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
);
218 void md_trim_bio(struct bio
*bio
, int offset
, int size
)
220 /* 'bio' is a cloned bio which we need to trim to match
221 * the given offset and size.
222 * This requires adjusting bi_sector, bi_size, and bi_io_vec
225 struct bio_vec
*bvec
;
229 if (offset
== 0 && size
== bio
->bi_size
)
232 bio
->bi_sector
+= offset
;
235 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
237 while (bio
->bi_idx
< bio
->bi_vcnt
&&
238 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
<= offset
) {
239 /* remove this whole bio_vec */
240 offset
-= bio
->bi_io_vec
[bio
->bi_idx
].bv_len
;
243 if (bio
->bi_idx
< bio
->bi_vcnt
) {
244 bio
->bi_io_vec
[bio
->bi_idx
].bv_offset
+= offset
;
245 bio
->bi_io_vec
[bio
->bi_idx
].bv_len
-= offset
;
247 /* avoid any complications with bi_idx being non-zero*/
249 memmove(bio
->bi_io_vec
, bio
->bi_io_vec
+bio
->bi_idx
,
250 (bio
->bi_vcnt
- bio
->bi_idx
) * sizeof(struct bio_vec
));
251 bio
->bi_vcnt
-= bio
->bi_idx
;
254 /* Make sure vcnt and last bv are not too big */
255 bio_for_each_segment(bvec
, bio
, i
) {
256 if (sofar
+ bvec
->bv_len
> size
)
257 bvec
->bv_len
= size
- sofar
;
258 if (bvec
->bv_len
== 0) {
262 sofar
+= bvec
->bv_len
;
265 EXPORT_SYMBOL_GPL(md_trim_bio
);
268 * We have a system wide 'event count' that is incremented
269 * on any 'interesting' event, and readers of /proc/mdstat
270 * can use 'poll' or 'select' to find out when the event
274 * start array, stop array, error, add device, remove device,
275 * start build, activate spare
277 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters
);
278 static atomic_t md_event_count
;
279 void md_new_event(mddev_t
*mddev
)
281 atomic_inc(&md_event_count
);
282 wake_up(&md_event_waiters
);
284 EXPORT_SYMBOL_GPL(md_new_event
);
286 /* Alternate version that can be called from interrupts
287 * when calling sysfs_notify isn't needed.
289 static void md_new_event_inintr(mddev_t
*mddev
)
291 atomic_inc(&md_event_count
);
292 wake_up(&md_event_waiters
);
296 * Enables to iterate over all existing md arrays
297 * all_mddevs_lock protects this list.
299 static LIST_HEAD(all_mddevs
);
300 static DEFINE_SPINLOCK(all_mddevs_lock
);
304 * iterates through all used mddevs in the system.
305 * We take care to grab the all_mddevs_lock whenever navigating
306 * the list, and to always hold a refcount when unlocked.
307 * Any code which breaks out of this loop while own
308 * a reference to the current mddev and must mddev_put it.
310 #define for_each_mddev(mddev,tmp) \
312 for (({ spin_lock(&all_mddevs_lock); \
313 tmp = all_mddevs.next; \
315 ({ if (tmp != &all_mddevs) \
316 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
317 spin_unlock(&all_mddevs_lock); \
318 if (mddev) mddev_put(mddev); \
319 mddev = list_entry(tmp, mddev_t, all_mddevs); \
320 tmp != &all_mddevs;}); \
321 ({ spin_lock(&all_mddevs_lock); \
326 /* Rather than calling directly into the personality make_request function,
327 * IO requests come here first so that we can check if the device is
328 * being suspended pending a reconfiguration.
329 * We hold a refcount over the call to ->make_request. By the time that
330 * call has finished, the bio has been linked into some internal structure
331 * and so is visible to ->quiesce(), so we don't need the refcount any more.
333 static int md_make_request(struct request_queue
*q
, struct bio
*bio
)
335 const int rw
= bio_data_dir(bio
);
336 mddev_t
*mddev
= q
->queuedata
;
339 unsigned int sectors
;
341 if (mddev
== NULL
|| mddev
->pers
== NULL
346 smp_rmb(); /* Ensure implications of 'active' are visible */
348 if (mddev
->suspended
) {
351 prepare_to_wait(&mddev
->sb_wait
, &__wait
,
352 TASK_UNINTERRUPTIBLE
);
353 if (!mddev
->suspended
)
359 finish_wait(&mddev
->sb_wait
, &__wait
);
361 atomic_inc(&mddev
->active_io
);
365 * save the sectors now since our bio can
366 * go away inside make_request
368 sectors
= bio_sectors(bio
);
369 rv
= mddev
->pers
->make_request(mddev
, bio
);
371 cpu
= part_stat_lock();
372 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
373 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
], sectors
);
376 if (atomic_dec_and_test(&mddev
->active_io
) && mddev
->suspended
)
377 wake_up(&mddev
->sb_wait
);
382 /* mddev_suspend makes sure no new requests are submitted
383 * to the device, and that any requests that have been submitted
384 * are completely handled.
385 * Once ->stop is called and completes, the module will be completely
388 void mddev_suspend(mddev_t
*mddev
)
390 BUG_ON(mddev
->suspended
);
391 mddev
->suspended
= 1;
393 wait_event(mddev
->sb_wait
, atomic_read(&mddev
->active_io
) == 0);
394 mddev
->pers
->quiesce(mddev
, 1);
396 EXPORT_SYMBOL_GPL(mddev_suspend
);
398 void mddev_resume(mddev_t
*mddev
)
400 mddev
->suspended
= 0;
401 wake_up(&mddev
->sb_wait
);
402 mddev
->pers
->quiesce(mddev
, 0);
404 md_wakeup_thread(mddev
->thread
);
405 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
407 EXPORT_SYMBOL_GPL(mddev_resume
);
409 int mddev_congested(mddev_t
*mddev
, int bits
)
411 return mddev
->suspended
;
413 EXPORT_SYMBOL(mddev_congested
);
416 * Generic flush handling for md
419 static void md_end_flush(struct bio
*bio
, int err
)
421 mdk_rdev_t
*rdev
= bio
->bi_private
;
422 mddev_t
*mddev
= rdev
->mddev
;
424 rdev_dec_pending(rdev
, mddev
);
426 if (atomic_dec_and_test(&mddev
->flush_pending
)) {
427 /* The pre-request flush has finished */
428 queue_work(md_wq
, &mddev
->flush_work
);
433 static void md_submit_flush_data(struct work_struct
*ws
);
435 static void submit_flushes(struct work_struct
*ws
)
437 mddev_t
*mddev
= container_of(ws
, mddev_t
, flush_work
);
440 INIT_WORK(&mddev
->flush_work
, md_submit_flush_data
);
441 atomic_set(&mddev
->flush_pending
, 1);
443 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
444 if (rdev
->raid_disk
>= 0 &&
445 !test_bit(Faulty
, &rdev
->flags
)) {
446 /* Take two references, one is dropped
447 * when request finishes, one after
448 * we reclaim rcu_read_lock
451 atomic_inc(&rdev
->nr_pending
);
452 atomic_inc(&rdev
->nr_pending
);
454 bi
= bio_alloc_mddev(GFP_KERNEL
, 0, mddev
);
455 bi
->bi_end_io
= md_end_flush
;
456 bi
->bi_private
= rdev
;
457 bi
->bi_bdev
= rdev
->bdev
;
458 atomic_inc(&mddev
->flush_pending
);
459 submit_bio(WRITE_FLUSH
, bi
);
461 rdev_dec_pending(rdev
, mddev
);
464 if (atomic_dec_and_test(&mddev
->flush_pending
))
465 queue_work(md_wq
, &mddev
->flush_work
);
468 static void md_submit_flush_data(struct work_struct
*ws
)
470 mddev_t
*mddev
= container_of(ws
, mddev_t
, flush_work
);
471 struct bio
*bio
= mddev
->flush_bio
;
473 if (bio
->bi_size
== 0)
474 /* an empty barrier - all done */
477 bio
->bi_rw
&= ~REQ_FLUSH
;
478 if (mddev
->pers
->make_request(mddev
, bio
))
479 generic_make_request(bio
);
482 mddev
->flush_bio
= NULL
;
483 wake_up(&mddev
->sb_wait
);
486 void md_flush_request(mddev_t
*mddev
, struct bio
*bio
)
488 spin_lock_irq(&mddev
->write_lock
);
489 wait_event_lock_irq(mddev
->sb_wait
,
491 mddev
->write_lock
, /*nothing*/);
492 mddev
->flush_bio
= bio
;
493 spin_unlock_irq(&mddev
->write_lock
);
495 INIT_WORK(&mddev
->flush_work
, submit_flushes
);
496 queue_work(md_wq
, &mddev
->flush_work
);
498 EXPORT_SYMBOL(md_flush_request
);
500 /* Support for plugging.
501 * This mirrors the plugging support in request_queue, but does not
502 * require having a whole queue or request structures.
503 * We allocate an md_plug_cb for each md device and each thread it gets
504 * plugged on. This links tot the private plug_handle structure in the
505 * personality data where we keep a count of the number of outstanding
506 * plugs so other code can see if a plug is active.
509 struct blk_plug_cb cb
;
513 static void plugger_unplug(struct blk_plug_cb
*cb
)
515 struct md_plug_cb
*mdcb
= container_of(cb
, struct md_plug_cb
, cb
);
516 if (atomic_dec_and_test(&mdcb
->mddev
->plug_cnt
))
517 md_wakeup_thread(mdcb
->mddev
->thread
);
521 /* Check that an unplug wakeup will come shortly.
522 * If not, wakeup the md thread immediately
524 int mddev_check_plugged(mddev_t
*mddev
)
526 struct blk_plug
*plug
= current
->plug
;
527 struct md_plug_cb
*mdcb
;
532 list_for_each_entry(mdcb
, &plug
->cb_list
, cb
.list
) {
533 if (mdcb
->cb
.callback
== plugger_unplug
&&
534 mdcb
->mddev
== mddev
) {
535 /* Already on the list, move to top */
536 if (mdcb
!= list_first_entry(&plug
->cb_list
,
539 list_move(&mdcb
->cb
.list
, &plug
->cb_list
);
543 /* Not currently on the callback list */
544 mdcb
= kmalloc(sizeof(*mdcb
), GFP_ATOMIC
);
549 mdcb
->cb
.callback
= plugger_unplug
;
550 atomic_inc(&mddev
->plug_cnt
);
551 list_add(&mdcb
->cb
.list
, &plug
->cb_list
);
554 EXPORT_SYMBOL_GPL(mddev_check_plugged
);
556 static inline mddev_t
*mddev_get(mddev_t
*mddev
)
558 atomic_inc(&mddev
->active
);
562 static void mddev_delayed_delete(struct work_struct
*ws
);
564 static void mddev_put(mddev_t
*mddev
)
566 struct bio_set
*bs
= NULL
;
568 if (!atomic_dec_and_lock(&mddev
->active
, &all_mddevs_lock
))
570 if (!mddev
->raid_disks
&& list_empty(&mddev
->disks
) &&
571 mddev
->ctime
== 0 && !mddev
->hold_active
) {
572 /* Array is not configured at all, and not held active,
574 list_del(&mddev
->all_mddevs
);
576 mddev
->bio_set
= NULL
;
577 if (mddev
->gendisk
) {
578 /* We did a probe so need to clean up. Call
579 * queue_work inside the spinlock so that
580 * flush_workqueue() after mddev_find will
581 * succeed in waiting for the work to be done.
583 INIT_WORK(&mddev
->del_work
, mddev_delayed_delete
);
584 queue_work(md_misc_wq
, &mddev
->del_work
);
588 spin_unlock(&all_mddevs_lock
);
593 void mddev_init(mddev_t
*mddev
)
595 mutex_init(&mddev
->open_mutex
);
596 mutex_init(&mddev
->reconfig_mutex
);
597 mutex_init(&mddev
->bitmap_info
.mutex
);
598 INIT_LIST_HEAD(&mddev
->disks
);
599 INIT_LIST_HEAD(&mddev
->all_mddevs
);
600 init_timer(&mddev
->safemode_timer
);
601 atomic_set(&mddev
->active
, 1);
602 atomic_set(&mddev
->openers
, 0);
603 atomic_set(&mddev
->active_io
, 0);
604 atomic_set(&mddev
->plug_cnt
, 0);
605 spin_lock_init(&mddev
->write_lock
);
606 atomic_set(&mddev
->flush_pending
, 0);
607 init_waitqueue_head(&mddev
->sb_wait
);
608 init_waitqueue_head(&mddev
->recovery_wait
);
609 mddev
->reshape_position
= MaxSector
;
610 mddev
->resync_min
= 0;
611 mddev
->resync_max
= MaxSector
;
612 mddev
->level
= LEVEL_NONE
;
614 EXPORT_SYMBOL_GPL(mddev_init
);
616 static mddev_t
* mddev_find(dev_t unit
)
618 mddev_t
*mddev
, *new = NULL
;
620 if (unit
&& MAJOR(unit
) != MD_MAJOR
)
621 unit
&= ~((1<<MdpMinorShift
)-1);
624 spin_lock(&all_mddevs_lock
);
627 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
628 if (mddev
->unit
== unit
) {
630 spin_unlock(&all_mddevs_lock
);
636 list_add(&new->all_mddevs
, &all_mddevs
);
637 spin_unlock(&all_mddevs_lock
);
638 new->hold_active
= UNTIL_IOCTL
;
642 /* find an unused unit number */
643 static int next_minor
= 512;
644 int start
= next_minor
;
648 dev
= MKDEV(MD_MAJOR
, next_minor
);
650 if (next_minor
> MINORMASK
)
652 if (next_minor
== start
) {
653 /* Oh dear, all in use. */
654 spin_unlock(&all_mddevs_lock
);
660 list_for_each_entry(mddev
, &all_mddevs
, all_mddevs
)
661 if (mddev
->unit
== dev
) {
667 new->md_minor
= MINOR(dev
);
668 new->hold_active
= UNTIL_STOP
;
669 list_add(&new->all_mddevs
, &all_mddevs
);
670 spin_unlock(&all_mddevs_lock
);
673 spin_unlock(&all_mddevs_lock
);
675 new = kzalloc(sizeof(*new), GFP_KERNEL
);
680 if (MAJOR(unit
) == MD_MAJOR
)
681 new->md_minor
= MINOR(unit
);
683 new->md_minor
= MINOR(unit
) >> MdpMinorShift
;
690 static inline int mddev_lock(mddev_t
* mddev
)
692 return mutex_lock_interruptible(&mddev
->reconfig_mutex
);
695 static inline int mddev_is_locked(mddev_t
*mddev
)
697 return mutex_is_locked(&mddev
->reconfig_mutex
);
700 static inline int mddev_trylock(mddev_t
* mddev
)
702 return mutex_trylock(&mddev
->reconfig_mutex
);
705 static struct attribute_group md_redundancy_group
;
707 static void mddev_unlock(mddev_t
* mddev
)
709 if (mddev
->to_remove
) {
710 /* These cannot be removed under reconfig_mutex as
711 * an access to the files will try to take reconfig_mutex
712 * while holding the file unremovable, which leads to
714 * So hold set sysfs_active while the remove in happeing,
715 * and anything else which might set ->to_remove or my
716 * otherwise change the sysfs namespace will fail with
717 * -EBUSY if sysfs_active is still set.
718 * We set sysfs_active under reconfig_mutex and elsewhere
719 * test it under the same mutex to ensure its correct value
722 struct attribute_group
*to_remove
= mddev
->to_remove
;
723 mddev
->to_remove
= NULL
;
724 mddev
->sysfs_active
= 1;
725 mutex_unlock(&mddev
->reconfig_mutex
);
727 if (mddev
->kobj
.sd
) {
728 if (to_remove
!= &md_redundancy_group
)
729 sysfs_remove_group(&mddev
->kobj
, to_remove
);
730 if (mddev
->pers
== NULL
||
731 mddev
->pers
->sync_request
== NULL
) {
732 sysfs_remove_group(&mddev
->kobj
, &md_redundancy_group
);
733 if (mddev
->sysfs_action
)
734 sysfs_put(mddev
->sysfs_action
);
735 mddev
->sysfs_action
= NULL
;
738 mddev
->sysfs_active
= 0;
740 mutex_unlock(&mddev
->reconfig_mutex
);
742 md_wakeup_thread(mddev
->thread
);
745 static mdk_rdev_t
* find_rdev_nr(mddev_t
*mddev
, int nr
)
749 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
750 if (rdev
->desc_nr
== nr
)
756 static mdk_rdev_t
* find_rdev(mddev_t
* mddev
, dev_t dev
)
760 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
761 if (rdev
->bdev
->bd_dev
== dev
)
767 static struct mdk_personality
*find_pers(int level
, char *clevel
)
769 struct mdk_personality
*pers
;
770 list_for_each_entry(pers
, &pers_list
, list
) {
771 if (level
!= LEVEL_NONE
&& pers
->level
== level
)
773 if (strcmp(pers
->name
, clevel
)==0)
779 /* return the offset of the super block in 512byte sectors */
780 static inline sector_t
calc_dev_sboffset(mdk_rdev_t
*rdev
)
782 sector_t num_sectors
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
783 return MD_NEW_SIZE_SECTORS(num_sectors
);
786 static int alloc_disk_sb(mdk_rdev_t
* rdev
)
791 rdev
->sb_page
= alloc_page(GFP_KERNEL
);
792 if (!rdev
->sb_page
) {
793 printk(KERN_ALERT
"md: out of memory.\n");
800 static void free_disk_sb(mdk_rdev_t
* rdev
)
803 put_page(rdev
->sb_page
);
805 rdev
->sb_page
= NULL
;
810 put_page(rdev
->bb_page
);
811 rdev
->bb_page
= NULL
;
816 static void super_written(struct bio
*bio
, int error
)
818 mdk_rdev_t
*rdev
= bio
->bi_private
;
819 mddev_t
*mddev
= rdev
->mddev
;
821 if (error
|| !test_bit(BIO_UPTODATE
, &bio
->bi_flags
)) {
822 printk("md: super_written gets error=%d, uptodate=%d\n",
823 error
, test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
824 WARN_ON(test_bit(BIO_UPTODATE
, &bio
->bi_flags
));
825 md_error(mddev
, rdev
);
828 if (atomic_dec_and_test(&mddev
->pending_writes
))
829 wake_up(&mddev
->sb_wait
);
833 void md_super_write(mddev_t
*mddev
, mdk_rdev_t
*rdev
,
834 sector_t sector
, int size
, struct page
*page
)
836 /* write first size bytes of page to sector of rdev
837 * Increment mddev->pending_writes before returning
838 * and decrement it on completion, waking up sb_wait
839 * if zero is reached.
840 * If an error occurred, call md_error
842 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, mddev
);
844 bio
->bi_bdev
= rdev
->meta_bdev
? rdev
->meta_bdev
: rdev
->bdev
;
845 bio
->bi_sector
= sector
;
846 bio_add_page(bio
, page
, size
, 0);
847 bio
->bi_private
= rdev
;
848 bio
->bi_end_io
= super_written
;
850 atomic_inc(&mddev
->pending_writes
);
851 submit_bio(REQ_WRITE
| REQ_SYNC
| REQ_FLUSH
| REQ_FUA
, bio
);
854 void md_super_wait(mddev_t
*mddev
)
856 /* wait for all superblock writes that were scheduled to complete */
859 prepare_to_wait(&mddev
->sb_wait
, &wq
, TASK_UNINTERRUPTIBLE
);
860 if (atomic_read(&mddev
->pending_writes
)==0)
864 finish_wait(&mddev
->sb_wait
, &wq
);
867 static void bi_complete(struct bio
*bio
, int error
)
869 complete((struct completion
*)bio
->bi_private
);
872 int sync_page_io(mdk_rdev_t
*rdev
, sector_t sector
, int size
,
873 struct page
*page
, int rw
, bool metadata_op
)
875 struct bio
*bio
= bio_alloc_mddev(GFP_NOIO
, 1, rdev
->mddev
);
876 struct completion event
;
881 bio
->bi_bdev
= (metadata_op
&& rdev
->meta_bdev
) ?
882 rdev
->meta_bdev
: rdev
->bdev
;
884 bio
->bi_sector
= sector
+ rdev
->sb_start
;
886 bio
->bi_sector
= sector
+ rdev
->data_offset
;
887 bio_add_page(bio
, page
, size
, 0);
888 init_completion(&event
);
889 bio
->bi_private
= &event
;
890 bio
->bi_end_io
= bi_complete
;
892 wait_for_completion(&event
);
894 ret
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
898 EXPORT_SYMBOL_GPL(sync_page_io
);
900 static int read_disk_sb(mdk_rdev_t
* rdev
, int size
)
902 char b
[BDEVNAME_SIZE
];
903 if (!rdev
->sb_page
) {
911 if (!sync_page_io(rdev
, 0, size
, rdev
->sb_page
, READ
, true))
917 printk(KERN_WARNING
"md: disabled device %s, could not read superblock.\n",
918 bdevname(rdev
->bdev
,b
));
922 static int uuid_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
924 return sb1
->set_uuid0
== sb2
->set_uuid0
&&
925 sb1
->set_uuid1
== sb2
->set_uuid1
&&
926 sb1
->set_uuid2
== sb2
->set_uuid2
&&
927 sb1
->set_uuid3
== sb2
->set_uuid3
;
930 static int sb_equal(mdp_super_t
*sb1
, mdp_super_t
*sb2
)
933 mdp_super_t
*tmp1
, *tmp2
;
935 tmp1
= kmalloc(sizeof(*tmp1
),GFP_KERNEL
);
936 tmp2
= kmalloc(sizeof(*tmp2
),GFP_KERNEL
);
938 if (!tmp1
|| !tmp2
) {
940 printk(KERN_INFO
"md.c sb_equal(): failed to allocate memory!\n");
948 * nr_disks is not constant
953 ret
= (memcmp(tmp1
, tmp2
, MD_SB_GENERIC_CONSTANT_WORDS
* 4) == 0);
961 static u32
md_csum_fold(u32 csum
)
963 csum
= (csum
& 0xffff) + (csum
>> 16);
964 return (csum
& 0xffff) + (csum
>> 16);
967 static unsigned int calc_sb_csum(mdp_super_t
* sb
)
970 u32
*sb32
= (u32
*)sb
;
972 unsigned int disk_csum
, csum
;
974 disk_csum
= sb
->sb_csum
;
977 for (i
= 0; i
< MD_SB_BYTES
/4 ; i
++)
979 csum
= (newcsum
& 0xffffffff) + (newcsum
>>32);
983 /* This used to use csum_partial, which was wrong for several
984 * reasons including that different results are returned on
985 * different architectures. It isn't critical that we get exactly
986 * the same return value as before (we always csum_fold before
987 * testing, and that removes any differences). However as we
988 * know that csum_partial always returned a 16bit value on
989 * alphas, do a fold to maximise conformity to previous behaviour.
991 sb
->sb_csum
= md_csum_fold(disk_csum
);
993 sb
->sb_csum
= disk_csum
;
1000 * Handle superblock details.
1001 * We want to be able to handle multiple superblock formats
1002 * so we have a common interface to them all, and an array of
1003 * different handlers.
1004 * We rely on user-space to write the initial superblock, and support
1005 * reading and updating of superblocks.
1006 * Interface methods are:
1007 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
1008 * loads and validates a superblock on dev.
1009 * if refdev != NULL, compare superblocks on both devices
1011 * 0 - dev has a superblock that is compatible with refdev
1012 * 1 - dev has a superblock that is compatible and newer than refdev
1013 * so dev should be used as the refdev in future
1014 * -EINVAL superblock incompatible or invalid
1015 * -othererror e.g. -EIO
1017 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
1018 * Verify that dev is acceptable into mddev.
1019 * The first time, mddev->raid_disks will be 0, and data from
1020 * dev should be merged in. Subsequent calls check that dev
1021 * is new enough. Return 0 or -EINVAL
1023 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
1024 * Update the superblock for rdev with data in mddev
1025 * This does not write to disc.
1031 struct module
*owner
;
1032 int (*load_super
)(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
,
1034 int (*validate_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
1035 void (*sync_super
)(mddev_t
*mddev
, mdk_rdev_t
*rdev
);
1036 unsigned long long (*rdev_size_change
)(mdk_rdev_t
*rdev
,
1037 sector_t num_sectors
);
1041 * Check that the given mddev has no bitmap.
1043 * This function is called from the run method of all personalities that do not
1044 * support bitmaps. It prints an error message and returns non-zero if mddev
1045 * has a bitmap. Otherwise, it returns 0.
1048 int md_check_no_bitmap(mddev_t
*mddev
)
1050 if (!mddev
->bitmap_info
.file
&& !mddev
->bitmap_info
.offset
)
1052 printk(KERN_ERR
"%s: bitmaps are not supported for %s\n",
1053 mdname(mddev
), mddev
->pers
->name
);
1056 EXPORT_SYMBOL(md_check_no_bitmap
);
1059 * load_super for 0.90.0
1061 static int super_90_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1063 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1068 * Calculate the position of the superblock (512byte sectors),
1069 * it's at the end of the disk.
1071 * It also happens to be a multiple of 4Kb.
1073 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1075 ret
= read_disk_sb(rdev
, MD_SB_BYTES
);
1076 if (ret
) return ret
;
1080 bdevname(rdev
->bdev
, b
);
1081 sb
= page_address(rdev
->sb_page
);
1083 if (sb
->md_magic
!= MD_SB_MAGIC
) {
1084 printk(KERN_ERR
"md: invalid raid superblock magic on %s\n",
1089 if (sb
->major_version
!= 0 ||
1090 sb
->minor_version
< 90 ||
1091 sb
->minor_version
> 91) {
1092 printk(KERN_WARNING
"Bad version number %d.%d on %s\n",
1093 sb
->major_version
, sb
->minor_version
,
1098 if (sb
->raid_disks
<= 0)
1101 if (md_csum_fold(calc_sb_csum(sb
)) != md_csum_fold(sb
->sb_csum
)) {
1102 printk(KERN_WARNING
"md: invalid superblock checksum on %s\n",
1107 rdev
->preferred_minor
= sb
->md_minor
;
1108 rdev
->data_offset
= 0;
1109 rdev
->sb_size
= MD_SB_BYTES
;
1110 rdev
->badblocks
.shift
= -1;
1112 if (sb
->level
== LEVEL_MULTIPATH
)
1115 rdev
->desc_nr
= sb
->this_disk
.number
;
1121 mdp_super_t
*refsb
= page_address(refdev
->sb_page
);
1122 if (!uuid_equal(refsb
, sb
)) {
1123 printk(KERN_WARNING
"md: %s has different UUID to %s\n",
1124 b
, bdevname(refdev
->bdev
,b2
));
1127 if (!sb_equal(refsb
, sb
)) {
1128 printk(KERN_WARNING
"md: %s has same UUID"
1129 " but different superblock to %s\n",
1130 b
, bdevname(refdev
->bdev
, b2
));
1134 ev2
= md_event(refsb
);
1140 rdev
->sectors
= rdev
->sb_start
;
1142 if (rdev
->sectors
< sb
->size
* 2 && sb
->level
> 1)
1143 /* "this cannot possibly happen" ... */
1151 * validate_super for 0.90.0
1153 static int super_90_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1156 mdp_super_t
*sb
= page_address(rdev
->sb_page
);
1157 __u64 ev1
= md_event(sb
);
1159 rdev
->raid_disk
= -1;
1160 clear_bit(Faulty
, &rdev
->flags
);
1161 clear_bit(In_sync
, &rdev
->flags
);
1162 clear_bit(WriteMostly
, &rdev
->flags
);
1164 if (mddev
->raid_disks
== 0) {
1165 mddev
->major_version
= 0;
1166 mddev
->minor_version
= sb
->minor_version
;
1167 mddev
->patch_version
= sb
->patch_version
;
1168 mddev
->external
= 0;
1169 mddev
->chunk_sectors
= sb
->chunk_size
>> 9;
1170 mddev
->ctime
= sb
->ctime
;
1171 mddev
->utime
= sb
->utime
;
1172 mddev
->level
= sb
->level
;
1173 mddev
->clevel
[0] = 0;
1174 mddev
->layout
= sb
->layout
;
1175 mddev
->raid_disks
= sb
->raid_disks
;
1176 mddev
->dev_sectors
= sb
->size
* 2;
1177 mddev
->events
= ev1
;
1178 mddev
->bitmap_info
.offset
= 0;
1179 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
1181 if (mddev
->minor_version
>= 91) {
1182 mddev
->reshape_position
= sb
->reshape_position
;
1183 mddev
->delta_disks
= sb
->delta_disks
;
1184 mddev
->new_level
= sb
->new_level
;
1185 mddev
->new_layout
= sb
->new_layout
;
1186 mddev
->new_chunk_sectors
= sb
->new_chunk
>> 9;
1188 mddev
->reshape_position
= MaxSector
;
1189 mddev
->delta_disks
= 0;
1190 mddev
->new_level
= mddev
->level
;
1191 mddev
->new_layout
= mddev
->layout
;
1192 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1195 if (sb
->state
& (1<<MD_SB_CLEAN
))
1196 mddev
->recovery_cp
= MaxSector
;
1198 if (sb
->events_hi
== sb
->cp_events_hi
&&
1199 sb
->events_lo
== sb
->cp_events_lo
) {
1200 mddev
->recovery_cp
= sb
->recovery_cp
;
1202 mddev
->recovery_cp
= 0;
1205 memcpy(mddev
->uuid
+0, &sb
->set_uuid0
, 4);
1206 memcpy(mddev
->uuid
+4, &sb
->set_uuid1
, 4);
1207 memcpy(mddev
->uuid
+8, &sb
->set_uuid2
, 4);
1208 memcpy(mddev
->uuid
+12,&sb
->set_uuid3
, 4);
1210 mddev
->max_disks
= MD_SB_DISKS
;
1212 if (sb
->state
& (1<<MD_SB_BITMAP_PRESENT
) &&
1213 mddev
->bitmap_info
.file
== NULL
)
1214 mddev
->bitmap_info
.offset
=
1215 mddev
->bitmap_info
.default_offset
;
1217 } else if (mddev
->pers
== NULL
) {
1218 /* Insist on good event counter while assembling, except
1219 * for spares (which don't need an event count) */
1221 if (sb
->disks
[rdev
->desc_nr
].state
& (
1222 (1<<MD_DISK_SYNC
) | (1 << MD_DISK_ACTIVE
)))
1223 if (ev1
< mddev
->events
)
1225 } else if (mddev
->bitmap
) {
1226 /* if adding to array with a bitmap, then we can accept an
1227 * older device ... but not too old.
