2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
61 static void autostart_arrays(int part
);
64 static LIST_HEAD(pers_list
);
65 static DEFINE_SPINLOCK(pers_lock
);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait
);
70 static struct workqueue_struct
*md_wq
;
71 static struct workqueue_struct
*md_misc_wq
;
73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min
= 1000;
95 static int sysctl_speed_limit_max
= 200000;
96 static inline int speed_min(mddev_t
*mddev
)
98 return mddev
->sync_speed_min
?
99 mddev
->sync_speed_min
: sysctl_speed_limit_min
;
102 static inline int speed_max(mddev_t
*mddev
)
104 return mddev
->sync_speed_max
?
105 mddev
->sync_speed_max
: sysctl_speed_limit_max
;
108 static struct ctl_table_header
*raid_table_header
;
110 static ctl_table raid_table
[] = {
112 .procname
= "speed_limit_min",
113 .data
= &sysctl_speed_limit_min
,
114 .maxlen
= sizeof(int),
115 .mode
= S_IRUGO
|S_IWUSR
,
116 .proc_handler
= proc_dointvec
,
119 .procname
= "speed_limit_max",
120 .data
= &sysctl_speed_limit_max
,
121 .maxlen
= sizeof(int),
122 .mode
= S_IRUGO
|S_IWUSR
,
123 .proc_handler
= proc_dointvec
,
128 static ctl_table raid_dir_table
[] = {
132 .mode
= S_IRUGO
|S_IXUGO
,
138 static ctl_table raid_root_table
[] = {
143 .child
= raid_dir_table
,
148 static const struct block_device_operations md_fops
;
150 static int start_readonly
;
153 * like bio_clone, but with a local bio set
156 static void mddev_bio_destructor(struct bio
*bio
)
158 mddev_t
*mddev
, **mddevp
;
163 bio_free(bio
, mddev
->bio_set
);
166 struct bio
*bio_alloc_mddev(gfp_t gfp_mask
, int nr_iovecs
,
172 if (!mddev
|| !mddev
->bio_set
)
173 return bio_alloc(gfp_mask
, nr_iovecs
);
175 b
= bio_alloc_bioset(gfp_mask
, nr_iovecs
,
181 b
->bi_destructor
= mddev_bio_destructor
;
184 EXPORT_SYMBOL_GPL(bio_alloc_mddev
);
186 struct bio
*bio_clone_mddev(struct bio
*bio
, gfp_t gfp_mask
,
192 if (!mddev
|| !mddev
->bio_set
)
193 return bio_clone(bio
, gfp_mask
);
195 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
,
201 b
->bi_destructor
= mddev_bio_destructor
;
203 if (bio_integrity(bio
)) {
206 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, mddev
->bio_set
);
216 EXPORT_SYMBOL_GPL(bio_clone_mddev
);
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
229 static atomic_t md_event_count
;
230 void md_new_event(mddev_t
*mddev
)
232 atomic_inc(&md_event_count
);
233 wake_up(&md_event_waiters
);
235 EXPORT_SYMBOL_GPL(md_new_event
);
237 /* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
240 static void md_new_event_inintr(mddev_t
*mddev
)
242 atomic_inc(&md_event_count
);
243 wake_up(&md_event_waiters
);
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
250 static LIST_HEAD(all_mddevs
);
251 static DEFINE_SPINLOCK(all_mddevs_lock
);
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
261 #define for_each_mddev(mddev,tmp) \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
277 /* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
284 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
286 const int rw
= bio_data_dir(bio
);
287 mddev_t
*mddev
= q
->queuedata
;
291 if (mddev
== NULL
|| mddev
->pers
== NULL
) {
296 if (mddev
->suspended
) {
299 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
300 TASK_UNINTERRUPTIBLE
);
301 if (!mddev
->suspended
)
307 finish_wait(&mddev
->sb_wait
, &__wait
);
309 atomic_inc(&mddev
->active_io
);
312 rv
= mddev
->pers
->make_request(mddev
, bio
);
314 cpu
= part_stat_lock();
315 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
316 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
320 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
321 wake_up(&mddev
->sb_wait
);
326 /* mddev_suspend makes sure no new requests are submitted
327 * to the device, and that any requests that have been submitted
328 * are completely handled.
329 * Once ->stop is called and completes, the module will be completely
332 void mddev_suspend(mddev_t
*mddev
)
334 BUG_ON(mddev
->suspended
);
335 mddev
->suspended
= 1;
337 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
338 mddev
->pers
->quiesce(mddev
, 1);
340 EXPORT_SYMBOL_GPL(mddev_suspend
);
342 void mddev_resume(mddev_t
*mddev
)
344 mddev
->suspended
= 0;
345 wake_up(&mddev
->sb_wait
);
346 mddev
->pers
->quiesce(mddev
, 0);
348 EXPORT_SYMBOL_GPL(mddev_resume
);
350 int mddev_congested(mddev_t
*mddev
, int bits
)
352 return mddev
->suspended
;
354 EXPORT_SYMBOL(mddev_congested
);
357 * Generic flush handling for md
360 static void md_end_flush(struct bio
*bio
, int err
)
362 mdk_rdev_t
*rdev
= bio
->bi_private
;
363 mddev_t
*mddev
= rdev
->mddev
;
365 rdev_dec_pending(rdev
, mddev
);
367 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
368 /* The pre-request flush has finished */
369 queue_work(md_wq
, &mddev
->flush_work
);
374 static void submit_flushes(mddev_t
*mddev
)
379 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
380 if (rdev
->raid_disk
>= 0 &&
381 !test_bit(Faulty
, &rdev
->flags
)) {
382 /* Take two references, one is dropped
383 * when request finishes, one after
384 * we reclaim rcu_read_lock
387 atomic_inc(&rdev
->nr_pending
);
388 atomic_inc(&rdev
->nr_pending
);
390 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
391 bi
->bi_end_io
= md_end_flush
;
392 bi
->bi_private
= rdev
;
393 bi
->bi_bdev
= rdev
->bdev
;
394 atomic_inc(&mddev
->flush_pending
);
395 submit_bio(WRITE_FLUSH
, bi
);
397 rdev_dec_pending(rdev
, mddev
);
402 static void md_submit_flush_data(struct work_struct
*ws
)
404 mddev_t
*mddev
= container_of(ws
, mddev_t
, flush_work
);
405 struct bio
*bio
= mddev
->flush_bio
;
407 atomic_set(&mddev
->flush_pending
, 1);
409 if (bio
->bi_size
== 0)
410 /* an empty barrier - all done */
413 bio
->bi_rw
&= ~REQ_FLUSH
;
414 if (mddev
->pers
->make_request(mddev
, bio
))
415 generic_make_request(bio
);
417 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
418 mddev
->flush_bio
= NULL
;
419 wake_up(&mddev
->sb_wait
);
423 void md_flush_request(mddev_t
*mddev
, struct bio
*bio
)
425 spin_lock_irq(&mddev
->write_lock
);
426 wait_event_lock_irq(mddev
->sb_wait
,
428 mddev
->write_lock
, /*nothing*/);
429 mddev
->flush_bio
= bio
;
430 spin_unlock_irq(&mddev
->write_lock
);
432 atomic_set(&mddev
->flush_pending
, 1);
433 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
435 submit_flushes(mddev
);
437 if (atomic_dec_and_test(&mddev
->flush_pending
))
438 queue_work(md_wq
, &mddev
->flush_work
);
440 EXPORT_SYMBOL(md_flush_request
);
442 /* Support for plugging.
443 * This mirrors the plugging support in request_queue, but does not
444 * require having a whole queue
446 static void plugger_work(struct work_struct
*work
)
448 struct plug_handle
*plug
=
449 container_of(work
, struct plug_handle
, unplug_work
);
450 plug
->unplug_fn(plug
);
452 static void plugger_timeout(unsigned long data
)
454 struct plug_handle
*plug
= (void *)data
;
455 kblockd_schedule_work(NULL
, &plug
->unplug_work
);
457 void plugger_init(struct plug_handle
*plug
,
458 void (*unplug_fn
)(struct plug_handle
*))
460 plug
->unplug_flag
= 0;
461 plug
->unplug_fn
= unplug_fn
;
462 init_timer(&plug
->unplug_timer
);
463 plug
->unplug_timer
.function
= plugger_timeout
;
464 plug
->unplug_timer
.data
= (unsigned long)plug
;
465 INIT_WORK(&plug
->unplug_work
, plugger_work
);
467 EXPORT_SYMBOL_GPL(plugger_init
);
469 void plugger_set_plug(struct plug_handle
*plug
)
471 if (!test_and_set_bit(PLUGGED_FLAG
, &plug
->unplug_flag
))
472 mod_timer(&plug
->unplug_timer
, jiffies
+ msecs_to_jiffies(3)+1);
474 EXPORT_SYMBOL_GPL(plugger_set_plug
);
476 int plugger_remove_plug(struct plug_handle
*plug
)
478 if (test_and_clear_bit(PLUGGED_FLAG
, &plug
->unplug_flag
)) {
479 del_timer(&plug
->unplug_timer
);
484 EXPORT_SYMBOL_GPL(plugger_remove_plug
);
487 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
489 atomic_inc(&mddev
->active
);
493 static void mddev_delayed_delete(struct work_struct
*ws
);
495 static void mddev_put(mddev_t
*mddev
)
497 struct bio_set
*bs
= NULL
;
499 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
501 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
502 mddev
->ctime
== 0 && !mddev
->hold_active
) {
503 /* Array is not configured at all, and not held active,
505 list_del(&mddev
->all_mddevs
);
507 mddev
->bio_set
= NULL
;
508 if (mddev
->gendisk
) {
509 /* We did a probe so need to clean up. Call
510 * queue_work inside the spinlock so that
511 * flush_workqueue() after mddev_find will
512 * succeed in waiting for the work to be done.
514 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
515 queue_work(md_misc_wq
, &mddev
->del_work
);
519 spin_unlock(&all_mddevs_lock
);
524 void mddev_init(mddev_t
*mddev
)
526 mutex_init(&mddev
->open_mutex
);
527 mutex_init(&mddev
->reconfig_mutex
);
528 mutex_init(&mddev
->bitmap_info
.mutex
);
529 INIT_LIST_HEAD(&mddev
->disks
);
530 INIT_LIST_HEAD(&mddev
->all_mddevs
);
531 init_timer(&mddev
->safemode_timer
);
532 atomic_set(&mddev
->active
, 1);
533 atomic_set(&mddev
->openers
, 0);
534 atomic_set(&mddev
->active_io
, 0);
535 spin_lock_init(&mddev
->write_lock
);
536 atomic_set(&mddev
->flush_pending
, 0);
537 init_waitqueue_head(&mddev
->sb_wait
);
538 init_waitqueue_head(&mddev
->recovery_wait
);
539 mddev
->reshape_position
= MaxSector
;
540 mddev
->resync_min
= 0;
541 mddev
->resync_max
= MaxSector
;
542 mddev
->level
= LEVEL_NONE
;
544 EXPORT_SYMBOL_GPL(mddev_init
);
546 static mddev_t
* mddev_find(dev_t unit
)
548 mddev_t
*mddev
, *new = NULL
;
551 spin_lock(&all_mddevs_lock
);
554 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
555 if (mddev
->unit
== unit
) {
557 spin_unlock(&all_mddevs_lock
);
563 list_add(&new->all_mddevs
, &all_mddevs
);
564 spin_unlock(&all_mddevs_lock
);
565 new->hold_active
= UNTIL_IOCTL
;
569 /* find an unused unit number */
570 static int next_minor
= 512;
571 int start
= next_minor
;
575 dev
= MKDEV(MD_MAJOR
, next_minor
);
577 if (next_minor
> MINORMASK
)
579 if (next_minor
== start
) {
580 /* Oh dear, all in use. */
581 spin_unlock(&all_mddevs_lock
);
587 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
588 if (mddev
->unit
== dev
) {
594 new->md_minor
= MINOR(dev
);
595 new->hold_active
= UNTIL_STOP
;
596 list_add(&new->all_mddevs
, &all_mddevs
);
597 spin_unlock(&all_mddevs_lock
);
600 spin_unlock(&all_mddevs_lock
);
602 new = kzalloc(sizeof(*new), GFP_KERNEL
);
607 if (MAJOR(unit
) == MD_MAJOR
)
608 new->md_minor
= MINOR(unit
);
610 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
617 static inline int mddev_lock(mddev_t
* mddev
)
619 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
622 static inline int mddev_is_locked(mddev_t
*mddev
)
624 return mutex_is_locked(&mddev
->reconfig_mutex
);
627 static inline int mddev_trylock(mddev_t
* mddev
)
629 return mutex_trylock(&mddev
->reconfig_mutex
);
632 static struct attribute_group md_redundancy_group
;
634 static void mddev_unlock(mddev_t
* mddev
)
636 if (mddev
->to_remove
) {
637 /* These cannot be removed under reconfig_mutex as
638 * an access to the files will try to take reconfig_mutex
639 * while holding the file unremovable, which leads to
641 * So hold set sysfs_active while the remove in happeing,
642 * and anything else which might set ->to_remove or my
643 * otherwise change the sysfs namespace will fail with
644 * -EBUSY if sysfs_active is still set.
645 * We set sysfs_active under reconfig_mutex and elsewhere
646 * test it under the same mutex to ensure its correct value
649 struct attribute_group
*to_remove
= mddev
->to_remove
;
650 mddev
->to_remove
= NULL
;
651 mddev
->sysfs_active
= 1;
652 mutex_unlock(&mddev
->reconfig_mutex
);
654 if (mddev
->kobj
.sd
) {
655 if (to_remove
!= &md_redundancy_group
)
656 sysfs_remove_group(&mddev
->kobj
, to_remove
);
657 if (mddev
->pers
== NULL
||
658 mddev
->pers
->sync_request
== NULL
) {
659 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
660 if (mddev
->sysfs_action
)
661 sysfs_put(mddev
->sysfs_action
);
662 mddev
->sysfs_action
= NULL
;
665 mddev
->sysfs_active
= 0;
667 mutex_unlock(&mddev
->reconfig_mutex
);
669 md_wakeup_thread(mddev
->thread
);
672 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
676 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
677 if (rdev
->desc_nr
== nr
)
683 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
687 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
688 if (rdev
->bdev
->bd_dev
== dev
)
694 static struct mdk_personality
*find_pers(int level
, char *clevel
)
696 struct mdk_personality
*pers
;
697 list_for_each_entry(pers
, &pers_list
, list
) {
698 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
700 if (strcmp(pers
->name
, clevel
)==0)
706 /* return the offset of the super block in 512byte sectors */
707 static inline sector_t
calc_dev_sboffset(struct block_device
*bdev
)
709 sector_t num_sectors
= bdev
->bd_inode
->i_size
/ 512;
710 return MD_NEW_SIZE_SECTORS(num_sectors
);
713 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
718 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
719 if (!rdev
->sb_page
) {
720 printk(KERN_ALERT
"md: out of memory.\n");
727 static void free_disk_sb(mdk_rdev_t
* rdev
)
730 put_page(rdev
->sb_page
);
732 rdev
->sb_page
= NULL
;
739 static void super_written(struct bio
*bio
, int error
)
741 mdk_rdev_t
*rdev
= bio
->bi_private
;
742 mddev_t
*mddev
= rdev
->mddev
;
744 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
745 printk("md: super_written gets error=%d, uptodate=%d\n",
746 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
747 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
748 md_error(mddev
, rdev
);
751 if (atomic_dec_and_test(&mddev
->pending_writes
))
752 wake_up(&mddev
->sb_wait
);
756 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
757 sector_t sector
, int size
, struct page
*page
)
759 /* write first size bytes of page to sector of rdev
760 * Increment mddev->pending_writes before returning
761 * and decrement it on completion, waking up sb_wait
762 * if zero is reached.
763 * If an error occurred, call md_error
765 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
767 bio
->bi_bdev
= rdev
->bdev
;
768 bio
->bi_sector
= sector
;
769 bio_add_page(bio
, page
, size
, 0);
770 bio
->bi_private
= rdev
;
771 bio
->bi_end_io
= super_written
;
773 atomic_inc(&mddev
->pending_writes
);
774 submit_bio(REQ_WRITE
| REQ_SYNC
| REQ_UNPLUG
| REQ_FLUSH
| REQ_FUA
,
778 void md_super_wait(mddev_t
*mddev
)
780 /* wait for all superblock writes that were scheduled to complete */
783 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
784 if (atomic_read(&mddev
->pending_writes
)==0)
788 finish_wait(&mddev
->sb_wait
, &wq
);
791 static void bi_complete(struct bio
*bio
, int error
)
793 complete((struct completion
*)bio
->bi_private
);
796 int sync_page_io(mdk_rdev_t
*rdev
, sector_t sector
, int size
,
797 struct page
*page
, int rw
)
799 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
800 struct completion event
;
803 rw
|= REQ_SYNC
| REQ_UNPLUG
;
805 bio
->bi_bdev
= rdev
->bdev
;
806 bio
->bi_sector
= sector
;
807 bio_add_page(bio
, page
, size
, 0);
808 init_completion(&event
);
809 bio
->bi_private
= &event
;
810 bio
->bi_end_io
= bi_complete
;
812 wait_for_completion(&event
);
814 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
818 EXPORT_SYMBOL_GPL(sync_page_io
);
820 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
822 char b
[BDEVNAME_SIZE
];
823 if (!rdev
->sb_page
) {
831 if (!sync_page_io(rdev
, rdev
->sb_start
, size
, rdev
->sb_page
, READ
))
837 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
838 bdevname(rdev
->bdev
,b
));
842 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
844 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
845 sb1
->set_uuid1
== sb2
->set_uuid1
&&
846 sb1
->set_uuid2
== sb2
->set_uuid2
&&
847 sb1
->set_uuid3
== sb2
->set_uuid3
;
850 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
853 mdp_super_t
*tmp1
, *tmp2
;
855 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
856 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
858 if (!tmp1
|| !tmp2
) {
860 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
868 * nr_disks is not constant
873 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
881 static u32
md_csum_fold(u32 csum
)
883 csum
= (csum
& 0xffff) + (csum
>> 16);
884 return (csum
& 0xffff) + (csum
>> 16);
887 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
890 u32
*sb32
= (u32
*)sb
;
892 unsigned int disk_csum
, csum
;
894 disk_csum
= sb
->sb_csum
;
897 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
899 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
903 /* This used to use csum_partial, which was wrong for several
904 * reasons including that different results are returned on
905 * different architectures. It isn't critical that we get exactly
906 * the same return value as before (we always csum_fold before
907 * testing, and that removes any differences). However as we
908 * know that csum_partial always returned a 16bit value on
909 * alphas, do a fold to maximise conformity to previous behaviour.
911 sb
->sb_csum
= md_csum_fold(disk_csum
);
913 sb
->sb_csum
= disk_csum
;
920 * Handle superblock details.
921 * We want to be able to handle multiple superblock formats
922 * so we have a common interface to them all, and an array of
923 * different handlers.
924 * We rely on user-space to write the initial superblock, and support
925 * reading and updating of superblocks.
926 * Interface methods are:
927 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
928 * loads and validates a superblock on dev.
929 * if refdev != NULL, compare superblocks on both devices
931 * 0 - dev has a superblock that is compatible with refdev
932 * 1 - dev has a superblock that is compatible and newer than refdev
933 * so dev should be used as the refdev in future
934 * -EINVAL superblock incompatible or invalid
935 * -othererror e.g. -EIO
937 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
938 * Verify that dev is acceptable into mddev.
939 * The first time, mddev->raid_disks will be 0, and data from
940 * dev should be merged in. Subsequent calls check that dev
941 * is new enough. Return 0 or -EINVAL
943 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
944 * Update the superblock for rdev with data in mddev
945 * This does not write to disc.
951 struct module
*owner
;
952 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
954 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
955 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
956 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
957 sector_t num_sectors
);
961 * Check that the given mddev has no bitmap.
963 * This function is called from the run method of all personalities that do not
964 * support bitmaps. It prints an error message and returns non-zero if mddev
965 * has a bitmap. Otherwise, it returns 0.
968 int md_check_no_bitmap(mddev_t
*mddev
)
970 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
972 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
973 mdname(mddev
), mddev
->pers
->name
);
976 EXPORT_SYMBOL(md_check_no_bitmap
);
979 * load_super for 0.90.0
981 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
983 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
988 * Calculate the position of the superblock (512byte sectors),
989 * it's at the end of the disk.
991 * It also happens to be a multiple of 4Kb.
