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/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min
= 1000;
97 static int sysctl_speed_limit_max
= 200000;
98 static inline int speed_min(struct mddev
*mddev
)
100 return mddev
->sync_speed_min
?
101 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
104 static inline int speed_max(struct mddev
*mddev
)
106 return mddev
->sync_speed_max
?
107 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
110 static struct ctl_table_header
*raid_table_header
;
112 static ctl_table raid_table
[] = {
114 .procname
= "speed_limit_min",
115 .data
= &sysctl_speed_limit_min
,
116 .maxlen
= sizeof(int),
117 .mode
= S_IRUGO
|S_IWUSR
,
118 .proc_handler
= proc_dointvec
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
134 .mode
= S_IRUGO
|S_IXUGO
,
140 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static const struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio
*bio
)
160 struct mddev
*mddev
, **mddevp
;
165 bio_free(bio
, mddev
->bio_set
);
168 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
172 struct mddev
**mddevp
;
174 if (!mddev
|| !mddev
->bio_set
)
175 return bio_alloc(gfp_mask
, nr_iovecs
);
177 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
183 b
->bi_destructor
= mddev_bio_destructor
;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
188 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
192 struct mddev
**mddevp
;
194 if (!mddev
|| !mddev
->bio_set
)
195 return bio_clone(bio
, gfp_mask
);
197 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
203 b
->bi_destructor
= mddev_bio_destructor
;
205 if (bio_integrity(bio
)) {
208 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
218 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
220 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
227 struct bio_vec
*bvec
;
231 if (offset
== 0 && size
== bio
->bi_size
)
234 bio
->bi_sector
+= offset
;
237 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
239 while (bio
->bi_idx
< bio
->bi_vcnt
&&
240 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
241 /* remove this whole bio_vec */
242 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
245 if (bio
->bi_idx
< bio
->bi_vcnt
) {
246 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
247 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
249 /* avoid any complications with bi_idx being non-zero*/
251 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
252 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
253 bio
->bi_vcnt
-= bio
->bi_idx
;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec
, bio
, i
) {
258 if (sofar
+ bvec
->bv_len
> size
)
259 bvec
->bv_len
= size
- sofar
;
260 if (bvec
->bv_len
== 0) {
264 sofar
+= bvec
->bv_len
;
267 EXPORT_SYMBOL_GPL(md_trim_bio
);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
280 static atomic_t md_event_count
;
281 void md_new_event(struct mddev
*mddev
)
283 atomic_inc(&md_event_count
);
284 wake_up(&md_event_waiters
);
286 EXPORT_SYMBOL_GPL(md_new_event
);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev
*mddev
)
293 atomic_inc(&md_event_count
);
294 wake_up(&md_event_waiters
);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs
);
302 static DEFINE_SPINLOCK(all_mddevs_lock
);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
337 const int rw
= bio_data_dir(bio
);
338 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 rv
= 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
);
384 /* mddev_suspend makes sure no new requests are submitted
385 * to the device, and that any requests that have been submitted
386 * are completely handled.
387 * Once ->stop is called and completes, the module will be completely
390 void mddev_suspend(struct mddev
*mddev
)
392 BUG_ON(mddev
->suspended
);
393 mddev
->suspended
= 1;
395 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
396 mddev
->pers
->quiesce(mddev
, 1);
398 EXPORT_SYMBOL_GPL(mddev_suspend
);
400 void mddev_resume(struct mddev
*mddev
)
402 mddev
->suspended
= 0;
403 wake_up(&mddev
->sb_wait
);
404 mddev
->pers
->quiesce(mddev
, 0);
406 md_wakeup_thread(mddev
->thread
);
407 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
409 EXPORT_SYMBOL_GPL(mddev_resume
);
411 int mddev_congested(struct mddev
*mddev
, int bits
)
413 return mddev
->suspended
;
415 EXPORT_SYMBOL(mddev_congested
);
418 * Generic flush handling for md
421 static void md_end_flush(struct bio
*bio
, int err
)
423 struct md_rdev
*rdev
= bio
->bi_private
;
424 struct mddev
*mddev
= rdev
->mddev
;
426 rdev_dec_pending(rdev
, mddev
);
428 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
429 /* The pre-request flush has finished */
430 queue_work(md_wq
, &mddev
->flush_work
);
435 static void md_submit_flush_data(struct work_struct
*ws
);
437 static void submit_flushes(struct work_struct
*ws
)
439 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
440 struct md_rdev
*rdev
;
442 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
443 atomic_set(&mddev
->flush_pending
, 1);
445 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
446 if (rdev
->raid_disk
>= 0 &&
447 !test_bit(Faulty
, &rdev
->flags
)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
453 atomic_inc(&rdev
->nr_pending
);
454 atomic_inc(&rdev
->nr_pending
);
456 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
457 bi
->bi_end_io
= md_end_flush
;
458 bi
->bi_private
= rdev
;
459 bi
->bi_bdev
= rdev
->bdev
;
460 atomic_inc(&mddev
->flush_pending
);
461 submit_bio(WRITE_FLUSH
, bi
);
463 rdev_dec_pending(rdev
, mddev
);
466 if (atomic_dec_and_test(&mddev
->flush_pending
))
467 queue_work(md_wq
, &mddev
->flush_work
);
470 static void md_submit_flush_data(struct work_struct
*ws
)
472 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
473 struct bio
*bio
= mddev
->flush_bio
;
475 if (bio
->bi_size
== 0)
476 /* an empty barrier - all done */
479 bio
->bi_rw
&= ~REQ_FLUSH
;
480 if (mddev
->pers
->make_request(mddev
, bio
))
481 generic_make_request(bio
);
484 mddev
->flush_bio
= NULL
;
485 wake_up(&mddev
->sb_wait
);
488 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
490 spin_lock_irq(&mddev
->write_lock
);
491 wait_event_lock_irq(mddev
->sb_wait
,
493 mddev
->write_lock
, /*nothing*/);
494 mddev
->flush_bio
= bio
;
495 spin_unlock_irq(&mddev
->write_lock
);
497 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
498 queue_work(md_wq
, &mddev
->flush_work
);
500 EXPORT_SYMBOL(md_flush_request
);
502 /* Support for plugging.
503 * This mirrors the plugging support in request_queue, but does not
504 * require having a whole queue or request structures.
505 * We allocate an md_plug_cb for each md device and each thread it gets
506 * plugged on. This links tot the private plug_handle structure in the
507 * personality data where we keep a count of the number of outstanding
508 * plugs so other code can see if a plug is active.
511 struct blk_plug_cb cb
;
515 static void plugger_unplug(struct blk_plug_cb
*cb
)
517 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
518 if (atomic_dec_and_test(&mdcb
->mddev
->plug_cnt
))
519 md_wakeup_thread(mdcb
->mddev
->thread
);
523 /* Check that an unplug wakeup will come shortly.
524 * If not, wakeup the md thread immediately
526 int mddev_check_plugged(struct mddev
*mddev
)
528 struct blk_plug
*plug
= current
->plug
;
529 struct md_plug_cb
*mdcb
;
534 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
535 if (mdcb
->cb
.callback
== plugger_unplug
&&
536 mdcb
->mddev
== mddev
) {
537 /* Already on the list, move to top */
538 if (mdcb
!= list_first_entry(&plug
->cb_list
,
541 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
545 /* Not currently on the callback list */
546 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
551 mdcb
->cb
.callback
= plugger_unplug
;
552 atomic_inc(&mddev
->plug_cnt
);
553 list_add(&mdcb
->cb
.list
, &plug
->cb_list
);
556 EXPORT_SYMBOL_GPL(mddev_check_plugged
);
558 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
560 atomic_inc(&mddev
->active
);
564 static void mddev_delayed_delete(struct work_struct
*ws
);
566 static void mddev_put(struct mddev
*mddev
)
568 struct bio_set
*bs
= NULL
;
570 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
572 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
573 mddev
->ctime
== 0 && !mddev
->hold_active
) {
574 /* Array is not configured at all, and not held active,
576 list_del(&mddev
->all_mddevs
);
578 mddev
->bio_set
= NULL
;
579 if (mddev
->gendisk
) {
580 /* We did a probe so need to clean up. Call
581 * queue_work inside the spinlock so that
582 * flush_workqueue() after mddev_find will
583 * succeed in waiting for the work to be done.
585 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
586 queue_work(md_misc_wq
, &mddev
->del_work
);
590 spin_unlock(&all_mddevs_lock
);
595 void mddev_init(struct mddev
*mddev
)
597 mutex_init(&mddev
->open_mutex
);
598 mutex_init(&mddev
->reconfig_mutex
);
599 mutex_init(&mddev
->bitmap_info
.mutex
);
600 INIT_LIST_HEAD(&mddev
->disks
);
601 INIT_LIST_HEAD(&mddev
->all_mddevs
);
602 init_timer(&mddev
->safemode_timer
);
603 atomic_set(&mddev
->active
, 1);
604 atomic_set(&mddev
->openers
, 0);
605 atomic_set(&mddev
->active_io
, 0);
606 atomic_set(&mddev
->plug_cnt
, 0);
607 spin_lock_init(&mddev
->write_lock
);
608 atomic_set(&mddev
->flush_pending
, 0);
609 init_waitqueue_head(&mddev
->sb_wait
);
610 init_waitqueue_head(&mddev
->recovery_wait
);
611 mddev
->reshape_position
= MaxSector
;
612 mddev
->resync_min
= 0;
613 mddev
->resync_max
= MaxSector
;
614 mddev
->level
= LEVEL_NONE
;
616 EXPORT_SYMBOL_GPL(mddev_init
);
618 static struct mddev
* mddev_find(dev_t unit
)
620 struct mddev
*mddev
, *new = NULL
;
622 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
623 unit
&= ~((1<<MdpMinorShift
)-1);
626 spin_lock(&all_mddevs_lock
);
629 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
630 if (mddev
->unit
== unit
) {
632 spin_unlock(&all_mddevs_lock
);
638 list_add(&new->all_mddevs
, &all_mddevs
);
639 spin_unlock(&all_mddevs_lock
);
640 new->hold_active
= UNTIL_IOCTL
;
644 /* find an unused unit number */
645 static int next_minor
= 512;
646 int start
= next_minor
;
650 dev
= MKDEV(MD_MAJOR
, next_minor
);
652 if (next_minor
> MINORMASK
)
654 if (next_minor
== start
) {
655 /* Oh dear, all in use. */
656 spin_unlock(&all_mddevs_lock
);
662 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
663 if (mddev
->unit
== dev
) {
669 new->md_minor
= MINOR(dev
);
670 new->hold_active
= UNTIL_STOP
;
671 list_add(&new->all_mddevs
, &all_mddevs
);
672 spin_unlock(&all_mddevs_lock
);
675 spin_unlock(&all_mddevs_lock
);
677 new = kzalloc(sizeof(*new), GFP_KERNEL
);
682 if (MAJOR(unit
) == MD_MAJOR
)
683 new->md_minor
= MINOR(unit
);
685 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
692 static inline int mddev_lock(struct mddev
* mddev
)
694 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
697 static inline int mddev_is_locked(struct mddev
*mddev
)
699 return mutex_is_locked(&mddev
->reconfig_mutex
);
702 static inline int mddev_trylock(struct mddev
* mddev
)
704 return mutex_trylock(&mddev
->reconfig_mutex
);
707 static struct attribute_group md_redundancy_group
;
709 static void mddev_unlock(struct mddev
* mddev
)
711 if (mddev
->to_remove
) {
712 /* These cannot be removed under reconfig_mutex as
713 * an access to the files will try to take reconfig_mutex
714 * while holding the file unremovable, which leads to
716 * So hold set sysfs_active while the remove in happeing,
717 * and anything else which might set ->to_remove or my
718 * otherwise change the sysfs namespace will fail with
719 * -EBUSY if sysfs_active is still set.
720 * We set sysfs_active under reconfig_mutex and elsewhere
721 * test it under the same mutex to ensure its correct value
724 struct attribute_group
*to_remove
= mddev
->to_remove
;
725 mddev
->to_remove
= NULL
;
726 mddev
->sysfs_active
= 1;
727 mutex_unlock(&mddev
->reconfig_mutex
);
729 if (mddev
->kobj
.sd
) {
730 if (to_remove
!= &md_redundancy_group
)
731 sysfs_remove_group(&mddev
->kobj
, to_remove
);
732 if (mddev
->pers
== NULL
||
733 mddev
->pers
->sync_request
== NULL
) {
734 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
735 if (mddev
->sysfs_action
)
736 sysfs_put(mddev
->sysfs_action
);
737 mddev
->sysfs_action
= NULL
;
740 mddev
->sysfs_active
= 0;
742 mutex_unlock(&mddev
->reconfig_mutex
);
744 /* was we've dropped the mutex we need a spinlock to
745 * make sur the thread doesn't disappear
747 spin_lock(&pers_lock
);
748 md_wakeup_thread(mddev
->thread
);
749 spin_unlock(&pers_lock
);
752 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
754 struct md_rdev
*rdev
;
756 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
757 if (rdev
->desc_nr
== nr
)
763 static struct md_rdev
* find_rdev(struct mddev
* mddev
, dev_t dev
)
765 struct md_rdev
*rdev
;
767 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
768 if (rdev
->bdev
->bd_dev
== dev
)
774 static struct mdk_personality
*find_pers(int level
, char *clevel
)
776 struct mdk_personality
*pers
;
777 list_for_each_entry(pers
, &pers_list
, list
) {
778 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
780 if (strcmp(pers
->name
, clevel
)==0)
786 /* return the offset of the super block in 512byte sectors */
787 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
789 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
790 return MD_NEW_SIZE_SECTORS(num_sectors
);
793 static int alloc_disk_sb(struct md_rdev
* rdev
)
798 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
799 if (!rdev
->sb_page
) {
800 printk(KERN_ALERT
"md: out of memory.\n");
807 static void free_disk_sb(struct md_rdev
* rdev
)
810 put_page(rdev
->sb_page
);
812 rdev
->sb_page
= NULL
;
817 put_page(rdev
->bb_page
);
818 rdev
->bb_page
= NULL
;
823 static void super_written(struct bio
*bio
, int error
)
825 struct md_rdev
*rdev
= bio
->bi_private
;
826 struct mddev
*mddev
= rdev
->mddev
;
828 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
829 printk("md: super_written gets error=%d, uptodate=%d\n",
830 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
831 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
832 md_error(mddev
, rdev
);
835 if (atomic_dec_and_test(&mddev
->pending_writes
))
836 wake_up(&mddev
->sb_wait
);
840 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
841 sector_t sector
, int size
, struct page
*page
)
843 /* write first size bytes of page to sector of rdev
844 * Increment mddev->pending_writes before returning
845 * and decrement it on completion, waking up sb_wait
846 * if zero is reached.
847 * If an error occurred, call md_error
849 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
851 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
852 bio
->bi_sector
= sector
;
853 bio_add_page(bio
, page
, size
, 0);
854 bio
->bi_private
= rdev
;
855 bio
->bi_end_io
= super_written
;
857 atomic_inc(&mddev
->pending_writes
);
858 submit_bio(WRITE_FLUSH_FUA
, bio
);
861 void md_super_wait(struct mddev
*mddev
)
863 /* wait for all superblock writes that were scheduled to complete */
866 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
867 if (atomic_read(&mddev
->pending_writes
)==0)
871 finish_wait(&mddev
->sb_wait
, &wq
);
874 static void bi_complete(struct bio
*bio
, int error
)
876 complete((struct completion
*)bio
->bi_private
);
879 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
880 struct page
*page
, int rw
, bool metadata_op
)
882 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
883 struct completion event
;
888 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
889 rdev
->meta_bdev
: rdev
->bdev
;
891 bio
->bi_sector
= sector
+ rdev
->sb_start
;
893 bio
->bi_sector
= sector
+ rdev
->data_offset
;
894 bio_add_page(bio
, page
, size
, 0);
895 init_completion(&event
);
896 bio
->bi_private
= &event
;
897 bio
->bi_end_io
= bi_complete
;
899 wait_for_completion(&event
);
901 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
905 EXPORT_SYMBOL_GPL(sync_page_io
);
907 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
909 char b
[BDEVNAME_SIZE
];
910 if (!rdev
->sb_page
) {
918 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
924 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
925 bdevname(rdev
->bdev
,b
));
929 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
931 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
932 sb1
->set_uuid1
== sb2
->set_uuid1
&&
933 sb1
->set_uuid2
== sb2
->set_uuid2
&&
934 sb1
->set_uuid3
== sb2
->set_uuid3
;
937 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
940 mdp_super_t
*tmp1
, *tmp2
;
942 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
943 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
945 if (!tmp1
|| !tmp2
) {
947 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
955 * nr_disks is not constant
960 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
968 static u32
md_csum_fold(u32 csum
)
970 csum
= (csum
& 0xffff) + (csum
>> 16);
971 return (csum
& 0xffff) + (csum
>> 16);
974 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
977 u32
*sb32
= (u32
*)sb
;
979 unsigned int disk_csum
, csum
;
981 disk_csum
= sb
->sb_csum
;
984 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
986 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
990 /* This used to use csum_partial, which was wrong for several
991 * reasons including that different results are returned on
992 * different architectures. It isn't critical that we get exactly
993 * the same return value as before (we always csum_fold before
994 * testing, and that removes any differences). However as we
995 * know that csum_partial always returned a 16bit value on
996 * alphas, do a fold to maximise conformity to previous behaviour.
998 sb
->sb_csum
= md_csum_fold(disk_csum
);
1000 sb
->sb_csum
= disk_csum
;
1007 * Handle superblock details.
1008 * We want to be able to handle multiple superblock formats
1009 * so we have a common interface to them all, and an array of
1010 * different handlers.
1011 * We rely on user-space to write the initial superblock, and support
1012 * reading and updating of superblocks.
1013 * Interface methods are:
1014 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1015 * loads and validates a superblock on dev.
1016 * if refdev != NULL, compare superblocks on both devices
1018 * 0 - dev has a superblock that is compatible with refdev
1019 * 1 - dev has a superblock that is compatible and newer than refdev
1020 * so dev should be used as the refdev in future
1021 * -EINVAL superblock incompatible or invalid
1022 * -othererror e.g. -EIO
1024 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1025 * Verify that dev is acceptable into mddev.
1026 * The first time, mddev->raid_disks will be 0, and data from
1027 * dev should be merged in. Subsequent calls check that dev
1028 * is new enough. Return 0 or -EINVAL
1030 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1031 * Update the superblock for rdev with data in mddev
1032 * This does not write to disc.
1038 struct module
*owner
;
1039 int (*load_super
)(struct md_rdev
*rdev
, struct md_rdev
*refdev
,
1041 int (*validate_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1042 void (*sync_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1043 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1044 sector_t num_sectors
);
1048 * Check that the given mddev has no bitmap.
1050 * This function is called from the run method of all personalities that do not
1051 * support bitmaps. It prints an error message and returns non-zero if mddev
1052 * has a bitmap. Otherwise, it returns 0.
1055 int md_check_no_bitmap(struct mddev
*mddev
)
1057 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1059 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1060 mdname(mddev
), mddev
->pers
->name
);
1063 EXPORT_SYMBOL(md_check_no_bitmap
);
1066 * load_super for 0.90.0
1068 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1070 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1075 * Calculate the position of the superblock (512byte sectors),
1076 * it's at the end of the disk.
1078 * It also happens to be a multiple of 4Kb.
1080 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1082 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1083 if (ret
) return ret
;
1087 bdevname(rdev
->bdev
, b
);
1088 sb
= page_address(rdev
->sb_page
);
1090 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1091 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1096 if (sb
->major_version
!= 0 ||
1097 sb
->minor_version
< 90 ||
1098 sb
->minor_version
> 91) {
1099 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1100 sb
->major_version
, sb
->minor_version
,
1105 if (sb
->raid_disks
<= 0)
1108 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1109 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1114 rdev
->preferred_minor
= sb
->md_minor
;
1115 rdev
->data_offset
= 0;
1116 rdev
->sb_size
= MD_SB_BYTES
;
1117 rdev
->badblocks
.shift
= -1;
1119 if (sb
->level
== LEVEL_MULTIPATH
)
1122 rdev
->desc_nr
= sb
->this_disk
.number
;
1128 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1129 if (!uuid_equal(refsb
, sb
)) {
1130 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1131 b
, bdevname(refdev
->bdev
,b2
));
1134 if (!sb_equal(refsb
, sb
)) {
1135 printk(KERN_WARNING
"md: %s has same UUID"
1136 " but different superblock to %s\n",
1137 b
, bdevname(refdev
->bdev
, b2
));
1141 ev2
= md_event(refsb
);
1147 rdev
->sectors
= rdev
->sb_start
;
1148 /* Limit to 4TB as metadata cannot record more than that */
1149 if (rdev
->sectors
>= (2ULL << 32))
1150 rdev
->sectors
= (2ULL << 32) - 2;
1152 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1153 /* "this cannot possibly happen" ... */
1161 * validate_super for 0.90.0
1163 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1166 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1167 __u64 ev1
= md_event(sb
);
1169 rdev
->raid_disk
= -1;
1170 clear_bit(Faulty
, &rdev
->flags
);
1171 clear_bit(In_sync
, &rdev
->flags
);
1172 clear_bit(WriteMostly
, &rdev
->flags
);
1174 if (mddev
->raid_disks
== 0) {
1175 mddev
->major_version
= 0;
1176 mddev
->minor_version
= sb
->minor_version
;
1177 mddev
->patch_version
= sb
->patch_version
;
1178 mddev
->external
= 0;
1179 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1180 mddev
->ctime
= sb
->ctime
;
1181 mddev
->utime
= sb
->utime
;
1182 mddev
->level
= sb
->level
;
1183 mddev
->clevel
[0] = 0;
1184 mddev
->layout
= sb
->layout
;
1185 mddev
->raid_disks
= sb
->raid_disks
;
1186 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1187 mddev
->events
= ev1
;
1188 mddev
->bitmap_info
.offset
= 0;
1189 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1191 if (mddev
->minor_version
>= 91) {
1192 mddev
->reshape_position
= sb
->reshape_position
;
1193 mddev
->delta_disks
= sb
->delta_disks
;
1194 mddev
->new_level
= sb
->new_level
;
1195 mddev
->new_layout
= sb
->new_layout
;
1196 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1198 mddev
->reshape_position
= MaxSector
;
1199 mddev
->delta_disks
= 0;
1200 mddev
->new_level
= mddev
->level
;
1201 mddev
->new_layout
= mddev
->layout
;
1202 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1205 if (sb
->state
& (1<<MD_SB_CLEAN
))
1206 mddev
->recovery_cp
= MaxSector
;
1208 if (sb
->events_hi
== sb
->cp_events_hi
&&
1209 sb
->events_lo
== sb
->cp_events_lo
) {
1210 mddev
->recovery_cp
= sb
->recovery_cp
;
1212 mddev
->recovery_cp
= 0;
1215 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1216 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1217 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1218 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1220 mddev
->max_disks
= MD_SB_DISKS
;
1222 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1223 mddev
->bitmap_info
.file
== NULL
)
1224 mddev
->bitmap_info
.offset
=
1225 mddev
->bitmap_info
.default_offset
;
1227 } else if (mddev
->pers
== NULL
) {
1228 /* Insist on good event counter while assembling, except
1229 * for spares (which don't need an event count) */
1231 if (sb
->disks
[rdev
->desc_nr
].state
& (
1232 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1233 if (ev1
< mddev
->events
)
1235 } else if (mddev
->bitmap
) {
1236 /* if adding to array with a bitmap, then we can accept an
1237 * older device ... but not too old.
