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/blk-integrity.h>
45 #include <linux/badblocks.h>
46 #include <linux/sysctl.h>
47 #include <linux/seq_file.h>
49 #include <linux/poll.h>
50 #include <linux/ctype.h>
51 #include <linux/string.h>
52 #include <linux/hdreg.h>
53 #include <linux/proc_fs.h>
54 #include <linux/random.h>
55 #include <linux/major.h>
56 #include <linux/module.h>
57 #include <linux/reboot.h>
58 #include <linux/file.h>
59 #include <linux/compat.h>
60 #include <linux/delay.h>
61 #include <linux/raid/md_p.h>
62 #include <linux/raid/md_u.h>
63 #include <linux/raid/detect.h>
64 #include <linux/slab.h>
65 #include <linux/percpu-refcount.h>
66 #include <linux/part_stat.h>
68 #include <trace/events/block.h>
70 #include "md-bitmap.h"
71 #include "md-cluster.h"
73 /* pers_list is a list of registered personalities protected by pers_lock. */
74 static LIST_HEAD(pers_list);
75 static DEFINE_SPINLOCK(pers_lock);
77 static const struct kobj_type md_ktype;
79 struct md_cluster_operations *md_cluster_ops;
80 EXPORT_SYMBOL(md_cluster_ops);
81 static struct module *md_cluster_mod;
83 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
84 static struct workqueue_struct *md_wq;
85 static struct workqueue_struct *md_misc_wq;
86 struct workqueue_struct *md_bitmap_wq;
88 static int remove_and_add_spares(struct mddev *mddev,
89 struct md_rdev *this);
90 static void mddev_detach(struct mddev *mddev);
91 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev);
92 static void md_wakeup_thread_directly(struct md_thread __rcu *thread);
95 * Default number of read corrections we'll attempt on an rdev
96 * before ejecting it from the array. We divide the read error
97 * count by 2 for every hour elapsed between read errors.
99 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
100 /* Default safemode delay: 200 msec */
101 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
103 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
104 * is 1000 KB/sec, so the extra system load does not show up that much.
105 * Increase it if you want to have more _guaranteed_ speed. Note that
106 * the RAID driver will use the maximum available bandwidth if the IO
107 * subsystem is idle. There is also an 'absolute maximum' reconstruction
108 * speed limit - in case reconstruction slows down your system despite
111 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
112 * or /sys/block/mdX/md/sync_speed_{min,max}
115 static int sysctl_speed_limit_min = 1000;
116 static int sysctl_speed_limit_max = 200000;
117 static inline int speed_min(struct mddev *mddev)
119 return mddev->sync_speed_min ?
120 mddev->sync_speed_min : sysctl_speed_limit_min;
123 static inline int speed_max(struct mddev *mddev)
125 return mddev->sync_speed_max ?
126 mddev->sync_speed_max : sysctl_speed_limit_max;
129 static void rdev_uninit_serial(struct md_rdev *rdev)
131 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
134 kvfree(rdev->serial);
138 static void rdevs_uninit_serial(struct mddev *mddev)
140 struct md_rdev *rdev;
142 rdev_for_each(rdev, mddev)
143 rdev_uninit_serial(rdev);
146 static int rdev_init_serial(struct md_rdev *rdev)
148 /* serial_nums equals with BARRIER_BUCKETS_NR */
149 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
150 struct serial_in_rdev *serial = NULL;
152 if (test_bit(CollisionCheck, &rdev->flags))
155 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
160 for (i = 0; i < serial_nums; i++) {
161 struct serial_in_rdev *serial_tmp = &serial[i];
163 spin_lock_init(&serial_tmp->serial_lock);
164 serial_tmp->serial_rb = RB_ROOT_CACHED;
165 init_waitqueue_head(&serial_tmp->serial_io_wait);
168 rdev->serial = serial;
169 set_bit(CollisionCheck, &rdev->flags);
174 static int rdevs_init_serial(struct mddev *mddev)
176 struct md_rdev *rdev;
179 rdev_for_each(rdev, mddev) {
180 ret = rdev_init_serial(rdev);
185 /* Free all resources if pool is not existed */
186 if (ret && !mddev->serial_info_pool)
187 rdevs_uninit_serial(mddev);
193 * rdev needs to enable serial stuffs if it meets the conditions:
194 * 1. it is multi-queue device flaged with writemostly.
195 * 2. the write-behind mode is enabled.
197 static int rdev_need_serial(struct md_rdev *rdev)
199 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
200 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
201 test_bit(WriteMostly, &rdev->flags));
205 * Init resource for rdev(s), then create serial_info_pool if:
206 * 1. rdev is the first device which return true from rdev_enable_serial.
207 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
209 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
214 if (rdev && !rdev_need_serial(rdev) &&
215 !test_bit(CollisionCheck, &rdev->flags))
219 mddev_suspend(mddev);
222 ret = rdevs_init_serial(mddev);
224 ret = rdev_init_serial(rdev);
228 if (mddev->serial_info_pool == NULL) {
230 * already in memalloc noio context by
233 mddev->serial_info_pool =
234 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
235 sizeof(struct serial_info));
236 if (!mddev->serial_info_pool) {
237 rdevs_uninit_serial(mddev);
238 pr_err("can't alloc memory pool for serialization\n");
248 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
249 * 1. rdev is the last device flaged with CollisionCheck.
250 * 2. when bitmap is destroyed while policy is not enabled.
251 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
253 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
256 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
259 if (mddev->serial_info_pool) {
260 struct md_rdev *temp;
261 int num = 0; /* used to track if other rdevs need the pool */
264 mddev_suspend(mddev);
265 rdev_for_each(temp, mddev) {
267 if (!mddev->serialize_policy ||
268 !rdev_need_serial(temp))
269 rdev_uninit_serial(temp);
272 } else if (temp != rdev &&
273 test_bit(CollisionCheck, &temp->flags))
278 rdev_uninit_serial(rdev);
281 pr_info("The mempool could be used by other devices\n");
283 mempool_destroy(mddev->serial_info_pool);
284 mddev->serial_info_pool = NULL;
291 static struct ctl_table_header *raid_table_header;
293 static struct ctl_table raid_table[] = {
295 .procname = "speed_limit_min",
296 .data = &sysctl_speed_limit_min,
297 .maxlen = sizeof(int),
298 .mode = S_IRUGO|S_IWUSR,
299 .proc_handler = proc_dointvec,
302 .procname = "speed_limit_max",
303 .data = &sysctl_speed_limit_max,
304 .maxlen = sizeof(int),
305 .mode = S_IRUGO|S_IWUSR,
306 .proc_handler = proc_dointvec,
311 static int start_readonly;
314 * The original mechanism for creating an md device is to create
315 * a device node in /dev and to open it. This causes races with device-close.
316 * The preferred method is to write to the "new_array" module parameter.
317 * This can avoid races.
318 * Setting create_on_open to false disables the original mechanism
319 * so all the races disappear.
321 static bool create_on_open = true;
324 * We have a system wide 'event count' that is incremented
325 * on any 'interesting' event, and readers of /proc/mdstat
326 * can use 'poll' or 'select' to find out when the event
330 * start array, stop array, error, add device, remove device,
331 * start build, activate spare
333 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
334 static atomic_t md_event_count;
335 void md_new_event(void)
337 atomic_inc(&md_event_count);
338 wake_up(&md_event_waiters);
340 EXPORT_SYMBOL_GPL(md_new_event);
343 * Enables to iterate over all existing md arrays
344 * all_mddevs_lock protects this list.
346 static LIST_HEAD(all_mddevs);
347 static DEFINE_SPINLOCK(all_mddevs_lock);
349 /* Rather than calling directly into the personality make_request function,
350 * IO requests come here first so that we can check if the device is
351 * being suspended pending a reconfiguration.
352 * We hold a refcount over the call to ->make_request. By the time that
353 * call has finished, the bio has been linked into some internal structure
354 * and so is visible to ->quiesce(), so we don't need the refcount any more.
356 static bool is_suspended(struct mddev *mddev, struct bio *bio)
358 if (is_md_suspended(mddev))
360 if (bio_data_dir(bio) != WRITE)
362 if (mddev->suspend_lo >= mddev->suspend_hi)
364 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
366 if (bio_end_sector(bio) < mddev->suspend_lo)
371 void md_handle_request(struct mddev *mddev, struct bio *bio)
374 if (is_suspended(mddev, bio)) {
376 /* Bail out if REQ_NOWAIT is set for the bio */
377 if (bio->bi_opf & REQ_NOWAIT) {
378 bio_wouldblock_error(bio);
382 prepare_to_wait(&mddev->sb_wait, &__wait,
383 TASK_UNINTERRUPTIBLE);
384 if (!is_suspended(mddev, bio))
388 finish_wait(&mddev->sb_wait, &__wait);
390 if (!percpu_ref_tryget_live(&mddev->active_io))
391 goto check_suspended;
393 if (!mddev->pers->make_request(mddev, bio)) {
394 percpu_ref_put(&mddev->active_io);
395 goto check_suspended;
398 percpu_ref_put(&mddev->active_io);
400 EXPORT_SYMBOL(md_handle_request);
402 static void md_submit_bio(struct bio *bio)
404 const int rw = bio_data_dir(bio);
405 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
407 if (mddev == NULL || mddev->pers == NULL) {
412 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
417 bio = bio_split_to_limits(bio);
421 if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
422 if (bio_sectors(bio) != 0)
423 bio->bi_status = BLK_STS_IOERR;
428 /* bio could be mergeable after passing to underlayer */
429 bio->bi_opf &= ~REQ_NOMERGE;
431 md_handle_request(mddev, bio);
434 /* mddev_suspend makes sure no new requests are submitted
435 * to the device, and that any requests that have been submitted
436 * are completely handled.
437 * Once mddev_detach() is called and completes, the module will be
440 void mddev_suspend(struct mddev *mddev)
442 struct md_thread *thread = rcu_dereference_protected(mddev->thread,
443 lockdep_is_held(&mddev->reconfig_mutex));
445 WARN_ON_ONCE(thread && current == thread->tsk);
446 if (mddev->suspended++)
448 wake_up(&mddev->sb_wait);
449 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
450 percpu_ref_kill(&mddev->active_io);
452 if (mddev->pers && mddev->pers->prepare_suspend)
453 mddev->pers->prepare_suspend(mddev);
455 wait_event(mddev->sb_wait, percpu_ref_is_zero(&mddev->active_io));
456 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
457 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
459 del_timer_sync(&mddev->safemode_timer);
460 /* restrict memory reclaim I/O during raid array is suspend */
461 mddev->noio_flag = memalloc_noio_save();
463 EXPORT_SYMBOL_GPL(mddev_suspend);
465 void mddev_resume(struct mddev *mddev)
467 lockdep_assert_held(&mddev->reconfig_mutex);
468 if (--mddev->suspended)
471 /* entred the memalloc scope from mddev_suspend() */
472 memalloc_noio_restore(mddev->noio_flag);
474 percpu_ref_resurrect(&mddev->active_io);
475 wake_up(&mddev->sb_wait);
477 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
478 md_wakeup_thread(mddev->thread);
479 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
481 EXPORT_SYMBOL_GPL(mddev_resume);
484 * Generic flush handling for md
487 static void md_end_flush(struct bio *bio)
489 struct md_rdev *rdev = bio->bi_private;
490 struct mddev *mddev = rdev->mddev;
494 rdev_dec_pending(rdev, mddev);
496 if (atomic_dec_and_test(&mddev->flush_pending)) {
497 /* The pair is percpu_ref_get() from md_flush_request() */
498 percpu_ref_put(&mddev->active_io);
500 /* The pre-request flush has finished */
501 queue_work(md_wq, &mddev->flush_work);
505 static void md_submit_flush_data(struct work_struct *ws);
507 static void submit_flushes(struct work_struct *ws)
509 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
510 struct md_rdev *rdev;
512 mddev->start_flush = ktime_get_boottime();
513 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
514 atomic_set(&mddev->flush_pending, 1);
516 rdev_for_each_rcu(rdev, mddev)
517 if (rdev->raid_disk >= 0 &&
518 !test_bit(Faulty, &rdev->flags)) {
521 atomic_inc(&rdev->nr_pending);
523 bi = bio_alloc_bioset(rdev->bdev, 0,
524 REQ_OP_WRITE | REQ_PREFLUSH,
525 GFP_NOIO, &mddev->bio_set);
526 bi->bi_end_io = md_end_flush;
527 bi->bi_private = rdev;
528 atomic_inc(&mddev->flush_pending);
533 if (atomic_dec_and_test(&mddev->flush_pending))
534 queue_work(md_wq, &mddev->flush_work);
537 static void md_submit_flush_data(struct work_struct *ws)
539 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
540 struct bio *bio = mddev->flush_bio;
543 * must reset flush_bio before calling into md_handle_request to avoid a
544 * deadlock, because other bios passed md_handle_request suspend check
545 * could wait for this and below md_handle_request could wait for those
546 * bios because of suspend check
548 spin_lock_irq(&mddev->lock);
549 mddev->prev_flush_start = mddev->start_flush;
550 mddev->flush_bio = NULL;
551 spin_unlock_irq(&mddev->lock);
552 wake_up(&mddev->sb_wait);
554 if (bio->bi_iter.bi_size == 0) {
555 /* an empty barrier - all done */
558 bio->bi_opf &= ~REQ_PREFLUSH;
559 md_handle_request(mddev, bio);
564 * Manages consolidation of flushes and submitting any flushes needed for
565 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
566 * being finished in another context. Returns false if the flushing is
567 * complete but still needs the I/O portion of the bio to be processed.
569 bool md_flush_request(struct mddev *mddev, struct bio *bio)
571 ktime_t req_start = ktime_get_boottime();
572 spin_lock_irq(&mddev->lock);
573 /* flush requests wait until ongoing flush completes,
574 * hence coalescing all the pending requests.
576 wait_event_lock_irq(mddev->sb_wait,
578 ktime_before(req_start, mddev->prev_flush_start),
580 /* new request after previous flush is completed */
581 if (ktime_after(req_start, mddev->prev_flush_start)) {
582 WARN_ON(mddev->flush_bio);
584 * Grab a reference to make sure mddev_suspend() will wait for
585 * this flush to be done.
587 * md_flush_reqeust() is called under md_handle_request() and
588 * 'active_io' is already grabbed, hence percpu_ref_is_zero()
589 * won't pass, percpu_ref_tryget_live() can't be used because
590 * percpu_ref_kill() can be called by mddev_suspend()
593 WARN_ON(percpu_ref_is_zero(&mddev->active_io));
594 percpu_ref_get(&mddev->active_io);
595 mddev->flush_bio = bio;
598 spin_unlock_irq(&mddev->lock);
601 INIT_WORK(&mddev->flush_work, submit_flushes);
602 queue_work(md_wq, &mddev->flush_work);
604 /* flush was performed for some other bio while we waited. */
605 if (bio->bi_iter.bi_size == 0)
606 /* an empty barrier - all done */
609 bio->bi_opf &= ~REQ_PREFLUSH;
615 EXPORT_SYMBOL(md_flush_request);
617 static inline struct mddev *mddev_get(struct mddev *mddev)
619 lockdep_assert_held(&all_mddevs_lock);
621 if (test_bit(MD_DELETED, &mddev->flags))
623 atomic_inc(&mddev->active);
627 static void mddev_delayed_delete(struct work_struct *ws);
629 void mddev_put(struct mddev *mddev)
631 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
633 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
634 mddev->ctime == 0 && !mddev->hold_active) {
635 /* Array is not configured at all, and not held active,
637 set_bit(MD_DELETED, &mddev->flags);
640 * Call queue_work inside the spinlock so that
641 * flush_workqueue() after mddev_find will succeed in waiting
642 * for the work to be done.
644 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
645 queue_work(md_misc_wq, &mddev->del_work);
647 spin_unlock(&all_mddevs_lock);
650 static void md_safemode_timeout(struct timer_list *t);
652 void mddev_init(struct mddev *mddev)
654 mutex_init(&mddev->open_mutex);
655 mutex_init(&mddev->reconfig_mutex);
656 mutex_init(&mddev->sync_mutex);
657 mutex_init(&mddev->bitmap_info.mutex);
658 INIT_LIST_HEAD(&mddev->disks);
659 INIT_LIST_HEAD(&mddev->all_mddevs);
660 INIT_LIST_HEAD(&mddev->deleting);
661 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
662 atomic_set(&mddev->active, 1);
663 atomic_set(&mddev->openers, 0);
664 atomic_set(&mddev->sync_seq, 0);
665 spin_lock_init(&mddev->lock);
666 atomic_set(&mddev->flush_pending, 0);
667 init_waitqueue_head(&mddev->sb_wait);
668 init_waitqueue_head(&mddev->recovery_wait);
669 mddev->reshape_position = MaxSector;
670 mddev->reshape_backwards = 0;
671 mddev->last_sync_action = "none";
672 mddev->resync_min = 0;
673 mddev->resync_max = MaxSector;
674 mddev->level = LEVEL_NONE;
676 EXPORT_SYMBOL_GPL(mddev_init);
678 static struct mddev *mddev_find_locked(dev_t unit)
682 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
683 if (mddev->unit == unit)
689 /* find an unused unit number */
690 static dev_t mddev_alloc_unit(void)
692 static int next_minor = 512;
693 int start = next_minor;
698 dev = MKDEV(MD_MAJOR, next_minor);
700 if (next_minor > MINORMASK)
702 if (next_minor == start)
703 return 0; /* Oh dear, all in use. */
704 is_free = !mddev_find_locked(dev);
710 static struct mddev *mddev_alloc(dev_t unit)
715 if (unit && MAJOR(unit) != MD_MAJOR)
716 unit &= ~((1 << MdpMinorShift) - 1);
718 new = kzalloc(sizeof(*new), GFP_KERNEL);
720 return ERR_PTR(-ENOMEM);
723 spin_lock(&all_mddevs_lock);
726 if (mddev_find_locked(unit))
729 if (MAJOR(unit) == MD_MAJOR)
730 new->md_minor = MINOR(unit);
732 new->md_minor = MINOR(unit) >> MdpMinorShift;
733 new->hold_active = UNTIL_IOCTL;
736 new->unit = mddev_alloc_unit();
739 new->md_minor = MINOR(new->unit);
740 new->hold_active = UNTIL_STOP;
743 list_add(&new->all_mddevs, &all_mddevs);
744 spin_unlock(&all_mddevs_lock);
747 spin_unlock(&all_mddevs_lock);
749 return ERR_PTR(error);
752 static void mddev_free(struct mddev *mddev)
754 spin_lock(&all_mddevs_lock);
755 list_del(&mddev->all_mddevs);
756 spin_unlock(&all_mddevs_lock);
761 static const struct attribute_group md_redundancy_group;
763 void mddev_unlock(struct mddev *mddev)
765 struct md_rdev *rdev;
769 if (!list_empty(&mddev->deleting))
770 list_splice_init(&mddev->deleting, &delete);
772 if (mddev->to_remove) {
773 /* These cannot be removed under reconfig_mutex as
774 * an access to the files will try to take reconfig_mutex
775 * while holding the file unremovable, which leads to
777 * So hold set sysfs_active while the remove in happeing,
778 * and anything else which might set ->to_remove or my
779 * otherwise change the sysfs namespace will fail with
780 * -EBUSY if sysfs_active is still set.
781 * We set sysfs_active under reconfig_mutex and elsewhere
782 * test it under the same mutex to ensure its correct value
785 const struct attribute_group *to_remove = mddev->to_remove;
786 mddev->to_remove = NULL;
787 mddev->sysfs_active = 1;
788 mutex_unlock(&mddev->reconfig_mutex);
790 if (mddev->kobj.sd) {
791 if (to_remove != &md_redundancy_group)
792 sysfs_remove_group(&mddev->kobj, to_remove);
793 if (mddev->pers == NULL ||
794 mddev->pers->sync_request == NULL) {
795 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
796 if (mddev->sysfs_action)
797 sysfs_put(mddev->sysfs_action);
798 if (mddev->sysfs_completed)
799 sysfs_put(mddev->sysfs_completed);
800 if (mddev->sysfs_degraded)
801 sysfs_put(mddev->sysfs_degraded);
802 mddev->sysfs_action = NULL;
803 mddev->sysfs_completed = NULL;
804 mddev->sysfs_degraded = NULL;
807 mddev->sysfs_active = 0;
809 mutex_unlock(&mddev->reconfig_mutex);
811 md_wakeup_thread(mddev->thread);
812 wake_up(&mddev->sb_wait);
814 list_for_each_entry_safe(rdev, tmp, &delete, same_set) {
815 list_del_init(&rdev->same_set);
816 kobject_del(&rdev->kobj);
817 export_rdev(rdev, mddev);
820 EXPORT_SYMBOL_GPL(mddev_unlock);
822 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
824 struct md_rdev *rdev;
826 rdev_for_each_rcu(rdev, mddev)
827 if (rdev->desc_nr == nr)
832 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
834 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
836 struct md_rdev *rdev;
838 rdev_for_each(rdev, mddev)
839 if (rdev->bdev->bd_dev == dev)
845 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
847 struct md_rdev *rdev;
849 rdev_for_each_rcu(rdev, mddev)
850 if (rdev->bdev->bd_dev == dev)
855 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
857 static struct md_personality *find_pers(int level, char *clevel)
859 struct md_personality *pers;
860 list_for_each_entry(pers, &pers_list, list) {
861 if (level != LEVEL_NONE && pers->level == level)
863 if (strcmp(pers->name, clevel)==0)
869 /* return the offset of the super block in 512byte sectors */
870 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
872 return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
875 static int alloc_disk_sb(struct md_rdev *rdev)
877 rdev->sb_page = alloc_page(GFP_KERNEL);
883 void md_rdev_clear(struct md_rdev *rdev)
886 put_page(rdev->sb_page);
888 rdev->sb_page = NULL;
893 put_page(rdev->bb_page);
894 rdev->bb_page = NULL;
896 badblocks_exit(&rdev->badblocks);
898 EXPORT_SYMBOL_GPL(md_rdev_clear);
900 static void super_written(struct bio *bio)
902 struct md_rdev *rdev = bio->bi_private;
903 struct mddev *mddev = rdev->mddev;
905 if (bio->bi_status) {
906 pr_err("md: %s gets error=%d\n", __func__,
907 blk_status_to_errno(bio->bi_status));
908 md_error(mddev, rdev);
909 if (!test_bit(Faulty, &rdev->flags)
910 && (bio->bi_opf & MD_FAILFAST)) {
911 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
912 set_bit(LastDev, &rdev->flags);
915 clear_bit(LastDev, &rdev->flags);
919 rdev_dec_pending(rdev, mddev);
921 if (atomic_dec_and_test(&mddev->pending_writes))
922 wake_up(&mddev->sb_wait);
925 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
926 sector_t sector, int size, struct page *page)
928 /* write first size bytes of page to sector of rdev
929 * Increment mddev->pending_writes before returning
930 * and decrement it on completion, waking up sb_wait
931 * if zero is reached.
932 * If an error occurred, call md_error
939 if (test_bit(Faulty, &rdev->flags))
942 bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
944 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA,
945 GFP_NOIO, &mddev->sync_set);
947 atomic_inc(&rdev->nr_pending);
949 bio->bi_iter.bi_sector = sector;
950 __bio_add_page(bio, page, size, 0);
951 bio->bi_private = rdev;
952 bio->bi_end_io = super_written;
954 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
955 test_bit(FailFast, &rdev->flags) &&
956 !test_bit(LastDev, &rdev->flags))
957 bio->bi_opf |= MD_FAILFAST;
959 atomic_inc(&mddev->pending_writes);
963 int md_super_wait(struct mddev *mddev)
965 /* wait for all superblock writes that were scheduled to complete */
966 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
967 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
972 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
973 struct page *page, blk_opf_t opf, bool metadata_op)
978 if (metadata_op && rdev->meta_bdev)
979 bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
981 bio_init(&bio, rdev->bdev, &bvec, 1, opf);
984 bio.bi_iter.bi_sector = sector + rdev->sb_start;
985 else if (rdev->mddev->reshape_position != MaxSector &&
986 (rdev->mddev->reshape_backwards ==
987 (sector >= rdev->mddev->reshape_position)))
988 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
990 bio.bi_iter.bi_sector = sector + rdev->data_offset;
991 __bio_add_page(&bio, page, size, 0);
993 submit_bio_wait(&bio);
995 return !bio.bi_status;
997 EXPORT_SYMBOL_GPL(sync_page_io);
999 static int read_disk_sb(struct md_rdev *rdev, int size)
1001 if (rdev->sb_loaded)
1004 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1006 rdev->sb_loaded = 1;
1010 pr_err("md: disabled device %pg, could not read superblock.\n",
1015 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1017 return sb1->set_uuid0 == sb2->set_uuid0 &&
1018 sb1->set_uuid1 == sb2->set_uuid1 &&
1019 sb1->set_uuid2 == sb2->set_uuid2 &&
1020 sb1->set_uuid3 == sb2->set_uuid3;
1023 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1026 mdp_super_t *tmp1, *tmp2;
1028 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1029 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1031 if (!tmp1 || !tmp2) {
1040 * nr_disks is not constant
1045 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1052 static u32 md_csum_fold(u32 csum)
1054 csum = (csum & 0xffff) + (csum >> 16);
1055 return (csum & 0xffff) + (csum >> 16);
1058 static unsigned int calc_sb_csum(mdp_super_t *sb)
1061 u32 *sb32 = (u32*)sb;
1063 unsigned int disk_csum, csum;
1065 disk_csum = sb->sb_csum;
1068 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1070 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1073 /* This used to use csum_partial, which was wrong for several
1074 * reasons including that different results are returned on
1075 * different architectures. It isn't critical that we get exactly
1076 * the same return value as before (we always csum_fold before
1077 * testing, and that removes any differences). However as we
1078 * know that csum_partial always returned a 16bit value on
1079 * alphas, do a fold to maximise conformity to previous behaviour.
1081 sb->sb_csum = md_csum_fold(disk_csum);
1083 sb->sb_csum = disk_csum;
1089 * Handle superblock details.
1090 * We want to be able to handle multiple superblock formats
1091 * so we have a common interface to them all, and an array of
1092 * different handlers.
1093 * We rely on user-space to write the initial superblock, and support
1094 * reading and updating of superblocks.
1095 * Interface methods are:
1096 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1097 * loads and validates a superblock on dev.
1098 * if refdev != NULL, compare superblocks on both devices
1100 * 0 - dev has a superblock that is compatible with refdev
1101 * 1 - dev has a superblock that is compatible and newer than refdev
1102 * so dev should be used as the refdev in future
1103 * -EINVAL superblock incompatible or invalid
1104 * -othererror e.g. -EIO
1106 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1107 * Verify that dev is acceptable into mddev.
1108 * The first time, mddev->raid_disks will be 0, and data from
1109 * dev should be merged in. Subsequent calls check that dev
1110 * is new enough. Return 0 or -EINVAL
1112 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1113 * Update the superblock for rdev with data in mddev
1114 * This does not write to disc.
1120 struct module *owner;
1121 int (*load_super)(struct md_rdev *rdev,
1122 struct md_rdev *refdev,
1124 int (*validate_super)(struct mddev *mddev,
1125 struct md_rdev *freshest,
1126 struct md_rdev *rdev);
1127 void (*sync_super)(struct mddev *mddev,
1128 struct md_rdev *rdev);
1129 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1130 sector_t num_sectors);
1131 int (*allow_new_offset)(struct md_rdev *rdev,
1132 unsigned long long new_offset);
1136 * Check that the given mddev has no bitmap.
1138 * This function is called from the run method of all personalities that do not
1139 * support bitmaps. It prints an error message and returns non-zero if mddev
1140 * has a bitmap. Otherwise, it returns 0.
1143 int md_check_no_bitmap(struct mddev *mddev)
1145 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1147 pr_warn("%s: bitmaps are not supported for %s\n",
1148 mdname(mddev), mddev->pers->name);
1151 EXPORT_SYMBOL(md_check_no_bitmap);
1154 * load_super for 0.90.0
1156 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1160 bool spare_disk = true;
1163 * Calculate the position of the superblock (512byte sectors),
1164 * it's at the end of the disk.
1166 * It also happens to be a multiple of 4Kb.
1168 rdev->sb_start = calc_dev_sboffset(rdev);
1170 ret = read_disk_sb(rdev, MD_SB_BYTES);
1176 sb = page_address(rdev->sb_page);
1178 if (sb->md_magic != MD_SB_MAGIC) {
1179 pr_warn("md: invalid raid superblock magic on %pg\n",
1184 if (sb->major_version != 0 ||
1185 sb->minor_version < 90 ||
1186 sb->minor_version > 91) {
1187 pr_warn("Bad version number %d.%d on %pg\n",
1188 sb->major_version, sb->minor_version, rdev->bdev);
1192 if (sb->raid_disks <= 0)
1195 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1196 pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1200 rdev->preferred_minor = sb->md_minor;
1201 rdev->data_offset = 0;
1202 rdev->new_data_offset = 0;
1203 rdev->sb_size = MD_SB_BYTES;
1204 rdev->badblocks.shift = -1;
1206 if (sb->level == LEVEL_MULTIPATH)
1209 rdev->desc_nr = sb->this_disk.number;
1211 /* not spare disk, or LEVEL_MULTIPATH */
1212 if (sb->level == LEVEL_MULTIPATH ||
1213 (rdev->desc_nr >= 0 &&
1214 rdev->desc_nr < MD_SB_DISKS &&
1215 sb->disks[rdev->desc_nr].state &
1216 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1226 mdp_super_t *refsb = page_address(refdev->sb_page);
1227 if (!md_uuid_equal(refsb, sb)) {
1228 pr_warn("md: %pg has different UUID to %pg\n",
1229 rdev->bdev, refdev->bdev);
1232 if (!md_sb_equal(refsb, sb)) {
1233 pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1234 rdev->bdev, refdev->bdev);
1238 ev2 = md_event(refsb);
1240 if (!spare_disk && ev1 > ev2)
1245 rdev->sectors = rdev->sb_start;
1246 /* Limit to 4TB as metadata cannot record more than that.