993 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
995 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1000 bdevname(rdev
->bdev
, b
);
1001 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1003 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1004 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1009 if (sb
->major_version
!= 0 ||
1010 sb
->minor_version
< 90 ||
1011 sb
->minor_version
> 91) {
1012 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1013 sb
->major_version
, sb
->minor_version
,
1018 if (sb
->raid_disks
<= 0)
1021 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1022 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1027 rdev
->preferred_minor
= sb
->md_minor
;
1028 rdev
->data_offset
= 0;
1029 rdev
->sb_size
= MD_SB_BYTES
;
1031 if (sb
->level
== LEVEL_MULTIPATH
)
1034 rdev
->desc_nr
= sb
->this_disk
.number
;
1040 mdp_super_t
*refsb
= (mdp_super_t
*)page_address(refdev
->sb_page
);
1041 if (!uuid_equal(refsb
, sb
)) {
1042 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1043 b
, bdevname(refdev
->bdev
,b2
));
1046 if (!sb_equal(refsb
, sb
)) {
1047 printk(KERN_WARNING
"md: %s has same UUID"
1048 " but different superblock to %s\n",
1049 b
, bdevname(refdev
->bdev
, b2
));
1053 ev2
= md_event(refsb
);
1059 rdev
->sectors
= rdev
->sb_start
;
1061 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
1062 /* "this cannot possibly happen" ... */
1070 * validate_super for 0.90.0
1072 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1075 mdp_super_t
*sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1076 __u64 ev1
= md_event(sb
);
1078 rdev
->raid_disk
= -1;
1079 clear_bit(Faulty
, &rdev
->flags
);
1080 clear_bit(In_sync
, &rdev
->flags
);
1081 clear_bit(WriteMostly
, &rdev
->flags
);
1083 if (mddev
->raid_disks
== 0) {
1084 mddev
->major_version
= 0;
1085 mddev
->minor_version
= sb
->minor_version
;
1086 mddev
->patch_version
= sb
->patch_version
;
1087 mddev
->external
= 0;
1088 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1089 mddev
->ctime
= sb
->ctime
;
1090 mddev
->utime
= sb
->utime
;
1091 mddev
->level
= sb
->level
;
1092 mddev
->clevel
[0] = 0;
1093 mddev
->layout
= sb
->layout
;
1094 mddev
->raid_disks
= sb
->raid_disks
;
1095 mddev
->dev_sectors
= sb
->size
* 2;
1096 mddev
->events
= ev1
;
1097 mddev
->bitmap_info
.offset
= 0;
1098 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1100 if (mddev
->minor_version
>= 91) {
1101 mddev
->reshape_position
= sb
->reshape_position
;
1102 mddev
->delta_disks
= sb
->delta_disks
;
1103 mddev
->new_level
= sb
->new_level
;
1104 mddev
->new_layout
= sb
->new_layout
;
1105 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1107 mddev
->reshape_position
= MaxSector
;
1108 mddev
->delta_disks
= 0;
1109 mddev
->new_level
= mddev
->level
;
1110 mddev
->new_layout
= mddev
->layout
;
1111 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1114 if (sb
->state
& (1<<MD_SB_CLEAN
))
1115 mddev
->recovery_cp
= MaxSector
;
1117 if (sb
->events_hi
== sb
->cp_events_hi
&&
1118 sb
->events_lo
== sb
->cp_events_lo
) {
1119 mddev
->recovery_cp
= sb
->recovery_cp
;
1121 mddev
->recovery_cp
= 0;
1124 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1125 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1126 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1127 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1129 mddev
->max_disks
= MD_SB_DISKS
;
1131 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1132 mddev
->bitmap_info
.file
== NULL
)
1133 mddev
->bitmap_info
.offset
=
1134 mddev
->bitmap_info
.default_offset
;
1136 } else if (mddev
->pers
== NULL
) {
1137 /* Insist on good event counter while assembling, except
1138 * for spares (which don't need an event count) */
1140 if (sb
->disks
[rdev
->desc_nr
].state
& (
1141 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1142 if (ev1
< mddev
->events
)
1144 } else if (mddev
->bitmap
) {
1145 /* if adding to array with a bitmap, then we can accept an
1146 * older device ... but not too old.
1148 if (ev1
< mddev
->bitmap
->events_cleared
)
1151 if (ev1
< mddev
->events
)
1152 /* just a hot-add of a new device, leave raid_disk at -1 */
1156 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1157 desc
= sb
->disks
+ rdev
->desc_nr
;
1159 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1160 set_bit(Faulty
, &rdev
->flags
);
1161 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1162 desc->raid_disk < mddev->raid_disks */) {
1163 set_bit(In_sync
, &rdev
->flags
);
1164 rdev
->raid_disk
= desc
->raid_disk
;
1165 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1166 /* active but not in sync implies recovery up to
1167 * reshape position. We don't know exactly where
1168 * that is, so set to zero for now */
1169 if (mddev
->minor_version
>= 91) {
1170 rdev
->recovery_offset
= 0;
1171 rdev
->raid_disk
= desc
->raid_disk
;
1174 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1175 set_bit(WriteMostly
, &rdev
->flags
);
1176 } else /* MULTIPATH are always insync */
1177 set_bit(In_sync
, &rdev
->flags
);
1182 * sync_super for 0.90.0
1184 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1188 int next_spare
= mddev
->raid_disks
;
1191 /* make rdev->sb match mddev data..
1194 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1195 * 3/ any empty disks < next_spare become removed
1197 * disks[0] gets initialised to REMOVED because
1198 * we cannot be sure from other fields if it has
1199 * been initialised or not.
1202 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1204 rdev
->sb_size
= MD_SB_BYTES
;
1206 sb
= (mdp_super_t
*)page_address(rdev
->sb_page
);
1208 memset(sb
, 0, sizeof(*sb
));
1210 sb
->md_magic
= MD_SB_MAGIC
;
1211 sb
->major_version
= mddev
->major_version
;
1212 sb
->patch_version
= mddev
->patch_version
;
1213 sb
->gvalid_words
= 0; /* ignored */
1214 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1215 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1216 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1217 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1219 sb
->ctime
= mddev
->ctime
;
1220 sb
->level
= mddev
->level
;
1221 sb
->size
= mddev
->dev_sectors
/ 2;
1222 sb
->raid_disks
= mddev
->raid_disks
;
1223 sb
->md_minor
= mddev
->md_minor
;
1224 sb
->not_persistent
= 0;
1225 sb
->utime
= mddev
->utime
;
1227 sb
->events_hi
= (mddev
->events
>>32);
1228 sb
->events_lo
= (u32
)mddev
->events
;
1230 if (mddev
->reshape_position
== MaxSector
)
1231 sb
->minor_version
= 90;
1233 sb
->minor_version
= 91;
1234 sb
->reshape_position
= mddev
->reshape_position
;
1235 sb
->new_level
= mddev
->new_level
;
1236 sb
->delta_disks
= mddev
->delta_disks
;
1237 sb
->new_layout
= mddev
->new_layout
;
1238 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1240 mddev
->minor_version
= sb
->minor_version
;
1243 sb
->recovery_cp
= mddev
->recovery_cp
;
1244 sb
->cp_events_hi
= (mddev
->events
>>32);
1245 sb
->cp_events_lo
= (u32
)mddev
->events
;
1246 if (mddev
->recovery_cp
== MaxSector
)
1247 sb
->state
= (1<< MD_SB_CLEAN
);
1249 sb
->recovery_cp
= 0;
1251 sb
->layout
= mddev
->layout
;
1252 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1254 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1255 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1257 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1258 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1261 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1263 if (rdev2
->raid_disk
>= 0 &&
1264 sb
->minor_version
>= 91)
1265 /* we have nowhere to store the recovery_offset,
1266 * but if it is not below the reshape_position,
1267 * we can piggy-back on that.
1270 if (rdev2
->raid_disk
< 0 ||
1271 test_bit(Faulty
, &rdev2
->flags
))
1274 desc_nr
= rdev2
->raid_disk
;
1276 desc_nr
= next_spare
++;
1277 rdev2
->desc_nr
= desc_nr
;
1278 d
= &sb
->disks
[rdev2
->desc_nr
];
1280 d
->number
= rdev2
->desc_nr
;
1281 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1282 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1284 d
->raid_disk
= rdev2
->raid_disk
;
1286 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1287 if (test_bit(Faulty
, &rdev2
->flags
))
1288 d
->state
= (1<<MD_DISK_FAULTY
);
1289 else if (is_active
) {
1290 d
->state
= (1<<MD_DISK_ACTIVE
);
1291 if (test_bit(In_sync
, &rdev2
->flags
))
1292 d
->state
|= (1<<MD_DISK_SYNC
);
1300 if (test_bit(WriteMostly
, &rdev2
->flags
))
1301 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1303 /* now set the "removed" and "faulty" bits on any missing devices */
1304 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1305 mdp_disk_t
*d
= &sb
->disks
[i
];
1306 if (d
->state
== 0 && d
->number
== 0) {
1309 d
->state
= (1<<MD_DISK_REMOVED
);
1310 d
->state
|= (1<<MD_DISK_FAULTY
);
1314 sb
->nr_disks
= nr_disks
;
1315 sb
->active_disks
= active
;
1316 sb
->working_disks
= working
;
1317 sb
->failed_disks
= failed
;
1318 sb
->spare_disks
= spare
;
1320 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1321 sb
->sb_csum
= calc_sb_csum(sb
);
1325 * rdev_size_change for 0.90.0
1327 static unsigned long long
1328 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1330 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1331 return 0; /* component must fit device */
1332 if (rdev
->mddev
->bitmap_info
.offset
)
1333 return 0; /* can't move bitmap */
1334 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
1335 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1336 num_sectors
= rdev
->sb_start
;
1337 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1339 md_super_wait(rdev
->mddev
);
1340 return num_sectors
/ 2; /* kB for sysfs */
1345 * version 1 superblock
1348 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1352 unsigned long long newcsum
;
1353 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1354 __le32
*isuper
= (__le32
*)sb
;
1357 disk_csum
= sb
->sb_csum
;
1360 for (i
=0; size
>=4; size
-= 4 )
1361 newcsum
+= le32_to_cpu(*isuper
++);
1364 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1366 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1367 sb
->sb_csum
= disk_csum
;
1368 return cpu_to_le32(csum
);
1371 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1373 struct mdp_superblock_1
*sb
;
1376 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1380 * Calculate the position of the superblock in 512byte sectors.
1381 * It is always aligned to a 4K boundary and
1382 * depeding on minor_version, it can be:
1383 * 0: At least 8K, but less than 12K, from end of device
1384 * 1: At start of device
1385 * 2: 4K from start of device.
1387 switch(minor_version
) {
1389 sb_start
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1391 sb_start
&= ~(sector_t
)(4*2-1);
1402 rdev
->sb_start
= sb_start
;
1404 /* superblock is rarely larger than 1K, but it can be larger,
1405 * and it is safe to read 4k, so we do that
1407 ret
= read_disk_sb(rdev
, 4096);
1408 if (ret
) return ret
;
1411 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1413 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1414 sb
->major_version
!= cpu_to_le32(1) ||
1415 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1416 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1417 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1420 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1421 printk("md: invalid superblock checksum on %s\n",
1422 bdevname(rdev
->bdev
,b
));
1425 if (le64_to_cpu(sb
->data_size
) < 10) {
1426 printk("md: data_size too small on %s\n",
1427 bdevname(rdev
->bdev
,b
));
1431 rdev
->preferred_minor
= 0xffff;
1432 rdev
->data_offset
= le64_to_cpu(sb
->data_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 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1447 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1453 struct mdp_superblock_1
*refsb
=
1454 (struct mdp_superblock_1
*)page_address(refdev
->sb_page
);
1456 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1457 sb
->level
!= refsb
->level
||
1458 sb
->layout
!= refsb
->layout
||
1459 sb
->chunksize
!= refsb
->chunksize
) {
1460 printk(KERN_WARNING
"md: %s has strangely different"
1461 " superblock to %s\n",
1462 bdevname(rdev
->bdev
,b
),
1463 bdevname(refdev
->bdev
,b2
));
1466 ev1
= le64_to_cpu(sb
->events
);
1467 ev2
= le64_to_cpu(refsb
->events
);
1475 rdev
->sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
1476 le64_to_cpu(sb
->data_offset
);
1478 rdev
->sectors
= rdev
->sb_start
;
1479 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1481 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1482 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1487 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1489 struct mdp_superblock_1
*sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1490 __u64 ev1
= le64_to_cpu(sb
->events
);
1492 rdev
->raid_disk
= -1;
1493 clear_bit(Faulty
, &rdev
->flags
);
1494 clear_bit(In_sync
, &rdev
->flags
);
1495 clear_bit(WriteMostly
, &rdev
->flags
);
1497 if (mddev
->raid_disks
== 0) {
1498 mddev
->major_version
= 1;
1499 mddev
->patch_version
= 0;
1500 mddev
->external
= 0;
1501 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1502 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1503 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1504 mddev
->level
= le32_to_cpu(sb
->level
);
1505 mddev
->clevel
[0] = 0;
1506 mddev
->layout
= le32_to_cpu(sb
->layout
);
1507 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1508 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1509 mddev
->events
= ev1
;
1510 mddev
->bitmap_info
.offset
= 0;
1511 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1513 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1514 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1516 mddev
->max_disks
= (4096-256)/2;
1518 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1519 mddev
->bitmap_info
.file
== NULL
)
1520 mddev
->bitmap_info
.offset
=
1521 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1523 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1524 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1525 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1526 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1527 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1528 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1530 mddev
->reshape_position
= MaxSector
;
1531 mddev
->delta_disks
= 0;
1532 mddev
->new_level
= mddev
->level
;
1533 mddev
->new_layout
= mddev
->layout
;
1534 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1537 } else if (mddev
->pers
== NULL
) {
1538 /* Insist of good event counter while assembling, except for
1539 * spares (which don't need an event count) */
1541 if (rdev
->desc_nr
>= 0 &&
1542 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1543 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1544 if (ev1
< mddev
->events
)
1546 } else if (mddev
->bitmap
) {
1547 /* If adding to array with a bitmap, then we can accept an
1548 * older device, but not too old.
1550 if (ev1
< mddev
->bitmap
->events_cleared
)
1553 if (ev1
< mddev
->events
)
1554 /* just a hot-add of a new device, leave raid_disk at -1 */
1557 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1559 if (rdev
->desc_nr
< 0 ||
1560 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1564 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1566 case 0xffff: /* spare */
1568 case 0xfffe: /* faulty */
1569 set_bit(Faulty
, &rdev
->flags
);
1572 if ((le32_to_cpu(sb
->feature_map
) &
1573 MD_FEATURE_RECOVERY_OFFSET
))
1574 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1576 set_bit(In_sync
, &rdev
->flags
);
1577 rdev
->raid_disk
= role
;
1580 if (sb
->devflags
& WriteMostly1
)
1581 set_bit(WriteMostly
, &rdev
->flags
);
1582 } else /* MULTIPATH are always insync */
1583 set_bit(In_sync
, &rdev
->flags
);
1588 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1590 struct mdp_superblock_1
*sb
;
1593 /* make rdev->sb match mddev and rdev data. */
1595 sb
= (struct mdp_superblock_1
*)page_address(rdev
->sb_page
);
1597 sb
->feature_map
= 0;
1599 sb
->recovery_offset
= cpu_to_le64(0);
1600 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1601 memset(sb
->pad2
, 0, sizeof(sb
->pad2
));
1602 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1604 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1605 sb
->events
= cpu_to_le64(mddev
->events
);
1607 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1609 sb
->resync_offset
= cpu_to_le64(0);
1611 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1613 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1614 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1615 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1616 sb
->level
= cpu_to_le32(mddev
->level
);
1617 sb
->layout
= cpu_to_le32(mddev
->layout
);
1619 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1620 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1621 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1624 if (rdev
->raid_disk
>= 0 &&
1625 !test_bit(In_sync
, &rdev
->flags
)) {
1627 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1628 sb
->recovery_offset
=
1629 cpu_to_le64(rdev
->recovery_offset
);
1632 if (mddev
->reshape_position
!= MaxSector
) {
1633 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1634 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1635 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1636 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1637 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1638 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1642 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1643 if (rdev2
->desc_nr
+1 > max_dev
)
1644 max_dev
= rdev2
->desc_nr
+1;
1646 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1648 sb
->max_dev
= cpu_to_le32(max_dev
);
1649 rdev
->sb_size
= max_dev
* 2 + 256;
1650 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1651 if (rdev
->sb_size
& bmask
)
1652 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1654 max_dev
= le32_to_cpu(sb
->max_dev
);
1656 for (i
=0; i
<max_dev
;i
++)
1657 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1659 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1661 if (test_bit(Faulty
, &rdev2
->flags
))
1662 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1663 else if (test_bit(In_sync
, &rdev2
->flags
))
1664 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1665 else if (rdev2
->raid_disk
>= 0)
1666 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1668 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1671 sb
->sb_csum
= calc_sb_1_csum(sb
);
1674 static unsigned long long
1675 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1677 struct mdp_superblock_1
*sb
;
1678 sector_t max_sectors
;
1679 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1680 return 0; /* component must fit device */
1681 if (rdev
->sb_start
< rdev
->data_offset
) {
1682 /* minor versions 1 and 2; superblock before data */
1683 max_sectors
= rdev
->bdev
->bd_inode
->i_size
>> 9;
1684 max_sectors
-= rdev
->data_offset
;
1685 if (!num_sectors
|| num_sectors
> max_sectors
)
1686 num_sectors
= max_sectors
;
1687 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1688 /* minor version 0 with bitmap we can't move */
1691 /* minor version 0; superblock after data */
1693 sb_start
= (rdev
->bdev
->bd_inode
->i_size
>> 9) - 8*2;
1694 sb_start
&= ~(sector_t
)(4*2 - 1);
1695 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1696 if (!num_sectors
|| num_sectors
> max_sectors
)
1697 num_sectors
= max_sectors
;
1698 rdev
->sb_start
= sb_start
;
1700 sb
= (struct mdp_superblock_1
*) page_address(rdev
->sb_page
);
1701 sb
->data_size
= cpu_to_le64(num_sectors
);
1702 sb
->super_offset
= rdev
->sb_start
;
1703 sb
->sb_csum
= calc_sb_1_csum(sb
);
1704 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1706 md_super_wait(rdev
->mddev
);
1707 return num_sectors
/ 2; /* kB for sysfs */
1710 static struct super_type super_types
[] = {
1713 .owner
= THIS_MODULE
,
1714 .load_super
= super_90_load
,
1715 .validate_super
= super_90_validate
,
1716 .sync_super
= super_90_sync
,
1717 .rdev_size_change
= super_90_rdev_size_change
,
1721 .owner
= THIS_MODULE
,
1722 .load_super
= super_1_load
,
1723 .validate_super
= super_1_validate
,
1724 .sync_super
= super_1_sync
,
1725 .rdev_size_change
= super_1_rdev_size_change
,
1729 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1731 mdk_rdev_t
*rdev
, *rdev2
;
1734 rdev_for_each_rcu(rdev
, mddev1
)
1735 rdev_for_each_rcu(rdev2
, mddev2
)
1736 if (rdev
->bdev
->bd_contains
==
1737 rdev2
->bdev
->bd_contains
) {
1745 static LIST_HEAD(pending_raid_disks
);
1748 * Try to register data integrity profile for an mddev
1750 * This is called when an array is started and after a disk has been kicked
1751 * from the array. It only succeeds if all working and active component devices
1752 * are integrity capable with matching profiles.
1754 int md_integrity_register(mddev_t
*mddev
)
1756 mdk_rdev_t
*rdev
, *reference
= NULL
;
1758 if (list_empty(&mddev
->disks
))
1759 return 0; /* nothing to do */
1760 if (blk_get_integrity(mddev
->gendisk
))
1761 return 0; /* already registered */
1762 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1763 /* skip spares and non-functional disks */
1764 if (test_bit(Faulty
, &rdev
->flags
))
1766 if (rdev
->raid_disk
< 0)
1769 * If at least one rdev is not integrity capable, we can not
1770 * enable data integrity for the md device.
1772 if (!bdev_get_integrity(rdev
->bdev
))
1775 /* Use the first rdev as the reference */
1779 /* does this rdev's profile match the reference profile? */
1780 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1781 rdev
->bdev
->bd_disk
) < 0)
1785 * All component devices are integrity capable and have matching
1786 * profiles, register the common profile for the md device.
