stop_writes(mddev
);
5044 mddev_unlock(mddev
);
5046 EXPORT_SYMBOL_GPL(md_stop_writes
);
5048 void md_stop(struct mddev
*mddev
)
5051 mddev
->pers
->stop(mddev
);
5052 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5053 mddev
->to_remove
= &md_redundancy_group
;
5054 module_put(mddev
->pers
->owner
);
5056 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5058 EXPORT_SYMBOL_GPL(md_stop
);
5060 static int md_set_readonly(struct mddev
*mddev
, int is_open
)
5063 mutex_lock(&mddev
->open_mutex
);
5064 if (atomic_read(&mddev
->openers
) > is_open
) {
5065 printk("md: %s still in use.\n",mdname(mddev
));
5070 __md_stop_writes(mddev
);
5076 set_disk_ro(mddev
->gendisk
, 1);
5077 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5078 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5082 mutex_unlock(&mddev
->open_mutex
);
5087 * 0 - completely stop and dis-assemble array
5088 * 2 - stop but do not disassemble array
5090 static int do_md_stop(struct mddev
* mddev
, int mode
, int is_open
)
5092 struct gendisk
*disk
= mddev
->gendisk
;
5093 struct md_rdev
*rdev
;
5095 mutex_lock(&mddev
->open_mutex
);
5096 if (atomic_read(&mddev
->openers
) > is_open
||
5097 mddev
->sysfs_active
) {
5098 printk("md: %s still in use.\n",mdname(mddev
));
5099 mutex_unlock(&mddev
->open_mutex
);
5105 set_disk_ro(disk
, 0);
5107 __md_stop_writes(mddev
);
5109 mddev
->queue
->merge_bvec_fn
= NULL
;
5110 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5112 /* tell userspace to handle 'inactive' */
5113 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5115 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5116 if (rdev
->raid_disk
>= 0)
5117 sysfs_unlink_rdev(mddev
, rdev
);
5119 set_capacity(disk
, 0);
5120 mutex_unlock(&mddev
->open_mutex
);
5122 revalidate_disk(disk
);
5127 mutex_unlock(&mddev
->open_mutex
);
5129 * Free resources if final stop
5132 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5134 bitmap_destroy(mddev
);
5135 if (mddev
->bitmap_info
.file
) {
5136 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5137 fput(mddev
->bitmap_info
.file
);
5138 mddev
->bitmap_info
.file
= NULL
;
5140 mddev
->bitmap_info
.offset
= 0;
5142 export_array(mddev
);
5145 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5146 if (mddev
->hold_active
== UNTIL_STOP
)
5147 mddev
->hold_active
= 0;
5149 blk_integrity_unregister(disk
);
5150 md_new_event(mddev
);
5151 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5156 static void autorun_array(struct mddev
*mddev
)
5158 struct md_rdev
*rdev
;
5161 if (list_empty(&mddev
->disks
))
5164 printk(KERN_INFO
"md: running: ");
5166 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5167 char b
[BDEVNAME_SIZE
];
5168 printk("<%s>", bdevname(rdev
->bdev
,b
));
5172 err
= do_md_run(mddev
);
5174 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5175 do_md_stop(mddev
, 0, 0);
5180 * lets try to run arrays based on all disks that have arrived
5181 * until now. (those are in pending_raid_disks)
5183 * the method: pick the first pending disk, collect all disks with
5184 * the same UUID, remove all from the pending list and put them into
5185 * the 'same_array' list. Then order this list based on superblock
5186 * update time (freshest comes first), kick out 'old' disks and
5187 * compare superblocks. If everything's fine then run it.
5189 * If "unit" is allocated, then bump its reference count
5191 static void autorun_devices(int part
)
5193 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5194 struct mddev
*mddev
;
5195 char b
[BDEVNAME_SIZE
];
5197 printk(KERN_INFO
"md: autorun ...\n");
5198 while (!list_empty(&pending_raid_disks
)) {
5201 LIST_HEAD(candidates
);
5202 rdev0
= list_entry(pending_raid_disks
.next
,
5203 struct md_rdev
, same_set
);
5205 printk(KERN_INFO
"md: considering %s ...\n",
5206 bdevname(rdev0
->bdev
,b
));
5207 INIT_LIST_HEAD(&candidates
);
5208 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5209 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5210 printk(KERN_INFO
"md: adding %s ...\n",
5211 bdevname(rdev
->bdev
,b
));
5212 list_move(&rdev
->same_set
, &candidates
);
5215 * now we have a set of devices, with all of them having
5216 * mostly sane superblocks. It's time to allocate the
5220 dev
= MKDEV(mdp_major
,
5221 rdev0
->preferred_minor
<< MdpMinorShift
);
5222 unit
= MINOR(dev
) >> MdpMinorShift
;
5224 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5227 if (rdev0
->preferred_minor
!= unit
) {
5228 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5229 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5233 md_probe(dev
, NULL
, NULL
);
5234 mddev
= mddev_find(dev
);
5235 if (!mddev
|| !mddev
->gendisk
) {
5239 "md: cannot allocate memory for md drive.\n");
5242 if (mddev_lock(mddev
))
5243 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5245 else if (mddev
->raid_disks
|| mddev
->major_version
5246 || !list_empty(&mddev
->disks
)) {
5248 "md: %s already running, cannot run %s\n",
5249 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5250 mddev_unlock(mddev
);
5252 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5253 mddev
->persistent
= 1;
5254 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5255 list_del_init(&rdev
->same_set
);
5256 if (bind_rdev_to_array(rdev
, mddev
))
5259 autorun_array(mddev
);
5260 mddev_unlock(mddev
);
5262 /* on success, candidates will be empty, on error
5265 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5266 list_del_init(&rdev
->same_set
);
5271 printk(KERN_INFO
"md: ... autorun DONE.\n");
5273 #endif /* !MODULE */
5275 static int get_version(void __user
* arg
)
5279 ver
.major
= MD_MAJOR_VERSION
;
5280 ver
.minor
= MD_MINOR_VERSION
;
5281 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5283 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5289 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5291 mdu_array_info_t info
;
5292 int nr
,working
,insync
,failed
,spare
;
5293 struct md_rdev
*rdev
;
5295 nr
=working
=insync
=failed
=spare
=0;
5296 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5298 if (test_bit(Faulty
, &rdev
->flags
))
5302 if (test_bit(In_sync
, &rdev
->flags
))
5309 info
.major_version
= mddev
->major_version
;
5310 info
.minor_version
= mddev
->minor_version
;
5311 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5312 info
.ctime
= mddev
->ctime
;
5313 info
.level
= mddev
->level
;
5314 info
.size
= mddev
->dev_sectors
/ 2;
5315 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5318 info
.raid_disks
= mddev
->raid_disks
;
5319 info
.md_minor
= mddev
->md_minor
;
5320 info
.not_persistent
= !mddev
->persistent
;
5322 info
.utime
= mddev
->utime
;
5325 info
.state
= (1<<MD_SB_CLEAN
);
5326 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5327 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5328 info
.active_disks
= insync
;
5329 info
.working_disks
= working
;
5330 info
.failed_disks
= failed
;
5331 info
.spare_disks
= spare
;
5333 info
.layout
= mddev
->layout
;
5334 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5336 if (copy_to_user(arg
, &info
, sizeof(info
)))
5342 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5344 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5345 char *ptr
, *buf
= NULL
;
5348 if (md_allow_write(mddev
))
5349 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5351 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5356 /* bitmap disabled, zero the first byte and copy out */
5357 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5358 file
->pathname
[0] = '\0';
5362 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5366 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5370 strcpy(file
->pathname
, ptr
);
5374 if (copy_to_user(arg
, file
, sizeof(*file
)))
5382 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5384 mdu_disk_info_t info
;
5385 struct md_rdev
*rdev
;
5387 if (copy_from_user(&info
, arg
, sizeof(info
)))
5390 rdev
= find_rdev_nr(mddev
, info
.number
);
5392 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5393 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5394 info
.raid_disk
= rdev
->raid_disk
;
5396 if (test_bit(Faulty
, &rdev
->flags
))
5397 info
.state
|= (1<<MD_DISK_FAULTY
);
5398 else if (test_bit(In_sync
, &rdev
->flags
)) {
5399 info
.state
|= (1<<MD_DISK_ACTIVE
);
5400 info
.state
|= (1<<MD_DISK_SYNC
);
5402 if (test_bit(WriteMostly
, &rdev
->flags
))
5403 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5405 info
.major
= info
.minor
= 0;
5406 info
.raid_disk
= -1;
5407 info
.state
= (1<<MD_DISK_REMOVED
);
5410 if (copy_to_user(arg
, &info
, sizeof(info
)))
5416 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5418 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5419 struct md_rdev
*rdev
;
5420 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5422 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5425 if (!mddev
->raid_disks
) {
5427 /* expecting a device which has a superblock */
5428 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5431 "md: md_import_device returned %ld\n",
5433 return PTR_ERR(rdev
);
5435 if (!list_empty(&mddev
->disks
)) {
5436 struct md_rdev
*rdev0
5437 = list_entry(mddev
->disks
.next
,
5438 struct md_rdev
, same_set
);
5439 err
= super_types
[mddev
->major_version
]
5440 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5443 "md: %s has different UUID to %s\n",
5444 bdevname(rdev
->bdev
,b
),
5445 bdevname(rdev0
->bdev
,b2
));
5450 err
= bind_rdev_to_array(rdev
, mddev
);
5457 * add_new_disk can be used once the array is assembled
5458 * to add "hot spares". They must already have a superblock
5463 if (!mddev
->pers
->hot_add_disk
) {
5465 "%s: personality does not support diskops!\n",
5469 if (mddev
->persistent
)
5470 rdev
= md_import_device(dev
, mddev
->major_version
,
5471 mddev
->minor_version
);
5473 rdev
= md_import_device(dev
, -1, -1);
5476 "md: md_import_device returned %ld\n",
5478 return PTR_ERR(rdev
);
5480 /* set saved_raid_disk if appropriate */
5481 if (!mddev
->persistent
) {
5482 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5483 info
->raid_disk
< mddev
->raid_disks
) {
5484 rdev
->raid_disk
= info
->raid_disk
;
5485 set_bit(In_sync
, &rdev
->flags
);
5487 rdev
->raid_disk
= -1;
5489 super_types
[mddev
->major_version
].
