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/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
57 #include "md-cluster.h"
60 static void autostart_arrays(int part
);
63 /* pers_list is a list of registered personalities protected
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
68 static LIST_HEAD(pers_list
);
69 static DEFINE_SPINLOCK(pers_lock
);
71 struct md_cluster_operations
*md_cluster_ops
;
72 EXPORT_SYMBOL(md_cluster_ops
);
73 struct module
*md_cluster_mod
;
74 EXPORT_SYMBOL(md_cluster_mod
);
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
77 static struct workqueue_struct
*md_wq
;
78 static struct workqueue_struct
*md_misc_wq
;
80 static int remove_and_add_spares(struct mddev
*mddev
,
81 struct md_rdev
*this);
82 static void mddev_detach(struct mddev
*mddev
);
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
103 static int sysctl_speed_limit_min
= 1000;
104 static int sysctl_speed_limit_max
= 200000;
105 static inline int speed_min(struct mddev
*mddev
)
107 return mddev
->sync_speed_min
?
108 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
111 static inline int speed_max(struct mddev
*mddev
)
113 return mddev
->sync_speed_max
?
114 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
117 static struct ctl_table_header
*raid_table_header
;
119 static struct ctl_table raid_table
[] = {
121 .procname
= "speed_limit_min",
122 .data
= &sysctl_speed_limit_min
,
123 .maxlen
= sizeof(int),
124 .mode
= S_IRUGO
|S_IWUSR
,
125 .proc_handler
= proc_dointvec
,
128 .procname
= "speed_limit_max",
129 .data
= &sysctl_speed_limit_max
,
130 .maxlen
= sizeof(int),
131 .mode
= S_IRUGO
|S_IWUSR
,
132 .proc_handler
= proc_dointvec
,
137 static struct ctl_table raid_dir_table
[] = {
141 .mode
= S_IRUGO
|S_IXUGO
,
147 static struct ctl_table raid_root_table
[] = {
152 .child
= raid_dir_table
,
157 static const struct block_device_operations md_fops
;
159 static int start_readonly
;
162 * like bio_clone, but with a local bio set
165 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
170 if (!mddev
|| !mddev
->bio_set
)
171 return bio_alloc(gfp_mask
, nr_iovecs
);
173 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, mddev
->bio_set
);
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
180 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
183 if (!mddev
|| !mddev
->bio_set
)
184 return bio_clone(bio
, gfp_mask
);
186 return bio_clone_bioset(bio
, gfp_mask
, mddev
->bio_set
);
188 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
201 static atomic_t md_event_count
;
202 void md_new_event(struct mddev
*mddev
)
204 atomic_inc(&md_event_count
);
205 wake_up(&md_event_waiters
);
207 EXPORT_SYMBOL_GPL(md_new_event
);
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
213 static LIST_HEAD(all_mddevs
);
214 static DEFINE_SPINLOCK(all_mddevs_lock
);
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
223 #define for_each_mddev(_mddev,_tmp) \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
245 static blk_qc_t
md_make_request(struct request_queue
*q
, struct bio
*bio
)
247 const int rw
= bio_data_dir(bio
);
248 struct mddev
*mddev
= q
->queuedata
;
249 unsigned int sectors
;
252 blk_queue_split(q
, &bio
, q
->bio_split
);
254 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
256 return BLK_QC_T_NONE
;
258 if (mddev
->ro
== 1 && unlikely(rw
== WRITE
)) {
259 if (bio_sectors(bio
) != 0)
260 bio
->bi_error
= -EROFS
;
262 return BLK_QC_T_NONE
;
264 smp_rmb(); /* Ensure implications of 'active' are visible */
266 if (mddev
->suspended
) {
269 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
270 TASK_UNINTERRUPTIBLE
);
271 if (!mddev
->suspended
)
277 finish_wait(&mddev
->sb_wait
, &__wait
);
279 atomic_inc(&mddev
->active_io
);
283 * save the sectors now since our bio can
284 * go away inside make_request
286 sectors
= bio_sectors(bio
);
287 /* bio could be mergeable after passing to underlayer */
288 bio
->bi_rw
&= ~REQ_NOMERGE
;
289 mddev
->pers
->make_request(mddev
, bio
);
291 cpu
= part_stat_lock();
292 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
293 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
296 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
297 wake_up(&mddev
->sb_wait
);
299 return BLK_QC_T_NONE
;
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once mddev_detach() is called and completes, the module will be
308 void mddev_suspend(struct mddev
*mddev
)
310 WARN_ON_ONCE(mddev
->thread
&& current
== mddev
->thread
->tsk
);
311 if (mddev
->suspended
++)
314 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
315 mddev
->pers
->quiesce(mddev
, 1);
317 del_timer_sync(&mddev
->safemode_timer
);
319 EXPORT_SYMBOL_GPL(mddev_suspend
);
321 void mddev_resume(struct mddev
*mddev
)
323 if (--mddev
->suspended
)
325 wake_up(&mddev
->sb_wait
);
326 mddev
->pers
->quiesce(mddev
, 0);
328 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
329 md_wakeup_thread(mddev
->thread
);
330 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
332 EXPORT_SYMBOL_GPL(mddev_resume
);
334 int mddev_congested(struct mddev
*mddev
, int bits
)
336 struct md_personality
*pers
= mddev
->pers
;
340 if (mddev
->suspended
)
342 else if (pers
&& pers
->congested
)
343 ret
= pers
->congested(mddev
, bits
);
347 EXPORT_SYMBOL_GPL(mddev_congested
);
348 static int md_congested(void *data
, int bits
)
350 struct mddev
*mddev
= data
;
351 return mddev_congested(mddev
, bits
);
355 * Generic flush handling for md
358 static void md_end_flush(struct bio
*bio
)
360 struct md_rdev
*rdev
= bio
->bi_private
;
361 struct mddev
*mddev
= rdev
->mddev
;
363 rdev_dec_pending(rdev
, mddev
);
365 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
366 /* The pre-request flush has finished */
367 queue_work(md_wq
, &mddev
->flush_work
);
372 static void md_submit_flush_data(struct work_struct
*ws
);
374 static void submit_flushes(struct work_struct
*ws
)
376 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
377 struct md_rdev
*rdev
;
379 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
380 atomic_set(&mddev
->flush_pending
, 1);
382 rdev_for_each_rcu(rdev
, mddev
)
383 if (rdev
->raid_disk
>= 0 &&
384 !test_bit(Faulty
, &rdev
->flags
)) {
385 /* Take two references, one is dropped
386 * when request finishes, one after
387 * we reclaim rcu_read_lock
390 atomic_inc(&rdev
->nr_pending
);
391 atomic_inc(&rdev
->nr_pending
);
393 bi
= bio_alloc_mddev(GFP_NOIO
, 0, mddev
);
394 bi
->bi_end_io
= md_end_flush
;
395 bi
->bi_private
= rdev
;
396 bi
->bi_bdev
= rdev
->bdev
;
397 atomic_inc(&mddev
->flush_pending
);
398 submit_bio(WRITE_FLUSH
, bi
);
400 rdev_dec_pending(rdev
, mddev
);
403 if (atomic_dec_and_test(&mddev
->flush_pending
))
404 queue_work(md_wq
, &mddev
->flush_work
);
407 static void md_submit_flush_data(struct work_struct
*ws
)
409 struct mddev
*mddev
= container_of(ws
, struct mddev
, flush_work
);
410 struct bio
*bio
= mddev
->flush_bio
;
412 if (bio
->bi_iter
.bi_size
== 0)
413 /* an empty barrier - all done */
416 bio
->bi_rw
&= ~REQ_FLUSH
;
417 mddev
->pers
->make_request(mddev
, bio
);
420 mddev
->flush_bio
= NULL
;
421 wake_up(&mddev
->sb_wait
);
424 void md_flush_request(struct mddev
*mddev
, struct bio
*bio
)
426 spin_lock_irq(&mddev
->lock
);
427 wait_event_lock_irq(mddev
->sb_wait
,
430 mddev
->flush_bio
= bio
;
431 spin_unlock_irq(&mddev
->lock
);
433 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
434 queue_work(md_wq
, &mddev
->flush_work
);
436 EXPORT_SYMBOL(md_flush_request
);
438 void md_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
440 struct mddev
*mddev
= cb
->data
;
441 md_wakeup_thread(mddev
->thread
);
444 EXPORT_SYMBOL(md_unplug
);
446 static inline struct mddev
*mddev_get(struct mddev
*mddev
)
448 atomic_inc(&mddev
->active
);
452 static void mddev_delayed_delete(struct work_struct
*ws
);
454 static void mddev_put(struct mddev
*mddev
)
456 struct bio_set
*bs
= NULL
;
458 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
460 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
461 mddev
->ctime
== 0 && !mddev
->hold_active
) {
462 /* Array is not configured at all, and not held active,
464 list_del_init(&mddev
->all_mddevs
);
466 mddev
->bio_set
= NULL
;
467 if (mddev
->gendisk
) {
468 /* We did a probe so need to clean up. Call
469 * queue_work inside the spinlock so that
470 * flush_workqueue() after mddev_find will
471 * succeed in waiting for the work to be done.
473 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
474 queue_work(md_misc_wq
, &mddev
->del_work
);
478 spin_unlock(&all_mddevs_lock
);
483 static void md_safemode_timeout(unsigned long data
);
485 void mddev_init(struct mddev
*mddev
)
487 mutex_init(&mddev
->open_mutex
);
488 mutex_init(&mddev
->reconfig_mutex
);
489 mutex_init(&mddev
->bitmap_info
.mutex
);
490 INIT_LIST_HEAD(&mddev
->disks
);
491 INIT_LIST_HEAD(&mddev
->all_mddevs
);
492 setup_timer(&mddev
->safemode_timer
, md_safemode_timeout
,
493 (unsigned long) mddev
);
494 atomic_set(&mddev
->active
, 1);
495 atomic_set(&mddev
->openers
, 0);
496 atomic_set(&mddev
->active_io
, 0);
497 spin_lock_init(&mddev
->lock
);
498 atomic_set(&mddev
->flush_pending
, 0);
499 init_waitqueue_head(&mddev
->sb_wait
);
500 init_waitqueue_head(&mddev
->recovery_wait
);
501 mddev
->reshape_position
= MaxSector
;
502 mddev
->reshape_backwards
= 0;
503 mddev
->last_sync_action
= "none";
504 mddev
->resync_min
= 0;
505 mddev
->resync_max
= MaxSector
;
506 mddev
->level
= LEVEL_NONE
;
508 EXPORT_SYMBOL_GPL(mddev_init
);
510 static struct mddev
*mddev_find(dev_t unit
)
512 struct mddev
*mddev
, *new = NULL
;
514 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
515 unit
&= ~((1<<MdpMinorShift
)-1);
518 spin_lock(&all_mddevs_lock
);
521 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
522 if (mddev
->unit
== unit
) {
524 spin_unlock(&all_mddevs_lock
);
530 list_add(&new->all_mddevs
, &all_mddevs
);
531 spin_unlock(&all_mddevs_lock
);
532 new->hold_active
= UNTIL_IOCTL
;
536 /* find an unused unit number */
537 static int next_minor
= 512;
538 int start
= next_minor
;
542 dev
= MKDEV(MD_MAJOR
, next_minor
);
544 if (next_minor
> MINORMASK
)
546 if (next_minor
== start
) {
547 /* Oh dear, all in use. */
548 spin_unlock(&all_mddevs_lock
);
554 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
555 if (mddev
->unit
== dev
) {
561 new->md_minor
= MINOR(dev
);
562 new->hold_active
= UNTIL_STOP
;
563 list_add(&new->all_mddevs
, &all_mddevs
);
564 spin_unlock(&all_mddevs_lock
);
567 spin_unlock(&all_mddevs_lock
);
569 new = kzalloc(sizeof(*new), GFP_KERNEL
);
574 if (MAJOR(unit
) == MD_MAJOR
)
575 new->md_minor
= MINOR(unit
);
577 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
584 static struct attribute_group md_redundancy_group
;
586 void mddev_unlock(struct mddev
*mddev
)
588 if (mddev
->to_remove
) {
589 /* These cannot be removed under reconfig_mutex as
590 * an access to the files will try to take reconfig_mutex
591 * while holding the file unremovable, which leads to
593 * So hold set sysfs_active while the remove in happeing,
594 * and anything else which might set ->to_remove or my
595 * otherwise change the sysfs namespace will fail with
596 * -EBUSY if sysfs_active is still set.
597 * We set sysfs_active under reconfig_mutex and elsewhere
598 * test it under the same mutex to ensure its correct value
601 struct attribute_group
*to_remove
= mddev
->to_remove
;
602 mddev
->to_remove
= NULL
;
603 mddev
->sysfs_active
= 1;
604 mutex_unlock(&mddev
->reconfig_mutex
);
606 if (mddev
->kobj
.sd
) {
607 if (to_remove
!= &md_redundancy_group
)
608 sysfs_remove_group(&mddev
->kobj
, to_remove
);
609 if (mddev
->pers
== NULL
||
610 mddev
->pers
->sync_request
== NULL
) {
611 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
612 if (mddev
->sysfs_action
)
613 sysfs_put(mddev
->sysfs_action
);
614 mddev
->sysfs_action
= NULL
;
617 mddev
->sysfs_active
= 0;
619 mutex_unlock(&mddev
->reconfig_mutex
);
621 /* As we've dropped the mutex we need a spinlock to
622 * make sure the thread doesn't disappear
624 spin_lock(&pers_lock
);
625 md_wakeup_thread(mddev
->thread
);
626 spin_unlock(&pers_lock
);
628 EXPORT_SYMBOL_GPL(mddev_unlock
);
630 struct md_rdev
*md_find_rdev_nr_rcu(struct mddev
*mddev
, int nr
)
632 struct md_rdev
*rdev
;
634 rdev_for_each_rcu(rdev
, mddev
)
635 if (rdev
->desc_nr
== nr
)
640 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu
);
642 static struct md_rdev
*find_rdev(struct mddev
*mddev
, dev_t dev
)
644 struct md_rdev
*rdev
;
646 rdev_for_each(rdev
, mddev
)
647 if (rdev
->bdev
->bd_dev
== dev
)
653 static struct md_rdev
*find_rdev_rcu(struct mddev
*mddev
, dev_t dev
)
655 struct md_rdev
*rdev
;
657 rdev_for_each_rcu(rdev
, mddev
)
658 if (rdev
->bdev
->bd_dev
== dev
)
664 static struct md_personality
*find_pers(int level
, char *clevel
)
666 struct md_personality
*pers
;
667 list_for_each_entry(pers
, &pers_list
, list
) {
668 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
670 if (strcmp(pers
->name
, clevel
)==0)
676 /* return the offset of the super block in 512byte sectors */
677 static inline sector_t
calc_dev_sboffset(struct md_rdev
*rdev
)
679 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
680 return MD_NEW_SIZE_SECTORS(num_sectors
);
683 static int alloc_disk_sb(struct md_rdev
*rdev
)
685 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
686 if (!rdev
->sb_page
) {
687 printk(KERN_ALERT
"md: out of memory.\n");
694 void md_rdev_clear(struct md_rdev
*rdev
)
697 put_page(rdev
->sb_page
);
699 rdev
->sb_page
= NULL
;
704 put_page(rdev
->bb_page
);
705 rdev
->bb_page
= NULL
;
707 badblocks_exit(&rdev
->badblocks
);
709 EXPORT_SYMBOL_GPL(md_rdev_clear
);
711 static void super_written(struct bio
*bio
)
713 struct md_rdev
*rdev
= bio
->bi_private
;
714 struct mddev
*mddev
= rdev
->mddev
;
717 printk("md: super_written gets error=%d\n", bio
->bi_error
);
718 md_error(mddev
, rdev
);
721 if (atomic_dec_and_test(&mddev
->pending_writes
))
722 wake_up(&mddev
->sb_wait
);
723 rdev_dec_pending(rdev
, mddev
);
727 void md_super_write(struct mddev
*mddev
, struct md_rdev
*rdev
,
728 sector_t sector
, int size
, struct page
*page
)
730 /* write first size bytes of page to sector of rdev
731 * Increment mddev->pending_writes before returning
732 * and decrement it on completion, waking up sb_wait
733 * if zero is reached.
734 * If an error occurred, call md_error
736 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
738 atomic_inc(&rdev
->nr_pending
);
740 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
741 bio
->bi_iter
.bi_sector
= sector
;
742 bio_add_page(bio
, page
, size
, 0);
743 bio
->bi_private
= rdev
;
744 bio
->bi_end_io
= super_written
;
746 atomic_inc(&mddev
->pending_writes
);
747 submit_bio(WRITE_FLUSH_FUA
, bio
);
750 void md_super_wait(struct mddev
*mddev
)
752 /* wait for all superblock writes that were scheduled to complete */
753 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->pending_writes
)==0);
756 int sync_page_io(struct md_rdev
*rdev
, sector_t sector
, int size
,
757 struct page
*page
, int rw
, bool metadata_op
)
759 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
762 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
763 rdev
->meta_bdev
: rdev
->bdev
;
765 bio
->bi_iter
.bi_sector
= sector
+ rdev
->sb_start
;
766 else if (rdev
->mddev
->reshape_position
!= MaxSector
&&
767 (rdev
->mddev
->reshape_backwards
==
768 (sector
>= rdev
->mddev
->reshape_position
)))
769 bio
->bi_iter
.bi_sector
= sector
+ rdev
->new_data_offset
;
771 bio
->bi_iter
.bi_sector
= sector
+ rdev
->data_offset
;
772 bio_add_page(bio
, page
, size
, 0);
773 submit_bio_wait(rw
, bio
);
775 ret
= !bio
->bi_error
;
779 EXPORT_SYMBOL_GPL(sync_page_io
);
781 static int read_disk_sb(struct md_rdev
*rdev
, int size
)
783 char b
[BDEVNAME_SIZE
];
788 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
794 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
795 bdevname(rdev
->bdev
,b
));
799 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
801 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
802 sb1
->set_uuid1
== sb2
->set_uuid1
&&
803 sb1
->set_uuid2
== sb2
->set_uuid2
&&
804 sb1
->set_uuid3
== sb2
->set_uuid3
;
807 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
810 mdp_super_t
*tmp1
, *tmp2
;
812 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
813 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
815 if (!tmp1
|| !tmp2
) {
817 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
825 * nr_disks is not constant
830 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
837 static u32
md_csum_fold(u32 csum
)
839 csum
= (csum
& 0xffff) + (csum
>> 16);
840 return (csum
& 0xffff) + (csum
>> 16);
843 static unsigned int calc_sb_csum(mdp_super_t
*sb
)
846 u32
*sb32
= (u32
*)sb
;
848 unsigned int disk_csum
, csum
;
850 disk_csum
= sb
->sb_csum
;
853 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
855 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
858 /* This used to use csum_partial, which was wrong for several
859 * reasons including that different results are returned on
860 * different architectures. It isn't critical that we get exactly
861 * the same return value as before (we always csum_fold before
862 * testing, and that removes any differences). However as we
863 * know that csum_partial always returned a 16bit value on
864 * alphas, do a fold to maximise conformity to previous behaviour.
866 sb
->sb_csum
= md_csum_fold(disk_csum
);
868 sb
->sb_csum
= disk_csum
;
874 * Handle superblock details.
875 * We want to be able to handle multiple superblock formats
876 * so we have a common interface to them all, and an array of
877 * different handlers.
878 * We rely on user-space to write the initial superblock, and support
879 * reading and updating of superblocks.
880 * Interface methods are:
881 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
882 * loads and validates a superblock on dev.
883 * if refdev != NULL, compare superblocks on both devices
885 * 0 - dev has a superblock that is compatible with refdev
886 * 1 - dev has a superblock that is compatible and newer than refdev
887 * so dev should be used as the refdev in future
888 * -EINVAL superblock incompatible or invalid
889 * -othererror e.g. -EIO
891 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
892 * Verify that dev is acceptable into mddev.
893 * The first time, mddev->raid_disks will be 0, and data from
894 * dev should be merged in. Subsequent calls check that dev
895 * is new enough. Return 0 or -EINVAL
897 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
898 * Update the superblock for rdev with data in mddev
899 * This does not write to disc.
905 struct module
*owner
;
906 int (*load_super
)(struct md_rdev
*rdev
,
907 struct md_rdev
*refdev
,
909 int (*validate_super
)(struct mddev
*mddev
,
910 struct md_rdev
*rdev
);
911 void (*sync_super
)(struct mddev
*mddev
,
912 struct md_rdev
*rdev
);
913 unsigned long long (*rdev_size_change
)(struct md_rdev
*rdev
,
914 sector_t num_sectors
);
915 int (*allow_new_offset
)(struct md_rdev
*rdev
,
916 unsigned long long new_offset
);
920 * Check that the given mddev has no bitmap.
922 * This function is called from the run method of all personalities that do not
923 * support bitmaps. It prints an error message and returns non-zero if mddev
924 * has a bitmap. Otherwise, it returns 0.
927 int md_check_no_bitmap(struct mddev
*mddev
)
929 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
931 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
932 mdname(mddev
), mddev
->pers
->name
);
935 EXPORT_SYMBOL(md_check_no_bitmap
);
938 * load_super for 0.90.0
940 static int super_90_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
942 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
947 * Calculate the position of the superblock (512byte sectors),
948 * it's at the end of the disk.
950 * It also happens to be a multiple of 4Kb.
952 rdev
->sb_start
= calc_dev_sboffset(rdev
);
954 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
959 bdevname(rdev
->bdev
, b
);
960 sb
= page_address(rdev
->sb_page
);
962 if (sb
->md_magic
!= MD_SB_MAGIC
) {
963 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
968 if (sb
->major_version
!= 0 ||
969 sb
->minor_version
< 90 ||
970 sb
->minor_version
> 91) {
971 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
972 sb
->major_version
, sb
->minor_version
,
977 if (sb
->raid_disks
<= 0)
980 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
981 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
986 rdev
->preferred_minor
= sb
->md_minor
;
987 rdev
->data_offset
= 0;
988 rdev
->new_data_offset
= 0;
989 rdev
->sb_size
= MD_SB_BYTES
;
990 rdev
->badblocks
.shift
= -1;
992 if (sb
->level
== LEVEL_MULTIPATH
)
995 rdev
->desc_nr
= sb
->this_disk
.number
;
1001 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1002 if (!uuid_equal(refsb
, sb
)) {
1003 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1004 b
, bdevname(refdev
->bdev
,b2
));
1007 if (!sb_equal(refsb
, sb
)) {
1008 printk(KERN_WARNING
"md: %s has same UUID"
1009 " but different superblock to %s\n",
1010 b
, bdevname(refdev
->bdev
, b2
));
1014 ev2
= md_event(refsb
);
1020 rdev
->sectors
= rdev
->sb_start
;
1021 /* Limit to 4TB as metadata cannot record more than that.
1022 * (not needed for Linear and RAID0 as metadata doesn't
1025 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)rdev
->sectors
>= (2ULL << 32) &&
1027 rdev
->sectors
= (sector_t
)(2ULL << 32) - 2;
1029 if (rdev
->sectors
< ((sector_t
)sb
->size
) * 2 && sb
->level
>= 1)
1030 /* "this cannot possibly happen" ... */
1038 * validate_super for 0.90.0
1040 static int super_90_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1043 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1044 __u64 ev1
= md_event(sb
);
1046 rdev
->raid_disk
= -1;
1047 clear_bit(Faulty
, &rdev
->flags
);
1048 clear_bit(In_sync
, &rdev
->flags
);
1049 clear_bit(Bitmap_sync
, &rdev
->flags
);
1050 clear_bit(WriteMostly
, &rdev
->flags
);
1052 if (mddev
->raid_disks
== 0) {
1053 mddev
->major_version
= 0;
1054 mddev
->minor_version
= sb
->minor_version
;
1055 mddev
->patch_version
= sb
->patch_version
;
1056 mddev
->external
= 0;
1057 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1058 mddev
->ctime
= sb
->ctime
;
1059 mddev
->utime
= sb
->utime
;
1060 mddev
->level
= sb
->level
;
1061 mddev
->clevel
[0] = 0;
1062 mddev
->layout
= sb
->layout
;
1063 mddev
->raid_disks
= sb
->raid_disks
;
1064 mddev
->dev_sectors
= ((sector_t
)sb
->size
) * 2;
1065 mddev
->events
= ev1
;
1066 mddev
->bitmap_info
.offset
= 0;
1067 mddev
->bitmap_info
.space
= 0;
1068 /* bitmap can use 60 K after the 4K superblocks */
1069 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1070 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
1071 mddev
->reshape_backwards
= 0;
1073 if (mddev
->minor_version
>= 91) {
1074 mddev
->reshape_position
= sb
->reshape_position
;
1075 mddev
->delta_disks
= sb
->delta_disks
;
1076 mddev
->new_level
= sb
->new_level
;
1077 mddev
->new_layout
= sb
->new_layout
;
1078 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1079 if (mddev
->delta_disks
< 0)
1080 mddev
->reshape_backwards
= 1;
1082 mddev
->reshape_position
= MaxSector
;
1083 mddev
->delta_disks
= 0;
1084 mddev
->new_level
= mddev
->level
;
1085 mddev
->new_layout
= mddev
->layout
;
1086 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1089 if (sb
->state
& (1<<MD_SB_CLEAN
))
1090 mddev
->recovery_cp
= MaxSector
;
1092 if (sb
->events_hi
== sb
->cp_events_hi
&&
1093 sb
->events_lo
== sb
->cp_events_lo
) {
1094 mddev
->recovery_cp
= sb
->recovery_cp
;
1096 mddev
->recovery_cp
= 0;
1099 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1100 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1101 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1102 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1104 mddev
->max_disks
= MD_SB_DISKS
;
1106 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1107 mddev
->bitmap_info
.file
== NULL
) {
1108 mddev
->bitmap_info
.offset
=
1109 mddev
->bitmap_info
.default_offset
;
1110 mddev
->bitmap_info
.space
=
1111 mddev
->bitmap_info
.default_space
;
1114 } else if (mddev
->pers
== NULL
) {
1115 /* Insist on good event counter while assembling, except
1116 * for spares (which don't need an event count) */
1118 if (sb
->disks
[rdev
->desc_nr
].state
& (
1119 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1120 if (ev1
< mddev
->events
)
1122 } else if (mddev
->bitmap
) {
1123 /* if adding to array with a bitmap, then we can accept an
1124 * older device ... but not too old.
1126 if (ev1
< mddev
->bitmap
->events_cleared
)
1128 if (ev1
< mddev
->events
)
1129 set_bit(Bitmap_sync
, &rdev
->flags
);
1131 if (ev1
< mddev
->events
)
1132 /* just a hot-add of a new device, leave raid_disk at -1 */
1136 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1137 desc
= sb
->disks
+ rdev
->desc_nr
;
1139 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1140 set_bit(Faulty
, &rdev
->flags
);
1141 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1142 desc->raid_disk < mddev->raid_disks */) {
1143 set_bit(In_sync
, &rdev
->flags
);
1144 rdev
->raid_disk
= desc
->raid_disk
;
1145 rdev
->saved_raid_disk
= desc
->raid_disk
;
1146 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1147 /* active but not in sync implies recovery up to
1148 * reshape position. We don't know exactly where
1149 * that is, so set to zero for now */
1150 if (mddev
->minor_version
>= 91) {
1151 rdev
->recovery_offset
= 0;
1152 rdev
->raid_disk
= desc
->raid_disk
;
1155 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1156 set_bit(WriteMostly
, &rdev
->flags
);
1157 } else /* MULTIPATH are always insync */
1158 set_bit(In_sync
, &rdev
->flags
);
1163 * sync_super for 0.90.0
1165 static void super_90_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1168 struct md_rdev
*rdev2
;
1169 int next_spare
= mddev
->raid_disks
;
1171 /* make rdev->sb match mddev data..
1174 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1175 * 3/ any empty disks < next_spare become removed
1177 * disks[0] gets initialised to REMOVED because
1178 * we cannot be sure from other fields if it has
1179 * been initialised or not.
