2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part
);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list
);
67 static DEFINE_SPINLOCK(pers_lock
);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
72 static struct workqueue_struct
*md_wq
;
73 static struct workqueue_struct
*md_misc_wq
;
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min
= 1000;
97 static int sysctl_speed_limit_max
= 200000;
98 static inline int speed_min(struct mddev
*mddev
)
100 return mddev
->sync_speed_min
?
101 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
104 static inline int speed_max(struct mddev
*mddev
)
106 return mddev
->sync_speed_max
?
107 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
110 static struct ctl_table_header
*raid_table_header
;
112 static ctl_table raid_table
[] = {
114 .procname
= "speed_limit_min",
115 .data
= &sysctl_speed_limit_min
,
116 .maxlen
= sizeof(int),
117 .mode
= S_IRUGO
|S_IWUSR
,
118 .proc_handler
= proc_dointvec
,
121 .procname
= "speed_limit_max",
122 .data
= &sysctl_speed_limit_max
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
130 static ctl_table raid_dir_table
[] = {
134 .mode
= S_IRUGO
|S_IXUGO
,
140 static ctl_table raid_root_table
[] = {
145 .child
= raid_dir_table
,
150 static const struct block_device_operations md_fops
;
152 static int start_readonly
;
155 * like bio_clone, but with a local bio set
158 static void mddev_bio_destructor(struct bio
*bio
)
160 struct mddev
*mddev
, **mddevp
;
165 bio_free(bio
, mddev
->bio_set
);
168 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
172 struct mddev
**mddevp
;
174 if (!mddev
|| !mddev
->bio_set
)
175 return bio_alloc(gfp_mask
, nr_iovecs
);
177 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
183 b
->bi_destructor
= mddev_bio_destructor
;
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
188 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
192 struct mddev
**mddevp
;
194 if (!mddev
|| !mddev
->bio_set
)
195 return bio_clone(bio
, gfp_mask
);
197 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
203 b
->bi_destructor
= mddev_bio_destructor
;
205 if (bio_integrity(bio
)) {
208 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
218 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
220 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
227 struct bio_vec
*bvec
;
231 if (offset
== 0 && size
== bio
->bi_size
)
234 bio
->bi_sector
+= offset
;
237 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
239 while (bio
->bi_idx
< bio
->bi_vcnt
&&
240 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
241 /* remove this whole bio_vec */
242 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
245 if (bio
->bi_idx
< bio
->bi_vcnt
) {
246 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
247 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
249 /* avoid any complications with bi_idx being non-zero*/
251 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
252 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
253 bio
->bi_vcnt
-= bio
->bi_idx
;
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec
, bio
, i
) {
258 if (sofar
+ bvec
->bv_len
> size
)
259 bvec
->bv_len
= size
- sofar
;
260 if (bvec
->bv_len
== 0) {
264 sofar
+= bvec
->bv_len
;
267 EXPORT_SYMBOL_GPL(md_trim_bio
);
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
280 static atomic_t md_event_count
;
281 void md_new_event(struct mddev
*mddev
)
283 atomic_inc(&md_event_count
);
284 wake_up(&md_event_waiters
);
286 EXPORT_SYMBOL_GPL(md_new_event
);
288 /* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
291 static void md_new_event_inintr(struct mddev
*mddev
)
293 atomic_inc(&md_event_count
);
294 wake_up(&md_event_waiters
);
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
301 static LIST_HEAD(all_mddevs
);
302 static DEFINE_SPINLOCK(all_mddevs_lock
);
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
312 #define for_each_mddev(_mddev,_tmp) \
314 for (({ spin_lock(&all_mddevs_lock); \
315 _tmp = all_mddevs.next; \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319 spin_unlock(&all_mddevs_lock); \
320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
323 ({ spin_lock(&all_mddevs_lock); \
324 _tmp = _tmp->next;}) \
328 /* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 static 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 EXPORT_SYMBOL_GPL(mddev_suspend
);
397 void mddev_resume(struct mddev
*mddev
)
399 mddev
->suspended
= 0;
400 wake_up(&mddev
->sb_wait
);
401 mddev
->pers
->quiesce(mddev
, 0);
403 md_wakeup_thread(mddev
->thread
);
404 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
406 EXPORT_SYMBOL_GPL(mddev_resume
);
408 int mddev_congested(struct mddev
*mddev
, int bits
)
410 return mddev
->suspended
;
412 EXPORT_SYMBOL(mddev_congested
);
415 * Generic flush handling for md
418 static void md_end_flush(struct bio
*bio
, int err
)
420 struct md_rdev
*rdev
= bio
->bi_private
;
421 struct mddev
*mddev
= rdev
->mddev
;
423 rdev_dec_pending(rdev
, mddev
);
425 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
426 /* The pre-request flush has finished */
427 queue_work(md_wq
, &mddev
->flush_work
);
432 static void md_submit_flush_data(struct work_struct
*ws
);
434 static void submit_flushes(struct work_struct
*ws
)
436 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
437 struct md_rdev
*rdev
;
439 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
440 atomic_set(&mddev
->flush_pending
, 1);
442 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
443 if (rdev
->raid_disk
>= 0 &&
444 !test_bit(Faulty
, &rdev
->flags
)) {
445 /* Take two references, one is dropped
446 * when request finishes, one after
447 * we reclaim rcu_read_lock
450 atomic_inc(&rdev
->nr_pending
);
451 atomic_inc(&rdev
->nr_pending
);
453 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
454 bi
->bi_end_io
= md_end_flush
;
455 bi
->bi_private
= rdev
;
456 bi
->bi_bdev
= rdev
->bdev
;
457 atomic_inc(&mddev
->flush_pending
);
458 submit_bio(WRITE_FLUSH
, bi
);
460 rdev_dec_pending(rdev
, mddev
);
463 if (atomic_dec_and_test(&mddev
->flush_pending
))
464 queue_work(md_wq
, &mddev
->flush_work
);
467 static void md_submit_flush_data(struct work_struct
*ws
)
469 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
470 struct bio
*bio
= mddev
->flush_bio
;
472 if (bio
->bi_size
== 0)
473 /* an empty barrier - all done */
476 bio
->bi_rw
&= ~REQ_FLUSH
;
477 mddev
->pers
->make_request(mddev
, bio
);
480 mddev
->flush_bio
= NULL
;
481 wake_up(&mddev
->sb_wait
);
484 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
486 spin_lock_irq(&mddev
->write_lock
);
487 wait_event_lock_irq(mddev
->sb_wait
,
489 mddev
->write_lock
, /*nothing*/);
490 mddev
->flush_bio
= bio
;
491 spin_unlock_irq(&mddev
->write_lock
);
493 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
494 queue_work(md_wq
, &mddev
->flush_work
);
496 EXPORT_SYMBOL(md_flush_request
);
498 /* Support for plugging.
499 * This mirrors the plugging support in request_queue, but does not
500 * require having a whole queue or request structures.
501 * We allocate an md_plug_cb for each md device and each thread it gets
502 * plugged on. This links tot the private plug_handle structure in the
503 * personality data where we keep a count of the number of outstanding
504 * plugs so other code can see if a plug is active.
507 struct blk_plug_cb cb
;
511 static void plugger_unplug(struct blk_plug_cb
*cb
)
513 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
514 if (atomic_dec_and_test(&mdcb
->mddev
->plug_cnt
))
515 md_wakeup_thread(mdcb
->mddev
->thread
);
519 /* Check that an unplug wakeup will come shortly.
520 * If not, wakeup the md thread immediately
522 int mddev_check_plugged(struct mddev
*mddev
)
524 struct blk_plug
*plug
= current
->plug
;
525 struct md_plug_cb
*mdcb
;
530 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
531 if (mdcb
->cb
.callback
== plugger_unplug
&&
532 mdcb
->mddev
== mddev
) {
533 /* Already on the list, move to top */
534 if (mdcb
!= list_first_entry(&plug
->cb_list
,
537 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
541 /* Not currently on the callback list */
542 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
547 mdcb
->cb
.callback
= plugger_unplug
;
548 atomic_inc(&mddev
->plug_cnt
);
549 list_add(&mdcb
->cb
.list
, &plug
->cb_list
);
552 EXPORT_SYMBOL_GPL(mddev_check_plugged
);
554 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
556 atomic_inc(&mddev
->active
);
560 static void mddev_delayed_delete(struct work_struct
*ws
);
562 static void mddev_put(struct mddev
*mddev
)
564 struct bio_set
*bs
= NULL
;
566 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
568 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
569 mddev
->ctime
== 0 && !mddev
->hold_active
) {
570 /* Array is not configured at all, and not held active,
572 list_del(&mddev
->all_mddevs
);
574 mddev
->bio_set
= NULL
;
575 if (mddev
->gendisk
) {
576 /* We did a probe so need to clean up. Call
577 * queue_work inside the spinlock so that
578 * flush_workqueue() after mddev_find will
579 * succeed in waiting for the work to be done.
581 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
582 queue_work(md_misc_wq
, &mddev
->del_work
);
586 spin_unlock(&all_mddevs_lock
);
591 void mddev_init(struct mddev
*mddev
)
593 mutex_init(&mddev
->open_mutex
);
594 mutex_init(&mddev
->reconfig_mutex
);
595 mutex_init(&mddev
->bitmap_info
.mutex
);
596 INIT_LIST_HEAD(&mddev
->disks
);
597 INIT_LIST_HEAD(&mddev
->all_mddevs
);
598 init_timer(&mddev
->safemode_timer
);
599 atomic_set(&mddev
->active
, 1);
600 atomic_set(&mddev
->openers
, 0);
601 atomic_set(&mddev
->active_io
, 0);
602 atomic_set(&mddev
->plug_cnt
, 0);
603 spin_lock_init(&mddev
->write_lock
);
604 atomic_set(&mddev
->flush_pending
, 0);
605 init_waitqueue_head(&mddev
->sb_wait
);
606 init_waitqueue_head(&mddev
->recovery_wait
);
607 mddev
->reshape_position
= MaxSector
;
608 mddev
->resync_min
= 0;
609 mddev
->resync_max
= MaxSector
;
610 mddev
->level
= LEVEL_NONE
;
612 EXPORT_SYMBOL_GPL(mddev_init
);
614 static struct mddev
* mddev_find(dev_t unit
)
616 struct mddev
*mddev
, *new = NULL
;
618 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
619 unit
&= ~((1<<MdpMinorShift
)-1);
622 spin_lock(&all_mddevs_lock
);
625 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
626 if (mddev
->unit
== unit
) {
628 spin_unlock(&all_mddevs_lock
);
634 list_add(&new->all_mddevs
, &all_mddevs
);
635 spin_unlock(&all_mddevs_lock
);
636 new->hold_active
= UNTIL_IOCTL
;
640 /* find an unused unit number */
641 static int next_minor
= 512;
642 int start
= next_minor
;
646 dev
= MKDEV(MD_MAJOR
, next_minor
);
648 if (next_minor
> MINORMASK
)
650 if (next_minor
== start
) {
651 /* Oh dear, all in use. */
652 spin_unlock(&all_mddevs_lock
);
658 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
659 if (mddev
->unit
== dev
) {
665 new->md_minor
= MINOR(dev
);
666 new->hold_active
= UNTIL_STOP
;
667 list_add(&new->all_mddevs
, &all_mddevs
);
668 spin_unlock(&all_mddevs_lock
);
671 spin_unlock(&all_mddevs_lock
);
673 new = kzalloc(sizeof(*new), GFP_KERNEL
);
678 if (MAJOR(unit
) == MD_MAJOR
)
679 new->md_minor
= MINOR(unit
);
681 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
688 static inline int mddev_lock(struct mddev
* mddev
)
690 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
693 static inline int mddev_is_locked(struct mddev
*mddev
)
695 return mutex_is_locked(&mddev
->reconfig_mutex
);
698 static inline int mddev_trylock(struct mddev
* mddev
)
700 return mutex_trylock(&mddev
->reconfig_mutex
);
703 static struct attribute_group md_redundancy_group
;
705 static void mddev_unlock(struct mddev
* mddev
)
707 if (mddev
->to_remove
) {
708 /* These cannot be removed under reconfig_mutex as
709 * an access to the files will try to take reconfig_mutex
710 * while holding the file unremovable, which leads to
712 * So hold set sysfs_active while the remove in happeing,
713 * and anything else which might set ->to_remove or my
714 * otherwise change the sysfs namespace will fail with
715 * -EBUSY if sysfs_active is still set.
716 * We set sysfs_active under reconfig_mutex and elsewhere
717 * test it under the same mutex to ensure its correct value
720 struct attribute_group
*to_remove
= mddev
->to_remove
;
721 mddev
->to_remove
= NULL
;
722 mddev
->sysfs_active
= 1;
723 mutex_unlock(&mddev
->reconfig_mutex
);
725 if (mddev
->kobj
.sd
) {
726 if (to_remove
!= &md_redundancy_group
)
727 sysfs_remove_group(&mddev
->kobj
, to_remove
);
728 if (mddev
->pers
== NULL
||
729 mddev
->pers
->sync_request
== NULL
) {
730 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
731 if (mddev
->sysfs_action
)
732 sysfs_put(mddev
->sysfs_action
);
733 mddev
->sysfs_action
= NULL
;
736 mddev
->sysfs_active
= 0;
738 mutex_unlock(&mddev
->reconfig_mutex
);
740 /* As we've dropped the mutex we need a spinlock to
741 * make sure the thread doesn't disappear
743 spin_lock(&pers_lock
);
744 md_wakeup_thread(mddev
->thread
);
745 spin_unlock(&pers_lock
);
748 static struct md_rdev
* find_rdev_nr(struct mddev
*mddev
, int nr
)
750 struct md_rdev
*rdev
;
752 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
753 if (rdev
->desc_nr
== nr
)
759 static struct md_rdev
* find_rdev(struct mddev
* mddev
, dev_t dev
)
761 struct md_rdev
*rdev
;
763 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
764 if (rdev
->bdev
->bd_dev
== dev
)
770 static struct md_personality
*find_pers(int level
, char *clevel
)
772 struct md_personality
*pers
;
773 list_for_each_entry(pers
, &pers_list
, list
) {
774 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
776 if (strcmp(pers
->name
, clevel
)==0)
782 /* return the offset of the super block in 512byte sectors */
783 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
785 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
786 return MD_NEW_SIZE_SECTORS(num_sectors
);
789 static int alloc_disk_sb(struct md_rdev
* rdev
)
794 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
795 if (!rdev
->sb_page
) {
796 printk(KERN_ALERT
"md: out of memory.\n");
803 static void free_disk_sb(struct md_rdev
* rdev
)
806 put_page(rdev
->sb_page
);
808 rdev
->sb_page
= NULL
;
813 put_page(rdev
->bb_page
);
814 rdev
->bb_page
= NULL
;
819 static void super_written(struct bio
*bio
, int error
)
821 struct md_rdev
*rdev
= bio
->bi_private
;
822 struct mddev
*mddev
= rdev
->mddev
;
824 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
825 printk("md: super_written gets error=%d, uptodate=%d\n",
826 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
827 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
828 md_error(mddev
, rdev
);
831 if (atomic_dec_and_test(&mddev
->pending_writes
))
832 wake_up(&mddev
->sb_wait
);
836 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
837 sector_t sector
, int size
, struct page
*page
)
839 /* write first size bytes of page to sector of rdev
840 * Increment mddev->pending_writes before returning
841 * and decrement it on completion, waking up sb_wait
842 * if zero is reached.
843 * If an error occurred, call md_error
845 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
847 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
848 bio
->bi_sector
= sector
;
849 bio_add_page(bio
, page
, size
, 0);
850 bio
->bi_private
= rdev
;
851 bio
->bi_end_io
= super_written
;
853 atomic_inc(&mddev
->pending_writes
);
854 submit_bio(WRITE_FLUSH_FUA
, bio
);
857 void md_super_wait(struct mddev
*mddev
)
859 /* wait for all superblock writes that were scheduled to complete */
862 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
863 if (atomic_read(&mddev
->pending_writes
)==0)
867 finish_wait(&mddev
->sb_wait
, &wq
);
870 static void bi_complete(struct bio
*bio
, int error
)
872 complete((struct completion
*)bio
->bi_private
);
875 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
876 struct page
*page
, int rw
, bool metadata_op
)
878 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
879 struct completion event
;
884 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
885 rdev
->meta_bdev
: rdev
->bdev
;
887 bio
->bi_sector
= sector
+ rdev
->sb_start
;
889 bio
->bi_sector
= sector
+ rdev
->data_offset
;
890 bio_add_page(bio
, page
, size
, 0);
891 init_completion(&event
);
892 bio
->bi_private
= &event
;
893 bio
->bi_end_io
= bi_complete
;
895 wait_for_completion(&event
);
897 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
901 EXPORT_SYMBOL_GPL(sync_page_io
);
903 static int read_disk_sb(struct md_rdev
* rdev
, int size
)
905 char b
[BDEVNAME_SIZE
];
906 if (!rdev
->sb_page
) {
914 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
920 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
921 bdevname(rdev
->bdev
,b
));
925 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
927 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
928 sb1
->set_uuid1
== sb2
->set_uuid1
&&
929 sb1
->set_uuid2
== sb2
->set_uuid2
&&
930 sb1
->set_uuid3
== sb2
->set_uuid3
;
933 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
936 mdp_super_t
*tmp1
, *tmp2
;
938 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
939 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
941 if (!tmp1
|| !tmp2
) {
943 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
951 * nr_disks is not constant
956 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
964 static u32
md_csum_fold(u32 csum
)
966 csum
= (csum
& 0xffff) + (csum
>> 16);
967 return (csum
& 0xffff) + (csum
>> 16);
970 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
973 u32
*sb32
= (u32
*)sb
;
975 unsigned int disk_csum
, csum
;
977 disk_csum
= sb
->sb_csum
;
980 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
982 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
986 /* This used to use csum_partial, which was wrong for several
987 * reasons including that different results are returned on
988 * different architectures. It isn't critical that we get exactly
989 * the same return value as before (we always csum_fold before
990 * testing, and that removes any differences). However as we
991 * know that csum_partial always returned a 16bit value on
992 * alphas, do a fold to maximise conformity to previous behaviour.
994 sb
->sb_csum
= md_csum_fold(disk_csum
);
996 sb
->sb_csum
= disk_csum
;
1003 * Handle superblock details.
1004 * We want to be able to handle multiple superblock formats
1005 * so we have a common interface to them all, and an array of
1006 * different handlers.
1007 * We rely on user-space to write the initial superblock, and support
1008 * reading and updating of superblocks.
1009 * Interface methods are:
1010 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1011 * loads and validates a superblock on dev.
1012 * if refdev != NULL, compare superblocks on both devices
1014 * 0 - dev has a superblock that is compatible with refdev
1015 * 1 - dev has a superblock that is compatible and newer than refdev
1016 * so dev should be used as the refdev in future
1017 * -EINVAL superblock incompatible or invalid
1018 * -othererror e.g. -EIO
1020 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1021 * Verify that dev is acceptable into mddev.
1022 * The first time, mddev->raid_disks will be 0, and data from
1023 * dev should be merged in. Subsequent calls check that dev
1024 * is new enough. Return 0 or -EINVAL
1026 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1027 * Update the superblock for rdev with data in mddev
1028 * This does not write to disc.
1034 struct module
*owner
;
1035 int (*load_super
)(struct md_rdev
*rdev
, struct md_rdev
*refdev
,
1037 int (*validate_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1038 void (*sync_super
)(struct mddev
*mddev
, struct md_rdev
*rdev
);
1039 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
1040 sector_t num_sectors
);
1044 * Check that the given mddev has no bitmap.
1046 * This function is called from the run method of all personalities that do not
1047 * support bitmaps. It prints an error message and returns non-zero if mddev
1048 * has a bitmap. Otherwise, it returns 0.
1051 int md_check_no_bitmap(struct mddev
*mddev
)
1053 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1055 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1056 mdname(mddev
), mddev
->pers
->name
);
1059 EXPORT_SYMBOL(md_check_no_bitmap
);
1062 * load_super for 0.90.0
1064 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1066 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1071 * Calculate the position of the superblock (512byte sectors),
1072 * it's at the end of the disk.
1074 * It also happens to be a multiple of 4Kb.
1076 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1078 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1079 if (ret
) return ret
;
1083 bdevname(rdev
->bdev
, b
);
1084 sb
= page_address(rdev
->sb_page
);
1086 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1087 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1092 if (sb
->major_version
!= 0 ||
1093 sb
->minor_version
< 90 ||
1094 sb
->minor_version
> 91) {
1095 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1096 sb
->major_version
, sb
->minor_version
,
1101 if (sb
->raid_disks
<= 0)
1104 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1105 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1110 rdev
->preferred_minor
= sb
->md_minor
;
1111 rdev
->data_offset
= 0;
1112 rdev
->sb_size
= MD_SB_BYTES
;
1113 rdev
->badblocks
.shift
= -1;
1115 if (sb
->level
== LEVEL_MULTIPATH
)
1118 rdev
->desc_nr
= sb
->this_disk
.number
;
1124 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1125 if (!uuid_equal(refsb
, sb
)) {
1126 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1127 b
, bdevname(refdev
->bdev
,b2
));
1130 if (!sb_equal(refsb
, sb
)) {
1131 printk(KERN_WARNING
"md: %s has same UUID"
1132 " but different superblock to %s\n",
1133 b
, bdevname(refdev
->bdev
, b2
));
1137 ev2
= md_event(refsb
);
1143 rdev
->sectors
= rdev
->sb_start
;
1144 /* Limit to 4TB as metadata cannot record more than that */
1145 if (rdev
->sectors
>= (2ULL << 32))
1146 rdev
->sectors
= (2ULL << 32) - 2;
1148 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1149 /* "this cannot possibly happen" ... */
1157 * validate_super for 0.90.0
1159 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1162 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1163 __u64 ev1
= md_event(sb
);
1165 rdev
->raid_disk
= -1;
1166 clear_bit(Faulty
, &rdev
->flags
);
1167 clear_bit(In_sync
, &rdev
->flags
);
1168 clear_bit(WriteMostly
, &rdev
->flags
);
1170 if (mddev
->raid_disks
== 0) {
1171 mddev
->major_version
= 0;
1172 mddev
->minor_version
= sb
->minor_version
;
1173 mddev
->patch_version
= sb
->patch_version
;
1174 mddev
->external
= 0;
1175 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1176 mddev
->ctime
= sb
->ctime
;
1177 mddev
->utime
= sb
->utime
;
1178 mddev
->level
= sb
->level
;
1179 mddev
->clevel
[0] = 0;
1180 mddev
->layout
= sb
->layout
;
1181 mddev
->raid_disks
= sb
->raid_disks
;
1182 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1183 mddev
->events
= ev1
;
1184 mddev
->bitmap_info
.offset
= 0;
1185 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1187 if (mddev
->minor_version
>= 91) {
1188 mddev
->reshape_position
= sb
->reshape_position
;
1189 mddev
->delta_disks
= sb
->delta_disks
;
1190 mddev
->new_level
= sb
->new_level
;
1191 mddev
->new_layout
= sb
->new_layout
;
1192 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1194 mddev
->reshape_position
= MaxSector
;
1195 mddev
->delta_disks
= 0;
1196 mddev
->new_level
= mddev
->level
;
1197 mddev
->new_layout
= mddev
->layout
;
1198 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1201 if (sb
->state
& (1<<MD_SB_CLEAN
))
1202 mddev
->recovery_cp
= MaxSector
;
1204 if (sb
->events_hi
== sb
->cp_events_hi
&&
1205 sb
->events_lo
== sb
->cp_events_lo
) {
1206 mddev
->recovery_cp
= sb
->recovery_cp
;
1208 mddev
->recovery_cp
= 0;
1211 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1212 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1213 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1214 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1216 mddev
->max_disks
= MD_SB_DISKS
;
1218 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1219 mddev
->bitmap_info
.file
== NULL
)
1220 mddev
->bitmap_info
.offset
=
1221 mddev
->bitmap_info
.default_offset
;
1223 } else if (mddev
->pers
== NULL
) {
1224 /* Insist on good event counter while assembling, except
1225 * for spares (which don't need an event count) */
1227 if (sb
->disks
[rdev
->desc_nr
].state
& (
1228 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1229 if (ev1
< mddev
->events
)
1231 } else if (mddev
->bitmap
) {
1232 /* if adding to array with a bitmap, then we can accept an
1233 * older device ... but not too old.
