2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min
= 1000;
97 static int sysctl_speed_limit_max
= 200000;
98 static inline int speed_min(struct mddev
*mddev
)
100 return mddev
->sync_speed_min
?
101 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
104 static inline int speed_max(struct mddev
*mddev
)
106 return mddev
->sync_speed_max
?
107 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
110 static struct ctl_table_header
*raid_table_header
;
112 static ctl_table raid_table
[] = {
114 .procname
= "speed_limit_min",
115 .data
= &sysctl_speed_limit_min
,
116 .maxlen
= sizeof(int),
117 .mode
= S_IRUGO
|S_IWUSR
,
118 .proc_handler
= proc_dointvec
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
134 .mode
= S_IRUGO
|S_IXUGO
,
140 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static const struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio
*bio
)
160 struct mddev
*mddev
, **mddevp
;
165 bio_free(bio
, mddev
->bio_set
);
168 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
172 struct mddev
**mddevp
;
174 if (!mddev
|| !mddev
->bio_set
)
175 return bio_alloc(gfp_mask
, nr_iovecs
);
177 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
183 b
->bi_destructor
= mddev_bio_destructor
;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
188 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
192 struct mddev
**mddevp
;
194 if (!mddev
|| !mddev
->bio_set
)
195 return bio_clone(bio
, gfp_mask
);
197 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
203 b
->bi_destructor
= mddev_bio_destructor
;
205 if (bio_integrity(bio
)) {
208 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
218 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
220 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
227 struct bio_vec
*bvec
;
231 if (offset
== 0 && size
== bio
->bi_size
)
234 bio
->bi_sector
+= offset
;
237 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
239 while (bio
->bi_idx
< bio
->bi_vcnt
&&
240 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
241 /* remove this whole bio_vec */
242 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
245 if (bio
->bi_idx
< bio
->bi_vcnt
) {
246 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
247 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
249 /* avoid any complications with bi_idx being non-zero*/
251 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
252 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
253 bio
->bi_vcnt
-= bio
->bi_idx
;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec
, bio
, i
) {
258 if (sofar
+ bvec
->bv_len
> size
)
259 bvec
->bv_len
= size
- sofar
;
260 if (bvec
->bv_len
== 0) {
264 sofar
+= bvec
->bv_len
;
267 EXPORT_SYMBOL_GPL(md_trim_bio
);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
280 static atomic_t md_event_count
;
281 void md_new_event(struct mddev
*mddev
)
283 atomic_inc(&md_event_count
);
284 wake_up(&md_event_waiters
);
286 EXPORT_SYMBOL_GPL(md_new_event
);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev
*mddev
)
293 atomic_inc(&md_event_count
);
294 wake_up(&md_event_waiters
);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs
);
302 static DEFINE_SPINLOCK(all_mddevs_lock
);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static void md_make_request(struct request_queue
*q
, struct bio
*bio
)
337 const int rw
= bio_data_dir(bio
);
338 struct mddev
*mddev
= q
->queuedata
;
340 unsigned int sectors
;
342 if (mddev
== NULL
|| mddev
->pers
== NULL
347 smp_rmb(); /* Ensure implications of 'active' are visible */
349 if (mddev
->suspended
) {
352 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
353 TASK_UNINTERRUPTIBLE
);
354 if (!mddev
->suspended
)
360 finish_wait(&mddev
->sb_wait
, &__wait
);
362 atomic_inc(&mddev
->active_io
);
366 * save the sectors now since our bio can
367 * go away inside make_request
369 sectors
= bio_sectors(bio
);
370 mddev
->pers
->make_request(mddev
, bio
);
372 cpu
= part_stat_lock();
373 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
374 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
377 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
378 wake_up(&mddev
->sb_wait
);
381 /* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
387 void mddev_suspend(struct mddev
*mddev
)
389 BUG_ON(mddev
->suspended
);
390 mddev
->suspended
= 1;
392 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
393 mddev
->pers
->quiesce(mddev
, 1);
395 del_timer_sync(&mddev
->safemode_timer
);
397 EXPORT_SYMBOL_GPL(mddev_suspend
);
399 void mddev_resume(struct mddev
*mddev
)
401 mddev
->suspended
= 0;
402 wake_up(&mddev
->sb_wait
);
403 mddev
->pers
->quiesce(mddev
, 0);
405 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
406 md_wakeup_thread(mddev
->thread
);
407 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
409 EXPORT_SYMBOL_GPL(mddev_resume
);
411 int mddev_congested(struct mddev
*mddev
, int bits
)
413 return mddev
->suspended
;
415 EXPORT_SYMBOL(mddev_congested
);
418 * Generic flush handling for md
421 static void md_end_flush(struct bio
*bio
, int err
)
423 struct md_rdev
*rdev
= bio
->bi_private
;
424 struct mddev
*mddev
= rdev
->mddev
;
426 rdev_dec_pending(rdev
, mddev
);
428 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
429 /* The pre-request flush has finished */
430 queue_work(md_wq
, &mddev
->flush_work
);
435 static void md_submit_flush_data(struct work_struct
*ws
);
437 static void submit_flushes(struct work_struct
*ws
)
439 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
440 struct md_rdev
*rdev
;
442 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
443 atomic_set(&mddev
->flush_pending
, 1);
445 rdev_for_each_rcu(rdev
, mddev
)
446 if (rdev
->raid_disk
>= 0 &&
447 !test_bit(Faulty
, &rdev
->flags
)) {
448 /* Take two references, one is dropped
449 * when request finishes, one after
450 * we reclaim rcu_read_lock
453 atomic_inc(&rdev
->nr_pending
);
454 atomic_inc(&rdev
->nr_pending
);
456 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
457 bi
->bi_end_io
= md_end_flush
;
458 bi
->bi_private
= rdev
;
459 bi
->bi_bdev
= rdev
->bdev
;
460 atomic_inc(&mddev
->flush_pending
);
461 submit_bio(WRITE_FLUSH
, bi
);
463 rdev_dec_pending(rdev
, mddev
);
466 if (atomic_dec_and_test(&mddev
->flush_pending
))
467 queue_work(md_wq
, &mddev
->flush_work
);
470 static void md_submit_flush_data(struct work_struct
*ws
)
472 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
473 struct bio
*bio
= mddev
->flush_bio
;
475 if (bio
->bi_size
== 0)
476 /* an empty barrier - all done */
479 bio
->bi_rw
&= ~REQ_FLUSH
;
480 mddev
->pers
->make_request(mddev
, bio
);
483 mddev
->flush_bio
= NULL
;
484 wake_up(&mddev
->sb_wait
);
487 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
489 spin_lock_irq(&mddev
->write_lock
);
490 wait_event_lock_irq(mddev
->sb_wait
,
492 mddev
->write_lock
, /*nothing*/);
493 mddev
->flush_bio
= bio
;
494 spin_unlock_irq(&mddev
->write_lock
);
496 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
497 queue_work(md_wq
, &mddev
->flush_work
);
499 EXPORT_SYMBOL(md_flush_request
);
501 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
503 struct mddev
*mddev
= cb
->data
;
504 md_wakeup_thread(mddev
->thread
);
507 EXPORT_SYMBOL(md_unplug
);
509 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
511 atomic_inc(&mddev
->active
);
515 static void mddev_delayed_delete(struct work_struct
*ws
);
517 static void mddev_put(struct mddev
*mddev
)
519 struct bio_set
*bs
= NULL
;
521 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
523 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
524 mddev
->ctime
== 0 && !mddev
->hold_active
) {
525 /* Array is not configured at all, and not held active,
527 list_del_init(&mddev
->all_mddevs
);
529 mddev
->bio_set
= NULL
;
530 if (mddev
->gendisk
) {
531 /* We did a probe so need to clean up. Call
532 * queue_work inside the spinlock so that
533 * flush_workqueue() after mddev_find will
534 * succeed in waiting for the work to be done.
536 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
537 queue_work(md_misc_wq
, &mddev
->del_work
);
541 spin_unlock(&all_mddevs_lock
);
546 void mddev_init(struct mddev
*mddev
)
548 mutex_init(&mddev
->open_mutex
);
549 mutex_init(&mddev
->reconfig_mutex
);
550 mutex_init(&mddev
->bitmap_info
.mutex
);
551 INIT_LIST_HEAD(&mddev
->disks
);
552 INIT_LIST_HEAD(&mddev
->all_mddevs
);
553 init_timer(&mddev
->safemode_timer
);
554 atomic_set(&mddev
->active
, 1);
555 atomic_set(&mddev
->openers
, 0);
556 atomic_set(&mddev
->active_io
, 0);
557 spin_lock_init(&mddev
->write_lock
);
558 atomic_set(&mddev
->flush_pending
, 0);
559 init_waitqueue_head(&mddev
->sb_wait
);
560 init_waitqueue_head(&mddev
->recovery_wait
);
561 mddev
->reshape_position
= MaxSector
;
562 mddev
->reshape_backwards
= 0;
563 mddev
->resync_min
= 0;
564 mddev
->resync_max
= MaxSector
;
565 mddev
->level
= LEVEL_NONE
;
567 EXPORT_SYMBOL_GPL(mddev_init
);
569 static struct mddev
* mddev_find(dev_t unit
)
571 struct mddev
*mddev
, *new = NULL
;
573 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
574 unit
&= ~((1<<MdpMinorShift
)-1);
577 spin_lock(&all_mddevs_lock
);
580 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
581 if (mddev
->unit
== unit
) {
583 spin_unlock(&all_mddevs_lock
);
589 list_add(&new->all_mddevs
, &all_mddevs
);
590 spin_unlock(&all_mddevs_lock
);
591 new->hold_active
= UNTIL_IOCTL
;
595 /* find an unused unit number */
596 static int next_minor
= 512;
597 int start
= next_minor
;
601 dev
= MKDEV(MD_MAJOR
, next_minor
);
603 if (next_minor
> MINORMASK
)
605 if (next_minor
== start
) {
606 /* Oh dear, all in use. */
607 spin_unlock(&all_mddevs_lock
);
613 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
614 if (mddev
->unit
== dev
) {
620 new->md_minor
= MINOR(dev
);
621 new->hold_active
= UNTIL_STOP
;
622 list_add(&new->all_mddevs
, &all_mddevs
);
623 spin_unlock(&all_mddevs_lock
);
626 spin_unlock(&all_mddevs_lock
);
628 new = kzalloc(sizeof(*new), GFP_KERNEL
);
633 if (MAJOR(unit
) == MD_MAJOR
)
634 new->md_minor
= MINOR(unit
);
636 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
643 static inline int mddev_lock(struct mddev
* mddev
)
645 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
648 static inline int mddev_is_locked(struct mddev
*mddev
)
650 return mutex_is_locked(&mddev
->reconfig_mutex
);
653 static inline int mddev_trylock(struct mddev
* mddev
)
655 return mutex_trylock(&mddev
->reconfig_mutex
);
658 static struct attribute_group md_redundancy_group
;
660 static void mddev_unlock(struct mddev
* mddev
)
662 if (mddev
->to_remove
) {
663 /* These cannot be removed under reconfig_mutex as
664 * an access to the files will try to take reconfig_mutex
665 * while holding the file unremovable, which leads to
667 * So hold set sysfs_active while the remove in happeing,
668 * and anything else which might set ->to_remove or my
669 * otherwise change the sysfs namespace will fail with
670 * -EBUSY if sysfs_active is still set.
671 * We set sysfs_active under reconfig_mutex and elsewhere
672 * test it under the same mutex to ensure its correct value
675 struct attribute_group
*to_remove
= mddev
->to_remove
;
676 mddev
->to_remove
= NULL
;
677 mddev
->sysfs_active
= 1;
678 mutex_unlock(&mddev
->reconfig_mutex
);
680 if (mddev
->kobj
.sd
) {
681 if (to_remove
!= &md_redundancy_group
)
682 sysfs_remove_group(&mddev
->kobj
, to_remove
);
683 if (mddev
->pers
== NULL
||
684 mddev
->pers
->sync_request
== NULL
) {
685 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
686 if (mddev
->sysfs_action
)
687 sysfs_put(mddev
->sysfs_action
);
688 mddev
->sysfs_action
= NULL
;
691 mddev
->sysfs_active
= 0;
693 mutex_unlock(&mddev
->reconfig_mutex
);
695 /* As we've dropped the mutex we need a spinlock to
696 * make sure the thread doesn't disappear
698 spin_lock(&pers_lock
);
699 md_wakeup_thread(mddev
->thread
);
700 spin_unlock(&pers_lock
);
703 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
705 struct md_rdev
*rdev
;
707 rdev_for_each(rdev
, mddev
)
708 if (rdev
->desc_nr
== nr
)
714 static struct md_rdev
* find_rdev(struct mddev
* mddev
, dev_t dev
)
716 struct md_rdev
*rdev
;
718 rdev_for_each(rdev
, mddev
)
719 if (rdev
->bdev
->bd_dev
== dev
)
725 static struct md_personality
*find_pers(int level
, char *clevel
)
727 struct md_personality
*pers
;
728 list_for_each_entry(pers
, &pers_list
, list
) {
729 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
731 if (strcmp(pers
->name
, clevel
)==0)
737 /* return the offset of the super block in 512byte sectors */
738 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
740 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
741 return MD_NEW_SIZE_SECTORS(num_sectors
);
744 static int alloc_disk_sb(struct md_rdev
* rdev
)
749 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
750 if (!rdev
->sb_page
) {
751 printk(KERN_ALERT
"md: out of memory.\n");
758 void md_rdev_clear(struct md_rdev
*rdev
)
761 put_page(rdev
->sb_page
);
763 rdev
->sb_page
= NULL
;
768 put_page(rdev
->bb_page
);
769 rdev
->bb_page
= NULL
;
771 kfree(rdev
->badblocks
.page
);
772 rdev
->badblocks
.page
= NULL
;
774 EXPORT_SYMBOL_GPL(md_rdev_clear
);
776 static void super_written(struct bio
*bio
, int error
)
778 struct md_rdev
*rdev
= bio
->bi_private
;
779 struct mddev
*mddev
= rdev
->mddev
;
781 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
782 printk("md: super_written gets error=%d, uptodate=%d\n",
783 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
784 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
785 md_error(mddev
, rdev
);
788 if (atomic_dec_and_test(&mddev
->pending_writes
))
789 wake_up(&mddev
->sb_wait
);
793 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
794 sector_t sector
, int size
, struct page
*page
)
796 /* write first size bytes of page to sector of rdev
797 * Increment mddev->pending_writes before returning
798 * and decrement it on completion, waking up sb_wait
799 * if zero is reached.
800 * If an error occurred, call md_error
802 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
804 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
805 bio
->bi_sector
= sector
;
806 bio_add_page(bio
, page
, size
, 0);
807 bio
->bi_private
= rdev
;
808 bio
->bi_end_io
= super_written
;
810 atomic_inc(&mddev
->pending_writes
);
811 submit_bio(WRITE_FLUSH_FUA
, bio
);
814 void md_super_wait(struct mddev
*mddev
)
816 /* wait for all superblock writes that were scheduled to complete */
819 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
820 if (atomic_read(&mddev
->pending_writes
)==0)
824 finish_wait(&mddev
->sb_wait
, &wq
);
827 static void bi_complete(struct bio
*bio
, int error
)
829 complete((struct completion
*)bio
->bi_private
);
832 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
833 struct page
*page
, int rw
, bool metadata_op
)
835 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
836 struct completion event
;
841 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
842 rdev
->meta_bdev
: rdev
->bdev
;
844 bio
->bi_sector
= sector
+ rdev
->sb_start
;
845 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
846 (rdev
->mddev
->reshape_backwards
==
847 (sector
>= rdev
->mddev
->reshape_position
)))
848 bio
->bi_sector
= sector
+ rdev
->new_data_offset
;
850 bio
->bi_sector
= sector
+ rdev
->data_offset
;
851 bio_add_page(bio
, page
, size
, 0);
852 init_completion(&event
);
853 bio
->bi_private
= &event
;
854 bio
->bi_end_io
= bi_complete
;
856 wait_for_completion(&event
);
858 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
862 EXPORT_SYMBOL_GPL(sync_page_io
);
864 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
866 char b
[BDEVNAME_SIZE
];
867 if (!rdev
->sb_page
) {
875 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
881 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
882 bdevname(rdev
->bdev
,b
));
886 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
888 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
889 sb1
->set_uuid1
== sb2
->set_uuid1
&&
890 sb1
->set_uuid2
== sb2
->set_uuid2
&&
891 sb1
->set_uuid3
== sb2
->set_uuid3
;
894 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
897 mdp_super_t
*tmp1
, *tmp2
;
899 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
900 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
902 if (!tmp1
|| !tmp2
) {
904 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
912 * nr_disks is not constant
917 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
925 static u32
md_csum_fold(u32 csum
)
927 csum
= (csum
& 0xffff) + (csum
>> 16);
928 return (csum
& 0xffff) + (csum
>> 16);
931 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
934 u32
*sb32
= (u32
*)sb
;
936 unsigned int disk_csum
, csum
;
938 disk_csum
= sb
->sb_csum
;
941 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
943 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
947 /* This used to use csum_partial, which was wrong for several
948 * reasons including that different results are returned on
949 * different architectures. It isn't critical that we get exactly
950 * the same return value as before (we always csum_fold before
951 * testing, and that removes any differences). However as we
952 * know that csum_partial always returned a 16bit value on
953 * alphas, do a fold to maximise conformity to previous behaviour.
955 sb
->sb_csum
= md_csum_fold(disk_csum
);
957 sb
->sb_csum
= disk_csum
;
964 * Handle superblock details.
965 * We want to be able to handle multiple superblock formats
966 * so we have a common interface to them all, and an array of
967 * different handlers.
968 * We rely on user-space to write the initial superblock, and support
969 * reading and updating of superblocks.
970 * Interface methods are:
971 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
972 * loads and validates a superblock on dev.
973 * if refdev != NULL, compare superblocks on both devices
975 * 0 - dev has a superblock that is compatible with refdev
976 * 1 - dev has a superblock that is compatible and newer than refdev
977 * so dev should be used as the refdev in future
978 * -EINVAL superblock incompatible or invalid
979 * -othererror e.g. -EIO
981 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
982 * Verify that dev is acceptable into mddev.
983 * The first time, mddev->raid_disks will be 0, and data from
984 * dev should be merged in. Subsequent calls check that dev
985 * is new enough. Return 0 or -EINVAL
987 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
988 * Update the superblock for rdev with data in mddev
989 * This does not write to disc.
995 struct module
*owner
;
996 int (*load_super
)(struct md_rdev
*rdev
,
997 struct md_rdev
*refdev
,
999 int (*validate_super
)(struct mddev
*mddev
,
1000 struct md_rdev
*rdev
);
1001 void (*sync_super
)(struct mddev
*mddev
,
1002 struct md_rdev
*rdev
);
1003 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1004 sector_t num_sectors
);
1005 int (*allow_new_offset
)(struct md_rdev
*rdev
,
1006 unsigned long long new_offset
);
1010 * Check that the given mddev has no bitmap.
1012 * This function is called from the run method of all personalities that do not
1013 * support bitmaps. It prints an error message and returns non-zero if mddev
1014 * has a bitmap. Otherwise, it returns 0.
1017 int md_check_no_bitmap(struct mddev
*mddev
)
1019 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1021 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1022 mdname(mddev
), mddev
->pers
->name
);
1025 EXPORT_SYMBOL(md_check_no_bitmap
);
1028 * load_super for 0.90.0
1030 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1032 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1037 * Calculate the position of the superblock (512byte sectors),
1038 * it's at the end of the disk.
1040 * It also happens to be a multiple of 4Kb.
1042 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1044 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1045 if (ret
) return ret
;
1049 bdevname(rdev
->bdev
, b
);
1050 sb
= page_address(rdev
->sb_page
);
1052 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1053 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1058 if (sb
->major_version
!= 0 ||
1059 sb
->minor_version
< 90 ||
1060 sb
->minor_version
> 91) {
1061 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1062 sb
->major_version
, sb
->minor_version
,
1067 if (sb
->raid_disks
<= 0)
1070 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1071 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1076 rdev
->preferred_minor
= sb
->md_minor
;
1077 rdev
->data_offset
= 0;
1078 rdev
->new_data_offset
= 0;
1079 rdev
->sb_size
= MD_SB_BYTES
;
1080 rdev
->badblocks
.shift
= -1;
1082 if (sb
->level
== LEVEL_MULTIPATH
)
1085 rdev
->desc_nr
= sb
->this_disk
.number
;
1091 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1092 if (!uuid_equal(refsb
, sb
)) {
1093 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1094 b
, bdevname(refdev
->bdev
,b2
));
1097 if (!sb_equal(refsb
, sb
)) {
1098 printk(KERN_WARNING
"md: %s has same UUID"
1099 " but different superblock to %s\n",
1100 b
, bdevname(refdev
->bdev
, b2
));
1104 ev2
= md_event(refsb
);
1110 rdev
->sectors
= rdev
->sb_start
;
1111 /* Limit to 4TB as metadata cannot record more than that.
1112 * (not needed for Linear and RAID0 as metadata doesn't
1115 if (rdev
->sectors
>= (2ULL << 32) && sb
->level
>= 1)
1116 rdev
->sectors
= (2ULL << 32) - 2;
1118 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1119 /* "this cannot possibly happen" ... */
1127 * validate_super for 0.90.0
1129 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1132 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1133 __u64 ev1
= md_event(sb
);
1135 rdev
->raid_disk
= -1;
1136 clear_bit(Faulty
, &rdev
->flags
);
1137 clear_bit(In_sync
, &rdev
->flags
);
1138 clear_bit(WriteMostly
, &rdev
->flags
);
1140 if (mddev
->raid_disks
== 0) {
1141 mddev
->major_version
= 0;
1142 mddev
->minor_version
= sb
->minor_version
;
1143 mddev
->patch_version
= sb
->patch_version
;
1144 mddev
->external
= 0;
1145 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1146 mddev
->ctime
= sb
->ctime
;
1147 mddev
->utime
= sb
->utime
;
1148 mddev
->level
= sb
->level
;
1149 mddev
->clevel
[0] = 0;
1150 mddev
->layout
= sb
->layout
;
1151 mddev
->raid_disks
= sb
->raid_disks
;
1152 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1153 mddev
->events
= ev1
;
1154 mddev
->bitmap_info
.offset
= 0;
1155 mddev
->bitmap_info
.space
= 0;
1156 /* bitmap can use 60 K after the 4K superblocks */
1157 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1158 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1159 mddev
->reshape_backwards
= 0;
1161 if (mddev
->minor_version
>= 91) {
1162 mddev
->reshape_position
= sb
->reshape_position
;
1163 mddev
->delta_disks
= sb
->delta_disks
;
1164 mddev
->new_level
= sb
->new_level
;
1165 mddev
->new_layout
= sb
->new_layout
;
1166 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1167 if (mddev
->delta_disks
< 0)
1168 mddev
->reshape_backwards
= 1;
1170 mddev
->reshape_position
= MaxSector
;
1171 mddev
->delta_disks
= 0;
1172 mddev
->new_level
= mddev
->level
;
1173 mddev
->new_layout
= mddev
->layout
;
1174 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1177 if (sb
->state
& (1<<MD_SB_CLEAN
))
1178 mddev
->recovery_cp
= MaxSector
;
1180 if (sb
->events_hi
== sb
->cp_events_hi
&&
1181 sb
->events_lo
== sb
->cp_events_lo
) {
1182 mddev
->recovery_cp
= sb
->recovery_cp
;
1184 mddev
->recovery_cp
= 0;
1187 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1188 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1189 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1190 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1192 mddev
->max_disks
= MD_SB_DISKS
;
1194 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1195 mddev
->bitmap_info
.file
== NULL
) {
1196 mddev
->bitmap_info
.offset
=
1197 mddev
->bitmap_info
.default_offset
;
1198 mddev
->bitmap_info
.space
=
1199 mddev
->bitmap_info
.space
;
1202 } else if (mddev
->pers
== NULL
) {
1203 /* Insist on good event counter while assembling, except
1204 * for spares (which don't need an event count) */
1206 if (sb
->disks
[rdev
->desc_nr
].state
& (
1207 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1208 if (ev1
< mddev
->events
)
1210 } else if (mddev
->bitmap
) {
1211 /* if adding to array with a bitmap, then we can accept an
1212 * older device ... but not too old.