1229 if (ev1
< mddev
->bitmap
->events_cleared
)
1232 if (ev1
< mddev
->events
)
1233 /* just a hot-add of a new device, leave raid_disk at -1 */
1237 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1238 desc
= sb
->disks
+ rdev
->desc_nr
;
1240 if (desc
->state
& (1<<MD_DISK_FAULTY
))
1241 set_bit(Faulty
, &rdev
->flags
);
1242 else if (desc
->state
& (1<<MD_DISK_SYNC
) /* &&
1243 desc->raid_disk < mddev->raid_disks */) {
1244 set_bit(In_sync
, &rdev
->flags
);
1245 rdev
->raid_disk
= desc
->raid_disk
;
1246 } else if (desc
->state
& (1<<MD_DISK_ACTIVE
)) {
1247 /* active but not in sync implies recovery up to
1248 * reshape position. We don't know exactly where
1249 * that is, so set to zero for now */
1250 if (mddev
->minor_version
>= 91) {
1251 rdev
->recovery_offset
= 0;
1252 rdev
->raid_disk
= desc
->raid_disk
;
1255 if (desc
->state
& (1<<MD_DISK_WRITEMOSTLY
))
1256 set_bit(WriteMostly
, &rdev
->flags
);
1257 } else /* MULTIPATH are always insync */
1258 set_bit(In_sync
, &rdev
->flags
);
1263 * sync_super for 0.90.0
1265 static void super_90_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1269 int next_spare
= mddev
->raid_disks
;
1272 /* make rdev->sb match mddev data..
1275 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1276 * 3/ any empty disks < next_spare become removed
1278 * disks[0] gets initialised to REMOVED because
1279 * we cannot be sure from other fields if it has
1280 * been initialised or not.
1283 int active
=0, working
=0,failed
=0,spare
=0,nr_disks
=0;
1285 rdev
->sb_size
= MD_SB_BYTES
;
1287 sb
= page_address(rdev
->sb_page
);
1289 memset(sb
, 0, sizeof(*sb
));
1291 sb
->md_magic
= MD_SB_MAGIC
;
1292 sb
->major_version
= mddev
->major_version
;
1293 sb
->patch_version
= mddev
->patch_version
;
1294 sb
->gvalid_words
= 0; /* ignored */
1295 memcpy(&sb
->set_uuid0
, mddev
->uuid
+0, 4);
1296 memcpy(&sb
->set_uuid1
, mddev
->uuid
+4, 4);
1297 memcpy(&sb
->set_uuid2
, mddev
->uuid
+8, 4);
1298 memcpy(&sb
->set_uuid3
, mddev
->uuid
+12,4);
1300 sb
->ctime
= mddev
->ctime
;
1301 sb
->level
= mddev
->level
;
1302 sb
->size
= mddev
->dev_sectors
/ 2;
1303 sb
->raid_disks
= mddev
->raid_disks
;
1304 sb
->md_minor
= mddev
->md_minor
;
1305 sb
->not_persistent
= 0;
1306 sb
->utime
= mddev
->utime
;
1308 sb
->events_hi
= (mddev
->events
>>32);
1309 sb
->events_lo
= (u32
)mddev
->events
;
1311 if (mddev
->reshape_position
== MaxSector
)
1312 sb
->minor_version
= 90;
1314 sb
->minor_version
= 91;
1315 sb
->reshape_position
= mddev
->reshape_position
;
1316 sb
->new_level
= mddev
->new_level
;
1317 sb
->delta_disks
= mddev
->delta_disks
;
1318 sb
->new_layout
= mddev
->new_layout
;
1319 sb
->new_chunk
= mddev
->new_chunk_sectors
<< 9;
1321 mddev
->minor_version
= sb
->minor_version
;
1324 sb
->recovery_cp
= mddev
->recovery_cp
;
1325 sb
->cp_events_hi
= (mddev
->events
>>32);
1326 sb
->cp_events_lo
= (u32
)mddev
->events
;
1327 if (mddev
->recovery_cp
== MaxSector
)
1328 sb
->state
= (1<< MD_SB_CLEAN
);
1330 sb
->recovery_cp
= 0;
1332 sb
->layout
= mddev
->layout
;
1333 sb
->chunk_size
= mddev
->chunk_sectors
<< 9;
1335 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
)
1336 sb
->state
|= (1<<MD_SB_BITMAP_PRESENT
);
1338 sb
->disks
[0].state
= (1<<MD_DISK_REMOVED
);
1339 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1342 int is_active
= test_bit(In_sync
, &rdev2
->flags
);
1344 if (rdev2
->raid_disk
>= 0 &&
1345 sb
->minor_version
>= 91)
1346 /* we have nowhere to store the recovery_offset,
1347 * but if it is not below the reshape_position,
1348 * we can piggy-back on that.
1351 if (rdev2
->raid_disk
< 0 ||
1352 test_bit(Faulty
, &rdev2
->flags
))
1355 desc_nr
= rdev2
->raid_disk
;
1357 desc_nr
= next_spare
++;
1358 rdev2
->desc_nr
= desc_nr
;
1359 d
= &sb
->disks
[rdev2
->desc_nr
];
1361 d
->number
= rdev2
->desc_nr
;
1362 d
->major
= MAJOR(rdev2
->bdev
->bd_dev
);
1363 d
->minor
= MINOR(rdev2
->bdev
->bd_dev
);
1365 d
->raid_disk
= rdev2
->raid_disk
;
1367 d
->raid_disk
= rdev2
->desc_nr
; /* compatibility */
1368 if (test_bit(Faulty
, &rdev2
->flags
))
1369 d
->state
= (1<<MD_DISK_FAULTY
);
1370 else if (is_active
) {
1371 d
->state
= (1<<MD_DISK_ACTIVE
);
1372 if (test_bit(In_sync
, &rdev2
->flags
))
1373 d
->state
|= (1<<MD_DISK_SYNC
);
1381 if (test_bit(WriteMostly
, &rdev2
->flags
))
1382 d
->state
|= (1<<MD_DISK_WRITEMOSTLY
);
1384 /* now set the "removed" and "faulty" bits on any missing devices */
1385 for (i
=0 ; i
< mddev
->raid_disks
; i
++) {
1386 mdp_disk_t
*d
= &sb
->disks
[i
];
1387 if (d
->state
== 0 && d
->number
== 0) {
1390 d
->state
= (1<<MD_DISK_REMOVED
);
1391 d
->state
|= (1<<MD_DISK_FAULTY
);
1395 sb
->nr_disks
= nr_disks
;
1396 sb
->active_disks
= active
;
1397 sb
->working_disks
= working
;
1398 sb
->failed_disks
= failed
;
1399 sb
->spare_disks
= spare
;
1401 sb
->this_disk
= sb
->disks
[rdev
->desc_nr
];
1402 sb
->sb_csum
= calc_sb_csum(sb
);
1406 * rdev_size_change for 0.90.0
1408 static unsigned long long
1409 super_90_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1411 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1412 return 0; /* component must fit device */
1413 if (rdev
->mddev
->bitmap_info
.offset
)
1414 return 0; /* can't move bitmap */
1415 rdev
->sb_start
= calc_dev_sboffset(rdev
);
1416 if (!num_sectors
|| num_sectors
> rdev
->sb_start
)
1417 num_sectors
= rdev
->sb_start
;
1418 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1420 md_super_wait(rdev
->mddev
);
1426 * version 1 superblock
1429 static __le32
calc_sb_1_csum(struct mdp_superblock_1
* sb
)
1433 unsigned long long newcsum
;
1434 int size
= 256 + le32_to_cpu(sb
->max_dev
)*2;
1435 __le32
*isuper
= (__le32
*)sb
;
1438 disk_csum
= sb
->sb_csum
;
1441 for (i
=0; size
>=4; size
-= 4 )
1442 newcsum
+= le32_to_cpu(*isuper
++);
1445 newcsum
+= le16_to_cpu(*(__le16
*) isuper
);
1447 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
1448 sb
->sb_csum
= disk_csum
;
1449 return cpu_to_le32(csum
);
1452 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
1454 static int super_1_load(mdk_rdev_t
*rdev
, mdk_rdev_t
*refdev
, int minor_version
)
1456 struct mdp_superblock_1
*sb
;
1459 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
1463 * Calculate the position of the superblock in 512byte sectors.
1464 * It is always aligned to a 4K boundary and
1465 * depeding on minor_version, it can be:
1466 * 0: At least 8K, but less than 12K, from end of device
1467 * 1: At start of device
1468 * 2: 4K from start of device.
1470 switch(minor_version
) {
1472 sb_start
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1474 sb_start
&= ~(sector_t
)(4*2-1);
1485 rdev
->sb_start
= sb_start
;
1487 /* superblock is rarely larger than 1K, but it can be larger,
1488 * and it is safe to read 4k, so we do that
1490 ret
= read_disk_sb(rdev
, 4096);
1491 if (ret
) return ret
;
1494 sb
= page_address(rdev
->sb_page
);
1496 if (sb
->magic
!= cpu_to_le32(MD_SB_MAGIC
) ||
1497 sb
->major_version
!= cpu_to_le32(1) ||
1498 le32_to_cpu(sb
->max_dev
) > (4096-256)/2 ||
1499 le64_to_cpu(sb
->super_offset
) != rdev
->sb_start
||
1500 (le32_to_cpu(sb
->feature_map
) & ~MD_FEATURE_ALL
) != 0)
1503 if (calc_sb_1_csum(sb
) != sb
->sb_csum
) {
1504 printk("md: invalid superblock checksum on %s\n",
1505 bdevname(rdev
->bdev
,b
));
1508 if (le64_to_cpu(sb
->data_size
) < 10) {
1509 printk("md: data_size too small on %s\n",
1510 bdevname(rdev
->bdev
,b
));
1514 rdev
->preferred_minor
= 0xffff;
1515 rdev
->data_offset
= le64_to_cpu(sb
->data_offset
);
1516 atomic_set(&rdev
->corrected_errors
, le32_to_cpu(sb
->cnt_corrected_read
));
1518 rdev
->sb_size
= le32_to_cpu(sb
->max_dev
) * 2 + 256;
1519 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1520 if (rdev
->sb_size
& bmask
)
1521 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1524 && rdev
->data_offset
< sb_start
+ (rdev
->sb_size
/512))
1527 if (sb
->level
== cpu_to_le32(LEVEL_MULTIPATH
))
1530 rdev
->desc_nr
= le32_to_cpu(sb
->dev_number
);
1532 if (!rdev
->bb_page
) {
1533 rdev
->bb_page
= alloc_page(GFP_KERNEL
);
1537 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BAD_BLOCKS
) &&
1538 rdev
->badblocks
.count
== 0) {
1539 /* need to load the bad block list.
1540 * Currently we limit it to one page.
1546 int sectors
= le16_to_cpu(sb
->bblog_size
);
1547 if (sectors
> (PAGE_SIZE
/ 512))
1549 offset
= le32_to_cpu(sb
->bblog_offset
);
1552 bb_sector
= (long long)offset
;
1553 if (!sync_page_io(rdev
, bb_sector
, sectors
<< 9,
1554 rdev
->bb_page
, READ
, true))
1556 bbp
= (u64
*)page_address(rdev
->bb_page
);
1557 rdev
->badblocks
.shift
= sb
->bblog_shift
;
1558 for (i
= 0 ; i
< (sectors
<< (9-3)) ; i
++, bbp
++) {
1559 u64 bb
= le64_to_cpu(*bbp
);
1560 int count
= bb
& (0x3ff);
1561 u64 sector
= bb
>> 10;
1562 sector
<<= sb
->bblog_shift
;
1563 count
<<= sb
->bblog_shift
;
1566 if (md_set_badblocks(&rdev
->badblocks
,
1567 sector
, count
, 1) == 0)
1570 } else if (sb
->bblog_offset
== 0)
1571 rdev
->badblocks
.shift
= -1;
1577 struct mdp_superblock_1
*refsb
= page_address(refdev
->sb_page
);
1579 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16) != 0 ||
1580 sb
->level
!= refsb
->level
||
1581 sb
->layout
!= refsb
->layout
||
1582 sb
->chunksize
!= refsb
->chunksize
) {
1583 printk(KERN_WARNING
"md: %s has strangely different"
1584 " superblock to %s\n",
1585 bdevname(rdev
->bdev
,b
),
1586 bdevname(refdev
->bdev
,b2
));
1589 ev1
= le64_to_cpu(sb
->events
);
1590 ev2
= le64_to_cpu(refsb
->events
);
1598 rdev
->sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
1599 le64_to_cpu(sb
->data_offset
);
1601 rdev
->sectors
= rdev
->sb_start
;
1602 if (rdev
->sectors
< le64_to_cpu(sb
->data_size
))
1604 rdev
->sectors
= le64_to_cpu(sb
->data_size
);
1605 if (le64_to_cpu(sb
->size
) > rdev
->sectors
)
1610 static int super_1_validate(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1612 struct mdp_superblock_1
*sb
= page_address(rdev
->sb_page
);
1613 __u64 ev1
= le64_to_cpu(sb
->events
);
1615 rdev
->raid_disk
= -1;
1616 clear_bit(Faulty
, &rdev
->flags
);
1617 clear_bit(In_sync
, &rdev
->flags
);
1618 clear_bit(WriteMostly
, &rdev
->flags
);
1620 if (mddev
->raid_disks
== 0) {
1621 mddev
->major_version
= 1;
1622 mddev
->patch_version
= 0;
1623 mddev
->external
= 0;
1624 mddev
->chunk_sectors
= le32_to_cpu(sb
->chunksize
);
1625 mddev
->ctime
= le64_to_cpu(sb
->ctime
) & ((1ULL << 32)-1);
1626 mddev
->utime
= le64_to_cpu(sb
->utime
) & ((1ULL << 32)-1);
1627 mddev
->level
= le32_to_cpu(sb
->level
);
1628 mddev
->clevel
[0] = 0;
1629 mddev
->layout
= le32_to_cpu(sb
->layout
);
1630 mddev
->raid_disks
= le32_to_cpu(sb
->raid_disks
);
1631 mddev
->dev_sectors
= le64_to_cpu(sb
->size
);
1632 mddev
->events
= ev1
;
1633 mddev
->bitmap_info
.offset
= 0;
1634 mddev
->bitmap_info
.default_offset
= 1024 >> 9;
1636 mddev
->recovery_cp
= le64_to_cpu(sb
->resync_offset
);
1637 memcpy(mddev
->uuid
, sb
->set_uuid
, 16);
1639 mddev
->max_disks
= (4096-256)/2;
1641 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
) &&
1642 mddev
->bitmap_info
.file
== NULL
)
1643 mddev
->bitmap_info
.offset
=
1644 (__s32
)le32_to_cpu(sb
->bitmap_offset
);
1646 if ((le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RESHAPE_ACTIVE
)) {
1647 mddev
->reshape_position
= le64_to_cpu(sb
->reshape_position
);
1648 mddev
->delta_disks
= le32_to_cpu(sb
->delta_disks
);
1649 mddev
->new_level
= le32_to_cpu(sb
->new_level
);
1650 mddev
->new_layout
= le32_to_cpu(sb
->new_layout
);
1651 mddev
->new_chunk_sectors
= le32_to_cpu(sb
->new_chunk
);
1653 mddev
->reshape_position
= MaxSector
;
1654 mddev
->delta_disks
= 0;
1655 mddev
->new_level
= mddev
->level
;
1656 mddev
->new_layout
= mddev
->layout
;
1657 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
1660 } else if (mddev
->pers
== NULL
) {
1661 /* Insist of good event counter while assembling, except for
1662 * spares (which don't need an event count) */
1664 if (rdev
->desc_nr
>= 0 &&
1665 rdev
->desc_nr
< le32_to_cpu(sb
->max_dev
) &&
1666 le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]) < 0xfffe)
1667 if (ev1
< mddev
->events
)
1669 } else if (mddev
->bitmap
) {
1670 /* If adding to array with a bitmap, then we can accept an
1671 * older device, but not too old.
1673 if (ev1
< mddev
->bitmap
->events_cleared
)
1676 if (ev1
< mddev
->events
)
1677 /* just a hot-add of a new device, leave raid_disk at -1 */
1680 if (mddev
->level
!= LEVEL_MULTIPATH
) {
1682 if (rdev
->desc_nr
< 0 ||
1683 rdev
->desc_nr
>= le32_to_cpu(sb
->max_dev
)) {
1687 role
= le16_to_cpu(sb
->dev_roles
[rdev
->desc_nr
]);
1689 case 0xffff: /* spare */
1691 case 0xfffe: /* faulty */
1692 set_bit(Faulty
, &rdev
->flags
);
1695 if ((le32_to_cpu(sb
->feature_map
) &
1696 MD_FEATURE_RECOVERY_OFFSET
))
1697 rdev
->recovery_offset
= le64_to_cpu(sb
->recovery_offset
);
1699 set_bit(In_sync
, &rdev
->flags
);
1700 rdev
->raid_disk
= role
;
1703 if (sb
->devflags
& WriteMostly1
)
1704 set_bit(WriteMostly
, &rdev
->flags
);
1705 } else /* MULTIPATH are always insync */
1706 set_bit(In_sync
, &rdev
->flags
);
1711 static void super_1_sync(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1713 struct mdp_superblock_1
*sb
;
1716 /* make rdev->sb match mddev and rdev data. */
1718 sb
= page_address(rdev
->sb_page
);
1720 sb
->feature_map
= 0;
1722 sb
->recovery_offset
= cpu_to_le64(0);
1723 memset(sb
->pad1
, 0, sizeof(sb
->pad1
));
1724 memset(sb
->pad3
, 0, sizeof(sb
->pad3
));
1726 sb
->utime
= cpu_to_le64((__u64
)mddev
->utime
);
1727 sb
->events
= cpu_to_le64(mddev
->events
);
1729 sb
->resync_offset
= cpu_to_le64(mddev
->recovery_cp
);
1731 sb
->resync_offset
= cpu_to_le64(0);
1733 sb
->cnt_corrected_read
= cpu_to_le32(atomic_read(&rdev
->corrected_errors
));
1735 sb
->raid_disks
= cpu_to_le32(mddev
->raid_disks
);
1736 sb
->size
= cpu_to_le64(mddev
->dev_sectors
);
1737 sb
->chunksize
= cpu_to_le32(mddev
->chunk_sectors
);
1738 sb
->level
= cpu_to_le32(mddev
->level
);
1739 sb
->layout
= cpu_to_le32(mddev
->layout
);
1741 if (test_bit(WriteMostly
, &rdev
->flags
))
1742 sb
->devflags
|= WriteMostly1
;
1744 sb
->devflags
&= ~WriteMostly1
;
1746 if (mddev
->bitmap
&& mddev
->bitmap_info
.file
== NULL
) {
1747 sb
->bitmap_offset
= cpu_to_le32((__u32
)mddev
->bitmap_info
.offset
);
1748 sb
->feature_map
= cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1751 if (rdev
->raid_disk
>= 0 &&
1752 !test_bit(In_sync
, &rdev
->flags
)) {
1754 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET
);
1755 sb
->recovery_offset
=
1756 cpu_to_le64(rdev
->recovery_offset
);
1759 if (mddev
->reshape_position
!= MaxSector
) {
1760 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1761 sb
->reshape_position
= cpu_to_le64(mddev
->reshape_position
);
1762 sb
->new_layout
= cpu_to_le32(mddev
->new_layout
);
1763 sb
->delta_disks
= cpu_to_le32(mddev
->delta_disks
);
1764 sb
->new_level
= cpu_to_le32(mddev
->new_level
);
1765 sb
->new_chunk
= cpu_to_le32(mddev
->new_chunk_sectors
);
1768 if (rdev
->badblocks
.count
== 0)
1769 /* Nothing to do for bad blocks*/ ;
1770 else if (sb
->bblog_offset
== 0)
1771 /* Cannot record bad blocks on this device */
1772 md_error(mddev
, rdev
);
1774 struct badblocks
*bb
= &rdev
->badblocks
;
1775 u64
*bbp
= (u64
*)page_address(rdev
->bb_page
);
1777 sb
->feature_map
|= cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1782 seq
= read_seqbegin(&bb
->lock
);
1784 memset(bbp
, 0xff, PAGE_SIZE
);
1786 for (i
= 0 ; i
< bb
->count
; i
++) {
1787 u64 internal_bb
= *p
++;
1788 u64 store_bb
= ((BB_OFFSET(internal_bb
) << 10)
1789 | BB_LEN(internal_bb
));
1790 *bbp
++ = cpu_to_le64(store_bb
);
1792 if (read_seqretry(&bb
->lock
, seq
))
1795 bb
->sector
= (rdev
->sb_start
+
1796 (int)le32_to_cpu(sb
->bblog_offset
));
1797 bb
->size
= le16_to_cpu(sb
->bblog_size
);
1803 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
1804 if (rdev2
->desc_nr
+1 > max_dev
)
1805 max_dev
= rdev2
->desc_nr
+1;
1807 if (max_dev
> le32_to_cpu(sb
->max_dev
)) {
1809 sb
->max_dev
= cpu_to_le32(max_dev
);
1810 rdev
->sb_size
= max_dev
* 2 + 256;
1811 bmask
= queue_logical_block_size(rdev
->bdev
->bd_disk
->queue
)-1;
1812 if (rdev
->sb_size
& bmask
)
1813 rdev
->sb_size
= (rdev
->sb_size
| bmask
) + 1;
1815 max_dev
= le32_to_cpu(sb
->max_dev
);
1817 for (i
=0; i
<max_dev
;i
++)
1818 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1820 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
1822 if (test_bit(Faulty
, &rdev2
->flags
))
1823 sb
->dev_roles
[i
] = cpu_to_le16(0xfffe);
1824 else if (test_bit(In_sync
, &rdev2
->flags
))
1825 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1826 else if (rdev2
->raid_disk
>= 0)
1827 sb
->dev_roles
[i
] = cpu_to_le16(rdev2
->raid_disk
);
1829 sb
->dev_roles
[i
] = cpu_to_le16(0xffff);
1832 sb
->sb_csum
= calc_sb_1_csum(sb
);
1835 static unsigned long long
1836 super_1_rdev_size_change(mdk_rdev_t
*rdev
, sector_t num_sectors
)
1838 struct mdp_superblock_1
*sb
;
1839 sector_t max_sectors
;
1840 if (num_sectors
&& num_sectors
< rdev
->mddev
->dev_sectors
)
1841 return 0; /* component must fit device */
1842 if (rdev
->sb_start
< rdev
->data_offset
) {
1843 /* minor versions 1 and 2; superblock before data */
1844 max_sectors
= i_size_read(rdev
->bdev
->bd_inode
) >> 9;
1845 max_sectors
-= rdev
->data_offset
;
1846 if (!num_sectors
|| num_sectors
> max_sectors
)
1847 num_sectors
= max_sectors
;
1848 } else if (rdev
->mddev
->bitmap_info
.offset
) {
1849 /* minor version 0 with bitmap we can't move */
1852 /* minor version 0; superblock after data */
1854 sb_start
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) - 8*2;
1855 sb_start
&= ~(sector_t
)(4*2 - 1);
1856 max_sectors
= rdev
->sectors
+ sb_start
- rdev
->sb_start
;
1857 if (!num_sectors
|| num_sectors
> max_sectors
)
1858 num_sectors
= max_sectors
;
1859 rdev
->sb_start
= sb_start
;
1861 sb
= page_address(rdev
->sb_page
);
1862 sb
->data_size
= cpu_to_le64(num_sectors
);
1863 sb
->super_offset
= rdev
->sb_start
;
1864 sb
->sb_csum
= calc_sb_1_csum(sb
);
1865 md_super_write(rdev
->mddev
, rdev
, rdev
->sb_start
, rdev
->sb_size
,
1867 md_super_wait(rdev
->mddev
);
1871 static struct super_type super_types
[] = {
1874 .owner
= THIS_MODULE
,
1875 .load_super
= super_90_load
,
1876 .validate_super
= super_90_validate
,
1877 .sync_super
= super_90_sync
,
1878 .rdev_size_change
= super_90_rdev_size_change
,
1882 .owner
= THIS_MODULE
,
1883 .load_super
= super_1_load
,
1884 .validate_super
= super_1_validate
,
1885 .sync_super
= super_1_sync
,
1886 .rdev_size_change
= super_1_rdev_size_change
,
1890 static void sync_super(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1892 if (mddev
->sync_super
) {
1893 mddev
->sync_super(mddev
, rdev
);
1897 BUG_ON(mddev
->major_version
>= ARRAY_SIZE(super_types
));
1899 super_types
[mddev
->major_version
].sync_super(mddev
, rdev
);
1902 static int match_mddev_units(mddev_t
*mddev1
, mddev_t
*mddev2
)
1904 mdk_rdev_t
*rdev
, *rdev2
;
1907 rdev_for_each_rcu(rdev
, mddev1
)
1908 rdev_for_each_rcu(rdev2
, mddev2
)
1909 if (rdev
->bdev
->bd_contains
==
1910 rdev2
->bdev
->bd_contains
) {
1918 static LIST_HEAD(pending_raid_disks
);
1921 * Try to register data integrity profile for an mddev
1923 * This is called when an array is started and after a disk has been kicked
1924 * from the array. It only succeeds if all working and active component devices
1925 * are integrity capable with matching profiles.
1927 int md_integrity_register(mddev_t
*mddev
)
1929 mdk_rdev_t
*rdev
, *reference
= NULL
;
1931 if (list_empty(&mddev
->disks
))
1932 return 0; /* nothing to do */
1933 if (!mddev
->gendisk
|| blk_get_integrity(mddev
->gendisk
))
1934 return 0; /* shouldn't register, or already is */
1935 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
1936 /* skip spares and non-functional disks */
1937 if (test_bit(Faulty
, &rdev
->flags
))
1939 if (rdev
->raid_disk
< 0)
1942 /* Use the first rdev as the reference */
1946 /* does this rdev's profile match the reference profile? */
1947 if (blk_integrity_compare(reference
->bdev
->bd_disk
,
1948 rdev
->bdev
->bd_disk
) < 0)
1951 if (!reference
|| !bdev_get_integrity(reference
->bdev
))
1954 * All component devices are integrity capable and have matching
1955 * profiles, register the common profile for the md device.
1957 if (blk_integrity_register(mddev
->gendisk
,
1958 bdev_get_integrity(reference
->bdev
)) != 0) {
1959 printk(KERN_ERR
"md: failed to register integrity for %s\n",
1963 printk(KERN_NOTICE
"md: data integrity enabled on %s\n", mdname(mddev
));
1964 if (bioset_integrity_create(mddev
->bio_set
, BIO_POOL_SIZE
)) {
1965 printk(KERN_ERR
"md: failed to create integrity pool for %s\n",
1971 EXPORT_SYMBOL(md_integrity_register
);
1973 /* Disable data integrity if non-capable/non-matching disk is being added */
1974 void md_integrity_add_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
1976 struct blk_integrity
*bi_rdev
= bdev_get_integrity(rdev
->bdev
);
1977 struct blk_integrity
*bi_mddev
= blk_get_integrity(mddev
->gendisk
);
1979 if (!bi_mddev
) /* nothing to do */
1981 if (rdev
->raid_disk
< 0) /* skip spares */
1983 if (bi_rdev
&& blk_integrity_compare(mddev
->gendisk
,
1984 rdev
->bdev
->bd_disk
) >= 0)
1986 printk(KERN_NOTICE
"disabling data integrity on %s\n", mdname(mddev
));
1987 blk_integrity_unregister(mddev
->gendisk
);
1989 EXPORT_SYMBOL(md_integrity_add_rdev
);
1991 static int bind_rdev_to_array(mdk_rdev_t
* rdev
, mddev_t
* mddev
)
1993 char b
[BDEVNAME_SIZE
];
2003 /* prevent duplicates */
2004 if (find_rdev(mddev
, rdev
->bdev
->bd_dev
))
2007 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2008 if (rdev
->sectors
&& (mddev
->dev_sectors
== 0 ||
2009 rdev
->sectors
< mddev
->dev_sectors
)) {
2011 /* Cannot change size, so fail
2012 * If mddev->level <= 0, then we don't care
2013 * about aligning sizes (e.g. linear)
2015 if (mddev
->level
> 0)
2018 mddev
->dev_sectors
= rdev
->sectors
;
2021 /* Verify rdev->desc_nr is unique.