1247 * (not needed for Linear and RAID0 as metadata doesn't
1250 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1251 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1253 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1254 /* "this cannot possibly happen" ... */
1262 * validate_super for 0.90.0
1263 * note: we are not using "freshest" for 0.9 superblock
1265 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1268 mdp_super_t *sb = page_address(rdev->sb_page);
1269 __u64 ev1 = md_event(sb);
1271 rdev->raid_disk = -1;
1272 clear_bit(Faulty, &rdev->flags);
1273 clear_bit(In_sync, &rdev->flags);
1274 clear_bit(Bitmap_sync, &rdev->flags);
1275 clear_bit(WriteMostly, &rdev->flags);
1277 if (mddev->raid_disks == 0) {
1278 mddev->major_version = 0;
1279 mddev->minor_version = sb->minor_version;
1280 mddev->patch_version = sb->patch_version;
1281 mddev->external = 0;
1282 mddev->chunk_sectors = sb->chunk_size >> 9;
1283 mddev->ctime = sb->ctime;
1284 mddev->utime = sb->utime;
1285 mddev->level = sb->level;
1286 mddev->clevel[0] = 0;
1287 mddev->layout = sb->layout;
1288 mddev->raid_disks = sb->raid_disks;
1289 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1290 mddev->events = ev1;
1291 mddev->bitmap_info.offset = 0;
1292 mddev->bitmap_info.space = 0;
1293 /* bitmap can use 60 K after the 4K superblocks */
1294 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1295 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1296 mddev->reshape_backwards = 0;
1298 if (mddev->minor_version >= 91) {
1299 mddev->reshape_position = sb->reshape_position;
1300 mddev->delta_disks = sb->delta_disks;
1301 mddev->new_level = sb->new_level;
1302 mddev->new_layout = sb->new_layout;
1303 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1304 if (mddev->delta_disks < 0)
1305 mddev->reshape_backwards = 1;
1307 mddev->reshape_position = MaxSector;
1308 mddev->delta_disks = 0;
1309 mddev->new_level = mddev->level;
1310 mddev->new_layout = mddev->layout;
1311 mddev->new_chunk_sectors = mddev->chunk_sectors;
1313 if (mddev->level == 0)
1316 if (sb->state & (1<<MD_SB_CLEAN))
1317 mddev->recovery_cp = MaxSector;
1319 if (sb->events_hi == sb->cp_events_hi &&
1320 sb->events_lo == sb->cp_events_lo) {
1321 mddev->recovery_cp = sb->recovery_cp;
1323 mddev->recovery_cp = 0;
1326 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1327 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1328 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1329 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1331 mddev->max_disks = MD_SB_DISKS;
1333 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1334 mddev->bitmap_info.file == NULL) {
1335 mddev->bitmap_info.offset =
1336 mddev->bitmap_info.default_offset;
1337 mddev->bitmap_info.space =
1338 mddev->bitmap_info.default_space;
1341 } else if (mddev->pers == NULL) {
1342 /* Insist on good event counter while assembling, except
1343 * for spares (which don't need an event count) */
1345 if (sb->disks[rdev->desc_nr].state & (
1346 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1347 if (ev1 < mddev->events)
1349 } else if (mddev->bitmap) {
1350 /* if adding to array with a bitmap, then we can accept an
1351 * older device ... but not too old.
1353 if (ev1 < mddev->bitmap->events_cleared)
1355 if (ev1 < mddev->events)
1356 set_bit(Bitmap_sync, &rdev->flags);
1358 if (ev1 < mddev->events)
1359 /* just a hot-add of a new device, leave raid_disk at -1 */
1363 if (mddev->level != LEVEL_MULTIPATH) {
1364 desc = sb->disks + rdev->desc_nr;
1366 if (desc->state & (1<<MD_DISK_FAULTY))
1367 set_bit(Faulty, &rdev->flags);
1368 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1369 desc->raid_disk < mddev->raid_disks */) {
1370 set_bit(In_sync, &rdev->flags);
1371 rdev->raid_disk = desc->raid_disk;
1372 rdev->saved_raid_disk = desc->raid_disk;
1373 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1374 /* active but not in sync implies recovery up to
1375 * reshape position. We don't know exactly where
1376 * that is, so set to zero for now */
1377 if (mddev->minor_version >= 91) {
1378 rdev->recovery_offset = 0;
1379 rdev->raid_disk = desc->raid_disk;
1382 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1383 set_bit(WriteMostly, &rdev->flags);
1384 if (desc->state & (1<<MD_DISK_FAILFAST))
1385 set_bit(FailFast, &rdev->flags);
1386 } else /* MULTIPATH are always insync */
1387 set_bit(In_sync, &rdev->flags);
1392 * sync_super for 0.90.0
1394 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1397 struct md_rdev *rdev2;
1398 int next_spare = mddev->raid_disks;
1400 /* make rdev->sb match mddev data..
1403 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1404 * 3/ any empty disks < next_spare become removed
1406 * disks[0] gets initialised to REMOVED because
1407 * we cannot be sure from other fields if it has
1408 * been initialised or not.
1411 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1413 rdev->sb_size = MD_SB_BYTES;
1415 sb = page_address(rdev->sb_page);
1417 memset(sb, 0, sizeof(*sb));
1419 sb->md_magic = MD_SB_MAGIC;
1420 sb->major_version = mddev->major_version;
1421 sb->patch_version = mddev->patch_version;
1422 sb->gvalid_words = 0; /* ignored */
1423 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1424 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1425 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1426 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1428 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1429 sb->level = mddev->level;
1430 sb->size = mddev->dev_sectors / 2;
1431 sb->raid_disks = mddev->raid_disks;
1432 sb->md_minor = mddev->md_minor;
1433 sb->not_persistent = 0;
1434 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1436 sb->events_hi = (mddev->events>>32);
1437 sb->events_lo = (u32)mddev->events;
1439 if (mddev->reshape_position == MaxSector)
1440 sb->minor_version = 90;
1442 sb->minor_version = 91;
1443 sb->reshape_position = mddev->reshape_position;
1444 sb->new_level = mddev->new_level;
1445 sb->delta_disks = mddev->delta_disks;
1446 sb->new_layout = mddev->new_layout;
1447 sb->new_chunk = mddev->new_chunk_sectors << 9;
1449 mddev->minor_version = sb->minor_version;
1452 sb->recovery_cp = mddev->recovery_cp;
1453 sb->cp_events_hi = (mddev->events>>32);
1454 sb->cp_events_lo = (u32)mddev->events;
1455 if (mddev->recovery_cp == MaxSector)
1456 sb->state = (1<< MD_SB_CLEAN);
1458 sb->recovery_cp = 0;
1460 sb->layout = mddev->layout;
1461 sb->chunk_size = mddev->chunk_sectors << 9;
1463 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1464 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1466 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1467 rdev_for_each(rdev2, mddev) {
1470 int is_active = test_bit(In_sync, &rdev2->flags);
1472 if (rdev2->raid_disk >= 0 &&
1473 sb->minor_version >= 91)
1474 /* we have nowhere to store the recovery_offset,
1475 * but if it is not below the reshape_position,
1476 * we can piggy-back on that.
1479 if (rdev2->raid_disk < 0 ||
1480 test_bit(Faulty, &rdev2->flags))
1483 desc_nr = rdev2->raid_disk;
1485 desc_nr = next_spare++;
1486 rdev2->desc_nr = desc_nr;
1487 d = &sb->disks[rdev2->desc_nr];
1489 d->number = rdev2->desc_nr;
1490 d->major = MAJOR(rdev2->bdev->bd_dev);
1491 d->minor = MINOR(rdev2->bdev->bd_dev);
1493 d->raid_disk = rdev2->raid_disk;
1495 d->raid_disk = rdev2->desc_nr; /* compatibility */
1496 if (test_bit(Faulty, &rdev2->flags))
1497 d->state = (1<<MD_DISK_FAULTY);
1498 else if (is_active) {
1499 d->state = (1<<MD_DISK_ACTIVE);
1500 if (test_bit(In_sync, &rdev2->flags))
1501 d->state |= (1<<MD_DISK_SYNC);
1509 if (test_bit(WriteMostly, &rdev2->flags))
1510 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1511 if (test_bit(FailFast, &rdev2->flags))
1512 d->state |= (1<<MD_DISK_FAILFAST);
1514 /* now set the "removed" and "faulty" bits on any missing devices */
1515 for (i=0 ; i < mddev->raid_disks ; i++) {
1516 mdp_disk_t *d = &sb->disks[i];
1517 if (d->state == 0 && d->number == 0) {
1520 d->state = (1<<MD_DISK_REMOVED);
1521 d->state |= (1<<MD_DISK_FAULTY);
1525 sb->nr_disks = nr_disks;
1526 sb->active_disks = active;
1527 sb->working_disks = working;
1528 sb->failed_disks = failed;
1529 sb->spare_disks = spare;
1531 sb->this_disk = sb->disks[rdev->desc_nr];
1532 sb->sb_csum = calc_sb_csum(sb);
1536 * rdev_size_change for 0.90.0
1538 static unsigned long long
1539 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1541 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1542 return 0; /* component must fit device */
1543 if (rdev->mddev->bitmap_info.offset)
1544 return 0; /* can't move bitmap */
1545 rdev->sb_start = calc_dev_sboffset(rdev);
1546 if (!num_sectors || num_sectors > rdev->sb_start)
1547 num_sectors = rdev->sb_start;
1548 /* Limit to 4TB as metadata cannot record more than that.
1549 * 4TB == 2^32 KB, or 2*2^32 sectors.
1551 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1552 num_sectors = (sector_t)(2ULL << 32) - 2;
1554 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1556 } while (md_super_wait(rdev->mddev) < 0);
1561 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1563 /* non-zero offset changes not possible with v0.90 */
1564 return new_offset == 0;
1568 * version 1 superblock
1571 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1575 unsigned long long newcsum;
1576 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1577 __le32 *isuper = (__le32*)sb;
1579 disk_csum = sb->sb_csum;
1582 for (; size >= 4; size -= 4)
1583 newcsum += le32_to_cpu(*isuper++);
1586 newcsum += le16_to_cpu(*(__le16*) isuper);
1588 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1589 sb->sb_csum = disk_csum;
1590 return cpu_to_le32(csum);
1593 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1595 struct mdp_superblock_1 *sb;
1600 bool spare_disk = true;
1603 * Calculate the position of the superblock in 512byte sectors.
1604 * It is always aligned to a 4K boundary and
1605 * depeding on minor_version, it can be:
1606 * 0: At least 8K, but less than 12K, from end of device
1607 * 1: At start of device
1608 * 2: 4K from start of device.
1610 switch(minor_version) {
1612 sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1613 sb_start &= ~(sector_t)(4*2-1);
1624 rdev->sb_start = sb_start;
1626 /* superblock is rarely larger than 1K, but it can be larger,
1627 * and it is safe to read 4k, so we do that
1629 ret = read_disk_sb(rdev, 4096);
1630 if (ret) return ret;
1632 sb = page_address(rdev->sb_page);
1634 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1635 sb->major_version != cpu_to_le32(1) ||
1636 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1637 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1638 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1641 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1642 pr_warn("md: invalid superblock checksum on %pg\n",
1646 if (le64_to_cpu(sb->data_size) < 10) {
1647 pr_warn("md: data_size too small on %pg\n",
1653 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1654 /* Some padding is non-zero, might be a new feature */
1657 rdev->preferred_minor = 0xffff;
1658 rdev->data_offset = le64_to_cpu(sb->data_offset);
1659 rdev->new_data_offset = rdev->data_offset;
1660 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1661 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1662 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1663 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1665 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1666 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1667 if (rdev->sb_size & bmask)
1668 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1671 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1674 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1677 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1680 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1682 if (!rdev->bb_page) {
1683 rdev->bb_page = alloc_page(GFP_KERNEL);
1687 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1688 rdev->badblocks.count == 0) {
1689 /* need to load the bad block list.
1690 * Currently we limit it to one page.
1696 int sectors = le16_to_cpu(sb->bblog_size);
1697 if (sectors > (PAGE_SIZE / 512))
1699 offset = le32_to_cpu(sb->bblog_offset);
1702 bb_sector = (long long)offset;
1703 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1704 rdev->bb_page, REQ_OP_READ, true))
1706 bbp = (__le64 *)page_address(rdev->bb_page);
1707 rdev->badblocks.shift = sb->bblog_shift;
1708 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1709 u64 bb = le64_to_cpu(*bbp);
1710 int count = bb & (0x3ff);
1711 u64 sector = bb >> 10;
1712 sector <<= sb->bblog_shift;
1713 count <<= sb->bblog_shift;
1716 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1719 } else if (sb->bblog_offset != 0)
1720 rdev->badblocks.shift = 0;
1722 if ((le32_to_cpu(sb->feature_map) &
1723 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1724 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1725 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1726 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1729 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1733 /* not spare disk, or LEVEL_MULTIPATH */
1734 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1735 (rdev->desc_nr >= 0 &&
1736 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1737 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1738 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1748 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1750 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1751 sb->level != refsb->level ||
1752 sb->layout != refsb->layout ||
1753 sb->chunksize != refsb->chunksize) {
1754 pr_warn("md: %pg has strangely different superblock to %pg\n",
1759 ev1 = le64_to_cpu(sb->events);
1760 ev2 = le64_to_cpu(refsb->events);
1762 if (!spare_disk && ev1 > ev2)
1768 sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1770 sectors = rdev->sb_start;
1771 if (sectors < le64_to_cpu(sb->data_size))
1773 rdev->sectors = le64_to_cpu(sb->data_size);
1777 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1779 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1780 __u64 ev1 = le64_to_cpu(sb->events);
1782 rdev->raid_disk = -1;
1783 clear_bit(Faulty, &rdev->flags);
1784 clear_bit(In_sync, &rdev->flags);
1785 clear_bit(Bitmap_sync, &rdev->flags);
1786 clear_bit(WriteMostly, &rdev->flags);
1788 if (mddev->raid_disks == 0) {
1789 mddev->major_version = 1;
1790 mddev->patch_version = 0;
1791 mddev->external = 0;
1792 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1793 mddev->ctime = le64_to_cpu(sb->ctime);
1794 mddev->utime = le64_to_cpu(sb->utime);
1795 mddev->level = le32_to_cpu(sb->level);
1796 mddev->clevel[0] = 0;
1797 mddev->layout = le32_to_cpu(sb->layout);
1798 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1799 mddev->dev_sectors = le64_to_cpu(sb->size);
1800 mddev->events = ev1;
1801 mddev->bitmap_info.offset = 0;
1802 mddev->bitmap_info.space = 0;
1803 /* Default location for bitmap is 1K after superblock
1804 * using 3K - total of 4K
1806 mddev->bitmap_info.default_offset = 1024 >> 9;
1807 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1808 mddev->reshape_backwards = 0;
1810 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1811 memcpy(mddev->uuid, sb->set_uuid, 16);
1813 mddev->max_disks = (4096-256)/2;
1815 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1816 mddev->bitmap_info.file == NULL) {
1817 mddev->bitmap_info.offset =
1818 (__s32)le32_to_cpu(sb->bitmap_offset);
1819 /* Metadata doesn't record how much space is available.
1820 * For 1.0, we assume we can use up to the superblock
1821 * if before, else to 4K beyond superblock.
1822 * For others, assume no change is possible.
1824 if (mddev->minor_version > 0)
1825 mddev->bitmap_info.space = 0;
1826 else if (mddev->bitmap_info.offset > 0)
1827 mddev->bitmap_info.space =
1828 8 - mddev->bitmap_info.offset;
1830 mddev->bitmap_info.space =
1831 -mddev->bitmap_info.offset;
1834 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1835 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1836 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1837 mddev->new_level = le32_to_cpu(sb->new_level);
1838 mddev->new_layout = le32_to_cpu(sb->new_layout);
1839 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1840 if (mddev->delta_disks < 0 ||
1841 (mddev->delta_disks == 0 &&
1842 (le32_to_cpu(sb->feature_map)
1843 & MD_FEATURE_RESHAPE_BACKWARDS)))
1844 mddev->reshape_backwards = 1;
1846 mddev->reshape_position = MaxSector;
1847 mddev->delta_disks = 0;
1848 mddev->new_level = mddev->level;
1849 mddev->new_layout = mddev->layout;
1850 mddev->new_chunk_sectors = mddev->chunk_sectors;
1853 if (mddev->level == 0 &&
1854 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1857 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1858 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1860 if (le32_to_cpu(sb->feature_map) &
1861 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1862 if (le32_to_cpu(sb->feature_map) &
1863 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1865 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1866 (le32_to_cpu(sb->feature_map) &
1867 MD_FEATURE_MULTIPLE_PPLS))
1869 set_bit(MD_HAS_PPL, &mddev->flags);
1871 } else if (mddev->pers == NULL) {
1872 /* Insist of good event counter while assembling, except for
1873 * spares (which don't need an event count).
1874 * Similar to mdadm, we allow event counter difference of 1
1875 * from the freshest device.
1877 if (rdev->desc_nr >= 0 &&
1878 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1879 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1880 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1881 if (ev1 + 1 < mddev->events)
1883 } else if (mddev->bitmap) {
1884 /* If adding to array with a bitmap, then we can accept an
1885 * older device, but not too old.
1887 if (ev1 < mddev->bitmap->events_cleared)
1889 if (ev1 < mddev->events)
1890 set_bit(Bitmap_sync, &rdev->flags);
1892 if (ev1 < mddev->events)
1893 /* just a hot-add of a new device, leave raid_disk at -1 */
1896 if (mddev->level != LEVEL_MULTIPATH) {
1898 if (rdev->desc_nr < 0 ||
1899 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1900 role = MD_DISK_ROLE_SPARE;
1902 } else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1904 * If we are assembling, and our event counter is smaller than the
1905 * highest event counter, we cannot trust our superblock about the role.
1906 * It could happen that our rdev was marked as Faulty, and all other
1907 * superblocks were updated with +1 event counter.
1908 * Then, before the next superblock update, which typically happens when
1909 * remove_and_add_spares() removes the device from the array, there was
1910 * a crash or reboot.
1911 * If we allow current rdev without consulting the freshest superblock,
1912 * we could cause data corruption.
1913 * Note that in this case our event counter is smaller by 1 than the
1914 * highest, otherwise, this rdev would not be allowed into array;
1915 * both kernel and mdadm allow event counter difference of 1.
1917 struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1918 u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1920 if (rdev->desc_nr >= freshest_max_dev) {
1921 /* this is unexpected, better not proceed */
1922 pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1923 mdname(mddev), rdev->bdev, rdev->desc_nr,
1924 freshest->bdev, freshest_max_dev);
1928 role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1929 pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1930 mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1932 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1935 case MD_DISK_ROLE_SPARE: /* spare */
1937 case MD_DISK_ROLE_FAULTY: /* faulty */
1938 set_bit(Faulty, &rdev->flags);
1940 case MD_DISK_ROLE_JOURNAL: /* journal device */
1941 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1942 /* journal device without journal feature */
1943 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1946 set_bit(Journal, &rdev->flags);
1947 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1948 rdev->raid_disk = 0;
1951 rdev->saved_raid_disk = role;
1952 if ((le32_to_cpu(sb->feature_map) &
1953 MD_FEATURE_RECOVERY_OFFSET)) {
1954 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1955 if (!(le32_to_cpu(sb->feature_map) &
1956 MD_FEATURE_RECOVERY_BITMAP))
1957 rdev->saved_raid_disk = -1;
1960 * If the array is FROZEN, then the device can't
1961 * be in_sync with rest of array.
1963 if (!test_bit(MD_RECOVERY_FROZEN,
1965 set_bit(In_sync, &rdev->flags);
1967 rdev->raid_disk = role;
1970 if (sb->devflags & WriteMostly1)
1971 set_bit(WriteMostly, &rdev->flags);
1972 if (sb->devflags & FailFast1)
1973 set_bit(FailFast, &rdev->flags);
1974 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1975 set_bit(Replacement, &rdev->flags);
1976 } else /* MULTIPATH are always insync */
1977 set_bit(In_sync, &rdev->flags);
1982 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1984 struct mdp_superblock_1 *sb;
1985 struct md_rdev *rdev2;
1987 /* make rdev->sb match mddev and rdev data. */
1989 sb = page_address(rdev->sb_page);
1991 sb->feature_map = 0;
1993 sb->recovery_offset = cpu_to_le64(0);
1994 memset(sb->pad3, 0, sizeof(sb->pad3));
1996 sb->utime = cpu_to_le64((__u64)mddev->utime);
1997 sb->events = cpu_to_le64(mddev->events);
1999 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2000 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2001 sb->resync_offset = cpu_to_le64(MaxSector);
2003 sb->resync_offset = cpu_to_le64(0);
2005 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2007 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2008 sb->size = cpu_to_le64(mddev->dev_sectors);
2009 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2010 sb->level = cpu_to_le32(mddev->level);
2011 sb->layout = cpu_to_le32(mddev->layout);
2012 if (test_bit(FailFast, &rdev->flags))
2013 sb->devflags |= FailFast1;
2015 sb->devflags &= ~FailFast1;
2017 if (test_bit(WriteMostly, &rdev->flags))
2018 sb->devflags |= WriteMostly1;
2020 sb->devflags &= ~WriteMostly1;
2021 sb->data_offset = cpu_to_le64(rdev->data_offset);
2022 sb->data_size = cpu_to_le64(rdev->sectors);
2024 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2025 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2026 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2029 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2030 !test_bit(In_sync, &rdev->flags)) {
2032 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2033 sb->recovery_offset =
2034 cpu_to_le64(rdev->recovery_offset);
2035 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2037 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2039 /* Note: recovery_offset and journal_tail share space */
2040 if (test_bit(Journal, &rdev->flags))
2041 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2042 if (test_bit(Replacement, &rdev->flags))
2044 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2046 if (mddev->reshape_position != MaxSector) {
2047 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2048 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2049 sb->new_layout = cpu_to_le32(mddev->new_layout);
2050 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2051 sb->new_level = cpu_to_le32(mddev->new_level);
2052 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2053 if (mddev->delta_disks == 0 &&
2054 mddev->reshape_backwards)
2056 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2057 if (rdev->new_data_offset != rdev->data_offset) {
2059 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2060 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2061 - rdev->data_offset));
2065 if (mddev_is_clustered(mddev))
2066 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2068 if (rdev->badblocks.count == 0)
2069 /* Nothing to do for bad blocks*/ ;
2070 else if (sb->bblog_offset == 0)
2071 /* Cannot record bad blocks on this device */
2072 md_error(mddev, rdev);
2074 struct badblocks *bb = &rdev->badblocks;
2075 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2077 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2082 seq = read_seqbegin(&bb->lock);
2084 memset(bbp, 0xff, PAGE_SIZE);
2086 for (i = 0 ; i < bb->count ; i++) {
2087 u64 internal_bb = p[i];
2088 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2089 | BB_LEN(internal_bb));
2090 bbp[i] = cpu_to_le64(store_bb);
2093 if (read_seqretry(&bb->lock, seq))
2096 bb->sector = (rdev->sb_start +
2097 (int)le32_to_cpu(sb->bblog_offset));
2098 bb->size = le16_to_cpu(sb->bblog_size);
2103 rdev_for_each(rdev2, mddev)
2104 if (rdev2->desc_nr+1 > max_dev)
2105 max_dev = rdev2->desc_nr+1;
2107 if (max_dev > le32_to_cpu(sb->max_dev)) {
2109 sb->max_dev = cpu_to_le32(max_dev);
2110 rdev->sb_size = max_dev * 2 + 256;
2111 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2112 if (rdev->sb_size & bmask)
2113 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2115 max_dev = le32_to_cpu(sb->max_dev);
2117 for (i=0; i<max_dev;i++)
2118 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2120 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2121 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2123 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2124 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2126 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2128 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2129 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2130 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2133 rdev_for_each(rdev2, mddev) {
2135 if (test_bit(Faulty, &rdev2->flags))
2136 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2137 else if (test_bit(In_sync, &rdev2->flags))
2138 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2139 else if (test_bit(Journal, &rdev2->flags))
2140 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2141 else if (rdev2->raid_disk >= 0)
2142 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2144 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2147 sb->sb_csum = calc_sb_1_csum(sb);
2150 static sector_t super_1_choose_bm_space(sector_t dev_size)
2154 /* if the device is bigger than 8Gig, save 64k for bitmap
2155 * usage, if bigger than 200Gig, save 128k
2157 if (dev_size < 64*2)
2159 else if (dev_size - 64*2 >= 200*1024*1024*2)
2161 else if (dev_size - 4*2 > 8*1024*1024*2)
2168 static unsigned long long
2169 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2171 struct mdp_superblock_1 *sb;
2172 sector_t max_sectors;
2173 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2174 return 0; /* component must fit device */
2175 if (rdev->data_offset != rdev->new_data_offset)
2176 return 0; /* too confusing */
2177 if (rdev->sb_start < rdev->data_offset) {
2178 /* minor versions 1 and 2; superblock before data */
2179 max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2180 if (!num_sectors || num_sectors > max_sectors)
2181 num_sectors = max_sectors;
2182 } else if (rdev->mddev->bitmap_info.offset) {
2183 /* minor version 0 with bitmap we can't move */
2186 /* minor version 0; superblock after data */
2187 sector_t sb_start, bm_space;
2188 sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2190 /* 8K is for superblock */
2191 sb_start = dev_size - 8*2;
2192 sb_start &= ~(sector_t)(4*2 - 1);
2194 bm_space = super_1_choose_bm_space(dev_size);
2196 /* Space that can be used to store date needs to decrease
2197 * superblock bitmap space and bad block space(4K)
2199 max_sectors = sb_start - bm_space - 4*2;
2201 if (!num_sectors || num_sectors > max_sectors)
2202 num_sectors = max_sectors;
2203 rdev->sb_start = sb_start;
2205 sb = page_address(rdev->sb_page);
2206 sb->data_size = cpu_to_le64(num_sectors);
2207 sb->super_offset = cpu_to_le64(rdev->sb_start);
2208 sb->sb_csum = calc_sb_1_csum(sb);
2210 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2212 } while (md_super_wait(rdev->mddev) < 0);
2218 super_1_allow_new_offset(struct md_rdev *rdev,
2219 unsigned long long new_offset)
2221 /* All necessary checks on new >= old have been done */
2222 struct bitmap *bitmap;
2223 if (new_offset >= rdev->data_offset)
2226 /* with 1.0 metadata, there is no metadata to tread on
2227 * so we can always move back */
2228 if (rdev->mddev->minor_version == 0)
2231 /* otherwise we must be sure not to step on
2232 * any metadata, so stay:
2233 * 36K beyond start of superblock
2234 * beyond end of badblocks
2235 * beyond write-intent bitmap
2237 if (rdev->sb_start + (32+4)*2 > new_offset)
2239 bitmap = rdev->mddev->bitmap;
2240 if (bitmap && !rdev->mddev->bitmap_info.file &&
2241 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2242 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2244 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2250 static struct super_type super_types[] = {
2253 .owner = THIS_MODULE,
2254 .load_super = super_90_load,
2255 .validate_super = super_90_validate,
2256 .sync_super = super_90_sync,
2257 .rdev_size_change = super_90_rdev_size_change,
2258 .allow_new_offset = super_90_allow_new_offset,
2262 .owner = THIS_MODULE,
2263 .load_super = super_1_load,
2264 .validate_super = super_1_validate,
2265 .sync_super = super_1_sync,
2266 .rdev_size_change = super_1_rdev_size_change,
2267 .allow_new_offset = super_1_allow_new_offset,
2271 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2273 if (mddev->sync_super) {
2274 mddev->sync_super(mddev, rdev);
2278 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2280 super_types[mddev->major_version].sync_super(mddev, rdev);
2283 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2285 struct md_rdev *rdev, *rdev2;
2288 rdev_for_each_rcu(rdev, mddev1) {
2289 if (test_bit(Faulty, &rdev->flags) ||
2290 test_bit(Journal, &rdev->flags) ||
2291 rdev->raid_disk == -1)
2293 rdev_for_each_rcu(rdev2, mddev2) {
2294 if (test_bit(Faulty, &rdev2->flags) ||
2295 test_bit(Journal, &rdev2->flags) ||
2296 rdev2->raid_disk == -1)
2298 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2308 static LIST_HEAD(pending_raid_disks);
2311 * Try to register data integrity profile for an mddev
2313 * This is called when an array is started and after a disk has been kicked
2314 * from the array. It only succeeds if all working and active component devices
2315 * are integrity capable with matching profiles.
2317 int md_integrity_register(struct mddev *mddev)
2319 struct md_rdev *rdev, *reference = NULL;
2321 if (list_empty(&mddev->disks))
2322 return 0; /* nothing to do */
2323 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2324 return 0; /* shouldn't register, or already is */
2325 rdev_for_each(rdev, mddev) {
2326 /* skip spares and non-functional disks */
2327 if (test_bit(Faulty, &rdev->flags))
2329 if (rdev->raid_disk < 0)
2332 /* Use the first rdev as the reference */
2336 /* does this rdev's profile match the reference profile? */
2337 if (blk_integrity_compare(reference->bdev->bd_disk,
2338 rdev->bdev->bd_disk) < 0)
2341 if (!reference || !bdev_get_integrity(reference->bdev))
2344 * All component devices are integrity capable and have matching
2345 * profiles, register the common profile for the md device.