1214 if (ev1
< mddev
->bitmap
->events_cleared
)
1217 if (ev1
< mddev
->events
)
1218 /* just a hot-add of a new device, leave raid_disk at -1 */
1222 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1223 desc
= sb
->disks
+ rdev
->desc_nr
;
1225 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1226 set_bit(Faulty
, &rdev
->flags
);
1227 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1228 desc->raid_disk < mddev->raid_disks */) {
1229 set_bit(In_sync
, &rdev
->flags
);
1230 rdev
->raid_disk
= desc
->raid_disk
;
1231 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1232 /* active but not in sync implies recovery up to
1233 * reshape position. We don't know exactly where
1234 * that is, so set to zero for now */
1235 if (mddev
->minor_version
>= 91) {
1236 rdev
->recovery_offset
= 0;
1237 rdev
->raid_disk
= desc
->raid_disk
;
1240 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1241 set_bit(WriteMostly
, &rdev
->flags
);
1242 } else /* MULTIPATH are always insync */
1243 set_bit(In_sync
, &rdev
->flags
);
1248 * sync_super for 0.90.0
1250 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1253 struct md_rdev
*rdev2
;
1254 int next_spare
= mddev
->raid_disks
;
1257 /* make rdev->sb match mddev data..
1260 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1261 * 3/ any empty disks < next_spare become removed
1263 * disks[0] gets initialised to REMOVED because
1264 * we cannot be sure from other fields if it has
1265 * been initialised or not.
1268 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1270 rdev
->sb_size
= MD_SB_BYTES
;
1272 sb
= page_address(rdev
->sb_page
);
1274 memset(sb
, 0, sizeof(*sb
));
1276 sb
->md_magic
= MD_SB_MAGIC
;
1277 sb
->major_version
= mddev
->major_version
;
1278 sb
->patch_version
= mddev
->patch_version
;
1279 sb
->gvalid_words
= 0; /* ignored */
1280 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1281 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1282 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1283 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1285 sb
->ctime
= mddev
->ctime
;
1286 sb
->level
= mddev
->level
;
1287 sb
->size
= mddev
->dev_sectors
/ 2;
1288 sb
->raid_disks
= mddev
->raid_disks
;
1289 sb
->md_minor
= mddev
->md_minor
;
1290 sb
->not_persistent
= 0;
1291 sb
->utime
= mddev
->utime
;
1293 sb
->events_hi
= (mddev
->events
>>32);
1294 sb
->events_lo
= (u32
)mddev
->events
;
1296 if (mddev
->reshape_position
== MaxSector
)
1297 sb
->minor_version
= 90;
1299 sb
->minor_version
= 91;
1300 sb
->reshape_position
= mddev
->reshape_position
;
1301 sb
->new_level
= mddev
->new_level
;
1302 sb
->delta_disks
= mddev
->delta_disks
;
1303 sb
->new_layout
= mddev
->new_layout
;
1304 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1306 mddev
->minor_version
= sb
->minor_version
;
1309 sb
->recovery_cp
= mddev
->recovery_cp
;
1310 sb
->cp_events_hi
= (mddev
->events
>>32);
1311 sb
->cp_events_lo
= (u32
)mddev
->events
;
1312 if (mddev
->recovery_cp
== MaxSector
)
1313 sb
->state
= (1<< MD_SB_CLEAN
);
1315 sb
->recovery_cp
= 0;
1317 sb
->layout
= mddev
->layout
;
1318 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1320 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1321 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1323 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1324 rdev_for_each(rdev2
, mddev
) {
1327 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1329 if (rdev2
->raid_disk
>= 0 &&
1330 sb
->minor_version
>= 91)
1331 /* we have nowhere to store the recovery_offset,
1332 * but if it is not below the reshape_position,
1333 * we can piggy-back on that.
1336 if (rdev2
->raid_disk
< 0 ||
1337 test_bit(Faulty
, &rdev2
->flags
))
1340 desc_nr
= rdev2
->raid_disk
;
1342 desc_nr
= next_spare
++;
1343 rdev2
->desc_nr
= desc_nr
;
1344 d
= &sb
->disks
[rdev2
->desc_nr
];
1346 d
->number
= rdev2
->desc_nr
;
1347 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1348 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1350 d
->raid_disk
= rdev2
->raid_disk
;
1352 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1353 if (test_bit(Faulty
, &rdev2
->flags
))
1354 d
->state
= (1<<MD_DISK_FAULTY
);
1355 else if (is_active
) {
1356 d
->state
= (1<<MD_DISK_ACTIVE
);
1357 if (test_bit(In_sync
, &rdev2
->flags
))
1358 d
->state
|= (1<<MD_DISK_SYNC
);
1366 if (test_bit(WriteMostly
, &rdev2
->flags
))
1367 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1369 /* now set the "removed" and "faulty" bits on any missing devices */
1370 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1371 mdp_disk_t
*d
= &sb
->disks
[i
];
1372 if (d
->state
== 0 && d
->number
== 0) {
1375 d
->state
= (1<<MD_DISK_REMOVED
);
1376 d
->state
|= (1<<MD_DISK_FAULTY
);
1380 sb
->nr_disks
= nr_disks
;
1381 sb
->active_disks
= active
;
1382 sb
->working_disks
= working
;
1383 sb
->failed_disks
= failed
;
1384 sb
->spare_disks
= spare
;
1386 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1387 sb
->sb_csum
= calc_sb_csum(sb
);
1391 * rdev_size_change for 0.90.0
1393 static unsigned long long
1394 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1396 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1397 return 0; /* component must fit device */
1398 if (rdev
->mddev
->bitmap_info
.offset
)
1399 return 0; /* can't move bitmap */
1400 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1401 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1402 num_sectors
= rdev
->sb_start
;
1403 /* Limit to 4TB as metadata cannot record more than that.
1404 * 4TB == 2^32 KB, or 2*2^32 sectors.
1406 if (num_sectors
>= (2ULL << 32) && rdev
->mddev
->level
>= 1)
1407 num_sectors
= (2ULL << 32) - 2;
1408 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1410 md_super_wait(rdev
->mddev
);
1415 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1417 /* non-zero offset changes not possible with v0.90 */
1418 return new_offset
== 0;
1422 * version 1 superblock
1425 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1429 unsigned long long newcsum
;
1430 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1431 __le32
*isuper
= (__le32
*)sb
;
1434 disk_csum
= sb
->sb_csum
;
1437 for (i
=0; size
>=4; size
-= 4 )
1438 newcsum
+= le32_to_cpu(*isuper
++);
1441 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1443 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1444 sb
->sb_csum
= disk_csum
;
1445 return cpu_to_le32(csum
);
1448 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1450 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1452 struct mdp_superblock_1
*sb
;
1456 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1460 * Calculate the position of the superblock in 512byte sectors.
1461 * It is always aligned to a 4K boundary and
1462 * depeding on minor_version, it can be:
1463 * 0: At least 8K, but less than 12K, from end of device
1464 * 1: At start of device
1465 * 2: 4K from start of device.
1467 switch(minor_version
) {
1469 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1471 sb_start
&= ~(sector_t
)(4*2-1);
1482 rdev
->sb_start
= sb_start
;
1484 /* superblock is rarely larger than 1K, but it can be larger,
1485 * and it is safe to read 4k, so we do that
1487 ret
= read_disk_sb(rdev
, 4096);
1488 if (ret
) return ret
;
1491 sb
= page_address(rdev
->sb_page
);
1493 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1494 sb
->major_version
!= cpu_to_le32(1) ||
1495 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1496 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1497 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1500 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1501 printk("md: invalid superblock checksum on %s\n",
1502 bdevname(rdev
->bdev
,b
));
1505 if (le64_to_cpu(sb
->data_size
) < 10) {
1506 printk("md: data_size too small on %s\n",
1507 bdevname(rdev
->bdev
,b
));
1512 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1513 /* Some padding is non-zero, might be a new feature */
1516 rdev
->preferred_minor
= 0xffff;
1517 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1518 rdev
->new_data_offset
= rdev
->data_offset
;
1519 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1520 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1521 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1522 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1524 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1525 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1526 if (rdev
->sb_size
& bmask
)
1527 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1530 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1533 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1536 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1539 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1541 if (!rdev
->bb_page
) {
1542 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1546 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1547 rdev
->badblocks
.count
== 0) {
1548 /* need to load the bad block list.
1549 * Currently we limit it to one page.
1555 int sectors
= le16_to_cpu(sb
->bblog_size
);
1556 if (sectors
> (PAGE_SIZE
/ 512))
1558 offset
= le32_to_cpu(sb
->bblog_offset
);
1561 bb_sector
= (long long)offset
;
1562 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1563 rdev
->bb_page
, READ
, true))
1565 bbp
= (u64
*)page_address(rdev
->bb_page
);
1566 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1567 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1568 u64 bb
= le64_to_cpu(*bbp
);
1569 int count
= bb
& (0x3ff);
1570 u64 sector
= bb
>> 10;
1571 sector
<<= sb
->bblog_shift
;
1572 count
<<= sb
->bblog_shift
;
1575 if (md_set_badblocks(&rdev
->badblocks
,
1576 sector
, count
, 1) == 0)
1579 } else if (sb
->bblog_offset
== 0)
1580 rdev
->badblocks
.shift
= -1;
1586 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1588 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1589 sb
->level
!= refsb
->level
||
1590 sb
->layout
!= refsb
->layout
||
1591 sb
->chunksize
!= refsb
->chunksize
) {
1592 printk(KERN_WARNING
"md: %s has strangely different"
1593 " superblock to %s\n",
1594 bdevname(rdev
->bdev
,b
),
1595 bdevname(refdev
->bdev
,b2
));
1598 ev1
= le64_to_cpu(sb
->events
);
1599 ev2
= le64_to_cpu(refsb
->events
);
1606 if (minor_version
) {
1607 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1608 sectors
-= rdev
->data_offset
;
1610 sectors
= rdev
->sb_start
;
1611 if (sectors
< le64_to_cpu(sb
->data_size
))
1613 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1617 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1619 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1620 __u64 ev1
= le64_to_cpu(sb
->events
);
1622 rdev
->raid_disk
= -1;
1623 clear_bit(Faulty
, &rdev
->flags
);
1624 clear_bit(In_sync
, &rdev
->flags
);
1625 clear_bit(WriteMostly
, &rdev
->flags
);
1627 if (mddev
->raid_disks
== 0) {
1628 mddev
->major_version
= 1;
1629 mddev
->patch_version
= 0;
1630 mddev
->external
= 0;
1631 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1632 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1633 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1634 mddev
->level
= le32_to_cpu(sb
->level
);
1635 mddev
->clevel
[0] = 0;
1636 mddev
->layout
= le32_to_cpu(sb
->layout
);
1637 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1638 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1639 mddev
->events
= ev1
;
1640 mddev
->bitmap_info
.offset
= 0;
1641 mddev
->bitmap_info
.space
= 0;
1642 /* Default location for bitmap is 1K after superblock
1643 * using 3K - total of 4K
1645 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1646 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1647 mddev
->reshape_backwards
= 0;
1649 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1650 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1652 mddev
->max_disks
= (4096-256)/2;
1654 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1655 mddev
->bitmap_info
.file
== NULL
) {
1656 mddev
->bitmap_info
.offset
=
1657 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1658 /* Metadata doesn't record how much space is available.
1659 * For 1.0, we assume we can use up to the superblock
1660 * if before, else to 4K beyond superblock.
1661 * For others, assume no change is possible.
1663 if (mddev
->minor_version
> 0)
1664 mddev
->bitmap_info
.space
= 0;
1665 else if (mddev
->bitmap_info
.offset
> 0)
1666 mddev
->bitmap_info
.space
=
1667 8 - mddev
->bitmap_info
.offset
;
1669 mddev
->bitmap_info
.space
=
1670 -mddev
->bitmap_info
.offset
;
1673 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1674 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1675 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1676 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1677 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1678 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1679 if (mddev
->delta_disks
< 0 ||
1680 (mddev
->delta_disks
== 0 &&
1681 (le32_to_cpu(sb
->feature_map
)
1682 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1683 mddev
->reshape_backwards
= 1;
1685 mddev
->reshape_position
= MaxSector
;
1686 mddev
->delta_disks
= 0;
1687 mddev
->new_level
= mddev
->level
;
1688 mddev
->new_layout
= mddev
->layout
;
1689 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1692 } else if (mddev
->pers
== NULL
) {
1693 /* Insist of good event counter while assembling, except for
1694 * spares (which don't need an event count) */
1696 if (rdev
->desc_nr
>= 0 &&
1697 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1698 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1699 if (ev1
< mddev
->events
)
1701 } else if (mddev
->bitmap
) {
1702 /* If adding to array with a bitmap, then we can accept an
1703 * older device, but not too old.
1705 if (ev1
< mddev
->bitmap
->events_cleared
)
1708 if (ev1
< mddev
->events
)
1709 /* just a hot-add of a new device, leave raid_disk at -1 */
1712 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1714 if (rdev
->desc_nr
< 0 ||
1715 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1719 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1721 case 0xffff: /* spare */
1723 case 0xfffe: /* faulty */
1724 set_bit(Faulty
, &rdev
->flags
);
1727 if ((le32_to_cpu(sb
->feature_map
) &
1728 MD_FEATURE_RECOVERY_OFFSET
))
1729 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1731 set_bit(In_sync
, &rdev
->flags
);
1732 rdev
->raid_disk
= role
;
1735 if (sb
->devflags
& WriteMostly1
)
1736 set_bit(WriteMostly
, &rdev
->flags
);
1737 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1738 set_bit(Replacement
, &rdev
->flags
);
1739 } else /* MULTIPATH are always insync */
1740 set_bit(In_sync
, &rdev
->flags
);
1745 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1747 struct mdp_superblock_1
*sb
;
1748 struct md_rdev
*rdev2
;
1750 /* make rdev->sb match mddev and rdev data. */
1752 sb
= page_address(rdev
->sb_page
);
1754 sb
->feature_map
= 0;
1756 sb
->recovery_offset
= cpu_to_le64(0);
1757 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1759 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1760 sb
->events
= cpu_to_le64(mddev
->events
);
1762 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1764 sb
->resync_offset
= cpu_to_le64(0);
1766 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1768 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1769 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1770 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1771 sb
->level
= cpu_to_le32(mddev
->level
);
1772 sb
->layout
= cpu_to_le32(mddev
->layout
);
1774 if (test_bit(WriteMostly
, &rdev
->flags
))
1775 sb
->devflags
|= WriteMostly1
;
1777 sb
->devflags
&= ~WriteMostly1
;
1778 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1779 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1781 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1782 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1783 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1786 if (rdev
->raid_disk
>= 0 &&
1787 !test_bit(In_sync
, &rdev
->flags
)) {
1789 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1790 sb
->recovery_offset
=
1791 cpu_to_le64(rdev
->recovery_offset
);
1793 if (test_bit(Replacement
, &rdev
->flags
))
1795 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1797 if (mddev
->reshape_position
!= MaxSector
) {
1798 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1799 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1800 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1801 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1802 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1803 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1804 if (mddev
->delta_disks
== 0 &&
1805 mddev
->reshape_backwards
)
1807 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1808 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1810 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1811 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1812 - rdev
->data_offset
));
1816 if (rdev
->badblocks
.count
== 0)
1817 /* Nothing to do for bad blocks*/ ;
1818 else if (sb
->bblog_offset
== 0)
1819 /* Cannot record bad blocks on this device */
1820 md_error(mddev
, rdev
);
1822 struct badblocks
*bb
= &rdev
->badblocks
;
1823 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1825 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1830 seq
= read_seqbegin(&bb
->lock
);
1832 memset(bbp
, 0xff, PAGE_SIZE
);
1834 for (i
= 0 ; i
< bb
->count
; i
++) {
1835 u64 internal_bb
= *p
++;
1836 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1837 | BB_LEN(internal_bb
));
1838 *bbp
++ = cpu_to_le64(store_bb
);
1841 if (read_seqretry(&bb
->lock
, seq
))
1844 bb
->sector
= (rdev
->sb_start
+
1845 (int)le32_to_cpu(sb
->bblog_offset
));
1846 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1851 rdev_for_each(rdev2
, mddev
)
1852 if (rdev2
->desc_nr
+1 > max_dev
)
1853 max_dev
= rdev2
->desc_nr
+1;
1855 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1857 sb
->max_dev
= cpu_to_le32(max_dev
);
1858 rdev
->sb_size
= max_dev
* 2 + 256;
1859 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1860 if (rdev
->sb_size
& bmask
)
1861 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1863 max_dev
= le32_to_cpu(sb
->max_dev
);
1865 for (i
=0; i
<max_dev
;i
++)
1866 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1868 rdev_for_each(rdev2
, mddev
) {
1870 if (test_bit(Faulty
, &rdev2
->flags
))
1871 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1872 else if (test_bit(In_sync
, &rdev2
->flags
))
1873 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1874 else if (rdev2
->raid_disk
>= 0)
1875 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1877 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1880 sb
->sb_csum
= calc_sb_1_csum(sb
);
1883 static unsigned long long
1884 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1886 struct mdp_superblock_1
*sb
;
1887 sector_t max_sectors
;
1888 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1889 return 0; /* component must fit device */
1890 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1891 return 0; /* too confusing */
1892 if (rdev
->sb_start
< rdev
->data_offset
) {
1893 /* minor versions 1 and 2; superblock before data */
1894 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1895 max_sectors
-= rdev
->data_offset
;
1896 if (!num_sectors
|| num_sectors
> max_sectors
)
1897 num_sectors
= max_sectors
;
1898 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1899 /* minor version 0 with bitmap we can't move */
1902 /* minor version 0; superblock after data */
1904 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1905 sb_start
&= ~(sector_t
)(4*2 - 1);
1906 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1907 if (!num_sectors
|| num_sectors
> max_sectors
)
1908 num_sectors
= max_sectors
;
1909 rdev
->sb_start
= sb_start
;
1911 sb
= page_address(rdev
->sb_page
);
1912 sb
->data_size
= cpu_to_le64(num_sectors
);
1913 sb
->super_offset
= rdev
->sb_start
;
1914 sb
->sb_csum
= calc_sb_1_csum(sb
);
1915 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1917 md_super_wait(rdev
->mddev
);
1923 super_1_allow_new_offset(struct md_rdev
*rdev
,
1924 unsigned long long new_offset
)
1926 /* All necessary checks on new >= old have been done */
1927 struct bitmap
*bitmap
;
1928 if (new_offset
>= rdev
->data_offset
)
1931 /* with 1.0 metadata, there is no metadata to tread on
1932 * so we can always move back */
1933 if (rdev
->mddev
->minor_version
== 0)
1936 /* otherwise we must be sure not to step on
1937 * any metadata, so stay:
1938 * 36K beyond start of superblock
1939 * beyond end of badblocks
1940 * beyond write-intent bitmap
1942 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1944 bitmap
= rdev
->mddev
->bitmap
;
1945 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1946 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1947 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1949 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1955 static struct super_type super_types
[] = {
1958 .owner
= THIS_MODULE
,
1959 .load_super
= super_90_load
,
1960 .validate_super
= super_90_validate
,
1961 .sync_super
= super_90_sync
,
1962 .rdev_size_change
= super_90_rdev_size_change
,
1963 .allow_new_offset
= super_90_allow_new_offset
,
1967 .owner
= THIS_MODULE
,
1968 .load_super
= super_1_load
,
1969 .validate_super
= super_1_validate
,
1970 .sync_super
= super_1_sync
,
1971 .rdev_size_change
= super_1_rdev_size_change
,
1972 .allow_new_offset
= super_1_allow_new_offset
,
1976 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1978 if (mddev
->sync_super
) {
1979 mddev
->sync_super(mddev
, rdev
);
1983 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1985 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1988 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1990 struct md_rdev
*rdev
, *rdev2
;
1993 rdev_for_each_rcu(rdev
, mddev1
)
1994 rdev_for_each_rcu(rdev2
, mddev2
)
1995 if (rdev
->bdev
->bd_contains
==
1996 rdev2
->bdev
->bd_contains
) {
2004 static LIST_HEAD(pending_raid_disks
);
2007 * Try to register data integrity profile for an mddev
2009 * This is called when an array is started and after a disk has been kicked
2010 * from the array. It only succeeds if all working and active component devices
2011 * are integrity capable with matching profiles.
2013 int md_integrity_register(struct mddev
*mddev
)
2015 struct md_rdev
*rdev
, *reference
= NULL
;
2017 if (list_empty(&mddev
->disks
))
2018 return 0; /* nothing to do */
2019 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
2020 return 0; /* shouldn't register, or already is */
2021 rdev_for_each(rdev
, mddev
) {
2022 /* skip spares and non-functional disks */
2023 if (test_bit(Faulty
, &rdev
->flags
))
2025 if (rdev
->raid_disk
< 0)
2028 /* Use the first rdev as the reference */
2032 /* does this rdev's profile match the reference profile? */
2033 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
2034 rdev
->bdev
->bd_disk
) < 0)
2037 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2040 * All component devices are integrity capable and have matching
2041 * profiles, register the common profile for the md device.
2043 if (blk_integrity_register(mddev
->gendisk
,
2044 bdev_get_integrity(reference
->bdev
)) != 0) {
2045 printk(KERN_ERR
"md: failed to register integrity for %s\n",
2049 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2050 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2051 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2057 EXPORT_SYMBOL(md_integrity_register
);
2059 /* Disable data integrity if non-capable/non-matching disk is being added */
2060 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2062 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2063 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2065 if (!bi_mddev
) /* nothing to do */
2067 if (rdev
->raid_disk
< 0) /* skip spares */
2069 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
2070 rdev
->bdev
->bd_disk
) >= 0)
2072 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
2073 blk_integrity_unregister(mddev
->gendisk
);
2075 EXPORT_SYMBOL(md_integrity_add_rdev
);
2077 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2079 char b
[BDEVNAME_SIZE
];
2089 /* prevent duplicates */
2090 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2093 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2094 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2095 rdev
->sectors
< mddev
->dev_sectors
)) {
2097 /* Cannot change size, so fail
2098 * If mddev->level <= 0, then we don't care
2099 * about aligning sizes (e.g. linear)
2101 if (mddev
->level
> 0)
2104 mddev
->dev_sectors
= rdev
->sectors
;
2107 /* Verify rdev->desc_nr is unique.
2108 * If it is -1, assign a free number, else
2109 * check number is not in use
2111 if (rdev
->desc_nr
< 0) {
2113 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2114 while (find_rdev_nr(mddev
, choice
))
2116 rdev
->desc_nr
= choice
;
2118 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2121 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2122 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2123 mdname(mddev
), mddev
->max_disks
);
2126 bdevname(rdev
->bdev
,b
);
2127 while ( (s
=strchr(b
, '/')) != NULL
)
2130 rdev
->mddev
= mddev
;
2131 printk(KERN_INFO
"md: bind<%s>\n", b
);
2133 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2136 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2137 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2138 /* failure here is OK */;
2139 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2141 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2142 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2144 /* May as well allow recovery to be retried once */
2145 mddev
->recovery_disabled
++;
2150 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2155 static void md_delayed_delete(struct work_struct
*ws
)
2157 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2158 kobject_del(&rdev
->kobj
);
2159 kobject_put(&rdev
->kobj
);
2162 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2164 char b
[BDEVNAME_SIZE
];
2169 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2170 list_del_rcu(&rdev
->same_set
);
2171 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2173 sysfs_remove_link(&rdev
->kobj
, "block");
2174 sysfs_put(rdev
->sysfs_state
);
2175 rdev
->sysfs_state
= NULL
;
2176 rdev
->badblocks
.count
= 0;
2177 /* We need to delay this, otherwise we can deadlock when
2178 * writing to 'remove' to "dev/state". We also need
2179 * to delay it due to rcu usage.