1182 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1184 rdev
->sb_size
= MD_SB_BYTES
;
1186 sb
= page_address(rdev
->sb_page
);
1188 memset(sb
, 0, sizeof(*sb
));
1190 sb
->md_magic
= MD_SB_MAGIC
;
1191 sb
->major_version
= mddev
->major_version
;
1192 sb
->patch_version
= mddev
->patch_version
;
1193 sb
->gvalid_words
= 0; /* ignored */
1194 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1195 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1196 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1197 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1199 sb
->ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
1200 sb
->level
= mddev
->level
;
1201 sb
->size
= mddev
->dev_sectors
/ 2;
1202 sb
->raid_disks
= mddev
->raid_disks
;
1203 sb
->md_minor
= mddev
->md_minor
;
1204 sb
->not_persistent
= 0;
1205 sb
->utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
1207 sb
->events_hi
= (mddev
->events
>>32);
1208 sb
->events_lo
= (u32
)mddev
->events
;
1210 if (mddev
->reshape_position
== MaxSector
)
1211 sb
->minor_version
= 90;
1213 sb
->minor_version
= 91;
1214 sb
->reshape_position
= mddev
->reshape_position
;
1215 sb
->new_level
= mddev
->new_level
;
1216 sb
->delta_disks
= mddev
->delta_disks
;
1217 sb
->new_layout
= mddev
->new_layout
;
1218 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1220 mddev
->minor_version
= sb
->minor_version
;
1223 sb
->recovery_cp
= mddev
->recovery_cp
;
1224 sb
->cp_events_hi
= (mddev
->events
>>32);
1225 sb
->cp_events_lo
= (u32
)mddev
->events
;
1226 if (mddev
->recovery_cp
== MaxSector
)
1227 sb
->state
= (1<< MD_SB_CLEAN
);
1229 sb
->recovery_cp
= 0;
1231 sb
->layout
= mddev
->layout
;
1232 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1234 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1235 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1237 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1238 rdev_for_each(rdev2
, mddev
) {
1241 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1243 if (rdev2
->raid_disk
>= 0 &&
1244 sb
->minor_version
>= 91)
1245 /* we have nowhere to store the recovery_offset,
1246 * but if it is not below the reshape_position,
1247 * we can piggy-back on that.
1250 if (rdev2
->raid_disk
< 0 ||
1251 test_bit(Faulty
, &rdev2
->flags
))
1254 desc_nr
= rdev2
->raid_disk
;
1256 desc_nr
= next_spare
++;
1257 rdev2
->desc_nr
= desc_nr
;
1258 d
= &sb
->disks
[rdev2
->desc_nr
];
1260 d
->number
= rdev2
->desc_nr
;
1261 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1262 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1264 d
->raid_disk
= rdev2
->raid_disk
;
1266 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1267 if (test_bit(Faulty
, &rdev2
->flags
))
1268 d
->state
= (1<<MD_DISK_FAULTY
);
1269 else if (is_active
) {
1270 d
->state
= (1<<MD_DISK_ACTIVE
);
1271 if (test_bit(In_sync
, &rdev2
->flags
))
1272 d
->state
|= (1<<MD_DISK_SYNC
);
1280 if (test_bit(WriteMostly
, &rdev2
->flags
))
1281 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1283 /* now set the "removed" and "faulty" bits on any missing devices */
1284 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1285 mdp_disk_t
*d
= &sb
->disks
[i
];
1286 if (d
->state
== 0 && d
->number
== 0) {
1289 d
->state
= (1<<MD_DISK_REMOVED
);
1290 d
->state
|= (1<<MD_DISK_FAULTY
);
1294 sb
->nr_disks
= nr_disks
;
1295 sb
->active_disks
= active
;
1296 sb
->working_disks
= working
;
1297 sb
->failed_disks
= failed
;
1298 sb
->spare_disks
= spare
;
1300 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1301 sb
->sb_csum
= calc_sb_csum(sb
);
1305 * rdev_size_change for 0.90.0
1307 static unsigned long long
1308 super_90_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1310 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1311 return 0; /* component must fit device */
1312 if (rdev
->mddev
->bitmap_info
.offset
)
1313 return 0; /* can't move bitmap */
1314 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1315 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1316 num_sectors
= rdev
->sb_start
;
1317 /* Limit to 4TB as metadata cannot record more than that.
1318 * 4TB == 2^32 KB, or 2*2^32 sectors.
1320 if (IS_ENABLED(CONFIG_LBDAF
) && (u64
)num_sectors
>= (2ULL << 32) &&
1321 rdev
->mddev
->level
>= 1)
1322 num_sectors
= (sector_t
)(2ULL << 32) - 2;
1323 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1325 md_super_wait(rdev
->mddev
);
1330 super_90_allow_new_offset(struct md_rdev
*rdev
, unsigned long long new_offset
)
1332 /* non-zero offset changes not possible with v0.90 */
1333 return new_offset
== 0;
1337 * version 1 superblock
1340 static __le32
calc_sb_1_csum(struct mdp_superblock_1
*sb
)
1344 unsigned long long newcsum
;
1345 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1346 __le32
*isuper
= (__le32
*)sb
;
1348 disk_csum
= sb
->sb_csum
;
1351 for (; size
>= 4; size
-= 4)
1352 newcsum
+= le32_to_cpu(*isuper
++);
1355 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1357 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1358 sb
->sb_csum
= disk_csum
;
1359 return cpu_to_le32(csum
);
1362 static int super_1_load(struct md_rdev
*rdev
, struct md_rdev
*refdev
, int minor_version
)
1364 struct mdp_superblock_1
*sb
;
1368 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1372 * Calculate the position of the superblock in 512byte sectors.
1373 * It is always aligned to a 4K boundary and
1374 * depeding on minor_version, it can be:
1375 * 0: At least 8K, but less than 12K, from end of device
1376 * 1: At start of device
1377 * 2: 4K from start of device.
1379 switch(minor_version
) {
1381 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1383 sb_start
&= ~(sector_t
)(4*2-1);
1394 rdev
->sb_start
= sb_start
;
1396 /* superblock is rarely larger than 1K, but it can be larger,
1397 * and it is safe to read 4k, so we do that
1399 ret
= read_disk_sb(rdev
, 4096);
1400 if (ret
) return ret
;
1402 sb
= page_address(rdev
->sb_page
);
1404 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1405 sb
->major_version
!= cpu_to_le32(1) ||
1406 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1407 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1408 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1411 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1412 printk("md: invalid superblock checksum on %s\n",
1413 bdevname(rdev
->bdev
,b
));
1416 if (le64_to_cpu(sb
->data_size
) < 10) {
1417 printk("md: data_size too small on %s\n",
1418 bdevname(rdev
->bdev
,b
));
1423 memcmp(sb
->pad3
, sb
->pad3
+1, sizeof(sb
->pad3
) - sizeof(sb
->pad3
[1])))
1424 /* Some padding is non-zero, might be a new feature */
1427 rdev
->preferred_minor
= 0xffff;
1428 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1429 rdev
->new_data_offset
= rdev
->data_offset
;
1430 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
) &&
1431 (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
1432 rdev
->new_data_offset
+= (s32
)le32_to_cpu(sb
->new_offset
);
1433 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1435 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1436 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1437 if (rdev
->sb_size
& bmask
)
1438 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1441 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1444 && rdev
->new_data_offset
< sb_start
+ (rdev
->sb_size
/512))
1447 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1450 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1452 if (!rdev
->bb_page
) {
1453 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1457 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1458 rdev
->badblocks
.count
== 0) {
1459 /* need to load the bad block list.
1460 * Currently we limit it to one page.
1466 int sectors
= le16_to_cpu(sb
->bblog_size
);
1467 if (sectors
> (PAGE_SIZE
/ 512))
1469 offset
= le32_to_cpu(sb
->bblog_offset
);
1472 bb_sector
= (long long)offset
;
1473 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1474 rdev
->bb_page
, READ
, true))
1476 bbp
= (u64
*)page_address(rdev
->bb_page
);
1477 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1478 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1479 u64 bb
= le64_to_cpu(*bbp
);
1480 int count
= bb
& (0x3ff);
1481 u64 sector
= bb
>> 10;
1482 sector
<<= sb
->bblog_shift
;
1483 count
<<= sb
->bblog_shift
;
1486 if (badblocks_set(&rdev
->badblocks
, sector
, count
, 1))
1489 } else if (sb
->bblog_offset
!= 0)
1490 rdev
->badblocks
.shift
= 0;
1496 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1498 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1499 sb
->level
!= refsb
->level
||
1500 sb
->layout
!= refsb
->layout
||
1501 sb
->chunksize
!= refsb
->chunksize
) {
1502 printk(KERN_WARNING
"md: %s has strangely different"
1503 " superblock to %s\n",
1504 bdevname(rdev
->bdev
,b
),
1505 bdevname(refdev
->bdev
,b2
));
1508 ev1
= le64_to_cpu(sb
->events
);
1509 ev2
= le64_to_cpu(refsb
->events
);
1516 if (minor_version
) {
1517 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9);
1518 sectors
-= rdev
->data_offset
;
1520 sectors
= rdev
->sb_start
;
1521 if (sectors
< le64_to_cpu(sb
->data_size
))
1523 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1527 static int super_1_validate(struct mddev
*mddev
, struct md_rdev
*rdev
)
1529 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1530 __u64 ev1
= le64_to_cpu(sb
->events
);
1532 rdev
->raid_disk
= -1;
1533 clear_bit(Faulty
, &rdev
->flags
);
1534 clear_bit(In_sync
, &rdev
->flags
);
1535 clear_bit(Bitmap_sync
, &rdev
->flags
);
1536 clear_bit(WriteMostly
, &rdev
->flags
);
1538 if (mddev
->raid_disks
== 0) {
1539 mddev
->major_version
= 1;
1540 mddev
->patch_version
= 0;
1541 mddev
->external
= 0;
1542 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1543 mddev
->ctime
= le64_to_cpu(sb
->ctime
);
1544 mddev
->utime
= le64_to_cpu(sb
->utime
);
1545 mddev
->level
= le32_to_cpu(sb
->level
);
1546 mddev
->clevel
[0] = 0;
1547 mddev
->layout
= le32_to_cpu(sb
->layout
);
1548 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1549 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1550 mddev
->events
= ev1
;
1551 mddev
->bitmap_info
.offset
= 0;
1552 mddev
->bitmap_info
.space
= 0;
1553 /* Default location for bitmap is 1K after superblock
1554 * using 3K - total of 4K
1556 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1557 mddev
->bitmap_info
.default_space
= (4096-1024) >> 9;
1558 mddev
->reshape_backwards
= 0;
1560 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1561 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1563 mddev
->max_disks
= (4096-256)/2;
1565 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1566 mddev
->bitmap_info
.file
== NULL
) {
1567 mddev
->bitmap_info
.offset
=
1568 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1569 /* Metadata doesn't record how much space is available.
1570 * For 1.0, we assume we can use up to the superblock
1571 * if before, else to 4K beyond superblock.
1572 * For others, assume no change is possible.
1574 if (mddev
->minor_version
> 0)
1575 mddev
->bitmap_info
.space
= 0;
1576 else if (mddev
->bitmap_info
.offset
> 0)
1577 mddev
->bitmap_info
.space
=
1578 8 - mddev
->bitmap_info
.offset
;
1580 mddev
->bitmap_info
.space
=
1581 -mddev
->bitmap_info
.offset
;
1584 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1585 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1586 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1587 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1588 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1589 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1590 if (mddev
->delta_disks
< 0 ||
1591 (mddev
->delta_disks
== 0 &&
1592 (le32_to_cpu(sb
->feature_map
)
1593 & MD_FEATURE_RESHAPE_BACKWARDS
)))
1594 mddev
->reshape_backwards
= 1;
1596 mddev
->reshape_position
= MaxSector
;
1597 mddev
->delta_disks
= 0;
1598 mddev
->new_level
= mddev
->level
;
1599 mddev
->new_layout
= mddev
->layout
;
1600 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1603 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
) {
1604 set_bit(MD_HAS_JOURNAL
, &mddev
->flags
);
1605 if (mddev
->recovery_cp
== MaxSector
)
1606 set_bit(MD_JOURNAL_CLEAN
, &mddev
->flags
);
1608 } else if (mddev
->pers
== NULL
) {
1609 /* Insist of good event counter while assembling, except for
1610 * spares (which don't need an event count) */
1612 if (rdev
->desc_nr
>= 0 &&
1613 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1614 (le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < MD_DISK_ROLE_MAX
||
1615 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) == MD_DISK_ROLE_JOURNAL
))
1616 if (ev1
< mddev
->events
)
1618 } else if (mddev
->bitmap
) {
1619 /* If adding to array with a bitmap, then we can accept an
1620 * older device, but not too old.
1622 if (ev1
< mddev
->bitmap
->events_cleared
)
1624 if (ev1
< mddev
->events
)
1625 set_bit(Bitmap_sync
, &rdev
->flags
);
1627 if (ev1
< mddev
->events
)
1628 /* just a hot-add of a new device, leave raid_disk at -1 */
1631 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1633 if (rdev
->desc_nr
< 0 ||
1634 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1635 role
= MD_DISK_ROLE_SPARE
;
1638 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1640 case MD_DISK_ROLE_SPARE
: /* spare */
1642 case MD_DISK_ROLE_FAULTY
: /* faulty */
1643 set_bit(Faulty
, &rdev
->flags
);
1645 case MD_DISK_ROLE_JOURNAL
: /* journal device */
1646 if (!(le32_to_cpu(sb
->feature_map
) & MD_FEATURE_JOURNAL
)) {
1647 /* journal device without journal feature */
1649 "md: journal device provided without journal feature, ignoring the device\n");
1652 set_bit(Journal
, &rdev
->flags
);
1653 rdev
->journal_tail
= le64_to_cpu(sb
->journal_tail
);
1654 rdev
->raid_disk
= 0;
1657 rdev
->saved_raid_disk
= role
;
1658 if ((le32_to_cpu(sb
->feature_map
) &
1659 MD_FEATURE_RECOVERY_OFFSET
)) {
1660 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1661 if (!(le32_to_cpu(sb
->feature_map
) &
1662 MD_FEATURE_RECOVERY_BITMAP
))
1663 rdev
->saved_raid_disk
= -1;
1665 set_bit(In_sync
, &rdev
->flags
);
1666 rdev
->raid_disk
= role
;
1669 if (sb
->devflags
& WriteMostly1
)
1670 set_bit(WriteMostly
, &rdev
->flags
);
1671 if (le32_to_cpu(sb
->feature_map
) & MD_FEATURE_REPLACEMENT
)
1672 set_bit(Replacement
, &rdev
->flags
);
1673 } else /* MULTIPATH are always insync */
1674 set_bit(In_sync
, &rdev
->flags
);
1679 static void super_1_sync(struct mddev
*mddev
, struct md_rdev
*rdev
)
1681 struct mdp_superblock_1
*sb
;
1682 struct md_rdev
*rdev2
;
1684 /* make rdev->sb match mddev and rdev data. */
1686 sb
= page_address(rdev
->sb_page
);
1688 sb
->feature_map
= 0;
1690 sb
->recovery_offset
= cpu_to_le64(0);
1691 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1693 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1694 sb
->events
= cpu_to_le64(mddev
->events
);
1696 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1697 else if (test_bit(MD_JOURNAL_CLEAN
, &mddev
->flags
))
1698 sb
->resync_offset
= cpu_to_le64(MaxSector
);
1700 sb
->resync_offset
= cpu_to_le64(0);
1702 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1704 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1705 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1706 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1707 sb
->level
= cpu_to_le32(mddev
->level
);
1708 sb
->layout
= cpu_to_le32(mddev
->layout
);
1710 if (test_bit(WriteMostly
, &rdev
->flags
))
1711 sb
->devflags
|= WriteMostly1
;
1713 sb
->devflags
&= ~WriteMostly1
;
1714 sb
->data_offset
= cpu_to_le64(rdev
->data_offset
);
1715 sb
->data_size
= cpu_to_le64(rdev
->sectors
);
1717 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1718 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1719 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1722 if (rdev
->raid_disk
>= 0 && !test_bit(Journal
, &rdev
->flags
) &&
1723 !test_bit(In_sync
, &rdev
->flags
)) {
1725 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1726 sb
->recovery_offset
=
1727 cpu_to_le64(rdev
->recovery_offset
);
1728 if (rdev
->saved_raid_disk
>= 0 && mddev
->bitmap
)
1730 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP
);
1732 /* Note: recovery_offset and journal_tail share space */
1733 if (test_bit(Journal
, &rdev
->flags
))
1734 sb
->journal_tail
= cpu_to_le64(rdev
->journal_tail
);
1735 if (test_bit(Replacement
, &rdev
->flags
))
1737 cpu_to_le32(MD_FEATURE_REPLACEMENT
);
1739 if (mddev
->reshape_position
!= MaxSector
) {
1740 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1741 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1742 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1743 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1744 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1745 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1746 if (mddev
->delta_disks
== 0 &&
1747 mddev
->reshape_backwards
)
1749 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1750 if (rdev
->new_data_offset
!= rdev
->data_offset
) {
1752 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET
);
1753 sb
->new_offset
= cpu_to_le32((__u32
)(rdev
->new_data_offset
1754 - rdev
->data_offset
));
1758 if (mddev_is_clustered(mddev
))
1759 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_CLUSTERED
);
1761 if (rdev
->badblocks
.count
== 0)
1762 /* Nothing to do for bad blocks*/ ;
1763 else if (sb
->bblog_offset
== 0)
1764 /* Cannot record bad blocks on this device */
1765 md_error(mddev
, rdev
);
1767 struct badblocks
*bb
= &rdev
->badblocks
;
1768 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1770 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1775 seq
= read_seqbegin(&bb
->lock
);
1777 memset(bbp
, 0xff, PAGE_SIZE
);
1779 for (i
= 0 ; i
< bb
->count
; i
++) {
1780 u64 internal_bb
= p
[i
];
1781 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1782 | BB_LEN(internal_bb
));
1783 bbp
[i
] = cpu_to_le64(store_bb
);
1786 if (read_seqretry(&bb
->lock
, seq
))
1789 bb
->sector
= (rdev
->sb_start
+
1790 (int)le32_to_cpu(sb
->bblog_offset
));
1791 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1796 rdev_for_each(rdev2
, mddev
)
1797 if (rdev2
->desc_nr
+1 > max_dev
)
1798 max_dev
= rdev2
->desc_nr
+1;
1800 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1802 sb
->max_dev
= cpu_to_le32(max_dev
);
1803 rdev
->sb_size
= max_dev
* 2 + 256;
1804 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1805 if (rdev
->sb_size
& bmask
)
1806 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1808 max_dev
= le32_to_cpu(sb
->max_dev
);
1810 for (i
=0; i
<max_dev
;i
++)
1811 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_FAULTY
);
1813 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
))
1814 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_JOURNAL
);
1816 rdev_for_each(rdev2
, mddev
) {
1818 if (test_bit(Faulty
, &rdev2
->flags
))
1819 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_FAULTY
);
1820 else if (test_bit(In_sync
, &rdev2
->flags
))
1821 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1822 else if (test_bit(Journal
, &rdev2
->flags
))
1823 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_JOURNAL
);
1824 else if (rdev2
->raid_disk
>= 0)
1825 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1827 sb
->dev_roles
[i
] = cpu_to_le16(MD_DISK_ROLE_SPARE
);
1830 sb
->sb_csum
= calc_sb_1_csum(sb
);
1833 static unsigned long long
1834 super_1_rdev_size_change(struct md_rdev
*rdev
, sector_t num_sectors
)
1836 struct mdp_superblock_1
*sb
;
1837 sector_t max_sectors
;
1838 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1839 return 0; /* component must fit device */
1840 if (rdev
->data_offset
!= rdev
->new_data_offset
)
1841 return 0; /* too confusing */
1842 if (rdev
->sb_start
< rdev
->data_offset
) {
1843 /* minor versions 1 and 2; superblock before data */
1844 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1845 max_sectors
-= rdev
->data_offset
;
1846 if (!num_sectors
|| num_sectors
> max_sectors
)
1847 num_sectors
= max_sectors
;
1848 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1849 /* minor version 0 with bitmap we can't move */
1852 /* minor version 0; superblock after data */
1854 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1855 sb_start
&= ~(sector_t
)(4*2 - 1);
1856 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1857 if (!num_sectors
|| num_sectors
> max_sectors
)
1858 num_sectors
= max_sectors
;
1859 rdev
->sb_start
= sb_start
;
1861 sb
= page_address(rdev
->sb_page
);
1862 sb
->data_size
= cpu_to_le64(num_sectors
);
1863 sb
->super_offset
= rdev
->sb_start
;
1864 sb
->sb_csum
= calc_sb_1_csum(sb
);
1865 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1867 md_super_wait(rdev
->mddev
);
1873 super_1_allow_new_offset(struct md_rdev
*rdev
,
1874 unsigned long long new_offset
)
1876 /* All necessary checks on new >= old have been done */
1877 struct bitmap
*bitmap
;
1878 if (new_offset
>= rdev
->data_offset
)
1881 /* with 1.0 metadata, there is no metadata to tread on
1882 * so we can always move back */
1883 if (rdev
->mddev
->minor_version
== 0)
1886 /* otherwise we must be sure not to step on
1887 * any metadata, so stay:
1888 * 36K beyond start of superblock
1889 * beyond end of badblocks
1890 * beyond write-intent bitmap
1892 if (rdev
->sb_start
+ (32+4)*2 > new_offset
)
1894 bitmap
= rdev
->mddev
->bitmap
;
1895 if (bitmap
&& !rdev
->mddev
->bitmap_info
.file
&&
1896 rdev
->sb_start
+ rdev
->mddev
->bitmap_info
.offset
+
1897 bitmap
->storage
.file_pages
* (PAGE_SIZE
>>9) > new_offset
)
1899 if (rdev
->badblocks
.sector
+ rdev
->badblocks
.size
> new_offset
)
1905 static struct super_type super_types
[] = {
1908 .owner
= THIS_MODULE
,
1909 .load_super
= super_90_load
,
1910 .validate_super
= super_90_validate
,
1911 .sync_super
= super_90_sync
,
1912 .rdev_size_change
= super_90_rdev_size_change
,
1913 .allow_new_offset
= super_90_allow_new_offset
,
1917 .owner
= THIS_MODULE
,
1918 .load_super
= super_1_load
,
1919 .validate_super
= super_1_validate
,
1920 .sync_super
= super_1_sync
,
1921 .rdev_size_change
= super_1_rdev_size_change
,
1922 .allow_new_offset
= super_1_allow_new_offset
,
1926 static void sync_super(struct mddev
*mddev
, struct md_rdev
*rdev
)
1928 if (mddev
->sync_super
) {
1929 mddev
->sync_super(mddev
, rdev
);
1933 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1935 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1938 static int match_mddev_units(struct mddev
*mddev1
, struct mddev
*mddev2
)
1940 struct md_rdev
*rdev
, *rdev2
;
1943 rdev_for_each_rcu(rdev
, mddev1
) {
1944 if (test_bit(Faulty
, &rdev
->flags
) ||
1945 test_bit(Journal
, &rdev
->flags
) ||
1946 rdev
->raid_disk
== -1)
1948 rdev_for_each_rcu(rdev2
, mddev2
) {
1949 if (test_bit(Faulty
, &rdev2
->flags
) ||
1950 test_bit(Journal
, &rdev2
->flags
) ||
1951 rdev2
->raid_disk
== -1)
1953 if (rdev
->bdev
->bd_contains
==
1954 rdev2
->bdev
->bd_contains
) {
1964 static LIST_HEAD(pending_raid_disks
);
1967 * Try to register data integrity profile for an mddev
1969 * This is called when an array is started and after a disk has been kicked
1970 * from the array. It only succeeds if all working and active component devices
1971 * are integrity capable with matching profiles.
1973 int md_integrity_register(struct mddev
*mddev
)
1975 struct md_rdev
*rdev
, *reference
= NULL
;
1977 if (list_empty(&mddev
->disks
))
1978 return 0; /* nothing to do */
1979 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1980 return 0; /* shouldn't register, or already is */
1981 rdev_for_each(rdev
, mddev
) {
1982 /* skip spares and non-functional disks */
1983 if (test_bit(Faulty
, &rdev
->flags
))
1985 if (rdev
->raid_disk
< 0)
1988 /* Use the first rdev as the reference */
1992 /* does this rdev's profile match the reference profile? */
1993 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1994 rdev
->bdev
->bd_disk
) < 0)
1997 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
2000 * All component devices are integrity capable and have matching
2001 * profiles, register the common profile for the md device.
2003 blk_integrity_register(mddev
->gendisk
,
2004 bdev_get_integrity(reference
->bdev
));
2006 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
2007 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
2008 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
2014 EXPORT_SYMBOL(md_integrity_register
);
2017 * Attempt to add an rdev, but only if it is consistent with the current
2020 int md_integrity_add_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
2022 struct blk_integrity
*bi_rdev
;
2023 struct blk_integrity
*bi_mddev
;
2024 char name
[BDEVNAME_SIZE
];
2026 if (!mddev
->gendisk
)
2029 bi_rdev
= bdev_get_integrity(rdev
->bdev
);
2030 bi_mddev
= blk_get_integrity(mddev
->gendisk
);
2032 if (!bi_mddev
) /* nothing to do */
2035 if (blk_integrity_compare(mddev
->gendisk
, rdev
->bdev
->bd_disk
) != 0) {
2036 printk(KERN_NOTICE
"%s: incompatible integrity profile for %s\n",
2037 mdname(mddev
), bdevname(rdev
->bdev
, name
));
2043 EXPORT_SYMBOL(md_integrity_add_rdev
);
2045 static int bind_rdev_to_array(struct md_rdev
*rdev
, struct mddev
*mddev
)
2047 char b
[BDEVNAME_SIZE
];
2051 /* prevent duplicates */
2052 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2055 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2056 if (!test_bit(Journal
, &rdev
->flags
) &&
2058 (mddev
->dev_sectors
== 0 || rdev
->sectors
< mddev
->dev_sectors
)) {
2060 /* Cannot change size, so fail
2061 * If mddev->level <= 0, then we don't care
2062 * about aligning sizes (e.g. linear)
2064 if (mddev
->level
> 0)
2067 mddev
->dev_sectors
= rdev
->sectors
;
2070 /* Verify rdev->desc_nr is unique.
2071 * If it is -1, assign a free number, else
2072 * check number is not in use
2075 if (rdev
->desc_nr
< 0) {
2078 choice
= mddev
->raid_disks
;
2079 while (md_find_rdev_nr_rcu(mddev
, choice
))
2081 rdev
->desc_nr
= choice
;
2083 if (md_find_rdev_nr_rcu(mddev
, rdev
->desc_nr
)) {
2089 if (!test_bit(Journal
, &rdev
->flags
) &&
2090 mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2091 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2092 mdname(mddev
), mddev
->max_disks
);
2095 bdevname(rdev
->bdev
,b
);
2096 strreplace(b
, '/', '!');
2098 rdev
->mddev
= mddev
;
2099 printk(KERN_INFO
"md: bind<%s>\n", b
);
2101 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2104 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2105 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2106 /* failure here is OK */;
2107 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2109 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2110 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2112 /* May as well allow recovery to be retried once */
2113 mddev
->recovery_disabled
++;
2118 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2123 static void md_delayed_delete(struct work_struct
*ws
)
2125 struct md_rdev
*rdev
= container_of(ws
, struct md_rdev
, del_work
);
2126 kobject_del(&rdev
->kobj
);
2127 kobject_put(&rdev
->kobj
);
2130 static void unbind_rdev_from_array(struct md_rdev
*rdev
)
2132 char b
[BDEVNAME_SIZE
];
2134 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2135 list_del_rcu(&rdev
->same_set
);
2136 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2138 sysfs_remove_link(&rdev
->kobj
, "block");
2139 sysfs_put(rdev
->sysfs_state
);
2140 rdev
->sysfs_state
= NULL
;
2141 rdev
->badblocks
.count
= 0;
2142 /* We need to delay this, otherwise we can deadlock when
2143 * writing to 'remove' to "dev/state". We also need
2144 * to delay it due to rcu usage.
2147 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2148 kobject_get(&rdev
->kobj
);
2149 queue_work(md_misc_wq
, &rdev
->del_work
);
2153 * prevent the device from being mounted, repartitioned or
2154 * otherwise reused by a RAID array (or any other kernel
2155 * subsystem), by bd_claiming the device.
