1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/module.h>
55 #include <linux/reboot.h>
56 #include <linux/file.h>
57 #include <linux/compat.h>
58 #include <linux/delay.h>
59 #include <linux/raid/md_p.h>
60 #include <linux/raid/md_u.h>
61 #include <linux/raid/detect.h>
62 #include <linux/slab.h>
63 #include <linux/percpu-refcount.h>
64 #include <linux/part_stat.h>
66 #include <trace/events/block.h>
68 #include "md-bitmap.h"
69 #include "md-cluster.h"
71 /* pers_list is a list of registered personalities protected
73 * pers_lock does extra service to protect accesses to
74 * mddev->thread when the mutex cannot be held.
76 static LIST_HEAD(pers_list);
77 static DEFINE_SPINLOCK(pers_lock);
79 static struct kobj_type md_ktype;
81 struct md_cluster_operations *md_cluster_ops;
82 EXPORT_SYMBOL(md_cluster_ops);
83 static struct module *md_cluster_mod;
85 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
86 static struct workqueue_struct *md_wq;
87 static struct workqueue_struct *md_misc_wq;
88 static struct workqueue_struct *md_rdev_misc_wq;
90 static int remove_and_add_spares(struct mddev *mddev,
91 struct md_rdev *this);
92 static void mddev_detach(struct mddev *mddev);
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 struct ctl_table raid_dir_table[] = {
315 .mode = S_IRUGO|S_IXUGO,
321 static struct ctl_table raid_root_table[] = {
326 .child = raid_dir_table,
331 static int start_readonly;
334 * The original mechanism for creating an md device is to create
335 * a device node in /dev and to open it. This causes races with device-close.
336 * The preferred method is to write to the "new_array" module parameter.
337 * This can avoid races.
338 * Setting create_on_open to false disables the original mechanism
339 * so all the races disappear.
341 static bool create_on_open = true;
343 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
346 if (!mddev || !bioset_initialized(&mddev->bio_set))
347 return bio_alloc(gfp_mask, nr_iovecs);
349 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
351 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
353 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
355 if (!mddev || !bioset_initialized(&mddev->sync_set))
356 return bio_alloc(GFP_NOIO, 1);
358 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
362 * We have a system wide 'event count' that is incremented
363 * on any 'interesting' event, and readers of /proc/mdstat
364 * can use 'poll' or 'select' to find out when the event
368 * start array, stop array, error, add device, remove device,
369 * start build, activate spare
371 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
372 static atomic_t md_event_count;
373 void md_new_event(struct mddev *mddev)
375 atomic_inc(&md_event_count);
376 wake_up(&md_event_waiters);
378 EXPORT_SYMBOL_GPL(md_new_event);
381 * Enables to iterate over all existing md arrays
382 * all_mddevs_lock protects this list.
384 static LIST_HEAD(all_mddevs);
385 static DEFINE_SPINLOCK(all_mddevs_lock);
388 * iterates through all used mddevs in the system.
389 * We take care to grab the all_mddevs_lock whenever navigating
390 * the list, and to always hold a refcount when unlocked.
391 * Any code which breaks out of this loop while own
392 * a reference to the current mddev and must mddev_put it.
394 #define for_each_mddev(_mddev,_tmp) \
396 for (({ spin_lock(&all_mddevs_lock); \
397 _tmp = all_mddevs.next; \
399 ({ if (_tmp != &all_mddevs) \
400 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
401 spin_unlock(&all_mddevs_lock); \
402 if (_mddev) mddev_put(_mddev); \
403 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
404 _tmp != &all_mddevs;}); \
405 ({ spin_lock(&all_mddevs_lock); \
406 _tmp = _tmp->next;}) \
409 /* Rather than calling directly into the personality make_request function,
410 * IO requests come here first so that we can check if the device is
411 * being suspended pending a reconfiguration.
412 * We hold a refcount over the call to ->make_request. By the time that
413 * call has finished, the bio has been linked into some internal structure
414 * and so is visible to ->quiesce(), so we don't need the refcount any more.
416 static bool is_suspended(struct mddev *mddev, struct bio *bio)
418 if (mddev->suspended)
420 if (bio_data_dir(bio) != WRITE)
422 if (mddev->suspend_lo >= mddev->suspend_hi)
424 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
426 if (bio_end_sector(bio) < mddev->suspend_lo)
431 void md_handle_request(struct mddev *mddev, struct bio *bio)
435 if (is_suspended(mddev, bio)) {
438 prepare_to_wait(&mddev->sb_wait, &__wait,
439 TASK_UNINTERRUPTIBLE);
440 if (!is_suspended(mddev, bio))
446 finish_wait(&mddev->sb_wait, &__wait);
448 atomic_inc(&mddev->active_io);
451 if (!mddev->pers->make_request(mddev, bio)) {
452 atomic_dec(&mddev->active_io);
453 wake_up(&mddev->sb_wait);
454 goto check_suspended;
457 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
458 wake_up(&mddev->sb_wait);
460 EXPORT_SYMBOL(md_handle_request);
462 static blk_qc_t md_submit_bio(struct bio *bio)
464 const int rw = bio_data_dir(bio);
465 const int sgrp = op_stat_group(bio_op(bio));
466 struct mddev *mddev = bio->bi_disk->private_data;
467 unsigned int sectors;
469 if (mddev == NULL || mddev->pers == NULL) {
471 return BLK_QC_T_NONE;
474 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
476 return BLK_QC_T_NONE;
479 blk_queue_split(&bio);
481 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
482 if (bio_sectors(bio) != 0)
483 bio->bi_status = BLK_STS_IOERR;
485 return BLK_QC_T_NONE;
489 * save the sectors now since our bio can
490 * go away inside make_request
492 sectors = bio_sectors(bio);
493 /* bio could be mergeable after passing to underlayer */
494 bio->bi_opf &= ~REQ_NOMERGE;
496 md_handle_request(mddev, bio);
499 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
500 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
503 return BLK_QC_T_NONE;
506 /* mddev_suspend makes sure no new requests are submitted
507 * to the device, and that any requests that have been submitted
508 * are completely handled.
509 * Once mddev_detach() is called and completes, the module will be
512 void mddev_suspend(struct mddev *mddev)
514 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
515 lockdep_assert_held(&mddev->reconfig_mutex);
516 if (mddev->suspended++)
519 wake_up(&mddev->sb_wait);
520 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
521 smp_mb__after_atomic();
522 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
523 mddev->pers->quiesce(mddev, 1);
524 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
525 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
527 del_timer_sync(&mddev->safemode_timer);
528 /* restrict memory reclaim I/O during raid array is suspend */
529 mddev->noio_flag = memalloc_noio_save();
531 EXPORT_SYMBOL_GPL(mddev_suspend);
533 void mddev_resume(struct mddev *mddev)
535 /* entred the memalloc scope from mddev_suspend() */
536 memalloc_noio_restore(mddev->noio_flag);
537 lockdep_assert_held(&mddev->reconfig_mutex);
538 if (--mddev->suspended)
540 wake_up(&mddev->sb_wait);
541 mddev->pers->quiesce(mddev, 0);
543 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
544 md_wakeup_thread(mddev->thread);
545 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
547 EXPORT_SYMBOL_GPL(mddev_resume);
550 * Generic flush handling for md
553 static void md_end_flush(struct bio *bio)
555 struct md_rdev *rdev = bio->bi_private;
556 struct mddev *mddev = rdev->mddev;
560 rdev_dec_pending(rdev, mddev);
562 if (atomic_dec_and_test(&mddev->flush_pending)) {
563 /* The pre-request flush has finished */
564 queue_work(md_wq, &mddev->flush_work);
568 static void md_submit_flush_data(struct work_struct *ws);
570 static void submit_flushes(struct work_struct *ws)
572 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
573 struct md_rdev *rdev;
575 mddev->start_flush = ktime_get_boottime();
576 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
577 atomic_set(&mddev->flush_pending, 1);
579 rdev_for_each_rcu(rdev, mddev)
580 if (rdev->raid_disk >= 0 &&
581 !test_bit(Faulty, &rdev->flags)) {
582 /* Take two references, one is dropped
583 * when request finishes, one after
584 * we reclaim rcu_read_lock
587 atomic_inc(&rdev->nr_pending);
588 atomic_inc(&rdev->nr_pending);
590 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
591 bi->bi_end_io = md_end_flush;
592 bi->bi_private = rdev;
593 bio_set_dev(bi, rdev->bdev);
594 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
595 atomic_inc(&mddev->flush_pending);
598 rdev_dec_pending(rdev, mddev);
601 if (atomic_dec_and_test(&mddev->flush_pending))
602 queue_work(md_wq, &mddev->flush_work);
605 static void md_submit_flush_data(struct work_struct *ws)
607 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
608 struct bio *bio = mddev->flush_bio;
611 * must reset flush_bio before calling into md_handle_request to avoid a
612 * deadlock, because other bios passed md_handle_request suspend check
613 * could wait for this and below md_handle_request could wait for those
614 * bios because of suspend check
616 spin_lock_irq(&mddev->lock);
617 mddev->last_flush = mddev->start_flush;
618 mddev->flush_bio = NULL;
619 spin_unlock_irq(&mddev->lock);
620 wake_up(&mddev->sb_wait);
622 if (bio->bi_iter.bi_size == 0) {
623 /* an empty barrier - all done */
626 bio->bi_opf &= ~REQ_PREFLUSH;
627 md_handle_request(mddev, bio);
632 * Manages consolidation of flushes and submitting any flushes needed for
633 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
634 * being finished in another context. Returns false if the flushing is
635 * complete but still needs the I/O portion of the bio to be processed.
637 bool md_flush_request(struct mddev *mddev, struct bio *bio)
639 ktime_t start = ktime_get_boottime();
640 spin_lock_irq(&mddev->lock);
641 wait_event_lock_irq(mddev->sb_wait,
643 ktime_after(mddev->last_flush, start),
645 if (!ktime_after(mddev->last_flush, start)) {
646 WARN_ON(mddev->flush_bio);
647 mddev->flush_bio = bio;
650 spin_unlock_irq(&mddev->lock);
653 INIT_WORK(&mddev->flush_work, submit_flushes);
654 queue_work(md_wq, &mddev->flush_work);
656 /* flush was performed for some other bio while we waited. */
657 if (bio->bi_iter.bi_size == 0)
658 /* an empty barrier - all done */
661 bio->bi_opf &= ~REQ_PREFLUSH;
667 EXPORT_SYMBOL(md_flush_request);
669 static inline struct mddev *mddev_get(struct mddev *mddev)
671 atomic_inc(&mddev->active);
675 static void mddev_delayed_delete(struct work_struct *ws);
677 static void mddev_put(struct mddev *mddev)
679 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
681 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
682 mddev->ctime == 0 && !mddev->hold_active) {
683 /* Array is not configured at all, and not held active,
685 list_del_init(&mddev->all_mddevs);
688 * Call queue_work inside the spinlock so that
689 * flush_workqueue() after mddev_find will succeed in waiting
690 * for the work to be done.
692 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
693 queue_work(md_misc_wq, &mddev->del_work);
695 spin_unlock(&all_mddevs_lock);
698 static void md_safemode_timeout(struct timer_list *t);
700 void mddev_init(struct mddev *mddev)
702 kobject_init(&mddev->kobj, &md_ktype);
703 mutex_init(&mddev->open_mutex);
704 mutex_init(&mddev->reconfig_mutex);
705 mutex_init(&mddev->bitmap_info.mutex);
706 INIT_LIST_HEAD(&mddev->disks);
707 INIT_LIST_HEAD(&mddev->all_mddevs);
708 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
709 atomic_set(&mddev->active, 1);
710 atomic_set(&mddev->openers, 0);
711 atomic_set(&mddev->active_io, 0);
712 spin_lock_init(&mddev->lock);
713 atomic_set(&mddev->flush_pending, 0);
714 init_waitqueue_head(&mddev->sb_wait);
715 init_waitqueue_head(&mddev->recovery_wait);
716 mddev->reshape_position = MaxSector;
717 mddev->reshape_backwards = 0;
718 mddev->last_sync_action = "none";
719 mddev->resync_min = 0;
720 mddev->resync_max = MaxSector;
721 mddev->level = LEVEL_NONE;
723 EXPORT_SYMBOL_GPL(mddev_init);
725 static struct mddev *mddev_find_locked(dev_t unit)
729 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
730 if (mddev->unit == unit)
736 static struct mddev *mddev_find(dev_t unit)
740 if (MAJOR(unit) != MD_MAJOR)
741 unit &= ~((1 << MdpMinorShift) - 1);
743 spin_lock(&all_mddevs_lock);
744 mddev = mddev_find_locked(unit);
747 spin_unlock(&all_mddevs_lock);
752 static struct mddev *mddev_find_or_alloc(dev_t unit)
754 struct mddev *mddev, *new = NULL;
756 if (unit && MAJOR(unit) != MD_MAJOR)
757 unit &= ~((1<<MdpMinorShift)-1);
760 spin_lock(&all_mddevs_lock);
763 mddev = mddev_find_locked(unit);
766 spin_unlock(&all_mddevs_lock);
772 list_add(&new->all_mddevs, &all_mddevs);
773 spin_unlock(&all_mddevs_lock);
774 new->hold_active = UNTIL_IOCTL;
778 /* find an unused unit number */
779 static int next_minor = 512;
780 int start = next_minor;
784 dev = MKDEV(MD_MAJOR, next_minor);
786 if (next_minor > MINORMASK)
788 if (next_minor == start) {
789 /* Oh dear, all in use. */
790 spin_unlock(&all_mddevs_lock);
795 is_free = !mddev_find_locked(dev);
798 new->md_minor = MINOR(dev);
799 new->hold_active = UNTIL_STOP;
800 list_add(&new->all_mddevs, &all_mddevs);
801 spin_unlock(&all_mddevs_lock);
804 spin_unlock(&all_mddevs_lock);
806 new = kzalloc(sizeof(*new), GFP_KERNEL);
811 if (MAJOR(unit) == MD_MAJOR)
812 new->md_minor = MINOR(unit);
814 new->md_minor = MINOR(unit) >> MdpMinorShift;
821 static struct attribute_group md_redundancy_group;
823 void mddev_unlock(struct mddev *mddev)
825 if (mddev->to_remove) {
826 /* These cannot be removed under reconfig_mutex as
827 * an access to the files will try to take reconfig_mutex
828 * while holding the file unremovable, which leads to
830 * So hold set sysfs_active while the remove in happeing,
831 * and anything else which might set ->to_remove or my
832 * otherwise change the sysfs namespace will fail with
833 * -EBUSY if sysfs_active is still set.
834 * We set sysfs_active under reconfig_mutex and elsewhere
835 * test it under the same mutex to ensure its correct value
838 struct attribute_group *to_remove = mddev->to_remove;
839 mddev->to_remove = NULL;
840 mddev->sysfs_active = 1;
841 mutex_unlock(&mddev->reconfig_mutex);
843 if (mddev->kobj.sd) {
844 if (to_remove != &md_redundancy_group)
845 sysfs_remove_group(&mddev->kobj, to_remove);
846 if (mddev->pers == NULL ||
847 mddev->pers->sync_request == NULL) {
848 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
849 if (mddev->sysfs_action)
850 sysfs_put(mddev->sysfs_action);
851 if (mddev->sysfs_completed)
852 sysfs_put(mddev->sysfs_completed);
853 if (mddev->sysfs_degraded)
854 sysfs_put(mddev->sysfs_degraded);
855 mddev->sysfs_action = NULL;
856 mddev->sysfs_completed = NULL;
857 mddev->sysfs_degraded = NULL;
860 mddev->sysfs_active = 0;
862 mutex_unlock(&mddev->reconfig_mutex);
864 /* As we've dropped the mutex we need a spinlock to
865 * make sure the thread doesn't disappear
867 spin_lock(&pers_lock);
868 md_wakeup_thread(mddev->thread);
869 wake_up(&mddev->sb_wait);
870 spin_unlock(&pers_lock);
872 EXPORT_SYMBOL_GPL(mddev_unlock);
874 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
876 struct md_rdev *rdev;
878 rdev_for_each_rcu(rdev, mddev)
879 if (rdev->desc_nr == nr)
884 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
886 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
888 struct md_rdev *rdev;
890 rdev_for_each(rdev, mddev)
891 if (rdev->bdev->bd_dev == dev)
897 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
899 struct md_rdev *rdev;
901 rdev_for_each_rcu(rdev, mddev)
902 if (rdev->bdev->bd_dev == dev)
907 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
909 static struct md_personality *find_pers(int level, char *clevel)
911 struct md_personality *pers;
912 list_for_each_entry(pers, &pers_list, list) {
913 if (level != LEVEL_NONE && pers->level == level)
915 if (strcmp(pers->name, clevel)==0)
921 /* return the offset of the super block in 512byte sectors */
922 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
924 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
925 return MD_NEW_SIZE_SECTORS(num_sectors);
928 static int alloc_disk_sb(struct md_rdev *rdev)
930 rdev->sb_page = alloc_page(GFP_KERNEL);
936 void md_rdev_clear(struct md_rdev *rdev)
939 put_page(rdev->sb_page);
941 rdev->sb_page = NULL;
946 put_page(rdev->bb_page);
947 rdev->bb_page = NULL;
949 badblocks_exit(&rdev->badblocks);
951 EXPORT_SYMBOL_GPL(md_rdev_clear);
953 static void super_written(struct bio *bio)
955 struct md_rdev *rdev = bio->bi_private;
956 struct mddev *mddev = rdev->mddev;
958 if (bio->bi_status) {
959 pr_err("md: %s gets error=%d\n", __func__,
960 blk_status_to_errno(bio->bi_status));
961 md_error(mddev, rdev);
962 if (!test_bit(Faulty, &rdev->flags)
963 && (bio->bi_opf & MD_FAILFAST)) {
964 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
965 set_bit(LastDev, &rdev->flags);
968 clear_bit(LastDev, &rdev->flags);
972 rdev_dec_pending(rdev, mddev);
974 if (atomic_dec_and_test(&mddev->pending_writes))
975 wake_up(&mddev->sb_wait);
978 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
979 sector_t sector, int size, struct page *page)
981 /* write first size bytes of page to sector of rdev
982 * Increment mddev->pending_writes before returning
983 * and decrement it on completion, waking up sb_wait
984 * if zero is reached.