2347 blk_integrity_register(mddev->gendisk,
2348 bdev_get_integrity(reference->bdev));
2350 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2351 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2352 (mddev->level != 1 && mddev->level != 10 &&
2353 bioset_integrity_create(&mddev->io_clone_set, BIO_POOL_SIZE))) {
2355 * No need to handle the failure of bioset_integrity_create,
2356 * because the function is called by md_run() -> pers->run(),
2357 * md_run calls bioset_exit -> bioset_integrity_free in case
2360 pr_err("md: failed to create integrity pool for %s\n",
2366 EXPORT_SYMBOL(md_integrity_register);
2369 * Attempt to add an rdev, but only if it is consistent with the current
2372 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2374 struct blk_integrity *bi_mddev;
2376 if (!mddev->gendisk)
2379 bi_mddev = blk_get_integrity(mddev->gendisk);
2381 if (!bi_mddev) /* nothing to do */
2384 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2385 pr_err("%s: incompatible integrity profile for %pg\n",
2386 mdname(mddev), rdev->bdev);
2392 EXPORT_SYMBOL(md_integrity_add_rdev);
2394 static bool rdev_read_only(struct md_rdev *rdev)
2396 return bdev_read_only(rdev->bdev) ||
2397 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2400 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2402 char b[BDEVNAME_SIZE];
2405 /* prevent duplicates */
2406 if (find_rdev(mddev, rdev->bdev->bd_dev))
2409 if (rdev_read_only(rdev) && mddev->pers)
2412 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2413 if (!test_bit(Journal, &rdev->flags) &&
2415 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2417 /* Cannot change size, so fail
2418 * If mddev->level <= 0, then we don't care
2419 * about aligning sizes (e.g. linear)
2421 if (mddev->level > 0)
2424 mddev->dev_sectors = rdev->sectors;
2427 /* Verify rdev->desc_nr is unique.
2428 * If it is -1, assign a free number, else
2429 * check number is not in use
2432 if (rdev->desc_nr < 0) {
2435 choice = mddev->raid_disks;
2436 while (md_find_rdev_nr_rcu(mddev, choice))
2438 rdev->desc_nr = choice;
2440 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2446 if (!test_bit(Journal, &rdev->flags) &&
2447 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2448 pr_warn("md: %s: array is limited to %d devices\n",
2449 mdname(mddev), mddev->max_disks);
2452 snprintf(b, sizeof(b), "%pg", rdev->bdev);
2453 strreplace(b, '/', '!');
2455 rdev->mddev = mddev;
2456 pr_debug("md: bind<%s>\n", b);
2458 if (mddev->raid_disks)
2459 mddev_create_serial_pool(mddev, rdev, false);
2461 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2464 /* failure here is OK */
2465 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2466 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2467 rdev->sysfs_unack_badblocks =
2468 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2469 rdev->sysfs_badblocks =
2470 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2472 list_add_rcu(&rdev->same_set, &mddev->disks);
2473 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2475 /* May as well allow recovery to be retried once */
2476 mddev->recovery_disabled++;
2481 pr_warn("md: failed to register dev-%s for %s\n",
2486 void md_autodetect_dev(dev_t dev);
2488 /* just for claiming the bdev */
2489 static struct md_rdev claim_rdev;
2491 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2493 pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2494 md_rdev_clear(rdev);
2496 if (test_bit(AutoDetected, &rdev->flags))
2497 md_autodetect_dev(rdev->bdev->bd_dev);
2499 blkdev_put(rdev->bdev,
2500 test_bit(Holder, &rdev->flags) ? rdev : &claim_rdev);
2502 kobject_put(&rdev->kobj);
2505 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2507 struct mddev *mddev = rdev->mddev;
2509 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2510 list_del_rcu(&rdev->same_set);
2511 pr_debug("md: unbind<%pg>\n", rdev->bdev);
2512 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2514 sysfs_remove_link(&rdev->kobj, "block");
2515 sysfs_put(rdev->sysfs_state);
2516 sysfs_put(rdev->sysfs_unack_badblocks);
2517 sysfs_put(rdev->sysfs_badblocks);
2518 rdev->sysfs_state = NULL;
2519 rdev->sysfs_unack_badblocks = NULL;
2520 rdev->sysfs_badblocks = NULL;
2521 rdev->badblocks.count = 0;
2526 * kobject_del() will wait for all in progress writers to be done, where
2527 * reconfig_mutex is held, hence it can't be called under
2528 * reconfig_mutex and it's delayed to mddev_unlock().
2530 list_add(&rdev->same_set, &mddev->deleting);
2533 static void export_array(struct mddev *mddev)
2535 struct md_rdev *rdev;
2537 while (!list_empty(&mddev->disks)) {
2538 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2540 md_kick_rdev_from_array(rdev);
2542 mddev->raid_disks = 0;
2543 mddev->major_version = 0;
2546 static bool set_in_sync(struct mddev *mddev)
2548 lockdep_assert_held(&mddev->lock);
2549 if (!mddev->in_sync) {
2550 mddev->sync_checkers++;
2551 spin_unlock(&mddev->lock);
2552 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2553 spin_lock(&mddev->lock);
2554 if (!mddev->in_sync &&
2555 percpu_ref_is_zero(&mddev->writes_pending)) {
2558 * Ensure ->in_sync is visible before we clear
2562 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2563 sysfs_notify_dirent_safe(mddev->sysfs_state);
2565 if (--mddev->sync_checkers == 0)
2566 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2568 if (mddev->safemode == 1)
2569 mddev->safemode = 0;
2570 return mddev->in_sync;
2573 static void sync_sbs(struct mddev *mddev, int nospares)
2575 /* Update each superblock (in-memory image), but
2576 * if we are allowed to, skip spares which already
2577 * have the right event counter, or have one earlier
2578 * (which would mean they aren't being marked as dirty
2579 * with the rest of the array)
2581 struct md_rdev *rdev;
2582 rdev_for_each(rdev, mddev) {
2583 if (rdev->sb_events == mddev->events ||
2585 rdev->raid_disk < 0 &&
2586 rdev->sb_events+1 == mddev->events)) {
2587 /* Don't update this superblock */
2588 rdev->sb_loaded = 2;
2590 sync_super(mddev, rdev);
2591 rdev->sb_loaded = 1;
2596 static bool does_sb_need_changing(struct mddev *mddev)
2598 struct md_rdev *rdev = NULL, *iter;
2599 struct mdp_superblock_1 *sb;
2602 /* Find a good rdev */
2603 rdev_for_each(iter, mddev)
2604 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2609 /* No good device found. */
2613 sb = page_address(rdev->sb_page);
2614 /* Check if a device has become faulty or a spare become active */
2615 rdev_for_each(rdev, mddev) {
2616 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2617 /* Device activated? */
2618 if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2619 !test_bit(Faulty, &rdev->flags))
2621 /* Device turned faulty? */
2622 if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2626 /* Check if any mddev parameters have changed */
2627 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2628 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2629 (mddev->layout != le32_to_cpu(sb->layout)) ||
2630 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2631 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2637 void md_update_sb(struct mddev *mddev, int force_change)
2639 struct md_rdev *rdev;
2642 int any_badblocks_changed = 0;
2645 if (!md_is_rdwr(mddev)) {
2647 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2652 if (mddev_is_clustered(mddev)) {
2653 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2655 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2657 ret = md_cluster_ops->metadata_update_start(mddev);
2658 /* Has someone else has updated the sb */
2659 if (!does_sb_need_changing(mddev)) {
2661 md_cluster_ops->metadata_update_cancel(mddev);
2662 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2663 BIT(MD_SB_CHANGE_DEVS) |
2664 BIT(MD_SB_CHANGE_CLEAN));
2670 * First make sure individual recovery_offsets are correct
2671 * curr_resync_completed can only be used during recovery.
2672 * During reshape/resync it might use array-addresses rather
2673 * that device addresses.
2675 rdev_for_each(rdev, mddev) {
2676 if (rdev->raid_disk >= 0 &&
2677 mddev->delta_disks >= 0 &&
2678 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2679 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2680 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2681 !test_bit(Journal, &rdev->flags) &&
2682 !test_bit(In_sync, &rdev->flags) &&
2683 mddev->curr_resync_completed > rdev->recovery_offset)
2684 rdev->recovery_offset = mddev->curr_resync_completed;
2687 if (!mddev->persistent) {
2688 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2689 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2690 if (!mddev->external) {
2691 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2692 rdev_for_each(rdev, mddev) {
2693 if (rdev->badblocks.changed) {
2694 rdev->badblocks.changed = 0;
2695 ack_all_badblocks(&rdev->badblocks);
2696 md_error(mddev, rdev);
2698 clear_bit(Blocked, &rdev->flags);
2699 clear_bit(BlockedBadBlocks, &rdev->flags);
2700 wake_up(&rdev->blocked_wait);
2703 wake_up(&mddev->sb_wait);
2707 spin_lock(&mddev->lock);
2709 mddev->utime = ktime_get_real_seconds();
2711 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2713 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2714 /* just a clean<-> dirty transition, possibly leave spares alone,
2715 * though if events isn't the right even/odd, we will have to do
2721 if (mddev->degraded)
2722 /* If the array is degraded, then skipping spares is both
2723 * dangerous and fairly pointless.
2724 * Dangerous because a device that was removed from the array
2725 * might have a event_count that still looks up-to-date,
2726 * so it can be re-added without a resync.
2727 * Pointless because if there are any spares to skip,
2728 * then a recovery will happen and soon that array won't
2729 * be degraded any more and the spare can go back to sleep then.
2733 sync_req = mddev->in_sync;
2735 /* If this is just a dirty<->clean transition, and the array is clean
2736 * and 'events' is odd, we can roll back to the previous clean state */
2738 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2739 && mddev->can_decrease_events
2740 && mddev->events != 1) {
2742 mddev->can_decrease_events = 0;
2744 /* otherwise we have to go forward and ... */
2746 mddev->can_decrease_events = nospares;
2750 * This 64-bit counter should never wrap.
2751 * Either we are in around ~1 trillion A.C., assuming
2752 * 1 reboot per second, or we have a bug...
2754 WARN_ON(mddev->events == 0);
2756 rdev_for_each(rdev, mddev) {
2757 if (rdev->badblocks.changed)
2758 any_badblocks_changed++;
2759 if (test_bit(Faulty, &rdev->flags))
2760 set_bit(FaultRecorded, &rdev->flags);
2763 sync_sbs(mddev, nospares);
2764 spin_unlock(&mddev->lock);
2766 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2767 mdname(mddev), mddev->in_sync);
2770 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2772 md_bitmap_update_sb(mddev->bitmap);
2773 rdev_for_each(rdev, mddev) {
2774 if (rdev->sb_loaded != 1)
2775 continue; /* no noise on spare devices */
2777 if (!test_bit(Faulty, &rdev->flags)) {
2778 md_super_write(mddev,rdev,
2779 rdev->sb_start, rdev->sb_size,
2781 pr_debug("md: (write) %pg's sb offset: %llu\n",
2783 (unsigned long long)rdev->sb_start);
2784 rdev->sb_events = mddev->events;
2785 if (rdev->badblocks.size) {
2786 md_super_write(mddev, rdev,
2787 rdev->badblocks.sector,
2788 rdev->badblocks.size << 9,
2790 rdev->badblocks.size = 0;
2794 pr_debug("md: %pg (skipping faulty)\n",
2797 if (mddev->level == LEVEL_MULTIPATH)
2798 /* only need to write one superblock... */
2801 if (md_super_wait(mddev) < 0)
2803 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2805 if (mddev_is_clustered(mddev) && ret == 0)
2806 md_cluster_ops->metadata_update_finish(mddev);
2808 if (mddev->in_sync != sync_req ||
2809 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2810 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2811 /* have to write it out again */
2813 wake_up(&mddev->sb_wait);
2814 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2815 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2817 rdev_for_each(rdev, mddev) {
2818 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2819 clear_bit(Blocked, &rdev->flags);
2821 if (any_badblocks_changed)
2822 ack_all_badblocks(&rdev->badblocks);
2823 clear_bit(BlockedBadBlocks, &rdev->flags);
2824 wake_up(&rdev->blocked_wait);
2827 EXPORT_SYMBOL(md_update_sb);
2829 static int add_bound_rdev(struct md_rdev *rdev)
2831 struct mddev *mddev = rdev->mddev;
2833 bool add_journal = test_bit(Journal, &rdev->flags);
2835 if (!mddev->pers->hot_remove_disk || add_journal) {
2836 /* If there is hot_add_disk but no hot_remove_disk
2837 * then added disks for geometry changes,
2838 * and should be added immediately.
2840 super_types[mddev->major_version].
2841 validate_super(mddev, NULL/*freshest*/, rdev);
2843 mddev_suspend(mddev);
2844 err = mddev->pers->hot_add_disk(mddev, rdev);
2846 mddev_resume(mddev);
2848 md_kick_rdev_from_array(rdev);
2852 sysfs_notify_dirent_safe(rdev->sysfs_state);
2854 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2855 if (mddev->degraded)
2856 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2857 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2859 md_wakeup_thread(mddev->thread);
2863 /* words written to sysfs files may, or may not, be \n terminated.
2864 * We want to accept with case. For this we use cmd_match.
2866 static int cmd_match(const char *cmd, const char *str)
2868 /* See if cmd, written into a sysfs file, matches
2869 * str. They must either be the same, or cmd can
2870 * have a trailing newline
2872 while (*cmd && *str && *cmd == *str) {
2883 struct rdev_sysfs_entry {
2884 struct attribute attr;
2885 ssize_t (*show)(struct md_rdev *, char *);
2886 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2890 state_show(struct md_rdev *rdev, char *page)
2894 unsigned long flags = READ_ONCE(rdev->flags);
2896 if (test_bit(Faulty, &flags) ||
2897 (!test_bit(ExternalBbl, &flags) &&
2898 rdev->badblocks.unacked_exist))
2899 len += sprintf(page+len, "faulty%s", sep);
2900 if (test_bit(In_sync, &flags))
2901 len += sprintf(page+len, "in_sync%s", sep);
2902 if (test_bit(Journal, &flags))
2903 len += sprintf(page+len, "journal%s", sep);
2904 if (test_bit(WriteMostly, &flags))
2905 len += sprintf(page+len, "write_mostly%s", sep);
2906 if (test_bit(Blocked, &flags) ||
2907 (rdev->badblocks.unacked_exist
2908 && !test_bit(Faulty, &flags)))
2909 len += sprintf(page+len, "blocked%s", sep);
2910 if (!test_bit(Faulty, &flags) &&
2911 !test_bit(Journal, &flags) &&
2912 !test_bit(In_sync, &flags))
2913 len += sprintf(page+len, "spare%s", sep);
2914 if (test_bit(WriteErrorSeen, &flags))
2915 len += sprintf(page+len, "write_error%s", sep);
2916 if (test_bit(WantReplacement, &flags))
2917 len += sprintf(page+len, "want_replacement%s", sep);
2918 if (test_bit(Replacement, &flags))
2919 len += sprintf(page+len, "replacement%s", sep);
2920 if (test_bit(ExternalBbl, &flags))
2921 len += sprintf(page+len, "external_bbl%s", sep);
2922 if (test_bit(FailFast, &flags))
2923 len += sprintf(page+len, "failfast%s", sep);
2928 return len+sprintf(page+len, "\n");
2932 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2935 * faulty - simulates an error
2936 * remove - disconnects the device
2937 * writemostly - sets write_mostly
2938 * -writemostly - clears write_mostly
2939 * blocked - sets the Blocked flags
2940 * -blocked - clears the Blocked and possibly simulates an error
2941 * insync - sets Insync providing device isn't active
2942 * -insync - clear Insync for a device with a slot assigned,
2943 * so that it gets rebuilt based on bitmap
2944 * write_error - sets WriteErrorSeen
2945 * -write_error - clears WriteErrorSeen
2946 * {,-}failfast - set/clear FailFast
2949 struct mddev *mddev = rdev->mddev;
2951 bool need_update_sb = false;
2953 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2954 md_error(rdev->mddev, rdev);
2956 if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2960 } else if (cmd_match(buf, "remove")) {
2961 if (rdev->mddev->pers) {
2962 clear_bit(Blocked, &rdev->flags);
2963 remove_and_add_spares(rdev->mddev, rdev);
2965 if (rdev->raid_disk >= 0)
2969 if (mddev_is_clustered(mddev))
2970 err = md_cluster_ops->remove_disk(mddev, rdev);
2973 md_kick_rdev_from_array(rdev);
2975 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2976 md_wakeup_thread(mddev->thread);
2981 } else if (cmd_match(buf, "writemostly")) {
2982 set_bit(WriteMostly, &rdev->flags);
2983 mddev_create_serial_pool(rdev->mddev, rdev, false);
2984 need_update_sb = true;
2986 } else if (cmd_match(buf, "-writemostly")) {
2987 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2988 clear_bit(WriteMostly, &rdev->flags);
2989 need_update_sb = true;
2991 } else if (cmd_match(buf, "blocked")) {
2992 set_bit(Blocked, &rdev->flags);
2994 } else if (cmd_match(buf, "-blocked")) {
2995 if (!test_bit(Faulty, &rdev->flags) &&
2996 !test_bit(ExternalBbl, &rdev->flags) &&
2997 rdev->badblocks.unacked_exist) {
2998 /* metadata handler doesn't understand badblocks,
2999 * so we need to fail the device
3001 md_error(rdev->mddev, rdev);
3003 clear_bit(Blocked, &rdev->flags);
3004 clear_bit(BlockedBadBlocks, &rdev->flags);
3005 wake_up(&rdev->blocked_wait);
3006 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3007 md_wakeup_thread(rdev->mddev->thread);
3010 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3011 set_bit(In_sync, &rdev->flags);
3013 } else if (cmd_match(buf, "failfast")) {
3014 set_bit(FailFast, &rdev->flags);
3015 need_update_sb = true;
3017 } else if (cmd_match(buf, "-failfast")) {
3018 clear_bit(FailFast, &rdev->flags);
3019 need_update_sb = true;
3021 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3022 !test_bit(Journal, &rdev->flags)) {
3023 if (rdev->mddev->pers == NULL) {
3024 clear_bit(In_sync, &rdev->flags);
3025 rdev->saved_raid_disk = rdev->raid_disk;
3026 rdev->raid_disk = -1;
3029 } else if (cmd_match(buf, "write_error")) {
3030 set_bit(WriteErrorSeen, &rdev->flags);
3032 } else if (cmd_match(buf, "-write_error")) {
3033 clear_bit(WriteErrorSeen, &rdev->flags);
3035 } else if (cmd_match(buf, "want_replacement")) {
3036 /* Any non-spare device that is not a replacement can
3037 * become want_replacement at any time, but we then need to
3038 * check if recovery is needed.
3040 if (rdev->raid_disk >= 0 &&
3041 !test_bit(Journal, &rdev->flags) &&
3042 !test_bit(Replacement, &rdev->flags))
3043 set_bit(WantReplacement, &rdev->flags);
3044 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3045 md_wakeup_thread(rdev->mddev->thread);
3047 } else if (cmd_match(buf, "-want_replacement")) {
3048 /* Clearing 'want_replacement' is always allowed.
3049 * Once replacements starts it is too late though.
3052 clear_bit(WantReplacement, &rdev->flags);
3053 } else if (cmd_match(buf, "replacement")) {
3054 /* Can only set a device as a replacement when array has not
3055 * yet been started. Once running, replacement is automatic
3056 * from spares, or by assigning 'slot'.
3058 if (rdev->mddev->pers)
3061 set_bit(Replacement, &rdev->flags);
3064 } else if (cmd_match(buf, "-replacement")) {
3065 /* Similarly, can only clear Replacement before start */
3066 if (rdev->mddev->pers)
3069 clear_bit(Replacement, &rdev->flags);
3072 } else if (cmd_match(buf, "re-add")) {
3073 if (!rdev->mddev->pers)
3075 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3076 rdev->saved_raid_disk >= 0) {
3077 /* clear_bit is performed _after_ all the devices
3078 * have their local Faulty bit cleared. If any writes
3079 * happen in the meantime in the local node, they
3080 * will land in the local bitmap, which will be synced
3081 * by this node eventually
3083 if (!mddev_is_clustered(rdev->mddev) ||
3084 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3085 clear_bit(Faulty, &rdev->flags);
3086 err = add_bound_rdev(rdev);
3090 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3091 set_bit(ExternalBbl, &rdev->flags);
3092 rdev->badblocks.shift = 0;
3094 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3095 clear_bit(ExternalBbl, &rdev->flags);
3099 md_update_sb(mddev, 1);
3101 sysfs_notify_dirent_safe(rdev->sysfs_state);
3102 return err ? err : len;
3104 static struct rdev_sysfs_entry rdev_state =
3105 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3108 errors_show(struct md_rdev *rdev, char *page)
3110 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3114 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3119 rv = kstrtouint(buf, 10, &n);
3122 atomic_set(&rdev->corrected_errors, n);
3125 static struct rdev_sysfs_entry rdev_errors =
3126 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3129 slot_show(struct md_rdev *rdev, char *page)
3131 if (test_bit(Journal, &rdev->flags))
3132 return sprintf(page, "journal\n");
3133 else if (rdev->raid_disk < 0)
3134 return sprintf(page, "none\n");
3136 return sprintf(page, "%d\n", rdev->raid_disk);
3140 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3145 if (test_bit(Journal, &rdev->flags))
3147 if (strncmp(buf, "none", 4)==0)
3150 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3157 if (rdev->mddev->pers && slot == -1) {
3158 /* Setting 'slot' on an active array requires also
3159 * updating the 'rd%d' link, and communicating
3160 * with the personality with ->hot_*_disk.
3161 * For now we only support removing
3162 * failed/spare devices. This normally happens automatically,
3163 * but not when the metadata is externally managed.
3165 if (rdev->raid_disk == -1)
3167 /* personality does all needed checks */
3168 if (rdev->mddev->pers->hot_remove_disk == NULL)
3170 clear_bit(Blocked, &rdev->flags);
3171 remove_and_add_spares(rdev->mddev, rdev);
3172 if (rdev->raid_disk >= 0)
3174 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3175 md_wakeup_thread(rdev->mddev->thread);
3176 } else if (rdev->mddev->pers) {
3177 /* Activating a spare .. or possibly reactivating
3178 * if we ever get bitmaps working here.
3182 if (rdev->raid_disk != -1)
3185 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3188 if (rdev->mddev->pers->hot_add_disk == NULL)
3191 if (slot >= rdev->mddev->raid_disks &&
3192 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3195 rdev->raid_disk = slot;
3196 if (test_bit(In_sync, &rdev->flags))
3197 rdev->saved_raid_disk = slot;
3199 rdev->saved_raid_disk = -1;
3200 clear_bit(In_sync, &rdev->flags);
3201 clear_bit(Bitmap_sync, &rdev->flags);
3202 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3204 rdev->raid_disk = -1;
3207 sysfs_notify_dirent_safe(rdev->sysfs_state);
3208 /* failure here is OK */;
3209 sysfs_link_rdev(rdev->mddev, rdev);
3210 /* don't wakeup anyone, leave that to userspace. */
3212 if (slot >= rdev->mddev->raid_disks &&
3213 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3215 rdev->raid_disk = slot;
3216 /* assume it is working */
3217 clear_bit(Faulty, &rdev->flags);
3218 clear_bit(WriteMostly, &rdev->flags);
3219 set_bit(In_sync, &rdev->flags);
3220 sysfs_notify_dirent_safe(rdev->sysfs_state);
3225 static struct rdev_sysfs_entry rdev_slot =
3226 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3229 offset_show(struct md_rdev *rdev, char *page)
3231 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3235 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3237 unsigned long long offset;
3238 if (kstrtoull(buf, 10, &offset) < 0)
3240 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3242 if (rdev->sectors && rdev->mddev->external)
3243 /* Must set offset before size, so overlap checks
3246 rdev->data_offset = offset;
3247 rdev->new_data_offset = offset;
3251 static struct rdev_sysfs_entry rdev_offset =
3252 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3254 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3256 return sprintf(page, "%llu\n",
3257 (unsigned long long)rdev->new_data_offset);
3260 static ssize_t new_offset_store(struct md_rdev *rdev,
3261 const char *buf, size_t len)
3263 unsigned long long new_offset;
3264 struct mddev *mddev = rdev->mddev;
3266 if (kstrtoull(buf, 10, &new_offset) < 0)
3269 if (mddev->sync_thread ||
3270 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3272 if (new_offset == rdev->data_offset)
3273 /* reset is always permitted */
3275 else if (new_offset > rdev->data_offset) {
3276 /* must not push array size beyond rdev_sectors */
3277 if (new_offset - rdev->data_offset
3278 + mddev->dev_sectors > rdev->sectors)
3281 /* Metadata worries about other space details. */
3283 /* decreasing the offset is inconsistent with a backwards
3286 if (new_offset < rdev->data_offset &&
3287 mddev->reshape_backwards)
3289 /* Increasing offset is inconsistent with forwards
3290 * reshape. reshape_direction should be set to
3291 * 'backwards' first.
3293 if (new_offset > rdev->data_offset &&
3294 !mddev->reshape_backwards)
3297 if (mddev->pers && mddev->persistent &&
3298 !super_types[mddev->major_version]
3299 .allow_new_offset(rdev, new_offset))
3301 rdev->new_data_offset = new_offset;
3302 if (new_offset > rdev->data_offset)
3303 mddev->reshape_backwards = 1;
3304 else if (new_offset < rdev->data_offset)
3305 mddev->reshape_backwards = 0;
3309 static struct rdev_sysfs_entry rdev_new_offset =
3310 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3313 rdev_size_show(struct md_rdev *rdev, char *page)
3315 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3318 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3320 /* check if two start/length pairs overlap */
3321 if (a->data_offset + a->sectors <= b->data_offset)
3323 if (b->data_offset + b->sectors <= a->data_offset)
3328 static bool md_rdev_overlaps(struct md_rdev *rdev)
3330 struct mddev *mddev;
3331 struct md_rdev *rdev2;
3333 spin_lock(&all_mddevs_lock);
3334 list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3335 if (test_bit(MD_DELETED, &mddev->flags))
3337 rdev_for_each(rdev2, mddev) {
3338 if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3339 md_rdevs_overlap(rdev, rdev2)) {
3340 spin_unlock(&all_mddevs_lock);
3345 spin_unlock(&all_mddevs_lock);
3349 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3351 unsigned long long blocks;
3354 if (kstrtoull(buf, 10, &blocks) < 0)
3357 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3358 return -EINVAL; /* sector conversion overflow */
3361 if (new != blocks * 2)
3362 return -EINVAL; /* unsigned long long to sector_t overflow */
3369 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3371 struct mddev *my_mddev = rdev->mddev;
3372 sector_t oldsectors = rdev->sectors;
3375 if (test_bit(Journal, &rdev->flags))
3377 if (strict_blocks_to_sectors(buf, §ors) < 0)
3379 if (rdev->data_offset != rdev->new_data_offset)
3380 return -EINVAL; /* too confusing */
3381 if (my_mddev->pers && rdev->raid_disk >= 0) {
3382 if (my_mddev->persistent) {
3383 sectors = super_types[my_mddev->major_version].
3384 rdev_size_change(rdev, sectors);
3387 } else if (!sectors)
3388 sectors = bdev_nr_sectors(rdev->bdev) -
3390 if (!my_mddev->pers->resize)
3391 /* Cannot change size for RAID0 or Linear etc */
3394 if (sectors < my_mddev->dev_sectors)
3395 return -EINVAL; /* component must fit device */
3397 rdev->sectors = sectors;
3400 * Check that all other rdevs with the same bdev do not overlap. This
3401 * check does not provide a hard guarantee, it just helps avoid
3402 * dangerous mistakes.
3404 if (sectors > oldsectors && my_mddev->external &&
3405 md_rdev_overlaps(rdev)) {
3407 * Someone else could have slipped in a size change here, but
3408 * doing so is just silly. We put oldsectors back because we
3409 * know it is safe, and trust userspace not to race with itself.
3411 rdev->sectors = oldsectors;
3417 static struct rdev_sysfs_entry rdev_size =
3418 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3420 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3422 unsigned long long recovery_start = rdev->recovery_offset;
3424 if (test_bit(In_sync, &rdev->flags) ||
3425 recovery_start == MaxSector)
3426 return sprintf(page, "none\n");
3428 return sprintf(page, "%llu\n", recovery_start);
3431 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3433 unsigned long long recovery_start;
3435 if (cmd_match(buf, "none"))
3436 recovery_start = MaxSector;
3437 else if (kstrtoull(buf, 10, &recovery_start))
3440 if (rdev->mddev->pers &&
3441 rdev->raid_disk >= 0)
3444 rdev->recovery_offset = recovery_start;
3445 if (recovery_start == MaxSector)
3446 set_bit(In_sync, &rdev->flags);
3448 clear_bit(In_sync, &rdev->flags);
3452 static struct rdev_sysfs_entry rdev_recovery_start =
3453 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3455 /* sysfs access to bad-blocks list.
3456 * We present two files.
3457 * 'bad-blocks' lists sector numbers and lengths of ranges that
3458 * are recorded as bad. The list is truncated to fit within
3459 * the one-page limit of sysfs.
3460 * Writing "sector length" to this file adds an acknowledged
3462 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3463 * been acknowledged. Writing to this file adds bad blocks
3464 * without acknowledging them. This is largely for testing.