2157 static int lock_rdev(struct md_rdev
*rdev
, dev_t dev
, int shared
)
2160 struct block_device
*bdev
;
2161 char b
[BDEVNAME_SIZE
];
2163 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2164 shared
? (struct md_rdev
*)lock_rdev
: rdev
);
2166 printk(KERN_ERR
"md: could not open %s.\n",
2167 __bdevname(dev
, b
));
2168 return PTR_ERR(bdev
);
2174 static void unlock_rdev(struct md_rdev
*rdev
)
2176 struct block_device
*bdev
= rdev
->bdev
;
2178 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2181 void md_autodetect_dev(dev_t dev
);
2183 static void export_rdev(struct md_rdev
*rdev
)
2185 char b
[BDEVNAME_SIZE
];
2187 printk(KERN_INFO
"md: export_rdev(%s)\n",
2188 bdevname(rdev
->bdev
,b
));
2189 md_rdev_clear(rdev
);
2191 if (test_bit(AutoDetected
, &rdev
->flags
))
2192 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2195 kobject_put(&rdev
->kobj
);
2198 void md_kick_rdev_from_array(struct md_rdev
*rdev
)
2200 unbind_rdev_from_array(rdev
);
2203 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array
);
2205 static void export_array(struct mddev
*mddev
)
2207 struct md_rdev
*rdev
;
2209 while (!list_empty(&mddev
->disks
)) {
2210 rdev
= list_first_entry(&mddev
->disks
, struct md_rdev
,
2212 md_kick_rdev_from_array(rdev
);
2214 mddev
->raid_disks
= 0;
2215 mddev
->major_version
= 0;
2218 static void sync_sbs(struct mddev
*mddev
, int nospares
)
2220 /* Update each superblock (in-memory image), but
2221 * if we are allowed to, skip spares which already
2222 * have the right event counter, or have one earlier
2223 * (which would mean they aren't being marked as dirty
2224 * with the rest of the array)
2226 struct md_rdev
*rdev
;
2227 rdev_for_each(rdev
, mddev
) {
2228 if (rdev
->sb_events
== mddev
->events
||
2230 rdev
->raid_disk
< 0 &&
2231 rdev
->sb_events
+1 == mddev
->events
)) {
2232 /* Don't update this superblock */
2233 rdev
->sb_loaded
= 2;
2235 sync_super(mddev
, rdev
);
2236 rdev
->sb_loaded
= 1;
2241 static bool does_sb_need_changing(struct mddev
*mddev
)
2243 struct md_rdev
*rdev
;
2244 struct mdp_superblock_1
*sb
;
2247 /* Find a good rdev */
2248 rdev_for_each(rdev
, mddev
)
2249 if ((rdev
->raid_disk
>= 0) && !test_bit(Faulty
, &rdev
->flags
))
2252 /* No good device found. */
2256 sb
= page_address(rdev
->sb_page
);
2257 /* Check if a device has become faulty or a spare become active */
2258 rdev_for_each(rdev
, mddev
) {
2259 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
2260 /* Device activated? */
2261 if (role
== 0xffff && rdev
->raid_disk
>=0 &&
2262 !test_bit(Faulty
, &rdev
->flags
))
2264 /* Device turned faulty? */
2265 if (test_bit(Faulty
, &rdev
->flags
) && (role
< 0xfffd))
2269 /* Check if any mddev parameters have changed */
2270 if ((mddev
->dev_sectors
!= le64_to_cpu(sb
->size
)) ||
2271 (mddev
->reshape_position
!= le64_to_cpu(sb
->reshape_position
)) ||
2272 (mddev
->layout
!= le64_to_cpu(sb
->layout
)) ||
2273 (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
)) ||
2274 (mddev
->chunk_sectors
!= le32_to_cpu(sb
->chunksize
)))
2280 void md_update_sb(struct mddev
*mddev
, int force_change
)
2282 struct md_rdev
*rdev
;
2285 int any_badblocks_changed
= 0;
2290 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2295 if (mddev_is_clustered(mddev
)) {
2296 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2298 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2300 ret
= md_cluster_ops
->metadata_update_start(mddev
);
2301 /* Has someone else has updated the sb */
2302 if (!does_sb_need_changing(mddev
)) {
2304 md_cluster_ops
->metadata_update_cancel(mddev
);
2305 bit_clear_unless(&mddev
->flags
, BIT(MD_CHANGE_PENDING
),
2306 BIT(MD_CHANGE_DEVS
) |
2307 BIT(MD_CHANGE_CLEAN
));
2312 /* First make sure individual recovery_offsets are correct */
2313 rdev_for_each(rdev
, mddev
) {
2314 if (rdev
->raid_disk
>= 0 &&
2315 mddev
->delta_disks
>= 0 &&
2316 !test_bit(Journal
, &rdev
->flags
) &&
2317 !test_bit(In_sync
, &rdev
->flags
) &&
2318 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2319 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2322 if (!mddev
->persistent
) {
2323 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2324 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2325 if (!mddev
->external
) {
2326 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2327 rdev_for_each(rdev
, mddev
) {
2328 if (rdev
->badblocks
.changed
) {
2329 rdev
->badblocks
.changed
= 0;
2330 ack_all_badblocks(&rdev
->badblocks
);
2331 md_error(mddev
, rdev
);
2333 clear_bit(Blocked
, &rdev
->flags
);
2334 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2335 wake_up(&rdev
->blocked_wait
);
2338 wake_up(&mddev
->sb_wait
);
2342 spin_lock(&mddev
->lock
);
2344 mddev
->utime
= ktime_get_real_seconds();
2346 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2348 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2349 /* just a clean<-> dirty transition, possibly leave spares alone,
2350 * though if events isn't the right even/odd, we will have to do
2356 if (mddev
->degraded
)
2357 /* If the array is degraded, then skipping spares is both
2358 * dangerous and fairly pointless.
2359 * Dangerous because a device that was removed from the array
2360 * might have a event_count that still looks up-to-date,
2361 * so it can be re-added without a resync.
2362 * Pointless because if there are any spares to skip,
2363 * then a recovery will happen and soon that array won't
2364 * be degraded any more and the spare can go back to sleep then.
2368 sync_req
= mddev
->in_sync
;
2370 /* If this is just a dirty<->clean transition, and the array is clean
2371 * and 'events' is odd, we can roll back to the previous clean state */
2373 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2374 && mddev
->can_decrease_events
2375 && mddev
->events
!= 1) {
2377 mddev
->can_decrease_events
= 0;
2379 /* otherwise we have to go forward and ... */
2381 mddev
->can_decrease_events
= nospares
;
2385 * This 64-bit counter should never wrap.
2386 * Either we are in around ~1 trillion A.C., assuming
2387 * 1 reboot per second, or we have a bug...
2389 WARN_ON(mddev
->events
== 0);
2391 rdev_for_each(rdev
, mddev
) {
2392 if (rdev
->badblocks
.changed
)
2393 any_badblocks_changed
++;
2394 if (test_bit(Faulty
, &rdev
->flags
))
2395 set_bit(FaultRecorded
, &rdev
->flags
);
2398 sync_sbs(mddev
, nospares
);
2399 spin_unlock(&mddev
->lock
);
2401 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2402 mdname(mddev
), mddev
->in_sync
);
2404 bitmap_update_sb(mddev
->bitmap
);
2405 rdev_for_each(rdev
, mddev
) {
2406 char b
[BDEVNAME_SIZE
];
2408 if (rdev
->sb_loaded
!= 1)
2409 continue; /* no noise on spare devices */
2411 if (!test_bit(Faulty
, &rdev
->flags
)) {
2412 md_super_write(mddev
,rdev
,
2413 rdev
->sb_start
, rdev
->sb_size
,
2415 pr_debug("md: (write) %s's sb offset: %llu\n",
2416 bdevname(rdev
->bdev
, b
),
2417 (unsigned long long)rdev
->sb_start
);
2418 rdev
->sb_events
= mddev
->events
;
2419 if (rdev
->badblocks
.size
) {
2420 md_super_write(mddev
, rdev
,
2421 rdev
->badblocks
.sector
,
2422 rdev
->badblocks
.size
<< 9,
2424 rdev
->badblocks
.size
= 0;
2428 pr_debug("md: %s (skipping faulty)\n",
2429 bdevname(rdev
->bdev
, b
));
2431 if (mddev
->level
== LEVEL_MULTIPATH
)
2432 /* only need to write one superblock... */
2435 md_super_wait(mddev
);
2436 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2438 if (mddev_is_clustered(mddev
) && ret
== 0)
2439 md_cluster_ops
->metadata_update_finish(mddev
);
2441 if (mddev
->in_sync
!= sync_req
||
2442 !bit_clear_unless(&mddev
->flags
, BIT(MD_CHANGE_PENDING
),
2443 BIT(MD_CHANGE_DEVS
) | BIT(MD_CHANGE_CLEAN
)))
2444 /* have to write it out again */
2446 wake_up(&mddev
->sb_wait
);
2447 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2448 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2450 rdev_for_each(rdev
, mddev
) {
2451 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2452 clear_bit(Blocked
, &rdev
->flags
);
2454 if (any_badblocks_changed
)
2455 ack_all_badblocks(&rdev
->badblocks
);
2456 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2457 wake_up(&rdev
->blocked_wait
);
2460 EXPORT_SYMBOL(md_update_sb
);
2462 static int add_bound_rdev(struct md_rdev
*rdev
)
2464 struct mddev
*mddev
= rdev
->mddev
;
2466 bool add_journal
= test_bit(Journal
, &rdev
->flags
);
2468 if (!mddev
->pers
->hot_remove_disk
|| add_journal
) {
2469 /* If there is hot_add_disk but no hot_remove_disk
2470 * then added disks for geometry changes,
2471 * and should be added immediately.
2473 super_types
[mddev
->major_version
].
2474 validate_super(mddev
, rdev
);
2476 mddev_suspend(mddev
);
2477 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
2479 mddev_resume(mddev
);
2481 unbind_rdev_from_array(rdev
);
2486 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2488 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2489 if (mddev
->degraded
)
2490 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
2491 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2492 md_new_event(mddev
);
2493 md_wakeup_thread(mddev
->thread
);
2497 /* words written to sysfs files may, or may not, be \n terminated.
2498 * We want to accept with case. For this we use cmd_match.
2500 static int cmd_match(const char *cmd
, const char *str
)
2502 /* See if cmd, written into a sysfs file, matches
2503 * str. They must either be the same, or cmd can
2504 * have a trailing newline
2506 while (*cmd
&& *str
&& *cmd
== *str
) {
2517 struct rdev_sysfs_entry
{
2518 struct attribute attr
;
2519 ssize_t (*show
)(struct md_rdev
*, char *);
2520 ssize_t (*store
)(struct md_rdev
*, const char *, size_t);
2524 state_show(struct md_rdev
*rdev
, char *page
)
2528 unsigned long flags
= ACCESS_ONCE(rdev
->flags
);
2530 if (test_bit(Faulty
, &flags
) ||
2531 rdev
->badblocks
.unacked_exist
) {
2532 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2535 if (test_bit(In_sync
, &flags
)) {
2536 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2539 if (test_bit(Journal
, &flags
)) {
2540 len
+= sprintf(page
+len
, "%sjournal",sep
);
2543 if (test_bit(WriteMostly
, &flags
)) {
2544 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2547 if (test_bit(Blocked
, &flags
) ||
2548 (rdev
->badblocks
.unacked_exist
2549 && !test_bit(Faulty
, &flags
))) {
2550 len
+= sprintf(page
+len
, "%sblocked", sep
);
2553 if (!test_bit(Faulty
, &flags
) &&
2554 !test_bit(Journal
, &flags
) &&
2555 !test_bit(In_sync
, &flags
)) {
2556 len
+= sprintf(page
+len
, "%sspare", sep
);
2559 if (test_bit(WriteErrorSeen
, &flags
)) {
2560 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2563 if (test_bit(WantReplacement
, &flags
)) {
2564 len
+= sprintf(page
+len
, "%swant_replacement", sep
);
2567 if (test_bit(Replacement
, &flags
)) {
2568 len
+= sprintf(page
+len
, "%sreplacement", sep
);
2572 return len
+sprintf(page
+len
, "\n");
2576 state_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2579 * faulty - simulates an error
2580 * remove - disconnects the device
2581 * writemostly - sets write_mostly
2582 * -writemostly - clears write_mostly
2583 * blocked - sets the Blocked flags
2584 * -blocked - clears the Blocked and possibly simulates an error
2585 * insync - sets Insync providing device isn't active
2586 * -insync - clear Insync for a device with a slot assigned,
2587 * so that it gets rebuilt based on bitmap
2588 * write_error - sets WriteErrorSeen
2589 * -write_error - clears WriteErrorSeen
2592 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2593 md_error(rdev
->mddev
, rdev
);
2594 if (test_bit(Faulty
, &rdev
->flags
))
2598 } else if (cmd_match(buf
, "remove")) {
2599 if (rdev
->raid_disk
>= 0)
2602 struct mddev
*mddev
= rdev
->mddev
;
2604 if (mddev_is_clustered(mddev
))
2605 err
= md_cluster_ops
->remove_disk(mddev
, rdev
);
2608 md_kick_rdev_from_array(rdev
);
2610 md_update_sb(mddev
, 1);
2611 md_new_event(mddev
);
2614 } else if (cmd_match(buf
, "writemostly")) {
2615 set_bit(WriteMostly
, &rdev
->flags
);
2617 } else if (cmd_match(buf
, "-writemostly")) {
2618 clear_bit(WriteMostly
, &rdev
->flags
);
2620 } else if (cmd_match(buf
, "blocked")) {
2621 set_bit(Blocked
, &rdev
->flags
);
2623 } else if (cmd_match(buf
, "-blocked")) {
2624 if (!test_bit(Faulty
, &rdev
->flags
) &&
2625 rdev
->badblocks
.unacked_exist
) {
2626 /* metadata handler doesn't understand badblocks,
2627 * so we need to fail the device
2629 md_error(rdev
->mddev
, rdev
);
2631 clear_bit(Blocked
, &rdev
->flags
);
2632 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2633 wake_up(&rdev
->blocked_wait
);
2634 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2635 md_wakeup_thread(rdev
->mddev
->thread
);
2638 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2639 set_bit(In_sync
, &rdev
->flags
);
2641 } else if (cmd_match(buf
, "-insync") && rdev
->raid_disk
>= 0 &&
2642 !test_bit(Journal
, &rdev
->flags
)) {
2643 if (rdev
->mddev
->pers
== NULL
) {
2644 clear_bit(In_sync
, &rdev
->flags
);
2645 rdev
->saved_raid_disk
= rdev
->raid_disk
;
2646 rdev
->raid_disk
= -1;
2649 } else if (cmd_match(buf
, "write_error")) {
2650 set_bit(WriteErrorSeen
, &rdev
->flags
);
2652 } else if (cmd_match(buf
, "-write_error")) {
2653 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2655 } else if (cmd_match(buf
, "want_replacement")) {
2656 /* Any non-spare device that is not a replacement can
2657 * become want_replacement at any time, but we then need to
2658 * check if recovery is needed.
2660 if (rdev
->raid_disk
>= 0 &&
2661 !test_bit(Journal
, &rdev
->flags
) &&
2662 !test_bit(Replacement
, &rdev
->flags
))
2663 set_bit(WantReplacement
, &rdev
->flags
);
2664 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2665 md_wakeup_thread(rdev
->mddev
->thread
);
2667 } else if (cmd_match(buf
, "-want_replacement")) {
2668 /* Clearing 'want_replacement' is always allowed.
2669 * Once replacements starts it is too late though.
2672 clear_bit(WantReplacement
, &rdev
->flags
);
2673 } else if (cmd_match(buf
, "replacement")) {
2674 /* Can only set a device as a replacement when array has not
2675 * yet been started. Once running, replacement is automatic
2676 * from spares, or by assigning 'slot'.
2678 if (rdev
->mddev
->pers
)
2681 set_bit(Replacement
, &rdev
->flags
);
2684 } else if (cmd_match(buf
, "-replacement")) {
2685 /* Similarly, can only clear Replacement before start */
2686 if (rdev
->mddev
->pers
)
2689 clear_bit(Replacement
, &rdev
->flags
);
2692 } else if (cmd_match(buf
, "re-add")) {
2693 if (test_bit(Faulty
, &rdev
->flags
) && (rdev
->raid_disk
== -1)) {
2694 /* clear_bit is performed _after_ all the devices
2695 * have their local Faulty bit cleared. If any writes
2696 * happen in the meantime in the local node, they
2697 * will land in the local bitmap, which will be synced
2698 * by this node eventually
2700 if (!mddev_is_clustered(rdev
->mddev
) ||
2701 (err
= md_cluster_ops
->gather_bitmaps(rdev
)) == 0) {
2702 clear_bit(Faulty
, &rdev
->flags
);
2703 err
= add_bound_rdev(rdev
);
2709 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2710 return err
? err
: len
;
2712 static struct rdev_sysfs_entry rdev_state
=
2713 __ATTR_PREALLOC(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2716 errors_show(struct md_rdev
*rdev
, char *page
)
2718 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2722 errors_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2727 rv
= kstrtouint(buf
, 10, &n
);
2730 atomic_set(&rdev
->corrected_errors
, n
);
2733 static struct rdev_sysfs_entry rdev_errors
=
2734 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2737 slot_show(struct md_rdev
*rdev
, char *page
)
2739 if (test_bit(Journal
, &rdev
->flags
))
2740 return sprintf(page
, "journal\n");
2741 else if (rdev
->raid_disk
< 0)
2742 return sprintf(page
, "none\n");
2744 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2748 slot_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2753 if (test_bit(Journal
, &rdev
->flags
))
2755 if (strncmp(buf
, "none", 4)==0)
2758 err
= kstrtouint(buf
, 10, (unsigned int *)&slot
);
2762 if (rdev
->mddev
->pers
&& slot
== -1) {
2763 /* Setting 'slot' on an active array requires also
2764 * updating the 'rd%d' link, and communicating
2765 * with the personality with ->hot_*_disk.
2766 * For now we only support removing
2767 * failed/spare devices. This normally happens automatically,
2768 * but not when the metadata is externally managed.
2770 if (rdev
->raid_disk
== -1)
2772 /* personality does all needed checks */
2773 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2775 clear_bit(Blocked
, &rdev
->flags
);
2776 remove_and_add_spares(rdev
->mddev
, rdev
);
2777 if (rdev
->raid_disk
>= 0)
2779 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2780 md_wakeup_thread(rdev
->mddev
->thread
);
2781 } else if (rdev
->mddev
->pers
) {
2782 /* Activating a spare .. or possibly reactivating
2783 * if we ever get bitmaps working here.
2787 if (rdev
->raid_disk
!= -1)
2790 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2793 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2796 if (slot
>= rdev
->mddev
->raid_disks
&&
2797 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2800 rdev
->raid_disk
= slot
;
2801 if (test_bit(In_sync
, &rdev
->flags
))
2802 rdev
->saved_raid_disk
= slot
;
2804 rdev
->saved_raid_disk
= -1;
2805 clear_bit(In_sync
, &rdev
->flags
);
2806 clear_bit(Bitmap_sync
, &rdev
->flags
);
2807 err
= rdev
->mddev
->pers
->
2808 hot_add_disk(rdev
->mddev
, rdev
);
2810 rdev
->raid_disk
= -1;
2813 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2814 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2815 /* failure here is OK */;
2816 /* don't wakeup anyone, leave that to userspace. */
2818 if (slot
>= rdev
->mddev
->raid_disks
&&
2819 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2821 rdev
->raid_disk
= slot
;
2822 /* assume it is working */
2823 clear_bit(Faulty
, &rdev
->flags
);
2824 clear_bit(WriteMostly
, &rdev
->flags
);
2825 set_bit(In_sync
, &rdev
->flags
);
2826 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2831 static struct rdev_sysfs_entry rdev_slot
=
2832 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2835 offset_show(struct md_rdev
*rdev
, char *page
)
2837 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2841 offset_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2843 unsigned long long offset
;
2844 if (kstrtoull(buf
, 10, &offset
) < 0)
2846 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2848 if (rdev
->sectors
&& rdev
->mddev
->external
)
2849 /* Must set offset before size, so overlap checks
2852 rdev
->data_offset
= offset
;
2853 rdev
->new_data_offset
= offset
;
2857 static struct rdev_sysfs_entry rdev_offset
=
2858 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2860 static ssize_t
new_offset_show(struct md_rdev
*rdev
, char *page
)
2862 return sprintf(page
, "%llu\n",
2863 (unsigned long long)rdev
->new_data_offset
);
2866 static ssize_t
new_offset_store(struct md_rdev
*rdev
,
2867 const char *buf
, size_t len
)
2869 unsigned long long new_offset
;
2870 struct mddev
*mddev
= rdev
->mddev
;
2872 if (kstrtoull(buf
, 10, &new_offset
) < 0)
2875 if (mddev
->sync_thread
||
2876 test_bit(MD_RECOVERY_RUNNING
,&mddev
->recovery
))
2878 if (new_offset
== rdev
->data_offset
)
2879 /* reset is always permitted */
2881 else if (new_offset
> rdev
->data_offset
) {
2882 /* must not push array size beyond rdev_sectors */
2883 if (new_offset
- rdev
->data_offset
2884 + mddev
->dev_sectors
> rdev
->sectors
)
2887 /* Metadata worries about other space details. */
2889 /* decreasing the offset is inconsistent with a backwards
2892 if (new_offset
< rdev
->data_offset
&&
2893 mddev
->reshape_backwards
)
2895 /* Increasing offset is inconsistent with forwards
2896 * reshape. reshape_direction should be set to
2897 * 'backwards' first.
2899 if (new_offset
> rdev
->data_offset
&&
2900 !mddev
->reshape_backwards
)
2903 if (mddev
->pers
&& mddev
->persistent
&&
2904 !super_types
[mddev
->major_version
]
2905 .allow_new_offset(rdev
, new_offset
))
2907 rdev
->new_data_offset
= new_offset
;
2908 if (new_offset
> rdev
->data_offset
)
2909 mddev
->reshape_backwards
= 1;
2910 else if (new_offset
< rdev
->data_offset
)
2911 mddev
->reshape_backwards
= 0;
2915 static struct rdev_sysfs_entry rdev_new_offset
=
2916 __ATTR(new_offset
, S_IRUGO
|S_IWUSR
, new_offset_show
, new_offset_store
);
2919 rdev_size_show(struct md_rdev
*rdev
, char *page
)
2921 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2924 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2926 /* check if two start/length pairs overlap */
2934 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2936 unsigned long long blocks
;
2939 if (kstrtoull(buf
, 10, &blocks
) < 0)
2942 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2943 return -EINVAL
; /* sector conversion overflow */
2946 if (new != blocks
* 2)
2947 return -EINVAL
; /* unsigned long long to sector_t overflow */
2954 rdev_size_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
2956 struct mddev
*my_mddev
= rdev
->mddev
;
2957 sector_t oldsectors
= rdev
->sectors
;
2960 if (test_bit(Journal
, &rdev
->flags
))
2962 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2964 if (rdev
->data_offset
!= rdev
->new_data_offset
)
2965 return -EINVAL
; /* too confusing */
2966 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2967 if (my_mddev
->persistent
) {
2968 sectors
= super_types
[my_mddev
->major_version
].
2969 rdev_size_change(rdev
, sectors
);
2972 } else if (!sectors
)
2973 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2975 if (!my_mddev
->pers
->resize
)
2976 /* Cannot change size for RAID0 or Linear etc */
2979 if (sectors
< my_mddev
->dev_sectors
)
2980 return -EINVAL
; /* component must fit device */
2982 rdev
->sectors
= sectors
;
2983 if (sectors
> oldsectors
&& my_mddev
->external
) {
2984 /* Need to check that all other rdevs with the same
2985 * ->bdev do not overlap. 'rcu' is sufficient to walk
2986 * the rdev lists safely.
2987 * This check does not provide a hard guarantee, it
2988 * just helps avoid dangerous mistakes.
2990 struct mddev
*mddev
;
2992 struct list_head
*tmp
;
2995 for_each_mddev(mddev
, tmp
) {
2996 struct md_rdev
*rdev2
;
2998 rdev_for_each(rdev2
, mddev
)
2999 if (rdev
->bdev
== rdev2
->bdev
&&
3001 overlaps(rdev
->data_offset
, rdev
->sectors
,
3014 /* Someone else could have slipped in a size
3015 * change here, but doing so is just silly.
3016 * We put oldsectors back because we *know* it is
3017 * safe, and trust userspace not to race with
3020 rdev
->sectors
= oldsectors
;
3027 static struct rdev_sysfs_entry rdev_size
=
3028 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
3030 static ssize_t
recovery_start_show(struct md_rdev
*rdev
, char *page
)
3032 unsigned long long recovery_start
= rdev
->recovery_offset
;
3034 if (test_bit(In_sync
, &rdev
->flags
) ||
3035 recovery_start
== MaxSector
)
3036 return sprintf(page
, "none\n");
3038 return sprintf(page
, "%llu\n", recovery_start
);
3041 static ssize_t
recovery_start_store(struct md_rdev
*rdev
, const char *buf
, size_t len
)
3043 unsigned long long recovery_start
;
3045 if (cmd_match(buf
, "none"))
3046 recovery_start
= MaxSector
;
3047 else if (kstrtoull(buf
, 10, &recovery_start
))
3050 if (rdev
->mddev
->pers
&&
3051 rdev
->raid_disk
>= 0)
3054 rdev
->recovery_offset
= recovery_start
;
3055 if (recovery_start
== MaxSector
)
3056 set_bit(In_sync
, &rdev
->flags
);
3058 clear_bit(In_sync
, &rdev
->flags
);
3062 static struct rdev_sysfs_entry rdev_recovery_start
=
3063 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
3065 /* sysfs access to bad-blocks list.
3066 * We present two files.
3067 * 'bad-blocks' lists sector numbers and lengths of ranges that
3068 * are recorded as bad. The list is truncated to fit within
3069 * the one-page limit of sysfs.
3070 * Writing "sector length" to this file adds an acknowledged
3072 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3073 * been acknowledged. Writing to this file adds bad blocks
3074 * without acknowledging them. This is largely for testing.
3076 static ssize_t
bb_show(struct md_rdev
*rdev
, char *page
)
3078 return badblocks_show(&rdev
->badblocks
, page
, 0);
3080 static ssize_t
bb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3082 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
3083 /* Maybe that ack was all we needed */
3084 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
3085 wake_up(&rdev
->blocked_wait
);
3088 static struct rdev_sysfs_entry rdev_bad_blocks
=
3089 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
3091 static ssize_t
ubb_show(struct md_rdev
*rdev
, char *page
)
3093 return badblocks_show(&rdev
->badblocks
, page
, 1);
3095 static ssize_t
ubb_store(struct md_rdev
*rdev
, const char *page
, size_t len
)
3097 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
3099 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
3100 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
3102 static struct attribute
*rdev_default_attrs
[] = {
3107 &rdev_new_offset
.attr
,
3109 &rdev_recovery_start
.attr
,
3110 &rdev_bad_blocks
.attr
,
3111 &rdev_unack_bad_blocks
.attr
,
3115 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
3117 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3118 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3124 return entry
->show(rdev
, page
);
3128 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
3129 const char *page
, size_t length
)
3131 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
3132 struct md_rdev
*rdev
= container_of(kobj
, struct md_rdev
, kobj
);
3134 struct mddev
*mddev
= rdev
->mddev
;
3138 if (!capable(CAP_SYS_ADMIN
))
3140 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
3142 if (rdev
->mddev
== NULL
)
3145 rv
= entry
->store(rdev
, page
, length
);
3146 mddev_unlock(mddev
);
3151 static void rdev_free(struct kobject
*ko
)
3153 struct md_rdev
*rdev
= container_of(ko
, struct md_rdev
, kobj
);
3156 static const struct sysfs_ops rdev_sysfs_ops
= {
3157 .show
= rdev_attr_show
,
3158 .store
= rdev_attr_store
,
3160 static struct kobj_type rdev_ktype
= {
3161 .release
= rdev_free
,
3162 .sysfs_ops
= &rdev_sysfs_ops
,
3163 .default_attrs
= rdev_default_attrs
,
3166 int md_rdev_init(struct md_rdev
*rdev
)
3169 rdev
->saved_raid_disk
= -1;
3170 rdev
->raid_disk
= -1;
3172 rdev
->data_offset
= 0;
3173 rdev
->new_data_offset
= 0;
3174 rdev
->sb_events
= 0;
3175 rdev
->last_read_error
.tv_sec
= 0;
3176 rdev
->last_read_error
.tv_nsec
= 0;
3177 rdev
->sb_loaded
= 0;
3178 rdev
->bb_page
= NULL
;
3179 atomic_set(&rdev
->nr_pending
, 0);
3180 atomic_set(&rdev
->read_errors
, 0);
3181 atomic_set(&rdev
->corrected_errors
, 0);
3183 INIT_LIST_HEAD(&rdev
->same_set
);
3184 init_waitqueue_head(&rdev
->blocked_wait
);
3186 /* Add space to store bad block list.
