1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/major.h>
55 #include <linux/module.h>
56 #include <linux/reboot.h>
57 #include <linux/file.h>
58 #include <linux/compat.h>
59 #include <linux/delay.h>
60 #include <linux/raid/md_p.h>
61 #include <linux/raid/md_u.h>
62 #include <linux/raid/detect.h>
63 #include <linux/slab.h>
64 #include <linux/percpu-refcount.h>
65 #include <linux/part_stat.h>
67 #include <trace/events/block.h>
69 #include "md-bitmap.h"
70 #include "md-cluster.h"
72 /* pers_list is a list of registered personalities protected
74 * pers_lock does extra service to protect accesses to
75 * mddev->thread when the mutex cannot be held.
77 static LIST_HEAD(pers_list);
78 static DEFINE_SPINLOCK(pers_lock);
80 static struct kobj_type md_ktype;
82 struct md_cluster_operations *md_cluster_ops;
83 EXPORT_SYMBOL(md_cluster_ops);
84 static struct module *md_cluster_mod;
86 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
87 static struct workqueue_struct *md_wq;
88 static struct workqueue_struct *md_misc_wq;
89 static struct workqueue_struct *md_rdev_misc_wq;
91 static int remove_and_add_spares(struct mddev *mddev,
92 struct md_rdev *this);
93 static void mddev_detach(struct mddev *mddev);
96 * Default number of read corrections we'll attempt on an rdev
97 * before ejecting it from the array. We divide the read error
98 * count by 2 for every hour elapsed between read errors.
100 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
101 /* Default safemode delay: 200 msec */
102 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
104 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
105 * is 1000 KB/sec, so the extra system load does not show up that much.
106 * Increase it if you want to have more _guaranteed_ speed. Note that
107 * the RAID driver will use the maximum available bandwidth if the IO
108 * subsystem is idle. There is also an 'absolute maximum' reconstruction
109 * speed limit - in case reconstruction slows down your system despite
112 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
113 * or /sys/block/mdX/md/sync_speed_{min,max}
116 static int sysctl_speed_limit_min = 1000;
117 static int sysctl_speed_limit_max = 200000;
118 static inline int speed_min(struct mddev *mddev)
120 return mddev->sync_speed_min ?
121 mddev->sync_speed_min : sysctl_speed_limit_min;
124 static inline int speed_max(struct mddev *mddev)
126 return mddev->sync_speed_max ?
127 mddev->sync_speed_max : sysctl_speed_limit_max;
130 static void rdev_uninit_serial(struct md_rdev *rdev)
132 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
135 kvfree(rdev->serial);
139 static void rdevs_uninit_serial(struct mddev *mddev)
141 struct md_rdev *rdev;
143 rdev_for_each(rdev, mddev)
144 rdev_uninit_serial(rdev);
147 static int rdev_init_serial(struct md_rdev *rdev)
149 /* serial_nums equals with BARRIER_BUCKETS_NR */
150 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
151 struct serial_in_rdev *serial = NULL;
153 if (test_bit(CollisionCheck, &rdev->flags))
156 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
161 for (i = 0; i < serial_nums; i++) {
162 struct serial_in_rdev *serial_tmp = &serial[i];
164 spin_lock_init(&serial_tmp->serial_lock);
165 serial_tmp->serial_rb = RB_ROOT_CACHED;
166 init_waitqueue_head(&serial_tmp->serial_io_wait);
169 rdev->serial = serial;
170 set_bit(CollisionCheck, &rdev->flags);
175 static int rdevs_init_serial(struct mddev *mddev)
177 struct md_rdev *rdev;
180 rdev_for_each(rdev, mddev) {
181 ret = rdev_init_serial(rdev);
186 /* Free all resources if pool is not existed */
187 if (ret && !mddev->serial_info_pool)
188 rdevs_uninit_serial(mddev);
194 * rdev needs to enable serial stuffs if it meets the conditions:
195 * 1. it is multi-queue device flaged with writemostly.
196 * 2. the write-behind mode is enabled.
198 static int rdev_need_serial(struct md_rdev *rdev)
200 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
201 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
202 test_bit(WriteMostly, &rdev->flags));
206 * Init resource for rdev(s), then create serial_info_pool if:
207 * 1. rdev is the first device which return true from rdev_enable_serial.
208 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
210 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
215 if (rdev && !rdev_need_serial(rdev) &&
216 !test_bit(CollisionCheck, &rdev->flags))
220 mddev_suspend(mddev);
223 ret = rdevs_init_serial(mddev);
225 ret = rdev_init_serial(rdev);
229 if (mddev->serial_info_pool == NULL) {
231 * already in memalloc noio context by
234 mddev->serial_info_pool =
235 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
236 sizeof(struct serial_info));
237 if (!mddev->serial_info_pool) {
238 rdevs_uninit_serial(mddev);
239 pr_err("can't alloc memory pool for serialization\n");
249 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
250 * 1. rdev is the last device flaged with CollisionCheck.
251 * 2. when bitmap is destroyed while policy is not enabled.
252 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
254 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
257 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
260 if (mddev->serial_info_pool) {
261 struct md_rdev *temp;
262 int num = 0; /* used to track if other rdevs need the pool */
265 mddev_suspend(mddev);
266 rdev_for_each(temp, mddev) {
268 if (!mddev->serialize_policy ||
269 !rdev_need_serial(temp))
270 rdev_uninit_serial(temp);
273 } else if (temp != rdev &&
274 test_bit(CollisionCheck, &temp->flags))
279 rdev_uninit_serial(rdev);
282 pr_info("The mempool could be used by other devices\n");
284 mempool_destroy(mddev->serial_info_pool);
285 mddev->serial_info_pool = NULL;
292 static struct ctl_table_header *raid_table_header;
294 static struct ctl_table raid_table[] = {
296 .procname = "speed_limit_min",
297 .data = &sysctl_speed_limit_min,
298 .maxlen = sizeof(int),
299 .mode = S_IRUGO|S_IWUSR,
300 .proc_handler = proc_dointvec,
303 .procname = "speed_limit_max",
304 .data = &sysctl_speed_limit_max,
305 .maxlen = sizeof(int),
306 .mode = S_IRUGO|S_IWUSR,
307 .proc_handler = proc_dointvec,
312 static struct ctl_table raid_dir_table[] = {
316 .mode = S_IRUGO|S_IXUGO,
322 static struct ctl_table raid_root_table[] = {
327 .child = raid_dir_table,
332 static int start_readonly;
335 * The original mechanism for creating an md device is to create
336 * a device node in /dev and to open it. This causes races with device-close.
337 * The preferred method is to write to the "new_array" module parameter.
338 * This can avoid races.
339 * Setting create_on_open to false disables the original mechanism
340 * so all the races disappear.
342 static bool create_on_open = true;
345 * We have a system wide 'event count' that is incremented
346 * on any 'interesting' event, and readers of /proc/mdstat
347 * can use 'poll' or 'select' to find out when the event
351 * start array, stop array, error, add device, remove device,
352 * start build, activate spare
354 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
355 static atomic_t md_event_count;
356 void md_new_event(struct mddev *mddev)
358 atomic_inc(&md_event_count);
359 wake_up(&md_event_waiters);
361 EXPORT_SYMBOL_GPL(md_new_event);
364 * Enables to iterate over all existing md arrays
365 * all_mddevs_lock protects this list.
367 static LIST_HEAD(all_mddevs);
368 static DEFINE_SPINLOCK(all_mddevs_lock);
371 * iterates through all used mddevs in the system.
372 * We take care to grab the all_mddevs_lock whenever navigating
373 * the list, and to always hold a refcount when unlocked.
374 * Any code which breaks out of this loop while own
375 * a reference to the current mddev and must mddev_put it.
377 #define for_each_mddev(_mddev,_tmp) \
379 for (({ spin_lock(&all_mddevs_lock); \
380 _tmp = all_mddevs.next; \
382 ({ if (_tmp != &all_mddevs) \
383 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
384 spin_unlock(&all_mddevs_lock); \
385 if (_mddev) mddev_put(_mddev); \
386 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
387 _tmp != &all_mddevs;}); \
388 ({ spin_lock(&all_mddevs_lock); \
389 _tmp = _tmp->next;}) \
392 /* Rather than calling directly into the personality make_request function,
393 * IO requests come here first so that we can check if the device is
394 * being suspended pending a reconfiguration.
395 * We hold a refcount over the call to ->make_request. By the time that
396 * call has finished, the bio has been linked into some internal structure
397 * and so is visible to ->quiesce(), so we don't need the refcount any more.
399 static bool is_suspended(struct mddev *mddev, struct bio *bio)
401 if (mddev->suspended)
403 if (bio_data_dir(bio) != WRITE)
405 if (mddev->suspend_lo >= mddev->suspend_hi)
407 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
409 if (bio_end_sector(bio) < mddev->suspend_lo)
414 void md_handle_request(struct mddev *mddev, struct bio *bio)
418 if (is_suspended(mddev, bio)) {
421 prepare_to_wait(&mddev->sb_wait, &__wait,
422 TASK_UNINTERRUPTIBLE);
423 if (!is_suspended(mddev, bio))
429 finish_wait(&mddev->sb_wait, &__wait);
431 atomic_inc(&mddev->active_io);
434 if (!mddev->pers->make_request(mddev, bio)) {
435 atomic_dec(&mddev->active_io);
436 wake_up(&mddev->sb_wait);
437 goto check_suspended;
440 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
441 wake_up(&mddev->sb_wait);
443 EXPORT_SYMBOL(md_handle_request);
445 static blk_qc_t md_submit_bio(struct bio *bio)
447 const int rw = bio_data_dir(bio);
448 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
450 if (mddev == NULL || mddev->pers == NULL) {
452 return BLK_QC_T_NONE;
455 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
457 return BLK_QC_T_NONE;
460 blk_queue_split(&bio);
462 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
463 if (bio_sectors(bio) != 0)
464 bio->bi_status = BLK_STS_IOERR;
466 return BLK_QC_T_NONE;
469 /* bio could be mergeable after passing to underlayer */
470 bio->bi_opf &= ~REQ_NOMERGE;
472 md_handle_request(mddev, bio);
474 return BLK_QC_T_NONE;
477 /* mddev_suspend makes sure no new requests are submitted
478 * to the device, and that any requests that have been submitted
479 * are completely handled.
480 * Once mddev_detach() is called and completes, the module will be
483 void mddev_suspend(struct mddev *mddev)
485 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
486 lockdep_assert_held(&mddev->reconfig_mutex);
487 if (mddev->suspended++)
490 wake_up(&mddev->sb_wait);
491 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
492 smp_mb__after_atomic();
493 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
494 mddev->pers->quiesce(mddev, 1);
495 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
496 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
498 del_timer_sync(&mddev->safemode_timer);
499 /* restrict memory reclaim I/O during raid array is suspend */
500 mddev->noio_flag = memalloc_noio_save();
502 EXPORT_SYMBOL_GPL(mddev_suspend);
504 void mddev_resume(struct mddev *mddev)
506 /* entred the memalloc scope from mddev_suspend() */
507 memalloc_noio_restore(mddev->noio_flag);
508 lockdep_assert_held(&mddev->reconfig_mutex);
509 if (--mddev->suspended)
511 wake_up(&mddev->sb_wait);
512 mddev->pers->quiesce(mddev, 0);
514 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
515 md_wakeup_thread(mddev->thread);
516 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
518 EXPORT_SYMBOL_GPL(mddev_resume);
521 * Generic flush handling for md
524 static void md_end_flush(struct bio *bio)
526 struct md_rdev *rdev = bio->bi_private;
527 struct mddev *mddev = rdev->mddev;
529 rdev_dec_pending(rdev, mddev);
531 if (atomic_dec_and_test(&mddev->flush_pending)) {
532 /* The pre-request flush has finished */
533 queue_work(md_wq, &mddev->flush_work);
538 static void md_submit_flush_data(struct work_struct *ws);
540 static void submit_flushes(struct work_struct *ws)
542 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
543 struct md_rdev *rdev;
545 mddev->start_flush = ktime_get_boottime();
546 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
547 atomic_set(&mddev->flush_pending, 1);
549 rdev_for_each_rcu(rdev, mddev)
550 if (rdev->raid_disk >= 0 &&
551 !test_bit(Faulty, &rdev->flags)) {
552 /* Take two references, one is dropped
553 * when request finishes, one after
554 * we reclaim rcu_read_lock
557 atomic_inc(&rdev->nr_pending);
558 atomic_inc(&rdev->nr_pending);
560 bi = bio_alloc_bioset(GFP_NOIO, 0, &mddev->bio_set);
561 bi->bi_end_io = md_end_flush;
562 bi->bi_private = rdev;
563 bio_set_dev(bi, rdev->bdev);
564 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
565 atomic_inc(&mddev->flush_pending);
568 rdev_dec_pending(rdev, mddev);
571 if (atomic_dec_and_test(&mddev->flush_pending))
572 queue_work(md_wq, &mddev->flush_work);
575 static void md_submit_flush_data(struct work_struct *ws)
577 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
578 struct bio *bio = mddev->flush_bio;
581 * must reset flush_bio before calling into md_handle_request to avoid a
582 * deadlock, because other bios passed md_handle_request suspend check
583 * could wait for this and below md_handle_request could wait for those
584 * bios because of suspend check
586 spin_lock_irq(&mddev->lock);
587 mddev->prev_flush_start = mddev->start_flush;
588 mddev->flush_bio = NULL;
589 spin_unlock_irq(&mddev->lock);
590 wake_up(&mddev->sb_wait);
592 if (bio->bi_iter.bi_size == 0) {
593 /* an empty barrier - all done */
596 bio->bi_opf &= ~REQ_PREFLUSH;
597 md_handle_request(mddev, bio);
602 * Manages consolidation of flushes and submitting any flushes needed for
603 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
604 * being finished in another context. Returns false if the flushing is
605 * complete but still needs the I/O portion of the bio to be processed.
607 bool md_flush_request(struct mddev *mddev, struct bio *bio)
609 ktime_t req_start = ktime_get_boottime();
610 spin_lock_irq(&mddev->lock);
611 /* flush requests wait until ongoing flush completes,
612 * hence coalescing all the pending requests.
614 wait_event_lock_irq(mddev->sb_wait,
616 ktime_before(req_start, mddev->prev_flush_start),
618 /* new request after previous flush is completed */
619 if (ktime_after(req_start, mddev->prev_flush_start)) {
620 WARN_ON(mddev->flush_bio);
621 mddev->flush_bio = bio;
624 spin_unlock_irq(&mddev->lock);
627 INIT_WORK(&mddev->flush_work, submit_flushes);
628 queue_work(md_wq, &mddev->flush_work);
630 /* flush was performed for some other bio while we waited. */
631 if (bio->bi_iter.bi_size == 0)
632 /* an empty barrier - all done */
635 bio->bi_opf &= ~REQ_PREFLUSH;
641 EXPORT_SYMBOL(md_flush_request);
643 static inline struct mddev *mddev_get(struct mddev *mddev)
645 atomic_inc(&mddev->active);
649 static void mddev_delayed_delete(struct work_struct *ws);
651 static void mddev_put(struct mddev *mddev)
653 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
655 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
656 mddev->ctime == 0 && !mddev->hold_active) {
657 /* Array is not configured at all, and not held active,
659 list_del_init(&mddev->all_mddevs);
662 * Call queue_work inside the spinlock so that
663 * flush_workqueue() after mddev_find will succeed in waiting
664 * for the work to be done.
666 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
667 queue_work(md_misc_wq, &mddev->del_work);
669 spin_unlock(&all_mddevs_lock);
672 static void md_safemode_timeout(struct timer_list *t);
674 void mddev_init(struct mddev *mddev)
676 kobject_init(&mddev->kobj, &md_ktype);
677 mutex_init(&mddev->open_mutex);
678 mutex_init(&mddev->reconfig_mutex);
679 mutex_init(&mddev->bitmap_info.mutex);
680 INIT_LIST_HEAD(&mddev->disks);
681 INIT_LIST_HEAD(&mddev->all_mddevs);
682 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
683 atomic_set(&mddev->active, 1);
684 atomic_set(&mddev->openers, 0);
685 atomic_set(&mddev->active_io, 0);
686 spin_lock_init(&mddev->lock);
687 atomic_set(&mddev->flush_pending, 0);
688 init_waitqueue_head(&mddev->sb_wait);
689 init_waitqueue_head(&mddev->recovery_wait);
690 mddev->reshape_position = MaxSector;
691 mddev->reshape_backwards = 0;
692 mddev->last_sync_action = "none";
693 mddev->resync_min = 0;
694 mddev->resync_max = MaxSector;
695 mddev->level = LEVEL_NONE;
697 EXPORT_SYMBOL_GPL(mddev_init);
699 static struct mddev *mddev_find_locked(dev_t unit)
703 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
704 if (mddev->unit == unit)
710 /* find an unused unit number */
711 static dev_t mddev_alloc_unit(void)
713 static int next_minor = 512;
714 int start = next_minor;
719 dev = MKDEV(MD_MAJOR, next_minor);
721 if (next_minor > MINORMASK)
723 if (next_minor == start)
724 return 0; /* Oh dear, all in use. */
725 is_free = !mddev_find_locked(dev);
731 static struct mddev *mddev_find(dev_t unit)
735 if (MAJOR(unit) != MD_MAJOR)
736 unit &= ~((1 << MdpMinorShift) - 1);
738 spin_lock(&all_mddevs_lock);
739 mddev = mddev_find_locked(unit);
742 spin_unlock(&all_mddevs_lock);
747 static struct mddev *mddev_alloc(dev_t unit)
752 if (unit && MAJOR(unit) != MD_MAJOR)
753 unit &= ~((1 << MdpMinorShift) - 1);
755 new = kzalloc(sizeof(*new), GFP_KERNEL);
757 return ERR_PTR(-ENOMEM);
760 spin_lock(&all_mddevs_lock);
763 if (mddev_find_locked(unit))
766 if (MAJOR(unit) == MD_MAJOR)
767 new->md_minor = MINOR(unit);
769 new->md_minor = MINOR(unit) >> MdpMinorShift;
770 new->hold_active = UNTIL_IOCTL;
773 new->unit = mddev_alloc_unit();
776 new->md_minor = MINOR(new->unit);
777 new->hold_active = UNTIL_STOP;
780 list_add(&new->all_mddevs, &all_mddevs);
781 spin_unlock(&all_mddevs_lock);
784 spin_unlock(&all_mddevs_lock);
786 return ERR_PTR(error);
789 static const struct attribute_group md_redundancy_group;
791 void mddev_unlock(struct mddev *mddev)
793 if (mddev->to_remove) {
794 /* These cannot be removed under reconfig_mutex as
795 * an access to the files will try to take reconfig_mutex
796 * while holding the file unremovable, which leads to
798 * So hold set sysfs_active while the remove in happeing,
799 * and anything else which might set ->to_remove or my
800 * otherwise change the sysfs namespace will fail with
801 * -EBUSY if sysfs_active is still set.
802 * We set sysfs_active under reconfig_mutex and elsewhere
803 * test it under the same mutex to ensure its correct value
806 const struct attribute_group *to_remove = mddev->to_remove;
807 mddev->to_remove = NULL;
808 mddev->sysfs_active = 1;
809 mutex_unlock(&mddev->reconfig_mutex);
811 if (mddev->kobj.sd) {
812 if (to_remove != &md_redundancy_group)
813 sysfs_remove_group(&mddev->kobj, to_remove);
814 if (mddev->pers == NULL ||
815 mddev->pers->sync_request == NULL) {
816 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
817 if (mddev->sysfs_action)
818 sysfs_put(mddev->sysfs_action);
819 if (mddev->sysfs_completed)
820 sysfs_put(mddev->sysfs_completed);
821 if (mddev->sysfs_degraded)
822 sysfs_put(mddev->sysfs_degraded);
823 mddev->sysfs_action = NULL;
824 mddev->sysfs_completed = NULL;
825 mddev->sysfs_degraded = NULL;
828 mddev->sysfs_active = 0;
830 mutex_unlock(&mddev->reconfig_mutex);
832 /* As we've dropped the mutex we need a spinlock to
833 * make sure the thread doesn't disappear
835 spin_lock(&pers_lock);
836 md_wakeup_thread(mddev->thread);
837 wake_up(&mddev->sb_wait);
838 spin_unlock(&pers_lock);
840 EXPORT_SYMBOL_GPL(mddev_unlock);
842 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
844 struct md_rdev *rdev;
846 rdev_for_each_rcu(rdev, mddev)
847 if (rdev->desc_nr == nr)
852 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
854 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
856 struct md_rdev *rdev;
858 rdev_for_each(rdev, mddev)
859 if (rdev->bdev->bd_dev == dev)
865 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
867 struct md_rdev *rdev;
869 rdev_for_each_rcu(rdev, mddev)
870 if (rdev->bdev->bd_dev == dev)
875 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
877 static struct md_personality *find_pers(int level, char *clevel)
879 struct md_personality *pers;
880 list_for_each_entry(pers, &pers_list, list) {
881 if (level != LEVEL_NONE && pers->level == level)
883 if (strcmp(pers->name, clevel)==0)
889 /* return the offset of the super block in 512byte sectors */
890 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
892 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
893 return MD_NEW_SIZE_SECTORS(num_sectors);
896 static int alloc_disk_sb(struct md_rdev *rdev)
898 rdev->sb_page = alloc_page(GFP_KERNEL);
904 void md_rdev_clear(struct md_rdev *rdev)
907 put_page(rdev->sb_page);
909 rdev->sb_page = NULL;
914 put_page(rdev->bb_page);
915 rdev->bb_page = NULL;
917 badblocks_exit(&rdev->badblocks);
919 EXPORT_SYMBOL_GPL(md_rdev_clear);
921 static void super_written(struct bio *bio)
923 struct md_rdev *rdev = bio->bi_private;
924 struct mddev *mddev = rdev->mddev;
926 if (bio->bi_status) {
927 pr_err("md: %s gets error=%d\n", __func__,
928 blk_status_to_errno(bio->bi_status));
929 md_error(mddev, rdev);
930 if (!test_bit(Faulty, &rdev->flags)
931 && (bio->bi_opf & MD_FAILFAST)) {
932 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
933 set_bit(LastDev, &rdev->flags);
936 clear_bit(LastDev, &rdev->flags);
938 if (atomic_dec_and_test(&mddev->pending_writes))
939 wake_up(&mddev->sb_wait);
940 rdev_dec_pending(rdev, mddev);
944 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
945 sector_t sector, int size, struct page *page)
947 /* write first size bytes of page to sector of rdev
948 * Increment mddev->pending_writes before returning
949 * and decrement it on completion, waking up sb_wait
950 * if zero is reached.
951 * If an error occurred, call md_error
959 if (test_bit(Faulty, &rdev->flags))
962 bio = bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
964 atomic_inc(&rdev->nr_pending);
966 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
967 bio->bi_iter.bi_sector = sector;
968 bio_add_page(bio, page, size, 0);
969 bio->bi_private = rdev;
970 bio->bi_end_io = super_written;
972 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
973 test_bit(FailFast, &rdev->flags) &&
974 !test_bit(LastDev, &rdev->flags))
976 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
978 atomic_inc(&mddev->pending_writes);
982 int md_super_wait(struct mddev *mddev)
984 /* wait for all superblock writes that were scheduled to complete */
985 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
986 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
991 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
992 struct page *page, int op, int op_flags, bool metadata_op)
997 bio_init(&bio, &bvec, 1);
999 if (metadata_op && rdev->meta_bdev)
1000 bio_set_dev(&bio, rdev->meta_bdev);
1002 bio_set_dev(&bio, rdev->bdev);
1003 bio.bi_opf = op | op_flags;
1005 bio.bi_iter.bi_sector = sector + rdev->sb_start;
1006 else if (rdev->mddev->reshape_position != MaxSector &&
1007 (rdev->mddev->reshape_backwards ==
1008 (sector >= rdev->mddev->reshape_position)))
1009 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1011 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1012 bio_add_page(&bio, page, size, 0);
1014 submit_bio_wait(&bio);
1016 return !bio.bi_status;
1018 EXPORT_SYMBOL_GPL(sync_page_io);
1020 static int read_disk_sb(struct md_rdev *rdev, int size)
1022 char b[BDEVNAME_SIZE];
1024 if (rdev->sb_loaded)
1027 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1029 rdev->sb_loaded = 1;
1033 pr_err("md: disabled device %s, could not read superblock.\n",
1034 bdevname(rdev->bdev,b));
1038 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1040 return sb1->set_uuid0 == sb2->set_uuid0 &&
1041 sb1->set_uuid1 == sb2->set_uuid1 &&
1042 sb1->set_uuid2 == sb2->set_uuid2 &&
1043 sb1->set_uuid3 == sb2->set_uuid3;
1046 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1049 mdp_super_t *tmp1, *tmp2;
1051 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1052 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1054 if (!tmp1 || !tmp2) {
1063 * nr_disks is not constant
1068 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1075 static u32 md_csum_fold(u32 csum)
1077 csum = (csum & 0xffff) + (csum >> 16);
1078 return (csum & 0xffff) + (csum >> 16);
1081 static unsigned int calc_sb_csum(mdp_super_t *sb)
1084 u32 *sb32 = (u32*)sb;
1086 unsigned int disk_csum, csum;
1088 disk_csum = sb->sb_csum;
1091 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1093 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1096 /* This used to use csum_partial, which was wrong for several
1097 * reasons including that different results are returned on
1098 * different architectures. It isn't critical that we get exactly
1099 * the same return value as before (we always csum_fold before
1100 * testing, and that removes any differences). However as we
1101 * know that csum_partial always returned a 16bit value on
1102 * alphas, do a fold to maximise conformity to previous behaviour.
1104 sb->sb_csum = md_csum_fold(disk_csum);
1106 sb->sb_csum = disk_csum;
1112 * Handle superblock details.
1113 * We want to be able to handle multiple superblock formats
1114 * so we have a common interface to them all, and an array of
1115 * different handlers.
1116 * We rely on user-space to write the initial superblock, and support
1117 * reading and updating of superblocks.
1118 * Interface methods are:
1119 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1120 * loads and validates a superblock on dev.
1121 * if refdev != NULL, compare superblocks on both devices
1123 * 0 - dev has a superblock that is compatible with refdev
1124 * 1 - dev has a superblock that is compatible and newer than refdev
1125 * so dev should be used as the refdev in future
1126 * -EINVAL superblock incompatible or invalid
1127 * -othererror e.g. -EIO
1129 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1130 * Verify that dev is acceptable into mddev.
1131 * The first time, mddev->raid_disks will be 0, and data from
1132 * dev should be merged in. Subsequent calls check that dev
1133 * is new enough. Return 0 or -EINVAL
1135 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1136 * Update the superblock for rdev with data in mddev
1137 * This does not write to disc.
1143 struct module *owner;
1144 int (*load_super)(struct md_rdev *rdev,
1145 struct md_rdev *refdev,
1147 int (*validate_super)(struct mddev *mddev,
1148 struct md_rdev *rdev);
1149 void (*sync_super)(struct mddev *mddev,
1150 struct md_rdev *rdev);
1151 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1152 sector_t num_sectors);
1153 int (*allow_new_offset)(struct md_rdev *rdev,
1154 unsigned long long new_offset);
1158 * Check that the given mddev has no bitmap.
1160 * This function is called from the run method of all personalities that do not
1161 * support bitmaps. It prints an error message and returns non-zero if mddev
1162 * has a bitmap. Otherwise, it returns 0.
1165 int md_check_no_bitmap(struct mddev *mddev)
1167 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1169 pr_warn("%s: bitmaps are not supported for %s\n",
1170 mdname(mddev), mddev->pers->name);
1173 EXPORT_SYMBOL(md_check_no_bitmap);
1176 * load_super for 0.90.0
1178 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1180 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1183 bool spare_disk = true;
1186 * Calculate the position of the superblock (512byte sectors),
1187 * it's at the end of the disk.
1189 * It also happens to be a multiple of 4Kb.
1191 rdev->sb_start = calc_dev_sboffset(rdev);
1193 ret = read_disk_sb(rdev, MD_SB_BYTES);
1199 bdevname(rdev->bdev, b);
1200 sb = page_address(rdev->sb_page);
1202 if (sb->md_magic != MD_SB_MAGIC) {
1203 pr_warn("md: invalid raid superblock magic on %s\n", b);
1207 if (sb->major_version != 0 ||
1208 sb->minor_version < 90 ||
1209 sb->minor_version > 91) {
1210 pr_warn("Bad version number %d.%d on %s\n",
1211 sb->major_version, sb->minor_version, b);
1215 if (sb->raid_disks <= 0)
1218 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1219 pr_warn("md: invalid superblock checksum on %s\n", b);
1223 rdev->preferred_minor = sb->md_minor;
1224 rdev->data_offset = 0;
1225 rdev->new_data_offset = 0;
1226 rdev->sb_size = MD_SB_BYTES;
1227 rdev->badblocks.shift = -1;
1229 if (sb->level == LEVEL_MULTIPATH)
1232 rdev->desc_nr = sb->this_disk.number;
1234 /* not spare disk, or LEVEL_MULTIPATH */
1235 if (sb->level == LEVEL_MULTIPATH ||
1236 (rdev->desc_nr >= 0 &&
1237 rdev->desc_nr < MD_SB_DISKS &&
1238 sb->disks[rdev->desc_nr].state &
1239 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1249 mdp_super_t *refsb = page_address(refdev->sb_page);
1250 if (!md_uuid_equal(refsb, sb)) {
1251 pr_warn("md: %s has different UUID to %s\n",
1252 b, bdevname(refdev->bdev,b2));
1255 if (!md_sb_equal(refsb, sb)) {
1256 pr_warn("md: %s has same UUID but different superblock to %s\n",
1257 b, bdevname(refdev->bdev, b2));
1261 ev2 = md_event(refsb);
1263 if (!spare_disk && ev1 > ev2)
1268 rdev->sectors = rdev->sb_start;
1269 /* Limit to 4TB as metadata cannot record more than that.