1239 if (ev1
< mddev
->bitmap
->events_cleared
)
1242 if (ev1
< mddev
->events
)
1243 /* just a hot-add of a new device, leave raid_disk at -1 */
1247 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1248 desc
= sb
->disks
+ rdev
->desc_nr
;
1250 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1251 set_bit(Faulty
, &rdev
->flags
);
1252 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1253 desc->raid_disk < mddev->raid_disks */) {
1254 set_bit(In_sync
, &rdev
->flags
);
1255 rdev
->raid_disk
= desc
->raid_disk
;
1256 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1257 /* active but not in sync implies recovery up to
1258 * reshape position. We don't know exactly where
1259 * that is, so set to zero for now */
1260 if (mddev
->minor_version
>= 91) {
1261 rdev
->recovery_offset
= 0;
1262 rdev
->raid_disk
= desc
->raid_disk
;
1265 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1266 set_bit(WriteMostly
, &rdev
->flags
);
1267 } else /* MULTIPATH are always insync */
1268 set_bit(In_sync
, &rdev
->flags
);
1273 * sync_super for 0.90.0
1275 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1278 struct md_rdev
*rdev2
;
1279 int next_spare
= mddev
->raid_disks
;
1282 /* make rdev->sb match mddev data..
1285 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1286 * 3/ any empty disks < next_spare become removed
1288 * disks[0] gets initialised to REMOVED because
1289 * we cannot be sure from other fields if it has
1290 * been initialised or not.
1293 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1295 rdev
->sb_size
= MD_SB_BYTES
;
1297 sb
= page_address(rdev
->sb_page
);
1299 memset(sb
, 0, sizeof(*sb
));
1301 sb
->md_magic
= MD_SB_MAGIC
;
1302 sb
->major_version
= mddev
->major_version
;
1303 sb
->patch_version
= mddev
->patch_version
;
1304 sb
->gvalid_words
= 0; /* ignored */
1305 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1306 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1307 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1308 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1310 sb
->ctime
= mddev
->ctime
;
1311 sb
->level
= mddev
->level
;
1312 sb
->size
= mddev
->dev_sectors
/ 2;
1313 sb
->raid_disks
= mddev
->raid_disks
;
1314 sb
->md_minor
= mddev
->md_minor
;
1315 sb
->not_persistent
= 0;
1316 sb
->utime
= mddev
->utime
;
1318 sb
->events_hi
= (mddev
->events
>>32);
1319 sb
->events_lo
= (u32
)mddev
->events
;
1321 if (mddev
->reshape_position
== MaxSector
)
1322 sb
->minor_version
= 90;
1324 sb
->minor_version
= 91;
1325 sb
->reshape_position
= mddev
->reshape_position
;
1326 sb
->new_level
= mddev
->new_level
;
1327 sb
->delta_disks
= mddev
->delta_disks
;
1328 sb
->new_layout
= mddev
->new_layout
;
1329 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1331 mddev
->minor_version
= sb
->minor_version
;
1334 sb
->recovery_cp
= mddev
->recovery_cp
;
1335 sb
->cp_events_hi
= (mddev
->events
>>32);
1336 sb
->cp_events_lo
= (u32
)mddev
->events
;
1337 if (mddev
->recovery_cp
== MaxSector
)
1338 sb
->state
= (1<< MD_SB_CLEAN
);
1340 sb
->recovery_cp
= 0;
1342 sb
->layout
= mddev
->layout
;
1343 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1345 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1346 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1348 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1349 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1352 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1354 if (rdev2
->raid_disk
>= 0 &&
1355 sb
->minor_version
>= 91)
1356 /* we have nowhere to store the recovery_offset,
1357 * but if it is not below the reshape_position,
1358 * we can piggy-back on that.
1361 if (rdev2
->raid_disk
< 0 ||
1362 test_bit(Faulty
, &rdev2
->flags
))
1365 desc_nr
= rdev2
->raid_disk
;
1367 desc_nr
= next_spare
++;
1368 rdev2
->desc_nr
= desc_nr
;
1369 d
= &sb
->disks
[rdev2
->desc_nr
];
1371 d
->number
= rdev2
->desc_nr
;
1372 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1373 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1375 d
->raid_disk
= rdev2
->raid_disk
;
1377 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1378 if (test_bit(Faulty
, &rdev2
->flags
))
1379 d
->state
= (1<<MD_DISK_FAULTY
);
1380 else if (is_active
) {
1381 d
->state
= (1<<MD_DISK_ACTIVE
);
1382 if (test_bit(In_sync
, &rdev2
->flags
))
1383 d
->state
|= (1<<MD_DISK_SYNC
);
1391 if (test_bit(WriteMostly
, &rdev2
->flags
))
1392 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1394 /* now set the "removed" and "faulty" bits on any missing devices */
1395 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1396 mdp_disk_t
*d
= &sb
->disks
[i
];
1397 if (d
->state
== 0 && d
->number
== 0) {
1400 d
->state
= (1<<MD_DISK_REMOVED
);
1401 d
->state
|= (1<<MD_DISK_FAULTY
);
1405 sb
->nr_disks
= nr_disks
;
1406 sb
->active_disks
= active
;
1407 sb
->working_disks
= working
;
1408 sb
->failed_disks
= failed
;
1409 sb
->spare_disks
= spare
;
1411 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1412 sb
->sb_csum
= calc_sb_csum(sb
);
1416 * rdev_size_change for 0.90.0
1418 static unsigned long long
1419 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1421 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1422 return 0; /* component must fit device */
1423 if (rdev
->mddev
->bitmap_info
.offset
)
1424 return 0; /* can't move bitmap */
1425 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1426 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1427 num_sectors
= rdev
->sb_start
;
1428 /* Limit to 4TB as metadata cannot record more than that.
1429 * 4TB == 2^32 KB, or 2*2^32 sectors.
1431 if (num_sectors
>= (2ULL << 32))
1432 num_sectors
= (2ULL << 32) - 2;
1433 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1435 md_super_wait(rdev
->mddev
);
1441 * version 1 superblock
1444 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1448 unsigned long long newcsum
;
1449 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1450 __le32
*isuper
= (__le32
*)sb
;
1453 disk_csum
= sb
->sb_csum
;
1456 for (i
=0; size
>=4; size
-= 4 )
1457 newcsum
+= le32_to_cpu(*isuper
++);
1460 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1462 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1463 sb
->sb_csum
= disk_csum
;
1464 return cpu_to_le32(csum
);
1467 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1469 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1471 struct mdp_superblock_1
*sb
;
1474 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1478 * Calculate the position of the superblock in 512byte sectors.
1479 * It is always aligned to a 4K boundary and
1480 * depeding on minor_version, it can be:
1481 * 0: At least 8K, but less than 12K, from end of device
1482 * 1: At start of device
1483 * 2: 4K from start of device.
1485 switch(minor_version
) {
1487 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1489 sb_start
&= ~(sector_t
)(4*2-1);
1500 rdev
->sb_start
= sb_start
;
1502 /* superblock is rarely larger than 1K, but it can be larger,
1503 * and it is safe to read 4k, so we do that
1505 ret
= read_disk_sb(rdev
, 4096);
1506 if (ret
) return ret
;
1509 sb
= page_address(rdev
->sb_page
);
1511 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1512 sb
->major_version
!= cpu_to_le32(1) ||
1513 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1514 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1515 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1518 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1519 printk("md: invalid superblock checksum on %s\n",
1520 bdevname(rdev
->bdev
,b
));
1523 if (le64_to_cpu(sb
->data_size
) < 10) {
1524 printk("md: data_size too small on %s\n",
1525 bdevname(rdev
->bdev
,b
));
1529 rdev
->preferred_minor
= 0xffff;
1530 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1531 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1533 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1534 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1535 if (rdev
->sb_size
& bmask
)
1536 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1539 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1542 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1545 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1547 if (!rdev
->bb_page
) {
1548 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1552 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1553 rdev
->badblocks
.count
== 0) {
1554 /* need to load the bad block list.
1555 * Currently we limit it to one page.
1561 int sectors
= le16_to_cpu(sb
->bblog_size
);
1562 if (sectors
> (PAGE_SIZE
/ 512))
1564 offset
= le32_to_cpu(sb
->bblog_offset
);
1567 bb_sector
= (long long)offset
;
1568 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1569 rdev
->bb_page
, READ
, true))
1571 bbp
= (u64
*)page_address(rdev
->bb_page
);
1572 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1573 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1574 u64 bb
= le64_to_cpu(*bbp
);
1575 int count
= bb
& (0x3ff);
1576 u64 sector
= bb
>> 10;
1577 sector
<<= sb
->bblog_shift
;
1578 count
<<= sb
->bblog_shift
;
1581 if (md_set_badblocks(&rdev
->badblocks
,
1582 sector
, count
, 1) == 0)
1585 } else if (sb
->bblog_offset
== 0)
1586 rdev
->badblocks
.shift
= -1;
1592 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1594 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1595 sb
->level
!= refsb
->level
||
1596 sb
->layout
!= refsb
->layout
||
1597 sb
->chunksize
!= refsb
->chunksize
) {
1598 printk(KERN_WARNING
"md: %s has strangely different"
1599 " superblock to %s\n",
1600 bdevname(rdev
->bdev
,b
),
1601 bdevname(refdev
->bdev
,b2
));
1604 ev1
= le64_to_cpu(sb
->events
);
1605 ev2
= le64_to_cpu(refsb
->events
);
1613 rdev
->sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
1614 le64_to_cpu(sb
->data_offset
);
1616 rdev
->sectors
= rdev
->sb_start
;
1617 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1619 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1620 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1625 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1627 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1628 __u64 ev1
= le64_to_cpu(sb
->events
);
1630 rdev
->raid_disk
= -1;
1631 clear_bit(Faulty
, &rdev
->flags
);
1632 clear_bit(In_sync
, &rdev
->flags
);
1633 clear_bit(WriteMostly
, &rdev
->flags
);
1635 if (mddev
->raid_disks
== 0) {
1636 mddev
->major_version
= 1;
1637 mddev
->patch_version
= 0;
1638 mddev
->external
= 0;
1639 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1640 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1641 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1642 mddev
->level
= le32_to_cpu(sb
->level
);
1643 mddev
->clevel
[0] = 0;
1644 mddev
->layout
= le32_to_cpu(sb
->layout
);
1645 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1646 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1647 mddev
->events
= ev1
;
1648 mddev
->bitmap_info
.offset
= 0;
1649 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1651 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1652 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1654 mddev
->max_disks
= (4096-256)/2;
1656 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1657 mddev
->bitmap_info
.file
== NULL
)
1658 mddev
->bitmap_info
.offset
=
1659 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1661 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1662 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1663 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1664 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1665 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1666 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1668 mddev
->reshape_position
= MaxSector
;
1669 mddev
->delta_disks
= 0;
1670 mddev
->new_level
= mddev
->level
;
1671 mddev
->new_layout
= mddev
->layout
;
1672 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1675 } else if (mddev
->pers
== NULL
) {
1676 /* Insist of good event counter while assembling, except for
1677 * spares (which don't need an event count) */
1679 if (rdev
->desc_nr
>= 0 &&
1680 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1681 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1682 if (ev1
< mddev
->events
)
1684 } else if (mddev
->bitmap
) {
1685 /* If adding to array with a bitmap, then we can accept an
1686 * older device, but not too old.
1688 if (ev1
< mddev
->bitmap
->events_cleared
)
1691 if (ev1
< mddev
->events
)
1692 /* just a hot-add of a new device, leave raid_disk at -1 */
1695 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1697 if (rdev
->desc_nr
< 0 ||
1698 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1702 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1704 case 0xffff: /* spare */
1706 case 0xfffe: /* faulty */
1707 set_bit(Faulty
, &rdev
->flags
);
1710 if ((le32_to_cpu(sb
->feature_map
) &
1711 MD_FEATURE_RECOVERY_OFFSET
))
1712 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1714 set_bit(In_sync
, &rdev
->flags
);
1715 rdev
->raid_disk
= role
;
1718 if (sb
->devflags
& WriteMostly1
)
1719 set_bit(WriteMostly
, &rdev
->flags
);
1720 } else /* MULTIPATH are always insync */
1721 set_bit(In_sync
, &rdev
->flags
);
1726 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1728 struct mdp_superblock_1
*sb
;
1729 struct md_rdev
*rdev2
;
1731 /* make rdev->sb match mddev and rdev data. */
1733 sb
= page_address(rdev
->sb_page
);
1735 sb
->feature_map
= 0;
1737 sb
->recovery_offset
= cpu_to_le64(0);
1738 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1739 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1741 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1742 sb
->events
= cpu_to_le64(mddev
->events
);
1744 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1746 sb
->resync_offset
= cpu_to_le64(0);
1748 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1750 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1751 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1752 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1753 sb
->level
= cpu_to_le32(mddev
->level
);
1754 sb
->layout
= cpu_to_le32(mddev
->layout
);
1756 if (test_bit(WriteMostly
, &rdev
->flags
))
1757 sb
->devflags
|= WriteMostly1
;
1759 sb
->devflags
&= ~WriteMostly1
;
1761 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1762 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1763 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1766 if (rdev
->raid_disk
>= 0 &&
1767 !test_bit(In_sync
, &rdev
->flags
)) {
1769 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1770 sb
->recovery_offset
=
1771 cpu_to_le64(rdev
->recovery_offset
);
1774 if (mddev
->reshape_position
!= MaxSector
) {
1775 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1776 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1777 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1778 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1779 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1780 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1783 if (rdev
->badblocks
.count
== 0)
1784 /* Nothing to do for bad blocks*/ ;
1785 else if (sb
->bblog_offset
== 0)
1786 /* Cannot record bad blocks on this device */
1787 md_error(mddev
, rdev
);
1789 struct badblocks
*bb
= &rdev
->badblocks
;
1790 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1792 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1797 seq
= read_seqbegin(&bb
->lock
);
1799 memset(bbp
, 0xff, PAGE_SIZE
);
1801 for (i
= 0 ; i
< bb
->count
; i
++) {
1802 u64 internal_bb
= *p
++;
1803 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1804 | BB_LEN(internal_bb
));
1805 *bbp
++ = cpu_to_le64(store_bb
);
1807 if (read_seqretry(&bb
->lock
, seq
))
1810 bb
->sector
= (rdev
->sb_start
+
1811 (int)le32_to_cpu(sb
->bblog_offset
));
1812 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1818 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1819 if (rdev2
->desc_nr
+1 > max_dev
)
1820 max_dev
= rdev2
->desc_nr
+1;
1822 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1824 sb
->max_dev
= cpu_to_le32(max_dev
);
1825 rdev
->sb_size
= max_dev
* 2 + 256;
1826 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1827 if (rdev
->sb_size
& bmask
)
1828 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1830 max_dev
= le32_to_cpu(sb
->max_dev
);
1832 for (i
=0; i
<max_dev
;i
++)
1833 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1835 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1837 if (test_bit(Faulty
, &rdev2
->flags
))
1838 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1839 else if (test_bit(In_sync
, &rdev2
->flags
))
1840 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1841 else if (rdev2
->raid_disk
>= 0)
1842 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1844 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1847 sb
->sb_csum
= calc_sb_1_csum(sb
);
1850 static unsigned long long
1851 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1853 struct mdp_superblock_1
*sb
;
1854 sector_t max_sectors
;
1855 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1856 return 0; /* component must fit device */
1857 if (rdev
->sb_start
< rdev
->data_offset
) {
1858 /* minor versions 1 and 2; superblock before data */
1859 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1860 max_sectors
-= rdev
->data_offset
;
1861 if (!num_sectors
|| num_sectors
> max_sectors
)
1862 num_sectors
= max_sectors
;
1863 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1864 /* minor version 0 with bitmap we can't move */
1867 /* minor version 0; superblock after data */
1869 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1870 sb_start
&= ~(sector_t
)(4*2 - 1);
1871 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1872 if (!num_sectors
|| num_sectors
> max_sectors
)
1873 num_sectors
= max_sectors
;
1874 rdev
->sb_start
= sb_start
;
1876 sb
= page_address(rdev
->sb_page
);
1877 sb
->data_size
= cpu_to_le64(num_sectors
);
1878 sb
->super_offset
= rdev
->sb_start
;
1879 sb
->sb_csum
= calc_sb_1_csum(sb
);
1880 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1882 md_super_wait(rdev
->mddev
);
1886 static struct super_type super_types
[] = {
1889 .owner
= THIS_MODULE
,
1890 .load_super
= super_90_load
,
1891 .validate_super
= super_90_validate
,
1892 .sync_super
= super_90_sync
,
1893 .rdev_size_change
= super_90_rdev_size_change
,
1897 .owner
= THIS_MODULE
,
1898 .load_super
= super_1_load
,
1899 .validate_super
= super_1_validate
,
1900 .sync_super
= super_1_sync
,
1901 .rdev_size_change
= super_1_rdev_size_change
,
1905 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1907 if (mddev
->sync_super
) {
1908 mddev
->sync_super(mddev
, rdev
);
1912 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1914 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1917 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1919 struct md_rdev
*rdev
, *rdev2
;
1922 rdev_for_each_rcu(rdev
, mddev1
)
1923 rdev_for_each_rcu(rdev2
, mddev2
)
1924 if (rdev
->bdev
->bd_contains
==
1925 rdev2
->bdev
->bd_contains
) {
1933 static LIST_HEAD(pending_raid_disks
);
1936 * Try to register data integrity profile for an mddev
1938 * This is called when an array is started and after a disk has been kicked
1939 * from the array. It only succeeds if all working and active component devices
1940 * are integrity capable with matching profiles.
1942 int md_integrity_register(struct mddev
*mddev
)
1944 struct md_rdev
*rdev
, *reference
= NULL
;
1946 if (list_empty(&mddev
->disks
))
1947 return 0; /* nothing to do */
1948 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1949 return 0; /* shouldn't register, or already is */
1950 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1951 /* skip spares and non-functional disks */
1952 if (test_bit(Faulty
, &rdev
->flags
))
1954 if (rdev
->raid_disk
< 0)
1957 /* Use the first rdev as the reference */
1961 /* does this rdev's profile match the reference profile? */
1962 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1963 rdev
->bdev
->bd_disk
) < 0)
1966 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1969 * All component devices are integrity capable and have matching
1970 * profiles, register the common profile for the md device.
1972 if (blk_integrity_register(mddev
->gendisk
,
1973 bdev_get_integrity(reference
->bdev
)) != 0) {
1974 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1978 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1979 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1980 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1986 EXPORT_SYMBOL(md_integrity_register
);
1988 /* Disable data integrity if non-capable/non-matching disk is being added */
1989 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
1991 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1992 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1994 if (!bi_mddev
) /* nothing to do */
1996 if (rdev
->raid_disk
< 0) /* skip spares */
1998 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1999 rdev
->bdev
->bd_disk
) >= 0)
2001 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2002 blk_integrity_unregister(mddev
->gendisk
);
2004 EXPORT_SYMBOL(md_integrity_add_rdev
);
2006 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2008 char b
[BDEVNAME_SIZE
];
2018 /* prevent duplicates */
2019 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2022 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2023 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2024 rdev
->sectors
< mddev
->dev_sectors
)) {
2026 /* Cannot change size, so fail
2027 * If mddev->level <= 0, then we don't care
2028 * about aligning sizes (e.g. linear)
2030 if (mddev
->level
> 0)
2033 mddev
->dev_sectors
= rdev
->sectors
;
2036 /* Verify rdev->desc_nr is unique.
2037 * If it is -1, assign a free number, else
2038 * check number is not in use
2040 if (rdev
->desc_nr
< 0) {
2042 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2043 while (find_rdev_nr(mddev
, choice
))
2045 rdev
->desc_nr
= choice
;
2047 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2050 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2051 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2052 mdname(mddev
), mddev
->max_disks
);
2055 bdevname(rdev
->bdev
,b
);
2056 while ( (s
=strchr(b
, '/')) != NULL
)
2059 rdev
->mddev
= mddev
;
2060 printk(KERN_INFO
"md: bind<%s>\n", b
);
2062 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2065 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2066 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2067 /* failure here is OK */;
2068 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2070 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2071 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2073 /* May as well allow recovery to be retried once */
2074 mddev
->recovery_disabled
++;
2079 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2084 static void md_delayed_delete(struct work_struct
*ws
)
2086 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2087 kobject_del(&rdev
->kobj
);
2088 kobject_put(&rdev
->kobj
);
2091 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2093 char b
[BDEVNAME_SIZE
];
2098 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2099 list_del_rcu(&rdev
->same_set
);
2100 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2102 sysfs_remove_link(&rdev
->kobj
, "block");
2103 sysfs_put(rdev
->sysfs_state
);
2104 rdev
->sysfs_state
= NULL
;
2105 kfree(rdev
->badblocks
.page
);
2106 rdev
->badblocks
.count
= 0;
2107 rdev
->badblocks
.page
= NULL
;
2108 /* We need to delay this, otherwise we can deadlock when
2109 * writing to 'remove' to "dev/state". We also need
2110 * to delay it due to rcu usage.
2113 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2114 kobject_get(&rdev
->kobj
);
2115 queue_work(md_misc_wq
, &rdev
->del_work
);
2119 * prevent the device from being mounted, repartitioned or
2120 * otherwise reused by a RAID array (or any other kernel
2121 * subsystem), by bd_claiming the device.