3187 * This reserves the space even on arrays where it cannot
3188 * be used - I wonder if that matters
3190 return badblocks_init(&rdev
->badblocks
, 0);
3192 EXPORT_SYMBOL_GPL(md_rdev_init
);
3194 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3196 * mark the device faulty if:
3198 * - the device is nonexistent (zero size)
3199 * - the device has no valid superblock
3201 * a faulty rdev _never_ has rdev->sb set.
3203 static struct md_rdev
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3205 char b
[BDEVNAME_SIZE
];
3207 struct md_rdev
*rdev
;
3210 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3212 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3213 return ERR_PTR(-ENOMEM
);
3216 err
= md_rdev_init(rdev
);
3219 err
= alloc_disk_sb(rdev
);
3223 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3227 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3229 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3232 "md: %s has zero or unknown size, marking faulty!\n",
3233 bdevname(rdev
->bdev
,b
));
3238 if (super_format
>= 0) {
3239 err
= super_types
[super_format
].
3240 load_super(rdev
, NULL
, super_minor
);
3241 if (err
== -EINVAL
) {
3243 "md: %s does not have a valid v%d.%d "
3244 "superblock, not importing!\n",
3245 bdevname(rdev
->bdev
,b
),
3246 super_format
, super_minor
);
3251 "md: could not read %s's sb, not importing!\n",
3252 bdevname(rdev
->bdev
,b
));
3262 md_rdev_clear(rdev
);
3264 return ERR_PTR(err
);
3268 * Check a full RAID array for plausibility
3271 static void analyze_sbs(struct mddev
*mddev
)
3274 struct md_rdev
*rdev
, *freshest
, *tmp
;
3275 char b
[BDEVNAME_SIZE
];
3278 rdev_for_each_safe(rdev
, tmp
, mddev
)
3279 switch (super_types
[mddev
->major_version
].
3280 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3288 "md: fatal superblock inconsistency in %s"
3289 " -- removing from array\n",
3290 bdevname(rdev
->bdev
,b
));
3291 md_kick_rdev_from_array(rdev
);
3294 super_types
[mddev
->major_version
].
3295 validate_super(mddev
, freshest
);
3298 rdev_for_each_safe(rdev
, tmp
, mddev
) {
3299 if (mddev
->max_disks
&&
3300 (rdev
->desc_nr
>= mddev
->max_disks
||
3301 i
> mddev
->max_disks
)) {
3303 "md: %s: %s: only %d devices permitted\n",
3304 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3306 md_kick_rdev_from_array(rdev
);
3309 if (rdev
!= freshest
) {
3310 if (super_types
[mddev
->major_version
].
3311 validate_super(mddev
, rdev
)) {
3312 printk(KERN_WARNING
"md: kicking non-fresh %s"
3314 bdevname(rdev
->bdev
,b
));
3315 md_kick_rdev_from_array(rdev
);
3319 if (mddev
->level
== LEVEL_MULTIPATH
) {
3320 rdev
->desc_nr
= i
++;
3321 rdev
->raid_disk
= rdev
->desc_nr
;
3322 set_bit(In_sync
, &rdev
->flags
);
3323 } else if (rdev
->raid_disk
>=
3324 (mddev
->raid_disks
- min(0, mddev
->delta_disks
)) &&
3325 !test_bit(Journal
, &rdev
->flags
)) {
3326 rdev
->raid_disk
= -1;
3327 clear_bit(In_sync
, &rdev
->flags
);
3332 /* Read a fixed-point number.
3333 * Numbers in sysfs attributes should be in "standard" units where
3334 * possible, so time should be in seconds.
3335 * However we internally use a a much smaller unit such as
3336 * milliseconds or jiffies.
3337 * This function takes a decimal number with a possible fractional
3338 * component, and produces an integer which is the result of
3339 * multiplying that number by 10^'scale'.
3340 * all without any floating-point arithmetic.
3342 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3344 unsigned long result
= 0;
3346 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3349 else if (decimals
< scale
) {
3352 result
= result
* 10 + value
;
3364 while (decimals
< scale
) {
3373 safe_delay_show(struct mddev
*mddev
, char *page
)
3375 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3376 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3379 safe_delay_store(struct mddev
*mddev
, const char *cbuf
, size_t len
)
3383 if (mddev_is_clustered(mddev
)) {
3384 pr_info("md: Safemode is disabled for clustered mode\n");
3388 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3391 mddev
->safemode_delay
= 0;
3393 unsigned long old_delay
= mddev
->safemode_delay
;
3394 unsigned long new_delay
= (msec
*HZ
)/1000;
3398 mddev
->safemode_delay
= new_delay
;
3399 if (new_delay
< old_delay
|| old_delay
== 0)
3400 mod_timer(&mddev
->safemode_timer
, jiffies
+1);
3404 static struct md_sysfs_entry md_safe_delay
=
3405 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3408 level_show(struct mddev
*mddev
, char *page
)
3410 struct md_personality
*p
;
3412 spin_lock(&mddev
->lock
);
3415 ret
= sprintf(page
, "%s\n", p
->name
);
3416 else if (mddev
->clevel
[0])
3417 ret
= sprintf(page
, "%s\n", mddev
->clevel
);
3418 else if (mddev
->level
!= LEVEL_NONE
)
3419 ret
= sprintf(page
, "%d\n", mddev
->level
);
3422 spin_unlock(&mddev
->lock
);
3427 level_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3432 struct md_personality
*pers
, *oldpers
;
3434 void *priv
, *oldpriv
;
3435 struct md_rdev
*rdev
;
3437 if (slen
== 0 || slen
>= sizeof(clevel
))
3440 rv
= mddev_lock(mddev
);
3444 if (mddev
->pers
== NULL
) {
3445 strncpy(mddev
->clevel
, buf
, slen
);
3446 if (mddev
->clevel
[slen
-1] == '\n')
3448 mddev
->clevel
[slen
] = 0;
3449 mddev
->level
= LEVEL_NONE
;
3457 /* request to change the personality. Need to ensure:
3458 * - array is not engaged in resync/recovery/reshape
3459 * - old personality can be suspended
3460 * - new personality will access other array.
3464 if (mddev
->sync_thread
||
3465 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3466 mddev
->reshape_position
!= MaxSector
||
3467 mddev
->sysfs_active
)
3471 if (!mddev
->pers
->quiesce
) {
3472 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3473 mdname(mddev
), mddev
->pers
->name
);
3477 /* Now find the new personality */
3478 strncpy(clevel
, buf
, slen
);
3479 if (clevel
[slen
-1] == '\n')
3482 if (kstrtol(clevel
, 10, &level
))
3485 if (request_module("md-%s", clevel
) != 0)
3486 request_module("md-level-%s", clevel
);
3487 spin_lock(&pers_lock
);
3488 pers
= find_pers(level
, clevel
);
3489 if (!pers
|| !try_module_get(pers
->owner
)) {
3490 spin_unlock(&pers_lock
);
3491 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3495 spin_unlock(&pers_lock
);
3497 if (pers
== mddev
->pers
) {
3498 /* Nothing to do! */
3499 module_put(pers
->owner
);
3503 if (!pers
->takeover
) {
3504 module_put(pers
->owner
);
3505 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3506 mdname(mddev
), clevel
);
3511 rdev_for_each(rdev
, mddev
)
3512 rdev
->new_raid_disk
= rdev
->raid_disk
;
3514 /* ->takeover must set new_* and/or delta_disks
3515 * if it succeeds, and may set them when it fails.
3517 priv
= pers
->takeover(mddev
);
3519 mddev
->new_level
= mddev
->level
;
3520 mddev
->new_layout
= mddev
->layout
;
3521 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3522 mddev
->raid_disks
-= mddev
->delta_disks
;
3523 mddev
->delta_disks
= 0;
3524 mddev
->reshape_backwards
= 0;
3525 module_put(pers
->owner
);
3526 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3527 mdname(mddev
), clevel
);
3532 /* Looks like we have a winner */
3533 mddev_suspend(mddev
);
3534 mddev_detach(mddev
);
3536 spin_lock(&mddev
->lock
);
3537 oldpers
= mddev
->pers
;
3538 oldpriv
= mddev
->private;
3540 mddev
->private = priv
;
3541 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3542 mddev
->level
= mddev
->new_level
;
3543 mddev
->layout
= mddev
->new_layout
;
3544 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3545 mddev
->delta_disks
= 0;
3546 mddev
->reshape_backwards
= 0;
3547 mddev
->degraded
= 0;
3548 spin_unlock(&mddev
->lock
);
3550 if (oldpers
->sync_request
== NULL
&&
3552 /* We are converting from a no-redundancy array
3553 * to a redundancy array and metadata is managed
3554 * externally so we need to be sure that writes
3555 * won't block due to a need to transition
3557 * until external management is started.
3560 mddev
->safemode_delay
= 0;
3561 mddev
->safemode
= 0;
3564 oldpers
->free(mddev
, oldpriv
);
3566 if (oldpers
->sync_request
== NULL
&&
3567 pers
->sync_request
!= NULL
) {
3568 /* need to add the md_redundancy_group */
3569 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3571 "md: cannot register extra attributes for %s\n",
3573 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, "sync_action");
3575 if (oldpers
->sync_request
!= NULL
&&
3576 pers
->sync_request
== NULL
) {
3577 /* need to remove the md_redundancy_group */
3578 if (mddev
->to_remove
== NULL
)
3579 mddev
->to_remove
= &md_redundancy_group
;
3582 rdev_for_each(rdev
, mddev
) {
3583 if (rdev
->raid_disk
< 0)
3585 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3586 rdev
->new_raid_disk
= -1;
3587 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3589 sysfs_unlink_rdev(mddev
, rdev
);
3591 rdev_for_each(rdev
, mddev
) {
3592 if (rdev
->raid_disk
< 0)
3594 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3596 rdev
->raid_disk
= rdev
->new_raid_disk
;
3597 if (rdev
->raid_disk
< 0)
3598 clear_bit(In_sync
, &rdev
->flags
);
3600 if (sysfs_link_rdev(mddev
, rdev
))
3601 printk(KERN_WARNING
"md: cannot register rd%d"
3602 " for %s after level change\n",
3603 rdev
->raid_disk
, mdname(mddev
));
3607 if (pers
->sync_request
== NULL
) {
3608 /* this is now an array without redundancy, so
3609 * it must always be in_sync
3612 del_timer_sync(&mddev
->safemode_timer
);
3614 blk_set_stacking_limits(&mddev
->queue
->limits
);
3616 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3617 mddev_resume(mddev
);
3619 md_update_sb(mddev
, 1);
3620 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3621 md_new_event(mddev
);
3624 mddev_unlock(mddev
);
3628 static struct md_sysfs_entry md_level
=
3629 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3632 layout_show(struct mddev
*mddev
, char *page
)
3634 /* just a number, not meaningful for all levels */
3635 if (mddev
->reshape_position
!= MaxSector
&&
3636 mddev
->layout
!= mddev
->new_layout
)
3637 return sprintf(page
, "%d (%d)\n",
3638 mddev
->new_layout
, mddev
->layout
);
3639 return sprintf(page
, "%d\n", mddev
->layout
);
3643 layout_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3648 err
= kstrtouint(buf
, 10, &n
);
3651 err
= mddev_lock(mddev
);
3656 if (mddev
->pers
->check_reshape
== NULL
)
3661 mddev
->new_layout
= n
;
3662 err
= mddev
->pers
->check_reshape(mddev
);
3664 mddev
->new_layout
= mddev
->layout
;
3667 mddev
->new_layout
= n
;
3668 if (mddev
->reshape_position
== MaxSector
)
3671 mddev_unlock(mddev
);
3674 static struct md_sysfs_entry md_layout
=
3675 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3678 raid_disks_show(struct mddev
*mddev
, char *page
)
3680 if (mddev
->raid_disks
== 0)
3682 if (mddev
->reshape_position
!= MaxSector
&&
3683 mddev
->delta_disks
!= 0)
3684 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3685 mddev
->raid_disks
- mddev
->delta_disks
);
3686 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3689 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
);
3692 raid_disks_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3697 err
= kstrtouint(buf
, 10, &n
);
3701 err
= mddev_lock(mddev
);
3705 err
= update_raid_disks(mddev
, n
);
3706 else if (mddev
->reshape_position
!= MaxSector
) {
3707 struct md_rdev
*rdev
;
3708 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3711 rdev_for_each(rdev
, mddev
) {
3713 rdev
->data_offset
< rdev
->new_data_offset
)
3716 rdev
->data_offset
> rdev
->new_data_offset
)
3720 mddev
->delta_disks
= n
- olddisks
;
3721 mddev
->raid_disks
= n
;
3722 mddev
->reshape_backwards
= (mddev
->delta_disks
< 0);
3724 mddev
->raid_disks
= n
;
3726 mddev_unlock(mddev
);
3727 return err
? err
: len
;
3729 static struct md_sysfs_entry md_raid_disks
=
3730 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3733 chunk_size_show(struct mddev
*mddev
, char *page
)
3735 if (mddev
->reshape_position
!= MaxSector
&&
3736 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3737 return sprintf(page
, "%d (%d)\n",
3738 mddev
->new_chunk_sectors
<< 9,
3739 mddev
->chunk_sectors
<< 9);
3740 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3744 chunk_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3749 err
= kstrtoul(buf
, 10, &n
);
3753 err
= mddev_lock(mddev
);
3757 if (mddev
->pers
->check_reshape
== NULL
)
3762 mddev
->new_chunk_sectors
= n
>> 9;
3763 err
= mddev
->pers
->check_reshape(mddev
);
3765 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3768 mddev
->new_chunk_sectors
= n
>> 9;
3769 if (mddev
->reshape_position
== MaxSector
)
3770 mddev
->chunk_sectors
= n
>> 9;
3772 mddev_unlock(mddev
);
3775 static struct md_sysfs_entry md_chunk_size
=
3776 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3779 resync_start_show(struct mddev
*mddev
, char *page
)
3781 if (mddev
->recovery_cp
== MaxSector
)
3782 return sprintf(page
, "none\n");
3783 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3787 resync_start_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3789 unsigned long long n
;
3792 if (cmd_match(buf
, "none"))
3795 err
= kstrtoull(buf
, 10, &n
);
3798 if (n
!= (sector_t
)n
)
3802 err
= mddev_lock(mddev
);
3805 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3809 mddev
->recovery_cp
= n
;
3811 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3813 mddev_unlock(mddev
);
3816 static struct md_sysfs_entry md_resync_start
=
3817 __ATTR_PREALLOC(resync_start
, S_IRUGO
|S_IWUSR
,
3818 resync_start_show
, resync_start_store
);
3821 * The array state can be:
3824 * No devices, no size, no level
3825 * Equivalent to STOP_ARRAY ioctl
3827 * May have some settings, but array is not active
3828 * all IO results in error
3829 * When written, doesn't tear down array, but just stops it
3830 * suspended (not supported yet)
3831 * All IO requests will block. The array can be reconfigured.
3832 * Writing this, if accepted, will block until array is quiescent
3834 * no resync can happen. no superblocks get written.
3835 * write requests fail
3837 * like readonly, but behaves like 'clean' on a write request.
3839 * clean - no pending writes, but otherwise active.
3840 * When written to inactive array, starts without resync
3841 * If a write request arrives then
3842 * if metadata is known, mark 'dirty' and switch to 'active'.
3843 * if not known, block and switch to write-pending
3844 * If written to an active array that has pending writes, then fails.
3846 * fully active: IO and resync can be happening.
3847 * When written to inactive array, starts with resync
3850 * clean, but writes are blocked waiting for 'active' to be written.
3853 * like active, but no writes have been seen for a while (100msec).
3856 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3857 write_pending
, active_idle
, bad_word
};
3858 static char *array_states
[] = {
3859 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3860 "write-pending", "active-idle", NULL
};
3862 static int match_word(const char *word
, char **list
)
3865 for (n
=0; list
[n
]; n
++)
3866 if (cmd_match(word
, list
[n
]))
3872 array_state_show(struct mddev
*mddev
, char *page
)
3874 enum array_state st
= inactive
;
3887 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3889 else if (mddev
->safemode
)
3895 if (list_empty(&mddev
->disks
) &&
3896 mddev
->raid_disks
== 0 &&
3897 mddev
->dev_sectors
== 0)
3902 return sprintf(page
, "%s\n", array_states
[st
]);
3905 static int do_md_stop(struct mddev
*mddev
, int ro
, struct block_device
*bdev
);
3906 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
);
3907 static int do_md_run(struct mddev
*mddev
);
3908 static int restart_array(struct mddev
*mddev
);
3911 array_state_store(struct mddev
*mddev
, const char *buf
, size_t len
)
3914 enum array_state st
= match_word(buf
, array_states
);
3916 if (mddev
->pers
&& (st
== active
|| st
== clean
) && mddev
->ro
!= 1) {
3917 /* don't take reconfig_mutex when toggling between
3920 spin_lock(&mddev
->lock
);
3922 restart_array(mddev
);
3923 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3924 wake_up(&mddev
->sb_wait
);
3926 } else /* st == clean */ {
3927 restart_array(mddev
);
3928 if (atomic_read(&mddev
->writes_pending
) == 0) {
3929 if (mddev
->in_sync
== 0) {
3931 if (mddev
->safemode
== 1)
3932 mddev
->safemode
= 0;
3933 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3939 spin_unlock(&mddev
->lock
);
3942 err
= mddev_lock(mddev
);
3950 /* stopping an active array */
3951 err
= do_md_stop(mddev
, 0, NULL
);
3954 /* stopping an active array */
3956 err
= do_md_stop(mddev
, 2, NULL
);
3958 err
= 0; /* already inactive */
3961 break; /* not supported yet */
3964 err
= md_set_readonly(mddev
, NULL
);
3967 set_disk_ro(mddev
->gendisk
, 1);
3968 err
= do_md_run(mddev
);
3974 err
= md_set_readonly(mddev
, NULL
);
3975 else if (mddev
->ro
== 1)
3976 err
= restart_array(mddev
);
3979 set_disk_ro(mddev
->gendisk
, 0);
3983 err
= do_md_run(mddev
);
3988 err
= restart_array(mddev
);
3991 spin_lock(&mddev
->lock
);
3992 if (atomic_read(&mddev
->writes_pending
) == 0) {
3993 if (mddev
->in_sync
== 0) {
3995 if (mddev
->safemode
== 1)
3996 mddev
->safemode
= 0;
3997 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
4002 spin_unlock(&mddev
->lock
);
4008 err
= restart_array(mddev
);
4011 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
4012 wake_up(&mddev
->sb_wait
);
4016 set_disk_ro(mddev
->gendisk
, 0);
4017 err
= do_md_run(mddev
);
4022 /* these cannot be set */
4027 if (mddev
->hold_active
== UNTIL_IOCTL
)
4028 mddev
->hold_active
= 0;
4029 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4031 mddev_unlock(mddev
);
4034 static struct md_sysfs_entry md_array_state
=
4035 __ATTR_PREALLOC(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
4038 max_corrected_read_errors_show(struct mddev
*mddev
, char *page
) {
4039 return sprintf(page
, "%d\n",
4040 atomic_read(&mddev
->max_corr_read_errors
));
4044 max_corrected_read_errors_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4049 rv
= kstrtouint(buf
, 10, &n
);
4052 atomic_set(&mddev
->max_corr_read_errors
, n
);
4056 static struct md_sysfs_entry max_corr_read_errors
=
4057 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
4058 max_corrected_read_errors_store
);
4061 null_show(struct mddev
*mddev
, char *page
)
4067 new_dev_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4069 /* buf must be %d:%d\n? giving major and minor numbers */
4070 /* The new device is added to the array.
4071 * If the array has a persistent superblock, we read the
4072 * superblock to initialise info and check validity.
4073 * Otherwise, only checking done is that in bind_rdev_to_array,
4074 * which mainly checks size.
4077 int major
= simple_strtoul(buf
, &e
, 10);
4080 struct md_rdev
*rdev
;
4083 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
4085 minor
= simple_strtoul(e
+1, &e
, 10);
4086 if (*e
&& *e
!= '\n')
4088 dev
= MKDEV(major
, minor
);
4089 if (major
!= MAJOR(dev
) ||
4090 minor
!= MINOR(dev
))
4093 flush_workqueue(md_misc_wq
);
4095 err
= mddev_lock(mddev
);
4098 if (mddev
->persistent
) {
4099 rdev
= md_import_device(dev
, mddev
->major_version
,
4100 mddev
->minor_version
);
4101 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
4102 struct md_rdev
*rdev0
4103 = list_entry(mddev
->disks
.next
,
4104 struct md_rdev
, same_set
);
4105 err
= super_types
[mddev
->major_version
]
4106 .load_super(rdev
, rdev0
, mddev
->minor_version
);
4110 } else if (mddev
->external
)
4111 rdev
= md_import_device(dev
, -2, -1);
4113 rdev
= md_import_device(dev
, -1, -1);
4116 mddev_unlock(mddev
);
4117 return PTR_ERR(rdev
);
4119 err
= bind_rdev_to_array(rdev
, mddev
);
4123 mddev_unlock(mddev
);
4124 return err
? err
: len
;
4127 static struct md_sysfs_entry md_new_device
=
4128 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
4131 bitmap_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4134 unsigned long chunk
, end_chunk
;
4137 err
= mddev_lock(mddev
);
4142 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4144 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
4145 if (buf
== end
) break;
4146 if (*end
== '-') { /* range */
4148 end_chunk
= simple_strtoul(buf
, &end
, 0);
4149 if (buf
== end
) break;
4151 if (*end
&& !isspace(*end
)) break;
4152 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
4153 buf
= skip_spaces(end
);
4155 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
4157 mddev_unlock(mddev
);
4161 static struct md_sysfs_entry md_bitmap
=
4162 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
4165 size_show(struct mddev
*mddev
, char *page
)
4167 return sprintf(page
, "%llu\n",
4168 (unsigned long long)mddev
->dev_sectors
/ 2);
4171 static int update_size(struct mddev
*mddev
, sector_t num_sectors
);
4174 size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4176 /* If array is inactive, we can reduce the component size, but
4177 * not increase it (except from 0).
4178 * If array is active, we can try an on-line resize
4181 int err
= strict_blocks_to_sectors(buf
, §ors
);
4185 err
= mddev_lock(mddev
);
4189 err
= update_size(mddev
, sectors
);
4190 md_update_sb(mddev
, 1);
4192 if (mddev
->dev_sectors
== 0 ||
4193 mddev
->dev_sectors
> sectors
)
4194 mddev
->dev_sectors
= sectors
;
4198 mddev_unlock(mddev
);
4199 return err
? err
: len
;
4202 static struct md_sysfs_entry md_size
=
4203 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
4205 /* Metadata version.
4207 * 'none' for arrays with no metadata (good luck...)
4208 * 'external' for arrays with externally managed metadata,
4209 * or N.M for internally known formats
4212 metadata_show(struct mddev
*mddev
, char *page
)
4214 if (mddev
->persistent
)
4215 return sprintf(page
, "%d.%d\n",
4216 mddev
->major_version
, mddev
->minor_version
);
4217 else if (mddev
->external
)
4218 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
4220 return sprintf(page
, "none\n");
4224 metadata_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4229 /* Changing the details of 'external' metadata is
4230 * always permitted. Otherwise there must be
4231 * no devices attached to the array.
4234 err
= mddev_lock(mddev
);
4238 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
4240 else if (!list_empty(&mddev
->disks
))
4244 if (cmd_match(buf
, "none")) {
4245 mddev
->persistent
= 0;
4246 mddev
->external
= 0;
4247 mddev
->major_version
= 0;
4248 mddev
->minor_version
= 90;
4251 if (strncmp(buf
, "external:", 9) == 0) {
4252 size_t namelen
= len
-9;
4253 if (namelen
>= sizeof(mddev
->metadata_type
))
4254 namelen
= sizeof(mddev
->metadata_type
)-1;
4255 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
4256 mddev
->metadata_type
[namelen
] = 0;
4257 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
4258 mddev
->metadata_type
[--namelen
] = 0;
4259 mddev
->persistent
= 0;
4260 mddev
->external
= 1;
4261 mddev
->major_version
= 0;
4262 mddev
->minor_version
= 90;
4265 major
= simple_strtoul(buf
, &e
, 10);
4267 if (e
==buf
|| *e
!= '.')