3466 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3468 return badblocks_show(&rdev->badblocks, page, 0);
3470 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3472 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3473 /* Maybe that ack was all we needed */
3474 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3475 wake_up(&rdev->blocked_wait);
3478 static struct rdev_sysfs_entry rdev_bad_blocks =
3479 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3481 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3483 return badblocks_show(&rdev->badblocks, page, 1);
3485 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3487 return badblocks_store(&rdev->badblocks, page, len, 1);
3489 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3490 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3493 ppl_sector_show(struct md_rdev *rdev, char *page)
3495 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3499 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3501 unsigned long long sector;
3503 if (kstrtoull(buf, 10, §or) < 0)
3505 if (sector != (sector_t)sector)
3508 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3509 rdev->raid_disk >= 0)
3512 if (rdev->mddev->persistent) {
3513 if (rdev->mddev->major_version == 0)
3515 if ((sector > rdev->sb_start &&
3516 sector - rdev->sb_start > S16_MAX) ||
3517 (sector < rdev->sb_start &&
3518 rdev->sb_start - sector > -S16_MIN))
3520 rdev->ppl.offset = sector - rdev->sb_start;
3521 } else if (!rdev->mddev->external) {
3524 rdev->ppl.sector = sector;
3528 static struct rdev_sysfs_entry rdev_ppl_sector =
3529 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3532 ppl_size_show(struct md_rdev *rdev, char *page)
3534 return sprintf(page, "%u\n", rdev->ppl.size);
3538 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3542 if (kstrtouint(buf, 10, &size) < 0)
3545 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3546 rdev->raid_disk >= 0)
3549 if (rdev->mddev->persistent) {
3550 if (rdev->mddev->major_version == 0)
3554 } else if (!rdev->mddev->external) {
3557 rdev->ppl.size = size;
3561 static struct rdev_sysfs_entry rdev_ppl_size =
3562 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3564 static struct attribute *rdev_default_attrs[] = {
3569 &rdev_new_offset.attr,
3571 &rdev_recovery_start.attr,
3572 &rdev_bad_blocks.attr,
3573 &rdev_unack_bad_blocks.attr,
3574 &rdev_ppl_sector.attr,
3575 &rdev_ppl_size.attr,
3578 ATTRIBUTE_GROUPS(rdev_default);
3580 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3582 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3583 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3589 return entry->show(rdev, page);
3593 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3594 const char *page, size_t length)
3596 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3597 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3598 struct kernfs_node *kn = NULL;
3600 struct mddev *mddev = rdev->mddev;
3604 if (!capable(CAP_SYS_ADMIN))
3607 if (entry->store == state_store && cmd_match(page, "remove"))
3608 kn = sysfs_break_active_protection(kobj, attr);
3610 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3612 if (rdev->mddev == NULL)
3615 rv = entry->store(rdev, page, length);
3616 mddev_unlock(mddev);
3620 sysfs_unbreak_active_protection(kn);
3625 static void rdev_free(struct kobject *ko)
3627 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3630 static const struct sysfs_ops rdev_sysfs_ops = {
3631 .show = rdev_attr_show,
3632 .store = rdev_attr_store,
3634 static const struct kobj_type rdev_ktype = {
3635 .release = rdev_free,
3636 .sysfs_ops = &rdev_sysfs_ops,
3637 .default_groups = rdev_default_groups,
3640 int md_rdev_init(struct md_rdev *rdev)
3643 rdev->saved_raid_disk = -1;
3644 rdev->raid_disk = -1;
3646 rdev->data_offset = 0;
3647 rdev->new_data_offset = 0;
3648 rdev->sb_events = 0;
3649 rdev->last_read_error = 0;
3650 rdev->sb_loaded = 0;
3651 rdev->bb_page = NULL;
3652 atomic_set(&rdev->nr_pending, 0);
3653 atomic_set(&rdev->read_errors, 0);
3654 atomic_set(&rdev->corrected_errors, 0);
3656 INIT_LIST_HEAD(&rdev->same_set);
3657 init_waitqueue_head(&rdev->blocked_wait);
3659 /* Add space to store bad block list.
3660 * This reserves the space even on arrays where it cannot
3661 * be used - I wonder if that matters
3663 return badblocks_init(&rdev->badblocks, 0);
3665 EXPORT_SYMBOL_GPL(md_rdev_init);
3668 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3670 * mark the device faulty if:
3672 * - the device is nonexistent (zero size)
3673 * - the device has no valid superblock
3675 * a faulty rdev _never_ has rdev->sb set.
3677 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3679 struct md_rdev *rdev;
3680 struct md_rdev *holder;
3684 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3686 return ERR_PTR(-ENOMEM);
3688 err = md_rdev_init(rdev);
3691 err = alloc_disk_sb(rdev);
3693 goto out_clear_rdev;
3695 if (super_format == -2) {
3696 holder = &claim_rdev;
3699 set_bit(Holder, &rdev->flags);
3702 rdev->bdev = blkdev_get_by_dev(newdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
3704 if (IS_ERR(rdev->bdev)) {
3705 pr_warn("md: could not open device unknown-block(%u,%u).\n",
3706 MAJOR(newdev), MINOR(newdev));
3707 err = PTR_ERR(rdev->bdev);
3708 goto out_clear_rdev;
3711 kobject_init(&rdev->kobj, &rdev_ktype);
3713 size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3715 pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3718 goto out_blkdev_put;
3721 if (super_format >= 0) {
3722 err = super_types[super_format].
3723 load_super(rdev, NULL, super_minor);
3724 if (err == -EINVAL) {
3725 pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3727 super_format, super_minor);
3728 goto out_blkdev_put;
3731 pr_warn("md: could not read %pg's sb, not importing!\n",
3733 goto out_blkdev_put;
3740 blkdev_put(rdev->bdev, holder);
3742 md_rdev_clear(rdev);
3745 return ERR_PTR(err);
3749 * Check a full RAID array for plausibility
3752 static int analyze_sbs(struct mddev *mddev)
3755 struct md_rdev *rdev, *freshest, *tmp;
3758 rdev_for_each_safe(rdev, tmp, mddev)
3759 switch (super_types[mddev->major_version].
3760 load_super(rdev, freshest, mddev->minor_version)) {
3767 pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3769 md_kick_rdev_from_array(rdev);
3772 /* Cannot find a valid fresh disk */
3774 pr_warn("md: cannot find a valid disk\n");
3778 super_types[mddev->major_version].
3779 validate_super(mddev, NULL/*freshest*/, freshest);
3782 rdev_for_each_safe(rdev, tmp, mddev) {
3783 if (mddev->max_disks &&
3784 (rdev->desc_nr >= mddev->max_disks ||
3785 i > mddev->max_disks)) {
3786 pr_warn("md: %s: %pg: only %d devices permitted\n",
3787 mdname(mddev), rdev->bdev,
3789 md_kick_rdev_from_array(rdev);
3792 if (rdev != freshest) {
3793 if (super_types[mddev->major_version].
3794 validate_super(mddev, freshest, rdev)) {
3795 pr_warn("md: kicking non-fresh %pg from array!\n",
3797 md_kick_rdev_from_array(rdev);
3801 if (mddev->level == LEVEL_MULTIPATH) {
3802 rdev->desc_nr = i++;
3803 rdev->raid_disk = rdev->desc_nr;
3804 set_bit(In_sync, &rdev->flags);
3805 } else if (rdev->raid_disk >=
3806 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3807 !test_bit(Journal, &rdev->flags)) {
3808 rdev->raid_disk = -1;
3809 clear_bit(In_sync, &rdev->flags);
3816 /* Read a fixed-point number.
3817 * Numbers in sysfs attributes should be in "standard" units where
3818 * possible, so time should be in seconds.
3819 * However we internally use a a much smaller unit such as
3820 * milliseconds or jiffies.
3821 * This function takes a decimal number with a possible fractional
3822 * component, and produces an integer which is the result of
3823 * multiplying that number by 10^'scale'.
3824 * all without any floating-point arithmetic.
3826 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3828 unsigned long result = 0;
3830 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3833 else if (decimals < scale) {
3836 result = result * 10 + value;
3848 *res = result * int_pow(10, scale - decimals);
3853 safe_delay_show(struct mddev *mddev, char *page)
3855 unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3857 return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3860 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3864 if (mddev_is_clustered(mddev)) {
3865 pr_warn("md: Safemode is disabled for clustered mode\n");
3869 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3872 mddev->safemode_delay = 0;
3874 unsigned long old_delay = mddev->safemode_delay;
3875 unsigned long new_delay = (msec*HZ)/1000;
3879 mddev->safemode_delay = new_delay;
3880 if (new_delay < old_delay || old_delay == 0)
3881 mod_timer(&mddev->safemode_timer, jiffies+1);
3885 static struct md_sysfs_entry md_safe_delay =
3886 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3889 level_show(struct mddev *mddev, char *page)
3891 struct md_personality *p;
3893 spin_lock(&mddev->lock);
3896 ret = sprintf(page, "%s\n", p->name);
3897 else if (mddev->clevel[0])
3898 ret = sprintf(page, "%s\n", mddev->clevel);
3899 else if (mddev->level != LEVEL_NONE)
3900 ret = sprintf(page, "%d\n", mddev->level);
3903 spin_unlock(&mddev->lock);
3908 level_store(struct mddev *mddev, const char *buf, size_t len)
3913 struct md_personality *pers, *oldpers;
3915 void *priv, *oldpriv;
3916 struct md_rdev *rdev;
3918 if (slen == 0 || slen >= sizeof(clevel))
3921 rv = mddev_lock(mddev);
3925 if (mddev->pers == NULL) {
3926 strncpy(mddev->clevel, buf, slen);
3927 if (mddev->clevel[slen-1] == '\n')
3929 mddev->clevel[slen] = 0;
3930 mddev->level = LEVEL_NONE;
3935 if (!md_is_rdwr(mddev))
3938 /* request to change the personality. Need to ensure:
3939 * - array is not engaged in resync/recovery/reshape
3940 * - old personality can be suspended
3941 * - new personality will access other array.
3945 if (mddev->sync_thread ||
3946 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3947 mddev->reshape_position != MaxSector ||
3948 mddev->sysfs_active)
3952 if (!mddev->pers->quiesce) {
3953 pr_warn("md: %s: %s does not support online personality change\n",
3954 mdname(mddev), mddev->pers->name);
3958 /* Now find the new personality */
3959 strncpy(clevel, buf, slen);
3960 if (clevel[slen-1] == '\n')
3963 if (kstrtol(clevel, 10, &level))
3966 if (request_module("md-%s", clevel) != 0)
3967 request_module("md-level-%s", clevel);
3968 spin_lock(&pers_lock);
3969 pers = find_pers(level, clevel);
3970 if (!pers || !try_module_get(pers->owner)) {
3971 spin_unlock(&pers_lock);
3972 pr_warn("md: personality %s not loaded\n", clevel);
3976 spin_unlock(&pers_lock);
3978 if (pers == mddev->pers) {
3979 /* Nothing to do! */
3980 module_put(pers->owner);
3984 if (!pers->takeover) {
3985 module_put(pers->owner);
3986 pr_warn("md: %s: %s does not support personality takeover\n",
3987 mdname(mddev), clevel);
3992 rdev_for_each(rdev, mddev)
3993 rdev->new_raid_disk = rdev->raid_disk;
3995 /* ->takeover must set new_* and/or delta_disks
3996 * if it succeeds, and may set them when it fails.
3998 priv = pers->takeover(mddev);
4000 mddev->new_level = mddev->level;
4001 mddev->new_layout = mddev->layout;
4002 mddev->new_chunk_sectors = mddev->chunk_sectors;
4003 mddev->raid_disks -= mddev->delta_disks;
4004 mddev->delta_disks = 0;
4005 mddev->reshape_backwards = 0;
4006 module_put(pers->owner);
4007 pr_warn("md: %s: %s would not accept array\n",
4008 mdname(mddev), clevel);
4013 /* Looks like we have a winner */
4014 mddev_suspend(mddev);
4015 mddev_detach(mddev);
4017 spin_lock(&mddev->lock);
4018 oldpers = mddev->pers;
4019 oldpriv = mddev->private;
4021 mddev->private = priv;
4022 strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4023 mddev->level = mddev->new_level;
4024 mddev->layout = mddev->new_layout;
4025 mddev->chunk_sectors = mddev->new_chunk_sectors;
4026 mddev->delta_disks = 0;
4027 mddev->reshape_backwards = 0;
4028 mddev->degraded = 0;
4029 spin_unlock(&mddev->lock);
4031 if (oldpers->sync_request == NULL &&
4033 /* We are converting from a no-redundancy array
4034 * to a redundancy array and metadata is managed
4035 * externally so we need to be sure that writes
4036 * won't block due to a need to transition
4038 * until external management is started.
4041 mddev->safemode_delay = 0;
4042 mddev->safemode = 0;
4045 oldpers->free(mddev, oldpriv);
4047 if (oldpers->sync_request == NULL &&
4048 pers->sync_request != NULL) {
4049 /* need to add the md_redundancy_group */
4050 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4051 pr_warn("md: cannot register extra attributes for %s\n",
4053 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4054 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4055 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4057 if (oldpers->sync_request != NULL &&
4058 pers->sync_request == NULL) {
4059 /* need to remove the md_redundancy_group */
4060 if (mddev->to_remove == NULL)
4061 mddev->to_remove = &md_redundancy_group;
4064 module_put(oldpers->owner);
4066 rdev_for_each(rdev, mddev) {
4067 if (rdev->raid_disk < 0)
4069 if (rdev->new_raid_disk >= mddev->raid_disks)
4070 rdev->new_raid_disk = -1;
4071 if (rdev->new_raid_disk == rdev->raid_disk)
4073 sysfs_unlink_rdev(mddev, rdev);
4075 rdev_for_each(rdev, mddev) {
4076 if (rdev->raid_disk < 0)
4078 if (rdev->new_raid_disk == rdev->raid_disk)
4080 rdev->raid_disk = rdev->new_raid_disk;
4081 if (rdev->raid_disk < 0)
4082 clear_bit(In_sync, &rdev->flags);
4084 if (sysfs_link_rdev(mddev, rdev))
4085 pr_warn("md: cannot register rd%d for %s after level change\n",
4086 rdev->raid_disk, mdname(mddev));
4090 if (pers->sync_request == NULL) {
4091 /* this is now an array without redundancy, so
4092 * it must always be in_sync
4095 del_timer_sync(&mddev->safemode_timer);
4097 blk_set_stacking_limits(&mddev->queue->limits);
4099 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4100 mddev_resume(mddev);
4102 md_update_sb(mddev, 1);
4103 sysfs_notify_dirent_safe(mddev->sysfs_level);
4107 mddev_unlock(mddev);
4111 static struct md_sysfs_entry md_level =
4112 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4115 layout_show(struct mddev *mddev, char *page)
4117 /* just a number, not meaningful for all levels */
4118 if (mddev->reshape_position != MaxSector &&
4119 mddev->layout != mddev->new_layout)
4120 return sprintf(page, "%d (%d)\n",
4121 mddev->new_layout, mddev->layout);
4122 return sprintf(page, "%d\n", mddev->layout);
4126 layout_store(struct mddev *mddev, const char *buf, size_t len)
4131 err = kstrtouint(buf, 10, &n);
4134 err = mddev_lock(mddev);
4139 if (mddev->pers->check_reshape == NULL)
4141 else if (!md_is_rdwr(mddev))
4144 mddev->new_layout = n;
4145 err = mddev->pers->check_reshape(mddev);
4147 mddev->new_layout = mddev->layout;
4150 mddev->new_layout = n;
4151 if (mddev->reshape_position == MaxSector)
4154 mddev_unlock(mddev);
4157 static struct md_sysfs_entry md_layout =
4158 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4161 raid_disks_show(struct mddev *mddev, char *page)
4163 if (mddev->raid_disks == 0)
4165 if (mddev->reshape_position != MaxSector &&
4166 mddev->delta_disks != 0)
4167 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4168 mddev->raid_disks - mddev->delta_disks);
4169 return sprintf(page, "%d\n", mddev->raid_disks);
4172 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4175 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4180 err = kstrtouint(buf, 10, &n);
4184 err = mddev_lock(mddev);
4188 err = update_raid_disks(mddev, n);
4189 else if (mddev->reshape_position != MaxSector) {
4190 struct md_rdev *rdev;
4191 int olddisks = mddev->raid_disks - mddev->delta_disks;
4194 rdev_for_each(rdev, mddev) {
4196 rdev->data_offset < rdev->new_data_offset)
4199 rdev->data_offset > rdev->new_data_offset)
4203 mddev->delta_disks = n - olddisks;
4204 mddev->raid_disks = n;
4205 mddev->reshape_backwards = (mddev->delta_disks < 0);
4207 mddev->raid_disks = n;
4209 mddev_unlock(mddev);
4210 return err ? err : len;
4212 static struct md_sysfs_entry md_raid_disks =
4213 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4216 uuid_show(struct mddev *mddev, char *page)
4218 return sprintf(page, "%pU\n", mddev->uuid);
4220 static struct md_sysfs_entry md_uuid =
4221 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4224 chunk_size_show(struct mddev *mddev, char *page)
4226 if (mddev->reshape_position != MaxSector &&
4227 mddev->chunk_sectors != mddev->new_chunk_sectors)
4228 return sprintf(page, "%d (%d)\n",
4229 mddev->new_chunk_sectors << 9,
4230 mddev->chunk_sectors << 9);
4231 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4235 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4240 err = kstrtoul(buf, 10, &n);
4244 err = mddev_lock(mddev);
4248 if (mddev->pers->check_reshape == NULL)
4250 else if (!md_is_rdwr(mddev))
4253 mddev->new_chunk_sectors = n >> 9;
4254 err = mddev->pers->check_reshape(mddev);
4256 mddev->new_chunk_sectors = mddev->chunk_sectors;
4259 mddev->new_chunk_sectors = n >> 9;
4260 if (mddev->reshape_position == MaxSector)
4261 mddev->chunk_sectors = n >> 9;
4263 mddev_unlock(mddev);
4266 static struct md_sysfs_entry md_chunk_size =
4267 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4270 resync_start_show(struct mddev *mddev, char *page)
4272 if (mddev->recovery_cp == MaxSector)
4273 return sprintf(page, "none\n");
4274 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4278 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4280 unsigned long long n;
4283 if (cmd_match(buf, "none"))
4286 err = kstrtoull(buf, 10, &n);
4289 if (n != (sector_t)n)
4293 err = mddev_lock(mddev);
4296 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4300 mddev->recovery_cp = n;
4302 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4304 mddev_unlock(mddev);
4307 static struct md_sysfs_entry md_resync_start =
4308 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4309 resync_start_show, resync_start_store);
4312 * The array state can be:
4315 * No devices, no size, no level
4316 * Equivalent to STOP_ARRAY ioctl
4318 * May have some settings, but array is not active
4319 * all IO results in error
4320 * When written, doesn't tear down array, but just stops it
4321 * suspended (not supported yet)
4322 * All IO requests will block. The array can be reconfigured.
4323 * Writing this, if accepted, will block until array is quiescent
4325 * no resync can happen. no superblocks get written.
4326 * write requests fail
4328 * like readonly, but behaves like 'clean' on a write request.
4330 * clean - no pending writes, but otherwise active.
4331 * When written to inactive array, starts without resync
4332 * If a write request arrives then
4333 * if metadata is known, mark 'dirty' and switch to 'active'.
4334 * if not known, block and switch to write-pending
4335 * If written to an active array that has pending writes, then fails.
4337 * fully active: IO and resync can be happening.
4338 * When written to inactive array, starts with resync
4341 * clean, but writes are blocked waiting for 'active' to be written.
4344 * like active, but no writes have been seen for a while (100msec).
4347 * Array is failed. It's useful because mounted-arrays aren't stopped
4348 * when array is failed, so this state will at least alert the user that
4349 * something is wrong.
4351 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4352 write_pending, active_idle, broken, bad_word};
4353 static char *array_states[] = {
4354 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4355 "write-pending", "active-idle", "broken", NULL };
4357 static int match_word(const char *word, char **list)
4360 for (n=0; list[n]; n++)
4361 if (cmd_match(word, list[n]))
4367 array_state_show(struct mddev *mddev, char *page)
4369 enum array_state st = inactive;
4371 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4380 spin_lock(&mddev->lock);
4381 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4383 else if (mddev->in_sync)
4385 else if (mddev->safemode)
4389 spin_unlock(&mddev->lock);
4392 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4395 if (list_empty(&mddev->disks) &&
4396 mddev->raid_disks == 0 &&
4397 mddev->dev_sectors == 0)
4402 return sprintf(page, "%s\n", array_states[st]);
4405 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4406 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4407 static int restart_array(struct mddev *mddev);
4410 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4413 enum array_state st = match_word(buf, array_states);
4415 if (mddev->pers && (st == active || st == clean) &&
4416 mddev->ro != MD_RDONLY) {
4417 /* don't take reconfig_mutex when toggling between
4420 spin_lock(&mddev->lock);
4422 restart_array(mddev);
4423 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4424 md_wakeup_thread(mddev->thread);
4425 wake_up(&mddev->sb_wait);
4426 } else /* st == clean */ {
4427 restart_array(mddev);
4428 if (!set_in_sync(mddev))
4432 sysfs_notify_dirent_safe(mddev->sysfs_state);
4433 spin_unlock(&mddev->lock);
4436 err = mddev_lock(mddev);
4444 /* stopping an active array */
4445 err = do_md_stop(mddev, 0, NULL);
4448 /* stopping an active array */
4450 err = do_md_stop(mddev, 2, NULL);
4452 err = 0; /* already inactive */
4455 break; /* not supported yet */
4458 err = md_set_readonly(mddev, NULL);
4460 mddev->ro = MD_RDONLY;
4461 set_disk_ro(mddev->gendisk, 1);
4462 err = do_md_run(mddev);
4467 if (md_is_rdwr(mddev))
4468 err = md_set_readonly(mddev, NULL);
4469 else if (mddev->ro == MD_RDONLY)
4470 err = restart_array(mddev);
4472 mddev->ro = MD_AUTO_READ;
4473 set_disk_ro(mddev->gendisk, 0);
4476 mddev->ro = MD_AUTO_READ;
4477 err = do_md_run(mddev);
4482 err = restart_array(mddev);
4485 spin_lock(&mddev->lock);
4486 if (!set_in_sync(mddev))
4488 spin_unlock(&mddev->lock);
4494 err = restart_array(mddev);
4497 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4498 wake_up(&mddev->sb_wait);
4501 mddev->ro = MD_RDWR;
4502 set_disk_ro(mddev->gendisk, 0);
4503 err = do_md_run(mddev);
4509 /* these cannot be set */
4514 if (mddev->hold_active == UNTIL_IOCTL)
4515 mddev->hold_active = 0;
4516 sysfs_notify_dirent_safe(mddev->sysfs_state);
4518 mddev_unlock(mddev);
4521 static struct md_sysfs_entry md_array_state =
4522 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4525 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4526 return sprintf(page, "%d\n",
4527 atomic_read(&mddev->max_corr_read_errors));
4531 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4536 rv = kstrtouint(buf, 10, &n);
4541 atomic_set(&mddev->max_corr_read_errors, n);
4545 static struct md_sysfs_entry max_corr_read_errors =
4546 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4547 max_corrected_read_errors_store);
4550 null_show(struct mddev *mddev, char *page)
4556 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4558 /* buf must be %d:%d\n? giving major and minor numbers */
4559 /* The new device is added to the array.
4560 * If the array has a persistent superblock, we read the
4561 * superblock to initialise info and check validity.
4562 * Otherwise, only checking done is that in bind_rdev_to_array,
4563 * which mainly checks size.
4566 int major = simple_strtoul(buf, &e, 10);
4569 struct md_rdev *rdev;
4572 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4574 minor = simple_strtoul(e+1, &e, 10);
4575 if (*e && *e != '\n')
4577 dev = MKDEV(major, minor);
4578 if (major != MAJOR(dev) ||
4579 minor != MINOR(dev))
4582 err = mddev_lock(mddev);
4585 if (mddev->persistent) {
4586 rdev = md_import_device(dev, mddev->major_version,
4587 mddev->minor_version);
4588 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4589 struct md_rdev *rdev0
4590 = list_entry(mddev->disks.next,
4591 struct md_rdev, same_set);
4592 err = super_types[mddev->major_version]
4593 .load_super(rdev, rdev0, mddev->minor_version);
4597 } else if (mddev->external)
4598 rdev = md_import_device(dev, -2, -1);
4600 rdev = md_import_device(dev, -1, -1);
4603 mddev_unlock(mddev);
4604 return PTR_ERR(rdev);
4606 err = bind_rdev_to_array(rdev, mddev);
4609 export_rdev(rdev, mddev);
4610 mddev_unlock(mddev);
4613 return err ? err : len;
4616 static struct md_sysfs_entry md_new_device =
4617 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4620 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4623 unsigned long chunk, end_chunk;
4626 err = mddev_lock(mddev);
4631 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4633 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4634 if (buf == end) break;
4635 if (*end == '-') { /* range */
4637 end_chunk = simple_strtoul(buf, &end, 0);
4638 if (buf == end) break;
4640 if (*end && !isspace(*end)) break;
4641 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4642 buf = skip_spaces(end);
4644 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4646 mddev_unlock(mddev);
4650 static struct md_sysfs_entry md_bitmap =
4651 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4654 size_show(struct mddev *mddev, char *page)
4656 return sprintf(page, "%llu\n",
4657 (unsigned long long)mddev->dev_sectors / 2);
4660 static int update_size(struct mddev *mddev, sector_t num_sectors);
4663 size_store(struct mddev *mddev, const char *buf, size_t len)
4665 /* If array is inactive, we can reduce the component size, but
4666 * not increase it (except from 0).
4667 * If array is active, we can try an on-line resize
4670 int err = strict_blocks_to_sectors(buf, §ors);
4674 err = mddev_lock(mddev);
4678 err = update_size(mddev, sectors);
4680 md_update_sb(mddev, 1);
4682 if (mddev->dev_sectors == 0 ||
4683 mddev->dev_sectors > sectors)
4684 mddev->dev_sectors = sectors;
4688 mddev_unlock(mddev);
4689 return err ? err : len;
4692 static struct md_sysfs_entry md_size =
4693 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4695 /* Metadata version.
4697 * 'none' for arrays with no metadata (good luck...)
4698 * 'external' for arrays with externally managed metadata,
4699 * or N.M for internally known formats
4702 metadata_show(struct mddev *mddev, char *page)
4704 if (mddev->persistent)
4705 return sprintf(page, "%d.%d\n",
4706 mddev->major_version, mddev->minor_version);
4707 else if (mddev->external)
4708 return sprintf(page, "external:%s\n", mddev->metadata_type);
4710 return sprintf(page, "none\n");
4714 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4719 /* Changing the details of 'external' metadata is
4720 * always permitted. Otherwise there must be
4721 * no devices attached to the array.
4724 err = mddev_lock(mddev);
4728 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4730 else if (!list_empty(&mddev->disks))
4734 if (cmd_match(buf, "none")) {
4735 mddev->persistent = 0;
4736 mddev->external = 0;
4737 mddev->major_version = 0;
4738 mddev->minor_version = 90;
4741 if (strncmp(buf, "external:", 9) == 0) {
4742 size_t namelen = len-9;
4743 if (namelen >= sizeof(mddev->metadata_type))
4744 namelen = sizeof(mddev->metadata_type)-1;
4745 strncpy(mddev->metadata_type, buf+9, namelen);
4746 mddev->metadata_type[namelen] = 0;
4747 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4748 mddev->metadata_type[--namelen] = 0;
4749 mddev->persistent = 0;
4750 mddev->external = 1;
4751 mddev->major_version = 0;
4752 mddev->minor_version = 90;
4755 major = simple_strtoul(buf, &e, 10);
4757 if (e==buf || *e != '.')
4760 minor = simple_strtoul(buf, &e, 10);
4761 if (e==buf || (*e && *e != '\n') )
4764 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4766 mddev->major_version = major;
4767 mddev->minor_version = minor;
4768 mddev->persistent = 1;
4769 mddev->external = 0;
4772 mddev_unlock(mddev);
4776 static struct md_sysfs_entry md_metadata =
4777 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4780 action_show(struct mddev *mddev, char *page)
4782 char *type = "idle";
4783 unsigned long recovery = mddev->recovery;
4784 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4786 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4787 (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4788 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4790 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4791 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4793 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4797 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4799 else if (mddev->reshape_position != MaxSector)
4802 return sprintf(page, "%s\n", type);
4805 static void stop_sync_thread(struct mddev *mddev)
4807 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4810 if (mddev_lock(mddev))
4814 * Check again in case MD_RECOVERY_RUNNING is cleared before lock is
4817 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4818 mddev_unlock(mddev);
4822 if (work_pending(&mddev->del_work))
4823 flush_workqueue(md_misc_wq);
4825 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4827 * Thread might be blocked waiting for metadata update which will now
4830 md_wakeup_thread_directly(mddev->sync_thread);
4832 mddev_unlock(mddev);
4835 static void idle_sync_thread(struct mddev *mddev)
4837 int sync_seq = atomic_read(&mddev->sync_seq);
4839 mutex_lock(&mddev->sync_mutex);
4840 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4841 stop_sync_thread(mddev);
4843 wait_event(resync_wait, sync_seq != atomic_read(&mddev->sync_seq) ||
4844 !test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4846 mutex_unlock(&mddev->sync_mutex);
4849 static void frozen_sync_thread(struct mddev *mddev)
4851 mutex_lock(&mddev->sync_mutex);
4852 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4853 stop_sync_thread(mddev);
4855 wait_event(resync_wait, mddev->sync_thread == NULL &&
4856 !test_bit(MD_RECOVERY_RUNNING, &mddev->recovery));
4858 mutex_unlock(&mddev->sync_mutex);
4862 action_store(struct mddev *mddev, const char *page, size_t len)
4864 if (!mddev->pers || !mddev->pers->sync_request)
4868 if (cmd_match(page, "idle"))
4869 idle_sync_thread(mddev);
4870 else if (cmd_match(page, "frozen"))
4871 frozen_sync_thread(mddev);
4872 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4874 else if (cmd_match(page, "resync"))
4875 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4876 else if (cmd_match(page, "recover")) {
4877 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4878 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4879 } else if (cmd_match(page, "reshape")) {
4881 if (mddev->pers->start_reshape == NULL)
4883 err = mddev_lock(mddev);
4885 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4887 } else if (mddev->reshape_position == MaxSector ||
4888 mddev->pers->check_reshape == NULL ||
4889 mddev->pers->check_reshape(mddev)) {
4890 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4891 err = mddev->pers->start_reshape(mddev);
4894 * If reshape is still in progress, and
4895 * md_check_recovery() can continue to reshape,
4896 * don't restart reshape because data can be
4897 * corrupted for raid456.