1270 * (not needed for Linear and RAID0 as metadata doesn't
1273 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1274 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1276 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1277 /* "this cannot possibly happen" ... */
1285 * validate_super for 0.90.0
1287 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1290 mdp_super_t *sb = page_address(rdev->sb_page);
1291 __u64 ev1 = md_event(sb);
1293 rdev->raid_disk = -1;
1294 clear_bit(Faulty, &rdev->flags);
1295 clear_bit(In_sync, &rdev->flags);
1296 clear_bit(Bitmap_sync, &rdev->flags);
1297 clear_bit(WriteMostly, &rdev->flags);
1299 if (mddev->raid_disks == 0) {
1300 mddev->major_version = 0;
1301 mddev->minor_version = sb->minor_version;
1302 mddev->patch_version = sb->patch_version;
1303 mddev->external = 0;
1304 mddev->chunk_sectors = sb->chunk_size >> 9;
1305 mddev->ctime = sb->ctime;
1306 mddev->utime = sb->utime;
1307 mddev->level = sb->level;
1308 mddev->clevel[0] = 0;
1309 mddev->layout = sb->layout;
1310 mddev->raid_disks = sb->raid_disks;
1311 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1312 mddev->events = ev1;
1313 mddev->bitmap_info.offset = 0;
1314 mddev->bitmap_info.space = 0;
1315 /* bitmap can use 60 K after the 4K superblocks */
1316 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1317 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1318 mddev->reshape_backwards = 0;
1320 if (mddev->minor_version >= 91) {
1321 mddev->reshape_position = sb->reshape_position;
1322 mddev->delta_disks = sb->delta_disks;
1323 mddev->new_level = sb->new_level;
1324 mddev->new_layout = sb->new_layout;
1325 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1326 if (mddev->delta_disks < 0)
1327 mddev->reshape_backwards = 1;
1329 mddev->reshape_position = MaxSector;
1330 mddev->delta_disks = 0;
1331 mddev->new_level = mddev->level;
1332 mddev->new_layout = mddev->layout;
1333 mddev->new_chunk_sectors = mddev->chunk_sectors;
1335 if (mddev->level == 0)
1338 if (sb->state & (1<<MD_SB_CLEAN))
1339 mddev->recovery_cp = MaxSector;
1341 if (sb->events_hi == sb->cp_events_hi &&
1342 sb->events_lo == sb->cp_events_lo) {
1343 mddev->recovery_cp = sb->recovery_cp;
1345 mddev->recovery_cp = 0;
1348 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1349 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1350 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1351 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1353 mddev->max_disks = MD_SB_DISKS;
1355 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1356 mddev->bitmap_info.file == NULL) {
1357 mddev->bitmap_info.offset =
1358 mddev->bitmap_info.default_offset;
1359 mddev->bitmap_info.space =
1360 mddev->bitmap_info.default_space;
1363 } else if (mddev->pers == NULL) {
1364 /* Insist on good event counter while assembling, except
1365 * for spares (which don't need an event count) */
1367 if (sb->disks[rdev->desc_nr].state & (
1368 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1369 if (ev1 < mddev->events)
1371 } else if (mddev->bitmap) {
1372 /* if adding to array with a bitmap, then we can accept an
1373 * older device ... but not too old.
1375 if (ev1 < mddev->bitmap->events_cleared)
1377 if (ev1 < mddev->events)
1378 set_bit(Bitmap_sync, &rdev->flags);
1380 if (ev1 < mddev->events)
1381 /* just a hot-add of a new device, leave raid_disk at -1 */
1385 if (mddev->level != LEVEL_MULTIPATH) {
1386 desc = sb->disks + rdev->desc_nr;
1388 if (desc->state & (1<<MD_DISK_FAULTY))
1389 set_bit(Faulty, &rdev->flags);
1390 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1391 desc->raid_disk < mddev->raid_disks */) {
1392 set_bit(In_sync, &rdev->flags);
1393 rdev->raid_disk = desc->raid_disk;
1394 rdev->saved_raid_disk = desc->raid_disk;
1395 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1396 /* active but not in sync implies recovery up to
1397 * reshape position. We don't know exactly where
1398 * that is, so set to zero for now */
1399 if (mddev->minor_version >= 91) {
1400 rdev->recovery_offset = 0;
1401 rdev->raid_disk = desc->raid_disk;
1404 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1405 set_bit(WriteMostly, &rdev->flags);
1406 if (desc->state & (1<<MD_DISK_FAILFAST))
1407 set_bit(FailFast, &rdev->flags);
1408 } else /* MULTIPATH are always insync */
1409 set_bit(In_sync, &rdev->flags);
1414 * sync_super for 0.90.0
1416 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1419 struct md_rdev *rdev2;
1420 int next_spare = mddev->raid_disks;
1422 /* make rdev->sb match mddev data..
1425 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1426 * 3/ any empty disks < next_spare become removed
1428 * disks[0] gets initialised to REMOVED because
1429 * we cannot be sure from other fields if it has
1430 * been initialised or not.
1433 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1435 rdev->sb_size = MD_SB_BYTES;
1437 sb = page_address(rdev->sb_page);
1439 memset(sb, 0, sizeof(*sb));
1441 sb->md_magic = MD_SB_MAGIC;
1442 sb->major_version = mddev->major_version;
1443 sb->patch_version = mddev->patch_version;
1444 sb->gvalid_words = 0; /* ignored */
1445 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1446 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1447 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1448 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1450 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1451 sb->level = mddev->level;
1452 sb->size = mddev->dev_sectors / 2;
1453 sb->raid_disks = mddev->raid_disks;
1454 sb->md_minor = mddev->md_minor;
1455 sb->not_persistent = 0;
1456 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1458 sb->events_hi = (mddev->events>>32);
1459 sb->events_lo = (u32)mddev->events;
1461 if (mddev->reshape_position == MaxSector)
1462 sb->minor_version = 90;
1464 sb->minor_version = 91;
1465 sb->reshape_position = mddev->reshape_position;
1466 sb->new_level = mddev->new_level;
1467 sb->delta_disks = mddev->delta_disks;
1468 sb->new_layout = mddev->new_layout;
1469 sb->new_chunk = mddev->new_chunk_sectors << 9;
1471 mddev->minor_version = sb->minor_version;
1474 sb->recovery_cp = mddev->recovery_cp;
1475 sb->cp_events_hi = (mddev->events>>32);
1476 sb->cp_events_lo = (u32)mddev->events;
1477 if (mddev->recovery_cp == MaxSector)
1478 sb->state = (1<< MD_SB_CLEAN);
1480 sb->recovery_cp = 0;
1482 sb->layout = mddev->layout;
1483 sb->chunk_size = mddev->chunk_sectors << 9;
1485 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1486 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1488 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1489 rdev_for_each(rdev2, mddev) {
1492 int is_active = test_bit(In_sync, &rdev2->flags);
1494 if (rdev2->raid_disk >= 0 &&
1495 sb->minor_version >= 91)
1496 /* we have nowhere to store the recovery_offset,
1497 * but if it is not below the reshape_position,
1498 * we can piggy-back on that.
1501 if (rdev2->raid_disk < 0 ||
1502 test_bit(Faulty, &rdev2->flags))
1505 desc_nr = rdev2->raid_disk;
1507 desc_nr = next_spare++;
1508 rdev2->desc_nr = desc_nr;
1509 d = &sb->disks[rdev2->desc_nr];
1511 d->number = rdev2->desc_nr;
1512 d->major = MAJOR(rdev2->bdev->bd_dev);
1513 d->minor = MINOR(rdev2->bdev->bd_dev);
1515 d->raid_disk = rdev2->raid_disk;
1517 d->raid_disk = rdev2->desc_nr; /* compatibility */
1518 if (test_bit(Faulty, &rdev2->flags))
1519 d->state = (1<<MD_DISK_FAULTY);
1520 else if (is_active) {
1521 d->state = (1<<MD_DISK_ACTIVE);
1522 if (test_bit(In_sync, &rdev2->flags))
1523 d->state |= (1<<MD_DISK_SYNC);
1531 if (test_bit(WriteMostly, &rdev2->flags))
1532 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1533 if (test_bit(FailFast, &rdev2->flags))
1534 d->state |= (1<<MD_DISK_FAILFAST);
1536 /* now set the "removed" and "faulty" bits on any missing devices */
1537 for (i=0 ; i < mddev->raid_disks ; i++) {
1538 mdp_disk_t *d = &sb->disks[i];
1539 if (d->state == 0 && d->number == 0) {
1542 d->state = (1<<MD_DISK_REMOVED);
1543 d->state |= (1<<MD_DISK_FAULTY);
1547 sb->nr_disks = nr_disks;
1548 sb->active_disks = active;
1549 sb->working_disks = working;
1550 sb->failed_disks = failed;
1551 sb->spare_disks = spare;
1553 sb->this_disk = sb->disks[rdev->desc_nr];
1554 sb->sb_csum = calc_sb_csum(sb);
1558 * rdev_size_change for 0.90.0
1560 static unsigned long long
1561 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1563 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1564 return 0; /* component must fit device */
1565 if (rdev->mddev->bitmap_info.offset)
1566 return 0; /* can't move bitmap */
1567 rdev->sb_start = calc_dev_sboffset(rdev);
1568 if (!num_sectors || num_sectors > rdev->sb_start)
1569 num_sectors = rdev->sb_start;
1570 /* Limit to 4TB as metadata cannot record more than that.
1571 * 4TB == 2^32 KB, or 2*2^32 sectors.
1573 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1574 num_sectors = (sector_t)(2ULL << 32) - 2;
1576 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1578 } while (md_super_wait(rdev->mddev) < 0);
1583 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1585 /* non-zero offset changes not possible with v0.90 */
1586 return new_offset == 0;
1590 * version 1 superblock
1593 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1597 unsigned long long newcsum;
1598 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1599 __le32 *isuper = (__le32*)sb;
1601 disk_csum = sb->sb_csum;
1604 for (; size >= 4; size -= 4)
1605 newcsum += le32_to_cpu(*isuper++);
1608 newcsum += le16_to_cpu(*(__le16*) isuper);
1610 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1611 sb->sb_csum = disk_csum;
1612 return cpu_to_le32(csum);
1615 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1617 struct mdp_superblock_1 *sb;
1621 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1623 bool spare_disk = true;
1626 * Calculate the position of the superblock in 512byte sectors.
1627 * It is always aligned to a 4K boundary and
1628 * depeding on minor_version, it can be:
1629 * 0: At least 8K, but less than 12K, from end of device
1630 * 1: At start of device
1631 * 2: 4K from start of device.
1633 switch(minor_version) {
1635 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1637 sb_start &= ~(sector_t)(4*2-1);
1648 rdev->sb_start = sb_start;
1650 /* superblock is rarely larger than 1K, but it can be larger,
1651 * and it is safe to read 4k, so we do that
1653 ret = read_disk_sb(rdev, 4096);
1654 if (ret) return ret;
1656 sb = page_address(rdev->sb_page);
1658 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1659 sb->major_version != cpu_to_le32(1) ||
1660 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1661 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1662 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1665 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1666 pr_warn("md: invalid superblock checksum on %s\n",
1667 bdevname(rdev->bdev,b));
1670 if (le64_to_cpu(sb->data_size) < 10) {
1671 pr_warn("md: data_size too small on %s\n",
1672 bdevname(rdev->bdev,b));
1677 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1678 /* Some padding is non-zero, might be a new feature */
1681 rdev->preferred_minor = 0xffff;
1682 rdev->data_offset = le64_to_cpu(sb->data_offset);
1683 rdev->new_data_offset = rdev->data_offset;
1684 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1685 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1686 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1687 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1689 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1690 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1691 if (rdev->sb_size & bmask)
1692 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1695 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1698 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1701 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1704 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1706 if (!rdev->bb_page) {
1707 rdev->bb_page = alloc_page(GFP_KERNEL);
1711 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1712 rdev->badblocks.count == 0) {
1713 /* need to load the bad block list.
1714 * Currently we limit it to one page.
1720 int sectors = le16_to_cpu(sb->bblog_size);
1721 if (sectors > (PAGE_SIZE / 512))
1723 offset = le32_to_cpu(sb->bblog_offset);
1726 bb_sector = (long long)offset;
1727 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1728 rdev->bb_page, REQ_OP_READ, 0, true))
1730 bbp = (__le64 *)page_address(rdev->bb_page);
1731 rdev->badblocks.shift = sb->bblog_shift;
1732 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1733 u64 bb = le64_to_cpu(*bbp);
1734 int count = bb & (0x3ff);
1735 u64 sector = bb >> 10;
1736 sector <<= sb->bblog_shift;
1737 count <<= sb->bblog_shift;
1740 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1743 } else if (sb->bblog_offset != 0)
1744 rdev->badblocks.shift = 0;
1746 if ((le32_to_cpu(sb->feature_map) &
1747 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1748 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1749 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1750 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1753 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1757 /* not spare disk, or LEVEL_MULTIPATH */
1758 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1759 (rdev->desc_nr >= 0 &&
1760 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1761 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1762 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1772 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1774 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1775 sb->level != refsb->level ||
1776 sb->layout != refsb->layout ||
1777 sb->chunksize != refsb->chunksize) {
1778 pr_warn("md: %s has strangely different superblock to %s\n",
1779 bdevname(rdev->bdev,b),
1780 bdevname(refdev->bdev,b2));
1783 ev1 = le64_to_cpu(sb->events);
1784 ev2 = le64_to_cpu(refsb->events);
1786 if (!spare_disk && ev1 > ev2)
1791 if (minor_version) {
1792 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1793 sectors -= rdev->data_offset;
1795 sectors = rdev->sb_start;
1796 if (sectors < le64_to_cpu(sb->data_size))
1798 rdev->sectors = le64_to_cpu(sb->data_size);
1802 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1804 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1805 __u64 ev1 = le64_to_cpu(sb->events);
1807 rdev->raid_disk = -1;
1808 clear_bit(Faulty, &rdev->flags);
1809 clear_bit(In_sync, &rdev->flags);
1810 clear_bit(Bitmap_sync, &rdev->flags);
1811 clear_bit(WriteMostly, &rdev->flags);
1813 if (mddev->raid_disks == 0) {
1814 mddev->major_version = 1;
1815 mddev->patch_version = 0;
1816 mddev->external = 0;
1817 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1818 mddev->ctime = le64_to_cpu(sb->ctime);
1819 mddev->utime = le64_to_cpu(sb->utime);
1820 mddev->level = le32_to_cpu(sb->level);
1821 mddev->clevel[0] = 0;
1822 mddev->layout = le32_to_cpu(sb->layout);
1823 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1824 mddev->dev_sectors = le64_to_cpu(sb->size);
1825 mddev->events = ev1;
1826 mddev->bitmap_info.offset = 0;
1827 mddev->bitmap_info.space = 0;
1828 /* Default location for bitmap is 1K after superblock
1829 * using 3K - total of 4K
1831 mddev->bitmap_info.default_offset = 1024 >> 9;
1832 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1833 mddev->reshape_backwards = 0;
1835 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1836 memcpy(mddev->uuid, sb->set_uuid, 16);
1838 mddev->max_disks = (4096-256)/2;
1840 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1841 mddev->bitmap_info.file == NULL) {
1842 mddev->bitmap_info.offset =
1843 (__s32)le32_to_cpu(sb->bitmap_offset);
1844 /* Metadata doesn't record how much space is available.
1845 * For 1.0, we assume we can use up to the superblock
1846 * if before, else to 4K beyond superblock.
1847 * For others, assume no change is possible.
1849 if (mddev->minor_version > 0)
1850 mddev->bitmap_info.space = 0;
1851 else if (mddev->bitmap_info.offset > 0)
1852 mddev->bitmap_info.space =
1853 8 - mddev->bitmap_info.offset;
1855 mddev->bitmap_info.space =
1856 -mddev->bitmap_info.offset;
1859 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1860 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1861 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1862 mddev->new_level = le32_to_cpu(sb->new_level);
1863 mddev->new_layout = le32_to_cpu(sb->new_layout);
1864 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1865 if (mddev->delta_disks < 0 ||
1866 (mddev->delta_disks == 0 &&
1867 (le32_to_cpu(sb->feature_map)
1868 & MD_FEATURE_RESHAPE_BACKWARDS)))
1869 mddev->reshape_backwards = 1;
1871 mddev->reshape_position = MaxSector;
1872 mddev->delta_disks = 0;
1873 mddev->new_level = mddev->level;
1874 mddev->new_layout = mddev->layout;
1875 mddev->new_chunk_sectors = mddev->chunk_sectors;
1878 if (mddev->level == 0 &&
1879 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1882 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1883 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1885 if (le32_to_cpu(sb->feature_map) &
1886 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1887 if (le32_to_cpu(sb->feature_map) &
1888 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1890 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1891 (le32_to_cpu(sb->feature_map) &
1892 MD_FEATURE_MULTIPLE_PPLS))
1894 set_bit(MD_HAS_PPL, &mddev->flags);
1896 } else if (mddev->pers == NULL) {
1897 /* Insist of good event counter while assembling, except for
1898 * spares (which don't need an event count) */
1900 if (rdev->desc_nr >= 0 &&
1901 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1902 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1903 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1904 if (ev1 < mddev->events)
1906 } else if (mddev->bitmap) {
1907 /* If adding to array with a bitmap, then we can accept an
1908 * older device, but not too old.
1910 if (ev1 < mddev->bitmap->events_cleared)
1912 if (ev1 < mddev->events)
1913 set_bit(Bitmap_sync, &rdev->flags);
1915 if (ev1 < mddev->events)
1916 /* just a hot-add of a new device, leave raid_disk at -1 */
1919 if (mddev->level != LEVEL_MULTIPATH) {
1921 if (rdev->desc_nr < 0 ||
1922 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1923 role = MD_DISK_ROLE_SPARE;
1926 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1928 case MD_DISK_ROLE_SPARE: /* spare */
1930 case MD_DISK_ROLE_FAULTY: /* faulty */
1931 set_bit(Faulty, &rdev->flags);
1933 case MD_DISK_ROLE_JOURNAL: /* journal device */
1934 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1935 /* journal device without journal feature */
1936 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1939 set_bit(Journal, &rdev->flags);
1940 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1941 rdev->raid_disk = 0;
1944 rdev->saved_raid_disk = role;
1945 if ((le32_to_cpu(sb->feature_map) &
1946 MD_FEATURE_RECOVERY_OFFSET)) {
1947 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1948 if (!(le32_to_cpu(sb->feature_map) &
1949 MD_FEATURE_RECOVERY_BITMAP))
1950 rdev->saved_raid_disk = -1;
1953 * If the array is FROZEN, then the device can't
1954 * be in_sync with rest of array.
1956 if (!test_bit(MD_RECOVERY_FROZEN,
1958 set_bit(In_sync, &rdev->flags);
1960 rdev->raid_disk = role;
1963 if (sb->devflags & WriteMostly1)
1964 set_bit(WriteMostly, &rdev->flags);
1965 if (sb->devflags & FailFast1)
1966 set_bit(FailFast, &rdev->flags);
1967 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1968 set_bit(Replacement, &rdev->flags);
1969 } else /* MULTIPATH are always insync */
1970 set_bit(In_sync, &rdev->flags);
1975 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1977 struct mdp_superblock_1 *sb;
1978 struct md_rdev *rdev2;
1980 /* make rdev->sb match mddev and rdev data. */
1982 sb = page_address(rdev->sb_page);
1984 sb->feature_map = 0;
1986 sb->recovery_offset = cpu_to_le64(0);
1987 memset(sb->pad3, 0, sizeof(sb->pad3));
1989 sb->utime = cpu_to_le64((__u64)mddev->utime);
1990 sb->events = cpu_to_le64(mddev->events);
1992 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1993 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1994 sb->resync_offset = cpu_to_le64(MaxSector);
1996 sb->resync_offset = cpu_to_le64(0);
1998 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2000 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2001 sb->size = cpu_to_le64(mddev->dev_sectors);
2002 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2003 sb->level = cpu_to_le32(mddev->level);
2004 sb->layout = cpu_to_le32(mddev->layout);
2005 if (test_bit(FailFast, &rdev->flags))
2006 sb->devflags |= FailFast1;
2008 sb->devflags &= ~FailFast1;
2010 if (test_bit(WriteMostly, &rdev->flags))
2011 sb->devflags |= WriteMostly1;
2013 sb->devflags &= ~WriteMostly1;
2014 sb->data_offset = cpu_to_le64(rdev->data_offset);
2015 sb->data_size = cpu_to_le64(rdev->sectors);
2017 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2018 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2019 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2022 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2023 !test_bit(In_sync, &rdev->flags)) {
2025 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2026 sb->recovery_offset =
2027 cpu_to_le64(rdev->recovery_offset);
2028 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2030 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2032 /* Note: recovery_offset and journal_tail share space */
2033 if (test_bit(Journal, &rdev->flags))
2034 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2035 if (test_bit(Replacement, &rdev->flags))
2037 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2039 if (mddev->reshape_position != MaxSector) {
2040 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2041 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2042 sb->new_layout = cpu_to_le32(mddev->new_layout);
2043 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2044 sb->new_level = cpu_to_le32(mddev->new_level);
2045 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2046 if (mddev->delta_disks == 0 &&
2047 mddev->reshape_backwards)
2049 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2050 if (rdev->new_data_offset != rdev->data_offset) {
2052 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2053 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2054 - rdev->data_offset));
2058 if (mddev_is_clustered(mddev))
2059 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2061 if (rdev->badblocks.count == 0)
2062 /* Nothing to do for bad blocks*/ ;
2063 else if (sb->bblog_offset == 0)
2064 /* Cannot record bad blocks on this device */
2065 md_error(mddev, rdev);
2067 struct badblocks *bb = &rdev->badblocks;
2068 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2070 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2075 seq = read_seqbegin(&bb->lock);
2077 memset(bbp, 0xff, PAGE_SIZE);
2079 for (i = 0 ; i < bb->count ; i++) {
2080 u64 internal_bb = p[i];
2081 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2082 | BB_LEN(internal_bb));
2083 bbp[i] = cpu_to_le64(store_bb);
2086 if (read_seqretry(&bb->lock, seq))
2089 bb->sector = (rdev->sb_start +
2090 (int)le32_to_cpu(sb->bblog_offset));
2091 bb->size = le16_to_cpu(sb->bblog_size);
2096 rdev_for_each(rdev2, mddev)
2097 if (rdev2->desc_nr+1 > max_dev)
2098 max_dev = rdev2->desc_nr+1;
2100 if (max_dev > le32_to_cpu(sb->max_dev)) {
2102 sb->max_dev = cpu_to_le32(max_dev);
2103 rdev->sb_size = max_dev * 2 + 256;
2104 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2105 if (rdev->sb_size & bmask)
2106 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2108 max_dev = le32_to_cpu(sb->max_dev);
2110 for (i=0; i<max_dev;i++)
2111 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2113 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2114 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2116 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2117 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2119 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2121 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2122 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2123 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2126 rdev_for_each(rdev2, mddev) {
2128 if (test_bit(Faulty, &rdev2->flags))
2129 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2130 else if (test_bit(In_sync, &rdev2->flags))
2131 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2132 else if (test_bit(Journal, &rdev2->flags))
2133 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2134 else if (rdev2->raid_disk >= 0)
2135 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2137 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2140 sb->sb_csum = calc_sb_1_csum(sb);
2143 static sector_t super_1_choose_bm_space(sector_t dev_size)
2147 /* if the device is bigger than 8Gig, save 64k for bitmap
2148 * usage, if bigger than 200Gig, save 128k
2150 if (dev_size < 64*2)
2152 else if (dev_size - 64*2 >= 200*1024*1024*2)
2154 else if (dev_size - 4*2 > 8*1024*1024*2)
2161 static unsigned long long
2162 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2164 struct mdp_superblock_1 *sb;
2165 sector_t max_sectors;
2166 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2167 return 0; /* component must fit device */
2168 if (rdev->data_offset != rdev->new_data_offset)
2169 return 0; /* too confusing */
2170 if (rdev->sb_start < rdev->data_offset) {
2171 /* minor versions 1 and 2; superblock before data */
2172 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2173 max_sectors -= rdev->data_offset;
2174 if (!num_sectors || num_sectors > max_sectors)
2175 num_sectors = max_sectors;
2176 } else if (rdev->mddev->bitmap_info.offset) {
2177 /* minor version 0 with bitmap we can't move */
2180 /* minor version 0; superblock after data */
2181 sector_t sb_start, bm_space;
2182 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2184 /* 8K is for superblock */
2185 sb_start = dev_size - 8*2;
2186 sb_start &= ~(sector_t)(4*2 - 1);
2188 bm_space = super_1_choose_bm_space(dev_size);
2190 /* Space that can be used to store date needs to decrease
2191 * superblock bitmap space and bad block space(4K)
2193 max_sectors = sb_start - bm_space - 4*2;
2195 if (!num_sectors || num_sectors > max_sectors)
2196 num_sectors = max_sectors;
2197 rdev->sb_start = sb_start;
2199 sb = page_address(rdev->sb_page);
2200 sb->data_size = cpu_to_le64(num_sectors);
2201 sb->super_offset = cpu_to_le64(rdev->sb_start);
2202 sb->sb_csum = calc_sb_1_csum(sb);
2204 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2206 } while (md_super_wait(rdev->mddev) < 0);
2212 super_1_allow_new_offset(struct md_rdev *rdev,
2213 unsigned long long new_offset)
2215 /* All necessary checks on new >= old have been done */
2216 struct bitmap *bitmap;
2217 if (new_offset >= rdev->data_offset)
2220 /* with 1.0 metadata, there is no metadata to tread on
2221 * so we can always move back */
2222 if (rdev->mddev->minor_version == 0)
2225 /* otherwise we must be sure not to step on
2226 * any metadata, so stay:
2227 * 36K beyond start of superblock
2228 * beyond end of badblocks
2229 * beyond write-intent bitmap
2231 if (rdev->sb_start + (32+4)*2 > new_offset)
2233 bitmap = rdev->mddev->bitmap;
2234 if (bitmap && !rdev->mddev->bitmap_info.file &&
2235 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2236 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2238 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2244 static struct super_type super_types[] = {
2247 .owner = THIS_MODULE,
2248 .load_super = super_90_load,
2249 .validate_super = super_90_validate,
2250 .sync_super = super_90_sync,
2251 .rdev_size_change = super_90_rdev_size_change,
2252 .allow_new_offset = super_90_allow_new_offset,
2256 .owner = THIS_MODULE,
2257 .load_super = super_1_load,
2258 .validate_super = super_1_validate,
2259 .sync_super = super_1_sync,
2260 .rdev_size_change = super_1_rdev_size_change,
2261 .allow_new_offset = super_1_allow_new_offset,
2265 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2267 if (mddev->sync_super) {
2268 mddev->sync_super(mddev, rdev);
2272 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2274 super_types[mddev->major_version].sync_super(mddev, rdev);
2277 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2279 struct md_rdev *rdev, *rdev2;
2282 rdev_for_each_rcu(rdev, mddev1) {
2283 if (test_bit(Faulty, &rdev->flags) ||
2284 test_bit(Journal, &rdev->flags) ||
2285 rdev->raid_disk == -1)
2287 rdev_for_each_rcu(rdev2, mddev2) {
2288 if (test_bit(Faulty, &rdev2->flags) ||
2289 test_bit(Journal, &rdev2->flags) ||
2290 rdev2->raid_disk == -1)
2292 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2302 static LIST_HEAD(pending_raid_disks);
2305 * Try to register data integrity profile for an mddev
2307 * This is called when an array is started and after a disk has been kicked
2308 * from the array. It only succeeds if all working and active component devices
2309 * are integrity capable with matching profiles.
2311 int md_integrity_register(struct mddev *mddev)
2313 struct md_rdev *rdev, *reference = NULL;
2315 if (list_empty(&mddev->disks))
2316 return 0; /* nothing to do */
2317 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2318 return 0; /* shouldn't register, or already is */
2319 rdev_for_each(rdev, mddev) {
2320 /* skip spares and non-functional disks */
2321 if (test_bit(Faulty, &rdev->flags))
2323 if (rdev->raid_disk < 0)
2326 /* Use the first rdev as the reference */
2330 /* does this rdev's profile match the reference profile? */
2331 if (blk_integrity_compare(reference->bdev->bd_disk,
2332 rdev->bdev->bd_disk) < 0)
2335 if (!reference || !bdev_get_integrity(reference->bdev))
2338 * All component devices are integrity capable and have matching
2339 * profiles, register the common profile for the md device.