2123 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2126 struct block_device
*bdev
;
2127 char b
[BDEVNAME_SIZE
];
2129 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2130 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2132 printk(KERN_ERR
"md: could not open %s.\n",
2133 __bdevname(dev
, b
));
2134 return PTR_ERR(bdev
);
2140 static void unlock_rdev(struct md_rdev
*rdev
)
2142 struct block_device
*bdev
= rdev
->bdev
;
2146 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2149 void md_autodetect_dev(dev_t dev
);
2151 static void export_rdev(struct md_rdev
* rdev
)
2153 char b
[BDEVNAME_SIZE
];
2154 printk(KERN_INFO
"md: export_rdev(%s)\n",
2155 bdevname(rdev
->bdev
,b
));
2160 if (test_bit(AutoDetected
, &rdev
->flags
))
2161 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2164 kobject_put(&rdev
->kobj
);
2167 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2169 unbind_rdev_from_array(rdev
);
2173 static void export_array(struct mddev
*mddev
)
2175 struct md_rdev
*rdev
, *tmp
;
2177 rdev_for_each(rdev
, tmp
, mddev
) {
2182 kick_rdev_from_array(rdev
);
2184 if (!list_empty(&mddev
->disks
))
2186 mddev
->raid_disks
= 0;
2187 mddev
->major_version
= 0;
2190 static void print_desc(mdp_disk_t
*desc
)
2192 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2193 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2196 static void print_sb_90(mdp_super_t
*sb
)
2201 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2202 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2203 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2205 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2206 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2207 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2208 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2209 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2210 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2211 sb
->failed_disks
, sb
->spare_disks
,
2212 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2215 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2218 desc
= sb
->disks
+ i
;
2219 if (desc
->number
|| desc
->major
|| desc
->minor
||
2220 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2221 printk(" D %2d: ", i
);
2225 printk(KERN_INFO
"md: THIS: ");
2226 print_desc(&sb
->this_disk
);
2229 static void print_sb_1(struct mdp_superblock_1
*sb
)
2233 uuid
= sb
->set_uuid
;
2235 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2236 "md: Name: \"%s\" CT:%llu\n",
2237 le32_to_cpu(sb
->major_version
),
2238 le32_to_cpu(sb
->feature_map
),
2241 (unsigned long long)le64_to_cpu(sb
->ctime
)
2242 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2244 uuid
= sb
->device_uuid
;
2246 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2248 "md: Dev:%08x UUID: %pU\n"
2249 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2250 "md: (MaxDev:%u) \n",
2251 le32_to_cpu(sb
->level
),
2252 (unsigned long long)le64_to_cpu(sb
->size
),
2253 le32_to_cpu(sb
->raid_disks
),
2254 le32_to_cpu(sb
->layout
),
2255 le32_to_cpu(sb
->chunksize
),
2256 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2257 (unsigned long long)le64_to_cpu(sb
->data_size
),
2258 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2259 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2260 le32_to_cpu(sb
->dev_number
),
2263 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2264 (unsigned long long)le64_to_cpu(sb
->events
),
2265 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2266 le32_to_cpu(sb
->sb_csum
),
2267 le32_to_cpu(sb
->max_dev
)
2271 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2273 char b
[BDEVNAME_SIZE
];
2274 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2275 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2276 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2278 if (rdev
->sb_loaded
) {
2279 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2280 switch (major_version
) {
2282 print_sb_90(page_address(rdev
->sb_page
));
2285 print_sb_1(page_address(rdev
->sb_page
));
2289 printk(KERN_INFO
"md: no rdev superblock!\n");
2292 static void md_print_devices(void)
2294 struct list_head
*tmp
;
2295 struct md_rdev
*rdev
;
2296 struct mddev
*mddev
;
2297 char b
[BDEVNAME_SIZE
];
2300 printk("md: **********************************\n");
2301 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2302 printk("md: **********************************\n");
2303 for_each_mddev(mddev
, tmp
) {
2306 bitmap_print_sb(mddev
->bitmap
);
2308 printk("%s: ", mdname(mddev
));
2309 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2310 printk("<%s>", bdevname(rdev
->bdev
,b
));
2313 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2314 print_rdev(rdev
, mddev
->major_version
);
2316 printk("md: **********************************\n");
2321 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2323 /* Update each superblock (in-memory image), but
2324 * if we are allowed to, skip spares which already
2325 * have the right event counter, or have one earlier
2326 * (which would mean they aren't being marked as dirty
2327 * with the rest of the array)
2329 struct md_rdev
*rdev
;
2330 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2331 if (rdev
->sb_events
== mddev
->events
||
2333 rdev
->raid_disk
< 0 &&
2334 rdev
->sb_events
+1 == mddev
->events
)) {
2335 /* Don't update this superblock */
2336 rdev
->sb_loaded
= 2;
2338 sync_super(mddev
, rdev
);
2339 rdev
->sb_loaded
= 1;
2344 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2346 struct md_rdev
*rdev
;
2349 int any_badblocks_changed
= 0;
2352 /* First make sure individual recovery_offsets are correct */
2353 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2354 if (rdev
->raid_disk
>= 0 &&
2355 mddev
->delta_disks
>= 0 &&
2356 !test_bit(In_sync
, &rdev
->flags
) &&
2357 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2358 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2361 if (!mddev
->persistent
) {
2362 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2363 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2364 if (!mddev
->external
) {
2365 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2366 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2367 if (rdev
->badblocks
.changed
) {
2368 md_ack_all_badblocks(&rdev
->badblocks
);
2369 md_error(mddev
, rdev
);
2371 clear_bit(Blocked
, &rdev
->flags
);
2372 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2373 wake_up(&rdev
->blocked_wait
);
2376 wake_up(&mddev
->sb_wait
);
2380 spin_lock_irq(&mddev
->write_lock
);
2382 mddev
->utime
= get_seconds();
2384 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2386 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2387 /* just a clean<-> dirty transition, possibly leave spares alone,
2388 * though if events isn't the right even/odd, we will have to do
2394 if (mddev
->degraded
)
2395 /* If the array is degraded, then skipping spares is both
2396 * dangerous and fairly pointless.
2397 * Dangerous because a device that was removed from the array
2398 * might have a event_count that still looks up-to-date,
2399 * so it can be re-added without a resync.
2400 * Pointless because if there are any spares to skip,
2401 * then a recovery will happen and soon that array won't
2402 * be degraded any more and the spare can go back to sleep then.
2406 sync_req
= mddev
->in_sync
;
2408 /* If this is just a dirty<->clean transition, and the array is clean
2409 * and 'events' is odd, we can roll back to the previous clean state */
2411 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2412 && mddev
->can_decrease_events
2413 && mddev
->events
!= 1) {
2415 mddev
->can_decrease_events
= 0;
2417 /* otherwise we have to go forward and ... */
2419 mddev
->can_decrease_events
= nospares
;
2422 if (!mddev
->events
) {
2424 * oops, this 64-bit counter should never wrap.
2425 * Either we are in around ~1 trillion A.C., assuming
2426 * 1 reboot per second, or we have a bug:
2432 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2433 if (rdev
->badblocks
.changed
)
2434 any_badblocks_changed
++;
2435 if (test_bit(Faulty
, &rdev
->flags
))
2436 set_bit(FaultRecorded
, &rdev
->flags
);
2439 sync_sbs(mddev
, nospares
);
2440 spin_unlock_irq(&mddev
->write_lock
);
2442 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2443 mdname(mddev
), mddev
->in_sync
);
2445 bitmap_update_sb(mddev
->bitmap
);
2446 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2447 char b
[BDEVNAME_SIZE
];
2449 if (rdev
->sb_loaded
!= 1)
2450 continue; /* no noise on spare devices */
2452 if (!test_bit(Faulty
, &rdev
->flags
)) {
2453 md_super_write(mddev
,rdev
,
2454 rdev
->sb_start
, rdev
->sb_size
,
2456 pr_debug("md: (write) %s's sb offset: %llu\n",
2457 bdevname(rdev
->bdev
, b
),
2458 (unsigned long long)rdev
->sb_start
);
2459 rdev
->sb_events
= mddev
->events
;
2460 if (rdev
->badblocks
.size
) {
2461 md_super_write(mddev
, rdev
,
2462 rdev
->badblocks
.sector
,
2463 rdev
->badblocks
.size
<< 9,
2465 rdev
->badblocks
.size
= 0;
2469 pr_debug("md: %s (skipping faulty)\n",
2470 bdevname(rdev
->bdev
, b
));
2471 if (mddev
->level
== LEVEL_MULTIPATH
)
2472 /* only need to write one superblock... */
2475 md_super_wait(mddev
);
2476 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2478 spin_lock_irq(&mddev
->write_lock
);
2479 if (mddev
->in_sync
!= sync_req
||
2480 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2481 /* have to write it out again */
2482 spin_unlock_irq(&mddev
->write_lock
);
2485 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2486 spin_unlock_irq(&mddev
->write_lock
);
2487 wake_up(&mddev
->sb_wait
);
2488 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2489 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2491 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2492 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2493 clear_bit(Blocked
, &rdev
->flags
);
2495 if (any_badblocks_changed
)
2496 md_ack_all_badblocks(&rdev
->badblocks
);
2497 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2498 wake_up(&rdev
->blocked_wait
);
2502 /* words written to sysfs files may, or may not, be \n terminated.
2503 * We want to accept with case. For this we use cmd_match.
2505 static int cmd_match(const char *cmd
, const char *str
)
2507 /* See if cmd, written into a sysfs file, matches
2508 * str. They must either be the same, or cmd can
2509 * have a trailing newline
2511 while (*cmd
&& *str
&& *cmd
== *str
) {
2522 struct rdev_sysfs_entry
{
2523 struct attribute attr
;
2524 ssize_t (*show
)(struct md_rdev
*, char *);
2525 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2529 state_show(struct md_rdev
*rdev
, char *page
)
2534 if (test_bit(Faulty
, &rdev
->flags
) ||
2535 rdev
->badblocks
.unacked_exist
) {
2536 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2539 if (test_bit(In_sync
, &rdev
->flags
)) {
2540 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2543 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2544 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2547 if (test_bit(Blocked
, &rdev
->flags
) ||
2548 rdev
->badblocks
.unacked_exist
) {
2549 len
+= sprintf(page
+len
, "%sblocked", sep
);
2552 if (!test_bit(Faulty
, &rdev
->flags
) &&
2553 !test_bit(In_sync
, &rdev
->flags
)) {
2554 len
+= sprintf(page
+len
, "%sspare", sep
);
2557 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2558 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2561 return len
+sprintf(page
+len
, "\n");
2565 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2568 * faulty - simulates an error
2569 * remove - disconnects the device
2570 * writemostly - sets write_mostly
2571 * -writemostly - clears write_mostly
2572 * blocked - sets the Blocked flags
2573 * -blocked - clears the Blocked and possibly simulates an error
2574 * insync - sets Insync providing device isn't active
2575 * write_error - sets WriteErrorSeen
2576 * -write_error - clears WriteErrorSeen
2579 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2580 md_error(rdev
->mddev
, rdev
);
2581 if (test_bit(Faulty
, &rdev
->flags
))
2585 } else if (cmd_match(buf
, "remove")) {
2586 if (rdev
->raid_disk
>= 0)
2589 struct mddev
*mddev
= rdev
->mddev
;
2590 kick_rdev_from_array(rdev
);
2592 md_update_sb(mddev
, 1);
2593 md_new_event(mddev
);
2596 } else if (cmd_match(buf
, "writemostly")) {
2597 set_bit(WriteMostly
, &rdev
->flags
);
2599 } else if (cmd_match(buf
, "-writemostly")) {
2600 clear_bit(WriteMostly
, &rdev
->flags
);
2602 } else if (cmd_match(buf
, "blocked")) {
2603 set_bit(Blocked
, &rdev
->flags
);
2605 } else if (cmd_match(buf
, "-blocked")) {
2606 if (!test_bit(Faulty
, &rdev
->flags
) &&
2607 rdev
->badblocks
.unacked_exist
) {
2608 /* metadata handler doesn't understand badblocks,
2609 * so we need to fail the device
2611 md_error(rdev
->mddev
, rdev
);
2613 clear_bit(Blocked
, &rdev
->flags
);
2614 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2615 wake_up(&rdev
->blocked_wait
);
2616 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2617 md_wakeup_thread(rdev
->mddev
->thread
);
2620 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2621 set_bit(In_sync
, &rdev
->flags
);
2623 } else if (cmd_match(buf
, "write_error")) {
2624 set_bit(WriteErrorSeen
, &rdev
->flags
);
2626 } else if (cmd_match(buf
, "-write_error")) {
2627 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2631 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2632 return err
? err
: len
;
2634 static struct rdev_sysfs_entry rdev_state
=
2635 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2638 errors_show(struct md_rdev
*rdev
, char *page
)
2640 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2644 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2647 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2648 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2649 atomic_set(&rdev
->corrected_errors
, n
);
2654 static struct rdev_sysfs_entry rdev_errors
=
2655 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2658 slot_show(struct md_rdev
*rdev
, char *page
)
2660 if (rdev
->raid_disk
< 0)
2661 return sprintf(page
, "none\n");
2663 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2667 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2671 int slot
= simple_strtoul(buf
, &e
, 10);
2672 if (strncmp(buf
, "none", 4)==0)
2674 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2676 if (rdev
->mddev
->pers
&& slot
== -1) {
2677 /* Setting 'slot' on an active array requires also
2678 * updating the 'rd%d' link, and communicating
2679 * with the personality with ->hot_*_disk.
2680 * For now we only support removing
2681 * failed/spare devices. This normally happens automatically,
2682 * but not when the metadata is externally managed.
2684 if (rdev
->raid_disk
== -1)
2686 /* personality does all needed checks */
2687 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2689 err
= rdev
->mddev
->pers
->
2690 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2693 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2694 rdev
->raid_disk
= -1;
2695 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2696 md_wakeup_thread(rdev
->mddev
->thread
);
2697 } else if (rdev
->mddev
->pers
) {
2698 struct md_rdev
*rdev2
;
2699 /* Activating a spare .. or possibly reactivating
2700 * if we ever get bitmaps working here.
2703 if (rdev
->raid_disk
!= -1)
2706 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2709 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2712 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2713 if (rdev2
->raid_disk
== slot
)
2716 if (slot
>= rdev
->mddev
->raid_disks
&&
2717 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2720 rdev
->raid_disk
= slot
;
2721 if (test_bit(In_sync
, &rdev
->flags
))
2722 rdev
->saved_raid_disk
= slot
;
2724 rdev
->saved_raid_disk
= -1;
2725 err
= rdev
->mddev
->pers
->
2726 hot_add_disk(rdev
->mddev
, rdev
);
2728 rdev
->raid_disk
= -1;
2731 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2732 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2733 /* failure here is OK */;
2734 /* don't wakeup anyone, leave that to userspace. */
2736 if (slot
>= rdev
->mddev
->raid_disks
&&
2737 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2739 rdev
->raid_disk
= slot
;
2740 /* assume it is working */
2741 clear_bit(Faulty
, &rdev
->flags
);
2742 clear_bit(WriteMostly
, &rdev
->flags
);
2743 set_bit(In_sync
, &rdev
->flags
);
2744 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2750 static struct rdev_sysfs_entry rdev_slot
=
2751 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2754 offset_show(struct md_rdev
*rdev
, char *page
)
2756 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2760 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2763 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2764 if (e
==buf
|| (*e
&& *e
!= '\n'))
2766 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2768 if (rdev
->sectors
&& rdev
->mddev
->external
)
2769 /* Must set offset before size, so overlap checks
2772 rdev
->data_offset
= offset
;
2776 static struct rdev_sysfs_entry rdev_offset
=
2777 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2780 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2782 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2785 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2787 /* check if two start/length pairs overlap */
2795 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2797 unsigned long long blocks
;
2800 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2803 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2804 return -EINVAL
; /* sector conversion overflow */
2807 if (new != blocks
* 2)
2808 return -EINVAL
; /* unsigned long long to sector_t overflow */
2815 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2817 struct mddev
*my_mddev
= rdev
->mddev
;
2818 sector_t oldsectors
= rdev
->sectors
;
2821 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2823 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2824 if (my_mddev
->persistent
) {
2825 sectors
= super_types
[my_mddev
->major_version
].
2826 rdev_size_change(rdev
, sectors
);
2829 } else if (!sectors
)
2830 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2833 if (sectors
< my_mddev
->dev_sectors
)
2834 return -EINVAL
; /* component must fit device */
2836 rdev
->sectors
= sectors
;
2837 if (sectors
> oldsectors
&& my_mddev
->external
) {
2838 /* need to check that all other rdevs with the same ->bdev
2839 * do not overlap. We need to unlock the mddev to avoid
2840 * a deadlock. We have already changed rdev->sectors, and if
2841 * we have to change it back, we will have the lock again.
2843 struct mddev
*mddev
;
2845 struct list_head
*tmp
;
2847 mddev_unlock(my_mddev
);
2848 for_each_mddev(mddev
, tmp
) {
2849 struct md_rdev
*rdev2
;
2852 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2853 if (rdev
->bdev
== rdev2
->bdev
&&
2855 overlaps(rdev
->data_offset
, rdev
->sectors
,
2861 mddev_unlock(mddev
);
2867 mddev_lock(my_mddev
);
2869 /* Someone else could have slipped in a size
2870 * change here, but doing so is just silly.
2871 * We put oldsectors back because we *know* it is
2872 * safe, and trust userspace not to race with
2875 rdev
->sectors
= oldsectors
;
2882 static struct rdev_sysfs_entry rdev_size
=
2883 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2886 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2888 unsigned long long recovery_start
= rdev
->recovery_offset
;
2890 if (test_bit(In_sync
, &rdev
->flags
) ||
2891 recovery_start
== MaxSector
)
2892 return sprintf(page
, "none\n");
2894 return sprintf(page
, "%llu\n", recovery_start
);
2897 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2899 unsigned long long recovery_start
;
2901 if (cmd_match(buf
, "none"))
2902 recovery_start
= MaxSector
;
2903 else if (strict_strtoull(buf
, 10, &recovery_start
))
2906 if (rdev
->mddev
->pers
&&
2907 rdev
->raid_disk
>= 0)
2910 rdev
->recovery_offset
= recovery_start
;
2911 if (recovery_start
== MaxSector
)
2912 set_bit(In_sync
, &rdev
->flags
);
2914 clear_bit(In_sync
, &rdev
->flags
);
2918 static struct rdev_sysfs_entry rdev_recovery_start
=
2919 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2923 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2925 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2927 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2929 return badblocks_show(&rdev
->badblocks
, page
, 0);
2931 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2933 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2934 /* Maybe that ack was all we needed */
2935 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2936 wake_up(&rdev
->blocked_wait
);
2939 static struct rdev_sysfs_entry rdev_bad_blocks
=
2940 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2943 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2945 return badblocks_show(&rdev
->badblocks
, page
, 1);
2947 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2949 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2951 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2952 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2954 static struct attribute
*rdev_default_attrs
[] = {
2960 &rdev_recovery_start
.attr
,
2961 &rdev_bad_blocks
.attr
,
2962 &rdev_unack_bad_blocks
.attr
,
2966 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2968 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2969 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2970 struct mddev
*mddev
= rdev
->mddev
;
2976 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2978 if (rdev
->mddev
== NULL
)
2981 rv
= entry
->show(rdev
, page
);
2982 mddev_unlock(mddev
);
2988 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2989 const char *page
, size_t length
)
2991 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2992 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2994 struct mddev
*mddev
= rdev
->mddev
;
2998 if (!capable(CAP_SYS_ADMIN
))
3000 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3002 if (rdev
->mddev
== NULL
)
3005 rv
= entry
->store(rdev
, page
, length
);
3006 mddev_unlock(mddev
);
3011 static void rdev_free(struct kobject
*ko
)
3013 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3016 static const struct sysfs_ops rdev_sysfs_ops
= {
3017 .show
= rdev_attr_show
,
3018 .store
= rdev_attr_store
,
3020 static struct kobj_type rdev_ktype
= {
3021 .release
= rdev_free
,
3022 .sysfs_ops
= &rdev_sysfs_ops
,
3023 .default_attrs
= rdev_default_attrs
,
3026 int md_rdev_init(struct md_rdev
*rdev
)
3029 rdev
->saved_raid_disk
= -1;
3030 rdev
->raid_disk
= -1;
3032 rdev
->data_offset
= 0;
3033 rdev
->sb_events
= 0;
3034 rdev
->last_read_error
.tv_sec
= 0;
3035 rdev
->last_read_error
.tv_nsec
= 0;
3036 rdev
->sb_loaded
= 0;
3037 rdev
->bb_page
= NULL
;
3038 atomic_set(&rdev
->nr_pending
, 0);
3039 atomic_set(&rdev
->read_errors
, 0);
3040 atomic_set(&rdev
->corrected_errors
, 0);
3042 INIT_LIST_HEAD(&rdev
->same_set
);
3043 init_waitqueue_head(&rdev
->blocked_wait
);
3045 /* Add space to store bad block list.
3046 * This reserves the space even on arrays where it cannot
3047 * be used - I wonder if that matters
3049 rdev
->badblocks
.count
= 0;
3050 rdev
->badblocks
.shift
= 0;
3051 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3052 seqlock_init(&rdev
->badblocks
.lock
);
3053 if (rdev
->badblocks
.page
== NULL
)
3058 EXPORT_SYMBOL_GPL(md_rdev_init
);
3060 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3062 * mark the device faulty if:
3064 * - the device is nonexistent (zero size)
3065 * - the device has no valid superblock
3067 * a faulty rdev _never_ has rdev->sb set.
3069 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3071 char b
[BDEVNAME_SIZE
];
3073 struct md_rdev
*rdev
;
3076 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3078 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3079 return ERR_PTR(-ENOMEM
);
3082 err
= md_rdev_init(rdev
);
3085 err
= alloc_disk_sb(rdev
);
3089 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3093 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3095 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3098 "md: %s has zero or unknown size, marking faulty!\n",
3099 bdevname(rdev
->bdev
,b
));
3104 if (super_format
>= 0) {
3105 err
= super_types
[super_format
].
3106 load_super(rdev
, NULL
, super_minor
);
3107 if (err
== -EINVAL
) {
3109 "md: %s does not have a valid v%d.%d "
3110 "superblock, not importing!\n",
3111 bdevname(rdev
->bdev
,b
),
3112 super_format
, super_minor
);
3117 "md: could not read %s's sb, not importing!\n",
3118 bdevname(rdev
->bdev
,b
));
3122 if (super_format
== -1)
3123 /* hot-add for 0.90, or non-persistent: so no badblocks */
3124 rdev
->badblocks
.shift
= -1;
3132 kfree(rdev
->badblocks
.page
);
3134 return ERR_PTR(err
);
3138 * Check a full RAID array for plausibility
3142 static void analyze_sbs(struct mddev
* mddev
)
3145 struct md_rdev
*rdev
, *freshest
, *tmp
;
3146 char b
[BDEVNAME_SIZE
];
3149 rdev_for_each(rdev
, tmp
, mddev
)
3150 switch (super_types
[mddev
->major_version
].