985 * If an error occurred, call md_error
993 if (test_bit(Faulty, &rdev->flags))
996 bio = md_bio_alloc_sync(mddev);
998 atomic_inc(&rdev->nr_pending);
1000 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
1001 bio->bi_iter.bi_sector = sector;
1002 bio_add_page(bio, page, size, 0);
1003 bio->bi_private = rdev;
1004 bio->bi_end_io = super_written;
1006 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1007 test_bit(FailFast, &rdev->flags) &&
1008 !test_bit(LastDev, &rdev->flags))
1010 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
1012 atomic_inc(&mddev->pending_writes);
1016 int md_super_wait(struct mddev *mddev)
1018 /* wait for all superblock writes that were scheduled to complete */
1019 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1020 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1025 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1026 struct page *page, int op, int op_flags, bool metadata_op)
1028 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1031 if (metadata_op && rdev->meta_bdev)
1032 bio_set_dev(bio, rdev->meta_bdev);
1034 bio_set_dev(bio, rdev->bdev);
1035 bio_set_op_attrs(bio, op, op_flags);
1037 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1038 else if (rdev->mddev->reshape_position != MaxSector &&
1039 (rdev->mddev->reshape_backwards ==
1040 (sector >= rdev->mddev->reshape_position)))
1041 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
1043 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1044 bio_add_page(bio, page, size, 0);
1046 submit_bio_wait(bio);
1048 ret = !bio->bi_status;
1052 EXPORT_SYMBOL_GPL(sync_page_io);
1054 static int read_disk_sb(struct md_rdev *rdev, int size)
1056 char b[BDEVNAME_SIZE];
1058 if (rdev->sb_loaded)
1061 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1063 rdev->sb_loaded = 1;
1067 pr_err("md: disabled device %s, could not read superblock.\n",
1068 bdevname(rdev->bdev,b));
1072 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1074 return sb1->set_uuid0 == sb2->set_uuid0 &&
1075 sb1->set_uuid1 == sb2->set_uuid1 &&
1076 sb1->set_uuid2 == sb2->set_uuid2 &&
1077 sb1->set_uuid3 == sb2->set_uuid3;
1080 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1083 mdp_super_t *tmp1, *tmp2;
1085 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1086 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1088 if (!tmp1 || !tmp2) {
1097 * nr_disks is not constant
1102 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1109 static u32 md_csum_fold(u32 csum)
1111 csum = (csum & 0xffff) + (csum >> 16);
1112 return (csum & 0xffff) + (csum >> 16);
1115 static unsigned int calc_sb_csum(mdp_super_t *sb)
1118 u32 *sb32 = (u32*)sb;
1120 unsigned int disk_csum, csum;
1122 disk_csum = sb->sb_csum;
1125 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1127 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1130 /* This used to use csum_partial, which was wrong for several
1131 * reasons including that different results are returned on
1132 * different architectures. It isn't critical that we get exactly
1133 * the same return value as before (we always csum_fold before
1134 * testing, and that removes any differences). However as we
1135 * know that csum_partial always returned a 16bit value on
1136 * alphas, do a fold to maximise conformity to previous behaviour.
1138 sb->sb_csum = md_csum_fold(disk_csum);
1140 sb->sb_csum = disk_csum;
1146 * Handle superblock details.
1147 * We want to be able to handle multiple superblock formats
1148 * so we have a common interface to them all, and an array of
1149 * different handlers.
1150 * We rely on user-space to write the initial superblock, and support
1151 * reading and updating of superblocks.
1152 * Interface methods are:
1153 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1154 * loads and validates a superblock on dev.
1155 * if refdev != NULL, compare superblocks on both devices
1157 * 0 - dev has a superblock that is compatible with refdev
1158 * 1 - dev has a superblock that is compatible and newer than refdev
1159 * so dev should be used as the refdev in future
1160 * -EINVAL superblock incompatible or invalid
1161 * -othererror e.g. -EIO
1163 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1164 * Verify that dev is acceptable into mddev.
1165 * The first time, mddev->raid_disks will be 0, and data from
1166 * dev should be merged in. Subsequent calls check that dev
1167 * is new enough. Return 0 or -EINVAL
1169 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1170 * Update the superblock for rdev with data in mddev
1171 * This does not write to disc.
1177 struct module *owner;
1178 int (*load_super)(struct md_rdev *rdev,
1179 struct md_rdev *refdev,
1181 int (*validate_super)(struct mddev *mddev,
1182 struct md_rdev *freshest,
1183 struct md_rdev *rdev);
1184 void (*sync_super)(struct mddev *mddev,
1185 struct md_rdev *rdev);
1186 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1187 sector_t num_sectors);
1188 int (*allow_new_offset)(struct md_rdev *rdev,
1189 unsigned long long new_offset);
1193 * Check that the given mddev has no bitmap.
1195 * This function is called from the run method of all personalities that do not
1196 * support bitmaps. It prints an error message and returns non-zero if mddev
1197 * has a bitmap. Otherwise, it returns 0.
1200 int md_check_no_bitmap(struct mddev *mddev)
1202 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1204 pr_warn("%s: bitmaps are not supported for %s\n",
1205 mdname(mddev), mddev->pers->name);
1208 EXPORT_SYMBOL(md_check_no_bitmap);
1211 * load_super for 0.90.0
1213 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1215 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1218 bool spare_disk = true;
1221 * Calculate the position of the superblock (512byte sectors),
1222 * it's at the end of the disk.
1224 * It also happens to be a multiple of 4Kb.
1226 rdev->sb_start = calc_dev_sboffset(rdev);
1228 ret = read_disk_sb(rdev, MD_SB_BYTES);
1234 bdevname(rdev->bdev, b);
1235 sb = page_address(rdev->sb_page);
1237 if (sb->md_magic != MD_SB_MAGIC) {
1238 pr_warn("md: invalid raid superblock magic on %s\n", b);
1242 if (sb->major_version != 0 ||
1243 sb->minor_version < 90 ||
1244 sb->minor_version > 91) {
1245 pr_warn("Bad version number %d.%d on %s\n",
1246 sb->major_version, sb->minor_version, b);
1250 if (sb->raid_disks <= 0)
1253 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1254 pr_warn("md: invalid superblock checksum on %s\n", b);
1258 rdev->preferred_minor = sb->md_minor;
1259 rdev->data_offset = 0;
1260 rdev->new_data_offset = 0;
1261 rdev->sb_size = MD_SB_BYTES;
1262 rdev->badblocks.shift = -1;
1264 if (sb->level == LEVEL_MULTIPATH)
1267 rdev->desc_nr = sb->this_disk.number;
1269 /* not spare disk, or LEVEL_MULTIPATH */
1270 if (sb->level == LEVEL_MULTIPATH ||
1271 (rdev->desc_nr >= 0 &&
1272 rdev->desc_nr < MD_SB_DISKS &&
1273 sb->disks[rdev->desc_nr].state &
1274 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1284 mdp_super_t *refsb = page_address(refdev->sb_page);
1285 if (!md_uuid_equal(refsb, sb)) {
1286 pr_warn("md: %s has different UUID to %s\n",
1287 b, bdevname(refdev->bdev,b2));
1290 if (!md_sb_equal(refsb, sb)) {
1291 pr_warn("md: %s has same UUID but different superblock to %s\n",
1292 b, bdevname(refdev->bdev, b2));
1296 ev2 = md_event(refsb);
1298 if (!spare_disk && ev1 > ev2)
1303 rdev->sectors = rdev->sb_start;
1304 /* Limit to 4TB as metadata cannot record more than that.
1305 * (not needed for Linear and RAID0 as metadata doesn't
1308 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1309 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1311 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1312 /* "this cannot possibly happen" ... */
1320 * validate_super for 0.90.0
1321 * note: we are not using "freshest" for 0.9 superblock
1323 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1326 mdp_super_t *sb = page_address(rdev->sb_page);
1327 __u64 ev1 = md_event(sb);
1329 rdev->raid_disk = -1;
1330 clear_bit(Faulty, &rdev->flags);
1331 clear_bit(In_sync, &rdev->flags);
1332 clear_bit(Bitmap_sync, &rdev->flags);
1333 clear_bit(WriteMostly, &rdev->flags);
1335 if (mddev->raid_disks == 0) {
1336 mddev->major_version = 0;
1337 mddev->minor_version = sb->minor_version;
1338 mddev->patch_version = sb->patch_version;
1339 mddev->external = 0;
1340 mddev->chunk_sectors = sb->chunk_size >> 9;
1341 mddev->ctime = sb->ctime;
1342 mddev->utime = sb->utime;
1343 mddev->level = sb->level;
1344 mddev->clevel[0] = 0;
1345 mddev->layout = sb->layout;
1346 mddev->raid_disks = sb->raid_disks;
1347 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1348 mddev->events = ev1;
1349 mddev->bitmap_info.offset = 0;
1350 mddev->bitmap_info.space = 0;
1351 /* bitmap can use 60 K after the 4K superblocks */
1352 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1353 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1354 mddev->reshape_backwards = 0;
1356 if (mddev->minor_version >= 91) {
1357 mddev->reshape_position = sb->reshape_position;
1358 mddev->delta_disks = sb->delta_disks;
1359 mddev->new_level = sb->new_level;
1360 mddev->new_layout = sb->new_layout;
1361 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1362 if (mddev->delta_disks < 0)
1363 mddev->reshape_backwards = 1;
1365 mddev->reshape_position = MaxSector;
1366 mddev->delta_disks = 0;
1367 mddev->new_level = mddev->level;
1368 mddev->new_layout = mddev->layout;
1369 mddev->new_chunk_sectors = mddev->chunk_sectors;
1371 if (mddev->level == 0)
1374 if (sb->state & (1<<MD_SB_CLEAN))
1375 mddev->recovery_cp = MaxSector;
1377 if (sb->events_hi == sb->cp_events_hi &&
1378 sb->events_lo == sb->cp_events_lo) {
1379 mddev->recovery_cp = sb->recovery_cp;
1381 mddev->recovery_cp = 0;
1384 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1385 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1386 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1387 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1389 mddev->max_disks = MD_SB_DISKS;
1391 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1392 mddev->bitmap_info.file == NULL) {
1393 mddev->bitmap_info.offset =
1394 mddev->bitmap_info.default_offset;
1395 mddev->bitmap_info.space =
1396 mddev->bitmap_info.default_space;
1399 } else if (mddev->pers == NULL) {
1400 /* Insist on good event counter while assembling, except
1401 * for spares (which don't need an event count) */
1403 if (sb->disks[rdev->desc_nr].state & (
1404 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1405 if (ev1 < mddev->events)
1407 } else if (mddev->bitmap) {
1408 /* if adding to array with a bitmap, then we can accept an
1409 * older device ... but not too old.
1411 if (ev1 < mddev->bitmap->events_cleared)
1413 if (ev1 < mddev->events)
1414 set_bit(Bitmap_sync, &rdev->flags);
1416 if (ev1 < mddev->events)
1417 /* just a hot-add of a new device, leave raid_disk at -1 */
1421 if (mddev->level != LEVEL_MULTIPATH) {
1422 desc = sb->disks + rdev->desc_nr;
1424 if (desc->state & (1<<MD_DISK_FAULTY))
1425 set_bit(Faulty, &rdev->flags);
1426 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1427 desc->raid_disk < mddev->raid_disks */) {
1428 set_bit(In_sync, &rdev->flags);
1429 rdev->raid_disk = desc->raid_disk;
1430 rdev->saved_raid_disk = desc->raid_disk;
1431 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1432 /* active but not in sync implies recovery up to
1433 * reshape position. We don't know exactly where
1434 * that is, so set to zero for now */
1435 if (mddev->minor_version >= 91) {
1436 rdev->recovery_offset = 0;
1437 rdev->raid_disk = desc->raid_disk;
1440 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1441 set_bit(WriteMostly, &rdev->flags);
1442 if (desc->state & (1<<MD_DISK_FAILFAST))
1443 set_bit(FailFast, &rdev->flags);
1444 } else /* MULTIPATH are always insync */
1445 set_bit(In_sync, &rdev->flags);
1450 * sync_super for 0.90.0
1452 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1455 struct md_rdev *rdev2;
1456 int next_spare = mddev->raid_disks;
1458 /* make rdev->sb match mddev data..
1461 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1462 * 3/ any empty disks < next_spare become removed
1464 * disks[0] gets initialised to REMOVED because
1465 * we cannot be sure from other fields if it has
1466 * been initialised or not.
1469 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1471 rdev->sb_size = MD_SB_BYTES;
1473 sb = page_address(rdev->sb_page);
1475 memset(sb, 0, sizeof(*sb));
1477 sb->md_magic = MD_SB_MAGIC;
1478 sb->major_version = mddev->major_version;
1479 sb->patch_version = mddev->patch_version;
1480 sb->gvalid_words = 0; /* ignored */
1481 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1482 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1483 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1484 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1486 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1487 sb->level = mddev->level;
1488 sb->size = mddev->dev_sectors / 2;
1489 sb->raid_disks = mddev->raid_disks;
1490 sb->md_minor = mddev->md_minor;
1491 sb->not_persistent = 0;
1492 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1494 sb->events_hi = (mddev->events>>32);
1495 sb->events_lo = (u32)mddev->events;
1497 if (mddev->reshape_position == MaxSector)
1498 sb->minor_version = 90;
1500 sb->minor_version = 91;
1501 sb->reshape_position = mddev->reshape_position;
1502 sb->new_level = mddev->new_level;
1503 sb->delta_disks = mddev->delta_disks;
1504 sb->new_layout = mddev->new_layout;
1505 sb->new_chunk = mddev->new_chunk_sectors << 9;
1507 mddev->minor_version = sb->minor_version;
1510 sb->recovery_cp = mddev->recovery_cp;
1511 sb->cp_events_hi = (mddev->events>>32);
1512 sb->cp_events_lo = (u32)mddev->events;
1513 if (mddev->recovery_cp == MaxSector)
1514 sb->state = (1<< MD_SB_CLEAN);
1516 sb->recovery_cp = 0;
1518 sb->layout = mddev->layout;
1519 sb->chunk_size = mddev->chunk_sectors << 9;
1521 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1522 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1524 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1525 rdev_for_each(rdev2, mddev) {
1528 int is_active = test_bit(In_sync, &rdev2->flags);
1530 if (rdev2->raid_disk >= 0 &&
1531 sb->minor_version >= 91)
1532 /* we have nowhere to store the recovery_offset,
1533 * but if it is not below the reshape_position,
1534 * we can piggy-back on that.
1537 if (rdev2->raid_disk < 0 ||
1538 test_bit(Faulty, &rdev2->flags))
1541 desc_nr = rdev2->raid_disk;
1543 desc_nr = next_spare++;
1544 rdev2->desc_nr = desc_nr;
1545 d = &sb->disks[rdev2->desc_nr];
1547 d->number = rdev2->desc_nr;
1548 d->major = MAJOR(rdev2->bdev->bd_dev);
1549 d->minor = MINOR(rdev2->bdev->bd_dev);
1551 d->raid_disk = rdev2->raid_disk;
1553 d->raid_disk = rdev2->desc_nr; /* compatibility */
1554 if (test_bit(Faulty, &rdev2->flags))
1555 d->state = (1<<MD_DISK_FAULTY);
1556 else if (is_active) {
1557 d->state = (1<<MD_DISK_ACTIVE);
1558 if (test_bit(In_sync, &rdev2->flags))
1559 d->state |= (1<<MD_DISK_SYNC);
1567 if (test_bit(WriteMostly, &rdev2->flags))
1568 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1569 if (test_bit(FailFast, &rdev2->flags))
1570 d->state |= (1<<MD_DISK_FAILFAST);
1572 /* now set the "removed" and "faulty" bits on any missing devices */
1573 for (i=0 ; i < mddev->raid_disks ; i++) {
1574 mdp_disk_t *d = &sb->disks[i];
1575 if (d->state == 0 && d->number == 0) {
1578 d->state = (1<<MD_DISK_REMOVED);
1579 d->state |= (1<<MD_DISK_FAULTY);
1583 sb->nr_disks = nr_disks;
1584 sb->active_disks = active;
1585 sb->working_disks = working;
1586 sb->failed_disks = failed;
1587 sb->spare_disks = spare;
1589 sb->this_disk = sb->disks[rdev->desc_nr];
1590 sb->sb_csum = calc_sb_csum(sb);
1594 * rdev_size_change for 0.90.0
1596 static unsigned long long
1597 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1599 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1600 return 0; /* component must fit device */
1601 if (rdev->mddev->bitmap_info.offset)
1602 return 0; /* can't move bitmap */
1603 rdev->sb_start = calc_dev_sboffset(rdev);
1604 if (!num_sectors || num_sectors > rdev->sb_start)
1605 num_sectors = rdev->sb_start;
1606 /* Limit to 4TB as metadata cannot record more than that.
