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
;
290 unsigned int sectors
;
292 if (mddev
== NULL
|| mddev
->pers
== NULL
297 smp_rmb(); /* Ensure implications of 'active' are visible */
299 if (mddev
->suspended
) {
302 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
303 TASK_UNINTERRUPTIBLE
);
304 if (!mddev
->suspended
)
310 finish_wait(&mddev
->sb_wait
, &__wait
);
312 atomic_inc(&mddev
->active_io
);
316 * save the sectors now since our bio can
317 * go away inside make_request
319 sectors
= bio_sectors(bio
);
320 rv
= mddev
->pers
->make_request(mddev
, bio
);
322 cpu
= part_stat_lock();
323 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
324 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
327 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
328 wake_up(&mddev
->sb_wait
);
333 /* mddev_suspend makes sure no new requests are submitted
334 * to the device, and that any requests that have been submitted
335 * are completely handled.
336 * Once ->stop is called and completes, the module will be completely
339 void mddev_suspend(mddev_t
*mddev
)
341 BUG_ON(mddev
->suspended
);
342 mddev
->suspended
= 1;
344 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
345 mddev
->pers
->quiesce(mddev
, 1);
347 EXPORT_SYMBOL_GPL(mddev_suspend
);
349 void mddev_resume(mddev_t
*mddev
)
351 mddev
->suspended
= 0;
352 wake_up(&mddev
->sb_wait
);
353 mddev
->pers
->quiesce(mddev
, 0);
355 md_wakeup_thread(mddev
->thread
);
356 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
358 EXPORT_SYMBOL_GPL(mddev_resume
);
360 int mddev_congested(mddev_t
*mddev
, int bits
)
362 return mddev
->suspended
;
364 EXPORT_SYMBOL(mddev_congested
);
367 * Generic flush handling for md
370 static void md_end_flush(struct bio
*bio
, int err
)
372 mdk_rdev_t
*rdev
= bio
->bi_private
;
373 mddev_t
*mddev
= rdev
->mddev
;
375 rdev_dec_pending(rdev
, mddev
);
377 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
378 /* The pre-request flush has finished */
379 queue_work(md_wq
, &mddev
->flush_work
);
384 static void md_submit_flush_data(struct work_struct
*ws
);
386 static void submit_flushes(struct work_struct
*ws
)
388 mddev_t
*mddev
= container_of(ws
, mddev_t
, flush_work
);
391 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
392 atomic_set(&mddev
->flush_pending
, 1);
394 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
395 if (rdev
->raid_disk
>= 0 &&
396 !test_bit(Faulty
, &rdev
->flags
)) {
397 /* Take two references, one is dropped
398 * when request finishes, one after
399 * we reclaim rcu_read_lock
402 atomic_inc(&rdev
->nr_pending
);
403 atomic_inc(&rdev
->nr_pending
);
405 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
406 bi
->bi_end_io
= md_end_flush
;
407 bi
->bi_private
= rdev
;
408 bi
->bi_bdev
= rdev
->bdev
;
409 atomic_inc(&mddev
->flush_pending
);
410 submit_bio(WRITE_FLUSH
, bi
);
412 rdev_dec_pending(rdev
, mddev
);
415 if (atomic_dec_and_test(&mddev
->flush_pending
))
416 queue_work(md_wq
, &mddev
->flush_work
);
419 static void md_submit_flush_data(struct work_struct
*ws
)
421 mddev_t
*mddev
= container_of(ws
, mddev_t
, flush_work
);
422 struct bio
*bio
= mddev
->flush_bio
;
424 if (bio
->bi_size
== 0)
425 /* an empty barrier - all done */
428 bio
->bi_rw
&= ~REQ_FLUSH
;
429 if (mddev
->pers
->make_request(mddev
, bio
))
430 generic_make_request(bio
);
433 mddev
->flush_bio
= NULL
;
434 wake_up(&mddev
->sb_wait
);
437 void md_flush_request(mddev_t
*mddev
, struct bio
*bio
)
439 spin_lock_irq(&mddev
->write_lock
);
440 wait_event_lock_irq(mddev
->sb_wait
,
442 mddev
->write_lock
, /*nothing*/);
443 mddev
->flush_bio
= bio
;
444 spin_unlock_irq(&mddev
->write_lock
);
446 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
447 queue_work(md_wq
, &mddev
->flush_work
);
449 EXPORT_SYMBOL(md_flush_request
);
451 /* Support for plugging.
452 * This mirrors the plugging support in request_queue, but does not
453 * require having a whole queue or request structures.
454 * We allocate an md_plug_cb for each md device and each thread it gets
455 * plugged on. This links tot the private plug_handle structure in the
456 * personality data where we keep a count of the number of outstanding
457 * plugs so other code can see if a plug is active.
460 struct blk_plug_cb cb
;
464 static void plugger_unplug(struct blk_plug_cb
*cb
)
466 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
467 if (atomic_dec_and_test(&mdcb
->mddev
->plug_cnt
))
468 md_wakeup_thread(mdcb
->mddev
->thread
);
472 /* Check that an unplug wakeup will come shortly.
473 * If not, wakeup the md thread immediately
475 int mddev_check_plugged(mddev_t
*mddev
)
477 struct blk_plug
*plug
= current
->plug
;
478 struct md_plug_cb
*mdcb
;
483 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
484 if (mdcb
->cb
.callback
== plugger_unplug
&&
485 mdcb
->mddev
== mddev
) {
486 /* Already on the list, move to top */
487 if (mdcb
!= list_first_entry(&plug
->cb_list
,
490 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
494 /* Not currently on the callback list */
495 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
500 mdcb
->cb
.callback
= plugger_unplug
;
501 atomic_inc(&mddev
->plug_cnt
);
502 list_add(&mdcb
->cb
.list
, &plug
->cb_list
);
505 EXPORT_SYMBOL_GPL(mddev_check_plugged
);
507 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
509 atomic_inc(&mddev
->active
);
513 static void mddev_delayed_delete(struct work_struct
*ws
);
515 static void mddev_put(mddev_t
*mddev
)
517 struct bio_set
*bs
= NULL
;
519 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
521 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
522 mddev
->ctime
== 0 && !mddev
->hold_active
) {
523 /* Array is not configured at all, and not held active,
525 list_del(&mddev
->all_mddevs
);
527 mddev
->bio_set
= NULL
;
528 if (mddev
->gendisk
) {
529 /* We did a probe so need to clean up. Call
530 * queue_work inside the spinlock so that
531 * flush_workqueue() after mddev_find will
532 * succeed in waiting for the work to be done.
534 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
535 queue_work(md_misc_wq
, &mddev
->del_work
);
539 spin_unlock(&all_mddevs_lock
);
544 void mddev_init(mddev_t
*mddev
)
546 mutex_init(&mddev
->open_mutex
);
547 mutex_init(&mddev
->reconfig_mutex
);
548 mutex_init(&mddev
->bitmap_info
.mutex
);
549 INIT_LIST_HEAD(&mddev
->disks
);
550 INIT_LIST_HEAD(&mddev
->all_mddevs
);
551 init_timer(&mddev
->safemode_timer
);
552 atomic_set(&mddev
->active
, 1);
553 atomic_set(&mddev
->openers
, 0);
554 atomic_set(&mddev
->active_io
, 0);
555 atomic_set(&mddev
->plug_cnt
, 0);
556 spin_lock_init(&mddev
->write_lock
);
557 atomic_set(&mddev
->flush_pending
, 0);
558 init_waitqueue_head(&mddev
->sb_wait
);
559 init_waitqueue_head(&mddev
->recovery_wait
);
560 mddev
->reshape_position
= MaxSector
;
561 mddev
->resync_min
= 0;
562 mddev
->resync_max
= MaxSector
;
563 mddev
->level
= LEVEL_NONE
;
565 EXPORT_SYMBOL_GPL(mddev_init
);
567 static mddev_t
* mddev_find(dev_t unit
)
569 mddev_t
*mddev
, *new = NULL
;
571 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
572 unit
&= ~((1<<MdpMinorShift
)-1);
575 spin_lock(&all_mddevs_lock
);
578 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
579 if (mddev
->unit
== unit
) {
581 spin_unlock(&all_mddevs_lock
);
587 list_add(&new->all_mddevs
, &all_mddevs
);
588 spin_unlock(&all_mddevs_lock
);
589 new->hold_active
= UNTIL_IOCTL
;
593 /* find an unused unit number */
594 static int next_minor
= 512;
595 int start
= next_minor
;
599 dev
= MKDEV(MD_MAJOR
, next_minor
);
601 if (next_minor
> MINORMASK
)
603 if (next_minor
== start
) {
604 /* Oh dear, all in use. */
605 spin_unlock(&all_mddevs_lock
);
611 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
612 if (mddev
->unit
== dev
) {
618 new->md_minor
= MINOR(dev
);
619 new->hold_active
= UNTIL_STOP
;
620 list_add(&new->all_mddevs
, &all_mddevs
);
621 spin_unlock(&all_mddevs_lock
);
624 spin_unlock(&all_mddevs_lock
);
626 new = kzalloc(sizeof(*new), GFP_KERNEL
);
631 if (MAJOR(unit
) == MD_MAJOR
)
632 new->md_minor
= MINOR(unit
);
634 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
641 static inline int mddev_lock(mddev_t
* mddev
)
643 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
646 static inline int mddev_is_locked(mddev_t
*mddev
)
648 return mutex_is_locked(&mddev
->reconfig_mutex
);
651 static inline int mddev_trylock(mddev_t
* mddev
)
653 return mutex_trylock(&mddev
->reconfig_mutex
);
656 static struct attribute_group md_redundancy_group
;
658 static void mddev_unlock(mddev_t
* mddev
)
660 if (mddev
->to_remove
) {
661 /* These cannot be removed under reconfig_mutex as
662 * an access to the files will try to take reconfig_mutex
663 * while holding the file unremovable, which leads to
665 * So hold set sysfs_active while the remove in happeing,
666 * and anything else which might set ->to_remove or my
667 * otherwise change the sysfs namespace will fail with
668 * -EBUSY if sysfs_active is still set.
669 * We set sysfs_active under reconfig_mutex and elsewhere
670 * test it under the same mutex to ensure its correct value
673 struct attribute_group
*to_remove
= mddev
->to_remove
;
674 mddev
->to_remove
= NULL
;
675 mddev
->sysfs_active
= 1;
676 mutex_unlock(&mddev
->reconfig_mutex
);
678 if (mddev
->kobj
.sd
) {
679 if (to_remove
!= &md_redundancy_group
)
680 sysfs_remove_group(&mddev
->kobj
, to_remove
);
681 if (mddev
->pers
== NULL
||
682 mddev
->pers
->sync_request
== NULL
) {
683 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
684 if (mddev
->sysfs_action
)
685 sysfs_put(mddev
->sysfs_action
);
686 mddev
->sysfs_action
= NULL
;
689 mddev
->sysfs_active
= 0;
691 mutex_unlock(&mddev
->reconfig_mutex
);
693 md_wakeup_thread(mddev
->thread
);
696 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
700 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
701 if (rdev
->desc_nr
== nr
)
707 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
711 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
712 if (rdev
->bdev
->bd_dev
== dev
)
718 static struct mdk_personality
*find_pers(int level
, char *clevel
)
720 struct mdk_personality
*pers
;
721 list_for_each_entry(pers
, &pers_list
, list
) {
722 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
724 if (strcmp(pers
->name
, clevel
)==0)
730 /* return the offset of the super block in 512byte sectors */
731 static inline sector_t
calc_dev_sboffset(mdk_rdev_t
*rdev
)
733 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
734 return MD_NEW_SIZE_SECTORS(num_sectors
);
737 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
742 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
743 if (!rdev
->sb_page
) {
744 printk(KERN_ALERT
"md: out of memory.\n");
751 static void free_disk_sb(mdk_rdev_t
* rdev
)
754 put_page(rdev
->sb_page
);
756 rdev
->sb_page
= NULL
;
761 put_page(rdev
->bb_page
);
762 rdev
->bb_page
= NULL
;
767 static void super_written(struct bio
*bio
, int error
)
769 mdk_rdev_t
*rdev
= bio
->bi_private
;
770 mddev_t
*mddev
= rdev
->mddev
;
772 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
773 printk("md: super_written gets error=%d, uptodate=%d\n",
774 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
775 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
776 md_error(mddev
, rdev
);
779 if (atomic_dec_and_test(&mddev
->pending_writes
))
780 wake_up(&mddev
->sb_wait
);
784 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
785 sector_t sector
, int size
, struct page
*page
)
787 /* write first size bytes of page to sector of rdev
788 * Increment mddev->pending_writes before returning
789 * and decrement it on completion, waking up sb_wait
790 * if zero is reached.
791 * If an error occurred, call md_error
793 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
795 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
796 bio
->bi_sector
= sector
;
797 bio_add_page(bio
, page
, size
, 0);
798 bio
->bi_private
= rdev
;
799 bio
->bi_end_io
= super_written
;
801 atomic_inc(&mddev
->pending_writes
);
802 submit_bio(REQ_WRITE
| REQ_SYNC
| REQ_FLUSH
| REQ_FUA
, bio
);
805 void md_super_wait(mddev_t
*mddev
)
807 /* wait for all superblock writes that were scheduled to complete */
810 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
811 if (atomic_read(&mddev
->pending_writes
)==0)
815 finish_wait(&mddev
->sb_wait
, &wq
);
818 static void bi_complete(struct bio
*bio
, int error
)
820 complete((struct completion
*)bio
->bi_private
);
823 int sync_page_io(mdk_rdev_t
*rdev
, sector_t sector
, int size
,
824 struct page
*page
, int rw
, bool metadata_op
)
826 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
827 struct completion event
;
832 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
833 rdev
->meta_bdev
: rdev
->bdev
;
835 bio
->bi_sector
= sector
+ rdev
->sb_start
;
837 bio
->bi_sector
= sector
+ rdev
->data_offset
;
838 bio_add_page(bio
, page
, size
, 0);
839 init_completion(&event
);
840 bio
->bi_private
= &event
;
841 bio
->bi_end_io
= bi_complete
;
843 wait_for_completion(&event
);
845 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
849 EXPORT_SYMBOL_GPL(sync_page_io
);
851 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
853 char b
[BDEVNAME_SIZE
];
854 if (!rdev
->sb_page
) {
862 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
868 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
869 bdevname(rdev
->bdev
,b
));
873 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
875 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
876 sb1
->set_uuid1
== sb2
->set_uuid1
&&
877 sb1
->set_uuid2
== sb2
->set_uuid2
&&
878 sb1
->set_uuid3
== sb2
->set_uuid3
;
881 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
884 mdp_super_t
*tmp1
, *tmp2
;
886 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
887 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
889 if (!tmp1
|| !tmp2
) {
891 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
899 * nr_disks is not constant
904 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
912 static u32
md_csum_fold(u32 csum
)
914 csum
= (csum
& 0xffff) + (csum
>> 16);
915 return (csum
& 0xffff) + (csum
>> 16);
918 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
921 u32
*sb32
= (u32
*)sb
;
923 unsigned int disk_csum
, csum
;
925 disk_csum
= sb
->sb_csum
;
928 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
930 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
934 /* This used to use csum_partial, which was wrong for several
935 * reasons including that different results are returned on
936 * different architectures. It isn't critical that we get exactly
937 * the same return value as before (we always csum_fold before
938 * testing, and that removes any differences). However as we
939 * know that csum_partial always returned a 16bit value on
940 * alphas, do a fold to maximise conformity to previous behaviour.
942 sb
->sb_csum
= md_csum_fold(disk_csum
);
944 sb
->sb_csum
= disk_csum
;
951 * Handle superblock details.
952 * We want to be able to handle multiple superblock formats
953 * so we have a common interface to them all, and an array of
954 * different handlers.
955 * We rely on user-space to write the initial superblock, and support
956 * reading and updating of superblocks.
957 * Interface methods are:
958 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
959 * loads and validates a superblock on dev.
960 * if refdev != NULL, compare superblocks on both devices
962 * 0 - dev has a superblock that is compatible with refdev
963 * 1 - dev has a superblock that is compatible and newer than refdev
964 * so dev should be used as the refdev in future
965 * -EINVAL superblock incompatible or invalid
966 * -othererror e.g. -EIO
968 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
969 * Verify that dev is acceptable into mddev.
970 * The first time, mddev->raid_disks will be 0, and data from
971 * dev should be merged in. Subsequent calls check that dev
972 * is new enough. Return 0 or -EINVAL
974 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
975 * Update the superblock for rdev with data in mddev
976 * This does not write to disc.
982 struct module
*owner
;
983 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
985 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
986 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
987 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
988 sector_t num_sectors
);
992 * Check that the given mddev has no bitmap.
994 * This function is called from the run method of all personalities that do not
995 * support bitmaps. It prints an error message and returns non-zero if mddev
996 * has a bitmap. Otherwise, it returns 0.
999 int md_check_no_bitmap(mddev_t
*mddev
)
1001 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1003 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1004 mdname(mddev
), mddev
->pers
->name
);
1007 EXPORT_SYMBOL(md_check_no_bitmap
);
1010 * load_super for 0.90.0
1012 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1014 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1019 * Calculate the position of the superblock (512byte sectors),
1020 * it's at the end of the disk.
1022 * It also happens to be a multiple of 4Kb.
1024 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1026 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1027 if (ret
) return ret
;
1031 bdevname(rdev
->bdev
, b
);
1032 sb
= page_address(rdev
->sb_page
);
1034 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1035 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1040 if (sb
->major_version
!= 0 ||
1041 sb
->minor_version
< 90 ||
1042 sb
->minor_version
> 91) {
1043 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1044 sb
->major_version
, sb
->minor_version
,
1049 if (sb
->raid_disks
<= 0)
1052 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1053 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1058 rdev
->preferred_minor
= sb
->md_minor
;
1059 rdev
->data_offset
= 0;
1060 rdev
->sb_size
= MD_SB_BYTES
;
1061 rdev
->badblocks
.shift
= -1;
1063 if (sb
->level
== LEVEL_MULTIPATH
)
1066 rdev
->desc_nr
= sb
->this_disk
.number
;
1072 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1073 if (!uuid_equal(refsb
, sb
)) {
1074 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1075 b
, bdevname(refdev
->bdev
,b2
));
1078 if (!sb_equal(refsb
, sb
)) {
1079 printk(KERN_WARNING
"md: %s has same UUID"
1080 " but different superblock to %s\n",
1081 b
, bdevname(refdev
->bdev
, b2
));
1085 ev2
= md_event(refsb
);
1091 rdev
->sectors
= rdev
->sb_start
;
1093 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
1094 /* "this cannot possibly happen" ... */
1102 * validate_super for 0.90.0
1104 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1107 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1108 __u64 ev1
= md_event(sb
);
1110 rdev
->raid_disk
= -1;
1111 clear_bit(Faulty
, &rdev
->flags
);
1112 clear_bit(In_sync
, &rdev
->flags
);
1113 clear_bit(WriteMostly
, &rdev
->flags
);
1115 if (mddev
->raid_disks
== 0) {
1116 mddev
->major_version
= 0;
1117 mddev
->minor_version
= sb
->minor_version
;
1118 mddev
->patch_version
= sb
->patch_version
;
1119 mddev
->external
= 0;
1120 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1121 mddev
->ctime
= sb
->ctime
;
1122 mddev
->utime
= sb
->utime
;
1123 mddev
->level
= sb
->level
;
1124 mddev
->clevel
[0] = 0;
1125 mddev
->layout
= sb
->layout
;
1126 mddev
->raid_disks
= sb
->raid_disks
;
1127 mddev
->dev_sectors
= sb
->size
* 2;
1128 mddev
->events
= ev1
;
1129 mddev
->bitmap_info
.offset
= 0;
1130 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1132 if (mddev
->minor_version
>= 91) {
1133 mddev
->reshape_position
= sb
->reshape_position
;
1134 mddev
->delta_disks
= sb
->delta_disks
;
1135 mddev
->new_level
= sb
->new_level
;
1136 mddev
->new_layout
= sb
->new_layout
;
1137 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1139 mddev
->reshape_position
= MaxSector
;
1140 mddev
->delta_disks
= 0;
1141 mddev
->new_level
= mddev
->level
;
1142 mddev
->new_layout
= mddev
->layout
;
1143 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1146 if (sb
->state
& (1<<MD_SB_CLEAN
))
1147 mddev
->recovery_cp
= MaxSector
;
1149 if (sb
->events_hi
== sb
->cp_events_hi
&&
1150 sb
->events_lo
== sb
->cp_events_lo
) {
1151 mddev
->recovery_cp
= sb
->recovery_cp
;
1153 mddev
->recovery_cp
= 0;
1156 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1157 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1158 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1159 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1161 mddev
->max_disks
= MD_SB_DISKS
;
1163 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1164 mddev
->bitmap_info
.file
== NULL
)
1165 mddev
->bitmap_info
.offset
=
1166 mddev
->bitmap_info
.default_offset
;
1168 } else if (mddev
->pers
== NULL
) {
1169 /* Insist on good event counter while assembling, except
1170 * for spares (which don't need an event count) */
1172 if (sb
->disks
[rdev
->desc_nr
].state
& (
1173 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1174 if (ev1
< mddev
->events
)
1176 } else if (mddev
->bitmap
) {
1177 /* if adding to array with a bitmap, then we can accept an
1178 * older device ... but not too old.
1180 if (ev1
< mddev
->bitmap
->events_cleared
)
1183 if (ev1
< mddev
->events
)
1184 /* just a hot-add of a new device, leave raid_disk at -1 */
1188 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1189 desc
= sb
->disks
+ rdev
->desc_nr
;
1191 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1192 set_bit(Faulty
, &rdev
->flags
);
1193 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1194 desc->raid_disk < mddev->raid_disks */) {
1195 set_bit(In_sync
, &rdev
->flags
);
1196 rdev
->raid_disk
= desc
->raid_disk
;
1197 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1198 /* active but not in sync implies recovery up to
1199 * reshape position. We don't know exactly where
1200 * that is, so set to zero for now */
1201 if (mddev
->minor_version
>= 91) {
1202 rdev
->recovery_offset
= 0;
1203 rdev
->raid_disk
= desc
->raid_disk
;
1206 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1207 set_bit(WriteMostly
, &rdev
->flags
);
1208 } else /* MULTIPATH are always insync */
1209 set_bit(In_sync
, &rdev
->flags
);
1214 * sync_super for 0.90.0
1216 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1220 int next_spare
= mddev
->raid_disks
;
1223 /* make rdev->sb match mddev data..