1235 if (ev1
< mddev
->bitmap
->events_cleared
)
1238 if (ev1
< mddev
->events
)
1239 /* just a hot-add of a new device, leave raid_disk at -1 */
1243 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1244 desc
= sb
->disks
+ rdev
->desc_nr
;
1246 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1247 set_bit(Faulty
, &rdev
->flags
);
1248 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1249 desc->raid_disk < mddev->raid_disks */) {
1250 set_bit(In_sync
, &rdev
->flags
);
1251 rdev
->raid_disk
= desc
->raid_disk
;
1252 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1253 /* active but not in sync implies recovery up to
1254 * reshape position. We don't know exactly where
1255 * that is, so set to zero for now */
1256 if (mddev
->minor_version
>= 91) {
1257 rdev
->recovery_offset
= 0;
1258 rdev
->raid_disk
= desc
->raid_disk
;
1261 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1262 set_bit(WriteMostly
, &rdev
->flags
);
1263 } else /* MULTIPATH are always insync */
1264 set_bit(In_sync
, &rdev
->flags
);
1269 * sync_super for 0.90.0
1271 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1274 struct md_rdev
*rdev2
;
1275 int next_spare
= mddev
->raid_disks
;
1278 /* make rdev->sb match mddev data..
1281 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1282 * 3/ any empty disks < next_spare become removed
1284 * disks[0] gets initialised to REMOVED because
1285 * we cannot be sure from other fields if it has
1286 * been initialised or not.
1289 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1291 rdev
->sb_size
= MD_SB_BYTES
;
1293 sb
= page_address(rdev
->sb_page
);
1295 memset(sb
, 0, sizeof(*sb
));
1297 sb
->md_magic
= MD_SB_MAGIC
;
1298 sb
->major_version
= mddev
->major_version
;
1299 sb
->patch_version
= mddev
->patch_version
;
1300 sb
->gvalid_words
= 0; /* ignored */
1301 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1302 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1303 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1304 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1306 sb
->ctime
= mddev
->ctime
;
1307 sb
->level
= mddev
->level
;
1308 sb
->size
= mddev
->dev_sectors
/ 2;
1309 sb
->raid_disks
= mddev
->raid_disks
;
1310 sb
->md_minor
= mddev
->md_minor
;
1311 sb
->not_persistent
= 0;
1312 sb
->utime
= mddev
->utime
;
1314 sb
->events_hi
= (mddev
->events
>>32);
1315 sb
->events_lo
= (u32
)mddev
->events
;
1317 if (mddev
->reshape_position
== MaxSector
)
1318 sb
->minor_version
= 90;
1320 sb
->minor_version
= 91;
1321 sb
->reshape_position
= mddev
->reshape_position
;
1322 sb
->new_level
= mddev
->new_level
;
1323 sb
->delta_disks
= mddev
->delta_disks
;
1324 sb
->new_layout
= mddev
->new_layout
;
1325 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1327 mddev
->minor_version
= sb
->minor_version
;
1330 sb
->recovery_cp
= mddev
->recovery_cp
;
1331 sb
->cp_events_hi
= (mddev
->events
>>32);
1332 sb
->cp_events_lo
= (u32
)mddev
->events
;
1333 if (mddev
->recovery_cp
== MaxSector
)
1334 sb
->state
= (1<< MD_SB_CLEAN
);
1336 sb
->recovery_cp
= 0;
1338 sb
->layout
= mddev
->layout
;
1339 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1341 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1342 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1344 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1345 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1348 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1350 if (rdev2
->raid_disk
>= 0 &&
1351 sb
->minor_version
>= 91)
1352 /* we have nowhere to store the recovery_offset,
1353 * but if it is not below the reshape_position,
1354 * we can piggy-back on that.
1357 if (rdev2
->raid_disk
< 0 ||
1358 test_bit(Faulty
, &rdev2
->flags
))
1361 desc_nr
= rdev2
->raid_disk
;
1363 desc_nr
= next_spare
++;
1364 rdev2
->desc_nr
= desc_nr
;
1365 d
= &sb
->disks
[rdev2
->desc_nr
];
1367 d
->number
= rdev2
->desc_nr
;
1368 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1369 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1371 d
->raid_disk
= rdev2
->raid_disk
;
1373 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1374 if (test_bit(Faulty
, &rdev2
->flags
))
1375 d
->state
= (1<<MD_DISK_FAULTY
);
1376 else if (is_active
) {
1377 d
->state
= (1<<MD_DISK_ACTIVE
);
1378 if (test_bit(In_sync
, &rdev2
->flags
))
1379 d
->state
|= (1<<MD_DISK_SYNC
);
1387 if (test_bit(WriteMostly
, &rdev2
->flags
))
1388 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1390 /* now set the "removed" and "faulty" bits on any missing devices */
1391 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1392 mdp_disk_t
*d
= &sb
->disks
[i
];
1393 if (d
->state
== 0 && d
->number
== 0) {
1396 d
->state
= (1<<MD_DISK_REMOVED
);
1397 d
->state
|= (1<<MD_DISK_FAULTY
);
1401 sb
->nr_disks
= nr_disks
;
1402 sb
->active_disks
= active
;
1403 sb
->working_disks
= working
;
1404 sb
->failed_disks
= failed
;
1405 sb
->spare_disks
= spare
;
1407 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1408 sb
->sb_csum
= calc_sb_csum(sb
);
1412 * rdev_size_change for 0.90.0
1414 static unsigned long long
1415 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1417 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1418 return 0; /* component must fit device */
1419 if (rdev
->mddev
->bitmap_info
.offset
)
1420 return 0; /* can't move bitmap */
1421 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1422 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1423 num_sectors
= rdev
->sb_start
;
1424 /* Limit to 4TB as metadata cannot record more than that.
1425 * 4TB == 2^32 KB, or 2*2^32 sectors.
1427 if (num_sectors
>= (2ULL << 32))
1428 num_sectors
= (2ULL << 32) - 2;
1429 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1431 md_super_wait(rdev
->mddev
);
1437 * version 1 superblock
1440 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1444 unsigned long long newcsum
;
1445 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1446 __le32
*isuper
= (__le32
*)sb
;
1449 disk_csum
= sb
->sb_csum
;
1452 for (i
=0; size
>=4; size
-= 4 )
1453 newcsum
+= le32_to_cpu(*isuper
++);
1456 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1458 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1459 sb
->sb_csum
= disk_csum
;
1460 return cpu_to_le32(csum
);
1463 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1465 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1467 struct mdp_superblock_1
*sb
;
1470 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1474 * Calculate the position of the superblock in 512byte sectors.
1475 * It is always aligned to a 4K boundary and
1476 * depeding on minor_version, it can be:
1477 * 0: At least 8K, but less than 12K, from end of device
1478 * 1: At start of device
1479 * 2: 4K from start of device.
1481 switch(minor_version
) {
1483 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1485 sb_start
&= ~(sector_t
)(4*2-1);
1496 rdev
->sb_start
= sb_start
;
1498 /* superblock is rarely larger than 1K, but it can be larger,
1499 * and it is safe to read 4k, so we do that
1501 ret
= read_disk_sb(rdev
, 4096);
1502 if (ret
) return ret
;
1505 sb
= page_address(rdev
->sb_page
);
1507 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1508 sb
->major_version
!= cpu_to_le32(1) ||
1509 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1510 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1511 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1514 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1515 printk("md: invalid superblock checksum on %s\n",
1516 bdevname(rdev
->bdev
,b
));
1519 if (le64_to_cpu(sb
->data_size
) < 10) {
1520 printk("md: data_size too small on %s\n",
1521 bdevname(rdev
->bdev
,b
));
1525 rdev
->preferred_minor
= 0xffff;
1526 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1527 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1529 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1530 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1531 if (rdev
->sb_size
& bmask
)
1532 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1535 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1538 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1541 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1543 if (!rdev
->bb_page
) {
1544 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1548 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1549 rdev
->badblocks
.count
== 0) {
1550 /* need to load the bad block list.
1551 * Currently we limit it to one page.
1557 int sectors
= le16_to_cpu(sb
->bblog_size
);
1558 if (sectors
> (PAGE_SIZE
/ 512))
1560 offset
= le32_to_cpu(sb
->bblog_offset
);
1563 bb_sector
= (long long)offset
;
1564 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1565 rdev
->bb_page
, READ
, true))
1567 bbp
= (u64
*)page_address(rdev
->bb_page
);
1568 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1569 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1570 u64 bb
= le64_to_cpu(*bbp
);
1571 int count
= bb
& (0x3ff);
1572 u64 sector
= bb
>> 10;
1573 sector
<<= sb
->bblog_shift
;
1574 count
<<= sb
->bblog_shift
;
1577 if (md_set_badblocks(&rdev
->badblocks
,
1578 sector
, count
, 1) == 0)
1581 } else if (sb
->bblog_offset
== 0)
1582 rdev
->badblocks
.shift
= -1;
1588 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1590 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1591 sb
->level
!= refsb
->level
||
1592 sb
->layout
!= refsb
->layout
||
1593 sb
->chunksize
!= refsb
->chunksize
) {
1594 printk(KERN_WARNING
"md: %s has strangely different"
1595 " superblock to %s\n",
1596 bdevname(rdev
->bdev
,b
),
1597 bdevname(refdev
->bdev
,b2
));
1600 ev1
= le64_to_cpu(sb
->events
);
1601 ev2
= le64_to_cpu(refsb
->events
);
1609 rdev
->sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
1610 le64_to_cpu(sb
->data_offset
);
1612 rdev
->sectors
= rdev
->sb_start
;
1613 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1615 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1616 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1621 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1623 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1624 __u64 ev1
= le64_to_cpu(sb
->events
);
1626 rdev
->raid_disk
= -1;
1627 clear_bit(Faulty
, &rdev
->flags
);
1628 clear_bit(In_sync
, &rdev
->flags
);
1629 clear_bit(WriteMostly
, &rdev
->flags
);
1631 if (mddev
->raid_disks
== 0) {
1632 mddev
->major_version
= 1;
1633 mddev
->patch_version
= 0;
1634 mddev
->external
= 0;
1635 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1636 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1637 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1638 mddev
->level
= le32_to_cpu(sb
->level
);
1639 mddev
->clevel
[0] = 0;
1640 mddev
->layout
= le32_to_cpu(sb
->layout
);
1641 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1642 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1643 mddev
->events
= ev1
;
1644 mddev
->bitmap_info
.offset
= 0;
1645 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1647 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1648 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1650 mddev
->max_disks
= (4096-256)/2;
1652 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1653 mddev
->bitmap_info
.file
== NULL
)
1654 mddev
->bitmap_info
.offset
=
1655 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1657 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1658 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1659 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1660 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1661 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1662 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1664 mddev
->reshape_position
= MaxSector
;
1665 mddev
->delta_disks
= 0;
1666 mddev
->new_level
= mddev
->level
;
1667 mddev
->new_layout
= mddev
->layout
;
1668 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1671 } else if (mddev
->pers
== NULL
) {
1672 /* Insist of good event counter while assembling, except for
1673 * spares (which don't need an event count) */
1675 if (rdev
->desc_nr
>= 0 &&
1676 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1677 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1678 if (ev1
< mddev
->events
)
1680 } else if (mddev
->bitmap
) {
1681 /* If adding to array with a bitmap, then we can accept an
1682 * older device, but not too old.
1684 if (ev1
< mddev
->bitmap
->events_cleared
)
1687 if (ev1
< mddev
->events
)
1688 /* just a hot-add of a new device, leave raid_disk at -1 */
1691 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1693 if (rdev
->desc_nr
< 0 ||
1694 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1698 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1700 case 0xffff: /* spare */
1702 case 0xfffe: /* faulty */
1703 set_bit(Faulty
, &rdev
->flags
);
1706 if ((le32_to_cpu(sb
->feature_map
) &
1707 MD_FEATURE_RECOVERY_OFFSET
))
1708 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1710 set_bit(In_sync
, &rdev
->flags
);
1711 rdev
->raid_disk
= role
;
1714 if (sb
->devflags
& WriteMostly1
)
1715 set_bit(WriteMostly
, &rdev
->flags
);
1716 } else /* MULTIPATH are always insync */
1717 set_bit(In_sync
, &rdev
->flags
);
1722 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1724 struct mdp_superblock_1
*sb
;
1725 struct md_rdev
*rdev2
;
1727 /* make rdev->sb match mddev and rdev data. */
1729 sb
= page_address(rdev
->sb_page
);
1731 sb
->feature_map
= 0;
1733 sb
->recovery_offset
= cpu_to_le64(0);
1734 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1735 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1737 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1738 sb
->events
= cpu_to_le64(mddev
->events
);
1740 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1742 sb
->resync_offset
= cpu_to_le64(0);
1744 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1746 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1747 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1748 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1749 sb
->level
= cpu_to_le32(mddev
->level
);
1750 sb
->layout
= cpu_to_le32(mddev
->layout
);
1752 if (test_bit(WriteMostly
, &rdev
->flags
))
1753 sb
->devflags
|= WriteMostly1
;
1755 sb
->devflags
&= ~WriteMostly1
;
1757 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1758 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1759 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1762 if (rdev
->raid_disk
>= 0 &&
1763 !test_bit(In_sync
, &rdev
->flags
)) {
1765 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1766 sb
->recovery_offset
=
1767 cpu_to_le64(rdev
->recovery_offset
);
1770 if (mddev
->reshape_position
!= MaxSector
) {
1771 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1772 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1773 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1774 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1775 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1776 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1779 if (rdev
->badblocks
.count
== 0)
1780 /* Nothing to do for bad blocks*/ ;
1781 else if (sb
->bblog_offset
== 0)
1782 /* Cannot record bad blocks on this device */
1783 md_error(mddev
, rdev
);
1785 struct badblocks
*bb
= &rdev
->badblocks
;
1786 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1788 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1793 seq
= read_seqbegin(&bb
->lock
);
1795 memset(bbp
, 0xff, PAGE_SIZE
);
1797 for (i
= 0 ; i
< bb
->count
; i
++) {
1798 u64 internal_bb
= *p
++;
1799 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1800 | BB_LEN(internal_bb
));
1801 *bbp
++ = cpu_to_le64(store_bb
);
1803 if (read_seqretry(&bb
->lock
, seq
))
1806 bb
->sector
= (rdev
->sb_start
+
1807 (int)le32_to_cpu(sb
->bblog_offset
));
1808 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1814 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1815 if (rdev2
->desc_nr
+1 > max_dev
)
1816 max_dev
= rdev2
->desc_nr
+1;
1818 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1820 sb
->max_dev
= cpu_to_le32(max_dev
);
1821 rdev
->sb_size
= max_dev
* 2 + 256;
1822 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1823 if (rdev
->sb_size
& bmask
)
1824 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1826 max_dev
= le32_to_cpu(sb
->max_dev
);
1828 for (i
=0; i
<max_dev
;i
++)
1829 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1831 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1833 if (test_bit(Faulty
, &rdev2
->flags
))
1834 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1835 else if (test_bit(In_sync
, &rdev2
->flags
))
1836 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1837 else if (rdev2
->raid_disk
>= 0)
1838 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1840 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1843 sb
->sb_csum
= calc_sb_1_csum(sb
);
1846 static unsigned long long
1847 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1849 struct mdp_superblock_1
*sb
;
1850 sector_t max_sectors
;
1851 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1852 return 0; /* component must fit device */
1853 if (rdev
->sb_start
< rdev
->data_offset
) {
1854 /* minor versions 1 and 2; superblock before data */
1855 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1856 max_sectors
-= rdev
->data_offset
;
1857 if (!num_sectors
|| num_sectors
> max_sectors
)
1858 num_sectors
= max_sectors
;
1859 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1860 /* minor version 0 with bitmap we can't move */
1863 /* minor version 0; superblock after data */
1865 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1866 sb_start
&= ~(sector_t
)(4*2 - 1);
1867 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1868 if (!num_sectors
|| num_sectors
> max_sectors
)
1869 num_sectors
= max_sectors
;
1870 rdev
->sb_start
= sb_start
;
1872 sb
= page_address(rdev
->sb_page
);
1873 sb
->data_size
= cpu_to_le64(num_sectors
);
1874 sb
->super_offset
= rdev
->sb_start
;
1875 sb
->sb_csum
= calc_sb_1_csum(sb
);
1876 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1878 md_super_wait(rdev
->mddev
);
1882 static struct super_type super_types
[] = {
1885 .owner
= THIS_MODULE
,
1886 .load_super
= super_90_load
,
1887 .validate_super
= super_90_validate
,
1888 .sync_super
= super_90_sync
,
1889 .rdev_size_change
= super_90_rdev_size_change
,
1893 .owner
= THIS_MODULE
,
1894 .load_super
= super_1_load
,
1895 .validate_super
= super_1_validate
,
1896 .sync_super
= super_1_sync
,
1897 .rdev_size_change
= super_1_rdev_size_change
,
1901 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1903 if (mddev
->sync_super
) {
1904 mddev
->sync_super(mddev
, rdev
);
1908 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1910 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1913 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1915 struct md_rdev
*rdev
, *rdev2
;
1918 rdev_for_each_rcu(rdev
, mddev1
)
1919 rdev_for_each_rcu(rdev2
, mddev2
)
1920 if (rdev
->bdev
->bd_contains
==
1921 rdev2
->bdev
->bd_contains
) {
1929 static LIST_HEAD(pending_raid_disks
);
1932 * Try to register data integrity profile for an mddev
1934 * This is called when an array is started and after a disk has been kicked
1935 * from the array. It only succeeds if all working and active component devices
1936 * are integrity capable with matching profiles.
1938 int md_integrity_register(struct mddev
*mddev
)
1940 struct md_rdev
*rdev
, *reference
= NULL
;
1942 if (list_empty(&mddev
->disks
))
1943 return 0; /* nothing to do */
1944 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1945 return 0; /* shouldn't register, or already is */
1946 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1947 /* skip spares and non-functional disks */
1948 if (test_bit(Faulty
, &rdev
->flags
))
1950 if (rdev
->raid_disk
< 0)
1953 /* Use the first rdev as the reference */
1957 /* does this rdev's profile match the reference profile? */
1958 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1959 rdev
->bdev
->bd_disk
) < 0)
1962 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1965 * All component devices are integrity capable and have matching
1966 * profiles, register the common profile for the md device.
1968 if (blk_integrity_register(mddev
->gendisk
,
1969 bdev_get_integrity(reference
->bdev
)) != 0) {
1970 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1974 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1975 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1976 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1982 EXPORT_SYMBOL(md_integrity_register
);
1984 /* Disable data integrity if non-capable/non-matching disk is being added */
1985 void md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
1987 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1988 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1990 if (!bi_mddev
) /* nothing to do */
1992 if (rdev
->raid_disk
< 0) /* skip spares */
1994 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1995 rdev
->bdev
->bd_disk
) >= 0)
1997 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1998 blk_integrity_unregister(mddev
->gendisk
);
2000 EXPORT_SYMBOL(md_integrity_add_rdev
);
2002 static int bind_rdev_to_array(struct md_rdev
* rdev
, struct mddev
* mddev
)
2004 char b
[BDEVNAME_SIZE
];
2014 /* prevent duplicates */
2015 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2018 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2019 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2020 rdev
->sectors
< mddev
->dev_sectors
)) {
2022 /* Cannot change size, so fail
2023 * If mddev->level <= 0, then we don't care
2024 * about aligning sizes (e.g. linear)
2026 if (mddev
->level
> 0)
2029 mddev
->dev_sectors
= rdev
->sectors
;
2032 /* Verify rdev->desc_nr is unique.
2033 * If it is -1, assign a free number, else
2034 * check number is not in use
2036 if (rdev
->desc_nr
< 0) {
2038 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2039 while (find_rdev_nr(mddev
, choice
))
2041 rdev
->desc_nr
= choice
;
2043 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2046 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2047 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2048 mdname(mddev
), mddev
->max_disks
);
2051 bdevname(rdev
->bdev
,b
);
2052 while ( (s
=strchr(b
, '/')) != NULL
)
2055 rdev
->mddev
= mddev
;
2056 printk(KERN_INFO
"md: bind<%s>\n", b
);
2058 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2061 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2062 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2063 /* failure here is OK */;
2064 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2066 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2067 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2069 /* May as well allow recovery to be retried once */
2070 mddev
->recovery_disabled
++;
2075 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2080 static void md_delayed_delete(struct work_struct
*ws
)
2082 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2083 kobject_del(&rdev
->kobj
);
2084 kobject_put(&rdev
->kobj
);
2087 static void unbind_rdev_from_array(struct md_rdev
* rdev
)
2089 char b
[BDEVNAME_SIZE
];
2094 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2095 list_del_rcu(&rdev
->same_set
);
2096 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2098 sysfs_remove_link(&rdev
->kobj
, "block");
2099 sysfs_put(rdev
->sysfs_state
);
2100 rdev
->sysfs_state
= NULL
;
2101 kfree(rdev
->badblocks
.page
);
2102 rdev
->badblocks
.count
= 0;
2103 rdev
->badblocks
.page
= NULL
;
2104 /* We need to delay this, otherwise we can deadlock when
2105 * writing to 'remove' to "dev/state". We also need
2106 * to delay it due to rcu usage.
2109 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2110 kobject_get(&rdev
->kobj
);
2111 queue_work(md_misc_wq
, &rdev
->del_work
);
2115 * prevent the device from being mounted, repartitioned or
2116 * otherwise reused by a RAID array (or any other kernel
2117 * subsystem), by bd_claiming the device.