5490 validate_super(mddev
, rdev
);
5491 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5492 (!test_bit(In_sync
, &rdev
->flags
) ||
5493 rdev
->raid_disk
!= info
->raid_disk
)) {
5494 /* This was a hot-add request, but events doesn't
5495 * match, so reject it.
5501 if (test_bit(In_sync
, &rdev
->flags
))
5502 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5504 rdev
->saved_raid_disk
= -1;
5506 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5507 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5508 set_bit(WriteMostly
, &rdev
->flags
);
5510 clear_bit(WriteMostly
, &rdev
->flags
);
5512 rdev
->raid_disk
= -1;
5513 err
= bind_rdev_to_array(rdev
, mddev
);
5514 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5515 /* If there is hot_add_disk but no hot_remove_disk
5516 * then added disks for geometry changes,
5517 * and should be added immediately.
5519 super_types
[mddev
->major_version
].
5520 validate_super(mddev
, rdev
);
5521 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5523 unbind_rdev_from_array(rdev
);
5528 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5530 md_update_sb(mddev
, 1);
5531 if (mddev
->degraded
)
5532 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5533 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5535 md_new_event(mddev
);
5536 md_wakeup_thread(mddev
->thread
);
5540 /* otherwise, add_new_disk is only allowed
5541 * for major_version==0 superblocks
5543 if (mddev
->major_version
!= 0) {
5544 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5549 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5551 rdev
= md_import_device(dev
, -1, 0);
5554 "md: error, md_import_device() returned %ld\n",
5556 return PTR_ERR(rdev
);
5558 rdev
->desc_nr
= info
->number
;
5559 if (info
->raid_disk
< mddev
->raid_disks
)
5560 rdev
->raid_disk
= info
->raid_disk
;
5562 rdev
->raid_disk
= -1;
5564 if (rdev
->raid_disk
< mddev
->raid_disks
)
5565 if (info
->state
& (1<<MD_DISK_SYNC
))
5566 set_bit(In_sync
, &rdev
->flags
);
5568 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5569 set_bit(WriteMostly
, &rdev
->flags
);
5571 if (!mddev
->persistent
) {
5572 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5573 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5575 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5576 rdev
->sectors
= rdev
->sb_start
;
5578 err
= bind_rdev_to_array(rdev
, mddev
);
5588 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5590 char b
[BDEVNAME_SIZE
];
5591 struct md_rdev
*rdev
;
5593 rdev
= find_rdev(mddev
, dev
);
5597 if (rdev
->raid_disk
>= 0)
5600 kick_rdev_from_array(rdev
);
5601 md_update_sb(mddev
, 1);
5602 md_new_event(mddev
);
5606 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5607 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5611 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5613 char b
[BDEVNAME_SIZE
];
5615 struct md_rdev
*rdev
;
5620 if (mddev
->major_version
!= 0) {
5621 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5622 " version-0 superblocks.\n",
5626 if (!mddev
->pers
->hot_add_disk
) {
5628 "%s: personality does not support diskops!\n",
5633 rdev
= md_import_device(dev
, -1, 0);
5636 "md: error, md_import_device() returned %ld\n",
5641 if (mddev
->persistent
)
5642 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5644 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5646 rdev
->sectors
= rdev
->sb_start
;
5648 if (test_bit(Faulty
, &rdev
->flags
)) {
5650 "md: can not hot-add faulty %s disk to %s!\n",
5651 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5655 clear_bit(In_sync
, &rdev
->flags
);
5657 rdev
->saved_raid_disk
= -1;
5658 err
= bind_rdev_to_array(rdev
, mddev
);
5663 * The rest should better be atomic, we can have disk failures
5664 * noticed in interrupt contexts ...
5667 rdev
->raid_disk
= -1;
5669 md_update_sb(mddev
, 1);
5672 * Kick recovery, maybe this spare has to be added to the
5673 * array immediately.
5675 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5676 md_wakeup_thread(mddev
->thread
);
5677 md_new_event(mddev
);
5685 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5690 if (!mddev
->pers
->quiesce
)
5692 if (mddev
->recovery
|| mddev
->sync_thread
)
5694 /* we should be able to change the bitmap.. */
5700 return -EEXIST
; /* cannot add when bitmap is present */
5701 mddev
->bitmap_info
.file
= fget(fd
);
5703 if (mddev
->bitmap_info
.file
== NULL
) {
5704 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5709 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5711 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5713 fput(mddev
->bitmap_info
.file
);
5714 mddev
->bitmap_info
.file
= NULL
;
5717 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5718 } else if (mddev
->bitmap
== NULL
)
5719 return -ENOENT
; /* cannot remove what isn't there */
5722 mddev
->pers
->quiesce(mddev
, 1);
5724 err
= bitmap_create(mddev
);
5726 err
= bitmap_load(mddev
);
5728 if (fd
< 0 || err
) {
5729 bitmap_destroy(mddev
);
5730 fd
= -1; /* make sure to put the file */
5732 mddev
->pers
->quiesce(mddev
, 0);
5735 if (mddev
->bitmap_info
.file
) {
5736 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5737 fput(mddev
->bitmap_info
.file
);
5739 mddev
->bitmap_info
.file
= NULL
;
5746 * set_array_info is used two different ways
5747 * The original usage is when creating a new array.
5748 * In this usage, raid_disks is > 0 and it together with
5749 * level, size, not_persistent,layout,chunksize determine the
5750 * shape of the array.
5751 * This will always create an array with a type-0.90.0 superblock.
5752 * The newer usage is when assembling an array.
5753 * In this case raid_disks will be 0, and the major_version field is
5754 * use to determine which style super-blocks are to be found on the devices.
5755 * The minor and patch _version numbers are also kept incase the
5756 * super_block handler wishes to interpret them.
5758 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
5761 if (info
->raid_disks
== 0) {
5762 /* just setting version number for superblock loading */
5763 if (info
->major_version
< 0 ||
5764 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5765 super_types
[info
->major_version
].name
== NULL
) {
5766 /* maybe try to auto-load a module? */
5768 "md: superblock version %d not known\n",
5769 info
->major_version
);
5772 mddev
->major_version
= info
->major_version
;
5773 mddev
->minor_version
= info
->minor_version
;
5774 mddev
->patch_version
= info
->patch_version
;
5775 mddev
->persistent
= !info
->not_persistent
;
5776 /* ensure mddev_put doesn't delete this now that there
5777 * is some minimal configuration.
5779 mddev
->ctime
= get_seconds();
5782 mddev
->major_version
= MD_MAJOR_VERSION
;
5783 mddev
->minor_version
= MD_MINOR_VERSION
;
5784 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5785 mddev
->ctime
= get_seconds();
5787 mddev
->level
= info
->level
;
5788 mddev
->clevel
[0] = 0;
5789 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5790 mddev
->raid_disks
= info
->raid_disks
;
5791 /* don't set md_minor, it is determined by which /dev/md* was
5794 if (info
->state
& (1<<MD_SB_CLEAN
))
5795 mddev
->recovery_cp
= MaxSector
;
5797 mddev
->recovery_cp
= 0;
5798 mddev
->persistent
= ! info
->not_persistent
;
5799 mddev
->external
= 0;
5801 mddev
->layout
= info
->layout
;
5802 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5804 mddev
->max_disks
= MD_SB_DISKS
;
5806 if (mddev
->persistent
)
5808 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5810 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5811 mddev
->bitmap_info
.offset
= 0;
5813 mddev
->reshape_position
= MaxSector
;
5816 * Generate a 128 bit UUID
5818 get_random_bytes(mddev
->uuid
, 16);
5820 mddev
->new_level
= mddev
->level
;
5821 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5822 mddev
->new_layout
= mddev
->layout
;
5823 mddev
->delta_disks
= 0;
5828 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
5830 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5832 if (mddev
->external_size
)
5835 mddev
->array_sectors
= array_sectors
;
5837 EXPORT_SYMBOL(md_set_array_sectors
);
5839 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
5841 struct md_rdev
*rdev
;
5843 int fit
= (num_sectors
== 0);
5845 if (mddev
->pers
->resize
== NULL
)
5847 /* The "num_sectors" is the number of sectors of each device that
5848 * is used. This can only make sense for arrays with redundancy.