4899 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4901 mddev_unlock(mddev);
4905 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4907 if (cmd_match(page, "check"))
4908 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4909 else if (!cmd_match(page, "repair"))
4911 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4912 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4913 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4915 if (mddev->ro == MD_AUTO_READ) {
4916 /* A write to sync_action is enough to justify
4917 * canceling read-auto mode
4919 mddev->ro = MD_RDWR;
4920 md_wakeup_thread(mddev->sync_thread);
4922 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4923 md_wakeup_thread(mddev->thread);
4924 sysfs_notify_dirent_safe(mddev->sysfs_action);
4928 static struct md_sysfs_entry md_scan_mode =
4929 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4932 last_sync_action_show(struct mddev *mddev, char *page)
4934 return sprintf(page, "%s\n", mddev->last_sync_action);
4937 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4940 mismatch_cnt_show(struct mddev *mddev, char *page)
4942 return sprintf(page, "%llu\n",
4943 (unsigned long long)
4944 atomic64_read(&mddev->resync_mismatches));
4947 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4950 sync_min_show(struct mddev *mddev, char *page)
4952 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4953 mddev->sync_speed_min ? "local": "system");
4957 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4962 if (strncmp(buf, "system", 6)==0) {
4965 rv = kstrtouint(buf, 10, &min);
4971 mddev->sync_speed_min = min;
4975 static struct md_sysfs_entry md_sync_min =
4976 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4979 sync_max_show(struct mddev *mddev, char *page)
4981 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4982 mddev->sync_speed_max ? "local": "system");
4986 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4991 if (strncmp(buf, "system", 6)==0) {
4994 rv = kstrtouint(buf, 10, &max);
5000 mddev->sync_speed_max = max;
5004 static struct md_sysfs_entry md_sync_max =
5005 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5008 degraded_show(struct mddev *mddev, char *page)
5010 return sprintf(page, "%d\n", mddev->degraded);
5012 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5015 sync_force_parallel_show(struct mddev *mddev, char *page)
5017 return sprintf(page, "%d\n", mddev->parallel_resync);
5021 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5025 if (kstrtol(buf, 10, &n))
5028 if (n != 0 && n != 1)
5031 mddev->parallel_resync = n;
5033 if (mddev->sync_thread)
5034 wake_up(&resync_wait);
5039 /* force parallel resync, even with shared block devices */
5040 static struct md_sysfs_entry md_sync_force_parallel =
5041 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5042 sync_force_parallel_show, sync_force_parallel_store);
5045 sync_speed_show(struct mddev *mddev, char *page)
5047 unsigned long resync, dt, db;
5048 if (mddev->curr_resync == MD_RESYNC_NONE)
5049 return sprintf(page, "none\n");
5050 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5051 dt = (jiffies - mddev->resync_mark) / HZ;
5053 db = resync - mddev->resync_mark_cnt;
5054 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5057 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5060 sync_completed_show(struct mddev *mddev, char *page)
5062 unsigned long long max_sectors, resync;
5064 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5065 return sprintf(page, "none\n");
5067 if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5068 mddev->curr_resync == MD_RESYNC_DELAYED)
5069 return sprintf(page, "delayed\n");
5071 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5072 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5073 max_sectors = mddev->resync_max_sectors;
5075 max_sectors = mddev->dev_sectors;
5077 resync = mddev->curr_resync_completed;
5078 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5081 static struct md_sysfs_entry md_sync_completed =
5082 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5085 min_sync_show(struct mddev *mddev, char *page)
5087 return sprintf(page, "%llu\n",
5088 (unsigned long long)mddev->resync_min);
5091 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5093 unsigned long long min;
5096 if (kstrtoull(buf, 10, &min))
5099 spin_lock(&mddev->lock);
5101 if (min > mddev->resync_max)
5105 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5108 /* Round down to multiple of 4K for safety */
5109 mddev->resync_min = round_down(min, 8);
5113 spin_unlock(&mddev->lock);
5117 static struct md_sysfs_entry md_min_sync =
5118 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5121 max_sync_show(struct mddev *mddev, char *page)
5123 if (mddev->resync_max == MaxSector)
5124 return sprintf(page, "max\n");
5126 return sprintf(page, "%llu\n",
5127 (unsigned long long)mddev->resync_max);
5130 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5133 spin_lock(&mddev->lock);
5134 if (strncmp(buf, "max", 3) == 0)
5135 mddev->resync_max = MaxSector;
5137 unsigned long long max;
5141 if (kstrtoull(buf, 10, &max))
5143 if (max < mddev->resync_min)
5147 if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5148 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5151 /* Must be a multiple of chunk_size */
5152 chunk = mddev->chunk_sectors;
5154 sector_t temp = max;
5157 if (sector_div(temp, chunk))
5160 mddev->resync_max = max;
5162 wake_up(&mddev->recovery_wait);
5165 spin_unlock(&mddev->lock);
5169 static struct md_sysfs_entry md_max_sync =
5170 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5173 suspend_lo_show(struct mddev *mddev, char *page)
5175 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5179 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5181 unsigned long long new;
5184 err = kstrtoull(buf, 10, &new);
5187 if (new != (sector_t)new)
5190 err = mddev_lock(mddev);
5194 if (mddev->pers == NULL ||
5195 mddev->pers->quiesce == NULL)
5197 mddev_suspend(mddev);
5198 mddev->suspend_lo = new;
5199 mddev_resume(mddev);
5203 mddev_unlock(mddev);
5206 static struct md_sysfs_entry md_suspend_lo =
5207 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5210 suspend_hi_show(struct mddev *mddev, char *page)
5212 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5216 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5218 unsigned long long new;
5221 err = kstrtoull(buf, 10, &new);
5224 if (new != (sector_t)new)
5227 err = mddev_lock(mddev);
5231 if (mddev->pers == NULL)
5234 mddev_suspend(mddev);
5235 mddev->suspend_hi = new;
5236 mddev_resume(mddev);
5240 mddev_unlock(mddev);
5243 static struct md_sysfs_entry md_suspend_hi =
5244 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5247 reshape_position_show(struct mddev *mddev, char *page)
5249 if (mddev->reshape_position != MaxSector)
5250 return sprintf(page, "%llu\n",
5251 (unsigned long long)mddev->reshape_position);
5252 strcpy(page, "none\n");
5257 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5259 struct md_rdev *rdev;
5260 unsigned long long new;
5263 err = kstrtoull(buf, 10, &new);
5266 if (new != (sector_t)new)
5268 err = mddev_lock(mddev);
5274 mddev->reshape_position = new;
5275 mddev->delta_disks = 0;
5276 mddev->reshape_backwards = 0;
5277 mddev->new_level = mddev->level;
5278 mddev->new_layout = mddev->layout;
5279 mddev->new_chunk_sectors = mddev->chunk_sectors;
5280 rdev_for_each(rdev, mddev)
5281 rdev->new_data_offset = rdev->data_offset;
5284 mddev_unlock(mddev);
5288 static struct md_sysfs_entry md_reshape_position =
5289 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5290 reshape_position_store);
5293 reshape_direction_show(struct mddev *mddev, char *page)
5295 return sprintf(page, "%s\n",
5296 mddev->reshape_backwards ? "backwards" : "forwards");
5300 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5305 if (cmd_match(buf, "forwards"))
5307 else if (cmd_match(buf, "backwards"))
5311 if (mddev->reshape_backwards == backwards)
5314 err = mddev_lock(mddev);
5317 /* check if we are allowed to change */
5318 if (mddev->delta_disks)
5320 else if (mddev->persistent &&
5321 mddev->major_version == 0)
5324 mddev->reshape_backwards = backwards;
5325 mddev_unlock(mddev);
5329 static struct md_sysfs_entry md_reshape_direction =
5330 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5331 reshape_direction_store);
5334 array_size_show(struct mddev *mddev, char *page)
5336 if (mddev->external_size)
5337 return sprintf(page, "%llu\n",
5338 (unsigned long long)mddev->array_sectors/2);
5340 return sprintf(page, "default\n");
5344 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5349 err = mddev_lock(mddev);
5353 /* cluster raid doesn't support change array_sectors */
5354 if (mddev_is_clustered(mddev)) {
5355 mddev_unlock(mddev);
5359 if (strncmp(buf, "default", 7) == 0) {
5361 sectors = mddev->pers->size(mddev, 0, 0);
5363 sectors = mddev->array_sectors;
5365 mddev->external_size = 0;
5367 if (strict_blocks_to_sectors(buf, §ors) < 0)
5369 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5372 mddev->external_size = 1;
5376 mddev->array_sectors = sectors;
5378 set_capacity_and_notify(mddev->gendisk,
5379 mddev->array_sectors);
5381 mddev_unlock(mddev);
5385 static struct md_sysfs_entry md_array_size =
5386 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5390 consistency_policy_show(struct mddev *mddev, char *page)
5394 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5395 ret = sprintf(page, "journal\n");
5396 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5397 ret = sprintf(page, "ppl\n");
5398 } else if (mddev->bitmap) {
5399 ret = sprintf(page, "bitmap\n");
5400 } else if (mddev->pers) {
5401 if (mddev->pers->sync_request)
5402 ret = sprintf(page, "resync\n");
5404 ret = sprintf(page, "none\n");
5406 ret = sprintf(page, "unknown\n");
5413 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5418 if (mddev->pers->change_consistency_policy)
5419 err = mddev->pers->change_consistency_policy(mddev, buf);
5422 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5423 set_bit(MD_HAS_PPL, &mddev->flags);
5428 return err ? err : len;
5431 static struct md_sysfs_entry md_consistency_policy =
5432 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5433 consistency_policy_store);
5435 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5437 return sprintf(page, "%d\n", mddev->fail_last_dev);
5441 * Setting fail_last_dev to true to allow last device to be forcibly removed
5442 * from RAID1/RAID10.
5445 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5450 ret = kstrtobool(buf, &value);
5454 if (value != mddev->fail_last_dev)
5455 mddev->fail_last_dev = value;
5459 static struct md_sysfs_entry md_fail_last_dev =
5460 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5461 fail_last_dev_store);
5463 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5465 if (mddev->pers == NULL || (mddev->pers->level != 1))
5466 return sprintf(page, "n/a\n");
5468 return sprintf(page, "%d\n", mddev->serialize_policy);
5472 * Setting serialize_policy to true to enforce write IO is not reordered
5476 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5481 err = kstrtobool(buf, &value);
5485 if (value == mddev->serialize_policy)
5488 err = mddev_lock(mddev);
5491 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5492 pr_err("md: serialize_policy is only effective for raid1\n");
5497 mddev_suspend(mddev);
5499 mddev_create_serial_pool(mddev, NULL, true);
5501 mddev_destroy_serial_pool(mddev, NULL, true);
5502 mddev->serialize_policy = value;
5503 mddev_resume(mddev);
5505 mddev_unlock(mddev);
5509 static struct md_sysfs_entry md_serialize_policy =
5510 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5511 serialize_policy_store);
5514 static struct attribute *md_default_attrs[] = {
5517 &md_raid_disks.attr,
5519 &md_chunk_size.attr,
5521 &md_resync_start.attr,
5523 &md_new_device.attr,
5524 &md_safe_delay.attr,
5525 &md_array_state.attr,
5526 &md_reshape_position.attr,
5527 &md_reshape_direction.attr,
5528 &md_array_size.attr,
5529 &max_corr_read_errors.attr,
5530 &md_consistency_policy.attr,
5531 &md_fail_last_dev.attr,
5532 &md_serialize_policy.attr,
5536 static const struct attribute_group md_default_group = {
5537 .attrs = md_default_attrs,
5540 static struct attribute *md_redundancy_attrs[] = {
5542 &md_last_scan_mode.attr,
5543 &md_mismatches.attr,
5546 &md_sync_speed.attr,
5547 &md_sync_force_parallel.attr,
5548 &md_sync_completed.attr,
5551 &md_suspend_lo.attr,
5552 &md_suspend_hi.attr,
5557 static const struct attribute_group md_redundancy_group = {
5559 .attrs = md_redundancy_attrs,
5562 static const struct attribute_group *md_attr_groups[] = {
5569 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5571 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5572 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5577 spin_lock(&all_mddevs_lock);
5578 if (!mddev_get(mddev)) {
5579 spin_unlock(&all_mddevs_lock);
5582 spin_unlock(&all_mddevs_lock);
5584 rv = entry->show(mddev, page);
5590 md_attr_store(struct kobject *kobj, struct attribute *attr,
5591 const char *page, size_t length)
5593 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5594 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5599 if (!capable(CAP_SYS_ADMIN))
5601 spin_lock(&all_mddevs_lock);
5602 if (!mddev_get(mddev)) {
5603 spin_unlock(&all_mddevs_lock);
5606 spin_unlock(&all_mddevs_lock);
5607 rv = entry->store(mddev, page, length);
5612 static void md_kobj_release(struct kobject *ko)
5614 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5616 if (mddev->sysfs_state)
5617 sysfs_put(mddev->sysfs_state);
5618 if (mddev->sysfs_level)
5619 sysfs_put(mddev->sysfs_level);
5621 del_gendisk(mddev->gendisk);
5622 put_disk(mddev->gendisk);
5625 static const struct sysfs_ops md_sysfs_ops = {
5626 .show = md_attr_show,
5627 .store = md_attr_store,
5629 static const struct kobj_type md_ktype = {
5630 .release = md_kobj_release,
5631 .sysfs_ops = &md_sysfs_ops,
5632 .default_groups = md_attr_groups,
5637 static void mddev_delayed_delete(struct work_struct *ws)
5639 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5641 kobject_put(&mddev->kobj);
5644 static void no_op(struct percpu_ref *r) {}
5646 int mddev_init_writes_pending(struct mddev *mddev)
5648 if (mddev->writes_pending.percpu_count_ptr)
5650 if (percpu_ref_init(&mddev->writes_pending, no_op,
5651 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5653 /* We want to start with the refcount at zero */
5654 percpu_ref_put(&mddev->writes_pending);
5657 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5659 struct mddev *md_alloc(dev_t dev, char *name)
5662 * If dev is zero, name is the name of a device to allocate with
5663 * an arbitrary minor number. It will be "md_???"
5664 * If dev is non-zero it must be a device number with a MAJOR of
5665 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5666 * the device is being created by opening a node in /dev.
5667 * If "name" is not NULL, the device is being created by
5668 * writing to /sys/module/md_mod/parameters/new_array.
5670 static DEFINE_MUTEX(disks_mutex);
5671 struct mddev *mddev;
5672 struct gendisk *disk;
5679 * Wait for any previous instance of this device to be completely
5680 * removed (mddev_delayed_delete).
5682 flush_workqueue(md_misc_wq);
5684 mutex_lock(&disks_mutex);
5685 mddev = mddev_alloc(dev);
5686 if (IS_ERR(mddev)) {
5687 error = PTR_ERR(mddev);
5691 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5692 shift = partitioned ? MdpMinorShift : 0;
5693 unit = MINOR(mddev->unit) >> shift;
5696 /* Need to ensure that 'name' is not a duplicate.
5698 struct mddev *mddev2;
5699 spin_lock(&all_mddevs_lock);
5701 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5702 if (mddev2->gendisk &&
5703 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5704 spin_unlock(&all_mddevs_lock);
5706 goto out_free_mddev;
5708 spin_unlock(&all_mddevs_lock);
5712 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5714 mddev->hold_active = UNTIL_STOP;
5717 disk = blk_alloc_disk(NUMA_NO_NODE);
5719 goto out_free_mddev;
5721 disk->major = MAJOR(mddev->unit);
5722 disk->first_minor = unit << shift;
5723 disk->minors = 1 << shift;
5725 strcpy(disk->disk_name, name);
5726 else if (partitioned)
5727 sprintf(disk->disk_name, "md_d%d", unit);
5729 sprintf(disk->disk_name, "md%d", unit);
5730 disk->fops = &md_fops;
5731 disk->private_data = mddev;
5733 mddev->queue = disk->queue;
5734 blk_set_stacking_limits(&mddev->queue->limits);
5735 blk_queue_write_cache(mddev->queue, true, true);
5736 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5737 mddev->gendisk = disk;
5738 error = add_disk(disk);
5742 kobject_init(&mddev->kobj, &md_ktype);
5743 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5746 * The disk is already live at this point. Clear the hold flag
5747 * and let mddev_put take care of the deletion, as it isn't any
5748 * different from a normal close on last release now.
5750 mddev->hold_active = 0;
5751 mutex_unlock(&disks_mutex);
5753 return ERR_PTR(error);
5756 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5757 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5758 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5759 mutex_unlock(&disks_mutex);
5767 mutex_unlock(&disks_mutex);
5768 return ERR_PTR(error);
5771 static int md_alloc_and_put(dev_t dev, char *name)
5773 struct mddev *mddev = md_alloc(dev, name);
5776 return PTR_ERR(mddev);
5781 static void md_probe(dev_t dev)
5783 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5786 md_alloc_and_put(dev, NULL);
5789 static int add_named_array(const char *val, const struct kernel_param *kp)
5792 * val must be "md_*" or "mdNNN".
5793 * For "md_*" we allocate an array with a large free minor number, and
5794 * set the name to val. val must not already be an active name.
5795 * For "mdNNN" we allocate an array with the minor number NNN
5796 * which must not already be in use.
5798 int len = strlen(val);
5799 char buf[DISK_NAME_LEN];
5800 unsigned long devnum;
5802 while (len && val[len-1] == '\n')
5804 if (len >= DISK_NAME_LEN)
5806 strscpy(buf, val, len+1);
5807 if (strncmp(buf, "md_", 3) == 0)
5808 return md_alloc_and_put(0, buf);
5809 if (strncmp(buf, "md", 2) == 0 &&
5811 kstrtoul(buf+2, 10, &devnum) == 0 &&
5812 devnum <= MINORMASK)
5813 return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5818 static void md_safemode_timeout(struct timer_list *t)
5820 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5822 mddev->safemode = 1;
5823 if (mddev->external)
5824 sysfs_notify_dirent_safe(mddev->sysfs_state);
5826 md_wakeup_thread(mddev->thread);
5829 static int start_dirty_degraded;
5830 static void active_io_release(struct percpu_ref *ref)
5832 struct mddev *mddev = container_of(ref, struct mddev, active_io);
5834 wake_up(&mddev->sb_wait);
5837 int md_run(struct mddev *mddev)
5840 struct md_rdev *rdev;
5841 struct md_personality *pers;
5844 if (list_empty(&mddev->disks))
5845 /* cannot run an array with no devices.. */
5850 /* Cannot run until previous stop completes properly */
5851 if (mddev->sysfs_active)
5855 * Analyze all RAID superblock(s)
5857 if (!mddev->raid_disks) {
5858 if (!mddev->persistent)
5860 err = analyze_sbs(mddev);
5865 if (mddev->level != LEVEL_NONE)
5866 request_module("md-level-%d", mddev->level);
5867 else if (mddev->clevel[0])
5868 request_module("md-%s", mddev->clevel);
5871 * Drop all container device buffers, from now on
5872 * the only valid external interface is through the md
5875 mddev->has_superblocks = false;
5876 rdev_for_each(rdev, mddev) {
5877 if (test_bit(Faulty, &rdev->flags))
5879 sync_blockdev(rdev->bdev);
5880 invalidate_bdev(rdev->bdev);
5881 if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5882 mddev->ro = MD_RDONLY;
5884 set_disk_ro(mddev->gendisk, 1);
5888 mddev->has_superblocks = true;
5890 /* perform some consistency tests on the device.
5891 * We don't want the data to overlap the metadata,
5892 * Internal Bitmap issues have been handled elsewhere.
5894 if (rdev->meta_bdev) {
5895 /* Nothing to check */;
5896 } else if (rdev->data_offset < rdev->sb_start) {
5897 if (mddev->dev_sectors &&
5898 rdev->data_offset + mddev->dev_sectors
5900 pr_warn("md: %s: data overlaps metadata\n",
5905 if (rdev->sb_start + rdev->sb_size/512
5906 > rdev->data_offset) {
5907 pr_warn("md: %s: metadata overlaps data\n",
5912 sysfs_notify_dirent_safe(rdev->sysfs_state);
5913 nowait = nowait && bdev_nowait(rdev->bdev);
5916 err = percpu_ref_init(&mddev->active_io, active_io_release,
5917 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5921 if (!bioset_initialized(&mddev->bio_set)) {
5922 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5924 goto exit_active_io;
5926 if (!bioset_initialized(&mddev->sync_set)) {
5927 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5932 if (!bioset_initialized(&mddev->io_clone_set)) {
5933 err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
5934 offsetof(struct md_io_clone, bio_clone), 0);
5939 spin_lock(&pers_lock);
5940 pers = find_pers(mddev->level, mddev->clevel);
5941 if (!pers || !try_module_get(pers->owner)) {
5942 spin_unlock(&pers_lock);
5943 if (mddev->level != LEVEL_NONE)
5944 pr_warn("md: personality for level %d is not loaded!\n",
5947 pr_warn("md: personality for level %s is not loaded!\n",
5952 spin_unlock(&pers_lock);
5953 if (mddev->level != pers->level) {
5954 mddev->level = pers->level;
5955 mddev->new_level = pers->level;
5957 strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5959 if (mddev->reshape_position != MaxSector &&
5960 pers->start_reshape == NULL) {
5961 /* This personality cannot handle reshaping... */
5962 module_put(pers->owner);
5967 if (pers->sync_request) {
5968 /* Warn if this is a potentially silly
5971 struct md_rdev *rdev2;
5974 rdev_for_each(rdev, mddev)
5975 rdev_for_each(rdev2, mddev) {
5977 rdev->bdev->bd_disk ==
5978 rdev2->bdev->bd_disk) {
5979 pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5988 pr_warn("True protection against single-disk failure might be compromised.\n");
5991 mddev->recovery = 0;
5992 /* may be over-ridden by personality */
5993 mddev->resync_max_sectors = mddev->dev_sectors;
5995 mddev->ok_start_degraded = start_dirty_degraded;
5997 if (start_readonly && md_is_rdwr(mddev))
5998 mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
6000 err = pers->run(mddev);
6002 pr_warn("md: pers->run() failed ...\n");
6003 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6004 WARN_ONCE(!mddev->external_size,
6005 "%s: default size too small, but 'external_size' not in effect?\n",
6007 pr_warn("md: invalid array_size %llu > default size %llu\n",
6008 (unsigned long long)mddev->array_sectors / 2,
6009 (unsigned long long)pers->size(mddev, 0, 0) / 2);
6012 if (err == 0 && pers->sync_request &&
6013 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6014 struct bitmap *bitmap;
6016 bitmap = md_bitmap_create(mddev, -1);
6017 if (IS_ERR(bitmap)) {
6018 err = PTR_ERR(bitmap);
6019 pr_warn("%s: failed to create bitmap (%d)\n",
6020 mdname(mddev), err);
6022 mddev->bitmap = bitmap;
6028 if (mddev->bitmap_info.max_write_behind > 0) {
6029 bool create_pool = false;
6031 rdev_for_each(rdev, mddev) {
6032 if (test_bit(WriteMostly, &rdev->flags) &&
6033 rdev_init_serial(rdev))
6036 if (create_pool && mddev->serial_info_pool == NULL) {
6037 mddev->serial_info_pool =
6038 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6039 sizeof(struct serial_info));
6040 if (!mddev->serial_info_pool) {
6050 rdev_for_each(rdev, mddev) {
6051 if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
6056 if (mddev->degraded)
6059 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6061 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6062 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6064 /* Set the NOWAIT flags if all underlying devices support it */
6066 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
6068 if (pers->sync_request) {
6069 if (mddev->kobj.sd &&
6070 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6071 pr_warn("md: cannot register extra attributes for %s\n",
6073 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6074 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6075 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6076 } else if (mddev->ro == MD_AUTO_READ)
6077 mddev->ro = MD_RDWR;
6079 atomic_set(&mddev->max_corr_read_errors,
6080 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6081 mddev->safemode = 0;
6082 if (mddev_is_clustered(mddev))
6083 mddev->safemode_delay = 0;
6085 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6088 spin_lock(&mddev->lock);
6090 spin_unlock(&mddev->lock);
6091 rdev_for_each(rdev, mddev)
6092 if (rdev->raid_disk >= 0)
6093 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6095 if (mddev->degraded && md_is_rdwr(mddev))
6096 /* This ensures that recovering status is reported immediately
6097 * via sysfs - until a lack of spares is confirmed.
6099 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6100 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6102 if (mddev->sb_flags)
6103 md_update_sb(mddev, 0);
6109 mddev_detach(mddev);
6111 pers->free(mddev, mddev->private);
6112 mddev->private = NULL;
6113 module_put(pers->owner);
6114 md_bitmap_destroy(mddev);
6116 bioset_exit(&mddev->io_clone_set);
6118 bioset_exit(&mddev->sync_set);
6120 bioset_exit(&mddev->bio_set);
6122 percpu_ref_exit(&mddev->active_io);
6125 EXPORT_SYMBOL_GPL(md_run);
6127 int do_md_run(struct mddev *mddev)
6131 set_bit(MD_NOT_READY, &mddev->flags);
6132 err = md_run(mddev);
6135 err = md_bitmap_load(mddev);
6137 md_bitmap_destroy(mddev);
6141 if (mddev_is_clustered(mddev))
6142 md_allow_write(mddev);
6144 /* run start up tasks that require md_thread */
6147 md_wakeup_thread(mddev->thread);
6148 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6150 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6151 clear_bit(MD_NOT_READY, &mddev->flags);
6153 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6154 sysfs_notify_dirent_safe(mddev->sysfs_state);
6155 sysfs_notify_dirent_safe(mddev->sysfs_action);
6156 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6158 clear_bit(MD_NOT_READY, &mddev->flags);
6162 int md_start(struct mddev *mddev)
6166 if (mddev->pers->start) {
6167 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6168 md_wakeup_thread(mddev->thread);
6169 ret = mddev->pers->start(mddev);
6170 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6171 md_wakeup_thread(mddev->sync_thread);
6175 EXPORT_SYMBOL_GPL(md_start);
6177 static int restart_array(struct mddev *mddev)
6179 struct gendisk *disk = mddev->gendisk;
6180 struct md_rdev *rdev;
6181 bool has_journal = false;
6182 bool has_readonly = false;
6184 /* Complain if it has no devices */
6185 if (list_empty(&mddev->disks))
6189 if (md_is_rdwr(mddev))
6193 rdev_for_each_rcu(rdev, mddev) {
6194 if (test_bit(Journal, &rdev->flags) &&
6195 !test_bit(Faulty, &rdev->flags))
6197 if (rdev_read_only(rdev))
6198 has_readonly = true;
6201 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6202 /* Don't restart rw with journal missing/faulty */
6207 mddev->safemode = 0;
6208 mddev->ro = MD_RDWR;
6209 set_disk_ro(disk, 0);
6210 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6211 /* Kick recovery or resync if necessary */
6212 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6213 md_wakeup_thread(mddev->thread);
6214 md_wakeup_thread(mddev->sync_thread);
6215 sysfs_notify_dirent_safe(mddev->sysfs_state);
6219 static void md_clean(struct mddev *mddev)
6221 mddev->array_sectors = 0;
6222 mddev->external_size = 0;
6223 mddev->dev_sectors = 0;
6224 mddev->raid_disks = 0;
6225 mddev->recovery_cp = 0;
6226 mddev->resync_min = 0;
6227 mddev->resync_max = MaxSector;
6228 mddev->reshape_position = MaxSector;
6229 /* we still need mddev->external in export_rdev, do not clear it yet */
6230 mddev->persistent = 0;
6231 mddev->level = LEVEL_NONE;
6232 mddev->clevel[0] = 0;
6234 mddev->sb_flags = 0;
6235 mddev->ro = MD_RDWR;
6236 mddev->metadata_type[0] = 0;
6237 mddev->chunk_sectors = 0;
6238 mddev->ctime = mddev->utime = 0;
6240 mddev->max_disks = 0;
6242 mddev->can_decrease_events = 0;
6243 mddev->delta_disks = 0;
6244 mddev->reshape_backwards = 0;
6245 mddev->new_level = LEVEL_NONE;
6246 mddev->new_layout = 0;
6247 mddev->new_chunk_sectors = 0;
6248 mddev->curr_resync = MD_RESYNC_NONE;
6249 atomic64_set(&mddev->resync_mismatches, 0);
6250 mddev->suspend_lo = mddev->suspend_hi = 0;
6251 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6252 mddev->recovery = 0;
6255 mddev->degraded = 0;
6256 mddev->safemode = 0;
6257 mddev->private = NULL;
6258 mddev->cluster_info = NULL;
6259 mddev->bitmap_info.offset = 0;
6260 mddev->bitmap_info.default_offset = 0;
6261 mddev->bitmap_info.default_space = 0;
6262 mddev->bitmap_info.chunksize = 0;
6263 mddev->bitmap_info.daemon_sleep = 0;
6264 mddev->bitmap_info.max_write_behind = 0;
6265 mddev->bitmap_info.nodes = 0;
6268 static void __md_stop_writes(struct mddev *mddev)
6270 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6271 if (work_pending(&mddev->del_work))
6272 flush_workqueue(md_misc_wq);
6273 if (mddev->sync_thread) {
6274 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6275 md_reap_sync_thread(mddev);
6278 del_timer_sync(&mddev->safemode_timer);
6280 if (mddev->pers && mddev->pers->quiesce) {
6281 mddev->pers->quiesce(mddev, 1);
6282 mddev->pers->quiesce(mddev, 0);
6284 md_bitmap_flush(mddev);
6286 if (md_is_rdwr(mddev) &&
6287 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6289 /* mark array as shutdown cleanly */
6290 if (!mddev_is_clustered(mddev))
6292 md_update_sb(mddev, 1);
6294 /* disable policy to guarantee rdevs free resources for serialization */
6295 mddev->serialize_policy = 0;
6296 mddev_destroy_serial_pool(mddev, NULL, true);
6299 void md_stop_writes(struct mddev *mddev)
6301 mddev_lock_nointr(mddev);
6302 __md_stop_writes(mddev);
6303 mddev_unlock(mddev);
6305 EXPORT_SYMBOL_GPL(md_stop_writes);
6307 static void mddev_detach(struct mddev *mddev)
6309 md_bitmap_wait_behind_writes(mddev);
6310 if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6311 mddev->pers->quiesce(mddev, 1);
6312 mddev->pers->quiesce(mddev, 0);
6314 md_unregister_thread(mddev, &mddev->thread);
6316 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6319 static void __md_stop(struct mddev *mddev)
6321 struct md_personality *pers = mddev->pers;
6322 md_bitmap_destroy(mddev);
6323 mddev_detach(mddev);
6324 /* Ensure ->event_work is done */
6325 if (mddev->event_work.func)
6326 flush_workqueue(md_misc_wq);
6327 spin_lock(&mddev->lock);
6329 spin_unlock(&mddev->lock);
6331 pers->free(mddev, mddev->private);
6332 mddev->private = NULL;
6333 if (pers->sync_request && mddev->to_remove == NULL)
6334 mddev->to_remove = &md_redundancy_group;
6335 module_put(pers->owner);
6336 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6338 percpu_ref_exit(&mddev->active_io);
6339 bioset_exit(&mddev->bio_set);
6340 bioset_exit(&mddev->sync_set);
6341 bioset_exit(&mddev->io_clone_set);
6344 void md_stop(struct mddev *mddev)
6346 lockdep_assert_held(&mddev->reconfig_mutex);
6348 /* stop the array and free an attached data structures.