2119 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2122 struct block_device
*bdev
;
2123 char b
[BDEVNAME_SIZE
];
2125 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2126 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2128 printk(KERN_ERR
"md: could not open %s.\n",
2129 __bdevname(dev
, b
));
2130 return PTR_ERR(bdev
);
2136 static void unlock_rdev(struct md_rdev
*rdev
)
2138 struct block_device
*bdev
= rdev
->bdev
;
2142 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2145 void md_autodetect_dev(dev_t dev
);
2147 static void export_rdev(struct md_rdev
* rdev
)
2149 char b
[BDEVNAME_SIZE
];
2150 printk(KERN_INFO
"md: export_rdev(%s)\n",
2151 bdevname(rdev
->bdev
,b
));
2156 if (test_bit(AutoDetected
, &rdev
->flags
))
2157 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2160 kobject_put(&rdev
->kobj
);
2163 static void kick_rdev_from_array(struct md_rdev
* rdev
)
2165 unbind_rdev_from_array(rdev
);
2169 static void export_array(struct mddev
*mddev
)
2171 struct md_rdev
*rdev
, *tmp
;
2173 rdev_for_each(rdev
, tmp
, mddev
) {
2178 kick_rdev_from_array(rdev
);
2180 if (!list_empty(&mddev
->disks
))
2182 mddev
->raid_disks
= 0;
2183 mddev
->major_version
= 0;
2186 static void print_desc(mdp_disk_t
*desc
)
2188 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2189 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2192 static void print_sb_90(mdp_super_t
*sb
)
2197 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2198 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2199 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2201 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2202 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2203 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2204 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2205 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2206 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2207 sb
->failed_disks
, sb
->spare_disks
,
2208 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2211 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2214 desc
= sb
->disks
+ i
;
2215 if (desc
->number
|| desc
->major
|| desc
->minor
||
2216 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2217 printk(" D %2d: ", i
);
2221 printk(KERN_INFO
"md: THIS: ");
2222 print_desc(&sb
->this_disk
);
2225 static void print_sb_1(struct mdp_superblock_1
*sb
)
2229 uuid
= sb
->set_uuid
;
2231 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2232 "md: Name: \"%s\" CT:%llu\n",
2233 le32_to_cpu(sb
->major_version
),
2234 le32_to_cpu(sb
->feature_map
),
2237 (unsigned long long)le64_to_cpu(sb
->ctime
)
2238 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2240 uuid
= sb
->device_uuid
;
2242 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2244 "md: Dev:%08x UUID: %pU\n"
2245 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2246 "md: (MaxDev:%u) \n",
2247 le32_to_cpu(sb
->level
),
2248 (unsigned long long)le64_to_cpu(sb
->size
),
2249 le32_to_cpu(sb
->raid_disks
),
2250 le32_to_cpu(sb
->layout
),
2251 le32_to_cpu(sb
->chunksize
),
2252 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2253 (unsigned long long)le64_to_cpu(sb
->data_size
),
2254 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2255 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2256 le32_to_cpu(sb
->dev_number
),
2259 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2260 (unsigned long long)le64_to_cpu(sb
->events
),
2261 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2262 le32_to_cpu(sb
->sb_csum
),
2263 le32_to_cpu(sb
->max_dev
)
2267 static void print_rdev(struct md_rdev
*rdev
, int major_version
)
2269 char b
[BDEVNAME_SIZE
];
2270 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2271 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2272 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2274 if (rdev
->sb_loaded
) {
2275 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2276 switch (major_version
) {
2278 print_sb_90(page_address(rdev
->sb_page
));
2281 print_sb_1(page_address(rdev
->sb_page
));
2285 printk(KERN_INFO
"md: no rdev superblock!\n");
2288 static void md_print_devices(void)
2290 struct list_head
*tmp
;
2291 struct md_rdev
*rdev
;
2292 struct mddev
*mddev
;
2293 char b
[BDEVNAME_SIZE
];
2296 printk("md: **********************************\n");
2297 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2298 printk("md: **********************************\n");
2299 for_each_mddev(mddev
, tmp
) {
2302 bitmap_print_sb(mddev
->bitmap
);
2304 printk("%s: ", mdname(mddev
));
2305 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2306 printk("<%s>", bdevname(rdev
->bdev
,b
));
2309 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2310 print_rdev(rdev
, mddev
->major_version
);
2312 printk("md: **********************************\n");
2317 static void sync_sbs(struct mddev
* mddev
, int nospares
)
2319 /* Update each superblock (in-memory image), but
2320 * if we are allowed to, skip spares which already
2321 * have the right event counter, or have one earlier
2322 * (which would mean they aren't being marked as dirty
2323 * with the rest of the array)
2325 struct md_rdev
*rdev
;
2326 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2327 if (rdev
->sb_events
== mddev
->events
||
2329 rdev
->raid_disk
< 0 &&
2330 rdev
->sb_events
+1 == mddev
->events
)) {
2331 /* Don't update this superblock */
2332 rdev
->sb_loaded
= 2;
2334 sync_super(mddev
, rdev
);
2335 rdev
->sb_loaded
= 1;
2340 static void md_update_sb(struct mddev
* mddev
, int force_change
)
2342 struct md_rdev
*rdev
;
2345 int any_badblocks_changed
= 0;
2348 /* First make sure individual recovery_offsets are correct */
2349 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2350 if (rdev
->raid_disk
>= 0 &&
2351 mddev
->delta_disks
>= 0 &&
2352 !test_bit(In_sync
, &rdev
->flags
) &&
2353 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2354 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2357 if (!mddev
->persistent
) {
2358 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2359 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2360 if (!mddev
->external
) {
2361 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2362 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2363 if (rdev
->badblocks
.changed
) {
2364 md_ack_all_badblocks(&rdev
->badblocks
);
2365 md_error(mddev
, rdev
);
2367 clear_bit(Blocked
, &rdev
->flags
);
2368 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2369 wake_up(&rdev
->blocked_wait
);
2372 wake_up(&mddev
->sb_wait
);
2376 spin_lock_irq(&mddev
->write_lock
);
2378 mddev
->utime
= get_seconds();
2380 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2382 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2383 /* just a clean<-> dirty transition, possibly leave spares alone,
2384 * though if events isn't the right even/odd, we will have to do
2390 if (mddev
->degraded
)
2391 /* If the array is degraded, then skipping spares is both
2392 * dangerous and fairly pointless.
2393 * Dangerous because a device that was removed from the array
2394 * might have a event_count that still looks up-to-date,
2395 * so it can be re-added without a resync.
2396 * Pointless because if there are any spares to skip,
2397 * then a recovery will happen and soon that array won't
2398 * be degraded any more and the spare can go back to sleep then.
2402 sync_req
= mddev
->in_sync
;
2404 /* If this is just a dirty<->clean transition, and the array is clean
2405 * and 'events' is odd, we can roll back to the previous clean state */
2407 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2408 && mddev
->can_decrease_events
2409 && mddev
->events
!= 1) {
2411 mddev
->can_decrease_events
= 0;
2413 /* otherwise we have to go forward and ... */
2415 mddev
->can_decrease_events
= nospares
;
2418 if (!mddev
->events
) {
2420 * oops, this 64-bit counter should never wrap.
2421 * Either we are in around ~1 trillion A.C., assuming
2422 * 1 reboot per second, or we have a bug:
2428 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2429 if (rdev
->badblocks
.changed
)
2430 any_badblocks_changed
++;
2431 if (test_bit(Faulty
, &rdev
->flags
))
2432 set_bit(FaultRecorded
, &rdev
->flags
);
2435 sync_sbs(mddev
, nospares
);
2436 spin_unlock_irq(&mddev
->write_lock
);
2438 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2439 mdname(mddev
), mddev
->in_sync
);
2441 bitmap_update_sb(mddev
->bitmap
);
2442 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2443 char b
[BDEVNAME_SIZE
];
2445 if (rdev
->sb_loaded
!= 1)
2446 continue; /* no noise on spare devices */
2448 if (!test_bit(Faulty
, &rdev
->flags
) &&
2449 rdev
->saved_raid_disk
== -1) {
2450 md_super_write(mddev
,rdev
,
2451 rdev
->sb_start
, rdev
->sb_size
,
2453 pr_debug("md: (write) %s's sb offset: %llu\n",
2454 bdevname(rdev
->bdev
, b
),
2455 (unsigned long long)rdev
->sb_start
);
2456 rdev
->sb_events
= mddev
->events
;
2457 if (rdev
->badblocks
.size
) {
2458 md_super_write(mddev
, rdev
,
2459 rdev
->badblocks
.sector
,
2460 rdev
->badblocks
.size
<< 9,
2462 rdev
->badblocks
.size
= 0;
2465 } else if (test_bit(Faulty
, &rdev
->flags
))
2466 pr_debug("md: %s (skipping faulty)\n",
2467 bdevname(rdev
->bdev
, b
));
2469 pr_debug("(skipping incremental s/r ");
2471 if (mddev
->level
== LEVEL_MULTIPATH
)
2472 /* only need to write one superblock... */
2475 md_super_wait(mddev
);
2476 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2478 spin_lock_irq(&mddev
->write_lock
);
2479 if (mddev
->in_sync
!= sync_req
||
2480 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2481 /* have to write it out again */
2482 spin_unlock_irq(&mddev
->write_lock
);
2485 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2486 spin_unlock_irq(&mddev
->write_lock
);
2487 wake_up(&mddev
->sb_wait
);
2488 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2489 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2491 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2492 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2493 clear_bit(Blocked
, &rdev
->flags
);
2495 if (any_badblocks_changed
)
2496 md_ack_all_badblocks(&rdev
->badblocks
);
2497 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2498 wake_up(&rdev
->blocked_wait
);
2502 /* words written to sysfs files may, or may not, be \n terminated.
2503 * We want to accept with case. For this we use cmd_match.
2505 static int cmd_match(const char *cmd
, const char *str
)
2507 /* See if cmd, written into a sysfs file, matches
2508 * str. They must either be the same, or cmd can
2509 * have a trailing newline
2511 while (*cmd
&& *str
&& *cmd
== *str
) {
2522 struct rdev_sysfs_entry
{
2523 struct attribute attr
;
2524 ssize_t (*show
)(struct md_rdev
*, char *);
2525 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2529 state_show(struct md_rdev
*rdev
, char *page
)
2534 if (test_bit(Faulty
, &rdev
->flags
) ||
2535 rdev
->badblocks
.unacked_exist
) {
2536 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2539 if (test_bit(In_sync
, &rdev
->flags
)) {
2540 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2543 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2544 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2547 if (test_bit(Blocked
, &rdev
->flags
) ||
2548 rdev
->badblocks
.unacked_exist
) {
2549 len
+= sprintf(page
+len
, "%sblocked", sep
);
2552 if (!test_bit(Faulty
, &rdev
->flags
) &&
2553 !test_bit(In_sync
, &rdev
->flags
)) {
2554 len
+= sprintf(page
+len
, "%sspare", sep
);
2557 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2558 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2561 return len
+sprintf(page
+len
, "\n");
2565 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2568 * faulty - simulates an error
2569 * remove - disconnects the device
2570 * writemostly - sets write_mostly
2571 * -writemostly - clears write_mostly
2572 * blocked - sets the Blocked flags
2573 * -blocked - clears the Blocked and possibly simulates an error
2574 * insync - sets Insync providing device isn't active
2575 * write_error - sets WriteErrorSeen
2576 * -write_error - clears WriteErrorSeen
2579 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2580 md_error(rdev
->mddev
, rdev
);
2581 if (test_bit(Faulty
, &rdev
->flags
))
2585 } else if (cmd_match(buf
, "remove")) {
2586 if (rdev
->raid_disk
>= 0)
2589 struct mddev
*mddev
= rdev
->mddev
;
2590 kick_rdev_from_array(rdev
);
2592 md_update_sb(mddev
, 1);
2593 md_new_event(mddev
);
2596 } else if (cmd_match(buf
, "writemostly")) {
2597 set_bit(WriteMostly
, &rdev
->flags
);
2599 } else if (cmd_match(buf
, "-writemostly")) {
2600 clear_bit(WriteMostly
, &rdev
->flags
);
2602 } else if (cmd_match(buf
, "blocked")) {
2603 set_bit(Blocked
, &rdev
->flags
);
2605 } else if (cmd_match(buf
, "-blocked")) {
2606 if (!test_bit(Faulty
, &rdev
->flags
) &&
2607 rdev
->badblocks
.unacked_exist
) {
2608 /* metadata handler doesn't understand badblocks,
2609 * so we need to fail the device
2611 md_error(rdev
->mddev
, rdev
);
2613 clear_bit(Blocked
, &rdev
->flags
);
2614 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2615 wake_up(&rdev
->blocked_wait
);
2616 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2617 md_wakeup_thread(rdev
->mddev
->thread
);
2620 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2621 set_bit(In_sync
, &rdev
->flags
);
2623 } else if (cmd_match(buf
, "write_error")) {
2624 set_bit(WriteErrorSeen
, &rdev
->flags
);
2626 } else if (cmd_match(buf
, "-write_error")) {
2627 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2631 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2632 return err
? err
: len
;
2634 static struct rdev_sysfs_entry rdev_state
=
2635 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2638 errors_show(struct md_rdev
*rdev
, char *page
)
2640 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2644 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2647 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2648 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2649 atomic_set(&rdev
->corrected_errors
, n
);
2654 static struct rdev_sysfs_entry rdev_errors
=
2655 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2658 slot_show(struct md_rdev
*rdev
, char *page
)
2660 if (rdev
->raid_disk
< 0)
2661 return sprintf(page
, "none\n");
2663 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2667 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2671 int slot
= simple_strtoul(buf
, &e
, 10);
2672 if (strncmp(buf
, "none", 4)==0)
2674 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2676 if (rdev
->mddev
->pers
&& slot
== -1) {
2677 /* Setting 'slot' on an active array requires also
2678 * updating the 'rd%d' link, and communicating
2679 * with the personality with ->hot_*_disk.
2680 * For now we only support removing
2681 * failed/spare devices. This normally happens automatically,
2682 * but not when the metadata is externally managed.
2684 if (rdev
->raid_disk
== -1)
2686 /* personality does all needed checks */
2687 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2689 err
= rdev
->mddev
->pers
->
2690 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2693 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2694 rdev
->raid_disk
= -1;
2695 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2696 md_wakeup_thread(rdev
->mddev
->thread
);
2697 } else if (rdev
->mddev
->pers
) {
2698 struct md_rdev
*rdev2
;
2699 /* Activating a spare .. or possibly reactivating
2700 * if we ever get bitmaps working here.
2703 if (rdev
->raid_disk
!= -1)
2706 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2709 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2712 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2713 if (rdev2
->raid_disk
== slot
)
2716 if (slot
>= rdev
->mddev
->raid_disks
&&
2717 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2720 rdev
->raid_disk
= slot
;
2721 if (test_bit(In_sync
, &rdev
->flags
))
2722 rdev
->saved_raid_disk
= slot
;
2724 rdev
->saved_raid_disk
= -1;
2725 clear_bit(In_sync
, &rdev
->flags
);
2726 err
= rdev
->mddev
->pers
->
2727 hot_add_disk(rdev
->mddev
, rdev
);
2729 rdev
->raid_disk
= -1;
2732 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2733 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2734 /* failure here is OK */;
2735 /* don't wakeup anyone, leave that to userspace. */
2737 if (slot
>= rdev
->mddev
->raid_disks
&&
2738 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2740 rdev
->raid_disk
= slot
;
2741 /* assume it is working */
2742 clear_bit(Faulty
, &rdev
->flags
);
2743 clear_bit(WriteMostly
, &rdev
->flags
);
2744 set_bit(In_sync
, &rdev
->flags
);
2745 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2751 static struct rdev_sysfs_entry rdev_slot
=
2752 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2755 offset_show(struct md_rdev
*rdev
, char *page
)
2757 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2761 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2764 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2765 if (e
==buf
|| (*e
&& *e
!= '\n'))
2767 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2769 if (rdev
->sectors
&& rdev
->mddev
->external
)
2770 /* Must set offset before size, so overlap checks
2773 rdev
->data_offset
= offset
;
2777 static struct rdev_sysfs_entry rdev_offset
=
2778 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2781 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2783 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2786 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2788 /* check if two start/length pairs overlap */
2796 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2798 unsigned long long blocks
;
2801 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2804 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2805 return -EINVAL
; /* sector conversion overflow */
2808 if (new != blocks
* 2)
2809 return -EINVAL
; /* unsigned long long to sector_t overflow */
2816 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2818 struct mddev
*my_mddev
= rdev
->mddev
;
2819 sector_t oldsectors
= rdev
->sectors
;
2822 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2824 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2825 if (my_mddev
->persistent
) {
2826 sectors
= super_types
[my_mddev
->major_version
].
2827 rdev_size_change(rdev
, sectors
);
2830 } else if (!sectors
)
2831 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2834 if (sectors
< my_mddev
->dev_sectors
)
2835 return -EINVAL
; /* component must fit device */
2837 rdev
->sectors
= sectors
;
2838 if (sectors
> oldsectors
&& my_mddev
->external
) {
2839 /* need to check that all other rdevs with the same ->bdev
2840 * do not overlap. We need to unlock the mddev to avoid
2841 * a deadlock. We have already changed rdev->sectors, and if
2842 * we have to change it back, we will have the lock again.
2844 struct mddev
*mddev
;
2846 struct list_head
*tmp
;
2848 mddev_unlock(my_mddev
);
2849 for_each_mddev(mddev
, tmp
) {
2850 struct md_rdev
*rdev2
;
2853 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2854 if (rdev
->bdev
== rdev2
->bdev
&&
2856 overlaps(rdev
->data_offset
, rdev
->sectors
,
2862 mddev_unlock(mddev
);
2868 mddev_lock(my_mddev
);
2870 /* Someone else could have slipped in a size
2871 * change here, but doing so is just silly.
2872 * We put oldsectors back because we *know* it is
2873 * safe, and trust userspace not to race with
2876 rdev
->sectors
= oldsectors
;
2883 static struct rdev_sysfs_entry rdev_size
=
2884 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2887 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
2889 unsigned long long recovery_start
= rdev
->recovery_offset
;
2891 if (test_bit(In_sync
, &rdev
->flags
) ||
2892 recovery_start
== MaxSector
)
2893 return sprintf(page
, "none\n");
2895 return sprintf(page
, "%llu\n", recovery_start
);
2898 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2900 unsigned long long recovery_start
;
2902 if (cmd_match(buf
, "none"))
2903 recovery_start
= MaxSector
;
2904 else if (strict_strtoull(buf
, 10, &recovery_start
))
2907 if (rdev
->mddev
->pers
&&
2908 rdev
->raid_disk
>= 0)
2911 rdev
->recovery_offset
= recovery_start
;
2912 if (recovery_start
== MaxSector
)
2913 set_bit(In_sync
, &rdev
->flags
);
2915 clear_bit(In_sync
, &rdev
->flags
);
2919 static struct rdev_sysfs_entry rdev_recovery_start
=
2920 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2924 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2926 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2928 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
2930 return badblocks_show(&rdev
->badblocks
, page
, 0);
2932 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2934 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2935 /* Maybe that ack was all we needed */
2936 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2937 wake_up(&rdev
->blocked_wait
);
2940 static struct rdev_sysfs_entry rdev_bad_blocks
=
2941 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2944 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
2946 return badblocks_show(&rdev
->badblocks
, page
, 1);
2948 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
2950 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2952 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2953 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2955 static struct attribute
*rdev_default_attrs
[] = {
2961 &rdev_recovery_start
.attr
,
2962 &rdev_bad_blocks
.attr
,
2963 &rdev_unack_bad_blocks
.attr
,
2967 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2969 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2970 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2971 struct mddev
*mddev
= rdev
->mddev
;
2977 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2979 if (rdev
->mddev
== NULL
)
2982 rv
= entry
->show(rdev
, page
);
2983 mddev_unlock(mddev
);
2989 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2990 const char *page
, size_t length
)
2992 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2993 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
2995 struct mddev
*mddev
= rdev
->mddev
;
2999 if (!capable(CAP_SYS_ADMIN
))
3001 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3003 if (rdev
->mddev
== NULL
)
3006 rv
= entry
->store(rdev
, page
, length
);
3007 mddev_unlock(mddev
);
3012 static void rdev_free(struct kobject
*ko
)
3014 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3017 static const struct sysfs_ops rdev_sysfs_ops
= {
3018 .show
= rdev_attr_show
,
3019 .store
= rdev_attr_store
,
3021 static struct kobj_type rdev_ktype
= {
3022 .release
= rdev_free
,
3023 .sysfs_ops
= &rdev_sysfs_ops
,
3024 .default_attrs
= rdev_default_attrs
,
3027 int md_rdev_init(struct md_rdev
*rdev
)
3030 rdev
->saved_raid_disk
= -1;
3031 rdev
->raid_disk
= -1;
3033 rdev
->data_offset
= 0;
3034 rdev
->sb_events
= 0;
3035 rdev
->last_read_error
.tv_sec
= 0;
3036 rdev
->last_read_error
.tv_nsec
= 0;
3037 rdev
->sb_loaded
= 0;
3038 rdev
->bb_page
= NULL
;
3039 atomic_set(&rdev
->nr_pending
, 0);
3040 atomic_set(&rdev
->read_errors
, 0);
3041 atomic_set(&rdev
->corrected_errors
, 0);
3043 INIT_LIST_HEAD(&rdev
->same_set
);
3044 init_waitqueue_head(&rdev
->blocked_wait
);
3046 /* Add space to store bad block list.
3047 * This reserves the space even on arrays where it cannot
3048 * be used - I wonder if that matters
3050 rdev
->badblocks
.count
= 0;
3051 rdev
->badblocks
.shift
= 0;
3052 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3053 seqlock_init(&rdev
->badblocks
.lock
);
3054 if (rdev
->badblocks
.page
== NULL
)
3059 EXPORT_SYMBOL_GPL(md_rdev_init
);
3061 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3063 * mark the device faulty if:
3065 * - the device is nonexistent (zero size)
3066 * - the device has no valid superblock
3068 * a faulty rdev _never_ has rdev->sb set.
3070 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3072 char b
[BDEVNAME_SIZE
];
3074 struct md_rdev
*rdev
;
3077 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3079 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3080 return ERR_PTR(-ENOMEM
);
3083 err
= md_rdev_init(rdev
);
3086 err
= alloc_disk_sb(rdev
);
3090 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3094 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3096 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3099 "md: %s has zero or unknown size, marking faulty!\n",
3100 bdevname(rdev
->bdev
,b
));
3105 if (super_format
>= 0) {
3106 err
= super_types
[super_format
].