4270 minor
= simple_strtoul(buf
, &e
, 10);
4271 if (e
==buf
|| (*e
&& *e
!= '\n') )
4274 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
4276 mddev
->major_version
= major
;
4277 mddev
->minor_version
= minor
;
4278 mddev
->persistent
= 1;
4279 mddev
->external
= 0;
4282 mddev_unlock(mddev
);
4286 static struct md_sysfs_entry md_metadata
=
4287 __ATTR_PREALLOC(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4290 action_show(struct mddev
*mddev
, char *page
)
4292 char *type
= "idle";
4293 unsigned long recovery
= mddev
->recovery
;
4294 if (test_bit(MD_RECOVERY_FROZEN
, &recovery
))
4296 else if (test_bit(MD_RECOVERY_RUNNING
, &recovery
) ||
4297 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &recovery
))) {
4298 if (test_bit(MD_RECOVERY_RESHAPE
, &recovery
))
4300 else if (test_bit(MD_RECOVERY_SYNC
, &recovery
)) {
4301 if (!test_bit(MD_RECOVERY_REQUESTED
, &recovery
))
4303 else if (test_bit(MD_RECOVERY_CHECK
, &recovery
))
4307 } else if (test_bit(MD_RECOVERY_RECOVER
, &recovery
))
4309 else if (mddev
->reshape_position
!= MaxSector
)
4312 return sprintf(page
, "%s\n", type
);
4316 action_store(struct mddev
*mddev
, const char *page
, size_t len
)
4318 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4322 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4323 if (cmd_match(page
, "frozen"))
4324 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4326 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4327 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
4328 mddev_lock(mddev
) == 0) {
4329 flush_workqueue(md_misc_wq
);
4330 if (mddev
->sync_thread
) {
4331 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4332 md_reap_sync_thread(mddev
);
4334 mddev_unlock(mddev
);
4336 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4338 else if (cmd_match(page
, "resync"))
4339 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4340 else if (cmd_match(page
, "recover")) {
4341 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4342 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4343 } else if (cmd_match(page
, "reshape")) {
4345 if (mddev
->pers
->start_reshape
== NULL
)
4347 err
= mddev_lock(mddev
);
4349 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4352 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4353 err
= mddev
->pers
->start_reshape(mddev
);
4355 mddev_unlock(mddev
);
4359 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4361 if (cmd_match(page
, "check"))
4362 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4363 else if (!cmd_match(page
, "repair"))
4365 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4366 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4367 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4369 if (mddev
->ro
== 2) {
4370 /* A write to sync_action is enough to justify
4371 * canceling read-auto mode
4374 md_wakeup_thread(mddev
->sync_thread
);
4376 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4377 md_wakeup_thread(mddev
->thread
);
4378 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4382 static struct md_sysfs_entry md_scan_mode
=
4383 __ATTR_PREALLOC(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4386 last_sync_action_show(struct mddev
*mddev
, char *page
)
4388 return sprintf(page
, "%s\n", mddev
->last_sync_action
);
4391 static struct md_sysfs_entry md_last_scan_mode
= __ATTR_RO(last_sync_action
);
4394 mismatch_cnt_show(struct mddev
*mddev
, char *page
)
4396 return sprintf(page
, "%llu\n",
4397 (unsigned long long)
4398 atomic64_read(&mddev
->resync_mismatches
));
4401 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4404 sync_min_show(struct mddev
*mddev
, char *page
)
4406 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4407 mddev
->sync_speed_min
? "local": "system");
4411 sync_min_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4416 if (strncmp(buf
, "system", 6)==0) {
4419 rv
= kstrtouint(buf
, 10, &min
);
4425 mddev
->sync_speed_min
= min
;
4429 static struct md_sysfs_entry md_sync_min
=
4430 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4433 sync_max_show(struct mddev
*mddev
, char *page
)
4435 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4436 mddev
->sync_speed_max
? "local": "system");
4440 sync_max_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4445 if (strncmp(buf
, "system", 6)==0) {
4448 rv
= kstrtouint(buf
, 10, &max
);
4454 mddev
->sync_speed_max
= max
;
4458 static struct md_sysfs_entry md_sync_max
=
4459 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4462 degraded_show(struct mddev
*mddev
, char *page
)
4464 return sprintf(page
, "%d\n", mddev
->degraded
);
4466 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4469 sync_force_parallel_show(struct mddev
*mddev
, char *page
)
4471 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4475 sync_force_parallel_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4479 if (kstrtol(buf
, 10, &n
))
4482 if (n
!= 0 && n
!= 1)
4485 mddev
->parallel_resync
= n
;
4487 if (mddev
->sync_thread
)
4488 wake_up(&resync_wait
);
4493 /* force parallel resync, even with shared block devices */
4494 static struct md_sysfs_entry md_sync_force_parallel
=
4495 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4496 sync_force_parallel_show
, sync_force_parallel_store
);
4499 sync_speed_show(struct mddev
*mddev
, char *page
)
4501 unsigned long resync
, dt
, db
;
4502 if (mddev
->curr_resync
== 0)
4503 return sprintf(page
, "none\n");
4504 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4505 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4507 db
= resync
- mddev
->resync_mark_cnt
;
4508 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4511 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4514 sync_completed_show(struct mddev
*mddev
, char *page
)
4516 unsigned long long max_sectors
, resync
;
4518 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4519 return sprintf(page
, "none\n");
4521 if (mddev
->curr_resync
== 1 ||
4522 mddev
->curr_resync
== 2)
4523 return sprintf(page
, "delayed\n");
4525 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
4526 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4527 max_sectors
= mddev
->resync_max_sectors
;
4529 max_sectors
= mddev
->dev_sectors
;
4531 resync
= mddev
->curr_resync_completed
;
4532 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4535 static struct md_sysfs_entry md_sync_completed
=
4536 __ATTR_PREALLOC(sync_completed
, S_IRUGO
, sync_completed_show
, NULL
);
4539 min_sync_show(struct mddev
*mddev
, char *page
)
4541 return sprintf(page
, "%llu\n",
4542 (unsigned long long)mddev
->resync_min
);
4545 min_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4547 unsigned long long min
;
4550 if (kstrtoull(buf
, 10, &min
))
4553 spin_lock(&mddev
->lock
);
4555 if (min
> mddev
->resync_max
)
4559 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4562 /* Round down to multiple of 4K for safety */
4563 mddev
->resync_min
= round_down(min
, 8);
4567 spin_unlock(&mddev
->lock
);
4571 static struct md_sysfs_entry md_min_sync
=
4572 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4575 max_sync_show(struct mddev
*mddev
, char *page
)
4577 if (mddev
->resync_max
== MaxSector
)
4578 return sprintf(page
, "max\n");
4580 return sprintf(page
, "%llu\n",
4581 (unsigned long long)mddev
->resync_max
);
4584 max_sync_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4587 spin_lock(&mddev
->lock
);
4588 if (strncmp(buf
, "max", 3) == 0)
4589 mddev
->resync_max
= MaxSector
;
4591 unsigned long long max
;
4595 if (kstrtoull(buf
, 10, &max
))
4597 if (max
< mddev
->resync_min
)
4601 if (max
< mddev
->resync_max
&&
4603 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4606 /* Must be a multiple of chunk_size */
4607 chunk
= mddev
->chunk_sectors
;
4609 sector_t temp
= max
;
4612 if (sector_div(temp
, chunk
))
4615 mddev
->resync_max
= max
;
4617 wake_up(&mddev
->recovery_wait
);
4620 spin_unlock(&mddev
->lock
);
4624 static struct md_sysfs_entry md_max_sync
=
4625 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4628 suspend_lo_show(struct mddev
*mddev
, char *page
)
4630 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4634 suspend_lo_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4636 unsigned long long old
, new;
4639 err
= kstrtoull(buf
, 10, &new);
4642 if (new != (sector_t
)new)
4645 err
= mddev_lock(mddev
);
4649 if (mddev
->pers
== NULL
||
4650 mddev
->pers
->quiesce
== NULL
)
4652 old
= mddev
->suspend_lo
;
4653 mddev
->suspend_lo
= new;
4655 /* Shrinking suspended region */
4656 mddev
->pers
->quiesce(mddev
, 2);
4658 /* Expanding suspended region - need to wait */
4659 mddev
->pers
->quiesce(mddev
, 1);
4660 mddev
->pers
->quiesce(mddev
, 0);
4664 mddev_unlock(mddev
);
4667 static struct md_sysfs_entry md_suspend_lo
=
4668 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4671 suspend_hi_show(struct mddev
*mddev
, char *page
)
4673 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4677 suspend_hi_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4679 unsigned long long old
, new;
4682 err
= kstrtoull(buf
, 10, &new);
4685 if (new != (sector_t
)new)
4688 err
= mddev_lock(mddev
);
4692 if (mddev
->pers
== NULL
||
4693 mddev
->pers
->quiesce
== NULL
)
4695 old
= mddev
->suspend_hi
;
4696 mddev
->suspend_hi
= new;
4698 /* Shrinking suspended region */
4699 mddev
->pers
->quiesce(mddev
, 2);
4701 /* Expanding suspended region - need to wait */
4702 mddev
->pers
->quiesce(mddev
, 1);
4703 mddev
->pers
->quiesce(mddev
, 0);
4707 mddev_unlock(mddev
);
4710 static struct md_sysfs_entry md_suspend_hi
=
4711 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4714 reshape_position_show(struct mddev
*mddev
, char *page
)
4716 if (mddev
->reshape_position
!= MaxSector
)
4717 return sprintf(page
, "%llu\n",
4718 (unsigned long long)mddev
->reshape_position
);
4719 strcpy(page
, "none\n");
4724 reshape_position_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4726 struct md_rdev
*rdev
;
4727 unsigned long long new;
4730 err
= kstrtoull(buf
, 10, &new);
4733 if (new != (sector_t
)new)
4735 err
= mddev_lock(mddev
);
4741 mddev
->reshape_position
= new;
4742 mddev
->delta_disks
= 0;
4743 mddev
->reshape_backwards
= 0;
4744 mddev
->new_level
= mddev
->level
;
4745 mddev
->new_layout
= mddev
->layout
;
4746 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4747 rdev_for_each(rdev
, mddev
)
4748 rdev
->new_data_offset
= rdev
->data_offset
;
4751 mddev_unlock(mddev
);
4755 static struct md_sysfs_entry md_reshape_position
=
4756 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4757 reshape_position_store
);
4760 reshape_direction_show(struct mddev
*mddev
, char *page
)
4762 return sprintf(page
, "%s\n",
4763 mddev
->reshape_backwards
? "backwards" : "forwards");
4767 reshape_direction_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4772 if (cmd_match(buf
, "forwards"))
4774 else if (cmd_match(buf
, "backwards"))
4778 if (mddev
->reshape_backwards
== backwards
)
4781 err
= mddev_lock(mddev
);
4784 /* check if we are allowed to change */
4785 if (mddev
->delta_disks
)
4787 else if (mddev
->persistent
&&
4788 mddev
->major_version
== 0)
4791 mddev
->reshape_backwards
= backwards
;
4792 mddev_unlock(mddev
);
4796 static struct md_sysfs_entry md_reshape_direction
=
4797 __ATTR(reshape_direction
, S_IRUGO
|S_IWUSR
, reshape_direction_show
,
4798 reshape_direction_store
);
4801 array_size_show(struct mddev
*mddev
, char *page
)
4803 if (mddev
->external_size
)
4804 return sprintf(page
, "%llu\n",
4805 (unsigned long long)mddev
->array_sectors
/2);
4807 return sprintf(page
, "default\n");
4811 array_size_store(struct mddev
*mddev
, const char *buf
, size_t len
)
4816 err
= mddev_lock(mddev
);
4820 /* cluster raid doesn't support change array_sectors */
4821 if (mddev_is_clustered(mddev
))
4824 if (strncmp(buf
, "default", 7) == 0) {
4826 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4828 sectors
= mddev
->array_sectors
;
4830 mddev
->external_size
= 0;
4832 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4834 else if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4837 mddev
->external_size
= 1;
4841 mddev
->array_sectors
= sectors
;
4843 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4844 revalidate_disk(mddev
->gendisk
);
4847 mddev_unlock(mddev
);
4851 static struct md_sysfs_entry md_array_size
=
4852 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4855 static struct attribute
*md_default_attrs
[] = {
4858 &md_raid_disks
.attr
,
4859 &md_chunk_size
.attr
,
4861 &md_resync_start
.attr
,
4863 &md_new_device
.attr
,
4864 &md_safe_delay
.attr
,
4865 &md_array_state
.attr
,
4866 &md_reshape_position
.attr
,
4867 &md_reshape_direction
.attr
,
4868 &md_array_size
.attr
,
4869 &max_corr_read_errors
.attr
,
4873 static struct attribute
*md_redundancy_attrs
[] = {
4875 &md_last_scan_mode
.attr
,
4876 &md_mismatches
.attr
,
4879 &md_sync_speed
.attr
,
4880 &md_sync_force_parallel
.attr
,
4881 &md_sync_completed
.attr
,
4884 &md_suspend_lo
.attr
,
4885 &md_suspend_hi
.attr
,
4890 static struct attribute_group md_redundancy_group
= {
4892 .attrs
= md_redundancy_attrs
,
4896 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4898 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4899 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4904 spin_lock(&all_mddevs_lock
);
4905 if (list_empty(&mddev
->all_mddevs
)) {
4906 spin_unlock(&all_mddevs_lock
);
4910 spin_unlock(&all_mddevs_lock
);
4912 rv
= entry
->show(mddev
, page
);
4918 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4919 const char *page
, size_t length
)
4921 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4922 struct mddev
*mddev
= container_of(kobj
, struct mddev
, kobj
);
4927 if (!capable(CAP_SYS_ADMIN
))
4929 spin_lock(&all_mddevs_lock
);
4930 if (list_empty(&mddev
->all_mddevs
)) {
4931 spin_unlock(&all_mddevs_lock
);
4935 spin_unlock(&all_mddevs_lock
);
4936 rv
= entry
->store(mddev
, page
, length
);
4941 static void md_free(struct kobject
*ko
)
4943 struct mddev
*mddev
= container_of(ko
, struct mddev
, kobj
);
4945 if (mddev
->sysfs_state
)
4946 sysfs_put(mddev
->sysfs_state
);
4949 blk_cleanup_queue(mddev
->queue
);
4950 if (mddev
->gendisk
) {
4951 del_gendisk(mddev
->gendisk
);
4952 put_disk(mddev
->gendisk
);
4958 static const struct sysfs_ops md_sysfs_ops
= {
4959 .show
= md_attr_show
,
4960 .store
= md_attr_store
,
4962 static struct kobj_type md_ktype
= {
4964 .sysfs_ops
= &md_sysfs_ops
,
4965 .default_attrs
= md_default_attrs
,
4970 static void mddev_delayed_delete(struct work_struct
*ws
)
4972 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
4974 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4975 kobject_del(&mddev
->kobj
);
4976 kobject_put(&mddev
->kobj
);
4979 static int md_alloc(dev_t dev
, char *name
)
4981 static DEFINE_MUTEX(disks_mutex
);
4982 struct mddev
*mddev
= mddev_find(dev
);
4983 struct gendisk
*disk
;
4992 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4993 shift
= partitioned
? MdpMinorShift
: 0;
4994 unit
= MINOR(mddev
->unit
) >> shift
;
4996 /* wait for any previous instance of this device to be
4997 * completely removed (mddev_delayed_delete).
4999 flush_workqueue(md_misc_wq
);
5001 mutex_lock(&disks_mutex
);
5007 /* Need to ensure that 'name' is not a duplicate.
5009 struct mddev
*mddev2
;
5010 spin_lock(&all_mddevs_lock
);
5012 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
5013 if (mddev2
->gendisk
&&
5014 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
5015 spin_unlock(&all_mddevs_lock
);
5018 spin_unlock(&all_mddevs_lock
);
5022 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
5025 mddev
->queue
->queuedata
= mddev
;
5027 blk_queue_make_request(mddev
->queue
, md_make_request
);
5028 blk_set_stacking_limits(&mddev
->queue
->limits
);
5030 disk
= alloc_disk(1 << shift
);
5032 blk_cleanup_queue(mddev
->queue
);
5033 mddev
->queue
= NULL
;
5036 disk
->major
= MAJOR(mddev
->unit
);
5037 disk
->first_minor
= unit
<< shift
;
5039 strcpy(disk
->disk_name
, name
);
5040 else if (partitioned
)
5041 sprintf(disk
->disk_name
, "md_d%d", unit
);
5043 sprintf(disk
->disk_name
, "md%d", unit
);
5044 disk
->fops
= &md_fops
;
5045 disk
->private_data
= mddev
;
5046 disk
->queue
= mddev
->queue
;
5047 blk_queue_write_cache(mddev
->queue
, true, true);
5048 /* Allow extended partitions. This makes the
5049 * 'mdp' device redundant, but we can't really
5052 disk
->flags
|= GENHD_FL_EXT_DEVT
;
5053 mddev
->gendisk
= disk
;
5054 /* As soon as we call add_disk(), another thread could get
5055 * through to md_open, so make sure it doesn't get too far
5057 mutex_lock(&mddev
->open_mutex
);
5060 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
5061 &disk_to_dev(disk
)->kobj
, "%s", "md");
5063 /* This isn't possible, but as kobject_init_and_add is marked
5064 * __must_check, we must do something with the result
5066 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
5070 if (mddev
->kobj
.sd
&&
5071 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
5072 printk(KERN_DEBUG
"pointless warning\n");
5073 mutex_unlock(&mddev
->open_mutex
);
5075 mutex_unlock(&disks_mutex
);
5076 if (!error
&& mddev
->kobj
.sd
) {
5077 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
5078 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
5084 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
5086 md_alloc(dev
, NULL
);
5090 static int add_named_array(const char *val
, struct kernel_param
*kp
)
5092 /* val must be "md_*" where * is not all digits.
5093 * We allocate an array with a large free minor number, and
5094 * set the name to val. val must not already be an active name.
5096 int len
= strlen(val
);
5097 char buf
[DISK_NAME_LEN
];
5099 while (len
&& val
[len
-1] == '\n')
5101 if (len
>= DISK_NAME_LEN
)
5103 strlcpy(buf
, val
, len
+1);
5104 if (strncmp(buf
, "md_", 3) != 0)
5106 return md_alloc(0, buf
);
5109 static void md_safemode_timeout(unsigned long data
)
5111 struct mddev
*mddev
= (struct mddev
*) data
;
5113 if (!atomic_read(&mddev
->writes_pending
)) {
5114 mddev
->safemode
= 1;
5115 if (mddev
->external
)
5116 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5118 md_wakeup_thread(mddev
->thread
);
5121 static int start_dirty_degraded
;
5123 int md_run(struct mddev
*mddev
)
5126 struct md_rdev
*rdev
;
5127 struct md_personality
*pers
;
5129 if (list_empty(&mddev
->disks
))
5130 /* cannot run an array with no devices.. */
5135 /* Cannot run until previous stop completes properly */
5136 if (mddev
->sysfs_active
)
5140 * Analyze all RAID superblock(s)
5142 if (!mddev
->raid_disks
) {
5143 if (!mddev
->persistent
)
5148 if (mddev
->level
!= LEVEL_NONE
)
5149 request_module("md-level-%d", mddev
->level
);
5150 else if (mddev
->clevel
[0])
5151 request_module("md-%s", mddev
->clevel
);
5154 * Drop all container device buffers, from now on
5155 * the only valid external interface is through the md
5158 rdev_for_each(rdev
, mddev
) {
5159 if (test_bit(Faulty
, &rdev
->flags
))
5161 sync_blockdev(rdev
->bdev
);
5162 invalidate_bdev(rdev
->bdev
);
5164 /* perform some consistency tests on the device.
5165 * We don't want the data to overlap the metadata,
5166 * Internal Bitmap issues have been handled elsewhere.
5168 if (rdev
->meta_bdev
) {
5169 /* Nothing to check */;
5170 } else if (rdev
->data_offset
< rdev
->sb_start
) {
5171 if (mddev
->dev_sectors
&&
5172 rdev
->data_offset
+ mddev
->dev_sectors
5174 printk("md: %s: data overlaps metadata\n",
5179 if (rdev
->sb_start
+ rdev
->sb_size
/512
5180 > rdev
->data_offset
) {
5181 printk("md: %s: metadata overlaps data\n",
5186 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5189 if (mddev
->bio_set
== NULL
)
5190 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
5192 spin_lock(&pers_lock
);
5193 pers
= find_pers(mddev
->level
, mddev
->clevel
);
5194 if (!pers
|| !try_module_get(pers
->owner
)) {
5195 spin_unlock(&pers_lock
);
5196 if (mddev
->level
!= LEVEL_NONE
)
5197 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
5200 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
5204 spin_unlock(&pers_lock
);
5205 if (mddev
->level
!= pers
->level
) {
5206 mddev
->level
= pers
->level
;
5207 mddev
->new_level
= pers
->level
;
5209 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
5211 if (mddev
->reshape_position
!= MaxSector
&&
5212 pers
->start_reshape
== NULL
) {
5213 /* This personality cannot handle reshaping... */
5214 module_put(pers
->owner
);
5218 if (pers
->sync_request
) {
5219 /* Warn if this is a potentially silly
5222 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5223 struct md_rdev
*rdev2
;
5226 rdev_for_each(rdev
, mddev
)
5227 rdev_for_each(rdev2
, mddev
) {
5229 rdev
->bdev
->bd_contains
==
5230 rdev2
->bdev
->bd_contains
) {
5232 "%s: WARNING: %s appears to be"
5233 " on the same physical disk as"
5236 bdevname(rdev
->bdev
,b
),
5237 bdevname(rdev2
->bdev
,b2
));
5244 "True protection against single-disk"
5245 " failure might be compromised.\n");
5248 mddev
->recovery
= 0;
5249 /* may be over-ridden by personality */
5250 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
5252 mddev
->ok_start_degraded
= start_dirty_degraded
;
5254 if (start_readonly
&& mddev
->ro
== 0)
5255 mddev
->ro
= 2; /* read-only, but switch on first write */
5257 err
= pers
->run(mddev
);
5259 printk(KERN_ERR
"md: pers->run() failed ...\n");
5260 else if (pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
5261 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
5262 " but 'external_size' not in effect?\n", __func__
);
5264 "md: invalid array_size %llu > default size %llu\n",
5265 (unsigned long long)mddev
->array_sectors
/ 2,
5266 (unsigned long long)pers
->size(mddev
, 0, 0) / 2);
5269 if (err
== 0 && pers
->sync_request
&&
5270 (mddev
->bitmap_info
.file
|| mddev
->bitmap_info
.offset
)) {
5271 struct bitmap
*bitmap
;
5273 bitmap
= bitmap_create(mddev
, -1);
5274 if (IS_ERR(bitmap
)) {
5275 err
= PTR_ERR(bitmap
);
5276 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
5277 mdname(mddev
), err
);
5279 mddev
->bitmap
= bitmap
;
5283 mddev_detach(mddev
);
5285 pers
->free(mddev
, mddev
->private);
5286 mddev
->private = NULL
;
5287 module_put(pers
->owner
);
5288 bitmap_destroy(mddev
);
5292 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
5293 mddev
->queue
->backing_dev_info
.congested_fn
= md_congested
;
5295 if (pers
->sync_request
) {
5296 if (mddev
->kobj
.sd
&&
5297 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
5299 "md: cannot register extra attributes for %s\n",
5301 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
5302 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
5305 atomic_set(&mddev
->writes_pending
,0);
5306 atomic_set(&mddev
->max_corr_read_errors
,
5307 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
5308 mddev
->safemode
= 0;
5309 if (mddev_is_clustered(mddev
))
5310 mddev
->safemode_delay
= 0;
5312 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
5315 spin_lock(&mddev
->lock
);
5317 spin_unlock(&mddev
->lock
);
5318 rdev_for_each(rdev
, mddev
)
5319 if (rdev
->raid_disk
>= 0)
5320 if (sysfs_link_rdev(mddev
, rdev
))
5321 /* failure here is OK */;
5323 if (mddev
->degraded
&& !mddev
->ro
)
5324 /* This ensures that recovering status is reported immediately
5325 * via sysfs - until a lack of spares is confirmed.
5327 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5328 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5330 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
5331 md_update_sb(mddev
, 0);
5333 md_new_event(mddev
);
5334 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5335 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
5336 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
5339 EXPORT_SYMBOL_GPL(md_run
);
5341 static int do_md_run(struct mddev
*mddev
)
5345 err
= md_run(mddev
);
5348 err
= bitmap_load(mddev
);
5350 bitmap_destroy(mddev
);
5354 if (mddev_is_clustered(mddev
))
5355 md_allow_write(mddev
);
5357 md_wakeup_thread(mddev
->thread
);
5358 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
5360 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
5361 revalidate_disk(mddev
->gendisk
);
5363 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5368 static int restart_array(struct mddev
*mddev
)
5370 struct gendisk
*disk
= mddev
->gendisk
;
5372 /* Complain if it has no devices */
5373 if (list_empty(&mddev
->disks
))
5379 if (test_bit(MD_HAS_JOURNAL
, &mddev
->flags
)) {
5380 struct md_rdev
*rdev
;
5381 bool has_journal
= false;
5384 rdev_for_each_rcu(rdev
, mddev
) {
5385 if (test_bit(Journal
, &rdev
->flags
) &&
5386 !test_bit(Faulty
, &rdev
->flags
)) {
5393 /* Don't restart rw with journal missing/faulty */
5398 mddev
->safemode
= 0;
5400 set_disk_ro(disk
, 0);
5401 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
5403 /* Kick recovery or resync if necessary */
5404 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5405 md_wakeup_thread(mddev
->thread
);
5406 md_wakeup_thread(mddev
->sync_thread
);
5407 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5411 static void md_clean(struct mddev
*mddev
)
5413 mddev
->array_sectors
= 0;
5414 mddev
->external_size
= 0;
5415 mddev
->dev_sectors
= 0;
5416 mddev
->raid_disks
= 0;
5417 mddev
->recovery_cp
= 0;
5418 mddev
->resync_min
= 0;
5419 mddev
->resync_max
= MaxSector
;
5420 mddev
->reshape_position
= MaxSector
;
5421 mddev
->external
= 0;
5422 mddev
->persistent
= 0;
5423 mddev
->level
= LEVEL_NONE
;
5424 mddev
->clevel
[0] = 0;
5427 mddev
->metadata_type
[0] = 0;
5428 mddev
->chunk_sectors
= 0;
5429 mddev
->ctime
= mddev
->utime
= 0;
5431 mddev
->max_disks
= 0;
5433 mddev
->can_decrease_events
= 0;
5434 mddev
->delta_disks
= 0;
5435 mddev
->reshape_backwards
= 0;
5436 mddev
->new_level
= LEVEL_NONE
;
5437 mddev
->new_layout
= 0;
5438 mddev
->new_chunk_sectors
= 0;
5439 mddev
->curr_resync
= 0;
5440 atomic64_set(&mddev
->resync_mismatches
, 0);
5441 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
5442 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
5443 mddev
->recovery
= 0;
5446 mddev
->degraded
= 0;
5447 mddev
->safemode
= 0;
5448 mddev
->private = NULL
;
5449 mddev
->bitmap_info
.offset
= 0;
5450 mddev
->bitmap_info
.default_offset
= 0;
5451 mddev
->bitmap_info
.default_space
= 0;
5452 mddev
->bitmap_info
.chunksize
= 0;
5453 mddev
->bitmap_info
.daemon_sleep
= 0;
5454 mddev
->bitmap_info
.max_write_behind
= 0;
5457 static void __md_stop_writes(struct mddev
*mddev
)
5459 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5460 flush_workqueue(md_misc_wq
);
5461 if (mddev
->sync_thread
) {
5462 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5463 md_reap_sync_thread(mddev
);
5466 del_timer_sync(&mddev
->safemode_timer
);
5468 bitmap_flush(mddev
);
5469 md_super_wait(mddev
);
5471 if (mddev
->ro
== 0 &&
5472 ((!mddev
->in_sync
&& !mddev_is_clustered(mddev
)) ||
5473 (mddev
->flags
& MD_UPDATE_SB_FLAGS
))) {
5474 /* mark array as shutdown cleanly */
5475 if (!mddev_is_clustered(mddev
))
5477 md_update_sb(mddev
, 1);
5481 void md_stop_writes(struct mddev
*mddev
)
5483 mddev_lock_nointr(mddev
);
5484 __md_stop_writes(mddev
);
5485 mddev_unlock(mddev
);
5487 EXPORT_SYMBOL_GPL(md_stop_writes
);
5489 static void mddev_detach(struct mddev
*mddev
)
5491 struct bitmap
*bitmap
= mddev
->bitmap
;
5492 /* wait for behind writes to complete */
5493 if (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
5494 printk(KERN_INFO
"md:%s: behind writes in progress - waiting to stop.\n",
5496 /* need to kick something here to make sure I/O goes? */
5497 wait_event(bitmap
->behind_wait
,
5498 atomic_read(&bitmap
->behind_writes
) == 0);
5500 if (mddev
->pers
&& mddev
->pers
->quiesce
) {
5501 mddev
->pers
->quiesce(mddev
, 1);
5502 mddev
->pers
->quiesce(mddev
, 0);
5504 md_unregister_thread(&mddev
->thread
);
5506 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
5509 static void __md_stop(struct mddev
*mddev
)
5511 struct md_personality
*pers
= mddev
->pers
;
5512 mddev_detach(mddev
);
5513 /* Ensure ->event_work is done */
5514 flush_workqueue(md_misc_wq
);
5515 spin_lock(&mddev
->lock
);
5517 spin_unlock(&mddev
->lock
);
5518 pers
->free(mddev
, mddev
->private);
5519 mddev
->private = NULL
;
5520 if (pers
->sync_request
&& mddev
->to_remove
== NULL
)
5521 mddev
->to_remove
= &md_redundancy_group
;
5522 module_put(pers
->owner
);
5523 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5526 void md_stop(struct mddev
*mddev
)
5528 /* stop the array and free an attached data structures.