1226 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1227 * 3/ any empty disks < next_spare become removed
1229 * disks[0] gets initialised to REMOVED because
1230 * we cannot be sure from other fields if it has
1231 * been initialised or not.
1234 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1236 rdev
->sb_size
= MD_SB_BYTES
;
1238 sb
= page_address(rdev
->sb_page
);
1240 memset(sb
, 0, sizeof(*sb
));
1242 sb
->md_magic
= MD_SB_MAGIC
;
1243 sb
->major_version
= mddev
->major_version
;
1244 sb
->patch_version
= mddev
->patch_version
;
1245 sb
->gvalid_words
= 0; /* ignored */
1246 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1247 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1248 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1249 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1251 sb
->ctime
= mddev
->ctime
;
1252 sb
->level
= mddev
->level
;
1253 sb
->size
= mddev
->dev_sectors
/ 2;
1254 sb
->raid_disks
= mddev
->raid_disks
;
1255 sb
->md_minor
= mddev
->md_minor
;
1256 sb
->not_persistent
= 0;
1257 sb
->utime
= mddev
->utime
;
1259 sb
->events_hi
= (mddev
->events
>>32);
1260 sb
->events_lo
= (u32
)mddev
->events
;
1262 if (mddev
->reshape_position
== MaxSector
)
1263 sb
->minor_version
= 90;
1265 sb
->minor_version
= 91;
1266 sb
->reshape_position
= mddev
->reshape_position
;
1267 sb
->new_level
= mddev
->new_level
;
1268 sb
->delta_disks
= mddev
->delta_disks
;
1269 sb
->new_layout
= mddev
->new_layout
;
1270 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1272 mddev
->minor_version
= sb
->minor_version
;
1275 sb
->recovery_cp
= mddev
->recovery_cp
;
1276 sb
->cp_events_hi
= (mddev
->events
>>32);
1277 sb
->cp_events_lo
= (u32
)mddev
->events
;
1278 if (mddev
->recovery_cp
== MaxSector
)
1279 sb
->state
= (1<< MD_SB_CLEAN
);
1281 sb
->recovery_cp
= 0;
1283 sb
->layout
= mddev
->layout
;
1284 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1286 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1287 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1289 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1290 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1293 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1295 if (rdev2
->raid_disk
>= 0 &&
1296 sb
->minor_version
>= 91)
1297 /* we have nowhere to store the recovery_offset,
1298 * but if it is not below the reshape_position,
1299 * we can piggy-back on that.
1302 if (rdev2
->raid_disk
< 0 ||
1303 test_bit(Faulty
, &rdev2
->flags
))
1306 desc_nr
= rdev2
->raid_disk
;
1308 desc_nr
= next_spare
++;
1309 rdev2
->desc_nr
= desc_nr
;
1310 d
= &sb
->disks
[rdev2
->desc_nr
];
1312 d
->number
= rdev2
->desc_nr
;
1313 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1314 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1316 d
->raid_disk
= rdev2
->raid_disk
;
1318 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1319 if (test_bit(Faulty
, &rdev2
->flags
))
1320 d
->state
= (1<<MD_DISK_FAULTY
);
1321 else if (is_active
) {
1322 d
->state
= (1<<MD_DISK_ACTIVE
);
1323 if (test_bit(In_sync
, &rdev2
->flags
))
1324 d
->state
|= (1<<MD_DISK_SYNC
);
1332 if (test_bit(WriteMostly
, &rdev2
->flags
))
1333 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1335 /* now set the "removed" and "faulty" bits on any missing devices */
1336 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1337 mdp_disk_t
*d
= &sb
->disks
[i
];
1338 if (d
->state
== 0 && d
->number
== 0) {
1341 d
->state
= (1<<MD_DISK_REMOVED
);
1342 d
->state
|= (1<<MD_DISK_FAULTY
);
1346 sb
->nr_disks
= nr_disks
;
1347 sb
->active_disks
= active
;
1348 sb
->working_disks
= working
;
1349 sb
->failed_disks
= failed
;
1350 sb
->spare_disks
= spare
;
1352 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1353 sb
->sb_csum
= calc_sb_csum(sb
);
1357 * rdev_size_change for 0.90.0
1359 static unsigned long long
1360 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1362 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1363 return 0; /* component must fit device */
1364 if (rdev
->mddev
->bitmap_info
.offset
)
1365 return 0; /* can't move bitmap */
1366 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1367 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1368 num_sectors
= rdev
->sb_start
;
1369 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1371 md_super_wait(rdev
->mddev
);
1377 * version 1 superblock
1380 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1384 unsigned long long newcsum
;
1385 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1386 __le32
*isuper
= (__le32
*)sb
;
1389 disk_csum
= sb
->sb_csum
;
1392 for (i
=0; size
>=4; size
-= 4 )
1393 newcsum
+= le32_to_cpu(*isuper
++);
1396 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1398 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1399 sb
->sb_csum
= disk_csum
;
1400 return cpu_to_le32(csum
);
1403 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1405 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1407 struct mdp_superblock_1
*sb
;
1410 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1414 * Calculate the position of the superblock in 512byte sectors.
1415 * It is always aligned to a 4K boundary and
1416 * depeding on minor_version, it can be:
1417 * 0: At least 8K, but less than 12K, from end of device
1418 * 1: At start of device
1419 * 2: 4K from start of device.
1421 switch(minor_version
) {
1423 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1425 sb_start
&= ~(sector_t
)(4*2-1);
1436 rdev
->sb_start
= sb_start
;
1438 /* superblock is rarely larger than 1K, but it can be larger,
1439 * and it is safe to read 4k, so we do that
1441 ret
= read_disk_sb(rdev
, 4096);
1442 if (ret
) return ret
;
1445 sb
= page_address(rdev
->sb_page
);
1447 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1448 sb
->major_version
!= cpu_to_le32(1) ||
1449 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1450 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1451 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1454 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1455 printk("md: invalid superblock checksum on %s\n",
1456 bdevname(rdev
->bdev
,b
));
1459 if (le64_to_cpu(sb
->data_size
) < 10) {
1460 printk("md: data_size too small on %s\n",
1461 bdevname(rdev
->bdev
,b
));
1465 rdev
->preferred_minor
= 0xffff;
1466 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1467 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1469 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1470 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1471 if (rdev
->sb_size
& bmask
)
1472 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1475 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1478 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1481 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1483 if (!rdev
->bb_page
) {
1484 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1488 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1489 rdev
->badblocks
.count
== 0) {
1490 /* need to load the bad block list.
1491 * Currently we limit it to one page.
1497 int sectors
= le16_to_cpu(sb
->bblog_size
);
1498 if (sectors
> (PAGE_SIZE
/ 512))
1500 offset
= le32_to_cpu(sb
->bblog_offset
);
1503 bb_sector
= (long long)offset
;
1504 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1505 rdev
->bb_page
, READ
, true))
1507 bbp
= (u64
*)page_address(rdev
->bb_page
);
1508 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1509 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1510 u64 bb
= le64_to_cpu(*bbp
);
1511 int count
= bb
& (0x3ff);
1512 u64 sector
= bb
>> 10;
1513 sector
<<= sb
->bblog_shift
;
1514 count
<<= sb
->bblog_shift
;
1517 if (md_set_badblocks(&rdev
->badblocks
,
1518 sector
, count
, 1) == 0)
1521 } else if (sb
->bblog_offset
== 0)
1522 rdev
->badblocks
.shift
= -1;
1528 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1530 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1531 sb
->level
!= refsb
->level
||
1532 sb
->layout
!= refsb
->layout
||
1533 sb
->chunksize
!= refsb
->chunksize
) {
1534 printk(KERN_WARNING
"md: %s has strangely different"
1535 " superblock to %s\n",
1536 bdevname(rdev
->bdev
,b
),
1537 bdevname(refdev
->bdev
,b2
));
1540 ev1
= le64_to_cpu(sb
->events
);
1541 ev2
= le64_to_cpu(refsb
->events
);
1549 rdev
->sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
1550 le64_to_cpu(sb
->data_offset
);
1552 rdev
->sectors
= rdev
->sb_start
;
1553 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1555 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1556 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1561 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1563 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1564 __u64 ev1
= le64_to_cpu(sb
->events
);
1566 rdev
->raid_disk
= -1;
1567 clear_bit(Faulty
, &rdev
->flags
);
1568 clear_bit(In_sync
, &rdev
->flags
);
1569 clear_bit(WriteMostly
, &rdev
->flags
);
1571 if (mddev
->raid_disks
== 0) {
1572 mddev
->major_version
= 1;
1573 mddev
->patch_version
= 0;
1574 mddev
->external
= 0;
1575 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1576 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1577 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1578 mddev
->level
= le32_to_cpu(sb
->level
);
1579 mddev
->clevel
[0] = 0;
1580 mddev
->layout
= le32_to_cpu(sb
->layout
);
1581 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1582 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1583 mddev
->events
= ev1
;
1584 mddev
->bitmap_info
.offset
= 0;
1585 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1587 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1588 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1590 mddev
->max_disks
= (4096-256)/2;
1592 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1593 mddev
->bitmap_info
.file
== NULL
)
1594 mddev
->bitmap_info
.offset
=
1595 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1597 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1598 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1599 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1600 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1601 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1602 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1604 mddev
->reshape_position
= MaxSector
;
1605 mddev
->delta_disks
= 0;
1606 mddev
->new_level
= mddev
->level
;
1607 mddev
->new_layout
= mddev
->layout
;
1608 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1611 } else if (mddev
->pers
== NULL
) {
1612 /* Insist of good event counter while assembling, except for
1613 * spares (which don't need an event count) */
1615 if (rdev
->desc_nr
>= 0 &&
1616 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1617 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1618 if (ev1
< mddev
->events
)
1620 } else if (mddev
->bitmap
) {
1621 /* If adding to array with a bitmap, then we can accept an
1622 * older device, but not too old.
1624 if (ev1
< mddev
->bitmap
->events_cleared
)
1627 if (ev1
< mddev
->events
)
1628 /* just a hot-add of a new device, leave raid_disk at -1 */
1631 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1633 if (rdev
->desc_nr
< 0 ||
1634 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1638 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1640 case 0xffff: /* spare */
1642 case 0xfffe: /* faulty */
1643 set_bit(Faulty
, &rdev
->flags
);
1646 if ((le32_to_cpu(sb
->feature_map
) &
1647 MD_FEATURE_RECOVERY_OFFSET
))
1648 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1650 set_bit(In_sync
, &rdev
->flags
);
1651 rdev
->raid_disk
= role
;
1654 if (sb
->devflags
& WriteMostly1
)
1655 set_bit(WriteMostly
, &rdev
->flags
);
1656 } else /* MULTIPATH are always insync */
1657 set_bit(In_sync
, &rdev
->flags
);
1662 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1664 struct mdp_superblock_1
*sb
;
1667 /* make rdev->sb match mddev and rdev data. */
1669 sb
= page_address(rdev
->sb_page
);
1671 sb
->feature_map
= 0;
1673 sb
->recovery_offset
= cpu_to_le64(0);
1674 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1675 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1677 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1678 sb
->events
= cpu_to_le64(mddev
->events
);
1680 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1682 sb
->resync_offset
= cpu_to_le64(0);
1684 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1686 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1687 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1688 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1689 sb
->level
= cpu_to_le32(mddev
->level
);
1690 sb
->layout
= cpu_to_le32(mddev
->layout
);
1692 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1693 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1694 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1697 if (rdev
->raid_disk
>= 0 &&
1698 !test_bit(In_sync
, &rdev
->flags
)) {
1700 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1701 sb
->recovery_offset
=
1702 cpu_to_le64(rdev
->recovery_offset
);
1705 if (mddev
->reshape_position
!= MaxSector
) {
1706 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1707 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1708 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1709 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1710 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1711 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1714 if (rdev
->badblocks
.count
== 0)
1715 /* Nothing to do for bad blocks*/ ;
1716 else if (sb
->bblog_offset
== 0)
1717 /* Cannot record bad blocks on this device */
1718 md_error(mddev
, rdev
);
1720 struct badblocks
*bb
= &rdev
->badblocks
;
1721 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1723 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1728 seq
= read_seqbegin(&bb
->lock
);
1730 memset(bbp
, 0xff, PAGE_SIZE
);
1732 for (i
= 0 ; i
< bb
->count
; i
++) {
1733 u64 internal_bb
= *p
++;
1734 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1735 | BB_LEN(internal_bb
));
1736 *bbp
++ = cpu_to_le64(store_bb
);
1738 if (read_seqretry(&bb
->lock
, seq
))
1741 bb
->sector
= (rdev
->sb_start
+
1742 (int)le32_to_cpu(sb
->bblog_offset
));
1743 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1749 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1750 if (rdev2
->desc_nr
+1 > max_dev
)
1751 max_dev
= rdev2
->desc_nr
+1;
1753 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1755 sb
->max_dev
= cpu_to_le32(max_dev
);
1756 rdev
->sb_size
= max_dev
* 2 + 256;
1757 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1758 if (rdev
->sb_size
& bmask
)
1759 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1761 max_dev
= le32_to_cpu(sb
->max_dev
);
1763 for (i
=0; i
<max_dev
;i
++)
1764 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1766 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1768 if (test_bit(Faulty
, &rdev2
->flags
))
1769 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1770 else if (test_bit(In_sync
, &rdev2
->flags
))
1771 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1772 else if (rdev2
->raid_disk
>= 0)
1773 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1775 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1778 sb
->sb_csum
= calc_sb_1_csum(sb
);
1781 static unsigned long long
1782 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1784 struct mdp_superblock_1
*sb
;
1785 sector_t max_sectors
;
1786 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1787 return 0; /* component must fit device */
1788 if (rdev
->sb_start
< rdev
->data_offset
) {
1789 /* minor versions 1 and 2; superblock before data */
1790 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1791 max_sectors
-= rdev
->data_offset
;
1792 if (!num_sectors
|| num_sectors
> max_sectors
)
1793 num_sectors
= max_sectors
;
1794 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1795 /* minor version 0 with bitmap we can't move */
1798 /* minor version 0; superblock after data */
1800 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1801 sb_start
&= ~(sector_t
)(4*2 - 1);
1802 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1803 if (!num_sectors
|| num_sectors
> max_sectors
)
1804 num_sectors
= max_sectors
;
1805 rdev
->sb_start
= sb_start
;
1807 sb
= page_address(rdev
->sb_page
);
1808 sb
->data_size
= cpu_to_le64(num_sectors
);
1809 sb
->super_offset
= rdev
->sb_start
;
1810 sb
->sb_csum
= calc_sb_1_csum(sb
);
1811 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1813 md_super_wait(rdev
->mddev
);
1817 static struct super_type super_types
[] = {
1820 .owner
= THIS_MODULE
,
1821 .load_super
= super_90_load
,
1822 .validate_super
= super_90_validate
,
1823 .sync_super
= super_90_sync
,
1824 .rdev_size_change
= super_90_rdev_size_change
,
1828 .owner
= THIS_MODULE
,
1829 .load_super
= super_1_load
,
1830 .validate_super
= super_1_validate
,
1831 .sync_super
= super_1_sync
,
1832 .rdev_size_change
= super_1_rdev_size_change
,
1836 static void sync_super(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1838 if (mddev
->sync_super
) {
1839 mddev
->sync_super(mddev
, rdev
);
1843 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1845 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1848 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1850 mdk_rdev_t
*rdev
, *rdev2
;
1853 rdev_for_each_rcu(rdev
, mddev1
)
1854 rdev_for_each_rcu(rdev2
, mddev2
)
1855 if (rdev
->bdev
->bd_contains
==
1856 rdev2
->bdev
->bd_contains
) {
1864 static LIST_HEAD(pending_raid_disks
);
1867 * Try to register data integrity profile for an mddev
1869 * This is called when an array is started and after a disk has been kicked
1870 * from the array. It only succeeds if all working and active component devices
1871 * are integrity capable with matching profiles.
1873 int md_integrity_register(mddev_t
*mddev
)
1875 mdk_rdev_t
*rdev
, *reference
= NULL
;
1877 if (list_empty(&mddev
->disks
))
1878 return 0; /* nothing to do */
1879 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1880 return 0; /* shouldn't register, or already is */
1881 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1882 /* skip spares and non-functional disks */
1883 if (test_bit(Faulty
, &rdev
->flags
))
1885 if (rdev
->raid_disk
< 0)
1888 /* Use the first rdev as the reference */
1892 /* does this rdev's profile match the reference profile? */
1893 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1894 rdev
->bdev
->bd_disk
) < 0)
1897 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1900 * All component devices are integrity capable and have matching
1901 * profiles, register the common profile for the md device.
1903 if (blk_integrity_register(mddev
->gendisk
,
1904 bdev_get_integrity(reference
->bdev
)) != 0) {
1905 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1909 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1910 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1911 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1917 EXPORT_SYMBOL(md_integrity_register
);
1919 /* Disable data integrity if non-capable/non-matching disk is being added */
1920 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1922 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1923 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1925 if (!bi_mddev
) /* nothing to do */
1927 if (rdev
->raid_disk
< 0) /* skip spares */
1929 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1930 rdev
->bdev
->bd_disk
) >= 0)
1932 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1933 blk_integrity_unregister(mddev
->gendisk
);
1935 EXPORT_SYMBOL(md_integrity_add_rdev
);
1937 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1939 char b
[BDEVNAME_SIZE
];
1949 /* prevent duplicates */
1950 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
1953 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1954 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
1955 rdev
->sectors
< mddev
->dev_sectors
)) {
1957 /* Cannot change size, so fail
1958 * If mddev->level <= 0, then we don't care
1959 * about aligning sizes (e.g. linear)
1961 if (mddev
->level
> 0)
1964 mddev
->dev_sectors
= rdev
->sectors
;
1967 /* Verify rdev->desc_nr is unique.
1968 * If it is -1, assign a free number, else
1969 * check number is not in use
1971 if (rdev
->desc_nr
< 0) {
1973 if (mddev
->pers
) choice
= mddev
->raid_disks
;
1974 while (find_rdev_nr(mddev
, choice
))
1976 rdev
->desc_nr
= choice
;
1978 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
1981 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
1982 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
1983 mdname(mddev
), mddev
->max_disks
);
1986 bdevname(rdev
->bdev
,b
);
1987 while ( (s
=strchr(b
, '/')) != NULL
)
1990 rdev
->mddev
= mddev
;
1991 printk(KERN_INFO
"md: bind<%s>\n", b
);
1993 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
1996 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
1997 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
1998 /* failure here is OK */;
1999 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2001 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2002 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2004 /* May as well allow recovery to be retried once */
2005 mddev
->recovery_disabled
++;
2010 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2015 static void md_delayed_delete(struct work_struct
*ws
)
2017 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
2018 kobject_del(&rdev
->kobj
);
2019 kobject_put(&rdev
->kobj
);
2022 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
2024 char b
[BDEVNAME_SIZE
];
2029 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2030 list_del_rcu(&rdev
->same_set
);
2031 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2033 sysfs_remove_link(&rdev
->kobj
, "block");
2034 sysfs_put(rdev
->sysfs_state
);
2035 rdev
->sysfs_state
= NULL
;
2036 kfree(rdev
->badblocks
.page
);
2037 rdev
->badblocks
.count
= 0;
2038 rdev
->badblocks
.page
= NULL
;
2039 /* We need to delay this, otherwise we can deadlock when
2040 * writing to 'remove' to "dev/state". We also need
2041 * to delay it due to rcu usage.
2044 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2045 kobject_get(&rdev
->kobj
);
2046 queue_work(md_misc_wq
, &rdev
->del_work
);
2050 * prevent the device from being mounted, repartitioned or
2051 * otherwise reused by a RAID array (or any other kernel
2052 * subsystem), by bd_claiming the device.