3107 load_super(rdev
, NULL
, super_minor
);
3108 if (err
== -EINVAL
) {
3110 "md: %s does not have a valid v%d.%d "
3111 "superblock, not importing!\n",
3112 bdevname(rdev
->bdev
,b
),
3113 super_format
, super_minor
);
3118 "md: could not read %s's sb, not importing!\n",
3119 bdevname(rdev
->bdev
,b
));
3123 if (super_format
== -1)
3124 /* hot-add for 0.90, or non-persistent: so no badblocks */
3125 rdev
->badblocks
.shift
= -1;
3133 kfree(rdev
->badblocks
.page
);
3135 return ERR_PTR(err
);
3139 * Check a full RAID array for plausibility
3143 static void analyze_sbs(struct mddev
* mddev
)
3146 struct md_rdev
*rdev
, *freshest
, *tmp
;
3147 char b
[BDEVNAME_SIZE
];
3150 rdev_for_each(rdev
, tmp
, mddev
)
3151 switch (super_types
[mddev
->major_version
].
3152 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3160 "md: fatal superblock inconsistency in %s"
3161 " -- removing from array\n",
3162 bdevname(rdev
->bdev
,b
));
3163 kick_rdev_from_array(rdev
);
3167 super_types
[mddev
->major_version
].
3168 validate_super(mddev
, freshest
);
3171 rdev_for_each(rdev
, tmp
, mddev
) {
3172 if (mddev
->max_disks
&&
3173 (rdev
->desc_nr
>= mddev
->max_disks
||
3174 i
> mddev
->max_disks
)) {
3176 "md: %s: %s: only %d devices permitted\n",
3177 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3179 kick_rdev_from_array(rdev
);
3182 if (rdev
!= freshest
)
3183 if (super_types
[mddev
->major_version
].
3184 validate_super(mddev
, rdev
)) {
3185 printk(KERN_WARNING
"md: kicking non-fresh %s"
3187 bdevname(rdev
->bdev
,b
));
3188 kick_rdev_from_array(rdev
);
3191 if (mddev
->level
== LEVEL_MULTIPATH
) {
3192 rdev
->desc_nr
= i
++;
3193 rdev
->raid_disk
= rdev
->desc_nr
;
3194 set_bit(In_sync
, &rdev
->flags
);
3195 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3196 rdev
->raid_disk
= -1;
3197 clear_bit(In_sync
, &rdev
->flags
);
3202 /* Read a fixed-point number.
3203 * Numbers in sysfs attributes should be in "standard" units where
3204 * possible, so time should be in seconds.
3205 * However we internally use a a much smaller unit such as
3206 * milliseconds or jiffies.
3207 * This function takes a decimal number with a possible fractional
3208 * component, and produces an integer which is the result of
3209 * multiplying that number by 10^'scale'.
3210 * all without any floating-point arithmetic.
3212 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3214 unsigned long result
= 0;
3216 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3219 else if (decimals
< scale
) {
3222 result
= result
* 10 + value
;
3234 while (decimals
< scale
) {
3243 static void md_safemode_timeout(unsigned long data
);
3246 safe_delay_show(struct mddev
*mddev
, char *page
)
3248 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3249 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3252 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3256 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3259 mddev
->safemode_delay
= 0;
3261 unsigned long old_delay
= mddev
->safemode_delay
;
3262 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3263 if (mddev
->safemode_delay
== 0)
3264 mddev
->safemode_delay
= 1;
3265 if (mddev
->safemode_delay
< old_delay
)
3266 md_safemode_timeout((unsigned long)mddev
);
3270 static struct md_sysfs_entry md_safe_delay
=
3271 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3274 level_show(struct mddev
*mddev
, char *page
)
3276 struct md_personality
*p
= mddev
->pers
;
3278 return sprintf(page
, "%s\n", p
->name
);
3279 else if (mddev
->clevel
[0])
3280 return sprintf(page
, "%s\n", mddev
->clevel
);
3281 else if (mddev
->level
!= LEVEL_NONE
)
3282 return sprintf(page
, "%d\n", mddev
->level
);
3288 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3292 struct md_personality
*pers
;
3295 struct md_rdev
*rdev
;
3297 if (mddev
->pers
== NULL
) {
3300 if (len
>= sizeof(mddev
->clevel
))
3302 strncpy(mddev
->clevel
, buf
, len
);
3303 if (mddev
->clevel
[len
-1] == '\n')
3305 mddev
->clevel
[len
] = 0;
3306 mddev
->level
= LEVEL_NONE
;
3310 /* request to change the personality. Need to ensure:
3311 * - array is not engaged in resync/recovery/reshape
3312 * - old personality can be suspended
3313 * - new personality will access other array.
3316 if (mddev
->sync_thread
||
3317 mddev
->reshape_position
!= MaxSector
||
3318 mddev
->sysfs_active
)
3321 if (!mddev
->pers
->quiesce
) {
3322 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3323 mdname(mddev
), mddev
->pers
->name
);
3327 /* Now find the new personality */
3328 if (len
== 0 || len
>= sizeof(clevel
))
3330 strncpy(clevel
, buf
, len
);
3331 if (clevel
[len
-1] == '\n')
3334 if (strict_strtol(clevel
, 10, &level
))
3337 if (request_module("md-%s", clevel
) != 0)
3338 request_module("md-level-%s", clevel
);
3339 spin_lock(&pers_lock
);
3340 pers
= find_pers(level
, clevel
);
3341 if (!pers
|| !try_module_get(pers
->owner
)) {
3342 spin_unlock(&pers_lock
);
3343 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3346 spin_unlock(&pers_lock
);
3348 if (pers
== mddev
->pers
) {
3349 /* Nothing to do! */
3350 module_put(pers
->owner
);
3353 if (!pers
->takeover
) {
3354 module_put(pers
->owner
);
3355 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3356 mdname(mddev
), clevel
);
3360 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3361 rdev
->new_raid_disk
= rdev
->raid_disk
;
3363 /* ->takeover must set new_* and/or delta_disks
3364 * if it succeeds, and may set them when it fails.
3366 priv
= pers
->takeover(mddev
);
3368 mddev
->new_level
= mddev
->level
;
3369 mddev
->new_layout
= mddev
->layout
;
3370 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3371 mddev
->raid_disks
-= mddev
->delta_disks
;
3372 mddev
->delta_disks
= 0;
3373 module_put(pers
->owner
);
3374 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3375 mdname(mddev
), clevel
);
3376 return PTR_ERR(priv
);
3379 /* Looks like we have a winner */
3380 mddev_suspend(mddev
);
3381 mddev
->pers
->stop(mddev
);
3383 if (mddev
->pers
->sync_request
== NULL
&&
3384 pers
->sync_request
!= NULL
) {
3385 /* need to add the md_redundancy_group */
3386 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3388 "md: cannot register extra attributes for %s\n",
3390 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3392 if (mddev
->pers
->sync_request
!= NULL
&&
3393 pers
->sync_request
== NULL
) {
3394 /* need to remove the md_redundancy_group */
3395 if (mddev
->to_remove
== NULL
)
3396 mddev
->to_remove
= &md_redundancy_group
;
3399 if (mddev
->pers
->sync_request
== NULL
&&
3401 /* We are converting from a no-redundancy array
3402 * to a redundancy array and metadata is managed
3403 * externally so we need to be sure that writes
3404 * won't block due to a need to transition
3406 * until external management is started.
3409 mddev
->safemode_delay
= 0;
3410 mddev
->safemode
= 0;
3413 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3414 if (rdev
->raid_disk
< 0)
3416 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3417 rdev
->new_raid_disk
= -1;
3418 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3420 sysfs_unlink_rdev(mddev
, rdev
);
3422 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3423 if (rdev
->raid_disk
< 0)
3425 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3427 rdev
->raid_disk
= rdev
->new_raid_disk
;
3428 if (rdev
->raid_disk
< 0)
3429 clear_bit(In_sync
, &rdev
->flags
);
3431 if (sysfs_link_rdev(mddev
, rdev
))
3432 printk(KERN_WARNING
"md: cannot register rd%d"
3433 " for %s after level change\n",
3434 rdev
->raid_disk
, mdname(mddev
));
3438 module_put(mddev
->pers
->owner
);
3440 mddev
->private = priv
;
3441 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3442 mddev
->level
= mddev
->new_level
;
3443 mddev
->layout
= mddev
->new_layout
;
3444 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3445 mddev
->delta_disks
= 0;
3446 mddev
->degraded
= 0;
3447 if (mddev
->pers
->sync_request
== NULL
) {
3448 /* this is now an array without redundancy, so
3449 * it must always be in_sync
3452 del_timer_sync(&mddev
->safemode_timer
);
3455 mddev_resume(mddev
);
3456 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3457 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3458 md_wakeup_thread(mddev
->thread
);
3459 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3460 md_new_event(mddev
);
3464 static struct md_sysfs_entry md_level
=
3465 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3469 layout_show(struct mddev
*mddev
, char *page
)
3471 /* just a number, not meaningful for all levels */
3472 if (mddev
->reshape_position
!= MaxSector
&&
3473 mddev
->layout
!= mddev
->new_layout
)
3474 return sprintf(page
, "%d (%d)\n",
3475 mddev
->new_layout
, mddev
->layout
);
3476 return sprintf(page
, "%d\n", mddev
->layout
);
3480 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3483 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3485 if (!*buf
|| (*e
&& *e
!= '\n'))
3490 if (mddev
->pers
->check_reshape
== NULL
)
3492 mddev
->new_layout
= n
;
3493 err
= mddev
->pers
->check_reshape(mddev
);
3495 mddev
->new_layout
= mddev
->layout
;
3499 mddev
->new_layout
= n
;
3500 if (mddev
->reshape_position
== MaxSector
)
3505 static struct md_sysfs_entry md_layout
=
3506 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3510 raid_disks_show(struct mddev
*mddev
, char *page
)
3512 if (mddev
->raid_disks
== 0)
3514 if (mddev
->reshape_position
!= MaxSector
&&
3515 mddev
->delta_disks
!= 0)
3516 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3517 mddev
->raid_disks
- mddev
->delta_disks
);
3518 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3521 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3524 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3528 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3530 if (!*buf
|| (*e
&& *e
!= '\n'))
3534 rv
= update_raid_disks(mddev
, n
);
3535 else if (mddev
->reshape_position
!= MaxSector
) {
3536 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3537 mddev
->delta_disks
= n
- olddisks
;
3538 mddev
->raid_disks
= n
;
3540 mddev
->raid_disks
= n
;
3541 return rv
? rv
: len
;
3543 static struct md_sysfs_entry md_raid_disks
=
3544 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3547 chunk_size_show(struct mddev
*mddev
, char *page
)
3549 if (mddev
->reshape_position
!= MaxSector
&&
3550 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3551 return sprintf(page
, "%d (%d)\n",
3552 mddev
->new_chunk_sectors
<< 9,
3553 mddev
->chunk_sectors
<< 9);
3554 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3558 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3561 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3563 if (!*buf
|| (*e
&& *e
!= '\n'))
3568 if (mddev
->pers
->check_reshape
== NULL
)
3570 mddev
->new_chunk_sectors
= n
>> 9;
3571 err
= mddev
->pers
->check_reshape(mddev
);
3573 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3577 mddev
->new_chunk_sectors
= n
>> 9;
3578 if (mddev
->reshape_position
== MaxSector
)
3579 mddev
->chunk_sectors
= n
>> 9;
3583 static struct md_sysfs_entry md_chunk_size
=
3584 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3587 resync_start_show(struct mddev
*mddev
, char *page
)
3589 if (mddev
->recovery_cp
== MaxSector
)
3590 return sprintf(page
, "none\n");
3591 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3595 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3598 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3600 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3602 if (cmd_match(buf
, "none"))
3604 else if (!*buf
|| (*e
&& *e
!= '\n'))
3607 mddev
->recovery_cp
= n
;
3610 static struct md_sysfs_entry md_resync_start
=
3611 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3614 * The array state can be:
3617 * No devices, no size, no level
3618 * Equivalent to STOP_ARRAY ioctl
3620 * May have some settings, but array is not active
3621 * all IO results in error
3622 * When written, doesn't tear down array, but just stops it
3623 * suspended (not supported yet)
3624 * All IO requests will block. The array can be reconfigured.
3625 * Writing this, if accepted, will block until array is quiescent
3627 * no resync can happen. no superblocks get written.
3628 * write requests fail
3630 * like readonly, but behaves like 'clean' on a write request.
3632 * clean - no pending writes, but otherwise active.
3633 * When written to inactive array, starts without resync
3634 * If a write request arrives then
3635 * if metadata is known, mark 'dirty' and switch to 'active'.
3636 * if not known, block and switch to write-pending
3637 * If written to an active array that has pending writes, then fails.
3639 * fully active: IO and resync can be happening.
3640 * When written to inactive array, starts with resync
3643 * clean, but writes are blocked waiting for 'active' to be written.
3646 * like active, but no writes have been seen for a while (100msec).
3649 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3650 write_pending
, active_idle
, bad_word
};
3651 static char *array_states
[] = {
3652 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3653 "write-pending", "active-idle", NULL
};
3655 static int match_word(const char *word
, char **list
)
3658 for (n
=0; list
[n
]; n
++)
3659 if (cmd_match(word
, list
[n
]))
3665 array_state_show(struct mddev
*mddev
, char *page
)
3667 enum array_state st
= inactive
;
3680 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3682 else if (mddev
->safemode
)
3688 if (list_empty(&mddev
->disks
) &&
3689 mddev
->raid_disks
== 0 &&
3690 mddev
->dev_sectors
== 0)
3695 return sprintf(page
, "%s\n", array_states
[st
]);
3698 static int do_md_stop(struct mddev
* mddev
, int ro
, int is_open
);
3699 static int md_set_readonly(struct mddev
* mddev
, int is_open
);
3700 static int do_md_run(struct mddev
* mddev
);
3701 static int restart_array(struct mddev
*mddev
);
3704 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3707 enum array_state st
= match_word(buf
, array_states
);
3712 /* stopping an active array */
3713 if (atomic_read(&mddev
->openers
) > 0)
3715 err
= do_md_stop(mddev
, 0, 0);
3718 /* stopping an active array */
3720 if (atomic_read(&mddev
->openers
) > 0)
3722 err
= do_md_stop(mddev
, 2, 0);
3724 err
= 0; /* already inactive */
3727 break; /* not supported yet */
3730 err
= md_set_readonly(mddev
, 0);
3733 set_disk_ro(mddev
->gendisk
, 1);
3734 err
= do_md_run(mddev
);
3740 err
= md_set_readonly(mddev
, 0);
3741 else if (mddev
->ro
== 1)
3742 err
= restart_array(mddev
);
3745 set_disk_ro(mddev
->gendisk
, 0);
3749 err
= do_md_run(mddev
);
3754 restart_array(mddev
);
3755 spin_lock_irq(&mddev
->write_lock
);
3756 if (atomic_read(&mddev
->writes_pending
) == 0) {
3757 if (mddev
->in_sync
== 0) {
3759 if (mddev
->safemode
== 1)
3760 mddev
->safemode
= 0;
3761 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3766 spin_unlock_irq(&mddev
->write_lock
);
3772 restart_array(mddev
);
3773 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3774 wake_up(&mddev
->sb_wait
);
3778 set_disk_ro(mddev
->gendisk
, 0);
3779 err
= do_md_run(mddev
);
3784 /* these cannot be set */
3790 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3794 static struct md_sysfs_entry md_array_state
=
3795 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3798 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
3799 return sprintf(page
, "%d\n",
3800 atomic_read(&mddev
->max_corr_read_errors
));
3804 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3807 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3809 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3810 atomic_set(&mddev
->max_corr_read_errors
, n
);
3816 static struct md_sysfs_entry max_corr_read_errors
=
3817 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3818 max_corrected_read_errors_store
);
3821 null_show(struct mddev
*mddev
, char *page
)
3827 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3829 /* buf must be %d:%d\n? giving major and minor numbers */
3830 /* The new device is added to the array.
3831 * If the array has a persistent superblock, we read the
3832 * superblock to initialise info and check validity.
3833 * Otherwise, only checking done is that in bind_rdev_to_array,
3834 * which mainly checks size.
3837 int major
= simple_strtoul(buf
, &e
, 10);
3840 struct md_rdev
*rdev
;
3843 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3845 minor
= simple_strtoul(e
+1, &e
, 10);
3846 if (*e
&& *e
!= '\n')
3848 dev
= MKDEV(major
, minor
);
3849 if (major
!= MAJOR(dev
) ||
3850 minor
!= MINOR(dev
))
3854 if (mddev
->persistent
) {
3855 rdev
= md_import_device(dev
, mddev
->major_version
,
3856 mddev
->minor_version
);
3857 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3858 struct md_rdev
*rdev0
3859 = list_entry(mddev
->disks
.next
,
3860 struct md_rdev
, same_set
);
3861 err
= super_types
[mddev
->major_version
]
3862 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3866 } else if (mddev
->external
)
3867 rdev
= md_import_device(dev
, -2, -1);
3869 rdev
= md_import_device(dev
, -1, -1);
3872 return PTR_ERR(rdev
);
3873 err
= bind_rdev_to_array(rdev
, mddev
);
3877 return err
? err
: len
;
3880 static struct md_sysfs_entry md_new_device
=
3881 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3884 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3887 unsigned long chunk
, end_chunk
;
3891 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3893 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3894 if (buf
== end
) break;
3895 if (*end
== '-') { /* range */
3897 end_chunk
= simple_strtoul(buf
, &end
, 0);
3898 if (buf
== end
) break;
3900 if (*end
&& !isspace(*end
)) break;
3901 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3902 buf
= skip_spaces(end
);
3904 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3909 static struct md_sysfs_entry md_bitmap
=
3910 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3913 size_show(struct mddev
*mddev
, char *page
)
3915 return sprintf(page
, "%llu\n",
3916 (unsigned long long)mddev
->dev_sectors
/ 2);
3919 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
3922 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3924 /* If array is inactive, we can reduce the component size, but
3925 * not increase it (except from 0).
3926 * If array is active, we can try an on-line resize
3929 int err
= strict_blocks_to_sectors(buf
, §ors
);
3934 err
= update_size(mddev
, sectors
);
3935 md_update_sb(mddev
, 1);
3937 if (mddev
->dev_sectors
== 0 ||
3938 mddev
->dev_sectors
> sectors
)
3939 mddev
->dev_sectors
= sectors
;
3943 return err
? err
: len
;
3946 static struct md_sysfs_entry md_size
=
3947 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3952 * 'none' for arrays with no metadata (good luck...)
3953 * 'external' for arrays with externally managed metadata,
3954 * or N.M for internally known formats
3957 metadata_show(struct mddev
*mddev
, char *page
)
3959 if (mddev
->persistent
)
3960 return sprintf(page
, "%d.%d\n",
3961 mddev
->major_version
, mddev
->minor_version
);
3962 else if (mddev
->external
)
3963 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3965 return sprintf(page
, "none\n");
3969 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3973 /* Changing the details of 'external' metadata is
3974 * always permitted. Otherwise there must be
3975 * no devices attached to the array.
3977 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3979 else if (!list_empty(&mddev
->disks
))
3982 if (cmd_match(buf
, "none")) {
3983 mddev
->persistent
= 0;
3984 mddev
->external
= 0;
3985 mddev
->major_version
= 0;
3986 mddev
->minor_version
= 90;
3989 if (strncmp(buf
, "external:", 9) == 0) {
3990 size_t namelen
= len
-9;
3991 if (namelen
>= sizeof(mddev
->metadata_type
))
3992 namelen
= sizeof(mddev
->metadata_type
)-1;
3993 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3994 mddev
->metadata_type
[namelen
] = 0;
3995 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3996 mddev
->metadata_type
[--namelen
] = 0;
3997 mddev
->persistent
= 0;
3998 mddev
->external
= 1;
3999 mddev
->major_version
= 0;
4000 mddev
->minor_version
= 90;
4003 major
= simple_strtoul(buf
, &e
, 10);
4004 if (e
==buf
|| *e
!= '.')