2022 * If it is -1, assign a free number, else
2023 * check number is not in use
2025 if (rdev
->desc_nr
< 0) {
2027 if (mddev
->pers
) choice
= mddev
->raid_disks
;
2028 while (find_rdev_nr(mddev
, choice
))
2030 rdev
->desc_nr
= choice
;
2032 if (find_rdev_nr(mddev
, rdev
->desc_nr
))
2035 if (mddev
->max_disks
&& rdev
->desc_nr
>= mddev
->max_disks
) {
2036 printk(KERN_WARNING
"md: %s: array is limited to %d devices\n",
2037 mdname(mddev
), mddev
->max_disks
);
2040 bdevname(rdev
->bdev
,b
);
2041 while ( (s
=strchr(b
, '/')) != NULL
)
2044 rdev
->mddev
= mddev
;
2045 printk(KERN_INFO
"md: bind<%s>\n", b
);
2047 if ((err
= kobject_add(&rdev
->kobj
, &mddev
->kobj
, "dev-%s", b
)))
2050 ko
= &part_to_dev(rdev
->bdev
->bd_part
)->kobj
;
2051 if (sysfs_create_link(&rdev
->kobj
, ko
, "block"))
2052 /* failure here is OK */;
2053 rdev
->sysfs_state
= sysfs_get_dirent_safe(rdev
->kobj
.sd
, "state");
2055 list_add_rcu(&rdev
->same_set
, &mddev
->disks
);
2056 bd_link_disk_holder(rdev
->bdev
, mddev
->gendisk
);
2058 /* May as well allow recovery to be retried once */
2059 mddev
->recovery_disabled
++;
2064 printk(KERN_WARNING
"md: failed to register dev-%s for %s\n",
2069 static void md_delayed_delete(struct work_struct
*ws
)
2071 mdk_rdev_t
*rdev
= container_of(ws
, mdk_rdev_t
, del_work
);
2072 kobject_del(&rdev
->kobj
);
2073 kobject_put(&rdev
->kobj
);
2076 static void unbind_rdev_from_array(mdk_rdev_t
* rdev
)
2078 char b
[BDEVNAME_SIZE
];
2083 bd_unlink_disk_holder(rdev
->bdev
, rdev
->mddev
->gendisk
);
2084 list_del_rcu(&rdev
->same_set
);
2085 printk(KERN_INFO
"md: unbind<%s>\n", bdevname(rdev
->bdev
,b
));
2087 sysfs_remove_link(&rdev
->kobj
, "block");
2088 sysfs_put(rdev
->sysfs_state
);
2089 rdev
->sysfs_state
= NULL
;
2090 kfree(rdev
->badblocks
.page
);
2091 rdev
->badblocks
.count
= 0;
2092 rdev
->badblocks
.page
= NULL
;
2093 /* We need to delay this, otherwise we can deadlock when
2094 * writing to 'remove' to "dev/state". We also need
2095 * to delay it due to rcu usage.
2098 INIT_WORK(&rdev
->del_work
, md_delayed_delete
);
2099 kobject_get(&rdev
->kobj
);
2100 queue_work(md_misc_wq
, &rdev
->del_work
);
2104 * prevent the device from being mounted, repartitioned or
2105 * otherwise reused by a RAID array (or any other kernel
2106 * subsystem), by bd_claiming the device.
2108 static int lock_rdev(mdk_rdev_t
*rdev
, dev_t dev
, int shared
)
2111 struct block_device
*bdev
;
2112 char b
[BDEVNAME_SIZE
];
2114 bdev
= blkdev_get_by_dev(dev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
,
2115 shared
? (mdk_rdev_t
*)lock_rdev
: rdev
);
2117 printk(KERN_ERR
"md: could not open %s.\n",
2118 __bdevname(dev
, b
));
2119 return PTR_ERR(bdev
);
2125 static void unlock_rdev(mdk_rdev_t
*rdev
)
2127 struct block_device
*bdev
= rdev
->bdev
;
2131 blkdev_put(bdev
, FMODE_READ
|FMODE_WRITE
|FMODE_EXCL
);
2134 void md_autodetect_dev(dev_t dev
);
2136 static void export_rdev(mdk_rdev_t
* rdev
)
2138 char b
[BDEVNAME_SIZE
];
2139 printk(KERN_INFO
"md: export_rdev(%s)\n",
2140 bdevname(rdev
->bdev
,b
));
2145 if (test_bit(AutoDetected
, &rdev
->flags
))
2146 md_autodetect_dev(rdev
->bdev
->bd_dev
);
2149 kobject_put(&rdev
->kobj
);
2152 static void kick_rdev_from_array(mdk_rdev_t
* rdev
)
2154 unbind_rdev_from_array(rdev
);
2158 static void export_array(mddev_t
*mddev
)
2160 mdk_rdev_t
*rdev
, *tmp
;
2162 rdev_for_each(rdev
, tmp
, mddev
) {
2167 kick_rdev_from_array(rdev
);
2169 if (!list_empty(&mddev
->disks
))
2171 mddev
->raid_disks
= 0;
2172 mddev
->major_version
= 0;
2175 static void print_desc(mdp_disk_t
*desc
)
2177 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc
->number
,
2178 desc
->major
,desc
->minor
,desc
->raid_disk
,desc
->state
);
2181 static void print_sb_90(mdp_super_t
*sb
)
2186 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2187 sb
->major_version
, sb
->minor_version
, sb
->patch_version
,
2188 sb
->set_uuid0
, sb
->set_uuid1
, sb
->set_uuid2
, sb
->set_uuid3
,
2190 printk(KERN_INFO
"md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2191 sb
->level
, sb
->size
, sb
->nr_disks
, sb
->raid_disks
,
2192 sb
->md_minor
, sb
->layout
, sb
->chunk_size
);
2193 printk(KERN_INFO
"md: UT:%08x ST:%d AD:%d WD:%d"
2194 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2195 sb
->utime
, sb
->state
, sb
->active_disks
, sb
->working_disks
,
2196 sb
->failed_disks
, sb
->spare_disks
,
2197 sb
->sb_csum
, (unsigned long)sb
->events_lo
);
2200 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2203 desc
= sb
->disks
+ i
;
2204 if (desc
->number
|| desc
->major
|| desc
->minor
||
2205 desc
->raid_disk
|| (desc
->state
&& (desc
->state
!= 4))) {
2206 printk(" D %2d: ", i
);
2210 printk(KERN_INFO
"md: THIS: ");
2211 print_desc(&sb
->this_disk
);
2214 static void print_sb_1(struct mdp_superblock_1
*sb
)
2218 uuid
= sb
->set_uuid
;
2220 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2221 "md: Name: \"%s\" CT:%llu\n",
2222 le32_to_cpu(sb
->major_version
),
2223 le32_to_cpu(sb
->feature_map
),
2226 (unsigned long long)le64_to_cpu(sb
->ctime
)
2227 & MD_SUPERBLOCK_1_TIME_SEC_MASK
);
2229 uuid
= sb
->device_uuid
;
2231 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2233 "md: Dev:%08x UUID: %pU\n"
2234 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2235 "md: (MaxDev:%u) \n",
2236 le32_to_cpu(sb
->level
),
2237 (unsigned long long)le64_to_cpu(sb
->size
),
2238 le32_to_cpu(sb
->raid_disks
),
2239 le32_to_cpu(sb
->layout
),
2240 le32_to_cpu(sb
->chunksize
),
2241 (unsigned long long)le64_to_cpu(sb
->data_offset
),
2242 (unsigned long long)le64_to_cpu(sb
->data_size
),
2243 (unsigned long long)le64_to_cpu(sb
->super_offset
),
2244 (unsigned long long)le64_to_cpu(sb
->recovery_offset
),
2245 le32_to_cpu(sb
->dev_number
),
2248 (unsigned long long)le64_to_cpu(sb
->utime
) & MD_SUPERBLOCK_1_TIME_SEC_MASK
,
2249 (unsigned long long)le64_to_cpu(sb
->events
),
2250 (unsigned long long)le64_to_cpu(sb
->resync_offset
),
2251 le32_to_cpu(sb
->sb_csum
),
2252 le32_to_cpu(sb
->max_dev
)
2256 static void print_rdev(mdk_rdev_t
*rdev
, int major_version
)
2258 char b
[BDEVNAME_SIZE
];
2259 printk(KERN_INFO
"md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2260 bdevname(rdev
->bdev
, b
), (unsigned long long)rdev
->sectors
,
2261 test_bit(Faulty
, &rdev
->flags
), test_bit(In_sync
, &rdev
->flags
),
2263 if (rdev
->sb_loaded
) {
2264 printk(KERN_INFO
"md: rdev superblock (MJ:%d):\n", major_version
);
2265 switch (major_version
) {
2267 print_sb_90(page_address(rdev
->sb_page
));
2270 print_sb_1(page_address(rdev
->sb_page
));
2274 printk(KERN_INFO
"md: no rdev superblock!\n");
2277 static void md_print_devices(void)
2279 struct list_head
*tmp
;
2282 char b
[BDEVNAME_SIZE
];
2285 printk("md: **********************************\n");
2286 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2287 printk("md: **********************************\n");
2288 for_each_mddev(mddev
, tmp
) {
2291 bitmap_print_sb(mddev
->bitmap
);
2293 printk("%s: ", mdname(mddev
));
2294 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2295 printk("<%s>", bdevname(rdev
->bdev
,b
));
2298 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
2299 print_rdev(rdev
, mddev
->major_version
);
2301 printk("md: **********************************\n");
2306 static void sync_sbs(mddev_t
* mddev
, int nospares
)
2308 /* Update each superblock (in-memory image), but
2309 * if we are allowed to, skip spares which already
2310 * have the right event counter, or have one earlier
2311 * (which would mean they aren't being marked as dirty
2312 * with the rest of the array)
2315 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2316 if (rdev
->sb_events
== mddev
->events
||
2318 rdev
->raid_disk
< 0 &&
2319 rdev
->sb_events
+1 == mddev
->events
)) {
2320 /* Don't update this superblock */
2321 rdev
->sb_loaded
= 2;
2323 sync_super(mddev
, rdev
);
2324 rdev
->sb_loaded
= 1;
2329 static void md_update_sb(mddev_t
* mddev
, int force_change
)
2334 int any_badblocks_changed
= 0;
2337 /* First make sure individual recovery_offsets are correct */
2338 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2339 if (rdev
->raid_disk
>= 0 &&
2340 mddev
->delta_disks
>= 0 &&
2341 !test_bit(In_sync
, &rdev
->flags
) &&
2342 mddev
->curr_resync_completed
> rdev
->recovery_offset
)
2343 rdev
->recovery_offset
= mddev
->curr_resync_completed
;
2346 if (!mddev
->persistent
) {
2347 clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
2348 clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2349 if (!mddev
->external
) {
2350 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2351 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2352 if (rdev
->badblocks
.changed
) {
2353 md_ack_all_badblocks(&rdev
->badblocks
);
2354 md_error(mddev
, rdev
);
2356 clear_bit(Blocked
, &rdev
->flags
);
2357 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2358 wake_up(&rdev
->blocked_wait
);
2361 wake_up(&mddev
->sb_wait
);
2365 spin_lock_irq(&mddev
->write_lock
);
2367 mddev
->utime
= get_seconds();
2369 if (test_and_clear_bit(MD_CHANGE_DEVS
, &mddev
->flags
))
2371 if (test_and_clear_bit(MD_CHANGE_CLEAN
, &mddev
->flags
))
2372 /* just a clean<-> dirty transition, possibly leave spares alone,
2373 * though if events isn't the right even/odd, we will have to do
2379 if (mddev
->degraded
)
2380 /* If the array is degraded, then skipping spares is both
2381 * dangerous and fairly pointless.
2382 * Dangerous because a device that was removed from the array
2383 * might have a event_count that still looks up-to-date,
2384 * so it can be re-added without a resync.
2385 * Pointless because if there are any spares to skip,
2386 * then a recovery will happen and soon that array won't
2387 * be degraded any more and the spare can go back to sleep then.
2391 sync_req
= mddev
->in_sync
;
2393 /* If this is just a dirty<->clean transition, and the array is clean
2394 * and 'events' is odd, we can roll back to the previous clean state */
2396 && (mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
2397 && mddev
->can_decrease_events
2398 && mddev
->events
!= 1) {
2400 mddev
->can_decrease_events
= 0;
2402 /* otherwise we have to go forward and ... */
2404 mddev
->can_decrease_events
= nospares
;
2407 if (!mddev
->events
) {
2409 * oops, this 64-bit counter should never wrap.
2410 * Either we are in around ~1 trillion A.C., assuming
2411 * 1 reboot per second, or we have a bug:
2417 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2418 if (rdev
->badblocks
.changed
)
2419 any_badblocks_changed
++;
2420 if (test_bit(Faulty
, &rdev
->flags
))
2421 set_bit(FaultRecorded
, &rdev
->flags
);
2424 sync_sbs(mddev
, nospares
);
2425 spin_unlock_irq(&mddev
->write_lock
);
2428 "md: updating %s RAID superblock on device (in sync %d)\n",
2429 mdname(mddev
),mddev
->in_sync
);
2431 bitmap_update_sb(mddev
->bitmap
);
2432 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2433 char b
[BDEVNAME_SIZE
];
2434 dprintk(KERN_INFO
"md: ");
2435 if (rdev
->sb_loaded
!= 1)
2436 continue; /* no noise on spare devices */
2437 if (test_bit(Faulty
, &rdev
->flags
))
2438 dprintk("(skipping faulty ");
2440 dprintk("%s ", bdevname(rdev
->bdev
,b
));
2441 if (!test_bit(Faulty
, &rdev
->flags
)) {
2442 md_super_write(mddev
,rdev
,
2443 rdev
->sb_start
, rdev
->sb_size
,
2445 dprintk(KERN_INFO
"(write) %s's sb offset: %llu\n",
2446 bdevname(rdev
->bdev
,b
),
2447 (unsigned long long)rdev
->sb_start
);
2448 rdev
->sb_events
= mddev
->events
;
2449 if (rdev
->badblocks
.size
) {
2450 md_super_write(mddev
, rdev
,
2451 rdev
->badblocks
.sector
,
2452 rdev
->badblocks
.size
<< 9,
2454 rdev
->badblocks
.size
= 0;
2459 if (mddev
->level
== LEVEL_MULTIPATH
)
2460 /* only need to write one superblock... */
2463 md_super_wait(mddev
);
2464 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2466 spin_lock_irq(&mddev
->write_lock
);
2467 if (mddev
->in_sync
!= sync_req
||
2468 test_bit(MD_CHANGE_DEVS
, &mddev
->flags
)) {
2469 /* have to write it out again */
2470 spin_unlock_irq(&mddev
->write_lock
);
2473 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
2474 spin_unlock_irq(&mddev
->write_lock
);
2475 wake_up(&mddev
->sb_wait
);
2476 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
2477 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
2479 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2480 if (test_and_clear_bit(FaultRecorded
, &rdev
->flags
))
2481 clear_bit(Blocked
, &rdev
->flags
);
2483 if (any_badblocks_changed
)
2484 md_ack_all_badblocks(&rdev
->badblocks
);
2485 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2486 wake_up(&rdev
->blocked_wait
);
2490 /* words written to sysfs files may, or may not, be \n terminated.
2491 * We want to accept with case. For this we use cmd_match.
2493 static int cmd_match(const char *cmd
, const char *str
)
2495 /* See if cmd, written into a sysfs file, matches
2496 * str. They must either be the same, or cmd can
2497 * have a trailing newline
2499 while (*cmd
&& *str
&& *cmd
== *str
) {
2510 struct rdev_sysfs_entry
{
2511 struct attribute attr
;
2512 ssize_t (*show
)(mdk_rdev_t
*, char *);
2513 ssize_t (*store
)(mdk_rdev_t
*, const char *, size_t);
2517 state_show(mdk_rdev_t
*rdev
, char *page
)
2522 if (test_bit(Faulty
, &rdev
->flags
) ||
2523 rdev
->badblocks
.unacked_exist
) {
2524 len
+= sprintf(page
+len
, "%sfaulty",sep
);
2527 if (test_bit(In_sync
, &rdev
->flags
)) {
2528 len
+= sprintf(page
+len
, "%sin_sync",sep
);
2531 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2532 len
+= sprintf(page
+len
, "%swrite_mostly",sep
);
2535 if (test_bit(Blocked
, &rdev
->flags
) ||
2536 rdev
->badblocks
.unacked_exist
) {
2537 len
+= sprintf(page
+len
, "%sblocked", sep
);
2540 if (!test_bit(Faulty
, &rdev
->flags
) &&
2541 !test_bit(In_sync
, &rdev
->flags
)) {
2542 len
+= sprintf(page
+len
, "%sspare", sep
);
2545 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
2546 len
+= sprintf(page
+len
, "%swrite_error", sep
);
2549 return len
+sprintf(page
+len
, "\n");
2553 state_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2556 * faulty - simulates an error
2557 * remove - disconnects the device
2558 * writemostly - sets write_mostly
2559 * -writemostly - clears write_mostly
2560 * blocked - sets the Blocked flags
2561 * -blocked - clears the Blocked and possibly simulates an error
2562 * insync - sets Insync providing device isn't active
2563 * write_error - sets WriteErrorSeen
2564 * -write_error - clears WriteErrorSeen
2567 if (cmd_match(buf
, "faulty") && rdev
->mddev
->pers
) {
2568 md_error(rdev
->mddev
, rdev
);
2569 if (test_bit(Faulty
, &rdev
->flags
))
2573 } else if (cmd_match(buf
, "remove")) {
2574 if (rdev
->raid_disk
>= 0)
2577 mddev_t
*mddev
= rdev
->mddev
;
2578 kick_rdev_from_array(rdev
);
2580 md_update_sb(mddev
, 1);
2581 md_new_event(mddev
);
2584 } else if (cmd_match(buf
, "writemostly")) {
2585 set_bit(WriteMostly
, &rdev
->flags
);
2587 } else if (cmd_match(buf
, "-writemostly")) {
2588 clear_bit(WriteMostly
, &rdev
->flags
);
2590 } else if (cmd_match(buf
, "blocked")) {
2591 set_bit(Blocked
, &rdev
->flags
);
2593 } else if (cmd_match(buf
, "-blocked")) {
2594 if (!test_bit(Faulty
, &rdev
->flags
) &&
2595 test_bit(BlockedBadBlocks
, &rdev
->flags
)) {
2596 /* metadata handler doesn't understand badblocks,
2597 * so we need to fail the device
2599 md_error(rdev
->mddev
, rdev
);
2601 clear_bit(Blocked
, &rdev
->flags
);
2602 clear_bit(BlockedBadBlocks
, &rdev
->flags
);
2603 wake_up(&rdev
->blocked_wait
);
2604 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2605 md_wakeup_thread(rdev
->mddev
->thread
);
2608 } else if (cmd_match(buf
, "insync") && rdev
->raid_disk
== -1) {
2609 set_bit(In_sync
, &rdev
->flags
);
2611 } else if (cmd_match(buf
, "write_error")) {
2612 set_bit(WriteErrorSeen
, &rdev
->flags
);
2614 } else if (cmd_match(buf
, "-write_error")) {
2615 clear_bit(WriteErrorSeen
, &rdev
->flags
);
2619 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2620 return err
? err
: len
;
2622 static struct rdev_sysfs_entry rdev_state
=
2623 __ATTR(state
, S_IRUGO
|S_IWUSR
, state_show
, state_store
);
2626 errors_show(mdk_rdev_t
*rdev
, char *page
)
2628 return sprintf(page
, "%d\n", atomic_read(&rdev
->corrected_errors
));
2632 errors_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2635 unsigned long n
= simple_strtoul(buf
, &e
, 10);
2636 if (*buf
&& (*e
== 0 || *e
== '\n')) {
2637 atomic_set(&rdev
->corrected_errors
, n
);
2642 static struct rdev_sysfs_entry rdev_errors
=
2643 __ATTR(errors
, S_IRUGO
|S_IWUSR
, errors_show
, errors_store
);
2646 slot_show(mdk_rdev_t
*rdev
, char *page
)
2648 if (rdev
->raid_disk
< 0)
2649 return sprintf(page
, "none\n");
2651 return sprintf(page
, "%d\n", rdev
->raid_disk
);
2655 slot_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2659 int slot
= simple_strtoul(buf
, &e
, 10);
2660 if (strncmp(buf
, "none", 4)==0)
2662 else if (e
==buf
|| (*e
&& *e
!= '\n'))
2664 if (rdev
->mddev
->pers
&& slot
== -1) {
2665 /* Setting 'slot' on an active array requires also
2666 * updating the 'rd%d' link, and communicating
2667 * with the personality with ->hot_*_disk.
2668 * For now we only support removing
2669 * failed/spare devices. This normally happens automatically,
2670 * but not when the metadata is externally managed.
2672 if (rdev
->raid_disk
== -1)
2674 /* personality does all needed checks */
2675 if (rdev
->mddev
->pers
->hot_remove_disk
== NULL
)
2677 err
= rdev
->mddev
->pers
->
2678 hot_remove_disk(rdev
->mddev
, rdev
->raid_disk
);
2681 sysfs_unlink_rdev(rdev
->mddev
, rdev
);
2682 rdev
->raid_disk
= -1;
2683 set_bit(MD_RECOVERY_NEEDED
, &rdev
->mddev
->recovery
);
2684 md_wakeup_thread(rdev
->mddev
->thread
);
2685 } else if (rdev
->mddev
->pers
) {
2687 /* Activating a spare .. or possibly reactivating
2688 * if we ever get bitmaps working here.
2691 if (rdev
->raid_disk
!= -1)
2694 if (test_bit(MD_RECOVERY_RUNNING
, &rdev
->mddev
->recovery
))
2697 if (rdev
->mddev
->pers
->hot_add_disk
== NULL
)
2700 list_for_each_entry(rdev2
, &rdev
->mddev
->disks
, same_set
)
2701 if (rdev2
->raid_disk
== slot
)
2704 if (slot
>= rdev
->mddev
->raid_disks
&&
2705 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2708 rdev
->raid_disk
= slot
;
2709 if (test_bit(In_sync
, &rdev
->flags
))
2710 rdev
->saved_raid_disk
= slot
;
2712 rdev
->saved_raid_disk
= -1;
2713 err
= rdev
->mddev
->pers
->
2714 hot_add_disk(rdev
->mddev
, rdev
);
2716 rdev
->raid_disk
= -1;
2719 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2720 if (sysfs_link_rdev(rdev
->mddev
, rdev
))
2721 /* failure here is OK */;
2722 /* don't wakeup anyone, leave that to userspace. */
2724 if (slot
>= rdev
->mddev
->raid_disks
&&
2725 slot
>= rdev
->mddev
->raid_disks
+ rdev
->mddev
->delta_disks
)
2727 rdev
->raid_disk
= slot
;
2728 /* assume it is working */
2729 clear_bit(Faulty
, &rdev
->flags
);
2730 clear_bit(WriteMostly
, &rdev
->flags
);
2731 set_bit(In_sync
, &rdev
->flags
);
2732 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
2738 static struct rdev_sysfs_entry rdev_slot
=
2739 __ATTR(slot
, S_IRUGO
|S_IWUSR
, slot_show
, slot_store
);
2742 offset_show(mdk_rdev_t
*rdev
, char *page
)
2744 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->data_offset
);
2748 offset_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2751 unsigned long long offset
= simple_strtoull(buf
, &e
, 10);
2752 if (e
==buf
|| (*e
&& *e
!= '\n'))
2754 if (rdev
->mddev
->pers
&& rdev
->raid_disk
>= 0)
2756 if (rdev
->sectors
&& rdev
->mddev
->external
)
2757 /* Must set offset before size, so overlap checks
2760 rdev
->data_offset
= offset
;
2764 static struct rdev_sysfs_entry rdev_offset
=
2765 __ATTR(offset
, S_IRUGO
|S_IWUSR
, offset_show
, offset_store
);
2768 rdev_size_show(mdk_rdev_t
*rdev
, char *page
)
2770 return sprintf(page
, "%llu\n", (unsigned long long)rdev
->sectors
/ 2);
2773 static int overlaps(sector_t s1
, sector_t l1
, sector_t s2
, sector_t l2
)
2775 /* check if two start/length pairs overlap */
2783 static int strict_blocks_to_sectors(const char *buf
, sector_t
*sectors
)
2785 unsigned long long blocks
;
2788 if (strict_strtoull(buf
, 10, &blocks
) < 0)
2791 if (blocks
& 1ULL << (8 * sizeof(blocks
) - 1))
2792 return -EINVAL
; /* sector conversion overflow */
2795 if (new != blocks
* 2)
2796 return -EINVAL
; /* unsigned long long to sector_t overflow */
2803 rdev_size_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2805 mddev_t
*my_mddev
= rdev
->mddev
;
2806 sector_t oldsectors
= rdev
->sectors
;
2809 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
2811 if (my_mddev
->pers
&& rdev
->raid_disk
>= 0) {
2812 if (my_mddev
->persistent
) {
2813 sectors
= super_types
[my_mddev
->major_version
].
2814 rdev_size_change(rdev
, sectors
);
2817 } else if (!sectors
)
2818 sectors
= (i_size_read(rdev
->bdev
->bd_inode
) >> 9) -
2821 if (sectors
< my_mddev
->dev_sectors
)
2822 return -EINVAL
; /* component must fit device */
2824 rdev
->sectors
= sectors
;
2825 if (sectors
> oldsectors
&& my_mddev
->external
) {
2826 /* need to check that all other rdevs with the same ->bdev
2827 * do not overlap. We need to unlock the mddev to avoid
2828 * a deadlock. We have already changed rdev->sectors, and if
2829 * we have to change it back, we will have the lock again.
2833 struct list_head
*tmp
;
2835 mddev_unlock(my_mddev
);
2836 for_each_mddev(mddev
, tmp
) {
2840 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
)
2841 if (rdev
->bdev
== rdev2
->bdev
&&
2843 overlaps(rdev
->data_offset
, rdev
->sectors
,
2849 mddev_unlock(mddev
);
2855 mddev_lock(my_mddev
);
2857 /* Someone else could have slipped in a size
2858 * change here, but doing so is just silly.
2859 * We put oldsectors back because we *know* it is
2860 * safe, and trust userspace not to race with
2863 rdev
->sectors
= oldsectors
;
2870 static struct rdev_sysfs_entry rdev_size
=
2871 __ATTR(size
, S_IRUGO
|S_IWUSR
, rdev_size_show
, rdev_size_store
);
2874 static ssize_t
recovery_start_show(mdk_rdev_t
*rdev
, char *page
)
2876 unsigned long long recovery_start
= rdev
->recovery_offset
;
2878 if (test_bit(In_sync
, &rdev
->flags
) ||
2879 recovery_start
== MaxSector
)
2880 return sprintf(page
, "none\n");
2882 return sprintf(page
, "%llu\n", recovery_start
);
2885 static ssize_t
recovery_start_store(mdk_rdev_t
*rdev
, const char *buf
, size_t len
)
2887 unsigned long long recovery_start
;
2889 if (cmd_match(buf
, "none"))
2890 recovery_start
= MaxSector
;
2891 else if (strict_strtoull(buf
, 10, &recovery_start
))
2894 if (rdev
->mddev
->pers
&&
2895 rdev
->raid_disk
>= 0)
2898 rdev
->recovery_offset
= recovery_start
;
2899 if (recovery_start
== MaxSector
)
2900 set_bit(In_sync
, &rdev
->flags
);
2902 clear_bit(In_sync
, &rdev
->flags
);
2906 static struct rdev_sysfs_entry rdev_recovery_start
=
2907 __ATTR(recovery_start
, S_IRUGO
|S_IWUSR
, recovery_start_show
, recovery_start_store
);
2911 badblocks_show(struct badblocks
*bb
, char *page
, int unack
);
2913 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
);
2915 static ssize_t
bb_show(mdk_rdev_t
*rdev
, char *page
)
2917 return badblocks_show(&rdev
->badblocks
, page
, 0);
2919 static ssize_t
bb_store(mdk_rdev_t
*rdev
, const char *page
, size_t len
)
2921 int rv
= badblocks_store(&rdev
->badblocks
, page
, len
, 0);
2922 /* Maybe that ack was all we needed */
2923 if (test_and_clear_bit(BlockedBadBlocks
, &rdev
->flags
))
2924 wake_up(&rdev
->blocked_wait
);
2927 static struct rdev_sysfs_entry rdev_bad_blocks
=
2928 __ATTR(bad_blocks
, S_IRUGO
|S_IWUSR
, bb_show
, bb_store
);
2931 static ssize_t
ubb_show(mdk_rdev_t
*rdev
, char *page
)
2933 return badblocks_show(&rdev
->badblocks
, page
, 1);
2935 static ssize_t
ubb_store(mdk_rdev_t
*rdev
, const char *page
, size_t len
)
2937 return badblocks_store(&rdev
->badblocks
, page
, len
, 1);
2939 static struct rdev_sysfs_entry rdev_unack_bad_blocks
=
2940 __ATTR(unacknowledged_bad_blocks
, S_IRUGO
|S_IWUSR
, ubb_show
, ubb_store
);
2942 static struct attribute
*rdev_default_attrs
[] = {
2948 &rdev_recovery_start
.attr
,
2949 &rdev_bad_blocks
.attr
,
2950 &rdev_unack_bad_blocks
.attr
,
2954 rdev_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
2956 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2957 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2958 mddev_t
*mddev
= rdev
->mddev
;
2964 rv
= mddev
? mddev_lock(mddev
) : -EBUSY
;
2966 if (rdev
->mddev
== NULL
)
2969 rv
= entry
->show(rdev
, page
);
2970 mddev_unlock(mddev
);
2976 rdev_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
2977 const char *page
, size_t length
)
2979 struct rdev_sysfs_entry
*entry
= container_of(attr
, struct rdev_sysfs_entry
, attr
);
2980 mdk_rdev_t
*rdev
= container_of(kobj
, mdk_rdev_t
, kobj
);
2982 mddev_t
*mddev
= rdev
->mddev
;
2986 if (!capable(CAP_SYS_ADMIN
))
2988 rv
= mddev
? mddev_lock(mddev
): -EBUSY
;
2990 if (rdev
->mddev
== NULL
)
2993 rv
= entry
->store(rdev
, page
, length
);
2994 mddev_unlock(mddev
);
2999 static void rdev_free(struct kobject
*ko
)
3001 mdk_rdev_t
*rdev
= container_of(ko
, mdk_rdev_t
, kobj
);
3004 static const struct sysfs_ops rdev_sysfs_ops
= {
3005 .show
= rdev_attr_show
,
3006 .store
= rdev_attr_store
,
3008 static struct kobj_type rdev_ktype
= {
3009 .release
= rdev_free
,
3010 .sysfs_ops
= &rdev_sysfs_ops
,
3011 .default_attrs
= rdev_default_attrs
,
3014 int md_rdev_init(mdk_rdev_t
*rdev
)
3017 rdev
->saved_raid_disk
= -1;
3018 rdev
->raid_disk
= -1;
3020 rdev
->data_offset
= 0;
3021 rdev
->sb_events
= 0;
3022 rdev
->last_read_error
.tv_sec
= 0;
3023 rdev
->last_read_error
.tv_nsec
= 0;
3024 rdev
->sb_loaded
= 0;
3025 rdev
->bb_page
= NULL
;
3026 atomic_set(&rdev
->nr_pending
, 0);
3027 atomic_set(&rdev
->read_errors
, 0);
3028 atomic_set(&rdev
->corrected_errors
, 0);
3030 INIT_LIST_HEAD(&rdev
->same_set
);
3031 init_waitqueue_head(&rdev
->blocked_wait
);
3033 /* Add space to store bad block list.