2054 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
2057 struct block_device
*bdev
;
2058 char b
[BDEVNAME_SIZE
];
2060 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2061 shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
2063 printk(KERN_ERR
"md: could not open %s.\n",
2064 __bdevname(dev
, b
));
2065 return PTR_ERR(bdev
);
2071 static void unlock_rdev(mdk_rdev_t
*rdev
)
2073 struct block_device
*bdev
= rdev
->bdev
;
2077 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2080 void md_autodetect_dev(dev_t dev
);
2082 static void export_rdev(mdk_rdev_t
* rdev
)
2084 char b
[BDEVNAME_SIZE
];
2085 printk(KERN_INFO
"md: export_rdev(%s)\n",
2086 bdevname(rdev
->bdev
,b
));
2091 if (test_bit(AutoDetected
, &rdev
->flags
))
2092 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2095 kobject_put(&rdev
->kobj
);
2098 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
2100 unbind_rdev_from_array(rdev
);
2104 static void export_array(mddev_t
*mddev
)
2106 mdk_rdev_t
*rdev
, *tmp
;
2108 rdev_for_each(rdev
, tmp
, mddev
) {
2113 kick_rdev_from_array(rdev
);
2115 if (!list_empty(&mddev
->disks
))
2117 mddev
->raid_disks
= 0;
2118 mddev
->major_version
= 0;
2121 static void print_desc(mdp_disk_t
*desc
)
2123 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2124 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2127 static void print_sb_90(mdp_super_t
*sb
)
2132 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2133 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2134 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2136 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2137 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2138 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2139 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2140 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2141 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2142 sb
->failed_disks
, sb
->spare_disks
,
2143 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2146 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2149 desc
= sb
->disks
+ i
;
2150 if (desc
->number
|| desc
->major
|| desc
->minor
||
2151 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2152 printk(" D %2d: ", i
);
2156 printk(KERN_INFO
"md: THIS: ");
2157 print_desc(&sb
->this_disk
);
2160 static void print_sb_1(struct mdp_superblock_1
*sb
)
2164 uuid
= sb
->set_uuid
;
2166 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2167 "md: Name: \"%s\" CT:%llu\n",
2168 le32_to_cpu(sb
->major_version
),
2169 le32_to_cpu(sb
->feature_map
),
2172 (unsigned long long)le64_to_cpu(sb
->ctime
)
2173 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2175 uuid
= sb
->device_uuid
;
2177 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2179 "md: Dev:%08x UUID: %pU\n"
2180 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2181 "md: (MaxDev:%u) \n",
2182 le32_to_cpu(sb
->level
),
2183 (unsigned long long)le64_to_cpu(sb
->size
),
2184 le32_to_cpu(sb
->raid_disks
),
2185 le32_to_cpu(sb
->layout
),
2186 le32_to_cpu(sb
->chunksize
),
2187 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2188 (unsigned long long)le64_to_cpu(sb
->data_size
),
2189 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2190 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2191 le32_to_cpu(sb
->dev_number
),
2194 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2195 (unsigned long long)le64_to_cpu(sb
->events
),
2196 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2197 le32_to_cpu(sb
->sb_csum
),
2198 le32_to_cpu(sb
->max_dev
)
2202 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2204 char b
[BDEVNAME_SIZE
];
2205 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2206 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2207 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2209 if (rdev
->sb_loaded
) {
2210 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2211 switch (major_version
) {
2213 print_sb_90(page_address(rdev
->sb_page
));
2216 print_sb_1(page_address(rdev
->sb_page
));
2220 printk(KERN_INFO
"md: no rdev superblock!\n");
2223 static void md_print_devices(void)
2225 struct list_head
*tmp
;
2228 char b
[BDEVNAME_SIZE
];
2231 printk("md: **********************************\n");
2232 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2233 printk("md: **********************************\n");
2234 for_each_mddev(mddev
, tmp
) {
2237 bitmap_print_sb(mddev
->bitmap
);
2239 printk("%s: ", mdname(mddev
));
2240 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2241 printk("<%s>", bdevname(rdev
->bdev
,b
));
2244 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2245 print_rdev(rdev
, mddev
->major_version
);
2247 printk("md: **********************************\n");
2252 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2254 /* Update each superblock (in-memory image), but
2255 * if we are allowed to, skip spares which already
2256 * have the right event counter, or have one earlier
2257 * (which would mean they aren't being marked as dirty
2258 * with the rest of the array)
2261 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2262 if (rdev
->sb_events
== mddev
->events
||
2264 rdev
->raid_disk
< 0 &&
2265 rdev
->sb_events
+1 == mddev
->events
)) {
2266 /* Don't update this superblock */
2267 rdev
->sb_loaded
= 2;
2269 sync_super(mddev
, rdev
);
2270 rdev
->sb_loaded
= 1;
2275 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2280 int any_badblocks_changed
= 0;
2283 /* First make sure individual recovery_offsets are correct */
2284 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2285 if (rdev
->raid_disk
>= 0 &&
2286 mddev
->delta_disks
>= 0 &&
2287 !test_bit(In_sync
, &rdev
->flags
) &&
2288 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2289 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2292 if (!mddev
->persistent
) {
2293 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2294 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2295 if (!mddev
->external
)
2296 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2297 wake_up(&mddev
->sb_wait
);
2301 spin_lock_irq(&mddev
->write_lock
);
2303 mddev
->utime
= get_seconds();
2305 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2307 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2308 /* just a clean<-> dirty transition, possibly leave spares alone,
2309 * though if events isn't the right even/odd, we will have to do
2315 if (mddev
->degraded
)
2316 /* If the array is degraded, then skipping spares is both
2317 * dangerous and fairly pointless.
2318 * Dangerous because a device that was removed from the array
2319 * might have a event_count that still looks up-to-date,
2320 * so it can be re-added without a resync.
2321 * Pointless because if there are any spares to skip,
2322 * then a recovery will happen and soon that array won't
2323 * be degraded any more and the spare can go back to sleep then.
2327 sync_req
= mddev
->in_sync
;
2329 /* If this is just a dirty<->clean transition, and the array is clean
2330 * and 'events' is odd, we can roll back to the previous clean state */
2332 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2333 && mddev
->can_decrease_events
2334 && mddev
->events
!= 1) {
2336 mddev
->can_decrease_events
= 0;
2338 /* otherwise we have to go forward and ... */
2340 mddev
->can_decrease_events
= nospares
;
2343 if (!mddev
->events
) {
2345 * oops, this 64-bit counter should never wrap.
2346 * Either we are in around ~1 trillion A.C., assuming
2347 * 1 reboot per second, or we have a bug:
2353 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2354 if (rdev
->badblocks
.changed
)
2355 any_badblocks_changed
++;
2357 sync_sbs(mddev
, nospares
);
2358 spin_unlock_irq(&mddev
->write_lock
);
2361 "md: updating %s RAID superblock on device (in sync %d)\n",
2362 mdname(mddev
),mddev
->in_sync
);
2364 bitmap_update_sb(mddev
->bitmap
);
2365 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2366 char b
[BDEVNAME_SIZE
];
2367 dprintk(KERN_INFO
"md: ");
2368 if (rdev
->sb_loaded
!= 1)
2369 continue; /* no noise on spare devices */
2370 if (test_bit(Faulty
, &rdev
->flags
))
2371 dprintk("(skipping faulty ");
2373 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2374 if (!test_bit(Faulty
, &rdev
->flags
)) {
2375 md_super_write(mddev
,rdev
,
2376 rdev
->sb_start
, rdev
->sb_size
,
2378 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2379 bdevname(rdev
->bdev
,b
),
2380 (unsigned long long)rdev
->sb_start
);
2381 rdev
->sb_events
= mddev
->events
;
2382 if (rdev
->badblocks
.size
) {
2383 md_super_write(mddev
, rdev
,
2384 rdev
->badblocks
.sector
,
2385 rdev
->badblocks
.size
<< 9,
2387 rdev
->badblocks
.size
= 0;
2392 if (mddev
->level
== LEVEL_MULTIPATH
)
2393 /* only need to write one superblock... */
2396 md_super_wait(mddev
);
2397 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2399 spin_lock_irq(&mddev
->write_lock
);
2400 if (mddev
->in_sync
!= sync_req
||
2401 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2402 /* have to write it out again */
2403 spin_unlock_irq(&mddev
->write_lock
);
2406 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2407 spin_unlock_irq(&mddev
->write_lock
);
2408 wake_up(&mddev
->sb_wait
);
2409 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2410 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2412 if (any_badblocks_changed
)
2413 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2414 md_ack_all_badblocks(&rdev
->badblocks
);
2417 /* words written to sysfs files may, or may not, be \n terminated.
2418 * We want to accept with case. For this we use cmd_match.
2420 static int cmd_match(const char *cmd
, const char *str
)
2422 /* See if cmd, written into a sysfs file, matches
2423 * str. They must either be the same, or cmd can
2424 * have a trailing newline
2426 while (*cmd
&& *str
&& *cmd
== *str
) {
2437 struct rdev_sysfs_entry
{
2438 struct attribute attr
;
2439 ssize_t (*show
)(mdk_rdev_t
*, char *);
2440 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2444 state_show(mdk_rdev_t
*rdev
, char *page
)
2449 if (test_bit(Faulty
, &rdev
->flags
)) {
2450 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2453 if (test_bit(In_sync
, &rdev
->flags
)) {
2454 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2457 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2458 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2461 if (test_bit(Blocked
, &rdev
->flags
)) {
2462 len
+= sprintf(page
+len
, "%sblocked", sep
);
2465 if (!test_bit(Faulty
, &rdev
->flags
) &&
2466 !test_bit(In_sync
, &rdev
->flags
)) {
2467 len
+= sprintf(page
+len
, "%sspare", sep
);
2470 return len
+sprintf(page
+len
, "\n");
2474 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2477 * faulty - simulates and error
2478 * remove - disconnects the device
2479 * writemostly - sets write_mostly
2480 * -writemostly - clears write_mostly
2481 * blocked - sets the Blocked flag
2482 * -blocked - clears the Blocked flag
2483 * insync - sets Insync providing device isn't active
2486 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2487 md_error(rdev
->mddev
, rdev
);
2489 } else if (cmd_match(buf
, "remove")) {
2490 if (rdev
->raid_disk
>= 0)
2493 mddev_t
*mddev
= rdev
->mddev
;
2494 kick_rdev_from_array(rdev
);
2496 md_update_sb(mddev
, 1);
2497 md_new_event(mddev
);
2500 } else if (cmd_match(buf
, "writemostly")) {
2501 set_bit(WriteMostly
, &rdev
->flags
);
2503 } else if (cmd_match(buf
, "-writemostly")) {
2504 clear_bit(WriteMostly
, &rdev
->flags
);
2506 } else if (cmd_match(buf
, "blocked")) {
2507 set_bit(Blocked
, &rdev
->flags
);
2509 } else if (cmd_match(buf
, "-blocked")) {
2510 clear_bit(Blocked
, &rdev
->flags
);
2511 wake_up(&rdev
->blocked_wait
);
2512 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2513 md_wakeup_thread(rdev
->mddev
->thread
);
2516 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2517 set_bit(In_sync
, &rdev
->flags
);
2521 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2522 return err
? err
: len
;
2524 static struct rdev_sysfs_entry rdev_state
=
2525 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2528 errors_show(mdk_rdev_t
*rdev
, char *page
)
2530 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2534 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2537 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2538 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2539 atomic_set(&rdev
->corrected_errors
, n
);
2544 static struct rdev_sysfs_entry rdev_errors
=
2545 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2548 slot_show(mdk_rdev_t
*rdev
, char *page
)
2550 if (rdev
->raid_disk
< 0)
2551 return sprintf(page
, "none\n");
2553 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2557 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2561 int slot
= simple_strtoul(buf
, &e
, 10);
2562 if (strncmp(buf
, "none", 4)==0)
2564 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2566 if (rdev
->mddev
->pers
&& slot
== -1) {
2567 /* Setting 'slot' on an active array requires also
2568 * updating the 'rd%d' link, and communicating
2569 * with the personality with ->hot_*_disk.
2570 * For now we only support removing
2571 * failed/spare devices. This normally happens automatically,
2572 * but not when the metadata is externally managed.
2574 if (rdev
->raid_disk
== -1)
2576 /* personality does all needed checks */
2577 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2579 err
= rdev
->mddev
->pers
->
2580 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2583 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2584 rdev
->raid_disk
= -1;
2585 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2586 md_wakeup_thread(rdev
->mddev
->thread
);
2587 } else if (rdev
->mddev
->pers
) {
2589 /* Activating a spare .. or possibly reactivating
2590 * if we ever get bitmaps working here.
2593 if (rdev
->raid_disk
!= -1)
2596 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2599 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2602 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2603 if (rdev2
->raid_disk
== slot
)
2606 if (slot
>= rdev
->mddev
->raid_disks
&&
2607 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2610 rdev
->raid_disk
= slot
;
2611 if (test_bit(In_sync
, &rdev
->flags
))
2612 rdev
->saved_raid_disk
= slot
;
2614 rdev
->saved_raid_disk
= -1;
2615 err
= rdev
->mddev
->pers
->
2616 hot_add_disk(rdev
->mddev
, rdev
);
2618 rdev
->raid_disk
= -1;
2621 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2622 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2623 /* failure here is OK */;
2624 /* don't wakeup anyone, leave that to userspace. */
2626 if (slot
>= rdev
->mddev
->raid_disks
&&
2627 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2629 rdev
->raid_disk
= slot
;
2630 /* assume it is working */
2631 clear_bit(Faulty
, &rdev
->flags
);
2632 clear_bit(WriteMostly
, &rdev
->flags
);
2633 set_bit(In_sync
, &rdev
->flags
);
2634 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2640 static struct rdev_sysfs_entry rdev_slot
=
2641 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2644 offset_show(mdk_rdev_t
*rdev
, char *page
)
2646 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2650 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2653 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2654 if (e
==buf
|| (*e
&& *e
!= '\n'))
2656 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2658 if (rdev
->sectors
&& rdev
->mddev
->external
)
2659 /* Must set offset before size, so overlap checks
2662 rdev
->data_offset
= offset
;
2666 static struct rdev_sysfs_entry rdev_offset
=
2667 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2670 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2672 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2675 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2677 /* check if two start/length pairs overlap */
2685 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2687 unsigned long long blocks
;
2690 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2693 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2694 return -EINVAL
; /* sector conversion overflow */
2697 if (new != blocks
* 2)
2698 return -EINVAL
; /* unsigned long long to sector_t overflow */
2705 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2707 mddev_t
*my_mddev
= rdev
->mddev
;
2708 sector_t oldsectors
= rdev
->sectors
;
2711 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2713 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2714 if (my_mddev
->persistent
) {
2715 sectors
= super_types
[my_mddev
->major_version
].
2716 rdev_size_change(rdev
, sectors
);
2719 } else if (!sectors
)
2720 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2723 if (sectors
< my_mddev
->dev_sectors
)
2724 return -EINVAL
; /* component must fit device */
2726 rdev
->sectors
= sectors
;
2727 if (sectors
> oldsectors
&& my_mddev
->external
) {
2728 /* need to check that all other rdevs with the same ->bdev
2729 * do not overlap. We need to unlock the mddev to avoid
2730 * a deadlock. We have already changed rdev->sectors, and if
2731 * we have to change it back, we will have the lock again.
2735 struct list_head
*tmp
;
2737 mddev_unlock(my_mddev
);
2738 for_each_mddev(mddev
, tmp
) {
2742 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2743 if (rdev
->bdev
== rdev2
->bdev
&&
2745 overlaps(rdev
->data_offset
, rdev
->sectors
,
2751 mddev_unlock(mddev
);
2757 mddev_lock(my_mddev
);
2759 /* Someone else could have slipped in a size
2760 * change here, but doing so is just silly.
2761 * We put oldsectors back because we *know* it is
2762 * safe, and trust userspace not to race with
2765 rdev
->sectors
= oldsectors
;
2772 static struct rdev_sysfs_entry rdev_size
=
2773 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2776 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2778 unsigned long long recovery_start
= rdev
->recovery_offset
;
2780 if (test_bit(In_sync
, &rdev
->flags
) ||
2781 recovery_start
== MaxSector
)
2782 return sprintf(page
, "none\n");
2784 return sprintf(page
, "%llu\n", recovery_start
);
2787 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2789 unsigned long long recovery_start
;
2791 if (cmd_match(buf
, "none"))
2792 recovery_start
= MaxSector
;
2793 else if (strict_strtoull(buf
, 10, &recovery_start
))
2796 if (rdev
->mddev
->pers
&&
2797 rdev
->raid_disk
>= 0)
2800 rdev
->recovery_offset
= recovery_start
;
2801 if (recovery_start
== MaxSector
)
2802 set_bit(In_sync
, &rdev
->flags
);
2804 clear_bit(In_sync
, &rdev
->flags
);
2808 static struct rdev_sysfs_entry rdev_recovery_start
=
2809 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2813 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2815 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2817 static ssize_t
bb_show(mdk_rdev_t
*rdev
, char *page
)
2819 return badblocks_show(&rdev
->badblocks
, page
, 0);
2821 static ssize_t
bb_store(mdk_rdev_t
*rdev
, const char *page
, size_t len
)
2823 return badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2825 static struct rdev_sysfs_entry rdev_bad_blocks
=
2826 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2829 static ssize_t
ubb_show(mdk_rdev_t
*rdev
, char *page
)
2831 return badblocks_show(&rdev
->badblocks
, page
, 1);
2833 static ssize_t
ubb_store(mdk_rdev_t
*rdev
, const char *page
, size_t len
)
2835 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2837 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2838 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2840 static struct attribute
*rdev_default_attrs
[] = {
2846 &rdev_recovery_start
.attr
,
2847 &rdev_bad_blocks
.attr
,
2848 &rdev_unack_bad_blocks
.attr
,
2852 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2854 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2855 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2856 mddev_t
*mddev
= rdev
->mddev
;
2862 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2864 if (rdev
->mddev
== NULL
)
2867 rv
= entry
->show(rdev
, page
);
2868 mddev_unlock(mddev
);
2874 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2875 const char *page
, size_t length
)
2877 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2878 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2880 mddev_t
*mddev
= rdev
->mddev
;
2884 if (!capable(CAP_SYS_ADMIN
))
2886 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2888 if (rdev
->mddev
== NULL
)
2891 rv
= entry
->store(rdev
, page
, length
);
2892 mddev_unlock(mddev
);
2897 static void rdev_free(struct kobject
*ko
)
2899 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
2902 static const struct sysfs_ops rdev_sysfs_ops
= {
2903 .show
= rdev_attr_show
,
2904 .store
= rdev_attr_store
,
2906 static struct kobj_type rdev_ktype
= {
2907 .release
= rdev_free
,
2908 .sysfs_ops
= &rdev_sysfs_ops
,
2909 .default_attrs
= rdev_default_attrs
,
2912 int md_rdev_init(mdk_rdev_t
*rdev
)
2915 rdev
->saved_raid_disk
= -1;
2916 rdev
->raid_disk
= -1;
2918 rdev
->data_offset
= 0;
2919 rdev
->sb_events
= 0;
2920 rdev
->last_read_error
.tv_sec
= 0;
2921 rdev
->last_read_error
.tv_nsec
= 0;
2922 rdev
->sb_loaded
= 0;
2923 rdev
->bb_page
= NULL
;
2924 atomic_set(&rdev
->nr_pending
, 0);
2925 atomic_set(&rdev
->read_errors
, 0);
2926 atomic_set(&rdev
->corrected_errors
, 0);
2928 INIT_LIST_HEAD(&rdev
->same_set
);
2929 init_waitqueue_head(&rdev
->blocked_wait
);
2931 /* Add space to store bad block list.
2932 * This reserves the space even on arrays where it cannot
2933 * be used - I wonder if that matters
2935 rdev
->badblocks
.count
= 0;
2936 rdev
->badblocks
.shift
= 0;
2937 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2938 seqlock_init(&rdev
->badblocks
.lock
);
2939 if (rdev
->badblocks
.page
== NULL
)
2944 EXPORT_SYMBOL_GPL(md_rdev_init
);
2946 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2948 * mark the device faulty if:
2950 * - the device is nonexistent (zero size)
2951 * - the device has no valid superblock
2953 * a faulty rdev _never_ has rdev->sb set.
2955 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
2957 char b
[BDEVNAME_SIZE
];
2962 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
2964 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
2965 return ERR_PTR(-ENOMEM
);
2968 err
= md_rdev_init(rdev
);
2971 err
= alloc_disk_sb(rdev
);
2975 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
2979 kobject_init(&rdev
->kobj
, &rdev_ktype
);
2981 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
2984 "md: %s has zero or unknown size, marking faulty!\n",
2985 bdevname(rdev
->bdev
,b
));
2990 if (super_format
>= 0) {
2991 err
= super_types
[super_format
].
2992 load_super(rdev
, NULL
, super_minor
);
2993 if (err
== -EINVAL
) {
2995 "md: %s does not have a valid v%d.%d "
2996 "superblock, not importing!\n",
2997 bdevname(rdev
->bdev
,b
),
2998 super_format
, super_minor
);
3003 "md: could not read %s's sb, not importing!\n",
3004 bdevname(rdev
->bdev
,b
));
3008 if (super_format
== -1)
3009 /* hot-add for 0.90, or non-persistent: so no badblocks */
3010 rdev
->badblocks
.shift
= -1;
3018 kfree(rdev
->badblocks
.page
);
3020 return ERR_PTR(err
);
3024 * Check a full RAID array for plausibility
3028 static void analyze_sbs(mddev_t
* mddev
)
3031 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
3032 char b
[BDEVNAME_SIZE
];
3035 rdev_for_each(rdev
, tmp
, mddev
)
3036 switch (super_types
[mddev
->major_version
].
3037 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3045 "md: fatal superblock inconsistency in %s"
3046 " -- removing from array\n",
3047 bdevname(rdev
->bdev
,b
));
3048 kick_rdev_from_array(rdev
);
3052 super_types
[mddev
->major_version
].
3053 validate_super(mddev
, freshest
);
3056 rdev_for_each(rdev
, tmp
, mddev
) {
3057 if (mddev
->max_disks
&&
3058 (rdev
->desc_nr
>= mddev
->max_disks
||
3059 i
> mddev
->max_disks
)) {
3061 "md: %s: %s: only %d devices permitted\n",
3062 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3064 kick_rdev_from_array(rdev
);
3067 if (rdev
!= freshest
)
3068 if (super_types
[mddev
->major_version
].
3069 validate_super(mddev
, rdev
)) {
3070 printk(KERN_WARNING
"md: kicking non-fresh %s"
3072 bdevname(rdev
->bdev
,b
));
3073 kick_rdev_from_array(rdev
);
3076 if (mddev
->level
== LEVEL_MULTIPATH
) {
3077 rdev
->desc_nr
= i
++;
3078 rdev
->raid_disk
= rdev
->desc_nr
;
3079 set_bit(In_sync
, &rdev
->flags
);
3080 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3081 rdev
->raid_disk
= -1;
3082 clear_bit(In_sync
, &rdev
->flags
);
3087 /* Read a fixed-point number.
3088 * Numbers in sysfs attributes should be in "standard" units where
3089 * possible, so time should be in seconds.
3090 * However we internally use a a much smaller unit such as
3091 * milliseconds or jiffies.
3092 * This function takes a decimal number with a possible fractional
3093 * component, and produces an integer which is the result of
3094 * multiplying that number by 10^'scale'.
3095 * all without any floating-point arithmetic.