1607 * 4TB == 2^32 KB, or 2*2^32 sectors.
1609 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1610 num_sectors = (sector_t)(2ULL << 32) - 2;
1612 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1614 } while (md_super_wait(rdev->mddev) < 0);
1619 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1621 /* non-zero offset changes not possible with v0.90 */
1622 return new_offset == 0;
1626 * version 1 superblock
1629 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1633 unsigned long long newcsum;
1634 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1635 __le32 *isuper = (__le32*)sb;
1637 disk_csum = sb->sb_csum;
1640 for (; size >= 4; size -= 4)
1641 newcsum += le32_to_cpu(*isuper++);
1644 newcsum += le16_to_cpu(*(__le16*) isuper);
1646 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1647 sb->sb_csum = disk_csum;
1648 return cpu_to_le32(csum);
1651 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1653 struct mdp_superblock_1 *sb;
1657 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1659 bool spare_disk = true;
1662 * Calculate the position of the superblock in 512byte sectors.
1663 * It is always aligned to a 4K boundary and
1664 * depeding on minor_version, it can be:
1665 * 0: At least 8K, but less than 12K, from end of device
1666 * 1: At start of device
1667 * 2: 4K from start of device.
1669 switch(minor_version) {
1671 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1673 sb_start &= ~(sector_t)(4*2-1);
1684 rdev->sb_start = sb_start;
1686 /* superblock is rarely larger than 1K, but it can be larger,
1687 * and it is safe to read 4k, so we do that
1689 ret = read_disk_sb(rdev, 4096);
1690 if (ret) return ret;
1692 sb = page_address(rdev->sb_page);
1694 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1695 sb->major_version != cpu_to_le32(1) ||
1696 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1697 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1698 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1701 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1702 pr_warn("md: invalid superblock checksum on %s\n",
1703 bdevname(rdev->bdev,b));
1706 if (le64_to_cpu(sb->data_size) < 10) {
1707 pr_warn("md: data_size too small on %s\n",
1708 bdevname(rdev->bdev,b));
1713 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1714 /* Some padding is non-zero, might be a new feature */
1717 rdev->preferred_minor = 0xffff;
1718 rdev->data_offset = le64_to_cpu(sb->data_offset);
1719 rdev->new_data_offset = rdev->data_offset;
1720 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1721 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1722 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1723 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1725 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1726 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1727 if (rdev->sb_size & bmask)
1728 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1731 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1734 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1737 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1740 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1742 if (!rdev->bb_page) {
1743 rdev->bb_page = alloc_page(GFP_KERNEL);
1747 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1748 rdev->badblocks.count == 0) {
1749 /* need to load the bad block list.
1750 * Currently we limit it to one page.
1756 int sectors = le16_to_cpu(sb->bblog_size);
1757 if (sectors > (PAGE_SIZE / 512))
1759 offset = le32_to_cpu(sb->bblog_offset);
1762 bb_sector = (long long)offset;
1763 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1764 rdev->bb_page, REQ_OP_READ, 0, true))
1766 bbp = (__le64 *)page_address(rdev->bb_page);
1767 rdev->badblocks.shift = sb->bblog_shift;
1768 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1769 u64 bb = le64_to_cpu(*bbp);
1770 int count = bb & (0x3ff);
1771 u64 sector = bb >> 10;
1772 sector <<= sb->bblog_shift;
1773 count <<= sb->bblog_shift;
1776 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1779 } else if (sb->bblog_offset != 0)
1780 rdev->badblocks.shift = 0;
1782 if ((le32_to_cpu(sb->feature_map) &
1783 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1784 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1785 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1786 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1789 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1793 /* not spare disk, or LEVEL_MULTIPATH */
1794 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1795 (rdev->desc_nr >= 0 &&
1796 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1797 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1798 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1808 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1810 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1811 sb->level != refsb->level ||
1812 sb->layout != refsb->layout ||
1813 sb->chunksize != refsb->chunksize) {
1814 pr_warn("md: %s has strangely different superblock to %s\n",
1815 bdevname(rdev->bdev,b),
1816 bdevname(refdev->bdev,b2));
1819 ev1 = le64_to_cpu(sb->events);
1820 ev2 = le64_to_cpu(refsb->events);
1822 if (!spare_disk && ev1 > ev2)
1827 if (minor_version) {
1828 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1829 sectors -= rdev->data_offset;
1831 sectors = rdev->sb_start;
1832 if (sectors < le64_to_cpu(sb->data_size))
1834 rdev->sectors = le64_to_cpu(sb->data_size);
1838 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1840 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1841 __u64 ev1 = le64_to_cpu(sb->events);
1843 rdev->raid_disk = -1;
1844 clear_bit(Faulty, &rdev->flags);
1845 clear_bit(In_sync, &rdev->flags);
1846 clear_bit(Bitmap_sync, &rdev->flags);
1847 clear_bit(WriteMostly, &rdev->flags);
1849 if (mddev->raid_disks == 0) {
1850 mddev->major_version = 1;
1851 mddev->patch_version = 0;
1852 mddev->external = 0;
1853 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1854 mddev->ctime = le64_to_cpu(sb->ctime);
1855 mddev->utime = le64_to_cpu(sb->utime);
1856 mddev->level = le32_to_cpu(sb->level);
1857 mddev->clevel[0] = 0;
1858 mddev->layout = le32_to_cpu(sb->layout);
1859 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1860 mddev->dev_sectors = le64_to_cpu(sb->size);
1861 mddev->events = ev1;
1862 mddev->bitmap_info.offset = 0;
1863 mddev->bitmap_info.space = 0;
1864 /* Default location for bitmap is 1K after superblock
1865 * using 3K - total of 4K
1867 mddev->bitmap_info.default_offset = 1024 >> 9;
1868 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1869 mddev->reshape_backwards = 0;
1871 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1872 memcpy(mddev->uuid, sb->set_uuid, 16);
1874 mddev->max_disks = (4096-256)/2;
1876 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1877 mddev->bitmap_info.file == NULL) {
1878 mddev->bitmap_info.offset =
1879 (__s32)le32_to_cpu(sb->bitmap_offset);
1880 /* Metadata doesn't record how much space is available.
1881 * For 1.0, we assume we can use up to the superblock
1882 * if before, else to 4K beyond superblock.
1883 * For others, assume no change is possible.
1885 if (mddev->minor_version > 0)
1886 mddev->bitmap_info.space = 0;
1887 else if (mddev->bitmap_info.offset > 0)
1888 mddev->bitmap_info.space =
1889 8 - mddev->bitmap_info.offset;
1891 mddev->bitmap_info.space =
1892 -mddev->bitmap_info.offset;
1895 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1896 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1897 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1898 mddev->new_level = le32_to_cpu(sb->new_level);
1899 mddev->new_layout = le32_to_cpu(sb->new_layout);
1900 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1901 if (mddev->delta_disks < 0 ||
1902 (mddev->delta_disks == 0 &&
1903 (le32_to_cpu(sb->feature_map)
1904 & MD_FEATURE_RESHAPE_BACKWARDS)))
1905 mddev->reshape_backwards = 1;
1907 mddev->reshape_position = MaxSector;
1908 mddev->delta_disks = 0;
1909 mddev->new_level = mddev->level;
1910 mddev->new_layout = mddev->layout;
1911 mddev->new_chunk_sectors = mddev->chunk_sectors;
1914 if (mddev->level == 0 &&
1915 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1918 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1919 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1921 if (le32_to_cpu(sb->feature_map) &
1922 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1923 if (le32_to_cpu(sb->feature_map) &
1924 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1926 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1927 (le32_to_cpu(sb->feature_map) &
1928 MD_FEATURE_MULTIPLE_PPLS))
1930 set_bit(MD_HAS_PPL, &mddev->flags);
1932 } else if (mddev->pers == NULL) {
1933 /* Insist of good event counter while assembling, except for
1934 * spares (which don't need an event count).
1935 * Similar to mdadm, we allow event counter difference of 1
1936 * from the freshest device.
1938 if (rdev->desc_nr >= 0 &&
1939 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1940 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1941 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1942 if (ev1 + 1 < mddev->events)
1944 } else if (mddev->bitmap) {
1945 /* If adding to array with a bitmap, then we can accept an
1946 * older device, but not too old.
1948 if (ev1 < mddev->bitmap->events_cleared)
1950 if (ev1 < mddev->events)
1951 set_bit(Bitmap_sync, &rdev->flags);
1953 if (ev1 < mddev->events)
1954 /* just a hot-add of a new device, leave raid_disk at -1 */
1957 if (mddev->level != LEVEL_MULTIPATH) {
1959 if (rdev->desc_nr < 0 ||
1960 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1961 role = MD_DISK_ROLE_SPARE;
1963 } else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1965 * If we are assembling, and our event counter is smaller than the
1966 * highest event counter, we cannot trust our superblock about the role.
1967 * It could happen that our rdev was marked as Faulty, and all other
1968 * superblocks were updated with +1 event counter.
1969 * Then, before the next superblock update, which typically happens when
1970 * remove_and_add_spares() removes the device from the array, there was
1971 * a crash or reboot.
1972 * If we allow current rdev without consulting the freshest superblock,
1973 * we could cause data corruption.
1974 * Note that in this case our event counter is smaller by 1 than the
1975 * highest, otherwise, this rdev would not be allowed into array;
1976 * both kernel and mdadm allow event counter difference of 1.
1978 struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1979 u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1981 if (rdev->desc_nr >= freshest_max_dev) {
1982 /* this is unexpected, better not proceed */
1983 pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1984 mdname(mddev), rdev->bdev, rdev->desc_nr,
1985 freshest->bdev, freshest_max_dev);
1989 role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1990 pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1991 mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1993 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1996 case MD_DISK_ROLE_SPARE: /* spare */
1998 case MD_DISK_ROLE_FAULTY: /* faulty */
1999 set_bit(Faulty, &rdev->flags);
2001 case MD_DISK_ROLE_JOURNAL: /* journal device */
2002 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
2003 /* journal device without journal feature */
2004 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
2007 set_bit(Journal, &rdev->flags);
2008 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
2009 rdev->raid_disk = 0;
2012 rdev->saved_raid_disk = role;
2013 if ((le32_to_cpu(sb->feature_map) &
2014 MD_FEATURE_RECOVERY_OFFSET)) {
2015 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
2016 if (!(le32_to_cpu(sb->feature_map) &
2017 MD_FEATURE_RECOVERY_BITMAP))
2018 rdev->saved_raid_disk = -1;
2021 * If the array is FROZEN, then the device can't
2022 * be in_sync with rest of array.
2024 if (!test_bit(MD_RECOVERY_FROZEN,
2026 set_bit(In_sync, &rdev->flags);
2028 rdev->raid_disk = role;
2031 if (sb->devflags & WriteMostly1)
2032 set_bit(WriteMostly, &rdev->flags);
2033 if (sb->devflags & FailFast1)
2034 set_bit(FailFast, &rdev->flags);
2035 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2036 set_bit(Replacement, &rdev->flags);
2037 } else /* MULTIPATH are always insync */
2038 set_bit(In_sync, &rdev->flags);
2043 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2045 struct mdp_superblock_1 *sb;
2046 struct md_rdev *rdev2;
2048 /* make rdev->sb match mddev and rdev data. */
2050 sb = page_address(rdev->sb_page);
2052 sb->feature_map = 0;
2054 sb->recovery_offset = cpu_to_le64(0);
2055 memset(sb->pad3, 0, sizeof(sb->pad3));
2057 sb->utime = cpu_to_le64((__u64)mddev->utime);
2058 sb->events = cpu_to_le64(mddev->events);
2060 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2061 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2062 sb->resync_offset = cpu_to_le64(MaxSector);
2064 sb->resync_offset = cpu_to_le64(0);
2066 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2068 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2069 sb->size = cpu_to_le64(mddev->dev_sectors);
2070 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2071 sb->level = cpu_to_le32(mddev->level);
2072 sb->layout = cpu_to_le32(mddev->layout);
2073 if (test_bit(FailFast, &rdev->flags))
2074 sb->devflags |= FailFast1;
2076 sb->devflags &= ~FailFast1;
2078 if (test_bit(WriteMostly, &rdev->flags))
2079 sb->devflags |= WriteMostly1;
2081 sb->devflags &= ~WriteMostly1;
2082 sb->data_offset = cpu_to_le64(rdev->data_offset);
2083 sb->data_size = cpu_to_le64(rdev->sectors);
2085 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2086 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2087 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2090 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2091 !test_bit(In_sync, &rdev->flags)) {
2093 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2094 sb->recovery_offset =
2095 cpu_to_le64(rdev->recovery_offset);
2096 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2098 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2100 /* Note: recovery_offset and journal_tail share space */
2101 if (test_bit(Journal, &rdev->flags))
2102 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2103 if (test_bit(Replacement, &rdev->flags))
2105 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2107 if (mddev->reshape_position != MaxSector) {
2108 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2109 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2110 sb->new_layout = cpu_to_le32(mddev->new_layout);
2111 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2112 sb->new_level = cpu_to_le32(mddev->new_level);
2113 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2114 if (mddev->delta_disks == 0 &&
2115 mddev->reshape_backwards)
2117 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2118 if (rdev->new_data_offset != rdev->data_offset) {
2120 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2121 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2122 - rdev->data_offset));
2126 if (mddev_is_clustered(mddev))
2127 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2129 if (rdev->badblocks.count == 0)
2130 /* Nothing to do for bad blocks*/ ;
2131 else if (sb->bblog_offset == 0)
2132 /* Cannot record bad blocks on this device */
2133 md_error(mddev, rdev);
2135 struct badblocks *bb = &rdev->badblocks;
2136 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2138 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2143 seq = read_seqbegin(&bb->lock);
2145 memset(bbp, 0xff, PAGE_SIZE);
2147 for (i = 0 ; i < bb->count ; i++) {
2148 u64 internal_bb = p[i];
2149 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2150 | BB_LEN(internal_bb));
2151 bbp[i] = cpu_to_le64(store_bb);
2154 if (read_seqretry(&bb->lock, seq))
2157 bb->sector = (rdev->sb_start +
2158 (int)le32_to_cpu(sb->bblog_offset));
2159 bb->size = le16_to_cpu(sb->bblog_size);
2164 rdev_for_each(rdev2, mddev)
2165 if (rdev2->desc_nr+1 > max_dev)
2166 max_dev = rdev2->desc_nr+1;
2168 if (max_dev > le32_to_cpu(sb->max_dev)) {
2170 sb->max_dev = cpu_to_le32(max_dev);
2171 rdev->sb_size = max_dev * 2 + 256;
2172 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2173 if (rdev->sb_size & bmask)
2174 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2176 max_dev = le32_to_cpu(sb->max_dev);
2178 for (i=0; i<max_dev;i++)
2179 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2181 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2182 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2184 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2185 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2187 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2189 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2190 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2191 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2194 rdev_for_each(rdev2, mddev) {
2196 if (test_bit(Faulty, &rdev2->flags))
2197 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2198 else if (test_bit(In_sync, &rdev2->flags))
2199 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2200 else if (test_bit(Journal, &rdev2->flags))
2201 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2202 else if (rdev2->raid_disk >= 0)
2203 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2205 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2208 sb->sb_csum = calc_sb_1_csum(sb);
2211 static sector_t super_1_choose_bm_space(sector_t dev_size)
2215 /* if the device is bigger than 8Gig, save 64k for bitmap
2216 * usage, if bigger than 200Gig, save 128k
2218 if (dev_size < 64*2)
2220 else if (dev_size - 64*2 >= 200*1024*1024*2)
2222 else if (dev_size - 4*2 > 8*1024*1024*2)
2229 static unsigned long long
2230 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2232 struct mdp_superblock_1 *sb;
2233 sector_t max_sectors;
2234 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2235 return 0; /* component must fit device */
2236 if (rdev->data_offset != rdev->new_data_offset)
2237 return 0; /* too confusing */
2238 if (rdev->sb_start < rdev->data_offset) {
2239 /* minor versions 1 and 2; superblock before data */
2240 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2241 max_sectors -= rdev->data_offset;
2242 if (!num_sectors || num_sectors > max_sectors)
2243 num_sectors = max_sectors;
2244 } else if (rdev->mddev->bitmap_info.offset) {
2245 /* minor version 0 with bitmap we can't move */
2248 /* minor version 0; superblock after data */
2249 sector_t sb_start, bm_space;
2250 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2252 /* 8K is for superblock */
2253 sb_start = dev_size - 8*2;
2254 sb_start &= ~(sector_t)(4*2 - 1);
2256 bm_space = super_1_choose_bm_space(dev_size);
2258 /* Space that can be used to store date needs to decrease
2259 * superblock bitmap space and bad block space(4K)
2261 max_sectors = sb_start - bm_space - 4*2;
2263 if (!num_sectors || num_sectors > max_sectors)
2264 num_sectors = max_sectors;
2265 rdev->sb_start = sb_start;
2267 sb = page_address(rdev->sb_page);
2268 sb->data_size = cpu_to_le64(num_sectors);
2269 sb->super_offset = cpu_to_le64(rdev->sb_start);
2270 sb->sb_csum = calc_sb_1_csum(sb);
2272 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2274 } while (md_super_wait(rdev->mddev) < 0);
2280 super_1_allow_new_offset(struct md_rdev *rdev,
2281 unsigned long long new_offset)
2283 /* All necessary checks on new >= old have been done */
2284 struct bitmap *bitmap;
2285 if (new_offset >= rdev->data_offset)
2288 /* with 1.0 metadata, there is no metadata to tread on
2289 * so we can always move back */
2290 if (rdev->mddev->minor_version == 0)
2293 /* otherwise we must be sure not to step on
2294 * any metadata, so stay:
2295 * 36K beyond start of superblock
2296 * beyond end of badblocks
2297 * beyond write-intent bitmap
2299 if (rdev->sb_start + (32+4)*2 > new_offset)
2301 bitmap = rdev->mddev->bitmap;
2302 if (bitmap && !rdev->mddev->bitmap_info.file &&
2303 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2304 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2306 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2312 static struct super_type super_types[] = {
2315 .owner = THIS_MODULE,
2316 .load_super = super_90_load,
2317 .validate_super = super_90_validate,
2318 .sync_super = super_90_sync,
2319 .rdev_size_change = super_90_rdev_size_change,
2320 .allow_new_offset = super_90_allow_new_offset,
2324 .owner = THIS_MODULE,
2325 .load_super = super_1_load,
2326 .validate_super = super_1_validate,
2327 .sync_super = super_1_sync,
2328 .rdev_size_change = super_1_rdev_size_change,
2329 .allow_new_offset = super_1_allow_new_offset,
2333 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2335 if (mddev->sync_super) {
2336 mddev->sync_super(mddev, rdev);
2340 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2342 super_types[mddev->major_version].sync_super(mddev, rdev);
2345 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2347 struct md_rdev *rdev, *rdev2;
2350 rdev_for_each_rcu(rdev, mddev1) {
2351 if (test_bit(Faulty, &rdev->flags) ||
2352 test_bit(Journal, &rdev->flags) ||
2353 rdev->raid_disk == -1)
2355 rdev_for_each_rcu(rdev2, mddev2) {
2356 if (test_bit(Faulty, &rdev2->flags) ||
2357 test_bit(Journal, &rdev2->flags) ||
2358 rdev2->raid_disk == -1)
2360 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2370 static LIST_HEAD(pending_raid_disks);
2373 * Try to register data integrity profile for an mddev
2375 * This is called when an array is started and after a disk has been kicked
2376 * from the array. It only succeeds if all working and active component devices
2377 * are integrity capable with matching profiles.