4007 minor
= simple_strtoul(buf
, &e
, 10);
4008 if (e
==buf
|| (*e
&& *e
!= '\n') )
4010 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4012 mddev
->major_version
= major
;
4013 mddev
->minor_version
= minor
;
4014 mddev
->persistent
= 1;
4015 mddev
->external
= 0;
4019 static struct md_sysfs_entry md_metadata
=
4020 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4023 action_show(struct mddev
*mddev
, char *page
)
4025 char *type
= "idle";
4026 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4028 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4029 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4030 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4032 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4033 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4035 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4039 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4042 return sprintf(page
, "%s\n", type
);
4045 static void reap_sync_thread(struct mddev
*mddev
);
4048 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4050 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4053 if (cmd_match(page
, "frozen"))
4054 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4056 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4058 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4059 if (mddev
->sync_thread
) {
4060 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4061 reap_sync_thread(mddev
);
4063 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4064 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4066 else if (cmd_match(page
, "resync"))
4067 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4068 else if (cmd_match(page
, "recover")) {
4069 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4070 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4071 } else if (cmd_match(page
, "reshape")) {
4073 if (mddev
->pers
->start_reshape
== NULL
)
4075 err
= mddev
->pers
->start_reshape(mddev
);
4078 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4080 if (cmd_match(page
, "check"))
4081 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4082 else if (!cmd_match(page
, "repair"))
4084 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4085 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4087 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4088 md_wakeup_thread(mddev
->thread
);
4089 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4094 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4096 return sprintf(page
, "%llu\n",
4097 (unsigned long long) mddev
->resync_mismatches
);
4100 static struct md_sysfs_entry md_scan_mode
=
4101 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4104 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4107 sync_min_show(struct mddev
*mddev
, char *page
)
4109 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4110 mddev
->sync_speed_min
? "local": "system");
4114 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4118 if (strncmp(buf
, "system", 6)==0) {
4119 mddev
->sync_speed_min
= 0;
4122 min
= simple_strtoul(buf
, &e
, 10);
4123 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4125 mddev
->sync_speed_min
= min
;
4129 static struct md_sysfs_entry md_sync_min
=
4130 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4133 sync_max_show(struct mddev
*mddev
, char *page
)
4135 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4136 mddev
->sync_speed_max
? "local": "system");
4140 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4144 if (strncmp(buf
, "system", 6)==0) {
4145 mddev
->sync_speed_max
= 0;
4148 max
= simple_strtoul(buf
, &e
, 10);
4149 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4151 mddev
->sync_speed_max
= max
;
4155 static struct md_sysfs_entry md_sync_max
=
4156 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4159 degraded_show(struct mddev
*mddev
, char *page
)
4161 return sprintf(page
, "%d\n", mddev
->degraded
);
4163 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4166 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4168 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4172 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4176 if (strict_strtol(buf
, 10, &n
))
4179 if (n
!= 0 && n
!= 1)
4182 mddev
->parallel_resync
= n
;
4184 if (mddev
->sync_thread
)
4185 wake_up(&resync_wait
);
4190 /* force parallel resync, even with shared block devices */
4191 static struct md_sysfs_entry md_sync_force_parallel
=
4192 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4193 sync_force_parallel_show
, sync_force_parallel_store
);
4196 sync_speed_show(struct mddev
*mddev
, char *page
)
4198 unsigned long resync
, dt
, db
;
4199 if (mddev
->curr_resync
== 0)
4200 return sprintf(page
, "none\n");
4201 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4202 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4204 db
= resync
- mddev
->resync_mark_cnt
;
4205 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4208 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4211 sync_completed_show(struct mddev
*mddev
, char *page
)
4213 unsigned long long max_sectors
, resync
;
4215 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4216 return sprintf(page
, "none\n");
4218 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4219 max_sectors
= mddev
->resync_max_sectors
;
4221 max_sectors
= mddev
->dev_sectors
;
4223 resync
= mddev
->curr_resync_completed
;
4224 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4227 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4230 min_sync_show(struct mddev
*mddev
, char *page
)
4232 return sprintf(page
, "%llu\n",
4233 (unsigned long long)mddev
->resync_min
);
4236 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4238 unsigned long long min
;
4239 if (strict_strtoull(buf
, 10, &min
))
4241 if (min
> mddev
->resync_max
)
4243 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4246 /* Must be a multiple of chunk_size */
4247 if (mddev
->chunk_sectors
) {
4248 sector_t temp
= min
;
4249 if (sector_div(temp
, mddev
->chunk_sectors
))
4252 mddev
->resync_min
= min
;
4257 static struct md_sysfs_entry md_min_sync
=
4258 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4261 max_sync_show(struct mddev
*mddev
, char *page
)
4263 if (mddev
->resync_max
== MaxSector
)
4264 return sprintf(page
, "max\n");
4266 return sprintf(page
, "%llu\n",
4267 (unsigned long long)mddev
->resync_max
);
4270 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4272 if (strncmp(buf
, "max", 3) == 0)
4273 mddev
->resync_max
= MaxSector
;
4275 unsigned long long max
;
4276 if (strict_strtoull(buf
, 10, &max
))
4278 if (max
< mddev
->resync_min
)
4280 if (max
< mddev
->resync_max
&&
4282 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4285 /* Must be a multiple of chunk_size */
4286 if (mddev
->chunk_sectors
) {
4287 sector_t temp
= max
;
4288 if (sector_div(temp
, mddev
->chunk_sectors
))
4291 mddev
->resync_max
= max
;
4293 wake_up(&mddev
->recovery_wait
);
4297 static struct md_sysfs_entry md_max_sync
=
4298 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4301 suspend_lo_show(struct mddev
*mddev
, char *page
)
4303 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4307 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4310 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4311 unsigned long long old
= mddev
->suspend_lo
;
4313 if (mddev
->pers
== NULL
||
4314 mddev
->pers
->quiesce
== NULL
)
4316 if (buf
== e
|| (*e
&& *e
!= '\n'))
4319 mddev
->suspend_lo
= new;
4321 /* Shrinking suspended region */
4322 mddev
->pers
->quiesce(mddev
, 2);
4324 /* Expanding suspended region - need to wait */
4325 mddev
->pers
->quiesce(mddev
, 1);
4326 mddev
->pers
->quiesce(mddev
, 0);
4330 static struct md_sysfs_entry md_suspend_lo
=
4331 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4335 suspend_hi_show(struct mddev
*mddev
, char *page
)
4337 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4341 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4344 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4345 unsigned long long old
= mddev
->suspend_hi
;
4347 if (mddev
->pers
== NULL
||
4348 mddev
->pers
->quiesce
== NULL
)
4350 if (buf
== e
|| (*e
&& *e
!= '\n'))
4353 mddev
->suspend_hi
= new;
4355 /* Shrinking suspended region */
4356 mddev
->pers
->quiesce(mddev
, 2);
4358 /* Expanding suspended region - need to wait */
4359 mddev
->pers
->quiesce(mddev
, 1);
4360 mddev
->pers
->quiesce(mddev
, 0);
4364 static struct md_sysfs_entry md_suspend_hi
=
4365 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4368 reshape_position_show(struct mddev
*mddev
, char *page
)
4370 if (mddev
->reshape_position
!= MaxSector
)
4371 return sprintf(page
, "%llu\n",
4372 (unsigned long long)mddev
->reshape_position
);
4373 strcpy(page
, "none\n");
4378 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4381 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4384 if (buf
== e
|| (*e
&& *e
!= '\n'))
4386 mddev
->reshape_position
= new;
4387 mddev
->delta_disks
= 0;
4388 mddev
->new_level
= mddev
->level
;
4389 mddev
->new_layout
= mddev
->layout
;
4390 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4394 static struct md_sysfs_entry md_reshape_position
=
4395 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4396 reshape_position_store
);
4399 array_size_show(struct mddev
*mddev
, char *page
)
4401 if (mddev
->external_size
)
4402 return sprintf(page
, "%llu\n",
4403 (unsigned long long)mddev
->array_sectors
/2);
4405 return sprintf(page
, "default\n");
4409 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4413 if (strncmp(buf
, "default", 7) == 0) {
4415 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4417 sectors
= mddev
->array_sectors
;
4419 mddev
->external_size
= 0;
4421 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4423 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4426 mddev
->external_size
= 1;
4429 mddev
->array_sectors
= sectors
;
4431 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4432 revalidate_disk(mddev
->gendisk
);
4437 static struct md_sysfs_entry md_array_size
=
4438 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4441 static struct attribute
*md_default_attrs
[] = {
4444 &md_raid_disks
.attr
,
4445 &md_chunk_size
.attr
,
4447 &md_resync_start
.attr
,
4449 &md_new_device
.attr
,
4450 &md_safe_delay
.attr
,
4451 &md_array_state
.attr
,
4452 &md_reshape_position
.attr
,
4453 &md_array_size
.attr
,
4454 &max_corr_read_errors
.attr
,
4458 static struct attribute
*md_redundancy_attrs
[] = {
4460 &md_mismatches
.attr
,
4463 &md_sync_speed
.attr
,
4464 &md_sync_force_parallel
.attr
,
4465 &md_sync_completed
.attr
,
4468 &md_suspend_lo
.attr
,
4469 &md_suspend_hi
.attr
,
4474 static struct attribute_group md_redundancy_group
= {
4476 .attrs
= md_redundancy_attrs
,
4481 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4483 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4484 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4489 rv
= mddev_lock(mddev
);
4491 rv
= entry
->show(mddev
, page
);
4492 mddev_unlock(mddev
);
4498 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4499 const char *page
, size_t length
)
4501 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4502 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4507 if (!capable(CAP_SYS_ADMIN
))
4509 rv
= mddev_lock(mddev
);
4510 if (mddev
->hold_active
== UNTIL_IOCTL
)
4511 mddev
->hold_active
= 0;
4513 rv
= entry
->store(mddev
, page
, length
);
4514 mddev_unlock(mddev
);
4519 static void md_free(struct kobject
*ko
)
4521 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4523 if (mddev
->sysfs_state
)
4524 sysfs_put(mddev
->sysfs_state
);
4526 if (mddev
->gendisk
) {
4527 del_gendisk(mddev
->gendisk
);
4528 put_disk(mddev
->gendisk
);
4531 blk_cleanup_queue(mddev
->queue
);
4536 static const struct sysfs_ops md_sysfs_ops
= {
4537 .show
= md_attr_show
,
4538 .store
= md_attr_store
,
4540 static struct kobj_type md_ktype
= {
4542 .sysfs_ops
= &md_sysfs_ops
,
4543 .default_attrs
= md_default_attrs
,
4548 static void mddev_delayed_delete(struct work_struct
*ws
)
4550 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4552 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4553 kobject_del(&mddev
->kobj
);
4554 kobject_put(&mddev
->kobj
);
4557 static int md_alloc(dev_t dev
, char *name
)
4559 static DEFINE_MUTEX(disks_mutex
);
4560 struct mddev
*mddev
= mddev_find(dev
);
4561 struct gendisk
*disk
;
4570 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4571 shift
= partitioned
? MdpMinorShift
: 0;
4572 unit
= MINOR(mddev
->unit
) >> shift
;
4574 /* wait for any previous instance of this device to be
4575 * completely removed (mddev_delayed_delete).
4577 flush_workqueue(md_misc_wq
);
4579 mutex_lock(&disks_mutex
);
4585 /* Need to ensure that 'name' is not a duplicate.
4587 struct mddev
*mddev2
;
4588 spin_lock(&all_mddevs_lock
);
4590 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4591 if (mddev2
->gendisk
&&
4592 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4593 spin_unlock(&all_mddevs_lock
);
4596 spin_unlock(&all_mddevs_lock
);
4600 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4603 mddev
->queue
->queuedata
= mddev
;
4605 blk_queue_make_request(mddev
->queue
, md_make_request
);
4607 disk
= alloc_disk(1 << shift
);
4609 blk_cleanup_queue(mddev
->queue
);
4610 mddev
->queue
= NULL
;
4613 disk
->major
= MAJOR(mddev
->unit
);
4614 disk
->first_minor
= unit
<< shift
;
4616 strcpy(disk
->disk_name
, name
);
4617 else if (partitioned
)
4618 sprintf(disk
->disk_name
, "md_d%d", unit
);
4620 sprintf(disk
->disk_name
, "md%d", unit
);
4621 disk
->fops
= &md_fops
;
4622 disk
->private_data
= mddev
;
4623 disk
->queue
= mddev
->queue
;
4624 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4625 /* Allow extended partitions. This makes the
4626 * 'mdp' device redundant, but we can't really
4629 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4630 mddev
->gendisk
= disk
;
4631 /* As soon as we call add_disk(), another thread could get
4632 * through to md_open, so make sure it doesn't get too far
4634 mutex_lock(&mddev
->open_mutex
);
4637 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4638 &disk_to_dev(disk
)->kobj
, "%s", "md");
4640 /* This isn't possible, but as kobject_init_and_add is marked
4641 * __must_check, we must do something with the result
4643 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4647 if (mddev
->kobj
.sd
&&
4648 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4649 printk(KERN_DEBUG
"pointless warning\n");
4650 mutex_unlock(&mddev
->open_mutex
);
4652 mutex_unlock(&disks_mutex
);
4653 if (!error
&& mddev
->kobj
.sd
) {
4654 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4655 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4661 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4663 md_alloc(dev
, NULL
);
4667 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4669 /* val must be "md_*" where * is not all digits.
4670 * We allocate an array with a large free minor number, and
4671 * set the name to val. val must not already be an active name.
4673 int len
= strlen(val
);
4674 char buf
[DISK_NAME_LEN
];
4676 while (len
&& val
[len
-1] == '\n')
4678 if (len
>= DISK_NAME_LEN
)
4680 strlcpy(buf
, val
, len
+1);
4681 if (strncmp(buf
, "md_", 3) != 0)
4683 return md_alloc(0, buf
);
4686 static void md_safemode_timeout(unsigned long data
)
4688 struct mddev
*mddev
= (struct mddev
*) data
;
4690 if (!atomic_read(&mddev
->writes_pending
)) {
4691 mddev
->safemode
= 1;
4692 if (mddev
->external
)
4693 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4695 md_wakeup_thread(mddev
->thread
);
4698 static int start_dirty_degraded
;
4700 int md_run(struct mddev
*mddev
)
4703 struct md_rdev
*rdev
;
4704 struct md_personality
*pers
;
4706 if (list_empty(&mddev
->disks
))
4707 /* cannot run an array with no devices.. */
4712 /* Cannot run until previous stop completes properly */
4713 if (mddev
->sysfs_active
)
4717 * Analyze all RAID superblock(s)
4719 if (!mddev
->raid_disks
) {
4720 if (!mddev
->persistent
)
4725 if (mddev
->level
!= LEVEL_NONE
)
4726 request_module("md-level-%d", mddev
->level
);
4727 else if (mddev
->clevel
[0])
4728 request_module("md-%s", mddev
->clevel
);
4731 * Drop all container device buffers, from now on
4732 * the only valid external interface is through the md
4735 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4736 if (test_bit(Faulty
, &rdev
->flags
))
4738 sync_blockdev(rdev
->bdev
);
4739 invalidate_bdev(rdev
->bdev
);
4741 /* perform some consistency tests on the device.
4742 * We don't want the data to overlap the metadata,
4743 * Internal Bitmap issues have been handled elsewhere.
4745 if (rdev
->meta_bdev
) {
4746 /* Nothing to check */;
4747 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4748 if (mddev
->dev_sectors
&&
4749 rdev
->data_offset
+ mddev
->dev_sectors
4751 printk("md: %s: data overlaps metadata\n",
4756 if (rdev
->sb_start
+ rdev
->sb_size
/512
4757 > rdev
->data_offset
) {
4758 printk("md: %s: metadata overlaps data\n",
4763 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4766 if (mddev
->bio_set
== NULL
)
4767 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
4768 sizeof(struct mddev
*));
4770 spin_lock(&pers_lock
);
4771 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4772 if (!pers
|| !try_module_get(pers
->owner
)) {
4773 spin_unlock(&pers_lock
);
4774 if (mddev
->level
!= LEVEL_NONE
)
4775 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4778 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4783 spin_unlock(&pers_lock
);
4784 if (mddev
->level
!= pers
->level
) {
4785 mddev
->level
= pers
->level
;
4786 mddev
->new_level
= pers
->level
;
4788 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4790 if (mddev
->reshape_position
!= MaxSector
&&
4791 pers
->start_reshape
== NULL
) {
4792 /* This personality cannot handle reshaping... */
4794 module_put(pers
->owner
);
4798 if (pers
->sync_request
) {
4799 /* Warn if this is a potentially silly
4802 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4803 struct md_rdev
*rdev2
;
4806 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4807 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4809 rdev
->bdev
->bd_contains
==
4810 rdev2
->bdev
->bd_contains
) {
4812 "%s: WARNING: %s appears to be"
4813 " on the same physical disk as"
4816 bdevname(rdev
->bdev
,b
),
4817 bdevname(rdev2
->bdev
,b2
));
4824 "True protection against single-disk"
4825 " failure might be compromised.\n");
4828 mddev
->recovery
= 0;
4829 /* may be over-ridden by personality */
4830 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4832 mddev
->ok_start_degraded
= start_dirty_degraded
;
4834 if (start_readonly
&& mddev
->ro
== 0)
4835 mddev
->ro
= 2; /* read-only, but switch on first write */
4837 err
= mddev
->pers
->run(mddev
);
4839 printk(KERN_ERR
"md: pers->run() failed ...\n");
4840 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4841 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4842 " but 'external_size' not in effect?\n", __func__
);
4844 "md: invalid array_size %llu > default size %llu\n",
4845 (unsigned long long)mddev
->array_sectors
/ 2,
4846 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4848 mddev
->pers
->stop(mddev
);
4850 if (err
== 0 && mddev
->pers
->sync_request
) {
4851 err
= bitmap_create(mddev
);
4853 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4854 mdname(mddev
), err
);
4855 mddev
->pers
->stop(mddev
);
4859 module_put(mddev
->pers
->owner
);
4861 bitmap_destroy(mddev
);
4864 if (mddev
->pers
->sync_request
) {
4865 if (mddev
->kobj
.sd
&&
4866 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4868 "md: cannot register extra attributes for %s\n",
4870 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4871 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4874 atomic_set(&mddev
->writes_pending
,0);
4875 atomic_set(&mddev
->max_corr_read_errors
,
4876 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4877 mddev
->safemode
= 0;
4878 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4879 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4880 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4884 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4885 if (rdev
->raid_disk
>= 0)
4886 if (sysfs_link_rdev(mddev
, rdev
))
4887 /* failure here is OK */;
4889 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4892 md_update_sb(mddev
, 0);
4894 md_new_event(mddev
);
4895 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4896 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4897 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4900 EXPORT_SYMBOL_GPL(md_run
);
4902 static int do_md_run(struct mddev
*mddev
)
4906 err
= md_run(mddev
);
4909 err
= bitmap_load(mddev
);
4911 bitmap_destroy(mddev
);
4915 md_wakeup_thread(mddev
->thread
);
4916 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4918 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4919 revalidate_disk(mddev
->gendisk
);
4921 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4926 static int restart_array(struct mddev
*mddev
)
4928 struct gendisk
*disk
= mddev
->gendisk
;
4930 /* Complain if it has no devices */
4931 if (list_empty(&mddev
->disks
))
4937 mddev
->safemode
= 0;
4939 set_disk_ro(disk
, 0);
4940 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4942 /* Kick recovery or resync if necessary */
4943 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4944 md_wakeup_thread(mddev
->thread
);
4945 md_wakeup_thread(mddev
->sync_thread
);
4946 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4950 /* similar to deny_write_access, but accounts for our holding a reference
4951 * to the file ourselves */
4952 static int deny_bitmap_write_access(struct file
* file
)
4954 struct inode
*inode
= file
->f_mapping
->host
;
4956 spin_lock(&inode
->i_lock
);
4957 if (atomic_read(&inode
->i_writecount
) > 1) {
4958 spin_unlock(&inode
->i_lock
);
4961 atomic_set(&inode
->i_writecount
, -1);
4962 spin_unlock(&inode
->i_lock
);
4967 void restore_bitmap_write_access(struct file
*file
)
4969 struct inode
*inode
= file
->f_mapping
->host
;
4971 spin_lock(&inode
->i_lock
);
4972 atomic_set(&inode
->i_writecount
, 1);
4973 spin_unlock(&inode
->i_lock
);
4976 static void md_clean(struct mddev
*mddev
)
4978 mddev
->array_sectors
= 0;
4979 mddev
->external_size
= 0;
4980 mddev
->dev_sectors
= 0;
4981 mddev
->raid_disks
= 0;
4982 mddev
->recovery_cp
= 0;
4983 mddev
->resync_min
= 0;
4984 mddev
->resync_max
= MaxSector
;
4985 mddev
->reshape_position
= MaxSector
;
4986 mddev
->external
= 0;
4987 mddev
->persistent
= 0;
4988 mddev
->level
= LEVEL_NONE
;
4989 mddev
->clevel
[0] = 0;
4992 mddev
->metadata_type
[0] = 0;
4993 mddev
->chunk_sectors
= 0;
4994 mddev
->ctime
= mddev
->utime
= 0;
4996 mddev
->max_disks
= 0;
4998 mddev
->can_decrease_events
= 0;
4999 mddev
->delta_disks
= 0;
5000 mddev
->new_level
= LEVEL_NONE
;
5001 mddev
->new_layout
= 0;
5002 mddev
->new_chunk_sectors
= 0;
5003 mddev
->curr_resync
= 0;
5004 mddev
->resync_mismatches
= 0;
5005 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5006 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5007 mddev
->recovery
= 0;
5010 mddev
->degraded
= 0;
5011 mddev
->safemode
= 0;
5012 mddev
->bitmap_info
.offset
= 0;
5013 mddev
->bitmap_info
.default_offset
= 0;
5014 mddev
->bitmap_info
.chunksize
= 0;
5015 mddev
->bitmap_info
.daemon_sleep
= 0;
5016 mddev
->bitmap_info
.max_write_behind
= 0;
5019 static void __md_stop_writes(struct mddev
*mddev
)
5021 if (mddev
->sync_thread
) {
5022 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5023 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5024 reap_sync_thread(mddev
);
5027 del_timer_sync(&mddev
->safemode_timer
);
5029 bitmap_flush(mddev
);
5030 md_super_wait(mddev
);
5032 if (!mddev
->in_sync
|| mddev
->flags
) {
5033 /* mark array as shutdown cleanly */
5035 md_update_sb(mddev
, 1);
5039 void md_stop_writes(struct mddev
*mddev
)
5042 __md_stop_writes(mddev
);
5043 mddev_unlock(mddev
);
5045 EXPORT_SYMBOL_GPL(md_stop_writes
);
5047 void md_stop(struct mddev
*mddev
)
5050 mddev
->pers
->stop(mddev
);
5051 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5052 mddev
->to_remove
= &md_redundancy_group
;
5053 module_put(mddev
->pers
->owner
);
5055 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5057 EXPORT_SYMBOL_GPL(md_stop
);
5059 static int md_set_readonly(struct mddev
*mddev
, int is_open
)
5062 mutex_lock(&mddev
->open_mutex
);
5063 if (atomic_read(&mddev
->openers
) > is_open
) {
5064 printk("md: %s still in use.\n",mdname(mddev
));
5069 __md_stop_writes(mddev
);
5075 set_disk_ro(mddev
->gendisk
, 1);
5076 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5077 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5081 mutex_unlock(&mddev
->open_mutex
);
5086 * 0 - completely stop and dis-assemble array
5087 * 2 - stop but do not disassemble array
5089 static int do_md_stop(struct mddev
* mddev
, int mode
, int is_open
)
5091 struct gendisk
*disk
= mddev
->gendisk
;
5092 struct md_rdev
*rdev
;
5094 mutex_lock(&mddev
->open_mutex
);
5095 if (atomic_read(&mddev
->openers
) > is_open
||
5096 mddev
->sysfs_active
) {
5097 printk("md: %s still in use.\n",mdname(mddev
));
5098 mutex_unlock(&mddev
->open_mutex
);
5104 set_disk_ro(disk
, 0);
5106 __md_stop_writes(mddev
);
5108 mddev
->queue
->merge_bvec_fn
= NULL
;
5109 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5111 /* tell userspace to handle 'inactive' */
5112 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5114 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5115 if (rdev
->raid_disk
>= 0)
5116 sysfs_unlink_rdev(mddev
, rdev
);
5118 set_capacity(disk
, 0);
5119 mutex_unlock(&mddev
->open_mutex
);
5121 revalidate_disk(disk
);
5126 mutex_unlock(&mddev
->open_mutex
);
5128 * Free resources if final stop
5131 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5133 bitmap_destroy(mddev
);
5134 if (mddev
->bitmap_info
.file
) {
5135 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5136 fput(mddev
->bitmap_info
.file
);
5137 mddev
->bitmap_info
.file
= NULL
;
5139 mddev
->bitmap_info
.offset
= 0;
5141 export_array(mddev
);
5144 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5145 if (mddev
->hold_active
== UNTIL_STOP
)
5146 mddev
->hold_active
= 0;
5148 blk_integrity_unregister(disk
);
5149 md_new_event(mddev
);
5150 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5155 static void autorun_array(struct mddev
*mddev
)
5157 struct md_rdev
*rdev
;
5160 if (list_empty(&mddev
->disks
))
5163 printk(KERN_INFO
"md: running: ");
5165 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5166 char b
[BDEVNAME_SIZE
];
5167 printk("<%s>", bdevname(rdev
->bdev
,b
));
5171 err
= do_md_run(mddev
);
5173 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5174 do_md_stop(mddev
, 0, 0);
5179 * lets try to run arrays based on all disks that have arrived
5180 * until now. (those are in pending_raid_disks)
5182 * the method: pick the first pending disk, collect all disks with
5183 * the same UUID, remove all from the pending list and put them into
5184 * the 'same_array' list. Then order this list based on superblock
5185 * update time (freshest comes first), kick out 'old' disks and
5186 * compare superblocks. If everything's fine then run it.