2341 blk_integrity_register(mddev->gendisk,
2342 bdev_get_integrity(reference->bdev));
2344 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2345 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2346 (mddev->level != 1 && mddev->level != 10 &&
2347 bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2349 * No need to handle the failure of bioset_integrity_create,
2350 * because the function is called by md_run() -> pers->run(),
2351 * md_run calls bioset_exit -> bioset_integrity_free in case
2354 pr_err("md: failed to create integrity pool for %s\n",
2360 EXPORT_SYMBOL(md_integrity_register);
2363 * Attempt to add an rdev, but only if it is consistent with the current
2366 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2368 struct blk_integrity *bi_mddev;
2369 char name[BDEVNAME_SIZE];
2371 if (!mddev->gendisk)
2374 bi_mddev = blk_get_integrity(mddev->gendisk);
2376 if (!bi_mddev) /* nothing to do */
2379 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2380 pr_err("%s: incompatible integrity profile for %s\n",
2381 mdname(mddev), bdevname(rdev->bdev, name));
2387 EXPORT_SYMBOL(md_integrity_add_rdev);
2389 static bool rdev_read_only(struct md_rdev *rdev)
2391 return bdev_read_only(rdev->bdev) ||
2392 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2395 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2397 char b[BDEVNAME_SIZE];
2400 /* prevent duplicates */
2401 if (find_rdev(mddev, rdev->bdev->bd_dev))
2404 if (rdev_read_only(rdev) && mddev->pers)
2407 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2408 if (!test_bit(Journal, &rdev->flags) &&
2410 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2412 /* Cannot change size, so fail
2413 * If mddev->level <= 0, then we don't care
2414 * about aligning sizes (e.g. linear)
2416 if (mddev->level > 0)
2419 mddev->dev_sectors = rdev->sectors;
2422 /* Verify rdev->desc_nr is unique.
2423 * If it is -1, assign a free number, else
2424 * check number is not in use
2427 if (rdev->desc_nr < 0) {
2430 choice = mddev->raid_disks;
2431 while (md_find_rdev_nr_rcu(mddev, choice))
2433 rdev->desc_nr = choice;
2435 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2441 if (!test_bit(Journal, &rdev->flags) &&
2442 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2443 pr_warn("md: %s: array is limited to %d devices\n",
2444 mdname(mddev), mddev->max_disks);
2447 bdevname(rdev->bdev,b);
2448 strreplace(b, '/', '!');
2450 rdev->mddev = mddev;
2451 pr_debug("md: bind<%s>\n", b);
2453 if (mddev->raid_disks)
2454 mddev_create_serial_pool(mddev, rdev, false);
2456 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2459 /* failure here is OK */
2460 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2461 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2462 rdev->sysfs_unack_badblocks =
2463 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2464 rdev->sysfs_badblocks =
2465 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2467 list_add_rcu(&rdev->same_set, &mddev->disks);
2468 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2470 /* May as well allow recovery to be retried once */
2471 mddev->recovery_disabled++;
2476 pr_warn("md: failed to register dev-%s for %s\n",
2481 static void rdev_delayed_delete(struct work_struct *ws)
2483 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2484 kobject_del(&rdev->kobj);
2485 kobject_put(&rdev->kobj);
2488 static void unbind_rdev_from_array(struct md_rdev *rdev)
2490 char b[BDEVNAME_SIZE];
2492 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2493 list_del_rcu(&rdev->same_set);
2494 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2495 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2497 sysfs_remove_link(&rdev->kobj, "block");
2498 sysfs_put(rdev->sysfs_state);
2499 sysfs_put(rdev->sysfs_unack_badblocks);
2500 sysfs_put(rdev->sysfs_badblocks);
2501 rdev->sysfs_state = NULL;
2502 rdev->sysfs_unack_badblocks = NULL;
2503 rdev->sysfs_badblocks = NULL;
2504 rdev->badblocks.count = 0;
2505 /* We need to delay this, otherwise we can deadlock when
2506 * writing to 'remove' to "dev/state". We also need
2507 * to delay it due to rcu usage.
2510 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2511 kobject_get(&rdev->kobj);
2512 queue_work(md_rdev_misc_wq, &rdev->del_work);
2516 * prevent the device from being mounted, repartitioned or
2517 * otherwise reused by a RAID array (or any other kernel
2518 * subsystem), by bd_claiming the device.
2520 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2523 struct block_device *bdev;
2525 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2526 shared ? (struct md_rdev *)lock_rdev : rdev);
2528 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2529 MAJOR(dev), MINOR(dev));
2530 return PTR_ERR(bdev);
2536 static void unlock_rdev(struct md_rdev *rdev)
2538 struct block_device *bdev = rdev->bdev;
2540 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2543 void md_autodetect_dev(dev_t dev);
2545 static void export_rdev(struct md_rdev *rdev)
2547 char b[BDEVNAME_SIZE];
2549 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2550 md_rdev_clear(rdev);
2552 if (test_bit(AutoDetected, &rdev->flags))
2553 md_autodetect_dev(rdev->bdev->bd_dev);
2556 kobject_put(&rdev->kobj);
2559 void md_kick_rdev_from_array(struct md_rdev *rdev)
2561 unbind_rdev_from_array(rdev);
2564 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2566 static void export_array(struct mddev *mddev)
2568 struct md_rdev *rdev;
2570 while (!list_empty(&mddev->disks)) {
2571 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2573 md_kick_rdev_from_array(rdev);
2575 mddev->raid_disks = 0;
2576 mddev->major_version = 0;
2579 static bool set_in_sync(struct mddev *mddev)
2581 lockdep_assert_held(&mddev->lock);
2582 if (!mddev->in_sync) {
2583 mddev->sync_checkers++;
2584 spin_unlock(&mddev->lock);
2585 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2586 spin_lock(&mddev->lock);
2587 if (!mddev->in_sync &&
2588 percpu_ref_is_zero(&mddev->writes_pending)) {
2591 * Ensure ->in_sync is visible before we clear
2595 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2596 sysfs_notify_dirent_safe(mddev->sysfs_state);
2598 if (--mddev->sync_checkers == 0)
2599 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2601 if (mddev->safemode == 1)
2602 mddev->safemode = 0;
2603 return mddev->in_sync;
2606 static void sync_sbs(struct mddev *mddev, int nospares)
2608 /* Update each superblock (in-memory image), but
2609 * if we are allowed to, skip spares which already
2610 * have the right event counter, or have one earlier
2611 * (which would mean they aren't being marked as dirty
2612 * with the rest of the array)
2614 struct md_rdev *rdev;
2615 rdev_for_each(rdev, mddev) {
2616 if (rdev->sb_events == mddev->events ||
2618 rdev->raid_disk < 0 &&
2619 rdev->sb_events+1 == mddev->events)) {
2620 /* Don't update this superblock */
2621 rdev->sb_loaded = 2;
2623 sync_super(mddev, rdev);
2624 rdev->sb_loaded = 1;
2629 static bool does_sb_need_changing(struct mddev *mddev)
2631 struct md_rdev *rdev = NULL, *iter;
2632 struct mdp_superblock_1 *sb;
2635 /* Find a good rdev */
2636 rdev_for_each(iter, mddev)
2637 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2642 /* No good device found. */
2646 sb = page_address(rdev->sb_page);
2647 /* Check if a device has become faulty or a spare become active */
2648 rdev_for_each(rdev, mddev) {
2649 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2650 /* Device activated? */
2651 if (role == 0xffff && rdev->raid_disk >=0 &&
2652 !test_bit(Faulty, &rdev->flags))
2654 /* Device turned faulty? */
2655 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2659 /* Check if any mddev parameters have changed */
2660 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2661 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2662 (mddev->layout != le32_to_cpu(sb->layout)) ||
2663 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2664 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2670 void md_update_sb(struct mddev *mddev, int force_change)
2672 struct md_rdev *rdev;
2675 int any_badblocks_changed = 0;
2680 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2685 if (mddev_is_clustered(mddev)) {
2686 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2688 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2690 ret = md_cluster_ops->metadata_update_start(mddev);
2691 /* Has someone else has updated the sb */
2692 if (!does_sb_need_changing(mddev)) {
2694 md_cluster_ops->metadata_update_cancel(mddev);
2695 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2696 BIT(MD_SB_CHANGE_DEVS) |
2697 BIT(MD_SB_CHANGE_CLEAN));
2703 * First make sure individual recovery_offsets are correct
2704 * curr_resync_completed can only be used during recovery.
2705 * During reshape/resync it might use array-addresses rather
2706 * that device addresses.
2708 rdev_for_each(rdev, mddev) {
2709 if (rdev->raid_disk >= 0 &&
2710 mddev->delta_disks >= 0 &&
2711 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2712 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2713 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2714 !test_bit(Journal, &rdev->flags) &&
2715 !test_bit(In_sync, &rdev->flags) &&
2716 mddev->curr_resync_completed > rdev->recovery_offset)
2717 rdev->recovery_offset = mddev->curr_resync_completed;
2720 if (!mddev->persistent) {
2721 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2722 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2723 if (!mddev->external) {
2724 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2725 rdev_for_each(rdev, mddev) {
2726 if (rdev->badblocks.changed) {
2727 rdev->badblocks.changed = 0;
2728 ack_all_badblocks(&rdev->badblocks);
2729 md_error(mddev, rdev);
2731 clear_bit(Blocked, &rdev->flags);
2732 clear_bit(BlockedBadBlocks, &rdev->flags);
2733 wake_up(&rdev->blocked_wait);
2736 wake_up(&mddev->sb_wait);
2740 spin_lock(&mddev->lock);
2742 mddev->utime = ktime_get_real_seconds();
2744 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2746 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2747 /* just a clean<-> dirty transition, possibly leave spares alone,
2748 * though if events isn't the right even/odd, we will have to do
2754 if (mddev->degraded)
2755 /* If the array is degraded, then skipping spares is both
2756 * dangerous and fairly pointless.
2757 * Dangerous because a device that was removed from the array
2758 * might have a event_count that still looks up-to-date,
2759 * so it can be re-added without a resync.
2760 * Pointless because if there are any spares to skip,
2761 * then a recovery will happen and soon that array won't
2762 * be degraded any more and the spare can go back to sleep then.
2766 sync_req = mddev->in_sync;
2768 /* If this is just a dirty<->clean transition, and the array is clean
2769 * and 'events' is odd, we can roll back to the previous clean state */
2771 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2772 && mddev->can_decrease_events
2773 && mddev->events != 1) {
2775 mddev->can_decrease_events = 0;
2777 /* otherwise we have to go forward and ... */
2779 mddev->can_decrease_events = nospares;
2783 * This 64-bit counter should never wrap.
2784 * Either we are in around ~1 trillion A.C., assuming
2785 * 1 reboot per second, or we have a bug...
2787 WARN_ON(mddev->events == 0);
2789 rdev_for_each(rdev, mddev) {
2790 if (rdev->badblocks.changed)
2791 any_badblocks_changed++;
2792 if (test_bit(Faulty, &rdev->flags))
2793 set_bit(FaultRecorded, &rdev->flags);
2796 sync_sbs(mddev, nospares);
2797 spin_unlock(&mddev->lock);
2799 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2800 mdname(mddev), mddev->in_sync);
2803 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2805 md_bitmap_update_sb(mddev->bitmap);
2806 rdev_for_each(rdev, mddev) {
2807 char b[BDEVNAME_SIZE];
2809 if (rdev->sb_loaded != 1)
2810 continue; /* no noise on spare devices */
2812 if (!test_bit(Faulty, &rdev->flags)) {
2813 md_super_write(mddev,rdev,
2814 rdev->sb_start, rdev->sb_size,
2816 pr_debug("md: (write) %s's sb offset: %llu\n",
2817 bdevname(rdev->bdev, b),
2818 (unsigned long long)rdev->sb_start);
2819 rdev->sb_events = mddev->events;
2820 if (rdev->badblocks.size) {
2821 md_super_write(mddev, rdev,
2822 rdev->badblocks.sector,
2823 rdev->badblocks.size << 9,
2825 rdev->badblocks.size = 0;
2829 pr_debug("md: %s (skipping faulty)\n",
2830 bdevname(rdev->bdev, b));
2832 if (mddev->level == LEVEL_MULTIPATH)
2833 /* only need to write one superblock... */
2836 if (md_super_wait(mddev) < 0)
2838 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2840 if (mddev_is_clustered(mddev) && ret == 0)
2841 md_cluster_ops->metadata_update_finish(mddev);
2843 if (mddev->in_sync != sync_req ||
2844 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2845 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2846 /* have to write it out again */
2848 wake_up(&mddev->sb_wait);
2849 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2850 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2852 rdev_for_each(rdev, mddev) {
2853 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2854 clear_bit(Blocked, &rdev->flags);
2856 if (any_badblocks_changed)
2857 ack_all_badblocks(&rdev->badblocks);
2858 clear_bit(BlockedBadBlocks, &rdev->flags);
2859 wake_up(&rdev->blocked_wait);
2862 EXPORT_SYMBOL(md_update_sb);
2864 static int add_bound_rdev(struct md_rdev *rdev)
2866 struct mddev *mddev = rdev->mddev;
2868 bool add_journal = test_bit(Journal, &rdev->flags);
2870 if (!mddev->pers->hot_remove_disk || add_journal) {
2871 /* If there is hot_add_disk but no hot_remove_disk
2872 * then added disks for geometry changes,
2873 * and should be added immediately.
2875 super_types[mddev->major_version].
2876 validate_super(mddev, rdev);
2878 mddev_suspend(mddev);
2879 err = mddev->pers->hot_add_disk(mddev, rdev);
2881 mddev_resume(mddev);
2883 md_kick_rdev_from_array(rdev);
2887 sysfs_notify_dirent_safe(rdev->sysfs_state);
2889 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2890 if (mddev->degraded)
2891 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2892 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2893 md_new_event(mddev);
2894 md_wakeup_thread(mddev->thread);
2898 /* words written to sysfs files may, or may not, be \n terminated.
2899 * We want to accept with case. For this we use cmd_match.
2901 static int cmd_match(const char *cmd, const char *str)
2903 /* See if cmd, written into a sysfs file, matches
2904 * str. They must either be the same, or cmd can
2905 * have a trailing newline
2907 while (*cmd && *str && *cmd == *str) {
2918 struct rdev_sysfs_entry {
2919 struct attribute attr;
2920 ssize_t (*show)(struct md_rdev *, char *);
2921 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2925 state_show(struct md_rdev *rdev, char *page)
2929 unsigned long flags = READ_ONCE(rdev->flags);
2931 if (test_bit(Faulty, &flags) ||
2932 (!test_bit(ExternalBbl, &flags) &&
2933 rdev->badblocks.unacked_exist))
2934 len += sprintf(page+len, "faulty%s", sep);
2935 if (test_bit(In_sync, &flags))
2936 len += sprintf(page+len, "in_sync%s", sep);
2937 if (test_bit(Journal, &flags))
2938 len += sprintf(page+len, "journal%s", sep);
2939 if (test_bit(WriteMostly, &flags))
2940 len += sprintf(page+len, "write_mostly%s", sep);
2941 if (test_bit(Blocked, &flags) ||
2942 (rdev->badblocks.unacked_exist
2943 && !test_bit(Faulty, &flags)))
2944 len += sprintf(page+len, "blocked%s", sep);
2945 if (!test_bit(Faulty, &flags) &&
2946 !test_bit(Journal, &flags) &&
2947 !test_bit(In_sync, &flags))
2948 len += sprintf(page+len, "spare%s", sep);
2949 if (test_bit(WriteErrorSeen, &flags))
2950 len += sprintf(page+len, "write_error%s", sep);
2951 if (test_bit(WantReplacement, &flags))
2952 len += sprintf(page+len, "want_replacement%s", sep);
2953 if (test_bit(Replacement, &flags))
2954 len += sprintf(page+len, "replacement%s", sep);
2955 if (test_bit(ExternalBbl, &flags))
2956 len += sprintf(page+len, "external_bbl%s", sep);
2957 if (test_bit(FailFast, &flags))
2958 len += sprintf(page+len, "failfast%s", sep);
2963 return len+sprintf(page+len, "\n");
2967 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2970 * faulty - simulates an error
2971 * remove - disconnects the device
2972 * writemostly - sets write_mostly
2973 * -writemostly - clears write_mostly
2974 * blocked - sets the Blocked flags
2975 * -blocked - clears the Blocked and possibly simulates an error
2976 * insync - sets Insync providing device isn't active
2977 * -insync - clear Insync for a device with a slot assigned,
2978 * so that it gets rebuilt based on bitmap
2979 * write_error - sets WriteErrorSeen
2980 * -write_error - clears WriteErrorSeen
2981 * {,-}failfast - set/clear FailFast
2984 struct mddev *mddev = rdev->mddev;
2986 bool need_update_sb = false;
2988 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2989 md_error(rdev->mddev, rdev);
2990 if (test_bit(Faulty, &rdev->flags))
2994 } else if (cmd_match(buf, "remove")) {
2995 if (rdev->mddev->pers) {
2996 clear_bit(Blocked, &rdev->flags);
2997 remove_and_add_spares(rdev->mddev, rdev);
2999 if (rdev->raid_disk >= 0)
3003 if (mddev_is_clustered(mddev))
3004 err = md_cluster_ops->remove_disk(mddev, rdev);
3007 md_kick_rdev_from_array(rdev);
3009 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3010 md_wakeup_thread(mddev->thread);
3012 md_new_event(mddev);
3015 } else if (cmd_match(buf, "writemostly")) {
3016 set_bit(WriteMostly, &rdev->flags);
3017 mddev_create_serial_pool(rdev->mddev, rdev, false);
3018 need_update_sb = true;
3020 } else if (cmd_match(buf, "-writemostly")) {
3021 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3022 clear_bit(WriteMostly, &rdev->flags);
3023 need_update_sb = true;
3025 } else if (cmd_match(buf, "blocked")) {
3026 set_bit(Blocked, &rdev->flags);
3028 } else if (cmd_match(buf, "-blocked")) {
3029 if (!test_bit(Faulty, &rdev->flags) &&
3030 !test_bit(ExternalBbl, &rdev->flags) &&
3031 rdev->badblocks.unacked_exist) {
3032 /* metadata handler doesn't understand badblocks,
3033 * so we need to fail the device
3035 md_error(rdev->mddev, rdev);
3037 clear_bit(Blocked, &rdev->flags);
3038 clear_bit(BlockedBadBlocks, &rdev->flags);
3039 wake_up(&rdev->blocked_wait);
3040 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3041 md_wakeup_thread(rdev->mddev->thread);
3044 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3045 set_bit(In_sync, &rdev->flags);
3047 } else if (cmd_match(buf, "failfast")) {
3048 set_bit(FailFast, &rdev->flags);
3049 need_update_sb = true;
3051 } else if (cmd_match(buf, "-failfast")) {
3052 clear_bit(FailFast, &rdev->flags);
3053 need_update_sb = true;
3055 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3056 !test_bit(Journal, &rdev->flags)) {
3057 if (rdev->mddev->pers == NULL) {
3058 clear_bit(In_sync, &rdev->flags);
3059 rdev->saved_raid_disk = rdev->raid_disk;
3060 rdev->raid_disk = -1;
3063 } else if (cmd_match(buf, "write_error")) {
3064 set_bit(WriteErrorSeen, &rdev->flags);
3066 } else if (cmd_match(buf, "-write_error")) {
3067 clear_bit(WriteErrorSeen, &rdev->flags);
3069 } else if (cmd_match(buf, "want_replacement")) {
3070 /* Any non-spare device that is not a replacement can
3071 * become want_replacement at any time, but we then need to
3072 * check if recovery is needed.
3074 if (rdev->raid_disk >= 0 &&
3075 !test_bit(Journal, &rdev->flags) &&
3076 !test_bit(Replacement, &rdev->flags))
3077 set_bit(WantReplacement, &rdev->flags);
3078 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3079 md_wakeup_thread(rdev->mddev->thread);
3081 } else if (cmd_match(buf, "-want_replacement")) {
3082 /* Clearing 'want_replacement' is always allowed.
3083 * Once replacements starts it is too late though.
3086 clear_bit(WantReplacement, &rdev->flags);
3087 } else if (cmd_match(buf, "replacement")) {
3088 /* Can only set a device as a replacement when array has not
3089 * yet been started. Once running, replacement is automatic
3090 * from spares, or by assigning 'slot'.
3092 if (rdev->mddev->pers)
3095 set_bit(Replacement, &rdev->flags);
3098 } else if (cmd_match(buf, "-replacement")) {
3099 /* Similarly, can only clear Replacement before start */
3100 if (rdev->mddev->pers)
3103 clear_bit(Replacement, &rdev->flags);
3106 } else if (cmd_match(buf, "re-add")) {
3107 if (!rdev->mddev->pers)
3109 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3110 rdev->saved_raid_disk >= 0) {
3111 /* clear_bit is performed _after_ all the devices
3112 * have their local Faulty bit cleared. If any writes
3113 * happen in the meantime in the local node, they
3114 * will land in the local bitmap, which will be synced
3115 * by this node eventually
3117 if (!mddev_is_clustered(rdev->mddev) ||
3118 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3119 clear_bit(Faulty, &rdev->flags);
3120 err = add_bound_rdev(rdev);
3124 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3125 set_bit(ExternalBbl, &rdev->flags);
3126 rdev->badblocks.shift = 0;
3128 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3129 clear_bit(ExternalBbl, &rdev->flags);
3133 md_update_sb(mddev, 1);
3135 sysfs_notify_dirent_safe(rdev->sysfs_state);
3136 return err ? err : len;
3138 static struct rdev_sysfs_entry rdev_state =
3139 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3142 errors_show(struct md_rdev *rdev, char *page)
3144 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3148 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3153 rv = kstrtouint(buf, 10, &n);
3156 atomic_set(&rdev->corrected_errors, n);
3159 static struct rdev_sysfs_entry rdev_errors =
3160 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3163 slot_show(struct md_rdev *rdev, char *page)
3165 if (test_bit(Journal, &rdev->flags))
3166 return sprintf(page, "journal\n");
3167 else if (rdev->raid_disk < 0)
3168 return sprintf(page, "none\n");
3170 return sprintf(page, "%d\n", rdev->raid_disk);
3174 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3179 if (test_bit(Journal, &rdev->flags))
3181 if (strncmp(buf, "none", 4)==0)
3184 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3188 if (rdev->mddev->pers && slot == -1) {
3189 /* Setting 'slot' on an active array requires also
3190 * updating the 'rd%d' link, and communicating
3191 * with the personality with ->hot_*_disk.
3192 * For now we only support removing
3193 * failed/spare devices. This normally happens automatically,
3194 * but not when the metadata is externally managed.
3196 if (rdev->raid_disk == -1)
3198 /* personality does all needed checks */
3199 if (rdev->mddev->pers->hot_remove_disk == NULL)
3201 clear_bit(Blocked, &rdev->flags);
3202 remove_and_add_spares(rdev->mddev, rdev);
3203 if (rdev->raid_disk >= 0)
3205 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3206 md_wakeup_thread(rdev->mddev->thread);
3207 } else if (rdev->mddev->pers) {
3208 /* Activating a spare .. or possibly reactivating
3209 * if we ever get bitmaps working here.
3213 if (rdev->raid_disk != -1)
3216 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3219 if (rdev->mddev->pers->hot_add_disk == NULL)
3222 if (slot >= rdev->mddev->raid_disks &&
3223 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3226 rdev->raid_disk = slot;
3227 if (test_bit(In_sync, &rdev->flags))
3228 rdev->saved_raid_disk = slot;
3230 rdev->saved_raid_disk = -1;
3231 clear_bit(In_sync, &rdev->flags);
3232 clear_bit(Bitmap_sync, &rdev->flags);
3233 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3235 rdev->raid_disk = -1;
3238 sysfs_notify_dirent_safe(rdev->sysfs_state);
3239 /* failure here is OK */;
3240 sysfs_link_rdev(rdev->mddev, rdev);
3241 /* don't wakeup anyone, leave that to userspace. */
3243 if (slot >= rdev->mddev->raid_disks &&
3244 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3246 rdev->raid_disk = slot;
3247 /* assume it is working */
3248 clear_bit(Faulty, &rdev->flags);
3249 clear_bit(WriteMostly, &rdev->flags);
3250 set_bit(In_sync, &rdev->flags);
3251 sysfs_notify_dirent_safe(rdev->sysfs_state);
3256 static struct rdev_sysfs_entry rdev_slot =
3257 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3260 offset_show(struct md_rdev *rdev, char *page)
3262 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3266 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3268 unsigned long long offset;
3269 if (kstrtoull(buf, 10, &offset) < 0)
3271 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3273 if (rdev->sectors && rdev->mddev->external)
3274 /* Must set offset before size, so overlap checks
3277 rdev->data_offset = offset;
3278 rdev->new_data_offset = offset;
3282 static struct rdev_sysfs_entry rdev_offset =
3283 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3285 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3287 return sprintf(page, "%llu\n",
3288 (unsigned long long)rdev->new_data_offset);
3291 static ssize_t new_offset_store(struct md_rdev *rdev,
3292 const char *buf, size_t len)
3294 unsigned long long new_offset;
3295 struct mddev *mddev = rdev->mddev;
3297 if (kstrtoull(buf, 10, &new_offset) < 0)
3300 if (mddev->sync_thread ||
3301 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3303 if (new_offset == rdev->data_offset)
3304 /* reset is always permitted */
3306 else if (new_offset > rdev->data_offset) {
3307 /* must not push array size beyond rdev_sectors */
3308 if (new_offset - rdev->data_offset
3309 + mddev->dev_sectors > rdev->sectors)
3312 /* Metadata worries about other space details. */
3314 /* decreasing the offset is inconsistent with a backwards
3317 if (new_offset < rdev->data_offset &&
3318 mddev->reshape_backwards)
3320 /* Increasing offset is inconsistent with forwards
3321 * reshape. reshape_direction should be set to
3322 * 'backwards' first.
3324 if (new_offset > rdev->data_offset &&
3325 !mddev->reshape_backwards)
3328 if (mddev->pers && mddev->persistent &&
3329 !super_types[mddev->major_version]
3330 .allow_new_offset(rdev, new_offset))
3332 rdev->new_data_offset = new_offset;
3333 if (new_offset > rdev->data_offset)
3334 mddev->reshape_backwards = 1;
3335 else if (new_offset < rdev->data_offset)
3336 mddev->reshape_backwards = 0;
3340 static struct rdev_sysfs_entry rdev_new_offset =
3341 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3344 rdev_size_show(struct md_rdev *rdev, char *page)
3346 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3349 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3351 /* check if two start/length pairs overlap */
3359 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3361 unsigned long long blocks;
3364 if (kstrtoull(buf, 10, &blocks) < 0)
3367 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3368 return -EINVAL; /* sector conversion overflow */
3371 if (new != blocks * 2)
3372 return -EINVAL; /* unsigned long long to sector_t overflow */
3379 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3381 struct mddev *my_mddev = rdev->mddev;
3382 sector_t oldsectors = rdev->sectors;
3385 if (test_bit(Journal, &rdev->flags))
3387 if (strict_blocks_to_sectors(buf, §ors) < 0)
3389 if (rdev->data_offset != rdev->new_data_offset)
3390 return -EINVAL; /* too confusing */
3391 if (my_mddev->pers && rdev->raid_disk >= 0) {
3392 if (my_mddev->persistent) {
3393 sectors = super_types[my_mddev->major_version].
3394 rdev_size_change(rdev, sectors);
3397 } else if (!sectors)
3398 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3400 if (!my_mddev->pers->resize)
3401 /* Cannot change size for RAID0 or Linear etc */
3404 if (sectors < my_mddev->dev_sectors)
3405 return -EINVAL; /* component must fit device */
3407 rdev->sectors = sectors;
3408 if (sectors > oldsectors && my_mddev->external) {
3409 /* Need to check that all other rdevs with the same
3410 * ->bdev do not overlap. 'rcu' is sufficient to walk
3411 * the rdev lists safely.
3412 * This check does not provide a hard guarantee, it
3413 * just helps avoid dangerous mistakes.
3415 struct mddev *mddev;
3417 struct list_head *tmp;
3420 for_each_mddev(mddev, tmp) {
3421 struct md_rdev *rdev2;
3423 rdev_for_each(rdev2, mddev)
3424 if (rdev->bdev == rdev2->bdev &&
3426 overlaps(rdev->data_offset, rdev->sectors,
3439 /* Someone else could have slipped in a size
3440 * change here, but doing so is just silly.
3441 * We put oldsectors back because we *know* it is
3442 * safe, and trust userspace not to race with
3445 rdev->sectors = oldsectors;
3452 static struct rdev_sysfs_entry rdev_size =
3453 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3455 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3457 unsigned long long recovery_start = rdev->recovery_offset;
3459 if (test_bit(In_sync, &rdev->flags) ||
3460 recovery_start == MaxSector)
3461 return sprintf(page, "none\n");
3463 return sprintf(page, "%llu\n", recovery_start);
3466 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3468 unsigned long long recovery_start;
3470 if (cmd_match(buf, "none"))
3471 recovery_start = MaxSector;
3472 else if (kstrtoull(buf, 10, &recovery_start))
3475 if (rdev->mddev->pers &&
3476 rdev->raid_disk >= 0)
3479 rdev->recovery_offset = recovery_start;
3480 if (recovery_start == MaxSector)
3481 set_bit(In_sync, &rdev->flags);
3483 clear_bit(In_sync, &rdev->flags);
3487 static struct rdev_sysfs_entry rdev_recovery_start =
3488 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3490 /* sysfs access to bad-blocks list.
3491 * We present two files.