6349 * This is called from dm-raid
6351 __md_stop_writes(mddev);
6353 percpu_ref_exit(&mddev->writes_pending);
6356 EXPORT_SYMBOL_GPL(md_stop);
6358 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6363 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6366 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6368 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6369 md_wakeup_thread(mddev->thread);
6371 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6372 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6375 * Thread might be blocked waiting for metadata update which will now
6378 md_wakeup_thread_directly(mddev->sync_thread);
6380 mddev_unlock(mddev);
6381 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6383 wait_event(mddev->sb_wait,
6384 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6385 mddev_lock_nointr(mddev);
6387 mutex_lock(&mddev->open_mutex);
6388 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6389 mddev->sync_thread ||
6390 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6391 pr_warn("md: %s still in use.\n",mdname(mddev));
6397 __md_stop_writes(mddev);
6399 if (mddev->ro == MD_RDONLY) {
6404 mddev->ro = MD_RDONLY;
6405 set_disk_ro(mddev->gendisk, 1);
6409 if ((mddev->pers && !err) || did_freeze) {
6410 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6411 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6412 md_wakeup_thread(mddev->thread);
6413 sysfs_notify_dirent_safe(mddev->sysfs_state);
6416 mutex_unlock(&mddev->open_mutex);
6421 * 0 - completely stop and dis-assemble array
6422 * 2 - stop but do not disassemble array
6424 static int do_md_stop(struct mddev *mddev, int mode,
6425 struct block_device *bdev)
6427 struct gendisk *disk = mddev->gendisk;
6428 struct md_rdev *rdev;
6431 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6433 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6434 md_wakeup_thread(mddev->thread);
6436 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6437 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6440 * Thread might be blocked waiting for metadata update which will now
6443 md_wakeup_thread_directly(mddev->sync_thread);
6445 mddev_unlock(mddev);
6446 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6447 !test_bit(MD_RECOVERY_RUNNING,
6448 &mddev->recovery)));
6449 mddev_lock_nointr(mddev);
6451 mutex_lock(&mddev->open_mutex);
6452 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6453 mddev->sysfs_active ||
6454 mddev->sync_thread ||
6455 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6456 pr_warn("md: %s still in use.\n",mdname(mddev));
6457 mutex_unlock(&mddev->open_mutex);
6459 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6460 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6461 md_wakeup_thread(mddev->thread);
6466 if (!md_is_rdwr(mddev))
6467 set_disk_ro(disk, 0);
6469 __md_stop_writes(mddev);
6472 /* tell userspace to handle 'inactive' */
6473 sysfs_notify_dirent_safe(mddev->sysfs_state);
6475 rdev_for_each(rdev, mddev)
6476 if (rdev->raid_disk >= 0)
6477 sysfs_unlink_rdev(mddev, rdev);
6479 set_capacity_and_notify(disk, 0);
6480 mutex_unlock(&mddev->open_mutex);
6483 if (!md_is_rdwr(mddev))
6484 mddev->ro = MD_RDWR;
6486 mutex_unlock(&mddev->open_mutex);
6488 * Free resources if final stop
6491 pr_info("md: %s stopped.\n", mdname(mddev));
6493 if (mddev->bitmap_info.file) {
6494 struct file *f = mddev->bitmap_info.file;
6495 spin_lock(&mddev->lock);
6496 mddev->bitmap_info.file = NULL;
6497 spin_unlock(&mddev->lock);
6500 mddev->bitmap_info.offset = 0;
6502 export_array(mddev);
6505 if (mddev->hold_active == UNTIL_STOP)
6506 mddev->hold_active = 0;
6509 sysfs_notify_dirent_safe(mddev->sysfs_state);
6514 static void autorun_array(struct mddev *mddev)
6516 struct md_rdev *rdev;
6519 if (list_empty(&mddev->disks))
6522 pr_info("md: running: ");
6524 rdev_for_each(rdev, mddev) {
6525 pr_cont("<%pg>", rdev->bdev);
6529 err = do_md_run(mddev);
6531 pr_warn("md: do_md_run() returned %d\n", err);
6532 do_md_stop(mddev, 0, NULL);
6537 * lets try to run arrays based on all disks that have arrived
6538 * until now. (those are in pending_raid_disks)
6540 * the method: pick the first pending disk, collect all disks with
6541 * the same UUID, remove all from the pending list and put them into
6542 * the 'same_array' list. Then order this list based on superblock
6543 * update time (freshest comes first), kick out 'old' disks and
6544 * compare superblocks. If everything's fine then run it.
6546 * If "unit" is allocated, then bump its reference count
6548 static void autorun_devices(int part)
6550 struct md_rdev *rdev0, *rdev, *tmp;
6551 struct mddev *mddev;
6553 pr_info("md: autorun ...\n");
6554 while (!list_empty(&pending_raid_disks)) {
6557 LIST_HEAD(candidates);
6558 rdev0 = list_entry(pending_raid_disks.next,
6559 struct md_rdev, same_set);
6561 pr_debug("md: considering %pg ...\n", rdev0->bdev);
6562 INIT_LIST_HEAD(&candidates);
6563 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6564 if (super_90_load(rdev, rdev0, 0) >= 0) {
6565 pr_debug("md: adding %pg ...\n",
6567 list_move(&rdev->same_set, &candidates);
6570 * now we have a set of devices, with all of them having
6571 * mostly sane superblocks. It's time to allocate the
6575 dev = MKDEV(mdp_major,
6576 rdev0->preferred_minor << MdpMinorShift);
6577 unit = MINOR(dev) >> MdpMinorShift;
6579 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6582 if (rdev0->preferred_minor != unit) {
6583 pr_warn("md: unit number in %pg is bad: %d\n",
6584 rdev0->bdev, rdev0->preferred_minor);
6588 mddev = md_alloc(dev, NULL);
6592 if (mddev_lock(mddev))
6593 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6594 else if (mddev->raid_disks || mddev->major_version
6595 || !list_empty(&mddev->disks)) {
6596 pr_warn("md: %s already running, cannot run %pg\n",
6597 mdname(mddev), rdev0->bdev);
6598 mddev_unlock(mddev);
6600 pr_debug("md: created %s\n", mdname(mddev));
6601 mddev->persistent = 1;
6602 rdev_for_each_list(rdev, tmp, &candidates) {
6603 list_del_init(&rdev->same_set);
6604 if (bind_rdev_to_array(rdev, mddev))
6605 export_rdev(rdev, mddev);
6607 autorun_array(mddev);
6608 mddev_unlock(mddev);
6610 /* on success, candidates will be empty, on error
6613 rdev_for_each_list(rdev, tmp, &candidates) {
6614 list_del_init(&rdev->same_set);
6615 export_rdev(rdev, mddev);
6619 pr_info("md: ... autorun DONE.\n");
6621 #endif /* !MODULE */
6623 static int get_version(void __user *arg)
6627 ver.major = MD_MAJOR_VERSION;
6628 ver.minor = MD_MINOR_VERSION;
6629 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6631 if (copy_to_user(arg, &ver, sizeof(ver)))
6637 static int get_array_info(struct mddev *mddev, void __user *arg)
6639 mdu_array_info_t info;
6640 int nr,working,insync,failed,spare;
6641 struct md_rdev *rdev;
6643 nr = working = insync = failed = spare = 0;
6645 rdev_for_each_rcu(rdev, mddev) {
6647 if (test_bit(Faulty, &rdev->flags))
6651 if (test_bit(In_sync, &rdev->flags))
6653 else if (test_bit(Journal, &rdev->flags))
6654 /* TODO: add journal count to md_u.h */
6662 info.major_version = mddev->major_version;
6663 info.minor_version = mddev->minor_version;
6664 info.patch_version = MD_PATCHLEVEL_VERSION;
6665 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6666 info.level = mddev->level;
6667 info.size = mddev->dev_sectors / 2;
6668 if (info.size != mddev->dev_sectors / 2) /* overflow */
6671 info.raid_disks = mddev->raid_disks;
6672 info.md_minor = mddev->md_minor;
6673 info.not_persistent= !mddev->persistent;
6675 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6678 info.state = (1<<MD_SB_CLEAN);
6679 if (mddev->bitmap && mddev->bitmap_info.offset)
6680 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6681 if (mddev_is_clustered(mddev))
6682 info.state |= (1<<MD_SB_CLUSTERED);
6683 info.active_disks = insync;
6684 info.working_disks = working;
6685 info.failed_disks = failed;
6686 info.spare_disks = spare;
6688 info.layout = mddev->layout;
6689 info.chunk_size = mddev->chunk_sectors << 9;
6691 if (copy_to_user(arg, &info, sizeof(info)))
6697 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6699 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6703 file = kzalloc(sizeof(*file), GFP_NOIO);
6708 spin_lock(&mddev->lock);
6709 /* bitmap enabled */
6710 if (mddev->bitmap_info.file) {
6711 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6712 sizeof(file->pathname));
6716 memmove(file->pathname, ptr,
6717 sizeof(file->pathname)-(ptr-file->pathname));
6719 spin_unlock(&mddev->lock);
6722 copy_to_user(arg, file, sizeof(*file)))
6729 static int get_disk_info(struct mddev *mddev, void __user * arg)
6731 mdu_disk_info_t info;
6732 struct md_rdev *rdev;
6734 if (copy_from_user(&info, arg, sizeof(info)))
6738 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6740 info.major = MAJOR(rdev->bdev->bd_dev);
6741 info.minor = MINOR(rdev->bdev->bd_dev);
6742 info.raid_disk = rdev->raid_disk;
6744 if (test_bit(Faulty, &rdev->flags))
6745 info.state |= (1<<MD_DISK_FAULTY);
6746 else if (test_bit(In_sync, &rdev->flags)) {
6747 info.state |= (1<<MD_DISK_ACTIVE);
6748 info.state |= (1<<MD_DISK_SYNC);
6750 if (test_bit(Journal, &rdev->flags))
6751 info.state |= (1<<MD_DISK_JOURNAL);
6752 if (test_bit(WriteMostly, &rdev->flags))
6753 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6754 if (test_bit(FailFast, &rdev->flags))
6755 info.state |= (1<<MD_DISK_FAILFAST);
6757 info.major = info.minor = 0;
6758 info.raid_disk = -1;
6759 info.state = (1<<MD_DISK_REMOVED);
6763 if (copy_to_user(arg, &info, sizeof(info)))
6769 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6771 struct md_rdev *rdev;
6772 dev_t dev = MKDEV(info->major,info->minor);
6774 if (mddev_is_clustered(mddev) &&
6775 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6776 pr_warn("%s: Cannot add to clustered mddev.\n",
6781 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6784 if (!mddev->raid_disks) {
6786 /* expecting a device which has a superblock */
6787 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6789 pr_warn("md: md_import_device returned %ld\n",
6791 return PTR_ERR(rdev);
6793 if (!list_empty(&mddev->disks)) {
6794 struct md_rdev *rdev0
6795 = list_entry(mddev->disks.next,
6796 struct md_rdev, same_set);
6797 err = super_types[mddev->major_version]
6798 .load_super(rdev, rdev0, mddev->minor_version);
6800 pr_warn("md: %pg has different UUID to %pg\n",
6803 export_rdev(rdev, mddev);
6807 err = bind_rdev_to_array(rdev, mddev);
6809 export_rdev(rdev, mddev);
6814 * md_add_new_disk can be used once the array is assembled
6815 * to add "hot spares". They must already have a superblock
6820 if (!mddev->pers->hot_add_disk) {
6821 pr_warn("%s: personality does not support diskops!\n",
6825 if (mddev->persistent)
6826 rdev = md_import_device(dev, mddev->major_version,
6827 mddev->minor_version);
6829 rdev = md_import_device(dev, -1, -1);
6831 pr_warn("md: md_import_device returned %ld\n",
6833 return PTR_ERR(rdev);
6835 /* set saved_raid_disk if appropriate */
6836 if (!mddev->persistent) {
6837 if (info->state & (1<<MD_DISK_SYNC) &&
6838 info->raid_disk < mddev->raid_disks) {
6839 rdev->raid_disk = info->raid_disk;
6840 clear_bit(Bitmap_sync, &rdev->flags);
6842 rdev->raid_disk = -1;
6843 rdev->saved_raid_disk = rdev->raid_disk;
6845 super_types[mddev->major_version].
6846 validate_super(mddev, NULL/*freshest*/, rdev);
6847 if ((info->state & (1<<MD_DISK_SYNC)) &&
6848 rdev->raid_disk != info->raid_disk) {
6849 /* This was a hot-add request, but events doesn't
6850 * match, so reject it.
6852 export_rdev(rdev, mddev);
6856 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6857 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6858 set_bit(WriteMostly, &rdev->flags);
6860 clear_bit(WriteMostly, &rdev->flags);
6861 if (info->state & (1<<MD_DISK_FAILFAST))
6862 set_bit(FailFast, &rdev->flags);
6864 clear_bit(FailFast, &rdev->flags);
6866 if (info->state & (1<<MD_DISK_JOURNAL)) {
6867 struct md_rdev *rdev2;
6868 bool has_journal = false;
6870 /* make sure no existing journal disk */
6871 rdev_for_each(rdev2, mddev) {
6872 if (test_bit(Journal, &rdev2->flags)) {
6877 if (has_journal || mddev->bitmap) {
6878 export_rdev(rdev, mddev);
6881 set_bit(Journal, &rdev->flags);
6884 * check whether the device shows up in other nodes
6886 if (mddev_is_clustered(mddev)) {
6887 if (info->state & (1 << MD_DISK_CANDIDATE))
6888 set_bit(Candidate, &rdev->flags);
6889 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6890 /* --add initiated by this node */
6891 err = md_cluster_ops->add_new_disk(mddev, rdev);
6893 export_rdev(rdev, mddev);
6899 rdev->raid_disk = -1;
6900 err = bind_rdev_to_array(rdev, mddev);
6903 export_rdev(rdev, mddev);
6905 if (mddev_is_clustered(mddev)) {
6906 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6908 err = md_cluster_ops->new_disk_ack(mddev,
6911 md_kick_rdev_from_array(rdev);
6915 md_cluster_ops->add_new_disk_cancel(mddev);
6917 err = add_bound_rdev(rdev);
6921 err = add_bound_rdev(rdev);
6926 /* otherwise, md_add_new_disk is only allowed
6927 * for major_version==0 superblocks
6929 if (mddev->major_version != 0) {
6930 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6934 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6936 rdev = md_import_device(dev, -1, 0);
6938 pr_warn("md: error, md_import_device() returned %ld\n",
6940 return PTR_ERR(rdev);
6942 rdev->desc_nr = info->number;
6943 if (info->raid_disk < mddev->raid_disks)
6944 rdev->raid_disk = info->raid_disk;
6946 rdev->raid_disk = -1;
6948 if (rdev->raid_disk < mddev->raid_disks)
6949 if (info->state & (1<<MD_DISK_SYNC))
6950 set_bit(In_sync, &rdev->flags);
6952 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6953 set_bit(WriteMostly, &rdev->flags);
6954 if (info->state & (1<<MD_DISK_FAILFAST))
6955 set_bit(FailFast, &rdev->flags);
6957 if (!mddev->persistent) {
6958 pr_debug("md: nonpersistent superblock ...\n");
6959 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6961 rdev->sb_start = calc_dev_sboffset(rdev);
6962 rdev->sectors = rdev->sb_start;
6964 err = bind_rdev_to_array(rdev, mddev);
6966 export_rdev(rdev, mddev);
6974 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6976 struct md_rdev *rdev;
6981 rdev = find_rdev(mddev, dev);
6985 if (rdev->raid_disk < 0)
6988 clear_bit(Blocked, &rdev->flags);
6989 remove_and_add_spares(mddev, rdev);
6991 if (rdev->raid_disk >= 0)
6995 if (mddev_is_clustered(mddev)) {
6996 if (md_cluster_ops->remove_disk(mddev, rdev))
7000 md_kick_rdev_from_array(rdev);
7001 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7003 md_wakeup_thread(mddev->thread);
7005 md_update_sb(mddev, 1);
7010 pr_debug("md: cannot remove active disk %pg from %s ...\n",
7011 rdev->bdev, mdname(mddev));
7015 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7018 struct md_rdev *rdev;
7023 if (mddev->major_version != 0) {
7024 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7028 if (!mddev->pers->hot_add_disk) {
7029 pr_warn("%s: personality does not support diskops!\n",
7034 rdev = md_import_device(dev, -1, 0);
7036 pr_warn("md: error, md_import_device() returned %ld\n",
7041 if (mddev->persistent)
7042 rdev->sb_start = calc_dev_sboffset(rdev);
7044 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7046 rdev->sectors = rdev->sb_start;
7048 if (test_bit(Faulty, &rdev->flags)) {
7049 pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
7050 rdev->bdev, mdname(mddev));
7055 clear_bit(In_sync, &rdev->flags);
7057 rdev->saved_raid_disk = -1;
7058 err = bind_rdev_to_array(rdev, mddev);
7063 * The rest should better be atomic, we can have disk failures
7064 * noticed in interrupt contexts ...
7067 rdev->raid_disk = -1;
7069 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7071 md_update_sb(mddev, 1);
7073 * If the new disk does not support REQ_NOWAIT,
7074 * disable on the whole MD.
7076 if (!bdev_nowait(rdev->bdev)) {
7077 pr_info("%s: Disabling nowait because %pg does not support nowait\n",
7078 mdname(mddev), rdev->bdev);
7079 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
7082 * Kick recovery, maybe this spare has to be added to the
7083 * array immediately.
7085 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7086 md_wakeup_thread(mddev->thread);
7091 export_rdev(rdev, mddev);
7095 static int set_bitmap_file(struct mddev *mddev, int fd)
7100 if (!mddev->pers->quiesce || !mddev->thread)
7102 if (mddev->recovery || mddev->sync_thread)
7104 /* we should be able to change the bitmap.. */
7108 struct inode *inode;
7111 if (mddev->bitmap || mddev->bitmap_info.file)
7112 return -EEXIST; /* cannot add when bitmap is present */
7114 if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7115 pr_warn("%s: bitmap files not supported by this kernel\n",
7119 pr_warn("%s: using deprecated bitmap file support\n",
7125 pr_warn("%s: error: failed to get bitmap file\n",
7130 inode = f->f_mapping->host;
7131 if (!S_ISREG(inode->i_mode)) {
7132 pr_warn("%s: error: bitmap file must be a regular file\n",
7135 } else if (!(f->f_mode & FMODE_WRITE)) {
7136 pr_warn("%s: error: bitmap file must open for write\n",
7139 } else if (atomic_read(&inode->i_writecount) != 1) {
7140 pr_warn("%s: error: bitmap file is already in use\n",
7148 mddev->bitmap_info.file = f;
7149 mddev->bitmap_info.offset = 0; /* file overrides offset */
7150 } else if (mddev->bitmap == NULL)
7151 return -ENOENT; /* cannot remove what isn't there */
7155 struct bitmap *bitmap;
7157 bitmap = md_bitmap_create(mddev, -1);
7158 mddev_suspend(mddev);
7159 if (!IS_ERR(bitmap)) {
7160 mddev->bitmap = bitmap;
7161 err = md_bitmap_load(mddev);
7163 err = PTR_ERR(bitmap);
7165 md_bitmap_destroy(mddev);
7168 mddev_resume(mddev);
7169 } else if (fd < 0) {
7170 mddev_suspend(mddev);
7171 md_bitmap_destroy(mddev);
7172 mddev_resume(mddev);
7176 struct file *f = mddev->bitmap_info.file;
7178 spin_lock(&mddev->lock);
7179 mddev->bitmap_info.file = NULL;
7180 spin_unlock(&mddev->lock);
7189 * md_set_array_info is used two different ways
7190 * The original usage is when creating a new array.
7191 * In this usage, raid_disks is > 0 and it together with
7192 * level, size, not_persistent,layout,chunksize determine the
7193 * shape of the array.
7194 * This will always create an array with a type-0.90.0 superblock.
7195 * The newer usage is when assembling an array.
7196 * In this case raid_disks will be 0, and the major_version field is
7197 * use to determine which style super-blocks are to be found on the devices.
7198 * The minor and patch _version numbers are also kept incase the
7199 * super_block handler wishes to interpret them.
7201 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7203 if (info->raid_disks == 0) {
7204 /* just setting version number for superblock loading */
7205 if (info->major_version < 0 ||
7206 info->major_version >= ARRAY_SIZE(super_types) ||
7207 super_types[info->major_version].name == NULL) {
7208 /* maybe try to auto-load a module? */
7209 pr_warn("md: superblock version %d not known\n",
7210 info->major_version);
7213 mddev->major_version = info->major_version;
7214 mddev->minor_version = info->minor_version;
7215 mddev->patch_version = info->patch_version;
7216 mddev->persistent = !info->not_persistent;
7217 /* ensure mddev_put doesn't delete this now that there
7218 * is some minimal configuration.
7220 mddev->ctime = ktime_get_real_seconds();
7223 mddev->major_version = MD_MAJOR_VERSION;
7224 mddev->minor_version = MD_MINOR_VERSION;
7225 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7226 mddev->ctime = ktime_get_real_seconds();
7228 mddev->level = info->level;
7229 mddev->clevel[0] = 0;
7230 mddev->dev_sectors = 2 * (sector_t)info->size;
7231 mddev->raid_disks = info->raid_disks;
7232 /* don't set md_minor, it is determined by which /dev/md* was
7235 if (info->state & (1<<MD_SB_CLEAN))
7236 mddev->recovery_cp = MaxSector;
7238 mddev->recovery_cp = 0;
7239 mddev->persistent = ! info->not_persistent;
7240 mddev->external = 0;
7242 mddev->layout = info->layout;
7243 if (mddev->level == 0)
7244 /* Cannot trust RAID0 layout info here */
7246 mddev->chunk_sectors = info->chunk_size >> 9;
7248 if (mddev->persistent) {
7249 mddev->max_disks = MD_SB_DISKS;
7251 mddev->sb_flags = 0;
7253 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7255 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7256 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7257 mddev->bitmap_info.offset = 0;
7259 mddev->reshape_position = MaxSector;
7262 * Generate a 128 bit UUID
7264 get_random_bytes(mddev->uuid, 16);
7266 mddev->new_level = mddev->level;
7267 mddev->new_chunk_sectors = mddev->chunk_sectors;
7268 mddev->new_layout = mddev->layout;
7269 mddev->delta_disks = 0;
7270 mddev->reshape_backwards = 0;
7275 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7277 lockdep_assert_held(&mddev->reconfig_mutex);
7279 if (mddev->external_size)
7282 mddev->array_sectors = array_sectors;
7284 EXPORT_SYMBOL(md_set_array_sectors);
7286 static int update_size(struct mddev *mddev, sector_t num_sectors)
7288 struct md_rdev *rdev;
7290 int fit = (num_sectors == 0);
7291 sector_t old_dev_sectors = mddev->dev_sectors;
7293 if (mddev->pers->resize == NULL)
7295 /* The "num_sectors" is the number of sectors of each device that
7296 * is used. This can only make sense for arrays with redundancy.
7297 * linear and raid0 always use whatever space is available. We can only
7298 * consider changing this number if no resync or reconstruction is
7299 * happening, and if the new size is acceptable. It must fit before the
7300 * sb_start or, if that is <data_offset, it must fit before the size
7301 * of each device. If num_sectors is zero, we find the largest size
7304 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7307 if (!md_is_rdwr(mddev))
7310 rdev_for_each(rdev, mddev) {
7311 sector_t avail = rdev->sectors;
7313 if (fit && (num_sectors == 0 || num_sectors > avail))
7314 num_sectors = avail;
7315 if (avail < num_sectors)
7318 rv = mddev->pers->resize(mddev, num_sectors);
7320 if (mddev_is_clustered(mddev))
7321 md_cluster_ops->update_size(mddev, old_dev_sectors);
7322 else if (mddev->queue) {
7323 set_capacity_and_notify(mddev->gendisk,
7324 mddev->array_sectors);
7330 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7333 struct md_rdev *rdev;
7334 /* change the number of raid disks */
7335 if (mddev->pers->check_reshape == NULL)
7337 if (!md_is_rdwr(mddev))
7339 if (raid_disks <= 0 ||
7340 (mddev->max_disks && raid_disks >= mddev->max_disks))
7342 if (mddev->sync_thread ||
7343 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7344 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7345 mddev->reshape_position != MaxSector)
7348 rdev_for_each(rdev, mddev) {
7349 if (mddev->raid_disks < raid_disks &&
7350 rdev->data_offset < rdev->new_data_offset)
7352 if (mddev->raid_disks > raid_disks &&
7353 rdev->data_offset > rdev->new_data_offset)
7357 mddev->delta_disks = raid_disks - mddev->raid_disks;
7358 if (mddev->delta_disks < 0)
7359 mddev->reshape_backwards = 1;
7360 else if (mddev->delta_disks > 0)
7361 mddev->reshape_backwards = 0;
7363 rv = mddev->pers->check_reshape(mddev);
7365 mddev->delta_disks = 0;
7366 mddev->reshape_backwards = 0;
7372 * update_array_info is used to change the configuration of an
7374 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7375 * fields in the info are checked against the array.
7376 * Any differences that cannot be handled will cause an error.
7377 * Normally, only one change can be managed at a time.
7379 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7385 /* calculate expected state,ignoring low bits */
7386 if (mddev->bitmap && mddev->bitmap_info.offset)
7387 state |= (1 << MD_SB_BITMAP_PRESENT);
7389 if (mddev->major_version != info->major_version ||
7390 mddev->minor_version != info->minor_version ||
7391 /* mddev->patch_version != info->patch_version || */
7392 mddev->ctime != info->ctime ||
7393 mddev->level != info->level ||
7394 /* mddev->layout != info->layout || */
7395 mddev->persistent != !info->not_persistent ||
7396 mddev->chunk_sectors != info->chunk_size >> 9 ||
7397 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7398 ((state^info->state) & 0xfffffe00)
7401 /* Check there is only one change */
7402 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7404 if (mddev->raid_disks != info->raid_disks)
7406 if (mddev->layout != info->layout)
7408 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7415 if (mddev->layout != info->layout) {
7417 * we don't need to do anything at the md level, the
7418 * personality will take care of it all.
7420 if (mddev->pers->check_reshape == NULL)
7423 mddev->new_layout = info->layout;
7424 rv = mddev->pers->check_reshape(mddev);
7426 mddev->new_layout = mddev->layout;
7430 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7431 rv = update_size(mddev, (sector_t)info->size * 2);
7433 if (mddev->raid_disks != info->raid_disks)
7434 rv = update_raid_disks(mddev, info->raid_disks);
7436 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7437 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7441 if (mddev->recovery || mddev->sync_thread) {
7445 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7446 struct bitmap *bitmap;
7447 /* add the bitmap */
7448 if (mddev->bitmap) {
7452 if (mddev->bitmap_info.default_offset == 0) {
7456 mddev->bitmap_info.offset =
7457 mddev->bitmap_info.default_offset;
7458 mddev->bitmap_info.space =
7459 mddev->bitmap_info.default_space;
7460 bitmap = md_bitmap_create(mddev, -1);
7461 mddev_suspend(mddev);
7462 if (!IS_ERR(bitmap)) {
7463 mddev->bitmap = bitmap;
7464 rv = md_bitmap_load(mddev);
7466 rv = PTR_ERR(bitmap);
7468 md_bitmap_destroy(mddev);
7469 mddev_resume(mddev);
7471 /* remove the bitmap */
7472 if (!mddev->bitmap) {
7476 if (mddev->bitmap->storage.file) {
7480 if (mddev->bitmap_info.nodes) {
7481 /* hold PW on all the bitmap lock */
7482 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7483 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7485 md_cluster_ops->unlock_all_bitmaps(mddev);
7489 mddev->bitmap_info.nodes = 0;
7490 md_cluster_ops->leave(mddev);
7491 module_put(md_cluster_mod);
7492 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7494 mddev_suspend(mddev);
7495 md_bitmap_destroy(mddev);
7496 mddev_resume(mddev);
7497 mddev->bitmap_info.offset = 0;
7500 md_update_sb(mddev, 1);
7506 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7508 struct md_rdev *rdev;
7511 if (mddev->pers == NULL)
7515 rdev = md_find_rdev_rcu(mddev, dev);
7519 md_error(mddev, rdev);
7520 if (test_bit(MD_BROKEN, &mddev->flags))
7528 * We have a problem here : there is no easy way to give a CHS
7529 * virtual geometry. We currently pretend that we have a 2 heads
7530 * 4 sectors (with a BIG number of cylinders...). This drives
7531 * dosfs just mad... ;-)
7533 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7535 struct mddev *mddev = bdev->bd_disk->private_data;
7539 geo->cylinders = mddev->array_sectors / 8;
7543 static inline bool md_ioctl_valid(unsigned int cmd)
7547 case GET_ARRAY_INFO:
7548 case GET_BITMAP_FILE:
7551 case HOT_REMOVE_DISK:
7553 case RESTART_ARRAY_RW:
7555 case SET_ARRAY_INFO:
7556 case SET_BITMAP_FILE:
7557 case SET_DISK_FAULTY:
7560 case CLUSTERED_DISK_NACK:
7567 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7569 mdu_array_info_t info;
7573 memset(&info, 0, sizeof(info));
7574 else if (copy_from_user(&info, argp, sizeof(info)))
7578 err = update_array_info(mddev, &info);
7580 pr_warn("md: couldn't update array info. %d\n", err);
7584 if (!list_empty(&mddev->disks)) {
7585 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7589 if (mddev->raid_disks) {
7590 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7594 err = md_set_array_info(mddev, &info);
7596 pr_warn("md: couldn't set array info. %d\n", err);
7601 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7602 unsigned int cmd, unsigned long arg)
7605 void __user *argp = (void __user *)arg;
7606 struct mddev *mddev = NULL;
7607 bool did_set_md_closing = false;
7609 if (!md_ioctl_valid(cmd))
7614 case GET_ARRAY_INFO:
7618 if (!capable(CAP_SYS_ADMIN))
7623 * Commands dealing with the RAID driver but not any
7628 err = get_version(argp);
7634 * Commands creating/starting a new array:
7637 mddev = bdev->bd_disk->private_data;
7644 /* Some actions do not requires the mutex */
7646 case GET_ARRAY_INFO:
7647 if (!mddev->raid_disks && !mddev->external)
7650 err = get_array_info(mddev, argp);
7654 if (!mddev->raid_disks && !mddev->external)
7657 err = get_disk_info(mddev, argp);
7660 case SET_DISK_FAULTY:
7661 err = set_disk_faulty(mddev, new_decode_dev(arg));
7664 case GET_BITMAP_FILE:
7665 err = get_bitmap_file(mddev, argp);
7670 if (cmd == HOT_REMOVE_DISK)
7671 /* need to ensure recovery thread has run */
7672 wait_event_interruptible_timeout(mddev->sb_wait,
7673 !test_bit(MD_RECOVERY_NEEDED,
7675 msecs_to_jiffies(5000));
7676 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7677 /* Need to flush page cache, and ensure no-one else opens
7680 mutex_lock(&mddev->open_mutex);
7681 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7682 mutex_unlock(&mddev->open_mutex);
7686 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7687 mutex_unlock(&mddev->open_mutex);
7691 did_set_md_closing = true;
7692 mutex_unlock(&mddev->open_mutex);
7693 sync_blockdev(bdev);
7695 err = mddev_lock(mddev);
7697 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7702 if (cmd == SET_ARRAY_INFO) {
7703 err = __md_set_array_info(mddev, argp);
7708 * Commands querying/configuring an existing array:
7710 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7711 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7712 if ((!mddev->raid_disks && !mddev->external)
7713 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7714 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7715 && cmd != GET_BITMAP_FILE) {
7721 * Commands even a read-only array can execute:
7724 case RESTART_ARRAY_RW:
7725 err = restart_array(mddev);
7729 err = do_md_stop(mddev, 0, bdev);
7733 err = md_set_readonly(mddev, bdev);
7736 case HOT_REMOVE_DISK:
7737 err = hot_remove_disk(mddev, new_decode_dev(arg));
7741 /* We can support ADD_NEW_DISK on read-only arrays
7742 * only if we are re-adding a preexisting device.