3034 * This reserves the space even on arrays where it cannot
3035 * be used - I wonder if that matters
3037 rdev
->badblocks
.count
= 0;
3038 rdev
->badblocks
.shift
= 0;
3039 rdev
->badblocks
.page
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3040 seqlock_init(&rdev
->badblocks
.lock
);
3041 if (rdev
->badblocks
.page
== NULL
)
3046 EXPORT_SYMBOL_GPL(md_rdev_init
);
3048 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3050 * mark the device faulty if:
3052 * - the device is nonexistent (zero size)
3053 * - the device has no valid superblock
3055 * a faulty rdev _never_ has rdev->sb set.
3057 static mdk_rdev_t
*md_import_device(dev_t newdev
, int super_format
, int super_minor
)
3059 char b
[BDEVNAME_SIZE
];
3064 rdev
= kzalloc(sizeof(*rdev
), GFP_KERNEL
);
3066 printk(KERN_ERR
"md: could not alloc mem for new device!\n");
3067 return ERR_PTR(-ENOMEM
);
3070 err
= md_rdev_init(rdev
);
3073 err
= alloc_disk_sb(rdev
);
3077 err
= lock_rdev(rdev
, newdev
, super_format
== -2);
3081 kobject_init(&rdev
->kobj
, &rdev_ktype
);
3083 size
= i_size_read(rdev
->bdev
->bd_inode
) >> BLOCK_SIZE_BITS
;
3086 "md: %s has zero or unknown size, marking faulty!\n",
3087 bdevname(rdev
->bdev
,b
));
3092 if (super_format
>= 0) {
3093 err
= super_types
[super_format
].
3094 load_super(rdev
, NULL
, super_minor
);
3095 if (err
== -EINVAL
) {
3097 "md: %s does not have a valid v%d.%d "
3098 "superblock, not importing!\n",
3099 bdevname(rdev
->bdev
,b
),
3100 super_format
, super_minor
);
3105 "md: could not read %s's sb, not importing!\n",
3106 bdevname(rdev
->bdev
,b
));
3110 if (super_format
== -1)
3111 /* hot-add for 0.90, or non-persistent: so no badblocks */
3112 rdev
->badblocks
.shift
= -1;
3120 kfree(rdev
->badblocks
.page
);
3122 return ERR_PTR(err
);
3126 * Check a full RAID array for plausibility
3130 static void analyze_sbs(mddev_t
* mddev
)
3133 mdk_rdev_t
*rdev
, *freshest
, *tmp
;
3134 char b
[BDEVNAME_SIZE
];
3137 rdev_for_each(rdev
, tmp
, mddev
)
3138 switch (super_types
[mddev
->major_version
].
3139 load_super(rdev
, freshest
, mddev
->minor_version
)) {
3147 "md: fatal superblock inconsistency in %s"
3148 " -- removing from array\n",
3149 bdevname(rdev
->bdev
,b
));
3150 kick_rdev_from_array(rdev
);
3154 super_types
[mddev
->major_version
].
3155 validate_super(mddev
, freshest
);
3158 rdev_for_each(rdev
, tmp
, mddev
) {
3159 if (mddev
->max_disks
&&
3160 (rdev
->desc_nr
>= mddev
->max_disks
||
3161 i
> mddev
->max_disks
)) {
3163 "md: %s: %s: only %d devices permitted\n",
3164 mdname(mddev
), bdevname(rdev
->bdev
, b
),
3166 kick_rdev_from_array(rdev
);
3169 if (rdev
!= freshest
)
3170 if (super_types
[mddev
->major_version
].
3171 validate_super(mddev
, rdev
)) {
3172 printk(KERN_WARNING
"md: kicking non-fresh %s"
3174 bdevname(rdev
->bdev
,b
));
3175 kick_rdev_from_array(rdev
);
3178 if (mddev
->level
== LEVEL_MULTIPATH
) {
3179 rdev
->desc_nr
= i
++;
3180 rdev
->raid_disk
= rdev
->desc_nr
;
3181 set_bit(In_sync
, &rdev
->flags
);
3182 } else if (rdev
->raid_disk
>= (mddev
->raid_disks
- min(0, mddev
->delta_disks
))) {
3183 rdev
->raid_disk
= -1;
3184 clear_bit(In_sync
, &rdev
->flags
);
3189 /* Read a fixed-point number.
3190 * Numbers in sysfs attributes should be in "standard" units where
3191 * possible, so time should be in seconds.
3192 * However we internally use a a much smaller unit such as
3193 * milliseconds or jiffies.
3194 * This function takes a decimal number with a possible fractional
3195 * component, and produces an integer which is the result of
3196 * multiplying that number by 10^'scale'.
3197 * all without any floating-point arithmetic.
3199 int strict_strtoul_scaled(const char *cp
, unsigned long *res
, int scale
)
3201 unsigned long result
= 0;
3203 while (isdigit(*cp
) || (*cp
== '.' && decimals
< 0)) {
3206 else if (decimals
< scale
) {
3209 result
= result
* 10 + value
;
3221 while (decimals
< scale
) {
3230 static void md_safemode_timeout(unsigned long data
);
3233 safe_delay_show(mddev_t
*mddev
, char *page
)
3235 int msec
= (mddev
->safemode_delay
*1000)/HZ
;
3236 return sprintf(page
, "%d.%03d\n", msec
/1000, msec
%1000);
3239 safe_delay_store(mddev_t
*mddev
, const char *cbuf
, size_t len
)
3243 if (strict_strtoul_scaled(cbuf
, &msec
, 3) < 0)
3246 mddev
->safemode_delay
= 0;
3248 unsigned long old_delay
= mddev
->safemode_delay
;
3249 mddev
->safemode_delay
= (msec
*HZ
)/1000;
3250 if (mddev
->safemode_delay
== 0)
3251 mddev
->safemode_delay
= 1;
3252 if (mddev
->safemode_delay
< old_delay
)
3253 md_safemode_timeout((unsigned long)mddev
);
3257 static struct md_sysfs_entry md_safe_delay
=
3258 __ATTR(safe_mode_delay
, S_IRUGO
|S_IWUSR
,safe_delay_show
, safe_delay_store
);
3261 level_show(mddev_t
*mddev
, char *page
)
3263 struct mdk_personality
*p
= mddev
->pers
;
3265 return sprintf(page
, "%s\n", p
->name
);
3266 else if (mddev
->clevel
[0])
3267 return sprintf(page
, "%s\n", mddev
->clevel
);
3268 else if (mddev
->level
!= LEVEL_NONE
)
3269 return sprintf(page
, "%d\n", mddev
->level
);
3275 level_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3279 struct mdk_personality
*pers
;
3284 if (mddev
->pers
== NULL
) {
3287 if (len
>= sizeof(mddev
->clevel
))
3289 strncpy(mddev
->clevel
, buf
, len
);
3290 if (mddev
->clevel
[len
-1] == '\n')
3292 mddev
->clevel
[len
] = 0;
3293 mddev
->level
= LEVEL_NONE
;
3297 /* request to change the personality. Need to ensure:
3298 * - array is not engaged in resync/recovery/reshape
3299 * - old personality can be suspended
3300 * - new personality will access other array.
3303 if (mddev
->sync_thread
||
3304 mddev
->reshape_position
!= MaxSector
||
3305 mddev
->sysfs_active
)
3308 if (!mddev
->pers
->quiesce
) {
3309 printk(KERN_WARNING
"md: %s: %s does not support online personality change\n",
3310 mdname(mddev
), mddev
->pers
->name
);
3314 /* Now find the new personality */
3315 if (len
== 0 || len
>= sizeof(clevel
))
3317 strncpy(clevel
, buf
, len
);
3318 if (clevel
[len
-1] == '\n')
3321 if (strict_strtol(clevel
, 10, &level
))
3324 if (request_module("md-%s", clevel
) != 0)
3325 request_module("md-level-%s", clevel
);
3326 spin_lock(&pers_lock
);
3327 pers
= find_pers(level
, clevel
);
3328 if (!pers
|| !try_module_get(pers
->owner
)) {
3329 spin_unlock(&pers_lock
);
3330 printk(KERN_WARNING
"md: personality %s not loaded\n", clevel
);
3333 spin_unlock(&pers_lock
);
3335 if (pers
== mddev
->pers
) {
3336 /* Nothing to do! */
3337 module_put(pers
->owner
);
3340 if (!pers
->takeover
) {
3341 module_put(pers
->owner
);
3342 printk(KERN_WARNING
"md: %s: %s does not support personality takeover\n",
3343 mdname(mddev
), clevel
);
3347 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
3348 rdev
->new_raid_disk
= rdev
->raid_disk
;
3350 /* ->takeover must set new_* and/or delta_disks
3351 * if it succeeds, and may set them when it fails.
3353 priv
= pers
->takeover(mddev
);
3355 mddev
->new_level
= mddev
->level
;
3356 mddev
->new_layout
= mddev
->layout
;
3357 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3358 mddev
->raid_disks
-= mddev
->delta_disks
;
3359 mddev
->delta_disks
= 0;
3360 module_put(pers
->owner
);
3361 printk(KERN_WARNING
"md: %s: %s would not accept array\n",
3362 mdname(mddev
), clevel
);
3363 return PTR_ERR(priv
);
3366 /* Looks like we have a winner */
3367 mddev_suspend(mddev
);
3368 mddev
->pers
->stop(mddev
);
3370 if (mddev
->pers
->sync_request
== NULL
&&
3371 pers
->sync_request
!= NULL
) {
3372 /* need to add the md_redundancy_group */
3373 if (sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
3375 "md: cannot register extra attributes for %s\n",
3377 mddev
->sysfs_action
= sysfs_get_dirent(mddev
->kobj
.sd
, NULL
, "sync_action");
3379 if (mddev
->pers
->sync_request
!= NULL
&&
3380 pers
->sync_request
== NULL
) {
3381 /* need to remove the md_redundancy_group */
3382 if (mddev
->to_remove
== NULL
)
3383 mddev
->to_remove
= &md_redundancy_group
;
3386 if (mddev
->pers
->sync_request
== NULL
&&
3388 /* We are converting from a no-redundancy array
3389 * to a redundancy array and metadata is managed
3390 * externally so we need to be sure that writes
3391 * won't block due to a need to transition
3393 * until external management is started.
3396 mddev
->safemode_delay
= 0;
3397 mddev
->safemode
= 0;
3400 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3401 if (rdev
->raid_disk
< 0)
3403 if (rdev
->new_raid_disk
>= mddev
->raid_disks
)
3404 rdev
->new_raid_disk
= -1;
3405 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3407 sysfs_unlink_rdev(mddev
, rdev
);
3409 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
3410 if (rdev
->raid_disk
< 0)
3412 if (rdev
->new_raid_disk
== rdev
->raid_disk
)
3414 rdev
->raid_disk
= rdev
->new_raid_disk
;
3415 if (rdev
->raid_disk
< 0)
3416 clear_bit(In_sync
, &rdev
->flags
);
3418 if (sysfs_link_rdev(mddev
, rdev
))
3419 printk(KERN_WARNING
"md: cannot register rd%d"
3420 " for %s after level change\n",
3421 rdev
->raid_disk
, mdname(mddev
));
3425 module_put(mddev
->pers
->owner
);
3427 mddev
->private = priv
;
3428 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
3429 mddev
->level
= mddev
->new_level
;
3430 mddev
->layout
= mddev
->new_layout
;
3431 mddev
->chunk_sectors
= mddev
->new_chunk_sectors
;
3432 mddev
->delta_disks
= 0;
3433 mddev
->degraded
= 0;
3434 if (mddev
->pers
->sync_request
== NULL
) {
3435 /* this is now an array without redundancy, so
3436 * it must always be in_sync
3439 del_timer_sync(&mddev
->safemode_timer
);
3442 mddev_resume(mddev
);
3443 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
3444 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3445 md_wakeup_thread(mddev
->thread
);
3446 sysfs_notify(&mddev
->kobj
, NULL
, "level");
3447 md_new_event(mddev
);
3451 static struct md_sysfs_entry md_level
=
3452 __ATTR(level
, S_IRUGO
|S_IWUSR
, level_show
, level_store
);
3456 layout_show(mddev_t
*mddev
, char *page
)
3458 /* just a number, not meaningful for all levels */
3459 if (mddev
->reshape_position
!= MaxSector
&&
3460 mddev
->layout
!= mddev
->new_layout
)
3461 return sprintf(page
, "%d (%d)\n",
3462 mddev
->new_layout
, mddev
->layout
);
3463 return sprintf(page
, "%d\n", mddev
->layout
);
3467 layout_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3470 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3472 if (!*buf
|| (*e
&& *e
!= '\n'))
3477 if (mddev
->pers
->check_reshape
== NULL
)
3479 mddev
->new_layout
= n
;
3480 err
= mddev
->pers
->check_reshape(mddev
);
3482 mddev
->new_layout
= mddev
->layout
;
3486 mddev
->new_layout
= n
;
3487 if (mddev
->reshape_position
== MaxSector
)
3492 static struct md_sysfs_entry md_layout
=
3493 __ATTR(layout
, S_IRUGO
|S_IWUSR
, layout_show
, layout_store
);
3497 raid_disks_show(mddev_t
*mddev
, char *page
)
3499 if (mddev
->raid_disks
== 0)
3501 if (mddev
->reshape_position
!= MaxSector
&&
3502 mddev
->delta_disks
!= 0)
3503 return sprintf(page
, "%d (%d)\n", mddev
->raid_disks
,
3504 mddev
->raid_disks
- mddev
->delta_disks
);
3505 return sprintf(page
, "%d\n", mddev
->raid_disks
);
3508 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
);
3511 raid_disks_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3515 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3517 if (!*buf
|| (*e
&& *e
!= '\n'))
3521 rv
= update_raid_disks(mddev
, n
);
3522 else if (mddev
->reshape_position
!= MaxSector
) {
3523 int olddisks
= mddev
->raid_disks
- mddev
->delta_disks
;
3524 mddev
->delta_disks
= n
- olddisks
;
3525 mddev
->raid_disks
= n
;
3527 mddev
->raid_disks
= n
;
3528 return rv
? rv
: len
;
3530 static struct md_sysfs_entry md_raid_disks
=
3531 __ATTR(raid_disks
, S_IRUGO
|S_IWUSR
, raid_disks_show
, raid_disks_store
);
3534 chunk_size_show(mddev_t
*mddev
, char *page
)
3536 if (mddev
->reshape_position
!= MaxSector
&&
3537 mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
)
3538 return sprintf(page
, "%d (%d)\n",
3539 mddev
->new_chunk_sectors
<< 9,
3540 mddev
->chunk_sectors
<< 9);
3541 return sprintf(page
, "%d\n", mddev
->chunk_sectors
<< 9);
3545 chunk_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3548 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3550 if (!*buf
|| (*e
&& *e
!= '\n'))
3555 if (mddev
->pers
->check_reshape
== NULL
)
3557 mddev
->new_chunk_sectors
= n
>> 9;
3558 err
= mddev
->pers
->check_reshape(mddev
);
3560 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3564 mddev
->new_chunk_sectors
= n
>> 9;
3565 if (mddev
->reshape_position
== MaxSector
)
3566 mddev
->chunk_sectors
= n
>> 9;
3570 static struct md_sysfs_entry md_chunk_size
=
3571 __ATTR(chunk_size
, S_IRUGO
|S_IWUSR
, chunk_size_show
, chunk_size_store
);
3574 resync_start_show(mddev_t
*mddev
, char *page
)
3576 if (mddev
->recovery_cp
== MaxSector
)
3577 return sprintf(page
, "none\n");
3578 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->recovery_cp
);
3582 resync_start_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3585 unsigned long long n
= simple_strtoull(buf
, &e
, 10);
3587 if (mddev
->pers
&& !test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
3589 if (cmd_match(buf
, "none"))
3591 else if (!*buf
|| (*e
&& *e
!= '\n'))
3594 mddev
->recovery_cp
= n
;
3597 static struct md_sysfs_entry md_resync_start
=
3598 __ATTR(resync_start
, S_IRUGO
|S_IWUSR
, resync_start_show
, resync_start_store
);
3601 * The array state can be:
3604 * No devices, no size, no level
3605 * Equivalent to STOP_ARRAY ioctl
3607 * May have some settings, but array is not active
3608 * all IO results in error
3609 * When written, doesn't tear down array, but just stops it
3610 * suspended (not supported yet)
3611 * All IO requests will block. The array can be reconfigured.
3612 * Writing this, if accepted, will block until array is quiescent
3614 * no resync can happen. no superblocks get written.
3615 * write requests fail
3617 * like readonly, but behaves like 'clean' on a write request.
3619 * clean - no pending writes, but otherwise active.
3620 * When written to inactive array, starts without resync
3621 * If a write request arrives then
3622 * if metadata is known, mark 'dirty' and switch to 'active'.
3623 * if not known, block and switch to write-pending
3624 * If written to an active array that has pending writes, then fails.
3626 * fully active: IO and resync can be happening.
3627 * When written to inactive array, starts with resync
3630 * clean, but writes are blocked waiting for 'active' to be written.
3633 * like active, but no writes have been seen for a while (100msec).
3636 enum array_state
{ clear
, inactive
, suspended
, readonly
, read_auto
, clean
, active
,
3637 write_pending
, active_idle
, bad_word
};
3638 static char *array_states
[] = {
3639 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3640 "write-pending", "active-idle", NULL
};
3642 static int match_word(const char *word
, char **list
)
3645 for (n
=0; list
[n
]; n
++)
3646 if (cmd_match(word
, list
[n
]))
3652 array_state_show(mddev_t
*mddev
, char *page
)
3654 enum array_state st
= inactive
;
3667 else if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
3669 else if (mddev
->safemode
)
3675 if (list_empty(&mddev
->disks
) &&
3676 mddev
->raid_disks
== 0 &&
3677 mddev
->dev_sectors
== 0)
3682 return sprintf(page
, "%s\n", array_states
[st
]);
3685 static int do_md_stop(mddev_t
* mddev
, int ro
, int is_open
);
3686 static int md_set_readonly(mddev_t
* mddev
, int is_open
);
3687 static int do_md_run(mddev_t
* mddev
);
3688 static int restart_array(mddev_t
*mddev
);
3691 array_state_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3694 enum array_state st
= match_word(buf
, array_states
);
3699 /* stopping an active array */
3700 if (atomic_read(&mddev
->openers
) > 0)
3702 err
= do_md_stop(mddev
, 0, 0);
3705 /* stopping an active array */
3707 if (atomic_read(&mddev
->openers
) > 0)
3709 err
= do_md_stop(mddev
, 2, 0);
3711 err
= 0; /* already inactive */
3714 break; /* not supported yet */
3717 err
= md_set_readonly(mddev
, 0);
3720 set_disk_ro(mddev
->gendisk
, 1);
3721 err
= do_md_run(mddev
);
3727 err
= md_set_readonly(mddev
, 0);
3728 else if (mddev
->ro
== 1)
3729 err
= restart_array(mddev
);
3732 set_disk_ro(mddev
->gendisk
, 0);
3736 err
= do_md_run(mddev
);
3741 restart_array(mddev
);
3742 spin_lock_irq(&mddev
->write_lock
);
3743 if (atomic_read(&mddev
->writes_pending
) == 0) {
3744 if (mddev
->in_sync
== 0) {
3746 if (mddev
->safemode
== 1)
3747 mddev
->safemode
= 0;
3748 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
3753 spin_unlock_irq(&mddev
->write_lock
);
3759 restart_array(mddev
);
3760 clear_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
3761 wake_up(&mddev
->sb_wait
);
3765 set_disk_ro(mddev
->gendisk
, 0);
3766 err
= do_md_run(mddev
);
3771 /* these cannot be set */
3777 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
3781 static struct md_sysfs_entry md_array_state
=
3782 __ATTR(array_state
, S_IRUGO
|S_IWUSR
, array_state_show
, array_state_store
);
3785 max_corrected_read_errors_show(mddev_t
*mddev
, char *page
) {
3786 return sprintf(page
, "%d\n",
3787 atomic_read(&mddev
->max_corr_read_errors
));
3791 max_corrected_read_errors_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3794 unsigned long n
= simple_strtoul(buf
, &e
, 10);
3796 if (*buf
&& (*e
== 0 || *e
== '\n')) {
3797 atomic_set(&mddev
->max_corr_read_errors
, n
);
3803 static struct md_sysfs_entry max_corr_read_errors
=
3804 __ATTR(max_read_errors
, S_IRUGO
|S_IWUSR
, max_corrected_read_errors_show
,
3805 max_corrected_read_errors_store
);
3808 null_show(mddev_t
*mddev
, char *page
)
3814 new_dev_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3816 /* buf must be %d:%d\n? giving major and minor numbers */
3817 /* The new device is added to the array.
3818 * If the array has a persistent superblock, we read the
3819 * superblock to initialise info and check validity.
3820 * Otherwise, only checking done is that in bind_rdev_to_array,
3821 * which mainly checks size.
3824 int major
= simple_strtoul(buf
, &e
, 10);
3830 if (!*buf
|| *e
!= ':' || !e
[1] || e
[1] == '\n')
3832 minor
= simple_strtoul(e
+1, &e
, 10);
3833 if (*e
&& *e
!= '\n')
3835 dev
= MKDEV(major
, minor
);
3836 if (major
!= MAJOR(dev
) ||
3837 minor
!= MINOR(dev
))
3841 if (mddev
->persistent
) {
3842 rdev
= md_import_device(dev
, mddev
->major_version
,
3843 mddev
->minor_version
);
3844 if (!IS_ERR(rdev
) && !list_empty(&mddev
->disks
)) {
3845 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
3846 mdk_rdev_t
, same_set
);
3847 err
= super_types
[mddev
->major_version
]
3848 .load_super(rdev
, rdev0
, mddev
->minor_version
);
3852 } else if (mddev
->external
)
3853 rdev
= md_import_device(dev
, -2, -1);
3855 rdev
= md_import_device(dev
, -1, -1);
3858 return PTR_ERR(rdev
);
3859 err
= bind_rdev_to_array(rdev
, mddev
);
3863 return err
? err
: len
;
3866 static struct md_sysfs_entry md_new_device
=
3867 __ATTR(new_dev
, S_IWUSR
, null_show
, new_dev_store
);
3870 bitmap_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3873 unsigned long chunk
, end_chunk
;
3877 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3879 chunk
= end_chunk
= simple_strtoul(buf
, &end
, 0);
3880 if (buf
== end
) break;
3881 if (*end
== '-') { /* range */
3883 end_chunk
= simple_strtoul(buf
, &end
, 0);
3884 if (buf
== end
) break;
3886 if (*end
&& !isspace(*end
)) break;
3887 bitmap_dirty_bits(mddev
->bitmap
, chunk
, end_chunk
);
3888 buf
= skip_spaces(end
);
3890 bitmap_unplug(mddev
->bitmap
); /* flush the bits to disk */
3895 static struct md_sysfs_entry md_bitmap
=
3896 __ATTR(bitmap_set_bits
, S_IWUSR
, null_show
, bitmap_store
);
3899 size_show(mddev_t
*mddev
, char *page
)
3901 return sprintf(page
, "%llu\n",
3902 (unsigned long long)mddev
->dev_sectors
/ 2);
3905 static int update_size(mddev_t
*mddev
, sector_t num_sectors
);
3908 size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3910 /* If array is inactive, we can reduce the component size, but
3911 * not increase it (except from 0).
3912 * If array is active, we can try an on-line resize
3915 int err
= strict_blocks_to_sectors(buf
, §ors
);
3920 err
= update_size(mddev
, sectors
);
3921 md_update_sb(mddev
, 1);
3923 if (mddev
->dev_sectors
== 0 ||
3924 mddev
->dev_sectors
> sectors
)
3925 mddev
->dev_sectors
= sectors
;
3929 return err
? err
: len
;
3932 static struct md_sysfs_entry md_size
=
3933 __ATTR(component_size
, S_IRUGO
|S_IWUSR
, size_show
, size_store
);
3938 * 'none' for arrays with no metadata (good luck...)
3939 * 'external' for arrays with externally managed metadata,
3940 * or N.M for internally known formats
3943 metadata_show(mddev_t
*mddev
, char *page
)
3945 if (mddev
->persistent
)
3946 return sprintf(page
, "%d.%d\n",
3947 mddev
->major_version
, mddev
->minor_version
);
3948 else if (mddev
->external
)
3949 return sprintf(page
, "external:%s\n", mddev
->metadata_type
);
3951 return sprintf(page
, "none\n");
3955 metadata_store(mddev_t
*mddev
, const char *buf
, size_t len
)
3959 /* Changing the details of 'external' metadata is
3960 * always permitted. Otherwise there must be
3961 * no devices attached to the array.
3963 if (mddev
->external
&& strncmp(buf
, "external:", 9) == 0)
3965 else if (!list_empty(&mddev
->disks
))
3968 if (cmd_match(buf
, "none")) {
3969 mddev
->persistent
= 0;
3970 mddev
->external
= 0;
3971 mddev
->major_version
= 0;
3972 mddev
->minor_version
= 90;
3975 if (strncmp(buf
, "external:", 9) == 0) {
3976 size_t namelen
= len
-9;
3977 if (namelen
>= sizeof(mddev
->metadata_type
))
3978 namelen
= sizeof(mddev
->metadata_type
)-1;
3979 strncpy(mddev
->metadata_type
, buf
+9, namelen
);
3980 mddev
->metadata_type
[namelen
] = 0;
3981 if (namelen
&& mddev
->metadata_type
[namelen
-1] == '\n')
3982 mddev
->metadata_type
[--namelen
] = 0;
3983 mddev
->persistent
= 0;
3984 mddev
->external
= 1;
3985 mddev
->major_version
= 0;
3986 mddev
->minor_version
= 90;
3989 major
= simple_strtoul(buf
, &e
, 10);
3990 if (e
==buf
|| *e
!= '.')