2379 int md_integrity_register(struct mddev *mddev)
2381 struct md_rdev *rdev, *reference = NULL;
2383 if (list_empty(&mddev->disks))
2384 return 0; /* nothing to do */
2385 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2386 return 0; /* shouldn't register, or already is */
2387 rdev_for_each(rdev, mddev) {
2388 /* skip spares and non-functional disks */
2389 if (test_bit(Faulty, &rdev->flags))
2391 if (rdev->raid_disk < 0)
2394 /* Use the first rdev as the reference */
2398 /* does this rdev's profile match the reference profile? */
2399 if (blk_integrity_compare(reference->bdev->bd_disk,
2400 rdev->bdev->bd_disk) < 0)
2403 if (!reference || !bdev_get_integrity(reference->bdev))
2406 * All component devices are integrity capable and have matching
2407 * profiles, register the common profile for the md device.
2409 blk_integrity_register(mddev->gendisk,
2410 bdev_get_integrity(reference->bdev));
2412 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2413 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2414 pr_err("md: failed to create integrity pool for %s\n",
2420 EXPORT_SYMBOL(md_integrity_register);
2423 * Attempt to add an rdev, but only if it is consistent with the current
2426 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2428 struct blk_integrity *bi_mddev;
2429 char name[BDEVNAME_SIZE];
2431 if (!mddev->gendisk)
2434 bi_mddev = blk_get_integrity(mddev->gendisk);
2436 if (!bi_mddev) /* nothing to do */
2439 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2440 pr_err("%s: incompatible integrity profile for %s\n",
2441 mdname(mddev), bdevname(rdev->bdev, name));
2447 EXPORT_SYMBOL(md_integrity_add_rdev);
2449 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2451 char b[BDEVNAME_SIZE];
2455 /* prevent duplicates */
2456 if (find_rdev(mddev, rdev->bdev->bd_dev))
2459 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2463 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2464 if (!test_bit(Journal, &rdev->flags) &&
2466 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2468 /* Cannot change size, so fail
2469 * If mddev->level <= 0, then we don't care
2470 * about aligning sizes (e.g. linear)
2472 if (mddev->level > 0)
2475 mddev->dev_sectors = rdev->sectors;
2478 /* Verify rdev->desc_nr is unique.
2479 * If it is -1, assign a free number, else
2480 * check number is not in use
2483 if (rdev->desc_nr < 0) {
2486 choice = mddev->raid_disks;
2487 while (md_find_rdev_nr_rcu(mddev, choice))
2489 rdev->desc_nr = choice;
2491 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2497 if (!test_bit(Journal, &rdev->flags) &&
2498 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2499 pr_warn("md: %s: array is limited to %d devices\n",
2500 mdname(mddev), mddev->max_disks);
2503 bdevname(rdev->bdev,b);
2504 strreplace(b, '/', '!');
2506 rdev->mddev = mddev;
2507 pr_debug("md: bind<%s>\n", b);
2509 if (mddev->raid_disks)
2510 mddev_create_serial_pool(mddev, rdev, false);
2512 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2515 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2516 /* failure here is OK */
2517 err = sysfs_create_link(&rdev->kobj, ko, "block");
2518 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2519 rdev->sysfs_unack_badblocks =
2520 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2521 rdev->sysfs_badblocks =
2522 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2524 list_add_rcu(&rdev->same_set, &mddev->disks);
2525 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2527 /* May as well allow recovery to be retried once */
2528 mddev->recovery_disabled++;
2533 pr_warn("md: failed to register dev-%s for %s\n",
2538 static void rdev_delayed_delete(struct work_struct *ws)
2540 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2541 kobject_del(&rdev->kobj);
2542 kobject_put(&rdev->kobj);
2545 static void unbind_rdev_from_array(struct md_rdev *rdev)
2547 char b[BDEVNAME_SIZE];
2549 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2550 list_del_rcu(&rdev->same_set);
2551 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2552 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2554 sysfs_remove_link(&rdev->kobj, "block");
2555 sysfs_put(rdev->sysfs_state);
2556 sysfs_put(rdev->sysfs_unack_badblocks);
2557 sysfs_put(rdev->sysfs_badblocks);
2558 rdev->sysfs_state = NULL;
2559 rdev->sysfs_unack_badblocks = NULL;
2560 rdev->sysfs_badblocks = NULL;
2561 rdev->badblocks.count = 0;
2562 /* We need to delay this, otherwise we can deadlock when
2563 * writing to 'remove' to "dev/state". We also need
2564 * to delay it due to rcu usage.
2567 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2568 kobject_get(&rdev->kobj);
2569 queue_work(md_rdev_misc_wq, &rdev->del_work);
2573 * prevent the device from being mounted, repartitioned or
2574 * otherwise reused by a RAID array (or any other kernel
2575 * subsystem), by bd_claiming the device.
2577 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2580 struct block_device *bdev;
2582 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2583 shared ? (struct md_rdev *)lock_rdev : rdev);
2585 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2586 MAJOR(dev), MINOR(dev));
2587 return PTR_ERR(bdev);
2593 static void unlock_rdev(struct md_rdev *rdev)
2595 struct block_device *bdev = rdev->bdev;
2597 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2600 void md_autodetect_dev(dev_t dev);
2602 static void export_rdev(struct md_rdev *rdev)
2604 char b[BDEVNAME_SIZE];
2606 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2607 md_rdev_clear(rdev);
2609 if (test_bit(AutoDetected, &rdev->flags))
2610 md_autodetect_dev(rdev->bdev->bd_dev);
2613 kobject_put(&rdev->kobj);
2616 void md_kick_rdev_from_array(struct md_rdev *rdev)
2618 unbind_rdev_from_array(rdev);
2621 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2623 static void export_array(struct mddev *mddev)
2625 struct md_rdev *rdev;
2627 while (!list_empty(&mddev->disks)) {
2628 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2630 md_kick_rdev_from_array(rdev);
2632 mddev->raid_disks = 0;
2633 mddev->major_version = 0;
2636 static bool set_in_sync(struct mddev *mddev)
2638 lockdep_assert_held(&mddev->lock);
2639 if (!mddev->in_sync) {
2640 mddev->sync_checkers++;
2641 spin_unlock(&mddev->lock);
2642 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2643 spin_lock(&mddev->lock);
2644 if (!mddev->in_sync &&
2645 percpu_ref_is_zero(&mddev->writes_pending)) {
2648 * Ensure ->in_sync is visible before we clear
2652 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2653 sysfs_notify_dirent_safe(mddev->sysfs_state);
2655 if (--mddev->sync_checkers == 0)
2656 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2658 if (mddev->safemode == 1)
2659 mddev->safemode = 0;
2660 return mddev->in_sync;
2663 static void sync_sbs(struct mddev *mddev, int nospares)
2665 /* Update each superblock (in-memory image), but
2666 * if we are allowed to, skip spares which already
2667 * have the right event counter, or have one earlier
2668 * (which would mean they aren't being marked as dirty
2669 * with the rest of the array)
2671 struct md_rdev *rdev;
2672 rdev_for_each(rdev, mddev) {
2673 if (rdev->sb_events == mddev->events ||
2675 rdev->raid_disk < 0 &&
2676 rdev->sb_events+1 == mddev->events)) {
2677 /* Don't update this superblock */
2678 rdev->sb_loaded = 2;
2680 sync_super(mddev, rdev);
2681 rdev->sb_loaded = 1;
2686 static bool does_sb_need_changing(struct mddev *mddev)
2688 struct md_rdev *rdev = NULL, *iter;
2689 struct mdp_superblock_1 *sb;
2692 /* Find a good rdev */
2693 rdev_for_each(iter, mddev)
2694 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2699 /* No good device found. */
2703 sb = page_address(rdev->sb_page);
2704 /* Check if a device has become faulty or a spare become active */
2705 rdev_for_each(rdev, mddev) {
2706 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2707 /* Device activated? */
2708 if (role == 0xffff && rdev->raid_disk >=0 &&
2709 !test_bit(Faulty, &rdev->flags))
2711 /* Device turned faulty? */
2712 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2716 /* Check if any mddev parameters have changed */
2717 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2718 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2719 (mddev->layout != le32_to_cpu(sb->layout)) ||
2720 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2721 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2727 void md_update_sb(struct mddev *mddev, int force_change)
2729 struct md_rdev *rdev;
2732 int any_badblocks_changed = 0;
2737 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2742 if (mddev_is_clustered(mddev)) {
2743 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2745 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2747 ret = md_cluster_ops->metadata_update_start(mddev);
2748 /* Has someone else has updated the sb */
2749 if (!does_sb_need_changing(mddev)) {
2751 md_cluster_ops->metadata_update_cancel(mddev);
2752 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2753 BIT(MD_SB_CHANGE_DEVS) |
2754 BIT(MD_SB_CHANGE_CLEAN));
2760 * First make sure individual recovery_offsets are correct
2761 * curr_resync_completed can only be used during recovery.
2762 * During reshape/resync it might use array-addresses rather
2763 * that device addresses.
2765 rdev_for_each(rdev, mddev) {
2766 if (rdev->raid_disk >= 0 &&
2767 mddev->delta_disks >= 0 &&
2768 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2769 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2770 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2771 !test_bit(Journal, &rdev->flags) &&
2772 !test_bit(In_sync, &rdev->flags) &&
2773 mddev->curr_resync_completed > rdev->recovery_offset)
2774 rdev->recovery_offset = mddev->curr_resync_completed;
2777 if (!mddev->persistent) {
2778 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2779 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2780 if (!mddev->external) {
2781 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2782 rdev_for_each(rdev, mddev) {
2783 if (rdev->badblocks.changed) {
2784 rdev->badblocks.changed = 0;
2785 ack_all_badblocks(&rdev->badblocks);
2786 md_error(mddev, rdev);
2788 clear_bit(Blocked, &rdev->flags);
2789 clear_bit(BlockedBadBlocks, &rdev->flags);
2790 wake_up(&rdev->blocked_wait);
2793 wake_up(&mddev->sb_wait);
2797 spin_lock(&mddev->lock);
2799 mddev->utime = ktime_get_real_seconds();
2801 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2803 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2804 /* just a clean<-> dirty transition, possibly leave spares alone,
2805 * though if events isn't the right even/odd, we will have to do
2811 if (mddev->degraded)
2812 /* If the array is degraded, then skipping spares is both
2813 * dangerous and fairly pointless.
2814 * Dangerous because a device that was removed from the array
2815 * might have a event_count that still looks up-to-date,
2816 * so it can be re-added without a resync.
2817 * Pointless because if there are any spares to skip,
2818 * then a recovery will happen and soon that array won't
2819 * be degraded any more and the spare can go back to sleep then.
2823 sync_req = mddev->in_sync;
2825 /* If this is just a dirty<->clean transition, and the array is clean
2826 * and 'events' is odd, we can roll back to the previous clean state */
2828 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2829 && mddev->can_decrease_events
2830 && mddev->events != 1) {
2832 mddev->can_decrease_events = 0;
2834 /* otherwise we have to go forward and ... */
2836 mddev->can_decrease_events = nospares;
2840 * This 64-bit counter should never wrap.
2841 * Either we are in around ~1 trillion A.C., assuming
2842 * 1 reboot per second, or we have a bug...
2844 WARN_ON(mddev->events == 0);
2846 rdev_for_each(rdev, mddev) {
2847 if (rdev->badblocks.changed)
2848 any_badblocks_changed++;
2849 if (test_bit(Faulty, &rdev->flags))
2850 set_bit(FaultRecorded, &rdev->flags);
2853 sync_sbs(mddev, nospares);
2854 spin_unlock(&mddev->lock);
2856 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2857 mdname(mddev), mddev->in_sync);
2860 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2862 md_bitmap_update_sb(mddev->bitmap);
2863 rdev_for_each(rdev, mddev) {
2864 char b[BDEVNAME_SIZE];
2866 if (rdev->sb_loaded != 1)
2867 continue; /* no noise on spare devices */
2869 if (!test_bit(Faulty, &rdev->flags)) {
2870 md_super_write(mddev,rdev,
2871 rdev->sb_start, rdev->sb_size,
2873 pr_debug("md: (write) %s's sb offset: %llu\n",
2874 bdevname(rdev->bdev, b),
2875 (unsigned long long)rdev->sb_start);
2876 rdev->sb_events = mddev->events;
2877 if (rdev->badblocks.size) {
2878 md_super_write(mddev, rdev,
2879 rdev->badblocks.sector,
2880 rdev->badblocks.size << 9,
2882 rdev->badblocks.size = 0;
2886 pr_debug("md: %s (skipping faulty)\n",
2887 bdevname(rdev->bdev, b));
2889 if (mddev->level == LEVEL_MULTIPATH)
2890 /* only need to write one superblock... */
2893 if (md_super_wait(mddev) < 0)
2895 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2897 if (mddev_is_clustered(mddev) && ret == 0)
2898 md_cluster_ops->metadata_update_finish(mddev);
2900 if (mddev->in_sync != sync_req ||
2901 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2902 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2903 /* have to write it out again */
2905 wake_up(&mddev->sb_wait);
2906 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2907 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2909 rdev_for_each(rdev, mddev) {
2910 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2911 clear_bit(Blocked, &rdev->flags);
2913 if (any_badblocks_changed)
2914 ack_all_badblocks(&rdev->badblocks);
2915 clear_bit(BlockedBadBlocks, &rdev->flags);
2916 wake_up(&rdev->blocked_wait);
2919 EXPORT_SYMBOL(md_update_sb);
2921 static int add_bound_rdev(struct md_rdev *rdev)
2923 struct mddev *mddev = rdev->mddev;
2925 bool add_journal = test_bit(Journal, &rdev->flags);
2927 if (!mddev->pers->hot_remove_disk || add_journal) {
2928 /* If there is hot_add_disk but no hot_remove_disk
2929 * then added disks for geometry changes,
2930 * and should be added immediately.