3492 * 'bad-blocks' lists sector numbers and lengths of ranges that
3493 * are recorded as bad. The list is truncated to fit within
3494 * the one-page limit of sysfs.
3495 * Writing "sector length" to this file adds an acknowledged
3497 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3498 * been acknowledged. Writing to this file adds bad blocks
3499 * without acknowledging them. This is largely for testing.
3501 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3503 return badblocks_show(&rdev->badblocks, page, 0);
3505 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3507 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3508 /* Maybe that ack was all we needed */
3509 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3510 wake_up(&rdev->blocked_wait);
3513 static struct rdev_sysfs_entry rdev_bad_blocks =
3514 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3516 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3518 return badblocks_show(&rdev->badblocks, page, 1);
3520 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3522 return badblocks_store(&rdev->badblocks, page, len, 1);
3524 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3525 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3528 ppl_sector_show(struct md_rdev *rdev, char *page)
3530 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3534 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3536 unsigned long long sector;
3538 if (kstrtoull(buf, 10, §or) < 0)
3540 if (sector != (sector_t)sector)
3543 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3544 rdev->raid_disk >= 0)
3547 if (rdev->mddev->persistent) {
3548 if (rdev->mddev->major_version == 0)
3550 if ((sector > rdev->sb_start &&
3551 sector - rdev->sb_start > S16_MAX) ||
3552 (sector < rdev->sb_start &&
3553 rdev->sb_start - sector > -S16_MIN))
3555 rdev->ppl.offset = sector - rdev->sb_start;
3556 } else if (!rdev->mddev->external) {
3559 rdev->ppl.sector = sector;
3563 static struct rdev_sysfs_entry rdev_ppl_sector =
3564 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3567 ppl_size_show(struct md_rdev *rdev, char *page)
3569 return sprintf(page, "%u\n", rdev->ppl.size);
3573 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3577 if (kstrtouint(buf, 10, &size) < 0)
3580 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3581 rdev->raid_disk >= 0)
3584 if (rdev->mddev->persistent) {
3585 if (rdev->mddev->major_version == 0)
3589 } else if (!rdev->mddev->external) {
3592 rdev->ppl.size = size;
3596 static struct rdev_sysfs_entry rdev_ppl_size =
3597 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3599 static struct attribute *rdev_default_attrs[] = {
3604 &rdev_new_offset.attr,
3606 &rdev_recovery_start.attr,
3607 &rdev_bad_blocks.attr,
3608 &rdev_unack_bad_blocks.attr,
3609 &rdev_ppl_sector.attr,
3610 &rdev_ppl_size.attr,
3614 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3616 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3617 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3623 return entry->show(rdev, page);
3627 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3628 const char *page, size_t length)
3630 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3631 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3633 struct mddev *mddev = rdev->mddev;
3637 if (!capable(CAP_SYS_ADMIN))
3639 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3641 if (rdev->mddev == NULL)
3644 rv = entry->store(rdev, page, length);
3645 mddev_unlock(mddev);
3650 static void rdev_free(struct kobject *ko)
3652 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3655 static const struct sysfs_ops rdev_sysfs_ops = {
3656 .show = rdev_attr_show,
3657 .store = rdev_attr_store,
3659 static struct kobj_type rdev_ktype = {
3660 .release = rdev_free,
3661 .sysfs_ops = &rdev_sysfs_ops,
3662 .default_attrs = rdev_default_attrs,
3665 int md_rdev_init(struct md_rdev *rdev)
3668 rdev->saved_raid_disk = -1;
3669 rdev->raid_disk = -1;
3671 rdev->data_offset = 0;
3672 rdev->new_data_offset = 0;
3673 rdev->sb_events = 0;
3674 rdev->last_read_error = 0;
3675 rdev->sb_loaded = 0;
3676 rdev->bb_page = NULL;
3677 atomic_set(&rdev->nr_pending, 0);
3678 atomic_set(&rdev->read_errors, 0);
3679 atomic_set(&rdev->corrected_errors, 0);
3681 INIT_LIST_HEAD(&rdev->same_set);
3682 init_waitqueue_head(&rdev->blocked_wait);
3684 /* Add space to store bad block list.
3685 * This reserves the space even on arrays where it cannot
3686 * be used - I wonder if that matters
3688 return badblocks_init(&rdev->badblocks, 0);
3690 EXPORT_SYMBOL_GPL(md_rdev_init);
3692 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3694 * mark the device faulty if:
3696 * - the device is nonexistent (zero size)
3697 * - the device has no valid superblock
3699 * a faulty rdev _never_ has rdev->sb set.
3701 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3703 char b[BDEVNAME_SIZE];
3705 struct md_rdev *rdev;
3708 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3710 return ERR_PTR(-ENOMEM);
3712 err = md_rdev_init(rdev);
3715 err = alloc_disk_sb(rdev);
3719 err = lock_rdev(rdev, newdev, super_format == -2);
3723 kobject_init(&rdev->kobj, &rdev_ktype);
3725 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3727 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3728 bdevname(rdev->bdev,b));
3733 if (super_format >= 0) {
3734 err = super_types[super_format].
3735 load_super(rdev, NULL, super_minor);
3736 if (err == -EINVAL) {
3737 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3738 bdevname(rdev->bdev,b),
3739 super_format, super_minor);
3743 pr_warn("md: could not read %s's sb, not importing!\n",
3744 bdevname(rdev->bdev,b));
3754 md_rdev_clear(rdev);
3756 return ERR_PTR(err);
3760 * Check a full RAID array for plausibility
3763 static int analyze_sbs(struct mddev *mddev)
3766 struct md_rdev *rdev, *freshest, *tmp;
3767 char b[BDEVNAME_SIZE];
3770 rdev_for_each_safe(rdev, tmp, mddev)
3771 switch (super_types[mddev->major_version].
3772 load_super(rdev, freshest, mddev->minor_version)) {
3779 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3780 bdevname(rdev->bdev,b));
3781 md_kick_rdev_from_array(rdev);
3784 /* Cannot find a valid fresh disk */
3786 pr_warn("md: cannot find a valid disk\n");
3790 super_types[mddev->major_version].
3791 validate_super(mddev, freshest);
3794 rdev_for_each_safe(rdev, tmp, mddev) {
3795 if (mddev->max_disks &&
3796 (rdev->desc_nr >= mddev->max_disks ||
3797 i > mddev->max_disks)) {
3798 pr_warn("md: %s: %s: only %d devices permitted\n",
3799 mdname(mddev), bdevname(rdev->bdev, b),
3801 md_kick_rdev_from_array(rdev);
3804 if (rdev != freshest) {
3805 if (super_types[mddev->major_version].
3806 validate_super(mddev, rdev)) {
3807 pr_warn("md: kicking non-fresh %s from array!\n",
3808 bdevname(rdev->bdev,b));
3809 md_kick_rdev_from_array(rdev);
3813 if (mddev->level == LEVEL_MULTIPATH) {
3814 rdev->desc_nr = i++;
3815 rdev->raid_disk = rdev->desc_nr;
3816 set_bit(In_sync, &rdev->flags);
3817 } else if (rdev->raid_disk >=
3818 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3819 !test_bit(Journal, &rdev->flags)) {
3820 rdev->raid_disk = -1;
3821 clear_bit(In_sync, &rdev->flags);
3828 /* Read a fixed-point number.
3829 * Numbers in sysfs attributes should be in "standard" units where
3830 * possible, so time should be in seconds.
3831 * However we internally use a a much smaller unit such as
3832 * milliseconds or jiffies.
3833 * This function takes a decimal number with a possible fractional
3834 * component, and produces an integer which is the result of
3835 * multiplying that number by 10^'scale'.
3836 * all without any floating-point arithmetic.
3838 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3840 unsigned long result = 0;
3842 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3845 else if (decimals < scale) {
3848 result = result * 10 + value;
3860 *res = result * int_pow(10, scale - decimals);
3865 safe_delay_show(struct mddev *mddev, char *page)
3867 int msec = (mddev->safemode_delay*1000)/HZ;
3868 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3871 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3875 if (mddev_is_clustered(mddev)) {
3876 pr_warn("md: Safemode is disabled for clustered mode\n");
3880 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3883 mddev->safemode_delay = 0;
3885 unsigned long old_delay = mddev->safemode_delay;
3886 unsigned long new_delay = (msec*HZ)/1000;
3890 mddev->safemode_delay = new_delay;
3891 if (new_delay < old_delay || old_delay == 0)
3892 mod_timer(&mddev->safemode_timer, jiffies+1);
3896 static struct md_sysfs_entry md_safe_delay =
3897 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3900 level_show(struct mddev *mddev, char *page)
3902 struct md_personality *p;
3904 spin_lock(&mddev->lock);
3907 ret = sprintf(page, "%s\n", p->name);
3908 else if (mddev->clevel[0])
3909 ret = sprintf(page, "%s\n", mddev->clevel);
3910 else if (mddev->level != LEVEL_NONE)
3911 ret = sprintf(page, "%d\n", mddev->level);
3914 spin_unlock(&mddev->lock);
3919 level_store(struct mddev *mddev, const char *buf, size_t len)
3924 struct md_personality *pers, *oldpers;
3926 void *priv, *oldpriv;
3927 struct md_rdev *rdev;
3929 if (slen == 0 || slen >= sizeof(clevel))
3932 rv = mddev_lock(mddev);
3936 if (mddev->pers == NULL) {
3937 strncpy(mddev->clevel, buf, slen);
3938 if (mddev->clevel[slen-1] == '\n')
3940 mddev->clevel[slen] = 0;
3941 mddev->level = LEVEL_NONE;
3949 /* request to change the personality. Need to ensure:
3950 * - array is not engaged in resync/recovery/reshape
3951 * - old personality can be suspended
3952 * - new personality will access other array.
3956 if (mddev->sync_thread ||
3957 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3958 mddev->reshape_position != MaxSector ||
3959 mddev->sysfs_active)
3963 if (!mddev->pers->quiesce) {
3964 pr_warn("md: %s: %s does not support online personality change\n",
3965 mdname(mddev), mddev->pers->name);
3969 /* Now find the new personality */
3970 strncpy(clevel, buf, slen);
3971 if (clevel[slen-1] == '\n')
3974 if (kstrtol(clevel, 10, &level))
3977 if (request_module("md-%s", clevel) != 0)
3978 request_module("md-level-%s", clevel);
3979 spin_lock(&pers_lock);
3980 pers = find_pers(level, clevel);
3981 if (!pers || !try_module_get(pers->owner)) {
3982 spin_unlock(&pers_lock);
3983 pr_warn("md: personality %s not loaded\n", clevel);
3987 spin_unlock(&pers_lock);
3989 if (pers == mddev->pers) {
3990 /* Nothing to do! */
3991 module_put(pers->owner);
3995 if (!pers->takeover) {
3996 module_put(pers->owner);
3997 pr_warn("md: %s: %s does not support personality takeover\n",
3998 mdname(mddev), clevel);
4003 rdev_for_each(rdev, mddev)
4004 rdev->new_raid_disk = rdev->raid_disk;
4006 /* ->takeover must set new_* and/or delta_disks
4007 * if it succeeds, and may set them when it fails.
4009 priv = pers->takeover(mddev);
4011 mddev->new_level = mddev->level;
4012 mddev->new_layout = mddev->layout;
4013 mddev->new_chunk_sectors = mddev->chunk_sectors;
4014 mddev->raid_disks -= mddev->delta_disks;
4015 mddev->delta_disks = 0;
4016 mddev->reshape_backwards = 0;
4017 module_put(pers->owner);
4018 pr_warn("md: %s: %s would not accept array\n",
4019 mdname(mddev), clevel);
4024 /* Looks like we have a winner */
4025 mddev_suspend(mddev);
4026 mddev_detach(mddev);
4028 spin_lock(&mddev->lock);
4029 oldpers = mddev->pers;
4030 oldpriv = mddev->private;
4032 mddev->private = priv;
4033 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4034 mddev->level = mddev->new_level;
4035 mddev->layout = mddev->new_layout;
4036 mddev->chunk_sectors = mddev->new_chunk_sectors;
4037 mddev->delta_disks = 0;
4038 mddev->reshape_backwards = 0;
4039 mddev->degraded = 0;
4040 spin_unlock(&mddev->lock);
4042 if (oldpers->sync_request == NULL &&
4044 /* We are converting from a no-redundancy array
4045 * to a redundancy array and metadata is managed
4046 * externally so we need to be sure that writes
4047 * won't block due to a need to transition
4049 * until external management is started.
4052 mddev->safemode_delay = 0;
4053 mddev->safemode = 0;
4056 oldpers->free(mddev, oldpriv);
4058 if (oldpers->sync_request == NULL &&
4059 pers->sync_request != NULL) {
4060 /* need to add the md_redundancy_group */
4061 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4062 pr_warn("md: cannot register extra attributes for %s\n",
4064 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4065 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4066 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4068 if (oldpers->sync_request != NULL &&
4069 pers->sync_request == NULL) {
4070 /* need to remove the md_redundancy_group */
4071 if (mddev->to_remove == NULL)
4072 mddev->to_remove = &md_redundancy_group;
4075 module_put(oldpers->owner);
4077 rdev_for_each(rdev, mddev) {
4078 if (rdev->raid_disk < 0)
4080 if (rdev->new_raid_disk >= mddev->raid_disks)
4081 rdev->new_raid_disk = -1;
4082 if (rdev->new_raid_disk == rdev->raid_disk)
4084 sysfs_unlink_rdev(mddev, rdev);
4086 rdev_for_each(rdev, mddev) {
4087 if (rdev->raid_disk < 0)
4089 if (rdev->new_raid_disk == rdev->raid_disk)
4091 rdev->raid_disk = rdev->new_raid_disk;
4092 if (rdev->raid_disk < 0)
4093 clear_bit(In_sync, &rdev->flags);
4095 if (sysfs_link_rdev(mddev, rdev))
4096 pr_warn("md: cannot register rd%d for %s after level change\n",
4097 rdev->raid_disk, mdname(mddev));
4101 if (pers->sync_request == NULL) {
4102 /* this is now an array without redundancy, so
4103 * it must always be in_sync
4106 del_timer_sync(&mddev->safemode_timer);
4108 blk_set_stacking_limits(&mddev->queue->limits);
4110 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4111 mddev_resume(mddev);
4113 md_update_sb(mddev, 1);
4114 sysfs_notify_dirent_safe(mddev->sysfs_level);
4115 md_new_event(mddev);
4118 mddev_unlock(mddev);
4122 static struct md_sysfs_entry md_level =
4123 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4126 layout_show(struct mddev *mddev, char *page)
4128 /* just a number, not meaningful for all levels */
4129 if (mddev->reshape_position != MaxSector &&
4130 mddev->layout != mddev->new_layout)
4131 return sprintf(page, "%d (%d)\n",
4132 mddev->new_layout, mddev->layout);
4133 return sprintf(page, "%d\n", mddev->layout);
4137 layout_store(struct mddev *mddev, const char *buf, size_t len)
4142 err = kstrtouint(buf, 10, &n);
4145 err = mddev_lock(mddev);
4150 if (mddev->pers->check_reshape == NULL)
4155 mddev->new_layout = n;
4156 err = mddev->pers->check_reshape(mddev);
4158 mddev->new_layout = mddev->layout;
4161 mddev->new_layout = n;
4162 if (mddev->reshape_position == MaxSector)
4165 mddev_unlock(mddev);
4168 static struct md_sysfs_entry md_layout =
4169 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4172 raid_disks_show(struct mddev *mddev, char *page)
4174 if (mddev->raid_disks == 0)
4176 if (mddev->reshape_position != MaxSector &&
4177 mddev->delta_disks != 0)
4178 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4179 mddev->raid_disks - mddev->delta_disks);
4180 return sprintf(page, "%d\n", mddev->raid_disks);
4183 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4186 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4191 err = kstrtouint(buf, 10, &n);
4195 err = mddev_lock(mddev);
4199 err = update_raid_disks(mddev, n);
4200 else if (mddev->reshape_position != MaxSector) {
4201 struct md_rdev *rdev;
4202 int olddisks = mddev->raid_disks - mddev->delta_disks;
4205 rdev_for_each(rdev, mddev) {
4207 rdev->data_offset < rdev->new_data_offset)
4210 rdev->data_offset > rdev->new_data_offset)
4214 mddev->delta_disks = n - olddisks;
4215 mddev->raid_disks = n;
4216 mddev->reshape_backwards = (mddev->delta_disks < 0);
4218 mddev->raid_disks = n;
4220 mddev_unlock(mddev);
4221 return err ? err : len;
4223 static struct md_sysfs_entry md_raid_disks =
4224 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4227 uuid_show(struct mddev *mddev, char *page)
4229 return sprintf(page, "%pU\n", mddev->uuid);
4231 static struct md_sysfs_entry md_uuid =
4232 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4235 chunk_size_show(struct mddev *mddev, char *page)
4237 if (mddev->reshape_position != MaxSector &&
4238 mddev->chunk_sectors != mddev->new_chunk_sectors)
4239 return sprintf(page, "%d (%d)\n",
4240 mddev->new_chunk_sectors << 9,
4241 mddev->chunk_sectors << 9);
4242 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4246 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4251 err = kstrtoul(buf, 10, &n);
4255 err = mddev_lock(mddev);
4259 if (mddev->pers->check_reshape == NULL)
4264 mddev->new_chunk_sectors = n >> 9;
4265 err = mddev->pers->check_reshape(mddev);
4267 mddev->new_chunk_sectors = mddev->chunk_sectors;
4270 mddev->new_chunk_sectors = n >> 9;
4271 if (mddev->reshape_position == MaxSector)
4272 mddev->chunk_sectors = n >> 9;
4274 mddev_unlock(mddev);
4277 static struct md_sysfs_entry md_chunk_size =
4278 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4281 resync_start_show(struct mddev *mddev, char *page)
4283 if (mddev->recovery_cp == MaxSector)
4284 return sprintf(page, "none\n");
4285 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4289 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4291 unsigned long long n;
4294 if (cmd_match(buf, "none"))
4297 err = kstrtoull(buf, 10, &n);
4300 if (n != (sector_t)n)
4304 err = mddev_lock(mddev);
4307 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4311 mddev->recovery_cp = n;
4313 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4315 mddev_unlock(mddev);
4318 static struct md_sysfs_entry md_resync_start =
4319 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4320 resync_start_show, resync_start_store);
4323 * The array state can be:
4326 * No devices, no size, no level
4327 * Equivalent to STOP_ARRAY ioctl
4329 * May have some settings, but array is not active
4330 * all IO results in error
4331 * When written, doesn't tear down array, but just stops it
4332 * suspended (not supported yet)
4333 * All IO requests will block. The array can be reconfigured.
4334 * Writing this, if accepted, will block until array is quiescent
4336 * no resync can happen. no superblocks get written.
4337 * write requests fail
4339 * like readonly, but behaves like 'clean' on a write request.
4341 * clean - no pending writes, but otherwise active.
4342 * When written to inactive array, starts without resync
4343 * If a write request arrives then
4344 * if metadata is known, mark 'dirty' and switch to 'active'.
4345 * if not known, block and switch to write-pending
4346 * If written to an active array that has pending writes, then fails.
4348 * fully active: IO and resync can be happening.
4349 * When written to inactive array, starts with resync
4352 * clean, but writes are blocked waiting for 'active' to be written.
4355 * like active, but no writes have been seen for a while (100msec).
4358 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4359 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4360 * when a member is gone, so this state will at least alert the
4361 * user that something is wrong.
4363 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4364 write_pending, active_idle, broken, bad_word};
4365 static char *array_states[] = {
4366 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4367 "write-pending", "active-idle", "broken", NULL };
4369 static int match_word(const char *word, char **list)
4372 for (n=0; list[n]; n++)
4373 if (cmd_match(word, list[n]))
4379 array_state_show(struct mddev *mddev, char *page)
4381 enum array_state st = inactive;
4383 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4392 spin_lock(&mddev->lock);
4393 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4395 else if (mddev->in_sync)
4397 else if (mddev->safemode)
4401 spin_unlock(&mddev->lock);
4404 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4407 if (list_empty(&mddev->disks) &&
4408 mddev->raid_disks == 0 &&
4409 mddev->dev_sectors == 0)
4414 return sprintf(page, "%s\n", array_states[st]);
4417 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4418 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4419 static int restart_array(struct mddev *mddev);
4422 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4425 enum array_state st = match_word(buf, array_states);
4427 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4428 /* don't take reconfig_mutex when toggling between
4431 spin_lock(&mddev->lock);
4433 restart_array(mddev);
4434 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4435 md_wakeup_thread(mddev->thread);
4436 wake_up(&mddev->sb_wait);
4437 } else /* st == clean */ {
4438 restart_array(mddev);
4439 if (!set_in_sync(mddev))
4443 sysfs_notify_dirent_safe(mddev->sysfs_state);
4444 spin_unlock(&mddev->lock);
4447 err = mddev_lock(mddev);
4455 /* stopping an active array */
4456 err = do_md_stop(mddev, 0, NULL);
4459 /* stopping an active array */
4461 err = do_md_stop(mddev, 2, NULL);
4463 err = 0; /* already inactive */
4466 break; /* not supported yet */
4469 err = md_set_readonly(mddev, NULL);
4472 set_disk_ro(mddev->gendisk, 1);
4473 err = do_md_run(mddev);
4479 err = md_set_readonly(mddev, NULL);
4480 else if (mddev->ro == 1)
4481 err = restart_array(mddev);
4484 set_disk_ro(mddev->gendisk, 0);
4488 err = do_md_run(mddev);
4493 err = restart_array(mddev);
4496 spin_lock(&mddev->lock);
4497 if (!set_in_sync(mddev))
4499 spin_unlock(&mddev->lock);
4505 err = restart_array(mddev);
4508 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4509 wake_up(&mddev->sb_wait);
4513 set_disk_ro(mddev->gendisk, 0);
4514 err = do_md_run(mddev);
4520 /* these cannot be set */
4525 if (mddev->hold_active == UNTIL_IOCTL)
4526 mddev->hold_active = 0;
4527 sysfs_notify_dirent_safe(mddev->sysfs_state);
4529 mddev_unlock(mddev);
4532 static struct md_sysfs_entry md_array_state =
4533 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4536 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4537 return sprintf(page, "%d\n",
4538 atomic_read(&mddev->max_corr_read_errors));
4542 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4547 rv = kstrtouint(buf, 10, &n);
4550 atomic_set(&mddev->max_corr_read_errors, n);
4554 static struct md_sysfs_entry max_corr_read_errors =
4555 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4556 max_corrected_read_errors_store);
4559 null_show(struct mddev *mddev, char *page)
4564 /* need to ensure rdev_delayed_delete() has completed */
4565 static void flush_rdev_wq(struct mddev *mddev)
4567 struct md_rdev *rdev;
4570 rdev_for_each_rcu(rdev, mddev)
4571 if (work_pending(&rdev->del_work)) {
4572 flush_workqueue(md_rdev_misc_wq);
4579 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4581 /* buf must be %d:%d\n? giving major and minor numbers */
4582 /* The new device is added to the array.
4583 * If the array has a persistent superblock, we read the
4584 * superblock to initialise info and check validity.
4585 * Otherwise, only checking done is that in bind_rdev_to_array,
4586 * which mainly checks size.
4589 int major = simple_strtoul(buf, &e, 10);
4592 struct md_rdev *rdev;
4595 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4597 minor = simple_strtoul(e+1, &e, 10);
4598 if (*e && *e != '\n')
4600 dev = MKDEV(major, minor);
4601 if (major != MAJOR(dev) ||
4602 minor != MINOR(dev))
4605 flush_rdev_wq(mddev);
4606 err = mddev_lock(mddev);
4609 if (mddev->persistent) {
4610 rdev = md_import_device(dev, mddev->major_version,
4611 mddev->minor_version);
4612 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4613 struct md_rdev *rdev0
4614 = list_entry(mddev->disks.next,
4615 struct md_rdev, same_set);
4616 err = super_types[mddev->major_version]
4617 .load_super(rdev, rdev0, mddev->minor_version);
4621 } else if (mddev->external)
4622 rdev = md_import_device(dev, -2, -1);
4624 rdev = md_import_device(dev, -1, -1);
4627 mddev_unlock(mddev);
4628 return PTR_ERR(rdev);
4630 err = bind_rdev_to_array(rdev, mddev);
4634 mddev_unlock(mddev);
4636 md_new_event(mddev);
4637 return err ? err : len;
4640 static struct md_sysfs_entry md_new_device =
4641 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4644 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4647 unsigned long chunk, end_chunk;
4650 err = mddev_lock(mddev);
4655 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4657 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4658 if (buf == end) break;
4659 if (*end == '-') { /* range */
4661 end_chunk = simple_strtoul(buf, &end, 0);
4662 if (buf == end) break;
4664 if (*end && !isspace(*end)) break;
4665 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4666 buf = skip_spaces(end);
4668 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4670 mddev_unlock(mddev);
4674 static struct md_sysfs_entry md_bitmap =
4675 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4678 size_show(struct mddev *mddev, char *page)
4680 return sprintf(page, "%llu\n",
4681 (unsigned long long)mddev->dev_sectors / 2);
4684 static int update_size(struct mddev *mddev, sector_t num_sectors);
4687 size_store(struct mddev *mddev, const char *buf, size_t len)
4689 /* If array is inactive, we can reduce the component size, but
4690 * not increase it (except from 0).
4691 * If array is active, we can try an on-line resize
4694 int err = strict_blocks_to_sectors(buf, §ors);
4698 err = mddev_lock(mddev);
4702 err = update_size(mddev, sectors);
4704 md_update_sb(mddev, 1);
4706 if (mddev->dev_sectors == 0 ||
4707 mddev->dev_sectors > sectors)
4708 mddev->dev_sectors = sectors;
4712 mddev_unlock(mddev);
4713 return err ? err : len;
4716 static struct md_sysfs_entry md_size =
4717 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4719 /* Metadata version.
4721 * 'none' for arrays with no metadata (good luck...)
4722 * 'external' for arrays with externally managed metadata,
4723 * or N.M for internally known formats
4726 metadata_show(struct mddev *mddev, char *page)
4728 if (mddev->persistent)
4729 return sprintf(page, "%d.%d\n",
4730 mddev->major_version, mddev->minor_version);
4731 else if (mddev->external)
4732 return sprintf(page, "external:%s\n", mddev->metadata_type);
4734 return sprintf(page, "none\n");
4738 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4743 /* Changing the details of 'external' metadata is
4744 * always permitted. Otherwise there must be
4745 * no devices attached to the array.
4748 err = mddev_lock(mddev);
4752 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4754 else if (!list_empty(&mddev->disks))
4758 if (cmd_match(buf, "none")) {
4759 mddev->persistent = 0;
4760 mddev->external = 0;
4761 mddev->major_version = 0;
4762 mddev->minor_version = 90;
4765 if (strncmp(buf, "external:", 9) == 0) {
4766 size_t namelen = len-9;
4767 if (namelen >= sizeof(mddev->metadata_type))
4768 namelen = sizeof(mddev->metadata_type)-1;
4769 strncpy(mddev->metadata_type, buf+9, namelen);
4770 mddev->metadata_type[namelen] = 0;
4771 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4772 mddev->metadata_type[--namelen] = 0;
4773 mddev->persistent = 0;
4774 mddev->external = 1;
4775 mddev->major_version = 0;
4776 mddev->minor_version = 90;
4779 major = simple_strtoul(buf, &e, 10);
4781 if (e==buf || *e != '.')