3151 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3159 "md: fatal superblock inconsistency in %s"
3160 " -- removing from array\n",
3161 bdevname(rdev
->bdev
,b
));
3162 kick_rdev_from_array(rdev
);
3166 super_types
[mddev
->major_version
].
3167 validate_super(mddev
, freshest
);
3170 rdev_for_each(rdev
, tmp
, mddev
) {
3171 if (mddev
->max_disks
&&
3172 (rdev
->desc_nr
>= mddev
->max_disks
||
3173 i
> mddev
->max_disks
)) {
3175 "md: %s: %s: only %d devices permitted\n",
3176 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3178 kick_rdev_from_array(rdev
);
3181 if (rdev
!= freshest
)
3182 if (super_types
[mddev
->major_version
].
3183 validate_super(mddev
, rdev
)) {
3184 printk(KERN_WARNING
"md: kicking non-fresh %s"
3186 bdevname(rdev
->bdev
,b
));
3187 kick_rdev_from_array(rdev
);
3190 if (mddev
->level
== LEVEL_MULTIPATH
) {
3191 rdev
->desc_nr
= i
++;
3192 rdev
->raid_disk
= rdev
->desc_nr
;
3193 set_bit(In_sync
, &rdev
->flags
);
3194 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3195 rdev
->raid_disk
= -1;
3196 clear_bit(In_sync
, &rdev
->flags
);
3201 /* Read a fixed-point number.
3202 * Numbers in sysfs attributes should be in "standard" units where
3203 * possible, so time should be in seconds.
3204 * However we internally use a a much smaller unit such as
3205 * milliseconds or jiffies.
3206 * This function takes a decimal number with a possible fractional
3207 * component, and produces an integer which is the result of
3208 * multiplying that number by 10^'scale'.
3209 * all without any floating-point arithmetic.
3211 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3213 unsigned long result
= 0;
3215 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3218 else if (decimals
< scale
) {
3221 result
= result
* 10 + value
;
3233 while (decimals
< scale
) {
3242 static void md_safemode_timeout(unsigned long data
);
3245 safe_delay_show(struct mddev
*mddev
, char *page
)
3247 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3248 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3251 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3255 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3258 mddev
->safemode_delay
= 0;
3260 unsigned long old_delay
= mddev
->safemode_delay
;
3261 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3262 if (mddev
->safemode_delay
== 0)
3263 mddev
->safemode_delay
= 1;
3264 if (mddev
->safemode_delay
< old_delay
)
3265 md_safemode_timeout((unsigned long)mddev
);
3269 static struct md_sysfs_entry md_safe_delay
=
3270 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3273 level_show(struct mddev
*mddev
, char *page
)
3275 struct mdk_personality
*p
= mddev
->pers
;
3277 return sprintf(page
, "%s\n", p
->name
);
3278 else if (mddev
->clevel
[0])
3279 return sprintf(page
, "%s\n", mddev
->clevel
);
3280 else if (mddev
->level
!= LEVEL_NONE
)
3281 return sprintf(page
, "%d\n", mddev
->level
);
3287 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3291 struct mdk_personality
*pers
;
3294 struct md_rdev
*rdev
;
3296 if (mddev
->pers
== NULL
) {
3299 if (len
>= sizeof(mddev
->clevel
))
3301 strncpy(mddev
->clevel
, buf
, len
);
3302 if (mddev
->clevel
[len
-1] == '\n')
3304 mddev
->clevel
[len
] = 0;
3305 mddev
->level
= LEVEL_NONE
;
3309 /* request to change the personality. Need to ensure:
3310 * - array is not engaged in resync/recovery/reshape
3311 * - old personality can be suspended
3312 * - new personality will access other array.
3315 if (mddev
->sync_thread
||
3316 mddev
->reshape_position
!= MaxSector
||
3317 mddev
->sysfs_active
)
3320 if (!mddev
->pers
->quiesce
) {
3321 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3322 mdname(mddev
), mddev
->pers
->name
);
3326 /* Now find the new personality */
3327 if (len
== 0 || len
>= sizeof(clevel
))
3329 strncpy(clevel
, buf
, len
);
3330 if (clevel
[len
-1] == '\n')
3333 if (strict_strtol(clevel
, 10, &level
))
3336 if (request_module("md-%s", clevel
) != 0)
3337 request_module("md-level-%s", clevel
);
3338 spin_lock(&pers_lock
);
3339 pers
= find_pers(level
, clevel
);
3340 if (!pers
|| !try_module_get(pers
->owner
)) {
3341 spin_unlock(&pers_lock
);
3342 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3345 spin_unlock(&pers_lock
);
3347 if (pers
== mddev
->pers
) {
3348 /* Nothing to do! */
3349 module_put(pers
->owner
);
3352 if (!pers
->takeover
) {
3353 module_put(pers
->owner
);
3354 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3355 mdname(mddev
), clevel
);
3359 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3360 rdev
->new_raid_disk
= rdev
->raid_disk
;
3362 /* ->takeover must set new_* and/or delta_disks
3363 * if it succeeds, and may set them when it fails.
3365 priv
= pers
->takeover(mddev
);
3367 mddev
->new_level
= mddev
->level
;
3368 mddev
->new_layout
= mddev
->layout
;
3369 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3370 mddev
->raid_disks
-= mddev
->delta_disks
;
3371 mddev
->delta_disks
= 0;
3372 module_put(pers
->owner
);
3373 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3374 mdname(mddev
), clevel
);
3375 return PTR_ERR(priv
);
3378 /* Looks like we have a winner */
3379 mddev_suspend(mddev
);
3380 mddev
->pers
->stop(mddev
);
3382 if (mddev
->pers
->sync_request
== NULL
&&
3383 pers
->sync_request
!= NULL
) {
3384 /* need to add the md_redundancy_group */
3385 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3387 "md: cannot register extra attributes for %s\n",
3389 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3391 if (mddev
->pers
->sync_request
!= NULL
&&
3392 pers
->sync_request
== NULL
) {
3393 /* need to remove the md_redundancy_group */
3394 if (mddev
->to_remove
== NULL
)
3395 mddev
->to_remove
= &md_redundancy_group
;
3398 if (mddev
->pers
->sync_request
== NULL
&&
3400 /* We are converting from a no-redundancy array
3401 * to a redundancy array and metadata is managed
3402 * externally so we need to be sure that writes
3403 * won't block due to a need to transition
3405 * until external management is started.
3408 mddev
->safemode_delay
= 0;
3409 mddev
->safemode
= 0;
3412 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3413 if (rdev
->raid_disk
< 0)
3415 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3416 rdev
->new_raid_disk
= -1;
3417 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3419 sysfs_unlink_rdev(mddev
, rdev
);
3421 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3422 if (rdev
->raid_disk
< 0)
3424 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3426 rdev
->raid_disk
= rdev
->new_raid_disk
;
3427 if (rdev
->raid_disk
< 0)
3428 clear_bit(In_sync
, &rdev
->flags
);
3430 if (sysfs_link_rdev(mddev
, rdev
))
3431 printk(KERN_WARNING
"md: cannot register rd%d"
3432 " for %s after level change\n",
3433 rdev
->raid_disk
, mdname(mddev
));
3437 module_put(mddev
->pers
->owner
);
3439 mddev
->private = priv
;
3440 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3441 mddev
->level
= mddev
->new_level
;
3442 mddev
->layout
= mddev
->new_layout
;
3443 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3444 mddev
->delta_disks
= 0;
3445 mddev
->degraded
= 0;
3446 if (mddev
->pers
->sync_request
== NULL
) {
3447 /* this is now an array without redundancy, so
3448 * it must always be in_sync
3451 del_timer_sync(&mddev
->safemode_timer
);
3454 mddev_resume(mddev
);
3455 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3456 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3457 md_wakeup_thread(mddev
->thread
);
3458 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3459 md_new_event(mddev
);
3463 static struct md_sysfs_entry md_level
=
3464 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3468 layout_show(struct mddev
*mddev
, char *page
)
3470 /* just a number, not meaningful for all levels */
3471 if (mddev
->reshape_position
!= MaxSector
&&
3472 mddev
->layout
!= mddev
->new_layout
)
3473 return sprintf(page
, "%d (%d)\n",
3474 mddev
->new_layout
, mddev
->layout
);
3475 return sprintf(page
, "%d\n", mddev
->layout
);
3479 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3482 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3484 if (!*buf
|| (*e
&& *e
!= '\n'))
3489 if (mddev
->pers
->check_reshape
== NULL
)
3491 mddev
->new_layout
= n
;
3492 err
= mddev
->pers
->check_reshape(mddev
);
3494 mddev
->new_layout
= mddev
->layout
;
3498 mddev
->new_layout
= n
;
3499 if (mddev
->reshape_position
== MaxSector
)
3504 static struct md_sysfs_entry md_layout
=
3505 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3509 raid_disks_show(struct mddev
*mddev
, char *page
)
3511 if (mddev
->raid_disks
== 0)
3513 if (mddev
->reshape_position
!= MaxSector
&&
3514 mddev
->delta_disks
!= 0)
3515 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3516 mddev
->raid_disks
- mddev
->delta_disks
);
3517 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3520 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3523 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3527 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3529 if (!*buf
|| (*e
&& *e
!= '\n'))
3533 rv
= update_raid_disks(mddev
, n
);
3534 else if (mddev
->reshape_position
!= MaxSector
) {
3535 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3536 mddev
->delta_disks
= n
- olddisks
;
3537 mddev
->raid_disks
= n
;
3539 mddev
->raid_disks
= n
;
3540 return rv
? rv
: len
;
3542 static struct md_sysfs_entry md_raid_disks
=
3543 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3546 chunk_size_show(struct mddev
*mddev
, char *page
)
3548 if (mddev
->reshape_position
!= MaxSector
&&
3549 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3550 return sprintf(page
, "%d (%d)\n",
3551 mddev
->new_chunk_sectors
<< 9,
3552 mddev
->chunk_sectors
<< 9);
3553 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3557 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3560 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3562 if (!*buf
|| (*e
&& *e
!= '\n'))
3567 if (mddev
->pers
->check_reshape
== NULL
)
3569 mddev
->new_chunk_sectors
= n
>> 9;
3570 err
= mddev
->pers
->check_reshape(mddev
);
3572 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3576 mddev
->new_chunk_sectors
= n
>> 9;
3577 if (mddev
->reshape_position
== MaxSector
)
3578 mddev
->chunk_sectors
= n
>> 9;
3582 static struct md_sysfs_entry md_chunk_size
=
3583 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3586 resync_start_show(struct mddev
*mddev
, char *page
)
3588 if (mddev
->recovery_cp
== MaxSector
)
3589 return sprintf(page
, "none\n");
3590 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3594 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3597 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3599 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3601 if (cmd_match(buf
, "none"))
3603 else if (!*buf
|| (*e
&& *e
!= '\n'))
3606 mddev
->recovery_cp
= n
;
3609 static struct md_sysfs_entry md_resync_start
=
3610 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3613 * The array state can be:
3616 * No devices, no size, no level
3617 * Equivalent to STOP_ARRAY ioctl
3619 * May have some settings, but array is not active
3620 * all IO results in error
3621 * When written, doesn't tear down array, but just stops it
3622 * suspended (not supported yet)
3623 * All IO requests will block. The array can be reconfigured.
3624 * Writing this, if accepted, will block until array is quiescent
3626 * no resync can happen. no superblocks get written.
3627 * write requests fail
3629 * like readonly, but behaves like 'clean' on a write request.
3631 * clean - no pending writes, but otherwise active.
3632 * When written to inactive array, starts without resync
3633 * If a write request arrives then
3634 * if metadata is known, mark 'dirty' and switch to 'active'.
3635 * if not known, block and switch to write-pending
3636 * If written to an active array that has pending writes, then fails.
3638 * fully active: IO and resync can be happening.
3639 * When written to inactive array, starts with resync
3642 * clean, but writes are blocked waiting for 'active' to be written.
3645 * like active, but no writes have been seen for a while (100msec).
3648 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3649 write_pending
, active_idle
, bad_word
};
3650 static char *array_states
[] = {
3651 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3652 "write-pending", "active-idle", NULL
};
3654 static int match_word(const char *word
, char **list
)
3657 for (n
=0; list
[n
]; n
++)
3658 if (cmd_match(word
, list
[n
]))
3664 array_state_show(struct mddev
*mddev
, char *page
)
3666 enum array_state st
= inactive
;
3679 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3681 else if (mddev
->safemode
)
3687 if (list_empty(&mddev
->disks
) &&
3688 mddev
->raid_disks
== 0 &&
3689 mddev
->dev_sectors
== 0)
3694 return sprintf(page
, "%s\n", array_states
[st
]);
3697 static int do_md_stop(struct mddev
* mddev
, int ro
, int is_open
);
3698 static int md_set_readonly(struct mddev
* mddev
, int is_open
);
3699 static int do_md_run(struct mddev
* mddev
);
3700 static int restart_array(struct mddev
*mddev
);
3703 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3706 enum array_state st
= match_word(buf
, array_states
);
3711 /* stopping an active array */
3712 if (atomic_read(&mddev
->openers
) > 0)
3714 err
= do_md_stop(mddev
, 0, 0);
3717 /* stopping an active array */
3719 if (atomic_read(&mddev
->openers
) > 0)
3721 err
= do_md_stop(mddev
, 2, 0);
3723 err
= 0; /* already inactive */
3726 break; /* not supported yet */
3729 err
= md_set_readonly(mddev
, 0);
3732 set_disk_ro(mddev
->gendisk
, 1);
3733 err
= do_md_run(mddev
);
3739 err
= md_set_readonly(mddev
, 0);
3740 else if (mddev
->ro
== 1)
3741 err
= restart_array(mddev
);
3744 set_disk_ro(mddev
->gendisk
, 0);
3748 err
= do_md_run(mddev
);
3753 restart_array(mddev
);
3754 spin_lock_irq(&mddev
->write_lock
);
3755 if (atomic_read(&mddev
->writes_pending
) == 0) {
3756 if (mddev
->in_sync
== 0) {
3758 if (mddev
->safemode
== 1)
3759 mddev
->safemode
= 0;
3760 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3765 spin_unlock_irq(&mddev
->write_lock
);
3771 restart_array(mddev
);
3772 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3773 wake_up(&mddev
->sb_wait
);
3777 set_disk_ro(mddev
->gendisk
, 0);
3778 err
= do_md_run(mddev
);
3783 /* these cannot be set */
3789 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3793 static struct md_sysfs_entry md_array_state
=
3794 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3797 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3798 return sprintf(page
, "%d\n",
3799 atomic_read(&mddev
->max_corr_read_errors
));
3803 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3806 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3808 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3809 atomic_set(&mddev
->max_corr_read_errors
, n
);
3815 static struct md_sysfs_entry max_corr_read_errors
=
3816 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3817 max_corrected_read_errors_store
);
3820 null_show(struct mddev
*mddev
, char *page
)
3826 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3828 /* buf must be %d:%d\n? giving major and minor numbers */
3829 /* The new device is added to the array.
3830 * If the array has a persistent superblock, we read the
3831 * superblock to initialise info and check validity.
3832 * Otherwise, only checking done is that in bind_rdev_to_array,
3833 * which mainly checks size.
3836 int major
= simple_strtoul(buf
, &e
, 10);
3839 struct md_rdev
*rdev
;
3842 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3844 minor
= simple_strtoul(e
+1, &e
, 10);
3845 if (*e
&& *e
!= '\n')
3847 dev
= MKDEV(major
, minor
);
3848 if (major
!= MAJOR(dev
) ||
3849 minor
!= MINOR(dev
))
3853 if (mddev
->persistent
) {
3854 rdev
= md_import_device(dev
, mddev
->major_version
,
3855 mddev
->minor_version
);
3856 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3857 struct md_rdev
*rdev0
3858 = list_entry(mddev
->disks
.next
,
3859 struct md_rdev
, same_set
);
3860 err
= super_types
[mddev
->major_version
]
3861 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3865 } else if (mddev
->external
)
3866 rdev
= md_import_device(dev
, -2, -1);
3868 rdev
= md_import_device(dev
, -1, -1);
3871 return PTR_ERR(rdev
);
3872 err
= bind_rdev_to_array(rdev
, mddev
);
3876 return err
? err
: len
;
3879 static struct md_sysfs_entry md_new_device
=
3880 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3883 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3886 unsigned long chunk
, end_chunk
;
3890 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3892 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3893 if (buf
== end
) break;
3894 if (*end
== '-') { /* range */
3896 end_chunk
= simple_strtoul(buf
, &end
, 0);
3897 if (buf
== end
) break;
3899 if (*end
&& !isspace(*end
)) break;
3900 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3901 buf
= skip_spaces(end
);
3903 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3908 static struct md_sysfs_entry md_bitmap
=
3909 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3912 size_show(struct mddev
*mddev
, char *page
)
3914 return sprintf(page
, "%llu\n",
3915 (unsigned long long)mddev
->dev_sectors
/ 2);
3918 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
3921 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3923 /* If array is inactive, we can reduce the component size, but
3924 * not increase it (except from 0).
3925 * If array is active, we can try an on-line resize
3928 int err
= strict_blocks_to_sectors(buf
, §ors
);
3933 err
= update_size(mddev
, sectors
);
3934 md_update_sb(mddev
, 1);
3936 if (mddev
->dev_sectors
== 0 ||
3937 mddev
->dev_sectors
> sectors
)
3938 mddev
->dev_sectors
= sectors
;
3942 return err
? err
: len
;
3945 static struct md_sysfs_entry md_size
=
3946 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3951 * 'none' for arrays with no metadata (good luck...)
3952 * 'external' for arrays with externally managed metadata,
3953 * or N.M for internally known formats
3956 metadata_show(struct mddev
*mddev
, char *page
)
3958 if (mddev
->persistent
)
3959 return sprintf(page
, "%d.%d\n",
3960 mddev
->major_version
, mddev
->minor_version
);
3961 else if (mddev
->external
)
3962 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3964 return sprintf(page
, "none\n");
3968 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3972 /* Changing the details of 'external' metadata is
3973 * always permitted. Otherwise there must be
3974 * no devices attached to the array.
3976 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3978 else if (!list_empty(&mddev
->disks
))
3981 if (cmd_match(buf
, "none")) {
3982 mddev
->persistent
= 0;
3983 mddev
->external
= 0;
3984 mddev
->major_version
= 0;
3985 mddev
->minor_version
= 90;
3988 if (strncmp(buf
, "external:", 9) == 0) {
3989 size_t namelen
= len
-9;
3990 if (namelen
>= sizeof(mddev
->metadata_type
))
3991 namelen
= sizeof(mddev
->metadata_type
)-1;
3992 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3993 mddev
->metadata_type
[namelen
] = 0;
3994 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3995 mddev
->metadata_type
[--namelen
] = 0;
3996 mddev
->persistent
= 0;
3997 mddev
->external
= 1;
3998 mddev
->major_version
= 0;
3999 mddev
->minor_version
= 90;
4002 major
= simple_strtoul(buf
, &e
, 10);
4003 if (e
==buf
|| *e
!= '.')