3097 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3099 unsigned long result
= 0;
3101 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3104 else if (decimals
< scale
) {
3107 result
= result
* 10 + value
;
3119 while (decimals
< scale
) {
3128 static void md_safemode_timeout(unsigned long data
);
3131 safe_delay_show(mddev_t
*mddev
, char *page
)
3133 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3134 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3137 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
3141 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3144 mddev
->safemode_delay
= 0;
3146 unsigned long old_delay
= mddev
->safemode_delay
;
3147 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3148 if (mddev
->safemode_delay
== 0)
3149 mddev
->safemode_delay
= 1;
3150 if (mddev
->safemode_delay
< old_delay
)
3151 md_safemode_timeout((unsigned long)mddev
);
3155 static struct md_sysfs_entry md_safe_delay
=
3156 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3159 level_show(mddev_t
*mddev
, char *page
)
3161 struct mdk_personality
*p
= mddev
->pers
;
3163 return sprintf(page
, "%s\n", p
->name
);
3164 else if (mddev
->clevel
[0])
3165 return sprintf(page
, "%s\n", mddev
->clevel
);
3166 else if (mddev
->level
!= LEVEL_NONE
)
3167 return sprintf(page
, "%d\n", mddev
->level
);
3173 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3177 struct mdk_personality
*pers
;
3182 if (mddev
->pers
== NULL
) {
3185 if (len
>= sizeof(mddev
->clevel
))
3187 strncpy(mddev
->clevel
, buf
, len
);
3188 if (mddev
->clevel
[len
-1] == '\n')
3190 mddev
->clevel
[len
] = 0;
3191 mddev
->level
= LEVEL_NONE
;
3195 /* request to change the personality. Need to ensure:
3196 * - array is not engaged in resync/recovery/reshape
3197 * - old personality can be suspended
3198 * - new personality will access other array.
3201 if (mddev
->sync_thread
||
3202 mddev
->reshape_position
!= MaxSector
||
3203 mddev
->sysfs_active
)
3206 if (!mddev
->pers
->quiesce
) {
3207 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3208 mdname(mddev
), mddev
->pers
->name
);
3212 /* Now find the new personality */
3213 if (len
== 0 || len
>= sizeof(clevel
))
3215 strncpy(clevel
, buf
, len
);
3216 if (clevel
[len
-1] == '\n')
3219 if (strict_strtol(clevel
, 10, &level
))
3222 if (request_module("md-%s", clevel
) != 0)
3223 request_module("md-level-%s", clevel
);
3224 spin_lock(&pers_lock
);
3225 pers
= find_pers(level
, clevel
);
3226 if (!pers
|| !try_module_get(pers
->owner
)) {
3227 spin_unlock(&pers_lock
);
3228 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3231 spin_unlock(&pers_lock
);
3233 if (pers
== mddev
->pers
) {
3234 /* Nothing to do! */
3235 module_put(pers
->owner
);
3238 if (!pers
->takeover
) {
3239 module_put(pers
->owner
);
3240 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3241 mdname(mddev
), clevel
);
3245 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3246 rdev
->new_raid_disk
= rdev
->raid_disk
;
3248 /* ->takeover must set new_* and/or delta_disks
3249 * if it succeeds, and may set them when it fails.
3251 priv
= pers
->takeover(mddev
);
3253 mddev
->new_level
= mddev
->level
;
3254 mddev
->new_layout
= mddev
->layout
;
3255 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3256 mddev
->raid_disks
-= mddev
->delta_disks
;
3257 mddev
->delta_disks
= 0;
3258 module_put(pers
->owner
);
3259 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3260 mdname(mddev
), clevel
);
3261 return PTR_ERR(priv
);
3264 /* Looks like we have a winner */
3265 mddev_suspend(mddev
);
3266 mddev
->pers
->stop(mddev
);
3268 if (mddev
->pers
->sync_request
== NULL
&&
3269 pers
->sync_request
!= NULL
) {
3270 /* need to add the md_redundancy_group */
3271 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3273 "md: cannot register extra attributes for %s\n",
3275 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3277 if (mddev
->pers
->sync_request
!= NULL
&&
3278 pers
->sync_request
== NULL
) {
3279 /* need to remove the md_redundancy_group */
3280 if (mddev
->to_remove
== NULL
)
3281 mddev
->to_remove
= &md_redundancy_group
;
3284 if (mddev
->pers
->sync_request
== NULL
&&
3286 /* We are converting from a no-redundancy array
3287 * to a redundancy array and metadata is managed
3288 * externally so we need to be sure that writes
3289 * won't block due to a need to transition
3291 * until external management is started.
3294 mddev
->safemode_delay
= 0;
3295 mddev
->safemode
= 0;
3298 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3299 if (rdev
->raid_disk
< 0)
3301 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3302 rdev
->new_raid_disk
= -1;
3303 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3305 sysfs_unlink_rdev(mddev
, rdev
);
3307 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3308 if (rdev
->raid_disk
< 0)
3310 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3312 rdev
->raid_disk
= rdev
->new_raid_disk
;
3313 if (rdev
->raid_disk
< 0)
3314 clear_bit(In_sync
, &rdev
->flags
);
3316 if (sysfs_link_rdev(mddev
, rdev
))
3317 printk(KERN_WARNING
"md: cannot register rd%d"
3318 " for %s after level change\n",
3319 rdev
->raid_disk
, mdname(mddev
));
3323 module_put(mddev
->pers
->owner
);
3325 mddev
->private = priv
;
3326 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3327 mddev
->level
= mddev
->new_level
;
3328 mddev
->layout
= mddev
->new_layout
;
3329 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3330 mddev
->delta_disks
= 0;
3331 mddev
->degraded
= 0;
3332 if (mddev
->pers
->sync_request
== NULL
) {
3333 /* this is now an array without redundancy, so
3334 * it must always be in_sync
3337 del_timer_sync(&mddev
->safemode_timer
);
3340 mddev_resume(mddev
);
3341 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3342 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3343 md_wakeup_thread(mddev
->thread
);
3344 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3345 md_new_event(mddev
);
3349 static struct md_sysfs_entry md_level
=
3350 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3354 layout_show(mddev_t
*mddev
, char *page
)
3356 /* just a number, not meaningful for all levels */
3357 if (mddev
->reshape_position
!= MaxSector
&&
3358 mddev
->layout
!= mddev
->new_layout
)
3359 return sprintf(page
, "%d (%d)\n",
3360 mddev
->new_layout
, mddev
->layout
);
3361 return sprintf(page
, "%d\n", mddev
->layout
);
3365 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3368 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3370 if (!*buf
|| (*e
&& *e
!= '\n'))
3375 if (mddev
->pers
->check_reshape
== NULL
)
3377 mddev
->new_layout
= n
;
3378 err
= mddev
->pers
->check_reshape(mddev
);
3380 mddev
->new_layout
= mddev
->layout
;
3384 mddev
->new_layout
= n
;
3385 if (mddev
->reshape_position
== MaxSector
)
3390 static struct md_sysfs_entry md_layout
=
3391 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3395 raid_disks_show(mddev_t
*mddev
, char *page
)
3397 if (mddev
->raid_disks
== 0)
3399 if (mddev
->reshape_position
!= MaxSector
&&
3400 mddev
->delta_disks
!= 0)
3401 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3402 mddev
->raid_disks
- mddev
->delta_disks
);
3403 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3406 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3409 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3413 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3415 if (!*buf
|| (*e
&& *e
!= '\n'))
3419 rv
= update_raid_disks(mddev
, n
);
3420 else if (mddev
->reshape_position
!= MaxSector
) {
3421 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3422 mddev
->delta_disks
= n
- olddisks
;
3423 mddev
->raid_disks
= n
;
3425 mddev
->raid_disks
= n
;
3426 return rv
? rv
: len
;
3428 static struct md_sysfs_entry md_raid_disks
=
3429 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3432 chunk_size_show(mddev_t
*mddev
, char *page
)
3434 if (mddev
->reshape_position
!= MaxSector
&&
3435 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3436 return sprintf(page
, "%d (%d)\n",
3437 mddev
->new_chunk_sectors
<< 9,
3438 mddev
->chunk_sectors
<< 9);
3439 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3443 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3446 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3448 if (!*buf
|| (*e
&& *e
!= '\n'))
3453 if (mddev
->pers
->check_reshape
== NULL
)
3455 mddev
->new_chunk_sectors
= n
>> 9;
3456 err
= mddev
->pers
->check_reshape(mddev
);
3458 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3462 mddev
->new_chunk_sectors
= n
>> 9;
3463 if (mddev
->reshape_position
== MaxSector
)
3464 mddev
->chunk_sectors
= n
>> 9;
3468 static struct md_sysfs_entry md_chunk_size
=
3469 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3472 resync_start_show(mddev_t
*mddev
, char *page
)
3474 if (mddev
->recovery_cp
== MaxSector
)
3475 return sprintf(page
, "none\n");
3476 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3480 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3483 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3485 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3487 if (cmd_match(buf
, "none"))
3489 else if (!*buf
|| (*e
&& *e
!= '\n'))
3492 mddev
->recovery_cp
= n
;
3495 static struct md_sysfs_entry md_resync_start
=
3496 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3499 * The array state can be:
3502 * No devices, no size, no level
3503 * Equivalent to STOP_ARRAY ioctl
3505 * May have some settings, but array is not active
3506 * all IO results in error
3507 * When written, doesn't tear down array, but just stops it
3508 * suspended (not supported yet)
3509 * All IO requests will block. The array can be reconfigured.
3510 * Writing this, if accepted, will block until array is quiescent
3512 * no resync can happen. no superblocks get written.
3513 * write requests fail
3515 * like readonly, but behaves like 'clean' on a write request.
3517 * clean - no pending writes, but otherwise active.
3518 * When written to inactive array, starts without resync
3519 * If a write request arrives then
3520 * if metadata is known, mark 'dirty' and switch to 'active'.
3521 * if not known, block and switch to write-pending
3522 * If written to an active array that has pending writes, then fails.
3524 * fully active: IO and resync can be happening.
3525 * When written to inactive array, starts with resync
3528 * clean, but writes are blocked waiting for 'active' to be written.
3531 * like active, but no writes have been seen for a while (100msec).
3534 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3535 write_pending
, active_idle
, bad_word
};
3536 static char *array_states
[] = {
3537 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3538 "write-pending", "active-idle", NULL
};
3540 static int match_word(const char *word
, char **list
)
3543 for (n
=0; list
[n
]; n
++)
3544 if (cmd_match(word
, list
[n
]))
3550 array_state_show(mddev_t
*mddev
, char *page
)
3552 enum array_state st
= inactive
;
3565 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3567 else if (mddev
->safemode
)
3573 if (list_empty(&mddev
->disks
) &&
3574 mddev
->raid_disks
== 0 &&
3575 mddev
->dev_sectors
== 0)
3580 return sprintf(page
, "%s\n", array_states
[st
]);
3583 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3584 static int md_set_readonly(mddev_t
* mddev
, int is_open
);
3585 static int do_md_run(mddev_t
* mddev
);
3586 static int restart_array(mddev_t
*mddev
);
3589 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3592 enum array_state st
= match_word(buf
, array_states
);
3597 /* stopping an active array */
3598 if (atomic_read(&mddev
->openers
) > 0)
3600 err
= do_md_stop(mddev
, 0, 0);
3603 /* stopping an active array */
3605 if (atomic_read(&mddev
->openers
) > 0)
3607 err
= do_md_stop(mddev
, 2, 0);
3609 err
= 0; /* already inactive */
3612 break; /* not supported yet */
3615 err
= md_set_readonly(mddev
, 0);
3618 set_disk_ro(mddev
->gendisk
, 1);
3619 err
= do_md_run(mddev
);
3625 err
= md_set_readonly(mddev
, 0);
3626 else if (mddev
->ro
== 1)
3627 err
= restart_array(mddev
);
3630 set_disk_ro(mddev
->gendisk
, 0);
3634 err
= do_md_run(mddev
);
3639 restart_array(mddev
);
3640 spin_lock_irq(&mddev
->write_lock
);
3641 if (atomic_read(&mddev
->writes_pending
) == 0) {
3642 if (mddev
->in_sync
== 0) {
3644 if (mddev
->safemode
== 1)
3645 mddev
->safemode
= 0;
3646 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3651 spin_unlock_irq(&mddev
->write_lock
);
3657 restart_array(mddev
);
3658 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3659 wake_up(&mddev
->sb_wait
);
3663 set_disk_ro(mddev
->gendisk
, 0);
3664 err
= do_md_run(mddev
);
3669 /* these cannot be set */
3675 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3679 static struct md_sysfs_entry md_array_state
=
3680 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3683 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3684 return sprintf(page
, "%d\n",
3685 atomic_read(&mddev
->max_corr_read_errors
));
3689 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3692 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3694 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3695 atomic_set(&mddev
->max_corr_read_errors
, n
);
3701 static struct md_sysfs_entry max_corr_read_errors
=
3702 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3703 max_corrected_read_errors_store
);
3706 null_show(mddev_t
*mddev
, char *page
)
3712 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3714 /* buf must be %d:%d\n? giving major and minor numbers */
3715 /* The new device is added to the array.
3716 * If the array has a persistent superblock, we read the
3717 * superblock to initialise info and check validity.
3718 * Otherwise, only checking done is that in bind_rdev_to_array,
3719 * which mainly checks size.
3722 int major
= simple_strtoul(buf
, &e
, 10);
3728 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3730 minor
= simple_strtoul(e
+1, &e
, 10);
3731 if (*e
&& *e
!= '\n')
3733 dev
= MKDEV(major
, minor
);
3734 if (major
!= MAJOR(dev
) ||
3735 minor
!= MINOR(dev
))
3739 if (mddev
->persistent
) {
3740 rdev
= md_import_device(dev
, mddev
->major_version
,
3741 mddev
->minor_version
);
3742 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3743 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3744 mdk_rdev_t
, same_set
);
3745 err
= super_types
[mddev
->major_version
]
3746 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3750 } else if (mddev
->external
)
3751 rdev
= md_import_device(dev
, -2, -1);
3753 rdev
= md_import_device(dev
, -1, -1);
3756 return PTR_ERR(rdev
);
3757 err
= bind_rdev_to_array(rdev
, mddev
);
3761 return err
? err
: len
;
3764 static struct md_sysfs_entry md_new_device
=
3765 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3768 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3771 unsigned long chunk
, end_chunk
;
3775 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3777 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3778 if (buf
== end
) break;
3779 if (*end
== '-') { /* range */
3781 end_chunk
= simple_strtoul(buf
, &end
, 0);
3782 if (buf
== end
) break;
3784 if (*end
&& !isspace(*end
)) break;
3785 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3786 buf
= skip_spaces(end
);
3788 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3793 static struct md_sysfs_entry md_bitmap
=
3794 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3797 size_show(mddev_t
*mddev
, char *page
)
3799 return sprintf(page
, "%llu\n",
3800 (unsigned long long)mddev
->dev_sectors
/ 2);
3803 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3806 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3808 /* If array is inactive, we can reduce the component size, but
3809 * not increase it (except from 0).
3810 * If array is active, we can try an on-line resize
3813 int err
= strict_blocks_to_sectors(buf
, §ors
);
3818 err
= update_size(mddev
, sectors
);
3819 md_update_sb(mddev
, 1);
3821 if (mddev
->dev_sectors
== 0 ||
3822 mddev
->dev_sectors
> sectors
)
3823 mddev
->dev_sectors
= sectors
;
3827 return err
? err
: len
;
3830 static struct md_sysfs_entry md_size
=
3831 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3836 * 'none' for arrays with no metadata (good luck...)
3837 * 'external' for arrays with externally managed metadata,
3838 * or N.M for internally known formats
3841 metadata_show(mddev_t
*mddev
, char *page
)
3843 if (mddev
->persistent
)
3844 return sprintf(page
, "%d.%d\n",
3845 mddev
->major_version
, mddev
->minor_version
);
3846 else if (mddev
->external
)
3847 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3849 return sprintf(page
, "none\n");
3853 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3857 /* Changing the details of 'external' metadata is
3858 * always permitted. Otherwise there must be
3859 * no devices attached to the array.
3861 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3863 else if (!list_empty(&mddev
->disks
))
3866 if (cmd_match(buf
, "none")) {
3867 mddev
->persistent
= 0;
3868 mddev
->external
= 0;
3869 mddev
->major_version
= 0;
3870 mddev
->minor_version
= 90;
3873 if (strncmp(buf
, "external:", 9) == 0) {
3874 size_t namelen
= len
-9;
3875 if (namelen
>= sizeof(mddev
->metadata_type
))
3876 namelen
= sizeof(mddev
->metadata_type
)-1;
3877 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3878 mddev
->metadata_type
[namelen
] = 0;
3879 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3880 mddev
->metadata_type
[--namelen
] = 0;
3881 mddev
->persistent
= 0;
3882 mddev
->external
= 1;
3883 mddev
->major_version
= 0;
3884 mddev
->minor_version
= 90;
3887 major
= simple_strtoul(buf
, &e
, 10);
3888 if (e
==buf
|| *e
!= '.')