2182 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2183 kobject_get(&rdev
->kobj
);
2184 queue_work(md_misc_wq
, &rdev
->del_work
);
2188 * prevent the device from being mounted, repartitioned or
2189 * otherwise reused by a RAID array (or any other kernel
2190 * subsystem), by bd_claiming the device.
2192 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2195 struct block_device
*bdev
;
2196 char b
[BDEVNAME_SIZE
];
2198 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2199 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2201 printk(KERN_ERR
"md: could not open %s.\n",
2202 __bdevname(dev
, b
));
2203 return PTR_ERR(bdev
);
2209 static void unlock_rdev(struct md_rdev
*rdev
)
2211 struct block_device
*bdev
= rdev
->bdev
;
2215 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2218 void md_autodetect_dev(dev_t dev
);
2220 static void export_rdev(struct md_rdev
* rdev
)
2222 char b
[BDEVNAME_SIZE
];
2223 printk(KERN_INFO
"md: export_rdev(%s)\n",
2224 bdevname(rdev
->bdev
,b
));
2227 md_rdev_clear(rdev
);
2229 if (test_bit(AutoDetected
, &rdev
->flags
))
2230 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2233 kobject_put(&rdev
->kobj
);
2236 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2238 unbind_rdev_from_array(rdev
);
2242 static void export_array(struct mddev
*mddev
)
2244 struct md_rdev
*rdev
, *tmp
;
2246 rdev_for_each_safe(rdev
, tmp
, mddev
) {
2251 kick_rdev_from_array(rdev
);
2253 if (!list_empty(&mddev
->disks
))
2255 mddev
->raid_disks
= 0;
2256 mddev
->major_version
= 0;
2259 static void print_desc(mdp_disk_t
*desc
)
2261 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2262 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2265 static void print_sb_90(mdp_super_t
*sb
)
2270 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2271 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2272 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2274 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2275 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2276 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2277 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2278 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2279 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2280 sb
->failed_disks
, sb
->spare_disks
,
2281 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2284 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2287 desc
= sb
->disks
+ i
;
2288 if (desc
->number
|| desc
->major
|| desc
->minor
||
2289 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2290 printk(" D %2d: ", i
);
2294 printk(KERN_INFO
"md: THIS: ");
2295 print_desc(&sb
->this_disk
);
2298 static void print_sb_1(struct mdp_superblock_1
*sb
)
2302 uuid
= sb
->set_uuid
;
2304 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2305 "md: Name: \"%s\" CT:%llu\n",
2306 le32_to_cpu(sb
->major_version
),
2307 le32_to_cpu(sb
->feature_map
),
2310 (unsigned long long)le64_to_cpu(sb
->ctime
)
2311 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2313 uuid
= sb
->device_uuid
;
2315 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2317 "md: Dev:%08x UUID: %pU\n"
2318 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2319 "md: (MaxDev:%u) \n",
2320 le32_to_cpu(sb
->level
),
2321 (unsigned long long)le64_to_cpu(sb
->size
),
2322 le32_to_cpu(sb
->raid_disks
),
2323 le32_to_cpu(sb
->layout
),
2324 le32_to_cpu(sb
->chunksize
),
2325 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2326 (unsigned long long)le64_to_cpu(sb
->data_size
),
2327 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2328 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2329 le32_to_cpu(sb
->dev_number
),
2332 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2333 (unsigned long long)le64_to_cpu(sb
->events
),
2334 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2335 le32_to_cpu(sb
->sb_csum
),
2336 le32_to_cpu(sb
->max_dev
)
2340 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2342 char b
[BDEVNAME_SIZE
];
2343 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2344 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2345 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2347 if (rdev
->sb_loaded
) {
2348 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2349 switch (major_version
) {
2351 print_sb_90(page_address(rdev
->sb_page
));
2354 print_sb_1(page_address(rdev
->sb_page
));
2358 printk(KERN_INFO
"md: no rdev superblock!\n");
2361 static void md_print_devices(void)
2363 struct list_head
*tmp
;
2364 struct md_rdev
*rdev
;
2365 struct mddev
*mddev
;
2366 char b
[BDEVNAME_SIZE
];
2369 printk("md: **********************************\n");
2370 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2371 printk("md: **********************************\n");
2372 for_each_mddev(mddev
, tmp
) {
2375 bitmap_print_sb(mddev
->bitmap
);
2377 printk("%s: ", mdname(mddev
));
2378 rdev_for_each(rdev
, mddev
)
2379 printk("<%s>", bdevname(rdev
->bdev
,b
));
2382 rdev_for_each(rdev
, mddev
)
2383 print_rdev(rdev
, mddev
->major_version
);
2385 printk("md: **********************************\n");
2390 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2392 /* Update each superblock (in-memory image), but
2393 * if we are allowed to, skip spares which already
2394 * have the right event counter, or have one earlier
2395 * (which would mean they aren't being marked as dirty
2396 * with the rest of the array)
2398 struct md_rdev
*rdev
;
2399 rdev_for_each(rdev
, mddev
) {
2400 if (rdev
->sb_events
== mddev
->events
||
2402 rdev
->raid_disk
< 0 &&
2403 rdev
->sb_events
+1 == mddev
->events
)) {
2404 /* Don't update this superblock */
2405 rdev
->sb_loaded
= 2;
2407 sync_super(mddev
, rdev
);
2408 rdev
->sb_loaded
= 1;
2413 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2415 struct md_rdev
*rdev
;
2418 int any_badblocks_changed
= 0;
2421 /* First make sure individual recovery_offsets are correct */
2422 rdev_for_each(rdev
, mddev
) {
2423 if (rdev
->raid_disk
>= 0 &&
2424 mddev
->delta_disks
>= 0 &&
2425 !test_bit(In_sync
, &rdev
->flags
) &&
2426 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2427 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2430 if (!mddev
->persistent
) {
2431 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2432 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2433 if (!mddev
->external
) {
2434 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2435 rdev_for_each(rdev
, mddev
) {
2436 if (rdev
->badblocks
.changed
) {
2437 rdev
->badblocks
.changed
= 0;
2438 md_ack_all_badblocks(&rdev
->badblocks
);
2439 md_error(mddev
, rdev
);
2441 clear_bit(Blocked
, &rdev
->flags
);
2442 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2443 wake_up(&rdev
->blocked_wait
);
2446 wake_up(&mddev
->sb_wait
);
2450 spin_lock_irq(&mddev
->write_lock
);
2452 mddev
->utime
= get_seconds();
2454 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2456 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2457 /* just a clean<-> dirty transition, possibly leave spares alone,
2458 * though if events isn't the right even/odd, we will have to do
2464 if (mddev
->degraded
)
2465 /* If the array is degraded, then skipping spares is both
2466 * dangerous and fairly pointless.
2467 * Dangerous because a device that was removed from the array
2468 * might have a event_count that still looks up-to-date,
2469 * so it can be re-added without a resync.
2470 * Pointless because if there are any spares to skip,
2471 * then a recovery will happen and soon that array won't
2472 * be degraded any more and the spare can go back to sleep then.
2476 sync_req
= mddev
->in_sync
;
2478 /* If this is just a dirty<->clean transition, and the array is clean
2479 * and 'events' is odd, we can roll back to the previous clean state */
2481 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2482 && mddev
->can_decrease_events
2483 && mddev
->events
!= 1) {
2485 mddev
->can_decrease_events
= 0;
2487 /* otherwise we have to go forward and ... */
2489 mddev
->can_decrease_events
= nospares
;
2492 if (!mddev
->events
) {
2494 * oops, this 64-bit counter should never wrap.
2495 * Either we are in around ~1 trillion A.C., assuming
2496 * 1 reboot per second, or we have a bug:
2502 rdev_for_each(rdev
, mddev
) {
2503 if (rdev
->badblocks
.changed
)
2504 any_badblocks_changed
++;
2505 if (test_bit(Faulty
, &rdev
->flags
))
2506 set_bit(FaultRecorded
, &rdev
->flags
);
2509 sync_sbs(mddev
, nospares
);
2510 spin_unlock_irq(&mddev
->write_lock
);
2512 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2513 mdname(mddev
), mddev
->in_sync
);
2515 bitmap_update_sb(mddev
->bitmap
);
2516 rdev_for_each(rdev
, mddev
) {
2517 char b
[BDEVNAME_SIZE
];
2519 if (rdev
->sb_loaded
!= 1)
2520 continue; /* no noise on spare devices */
2522 if (!test_bit(Faulty
, &rdev
->flags
) &&
2523 rdev
->saved_raid_disk
== -1) {
2524 md_super_write(mddev
,rdev
,
2525 rdev
->sb_start
, rdev
->sb_size
,
2527 pr_debug("md: (write) %s's sb offset: %llu\n",
2528 bdevname(rdev
->bdev
, b
),
2529 (unsigned long long)rdev
->sb_start
);
2530 rdev
->sb_events
= mddev
->events
;
2531 if (rdev
->badblocks
.size
) {
2532 md_super_write(mddev
, rdev
,
2533 rdev
->badblocks
.sector
,
2534 rdev
->badblocks
.size
<< 9,
2536 rdev
->badblocks
.size
= 0;
2539 } else if (test_bit(Faulty
, &rdev
->flags
))
2540 pr_debug("md: %s (skipping faulty)\n",
2541 bdevname(rdev
->bdev
, b
));
2543 pr_debug("(skipping incremental s/r ");
2545 if (mddev
->level
== LEVEL_MULTIPATH
)
2546 /* only need to write one superblock... */
2549 md_super_wait(mddev
);
2550 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2552 spin_lock_irq(&mddev
->write_lock
);
2553 if (mddev
->in_sync
!= sync_req
||
2554 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2555 /* have to write it out again */
2556 spin_unlock_irq(&mddev
->write_lock
);
2559 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2560 spin_unlock_irq(&mddev
->write_lock
);
2561 wake_up(&mddev
->sb_wait
);
2562 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2563 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2565 rdev_for_each(rdev
, mddev
) {
2566 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2567 clear_bit(Blocked
, &rdev
->flags
);
2569 if (any_badblocks_changed
)
2570 md_ack_all_badblocks(&rdev
->badblocks
);
2571 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2572 wake_up(&rdev
->blocked_wait
);
2576 /* words written to sysfs files may, or may not, be \n terminated.
2577 * We want to accept with case. For this we use cmd_match.
2579 static int cmd_match(const char *cmd
, const char *str
)
2581 /* See if cmd, written into a sysfs file, matches
2582 * str. They must either be the same, or cmd can
2583 * have a trailing newline
2585 while (*cmd
&& *str
&& *cmd
== *str
) {
2596 struct rdev_sysfs_entry
{
2597 struct attribute attr
;
2598 ssize_t (*show
)(struct md_rdev
*, char *);
2599 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2603 state_show(struct md_rdev
*rdev
, char *page
)
2608 if (test_bit(Faulty
, &rdev
->flags
) ||
2609 rdev
->badblocks
.unacked_exist
) {
2610 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2613 if (test_bit(In_sync
, &rdev
->flags
)) {
2614 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2617 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2618 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2621 if (test_bit(Blocked
, &rdev
->flags
) ||
2622 (rdev
->badblocks
.unacked_exist
2623 && !test_bit(Faulty
, &rdev
->flags
))) {
2624 len
+= sprintf(page
+len
, "%sblocked", sep
);
2627 if (!test_bit(Faulty
, &rdev
->flags
) &&
2628 !test_bit(In_sync
, &rdev
->flags
)) {
2629 len
+= sprintf(page
+len
, "%sspare", sep
);
2632 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2633 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2636 if (test_bit(WantReplacement
, &rdev
->flags
)) {
2637 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2640 if (test_bit(Replacement
, &rdev
->flags
)) {
2641 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2645 return len
+sprintf(page
+len
, "\n");
2649 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2652 * faulty - simulates an error
2653 * remove - disconnects the device
2654 * writemostly - sets write_mostly
2655 * -writemostly - clears write_mostly
2656 * blocked - sets the Blocked flags
2657 * -blocked - clears the Blocked and possibly simulates an error
2658 * insync - sets Insync providing device isn't active
2659 * write_error - sets WriteErrorSeen
2660 * -write_error - clears WriteErrorSeen
2663 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2664 md_error(rdev
->mddev
, rdev
);
2665 if (test_bit(Faulty
, &rdev
->flags
))
2669 } else if (cmd_match(buf
, "remove")) {
2670 if (rdev
->raid_disk
>= 0)
2673 struct mddev
*mddev
= rdev
->mddev
;
2674 kick_rdev_from_array(rdev
);
2676 md_update_sb(mddev
, 1);
2677 md_new_event(mddev
);
2680 } else if (cmd_match(buf
, "writemostly")) {
2681 set_bit(WriteMostly
, &rdev
->flags
);
2683 } else if (cmd_match(buf
, "-writemostly")) {
2684 clear_bit(WriteMostly
, &rdev
->flags
);
2686 } else if (cmd_match(buf
, "blocked")) {
2687 set_bit(Blocked
, &rdev
->flags
);
2689 } else if (cmd_match(buf
, "-blocked")) {
2690 if (!test_bit(Faulty
, &rdev
->flags
) &&
2691 rdev
->badblocks
.unacked_exist
) {
2692 /* metadata handler doesn't understand badblocks,
2693 * so we need to fail the device
2695 md_error(rdev
->mddev
, rdev
);
2697 clear_bit(Blocked
, &rdev
->flags
);
2698 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2699 wake_up(&rdev
->blocked_wait
);
2700 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2701 md_wakeup_thread(rdev
->mddev
->thread
);
2704 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2705 set_bit(In_sync
, &rdev
->flags
);
2707 } else if (cmd_match(buf
, "write_error")) {
2708 set_bit(WriteErrorSeen
, &rdev
->flags
);
2710 } else if (cmd_match(buf
, "-write_error")) {
2711 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2713 } else if (cmd_match(buf
, "want_replacement")) {
2714 /* Any non-spare device that is not a replacement can
2715 * become want_replacement at any time, but we then need to
2716 * check if recovery is needed.
2718 if (rdev
->raid_disk
>= 0 &&
2719 !test_bit(Replacement
, &rdev
->flags
))
2720 set_bit(WantReplacement
, &rdev
->flags
);
2721 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2722 md_wakeup_thread(rdev
->mddev
->thread
);
2724 } else if (cmd_match(buf
, "-want_replacement")) {
2725 /* Clearing 'want_replacement' is always allowed.
2726 * Once replacements starts it is too late though.
2729 clear_bit(WantReplacement
, &rdev
->flags
);
2730 } else if (cmd_match(buf
, "replacement")) {
2731 /* Can only set a device as a replacement when array has not
2732 * yet been started. Once running, replacement is automatic
2733 * from spares, or by assigning 'slot'.
2735 if (rdev
->mddev
->pers
)
2738 set_bit(Replacement
, &rdev
->flags
);
2741 } else if (cmd_match(buf
, "-replacement")) {
2742 /* Similarly, can only clear Replacement before start */
2743 if (rdev
->mddev
->pers
)
2746 clear_bit(Replacement
, &rdev
->flags
);
2751 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2752 return err
? err
: len
;
2754 static struct rdev_sysfs_entry rdev_state
=
2755 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2758 errors_show(struct md_rdev
*rdev
, char *page
)
2760 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2764 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2767 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2768 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2769 atomic_set(&rdev
->corrected_errors
, n
);
2774 static struct rdev_sysfs_entry rdev_errors
=
2775 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2778 slot_show(struct md_rdev
*rdev
, char *page
)
2780 if (rdev
->raid_disk
< 0)
2781 return sprintf(page
, "none\n");
2783 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2787 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2791 int slot
= simple_strtoul(buf
, &e
, 10);
2792 if (strncmp(buf
, "none", 4)==0)
2794 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2796 if (rdev
->mddev
->pers
&& slot
== -1) {
2797 /* Setting 'slot' on an active array requires also
2798 * updating the 'rd%d' link, and communicating
2799 * with the personality with ->hot_*_disk.
2800 * For now we only support removing
2801 * failed/spare devices. This normally happens automatically,
2802 * but not when the metadata is externally managed.
2804 if (rdev
->raid_disk
== -1)
2806 /* personality does all needed checks */
2807 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2809 err
= rdev
->mddev
->pers
->
2810 hot_remove_disk(rdev
->mddev
, rdev
);
2813 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2814 rdev
->raid_disk
= -1;
2815 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2816 md_wakeup_thread(rdev
->mddev
->thread
);
2817 } else if (rdev
->mddev
->pers
) {
2818 /* Activating a spare .. or possibly reactivating
2819 * if we ever get bitmaps working here.
2822 if (rdev
->raid_disk
!= -1)
2825 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2828 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2831 if (slot
>= rdev
->mddev
->raid_disks
&&
2832 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2835 rdev
->raid_disk
= slot
;
2836 if (test_bit(In_sync
, &rdev
->flags
))
2837 rdev
->saved_raid_disk
= slot
;
2839 rdev
->saved_raid_disk
= -1;
2840 clear_bit(In_sync
, &rdev
->flags
);
2841 err
= rdev
->mddev
->pers
->
2842 hot_add_disk(rdev
->mddev
, rdev
);
2844 rdev
->raid_disk
= -1;
2847 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2848 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2849 /* failure here is OK */;
2850 /* don't wakeup anyone, leave that to userspace. */
2852 if (slot
>= rdev
->mddev
->raid_disks
&&
2853 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2855 rdev
->raid_disk
= slot
;
2856 /* assume it is working */
2857 clear_bit(Faulty
, &rdev
->flags
);
2858 clear_bit(WriteMostly
, &rdev
->flags
);
2859 set_bit(In_sync
, &rdev
->flags
);
2860 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2866 static struct rdev_sysfs_entry rdev_slot
=
2867 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2870 offset_show(struct md_rdev
*rdev
, char *page
)
2872 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2876 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2878 unsigned long long offset
;
2879 if (strict_strtoull(buf
, 10, &offset
) < 0)
2881 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2883 if (rdev
->sectors
&& rdev
->mddev
->external
)
2884 /* Must set offset before size, so overlap checks
2887 rdev
->data_offset
= offset
;
2888 rdev
->new_data_offset
= offset
;
2892 static struct rdev_sysfs_entry rdev_offset
=
2893 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2895 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2897 return sprintf(page
, "%llu\n",
2898 (unsigned long long)rdev
->new_data_offset
);
2901 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2902 const char *buf
, size_t len
)
2904 unsigned long long new_offset
;
2905 struct mddev
*mddev
= rdev
->mddev
;
2907 if (strict_strtoull(buf
, 10, &new_offset
) < 0)
2910 if (mddev
->sync_thread
)
2912 if (new_offset
== rdev
->data_offset
)
2913 /* reset is always permitted */
2915 else if (new_offset
> rdev
->data_offset
) {
2916 /* must not push array size beyond rdev_sectors */
2917 if (new_offset
- rdev
->data_offset
2918 + mddev
->dev_sectors
> rdev
->sectors
)
2921 /* Metadata worries about other space details. */
2923 /* decreasing the offset is inconsistent with a backwards
2926 if (new_offset
< rdev
->data_offset
&&
2927 mddev
->reshape_backwards
)
2929 /* Increasing offset is inconsistent with forwards
2930 * reshape. reshape_direction should be set to
2931 * 'backwards' first.
2933 if (new_offset
> rdev
->data_offset
&&
2934 !mddev
->reshape_backwards
)
2937 if (mddev
->pers
&& mddev
->persistent
&&
2938 !super_types
[mddev
->major_version
]
2939 .allow_new_offset(rdev
, new_offset
))
2941 rdev
->new_data_offset
= new_offset
;
2942 if (new_offset
> rdev
->data_offset
)
2943 mddev
->reshape_backwards
= 1;
2944 else if (new_offset
< rdev
->data_offset
)
2945 mddev
->reshape_backwards
= 0;
2949 static struct rdev_sysfs_entry rdev_new_offset
=
2950 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2953 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2955 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2958 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2960 /* check if two start/length pairs overlap */
2968 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2970 unsigned long long blocks
;
2973 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2976 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2977 return -EINVAL
; /* sector conversion overflow */
2980 if (new != blocks
* 2)
2981 return -EINVAL
; /* unsigned long long to sector_t overflow */
2988 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2990 struct mddev
*my_mddev
= rdev
->mddev
;
2991 sector_t oldsectors
= rdev
->sectors
;
2994 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2996 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2997 return -EINVAL
; /* too confusing */
2998 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2999 if (my_mddev
->persistent
) {
3000 sectors
= super_types
[my_mddev
->major_version
].
3001 rdev_size_change(rdev
, sectors
);
3004 } else if (!sectors
)
3005 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
3008 if (sectors
< my_mddev
->dev_sectors
)
3009 return -EINVAL
; /* component must fit device */
3011 rdev
->sectors
= sectors
;
3012 if (sectors
> oldsectors
&& my_mddev
->external
) {
3013 /* need to check that all other rdevs with the same ->bdev
3014 * do not overlap. We need to unlock the mddev to avoid
3015 * a deadlock. We have already changed rdev->sectors, and if
3016 * we have to change it back, we will have the lock again.
3018 struct mddev
*mddev
;
3020 struct list_head
*tmp
;
3022 mddev_unlock(my_mddev
);
3023 for_each_mddev(mddev
, tmp
) {
3024 struct md_rdev
*rdev2
;
3027 rdev_for_each(rdev2
, mddev
)
3028 if (rdev
->bdev
== rdev2
->bdev
&&
3030 overlaps(rdev
->data_offset
, rdev
->sectors
,
3036 mddev_unlock(mddev
);
3042 mddev_lock(my_mddev
);
3044 /* Someone else could have slipped in a size
3045 * change here, but doing so is just silly.
3046 * We put oldsectors back because we *know* it is
3047 * safe, and trust userspace not to race with
3050 rdev
->sectors
= oldsectors
;
3057 static struct rdev_sysfs_entry rdev_size
=
3058 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3061 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3063 unsigned long long recovery_start
= rdev
->recovery_offset
;
3065 if (test_bit(In_sync
, &rdev
->flags
) ||
3066 recovery_start
== MaxSector
)
3067 return sprintf(page
, "none\n");
3069 return sprintf(page
, "%llu\n", recovery_start
);
3072 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3074 unsigned long long recovery_start
;
3076 if (cmd_match(buf
, "none"))
3077 recovery_start
= MaxSector
;
3078 else if (strict_strtoull(buf
, 10, &recovery_start
))
3081 if (rdev
->mddev
->pers
&&
3082 rdev
->raid_disk
>= 0)
3085 rdev
->recovery_offset
= recovery_start
;
3086 if (recovery_start
== MaxSector
)
3087 set_bit(In_sync
, &rdev
->flags
);
3089 clear_bit(In_sync
, &rdev
->flags
);
3093 static struct rdev_sysfs_entry rdev_recovery_start
=
3094 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3098 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
3100 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
3102 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3104 return badblocks_show(&rdev
->badblocks
, page
, 0);
3106 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3108 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3109 /* Maybe that ack was all we needed */
3110 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3111 wake_up(&rdev
->blocked_wait
);
3114 static struct rdev_sysfs_entry rdev_bad_blocks
=
3115 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3118 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3120 return badblocks_show(&rdev
->badblocks
, page
, 1);
3122 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3124 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3126 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3127 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3129 static struct attribute
*rdev_default_attrs
[] = {
3134 &rdev_new_offset
.attr
,
3136 &rdev_recovery_start
.attr
,
3137 &rdev_bad_blocks
.attr
,
3138 &rdev_unack_bad_blocks
.attr
,
3142 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3144 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3145 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3146 struct mddev
*mddev
= rdev
->mddev
;
3152 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
3154 if (rdev
->mddev
== NULL
)
3157 rv
= entry
->show(rdev
, page
);
3158 mddev_unlock(mddev
);
3164 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3165 const char *page
, size_t length
)
3167 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3168 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3170 struct mddev
*mddev
= rdev
->mddev
;
3174 if (!capable(CAP_SYS_ADMIN
))
3176 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3178 if (rdev
->mddev
== NULL
)
3181 rv
= entry
->store(rdev
, page
, length
);
3182 mddev_unlock(mddev
);
3187 static void rdev_free(struct kobject
*ko
)
3189 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3192 static const struct sysfs_ops rdev_sysfs_ops
= {
3193 .show
= rdev_attr_show
,
3194 .store
= rdev_attr_store
,
3196 static struct kobj_type rdev_ktype
= {
3197 .release
= rdev_free
,
3198 .sysfs_ops
= &rdev_sysfs_ops
,
3199 .default_attrs
= rdev_default_attrs
,
3202 int md_rdev_init(struct md_rdev
*rdev
)
3205 rdev
->saved_raid_disk
= -1;
3206 rdev
->raid_disk
= -1;
3208 rdev
->data_offset
= 0;
3209 rdev
->new_data_offset
= 0;
3210 rdev
->sb_events
= 0;
3211 rdev
->last_read_error
.tv_sec
= 0;
3212 rdev
->last_read_error
.tv_nsec
= 0;
3213 rdev
->sb_loaded
= 0;
3214 rdev
->bb_page
= NULL
;
3215 atomic_set(&rdev
->nr_pending
, 0);
3216 atomic_set(&rdev
->read_errors
, 0);
3217 atomic_set(&rdev
->corrected_errors
, 0);
3219 INIT_LIST_HEAD(&rdev
->same_set
);
3220 init_waitqueue_head(&rdev
->blocked_wait
);
3222 /* Add space to store bad block list.