3993 minor
= simple_strtoul(buf
, &e
, 10);
3994 if (e
==buf
|| (*e
&& *e
!= '\n') )
3996 if (major
>= ARRAY_SIZE(super_types
) || super_types
[major
].name
== NULL
)
3998 mddev
->major_version
= major
;
3999 mddev
->minor_version
= minor
;
4000 mddev
->persistent
= 1;
4001 mddev
->external
= 0;
4005 static struct md_sysfs_entry md_metadata
=
4006 __ATTR(metadata_version
, S_IRUGO
|S_IWUSR
, metadata_show
, metadata_store
);
4009 action_show(mddev_t
*mddev
, char *page
)
4011 char *type
= "idle";
4012 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
4014 else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4015 (!mddev
->ro
&& test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))) {
4016 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
4018 else if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
4019 if (!test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
4021 else if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
4025 } else if (test_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
))
4028 return sprintf(page
, "%s\n", type
);
4031 static void reap_sync_thread(mddev_t
*mddev
);
4034 action_store(mddev_t
*mddev
, const char *page
, size_t len
)
4036 if (!mddev
->pers
|| !mddev
->pers
->sync_request
)
4039 if (cmd_match(page
, "frozen"))
4040 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4042 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
4044 if (cmd_match(page
, "idle") || cmd_match(page
, "frozen")) {
4045 if (mddev
->sync_thread
) {
4046 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
4047 reap_sync_thread(mddev
);
4049 } else if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) ||
4050 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
4052 else if (cmd_match(page
, "resync"))
4053 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4054 else if (cmd_match(page
, "recover")) {
4055 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
4056 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4057 } else if (cmd_match(page
, "reshape")) {
4059 if (mddev
->pers
->start_reshape
== NULL
)
4061 err
= mddev
->pers
->start_reshape(mddev
);
4064 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4066 if (cmd_match(page
, "check"))
4067 set_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
4068 else if (!cmd_match(page
, "repair"))
4070 set_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
4071 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
4073 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4074 md_wakeup_thread(mddev
->thread
);
4075 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4080 mismatch_cnt_show(mddev_t
*mddev
, char *page
)
4082 return sprintf(page
, "%llu\n",
4083 (unsigned long long) mddev
->resync_mismatches
);
4086 static struct md_sysfs_entry md_scan_mode
=
4087 __ATTR(sync_action
, S_IRUGO
|S_IWUSR
, action_show
, action_store
);
4090 static struct md_sysfs_entry md_mismatches
= __ATTR_RO(mismatch_cnt
);
4093 sync_min_show(mddev_t
*mddev
, char *page
)
4095 return sprintf(page
, "%d (%s)\n", speed_min(mddev
),
4096 mddev
->sync_speed_min
? "local": "system");
4100 sync_min_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4104 if (strncmp(buf
, "system", 6)==0) {
4105 mddev
->sync_speed_min
= 0;
4108 min
= simple_strtoul(buf
, &e
, 10);
4109 if (buf
== e
|| (*e
&& *e
!= '\n') || min
<= 0)
4111 mddev
->sync_speed_min
= min
;
4115 static struct md_sysfs_entry md_sync_min
=
4116 __ATTR(sync_speed_min
, S_IRUGO
|S_IWUSR
, sync_min_show
, sync_min_store
);
4119 sync_max_show(mddev_t
*mddev
, char *page
)
4121 return sprintf(page
, "%d (%s)\n", speed_max(mddev
),
4122 mddev
->sync_speed_max
? "local": "system");
4126 sync_max_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4130 if (strncmp(buf
, "system", 6)==0) {
4131 mddev
->sync_speed_max
= 0;
4134 max
= simple_strtoul(buf
, &e
, 10);
4135 if (buf
== e
|| (*e
&& *e
!= '\n') || max
<= 0)
4137 mddev
->sync_speed_max
= max
;
4141 static struct md_sysfs_entry md_sync_max
=
4142 __ATTR(sync_speed_max
, S_IRUGO
|S_IWUSR
, sync_max_show
, sync_max_store
);
4145 degraded_show(mddev_t
*mddev
, char *page
)
4147 return sprintf(page
, "%d\n", mddev
->degraded
);
4149 static struct md_sysfs_entry md_degraded
= __ATTR_RO(degraded
);
4152 sync_force_parallel_show(mddev_t
*mddev
, char *page
)
4154 return sprintf(page
, "%d\n", mddev
->parallel_resync
);
4158 sync_force_parallel_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4162 if (strict_strtol(buf
, 10, &n
))
4165 if (n
!= 0 && n
!= 1)
4168 mddev
->parallel_resync
= n
;
4170 if (mddev
->sync_thread
)
4171 wake_up(&resync_wait
);
4176 /* force parallel resync, even with shared block devices */
4177 static struct md_sysfs_entry md_sync_force_parallel
=
4178 __ATTR(sync_force_parallel
, S_IRUGO
|S_IWUSR
,
4179 sync_force_parallel_show
, sync_force_parallel_store
);
4182 sync_speed_show(mddev_t
*mddev
, char *page
)
4184 unsigned long resync
, dt
, db
;
4185 if (mddev
->curr_resync
== 0)
4186 return sprintf(page
, "none\n");
4187 resync
= mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
);
4188 dt
= (jiffies
- mddev
->resync_mark
) / HZ
;
4190 db
= resync
- mddev
->resync_mark_cnt
;
4191 return sprintf(page
, "%lu\n", db
/dt
/2); /* K/sec */
4194 static struct md_sysfs_entry md_sync_speed
= __ATTR_RO(sync_speed
);
4197 sync_completed_show(mddev_t
*mddev
, char *page
)
4199 unsigned long long max_sectors
, resync
;
4201 if (!test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4202 return sprintf(page
, "none\n");
4204 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
4205 max_sectors
= mddev
->resync_max_sectors
;
4207 max_sectors
= mddev
->dev_sectors
;
4209 resync
= mddev
->curr_resync_completed
;
4210 return sprintf(page
, "%llu / %llu\n", resync
, max_sectors
);
4213 static struct md_sysfs_entry md_sync_completed
= __ATTR_RO(sync_completed
);
4216 min_sync_show(mddev_t
*mddev
, char *page
)
4218 return sprintf(page
, "%llu\n",
4219 (unsigned long long)mddev
->resync_min
);
4222 min_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4224 unsigned long long min
;
4225 if (strict_strtoull(buf
, 10, &min
))
4227 if (min
> mddev
->resync_max
)
4229 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4232 /* Must be a multiple of chunk_size */
4233 if (mddev
->chunk_sectors
) {
4234 sector_t temp
= min
;
4235 if (sector_div(temp
, mddev
->chunk_sectors
))
4238 mddev
->resync_min
= min
;
4243 static struct md_sysfs_entry md_min_sync
=
4244 __ATTR(sync_min
, S_IRUGO
|S_IWUSR
, min_sync_show
, min_sync_store
);
4247 max_sync_show(mddev_t
*mddev
, char *page
)
4249 if (mddev
->resync_max
== MaxSector
)
4250 return sprintf(page
, "max\n");
4252 return sprintf(page
, "%llu\n",
4253 (unsigned long long)mddev
->resync_max
);
4256 max_sync_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4258 if (strncmp(buf
, "max", 3) == 0)
4259 mddev
->resync_max
= MaxSector
;
4261 unsigned long long max
;
4262 if (strict_strtoull(buf
, 10, &max
))
4264 if (max
< mddev
->resync_min
)
4266 if (max
< mddev
->resync_max
&&
4268 test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
))
4271 /* Must be a multiple of chunk_size */
4272 if (mddev
->chunk_sectors
) {
4273 sector_t temp
= max
;
4274 if (sector_div(temp
, mddev
->chunk_sectors
))
4277 mddev
->resync_max
= max
;
4279 wake_up(&mddev
->recovery_wait
);
4283 static struct md_sysfs_entry md_max_sync
=
4284 __ATTR(sync_max
, S_IRUGO
|S_IWUSR
, max_sync_show
, max_sync_store
);
4287 suspend_lo_show(mddev_t
*mddev
, char *page
)
4289 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_lo
);
4293 suspend_lo_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4296 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4297 unsigned long long old
= mddev
->suspend_lo
;
4299 if (mddev
->pers
== NULL
||
4300 mddev
->pers
->quiesce
== NULL
)
4302 if (buf
== e
|| (*e
&& *e
!= '\n'))
4305 mddev
->suspend_lo
= new;
4307 /* Shrinking suspended region */
4308 mddev
->pers
->quiesce(mddev
, 2);
4310 /* Expanding suspended region - need to wait */
4311 mddev
->pers
->quiesce(mddev
, 1);
4312 mddev
->pers
->quiesce(mddev
, 0);
4316 static struct md_sysfs_entry md_suspend_lo
=
4317 __ATTR(suspend_lo
, S_IRUGO
|S_IWUSR
, suspend_lo_show
, suspend_lo_store
);
4321 suspend_hi_show(mddev_t
*mddev
, char *page
)
4323 return sprintf(page
, "%llu\n", (unsigned long long)mddev
->suspend_hi
);
4327 suspend_hi_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4330 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4331 unsigned long long old
= mddev
->suspend_hi
;
4333 if (mddev
->pers
== NULL
||
4334 mddev
->pers
->quiesce
== NULL
)
4336 if (buf
== e
|| (*e
&& *e
!= '\n'))
4339 mddev
->suspend_hi
= new;
4341 /* Shrinking suspended region */
4342 mddev
->pers
->quiesce(mddev
, 2);
4344 /* Expanding suspended region - need to wait */
4345 mddev
->pers
->quiesce(mddev
, 1);
4346 mddev
->pers
->quiesce(mddev
, 0);
4350 static struct md_sysfs_entry md_suspend_hi
=
4351 __ATTR(suspend_hi
, S_IRUGO
|S_IWUSR
, suspend_hi_show
, suspend_hi_store
);
4354 reshape_position_show(mddev_t
*mddev
, char *page
)
4356 if (mddev
->reshape_position
!= MaxSector
)
4357 return sprintf(page
, "%llu\n",
4358 (unsigned long long)mddev
->reshape_position
);
4359 strcpy(page
, "none\n");
4364 reshape_position_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4367 unsigned long long new = simple_strtoull(buf
, &e
, 10);
4370 if (buf
== e
|| (*e
&& *e
!= '\n'))
4372 mddev
->reshape_position
= new;
4373 mddev
->delta_disks
= 0;
4374 mddev
->new_level
= mddev
->level
;
4375 mddev
->new_layout
= mddev
->layout
;
4376 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
4380 static struct md_sysfs_entry md_reshape_position
=
4381 __ATTR(reshape_position
, S_IRUGO
|S_IWUSR
, reshape_position_show
,
4382 reshape_position_store
);
4385 array_size_show(mddev_t
*mddev
, char *page
)
4387 if (mddev
->external_size
)
4388 return sprintf(page
, "%llu\n",
4389 (unsigned long long)mddev
->array_sectors
/2);
4391 return sprintf(page
, "default\n");
4395 array_size_store(mddev_t
*mddev
, const char *buf
, size_t len
)
4399 if (strncmp(buf
, "default", 7) == 0) {
4401 sectors
= mddev
->pers
->size(mddev
, 0, 0);
4403 sectors
= mddev
->array_sectors
;
4405 mddev
->external_size
= 0;
4407 if (strict_blocks_to_sectors(buf
, §ors
) < 0)
4409 if (mddev
->pers
&& mddev
->pers
->size(mddev
, 0, 0) < sectors
)
4412 mddev
->external_size
= 1;
4415 mddev
->array_sectors
= sectors
;
4417 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4418 revalidate_disk(mddev
->gendisk
);
4423 static struct md_sysfs_entry md_array_size
=
4424 __ATTR(array_size
, S_IRUGO
|S_IWUSR
, array_size_show
,
4427 static struct attribute
*md_default_attrs
[] = {
4430 &md_raid_disks
.attr
,
4431 &md_chunk_size
.attr
,
4433 &md_resync_start
.attr
,
4435 &md_new_device
.attr
,
4436 &md_safe_delay
.attr
,
4437 &md_array_state
.attr
,
4438 &md_reshape_position
.attr
,
4439 &md_array_size
.attr
,
4440 &max_corr_read_errors
.attr
,
4444 static struct attribute
*md_redundancy_attrs
[] = {
4446 &md_mismatches
.attr
,
4449 &md_sync_speed
.attr
,
4450 &md_sync_force_parallel
.attr
,
4451 &md_sync_completed
.attr
,
4454 &md_suspend_lo
.attr
,
4455 &md_suspend_hi
.attr
,
4460 static struct attribute_group md_redundancy_group
= {
4462 .attrs
= md_redundancy_attrs
,
4467 md_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
4469 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4470 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4475 rv
= mddev_lock(mddev
);
4477 rv
= entry
->show(mddev
, page
);
4478 mddev_unlock(mddev
);
4484 md_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
4485 const char *page
, size_t length
)
4487 struct md_sysfs_entry
*entry
= container_of(attr
, struct md_sysfs_entry
, attr
);
4488 mddev_t
*mddev
= container_of(kobj
, struct mddev_s
, kobj
);
4493 if (!capable(CAP_SYS_ADMIN
))
4495 rv
= mddev_lock(mddev
);
4496 if (mddev
->hold_active
== UNTIL_IOCTL
)
4497 mddev
->hold_active
= 0;
4499 rv
= entry
->store(mddev
, page
, length
);
4500 mddev_unlock(mddev
);
4505 static void md_free(struct kobject
*ko
)
4507 mddev_t
*mddev
= container_of(ko
, mddev_t
, kobj
);
4509 if (mddev
->sysfs_state
)
4510 sysfs_put(mddev
->sysfs_state
);
4512 if (mddev
->gendisk
) {
4513 del_gendisk(mddev
->gendisk
);
4514 put_disk(mddev
->gendisk
);
4517 blk_cleanup_queue(mddev
->queue
);
4522 static const struct sysfs_ops md_sysfs_ops
= {
4523 .show
= md_attr_show
,
4524 .store
= md_attr_store
,
4526 static struct kobj_type md_ktype
= {
4528 .sysfs_ops
= &md_sysfs_ops
,
4529 .default_attrs
= md_default_attrs
,
4534 static void mddev_delayed_delete(struct work_struct
*ws
)
4536 mddev_t
*mddev
= container_of(ws
, mddev_t
, del_work
);
4538 sysfs_remove_group(&mddev
->kobj
, &md_bitmap_group
);
4539 kobject_del(&mddev
->kobj
);
4540 kobject_put(&mddev
->kobj
);
4543 static int md_alloc(dev_t dev
, char *name
)
4545 static DEFINE_MUTEX(disks_mutex
);
4546 mddev_t
*mddev
= mddev_find(dev
);
4547 struct gendisk
*disk
;
4556 partitioned
= (MAJOR(mddev
->unit
) != MD_MAJOR
);
4557 shift
= partitioned
? MdpMinorShift
: 0;
4558 unit
= MINOR(mddev
->unit
) >> shift
;
4560 /* wait for any previous instance of this device to be
4561 * completely removed (mddev_delayed_delete).
4563 flush_workqueue(md_misc_wq
);
4565 mutex_lock(&disks_mutex
);
4571 /* Need to ensure that 'name' is not a duplicate.
4574 spin_lock(&all_mddevs_lock
);
4576 list_for_each_entry(mddev2
, &all_mddevs
, all_mddevs
)
4577 if (mddev2
->gendisk
&&
4578 strcmp(mddev2
->gendisk
->disk_name
, name
) == 0) {
4579 spin_unlock(&all_mddevs_lock
);
4582 spin_unlock(&all_mddevs_lock
);
4586 mddev
->queue
= blk_alloc_queue(GFP_KERNEL
);
4589 mddev
->queue
->queuedata
= mddev
;
4591 blk_queue_make_request(mddev
->queue
, md_make_request
);
4593 disk
= alloc_disk(1 << shift
);
4595 blk_cleanup_queue(mddev
->queue
);
4596 mddev
->queue
= NULL
;
4599 disk
->major
= MAJOR(mddev
->unit
);
4600 disk
->first_minor
= unit
<< shift
;
4602 strcpy(disk
->disk_name
, name
);
4603 else if (partitioned
)
4604 sprintf(disk
->disk_name
, "md_d%d", unit
);
4606 sprintf(disk
->disk_name
, "md%d", unit
);
4607 disk
->fops
= &md_fops
;
4608 disk
->private_data
= mddev
;
4609 disk
->queue
= mddev
->queue
;
4610 blk_queue_flush(mddev
->queue
, REQ_FLUSH
| REQ_FUA
);
4611 /* Allow extended partitions. This makes the
4612 * 'mdp' device redundant, but we can't really
4615 disk
->flags
|= GENHD_FL_EXT_DEVT
;
4616 mddev
->gendisk
= disk
;
4617 /* As soon as we call add_disk(), another thread could get
4618 * through to md_open, so make sure it doesn't get too far
4620 mutex_lock(&mddev
->open_mutex
);
4623 error
= kobject_init_and_add(&mddev
->kobj
, &md_ktype
,
4624 &disk_to_dev(disk
)->kobj
, "%s", "md");
4626 /* This isn't possible, but as kobject_init_and_add is marked
4627 * __must_check, we must do something with the result
4629 printk(KERN_WARNING
"md: cannot register %s/md - name in use\n",
4633 if (mddev
->kobj
.sd
&&
4634 sysfs_create_group(&mddev
->kobj
, &md_bitmap_group
))
4635 printk(KERN_DEBUG
"pointless warning\n");
4636 mutex_unlock(&mddev
->open_mutex
);
4638 mutex_unlock(&disks_mutex
);
4639 if (!error
&& mddev
->kobj
.sd
) {
4640 kobject_uevent(&mddev
->kobj
, KOBJ_ADD
);
4641 mddev
->sysfs_state
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "array_state");
4647 static struct kobject
*md_probe(dev_t dev
, int *part
, void *data
)
4649 md_alloc(dev
, NULL
);
4653 static int add_named_array(const char *val
, struct kernel_param
*kp
)
4655 /* val must be "md_*" where * is not all digits.
4656 * We allocate an array with a large free minor number, and
4657 * set the name to val. val must not already be an active name.
4659 int len
= strlen(val
);
4660 char buf
[DISK_NAME_LEN
];
4662 while (len
&& val
[len
-1] == '\n')
4664 if (len
>= DISK_NAME_LEN
)
4666 strlcpy(buf
, val
, len
+1);
4667 if (strncmp(buf
, "md_", 3) != 0)
4669 return md_alloc(0, buf
);
4672 static void md_safemode_timeout(unsigned long data
)
4674 mddev_t
*mddev
= (mddev_t
*) data
;
4676 if (!atomic_read(&mddev
->writes_pending
)) {
4677 mddev
->safemode
= 1;
4678 if (mddev
->external
)
4679 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4681 md_wakeup_thread(mddev
->thread
);
4684 static int start_dirty_degraded
;
4686 int md_run(mddev_t
*mddev
)
4690 struct mdk_personality
*pers
;
4692 if (list_empty(&mddev
->disks
))
4693 /* cannot run an array with no devices.. */
4698 /* Cannot run until previous stop completes properly */
4699 if (mddev
->sysfs_active
)
4703 * Analyze all RAID superblock(s)
4705 if (!mddev
->raid_disks
) {
4706 if (!mddev
->persistent
)
4711 if (mddev
->level
!= LEVEL_NONE
)
4712 request_module("md-level-%d", mddev
->level
);
4713 else if (mddev
->clevel
[0])
4714 request_module("md-%s", mddev
->clevel
);
4717 * Drop all container device buffers, from now on
4718 * the only valid external interface is through the md
4721 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
4722 if (test_bit(Faulty
, &rdev
->flags
))
4724 sync_blockdev(rdev
->bdev
);
4725 invalidate_bdev(rdev
->bdev
);
4727 /* perform some consistency tests on the device.
4728 * We don't want the data to overlap the metadata,
4729 * Internal Bitmap issues have been handled elsewhere.
4731 if (rdev
->meta_bdev
) {
4732 /* Nothing to check */;
4733 } else if (rdev
->data_offset
< rdev
->sb_start
) {
4734 if (mddev
->dev_sectors
&&
4735 rdev
->data_offset
+ mddev
->dev_sectors
4737 printk("md: %s: data overlaps metadata\n",
4742 if (rdev
->sb_start
+ rdev
->sb_size
/512
4743 > rdev
->data_offset
) {
4744 printk("md: %s: metadata overlaps data\n",
4749 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
4752 if (mddev
->bio_set
== NULL
)
4753 mddev
->bio_set
= bioset_create(BIO_POOL_SIZE
,
4756 spin_lock(&pers_lock
);
4757 pers
= find_pers(mddev
->level
, mddev
->clevel
);
4758 if (!pers
|| !try_module_get(pers
->owner
)) {
4759 spin_unlock(&pers_lock
);
4760 if (mddev
->level
!= LEVEL_NONE
)
4761 printk(KERN_WARNING
"md: personality for level %d is not loaded!\n",
4764 printk(KERN_WARNING
"md: personality for level %s is not loaded!\n",
4769 spin_unlock(&pers_lock
);
4770 if (mddev
->level
!= pers
->level
) {
4771 mddev
->level
= pers
->level
;
4772 mddev
->new_level
= pers
->level
;
4774 strlcpy(mddev
->clevel
, pers
->name
, sizeof(mddev
->clevel
));
4776 if (mddev
->reshape_position
!= MaxSector
&&
4777 pers
->start_reshape
== NULL
) {
4778 /* This personality cannot handle reshaping... */
4780 module_put(pers
->owner
);
4784 if (pers
->sync_request
) {
4785 /* Warn if this is a potentially silly
4788 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
4792 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4793 list_for_each_entry(rdev2
, &mddev
->disks
, same_set
) {
4795 rdev
->bdev
->bd_contains
==
4796 rdev2
->bdev
->bd_contains
) {
4798 "%s: WARNING: %s appears to be"
4799 " on the same physical disk as"
4802 bdevname(rdev
->bdev
,b
),
4803 bdevname(rdev2
->bdev
,b2
));
4810 "True protection against single-disk"
4811 " failure might be compromised.\n");
4814 mddev
->recovery
= 0;
4815 /* may be over-ridden by personality */
4816 mddev
->resync_max_sectors
= mddev
->dev_sectors
;
4818 mddev
->ok_start_degraded
= start_dirty_degraded
;
4820 if (start_readonly
&& mddev
->ro
== 0)
4821 mddev
->ro
= 2; /* read-only, but switch on first write */
4823 err
= mddev
->pers
->run(mddev
);
4825 printk(KERN_ERR
"md: pers->run() failed ...\n");
4826 else if (mddev
->pers
->size(mddev
, 0, 0) < mddev
->array_sectors
) {
4827 WARN_ONCE(!mddev
->external_size
, "%s: default size too small,"
4828 " but 'external_size' not in effect?\n", __func__
);
4830 "md: invalid array_size %llu > default size %llu\n",
4831 (unsigned long long)mddev
->array_sectors
/ 2,
4832 (unsigned long long)mddev
->pers
->size(mddev
, 0, 0) / 2);
4834 mddev
->pers
->stop(mddev
);
4836 if (err
== 0 && mddev
->pers
->sync_request
) {
4837 err
= bitmap_create(mddev
);
4839 printk(KERN_ERR
"%s: failed to create bitmap (%d)\n",
4840 mdname(mddev
), err
);
4841 mddev
->pers
->stop(mddev
);
4845 module_put(mddev
->pers
->owner
);
4847 bitmap_destroy(mddev
);
4850 if (mddev
->pers
->sync_request
) {
4851 if (mddev
->kobj
.sd
&&
4852 sysfs_create_group(&mddev
->kobj
, &md_redundancy_group
))
4854 "md: cannot register extra attributes for %s\n",
4856 mddev
->sysfs_action
= sysfs_get_dirent_safe(mddev
->kobj
.sd
, "sync_action");
4857 } else if (mddev
->ro
== 2) /* auto-readonly not meaningful */
4860 atomic_set(&mddev
->writes_pending
,0);
4861 atomic_set(&mddev
->max_corr_read_errors
,
4862 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS
);
4863 mddev
->safemode
= 0;
4864 mddev
->safemode_timer
.function
= md_safemode_timeout
;
4865 mddev
->safemode_timer
.data
= (unsigned long) mddev
;
4866 mddev
->safemode_delay
= (200 * HZ
)/1000 +1; /* 200 msec delay */
4870 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
4871 if (rdev
->raid_disk
>= 0)
4872 if (sysfs_link_rdev(mddev
, rdev
))
4873 /* failure here is OK */;
4875 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4878 md_update_sb(mddev
, 0);
4880 md_new_event(mddev
);
4881 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4882 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
4883 sysfs_notify(&mddev
->kobj
, NULL
, "degraded");
4886 EXPORT_SYMBOL_GPL(md_run
);
4888 static int do_md_run(mddev_t
*mddev
)
4892 err
= md_run(mddev
);
4895 err
= bitmap_load(mddev
);
4897 bitmap_destroy(mddev
);
4901 md_wakeup_thread(mddev
->thread
);
4902 md_wakeup_thread(mddev
->sync_thread
); /* possibly kick off a reshape */
4904 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
4905 revalidate_disk(mddev
->gendisk
);
4907 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
4912 static int restart_array(mddev_t
*mddev
)
4914 struct gendisk
*disk
= mddev
->gendisk
;
4916 /* Complain if it has no devices */
4917 if (list_empty(&mddev
->disks
))
4923 mddev
->safemode
= 0;
4925 set_disk_ro(disk
, 0);
4926 printk(KERN_INFO
"md: %s switched to read-write mode.\n",
4928 /* Kick recovery or resync if necessary */
4929 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
4930 md_wakeup_thread(mddev
->thread
);
4931 md_wakeup_thread(mddev
->sync_thread
);
4932 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
4936 /* similar to deny_write_access, but accounts for our holding a reference
4937 * to the file ourselves */
4938 static int deny_bitmap_write_access(struct file
* file
)
4940 struct inode
*inode
= file
->f_mapping
->host
;
4942 spin_lock(&inode
->i_lock
);
4943 if (atomic_read(&inode
->i_writecount
) > 1) {
4944 spin_unlock(&inode
->i_lock
);
4947 atomic_set(&inode
->i_writecount
, -1);
4948 spin_unlock(&inode
->i_lock
);
4953 void restore_bitmap_write_access(struct file
*file
)
4955 struct inode
*inode
= file
->f_mapping
->host
;
4957 spin_lock(&inode
->i_lock
);
4958 atomic_set(&inode
->i_writecount
, 1);
4959 spin_unlock(&inode
->i_lock
);
4962 static void md_clean(mddev_t
*mddev
)
4964 mddev
->array_sectors
= 0;
4965 mddev
->external_size
= 0;
4966 mddev
->dev_sectors
= 0;
4967 mddev
->raid_disks
= 0;
4968 mddev
->recovery_cp
= 0;
4969 mddev
->resync_min
= 0;
4970 mddev
->resync_max
= MaxSector
;
4971 mddev
->reshape_position
= MaxSector
;
4972 mddev
->external
= 0;
4973 mddev
->persistent
= 0;
4974 mddev
->level
= LEVEL_NONE
;
4975 mddev
->clevel
[0] = 0;
4978 mddev
->metadata_type
[0] = 0;
4979 mddev
->chunk_sectors
= 0;
4980 mddev
->ctime
= mddev
->utime
= 0;
4982 mddev
->max_disks
= 0;
4984 mddev
->can_decrease_events
= 0;
4985 mddev
->delta_disks
= 0;
4986 mddev
->new_level
= LEVEL_NONE
;
4987 mddev
->new_layout
= 0;
4988 mddev
->new_chunk_sectors
= 0;
4989 mddev
->curr_resync
= 0;
4990 mddev
->resync_mismatches
= 0;
4991 mddev
->suspend_lo
= mddev
->suspend_hi
= 0;
4992 mddev
->sync_speed_min
= mddev
->sync_speed_max
= 0;
4993 mddev
->recovery
= 0;
4996 mddev
->degraded
= 0;
4997 mddev
->safemode
= 0;
4998 mddev
->bitmap_info
.offset
= 0;
4999 mddev
->bitmap_info
.default_offset
= 0;
5000 mddev
->bitmap_info
.chunksize
= 0;
5001 mddev
->bitmap_info
.daemon_sleep
= 0;
5002 mddev
->bitmap_info
.max_write_behind
= 0;
5005 static void __md_stop_writes(mddev_t
*mddev
)
5007 if (mddev
->sync_thread
) {
5008 set_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5009 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
5010 reap_sync_thread(mddev
);
5013 del_timer_sync(&mddev
->safemode_timer
);
5015 bitmap_flush(mddev
);
5016 md_super_wait(mddev
);
5018 if (!mddev
->in_sync
|| mddev
->flags
) {
5019 /* mark array as shutdown cleanly */
5021 md_update_sb(mddev
, 1);
5025 void md_stop_writes(mddev_t
*mddev
)
5028 __md_stop_writes(mddev
);
5029 mddev_unlock(mddev
);
5031 EXPORT_SYMBOL_GPL(md_stop_writes
);
5033 void md_stop(mddev_t
*mddev
)
5036 mddev
->pers
->stop(mddev
);
5037 if (mddev
->pers
->sync_request
&& mddev
->to_remove
== NULL
)
5038 mddev
->to_remove
= &md_redundancy_group
;
5039 module_put(mddev
->pers
->owner
);
5041 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5043 EXPORT_SYMBOL_GPL(md_stop
);
5045 static int md_set_readonly(mddev_t
*mddev
, int is_open
)
5048 mutex_lock(&mddev
->open_mutex
);
5049 if (atomic_read(&mddev
->openers
) > is_open
) {
5050 printk("md: %s still in use.\n",mdname(mddev
));
5055 __md_stop_writes(mddev
);
5061 set_disk_ro(mddev
->gendisk
, 1);
5062 clear_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
);
5063 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5067 mutex_unlock(&mddev
->open_mutex
);
5072 * 0 - completely stop and dis-assemble array
5073 * 2 - stop but do not disassemble array
5075 static int do_md_stop(mddev_t
* mddev
, int mode
, int is_open
)
5077 struct gendisk
*disk
= mddev
->gendisk
;
5080 mutex_lock(&mddev
->open_mutex
);
5081 if (atomic_read(&mddev
->openers
) > is_open
||
5082 mddev
->sysfs_active
) {
5083 printk("md: %s still in use.\n",mdname(mddev
));
5084 mutex_unlock(&mddev
->open_mutex
);
5090 set_disk_ro(disk
, 0);
5092 __md_stop_writes(mddev
);
5094 mddev
->queue
->merge_bvec_fn
= NULL
;
5095 mddev
->queue
->backing_dev_info
.congested_fn
= NULL
;
5097 /* tell userspace to handle 'inactive' */
5098 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5100 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
5101 if (rdev
->raid_disk
>= 0)
5102 sysfs_unlink_rdev(mddev
, rdev
);
5104 set_capacity(disk
, 0);
5105 mutex_unlock(&mddev
->open_mutex
);
5107 revalidate_disk(disk
);
5112 mutex_unlock(&mddev
->open_mutex
);
5114 * Free resources if final stop
5117 printk(KERN_INFO
"md: %s stopped.\n", mdname(mddev
));
5119 bitmap_destroy(mddev
);
5120 if (mddev
->bitmap_info
.file
) {
5121 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5122 fput(mddev
->bitmap_info
.file
);
5123 mddev
->bitmap_info
.file
= NULL
;
5125 mddev
->bitmap_info
.offset
= 0;
5127 export_array(mddev
);
5130 kobject_uevent(&disk_to_dev(mddev
->gendisk
)->kobj
, KOBJ_CHANGE
);
5131 if (mddev
->hold_active
== UNTIL_STOP
)
5132 mddev
->hold_active
= 0;
5134 blk_integrity_unregister(disk
);
5135 md_new_event(mddev
);
5136 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
5141 static void autorun_array(mddev_t
*mddev
)
5146 if (list_empty(&mddev
->disks
))
5149 printk(KERN_INFO
"md: running: ");
5151 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5152 char b
[BDEVNAME_SIZE
];
5153 printk("<%s>", bdevname(rdev
->bdev
,b
));
5157 err
= do_md_run(mddev
);
5159 printk(KERN_WARNING
"md: do_md_run() returned %d\n", err
);
5160 do_md_stop(mddev
, 0, 0);
5165 * lets try to run arrays based on all disks that have arrived
5166 * until now. (those are in pending_raid_disks)
5168 * the method: pick the first pending disk, collect all disks with
5169 * the same UUID, remove all from the pending list and put them into
5170 * the 'same_array' list. Then order this list based on superblock
5171 * update time (freshest comes first), kick out 'old' disks and
5172 * compare superblocks. If everything's fine then run it.