5529 * This is called from dm-raid
5532 bitmap_destroy(mddev
);
5534 bioset_free(mddev
->bio_set
);
5537 EXPORT_SYMBOL_GPL(md_stop
);
5539 static int md_set_readonly(struct mddev
*mddev
, struct block_device
*bdev
)
5544 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5546 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5547 md_wakeup_thread(mddev
->thread
);
5549 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5550 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5551 if (mddev
->sync_thread
)
5552 /* Thread might be blocked waiting for metadata update
5553 * which will now never happen */
5554 wake_up_process(mddev
->sync_thread
->tsk
);
5556 if (mddev
->external
&& test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
5558 mddev_unlock(mddev
);
5559 wait_event(resync_wait
, !test_bit(MD_RECOVERY_RUNNING
,
5561 wait_event(mddev
->sb_wait
,
5562 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
5563 mddev_lock_nointr(mddev
);
5565 mutex_lock(&mddev
->open_mutex
);
5566 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5567 mddev
->sync_thread
||
5568 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5569 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5570 printk("md: %s still in use.\n",mdname(mddev
));
5572 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5573 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5574 md_wakeup_thread(mddev
->thread
);
5580 __md_stop_writes(mddev
);
5586 set_disk_ro(mddev
->gendisk
, 1);
5587 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5588 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5589 md_wakeup_thread(mddev
->thread
);
5590 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5594 mutex_unlock(&mddev
->open_mutex
);
5599 * 0 - completely stop and dis-assemble array
5600 * 2 - stop but do not disassemble array
5602 static int do_md_stop(struct mddev
*mddev
, int mode
,
5603 struct block_device
*bdev
)
5605 struct gendisk
*disk
= mddev
->gendisk
;
5606 struct md_rdev
*rdev
;
5609 if (!test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
)) {
5611 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5612 md_wakeup_thread(mddev
->thread
);
5614 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
5615 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5616 if (mddev
->sync_thread
)
5617 /* Thread might be blocked waiting for metadata update
5618 * which will now never happen */
5619 wake_up_process(mddev
->sync_thread
->tsk
);
5621 mddev_unlock(mddev
);
5622 wait_event(resync_wait
, (mddev
->sync_thread
== NULL
&&
5623 !test_bit(MD_RECOVERY_RUNNING
,
5624 &mddev
->recovery
)));
5625 mddev_lock_nointr(mddev
);
5627 mutex_lock(&mddev
->open_mutex
);
5628 if ((mddev
->pers
&& atomic_read(&mddev
->openers
) > !!bdev
) ||
5629 mddev
->sysfs_active
||
5630 mddev
->sync_thread
||
5631 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
5632 (bdev
&& !test_bit(MD_STILL_CLOSED
, &mddev
->flags
))) {
5633 printk("md: %s still in use.\n",mdname(mddev
));
5634 mutex_unlock(&mddev
->open_mutex
);
5636 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5637 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5638 md_wakeup_thread(mddev
->thread
);
5644 set_disk_ro(disk
, 0);
5646 __md_stop_writes(mddev
);
5648 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5650 /* tell userspace to handle 'inactive' */
5651 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5653 rdev_for_each(rdev
, mddev
)
5654 if (rdev
->raid_disk
>= 0)
5655 sysfs_unlink_rdev(mddev
, rdev
);
5657 set_capacity(disk
, 0);
5658 mutex_unlock(&mddev
->open_mutex
);
5660 revalidate_disk(disk
);
5665 mutex_unlock(&mddev
->open_mutex
);
5667 * Free resources if final stop
5670 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5672 bitmap_destroy(mddev
);
5673 if (mddev
->bitmap_info
.file
) {
5674 struct file
*f
= mddev
->bitmap_info
.file
;
5675 spin_lock(&mddev
->lock
);
5676 mddev
->bitmap_info
.file
= NULL
;
5677 spin_unlock(&mddev
->lock
);
5680 mddev
->bitmap_info
.offset
= 0;
5682 export_array(mddev
);
5685 if (mddev
->hold_active
== UNTIL_STOP
)
5686 mddev
->hold_active
= 0;
5688 md_new_event(mddev
);
5689 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5694 static void autorun_array(struct mddev
*mddev
)
5696 struct md_rdev
*rdev
;
5699 if (list_empty(&mddev
->disks
))
5702 printk(KERN_INFO
"md: running: ");
5704 rdev_for_each(rdev
, mddev
) {
5705 char b
[BDEVNAME_SIZE
];
5706 printk("<%s>", bdevname(rdev
->bdev
,b
));
5710 err
= do_md_run(mddev
);
5712 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5713 do_md_stop(mddev
, 0, NULL
);
5718 * lets try to run arrays based on all disks that have arrived
5719 * until now. (those are in pending_raid_disks)
5721 * the method: pick the first pending disk, collect all disks with
5722 * the same UUID, remove all from the pending list and put them into
5723 * the 'same_array' list. Then order this list based on superblock
5724 * update time (freshest comes first), kick out 'old' disks and
5725 * compare superblocks. If everything's fine then run it.
5727 * If "unit" is allocated, then bump its reference count
5729 static void autorun_devices(int part
)
5731 struct md_rdev
*rdev0
, *rdev
, *tmp
;
5732 struct mddev
*mddev
;
5733 char b
[BDEVNAME_SIZE
];
5735 printk(KERN_INFO
"md: autorun ...\n");
5736 while (!list_empty(&pending_raid_disks
)) {
5739 LIST_HEAD(candidates
);
5740 rdev0
= list_entry(pending_raid_disks
.next
,
5741 struct md_rdev
, same_set
);
5743 printk(KERN_INFO
"md: considering %s ...\n",
5744 bdevname(rdev0
->bdev
,b
));
5745 INIT_LIST_HEAD(&candidates
);
5746 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5747 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5748 printk(KERN_INFO
"md: adding %s ...\n",
5749 bdevname(rdev
->bdev
,b
));
5750 list_move(&rdev
->same_set
, &candidates
);
5753 * now we have a set of devices, with all of them having
5754 * mostly sane superblocks. It's time to allocate the
5758 dev
= MKDEV(mdp_major
,
5759 rdev0
->preferred_minor
<< MdpMinorShift
);
5760 unit
= MINOR(dev
) >> MdpMinorShift
;
5762 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5765 if (rdev0
->preferred_minor
!= unit
) {
5766 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5767 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5771 md_probe(dev
, NULL
, NULL
);
5772 mddev
= mddev_find(dev
);
5773 if (!mddev
|| !mddev
->gendisk
) {
5777 "md: cannot allocate memory for md drive.\n");
5780 if (mddev_lock(mddev
))
5781 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5783 else if (mddev
->raid_disks
|| mddev
->major_version
5784 || !list_empty(&mddev
->disks
)) {
5786 "md: %s already running, cannot run %s\n",
5787 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5788 mddev_unlock(mddev
);
5790 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5791 mddev
->persistent
= 1;
5792 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5793 list_del_init(&rdev
->same_set
);
5794 if (bind_rdev_to_array(rdev
, mddev
))
5797 autorun_array(mddev
);
5798 mddev_unlock(mddev
);
5800 /* on success, candidates will be empty, on error
5803 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5804 list_del_init(&rdev
->same_set
);
5809 printk(KERN_INFO
"md: ... autorun DONE.\n");
5811 #endif /* !MODULE */
5813 static int get_version(void __user
*arg
)
5817 ver
.major
= MD_MAJOR_VERSION
;
5818 ver
.minor
= MD_MINOR_VERSION
;
5819 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5821 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5827 static int get_array_info(struct mddev
*mddev
, void __user
*arg
)
5829 mdu_array_info_t info
;
5830 int nr
,working
,insync
,failed
,spare
;
5831 struct md_rdev
*rdev
;
5833 nr
= working
= insync
= failed
= spare
= 0;
5835 rdev_for_each_rcu(rdev
, mddev
) {
5837 if (test_bit(Faulty
, &rdev
->flags
))
5841 if (test_bit(In_sync
, &rdev
->flags
))
5849 info
.major_version
= mddev
->major_version
;
5850 info
.minor_version
= mddev
->minor_version
;
5851 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5852 info
.ctime
= clamp_t(time64_t
, mddev
->ctime
, 0, U32_MAX
);
5853 info
.level
= mddev
->level
;
5854 info
.size
= mddev
->dev_sectors
/ 2;
5855 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5858 info
.raid_disks
= mddev
->raid_disks
;
5859 info
.md_minor
= mddev
->md_minor
;
5860 info
.not_persistent
= !mddev
->persistent
;
5862 info
.utime
= clamp_t(time64_t
, mddev
->utime
, 0, U32_MAX
);
5865 info
.state
= (1<<MD_SB_CLEAN
);
5866 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5867 info
.state
|= (1<<MD_SB_BITMAP_PRESENT
);
5868 if (mddev_is_clustered(mddev
))
5869 info
.state
|= (1<<MD_SB_CLUSTERED
);
5870 info
.active_disks
= insync
;
5871 info
.working_disks
= working
;
5872 info
.failed_disks
= failed
;
5873 info
.spare_disks
= spare
;
5875 info
.layout
= mddev
->layout
;
5876 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5878 if (copy_to_user(arg
, &info
, sizeof(info
)))
5884 static int get_bitmap_file(struct mddev
*mddev
, void __user
* arg
)
5886 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5890 file
= kzalloc(sizeof(*file
), GFP_NOIO
);
5895 spin_lock(&mddev
->lock
);
5896 /* bitmap enabled */
5897 if (mddev
->bitmap_info
.file
) {
5898 ptr
= file_path(mddev
->bitmap_info
.file
, file
->pathname
,
5899 sizeof(file
->pathname
));
5903 memmove(file
->pathname
, ptr
,
5904 sizeof(file
->pathname
)-(ptr
-file
->pathname
));
5906 spin_unlock(&mddev
->lock
);
5909 copy_to_user(arg
, file
, sizeof(*file
)))
5916 static int get_disk_info(struct mddev
*mddev
, void __user
* arg
)
5918 mdu_disk_info_t info
;
5919 struct md_rdev
*rdev
;
5921 if (copy_from_user(&info
, arg
, sizeof(info
)))
5925 rdev
= md_find_rdev_nr_rcu(mddev
, info
.number
);
5927 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5928 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5929 info
.raid_disk
= rdev
->raid_disk
;
5931 if (test_bit(Faulty
, &rdev
->flags
))
5932 info
.state
|= (1<<MD_DISK_FAULTY
);
5933 else if (test_bit(In_sync
, &rdev
->flags
)) {
5934 info
.state
|= (1<<MD_DISK_ACTIVE
);
5935 info
.state
|= (1<<MD_DISK_SYNC
);
5937 if (test_bit(Journal
, &rdev
->flags
))
5938 info
.state
|= (1<<MD_DISK_JOURNAL
);
5939 if (test_bit(WriteMostly
, &rdev
->flags
))
5940 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5942 info
.major
= info
.minor
= 0;
5943 info
.raid_disk
= -1;
5944 info
.state
= (1<<MD_DISK_REMOVED
);
5948 if (copy_to_user(arg
, &info
, sizeof(info
)))
5954 static int add_new_disk(struct mddev
*mddev
, mdu_disk_info_t
*info
)
5956 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5957 struct md_rdev
*rdev
;
5958 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5960 if (mddev_is_clustered(mddev
) &&
5961 !(info
->state
& ((1 << MD_DISK_CLUSTER_ADD
) | (1 << MD_DISK_CANDIDATE
)))) {
5962 pr_err("%s: Cannot add to clustered mddev.\n",
5967 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5970 if (!mddev
->raid_disks
) {
5972 /* expecting a device which has a superblock */
5973 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5976 "md: md_import_device returned %ld\n",
5978 return PTR_ERR(rdev
);
5980 if (!list_empty(&mddev
->disks
)) {
5981 struct md_rdev
*rdev0
5982 = list_entry(mddev
->disks
.next
,
5983 struct md_rdev
, same_set
);
5984 err
= super_types
[mddev
->major_version
]
5985 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5988 "md: %s has different UUID to %s\n",
5989 bdevname(rdev
->bdev
,b
),
5990 bdevname(rdev0
->bdev
,b2
));
5995 err
= bind_rdev_to_array(rdev
, mddev
);
6002 * add_new_disk can be used once the array is assembled
6003 * to add "hot spares". They must already have a superblock
6008 if (!mddev
->pers
->hot_add_disk
) {
6010 "%s: personality does not support diskops!\n",
6014 if (mddev
->persistent
)
6015 rdev
= md_import_device(dev
, mddev
->major_version
,
6016 mddev
->minor_version
);
6018 rdev
= md_import_device(dev
, -1, -1);
6021 "md: md_import_device returned %ld\n",
6023 return PTR_ERR(rdev
);
6025 /* set saved_raid_disk if appropriate */
6026 if (!mddev
->persistent
) {
6027 if (info
->state
& (1<<MD_DISK_SYNC
) &&
6028 info
->raid_disk
< mddev
->raid_disks
) {
6029 rdev
->raid_disk
= info
->raid_disk
;
6030 set_bit(In_sync
, &rdev
->flags
);
6031 clear_bit(Bitmap_sync
, &rdev
->flags
);
6033 rdev
->raid_disk
= -1;
6034 rdev
->saved_raid_disk
= rdev
->raid_disk
;
6036 super_types
[mddev
->major_version
].
6037 validate_super(mddev
, rdev
);
6038 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
6039 rdev
->raid_disk
!= info
->raid_disk
) {
6040 /* This was a hot-add request, but events doesn't
6041 * match, so reject it.
6047 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
6048 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6049 set_bit(WriteMostly
, &rdev
->flags
);
6051 clear_bit(WriteMostly
, &rdev
->flags
);
6053 if (info
->state
& (1<<MD_DISK_JOURNAL
)) {
6054 struct md_rdev
*rdev2
;
6055 bool has_journal
= false;
6057 /* make sure no existing journal disk */
6058 rdev_for_each(rdev2
, mddev
) {
6059 if (test_bit(Journal
, &rdev2
->flags
)) {
6068 set_bit(Journal
, &rdev
->flags
);
6071 * check whether the device shows up in other nodes
6073 if (mddev_is_clustered(mddev
)) {
6074 if (info
->state
& (1 << MD_DISK_CANDIDATE
))
6075 set_bit(Candidate
, &rdev
->flags
);
6076 else if (info
->state
& (1 << MD_DISK_CLUSTER_ADD
)) {
6077 /* --add initiated by this node */
6078 err
= md_cluster_ops
->add_new_disk(mddev
, rdev
);
6086 rdev
->raid_disk
= -1;
6087 err
= bind_rdev_to_array(rdev
, mddev
);
6092 if (mddev_is_clustered(mddev
)) {
6093 if (info
->state
& (1 << MD_DISK_CANDIDATE
))
6094 md_cluster_ops
->new_disk_ack(mddev
, (err
== 0));
6097 md_cluster_ops
->add_new_disk_cancel(mddev
);
6099 err
= add_bound_rdev(rdev
);
6103 err
= add_bound_rdev(rdev
);
6108 /* otherwise, add_new_disk is only allowed
6109 * for major_version==0 superblocks
6111 if (mddev
->major_version
!= 0) {
6112 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
6117 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
6119 rdev
= md_import_device(dev
, -1, 0);
6122 "md: error, md_import_device() returned %ld\n",
6124 return PTR_ERR(rdev
);
6126 rdev
->desc_nr
= info
->number
;
6127 if (info
->raid_disk
< mddev
->raid_disks
)
6128 rdev
->raid_disk
= info
->raid_disk
;
6130 rdev
->raid_disk
= -1;
6132 if (rdev
->raid_disk
< mddev
->raid_disks
)
6133 if (info
->state
& (1<<MD_DISK_SYNC
))
6134 set_bit(In_sync
, &rdev
->flags
);
6136 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
6137 set_bit(WriteMostly
, &rdev
->flags
);
6139 if (!mddev
->persistent
) {
6140 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
6141 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6143 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6144 rdev
->sectors
= rdev
->sb_start
;
6146 err
= bind_rdev_to_array(rdev
, mddev
);
6156 static int hot_remove_disk(struct mddev
*mddev
, dev_t dev
)
6158 char b
[BDEVNAME_SIZE
];
6159 struct md_rdev
*rdev
;
6161 rdev
= find_rdev(mddev
, dev
);
6165 if (rdev
->raid_disk
< 0)
6168 clear_bit(Blocked
, &rdev
->flags
);
6169 remove_and_add_spares(mddev
, rdev
);
6171 if (rdev
->raid_disk
>= 0)
6175 if (mddev_is_clustered(mddev
))
6176 md_cluster_ops
->remove_disk(mddev
, rdev
);
6178 md_kick_rdev_from_array(rdev
);
6179 md_update_sb(mddev
, 1);
6180 md_new_event(mddev
);
6184 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
6185 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6189 static int hot_add_disk(struct mddev
*mddev
, dev_t dev
)
6191 char b
[BDEVNAME_SIZE
];
6193 struct md_rdev
*rdev
;
6198 if (mddev
->major_version
!= 0) {
6199 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
6200 " version-0 superblocks.\n",
6204 if (!mddev
->pers
->hot_add_disk
) {
6206 "%s: personality does not support diskops!\n",
6211 rdev
= md_import_device(dev
, -1, 0);
6214 "md: error, md_import_device() returned %ld\n",
6219 if (mddev
->persistent
)
6220 rdev
->sb_start
= calc_dev_sboffset(rdev
);
6222 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
6224 rdev
->sectors
= rdev
->sb_start
;
6226 if (test_bit(Faulty
, &rdev
->flags
)) {
6228 "md: can not hot-add faulty %s disk to %s!\n",
6229 bdevname(rdev
->bdev
,b
), mdname(mddev
));
6234 clear_bit(In_sync
, &rdev
->flags
);
6236 rdev
->saved_raid_disk
= -1;
6237 err
= bind_rdev_to_array(rdev
, mddev
);
6242 * The rest should better be atomic, we can have disk failures
6243 * noticed in interrupt contexts ...
6246 rdev
->raid_disk
= -1;
6248 md_update_sb(mddev
, 1);
6250 * Kick recovery, maybe this spare has to be added to the
6251 * array immediately.
6253 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6254 md_wakeup_thread(mddev
->thread
);
6255 md_new_event(mddev
);
6263 static int set_bitmap_file(struct mddev
*mddev
, int fd
)
6268 if (!mddev
->pers
->quiesce
|| !mddev
->thread
)
6270 if (mddev
->recovery
|| mddev
->sync_thread
)
6272 /* we should be able to change the bitmap.. */
6276 struct inode
*inode
;
6279 if (mddev
->bitmap
|| mddev
->bitmap_info
.file
)
6280 return -EEXIST
; /* cannot add when bitmap is present */
6284 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
6289 inode
= f
->f_mapping
->host
;
6290 if (!S_ISREG(inode
->i_mode
)) {
6291 printk(KERN_ERR
"%s: error: bitmap file must be a regular file\n",
6294 } else if (!(f
->f_mode
& FMODE_WRITE
)) {
6295 printk(KERN_ERR
"%s: error: bitmap file must open for write\n",
6298 } else if (atomic_read(&inode
->i_writecount
) != 1) {
6299 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
6307 mddev
->bitmap_info
.file
= f
;
6308 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
6309 } else if (mddev
->bitmap
== NULL
)
6310 return -ENOENT
; /* cannot remove what isn't there */
6313 mddev
->pers
->quiesce(mddev
, 1);
6315 struct bitmap
*bitmap
;
6317 bitmap
= bitmap_create(mddev
, -1);
6318 if (!IS_ERR(bitmap
)) {
6319 mddev
->bitmap
= bitmap
;
6320 err
= bitmap_load(mddev
);
6322 err
= PTR_ERR(bitmap
);
6324 if (fd
< 0 || err
) {
6325 bitmap_destroy(mddev
);
6326 fd
= -1; /* make sure to put the file */
6328 mddev
->pers
->quiesce(mddev
, 0);
6331 struct file
*f
= mddev
->bitmap_info
.file
;
6333 spin_lock(&mddev
->lock
);
6334 mddev
->bitmap_info
.file
= NULL
;
6335 spin_unlock(&mddev
->lock
);
6344 * set_array_info is used two different ways
6345 * The original usage is when creating a new array.
6346 * In this usage, raid_disks is > 0 and it together with
6347 * level, size, not_persistent,layout,chunksize determine the
6348 * shape of the array.
6349 * This will always create an array with a type-0.90.0 superblock.
6350 * The newer usage is when assembling an array.
6351 * In this case raid_disks will be 0, and the major_version field is
6352 * use to determine which style super-blocks are to be found on the devices.
6353 * The minor and patch _version numbers are also kept incase the
6354 * super_block handler wishes to interpret them.
6356 static int set_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6359 if (info
->raid_disks
== 0) {
6360 /* just setting version number for superblock loading */
6361 if (info
->major_version
< 0 ||
6362 info
->major_version
>= ARRAY_SIZE(super_types
) ||
6363 super_types
[info
->major_version
].name
== NULL
) {
6364 /* maybe try to auto-load a module? */
6366 "md: superblock version %d not known\n",
6367 info
->major_version
);
6370 mddev
->major_version
= info
->major_version
;
6371 mddev
->minor_version
= info
->minor_version
;
6372 mddev
->patch_version
= info
->patch_version
;
6373 mddev
->persistent
= !info
->not_persistent
;
6374 /* ensure mddev_put doesn't delete this now that there
6375 * is some minimal configuration.
6377 mddev
->ctime
= ktime_get_real_seconds();
6380 mddev
->major_version
= MD_MAJOR_VERSION
;
6381 mddev
->minor_version
= MD_MINOR_VERSION
;
6382 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
6383 mddev
->ctime
= ktime_get_real_seconds();
6385 mddev
->level
= info
->level
;
6386 mddev
->clevel
[0] = 0;
6387 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
6388 mddev
->raid_disks
= info
->raid_disks
;
6389 /* don't set md_minor, it is determined by which /dev/md* was
6392 if (info
->state
& (1<<MD_SB_CLEAN
))
6393 mddev
->recovery_cp
= MaxSector
;
6395 mddev
->recovery_cp
= 0;
6396 mddev
->persistent
= ! info
->not_persistent
;
6397 mddev
->external
= 0;
6399 mddev
->layout
= info
->layout
;
6400 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
6402 mddev
->max_disks
= MD_SB_DISKS
;
6404 if (mddev
->persistent
)
6406 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6408 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
6409 mddev
->bitmap_info
.default_space
= 64*2 - (MD_SB_BYTES
>> 9);
6410 mddev
->bitmap_info
.offset
= 0;
6412 mddev
->reshape_position
= MaxSector
;
6415 * Generate a 128 bit UUID
6417 get_random_bytes(mddev
->uuid
, 16);
6419 mddev
->new_level
= mddev
->level
;
6420 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
6421 mddev
->new_layout
= mddev
->layout
;
6422 mddev
->delta_disks
= 0;
6423 mddev
->reshape_backwards
= 0;
6428 void md_set_array_sectors(struct mddev
*mddev
, sector_t array_sectors
)
6430 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
6432 if (mddev
->external_size
)
6435 mddev
->array_sectors
= array_sectors
;
6437 EXPORT_SYMBOL(md_set_array_sectors
);
6439 static int update_size(struct mddev
*mddev
, sector_t num_sectors
)
6441 struct md_rdev
*rdev
;
6443 int fit
= (num_sectors
== 0);
6445 /* cluster raid doesn't support update size */
6446 if (mddev_is_clustered(mddev
))
6449 if (mddev
->pers
->resize
== NULL
)
6451 /* The "num_sectors" is the number of sectors of each device that
6452 * is used. This can only make sense for arrays with redundancy.
6453 * linear and raid0 always use whatever space is available. We can only
6454 * consider changing this number if no resync or reconstruction is
6455 * happening, and if the new size is acceptable. It must fit before the
6456 * sb_start or, if that is <data_offset, it must fit before the size
6457 * of each device. If num_sectors is zero, we find the largest size
6460 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6466 rdev_for_each(rdev
, mddev
) {
6467 sector_t avail
= rdev
->sectors
;
6469 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
6470 num_sectors
= avail
;
6471 if (avail
< num_sectors
)
6474 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
6476 revalidate_disk(mddev
->gendisk
);
6480 static int update_raid_disks(struct mddev
*mddev
, int raid_disks
)
6483 struct md_rdev
*rdev
;
6484 /* change the number of raid disks */
6485 if (mddev
->pers
->check_reshape
== NULL
)
6489 if (raid_disks
<= 0 ||
6490 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
6492 if (mddev
->sync_thread
||
6493 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
6494 mddev
->reshape_position
!= MaxSector
)
6497 rdev_for_each(rdev
, mddev
) {
6498 if (mddev
->raid_disks
< raid_disks
&&
6499 rdev
->data_offset
< rdev
->new_data_offset
)
6501 if (mddev
->raid_disks
> raid_disks
&&
6502 rdev
->data_offset
> rdev
->new_data_offset
)
6506 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
6507 if (mddev
->delta_disks
< 0)
6508 mddev
->reshape_backwards
= 1;
6509 else if (mddev
->delta_disks
> 0)
6510 mddev
->reshape_backwards
= 0;
6512 rv
= mddev
->pers
->check_reshape(mddev
);
6514 mddev
->delta_disks
= 0;
6515 mddev
->reshape_backwards
= 0;
6521 * update_array_info is used to change the configuration of an
6523 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6524 * fields in the info are checked against the array.
6525 * Any differences that cannot be handled will cause an error.
6526 * Normally, only one change can be managed at a time.
6528 static int update_array_info(struct mddev
*mddev
, mdu_array_info_t
*info
)
6534 /* calculate expected state,ignoring low bits */
6535 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
6536 state
|= (1 << MD_SB_BITMAP_PRESENT
);
6538 if (mddev
->major_version
!= info
->major_version
||
6539 mddev
->minor_version
!= info
->minor_version
||
6540 /* mddev->patch_version != info->patch_version || */
6541 mddev
->ctime
!= info
->ctime
||
6542 mddev
->level
!= info
->level
||
6543 /* mddev->layout != info->layout || */
6544 mddev
->persistent
!= !info
->not_persistent
||
6545 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
6546 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6547 ((state
^info
->state
) & 0xfffffe00)
6550 /* Check there is only one change */
6551 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6553 if (mddev
->raid_disks
!= info
->raid_disks
)
6555 if (mddev
->layout
!= info
->layout
)
6557 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
6564 if (mddev
->layout
!= info
->layout
) {
6566 * we don't need to do anything at the md level, the
6567 * personality will take care of it all.