3891 minor
= simple_strtoul(buf
, &e
, 10);
3892 if (e
==buf
|| (*e
&& *e
!= '\n') )
3894 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3896 mddev
->major_version
= major
;
3897 mddev
->minor_version
= minor
;
3898 mddev
->persistent
= 1;
3899 mddev
->external
= 0;
3903 static struct md_sysfs_entry md_metadata
=
3904 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
3907 action_show(mddev_t
*mddev
, char *page
)
3909 char *type
= "idle";
3910 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3912 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3913 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
3914 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
3916 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
3917 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
3919 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
3923 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
3926 return sprintf(page
, "%s\n", type
);
3929 static void reap_sync_thread(mddev_t
*mddev
);
3932 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
3934 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
3937 if (cmd_match(page
, "frozen"))
3938 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3940 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
3942 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
3943 if (mddev
->sync_thread
) {
3944 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
3945 reap_sync_thread(mddev
);
3947 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
3948 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
3950 else if (cmd_match(page
, "resync"))
3951 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3952 else if (cmd_match(page
, "recover")) {
3953 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
3954 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3955 } else if (cmd_match(page
, "reshape")) {
3957 if (mddev
->pers
->start_reshape
== NULL
)
3959 err
= mddev
->pers
->start_reshape(mddev
);
3962 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
3964 if (cmd_match(page
, "check"))
3965 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
3966 else if (!cmd_match(page
, "repair"))
3968 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
3969 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
3971 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3972 md_wakeup_thread(mddev
->thread
);
3973 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
3978 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
3980 return sprintf(page
, "%llu\n",
3981 (unsigned long long) mddev
->resync_mismatches
);
3984 static struct md_sysfs_entry md_scan_mode
=
3985 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
3988 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
3991 sync_min_show(mddev_t
*mddev
, char *page
)
3993 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
3994 mddev
->sync_speed_min
? "local": "system");
3998 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4002 if (strncmp(buf
, "system", 6)==0) {
4003 mddev
->sync_speed_min
= 0;
4006 min
= simple_strtoul(buf
, &e
, 10);
4007 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4009 mddev
->sync_speed_min
= min
;
4013 static struct md_sysfs_entry md_sync_min
=
4014 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4017 sync_max_show(mddev_t
*mddev
, char *page
)
4019 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4020 mddev
->sync_speed_max
? "local": "system");
4024 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4028 if (strncmp(buf
, "system", 6)==0) {
4029 mddev
->sync_speed_max
= 0;
4032 max
= simple_strtoul(buf
, &e
, 10);
4033 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4035 mddev
->sync_speed_max
= max
;
4039 static struct md_sysfs_entry md_sync_max
=
4040 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4043 degraded_show(mddev_t
*mddev
, char *page
)
4045 return sprintf(page
, "%d\n", mddev
->degraded
);
4047 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4050 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
4052 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4056 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4060 if (strict_strtol(buf
, 10, &n
))
4063 if (n
!= 0 && n
!= 1)
4066 mddev
->parallel_resync
= n
;
4068 if (mddev
->sync_thread
)
4069 wake_up(&resync_wait
);
4074 /* force parallel resync, even with shared block devices */
4075 static struct md_sysfs_entry md_sync_force_parallel
=
4076 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4077 sync_force_parallel_show
, sync_force_parallel_store
);
4080 sync_speed_show(mddev_t
*mddev
, char *page
)
4082 unsigned long resync
, dt
, db
;
4083 if (mddev
->curr_resync
== 0)
4084 return sprintf(page
, "none\n");
4085 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4086 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4088 db
= resync
- mddev
->resync_mark_cnt
;
4089 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4092 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4095 sync_completed_show(mddev_t
*mddev
, char *page
)
4097 unsigned long long max_sectors
, resync
;
4099 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4100 return sprintf(page
, "none\n");
4102 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4103 max_sectors
= mddev
->resync_max_sectors
;
4105 max_sectors
= mddev
->dev_sectors
;
4107 resync
= mddev
->curr_resync_completed
;
4108 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4111 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4114 min_sync_show(mddev_t
*mddev
, char *page
)
4116 return sprintf(page
, "%llu\n",
4117 (unsigned long long)mddev
->resync_min
);
4120 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4122 unsigned long long min
;
4123 if (strict_strtoull(buf
, 10, &min
))
4125 if (min
> mddev
->resync_max
)
4127 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4130 /* Must be a multiple of chunk_size */
4131 if (mddev
->chunk_sectors
) {
4132 sector_t temp
= min
;
4133 if (sector_div(temp
, mddev
->chunk_sectors
))
4136 mddev
->resync_min
= min
;
4141 static struct md_sysfs_entry md_min_sync
=
4142 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4145 max_sync_show(mddev_t
*mddev
, char *page
)
4147 if (mddev
->resync_max
== MaxSector
)
4148 return sprintf(page
, "max\n");
4150 return sprintf(page
, "%llu\n",
4151 (unsigned long long)mddev
->resync_max
);
4154 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4156 if (strncmp(buf
, "max", 3) == 0)
4157 mddev
->resync_max
= MaxSector
;
4159 unsigned long long max
;
4160 if (strict_strtoull(buf
, 10, &max
))
4162 if (max
< mddev
->resync_min
)
4164 if (max
< mddev
->resync_max
&&
4166 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4169 /* Must be a multiple of chunk_size */
4170 if (mddev
->chunk_sectors
) {
4171 sector_t temp
= max
;
4172 if (sector_div(temp
, mddev
->chunk_sectors
))
4175 mddev
->resync_max
= max
;
4177 wake_up(&mddev
->recovery_wait
);
4181 static struct md_sysfs_entry md_max_sync
=
4182 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4185 suspend_lo_show(mddev_t
*mddev
, char *page
)
4187 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4191 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4194 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4195 unsigned long long old
= mddev
->suspend_lo
;
4197 if (mddev
->pers
== NULL
||
4198 mddev
->pers
->quiesce
== NULL
)
4200 if (buf
== e
|| (*e
&& *e
!= '\n'))
4203 mddev
->suspend_lo
= new;
4205 /* Shrinking suspended region */
4206 mddev
->pers
->quiesce(mddev
, 2);
4208 /* Expanding suspended region - need to wait */
4209 mddev
->pers
->quiesce(mddev
, 1);
4210 mddev
->pers
->quiesce(mddev
, 0);
4214 static struct md_sysfs_entry md_suspend_lo
=
4215 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4219 suspend_hi_show(mddev_t
*mddev
, char *page
)
4221 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4225 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4228 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4229 unsigned long long old
= mddev
->suspend_hi
;
4231 if (mddev
->pers
== NULL
||
4232 mddev
->pers
->quiesce
== NULL
)
4234 if (buf
== e
|| (*e
&& *e
!= '\n'))
4237 mddev
->suspend_hi
= new;
4239 /* Shrinking suspended region */
4240 mddev
->pers
->quiesce(mddev
, 2);
4242 /* Expanding suspended region - need to wait */
4243 mddev
->pers
->quiesce(mddev
, 1);
4244 mddev
->pers
->quiesce(mddev
, 0);
4248 static struct md_sysfs_entry md_suspend_hi
=
4249 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4252 reshape_position_show(mddev_t
*mddev
, char *page
)
4254 if (mddev
->reshape_position
!= MaxSector
)
4255 return sprintf(page
, "%llu\n",
4256 (unsigned long long)mddev
->reshape_position
);
4257 strcpy(page
, "none\n");
4262 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4265 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4268 if (buf
== e
|| (*e
&& *e
!= '\n'))
4270 mddev
->reshape_position
= new;
4271 mddev
->delta_disks
= 0;
4272 mddev
->new_level
= mddev
->level
;
4273 mddev
->new_layout
= mddev
->layout
;
4274 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4278 static struct md_sysfs_entry md_reshape_position
=
4279 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4280 reshape_position_store
);
4283 array_size_show(mddev_t
*mddev
, char *page
)
4285 if (mddev
->external_size
)
4286 return sprintf(page
, "%llu\n",
4287 (unsigned long long)mddev
->array_sectors
/2);
4289 return sprintf(page
, "default\n");
4293 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4297 if (strncmp(buf
, "default", 7) == 0) {
4299 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4301 sectors
= mddev
->array_sectors
;
4303 mddev
->external_size
= 0;
4305 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4307 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4310 mddev
->external_size
= 1;
4313 mddev
->array_sectors
= sectors
;
4315 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4316 revalidate_disk(mddev
->gendisk
);
4321 static struct md_sysfs_entry md_array_size
=
4322 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4325 static struct attribute
*md_default_attrs
[] = {
4328 &md_raid_disks
.attr
,
4329 &md_chunk_size
.attr
,
4331 &md_resync_start
.attr
,
4333 &md_new_device
.attr
,
4334 &md_safe_delay
.attr
,
4335 &md_array_state
.attr
,
4336 &md_reshape_position
.attr
,
4337 &md_array_size
.attr
,
4338 &max_corr_read_errors
.attr
,
4342 static struct attribute
*md_redundancy_attrs
[] = {
4344 &md_mismatches
.attr
,
4347 &md_sync_speed
.attr
,
4348 &md_sync_force_parallel
.attr
,
4349 &md_sync_completed
.attr
,
4352 &md_suspend_lo
.attr
,
4353 &md_suspend_hi
.attr
,
4358 static struct attribute_group md_redundancy_group
= {
4360 .attrs
= md_redundancy_attrs
,
4365 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4367 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4368 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4373 rv
= mddev_lock(mddev
);
4375 rv
= entry
->show(mddev
, page
);
4376 mddev_unlock(mddev
);
4382 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4383 const char *page
, size_t length
)
4385 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4386 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4391 if (!capable(CAP_SYS_ADMIN
))
4393 rv
= mddev_lock(mddev
);
4394 if (mddev
->hold_active
== UNTIL_IOCTL
)
4395 mddev
->hold_active
= 0;
4397 rv
= entry
->store(mddev
, page
, length
);
4398 mddev_unlock(mddev
);
4403 static void md_free(struct kobject
*ko
)
4405 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4407 if (mddev
->sysfs_state
)
4408 sysfs_put(mddev
->sysfs_state
);
4410 if (mddev
->gendisk
) {
4411 del_gendisk(mddev
->gendisk
);
4412 put_disk(mddev
->gendisk
);
4415 blk_cleanup_queue(mddev
->queue
);
4420 static const struct sysfs_ops md_sysfs_ops
= {
4421 .show
= md_attr_show
,
4422 .store
= md_attr_store
,
4424 static struct kobj_type md_ktype
= {
4426 .sysfs_ops
= &md_sysfs_ops
,
4427 .default_attrs
= md_default_attrs
,
4432 static void mddev_delayed_delete(struct work_struct
*ws
)
4434 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4436 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4437 kobject_del(&mddev
->kobj
);
4438 kobject_put(&mddev
->kobj
);
4441 static int md_alloc(dev_t dev
, char *name
)
4443 static DEFINE_MUTEX(disks_mutex
);
4444 mddev_t
*mddev
= mddev_find(dev
);
4445 struct gendisk
*disk
;
4454 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4455 shift
= partitioned
? MdpMinorShift
: 0;
4456 unit
= MINOR(mddev
->unit
) >> shift
;
4458 /* wait for any previous instance of this device to be
4459 * completely removed (mddev_delayed_delete).
4461 flush_workqueue(md_misc_wq
);
4463 mutex_lock(&disks_mutex
);
4469 /* Need to ensure that 'name' is not a duplicate.
4472 spin_lock(&all_mddevs_lock
);
4474 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4475 if (mddev2
->gendisk
&&
4476 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4477 spin_unlock(&all_mddevs_lock
);
4480 spin_unlock(&all_mddevs_lock
);
4484 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4487 mddev
->queue
->queuedata
= mddev
;
4489 blk_queue_make_request(mddev
->queue
, md_make_request
);
4491 disk
= alloc_disk(1 << shift
);
4493 blk_cleanup_queue(mddev
->queue
);
4494 mddev
->queue
= NULL
;
4497 disk
->major
= MAJOR(mddev
->unit
);
4498 disk
->first_minor
= unit
<< shift
;
4500 strcpy(disk
->disk_name
, name
);
4501 else if (partitioned
)
4502 sprintf(disk
->disk_name
, "md_d%d", unit
);
4504 sprintf(disk
->disk_name
, "md%d", unit
);
4505 disk
->fops
= &md_fops
;
4506 disk
->private_data
= mddev
;
4507 disk
->queue
= mddev
->queue
;
4508 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4509 /* Allow extended partitions. This makes the
4510 * 'mdp' device redundant, but we can't really
4513 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4514 mddev
->gendisk
= disk
;
4515 /* As soon as we call add_disk(), another thread could get
4516 * through to md_open, so make sure it doesn't get too far
4518 mutex_lock(&mddev
->open_mutex
);
4521 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4522 &disk_to_dev(disk
)->kobj
, "%s", "md");
4524 /* This isn't possible, but as kobject_init_and_add is marked
4525 * __must_check, we must do something with the result
4527 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4531 if (mddev
->kobj
.sd
&&
4532 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4533 printk(KERN_DEBUG
"pointless warning\n");
4534 mutex_unlock(&mddev
->open_mutex
);
4536 mutex_unlock(&disks_mutex
);
4537 if (!error
&& mddev
->kobj
.sd
) {
4538 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4539 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4545 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4547 md_alloc(dev
, NULL
);
4551 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4553 /* val must be "md_*" where * is not all digits.
4554 * We allocate an array with a large free minor number, and
4555 * set the name to val. val must not already be an active name.
4557 int len
= strlen(val
);
4558 char buf
[DISK_NAME_LEN
];
4560 while (len
&& val
[len
-1] == '\n')
4562 if (len
>= DISK_NAME_LEN
)
4564 strlcpy(buf
, val
, len
+1);
4565 if (strncmp(buf
, "md_", 3) != 0)
4567 return md_alloc(0, buf
);
4570 static void md_safemode_timeout(unsigned long data
)
4572 mddev_t
*mddev
= (mddev_t
*) data
;
4574 if (!atomic_read(&mddev
->writes_pending
)) {
4575 mddev
->safemode
= 1;
4576 if (mddev
->external
)
4577 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4579 md_wakeup_thread(mddev
->thread
);
4582 static int start_dirty_degraded
;
4584 int md_run(mddev_t
*mddev
)
4588 struct mdk_personality
*pers
;
4590 if (list_empty(&mddev
->disks
))
4591 /* cannot run an array with no devices.. */
4596 /* Cannot run until previous stop completes properly */
4597 if (mddev
->sysfs_active
)
4601 * Analyze all RAID superblock(s)
4603 if (!mddev
->raid_disks
) {
4604 if (!mddev
->persistent
)
4609 if (mddev
->level
!= LEVEL_NONE
)
4610 request_module("md-level-%d", mddev
->level
);
4611 else if (mddev
->clevel
[0])
4612 request_module("md-%s", mddev
->clevel
);
4615 * Drop all container device buffers, from now on
4616 * the only valid external interface is through the md
4619 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4620 if (test_bit(Faulty
, &rdev
->flags
))
4622 sync_blockdev(rdev
->bdev
);
4623 invalidate_bdev(rdev
->bdev
);
4625 /* perform some consistency tests on the device.
4626 * We don't want the data to overlap the metadata,
4627 * Internal Bitmap issues have been handled elsewhere.
4629 if (rdev
->meta_bdev
) {
4630 /* Nothing to check */;
4631 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4632 if (mddev
->dev_sectors
&&
4633 rdev
->data_offset
+ mddev
->dev_sectors
4635 printk("md: %s: data overlaps metadata\n",
4640 if (rdev
->sb_start
+ rdev
->sb_size
/512
4641 > rdev
->data_offset
) {
4642 printk("md: %s: metadata overlaps data\n",
4647 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4650 if (mddev
->bio_set
== NULL
)
4651 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
4654 spin_lock(&pers_lock
);
4655 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4656 if (!pers
|| !try_module_get(pers
->owner
)) {
4657 spin_unlock(&pers_lock
);
4658 if (mddev
->level
!= LEVEL_NONE
)
4659 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4662 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4667 spin_unlock(&pers_lock
);
4668 if (mddev
->level
!= pers
->level
) {
4669 mddev
->level
= pers
->level
;
4670 mddev
->new_level
= pers
->level
;
4672 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4674 if (mddev
->reshape_position
!= MaxSector
&&
4675 pers
->start_reshape
== NULL
) {
4676 /* This personality cannot handle reshaping... */
4678 module_put(pers
->owner
);
4682 if (pers
->sync_request
) {
4683 /* Warn if this is a potentially silly
4686 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4690 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4691 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4693 rdev
->bdev
->bd_contains
==
4694 rdev2
->bdev
->bd_contains
) {
4696 "%s: WARNING: %s appears to be"
4697 " on the same physical disk as"
4700 bdevname(rdev
->bdev
,b
),
4701 bdevname(rdev2
->bdev
,b2
));
4708 "True protection against single-disk"
4709 " failure might be compromised.\n");
4712 mddev
->recovery
= 0;
4713 /* may be over-ridden by personality */
4714 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4716 mddev
->ok_start_degraded
= start_dirty_degraded
;
4718 if (start_readonly
&& mddev
->ro
== 0)
4719 mddev
->ro
= 2; /* read-only, but switch on first write */
4721 err
= mddev
->pers
->run(mddev
);
4723 printk(KERN_ERR
"md: pers->run() failed ...\n");
4724 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4725 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4726 " but 'external_size' not in effect?\n", __func__
);
4728 "md: invalid array_size %llu > default size %llu\n",
4729 (unsigned long long)mddev
->array_sectors
/ 2,
4730 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4732 mddev
->pers
->stop(mddev
);
4734 if (err
== 0 && mddev
->pers
->sync_request
) {
4735 err
= bitmap_create(mddev
);
4737 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4738 mdname(mddev
), err
);
4739 mddev
->pers
->stop(mddev
);
4743 module_put(mddev
->pers
->owner
);
4745 bitmap_destroy(mddev
);
4748 if (mddev
->pers
->sync_request
) {
4749 if (mddev
->kobj
.sd
&&
4750 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4752 "md: cannot register extra attributes for %s\n",
4754 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4755 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4758 atomic_set(&mddev
->writes_pending
,0);
4759 atomic_set(&mddev
->max_corr_read_errors
,
4760 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4761 mddev
->safemode
= 0;
4762 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4763 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4764 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4768 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4769 if (rdev
->raid_disk
>= 0)
4770 if (sysfs_link_rdev(mddev
, rdev
))
4771 /* failure here is OK */;
4773 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4776 md_update_sb(mddev
, 0);
4778 md_new_event(mddev
);
4779 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4780 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4781 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4784 EXPORT_SYMBOL_GPL(md_run
);
4786 static int do_md_run(mddev_t
*mddev
)
4790 err
= md_run(mddev
);
4793 err
= bitmap_load(mddev
);
4795 bitmap_destroy(mddev
);
4799 md_wakeup_thread(mddev
->thread
);
4800 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4802 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4803 revalidate_disk(mddev
->gendisk
);
4805 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4810 static int restart_array(mddev_t
*mddev
)
4812 struct gendisk
*disk
= mddev
->gendisk
;
4814 /* Complain if it has no devices */
4815 if (list_empty(&mddev
->disks
))
4821 mddev
->safemode
= 0;
4823 set_disk_ro(disk
, 0);
4824 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4826 /* Kick recovery or resync if necessary */
4827 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4828 md_wakeup_thread(mddev
->thread
);
4829 md_wakeup_thread(mddev
->sync_thread
);
4830 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4834 /* similar to deny_write_access, but accounts for our holding a reference
4835 * to the file ourselves */
4836 static int deny_bitmap_write_access(struct file
* file
)
4838 struct inode
*inode
= file
->f_mapping
->host
;
4840 spin_lock(&inode
->i_lock
);
4841 if (atomic_read(&inode
->i_writecount
) > 1) {
4842 spin_unlock(&inode
->i_lock
);
4845 atomic_set(&inode
->i_writecount
, -1);
4846 spin_unlock(&inode
->i_lock
);
4851 void restore_bitmap_write_access(struct file
*file
)
4853 struct inode
*inode
= file
->f_mapping
->host
;
4855 spin_lock(&inode
->i_lock
);
4856 atomic_set(&inode
->i_writecount
, 1);
4857 spin_unlock(&inode
->i_lock
);
4860 static void md_clean(mddev_t
*mddev
)
4862 mddev
->array_sectors
= 0;
4863 mddev
->external_size
= 0;
4864 mddev
->dev_sectors
= 0;
4865 mddev
->raid_disks
= 0;
4866 mddev
->recovery_cp
= 0;
4867 mddev
->resync_min
= 0;
4868 mddev
->resync_max
= MaxSector
;
4869 mddev
->reshape_position
= MaxSector
;
4870 mddev
->external
= 0;
4871 mddev
->persistent
= 0;
4872 mddev
->level
= LEVEL_NONE
;
4873 mddev
->clevel
[0] = 0;
4876 mddev
->metadata_type
[0] = 0;
4877 mddev
->chunk_sectors
= 0;
4878 mddev
->ctime
= mddev
->utime
= 0;
4880 mddev
->max_disks
= 0;
4882 mddev
->can_decrease_events
= 0;
4883 mddev
->delta_disks
= 0;
4884 mddev
->new_level
= LEVEL_NONE
;
4885 mddev
->new_layout
= 0;
4886 mddev
->new_chunk_sectors
= 0;
4887 mddev
->curr_resync
= 0;
4888 mddev
->resync_mismatches
= 0;
4889 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4890 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4891 mddev
->recovery
= 0;
4894 mddev
->degraded
= 0;
4895 mddev
->safemode
= 0;
4896 mddev
->bitmap_info
.offset
= 0;
4897 mddev
->bitmap_info
.default_offset
= 0;
4898 mddev
->bitmap_info
.chunksize
= 0;
4899 mddev
->bitmap_info
.daemon_sleep
= 0;
4900 mddev
->bitmap_info
.max_write_behind
= 0;
4903 static void __md_stop_writes(mddev_t
*mddev
)
4905 if (mddev
->sync_thread
) {
4906 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4907 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4908 reap_sync_thread(mddev
);
4911 del_timer_sync(&mddev
->safemode_timer
);
4913 bitmap_flush(mddev
);
4914 md_super_wait(mddev
);
4916 if (!mddev
->in_sync
|| mddev
->flags
) {
4917 /* mark array as shutdown cleanly */
4919 md_update_sb(mddev
, 1);
4923 void md_stop_writes(mddev_t
*mddev
)
4926 __md_stop_writes(mddev
);
4927 mddev_unlock(mddev
);
4929 EXPORT_SYMBOL_GPL(md_stop_writes
);
4931 void md_stop(mddev_t
*mddev
)
4934 mddev
->pers
->stop(mddev
);
4935 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
4936 mddev
->to_remove
= &md_redundancy_group
;
4937 module_put(mddev
->pers
->owner
);
4939 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4941 EXPORT_SYMBOL_GPL(md_stop
);
4943 static int md_set_readonly(mddev_t
*mddev
, int is_open
)
4946 mutex_lock(&mddev
->open_mutex
);
4947 if (atomic_read(&mddev
->openers
) > is_open
) {
4948 printk("md: %s still in use.\n",mdname(mddev
));
4953 __md_stop_writes(mddev
);
4959 set_disk_ro(mddev
->gendisk
, 1);
4960 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4961 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4965 mutex_unlock(&mddev
->open_mutex
);
4970 * 0 - completely stop and dis-assemble array
4971 * 2 - stop but do not disassemble array
4973 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
4975 struct gendisk
*disk
= mddev
->gendisk
;
4978 mutex_lock(&mddev
->open_mutex
);
4979 if (atomic_read(&mddev
->openers
) > is_open
||
4980 mddev
->sysfs_active
) {
4981 printk("md: %s still in use.\n",mdname(mddev
));
4982 mutex_unlock(&mddev
->open_mutex
);
4988 set_disk_ro(disk
, 0);
4990 __md_stop_writes(mddev
);
4992 mddev
->queue
->merge_bvec_fn
= NULL
;
4993 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
4995 /* tell userspace to handle 'inactive' */
4996 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4998 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4999 if (rdev
->raid_disk
>= 0)
5000 sysfs_unlink_rdev(mddev
, rdev
);
5002 set_capacity(disk
, 0);
5003 mutex_unlock(&mddev
->open_mutex
);
5005 revalidate_disk(disk
);
5010 mutex_unlock(&mddev
->open_mutex
);
5012 * Free resources if final stop
5015 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5017 bitmap_destroy(mddev
);
5018 if (mddev
->bitmap_info
.file
) {
5019 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5020 fput(mddev
->bitmap_info
.file
);
5021 mddev
->bitmap_info
.file
= NULL
;
5023 mddev
->bitmap_info
.offset
= 0;
5025 export_array(mddev
);
5028 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5029 if (mddev
->hold_active
== UNTIL_STOP
)
5030 mddev
->hold_active
= 0;
5032 blk_integrity_unregister(disk
);
5033 md_new_event(mddev
);
5034 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5039 static void autorun_array(mddev_t
*mddev
)
5044 if (list_empty(&mddev
->disks
))
5047 printk(KERN_INFO
"md: running: ");
5049 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5050 char b
[BDEVNAME_SIZE
];
5051 printk("<%s>", bdevname(rdev
->bdev
,b
));
5055 err
= do_md_run(mddev
);
5057 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5058 do_md_stop(mddev
, 0, 0);
5063 * lets try to run arrays based on all disks that have arrived
5064 * until now. (those are in pending_raid_disks)
5066 * the method: pick the first pending disk, collect all disks with
5067 * the same UUID, remove all from the pending list and put them into
5068 * the 'same_array' list. Then order this list based on superblock
5069 * update time (freshest comes first), kick out 'old' disks and
5070 * compare superblocks. If everything's fine then run it.