1788 if (blk_integrity_register(mddev
->gendisk
,
1789 bdev_get_integrity(reference
->bdev
)) != 0) {
1790 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1794 printk(KERN_NOTICE
"md: data integrity on %s enabled\n",
1798 EXPORT_SYMBOL(md_integrity_register
);
1800 /* Disable data integrity if non-capable/non-matching disk is being added */
1801 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1803 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1804 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1806 if (!bi_mddev
) /* nothing to do */
1808 if (rdev
->raid_disk
< 0) /* skip spares */
1810 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1811 rdev
->bdev
->bd_disk
) >= 0)
1813 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1814 blk_integrity_unregister(mddev
->gendisk
);
1816 EXPORT_SYMBOL(md_integrity_add_rdev
);
1818 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1820 char b
[BDEVNAME_SIZE
];
1830 /* prevent duplicates */
1831 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1834 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1835 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1836 rdev
->sectors
< mddev
->dev_sectors
)) {
1838 /* Cannot change size, so fail
1839 * If mddev->level <= 0, then we don't care
1840 * about aligning sizes (e.g. linear)
1842 if (mddev
->level
> 0)
1845 mddev
->dev_sectors
= rdev
->sectors
;
1848 /* Verify rdev->desc_nr is unique.
1849 * If it is -1, assign a free number, else
1850 * check number is not in use
1852 if (rdev
->desc_nr
< 0) {
1854 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1855 while (find_rdev_nr(mddev
, choice
))
1857 rdev
->desc_nr
= choice
;
1859 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1862 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1863 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1864 mdname(mddev
), mddev
->max_disks
);
1867 bdevname(rdev
->bdev
,b
);
1868 while ( (s
=strchr(b
, '/')) != NULL
)
1871 rdev
->mddev
= mddev
;
1872 printk(KERN_INFO
"md: bind<%s>\n", b
);
1874 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1877 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1878 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
1879 /* failure here is OK */;
1880 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
1882 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
1883 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
1885 /* May as well allow recovery to be retried once */
1886 mddev
->recovery_disabled
= 0;
1891 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
1896 static void md_delayed_delete(struct work_struct
*ws
)
1898 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
1899 kobject_del(&rdev
->kobj
);
1900 kobject_put(&rdev
->kobj
);
1903 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
1905 char b
[BDEVNAME_SIZE
];
1910 list_del_rcu(&rdev
->same_set
);
1911 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
1913 sysfs_remove_link(&rdev
->kobj
, "block");
1914 sysfs_put(rdev
->sysfs_state
);
1915 rdev
->sysfs_state
= NULL
;
1916 /* We need to delay this, otherwise we can deadlock when
1917 * writing to 'remove' to "dev/state". We also need
1918 * to delay it due to rcu usage.
1921 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
1922 kobject_get(&rdev
->kobj
);
1923 queue_work(md_misc_wq
, &rdev
->del_work
);
1927 * prevent the device from being mounted, repartitioned or
1928 * otherwise reused by a RAID array (or any other kernel
1929 * subsystem), by bd_claiming the device.
1931 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
1934 struct block_device
*bdev
;
1935 char b
[BDEVNAME_SIZE
];
1937 bdev
= open_by_devnum(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
1938 shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
1940 printk(KERN_ERR
"md: could not open %s.\n",
1941 __bdevname(dev
, b
));
1942 return PTR_ERR(bdev
);
1945 set_bit(AllReserved
, &rdev
->flags
);
1950 static void unlock_rdev(mdk_rdev_t
*rdev
)
1952 struct block_device
*bdev
= rdev
->bdev
;
1956 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
1959 void md_autodetect_dev(dev_t dev
);
1961 static void export_rdev(mdk_rdev_t
* rdev
)
1963 char b
[BDEVNAME_SIZE
];
1964 printk(KERN_INFO
"md: export_rdev(%s)\n",
1965 bdevname(rdev
->bdev
,b
));
1970 if (test_bit(AutoDetected
, &rdev
->flags
))
1971 md_autodetect_dev(rdev
->bdev
->bd_dev
);
1974 kobject_put(&rdev
->kobj
);
1977 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
1979 unbind_rdev_from_array(rdev
);
1983 static void export_array(mddev_t
*mddev
)
1985 mdk_rdev_t
*rdev
, *tmp
;
1987 rdev_for_each(rdev
, tmp
, mddev
) {
1992 kick_rdev_from_array(rdev
);
1994 if (!list_empty(&mddev
->disks
))
1996 mddev
->raid_disks
= 0;
1997 mddev
->major_version
= 0;
2000 static void print_desc(mdp_disk_t
*desc
)
2002 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2003 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2006 static void print_sb_90(mdp_super_t
*sb
)
2011 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2012 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2013 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2015 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2016 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2017 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2018 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2019 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2020 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2021 sb
->failed_disks
, sb
->spare_disks
,
2022 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2025 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2028 desc
= sb
->disks
+ i
;
2029 if (desc
->number
|| desc
->major
|| desc
->minor
||
2030 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2031 printk(" D %2d: ", i
);
2035 printk(KERN_INFO
"md: THIS: ");
2036 print_desc(&sb
->this_disk
);
2039 static void print_sb_1(struct mdp_superblock_1
*sb
)
2043 uuid
= sb
->set_uuid
;
2045 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2046 "md: Name: \"%s\" CT:%llu\n",
2047 le32_to_cpu(sb
->major_version
),
2048 le32_to_cpu(sb
->feature_map
),
2051 (unsigned long long)le64_to_cpu(sb
->ctime
)
2052 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2054 uuid
= sb
->device_uuid
;
2056 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2058 "md: Dev:%08x UUID: %pU\n"
2059 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2060 "md: (MaxDev:%u) \n",
2061 le32_to_cpu(sb
->level
),
2062 (unsigned long long)le64_to_cpu(sb
->size
),
2063 le32_to_cpu(sb
->raid_disks
),
2064 le32_to_cpu(sb
->layout
),
2065 le32_to_cpu(sb
->chunksize
),
2066 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2067 (unsigned long long)le64_to_cpu(sb
->data_size
),
2068 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2069 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2070 le32_to_cpu(sb
->dev_number
),
2073 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2074 (unsigned long long)le64_to_cpu(sb
->events
),
2075 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2076 le32_to_cpu(sb
->sb_csum
),
2077 le32_to_cpu(sb
->max_dev
)
2081 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2083 char b
[BDEVNAME_SIZE
];
2084 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2085 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2086 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2088 if (rdev
->sb_loaded
) {
2089 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2090 switch (major_version
) {
2092 print_sb_90((mdp_super_t
*)page_address(rdev
->sb_page
));
2095 print_sb_1((struct mdp_superblock_1
*)page_address(rdev
->sb_page
));
2099 printk(KERN_INFO
"md: no rdev superblock!\n");
2102 static void md_print_devices(void)
2104 struct list_head
*tmp
;
2107 char b
[BDEVNAME_SIZE
];
2110 printk("md: **********************************\n");
2111 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2112 printk("md: **********************************\n");
2113 for_each_mddev(mddev
, tmp
) {
2116 bitmap_print_sb(mddev
->bitmap
);
2118 printk("%s: ", mdname(mddev
));
2119 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2120 printk("<%s>", bdevname(rdev
->bdev
,b
));
2123 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2124 print_rdev(rdev
, mddev
->major_version
);
2126 printk("md: **********************************\n");
2131 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2133 /* Update each superblock (in-memory image), but
2134 * if we are allowed to, skip spares which already
2135 * have the right event counter, or have one earlier
2136 * (which would mean they aren't being marked as dirty
2137 * with the rest of the array)
2140 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2141 if (rdev
->sb_events
== mddev
->events
||
2143 rdev
->raid_disk
< 0 &&
2144 rdev
->sb_events
+1 == mddev
->events
)) {
2145 /* Don't update this superblock */
2146 rdev
->sb_loaded
= 2;
2148 super_types
[mddev
->major_version
].
2149 sync_super(mddev
, rdev
);
2150 rdev
->sb_loaded
= 1;
2155 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2162 /* First make sure individual recovery_offsets are correct */
2163 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2164 if (rdev
->raid_disk
>= 0 &&
2165 mddev
->delta_disks
>= 0 &&
2166 !test_bit(In_sync
, &rdev
->flags
) &&
2167 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2168 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2171 if (!mddev
->persistent
) {
2172 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2173 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2174 if (!mddev
->external
)
2175 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2176 wake_up(&mddev
->sb_wait
);
2180 spin_lock_irq(&mddev
->write_lock
);
2182 mddev
->utime
= get_seconds();
2184 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2186 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2187 /* just a clean<-> dirty transition, possibly leave spares alone,
2188 * though if events isn't the right even/odd, we will have to do
2194 if (mddev
->degraded
)
2195 /* If the array is degraded, then skipping spares is both
2196 * dangerous and fairly pointless.
2197 * Dangerous because a device that was removed from the array
2198 * might have a event_count that still looks up-to-date,
2199 * so it can be re-added without a resync.
2200 * Pointless because if there are any spares to skip,
2201 * then a recovery will happen and soon that array won't
2202 * be degraded any more and the spare can go back to sleep then.
2206 sync_req
= mddev
->in_sync
;
2208 /* If this is just a dirty<->clean transition, and the array is clean
2209 * and 'events' is odd, we can roll back to the previous clean state */
2211 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2212 && mddev
->can_decrease_events
2213 && mddev
->events
!= 1) {
2215 mddev
->can_decrease_events
= 0;
2217 /* otherwise we have to go forward and ... */
2219 mddev
->can_decrease_events
= nospares
;
2222 if (!mddev
->events
) {
2224 * oops, this 64-bit counter should never wrap.
2225 * Either we are in around ~1 trillion A.C., assuming
2226 * 1 reboot per second, or we have a bug:
2231 sync_sbs(mddev
, nospares
);
2232 spin_unlock_irq(&mddev
->write_lock
);
2235 "md: updating %s RAID superblock on device (in sync %d)\n",
2236 mdname(mddev
),mddev
->in_sync
);
2238 bitmap_update_sb(mddev
->bitmap
);
2239 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2240 char b
[BDEVNAME_SIZE
];
2241 dprintk(KERN_INFO
"md: ");
2242 if (rdev
->sb_loaded
!= 1)
2243 continue; /* no noise on spare devices */
2244 if (test_bit(Faulty
, &rdev
->flags
))
2245 dprintk("(skipping faulty ");
2247 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2248 if (!test_bit(Faulty
, &rdev
->flags
)) {
2249 md_super_write(mddev
,rdev
,
2250 rdev
->sb_start
, rdev
->sb_size
,
2252 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2253 bdevname(rdev
->bdev
,b
),
2254 (unsigned long long)rdev
->sb_start
);
2255 rdev
->sb_events
= mddev
->events
;
2259 if (mddev
->level
== LEVEL_MULTIPATH
)
2260 /* only need to write one superblock... */
2263 md_super_wait(mddev
);
2264 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2266 spin_lock_irq(&mddev
->write_lock
);
2267 if (mddev
->in_sync
!= sync_req
||
2268 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2269 /* have to write it out again */
2270 spin_unlock_irq(&mddev
->write_lock
);
2273 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2274 spin_unlock_irq(&mddev
->write_lock
);
2275 wake_up(&mddev
->sb_wait
);
2276 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2277 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2281 /* words written to sysfs files may, or may not, be \n terminated.
2282 * We want to accept with case. For this we use cmd_match.
2284 static int cmd_match(const char *cmd
, const char *str
)
2286 /* See if cmd, written into a sysfs file, matches
2287 * str. They must either be the same, or cmd can
2288 * have a trailing newline
2290 while (*cmd
&& *str
&& *cmd
== *str
) {
2301 struct rdev_sysfs_entry
{
2302 struct attribute attr
;
2303 ssize_t (*show
)(mdk_rdev_t
*, char *);
2304 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2308 state_show(mdk_rdev_t
*rdev
, char *page
)
2313 if (test_bit(Faulty
, &rdev
->flags
)) {
2314 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2317 if (test_bit(In_sync
, &rdev
->flags
)) {
2318 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2321 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2322 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2325 if (test_bit(Blocked
, &rdev
->flags
)) {
2326 len
+= sprintf(page
+len
, "%sblocked", sep
);
2329 if (!test_bit(Faulty
, &rdev
->flags
) &&
2330 !test_bit(In_sync
, &rdev
->flags
)) {
2331 len
+= sprintf(page
+len
, "%sspare", sep
);
2334 return len
+sprintf(page
+len
, "\n");
2338 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2341 * faulty - simulates and error
2342 * remove - disconnects the device
2343 * writemostly - sets write_mostly
2344 * -writemostly - clears write_mostly
2345 * blocked - sets the Blocked flag
2346 * -blocked - clears the Blocked flag
2347 * insync - sets Insync providing device isn't active
2350 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2351 md_error(rdev
->mddev
, rdev
);
2353 } else if (cmd_match(buf
, "remove")) {
2354 if (rdev
->raid_disk
>= 0)
2357 mddev_t
*mddev
= rdev
->mddev
;
2358 kick_rdev_from_array(rdev
);
2360 md_update_sb(mddev
, 1);
2361 md_new_event(mddev
);
2364 } else if (cmd_match(buf
, "writemostly")) {
2365 set_bit(WriteMostly
, &rdev
->flags
);
2367 } else if (cmd_match(buf
, "-writemostly")) {
2368 clear_bit(WriteMostly
, &rdev
->flags
);
2370 } else if (cmd_match(buf
, "blocked")) {
2371 set_bit(Blocked
, &rdev
->flags
);
2373 } else if (cmd_match(buf
, "-blocked")) {
2374 clear_bit(Blocked
, &rdev
->flags
);
2375 wake_up(&rdev
->blocked_wait
);
2376 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2377 md_wakeup_thread(rdev
->mddev
->thread
);
2380 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2381 set_bit(In_sync
, &rdev
->flags
);
2385 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2386 return err
? err
: len
;
2388 static struct rdev_sysfs_entry rdev_state
=
2389 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2392 errors_show(mdk_rdev_t
*rdev
, char *page
)
2394 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2398 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2401 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2402 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2403 atomic_set(&rdev
->corrected_errors
, n
);
2408 static struct rdev_sysfs_entry rdev_errors
=
2409 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2412 slot_show(mdk_rdev_t
*rdev
, char *page
)
2414 if (rdev
->raid_disk
< 0)
2415 return sprintf(page
, "none\n");
2417 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2421 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2426 int slot
= simple_strtoul(buf
, &e
, 10);
2427 if (strncmp(buf
, "none", 4)==0)
2429 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2431 if (rdev
->mddev
->pers
&& slot
== -1) {
2432 /* Setting 'slot' on an active array requires also
2433 * updating the 'rd%d' link, and communicating
2434 * with the personality with ->hot_*_disk.
2435 * For now we only support removing
2436 * failed/spare devices. This normally happens automatically,
2437 * but not when the metadata is externally managed.
2439 if (rdev
->raid_disk
== -1)
2441 /* personality does all needed checks */
2442 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2444 err
= rdev
->mddev
->pers
->
2445 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2448 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2449 sysfs_remove_link(&rdev
->mddev
->kobj
, nm
);
2450 rdev
->raid_disk
= -1;
2451 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2452 md_wakeup_thread(rdev
->mddev
->thread
);
2453 } else if (rdev
->mddev
->pers
) {
2455 /* Activating a spare .. or possibly reactivating
2456 * if we ever get bitmaps working here.
2459 if (rdev
->raid_disk
!= -1)
2462 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2465 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2466 if (rdev2
->raid_disk
== slot
)
2469 rdev
->raid_disk
= slot
;
2470 if (test_bit(In_sync
, &rdev
->flags
))
2471 rdev
->saved_raid_disk
= slot
;
2473 rdev
->saved_raid_disk
= -1;
2474 err
= rdev
->mddev
->pers
->
2475 hot_add_disk(rdev
->mddev
, rdev
);
2477 rdev
->raid_disk
= -1;
2480 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2481 sprintf(nm
, "rd%d", rdev
->raid_disk
);
2482 if (sysfs_create_link(&rdev
->mddev
->kobj
, &rdev
->kobj
, nm
))
2483 /* failure here is OK */;
2484 /* don't wakeup anyone, leave that to userspace. */
2486 if (slot
>= rdev
->mddev
->raid_disks
)
2488 rdev
->raid_disk
= slot
;
2489 /* assume it is working */
2490 clear_bit(Faulty
, &rdev
->flags
);
2491 clear_bit(WriteMostly
, &rdev
->flags
);
2492 set_bit(In_sync
, &rdev
->flags
);
2493 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2499 static struct rdev_sysfs_entry rdev_slot
=
2500 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2503 offset_show(mdk_rdev_t
*rdev
, char *page
)
2505 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2509 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2512 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2513 if (e
==buf
|| (*e
&& *e
!= '\n'))
2515 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2517 if (rdev
->sectors
&& rdev
->mddev
->external
)
2518 /* Must set offset before size, so overlap checks
2521 rdev
->data_offset
= offset
;
2525 static struct rdev_sysfs_entry rdev_offset
=
2526 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2529 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2531 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2534 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2536 /* check if two start/length pairs overlap */
2544 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2546 unsigned long long blocks
;
2549 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2552 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2553 return -EINVAL
; /* sector conversion overflow */
2556 if (new != blocks
* 2)
2557 return -EINVAL
; /* unsigned long long to sector_t overflow */
2564 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2566 mddev_t
*my_mddev
= rdev
->mddev
;
2567 sector_t oldsectors
= rdev
->sectors
;
2570 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2572 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2573 if (my_mddev
->persistent
) {
2574 sectors
= super_types
[my_mddev
->major_version
].
2575 rdev_size_change(rdev
, sectors
);
2578 } else if (!sectors
)
2579 sectors
= (rdev
->bdev
->bd_inode
->i_size
>> 9) -
2582 if (sectors
< my_mddev
->dev_sectors
)
2583 return -EINVAL
; /* component must fit device */
2585 rdev
->sectors
= sectors
;
2586 if (sectors
> oldsectors
&& my_mddev
->external
) {
2587 /* need to check that all other rdevs with the same ->bdev
2588 * do not overlap. We need to unlock the mddev to avoid
2589 * a deadlock. We have already changed rdev->sectors, and if
2590 * we have to change it back, we will have the lock again.
2594 struct list_head
*tmp
;
2596 mddev_unlock(my_mddev
);
2597 for_each_mddev(mddev
, tmp
) {
2601 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2602 if (test_bit(AllReserved
, &rdev2
->flags
) ||
2603 (rdev
->bdev
== rdev2
->bdev
&&
2605 overlaps(rdev
->data_offset
, rdev
->sectors
,
2611 mddev_unlock(mddev
);
2617 mddev_lock(my_mddev
);
2619 /* Someone else could have slipped in a size
2620 * change here, but doing so is just silly.
2621 * We put oldsectors back because we *know* it is
2622 * safe, and trust userspace not to race with
2625 rdev
->sectors
= oldsectors
;
2632 static struct rdev_sysfs_entry rdev_size
=
2633 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2636 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2638 unsigned long long recovery_start
= rdev
->recovery_offset
;
2640 if (test_bit(In_sync
, &rdev
->flags
) ||
2641 recovery_start
== MaxSector
)
2642 return sprintf(page
, "none\n");
2644 return sprintf(page
, "%llu\n", recovery_start
);
2647 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2649 unsigned long long recovery_start
;
2651 if (cmd_match(buf
, "none"))
2652 recovery_start
= MaxSector
;
2653 else if (strict_strtoull(buf
, 10, &recovery_start
))
2656 if (rdev
->mddev
->pers
&&
2657 rdev
->raid_disk
>= 0)
2660 rdev
->recovery_offset
= recovery_start
;
2661 if (recovery_start
== MaxSector
)
2662 set_bit(In_sync
, &rdev
->flags
);
2664 clear_bit(In_sync
, &rdev
->flags
);
2668 static struct rdev_sysfs_entry rdev_recovery_start
=
2669 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2671 static struct attribute
*rdev_default_attrs
[] = {
2677 &rdev_recovery_start
.attr
,
2681 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2683 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2684 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2685 mddev_t
*mddev
= rdev
->mddev
;
2691 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2693 if (rdev
->mddev
== NULL
)
2696 rv
= entry
->show(rdev
, page
);
2697 mddev_unlock(mddev
);
2703 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2704 const char *page
, size_t length
)
2706 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2707 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2709 mddev_t
*mddev
= rdev
->mddev
;
2713 if (!capable(CAP_SYS_ADMIN
))
2715 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2717 if (rdev
->mddev
== NULL
)
2720 rv
= entry
->store(rdev
, page
, length
);
2721 mddev_unlock(mddev
);
2726 static void rdev_free(struct kobject
*ko
)
2728 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2731 static const struct sysfs_ops rdev_sysfs_ops
= {
2732 .show
= rdev_attr_show
,
2733 .store
= rdev_attr_store
,
2735 static struct kobj_type rdev_ktype
= {
2736 .release
= rdev_free
,
2737 .sysfs_ops
= &rdev_sysfs_ops
,
2738 .default_attrs
= rdev_default_attrs
,
2741 void md_rdev_init(mdk_rdev_t
*rdev
)
2744 rdev
->saved_raid_disk
= -1;
2745 rdev
->raid_disk
= -1;
2747 rdev
->data_offset
= 0;
2748 rdev
->sb_events
= 0;
2749 rdev
->last_read_error
.tv_sec
= 0;
2750 rdev
->last_read_error
.tv_nsec
= 0;
2751 atomic_set(&rdev
->nr_pending
, 0);
2752 atomic_set(&rdev
->read_errors
, 0);
2753 atomic_set(&rdev
->corrected_errors
, 0);
2755 INIT_LIST_HEAD(&rdev
->same_set
);
2756 init_waitqueue_head(&rdev
->blocked_wait
);
2758 EXPORT_SYMBOL_GPL(md_rdev_init
);
2760 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2762 * mark the device faulty if:
2764 * - the device is nonexistent (zero size)
2765 * - the device has no valid superblock
2767 * a faulty rdev _never_ has rdev->sb set.