5849 * linear and raid0 always use whatever space is available. We can only
5850 * consider changing this number if no resync or reconstruction is
5851 * happening, and if the new size is acceptable. It must fit before the
5852 * sb_start or, if that is <data_offset, it must fit before the size
5853 * of each device. If num_sectors is zero, we find the largest size
5856 if (mddev
->sync_thread
)
5859 /* Sorry, cannot grow a bitmap yet, just remove it,
5863 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5864 sector_t avail
= rdev
->sectors
;
5866 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5867 num_sectors
= avail
;
5868 if (avail
< num_sectors
)
5871 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5873 revalidate_disk(mddev
->gendisk
);
5877 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
5880 /* change the number of raid disks */
5881 if (mddev
->pers
->check_reshape
== NULL
)
5883 if (raid_disks
<= 0 ||
5884 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5886 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5888 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5890 rv
= mddev
->pers
->check_reshape(mddev
);
5892 mddev
->delta_disks
= 0;
5898 * update_array_info is used to change the configuration of an
5900 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5901 * fields in the info are checked against the array.
5902 * Any differences that cannot be handled will cause an error.
5903 * Normally, only one change can be managed at a time.
5905 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
5911 /* calculate expected state,ignoring low bits */
5912 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5913 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5915 if (mddev
->major_version
!= info
->major_version
||
5916 mddev
->minor_version
!= info
->minor_version
||
5917 /* mddev->patch_version != info->patch_version || */
5918 mddev
->ctime
!= info
->ctime
||
5919 mddev
->level
!= info
->level
||
5920 /* mddev->layout != info->layout || */
5921 !mddev
->persistent
!= info
->not_persistent
||
5922 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5923 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5924 ((state
^info
->state
) & 0xfffffe00)
5927 /* Check there is only one change */
5928 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5930 if (mddev
->raid_disks
!= info
->raid_disks
)
5932 if (mddev
->layout
!= info
->layout
)
5934 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5941 if (mddev
->layout
!= info
->layout
) {
5943 * we don't need to do anything at the md level, the
5944 * personality will take care of it all.
5946 if (mddev
->pers
->check_reshape
== NULL
)
5949 mddev
->new_layout
= info
->layout
;
5950 rv
= mddev
->pers
->check_reshape(mddev
);
5952 mddev
->new_layout
= mddev
->layout
;
5956 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5957 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5959 if (mddev
->raid_disks
!= info
->raid_disks
)
5960 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5962 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5963 if (mddev
->pers
->quiesce
== NULL
)
5965 if (mddev
->recovery
|| mddev
->sync_thread
)
5967 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5968 /* add the bitmap */
5971 if (mddev
->bitmap_info
.default_offset
== 0)
5973 mddev
->bitmap_info
.offset
=
5974 mddev
->bitmap_info
.default_offset
;
5975 mddev
->pers
->quiesce(mddev
, 1);
5976 rv
= bitmap_create(mddev
);
5978 rv
= bitmap_load(mddev
);
5980 bitmap_destroy(mddev
);
5981 mddev
->pers
->quiesce(mddev
, 0);
5983 /* remove the bitmap */
5986 if (mddev
->bitmap
->file
)
5988 mddev
->pers
->quiesce(mddev
, 1);
5989 bitmap_destroy(mddev
);
5990 mddev
->pers
->quiesce(mddev
, 0);
5991 mddev
->bitmap_info
.offset
= 0;
5994 md_update_sb(mddev
, 1);
5998 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6000 struct md_rdev
*rdev
;
6002 if (mddev
->pers
== NULL
)
6005 rdev
= find_rdev(mddev
, dev
);
6009 md_error(mddev
, rdev
);
6010 if (!test_bit(Faulty
, &rdev
->flags
))
6016 * We have a problem here : there is no easy way to give a CHS
6017 * virtual geometry. We currently pretend that we have a 2 heads
6018 * 4 sectors (with a BIG number of cylinders...). This drives
6019 * dosfs just mad... ;-)
6021 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6023 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6027 geo
->cylinders
= mddev
->array_sectors
/ 8;
6031 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6032 unsigned int cmd
, unsigned long arg
)
6035 void __user
*argp
= (void __user
*)arg
;
6036 struct mddev
*mddev
= NULL
;
6039 if (!capable(CAP_SYS_ADMIN
))
6043 * Commands dealing with the RAID driver but not any
6049 err
= get_version(argp
);
6052 case PRINT_RAID_DEBUG
:
6060 autostart_arrays(arg
);
6067 * Commands creating/starting a new array:
6070 mddev
= bdev
->bd_disk
->private_data
;
6077 err
= mddev_lock(mddev
);
6080 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6087 case SET_ARRAY_INFO
:
6089 mdu_array_info_t info
;
6091 memset(&info
, 0, sizeof(info
));
6092 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6097 err
= update_array_info(mddev
, &info
);
6099 printk(KERN_WARNING
"md: couldn't update"
6100 " array info. %d\n", err
);
6105 if (!list_empty(&mddev
->disks
)) {
6107 "md: array %s already has disks!\n",
6112 if (mddev
->raid_disks
) {
6114 "md: array %s already initialised!\n",
6119 err
= set_array_info(mddev
, &info
);
6121 printk(KERN_WARNING
"md: couldn't set"
6122 " array info. %d\n", err
);
6132 * Commands querying/configuring an existing array:
6134 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6135 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6136 if ((!mddev
->raid_disks
&& !mddev
->external
)
6137 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6138 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6139 && cmd
!= GET_BITMAP_FILE
) {
6145 * Commands even a read-only array can execute:
6149 case GET_ARRAY_INFO
:
6150 err
= get_array_info(mddev
, argp
);
6153 case GET_BITMAP_FILE
:
6154 err
= get_bitmap_file(mddev
, argp
);
6158 err
= get_disk_info(mddev
, argp
);
6161 case RESTART_ARRAY_RW
:
6162 err
= restart_array(mddev
);
6166 err
= do_md_stop(mddev
, 0, 1);
6170 err
= md_set_readonly(mddev
, 1);
6174 if (get_user(ro
, (int __user
*)(arg
))) {
6180 /* if the bdev is going readonly the value of mddev->ro
6181 * does not matter, no writes are coming
6186 /* are we are already prepared for writes? */
6190 /* transitioning to readauto need only happen for
6191 * arrays that call md_write_start
6194 err
= restart_array(mddev
);
6197 set_disk_ro(mddev
->gendisk
, 0);
6204 * The remaining ioctls are changing the state of the
6205 * superblock, so we do not allow them on read-only arrays.
6206 * However non-MD ioctls (e.g. get-size) will still come through
6207 * here and hit the 'default' below, so only disallow
6208 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6210 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6211 if (mddev
->ro
== 2) {
6213 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6214 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6215 md_wakeup_thread(mddev
->thread
);
6226 mdu_disk_info_t info
;
6227 if (copy_from_user(&info
, argp
, sizeof(info
)))
6230 err
= add_new_disk(mddev
, &info
);
6234 case HOT_REMOVE_DISK
:
6235 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6239 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6242 case SET_DISK_FAULTY
:
6243 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6247 err
= do_md_run(mddev
);
6250 case SET_BITMAP_FILE
:
6251 err
= set_bitmap_file(mddev
, (int)arg
);
6261 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6263 mddev
->hold_active
= 0;
6264 mddev_unlock(mddev
);
6273 #ifdef CONFIG_COMPAT
6274 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6275 unsigned int cmd
, unsigned long arg
)
6278 case HOT_REMOVE_DISK
:
6280 case SET_DISK_FAULTY
:
6281 case SET_BITMAP_FILE
:
6282 /* These take in integer arg, do not convert */
6285 arg
= (unsigned long)compat_ptr(arg
);
6289 return md_ioctl(bdev
, mode
, cmd
, arg
);
6291 #endif /* CONFIG_COMPAT */
6293 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6296 * Succeed if we can lock the mddev, which confirms that
6297 * it isn't being stopped right now.