2932 super_types[mddev->major_version].
2933 validate_super(mddev, NULL/*freshest*/, rdev);
2935 mddev_suspend(mddev);
2936 err = mddev->pers->hot_add_disk(mddev, rdev);
2938 mddev_resume(mddev);
2940 md_kick_rdev_from_array(rdev);
2944 sysfs_notify_dirent_safe(rdev->sysfs_state);
2946 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2947 if (mddev->degraded)
2948 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2949 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2950 md_new_event(mddev);
2951 md_wakeup_thread(mddev->thread);
2955 /* words written to sysfs files may, or may not, be \n terminated.
2956 * We want to accept with case. For this we use cmd_match.
2958 static int cmd_match(const char *cmd, const char *str)
2960 /* See if cmd, written into a sysfs file, matches
2961 * str. They must either be the same, or cmd can
2962 * have a trailing newline
2964 while (*cmd && *str && *cmd == *str) {
2975 struct rdev_sysfs_entry {
2976 struct attribute attr;
2977 ssize_t (*show)(struct md_rdev *, char *);
2978 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2982 state_show(struct md_rdev *rdev, char *page)
2986 unsigned long flags = READ_ONCE(rdev->flags);
2988 if (test_bit(Faulty, &flags) ||
2989 (!test_bit(ExternalBbl, &flags) &&
2990 rdev->badblocks.unacked_exist))
2991 len += sprintf(page+len, "faulty%s", sep);
2992 if (test_bit(In_sync, &flags))
2993 len += sprintf(page+len, "in_sync%s", sep);
2994 if (test_bit(Journal, &flags))
2995 len += sprintf(page+len, "journal%s", sep);
2996 if (test_bit(WriteMostly, &flags))
2997 len += sprintf(page+len, "write_mostly%s", sep);
2998 if (test_bit(Blocked, &flags) ||
2999 (rdev->badblocks.unacked_exist
3000 && !test_bit(Faulty, &flags)))
3001 len += sprintf(page+len, "blocked%s", sep);
3002 if (!test_bit(Faulty, &flags) &&
3003 !test_bit(Journal, &flags) &&
3004 !test_bit(In_sync, &flags))
3005 len += sprintf(page+len, "spare%s", sep);
3006 if (test_bit(WriteErrorSeen, &flags))
3007 len += sprintf(page+len, "write_error%s", sep);
3008 if (test_bit(WantReplacement, &flags))
3009 len += sprintf(page+len, "want_replacement%s", sep);
3010 if (test_bit(Replacement, &flags))
3011 len += sprintf(page+len, "replacement%s", sep);
3012 if (test_bit(ExternalBbl, &flags))
3013 len += sprintf(page+len, "external_bbl%s", sep);
3014 if (test_bit(FailFast, &flags))
3015 len += sprintf(page+len, "failfast%s", sep);
3020 return len+sprintf(page+len, "\n");
3024 state_store(struct md_rdev *rdev, const char *buf, size_t len)
3027 * faulty - simulates an error
3028 * remove - disconnects the device
3029 * writemostly - sets write_mostly
3030 * -writemostly - clears write_mostly
3031 * blocked - sets the Blocked flags
3032 * -blocked - clears the Blocked and possibly simulates an error
3033 * insync - sets Insync providing device isn't active
3034 * -insync - clear Insync for a device with a slot assigned,
3035 * so that it gets rebuilt based on bitmap
3036 * write_error - sets WriteErrorSeen
3037 * -write_error - clears WriteErrorSeen
3038 * {,-}failfast - set/clear FailFast
3041 struct mddev *mddev = rdev->mddev;
3043 bool need_update_sb = false;
3045 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
3046 md_error(rdev->mddev, rdev);
3047 if (test_bit(Faulty, &rdev->flags))
3051 } else if (cmd_match(buf, "remove")) {
3052 if (rdev->mddev->pers) {
3053 clear_bit(Blocked, &rdev->flags);
3054 remove_and_add_spares(rdev->mddev, rdev);
3056 if (rdev->raid_disk >= 0)
3060 if (mddev_is_clustered(mddev))
3061 err = md_cluster_ops->remove_disk(mddev, rdev);
3064 md_kick_rdev_from_array(rdev);
3066 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3067 md_wakeup_thread(mddev->thread);
3069 md_new_event(mddev);
3072 } else if (cmd_match(buf, "writemostly")) {
3073 set_bit(WriteMostly, &rdev->flags);
3074 mddev_create_serial_pool(rdev->mddev, rdev, false);
3075 need_update_sb = true;
3077 } else if (cmd_match(buf, "-writemostly")) {
3078 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3079 clear_bit(WriteMostly, &rdev->flags);
3080 need_update_sb = true;
3082 } else if (cmd_match(buf, "blocked")) {
3083 set_bit(Blocked, &rdev->flags);
3085 } else if (cmd_match(buf, "-blocked")) {
3086 if (!test_bit(Faulty, &rdev->flags) &&
3087 !test_bit(ExternalBbl, &rdev->flags) &&
3088 rdev->badblocks.unacked_exist) {
3089 /* metadata handler doesn't understand badblocks,
3090 * so we need to fail the device
3092 md_error(rdev->mddev, rdev);
3094 clear_bit(Blocked, &rdev->flags);
3095 clear_bit(BlockedBadBlocks, &rdev->flags);
3096 wake_up(&rdev->blocked_wait);
3097 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3098 md_wakeup_thread(rdev->mddev->thread);
3101 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3102 set_bit(In_sync, &rdev->flags);
3104 } else if (cmd_match(buf, "failfast")) {
3105 set_bit(FailFast, &rdev->flags);
3106 need_update_sb = true;
3108 } else if (cmd_match(buf, "-failfast")) {
3109 clear_bit(FailFast, &rdev->flags);
3110 need_update_sb = true;
3112 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3113 !test_bit(Journal, &rdev->flags)) {
3114 if (rdev->mddev->pers == NULL) {
3115 clear_bit(In_sync, &rdev->flags);
3116 rdev->saved_raid_disk = rdev->raid_disk;
3117 rdev->raid_disk = -1;
3120 } else if (cmd_match(buf, "write_error")) {
3121 set_bit(WriteErrorSeen, &rdev->flags);
3123 } else if (cmd_match(buf, "-write_error")) {
3124 clear_bit(WriteErrorSeen, &rdev->flags);
3126 } else if (cmd_match(buf, "want_replacement")) {
3127 /* Any non-spare device that is not a replacement can
3128 * become want_replacement at any time, but we then need to
3129 * check if recovery is needed.
3131 if (rdev->raid_disk >= 0 &&
3132 !test_bit(Journal, &rdev->flags) &&
3133 !test_bit(Replacement, &rdev->flags))
3134 set_bit(WantReplacement, &rdev->flags);
3135 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3136 md_wakeup_thread(rdev->mddev->thread);
3138 } else if (cmd_match(buf, "-want_replacement")) {
3139 /* Clearing 'want_replacement' is always allowed.
3140 * Once replacements starts it is too late though.
3143 clear_bit(WantReplacement, &rdev->flags);
3144 } else if (cmd_match(buf, "replacement")) {
3145 /* Can only set a device as a replacement when array has not
3146 * yet been started. Once running, replacement is automatic
3147 * from spares, or by assigning 'slot'.
3149 if (rdev->mddev->pers)
3152 set_bit(Replacement, &rdev->flags);
3155 } else if (cmd_match(buf, "-replacement")) {
3156 /* Similarly, can only clear Replacement before start */
3157 if (rdev->mddev->pers)
3160 clear_bit(Replacement, &rdev->flags);
3163 } else if (cmd_match(buf, "re-add")) {
3164 if (!rdev->mddev->pers)
3166 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3167 rdev->saved_raid_disk >= 0) {
3168 /* clear_bit is performed _after_ all the devices
3169 * have their local Faulty bit cleared. If any writes
3170 * happen in the meantime in the local node, they
3171 * will land in the local bitmap, which will be synced
3172 * by this node eventually
3174 if (!mddev_is_clustered(rdev->mddev) ||
3175 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3176 clear_bit(Faulty, &rdev->flags);
3177 err = add_bound_rdev(rdev);
3181 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3182 set_bit(ExternalBbl, &rdev->flags);
3183 rdev->badblocks.shift = 0;
3185 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3186 clear_bit(ExternalBbl, &rdev->flags);
3190 md_update_sb(mddev, 1);
3192 sysfs_notify_dirent_safe(rdev->sysfs_state);
3193 return err ? err : len;
3195 static struct rdev_sysfs_entry rdev_state =
3196 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3199 errors_show(struct md_rdev *rdev, char *page)
3201 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3205 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3210 rv = kstrtouint(buf, 10, &n);
3213 atomic_set(&rdev->corrected_errors, n);
3216 static struct rdev_sysfs_entry rdev_errors =
3217 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3220 slot_show(struct md_rdev *rdev, char *page)
3222 if (test_bit(Journal, &rdev->flags))
3223 return sprintf(page, "journal\n");
3224 else if (rdev->raid_disk < 0)
3225 return sprintf(page, "none\n");
3227 return sprintf(page, "%d\n", rdev->raid_disk);
3231 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3236 if (test_bit(Journal, &rdev->flags))
3238 if (strncmp(buf, "none", 4)==0)
3241 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3248 if (rdev->mddev->pers && slot == -1) {
3249 /* Setting 'slot' on an active array requires also
3250 * updating the 'rd%d' link, and communicating
3251 * with the personality with ->hot_*_disk.
3252 * For now we only support removing
3253 * failed/spare devices. This normally happens automatically,
3254 * but not when the metadata is externally managed.
3256 if (rdev->raid_disk == -1)
3258 /* personality does all needed checks */
3259 if (rdev->mddev->pers->hot_remove_disk == NULL)
3261 clear_bit(Blocked, &rdev->flags);
3262 remove_and_add_spares(rdev->mddev, rdev);
3263 if (rdev->raid_disk >= 0)
3265 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3266 md_wakeup_thread(rdev->mddev->thread);
3267 } else if (rdev->mddev->pers) {
3268 /* Activating a spare .. or possibly reactivating
3269 * if we ever get bitmaps working here.
3273 if (rdev->raid_disk != -1)
3276 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3279 if (rdev->mddev->pers->hot_add_disk == NULL)
3282 if (slot >= rdev->mddev->raid_disks &&
3283 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3286 rdev->raid_disk = slot;
3287 if (test_bit(In_sync, &rdev->flags))
3288 rdev->saved_raid_disk = slot;
3290 rdev->saved_raid_disk = -1;
3291 clear_bit(In_sync, &rdev->flags);
3292 clear_bit(Bitmap_sync, &rdev->flags);
3293 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3295 rdev->raid_disk = -1;
3298 sysfs_notify_dirent_safe(rdev->sysfs_state);
3299 /* failure here is OK */;
3300 sysfs_link_rdev(rdev->mddev, rdev);
3301 /* don't wakeup anyone, leave that to userspace. */
3303 if (slot >= rdev->mddev->raid_disks &&
3304 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3306 rdev->raid_disk = slot;
3307 /* assume it is working */
3308 clear_bit(Faulty, &rdev->flags);
3309 clear_bit(WriteMostly, &rdev->flags);
3310 set_bit(In_sync, &rdev->flags);
3311 sysfs_notify_dirent_safe(rdev->sysfs_state);
3316 static struct rdev_sysfs_entry rdev_slot =
3317 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3320 offset_show(struct md_rdev *rdev, char *page)
3322 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3326 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3328 unsigned long long offset;
3329 if (kstrtoull(buf, 10, &offset) < 0)
3331 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3333 if (rdev->sectors && rdev->mddev->external)
3334 /* Must set offset before size, so overlap checks
3337 rdev->data_offset = offset;
3338 rdev->new_data_offset = offset;
3342 static struct rdev_sysfs_entry rdev_offset =
3343 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3345 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3347 return sprintf(page, "%llu\n",
3348 (unsigned long long)rdev->new_data_offset);
3351 static ssize_t new_offset_store(struct md_rdev *rdev,
3352 const char *buf, size_t len)
3354 unsigned long long new_offset;
3355 struct mddev *mddev = rdev->mddev;
3357 if (kstrtoull(buf, 10, &new_offset) < 0)
3360 if (mddev->sync_thread ||
3361 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3363 if (new_offset == rdev->data_offset)
3364 /* reset is always permitted */
3366 else if (new_offset > rdev->data_offset) {
3367 /* must not push array size beyond rdev_sectors */
3368 if (new_offset - rdev->data_offset
3369 + mddev->dev_sectors > rdev->sectors)
3372 /* Metadata worries about other space details. */
3374 /* decreasing the offset is inconsistent with a backwards
3377 if (new_offset < rdev->data_offset &&
3378 mddev->reshape_backwards)
3380 /* Increasing offset is inconsistent with forwards
3381 * reshape. reshape_direction should be set to
3382 * 'backwards' first.
3384 if (new_offset > rdev->data_offset &&
3385 !mddev->reshape_backwards)
3388 if (mddev->pers && mddev->persistent &&
3389 !super_types[mddev->major_version]
3390 .allow_new_offset(rdev, new_offset))
3392 rdev->new_data_offset = new_offset;
3393 if (new_offset > rdev->data_offset)
3394 mddev->reshape_backwards = 1;
3395 else if (new_offset < rdev->data_offset)
3396 mddev->reshape_backwards = 0;
3400 static struct rdev_sysfs_entry rdev_new_offset =
3401 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3404 rdev_size_show(struct md_rdev *rdev, char *page)
3406 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3409 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3411 /* check if two start/length pairs overlap */
3419 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3421 unsigned long long blocks;
3424 if (kstrtoull(buf, 10, &blocks) < 0)
3427 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3428 return -EINVAL; /* sector conversion overflow */
3431 if (new != blocks * 2)
3432 return -EINVAL; /* unsigned long long to sector_t overflow */
3439 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3441 struct mddev *my_mddev = rdev->mddev;
3442 sector_t oldsectors = rdev->sectors;
3445 if (test_bit(Journal, &rdev->flags))
3447 if (strict_blocks_to_sectors(buf, §ors) < 0)
3449 if (rdev->data_offset != rdev->new_data_offset)
3450 return -EINVAL; /* too confusing */
3451 if (my_mddev->pers && rdev->raid_disk >= 0) {
3452 if (my_mddev->persistent) {
3453 sectors = super_types[my_mddev->major_version].
3454 rdev_size_change(rdev, sectors);
3457 } else if (!sectors)
3458 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3460 if (!my_mddev->pers->resize)
3461 /* Cannot change size for RAID0 or Linear etc */
3464 if (sectors < my_mddev->dev_sectors)
3465 return -EINVAL; /* component must fit device */
3467 rdev->sectors = sectors;
3468 if (sectors > oldsectors && my_mddev->external) {
3469 /* Need to check that all other rdevs with the same
3470 * ->bdev do not overlap. 'rcu' is sufficient to walk
3471 * the rdev lists safely.
3472 * This check does not provide a hard guarantee, it
3473 * just helps avoid dangerous mistakes.
3475 struct mddev *mddev;
3477 struct list_head *tmp;
3480 for_each_mddev(mddev, tmp) {
3481 struct md_rdev *rdev2;
3483 rdev_for_each(rdev2, mddev)
3484 if (rdev->bdev == rdev2->bdev &&
3486 overlaps(rdev->data_offset, rdev->sectors,
3499 /* Someone else could have slipped in a size
3500 * change here, but doing so is just silly.
3501 * We put oldsectors back because we *know* it is
3502 * safe, and trust userspace not to race with
3505 rdev->sectors = oldsectors;
3512 static struct rdev_sysfs_entry rdev_size =
3513 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3515 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3517 unsigned long long recovery_start = rdev->recovery_offset;
3519 if (test_bit(In_sync, &rdev->flags) ||
3520 recovery_start == MaxSector)
3521 return sprintf(page, "none\n");
3523 return sprintf(page, "%llu\n", recovery_start);
3526 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3528 unsigned long long recovery_start;
3530 if (cmd_match(buf, "none"))
3531 recovery_start = MaxSector;
3532 else if (kstrtoull(buf, 10, &recovery_start))
3535 if (rdev->mddev->pers &&
3536 rdev->raid_disk >= 0)
3539 rdev->recovery_offset = recovery_start;
3540 if (recovery_start == MaxSector)
3541 set_bit(In_sync, &rdev->flags);
3543 clear_bit(In_sync, &rdev->flags);
3547 static struct rdev_sysfs_entry rdev_recovery_start =
3548 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3550 /* sysfs access to bad-blocks list.
3551 * We present two files.
3552 * 'bad-blocks' lists sector numbers and lengths of ranges that
3553 * are recorded as bad. The list is truncated to fit within
3554 * the one-page limit of sysfs.
3555 * Writing "sector length" to this file adds an acknowledged
3557 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3558 * been acknowledged. Writing to this file adds bad blocks
3559 * without acknowledging them. This is largely for testing.