5188 * If "unit" is allocated, then bump its reference count
5190 static void autorun_devices(int part
)
5192 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5193 struct mddev
*mddev
;
5194 char b
[BDEVNAME_SIZE
];
5196 printk(KERN_INFO
"md: autorun ...\n");
5197 while (!list_empty(&pending_raid_disks
)) {
5200 LIST_HEAD(candidates
);
5201 rdev0
= list_entry(pending_raid_disks
.next
,
5202 struct md_rdev
, same_set
);
5204 printk(KERN_INFO
"md: considering %s ...\n",
5205 bdevname(rdev0
->bdev
,b
));
5206 INIT_LIST_HEAD(&candidates
);
5207 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5208 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5209 printk(KERN_INFO
"md: adding %s ...\n",
5210 bdevname(rdev
->bdev
,b
));
5211 list_move(&rdev
->same_set
, &candidates
);
5214 * now we have a set of devices, with all of them having
5215 * mostly sane superblocks. It's time to allocate the
5219 dev
= MKDEV(mdp_major
,
5220 rdev0
->preferred_minor
<< MdpMinorShift
);
5221 unit
= MINOR(dev
) >> MdpMinorShift
;
5223 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5226 if (rdev0
->preferred_minor
!= unit
) {
5227 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5228 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5232 md_probe(dev
, NULL
, NULL
);
5233 mddev
= mddev_find(dev
);
5234 if (!mddev
|| !mddev
->gendisk
) {
5238 "md: cannot allocate memory for md drive.\n");
5241 if (mddev_lock(mddev
))
5242 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5244 else if (mddev
->raid_disks
|| mddev
->major_version
5245 || !list_empty(&mddev
->disks
)) {
5247 "md: %s already running, cannot run %s\n",
5248 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5249 mddev_unlock(mddev
);
5251 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5252 mddev
->persistent
= 1;
5253 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5254 list_del_init(&rdev
->same_set
);
5255 if (bind_rdev_to_array(rdev
, mddev
))
5258 autorun_array(mddev
);
5259 mddev_unlock(mddev
);
5261 /* on success, candidates will be empty, on error
5264 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5265 list_del_init(&rdev
->same_set
);
5270 printk(KERN_INFO
"md: ... autorun DONE.\n");
5272 #endif /* !MODULE */
5274 static int get_version(void __user
* arg
)
5278 ver
.major
= MD_MAJOR_VERSION
;
5279 ver
.minor
= MD_MINOR_VERSION
;
5280 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5282 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5288 static int get_array_info(struct mddev
* mddev
, void __user
* arg
)
5290 mdu_array_info_t info
;
5291 int nr
,working
,insync
,failed
,spare
;
5292 struct md_rdev
*rdev
;
5294 nr
=working
=insync
=failed
=spare
=0;
5295 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5297 if (test_bit(Faulty
, &rdev
->flags
))
5301 if (test_bit(In_sync
, &rdev
->flags
))
5308 info
.major_version
= mddev
->major_version
;
5309 info
.minor_version
= mddev
->minor_version
;
5310 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5311 info
.ctime
= mddev
->ctime
;
5312 info
.level
= mddev
->level
;
5313 info
.size
= mddev
->dev_sectors
/ 2;
5314 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5317 info
.raid_disks
= mddev
->raid_disks
;
5318 info
.md_minor
= mddev
->md_minor
;
5319 info
.not_persistent
= !mddev
->persistent
;
5321 info
.utime
= mddev
->utime
;
5324 info
.state
= (1<<MD_SB_CLEAN
);
5325 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5326 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5327 info
.active_disks
= insync
;
5328 info
.working_disks
= working
;
5329 info
.failed_disks
= failed
;
5330 info
.spare_disks
= spare
;
5332 info
.layout
= mddev
->layout
;
5333 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5335 if (copy_to_user(arg
, &info
, sizeof(info
)))
5341 static int get_bitmap_file(struct mddev
* mddev
, void __user
* arg
)
5343 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5344 char *ptr
, *buf
= NULL
;
5347 if (md_allow_write(mddev
))
5348 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5350 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5355 /* bitmap disabled, zero the first byte and copy out */
5356 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5357 file
->pathname
[0] = '\0';
5361 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5365 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5369 strcpy(file
->pathname
, ptr
);
5373 if (copy_to_user(arg
, file
, sizeof(*file
)))
5381 static int get_disk_info(struct mddev
* mddev
, void __user
* arg
)
5383 mdu_disk_info_t info
;
5384 struct md_rdev
*rdev
;
5386 if (copy_from_user(&info
, arg
, sizeof(info
)))
5389 rdev
= find_rdev_nr(mddev
, info
.number
);
5391 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5392 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5393 info
.raid_disk
= rdev
->raid_disk
;
5395 if (test_bit(Faulty
, &rdev
->flags
))
5396 info
.state
|= (1<<MD_DISK_FAULTY
);
5397 else if (test_bit(In_sync
, &rdev
->flags
)) {
5398 info
.state
|= (1<<MD_DISK_ACTIVE
);
5399 info
.state
|= (1<<MD_DISK_SYNC
);
5401 if (test_bit(WriteMostly
, &rdev
->flags
))
5402 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5404 info
.major
= info
.minor
= 0;
5405 info
.raid_disk
= -1;
5406 info
.state
= (1<<MD_DISK_REMOVED
);
5409 if (copy_to_user(arg
, &info
, sizeof(info
)))
5415 static int add_new_disk(struct mddev
* mddev
, mdu_disk_info_t
*info
)
5417 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5418 struct md_rdev
*rdev
;
5419 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5421 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5424 if (!mddev
->raid_disks
) {
5426 /* expecting a device which has a superblock */
5427 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5430 "md: md_import_device returned %ld\n",
5432 return PTR_ERR(rdev
);
5434 if (!list_empty(&mddev
->disks
)) {
5435 struct md_rdev
*rdev0
5436 = list_entry(mddev
->disks
.next
,
5437 struct md_rdev
, same_set
);
5438 err
= super_types
[mddev
->major_version
]
5439 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5442 "md: %s has different UUID to %s\n",
5443 bdevname(rdev
->bdev
,b
),
5444 bdevname(rdev0
->bdev
,b2
));
5449 err
= bind_rdev_to_array(rdev
, mddev
);
5456 * add_new_disk can be used once the array is assembled
5457 * to add "hot spares". They must already have a superblock
5462 if (!mddev
->pers
->hot_add_disk
) {
5464 "%s: personality does not support diskops!\n",
5468 if (mddev
->persistent
)
5469 rdev
= md_import_device(dev
, mddev
->major_version
,
5470 mddev
->minor_version
);
5472 rdev
= md_import_device(dev
, -1, -1);
5475 "md: md_import_device returned %ld\n",
5477 return PTR_ERR(rdev
);
5479 /* set saved_raid_disk if appropriate */
5480 if (!mddev
->persistent
) {
5481 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5482 info
->raid_disk
< mddev
->raid_disks
) {
5483 rdev
->raid_disk
= info
->raid_disk
;
5484 set_bit(In_sync
, &rdev
->flags
);
5486 rdev
->raid_disk
= -1;
5488 super_types
[mddev
->major_version
].
5489 validate_super(mddev
, rdev
);
5490 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5491 (!test_bit(In_sync
, &rdev
->flags
) ||
5492 rdev
->raid_disk
!= info
->raid_disk
)) {
5493 /* This was a hot-add request, but events doesn't
5494 * match, so reject it.
5500 if (test_bit(In_sync
, &rdev
->flags
))
5501 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5503 rdev
->saved_raid_disk
= -1;
5505 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5506 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5507 set_bit(WriteMostly
, &rdev
->flags
);
5509 clear_bit(WriteMostly
, &rdev
->flags
);
5511 rdev
->raid_disk
= -1;
5512 err
= bind_rdev_to_array(rdev
, mddev
);
5513 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5514 /* If there is hot_add_disk but no hot_remove_disk
5515 * then added disks for geometry changes,
5516 * and should be added immediately.
5518 super_types
[mddev
->major_version
].
5519 validate_super(mddev
, rdev
);
5520 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5522 unbind_rdev_from_array(rdev
);
5527 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5529 md_update_sb(mddev
, 1);
5530 if (mddev
->degraded
)
5531 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5532 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5534 md_new_event(mddev
);
5535 md_wakeup_thread(mddev
->thread
);
5539 /* otherwise, add_new_disk is only allowed
5540 * for major_version==0 superblocks
5542 if (mddev
->major_version
!= 0) {
5543 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5548 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5550 rdev
= md_import_device(dev
, -1, 0);
5553 "md: error, md_import_device() returned %ld\n",
5555 return PTR_ERR(rdev
);
5557 rdev
->desc_nr
= info
->number
;
5558 if (info
->raid_disk
< mddev
->raid_disks
)
5559 rdev
->raid_disk
= info
->raid_disk
;
5561 rdev
->raid_disk
= -1;
5563 if (rdev
->raid_disk
< mddev
->raid_disks
)
5564 if (info
->state
& (1<<MD_DISK_SYNC
))
5565 set_bit(In_sync
, &rdev
->flags
);
5567 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5568 set_bit(WriteMostly
, &rdev
->flags
);
5570 if (!mddev
->persistent
) {
5571 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5572 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5574 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5575 rdev
->sectors
= rdev
->sb_start
;
5577 err
= bind_rdev_to_array(rdev
, mddev
);
5587 static int hot_remove_disk(struct mddev
* mddev
, dev_t dev
)
5589 char b
[BDEVNAME_SIZE
];
5590 struct md_rdev
*rdev
;
5592 rdev
= find_rdev(mddev
, dev
);
5596 if (rdev
->raid_disk
>= 0)
5599 kick_rdev_from_array(rdev
);
5600 md_update_sb(mddev
, 1);
5601 md_new_event(mddev
);
5605 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5606 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5610 static int hot_add_disk(struct mddev
* mddev
, dev_t dev
)
5612 char b
[BDEVNAME_SIZE
];
5614 struct md_rdev
*rdev
;
5619 if (mddev
->major_version
!= 0) {
5620 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5621 " version-0 superblocks.\n",
5625 if (!mddev
->pers
->hot_add_disk
) {
5627 "%s: personality does not support diskops!\n",
5632 rdev
= md_import_device(dev
, -1, 0);
5635 "md: error, md_import_device() returned %ld\n",
5640 if (mddev
->persistent
)
5641 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5643 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5645 rdev
->sectors
= rdev
->sb_start
;
5647 if (test_bit(Faulty
, &rdev
->flags
)) {
5649 "md: can not hot-add faulty %s disk to %s!\n",
5650 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5654 clear_bit(In_sync
, &rdev
->flags
);
5656 rdev
->saved_raid_disk
= -1;
5657 err
= bind_rdev_to_array(rdev
, mddev
);
5662 * The rest should better be atomic, we can have disk failures
5663 * noticed in interrupt contexts ...
5666 rdev
->raid_disk
= -1;
5668 md_update_sb(mddev
, 1);
5671 * Kick recovery, maybe this spare has to be added to the
5672 * array immediately.
5674 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5675 md_wakeup_thread(mddev
->thread
);
5676 md_new_event(mddev
);
5684 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
5689 if (!mddev
->pers
->quiesce
)
5691 if (mddev
->recovery
|| mddev
->sync_thread
)
5693 /* we should be able to change the bitmap.. */
5699 return -EEXIST
; /* cannot add when bitmap is present */
5700 mddev
->bitmap_info
.file
= fget(fd
);
5702 if (mddev
->bitmap_info
.file
== NULL
) {
5703 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5708 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5710 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5712 fput(mddev
->bitmap_info
.file
);
5713 mddev
->bitmap_info
.file
= NULL
;
5716 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5717 } else if (mddev
->bitmap
== NULL
)
5718 return -ENOENT
; /* cannot remove what isn't there */
5721 mddev
->pers
->quiesce(mddev
, 1);
5723 err
= bitmap_create(mddev
);
5725 err
= bitmap_load(mddev
);
5727 if (fd
< 0 || err
) {
5728 bitmap_destroy(mddev
);
5729 fd
= -1; /* make sure to put the file */
5731 mddev
->pers
->quiesce(mddev
, 0);
5734 if (mddev
->bitmap_info
.file
) {
5735 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5736 fput(mddev
->bitmap_info
.file
);
5738 mddev
->bitmap_info
.file
= NULL
;
5745 * set_array_info is used two different ways
5746 * The original usage is when creating a new array.
5747 * In this usage, raid_disks is > 0 and it together with
5748 * level, size, not_persistent,layout,chunksize determine the
5749 * shape of the array.
5750 * This will always create an array with a type-0.90.0 superblock.
5751 * The newer usage is when assembling an array.
5752 * In this case raid_disks will be 0, and the major_version field is
5753 * use to determine which style super-blocks are to be found on the devices.
5754 * The minor and patch _version numbers are also kept incase the
5755 * super_block handler wishes to interpret them.
5757 static int set_array_info(struct mddev
* mddev
, mdu_array_info_t
*info
)
5760 if (info
->raid_disks
== 0) {
5761 /* just setting version number for superblock loading */
5762 if (info
->major_version
< 0 ||
5763 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5764 super_types
[info
->major_version
].name
== NULL
) {
5765 /* maybe try to auto-load a module? */
5767 "md: superblock version %d not known\n",
5768 info
->major_version
);
5771 mddev
->major_version
= info
->major_version
;
5772 mddev
->minor_version
= info
->minor_version
;
5773 mddev
->patch_version
= info
->patch_version
;
5774 mddev
->persistent
= !info
->not_persistent
;
5775 /* ensure mddev_put doesn't delete this now that there
5776 * is some minimal configuration.
5778 mddev
->ctime
= get_seconds();
5781 mddev
->major_version
= MD_MAJOR_VERSION
;
5782 mddev
->minor_version
= MD_MINOR_VERSION
;
5783 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5784 mddev
->ctime
= get_seconds();
5786 mddev
->level
= info
->level
;
5787 mddev
->clevel
[0] = 0;
5788 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5789 mddev
->raid_disks
= info
->raid_disks
;
5790 /* don't set md_minor, it is determined by which /dev/md* was
5793 if (info
->state
& (1<<MD_SB_CLEAN
))
5794 mddev
->recovery_cp
= MaxSector
;
5796 mddev
->recovery_cp
= 0;
5797 mddev
->persistent
= ! info
->not_persistent
;
5798 mddev
->external
= 0;
5800 mddev
->layout
= info
->layout
;
5801 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5803 mddev
->max_disks
= MD_SB_DISKS
;
5805 if (mddev
->persistent
)
5807 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5809 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5810 mddev
->bitmap_info
.offset
= 0;
5812 mddev
->reshape_position
= MaxSector
;
5815 * Generate a 128 bit UUID
5817 get_random_bytes(mddev
->uuid
, 16);
5819 mddev
->new_level
= mddev
->level
;
5820 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5821 mddev
->new_layout
= mddev
->layout
;
5822 mddev
->delta_disks
= 0;
5827 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
5829 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5831 if (mddev
->external_size
)
5834 mddev
->array_sectors
= array_sectors
;
5836 EXPORT_SYMBOL(md_set_array_sectors
);
5838 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
5840 struct md_rdev
*rdev
;
5842 int fit
= (num_sectors
== 0);
5844 if (mddev
->pers
->resize
== NULL
)
5846 /* The "num_sectors" is the number of sectors of each device that
5847 * is used. This can only make sense for arrays with redundancy.
5848 * linear and raid0 always use whatever space is available. We can only
5849 * consider changing this number if no resync or reconstruction is
5850 * happening, and if the new size is acceptable. It must fit before the
5851 * sb_start or, if that is <data_offset, it must fit before the size
5852 * of each device. If num_sectors is zero, we find the largest size
5855 if (mddev
->sync_thread
)
5858 /* Sorry, cannot grow a bitmap yet, just remove it,
5862 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5863 sector_t avail
= rdev
->sectors
;
5865 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5866 num_sectors
= avail
;
5867 if (avail
< num_sectors
)
5870 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5872 revalidate_disk(mddev
->gendisk
);
5876 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
5879 /* change the number of raid disks */
5880 if (mddev
->pers
->check_reshape
== NULL
)
5882 if (raid_disks
<= 0 ||
5883 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5885 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5887 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5889 rv
= mddev
->pers
->check_reshape(mddev
);
5891 mddev
->delta_disks
= 0;
5897 * update_array_info is used to change the configuration of an
5899 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5900 * fields in the info are checked against the array.
5901 * Any differences that cannot be handled will cause an error.
5902 * Normally, only one change can be managed at a time.
5904 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
5910 /* calculate expected state,ignoring low bits */
5911 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5912 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5914 if (mddev
->major_version
!= info
->major_version
||
5915 mddev
->minor_version
!= info
->minor_version
||
5916 /* mddev->patch_version != info->patch_version || */
5917 mddev
->ctime
!= info
->ctime
||
5918 mddev
->level
!= info
->level
||
5919 /* mddev->layout != info->layout || */
5920 !mddev
->persistent
!= info
->not_persistent
||
5921 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5922 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5923 ((state
^info
->state
) & 0xfffffe00)
5926 /* Check there is only one change */
5927 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5929 if (mddev
->raid_disks
!= info
->raid_disks
)
5931 if (mddev
->layout
!= info
->layout
)
5933 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5940 if (mddev
->layout
!= info
->layout
) {
5942 * we don't need to do anything at the md level, the
5943 * personality will take care of it all.
5945 if (mddev
->pers
->check_reshape
== NULL
)
5948 mddev
->new_layout
= info
->layout
;
5949 rv
= mddev
->pers
->check_reshape(mddev
);
5951 mddev
->new_layout
= mddev
->layout
;
5955 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5956 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5958 if (mddev
->raid_disks
!= info
->raid_disks
)
5959 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5961 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5962 if (mddev
->pers
->quiesce
== NULL
)
5964 if (mddev
->recovery
|| mddev
->sync_thread
)
5966 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5967 /* add the bitmap */
5970 if (mddev
->bitmap_info
.default_offset
== 0)
5972 mddev
->bitmap_info
.offset
=
5973 mddev
->bitmap_info
.default_offset
;
5974 mddev
->pers
->quiesce(mddev
, 1);
5975 rv
= bitmap_create(mddev
);
5977 rv
= bitmap_load(mddev
);
5979 bitmap_destroy(mddev
);
5980 mddev
->pers
->quiesce(mddev
, 0);
5982 /* remove the bitmap */
5985 if (mddev
->bitmap
->file
)
5987 mddev
->pers
->quiesce(mddev
, 1);
5988 bitmap_destroy(mddev
);
5989 mddev
->pers
->quiesce(mddev
, 0);
5990 mddev
->bitmap_info
.offset
= 0;
5993 md_update_sb(mddev
, 1);
5997 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
5999 struct md_rdev
*rdev
;
6001 if (mddev
->pers
== NULL
)
6004 rdev
= find_rdev(mddev
, dev
);
6008 md_error(mddev
, rdev
);
6009 if (!test_bit(Faulty
, &rdev
->flags
))
6015 * We have a problem here : there is no easy way to give a CHS
6016 * virtual geometry. We currently pretend that we have a 2 heads
6017 * 4 sectors (with a BIG number of cylinders...). This drives
6018 * dosfs just mad... ;-)
6020 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6022 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6026 geo
->cylinders
= mddev
->array_sectors
/ 8;
6030 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6031 unsigned int cmd
, unsigned long arg
)
6034 void __user
*argp
= (void __user
*)arg
;
6035 struct mddev
*mddev
= NULL
;
6038 if (!capable(CAP_SYS_ADMIN
))
6042 * Commands dealing with the RAID driver but not any
6048 err
= get_version(argp
);
6051 case PRINT_RAID_DEBUG
:
6059 autostart_arrays(arg
);
6066 * Commands creating/starting a new array:
6069 mddev
= bdev
->bd_disk
->private_data
;
6076 err
= mddev_lock(mddev
);
6079 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6086 case SET_ARRAY_INFO
:
6088 mdu_array_info_t info
;
6090 memset(&info
, 0, sizeof(info
));
6091 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6096 err
= update_array_info(mddev
, &info
);
6098 printk(KERN_WARNING
"md: couldn't update"
6099 " array info. %d\n", err
);
6104 if (!list_empty(&mddev
->disks
)) {
6106 "md: array %s already has disks!\n",
6111 if (mddev
->raid_disks
) {
6113 "md: array %s already initialised!\n",
6118 err
= set_array_info(mddev
, &info
);
6120 printk(KERN_WARNING
"md: couldn't set"
6121 " array info. %d\n", err
);
6131 * Commands querying/configuring an existing array:
6133 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6134 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6135 if ((!mddev
->raid_disks
&& !mddev
->external
)
6136 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6137 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6138 && cmd
!= GET_BITMAP_FILE
) {
6144 * Commands even a read-only array can execute:
6148 case GET_ARRAY_INFO
:
6149 err
= get_array_info(mddev
, argp
);
6152 case GET_BITMAP_FILE
:
6153 err
= get_bitmap_file(mddev
, argp
);
6157 err
= get_disk_info(mddev
, argp
);
6160 case RESTART_ARRAY_RW
:
6161 err
= restart_array(mddev
);
6165 err
= do_md_stop(mddev
, 0, 1);
6169 err
= md_set_readonly(mddev
, 1);
6173 if (get_user(ro
, (int __user
*)(arg
))) {
6179 /* if the bdev is going readonly the value of mddev->ro
6180 * does not matter, no writes are coming
6185 /* are we are already prepared for writes? */
6189 /* transitioning to readauto need only happen for
6190 * arrays that call md_write_start
6193 err
= restart_array(mddev
);
6196 set_disk_ro(mddev
->gendisk
, 0);
6203 * The remaining ioctls are changing the state of the
6204 * superblock, so we do not allow them on read-only arrays.
6205 * However non-MD ioctls (e.g. get-size) will still come through
6206 * here and hit the 'default' below, so only disallow
6207 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6209 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6210 if (mddev
->ro
== 2) {
6212 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6213 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6214 md_wakeup_thread(mddev
->thread
);
6225 mdu_disk_info_t info
;
6226 if (copy_from_user(&info
, argp
, sizeof(info
)))
6229 err
= add_new_disk(mddev
, &info
);
6233 case HOT_REMOVE_DISK
:
6234 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6238 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6241 case SET_DISK_FAULTY
:
6242 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6246 err
= do_md_run(mddev
);
6249 case SET_BITMAP_FILE
:
6250 err
= set_bitmap_file(mddev
, (int)arg
);
6260 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6262 mddev
->hold_active
= 0;
6263 mddev_unlock(mddev
);
6272 #ifdef CONFIG_COMPAT
6273 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6274 unsigned int cmd
, unsigned long arg
)
6277 case HOT_REMOVE_DISK
:
6279 case SET_DISK_FAULTY
:
6280 case SET_BITMAP_FILE
:
6281 /* These take in integer arg, do not convert */
6284 arg
= (unsigned long)compat_ptr(arg
);
6288 return md_ioctl(bdev
, mode
, cmd
, arg
);
6290 #endif /* CONFIG_COMPAT */
6292 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6295 * Succeed if we can lock the mddev, which confirms that
6296 * it isn't being stopped right now.