6299 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6302 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6303 /* we are racing with mddev_put which is discarding this
6307 /* Wait until bdev->bd_disk is definitely gone */
6308 flush_workqueue(md_misc_wq
);
6309 /* Then retry the open from the top */
6310 return -ERESTARTSYS
;
6312 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6314 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6318 atomic_inc(&mddev
->openers
);
6319 mutex_unlock(&mddev
->open_mutex
);
6321 check_disk_change(bdev
);
6326 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6328 struct mddev
*mddev
= disk
->private_data
;
6331 atomic_dec(&mddev
->openers
);
6337 static int md_media_changed(struct gendisk
*disk
)
6339 struct mddev
*mddev
= disk
->private_data
;
6341 return mddev
->changed
;
6344 static int md_revalidate(struct gendisk
*disk
)
6346 struct mddev
*mddev
= disk
->private_data
;
6351 static const struct block_device_operations md_fops
=
6353 .owner
= THIS_MODULE
,
6355 .release
= md_release
,
6357 #ifdef CONFIG_COMPAT
6358 .compat_ioctl
= md_compat_ioctl
,
6360 .getgeo
= md_getgeo
,
6361 .media_changed
= md_media_changed
,
6362 .revalidate_disk
= md_revalidate
,
6365 static int md_thread(void * arg
)
6367 struct md_thread
*thread
= arg
;
6370 * md_thread is a 'system-thread', it's priority should be very
6371 * high. We avoid resource deadlocks individually in each
6372 * raid personality. (RAID5 does preallocation) We also use RR and
6373 * the very same RT priority as kswapd, thus we will never get
6374 * into a priority inversion deadlock.
6376 * we definitely have to have equal or higher priority than
6377 * bdflush, otherwise bdflush will deadlock if there are too
6378 * many dirty RAID5 blocks.
6381 allow_signal(SIGKILL
);
6382 while (!kthread_should_stop()) {
6384 /* We need to wait INTERRUPTIBLE so that
6385 * we don't add to the load-average.
6386 * That means we need to be sure no signals are
6389 if (signal_pending(current
))
6390 flush_signals(current
);
6392 wait_event_interruptible_timeout
6394 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6395 || kthread_should_stop(),
6398 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6399 if (!kthread_should_stop())
6400 thread
->run(thread
->mddev
);
6406 void md_wakeup_thread(struct md_thread
*thread
)
6409 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6410 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6411 wake_up(&thread
->wqueue
);
6415 struct md_thread
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6418 struct md_thread
*thread
;
6420 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6424 init_waitqueue_head(&thread
->wqueue
);
6427 thread
->mddev
= mddev
;
6428 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6429 thread
->tsk
= kthread_run(md_thread
, thread
,
6431 mdname(thread
->mddev
),
6432 name
?: mddev
->pers
->name
);
6433 if (IS_ERR(thread
->tsk
)) {
6440 void md_unregister_thread(struct md_thread
**threadp
)
6442 struct md_thread
*thread
= *threadp
;
6445 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6446 /* Locking ensures that mddev_unlock does not wake_up a
6447 * non-existent thread
6449 spin_lock(&pers_lock
);
6451 spin_unlock(&pers_lock
);
6453 kthread_stop(thread
->tsk
);
6457 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6464 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6467 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6469 mddev
->pers
->error_handler(mddev
,rdev
);
6470 if (mddev
->degraded
)
6471 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6472 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6473 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6474 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6475 md_wakeup_thread(mddev
->thread
);
6476 if (mddev
->event_work
.func
)
6477 queue_work(md_misc_wq
, &mddev
->event_work
);
6478 md_new_event_inintr(mddev
);
6481 /* seq_file implementation /proc/mdstat */
6483 static void status_unused(struct seq_file
*seq
)
6486 struct md_rdev
*rdev
;
6488 seq_printf(seq
, "unused devices: ");
6490 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6491 char b
[BDEVNAME_SIZE
];
6493 seq_printf(seq
, "%s ",
6494 bdevname(rdev
->bdev
,b
));
6497 seq_printf(seq
, "<none>");
6499 seq_printf(seq
, "\n");
6503 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6505 sector_t max_sectors
, resync
, res
;
6506 unsigned long dt
, db
;
6509 unsigned int per_milli
;
6511 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6513 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6514 max_sectors
= mddev
->resync_max_sectors
;
6516 max_sectors
= mddev
->dev_sectors
;
6519 * Should not happen.
6525 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6526 * in a sector_t, and (max_sectors>>scale) will fit in a
6527 * u32, as those are the requirements for sector_div.
6528 * Thus 'scale' must be at least 10
6531 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6532 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6535 res
= (resync
>>scale
)*1000;
6536 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6540 int i
, x
= per_milli
/50, y
= 20-x
;
6541 seq_printf(seq
, "[");
6542 for (i
= 0; i
< x
; i
++)
6543 seq_printf(seq
, "=");
6544 seq_printf(seq
, ">");
6545 for (i
= 0; i
< y
; i
++)
6546 seq_printf(seq
, ".");
6547 seq_printf(seq
, "] ");
6549 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6550 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6552 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6554 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6555 "resync" : "recovery"))),
6556 per_milli
/10, per_milli
% 10,
6557 (unsigned long long) resync
/2,
6558 (unsigned long long) max_sectors
/2);
6561 * dt: time from mark until now
6562 * db: blocks written from mark until now
6563 * rt: remaining time
6565 * rt is a sector_t, so could be 32bit or 64bit.
6566 * So we divide before multiply in case it is 32bit and close
6568 * We scale the divisor (db) by 32 to avoid losing precision
6569 * near the end of resync when the number of remaining sectors
6571 * We then divide rt by 32 after multiplying by db to compensate.
6572 * The '+1' avoids division by zero if db is very small.
6574 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6576 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6577 - mddev
->resync_mark_cnt
;
6579 rt
= max_sectors
- resync
; /* number of remaining sectors */
6580 sector_div(rt
, db
/32+1);
6584 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6585 ((unsigned long)rt
% 60)/6);
6587 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6590 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6592 struct list_head
*tmp
;
6594 struct mddev
*mddev
;
6602 spin_lock(&all_mddevs_lock
);
6603 list_for_each(tmp
,&all_mddevs
)
6605 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6607 spin_unlock(&all_mddevs_lock
);
6610 spin_unlock(&all_mddevs_lock
);
6612 return (void*)2;/* tail */
6616 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6618 struct list_head
*tmp
;
6619 struct mddev
*next_mddev
, *mddev
= v
;
6625 spin_lock(&all_mddevs_lock
);
6627 tmp
= all_mddevs
.next
;
6629 tmp
= mddev
->all_mddevs
.next
;
6630 if (tmp
!= &all_mddevs
)
6631 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6633 next_mddev
= (void*)2;
6636 spin_unlock(&all_mddevs_lock
);
6644 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6646 struct mddev
*mddev
= v
;
6648 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6652 static int md_seq_show(struct seq_file
*seq
, void *v
)
6654 struct mddev
*mddev
= v
;
6656 struct md_rdev
*rdev
;
6657 struct bitmap
*bitmap
;
6659 if (v
== (void*)1) {
6660 struct md_personality
*pers
;
6661 seq_printf(seq
, "Personalities : ");
6662 spin_lock(&pers_lock
);
6663 list_for_each_entry(pers
, &pers_list
, list
)
6664 seq_printf(seq
, "[%s] ", pers
->name
);
6666 spin_unlock(&pers_lock
);
6667 seq_printf(seq
, "\n");
6668 seq
->poll_event
= atomic_read(&md_event_count
);
6671 if (v
== (void*)2) {
6676 if (mddev_lock(mddev
) < 0)
6679 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6680 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6681 mddev
->pers
? "" : "in");
6684 seq_printf(seq
, " (read-only)");
6686 seq_printf(seq
, " (auto-read-only)");
6687 seq_printf(seq
, " %s", mddev
->pers
->name
);
6691 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6692 char b
[BDEVNAME_SIZE
];
6693 seq_printf(seq
, " %s[%d]",
6694 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6695 if (test_bit(WriteMostly
, &rdev
->flags
))
6696 seq_printf(seq
, "(W)");
6697 if (test_bit(Faulty
, &rdev
->flags
)) {
6698 seq_printf(seq
, "(F)");
6700 } else if (rdev
->raid_disk
< 0)
6701 seq_printf(seq
, "(S)"); /* spare */
6702 sectors
+= rdev
->sectors
;
6705 if (!list_empty(&mddev
->disks
)) {
6707 seq_printf(seq
, "\n %llu blocks",
6708 (unsigned long long)
6709 mddev
->array_sectors
/ 2);
6711 seq_printf(seq
, "\n %llu blocks",
6712 (unsigned long long)sectors
/ 2);
6714 if (mddev
->persistent
) {
6715 if (mddev
->major_version
!= 0 ||
6716 mddev
->minor_version
!= 90) {
6717 seq_printf(seq
," super %d.%d",
6718 mddev
->major_version
,
6719 mddev
->minor_version
);
6721 } else if (mddev
->external
)
6722 seq_printf(seq
, " super external:%s",
6723 mddev
->metadata_type
);
6725 seq_printf(seq
, " super non-persistent");
6728 mddev
->pers
->status(seq
, mddev
);
6729 seq_printf(seq
, "\n ");
6730 if (mddev
->pers
->sync_request
) {
6731 if (mddev
->curr_resync
> 2) {
6732 status_resync(seq
, mddev
);
6733 seq_printf(seq
, "\n ");
6734 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6735 seq_printf(seq
, "\tresync=DELAYED\n ");
6736 else if (mddev
->recovery_cp
< MaxSector
)
6737 seq_printf(seq
, "\tresync=PENDING\n ");
6740 seq_printf(seq
, "\n ");
6742 if ((bitmap
= mddev
->bitmap
)) {
6743 unsigned long chunk_kb
;
6744 unsigned long flags
;
6745 spin_lock_irqsave(&bitmap
->lock
, flags
);
6746 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6747 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6749 bitmap
->pages
- bitmap
->missing_pages
,
6751 (bitmap
->pages
- bitmap
->missing_pages
)
6752 << (PAGE_SHIFT
- 10),
6753 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6754 chunk_kb
? "KB" : "B");
6756 seq_printf(seq
, ", file: ");
6757 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6760 seq_printf(seq
, "\n");
6761 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6764 seq_printf(seq
, "\n");
6766 mddev_unlock(mddev
);
6771 static const struct seq_operations md_seq_ops
= {
6772 .start
= md_seq_start
,
6773 .next
= md_seq_next
,
6774 .stop
= md_seq_stop
,
6775 .show
= md_seq_show
,
6778 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6780 struct seq_file
*seq
;
6783 error
= seq_open(file
, &md_seq_ops
);
6787 seq
= file
->private_data
;
6788 seq
->poll_event
= atomic_read(&md_event_count
);
6792 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6794 struct seq_file
*seq
= filp
->private_data
;
6797 poll_wait(filp
, &md_event_waiters
, wait
);
6799 /* always allow read */
6800 mask
= POLLIN
| POLLRDNORM
;
6802 if (seq
->poll_event
!= atomic_read(&md_event_count
))
6803 mask
|= POLLERR
| POLLPRI
;
6807 static const struct file_operations md_seq_fops
= {
6808 .owner
= THIS_MODULE
,
6809 .open
= md_seq_open
,
6811 .llseek
= seq_lseek
,
6812 .release
= seq_release_private
,
6813 .poll
= mdstat_poll
,
6816 int register_md_personality(struct md_personality
*p
)
6818 spin_lock(&pers_lock
);
6819 list_add_tail(&p
->list
, &pers_list
);
6820 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6821 spin_unlock(&pers_lock
);
6825 int unregister_md_personality(struct md_personality
*p
)
6827 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6828 spin_lock(&pers_lock
);
6829 list_del_init(&p
->list
);
6830 spin_unlock(&pers_lock
);
6834 static int is_mddev_idle(struct mddev
*mddev
, int init
)
6836 struct md_rdev
* rdev
;
6842 rdev_for_each_rcu(rdev
, mddev
) {
6843 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6844 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6845 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6846 atomic_read(&disk
->sync_io
);
6847 /* sync IO will cause sync_io to increase before the disk_stats
6848 * as sync_io is counted when a request starts, and
6849 * disk_stats is counted when it completes.