2769 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2771 char b
[BDEVNAME_SIZE
];
2776 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2778 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2779 return ERR_PTR(-ENOMEM
);
2783 if ((err
= alloc_disk_sb(rdev
)))
2786 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2790 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2792 size
= rdev
->bdev
->bd_inode
->i_size
>> BLOCK_SIZE_BITS
;
2795 "md: %s has zero or unknown size, marking faulty!\n",
2796 bdevname(rdev
->bdev
,b
));
2801 if (super_format
>= 0) {
2802 err
= super_types
[super_format
].
2803 load_super(rdev
, NULL
, super_minor
);
2804 if (err
== -EINVAL
) {
2806 "md: %s does not have a valid v%d.%d "
2807 "superblock, not importing!\n",
2808 bdevname(rdev
->bdev
,b
),
2809 super_format
, super_minor
);
2814 "md: could not read %s's sb, not importing!\n",
2815 bdevname(rdev
->bdev
,b
));
2823 if (rdev
->sb_page
) {
2829 return ERR_PTR(err
);
2833 * Check a full RAID array for plausibility
2837 static void analyze_sbs(mddev_t
* mddev
)
2840 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
2841 char b
[BDEVNAME_SIZE
];
2844 rdev_for_each(rdev
, tmp
, mddev
)
2845 switch (super_types
[mddev
->major_version
].
2846 load_super(rdev
, freshest
, mddev
->minor_version
)) {
2854 "md: fatal superblock inconsistency in %s"
2855 " -- removing from array\n",
2856 bdevname(rdev
->bdev
,b
));
2857 kick_rdev_from_array(rdev
);
2861 super_types
[mddev
->major_version
].
2862 validate_super(mddev
, freshest
);
2865 rdev_for_each(rdev
, tmp
, mddev
) {
2866 if (mddev
->max_disks
&&
2867 (rdev
->desc_nr
>= mddev
->max_disks
||
2868 i
> mddev
->max_disks
)) {
2870 "md: %s: %s: only %d devices permitted\n",
2871 mdname(mddev
), bdevname(rdev
->bdev
, b
),
2873 kick_rdev_from_array(rdev
);
2876 if (rdev
!= freshest
)
2877 if (super_types
[mddev
->major_version
].
2878 validate_super(mddev
, rdev
)) {
2879 printk(KERN_WARNING
"md: kicking non-fresh %s"
2881 bdevname(rdev
->bdev
,b
));
2882 kick_rdev_from_array(rdev
);
2885 if (mddev
->level
== LEVEL_MULTIPATH
) {
2886 rdev
->desc_nr
= i
++;
2887 rdev
->raid_disk
= rdev
->desc_nr
;
2888 set_bit(In_sync
, &rdev
->flags
);
2889 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
2890 rdev
->raid_disk
= -1;
2891 clear_bit(In_sync
, &rdev
->flags
);
2896 /* Read a fixed-point number.
2897 * Numbers in sysfs attributes should be in "standard" units where
2898 * possible, so time should be in seconds.
2899 * However we internally use a a much smaller unit such as
2900 * milliseconds or jiffies.
2901 * This function takes a decimal number with a possible fractional
2902 * component, and produces an integer which is the result of
2903 * multiplying that number by 10^'scale'.
2904 * all without any floating-point arithmetic.
2906 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
2908 unsigned long result
= 0;
2910 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
2913 else if (decimals
< scale
) {
2916 result
= result
* 10 + value
;
2928 while (decimals
< scale
) {
2937 static void md_safemode_timeout(unsigned long data
);
2940 safe_delay_show(mddev_t
*mddev
, char *page
)
2942 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
2943 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
2946 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
2950 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
2953 mddev
->safemode_delay
= 0;
2955 unsigned long old_delay
= mddev
->safemode_delay
;
2956 mddev
->safemode_delay
= (msec
*HZ
)/1000;
2957 if (mddev
->safemode_delay
== 0)
2958 mddev
->safemode_delay
= 1;
2959 if (mddev
->safemode_delay
< old_delay
)
2960 md_safemode_timeout((unsigned long)mddev
);
2964 static struct md_sysfs_entry md_safe_delay
=
2965 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
2968 level_show(mddev_t
*mddev
, char *page
)
2970 struct mdk_personality
*p
= mddev
->pers
;
2972 return sprintf(page
, "%s\n", p
->name
);
2973 else if (mddev
->clevel
[0])
2974 return sprintf(page
, "%s\n", mddev
->clevel
);
2975 else if (mddev
->level
!= LEVEL_NONE
)
2976 return sprintf(page
, "%d\n", mddev
->level
);
2982 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
2986 struct mdk_personality
*pers
;
2991 if (mddev
->pers
== NULL
) {
2994 if (len
>= sizeof(mddev
->clevel
))
2996 strncpy(mddev
->clevel
, buf
, len
);
2997 if (mddev
->clevel
[len
-1] == '\n')
2999 mddev
->clevel
[len
] = 0;
3000 mddev
->level
= LEVEL_NONE
;
3004 /* request to change the personality. Need to ensure:
3005 * - array is not engaged in resync/recovery/reshape
3006 * - old personality can be suspended
3007 * - new personality will access other array.
3010 if (mddev
->sync_thread
||
3011 mddev
->reshape_position
!= MaxSector
||
3012 mddev
->sysfs_active
)
3015 if (!mddev
->pers
->quiesce
) {
3016 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3017 mdname(mddev
), mddev
->pers
->name
);
3021 /* Now find the new personality */
3022 if (len
== 0 || len
>= sizeof(clevel
))
3024 strncpy(clevel
, buf
, len
);
3025 if (clevel
[len
-1] == '\n')
3028 if (strict_strtol(clevel
, 10, &level
))
3031 if (request_module("md-%s", clevel
) != 0)
3032 request_module("md-level-%s", clevel
);
3033 spin_lock(&pers_lock
);
3034 pers
= find_pers(level
, clevel
);
3035 if (!pers
|| !try_module_get(pers
->owner
)) {
3036 spin_unlock(&pers_lock
);
3037 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3040 spin_unlock(&pers_lock
);
3042 if (pers
== mddev
->pers
) {
3043 /* Nothing to do! */
3044 module_put(pers
->owner
);
3047 if (!pers
->takeover
) {
3048 module_put(pers
->owner
);
3049 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3050 mdname(mddev
), clevel
);
3054 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3055 rdev
->new_raid_disk
= rdev
->raid_disk
;
3057 /* ->takeover must set new_* and/or delta_disks
3058 * if it succeeds, and may set them when it fails.
3060 priv
= pers
->takeover(mddev
);
3062 mddev
->new_level
= mddev
->level
;
3063 mddev
->new_layout
= mddev
->layout
;
3064 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3065 mddev
->raid_disks
-= mddev
->delta_disks
;
3066 mddev
->delta_disks
= 0;
3067 module_put(pers
->owner
);
3068 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3069 mdname(mddev
), clevel
);
3070 return PTR_ERR(priv
);
3073 /* Looks like we have a winner */
3074 mddev_suspend(mddev
);
3075 mddev
->pers
->stop(mddev
);
3077 if (mddev
->pers
->sync_request
== NULL
&&
3078 pers
->sync_request
!= NULL
) {
3079 /* need to add the md_redundancy_group */
3080 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3082 "md: cannot register extra attributes for %s\n",
3084 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3086 if (mddev
->pers
->sync_request
!= NULL
&&
3087 pers
->sync_request
== NULL
) {
3088 /* need to remove the md_redundancy_group */
3089 if (mddev
->to_remove
== NULL
)
3090 mddev
->to_remove
= &md_redundancy_group
;
3093 if (mddev
->pers
->sync_request
== NULL
&&
3095 /* We are converting from a no-redundancy array
3096 * to a redundancy array and metadata is managed
3097 * externally so we need to be sure that writes
3098 * won't block due to a need to transition
3100 * until external management is started.
3103 mddev
->safemode_delay
= 0;
3104 mddev
->safemode
= 0;
3107 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3109 if (rdev
->raid_disk
< 0)
3111 if (rdev
->new_raid_disk
> mddev
->raid_disks
)
3112 rdev
->new_raid_disk
= -1;
3113 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3115 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3116 sysfs_remove_link(&mddev
->kobj
, nm
);
3118 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3119 if (rdev
->raid_disk
< 0)
3121 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3123 rdev
->raid_disk
= rdev
->new_raid_disk
;
3124 if (rdev
->raid_disk
< 0)
3125 clear_bit(In_sync
, &rdev
->flags
);
3128 sprintf(nm
, "rd%d", rdev
->raid_disk
);
3129 if(sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
3130 printk("md: cannot register %s for %s after level change\n",
3135 module_put(mddev
->pers
->owner
);
3137 mddev
->private = priv
;
3138 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3139 mddev
->level
= mddev
->new_level
;
3140 mddev
->layout
= mddev
->new_layout
;
3141 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3142 mddev
->delta_disks
= 0;
3143 if (mddev
->pers
->sync_request
== NULL
) {
3144 /* this is now an array without redundancy, so
3145 * it must always be in_sync
3148 del_timer_sync(&mddev
->safemode_timer
);
3151 mddev_resume(mddev
);
3152 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3153 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3154 md_wakeup_thread(mddev
->thread
);
3155 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3156 md_new_event(mddev
);
3160 static struct md_sysfs_entry md_level
=
3161 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3165 layout_show(mddev_t
*mddev
, char *page
)
3167 /* just a number, not meaningful for all levels */
3168 if (mddev
->reshape_position
!= MaxSector
&&
3169 mddev
->layout
!= mddev
->new_layout
)
3170 return sprintf(page
, "%d (%d)\n",
3171 mddev
->new_layout
, mddev
->layout
);
3172 return sprintf(page
, "%d\n", mddev
->layout
);
3176 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3179 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3181 if (!*buf
|| (*e
&& *e
!= '\n'))
3186 if (mddev
->pers
->check_reshape
== NULL
)
3188 mddev
->new_layout
= n
;
3189 err
= mddev
->pers
->check_reshape(mddev
);
3191 mddev
->new_layout
= mddev
->layout
;
3195 mddev
->new_layout
= n
;
3196 if (mddev
->reshape_position
== MaxSector
)
3201 static struct md_sysfs_entry md_layout
=
3202 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3206 raid_disks_show(mddev_t
*mddev
, char *page
)
3208 if (mddev
->raid_disks
== 0)
3210 if (mddev
->reshape_position
!= MaxSector
&&
3211 mddev
->delta_disks
!= 0)
3212 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3213 mddev
->raid_disks
- mddev
->delta_disks
);
3214 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3217 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3220 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3224 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3226 if (!*buf
|| (*e
&& *e
!= '\n'))
3230 rv
= update_raid_disks(mddev
, n
);
3231 else if (mddev
->reshape_position
!= MaxSector
) {
3232 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3233 mddev
->delta_disks
= n
- olddisks
;
3234 mddev
->raid_disks
= n
;
3236 mddev
->raid_disks
= n
;
3237 return rv
? rv
: len
;
3239 static struct md_sysfs_entry md_raid_disks
=
3240 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3243 chunk_size_show(mddev_t
*mddev
, char *page
)
3245 if (mddev
->reshape_position
!= MaxSector
&&
3246 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3247 return sprintf(page
, "%d (%d)\n",
3248 mddev
->new_chunk_sectors
<< 9,
3249 mddev
->chunk_sectors
<< 9);
3250 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3254 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3257 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3259 if (!*buf
|| (*e
&& *e
!= '\n'))
3264 if (mddev
->pers
->check_reshape
== NULL
)
3266 mddev
->new_chunk_sectors
= n
>> 9;
3267 err
= mddev
->pers
->check_reshape(mddev
);
3269 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3273 mddev
->new_chunk_sectors
= n
>> 9;
3274 if (mddev
->reshape_position
== MaxSector
)
3275 mddev
->chunk_sectors
= n
>> 9;
3279 static struct md_sysfs_entry md_chunk_size
=
3280 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3283 resync_start_show(mddev_t
*mddev
, char *page
)
3285 if (mddev
->recovery_cp
== MaxSector
)
3286 return sprintf(page
, "none\n");
3287 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3291 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3294 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3298 if (cmd_match(buf
, "none"))
3300 else if (!*buf
|| (*e
&& *e
!= '\n'))
3303 mddev
->recovery_cp
= n
;
3306 static struct md_sysfs_entry md_resync_start
=
3307 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3310 * The array state can be:
3313 * No devices, no size, no level
3314 * Equivalent to STOP_ARRAY ioctl
3316 * May have some settings, but array is not active
3317 * all IO results in error
3318 * When written, doesn't tear down array, but just stops it
3319 * suspended (not supported yet)
3320 * All IO requests will block. The array can be reconfigured.
3321 * Writing this, if accepted, will block until array is quiescent
3323 * no resync can happen. no superblocks get written.
3324 * write requests fail
3326 * like readonly, but behaves like 'clean' on a write request.
3328 * clean - no pending writes, but otherwise active.
3329 * When written to inactive array, starts without resync
3330 * If a write request arrives then
3331 * if metadata is known, mark 'dirty' and switch to 'active'.
3332 * if not known, block and switch to write-pending
3333 * If written to an active array that has pending writes, then fails.
3335 * fully active: IO and resync can be happening.
3336 * When written to inactive array, starts with resync
3339 * clean, but writes are blocked waiting for 'active' to be written.
3342 * like active, but no writes have been seen for a while (100msec).
3345 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3346 write_pending
, active_idle
, bad_word
};
3347 static char *array_states
[] = {
3348 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3349 "write-pending", "active-idle", NULL
};
3351 static int match_word(const char *word
, char **list
)
3354 for (n
=0; list
[n
]; n
++)
3355 if (cmd_match(word
, list
[n
]))
3361 array_state_show(mddev_t
*mddev
, char *page
)
3363 enum array_state st
= inactive
;
3376 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3378 else if (mddev
->safemode
)
3384 if (list_empty(&mddev
->disks
) &&
3385 mddev
->raid_disks
== 0 &&
3386 mddev
->dev_sectors
== 0)
3391 return sprintf(page
, "%s\n", array_states
[st
]);
3394 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3395 static int md_set_readonly(mddev_t
* mddev
, int is_open
);
3396 static int do_md_run(mddev_t
* mddev
);
3397 static int restart_array(mddev_t
*mddev
);
3400 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3403 enum array_state st
= match_word(buf
, array_states
);
3408 /* stopping an active array */
3409 if (atomic_read(&mddev
->openers
) > 0)
3411 err
= do_md_stop(mddev
, 0, 0);
3414 /* stopping an active array */
3416 if (atomic_read(&mddev
->openers
) > 0)
3418 err
= do_md_stop(mddev
, 2, 0);
3420 err
= 0; /* already inactive */
3423 break; /* not supported yet */
3426 err
= md_set_readonly(mddev
, 0);
3429 set_disk_ro(mddev
->gendisk
, 1);
3430 err
= do_md_run(mddev
);
3436 err
= md_set_readonly(mddev
, 0);
3437 else if (mddev
->ro
== 1)
3438 err
= restart_array(mddev
);
3441 set_disk_ro(mddev
->gendisk
, 0);
3445 err
= do_md_run(mddev
);
3450 restart_array(mddev
);
3451 spin_lock_irq(&mddev
->write_lock
);
3452 if (atomic_read(&mddev
->writes_pending
) == 0) {
3453 if (mddev
->in_sync
== 0) {
3455 if (mddev
->safemode
== 1)
3456 mddev
->safemode
= 0;
3457 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3462 spin_unlock_irq(&mddev
->write_lock
);
3468 restart_array(mddev
);
3469 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3470 wake_up(&mddev
->sb_wait
);
3474 set_disk_ro(mddev
->gendisk
, 0);
3475 err
= do_md_run(mddev
);
3480 /* these cannot be set */
3486 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3490 static struct md_sysfs_entry md_array_state
=
3491 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3494 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3495 return sprintf(page
, "%d\n",
3496 atomic_read(&mddev
->max_corr_read_errors
));
3500 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3503 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3505 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3506 atomic_set(&mddev
->max_corr_read_errors
, n
);
3512 static struct md_sysfs_entry max_corr_read_errors
=
3513 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3514 max_corrected_read_errors_store
);
3517 null_show(mddev_t
*mddev
, char *page
)
3523 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3525 /* buf must be %d:%d\n? giving major and minor numbers */
3526 /* The new device is added to the array.
3527 * If the array has a persistent superblock, we read the
3528 * superblock to initialise info and check validity.
3529 * Otherwise, only checking done is that in bind_rdev_to_array,
3530 * which mainly checks size.
3533 int major
= simple_strtoul(buf
, &e
, 10);
3539 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3541 minor
= simple_strtoul(e
+1, &e
, 10);
3542 if (*e
&& *e
!= '\n')
3544 dev
= MKDEV(major
, minor
);
3545 if (major
!= MAJOR(dev
) ||
3546 minor
!= MINOR(dev
))
3550 if (mddev
->persistent
) {
3551 rdev
= md_import_device(dev
, mddev
->major_version
,
3552 mddev
->minor_version
);
3553 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3554 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3555 mdk_rdev_t
, same_set
);
3556 err
= super_types
[mddev
->major_version
]
3557 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3561 } else if (mddev
->external
)
3562 rdev
= md_import_device(dev
, -2, -1);
3564 rdev
= md_import_device(dev
, -1, -1);
3567 return PTR_ERR(rdev
);
3568 err
= bind_rdev_to_array(rdev
, mddev
);
3572 return err
? err
: len
;
3575 static struct md_sysfs_entry md_new_device
=
3576 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3579 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3582 unsigned long chunk
, end_chunk
;
3586 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3588 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3589 if (buf
== end
) break;
3590 if (*end
== '-') { /* range */
3592 end_chunk
= simple_strtoul(buf
, &end
, 0);
3593 if (buf
== end
) break;
3595 if (*end
&& !isspace(*end
)) break;
3596 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3597 buf
= skip_spaces(end
);
3599 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3604 static struct md_sysfs_entry md_bitmap
=
3605 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3608 size_show(mddev_t
*mddev
, char *page
)
3610 return sprintf(page
, "%llu\n",
3611 (unsigned long long)mddev
->dev_sectors
/ 2);
3614 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3617 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3619 /* If array is inactive, we can reduce the component size, but
3620 * not increase it (except from 0).
3621 * If array is active, we can try an on-line resize
3624 int err
= strict_blocks_to_sectors(buf
, §ors
);
3629 err
= update_size(mddev
, sectors
);
3630 md_update_sb(mddev
, 1);
3632 if (mddev
->dev_sectors
== 0 ||
3633 mddev
->dev_sectors
> sectors
)
3634 mddev
->dev_sectors
= sectors
;
3638 return err
? err
: len
;
3641 static struct md_sysfs_entry md_size
=
3642 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3647 * 'none' for arrays with no metadata (good luck...)
3648 * 'external' for arrays with externally managed metadata,
3649 * or N.M for internally known formats
3652 metadata_show(mddev_t
*mddev
, char *page
)
3654 if (mddev
->persistent
)
3655 return sprintf(page
, "%d.%d\n",
3656 mddev
->major_version
, mddev
->minor_version
);
3657 else if (mddev
->external
)
3658 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3660 return sprintf(page
, "none\n");
3664 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3668 /* Changing the details of 'external' metadata is
3669 * always permitted. Otherwise there must be
3670 * no devices attached to the array.
3672 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3674 else if (!list_empty(&mddev
->disks
))
3677 if (cmd_match(buf
, "none")) {
3678 mddev
->persistent
= 0;
3679 mddev
->external
= 0;
3680 mddev
->major_version
= 0;
3681 mddev
->minor_version
= 90;
3684 if (strncmp(buf
, "external:", 9) == 0) {
3685 size_t namelen
= len
-9;
3686 if (namelen
>= sizeof(mddev
->metadata_type
))
3687 namelen
= sizeof(mddev
->metadata_type
)-1;
3688 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3689 mddev
->metadata_type
[namelen
] = 0;
3690 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3691 mddev
->metadata_type
[--namelen
] = 0;
3692 mddev
->persistent
= 0;
3693 mddev
->external
= 1;
3694 mddev
->major_version
= 0;
3695 mddev
->minor_version
= 90;
3698 major
= simple_strtoul(buf
, &e
, 10);
3699 if (e
==buf
|| *e
!= '.')