3223 * This reserves the space even on arrays where it cannot
3224 * be used - I wonder if that matters
3226 rdev
->badblocks
.count
= 0;
3227 rdev
->badblocks
.shift
= 0;
3228 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3229 seqlock_init(&rdev
->badblocks
.lock
);
3230 if (rdev
->badblocks
.page
== NULL
)
3235 EXPORT_SYMBOL_GPL(md_rdev_init
);
3237 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3239 * mark the device faulty if:
3241 * - the device is nonexistent (zero size)
3242 * - the device has no valid superblock
3244 * a faulty rdev _never_ has rdev->sb set.
3246 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3248 char b
[BDEVNAME_SIZE
];
3250 struct md_rdev
*rdev
;
3253 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3255 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3256 return ERR_PTR(-ENOMEM
);
3259 err
= md_rdev_init(rdev
);
3262 err
= alloc_disk_sb(rdev
);
3266 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3270 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3272 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3275 "md: %s has zero or unknown size, marking faulty!\n",
3276 bdevname(rdev
->bdev
,b
));
3281 if (super_format
>= 0) {
3282 err
= super_types
[super_format
].
3283 load_super(rdev
, NULL
, super_minor
);
3284 if (err
== -EINVAL
) {
3286 "md: %s does not have a valid v%d.%d "
3287 "superblock, not importing!\n",
3288 bdevname(rdev
->bdev
,b
),
3289 super_format
, super_minor
);
3294 "md: could not read %s's sb, not importing!\n",
3295 bdevname(rdev
->bdev
,b
));
3299 if (super_format
== -1)
3300 /* hot-add for 0.90, or non-persistent: so no badblocks */
3301 rdev
->badblocks
.shift
= -1;
3308 md_rdev_clear(rdev
);
3310 return ERR_PTR(err
);
3314 * Check a full RAID array for plausibility
3318 static void analyze_sbs(struct mddev
* mddev
)
3321 struct md_rdev
*rdev
, *freshest
, *tmp
;
3322 char b
[BDEVNAME_SIZE
];
3325 rdev_for_each_safe(rdev
, tmp
, mddev
)
3326 switch (super_types
[mddev
->major_version
].
3327 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3335 "md: fatal superblock inconsistency in %s"
3336 " -- removing from array\n",
3337 bdevname(rdev
->bdev
,b
));
3338 kick_rdev_from_array(rdev
);
3342 super_types
[mddev
->major_version
].
3343 validate_super(mddev
, freshest
);
3346 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3347 if (mddev
->max_disks
&&
3348 (rdev
->desc_nr
>= mddev
->max_disks
||
3349 i
> mddev
->max_disks
)) {
3351 "md: %s: %s: only %d devices permitted\n",
3352 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3354 kick_rdev_from_array(rdev
);
3357 if (rdev
!= freshest
)
3358 if (super_types
[mddev
->major_version
].
3359 validate_super(mddev
, rdev
)) {
3360 printk(KERN_WARNING
"md: kicking non-fresh %s"
3362 bdevname(rdev
->bdev
,b
));
3363 kick_rdev_from_array(rdev
);
3366 if (mddev
->level
== LEVEL_MULTIPATH
) {
3367 rdev
->desc_nr
= i
++;
3368 rdev
->raid_disk
= rdev
->desc_nr
;
3369 set_bit(In_sync
, &rdev
->flags
);
3370 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3371 rdev
->raid_disk
= -1;
3372 clear_bit(In_sync
, &rdev
->flags
);
3377 /* Read a fixed-point number.
3378 * Numbers in sysfs attributes should be in "standard" units where
3379 * possible, so time should be in seconds.
3380 * However we internally use a a much smaller unit such as
3381 * milliseconds or jiffies.
3382 * This function takes a decimal number with a possible fractional
3383 * component, and produces an integer which is the result of
3384 * multiplying that number by 10^'scale'.
3385 * all without any floating-point arithmetic.
3387 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3389 unsigned long result
= 0;
3391 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3394 else if (decimals
< scale
) {
3397 result
= result
* 10 + value
;
3409 while (decimals
< scale
) {
3418 static void md_safemode_timeout(unsigned long data
);
3421 safe_delay_show(struct mddev
*mddev
, char *page
)
3423 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3424 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3427 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3431 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3434 mddev
->safemode_delay
= 0;
3436 unsigned long old_delay
= mddev
->safemode_delay
;
3437 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3438 if (mddev
->safemode_delay
== 0)
3439 mddev
->safemode_delay
= 1;
3440 if (mddev
->safemode_delay
< old_delay
)
3441 md_safemode_timeout((unsigned long)mddev
);
3445 static struct md_sysfs_entry md_safe_delay
=
3446 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3449 level_show(struct mddev
*mddev
, char *page
)
3451 struct md_personality
*p
= mddev
->pers
;
3453 return sprintf(page
, "%s\n", p
->name
);
3454 else if (mddev
->clevel
[0])
3455 return sprintf(page
, "%s\n", mddev
->clevel
);
3456 else if (mddev
->level
!= LEVEL_NONE
)
3457 return sprintf(page
, "%d\n", mddev
->level
);
3463 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3467 struct md_personality
*pers
;
3470 struct md_rdev
*rdev
;
3472 if (mddev
->pers
== NULL
) {
3475 if (len
>= sizeof(mddev
->clevel
))
3477 strncpy(mddev
->clevel
, buf
, len
);
3478 if (mddev
->clevel
[len
-1] == '\n')
3480 mddev
->clevel
[len
] = 0;
3481 mddev
->level
= LEVEL_NONE
;
3485 /* request to change the personality. Need to ensure:
3486 * - array is not engaged in resync/recovery/reshape
3487 * - old personality can be suspended
3488 * - new personality will access other array.
3491 if (mddev
->sync_thread
||
3492 mddev
->reshape_position
!= MaxSector
||
3493 mddev
->sysfs_active
)
3496 if (!mddev
->pers
->quiesce
) {
3497 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3498 mdname(mddev
), mddev
->pers
->name
);
3502 /* Now find the new personality */
3503 if (len
== 0 || len
>= sizeof(clevel
))
3505 strncpy(clevel
, buf
, len
);
3506 if (clevel
[len
-1] == '\n')
3509 if (strict_strtol(clevel
, 10, &level
))
3512 if (request_module("md-%s", clevel
) != 0)
3513 request_module("md-level-%s", clevel
);
3514 spin_lock(&pers_lock
);
3515 pers
= find_pers(level
, clevel
);
3516 if (!pers
|| !try_module_get(pers
->owner
)) {
3517 spin_unlock(&pers_lock
);
3518 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3521 spin_unlock(&pers_lock
);
3523 if (pers
== mddev
->pers
) {
3524 /* Nothing to do! */
3525 module_put(pers
->owner
);
3528 if (!pers
->takeover
) {
3529 module_put(pers
->owner
);
3530 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3531 mdname(mddev
), clevel
);
3535 rdev_for_each(rdev
, mddev
)
3536 rdev
->new_raid_disk
= rdev
->raid_disk
;
3538 /* ->takeover must set new_* and/or delta_disks
3539 * if it succeeds, and may set them when it fails.
3541 priv
= pers
->takeover(mddev
);
3543 mddev
->new_level
= mddev
->level
;
3544 mddev
->new_layout
= mddev
->layout
;
3545 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3546 mddev
->raid_disks
-= mddev
->delta_disks
;
3547 mddev
->delta_disks
= 0;
3548 mddev
->reshape_backwards
= 0;
3549 module_put(pers
->owner
);
3550 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3551 mdname(mddev
), clevel
);
3552 return PTR_ERR(priv
);
3555 /* Looks like we have a winner */
3556 mddev_suspend(mddev
);
3557 mddev
->pers
->stop(mddev
);
3559 if (mddev
->pers
->sync_request
== NULL
&&
3560 pers
->sync_request
!= NULL
) {
3561 /* need to add the md_redundancy_group */
3562 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3564 "md: cannot register extra attributes for %s\n",
3566 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3568 if (mddev
->pers
->sync_request
!= NULL
&&
3569 pers
->sync_request
== NULL
) {
3570 /* need to remove the md_redundancy_group */
3571 if (mddev
->to_remove
== NULL
)
3572 mddev
->to_remove
= &md_redundancy_group
;
3575 if (mddev
->pers
->sync_request
== NULL
&&
3577 /* We are converting from a no-redundancy array
3578 * to a redundancy array and metadata is managed
3579 * externally so we need to be sure that writes
3580 * won't block due to a need to transition
3582 * until external management is started.
3585 mddev
->safemode_delay
= 0;
3586 mddev
->safemode
= 0;
3589 rdev_for_each(rdev
, mddev
) {
3590 if (rdev
->raid_disk
< 0)
3592 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3593 rdev
->new_raid_disk
= -1;
3594 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3596 sysfs_unlink_rdev(mddev
, rdev
);
3598 rdev_for_each(rdev
, mddev
) {
3599 if (rdev
->raid_disk
< 0)
3601 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3603 rdev
->raid_disk
= rdev
->new_raid_disk
;
3604 if (rdev
->raid_disk
< 0)
3605 clear_bit(In_sync
, &rdev
->flags
);
3607 if (sysfs_link_rdev(mddev
, rdev
))
3608 printk(KERN_WARNING
"md: cannot register rd%d"
3609 " for %s after level change\n",
3610 rdev
->raid_disk
, mdname(mddev
));
3614 module_put(mddev
->pers
->owner
);
3616 mddev
->private = priv
;
3617 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3618 mddev
->level
= mddev
->new_level
;
3619 mddev
->layout
= mddev
->new_layout
;
3620 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3621 mddev
->delta_disks
= 0;
3622 mddev
->reshape_backwards
= 0;
3623 mddev
->degraded
= 0;
3624 if (mddev
->pers
->sync_request
== NULL
) {
3625 /* this is now an array without redundancy, so
3626 * it must always be in_sync
3629 del_timer_sync(&mddev
->safemode_timer
);
3632 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3633 mddev_resume(mddev
);
3634 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3635 md_new_event(mddev
);
3639 static struct md_sysfs_entry md_level
=
3640 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3644 layout_show(struct mddev
*mddev
, char *page
)
3646 /* just a number, not meaningful for all levels */
3647 if (mddev
->reshape_position
!= MaxSector
&&
3648 mddev
->layout
!= mddev
->new_layout
)
3649 return sprintf(page
, "%d (%d)\n",
3650 mddev
->new_layout
, mddev
->layout
);
3651 return sprintf(page
, "%d\n", mddev
->layout
);
3655 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3658 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3660 if (!*buf
|| (*e
&& *e
!= '\n'))
3665 if (mddev
->pers
->check_reshape
== NULL
)
3667 mddev
->new_layout
= n
;
3668 err
= mddev
->pers
->check_reshape(mddev
);
3670 mddev
->new_layout
= mddev
->layout
;
3674 mddev
->new_layout
= n
;
3675 if (mddev
->reshape_position
== MaxSector
)
3680 static struct md_sysfs_entry md_layout
=
3681 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3685 raid_disks_show(struct mddev
*mddev
, char *page
)
3687 if (mddev
->raid_disks
== 0)
3689 if (mddev
->reshape_position
!= MaxSector
&&
3690 mddev
->delta_disks
!= 0)
3691 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3692 mddev
->raid_disks
- mddev
->delta_disks
);
3693 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3696 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3699 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3703 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3705 if (!*buf
|| (*e
&& *e
!= '\n'))
3709 rv
= update_raid_disks(mddev
, n
);
3710 else if (mddev
->reshape_position
!= MaxSector
) {
3711 struct md_rdev
*rdev
;
3712 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3714 rdev_for_each(rdev
, mddev
) {
3716 rdev
->data_offset
< rdev
->new_data_offset
)
3719 rdev
->data_offset
> rdev
->new_data_offset
)
3722 mddev
->delta_disks
= n
- olddisks
;
3723 mddev
->raid_disks
= n
;
3724 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3726 mddev
->raid_disks
= n
;
3727 return rv
? rv
: len
;
3729 static struct md_sysfs_entry md_raid_disks
=
3730 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3733 chunk_size_show(struct mddev
*mddev
, char *page
)
3735 if (mddev
->reshape_position
!= MaxSector
&&
3736 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3737 return sprintf(page
, "%d (%d)\n",
3738 mddev
->new_chunk_sectors
<< 9,
3739 mddev
->chunk_sectors
<< 9);
3740 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3744 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3747 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3749 if (!*buf
|| (*e
&& *e
!= '\n'))
3754 if (mddev
->pers
->check_reshape
== NULL
)
3756 mddev
->new_chunk_sectors
= n
>> 9;
3757 err
= mddev
->pers
->check_reshape(mddev
);
3759 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3763 mddev
->new_chunk_sectors
= n
>> 9;
3764 if (mddev
->reshape_position
== MaxSector
)
3765 mddev
->chunk_sectors
= n
>> 9;
3769 static struct md_sysfs_entry md_chunk_size
=
3770 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3773 resync_start_show(struct mddev
*mddev
, char *page
)
3775 if (mddev
->recovery_cp
== MaxSector
)
3776 return sprintf(page
, "none\n");
3777 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3781 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3784 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3786 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3788 if (cmd_match(buf
, "none"))
3790 else if (!*buf
|| (*e
&& *e
!= '\n'))
3793 mddev
->recovery_cp
= n
;
3796 static struct md_sysfs_entry md_resync_start
=
3797 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3800 * The array state can be:
3803 * No devices, no size, no level
3804 * Equivalent to STOP_ARRAY ioctl
3806 * May have some settings, but array is not active
3807 * all IO results in error
3808 * When written, doesn't tear down array, but just stops it
3809 * suspended (not supported yet)
3810 * All IO requests will block. The array can be reconfigured.
3811 * Writing this, if accepted, will block until array is quiescent
3813 * no resync can happen. no superblocks get written.
3814 * write requests fail
3816 * like readonly, but behaves like 'clean' on a write request.
3818 * clean - no pending writes, but otherwise active.
3819 * When written to inactive array, starts without resync
3820 * If a write request arrives then
3821 * if metadata is known, mark 'dirty' and switch to 'active'.
3822 * if not known, block and switch to write-pending
3823 * If written to an active array that has pending writes, then fails.
3825 * fully active: IO and resync can be happening.
3826 * When written to inactive array, starts with resync
3829 * clean, but writes are blocked waiting for 'active' to be written.
3832 * like active, but no writes have been seen for a while (100msec).
3835 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3836 write_pending
, active_idle
, bad_word
};
3837 static char *array_states
[] = {
3838 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3839 "write-pending", "active-idle", NULL
};
3841 static int match_word(const char *word
, char **list
)
3844 for (n
=0; list
[n
]; n
++)
3845 if (cmd_match(word
, list
[n
]))
3851 array_state_show(struct mddev
*mddev
, char *page
)
3853 enum array_state st
= inactive
;
3866 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3868 else if (mddev
->safemode
)
3874 if (list_empty(&mddev
->disks
) &&
3875 mddev
->raid_disks
== 0 &&
3876 mddev
->dev_sectors
== 0)
3881 return sprintf(page
, "%s\n", array_states
[st
]);
3884 static int do_md_stop(struct mddev
* mddev
, int ro
, struct block_device
*bdev
);
3885 static int md_set_readonly(struct mddev
* mddev
, struct block_device
*bdev
);
3886 static int do_md_run(struct mddev
* mddev
);
3887 static int restart_array(struct mddev
*mddev
);
3890 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3893 enum array_state st
= match_word(buf
, array_states
);
3898 /* stopping an active array */
3899 err
= do_md_stop(mddev
, 0, NULL
);
3902 /* stopping an active array */
3904 err
= do_md_stop(mddev
, 2, NULL
);
3906 err
= 0; /* already inactive */
3909 break; /* not supported yet */
3912 err
= md_set_readonly(mddev
, NULL
);
3915 set_disk_ro(mddev
->gendisk
, 1);
3916 err
= do_md_run(mddev
);
3922 err
= md_set_readonly(mddev
, NULL
);
3923 else if (mddev
->ro
== 1)
3924 err
= restart_array(mddev
);
3927 set_disk_ro(mddev
->gendisk
, 0);
3931 err
= do_md_run(mddev
);
3936 restart_array(mddev
);
3937 spin_lock_irq(&mddev
->write_lock
);
3938 if (atomic_read(&mddev
->writes_pending
) == 0) {
3939 if (mddev
->in_sync
== 0) {
3941 if (mddev
->safemode
== 1)
3942 mddev
->safemode
= 0;
3943 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3948 spin_unlock_irq(&mddev
->write_lock
);
3954 restart_array(mddev
);
3955 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3956 wake_up(&mddev
->sb_wait
);
3960 set_disk_ro(mddev
->gendisk
, 0);
3961 err
= do_md_run(mddev
);
3966 /* these cannot be set */
3972 if (mddev
->hold_active
== UNTIL_IOCTL
)
3973 mddev
->hold_active
= 0;
3974 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3978 static struct md_sysfs_entry md_array_state
=
3979 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3982 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3983 return sprintf(page
, "%d\n",
3984 atomic_read(&mddev
->max_corr_read_errors
));
3988 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3991 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3993 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3994 atomic_set(&mddev
->max_corr_read_errors
, n
);
4000 static struct md_sysfs_entry max_corr_read_errors
=
4001 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4002 max_corrected_read_errors_store
);
4005 null_show(struct mddev
*mddev
, char *page
)
4011 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4013 /* buf must be %d:%d\n? giving major and minor numbers */
4014 /* The new device is added to the array.
4015 * If the array has a persistent superblock, we read the
4016 * superblock to initialise info and check validity.
4017 * Otherwise, only checking done is that in bind_rdev_to_array,
4018 * which mainly checks size.
4021 int major
= simple_strtoul(buf
, &e
, 10);
4024 struct md_rdev
*rdev
;
4027 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4029 minor
= simple_strtoul(e
+1, &e
, 10);
4030 if (*e
&& *e
!= '\n')
4032 dev
= MKDEV(major
, minor
);
4033 if (major
!= MAJOR(dev
) ||
4034 minor
!= MINOR(dev
))
4038 if (mddev
->persistent
) {
4039 rdev
= md_import_device(dev
, mddev
->major_version
,
4040 mddev
->minor_version
);
4041 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4042 struct md_rdev
*rdev0
4043 = list_entry(mddev
->disks
.next
,
4044 struct md_rdev
, same_set
);
4045 err
= super_types
[mddev
->major_version
]
4046 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4050 } else if (mddev
->external
)
4051 rdev
= md_import_device(dev
, -2, -1);
4053 rdev
= md_import_device(dev
, -1, -1);
4056 return PTR_ERR(rdev
);
4057 err
= bind_rdev_to_array(rdev
, mddev
);
4061 return err
? err
: len
;
4064 static struct md_sysfs_entry md_new_device
=
4065 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4068 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4071 unsigned long chunk
, end_chunk
;
4075 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4077 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4078 if (buf
== end
) break;
4079 if (*end
== '-') { /* range */
4081 end_chunk
= simple_strtoul(buf
, &end
, 0);
4082 if (buf
== end
) break;
4084 if (*end
&& !isspace(*end
)) break;
4085 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4086 buf
= skip_spaces(end
);
4088 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4093 static struct md_sysfs_entry md_bitmap
=
4094 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4097 size_show(struct mddev
*mddev
, char *page
)
4099 return sprintf(page
, "%llu\n",
4100 (unsigned long long)mddev
->dev_sectors
/ 2);
4103 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4106 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4108 /* If array is inactive, we can reduce the component size, but
4109 * not increase it (except from 0).
4110 * If array is active, we can try an on-line resize
4113 int err
= strict_blocks_to_sectors(buf
, §ors
);
4118 err
= update_size(mddev
, sectors
);
4119 md_update_sb(mddev
, 1);
4121 if (mddev
->dev_sectors
== 0 ||
4122 mddev
->dev_sectors
> sectors
)
4123 mddev
->dev_sectors
= sectors
;
4127 return err
? err
: len
;
4130 static struct md_sysfs_entry md_size
=
4131 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4136 * 'none' for arrays with no metadata (good luck...)
4137 * 'external' for arrays with externally managed metadata,
4138 * or N.M for internally known formats
4141 metadata_show(struct mddev
*mddev
, char *page
)
4143 if (mddev
->persistent
)
4144 return sprintf(page
, "%d.%d\n",
4145 mddev
->major_version
, mddev
->minor_version
);
4146 else if (mddev
->external
)
4147 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4149 return sprintf(page
, "none\n");
4153 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4157 /* Changing the details of 'external' metadata is
4158 * always permitted. Otherwise there must be
4159 * no devices attached to the array.
4161 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4163 else if (!list_empty(&mddev
->disks
))
4166 if (cmd_match(buf
, "none")) {
4167 mddev
->persistent
= 0;
4168 mddev
->external
= 0;
4169 mddev
->major_version
= 0;
4170 mddev
->minor_version
= 90;
4173 if (strncmp(buf
, "external:", 9) == 0) {
4174 size_t namelen
= len
-9;
4175 if (namelen
>= sizeof(mddev
->metadata_type
))
4176 namelen
= sizeof(mddev
->metadata_type
)-1;
4177 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4178 mddev
->metadata_type
[namelen
] = 0;
4179 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4180 mddev
->metadata_type
[--namelen
] = 0;
4181 mddev
->persistent
= 0;
4182 mddev
->external
= 1;
4183 mddev
->major_version
= 0;
4184 mddev
->minor_version
= 90;
4187 major
= simple_strtoul(buf
, &e
, 10);
4188 if (e
==buf
|| *e
!= '.')