6298 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
6301 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6302 /* we are racing with mddev_put which is discarding this
6306 /* Wait until bdev->bd_disk is definitely gone */
6307 flush_workqueue(md_misc_wq
);
6308 /* Then retry the open from the top */
6309 return -ERESTARTSYS
;
6311 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6313 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6317 atomic_inc(&mddev
->openers
);
6318 mutex_unlock(&mddev
->open_mutex
);
6320 check_disk_change(bdev
);
6325 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6327 struct mddev
*mddev
= disk
->private_data
;
6330 atomic_dec(&mddev
->openers
);
6336 static int md_media_changed(struct gendisk
*disk
)
6338 struct mddev
*mddev
= disk
->private_data
;
6340 return mddev
->changed
;
6343 static int md_revalidate(struct gendisk
*disk
)
6345 struct mddev
*mddev
= disk
->private_data
;
6350 static const struct block_device_operations md_fops
=
6352 .owner
= THIS_MODULE
,
6354 .release
= md_release
,
6356 #ifdef CONFIG_COMPAT
6357 .compat_ioctl
= md_compat_ioctl
,
6359 .getgeo
= md_getgeo
,
6360 .media_changed
= md_media_changed
,
6361 .revalidate_disk
= md_revalidate
,
6364 static int md_thread(void * arg
)
6366 struct md_thread
*thread
= arg
;
6369 * md_thread is a 'system-thread', it's priority should be very
6370 * high. We avoid resource deadlocks individually in each
6371 * raid personality. (RAID5 does preallocation) We also use RR and
6372 * the very same RT priority as kswapd, thus we will never get
6373 * into a priority inversion deadlock.
6375 * we definitely have to have equal or higher priority than
6376 * bdflush, otherwise bdflush will deadlock if there are too
6377 * many dirty RAID5 blocks.
6380 allow_signal(SIGKILL
);
6381 while (!kthread_should_stop()) {
6383 /* We need to wait INTERRUPTIBLE so that
6384 * we don't add to the load-average.
6385 * That means we need to be sure no signals are
6388 if (signal_pending(current
))
6389 flush_signals(current
);
6391 wait_event_interruptible_timeout
6393 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6394 || kthread_should_stop(),
6397 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6398 if (!kthread_should_stop())
6399 thread
->run(thread
->mddev
);
6405 void md_wakeup_thread(struct md_thread
*thread
)
6408 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6409 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6410 wake_up(&thread
->wqueue
);
6414 struct md_thread
*md_register_thread(void (*run
) (struct mddev
*), struct mddev
*mddev
,
6417 struct md_thread
*thread
;
6419 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
6423 init_waitqueue_head(&thread
->wqueue
);
6426 thread
->mddev
= mddev
;
6427 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6428 thread
->tsk
= kthread_run(md_thread
, thread
,
6430 mdname(thread
->mddev
),
6431 name
?: mddev
->pers
->name
);
6432 if (IS_ERR(thread
->tsk
)) {
6439 void md_unregister_thread(struct md_thread
**threadp
)
6441 struct md_thread
*thread
= *threadp
;
6444 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6445 /* Locking ensures that mddev_unlock does not wake_up a
6446 * non-existent thread
6448 spin_lock(&pers_lock
);
6450 spin_unlock(&pers_lock
);
6452 kthread_stop(thread
->tsk
);
6456 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
6463 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6466 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6468 mddev
->pers
->error_handler(mddev
,rdev
);
6469 if (mddev
->degraded
)
6470 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6471 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6472 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6473 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6474 md_wakeup_thread(mddev
->thread
);
6475 if (mddev
->event_work
.func
)
6476 queue_work(md_misc_wq
, &mddev
->event_work
);
6477 md_new_event_inintr(mddev
);
6480 /* seq_file implementation /proc/mdstat */
6482 static void status_unused(struct seq_file
*seq
)
6485 struct md_rdev
*rdev
;
6487 seq_printf(seq
, "unused devices: ");
6489 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6490 char b
[BDEVNAME_SIZE
];
6492 seq_printf(seq
, "%s ",
6493 bdevname(rdev
->bdev
,b
));
6496 seq_printf(seq
, "<none>");
6498 seq_printf(seq
, "\n");
6502 static void status_resync(struct seq_file
*seq
, struct mddev
* mddev
)
6504 sector_t max_sectors
, resync
, res
;
6505 unsigned long dt
, db
;
6508 unsigned int per_milli
;
6510 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6512 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6513 max_sectors
= mddev
->resync_max_sectors
;
6515 max_sectors
= mddev
->dev_sectors
;
6518 * Should not happen.
6524 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6525 * in a sector_t, and (max_sectors>>scale) will fit in a
6526 * u32, as those are the requirements for sector_div.
6527 * Thus 'scale' must be at least 10
6530 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6531 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6534 res
= (resync
>>scale
)*1000;
6535 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6539 int i
, x
= per_milli
/50, y
= 20-x
;
6540 seq_printf(seq
, "[");
6541 for (i
= 0; i
< x
; i
++)
6542 seq_printf(seq
, "=");
6543 seq_printf(seq
, ">");
6544 for (i
= 0; i
< y
; i
++)
6545 seq_printf(seq
, ".");
6546 seq_printf(seq
, "] ");
6548 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6549 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6551 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6553 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6554 "resync" : "recovery"))),
6555 per_milli
/10, per_milli
% 10,
6556 (unsigned long long) resync
/2,
6557 (unsigned long long) max_sectors
/2);
6560 * dt: time from mark until now
6561 * db: blocks written from mark until now
6562 * rt: remaining time
6564 * rt is a sector_t, so could be 32bit or 64bit.
6565 * So we divide before multiply in case it is 32bit and close
6567 * We scale the divisor (db) by 32 to avoid losing precision
6568 * near the end of resync when the number of remaining sectors
6570 * We then divide rt by 32 after multiplying by db to compensate.
6571 * The '+1' avoids division by zero if db is very small.
6573 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6575 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6576 - mddev
->resync_mark_cnt
;
6578 rt
= max_sectors
- resync
; /* number of remaining sectors */
6579 sector_div(rt
, db
/32+1);
6583 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6584 ((unsigned long)rt
% 60)/6);
6586 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6589 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6591 struct list_head
*tmp
;
6593 struct mddev
*mddev
;
6601 spin_lock(&all_mddevs_lock
);
6602 list_for_each(tmp
,&all_mddevs
)
6604 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
6606 spin_unlock(&all_mddevs_lock
);
6609 spin_unlock(&all_mddevs_lock
);
6611 return (void*)2;/* tail */
6615 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6617 struct list_head
*tmp
;
6618 struct mddev
*next_mddev
, *mddev
= v
;
6624 spin_lock(&all_mddevs_lock
);
6626 tmp
= all_mddevs
.next
;
6628 tmp
= mddev
->all_mddevs
.next
;
6629 if (tmp
!= &all_mddevs
)
6630 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
6632 next_mddev
= (void*)2;
6635 spin_unlock(&all_mddevs_lock
);
6643 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6645 struct mddev
*mddev
= v
;
6647 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6651 static int md_seq_show(struct seq_file
*seq
, void *v
)
6653 struct mddev
*mddev
= v
;
6655 struct md_rdev
*rdev
;
6656 struct bitmap
*bitmap
;
6658 if (v
== (void*)1) {
6659 struct md_personality
*pers
;
6660 seq_printf(seq
, "Personalities : ");
6661 spin_lock(&pers_lock
);
6662 list_for_each_entry(pers
, &pers_list
, list
)
6663 seq_printf(seq
, "[%s] ", pers
->name
);
6665 spin_unlock(&pers_lock
);
6666 seq_printf(seq
, "\n");
6667 seq
->poll_event
= atomic_read(&md_event_count
);
6670 if (v
== (void*)2) {
6675 if (mddev_lock(mddev
) < 0)
6678 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6679 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6680 mddev
->pers
? "" : "in");
6683 seq_printf(seq
, " (read-only)");
6685 seq_printf(seq
, " (auto-read-only)");
6686 seq_printf(seq
, " %s", mddev
->pers
->name
);
6690 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6691 char b
[BDEVNAME_SIZE
];
6692 seq_printf(seq
, " %s[%d]",
6693 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6694 if (test_bit(WriteMostly
, &rdev
->flags
))
6695 seq_printf(seq
, "(W)");
6696 if (test_bit(Faulty
, &rdev
->flags
)) {
6697 seq_printf(seq
, "(F)");
6699 } else if (rdev
->raid_disk
< 0)
6700 seq_printf(seq
, "(S)"); /* spare */
6701 sectors
+= rdev
->sectors
;
6704 if (!list_empty(&mddev
->disks
)) {
6706 seq_printf(seq
, "\n %llu blocks",
6707 (unsigned long long)
6708 mddev
->array_sectors
/ 2);
6710 seq_printf(seq
, "\n %llu blocks",
6711 (unsigned long long)sectors
/ 2);
6713 if (mddev
->persistent
) {
6714 if (mddev
->major_version
!= 0 ||
6715 mddev
->minor_version
!= 90) {
6716 seq_printf(seq
," super %d.%d",
6717 mddev
->major_version
,
6718 mddev
->minor_version
);
6720 } else if (mddev
->external
)
6721 seq_printf(seq
, " super external:%s",
6722 mddev
->metadata_type
);
6724 seq_printf(seq
, " super non-persistent");
6727 mddev
->pers
->status(seq
, mddev
);
6728 seq_printf(seq
, "\n ");
6729 if (mddev
->pers
->sync_request
) {
6730 if (mddev
->curr_resync
> 2) {
6731 status_resync(seq
, mddev
);
6732 seq_printf(seq
, "\n ");
6733 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6734 seq_printf(seq
, "\tresync=DELAYED\n ");
6735 else if (mddev
->recovery_cp
< MaxSector
)
6736 seq_printf(seq
, "\tresync=PENDING\n ");
6739 seq_printf(seq
, "\n ");
6741 if ((bitmap
= mddev
->bitmap
)) {
6742 unsigned long chunk_kb
;
6743 unsigned long flags
;
6744 spin_lock_irqsave(&bitmap
->lock
, flags
);
6745 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6746 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6748 bitmap
->pages
- bitmap
->missing_pages
,
6750 (bitmap
->pages
- bitmap
->missing_pages
)
6751 << (PAGE_SHIFT
- 10),
6752 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6753 chunk_kb
? "KB" : "B");
6755 seq_printf(seq
, ", file: ");
6756 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6759 seq_printf(seq
, "\n");
6760 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6763 seq_printf(seq
, "\n");
6765 mddev_unlock(mddev
);
6770 static const struct seq_operations md_seq_ops
= {
6771 .start
= md_seq_start
,
6772 .next
= md_seq_next
,
6773 .stop
= md_seq_stop
,
6774 .show
= md_seq_show
,
6777 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6779 struct seq_file
*seq
;
6782 error
= seq_open(file
, &md_seq_ops
);
6786 seq
= file
->private_data
;
6787 seq
->poll_event
= atomic_read(&md_event_count
);
6791 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6793 struct seq_file
*seq
= filp
->private_data
;
6796 poll_wait(filp
, &md_event_waiters
, wait
);
6798 /* always allow read */
6799 mask
= POLLIN
| POLLRDNORM
;
6801 if (seq
->poll_event
!= atomic_read(&md_event_count
))
6802 mask
|= POLLERR
| POLLPRI
;
6806 static const struct file_operations md_seq_fops
= {
6807 .owner
= THIS_MODULE
,
6808 .open
= md_seq_open
,
6810 .llseek
= seq_lseek
,
6811 .release
= seq_release_private
,
6812 .poll
= mdstat_poll
,
6815 int register_md_personality(struct md_personality
*p
)
6817 spin_lock(&pers_lock
);
6818 list_add_tail(&p
->list
, &pers_list
);
6819 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6820 spin_unlock(&pers_lock
);
6824 int unregister_md_personality(struct md_personality
*p
)
6826 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6827 spin_lock(&pers_lock
);
6828 list_del_init(&p
->list
);
6829 spin_unlock(&pers_lock
);
6833 static int is_mddev_idle(struct mddev
*mddev
, int init
)
6835 struct md_rdev
* rdev
;
6841 rdev_for_each_rcu(rdev
, mddev
) {
6842 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6843 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6844 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6845 atomic_read(&disk
->sync_io
);
6846 /* sync IO will cause sync_io to increase before the disk_stats
6847 * as sync_io is counted when a request starts, and
6848 * disk_stats is counted when it completes.
6849 * So resync activity will cause curr_events to be smaller than
6850 * when there was no such activity.
6851 * non-sync IO will cause disk_stat to increase without
6852 * increasing sync_io so curr_events will (eventually)
6853 * be larger than it was before. Once it becomes
6854 * substantially larger, the test below will cause
6855 * the array to appear non-idle, and resync will slow
6857 * If there is a lot of outstanding resync activity when
6858 * we set last_event to curr_events, then all that activity
6859 * completing might cause the array to appear non-idle
6860 * and resync will be slowed down even though there might
6861 * not have been non-resync activity. This will only
6862 * happen once though. 'last_events' will soon reflect
6863 * the state where there is little or no outstanding
6864 * resync requests, and further resync activity will
6865 * always make curr_events less than last_events.
6868 if (init
|| curr_events
- rdev
->last_events
> 64) {
6869 rdev
->last_events
= curr_events
;
6877 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
6879 /* another "blocks" (512byte) blocks have been synced */
6880 atomic_sub(blocks
, &mddev
->recovery_active
);
6881 wake_up(&mddev
->recovery_wait
);
6883 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6884 md_wakeup_thread(mddev
->thread
);
6885 // stop recovery, signal do_sync ....
6890 /* md_write_start(mddev, bi)
6891 * If we need to update some array metadata (e.g. 'active' flag
6892 * in superblock) before writing, schedule a superblock update
6893 * and wait for it to complete.
6895 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
6898 if (bio_data_dir(bi
) != WRITE
)
6901 BUG_ON(mddev
->ro
== 1);
6902 if (mddev
->ro
== 2) {
6903 /* need to switch to read/write */
6905 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6906 md_wakeup_thread(mddev
->thread
);
6907 md_wakeup_thread(mddev
->sync_thread
);
6910 atomic_inc(&mddev
->writes_pending
);
6911 if (mddev
->safemode
== 1)
6912 mddev
->safemode
= 0;
6913 if (mddev
->in_sync
) {
6914 spin_lock_irq(&mddev
->write_lock
);
6915 if (mddev
->in_sync
) {
6917 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6918 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6919 md_wakeup_thread(mddev
->thread
);
6922 spin_unlock_irq(&mddev
->write_lock
);
6925 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6926 wait_event(mddev
->sb_wait
,
6927 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6930 void md_write_end(struct mddev
*mddev
)
6932 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6933 if (mddev
->safemode
== 2)
6934 md_wakeup_thread(mddev
->thread
);
6935 else if (mddev
->safemode_delay
)
6936 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6940 /* md_allow_write(mddev)
6941 * Calling this ensures that the array is marked 'active' so that writes
6942 * may proceed without blocking. It is important to call this before
6943 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6944 * Must be called with mddev_lock held.
6946 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6947 * is dropped, so return -EAGAIN after notifying userspace.
6949 int md_allow_write(struct mddev
*mddev
)
6955 if (!mddev
->pers
->sync_request
)
6958 spin_lock_irq(&mddev
->write_lock
);
6959 if (mddev
->in_sync
) {
6961 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6962 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6963 if (mddev
->safemode_delay
&&
6964 mddev
->safemode
== 0)
6965 mddev
->safemode
= 1;
6966 spin_unlock_irq(&mddev
->write_lock
);
6967 md_update_sb(mddev
, 0);
6968 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6970 spin_unlock_irq(&mddev
->write_lock
);
6972 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6977 EXPORT_SYMBOL_GPL(md_allow_write
);
6979 #define SYNC_MARKS 10
6980 #define SYNC_MARK_STEP (3*HZ)
6981 void md_do_sync(struct mddev
*mddev
)
6983 struct mddev
*mddev2
;
6984 unsigned int currspeed
= 0,
6986 sector_t max_sectors
,j
, io_sectors
;
6987 unsigned long mark
[SYNC_MARKS
];
6988 sector_t mark_cnt
[SYNC_MARKS
];
6990 struct list_head
*tmp
;
6991 sector_t last_check
;
6993 struct md_rdev
*rdev
;
6996 /* just incase thread restarts... */
6997 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6999 if (mddev
->ro
) /* never try to sync a read-only array */
7002 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7003 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
7004 desc
= "data-check";
7005 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7006 desc
= "requested-resync";
7009 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7014 /* we overload curr_resync somewhat here.
7015 * 0 == not engaged in resync at all
7016 * 2 == checking that there is no conflict with another sync
7017 * 1 == like 2, but have yielded to allow conflicting resync to
7019 * other == active in resync - this many blocks
7021 * Before starting a resync we must have set curr_resync to
7022 * 2, and then checked that every "conflicting" array has curr_resync
7023 * less than ours. When we find one that is the same or higher
7024 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7025 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7026 * This will mean we have to start checking from the beginning again.
7031 mddev
->curr_resync
= 2;
7034 if (kthread_should_stop())
7035 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7037 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7039 for_each_mddev(mddev2
, tmp
) {
7040 if (mddev2
== mddev
)
7042 if (!mddev
->parallel_resync
7043 && mddev2
->curr_resync
7044 && match_mddev_units(mddev
, mddev2
)) {
7046 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7047 /* arbitrarily yield */
7048 mddev
->curr_resync
= 1;
7049 wake_up(&resync_wait
);
7051 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7052 /* no need to wait here, we can wait the next
7053 * time 'round when curr_resync == 2
7056 /* We need to wait 'interruptible' so as not to
7057 * contribute to the load average, and not to
7058 * be caught by 'softlockup'
7060 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7061 if (!kthread_should_stop() &&
7062 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7063 printk(KERN_INFO
"md: delaying %s of %s"
7064 " until %s has finished (they"
7065 " share one or more physical units)\n",
7066 desc
, mdname(mddev
), mdname(mddev2
));
7068 if (signal_pending(current
))
7069 flush_signals(current
);
7071 finish_wait(&resync_wait
, &wq
);
7074 finish_wait(&resync_wait
, &wq
);
7077 } while (mddev
->curr_resync
< 2);
7080 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7081 /* resync follows the size requested by the personality,
7082 * which defaults to physical size, but can be virtual size
7084 max_sectors
= mddev
->resync_max_sectors
;
7085 mddev
->resync_mismatches
= 0;
7086 /* we don't use the checkpoint if there's a bitmap */
7087 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7088 j
= mddev
->resync_min
;
7089 else if (!mddev
->bitmap
)
7090 j
= mddev
->recovery_cp
;
7092 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7093 max_sectors
= mddev
->dev_sectors
;
7095 /* recovery follows the physical size of devices */
7096 max_sectors
= mddev
->dev_sectors
;
7099 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7100 if (rdev
->raid_disk
>= 0 &&
7101 !test_bit(Faulty
, &rdev
->flags
) &&
7102 !test_bit(In_sync
, &rdev
->flags
) &&
7103 rdev
->recovery_offset
< j
)
7104 j
= rdev
->recovery_offset
;
7108 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7109 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7110 " %d KB/sec/disk.\n", speed_min(mddev
));
7111 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7112 "(but not more than %d KB/sec) for %s.\n",
7113 speed_max(mddev
), desc
);
7115 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7118 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7120 mark_cnt
[m
] = io_sectors
;
7123 mddev
->resync_mark
= mark
[last_mark
];
7124 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7127 * Tune reconstruction:
7129 window
= 32*(PAGE_SIZE
/512);
7130 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7131 window
/2, (unsigned long long)max_sectors
/2);
7133 atomic_set(&mddev
->recovery_active
, 0);
7138 "md: resuming %s of %s from checkpoint.\n",
7139 desc
, mdname(mddev
));
7140 mddev
->curr_resync
= j
;
7142 mddev
->curr_resync_completed
= j
;
7144 while (j
< max_sectors
) {
7149 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7150 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7151 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7152 > (max_sectors
>> 4)) ||
7153 (j
- mddev
->curr_resync_completed
)*2
7154 >= mddev
->resync_max
- mddev
->curr_resync_completed
7156 /* time to update curr_resync_completed */
7157 wait_event(mddev
->recovery_wait
,
7158 atomic_read(&mddev
->recovery_active
) == 0);
7159 mddev
->curr_resync_completed
= j
;
7160 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7161 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7164 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7165 /* As this condition is controlled by user-space,
7166 * we can block indefinitely, so use '_interruptible'
7167 * to avoid triggering warnings.
7169 flush_signals(current
); /* just in case */
7170 wait_event_interruptible(mddev
->recovery_wait
,
7171 mddev
->resync_max
> j
7172 || kthread_should_stop());
7175 if (kthread_should_stop())
7178 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7179 currspeed
< speed_min(mddev
));
7181 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7185 if (!skipped
) { /* actual IO requested */
7186 io_sectors
+= sectors
;
7187 atomic_add(sectors
, &mddev
->recovery_active
);
7190 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7194 if (j
>1) mddev
->curr_resync
= j
;
7195 mddev
->curr_mark_cnt
= io_sectors
;
7196 if (last_check
== 0)
7197 /* this is the earliest that rebuild will be
7198 * visible in /proc/mdstat
7200 md_new_event(mddev
);
7202 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7205 last_check
= io_sectors
;
7207 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7209 int next
= (last_mark
+1) % SYNC_MARKS
;
7211 mddev
->resync_mark
= mark
[next
];
7212 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7213 mark
[next
] = jiffies
;
7214 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7219 if (kthread_should_stop())
7224 * this loop exits only if either when we are slower than
7225 * the 'hard' speed limit, or the system was IO-idle for
7227 * the system might be non-idle CPU-wise, but we only care
7228 * about not overloading the IO subsystem. (things like an
7229 * e2fsck being done on the RAID array should execute fast)
7233 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7234 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7236 if (currspeed
> speed_min(mddev
)) {
7237 if ((currspeed
> speed_max(mddev
)) ||
7238 !is_mddev_idle(mddev
, 0)) {
7244 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7246 * this also signals 'finished resyncing' to md_stop
7249 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7251 /* tell personality that we are finished */
7252 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7254 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7255 mddev
->curr_resync
> 2) {
7256 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7257 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7258 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7260 "md: checkpointing %s of %s.\n",
7261 desc
, mdname(mddev
));
7262 mddev
->recovery_cp
= mddev
->curr_resync
;
7265 mddev
->recovery_cp
= MaxSector
;
7267 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7268 mddev
->curr_resync
= MaxSector
;
7270 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7271 if (rdev
->raid_disk
>= 0 &&
7272 mddev
->delta_disks
>= 0 &&
7273 !test_bit(Faulty
, &rdev
->flags
) &&
7274 !test_bit(In_sync
, &rdev
->flags
) &&
7275 rdev
->recovery_offset
< mddev
->curr_resync
)
7276 rdev
->recovery_offset
= mddev
->curr_resync
;
7280 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7283 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7284 /* We completed so min/max setting can be forgotten if used. */
7285 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7286 mddev
->resync_min
= 0;
7287 mddev
->resync_max
= MaxSector
;
7288 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7289 mddev
->resync_min
= mddev
->curr_resync_completed
;
7290 mddev
->curr_resync
= 0;
7291 wake_up(&resync_wait
);
7292 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7293 md_wakeup_thread(mddev
->thread
);
7298 * got a signal, exit.