6569 if (mddev
->pers
->check_reshape
== NULL
)
6572 mddev
->new_layout
= info
->layout
;
6573 rv
= mddev
->pers
->check_reshape(mddev
);
6575 mddev
->new_layout
= mddev
->layout
;
6579 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
6580 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
6582 if (mddev
->raid_disks
!= info
->raid_disks
)
6583 rv
= update_raid_disks(mddev
, info
->raid_disks
);
6585 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
6586 if (mddev
->pers
->quiesce
== NULL
|| mddev
->thread
== NULL
) {
6590 if (mddev
->recovery
|| mddev
->sync_thread
) {
6594 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
6595 struct bitmap
*bitmap
;
6596 /* add the bitmap */
6597 if (mddev
->bitmap
) {
6601 if (mddev
->bitmap_info
.default_offset
== 0) {
6605 mddev
->bitmap_info
.offset
=
6606 mddev
->bitmap_info
.default_offset
;
6607 mddev
->bitmap_info
.space
=
6608 mddev
->bitmap_info
.default_space
;
6609 mddev
->pers
->quiesce(mddev
, 1);
6610 bitmap
= bitmap_create(mddev
, -1);
6611 if (!IS_ERR(bitmap
)) {
6612 mddev
->bitmap
= bitmap
;
6613 rv
= bitmap_load(mddev
);
6615 rv
= PTR_ERR(bitmap
);
6617 bitmap_destroy(mddev
);
6618 mddev
->pers
->quiesce(mddev
, 0);
6620 /* remove the bitmap */
6621 if (!mddev
->bitmap
) {
6625 if (mddev
->bitmap
->storage
.file
) {
6629 if (mddev
->bitmap_info
.nodes
) {
6630 /* hold PW on all the bitmap lock */
6631 if (md_cluster_ops
->lock_all_bitmaps(mddev
) <= 0) {
6632 printk("md: can't change bitmap to none since the"
6633 " array is in use by more than one node\n");
6635 md_cluster_ops
->unlock_all_bitmaps(mddev
);
6639 mddev
->bitmap_info
.nodes
= 0;
6640 md_cluster_ops
->leave(mddev
);
6642 mddev
->pers
->quiesce(mddev
, 1);
6643 bitmap_destroy(mddev
);
6644 mddev
->pers
->quiesce(mddev
, 0);
6645 mddev
->bitmap_info
.offset
= 0;
6648 md_update_sb(mddev
, 1);
6654 static int set_disk_faulty(struct mddev
*mddev
, dev_t dev
)
6656 struct md_rdev
*rdev
;
6659 if (mddev
->pers
== NULL
)
6663 rdev
= find_rdev_rcu(mddev
, dev
);
6667 md_error(mddev
, rdev
);
6668 if (!test_bit(Faulty
, &rdev
->flags
))
6676 * We have a problem here : there is no easy way to give a CHS
6677 * virtual geometry. We currently pretend that we have a 2 heads
6678 * 4 sectors (with a BIG number of cylinders...). This drives
6679 * dosfs just mad... ;-)
6681 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6683 struct mddev
*mddev
= bdev
->bd_disk
->private_data
;
6687 geo
->cylinders
= mddev
->array_sectors
/ 8;
6691 static inline bool md_ioctl_valid(unsigned int cmd
)
6696 case GET_ARRAY_INFO
:
6697 case GET_BITMAP_FILE
:
6700 case HOT_REMOVE_DISK
:
6703 case RESTART_ARRAY_RW
:
6705 case SET_ARRAY_INFO
:
6706 case SET_BITMAP_FILE
:
6707 case SET_DISK_FAULTY
:
6710 case CLUSTERED_DISK_NACK
:
6717 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6718 unsigned int cmd
, unsigned long arg
)
6721 void __user
*argp
= (void __user
*)arg
;
6722 struct mddev
*mddev
= NULL
;
6725 if (!md_ioctl_valid(cmd
))
6730 case GET_ARRAY_INFO
:
6734 if (!capable(CAP_SYS_ADMIN
))
6739 * Commands dealing with the RAID driver but not any
6744 err
= get_version(argp
);
6750 autostart_arrays(arg
);
6757 * Commands creating/starting a new array:
6760 mddev
= bdev
->bd_disk
->private_data
;
6767 /* Some actions do not requires the mutex */
6769 case GET_ARRAY_INFO
:
6770 if (!mddev
->raid_disks
&& !mddev
->external
)
6773 err
= get_array_info(mddev
, argp
);
6777 if (!mddev
->raid_disks
&& !mddev
->external
)
6780 err
= get_disk_info(mddev
, argp
);
6783 case SET_DISK_FAULTY
:
6784 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6787 case GET_BITMAP_FILE
:
6788 err
= get_bitmap_file(mddev
, argp
);
6793 if (cmd
== ADD_NEW_DISK
)
6794 /* need to ensure md_delayed_delete() has completed */
6795 flush_workqueue(md_misc_wq
);
6797 if (cmd
== HOT_REMOVE_DISK
)
6798 /* need to ensure recovery thread has run */
6799 wait_event_interruptible_timeout(mddev
->sb_wait
,
6800 !test_bit(MD_RECOVERY_NEEDED
,
6802 msecs_to_jiffies(5000));
6803 if (cmd
== STOP_ARRAY
|| cmd
== STOP_ARRAY_RO
) {
6804 /* Need to flush page cache, and ensure no-one else opens
6807 mutex_lock(&mddev
->open_mutex
);
6808 if (mddev
->pers
&& atomic_read(&mddev
->openers
) > 1) {
6809 mutex_unlock(&mddev
->open_mutex
);
6813 set_bit(MD_STILL_CLOSED
, &mddev
->flags
);
6814 mutex_unlock(&mddev
->open_mutex
);
6815 sync_blockdev(bdev
);
6817 err
= mddev_lock(mddev
);
6820 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6825 if (cmd
== SET_ARRAY_INFO
) {
6826 mdu_array_info_t info
;
6828 memset(&info
, 0, sizeof(info
));
6829 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6834 err
= update_array_info(mddev
, &info
);
6836 printk(KERN_WARNING
"md: couldn't update"
6837 " array info. %d\n", err
);
6842 if (!list_empty(&mddev
->disks
)) {
6844 "md: array %s already has disks!\n",
6849 if (mddev
->raid_disks
) {
6851 "md: array %s already initialised!\n",
6856 err
= set_array_info(mddev
, &info
);
6858 printk(KERN_WARNING
"md: couldn't set"
6859 " array info. %d\n", err
);
6866 * Commands querying/configuring an existing array:
6868 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6869 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6870 if ((!mddev
->raid_disks
&& !mddev
->external
)
6871 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6872 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6873 && cmd
!= GET_BITMAP_FILE
) {
6879 * Commands even a read-only array can execute:
6882 case RESTART_ARRAY_RW
:
6883 err
= restart_array(mddev
);
6887 err
= do_md_stop(mddev
, 0, bdev
);
6891 err
= md_set_readonly(mddev
, bdev
);
6894 case HOT_REMOVE_DISK
:
6895 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6899 /* We can support ADD_NEW_DISK on read-only arrays
6900 * only if we are re-adding a preexisting device.
6901 * So require mddev->pers and MD_DISK_SYNC.
6904 mdu_disk_info_t info
;
6905 if (copy_from_user(&info
, argp
, sizeof(info
)))
6907 else if (!(info
.state
& (1<<MD_DISK_SYNC
)))
6908 /* Need to clear read-only for this */
6911 err
= add_new_disk(mddev
, &info
);
6917 if (get_user(ro
, (int __user
*)(arg
))) {
6923 /* if the bdev is going readonly the value of mddev->ro
6924 * does not matter, no writes are coming
6929 /* are we are already prepared for writes? */
6933 /* transitioning to readauto need only happen for
6934 * arrays that call md_write_start
6937 err
= restart_array(mddev
);
6940 set_disk_ro(mddev
->gendisk
, 0);
6947 * The remaining ioctls are changing the state of the
6948 * superblock, so we do not allow them on read-only arrays.
6950 if (mddev
->ro
&& mddev
->pers
) {
6951 if (mddev
->ro
== 2) {
6953 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6954 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6955 /* mddev_unlock will wake thread */
6956 /* If a device failed while we were read-only, we
6957 * need to make sure the metadata is updated now.
6959 if (test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
6960 mddev_unlock(mddev
);
6961 wait_event(mddev
->sb_wait
,
6962 !test_bit(MD_CHANGE_DEVS
, &mddev
->flags
) &&
6963 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6964 mddev_lock_nointr(mddev
);
6975 mdu_disk_info_t info
;
6976 if (copy_from_user(&info
, argp
, sizeof(info
)))
6979 err
= add_new_disk(mddev
, &info
);
6983 case CLUSTERED_DISK_NACK
:
6984 if (mddev_is_clustered(mddev
))
6985 md_cluster_ops
->new_disk_ack(mddev
, false);
6991 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6995 err
= do_md_run(mddev
);
6998 case SET_BITMAP_FILE
:
6999 err
= set_bitmap_file(mddev
, (int)arg
);
7008 if (mddev
->hold_active
== UNTIL_IOCTL
&&
7010 mddev
->hold_active
= 0;
7011 mddev_unlock(mddev
);
7015 #ifdef CONFIG_COMPAT
7016 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
7017 unsigned int cmd
, unsigned long arg
)
7020 case HOT_REMOVE_DISK
:
7022 case SET_DISK_FAULTY
:
7023 case SET_BITMAP_FILE
:
7024 /* These take in integer arg, do not convert */
7027 arg
= (unsigned long)compat_ptr(arg
);
7031 return md_ioctl(bdev
, mode
, cmd
, arg
);
7033 #endif /* CONFIG_COMPAT */
7035 static int md_open(struct block_device
*bdev
, fmode_t mode
)
7038 * Succeed if we can lock the mddev, which confirms that
7039 * it isn't being stopped right now.
7041 struct mddev
*mddev
= mddev_find(bdev
->bd_dev
);
7047 if (mddev
->gendisk
!= bdev
->bd_disk
) {
7048 /* we are racing with mddev_put which is discarding this
7052 /* Wait until bdev->bd_disk is definitely gone */
7053 flush_workqueue(md_misc_wq
);
7054 /* Then retry the open from the top */
7055 return -ERESTARTSYS
;
7057 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
7059 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
7063 atomic_inc(&mddev
->openers
);
7064 clear_bit(MD_STILL_CLOSED
, &mddev
->flags
);
7065 mutex_unlock(&mddev
->open_mutex
);
7067 check_disk_change(bdev
);
7072 static void md_release(struct gendisk
*disk
, fmode_t mode
)
7074 struct mddev
*mddev
= disk
->private_data
;
7077 atomic_dec(&mddev
->openers
);
7081 static int md_media_changed(struct gendisk
*disk
)
7083 struct mddev
*mddev
= disk
->private_data
;
7085 return mddev
->changed
;
7088 static int md_revalidate(struct gendisk
*disk
)
7090 struct mddev
*mddev
= disk
->private_data
;
7095 static const struct block_device_operations md_fops
=
7097 .owner
= THIS_MODULE
,
7099 .release
= md_release
,
7101 #ifdef CONFIG_COMPAT
7102 .compat_ioctl
= md_compat_ioctl
,
7104 .getgeo
= md_getgeo
,
7105 .media_changed
= md_media_changed
,
7106 .revalidate_disk
= md_revalidate
,
7109 static int md_thread(void *arg
)
7111 struct md_thread
*thread
= arg
;
7114 * md_thread is a 'system-thread', it's priority should be very
7115 * high. We avoid resource deadlocks individually in each
7116 * raid personality. (RAID5 does preallocation) We also use RR and
7117 * the very same RT priority as kswapd, thus we will never get
7118 * into a priority inversion deadlock.
7120 * we definitely have to have equal or higher priority than
7121 * bdflush, otherwise bdflush will deadlock if there are too
7122 * many dirty RAID5 blocks.
7125 allow_signal(SIGKILL
);
7126 while (!kthread_should_stop()) {
7128 /* We need to wait INTERRUPTIBLE so that
7129 * we don't add to the load-average.
7130 * That means we need to be sure no signals are
7133 if (signal_pending(current
))
7134 flush_signals(current
);
7136 wait_event_interruptible_timeout
7138 test_bit(THREAD_WAKEUP
, &thread
->flags
)
7139 || kthread_should_stop(),
7142 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
7143 if (!kthread_should_stop())
7144 thread
->run(thread
);
7150 void md_wakeup_thread(struct md_thread
*thread
)
7153 pr_debug("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
7154 set_bit(THREAD_WAKEUP
, &thread
->flags
);
7155 wake_up(&thread
->wqueue
);
7158 EXPORT_SYMBOL(md_wakeup_thread
);
7160 struct md_thread
*md_register_thread(void (*run
) (struct md_thread
*),
7161 struct mddev
*mddev
, const char *name
)
7163 struct md_thread
*thread
;
7165 thread
= kzalloc(sizeof(struct md_thread
), GFP_KERNEL
);
7169 init_waitqueue_head(&thread
->wqueue
);
7172 thread
->mddev
= mddev
;
7173 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
7174 thread
->tsk
= kthread_run(md_thread
, thread
,
7176 mdname(thread
->mddev
),
7178 if (IS_ERR(thread
->tsk
)) {
7184 EXPORT_SYMBOL(md_register_thread
);
7186 void md_unregister_thread(struct md_thread
**threadp
)
7188 struct md_thread
*thread
= *threadp
;
7191 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
7192 /* Locking ensures that mddev_unlock does not wake_up a
7193 * non-existent thread
7195 spin_lock(&pers_lock
);
7197 spin_unlock(&pers_lock
);
7199 kthread_stop(thread
->tsk
);
7202 EXPORT_SYMBOL(md_unregister_thread
);
7204 void md_error(struct mddev
*mddev
, struct md_rdev
*rdev
)
7206 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
7209 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
7211 mddev
->pers
->error_handler(mddev
,rdev
);
7212 if (mddev
->degraded
)
7213 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7214 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7215 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7216 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7217 md_wakeup_thread(mddev
->thread
);
7218 if (mddev
->event_work
.func
)
7219 queue_work(md_misc_wq
, &mddev
->event_work
);
7220 md_new_event(mddev
);
7222 EXPORT_SYMBOL(md_error
);
7224 /* seq_file implementation /proc/mdstat */
7226 static void status_unused(struct seq_file
*seq
)
7229 struct md_rdev
*rdev
;
7231 seq_printf(seq
, "unused devices: ");
7233 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
7234 char b
[BDEVNAME_SIZE
];
7236 seq_printf(seq
, "%s ",
7237 bdevname(rdev
->bdev
,b
));
7240 seq_printf(seq
, "<none>");
7242 seq_printf(seq
, "\n");
7245 static int status_resync(struct seq_file
*seq
, struct mddev
*mddev
)
7247 sector_t max_sectors
, resync
, res
;
7248 unsigned long dt
, db
;
7251 unsigned int per_milli
;
7253 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7254 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7255 max_sectors
= mddev
->resync_max_sectors
;
7257 max_sectors
= mddev
->dev_sectors
;
7259 resync
= mddev
->curr_resync
;
7261 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7262 /* Still cleaning up */
7263 resync
= max_sectors
;
7265 resync
-= atomic_read(&mddev
->recovery_active
);
7268 if (mddev
->recovery_cp
< MaxSector
) {
7269 seq_printf(seq
, "\tresync=PENDING");
7275 seq_printf(seq
, "\tresync=DELAYED");
7279 WARN_ON(max_sectors
== 0);
7280 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7281 * in a sector_t, and (max_sectors>>scale) will fit in a
7282 * u32, as those are the requirements for sector_div.
7283 * Thus 'scale' must be at least 10
7286 if (sizeof(sector_t
) > sizeof(unsigned long)) {
7287 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
7290 res
= (resync
>>scale
)*1000;
7291 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
7295 int i
, x
= per_milli
/50, y
= 20-x
;
7296 seq_printf(seq
, "[");
7297 for (i
= 0; i
< x
; i
++)
7298 seq_printf(seq
, "=");
7299 seq_printf(seq
, ">");
7300 for (i
= 0; i
< y
; i
++)
7301 seq_printf(seq
, ".");
7302 seq_printf(seq
, "] ");
7304 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
7305 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
7307 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
7309 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
7310 "resync" : "recovery"))),
7311 per_milli
/10, per_milli
% 10,
7312 (unsigned long long) resync
/2,
7313 (unsigned long long) max_sectors
/2);
7316 * dt: time from mark until now
7317 * db: blocks written from mark until now
7318 * rt: remaining time
7320 * rt is a sector_t, so could be 32bit or 64bit.
7321 * So we divide before multiply in case it is 32bit and close
7323 * We scale the divisor (db) by 32 to avoid losing precision
7324 * near the end of resync when the number of remaining sectors
7326 * We then divide rt by 32 after multiplying by db to compensate.
7327 * The '+1' avoids division by zero if db is very small.
7329 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
7331 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
7332 - mddev
->resync_mark_cnt
;
7334 rt
= max_sectors
- resync
; /* number of remaining sectors */
7335 sector_div(rt
, db
/32+1);
7339 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
7340 ((unsigned long)rt
% 60)/6);
7342 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
7346 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
7348 struct list_head
*tmp
;
7350 struct mddev
*mddev
;
7358 spin_lock(&all_mddevs_lock
);
7359 list_for_each(tmp
,&all_mddevs
)
7361 mddev
= list_entry(tmp
, struct mddev
, all_mddevs
);
7363 spin_unlock(&all_mddevs_lock
);
7366 spin_unlock(&all_mddevs_lock
);
7368 return (void*)2;/* tail */
7372 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
7374 struct list_head
*tmp
;
7375 struct mddev
*next_mddev
, *mddev
= v
;
7381 spin_lock(&all_mddevs_lock
);
7383 tmp
= all_mddevs
.next
;
7385 tmp
= mddev
->all_mddevs
.next
;
7386 if (tmp
!= &all_mddevs
)
7387 next_mddev
= mddev_get(list_entry(tmp
,struct mddev
,all_mddevs
));
7389 next_mddev
= (void*)2;
7392 spin_unlock(&all_mddevs_lock
);
7400 static void md_seq_stop(struct seq_file
*seq
, void *v
)
7402 struct mddev
*mddev
= v
;
7404 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
7408 static int md_seq_show(struct seq_file
*seq
, void *v
)
7410 struct mddev
*mddev
= v
;
7412 struct md_rdev
*rdev
;
7414 if (v
== (void*)1) {
7415 struct md_personality
*pers
;
7416 seq_printf(seq
, "Personalities : ");
7417 spin_lock(&pers_lock
);
7418 list_for_each_entry(pers
, &pers_list
, list
)
7419 seq_printf(seq
, "[%s] ", pers
->name
);
7421 spin_unlock(&pers_lock
);
7422 seq_printf(seq
, "\n");
7423 seq
->poll_event
= atomic_read(&md_event_count
);
7426 if (v
== (void*)2) {
7431 spin_lock(&mddev
->lock
);
7432 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
7433 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
7434 mddev
->pers
? "" : "in");
7437 seq_printf(seq
, " (read-only)");
7439 seq_printf(seq
, " (auto-read-only)");
7440 seq_printf(seq
, " %s", mddev
->pers
->name
);
7445 rdev_for_each_rcu(rdev
, mddev
) {
7446 char b
[BDEVNAME_SIZE
];
7447 seq_printf(seq
, " %s[%d]",
7448 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
7449 if (test_bit(WriteMostly
, &rdev
->flags
))
7450 seq_printf(seq
, "(W)");
7451 if (test_bit(Journal
, &rdev
->flags
))
7452 seq_printf(seq
, "(J)");
7453 if (test_bit(Faulty
, &rdev
->flags
)) {
7454 seq_printf(seq
, "(F)");
7457 if (rdev
->raid_disk
< 0)
7458 seq_printf(seq
, "(S)"); /* spare */
7459 if (test_bit(Replacement
, &rdev
->flags
))
7460 seq_printf(seq
, "(R)");
7461 sectors
+= rdev
->sectors
;
7465 if (!list_empty(&mddev
->disks
)) {
7467 seq_printf(seq
, "\n %llu blocks",
7468 (unsigned long long)
7469 mddev
->array_sectors
/ 2);
7471 seq_printf(seq
, "\n %llu blocks",
7472 (unsigned long long)sectors
/ 2);
7474 if (mddev
->persistent
) {
7475 if (mddev
->major_version
!= 0 ||
7476 mddev
->minor_version
!= 90) {
7477 seq_printf(seq
," super %d.%d",
7478 mddev
->major_version
,
7479 mddev
->minor_version
);
7481 } else if (mddev
->external
)
7482 seq_printf(seq
, " super external:%s",
7483 mddev
->metadata_type
);
7485 seq_printf(seq
, " super non-persistent");
7488 mddev
->pers
->status(seq
, mddev
);
7489 seq_printf(seq
, "\n ");
7490 if (mddev
->pers
->sync_request
) {
7491 if (status_resync(seq
, mddev
))
7492 seq_printf(seq
, "\n ");
7495 seq_printf(seq
, "\n ");
7497 bitmap_status(seq
, mddev
->bitmap
);
7499 seq_printf(seq
, "\n");
7501 spin_unlock(&mddev
->lock
);
7506 static const struct seq_operations md_seq_ops
= {
7507 .start
= md_seq_start
,
7508 .next
= md_seq_next
,
7509 .stop
= md_seq_stop
,
7510 .show
= md_seq_show
,
7513 static int md_seq_open(struct inode
*inode
, struct file
*file
)
7515 struct seq_file
*seq
;
7518 error
= seq_open(file
, &md_seq_ops
);
7522 seq
= file
->private_data
;
7523 seq
->poll_event
= atomic_read(&md_event_count
);
7527 static int md_unloading
;
7528 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
7530 struct seq_file
*seq
= filp
->private_data
;
7534 return POLLIN
|POLLRDNORM
|POLLERR
|POLLPRI
;
7535 poll_wait(filp
, &md_event_waiters
, wait
);
7537 /* always allow read */
7538 mask
= POLLIN
| POLLRDNORM
;
7540 if (seq
->poll_event
!= atomic_read(&md_event_count
))
7541 mask
|= POLLERR
| POLLPRI
;
7545 static const struct file_operations md_seq_fops
= {
7546 .owner
= THIS_MODULE
,
7547 .open
= md_seq_open
,
7549 .llseek
= seq_lseek
,
7550 .release
= seq_release_private
,
7551 .poll
= mdstat_poll
,
7554 int register_md_personality(struct md_personality
*p
)
7556 printk(KERN_INFO
"md: %s personality registered for level %d\n",
7558 spin_lock(&pers_lock
);
7559 list_add_tail(&p
->list
, &pers_list
);
7560 spin_unlock(&pers_lock
);
7563 EXPORT_SYMBOL(register_md_personality
);
7565 int unregister_md_personality(struct md_personality
*p
)
7567 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
7568 spin_lock(&pers_lock
);
7569 list_del_init(&p
->list
);
7570 spin_unlock(&pers_lock
);
7573 EXPORT_SYMBOL(unregister_md_personality
);
7575 int register_md_cluster_operations(struct md_cluster_operations
*ops
,
7576 struct module
*module
)
7579 spin_lock(&pers_lock
);
7580 if (md_cluster_ops
!= NULL
)
7583 md_cluster_ops
= ops
;
7584 md_cluster_mod
= module
;
7586 spin_unlock(&pers_lock
);
7589 EXPORT_SYMBOL(register_md_cluster_operations
);
7591 int unregister_md_cluster_operations(void)
7593 spin_lock(&pers_lock
);
7594 md_cluster_ops
= NULL
;
7595 spin_unlock(&pers_lock
);
7598 EXPORT_SYMBOL(unregister_md_cluster_operations
);
7600 int md_setup_cluster(struct mddev
*mddev
, int nodes
)
7604 err
= request_module("md-cluster");
7606 pr_err("md-cluster module not found.\n");
7610 spin_lock(&pers_lock
);
7611 if (!md_cluster_ops
|| !try_module_get(md_cluster_mod
)) {
7612 spin_unlock(&pers_lock
);
7615 spin_unlock(&pers_lock
);
7617 return md_cluster_ops
->join(mddev
, nodes
);
7620 void md_cluster_stop(struct mddev
*mddev
)
7622 if (!md_cluster_ops
)
7624 md_cluster_ops
->leave(mddev
);
7625 module_put(md_cluster_mod
);
7628 static int is_mddev_idle(struct mddev
*mddev
, int init
)
7630 struct md_rdev
*rdev
;
7636 rdev_for_each_rcu(rdev
, mddev
) {
7637 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
7638 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
7639 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
7640 atomic_read(&disk
->sync_io
);
7641 /* sync IO will cause sync_io to increase before the disk_stats
7642 * as sync_io is counted when a request starts, and
7643 * disk_stats is counted when it completes.
7644 * So resync activity will cause curr_events to be smaller than
7645 * when there was no such activity.
7646 * non-sync IO will cause disk_stat to increase without
7647 * increasing sync_io so curr_events will (eventually)
7648 * be larger than it was before. Once it becomes
7649 * substantially larger, the test below will cause
7650 * the array to appear non-idle, and resync will slow
7652 * If there is a lot of outstanding resync activity when
7653 * we set last_event to curr_events, then all that activity
7654 * completing might cause the array to appear non-idle
7655 * and resync will be slowed down even though there might
7656 * not have been non-resync activity. This will only
7657 * happen once though. 'last_events' will soon reflect
7658 * the state where there is little or no outstanding
7659 * resync requests, and further resync activity will
7660 * always make curr_events less than last_events.
7663 if (init
|| curr_events
- rdev
->last_events
> 64) {
7664 rdev
->last_events
= curr_events
;
7672 void md_done_sync(struct mddev
*mddev
, int blocks
, int ok
)
7674 /* another "blocks" (512byte) blocks have been synced */
7675 atomic_sub(blocks
, &mddev
->recovery_active
);
7676 wake_up(&mddev
->recovery_wait
);
7678 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7679 set_bit(MD_RECOVERY_ERROR
, &mddev
->recovery
);
7680 md_wakeup_thread(mddev
->thread
);
7681 // stop recovery, signal do_sync ....
7684 EXPORT_SYMBOL(md_done_sync
);
7686 /* md_write_start(mddev, bi)
7687 * If we need to update some array metadata (e.g. 'active' flag
7688 * in superblock) before writing, schedule a superblock update
7689 * and wait for it to complete.
7691 void md_write_start(struct mddev
*mddev
, struct bio
*bi
)
7694 if (bio_data_dir(bi
) != WRITE
)
7697 BUG_ON(mddev
->ro
== 1);
7698 if (mddev
->ro
== 2) {
7699 /* need to switch to read/write */
7701 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7702 md_wakeup_thread(mddev
->thread
);
7703 md_wakeup_thread(mddev
->sync_thread
);
7706 atomic_inc(&mddev
->writes_pending
);
7707 if (mddev
->safemode
== 1)
7708 mddev
->safemode
= 0;
7709 if (mddev
->in_sync
) {
7710 spin_lock(&mddev
->lock
);
7711 if (mddev
->in_sync
) {
7713 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7714 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7715 md_wakeup_thread(mddev
->thread
);
7718 spin_unlock(&mddev
->lock
);
7721 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7722 wait_event(mddev
->sb_wait
,
7723 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
7725 EXPORT_SYMBOL(md_write_start
);
7727 void md_write_end(struct mddev
*mddev
)
7729 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
7730 if (mddev
->safemode
== 2)
7731 md_wakeup_thread(mddev
->thread
);
7732 else if (mddev
->safemode_delay
)
7733 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
7736 EXPORT_SYMBOL(md_write_end
);
7738 /* md_allow_write(mddev)
7739 * Calling this ensures that the array is marked 'active' so that writes
7740 * may proceed without blocking. It is important to call this before
7741 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7742 * Must be called with mddev_lock held.
7744 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7745 * is dropped, so return -EAGAIN after notifying userspace.
7747 int md_allow_write(struct mddev
*mddev
)
7753 if (!mddev
->pers
->sync_request
)
7756 spin_lock(&mddev
->lock
);
7757 if (mddev
->in_sync
) {
7759 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7760 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
7761 if (mddev
->safemode_delay
&&
7762 mddev
->safemode
== 0)
7763 mddev
->safemode
= 1;
7764 spin_unlock(&mddev
->lock
);
7765 md_update_sb(mddev
, 0);
7766 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7768 spin_unlock(&mddev
->lock
);
7770 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
7775 EXPORT_SYMBOL_GPL(md_allow_write
);
7777 #define SYNC_MARKS 10
7778 #define SYNC_MARK_STEP (3*HZ)
7779 #define UPDATE_FREQUENCY (5*60*HZ)
7780 void md_do_sync(struct md_thread
*thread
)
7782 struct mddev
*mddev
= thread
->mddev
;
7783 struct mddev
*mddev2
;
7784 unsigned int currspeed
= 0,
7786 sector_t max_sectors
,j
, io_sectors
, recovery_done
;
7787 unsigned long mark
[SYNC_MARKS
];
7788 unsigned long update_time
;
7789 sector_t mark_cnt
[SYNC_MARKS
];
7791 struct list_head
*tmp
;
7792 sector_t last_check
;
7794 struct md_rdev
*rdev
;
7795 char *desc
, *action
= NULL
;
7796 struct blk_plug plug
;
7799 /* just incase thread restarts... */
7800 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
7802 if (mddev
->ro
) {/* never try to sync a read-only array */
7803 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7807 if (mddev_is_clustered(mddev
)) {
7808 ret
= md_cluster_ops
->resync_start(mddev
);
7812 if (!(test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ||
7813 test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) ||
7814 test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
7815 && ((unsigned long long)mddev
->curr_resync_completed
7816 < (unsigned long long)mddev
->resync_max_sectors
))
7820 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7821 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
7822 desc
= "data-check";
7824 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7825 desc
= "requested-resync";
7829 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7834 mddev
->last_sync_action
= action
?: desc
;
7836 /* we overload curr_resync somewhat here.
7837 * 0 == not engaged in resync at all
7838 * 2 == checking that there is no conflict with another sync
7839 * 1 == like 2, but have yielded to allow conflicting resync to
7841 * other == active in resync - this many blocks
7843 * Before starting a resync we must have set curr_resync to
7844 * 2, and then checked that every "conflicting" array has curr_resync
7845 * less than ours. When we find one that is the same or higher
7846 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7847 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7848 * This will mean we have to start checking from the beginning again.