4191 minor
= simple_strtoul(buf
, &e
, 10);
4192 if (e
==buf
|| (*e
&& *e
!= '\n') )
4194 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4196 mddev
->major_version
= major
;
4197 mddev
->minor_version
= minor
;
4198 mddev
->persistent
= 1;
4199 mddev
->external
= 0;
4203 static struct md_sysfs_entry md_metadata
=
4204 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4207 action_show(struct mddev
*mddev
, char *page
)
4209 char *type
= "idle";
4210 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4212 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4213 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4214 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4216 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4217 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4219 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4223 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4226 return sprintf(page
, "%s\n", type
);
4229 static void reap_sync_thread(struct mddev
*mddev
);
4232 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4234 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4237 if (cmd_match(page
, "frozen"))
4238 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4240 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4242 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4243 if (mddev
->sync_thread
) {
4244 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4245 reap_sync_thread(mddev
);
4247 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4248 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4250 else if (cmd_match(page
, "resync"))
4251 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4252 else if (cmd_match(page
, "recover")) {
4253 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4254 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4255 } else if (cmd_match(page
, "reshape")) {
4257 if (mddev
->pers
->start_reshape
== NULL
)
4259 err
= mddev
->pers
->start_reshape(mddev
);
4262 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4264 if (cmd_match(page
, "check"))
4265 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4266 else if (!cmd_match(page
, "repair"))
4268 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4269 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4271 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4272 md_wakeup_thread(mddev
->thread
);
4273 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4278 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4280 return sprintf(page
, "%llu\n",
4281 (unsigned long long) mddev
->resync_mismatches
);
4284 static struct md_sysfs_entry md_scan_mode
=
4285 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4288 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4291 sync_min_show(struct mddev
*mddev
, char *page
)
4293 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4294 mddev
->sync_speed_min
? "local": "system");
4298 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4302 if (strncmp(buf
, "system", 6)==0) {
4303 mddev
->sync_speed_min
= 0;
4306 min
= simple_strtoul(buf
, &e
, 10);
4307 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4309 mddev
->sync_speed_min
= min
;
4313 static struct md_sysfs_entry md_sync_min
=
4314 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4317 sync_max_show(struct mddev
*mddev
, char *page
)
4319 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4320 mddev
->sync_speed_max
? "local": "system");
4324 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4328 if (strncmp(buf
, "system", 6)==0) {
4329 mddev
->sync_speed_max
= 0;
4332 max
= simple_strtoul(buf
, &e
, 10);
4333 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4335 mddev
->sync_speed_max
= max
;
4339 static struct md_sysfs_entry md_sync_max
=
4340 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4343 degraded_show(struct mddev
*mddev
, char *page
)
4345 return sprintf(page
, "%d\n", mddev
->degraded
);
4347 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4350 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4352 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4356 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4360 if (strict_strtol(buf
, 10, &n
))
4363 if (n
!= 0 && n
!= 1)
4366 mddev
->parallel_resync
= n
;
4368 if (mddev
->sync_thread
)
4369 wake_up(&resync_wait
);
4374 /* force parallel resync, even with shared block devices */
4375 static struct md_sysfs_entry md_sync_force_parallel
=
4376 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4377 sync_force_parallel_show
, sync_force_parallel_store
);
4380 sync_speed_show(struct mddev
*mddev
, char *page
)
4382 unsigned long resync
, dt
, db
;
4383 if (mddev
->curr_resync
== 0)
4384 return sprintf(page
, "none\n");
4385 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4386 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4388 db
= resync
- mddev
->resync_mark_cnt
;
4389 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4392 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4395 sync_completed_show(struct mddev
*mddev
, char *page
)
4397 unsigned long long max_sectors
, resync
;
4399 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4400 return sprintf(page
, "none\n");
4402 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4403 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4404 max_sectors
= mddev
->resync_max_sectors
;
4406 max_sectors
= mddev
->dev_sectors
;
4408 resync
= mddev
->curr_resync_completed
;
4409 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4412 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4415 min_sync_show(struct mddev
*mddev
, char *page
)
4417 return sprintf(page
, "%llu\n",
4418 (unsigned long long)mddev
->resync_min
);
4421 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4423 unsigned long long min
;
4424 if (strict_strtoull(buf
, 10, &min
))
4426 if (min
> mddev
->resync_max
)
4428 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4431 /* Must be a multiple of chunk_size */
4432 if (mddev
->chunk_sectors
) {
4433 sector_t temp
= min
;
4434 if (sector_div(temp
, mddev
->chunk_sectors
))
4437 mddev
->resync_min
= min
;
4442 static struct md_sysfs_entry md_min_sync
=
4443 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4446 max_sync_show(struct mddev
*mddev
, char *page
)
4448 if (mddev
->resync_max
== MaxSector
)
4449 return sprintf(page
, "max\n");
4451 return sprintf(page
, "%llu\n",
4452 (unsigned long long)mddev
->resync_max
);
4455 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4457 if (strncmp(buf
, "max", 3) == 0)
4458 mddev
->resync_max
= MaxSector
;
4460 unsigned long long max
;
4461 if (strict_strtoull(buf
, 10, &max
))
4463 if (max
< mddev
->resync_min
)
4465 if (max
< mddev
->resync_max
&&
4467 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4470 /* Must be a multiple of chunk_size */
4471 if (mddev
->chunk_sectors
) {
4472 sector_t temp
= max
;
4473 if (sector_div(temp
, mddev
->chunk_sectors
))
4476 mddev
->resync_max
= max
;
4478 wake_up(&mddev
->recovery_wait
);
4482 static struct md_sysfs_entry md_max_sync
=
4483 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4486 suspend_lo_show(struct mddev
*mddev
, char *page
)
4488 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4492 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4495 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4496 unsigned long long old
= mddev
->suspend_lo
;
4498 if (mddev
->pers
== NULL
||
4499 mddev
->pers
->quiesce
== NULL
)
4501 if (buf
== e
|| (*e
&& *e
!= '\n'))
4504 mddev
->suspend_lo
= new;
4506 /* Shrinking suspended region */
4507 mddev
->pers
->quiesce(mddev
, 2);
4509 /* Expanding suspended region - need to wait */
4510 mddev
->pers
->quiesce(mddev
, 1);
4511 mddev
->pers
->quiesce(mddev
, 0);
4515 static struct md_sysfs_entry md_suspend_lo
=
4516 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4520 suspend_hi_show(struct mddev
*mddev
, char *page
)
4522 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4526 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4529 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4530 unsigned long long old
= mddev
->suspend_hi
;
4532 if (mddev
->pers
== NULL
||
4533 mddev
->pers
->quiesce
== NULL
)
4535 if (buf
== e
|| (*e
&& *e
!= '\n'))
4538 mddev
->suspend_hi
= new;
4540 /* Shrinking suspended region */
4541 mddev
->pers
->quiesce(mddev
, 2);
4543 /* Expanding suspended region - need to wait */
4544 mddev
->pers
->quiesce(mddev
, 1);
4545 mddev
->pers
->quiesce(mddev
, 0);
4549 static struct md_sysfs_entry md_suspend_hi
=
4550 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4553 reshape_position_show(struct mddev
*mddev
, char *page
)
4555 if (mddev
->reshape_position
!= MaxSector
)
4556 return sprintf(page
, "%llu\n",
4557 (unsigned long long)mddev
->reshape_position
);
4558 strcpy(page
, "none\n");
4563 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4565 struct md_rdev
*rdev
;
4567 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4570 if (buf
== e
|| (*e
&& *e
!= '\n'))
4572 mddev
->reshape_position
= new;
4573 mddev
->delta_disks
= 0;
4574 mddev
->reshape_backwards
= 0;
4575 mddev
->new_level
= mddev
->level
;
4576 mddev
->new_layout
= mddev
->layout
;
4577 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4578 rdev_for_each(rdev
, mddev
)
4579 rdev
->new_data_offset
= rdev
->data_offset
;
4583 static struct md_sysfs_entry md_reshape_position
=
4584 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4585 reshape_position_store
);
4588 reshape_direction_show(struct mddev
*mddev
, char *page
)
4590 return sprintf(page
, "%s\n",
4591 mddev
->reshape_backwards
? "backwards" : "forwards");
4595 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4598 if (cmd_match(buf
, "forwards"))
4600 else if (cmd_match(buf
, "backwards"))
4604 if (mddev
->reshape_backwards
== backwards
)
4607 /* check if we are allowed to change */
4608 if (mddev
->delta_disks
)
4611 if (mddev
->persistent
&&
4612 mddev
->major_version
== 0)
4615 mddev
->reshape_backwards
= backwards
;
4619 static struct md_sysfs_entry md_reshape_direction
=
4620 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4621 reshape_direction_store
);
4624 array_size_show(struct mddev
*mddev
, char *page
)
4626 if (mddev
->external_size
)
4627 return sprintf(page
, "%llu\n",
4628 (unsigned long long)mddev
->array_sectors
/2);
4630 return sprintf(page
, "default\n");
4634 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4638 if (strncmp(buf
, "default", 7) == 0) {
4640 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4642 sectors
= mddev
->array_sectors
;
4644 mddev
->external_size
= 0;
4646 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4648 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4651 mddev
->external_size
= 1;
4654 mddev
->array_sectors
= sectors
;
4656 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4657 revalidate_disk(mddev
->gendisk
);
4662 static struct md_sysfs_entry md_array_size
=
4663 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4666 static struct attribute
*md_default_attrs
[] = {
4669 &md_raid_disks
.attr
,
4670 &md_chunk_size
.attr
,
4672 &md_resync_start
.attr
,
4674 &md_new_device
.attr
,
4675 &md_safe_delay
.attr
,
4676 &md_array_state
.attr
,
4677 &md_reshape_position
.attr
,
4678 &md_reshape_direction
.attr
,
4679 &md_array_size
.attr
,
4680 &max_corr_read_errors
.attr
,
4684 static struct attribute
*md_redundancy_attrs
[] = {
4686 &md_mismatches
.attr
,
4689 &md_sync_speed
.attr
,
4690 &md_sync_force_parallel
.attr
,
4691 &md_sync_completed
.attr
,
4694 &md_suspend_lo
.attr
,
4695 &md_suspend_hi
.attr
,
4700 static struct attribute_group md_redundancy_group
= {
4702 .attrs
= md_redundancy_attrs
,
4707 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4709 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4710 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4715 spin_lock(&all_mddevs_lock
);
4716 if (list_empty(&mddev
->all_mddevs
)) {
4717 spin_unlock(&all_mddevs_lock
);
4721 spin_unlock(&all_mddevs_lock
);
4723 rv
= mddev_lock(mddev
);
4725 rv
= entry
->show(mddev
, page
);
4726 mddev_unlock(mddev
);
4733 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4734 const char *page
, size_t length
)
4736 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4737 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4742 if (!capable(CAP_SYS_ADMIN
))
4744 spin_lock(&all_mddevs_lock
);
4745 if (list_empty(&mddev
->all_mddevs
)) {
4746 spin_unlock(&all_mddevs_lock
);
4750 spin_unlock(&all_mddevs_lock
);
4751 rv
= mddev_lock(mddev
);
4753 rv
= entry
->store(mddev
, page
, length
);
4754 mddev_unlock(mddev
);
4760 static void md_free(struct kobject
*ko
)
4762 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4764 if (mddev
->sysfs_state
)
4765 sysfs_put(mddev
->sysfs_state
);
4767 if (mddev
->gendisk
) {
4768 del_gendisk(mddev
->gendisk
);
4769 put_disk(mddev
->gendisk
);
4772 blk_cleanup_queue(mddev
->queue
);
4777 static const struct sysfs_ops md_sysfs_ops
= {
4778 .show
= md_attr_show
,
4779 .store
= md_attr_store
,
4781 static struct kobj_type md_ktype
= {
4783 .sysfs_ops
= &md_sysfs_ops
,
4784 .default_attrs
= md_default_attrs
,
4789 static void mddev_delayed_delete(struct work_struct
*ws
)
4791 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4793 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4794 kobject_del(&mddev
->kobj
);
4795 kobject_put(&mddev
->kobj
);
4798 static int md_alloc(dev_t dev
, char *name
)
4800 static DEFINE_MUTEX(disks_mutex
);
4801 struct mddev
*mddev
= mddev_find(dev
);
4802 struct gendisk
*disk
;
4811 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4812 shift
= partitioned
? MdpMinorShift
: 0;
4813 unit
= MINOR(mddev
->unit
) >> shift
;
4815 /* wait for any previous instance of this device to be
4816 * completely removed (mddev_delayed_delete).
4818 flush_workqueue(md_misc_wq
);
4820 mutex_lock(&disks_mutex
);
4826 /* Need to ensure that 'name' is not a duplicate.
4828 struct mddev
*mddev2
;
4829 spin_lock(&all_mddevs_lock
);
4831 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4832 if (mddev2
->gendisk
&&
4833 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4834 spin_unlock(&all_mddevs_lock
);
4837 spin_unlock(&all_mddevs_lock
);
4841 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4844 mddev
->queue
->queuedata
= mddev
;
4846 blk_queue_make_request(mddev
->queue
, md_make_request
);
4847 blk_set_stacking_limits(&mddev
->queue
->limits
);
4849 disk
= alloc_disk(1 << shift
);
4851 blk_cleanup_queue(mddev
->queue
);
4852 mddev
->queue
= NULL
;
4855 disk
->major
= MAJOR(mddev
->unit
);
4856 disk
->first_minor
= unit
<< shift
;
4858 strcpy(disk
->disk_name
, name
);
4859 else if (partitioned
)
4860 sprintf(disk
->disk_name
, "md_d%d", unit
);
4862 sprintf(disk
->disk_name
, "md%d", unit
);
4863 disk
->fops
= &md_fops
;
4864 disk
->private_data
= mddev
;
4865 disk
->queue
= mddev
->queue
;
4866 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4867 /* Allow extended partitions. This makes the
4868 * 'mdp' device redundant, but we can't really
4871 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4872 mddev
->gendisk
= disk
;
4873 /* As soon as we call add_disk(), another thread could get
4874 * through to md_open, so make sure it doesn't get too far
4876 mutex_lock(&mddev
->open_mutex
);
4879 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4880 &disk_to_dev(disk
)->kobj
, "%s", "md");
4882 /* This isn't possible, but as kobject_init_and_add is marked
4883 * __must_check, we must do something with the result
4885 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4889 if (mddev
->kobj
.sd
&&
4890 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4891 printk(KERN_DEBUG
"pointless warning\n");
4892 mutex_unlock(&mddev
->open_mutex
);
4894 mutex_unlock(&disks_mutex
);
4895 if (!error
&& mddev
->kobj
.sd
) {
4896 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4897 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4903 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4905 md_alloc(dev
, NULL
);
4909 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4911 /* val must be "md_*" where * is not all digits.
4912 * We allocate an array with a large free minor number, and
4913 * set the name to val. val must not already be an active name.
4915 int len
= strlen(val
);
4916 char buf
[DISK_NAME_LEN
];
4918 while (len
&& val
[len
-1] == '\n')
4920 if (len
>= DISK_NAME_LEN
)
4922 strlcpy(buf
, val
, len
+1);
4923 if (strncmp(buf
, "md_", 3) != 0)
4925 return md_alloc(0, buf
);
4928 static void md_safemode_timeout(unsigned long data
)
4930 struct mddev
*mddev
= (struct mddev
*) data
;
4932 if (!atomic_read(&mddev
->writes_pending
)) {
4933 mddev
->safemode
= 1;
4934 if (mddev
->external
)
4935 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4937 md_wakeup_thread(mddev
->thread
);
4940 static int start_dirty_degraded
;
4942 int md_run(struct mddev
*mddev
)
4945 struct md_rdev
*rdev
;
4946 struct md_personality
*pers
;
4948 if (list_empty(&mddev
->disks
))
4949 /* cannot run an array with no devices.. */
4954 /* Cannot run until previous stop completes properly */
4955 if (mddev
->sysfs_active
)
4959 * Analyze all RAID superblock(s)
4961 if (!mddev
->raid_disks
) {
4962 if (!mddev
->persistent
)
4967 if (mddev
->level
!= LEVEL_NONE
)
4968 request_module("md-level-%d", mddev
->level
);
4969 else if (mddev
->clevel
[0])
4970 request_module("md-%s", mddev
->clevel
);
4973 * Drop all container device buffers, from now on
4974 * the only valid external interface is through the md
4977 rdev_for_each(rdev
, mddev
) {
4978 if (test_bit(Faulty
, &rdev
->flags
))
4980 sync_blockdev(rdev
->bdev
);
4981 invalidate_bdev(rdev
->bdev
);
4983 /* perform some consistency tests on the device.
4984 * We don't want the data to overlap the metadata,
4985 * Internal Bitmap issues have been handled elsewhere.
4987 if (rdev
->meta_bdev
) {
4988 /* Nothing to check */;
4989 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4990 if (mddev
->dev_sectors
&&
4991 rdev
->data_offset
+ mddev
->dev_sectors
4993 printk("md: %s: data overlaps metadata\n",
4998 if (rdev
->sb_start
+ rdev
->sb_size
/512
4999 > rdev
->data_offset
) {
5000 printk("md: %s: metadata overlaps data\n",
5005 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5008 if (mddev
->bio_set
== NULL
)
5009 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
5010 sizeof(struct mddev
*));
5012 spin_lock(&pers_lock
);
5013 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5014 if (!pers
|| !try_module_get(pers
->owner
)) {
5015 spin_unlock(&pers_lock
);
5016 if (mddev
->level
!= LEVEL_NONE
)
5017 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5020 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5025 spin_unlock(&pers_lock
);
5026 if (mddev
->level
!= pers
->level
) {
5027 mddev
->level
= pers
->level
;
5028 mddev
->new_level
= pers
->level
;
5030 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5032 if (mddev
->reshape_position
!= MaxSector
&&
5033 pers
->start_reshape
== NULL
) {
5034 /* This personality cannot handle reshaping... */
5036 module_put(pers
->owner
);
5040 if (pers
->sync_request
) {
5041 /* Warn if this is a potentially silly
5044 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5045 struct md_rdev
*rdev2
;
5048 rdev_for_each(rdev
, mddev
)
5049 rdev_for_each(rdev2
, mddev
) {
5051 rdev
->bdev
->bd_contains
==
5052 rdev2
->bdev
->bd_contains
) {
5054 "%s: WARNING: %s appears to be"
5055 " on the same physical disk as"
5058 bdevname(rdev
->bdev
,b
),
5059 bdevname(rdev2
->bdev
,b2
));
5066 "True protection against single-disk"
5067 " failure might be compromised.\n");
5070 mddev
->recovery
= 0;
5071 /* may be over-ridden by personality */
5072 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5074 mddev
->ok_start_degraded
= start_dirty_degraded
;
5076 if (start_readonly
&& mddev
->ro
== 0)
5077 mddev
->ro
= 2; /* read-only, but switch on first write */
5079 err
= mddev
->pers
->run(mddev
);
5081 printk(KERN_ERR
"md: pers->run() failed ...\n");
5082 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5083 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5084 " but 'external_size' not in effect?\n", __func__
);
5086 "md: invalid array_size %llu > default size %llu\n",
5087 (unsigned long long)mddev
->array_sectors
/ 2,
5088 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
5090 mddev
->pers
->stop(mddev
);
5092 if (err
== 0 && mddev
->pers
->sync_request
&&
5093 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5094 err
= bitmap_create(mddev
);
5096 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5097 mdname(mddev
), err
);
5098 mddev
->pers
->stop(mddev
);
5102 module_put(mddev
->pers
->owner
);
5104 bitmap_destroy(mddev
);
5107 if (mddev
->pers
->sync_request
) {
5108 if (mddev
->kobj
.sd
&&
5109 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5111 "md: cannot register extra attributes for %s\n",
5113 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5114 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5117 atomic_set(&mddev
->writes_pending
,0);
5118 atomic_set(&mddev
->max_corr_read_errors
,
5119 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5120 mddev
->safemode
= 0;
5121 mddev
->safemode_timer
.function
= md_safemode_timeout
;
5122 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
5123 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5127 rdev_for_each(rdev
, mddev
)
5128 if (rdev
->raid_disk
>= 0)
5129 if (sysfs_link_rdev(mddev
, rdev
))
5130 /* failure here is OK */;
5132 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5135 md_update_sb(mddev
, 0);
5137 md_new_event(mddev
);
5138 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5139 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5140 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5143 EXPORT_SYMBOL_GPL(md_run
);
5145 static int do_md_run(struct mddev
*mddev
)
5149 err
= md_run(mddev
);
5152 err
= bitmap_load(mddev
);
5154 bitmap_destroy(mddev
);
5158 md_wakeup_thread(mddev
->thread
);
5159 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5161 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5162 revalidate_disk(mddev
->gendisk
);
5164 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5169 static int restart_array(struct mddev
*mddev
)
5171 struct gendisk
*disk
= mddev
->gendisk
;
5173 /* Complain if it has no devices */
5174 if (list_empty(&mddev
->disks
))
5180 mddev
->safemode
= 0;
5182 set_disk_ro(disk
, 0);
5183 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5185 /* Kick recovery or resync if necessary */
5186 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5187 md_wakeup_thread(mddev
->thread
);
5188 md_wakeup_thread(mddev
->sync_thread
);
5189 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5193 /* similar to deny_write_access, but accounts for our holding a reference
5194 * to the file ourselves */
5195 static int deny_bitmap_write_access(struct file
* file
)
5197 struct inode
*inode
= file
->f_mapping
->host
;
5199 spin_lock(&inode
->i_lock
);
5200 if (atomic_read(&inode
->i_writecount
) > 1) {
5201 spin_unlock(&inode
->i_lock
);
5204 atomic_set(&inode
->i_writecount
, -1);
5205 spin_unlock(&inode
->i_lock
);
5210 void restore_bitmap_write_access(struct file
*file
)
5212 struct inode
*inode
= file
->f_mapping
->host
;
5214 spin_lock(&inode
->i_lock
);
5215 atomic_set(&inode
->i_writecount
, 1);
5216 spin_unlock(&inode
->i_lock
);
5219 static void md_clean(struct mddev
*mddev
)
5221 mddev
->array_sectors
= 0;
5222 mddev
->external_size
= 0;
5223 mddev
->dev_sectors
= 0;
5224 mddev
->raid_disks
= 0;
5225 mddev
->recovery_cp
= 0;
5226 mddev
->resync_min
= 0;
5227 mddev
->resync_max
= MaxSector
;
5228 mddev
->reshape_position
= MaxSector
;
5229 mddev
->external
= 0;
5230 mddev
->persistent
= 0;
5231 mddev
->level
= LEVEL_NONE
;
5232 mddev
->clevel
[0] = 0;
5235 mddev
->metadata_type
[0] = 0;
5236 mddev
->chunk_sectors
= 0;
5237 mddev
->ctime
= mddev
->utime
= 0;
5239 mddev
->max_disks
= 0;
5241 mddev
->can_decrease_events
= 0;
5242 mddev
->delta_disks
= 0;
5243 mddev
->reshape_backwards
= 0;
5244 mddev
->new_level
= LEVEL_NONE
;
5245 mddev
->new_layout
= 0;
5246 mddev
->new_chunk_sectors
= 0;
5247 mddev
->curr_resync
= 0;
5248 mddev
->resync_mismatches
= 0;
5249 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5250 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5251 mddev
->recovery
= 0;
5254 mddev
->degraded
= 0;
5255 mddev
->safemode
= 0;
5256 mddev
->merge_check_needed
= 0;
5257 mddev
->bitmap_info
.offset
= 0;
5258 mddev
->bitmap_info
.default_offset
= 0;
5259 mddev
->bitmap_info
.default_space
= 0;
5260 mddev
->bitmap_info
.chunksize
= 0;
5261 mddev
->bitmap_info
.daemon_sleep
= 0;
5262 mddev
->bitmap_info
.max_write_behind
= 0;
5265 static void __md_stop_writes(struct mddev
*mddev
)
5267 if (mddev
->sync_thread
) {
5268 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5269 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5270 reap_sync_thread(mddev
);
5273 del_timer_sync(&mddev
->safemode_timer
);
5275 bitmap_flush(mddev
);
5276 md_super_wait(mddev
);
5278 if (!mddev
->in_sync
|| mddev
->flags
) {
5279 /* mark array as shutdown cleanly */
5281 md_update_sb(mddev
, 1);
5285 void md_stop_writes(struct mddev
*mddev
)
5288 __md_stop_writes(mddev
);
5289 mddev_unlock(mddev
);
5291 EXPORT_SYMBOL_GPL(md_stop_writes
);
5293 void md_stop(struct mddev
*mddev
)
5296 mddev
->pers
->stop(mddev
);
5297 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5298 mddev
->to_remove
= &md_redundancy_group
;
5299 module_put(mddev
->pers
->owner
);
5301 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5303 EXPORT_SYMBOL_GPL(md_stop
);
5305 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5308 mutex_lock(&mddev
->open_mutex
);
5309 if (atomic_read(&mddev
->openers
) > !!bdev
) {
5310 printk("md: %s still in use.\n",mdname(mddev
));
5315 sync_blockdev(bdev
);
5317 __md_stop_writes(mddev
);
5323 set_disk_ro(mddev
->gendisk
, 1);
5324 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5325 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5329 mutex_unlock(&mddev
->open_mutex
);
5334 * 0 - completely stop and dis-assemble array
5335 * 2 - stop but do not disassemble array
5337 static int do_md_stop(struct mddev
* mddev
, int mode
,
5338 struct block_device
*bdev
)
5340 struct gendisk
*disk
= mddev
->gendisk
;
5341 struct md_rdev
*rdev
;
5343 mutex_lock(&mddev
->open_mutex
);
5344 if (atomic_read(&mddev
->openers
) > !!bdev
||
5345 mddev
->sysfs_active
) {
5346 printk("md: %s still in use.\n",mdname(mddev
));
5347 mutex_unlock(&mddev
->open_mutex
);
5351 /* It is possible IO was issued on some other
5352 * open file which was closed before we took ->open_mutex.