3702 minor
= simple_strtoul(buf
, &e
, 10);
3703 if (e
==buf
|| (*e
&& *e
!= '\n') )
3705 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3707 mddev
->major_version
= major
;
3708 mddev
->minor_version
= minor
;
3709 mddev
->persistent
= 1;
3710 mddev
->external
= 0;
3714 static struct md_sysfs_entry md_metadata
=
3715 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3718 action_show(mddev_t
*mddev
, char *page
)
3720 char *type
= "idle";
3721 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3723 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3724 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3725 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3727 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3728 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3730 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3734 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3737 return sprintf(page
, "%s\n", type
);
3741 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3743 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3746 if (cmd_match(page
, "frozen"))
3747 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3749 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3751 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3752 if (mddev
->sync_thread
) {
3753 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3754 md_unregister_thread(mddev
->sync_thread
);
3755 mddev
->sync_thread
= NULL
;
3756 mddev
->recovery
= 0;
3758 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3759 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3761 else if (cmd_match(page
, "resync"))
3762 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3763 else if (cmd_match(page
, "recover")) {
3764 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3765 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3766 } else if (cmd_match(page
, "reshape")) {
3768 if (mddev
->pers
->start_reshape
== NULL
)
3770 err
= mddev
->pers
->start_reshape(mddev
);
3773 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3775 if (cmd_match(page
, "check"))
3776 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3777 else if (!cmd_match(page
, "repair"))
3779 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3780 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3782 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3783 md_wakeup_thread(mddev
->thread
);
3784 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
3789 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3791 return sprintf(page
, "%llu\n",
3792 (unsigned long long) mddev
->resync_mismatches
);
3795 static struct md_sysfs_entry md_scan_mode
=
3796 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3799 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3802 sync_min_show(mddev_t
*mddev
, char *page
)
3804 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3805 mddev
->sync_speed_min
? "local": "system");
3809 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3813 if (strncmp(buf
, "system", 6)==0) {
3814 mddev
->sync_speed_min
= 0;
3817 min
= simple_strtoul(buf
, &e
, 10);
3818 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
3820 mddev
->sync_speed_min
= min
;
3824 static struct md_sysfs_entry md_sync_min
=
3825 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
3828 sync_max_show(mddev_t
*mddev
, char *page
)
3830 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
3831 mddev
->sync_speed_max
? "local": "system");
3835 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3839 if (strncmp(buf
, "system", 6)==0) {
3840 mddev
->sync_speed_max
= 0;
3843 max
= simple_strtoul(buf
, &e
, 10);
3844 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
3846 mddev
->sync_speed_max
= max
;
3850 static struct md_sysfs_entry md_sync_max
=
3851 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
3854 degraded_show(mddev_t
*mddev
, char *page
)
3856 return sprintf(page
, "%d\n", mddev
->degraded
);
3858 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
3861 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
3863 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
3867 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3871 if (strict_strtol(buf
, 10, &n
))
3874 if (n
!= 0 && n
!= 1)
3877 mddev
->parallel_resync
= n
;
3879 if (mddev
->sync_thread
)
3880 wake_up(&resync_wait
);
3885 /* force parallel resync, even with shared block devices */
3886 static struct md_sysfs_entry md_sync_force_parallel
=
3887 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
3888 sync_force_parallel_show
, sync_force_parallel_store
);
3891 sync_speed_show(mddev_t
*mddev
, char *page
)
3893 unsigned long resync
, dt
, db
;
3894 if (mddev
->curr_resync
== 0)
3895 return sprintf(page
, "none\n");
3896 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
3897 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
3899 db
= resync
- mddev
->resync_mark_cnt
;
3900 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
3903 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
3906 sync_completed_show(mddev_t
*mddev
, char *page
)
3908 unsigned long max_sectors
, resync
;
3910 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3911 return sprintf(page
, "none\n");
3913 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
3914 max_sectors
= mddev
->resync_max_sectors
;
3916 max_sectors
= mddev
->dev_sectors
;
3918 resync
= mddev
->curr_resync_completed
;
3919 return sprintf(page
, "%lu / %lu\n", resync
, max_sectors
);
3922 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
3925 min_sync_show(mddev_t
*mddev
, char *page
)
3927 return sprintf(page
, "%llu\n",
3928 (unsigned long long)mddev
->resync_min
);
3931 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3933 unsigned long long min
;
3934 if (strict_strtoull(buf
, 10, &min
))
3936 if (min
> mddev
->resync_max
)
3938 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3941 /* Must be a multiple of chunk_size */
3942 if (mddev
->chunk_sectors
) {
3943 sector_t temp
= min
;
3944 if (sector_div(temp
, mddev
->chunk_sectors
))
3947 mddev
->resync_min
= min
;
3952 static struct md_sysfs_entry md_min_sync
=
3953 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
3956 max_sync_show(mddev_t
*mddev
, char *page
)
3958 if (mddev
->resync_max
== MaxSector
)
3959 return sprintf(page
, "max\n");
3961 return sprintf(page
, "%llu\n",
3962 (unsigned long long)mddev
->resync_max
);
3965 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3967 if (strncmp(buf
, "max", 3) == 0)
3968 mddev
->resync_max
= MaxSector
;
3970 unsigned long long max
;
3971 if (strict_strtoull(buf
, 10, &max
))
3973 if (max
< mddev
->resync_min
)
3975 if (max
< mddev
->resync_max
&&
3977 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
3980 /* Must be a multiple of chunk_size */
3981 if (mddev
->chunk_sectors
) {
3982 sector_t temp
= max
;
3983 if (sector_div(temp
, mddev
->chunk_sectors
))
3986 mddev
->resync_max
= max
;
3988 wake_up(&mddev
->recovery_wait
);
3992 static struct md_sysfs_entry md_max_sync
=
3993 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
3996 suspend_lo_show(mddev_t
*mddev
, char *page
)
3998 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4002 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4005 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4007 if (mddev
->pers
== NULL
||
4008 mddev
->pers
->quiesce
== NULL
)
4010 if (buf
== e
|| (*e
&& *e
!= '\n'))
4012 if (new >= mddev
->suspend_hi
||
4013 (new > mddev
->suspend_lo
&& new < mddev
->suspend_hi
)) {
4014 mddev
->suspend_lo
= new;
4015 mddev
->pers
->quiesce(mddev
, 2);
4020 static struct md_sysfs_entry md_suspend_lo
=
4021 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4025 suspend_hi_show(mddev_t
*mddev
, char *page
)
4027 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4031 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4034 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4036 if (mddev
->pers
== NULL
||
4037 mddev
->pers
->quiesce
== NULL
)
4039 if (buf
== e
|| (*e
&& *e
!= '\n'))
4041 if ((new <= mddev
->suspend_lo
&& mddev
->suspend_lo
>= mddev
->suspend_hi
) ||
4042 (new > mddev
->suspend_lo
&& new > mddev
->suspend_hi
)) {
4043 mddev
->suspend_hi
= new;
4044 mddev
->pers
->quiesce(mddev
, 1);
4045 mddev
->pers
->quiesce(mddev
, 0);
4050 static struct md_sysfs_entry md_suspend_hi
=
4051 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4054 reshape_position_show(mddev_t
*mddev
, char *page
)
4056 if (mddev
->reshape_position
!= MaxSector
)
4057 return sprintf(page
, "%llu\n",
4058 (unsigned long long)mddev
->reshape_position
);
4059 strcpy(page
, "none\n");
4064 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4067 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4070 if (buf
== e
|| (*e
&& *e
!= '\n'))
4072 mddev
->reshape_position
= new;
4073 mddev
->delta_disks
= 0;
4074 mddev
->new_level
= mddev
->level
;
4075 mddev
->new_layout
= mddev
->layout
;
4076 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4080 static struct md_sysfs_entry md_reshape_position
=
4081 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4082 reshape_position_store
);
4085 array_size_show(mddev_t
*mddev
, char *page
)
4087 if (mddev
->external_size
)
4088 return sprintf(page
, "%llu\n",
4089 (unsigned long long)mddev
->array_sectors
/2);
4091 return sprintf(page
, "default\n");
4095 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4099 if (strncmp(buf
, "default", 7) == 0) {
4101 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4103 sectors
= mddev
->array_sectors
;
4105 mddev
->external_size
= 0;
4107 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4109 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4112 mddev
->external_size
= 1;
4115 mddev
->array_sectors
= sectors
;
4116 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4118 revalidate_disk(mddev
->gendisk
);
4123 static struct md_sysfs_entry md_array_size
=
4124 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4127 static struct attribute
*md_default_attrs
[] = {
4130 &md_raid_disks
.attr
,
4131 &md_chunk_size
.attr
,
4133 &md_resync_start
.attr
,
4135 &md_new_device
.attr
,
4136 &md_safe_delay
.attr
,
4137 &md_array_state
.attr
,
4138 &md_reshape_position
.attr
,
4139 &md_array_size
.attr
,
4140 &max_corr_read_errors
.attr
,
4144 static struct attribute
*md_redundancy_attrs
[] = {
4146 &md_mismatches
.attr
,
4149 &md_sync_speed
.attr
,
4150 &md_sync_force_parallel
.attr
,
4151 &md_sync_completed
.attr
,
4154 &md_suspend_lo
.attr
,
4155 &md_suspend_hi
.attr
,
4160 static struct attribute_group md_redundancy_group
= {
4162 .attrs
= md_redundancy_attrs
,
4167 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4169 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4170 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4175 rv
= mddev_lock(mddev
);
4177 rv
= entry
->show(mddev
, page
);
4178 mddev_unlock(mddev
);
4184 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4185 const char *page
, size_t length
)
4187 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4188 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4193 if (!capable(CAP_SYS_ADMIN
))
4195 rv
= mddev_lock(mddev
);
4196 if (mddev
->hold_active
== UNTIL_IOCTL
)
4197 mddev
->hold_active
= 0;
4199 rv
= entry
->store(mddev
, page
, length
);
4200 mddev_unlock(mddev
);
4205 static void md_free(struct kobject
*ko
)
4207 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4209 if (mddev
->sysfs_state
)
4210 sysfs_put(mddev
->sysfs_state
);
4212 if (mddev
->gendisk
) {
4213 del_gendisk(mddev
->gendisk
);
4214 put_disk(mddev
->gendisk
);
4217 blk_cleanup_queue(mddev
->queue
);
4222 static const struct sysfs_ops md_sysfs_ops
= {
4223 .show
= md_attr_show
,
4224 .store
= md_attr_store
,
4226 static struct kobj_type md_ktype
= {
4228 .sysfs_ops
= &md_sysfs_ops
,
4229 .default_attrs
= md_default_attrs
,
4234 static void mddev_delayed_delete(struct work_struct
*ws
)
4236 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4238 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4239 kobject_del(&mddev
->kobj
);
4240 kobject_put(&mddev
->kobj
);
4243 static int md_alloc(dev_t dev
, char *name
)
4245 static DEFINE_MUTEX(disks_mutex
);
4246 mddev_t
*mddev
= mddev_find(dev
);
4247 struct gendisk
*disk
;
4256 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4257 shift
= partitioned
? MdpMinorShift
: 0;
4258 unit
= MINOR(mddev
->unit
) >> shift
;
4260 /* wait for any previous instance of this device to be
4261 * completely removed (mddev_delayed_delete).
4263 flush_workqueue(md_misc_wq
);
4265 mutex_lock(&disks_mutex
);
4271 /* Need to ensure that 'name' is not a duplicate.
4274 spin_lock(&all_mddevs_lock
);
4276 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4277 if (mddev2
->gendisk
&&
4278 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4279 spin_unlock(&all_mddevs_lock
);
4282 spin_unlock(&all_mddevs_lock
);
4286 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4289 mddev
->queue
->queuedata
= mddev
;
4291 /* Can be unlocked because the queue is new: no concurrency */
4292 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER
, mddev
->queue
);
4294 blk_queue_make_request(mddev
->queue
, md_make_request
);
4296 disk
= alloc_disk(1 << shift
);
4298 blk_cleanup_queue(mddev
->queue
);
4299 mddev
->queue
= NULL
;
4302 disk
->major
= MAJOR(mddev
->unit
);
4303 disk
->first_minor
= unit
<< shift
;
4305 strcpy(disk
->disk_name
, name
);
4306 else if (partitioned
)
4307 sprintf(disk
->disk_name
, "md_d%d", unit
);
4309 sprintf(disk
->disk_name
, "md%d", unit
);
4310 disk
->fops
= &md_fops
;
4311 disk
->private_data
= mddev
;
4312 disk
->queue
= mddev
->queue
;
4313 /* Allow extended partitions. This makes the
4314 * 'mdp' device redundant, but we can't really
4317 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4319 mddev
->gendisk
= disk
;
4320 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4321 &disk_to_dev(disk
)->kobj
, "%s", "md");
4323 /* This isn't possible, but as kobject_init_and_add is marked
4324 * __must_check, we must do something with the result
4326 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4330 if (mddev
->kobj
.sd
&&
4331 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4332 printk(KERN_DEBUG
"pointless warning\n");
4334 mutex_unlock(&disks_mutex
);
4335 if (!error
&& mddev
->kobj
.sd
) {
4336 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4337 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4343 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4345 md_alloc(dev
, NULL
);
4349 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4351 /* val must be "md_*" where * is not all digits.
4352 * We allocate an array with a large free minor number, and
4353 * set the name to val. val must not already be an active name.
4355 int len
= strlen(val
);
4356 char buf
[DISK_NAME_LEN
];
4358 while (len
&& val
[len
-1] == '\n')
4360 if (len
>= DISK_NAME_LEN
)
4362 strlcpy(buf
, val
, len
+1);
4363 if (strncmp(buf
, "md_", 3) != 0)
4365 return md_alloc(0, buf
);
4368 static void md_safemode_timeout(unsigned long data
)
4370 mddev_t
*mddev
= (mddev_t
*) data
;
4372 if (!atomic_read(&mddev
->writes_pending
)) {
4373 mddev
->safemode
= 1;
4374 if (mddev
->external
)
4375 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4377 md_wakeup_thread(mddev
->thread
);
4380 static int start_dirty_degraded
;
4382 int md_run(mddev_t
*mddev
)
4386 struct mdk_personality
*pers
;
4388 if (list_empty(&mddev
->disks
))
4389 /* cannot run an array with no devices.. */
4394 /* Cannot run until previous stop completes properly */
4395 if (mddev
->sysfs_active
)
4399 * Analyze all RAID superblock(s)
4401 if (!mddev
->raid_disks
) {
4402 if (!mddev
->persistent
)
4407 if (mddev
->level
!= LEVEL_NONE
)
4408 request_module("md-level-%d", mddev
->level
);
4409 else if (mddev
->clevel
[0])
4410 request_module("md-%s", mddev
->clevel
);
4413 * Drop all container device buffers, from now on
4414 * the only valid external interface is through the md
4417 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4418 if (test_bit(Faulty
, &rdev
->flags
))
4420 sync_blockdev(rdev
->bdev
);
4421 invalidate_bdev(rdev
->bdev
);
4423 /* perform some consistency tests on the device.
4424 * We don't want the data to overlap the metadata,
4425 * Internal Bitmap issues have been handled elsewhere.