7301 "md: md_do_sync() got signal ... exiting\n");
7302 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7306 EXPORT_SYMBOL_GPL(md_do_sync
);
7308 static int remove_and_add_spares(struct mddev
*mddev
)
7310 struct md_rdev
*rdev
;
7313 mddev
->curr_resync_completed
= 0;
7315 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7316 if (rdev
->raid_disk
>= 0 &&
7317 !test_bit(Blocked
, &rdev
->flags
) &&
7318 (test_bit(Faulty
, &rdev
->flags
) ||
7319 ! test_bit(In_sync
, &rdev
->flags
)) &&
7320 atomic_read(&rdev
->nr_pending
)==0) {
7321 if (mddev
->pers
->hot_remove_disk(
7322 mddev
, rdev
->raid_disk
)==0) {
7323 sysfs_unlink_rdev(mddev
, rdev
);
7324 rdev
->raid_disk
= -1;
7328 if (mddev
->degraded
) {
7329 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7330 if (rdev
->raid_disk
>= 0 &&
7331 !test_bit(In_sync
, &rdev
->flags
) &&
7332 !test_bit(Faulty
, &rdev
->flags
))
7334 if (rdev
->raid_disk
< 0
7335 && !test_bit(Faulty
, &rdev
->flags
)) {
7336 rdev
->recovery_offset
= 0;
7338 hot_add_disk(mddev
, rdev
) == 0) {
7339 if (sysfs_link_rdev(mddev
, rdev
))
7340 /* failure here is OK */;
7342 md_new_event(mddev
);
7343 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7352 static void reap_sync_thread(struct mddev
*mddev
)
7354 struct md_rdev
*rdev
;
7356 /* resync has finished, collect result */
7357 md_unregister_thread(&mddev
->sync_thread
);
7358 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7359 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7361 /* activate any spares */
7362 if (mddev
->pers
->spare_active(mddev
))
7363 sysfs_notify(&mddev
->kobj
, NULL
,
7366 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7367 mddev
->pers
->finish_reshape
)
7368 mddev
->pers
->finish_reshape(mddev
);
7370 /* If array is no-longer degraded, then any saved_raid_disk
7371 * information must be scrapped. Also if any device is now
7372 * In_sync we must scrape the saved_raid_disk for that device
7373 * do the superblock for an incrementally recovered device
7376 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7377 if (!mddev
->degraded
||
7378 test_bit(In_sync
, &rdev
->flags
))
7379 rdev
->saved_raid_disk
= -1;
7381 md_update_sb(mddev
, 1);
7382 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7383 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7384 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7385 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7386 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7387 /* flag recovery needed just to double check */
7388 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7389 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7390 md_new_event(mddev
);
7391 if (mddev
->event_work
.func
)
7392 queue_work(md_misc_wq
, &mddev
->event_work
);
7396 * This routine is regularly called by all per-raid-array threads to
7397 * deal with generic issues like resync and super-block update.
7398 * Raid personalities that don't have a thread (linear/raid0) do not
7399 * need this as they never do any recovery or update the superblock.
7401 * It does not do any resync itself, but rather "forks" off other threads
7402 * to do that as needed.
7403 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7404 * "->recovery" and create a thread at ->sync_thread.
7405 * When the thread finishes it sets MD_RECOVERY_DONE
7406 * and wakeups up this thread which will reap the thread and finish up.
7407 * This thread also removes any faulty devices (with nr_pending == 0).
7409 * The overall approach is:
7410 * 1/ if the superblock needs updating, update it.
7411 * 2/ If a recovery thread is running, don't do anything else.
7412 * 3/ If recovery has finished, clean up, possibly marking spares active.
7413 * 4/ If there are any faulty devices, remove them.
7414 * 5/ If array is degraded, try to add spares devices
7415 * 6/ If array has spares or is not in-sync, start a resync thread.
7417 void md_check_recovery(struct mddev
*mddev
)
7419 if (mddev
->suspended
)
7423 bitmap_daemon_work(mddev
);
7425 if (signal_pending(current
)) {
7426 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7427 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7429 mddev
->safemode
= 2;
7431 flush_signals(current
);
7434 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7437 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7438 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7439 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7440 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7441 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7442 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7446 if (mddev_trylock(mddev
)) {
7450 /* Only thing we do on a ro array is remove
7453 struct md_rdev
*rdev
;
7454 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7455 if (rdev
->raid_disk
>= 0 &&
7456 !test_bit(Blocked
, &rdev
->flags
) &&
7457 test_bit(Faulty
, &rdev
->flags
) &&
7458 atomic_read(&rdev
->nr_pending
)==0) {
7459 if (mddev
->pers
->hot_remove_disk(
7460 mddev
, rdev
->raid_disk
)==0) {
7461 sysfs_unlink_rdev(mddev
, rdev
);
7462 rdev
->raid_disk
= -1;
7465 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7469 if (!mddev
->external
) {
7471 spin_lock_irq(&mddev
->write_lock
);
7472 if (mddev
->safemode
&&
7473 !atomic_read(&mddev
->writes_pending
) &&
7475 mddev
->recovery_cp
== MaxSector
) {
7478 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7480 if (mddev
->safemode
== 1)
7481 mddev
->safemode
= 0;
7482 spin_unlock_irq(&mddev
->write_lock
);
7484 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7488 md_update_sb(mddev
, 0);
7490 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7491 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7492 /* resync/recovery still happening */
7493 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7496 if (mddev
->sync_thread
) {
7497 reap_sync_thread(mddev
);
7500 /* Set RUNNING before clearing NEEDED to avoid
7501 * any transients in the value of "sync_action".
7503 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7504 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7505 /* Clear some bits that don't mean anything, but
7508 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7509 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7511 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7513 /* no recovery is running.
7514 * remove any failed drives, then
7515 * add spares if possible.
7516 * Spare are also removed and re-added, to allow
7517 * the personality to fail the re-add.
7520 if (mddev
->reshape_position
!= MaxSector
) {
7521 if (mddev
->pers
->check_reshape
== NULL
||
7522 mddev
->pers
->check_reshape(mddev
) != 0)
7523 /* Cannot proceed */
7525 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7526 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7527 } else if ((spares
= remove_and_add_spares(mddev
))) {
7528 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7529 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7530 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7531 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7532 } else if (mddev
->recovery_cp
< MaxSector
) {
7533 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7534 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7535 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7536 /* nothing to be done ... */
7539 if (mddev
->pers
->sync_request
) {
7540 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7541 /* We are adding a device or devices to an array
7542 * which has the bitmap stored on all devices.
7543 * So make sure all bitmap pages get written
7545 bitmap_write_all(mddev
->bitmap
);
7547 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7550 if (!mddev
->sync_thread
) {
7551 printk(KERN_ERR
"%s: could not start resync"
7554 /* leave the spares where they are, it shouldn't hurt */
7555 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7556 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7557 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7558 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7559 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7561 md_wakeup_thread(mddev
->sync_thread
);
7562 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7563 md_new_event(mddev
);
7566 if (!mddev
->sync_thread
) {
7567 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7568 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7570 if (mddev
->sysfs_action
)
7571 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7573 mddev_unlock(mddev
);
7577 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
7579 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7580 wait_event_timeout(rdev
->blocked_wait
,
7581 !test_bit(Blocked
, &rdev
->flags
) &&
7582 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7583 msecs_to_jiffies(5000));
7584 rdev_dec_pending(rdev
, mddev
);
7586 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7589 /* Bad block management.
7590 * We can record which blocks on each device are 'bad' and so just
7591 * fail those blocks, or that stripe, rather than the whole device.
7592 * Entries in the bad-block table are 64bits wide. This comprises:
7593 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7594 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7595 * A 'shift' can be set so that larger blocks are tracked and
7596 * consequently larger devices can be covered.
7597 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7599 * Locking of the bad-block table uses a seqlock so md_is_badblock
7600 * might need to retry if it is very unlucky.
7601 * We will sometimes want to check for bad blocks in a bi_end_io function,
7602 * so we use the write_seqlock_irq variant.
7604 * When looking for a bad block we specify a range and want to
7605 * know if any block in the range is bad. So we binary-search
7606 * to the last range that starts at-or-before the given endpoint,
7607 * (or "before the sector after the target range")
7608 * then see if it ends after the given start.
7610 * 0 if there are no known bad blocks in the range
7611 * 1 if there are known bad block which are all acknowledged
7612 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7613 * plus the start/length of the first bad section we overlap.
7615 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7616 sector_t
*first_bad
, int *bad_sectors
)
7622 sector_t target
= s
+ sectors
;
7625 if (bb
->shift
> 0) {
7626 /* round the start down, and the end up */
7628 target
+= (1<<bb
->shift
) - 1;
7629 target
>>= bb
->shift
;
7630 sectors
= target
- s
;
7632 /* 'target' is now the first block after the bad range */
7635 seq
= read_seqbegin(&bb
->lock
);
7639 /* Binary search between lo and hi for 'target'
7640 * i.e. for the last range that starts before 'target'
7642 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7643 * are known not to be the last range before target.
7644 * VARIANT: hi-lo is the number of possible
7645 * ranges, and decreases until it reaches 1
7647 while (hi
- lo
> 1) {
7648 int mid
= (lo
+ hi
) / 2;
7649 sector_t a
= BB_OFFSET(p
[mid
]);
7651 /* This could still be the one, earlier ranges
7655 /* This and later ranges are definitely out. */
7658 /* 'lo' might be the last that started before target, but 'hi' isn't */
7660 /* need to check all range that end after 's' to see if
7661 * any are unacknowledged.
7664 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7665 if (BB_OFFSET(p
[lo
]) < target
) {
7666 /* starts before the end, and finishes after
7667 * the start, so they must overlap
7669 if (rv
!= -1 && BB_ACK(p
[lo
]))
7673 *first_bad
= BB_OFFSET(p
[lo
]);
7674 *bad_sectors
= BB_LEN(p
[lo
]);
7680 if (read_seqretry(&bb
->lock
, seq
))
7685 EXPORT_SYMBOL_GPL(md_is_badblock
);
7688 * Add a range of bad blocks to the table.
7689 * This might extend the table, or might contract it
7690 * if two adjacent ranges can be merged.
7691 * We binary-search to find the 'insertion' point, then
7692 * decide how best to handle it.
7694 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7702 /* badblocks are disabled */
7706 /* round the start down, and the end up */
7707 sector_t next
= s
+ sectors
;
7709 next
+= (1<<bb
->shift
) - 1;
7714 write_seqlock_irq(&bb
->lock
);
7719 /* Find the last range that starts at-or-before 's' */
7720 while (hi
- lo
> 1) {
7721 int mid
= (lo
+ hi
) / 2;
7722 sector_t a
= BB_OFFSET(p
[mid
]);
7728 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7732 /* we found a range that might merge with the start
7735 sector_t a
= BB_OFFSET(p
[lo
]);
7736 sector_t e
= a
+ BB_LEN(p
[lo
]);
7737 int ack
= BB_ACK(p
[lo
]);
7739 /* Yes, we can merge with a previous range */
7740 if (s
== a
&& s
+ sectors
>= e
)
7741 /* new range covers old */
7744 ack
= ack
&& acknowledged
;
7746 if (e
< s
+ sectors
)
7748 if (e
- a
<= BB_MAX_LEN
) {
7749 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
7752 /* does not all fit in one range,
7753 * make p[lo] maximal
7755 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
7756 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7762 if (sectors
&& hi
< bb
->count
) {
7763 /* 'hi' points to the first range that starts after 's'.
7764 * Maybe we can merge with the start of that range */
7765 sector_t a
= BB_OFFSET(p
[hi
]);
7766 sector_t e
= a
+ BB_LEN(p
[hi
]);
7767 int ack
= BB_ACK(p
[hi
]);
7768 if (a
<= s
+ sectors
) {
7769 /* merging is possible */
7770 if (e
<= s
+ sectors
) {
7775 ack
= ack
&& acknowledged
;
7778 if (e
- a
<= BB_MAX_LEN
) {
7779 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
7782 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7790 if (sectors
== 0 && hi
< bb
->count
) {
7791 /* we might be able to combine lo and hi */
7792 /* Note: 's' is at the end of 'lo' */
7793 sector_t a
= BB_OFFSET(p
[hi
]);
7794 int lolen
= BB_LEN(p
[lo
]);
7795 int hilen
= BB_LEN(p
[hi
]);
7796 int newlen
= lolen
+ hilen
- (s
- a
);
7797 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
7798 /* yes, we can combine them */
7799 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
7800 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
7801 memmove(p
+ hi
, p
+ hi
+ 1,
7802 (bb
->count
- hi
- 1) * 8);
7807 /* didn't merge (it all).
7808 * Need to add a range just before 'hi' */
7809 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7810 /* No room for more */
7814 int this_sectors
= sectors
;
7815 memmove(p
+ hi
+ 1, p
+ hi
,
7816 (bb
->count
- hi
) * 8);
7819 if (this_sectors
> BB_MAX_LEN
)
7820 this_sectors
= BB_MAX_LEN
;
7821 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
7822 sectors
-= this_sectors
;
7829 bb
->unacked_exist
= 1;
7830 write_sequnlock_irq(&bb
->lock
);
7835 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
7838 int rv
= md_set_badblocks(&rdev
->badblocks
,
7839 s
+ rdev
->data_offset
, sectors
, acknowledged
);
7841 /* Make sure they get written out promptly */
7842 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
7843 md_wakeup_thread(rdev
->mddev
->thread
);
7847 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
7850 * Remove a range of bad blocks from the table.
7851 * This may involve extending the table if we spilt a region,
7852 * but it must not fail. So if the table becomes full, we just
7853 * drop the remove request.
7855 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
7859 sector_t target
= s
+ sectors
;
7862 if (bb
->shift
> 0) {
7863 /* When clearing we round the start up and the end down.
7864 * This should not matter as the shift should align with
7865 * the block size and no rounding should ever be needed.
7866 * However it is better the think a block is bad when it
7867 * isn't than to think a block is not bad when it is.
7869 s
+= (1<<bb
->shift
) - 1;
7871 target
>>= bb
->shift
;
7872 sectors
= target
- s
;
7875 write_seqlock_irq(&bb
->lock
);
7880 /* Find the last range that starts before 'target' */
7881 while (hi
- lo
> 1) {
7882 int mid
= (lo
+ hi
) / 2;
7883 sector_t a
= BB_OFFSET(p
[mid
]);
7890 /* p[lo] is the last range that could overlap the
7891 * current range. Earlier ranges could also overlap,
7892 * but only this one can overlap the end of the range.
7894 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
7895 /* Partial overlap, leave the tail of this range */
7896 int ack
= BB_ACK(p
[lo
]);
7897 sector_t a
= BB_OFFSET(p
[lo
]);
7898 sector_t end
= a
+ BB_LEN(p
[lo
]);
7901 /* we need to split this range */
7902 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7906 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
7908 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
7911 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
7912 /* there is no longer an overlap */
7917 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7918 /* This range does overlap */
7919 if (BB_OFFSET(p
[lo
]) < s
) {
7920 /* Keep the early parts of this range. */
7921 int ack
= BB_ACK(p
[lo
]);
7922 sector_t start
= BB_OFFSET(p
[lo
]);
7923 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
7924 /* now low doesn't overlap, so.. */
7929 /* 'lo' is strictly before, 'hi' is strictly after,
7930 * anything between needs to be discarded
7933 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
7934 bb
->count
-= (hi
- lo
- 1);
7940 write_sequnlock_irq(&bb
->lock
);
7944 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
)
7946 return md_clear_badblocks(&rdev
->badblocks
,
7947 s
+ rdev
->data_offset
,
7950 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
7953 * Acknowledge all bad blocks in a list.
7954 * This only succeeds if ->changed is clear. It is used by
7955 * in-kernel metadata updates
7957 void md_ack_all_badblocks(struct badblocks
*bb
)
7959 if (bb
->page
== NULL
|| bb
->changed
)
7960 /* no point even trying */
7962 write_seqlock_irq(&bb
->lock
);
7964 if (bb
->changed
== 0) {
7967 for (i
= 0; i
< bb
->count
; i
++) {
7968 if (!BB_ACK(p
[i
])) {
7969 sector_t start
= BB_OFFSET(p
[i
]);
7970 int len
= BB_LEN(p
[i
]);
7971 p
[i
] = BB_MAKE(start
, len
, 1);
7974 bb
->unacked_exist
= 0;
7976 write_sequnlock_irq(&bb
->lock
);
7978 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
7980 /* sysfs access to bad-blocks list.
7981 * We present two files.
7982 * 'bad-blocks' lists sector numbers and lengths of ranges that
7983 * are recorded as bad. The list is truncated to fit within
7984 * the one-page limit of sysfs.
7985 * Writing "sector length" to this file adds an acknowledged
7987 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7988 * been acknowledged. Writing to this file adds bad blocks
7989 * without acknowledging them. This is largely for testing.
7993 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
8004 seq
= read_seqbegin(&bb
->lock
);
8009 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
8010 sector_t s
= BB_OFFSET(p
[i
]);
8011 unsigned int length
= BB_LEN(p
[i
]);
8012 int ack
= BB_ACK(p
[i
]);
8018 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
8019 (unsigned long long)s
<< bb
->shift
,
8020 length
<< bb
->shift
);
8022 if (unack
&& len
== 0)
8023 bb
->unacked_exist
= 0;
8025 if (read_seqretry(&bb
->lock
, seq
))
8034 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8036 unsigned long long sector
;
8040 /* Allow clearing via sysfs *only* for testing/debugging.
8041 * Normally only a successful write may clear a badblock
8044 if (page
[0] == '-') {
8048 #endif /* DO_DEBUG */
8050 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8052 if (newline
!= '\n')
8064 md_clear_badblocks(bb
, sector
, length
);
8067 #endif /* DO_DEBUG */
8068 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8074 static int md_notify_reboot(struct notifier_block
*this,
8075 unsigned long code
, void *x
)
8077 struct list_head
*tmp
;
8078 struct mddev
*mddev
;
8081 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
8083 printk(KERN_INFO
"md: stopping all md devices.\n");
8085 for_each_mddev(mddev
, tmp
) {
8086 if (mddev_trylock(mddev
)) {
8087 /* Force a switch to readonly even array
8088 * appears to still be in use. Hence
8091 md_set_readonly(mddev
, 100);
8092 mddev_unlock(mddev
);
8097 * certain more exotic SCSI devices are known to be
8098 * volatile wrt too early system reboots. While the
8099 * right place to handle this issue is the given
8100 * driver, we do want to have a safe RAID driver ...
8108 static struct notifier_block md_notifier
= {
8109 .notifier_call
= md_notify_reboot
,
8111 .priority
= INT_MAX
, /* before any real devices */
8114 static void md_geninit(void)
8116 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8118 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8121 static int __init
md_init(void)
8125 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8129 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8133 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8136 if ((ret
= register_blkdev(0, "mdp")) < 0)
8140 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8141 md_probe
, NULL
, NULL
);
8142 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8143 md_probe
, NULL
, NULL
);
8145 register_reboot_notifier(&md_notifier
);
8146 raid_table_header
= register_sysctl_table(raid_root_table
);
8152 unregister_blkdev(MD_MAJOR
, "md");
8154 destroy_workqueue(md_misc_wq
);
8156 destroy_workqueue(md_wq
);
8164 * Searches all registered partitions for autorun RAID arrays
8168 static LIST_HEAD(all_detected_devices
);
8169 struct detected_devices_node
{
8170 struct list_head list
;
8174 void md_autodetect_dev(dev_t dev
)
8176 struct detected_devices_node
*node_detected_dev
;
8178 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8179 if (node_detected_dev
) {
8180 node_detected_dev
->dev
= dev
;
8181 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8183 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8184 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8189 static void autostart_arrays(int part
)
8191 struct md_rdev
*rdev
;
8192 struct detected_devices_node
*node_detected_dev
;
8194 int i_scanned
, i_passed
;
8199 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8201 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8203 node_detected_dev
= list_entry(all_detected_devices
.next
,
8204 struct detected_devices_node
, list
);
8205 list_del(&node_detected_dev
->list
);
8206 dev
= node_detected_dev
->dev
;
8207 kfree(node_detected_dev
);
8208 rdev
= md_import_device(dev
,0, 90);
8212 if (test_bit(Faulty
, &rdev
->flags
)) {
8216 set_bit(AutoDetected
, &rdev
->flags
);
8217 list_add(&rdev
->same_set
, &pending_raid_disks
);
8221 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8222 i_scanned
, i_passed
);
8224 autorun_devices(part
);
8227 #endif /* !MODULE */
8229 static __exit
void md_exit(void)
8231 struct mddev
*mddev
;
8232 struct list_head
*tmp
;
8234 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8235 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8237 unregister_blkdev(MD_MAJOR
,"md");
8238 unregister_blkdev(mdp_major
, "mdp");
8239 unregister_reboot_notifier(&md_notifier
);
8240 unregister_sysctl_table(raid_table_header
);
8241 remove_proc_entry("mdstat", NULL
);
8242 for_each_mddev(mddev
, tmp
) {
8243 export_array(mddev
);
8244 mddev
->hold_active
= 0;
8246 destroy_workqueue(md_misc_wq
);
8247 destroy_workqueue(md_wq
);
8250 subsys_initcall(md_init
);
8251 module_exit(md_exit
)
8253 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8255 return sprintf(buffer
, "%d", start_readonly
);
8257 static int set_ro(const char *val
, struct kernel_param
*kp
)
8260 int num
= simple_strtoul(val
, &e
, 10);
8261 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8262 start_readonly
= num
;
8268 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8269 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8271 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8273 EXPORT_SYMBOL(register_md_personality
);
8274 EXPORT_SYMBOL(unregister_md_personality
);
8275 EXPORT_SYMBOL(md_error
);
8276 EXPORT_SYMBOL(md_done_sync
);
8277 EXPORT_SYMBOL(md_write_start
);
8278 EXPORT_SYMBOL(md_write_end
);
8279 EXPORT_SYMBOL(md_register_thread
);
8280 EXPORT_SYMBOL(md_unregister_thread
);
8281 EXPORT_SYMBOL(md_wakeup_thread
);
8282 EXPORT_SYMBOL(md_check_recovery
);
8283 MODULE_LICENSE("GPL");
8284 MODULE_DESCRIPTION("MD RAID framework");
8286 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);