5174 * If "unit" is allocated, then bump its reference count
5176 static void autorun_devices(int part
)
5178 mdk_rdev_t
*rdev0
, *rdev
, *tmp
;
5180 char b
[BDEVNAME_SIZE
];
5182 printk(KERN_INFO
"md: autorun ...\n");
5183 while (!list_empty(&pending_raid_disks
)) {
5186 LIST_HEAD(candidates
);
5187 rdev0
= list_entry(pending_raid_disks
.next
,
5188 mdk_rdev_t
, same_set
);
5190 printk(KERN_INFO
"md: considering %s ...\n",
5191 bdevname(rdev0
->bdev
,b
));
5192 INIT_LIST_HEAD(&candidates
);
5193 rdev_for_each_list(rdev
, tmp
, &pending_raid_disks
)
5194 if (super_90_load(rdev
, rdev0
, 0) >= 0) {
5195 printk(KERN_INFO
"md: adding %s ...\n",
5196 bdevname(rdev
->bdev
,b
));
5197 list_move(&rdev
->same_set
, &candidates
);
5200 * now we have a set of devices, with all of them having
5201 * mostly sane superblocks. It's time to allocate the
5205 dev
= MKDEV(mdp_major
,
5206 rdev0
->preferred_minor
<< MdpMinorShift
);
5207 unit
= MINOR(dev
) >> MdpMinorShift
;
5209 dev
= MKDEV(MD_MAJOR
, rdev0
->preferred_minor
);
5212 if (rdev0
->preferred_minor
!= unit
) {
5213 printk(KERN_INFO
"md: unit number in %s is bad: %d\n",
5214 bdevname(rdev0
->bdev
, b
), rdev0
->preferred_minor
);
5218 md_probe(dev
, NULL
, NULL
);
5219 mddev
= mddev_find(dev
);
5220 if (!mddev
|| !mddev
->gendisk
) {
5224 "md: cannot allocate memory for md drive.\n");
5227 if (mddev_lock(mddev
))
5228 printk(KERN_WARNING
"md: %s locked, cannot run\n",
5230 else if (mddev
->raid_disks
|| mddev
->major_version
5231 || !list_empty(&mddev
->disks
)) {
5233 "md: %s already running, cannot run %s\n",
5234 mdname(mddev
), bdevname(rdev0
->bdev
,b
));
5235 mddev_unlock(mddev
);
5237 printk(KERN_INFO
"md: created %s\n", mdname(mddev
));
5238 mddev
->persistent
= 1;
5239 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5240 list_del_init(&rdev
->same_set
);
5241 if (bind_rdev_to_array(rdev
, mddev
))
5244 autorun_array(mddev
);
5245 mddev_unlock(mddev
);
5247 /* on success, candidates will be empty, on error
5250 rdev_for_each_list(rdev
, tmp
, &candidates
) {
5251 list_del_init(&rdev
->same_set
);
5256 printk(KERN_INFO
"md: ... autorun DONE.\n");
5258 #endif /* !MODULE */
5260 static int get_version(void __user
* arg
)
5264 ver
.major
= MD_MAJOR_VERSION
;
5265 ver
.minor
= MD_MINOR_VERSION
;
5266 ver
.patchlevel
= MD_PATCHLEVEL_VERSION
;
5268 if (copy_to_user(arg
, &ver
, sizeof(ver
)))
5274 static int get_array_info(mddev_t
* mddev
, void __user
* arg
)
5276 mdu_array_info_t info
;
5277 int nr
,working
,insync
,failed
,spare
;
5280 nr
=working
=insync
=failed
=spare
=0;
5281 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5283 if (test_bit(Faulty
, &rdev
->flags
))
5287 if (test_bit(In_sync
, &rdev
->flags
))
5294 info
.major_version
= mddev
->major_version
;
5295 info
.minor_version
= mddev
->minor_version
;
5296 info
.patch_version
= MD_PATCHLEVEL_VERSION
;
5297 info
.ctime
= mddev
->ctime
;
5298 info
.level
= mddev
->level
;
5299 info
.size
= mddev
->dev_sectors
/ 2;
5300 if (info
.size
!= mddev
->dev_sectors
/ 2) /* overflow */
5303 info
.raid_disks
= mddev
->raid_disks
;
5304 info
.md_minor
= mddev
->md_minor
;
5305 info
.not_persistent
= !mddev
->persistent
;
5307 info
.utime
= mddev
->utime
;
5310 info
.state
= (1<<MD_SB_CLEAN
);
5311 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5312 info
.state
= (1<<MD_SB_BITMAP_PRESENT
);
5313 info
.active_disks
= insync
;
5314 info
.working_disks
= working
;
5315 info
.failed_disks
= failed
;
5316 info
.spare_disks
= spare
;
5318 info
.layout
= mddev
->layout
;
5319 info
.chunk_size
= mddev
->chunk_sectors
<< 9;
5321 if (copy_to_user(arg
, &info
, sizeof(info
)))
5327 static int get_bitmap_file(mddev_t
* mddev
, void __user
* arg
)
5329 mdu_bitmap_file_t
*file
= NULL
; /* too big for stack allocation */
5330 char *ptr
, *buf
= NULL
;
5333 if (md_allow_write(mddev
))
5334 file
= kmalloc(sizeof(*file
), GFP_NOIO
);
5336 file
= kmalloc(sizeof(*file
), GFP_KERNEL
);
5341 /* bitmap disabled, zero the first byte and copy out */
5342 if (!mddev
->bitmap
|| !mddev
->bitmap
->file
) {
5343 file
->pathname
[0] = '\0';
5347 buf
= kmalloc(sizeof(file
->pathname
), GFP_KERNEL
);
5351 ptr
= d_path(&mddev
->bitmap
->file
->f_path
, buf
, sizeof(file
->pathname
));
5355 strcpy(file
->pathname
, ptr
);
5359 if (copy_to_user(arg
, file
, sizeof(*file
)))
5367 static int get_disk_info(mddev_t
* mddev
, void __user
* arg
)
5369 mdu_disk_info_t info
;
5372 if (copy_from_user(&info
, arg
, sizeof(info
)))
5375 rdev
= find_rdev_nr(mddev
, info
.number
);
5377 info
.major
= MAJOR(rdev
->bdev
->bd_dev
);
5378 info
.minor
= MINOR(rdev
->bdev
->bd_dev
);
5379 info
.raid_disk
= rdev
->raid_disk
;
5381 if (test_bit(Faulty
, &rdev
->flags
))
5382 info
.state
|= (1<<MD_DISK_FAULTY
);
5383 else if (test_bit(In_sync
, &rdev
->flags
)) {
5384 info
.state
|= (1<<MD_DISK_ACTIVE
);
5385 info
.state
|= (1<<MD_DISK_SYNC
);
5387 if (test_bit(WriteMostly
, &rdev
->flags
))
5388 info
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
5390 info
.major
= info
.minor
= 0;
5391 info
.raid_disk
= -1;
5392 info
.state
= (1<<MD_DISK_REMOVED
);
5395 if (copy_to_user(arg
, &info
, sizeof(info
)))
5401 static int add_new_disk(mddev_t
* mddev
, mdu_disk_info_t
*info
)
5403 char b
[BDEVNAME_SIZE
], b2
[BDEVNAME_SIZE
];
5405 dev_t dev
= MKDEV(info
->major
,info
->minor
);
5407 if (info
->major
!= MAJOR(dev
) || info
->minor
!= MINOR(dev
))
5410 if (!mddev
->raid_disks
) {
5412 /* expecting a device which has a superblock */
5413 rdev
= md_import_device(dev
, mddev
->major_version
, mddev
->minor_version
);
5416 "md: md_import_device returned %ld\n",
5418 return PTR_ERR(rdev
);
5420 if (!list_empty(&mddev
->disks
)) {
5421 mdk_rdev_t
*rdev0
= list_entry(mddev
->disks
.next
,
5422 mdk_rdev_t
, same_set
);
5423 err
= super_types
[mddev
->major_version
]
5424 .load_super(rdev
, rdev0
, mddev
->minor_version
);
5427 "md: %s has different UUID to %s\n",
5428 bdevname(rdev
->bdev
,b
),
5429 bdevname(rdev0
->bdev
,b2
));
5434 err
= bind_rdev_to_array(rdev
, mddev
);
5441 * add_new_disk can be used once the array is assembled
5442 * to add "hot spares". They must already have a superblock
5447 if (!mddev
->pers
->hot_add_disk
) {
5449 "%s: personality does not support diskops!\n",
5453 if (mddev
->persistent
)
5454 rdev
= md_import_device(dev
, mddev
->major_version
,
5455 mddev
->minor_version
);
5457 rdev
= md_import_device(dev
, -1, -1);
5460 "md: md_import_device returned %ld\n",
5462 return PTR_ERR(rdev
);
5464 /* set saved_raid_disk if appropriate */
5465 if (!mddev
->persistent
) {
5466 if (info
->state
& (1<<MD_DISK_SYNC
) &&
5467 info
->raid_disk
< mddev
->raid_disks
) {
5468 rdev
->raid_disk
= info
->raid_disk
;
5469 set_bit(In_sync
, &rdev
->flags
);
5471 rdev
->raid_disk
= -1;
5473 super_types
[mddev
->major_version
].
5474 validate_super(mddev
, rdev
);
5475 if ((info
->state
& (1<<MD_DISK_SYNC
)) &&
5476 (!test_bit(In_sync
, &rdev
->flags
) ||
5477 rdev
->raid_disk
!= info
->raid_disk
)) {
5478 /* This was a hot-add request, but events doesn't
5479 * match, so reject it.
5485 if (test_bit(In_sync
, &rdev
->flags
))
5486 rdev
->saved_raid_disk
= rdev
->raid_disk
;
5488 rdev
->saved_raid_disk
= -1;
5490 clear_bit(In_sync
, &rdev
->flags
); /* just to be sure */
5491 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5492 set_bit(WriteMostly
, &rdev
->flags
);
5494 clear_bit(WriteMostly
, &rdev
->flags
);
5496 rdev
->raid_disk
= -1;
5497 err
= bind_rdev_to_array(rdev
, mddev
);
5498 if (!err
&& !mddev
->pers
->hot_remove_disk
) {
5499 /* If there is hot_add_disk but no hot_remove_disk
5500 * then added disks for geometry changes,
5501 * and should be added immediately.
5503 super_types
[mddev
->major_version
].
5504 validate_super(mddev
, rdev
);
5505 err
= mddev
->pers
->hot_add_disk(mddev
, rdev
);
5507 unbind_rdev_from_array(rdev
);
5512 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
5514 md_update_sb(mddev
, 1);
5515 if (mddev
->degraded
)
5516 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
5517 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5519 md_new_event(mddev
);
5520 md_wakeup_thread(mddev
->thread
);
5524 /* otherwise, add_new_disk is only allowed
5525 * for major_version==0 superblocks
5527 if (mddev
->major_version
!= 0) {
5528 printk(KERN_WARNING
"%s: ADD_NEW_DISK not supported\n",
5533 if (!(info
->state
& (1<<MD_DISK_FAULTY
))) {
5535 rdev
= md_import_device(dev
, -1, 0);
5538 "md: error, md_import_device() returned %ld\n",
5540 return PTR_ERR(rdev
);
5542 rdev
->desc_nr
= info
->number
;
5543 if (info
->raid_disk
< mddev
->raid_disks
)
5544 rdev
->raid_disk
= info
->raid_disk
;
5546 rdev
->raid_disk
= -1;
5548 if (rdev
->raid_disk
< mddev
->raid_disks
)
5549 if (info
->state
& (1<<MD_DISK_SYNC
))
5550 set_bit(In_sync
, &rdev
->flags
);
5552 if (info
->state
& (1<<MD_DISK_WRITEMOSTLY
))
5553 set_bit(WriteMostly
, &rdev
->flags
);
5555 if (!mddev
->persistent
) {
5556 printk(KERN_INFO
"md: nonpersistent superblock ...\n");
5557 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5559 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5560 rdev
->sectors
= rdev
->sb_start
;
5562 err
= bind_rdev_to_array(rdev
, mddev
);
5572 static int hot_remove_disk(mddev_t
* mddev
, dev_t dev
)
5574 char b
[BDEVNAME_SIZE
];
5577 rdev
= find_rdev(mddev
, dev
);
5581 if (rdev
->raid_disk
>= 0)
5584 kick_rdev_from_array(rdev
);
5585 md_update_sb(mddev
, 1);
5586 md_new_event(mddev
);
5590 printk(KERN_WARNING
"md: cannot remove active disk %s from %s ...\n",
5591 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5595 static int hot_add_disk(mddev_t
* mddev
, dev_t dev
)
5597 char b
[BDEVNAME_SIZE
];
5604 if (mddev
->major_version
!= 0) {
5605 printk(KERN_WARNING
"%s: HOT_ADD may only be used with"
5606 " version-0 superblocks.\n",
5610 if (!mddev
->pers
->hot_add_disk
) {
5612 "%s: personality does not support diskops!\n",
5617 rdev
= md_import_device(dev
, -1, 0);
5620 "md: error, md_import_device() returned %ld\n",
5625 if (mddev
->persistent
)
5626 rdev
->sb_start
= calc_dev_sboffset(rdev
);
5628 rdev
->sb_start
= i_size_read(rdev
->bdev
->bd_inode
) / 512;
5630 rdev
->sectors
= rdev
->sb_start
;
5632 if (test_bit(Faulty
, &rdev
->flags
)) {
5634 "md: can not hot-add faulty %s disk to %s!\n",
5635 bdevname(rdev
->bdev
,b
), mdname(mddev
));
5639 clear_bit(In_sync
, &rdev
->flags
);
5641 rdev
->saved_raid_disk
= -1;
5642 err
= bind_rdev_to_array(rdev
, mddev
);
5647 * The rest should better be atomic, we can have disk failures
5648 * noticed in interrupt contexts ...
5651 rdev
->raid_disk
= -1;
5653 md_update_sb(mddev
, 1);
5656 * Kick recovery, maybe this spare has to be added to the
5657 * array immediately.
5659 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
5660 md_wakeup_thread(mddev
->thread
);
5661 md_new_event(mddev
);
5669 static int set_bitmap_file(mddev_t
*mddev
, int fd
)
5674 if (!mddev
->pers
->quiesce
)
5676 if (mddev
->recovery
|| mddev
->sync_thread
)
5678 /* we should be able to change the bitmap.. */
5684 return -EEXIST
; /* cannot add when bitmap is present */
5685 mddev
->bitmap_info
.file
= fget(fd
);
5687 if (mddev
->bitmap_info
.file
== NULL
) {
5688 printk(KERN_ERR
"%s: error: failed to get bitmap file\n",
5693 err
= deny_bitmap_write_access(mddev
->bitmap_info
.file
);
5695 printk(KERN_ERR
"%s: error: bitmap file is already in use\n",
5697 fput(mddev
->bitmap_info
.file
);
5698 mddev
->bitmap_info
.file
= NULL
;
5701 mddev
->bitmap_info
.offset
= 0; /* file overrides offset */
5702 } else if (mddev
->bitmap
== NULL
)
5703 return -ENOENT
; /* cannot remove what isn't there */
5706 mddev
->pers
->quiesce(mddev
, 1);
5708 err
= bitmap_create(mddev
);
5710 err
= bitmap_load(mddev
);
5712 if (fd
< 0 || err
) {
5713 bitmap_destroy(mddev
);
5714 fd
= -1; /* make sure to put the file */
5716 mddev
->pers
->quiesce(mddev
, 0);
5719 if (mddev
->bitmap_info
.file
) {
5720 restore_bitmap_write_access(mddev
->bitmap_info
.file
);
5721 fput(mddev
->bitmap_info
.file
);
5723 mddev
->bitmap_info
.file
= NULL
;
5730 * set_array_info is used two different ways
5731 * The original usage is when creating a new array.
5732 * In this usage, raid_disks is > 0 and it together with
5733 * level, size, not_persistent,layout,chunksize determine the
5734 * shape of the array.
5735 * This will always create an array with a type-0.90.0 superblock.
5736 * The newer usage is when assembling an array.
5737 * In this case raid_disks will be 0, and the major_version field is
5738 * use to determine which style super-blocks are to be found on the devices.
5739 * The minor and patch _version numbers are also kept incase the
5740 * super_block handler wishes to interpret them.
5742 static int set_array_info(mddev_t
* mddev
, mdu_array_info_t
*info
)
5745 if (info
->raid_disks
== 0) {
5746 /* just setting version number for superblock loading */
5747 if (info
->major_version
< 0 ||
5748 info
->major_version
>= ARRAY_SIZE(super_types
) ||
5749 super_types
[info
->major_version
].name
== NULL
) {
5750 /* maybe try to auto-load a module? */
5752 "md: superblock version %d not known\n",
5753 info
->major_version
);
5756 mddev
->major_version
= info
->major_version
;
5757 mddev
->minor_version
= info
->minor_version
;
5758 mddev
->patch_version
= info
->patch_version
;
5759 mddev
->persistent
= !info
->not_persistent
;
5760 /* ensure mddev_put doesn't delete this now that there
5761 * is some minimal configuration.
5763 mddev
->ctime
= get_seconds();
5766 mddev
->major_version
= MD_MAJOR_VERSION
;
5767 mddev
->minor_version
= MD_MINOR_VERSION
;
5768 mddev
->patch_version
= MD_PATCHLEVEL_VERSION
;
5769 mddev
->ctime
= get_seconds();
5771 mddev
->level
= info
->level
;
5772 mddev
->clevel
[0] = 0;
5773 mddev
->dev_sectors
= 2 * (sector_t
)info
->size
;
5774 mddev
->raid_disks
= info
->raid_disks
;
5775 /* don't set md_minor, it is determined by which /dev/md* was
5778 if (info
->state
& (1<<MD_SB_CLEAN
))
5779 mddev
->recovery_cp
= MaxSector
;
5781 mddev
->recovery_cp
= 0;
5782 mddev
->persistent
= ! info
->not_persistent
;
5783 mddev
->external
= 0;
5785 mddev
->layout
= info
->layout
;
5786 mddev
->chunk_sectors
= info
->chunk_size
>> 9;
5788 mddev
->max_disks
= MD_SB_DISKS
;
5790 if (mddev
->persistent
)
5792 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
5794 mddev
->bitmap_info
.default_offset
= MD_SB_BYTES
>> 9;
5795 mddev
->bitmap_info
.offset
= 0;
5797 mddev
->reshape_position
= MaxSector
;
5800 * Generate a 128 bit UUID
5802 get_random_bytes(mddev
->uuid
, 16);
5804 mddev
->new_level
= mddev
->level
;
5805 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
5806 mddev
->new_layout
= mddev
->layout
;
5807 mddev
->delta_disks
= 0;
5812 void md_set_array_sectors(mddev_t
*mddev
, sector_t array_sectors
)
5814 WARN(!mddev_is_locked(mddev
), "%s: unlocked mddev!\n", __func__
);
5816 if (mddev
->external_size
)
5819 mddev
->array_sectors
= array_sectors
;
5821 EXPORT_SYMBOL(md_set_array_sectors
);
5823 static int update_size(mddev_t
*mddev
, sector_t num_sectors
)
5827 int fit
= (num_sectors
== 0);
5829 if (mddev
->pers
->resize
== NULL
)
5831 /* The "num_sectors" is the number of sectors of each device that
5832 * is used. This can only make sense for arrays with redundancy.
5833 * linear and raid0 always use whatever space is available. We can only
5834 * consider changing this number if no resync or reconstruction is
5835 * happening, and if the new size is acceptable. It must fit before the
5836 * sb_start or, if that is <data_offset, it must fit before the size
5837 * of each device. If num_sectors is zero, we find the largest size
5840 if (mddev
->sync_thread
)
5843 /* Sorry, cannot grow a bitmap yet, just remove it,
5847 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
5848 sector_t avail
= rdev
->sectors
;
5850 if (fit
&& (num_sectors
== 0 || num_sectors
> avail
))
5851 num_sectors
= avail
;
5852 if (avail
< num_sectors
)
5855 rv
= mddev
->pers
->resize(mddev
, num_sectors
);
5857 revalidate_disk(mddev
->gendisk
);
5861 static int update_raid_disks(mddev_t
*mddev
, int raid_disks
)
5864 /* change the number of raid disks */
5865 if (mddev
->pers
->check_reshape
== NULL
)
5867 if (raid_disks
<= 0 ||
5868 (mddev
->max_disks
&& raid_disks
>= mddev
->max_disks
))
5870 if (mddev
->sync_thread
|| mddev
->reshape_position
!= MaxSector
)
5872 mddev
->delta_disks
= raid_disks
- mddev
->raid_disks
;
5874 rv
= mddev
->pers
->check_reshape(mddev
);
5876 mddev
->delta_disks
= 0;
5882 * update_array_info is used to change the configuration of an
5884 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5885 * fields in the info are checked against the array.
5886 * Any differences that cannot be handled will cause an error.
5887 * Normally, only one change can be managed at a time.
5889 static int update_array_info(mddev_t
*mddev
, mdu_array_info_t
*info
)
5895 /* calculate expected state,ignoring low bits */
5896 if (mddev
->bitmap
&& mddev
->bitmap_info
.offset
)
5897 state
|= (1 << MD_SB_BITMAP_PRESENT
);
5899 if (mddev
->major_version
!= info
->major_version
||
5900 mddev
->minor_version
!= info
->minor_version
||
5901 /* mddev->patch_version != info->patch_version || */
5902 mddev
->ctime
!= info
->ctime
||
5903 mddev
->level
!= info
->level
||
5904 /* mddev->layout != info->layout || */
5905 !mddev
->persistent
!= info
->not_persistent
||
5906 mddev
->chunk_sectors
!= info
->chunk_size
>> 9 ||
5907 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5908 ((state
^info
->state
) & 0xfffffe00)
5911 /* Check there is only one change */
5912 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5914 if (mddev
->raid_disks
!= info
->raid_disks
)
5916 if (mddev
->layout
!= info
->layout
)
5918 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
))
5925 if (mddev
->layout
!= info
->layout
) {
5927 * we don't need to do anything at the md level, the
5928 * personality will take care of it all.
5930 if (mddev
->pers
->check_reshape
== NULL
)
5933 mddev
->new_layout
= info
->layout
;
5934 rv
= mddev
->pers
->check_reshape(mddev
);
5936 mddev
->new_layout
= mddev
->layout
;
5940 if (info
->size
>= 0 && mddev
->dev_sectors
/ 2 != info
->size
)
5941 rv
= update_size(mddev
, (sector_t
)info
->size
* 2);
5943 if (mddev
->raid_disks
!= info
->raid_disks
)
5944 rv
= update_raid_disks(mddev
, info
->raid_disks
);
5946 if ((state
^ info
->state
) & (1<<MD_SB_BITMAP_PRESENT
)) {
5947 if (mddev
->pers
->quiesce
== NULL
)
5949 if (mddev
->recovery
|| mddev
->sync_thread
)
5951 if (info
->state
& (1<<MD_SB_BITMAP_PRESENT
)) {
5952 /* add the bitmap */
5955 if (mddev
->bitmap_info
.default_offset
== 0)
5957 mddev
->bitmap_info
.offset
=
5958 mddev
->bitmap_info
.default_offset
;
5959 mddev
->pers
->quiesce(mddev
, 1);
5960 rv
= bitmap_create(mddev
);
5962 rv
= bitmap_load(mddev
);
5964 bitmap_destroy(mddev
);
5965 mddev
->pers
->quiesce(mddev
, 0);
5967 /* remove the bitmap */
5970 if (mddev
->bitmap
->file
)
5972 mddev
->pers
->quiesce(mddev
, 1);
5973 bitmap_destroy(mddev
);
5974 mddev
->pers
->quiesce(mddev
, 0);
5975 mddev
->bitmap_info
.offset
= 0;
5978 md_update_sb(mddev
, 1);
5982 static int set_disk_faulty(mddev_t
*mddev
, dev_t dev
)
5986 if (mddev
->pers
== NULL
)
5989 rdev
= find_rdev(mddev
, dev
);
5993 md_error(mddev
, rdev
);
5994 if (!test_bit(Faulty
, &rdev
->flags
))
6000 * We have a problem here : there is no easy way to give a CHS
6001 * virtual geometry. We currently pretend that we have a 2 heads
6002 * 4 sectors (with a BIG number of cylinders...). This drives
6003 * dosfs just mad... ;-)
6005 static int md_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
6007 mddev_t
*mddev
= bdev
->bd_disk
->private_data
;
6011 geo
->cylinders
= mddev
->array_sectors
/ 8;
6015 static int md_ioctl(struct block_device
*bdev
, fmode_t mode
,
6016 unsigned int cmd
, unsigned long arg
)
6019 void __user
*argp
= (void __user
*)arg
;
6020 mddev_t
*mddev
= NULL
;
6023 if (!capable(CAP_SYS_ADMIN
))
6027 * Commands dealing with the RAID driver but not any
6033 err
= get_version(argp
);
6036 case PRINT_RAID_DEBUG
:
6044 autostart_arrays(arg
);
6051 * Commands creating/starting a new array:
6054 mddev
= bdev
->bd_disk
->private_data
;
6061 err
= mddev_lock(mddev
);
6064 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6071 case SET_ARRAY_INFO
:
6073 mdu_array_info_t info
;
6075 memset(&info
, 0, sizeof(info
));
6076 else if (copy_from_user(&info
, argp
, sizeof(info
))) {
6081 err
= update_array_info(mddev
, &info
);
6083 printk(KERN_WARNING
"md: couldn't update"
6084 " array info. %d\n", err
);
6089 if (!list_empty(&mddev
->disks
)) {
6091 "md: array %s already has disks!\n",
6096 if (mddev
->raid_disks
) {
6098 "md: array %s already initialised!\n",
6103 err
= set_array_info(mddev
, &info
);
6105 printk(KERN_WARNING
"md: couldn't set"
6106 " array info. %d\n", err
);
6116 * Commands querying/configuring an existing array:
6118 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6119 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6120 if ((!mddev
->raid_disks
&& !mddev
->external
)
6121 && cmd
!= ADD_NEW_DISK
&& cmd
!= STOP_ARRAY
6122 && cmd
!= RUN_ARRAY
&& cmd
!= SET_BITMAP_FILE
6123 && cmd
!= GET_BITMAP_FILE
) {
6129 * Commands even a read-only array can execute:
6133 case GET_ARRAY_INFO
:
6134 err
= get_array_info(mddev
, argp
);
6137 case GET_BITMAP_FILE
:
6138 err
= get_bitmap_file(mddev
, argp
);
6142 err
= get_disk_info(mddev
, argp
);
6145 case RESTART_ARRAY_RW
:
6146 err
= restart_array(mddev
);
6150 err
= do_md_stop(mddev
, 0, 1);
6154 err
= md_set_readonly(mddev
, 1);
6158 if (get_user(ro
, (int __user
*)(arg
))) {
6164 /* if the bdev is going readonly the value of mddev->ro
6165 * does not matter, no writes are coming
6170 /* are we are already prepared for writes? */
6174 /* transitioning to readauto need only happen for
6175 * arrays that call md_write_start
6178 err
= restart_array(mddev
);
6181 set_disk_ro(mddev
->gendisk
, 0);
6188 * The remaining ioctls are changing the state of the
6189 * superblock, so we do not allow them on read-only arrays.