4784 minor = simple_strtoul(buf, &e, 10);
4785 if (e==buf || (*e && *e != '\n') )
4788 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4790 mddev->major_version = major;
4791 mddev->minor_version = minor;
4792 mddev->persistent = 1;
4793 mddev->external = 0;
4796 mddev_unlock(mddev);
4800 static struct md_sysfs_entry md_metadata =
4801 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4804 action_show(struct mddev *mddev, char *page)
4806 char *type = "idle";
4807 unsigned long recovery = mddev->recovery;
4808 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4810 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4811 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4812 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4814 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4815 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4817 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4821 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4823 else if (mddev->reshape_position != MaxSector)
4826 return sprintf(page, "%s\n", type);
4830 action_store(struct mddev *mddev, const char *page, size_t len)
4832 if (!mddev->pers || !mddev->pers->sync_request)
4836 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4837 if (cmd_match(page, "frozen"))
4838 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4840 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4841 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4842 mddev_lock(mddev) == 0) {
4843 if (work_pending(&mddev->del_work))
4844 flush_workqueue(md_misc_wq);
4845 if (mddev->sync_thread) {
4846 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4847 md_reap_sync_thread(mddev);
4849 mddev_unlock(mddev);
4851 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4853 else if (cmd_match(page, "resync"))
4854 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4855 else if (cmd_match(page, "recover")) {
4856 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4857 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4858 } else if (cmd_match(page, "reshape")) {
4860 if (mddev->pers->start_reshape == NULL)
4862 err = mddev_lock(mddev);
4864 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4867 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4868 err = mddev->pers->start_reshape(mddev);
4870 mddev_unlock(mddev);
4874 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4876 if (cmd_match(page, "check"))
4877 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4878 else if (!cmd_match(page, "repair"))
4880 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4881 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4882 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4884 if (mddev->ro == 2) {
4885 /* A write to sync_action is enough to justify
4886 * canceling read-auto mode
4889 md_wakeup_thread(mddev->sync_thread);
4891 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4892 md_wakeup_thread(mddev->thread);
4893 sysfs_notify_dirent_safe(mddev->sysfs_action);
4897 static struct md_sysfs_entry md_scan_mode =
4898 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4901 last_sync_action_show(struct mddev *mddev, char *page)
4903 return sprintf(page, "%s\n", mddev->last_sync_action);
4906 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4909 mismatch_cnt_show(struct mddev *mddev, char *page)
4911 return sprintf(page, "%llu\n",
4912 (unsigned long long)
4913 atomic64_read(&mddev->resync_mismatches));
4916 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4919 sync_min_show(struct mddev *mddev, char *page)
4921 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4922 mddev->sync_speed_min ? "local": "system");
4926 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4931 if (strncmp(buf, "system", 6)==0) {
4934 rv = kstrtouint(buf, 10, &min);
4940 mddev->sync_speed_min = min;
4944 static struct md_sysfs_entry md_sync_min =
4945 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4948 sync_max_show(struct mddev *mddev, char *page)
4950 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4951 mddev->sync_speed_max ? "local": "system");
4955 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4960 if (strncmp(buf, "system", 6)==0) {
4963 rv = kstrtouint(buf, 10, &max);
4969 mddev->sync_speed_max = max;
4973 static struct md_sysfs_entry md_sync_max =
4974 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4977 degraded_show(struct mddev *mddev, char *page)
4979 return sprintf(page, "%d\n", mddev->degraded);
4981 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4984 sync_force_parallel_show(struct mddev *mddev, char *page)
4986 return sprintf(page, "%d\n", mddev->parallel_resync);
4990 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4994 if (kstrtol(buf, 10, &n))
4997 if (n != 0 && n != 1)
5000 mddev->parallel_resync = n;
5002 if (mddev->sync_thread)
5003 wake_up(&resync_wait);
5008 /* force parallel resync, even with shared block devices */
5009 static struct md_sysfs_entry md_sync_force_parallel =
5010 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5011 sync_force_parallel_show, sync_force_parallel_store);
5014 sync_speed_show(struct mddev *mddev, char *page)
5016 unsigned long resync, dt, db;
5017 if (mddev->curr_resync == 0)
5018 return sprintf(page, "none\n");
5019 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5020 dt = (jiffies - mddev->resync_mark) / HZ;
5022 db = resync - mddev->resync_mark_cnt;
5023 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5026 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5029 sync_completed_show(struct mddev *mddev, char *page)
5031 unsigned long long max_sectors, resync;
5033 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5034 return sprintf(page, "none\n");
5036 if (mddev->curr_resync == 1 ||
5037 mddev->curr_resync == 2)
5038 return sprintf(page, "delayed\n");
5040 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5041 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5042 max_sectors = mddev->resync_max_sectors;
5044 max_sectors = mddev->dev_sectors;
5046 resync = mddev->curr_resync_completed;
5047 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5050 static struct md_sysfs_entry md_sync_completed =
5051 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5054 min_sync_show(struct mddev *mddev, char *page)
5056 return sprintf(page, "%llu\n",
5057 (unsigned long long)mddev->resync_min);
5060 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5062 unsigned long long min;
5065 if (kstrtoull(buf, 10, &min))
5068 spin_lock(&mddev->lock);
5070 if (min > mddev->resync_max)
5074 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5077 /* Round down to multiple of 4K for safety */
5078 mddev->resync_min = round_down(min, 8);
5082 spin_unlock(&mddev->lock);
5086 static struct md_sysfs_entry md_min_sync =
5087 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5090 max_sync_show(struct mddev *mddev, char *page)
5092 if (mddev->resync_max == MaxSector)
5093 return sprintf(page, "max\n");
5095 return sprintf(page, "%llu\n",
5096 (unsigned long long)mddev->resync_max);
5099 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5102 spin_lock(&mddev->lock);
5103 if (strncmp(buf, "max", 3) == 0)
5104 mddev->resync_max = MaxSector;
5106 unsigned long long max;
5110 if (kstrtoull(buf, 10, &max))
5112 if (max < mddev->resync_min)
5116 if (max < mddev->resync_max &&
5118 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5121 /* Must be a multiple of chunk_size */
5122 chunk = mddev->chunk_sectors;
5124 sector_t temp = max;
5127 if (sector_div(temp, chunk))
5130 mddev->resync_max = max;
5132 wake_up(&mddev->recovery_wait);
5135 spin_unlock(&mddev->lock);
5139 static struct md_sysfs_entry md_max_sync =
5140 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5143 suspend_lo_show(struct mddev *mddev, char *page)
5145 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5149 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5151 unsigned long long new;
5154 err = kstrtoull(buf, 10, &new);
5157 if (new != (sector_t)new)
5160 err = mddev_lock(mddev);
5164 if (mddev->pers == NULL ||
5165 mddev->pers->quiesce == NULL)
5167 mddev_suspend(mddev);
5168 mddev->suspend_lo = new;
5169 mddev_resume(mddev);
5173 mddev_unlock(mddev);
5176 static struct md_sysfs_entry md_suspend_lo =
5177 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5180 suspend_hi_show(struct mddev *mddev, char *page)
5182 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5186 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5188 unsigned long long new;
5191 err = kstrtoull(buf, 10, &new);
5194 if (new != (sector_t)new)
5197 err = mddev_lock(mddev);
5201 if (mddev->pers == NULL)
5204 mddev_suspend(mddev);
5205 mddev->suspend_hi = new;
5206 mddev_resume(mddev);
5210 mddev_unlock(mddev);
5213 static struct md_sysfs_entry md_suspend_hi =
5214 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5217 reshape_position_show(struct mddev *mddev, char *page)
5219 if (mddev->reshape_position != MaxSector)
5220 return sprintf(page, "%llu\n",
5221 (unsigned long long)mddev->reshape_position);
5222 strcpy(page, "none\n");
5227 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5229 struct md_rdev *rdev;
5230 unsigned long long new;
5233 err = kstrtoull(buf, 10, &new);
5236 if (new != (sector_t)new)
5238 err = mddev_lock(mddev);
5244 mddev->reshape_position = new;
5245 mddev->delta_disks = 0;
5246 mddev->reshape_backwards = 0;
5247 mddev->new_level = mddev->level;
5248 mddev->new_layout = mddev->layout;
5249 mddev->new_chunk_sectors = mddev->chunk_sectors;
5250 rdev_for_each(rdev, mddev)
5251 rdev->new_data_offset = rdev->data_offset;
5254 mddev_unlock(mddev);
5258 static struct md_sysfs_entry md_reshape_position =
5259 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5260 reshape_position_store);
5263 reshape_direction_show(struct mddev *mddev, char *page)
5265 return sprintf(page, "%s\n",
5266 mddev->reshape_backwards ? "backwards" : "forwards");
5270 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5275 if (cmd_match(buf, "forwards"))
5277 else if (cmd_match(buf, "backwards"))
5281 if (mddev->reshape_backwards == backwards)
5284 err = mddev_lock(mddev);
5287 /* check if we are allowed to change */
5288 if (mddev->delta_disks)
5290 else if (mddev->persistent &&
5291 mddev->major_version == 0)
5294 mddev->reshape_backwards = backwards;
5295 mddev_unlock(mddev);
5299 static struct md_sysfs_entry md_reshape_direction =
5300 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5301 reshape_direction_store);
5304 array_size_show(struct mddev *mddev, char *page)
5306 if (mddev->external_size)
5307 return sprintf(page, "%llu\n",
5308 (unsigned long long)mddev->array_sectors/2);
5310 return sprintf(page, "default\n");
5314 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5319 err = mddev_lock(mddev);
5323 /* cluster raid doesn't support change array_sectors */
5324 if (mddev_is_clustered(mddev)) {
5325 mddev_unlock(mddev);
5329 if (strncmp(buf, "default", 7) == 0) {
5331 sectors = mddev->pers->size(mddev, 0, 0);
5333 sectors = mddev->array_sectors;
5335 mddev->external_size = 0;
5337 if (strict_blocks_to_sectors(buf, §ors) < 0)
5339 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5342 mddev->external_size = 1;
5346 mddev->array_sectors = sectors;
5348 set_capacity_and_notify(mddev->gendisk,
5349 mddev->array_sectors);
5351 mddev_unlock(mddev);
5355 static struct md_sysfs_entry md_array_size =
5356 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5360 consistency_policy_show(struct mddev *mddev, char *page)
5364 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5365 ret = sprintf(page, "journal\n");
5366 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5367 ret = sprintf(page, "ppl\n");
5368 } else if (mddev->bitmap) {
5369 ret = sprintf(page, "bitmap\n");
5370 } else if (mddev->pers) {
5371 if (mddev->pers->sync_request)
5372 ret = sprintf(page, "resync\n");
5374 ret = sprintf(page, "none\n");
5376 ret = sprintf(page, "unknown\n");
5383 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5388 if (mddev->pers->change_consistency_policy)
5389 err = mddev->pers->change_consistency_policy(mddev, buf);
5392 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5393 set_bit(MD_HAS_PPL, &mddev->flags);
5398 return err ? err : len;
5401 static struct md_sysfs_entry md_consistency_policy =
5402 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5403 consistency_policy_store);
5405 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5407 return sprintf(page, "%d\n", mddev->fail_last_dev);
5411 * Setting fail_last_dev to true to allow last device to be forcibly removed
5412 * from RAID1/RAID10.
5415 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5420 ret = kstrtobool(buf, &value);
5424 if (value != mddev->fail_last_dev)
5425 mddev->fail_last_dev = value;
5429 static struct md_sysfs_entry md_fail_last_dev =
5430 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5431 fail_last_dev_store);
5433 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5435 if (mddev->pers == NULL || (mddev->pers->level != 1))
5436 return sprintf(page, "n/a\n");
5438 return sprintf(page, "%d\n", mddev->serialize_policy);
5442 * Setting serialize_policy to true to enforce write IO is not reordered
5446 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5451 err = kstrtobool(buf, &value);
5455 if (value == mddev->serialize_policy)
5458 err = mddev_lock(mddev);
5461 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5462 pr_err("md: serialize_policy is only effective for raid1\n");
5467 mddev_suspend(mddev);
5469 mddev_create_serial_pool(mddev, NULL, true);
5471 mddev_destroy_serial_pool(mddev, NULL, true);
5472 mddev->serialize_policy = value;
5473 mddev_resume(mddev);
5475 mddev_unlock(mddev);
5479 static struct md_sysfs_entry md_serialize_policy =
5480 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5481 serialize_policy_store);
5484 static struct attribute *md_default_attrs[] = {
5487 &md_raid_disks.attr,
5489 &md_chunk_size.attr,
5491 &md_resync_start.attr,
5493 &md_new_device.attr,
5494 &md_safe_delay.attr,
5495 &md_array_state.attr,
5496 &md_reshape_position.attr,
5497 &md_reshape_direction.attr,
5498 &md_array_size.attr,
5499 &max_corr_read_errors.attr,
5500 &md_consistency_policy.attr,
5501 &md_fail_last_dev.attr,
5502 &md_serialize_policy.attr,
5506 static struct attribute *md_redundancy_attrs[] = {
5508 &md_last_scan_mode.attr,
5509 &md_mismatches.attr,
5512 &md_sync_speed.attr,
5513 &md_sync_force_parallel.attr,
5514 &md_sync_completed.attr,
5517 &md_suspend_lo.attr,
5518 &md_suspend_hi.attr,
5523 static const struct attribute_group md_redundancy_group = {
5525 .attrs = md_redundancy_attrs,
5529 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5531 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5532 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5537 spin_lock(&all_mddevs_lock);
5538 if (list_empty(&mddev->all_mddevs)) {
5539 spin_unlock(&all_mddevs_lock);
5543 spin_unlock(&all_mddevs_lock);
5545 rv = entry->show(mddev, page);
5551 md_attr_store(struct kobject *kobj, struct attribute *attr,
5552 const char *page, size_t length)
5554 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5555 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5560 if (!capable(CAP_SYS_ADMIN))
5562 spin_lock(&all_mddevs_lock);
5563 if (list_empty(&mddev->all_mddevs)) {
5564 spin_unlock(&all_mddevs_lock);
5568 spin_unlock(&all_mddevs_lock);
5569 rv = entry->store(mddev, page, length);
5574 static void md_free(struct kobject *ko)
5576 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5578 if (mddev->sysfs_state)
5579 sysfs_put(mddev->sysfs_state);
5580 if (mddev->sysfs_level)
5581 sysfs_put(mddev->sysfs_level);
5583 if (mddev->gendisk) {
5584 del_gendisk(mddev->gendisk);
5585 blk_cleanup_disk(mddev->gendisk);
5587 percpu_ref_exit(&mddev->writes_pending);
5589 bioset_exit(&mddev->bio_set);
5590 bioset_exit(&mddev->sync_set);
5594 static const struct sysfs_ops md_sysfs_ops = {
5595 .show = md_attr_show,
5596 .store = md_attr_store,
5598 static struct kobj_type md_ktype = {
5600 .sysfs_ops = &md_sysfs_ops,
5601 .default_attrs = md_default_attrs,
5606 static void mddev_delayed_delete(struct work_struct *ws)
5608 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5610 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5611 kobject_del(&mddev->kobj);
5612 kobject_put(&mddev->kobj);
5615 static void no_op(struct percpu_ref *r) {}
5617 int mddev_init_writes_pending(struct mddev *mddev)
5619 if (mddev->writes_pending.percpu_count_ptr)
5621 if (percpu_ref_init(&mddev->writes_pending, no_op,
5622 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5624 /* We want to start with the refcount at zero */
5625 percpu_ref_put(&mddev->writes_pending);
5628 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5630 static int md_alloc(dev_t dev, char *name)
5633 * If dev is zero, name is the name of a device to allocate with
5634 * an arbitrary minor number. It will be "md_???"
5635 * If dev is non-zero it must be a device number with a MAJOR of
5636 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5637 * the device is being created by opening a node in /dev.
5638 * If "name" is not NULL, the device is being created by
5639 * writing to /sys/module/md_mod/parameters/new_array.
5641 static DEFINE_MUTEX(disks_mutex);
5642 struct mddev *mddev;
5643 struct gendisk *disk;
5650 * Wait for any previous instance of this device to be completely
5651 * removed (mddev_delayed_delete).
5653 flush_workqueue(md_misc_wq);
5654 flush_workqueue(md_rdev_misc_wq);
5656 mutex_lock(&disks_mutex);
5657 mddev = mddev_alloc(dev);
5658 if (IS_ERR(mddev)) {
5659 mutex_unlock(&disks_mutex);
5660 return PTR_ERR(mddev);
5663 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5664 shift = partitioned ? MdpMinorShift : 0;
5665 unit = MINOR(mddev->unit) >> shift;
5668 /* Need to ensure that 'name' is not a duplicate.
5670 struct mddev *mddev2;
5671 spin_lock(&all_mddevs_lock);
5673 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5674 if (mddev2->gendisk &&
5675 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5676 spin_unlock(&all_mddevs_lock);
5680 spin_unlock(&all_mddevs_lock);
5684 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5686 mddev->hold_active = UNTIL_STOP;
5689 disk = blk_alloc_disk(NUMA_NO_NODE);
5693 disk->major = MAJOR(mddev->unit);
5694 disk->first_minor = unit << shift;
5695 disk->minors = 1 << shift;
5697 strcpy(disk->disk_name, name);
5698 else if (partitioned)
5699 sprintf(disk->disk_name, "md_d%d", unit);
5701 sprintf(disk->disk_name, "md%d", unit);
5702 disk->fops = &md_fops;
5703 disk->private_data = mddev;
5705 mddev->queue = disk->queue;
5706 blk_set_stacking_limits(&mddev->queue->limits);
5707 blk_queue_write_cache(mddev->queue, true, true);
5708 /* Allow extended partitions. This makes the
5709 * 'mdp' device redundant, but we can't really
5712 disk->flags |= GENHD_FL_EXT_DEVT;
5713 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5714 mddev->gendisk = disk;
5717 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5719 /* This isn't possible, but as kobject_init_and_add is marked
5720 * __must_check, we must do something with the result
5722 pr_debug("md: cannot register %s/md - name in use\n",
5726 if (mddev->kobj.sd &&
5727 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5728 pr_debug("pointless warning\n");
5730 mutex_unlock(&disks_mutex);
5731 if (!error && mddev->kobj.sd) {
5732 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5733 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5734 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5740 static void md_probe(dev_t dev)
5742 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5745 md_alloc(dev, NULL);
5748 static int add_named_array(const char *val, const struct kernel_param *kp)
5751 * val must be "md_*" or "mdNNN".
5752 * For "md_*" we allocate an array with a large free minor number, and
5753 * set the name to val. val must not already be an active name.
5754 * For "mdNNN" we allocate an array with the minor number NNN
5755 * which must not already be in use.
5757 int len = strlen(val);
5758 char buf[DISK_NAME_LEN];
5759 unsigned long devnum;
5761 while (len && val[len-1] == '\n')
5763 if (len >= DISK_NAME_LEN)
5765 strlcpy(buf, val, len+1);
5766 if (strncmp(buf, "md_", 3) == 0)
5767 return md_alloc(0, buf);
5768 if (strncmp(buf, "md", 2) == 0 &&
5770 kstrtoul(buf+2, 10, &devnum) == 0 &&
5771 devnum <= MINORMASK)
5772 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5777 static void md_safemode_timeout(struct timer_list *t)
5779 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5781 mddev->safemode = 1;
5782 if (mddev->external)
5783 sysfs_notify_dirent_safe(mddev->sysfs_state);
5785 md_wakeup_thread(mddev->thread);
5788 static int start_dirty_degraded;
5790 int md_run(struct mddev *mddev)
5793 struct md_rdev *rdev;
5794 struct md_personality *pers;
5796 if (list_empty(&mddev->disks))
5797 /* cannot run an array with no devices.. */
5802 /* Cannot run until previous stop completes properly */
5803 if (mddev->sysfs_active)
5807 * Analyze all RAID superblock(s)
5809 if (!mddev->raid_disks) {
5810 if (!mddev->persistent)
5812 err = analyze_sbs(mddev);
5817 if (mddev->level != LEVEL_NONE)
5818 request_module("md-level-%d", mddev->level);
5819 else if (mddev->clevel[0])
5820 request_module("md-%s", mddev->clevel);
5823 * Drop all container device buffers, from now on
5824 * the only valid external interface is through the md
5827 mddev->has_superblocks = false;
5828 rdev_for_each(rdev, mddev) {
5829 if (test_bit(Faulty, &rdev->flags))
5831 sync_blockdev(rdev->bdev);
5832 invalidate_bdev(rdev->bdev);
5833 if (mddev->ro != 1 && rdev_read_only(rdev)) {
5836 set_disk_ro(mddev->gendisk, 1);
5840 mddev->has_superblocks = true;
5842 /* perform some consistency tests on the device.
5843 * We don't want the data to overlap the metadata,
5844 * Internal Bitmap issues have been handled elsewhere.
5846 if (rdev->meta_bdev) {
5847 /* Nothing to check */;
5848 } else if (rdev->data_offset < rdev->sb_start) {
5849 if (mddev->dev_sectors &&
5850 rdev->data_offset + mddev->dev_sectors
5852 pr_warn("md: %s: data overlaps metadata\n",
5857 if (rdev->sb_start + rdev->sb_size/512
5858 > rdev->data_offset) {
5859 pr_warn("md: %s: metadata overlaps data\n",
5864 sysfs_notify_dirent_safe(rdev->sysfs_state);
5867 if (!bioset_initialized(&mddev->bio_set)) {
5868 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5872 if (!bioset_initialized(&mddev->sync_set)) {
5873 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5878 spin_lock(&pers_lock);
5879 pers = find_pers(mddev->level, mddev->clevel);
5880 if (!pers || !try_module_get(pers->owner)) {
5881 spin_unlock(&pers_lock);
5882 if (mddev->level != LEVEL_NONE)
5883 pr_warn("md: personality for level %d is not loaded!\n",
5886 pr_warn("md: personality for level %s is not loaded!\n",
5891 spin_unlock(&pers_lock);
5892 if (mddev->level != pers->level) {
5893 mddev->level = pers->level;
5894 mddev->new_level = pers->level;
5896 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5898 if (mddev->reshape_position != MaxSector &&
5899 pers->start_reshape == NULL) {
5900 /* This personality cannot handle reshaping... */
5901 module_put(pers->owner);
5906 if (pers->sync_request) {
5907 /* Warn if this is a potentially silly
5910 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5911 struct md_rdev *rdev2;
5914 rdev_for_each(rdev, mddev)
5915 rdev_for_each(rdev2, mddev) {
5917 rdev->bdev->bd_disk ==
5918 rdev2->bdev->bd_disk) {
5919 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5921 bdevname(rdev->bdev,b),
5922 bdevname(rdev2->bdev,b2));
5928 pr_warn("True protection against single-disk failure might be compromised.\n");
5931 mddev->recovery = 0;
5932 /* may be over-ridden by personality */
5933 mddev->resync_max_sectors = mddev->dev_sectors;
5935 mddev->ok_start_degraded = start_dirty_degraded;
5937 if (start_readonly && mddev->ro == 0)
5938 mddev->ro = 2; /* read-only, but switch on first write */
5940 err = pers->run(mddev);
5942 pr_warn("md: pers->run() failed ...\n");
5943 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5944 WARN_ONCE(!mddev->external_size,
5945 "%s: default size too small, but 'external_size' not in effect?\n",
5947 pr_warn("md: invalid array_size %llu > default size %llu\n",
5948 (unsigned long long)mddev->array_sectors / 2,
5949 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5952 if (err == 0 && pers->sync_request &&
5953 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5954 struct bitmap *bitmap;
5956 bitmap = md_bitmap_create(mddev, -1);
5957 if (IS_ERR(bitmap)) {
5958 err = PTR_ERR(bitmap);
5959 pr_warn("%s: failed to create bitmap (%d)\n",
5960 mdname(mddev), err);
5962 mddev->bitmap = bitmap;
5968 if (mddev->bitmap_info.max_write_behind > 0) {
5969 bool create_pool = false;
5971 rdev_for_each(rdev, mddev) {
5972 if (test_bit(WriteMostly, &rdev->flags) &&
5973 rdev_init_serial(rdev))
5976 if (create_pool && mddev->serial_info_pool == NULL) {
5977 mddev->serial_info_pool =
5978 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5979 sizeof(struct serial_info));
5980 if (!mddev->serial_info_pool) {
5990 rdev_for_each(rdev, mddev) {
5991 if (rdev->raid_disk >= 0 &&
5992 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5997 if (mddev->degraded)
6000 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6002 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6003 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6005 if (pers->sync_request) {
6006 if (mddev->kobj.sd &&
6007 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6008 pr_warn("md: cannot register extra attributes for %s\n",
6010 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6011 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6012 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6013 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6016 atomic_set(&mddev->max_corr_read_errors,
6017 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6018 mddev->safemode = 0;
6019 if (mddev_is_clustered(mddev))
6020 mddev->safemode_delay = 0;
6022 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6025 spin_lock(&mddev->lock);
6027 spin_unlock(&mddev->lock);
6028 rdev_for_each(rdev, mddev)
6029 if (rdev->raid_disk >= 0)
6030 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6032 if (mddev->degraded && !mddev->ro)
6033 /* This ensures that recovering status is reported immediately
6034 * via sysfs - until a lack of spares is confirmed.
6036 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6037 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6039 if (mddev->sb_flags)
6040 md_update_sb(mddev, 0);
6042 md_new_event(mddev);
6046 mddev_detach(mddev);
6048 pers->free(mddev, mddev->private);
6049 mddev->private = NULL;
6050 module_put(pers->owner);
6051 md_bitmap_destroy(mddev);
6053 bioset_exit(&mddev->sync_set);
6055 bioset_exit(&mddev->bio_set);
6058 EXPORT_SYMBOL_GPL(md_run);
6060 int do_md_run(struct mddev *mddev)
6064 set_bit(MD_NOT_READY, &mddev->flags);
6065 err = md_run(mddev);
6068 err = md_bitmap_load(mddev);
6070 md_bitmap_destroy(mddev);
6074 if (mddev_is_clustered(mddev))
6075 md_allow_write(mddev);
6077 /* run start up tasks that require md_thread */
6080 md_wakeup_thread(mddev->thread);
6081 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6083 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6084 clear_bit(MD_NOT_READY, &mddev->flags);
6086 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6087 sysfs_notify_dirent_safe(mddev->sysfs_state);
6088 sysfs_notify_dirent_safe(mddev->sysfs_action);
6089 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6091 clear_bit(MD_NOT_READY, &mddev->flags);
6095 int md_start(struct mddev *mddev)
6099 if (mddev->pers->start) {
6100 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6101 md_wakeup_thread(mddev->thread);
6102 ret = mddev->pers->start(mddev);
6103 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6104 md_wakeup_thread(mddev->sync_thread);
6108 EXPORT_SYMBOL_GPL(md_start);
6110 static int restart_array(struct mddev *mddev)
6112 struct gendisk *disk = mddev->gendisk;
6113 struct md_rdev *rdev;
6114 bool has_journal = false;
6115 bool has_readonly = false;
6117 /* Complain if it has no devices */
6118 if (list_empty(&mddev->disks))
6126 rdev_for_each_rcu(rdev, mddev) {
6127 if (test_bit(Journal, &rdev->flags) &&
6128 !test_bit(Faulty, &rdev->flags))
6130 if (rdev_read_only(rdev))
6131 has_readonly = true;
6134 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6135 /* Don't restart rw with journal missing/faulty */
6140 mddev->safemode = 0;
6142 set_disk_ro(disk, 0);
6143 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6144 /* Kick recovery or resync if necessary */
6145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6146 md_wakeup_thread(mddev->thread);
6147 md_wakeup_thread(mddev->sync_thread);
6148 sysfs_notify_dirent_safe(mddev->sysfs_state);
6152 static void md_clean(struct mddev *mddev)
6154 mddev->array_sectors = 0;
6155 mddev->external_size = 0;
6156 mddev->dev_sectors = 0;
6157 mddev->raid_disks = 0;
6158 mddev->recovery_cp = 0;
6159 mddev->resync_min = 0;
6160 mddev->resync_max = MaxSector;
6161 mddev->reshape_position = MaxSector;
6162 mddev->external = 0;
6163 mddev->persistent = 0;
6164 mddev->level = LEVEL_NONE;
6165 mddev->clevel[0] = 0;
6167 mddev->sb_flags = 0;
6169 mddev->metadata_type[0] = 0;
6170 mddev->chunk_sectors = 0;
6171 mddev->ctime = mddev->utime = 0;
6173 mddev->max_disks = 0;
6175 mddev->can_decrease_events = 0;
6176 mddev->delta_disks = 0;
6177 mddev->reshape_backwards = 0;
6178 mddev->new_level = LEVEL_NONE;
6179 mddev->new_layout = 0;
6180 mddev->new_chunk_sectors = 0;
6181 mddev->curr_resync = 0;
6182 atomic64_set(&mddev->resync_mismatches, 0);
6183 mddev->suspend_lo = mddev->suspend_hi = 0;
6184 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6185 mddev->recovery = 0;
6188 mddev->degraded = 0;
6189 mddev->safemode = 0;
6190 mddev->private = NULL;
6191 mddev->cluster_info = NULL;
6192 mddev->bitmap_info.offset = 0;
6193 mddev->bitmap_info.default_offset = 0;
6194 mddev->bitmap_info.default_space = 0;
6195 mddev->bitmap_info.chunksize = 0;
6196 mddev->bitmap_info.daemon_sleep = 0;
6197 mddev->bitmap_info.max_write_behind = 0;
6198 mddev->bitmap_info.nodes = 0;
6201 static void __md_stop_writes(struct mddev *mddev)
6203 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6204 if (work_pending(&mddev->del_work))
6205 flush_workqueue(md_misc_wq);
6206 if (mddev->sync_thread) {
6207 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6208 md_reap_sync_thread(mddev);
6211 del_timer_sync(&mddev->safemode_timer);
6213 if (mddev->pers && mddev->pers->quiesce) {
6214 mddev->pers->quiesce(mddev, 1);
6215 mddev->pers->quiesce(mddev, 0);
6217 md_bitmap_flush(mddev);
6219 if (mddev->ro == 0 &&
6220 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6222 /* mark array as shutdown cleanly */
6223 if (!mddev_is_clustered(mddev))
6225 md_update_sb(mddev, 1);
6227 /* disable policy to guarantee rdevs free resources for serialization */
6228 mddev->serialize_policy = 0;
6229 mddev_destroy_serial_pool(mddev, NULL, true);
6232 void md_stop_writes(struct mddev *mddev)
6234 mddev_lock_nointr(mddev);
6235 __md_stop_writes(mddev);
6236 mddev_unlock(mddev);
6238 EXPORT_SYMBOL_GPL(md_stop_writes);
6240 static void mddev_detach(struct mddev *mddev)
6242 md_bitmap_wait_behind_writes(mddev);
6243 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6244 mddev->pers->quiesce(mddev, 1);
6245 mddev->pers->quiesce(mddev, 0);
6247 md_unregister_thread(&mddev->thread);
6249 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6252 static void __md_stop(struct mddev *mddev)
6254 struct md_personality *pers = mddev->pers;
6255 md_bitmap_destroy(mddev);
6256 mddev_detach(mddev);
6257 /* Ensure ->event_work is done */
6258 if (mddev->event_work.func)
6259 flush_workqueue(md_misc_wq);
6260 spin_lock(&mddev->lock);
6262 spin_unlock(&mddev->lock);
6263 pers->free(mddev, mddev->private);
6264 mddev->private = NULL;
6265 if (pers->sync_request && mddev->to_remove == NULL)
6266 mddev->to_remove = &md_redundancy_group;
6267 module_put(pers->owner);
6268 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6271 void md_stop(struct mddev *mddev)
6273 /* stop the array and free an attached data structures.