4427 if (rdev
->data_offset
< rdev
->sb_start
) {
4428 if (mddev
->dev_sectors
&&
4429 rdev
->data_offset
+ mddev
->dev_sectors
4431 printk("md: %s: data overlaps metadata\n",
4436 if (rdev
->sb_start
+ rdev
->sb_size
/512
4437 > rdev
->data_offset
) {
4438 printk("md: %s: metadata overlaps data\n",
4443 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4446 if (mddev
->bio_set
== NULL
)
4447 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
, sizeof(mddev
));
4449 spin_lock(&pers_lock
);
4450 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4451 if (!pers
|| !try_module_get(pers
->owner
)) {
4452 spin_unlock(&pers_lock
);
4453 if (mddev
->level
!= LEVEL_NONE
)
4454 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4457 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4462 spin_unlock(&pers_lock
);
4463 if (mddev
->level
!= pers
->level
) {
4464 mddev
->level
= pers
->level
;
4465 mddev
->new_level
= pers
->level
;
4467 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4469 if (mddev
->reshape_position
!= MaxSector
&&
4470 pers
->start_reshape
== NULL
) {
4471 /* This personality cannot handle reshaping... */
4473 module_put(pers
->owner
);
4477 if (pers
->sync_request
) {
4478 /* Warn if this is a potentially silly
4481 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4485 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4486 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4488 rdev
->bdev
->bd_contains
==
4489 rdev2
->bdev
->bd_contains
) {
4491 "%s: WARNING: %s appears to be"
4492 " on the same physical disk as"
4495 bdevname(rdev
->bdev
,b
),
4496 bdevname(rdev2
->bdev
,b2
));
4503 "True protection against single-disk"
4504 " failure might be compromised.\n");
4507 mddev
->recovery
= 0;
4508 /* may be over-ridden by personality */
4509 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4511 mddev
->ok_start_degraded
= start_dirty_degraded
;
4513 if (start_readonly
&& mddev
->ro
== 0)
4514 mddev
->ro
= 2; /* read-only, but switch on first write */
4516 err
= mddev
->pers
->run(mddev
);
4518 printk(KERN_ERR
"md: pers->run() failed ...\n");
4519 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4520 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4521 " but 'external_size' not in effect?\n", __func__
);
4523 "md: invalid array_size %llu > default size %llu\n",
4524 (unsigned long long)mddev
->array_sectors
/ 2,
4525 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4527 mddev
->pers
->stop(mddev
);
4529 if (err
== 0 && mddev
->pers
->sync_request
) {
4530 err
= bitmap_create(mddev
);
4532 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4533 mdname(mddev
), err
);
4534 mddev
->pers
->stop(mddev
);
4538 module_put(mddev
->pers
->owner
);
4540 bitmap_destroy(mddev
);
4543 if (mddev
->pers
->sync_request
) {
4544 if (mddev
->kobj
.sd
&&
4545 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4547 "md: cannot register extra attributes for %s\n",
4549 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4550 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4553 atomic_set(&mddev
->writes_pending
,0);
4554 atomic_set(&mddev
->max_corr_read_errors
,
4555 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4556 mddev
->safemode
= 0;
4557 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4558 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4559 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4562 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4563 if (rdev
->raid_disk
>= 0) {
4565 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4566 if (sysfs_create_link(&mddev
->kobj
, &rdev
->kobj
, nm
))
4567 /* failure here is OK */;
4570 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4573 md_update_sb(mddev
, 0);
4575 md_wakeup_thread(mddev
->thread
);
4576 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4578 md_new_event(mddev
);
4579 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4580 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4581 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4584 EXPORT_SYMBOL_GPL(md_run
);
4586 static int do_md_run(mddev_t
*mddev
)
4590 err
= md_run(mddev
);
4593 err
= bitmap_load(mddev
);
4595 bitmap_destroy(mddev
);
4598 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4599 revalidate_disk(mddev
->gendisk
);
4600 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4605 static int restart_array(mddev_t
*mddev
)
4607 struct gendisk
*disk
= mddev
->gendisk
;
4609 /* Complain if it has no devices */
4610 if (list_empty(&mddev
->disks
))
4616 mddev
->safemode
= 0;
4618 set_disk_ro(disk
, 0);
4619 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4621 /* Kick recovery or resync if necessary */
4622 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4623 md_wakeup_thread(mddev
->thread
);
4624 md_wakeup_thread(mddev
->sync_thread
);
4625 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4629 /* similar to deny_write_access, but accounts for our holding a reference
4630 * to the file ourselves */
4631 static int deny_bitmap_write_access(struct file
* file
)
4633 struct inode
*inode
= file
->f_mapping
->host
;
4635 spin_lock(&inode
->i_lock
);
4636 if (atomic_read(&inode
->i_writecount
) > 1) {
4637 spin_unlock(&inode
->i_lock
);
4640 atomic_set(&inode
->i_writecount
, -1);
4641 spin_unlock(&inode
->i_lock
);
4646 void restore_bitmap_write_access(struct file
*file
)
4648 struct inode
*inode
= file
->f_mapping
->host
;
4650 spin_lock(&inode
->i_lock
);
4651 atomic_set(&inode
->i_writecount
, 1);
4652 spin_unlock(&inode
->i_lock
);
4655 static void md_clean(mddev_t
*mddev
)
4657 mddev
->array_sectors
= 0;
4658 mddev
->external_size
= 0;
4659 mddev
->dev_sectors
= 0;
4660 mddev
->raid_disks
= 0;
4661 mddev
->recovery_cp
= 0;
4662 mddev
->resync_min
= 0;
4663 mddev
->resync_max
= MaxSector
;
4664 mddev
->reshape_position
= MaxSector
;
4665 mddev
->external
= 0;
4666 mddev
->persistent
= 0;
4667 mddev
->level
= LEVEL_NONE
;
4668 mddev
->clevel
[0] = 0;
4671 mddev
->metadata_type
[0] = 0;
4672 mddev
->chunk_sectors
= 0;
4673 mddev
->ctime
= mddev
->utime
= 0;
4675 mddev
->max_disks
= 0;
4677 mddev
->can_decrease_events
= 0;
4678 mddev
->delta_disks
= 0;
4679 mddev
->new_level
= LEVEL_NONE
;
4680 mddev
->new_layout
= 0;
4681 mddev
->new_chunk_sectors
= 0;
4682 mddev
->curr_resync
= 0;
4683 mddev
->resync_mismatches
= 0;
4684 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4685 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4686 mddev
->recovery
= 0;
4688 mddev
->degraded
= 0;
4689 mddev
->safemode
= 0;
4690 mddev
->bitmap_info
.offset
= 0;
4691 mddev
->bitmap_info
.default_offset
= 0;
4692 mddev
->bitmap_info
.chunksize
= 0;
4693 mddev
->bitmap_info
.daemon_sleep
= 0;
4694 mddev
->bitmap_info
.max_write_behind
= 0;
4698 void md_stop_writes(mddev_t
*mddev
)
4700 if (mddev
->sync_thread
) {
4701 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4702 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4703 md_unregister_thread(mddev
->sync_thread
);
4704 mddev
->sync_thread
= NULL
;
4707 del_timer_sync(&mddev
->safemode_timer
);
4709 bitmap_flush(mddev
);
4710 md_super_wait(mddev
);
4712 if (!mddev
->in_sync
|| mddev
->flags
) {
4713 /* mark array as shutdown cleanly */
4715 md_update_sb(mddev
, 1);
4718 EXPORT_SYMBOL_GPL(md_stop_writes
);
4720 void md_stop(mddev_t
*mddev
)
4722 mddev
->pers
->stop(mddev
);
4723 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
4724 mddev
->to_remove
= &md_redundancy_group
;
4725 module_put(mddev
->pers
->owner
);
4727 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4729 EXPORT_SYMBOL_GPL(md_stop
);
4731 static int md_set_readonly(mddev_t
*mddev
, int is_open
)
4734 mutex_lock(&mddev
->open_mutex
);
4735 if (atomic_read(&mddev
->openers
) > is_open
) {
4736 printk("md: %s still in use.\n",mdname(mddev
));
4741 md_stop_writes(mddev
);
4747 set_disk_ro(mddev
->gendisk
, 1);
4748 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4749 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4753 mutex_unlock(&mddev
->open_mutex
);
4758 * 0 - completely stop and dis-assemble array
4759 * 2 - stop but do not disassemble array
4761 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4763 struct gendisk
*disk
= mddev
->gendisk
;
4766 mutex_lock(&mddev
->open_mutex
);
4767 if (atomic_read(&mddev
->openers
) > is_open
||
4768 mddev
->sysfs_active
) {
4769 printk("md: %s still in use.\n",mdname(mddev
));
4770 mutex_unlock(&mddev
->open_mutex
);
4776 set_disk_ro(disk
, 0);
4778 md_stop_writes(mddev
);
4780 mddev
->queue
->merge_bvec_fn
= NULL
;
4781 mddev
->queue
->unplug_fn
= NULL
;
4782 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4784 /* tell userspace to handle 'inactive' */
4785 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4787 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4788 if (rdev
->raid_disk
>= 0) {
4790 sprintf(nm
, "rd%d", rdev
->raid_disk
);
4791 sysfs_remove_link(&mddev
->kobj
, nm
);
4794 set_capacity(disk
, 0);
4795 mutex_unlock(&mddev
->open_mutex
);
4796 revalidate_disk(disk
);
4801 mutex_unlock(&mddev
->open_mutex
);
4803 * Free resources if final stop
4806 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
4808 bitmap_destroy(mddev
);
4809 if (mddev
->bitmap_info
.file
) {
4810 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
4811 fput(mddev
->bitmap_info
.file
);
4812 mddev
->bitmap_info
.file
= NULL
;
4814 mddev
->bitmap_info
.offset
= 0;
4816 export_array(mddev
);
4819 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4820 if (mddev
->hold_active
== UNTIL_STOP
)
4821 mddev
->hold_active
= 0;
4823 blk_integrity_unregister(disk
);
4824 md_new_event(mddev
);
4825 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4830 static void autorun_array(mddev_t
*mddev
)
4835 if (list_empty(&mddev
->disks
))
4838 printk(KERN_INFO
"md: running: ");
4840 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4841 char b
[BDEVNAME_SIZE
];
4842 printk("<%s>", bdevname(rdev
->bdev
,b
));
4846 err
= do_md_run(mddev
);
4848 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
4849 do_md_stop(mddev
, 0, 0);
4854 * lets try to run arrays based on all disks that have arrived
4855 * until now. (those are in pending_raid_disks)
4857 * the method: pick the first pending disk, collect all disks with
4858 * the same UUID, remove all from the pending list and put them into
4859 * the 'same_array' list. Then order this list based on superblock
4860 * update time (freshest comes first), kick out 'old' disks and
4861 * compare superblocks. If everything's fine then run it.
4863 * If "unit" is allocated, then bump its reference count
4865 static void autorun_devices(int part
)
4867 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
4869 char b
[BDEVNAME_SIZE
];
4871 printk(KERN_INFO
"md: autorun ...\n");
4872 while (!list_empty(&pending_raid_disks
)) {
4875 LIST_HEAD(candidates
);
4876 rdev0
= list_entry(pending_raid_disks
.next
,
4877 mdk_rdev_t
, same_set
);
4879 printk(KERN_INFO
"md: considering %s ...\n",
4880 bdevname(rdev0
->bdev
,b
));
4881 INIT_LIST_HEAD(&candidates
);
4882 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
4883 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
4884 printk(KERN_INFO
"md: adding %s ...\n",
4885 bdevname(rdev
->bdev
,b
));
4886 list_move(&rdev
->same_set
, &candidates
);
4889 * now we have a set of devices, with all of them having
4890 * mostly sane superblocks. It's time to allocate the
4894 dev
= MKDEV(mdp_major
,
4895 rdev0
->preferred_minor
<< MdpMinorShift
);
4896 unit
= MINOR(dev
) >> MdpMinorShift
;
4898 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
4901 if (rdev0
->preferred_minor
!= unit
) {
4902 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
4903 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
4907 md_probe(dev
, NULL
, NULL
);
4908 mddev
= mddev_find(dev
);
4909 if (!mddev
|| !mddev
->gendisk
) {
4913 "md: cannot allocate memory for md drive.\n");
4916 if (mddev_lock(mddev
))
4917 printk(KERN_WARNING
"md: %s locked, cannot run\n",
4919 else if (mddev
->raid_disks
|| mddev
->major_version
4920 || !list_empty(&mddev
->disks
)) {
4922 "md: %s already running, cannot run %s\n",
4923 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
4924 mddev_unlock(mddev
);
4926 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
4927 mddev
->persistent
= 1;
4928 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4929 list_del_init(&rdev
->same_set
);
4930 if (bind_rdev_to_array(rdev
, mddev
))
4933 autorun_array(mddev
);
4934 mddev_unlock(mddev
);
4936 /* on success, candidates will be empty, on error
4939 rdev_for_each_list(rdev
, tmp
, &candidates
) {
4940 list_del_init(&rdev
->same_set
);
4945 printk(KERN_INFO
"md: ... autorun DONE.\n");
4947 #endif /* !MODULE */
4949 static int get_version(void __user
* arg
)
4953 ver
.major
= MD_MAJOR_VERSION
;
4954 ver
.minor
= MD_MINOR_VERSION
;
4955 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
4957 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
4963 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
4965 mdu_array_info_t info
;
4966 int nr
,working
,insync
,failed
,spare
;
4969 nr
=working
=insync
=failed
=spare
=0;
4970 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4972 if (test_bit(Faulty
, &rdev
->flags
))
4976 if (test_bit(In_sync
, &rdev
->flags
))
4983 info
.major_version
= mddev
->major_version
;
4984 info
.minor_version
= mddev
->minor_version
;
4985 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
4986 info
.ctime
= mddev
->ctime
;
4987 info
.level
= mddev
->level
;
4988 info
.size
= mddev
->dev_sectors
/ 2;
4989 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
4992 info
.raid_disks
= mddev
->raid_disks
;
4993 info
.md_minor
= mddev
->md_minor
;
4994 info
.not_persistent
= !mddev
->persistent
;
4996 info
.utime
= mddev
->utime
;
4999 info
.state
= (1<<MD_SB_CLEAN
);
5000 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5001 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5002 info
.active_disks
= insync
;
5003 info
.working_disks
= working
;
5004 info
.failed_disks
= failed
;
5005 info
.spare_disks
= spare
;
5007 info
.layout
= mddev
->layout
;
5008 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5010 if (copy_to_user(arg
, &info
, sizeof(info
)))
5016 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
5018 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5019 char *ptr
, *buf
= NULL
;
5022 if (md_allow_write(mddev
))
5023 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5025 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5030 /* bitmap disabled, zero the first byte and copy out */
5031 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5032 file
->pathname
[0] = '\0';
5036 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5040 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5044 strcpy(file
->pathname
, ptr
);
5048 if (copy_to_user(arg
, file
, sizeof(*file
)))
5056 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
5058 mdu_disk_info_t info
;
5061 if (copy_from_user(&info
, arg
, sizeof(info
)))
5064 rdev
= find_rdev_nr(mddev
, info
.number
);
5066 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5067 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5068 info
.raid_disk
= rdev
->raid_disk
;
5070 if (test_bit(Faulty
, &rdev
->flags
))
5071 info
.state
|= (1<<MD_DISK_FAULTY
);
5072 else if (test_bit(In_sync
, &rdev
->flags
)) {
5073 info
.state
|= (1<<MD_DISK_ACTIVE
);
5074 info
.state
|= (1<<MD_DISK_SYNC
);
5076 if (test_bit(WriteMostly
, &rdev
->flags
))
5077 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5079 info
.major
= info
.minor
= 0;
5080 info
.raid_disk
= -1;
5081 info
.state
= (1<<MD_DISK_REMOVED
);
5084 if (copy_to_user(arg
, &info
, sizeof(info
)))
5090 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
5092 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5094 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5096 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5099 if (!mddev
->raid_disks
) {
5101 /* expecting a device which has a superblock */
5102 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5105 "md: md_import_device returned %ld\n",
5107 return PTR_ERR(rdev
);
5109 if (!list_empty(&mddev
->disks
)) {
5110 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
5111 mdk_rdev_t
, same_set
);
5112 err
= super_types
[mddev
->major_version
]
5113 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5116 "md: %s has different UUID to %s\n",
5117 bdevname(rdev
->bdev
,b
),
5118 bdevname(rdev0
->bdev
,b2
));
5123 err
= bind_rdev_to_array(rdev
, mddev
);
5130 * add_new_disk can be used once the array is assembled
5131 * to add "hot spares". They must already have a superblock
5136 if (!mddev
->pers
->hot_add_disk
) {
5138 "%s: personality does not support diskops!\n",
5142 if (mddev
->persistent
)
5143 rdev
= md_import_device(dev
, mddev
->major_version
,
5144 mddev
->minor_version
);
5146 rdev
= md_import_device(dev
, -1, -1);
5149 "md: md_import_device returned %ld\n",
5151 return PTR_ERR(rdev
);
5153 /* set save_raid_disk if appropriate */
5154 if (!mddev
->persistent
) {
5155 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5156 info
->raid_disk
< mddev
->raid_disks
)
5157 rdev
->raid_disk
= info
->raid_disk
;
5159 rdev
->raid_disk
= -1;
5161 super_types
[mddev
->major_version
].
5162 validate_super(mddev
, rdev
);
5163 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5165 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5166 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5167 set_bit(WriteMostly
, &rdev
->flags
);
5169 clear_bit(WriteMostly
, &rdev
->flags
);
5171 rdev
->raid_disk
= -1;
5172 err
= bind_rdev_to_array(rdev
, mddev
);
5173 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5174 /* If there is hot_add_disk but no hot_remove_disk
5175 * then added disks for geometry changes,
5176 * and should be added immediately.
5178 super_types
[mddev
->major_version
].
5179 validate_super(mddev
, rdev
);
5180 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5182 unbind_rdev_from_array(rdev
);
5187 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5189 md_update_sb(mddev
, 1);
5190 if (mddev
->degraded
)
5191 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5192 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5193 md_wakeup_thread(mddev
->thread
);
5197 /* otherwise, add_new_disk is only allowed
5198 * for major_version==0 superblocks
5200 if (mddev
->major_version
!= 0) {
5201 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5206 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5208 rdev
= md_import_device(dev
, -1, 0);
5211 "md: error, md_import_device() returned %ld\n",
5213 return PTR_ERR(rdev
);
5215 rdev
->desc_nr
= info
->number
;
5216 if (info
->raid_disk
< mddev
->raid_disks
)
5217 rdev
->raid_disk
= info
->raid_disk
;
5219 rdev
->raid_disk
= -1;
5221 if (rdev
->raid_disk
< mddev
->raid_disks
)
5222 if (info
->state
& (1<<MD_DISK_SYNC
))
5223 set_bit(In_sync
, &rdev
->flags
);
5225 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5226 set_bit(WriteMostly
, &rdev
->flags
);
5228 if (!mddev
->persistent
) {
5229 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5230 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5232 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5233 rdev
->sectors
= rdev
->sb_start
;
5235 err
= bind_rdev_to_array(rdev
, mddev
);
5245 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5247 char b
[BDEVNAME_SIZE
];
5250 rdev
= find_rdev(mddev
, dev
);
5254 if (rdev
->raid_disk
>= 0)
5257 kick_rdev_from_array(rdev
);
5258 md_update_sb(mddev
, 1);
5259 md_new_event(mddev
);
5263 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5264 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5268 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5270 char b
[BDEVNAME_SIZE
];
5277 if (mddev
->major_version
!= 0) {
5278 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5279 " version-0 superblocks.\n",
5283 if (!mddev
->pers
->hot_add_disk
) {
5285 "%s: personality does not support diskops!\n",
5290 rdev
= md_import_device(dev
, -1, 0);
5293 "md: error, md_import_device() returned %ld\n",
5298 if (mddev
->persistent
)
5299 rdev
->sb_start
= calc_dev_sboffset(rdev
->bdev
);
5301 rdev
->sb_start
= rdev
->bdev
->bd_inode
->i_size
/ 512;
5303 rdev
->sectors
= rdev
->sb_start
;
5305 if (test_bit(Faulty
, &rdev
->flags
)) {
5307 "md: can not hot-add faulty %s disk to %s!\n",
5308 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5312 clear_bit(In_sync
, &rdev
->flags
);
5314 rdev
->saved_raid_disk
= -1;
5315 err
= bind_rdev_to_array(rdev
, mddev
);
5320 * The rest should better be atomic, we can have disk failures
5321 * noticed in interrupt contexts ...
5324 rdev
->raid_disk
= -1;
5326 md_update_sb(mddev
, 1);
5329 * Kick recovery, maybe this spare has to be added to the
5330 * array immediately.
5332 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5333 md_wakeup_thread(mddev
->thread
);
5334 md_new_event(mddev
);
5342 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5347 if (!mddev
->pers
->quiesce
)
5349 if (mddev
->recovery
|| mddev
->sync_thread
)
5351 /* we should be able to change the bitmap.. */
5357 return -EEXIST
; /* cannot add when bitmap is present */
5358 mddev
->bitmap_info
.file
= fget(fd
);
5360 if (mddev
->bitmap_info
.file
== NULL
) {
5361 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5366 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5368 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5370 fput(mddev
->bitmap_info
.file
);
5371 mddev
->bitmap_info
.file
= NULL
;
5374 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5375 } else if (mddev
->bitmap
== NULL
)
5376 return -ENOENT
; /* cannot remove what isn't there */
5379 mddev
->pers
->quiesce(mddev
, 1);
5381 err
= bitmap_create(mddev
);
5383 err
= bitmap_load(mddev
);
5385 if (fd
< 0 || err
) {
5386 bitmap_destroy(mddev
);
5387 fd
= -1; /* make sure to put the file */
5389 mddev
->pers
->quiesce(mddev
, 0);
5392 if (mddev
->bitmap_info
.file
) {
5393 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5394 fput(mddev
->bitmap_info
.file
);
5396 mddev
->bitmap_info
.file
= NULL
;
5403 * set_array_info is used two different ways
5404 * The original usage is when creating a new array.
5405 * In this usage, raid_disks is > 0 and it together with
5406 * level, size, not_persistent,layout,chunksize determine the
5407 * shape of the array.
5408 * This will always create an array with a type-0.90.0 superblock.
5409 * The newer usage is when assembling an array.
5410 * In this case raid_disks will be 0, and the major_version field is
5411 * use to determine which style super-blocks are to be found on the devices.
5412 * The minor and patch _version numbers are also kept incase the
5413 * super_block handler wishes to interpret them.
5415 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5418 if (info
->raid_disks
== 0) {
5419 /* just setting version number for superblock loading */
5420 if (info
->major_version
< 0 ||
5421 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5422 super_types
[info
->major_version
].name
== NULL
) {
5423 /* maybe try to auto-load a module? */
5425 "md: superblock version %d not known\n",
5426 info
->major_version
);
5429 mddev
->major_version
= info
->major_version
;
5430 mddev
->minor_version
= info
->minor_version
;
5431 mddev
->patch_version
= info
->patch_version
;
5432 mddev
->persistent
= !info
->not_persistent
;
5433 /* ensure mddev_put doesn't delete this now that there
5434 * is some minimal configuration.
5436 mddev
->ctime
= get_seconds();
5439 mddev
->major_version
= MD_MAJOR_VERSION
;
5440 mddev
->minor_version
= MD_MINOR_VERSION
;
5441 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5442 mddev
->ctime
= get_seconds();
5444 mddev
->level
= info
->level
;
5445 mddev
->clevel
[0] = 0;
5446 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5447 mddev
->raid_disks
= info
->raid_disks
;
5448 /* don't set md_minor, it is determined by which /dev/md* was
5451 if (info
->state
& (1<<MD_SB_CLEAN
))
5452 mddev
->recovery_cp
= MaxSector
;
5454 mddev
->recovery_cp
= 0;
5455 mddev
->persistent
= ! info
->not_persistent
;
5456 mddev
->external
= 0;
5458 mddev
->layout
= info
->layout
;
5459 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5461 mddev
->max_disks
= MD_SB_DISKS
;
5463 if (mddev
->persistent
)
5465 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5467 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5468 mddev
->bitmap_info
.offset
= 0;
5470 mddev
->reshape_position
= MaxSector
;
5473 * Generate a 128 bit UUID
5475 get_random_bytes(mddev
->uuid
, 16);
5477 mddev
->new_level
= mddev
->level
;
5478 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5479 mddev
->new_layout
= mddev
->layout
;
5480 mddev
->delta_disks
= 0;
5485 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5487 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5489 if (mddev
->external_size
)
5492 mddev
->array_sectors
= array_sectors
;
5494 EXPORT_SYMBOL(md_set_array_sectors
);
5496 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5500 int fit
= (num_sectors
== 0);
5502 if (mddev
->pers
->resize
== NULL
)
5504 /* The "num_sectors" is the number of sectors of each device that
5505 * is used. This can only make sense for arrays with redundancy.
5506 * linear and raid0 always use whatever space is available. We can only
5507 * consider changing this number if no resync or reconstruction is
5508 * happening, and if the new size is acceptable. It must fit before the
5509 * sb_start or, if that is <data_offset, it must fit before the size
5510 * of each device. If num_sectors is zero, we find the largest size
5514 if (mddev
->sync_thread
)
5517 /* Sorry, cannot grow a bitmap yet, just remove it,
5521 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5522 sector_t avail
= rdev
->sectors
;
5524 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5525 num_sectors
= avail
;
5526 if (avail
< num_sectors
)
5529 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5531 revalidate_disk(mddev
->gendisk
);
5535 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5538 /* change the number of raid disks */
5539 if (mddev
->pers
->check_reshape
== NULL
)
5541 if (raid_disks
<= 0 ||
5542 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5544 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5546 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5548 rv
= mddev
->pers
->check_reshape(mddev
);
5554 * update_array_info is used to change the configuration of an
5556 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5557 * fields in the info are checked against the array.