6190 * However non-MD ioctls (e.g. get-size) will still come through
6191 * here and hit the 'default' below, so only disallow
6192 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6194 if (_IOC_TYPE(cmd
) == MD_MAJOR
&& mddev
->ro
&& mddev
->pers
) {
6195 if (mddev
->ro
== 2) {
6197 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6198 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6199 md_wakeup_thread(mddev
->thread
);
6210 mdu_disk_info_t info
;
6211 if (copy_from_user(&info
, argp
, sizeof(info
)))
6214 err
= add_new_disk(mddev
, &info
);
6218 case HOT_REMOVE_DISK
:
6219 err
= hot_remove_disk(mddev
, new_decode_dev(arg
));
6223 err
= hot_add_disk(mddev
, new_decode_dev(arg
));
6226 case SET_DISK_FAULTY
:
6227 err
= set_disk_faulty(mddev
, new_decode_dev(arg
));
6231 err
= do_md_run(mddev
);
6234 case SET_BITMAP_FILE
:
6235 err
= set_bitmap_file(mddev
, (int)arg
);
6245 if (mddev
->hold_active
== UNTIL_IOCTL
&&
6247 mddev
->hold_active
= 0;
6248 mddev_unlock(mddev
);
6257 #ifdef CONFIG_COMPAT
6258 static int md_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
6259 unsigned int cmd
, unsigned long arg
)
6262 case HOT_REMOVE_DISK
:
6264 case SET_DISK_FAULTY
:
6265 case SET_BITMAP_FILE
:
6266 /* These take in integer arg, do not convert */
6269 arg
= (unsigned long)compat_ptr(arg
);
6273 return md_ioctl(bdev
, mode
, cmd
, arg
);
6275 #endif /* CONFIG_COMPAT */
6277 static int md_open(struct block_device
*bdev
, fmode_t mode
)
6280 * Succeed if we can lock the mddev, which confirms that
6281 * it isn't being stopped right now.
6283 mddev_t
*mddev
= mddev_find(bdev
->bd_dev
);
6286 if (mddev
->gendisk
!= bdev
->bd_disk
) {
6287 /* we are racing with mddev_put which is discarding this
6291 /* Wait until bdev->bd_disk is definitely gone */
6292 flush_workqueue(md_misc_wq
);
6293 /* Then retry the open from the top */
6294 return -ERESTARTSYS
;
6296 BUG_ON(mddev
!= bdev
->bd_disk
->private_data
);
6298 if ((err
= mutex_lock_interruptible(&mddev
->open_mutex
)))
6302 atomic_inc(&mddev
->openers
);
6303 mutex_unlock(&mddev
->open_mutex
);
6305 check_disk_change(bdev
);
6310 static int md_release(struct gendisk
*disk
, fmode_t mode
)
6312 mddev_t
*mddev
= disk
->private_data
;
6315 atomic_dec(&mddev
->openers
);
6321 static int md_media_changed(struct gendisk
*disk
)
6323 mddev_t
*mddev
= disk
->private_data
;
6325 return mddev
->changed
;
6328 static int md_revalidate(struct gendisk
*disk
)
6330 mddev_t
*mddev
= disk
->private_data
;
6335 static const struct block_device_operations md_fops
=
6337 .owner
= THIS_MODULE
,
6339 .release
= md_release
,
6341 #ifdef CONFIG_COMPAT
6342 .compat_ioctl
= md_compat_ioctl
,
6344 .getgeo
= md_getgeo
,
6345 .media_changed
= md_media_changed
,
6346 .revalidate_disk
= md_revalidate
,
6349 static int md_thread(void * arg
)
6351 mdk_thread_t
*thread
= arg
;
6354 * md_thread is a 'system-thread', it's priority should be very
6355 * high. We avoid resource deadlocks individually in each
6356 * raid personality. (RAID5 does preallocation) We also use RR and
6357 * the very same RT priority as kswapd, thus we will never get
6358 * into a priority inversion deadlock.
6360 * we definitely have to have equal or higher priority than
6361 * bdflush, otherwise bdflush will deadlock if there are too
6362 * many dirty RAID5 blocks.
6365 allow_signal(SIGKILL
);
6366 while (!kthread_should_stop()) {
6368 /* We need to wait INTERRUPTIBLE so that
6369 * we don't add to the load-average.
6370 * That means we need to be sure no signals are
6373 if (signal_pending(current
))
6374 flush_signals(current
);
6376 wait_event_interruptible_timeout
6378 test_bit(THREAD_WAKEUP
, &thread
->flags
)
6379 || kthread_should_stop(),
6382 clear_bit(THREAD_WAKEUP
, &thread
->flags
);
6383 if (!kthread_should_stop())
6384 thread
->run(thread
->mddev
);
6390 void md_wakeup_thread(mdk_thread_t
*thread
)
6393 dprintk("md: waking up MD thread %s.\n", thread
->tsk
->comm
);
6394 set_bit(THREAD_WAKEUP
, &thread
->flags
);
6395 wake_up(&thread
->wqueue
);
6399 mdk_thread_t
*md_register_thread(void (*run
) (mddev_t
*), mddev_t
*mddev
,
6402 mdk_thread_t
*thread
;
6404 thread
= kzalloc(sizeof(mdk_thread_t
), GFP_KERNEL
);
6408 init_waitqueue_head(&thread
->wqueue
);
6411 thread
->mddev
= mddev
;
6412 thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
6413 thread
->tsk
= kthread_run(md_thread
, thread
,
6415 mdname(thread
->mddev
),
6416 name
?: mddev
->pers
->name
);
6417 if (IS_ERR(thread
->tsk
)) {
6424 void md_unregister_thread(mdk_thread_t
*thread
)
6428 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread
->tsk
));
6430 kthread_stop(thread
->tsk
);
6434 void md_error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
6441 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
6444 if (!mddev
->pers
|| !mddev
->pers
->error_handler
)
6446 mddev
->pers
->error_handler(mddev
,rdev
);
6447 if (mddev
->degraded
)
6448 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
6449 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
6450 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6451 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6452 md_wakeup_thread(mddev
->thread
);
6453 if (mddev
->event_work
.func
)
6454 queue_work(md_misc_wq
, &mddev
->event_work
);
6455 md_new_event_inintr(mddev
);
6458 /* seq_file implementation /proc/mdstat */
6460 static void status_unused(struct seq_file
*seq
)
6465 seq_printf(seq
, "unused devices: ");
6467 list_for_each_entry(rdev
, &pending_raid_disks
, same_set
) {
6468 char b
[BDEVNAME_SIZE
];
6470 seq_printf(seq
, "%s ",
6471 bdevname(rdev
->bdev
,b
));
6474 seq_printf(seq
, "<none>");
6476 seq_printf(seq
, "\n");
6480 static void status_resync(struct seq_file
*seq
, mddev_t
* mddev
)
6482 sector_t max_sectors
, resync
, res
;
6483 unsigned long dt
, db
;
6486 unsigned int per_milli
;
6488 resync
= mddev
->curr_resync
- atomic_read(&mddev
->recovery_active
);
6490 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
6491 max_sectors
= mddev
->resync_max_sectors
;
6493 max_sectors
= mddev
->dev_sectors
;
6496 * Should not happen.
6502 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6503 * in a sector_t, and (max_sectors>>scale) will fit in a
6504 * u32, as those are the requirements for sector_div.
6505 * Thus 'scale' must be at least 10
6508 if (sizeof(sector_t
) > sizeof(unsigned long)) {
6509 while ( max_sectors
/2 > (1ULL<<(scale
+32)))
6512 res
= (resync
>>scale
)*1000;
6513 sector_div(res
, (u32
)((max_sectors
>>scale
)+1));
6517 int i
, x
= per_milli
/50, y
= 20-x
;
6518 seq_printf(seq
, "[");
6519 for (i
= 0; i
< x
; i
++)
6520 seq_printf(seq
, "=");
6521 seq_printf(seq
, ">");
6522 for (i
= 0; i
< y
; i
++)
6523 seq_printf(seq
, ".");
6524 seq_printf(seq
, "] ");
6526 seq_printf(seq
, " %s =%3u.%u%% (%llu/%llu)",
6527 (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
)?
6529 (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)?
6531 (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) ?
6532 "resync" : "recovery"))),
6533 per_milli
/10, per_milli
% 10,
6534 (unsigned long long) resync
/2,
6535 (unsigned long long) max_sectors
/2);
6538 * dt: time from mark until now
6539 * db: blocks written from mark until now
6540 * rt: remaining time
6542 * rt is a sector_t, so could be 32bit or 64bit.
6543 * So we divide before multiply in case it is 32bit and close
6545 * We scale the divisor (db) by 32 to avoid losing precision
6546 * near the end of resync when the number of remaining sectors
6548 * We then divide rt by 32 after multiplying by db to compensate.
6549 * The '+1' avoids division by zero if db is very small.
6551 dt
= ((jiffies
- mddev
->resync_mark
) / HZ
);
6553 db
= (mddev
->curr_mark_cnt
- atomic_read(&mddev
->recovery_active
))
6554 - mddev
->resync_mark_cnt
;
6556 rt
= max_sectors
- resync
; /* number of remaining sectors */
6557 sector_div(rt
, db
/32+1);
6561 seq_printf(seq
, " finish=%lu.%lumin", (unsigned long)rt
/ 60,
6562 ((unsigned long)rt
% 60)/6);
6564 seq_printf(seq
, " speed=%ldK/sec", db
/2/dt
);
6567 static void *md_seq_start(struct seq_file
*seq
, loff_t
*pos
)
6569 struct list_head
*tmp
;
6579 spin_lock(&all_mddevs_lock
);
6580 list_for_each(tmp
,&all_mddevs
)
6582 mddev
= list_entry(tmp
, mddev_t
, all_mddevs
);
6584 spin_unlock(&all_mddevs_lock
);
6587 spin_unlock(&all_mddevs_lock
);
6589 return (void*)2;/* tail */
6593 static void *md_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
6595 struct list_head
*tmp
;
6596 mddev_t
*next_mddev
, *mddev
= v
;
6602 spin_lock(&all_mddevs_lock
);
6604 tmp
= all_mddevs
.next
;
6606 tmp
= mddev
->all_mddevs
.next
;
6607 if (tmp
!= &all_mddevs
)
6608 next_mddev
= mddev_get(list_entry(tmp
,mddev_t
,all_mddevs
));
6610 next_mddev
= (void*)2;
6613 spin_unlock(&all_mddevs_lock
);
6621 static void md_seq_stop(struct seq_file
*seq
, void *v
)
6625 if (mddev
&& v
!= (void*)1 && v
!= (void*)2)
6629 static int md_seq_show(struct seq_file
*seq
, void *v
)
6634 struct bitmap
*bitmap
;
6636 if (v
== (void*)1) {
6637 struct mdk_personality
*pers
;
6638 seq_printf(seq
, "Personalities : ");
6639 spin_lock(&pers_lock
);
6640 list_for_each_entry(pers
, &pers_list
, list
)
6641 seq_printf(seq
, "[%s] ", pers
->name
);
6643 spin_unlock(&pers_lock
);
6644 seq_printf(seq
, "\n");
6645 seq
->poll_event
= atomic_read(&md_event_count
);
6648 if (v
== (void*)2) {
6653 if (mddev_lock(mddev
) < 0)
6656 if (mddev
->pers
|| mddev
->raid_disks
|| !list_empty(&mddev
->disks
)) {
6657 seq_printf(seq
, "%s : %sactive", mdname(mddev
),
6658 mddev
->pers
? "" : "in");
6661 seq_printf(seq
, " (read-only)");
6663 seq_printf(seq
, " (auto-read-only)");
6664 seq_printf(seq
, " %s", mddev
->pers
->name
);
6668 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
6669 char b
[BDEVNAME_SIZE
];
6670 seq_printf(seq
, " %s[%d]",
6671 bdevname(rdev
->bdev
,b
), rdev
->desc_nr
);
6672 if (test_bit(WriteMostly
, &rdev
->flags
))
6673 seq_printf(seq
, "(W)");
6674 if (test_bit(Faulty
, &rdev
->flags
)) {
6675 seq_printf(seq
, "(F)");
6677 } else if (rdev
->raid_disk
< 0)
6678 seq_printf(seq
, "(S)"); /* spare */
6679 sectors
+= rdev
->sectors
;
6682 if (!list_empty(&mddev
->disks
)) {
6684 seq_printf(seq
, "\n %llu blocks",
6685 (unsigned long long)
6686 mddev
->array_sectors
/ 2);
6688 seq_printf(seq
, "\n %llu blocks",
6689 (unsigned long long)sectors
/ 2);
6691 if (mddev
->persistent
) {
6692 if (mddev
->major_version
!= 0 ||
6693 mddev
->minor_version
!= 90) {
6694 seq_printf(seq
," super %d.%d",
6695 mddev
->major_version
,
6696 mddev
->minor_version
);
6698 } else if (mddev
->external
)
6699 seq_printf(seq
, " super external:%s",
6700 mddev
->metadata_type
);
6702 seq_printf(seq
, " super non-persistent");
6705 mddev
->pers
->status(seq
, mddev
);
6706 seq_printf(seq
, "\n ");
6707 if (mddev
->pers
->sync_request
) {
6708 if (mddev
->curr_resync
> 2) {
6709 status_resync(seq
, mddev
);
6710 seq_printf(seq
, "\n ");
6711 } else if (mddev
->curr_resync
== 1 || mddev
->curr_resync
== 2)
6712 seq_printf(seq
, "\tresync=DELAYED\n ");
6713 else if (mddev
->recovery_cp
< MaxSector
)
6714 seq_printf(seq
, "\tresync=PENDING\n ");
6717 seq_printf(seq
, "\n ");
6719 if ((bitmap
= mddev
->bitmap
)) {
6720 unsigned long chunk_kb
;
6721 unsigned long flags
;
6722 spin_lock_irqsave(&bitmap
->lock
, flags
);
6723 chunk_kb
= mddev
->bitmap_info
.chunksize
>> 10;
6724 seq_printf(seq
, "bitmap: %lu/%lu pages [%luKB], "
6726 bitmap
->pages
- bitmap
->missing_pages
,
6728 (bitmap
->pages
- bitmap
->missing_pages
)
6729 << (PAGE_SHIFT
- 10),
6730 chunk_kb
? chunk_kb
: mddev
->bitmap_info
.chunksize
,
6731 chunk_kb
? "KB" : "B");
6733 seq_printf(seq
, ", file: ");
6734 seq_path(seq
, &bitmap
->file
->f_path
, " \t\n");
6737 seq_printf(seq
, "\n");
6738 spin_unlock_irqrestore(&bitmap
->lock
, flags
);
6741 seq_printf(seq
, "\n");
6743 mddev_unlock(mddev
);
6748 static const struct seq_operations md_seq_ops
= {
6749 .start
= md_seq_start
,
6750 .next
= md_seq_next
,
6751 .stop
= md_seq_stop
,
6752 .show
= md_seq_show
,
6755 static int md_seq_open(struct inode
*inode
, struct file
*file
)
6757 struct seq_file
*seq
;
6760 error
= seq_open(file
, &md_seq_ops
);
6764 seq
= file
->private_data
;
6765 seq
->poll_event
= atomic_read(&md_event_count
);
6769 static unsigned int mdstat_poll(struct file
*filp
, poll_table
*wait
)
6771 struct seq_file
*seq
= filp
->private_data
;
6774 poll_wait(filp
, &md_event_waiters
, wait
);
6776 /* always allow read */
6777 mask
= POLLIN
| POLLRDNORM
;
6779 if (seq
->poll_event
!= atomic_read(&md_event_count
))
6780 mask
|= POLLERR
| POLLPRI
;
6784 static const struct file_operations md_seq_fops
= {
6785 .owner
= THIS_MODULE
,
6786 .open
= md_seq_open
,
6788 .llseek
= seq_lseek
,
6789 .release
= seq_release_private
,
6790 .poll
= mdstat_poll
,
6793 int register_md_personality(struct mdk_personality
*p
)
6795 spin_lock(&pers_lock
);
6796 list_add_tail(&p
->list
, &pers_list
);
6797 printk(KERN_INFO
"md: %s personality registered for level %d\n", p
->name
, p
->level
);
6798 spin_unlock(&pers_lock
);
6802 int unregister_md_personality(struct mdk_personality
*p
)
6804 printk(KERN_INFO
"md: %s personality unregistered\n", p
->name
);
6805 spin_lock(&pers_lock
);
6806 list_del_init(&p
->list
);
6807 spin_unlock(&pers_lock
);
6811 static int is_mddev_idle(mddev_t
*mddev
, int init
)
6819 rdev_for_each_rcu(rdev
, mddev
) {
6820 struct gendisk
*disk
= rdev
->bdev
->bd_contains
->bd_disk
;
6821 curr_events
= (int)part_stat_read(&disk
->part0
, sectors
[0]) +
6822 (int)part_stat_read(&disk
->part0
, sectors
[1]) -
6823 atomic_read(&disk
->sync_io
);
6824 /* sync IO will cause sync_io to increase before the disk_stats
6825 * as sync_io is counted when a request starts, and
6826 * disk_stats is counted when it completes.
6827 * So resync activity will cause curr_events to be smaller than
6828 * when there was no such activity.
6829 * non-sync IO will cause disk_stat to increase without
6830 * increasing sync_io so curr_events will (eventually)
6831 * be larger than it was before. Once it becomes
6832 * substantially larger, the test below will cause
6833 * the array to appear non-idle, and resync will slow
6835 * If there is a lot of outstanding resync activity when
6836 * we set last_event to curr_events, then all that activity
6837 * completing might cause the array to appear non-idle
6838 * and resync will be slowed down even though there might
6839 * not have been non-resync activity. This will only
6840 * happen once though. 'last_events' will soon reflect
6841 * the state where there is little or no outstanding
6842 * resync requests, and further resync activity will
6843 * always make curr_events less than last_events.
6846 if (init
|| curr_events
- rdev
->last_events
> 64) {
6847 rdev
->last_events
= curr_events
;
6855 void md_done_sync(mddev_t
*mddev
, int blocks
, int ok
)
6857 /* another "blocks" (512byte) blocks have been synced */
6858 atomic_sub(blocks
, &mddev
->recovery_active
);
6859 wake_up(&mddev
->recovery_wait
);
6861 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
6862 md_wakeup_thread(mddev
->thread
);
6863 // stop recovery, signal do_sync ....
6868 /* md_write_start(mddev, bi)
6869 * If we need to update some array metadata (e.g. 'active' flag
6870 * in superblock) before writing, schedule a superblock update
6871 * and wait for it to complete.
6873 void md_write_start(mddev_t
*mddev
, struct bio
*bi
)
6876 if (bio_data_dir(bi
) != WRITE
)
6879 BUG_ON(mddev
->ro
== 1);
6880 if (mddev
->ro
== 2) {
6881 /* need to switch to read/write */
6883 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
6884 md_wakeup_thread(mddev
->thread
);
6885 md_wakeup_thread(mddev
->sync_thread
);
6888 atomic_inc(&mddev
->writes_pending
);
6889 if (mddev
->safemode
== 1)
6890 mddev
->safemode
= 0;
6891 if (mddev
->in_sync
) {
6892 spin_lock_irq(&mddev
->write_lock
);
6893 if (mddev
->in_sync
) {
6895 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6896 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6897 md_wakeup_thread(mddev
->thread
);
6900 spin_unlock_irq(&mddev
->write_lock
);
6903 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6904 wait_event(mddev
->sb_wait
,
6905 !test_bit(MD_CHANGE_PENDING
, &mddev
->flags
));
6908 void md_write_end(mddev_t
*mddev
)
6910 if (atomic_dec_and_test(&mddev
->writes_pending
)) {
6911 if (mddev
->safemode
== 2)
6912 md_wakeup_thread(mddev
->thread
);
6913 else if (mddev
->safemode_delay
)
6914 mod_timer(&mddev
->safemode_timer
, jiffies
+ mddev
->safemode_delay
);
6918 /* md_allow_write(mddev)
6919 * Calling this ensures that the array is marked 'active' so that writes
6920 * may proceed without blocking. It is important to call this before
6921 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6922 * Must be called with mddev_lock held.
6924 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6925 * is dropped, so return -EAGAIN after notifying userspace.
6927 int md_allow_write(mddev_t
*mddev
)
6933 if (!mddev
->pers
->sync_request
)
6936 spin_lock_irq(&mddev
->write_lock
);
6937 if (mddev
->in_sync
) {
6939 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
6940 set_bit(MD_CHANGE_PENDING
, &mddev
->flags
);
6941 if (mddev
->safemode_delay
&&
6942 mddev
->safemode
== 0)
6943 mddev
->safemode
= 1;
6944 spin_unlock_irq(&mddev
->write_lock
);
6945 md_update_sb(mddev
, 0);
6946 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
6948 spin_unlock_irq(&mddev
->write_lock
);
6950 if (test_bit(MD_CHANGE_PENDING
, &mddev
->flags
))
6955 EXPORT_SYMBOL_GPL(md_allow_write
);
6957 #define SYNC_MARKS 10
6958 #define SYNC_MARK_STEP (3*HZ)
6959 void md_do_sync(mddev_t
*mddev
)
6962 unsigned int currspeed
= 0,
6964 sector_t max_sectors
,j
, io_sectors
;
6965 unsigned long mark
[SYNC_MARKS
];
6966 sector_t mark_cnt
[SYNC_MARKS
];
6968 struct list_head
*tmp
;
6969 sector_t last_check
;
6974 /* just incase thread restarts... */
6975 if (test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
))
6977 if (mddev
->ro
) /* never try to sync a read-only array */
6980 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
6981 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
6982 desc
= "data-check";
6983 else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
6984 desc
= "requested-resync";
6987 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
6992 /* we overload curr_resync somewhat here.
6993 * 0 == not engaged in resync at all
6994 * 2 == checking that there is no conflict with another sync
6995 * 1 == like 2, but have yielded to allow conflicting resync to
6997 * other == active in resync - this many blocks
6999 * Before starting a resync we must have set curr_resync to
7000 * 2, and then checked that every "conflicting" array has curr_resync
7001 * less than ours. When we find one that is the same or higher
7002 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7003 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7004 * This will mean we have to start checking from the beginning again.
7009 mddev
->curr_resync
= 2;
7012 if (kthread_should_stop())
7013 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7015 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7017 for_each_mddev(mddev2
, tmp
) {
7018 if (mddev2
== mddev
)
7020 if (!mddev
->parallel_resync
7021 && mddev2
->curr_resync
7022 && match_mddev_units(mddev
, mddev2
)) {
7024 if (mddev
< mddev2
&& mddev
->curr_resync
== 2) {
7025 /* arbitrarily yield */
7026 mddev
->curr_resync
= 1;
7027 wake_up(&resync_wait
);
7029 if (mddev
> mddev2
&& mddev
->curr_resync
== 1)
7030 /* no need to wait here, we can wait the next
7031 * time 'round when curr_resync == 2
7034 /* We need to wait 'interruptible' so as not to
7035 * contribute to the load average, and not to
7036 * be caught by 'softlockup'
7038 prepare_to_wait(&resync_wait
, &wq
, TASK_INTERRUPTIBLE
);
7039 if (!kthread_should_stop() &&
7040 mddev2
->curr_resync
>= mddev
->curr_resync
) {
7041 printk(KERN_INFO
"md: delaying %s of %s"
7042 " until %s has finished (they"
7043 " share one or more physical units)\n",
7044 desc
, mdname(mddev
), mdname(mddev2
));
7046 if (signal_pending(current
))
7047 flush_signals(current
);
7049 finish_wait(&resync_wait
, &wq
);
7052 finish_wait(&resync_wait
, &wq
);
7055 } while (mddev
->curr_resync
< 2);
7058 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7059 /* resync follows the size requested by the personality,
7060 * which defaults to physical size, but can be virtual size
7062 max_sectors
= mddev
->resync_max_sectors
;
7063 mddev
->resync_mismatches
= 0;
7064 /* we don't use the checkpoint if there's a bitmap */
7065 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7066 j
= mddev
->resync_min
;
7067 else if (!mddev
->bitmap
)
7068 j
= mddev
->recovery_cp
;
7070 } else if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
))
7071 max_sectors
= mddev
->dev_sectors
;
7073 /* recovery follows the physical size of devices */
7074 max_sectors
= mddev
->dev_sectors
;
7077 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7078 if (rdev
->raid_disk
>= 0 &&
7079 !test_bit(Faulty
, &rdev
->flags
) &&
7080 !test_bit(In_sync
, &rdev
->flags
) &&
7081 rdev
->recovery_offset
< j
)
7082 j
= rdev
->recovery_offset
;
7086 printk(KERN_INFO
"md: %s of RAID array %s\n", desc
, mdname(mddev
));
7087 printk(KERN_INFO
"md: minimum _guaranteed_ speed:"
7088 " %d KB/sec/disk.\n", speed_min(mddev
));
7089 printk(KERN_INFO
"md: using maximum available idle IO bandwidth "
7090 "(but not more than %d KB/sec) for %s.\n",
7091 speed_max(mddev
), desc
);
7093 is_mddev_idle(mddev
, 1); /* this initializes IO event counters */
7096 for (m
= 0; m
< SYNC_MARKS
; m
++) {
7098 mark_cnt
[m
] = io_sectors
;
7101 mddev
->resync_mark
= mark
[last_mark
];
7102 mddev
->resync_mark_cnt
= mark_cnt
[last_mark
];
7105 * Tune reconstruction:
7107 window
= 32*(PAGE_SIZE
/512);
7108 printk(KERN_INFO
"md: using %dk window, over a total of %lluk.\n",
7109 window
/2, (unsigned long long)max_sectors
/2);
7111 atomic_set(&mddev
->recovery_active
, 0);
7116 "md: resuming %s of %s from checkpoint.\n",
7117 desc
, mdname(mddev
));
7118 mddev
->curr_resync
= j
;
7120 mddev
->curr_resync_completed
= j
;
7122 while (j
< max_sectors
) {
7127 if (!test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7128 ((mddev
->curr_resync
> mddev
->curr_resync_completed
&&
7129 (mddev
->curr_resync
- mddev
->curr_resync_completed
)
7130 > (max_sectors
>> 4)) ||
7131 (j
- mddev
->curr_resync_completed
)*2
7132 >= mddev
->resync_max
- mddev
->curr_resync_completed
7134 /* time to update curr_resync_completed */
7135 wait_event(mddev
->recovery_wait
,
7136 atomic_read(&mddev
->recovery_active
) == 0);
7137 mddev
->curr_resync_completed
= j
;
7138 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7139 sysfs_notify(&mddev
->kobj
, NULL
, "sync_completed");
7142 while (j
>= mddev
->resync_max
&& !kthread_should_stop()) {
7143 /* As this condition is controlled by user-space,
7144 * we can block indefinitely, so use '_interruptible'
7145 * to avoid triggering warnings.
7147 flush_signals(current
); /* just in case */
7148 wait_event_interruptible(mddev
->recovery_wait
,
7149 mddev
->resync_max
> j
7150 || kthread_should_stop());
7153 if (kthread_should_stop())
7156 sectors
= mddev
->pers
->sync_request(mddev
, j
, &skipped
,
7157 currspeed
< speed_min(mddev
));
7159 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7163 if (!skipped
) { /* actual IO requested */
7164 io_sectors
+= sectors
;
7165 atomic_add(sectors
, &mddev
->recovery_active
);
7168 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7172 if (j
>1) mddev
->curr_resync
= j
;
7173 mddev
->curr_mark_cnt
= io_sectors
;
7174 if (last_check
== 0)
7175 /* this is the earliest that rebuild will be
7176 * visible in /proc/mdstat
7178 md_new_event(mddev
);
7180 if (last_check
+ window
> io_sectors
|| j
== max_sectors
)
7183 last_check
= io_sectors
;
7185 if (time_after_eq(jiffies
, mark
[last_mark
] + SYNC_MARK_STEP
)) {
7187 int next
= (last_mark
+1) % SYNC_MARKS
;
7189 mddev
->resync_mark
= mark
[next
];
7190 mddev
->resync_mark_cnt
= mark_cnt
[next
];
7191 mark
[next
] = jiffies
;
7192 mark_cnt
[next
] = io_sectors
- atomic_read(&mddev
->recovery_active
);
7197 if (kthread_should_stop())
7202 * this loop exits only if either when we are slower than
7203 * the 'hard' speed limit, or the system was IO-idle for
7205 * the system might be non-idle CPU-wise, but we only care
7206 * about not overloading the IO subsystem. (things like an
7207 * e2fsck being done on the RAID array should execute fast)
7211 currspeed
= ((unsigned long)(io_sectors
-mddev
->resync_mark_cnt
))/2
7212 /((jiffies
-mddev
->resync_mark
)/HZ
+1) +1;
7214 if (currspeed
> speed_min(mddev
)) {
7215 if ((currspeed
> speed_max(mddev
)) ||
7216 !is_mddev_idle(mddev
, 0)) {
7222 printk(KERN_INFO
"md: %s: %s done.\n",mdname(mddev
), desc
);
7224 * this also signals 'finished resyncing' to md_stop
7227 wait_event(mddev
->recovery_wait
, !atomic_read(&mddev
->recovery_active
));
7229 /* tell personality that we are finished */
7230 mddev
->pers
->sync_request(mddev
, max_sectors
, &skipped
, 1);
7232 if (!test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
) &&
7233 mddev
->curr_resync
> 2) {
7234 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
7235 if (test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7236 if (mddev
->curr_resync
>= mddev
->recovery_cp
) {
7238 "md: checkpointing %s of %s.\n",
7239 desc
, mdname(mddev
));
7240 mddev
->recovery_cp
= mddev
->curr_resync
;
7243 mddev
->recovery_cp
= MaxSector
;
7245 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
))
7246 mddev
->curr_resync
= MaxSector
;
7248 list_for_each_entry_rcu(rdev
, &mddev
->disks
, same_set
)
7249 if (rdev
->raid_disk
>= 0 &&
7250 mddev
->delta_disks
>= 0 &&
7251 !test_bit(Faulty
, &rdev
->flags
) &&
7252 !test_bit(In_sync
, &rdev
->flags
) &&
7253 rdev
->recovery_offset
< mddev
->curr_resync
)
7254 rdev
->recovery_offset
= mddev
->curr_resync
;
7258 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7261 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
)) {
7262 /* We completed so min/max setting can be forgotten if used. */
7263 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7264 mddev
->resync_min
= 0;
7265 mddev
->resync_max
= MaxSector
;
7266 } else if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
7267 mddev
->resync_min
= mddev
->curr_resync_completed
;
7268 mddev
->curr_resync
= 0;
7269 wake_up(&resync_wait
);
7270 set_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7271 md_wakeup_thread(mddev
->thread
);
7276 * got a signal, exit.