5353 * As that was not the last close __blkdev_put will not
5354 * have called sync_blockdev, so we must.
5356 sync_blockdev(bdev
);
5360 set_disk_ro(disk
, 0);
5362 __md_stop_writes(mddev
);
5364 mddev
->queue
->merge_bvec_fn
= NULL
;
5365 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5367 /* tell userspace to handle 'inactive' */
5368 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5370 rdev_for_each(rdev
, mddev
)
5371 if (rdev
->raid_disk
>= 0)
5372 sysfs_unlink_rdev(mddev
, rdev
);
5374 set_capacity(disk
, 0);
5375 mutex_unlock(&mddev
->open_mutex
);
5377 revalidate_disk(disk
);
5382 mutex_unlock(&mddev
->open_mutex
);
5384 * Free resources if final stop
5387 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5389 bitmap_destroy(mddev
);
5390 if (mddev
->bitmap_info
.file
) {
5391 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5392 fput(mddev
->bitmap_info
.file
);
5393 mddev
->bitmap_info
.file
= NULL
;
5395 mddev
->bitmap_info
.offset
= 0;
5397 export_array(mddev
);
5400 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5401 if (mddev
->hold_active
== UNTIL_STOP
)
5402 mddev
->hold_active
= 0;
5404 blk_integrity_unregister(disk
);
5405 md_new_event(mddev
);
5406 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5411 static void autorun_array(struct mddev
*mddev
)
5413 struct md_rdev
*rdev
;
5416 if (list_empty(&mddev
->disks
))
5419 printk(KERN_INFO
"md: running: ");
5421 rdev_for_each(rdev
, mddev
) {
5422 char b
[BDEVNAME_SIZE
];
5423 printk("<%s>", bdevname(rdev
->bdev
,b
));
5427 err
= do_md_run(mddev
);
5429 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5430 do_md_stop(mddev
, 0, NULL
);
5435 * lets try to run arrays based on all disks that have arrived
5436 * until now. (those are in pending_raid_disks)
5438 * the method: pick the first pending disk, collect all disks with
5439 * the same UUID, remove all from the pending list and put them into
5440 * the 'same_array' list. Then order this list based on superblock
5441 * update time (freshest comes first), kick out 'old' disks and
5442 * compare superblocks. If everything's fine then run it.
5444 * If "unit" is allocated, then bump its reference count
5446 static void autorun_devices(int part
)
5448 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5449 struct mddev
*mddev
;
5450 char b
[BDEVNAME_SIZE
];
5452 printk(KERN_INFO
"md: autorun ...\n");
5453 while (!list_empty(&pending_raid_disks
)) {
5456 LIST_HEAD(candidates
);
5457 rdev0
= list_entry(pending_raid_disks
.next
,
5458 struct md_rdev
, same_set
);
5460 printk(KERN_INFO
"md: considering %s ...\n",
5461 bdevname(rdev0
->bdev
,b
));
5462 INIT_LIST_HEAD(&candidates
);
5463 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5464 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5465 printk(KERN_INFO
"md: adding %s ...\n",
5466 bdevname(rdev
->bdev
,b
));
5467 list_move(&rdev
->same_set
, &candidates
);
5470 * now we have a set of devices, with all of them having
5471 * mostly sane superblocks. It's time to allocate the
5475 dev
= MKDEV(mdp_major
,
5476 rdev0
->preferred_minor
<< MdpMinorShift
);
5477 unit
= MINOR(dev
) >> MdpMinorShift
;
5479 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5482 if (rdev0
->preferred_minor
!= unit
) {
5483 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5484 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5488 md_probe(dev
, NULL
, NULL
);
5489 mddev
= mddev_find(dev
);
5490 if (!mddev
|| !mddev
->gendisk
) {
5494 "md: cannot allocate memory for md drive.\n");
5497 if (mddev_lock(mddev
))
5498 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5500 else if (mddev
->raid_disks
|| mddev
->major_version
5501 || !list_empty(&mddev
->disks
)) {
5503 "md: %s already running, cannot run %s\n",
5504 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5505 mddev_unlock(mddev
);
5507 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5508 mddev
->persistent
= 1;
5509 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5510 list_del_init(&rdev
->same_set
);
5511 if (bind_rdev_to_array(rdev
, mddev
))
5514 autorun_array(mddev
);
5515 mddev_unlock(mddev
);
5517 /* on success, candidates will be empty, on error
5520 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5521 list_del_init(&rdev
->same_set
);
5526 printk(KERN_INFO
"md: ... autorun DONE.\n");
5528 #endif /* !MODULE */
5530 static int get_version(void __user
* arg
)
5534 ver
.major
= MD_MAJOR_VERSION
;
5535 ver
.minor
= MD_MINOR_VERSION
;
5536 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5538 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5544 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5546 mdu_array_info_t info
;
5547 int nr
,working
,insync
,failed
,spare
;
5548 struct md_rdev
*rdev
;
5550 nr
=working
=insync
=failed
=spare
=0;
5551 rdev_for_each(rdev
, mddev
) {
5553 if (test_bit(Faulty
, &rdev
->flags
))
5557 if (test_bit(In_sync
, &rdev
->flags
))
5564 info
.major_version
= mddev
->major_version
;
5565 info
.minor_version
= mddev
->minor_version
;
5566 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5567 info
.ctime
= mddev
->ctime
;
5568 info
.level
= mddev
->level
;
5569 info
.size
= mddev
->dev_sectors
/ 2;
5570 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5573 info
.raid_disks
= mddev
->raid_disks
;
5574 info
.md_minor
= mddev
->md_minor
;
5575 info
.not_persistent
= !mddev
->persistent
;
5577 info
.utime
= mddev
->utime
;
5580 info
.state
= (1<<MD_SB_CLEAN
);
5581 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5582 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5583 info
.active_disks
= insync
;
5584 info
.working_disks
= working
;
5585 info
.failed_disks
= failed
;
5586 info
.spare_disks
= spare
;
5588 info
.layout
= mddev
->layout
;
5589 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5591 if (copy_to_user(arg
, &info
, sizeof(info
)))
5597 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5599 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5600 char *ptr
, *buf
= NULL
;
5603 if (md_allow_write(mddev
))
5604 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5606 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5611 /* bitmap disabled, zero the first byte and copy out */
5612 if (!mddev
->bitmap
|| !mddev
->bitmap
->storage
.file
) {
5613 file
->pathname
[0] = '\0';
5617 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5621 ptr
= d_path(&mddev
->bitmap
->storage
.file
->f_path
,
5622 buf
, sizeof(file
->pathname
));
5626 strcpy(file
->pathname
, ptr
);
5630 if (copy_to_user(arg
, file
, sizeof(*file
)))
5638 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5640 mdu_disk_info_t info
;
5641 struct md_rdev
*rdev
;
5643 if (copy_from_user(&info
, arg
, sizeof(info
)))
5646 rdev
= find_rdev_nr(mddev
, info
.number
);
5648 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5649 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5650 info
.raid_disk
= rdev
->raid_disk
;
5652 if (test_bit(Faulty
, &rdev
->flags
))
5653 info
.state
|= (1<<MD_DISK_FAULTY
);
5654 else if (test_bit(In_sync
, &rdev
->flags
)) {
5655 info
.state
|= (1<<MD_DISK_ACTIVE
);
5656 info
.state
|= (1<<MD_DISK_SYNC
);
5658 if (test_bit(WriteMostly
, &rdev
->flags
))
5659 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5661 info
.major
= info
.minor
= 0;
5662 info
.raid_disk
= -1;
5663 info
.state
= (1<<MD_DISK_REMOVED
);
5666 if (copy_to_user(arg
, &info
, sizeof(info
)))
5672 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5674 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5675 struct md_rdev
*rdev
;
5676 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5678 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5681 if (!mddev
->raid_disks
) {
5683 /* expecting a device which has a superblock */
5684 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5687 "md: md_import_device returned %ld\n",
5689 return PTR_ERR(rdev
);
5691 if (!list_empty(&mddev
->disks
)) {
5692 struct md_rdev
*rdev0
5693 = list_entry(mddev
->disks
.next
,
5694 struct md_rdev
, same_set
);
5695 err
= super_types
[mddev
->major_version
]
5696 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5699 "md: %s has different UUID to %s\n",
5700 bdevname(rdev
->bdev
,b
),
5701 bdevname(rdev0
->bdev
,b2
));
5706 err
= bind_rdev_to_array(rdev
, mddev
);
5713 * add_new_disk can be used once the array is assembled
5714 * to add "hot spares". They must already have a superblock
5719 if (!mddev
->pers
->hot_add_disk
) {
5721 "%s: personality does not support diskops!\n",
5725 if (mddev
->persistent
)
5726 rdev
= md_import_device(dev
, mddev
->major_version
,
5727 mddev
->minor_version
);
5729 rdev
= md_import_device(dev
, -1, -1);
5732 "md: md_import_device returned %ld\n",
5734 return PTR_ERR(rdev
);
5736 /* set saved_raid_disk if appropriate */
5737 if (!mddev
->persistent
) {
5738 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5739 info
->raid_disk
< mddev
->raid_disks
) {
5740 rdev
->raid_disk
= info
->raid_disk
;
5741 set_bit(In_sync
, &rdev
->flags
);
5743 rdev
->raid_disk
= -1;
5745 super_types
[mddev
->major_version
].
5746 validate_super(mddev
, rdev
);
5747 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5748 rdev
->raid_disk
!= info
->raid_disk
) {
5749 /* This was a hot-add request, but events doesn't
5750 * match, so reject it.
5756 if (test_bit(In_sync
, &rdev
->flags
))
5757 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5759 rdev
->saved_raid_disk
= -1;
5761 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5762 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5763 set_bit(WriteMostly
, &rdev
->flags
);
5765 clear_bit(WriteMostly
, &rdev
->flags
);
5767 rdev
->raid_disk
= -1;
5768 err
= bind_rdev_to_array(rdev
, mddev
);
5769 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5770 /* If there is hot_add_disk but no hot_remove_disk
5771 * then added disks for geometry changes,
5772 * and should be added immediately.
5774 super_types
[mddev
->major_version
].
5775 validate_super(mddev
, rdev
);
5776 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5778 unbind_rdev_from_array(rdev
);
5783 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5785 md_update_sb(mddev
, 1);
5786 if (mddev
->degraded
)
5787 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5788 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5790 md_new_event(mddev
);
5791 md_wakeup_thread(mddev
->thread
);
5795 /* otherwise, add_new_disk is only allowed
5796 * for major_version==0 superblocks
5798 if (mddev
->major_version
!= 0) {
5799 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5804 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5806 rdev
= md_import_device(dev
, -1, 0);
5809 "md: error, md_import_device() returned %ld\n",
5811 return PTR_ERR(rdev
);
5813 rdev
->desc_nr
= info
->number
;
5814 if (info
->raid_disk
< mddev
->raid_disks
)
5815 rdev
->raid_disk
= info
->raid_disk
;
5817 rdev
->raid_disk
= -1;
5819 if (rdev
->raid_disk
< mddev
->raid_disks
)
5820 if (info
->state
& (1<<MD_DISK_SYNC
))
5821 set_bit(In_sync
, &rdev
->flags
);
5823 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5824 set_bit(WriteMostly
, &rdev
->flags
);
5826 if (!mddev
->persistent
) {
5827 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5828 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5830 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5831 rdev
->sectors
= rdev
->sb_start
;
5833 err
= bind_rdev_to_array(rdev
, mddev
);
5843 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5845 char b
[BDEVNAME_SIZE
];
5846 struct md_rdev
*rdev
;
5848 rdev
= find_rdev(mddev
, dev
);
5852 if (rdev
->raid_disk
>= 0)
5855 kick_rdev_from_array(rdev
);
5856 md_update_sb(mddev
, 1);
5857 md_new_event(mddev
);
5861 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5862 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5866 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5868 char b
[BDEVNAME_SIZE
];
5870 struct md_rdev
*rdev
;
5875 if (mddev
->major_version
!= 0) {
5876 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5877 " version-0 superblocks.\n",
5881 if (!mddev
->pers
->hot_add_disk
) {
5883 "%s: personality does not support diskops!\n",
5888 rdev
= md_import_device(dev
, -1, 0);
5891 "md: error, md_import_device() returned %ld\n",
5896 if (mddev
->persistent
)
5897 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5899 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5901 rdev
->sectors
= rdev
->sb_start
;
5903 if (test_bit(Faulty
, &rdev
->flags
)) {
5905 "md: can not hot-add faulty %s disk to %s!\n",
5906 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5910 clear_bit(In_sync
, &rdev
->flags
);
5912 rdev
->saved_raid_disk
= -1;
5913 err
= bind_rdev_to_array(rdev
, mddev
);
5918 * The rest should better be atomic, we can have disk failures
5919 * noticed in interrupt contexts ...
5922 rdev
->raid_disk
= -1;
5924 md_update_sb(mddev
, 1);
5927 * Kick recovery, maybe this spare has to be added to the
5928 * array immediately.
5930 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5931 md_wakeup_thread(mddev
->thread
);
5932 md_new_event(mddev
);
5940 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5945 if (!mddev
->pers
->quiesce
)
5947 if (mddev
->recovery
|| mddev
->sync_thread
)
5949 /* we should be able to change the bitmap.. */
5955 return -EEXIST
; /* cannot add when bitmap is present */
5956 mddev
->bitmap_info
.file
= fget(fd
);
5958 if (mddev
->bitmap_info
.file
== NULL
) {
5959 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5964 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5966 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5968 fput(mddev
->bitmap_info
.file
);
5969 mddev
->bitmap_info
.file
= NULL
;
5972 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5973 } else if (mddev
->bitmap
== NULL
)
5974 return -ENOENT
; /* cannot remove what isn't there */
5977 mddev
->pers
->quiesce(mddev
, 1);
5979 err
= bitmap_create(mddev
);
5981 err
= bitmap_load(mddev
);
5983 if (fd
< 0 || err
) {
5984 bitmap_destroy(mddev
);
5985 fd
= -1; /* make sure to put the file */
5987 mddev
->pers
->quiesce(mddev
, 0);
5990 if (mddev
->bitmap_info
.file
) {
5991 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5992 fput(mddev
->bitmap_info
.file
);
5994 mddev
->bitmap_info
.file
= NULL
;
6001 * set_array_info is used two different ways
6002 * The original usage is when creating a new array.
6003 * In this usage, raid_disks is > 0 and it together with
6004 * level, size, not_persistent,layout,chunksize determine the
6005 * shape of the array.
6006 * This will always create an array with a type-0.90.0 superblock.
6007 * The newer usage is when assembling an array.
6008 * In this case raid_disks will be 0, and the major_version field is
6009 * use to determine which style super-blocks are to be found on the devices.
6010 * The minor and patch _version numbers are also kept incase the
6011 * super_block handler wishes to interpret them.
6013 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
6016 if (info
->raid_disks
== 0) {
6017 /* just setting version number for superblock loading */
6018 if (info
->major_version
< 0 ||
6019 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6020 super_types
[info
->major_version
].name
== NULL
) {
6021 /* maybe try to auto-load a module? */
6023 "md: superblock version %d not known\n",
6024 info
->major_version
);
6027 mddev
->major_version
= info
->major_version
;
6028 mddev
->minor_version
= info
->minor_version
;
6029 mddev
->patch_version
= info
->patch_version
;
6030 mddev
->persistent
= !info
->not_persistent
;
6031 /* ensure mddev_put doesn't delete this now that there
6032 * is some minimal configuration.
6034 mddev
->ctime
= get_seconds();
6037 mddev
->major_version
= MD_MAJOR_VERSION
;
6038 mddev
->minor_version
= MD_MINOR_VERSION
;
6039 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6040 mddev
->ctime
= get_seconds();
6042 mddev
->level
= info
->level
;
6043 mddev
->clevel
[0] = 0;
6044 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6045 mddev
->raid_disks
= info
->raid_disks
;
6046 /* don't set md_minor, it is determined by which /dev/md* was
6049 if (info
->state
& (1<<MD_SB_CLEAN
))
6050 mddev
->recovery_cp
= MaxSector
;
6052 mddev
->recovery_cp
= 0;
6053 mddev
->persistent
= ! info
->not_persistent
;
6054 mddev
->external
= 0;
6056 mddev
->layout
= info
->layout
;
6057 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6059 mddev
->max_disks
= MD_SB_DISKS
;
6061 if (mddev
->persistent
)
6063 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6065 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6066 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6067 mddev
->bitmap_info
.offset
= 0;
6069 mddev
->reshape_position
= MaxSector
;
6072 * Generate a 128 bit UUID
6074 get_random_bytes(mddev
->uuid
, 16);
6076 mddev
->new_level
= mddev
->level
;
6077 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6078 mddev
->new_layout
= mddev
->layout
;
6079 mddev
->delta_disks
= 0;
6080 mddev
->reshape_backwards
= 0;
6085 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6087 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6089 if (mddev
->external_size
)
6092 mddev
->array_sectors
= array_sectors
;
6094 EXPORT_SYMBOL(md_set_array_sectors
);
6096 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6098 struct md_rdev
*rdev
;
6100 int fit
= (num_sectors
== 0);
6102 if (mddev
->pers
->resize
== NULL
)
6104 /* The "num_sectors" is the number of sectors of each device that
6105 * is used. This can only make sense for arrays with redundancy.
6106 * linear and raid0 always use whatever space is available. We can only
6107 * consider changing this number if no resync or reconstruction is
6108 * happening, and if the new size is acceptable. It must fit before the
6109 * sb_start or, if that is <data_offset, it must fit before the size
6110 * of each device. If num_sectors is zero, we find the largest size
6113 if (mddev
->sync_thread
)
6116 rdev_for_each(rdev
, mddev
) {
6117 sector_t avail
= rdev
->sectors
;
6119 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6120 num_sectors
= avail
;
6121 if (avail
< num_sectors
)
6124 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6126 revalidate_disk(mddev
->gendisk
);
6130 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6133 struct md_rdev
*rdev
;
6134 /* change the number of raid disks */
6135 if (mddev
->pers
->check_reshape
== NULL
)
6137 if (raid_disks
<= 0 ||
6138 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6140 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
6143 rdev_for_each(rdev
, mddev
) {
6144 if (mddev
->raid_disks
< raid_disks
&&
6145 rdev
->data_offset
< rdev
->new_data_offset
)
6147 if (mddev
->raid_disks
> raid_disks
&&
6148 rdev
->data_offset
> rdev
->new_data_offset
)
6152 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6153 if (mddev
->delta_disks
< 0)
6154 mddev
->reshape_backwards
= 1;
6155 else if (mddev
->delta_disks
> 0)
6156 mddev
->reshape_backwards
= 0;
6158 rv
= mddev
->pers
->check_reshape(mddev
);
6160 mddev
->delta_disks
= 0;
6161 mddev
->reshape_backwards
= 0;
6168 * update_array_info is used to change the configuration of an
6170 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6171 * fields in the info are checked against the array.
6172 * Any differences that cannot be handled will cause an error.
6173 * Normally, only one change can be managed at a time.
6175 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6181 /* calculate expected state,ignoring low bits */
6182 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6183 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6185 if (mddev
->major_version
!= info
->major_version
||
6186 mddev
->minor_version
!= info
->minor_version
||
6187 /* mddev->patch_version != info->patch_version || */
6188 mddev
->ctime
!= info
->ctime
||
6189 mddev
->level
!= info
->level
||
6190 /* mddev->layout != info->layout || */
6191 !mddev
->persistent
!= info
->not_persistent
||
6192 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6193 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6194 ((state
^info
->state
) & 0xfffffe00)
6197 /* Check there is only one change */
6198 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6200 if (mddev
->raid_disks
!= info
->raid_disks
)
6202 if (mddev
->layout
!= info
->layout
)
6204 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6211 if (mddev
->layout
!= info
->layout
) {
6213 * we don't need to do anything at the md level, the
6214 * personality will take care of it all.
6216 if (mddev
->pers
->check_reshape
== NULL
)
6219 mddev
->new_layout
= info
->layout
;
6220 rv
= mddev
->pers
->check_reshape(mddev
);
6222 mddev
->new_layout
= mddev
->layout
;
6226 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6227 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6229 if (mddev
->raid_disks
!= info
->raid_disks
)
6230 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6232 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6233 if (mddev
->pers
->quiesce
== NULL
)
6235 if (mddev
->recovery
|| mddev
->sync_thread
)
6237 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6238 /* add the bitmap */
6241 if (mddev
->bitmap_info
.default_offset
== 0)
6243 mddev
->bitmap_info
.offset
=
6244 mddev
->bitmap_info
.default_offset
;
6245 mddev
->bitmap_info
.space
=
6246 mddev
->bitmap_info
.default_space
;
6247 mddev
->pers
->quiesce(mddev
, 1);
6248 rv
= bitmap_create(mddev
);
6250 rv
= bitmap_load(mddev
);
6252 bitmap_destroy(mddev
);
6253 mddev
->pers
->quiesce(mddev
, 0);
6255 /* remove the bitmap */
6258 if (mddev
->bitmap
->storage
.file
)
6260 mddev
->pers
->quiesce(mddev
, 1);
6261 bitmap_destroy(mddev
);
6262 mddev
->pers
->quiesce(mddev
, 0);
6263 mddev
->bitmap_info
.offset
= 0;
6266 md_update_sb(mddev
, 1);
6270 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6272 struct md_rdev
*rdev
;
6274 if (mddev
->pers
== NULL
)
6277 rdev
= find_rdev(mddev
, dev
);
6281 md_error(mddev
, rdev
);
6282 if (!test_bit(Faulty
, &rdev
->flags
))
6288 * We have a problem here : there is no easy way to give a CHS
6289 * virtual geometry. We currently pretend that we have a 2 heads
6290 * 4 sectors (with a BIG number of cylinders...). This drives
6291 * dosfs just mad... ;-)
6293 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6295 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6299 geo
->cylinders
= mddev
->array_sectors
/ 8;
6303 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6304 unsigned int cmd
, unsigned long arg
)
6307 void __user
*argp
= (void __user
*)arg
;
6308 struct mddev
*mddev
= NULL
;
6313 case GET_ARRAY_INFO
:
6317 if (!capable(CAP_SYS_ADMIN
))
6322 * Commands dealing with the RAID driver but not any
6328 err
= get_version(argp
);
6331 case PRINT_RAID_DEBUG
:
6339 autostart_arrays(arg
);
6346 * Commands creating/starting a new array:
6349 mddev
= bdev
->bd_disk
->private_data
;
6356 err
= mddev_lock(mddev
);
6359 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6366 case SET_ARRAY_INFO
:
6368 mdu_array_info_t info
;
6370 memset(&info
, 0, sizeof(info
));
6371 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6376 err
= update_array_info(mddev
, &info
);
6378 printk(KERN_WARNING
"md: couldn't update"
6379 " array info. %d\n", err
);
6384 if (!list_empty(&mddev
->disks
)) {
6386 "md: array %s already has disks!\n",
6391 if (mddev
->raid_disks
) {
6393 "md: array %s already initialised!\n",
6398 err
= set_array_info(mddev
, &info
);
6400 printk(KERN_WARNING
"md: couldn't set"
6401 " array info. %d\n", err
);
6411 * Commands querying/configuring an existing array:
6413 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6414 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6415 if ((!mddev
->raid_disks
&& !mddev
->external
)
6416 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6417 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6418 && cmd
!= GET_BITMAP_FILE
) {
6424 * Commands even a read-only array can execute:
6428 case GET_ARRAY_INFO
:
6429 err
= get_array_info(mddev
, argp
);
6432 case GET_BITMAP_FILE
:
6433 err
= get_bitmap_file(mddev
, argp
);
6437 err
= get_disk_info(mddev
, argp
);
6440 case RESTART_ARRAY_RW
:
6441 err
= restart_array(mddev
);
6445 err
= do_md_stop(mddev
, 0, bdev
);
6449 err
= md_set_readonly(mddev
, bdev
);
6453 if (get_user(ro
, (int __user
*)(arg
))) {
6459 /* if the bdev is going readonly the value of mddev->ro
6460 * does not matter, no writes are coming
6465 /* are we are already prepared for writes? */
6469 /* transitioning to readauto need only happen for
6470 * arrays that call md_write_start
6473 err
= restart_array(mddev
);
6476 set_disk_ro(mddev
->gendisk
, 0);
6483 * The remaining ioctls are changing the state of the
6484 * superblock, so we do not allow them on read-only arrays.