6850 * So resync activity will cause curr_events to be smaller than
6851 * when there was no such activity.
6852 * non-sync IO will cause disk_stat to increase without
6853 * increasing sync_io so curr_events will (eventually)
6854 * be larger than it was before. Once it becomes
6855 * substantially larger, the test below will cause
6856 * the array to appear non-idle, and resync will slow
6858 * If there is a lot of outstanding resync activity when
6859 * we set last_event to curr_events, then all that activity
6860 * completing might cause the array to appear non-idle
6861 * and resync will be slowed down even though there might
6862 * not have been non-resync activity. This will only
6863 * happen once though. 'last_events' will soon reflect
6864 * the state where there is little or no outstanding
6865 * resync requests, and further resync activity will
6866 * always make curr_events less than last_events.
6869 if (init
|| curr_events
- rdev
->last_events
> 64) {
6870 rdev
->last_events
= curr_events
;
6878 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
6880 /* another "blocks" (512byte) blocks have been synced */
6881 atomic_sub(blocks
, &mddev
->recovery_active
);
6882 wake_up(&mddev
->recovery_wait
);
6884 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6885 md_wakeup_thread(mddev
->thread
);
6886 // stop recovery, signal do_sync ....
6891 /* md_write_start(mddev, bi)
6892 * If we need to update some array metadata (e.g. 'active' flag
6893 * in superblock) before writing, schedule a superblock update
6894 * and wait for it to complete.
6896 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
6899 if (bio_data_dir(bi
) != WRITE
)
6902 BUG_ON(mddev
->ro
== 1);
6903 if (mddev
->ro
== 2) {
6904 /* need to switch to read/write */
6906 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6907 md_wakeup_thread(mddev
->thread
);
6908 md_wakeup_thread(mddev
->sync_thread
);
6911 atomic_inc(&mddev
->writes_pending
);
6912 if (mddev
->safemode
== 1)
6913 mddev
->safemode
= 0;
6914 if (mddev
->in_sync
) {
6915 spin_lock_irq(&mddev
->write_lock
);
6916 if (mddev
->in_sync
) {
6918 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6919 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6920 md_wakeup_thread(mddev
->thread
);
6923 spin_unlock_irq(&mddev
->write_lock
);
6926 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6927 wait_event(mddev
->sb_wait
,
6928 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6931 void md_write_end(struct mddev
*mddev
)
6933 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6934 if (mddev
->safemode
== 2)
6935 md_wakeup_thread(mddev
->thread
);
6936 else if (mddev
->safemode_delay
)
6937 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6941 /* md_allow_write(mddev)
6942 * Calling this ensures that the array is marked 'active' so that writes
6943 * may proceed without blocking. It is important to call this before
6944 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6945 * Must be called with mddev_lock held.
6947 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6948 * is dropped, so return -EAGAIN after notifying userspace.
6950 int md_allow_write(struct mddev
*mddev
)
6956 if (!mddev
->pers
->sync_request
)
6959 spin_lock_irq(&mddev
->write_lock
);
6960 if (mddev
->in_sync
) {
6962 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6963 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6964 if (mddev
->safemode_delay
&&
6965 mddev
->safemode
== 0)
6966 mddev
->safemode
= 1;
6967 spin_unlock_irq(&mddev
->write_lock
);
6968 md_update_sb(mddev
, 0);
6969 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6971 spin_unlock_irq(&mddev
->write_lock
);
6973 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6978 EXPORT_SYMBOL_GPL(md_allow_write
);
6980 #define SYNC_MARKS 10
6981 #define SYNC_MARK_STEP (3*HZ)
6982 void md_do_sync(struct mddev
*mddev
)
6984 struct mddev
*mddev2
;
6985 unsigned int currspeed
= 0,
6987 sector_t max_sectors
,j
, io_sectors
;
6988 unsigned long mark
[SYNC_MARKS
];
6989 sector_t mark_cnt
[SYNC_MARKS
];
6991 struct list_head
*tmp
;
6992 sector_t last_check
;
6994 struct md_rdev
*rdev
;
6997 /* just incase thread restarts... */
6998 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7000 if (mddev
->ro
) /* never try to sync a read-only array */
7003 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7004 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7005 desc
= "data-check";
7006 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7007 desc
= "requested-resync";
7010 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7015 /* we overload curr_resync somewhat here.
7016 * 0 == not engaged in resync at all
7017 * 2 == checking that there is no conflict with another sync
7018 * 1 == like 2, but have yielded to allow conflicting resync to
7020 * other == active in resync - this many blocks
7022 * Before starting a resync we must have set curr_resync to
7023 * 2, and then checked that every "conflicting" array has curr_resync
7024 * less than ours. When we find one that is the same or higher
7025 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7026 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7027 * This will mean we have to start checking from the beginning again.
7032 mddev
->curr_resync
= 2;
7035 if (kthread_should_stop())
7036 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7038 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7040 for_each_mddev(mddev2
, tmp
) {
7041 if (mddev2
== mddev
)
7043 if (!mddev
->parallel_resync
7044 && mddev2
->curr_resync
7045 && match_mddev_units(mddev
, mddev2
)) {
7047 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7048 /* arbitrarily yield */
7049 mddev
->curr_resync
= 1;
7050 wake_up(&resync_wait
);
7052 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7053 /* no need to wait here, we can wait the next
7054 * time 'round when curr_resync == 2
7057 /* We need to wait 'interruptible' so as not to
7058 * contribute to the load average, and not to
7059 * be caught by 'softlockup'
7061 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7062 if (!kthread_should_stop() &&
7063 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7064 printk(KERN_INFO
"md: delaying %s of %s"
7065 " until %s has finished (they"
7066 " share one or more physical units)\n",
7067 desc
, mdname(mddev
), mdname(mddev2
));
7069 if (signal_pending(current
))
7070 flush_signals(current
);
7072 finish_wait(&resync_wait
, &wq
);
7075 finish_wait(&resync_wait
, &wq
);
7078 } while (mddev
->curr_resync
< 2);
7081 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7082 /* resync follows the size requested by the personality,
7083 * which defaults to physical size, but can be virtual size
7085 max_sectors
= mddev
->resync_max_sectors
;
7086 mddev
->resync_mismatches
= 0;
7087 /* we don't use the checkpoint if there's a bitmap */
7088 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7089 j
= mddev
->resync_min
;
7090 else if (!mddev
->bitmap
)
7091 j
= mddev
->recovery_cp
;
7093 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7094 max_sectors
= mddev
->dev_sectors
;
7096 /* recovery follows the physical size of devices */
7097 max_sectors
= mddev
->dev_sectors
;
7100 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7101 if (rdev
->raid_disk
>= 0 &&
7102 !test_bit(Faulty
, &rdev
->flags
) &&
7103 !test_bit(In_sync
, &rdev
->flags
) &&
7104 rdev
->recovery_offset
< j
)
7105 j
= rdev
->recovery_offset
;
7109 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7110 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7111 " %d KB/sec/disk.\n", speed_min(mddev
));
7112 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7113 "(but not more than %d KB/sec) for %s.\n",
7114 speed_max(mddev
), desc
);
7116 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7119 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7121 mark_cnt
[m
] = io_sectors
;
7124 mddev
->resync_mark
= mark
[last_mark
];
7125 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7128 * Tune reconstruction:
7130 window
= 32*(PAGE_SIZE
/512);
7131 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7132 window
/2, (unsigned long long)max_sectors
/2);
7134 atomic_set(&mddev
->recovery_active
, 0);
7139 "md: resuming %s of %s from checkpoint.\n",
7140 desc
, mdname(mddev
));
7141 mddev
->curr_resync
= j
;
7143 mddev
->curr_resync_completed
= j
;
7145 while (j
< max_sectors
) {
7150 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7151 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7152 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7153 > (max_sectors
>> 4)) ||
7154 (j
- mddev
->curr_resync_completed
)*2
7155 >= mddev
->resync_max
- mddev
->curr_resync_completed
7157 /* time to update curr_resync_completed */
7158 wait_event(mddev
->recovery_wait
,
7159 atomic_read(&mddev
->recovery_active
) == 0);
7160 mddev
->curr_resync_completed
= j
;
7161 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7162 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7165 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7166 /* As this condition is controlled by user-space,
7167 * we can block indefinitely, so use '_interruptible'
7168 * to avoid triggering warnings.