4006 minor
= simple_strtoul(buf
, &e
, 10);
4007 if (e
==buf
|| (*e
&& *e
!= '\n') )
4009 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4011 mddev
->major_version
= major
;
4012 mddev
->minor_version
= minor
;
4013 mddev
->persistent
= 1;
4014 mddev
->external
= 0;
4018 static struct md_sysfs_entry md_metadata
=
4019 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4022 action_show(struct mddev
*mddev
, char *page
)
4024 char *type
= "idle";
4025 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4027 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4028 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4029 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4031 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4032 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4034 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4038 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4041 return sprintf(page
, "%s\n", type
);
4044 static void reap_sync_thread(struct mddev
*mddev
);
4047 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4049 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4052 if (cmd_match(page
, "frozen"))
4053 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4055 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4057 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4058 if (mddev
->sync_thread
) {
4059 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4060 reap_sync_thread(mddev
);
4062 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4063 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4065 else if (cmd_match(page
, "resync"))
4066 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4067 else if (cmd_match(page
, "recover")) {
4068 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4069 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4070 } else if (cmd_match(page
, "reshape")) {
4072 if (mddev
->pers
->start_reshape
== NULL
)
4074 err
= mddev
->pers
->start_reshape(mddev
);
4077 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4079 if (cmd_match(page
, "check"))
4080 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4081 else if (!cmd_match(page
, "repair"))
4083 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4084 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4086 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4087 md_wakeup_thread(mddev
->thread
);
4088 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4093 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4095 return sprintf(page
, "%llu\n",
4096 (unsigned long long) mddev
->resync_mismatches
);
4099 static struct md_sysfs_entry md_scan_mode
=
4100 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4103 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4106 sync_min_show(struct mddev
*mddev
, char *page
)
4108 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4109 mddev
->sync_speed_min
? "local": "system");
4113 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4117 if (strncmp(buf
, "system", 6)==0) {
4118 mddev
->sync_speed_min
= 0;
4121 min
= simple_strtoul(buf
, &e
, 10);
4122 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4124 mddev
->sync_speed_min
= min
;
4128 static struct md_sysfs_entry md_sync_min
=
4129 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4132 sync_max_show(struct mddev
*mddev
, char *page
)
4134 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4135 mddev
->sync_speed_max
? "local": "system");
4139 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4143 if (strncmp(buf
, "system", 6)==0) {
4144 mddev
->sync_speed_max
= 0;
4147 max
= simple_strtoul(buf
, &e
, 10);
4148 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4150 mddev
->sync_speed_max
= max
;
4154 static struct md_sysfs_entry md_sync_max
=
4155 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4158 degraded_show(struct mddev
*mddev
, char *page
)
4160 return sprintf(page
, "%d\n", mddev
->degraded
);
4162 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4165 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4167 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4171 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4175 if (strict_strtol(buf
, 10, &n
))
4178 if (n
!= 0 && n
!= 1)
4181 mddev
->parallel_resync
= n
;
4183 if (mddev
->sync_thread
)
4184 wake_up(&resync_wait
);
4189 /* force parallel resync, even with shared block devices */
4190 static struct md_sysfs_entry md_sync_force_parallel
=
4191 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4192 sync_force_parallel_show
, sync_force_parallel_store
);
4195 sync_speed_show(struct mddev
*mddev
, char *page
)
4197 unsigned long resync
, dt
, db
;
4198 if (mddev
->curr_resync
== 0)
4199 return sprintf(page
, "none\n");
4200 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4201 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4203 db
= resync
- mddev
->resync_mark_cnt
;
4204 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4207 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4210 sync_completed_show(struct mddev
*mddev
, char *page
)
4212 unsigned long long max_sectors
, resync
;
4214 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4215 return sprintf(page
, "none\n");
4217 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4218 max_sectors
= mddev
->resync_max_sectors
;
4220 max_sectors
= mddev
->dev_sectors
;
4222 resync
= mddev
->curr_resync_completed
;
4223 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4226 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4229 min_sync_show(struct mddev
*mddev
, char *page
)
4231 return sprintf(page
, "%llu\n",
4232 (unsigned long long)mddev
->resync_min
);
4235 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4237 unsigned long long min
;
4238 if (strict_strtoull(buf
, 10, &min
))
4240 if (min
> mddev
->resync_max
)
4242 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4245 /* Must be a multiple of chunk_size */
4246 if (mddev
->chunk_sectors
) {
4247 sector_t temp
= min
;
4248 if (sector_div(temp
, mddev
->chunk_sectors
))
4251 mddev
->resync_min
= min
;
4256 static struct md_sysfs_entry md_min_sync
=
4257 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4260 max_sync_show(struct mddev
*mddev
, char *page
)
4262 if (mddev
->resync_max
== MaxSector
)
4263 return sprintf(page
, "max\n");
4265 return sprintf(page
, "%llu\n",
4266 (unsigned long long)mddev
->resync_max
);
4269 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4271 if (strncmp(buf
, "max", 3) == 0)
4272 mddev
->resync_max
= MaxSector
;
4274 unsigned long long max
;
4275 if (strict_strtoull(buf
, 10, &max
))
4277 if (max
< mddev
->resync_min
)
4279 if (max
< mddev
->resync_max
&&
4281 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4284 /* Must be a multiple of chunk_size */
4285 if (mddev
->chunk_sectors
) {
4286 sector_t temp
= max
;
4287 if (sector_div(temp
, mddev
->chunk_sectors
))
4290 mddev
->resync_max
= max
;
4292 wake_up(&mddev
->recovery_wait
);
4296 static struct md_sysfs_entry md_max_sync
=
4297 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4300 suspend_lo_show(struct mddev
*mddev
, char *page
)
4302 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4306 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4309 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4310 unsigned long long old
= mddev
->suspend_lo
;
4312 if (mddev
->pers
== NULL
||
4313 mddev
->pers
->quiesce
== NULL
)
4315 if (buf
== e
|| (*e
&& *e
!= '\n'))
4318 mddev
->suspend_lo
= new;
4320 /* Shrinking suspended region */
4321 mddev
->pers
->quiesce(mddev
, 2);
4323 /* Expanding suspended region - need to wait */
4324 mddev
->pers
->quiesce(mddev
, 1);
4325 mddev
->pers
->quiesce(mddev
, 0);
4329 static struct md_sysfs_entry md_suspend_lo
=
4330 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4334 suspend_hi_show(struct mddev
*mddev
, char *page
)
4336 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4340 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4343 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4344 unsigned long long old
= mddev
->suspend_hi
;
4346 if (mddev
->pers
== NULL
||
4347 mddev
->pers
->quiesce
== NULL
)
4349 if (buf
== e
|| (*e
&& *e
!= '\n'))
4352 mddev
->suspend_hi
= new;
4354 /* Shrinking suspended region */
4355 mddev
->pers
->quiesce(mddev
, 2);
4357 /* Expanding suspended region - need to wait */
4358 mddev
->pers
->quiesce(mddev
, 1);
4359 mddev
->pers
->quiesce(mddev
, 0);
4363 static struct md_sysfs_entry md_suspend_hi
=
4364 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4367 reshape_position_show(struct mddev
*mddev
, char *page
)
4369 if (mddev
->reshape_position
!= MaxSector
)
4370 return sprintf(page
, "%llu\n",
4371 (unsigned long long)mddev
->reshape_position
);
4372 strcpy(page
, "none\n");
4377 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4380 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4383 if (buf
== e
|| (*e
&& *e
!= '\n'))
4385 mddev
->reshape_position
= new;
4386 mddev
->delta_disks
= 0;
4387 mddev
->new_level
= mddev
->level
;
4388 mddev
->new_layout
= mddev
->layout
;
4389 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4393 static struct md_sysfs_entry md_reshape_position
=
4394 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4395 reshape_position_store
);
4398 array_size_show(struct mddev
*mddev
, char *page
)
4400 if (mddev
->external_size
)
4401 return sprintf(page
, "%llu\n",
4402 (unsigned long long)mddev
->array_sectors
/2);
4404 return sprintf(page
, "default\n");
4408 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4412 if (strncmp(buf
, "default", 7) == 0) {
4414 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4416 sectors
= mddev
->array_sectors
;
4418 mddev
->external_size
= 0;
4420 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4422 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4425 mddev
->external_size
= 1;
4428 mddev
->array_sectors
= sectors
;
4430 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4431 revalidate_disk(mddev
->gendisk
);
4436 static struct md_sysfs_entry md_array_size
=
4437 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4440 static struct attribute
*md_default_attrs
[] = {
4443 &md_raid_disks
.attr
,
4444 &md_chunk_size
.attr
,
4446 &md_resync_start
.attr
,
4448 &md_new_device
.attr
,
4449 &md_safe_delay
.attr
,
4450 &md_array_state
.attr
,
4451 &md_reshape_position
.attr
,
4452 &md_array_size
.attr
,
4453 &max_corr_read_errors
.attr
,
4457 static struct attribute
*md_redundancy_attrs
[] = {
4459 &md_mismatches
.attr
,
4462 &md_sync_speed
.attr
,
4463 &md_sync_force_parallel
.attr
,
4464 &md_sync_completed
.attr
,
4467 &md_suspend_lo
.attr
,
4468 &md_suspend_hi
.attr
,
4473 static struct attribute_group md_redundancy_group
= {
4475 .attrs
= md_redundancy_attrs
,
4480 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4482 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4483 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4488 rv
= mddev_lock(mddev
);
4490 rv
= entry
->show(mddev
, page
);
4491 mddev_unlock(mddev
);
4497 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4498 const char *page
, size_t length
)
4500 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4501 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4506 if (!capable(CAP_SYS_ADMIN
))
4508 rv
= mddev_lock(mddev
);
4509 if (mddev
->hold_active
== UNTIL_IOCTL
)
4510 mddev
->hold_active
= 0;
4512 rv
= entry
->store(mddev
, page
, length
);
4513 mddev_unlock(mddev
);
4518 static void md_free(struct kobject
*ko
)
4520 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4522 if (mddev
->sysfs_state
)
4523 sysfs_put(mddev
->sysfs_state
);
4525 if (mddev
->gendisk
) {
4526 del_gendisk(mddev
->gendisk
);
4527 put_disk(mddev
->gendisk
);
4530 blk_cleanup_queue(mddev
->queue
);
4535 static const struct sysfs_ops md_sysfs_ops
= {
4536 .show
= md_attr_show
,
4537 .store
= md_attr_store
,
4539 static struct kobj_type md_ktype
= {
4541 .sysfs_ops
= &md_sysfs_ops
,
4542 .default_attrs
= md_default_attrs
,
4547 static void mddev_delayed_delete(struct work_struct
*ws
)
4549 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4551 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4552 kobject_del(&mddev
->kobj
);
4553 kobject_put(&mddev
->kobj
);
4556 static int md_alloc(dev_t dev
, char *name
)
4558 static DEFINE_MUTEX(disks_mutex
);
4559 struct mddev
*mddev
= mddev_find(dev
);
4560 struct gendisk
*disk
;
4569 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4570 shift
= partitioned
? MdpMinorShift
: 0;
4571 unit
= MINOR(mddev
->unit
) >> shift
;
4573 /* wait for any previous instance of this device to be
4574 * completely removed (mddev_delayed_delete).
4576 flush_workqueue(md_misc_wq
);
4578 mutex_lock(&disks_mutex
);
4584 /* Need to ensure that 'name' is not a duplicate.
4586 struct mddev
*mddev2
;
4587 spin_lock(&all_mddevs_lock
);
4589 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4590 if (mddev2
->gendisk
&&
4591 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4592 spin_unlock(&all_mddevs_lock
);
4595 spin_unlock(&all_mddevs_lock
);
4599 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4602 mddev
->queue
->queuedata
= mddev
;
4604 blk_queue_make_request(mddev
->queue
, md_make_request
);
4606 disk
= alloc_disk(1 << shift
);
4608 blk_cleanup_queue(mddev
->queue
);
4609 mddev
->queue
= NULL
;
4612 disk
->major
= MAJOR(mddev
->unit
);
4613 disk
->first_minor
= unit
<< shift
;
4615 strcpy(disk
->disk_name
, name
);
4616 else if (partitioned
)
4617 sprintf(disk
->disk_name
, "md_d%d", unit
);
4619 sprintf(disk
->disk_name
, "md%d", unit
);
4620 disk
->fops
= &md_fops
;
4621 disk
->private_data
= mddev
;
4622 disk
->queue
= mddev
->queue
;
4623 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4624 /* Allow extended partitions. This makes the
4625 * 'mdp' device redundant, but we can't really
4628 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4629 mddev
->gendisk
= disk
;
4630 /* As soon as we call add_disk(), another thread could get
4631 * through to md_open, so make sure it doesn't get too far
4633 mutex_lock(&mddev
->open_mutex
);
4636 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4637 &disk_to_dev(disk
)->kobj
, "%s", "md");
4639 /* This isn't possible, but as kobject_init_and_add is marked
4640 * __must_check, we must do something with the result
4642 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4646 if (mddev
->kobj
.sd
&&
4647 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4648 printk(KERN_DEBUG
"pointless warning\n");
4649 mutex_unlock(&mddev
->open_mutex
);
4651 mutex_unlock(&disks_mutex
);
4652 if (!error
&& mddev
->kobj
.sd
) {
4653 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4654 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4660 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4662 md_alloc(dev
, NULL
);
4666 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4668 /* val must be "md_*" where * is not all digits.
4669 * We allocate an array with a large free minor number, and
4670 * set the name to val. val must not already be an active name.
4672 int len
= strlen(val
);
4673 char buf
[DISK_NAME_LEN
];
4675 while (len
&& val
[len
-1] == '\n')
4677 if (len
>= DISK_NAME_LEN
)
4679 strlcpy(buf
, val
, len
+1);
4680 if (strncmp(buf
, "md_", 3) != 0)
4682 return md_alloc(0, buf
);
4685 static void md_safemode_timeout(unsigned long data
)
4687 struct mddev
*mddev
= (struct mddev
*) data
;
4689 if (!atomic_read(&mddev
->writes_pending
)) {
4690 mddev
->safemode
= 1;
4691 if (mddev
->external
)
4692 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4694 md_wakeup_thread(mddev
->thread
);
4697 static int start_dirty_degraded
;
4699 int md_run(struct mddev
*mddev
)
4702 struct md_rdev
*rdev
;
4703 struct mdk_personality
*pers
;
4705 if (list_empty(&mddev
->disks
))
4706 /* cannot run an array with no devices.. */
4711 /* Cannot run until previous stop completes properly */
4712 if (mddev
->sysfs_active
)
4716 * Analyze all RAID superblock(s)
4718 if (!mddev
->raid_disks
) {
4719 if (!mddev
->persistent
)
4724 if (mddev
->level
!= LEVEL_NONE
)
4725 request_module("md-level-%d", mddev
->level
);
4726 else if (mddev
->clevel
[0])
4727 request_module("md-%s", mddev
->clevel
);
4730 * Drop all container device buffers, from now on
4731 * the only valid external interface is through the md
4734 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4735 if (test_bit(Faulty
, &rdev
->flags
))
4737 sync_blockdev(rdev
->bdev
);
4738 invalidate_bdev(rdev
->bdev
);
4740 /* perform some consistency tests on the device.
4741 * We don't want the data to overlap the metadata,
4742 * Internal Bitmap issues have been handled elsewhere.
4744 if (rdev
->meta_bdev
) {
4745 /* Nothing to check */;
4746 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4747 if (mddev
->dev_sectors
&&
4748 rdev
->data_offset
+ mddev
->dev_sectors
4750 printk("md: %s: data overlaps metadata\n",
4755 if (rdev
->sb_start
+ rdev
->sb_size
/512
4756 > rdev
->data_offset
) {
4757 printk("md: %s: metadata overlaps data\n",
4762 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4765 if (mddev
->bio_set
== NULL
)
4766 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
4767 sizeof(struct mddev
*));
4769 spin_lock(&pers_lock
);
4770 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4771 if (!pers
|| !try_module_get(pers
->owner
)) {
4772 spin_unlock(&pers_lock
);
4773 if (mddev
->level
!= LEVEL_NONE
)
4774 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4777 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4782 spin_unlock(&pers_lock
);
4783 if (mddev
->level
!= pers
->level
) {
4784 mddev
->level
= pers
->level
;
4785 mddev
->new_level
= pers
->level
;
4787 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4789 if (mddev
->reshape_position
!= MaxSector
&&
4790 pers
->start_reshape
== NULL
) {
4791 /* This personality cannot handle reshaping... */
4793 module_put(pers
->owner
);
4797 if (pers
->sync_request
) {
4798 /* Warn if this is a potentially silly
4801 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4802 struct md_rdev
*rdev2
;
4805 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4806 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4808 rdev
->bdev
->bd_contains
==
4809 rdev2
->bdev
->bd_contains
) {
4811 "%s: WARNING: %s appears to be"
4812 " on the same physical disk as"
4815 bdevname(rdev
->bdev
,b
),
4816 bdevname(rdev2
->bdev
,b2
));
4823 "True protection against single-disk"
4824 " failure might be compromised.\n");
4827 mddev
->recovery
= 0;
4828 /* may be over-ridden by personality */
4829 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4831 mddev
->ok_start_degraded
= start_dirty_degraded
;
4833 if (start_readonly
&& mddev
->ro
== 0)
4834 mddev
->ro
= 2; /* read-only, but switch on first write */
4836 err
= mddev
->pers
->run(mddev
);
4838 printk(KERN_ERR
"md: pers->run() failed ...\n");
4839 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4840 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4841 " but 'external_size' not in effect?\n", __func__
);
4843 "md: invalid array_size %llu > default size %llu\n",
4844 (unsigned long long)mddev
->array_sectors
/ 2,
4845 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4847 mddev
->pers
->stop(mddev
);
4849 if (err
== 0 && mddev
->pers
->sync_request
) {
4850 err
= bitmap_create(mddev
);
4852 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4853 mdname(mddev
), err
);
4854 mddev
->pers
->stop(mddev
);
4858 module_put(mddev
->pers
->owner
);
4860 bitmap_destroy(mddev
);
4863 if (mddev
->pers
->sync_request
) {
4864 if (mddev
->kobj
.sd
&&
4865 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4867 "md: cannot register extra attributes for %s\n",
4869 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4870 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4873 atomic_set(&mddev
->writes_pending
,0);
4874 atomic_set(&mddev
->max_corr_read_errors
,
4875 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4876 mddev
->safemode
= 0;
4877 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4878 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4879 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4883 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4884 if (rdev
->raid_disk
>= 0)
4885 if (sysfs_link_rdev(mddev
, rdev
))
4886 /* failure here is OK */;
4888 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4891 md_update_sb(mddev
, 0);
4893 md_new_event(mddev
);
4894 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4895 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4896 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4899 EXPORT_SYMBOL_GPL(md_run
);
4901 static int do_md_run(struct mddev
*mddev
)
4905 err
= md_run(mddev
);
4908 err
= bitmap_load(mddev
);
4910 bitmap_destroy(mddev
);
4914 md_wakeup_thread(mddev
->thread
);
4915 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4917 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4918 revalidate_disk(mddev
->gendisk
);
4920 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4925 static int restart_array(struct mddev
*mddev
)
4927 struct gendisk
*disk
= mddev
->gendisk
;
4929 /* Complain if it has no devices */
4930 if (list_empty(&mddev
->disks
))
4936 mddev
->safemode
= 0;
4938 set_disk_ro(disk
, 0);
4939 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4941 /* Kick recovery or resync if necessary */
4942 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4943 md_wakeup_thread(mddev
->thread
);
4944 md_wakeup_thread(mddev
->sync_thread
);
4945 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4949 /* similar to deny_write_access, but accounts for our holding a reference
4950 * to the file ourselves */
4951 static int deny_bitmap_write_access(struct file
* file
)
4953 struct inode
*inode
= file
->f_mapping
->host
;
4955 spin_lock(&inode
->i_lock
);
4956 if (atomic_read(&inode
->i_writecount
) > 1) {
4957 spin_unlock(&inode
->i_lock
);
4960 atomic_set(&inode
->i_writecount
, -1);
4961 spin_unlock(&inode
->i_lock
);
4966 void restore_bitmap_write_access(struct file
*file
)
4968 struct inode
*inode
= file
->f_mapping
->host
;
4970 spin_lock(&inode
->i_lock
);
4971 atomic_set(&inode
->i_writecount
, 1);
4972 spin_unlock(&inode
->i_lock
);
4975 static void md_clean(struct mddev
*mddev
)
4977 mddev
->array_sectors
= 0;
4978 mddev
->external_size
= 0;
4979 mddev
->dev_sectors
= 0;
4980 mddev
->raid_disks
= 0;
4981 mddev
->recovery_cp
= 0;
4982 mddev
->resync_min
= 0;
4983 mddev
->resync_max
= MaxSector
;
4984 mddev
->reshape_position
= MaxSector
;
4985 mddev
->external
= 0;
4986 mddev
->persistent
= 0;
4987 mddev
->level
= LEVEL_NONE
;
4988 mddev
->clevel
[0] = 0;
4991 mddev
->metadata_type
[0] = 0;
4992 mddev
->chunk_sectors
= 0;
4993 mddev
->ctime
= mddev
->utime
= 0;
4995 mddev
->max_disks
= 0;
4997 mddev
->can_decrease_events
= 0;
4998 mddev
->delta_disks
= 0;
4999 mddev
->new_level
= LEVEL_NONE
;
5000 mddev
->new_layout
= 0;
5001 mddev
->new_chunk_sectors
= 0;
5002 mddev
->curr_resync
= 0;
5003 mddev
->resync_mismatches
= 0;
5004 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5005 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5006 mddev
->recovery
= 0;
5009 mddev
->degraded
= 0;
5010 mddev
->safemode
= 0;
5011 mddev
->bitmap_info
.offset
= 0;
5012 mddev
->bitmap_info
.default_offset
= 0;
5013 mddev
->bitmap_info
.chunksize
= 0;
5014 mddev
->bitmap_info
.daemon_sleep
= 0;
5015 mddev
->bitmap_info
.max_write_behind
= 0;
5018 static void __md_stop_writes(struct mddev
*mddev
)
5020 if (mddev
->sync_thread
) {
5021 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5022 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5023 reap_sync_thread(mddev
);
5026 del_timer_sync(&mddev
->safemode_timer
);
5028 bitmap_flush(mddev
);
5029 md_super_wait(mddev
);
5031 if (!mddev
->in_sync
|| mddev
->flags
) {
5032 /* mark array as shutdown cleanly */
5034 md_update_sb(mddev
, 1);
5038 void md_stop_writes(struct mddev
*mddev
)
5041 __md_stop_writes(mddev
);
5042 mddev_unlock(mddev
);
5044 EXPORT_SYMBOL_GPL(md_stop_writes
);
5046 void md_stop(struct mddev
*mddev
)
5049 mddev
->pers
->stop(mddev
);
5050 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5051 mddev
->to_remove
= &md_redundancy_group
;
5052 module_put(mddev
->pers
->owner
);
5054 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5056 EXPORT_SYMBOL_GPL(md_stop
);
5058 static int md_set_readonly(struct mddev
*mddev
, int is_open
)
5061 mutex_lock(&mddev
->open_mutex
);
5062 if (atomic_read(&mddev
->openers
) > is_open
) {
5063 printk("md: %s still in use.\n",mdname(mddev
));
5068 __md_stop_writes(mddev
);
5074 set_disk_ro(mddev
->gendisk
, 1);
5075 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5076 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5080 mutex_unlock(&mddev
->open_mutex
);
5085 * 0 - completely stop and dis-assemble array
5086 * 2 - stop but do not disassemble array
5088 static int do_md_stop(struct mddev
* mddev
, int mode
, int is_open
)
5090 struct gendisk
*disk
= mddev
->gendisk
;
5091 struct md_rdev
*rdev
;
5093 mutex_lock(&mddev
->open_mutex
);
5094 if (atomic_read(&mddev
->openers
) > is_open
||
5095 mddev
->sysfs_active
) {
5096 printk("md: %s still in use.\n",mdname(mddev
));
5097 mutex_unlock(&mddev
->open_mutex
);
5103 set_disk_ro(disk
, 0);
5105 __md_stop_writes(mddev
);
5107 mddev
->queue
->merge_bvec_fn
= NULL
;
5108 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5110 /* tell userspace to handle 'inactive' */
5111 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5113 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5114 if (rdev
->raid_disk
>= 0)
5115 sysfs_unlink_rdev(mddev
, rdev
);
5117 set_capacity(disk
, 0);
5118 mutex_unlock(&mddev
->open_mutex
);
5120 revalidate_disk(disk
);
5125 mutex_unlock(&mddev
->open_mutex
);
5127 * Free resources if final stop
5130 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5132 bitmap_destroy(mddev
);
5133 if (mddev
->bitmap_info
.file
) {
5134 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5135 fput(mddev
->bitmap_info
.file
);
5136 mddev
->bitmap_info
.file
= NULL
;
5138 mddev
->bitmap_info
.offset
= 0;
5140 export_array(mddev
);
5143 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5144 if (mddev
->hold_active
== UNTIL_STOP
)
5145 mddev
->hold_active
= 0;
5147 blk_integrity_unregister(disk
);
5148 md_new_event(mddev
);
5149 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5154 static void autorun_array(struct mddev
*mddev
)
5156 struct md_rdev
*rdev
;
5159 if (list_empty(&mddev
->disks
))
5162 printk(KERN_INFO
"md: running: ");
5164 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5165 char b
[BDEVNAME_SIZE
];
5166 printk("<%s>", bdevname(rdev
->bdev
,b
));
5170 err
= do_md_run(mddev
);
5172 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5173 do_md_stop(mddev
, 0, 0);
5178 * lets try to run arrays based on all disks that have arrived
5179 * until now. (those are in pending_raid_disks)
5181 * the method: pick the first pending disk, collect all disks with
5182 * the same UUID, remove all from the pending list and put them into
5183 * the 'same_array' list. Then order this list based on superblock
5184 * update time (freshest comes first), kick out 'old' disks and
5185 * compare superblocks. If everything's fine then run it.