6485 * However non-MD ioctls (e.g. get-size) will still come through
6486 * here and hit the 'default' below, so only disallow
6487 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6489 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6490 if (mddev
->ro
== 2) {
6492 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6493 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6494 md_wakeup_thread(mddev
->thread
);
6505 mdu_disk_info_t info
;
6506 if (copy_from_user(&info
, argp
, sizeof(info
)))
6509 err
= add_new_disk(mddev
, &info
);
6513 case HOT_REMOVE_DISK
:
6514 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6518 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6521 case SET_DISK_FAULTY
:
6522 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6526 err
= do_md_run(mddev
);
6529 case SET_BITMAP_FILE
:
6530 err
= set_bitmap_file(mddev
, (int)arg
);
6540 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6542 mddev
->hold_active
= 0;
6543 mddev_unlock(mddev
);
6552 #ifdef CONFIG_COMPAT
6553 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6554 unsigned int cmd
, unsigned long arg
)
6557 case HOT_REMOVE_DISK
:
6559 case SET_DISK_FAULTY
:
6560 case SET_BITMAP_FILE
:
6561 /* These take in integer arg, do not convert */
6564 arg
= (unsigned long)compat_ptr(arg
);
6568 return md_ioctl(bdev
, mode
, cmd
, arg
);
6570 #endif /* CONFIG_COMPAT */
6572 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6575 * Succeed if we can lock the mddev, which confirms that
6576 * it isn't being stopped right now.
6578 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6584 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6585 /* we are racing with mddev_put which is discarding this
6589 /* Wait until bdev->bd_disk is definitely gone */
6590 flush_workqueue(md_misc_wq
);
6591 /* Then retry the open from the top */
6592 return -ERESTARTSYS
;
6594 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6596 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6600 atomic_inc(&mddev
->openers
);
6601 mutex_unlock(&mddev
->open_mutex
);
6603 check_disk_change(bdev
);
6608 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6610 struct mddev
*mddev
= disk
->private_data
;
6613 atomic_dec(&mddev
->openers
);
6619 static int md_media_changed(struct gendisk
*disk
)
6621 struct mddev
*mddev
= disk
->private_data
;
6623 return mddev
->changed
;
6626 static int md_revalidate(struct gendisk
*disk
)
6628 struct mddev
*mddev
= disk
->private_data
;
6633 static const struct block_device_operations md_fops
=
6635 .owner
= THIS_MODULE
,
6637 .release
= md_release
,
6639 #ifdef CONFIG_COMPAT
6640 .compat_ioctl
= md_compat_ioctl
,
6642 .getgeo
= md_getgeo
,
6643 .media_changed
= md_media_changed
,
6644 .revalidate_disk
= md_revalidate
,
6647 static int md_thread(void * arg
)
6649 struct md_thread
*thread
= arg
;
6652 * md_thread is a 'system-thread', it's priority should be very
6653 * high. We avoid resource deadlocks individually in each
6654 * raid personality. (RAID5 does preallocation) We also use RR and
6655 * the very same RT priority as kswapd, thus we will never get
6656 * into a priority inversion deadlock.
6658 * we definitely have to have equal or higher priority than
6659 * bdflush, otherwise bdflush will deadlock if there are too
6660 * many dirty RAID5 blocks.
6663 allow_signal(SIGKILL
);
6664 while (!kthread_should_stop()) {
6666 /* We need to wait INTERRUPTIBLE so that
6667 * we don't add to the load-average.
6668 * That means we need to be sure no signals are
6671 if (signal_pending(current
))
6672 flush_signals(current
);
6674 wait_event_interruptible_timeout
6676 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6677 || kthread_should_stop(),
6680 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6681 if (!kthread_should_stop())
6682 thread
->run(thread
->mddev
);
6688 void md_wakeup_thread(struct md_thread
*thread
)
6691 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6692 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6693 wake_up(&thread
->wqueue
);
6697 struct md_thread
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6700 struct md_thread
*thread
;
6702 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6706 init_waitqueue_head(&thread
->wqueue
);
6709 thread
->mddev
= mddev
;
6710 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6711 thread
->tsk
= kthread_run(md_thread
, thread
,
6713 mdname(thread
->mddev
),
6715 if (IS_ERR(thread
->tsk
)) {
6722 void md_unregister_thread(struct md_thread
**threadp
)
6724 struct md_thread
*thread
= *threadp
;
6727 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6728 /* Locking ensures that mddev_unlock does not wake_up a
6729 * non-existent thread
6731 spin_lock(&pers_lock
);
6733 spin_unlock(&pers_lock
);
6735 kthread_stop(thread
->tsk
);
6739 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6746 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6749 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6751 mddev
->pers
->error_handler(mddev
,rdev
);
6752 if (mddev
->degraded
)
6753 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6754 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6755 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6756 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6757 md_wakeup_thread(mddev
->thread
);
6758 if (mddev
->event_work
.func
)
6759 queue_work(md_misc_wq
, &mddev
->event_work
);
6760 md_new_event_inintr(mddev
);
6763 /* seq_file implementation /proc/mdstat */
6765 static void status_unused(struct seq_file
*seq
)
6768 struct md_rdev
*rdev
;
6770 seq_printf(seq
, "unused devices: ");
6772 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6773 char b
[BDEVNAME_SIZE
];
6775 seq_printf(seq
, "%s ",
6776 bdevname(rdev
->bdev
,b
));
6779 seq_printf(seq
, "<none>");
6781 seq_printf(seq
, "\n");
6785 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6787 sector_t max_sectors
, resync
, res
;
6788 unsigned long dt
, db
;
6791 unsigned int per_milli
;
6793 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6795 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
6796 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6797 max_sectors
= mddev
->resync_max_sectors
;
6799 max_sectors
= mddev
->dev_sectors
;
6802 * Should not happen.
6808 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6809 * in a sector_t, and (max_sectors>>scale) will fit in a
6810 * u32, as those are the requirements for sector_div.
6811 * Thus 'scale' must be at least 10
6814 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6815 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6818 res
= (resync
>>scale
)*1000;
6819 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6823 int i
, x
= per_milli
/50, y
= 20-x
;
6824 seq_printf(seq
, "[");
6825 for (i
= 0; i
< x
; i
++)
6826 seq_printf(seq
, "=");
6827 seq_printf(seq
, ">");
6828 for (i
= 0; i
< y
; i
++)
6829 seq_printf(seq
, ".");
6830 seq_printf(seq
, "] ");
6832 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6833 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6835 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6837 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6838 "resync" : "recovery"))),
6839 per_milli
/10, per_milli
% 10,
6840 (unsigned long long) resync
/2,
6841 (unsigned long long) max_sectors
/2);
6844 * dt: time from mark until now
6845 * db: blocks written from mark until now
6846 * rt: remaining time
6848 * rt is a sector_t, so could be 32bit or 64bit.
6849 * So we divide before multiply in case it is 32bit and close
6851 * We scale the divisor (db) by 32 to avoid losing precision
6852 * near the end of resync when the number of remaining sectors
6854 * We then divide rt by 32 after multiplying by db to compensate.
6855 * The '+1' avoids division by zero if db is very small.
6857 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6859 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6860 - mddev
->resync_mark_cnt
;
6862 rt
= max_sectors
- resync
; /* number of remaining sectors */
6863 sector_div(rt
, db
/32+1);
6867 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6868 ((unsigned long)rt
% 60)/6);
6870 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6873 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6875 struct list_head
*tmp
;
6877 struct mddev
*mddev
;
6885 spin_lock(&all_mddevs_lock
);
6886 list_for_each(tmp
,&all_mddevs
)
6888 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6890 spin_unlock(&all_mddevs_lock
);
6893 spin_unlock(&all_mddevs_lock
);
6895 return (void*)2;/* tail */
6899 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6901 struct list_head
*tmp
;
6902 struct mddev
*next_mddev
, *mddev
= v
;
6908 spin_lock(&all_mddevs_lock
);
6910 tmp
= all_mddevs
.next
;
6912 tmp
= mddev
->all_mddevs
.next
;
6913 if (tmp
!= &all_mddevs
)
6914 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6916 next_mddev
= (void*)2;
6919 spin_unlock(&all_mddevs_lock
);
6927 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6929 struct mddev
*mddev
= v
;
6931 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6935 static int md_seq_show(struct seq_file
*seq
, void *v
)
6937 struct mddev
*mddev
= v
;
6939 struct md_rdev
*rdev
;
6941 if (v
== (void*)1) {
6942 struct md_personality
*pers
;
6943 seq_printf(seq
, "Personalities : ");
6944 spin_lock(&pers_lock
);
6945 list_for_each_entry(pers
, &pers_list
, list
)
6946 seq_printf(seq
, "[%s] ", pers
->name
);
6948 spin_unlock(&pers_lock
);
6949 seq_printf(seq
, "\n");
6950 seq
->poll_event
= atomic_read(&md_event_count
);
6953 if (v
== (void*)2) {
6958 if (mddev_lock(mddev
) < 0)
6961 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6962 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6963 mddev
->pers
? "" : "in");
6966 seq_printf(seq
, " (read-only)");
6968 seq_printf(seq
, " (auto-read-only)");
6969 seq_printf(seq
, " %s", mddev
->pers
->name
);
6973 rdev_for_each(rdev
, mddev
) {
6974 char b
[BDEVNAME_SIZE
];
6975 seq_printf(seq
, " %s[%d]",
6976 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6977 if (test_bit(WriteMostly
, &rdev
->flags
))
6978 seq_printf(seq
, "(W)");
6979 if (test_bit(Faulty
, &rdev
->flags
)) {
6980 seq_printf(seq
, "(F)");
6983 if (rdev
->raid_disk
< 0)
6984 seq_printf(seq
, "(S)"); /* spare */
6985 if (test_bit(Replacement
, &rdev
->flags
))
6986 seq_printf(seq
, "(R)");
6987 sectors
+= rdev
->sectors
;
6990 if (!list_empty(&mddev
->disks
)) {
6992 seq_printf(seq
, "\n %llu blocks",
6993 (unsigned long long)
6994 mddev
->array_sectors
/ 2);
6996 seq_printf(seq
, "\n %llu blocks",
6997 (unsigned long long)sectors
/ 2);
6999 if (mddev
->persistent
) {
7000 if (mddev
->major_version
!= 0 ||
7001 mddev
->minor_version
!= 90) {
7002 seq_printf(seq
," super %d.%d",
7003 mddev
->major_version
,
7004 mddev
->minor_version
);
7006 } else if (mddev
->external
)
7007 seq_printf(seq
, " super external:%s",
7008 mddev
->metadata_type
);
7010 seq_printf(seq
, " super non-persistent");
7013 mddev
->pers
->status(seq
, mddev
);
7014 seq_printf(seq
, "\n ");
7015 if (mddev
->pers
->sync_request
) {
7016 if (mddev
->curr_resync
> 2) {
7017 status_resync(seq
, mddev
);
7018 seq_printf(seq
, "\n ");
7019 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
7020 seq_printf(seq
, "\tresync=DELAYED\n ");
7021 else if (mddev
->recovery_cp
< MaxSector
)
7022 seq_printf(seq
, "\tresync=PENDING\n ");
7025 seq_printf(seq
, "\n ");
7027 bitmap_status(seq
, mddev
->bitmap
);
7029 seq_printf(seq
, "\n");
7031 mddev_unlock(mddev
);
7036 static const struct seq_operations md_seq_ops
= {
7037 .start
= md_seq_start
,
7038 .next
= md_seq_next
,
7039 .stop
= md_seq_stop
,
7040 .show
= md_seq_show
,
7043 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7045 struct seq_file
*seq
;
7048 error
= seq_open(file
, &md_seq_ops
);
7052 seq
= file
->private_data
;
7053 seq
->poll_event
= atomic_read(&md_event_count
);
7057 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7059 struct seq_file
*seq
= filp
->private_data
;
7062 poll_wait(filp
, &md_event_waiters
, wait
);
7064 /* always allow read */
7065 mask
= POLLIN
| POLLRDNORM
;
7067 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7068 mask
|= POLLERR
| POLLPRI
;
7072 static const struct file_operations md_seq_fops
= {
7073 .owner
= THIS_MODULE
,
7074 .open
= md_seq_open
,
7076 .llseek
= seq_lseek
,
7077 .release
= seq_release_private
,
7078 .poll
= mdstat_poll
,
7081 int register_md_personality(struct md_personality
*p
)
7083 spin_lock(&pers_lock
);
7084 list_add_tail(&p
->list
, &pers_list
);
7085 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
7086 spin_unlock(&pers_lock
);
7090 int unregister_md_personality(struct md_personality
*p
)
7092 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7093 spin_lock(&pers_lock
);
7094 list_del_init(&p
->list
);
7095 spin_unlock(&pers_lock
);
7099 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7101 struct md_rdev
* rdev
;
7107 rdev_for_each_rcu(rdev
, mddev
) {
7108 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7109 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7110 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7111 atomic_read(&disk
->sync_io
);
7112 /* sync IO will cause sync_io to increase before the disk_stats
7113 * as sync_io is counted when a request starts, and
7114 * disk_stats is counted when it completes.
7115 * So resync activity will cause curr_events to be smaller than
7116 * when there was no such activity.
7117 * non-sync IO will cause disk_stat to increase without
7118 * increasing sync_io so curr_events will (eventually)
7119 * be larger than it was before. Once it becomes
7120 * substantially larger, the test below will cause
7121 * the array to appear non-idle, and resync will slow
7123 * If there is a lot of outstanding resync activity when
7124 * we set last_event to curr_events, then all that activity
7125 * completing might cause the array to appear non-idle
7126 * and resync will be slowed down even though there might
7127 * not have been non-resync activity. This will only
7128 * happen once though. 'last_events' will soon reflect
7129 * the state where there is little or no outstanding
7130 * resync requests, and further resync activity will
7131 * always make curr_events less than last_events.
7134 if (init
|| curr_events
- rdev
->last_events
> 64) {
7135 rdev
->last_events
= curr_events
;
7143 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7145 /* another "blocks" (512byte) blocks have been synced */
7146 atomic_sub(blocks
, &mddev
->recovery_active
);
7147 wake_up(&mddev
->recovery_wait
);
7149 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7150 md_wakeup_thread(mddev
->thread
);
7151 // stop recovery, signal do_sync ....
7156 /* md_write_start(mddev, bi)
7157 * If we need to update some array metadata (e.g. 'active' flag
7158 * in superblock) before writing, schedule a superblock update
7159 * and wait for it to complete.
7161 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7164 if (bio_data_dir(bi
) != WRITE
)
7167 BUG_ON(mddev
->ro
== 1);
7168 if (mddev
->ro
== 2) {
7169 /* need to switch to read/write */
7171 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7172 md_wakeup_thread(mddev
->thread
);
7173 md_wakeup_thread(mddev
->sync_thread
);
7176 atomic_inc(&mddev
->writes_pending
);
7177 if (mddev
->safemode
== 1)
7178 mddev
->safemode
= 0;
7179 if (mddev
->in_sync
) {
7180 spin_lock_irq(&mddev
->write_lock
);
7181 if (mddev
->in_sync
) {
7183 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7184 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7185 md_wakeup_thread(mddev
->thread
);
7188 spin_unlock_irq(&mddev
->write_lock
);
7191 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7192 wait_event(mddev
->sb_wait
,
7193 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7196 void md_write_end(struct mddev
*mddev
)
7198 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7199 if (mddev
->safemode
== 2)
7200 md_wakeup_thread(mddev
->thread
);
7201 else if (mddev
->safemode_delay
)
7202 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7206 /* md_allow_write(mddev)
7207 * Calling this ensures that the array is marked 'active' so that writes
7208 * may proceed without blocking. It is important to call this before
7209 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7210 * Must be called with mddev_lock held.
7212 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7213 * is dropped, so return -EAGAIN after notifying userspace.
7215 int md_allow_write(struct mddev
*mddev
)
7221 if (!mddev
->pers
->sync_request
)
7224 spin_lock_irq(&mddev
->write_lock
);
7225 if (mddev
->in_sync
) {
7227 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7228 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7229 if (mddev
->safemode_delay
&&
7230 mddev
->safemode
== 0)
7231 mddev
->safemode
= 1;
7232 spin_unlock_irq(&mddev
->write_lock
);
7233 md_update_sb(mddev
, 0);
7234 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7236 spin_unlock_irq(&mddev
->write_lock
);
7238 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7243 EXPORT_SYMBOL_GPL(md_allow_write
);
7245 #define SYNC_MARKS 10
7246 #define SYNC_MARK_STEP (3*HZ)
7247 void md_do_sync(struct mddev
*mddev
)
7249 struct mddev
*mddev2
;
7250 unsigned int currspeed
= 0,
7252 sector_t max_sectors
,j
, io_sectors
;
7253 unsigned long mark
[SYNC_MARKS
];
7254 sector_t mark_cnt
[SYNC_MARKS
];
7256 struct list_head
*tmp
;
7257 sector_t last_check
;
7259 struct md_rdev
*rdev
;
7261 struct blk_plug plug
;
7263 /* just incase thread restarts... */
7264 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7266 if (mddev
->ro
) /* never try to sync a read-only array */
7269 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7270 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7271 desc
= "data-check";
7272 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7273 desc
= "requested-resync";
7276 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7281 /* we overload curr_resync somewhat here.
7282 * 0 == not engaged in resync at all
7283 * 2 == checking that there is no conflict with another sync
7284 * 1 == like 2, but have yielded to allow conflicting resync to
7286 * other == active in resync - this many blocks
7288 * Before starting a resync we must have set curr_resync to
7289 * 2, and then checked that every "conflicting" array has curr_resync
7290 * less than ours. When we find one that is the same or higher
7291 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7292 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7293 * This will mean we have to start checking from the beginning again.
7298 mddev
->curr_resync
= 2;
7301 if (kthread_should_stop())
7302 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7304 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7306 for_each_mddev(mddev2
, tmp
) {
7307 if (mddev2
== mddev
)
7309 if (!mddev
->parallel_resync
7310 && mddev2
->curr_resync
7311 && match_mddev_units(mddev
, mddev2
)) {
7313 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7314 /* arbitrarily yield */
7315 mddev
->curr_resync
= 1;
7316 wake_up(&resync_wait
);
7318 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7319 /* no need to wait here, we can wait the next
7320 * time 'round when curr_resync == 2
7323 /* We need to wait 'interruptible' so as not to
7324 * contribute to the load average, and not to
7325 * be caught by 'softlockup'
7327 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7328 if (!kthread_should_stop() &&
7329 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7330 printk(KERN_INFO
"md: delaying %s of %s"
7331 " until %s has finished (they"
7332 " share one or more physical units)\n",
7333 desc
, mdname(mddev
), mdname(mddev2
));
7335 if (signal_pending(current
))
7336 flush_signals(current
);
7338 finish_wait(&resync_wait
, &wq
);
7341 finish_wait(&resync_wait
, &wq
);
7344 } while (mddev
->curr_resync
< 2);
7347 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7348 /* resync follows the size requested by the personality,
7349 * which defaults to physical size, but can be virtual size
7351 max_sectors
= mddev
->resync_max_sectors
;
7352 mddev
->resync_mismatches
= 0;
7353 /* we don't use the checkpoint if there's a bitmap */
7354 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7355 j
= mddev
->resync_min
;
7356 else if (!mddev
->bitmap
)
7357 j
= mddev
->recovery_cp
;
7359 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7360 max_sectors
= mddev
->resync_max_sectors
;
7362 /* recovery follows the physical size of devices */
7363 max_sectors
= mddev
->dev_sectors
;
7366 rdev_for_each_rcu(rdev
, mddev
)
7367 if (rdev
->raid_disk
>= 0 &&
7368 !test_bit(Faulty
, &rdev
->flags
) &&
7369 !test_bit(In_sync
, &rdev
->flags
) &&
7370 rdev
->recovery_offset
< j
)
7371 j
= rdev
->recovery_offset
;
7375 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7376 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7377 " %d KB/sec/disk.\n", speed_min(mddev
));
7378 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7379 "(but not more than %d KB/sec) for %s.\n",
7380 speed_max(mddev
), desc
);
7382 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7385 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7387 mark_cnt
[m
] = io_sectors
;
7390 mddev
->resync_mark
= mark
[last_mark
];
7391 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7394 * Tune reconstruction:
7396 window
= 32*(PAGE_SIZE
/512);
7397 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7398 window
/2, (unsigned long long)max_sectors
/2);
7400 atomic_set(&mddev
->recovery_active
, 0);
7405 "md: resuming %s of %s from checkpoint.\n",
7406 desc
, mdname(mddev
));
7407 mddev
->curr_resync
= j
;
7409 mddev
->curr_resync_completed
= j
;
7411 blk_start_plug(&plug
);
7412 while (j
< max_sectors
) {
7417 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7418 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7419 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7420 > (max_sectors
>> 4)) ||
7421 (j
- mddev
->curr_resync_completed
)*2
7422 >= mddev
->resync_max
- mddev
->curr_resync_completed
7424 /* time to update curr_resync_completed */
7425 wait_event(mddev
->recovery_wait
,
7426 atomic_read(&mddev
->recovery_active
) == 0);
7427 mddev
->curr_resync_completed
= j
;
7428 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7429 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7432 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7433 /* As this condition is controlled by user-space,
7434 * we can block indefinitely, so use '_interruptible'
7435 * to avoid triggering warnings.
7437 flush_signals(current
); /* just in case */
7438 wait_event_interruptible(mddev
->recovery_wait
,
7439 mddev
->resync_max
> j
7440 || kthread_should_stop());
7443 if (kthread_should_stop())
7446 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7447 currspeed
< speed_min(mddev
));
7449 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7453 if (!skipped
) { /* actual IO requested */
7454 io_sectors
+= sectors
;
7455 atomic_add(sectors
, &mddev
->recovery_active
);
7458 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7462 if (j
>1) mddev
->curr_resync
= j
;
7463 mddev
->curr_mark_cnt
= io_sectors
;
7464 if (last_check
== 0)
7465 /* this is the earliest that rebuild will be
7466 * visible in /proc/mdstat
7468 md_new_event(mddev
);
7470 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7473 last_check
= io_sectors
;
7475 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7477 int next
= (last_mark
+1) % SYNC_MARKS
;
7479 mddev
->resync_mark
= mark
[next
];
7480 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7481 mark
[next
] = jiffies
;
7482 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7487 if (kthread_should_stop())
7492 * this loop exits only if either when we are slower than
7493 * the 'hard' speed limit, or the system was IO-idle for
7495 * the system might be non-idle CPU-wise, but we only care
7496 * about not overloading the IO subsystem. (things like an
7497 * e2fsck being done on the RAID array should execute fast)
7501 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7502 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7504 if (currspeed
> speed_min(mddev
)) {
7505 if ((currspeed
> speed_max(mddev
)) ||
7506 !is_mddev_idle(mddev
, 0)) {
7512 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7514 * this also signals 'finished resyncing' to md_stop
7517 blk_finish_plug(&plug
);
7518 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7520 /* tell personality that we are finished */
7521 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7523 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7524 mddev
->curr_resync
> 2) {
7525 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7526 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7527 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7529 "md: checkpointing %s of %s.\n",
7530 desc
, mdname(mddev
));
7531 mddev
->recovery_cp
=
7532 mddev
->curr_resync_completed
;
7535 mddev
->recovery_cp
= MaxSector
;
7537 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7538 mddev
->curr_resync
= MaxSector
;
7540 rdev_for_each_rcu(rdev
, mddev
)
7541 if (rdev
->raid_disk
>= 0 &&
7542 mddev
->delta_disks
>= 0 &&
7543 !test_bit(Faulty
, &rdev
->flags
) &&
7544 !test_bit(In_sync
, &rdev
->flags
) &&
7545 rdev
->recovery_offset
< mddev
->curr_resync
)
7546 rdev
->recovery_offset
= mddev
->curr_resync
;
7551 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7553 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7554 /* We completed so min/max setting can be forgotten if used. */
7555 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7556 mddev
->resync_min
= 0;
7557 mddev
->resync_max
= MaxSector
;
7558 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7559 mddev
->resync_min
= mddev
->curr_resync_completed
;
7560 mddev
->curr_resync
= 0;
7561 wake_up(&resync_wait
);
7562 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7563 md_wakeup_thread(mddev
->thread
);
7568 * got a signal, exit.