6274 * This is called from dm-raid
6276 __md_stop_writes(mddev);
6278 bioset_exit(&mddev->bio_set);
6279 bioset_exit(&mddev->sync_set);
6282 EXPORT_SYMBOL_GPL(md_stop);
6284 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6289 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6291 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6292 md_wakeup_thread(mddev->thread);
6294 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6295 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6296 if (mddev->sync_thread)
6297 /* Thread might be blocked waiting for metadata update
6298 * which will now never happen */
6299 wake_up_process(mddev->sync_thread->tsk);
6301 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6303 mddev_unlock(mddev);
6304 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6306 wait_event(mddev->sb_wait,
6307 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6308 mddev_lock_nointr(mddev);
6310 mutex_lock(&mddev->open_mutex);
6311 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6312 mddev->sync_thread ||
6313 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6314 pr_warn("md: %s still in use.\n",mdname(mddev));
6316 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6317 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6318 md_wakeup_thread(mddev->thread);
6324 __md_stop_writes(mddev);
6330 set_disk_ro(mddev->gendisk, 1);
6331 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6332 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6333 md_wakeup_thread(mddev->thread);
6334 sysfs_notify_dirent_safe(mddev->sysfs_state);
6338 mutex_unlock(&mddev->open_mutex);
6343 * 0 - completely stop and dis-assemble array
6344 * 2 - stop but do not disassemble array
6346 static int do_md_stop(struct mddev *mddev, int mode,
6347 struct block_device *bdev)
6349 struct gendisk *disk = mddev->gendisk;
6350 struct md_rdev *rdev;
6353 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6355 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6356 md_wakeup_thread(mddev->thread);
6358 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6359 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6360 if (mddev->sync_thread)
6361 /* Thread might be blocked waiting for metadata update
6362 * which will now never happen */
6363 wake_up_process(mddev->sync_thread->tsk);
6365 mddev_unlock(mddev);
6366 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6367 !test_bit(MD_RECOVERY_RUNNING,
6368 &mddev->recovery)));
6369 mddev_lock_nointr(mddev);
6371 mutex_lock(&mddev->open_mutex);
6372 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6373 mddev->sysfs_active ||
6374 mddev->sync_thread ||
6375 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6376 pr_warn("md: %s still in use.\n",mdname(mddev));
6377 mutex_unlock(&mddev->open_mutex);
6379 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6380 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6381 md_wakeup_thread(mddev->thread);
6387 set_disk_ro(disk, 0);
6389 __md_stop_writes(mddev);
6392 /* tell userspace to handle 'inactive' */
6393 sysfs_notify_dirent_safe(mddev->sysfs_state);
6395 rdev_for_each(rdev, mddev)
6396 if (rdev->raid_disk >= 0)
6397 sysfs_unlink_rdev(mddev, rdev);
6399 set_capacity_and_notify(disk, 0);
6400 mutex_unlock(&mddev->open_mutex);
6406 mutex_unlock(&mddev->open_mutex);
6408 * Free resources if final stop
6411 pr_info("md: %s stopped.\n", mdname(mddev));
6413 if (mddev->bitmap_info.file) {
6414 struct file *f = mddev->bitmap_info.file;
6415 spin_lock(&mddev->lock);
6416 mddev->bitmap_info.file = NULL;
6417 spin_unlock(&mddev->lock);
6420 mddev->bitmap_info.offset = 0;
6422 export_array(mddev);
6425 if (mddev->hold_active == UNTIL_STOP)
6426 mddev->hold_active = 0;
6428 md_new_event(mddev);
6429 sysfs_notify_dirent_safe(mddev->sysfs_state);
6434 static void autorun_array(struct mddev *mddev)
6436 struct md_rdev *rdev;
6439 if (list_empty(&mddev->disks))
6442 pr_info("md: running: ");
6444 rdev_for_each(rdev, mddev) {
6445 char b[BDEVNAME_SIZE];
6446 pr_cont("<%s>", bdevname(rdev->bdev,b));
6450 err = do_md_run(mddev);
6452 pr_warn("md: do_md_run() returned %d\n", err);
6453 do_md_stop(mddev, 0, NULL);
6458 * lets try to run arrays based on all disks that have arrived
6459 * until now. (those are in pending_raid_disks)
6461 * the method: pick the first pending disk, collect all disks with
6462 * the same UUID, remove all from the pending list and put them into
6463 * the 'same_array' list. Then order this list based on superblock
6464 * update time (freshest comes first), kick out 'old' disks and
6465 * compare superblocks. If everything's fine then run it.
6467 * If "unit" is allocated, then bump its reference count
6469 static void autorun_devices(int part)
6471 struct md_rdev *rdev0, *rdev, *tmp;
6472 struct mddev *mddev;
6473 char b[BDEVNAME_SIZE];
6475 pr_info("md: autorun ...\n");
6476 while (!list_empty(&pending_raid_disks)) {
6479 LIST_HEAD(candidates);
6480 rdev0 = list_entry(pending_raid_disks.next,
6481 struct md_rdev, same_set);
6483 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6484 INIT_LIST_HEAD(&candidates);
6485 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6486 if (super_90_load(rdev, rdev0, 0) >= 0) {
6487 pr_debug("md: adding %s ...\n",
6488 bdevname(rdev->bdev,b));
6489 list_move(&rdev->same_set, &candidates);
6492 * now we have a set of devices, with all of them having
6493 * mostly sane superblocks. It's time to allocate the
6497 dev = MKDEV(mdp_major,
6498 rdev0->preferred_minor << MdpMinorShift);
6499 unit = MINOR(dev) >> MdpMinorShift;
6501 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6504 if (rdev0->preferred_minor != unit) {
6505 pr_warn("md: unit number in %s is bad: %d\n",
6506 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6511 mddev = mddev_find(dev);
6515 if (mddev_lock(mddev))
6516 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6517 else if (mddev->raid_disks || mddev->major_version
6518 || !list_empty(&mddev->disks)) {
6519 pr_warn("md: %s already running, cannot run %s\n",
6520 mdname(mddev), bdevname(rdev0->bdev,b));
6521 mddev_unlock(mddev);
6523 pr_debug("md: created %s\n", mdname(mddev));
6524 mddev->persistent = 1;
6525 rdev_for_each_list(rdev, tmp, &candidates) {
6526 list_del_init(&rdev->same_set);
6527 if (bind_rdev_to_array(rdev, mddev))
6530 autorun_array(mddev);
6531 mddev_unlock(mddev);
6533 /* on success, candidates will be empty, on error
6536 rdev_for_each_list(rdev, tmp, &candidates) {
6537 list_del_init(&rdev->same_set);
6542 pr_info("md: ... autorun DONE.\n");
6544 #endif /* !MODULE */
6546 static int get_version(void __user *arg)
6550 ver.major = MD_MAJOR_VERSION;
6551 ver.minor = MD_MINOR_VERSION;
6552 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6554 if (copy_to_user(arg, &ver, sizeof(ver)))
6560 static int get_array_info(struct mddev *mddev, void __user *arg)
6562 mdu_array_info_t info;
6563 int nr,working,insync,failed,spare;
6564 struct md_rdev *rdev;
6566 nr = working = insync = failed = spare = 0;
6568 rdev_for_each_rcu(rdev, mddev) {
6570 if (test_bit(Faulty, &rdev->flags))
6574 if (test_bit(In_sync, &rdev->flags))
6576 else if (test_bit(Journal, &rdev->flags))
6577 /* TODO: add journal count to md_u.h */
6585 info.major_version = mddev->major_version;
6586 info.minor_version = mddev->minor_version;
6587 info.patch_version = MD_PATCHLEVEL_VERSION;
6588 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6589 info.level = mddev->level;
6590 info.size = mddev->dev_sectors / 2;
6591 if (info.size != mddev->dev_sectors / 2) /* overflow */
6594 info.raid_disks = mddev->raid_disks;
6595 info.md_minor = mddev->md_minor;
6596 info.not_persistent= !mddev->persistent;
6598 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6601 info.state = (1<<MD_SB_CLEAN);
6602 if (mddev->bitmap && mddev->bitmap_info.offset)
6603 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6604 if (mddev_is_clustered(mddev))
6605 info.state |= (1<<MD_SB_CLUSTERED);
6606 info.active_disks = insync;
6607 info.working_disks = working;
6608 info.failed_disks = failed;
6609 info.spare_disks = spare;
6611 info.layout = mddev->layout;
6612 info.chunk_size = mddev->chunk_sectors << 9;
6614 if (copy_to_user(arg, &info, sizeof(info)))
6620 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6622 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6626 file = kzalloc(sizeof(*file), GFP_NOIO);
6631 spin_lock(&mddev->lock);
6632 /* bitmap enabled */
6633 if (mddev->bitmap_info.file) {
6634 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6635 sizeof(file->pathname));
6639 memmove(file->pathname, ptr,
6640 sizeof(file->pathname)-(ptr-file->pathname));
6642 spin_unlock(&mddev->lock);
6645 copy_to_user(arg, file, sizeof(*file)))
6652 static int get_disk_info(struct mddev *mddev, void __user * arg)
6654 mdu_disk_info_t info;
6655 struct md_rdev *rdev;
6657 if (copy_from_user(&info, arg, sizeof(info)))
6661 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6663 info.major = MAJOR(rdev->bdev->bd_dev);
6664 info.minor = MINOR(rdev->bdev->bd_dev);
6665 info.raid_disk = rdev->raid_disk;
6667 if (test_bit(Faulty, &rdev->flags))
6668 info.state |= (1<<MD_DISK_FAULTY);
6669 else if (test_bit(In_sync, &rdev->flags)) {
6670 info.state |= (1<<MD_DISK_ACTIVE);
6671 info.state |= (1<<MD_DISK_SYNC);
6673 if (test_bit(Journal, &rdev->flags))
6674 info.state |= (1<<MD_DISK_JOURNAL);
6675 if (test_bit(WriteMostly, &rdev->flags))
6676 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6677 if (test_bit(FailFast, &rdev->flags))
6678 info.state |= (1<<MD_DISK_FAILFAST);
6680 info.major = info.minor = 0;
6681 info.raid_disk = -1;
6682 info.state = (1<<MD_DISK_REMOVED);
6686 if (copy_to_user(arg, &info, sizeof(info)))
6692 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6694 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6695 struct md_rdev *rdev;
6696 dev_t dev = MKDEV(info->major,info->minor);
6698 if (mddev_is_clustered(mddev) &&
6699 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6700 pr_warn("%s: Cannot add to clustered mddev.\n",
6705 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6708 if (!mddev->raid_disks) {
6710 /* expecting a device which has a superblock */
6711 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6713 pr_warn("md: md_import_device returned %ld\n",
6715 return PTR_ERR(rdev);
6717 if (!list_empty(&mddev->disks)) {
6718 struct md_rdev *rdev0
6719 = list_entry(mddev->disks.next,
6720 struct md_rdev, same_set);
6721 err = super_types[mddev->major_version]
6722 .load_super(rdev, rdev0, mddev->minor_version);
6724 pr_warn("md: %s has different UUID to %s\n",
6725 bdevname(rdev->bdev,b),
6726 bdevname(rdev0->bdev,b2));
6731 err = bind_rdev_to_array(rdev, mddev);
6738 * md_add_new_disk can be used once the array is assembled
6739 * to add "hot spares". They must already have a superblock
6744 if (!mddev->pers->hot_add_disk) {
6745 pr_warn("%s: personality does not support diskops!\n",
6749 if (mddev->persistent)
6750 rdev = md_import_device(dev, mddev->major_version,
6751 mddev->minor_version);
6753 rdev = md_import_device(dev, -1, -1);
6755 pr_warn("md: md_import_device returned %ld\n",
6757 return PTR_ERR(rdev);
6759 /* set saved_raid_disk if appropriate */
6760 if (!mddev->persistent) {
6761 if (info->state & (1<<MD_DISK_SYNC) &&
6762 info->raid_disk < mddev->raid_disks) {
6763 rdev->raid_disk = info->raid_disk;
6764 set_bit(In_sync, &rdev->flags);
6765 clear_bit(Bitmap_sync, &rdev->flags);
6767 rdev->raid_disk = -1;
6768 rdev->saved_raid_disk = rdev->raid_disk;
6770 super_types[mddev->major_version].
6771 validate_super(mddev, rdev);
6772 if ((info->state & (1<<MD_DISK_SYNC)) &&
6773 rdev->raid_disk != info->raid_disk) {
6774 /* This was a hot-add request, but events doesn't
6775 * match, so reject it.
6781 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6782 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6783 set_bit(WriteMostly, &rdev->flags);
6785 clear_bit(WriteMostly, &rdev->flags);
6786 if (info->state & (1<<MD_DISK_FAILFAST))
6787 set_bit(FailFast, &rdev->flags);
6789 clear_bit(FailFast, &rdev->flags);
6791 if (info->state & (1<<MD_DISK_JOURNAL)) {
6792 struct md_rdev *rdev2;
6793 bool has_journal = false;
6795 /* make sure no existing journal disk */
6796 rdev_for_each(rdev2, mddev) {
6797 if (test_bit(Journal, &rdev2->flags)) {
6802 if (has_journal || mddev->bitmap) {
6806 set_bit(Journal, &rdev->flags);
6809 * check whether the device shows up in other nodes
6811 if (mddev_is_clustered(mddev)) {
6812 if (info->state & (1 << MD_DISK_CANDIDATE))
6813 set_bit(Candidate, &rdev->flags);
6814 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6815 /* --add initiated by this node */
6816 err = md_cluster_ops->add_new_disk(mddev, rdev);
6824 rdev->raid_disk = -1;
6825 err = bind_rdev_to_array(rdev, mddev);
6830 if (mddev_is_clustered(mddev)) {
6831 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6833 err = md_cluster_ops->new_disk_ack(mddev,
6836 md_kick_rdev_from_array(rdev);
6840 md_cluster_ops->add_new_disk_cancel(mddev);
6842 err = add_bound_rdev(rdev);
6846 err = add_bound_rdev(rdev);
6851 /* otherwise, md_add_new_disk is only allowed
6852 * for major_version==0 superblocks
6854 if (mddev->major_version != 0) {
6855 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6859 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6861 rdev = md_import_device(dev, -1, 0);
6863 pr_warn("md: error, md_import_device() returned %ld\n",
6865 return PTR_ERR(rdev);
6867 rdev->desc_nr = info->number;
6868 if (info->raid_disk < mddev->raid_disks)
6869 rdev->raid_disk = info->raid_disk;
6871 rdev->raid_disk = -1;
6873 if (rdev->raid_disk < mddev->raid_disks)
6874 if (info->state & (1<<MD_DISK_SYNC))
6875 set_bit(In_sync, &rdev->flags);
6877 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6878 set_bit(WriteMostly, &rdev->flags);
6879 if (info->state & (1<<MD_DISK_FAILFAST))
6880 set_bit(FailFast, &rdev->flags);
6882 if (!mddev->persistent) {
6883 pr_debug("md: nonpersistent superblock ...\n");
6884 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6886 rdev->sb_start = calc_dev_sboffset(rdev);
6887 rdev->sectors = rdev->sb_start;
6889 err = bind_rdev_to_array(rdev, mddev);
6899 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6901 char b[BDEVNAME_SIZE];
6902 struct md_rdev *rdev;
6907 rdev = find_rdev(mddev, dev);
6911 if (rdev->raid_disk < 0)
6914 clear_bit(Blocked, &rdev->flags);
6915 remove_and_add_spares(mddev, rdev);
6917 if (rdev->raid_disk >= 0)
6921 if (mddev_is_clustered(mddev)) {
6922 if (md_cluster_ops->remove_disk(mddev, rdev))
6926 md_kick_rdev_from_array(rdev);
6927 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6929 md_wakeup_thread(mddev->thread);
6931 md_update_sb(mddev, 1);
6932 md_new_event(mddev);
6936 pr_debug("md: cannot remove active disk %s from %s ...\n",
6937 bdevname(rdev->bdev,b), mdname(mddev));
6941 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6943 char b[BDEVNAME_SIZE];
6945 struct md_rdev *rdev;
6950 if (mddev->major_version != 0) {
6951 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6955 if (!mddev->pers->hot_add_disk) {
6956 pr_warn("%s: personality does not support diskops!\n",
6961 rdev = md_import_device(dev, -1, 0);
6963 pr_warn("md: error, md_import_device() returned %ld\n",
6968 if (mddev->persistent)
6969 rdev->sb_start = calc_dev_sboffset(rdev);
6971 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6973 rdev->sectors = rdev->sb_start;
6975 if (test_bit(Faulty, &rdev->flags)) {
6976 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6977 bdevname(rdev->bdev,b), mdname(mddev));
6982 clear_bit(In_sync, &rdev->flags);
6984 rdev->saved_raid_disk = -1;
6985 err = bind_rdev_to_array(rdev, mddev);
6990 * The rest should better be atomic, we can have disk failures
6991 * noticed in interrupt contexts ...
6994 rdev->raid_disk = -1;
6996 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6998 md_update_sb(mddev, 1);
7000 * Kick recovery, maybe this spare has to be added to the
7001 * array immediately.
7003 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7004 md_wakeup_thread(mddev->thread);
7005 md_new_event(mddev);
7013 static int set_bitmap_file(struct mddev *mddev, int fd)
7018 if (!mddev->pers->quiesce || !mddev->thread)
7020 if (mddev->recovery || mddev->sync_thread)
7022 /* we should be able to change the bitmap.. */
7026 struct inode *inode;
7029 if (mddev->bitmap || mddev->bitmap_info.file)
7030 return -EEXIST; /* cannot add when bitmap is present */
7034 pr_warn("%s: error: failed to get bitmap file\n",
7039 inode = f->f_mapping->host;
7040 if (!S_ISREG(inode->i_mode)) {
7041 pr_warn("%s: error: bitmap file must be a regular file\n",
7044 } else if (!(f->f_mode & FMODE_WRITE)) {
7045 pr_warn("%s: error: bitmap file must open for write\n",
7048 } else if (atomic_read(&inode->i_writecount) != 1) {
7049 pr_warn("%s: error: bitmap file is already in use\n",
7057 mddev->bitmap_info.file = f;
7058 mddev->bitmap_info.offset = 0; /* file overrides offset */
7059 } else if (mddev->bitmap == NULL)
7060 return -ENOENT; /* cannot remove what isn't there */
7064 struct bitmap *bitmap;
7066 bitmap = md_bitmap_create(mddev, -1);
7067 mddev_suspend(mddev);
7068 if (!IS_ERR(bitmap)) {
7069 mddev->bitmap = bitmap;
7070 err = md_bitmap_load(mddev);
7072 err = PTR_ERR(bitmap);
7074 md_bitmap_destroy(mddev);
7077 mddev_resume(mddev);
7078 } else if (fd < 0) {
7079 mddev_suspend(mddev);
7080 md_bitmap_destroy(mddev);
7081 mddev_resume(mddev);
7085 struct file *f = mddev->bitmap_info.file;
7087 spin_lock(&mddev->lock);
7088 mddev->bitmap_info.file = NULL;
7089 spin_unlock(&mddev->lock);
7098 * md_set_array_info is used two different ways
7099 * The original usage is when creating a new array.
7100 * In this usage, raid_disks is > 0 and it together with
7101 * level, size, not_persistent,layout,chunksize determine the
7102 * shape of the array.
7103 * This will always create an array with a type-0.90.0 superblock.
7104 * The newer usage is when assembling an array.
7105 * In this case raid_disks will be 0, and the major_version field is
7106 * use to determine which style super-blocks are to be found on the devices.
7107 * The minor and patch _version numbers are also kept incase the
7108 * super_block handler wishes to interpret them.
7110 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7112 if (info->raid_disks == 0) {
7113 /* just setting version number for superblock loading */
7114 if (info->major_version < 0 ||
7115 info->major_version >= ARRAY_SIZE(super_types) ||
7116 super_types[info->major_version].name == NULL) {
7117 /* maybe try to auto-load a module? */
7118 pr_warn("md: superblock version %d not known\n",
7119 info->major_version);
7122 mddev->major_version = info->major_version;
7123 mddev->minor_version = info->minor_version;
7124 mddev->patch_version = info->patch_version;
7125 mddev->persistent = !info->not_persistent;
7126 /* ensure mddev_put doesn't delete this now that there
7127 * is some minimal configuration.
7129 mddev->ctime = ktime_get_real_seconds();
7132 mddev->major_version = MD_MAJOR_VERSION;
7133 mddev->minor_version = MD_MINOR_VERSION;
7134 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7135 mddev->ctime = ktime_get_real_seconds();
7137 mddev->level = info->level;
7138 mddev->clevel[0] = 0;
7139 mddev->dev_sectors = 2 * (sector_t)info->size;
7140 mddev->raid_disks = info->raid_disks;
7141 /* don't set md_minor, it is determined by which /dev/md* was
7144 if (info->state & (1<<MD_SB_CLEAN))
7145 mddev->recovery_cp = MaxSector;
7147 mddev->recovery_cp = 0;
7148 mddev->persistent = ! info->not_persistent;
7149 mddev->external = 0;
7151 mddev->layout = info->layout;
7152 if (mddev->level == 0)
7153 /* Cannot trust RAID0 layout info here */
7155 mddev->chunk_sectors = info->chunk_size >> 9;
7157 if (mddev->persistent) {
7158 mddev->max_disks = MD_SB_DISKS;
7160 mddev->sb_flags = 0;
7162 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7164 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7165 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7166 mddev->bitmap_info.offset = 0;
7168 mddev->reshape_position = MaxSector;
7171 * Generate a 128 bit UUID
7173 get_random_bytes(mddev->uuid, 16);
7175 mddev->new_level = mddev->level;
7176 mddev->new_chunk_sectors = mddev->chunk_sectors;
7177 mddev->new_layout = mddev->layout;
7178 mddev->delta_disks = 0;
7179 mddev->reshape_backwards = 0;
7184 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7186 lockdep_assert_held(&mddev->reconfig_mutex);
7188 if (mddev->external_size)
7191 mddev->array_sectors = array_sectors;
7193 EXPORT_SYMBOL(md_set_array_sectors);
7195 static int update_size(struct mddev *mddev, sector_t num_sectors)
7197 struct md_rdev *rdev;
7199 int fit = (num_sectors == 0);
7200 sector_t old_dev_sectors = mddev->dev_sectors;
7202 if (mddev->pers->resize == NULL)
7204 /* The "num_sectors" is the number of sectors of each device that
7205 * is used. This can only make sense for arrays with redundancy.
7206 * linear and raid0 always use whatever space is available. We can only
7207 * consider changing this number if no resync or reconstruction is
7208 * happening, and if the new size is acceptable. It must fit before the
7209 * sb_start or, if that is <data_offset, it must fit before the size
7210 * of each device. If num_sectors is zero, we find the largest size
7213 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7219 rdev_for_each(rdev, mddev) {
7220 sector_t avail = rdev->sectors;
7222 if (fit && (num_sectors == 0 || num_sectors > avail))
7223 num_sectors = avail;
7224 if (avail < num_sectors)
7227 rv = mddev->pers->resize(mddev, num_sectors);
7229 if (mddev_is_clustered(mddev))
7230 md_cluster_ops->update_size(mddev, old_dev_sectors);
7231 else if (mddev->queue) {
7232 set_capacity_and_notify(mddev->gendisk,
7233 mddev->array_sectors);
7239 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7242 struct md_rdev *rdev;
7243 /* change the number of raid disks */
7244 if (mddev->pers->check_reshape == NULL)
7248 if (raid_disks <= 0 ||
7249 (mddev->max_disks && raid_disks >= mddev->max_disks))
7251 if (mddev->sync_thread ||
7252 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7253 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7254 mddev->reshape_position != MaxSector)
7257 rdev_for_each(rdev, mddev) {
7258 if (mddev->raid_disks < raid_disks &&
7259 rdev->data_offset < rdev->new_data_offset)
7261 if (mddev->raid_disks > raid_disks &&
7262 rdev->data_offset > rdev->new_data_offset)
7266 mddev->delta_disks = raid_disks - mddev->raid_disks;
7267 if (mddev->delta_disks < 0)
7268 mddev->reshape_backwards = 1;
7269 else if (mddev->delta_disks > 0)
7270 mddev->reshape_backwards = 0;
7272 rv = mddev->pers->check_reshape(mddev);
7274 mddev->delta_disks = 0;
7275 mddev->reshape_backwards = 0;
7281 * update_array_info is used to change the configuration of an
7283 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7284 * fields in the info are checked against the array.
7285 * Any differences that cannot be handled will cause an error.
7286 * Normally, only one change can be managed at a time.
7288 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7294 /* calculate expected state,ignoring low bits */
7295 if (mddev->bitmap && mddev->bitmap_info.offset)
7296 state |= (1 << MD_SB_BITMAP_PRESENT);
7298 if (mddev->major_version != info->major_version ||
7299 mddev->minor_version != info->minor_version ||
7300 /* mddev->patch_version != info->patch_version || */
7301 mddev->ctime != info->ctime ||
7302 mddev->level != info->level ||
7303 /* mddev->layout != info->layout || */
7304 mddev->persistent != !info->not_persistent ||
7305 mddev->chunk_sectors != info->chunk_size >> 9 ||
7306 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7307 ((state^info->state) & 0xfffffe00)
7310 /* Check there is only one change */
7311 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7313 if (mddev->raid_disks != info->raid_disks)
7315 if (mddev->layout != info->layout)
7317 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7324 if (mddev->layout != info->layout) {
7326 * we don't need to do anything at the md level, the
7327 * personality will take care of it all.
7329 if (mddev->pers->check_reshape == NULL)
7332 mddev->new_layout = info->layout;
7333 rv = mddev->pers->check_reshape(mddev);
7335 mddev->new_layout = mddev->layout;
7339 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7340 rv = update_size(mddev, (sector_t)info->size * 2);
7342 if (mddev->raid_disks != info->raid_disks)
7343 rv = update_raid_disks(mddev, info->raid_disks);
7345 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7346 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7350 if (mddev->recovery || mddev->sync_thread) {
7354 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7355 struct bitmap *bitmap;
7356 /* add the bitmap */
7357 if (mddev->bitmap) {
7361 if (mddev->bitmap_info.default_offset == 0) {
7365 mddev->bitmap_info.offset =
7366 mddev->bitmap_info.default_offset;
7367 mddev->bitmap_info.space =
7368 mddev->bitmap_info.default_space;
7369 bitmap = md_bitmap_create(mddev, -1);
7370 mddev_suspend(mddev);
7371 if (!IS_ERR(bitmap)) {
7372 mddev->bitmap = bitmap;
7373 rv = md_bitmap_load(mddev);
7375 rv = PTR_ERR(bitmap);
7377 md_bitmap_destroy(mddev);
7378 mddev_resume(mddev);
7380 /* remove the bitmap */
7381 if (!mddev->bitmap) {
7385 if (mddev->bitmap->storage.file) {
7389 if (mddev->bitmap_info.nodes) {
7390 /* hold PW on all the bitmap lock */
7391 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7392 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7394 md_cluster_ops->unlock_all_bitmaps(mddev);
7398 mddev->bitmap_info.nodes = 0;
7399 md_cluster_ops->leave(mddev);
7400 module_put(md_cluster_mod);
7401 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7403 mddev_suspend(mddev);
7404 md_bitmap_destroy(mddev);
7405 mddev_resume(mddev);
7406 mddev->bitmap_info.offset = 0;
7409 md_update_sb(mddev, 1);
7415 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7417 struct md_rdev *rdev;
7420 if (mddev->pers == NULL)
7424 rdev = md_find_rdev_rcu(mddev, dev);
7428 md_error(mddev, rdev);
7429 if (!test_bit(Faulty, &rdev->flags))
7437 * We have a problem here : there is no easy way to give a CHS
7438 * virtual geometry. We currently pretend that we have a 2 heads
7439 * 4 sectors (with a BIG number of cylinders...). This drives
7440 * dosfs just mad... ;-)
7442 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7444 struct mddev *mddev = bdev->bd_disk->private_data;
7448 geo->cylinders = mddev->array_sectors / 8;
7452 static inline bool md_ioctl_valid(unsigned int cmd)
7456 case GET_ARRAY_INFO:
7457 case GET_BITMAP_FILE:
7460 case HOT_REMOVE_DISK:
7462 case RESTART_ARRAY_RW:
7464 case SET_ARRAY_INFO:
7465 case SET_BITMAP_FILE:
7466 case SET_DISK_FAULTY:
7469 case CLUSTERED_DISK_NACK:
7476 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7477 unsigned int cmd, unsigned long arg)
7480 void __user *argp = (void __user *)arg;
7481 struct mddev *mddev = NULL;
7482 bool did_set_md_closing = false;
7484 if (!md_ioctl_valid(cmd))
7489 case GET_ARRAY_INFO:
7493 if (!capable(CAP_SYS_ADMIN))
7498 * Commands dealing with the RAID driver but not any
7503 err = get_version(argp);
7509 * Commands creating/starting a new array:
7512 mddev = bdev->bd_disk->private_data;
7519 /* Some actions do not requires the mutex */
7521 case GET_ARRAY_INFO:
7522 if (!mddev->raid_disks && !mddev->external)
7525 err = get_array_info(mddev, argp);
7529 if (!mddev->raid_disks && !mddev->external)
7532 err = get_disk_info(mddev, argp);
7535 case SET_DISK_FAULTY:
7536 err = set_disk_faulty(mddev, new_decode_dev(arg));
7539 case GET_BITMAP_FILE:
7540 err = get_bitmap_file(mddev, argp);
7545 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7546 flush_rdev_wq(mddev);
7548 if (cmd == HOT_REMOVE_DISK)
7549 /* need to ensure recovery thread has run */
7550 wait_event_interruptible_timeout(mddev->sb_wait,
7551 !test_bit(MD_RECOVERY_NEEDED,
7553 msecs_to_jiffies(5000));
7554 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7555 /* Need to flush page cache, and ensure no-one else opens
7558 mutex_lock(&mddev->open_mutex);
7559 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7560 mutex_unlock(&mddev->open_mutex);
7564 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7565 mutex_unlock(&mddev->open_mutex);
7569 did_set_md_closing = true;
7570 mutex_unlock(&mddev->open_mutex);
7571 sync_blockdev(bdev);
7573 err = mddev_lock(mddev);
7575 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7580 if (cmd == SET_ARRAY_INFO) {
7581 mdu_array_info_t info;
7583 memset(&info, 0, sizeof(info));
7584 else if (copy_from_user(&info, argp, sizeof(info))) {
7589 err = update_array_info(mddev, &info);
7591 pr_warn("md: couldn't update array info. %d\n", err);
7596 if (!list_empty(&mddev->disks)) {
7597 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7601 if (mddev->raid_disks) {
7602 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7606 err = md_set_array_info(mddev, &info);
7608 pr_warn("md: couldn't set array info. %d\n", err);
7615 * Commands querying/configuring an existing array:
7617 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7618 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7619 if ((!mddev->raid_disks && !mddev->external)
7620 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7621 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7622 && cmd != GET_BITMAP_FILE) {
7628 * Commands even a read-only array can execute:
7631 case RESTART_ARRAY_RW:
7632 err = restart_array(mddev);
7636 err = do_md_stop(mddev, 0, bdev);
7640 err = md_set_readonly(mddev, bdev);
7643 case HOT_REMOVE_DISK:
7644 err = hot_remove_disk(mddev, new_decode_dev(arg));
7648 /* We can support ADD_NEW_DISK on read-only arrays
7649 * only if we are re-adding a preexisting device.