7743 * So require mddev->pers and MD_DISK_SYNC.
7746 mdu_disk_info_t info;
7747 if (copy_from_user(&info, argp, sizeof(info)))
7749 else if (!(info.state & (1<<MD_DISK_SYNC)))
7750 /* Need to clear read-only for this */
7753 err = md_add_new_disk(mddev, &info);
7760 * The remaining ioctls are changing the state of the
7761 * superblock, so we do not allow them on read-only arrays.
7763 if (!md_is_rdwr(mddev) && mddev->pers) {
7764 if (mddev->ro != MD_AUTO_READ) {
7768 mddev->ro = MD_RDWR;
7769 sysfs_notify_dirent_safe(mddev->sysfs_state);
7770 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7771 /* mddev_unlock will wake thread */
7772 /* If a device failed while we were read-only, we
7773 * need to make sure the metadata is updated now.
7775 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7776 mddev_unlock(mddev);
7777 wait_event(mddev->sb_wait,
7778 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7779 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7780 mddev_lock_nointr(mddev);
7787 mdu_disk_info_t info;
7788 if (copy_from_user(&info, argp, sizeof(info)))
7791 err = md_add_new_disk(mddev, &info);
7795 case CLUSTERED_DISK_NACK:
7796 if (mddev_is_clustered(mddev))
7797 md_cluster_ops->new_disk_ack(mddev, false);
7803 err = hot_add_disk(mddev, new_decode_dev(arg));
7807 err = do_md_run(mddev);
7810 case SET_BITMAP_FILE:
7811 err = set_bitmap_file(mddev, (int)arg);
7820 if (mddev->hold_active == UNTIL_IOCTL &&
7822 mddev->hold_active = 0;
7823 mddev_unlock(mddev);
7825 if(did_set_md_closing)
7826 clear_bit(MD_CLOSING, &mddev->flags);
7829 #ifdef CONFIG_COMPAT
7830 static int md_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
7831 unsigned int cmd, unsigned long arg)
7834 case HOT_REMOVE_DISK:
7836 case SET_DISK_FAULTY:
7837 case SET_BITMAP_FILE:
7838 /* These take in integer arg, do not convert */
7841 arg = (unsigned long)compat_ptr(arg);
7845 return md_ioctl(bdev, mode, cmd, arg);
7847 #endif /* CONFIG_COMPAT */
7849 static int md_set_read_only(struct block_device *bdev, bool ro)
7851 struct mddev *mddev = bdev->bd_disk->private_data;
7854 err = mddev_lock(mddev);
7858 if (!mddev->raid_disks && !mddev->external) {
7864 * Transitioning to read-auto need only happen for arrays that call
7865 * md_write_start and which are not ready for writes yet.
7867 if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7868 err = restart_array(mddev);
7871 mddev->ro = MD_AUTO_READ;
7875 mddev_unlock(mddev);
7879 static int md_open(struct gendisk *disk, blk_mode_t mode)
7881 struct mddev *mddev;
7884 spin_lock(&all_mddevs_lock);
7885 mddev = mddev_get(disk->private_data);
7886 spin_unlock(&all_mddevs_lock);
7890 err = mutex_lock_interruptible(&mddev->open_mutex);
7895 if (test_bit(MD_CLOSING, &mddev->flags))
7898 atomic_inc(&mddev->openers);
7899 mutex_unlock(&mddev->open_mutex);
7901 disk_check_media_change(disk);
7905 mutex_unlock(&mddev->open_mutex);
7911 static void md_release(struct gendisk *disk)
7913 struct mddev *mddev = disk->private_data;
7916 atomic_dec(&mddev->openers);
7920 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7922 struct mddev *mddev = disk->private_data;
7923 unsigned int ret = 0;
7926 ret = DISK_EVENT_MEDIA_CHANGE;
7931 static void md_free_disk(struct gendisk *disk)
7933 struct mddev *mddev = disk->private_data;
7935 percpu_ref_exit(&mddev->writes_pending);
7939 const struct block_device_operations md_fops =
7941 .owner = THIS_MODULE,
7942 .submit_bio = md_submit_bio,
7944 .release = md_release,
7946 #ifdef CONFIG_COMPAT
7947 .compat_ioctl = md_compat_ioctl,
7949 .getgeo = md_getgeo,
7950 .check_events = md_check_events,
7951 .set_read_only = md_set_read_only,
7952 .free_disk = md_free_disk,
7955 static int md_thread(void *arg)
7957 struct md_thread *thread = arg;
7960 * md_thread is a 'system-thread', it's priority should be very
7961 * high. We avoid resource deadlocks individually in each
7962 * raid personality. (RAID5 does preallocation) We also use RR and
7963 * the very same RT priority as kswapd, thus we will never get
7964 * into a priority inversion deadlock.
7966 * we definitely have to have equal or higher priority than
7967 * bdflush, otherwise bdflush will deadlock if there are too
7968 * many dirty RAID5 blocks.
7971 allow_signal(SIGKILL);
7972 while (!kthread_should_stop()) {
7974 /* We need to wait INTERRUPTIBLE so that
7975 * we don't add to the load-average.
7976 * That means we need to be sure no signals are
7979 if (signal_pending(current))
7980 flush_signals(current);
7982 wait_event_interruptible_timeout
7984 test_bit(THREAD_WAKEUP, &thread->flags)
7985 || kthread_should_stop() || kthread_should_park(),
7988 clear_bit(THREAD_WAKEUP, &thread->flags);
7989 if (kthread_should_park())
7991 if (!kthread_should_stop())
7992 thread->run(thread);
7998 static void md_wakeup_thread_directly(struct md_thread __rcu *thread)
8000 struct md_thread *t;
8003 t = rcu_dereference(thread);
8005 wake_up_process(t->tsk);
8009 void md_wakeup_thread(struct md_thread __rcu *thread)
8011 struct md_thread *t;
8014 t = rcu_dereference(thread);
8016 pr_debug("md: waking up MD thread %s.\n", t->tsk->comm);
8017 set_bit(THREAD_WAKEUP, &t->flags);
8018 wake_up(&t->wqueue);
8022 EXPORT_SYMBOL(md_wakeup_thread);
8024 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
8025 struct mddev *mddev, const char *name)
8027 struct md_thread *thread;
8029 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
8033 init_waitqueue_head(&thread->wqueue);
8036 thread->mddev = mddev;
8037 thread->timeout = MAX_SCHEDULE_TIMEOUT;
8038 thread->tsk = kthread_run(md_thread, thread,
8040 mdname(thread->mddev),
8042 if (IS_ERR(thread->tsk)) {
8048 EXPORT_SYMBOL(md_register_thread);
8050 void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp)
8052 struct md_thread *thread = rcu_dereference_protected(*threadp,
8053 lockdep_is_held(&mddev->reconfig_mutex));
8058 rcu_assign_pointer(*threadp, NULL);
8061 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8062 kthread_stop(thread->tsk);
8065 EXPORT_SYMBOL(md_unregister_thread);
8067 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8069 if (!rdev || test_bit(Faulty, &rdev->flags))
8072 if (!mddev->pers || !mddev->pers->error_handler)
8074 mddev->pers->error_handler(mddev, rdev);
8076 if (mddev->pers->level == 0 || mddev->pers->level == LEVEL_LINEAR)
8079 if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
8080 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8081 sysfs_notify_dirent_safe(rdev->sysfs_state);
8082 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8083 if (!test_bit(MD_BROKEN, &mddev->flags)) {
8084 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8085 md_wakeup_thread(mddev->thread);
8087 if (mddev->event_work.func)
8088 queue_work(md_misc_wq, &mddev->event_work);
8091 EXPORT_SYMBOL(md_error);
8093 /* seq_file implementation /proc/mdstat */
8095 static void status_unused(struct seq_file *seq)
8098 struct md_rdev *rdev;
8100 seq_printf(seq, "unused devices: ");
8102 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8104 seq_printf(seq, "%pg ", rdev->bdev);
8107 seq_printf(seq, "<none>");
8109 seq_printf(seq, "\n");
8112 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8114 sector_t max_sectors, resync, res;
8115 unsigned long dt, db = 0;
8116 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8117 int scale, recovery_active;
8118 unsigned int per_milli;
8120 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8121 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8122 max_sectors = mddev->resync_max_sectors;
8124 max_sectors = mddev->dev_sectors;
8126 resync = mddev->curr_resync;
8127 if (resync < MD_RESYNC_ACTIVE) {
8128 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8129 /* Still cleaning up */
8130 resync = max_sectors;
8131 } else if (resync > max_sectors) {
8132 resync = max_sectors;
8134 res = atomic_read(&mddev->recovery_active);
8136 * Resync has started, but the subtraction has overflowed or
8137 * yielded one of the special values. Force it to active to
8138 * ensure the status reports an active resync.
8140 if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8141 resync = MD_RESYNC_ACTIVE;
8146 if (resync == MD_RESYNC_NONE) {
8147 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8148 struct md_rdev *rdev;
8150 rdev_for_each(rdev, mddev)
8151 if (rdev->raid_disk >= 0 &&
8152 !test_bit(Faulty, &rdev->flags) &&
8153 rdev->recovery_offset != MaxSector &&
8154 rdev->recovery_offset) {
8155 seq_printf(seq, "\trecover=REMOTE");
8158 if (mddev->reshape_position != MaxSector)
8159 seq_printf(seq, "\treshape=REMOTE");
8161 seq_printf(seq, "\tresync=REMOTE");
8164 if (mddev->recovery_cp < MaxSector) {
8165 seq_printf(seq, "\tresync=PENDING");
8170 if (resync < MD_RESYNC_ACTIVE) {
8171 seq_printf(seq, "\tresync=DELAYED");
8175 WARN_ON(max_sectors == 0);
8176 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8177 * in a sector_t, and (max_sectors>>scale) will fit in a
8178 * u32, as those are the requirements for sector_div.
8179 * Thus 'scale' must be at least 10
8182 if (sizeof(sector_t) > sizeof(unsigned long)) {
8183 while ( max_sectors/2 > (1ULL<<(scale+32)))
8186 res = (resync>>scale)*1000;
8187 sector_div(res, (u32)((max_sectors>>scale)+1));
8191 int i, x = per_milli/50, y = 20-x;
8192 seq_printf(seq, "[");
8193 for (i = 0; i < x; i++)
8194 seq_printf(seq, "=");
8195 seq_printf(seq, ">");
8196 for (i = 0; i < y; i++)
8197 seq_printf(seq, ".");
8198 seq_printf(seq, "] ");
8200 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8201 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8203 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8205 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8206 "resync" : "recovery"))),
8207 per_milli/10, per_milli % 10,
8208 (unsigned long long) resync/2,
8209 (unsigned long long) max_sectors/2);
8212 * dt: time from mark until now
8213 * db: blocks written from mark until now
8214 * rt: remaining time
8216 * rt is a sector_t, which is always 64bit now. We are keeping
8217 * the original algorithm, but it is not really necessary.
8219 * Original algorithm:
8220 * So we divide before multiply in case it is 32bit and close
8222 * We scale the divisor (db) by 32 to avoid losing precision
8223 * near the end of resync when the number of remaining sectors
8225 * We then divide rt by 32 after multiplying by db to compensate.
8226 * The '+1' avoids division by zero if db is very small.
8228 dt = ((jiffies - mddev->resync_mark) / HZ);
8231 curr_mark_cnt = mddev->curr_mark_cnt;
8232 recovery_active = atomic_read(&mddev->recovery_active);
8233 resync_mark_cnt = mddev->resync_mark_cnt;
8235 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8236 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8238 rt = max_sectors - resync; /* number of remaining sectors */
8239 rt = div64_u64(rt, db/32+1);
8243 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8244 ((unsigned long)rt % 60)/6);
8246 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8250 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8252 struct list_head *tmp;
8254 struct mddev *mddev;
8266 spin_lock(&all_mddevs_lock);
8267 list_for_each(tmp,&all_mddevs)
8269 mddev = list_entry(tmp, struct mddev, all_mddevs);
8270 if (!mddev_get(mddev))
8272 spin_unlock(&all_mddevs_lock);
8275 spin_unlock(&all_mddevs_lock);
8277 return (void*)2;/* tail */
8281 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8283 struct list_head *tmp;
8284 struct mddev *next_mddev, *mddev = v;
8285 struct mddev *to_put = NULL;
8291 spin_lock(&all_mddevs_lock);
8292 if (v == (void*)1) {
8293 tmp = all_mddevs.next;
8296 tmp = mddev->all_mddevs.next;
8300 if (tmp == &all_mddevs) {
8301 next_mddev = (void*)2;
8305 next_mddev = list_entry(tmp, struct mddev, all_mddevs);
8306 if (mddev_get(next_mddev))
8309 tmp = mddev->all_mddevs.next;
8311 spin_unlock(&all_mddevs_lock);
8319 static void md_seq_stop(struct seq_file *seq, void *v)
8321 struct mddev *mddev = v;
8323 if (mddev && v != (void*)1 && v != (void*)2)
8327 static int md_seq_show(struct seq_file *seq, void *v)
8329 struct mddev *mddev = v;
8331 struct md_rdev *rdev;
8333 if (v == (void*)1) {
8334 struct md_personality *pers;
8335 seq_printf(seq, "Personalities : ");
8336 spin_lock(&pers_lock);
8337 list_for_each_entry(pers, &pers_list, list)
8338 seq_printf(seq, "[%s] ", pers->name);
8340 spin_unlock(&pers_lock);
8341 seq_printf(seq, "\n");
8342 seq->poll_event = atomic_read(&md_event_count);
8345 if (v == (void*)2) {
8350 spin_lock(&mddev->lock);
8351 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8352 seq_printf(seq, "%s : %sactive", mdname(mddev),
8353 mddev->pers ? "" : "in");
8355 if (mddev->ro == MD_RDONLY)
8356 seq_printf(seq, " (read-only)");
8357 if (mddev->ro == MD_AUTO_READ)
8358 seq_printf(seq, " (auto-read-only)");
8359 seq_printf(seq, " %s", mddev->pers->name);
8364 rdev_for_each_rcu(rdev, mddev) {
8365 seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8367 if (test_bit(WriteMostly, &rdev->flags))
8368 seq_printf(seq, "(W)");
8369 if (test_bit(Journal, &rdev->flags))
8370 seq_printf(seq, "(J)");
8371 if (test_bit(Faulty, &rdev->flags)) {
8372 seq_printf(seq, "(F)");
8375 if (rdev->raid_disk < 0)
8376 seq_printf(seq, "(S)"); /* spare */
8377 if (test_bit(Replacement, &rdev->flags))
8378 seq_printf(seq, "(R)");
8379 sectors += rdev->sectors;
8383 if (!list_empty(&mddev->disks)) {
8385 seq_printf(seq, "\n %llu blocks",
8386 (unsigned long long)
8387 mddev->array_sectors / 2);
8389 seq_printf(seq, "\n %llu blocks",
8390 (unsigned long long)sectors / 2);
8392 if (mddev->persistent) {
8393 if (mddev->major_version != 0 ||
8394 mddev->minor_version != 90) {
8395 seq_printf(seq," super %d.%d",
8396 mddev->major_version,
8397 mddev->minor_version);
8399 } else if (mddev->external)
8400 seq_printf(seq, " super external:%s",
8401 mddev->metadata_type);
8403 seq_printf(seq, " super non-persistent");
8406 mddev->pers->status(seq, mddev);
8407 seq_printf(seq, "\n ");
8408 if (mddev->pers->sync_request) {
8409 if (status_resync(seq, mddev))
8410 seq_printf(seq, "\n ");
8413 seq_printf(seq, "\n ");
8415 md_bitmap_status(seq, mddev->bitmap);
8417 seq_printf(seq, "\n");
8419 spin_unlock(&mddev->lock);
8424 static const struct seq_operations md_seq_ops = {
8425 .start = md_seq_start,
8426 .next = md_seq_next,
8427 .stop = md_seq_stop,
8428 .show = md_seq_show,
8431 static int md_seq_open(struct inode *inode, struct file *file)
8433 struct seq_file *seq;
8436 error = seq_open(file, &md_seq_ops);
8440 seq = file->private_data;
8441 seq->poll_event = atomic_read(&md_event_count);
8445 static int md_unloading;
8446 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8448 struct seq_file *seq = filp->private_data;
8452 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8453 poll_wait(filp, &md_event_waiters, wait);
8455 /* always allow read */
8456 mask = EPOLLIN | EPOLLRDNORM;
8458 if (seq->poll_event != atomic_read(&md_event_count))
8459 mask |= EPOLLERR | EPOLLPRI;
8463 static const struct proc_ops mdstat_proc_ops = {
8464 .proc_open = md_seq_open,
8465 .proc_read = seq_read,
8466 .proc_lseek = seq_lseek,
8467 .proc_release = seq_release,
8468 .proc_poll = mdstat_poll,
8471 int register_md_personality(struct md_personality *p)
8473 pr_debug("md: %s personality registered for level %d\n",
8475 spin_lock(&pers_lock);
8476 list_add_tail(&p->list, &pers_list);
8477 spin_unlock(&pers_lock);
8480 EXPORT_SYMBOL(register_md_personality);
8482 int unregister_md_personality(struct md_personality *p)
8484 pr_debug("md: %s personality unregistered\n", p->name);
8485 spin_lock(&pers_lock);
8486 list_del_init(&p->list);
8487 spin_unlock(&pers_lock);
8490 EXPORT_SYMBOL(unregister_md_personality);
8492 int register_md_cluster_operations(struct md_cluster_operations *ops,
8493 struct module *module)
8496 spin_lock(&pers_lock);
8497 if (md_cluster_ops != NULL)
8500 md_cluster_ops = ops;
8501 md_cluster_mod = module;
8503 spin_unlock(&pers_lock);
8506 EXPORT_SYMBOL(register_md_cluster_operations);
8508 int unregister_md_cluster_operations(void)
8510 spin_lock(&pers_lock);
8511 md_cluster_ops = NULL;
8512 spin_unlock(&pers_lock);
8515 EXPORT_SYMBOL(unregister_md_cluster_operations);
8517 int md_setup_cluster(struct mddev *mddev, int nodes)
8520 if (!md_cluster_ops)
8521 request_module("md-cluster");
8522 spin_lock(&pers_lock);
8523 /* ensure module won't be unloaded */
8524 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8525 pr_warn("can't find md-cluster module or get its reference.\n");
8526 spin_unlock(&pers_lock);
8529 spin_unlock(&pers_lock);
8531 ret = md_cluster_ops->join(mddev, nodes);
8533 mddev->safemode_delay = 0;
8537 void md_cluster_stop(struct mddev *mddev)
8539 if (!md_cluster_ops)
8541 md_cluster_ops->leave(mddev);
8542 module_put(md_cluster_mod);
8545 static int is_mddev_idle(struct mddev *mddev, int init)
8547 struct md_rdev *rdev;
8553 rdev_for_each_rcu(rdev, mddev) {
8554 struct gendisk *disk = rdev->bdev->bd_disk;
8555 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8556 atomic_read(&disk->sync_io);
8557 /* sync IO will cause sync_io to increase before the disk_stats
8558 * as sync_io is counted when a request starts, and
8559 * disk_stats is counted when it completes.
8560 * So resync activity will cause curr_events to be smaller than
8561 * when there was no such activity.
8562 * non-sync IO will cause disk_stat to increase without
8563 * increasing sync_io so curr_events will (eventually)
8564 * be larger than it was before. Once it becomes
8565 * substantially larger, the test below will cause
8566 * the array to appear non-idle, and resync will slow
8568 * If there is a lot of outstanding resync activity when
8569 * we set last_event to curr_events, then all that activity
8570 * completing might cause the array to appear non-idle
8571 * and resync will be slowed down even though there might
8572 * not have been non-resync activity. This will only
8573 * happen once though. 'last_events' will soon reflect
8574 * the state where there is little or no outstanding
8575 * resync requests, and further resync activity will
8576 * always make curr_events less than last_events.
8579 if (init || curr_events - rdev->last_events > 64) {
8580 rdev->last_events = curr_events;
8588 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8590 /* another "blocks" (512byte) blocks have been synced */
8591 atomic_sub(blocks, &mddev->recovery_active);
8592 wake_up(&mddev->recovery_wait);
8594 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8595 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8596 md_wakeup_thread(mddev->thread);
8597 // stop recovery, signal do_sync ....
8600 EXPORT_SYMBOL(md_done_sync);
8602 /* md_write_start(mddev, bi)
8603 * If we need to update some array metadata (e.g. 'active' flag
8604 * in superblock) before writing, schedule a superblock update
8605 * and wait for it to complete.
8606 * A return value of 'false' means that the write wasn't recorded
8607 * and cannot proceed as the array is being suspend.
8609 bool md_write_start(struct mddev *mddev, struct bio *bi)
8613 if (bio_data_dir(bi) != WRITE)
8616 BUG_ON(mddev->ro == MD_RDONLY);
8617 if (mddev->ro == MD_AUTO_READ) {
8618 /* need to switch to read/write */
8619 mddev->ro = MD_RDWR;
8620 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8621 md_wakeup_thread(mddev->thread);
8622 md_wakeup_thread(mddev->sync_thread);
8626 percpu_ref_get(&mddev->writes_pending);
8627 smp_mb(); /* Match smp_mb in set_in_sync() */
8628 if (mddev->safemode == 1)
8629 mddev->safemode = 0;
8630 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8631 if (mddev->in_sync || mddev->sync_checkers) {
8632 spin_lock(&mddev->lock);
8633 if (mddev->in_sync) {
8635 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8636 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8637 md_wakeup_thread(mddev->thread);
8640 spin_unlock(&mddev->lock);
8644 sysfs_notify_dirent_safe(mddev->sysfs_state);
8645 if (!mddev->has_superblocks)
8647 wait_event(mddev->sb_wait,
8648 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8649 is_md_suspended(mddev));
8650 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8651 percpu_ref_put(&mddev->writes_pending);
8656 EXPORT_SYMBOL(md_write_start);
8658 /* md_write_inc can only be called when md_write_start() has
8659 * already been called at least once of the current request.
8660 * It increments the counter and is useful when a single request
8661 * is split into several parts. Each part causes an increment and
8662 * so needs a matching md_write_end().
8663 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8664 * a spinlocked region.
8666 void md_write_inc(struct mddev *mddev, struct bio *bi)
8668 if (bio_data_dir(bi) != WRITE)
8670 WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8671 percpu_ref_get(&mddev->writes_pending);
8673 EXPORT_SYMBOL(md_write_inc);
8675 void md_write_end(struct mddev *mddev)
8677 percpu_ref_put(&mddev->writes_pending);
8679 if (mddev->safemode == 2)
8680 md_wakeup_thread(mddev->thread);
8681 else if (mddev->safemode_delay)
8682 /* The roundup() ensures this only performs locking once
8683 * every ->safemode_delay jiffies
8685 mod_timer(&mddev->safemode_timer,
8686 roundup(jiffies, mddev->safemode_delay) +
8687 mddev->safemode_delay);
8690 EXPORT_SYMBOL(md_write_end);
8692 /* This is used by raid0 and raid10 */
8693 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8694 struct bio *bio, sector_t start, sector_t size)
8696 struct bio *discard_bio = NULL;
8698 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8699 &discard_bio) || !discard_bio)
8702 bio_chain(discard_bio, bio);
8703 bio_clone_blkg_association(discard_bio, bio);
8705 trace_block_bio_remap(discard_bio,
8706 disk_devt(mddev->gendisk),
8707 bio->bi_iter.bi_sector);
8708 submit_bio_noacct(discard_bio);
8710 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8712 static void md_end_clone_io(struct bio *bio)
8714 struct md_io_clone *md_io_clone = bio->bi_private;
8715 struct bio *orig_bio = md_io_clone->orig_bio;
8716 struct mddev *mddev = md_io_clone->mddev;
8718 if (bio->bi_status && !orig_bio->bi_status)
8719 orig_bio->bi_status = bio->bi_status;
8721 if (md_io_clone->start_time)
8722 bio_end_io_acct(orig_bio, md_io_clone->start_time);
8725 bio_endio(orig_bio);
8726 percpu_ref_put(&mddev->active_io);
8729 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8731 struct block_device *bdev = (*bio)->bi_bdev;
8732 struct md_io_clone *md_io_clone;
8734 bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8736 md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8737 md_io_clone->orig_bio = *bio;
8738 md_io_clone->mddev = mddev;
8739 if (blk_queue_io_stat(bdev->bd_disk->queue))
8740 md_io_clone->start_time = bio_start_io_acct(*bio);
8742 clone->bi_end_io = md_end_clone_io;
8743 clone->bi_private = md_io_clone;
8747 void md_account_bio(struct mddev *mddev, struct bio **bio)
8749 percpu_ref_get(&mddev->active_io);
8750 md_clone_bio(mddev, bio);
8752 EXPORT_SYMBOL_GPL(md_account_bio);
8754 /* md_allow_write(mddev)
8755 * Calling this ensures that the array is marked 'active' so that writes
8756 * may proceed without blocking. It is important to call this before
8757 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8758 * Must be called with mddev_lock held.
8760 void md_allow_write(struct mddev *mddev)
8764 if (!md_is_rdwr(mddev))
8766 if (!mddev->pers->sync_request)
8769 spin_lock(&mddev->lock);
8770 if (mddev->in_sync) {
8772 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8773 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8774 if (mddev->safemode_delay &&
8775 mddev->safemode == 0)
8776 mddev->safemode = 1;
8777 spin_unlock(&mddev->lock);
8778 md_update_sb(mddev, 0);
8779 sysfs_notify_dirent_safe(mddev->sysfs_state);
8780 /* wait for the dirty state to be recorded in the metadata */
8781 wait_event(mddev->sb_wait,
8782 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8784 spin_unlock(&mddev->lock);
8786 EXPORT_SYMBOL_GPL(md_allow_write);
8788 #define SYNC_MARKS 10
8789 #define SYNC_MARK_STEP (3*HZ)
8790 #define UPDATE_FREQUENCY (5*60*HZ)
8791 void md_do_sync(struct md_thread *thread)
8793 struct mddev *mddev = thread->mddev;
8794 struct mddev *mddev2;
8795 unsigned int currspeed = 0, window;
8796 sector_t max_sectors,j, io_sectors, recovery_done;
8797 unsigned long mark[SYNC_MARKS];
8798 unsigned long update_time;
8799 sector_t mark_cnt[SYNC_MARKS];
8801 sector_t last_check;
8803 struct md_rdev *rdev;
8804 char *desc, *action = NULL;
8805 struct blk_plug plug;
8808 /* just incase thread restarts... */
8809 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8810 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8812 if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8813 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8817 if (mddev_is_clustered(mddev)) {
8818 ret = md_cluster_ops->resync_start(mddev);
8822 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8823 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8824 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8825 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8826 && ((unsigned long long)mddev->curr_resync_completed
8827 < (unsigned long long)mddev->resync_max_sectors))
8831 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8832 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8833 desc = "data-check";
8835 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8836 desc = "requested-resync";
8840 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8845 mddev->last_sync_action = action ?: desc;
8848 * Before starting a resync we must have set curr_resync to
8849 * 2, and then checked that every "conflicting" array has curr_resync
8850 * less than ours. When we find one that is the same or higher
8851 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8852 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8853 * This will mean we have to start checking from the beginning again.