3561 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3563 return badblocks_show(&rdev->badblocks, page, 0);
3565 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3567 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3568 /* Maybe that ack was all we needed */
3569 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3570 wake_up(&rdev->blocked_wait);
3573 static struct rdev_sysfs_entry rdev_bad_blocks =
3574 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3576 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3578 return badblocks_show(&rdev->badblocks, page, 1);
3580 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3582 return badblocks_store(&rdev->badblocks, page, len, 1);
3584 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3585 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3588 ppl_sector_show(struct md_rdev *rdev, char *page)
3590 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3594 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3596 unsigned long long sector;
3598 if (kstrtoull(buf, 10, §or) < 0)
3600 if (sector != (sector_t)sector)
3603 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3604 rdev->raid_disk >= 0)
3607 if (rdev->mddev->persistent) {
3608 if (rdev->mddev->major_version == 0)
3610 if ((sector > rdev->sb_start &&
3611 sector - rdev->sb_start > S16_MAX) ||
3612 (sector < rdev->sb_start &&
3613 rdev->sb_start - sector > -S16_MIN))
3615 rdev->ppl.offset = sector - rdev->sb_start;
3616 } else if (!rdev->mddev->external) {
3619 rdev->ppl.sector = sector;
3623 static struct rdev_sysfs_entry rdev_ppl_sector =
3624 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3627 ppl_size_show(struct md_rdev *rdev, char *page)
3629 return sprintf(page, "%u\n", rdev->ppl.size);
3633 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3637 if (kstrtouint(buf, 10, &size) < 0)
3640 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3641 rdev->raid_disk >= 0)
3644 if (rdev->mddev->persistent) {
3645 if (rdev->mddev->major_version == 0)
3649 } else if (!rdev->mddev->external) {
3652 rdev->ppl.size = size;
3656 static struct rdev_sysfs_entry rdev_ppl_size =
3657 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3659 static struct attribute *rdev_default_attrs[] = {
3664 &rdev_new_offset.attr,
3666 &rdev_recovery_start.attr,
3667 &rdev_bad_blocks.attr,
3668 &rdev_unack_bad_blocks.attr,
3669 &rdev_ppl_sector.attr,
3670 &rdev_ppl_size.attr,
3674 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3676 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3677 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3683 return entry->show(rdev, page);
3687 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3688 const char *page, size_t length)
3690 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3691 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3693 struct mddev *mddev = rdev->mddev;
3697 if (!capable(CAP_SYS_ADMIN))
3699 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3701 if (rdev->mddev == NULL)
3704 rv = entry->store(rdev, page, length);
3705 mddev_unlock(mddev);
3710 static void rdev_free(struct kobject *ko)
3712 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3715 static const struct sysfs_ops rdev_sysfs_ops = {
3716 .show = rdev_attr_show,
3717 .store = rdev_attr_store,
3719 static struct kobj_type rdev_ktype = {
3720 .release = rdev_free,
3721 .sysfs_ops = &rdev_sysfs_ops,
3722 .default_attrs = rdev_default_attrs,
3725 int md_rdev_init(struct md_rdev *rdev)
3728 rdev->saved_raid_disk = -1;
3729 rdev->raid_disk = -1;
3731 rdev->data_offset = 0;
3732 rdev->new_data_offset = 0;
3733 rdev->sb_events = 0;
3734 rdev->last_read_error = 0;
3735 rdev->sb_loaded = 0;
3736 rdev->bb_page = NULL;
3737 atomic_set(&rdev->nr_pending, 0);
3738 atomic_set(&rdev->read_errors, 0);
3739 atomic_set(&rdev->corrected_errors, 0);
3741 INIT_LIST_HEAD(&rdev->same_set);
3742 init_waitqueue_head(&rdev->blocked_wait);
3744 /* Add space to store bad block list.
3745 * This reserves the space even on arrays where it cannot
3746 * be used - I wonder if that matters
3748 return badblocks_init(&rdev->badblocks, 0);
3750 EXPORT_SYMBOL_GPL(md_rdev_init);
3752 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3754 * mark the device faulty if:
3756 * - the device is nonexistent (zero size)
3757 * - the device has no valid superblock
3759 * a faulty rdev _never_ has rdev->sb set.
3761 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3763 char b[BDEVNAME_SIZE];
3765 struct md_rdev *rdev;
3768 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3770 return ERR_PTR(-ENOMEM);
3772 err = md_rdev_init(rdev);
3775 err = alloc_disk_sb(rdev);
3779 err = lock_rdev(rdev, newdev, super_format == -2);
3783 kobject_init(&rdev->kobj, &rdev_ktype);
3785 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3787 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3788 bdevname(rdev->bdev,b));
3793 if (super_format >= 0) {
3794 err = super_types[super_format].
3795 load_super(rdev, NULL, super_minor);
3796 if (err == -EINVAL) {
3797 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3798 bdevname(rdev->bdev,b),
3799 super_format, super_minor);
3803 pr_warn("md: could not read %s's sb, not importing!\n",
3804 bdevname(rdev->bdev,b));
3814 md_rdev_clear(rdev);
3816 return ERR_PTR(err);
3820 * Check a full RAID array for plausibility
3823 static int analyze_sbs(struct mddev *mddev)
3826 struct md_rdev *rdev, *freshest, *tmp;
3827 char b[BDEVNAME_SIZE];
3830 rdev_for_each_safe(rdev, tmp, mddev)
3831 switch (super_types[mddev->major_version].
3832 load_super(rdev, freshest, mddev->minor_version)) {
3839 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3840 bdevname(rdev->bdev,b));
3841 md_kick_rdev_from_array(rdev);
3844 /* Cannot find a valid fresh disk */
3846 pr_warn("md: cannot find a valid disk\n");
3850 super_types[mddev->major_version].
3851 validate_super(mddev, NULL/*freshest*/, freshest);
3854 rdev_for_each_safe(rdev, tmp, mddev) {
3855 if (mddev->max_disks &&
3856 (rdev->desc_nr >= mddev->max_disks ||
3857 i > mddev->max_disks)) {
3858 pr_warn("md: %s: %s: only %d devices permitted\n",
3859 mdname(mddev), bdevname(rdev->bdev, b),
3861 md_kick_rdev_from_array(rdev);
3864 if (rdev != freshest) {
3865 if (super_types[mddev->major_version].
3866 validate_super(mddev, freshest, rdev)) {
3867 pr_warn("md: kicking non-fresh %s from array!\n",
3868 bdevname(rdev->bdev,b));
3869 md_kick_rdev_from_array(rdev);
3873 if (mddev->level == LEVEL_MULTIPATH) {
3874 rdev->desc_nr = i++;
3875 rdev->raid_disk = rdev->desc_nr;
3876 set_bit(In_sync, &rdev->flags);
3877 } else if (rdev->raid_disk >=
3878 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3879 !test_bit(Journal, &rdev->flags)) {
3880 rdev->raid_disk = -1;
3881 clear_bit(In_sync, &rdev->flags);
3888 /* Read a fixed-point number.
3889 * Numbers in sysfs attributes should be in "standard" units where
3890 * possible, so time should be in seconds.
3891 * However we internally use a a much smaller unit such as
3892 * milliseconds or jiffies.
3893 * This function takes a decimal number with a possible fractional
3894 * component, and produces an integer which is the result of
3895 * multiplying that number by 10^'scale'.
3896 * all without any floating-point arithmetic.
3898 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3900 unsigned long result = 0;
3902 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3905 else if (decimals < scale) {
3908 result = result * 10 + value;
3920 *res = result * int_pow(10, scale - decimals);
3925 safe_delay_show(struct mddev *mddev, char *page)
3927 unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3929 return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3932 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3936 if (mddev_is_clustered(mddev)) {
3937 pr_warn("md: Safemode is disabled for clustered mode\n");
3941 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3944 mddev->safemode_delay = 0;
3946 unsigned long old_delay = mddev->safemode_delay;
3947 unsigned long new_delay = (msec*HZ)/1000;
3951 mddev->safemode_delay = new_delay;
3952 if (new_delay < old_delay || old_delay == 0)
3953 mod_timer(&mddev->safemode_timer, jiffies+1);
3957 static struct md_sysfs_entry md_safe_delay =
3958 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3961 level_show(struct mddev *mddev, char *page)
3963 struct md_personality *p;
3965 spin_lock(&mddev->lock);
3968 ret = sprintf(page, "%s\n", p->name);
3969 else if (mddev->clevel[0])
3970 ret = sprintf(page, "%s\n", mddev->clevel);
3971 else if (mddev->level != LEVEL_NONE)
3972 ret = sprintf(page, "%d\n", mddev->level);
3975 spin_unlock(&mddev->lock);
3980 level_store(struct mddev *mddev, const char *buf, size_t len)
3985 struct md_personality *pers, *oldpers;
3987 void *priv, *oldpriv;
3988 struct md_rdev *rdev;
3990 if (slen == 0 || slen >= sizeof(clevel))
3993 rv = mddev_lock(mddev);
3997 if (mddev->pers == NULL) {
3998 strncpy(mddev->clevel, buf, slen);
3999 if (mddev->clevel[slen-1] == '\n')
4001 mddev->clevel[slen] = 0;
4002 mddev->level = LEVEL_NONE;
4010 /* request to change the personality. Need to ensure:
4011 * - array is not engaged in resync/recovery/reshape
4012 * - old personality can be suspended
4013 * - new personality will access other array.
4017 if (mddev->sync_thread ||
4018 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4019 mddev->reshape_position != MaxSector ||
4020 mddev->sysfs_active)
4024 if (!mddev->pers->quiesce) {
4025 pr_warn("md: %s: %s does not support online personality change\n",
4026 mdname(mddev), mddev->pers->name);
4030 /* Now find the new personality */
4031 strncpy(clevel, buf, slen);
4032 if (clevel[slen-1] == '\n')
4035 if (kstrtol(clevel, 10, &level))
4038 if (request_module("md-%s", clevel) != 0)
4039 request_module("md-level-%s", clevel);
4040 spin_lock(&pers_lock);
4041 pers = find_pers(level, clevel);
4042 if (!pers || !try_module_get(pers->owner)) {
4043 spin_unlock(&pers_lock);
4044 pr_warn("md: personality %s not loaded\n", clevel);
4048 spin_unlock(&pers_lock);
4050 if (pers == mddev->pers) {
4051 /* Nothing to do! */
4052 module_put(pers->owner);
4056 if (!pers->takeover) {
4057 module_put(pers->owner);
4058 pr_warn("md: %s: %s does not support personality takeover\n",
4059 mdname(mddev), clevel);
4064 rdev_for_each(rdev, mddev)
4065 rdev->new_raid_disk = rdev->raid_disk;
4067 /* ->takeover must set new_* and/or delta_disks
4068 * if it succeeds, and may set them when it fails.
4070 priv = pers->takeover(mddev);
4072 mddev->new_level = mddev->level;
4073 mddev->new_layout = mddev->layout;
4074 mddev->new_chunk_sectors = mddev->chunk_sectors;
4075 mddev->raid_disks -= mddev->delta_disks;
4076 mddev->delta_disks = 0;
4077 mddev->reshape_backwards = 0;
4078 module_put(pers->owner);
4079 pr_warn("md: %s: %s would not accept array\n",
4080 mdname(mddev), clevel);
4085 /* Looks like we have a winner */
4086 mddev_suspend(mddev);
4087 mddev_detach(mddev);
4089 spin_lock(&mddev->lock);
4090 oldpers = mddev->pers;
4091 oldpriv = mddev->private;
4093 mddev->private = priv;
4094 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4095 mddev->level = mddev->new_level;
4096 mddev->layout = mddev->new_layout;
4097 mddev->chunk_sectors = mddev->new_chunk_sectors;
4098 mddev->delta_disks = 0;
4099 mddev->reshape_backwards = 0;
4100 mddev->degraded = 0;
4101 spin_unlock(&mddev->lock);
4103 if (oldpers->sync_request == NULL &&
4105 /* We are converting from a no-redundancy array
4106 * to a redundancy array and metadata is managed
4107 * externally so we need to be sure that writes
4108 * won't block due to a need to transition
4110 * until external management is started.
4113 mddev->safemode_delay = 0;
4114 mddev->safemode = 0;
4117 oldpers->free(mddev, oldpriv);
4119 if (oldpers->sync_request == NULL &&
4120 pers->sync_request != NULL) {
4121 /* need to add the md_redundancy_group */
4122 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4123 pr_warn("md: cannot register extra attributes for %s\n",
4125 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4126 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4127 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4129 if (oldpers->sync_request != NULL &&
4130 pers->sync_request == NULL) {
4131 /* need to remove the md_redundancy_group */
4132 if (mddev->to_remove == NULL)
4133 mddev->to_remove = &md_redundancy_group;
4136 module_put(oldpers->owner);
4138 rdev_for_each(rdev, mddev) {
4139 if (rdev->raid_disk < 0)
4141 if (rdev->new_raid_disk >= mddev->raid_disks)
4142 rdev->new_raid_disk = -1;
4143 if (rdev->new_raid_disk == rdev->raid_disk)
4145 sysfs_unlink_rdev(mddev, rdev);
4147 rdev_for_each(rdev, mddev) {
4148 if (rdev->raid_disk < 0)
4150 if (rdev->new_raid_disk == rdev->raid_disk)
4152 rdev->raid_disk = rdev->new_raid_disk;
4153 if (rdev->raid_disk < 0)
4154 clear_bit(In_sync, &rdev->flags);
4156 if (sysfs_link_rdev(mddev, rdev))
4157 pr_warn("md: cannot register rd%d for %s after level change\n",
4158 rdev->raid_disk, mdname(mddev));
4162 if (pers->sync_request == NULL) {
4163 /* this is now an array without redundancy, so
4164 * it must always be in_sync
4167 del_timer_sync(&mddev->safemode_timer);
4169 blk_set_stacking_limits(&mddev->queue->limits);
4171 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4172 mddev_resume(mddev);
4174 md_update_sb(mddev, 1);
4175 sysfs_notify_dirent_safe(mddev->sysfs_level);
4176 md_new_event(mddev);
4179 mddev_unlock(mddev);
4183 static struct md_sysfs_entry md_level =
4184 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4187 layout_show(struct mddev *mddev, char *page)
4189 /* just a number, not meaningful for all levels */
4190 if (mddev->reshape_position != MaxSector &&
4191 mddev->layout != mddev->new_layout)
4192 return sprintf(page, "%d (%d)\n",
4193 mddev->new_layout, mddev->layout);
4194 return sprintf(page, "%d\n", mddev->layout);
4198 layout_store(struct mddev *mddev, const char *buf, size_t len)
4203 err = kstrtouint(buf, 10, &n);
4206 err = mddev_lock(mddev);
4211 if (mddev->pers->check_reshape == NULL)
4216 mddev->new_layout = n;
4217 err = mddev->pers->check_reshape(mddev);
4219 mddev->new_layout = mddev->layout;
4222 mddev->new_layout = n;
4223 if (mddev->reshape_position == MaxSector)
4226 mddev_unlock(mddev);
4229 static struct md_sysfs_entry md_layout =
4230 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4233 raid_disks_show(struct mddev *mddev, char *page)
4235 if (mddev->raid_disks == 0)
4237 if (mddev->reshape_position != MaxSector &&
4238 mddev->delta_disks != 0)
4239 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4240 mddev->raid_disks - mddev->delta_disks);
4241 return sprintf(page, "%d\n", mddev->raid_disks);
4244 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4247 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4252 err = kstrtouint(buf, 10, &n);
4256 err = mddev_lock(mddev);
4260 err = update_raid_disks(mddev, n);
4261 else if (mddev->reshape_position != MaxSector) {
4262 struct md_rdev *rdev;
4263 int olddisks = mddev->raid_disks - mddev->delta_disks;
4266 rdev_for_each(rdev, mddev) {
4268 rdev->data_offset < rdev->new_data_offset)
4271 rdev->data_offset > rdev->new_data_offset)
4275 mddev->delta_disks = n - olddisks;
4276 mddev->raid_disks = n;
4277 mddev->reshape_backwards = (mddev->delta_disks < 0);
4279 mddev->raid_disks = n;
4281 mddev_unlock(mddev);
4282 return err ? err : len;
4284 static struct md_sysfs_entry md_raid_disks =
4285 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4288 uuid_show(struct mddev *mddev, char *page)
4290 return sprintf(page, "%pU\n", mddev->uuid);
4292 static struct md_sysfs_entry md_uuid =
4293 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4296 chunk_size_show(struct mddev *mddev, char *page)
4298 if (mddev->reshape_position != MaxSector &&
4299 mddev->chunk_sectors != mddev->new_chunk_sectors)
4300 return sprintf(page, "%d (%d)\n",
4301 mddev->new_chunk_sectors << 9,
4302 mddev->chunk_sectors << 9);
4303 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4307 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4312 err = kstrtoul(buf, 10, &n);
4316 err = mddev_lock(mddev);
4320 if (mddev->pers->check_reshape == NULL)
4325 mddev->new_chunk_sectors = n >> 9;
4326 err = mddev->pers->check_reshape(mddev);
4328 mddev->new_chunk_sectors = mddev->chunk_sectors;
4331 mddev->new_chunk_sectors = n >> 9;
4332 if (mddev->reshape_position == MaxSector)
4333 mddev->chunk_sectors = n >> 9;
4335 mddev_unlock(mddev);
4338 static struct md_sysfs_entry md_chunk_size =
4339 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4342 resync_start_show(struct mddev *mddev, char *page)
4344 if (mddev->recovery_cp == MaxSector)
4345 return sprintf(page, "none\n");
4346 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4350 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4352 unsigned long long n;
4355 if (cmd_match(buf, "none"))
4358 err = kstrtoull(buf, 10, &n);
4361 if (n != (sector_t)n)
4365 err = mddev_lock(mddev);
4368 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4372 mddev->recovery_cp = n;
4374 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4376 mddev_unlock(mddev);
4379 static struct md_sysfs_entry md_resync_start =
4380 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4381 resync_start_show, resync_start_store);
4384 * The array state can be:
4387 * No devices, no size, no level
4388 * Equivalent to STOP_ARRAY ioctl
4390 * May have some settings, but array is not active
4391 * all IO results in error
4392 * When written, doesn't tear down array, but just stops it
4393 * suspended (not supported yet)
4394 * All IO requests will block. The array can be reconfigured.