7650 * So require mddev->pers and MD_DISK_SYNC.
7653 mdu_disk_info_t info;
7654 if (copy_from_user(&info, argp, sizeof(info)))
7656 else if (!(info.state & (1<<MD_DISK_SYNC)))
7657 /* Need to clear read-only for this */
7660 err = md_add_new_disk(mddev, &info);
7667 * The remaining ioctls are changing the state of the
7668 * superblock, so we do not allow them on read-only arrays.
7670 if (mddev->ro && mddev->pers) {
7671 if (mddev->ro == 2) {
7673 sysfs_notify_dirent_safe(mddev->sysfs_state);
7674 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7675 /* mddev_unlock will wake thread */
7676 /* If a device failed while we were read-only, we
7677 * need to make sure the metadata is updated now.
7679 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7680 mddev_unlock(mddev);
7681 wait_event(mddev->sb_wait,
7682 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7683 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7684 mddev_lock_nointr(mddev);
7695 mdu_disk_info_t info;
7696 if (copy_from_user(&info, argp, sizeof(info)))
7699 err = md_add_new_disk(mddev, &info);
7703 case CLUSTERED_DISK_NACK:
7704 if (mddev_is_clustered(mddev))
7705 md_cluster_ops->new_disk_ack(mddev, false);
7711 err = hot_add_disk(mddev, new_decode_dev(arg));
7715 err = do_md_run(mddev);
7718 case SET_BITMAP_FILE:
7719 err = set_bitmap_file(mddev, (int)arg);
7728 if (mddev->hold_active == UNTIL_IOCTL &&
7730 mddev->hold_active = 0;
7731 mddev_unlock(mddev);
7733 if(did_set_md_closing)
7734 clear_bit(MD_CLOSING, &mddev->flags);
7737 #ifdef CONFIG_COMPAT
7738 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7739 unsigned int cmd, unsigned long arg)
7742 case HOT_REMOVE_DISK:
7744 case SET_DISK_FAULTY:
7745 case SET_BITMAP_FILE:
7746 /* These take in integer arg, do not convert */
7749 arg = (unsigned long)compat_ptr(arg);
7753 return md_ioctl(bdev, mode, cmd, arg);
7755 #endif /* CONFIG_COMPAT */
7757 static int md_set_read_only(struct block_device *bdev, bool ro)
7759 struct mddev *mddev = bdev->bd_disk->private_data;
7762 err = mddev_lock(mddev);
7766 if (!mddev->raid_disks && !mddev->external) {
7772 * Transitioning to read-auto need only happen for arrays that call
7773 * md_write_start and which are not ready for writes yet.
7775 if (!ro && mddev->ro == 1 && mddev->pers) {
7776 err = restart_array(mddev);
7783 mddev_unlock(mddev);
7787 static int md_open(struct block_device *bdev, fmode_t mode)
7790 * Succeed if we can lock the mddev, which confirms that
7791 * it isn't being stopped right now.
7793 struct mddev *mddev = mddev_find(bdev->bd_dev);
7799 if (mddev->gendisk != bdev->bd_disk) {
7800 /* we are racing with mddev_put which is discarding this
7804 /* Wait until bdev->bd_disk is definitely gone */
7805 if (work_pending(&mddev->del_work))
7806 flush_workqueue(md_misc_wq);
7809 BUG_ON(mddev != bdev->bd_disk->private_data);
7811 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7814 if (test_bit(MD_CLOSING, &mddev->flags)) {
7815 mutex_unlock(&mddev->open_mutex);
7821 atomic_inc(&mddev->openers);
7822 mutex_unlock(&mddev->open_mutex);
7824 bdev_check_media_change(bdev);
7831 static void md_release(struct gendisk *disk, fmode_t mode)
7833 struct mddev *mddev = disk->private_data;
7836 atomic_dec(&mddev->openers);
7840 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7842 struct mddev *mddev = disk->private_data;
7843 unsigned int ret = 0;
7846 ret = DISK_EVENT_MEDIA_CHANGE;
7851 const struct block_device_operations md_fops =
7853 .owner = THIS_MODULE,
7854 .submit_bio = md_submit_bio,
7856 .release = md_release,
7858 #ifdef CONFIG_COMPAT
7859 .compat_ioctl = md_compat_ioctl,
7861 .getgeo = md_getgeo,
7862 .check_events = md_check_events,
7863 .set_read_only = md_set_read_only,
7866 static int md_thread(void *arg)
7868 struct md_thread *thread = arg;
7871 * md_thread is a 'system-thread', it's priority should be very
7872 * high. We avoid resource deadlocks individually in each
7873 * raid personality. (RAID5 does preallocation) We also use RR and
7874 * the very same RT priority as kswapd, thus we will never get
7875 * into a priority inversion deadlock.
7877 * we definitely have to have equal or higher priority than
7878 * bdflush, otherwise bdflush will deadlock if there are too
7879 * many dirty RAID5 blocks.
7882 allow_signal(SIGKILL);
7883 while (!kthread_should_stop()) {
7885 /* We need to wait INTERRUPTIBLE so that
7886 * we don't add to the load-average.
7887 * That means we need to be sure no signals are
7890 if (signal_pending(current))
7891 flush_signals(current);
7893 wait_event_interruptible_timeout
7895 test_bit(THREAD_WAKEUP, &thread->flags)
7896 || kthread_should_stop() || kthread_should_park(),
7899 clear_bit(THREAD_WAKEUP, &thread->flags);
7900 if (kthread_should_park())
7902 if (!kthread_should_stop())
7903 thread->run(thread);
7909 void md_wakeup_thread(struct md_thread *thread)
7912 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7913 set_bit(THREAD_WAKEUP, &thread->flags);
7914 wake_up(&thread->wqueue);
7917 EXPORT_SYMBOL(md_wakeup_thread);
7919 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7920 struct mddev *mddev, const char *name)
7922 struct md_thread *thread;
7924 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7928 init_waitqueue_head(&thread->wqueue);
7931 thread->mddev = mddev;
7932 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7933 thread->tsk = kthread_run(md_thread, thread,
7935 mdname(thread->mddev),
7937 if (IS_ERR(thread->tsk)) {
7943 EXPORT_SYMBOL(md_register_thread);
7945 void md_unregister_thread(struct md_thread **threadp)
7947 struct md_thread *thread;
7950 * Locking ensures that mddev_unlock does not wake_up a
7951 * non-existent thread
7953 spin_lock(&pers_lock);
7956 spin_unlock(&pers_lock);
7960 spin_unlock(&pers_lock);
7962 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7963 kthread_stop(thread->tsk);
7966 EXPORT_SYMBOL(md_unregister_thread);
7968 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7970 if (!rdev || test_bit(Faulty, &rdev->flags))
7973 if (!mddev->pers || !mddev->pers->error_handler)
7975 mddev->pers->error_handler(mddev,rdev);
7976 if (mddev->degraded)
7977 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7978 sysfs_notify_dirent_safe(rdev->sysfs_state);
7979 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7980 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7981 md_wakeup_thread(mddev->thread);
7982 if (mddev->event_work.func)
7983 queue_work(md_misc_wq, &mddev->event_work);
7984 md_new_event(mddev);
7986 EXPORT_SYMBOL(md_error);
7988 /* seq_file implementation /proc/mdstat */
7990 static void status_unused(struct seq_file *seq)
7993 struct md_rdev *rdev;
7995 seq_printf(seq, "unused devices: ");
7997 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7998 char b[BDEVNAME_SIZE];
8000 seq_printf(seq, "%s ",
8001 bdevname(rdev->bdev,b));
8004 seq_printf(seq, "<none>");
8006 seq_printf(seq, "\n");
8009 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8011 sector_t max_sectors, resync, res;
8012 unsigned long dt, db = 0;
8013 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8014 int scale, recovery_active;
8015 unsigned int per_milli;
8017 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8018 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8019 max_sectors = mddev->resync_max_sectors;
8021 max_sectors = mddev->dev_sectors;
8023 resync = mddev->curr_resync;
8025 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8026 /* Still cleaning up */
8027 resync = max_sectors;
8028 } else if (resync > max_sectors)
8029 resync = max_sectors;
8031 resync -= atomic_read(&mddev->recovery_active);
8034 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8035 struct md_rdev *rdev;
8037 rdev_for_each(rdev, mddev)
8038 if (rdev->raid_disk >= 0 &&
8039 !test_bit(Faulty, &rdev->flags) &&
8040 rdev->recovery_offset != MaxSector &&
8041 rdev->recovery_offset) {
8042 seq_printf(seq, "\trecover=REMOTE");
8045 if (mddev->reshape_position != MaxSector)
8046 seq_printf(seq, "\treshape=REMOTE");
8048 seq_printf(seq, "\tresync=REMOTE");
8051 if (mddev->recovery_cp < MaxSector) {
8052 seq_printf(seq, "\tresync=PENDING");
8058 seq_printf(seq, "\tresync=DELAYED");
8062 WARN_ON(max_sectors == 0);
8063 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8064 * in a sector_t, and (max_sectors>>scale) will fit in a
8065 * u32, as those are the requirements for sector_div.
8066 * Thus 'scale' must be at least 10
8069 if (sizeof(sector_t) > sizeof(unsigned long)) {
8070 while ( max_sectors/2 > (1ULL<<(scale+32)))
8073 res = (resync>>scale)*1000;
8074 sector_div(res, (u32)((max_sectors>>scale)+1));
8078 int i, x = per_milli/50, y = 20-x;
8079 seq_printf(seq, "[");
8080 for (i = 0; i < x; i++)
8081 seq_printf(seq, "=");
8082 seq_printf(seq, ">");
8083 for (i = 0; i < y; i++)
8084 seq_printf(seq, ".");
8085 seq_printf(seq, "] ");
8087 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8088 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8090 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8092 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8093 "resync" : "recovery"))),
8094 per_milli/10, per_milli % 10,
8095 (unsigned long long) resync/2,
8096 (unsigned long long) max_sectors/2);
8099 * dt: time from mark until now
8100 * db: blocks written from mark until now
8101 * rt: remaining time
8103 * rt is a sector_t, which is always 64bit now. We are keeping
8104 * the original algorithm, but it is not really necessary.
8106 * Original algorithm:
8107 * So we divide before multiply in case it is 32bit and close
8109 * We scale the divisor (db) by 32 to avoid losing precision
8110 * near the end of resync when the number of remaining sectors
8112 * We then divide rt by 32 after multiplying by db to compensate.
8113 * The '+1' avoids division by zero if db is very small.
8115 dt = ((jiffies - mddev->resync_mark) / HZ);
8118 curr_mark_cnt = mddev->curr_mark_cnt;
8119 recovery_active = atomic_read(&mddev->recovery_active);
8120 resync_mark_cnt = mddev->resync_mark_cnt;
8122 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8123 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8125 rt = max_sectors - resync; /* number of remaining sectors */
8126 rt = div64_u64(rt, db/32+1);
8130 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8131 ((unsigned long)rt % 60)/6);
8133 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8137 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8139 struct list_head *tmp;
8141 struct mddev *mddev;
8153 spin_lock(&all_mddevs_lock);
8154 list_for_each(tmp,&all_mddevs)
8156 mddev = list_entry(tmp, struct mddev, all_mddevs);
8158 spin_unlock(&all_mddevs_lock);
8161 spin_unlock(&all_mddevs_lock);
8163 return (void*)2;/* tail */
8167 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8169 struct list_head *tmp;
8170 struct mddev *next_mddev, *mddev = v;
8176 spin_lock(&all_mddevs_lock);
8178 tmp = all_mddevs.next;
8180 tmp = mddev->all_mddevs.next;
8181 if (tmp != &all_mddevs)
8182 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8184 next_mddev = (void*)2;
8187 spin_unlock(&all_mddevs_lock);
8195 static void md_seq_stop(struct seq_file *seq, void *v)
8197 struct mddev *mddev = v;
8199 if (mddev && v != (void*)1 && v != (void*)2)
8203 static int md_seq_show(struct seq_file *seq, void *v)
8205 struct mddev *mddev = v;
8207 struct md_rdev *rdev;
8209 if (v == (void*)1) {
8210 struct md_personality *pers;
8211 seq_printf(seq, "Personalities : ");
8212 spin_lock(&pers_lock);
8213 list_for_each_entry(pers, &pers_list, list)
8214 seq_printf(seq, "[%s] ", pers->name);
8216 spin_unlock(&pers_lock);
8217 seq_printf(seq, "\n");
8218 seq->poll_event = atomic_read(&md_event_count);
8221 if (v == (void*)2) {
8226 spin_lock(&mddev->lock);
8227 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8228 seq_printf(seq, "%s : %sactive", mdname(mddev),
8229 mddev->pers ? "" : "in");
8232 seq_printf(seq, " (read-only)");
8234 seq_printf(seq, " (auto-read-only)");
8235 seq_printf(seq, " %s", mddev->pers->name);
8240 rdev_for_each_rcu(rdev, mddev) {
8241 char b[BDEVNAME_SIZE];
8242 seq_printf(seq, " %s[%d]",
8243 bdevname(rdev->bdev,b), rdev->desc_nr);
8244 if (test_bit(WriteMostly, &rdev->flags))
8245 seq_printf(seq, "(W)");
8246 if (test_bit(Journal, &rdev->flags))
8247 seq_printf(seq, "(J)");
8248 if (test_bit(Faulty, &rdev->flags)) {
8249 seq_printf(seq, "(F)");
8252 if (rdev->raid_disk < 0)
8253 seq_printf(seq, "(S)"); /* spare */
8254 if (test_bit(Replacement, &rdev->flags))
8255 seq_printf(seq, "(R)");
8256 sectors += rdev->sectors;
8260 if (!list_empty(&mddev->disks)) {
8262 seq_printf(seq, "\n %llu blocks",
8263 (unsigned long long)
8264 mddev->array_sectors / 2);
8266 seq_printf(seq, "\n %llu blocks",
8267 (unsigned long long)sectors / 2);
8269 if (mddev->persistent) {
8270 if (mddev->major_version != 0 ||
8271 mddev->minor_version != 90) {
8272 seq_printf(seq," super %d.%d",
8273 mddev->major_version,
8274 mddev->minor_version);
8276 } else if (mddev->external)
8277 seq_printf(seq, " super external:%s",
8278 mddev->metadata_type);
8280 seq_printf(seq, " super non-persistent");
8283 mddev->pers->status(seq, mddev);
8284 seq_printf(seq, "\n ");
8285 if (mddev->pers->sync_request) {
8286 if (status_resync(seq, mddev))
8287 seq_printf(seq, "\n ");
8290 seq_printf(seq, "\n ");
8292 md_bitmap_status(seq, mddev->bitmap);
8294 seq_printf(seq, "\n");
8296 spin_unlock(&mddev->lock);
8301 static const struct seq_operations md_seq_ops = {
8302 .start = md_seq_start,
8303 .next = md_seq_next,
8304 .stop = md_seq_stop,
8305 .show = md_seq_show,
8308 static int md_seq_open(struct inode *inode, struct file *file)
8310 struct seq_file *seq;
8313 error = seq_open(file, &md_seq_ops);
8317 seq = file->private_data;
8318 seq->poll_event = atomic_read(&md_event_count);
8322 static int md_unloading;
8323 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8325 struct seq_file *seq = filp->private_data;
8329 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8330 poll_wait(filp, &md_event_waiters, wait);
8332 /* always allow read */
8333 mask = EPOLLIN | EPOLLRDNORM;
8335 if (seq->poll_event != atomic_read(&md_event_count))
8336 mask |= EPOLLERR | EPOLLPRI;
8340 static const struct proc_ops mdstat_proc_ops = {
8341 .proc_open = md_seq_open,
8342 .proc_read = seq_read,
8343 .proc_lseek = seq_lseek,
8344 .proc_release = seq_release,
8345 .proc_poll = mdstat_poll,
8348 int register_md_personality(struct md_personality *p)
8350 pr_debug("md: %s personality registered for level %d\n",
8352 spin_lock(&pers_lock);
8353 list_add_tail(&p->list, &pers_list);
8354 spin_unlock(&pers_lock);
8357 EXPORT_SYMBOL(register_md_personality);
8359 int unregister_md_personality(struct md_personality *p)
8361 pr_debug("md: %s personality unregistered\n", p->name);
8362 spin_lock(&pers_lock);
8363 list_del_init(&p->list);
8364 spin_unlock(&pers_lock);
8367 EXPORT_SYMBOL(unregister_md_personality);
8369 int register_md_cluster_operations(struct md_cluster_operations *ops,
8370 struct module *module)
8373 spin_lock(&pers_lock);
8374 if (md_cluster_ops != NULL)
8377 md_cluster_ops = ops;
8378 md_cluster_mod = module;
8380 spin_unlock(&pers_lock);
8383 EXPORT_SYMBOL(register_md_cluster_operations);
8385 int unregister_md_cluster_operations(void)
8387 spin_lock(&pers_lock);
8388 md_cluster_ops = NULL;
8389 spin_unlock(&pers_lock);
8392 EXPORT_SYMBOL(unregister_md_cluster_operations);
8394 int md_setup_cluster(struct mddev *mddev, int nodes)
8397 if (!md_cluster_ops)
8398 request_module("md-cluster");
8399 spin_lock(&pers_lock);
8400 /* ensure module won't be unloaded */
8401 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8402 pr_warn("can't find md-cluster module or get it's reference.\n");
8403 spin_unlock(&pers_lock);
8406 spin_unlock(&pers_lock);
8408 ret = md_cluster_ops->join(mddev, nodes);
8410 mddev->safemode_delay = 0;
8414 void md_cluster_stop(struct mddev *mddev)
8416 if (!md_cluster_ops)
8418 md_cluster_ops->leave(mddev);
8419 module_put(md_cluster_mod);
8422 static int is_mddev_idle(struct mddev *mddev, int init)
8424 struct md_rdev *rdev;
8430 rdev_for_each_rcu(rdev, mddev) {
8431 struct gendisk *disk = rdev->bdev->bd_disk;
8432 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8433 atomic_read(&disk->sync_io);
8434 /* sync IO will cause sync_io to increase before the disk_stats
8435 * as sync_io is counted when a request starts, and
8436 * disk_stats is counted when it completes.
8437 * So resync activity will cause curr_events to be smaller than
8438 * when there was no such activity.
8439 * non-sync IO will cause disk_stat to increase without
8440 * increasing sync_io so curr_events will (eventually)
8441 * be larger than it was before. Once it becomes
8442 * substantially larger, the test below will cause
8443 * the array to appear non-idle, and resync will slow
8445 * If there is a lot of outstanding resync activity when
8446 * we set last_event to curr_events, then all that activity
8447 * completing might cause the array to appear non-idle
8448 * and resync will be slowed down even though there might
8449 * not have been non-resync activity. This will only
8450 * happen once though. 'last_events' will soon reflect
8451 * the state where there is little or no outstanding
8452 * resync requests, and further resync activity will
8453 * always make curr_events less than last_events.
8456 if (init || curr_events - rdev->last_events > 64) {
8457 rdev->last_events = curr_events;
8465 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8467 /* another "blocks" (512byte) blocks have been synced */
8468 atomic_sub(blocks, &mddev->recovery_active);
8469 wake_up(&mddev->recovery_wait);
8471 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8472 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8473 md_wakeup_thread(mddev->thread);
8474 // stop recovery, signal do_sync ....
8477 EXPORT_SYMBOL(md_done_sync);
8479 /* md_write_start(mddev, bi)
8480 * If we need to update some array metadata (e.g. 'active' flag
8481 * in superblock) before writing, schedule a superblock update
8482 * and wait for it to complete.
8483 * A return value of 'false' means that the write wasn't recorded
8484 * and cannot proceed as the array is being suspend.
8486 bool md_write_start(struct mddev *mddev, struct bio *bi)
8490 if (bio_data_dir(bi) != WRITE)
8493 BUG_ON(mddev->ro == 1);
8494 if (mddev->ro == 2) {
8495 /* need to switch to read/write */
8497 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8498 md_wakeup_thread(mddev->thread);
8499 md_wakeup_thread(mddev->sync_thread);
8503 percpu_ref_get(&mddev->writes_pending);
8504 smp_mb(); /* Match smp_mb in set_in_sync() */
8505 if (mddev->safemode == 1)
8506 mddev->safemode = 0;
8507 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8508 if (mddev->in_sync || mddev->sync_checkers) {
8509 spin_lock(&mddev->lock);
8510 if (mddev->in_sync) {
8512 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8513 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8514 md_wakeup_thread(mddev->thread);
8517 spin_unlock(&mddev->lock);
8521 sysfs_notify_dirent_safe(mddev->sysfs_state);
8522 if (!mddev->has_superblocks)
8524 wait_event(mddev->sb_wait,
8525 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8527 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8528 percpu_ref_put(&mddev->writes_pending);
8533 EXPORT_SYMBOL(md_write_start);
8535 /* md_write_inc can only be called when md_write_start() has
8536 * already been called at least once of the current request.
8537 * It increments the counter and is useful when a single request
8538 * is split into several parts. Each part causes an increment and
8539 * so needs a matching md_write_end().
8540 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8541 * a spinlocked region.
8543 void md_write_inc(struct mddev *mddev, struct bio *bi)
8545 if (bio_data_dir(bi) != WRITE)
8547 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8548 percpu_ref_get(&mddev->writes_pending);
8550 EXPORT_SYMBOL(md_write_inc);
8552 void md_write_end(struct mddev *mddev)
8554 percpu_ref_put(&mddev->writes_pending);
8556 if (mddev->safemode == 2)
8557 md_wakeup_thread(mddev->thread);
8558 else if (mddev->safemode_delay)
8559 /* The roundup() ensures this only performs locking once
8560 * every ->safemode_delay jiffies
8562 mod_timer(&mddev->safemode_timer,
8563 roundup(jiffies, mddev->safemode_delay) +
8564 mddev->safemode_delay);
8567 EXPORT_SYMBOL(md_write_end);
8569 /* This is used by raid0 and raid10 */
8570 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8571 struct bio *bio, sector_t start, sector_t size)
8573 struct bio *discard_bio = NULL;
8575 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO, 0,
8576 &discard_bio) || !discard_bio)
8579 bio_chain(discard_bio, bio);
8580 bio_clone_blkg_association(discard_bio, bio);
8582 trace_block_bio_remap(discard_bio,
8583 disk_devt(mddev->gendisk),
8584 bio->bi_iter.bi_sector);
8585 submit_bio_noacct(discard_bio);
8587 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8589 int acct_bioset_init(struct mddev *mddev)
8593 if (!bioset_initialized(&mddev->io_acct_set))
8594 err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
8595 offsetof(struct md_io_acct, bio_clone), 0);
8598 EXPORT_SYMBOL_GPL(acct_bioset_init);
8600 void acct_bioset_exit(struct mddev *mddev)
8602 bioset_exit(&mddev->io_acct_set);
8604 EXPORT_SYMBOL_GPL(acct_bioset_exit);
8606 static void md_end_io_acct(struct bio *bio)
8608 struct md_io_acct *md_io_acct = bio->bi_private;
8609 struct bio *orig_bio = md_io_acct->orig_bio;
8611 orig_bio->bi_status = bio->bi_status;
8613 bio_end_io_acct(orig_bio, md_io_acct->start_time);
8615 bio_endio(orig_bio);
8619 * Used by personalities that don't already clone the bio and thus can't
8620 * easily add the timestamp to their extended bio structure.
8622 void md_account_bio(struct mddev *mddev, struct bio **bio)
8624 struct md_io_acct *md_io_acct;
8627 if (!blk_queue_io_stat((*bio)->bi_bdev->bd_disk->queue))
8630 clone = bio_clone_fast(*bio, GFP_NOIO, &mddev->io_acct_set);
8631 md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8632 md_io_acct->orig_bio = *bio;
8633 md_io_acct->start_time = bio_start_io_acct(*bio);
8635 clone->bi_end_io = md_end_io_acct;
8636 clone->bi_private = md_io_acct;
8639 EXPORT_SYMBOL_GPL(md_account_bio);
8641 /* md_allow_write(mddev)
8642 * Calling this ensures that the array is marked 'active' so that writes
8643 * may proceed without blocking. It is important to call this before
8644 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8645 * Must be called with mddev_lock held.
8647 void md_allow_write(struct mddev *mddev)
8653 if (!mddev->pers->sync_request)
8656 spin_lock(&mddev->lock);
8657 if (mddev->in_sync) {
8659 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8660 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8661 if (mddev->safemode_delay &&
8662 mddev->safemode == 0)
8663 mddev->safemode = 1;
8664 spin_unlock(&mddev->lock);
8665 md_update_sb(mddev, 0);
8666 sysfs_notify_dirent_safe(mddev->sysfs_state);
8667 /* wait for the dirty state to be recorded in the metadata */
8668 wait_event(mddev->sb_wait,
8669 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8671 spin_unlock(&mddev->lock);
8673 EXPORT_SYMBOL_GPL(md_allow_write);
8675 #define SYNC_MARKS 10
8676 #define SYNC_MARK_STEP (3*HZ)
8677 #define UPDATE_FREQUENCY (5*60*HZ)
8678 void md_do_sync(struct md_thread *thread)
8680 struct mddev *mddev = thread->mddev;
8681 struct mddev *mddev2;
8682 unsigned int currspeed = 0, window;
8683 sector_t max_sectors,j, io_sectors, recovery_done;
8684 unsigned long mark[SYNC_MARKS];
8685 unsigned long update_time;
8686 sector_t mark_cnt[SYNC_MARKS];
8688 struct list_head *tmp;
8689 sector_t last_check;
8691 struct md_rdev *rdev;
8692 char *desc, *action = NULL;
8693 struct blk_plug plug;
8696 /* just incase thread restarts... */
8697 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8698 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8700 if (mddev->ro) {/* never try to sync a read-only array */
8701 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8705 if (mddev_is_clustered(mddev)) {
8706 ret = md_cluster_ops->resync_start(mddev);
8710 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8711 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8712 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8713 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8714 && ((unsigned long long)mddev->curr_resync_completed
8715 < (unsigned long long)mddev->resync_max_sectors))
8719 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8720 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8721 desc = "data-check";
8723 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8724 desc = "requested-resync";
8728 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8733 mddev->last_sync_action = action ?: desc;
8735 /* we overload curr_resync somewhat here.
8736 * 0 == not engaged in resync at all
8737 * 2 == checking that there is no conflict with another sync
8738 * 1 == like 2, but have yielded to allow conflicting resync to
8740 * other == active in resync - this many blocks
8742 * Before starting a resync we must have set curr_resync to
8743 * 2, and then checked that every "conflicting" array has curr_resync
8744 * less than ours. When we find one that is the same or higher
8745 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8746 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8747 * This will mean we have to start checking from the beginning again.