7170 flush_signals(current
); /* just in case */
7171 wait_event_interruptible(mddev
->recovery_wait
,
7172 mddev
->resync_max
> j
7173 || kthread_should_stop());
7176 if (kthread_should_stop())
7179 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7180 currspeed
< speed_min(mddev
));
7182 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7186 if (!skipped
) { /* actual IO requested */
7187 io_sectors
+= sectors
;
7188 atomic_add(sectors
, &mddev
->recovery_active
);
7191 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7195 if (j
>1) mddev
->curr_resync
= j
;
7196 mddev
->curr_mark_cnt
= io_sectors
;
7197 if (last_check
== 0)
7198 /* this is the earliest that rebuild will be
7199 * visible in /proc/mdstat
7201 md_new_event(mddev
);
7203 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7206 last_check
= io_sectors
;
7208 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7210 int next
= (last_mark
+1) % SYNC_MARKS
;
7212 mddev
->resync_mark
= mark
[next
];
7213 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7214 mark
[next
] = jiffies
;
7215 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7220 if (kthread_should_stop())
7225 * this loop exits only if either when we are slower than
7226 * the 'hard' speed limit, or the system was IO-idle for
7228 * the system might be non-idle CPU-wise, but we only care
7229 * about not overloading the IO subsystem. (things like an
7230 * e2fsck being done on the RAID array should execute fast)
7234 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7235 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7237 if (currspeed
> speed_min(mddev
)) {
7238 if ((currspeed
> speed_max(mddev
)) ||
7239 !is_mddev_idle(mddev
, 0)) {
7245 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7247 * this also signals 'finished resyncing' to md_stop
7250 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7252 /* tell personality that we are finished */
7253 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7255 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7256 mddev
->curr_resync
> 2) {
7257 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7258 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7259 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7261 "md: checkpointing %s of %s.\n",
7262 desc
, mdname(mddev
));
7263 mddev
->recovery_cp
= mddev
->curr_resync
;
7266 mddev
->recovery_cp
= MaxSector
;
7268 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7269 mddev
->curr_resync
= MaxSector
;
7271 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7272 if (rdev
->raid_disk
>= 0 &&
7273 mddev
->delta_disks
>= 0 &&
7274 !test_bit(Faulty
, &rdev
->flags
) &&
7275 !test_bit(In_sync
, &rdev
->flags
) &&
7276 rdev
->recovery_offset
< mddev
->curr_resync
)
7277 rdev
->recovery_offset
= mddev
->curr_resync
;
7281 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7284 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7285 /* We completed so min/max setting can be forgotten if used. */
7286 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7287 mddev
->resync_min
= 0;
7288 mddev
->resync_max
= MaxSector
;
7289 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7290 mddev
->resync_min
= mddev
->curr_resync_completed
;
7291 mddev
->curr_resync
= 0;
7292 wake_up(&resync_wait
);
7293 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7294 md_wakeup_thread(mddev
->thread
);
7299 * got a signal, exit.
7302 "md: md_do_sync() got signal ... exiting\n");
7303 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7307 EXPORT_SYMBOL_GPL(md_do_sync
);
7309 static int remove_and_add_spares(struct mddev
*mddev
)
7311 struct md_rdev
*rdev
;
7314 mddev
->curr_resync_completed
= 0;
7316 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7317 if (rdev
->raid_disk
>= 0 &&
7318 !test_bit(Blocked
, &rdev
->flags
) &&
7319 (test_bit(Faulty
, &rdev
->flags
) ||
7320 ! test_bit(In_sync
, &rdev
->flags
)) &&
7321 atomic_read(&rdev
->nr_pending
)==0) {
7322 if (mddev
->pers
->hot_remove_disk(
7323 mddev
, rdev
->raid_disk
)==0) {
7324 sysfs_unlink_rdev(mddev
, rdev
);
7325 rdev
->raid_disk
= -1;
7329 if (mddev
->degraded
) {
7330 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7331 if (rdev
->raid_disk
>= 0 &&
7332 !test_bit(In_sync
, &rdev
->flags
) &&
7333 !test_bit(Faulty
, &rdev
->flags
))
7335 if (rdev
->raid_disk
< 0
7336 && !test_bit(Faulty
, &rdev
->flags
)) {
7337 rdev
->recovery_offset
= 0;
7339 hot_add_disk(mddev
, rdev
) == 0) {
7340 if (sysfs_link_rdev(mddev
, rdev
))
7341 /* failure here is OK */;
7343 md_new_event(mddev
);
7344 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7353 static void reap_sync_thread(struct mddev
*mddev
)
7355 struct md_rdev
*rdev
;
7357 /* resync has finished, collect result */
7358 md_unregister_thread(&mddev
->sync_thread
);
7359 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7360 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7362 /* activate any spares */
7363 if (mddev
->pers
->spare_active(mddev
))
7364 sysfs_notify(&mddev
->kobj
, NULL
,
7367 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7368 mddev
->pers
->finish_reshape
)
7369 mddev
->pers
->finish_reshape(mddev
);
7371 /* If array is no-longer degraded, then any saved_raid_disk
7372 * information must be scrapped. Also if any device is now
7373 * In_sync we must scrape the saved_raid_disk for that device
7374 * do the superblock for an incrementally recovered device
7377 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7378 if (!mddev
->degraded
||
7379 test_bit(In_sync
, &rdev
->flags
))
7380 rdev
->saved_raid_disk
= -1;
7382 md_update_sb(mddev
, 1);
7383 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7384 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7385 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7386 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7387 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7388 /* flag recovery needed just to double check */
7389 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7390 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7391 md_new_event(mddev
);
7392 if (mddev
->event_work
.func
)
7393 queue_work(md_misc_wq
, &mddev
->event_work
);
7397 * This routine is regularly called by all per-raid-array threads to
7398 * deal with generic issues like resync and super-block update.
7399 * Raid personalities that don't have a thread (linear/raid0) do not
7400 * need this as they never do any recovery or update the superblock.
7402 * It does not do any resync itself, but rather "forks" off other threads
7403 * to do that as needed.
7404 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7405 * "->recovery" and create a thread at ->sync_thread.
7406 * When the thread finishes it sets MD_RECOVERY_DONE
7407 * and wakeups up this thread which will reap the thread and finish up.
7408 * This thread also removes any faulty devices (with nr_pending == 0).
7410 * The overall approach is:
7411 * 1/ if the superblock needs updating, update it.
7412 * 2/ If a recovery thread is running, don't do anything else.
7413 * 3/ If recovery has finished, clean up, possibly marking spares active.
7414 * 4/ If there are any faulty devices, remove them.
7415 * 5/ If array is degraded, try to add spares devices
7416 * 6/ If array has spares or is not in-sync, start a resync thread.