4395 * Writing this, if accepted, will block until array is quiescent
4397 * no resync can happen. no superblocks get written.
4398 * write requests fail
4400 * like readonly, but behaves like 'clean' on a write request.
4402 * clean - no pending writes, but otherwise active.
4403 * When written to inactive array, starts without resync
4404 * If a write request arrives then
4405 * if metadata is known, mark 'dirty' and switch to 'active'.
4406 * if not known, block and switch to write-pending
4407 * If written to an active array that has pending writes, then fails.
4409 * fully active: IO and resync can be happening.
4410 * When written to inactive array, starts with resync
4413 * clean, but writes are blocked waiting for 'active' to be written.
4416 * like active, but no writes have been seen for a while (100msec).
4419 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4420 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4421 * when a member is gone, so this state will at least alert the
4422 * user that something is wrong.
4424 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4425 write_pending, active_idle, broken, bad_word};
4426 static char *array_states[] = {
4427 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4428 "write-pending", "active-idle", "broken", NULL };
4430 static int match_word(const char *word, char **list)
4433 for (n=0; list[n]; n++)
4434 if (cmd_match(word, list[n]))
4440 array_state_show(struct mddev *mddev, char *page)
4442 enum array_state st = inactive;
4444 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4453 spin_lock(&mddev->lock);
4454 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4456 else if (mddev->in_sync)
4458 else if (mddev->safemode)
4462 spin_unlock(&mddev->lock);
4465 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4468 if (list_empty(&mddev->disks) &&
4469 mddev->raid_disks == 0 &&
4470 mddev->dev_sectors == 0)
4475 return sprintf(page, "%s\n", array_states[st]);
4478 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4479 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4480 static int restart_array(struct mddev *mddev);
4483 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4486 enum array_state st = match_word(buf, array_states);
4488 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4489 /* don't take reconfig_mutex when toggling between
4492 spin_lock(&mddev->lock);
4494 restart_array(mddev);
4495 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4496 md_wakeup_thread(mddev->thread);
4497 wake_up(&mddev->sb_wait);
4498 } else /* st == clean */ {
4499 restart_array(mddev);
4500 if (!set_in_sync(mddev))
4504 sysfs_notify_dirent_safe(mddev->sysfs_state);
4505 spin_unlock(&mddev->lock);
4508 err = mddev_lock(mddev);
4516 /* stopping an active array */
4517 err = do_md_stop(mddev, 0, NULL);
4520 /* stopping an active array */
4522 err = do_md_stop(mddev, 2, NULL);
4524 err = 0; /* already inactive */
4527 break; /* not supported yet */
4530 err = md_set_readonly(mddev, NULL);
4533 set_disk_ro(mddev->gendisk, 1);
4534 err = do_md_run(mddev);
4540 err = md_set_readonly(mddev, NULL);
4541 else if (mddev->ro == 1)
4542 err = restart_array(mddev);
4545 set_disk_ro(mddev->gendisk, 0);
4549 err = do_md_run(mddev);
4554 err = restart_array(mddev);
4557 spin_lock(&mddev->lock);
4558 if (!set_in_sync(mddev))
4560 spin_unlock(&mddev->lock);
4566 err = restart_array(mddev);
4569 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4570 wake_up(&mddev->sb_wait);
4574 set_disk_ro(mddev->gendisk, 0);
4575 err = do_md_run(mddev);
4581 /* these cannot be set */
4586 if (mddev->hold_active == UNTIL_IOCTL)
4587 mddev->hold_active = 0;
4588 sysfs_notify_dirent_safe(mddev->sysfs_state);
4590 mddev_unlock(mddev);
4593 static struct md_sysfs_entry md_array_state =
4594 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4597 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4598 return sprintf(page, "%d\n",
4599 atomic_read(&mddev->max_corr_read_errors));
4603 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4608 rv = kstrtouint(buf, 10, &n);
4613 atomic_set(&mddev->max_corr_read_errors, n);
4617 static struct md_sysfs_entry max_corr_read_errors =
4618 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4619 max_corrected_read_errors_store);
4622 null_show(struct mddev *mddev, char *page)
4627 /* need to ensure rdev_delayed_delete() has completed */
4628 static void flush_rdev_wq(struct mddev *mddev)
4630 struct md_rdev *rdev;
4633 rdev_for_each_rcu(rdev, mddev)
4634 if (work_pending(&rdev->del_work)) {
4635 flush_workqueue(md_rdev_misc_wq);
4642 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4644 /* buf must be %d:%d\n? giving major and minor numbers */
4645 /* The new device is added to the array.
4646 * If the array has a persistent superblock, we read the
4647 * superblock to initialise info and check validity.
4648 * Otherwise, only checking done is that in bind_rdev_to_array,
4649 * which mainly checks size.
4652 int major = simple_strtoul(buf, &e, 10);
4655 struct md_rdev *rdev;
4658 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4660 minor = simple_strtoul(e+1, &e, 10);
4661 if (*e && *e != '\n')
4663 dev = MKDEV(major, minor);
4664 if (major != MAJOR(dev) ||
4665 minor != MINOR(dev))
4668 flush_rdev_wq(mddev);
4669 err = mddev_lock(mddev);
4672 if (mddev->persistent) {
4673 rdev = md_import_device(dev, mddev->major_version,
4674 mddev->minor_version);
4675 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4676 struct md_rdev *rdev0
4677 = list_entry(mddev->disks.next,
4678 struct md_rdev, same_set);
4679 err = super_types[mddev->major_version]
4680 .load_super(rdev, rdev0, mddev->minor_version);
4684 } else if (mddev->external)
4685 rdev = md_import_device(dev, -2, -1);
4687 rdev = md_import_device(dev, -1, -1);
4690 mddev_unlock(mddev);
4691 return PTR_ERR(rdev);
4693 err = bind_rdev_to_array(rdev, mddev);
4697 mddev_unlock(mddev);
4699 md_new_event(mddev);
4700 return err ? err : len;
4703 static struct md_sysfs_entry md_new_device =
4704 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4707 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4710 unsigned long chunk, end_chunk;
4713 err = mddev_lock(mddev);
4718 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4720 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4721 if (buf == end) break;
4722 if (*end == '-') { /* range */
4724 end_chunk = simple_strtoul(buf, &end, 0);
4725 if (buf == end) break;
4727 if (*end && !isspace(*end)) break;
4728 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4729 buf = skip_spaces(end);
4731 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4733 mddev_unlock(mddev);
4737 static struct md_sysfs_entry md_bitmap =
4738 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4741 size_show(struct mddev *mddev, char *page)
4743 return sprintf(page, "%llu\n",
4744 (unsigned long long)mddev->dev_sectors / 2);
4747 static int update_size(struct mddev *mddev, sector_t num_sectors);
4750 size_store(struct mddev *mddev, const char *buf, size_t len)
4752 /* If array is inactive, we can reduce the component size, but
4753 * not increase it (except from 0).
4754 * If array is active, we can try an on-line resize
4757 int err = strict_blocks_to_sectors(buf, §ors);
4761 err = mddev_lock(mddev);
4765 err = update_size(mddev, sectors);
4767 md_update_sb(mddev, 1);
4769 if (mddev->dev_sectors == 0 ||
4770 mddev->dev_sectors > sectors)
4771 mddev->dev_sectors = sectors;
4775 mddev_unlock(mddev);
4776 return err ? err : len;
4779 static struct md_sysfs_entry md_size =
4780 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4782 /* Metadata version.
4784 * 'none' for arrays with no metadata (good luck...)
4785 * 'external' for arrays with externally managed metadata,
4786 * or N.M for internally known formats
4789 metadata_show(struct mddev *mddev, char *page)
4791 if (mddev->persistent)
4792 return sprintf(page, "%d.%d\n",
4793 mddev->major_version, mddev->minor_version);
4794 else if (mddev->external)
4795 return sprintf(page, "external:%s\n", mddev->metadata_type);
4797 return sprintf(page, "none\n");
4801 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4806 /* Changing the details of 'external' metadata is
4807 * always permitted. Otherwise there must be
4808 * no devices attached to the array.
4811 err = mddev_lock(mddev);
4815 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4817 else if (!list_empty(&mddev->disks))
4821 if (cmd_match(buf, "none")) {
4822 mddev->persistent = 0;
4823 mddev->external = 0;
4824 mddev->major_version = 0;
4825 mddev->minor_version = 90;
4828 if (strncmp(buf, "external:", 9) == 0) {
4829 size_t namelen = len-9;
4830 if (namelen >= sizeof(mddev->metadata_type))
4831 namelen = sizeof(mddev->metadata_type)-1;
4832 strncpy(mddev->metadata_type, buf+9, namelen);
4833 mddev->metadata_type[namelen] = 0;
4834 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4835 mddev->metadata_type[--namelen] = 0;
4836 mddev->persistent = 0;
4837 mddev->external = 1;
4838 mddev->major_version = 0;
4839 mddev->minor_version = 90;
4842 major = simple_strtoul(buf, &e, 10);
4844 if (e==buf || *e != '.')
4847 minor = simple_strtoul(buf, &e, 10);
4848 if (e==buf || (*e && *e != '\n') )
4851 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4853 mddev->major_version = major;
4854 mddev->minor_version = minor;
4855 mddev->persistent = 1;
4856 mddev->external = 0;
4859 mddev_unlock(mddev);
4863 static struct md_sysfs_entry md_metadata =
4864 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4867 action_show(struct mddev *mddev, char *page)
4869 char *type = "idle";
4870 unsigned long recovery = mddev->recovery;
4871 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4873 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4874 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4875 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4877 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4878 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4880 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4884 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4886 else if (mddev->reshape_position != MaxSector)
4889 return sprintf(page, "%s\n", type);
4893 action_store(struct mddev *mddev, const char *page, size_t len)
4895 if (!mddev->pers || !mddev->pers->sync_request)
4899 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4900 if (cmd_match(page, "frozen"))
4901 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4903 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4904 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4905 mddev_lock(mddev) == 0) {
4906 if (work_pending(&mddev->del_work))
4907 flush_workqueue(md_misc_wq);
4908 if (mddev->sync_thread) {
4909 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4910 md_reap_sync_thread(mddev);
4912 mddev_unlock(mddev);
4914 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4916 else if (cmd_match(page, "resync"))
4917 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4918 else if (cmd_match(page, "recover")) {
4919 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4920 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4921 } else if (cmd_match(page, "reshape")) {
4923 if (mddev->pers->start_reshape == NULL)
4925 err = mddev_lock(mddev);
4927 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4929 } else if (mddev->reshape_position == MaxSector ||
4930 mddev->pers->check_reshape == NULL ||
4931 mddev->pers->check_reshape(mddev)) {
4932 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4933 err = mddev->pers->start_reshape(mddev);
4936 * If reshape is still in progress, and
4937 * md_check_recovery() can continue to reshape,
4938 * don't restart reshape because data can be
4939 * corrupted for raid456.
4941 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4943 mddev_unlock(mddev);
4947 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4949 if (cmd_match(page, "check"))
4950 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4951 else if (!cmd_match(page, "repair"))
4953 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4954 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4955 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4957 if (mddev->ro == 2) {
4958 /* A write to sync_action is enough to justify
4959 * canceling read-auto mode
4962 md_wakeup_thread(mddev->sync_thread);
4964 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4965 md_wakeup_thread(mddev->thread);
4966 sysfs_notify_dirent_safe(mddev->sysfs_action);
4970 static struct md_sysfs_entry md_scan_mode =
4971 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4974 last_sync_action_show(struct mddev *mddev, char *page)
4976 return sprintf(page, "%s\n", mddev->last_sync_action);
4979 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4982 mismatch_cnt_show(struct mddev *mddev, char *page)
4984 return sprintf(page, "%llu\n",
4985 (unsigned long long)
4986 atomic64_read(&mddev->resync_mismatches));
4989 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4992 sync_min_show(struct mddev *mddev, char *page)
4994 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4995 mddev->sync_speed_min ? "local": "system");
4999 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
5004 if (strncmp(buf, "system", 6)==0) {
5007 rv = kstrtouint(buf, 10, &min);
5013 mddev->sync_speed_min = min;
5017 static struct md_sysfs_entry md_sync_min =
5018 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
5021 sync_max_show(struct mddev *mddev, char *page)
5023 return sprintf(page, "%d (%s)\n", speed_max(mddev),
5024 mddev->sync_speed_max ? "local": "system");
5028 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
5033 if (strncmp(buf, "system", 6)==0) {
5036 rv = kstrtouint(buf, 10, &max);
5042 mddev->sync_speed_max = max;
5046 static struct md_sysfs_entry md_sync_max =
5047 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5050 degraded_show(struct mddev *mddev, char *page)
5052 return sprintf(page, "%d\n", mddev->degraded);
5054 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5057 sync_force_parallel_show(struct mddev *mddev, char *page)
5059 return sprintf(page, "%d\n", mddev->parallel_resync);
5063 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5067 if (kstrtol(buf, 10, &n))
5070 if (n != 0 && n != 1)
5073 mddev->parallel_resync = n;
5075 if (mddev->sync_thread)
5076 wake_up(&resync_wait);
5081 /* force parallel resync, even with shared block devices */
5082 static struct md_sysfs_entry md_sync_force_parallel =
5083 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5084 sync_force_parallel_show, sync_force_parallel_store);
5087 sync_speed_show(struct mddev *mddev, char *page)
5089 unsigned long resync, dt, db;
5090 if (mddev->curr_resync == 0)
5091 return sprintf(page, "none\n");
5092 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5093 dt = (jiffies - mddev->resync_mark) / HZ;
5095 db = resync - mddev->resync_mark_cnt;
5096 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5099 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5102 sync_completed_show(struct mddev *mddev, char *page)
5104 unsigned long long max_sectors, resync;
5106 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5107 return sprintf(page, "none\n");
5109 if (mddev->curr_resync == 1 ||
5110 mddev->curr_resync == 2)
5111 return sprintf(page, "delayed\n");
5113 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5114 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5115 max_sectors = mddev->resync_max_sectors;
5117 max_sectors = mddev->dev_sectors;
5119 resync = mddev->curr_resync_completed;
5120 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5123 static struct md_sysfs_entry md_sync_completed =
5124 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5127 min_sync_show(struct mddev *mddev, char *page)
5129 return sprintf(page, "%llu\n",
5130 (unsigned long long)mddev->resync_min);
5133 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5135 unsigned long long min;
5138 if (kstrtoull(buf, 10, &min))
5141 spin_lock(&mddev->lock);
5143 if (min > mddev->resync_max)
5147 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5150 /* Round down to multiple of 4K for safety */
5151 mddev->resync_min = round_down(min, 8);
5155 spin_unlock(&mddev->lock);
5159 static struct md_sysfs_entry md_min_sync =
5160 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5163 max_sync_show(struct mddev *mddev, char *page)
5165 if (mddev->resync_max == MaxSector)
5166 return sprintf(page, "max\n");
5168 return sprintf(page, "%llu\n",
5169 (unsigned long long)mddev->resync_max);
5172 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5175 spin_lock(&mddev->lock);
5176 if (strncmp(buf, "max", 3) == 0)
5177 mddev->resync_max = MaxSector;
5179 unsigned long long max;
5183 if (kstrtoull(buf, 10, &max))
5185 if (max < mddev->resync_min)
5189 if (max < mddev->resync_max &&
5191 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5194 /* Must be a multiple of chunk_size */
5195 chunk = mddev->chunk_sectors;
5197 sector_t temp = max;
5200 if (sector_div(temp, chunk))
5203 mddev->resync_max = max;
5205 wake_up(&mddev->recovery_wait);
5208 spin_unlock(&mddev->lock);
5212 static struct md_sysfs_entry md_max_sync =
5213 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5216 suspend_lo_show(struct mddev *mddev, char *page)
5218 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5222 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5224 unsigned long long new;
5227 err = kstrtoull(buf, 10, &new);
5230 if (new != (sector_t)new)
5233 err = mddev_lock(mddev);
5237 if (mddev->pers == NULL ||
5238 mddev->pers->quiesce == NULL)
5240 mddev_suspend(mddev);
5241 mddev->suspend_lo = new;
5242 mddev_resume(mddev);
5246 mddev_unlock(mddev);
5249 static struct md_sysfs_entry md_suspend_lo =
5250 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5253 suspend_hi_show(struct mddev *mddev, char *page)
5255 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5259 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5261 unsigned long long new;
5264 err = kstrtoull(buf, 10, &new);
5267 if (new != (sector_t)new)
5270 err = mddev_lock(mddev);
5274 if (mddev->pers == NULL)
5277 mddev_suspend(mddev);
5278 mddev->suspend_hi = new;
5279 mddev_resume(mddev);
5283 mddev_unlock(mddev);
5286 static struct md_sysfs_entry md_suspend_hi =
5287 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5290 reshape_position_show(struct mddev *mddev, char *page)
5292 if (mddev->reshape_position != MaxSector)
5293 return sprintf(page, "%llu\n",
5294 (unsigned long long)mddev->reshape_position);
5295 strcpy(page, "none\n");
5300 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5302 struct md_rdev *rdev;
5303 unsigned long long new;
5306 err = kstrtoull(buf, 10, &new);
5309 if (new != (sector_t)new)
5311 err = mddev_lock(mddev);
5317 mddev->reshape_position = new;
5318 mddev->delta_disks = 0;
5319 mddev->reshape_backwards = 0;
5320 mddev->new_level = mddev->level;
5321 mddev->new_layout = mddev->layout;
5322 mddev->new_chunk_sectors = mddev->chunk_sectors;
5323 rdev_for_each(rdev, mddev)
5324 rdev->new_data_offset = rdev->data_offset;
5327 mddev_unlock(mddev);
5331 static struct md_sysfs_entry md_reshape_position =
5332 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5333 reshape_position_store);
5336 reshape_direction_show(struct mddev *mddev, char *page)
5338 return sprintf(page, "%s\n",
5339 mddev->reshape_backwards ? "backwards" : "forwards");
5343 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5348 if (cmd_match(buf, "forwards"))
5350 else if (cmd_match(buf, "backwards"))
5354 if (mddev->reshape_backwards == backwards)
5357 err = mddev_lock(mddev);
5360 /* check if we are allowed to change */
5361 if (mddev->delta_disks)
5363 else if (mddev->persistent &&
5364 mddev->major_version == 0)
5367 mddev->reshape_backwards = backwards;
5368 mddev_unlock(mddev);
5372 static struct md_sysfs_entry md_reshape_direction =
5373 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5374 reshape_direction_store);
5377 array_size_show(struct mddev *mddev, char *page)
5379 if (mddev->external_size)
5380 return sprintf(page, "%llu\n",
5381 (unsigned long long)mddev->array_sectors/2);
5383 return sprintf(page, "default\n");
5387 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5392 err = mddev_lock(mddev);
5396 /* cluster raid doesn't support change array_sectors */
5397 if (mddev_is_clustered(mddev)) {
5398 mddev_unlock(mddev);
5402 if (strncmp(buf, "default", 7) == 0) {
5404 sectors = mddev->pers->size(mddev, 0, 0);
5406 sectors = mddev->array_sectors;
5408 mddev->external_size = 0;
5410 if (strict_blocks_to_sectors(buf, §ors) < 0)
5412 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5415 mddev->external_size = 1;
5419 mddev->array_sectors = sectors;
5421 set_capacity(mddev->gendisk, mddev->array_sectors);
5422 revalidate_disk_size(mddev->gendisk, true);
5425 mddev_unlock(mddev);
5429 static struct md_sysfs_entry md_array_size =
5430 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5434 consistency_policy_show(struct mddev *mddev, char *page)
5438 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5439 ret = sprintf(page, "journal\n");
5440 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5441 ret = sprintf(page, "ppl\n");
5442 } else if (mddev->bitmap) {
5443 ret = sprintf(page, "bitmap\n");
5444 } else if (mddev->pers) {
5445 if (mddev->pers->sync_request)
5446 ret = sprintf(page, "resync\n");
5448 ret = sprintf(page, "none\n");
5450 ret = sprintf(page, "unknown\n");
5457 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5462 if (mddev->pers->change_consistency_policy)
5463 err = mddev->pers->change_consistency_policy(mddev, buf);
5466 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5467 set_bit(MD_HAS_PPL, &mddev->flags);
5472 return err ? err : len;
5475 static struct md_sysfs_entry md_consistency_policy =
5476 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5477 consistency_policy_store);
5479 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5481 return sprintf(page, "%d\n", mddev->fail_last_dev);
5485 * Setting fail_last_dev to true to allow last device to be forcibly removed
5486 * from RAID1/RAID10.