8752 int mddev2_minor = -1;
8753 mddev->curr_resync = 2;
8756 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8758 for_each_mddev(mddev2, tmp) {
8759 if (mddev2 == mddev)
8761 if (!mddev->parallel_resync
8762 && mddev2->curr_resync
8763 && match_mddev_units(mddev, mddev2)) {
8765 if (mddev < mddev2 && mddev->curr_resync == 2) {
8766 /* arbitrarily yield */
8767 mddev->curr_resync = 1;
8768 wake_up(&resync_wait);
8770 if (mddev > mddev2 && mddev->curr_resync == 1)
8771 /* no need to wait here, we can wait the next
8772 * time 'round when curr_resync == 2
8775 /* We need to wait 'interruptible' so as not to
8776 * contribute to the load average, and not to
8777 * be caught by 'softlockup'
8779 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8780 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8781 mddev2->curr_resync >= mddev->curr_resync) {
8782 if (mddev2_minor != mddev2->md_minor) {
8783 mddev2_minor = mddev2->md_minor;
8784 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8785 desc, mdname(mddev),
8789 if (signal_pending(current))
8790 flush_signals(current);
8792 finish_wait(&resync_wait, &wq);
8795 finish_wait(&resync_wait, &wq);
8798 } while (mddev->curr_resync < 2);
8801 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8802 /* resync follows the size requested by the personality,
8803 * which defaults to physical size, but can be virtual size
8805 max_sectors = mddev->resync_max_sectors;
8806 atomic64_set(&mddev->resync_mismatches, 0);
8807 /* we don't use the checkpoint if there's a bitmap */
8808 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8809 j = mddev->resync_min;
8810 else if (!mddev->bitmap)
8811 j = mddev->recovery_cp;
8813 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8814 max_sectors = mddev->resync_max_sectors;
8816 * If the original node aborts reshaping then we continue the
8817 * reshaping, so set j again to avoid restart reshape from the
8820 if (mddev_is_clustered(mddev) &&
8821 mddev->reshape_position != MaxSector)
8822 j = mddev->reshape_position;
8824 /* recovery follows the physical size of devices */
8825 max_sectors = mddev->dev_sectors;
8828 rdev_for_each_rcu(rdev, mddev)
8829 if (rdev->raid_disk >= 0 &&
8830 !test_bit(Journal, &rdev->flags) &&
8831 !test_bit(Faulty, &rdev->flags) &&
8832 !test_bit(In_sync, &rdev->flags) &&
8833 rdev->recovery_offset < j)
8834 j = rdev->recovery_offset;
8837 /* If there is a bitmap, we need to make sure all
8838 * writes that started before we added a spare
8839 * complete before we start doing a recovery.
8840 * Otherwise the write might complete and (via
8841 * bitmap_endwrite) set a bit in the bitmap after the
8842 * recovery has checked that bit and skipped that
8845 if (mddev->bitmap) {
8846 mddev->pers->quiesce(mddev, 1);
8847 mddev->pers->quiesce(mddev, 0);
8851 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8852 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8853 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8854 speed_max(mddev), desc);
8856 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8859 for (m = 0; m < SYNC_MARKS; m++) {
8861 mark_cnt[m] = io_sectors;
8864 mddev->resync_mark = mark[last_mark];
8865 mddev->resync_mark_cnt = mark_cnt[last_mark];
8868 * Tune reconstruction:
8870 window = 32 * (PAGE_SIZE / 512);
8871 pr_debug("md: using %dk window, over a total of %lluk.\n",
8872 window/2, (unsigned long long)max_sectors/2);
8874 atomic_set(&mddev->recovery_active, 0);
8878 pr_debug("md: resuming %s of %s from checkpoint.\n",
8879 desc, mdname(mddev));
8880 mddev->curr_resync = j;
8882 mddev->curr_resync = 3; /* no longer delayed */
8883 mddev->curr_resync_completed = j;
8884 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8885 md_new_event(mddev);
8886 update_time = jiffies;
8888 blk_start_plug(&plug);
8889 while (j < max_sectors) {
8894 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8895 ((mddev->curr_resync > mddev->curr_resync_completed &&
8896 (mddev->curr_resync - mddev->curr_resync_completed)
8897 > (max_sectors >> 4)) ||
8898 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8899 (j - mddev->curr_resync_completed)*2
8900 >= mddev->resync_max - mddev->curr_resync_completed ||
8901 mddev->curr_resync_completed > mddev->resync_max
8903 /* time to update curr_resync_completed */
8904 wait_event(mddev->recovery_wait,
8905 atomic_read(&mddev->recovery_active) == 0);
8906 mddev->curr_resync_completed = j;
8907 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8908 j > mddev->recovery_cp)
8909 mddev->recovery_cp = j;
8910 update_time = jiffies;
8911 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8912 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8915 while (j >= mddev->resync_max &&
8916 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8917 /* As this condition is controlled by user-space,
8918 * we can block indefinitely, so use '_interruptible'
8919 * to avoid triggering warnings.
8921 flush_signals(current); /* just in case */
8922 wait_event_interruptible(mddev->recovery_wait,
8923 mddev->resync_max > j
8924 || test_bit(MD_RECOVERY_INTR,
8928 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8931 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8933 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8937 if (!skipped) { /* actual IO requested */
8938 io_sectors += sectors;
8939 atomic_add(sectors, &mddev->recovery_active);
8942 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8946 if (j > max_sectors)
8947 /* when skipping, extra large numbers can be returned. */
8950 mddev->curr_resync = j;
8951 mddev->curr_mark_cnt = io_sectors;
8952 if (last_check == 0)
8953 /* this is the earliest that rebuild will be
8954 * visible in /proc/mdstat
8956 md_new_event(mddev);
8958 if (last_check + window > io_sectors || j == max_sectors)
8961 last_check = io_sectors;
8963 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8965 int next = (last_mark+1) % SYNC_MARKS;
8967 mddev->resync_mark = mark[next];
8968 mddev->resync_mark_cnt = mark_cnt[next];
8969 mark[next] = jiffies;
8970 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8974 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8978 * this loop exits only if either when we are slower than
8979 * the 'hard' speed limit, or the system was IO-idle for
8981 * the system might be non-idle CPU-wise, but we only care
8982 * about not overloading the IO subsystem. (things like an
8983 * e2fsck being done on the RAID array should execute fast)
8987 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8988 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8989 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8991 if (currspeed > speed_min(mddev)) {
8992 if (currspeed > speed_max(mddev)) {
8996 if (!is_mddev_idle(mddev, 0)) {
8998 * Give other IO more of a chance.
8999 * The faster the devices, the less we wait.
9001 wait_event(mddev->recovery_wait,
9002 !atomic_read(&mddev->recovery_active));
9006 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9007 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9008 ? "interrupted" : "done");
9010 * this also signals 'finished resyncing' to md_stop
9012 blk_finish_plug(&plug);
9013 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9015 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9016 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9017 mddev->curr_resync > 3) {
9018 mddev->curr_resync_completed = mddev->curr_resync;
9019 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9021 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9023 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9024 mddev->curr_resync > 3) {
9025 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9026 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9027 if (mddev->curr_resync >= mddev->recovery_cp) {
9028 pr_debug("md: checkpointing %s of %s.\n",
9029 desc, mdname(mddev));
9030 if (test_bit(MD_RECOVERY_ERROR,
9032 mddev->recovery_cp =
9033 mddev->curr_resync_completed;
9035 mddev->recovery_cp =
9039 mddev->recovery_cp = MaxSector;
9041 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9042 mddev->curr_resync = MaxSector;
9043 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9044 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9046 rdev_for_each_rcu(rdev, mddev)
9047 if (rdev->raid_disk >= 0 &&
9048 mddev->delta_disks >= 0 &&
9049 !test_bit(Journal, &rdev->flags) &&
9050 !test_bit(Faulty, &rdev->flags) &&
9051 !test_bit(In_sync, &rdev->flags) &&
9052 rdev->recovery_offset < mddev->curr_resync)
9053 rdev->recovery_offset = mddev->curr_resync;
9059 /* set CHANGE_PENDING here since maybe another update is needed,
9060 * so other nodes are informed. It should be harmless for normal
9062 set_mask_bits(&mddev->sb_flags, 0,
9063 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9065 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9066 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9067 mddev->delta_disks > 0 &&
9068 mddev->pers->finish_reshape &&
9069 mddev->pers->size &&
9071 mddev_lock_nointr(mddev);
9072 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9073 mddev_unlock(mddev);
9074 if (!mddev_is_clustered(mddev))
9075 set_capacity_and_notify(mddev->gendisk,
9076 mddev->array_sectors);
9079 spin_lock(&mddev->lock);
9080 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9081 /* We completed so min/max setting can be forgotten if used. */
9082 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9083 mddev->resync_min = 0;
9084 mddev->resync_max = MaxSector;
9085 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9086 mddev->resync_min = mddev->curr_resync_completed;
9087 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9088 mddev->curr_resync = 0;
9089 spin_unlock(&mddev->lock);
9091 wake_up(&resync_wait);
9092 md_wakeup_thread(mddev->thread);
9095 EXPORT_SYMBOL_GPL(md_do_sync);
9097 static int remove_and_add_spares(struct mddev *mddev,
9098 struct md_rdev *this)
9100 struct md_rdev *rdev;
9103 bool remove_some = false;
9105 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9106 /* Mustn't remove devices when resync thread is running */
9109 rdev_for_each(rdev, mddev) {
9110 if ((this == NULL || rdev == this) &&
9111 rdev->raid_disk >= 0 &&
9112 !test_bit(Blocked, &rdev->flags) &&
9113 test_bit(Faulty, &rdev->flags) &&
9114 atomic_read(&rdev->nr_pending)==0) {
9115 /* Faulty non-Blocked devices with nr_pending == 0
9116 * never get nr_pending incremented,
9117 * never get Faulty cleared, and never get Blocked set.
9118 * So we can synchronize_rcu now rather than once per device
9121 set_bit(RemoveSynchronized, &rdev->flags);
9127 rdev_for_each(rdev, mddev) {
9128 if ((this == NULL || rdev == this) &&
9129 rdev->raid_disk >= 0 &&
9130 !test_bit(Blocked, &rdev->flags) &&
9131 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9132 (!test_bit(In_sync, &rdev->flags) &&
9133 !test_bit(Journal, &rdev->flags))) &&
9134 atomic_read(&rdev->nr_pending)==0)) {
9135 if (mddev->pers->hot_remove_disk(
9136 mddev, rdev) == 0) {
9137 sysfs_unlink_rdev(mddev, rdev);
9138 rdev->saved_raid_disk = rdev->raid_disk;
9139 rdev->raid_disk = -1;
9143 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9144 clear_bit(RemoveSynchronized, &rdev->flags);
9147 if (removed && mddev->kobj.sd)
9148 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9150 if (this && removed)
9153 rdev_for_each(rdev, mddev) {
9154 if (this && this != rdev)
9156 if (test_bit(Candidate, &rdev->flags))
9158 if (rdev->raid_disk >= 0 &&
9159 !test_bit(In_sync, &rdev->flags) &&
9160 !test_bit(Journal, &rdev->flags) &&
9161 !test_bit(Faulty, &rdev->flags))
9163 if (rdev->raid_disk >= 0)
9165 if (test_bit(Faulty, &rdev->flags))
9167 if (!test_bit(Journal, &rdev->flags)) {
9169 ! (rdev->saved_raid_disk >= 0 &&
9170 !test_bit(Bitmap_sync, &rdev->flags)))
9173 rdev->recovery_offset = 0;
9175 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9176 /* failure here is OK */
9177 sysfs_link_rdev(mddev, rdev);
9178 if (!test_bit(Journal, &rdev->flags))
9180 md_new_event(mddev);
9181 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9186 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9190 static void md_start_sync(struct work_struct *ws)
9192 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9194 mddev->sync_thread = md_register_thread(md_do_sync,
9197 if (!mddev->sync_thread) {
9198 pr_warn("%s: could not start resync thread...\n",
9200 /* leave the spares where they are, it shouldn't hurt */
9201 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9202 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9203 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9204 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9205 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9206 wake_up(&resync_wait);
9207 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9209 if (mddev->sysfs_action)
9210 sysfs_notify_dirent_safe(mddev->sysfs_action);
9212 md_wakeup_thread(mddev->sync_thread);
9213 sysfs_notify_dirent_safe(mddev->sysfs_action);
9214 md_new_event(mddev);
9218 * This routine is regularly called by all per-raid-array threads to
9219 * deal with generic issues like resync and super-block update.
9220 * Raid personalities that don't have a thread (linear/raid0) do not
9221 * need this as they never do any recovery or update the superblock.
9223 * It does not do any resync itself, but rather "forks" off other threads
9224 * to do that as needed.
9225 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9226 * "->recovery" and create a thread at ->sync_thread.
9227 * When the thread finishes it sets MD_RECOVERY_DONE
9228 * and wakeups up this thread which will reap the thread and finish up.
9229 * This thread also removes any faulty devices (with nr_pending == 0).
9231 * The overall approach is:
9232 * 1/ if the superblock needs updating, update it.
9233 * 2/ If a recovery thread is running, don't do anything else.
9234 * 3/ If recovery has finished, clean up, possibly marking spares active.
9235 * 4/ If there are any faulty devices, remove them.
9236 * 5/ If array is degraded, try to add spares devices
9237 * 6/ If array has spares or is not in-sync, start a resync thread.
9239 void md_check_recovery(struct mddev *mddev)
9241 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9242 /* Write superblock - thread that called mddev_suspend()
9243 * holds reconfig_mutex for us.
9245 set_bit(MD_UPDATING_SB, &mddev->flags);
9246 smp_mb__after_atomic();
9247 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9248 md_update_sb(mddev, 0);
9249 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9250 wake_up(&mddev->sb_wait);
9253 if (mddev->suspended)
9257 md_bitmap_daemon_work(mddev);
9259 if (signal_pending(current)) {
9260 if (mddev->pers->sync_request && !mddev->external) {
9261 pr_debug("md: %s in immediate safe mode\n",
9263 mddev->safemode = 2;
9265 flush_signals(current);
9268 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9271 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9272 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9273 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9274 (mddev->external == 0 && mddev->safemode == 1) ||
9275 (mddev->safemode == 2
9276 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9280 if (mddev_trylock(mddev)) {
9282 bool try_set_sync = mddev->safemode != 0;
9284 if (!mddev->external && mddev->safemode == 1)
9285 mddev->safemode = 0;
9288 struct md_rdev *rdev;
9289 if (!mddev->external && mddev->in_sync)
9290 /* 'Blocked' flag not needed as failed devices
9291 * will be recorded if array switched to read/write.
9292 * Leaving it set will prevent the device
9293 * from being removed.
9295 rdev_for_each(rdev, mddev)
9296 clear_bit(Blocked, &rdev->flags);
9297 /* On a read-only array we can:
9298 * - remove failed devices
9299 * - add already-in_sync devices if the array itself
9301 * As we only add devices that are already in-sync,
9302 * we can activate the spares immediately.
9304 remove_and_add_spares(mddev, NULL);
9305 /* There is no thread, but we need to call
9306 * ->spare_active and clear saved_raid_disk
9308 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9309 md_reap_sync_thread(mddev);
9310 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9311 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9312 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9316 if (mddev_is_clustered(mddev)) {
9317 struct md_rdev *rdev, *tmp;
9318 /* kick the device if another node issued a
9321 rdev_for_each_safe(rdev, tmp, mddev) {
9322 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9323 rdev->raid_disk < 0)
9324 md_kick_rdev_from_array(rdev);
9328 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9329 spin_lock(&mddev->lock);
9331 spin_unlock(&mddev->lock);
9334 if (mddev->sb_flags)
9335 md_update_sb(mddev, 0);
9337 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9338 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9339 /* resync/recovery still happening */
9340 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9343 if (mddev->sync_thread) {
9344 md_reap_sync_thread(mddev);
9347 /* Set RUNNING before clearing NEEDED to avoid
9348 * any transients in the value of "sync_action".
9350 mddev->curr_resync_completed = 0;
9351 spin_lock(&mddev->lock);
9352 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9353 spin_unlock(&mddev->lock);
9354 /* Clear some bits that don't mean anything, but
9357 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9358 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9360 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9361 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9363 /* no recovery is running.
9364 * remove any failed drives, then
9365 * add spares if possible.
9366 * Spares are also removed and re-added, to allow
9367 * the personality to fail the re-add.
9370 if (mddev->reshape_position != MaxSector) {
9371 if (mddev->pers->check_reshape == NULL ||
9372 mddev->pers->check_reshape(mddev) != 0)
9373 /* Cannot proceed */
9375 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9376 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9377 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9378 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9379 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9380 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9381 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9382 } else if (mddev->recovery_cp < MaxSector) {
9383 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9384 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9385 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9386 /* nothing to be done ... */
9389 if (mddev->pers->sync_request) {
9391 /* We are adding a device or devices to an array
9392 * which has the bitmap stored on all devices.
9393 * So make sure all bitmap pages get written
9395 md_bitmap_write_all(mddev->bitmap);
9397 INIT_WORK(&mddev->del_work, md_start_sync);
9398 queue_work(md_misc_wq, &mddev->del_work);
9402 if (!mddev->sync_thread) {
9403 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9404 wake_up(&resync_wait);
9405 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9407 if (mddev->sysfs_action)
9408 sysfs_notify_dirent_safe(mddev->sysfs_action);
9411 wake_up(&mddev->sb_wait);
9412 mddev_unlock(mddev);
9415 EXPORT_SYMBOL(md_check_recovery);
9417 void md_reap_sync_thread(struct mddev *mddev)
9419 struct md_rdev *rdev;
9420 sector_t old_dev_sectors = mddev->dev_sectors;
9421 bool is_reshaped = false;
9423 /* resync has finished, collect result */
9424 md_unregister_thread(&mddev->sync_thread);
9425 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9426 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9427 mddev->degraded != mddev->raid_disks) {
9429 /* activate any spares */
9430 if (mddev->pers->spare_active(mddev)) {
9431 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9432 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9435 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9436 mddev->pers->finish_reshape) {
9437 mddev->pers->finish_reshape(mddev);
9438 if (mddev_is_clustered(mddev))
9442 /* If array is no-longer degraded, then any saved_raid_disk
9443 * information must be scrapped.
9445 if (!mddev->degraded)
9446 rdev_for_each(rdev, mddev)
9447 rdev->saved_raid_disk = -1;
9449 md_update_sb(mddev, 1);
9450 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9451 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9453 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9454 md_cluster_ops->resync_finish(mddev);
9455 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9456 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9457 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9458 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9459 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9460 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9462 * We call md_cluster_ops->update_size here because sync_size could
9463 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9464 * so it is time to update size across cluster.
9466 if (mddev_is_clustered(mddev) && is_reshaped
9467 && !test_bit(MD_CLOSING, &mddev->flags))
9468 md_cluster_ops->update_size(mddev, old_dev_sectors);
9469 wake_up(&resync_wait);
9470 /* flag recovery needed just to double check */
9471 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9472 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9473 sysfs_notify_dirent_safe(mddev->sysfs_action);
9474 md_new_event(mddev);
9475 if (mddev->event_work.func)
9476 queue_work(md_misc_wq, &mddev->event_work);
9478 EXPORT_SYMBOL(md_reap_sync_thread);
9480 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9482 sysfs_notify_dirent_safe(rdev->sysfs_state);
9483 wait_event_timeout(rdev->blocked_wait,
9484 !test_bit(Blocked, &rdev->flags) &&
9485 !test_bit(BlockedBadBlocks, &rdev->flags),
9486 msecs_to_jiffies(5000));
9487 rdev_dec_pending(rdev, mddev);
9489 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9491 void md_finish_reshape(struct mddev *mddev)
9493 /* called be personality module when reshape completes. */
9494 struct md_rdev *rdev;
9496 rdev_for_each(rdev, mddev) {
9497 if (rdev->data_offset > rdev->new_data_offset)
9498 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9500 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9501 rdev->data_offset = rdev->new_data_offset;
9504 EXPORT_SYMBOL(md_finish_reshape);
9506 /* Bad block management */
9508 /* Returns 1 on success, 0 on failure */
9509 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9512 struct mddev *mddev = rdev->mddev;
9515 s += rdev->new_data_offset;
9517 s += rdev->data_offset;
9518 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9520 /* Make sure they get written out promptly */
9521 if (test_bit(ExternalBbl, &rdev->flags))
9522 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9523 sysfs_notify_dirent_safe(rdev->sysfs_state);
9524 set_mask_bits(&mddev->sb_flags, 0,
9525 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9526 md_wakeup_thread(rdev->mddev->thread);
9531 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9533 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9538 s += rdev->new_data_offset;
9540 s += rdev->data_offset;
9541 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9542 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9543 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9546 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9548 static int md_notify_reboot(struct notifier_block *this,
9549 unsigned long code, void *x)
9551 struct list_head *tmp;
9552 struct mddev *mddev;
9555 for_each_mddev(mddev, tmp) {
9556 if (mddev_trylock(mddev)) {
9558 __md_stop_writes(mddev);
9559 if (mddev->persistent)
9560 mddev->safemode = 2;
9561 mddev_unlock(mddev);
9566 * certain more exotic SCSI devices are known to be
9567 * volatile wrt too early system reboots. While the
9568 * right place to handle this issue is the given
9569 * driver, we do want to have a safe RAID driver ...
9577 static struct notifier_block md_notifier = {
9578 .notifier_call = md_notify_reboot,
9580 .priority = INT_MAX, /* before any real devices */
9583 static void md_geninit(void)
9585 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9587 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9590 static int __init md_init(void)
9594 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9598 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9602 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9603 if (!md_rdev_misc_wq)
9604 goto err_rdev_misc_wq;
9606 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9610 ret = __register_blkdev(0, "mdp", md_probe);
9615 register_reboot_notifier(&md_notifier);
9616 raid_table_header = register_sysctl_table(raid_root_table);
9622 unregister_blkdev(MD_MAJOR, "md");
9624 destroy_workqueue(md_rdev_misc_wq);
9626 destroy_workqueue(md_misc_wq);
9628 destroy_workqueue(md_wq);
9633 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9635 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9636 struct md_rdev *rdev2, *tmp;
9638 char b[BDEVNAME_SIZE];
9641 * If size is changed in another node then we need to
9642 * do resize as well.
9644 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9645 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9647 pr_info("md-cluster: resize failed\n");
9649 md_bitmap_update_sb(mddev->bitmap);
9652 /* Check for change of roles in the active devices */
9653 rdev_for_each_safe(rdev2, tmp, mddev) {
9654 if (test_bit(Faulty, &rdev2->flags))
9657 /* Check if the roles changed */
9658 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9660 if (test_bit(Candidate, &rdev2->flags)) {
9661 if (role == 0xfffe) {
9662 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9663 md_kick_rdev_from_array(rdev2);
9667 clear_bit(Candidate, &rdev2->flags);
9670 if (role != rdev2->raid_disk) {
9672 * got activated except reshape is happening.
9674 if (rdev2->raid_disk == -1 && role != 0xffff &&
9675 !(le32_to_cpu(sb->feature_map) &
9676 MD_FEATURE_RESHAPE_ACTIVE)) {
9677 rdev2->saved_raid_disk = role;
9678 ret = remove_and_add_spares(mddev, rdev2);
9679 pr_info("Activated spare: %s\n",
9680 bdevname(rdev2->bdev,b));
9681 /* wakeup mddev->thread here, so array could
9682 * perform resync with the new activated disk */
9683 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9684 md_wakeup_thread(mddev->thread);
9687 * We just want to do the minimum to mark the disk
9688 * as faulty. The recovery is performed by the
9689 * one who initiated the error.
9691 if ((role == 0xfffe) || (role == 0xfffd)) {
9692 md_error(mddev, rdev2);
9693 clear_bit(Blocked, &rdev2->flags);
9698 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9699 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9701 pr_warn("md: updating array disks failed. %d\n", ret);
9705 * Since mddev->delta_disks has already updated in update_raid_disks,
9706 * so it is time to check reshape.
9708 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9709 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9711 * reshape is happening in the remote node, we need to
9712 * update reshape_position and call start_reshape.
9714 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9715 if (mddev->pers->update_reshape_pos)
9716 mddev->pers->update_reshape_pos(mddev);
9717 if (mddev->pers->start_reshape)
9718 mddev->pers->start_reshape(mddev);
9719 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9720 mddev->reshape_position != MaxSector &&
9721 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9722 /* reshape is just done in another node. */
9723 mddev->reshape_position = MaxSector;
9724 if (mddev->pers->update_reshape_pos)
9725 mddev->pers->update_reshape_pos(mddev);
9728 /* Finally set the event to be up to date */
9729 mddev->events = le64_to_cpu(sb->events);
9732 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9735 struct page *swapout = rdev->sb_page;
9736 struct mdp_superblock_1 *sb;
9738 /* Store the sb page of the rdev in the swapout temporary
9739 * variable in case we err in the future
9741 rdev->sb_page = NULL;
9742 err = alloc_disk_sb(rdev);
9744 ClearPageUptodate(rdev->sb_page);
9745 rdev->sb_loaded = 0;
9746 err = super_types[mddev->major_version].
9747 load_super(rdev, NULL, mddev->minor_version);
9750 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9751 __func__, __LINE__, rdev->desc_nr, err);
9753 put_page(rdev->sb_page);
9754 rdev->sb_page = swapout;
9755 rdev->sb_loaded = 1;
9759 sb = page_address(rdev->sb_page);
9760 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9764 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9765 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9767 /* The other node finished recovery, call spare_active to set
9768 * device In_sync and mddev->degraded
9770 if (rdev->recovery_offset == MaxSector &&
9771 !test_bit(In_sync, &rdev->flags) &&
9772 mddev->pers->spare_active(mddev))
9773 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9779 void md_reload_sb(struct mddev *mddev, int nr)
9781 struct md_rdev *rdev = NULL, *iter;
9785 rdev_for_each_rcu(iter, mddev) {
9786 if (iter->desc_nr == nr) {
9793 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9797 err = read_rdev(mddev, rdev);
9801 check_sb_changes(mddev, rdev);
9803 /* Read all rdev's to update recovery_offset */
9804 rdev_for_each_rcu(rdev, mddev) {
9805 if (!test_bit(Faulty, &rdev->flags))
9806 read_rdev(mddev, rdev);
9809 EXPORT_SYMBOL(md_reload_sb);
9814 * Searches all registered partitions for autorun RAID arrays
9818 static DEFINE_MUTEX(detected_devices_mutex);
9819 static LIST_HEAD(all_detected_devices);
9820 struct detected_devices_node {
9821 struct list_head list;
9825 void md_autodetect_dev(dev_t dev)
9827 struct detected_devices_node *node_detected_dev;
9829 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9830 if (node_detected_dev) {
9831 node_detected_dev->dev = dev;
9832 mutex_lock(&detected_devices_mutex);
9833 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9834 mutex_unlock(&detected_devices_mutex);
9838 void md_autostart_arrays(int part)
9840 struct md_rdev *rdev;
9841 struct detected_devices_node *node_detected_dev;
9843 int i_scanned, i_passed;
9848 pr_info("md: Autodetecting RAID arrays.\n");
9850 mutex_lock(&detected_devices_mutex);
9851 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9853 node_detected_dev = list_entry(all_detected_devices.next,
9854 struct detected_devices_node, list);
9855 list_del(&node_detected_dev->list);
9856 dev = node_detected_dev->dev;
9857 kfree(node_detected_dev);
9858 mutex_unlock(&detected_devices_mutex);
9859 rdev = md_import_device(dev,0, 90);
9860 mutex_lock(&detected_devices_mutex);
9864 if (test_bit(Faulty, &rdev->flags))
9867 set_bit(AutoDetected, &rdev->flags);
9868 list_add(&rdev->same_set, &pending_raid_disks);
9871 mutex_unlock(&detected_devices_mutex);
9873 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9875 autorun_devices(part);
9878 #endif /* !MODULE */
9880 static __exit void md_exit(void)
9882 struct mddev *mddev;
9883 struct list_head *tmp;
9886 unregister_blkdev(MD_MAJOR,"md");
9887 unregister_blkdev(mdp_major, "mdp");
9888 unregister_reboot_notifier(&md_notifier);
9889 unregister_sysctl_table(raid_table_header);
9891 /* We cannot unload the modules while some process is
9892 * waiting for us in select() or poll() - wake them up
9895 while (waitqueue_active(&md_event_waiters)) {
9896 /* not safe to leave yet */
9897 wake_up(&md_event_waiters);
9901 remove_proc_entry("mdstat", NULL);
9903 for_each_mddev(mddev, tmp) {
9904 export_array(mddev);
9906 mddev->hold_active = 0;
9908 * for_each_mddev() will call mddev_put() at the end of each
9909 * iteration. As the mddev is now fully clear, this will
9910 * schedule the mddev for destruction by a workqueue, and the
9911 * destroy_workqueue() below will wait for that to complete.
9914 destroy_workqueue(md_rdev_misc_wq);
9915 destroy_workqueue(md_misc_wq);
9916 destroy_workqueue(md_wq);
9919 subsys_initcall(md_init);
9920 module_exit(md_exit)
9922 static int get_ro(char *buffer, const struct kernel_param *kp)
9924 return sprintf(buffer, "%d\n", start_readonly);
9926 static int set_ro(const char *val, const struct kernel_param *kp)
9928 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9931 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9932 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9933 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9934 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9936 MODULE_LICENSE("GPL");
9937 MODULE_DESCRIPTION("MD RAID framework");
9939 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);