7418 void md_check_recovery(struct mddev
*mddev
)
7420 if (mddev
->suspended
)
7424 bitmap_daemon_work(mddev
);
7426 if (signal_pending(current
)) {
7427 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7428 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7430 mddev
->safemode
= 2;
7432 flush_signals(current
);
7435 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7438 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7439 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7440 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7441 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7442 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7443 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7447 if (mddev_trylock(mddev
)) {
7451 /* Only thing we do on a ro array is remove
7454 struct md_rdev
*rdev
;
7455 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7456 if (rdev
->raid_disk
>= 0 &&
7457 !test_bit(Blocked
, &rdev
->flags
) &&
7458 test_bit(Faulty
, &rdev
->flags
) &&
7459 atomic_read(&rdev
->nr_pending
)==0) {
7460 if (mddev
->pers
->hot_remove_disk(
7461 mddev
, rdev
->raid_disk
)==0) {
7462 sysfs_unlink_rdev(mddev
, rdev
);
7463 rdev
->raid_disk
= -1;
7466 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7470 if (!mddev
->external
) {
7472 spin_lock_irq(&mddev
->write_lock
);
7473 if (mddev
->safemode
&&
7474 !atomic_read(&mddev
->writes_pending
) &&
7476 mddev
->recovery_cp
== MaxSector
) {
7479 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7481 if (mddev
->safemode
== 1)
7482 mddev
->safemode
= 0;
7483 spin_unlock_irq(&mddev
->write_lock
);
7485 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7489 md_update_sb(mddev
, 0);
7491 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7492 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7493 /* resync/recovery still happening */
7494 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7497 if (mddev
->sync_thread
) {
7498 reap_sync_thread(mddev
);
7501 /* Set RUNNING before clearing NEEDED to avoid
7502 * any transients in the value of "sync_action".
7504 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7505 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7506 /* Clear some bits that don't mean anything, but
7509 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7510 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7512 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7514 /* no recovery is running.
7515 * remove any failed drives, then
7516 * add spares if possible.
7517 * Spare are also removed and re-added, to allow
7518 * the personality to fail the re-add.
7521 if (mddev
->reshape_position
!= MaxSector
) {
7522 if (mddev
->pers
->check_reshape
== NULL
||
7523 mddev
->pers
->check_reshape(mddev
) != 0)
7524 /* Cannot proceed */
7526 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7527 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7528 } else if ((spares
= remove_and_add_spares(mddev
))) {
7529 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7530 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7531 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7532 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7533 } else if (mddev
->recovery_cp
< MaxSector
) {
7534 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7535 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7536 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7537 /* nothing to be done ... */
7540 if (mddev
->pers
->sync_request
) {
7541 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7542 /* We are adding a device or devices to an array
7543 * which has the bitmap stored on all devices.
7544 * So make sure all bitmap pages get written
7546 bitmap_write_all(mddev
->bitmap
);
7548 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7551 if (!mddev
->sync_thread
) {
7552 printk(KERN_ERR
"%s: could not start resync"
7555 /* leave the spares where they are, it shouldn't hurt */
7556 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7557 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7558 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7559 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7560 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7562 md_wakeup_thread(mddev
->sync_thread
);
7563 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7564 md_new_event(mddev
);
7567 if (!mddev
->sync_thread
) {
7568 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7569 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7571 if (mddev
->sysfs_action
)
7572 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7574 mddev_unlock(mddev
);
7578 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7580 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7581 wait_event_timeout(rdev
->blocked_wait
,
7582 !test_bit(Blocked
, &rdev
->flags
) &&
7583 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7584 msecs_to_jiffies(5000));
7585 rdev_dec_pending(rdev
, mddev
);
7587 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7590 /* Bad block management.
7591 * We can record which blocks on each device are 'bad' and so just
7592 * fail those blocks, or that stripe, rather than the whole device.
7593 * Entries in the bad-block table are 64bits wide. This comprises:
7594 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7595 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7596 * A 'shift' can be set so that larger blocks are tracked and
7597 * consequently larger devices can be covered.
7598 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7600 * Locking of the bad-block table uses a seqlock so md_is_badblock
7601 * might need to retry if it is very unlucky.
7602 * We will sometimes want to check for bad blocks in a bi_end_io function,
7603 * so we use the write_seqlock_irq variant.
7605 * When looking for a bad block we specify a range and want to
7606 * know if any block in the range is bad. So we binary-search
7607 * to the last range that starts at-or-before the given endpoint,
7608 * (or "before the sector after the target range")
7609 * then see if it ends after the given start.
7611 * 0 if there are no known bad blocks in the range
7612 * 1 if there are known bad block which are all acknowledged
7613 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7614 * plus the start/length of the first bad section we overlap.
7616 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7617 sector_t
*first_bad
, int *bad_sectors
)
7623 sector_t target
= s
+ sectors
;
7626 if (bb
->shift
> 0) {
7627 /* round the start down, and the end up */
7629 target
+= (1<<bb
->shift
) - 1;
7630 target
>>= bb
->shift
;
7631 sectors
= target
- s
;
7633 /* 'target' is now the first block after the bad range */
7636 seq
= read_seqbegin(&bb
->lock
);
7640 /* Binary search between lo and hi for 'target'
7641 * i.e. for the last range that starts before 'target'
7643 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7644 * are known not to be the last range before target.
7645 * VARIANT: hi-lo is the number of possible
7646 * ranges, and decreases until it reaches 1
7648 while (hi
- lo
> 1) {
7649 int mid
= (lo
+ hi
) / 2;
7650 sector_t a
= BB_OFFSET(p
[mid
]);
7652 /* This could still be the one, earlier ranges
7656 /* This and later ranges are definitely out. */
7659 /* 'lo' might be the last that started before target, but 'hi' isn't */
7661 /* need to check all range that end after 's' to see if
7662 * any are unacknowledged.
7665 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7666 if (BB_OFFSET(p
[lo
]) < target
) {
7667 /* starts before the end, and finishes after
7668 * the start, so they must overlap
7670 if (rv
!= -1 && BB_ACK(p
[lo
]))
7674 *first_bad
= BB_OFFSET(p
[lo
]);
7675 *bad_sectors
= BB_LEN(p
[lo
]);
7681 if (read_seqretry(&bb
->lock
, seq
))
7686 EXPORT_SYMBOL_GPL(md_is_badblock
);
7689 * Add a range of bad blocks to the table.
7690 * This might extend the table, or might contract it
7691 * if two adjacent ranges can be merged.
7692 * We binary-search to find the 'insertion' point, then
7693 * decide how best to handle it.
7695 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7703 /* badblocks are disabled */
7707 /* round the start down, and the end up */
7708 sector_t next
= s
+ sectors
;
7710 next
+= (1<<bb
->shift
) - 1;
7715 write_seqlock_irq(&bb
->lock
);
7720 /* Find the last range that starts at-or-before 's' */
7721 while (hi
- lo
> 1) {
7722 int mid
= (lo
+ hi
) / 2;
7723 sector_t a
= BB_OFFSET(p
[mid
]);
7729 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7733 /* we found a range that might merge with the start
7736 sector_t a
= BB_OFFSET(p
[lo
]);
7737 sector_t e
= a
+ BB_LEN(p
[lo
]);
7738 int ack
= BB_ACK(p
[lo
]);
7740 /* Yes, we can merge with a previous range */
7741 if (s
== a
&& s
+ sectors
>= e
)
7742 /* new range covers old */
7745 ack
= ack
&& acknowledged
;
7747 if (e
< s
+ sectors
)
7749 if (e
- a
<= BB_MAX_LEN
) {
7750 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
7753 /* does not all fit in one range,
7754 * make p[lo] maximal
7756 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
7757 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7763 if (sectors
&& hi
< bb
->count
) {
7764 /* 'hi' points to the first range that starts after 's'.
7765 * Maybe we can merge with the start of that range */
7766 sector_t a
= BB_OFFSET(p
[hi
]);
7767 sector_t e
= a
+ BB_LEN(p
[hi
]);
7768 int ack
= BB_ACK(p
[hi
]);
7769 if (a
<= s
+ sectors
) {
7770 /* merging is possible */
7771 if (e
<= s
+ sectors
) {
7776 ack
= ack
&& acknowledged
;
7779 if (e
- a
<= BB_MAX_LEN
) {
7780 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
7783 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7791 if (sectors
== 0 && hi
< bb
->count
) {
7792 /* we might be able to combine lo and hi */
7793 /* Note: 's' is at the end of 'lo' */
7794 sector_t a
= BB_OFFSET(p
[hi
]);
7795 int lolen
= BB_LEN(p
[lo
]);
7796 int hilen
= BB_LEN(p
[hi
]);
7797 int newlen
= lolen
+ hilen
- (s
- a
);
7798 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
7799 /* yes, we can combine them */
7800 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
7801 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
7802 memmove(p
+ hi
, p
+ hi
+ 1,
7803 (bb
->count
- hi
- 1) * 8);
7808 /* didn't merge (it all).
7809 * Need to add a range just before 'hi' */
7810 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7811 /* No room for more */
7815 int this_sectors
= sectors
;
7816 memmove(p
+ hi
+ 1, p
+ hi
,
7817 (bb
->count
- hi
) * 8);
7820 if (this_sectors
> BB_MAX_LEN
)
7821 this_sectors
= BB_MAX_LEN
;
7822 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
7823 sectors
-= this_sectors
;
7830 bb
->unacked_exist
= 1;
7831 write_sequnlock_irq(&bb
->lock
);
7836 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
7839 int rv
= md_set_badblocks(&rdev
->badblocks
,
7840 s
+ rdev
->data_offset
, sectors
, acknowledged
);
7842 /* Make sure they get written out promptly */
7843 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
7844 md_wakeup_thread(rdev
->mddev
->thread
);
7848 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
7851 * Remove a range of bad blocks from the table.