5072 * If "unit" is allocated, then bump its reference count
5074 static void autorun_devices(int part
)
5076 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
5078 char b
[BDEVNAME_SIZE
];
5080 printk(KERN_INFO
"md: autorun ...\n");
5081 while (!list_empty(&pending_raid_disks
)) {
5084 LIST_HEAD(candidates
);
5085 rdev0
= list_entry(pending_raid_disks
.next
,
5086 mdk_rdev_t
, same_set
);
5088 printk(KERN_INFO
"md: considering %s ...\n",
5089 bdevname(rdev0
->bdev
,b
));
5090 INIT_LIST_HEAD(&candidates
);
5091 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5092 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5093 printk(KERN_INFO
"md: adding %s ...\n",
5094 bdevname(rdev
->bdev
,b
));
5095 list_move(&rdev
->same_set
, &candidates
);
5098 * now we have a set of devices, with all of them having
5099 * mostly sane superblocks. It's time to allocate the
5103 dev
= MKDEV(mdp_major
,
5104 rdev0
->preferred_minor
<< MdpMinorShift
);
5105 unit
= MINOR(dev
) >> MdpMinorShift
;
5107 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5110 if (rdev0
->preferred_minor
!= unit
) {
5111 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5112 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5116 md_probe(dev
, NULL
, NULL
);
5117 mddev
= mddev_find(dev
);
5118 if (!mddev
|| !mddev
->gendisk
) {
5122 "md: cannot allocate memory for md drive.\n");
5125 if (mddev_lock(mddev
))
5126 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5128 else if (mddev
->raid_disks
|| mddev
->major_version
5129 || !list_empty(&mddev
->disks
)) {
5131 "md: %s already running, cannot run %s\n",
5132 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5133 mddev_unlock(mddev
);
5135 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5136 mddev
->persistent
= 1;
5137 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5138 list_del_init(&rdev
->same_set
);
5139 if (bind_rdev_to_array(rdev
, mddev
))
5142 autorun_array(mddev
);
5143 mddev_unlock(mddev
);
5145 /* on success, candidates will be empty, on error
5148 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5149 list_del_init(&rdev
->same_set
);
5154 printk(KERN_INFO
"md: ... autorun DONE.\n");
5156 #endif /* !MODULE */
5158 static int get_version(void __user
* arg
)
5162 ver
.major
= MD_MAJOR_VERSION
;
5163 ver
.minor
= MD_MINOR_VERSION
;
5164 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5166 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5172 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
5174 mdu_array_info_t info
;
5175 int nr
,working
,insync
,failed
,spare
;
5178 nr
=working
=insync
=failed
=spare
=0;
5179 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5181 if (test_bit(Faulty
, &rdev
->flags
))
5185 if (test_bit(In_sync
, &rdev
->flags
))
5192 info
.major_version
= mddev
->major_version
;
5193 info
.minor_version
= mddev
->minor_version
;
5194 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5195 info
.ctime
= mddev
->ctime
;
5196 info
.level
= mddev
->level
;
5197 info
.size
= mddev
->dev_sectors
/ 2;
5198 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5201 info
.raid_disks
= mddev
->raid_disks
;
5202 info
.md_minor
= mddev
->md_minor
;
5203 info
.not_persistent
= !mddev
->persistent
;
5205 info
.utime
= mddev
->utime
;
5208 info
.state
= (1<<MD_SB_CLEAN
);
5209 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5210 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5211 info
.active_disks
= insync
;
5212 info
.working_disks
= working
;
5213 info
.failed_disks
= failed
;
5214 info
.spare_disks
= spare
;
5216 info
.layout
= mddev
->layout
;
5217 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5219 if (copy_to_user(arg
, &info
, sizeof(info
)))
5225 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
5227 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5228 char *ptr
, *buf
= NULL
;
5231 if (md_allow_write(mddev
))
5232 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5234 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5239 /* bitmap disabled, zero the first byte and copy out */
5240 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5241 file
->pathname
[0] = '\0';
5245 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5249 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5253 strcpy(file
->pathname
, ptr
);
5257 if (copy_to_user(arg
, file
, sizeof(*file
)))
5265 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
5267 mdu_disk_info_t info
;
5270 if (copy_from_user(&info
, arg
, sizeof(info
)))
5273 rdev
= find_rdev_nr(mddev
, info
.number
);
5275 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5276 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5277 info
.raid_disk
= rdev
->raid_disk
;
5279 if (test_bit(Faulty
, &rdev
->flags
))
5280 info
.state
|= (1<<MD_DISK_FAULTY
);
5281 else if (test_bit(In_sync
, &rdev
->flags
)) {
5282 info
.state
|= (1<<MD_DISK_ACTIVE
);
5283 info
.state
|= (1<<MD_DISK_SYNC
);
5285 if (test_bit(WriteMostly
, &rdev
->flags
))
5286 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5288 info
.major
= info
.minor
= 0;
5289 info
.raid_disk
= -1;
5290 info
.state
= (1<<MD_DISK_REMOVED
);
5293 if (copy_to_user(arg
, &info
, sizeof(info
)))
5299 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
5301 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5303 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5305 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5308 if (!mddev
->raid_disks
) {
5310 /* expecting a device which has a superblock */
5311 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5314 "md: md_import_device returned %ld\n",
5316 return PTR_ERR(rdev
);
5318 if (!list_empty(&mddev
->disks
)) {
5319 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
5320 mdk_rdev_t
, same_set
);
5321 err
= super_types
[mddev
->major_version
]
5322 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5325 "md: %s has different UUID to %s\n",
5326 bdevname(rdev
->bdev
,b
),
5327 bdevname(rdev0
->bdev
,b2
));
5332 err
= bind_rdev_to_array(rdev
, mddev
);
5339 * add_new_disk can be used once the array is assembled
5340 * to add "hot spares". They must already have a superblock
5345 if (!mddev
->pers
->hot_add_disk
) {
5347 "%s: personality does not support diskops!\n",
5351 if (mddev
->persistent
)
5352 rdev
= md_import_device(dev
, mddev
->major_version
,
5353 mddev
->minor_version
);
5355 rdev
= md_import_device(dev
, -1, -1);
5358 "md: md_import_device returned %ld\n",
5360 return PTR_ERR(rdev
);
5362 /* set saved_raid_disk if appropriate */
5363 if (!mddev
->persistent
) {
5364 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5365 info
->raid_disk
< mddev
->raid_disks
) {
5366 rdev
->raid_disk
= info
->raid_disk
;
5367 set_bit(In_sync
, &rdev
->flags
);
5369 rdev
->raid_disk
= -1;
5371 super_types
[mddev
->major_version
].
5372 validate_super(mddev
, rdev
);
5373 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5374 (!test_bit(In_sync
, &rdev
->flags
) ||
5375 rdev
->raid_disk
!= info
->raid_disk
)) {
5376 /* This was a hot-add request, but events doesn't
5377 * match, so reject it.
5383 if (test_bit(In_sync
, &rdev
->flags
))
5384 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5386 rdev
->saved_raid_disk
= -1;
5388 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5389 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5390 set_bit(WriteMostly
, &rdev
->flags
);
5392 clear_bit(WriteMostly
, &rdev
->flags
);
5394 rdev
->raid_disk
= -1;
5395 err
= bind_rdev_to_array(rdev
, mddev
);
5396 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5397 /* If there is hot_add_disk but no hot_remove_disk
5398 * then added disks for geometry changes,
5399 * and should be added immediately.
5401 super_types
[mddev
->major_version
].
5402 validate_super(mddev
, rdev
);
5403 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5405 unbind_rdev_from_array(rdev
);
5410 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5412 md_update_sb(mddev
, 1);
5413 if (mddev
->degraded
)
5414 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5415 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5417 md_new_event(mddev
);
5418 md_wakeup_thread(mddev
->thread
);
5422 /* otherwise, add_new_disk is only allowed
5423 * for major_version==0 superblocks
5425 if (mddev
->major_version
!= 0) {
5426 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5431 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5433 rdev
= md_import_device(dev
, -1, 0);
5436 "md: error, md_import_device() returned %ld\n",
5438 return PTR_ERR(rdev
);
5440 rdev
->desc_nr
= info
->number
;
5441 if (info
->raid_disk
< mddev
->raid_disks
)
5442 rdev
->raid_disk
= info
->raid_disk
;
5444 rdev
->raid_disk
= -1;
5446 if (rdev
->raid_disk
< mddev
->raid_disks
)
5447 if (info
->state
& (1<<MD_DISK_SYNC
))
5448 set_bit(In_sync
, &rdev
->flags
);
5450 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5451 set_bit(WriteMostly
, &rdev
->flags
);
5453 if (!mddev
->persistent
) {
5454 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5455 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5457 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5458 rdev
->sectors
= rdev
->sb_start
;
5460 err
= bind_rdev_to_array(rdev
, mddev
);
5470 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5472 char b
[BDEVNAME_SIZE
];
5475 rdev
= find_rdev(mddev
, dev
);
5479 if (rdev
->raid_disk
>= 0)
5482 kick_rdev_from_array(rdev
);
5483 md_update_sb(mddev
, 1);
5484 md_new_event(mddev
);
5488 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5489 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5493 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5495 char b
[BDEVNAME_SIZE
];
5502 if (mddev
->major_version
!= 0) {
5503 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5504 " version-0 superblocks.\n",
5508 if (!mddev
->pers
->hot_add_disk
) {
5510 "%s: personality does not support diskops!\n",
5515 rdev
= md_import_device(dev
, -1, 0);
5518 "md: error, md_import_device() returned %ld\n",
5523 if (mddev
->persistent
)
5524 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5526 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5528 rdev
->sectors
= rdev
->sb_start
;
5530 if (test_bit(Faulty
, &rdev
->flags
)) {
5532 "md: can not hot-add faulty %s disk to %s!\n",
5533 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5537 clear_bit(In_sync
, &rdev
->flags
);
5539 rdev
->saved_raid_disk
= -1;
5540 err
= bind_rdev_to_array(rdev
, mddev
);
5545 * The rest should better be atomic, we can have disk failures
5546 * noticed in interrupt contexts ...
5549 rdev
->raid_disk
= -1;
5551 md_update_sb(mddev
, 1);
5554 * Kick recovery, maybe this spare has to be added to the
5555 * array immediately.
5557 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5558 md_wakeup_thread(mddev
->thread
);
5559 md_new_event(mddev
);
5567 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5572 if (!mddev
->pers
->quiesce
)
5574 if (mddev
->recovery
|| mddev
->sync_thread
)
5576 /* we should be able to change the bitmap.. */
5582 return -EEXIST
; /* cannot add when bitmap is present */
5583 mddev
->bitmap_info
.file
= fget(fd
);
5585 if (mddev
->bitmap_info
.file
== NULL
) {
5586 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5591 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5593 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5595 fput(mddev
->bitmap_info
.file
);
5596 mddev
->bitmap_info
.file
= NULL
;
5599 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5600 } else if (mddev
->bitmap
== NULL
)
5601 return -ENOENT
; /* cannot remove what isn't there */
5604 mddev
->pers
->quiesce(mddev
, 1);
5606 err
= bitmap_create(mddev
);
5608 err
= bitmap_load(mddev
);
5610 if (fd
< 0 || err
) {
5611 bitmap_destroy(mddev
);
5612 fd
= -1; /* make sure to put the file */
5614 mddev
->pers
->quiesce(mddev
, 0);
5617 if (mddev
->bitmap_info
.file
) {
5618 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5619 fput(mddev
->bitmap_info
.file
);
5621 mddev
->bitmap_info
.file
= NULL
;
5628 * set_array_info is used two different ways
5629 * The original usage is when creating a new array.
5630 * In this usage, raid_disks is > 0 and it together with
5631 * level, size, not_persistent,layout,chunksize determine the
5632 * shape of the array.
5633 * This will always create an array with a type-0.90.0 superblock.
5634 * The newer usage is when assembling an array.
5635 * In this case raid_disks will be 0, and the major_version field is
5636 * use to determine which style super-blocks are to be found on the devices.
5637 * The minor and patch _version numbers are also kept incase the
5638 * super_block handler wishes to interpret them.
5640 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5643 if (info
->raid_disks
== 0) {
5644 /* just setting version number for superblock loading */
5645 if (info
->major_version
< 0 ||
5646 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5647 super_types
[info
->major_version
].name
== NULL
) {
5648 /* maybe try to auto-load a module? */
5650 "md: superblock version %d not known\n",
5651 info
->major_version
);
5654 mddev
->major_version
= info
->major_version
;
5655 mddev
->minor_version
= info
->minor_version
;
5656 mddev
->patch_version
= info
->patch_version
;
5657 mddev
->persistent
= !info
->not_persistent
;
5658 /* ensure mddev_put doesn't delete this now that there
5659 * is some minimal configuration.
5661 mddev
->ctime
= get_seconds();
5664 mddev
->major_version
= MD_MAJOR_VERSION
;
5665 mddev
->minor_version
= MD_MINOR_VERSION
;
5666 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5667 mddev
->ctime
= get_seconds();
5669 mddev
->level
= info
->level
;
5670 mddev
->clevel
[0] = 0;
5671 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5672 mddev
->raid_disks
= info
->raid_disks
;
5673 /* don't set md_minor, it is determined by which /dev/md* was
5676 if (info
->state
& (1<<MD_SB_CLEAN
))
5677 mddev
->recovery_cp
= MaxSector
;
5679 mddev
->recovery_cp
= 0;
5680 mddev
->persistent
= ! info
->not_persistent
;
5681 mddev
->external
= 0;
5683 mddev
->layout
= info
->layout
;
5684 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5686 mddev
->max_disks
= MD_SB_DISKS
;
5688 if (mddev
->persistent
)
5690 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5692 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5693 mddev
->bitmap_info
.offset
= 0;
5695 mddev
->reshape_position
= MaxSector
;
5698 * Generate a 128 bit UUID
5700 get_random_bytes(mddev
->uuid
, 16);
5702 mddev
->new_level
= mddev
->level
;
5703 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5704 mddev
->new_layout
= mddev
->layout
;
5705 mddev
->delta_disks
= 0;
5710 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5712 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5714 if (mddev
->external_size
)
5717 mddev
->array_sectors
= array_sectors
;
5719 EXPORT_SYMBOL(md_set_array_sectors
);
5721 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5725 int fit
= (num_sectors
== 0);
5727 if (mddev
->pers
->resize
== NULL
)
5729 /* The "num_sectors" is the number of sectors of each device that
5730 * is used. This can only make sense for arrays with redundancy.
5731 * linear and raid0 always use whatever space is available. We can only
5732 * consider changing this number if no resync or reconstruction is
5733 * happening, and if the new size is acceptable. It must fit before the
5734 * sb_start or, if that is <data_offset, it must fit before the size
5735 * of each device. If num_sectors is zero, we find the largest size
5738 if (mddev
->sync_thread
)
5741 /* Sorry, cannot grow a bitmap yet, just remove it,
5745 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5746 sector_t avail
= rdev
->sectors
;
5748 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5749 num_sectors
= avail
;
5750 if (avail
< num_sectors
)
5753 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5755 revalidate_disk(mddev
->gendisk
);
5759 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5762 /* change the number of raid disks */
5763 if (mddev
->pers
->check_reshape
== NULL
)
5765 if (raid_disks
<= 0 ||
5766 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5768 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5770 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5772 rv
= mddev
->pers
->check_reshape(mddev
);
5774 mddev
->delta_disks
= 0;
5780 * update_array_info is used to change the configuration of an
5782 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5783 * fields in the info are checked against the array.
5784 * Any differences that cannot be handled will cause an error.
5785 * Normally, only one change can be managed at a time.
5787 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5793 /* calculate expected state,ignoring low bits */
5794 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5795 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5797 if (mddev
->major_version
!= info
->major_version
||
5798 mddev
->minor_version
!= info
->minor_version
||
5799 /* mddev->patch_version != info->patch_version || */
5800 mddev
->ctime
!= info
->ctime
||
5801 mddev
->level
!= info
->level
||
5802 /* mddev->layout != info->layout || */
5803 !mddev
->persistent
!= info
->not_persistent
||
5804 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5805 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5806 ((state
^info
->state
) & 0xfffffe00)
5809 /* Check there is only one change */
5810 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5812 if (mddev
->raid_disks
!= info
->raid_disks
)
5814 if (mddev
->layout
!= info
->layout
)
5816 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5823 if (mddev
->layout
!= info
->layout
) {
5825 * we don't need to do anything at the md level, the
5826 * personality will take care of it all.
5828 if (mddev
->pers
->check_reshape
== NULL
)
5831 mddev
->new_layout
= info
->layout
;
5832 rv
= mddev
->pers
->check_reshape(mddev
);
5834 mddev
->new_layout
= mddev
->layout
;
5838 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5839 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5841 if (mddev
->raid_disks
!= info
->raid_disks
)
5842 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5844 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5845 if (mddev
->pers
->quiesce
== NULL
)
5847 if (mddev
->recovery
|| mddev
->sync_thread
)
5849 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5850 /* add the bitmap */
5853 if (mddev
->bitmap_info
.default_offset
== 0)
5855 mddev
->bitmap_info
.offset
=
5856 mddev
->bitmap_info
.default_offset
;
5857 mddev
->pers
->quiesce(mddev
, 1);
5858 rv
= bitmap_create(mddev
);
5860 rv
= bitmap_load(mddev
);
5862 bitmap_destroy(mddev
);
5863 mddev
->pers
->quiesce(mddev
, 0);
5865 /* remove the bitmap */
5868 if (mddev
->bitmap
->file
)
5870 mddev
->pers
->quiesce(mddev
, 1);
5871 bitmap_destroy(mddev
);
5872 mddev
->pers
->quiesce(mddev
, 0);
5873 mddev
->bitmap_info
.offset
= 0;
5876 md_update_sb(mddev
, 1);
5880 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5884 if (mddev
->pers
== NULL
)
5887 rdev
= find_rdev(mddev
, dev
);
5891 md_error(mddev
, rdev
);
5896 * We have a problem here : there is no easy way to give a CHS
5897 * virtual geometry. We currently pretend that we have a 2 heads
5898 * 4 sectors (with a BIG number of cylinders...). This drives
5899 * dosfs just mad... ;-)
5901 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
5903 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
5907 geo
->cylinders
= mddev
->array_sectors
/ 8;
5911 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
5912 unsigned int cmd
, unsigned long arg
)
5915 void __user
*argp
= (void __user
*)arg
;
5916 mddev_t
*mddev
= NULL
;
5919 if (!capable(CAP_SYS_ADMIN
))
5923 * Commands dealing with the RAID driver but not any
5929 err
= get_version(argp
);
5932 case PRINT_RAID_DEBUG
:
5940 autostart_arrays(arg
);
5947 * Commands creating/starting a new array:
5950 mddev
= bdev
->bd_disk
->private_data
;
5957 err
= mddev_lock(mddev
);
5960 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5967 case SET_ARRAY_INFO
:
5969 mdu_array_info_t info
;
5971 memset(&info
, 0, sizeof(info
));
5972 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
5977 err
= update_array_info(mddev
, &info
);
5979 printk(KERN_WARNING
"md: couldn't update"
5980 " array info. %d\n", err
);
5985 if (!list_empty(&mddev
->disks
)) {
5987 "md: array %s already has disks!\n",
5992 if (mddev
->raid_disks
) {
5994 "md: array %s already initialised!\n",
5999 err
= set_array_info(mddev
, &info
);
6001 printk(KERN_WARNING
"md: couldn't set"
6002 " array info. %d\n", err
);
6012 * Commands querying/configuring an existing array:
6014 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6015 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6016 if ((!mddev
->raid_disks
&& !mddev
->external
)
6017 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6018 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6019 && cmd
!= GET_BITMAP_FILE
) {
6025 * Commands even a read-only array can execute:
6029 case GET_ARRAY_INFO
:
6030 err
= get_array_info(mddev
, argp
);
6033 case GET_BITMAP_FILE
:
6034 err
= get_bitmap_file(mddev
, argp
);
6038 err
= get_disk_info(mddev
, argp
);
6041 case RESTART_ARRAY_RW
:
6042 err
= restart_array(mddev
);
6046 err
= do_md_stop(mddev
, 0, 1);
6050 err
= md_set_readonly(mddev
, 1);
6054 if (get_user(ro
, (int __user
*)(arg
))) {
6060 /* if the bdev is going readonly the value of mddev->ro
6061 * does not matter, no writes are coming
6066 /* are we are already prepared for writes? */
6070 /* transitioning to readauto need only happen for
6071 * arrays that call md_write_start
6074 err
= restart_array(mddev
);
6077 set_disk_ro(mddev
->gendisk
, 0);
6084 * The remaining ioctls are changing the state of the
6085 * superblock, so we do not allow them on read-only arrays.
6086 * However non-MD ioctls (e.g. get-size) will still come through
6087 * here and hit the 'default' below, so only disallow
6088 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6090 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6091 if (mddev
->ro
== 2) {
6093 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6094 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6095 md_wakeup_thread(mddev
->thread
);
6106 mdu_disk_info_t info
;
6107 if (copy_from_user(&info
, argp
, sizeof(info
)))
6110 err
= add_new_disk(mddev
, &info
);
6114 case HOT_REMOVE_DISK
:
6115 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6119 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6122 case SET_DISK_FAULTY
:
6123 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6127 err
= do_md_run(mddev
);
6130 case SET_BITMAP_FILE
:
6131 err
= set_bitmap_file(mddev
, (int)arg
);
6141 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6143 mddev
->hold_active
= 0;
6144 mddev_unlock(mddev
);
6153 #ifdef CONFIG_COMPAT
6154 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6155 unsigned int cmd
, unsigned long arg
)
6158 case HOT_REMOVE_DISK
:
6160 case SET_DISK_FAULTY
:
6161 case SET_BITMAP_FILE
:
6162 /* These take in integer arg, do not convert */
6165 arg
= (unsigned long)compat_ptr(arg
);
6169 return md_ioctl(bdev
, mode
, cmd
, arg
);
6171 #endif /* CONFIG_COMPAT */
6173 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6176 * Succeed if we can lock the mddev, which confirms that
6177 * it isn't being stopped right now.
6179 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
6182 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6183 /* we are racing with mddev_put which is discarding this
6187 /* Wait until bdev->bd_disk is definitely gone */
6188 flush_workqueue(md_misc_wq
);
6189 /* Then retry the open from the top */
6190 return -ERESTARTSYS
;
6192 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6194 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6198 atomic_inc(&mddev
->openers
);
6199 mutex_unlock(&mddev
->open_mutex
);
6201 check_disk_change(bdev
);
6206 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6208 mddev_t
*mddev
= disk
->private_data
;
6211 atomic_dec(&mddev
->openers
);
6217 static int md_media_changed(struct gendisk
*disk
)
6219 mddev_t
*mddev
= disk
->private_data
;
6221 return mddev
->changed
;
6224 static int md_revalidate(struct gendisk
*disk
)
6226 mddev_t
*mddev
= disk
->private_data
;
6231 static const struct block_device_operations md_fops
=
6233 .owner
= THIS_MODULE
,
6235 .release
= md_release
,
6237 #ifdef CONFIG_COMPAT
6238 .compat_ioctl
= md_compat_ioctl
,
6240 .getgeo
= md_getgeo
,
6241 .media_changed
= md_media_changed
,
6242 .revalidate_disk
= md_revalidate
,
6245 static int md_thread(void * arg
)
6247 mdk_thread_t
*thread
= arg
;
6250 * md_thread is a 'system-thread', it's priority should be very
6251 * high. We avoid resource deadlocks individually in each
6252 * raid personality. (RAID5 does preallocation) We also use RR and
6253 * the very same RT priority as kswapd, thus we will never get
6254 * into a priority inversion deadlock.