5187 * If "unit" is allocated, then bump its reference count
5189 static void autorun_devices(int part
)
5191 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5192 struct mddev
*mddev
;
5193 char b
[BDEVNAME_SIZE
];
5195 printk(KERN_INFO
"md: autorun ...\n");
5196 while (!list_empty(&pending_raid_disks
)) {
5199 LIST_HEAD(candidates
);
5200 rdev0
= list_entry(pending_raid_disks
.next
,
5201 struct md_rdev
, same_set
);
5203 printk(KERN_INFO
"md: considering %s ...\n",
5204 bdevname(rdev0
->bdev
,b
));
5205 INIT_LIST_HEAD(&candidates
);
5206 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5207 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5208 printk(KERN_INFO
"md: adding %s ...\n",
5209 bdevname(rdev
->bdev
,b
));
5210 list_move(&rdev
->same_set
, &candidates
);
5213 * now we have a set of devices, with all of them having
5214 * mostly sane superblocks. It's time to allocate the
5218 dev
= MKDEV(mdp_major
,
5219 rdev0
->preferred_minor
<< MdpMinorShift
);
5220 unit
= MINOR(dev
) >> MdpMinorShift
;
5222 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5225 if (rdev0
->preferred_minor
!= unit
) {
5226 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5227 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5231 md_probe(dev
, NULL
, NULL
);
5232 mddev
= mddev_find(dev
);
5233 if (!mddev
|| !mddev
->gendisk
) {
5237 "md: cannot allocate memory for md drive.\n");
5240 if (mddev_lock(mddev
))
5241 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5243 else if (mddev
->raid_disks
|| mddev
->major_version
5244 || !list_empty(&mddev
->disks
)) {
5246 "md: %s already running, cannot run %s\n",
5247 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5248 mddev_unlock(mddev
);
5250 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5251 mddev
->persistent
= 1;
5252 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5253 list_del_init(&rdev
->same_set
);
5254 if (bind_rdev_to_array(rdev
, mddev
))
5257 autorun_array(mddev
);
5258 mddev_unlock(mddev
);
5260 /* on success, candidates will be empty, on error
5263 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5264 list_del_init(&rdev
->same_set
);
5269 printk(KERN_INFO
"md: ... autorun DONE.\n");
5271 #endif /* !MODULE */
5273 static int get_version(void __user
* arg
)
5277 ver
.major
= MD_MAJOR_VERSION
;
5278 ver
.minor
= MD_MINOR_VERSION
;
5279 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5281 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5287 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5289 mdu_array_info_t info
;
5290 int nr
,working
,insync
,failed
,spare
;
5291 struct md_rdev
*rdev
;
5293 nr
=working
=insync
=failed
=spare
=0;
5294 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5296 if (test_bit(Faulty
, &rdev
->flags
))
5300 if (test_bit(In_sync
, &rdev
->flags
))
5307 info
.major_version
= mddev
->major_version
;
5308 info
.minor_version
= mddev
->minor_version
;
5309 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5310 info
.ctime
= mddev
->ctime
;
5311 info
.level
= mddev
->level
;
5312 info
.size
= mddev
->dev_sectors
/ 2;
5313 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5316 info
.raid_disks
= mddev
->raid_disks
;
5317 info
.md_minor
= mddev
->md_minor
;
5318 info
.not_persistent
= !mddev
->persistent
;
5320 info
.utime
= mddev
->utime
;
5323 info
.state
= (1<<MD_SB_CLEAN
);
5324 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5325 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5326 info
.active_disks
= insync
;
5327 info
.working_disks
= working
;
5328 info
.failed_disks
= failed
;
5329 info
.spare_disks
= spare
;
5331 info
.layout
= mddev
->layout
;
5332 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5334 if (copy_to_user(arg
, &info
, sizeof(info
)))
5340 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5342 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5343 char *ptr
, *buf
= NULL
;
5346 if (md_allow_write(mddev
))
5347 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5349 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5354 /* bitmap disabled, zero the first byte and copy out */
5355 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5356 file
->pathname
[0] = '\0';
5360 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5364 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5368 strcpy(file
->pathname
, ptr
);
5372 if (copy_to_user(arg
, file
, sizeof(*file
)))
5380 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5382 mdu_disk_info_t info
;
5383 struct md_rdev
*rdev
;
5385 if (copy_from_user(&info
, arg
, sizeof(info
)))
5388 rdev
= find_rdev_nr(mddev
, info
.number
);
5390 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5391 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5392 info
.raid_disk
= rdev
->raid_disk
;
5394 if (test_bit(Faulty
, &rdev
->flags
))
5395 info
.state
|= (1<<MD_DISK_FAULTY
);
5396 else if (test_bit(In_sync
, &rdev
->flags
)) {
5397 info
.state
|= (1<<MD_DISK_ACTIVE
);
5398 info
.state
|= (1<<MD_DISK_SYNC
);
5400 if (test_bit(WriteMostly
, &rdev
->flags
))
5401 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5403 info
.major
= info
.minor
= 0;
5404 info
.raid_disk
= -1;
5405 info
.state
= (1<<MD_DISK_REMOVED
);
5408 if (copy_to_user(arg
, &info
, sizeof(info
)))
5414 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5416 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5417 struct md_rdev
*rdev
;
5418 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5420 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5423 if (!mddev
->raid_disks
) {
5425 /* expecting a device which has a superblock */
5426 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5429 "md: md_import_device returned %ld\n",
5431 return PTR_ERR(rdev
);
5433 if (!list_empty(&mddev
->disks
)) {
5434 struct md_rdev
*rdev0
5435 = list_entry(mddev
->disks
.next
,
5436 struct md_rdev
, same_set
);
5437 err
= super_types
[mddev
->major_version
]
5438 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5441 "md: %s has different UUID to %s\n",
5442 bdevname(rdev
->bdev
,b
),
5443 bdevname(rdev0
->bdev
,b2
));
5448 err
= bind_rdev_to_array(rdev
, mddev
);
5455 * add_new_disk can be used once the array is assembled
5456 * to add "hot spares". They must already have a superblock
5461 if (!mddev
->pers
->hot_add_disk
) {
5463 "%s: personality does not support diskops!\n",
5467 if (mddev
->persistent
)
5468 rdev
= md_import_device(dev
, mddev
->major_version
,
5469 mddev
->minor_version
);
5471 rdev
= md_import_device(dev
, -1, -1);
5474 "md: md_import_device returned %ld\n",
5476 return PTR_ERR(rdev
);
5478 /* set saved_raid_disk if appropriate */
5479 if (!mddev
->persistent
) {
5480 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5481 info
->raid_disk
< mddev
->raid_disks
) {
5482 rdev
->raid_disk
= info
->raid_disk
;
5483 set_bit(In_sync
, &rdev
->flags
);
5485 rdev
->raid_disk
= -1;
5487 super_types
[mddev
->major_version
].
5488 validate_super(mddev
, rdev
);
5489 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5490 (!test_bit(In_sync
, &rdev
->flags
) ||
5491 rdev
->raid_disk
!= info
->raid_disk
)) {
5492 /* This was a hot-add request, but events doesn't
5493 * match, so reject it.
5499 if (test_bit(In_sync
, &rdev
->flags
))
5500 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5502 rdev
->saved_raid_disk
= -1;
5504 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5505 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5506 set_bit(WriteMostly
, &rdev
->flags
);
5508 clear_bit(WriteMostly
, &rdev
->flags
);
5510 rdev
->raid_disk
= -1;
5511 err
= bind_rdev_to_array(rdev
, mddev
);
5512 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5513 /* If there is hot_add_disk but no hot_remove_disk
5514 * then added disks for geometry changes,
5515 * and should be added immediately.
5517 super_types
[mddev
->major_version
].
5518 validate_super(mddev
, rdev
);
5519 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5521 unbind_rdev_from_array(rdev
);
5526 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5528 md_update_sb(mddev
, 1);
5529 if (mddev
->degraded
)
5530 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5531 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5533 md_new_event(mddev
);
5534 md_wakeup_thread(mddev
->thread
);
5538 /* otherwise, add_new_disk is only allowed
5539 * for major_version==0 superblocks
5541 if (mddev
->major_version
!= 0) {
5542 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5547 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5549 rdev
= md_import_device(dev
, -1, 0);
5552 "md: error, md_import_device() returned %ld\n",
5554 return PTR_ERR(rdev
);
5556 rdev
->desc_nr
= info
->number
;
5557 if (info
->raid_disk
< mddev
->raid_disks
)
5558 rdev
->raid_disk
= info
->raid_disk
;
5560 rdev
->raid_disk
= -1;
5562 if (rdev
->raid_disk
< mddev
->raid_disks
)
5563 if (info
->state
& (1<<MD_DISK_SYNC
))
5564 set_bit(In_sync
, &rdev
->flags
);
5566 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5567 set_bit(WriteMostly
, &rdev
->flags
);
5569 if (!mddev
->persistent
) {
5570 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5571 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5573 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5574 rdev
->sectors
= rdev
->sb_start
;
5576 err
= bind_rdev_to_array(rdev
, mddev
);
5586 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5588 char b
[BDEVNAME_SIZE
];
5589 struct md_rdev
*rdev
;
5591 rdev
= find_rdev(mddev
, dev
);
5595 if (rdev
->raid_disk
>= 0)
5598 kick_rdev_from_array(rdev
);
5599 md_update_sb(mddev
, 1);
5600 md_new_event(mddev
);
5604 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5605 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5609 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5611 char b
[BDEVNAME_SIZE
];
5613 struct md_rdev
*rdev
;
5618 if (mddev
->major_version
!= 0) {
5619 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5620 " version-0 superblocks.\n",
5624 if (!mddev
->pers
->hot_add_disk
) {
5626 "%s: personality does not support diskops!\n",
5631 rdev
= md_import_device(dev
, -1, 0);
5634 "md: error, md_import_device() returned %ld\n",
5639 if (mddev
->persistent
)
5640 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5642 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5644 rdev
->sectors
= rdev
->sb_start
;
5646 if (test_bit(Faulty
, &rdev
->flags
)) {
5648 "md: can not hot-add faulty %s disk to %s!\n",
5649 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5653 clear_bit(In_sync
, &rdev
->flags
);
5655 rdev
->saved_raid_disk
= -1;
5656 err
= bind_rdev_to_array(rdev
, mddev
);
5661 * The rest should better be atomic, we can have disk failures
5662 * noticed in interrupt contexts ...
5665 rdev
->raid_disk
= -1;
5667 md_update_sb(mddev
, 1);
5670 * Kick recovery, maybe this spare has to be added to the
5671 * array immediately.
5673 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5674 md_wakeup_thread(mddev
->thread
);
5675 md_new_event(mddev
);
5683 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5688 if (!mddev
->pers
->quiesce
)
5690 if (mddev
->recovery
|| mddev
->sync_thread
)
5692 /* we should be able to change the bitmap.. */
5698 return -EEXIST
; /* cannot add when bitmap is present */
5699 mddev
->bitmap_info
.file
= fget(fd
);
5701 if (mddev
->bitmap_info
.file
== NULL
) {
5702 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5707 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5709 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5711 fput(mddev
->bitmap_info
.file
);
5712 mddev
->bitmap_info
.file
= NULL
;
5715 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5716 } else if (mddev
->bitmap
== NULL
)
5717 return -ENOENT
; /* cannot remove what isn't there */
5720 mddev
->pers
->quiesce(mddev
, 1);
5722 err
= bitmap_create(mddev
);
5724 err
= bitmap_load(mddev
);
5726 if (fd
< 0 || err
) {
5727 bitmap_destroy(mddev
);
5728 fd
= -1; /* make sure to put the file */
5730 mddev
->pers
->quiesce(mddev
, 0);
5733 if (mddev
->bitmap_info
.file
) {
5734 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5735 fput(mddev
->bitmap_info
.file
);
5737 mddev
->bitmap_info
.file
= NULL
;
5744 * set_array_info is used two different ways
5745 * The original usage is when creating a new array.
5746 * In this usage, raid_disks is > 0 and it together with
5747 * level, size, not_persistent,layout,chunksize determine the
5748 * shape of the array.
5749 * This will always create an array with a type-0.90.0 superblock.
5750 * The newer usage is when assembling an array.
5751 * In this case raid_disks will be 0, and the major_version field is
5752 * use to determine which style super-blocks are to be found on the devices.
5753 * The minor and patch _version numbers are also kept incase the
5754 * super_block handler wishes to interpret them.
5756 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
5759 if (info
->raid_disks
== 0) {
5760 /* just setting version number for superblock loading */
5761 if (info
->major_version
< 0 ||
5762 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5763 super_types
[info
->major_version
].name
== NULL
) {
5764 /* maybe try to auto-load a module? */
5766 "md: superblock version %d not known\n",
5767 info
->major_version
);
5770 mddev
->major_version
= info
->major_version
;
5771 mddev
->minor_version
= info
->minor_version
;
5772 mddev
->patch_version
= info
->patch_version
;
5773 mddev
->persistent
= !info
->not_persistent
;
5774 /* ensure mddev_put doesn't delete this now that there
5775 * is some minimal configuration.
5777 mddev
->ctime
= get_seconds();
5780 mddev
->major_version
= MD_MAJOR_VERSION
;
5781 mddev
->minor_version
= MD_MINOR_VERSION
;
5782 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5783 mddev
->ctime
= get_seconds();
5785 mddev
->level
= info
->level
;
5786 mddev
->clevel
[0] = 0;
5787 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5788 mddev
->raid_disks
= info
->raid_disks
;
5789 /* don't set md_minor, it is determined by which /dev/md* was
5792 if (info
->state
& (1<<MD_SB_CLEAN
))
5793 mddev
->recovery_cp
= MaxSector
;
5795 mddev
->recovery_cp
= 0;
5796 mddev
->persistent
= ! info
->not_persistent
;
5797 mddev
->external
= 0;
5799 mddev
->layout
= info
->layout
;
5800 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5802 mddev
->max_disks
= MD_SB_DISKS
;
5804 if (mddev
->persistent
)
5806 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5808 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5809 mddev
->bitmap_info
.offset
= 0;
5811 mddev
->reshape_position
= MaxSector
;
5814 * Generate a 128 bit UUID
5816 get_random_bytes(mddev
->uuid
, 16);
5818 mddev
->new_level
= mddev
->level
;
5819 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5820 mddev
->new_layout
= mddev
->layout
;
5821 mddev
->delta_disks
= 0;
5826 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
5828 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5830 if (mddev
->external_size
)
5833 mddev
->array_sectors
= array_sectors
;
5835 EXPORT_SYMBOL(md_set_array_sectors
);
5837 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
5839 struct md_rdev
*rdev
;
5841 int fit
= (num_sectors
== 0);
5843 if (mddev
->pers
->resize
== NULL
)
5845 /* The "num_sectors" is the number of sectors of each device that
5846 * is used. This can only make sense for arrays with redundancy.
5847 * linear and raid0 always use whatever space is available. We can only
5848 * consider changing this number if no resync or reconstruction is
5849 * happening, and if the new size is acceptable. It must fit before the
5850 * sb_start or, if that is <data_offset, it must fit before the size
5851 * of each device. If num_sectors is zero, we find the largest size
5854 if (mddev
->sync_thread
)
5857 /* Sorry, cannot grow a bitmap yet, just remove it,
5861 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5862 sector_t avail
= rdev
->sectors
;
5864 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5865 num_sectors
= avail
;
5866 if (avail
< num_sectors
)
5869 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5871 revalidate_disk(mddev
->gendisk
);
5875 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
5878 /* change the number of raid disks */
5879 if (mddev
->pers
->check_reshape
== NULL
)
5881 if (raid_disks
<= 0 ||
5882 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5884 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5886 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5888 rv
= mddev
->pers
->check_reshape(mddev
);
5890 mddev
->delta_disks
= 0;
5896 * update_array_info is used to change the configuration of an
5898 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5899 * fields in the info are checked against the array.
5900 * Any differences that cannot be handled will cause an error.
5901 * Normally, only one change can be managed at a time.
5903 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
5909 /* calculate expected state,ignoring low bits */
5910 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5911 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5913 if (mddev
->major_version
!= info
->major_version
||
5914 mddev
->minor_version
!= info
->minor_version
||
5915 /* mddev->patch_version != info->patch_version || */
5916 mddev
->ctime
!= info
->ctime
||
5917 mddev
->level
!= info
->level
||
5918 /* mddev->layout != info->layout || */
5919 !mddev
->persistent
!= info
->not_persistent
||
5920 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5921 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5922 ((state
^info
->state
) & 0xfffffe00)
5925 /* Check there is only one change */
5926 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5928 if (mddev
->raid_disks
!= info
->raid_disks
)
5930 if (mddev
->layout
!= info
->layout
)
5932 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5939 if (mddev
->layout
!= info
->layout
) {
5941 * we don't need to do anything at the md level, the
5942 * personality will take care of it all.
5944 if (mddev
->pers
->check_reshape
== NULL
)
5947 mddev
->new_layout
= info
->layout
;
5948 rv
= mddev
->pers
->check_reshape(mddev
);
5950 mddev
->new_layout
= mddev
->layout
;
5954 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5955 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5957 if (mddev
->raid_disks
!= info
->raid_disks
)
5958 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5960 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5961 if (mddev
->pers
->quiesce
== NULL
)
5963 if (mddev
->recovery
|| mddev
->sync_thread
)
5965 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5966 /* add the bitmap */
5969 if (mddev
->bitmap_info
.default_offset
== 0)
5971 mddev
->bitmap_info
.offset
=
5972 mddev
->bitmap_info
.default_offset
;
5973 mddev
->pers
->quiesce(mddev
, 1);
5974 rv
= bitmap_create(mddev
);
5976 rv
= bitmap_load(mddev
);
5978 bitmap_destroy(mddev
);
5979 mddev
->pers
->quiesce(mddev
, 0);
5981 /* remove the bitmap */
5984 if (mddev
->bitmap
->file
)
5986 mddev
->pers
->quiesce(mddev
, 1);
5987 bitmap_destroy(mddev
);
5988 mddev
->pers
->quiesce(mddev
, 0);
5989 mddev
->bitmap_info
.offset
= 0;
5992 md_update_sb(mddev
, 1);
5996 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
5998 struct md_rdev
*rdev
;
6000 if (mddev
->pers
== NULL
)
6003 rdev
= find_rdev(mddev
, dev
);
6007 md_error(mddev
, rdev
);
6008 if (!test_bit(Faulty
, &rdev
->flags
))
6014 * We have a problem here : there is no easy way to give a CHS
6015 * virtual geometry. We currently pretend that we have a 2 heads
6016 * 4 sectors (with a BIG number of cylinders...). This drives
6017 * dosfs just mad... ;-)
6019 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6021 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6025 geo
->cylinders
= mddev
->array_sectors
/ 8;
6029 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6030 unsigned int cmd
, unsigned long arg
)
6033 void __user
*argp
= (void __user
*)arg
;
6034 struct mddev
*mddev
= NULL
;
6037 if (!capable(CAP_SYS_ADMIN
))
6041 * Commands dealing with the RAID driver but not any
6047 err
= get_version(argp
);
6050 case PRINT_RAID_DEBUG
:
6058 autostart_arrays(arg
);
6065 * Commands creating/starting a new array:
6068 mddev
= bdev
->bd_disk
->private_data
;
6075 err
= mddev_lock(mddev
);
6078 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6085 case SET_ARRAY_INFO
:
6087 mdu_array_info_t info
;
6089 memset(&info
, 0, sizeof(info
));
6090 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6095 err
= update_array_info(mddev
, &info
);
6097 printk(KERN_WARNING
"md: couldn't update"
6098 " array info. %d\n", err
);
6103 if (!list_empty(&mddev
->disks
)) {
6105 "md: array %s already has disks!\n",
6110 if (mddev
->raid_disks
) {
6112 "md: array %s already initialised!\n",
6117 err
= set_array_info(mddev
, &info
);
6119 printk(KERN_WARNING
"md: couldn't set"
6120 " array info. %d\n", err
);
6130 * Commands querying/configuring an existing array:
6132 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6133 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6134 if ((!mddev
->raid_disks
&& !mddev
->external
)
6135 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6136 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6137 && cmd
!= GET_BITMAP_FILE
) {
6143 * Commands even a read-only array can execute:
6147 case GET_ARRAY_INFO
:
6148 err
= get_array_info(mddev
, argp
);
6151 case GET_BITMAP_FILE
:
6152 err
= get_bitmap_file(mddev
, argp
);
6156 err
= get_disk_info(mddev
, argp
);
6159 case RESTART_ARRAY_RW
:
6160 err
= restart_array(mddev
);
6164 err
= do_md_stop(mddev
, 0, 1);
6168 err
= md_set_readonly(mddev
, 1);
6172 if (get_user(ro
, (int __user
*)(arg
))) {
6178 /* if the bdev is going readonly the value of mddev->ro
6179 * does not matter, no writes are coming
6184 /* are we are already prepared for writes? */
6188 /* transitioning to readauto need only happen for
6189 * arrays that call md_write_start
6192 err
= restart_array(mddev
);
6195 set_disk_ro(mddev
->gendisk
, 0);
6202 * The remaining ioctls are changing the state of the
6203 * superblock, so we do not allow them on read-only arrays.
6204 * However non-MD ioctls (e.g. get-size) will still come through
6205 * here and hit the 'default' below, so only disallow
6206 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6208 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6209 if (mddev
->ro
== 2) {
6211 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6212 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6213 md_wakeup_thread(mddev
->thread
);
6224 mdu_disk_info_t info
;
6225 if (copy_from_user(&info
, argp
, sizeof(info
)))
6228 err
= add_new_disk(mddev
, &info
);
6232 case HOT_REMOVE_DISK
:
6233 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6237 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6240 case SET_DISK_FAULTY
:
6241 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6245 err
= do_md_run(mddev
);
6248 case SET_BITMAP_FILE
:
6249 err
= set_bitmap_file(mddev
, (int)arg
);
6259 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6261 mddev
->hold_active
= 0;
6262 mddev_unlock(mddev
);
6271 #ifdef CONFIG_COMPAT
6272 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6273 unsigned int cmd
, unsigned long arg
)
6276 case HOT_REMOVE_DISK
:
6278 case SET_DISK_FAULTY
:
6279 case SET_BITMAP_FILE
:
6280 /* These take in integer arg, do not convert */
6283 arg
= (unsigned long)compat_ptr(arg
);
6287 return md_ioctl(bdev
, mode
, cmd
, arg
);
6289 #endif /* CONFIG_COMPAT */
6291 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6294 * Succeed if we can lock the mddev, which confirms that
6295 * it isn't being stopped right now.