7852 * This may involve extending the table if we spilt a region,
7853 * but it must not fail. So if the table becomes full, we just
7854 * drop the remove request.
7856 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
7860 sector_t target
= s
+ sectors
;
7863 if (bb
->shift
> 0) {
7864 /* When clearing we round the start up and the end down.
7865 * This should not matter as the shift should align with
7866 * the block size and no rounding should ever be needed.
7867 * However it is better the think a block is bad when it
7868 * isn't than to think a block is not bad when it is.
7870 s
+= (1<<bb
->shift
) - 1;
7872 target
>>= bb
->shift
;
7873 sectors
= target
- s
;
7876 write_seqlock_irq(&bb
->lock
);
7881 /* Find the last range that starts before 'target' */
7882 while (hi
- lo
> 1) {
7883 int mid
= (lo
+ hi
) / 2;
7884 sector_t a
= BB_OFFSET(p
[mid
]);
7891 /* p[lo] is the last range that could overlap the
7892 * current range. Earlier ranges could also overlap,
7893 * but only this one can overlap the end of the range.
7895 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
7896 /* Partial overlap, leave the tail of this range */
7897 int ack
= BB_ACK(p
[lo
]);
7898 sector_t a
= BB_OFFSET(p
[lo
]);
7899 sector_t end
= a
+ BB_LEN(p
[lo
]);
7902 /* we need to split this range */
7903 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7907 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
7909 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
7912 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
7913 /* there is no longer an overlap */
7918 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7919 /* This range does overlap */
7920 if (BB_OFFSET(p
[lo
]) < s
) {
7921 /* Keep the early parts of this range. */
7922 int ack
= BB_ACK(p
[lo
]);
7923 sector_t start
= BB_OFFSET(p
[lo
]);
7924 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
7925 /* now low doesn't overlap, so.. */
7930 /* 'lo' is strictly before, 'hi' is strictly after,
7931 * anything between needs to be discarded
7934 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
7935 bb
->count
-= (hi
- lo
- 1);
7941 write_sequnlock_irq(&bb
->lock
);
7945 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
)
7947 return md_clear_badblocks(&rdev
->badblocks
,
7948 s
+ rdev
->data_offset
,
7951 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
7954 * Acknowledge all bad blocks in a list.
7955 * This only succeeds if ->changed is clear. It is used by
7956 * in-kernel metadata updates
7958 void md_ack_all_badblocks(struct badblocks
*bb
)
7960 if (bb
->page
== NULL
|| bb
->changed
)
7961 /* no point even trying */
7963 write_seqlock_irq(&bb
->lock
);
7965 if (bb
->changed
== 0) {
7968 for (i
= 0; i
< bb
->count
; i
++) {
7969 if (!BB_ACK(p
[i
])) {
7970 sector_t start
= BB_OFFSET(p
[i
]);
7971 int len
= BB_LEN(p
[i
]);
7972 p
[i
] = BB_MAKE(start
, len
, 1);
7975 bb
->unacked_exist
= 0;
7977 write_sequnlock_irq(&bb
->lock
);
7979 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
7981 /* sysfs access to bad-blocks list.
7982 * We present two files.
7983 * 'bad-blocks' lists sector numbers and lengths of ranges that
7984 * are recorded as bad. The list is truncated to fit within
7985 * the one-page limit of sysfs.
7986 * Writing "sector length" to this file adds an acknowledged
7988 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7989 * been acknowledged. Writing to this file adds bad blocks
7990 * without acknowledging them. This is largely for testing.
7994 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8005 seq
= read_seqbegin(&bb
->lock
);
8010 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8011 sector_t s
= BB_OFFSET(p
[i
]);
8012 unsigned int length
= BB_LEN(p
[i
]);
8013 int ack
= BB_ACK(p
[i
]);
8019 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8020 (unsigned long long)s
<< bb
->shift
,
8021 length
<< bb
->shift
);
8023 if (unack
&& len
== 0)
8024 bb
->unacked_exist
= 0;
8026 if (read_seqretry(&bb
->lock
, seq
))
8035 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8037 unsigned long long sector
;
8041 /* Allow clearing via sysfs *only* for testing/debugging.
8042 * Normally only a successful write may clear a badblock
8045 if (page
[0] == '-') {
8049 #endif /* DO_DEBUG */
8051 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8053 if (newline
!= '\n')
8065 md_clear_badblocks(bb
, sector
, length
);
8068 #endif /* DO_DEBUG */
8069 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8075 static int md_notify_reboot(struct notifier_block
*this,
8076 unsigned long code
, void *x
)
8078 struct list_head
*tmp
;
8079 struct mddev
*mddev
;
8082 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
8084 printk(KERN_INFO
"md: stopping all md devices.\n");
8086 for_each_mddev(mddev
, tmp
) {
8087 if (mddev_trylock(mddev
)) {
8088 /* Force a switch to readonly even array
8089 * appears to still be in use. Hence
8092 md_set_readonly(mddev
, 100);
8093 mddev_unlock(mddev
);
8098 * certain more exotic SCSI devices are known to be
8099 * volatile wrt too early system reboots. While the
8100 * right place to handle this issue is the given
8101 * driver, we do want to have a safe RAID driver ...
8109 static struct notifier_block md_notifier
= {
8110 .notifier_call
= md_notify_reboot
,
8112 .priority
= INT_MAX
, /* before any real devices */
8115 static void md_geninit(void)
8117 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8119 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8122 static int __init
md_init(void)
8126 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8130 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8134 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8137 if ((ret
= register_blkdev(0, "mdp")) < 0)
8141 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8142 md_probe
, NULL
, NULL
);
8143 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8144 md_probe
, NULL
, NULL
);
8146 register_reboot_notifier(&md_notifier
);
8147 raid_table_header
= register_sysctl_table(raid_root_table
);
8153 unregister_blkdev(MD_MAJOR
, "md");
8155 destroy_workqueue(md_misc_wq
);
8157 destroy_workqueue(md_wq
);
8165 * Searches all registered partitions for autorun RAID arrays
8169 static LIST_HEAD(all_detected_devices
);
8170 struct detected_devices_node
{
8171 struct list_head list
;
8175 void md_autodetect_dev(dev_t dev
)
8177 struct detected_devices_node
*node_detected_dev
;
8179 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8180 if (node_detected_dev
) {
8181 node_detected_dev
->dev
= dev
;
8182 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8184 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8185 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8190 static void autostart_arrays(int part
)
8192 struct md_rdev
*rdev
;
8193 struct detected_devices_node
*node_detected_dev
;
8195 int i_scanned
, i_passed
;
8200 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8202 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8204 node_detected_dev
= list_entry(all_detected_devices
.next
,
8205 struct detected_devices_node
, list
);
8206 list_del(&node_detected_dev
->list
);
8207 dev
= node_detected_dev
->dev
;
8208 kfree(node_detected_dev
);
8209 rdev
= md_import_device(dev
,0, 90);
8213 if (test_bit(Faulty
, &rdev
->flags
)) {
8217 set_bit(AutoDetected
, &rdev
->flags
);
8218 list_add(&rdev
->same_set
, &pending_raid_disks
);
8222 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8223 i_scanned
, i_passed
);
8225 autorun_devices(part
);
8228 #endif /* !MODULE */
8230 static __exit
void md_exit(void)
8232 struct mddev
*mddev
;
8233 struct list_head
*tmp
;
8235 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8236 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8238 unregister_blkdev(MD_MAJOR
,"md");
8239 unregister_blkdev(mdp_major
, "mdp");
8240 unregister_reboot_notifier(&md_notifier
);
8241 unregister_sysctl_table(raid_table_header
);
8242 remove_proc_entry("mdstat", NULL
);
8243 for_each_mddev(mddev
, tmp
) {
8244 export_array(mddev
);
8245 mddev
->hold_active
= 0;
8247 destroy_workqueue(md_misc_wq
);
8248 destroy_workqueue(md_wq
);
8251 subsys_initcall(md_init
);
8252 module_exit(md_exit
)
8254 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8256 return sprintf(buffer
, "%d", start_readonly
);
8258 static int set_ro(const char *val
, struct kernel_param
*kp
)
8261 int num
= simple_strtoul(val
, &e
, 10);
8262 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8263 start_readonly
= num
;
8269 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8270 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8272 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8274 EXPORT_SYMBOL(register_md_personality
);
8275 EXPORT_SYMBOL(unregister_md_personality
);
8276 EXPORT_SYMBOL(md_error
);
8277 EXPORT_SYMBOL(md_done_sync
);
8278 EXPORT_SYMBOL(md_write_start
);
8279 EXPORT_SYMBOL(md_write_end
);
8280 EXPORT_SYMBOL(md_register_thread
);
8281 EXPORT_SYMBOL(md_unregister_thread
);
8282 EXPORT_SYMBOL(md_wakeup_thread
);
8283 EXPORT_SYMBOL(md_check_recovery
);
8284 MODULE_LICENSE("GPL");
8285 MODULE_DESCRIPTION("MD RAID framework");
8287 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);