6256 * we definitely have to have equal or higher priority than
6257 * bdflush, otherwise bdflush will deadlock if there are too
6258 * many dirty RAID5 blocks.
6261 allow_signal(SIGKILL
);
6262 while (!kthread_should_stop()) {
6264 /* We need to wait INTERRUPTIBLE so that
6265 * we don't add to the load-average.
6266 * That means we need to be sure no signals are
6269 if (signal_pending(current
))
6270 flush_signals(current
);
6272 wait_event_interruptible_timeout
6274 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6275 || kthread_should_stop(),
6278 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6279 if (!kthread_should_stop())
6280 thread
->run(thread
->mddev
);
6286 void md_wakeup_thread(mdk_thread_t
*thread
)
6289 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6290 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6291 wake_up(&thread
->wqueue
);
6295 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
6298 mdk_thread_t
*thread
;
6300 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
6304 init_waitqueue_head(&thread
->wqueue
);
6307 thread
->mddev
= mddev
;
6308 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6309 thread
->tsk
= kthread_run(md_thread
, thread
,
6311 mdname(thread
->mddev
),
6312 name
?: mddev
->pers
->name
);
6313 if (IS_ERR(thread
->tsk
)) {
6320 void md_unregister_thread(mdk_thread_t
*thread
)
6324 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6326 kthread_stop(thread
->tsk
);
6330 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
6337 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6340 if (mddev
->external
)
6341 set_bit(Blocked
, &rdev
->flags
);
6343 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6345 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6346 __builtin_return_address(0),__builtin_return_address(1),
6347 __builtin_return_address(2),__builtin_return_address(3));
6351 if (!mddev
->pers
->error_handler
)
6353 mddev
->pers
->error_handler(mddev
,rdev
);
6354 if (mddev
->degraded
)
6355 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6356 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6357 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6358 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6359 md_wakeup_thread(mddev
->thread
);
6360 if (mddev
->event_work
.func
)
6361 queue_work(md_misc_wq
, &mddev
->event_work
);
6362 md_new_event_inintr(mddev
);
6365 /* seq_file implementation /proc/mdstat */
6367 static void status_unused(struct seq_file
*seq
)
6372 seq_printf(seq
, "unused devices: ");
6374 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6375 char b
[BDEVNAME_SIZE
];
6377 seq_printf(seq
, "%s ",
6378 bdevname(rdev
->bdev
,b
));
6381 seq_printf(seq
, "<none>");
6383 seq_printf(seq
, "\n");
6387 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6389 sector_t max_sectors
, resync
, res
;
6390 unsigned long dt
, db
;
6393 unsigned int per_milli
;
6395 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6397 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6398 max_sectors
= mddev
->resync_max_sectors
;
6400 max_sectors
= mddev
->dev_sectors
;
6403 * Should not happen.
6409 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6410 * in a sector_t, and (max_sectors>>scale) will fit in a
6411 * u32, as those are the requirements for sector_div.
6412 * Thus 'scale' must be at least 10
6415 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6416 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6419 res
= (resync
>>scale
)*1000;
6420 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6424 int i
, x
= per_milli
/50, y
= 20-x
;
6425 seq_printf(seq
, "[");
6426 for (i
= 0; i
< x
; i
++)
6427 seq_printf(seq
, "=");
6428 seq_printf(seq
, ">");
6429 for (i
= 0; i
< y
; i
++)
6430 seq_printf(seq
, ".");
6431 seq_printf(seq
, "] ");
6433 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6434 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6436 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6438 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6439 "resync" : "recovery"))),
6440 per_milli
/10, per_milli
% 10,
6441 (unsigned long long) resync
/2,
6442 (unsigned long long) max_sectors
/2);
6445 * dt: time from mark until now
6446 * db: blocks written from mark until now
6447 * rt: remaining time
6449 * rt is a sector_t, so could be 32bit or 64bit.
6450 * So we divide before multiply in case it is 32bit and close
6452 * We scale the divisor (db) by 32 to avoid losing precision
6453 * near the end of resync when the number of remaining sectors
6455 * We then divide rt by 32 after multiplying by db to compensate.
6456 * The '+1' avoids division by zero if db is very small.
6458 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6460 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6461 - mddev
->resync_mark_cnt
;
6463 rt
= max_sectors
- resync
; /* number of remaining sectors */
6464 sector_div(rt
, db
/32+1);
6468 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6469 ((unsigned long)rt
% 60)/6);
6471 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6474 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6476 struct list_head
*tmp
;
6486 spin_lock(&all_mddevs_lock
);
6487 list_for_each(tmp
,&all_mddevs
)
6489 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6491 spin_unlock(&all_mddevs_lock
);
6494 spin_unlock(&all_mddevs_lock
);
6496 return (void*)2;/* tail */
6500 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6502 struct list_head
*tmp
;
6503 mddev_t
*next_mddev
, *mddev
= v
;
6509 spin_lock(&all_mddevs_lock
);
6511 tmp
= all_mddevs
.next
;
6513 tmp
= mddev
->all_mddevs
.next
;
6514 if (tmp
!= &all_mddevs
)
6515 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6517 next_mddev
= (void*)2;
6520 spin_unlock(&all_mddevs_lock
);
6528 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6532 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6536 struct mdstat_info
{
6540 static int md_seq_show(struct seq_file
*seq
, void *v
)
6545 struct mdstat_info
*mi
= seq
->private;
6546 struct bitmap
*bitmap
;
6548 if (v
== (void*)1) {
6549 struct mdk_personality
*pers
;
6550 seq_printf(seq
, "Personalities : ");
6551 spin_lock(&pers_lock
);
6552 list_for_each_entry(pers
, &pers_list
, list
)
6553 seq_printf(seq
, "[%s] ", pers
->name
);
6555 spin_unlock(&pers_lock
);
6556 seq_printf(seq
, "\n");
6557 mi
->event
= atomic_read(&md_event_count
);
6560 if (v
== (void*)2) {
6565 if (mddev_lock(mddev
) < 0)
6568 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6569 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6570 mddev
->pers
? "" : "in");
6573 seq_printf(seq
, " (read-only)");
6575 seq_printf(seq
, " (auto-read-only)");
6576 seq_printf(seq
, " %s", mddev
->pers
->name
);
6580 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6581 char b
[BDEVNAME_SIZE
];
6582 seq_printf(seq
, " %s[%d]",
6583 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6584 if (test_bit(WriteMostly
, &rdev
->flags
))
6585 seq_printf(seq
, "(W)");
6586 if (test_bit(Faulty
, &rdev
->flags
)) {
6587 seq_printf(seq
, "(F)");
6589 } else if (rdev
->raid_disk
< 0)
6590 seq_printf(seq
, "(S)"); /* spare */
6591 sectors
+= rdev
->sectors
;
6594 if (!list_empty(&mddev
->disks
)) {
6596 seq_printf(seq
, "\n %llu blocks",
6597 (unsigned long long)
6598 mddev
->array_sectors
/ 2);
6600 seq_printf(seq
, "\n %llu blocks",
6601 (unsigned long long)sectors
/ 2);
6603 if (mddev
->persistent
) {
6604 if (mddev
->major_version
!= 0 ||
6605 mddev
->minor_version
!= 90) {
6606 seq_printf(seq
," super %d.%d",
6607 mddev
->major_version
,
6608 mddev
->minor_version
);
6610 } else if (mddev
->external
)
6611 seq_printf(seq
, " super external:%s",
6612 mddev
->metadata_type
);
6614 seq_printf(seq
, " super non-persistent");
6617 mddev
->pers
->status(seq
, mddev
);
6618 seq_printf(seq
, "\n ");
6619 if (mddev
->pers
->sync_request
) {
6620 if (mddev
->curr_resync
> 2) {
6621 status_resync(seq
, mddev
);
6622 seq_printf(seq
, "\n ");
6623 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6624 seq_printf(seq
, "\tresync=DELAYED\n ");
6625 else if (mddev
->recovery_cp
< MaxSector
)
6626 seq_printf(seq
, "\tresync=PENDING\n ");
6629 seq_printf(seq
, "\n ");
6631 if ((bitmap
= mddev
->bitmap
)) {
6632 unsigned long chunk_kb
;
6633 unsigned long flags
;
6634 spin_lock_irqsave(&bitmap
->lock
, flags
);
6635 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6636 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6638 bitmap
->pages
- bitmap
->missing_pages
,
6640 (bitmap
->pages
- bitmap
->missing_pages
)
6641 << (PAGE_SHIFT
- 10),
6642 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6643 chunk_kb
? "KB" : "B");
6645 seq_printf(seq
, ", file: ");
6646 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6649 seq_printf(seq
, "\n");
6650 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6653 seq_printf(seq
, "\n");
6655 mddev_unlock(mddev
);
6660 static const struct seq_operations md_seq_ops
= {
6661 .start
= md_seq_start
,
6662 .next
= md_seq_next
,
6663 .stop
= md_seq_stop
,
6664 .show
= md_seq_show
,
6667 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6670 struct mdstat_info
*mi
= kmalloc(sizeof(*mi
), GFP_KERNEL
);
6674 error
= seq_open(file
, &md_seq_ops
);
6678 struct seq_file
*p
= file
->private_data
;
6680 mi
->event
= atomic_read(&md_event_count
);
6685 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6687 struct seq_file
*m
= filp
->private_data
;
6688 struct mdstat_info
*mi
= m
->private;
6691 poll_wait(filp
, &md_event_waiters
, wait
);
6693 /* always allow read */
6694 mask
= POLLIN
| POLLRDNORM
;
6696 if (mi
->event
!= atomic_read(&md_event_count
))
6697 mask
|= POLLERR
| POLLPRI
;
6701 static const struct file_operations md_seq_fops
= {
6702 .owner
= THIS_MODULE
,
6703 .open
= md_seq_open
,
6705 .llseek
= seq_lseek
,
6706 .release
= seq_release_private
,
6707 .poll
= mdstat_poll
,
6710 int register_md_personality(struct mdk_personality
*p
)
6712 spin_lock(&pers_lock
);
6713 list_add_tail(&p
->list
, &pers_list
);
6714 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6715 spin_unlock(&pers_lock
);
6719 int unregister_md_personality(struct mdk_personality
*p
)
6721 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6722 spin_lock(&pers_lock
);
6723 list_del_init(&p
->list
);
6724 spin_unlock(&pers_lock
);
6728 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6736 rdev_for_each_rcu(rdev
, mddev
) {
6737 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6738 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6739 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6740 atomic_read(&disk
->sync_io
);
6741 /* sync IO will cause sync_io to increase before the disk_stats
6742 * as sync_io is counted when a request starts, and
6743 * disk_stats is counted when it completes.
6744 * So resync activity will cause curr_events to be smaller than
6745 * when there was no such activity.
6746 * non-sync IO will cause disk_stat to increase without
6747 * increasing sync_io so curr_events will (eventually)
6748 * be larger than it was before. Once it becomes
6749 * substantially larger, the test below will cause
6750 * the array to appear non-idle, and resync will slow
6752 * If there is a lot of outstanding resync activity when
6753 * we set last_event to curr_events, then all that activity
6754 * completing might cause the array to appear non-idle
6755 * and resync will be slowed down even though there might
6756 * not have been non-resync activity. This will only
6757 * happen once though. 'last_events' will soon reflect
6758 * the state where there is little or no outstanding
6759 * resync requests, and further resync activity will
6760 * always make curr_events less than last_events.
6763 if (init
|| curr_events
- rdev
->last_events
> 64) {
6764 rdev
->last_events
= curr_events
;
6772 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6774 /* another "blocks" (512byte) blocks have been synced */
6775 atomic_sub(blocks
, &mddev
->recovery_active
);
6776 wake_up(&mddev
->recovery_wait
);
6778 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6779 md_wakeup_thread(mddev
->thread
);
6780 // stop recovery, signal do_sync ....
6785 /* md_write_start(mddev, bi)
6786 * If we need to update some array metadata (e.g. 'active' flag
6787 * in superblock) before writing, schedule a superblock update
6788 * and wait for it to complete.
6790 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6793 if (bio_data_dir(bi
) != WRITE
)
6796 BUG_ON(mddev
->ro
== 1);
6797 if (mddev
->ro
== 2) {
6798 /* need to switch to read/write */
6800 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6801 md_wakeup_thread(mddev
->thread
);
6802 md_wakeup_thread(mddev
->sync_thread
);
6805 atomic_inc(&mddev
->writes_pending
);
6806 if (mddev
->safemode
== 1)
6807 mddev
->safemode
= 0;
6808 if (mddev
->in_sync
) {
6809 spin_lock_irq(&mddev
->write_lock
);
6810 if (mddev
->in_sync
) {
6812 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6813 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6814 md_wakeup_thread(mddev
->thread
);
6817 spin_unlock_irq(&mddev
->write_lock
);
6820 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6821 wait_event(mddev
->sb_wait
,
6822 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6825 void md_write_end(mddev_t
*mddev
)
6827 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6828 if (mddev
->safemode
== 2)
6829 md_wakeup_thread(mddev
->thread
);
6830 else if (mddev
->safemode_delay
)
6831 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6835 /* md_allow_write(mddev)
6836 * Calling this ensures that the array is marked 'active' so that writes
6837 * may proceed without blocking. It is important to call this before
6838 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6839 * Must be called with mddev_lock held.
6841 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6842 * is dropped, so return -EAGAIN after notifying userspace.
6844 int md_allow_write(mddev_t
*mddev
)
6850 if (!mddev
->pers
->sync_request
)
6853 spin_lock_irq(&mddev
->write_lock
);
6854 if (mddev
->in_sync
) {
6856 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6857 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6858 if (mddev
->safemode_delay
&&
6859 mddev
->safemode
== 0)
6860 mddev
->safemode
= 1;
6861 spin_unlock_irq(&mddev
->write_lock
);
6862 md_update_sb(mddev
, 0);
6863 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6865 spin_unlock_irq(&mddev
->write_lock
);
6867 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6872 EXPORT_SYMBOL_GPL(md_allow_write
);
6874 #define SYNC_MARKS 10
6875 #define SYNC_MARK_STEP (3*HZ)
6876 void md_do_sync(mddev_t
*mddev
)
6879 unsigned int currspeed
= 0,
6881 sector_t max_sectors
,j
, io_sectors
;
6882 unsigned long mark
[SYNC_MARKS
];
6883 sector_t mark_cnt
[SYNC_MARKS
];
6885 struct list_head
*tmp
;
6886 sector_t last_check
;
6891 /* just incase thread restarts... */
6892 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6894 if (mddev
->ro
) /* never try to sync a read-only array */
6897 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6898 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6899 desc
= "data-check";
6900 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6901 desc
= "requested-resync";
6904 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6909 /* we overload curr_resync somewhat here.
6910 * 0 == not engaged in resync at all
6911 * 2 == checking that there is no conflict with another sync
6912 * 1 == like 2, but have yielded to allow conflicting resync to
6914 * other == active in resync - this many blocks
6916 * Before starting a resync we must have set curr_resync to
6917 * 2, and then checked that every "conflicting" array has curr_resync
6918 * less than ours. When we find one that is the same or higher
6919 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6920 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6921 * This will mean we have to start checking from the beginning again.
6926 mddev
->curr_resync
= 2;
6929 if (kthread_should_stop())
6930 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6932 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
6934 for_each_mddev(mddev2
, tmp
) {
6935 if (mddev2
== mddev
)
6937 if (!mddev
->parallel_resync
6938 && mddev2
->curr_resync
6939 && match_mddev_units(mddev
, mddev2
)) {
6941 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
6942 /* arbitrarily yield */
6943 mddev
->curr_resync
= 1;
6944 wake_up(&resync_wait
);
6946 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
6947 /* no need to wait here, we can wait the next
6948 * time 'round when curr_resync == 2
6951 /* We need to wait 'interruptible' so as not to
6952 * contribute to the load average, and not to
6953 * be caught by 'softlockup'
6955 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
6956 if (!kthread_should_stop() &&
6957 mddev2
->curr_resync
>= mddev
->curr_resync
) {
6958 printk(KERN_INFO
"md: delaying %s of %s"
6959 " until %s has finished (they"
6960 " share one or more physical units)\n",
6961 desc
, mdname(mddev
), mdname(mddev2
));
6963 if (signal_pending(current
))
6964 flush_signals(current
);
6966 finish_wait(&resync_wait
, &wq
);
6969 finish_wait(&resync_wait
, &wq
);
6972 } while (mddev
->curr_resync
< 2);
6975 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6976 /* resync follows the size requested by the personality,
6977 * which defaults to physical size, but can be virtual size
6979 max_sectors
= mddev
->resync_max_sectors
;
6980 mddev
->resync_mismatches
= 0;
6981 /* we don't use the checkpoint if there's a bitmap */
6982 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6983 j
= mddev
->resync_min
;
6984 else if (!mddev
->bitmap
)
6985 j
= mddev
->recovery_cp
;
6987 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6988 max_sectors
= mddev
->dev_sectors
;
6990 /* recovery follows the physical size of devices */
6991 max_sectors
= mddev
->dev_sectors
;
6994 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
6995 if (rdev
->raid_disk
>= 0 &&
6996 !test_bit(Faulty
, &rdev
->flags
) &&
6997 !test_bit(In_sync
, &rdev
->flags
) &&
6998 rdev
->recovery_offset
< j
)
6999 j
= rdev
->recovery_offset
;
7003 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7004 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7005 " %d KB/sec/disk.\n", speed_min(mddev
));
7006 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7007 "(but not more than %d KB/sec) for %s.\n",
7008 speed_max(mddev
), desc
);
7010 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7013 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7015 mark_cnt
[m
] = io_sectors
;
7018 mddev
->resync_mark
= mark
[last_mark
];
7019 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7022 * Tune reconstruction:
7024 window
= 32*(PAGE_SIZE
/512);
7025 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7026 window
/2, (unsigned long long)max_sectors
/2);
7028 atomic_set(&mddev
->recovery_active
, 0);
7033 "md: resuming %s of %s from checkpoint.\n",
7034 desc
, mdname(mddev
));
7035 mddev
->curr_resync
= j
;
7037 mddev
->curr_resync_completed
= j
;
7039 while (j
< max_sectors
) {
7044 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7045 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7046 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7047 > (max_sectors
>> 4)) ||
7048 (j
- mddev
->curr_resync_completed
)*2
7049 >= mddev
->resync_max
- mddev
->curr_resync_completed
7051 /* time to update curr_resync_completed */
7052 wait_event(mddev
->recovery_wait
,
7053 atomic_read(&mddev
->recovery_active
) == 0);
7054 mddev
->curr_resync_completed
= j
;
7055 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7056 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7059 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7060 /* As this condition is controlled by user-space,
7061 * we can block indefinitely, so use '_interruptible'
7062 * to avoid triggering warnings.
7064 flush_signals(current
); /* just in case */
7065 wait_event_interruptible(mddev
->recovery_wait
,
7066 mddev
->resync_max
> j
7067 || kthread_should_stop());
7070 if (kthread_should_stop())
7073 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7074 currspeed
< speed_min(mddev
));
7076 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7080 if (!skipped
) { /* actual IO requested */
7081 io_sectors
+= sectors
;
7082 atomic_add(sectors
, &mddev
->recovery_active
);
7086 if (j
>1) mddev
->curr_resync
= j
;
7087 mddev
->curr_mark_cnt
= io_sectors
;
7088 if (last_check
== 0)
7089 /* this is the earliers that rebuilt will be
7090 * visible in /proc/mdstat
7092 md_new_event(mddev
);
7094 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7097 last_check
= io_sectors
;
7099 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7103 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7105 int next
= (last_mark
+1) % SYNC_MARKS
;
7107 mddev
->resync_mark
= mark
[next
];
7108 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7109 mark
[next
] = jiffies
;
7110 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7115 if (kthread_should_stop())
7120 * this loop exits only if either when we are slower than
7121 * the 'hard' speed limit, or the system was IO-idle for
7123 * the system might be non-idle CPU-wise, but we only care
7124 * about not overloading the IO subsystem. (things like an
7125 * e2fsck being done on the RAID array should execute fast)
7129 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7130 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7132 if (currspeed
> speed_min(mddev
)) {
7133 if ((currspeed
> speed_max(mddev
)) ||
7134 !is_mddev_idle(mddev
, 0)) {
7140 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7142 * this also signals 'finished resyncing' to md_stop
7145 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7147 /* tell personality that we are finished */
7148 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7150 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7151 mddev
->curr_resync
> 2) {
7152 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7153 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7154 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7156 "md: checkpointing %s of %s.\n",
7157 desc
, mdname(mddev
));
7158 mddev
->recovery_cp
= mddev
->curr_resync
;
7161 mddev
->recovery_cp
= MaxSector
;
7163 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7164 mddev
->curr_resync
= MaxSector
;
7166 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7167 if (rdev
->raid_disk
>= 0 &&
7168 mddev
->delta_disks
>= 0 &&
7169 !test_bit(Faulty
, &rdev
->flags
) &&
7170 !test_bit(In_sync
, &rdev
->flags
) &&
7171 rdev
->recovery_offset
< mddev
->curr_resync
)
7172 rdev
->recovery_offset
= mddev
->curr_resync
;
7176 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7179 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7180 /* We completed so min/max setting can be forgotten if used. */
7181 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7182 mddev
->resync_min
= 0;
7183 mddev
->resync_max
= MaxSector
;
7184 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7185 mddev
->resync_min
= mddev
->curr_resync_completed
;
7186 mddev
->curr_resync
= 0;
7187 wake_up(&resync_wait
);
7188 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7189 md_wakeup_thread(mddev
->thread
);
7194 * got a signal, exit.