8858 int mddev2_minor = -1;
8859 mddev->curr_resync = MD_RESYNC_DELAYED;
8862 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8864 spin_lock(&all_mddevs_lock);
8865 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8866 if (test_bit(MD_DELETED, &mddev2->flags))
8868 if (mddev2 == mddev)
8870 if (!mddev->parallel_resync
8871 && mddev2->curr_resync
8872 && match_mddev_units(mddev, mddev2)) {
8874 if (mddev < mddev2 &&
8875 mddev->curr_resync == MD_RESYNC_DELAYED) {
8876 /* arbitrarily yield */
8877 mddev->curr_resync = MD_RESYNC_YIELDED;
8878 wake_up(&resync_wait);
8880 if (mddev > mddev2 &&
8881 mddev->curr_resync == MD_RESYNC_YIELDED)
8882 /* no need to wait here, we can wait the next
8883 * time 'round when curr_resync == 2
8886 /* We need to wait 'interruptible' so as not to
8887 * contribute to the load average, and not to
8888 * be caught by 'softlockup'
8890 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8891 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8892 mddev2->curr_resync >= mddev->curr_resync) {
8893 if (mddev2_minor != mddev2->md_minor) {
8894 mddev2_minor = mddev2->md_minor;
8895 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8896 desc, mdname(mddev),
8899 spin_unlock(&all_mddevs_lock);
8901 if (signal_pending(current))
8902 flush_signals(current);
8904 finish_wait(&resync_wait, &wq);
8907 finish_wait(&resync_wait, &wq);
8910 spin_unlock(&all_mddevs_lock);
8911 } while (mddev->curr_resync < MD_RESYNC_DELAYED);
8914 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8915 /* resync follows the size requested by the personality,
8916 * which defaults to physical size, but can be virtual size
8918 max_sectors = mddev->resync_max_sectors;
8919 atomic64_set(&mddev->resync_mismatches, 0);
8920 /* we don't use the checkpoint if there's a bitmap */
8921 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8922 j = mddev->resync_min;
8923 else if (!mddev->bitmap)
8924 j = mddev->recovery_cp;
8926 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8927 max_sectors = mddev->resync_max_sectors;
8929 * If the original node aborts reshaping then we continue the
8930 * reshaping, so set j again to avoid restart reshape from the
8933 if (mddev_is_clustered(mddev) &&
8934 mddev->reshape_position != MaxSector)
8935 j = mddev->reshape_position;
8937 /* recovery follows the physical size of devices */
8938 max_sectors = mddev->dev_sectors;
8941 rdev_for_each_rcu(rdev, mddev)
8942 if (rdev->raid_disk >= 0 &&
8943 !test_bit(Journal, &rdev->flags) &&
8944 !test_bit(Faulty, &rdev->flags) &&
8945 !test_bit(In_sync, &rdev->flags) &&
8946 rdev->recovery_offset < j)
8947 j = rdev->recovery_offset;
8950 /* If there is a bitmap, we need to make sure all
8951 * writes that started before we added a spare
8952 * complete before we start doing a recovery.
8953 * Otherwise the write might complete and (via
8954 * bitmap_endwrite) set a bit in the bitmap after the
8955 * recovery has checked that bit and skipped that
8958 if (mddev->bitmap) {
8959 mddev->pers->quiesce(mddev, 1);
8960 mddev->pers->quiesce(mddev, 0);
8964 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8965 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8966 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8967 speed_max(mddev), desc);
8969 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8972 for (m = 0; m < SYNC_MARKS; m++) {
8974 mark_cnt[m] = io_sectors;
8977 mddev->resync_mark = mark[last_mark];
8978 mddev->resync_mark_cnt = mark_cnt[last_mark];
8981 * Tune reconstruction:
8983 window = 32 * (PAGE_SIZE / 512);
8984 pr_debug("md: using %dk window, over a total of %lluk.\n",
8985 window/2, (unsigned long long)max_sectors/2);
8987 atomic_set(&mddev->recovery_active, 0);
8990 if (j >= MD_RESYNC_ACTIVE) {
8991 pr_debug("md: resuming %s of %s from checkpoint.\n",
8992 desc, mdname(mddev));
8993 mddev->curr_resync = j;
8995 mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
8996 mddev->curr_resync_completed = j;
8997 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8999 update_time = jiffies;
9001 blk_start_plug(&plug);
9002 while (j < max_sectors) {
9007 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9008 ((mddev->curr_resync > mddev->curr_resync_completed &&
9009 (mddev->curr_resync - mddev->curr_resync_completed)
9010 > (max_sectors >> 4)) ||
9011 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
9012 (j - mddev->curr_resync_completed)*2
9013 >= mddev->resync_max - mddev->curr_resync_completed ||
9014 mddev->curr_resync_completed > mddev->resync_max
9016 /* time to update curr_resync_completed */
9017 wait_event(mddev->recovery_wait,
9018 atomic_read(&mddev->recovery_active) == 0);
9019 mddev->curr_resync_completed = j;
9020 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
9021 j > mddev->recovery_cp)
9022 mddev->recovery_cp = j;
9023 update_time = jiffies;
9024 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
9025 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9028 while (j >= mddev->resync_max &&
9029 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9030 /* As this condition is controlled by user-space,
9031 * we can block indefinitely, so use '_interruptible'
9032 * to avoid triggering warnings.
9034 flush_signals(current); /* just in case */
9035 wait_event_interruptible(mddev->recovery_wait,
9036 mddev->resync_max > j
9037 || test_bit(MD_RECOVERY_INTR,
9041 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9044 sectors = mddev->pers->sync_request(mddev, j, &skipped);
9046 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9050 if (!skipped) { /* actual IO requested */
9051 io_sectors += sectors;
9052 atomic_add(sectors, &mddev->recovery_active);
9055 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9059 if (j > max_sectors)
9060 /* when skipping, extra large numbers can be returned. */
9062 if (j >= MD_RESYNC_ACTIVE)
9063 mddev->curr_resync = j;
9064 mddev->curr_mark_cnt = io_sectors;
9065 if (last_check == 0)
9066 /* this is the earliest that rebuild will be
9067 * visible in /proc/mdstat
9071 if (last_check + window > io_sectors || j == max_sectors)
9074 last_check = io_sectors;
9076 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9078 int next = (last_mark+1) % SYNC_MARKS;
9080 mddev->resync_mark = mark[next];
9081 mddev->resync_mark_cnt = mark_cnt[next];
9082 mark[next] = jiffies;
9083 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9087 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9091 * this loop exits only if either when we are slower than
9092 * the 'hard' speed limit, or the system was IO-idle for
9094 * the system might be non-idle CPU-wise, but we only care
9095 * about not overloading the IO subsystem. (things like an
9096 * e2fsck being done on the RAID array should execute fast)
9100 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9101 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9102 /((jiffies-mddev->resync_mark)/HZ +1) +1;
9104 if (currspeed > speed_min(mddev)) {
9105 if (currspeed > speed_max(mddev)) {
9109 if (!is_mddev_idle(mddev, 0)) {
9111 * Give other IO more of a chance.
9112 * The faster the devices, the less we wait.
9114 wait_event(mddev->recovery_wait,
9115 !atomic_read(&mddev->recovery_active));
9119 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9120 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9121 ? "interrupted" : "done");
9123 * this also signals 'finished resyncing' to md_stop
9125 blk_finish_plug(&plug);
9126 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9128 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9129 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9130 mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9131 mddev->curr_resync_completed = mddev->curr_resync;
9132 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9134 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9136 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9137 mddev->curr_resync > MD_RESYNC_ACTIVE) {
9138 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9139 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9140 if (mddev->curr_resync >= mddev->recovery_cp) {
9141 pr_debug("md: checkpointing %s of %s.\n",
9142 desc, mdname(mddev));
9143 if (test_bit(MD_RECOVERY_ERROR,
9145 mddev->recovery_cp =
9146 mddev->curr_resync_completed;
9148 mddev->recovery_cp =
9152 mddev->recovery_cp = MaxSector;
9154 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9155 mddev->curr_resync = MaxSector;
9156 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9157 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9159 rdev_for_each_rcu(rdev, mddev)
9160 if (rdev->raid_disk >= 0 &&
9161 mddev->delta_disks >= 0 &&
9162 !test_bit(Journal, &rdev->flags) &&
9163 !test_bit(Faulty, &rdev->flags) &&
9164 !test_bit(In_sync, &rdev->flags) &&
9165 rdev->recovery_offset < mddev->curr_resync)
9166 rdev->recovery_offset = mddev->curr_resync;
9172 /* set CHANGE_PENDING here since maybe another update is needed,
9173 * so other nodes are informed. It should be harmless for normal
9175 set_mask_bits(&mddev->sb_flags, 0,
9176 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9178 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9179 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9180 mddev->delta_disks > 0 &&
9181 mddev->pers->finish_reshape &&
9182 mddev->pers->size &&
9184 mddev_lock_nointr(mddev);
9185 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9186 mddev_unlock(mddev);
9187 if (!mddev_is_clustered(mddev))
9188 set_capacity_and_notify(mddev->gendisk,
9189 mddev->array_sectors);
9192 spin_lock(&mddev->lock);
9193 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9194 /* We completed so min/max setting can be forgotten if used. */
9195 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9196 mddev->resync_min = 0;
9197 mddev->resync_max = MaxSector;
9198 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9199 mddev->resync_min = mddev->curr_resync_completed;
9200 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9201 mddev->curr_resync = MD_RESYNC_NONE;
9202 spin_unlock(&mddev->lock);
9204 wake_up(&resync_wait);
9205 wake_up(&mddev->sb_wait);
9206 md_wakeup_thread(mddev->thread);
9209 EXPORT_SYMBOL_GPL(md_do_sync);
9211 static int remove_and_add_spares(struct mddev *mddev,
9212 struct md_rdev *this)
9214 struct md_rdev *rdev;
9217 bool remove_some = false;
9219 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9220 /* Mustn't remove devices when resync thread is running */
9223 rdev_for_each(rdev, mddev) {
9224 if ((this == NULL || rdev == this) &&
9225 rdev->raid_disk >= 0 &&
9226 !test_bit(Blocked, &rdev->flags) &&
9227 test_bit(Faulty, &rdev->flags) &&
9228 atomic_read(&rdev->nr_pending)==0) {
9229 /* Faulty non-Blocked devices with nr_pending == 0
9230 * never get nr_pending incremented,
9231 * never get Faulty cleared, and never get Blocked set.
9232 * So we can synchronize_rcu now rather than once per device
9235 set_bit(RemoveSynchronized, &rdev->flags);
9241 rdev_for_each(rdev, mddev) {
9242 if ((this == NULL || rdev == this) &&
9243 rdev->raid_disk >= 0 &&
9244 !test_bit(Blocked, &rdev->flags) &&
9245 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9246 (!test_bit(In_sync, &rdev->flags) &&
9247 !test_bit(Journal, &rdev->flags))) &&
9248 atomic_read(&rdev->nr_pending)==0)) {
9249 if (mddev->pers->hot_remove_disk(
9250 mddev, rdev) == 0) {
9251 sysfs_unlink_rdev(mddev, rdev);
9252 rdev->saved_raid_disk = rdev->raid_disk;
9253 rdev->raid_disk = -1;
9257 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9258 clear_bit(RemoveSynchronized, &rdev->flags);
9261 if (removed && mddev->kobj.sd)
9262 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9264 if (this && removed)
9267 rdev_for_each(rdev, mddev) {
9268 if (this && this != rdev)
9270 if (test_bit(Candidate, &rdev->flags))
9272 if (rdev->raid_disk >= 0 &&
9273 !test_bit(In_sync, &rdev->flags) &&
9274 !test_bit(Journal, &rdev->flags) &&
9275 !test_bit(Faulty, &rdev->flags))
9277 if (rdev->raid_disk >= 0)
9279 if (test_bit(Faulty, &rdev->flags))
9281 if (!test_bit(Journal, &rdev->flags)) {
9282 if (!md_is_rdwr(mddev) &&
9283 !(rdev->saved_raid_disk >= 0 &&
9284 !test_bit(Bitmap_sync, &rdev->flags)))
9287 rdev->recovery_offset = 0;
9289 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9290 /* failure here is OK */
9291 sysfs_link_rdev(mddev, rdev);
9292 if (!test_bit(Journal, &rdev->flags))
9295 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9300 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9304 static void md_start_sync(struct work_struct *ws)
9306 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9308 rcu_assign_pointer(mddev->sync_thread,
9309 md_register_thread(md_do_sync, mddev, "resync"));
9310 if (!mddev->sync_thread) {
9311 pr_warn("%s: could not start resync thread...\n",
9313 /* leave the spares where they are, it shouldn't hurt */
9314 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9315 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9316 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9317 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9318 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9319 wake_up(&resync_wait);
9320 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9322 if (mddev->sysfs_action)
9323 sysfs_notify_dirent_safe(mddev->sysfs_action);
9325 md_wakeup_thread(mddev->sync_thread);
9326 sysfs_notify_dirent_safe(mddev->sysfs_action);
9331 * This routine is regularly called by all per-raid-array threads to
9332 * deal with generic issues like resync and super-block update.
9333 * Raid personalities that don't have a thread (linear/raid0) do not
9334 * need this as they never do any recovery or update the superblock.
9336 * It does not do any resync itself, but rather "forks" off other threads
9337 * to do that as needed.
9338 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9339 * "->recovery" and create a thread at ->sync_thread.
9340 * When the thread finishes it sets MD_RECOVERY_DONE
9341 * and wakeups up this thread which will reap the thread and finish up.
9342 * This thread also removes any faulty devices (with nr_pending == 0).
9344 * The overall approach is:
9345 * 1/ if the superblock needs updating, update it.
9346 * 2/ If a recovery thread is running, don't do anything else.
9347 * 3/ If recovery has finished, clean up, possibly marking spares active.
9348 * 4/ If there are any faulty devices, remove them.
9349 * 5/ If array is degraded, try to add spares devices
9350 * 6/ If array has spares or is not in-sync, start a resync thread.
9352 void md_check_recovery(struct mddev *mddev)
9354 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9355 /* Write superblock - thread that called mddev_suspend()
9356 * holds reconfig_mutex for us.
9358 set_bit(MD_UPDATING_SB, &mddev->flags);
9359 smp_mb__after_atomic();
9360 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9361 md_update_sb(mddev, 0);
9362 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9363 wake_up(&mddev->sb_wait);
9366 if (is_md_suspended(mddev))
9370 md_bitmap_daemon_work(mddev);
9372 if (signal_pending(current)) {
9373 if (mddev->pers->sync_request && !mddev->external) {
9374 pr_debug("md: %s in immediate safe mode\n",
9376 mddev->safemode = 2;
9378 flush_signals(current);
9381 if (!md_is_rdwr(mddev) &&
9382 !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9385 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9386 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9387 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9388 (mddev->external == 0 && mddev->safemode == 1) ||
9389 (mddev->safemode == 2
9390 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9394 if (mddev_trylock(mddev)) {
9396 bool try_set_sync = mddev->safemode != 0;
9398 if (!mddev->external && mddev->safemode == 1)
9399 mddev->safemode = 0;
9401 if (!md_is_rdwr(mddev)) {
9402 struct md_rdev *rdev;
9403 if (!mddev->external && mddev->in_sync)
9404 /* 'Blocked' flag not needed as failed devices
9405 * will be recorded if array switched to read/write.
9406 * Leaving it set will prevent the device
9407 * from being removed.
9409 rdev_for_each(rdev, mddev)
9410 clear_bit(Blocked, &rdev->flags);
9411 /* On a read-only array we can:
9412 * - remove failed devices
9413 * - add already-in_sync devices if the array itself
9415 * As we only add devices that are already in-sync,
9416 * we can activate the spares immediately.
9418 remove_and_add_spares(mddev, NULL);
9419 /* There is no thread, but we need to call
9420 * ->spare_active and clear saved_raid_disk
9422 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9423 md_reap_sync_thread(mddev);
9424 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9425 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9426 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9430 if (mddev_is_clustered(mddev)) {
9431 struct md_rdev *rdev, *tmp;
9432 /* kick the device if another node issued a
9435 rdev_for_each_safe(rdev, tmp, mddev) {
9436 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9437 rdev->raid_disk < 0)
9438 md_kick_rdev_from_array(rdev);
9442 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9443 spin_lock(&mddev->lock);
9445 spin_unlock(&mddev->lock);
9448 if (mddev->sb_flags)
9449 md_update_sb(mddev, 0);
9452 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9455 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9456 if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9457 /* resync/recovery still happening */
9458 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9462 if (WARN_ON_ONCE(!mddev->sync_thread))
9465 md_reap_sync_thread(mddev);
9469 /* Set RUNNING before clearing NEEDED to avoid
9470 * any transients in the value of "sync_action".
9472 mddev->curr_resync_completed = 0;
9473 spin_lock(&mddev->lock);
9474 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9475 spin_unlock(&mddev->lock);
9476 /* Clear some bits that don't mean anything, but
9479 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9480 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9482 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9483 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9485 /* no recovery is running.
9486 * remove any failed drives, then
9487 * add spares if possible.
9488 * Spares are also removed and re-added, to allow
9489 * the personality to fail the re-add.
9492 if (mddev->reshape_position != MaxSector) {
9493 if (mddev->pers->check_reshape == NULL ||
9494 mddev->pers->check_reshape(mddev) != 0)
9495 /* Cannot proceed */
9497 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9498 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9499 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9500 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9501 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9502 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9503 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9504 } else if (mddev->recovery_cp < MaxSector) {
9505 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9506 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9507 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9508 /* nothing to be done ... */
9511 if (mddev->pers->sync_request) {
9513 /* We are adding a device or devices to an array
9514 * which has the bitmap stored on all devices.
9515 * So make sure all bitmap pages get written
9517 md_bitmap_write_all(mddev->bitmap);
9519 INIT_WORK(&mddev->del_work, md_start_sync);
9520 queue_work(md_misc_wq, &mddev->del_work);
9524 if (!mddev->sync_thread) {
9525 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9526 wake_up(&resync_wait);
9527 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9529 if (mddev->sysfs_action)
9530 sysfs_notify_dirent_safe(mddev->sysfs_action);
9533 wake_up(&mddev->sb_wait);
9534 mddev_unlock(mddev);
9537 EXPORT_SYMBOL(md_check_recovery);
9539 void md_reap_sync_thread(struct mddev *mddev)
9541 struct md_rdev *rdev;
9542 sector_t old_dev_sectors = mddev->dev_sectors;
9543 bool is_reshaped = false;
9545 /* resync has finished, collect result */
9546 md_unregister_thread(mddev, &mddev->sync_thread);
9547 atomic_inc(&mddev->sync_seq);
9549 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9550 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9551 mddev->degraded != mddev->raid_disks) {
9553 /* activate any spares */
9554 if (mddev->pers->spare_active(mddev)) {
9555 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9556 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9559 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9560 mddev->pers->finish_reshape) {
9561 mddev->pers->finish_reshape(mddev);
9562 if (mddev_is_clustered(mddev))
9566 /* If array is no-longer degraded, then any saved_raid_disk
9567 * information must be scrapped.
9569 if (!mddev->degraded)
9570 rdev_for_each(rdev, mddev)
9571 rdev->saved_raid_disk = -1;
9573 md_update_sb(mddev, 1);
9574 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9575 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9577 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9578 md_cluster_ops->resync_finish(mddev);
9579 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9580 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9581 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9582 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9583 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9584 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9586 * We call md_cluster_ops->update_size here because sync_size could
9587 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9588 * so it is time to update size across cluster.
9590 if (mddev_is_clustered(mddev) && is_reshaped
9591 && !test_bit(MD_CLOSING, &mddev->flags))
9592 md_cluster_ops->update_size(mddev, old_dev_sectors);
9593 /* flag recovery needed just to double check */
9594 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9595 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9596 sysfs_notify_dirent_safe(mddev->sysfs_action);
9598 if (mddev->event_work.func)
9599 queue_work(md_misc_wq, &mddev->event_work);
9600 wake_up(&resync_wait);
9602 EXPORT_SYMBOL(md_reap_sync_thread);
9604 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9606 sysfs_notify_dirent_safe(rdev->sysfs_state);
9607 wait_event_timeout(rdev->blocked_wait,
9608 !test_bit(Blocked, &rdev->flags) &&
9609 !test_bit(BlockedBadBlocks, &rdev->flags),
9610 msecs_to_jiffies(5000));
9611 rdev_dec_pending(rdev, mddev);
9613 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9615 void md_finish_reshape(struct mddev *mddev)
9617 /* called be personality module when reshape completes. */
9618 struct md_rdev *rdev;
9620 rdev_for_each(rdev, mddev) {
9621 if (rdev->data_offset > rdev->new_data_offset)
9622 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9624 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9625 rdev->data_offset = rdev->new_data_offset;
9628 EXPORT_SYMBOL(md_finish_reshape);
9630 /* Bad block management */
9632 /* Returns 1 on success, 0 on failure */
9633 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9636 struct mddev *mddev = rdev->mddev;
9639 s += rdev->new_data_offset;
9641 s += rdev->data_offset;
9642 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9644 /* Make sure they get written out promptly */
9645 if (test_bit(ExternalBbl, &rdev->flags))
9646 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9647 sysfs_notify_dirent_safe(rdev->sysfs_state);
9648 set_mask_bits(&mddev->sb_flags, 0,
9649 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9650 md_wakeup_thread(rdev->mddev->thread);
9655 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9657 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9662 s += rdev->new_data_offset;
9664 s += rdev->data_offset;
9665 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9666 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9667 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9670 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9672 static int md_notify_reboot(struct notifier_block *this,
9673 unsigned long code, void *x)
9675 struct mddev *mddev, *n;
9678 spin_lock(&all_mddevs_lock);
9679 list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9680 if (!mddev_get(mddev))
9682 spin_unlock(&all_mddevs_lock);
9683 if (mddev_trylock(mddev)) {
9685 __md_stop_writes(mddev);
9686 if (mddev->persistent)
9687 mddev->safemode = 2;
9688 mddev_unlock(mddev);
9692 spin_lock(&all_mddevs_lock);
9694 spin_unlock(&all_mddevs_lock);
9697 * certain more exotic SCSI devices are known to be
9698 * volatile wrt too early system reboots. While the
9699 * right place to handle this issue is the given
9700 * driver, we do want to have a safe RAID driver ...
9708 static struct notifier_block md_notifier = {
9709 .notifier_call = md_notify_reboot,
9711 .priority = INT_MAX, /* before any real devices */
9714 static void md_geninit(void)
9716 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9718 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9721 static int __init md_init(void)
9725 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9729 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9733 md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9738 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9742 ret = __register_blkdev(0, "mdp", md_probe);
9747 register_reboot_notifier(&md_notifier);
9748 raid_table_header = register_sysctl("dev/raid", raid_table);
9754 unregister_blkdev(MD_MAJOR, "md");
9756 destroy_workqueue(md_bitmap_wq);
9758 destroy_workqueue(md_misc_wq);
9760 destroy_workqueue(md_wq);
9765 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9767 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9768 struct md_rdev *rdev2, *tmp;
9772 * If size is changed in another node then we need to
9773 * do resize as well.
9775 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9776 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9778 pr_info("md-cluster: resize failed\n");
9780 md_bitmap_update_sb(mddev->bitmap);
9783 /* Check for change of roles in the active devices */
9784 rdev_for_each_safe(rdev2, tmp, mddev) {
9785 if (test_bit(Faulty, &rdev2->flags))
9788 /* Check if the roles changed */
9789 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9791 if (test_bit(Candidate, &rdev2->flags)) {
9792 if (role == MD_DISK_ROLE_FAULTY) {
9793 pr_info("md: Removing Candidate device %pg because add failed\n",
9795 md_kick_rdev_from_array(rdev2);
9799 clear_bit(Candidate, &rdev2->flags);
9802 if (role != rdev2->raid_disk) {
9804 * got activated except reshape is happening.
9806 if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9807 !(le32_to_cpu(sb->feature_map) &
9808 MD_FEATURE_RESHAPE_ACTIVE)) {
9809 rdev2->saved_raid_disk = role;
9810 ret = remove_and_add_spares(mddev, rdev2);
9811 pr_info("Activated spare: %pg\n",
9813 /* wakeup mddev->thread here, so array could
9814 * perform resync with the new activated disk */
9815 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9816 md_wakeup_thread(mddev->thread);
9819 * We just want to do the minimum to mark the disk
9820 * as faulty. The recovery is performed by the
9821 * one who initiated the error.
9823 if (role == MD_DISK_ROLE_FAULTY ||
9824 role == MD_DISK_ROLE_JOURNAL) {
9825 md_error(mddev, rdev2);
9826 clear_bit(Blocked, &rdev2->flags);
9831 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9832 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9834 pr_warn("md: updating array disks failed. %d\n", ret);
9838 * Since mddev->delta_disks has already updated in update_raid_disks,
9839 * so it is time to check reshape.
9841 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9842 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9844 * reshape is happening in the remote node, we need to
9845 * update reshape_position and call start_reshape.
9847 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9848 if (mddev->pers->update_reshape_pos)
9849 mddev->pers->update_reshape_pos(mddev);
9850 if (mddev->pers->start_reshape)
9851 mddev->pers->start_reshape(mddev);
9852 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9853 mddev->reshape_position != MaxSector &&
9854 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9855 /* reshape is just done in another node. */
9856 mddev->reshape_position = MaxSector;
9857 if (mddev->pers->update_reshape_pos)
9858 mddev->pers->update_reshape_pos(mddev);
9861 /* Finally set the event to be up to date */
9862 mddev->events = le64_to_cpu(sb->events);
9865 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9868 struct page *swapout = rdev->sb_page;
9869 struct mdp_superblock_1 *sb;
9871 /* Store the sb page of the rdev in the swapout temporary
9872 * variable in case we err in the future
9874 rdev->sb_page = NULL;
9875 err = alloc_disk_sb(rdev);
9877 ClearPageUptodate(rdev->sb_page);
9878 rdev->sb_loaded = 0;
9879 err = super_types[mddev->major_version].
9880 load_super(rdev, NULL, mddev->minor_version);
9883 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9884 __func__, __LINE__, rdev->desc_nr, err);
9886 put_page(rdev->sb_page);
9887 rdev->sb_page = swapout;
9888 rdev->sb_loaded = 1;
9892 sb = page_address(rdev->sb_page);
9893 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9897 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9898 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9900 /* The other node finished recovery, call spare_active to set
9901 * device In_sync and mddev->degraded
9903 if (rdev->recovery_offset == MaxSector &&
9904 !test_bit(In_sync, &rdev->flags) &&
9905 mddev->pers->spare_active(mddev))
9906 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9912 void md_reload_sb(struct mddev *mddev, int nr)
9914 struct md_rdev *rdev = NULL, *iter;
9918 rdev_for_each_rcu(iter, mddev) {
9919 if (iter->desc_nr == nr) {
9926 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9930 err = read_rdev(mddev, rdev);
9934 check_sb_changes(mddev, rdev);
9936 /* Read all rdev's to update recovery_offset */
9937 rdev_for_each_rcu(rdev, mddev) {
9938 if (!test_bit(Faulty, &rdev->flags))
9939 read_rdev(mddev, rdev);
9942 EXPORT_SYMBOL(md_reload_sb);
9947 * Searches all registered partitions for autorun RAID arrays
9951 static DEFINE_MUTEX(detected_devices_mutex);
9952 static LIST_HEAD(all_detected_devices);
9953 struct detected_devices_node {
9954 struct list_head list;
9958 void md_autodetect_dev(dev_t dev)
9960 struct detected_devices_node *node_detected_dev;
9962 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9963 if (node_detected_dev) {
9964 node_detected_dev->dev = dev;
9965 mutex_lock(&detected_devices_mutex);
9966 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9967 mutex_unlock(&detected_devices_mutex);
9971 void md_autostart_arrays(int part)
9973 struct md_rdev *rdev;
9974 struct detected_devices_node *node_detected_dev;
9976 int i_scanned, i_passed;
9981 pr_info("md: Autodetecting RAID arrays.\n");
9983 mutex_lock(&detected_devices_mutex);
9984 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9986 node_detected_dev = list_entry(all_detected_devices.next,
9987 struct detected_devices_node, list);
9988 list_del(&node_detected_dev->list);
9989 dev = node_detected_dev->dev;
9990 kfree(node_detected_dev);
9991 mutex_unlock(&detected_devices_mutex);
9992 rdev = md_import_device(dev,0, 90);
9993 mutex_lock(&detected_devices_mutex);
9997 if (test_bit(Faulty, &rdev->flags))
10000 set_bit(AutoDetected, &rdev->flags);
10001 list_add(&rdev->same_set, &pending_raid_disks);
10004 mutex_unlock(&detected_devices_mutex);
10006 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
10008 autorun_devices(part);
10011 #endif /* !MODULE */
10013 static __exit void md_exit(void)
10015 struct mddev *mddev, *n;
10018 unregister_blkdev(MD_MAJOR,"md");
10019 unregister_blkdev(mdp_major, "mdp");
10020 unregister_reboot_notifier(&md_notifier);
10021 unregister_sysctl_table(raid_table_header);
10023 /* We cannot unload the modules while some process is
10024 * waiting for us in select() or poll() - wake them up
10027 while (waitqueue_active(&md_event_waiters)) {
10028 /* not safe to leave yet */
10029 wake_up(&md_event_waiters);
10033 remove_proc_entry("mdstat", NULL);
10035 spin_lock(&all_mddevs_lock);
10036 list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
10037 if (!mddev_get(mddev))
10039 spin_unlock(&all_mddevs_lock);
10040 export_array(mddev);
10042 mddev->hold_active = 0;
10044 * As the mddev is now fully clear, mddev_put will schedule
10045 * the mddev for destruction by a workqueue, and the
10046 * destroy_workqueue() below will wait for that to complete.
10049 spin_lock(&all_mddevs_lock);
10051 spin_unlock(&all_mddevs_lock);
10053 destroy_workqueue(md_misc_wq);
10054 destroy_workqueue(md_bitmap_wq);
10055 destroy_workqueue(md_wq);
10058 subsys_initcall(md_init);
10059 module_exit(md_exit)
10061 static int get_ro(char *buffer, const struct kernel_param *kp)
10063 return sprintf(buffer, "%d\n", start_readonly);
10065 static int set_ro(const char *val, const struct kernel_param *kp)
10067 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10070 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10071 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10072 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10073 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10075 MODULE_LICENSE("GPL");
10076 MODULE_DESCRIPTION("MD RAID framework");
10077 MODULE_ALIAS("md");
10078 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);