7571 "md: md_do_sync() got signal ... exiting\n");
7572 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7576 EXPORT_SYMBOL_GPL(md_do_sync
);
7578 static int remove_and_add_spares(struct mddev
*mddev
)
7580 struct md_rdev
*rdev
;
7584 mddev
->curr_resync_completed
= 0;
7586 rdev_for_each(rdev
, mddev
)
7587 if (rdev
->raid_disk
>= 0 &&
7588 !test_bit(Blocked
, &rdev
->flags
) &&
7589 (test_bit(Faulty
, &rdev
->flags
) ||
7590 ! test_bit(In_sync
, &rdev
->flags
)) &&
7591 atomic_read(&rdev
->nr_pending
)==0) {
7592 if (mddev
->pers
->hot_remove_disk(
7593 mddev
, rdev
) == 0) {
7594 sysfs_unlink_rdev(mddev
, rdev
);
7595 rdev
->raid_disk
= -1;
7600 sysfs_notify(&mddev
->kobj
, NULL
,
7604 rdev_for_each(rdev
, mddev
) {
7605 if (rdev
->raid_disk
>= 0 &&
7606 !test_bit(In_sync
, &rdev
->flags
) &&
7607 !test_bit(Faulty
, &rdev
->flags
))
7609 if (rdev
->raid_disk
< 0
7610 && !test_bit(Faulty
, &rdev
->flags
)) {
7611 rdev
->recovery_offset
= 0;
7613 hot_add_disk(mddev
, rdev
) == 0) {
7614 if (sysfs_link_rdev(mddev
, rdev
))
7615 /* failure here is OK */;
7617 md_new_event(mddev
);
7618 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7623 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7627 static void reap_sync_thread(struct mddev
*mddev
)
7629 struct md_rdev
*rdev
;
7631 /* resync has finished, collect result */
7632 md_unregister_thread(&mddev
->sync_thread
);
7633 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7634 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7636 /* activate any spares */
7637 if (mddev
->pers
->spare_active(mddev
)) {
7638 sysfs_notify(&mddev
->kobj
, NULL
,
7640 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7643 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7644 mddev
->pers
->finish_reshape
)
7645 mddev
->pers
->finish_reshape(mddev
);
7647 /* If array is no-longer degraded, then any saved_raid_disk
7648 * information must be scrapped. Also if any device is now
7649 * In_sync we must scrape the saved_raid_disk for that device
7650 * do the superblock for an incrementally recovered device
7653 rdev_for_each(rdev
, mddev
)
7654 if (!mddev
->degraded
||
7655 test_bit(In_sync
, &rdev
->flags
))
7656 rdev
->saved_raid_disk
= -1;
7658 md_update_sb(mddev
, 1);
7659 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7660 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7661 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7662 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7663 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7664 /* flag recovery needed just to double check */
7665 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7666 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7667 md_new_event(mddev
);
7668 if (mddev
->event_work
.func
)
7669 queue_work(md_misc_wq
, &mddev
->event_work
);
7673 * This routine is regularly called by all per-raid-array threads to
7674 * deal with generic issues like resync and super-block update.
7675 * Raid personalities that don't have a thread (linear/raid0) do not
7676 * need this as they never do any recovery or update the superblock.
7678 * It does not do any resync itself, but rather "forks" off other threads
7679 * to do that as needed.
7680 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7681 * "->recovery" and create a thread at ->sync_thread.
7682 * When the thread finishes it sets MD_RECOVERY_DONE
7683 * and wakeups up this thread which will reap the thread and finish up.
7684 * This thread also removes any faulty devices (with nr_pending == 0).
7686 * The overall approach is:
7687 * 1/ if the superblock needs updating, update it.
7688 * 2/ If a recovery thread is running, don't do anything else.
7689 * 3/ If recovery has finished, clean up, possibly marking spares active.
7690 * 4/ If there are any faulty devices, remove them.
7691 * 5/ If array is degraded, try to add spares devices
7692 * 6/ If array has spares or is not in-sync, start a resync thread.
7694 void md_check_recovery(struct mddev
*mddev
)
7696 if (mddev
->suspended
)
7700 bitmap_daemon_work(mddev
);
7702 if (signal_pending(current
)) {
7703 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7704 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7706 mddev
->safemode
= 2;
7708 flush_signals(current
);
7711 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7714 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7715 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7716 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7717 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7718 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7719 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7723 if (mddev_trylock(mddev
)) {
7727 /* Only thing we do on a ro array is remove
7730 struct md_rdev
*rdev
;
7731 rdev_for_each(rdev
, mddev
)
7732 if (rdev
->raid_disk
>= 0 &&
7733 !test_bit(Blocked
, &rdev
->flags
) &&
7734 test_bit(Faulty
, &rdev
->flags
) &&
7735 atomic_read(&rdev
->nr_pending
)==0) {
7736 if (mddev
->pers
->hot_remove_disk(
7737 mddev
, rdev
) == 0) {
7738 sysfs_unlink_rdev(mddev
, rdev
);
7739 rdev
->raid_disk
= -1;
7742 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7746 if (!mddev
->external
) {
7748 spin_lock_irq(&mddev
->write_lock
);
7749 if (mddev
->safemode
&&
7750 !atomic_read(&mddev
->writes_pending
) &&
7752 mddev
->recovery_cp
== MaxSector
) {
7755 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7757 if (mddev
->safemode
== 1)
7758 mddev
->safemode
= 0;
7759 spin_unlock_irq(&mddev
->write_lock
);
7761 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7765 md_update_sb(mddev
, 0);
7767 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7768 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7769 /* resync/recovery still happening */
7770 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7773 if (mddev
->sync_thread
) {
7774 reap_sync_thread(mddev
);
7777 /* Set RUNNING before clearing NEEDED to avoid
7778 * any transients in the value of "sync_action".
7780 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7781 /* Clear some bits that don't mean anything, but
7784 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7785 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7787 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7788 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7790 /* no recovery is running.
7791 * remove any failed drives, then
7792 * add spares if possible.
7793 * Spare are also removed and re-added, to allow
7794 * the personality to fail the re-add.
7797 if (mddev
->reshape_position
!= MaxSector
) {
7798 if (mddev
->pers
->check_reshape
== NULL
||
7799 mddev
->pers
->check_reshape(mddev
) != 0)
7800 /* Cannot proceed */
7802 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7803 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7804 } else if ((spares
= remove_and_add_spares(mddev
))) {
7805 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7806 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7807 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7808 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7809 } else if (mddev
->recovery_cp
< MaxSector
) {
7810 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7811 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7812 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7813 /* nothing to be done ... */
7816 if (mddev
->pers
->sync_request
) {
7818 /* We are adding a device or devices to an array
7819 * which has the bitmap stored on all devices.
7820 * So make sure all bitmap pages get written
7822 bitmap_write_all(mddev
->bitmap
);
7824 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7827 if (!mddev
->sync_thread
) {
7828 printk(KERN_ERR
"%s: could not start resync"
7831 /* leave the spares where they are, it shouldn't hurt */
7832 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7833 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7834 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7835 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7836 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7838 md_wakeup_thread(mddev
->sync_thread
);
7839 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7840 md_new_event(mddev
);
7843 if (!mddev
->sync_thread
) {
7844 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7845 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7847 if (mddev
->sysfs_action
)
7848 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7850 mddev_unlock(mddev
);
7854 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7856 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7857 wait_event_timeout(rdev
->blocked_wait
,
7858 !test_bit(Blocked
, &rdev
->flags
) &&
7859 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7860 msecs_to_jiffies(5000));
7861 rdev_dec_pending(rdev
, mddev
);
7863 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7865 void md_finish_reshape(struct mddev
*mddev
)
7867 /* called be personality module when reshape completes. */
7868 struct md_rdev
*rdev
;
7870 rdev_for_each(rdev
, mddev
) {
7871 if (rdev
->data_offset
> rdev
->new_data_offset
)
7872 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
7874 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
7875 rdev
->data_offset
= rdev
->new_data_offset
;
7878 EXPORT_SYMBOL(md_finish_reshape
);
7880 /* Bad block management.
7881 * We can record which blocks on each device are 'bad' and so just
7882 * fail those blocks, or that stripe, rather than the whole device.
7883 * Entries in the bad-block table are 64bits wide. This comprises:
7884 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7885 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7886 * A 'shift' can be set so that larger blocks are tracked and
7887 * consequently larger devices can be covered.
7888 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7890 * Locking of the bad-block table uses a seqlock so md_is_badblock
7891 * might need to retry if it is very unlucky.
7892 * We will sometimes want to check for bad blocks in a bi_end_io function,
7893 * so we use the write_seqlock_irq variant.
7895 * When looking for a bad block we specify a range and want to
7896 * know if any block in the range is bad. So we binary-search
7897 * to the last range that starts at-or-before the given endpoint,
7898 * (or "before the sector after the target range")
7899 * then see if it ends after the given start.
7901 * 0 if there are no known bad blocks in the range
7902 * 1 if there are known bad block which are all acknowledged
7903 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7904 * plus the start/length of the first bad section we overlap.
7906 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7907 sector_t
*first_bad
, int *bad_sectors
)
7913 sector_t target
= s
+ sectors
;
7916 if (bb
->shift
> 0) {
7917 /* round the start down, and the end up */
7919 target
+= (1<<bb
->shift
) - 1;
7920 target
>>= bb
->shift
;
7921 sectors
= target
- s
;
7923 /* 'target' is now the first block after the bad range */
7926 seq
= read_seqbegin(&bb
->lock
);
7930 /* Binary search between lo and hi for 'target'
7931 * i.e. for the last range that starts before 'target'
7933 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7934 * are known not to be the last range before target.
7935 * VARIANT: hi-lo is the number of possible
7936 * ranges, and decreases until it reaches 1
7938 while (hi
- lo
> 1) {
7939 int mid
= (lo
+ hi
) / 2;
7940 sector_t a
= BB_OFFSET(p
[mid
]);
7942 /* This could still be the one, earlier ranges
7946 /* This and later ranges are definitely out. */
7949 /* 'lo' might be the last that started before target, but 'hi' isn't */
7951 /* need to check all range that end after 's' to see if
7952 * any are unacknowledged.
7955 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7956 if (BB_OFFSET(p
[lo
]) < target
) {
7957 /* starts before the end, and finishes after
7958 * the start, so they must overlap
7960 if (rv
!= -1 && BB_ACK(p
[lo
]))
7964 *first_bad
= BB_OFFSET(p
[lo
]);
7965 *bad_sectors
= BB_LEN(p
[lo
]);
7971 if (read_seqretry(&bb
->lock
, seq
))
7976 EXPORT_SYMBOL_GPL(md_is_badblock
);
7979 * Add a range of bad blocks to the table.
7980 * This might extend the table, or might contract it
7981 * if two adjacent ranges can be merged.
7982 * We binary-search to find the 'insertion' point, then
7983 * decide how best to handle it.
7985 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7993 /* badblocks are disabled */
7997 /* round the start down, and the end up */
7998 sector_t next
= s
+ sectors
;
8000 next
+= (1<<bb
->shift
) - 1;
8005 write_seqlock_irq(&bb
->lock
);
8010 /* Find the last range that starts at-or-before 's' */
8011 while (hi
- lo
> 1) {
8012 int mid
= (lo
+ hi
) / 2;
8013 sector_t a
= BB_OFFSET(p
[mid
]);
8019 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
8023 /* we found a range that might merge with the start
8026 sector_t a
= BB_OFFSET(p
[lo
]);
8027 sector_t e
= a
+ BB_LEN(p
[lo
]);
8028 int ack
= BB_ACK(p
[lo
]);
8030 /* Yes, we can merge with a previous range */
8031 if (s
== a
&& s
+ sectors
>= e
)
8032 /* new range covers old */
8035 ack
= ack
&& acknowledged
;
8037 if (e
< s
+ sectors
)
8039 if (e
- a
<= BB_MAX_LEN
) {
8040 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
8043 /* does not all fit in one range,
8044 * make p[lo] maximal
8046 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
8047 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8053 if (sectors
&& hi
< bb
->count
) {
8054 /* 'hi' points to the first range that starts after 's'.
8055 * Maybe we can merge with the start of that range */
8056 sector_t a
= BB_OFFSET(p
[hi
]);
8057 sector_t e
= a
+ BB_LEN(p
[hi
]);
8058 int ack
= BB_ACK(p
[hi
]);
8059 if (a
<= s
+ sectors
) {
8060 /* merging is possible */
8061 if (e
<= s
+ sectors
) {
8066 ack
= ack
&& acknowledged
;
8069 if (e
- a
<= BB_MAX_LEN
) {
8070 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
8073 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
8081 if (sectors
== 0 && hi
< bb
->count
) {
8082 /* we might be able to combine lo and hi */
8083 /* Note: 's' is at the end of 'lo' */
8084 sector_t a
= BB_OFFSET(p
[hi
]);
8085 int lolen
= BB_LEN(p
[lo
]);
8086 int hilen
= BB_LEN(p
[hi
]);
8087 int newlen
= lolen
+ hilen
- (s
- a
);
8088 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
8089 /* yes, we can combine them */
8090 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
8091 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
8092 memmove(p
+ hi
, p
+ hi
+ 1,
8093 (bb
->count
- hi
- 1) * 8);
8098 /* didn't merge (it all).
8099 * Need to add a range just before 'hi' */
8100 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8101 /* No room for more */
8105 int this_sectors
= sectors
;
8106 memmove(p
+ hi
+ 1, p
+ hi
,
8107 (bb
->count
- hi
) * 8);
8110 if (this_sectors
> BB_MAX_LEN
)
8111 this_sectors
= BB_MAX_LEN
;
8112 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
8113 sectors
-= this_sectors
;
8120 bb
->unacked_exist
= 1;
8121 write_sequnlock_irq(&bb
->lock
);
8126 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8131 s
+= rdev
->new_data_offset
;
8133 s
+= rdev
->data_offset
;
8134 rv
= md_set_badblocks(&rdev
->badblocks
,
8137 /* Make sure they get written out promptly */
8138 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8139 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
8140 md_wakeup_thread(rdev
->mddev
->thread
);
8144 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8147 * Remove a range of bad blocks from the table.
8148 * This may involve extending the table if we spilt a region,
8149 * but it must not fail. So if the table becomes full, we just
8150 * drop the remove request.
8152 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
8156 sector_t target
= s
+ sectors
;
8159 if (bb
->shift
> 0) {
8160 /* When clearing we round the start up and the end down.
8161 * This should not matter as the shift should align with
8162 * the block size and no rounding should ever be needed.
8163 * However it is better the think a block is bad when it
8164 * isn't than to think a block is not bad when it is.
8166 s
+= (1<<bb
->shift
) - 1;
8168 target
>>= bb
->shift
;
8169 sectors
= target
- s
;
8172 write_seqlock_irq(&bb
->lock
);
8177 /* Find the last range that starts before 'target' */
8178 while (hi
- lo
> 1) {
8179 int mid
= (lo
+ hi
) / 2;
8180 sector_t a
= BB_OFFSET(p
[mid
]);
8187 /* p[lo] is the last range that could overlap the
8188 * current range. Earlier ranges could also overlap,
8189 * but only this one can overlap the end of the range.
8191 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
8192 /* Partial overlap, leave the tail of this range */
8193 int ack
= BB_ACK(p
[lo
]);
8194 sector_t a
= BB_OFFSET(p
[lo
]);
8195 sector_t end
= a
+ BB_LEN(p
[lo
]);
8198 /* we need to split this range */
8199 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
8203 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
8205 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
8208 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
8209 /* there is no longer an overlap */
8214 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
8215 /* This range does overlap */
8216 if (BB_OFFSET(p
[lo
]) < s
) {
8217 /* Keep the early parts of this range. */
8218 int ack
= BB_ACK(p
[lo
]);
8219 sector_t start
= BB_OFFSET(p
[lo
]);
8220 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
8221 /* now low doesn't overlap, so.. */
8226 /* 'lo' is strictly before, 'hi' is strictly after,
8227 * anything between needs to be discarded
8230 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
8231 bb
->count
-= (hi
- lo
- 1);
8237 write_sequnlock_irq(&bb
->lock
);
8241 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8245 s
+= rdev
->new_data_offset
;
8247 s
+= rdev
->data_offset
;
8248 return md_clear_badblocks(&rdev
->badblocks
,
8251 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8254 * Acknowledge all bad blocks in a list.
8255 * This only succeeds if ->changed is clear. It is used by
8256 * in-kernel metadata updates
8258 void md_ack_all_badblocks(struct badblocks
*bb
)
8260 if (bb
->page
== NULL
|| bb
->changed
)
8261 /* no point even trying */
8263 write_seqlock_irq(&bb
->lock
);
8265 if (bb
->changed
== 0 && bb
->unacked_exist
) {
8268 for (i
= 0; i
< bb
->count
; i
++) {
8269 if (!BB_ACK(p
[i
])) {
8270 sector_t start
= BB_OFFSET(p
[i
]);
8271 int len
= BB_LEN(p
[i
]);
8272 p
[i
] = BB_MAKE(start
, len
, 1);
8275 bb
->unacked_exist
= 0;
8277 write_sequnlock_irq(&bb
->lock
);
8279 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
8281 /* sysfs access to bad-blocks list.
8282 * We present two files.
8283 * 'bad-blocks' lists sector numbers and lengths of ranges that
8284 * are recorded as bad. The list is truncated to fit within
8285 * the one-page limit of sysfs.
8286 * Writing "sector length" to this file adds an acknowledged
8288 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8289 * been acknowledged. Writing to this file adds bad blocks
8290 * without acknowledging them. This is largely for testing.
8294 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8305 seq
= read_seqbegin(&bb
->lock
);
8310 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8311 sector_t s
= BB_OFFSET(p
[i
]);
8312 unsigned int length
= BB_LEN(p
[i
]);
8313 int ack
= BB_ACK(p
[i
]);
8319 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8320 (unsigned long long)s
<< bb
->shift
,
8321 length
<< bb
->shift
);
8323 if (unack
&& len
== 0)
8324 bb
->unacked_exist
= 0;
8326 if (read_seqretry(&bb
->lock
, seq
))
8335 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8337 unsigned long long sector
;
8341 /* Allow clearing via sysfs *only* for testing/debugging.
8342 * Normally only a successful write may clear a badblock
8345 if (page
[0] == '-') {
8349 #endif /* DO_DEBUG */
8351 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8353 if (newline
!= '\n')
8365 md_clear_badblocks(bb
, sector
, length
);
8368 #endif /* DO_DEBUG */
8369 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8375 static int md_notify_reboot(struct notifier_block
*this,
8376 unsigned long code
, void *x
)
8378 struct list_head
*tmp
;
8379 struct mddev
*mddev
;
8382 for_each_mddev(mddev
, tmp
) {
8383 if (mddev_trylock(mddev
)) {
8385 __md_stop_writes(mddev
);
8386 mddev
->safemode
= 2;
8387 mddev_unlock(mddev
);
8392 * certain more exotic SCSI devices are known to be
8393 * volatile wrt too early system reboots. While the
8394 * right place to handle this issue is the given
8395 * driver, we do want to have a safe RAID driver ...
8403 static struct notifier_block md_notifier
= {
8404 .notifier_call
= md_notify_reboot
,
8406 .priority
= INT_MAX
, /* before any real devices */
8409 static void md_geninit(void)
8411 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8413 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8416 static int __init
md_init(void)
8420 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8424 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8428 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8431 if ((ret
= register_blkdev(0, "mdp")) < 0)
8435 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8436 md_probe
, NULL
, NULL
);
8437 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8438 md_probe
, NULL
, NULL
);
8440 register_reboot_notifier(&md_notifier
);
8441 raid_table_header
= register_sysctl_table(raid_root_table
);
8447 unregister_blkdev(MD_MAJOR
, "md");
8449 destroy_workqueue(md_misc_wq
);
8451 destroy_workqueue(md_wq
);
8459 * Searches all registered partitions for autorun RAID arrays
8463 static LIST_HEAD(all_detected_devices
);
8464 struct detected_devices_node
{
8465 struct list_head list
;
8469 void md_autodetect_dev(dev_t dev
)
8471 struct detected_devices_node
*node_detected_dev
;
8473 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8474 if (node_detected_dev
) {
8475 node_detected_dev
->dev
= dev
;
8476 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8478 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8479 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8484 static void autostart_arrays(int part
)
8486 struct md_rdev
*rdev
;
8487 struct detected_devices_node
*node_detected_dev
;
8489 int i_scanned
, i_passed
;
8494 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8496 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8498 node_detected_dev
= list_entry(all_detected_devices
.next
,
8499 struct detected_devices_node
, list
);
8500 list_del(&node_detected_dev
->list
);
8501 dev
= node_detected_dev
->dev
;
8502 kfree(node_detected_dev
);
8503 rdev
= md_import_device(dev
,0, 90);
8507 if (test_bit(Faulty
, &rdev
->flags
)) {
8511 set_bit(AutoDetected
, &rdev
->flags
);
8512 list_add(&rdev
->same_set
, &pending_raid_disks
);
8516 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8517 i_scanned
, i_passed
);
8519 autorun_devices(part
);
8522 #endif /* !MODULE */
8524 static __exit
void md_exit(void)
8526 struct mddev
*mddev
;
8527 struct list_head
*tmp
;
8529 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8530 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8532 unregister_blkdev(MD_MAJOR
,"md");
8533 unregister_blkdev(mdp_major
, "mdp");
8534 unregister_reboot_notifier(&md_notifier
);
8535 unregister_sysctl_table(raid_table_header
);
8536 remove_proc_entry("mdstat", NULL
);
8537 for_each_mddev(mddev
, tmp
) {
8538 export_array(mddev
);
8539 mddev
->hold_active
= 0;
8541 destroy_workqueue(md_misc_wq
);
8542 destroy_workqueue(md_wq
);
8545 subsys_initcall(md_init
);
8546 module_exit(md_exit
)
8548 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8550 return sprintf(buffer
, "%d", start_readonly
);
8552 static int set_ro(const char *val
, struct kernel_param
*kp
)
8555 int num
= simple_strtoul(val
, &e
, 10);
8556 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8557 start_readonly
= num
;
8563 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8564 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8566 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8568 EXPORT_SYMBOL(register_md_personality
);
8569 EXPORT_SYMBOL(unregister_md_personality
);
8570 EXPORT_SYMBOL(md_error
);
8571 EXPORT_SYMBOL(md_done_sync
);
8572 EXPORT_SYMBOL(md_write_start
);
8573 EXPORT_SYMBOL(md_write_end
);
8574 EXPORT_SYMBOL(md_register_thread
);
8575 EXPORT_SYMBOL(md_unregister_thread
);
8576 EXPORT_SYMBOL(md_wakeup_thread
);
8577 EXPORT_SYMBOL(md_check_recovery
);
8578 MODULE_LICENSE("GPL");
8579 MODULE_DESCRIPTION("MD RAID framework");
8581 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);