7197 "md: md_do_sync() got signal ... exiting\n");
7198 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7202 EXPORT_SYMBOL_GPL(md_do_sync
);
7204 static int remove_and_add_spares(mddev_t
*mddev
)
7209 mddev
->curr_resync_completed
= 0;
7211 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7212 if (rdev
->raid_disk
>= 0 &&
7213 !test_bit(Blocked
, &rdev
->flags
) &&
7214 (test_bit(Faulty
, &rdev
->flags
) ||
7215 ! test_bit(In_sync
, &rdev
->flags
)) &&
7216 atomic_read(&rdev
->nr_pending
)==0) {
7217 if (mddev
->pers
->hot_remove_disk(
7218 mddev
, rdev
->raid_disk
)==0) {
7219 sysfs_unlink_rdev(mddev
, rdev
);
7220 rdev
->raid_disk
= -1;
7224 if (mddev
->degraded
) {
7225 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7226 if (rdev
->raid_disk
>= 0 &&
7227 !test_bit(In_sync
, &rdev
->flags
) &&
7228 !test_bit(Faulty
, &rdev
->flags
) &&
7229 !test_bit(Blocked
, &rdev
->flags
))
7231 if (rdev
->raid_disk
< 0
7232 && !test_bit(Faulty
, &rdev
->flags
)) {
7233 rdev
->recovery_offset
= 0;
7235 hot_add_disk(mddev
, rdev
) == 0) {
7236 if (sysfs_link_rdev(mddev
, rdev
))
7237 /* failure here is OK */;
7239 md_new_event(mddev
);
7240 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7249 static void reap_sync_thread(mddev_t
*mddev
)
7253 /* resync has finished, collect result */
7254 md_unregister_thread(mddev
->sync_thread
);
7255 mddev
->sync_thread
= NULL
;
7256 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7257 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7259 /* activate any spares */
7260 if (mddev
->pers
->spare_active(mddev
))
7261 sysfs_notify(&mddev
->kobj
, NULL
,
7264 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7265 mddev
->pers
->finish_reshape
)
7266 mddev
->pers
->finish_reshape(mddev
);
7267 md_update_sb(mddev
, 1);
7269 /* if array is no-longer degraded, then any saved_raid_disk
7270 * information must be scrapped
7272 if (!mddev
->degraded
)
7273 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7274 rdev
->saved_raid_disk
= -1;
7276 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7277 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7278 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7279 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7280 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7281 /* flag recovery needed just to double check */
7282 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7283 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7284 md_new_event(mddev
);
7285 if (mddev
->event_work
.func
)
7286 queue_work(md_misc_wq
, &mddev
->event_work
);
7290 * This routine is regularly called by all per-raid-array threads to
7291 * deal with generic issues like resync and super-block update.
7292 * Raid personalities that don't have a thread (linear/raid0) do not
7293 * need this as they never do any recovery or update the superblock.
7295 * It does not do any resync itself, but rather "forks" off other threads
7296 * to do that as needed.
7297 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7298 * "->recovery" and create a thread at ->sync_thread.
7299 * When the thread finishes it sets MD_RECOVERY_DONE
7300 * and wakeups up this thread which will reap the thread and finish up.
7301 * This thread also removes any faulty devices (with nr_pending == 0).
7303 * The overall approach is:
7304 * 1/ if the superblock needs updating, update it.
7305 * 2/ If a recovery thread is running, don't do anything else.
7306 * 3/ If recovery has finished, clean up, possibly marking spares active.
7307 * 4/ If there are any faulty devices, remove them.
7308 * 5/ If array is degraded, try to add spares devices
7309 * 6/ If array has spares or is not in-sync, start a resync thread.
7311 void md_check_recovery(mddev_t
*mddev
)
7313 if (mddev
->suspended
)
7317 bitmap_daemon_work(mddev
);
7319 if (signal_pending(current
)) {
7320 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7321 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7323 mddev
->safemode
= 2;
7325 flush_signals(current
);
7328 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7331 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7332 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7333 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7334 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7335 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7336 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7340 if (mddev_trylock(mddev
)) {
7344 /* Only thing we do on a ro array is remove
7348 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7349 if (rdev
->raid_disk
>= 0 &&
7350 !test_bit(Blocked
, &rdev
->flags
) &&
7351 test_bit(Faulty
, &rdev
->flags
) &&
7352 atomic_read(&rdev
->nr_pending
)==0) {
7353 if (mddev
->pers
->hot_remove_disk(
7354 mddev
, rdev
->raid_disk
)==0) {
7355 sysfs_unlink_rdev(mddev
, rdev
);
7356 rdev
->raid_disk
= -1;
7359 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7363 if (!mddev
->external
) {
7365 spin_lock_irq(&mddev
->write_lock
);
7366 if (mddev
->safemode
&&
7367 !atomic_read(&mddev
->writes_pending
) &&
7369 mddev
->recovery_cp
== MaxSector
) {
7372 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7374 if (mddev
->safemode
== 1)
7375 mddev
->safemode
= 0;
7376 spin_unlock_irq(&mddev
->write_lock
);
7378 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7382 md_update_sb(mddev
, 0);
7384 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7385 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7386 /* resync/recovery still happening */
7387 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7390 if (mddev
->sync_thread
) {
7391 reap_sync_thread(mddev
);
7394 /* Set RUNNING before clearing NEEDED to avoid
7395 * any transients in the value of "sync_action".
7397 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7398 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7399 /* Clear some bits that don't mean anything, but
7402 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7403 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7405 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7407 /* no recovery is running.
7408 * remove any failed drives, then
7409 * add spares if possible.
7410 * Spare are also removed and re-added, to allow
7411 * the personality to fail the re-add.
7414 if (mddev
->reshape_position
!= MaxSector
) {
7415 if (mddev
->pers
->check_reshape
== NULL
||
7416 mddev
->pers
->check_reshape(mddev
) != 0)
7417 /* Cannot proceed */
7419 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7420 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7421 } else if ((spares
= remove_and_add_spares(mddev
))) {
7422 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7423 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7424 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7425 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7426 } else if (mddev
->recovery_cp
< MaxSector
) {
7427 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7428 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7429 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7430 /* nothing to be done ... */
7433 if (mddev
->pers
->sync_request
) {
7434 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7435 /* We are adding a device or devices to an array
7436 * which has the bitmap stored on all devices.
7437 * So make sure all bitmap pages get written
7439 bitmap_write_all(mddev
->bitmap
);
7441 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7444 if (!mddev
->sync_thread
) {
7445 printk(KERN_ERR
"%s: could not start resync"
7448 /* leave the spares where they are, it shouldn't hurt */
7449 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7450 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7451 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7452 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7453 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7455 md_wakeup_thread(mddev
->sync_thread
);
7456 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7457 md_new_event(mddev
);
7460 if (!mddev
->sync_thread
) {
7461 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7462 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7464 if (mddev
->sysfs_action
)
7465 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7467 mddev_unlock(mddev
);
7471 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7473 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7474 wait_event_timeout(rdev
->blocked_wait
,
7475 !test_bit(Blocked
, &rdev
->flags
),
7476 msecs_to_jiffies(5000));
7477 rdev_dec_pending(rdev
, mddev
);
7479 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7482 /* Bad block management.
7483 * We can record which blocks on each device are 'bad' and so just
7484 * fail those blocks, or that stripe, rather than the whole device.
7485 * Entries in the bad-block table are 64bits wide. This comprises:
7486 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7487 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7488 * A 'shift' can be set so that larger blocks are tracked and
7489 * consequently larger devices can be covered.
7490 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7492 * Locking of the bad-block table uses a seqlock so md_is_badblock
7493 * might need to retry if it is very unlucky.
7494 * We will sometimes want to check for bad blocks in a bi_end_io function,
7495 * so we use the write_seqlock_irq variant.
7497 * When looking for a bad block we specify a range and want to
7498 * know if any block in the range is bad. So we binary-search
7499 * to the last range that starts at-or-before the given endpoint,
7500 * (or "before the sector after the target range")
7501 * then see if it ends after the given start.
7503 * 0 if there are no known bad blocks in the range
7504 * 1 if there are known bad block which are all acknowledged
7505 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7506 * plus the start/length of the first bad section we overlap.
7508 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7509 sector_t
*first_bad
, int *bad_sectors
)
7515 sector_t target
= s
+ sectors
;
7518 if (bb
->shift
> 0) {
7519 /* round the start down, and the end up */
7521 target
+= (1<<bb
->shift
) - 1;
7522 target
>>= bb
->shift
;
7523 sectors
= target
- s
;
7525 /* 'target' is now the first block after the bad range */
7528 seq
= read_seqbegin(&bb
->lock
);
7532 /* Binary search between lo and hi for 'target'
7533 * i.e. for the last range that starts before 'target'
7535 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7536 * are known not to be the last range before target.
7537 * VARIANT: hi-lo is the number of possible
7538 * ranges, and decreases until it reaches 1
7540 while (hi
- lo
> 1) {
7541 int mid
= (lo
+ hi
) / 2;
7542 sector_t a
= BB_OFFSET(p
[mid
]);
7544 /* This could still be the one, earlier ranges
7548 /* This and later ranges are definitely out. */
7551 /* 'lo' might be the last that started before target, but 'hi' isn't */
7553 /* need to check all range that end after 's' to see if
7554 * any are unacknowledged.
7557 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7558 if (BB_OFFSET(p
[lo
]) < target
) {
7559 /* starts before the end, and finishes after
7560 * the start, so they must overlap
7562 if (rv
!= -1 && BB_ACK(p
[lo
]))
7566 *first_bad
= BB_OFFSET(p
[lo
]);
7567 *bad_sectors
= BB_LEN(p
[lo
]);
7573 if (read_seqretry(&bb
->lock
, seq
))
7578 EXPORT_SYMBOL_GPL(md_is_badblock
);
7581 * Add a range of bad blocks to the table.
7582 * This might extend the table, or might contract it
7583 * if two adjacent ranges can be merged.
7584 * We binary-search to find the 'insertion' point, then
7585 * decide how best to handle it.
7587 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7595 /* badblocks are disabled */
7599 /* round the start down, and the end up */
7600 sector_t next
= s
+ sectors
;
7602 next
+= (1<<bb
->shift
) - 1;
7607 write_seqlock_irq(&bb
->lock
);
7612 /* Find the last range that starts at-or-before 's' */
7613 while (hi
- lo
> 1) {
7614 int mid
= (lo
+ hi
) / 2;
7615 sector_t a
= BB_OFFSET(p
[mid
]);
7621 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7625 /* we found a range that might merge with the start
7628 sector_t a
= BB_OFFSET(p
[lo
]);
7629 sector_t e
= a
+ BB_LEN(p
[lo
]);
7630 int ack
= BB_ACK(p
[lo
]);
7632 /* Yes, we can merge with a previous range */
7633 if (s
== a
&& s
+ sectors
>= e
)
7634 /* new range covers old */
7637 ack
= ack
&& acknowledged
;
7639 if (e
< s
+ sectors
)
7641 if (e
- a
<= BB_MAX_LEN
) {
7642 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
7645 /* does not all fit in one range,
7646 * make p[lo] maximal
7648 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
7649 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7655 if (sectors
&& hi
< bb
->count
) {
7656 /* 'hi' points to the first range that starts after 's'.
7657 * Maybe we can merge with the start of that range */
7658 sector_t a
= BB_OFFSET(p
[hi
]);
7659 sector_t e
= a
+ BB_LEN(p
[hi
]);
7660 int ack
= BB_ACK(p
[hi
]);
7661 if (a
<= s
+ sectors
) {
7662 /* merging is possible */
7663 if (e
<= s
+ sectors
) {
7668 ack
= ack
&& acknowledged
;
7671 if (e
- a
<= BB_MAX_LEN
) {
7672 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
7675 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7683 if (sectors
== 0 && hi
< bb
->count
) {
7684 /* we might be able to combine lo and hi */
7685 /* Note: 's' is at the end of 'lo' */
7686 sector_t a
= BB_OFFSET(p
[hi
]);
7687 int lolen
= BB_LEN(p
[lo
]);
7688 int hilen
= BB_LEN(p
[hi
]);
7689 int newlen
= lolen
+ hilen
- (s
- a
);
7690 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
7691 /* yes, we can combine them */
7692 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
7693 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
7694 memmove(p
+ hi
, p
+ hi
+ 1,
7695 (bb
->count
- hi
- 1) * 8);
7700 /* didn't merge (it all).
7701 * Need to add a range just before 'hi' */
7702 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7703 /* No room for more */
7707 int this_sectors
= sectors
;
7708 memmove(p
+ hi
+ 1, p
+ hi
,
7709 (bb
->count
- hi
) * 8);
7712 if (this_sectors
> BB_MAX_LEN
)
7713 this_sectors
= BB_MAX_LEN
;
7714 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
7715 sectors
-= this_sectors
;
7721 write_sequnlock_irq(&bb
->lock
);
7726 int rdev_set_badblocks(mdk_rdev_t
*rdev
, sector_t s
, int sectors
,
7729 int rv
= md_set_badblocks(&rdev
->badblocks
,
7730 s
+ rdev
->data_offset
, sectors
, acknowledged
);
7732 /* Make sure they get written out promptly */
7733 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
7734 md_wakeup_thread(rdev
->mddev
->thread
);
7738 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
7741 * Remove a range of bad blocks from the table.
7742 * This may involve extending the table if we spilt a region,
7743 * but it must not fail. So if the table becomes full, we just
7744 * drop the remove request.
7746 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
7750 sector_t target
= s
+ sectors
;
7753 if (bb
->shift
> 0) {
7754 /* When clearing we round the start up and the end down.
7755 * This should not matter as the shift should align with
7756 * the block size and no rounding should ever be needed.
7757 * However it is better the think a block is bad when it
7758 * isn't than to think a block is not bad when it is.
7760 s
+= (1<<bb
->shift
) - 1;
7762 target
>>= bb
->shift
;
7763 sectors
= target
- s
;
7766 write_seqlock_irq(&bb
->lock
);
7771 /* Find the last range that starts before 'target' */
7772 while (hi
- lo
> 1) {
7773 int mid
= (lo
+ hi
) / 2;
7774 sector_t a
= BB_OFFSET(p
[mid
]);
7781 /* p[lo] is the last range that could overlap the
7782 * current range. Earlier ranges could also overlap,
7783 * but only this one can overlap the end of the range.
7785 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
7786 /* Partial overlap, leave the tail of this range */
7787 int ack
= BB_ACK(p
[lo
]);
7788 sector_t a
= BB_OFFSET(p
[lo
]);
7789 sector_t end
= a
+ BB_LEN(p
[lo
]);
7792 /* we need to split this range */
7793 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7797 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
7799 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
7802 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
7803 /* there is no longer an overlap */
7808 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7809 /* This range does overlap */
7810 if (BB_OFFSET(p
[lo
]) < s
) {
7811 /* Keep the early parts of this range. */
7812 int ack
= BB_ACK(p
[lo
]);
7813 sector_t start
= BB_OFFSET(p
[lo
]);
7814 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
7815 /* now low doesn't overlap, so.. */
7820 /* 'lo' is strictly before, 'hi' is strictly after,
7821 * anything between needs to be discarded
7824 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
7825 bb
->count
-= (hi
- lo
- 1);
7831 write_sequnlock_irq(&bb
->lock
);
7835 int rdev_clear_badblocks(mdk_rdev_t
*rdev
, sector_t s
, int sectors
)
7837 return md_clear_badblocks(&rdev
->badblocks
,
7838 s
+ rdev
->data_offset
,
7841 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
7844 * Acknowledge all bad blocks in a list.
7845 * This only succeeds if ->changed is clear. It is used by
7846 * in-kernel metadata updates
7848 void md_ack_all_badblocks(struct badblocks
*bb
)
7850 if (bb
->page
== NULL
|| bb
->changed
)
7851 /* no point even trying */
7853 write_seqlock_irq(&bb
->lock
);
7855 if (bb
->changed
== 0) {
7858 for (i
= 0; i
< bb
->count
; i
++) {
7859 if (!BB_ACK(p
[i
])) {
7860 sector_t start
= BB_OFFSET(p
[i
]);
7861 int len
= BB_LEN(p
[i
]);
7862 p
[i
] = BB_MAKE(start
, len
, 1);
7866 write_sequnlock_irq(&bb
->lock
);
7868 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
7870 /* sysfs access to bad-blocks list.
7871 * We present two files.
7872 * 'bad-blocks' lists sector numbers and lengths of ranges that
7873 * are recorded as bad. The list is truncated to fit within
7874 * the one-page limit of sysfs.
7875 * Writing "sector length" to this file adds an acknowledged
7877 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7878 * been acknowledged. Writing to this file adds bad blocks
7879 * without acknowledging them. This is largely for testing.
7883 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
7894 seq
= read_seqbegin(&bb
->lock
);
7899 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
7900 sector_t s
= BB_OFFSET(p
[i
]);
7901 unsigned int length
= BB_LEN(p
[i
]);
7902 int ack
= BB_ACK(p
[i
]);
7908 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
7909 (unsigned long long)s
<< bb
->shift
,
7910 length
<< bb
->shift
);
7913 if (read_seqretry(&bb
->lock
, seq
))
7922 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
7924 unsigned long long sector
;
7928 /* Allow clearing via sysfs *only* for testing/debugging.
7929 * Normally only a successful write may clear a badblock
7932 if (page
[0] == '-') {
7936 #endif /* DO_DEBUG */
7938 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
7940 if (newline
!= '\n')
7952 md_clear_badblocks(bb
, sector
, length
);
7955 #endif /* DO_DEBUG */
7956 if (md_set_badblocks(bb
, sector
, length
, !unack
))
7962 static int md_notify_reboot(struct notifier_block
*this,
7963 unsigned long code
, void *x
)
7965 struct list_head
*tmp
;
7968 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
7970 printk(KERN_INFO
"md: stopping all md devices.\n");
7972 for_each_mddev(mddev
, tmp
)
7973 if (mddev_trylock(mddev
)) {
7974 /* Force a switch to readonly even array
7975 * appears to still be in use. Hence
7978 md_set_readonly(mddev
, 100);
7979 mddev_unlock(mddev
);
7982 * certain more exotic SCSI devices are known to be
7983 * volatile wrt too early system reboots. While the
7984 * right place to handle this issue is the given
7985 * driver, we do want to have a safe RAID driver ...
7992 static struct notifier_block md_notifier
= {
7993 .notifier_call
= md_notify_reboot
,
7995 .priority
= INT_MAX
, /* before any real devices */
7998 static void md_geninit(void)
8000 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8002 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8005 static int __init
md_init(void)
8009 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8013 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8017 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8020 if ((ret
= register_blkdev(0, "mdp")) < 0)
8024 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8025 md_probe
, NULL
, NULL
);
8026 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8027 md_probe
, NULL
, NULL
);
8029 register_reboot_notifier(&md_notifier
);
8030 raid_table_header
= register_sysctl_table(raid_root_table
);
8036 unregister_blkdev(MD_MAJOR
, "md");
8038 destroy_workqueue(md_misc_wq
);
8040 destroy_workqueue(md_wq
);
8048 * Searches all registered partitions for autorun RAID arrays
8052 static LIST_HEAD(all_detected_devices
);
8053 struct detected_devices_node
{
8054 struct list_head list
;
8058 void md_autodetect_dev(dev_t dev
)
8060 struct detected_devices_node
*node_detected_dev
;
8062 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8063 if (node_detected_dev
) {
8064 node_detected_dev
->dev
= dev
;
8065 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8067 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8068 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8073 static void autostart_arrays(int part
)
8076 struct detected_devices_node
*node_detected_dev
;
8078 int i_scanned
, i_passed
;
8083 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8085 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8087 node_detected_dev
= list_entry(all_detected_devices
.next
,
8088 struct detected_devices_node
, list
);
8089 list_del(&node_detected_dev
->list
);
8090 dev
= node_detected_dev
->dev
;
8091 kfree(node_detected_dev
);
8092 rdev
= md_import_device(dev
,0, 90);
8096 if (test_bit(Faulty
, &rdev
->flags
)) {
8100 set_bit(AutoDetected
, &rdev
->flags
);
8101 list_add(&rdev
->same_set
, &pending_raid_disks
);
8105 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8106 i_scanned
, i_passed
);
8108 autorun_devices(part
);
8111 #endif /* !MODULE */
8113 static __exit
void md_exit(void)
8116 struct list_head
*tmp
;
8118 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8119 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8121 unregister_blkdev(MD_MAJOR
,"md");
8122 unregister_blkdev(mdp_major
, "mdp");
8123 unregister_reboot_notifier(&md_notifier
);
8124 unregister_sysctl_table(raid_table_header
);
8125 remove_proc_entry("mdstat", NULL
);
8126 for_each_mddev(mddev
, tmp
) {
8127 export_array(mddev
);
8128 mddev
->hold_active
= 0;
8130 destroy_workqueue(md_misc_wq
);
8131 destroy_workqueue(md_wq
);
8134 subsys_initcall(md_init
);
8135 module_exit(md_exit
)
8137 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8139 return sprintf(buffer
, "%d", start_readonly
);
8141 static int set_ro(const char *val
, struct kernel_param
*kp
)
8144 int num
= simple_strtoul(val
, &e
, 10);
8145 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8146 start_readonly
= num
;
8152 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8153 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8155 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8157 EXPORT_SYMBOL(register_md_personality
);
8158 EXPORT_SYMBOL(unregister_md_personality
);
8159 EXPORT_SYMBOL(md_error
);
8160 EXPORT_SYMBOL(md_done_sync
);
8161 EXPORT_SYMBOL(md_write_start
);
8162 EXPORT_SYMBOL(md_write_end
);
8163 EXPORT_SYMBOL(md_register_thread
);
8164 EXPORT_SYMBOL(md_unregister_thread
);
8165 EXPORT_SYMBOL(md_wakeup_thread
);
8166 EXPORT_SYMBOL(md_check_recovery
);
8167 MODULE_LICENSE("GPL");
8168 MODULE_DESCRIPTION("MD RAID framework");
8170 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);