5558 * Any differences that cannot be handled will cause an error.
5559 * Normally, only one change can be managed at a time.
5561 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5567 /* calculate expected state,ignoring low bits */
5568 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5569 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5571 if (mddev
->major_version
!= info
->major_version
||
5572 mddev
->minor_version
!= info
->minor_version
||
5573 /* mddev->patch_version != info->patch_version || */
5574 mddev
->ctime
!= info
->ctime
||
5575 mddev
->level
!= info
->level
||
5576 /* mddev->layout != info->layout || */
5577 !mddev
->persistent
!= info
->not_persistent
||
5578 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5579 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5580 ((state
^info
->state
) & 0xfffffe00)
5583 /* Check there is only one change */
5584 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5586 if (mddev
->raid_disks
!= info
->raid_disks
)
5588 if (mddev
->layout
!= info
->layout
)
5590 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5597 if (mddev
->layout
!= info
->layout
) {
5599 * we don't need to do anything at the md level, the
5600 * personality will take care of it all.
5602 if (mddev
->pers
->check_reshape
== NULL
)
5605 mddev
->new_layout
= info
->layout
;
5606 rv
= mddev
->pers
->check_reshape(mddev
);
5608 mddev
->new_layout
= mddev
->layout
;
5612 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5613 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5615 if (mddev
->raid_disks
!= info
->raid_disks
)
5616 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5618 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5619 if (mddev
->pers
->quiesce
== NULL
)
5621 if (mddev
->recovery
|| mddev
->sync_thread
)
5623 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5624 /* add the bitmap */
5627 if (mddev
->bitmap_info
.default_offset
== 0)
5629 mddev
->bitmap_info
.offset
=
5630 mddev
->bitmap_info
.default_offset
;
5631 mddev
->pers
->quiesce(mddev
, 1);
5632 rv
= bitmap_create(mddev
);
5634 rv
= bitmap_load(mddev
);
5636 bitmap_destroy(mddev
);
5637 mddev
->pers
->quiesce(mddev
, 0);
5639 /* remove the bitmap */
5642 if (mddev
->bitmap
->file
)
5644 mddev
->pers
->quiesce(mddev
, 1);
5645 bitmap_destroy(mddev
);
5646 mddev
->pers
->quiesce(mddev
, 0);
5647 mddev
->bitmap_info
.offset
= 0;
5650 md_update_sb(mddev
, 1);
5654 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5658 if (mddev
->pers
== NULL
)
5661 rdev
= find_rdev(mddev
, dev
);
5665 md_error(mddev
, rdev
);
5670 * We have a problem here : there is no easy way to give a CHS
5671 * virtual geometry. We currently pretend that we have a 2 heads
5672 * 4 sectors (with a BIG number of cylinders...). This drives
5673 * dosfs just mad... ;-)
5675 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5677 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5681 geo
->cylinders
= mddev
->array_sectors
/ 8;
5685 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5686 unsigned int cmd
, unsigned long arg
)
5689 void __user
*argp
= (void __user
*)arg
;
5690 mddev_t
*mddev
= NULL
;
5693 if (!capable(CAP_SYS_ADMIN
))
5697 * Commands dealing with the RAID driver but not any
5703 err
= get_version(argp
);
5706 case PRINT_RAID_DEBUG
:
5714 autostart_arrays(arg
);
5721 * Commands creating/starting a new array:
5724 mddev
= bdev
->bd_disk
->private_data
;
5731 err
= mddev_lock(mddev
);
5734 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5741 case SET_ARRAY_INFO
:
5743 mdu_array_info_t info
;
5745 memset(&info
, 0, sizeof(info
));
5746 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5751 err
= update_array_info(mddev
, &info
);
5753 printk(KERN_WARNING
"md: couldn't update"
5754 " array info. %d\n", err
);
5759 if (!list_empty(&mddev
->disks
)) {
5761 "md: array %s already has disks!\n",
5766 if (mddev
->raid_disks
) {
5768 "md: array %s already initialised!\n",
5773 err
= set_array_info(mddev
, &info
);
5775 printk(KERN_WARNING
"md: couldn't set"
5776 " array info. %d\n", err
);
5786 * Commands querying/configuring an existing array:
5788 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5789 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5790 if ((!mddev
->raid_disks
&& !mddev
->external
)
5791 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
5792 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
5793 && cmd
!= GET_BITMAP_FILE
) {
5799 * Commands even a read-only array can execute:
5803 case GET_ARRAY_INFO
:
5804 err
= get_array_info(mddev
, argp
);
5807 case GET_BITMAP_FILE
:
5808 err
= get_bitmap_file(mddev
, argp
);
5812 err
= get_disk_info(mddev
, argp
);
5815 case RESTART_ARRAY_RW
:
5816 err
= restart_array(mddev
);
5820 err
= do_md_stop(mddev
, 0, 1);
5824 err
= md_set_readonly(mddev
, 1);
5828 if (get_user(ro
, (int __user
*)(arg
))) {
5834 /* if the bdev is going readonly the value of mddev->ro
5835 * does not matter, no writes are coming
5840 /* are we are already prepared for writes? */
5844 /* transitioning to readauto need only happen for
5845 * arrays that call md_write_start
5848 err
= restart_array(mddev
);
5851 set_disk_ro(mddev
->gendisk
, 0);
5858 * The remaining ioctls are changing the state of the
5859 * superblock, so we do not allow them on read-only arrays.
5860 * However non-MD ioctls (e.g. get-size) will still come through
5861 * here and hit the 'default' below, so only disallow
5862 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5864 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
5865 if (mddev
->ro
== 2) {
5867 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5868 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5869 md_wakeup_thread(mddev
->thread
);
5880 mdu_disk_info_t info
;
5881 if (copy_from_user(&info
, argp
, sizeof(info
)))
5884 err
= add_new_disk(mddev
, &info
);
5888 case HOT_REMOVE_DISK
:
5889 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
5893 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
5896 case SET_DISK_FAULTY
:
5897 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
5901 err
= do_md_run(mddev
);
5904 case SET_BITMAP_FILE
:
5905 err
= set_bitmap_file(mddev
, (int)arg
);
5915 if (mddev
->hold_active
== UNTIL_IOCTL
&&
5917 mddev
->hold_active
= 0;
5918 mddev_unlock(mddev
);
5927 #ifdef CONFIG_COMPAT
5928 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
5929 unsigned int cmd
, unsigned long arg
)
5932 case HOT_REMOVE_DISK
:
5934 case SET_DISK_FAULTY
:
5935 case SET_BITMAP_FILE
:
5936 /* These take in integer arg, do not convert */
5939 arg
= (unsigned long)compat_ptr(arg
);
5943 return md_ioctl(bdev
, mode
, cmd
, arg
);
5945 #endif /* CONFIG_COMPAT */
5947 static int md_open(struct block_device
*bdev
, fmode_t mode
)
5950 * Succeed if we can lock the mddev, which confirms that
5951 * it isn't being stopped right now.
5953 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
5956 if (mddev
->gendisk
!= bdev
->bd_disk
) {
5957 /* we are racing with mddev_put which is discarding this
5961 /* Wait until bdev->bd_disk is definitely gone */
5962 flush_workqueue(md_misc_wq
);
5963 /* Then retry the open from the top */
5964 return -ERESTARTSYS
;
5966 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
5968 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
5972 atomic_inc(&mddev
->openers
);
5973 mutex_unlock(&mddev
->open_mutex
);
5975 check_disk_size_change(mddev
->gendisk
, bdev
);
5980 static int md_release(struct gendisk
*disk
, fmode_t mode
)
5982 mddev_t
*mddev
= disk
->private_data
;
5985 atomic_dec(&mddev
->openers
);
5990 static const struct block_device_operations md_fops
=
5992 .owner
= THIS_MODULE
,
5994 .release
= md_release
,
5996 #ifdef CONFIG_COMPAT
5997 .compat_ioctl
= md_compat_ioctl
,
5999 .getgeo
= md_getgeo
,
6002 static int md_thread(void * arg
)
6004 mdk_thread_t
*thread
= arg
;
6007 * md_thread is a 'system-thread', it's priority should be very
6008 * high. We avoid resource deadlocks individually in each
6009 * raid personality. (RAID5 does preallocation) We also use RR and
6010 * the very same RT priority as kswapd, thus we will never get
6011 * into a priority inversion deadlock.
6013 * we definitely have to have equal or higher priority than
6014 * bdflush, otherwise bdflush will deadlock if there are too
6015 * many dirty RAID5 blocks.
6018 allow_signal(SIGKILL
);
6019 while (!kthread_should_stop()) {
6021 /* We need to wait INTERRUPTIBLE so that
6022 * we don't add to the load-average.
6023 * That means we need to be sure no signals are
6026 if (signal_pending(current
))
6027 flush_signals(current
);
6029 wait_event_interruptible_timeout
6031 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6032 || kthread_should_stop(),
6035 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6037 thread
->run(thread
->mddev
);
6043 void md_wakeup_thread(mdk_thread_t
*thread
)
6046 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6047 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6048 wake_up(&thread
->wqueue
);
6052 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
6055 mdk_thread_t
*thread
;
6057 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
6061 init_waitqueue_head(&thread
->wqueue
);
6064 thread
->mddev
= mddev
;
6065 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6066 thread
->tsk
= kthread_run(md_thread
, thread
,
6068 mdname(thread
->mddev
),
6069 name
?: mddev
->pers
->name
);
6070 if (IS_ERR(thread
->tsk
)) {
6077 void md_unregister_thread(mdk_thread_t
*thread
)
6081 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6083 kthread_stop(thread
->tsk
);
6087 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
6094 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6097 if (mddev
->external
)
6098 set_bit(Blocked
, &rdev
->flags
);
6100 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6102 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6103 __builtin_return_address(0),__builtin_return_address(1),
6104 __builtin_return_address(2),__builtin_return_address(3));
6108 if (!mddev
->pers
->error_handler
)
6110 mddev
->pers
->error_handler(mddev
,rdev
);
6111 if (mddev
->degraded
)
6112 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6113 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6114 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6115 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6116 md_wakeup_thread(mddev
->thread
);
6117 if (mddev
->event_work
.func
)
6118 queue_work(md_misc_wq
, &mddev
->event_work
);
6119 md_new_event_inintr(mddev
);
6122 /* seq_file implementation /proc/mdstat */
6124 static void status_unused(struct seq_file
*seq
)
6129 seq_printf(seq
, "unused devices: ");
6131 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6132 char b
[BDEVNAME_SIZE
];
6134 seq_printf(seq
, "%s ",
6135 bdevname(rdev
->bdev
,b
));
6138 seq_printf(seq
, "<none>");
6140 seq_printf(seq
, "\n");
6144 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6146 sector_t max_sectors
, resync
, res
;
6147 unsigned long dt
, db
;
6150 unsigned int per_milli
;
6152 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6154 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6155 max_sectors
= mddev
->resync_max_sectors
;
6157 max_sectors
= mddev
->dev_sectors
;
6160 * Should not happen.
6166 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6167 * in a sector_t, and (max_sectors>>scale) will fit in a
6168 * u32, as those are the requirements for sector_div.
6169 * Thus 'scale' must be at least 10
6172 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6173 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6176 res
= (resync
>>scale
)*1000;
6177 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6181 int i
, x
= per_milli
/50, y
= 20-x
;
6182 seq_printf(seq
, "[");
6183 for (i
= 0; i
< x
; i
++)
6184 seq_printf(seq
, "=");
6185 seq_printf(seq
, ">");
6186 for (i
= 0; i
< y
; i
++)
6187 seq_printf(seq
, ".");
6188 seq_printf(seq
, "] ");
6190 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6191 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6193 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6195 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6196 "resync" : "recovery"))),
6197 per_milli
/10, per_milli
% 10,
6198 (unsigned long long) resync
/2,
6199 (unsigned long long) max_sectors
/2);
6202 * dt: time from mark until now
6203 * db: blocks written from mark until now
6204 * rt: remaining time
6206 * rt is a sector_t, so could be 32bit or 64bit.
6207 * So we divide before multiply in case it is 32bit and close
6209 * We scale the divisor (db) by 32 to avoid loosing precision
6210 * near the end of resync when the number of remaining sectors
6212 * We then divide rt by 32 after multiplying by db to compensate.
6213 * The '+1' avoids division by zero if db is very small.
6215 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6217 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6218 - mddev
->resync_mark_cnt
;
6220 rt
= max_sectors
- resync
; /* number of remaining sectors */
6221 sector_div(rt
, db
/32+1);
6225 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6226 ((unsigned long)rt
% 60)/6);
6228 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6231 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6233 struct list_head
*tmp
;
6243 spin_lock(&all_mddevs_lock
);
6244 list_for_each(tmp
,&all_mddevs
)
6246 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6248 spin_unlock(&all_mddevs_lock
);
6251 spin_unlock(&all_mddevs_lock
);
6253 return (void*)2;/* tail */
6257 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6259 struct list_head
*tmp
;
6260 mddev_t
*next_mddev
, *mddev
= v
;
6266 spin_lock(&all_mddevs_lock
);
6268 tmp
= all_mddevs
.next
;
6270 tmp
= mddev
->all_mddevs
.next
;
6271 if (tmp
!= &all_mddevs
)
6272 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6274 next_mddev
= (void*)2;
6277 spin_unlock(&all_mddevs_lock
);
6285 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6289 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6293 struct mdstat_info
{
6297 static int md_seq_show(struct seq_file
*seq
, void *v
)
6302 struct mdstat_info
*mi
= seq
->private;
6303 struct bitmap
*bitmap
;
6305 if (v
== (void*)1) {
6306 struct mdk_personality
*pers
;
6307 seq_printf(seq
, "Personalities : ");
6308 spin_lock(&pers_lock
);
6309 list_for_each_entry(pers
, &pers_list
, list
)
6310 seq_printf(seq
, "[%s] ", pers
->name
);
6312 spin_unlock(&pers_lock
);
6313 seq_printf(seq
, "\n");
6314 mi
->event
= atomic_read(&md_event_count
);
6317 if (v
== (void*)2) {
6322 if (mddev_lock(mddev
) < 0)
6325 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6326 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6327 mddev
->pers
? "" : "in");
6330 seq_printf(seq
, " (read-only)");
6332 seq_printf(seq
, " (auto-read-only)");
6333 seq_printf(seq
, " %s", mddev
->pers
->name
);
6337 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6338 char b
[BDEVNAME_SIZE
];
6339 seq_printf(seq
, " %s[%d]",
6340 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6341 if (test_bit(WriteMostly
, &rdev
->flags
))
6342 seq_printf(seq
, "(W)");
6343 if (test_bit(Faulty
, &rdev
->flags
)) {
6344 seq_printf(seq
, "(F)");
6346 } else if (rdev
->raid_disk
< 0)
6347 seq_printf(seq
, "(S)"); /* spare */
6348 sectors
+= rdev
->sectors
;
6351 if (!list_empty(&mddev
->disks
)) {
6353 seq_printf(seq
, "\n %llu blocks",
6354 (unsigned long long)
6355 mddev
->array_sectors
/ 2);
6357 seq_printf(seq
, "\n %llu blocks",
6358 (unsigned long long)sectors
/ 2);
6360 if (mddev
->persistent
) {
6361 if (mddev
->major_version
!= 0 ||
6362 mddev
->minor_version
!= 90) {
6363 seq_printf(seq
," super %d.%d",
6364 mddev
->major_version
,
6365 mddev
->minor_version
);
6367 } else if (mddev
->external
)
6368 seq_printf(seq
, " super external:%s",
6369 mddev
->metadata_type
);
6371 seq_printf(seq
, " super non-persistent");
6374 mddev
->pers
->status(seq
, mddev
);
6375 seq_printf(seq
, "\n ");
6376 if (mddev
->pers
->sync_request
) {
6377 if (mddev
->curr_resync
> 2) {
6378 status_resync(seq
, mddev
);
6379 seq_printf(seq
, "\n ");
6380 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6381 seq_printf(seq
, "\tresync=DELAYED\n ");
6382 else if (mddev
->recovery_cp
< MaxSector
)
6383 seq_printf(seq
, "\tresync=PENDING\n ");
6386 seq_printf(seq
, "\n ");
6388 if ((bitmap
= mddev
->bitmap
)) {
6389 unsigned long chunk_kb
;
6390 unsigned long flags
;
6391 spin_lock_irqsave(&bitmap
->lock
, flags
);
6392 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6393 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6395 bitmap
->pages
- bitmap
->missing_pages
,
6397 (bitmap
->pages
- bitmap
->missing_pages
)
6398 << (PAGE_SHIFT
- 10),
6399 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6400 chunk_kb
? "KB" : "B");
6402 seq_printf(seq
, ", file: ");
6403 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6406 seq_printf(seq
, "\n");
6407 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6410 seq_printf(seq
, "\n");
6412 mddev_unlock(mddev
);
6417 static const struct seq_operations md_seq_ops
= {
6418 .start
= md_seq_start
,
6419 .next
= md_seq_next
,
6420 .stop
= md_seq_stop
,
6421 .show
= md_seq_show
,
6424 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6427 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6431 error
= seq_open(file
, &md_seq_ops
);
6435 struct seq_file
*p
= file
->private_data
;
6437 mi
->event
= atomic_read(&md_event_count
);
6442 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6444 struct seq_file
*m
= filp
->private_data
;
6445 struct mdstat_info
*mi
= m
->private;
6448 poll_wait(filp
, &md_event_waiters
, wait
);
6450 /* always allow read */
6451 mask
= POLLIN
| POLLRDNORM
;
6453 if (mi
->event
!= atomic_read(&md_event_count
))
6454 mask
|= POLLERR
| POLLPRI
;
6458 static const struct file_operations md_seq_fops
= {
6459 .owner
= THIS_MODULE
,
6460 .open
= md_seq_open
,
6462 .llseek
= seq_lseek
,
6463 .release
= seq_release_private
,
6464 .poll
= mdstat_poll
,
6467 int register_md_personality(struct mdk_personality
*p
)
6469 spin_lock(&pers_lock
);
6470 list_add_tail(&p
->list
, &pers_list
);
6471 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6472 spin_unlock(&pers_lock
);
6476 int unregister_md_personality(struct mdk_personality
*p
)
6478 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6479 spin_lock(&pers_lock
);
6480 list_del_init(&p
->list
);
6481 spin_unlock(&pers_lock
);
6485 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6493 rdev_for_each_rcu(rdev
, mddev
) {
6494 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6495 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6496 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6497 atomic_read(&disk
->sync_io
);
6498 /* sync IO will cause sync_io to increase before the disk_stats
6499 * as sync_io is counted when a request starts, and
6500 * disk_stats is counted when it completes.
6501 * So resync activity will cause curr_events to be smaller than
6502 * when there was no such activity.
6503 * non-sync IO will cause disk_stat to increase without
6504 * increasing sync_io so curr_events will (eventually)
6505 * be larger than it was before. Once it becomes
6506 * substantially larger, the test below will cause
6507 * the array to appear non-idle, and resync will slow
6509 * If there is a lot of outstanding resync activity when
6510 * we set last_event to curr_events, then all that activity
6511 * completing might cause the array to appear non-idle
6512 * and resync will be slowed down even though there might
6513 * not have been non-resync activity. This will only
6514 * happen once though. 'last_events' will soon reflect
6515 * the state where there is little or no outstanding
6516 * resync requests, and further resync activity will
6517 * always make curr_events less than last_events.
6520 if (init
|| curr_events
- rdev
->last_events
> 64) {
6521 rdev
->last_events
= curr_events
;
6529 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6531 /* another "blocks" (512byte) blocks have been synced */
6532 atomic_sub(blocks
, &mddev
->recovery_active
);
6533 wake_up(&mddev
->recovery_wait
);
6535 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6536 md_wakeup_thread(mddev
->thread
);
6537 // stop recovery, signal do_sync ....
6542 /* md_write_start(mddev, bi)
6543 * If we need to update some array metadata (e.g. 'active' flag
6544 * in superblock) before writing, schedule a superblock update
6545 * and wait for it to complete.