6297 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6300 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6301 /* we are racing with mddev_put which is discarding this
6305 /* Wait until bdev->bd_disk is definitely gone */
6306 flush_workqueue(md_misc_wq
);
6307 /* Then retry the open from the top */
6308 return -ERESTARTSYS
;
6310 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6312 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6316 atomic_inc(&mddev
->openers
);
6317 mutex_unlock(&mddev
->open_mutex
);
6319 check_disk_change(bdev
);
6324 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6326 struct mddev
*mddev
= disk
->private_data
;
6329 atomic_dec(&mddev
->openers
);
6335 static int md_media_changed(struct gendisk
*disk
)
6337 struct mddev
*mddev
= disk
->private_data
;
6339 return mddev
->changed
;
6342 static int md_revalidate(struct gendisk
*disk
)
6344 struct mddev
*mddev
= disk
->private_data
;
6349 static const struct block_device_operations md_fops
=
6351 .owner
= THIS_MODULE
,
6353 .release
= md_release
,
6355 #ifdef CONFIG_COMPAT
6356 .compat_ioctl
= md_compat_ioctl
,
6358 .getgeo
= md_getgeo
,
6359 .media_changed
= md_media_changed
,
6360 .revalidate_disk
= md_revalidate
,
6363 static int md_thread(void * arg
)
6365 mdk_thread_t
*thread
= arg
;
6368 * md_thread is a 'system-thread', it's priority should be very
6369 * high. We avoid resource deadlocks individually in each
6370 * raid personality. (RAID5 does preallocation) We also use RR and
6371 * the very same RT priority as kswapd, thus we will never get
6372 * into a priority inversion deadlock.
6374 * we definitely have to have equal or higher priority than
6375 * bdflush, otherwise bdflush will deadlock if there are too
6376 * many dirty RAID5 blocks.
6379 allow_signal(SIGKILL
);
6380 while (!kthread_should_stop()) {
6382 /* We need to wait INTERRUPTIBLE so that
6383 * we don't add to the load-average.
6384 * That means we need to be sure no signals are
6387 if (signal_pending(current
))
6388 flush_signals(current
);
6390 wait_event_interruptible_timeout
6392 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6393 || kthread_should_stop(),
6396 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6397 if (!kthread_should_stop())
6398 thread
->run(thread
->mddev
);
6404 void md_wakeup_thread(mdk_thread_t
*thread
)
6407 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6408 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6409 wake_up(&thread
->wqueue
);
6413 mdk_thread_t
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6416 mdk_thread_t
*thread
;
6418 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
6422 init_waitqueue_head(&thread
->wqueue
);
6425 thread
->mddev
= mddev
;
6426 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6427 thread
->tsk
= kthread_run(md_thread
, thread
,
6429 mdname(thread
->mddev
),
6430 name
?: mddev
->pers
->name
);
6431 if (IS_ERR(thread
->tsk
)) {
6438 void md_unregister_thread(mdk_thread_t
**threadp
)
6440 mdk_thread_t
*thread
= *threadp
;
6443 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6444 /* Locking ensures that mddev_unlock does not wake_up a
6445 * non-existent thread
6447 spin_lock(&pers_lock
);
6449 spin_unlock(&pers_lock
);
6451 kthread_stop(thread
->tsk
);
6455 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6462 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6465 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6467 mddev
->pers
->error_handler(mddev
,rdev
);
6468 if (mddev
->degraded
)
6469 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6470 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6471 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6472 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6473 md_wakeup_thread(mddev
->thread
);
6474 if (mddev
->event_work
.func
)
6475 queue_work(md_misc_wq
, &mddev
->event_work
);
6476 md_new_event_inintr(mddev
);
6479 /* seq_file implementation /proc/mdstat */
6481 static void status_unused(struct seq_file
*seq
)
6484 struct md_rdev
*rdev
;
6486 seq_printf(seq
, "unused devices: ");
6488 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6489 char b
[BDEVNAME_SIZE
];
6491 seq_printf(seq
, "%s ",
6492 bdevname(rdev
->bdev
,b
));
6495 seq_printf(seq
, "<none>");
6497 seq_printf(seq
, "\n");
6501 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6503 sector_t max_sectors
, resync
, res
;
6504 unsigned long dt
, db
;
6507 unsigned int per_milli
;
6509 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6511 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6512 max_sectors
= mddev
->resync_max_sectors
;
6514 max_sectors
= mddev
->dev_sectors
;
6517 * Should not happen.
6523 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6524 * in a sector_t, and (max_sectors>>scale) will fit in a
6525 * u32, as those are the requirements for sector_div.
6526 * Thus 'scale' must be at least 10
6529 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6530 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6533 res
= (resync
>>scale
)*1000;
6534 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6538 int i
, x
= per_milli
/50, y
= 20-x
;
6539 seq_printf(seq
, "[");
6540 for (i
= 0; i
< x
; i
++)
6541 seq_printf(seq
, "=");
6542 seq_printf(seq
, ">");
6543 for (i
= 0; i
< y
; i
++)
6544 seq_printf(seq
, ".");
6545 seq_printf(seq
, "] ");
6547 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6548 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6550 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6552 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6553 "resync" : "recovery"))),
6554 per_milli
/10, per_milli
% 10,
6555 (unsigned long long) resync
/2,
6556 (unsigned long long) max_sectors
/2);
6559 * dt: time from mark until now
6560 * db: blocks written from mark until now
6561 * rt: remaining time
6563 * rt is a sector_t, so could be 32bit or 64bit.
6564 * So we divide before multiply in case it is 32bit and close
6566 * We scale the divisor (db) by 32 to avoid losing precision
6567 * near the end of resync when the number of remaining sectors
6569 * We then divide rt by 32 after multiplying by db to compensate.
6570 * The '+1' avoids division by zero if db is very small.
6572 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6574 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6575 - mddev
->resync_mark_cnt
;
6577 rt
= max_sectors
- resync
; /* number of remaining sectors */
6578 sector_div(rt
, db
/32+1);
6582 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6583 ((unsigned long)rt
% 60)/6);
6585 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6588 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6590 struct list_head
*tmp
;
6592 struct mddev
*mddev
;
6600 spin_lock(&all_mddevs_lock
);
6601 list_for_each(tmp
,&all_mddevs
)
6603 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6605 spin_unlock(&all_mddevs_lock
);
6608 spin_unlock(&all_mddevs_lock
);
6610 return (void*)2;/* tail */
6614 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6616 struct list_head
*tmp
;
6617 struct mddev
*next_mddev
, *mddev
= v
;
6623 spin_lock(&all_mddevs_lock
);
6625 tmp
= all_mddevs
.next
;
6627 tmp
= mddev
->all_mddevs
.next
;
6628 if (tmp
!= &all_mddevs
)
6629 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6631 next_mddev
= (void*)2;
6634 spin_unlock(&all_mddevs_lock
);
6642 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6644 struct mddev
*mddev
= v
;
6646 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6650 static int md_seq_show(struct seq_file
*seq
, void *v
)
6652 struct mddev
*mddev
= v
;
6654 struct md_rdev
*rdev
;
6655 struct bitmap
*bitmap
;
6657 if (v
== (void*)1) {
6658 struct mdk_personality
*pers
;
6659 seq_printf(seq
, "Personalities : ");
6660 spin_lock(&pers_lock
);
6661 list_for_each_entry(pers
, &pers_list
, list
)
6662 seq_printf(seq
, "[%s] ", pers
->name
);
6664 spin_unlock(&pers_lock
);
6665 seq_printf(seq
, "\n");
6666 seq
->poll_event
= atomic_read(&md_event_count
);
6669 if (v
== (void*)2) {
6674 if (mddev_lock(mddev
) < 0)
6677 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6678 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6679 mddev
->pers
? "" : "in");
6682 seq_printf(seq
, " (read-only)");
6684 seq_printf(seq
, " (auto-read-only)");
6685 seq_printf(seq
, " %s", mddev
->pers
->name
);
6689 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6690 char b
[BDEVNAME_SIZE
];
6691 seq_printf(seq
, " %s[%d]",
6692 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6693 if (test_bit(WriteMostly
, &rdev
->flags
))
6694 seq_printf(seq
, "(W)");
6695 if (test_bit(Faulty
, &rdev
->flags
)) {
6696 seq_printf(seq
, "(F)");
6698 } else if (rdev
->raid_disk
< 0)
6699 seq_printf(seq
, "(S)"); /* spare */
6700 sectors
+= rdev
->sectors
;
6703 if (!list_empty(&mddev
->disks
)) {
6705 seq_printf(seq
, "\n %llu blocks",
6706 (unsigned long long)
6707 mddev
->array_sectors
/ 2);
6709 seq_printf(seq
, "\n %llu blocks",
6710 (unsigned long long)sectors
/ 2);
6712 if (mddev
->persistent
) {
6713 if (mddev
->major_version
!= 0 ||
6714 mddev
->minor_version
!= 90) {
6715 seq_printf(seq
," super %d.%d",
6716 mddev
->major_version
,
6717 mddev
->minor_version
);
6719 } else if (mddev
->external
)
6720 seq_printf(seq
, " super external:%s",
6721 mddev
->metadata_type
);
6723 seq_printf(seq
, " super non-persistent");
6726 mddev
->pers
->status(seq
, mddev
);
6727 seq_printf(seq
, "\n ");
6728 if (mddev
->pers
->sync_request
) {
6729 if (mddev
->curr_resync
> 2) {
6730 status_resync(seq
, mddev
);
6731 seq_printf(seq
, "\n ");
6732 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6733 seq_printf(seq
, "\tresync=DELAYED\n ");
6734 else if (mddev
->recovery_cp
< MaxSector
)
6735 seq_printf(seq
, "\tresync=PENDING\n ");
6738 seq_printf(seq
, "\n ");
6740 if ((bitmap
= mddev
->bitmap
)) {
6741 unsigned long chunk_kb
;
6742 unsigned long flags
;
6743 spin_lock_irqsave(&bitmap
->lock
, flags
);
6744 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6745 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6747 bitmap
->pages
- bitmap
->missing_pages
,
6749 (bitmap
->pages
- bitmap
->missing_pages
)
6750 << (PAGE_SHIFT
- 10),
6751 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6752 chunk_kb
? "KB" : "B");
6754 seq_printf(seq
, ", file: ");
6755 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6758 seq_printf(seq
, "\n");
6759 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6762 seq_printf(seq
, "\n");
6764 mddev_unlock(mddev
);
6769 static const struct seq_operations md_seq_ops
= {
6770 .start
= md_seq_start
,
6771 .next
= md_seq_next
,
6772 .stop
= md_seq_stop
,
6773 .show
= md_seq_show
,
6776 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6778 struct seq_file
*seq
;
6781 error
= seq_open(file
, &md_seq_ops
);
6785 seq
= file
->private_data
;
6786 seq
->poll_event
= atomic_read(&md_event_count
);
6790 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6792 struct seq_file
*seq
= filp
->private_data
;
6795 poll_wait(filp
, &md_event_waiters
, wait
);
6797 /* always allow read */
6798 mask
= POLLIN
| POLLRDNORM
;
6800 if (seq
->poll_event
!= atomic_read(&md_event_count
))
6801 mask
|= POLLERR
| POLLPRI
;
6805 static const struct file_operations md_seq_fops
= {
6806 .owner
= THIS_MODULE
,
6807 .open
= md_seq_open
,
6809 .llseek
= seq_lseek
,
6810 .release
= seq_release_private
,
6811 .poll
= mdstat_poll
,
6814 int register_md_personality(struct mdk_personality
*p
)
6816 spin_lock(&pers_lock
);
6817 list_add_tail(&p
->list
, &pers_list
);
6818 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6819 spin_unlock(&pers_lock
);
6823 int unregister_md_personality(struct mdk_personality
*p
)
6825 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6826 spin_lock(&pers_lock
);
6827 list_del_init(&p
->list
);
6828 spin_unlock(&pers_lock
);
6832 static int is_mddev_idle(struct mddev
*mddev
, int init
)
6834 struct md_rdev
* rdev
;
6840 rdev_for_each_rcu(rdev
, mddev
) {
6841 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6842 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6843 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6844 atomic_read(&disk
->sync_io
);
6845 /* sync IO will cause sync_io to increase before the disk_stats
6846 * as sync_io is counted when a request starts, and
6847 * disk_stats is counted when it completes.
6848 * So resync activity will cause curr_events to be smaller than
6849 * when there was no such activity.
6850 * non-sync IO will cause disk_stat to increase without
6851 * increasing sync_io so curr_events will (eventually)
6852 * be larger than it was before. Once it becomes
6853 * substantially larger, the test below will cause
6854 * the array to appear non-idle, and resync will slow
6856 * If there is a lot of outstanding resync activity when
6857 * we set last_event to curr_events, then all that activity
6858 * completing might cause the array to appear non-idle
6859 * and resync will be slowed down even though there might
6860 * not have been non-resync activity. This will only
6861 * happen once though. 'last_events' will soon reflect
6862 * the state where there is little or no outstanding
6863 * resync requests, and further resync activity will
6864 * always make curr_events less than last_events.
6867 if (init
|| curr_events
- rdev
->last_events
> 64) {
6868 rdev
->last_events
= curr_events
;
6876 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
6878 /* another "blocks" (512byte) blocks have been synced */
6879 atomic_sub(blocks
, &mddev
->recovery_active
);
6880 wake_up(&mddev
->recovery_wait
);
6882 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6883 md_wakeup_thread(mddev
->thread
);
6884 // stop recovery, signal do_sync ....
6889 /* md_write_start(mddev, bi)
6890 * If we need to update some array metadata (e.g. 'active' flag
6891 * in superblock) before writing, schedule a superblock update
6892 * and wait for it to complete.
6894 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
6897 if (bio_data_dir(bi
) != WRITE
)
6900 BUG_ON(mddev
->ro
== 1);
6901 if (mddev
->ro
== 2) {
6902 /* need to switch to read/write */
6904 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6905 md_wakeup_thread(mddev
->thread
);
6906 md_wakeup_thread(mddev
->sync_thread
);
6909 atomic_inc(&mddev
->writes_pending
);
6910 if (mddev
->safemode
== 1)
6911 mddev
->safemode
= 0;
6912 if (mddev
->in_sync
) {
6913 spin_lock_irq(&mddev
->write_lock
);
6914 if (mddev
->in_sync
) {
6916 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6917 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6918 md_wakeup_thread(mddev
->thread
);
6921 spin_unlock_irq(&mddev
->write_lock
);
6924 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6925 wait_event(mddev
->sb_wait
,
6926 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6929 void md_write_end(struct mddev
*mddev
)
6931 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6932 if (mddev
->safemode
== 2)
6933 md_wakeup_thread(mddev
->thread
);
6934 else if (mddev
->safemode_delay
)
6935 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6939 /* md_allow_write(mddev)
6940 * Calling this ensures that the array is marked 'active' so that writes
6941 * may proceed without blocking. It is important to call this before
6942 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6943 * Must be called with mddev_lock held.
6945 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6946 * is dropped, so return -EAGAIN after notifying userspace.
6948 int md_allow_write(struct mddev
*mddev
)
6954 if (!mddev
->pers
->sync_request
)
6957 spin_lock_irq(&mddev
->write_lock
);
6958 if (mddev
->in_sync
) {
6960 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6961 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6962 if (mddev
->safemode_delay
&&
6963 mddev
->safemode
== 0)
6964 mddev
->safemode
= 1;
6965 spin_unlock_irq(&mddev
->write_lock
);
6966 md_update_sb(mddev
, 0);
6967 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6969 spin_unlock_irq(&mddev
->write_lock
);
6971 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6976 EXPORT_SYMBOL_GPL(md_allow_write
);
6978 #define SYNC_MARKS 10
6979 #define SYNC_MARK_STEP (3*HZ)
6980 void md_do_sync(struct mddev
*mddev
)
6982 struct mddev
*mddev2
;
6983 unsigned int currspeed
= 0,
6985 sector_t max_sectors
,j
, io_sectors
;
6986 unsigned long mark
[SYNC_MARKS
];
6987 sector_t mark_cnt
[SYNC_MARKS
];
6989 struct list_head
*tmp
;
6990 sector_t last_check
;
6992 struct md_rdev
*rdev
;
6995 /* just incase thread restarts... */
6996 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6998 if (mddev
->ro
) /* never try to sync a read-only array */
7001 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7002 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7003 desc
= "data-check";
7004 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7005 desc
= "requested-resync";
7008 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7013 /* we overload curr_resync somewhat here.
7014 * 0 == not engaged in resync at all
7015 * 2 == checking that there is no conflict with another sync
7016 * 1 == like 2, but have yielded to allow conflicting resync to
7018 * other == active in resync - this many blocks
7020 * Before starting a resync we must have set curr_resync to
7021 * 2, and then checked that every "conflicting" array has curr_resync
7022 * less than ours. When we find one that is the same or higher
7023 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7024 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7025 * This will mean we have to start checking from the beginning again.
7030 mddev
->curr_resync
= 2;
7033 if (kthread_should_stop())
7034 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7036 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7038 for_each_mddev(mddev2
, tmp
) {
7039 if (mddev2
== mddev
)
7041 if (!mddev
->parallel_resync
7042 && mddev2
->curr_resync
7043 && match_mddev_units(mddev
, mddev2
)) {
7045 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7046 /* arbitrarily yield */
7047 mddev
->curr_resync
= 1;
7048 wake_up(&resync_wait
);
7050 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7051 /* no need to wait here, we can wait the next
7052 * time 'round when curr_resync == 2
7055 /* We need to wait 'interruptible' so as not to
7056 * contribute to the load average, and not to
7057 * be caught by 'softlockup'
7059 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7060 if (!kthread_should_stop() &&
7061 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7062 printk(KERN_INFO
"md: delaying %s of %s"
7063 " until %s has finished (they"
7064 " share one or more physical units)\n",
7065 desc
, mdname(mddev
), mdname(mddev2
));
7067 if (signal_pending(current
))
7068 flush_signals(current
);
7070 finish_wait(&resync_wait
, &wq
);
7073 finish_wait(&resync_wait
, &wq
);
7076 } while (mddev
->curr_resync
< 2);
7079 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7080 /* resync follows the size requested by the personality,
7081 * which defaults to physical size, but can be virtual size
7083 max_sectors
= mddev
->resync_max_sectors
;
7084 mddev
->resync_mismatches
= 0;
7085 /* we don't use the checkpoint if there's a bitmap */
7086 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7087 j
= mddev
->resync_min
;
7088 else if (!mddev
->bitmap
)
7089 j
= mddev
->recovery_cp
;
7091 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7092 max_sectors
= mddev
->dev_sectors
;
7094 /* recovery follows the physical size of devices */
7095 max_sectors
= mddev
->dev_sectors
;
7098 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7099 if (rdev
->raid_disk
>= 0 &&
7100 !test_bit(Faulty
, &rdev
->flags
) &&
7101 !test_bit(In_sync
, &rdev
->flags
) &&
7102 rdev
->recovery_offset
< j
)
7103 j
= rdev
->recovery_offset
;
7107 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7108 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7109 " %d KB/sec/disk.\n", speed_min(mddev
));
7110 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7111 "(but not more than %d KB/sec) for %s.\n",
7112 speed_max(mddev
), desc
);
7114 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7117 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7119 mark_cnt
[m
] = io_sectors
;
7122 mddev
->resync_mark
= mark
[last_mark
];
7123 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7126 * Tune reconstruction:
7128 window
= 32*(PAGE_SIZE
/512);
7129 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7130 window
/2, (unsigned long long)max_sectors
/2);
7132 atomic_set(&mddev
->recovery_active
, 0);
7137 "md: resuming %s of %s from checkpoint.\n",
7138 desc
, mdname(mddev
));
7139 mddev
->curr_resync
= j
;
7141 mddev
->curr_resync_completed
= j
;
7143 while (j
< max_sectors
) {
7148 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7149 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7150 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7151 > (max_sectors
>> 4)) ||
7152 (j
- mddev
->curr_resync_completed
)*2
7153 >= mddev
->resync_max
- mddev
->curr_resync_completed
7155 /* time to update curr_resync_completed */
7156 wait_event(mddev
->recovery_wait
,
7157 atomic_read(&mddev
->recovery_active
) == 0);
7158 mddev
->curr_resync_completed
= j
;
7159 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7160 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7163 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7164 /* As this condition is controlled by user-space,
7165 * we can block indefinitely, so use '_interruptible'
7166 * to avoid triggering warnings.
7168 flush_signals(current
); /* just in case */
7169 wait_event_interruptible(mddev
->recovery_wait
,
7170 mddev
->resync_max
> j
7171 || kthread_should_stop());
7174 if (kthread_should_stop())
7177 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7178 currspeed
< speed_min(mddev
));
7180 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7184 if (!skipped
) { /* actual IO requested */
7185 io_sectors
+= sectors
;
7186 atomic_add(sectors
, &mddev
->recovery_active
);
7189 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7193 if (j
>1) mddev
->curr_resync
= j
;
7194 mddev
->curr_mark_cnt
= io_sectors
;
7195 if (last_check
== 0)
7196 /* this is the earliest that rebuild will be
7197 * visible in /proc/mdstat
7199 md_new_event(mddev
);
7201 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7204 last_check
= io_sectors
;
7206 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7208 int next
= (last_mark
+1) % SYNC_MARKS
;
7210 mddev
->resync_mark
= mark
[next
];
7211 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7212 mark
[next
] = jiffies
;
7213 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7218 if (kthread_should_stop())
7223 * this loop exits only if either when we are slower than
7224 * the 'hard' speed limit, or the system was IO-idle for
7226 * the system might be non-idle CPU-wise, but we only care
7227 * about not overloading the IO subsystem. (things like an
7228 * e2fsck being done on the RAID array should execute fast)
7232 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7233 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7235 if (currspeed
> speed_min(mddev
)) {
7236 if ((currspeed
> speed_max(mddev
)) ||
7237 !is_mddev_idle(mddev
, 0)) {
7243 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7245 * this also signals 'finished resyncing' to md_stop
7248 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7250 /* tell personality that we are finished */
7251 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7253 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7254 mddev
->curr_resync
> 2) {
7255 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7256 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7257 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7259 "md: checkpointing %s of %s.\n",
7260 desc
, mdname(mddev
));
7261 mddev
->recovery_cp
= mddev
->curr_resync
;
7264 mddev
->recovery_cp
= MaxSector
;
7266 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7267 mddev
->curr_resync
= MaxSector
;
7269 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7270 if (rdev
->raid_disk
>= 0 &&
7271 mddev
->delta_disks
>= 0 &&
7272 !test_bit(Faulty
, &rdev
->flags
) &&
7273 !test_bit(In_sync
, &rdev
->flags
) &&
7274 rdev
->recovery_offset
< mddev
->curr_resync
)
7275 rdev
->recovery_offset
= mddev
->curr_resync
;
7279 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7282 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7283 /* We completed so min/max setting can be forgotten if used. */
7284 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7285 mddev
->resync_min
= 0;
7286 mddev
->resync_max
= MaxSector
;
7287 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7288 mddev
->resync_min
= mddev
->curr_resync_completed
;
7289 mddev
->curr_resync
= 0;
7290 wake_up(&resync_wait
);
7291 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7292 md_wakeup_thread(mddev
->thread
);
7297 * got a signal, exit.