7853 mddev
->curr_resync
= 2;
7856 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7858 for_each_mddev(mddev2
, tmp
) {
7859 if (mddev2
== mddev
)
7861 if (!mddev
->parallel_resync
7862 && mddev2
->curr_resync
7863 && match_mddev_units(mddev
, mddev2
)) {
7865 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7866 /* arbitrarily yield */
7867 mddev
->curr_resync
= 1;
7868 wake_up(&resync_wait
);
7870 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7871 /* no need to wait here, we can wait the next
7872 * time 'round when curr_resync == 2
7875 /* We need to wait 'interruptible' so as not to
7876 * contribute to the load average, and not to
7877 * be caught by 'softlockup'
7879 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7880 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7881 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7882 printk(KERN_INFO
"md: delaying %s of %s"
7883 " until %s has finished (they"
7884 " share one or more physical units)\n",
7885 desc
, mdname(mddev
), mdname(mddev2
));
7887 if (signal_pending(current
))
7888 flush_signals(current
);
7890 finish_wait(&resync_wait
, &wq
);
7893 finish_wait(&resync_wait
, &wq
);
7896 } while (mddev
->curr_resync
< 2);
7899 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7900 /* resync follows the size requested by the personality,
7901 * which defaults to physical size, but can be virtual size
7903 max_sectors
= mddev
->resync_max_sectors
;
7904 atomic64_set(&mddev
->resync_mismatches
, 0);
7905 /* we don't use the checkpoint if there's a bitmap */
7906 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7907 j
= mddev
->resync_min
;
7908 else if (!mddev
->bitmap
)
7909 j
= mddev
->recovery_cp
;
7911 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7912 max_sectors
= mddev
->resync_max_sectors
;
7914 /* recovery follows the physical size of devices */
7915 max_sectors
= mddev
->dev_sectors
;
7918 rdev_for_each_rcu(rdev
, mddev
)
7919 if (rdev
->raid_disk
>= 0 &&
7920 !test_bit(Journal
, &rdev
->flags
) &&
7921 !test_bit(Faulty
, &rdev
->flags
) &&
7922 !test_bit(In_sync
, &rdev
->flags
) &&
7923 rdev
->recovery_offset
< j
)
7924 j
= rdev
->recovery_offset
;
7927 /* If there is a bitmap, we need to make sure all
7928 * writes that started before we added a spare
7929 * complete before we start doing a recovery.
7930 * Otherwise the write might complete and (via
7931 * bitmap_endwrite) set a bit in the bitmap after the
7932 * recovery has checked that bit and skipped that
7935 if (mddev
->bitmap
) {
7936 mddev
->pers
->quiesce(mddev
, 1);
7937 mddev
->pers
->quiesce(mddev
, 0);
7941 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7942 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7943 " %d KB/sec/disk.\n", speed_min(mddev
));
7944 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7945 "(but not more than %d KB/sec) for %s.\n",
7946 speed_max(mddev
), desc
);
7948 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7951 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7953 mark_cnt
[m
] = io_sectors
;
7956 mddev
->resync_mark
= mark
[last_mark
];
7957 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7960 * Tune reconstruction:
7962 window
= 32*(PAGE_SIZE
/512);
7963 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7964 window
/2, (unsigned long long)max_sectors
/2);
7966 atomic_set(&mddev
->recovery_active
, 0);
7971 "md: resuming %s of %s from checkpoint.\n",
7972 desc
, mdname(mddev
));
7973 mddev
->curr_resync
= j
;
7975 mddev
->curr_resync
= 3; /* no longer delayed */
7976 mddev
->curr_resync_completed
= j
;
7977 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7978 md_new_event(mddev
);
7979 update_time
= jiffies
;
7981 blk_start_plug(&plug
);
7982 while (j
< max_sectors
) {
7987 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7988 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7989 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7990 > (max_sectors
>> 4)) ||
7991 time_after_eq(jiffies
, update_time
+ UPDATE_FREQUENCY
) ||
7992 (j
- mddev
->curr_resync_completed
)*2
7993 >= mddev
->resync_max
- mddev
->curr_resync_completed
||
7994 mddev
->curr_resync_completed
> mddev
->resync_max
7996 /* time to update curr_resync_completed */
7997 wait_event(mddev
->recovery_wait
,
7998 atomic_read(&mddev
->recovery_active
) == 0);
7999 mddev
->curr_resync_completed
= j
;
8000 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
8001 j
> mddev
->recovery_cp
)
8002 mddev
->recovery_cp
= j
;
8003 update_time
= jiffies
;
8004 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
8005 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8008 while (j
>= mddev
->resync_max
&&
8009 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8010 /* As this condition is controlled by user-space,
8011 * we can block indefinitely, so use '_interruptible'
8012 * to avoid triggering warnings.
8014 flush_signals(current
); /* just in case */
8015 wait_event_interruptible(mddev
->recovery_wait
,
8016 mddev
->resync_max
> j
8017 || test_bit(MD_RECOVERY_INTR
,
8021 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8024 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
);
8026 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8030 if (!skipped
) { /* actual IO requested */
8031 io_sectors
+= sectors
;
8032 atomic_add(sectors
, &mddev
->recovery_active
);
8035 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8039 if (j
> max_sectors
)
8040 /* when skipping, extra large numbers can be returned. */
8043 mddev
->curr_resync
= j
;
8044 mddev
->curr_mark_cnt
= io_sectors
;
8045 if (last_check
== 0)
8046 /* this is the earliest that rebuild will be
8047 * visible in /proc/mdstat
8049 md_new_event(mddev
);
8051 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
8054 last_check
= io_sectors
;
8056 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
8058 int next
= (last_mark
+1) % SYNC_MARKS
;
8060 mddev
->resync_mark
= mark
[next
];
8061 mddev
->resync_mark_cnt
= mark_cnt
[next
];
8062 mark
[next
] = jiffies
;
8063 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
8067 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8071 * this loop exits only if either when we are slower than
8072 * the 'hard' speed limit, or the system was IO-idle for
8074 * the system might be non-idle CPU-wise, but we only care
8075 * about not overloading the IO subsystem. (things like an
8076 * e2fsck being done on the RAID array should execute fast)
8080 recovery_done
= io_sectors
- atomic_read(&mddev
->recovery_active
);
8081 currspeed
= ((unsigned long)(recovery_done
- mddev
->resync_mark_cnt
))/2
8082 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
8084 if (currspeed
> speed_min(mddev
)) {
8085 if (currspeed
> speed_max(mddev
)) {
8089 if (!is_mddev_idle(mddev
, 0)) {
8091 * Give other IO more of a chance.
8092 * The faster the devices, the less we wait.
8094 wait_event(mddev
->recovery_wait
,
8095 !atomic_read(&mddev
->recovery_active
));
8099 printk(KERN_INFO
"md: %s: %s %s.\n",mdname(mddev
), desc
,
8100 test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)
8101 ? "interrupted" : "done");
8103 * this also signals 'finished resyncing' to md_stop
8105 blk_finish_plug(&plug
);
8106 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
8108 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8109 !test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8110 mddev
->curr_resync
> 2) {
8111 mddev
->curr_resync_completed
= mddev
->curr_resync
;
8112 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
8114 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
);
8116 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
8117 mddev
->curr_resync
> 2) {
8118 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
8119 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8120 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
8122 "md: checkpointing %s of %s.\n",
8123 desc
, mdname(mddev
));
8124 if (test_bit(MD_RECOVERY_ERROR
,
8126 mddev
->recovery_cp
=
8127 mddev
->curr_resync_completed
;
8129 mddev
->recovery_cp
=
8133 mddev
->recovery_cp
= MaxSector
;
8135 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
8136 mddev
->curr_resync
= MaxSector
;
8138 rdev_for_each_rcu(rdev
, mddev
)
8139 if (rdev
->raid_disk
>= 0 &&
8140 mddev
->delta_disks
>= 0 &&
8141 !test_bit(Journal
, &rdev
->flags
) &&
8142 !test_bit(Faulty
, &rdev
->flags
) &&
8143 !test_bit(In_sync
, &rdev
->flags
) &&
8144 rdev
->recovery_offset
< mddev
->curr_resync
)
8145 rdev
->recovery_offset
= mddev
->curr_resync
;
8150 if (mddev_is_clustered(mddev
) &&
8152 /* set CHANGE_PENDING here since maybe another
8153 * update is needed, so other nodes are informed */
8154 set_mask_bits(&mddev
->flags
, 0,
8155 BIT(MD_CHANGE_PENDING
) | BIT(MD_CHANGE_DEVS
));
8156 md_wakeup_thread(mddev
->thread
);
8157 wait_event(mddev
->sb_wait
,
8158 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
8159 md_cluster_ops
->resync_finish(mddev
);
8161 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8163 spin_lock(&mddev
->lock
);
8164 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
8165 /* We completed so min/max setting can be forgotten if used. */
8166 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8167 mddev
->resync_min
= 0;
8168 mddev
->resync_max
= MaxSector
;
8169 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
8170 mddev
->resync_min
= mddev
->curr_resync_completed
;
8171 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8172 mddev
->curr_resync
= 0;
8173 spin_unlock(&mddev
->lock
);
8175 wake_up(&resync_wait
);
8176 md_wakeup_thread(mddev
->thread
);
8179 EXPORT_SYMBOL_GPL(md_do_sync
);
8181 static int remove_and_add_spares(struct mddev
*mddev
,
8182 struct md_rdev
*this)
8184 struct md_rdev
*rdev
;
8188 rdev_for_each(rdev
, mddev
)
8189 if ((this == NULL
|| rdev
== this) &&
8190 rdev
->raid_disk
>= 0 &&
8191 !test_bit(Blocked
, &rdev
->flags
) &&
8192 (test_bit(Faulty
, &rdev
->flags
) ||
8193 (!test_bit(In_sync
, &rdev
->flags
) &&
8194 !test_bit(Journal
, &rdev
->flags
))) &&
8195 atomic_read(&rdev
->nr_pending
)==0) {
8196 if (mddev
->pers
->hot_remove_disk(
8197 mddev
, rdev
) == 0) {
8198 sysfs_unlink_rdev(mddev
, rdev
);
8199 rdev
->raid_disk
= -1;
8203 if (removed
&& mddev
->kobj
.sd
)
8204 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
8206 if (this && removed
)
8209 rdev_for_each(rdev
, mddev
) {
8210 if (this && this != rdev
)
8212 if (test_bit(Candidate
, &rdev
->flags
))
8214 if (rdev
->raid_disk
>= 0 &&
8215 !test_bit(In_sync
, &rdev
->flags
) &&
8216 !test_bit(Journal
, &rdev
->flags
) &&
8217 !test_bit(Faulty
, &rdev
->flags
))
8219 if (rdev
->raid_disk
>= 0)
8221 if (test_bit(Faulty
, &rdev
->flags
))
8223 if (!test_bit(Journal
, &rdev
->flags
)) {
8225 ! (rdev
->saved_raid_disk
>= 0 &&
8226 !test_bit(Bitmap_sync
, &rdev
->flags
)))
8229 rdev
->recovery_offset
= 0;
8232 hot_add_disk(mddev
, rdev
) == 0) {
8233 if (sysfs_link_rdev(mddev
, rdev
))
8234 /* failure here is OK */;
8235 if (!test_bit(Journal
, &rdev
->flags
))
8237 md_new_event(mddev
);
8238 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8243 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8247 static void md_start_sync(struct work_struct
*ws
)
8249 struct mddev
*mddev
= container_of(ws
, struct mddev
, del_work
);
8252 mddev
->sync_thread
= md_register_thread(md_do_sync
,
8255 if (!mddev
->sync_thread
) {
8256 if (!(mddev_is_clustered(mddev
) && ret
== -EAGAIN
))
8257 printk(KERN_ERR
"%s: could not start resync"
8260 /* leave the spares where they are, it shouldn't hurt */
8261 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8262 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8263 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8264 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8265 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8266 wake_up(&resync_wait
);
8267 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8269 if (mddev
->sysfs_action
)
8270 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8272 md_wakeup_thread(mddev
->sync_thread
);
8273 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8274 md_new_event(mddev
);
8278 * This routine is regularly called by all per-raid-array threads to
8279 * deal with generic issues like resync and super-block update.
8280 * Raid personalities that don't have a thread (linear/raid0) do not
8281 * need this as they never do any recovery or update the superblock.
8283 * It does not do any resync itself, but rather "forks" off other threads
8284 * to do that as needed.
8285 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8286 * "->recovery" and create a thread at ->sync_thread.
8287 * When the thread finishes it sets MD_RECOVERY_DONE
8288 * and wakeups up this thread which will reap the thread and finish up.
8289 * This thread also removes any faulty devices (with nr_pending == 0).
8291 * The overall approach is:
8292 * 1/ if the superblock needs updating, update it.
8293 * 2/ If a recovery thread is running, don't do anything else.
8294 * 3/ If recovery has finished, clean up, possibly marking spares active.
8295 * 4/ If there are any faulty devices, remove them.
8296 * 5/ If array is degraded, try to add spares devices
8297 * 6/ If array has spares or is not in-sync, start a resync thread.
8299 void md_check_recovery(struct mddev
*mddev
)
8301 if (mddev
->suspended
)
8305 bitmap_daemon_work(mddev
);
8307 if (signal_pending(current
)) {
8308 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
8309 printk(KERN_INFO
"md: %s in immediate safe mode\n",
8311 mddev
->safemode
= 2;
8313 flush_signals(current
);
8316 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
8319 (mddev
->flags
& MD_UPDATE_SB_FLAGS
& ~ (1<<MD_CHANGE_PENDING
)) ||
8320 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8321 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
8322 test_bit(MD_RELOAD_SB
, &mddev
->flags
) ||
8323 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
8324 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
8325 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
8329 if (mddev_trylock(mddev
)) {
8333 struct md_rdev
*rdev
;
8334 if (!mddev
->external
&& mddev
->in_sync
)
8335 /* 'Blocked' flag not needed as failed devices
8336 * will be recorded if array switched to read/write.
8337 * Leaving it set will prevent the device
8338 * from being removed.
8340 rdev_for_each(rdev
, mddev
)
8341 clear_bit(Blocked
, &rdev
->flags
);
8342 /* On a read-only array we can:
8343 * - remove failed devices
8344 * - add already-in_sync devices if the array itself
8346 * As we only add devices that are already in-sync,
8347 * we can activate the spares immediately.
8349 remove_and_add_spares(mddev
, NULL
);
8350 /* There is no thread, but we need to call
8351 * ->spare_active and clear saved_raid_disk
8353 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8354 md_reap_sync_thread(mddev
);
8355 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8356 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8357 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
8361 if (mddev_is_clustered(mddev
)) {
8362 struct md_rdev
*rdev
;
8363 /* kick the device if another node issued a
8366 rdev_for_each(rdev
, mddev
) {
8367 if (test_and_clear_bit(ClusterRemove
, &rdev
->flags
) &&
8368 rdev
->raid_disk
< 0)
8369 md_kick_rdev_from_array(rdev
);
8372 if (test_and_clear_bit(MD_RELOAD_SB
, &mddev
->flags
))
8373 md_reload_sb(mddev
, mddev
->good_device_nr
);
8376 if (!mddev
->external
) {
8378 spin_lock(&mddev
->lock
);
8379 if (mddev
->safemode
&&
8380 !atomic_read(&mddev
->writes_pending
) &&
8382 mddev
->recovery_cp
== MaxSector
) {
8385 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
8387 if (mddev
->safemode
== 1)
8388 mddev
->safemode
= 0;
8389 spin_unlock(&mddev
->lock
);
8391 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
8394 if (mddev
->flags
& MD_UPDATE_SB_FLAGS
)
8395 md_update_sb(mddev
, 0);
8397 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
8398 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
8399 /* resync/recovery still happening */
8400 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8403 if (mddev
->sync_thread
) {
8404 md_reap_sync_thread(mddev
);
8407 /* Set RUNNING before clearing NEEDED to avoid
8408 * any transients in the value of "sync_action".
8410 mddev
->curr_resync_completed
= 0;
8411 spin_lock(&mddev
->lock
);
8412 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8413 spin_unlock(&mddev
->lock
);
8414 /* Clear some bits that don't mean anything, but
8417 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
8418 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8420 if (!test_and_clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
8421 test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
8423 /* no recovery is running.
8424 * remove any failed drives, then
8425 * add spares if possible.
8426 * Spares are also removed and re-added, to allow
8427 * the personality to fail the re-add.
8430 if (mddev
->reshape_position
!= MaxSector
) {
8431 if (mddev
->pers
->check_reshape
== NULL
||
8432 mddev
->pers
->check_reshape(mddev
) != 0)
8433 /* Cannot proceed */
8435 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8436 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8437 } else if ((spares
= remove_and_add_spares(mddev
, NULL
))) {
8438 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8439 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8440 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8441 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8442 } else if (mddev
->recovery_cp
< MaxSector
) {
8443 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8444 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
8445 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
8446 /* nothing to be done ... */
8449 if (mddev
->pers
->sync_request
) {
8451 /* We are adding a device or devices to an array
8452 * which has the bitmap stored on all devices.
8453 * So make sure all bitmap pages get written
8455 bitmap_write_all(mddev
->bitmap
);
8457 INIT_WORK(&mddev
->del_work
, md_start_sync
);
8458 queue_work(md_misc_wq
, &mddev
->del_work
);
8462 if (!mddev
->sync_thread
) {
8463 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8464 wake_up(&resync_wait
);
8465 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
8467 if (mddev
->sysfs_action
)
8468 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8471 wake_up(&mddev
->sb_wait
);
8472 mddev_unlock(mddev
);
8475 EXPORT_SYMBOL(md_check_recovery
);
8477 void md_reap_sync_thread(struct mddev
*mddev
)
8479 struct md_rdev
*rdev
;
8481 /* resync has finished, collect result */
8482 md_unregister_thread(&mddev
->sync_thread
);
8483 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
8484 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
8486 /* activate any spares */
8487 if (mddev
->pers
->spare_active(mddev
)) {
8488 sysfs_notify(&mddev
->kobj
, NULL
,
8490 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
8493 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
8494 mddev
->pers
->finish_reshape
)
8495 mddev
->pers
->finish_reshape(mddev
);
8497 /* If array is no-longer degraded, then any saved_raid_disk
8498 * information must be scrapped.
8500 if (!mddev
->degraded
)
8501 rdev_for_each(rdev
, mddev
)
8502 rdev
->saved_raid_disk
= -1;
8504 md_update_sb(mddev
, 1);
8505 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
8506 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
8507 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
8508 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
8509 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
8510 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
8511 wake_up(&resync_wait
);
8512 /* flag recovery needed just to double check */
8513 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8514 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
8515 md_new_event(mddev
);
8516 if (mddev
->event_work
.func
)
8517 queue_work(md_misc_wq
, &mddev
->event_work
);
8519 EXPORT_SYMBOL(md_reap_sync_thread
);
8521 void md_wait_for_blocked_rdev(struct md_rdev
*rdev
, struct mddev
*mddev
)
8523 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8524 wait_event_timeout(rdev
->blocked_wait
,
8525 !test_bit(Blocked
, &rdev
->flags
) &&
8526 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
8527 msecs_to_jiffies(5000));
8528 rdev_dec_pending(rdev
, mddev
);
8530 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
8532 void md_finish_reshape(struct mddev
*mddev
)
8534 /* called be personality module when reshape completes. */
8535 struct md_rdev
*rdev
;
8537 rdev_for_each(rdev
, mddev
) {
8538 if (rdev
->data_offset
> rdev
->new_data_offset
)
8539 rdev
->sectors
+= rdev
->data_offset
- rdev
->new_data_offset
;
8541 rdev
->sectors
-= rdev
->new_data_offset
- rdev
->data_offset
;
8542 rdev
->data_offset
= rdev
->new_data_offset
;
8545 EXPORT_SYMBOL(md_finish_reshape
);
8547 /* Bad block management */
8549 /* Returns 1 on success, 0 on failure */
8550 int rdev_set_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8553 struct mddev
*mddev
= rdev
->mddev
;
8556 s
+= rdev
->new_data_offset
;
8558 s
+= rdev
->data_offset
;
8559 rv
= badblocks_set(&rdev
->badblocks
, s
, sectors
, 0);
8561 /* Make sure they get written out promptly */
8562 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
8563 set_mask_bits(&mddev
->flags
, 0,
8564 BIT(MD_CHANGE_CLEAN
) | BIT(MD_CHANGE_PENDING
));
8565 md_wakeup_thread(rdev
->mddev
->thread
);
8570 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
8572 int rdev_clear_badblocks(struct md_rdev
*rdev
, sector_t s
, int sectors
,
8576 s
+= rdev
->new_data_offset
;
8578 s
+= rdev
->data_offset
;
8579 return badblocks_clear(&rdev
->badblocks
,
8582 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
8584 static int md_notify_reboot(struct notifier_block
*this,
8585 unsigned long code
, void *x
)
8587 struct list_head
*tmp
;
8588 struct mddev
*mddev
;
8591 for_each_mddev(mddev
, tmp
) {
8592 if (mddev_trylock(mddev
)) {
8594 __md_stop_writes(mddev
);
8595 if (mddev
->persistent
)
8596 mddev
->safemode
= 2;
8597 mddev_unlock(mddev
);
8602 * certain more exotic SCSI devices are known to be
8603 * volatile wrt too early system reboots. While the
8604 * right place to handle this issue is the given
8605 * driver, we do want to have a safe RAID driver ...
8613 static struct notifier_block md_notifier
= {
8614 .notifier_call
= md_notify_reboot
,
8616 .priority
= INT_MAX
, /* before any real devices */
8619 static void md_geninit(void)
8621 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8623 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8626 static int __init
md_init(void)
8630 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8634 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8638 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8641 if ((ret
= register_blkdev(0, "mdp")) < 0)
8645 blk_register_region(MKDEV(MD_MAJOR
, 0), 512, THIS_MODULE
,
8646 md_probe
, NULL
, NULL
);
8647 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8648 md_probe
, NULL
, NULL
);
8650 register_reboot_notifier(&md_notifier
);
8651 raid_table_header
= register_sysctl_table(raid_root_table
);
8657 unregister_blkdev(MD_MAJOR
, "md");
8659 destroy_workqueue(md_misc_wq
);
8661 destroy_workqueue(md_wq
);
8666 static void check_sb_changes(struct mddev
*mddev
, struct md_rdev
*rdev
)
8668 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
8669 struct md_rdev
*rdev2
;
8671 char b
[BDEVNAME_SIZE
];
8673 /* Check for change of roles in the active devices */
8674 rdev_for_each(rdev2
, mddev
) {
8675 if (test_bit(Faulty
, &rdev2
->flags
))
8678 /* Check if the roles changed */
8679 role
= le16_to_cpu(sb
->dev_roles
[rdev2
->desc_nr
]);
8681 if (test_bit(Candidate
, &rdev2
->flags
)) {
8682 if (role
== 0xfffe) {
8683 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2
->bdev
,b
));
8684 md_kick_rdev_from_array(rdev2
);
8688 clear_bit(Candidate
, &rdev2
->flags
);
8691 if (role
!= rdev2
->raid_disk
) {
8693 if (rdev2
->raid_disk
== -1 && role
!= 0xffff) {
8694 rdev2
->saved_raid_disk
= role
;
8695 ret
= remove_and_add_spares(mddev
, rdev2
);
8696 pr_info("Activated spare: %s\n",
8697 bdevname(rdev2
->bdev
,b
));
8698 /* wakeup mddev->thread here, so array could
8699 * perform resync with the new activated disk */
8700 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
8701 md_wakeup_thread(mddev
->thread
);
8705 * We just want to do the minimum to mark the disk
8706 * as faulty. The recovery is performed by the
8707 * one who initiated the error.
8709 if ((role
== 0xfffe) || (role
== 0xfffd)) {
8710 md_error(mddev
, rdev2
);
8711 clear_bit(Blocked
, &rdev2
->flags
);
8716 if (mddev
->raid_disks
!= le32_to_cpu(sb
->raid_disks
))
8717 update_raid_disks(mddev
, le32_to_cpu(sb
->raid_disks
));
8719 /* Finally set the event to be up to date */
8720 mddev
->events
= le64_to_cpu(sb
->events
);
8723 static int read_rdev(struct mddev
*mddev
, struct md_rdev
*rdev
)
8726 struct page
*swapout
= rdev
->sb_page
;
8727 struct mdp_superblock_1
*sb
;
8729 /* Store the sb page of the rdev in the swapout temporary
8730 * variable in case we err in the future
8732 rdev
->sb_page
= NULL
;
8733 alloc_disk_sb(rdev
);
8734 ClearPageUptodate(rdev
->sb_page
);
8735 rdev
->sb_loaded
= 0;
8736 err
= super_types
[mddev
->major_version
].load_super(rdev
, NULL
, mddev
->minor_version
);
8739 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8740 __func__
, __LINE__
, rdev
->desc_nr
, err
);
8741 put_page(rdev
->sb_page
);
8742 rdev
->sb_page
= swapout
;
8743 rdev
->sb_loaded
= 1;
8747 sb
= page_address(rdev
->sb_page
);
8748 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8752 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
))
8753 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
8755 /* The other node finished recovery, call spare_active to set
8756 * device In_sync and mddev->degraded
8758 if (rdev
->recovery_offset
== MaxSector
&&
8759 !test_bit(In_sync
, &rdev
->flags
) &&
8760 mddev
->pers
->spare_active(mddev
))
8761 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
8767 void md_reload_sb(struct mddev
*mddev
, int nr
)
8769 struct md_rdev
*rdev
;
8773 rdev_for_each_rcu(rdev
, mddev
) {
8774 if (rdev
->desc_nr
== nr
)
8778 if (!rdev
|| rdev
->desc_nr
!= nr
) {
8779 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__
, __LINE__
, nr
);
8783 err
= read_rdev(mddev
, rdev
);
8787 check_sb_changes(mddev
, rdev
);
8789 /* Read all rdev's to update recovery_offset */
8790 rdev_for_each_rcu(rdev
, mddev
)
8791 read_rdev(mddev
, rdev
);
8793 EXPORT_SYMBOL(md_reload_sb
);
8798 * Searches all registered partitions for autorun RAID arrays
8802 static LIST_HEAD(all_detected_devices
);
8803 struct detected_devices_node
{
8804 struct list_head list
;
8808 void md_autodetect_dev(dev_t dev
)
8810 struct detected_devices_node
*node_detected_dev
;
8812 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8813 if (node_detected_dev
) {
8814 node_detected_dev
->dev
= dev
;
8815 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8817 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8818 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8822 static void autostart_arrays(int part
)
8824 struct md_rdev
*rdev
;
8825 struct detected_devices_node
*node_detected_dev
;
8827 int i_scanned
, i_passed
;
8832 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8834 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8836 node_detected_dev
= list_entry(all_detected_devices
.next
,
8837 struct detected_devices_node
, list
);
8838 list_del(&node_detected_dev
->list
);
8839 dev
= node_detected_dev
->dev
;
8840 kfree(node_detected_dev
);
8841 rdev
= md_import_device(dev
,0, 90);
8845 if (test_bit(Faulty
, &rdev
->flags
))
8848 set_bit(AutoDetected
, &rdev
->flags
);
8849 list_add(&rdev
->same_set
, &pending_raid_disks
);
8853 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8854 i_scanned
, i_passed
);
8856 autorun_devices(part
);
8859 #endif /* !MODULE */
8861 static __exit
void md_exit(void)
8863 struct mddev
*mddev
;
8864 struct list_head
*tmp
;
8867 blk_unregister_region(MKDEV(MD_MAJOR
,0), 512);
8868 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8870 unregister_blkdev(MD_MAJOR
,"md");
8871 unregister_blkdev(mdp_major
, "mdp");
8872 unregister_reboot_notifier(&md_notifier
);
8873 unregister_sysctl_table(raid_table_header
);
8875 /* We cannot unload the modules while some process is
8876 * waiting for us in select() or poll() - wake them up
8879 while (waitqueue_active(&md_event_waiters
)) {
8880 /* not safe to leave yet */
8881 wake_up(&md_event_waiters
);
8885 remove_proc_entry("mdstat", NULL
);
8887 for_each_mddev(mddev
, tmp
) {
8888 export_array(mddev
);
8889 mddev
->hold_active
= 0;
8891 destroy_workqueue(md_misc_wq
);
8892 destroy_workqueue(md_wq
);
8895 subsys_initcall(md_init
);
8896 module_exit(md_exit
)
8898 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8900 return sprintf(buffer
, "%d", start_readonly
);
8902 static int set_ro(const char *val
, struct kernel_param
*kp
)
8904 return kstrtouint(val
, 10, (unsigned int *)&start_readonly
);
8907 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8908 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8909 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8911 MODULE_LICENSE("GPL");
8912 MODULE_DESCRIPTION("MD RAID framework");
8914 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);