7279 "md: md_do_sync() got signal ... exiting\n");
7280 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7284 EXPORT_SYMBOL_GPL(md_do_sync
);
7286 static int remove_and_add_spares(mddev_t
*mddev
)
7291 mddev
->curr_resync_completed
= 0;
7293 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7294 if (rdev
->raid_disk
>= 0 &&
7295 !test_bit(Blocked
, &rdev
->flags
) &&
7296 (test_bit(Faulty
, &rdev
->flags
) ||
7297 ! test_bit(In_sync
, &rdev
->flags
)) &&
7298 atomic_read(&rdev
->nr_pending
)==0) {
7299 if (mddev
->pers
->hot_remove_disk(
7300 mddev
, rdev
->raid_disk
)==0) {
7301 sysfs_unlink_rdev(mddev
, rdev
);
7302 rdev
->raid_disk
= -1;
7306 if (mddev
->degraded
) {
7307 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
7308 if (rdev
->raid_disk
>= 0 &&
7309 !test_bit(In_sync
, &rdev
->flags
) &&
7310 !test_bit(Faulty
, &rdev
->flags
))
7312 if (rdev
->raid_disk
< 0
7313 && !test_bit(Faulty
, &rdev
->flags
)) {
7314 rdev
->recovery_offset
= 0;
7316 hot_add_disk(mddev
, rdev
) == 0) {
7317 if (sysfs_link_rdev(mddev
, rdev
))
7318 /* failure here is OK */;
7320 md_new_event(mddev
);
7321 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
7330 static void reap_sync_thread(mddev_t
*mddev
)
7334 /* resync has finished, collect result */
7335 md_unregister_thread(mddev
->sync_thread
);
7336 mddev
->sync_thread
= NULL
;
7337 if (!test_bit(MD_RECOVERY_INTR
, &mddev
->recovery
) &&
7338 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
7340 /* activate any spares */
7341 if (mddev
->pers
->spare_active(mddev
))
7342 sysfs_notify(&mddev
->kobj
, NULL
,
7345 if (test_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
) &&
7346 mddev
->pers
->finish_reshape
)
7347 mddev
->pers
->finish_reshape(mddev
);
7348 md_update_sb(mddev
, 1);
7350 /* if array is no-longer degraded, then any saved_raid_disk
7351 * information must be scrapped
7353 if (!mddev
->degraded
)
7354 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7355 rdev
->saved_raid_disk
= -1;
7357 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7358 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7359 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7360 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7361 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7362 /* flag recovery needed just to double check */
7363 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7364 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7365 md_new_event(mddev
);
7366 if (mddev
->event_work
.func
)
7367 queue_work(md_misc_wq
, &mddev
->event_work
);
7371 * This routine is regularly called by all per-raid-array threads to
7372 * deal with generic issues like resync and super-block update.
7373 * Raid personalities that don't have a thread (linear/raid0) do not
7374 * need this as they never do any recovery or update the superblock.
7376 * It does not do any resync itself, but rather "forks" off other threads
7377 * to do that as needed.
7378 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7379 * "->recovery" and create a thread at ->sync_thread.
7380 * When the thread finishes it sets MD_RECOVERY_DONE
7381 * and wakeups up this thread which will reap the thread and finish up.
7382 * This thread also removes any faulty devices (with nr_pending == 0).
7384 * The overall approach is:
7385 * 1/ if the superblock needs updating, update it.
7386 * 2/ If a recovery thread is running, don't do anything else.
7387 * 3/ If recovery has finished, clean up, possibly marking spares active.
7388 * 4/ If there are any faulty devices, remove them.
7389 * 5/ If array is degraded, try to add spares devices
7390 * 6/ If array has spares or is not in-sync, start a resync thread.
7392 void md_check_recovery(mddev_t
*mddev
)
7394 if (mddev
->suspended
)
7398 bitmap_daemon_work(mddev
);
7400 if (signal_pending(current
)) {
7401 if (mddev
->pers
->sync_request
&& !mddev
->external
) {
7402 printk(KERN_INFO
"md: %s in immediate safe mode\n",
7404 mddev
->safemode
= 2;
7406 flush_signals(current
);
7409 if (mddev
->ro
&& !test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
))
7412 (mddev
->flags
& ~ (1<<MD_CHANGE_PENDING
)) ||
7413 test_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
) ||
7414 test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
) ||
7415 (mddev
->external
== 0 && mddev
->safemode
== 1) ||
7416 (mddev
->safemode
== 2 && ! atomic_read(&mddev
->writes_pending
)
7417 && !mddev
->in_sync
&& mddev
->recovery_cp
== MaxSector
)
7421 if (mddev_trylock(mddev
)) {
7425 /* Only thing we do on a ro array is remove
7429 list_for_each_entry(rdev
, &mddev
->disks
, same_set
)
7430 if (rdev
->raid_disk
>= 0 &&
7431 !test_bit(Blocked
, &rdev
->flags
) &&
7432 test_bit(Faulty
, &rdev
->flags
) &&
7433 atomic_read(&rdev
->nr_pending
)==0) {
7434 if (mddev
->pers
->hot_remove_disk(
7435 mddev
, rdev
->raid_disk
)==0) {
7436 sysfs_unlink_rdev(mddev
, rdev
);
7437 rdev
->raid_disk
= -1;
7440 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7444 if (!mddev
->external
) {
7446 spin_lock_irq(&mddev
->write_lock
);
7447 if (mddev
->safemode
&&
7448 !atomic_read(&mddev
->writes_pending
) &&
7450 mddev
->recovery_cp
== MaxSector
) {
7453 set_bit(MD_CHANGE_CLEAN
, &mddev
->flags
);
7455 if (mddev
->safemode
== 1)
7456 mddev
->safemode
= 0;
7457 spin_unlock_irq(&mddev
->write_lock
);
7459 sysfs_notify_dirent_safe(mddev
->sysfs_state
);
7463 md_update_sb(mddev
, 0);
7465 if (test_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
) &&
7466 !test_bit(MD_RECOVERY_DONE
, &mddev
->recovery
)) {
7467 /* resync/recovery still happening */
7468 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7471 if (mddev
->sync_thread
) {
7472 reap_sync_thread(mddev
);
7475 /* Set RUNNING before clearing NEEDED to avoid
7476 * any transients in the value of "sync_action".
7478 set_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7479 clear_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
7480 /* Clear some bits that don't mean anything, but
7483 clear_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
7484 clear_bit(MD_RECOVERY_DONE
, &mddev
->recovery
);
7486 if (test_bit(MD_RECOVERY_FROZEN
, &mddev
->recovery
))
7488 /* no recovery is running.
7489 * remove any failed drives, then
7490 * add spares if possible.
7491 * Spare are also removed and re-added, to allow
7492 * the personality to fail the re-add.
7495 if (mddev
->reshape_position
!= MaxSector
) {
7496 if (mddev
->pers
->check_reshape
== NULL
||
7497 mddev
->pers
->check_reshape(mddev
) != 0)
7498 /* Cannot proceed */
7500 set_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7501 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7502 } else if ((spares
= remove_and_add_spares(mddev
))) {
7503 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7504 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7505 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7506 set_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7507 } else if (mddev
->recovery_cp
< MaxSector
) {
7508 set_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7509 clear_bit(MD_RECOVERY_RECOVER
, &mddev
->recovery
);
7510 } else if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
7511 /* nothing to be done ... */
7514 if (mddev
->pers
->sync_request
) {
7515 if (spares
&& mddev
->bitmap
&& ! mddev
->bitmap
->file
) {
7516 /* We are adding a device or devices to an array
7517 * which has the bitmap stored on all devices.
7518 * So make sure all bitmap pages get written
7520 bitmap_write_all(mddev
->bitmap
);
7522 mddev
->sync_thread
= md_register_thread(md_do_sync
,
7525 if (!mddev
->sync_thread
) {
7526 printk(KERN_ERR
"%s: could not start resync"
7529 /* leave the spares where they are, it shouldn't hurt */
7530 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7531 clear_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
);
7532 clear_bit(MD_RECOVERY_RESHAPE
, &mddev
->recovery
);
7533 clear_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
);
7534 clear_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
);
7536 md_wakeup_thread(mddev
->sync_thread
);
7537 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7538 md_new_event(mddev
);
7541 if (!mddev
->sync_thread
) {
7542 clear_bit(MD_RECOVERY_RUNNING
, &mddev
->recovery
);
7543 if (test_and_clear_bit(MD_RECOVERY_RECOVER
,
7545 if (mddev
->sysfs_action
)
7546 sysfs_notify_dirent_safe(mddev
->sysfs_action
);
7548 mddev_unlock(mddev
);
7552 void md_wait_for_blocked_rdev(mdk_rdev_t
*rdev
, mddev_t
*mddev
)
7554 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
7555 wait_event_timeout(rdev
->blocked_wait
,
7556 !test_bit(Blocked
, &rdev
->flags
) &&
7557 !test_bit(BlockedBadBlocks
, &rdev
->flags
),
7558 msecs_to_jiffies(5000));
7559 rdev_dec_pending(rdev
, mddev
);
7561 EXPORT_SYMBOL(md_wait_for_blocked_rdev
);
7564 /* Bad block management.
7565 * We can record which blocks on each device are 'bad' and so just
7566 * fail those blocks, or that stripe, rather than the whole device.
7567 * Entries in the bad-block table are 64bits wide. This comprises:
7568 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7569 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7570 * A 'shift' can be set so that larger blocks are tracked and
7571 * consequently larger devices can be covered.
7572 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7574 * Locking of the bad-block table uses a seqlock so md_is_badblock
7575 * might need to retry if it is very unlucky.
7576 * We will sometimes want to check for bad blocks in a bi_end_io function,
7577 * so we use the write_seqlock_irq variant.
7579 * When looking for a bad block we specify a range and want to
7580 * know if any block in the range is bad. So we binary-search
7581 * to the last range that starts at-or-before the given endpoint,
7582 * (or "before the sector after the target range")
7583 * then see if it ends after the given start.
7585 * 0 if there are no known bad blocks in the range
7586 * 1 if there are known bad block which are all acknowledged
7587 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7588 * plus the start/length of the first bad section we overlap.
7590 int md_is_badblock(struct badblocks
*bb
, sector_t s
, int sectors
,
7591 sector_t
*first_bad
, int *bad_sectors
)
7597 sector_t target
= s
+ sectors
;
7600 if (bb
->shift
> 0) {
7601 /* round the start down, and the end up */
7603 target
+= (1<<bb
->shift
) - 1;
7604 target
>>= bb
->shift
;
7605 sectors
= target
- s
;
7607 /* 'target' is now the first block after the bad range */
7610 seq
= read_seqbegin(&bb
->lock
);
7614 /* Binary search between lo and hi for 'target'
7615 * i.e. for the last range that starts before 'target'
7617 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7618 * are known not to be the last range before target.
7619 * VARIANT: hi-lo is the number of possible
7620 * ranges, and decreases until it reaches 1
7622 while (hi
- lo
> 1) {
7623 int mid
= (lo
+ hi
) / 2;
7624 sector_t a
= BB_OFFSET(p
[mid
]);
7626 /* This could still be the one, earlier ranges
7630 /* This and later ranges are definitely out. */
7633 /* 'lo' might be the last that started before target, but 'hi' isn't */
7635 /* need to check all range that end after 's' to see if
7636 * any are unacknowledged.
7639 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7640 if (BB_OFFSET(p
[lo
]) < target
) {
7641 /* starts before the end, and finishes after
7642 * the start, so they must overlap
7644 if (rv
!= -1 && BB_ACK(p
[lo
]))
7648 *first_bad
= BB_OFFSET(p
[lo
]);
7649 *bad_sectors
= BB_LEN(p
[lo
]);
7655 if (read_seqretry(&bb
->lock
, seq
))
7660 EXPORT_SYMBOL_GPL(md_is_badblock
);
7663 * Add a range of bad blocks to the table.
7664 * This might extend the table, or might contract it
7665 * if two adjacent ranges can be merged.
7666 * We binary-search to find the 'insertion' point, then
7667 * decide how best to handle it.
7669 static int md_set_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
,
7677 /* badblocks are disabled */
7681 /* round the start down, and the end up */
7682 sector_t next
= s
+ sectors
;
7684 next
+= (1<<bb
->shift
) - 1;
7689 write_seqlock_irq(&bb
->lock
);
7694 /* Find the last range that starts at-or-before 's' */
7695 while (hi
- lo
> 1) {
7696 int mid
= (lo
+ hi
) / 2;
7697 sector_t a
= BB_OFFSET(p
[mid
]);
7703 if (hi
> lo
&& BB_OFFSET(p
[lo
]) > s
)
7707 /* we found a range that might merge with the start
7710 sector_t a
= BB_OFFSET(p
[lo
]);
7711 sector_t e
= a
+ BB_LEN(p
[lo
]);
7712 int ack
= BB_ACK(p
[lo
]);
7714 /* Yes, we can merge with a previous range */
7715 if (s
== a
&& s
+ sectors
>= e
)
7716 /* new range covers old */
7719 ack
= ack
&& acknowledged
;
7721 if (e
< s
+ sectors
)
7723 if (e
- a
<= BB_MAX_LEN
) {
7724 p
[lo
] = BB_MAKE(a
, e
-a
, ack
);
7727 /* does not all fit in one range,
7728 * make p[lo] maximal
7730 if (BB_LEN(p
[lo
]) != BB_MAX_LEN
)
7731 p
[lo
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7737 if (sectors
&& hi
< bb
->count
) {
7738 /* 'hi' points to the first range that starts after 's'.
7739 * Maybe we can merge with the start of that range */
7740 sector_t a
= BB_OFFSET(p
[hi
]);
7741 sector_t e
= a
+ BB_LEN(p
[hi
]);
7742 int ack
= BB_ACK(p
[hi
]);
7743 if (a
<= s
+ sectors
) {
7744 /* merging is possible */
7745 if (e
<= s
+ sectors
) {
7750 ack
= ack
&& acknowledged
;
7753 if (e
- a
<= BB_MAX_LEN
) {
7754 p
[hi
] = BB_MAKE(a
, e
-a
, ack
);
7757 p
[hi
] = BB_MAKE(a
, BB_MAX_LEN
, ack
);
7765 if (sectors
== 0 && hi
< bb
->count
) {
7766 /* we might be able to combine lo and hi */
7767 /* Note: 's' is at the end of 'lo' */
7768 sector_t a
= BB_OFFSET(p
[hi
]);
7769 int lolen
= BB_LEN(p
[lo
]);
7770 int hilen
= BB_LEN(p
[hi
]);
7771 int newlen
= lolen
+ hilen
- (s
- a
);
7772 if (s
>= a
&& newlen
< BB_MAX_LEN
) {
7773 /* yes, we can combine them */
7774 int ack
= BB_ACK(p
[lo
]) && BB_ACK(p
[hi
]);
7775 p
[lo
] = BB_MAKE(BB_OFFSET(p
[lo
]), newlen
, ack
);
7776 memmove(p
+ hi
, p
+ hi
+ 1,
7777 (bb
->count
- hi
- 1) * 8);
7782 /* didn't merge (it all).
7783 * Need to add a range just before 'hi' */
7784 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7785 /* No room for more */
7789 int this_sectors
= sectors
;
7790 memmove(p
+ hi
+ 1, p
+ hi
,
7791 (bb
->count
- hi
) * 8);
7794 if (this_sectors
> BB_MAX_LEN
)
7795 this_sectors
= BB_MAX_LEN
;
7796 p
[hi
] = BB_MAKE(s
, this_sectors
, acknowledged
);
7797 sectors
-= this_sectors
;
7804 bb
->unacked_exist
= 1;
7805 write_sequnlock_irq(&bb
->lock
);
7810 int rdev_set_badblocks(mdk_rdev_t
*rdev
, sector_t s
, int sectors
,
7813 int rv
= md_set_badblocks(&rdev
->badblocks
,
7814 s
+ rdev
->data_offset
, sectors
, acknowledged
);
7816 /* Make sure they get written out promptly */
7817 set_bit(MD_CHANGE_CLEAN
, &rdev
->mddev
->flags
);
7818 md_wakeup_thread(rdev
->mddev
->thread
);
7822 EXPORT_SYMBOL_GPL(rdev_set_badblocks
);
7825 * Remove a range of bad blocks from the table.
7826 * This may involve extending the table if we spilt a region,
7827 * but it must not fail. So if the table becomes full, we just
7828 * drop the remove request.
7830 static int md_clear_badblocks(struct badblocks
*bb
, sector_t s
, int sectors
)
7834 sector_t target
= s
+ sectors
;
7837 if (bb
->shift
> 0) {
7838 /* When clearing we round the start up and the end down.
7839 * This should not matter as the shift should align with
7840 * the block size and no rounding should ever be needed.
7841 * However it is better the think a block is bad when it
7842 * isn't than to think a block is not bad when it is.
7844 s
+= (1<<bb
->shift
) - 1;
7846 target
>>= bb
->shift
;
7847 sectors
= target
- s
;
7850 write_seqlock_irq(&bb
->lock
);
7855 /* Find the last range that starts before 'target' */
7856 while (hi
- lo
> 1) {
7857 int mid
= (lo
+ hi
) / 2;
7858 sector_t a
= BB_OFFSET(p
[mid
]);
7865 /* p[lo] is the last range that could overlap the
7866 * current range. Earlier ranges could also overlap,
7867 * but only this one can overlap the end of the range.
7869 if (BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > target
) {
7870 /* Partial overlap, leave the tail of this range */
7871 int ack
= BB_ACK(p
[lo
]);
7872 sector_t a
= BB_OFFSET(p
[lo
]);
7873 sector_t end
= a
+ BB_LEN(p
[lo
]);
7876 /* we need to split this range */
7877 if (bb
->count
>= MD_MAX_BADBLOCKS
) {
7881 memmove(p
+lo
+1, p
+lo
, (bb
->count
- lo
) * 8);
7883 p
[lo
] = BB_MAKE(a
, s
-a
, ack
);
7886 p
[lo
] = BB_MAKE(target
, end
- target
, ack
);
7887 /* there is no longer an overlap */
7892 BB_OFFSET(p
[lo
]) + BB_LEN(p
[lo
]) > s
) {
7893 /* This range does overlap */
7894 if (BB_OFFSET(p
[lo
]) < s
) {
7895 /* Keep the early parts of this range. */
7896 int ack
= BB_ACK(p
[lo
]);
7897 sector_t start
= BB_OFFSET(p
[lo
]);
7898 p
[lo
] = BB_MAKE(start
, s
- start
, ack
);
7899 /* now low doesn't overlap, so.. */
7904 /* 'lo' is strictly before, 'hi' is strictly after,
7905 * anything between needs to be discarded
7908 memmove(p
+lo
+1, p
+hi
, (bb
->count
- hi
) * 8);
7909 bb
->count
-= (hi
- lo
- 1);
7915 write_sequnlock_irq(&bb
->lock
);
7919 int rdev_clear_badblocks(mdk_rdev_t
*rdev
, sector_t s
, int sectors
)
7921 return md_clear_badblocks(&rdev
->badblocks
,
7922 s
+ rdev
->data_offset
,
7925 EXPORT_SYMBOL_GPL(rdev_clear_badblocks
);
7928 * Acknowledge all bad blocks in a list.
7929 * This only succeeds if ->changed is clear. It is used by
7930 * in-kernel metadata updates
7932 void md_ack_all_badblocks(struct badblocks
*bb
)
7934 if (bb
->page
== NULL
|| bb
->changed
)
7935 /* no point even trying */
7937 write_seqlock_irq(&bb
->lock
);
7939 if (bb
->changed
== 0) {
7942 for (i
= 0; i
< bb
->count
; i
++) {
7943 if (!BB_ACK(p
[i
])) {
7944 sector_t start
= BB_OFFSET(p
[i
]);
7945 int len
= BB_LEN(p
[i
]);
7946 p
[i
] = BB_MAKE(start
, len
, 1);
7949 bb
->unacked_exist
= 0;
7951 write_sequnlock_irq(&bb
->lock
);
7953 EXPORT_SYMBOL_GPL(md_ack_all_badblocks
);
7955 /* sysfs access to bad-blocks list.
7956 * We present two files.
7957 * 'bad-blocks' lists sector numbers and lengths of ranges that
7958 * are recorded as bad. The list is truncated to fit within
7959 * the one-page limit of sysfs.
7960 * Writing "sector length" to this file adds an acknowledged
7962 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7963 * been acknowledged. Writing to this file adds bad blocks
7964 * without acknowledging them. This is largely for testing.
7968 badblocks_show(struct badblocks
*bb
, char *page
, int unack
)
7979 seq
= read_seqbegin(&bb
->lock
);
7984 while (len
< PAGE_SIZE
&& i
< bb
->count
) {
7985 sector_t s
= BB_OFFSET(p
[i
]);
7986 unsigned int length
= BB_LEN(p
[i
]);
7987 int ack
= BB_ACK(p
[i
]);
7993 len
+= snprintf(page
+len
, PAGE_SIZE
-len
, "%llu %u\n",
7994 (unsigned long long)s
<< bb
->shift
,
7995 length
<< bb
->shift
);
7997 if (unack
&& len
== 0)
7998 bb
->unacked_exist
= 0;
8000 if (read_seqretry(&bb
->lock
, seq
))
8009 badblocks_store(struct badblocks
*bb
, const char *page
, size_t len
, int unack
)
8011 unsigned long long sector
;
8015 /* Allow clearing via sysfs *only* for testing/debugging.
8016 * Normally only a successful write may clear a badblock
8019 if (page
[0] == '-') {
8023 #endif /* DO_DEBUG */
8025 switch (sscanf(page
, "%llu %d%c", §or
, &length
, &newline
)) {
8027 if (newline
!= '\n')
8039 md_clear_badblocks(bb
, sector
, length
);
8042 #endif /* DO_DEBUG */
8043 if (md_set_badblocks(bb
, sector
, length
, !unack
))
8049 static int md_notify_reboot(struct notifier_block
*this,
8050 unsigned long code
, void *x
)
8052 struct list_head
*tmp
;
8055 if ((code
== SYS_DOWN
) || (code
== SYS_HALT
) || (code
== SYS_POWER_OFF
)) {
8057 printk(KERN_INFO
"md: stopping all md devices.\n");
8059 for_each_mddev(mddev
, tmp
)
8060 if (mddev_trylock(mddev
)) {
8061 /* Force a switch to readonly even array
8062 * appears to still be in use. Hence
8065 md_set_readonly(mddev
, 100);
8066 mddev_unlock(mddev
);
8069 * certain more exotic SCSI devices are known to be
8070 * volatile wrt too early system reboots. While the
8071 * right place to handle this issue is the given
8072 * driver, we do want to have a safe RAID driver ...
8079 static struct notifier_block md_notifier
= {
8080 .notifier_call
= md_notify_reboot
,
8082 .priority
= INT_MAX
, /* before any real devices */
8085 static void md_geninit(void)
8087 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t
));
8089 proc_create("mdstat", S_IRUGO
, NULL
, &md_seq_fops
);
8092 static int __init
md_init(void)
8096 md_wq
= alloc_workqueue("md", WQ_MEM_RECLAIM
, 0);
8100 md_misc_wq
= alloc_workqueue("md_misc", 0, 0);
8104 if ((ret
= register_blkdev(MD_MAJOR
, "md")) < 0)
8107 if ((ret
= register_blkdev(0, "mdp")) < 0)
8111 blk_register_region(MKDEV(MD_MAJOR
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8112 md_probe
, NULL
, NULL
);
8113 blk_register_region(MKDEV(mdp_major
, 0), 1UL<<MINORBITS
, THIS_MODULE
,
8114 md_probe
, NULL
, NULL
);
8116 register_reboot_notifier(&md_notifier
);
8117 raid_table_header
= register_sysctl_table(raid_root_table
);
8123 unregister_blkdev(MD_MAJOR
, "md");
8125 destroy_workqueue(md_misc_wq
);
8127 destroy_workqueue(md_wq
);
8135 * Searches all registered partitions for autorun RAID arrays
8139 static LIST_HEAD(all_detected_devices
);
8140 struct detected_devices_node
{
8141 struct list_head list
;
8145 void md_autodetect_dev(dev_t dev
)
8147 struct detected_devices_node
*node_detected_dev
;
8149 node_detected_dev
= kzalloc(sizeof(*node_detected_dev
), GFP_KERNEL
);
8150 if (node_detected_dev
) {
8151 node_detected_dev
->dev
= dev
;
8152 list_add_tail(&node_detected_dev
->list
, &all_detected_devices
);
8154 printk(KERN_CRIT
"md: md_autodetect_dev: kzalloc failed"
8155 ", skipping dev(%d,%d)\n", MAJOR(dev
), MINOR(dev
));
8160 static void autostart_arrays(int part
)
8163 struct detected_devices_node
*node_detected_dev
;
8165 int i_scanned
, i_passed
;
8170 printk(KERN_INFO
"md: Autodetecting RAID arrays.\n");
8172 while (!list_empty(&all_detected_devices
) && i_scanned
< INT_MAX
) {
8174 node_detected_dev
= list_entry(all_detected_devices
.next
,
8175 struct detected_devices_node
, list
);
8176 list_del(&node_detected_dev
->list
);
8177 dev
= node_detected_dev
->dev
;
8178 kfree(node_detected_dev
);
8179 rdev
= md_import_device(dev
,0, 90);
8183 if (test_bit(Faulty
, &rdev
->flags
)) {
8187 set_bit(AutoDetected
, &rdev
->flags
);
8188 list_add(&rdev
->same_set
, &pending_raid_disks
);
8192 printk(KERN_INFO
"md: Scanned %d and added %d devices.\n",
8193 i_scanned
, i_passed
);
8195 autorun_devices(part
);
8198 #endif /* !MODULE */
8200 static __exit
void md_exit(void)
8203 struct list_head
*tmp
;
8205 blk_unregister_region(MKDEV(MD_MAJOR
,0), 1U << MINORBITS
);
8206 blk_unregister_region(MKDEV(mdp_major
,0), 1U << MINORBITS
);
8208 unregister_blkdev(MD_MAJOR
,"md");
8209 unregister_blkdev(mdp_major
, "mdp");
8210 unregister_reboot_notifier(&md_notifier
);
8211 unregister_sysctl_table(raid_table_header
);
8212 remove_proc_entry("mdstat", NULL
);
8213 for_each_mddev(mddev
, tmp
) {
8214 export_array(mddev
);
8215 mddev
->hold_active
= 0;
8217 destroy_workqueue(md_misc_wq
);
8218 destroy_workqueue(md_wq
);
8221 subsys_initcall(md_init
);
8222 module_exit(md_exit
)
8224 static int get_ro(char *buffer
, struct kernel_param
*kp
)
8226 return sprintf(buffer
, "%d", start_readonly
);
8228 static int set_ro(const char *val
, struct kernel_param
*kp
)
8231 int num
= simple_strtoul(val
, &e
, 10);
8232 if (*val
&& (*e
== '\0' || *e
== '\n')) {
8233 start_readonly
= num
;
8239 module_param_call(start_ro
, set_ro
, get_ro
, NULL
, S_IRUSR
|S_IWUSR
);
8240 module_param(start_dirty_degraded
, int, S_IRUGO
|S_IWUSR
);
8242 module_param_call(new_array
, add_named_array
, NULL
, NULL
, S_IWUSR
);
8244 EXPORT_SYMBOL(register_md_personality
);
8245 EXPORT_SYMBOL(unregister_md_personality
);
8246 EXPORT_SYMBOL(md_error
);
8247 EXPORT_SYMBOL(md_done_sync
);
8248 EXPORT_SYMBOL(md_write_start
);
8249 EXPORT_SYMBOL(md_write_end
);
8250 EXPORT_SYMBOL(md_register_thread
);
8251 EXPORT_SYMBOL(md_unregister_thread
);
8252 EXPORT_SYMBOL(md_wakeup_thread
);
8253 EXPORT_SYMBOL(md_check_recovery
);
8254 MODULE_LICENSE("GPL");
8255 MODULE_DESCRIPTION("MD RAID framework");
8257 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR
);