7300 "md: md_do_sync() got signal ... exiting\n");
7301 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7305 EXPORT_SYMBOL_GPL(md_do_sync
);
7307 static int remove_and_add_spares(struct mddev
*mddev
)
7309 struct md_rdev
*rdev
;
7312 mddev
->curr_resync_completed
= 0;
7314 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7315 if (rdev
->raid_disk
>= 0 &&
7316 !test_bit(Blocked
, &rdev
->flags
) &&
7317 (test_bit(Faulty
, &rdev
->flags
) ||
7318 ! test_bit(In_sync
, &rdev
->flags
)) &&
7319 atomic_read(&rdev
->nr_pending
)==0) {
7320 if (mddev
->pers
->hot_remove_disk(
7321 mddev
, rdev
->raid_disk
)==0) {
7322 sysfs_unlink_rdev(mddev
, rdev
);
7323 rdev
->raid_disk
= -1;
7327 if (mddev
->degraded
) {
7328 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7329 if (rdev
->raid_disk
>= 0 &&
7330 !test_bit(In_sync
, &rdev
->flags
) &&
7331 !test_bit(Faulty
, &rdev
->flags
))
7333 if (rdev
->raid_disk
< 0
7334 && !test_bit(Faulty
, &rdev
->flags
)) {
7335 rdev
->recovery_offset
= 0;
7337 hot_add_disk(mddev
, rdev
) == 0) {
7338 if (sysfs_link_rdev(mddev
, rdev
))
7339 /* failure here is OK */;
7341 md_new_event(mddev
);
7342 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7351 static void reap_sync_thread(struct mddev
*mddev
)
7353 struct md_rdev
*rdev
;
7355 /* resync has finished, collect result */
7356 md_unregister_thread(&mddev
->sync_thread
);
7357 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7358 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7360 /* activate any spares */
7361 if (mddev
->pers
->spare_active(mddev
))
7362 sysfs_notify(&mddev
->kobj
, NULL
,
7365 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7366 mddev
->pers
->finish_reshape
)
7367 mddev
->pers
->finish_reshape(mddev
);
7368 md_update_sb(mddev
, 1);
7370 /* if array is no-longer degraded, then any saved_raid_disk
7371 * information must be scrapped
7373 if (!mddev
->degraded
)
7374 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7375 rdev
->saved_raid_disk
= -1;
7377 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7378 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7379 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7380 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7381 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7382 /* flag recovery needed just to double check */
7383 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7384 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7385 md_new_event(mddev
);
7386 if (mddev
->event_work
.func
)
7387 queue_work(md_misc_wq
, &mddev
->event_work
);
7391 * This routine is regularly called by all per-raid-array threads to
7392 * deal with generic issues like resync and super-block update.
7393 * Raid personalities that don't have a thread (linear/raid0) do not
7394 * need this as they never do any recovery or update the superblock.
7396 * It does not do any resync itself, but rather "forks" off other threads
7397 * to do that as needed.
7398 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7399 * "->recovery" and create a thread at ->sync_thread.
7400 * When the thread finishes it sets MD_RECOVERY_DONE
7401 * and wakeups up this thread which will reap the thread and finish up.
7402 * This thread also removes any faulty devices (with nr_pending == 0).
7404 * The overall approach is:
7405 * 1/ if the superblock needs updating, update it.
7406 * 2/ If a recovery thread is running, don't do anything else.
7407 * 3/ If recovery has finished, clean up, possibly marking spares active.
7408 * 4/ If there are any faulty devices, remove them.
7409 * 5/ If array is degraded, try to add spares devices
7410 * 6/ If array has spares or is not in-sync, start a resync thread.
7412 void md_check_recovery(struct mddev
*mddev
)
7414 if (mddev
->suspended
)
7418 bitmap_daemon_work(mddev
);
7420 if (signal_pending(current
)) {
7421 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7422 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7424 mddev
->safemode
= 2;
7426 flush_signals(current
);
7429 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7432 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7433 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7434 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7435 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7436 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7437 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7441 if (mddev_trylock(mddev
)) {
7445 /* Only thing we do on a ro array is remove
7448 struct md_rdev
*rdev
;
7449 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7450 if (rdev
->raid_disk
>= 0 &&
7451 !test_bit(Blocked
, &rdev
->flags
) &&
7452 test_bit(Faulty
, &rdev
->flags
) &&
7453 atomic_read(&rdev
->nr_pending
)==0) {
7454 if (mddev
->pers
->hot_remove_disk(
7455 mddev
, rdev
->raid_disk
)==0) {
7456 sysfs_unlink_rdev(mddev
, rdev
);
7457 rdev
->raid_disk
= -1;
7460 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7464 if (!mddev
->external
) {
7466 spin_lock_irq(&mddev
->write_lock
);
7467 if (mddev
->safemode
&&
7468 !atomic_read(&mddev
->writes_pending
) &&
7470 mddev
->recovery_cp
== MaxSector
) {
7473 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7475 if (mddev
->safemode
== 1)
7476 mddev
->safemode
= 0;
7477 spin_unlock_irq(&mddev
->write_lock
);
7479 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7483 md_update_sb(mddev
, 0);
7485 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7486 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7487 /* resync/recovery still happening */
7488 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7491 if (mddev
->sync_thread
) {
7492 reap_sync_thread(mddev
);
7495 /* Set RUNNING before clearing NEEDED to avoid
7496 * any transients in the value of "sync_action".
7498 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7499 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7500 /* Clear some bits that don't mean anything, but
7503 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7504 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7506 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7508 /* no recovery is running.
7509 * remove any failed drives, then
7510 * add spares if possible.
7511 * Spare are also removed and re-added, to allow
7512 * the personality to fail the re-add.
7515 if (mddev
->reshape_position
!= MaxSector
) {
7516 if (mddev
->pers
->check_reshape
== NULL
||
7517 mddev
->pers
->check_reshape(mddev
) != 0)
7518 /* Cannot proceed */
7520 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7521 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7522 } else if ((spares
= remove_and_add_spares(mddev
))) {
7523 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7524 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7525 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7526 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7527 } else if (mddev
->recovery_cp
< MaxSector
) {
7528 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7529 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7530 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7531 /* nothing to be done ... */
7534 if (mddev
->pers
->sync_request
) {
7535 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7536 /* We are adding a device or devices to an array
7537 * which has the bitmap stored on all devices.
7538 * So make sure all bitmap pages get written
7540 bitmap_write_all(mddev
->bitmap
);
7542 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7545 if (!mddev
->sync_thread
) {
7546 printk(KERN_ERR
"%s: could not start resync"
7549 /* leave the spares where they are, it shouldn't hurt */
7550 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7551 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7552 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7553 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7554 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7556 md_wakeup_thread(mddev
->sync_thread
);
7557 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7558 md_new_event(mddev
);
7561 if (!mddev
->sync_thread
) {
7562 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7563 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7565 if (mddev
->sysfs_action
)
7566 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7568 mddev_unlock(mddev
);
7572 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7574 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7575 wait_event_timeout(rdev
->blocked_wait
,
7576 !test_bit(Blocked
, &rdev
->flags
) &&
7577 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7578 msecs_to_jiffies(5000));
7579 rdev_dec_pending(rdev
, mddev
);
7581 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7584 /* Bad block management.
7585 * We can record which blocks on each device are 'bad' and so just
7586 * fail those blocks, or that stripe, rather than the whole device.
7587 * Entries in the bad-block table are 64bits wide. This comprises:
7588 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7589 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7590 * A 'shift' can be set so that larger blocks are tracked and
7591 * consequently larger devices can be covered.
7592 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7594 * Locking of the bad-block table uses a seqlock so md_is_badblock
7595 * might need to retry if it is very unlucky.
7596 * We will sometimes want to check for bad blocks in a bi_end_io function,
7597 * so we use the write_seqlock_irq variant.
7599 * When looking for a bad block we specify a range and want to
7600 * know if any block in the range is bad. So we binary-search
7601 * to the last range that starts at-or-before the given endpoint,
7602 * (or "before the sector after the target range")
7603 * then see if it ends after the given start.
7605 * 0 if there are no known bad blocks in the range
7606 * 1 if there are known bad block which are all acknowledged
7607 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7608 * plus the start/length of the first bad section we overlap.
7610 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7611 sector_t
*first_bad
, int *bad_sectors
)
7617 sector_t target
= s
+ sectors
;
7620 if (bb
->shift
> 0) {
7621 /* round the start down, and the end up */
7623 target
+= (1<<bb
->shift
) - 1;
7624 target
>>= bb
->shift
;
7625 sectors
= target
- s
;
7627 /* 'target' is now the first block after the bad range */
7630 seq
= read_seqbegin(&bb
->lock
);
7634 /* Binary search between lo and hi for 'target'
7635 * i.e. for the last range that starts before 'target'
7637 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7638 * are known not to be the last range before target.
7639 * VARIANT: hi-lo is the number of possible
7640 * ranges, and decreases until it reaches 1
7642 while (hi
- lo
> 1) {
7643 int mid
= (lo
+ hi
) / 2;
7644 sector_t a
= BB_OFFSET(p
[mid
]);
7646 /* This could still be the one, earlier ranges
7650 /* This and later ranges are definitely out. */
7653 /* 'lo' might be the last that started before target, but 'hi' isn't */
7655 /* need to check all range that end after 's' to see if
7656 * any are unacknowledged.
7659 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7660 if (BB_OFFSET(p
[lo
]) < target
) {
7661 /* starts before the end, and finishes after
7662 * the start, so they must overlap
7664 if (rv
!= -1 && BB_ACK(p
[lo
]))
7668 *first_bad
= BB_OFFSET(p
[lo
]);
7669 *bad_sectors
= BB_LEN(p
[lo
]);
7675 if (read_seqretry(&bb
->lock
, seq
))
7680 EXPORT_SYMBOL_GPL(md_is_badblock
);
7683 * Add a range of bad blocks to the table.
7684 * This might extend the table, or might contract it
7685 * if two adjacent ranges can be merged.
7686 * We binary-search to find the 'insertion' point, then
7687 * decide how best to handle it.
7689 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7697 /* badblocks are disabled */
7701 /* round the start down, and the end up */
7702 sector_t next
= s
+ sectors
;
7704 next
+= (1<<bb
->shift
) - 1;
7709 write_seqlock_irq(&bb
->lock
);
7714 /* Find the last range that starts at-or-before 's' */
7715 while (hi
- lo
> 1) {
7716 int mid
= (lo
+ hi
) / 2;
7717 sector_t a
= BB_OFFSET(p
[mid
]);
7723 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7727 /* we found a range that might merge with the start
7730 sector_t a
= BB_OFFSET(p
[lo
]);
7731 sector_t e
= a
+ BB_LEN(p
[lo
]);
7732 int ack
= BB_ACK(p
[lo
]);
7734 /* Yes, we can merge with a previous range */
7735 if (s
== a
&& s
+ sectors
>= e
)
7736 /* new range covers old */
7739 ack
= ack
&& acknowledged
;
7741 if (e
< s
+ sectors
)
7743 if (e
- a
<= BB_MAX_LEN
) {
7744 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
7747 /* does not all fit in one range,
7748 * make p[lo] maximal
7750 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
7751 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7757 if (sectors
&& hi
< bb
->count
) {
7758 /* 'hi' points to the first range that starts after 's'.
7759 * Maybe we can merge with the start of that range */
7760 sector_t a
= BB_OFFSET(p
[hi
]);
7761 sector_t e
= a
+ BB_LEN(p
[hi
]);
7762 int ack
= BB_ACK(p
[hi
]);
7763 if (a
<= s
+ sectors
) {
7764 /* merging is possible */
7765 if (e
<= s
+ sectors
) {
7770 ack
= ack
&& acknowledged
;
7773 if (e
- a
<= BB_MAX_LEN
) {
7774 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
7777 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7785 if (sectors
== 0 && hi
< bb
->count
) {
7786 /* we might be able to combine lo and hi */
7787 /* Note: 's' is at the end of 'lo' */
7788 sector_t a
= BB_OFFSET(p
[hi
]);
7789 int lolen
= BB_LEN(p
[lo
]);
7790 int hilen
= BB_LEN(p
[hi
]);
7791 int newlen
= lolen
+ hilen
- (s
- a
);
7792 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
7793 /* yes, we can combine them */
7794 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
7795 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
7796 memmove(p
+ hi
, p
+ hi
+ 1,
7797 (bb
->count
- hi
- 1) * 8);
7802 /* didn't merge (it all).
7803 * Need to add a range just before 'hi' */
7804 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7805 /* No room for more */
7809 int this_sectors
= sectors
;
7810 memmove(p
+ hi
+ 1, p
+ hi
,
7811 (bb
->count
- hi
) * 8);
7814 if (this_sectors
> BB_MAX_LEN
)
7815 this_sectors
= BB_MAX_LEN
;
7816 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
7817 sectors
-= this_sectors
;
7824 bb
->unacked_exist
= 1;
7825 write_sequnlock_irq(&bb
->lock
);
7830 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
7833 int rv
= md_set_badblocks(&rdev
->badblocks
,
7834 s
+ rdev
->data_offset
, sectors
, acknowledged
);
7836 /* Make sure they get written out promptly */
7837 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
7838 md_wakeup_thread(rdev
->mddev
->thread
);
7842 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
7845 * Remove a range of bad blocks from the table.
7846 * This may involve extending the table if we spilt a region,
7847 * but it must not fail. So if the table becomes full, we just
7848 * drop the remove request.
7850 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
7854 sector_t target
= s
+ sectors
;
7857 if (bb
->shift
> 0) {
7858 /* When clearing we round the start up and the end down.
7859 * This should not matter as the shift should align with
7860 * the block size and no rounding should ever be needed.
7861 * However it is better the think a block is bad when it
7862 * isn't than to think a block is not bad when it is.
7864 s
+= (1<<bb
->shift
) - 1;
7866 target
>>= bb
->shift
;
7867 sectors
= target
- s
;
7870 write_seqlock_irq(&bb
->lock
);
7875 /* Find the last range that starts before 'target' */
7876 while (hi
- lo
> 1) {
7877 int mid
= (lo
+ hi
) / 2;
7878 sector_t a
= BB_OFFSET(p
[mid
]);
7885 /* p[lo] is the last range that could overlap the
7886 * current range. Earlier ranges could also overlap,
7887 * but only this one can overlap the end of the range.
7889 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
7890 /* Partial overlap, leave the tail of this range */
7891 int ack
= BB_ACK(p
[lo
]);
7892 sector_t a
= BB_OFFSET(p
[lo
]);
7893 sector_t end
= a
+ BB_LEN(p
[lo
]);
7896 /* we need to split this range */
7897 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7901 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
7903 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
7906 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
7907 /* there is no longer an overlap */
7912 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7913 /* This range does overlap */
7914 if (BB_OFFSET(p
[lo
]) < s
) {
7915 /* Keep the early parts of this range. */
7916 int ack
= BB_ACK(p
[lo
]);
7917 sector_t start
= BB_OFFSET(p
[lo
]);
7918 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
7919 /* now low doesn't overlap, so.. */
7924 /* 'lo' is strictly before, 'hi' is strictly after,
7925 * anything between needs to be discarded
7928 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
7929 bb
->count
-= (hi
- lo
- 1);
7935 write_sequnlock_irq(&bb
->lock
);
7939 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
)
7941 return md_clear_badblocks(&rdev
->badblocks
,
7942 s
+ rdev
->data_offset
,
7945 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
7948 * Acknowledge all bad blocks in a list.
7949 * This only succeeds if ->changed is clear. It is used by
7950 * in-kernel metadata updates
7952 void md_ack_all_badblocks(struct badblocks
*bb
)
7954 if (bb
->page
== NULL
|| bb
->changed
)
7955 /* no point even trying */
7957 write_seqlock_irq(&bb
->lock
);
7959 if (bb
->changed
== 0) {
7962 for (i
= 0; i
< bb
->count
; i
++) {
7963 if (!BB_ACK(p
[i
])) {
7964 sector_t start
= BB_OFFSET(p
[i
]);
7965 int len
= BB_LEN(p
[i
]);
7966 p
[i
] = BB_MAKE(start
, len
, 1);
7969 bb
->unacked_exist
= 0;
7971 write_sequnlock_irq(&bb
->lock
);
7973 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
7975 /* sysfs access to bad-blocks list.
7976 * We present two files.
7977 * 'bad-blocks' lists sector numbers and lengths of ranges that
7978 * are recorded as bad. The list is truncated to fit within
7979 * the one-page limit of sysfs.
7980 * Writing "sector length" to this file adds an acknowledged
7982 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7983 * been acknowledged. Writing to this file adds bad blocks
7984 * without acknowledging them. This is largely for testing.
7988 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
7999 seq
= read_seqbegin(&bb
->lock
);
8004 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8005 sector_t s
= BB_OFFSET(p
[i
]);
8006 unsigned int length
= BB_LEN(p
[i
]);
8007 int ack
= BB_ACK(p
[i
]);
8013 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8014 (unsigned long long)s
<< bb
->shift
,
8015 length
<< bb
->shift
);
8017 if (unack
&& len
== 0)
8018 bb
->unacked_exist
= 0;
8020 if (read_seqretry(&bb
->lock
, seq
))
8029 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8031 unsigned long long sector
;
8035 /* Allow clearing via sysfs *only* for testing/debugging.
8036 * Normally only a successful write may clear a badblock
8039 if (page
[0] == '-') {
8043 #endif /* DO_DEBUG */
8045 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8047 if (newline
!= '\n')
8059 md_clear_badblocks(bb
, sector
, length
);
8062 #endif /* DO_DEBUG */
8063 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8069 static int md_notify_reboot(struct notifier_block
*this,
8070 unsigned long code
, void *x
)
8072 struct list_head
*tmp
;
8073 struct mddev
*mddev
;
8076 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
8078 printk(KERN_INFO
"md: stopping all md devices.\n");
8080 for_each_mddev(mddev
, tmp
) {
8081 if (mddev_trylock(mddev
)) {
8082 /* Force a switch to readonly even array
8083 * appears to still be in use. Hence
8086 md_set_readonly(mddev
, 100);
8087 mddev_unlock(mddev
);
8092 * certain more exotic SCSI devices are known to be
8093 * volatile wrt too early system reboots. While the
8094 * right place to handle this issue is the given
8095 * driver, we do want to have a safe RAID driver ...
8103 static struct notifier_block md_notifier
= {
8104 .notifier_call
= md_notify_reboot
,
8106 .priority
= INT_MAX
, /* before any real devices */
8109 static void md_geninit(void)
8111 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8113 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8116 static int __init
md_init(void)
8120 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8124 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8128 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8131 if ((ret
= register_blkdev(0, "mdp")) < 0)
8135 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8136 md_probe
, NULL
, NULL
);
8137 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8138 md_probe
, NULL
, NULL
);
8140 register_reboot_notifier(&md_notifier
);
8141 raid_table_header
= register_sysctl_table(raid_root_table
);
8147 unregister_blkdev(MD_MAJOR
, "md");
8149 destroy_workqueue(md_misc_wq
);
8151 destroy_workqueue(md_wq
);
8159 * Searches all registered partitions for autorun RAID arrays
8163 static LIST_HEAD(all_detected_devices
);
8164 struct detected_devices_node
{
8165 struct list_head list
;
8169 void md_autodetect_dev(dev_t dev
)
8171 struct detected_devices_node
*node_detected_dev
;
8173 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8174 if (node_detected_dev
) {
8175 node_detected_dev
->dev
= dev
;
8176 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8178 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8179 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8184 static void autostart_arrays(int part
)
8186 struct md_rdev
*rdev
;
8187 struct detected_devices_node
*node_detected_dev
;
8189 int i_scanned
, i_passed
;
8194 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8196 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8198 node_detected_dev
= list_entry(all_detected_devices
.next
,
8199 struct detected_devices_node
, list
);
8200 list_del(&node_detected_dev
->list
);
8201 dev
= node_detected_dev
->dev
;
8202 kfree(node_detected_dev
);
8203 rdev
= md_import_device(dev
,0, 90);
8207 if (test_bit(Faulty
, &rdev
->flags
)) {
8211 set_bit(AutoDetected
, &rdev
->flags
);
8212 list_add(&rdev
->same_set
, &pending_raid_disks
);
8216 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8217 i_scanned
, i_passed
);
8219 autorun_devices(part
);
8222 #endif /* !MODULE */
8224 static __exit
void md_exit(void)
8226 struct mddev
*mddev
;
8227 struct list_head
*tmp
;
8229 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8230 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8232 unregister_blkdev(MD_MAJOR
,"md");
8233 unregister_blkdev(mdp_major
, "mdp");
8234 unregister_reboot_notifier(&md_notifier
);
8235 unregister_sysctl_table(raid_table_header
);
8236 remove_proc_entry("mdstat", NULL
);
8237 for_each_mddev(mddev
, tmp
) {
8238 export_array(mddev
);
8239 mddev
->hold_active
= 0;
8241 destroy_workqueue(md_misc_wq
);
8242 destroy_workqueue(md_wq
);
8245 subsys_initcall(md_init
);
8246 module_exit(md_exit
)
8248 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8250 return sprintf(buffer
, "%d", start_readonly
);
8252 static int set_ro(const char *val
, struct kernel_param
*kp
)
8255 int num
= simple_strtoul(val
, &e
, 10);
8256 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8257 start_readonly
= num
;
8263 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8264 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8266 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8268 EXPORT_SYMBOL(register_md_personality
);
8269 EXPORT_SYMBOL(unregister_md_personality
);
8270 EXPORT_SYMBOL(md_error
);
8271 EXPORT_SYMBOL(md_done_sync
);
8272 EXPORT_SYMBOL(md_write_start
);
8273 EXPORT_SYMBOL(md_write_end
);
8274 EXPORT_SYMBOL(md_register_thread
);
8275 EXPORT_SYMBOL(md_unregister_thread
);
8276 EXPORT_SYMBOL(md_wakeup_thread
);
8277 EXPORT_SYMBOL(md_check_recovery
);
8278 MODULE_LICENSE("GPL");
8279 MODULE_DESCRIPTION("MD RAID framework");
8281 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);