6547 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6550 if (bio_data_dir(bi
) != WRITE
)
6553 BUG_ON(mddev
->ro
== 1);
6554 if (mddev
->ro
== 2) {
6555 /* need to switch to read/write */
6557 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6558 md_wakeup_thread(mddev
->thread
);
6559 md_wakeup_thread(mddev
->sync_thread
);
6562 atomic_inc(&mddev
->writes_pending
);
6563 if (mddev
->safemode
== 1)
6564 mddev
->safemode
= 0;
6565 if (mddev
->in_sync
) {
6566 spin_lock_irq(&mddev
->write_lock
);
6567 if (mddev
->in_sync
) {
6569 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6570 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6571 md_wakeup_thread(mddev
->thread
);
6574 spin_unlock_irq(&mddev
->write_lock
);
6577 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6578 wait_event(mddev
->sb_wait
,
6579 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6582 void md_write_end(mddev_t
*mddev
)
6584 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6585 if (mddev
->safemode
== 2)
6586 md_wakeup_thread(mddev
->thread
);
6587 else if (mddev
->safemode_delay
)
6588 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6592 /* md_allow_write(mddev)
6593 * Calling this ensures that the array is marked 'active' so that writes
6594 * may proceed without blocking. It is important to call this before
6595 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6596 * Must be called with mddev_lock held.
6598 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6599 * is dropped, so return -EAGAIN after notifying userspace.
6601 int md_allow_write(mddev_t
*mddev
)
6607 if (!mddev
->pers
->sync_request
)
6610 spin_lock_irq(&mddev
->write_lock
);
6611 if (mddev
->in_sync
) {
6613 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6614 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6615 if (mddev
->safemode_delay
&&
6616 mddev
->safemode
== 0)
6617 mddev
->safemode
= 1;
6618 spin_unlock_irq(&mddev
->write_lock
);
6619 md_update_sb(mddev
, 0);
6620 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6622 spin_unlock_irq(&mddev
->write_lock
);
6624 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6629 EXPORT_SYMBOL_GPL(md_allow_write
);
6631 void md_unplug(mddev_t
*mddev
)
6634 blk_unplug(mddev
->queue
);
6636 mddev
->plug
->unplug_fn(mddev
->plug
);
6639 #define SYNC_MARKS 10
6640 #define SYNC_MARK_STEP (3*HZ)
6641 void md_do_sync(mddev_t
*mddev
)
6644 unsigned int currspeed
= 0,
6646 sector_t max_sectors
,j
, io_sectors
;
6647 unsigned long mark
[SYNC_MARKS
];
6648 sector_t mark_cnt
[SYNC_MARKS
];
6650 struct list_head
*tmp
;
6651 sector_t last_check
;
6656 /* just incase thread restarts... */
6657 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6659 if (mddev
->ro
) /* never try to sync a read-only array */
6662 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6663 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6664 desc
= "data-check";
6665 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6666 desc
= "requested-resync";
6669 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6674 /* we overload curr_resync somewhat here.
6675 * 0 == not engaged in resync at all
6676 * 2 == checking that there is no conflict with another sync
6677 * 1 == like 2, but have yielded to allow conflicting resync to
6679 * other == active in resync - this many blocks
6681 * Before starting a resync we must have set curr_resync to
6682 * 2, and then checked that every "conflicting" array has curr_resync
6683 * less than ours. When we find one that is the same or higher
6684 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6685 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6686 * This will mean we have to start checking from the beginning again.
6691 mddev
->curr_resync
= 2;
6694 if (kthread_should_stop())
6695 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6697 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6699 for_each_mddev(mddev2
, tmp
) {
6700 if (mddev2
== mddev
)
6702 if (!mddev
->parallel_resync
6703 && mddev2
->curr_resync
6704 && match_mddev_units(mddev
, mddev2
)) {
6706 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6707 /* arbitrarily yield */
6708 mddev
->curr_resync
= 1;
6709 wake_up(&resync_wait
);
6711 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6712 /* no need to wait here, we can wait the next
6713 * time 'round when curr_resync == 2
6716 /* We need to wait 'interruptible' so as not to
6717 * contribute to the load average, and not to
6718 * be caught by 'softlockup'
6720 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6721 if (!kthread_should_stop() &&
6722 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6723 printk(KERN_INFO
"md: delaying %s of %s"
6724 " until %s has finished (they"
6725 " share one or more physical units)\n",
6726 desc
, mdname(mddev
), mdname(mddev2
));
6728 if (signal_pending(current
))
6729 flush_signals(current
);
6731 finish_wait(&resync_wait
, &wq
);
6734 finish_wait(&resync_wait
, &wq
);
6737 } while (mddev
->curr_resync
< 2);
6740 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6741 /* resync follows the size requested by the personality,
6742 * which defaults to physical size, but can be virtual size
6744 max_sectors
= mddev
->resync_max_sectors
;
6745 mddev
->resync_mismatches
= 0;
6746 /* we don't use the checkpoint if there's a bitmap */
6747 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6748 j
= mddev
->resync_min
;
6749 else if (!mddev
->bitmap
)
6750 j
= mddev
->recovery_cp
;
6752 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6753 max_sectors
= mddev
->dev_sectors
;
6755 /* recovery follows the physical size of devices */
6756 max_sectors
= mddev
->dev_sectors
;
6759 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6760 if (rdev
->raid_disk
>= 0 &&
6761 !test_bit(Faulty
, &rdev
->flags
) &&
6762 !test_bit(In_sync
, &rdev
->flags
) &&
6763 rdev
->recovery_offset
< j
)
6764 j
= rdev
->recovery_offset
;
6768 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
6769 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
6770 " %d KB/sec/disk.\n", speed_min(mddev
));
6771 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
6772 "(but not more than %d KB/sec) for %s.\n",
6773 speed_max(mddev
), desc
);
6775 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
6778 for (m
= 0; m
< SYNC_MARKS
; m
++) {
6780 mark_cnt
[m
] = io_sectors
;
6783 mddev
->resync_mark
= mark
[last_mark
];
6784 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
6787 * Tune reconstruction:
6789 window
= 32*(PAGE_SIZE
/512);
6790 printk(KERN_INFO
"md: using %dk window, over a total of %llu blocks.\n",
6791 window
/2,(unsigned long long) max_sectors
/2);
6793 atomic_set(&mddev
->recovery_active
, 0);
6798 "md: resuming %s of %s from checkpoint.\n",
6799 desc
, mdname(mddev
));
6800 mddev
->curr_resync
= j
;
6802 mddev
->curr_resync_completed
= mddev
->curr_resync
;
6804 while (j
< max_sectors
) {
6809 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
6810 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
6811 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
6812 > (max_sectors
>> 4)) ||
6813 (j
- mddev
->curr_resync_completed
)*2
6814 >= mddev
->resync_max
- mddev
->curr_resync_completed
6816 /* time to update curr_resync_completed */
6818 wait_event(mddev
->recovery_wait
,
6819 atomic_read(&mddev
->recovery_active
) == 0);
6820 mddev
->curr_resync_completed
=
6822 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6823 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6826 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
6827 /* As this condition is controlled by user-space,
6828 * we can block indefinitely, so use '_interruptible'
6829 * to avoid triggering warnings.
6831 flush_signals(current
); /* just in case */
6832 wait_event_interruptible(mddev
->recovery_wait
,
6833 mddev
->resync_max
> j
6834 || kthread_should_stop());
6837 if (kthread_should_stop())
6840 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
6841 currspeed
< speed_min(mddev
));
6843 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6847 if (!skipped
) { /* actual IO requested */
6848 io_sectors
+= sectors
;
6849 atomic_add(sectors
, &mddev
->recovery_active
);
6853 if (j
>1) mddev
->curr_resync
= j
;
6854 mddev
->curr_mark_cnt
= io_sectors
;
6855 if (last_check
== 0)
6856 /* this is the earliers that rebuilt will be
6857 * visible in /proc/mdstat
6859 md_new_event(mddev
);
6861 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
6864 last_check
= io_sectors
;
6866 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6870 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
6872 int next
= (last_mark
+1) % SYNC_MARKS
;
6874 mddev
->resync_mark
= mark
[next
];
6875 mddev
->resync_mark_cnt
= mark_cnt
[next
];
6876 mark
[next
] = jiffies
;
6877 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
6882 if (kthread_should_stop())
6887 * this loop exits only if either when we are slower than
6888 * the 'hard' speed limit, or the system was IO-idle for
6890 * the system might be non-idle CPU-wise, but we only care
6891 * about not overloading the IO subsystem. (things like an
6892 * e2fsck being done on the RAID array should execute fast)
6897 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
6898 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
6900 if (currspeed
> speed_min(mddev
)) {
6901 if ((currspeed
> speed_max(mddev
)) ||
6902 !is_mddev_idle(mddev
, 0)) {
6908 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
6910 * this also signals 'finished resyncing' to md_stop
6915 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
6917 /* tell personality that we are finished */
6918 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
6920 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
6921 mddev
->curr_resync
> 2) {
6922 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6923 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6924 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
6926 "md: checkpointing %s of %s.\n",
6927 desc
, mdname(mddev
));
6928 mddev
->recovery_cp
= mddev
->curr_resync
;
6931 mddev
->recovery_cp
= MaxSector
;
6933 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6934 mddev
->curr_resync
= MaxSector
;
6936 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6937 if (rdev
->raid_disk
>= 0 &&
6938 mddev
->delta_disks
>= 0 &&
6939 !test_bit(Faulty
, &rdev
->flags
) &&
6940 !test_bit(In_sync
, &rdev
->flags
) &&
6941 rdev
->recovery_offset
< mddev
->curr_resync
)
6942 rdev
->recovery_offset
= mddev
->curr_resync
;
6946 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
6949 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
6950 /* We completed so min/max setting can be forgotten if used. */
6951 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6952 mddev
->resync_min
= 0;
6953 mddev
->resync_max
= MaxSector
;
6954 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6955 mddev
->resync_min
= mddev
->curr_resync_completed
;
6956 mddev
->curr_resync
= 0;
6957 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6958 mddev
->curr_resync_completed
= 0;
6959 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
6960 wake_up(&resync_wait
);
6961 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
6962 md_wakeup_thread(mddev
->thread
);
6967 * got a signal, exit.
6970 "md: md_do_sync() got signal ... exiting\n");
6971 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6975 EXPORT_SYMBOL_GPL(md_do_sync
);
6978 static int remove_and_add_spares(mddev_t
*mddev
)
6983 mddev
->curr_resync_completed
= 0;
6985 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
6986 if (rdev
->raid_disk
>= 0 &&
6987 !test_bit(Blocked
, &rdev
->flags
) &&
6988 (test_bit(Faulty
, &rdev
->flags
) ||
6989 ! test_bit(In_sync
, &rdev
->flags
)) &&
6990 atomic_read(&rdev
->nr_pending
)==0) {
6991 if (mddev
->pers
->hot_remove_disk(
6992 mddev
, rdev
->raid_disk
)==0) {
6994 sprintf(nm
,"rd%d", rdev
->raid_disk
);
6995 sysfs_remove_link(&mddev
->kobj
, nm
);
6996 rdev
->raid_disk
= -1;
7000 if (mddev
->degraded
&& ! mddev
->ro
&& !mddev
->recovery_disabled
) {
7001 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7002 if (rdev
->raid_disk
>= 0 &&
7003 !test_bit(In_sync
, &rdev
->flags
) &&
7004 !test_bit(Blocked
, &rdev
->flags
))
7006 if (rdev
->raid_disk
< 0
7007 && !test_bit(Faulty
, &rdev
->flags
)) {
7008 rdev
->recovery_offset
= 0;
7010 hot_add_disk(mddev
, rdev
) == 0) {
7012 sprintf(nm
, "rd%d", rdev
->raid_disk
);
7013 if (sysfs_create_link(&mddev
->kobj
,
7015 /* failure here is OK */;
7017 md_new_event(mddev
);
7018 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7027 * This routine is regularly called by all per-raid-array threads to
7028 * deal with generic issues like resync and super-block update.
7029 * Raid personalities that don't have a thread (linear/raid0) do not
7030 * need this as they never do any recovery or update the superblock.
7032 * It does not do any resync itself, but rather "forks" off other threads
7033 * to do that as needed.
7034 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7035 * "->recovery" and create a thread at ->sync_thread.
7036 * When the thread finishes it sets MD_RECOVERY_DONE
7037 * and wakeups up this thread which will reap the thread and finish up.
7038 * This thread also removes any faulty devices (with nr_pending == 0).
7040 * The overall approach is:
7041 * 1/ if the superblock needs updating, update it.
7042 * 2/ If a recovery thread is running, don't do anything else.
7043 * 3/ If recovery has finished, clean up, possibly marking spares active.
7044 * 4/ If there are any faulty devices, remove them.
7045 * 5/ If array is degraded, try to add spares devices
7046 * 6/ If array has spares or is not in-sync, start a resync thread.
7048 void md_check_recovery(mddev_t
*mddev
)
7054 bitmap_daemon_work(mddev
);
7059 if (signal_pending(current
)) {
7060 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7061 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7063 mddev
->safemode
= 2;
7065 flush_signals(current
);
7068 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7071 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7072 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7073 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7074 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7075 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7076 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7080 if (mddev_trylock(mddev
)) {
7084 /* Only thing we do on a ro array is remove
7087 remove_and_add_spares(mddev
);
7088 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7092 if (!mddev
->external
) {
7094 spin_lock_irq(&mddev
->write_lock
);
7095 if (mddev
->safemode
&&
7096 !atomic_read(&mddev
->writes_pending
) &&
7098 mddev
->recovery_cp
== MaxSector
) {
7101 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7103 if (mddev
->safemode
== 1)
7104 mddev
->safemode
= 0;
7105 spin_unlock_irq(&mddev
->write_lock
);
7107 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7111 md_update_sb(mddev
, 0);
7113 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7114 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7115 /* resync/recovery still happening */
7116 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7119 if (mddev
->sync_thread
) {
7120 /* resync has finished, collect result */
7121 md_unregister_thread(mddev
->sync_thread
);
7122 mddev
->sync_thread
= NULL
;
7123 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7124 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7126 /* activate any spares */
7127 if (mddev
->pers
->spare_active(mddev
))
7128 sysfs_notify(&mddev
->kobj
, NULL
,
7131 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7132 mddev
->pers
->finish_reshape
)
7133 mddev
->pers
->finish_reshape(mddev
);
7134 md_update_sb(mddev
, 1);
7136 /* if array is no-longer degraded, then any saved_raid_disk
7137 * information must be scrapped
7139 if (!mddev
->degraded
)
7140 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7141 rdev
->saved_raid_disk
= -1;
7143 mddev
->recovery
= 0;
7144 /* flag recovery needed just to double check */
7145 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7146 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7147 md_new_event(mddev
);
7150 /* Set RUNNING before clearing NEEDED to avoid
7151 * any transients in the value of "sync_action".
7153 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7154 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7155 /* Clear some bits that don't mean anything, but
7158 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7159 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7161 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7163 /* no recovery is running.
7164 * remove any failed drives, then
7165 * add spares if possible.
7166 * Spare are also removed and re-added, to allow
7167 * the personality to fail the re-add.
7170 if (mddev
->reshape_position
!= MaxSector
) {
7171 if (mddev
->pers
->check_reshape
== NULL
||
7172 mddev
->pers
->check_reshape(mddev
) != 0)
7173 /* Cannot proceed */
7175 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7176 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7177 } else if ((spares
= remove_and_add_spares(mddev
))) {
7178 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7179 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7180 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7181 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7182 } else if (mddev
->recovery_cp
< MaxSector
) {
7183 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7184 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7185 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7186 /* nothing to be done ... */
7189 if (mddev
->pers
->sync_request
) {
7190 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7191 /* We are adding a device or devices to an array
7192 * which has the bitmap stored on all devices.
7193 * So make sure all bitmap pages get written
7195 bitmap_write_all(mddev
->bitmap
);
7197 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7200 if (!mddev
->sync_thread
) {
7201 printk(KERN_ERR
"%s: could not start resync"
7204 /* leave the spares where they are, it shouldn't hurt */
7205 mddev
->recovery
= 0;
7207 md_wakeup_thread(mddev
->sync_thread
);
7208 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7209 md_new_event(mddev
);
7212 if (!mddev
->sync_thread
) {
7213 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7214 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7216 if (mddev
->sysfs_action
)
7217 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7219 mddev_unlock(mddev
);
7223 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7225 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7226 wait_event_timeout(rdev
->blocked_wait
,
7227 !test_bit(Blocked
, &rdev
->flags
),
7228 msecs_to_jiffies(5000));
7229 rdev_dec_pending(rdev
, mddev
);
7231 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7233 static int md_notify_reboot(struct notifier_block
*this,
7234 unsigned long code
, void *x
)
7236 struct list_head
*tmp
;
7239 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7241 printk(KERN_INFO
"md: stopping all md devices.\n");
7243 for_each_mddev(mddev
, tmp
)
7244 if (mddev_trylock(mddev
)) {
7245 /* Force a switch to readonly even array
7246 * appears to still be in use. Hence
7249 md_set_readonly(mddev
, 100);
7250 mddev_unlock(mddev
);
7253 * certain more exotic SCSI devices are known to be
7254 * volatile wrt too early system reboots. While the
7255 * right place to handle this issue is the given
7256 * driver, we do want to have a safe RAID driver ...
7263 static struct notifier_block md_notifier
= {
7264 .notifier_call
= md_notify_reboot
,
7266 .priority
= INT_MAX
, /* before any real devices */
7269 static void md_geninit(void)
7271 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
7273 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
7276 static int __init
md_init(void)
7280 md_wq
= alloc_workqueue("md", WQ_RESCUER
, 0);
7284 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
7288 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
7291 if ((ret
= register_blkdev(0, "mdp")) < 0)
7295 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7296 md_probe
, NULL
, NULL
);
7297 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
7298 md_probe
, NULL
, NULL
);
7300 register_reboot_notifier(&md_notifier
);
7301 raid_table_header
= register_sysctl_table(raid_root_table
);
7307 unregister_blkdev(MD_MAJOR
, "md");
7309 destroy_workqueue(md_misc_wq
);
7311 destroy_workqueue(md_wq
);
7319 * Searches all registered partitions for autorun RAID arrays
7323 static LIST_HEAD(all_detected_devices
);
7324 struct detected_devices_node
{
7325 struct list_head list
;
7329 void md_autodetect_dev(dev_t dev
)
7331 struct detected_devices_node
*node_detected_dev
;
7333 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
7334 if (node_detected_dev
) {
7335 node_detected_dev
->dev
= dev
;
7336 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
7338 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
7339 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
7344 static void autostart_arrays(int part
)
7347 struct detected_devices_node
*node_detected_dev
;
7349 int i_scanned
, i_passed
;
7354 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
7356 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
7358 node_detected_dev
= list_entry(all_detected_devices
.next
,
7359 struct detected_devices_node
, list
);
7360 list_del(&node_detected_dev
->list
);
7361 dev
= node_detected_dev
->dev
;
7362 kfree(node_detected_dev
);
7363 rdev
= md_import_device(dev
,0, 90);
7367 if (test_bit(Faulty
, &rdev
->flags
)) {
7371 set_bit(AutoDetected
, &rdev
->flags
);
7372 list_add(&rdev
->same_set
, &pending_raid_disks
);
7376 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
7377 i_scanned
, i_passed
);
7379 autorun_devices(part
);
7382 #endif /* !MODULE */
7384 static __exit
void md_exit(void)
7387 struct list_head
*tmp
;
7389 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
7390 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
7392 unregister_blkdev(MD_MAJOR
,"md");
7393 unregister_blkdev(mdp_major
, "mdp");
7394 unregister_reboot_notifier(&md_notifier
);
7395 unregister_sysctl_table(raid_table_header
);
7396 remove_proc_entry("mdstat", NULL
);
7397 for_each_mddev(mddev
, tmp
) {
7398 export_array(mddev
);
7399 mddev
->hold_active
= 0;
7401 destroy_workqueue(md_misc_wq
);
7402 destroy_workqueue(md_wq
);
7405 subsys_initcall(md_init
);
7406 module_exit(md_exit
)
7408 static int get_ro(char *buffer
, struct kernel_param
*kp
)
7410 return sprintf(buffer
, "%d", start_readonly
);
7412 static int set_ro(const char *val
, struct kernel_param
*kp
)
7415 int num
= simple_strtoul(val
, &e
, 10);
7416 if (*val
&& (*e
== '\0' || *e
== '\n')) {
7417 start_readonly
= num
;
7423 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
7424 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
7426 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
7428 EXPORT_SYMBOL(register_md_personality
);
7429 EXPORT_SYMBOL(unregister_md_personality
);
7430 EXPORT_SYMBOL(md_error
);
7431 EXPORT_SYMBOL(md_done_sync
);
7432 EXPORT_SYMBOL(md_write_start
);
7433 EXPORT_SYMBOL(md_write_end
);
7434 EXPORT_SYMBOL(md_register_thread
);
7435 EXPORT_SYMBOL(md_unregister_thread
);
7436 EXPORT_SYMBOL(md_wakeup_thread
);
7437 EXPORT_SYMBOL(md_check_recovery
);
7438 MODULE_LICENSE("GPL");
7439 MODULE_DESCRIPTION("MD RAID framework");
7441 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);