5489 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5494 ret = kstrtobool(buf, &value);
5498 if (value != mddev->fail_last_dev)
5499 mddev->fail_last_dev = value;
5503 static struct md_sysfs_entry md_fail_last_dev =
5504 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5505 fail_last_dev_store);
5507 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5509 if (mddev->pers == NULL || (mddev->pers->level != 1))
5510 return sprintf(page, "n/a\n");
5512 return sprintf(page, "%d\n", mddev->serialize_policy);
5516 * Setting serialize_policy to true to enforce write IO is not reordered
5520 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5525 err = kstrtobool(buf, &value);
5529 if (value == mddev->serialize_policy)
5532 err = mddev_lock(mddev);
5535 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5536 pr_err("md: serialize_policy is only effective for raid1\n");
5541 mddev_suspend(mddev);
5543 mddev_create_serial_pool(mddev, NULL, true);
5545 mddev_destroy_serial_pool(mddev, NULL, true);
5546 mddev->serialize_policy = value;
5547 mddev_resume(mddev);
5549 mddev_unlock(mddev);
5553 static struct md_sysfs_entry md_serialize_policy =
5554 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5555 serialize_policy_store);
5558 static struct attribute *md_default_attrs[] = {
5561 &md_raid_disks.attr,
5563 &md_chunk_size.attr,
5565 &md_resync_start.attr,
5567 &md_new_device.attr,
5568 &md_safe_delay.attr,
5569 &md_array_state.attr,
5570 &md_reshape_position.attr,
5571 &md_reshape_direction.attr,
5572 &md_array_size.attr,
5573 &max_corr_read_errors.attr,
5574 &md_consistency_policy.attr,
5575 &md_fail_last_dev.attr,
5576 &md_serialize_policy.attr,
5580 static struct attribute *md_redundancy_attrs[] = {
5582 &md_last_scan_mode.attr,
5583 &md_mismatches.attr,
5586 &md_sync_speed.attr,
5587 &md_sync_force_parallel.attr,
5588 &md_sync_completed.attr,
5591 &md_suspend_lo.attr,
5592 &md_suspend_hi.attr,
5597 static struct attribute_group md_redundancy_group = {
5599 .attrs = md_redundancy_attrs,
5603 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5605 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5606 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5611 spin_lock(&all_mddevs_lock);
5612 if (list_empty(&mddev->all_mddevs)) {
5613 spin_unlock(&all_mddevs_lock);
5617 spin_unlock(&all_mddevs_lock);
5619 rv = entry->show(mddev, page);
5625 md_attr_store(struct kobject *kobj, struct attribute *attr,
5626 const char *page, size_t length)
5628 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5629 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5634 if (!capable(CAP_SYS_ADMIN))
5636 spin_lock(&all_mddevs_lock);
5637 if (list_empty(&mddev->all_mddevs)) {
5638 spin_unlock(&all_mddevs_lock);
5642 spin_unlock(&all_mddevs_lock);
5643 rv = entry->store(mddev, page, length);
5648 static void md_free(struct kobject *ko)
5650 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5652 if (mddev->sysfs_state)
5653 sysfs_put(mddev->sysfs_state);
5654 if (mddev->sysfs_level)
5655 sysfs_put(mddev->sysfs_level);
5658 del_gendisk(mddev->gendisk);
5660 blk_cleanup_queue(mddev->queue);
5662 put_disk(mddev->gendisk);
5663 percpu_ref_exit(&mddev->writes_pending);
5665 bioset_exit(&mddev->bio_set);
5666 bioset_exit(&mddev->sync_set);
5670 static const struct sysfs_ops md_sysfs_ops = {
5671 .show = md_attr_show,
5672 .store = md_attr_store,
5674 static struct kobj_type md_ktype = {
5676 .sysfs_ops = &md_sysfs_ops,
5677 .default_attrs = md_default_attrs,
5682 static void mddev_delayed_delete(struct work_struct *ws)
5684 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5686 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5687 kobject_del(&mddev->kobj);
5688 kobject_put(&mddev->kobj);
5691 static void no_op(struct percpu_ref *r) {}
5693 int mddev_init_writes_pending(struct mddev *mddev)
5695 if (mddev->writes_pending.percpu_count_ptr)
5697 if (percpu_ref_init(&mddev->writes_pending, no_op,
5698 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5700 /* We want to start with the refcount at zero */
5701 percpu_ref_put(&mddev->writes_pending);
5704 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5706 static int md_alloc(dev_t dev, char *name)
5709 * If dev is zero, name is the name of a device to allocate with
5710 * an arbitrary minor number. It will be "md_???"
5711 * If dev is non-zero it must be a device number with a MAJOR of
5712 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5713 * the device is being created by opening a node in /dev.
5714 * If "name" is not NULL, the device is being created by
5715 * writing to /sys/module/md_mod/parameters/new_array.
5717 static DEFINE_MUTEX(disks_mutex);
5718 struct mddev *mddev = mddev_find_or_alloc(dev);
5719 struct gendisk *disk;
5728 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5729 shift = partitioned ? MdpMinorShift : 0;
5730 unit = MINOR(mddev->unit) >> shift;
5732 /* wait for any previous instance of this device to be
5733 * completely removed (mddev_delayed_delete).
5735 flush_workqueue(md_misc_wq);
5736 flush_workqueue(md_rdev_misc_wq);
5738 mutex_lock(&disks_mutex);
5744 /* Need to ensure that 'name' is not a duplicate.
5746 struct mddev *mddev2;
5747 spin_lock(&all_mddevs_lock);
5749 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5750 if (mddev2->gendisk &&
5751 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5752 spin_unlock(&all_mddevs_lock);
5755 spin_unlock(&all_mddevs_lock);
5759 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5761 mddev->hold_active = UNTIL_STOP;
5764 mddev->queue = blk_alloc_queue(NUMA_NO_NODE);
5768 blk_set_stacking_limits(&mddev->queue->limits);
5770 disk = alloc_disk(1 << shift);
5772 blk_cleanup_queue(mddev->queue);
5773 mddev->queue = NULL;
5776 disk->major = MAJOR(mddev->unit);
5777 disk->first_minor = unit << shift;
5779 strcpy(disk->disk_name, name);
5780 else if (partitioned)
5781 sprintf(disk->disk_name, "md_d%d", unit);
5783 sprintf(disk->disk_name, "md%d", unit);
5784 disk->fops = &md_fops;
5785 disk->private_data = mddev;
5786 disk->queue = mddev->queue;
5787 blk_queue_write_cache(mddev->queue, true, true);
5788 /* Allow extended partitions. This makes the
5789 * 'mdp' device redundant, but we can't really
5792 disk->flags |= GENHD_FL_EXT_DEVT;
5793 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5794 mddev->gendisk = disk;
5797 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5799 /* This isn't possible, but as kobject_init_and_add is marked
5800 * __must_check, we must do something with the result
5802 pr_debug("md: cannot register %s/md - name in use\n",
5806 if (mddev->kobj.sd &&
5807 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5808 pr_debug("pointless warning\n");
5810 mutex_unlock(&disks_mutex);
5811 if (!error && mddev->kobj.sd) {
5812 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5813 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5814 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5820 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5823 md_alloc(dev, NULL);
5827 static int add_named_array(const char *val, const struct kernel_param *kp)
5830 * val must be "md_*" or "mdNNN".
5831 * For "md_*" we allocate an array with a large free minor number, and
5832 * set the name to val. val must not already be an active name.
5833 * For "mdNNN" we allocate an array with the minor number NNN
5834 * which must not already be in use.
5836 int len = strlen(val);
5837 char buf[DISK_NAME_LEN];
5838 unsigned long devnum;
5840 while (len && val[len-1] == '\n')
5842 if (len >= DISK_NAME_LEN)
5844 strlcpy(buf, val, len+1);
5845 if (strncmp(buf, "md_", 3) == 0)
5846 return md_alloc(0, buf);
5847 if (strncmp(buf, "md", 2) == 0 &&
5849 kstrtoul(buf+2, 10, &devnum) == 0 &&
5850 devnum <= MINORMASK)
5851 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5856 static void md_safemode_timeout(struct timer_list *t)
5858 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5860 mddev->safemode = 1;
5861 if (mddev->external)
5862 sysfs_notify_dirent_safe(mddev->sysfs_state);
5864 md_wakeup_thread(mddev->thread);
5867 static int start_dirty_degraded;
5869 int md_run(struct mddev *mddev)
5872 struct md_rdev *rdev;
5873 struct md_personality *pers;
5875 if (list_empty(&mddev->disks))
5876 /* cannot run an array with no devices.. */
5881 /* Cannot run until previous stop completes properly */
5882 if (mddev->sysfs_active)
5886 * Analyze all RAID superblock(s)
5888 if (!mddev->raid_disks) {
5889 if (!mddev->persistent)
5891 err = analyze_sbs(mddev);
5896 if (mddev->level != LEVEL_NONE)
5897 request_module("md-level-%d", mddev->level);
5898 else if (mddev->clevel[0])
5899 request_module("md-%s", mddev->clevel);
5902 * Drop all container device buffers, from now on
5903 * the only valid external interface is through the md
5906 mddev->has_superblocks = false;
5907 rdev_for_each(rdev, mddev) {
5908 if (test_bit(Faulty, &rdev->flags))
5910 sync_blockdev(rdev->bdev);
5911 invalidate_bdev(rdev->bdev);
5912 if (mddev->ro != 1 &&
5913 (bdev_read_only(rdev->bdev) ||
5914 bdev_read_only(rdev->meta_bdev))) {
5917 set_disk_ro(mddev->gendisk, 1);
5921 mddev->has_superblocks = true;
5923 /* perform some consistency tests on the device.
5924 * We don't want the data to overlap the metadata,
5925 * Internal Bitmap issues have been handled elsewhere.
5927 if (rdev->meta_bdev) {
5928 /* Nothing to check */;
5929 } else if (rdev->data_offset < rdev->sb_start) {
5930 if (mddev->dev_sectors &&
5931 rdev->data_offset + mddev->dev_sectors
5933 pr_warn("md: %s: data overlaps metadata\n",
5938 if (rdev->sb_start + rdev->sb_size/512
5939 > rdev->data_offset) {
5940 pr_warn("md: %s: metadata overlaps data\n",
5945 sysfs_notify_dirent_safe(rdev->sysfs_state);
5948 if (!bioset_initialized(&mddev->bio_set)) {
5949 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5953 if (!bioset_initialized(&mddev->sync_set)) {
5954 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5959 spin_lock(&pers_lock);
5960 pers = find_pers(mddev->level, mddev->clevel);
5961 if (!pers || !try_module_get(pers->owner)) {
5962 spin_unlock(&pers_lock);
5963 if (mddev->level != LEVEL_NONE)
5964 pr_warn("md: personality for level %d is not loaded!\n",
5967 pr_warn("md: personality for level %s is not loaded!\n",
5972 spin_unlock(&pers_lock);
5973 if (mddev->level != pers->level) {
5974 mddev->level = pers->level;
5975 mddev->new_level = pers->level;
5977 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5979 if (mddev->reshape_position != MaxSector &&
5980 pers->start_reshape == NULL) {
5981 /* This personality cannot handle reshaping... */
5982 module_put(pers->owner);
5987 if (pers->sync_request) {
5988 /* Warn if this is a potentially silly
5991 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5992 struct md_rdev *rdev2;
5995 rdev_for_each(rdev, mddev)
5996 rdev_for_each(rdev2, mddev) {
5998 rdev->bdev->bd_disk ==
5999 rdev2->bdev->bd_disk) {
6000 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
6002 bdevname(rdev->bdev,b),
6003 bdevname(rdev2->bdev,b2));
6009 pr_warn("True protection against single-disk failure might be compromised.\n");
6012 mddev->recovery = 0;
6013 /* may be over-ridden by personality */
6014 mddev->resync_max_sectors = mddev->dev_sectors;
6016 mddev->ok_start_degraded = start_dirty_degraded;
6018 if (start_readonly && mddev->ro == 0)
6019 mddev->ro = 2; /* read-only, but switch on first write */
6021 err = pers->run(mddev);
6023 pr_warn("md: pers->run() failed ...\n");
6024 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6025 WARN_ONCE(!mddev->external_size,
6026 "%s: default size too small, but 'external_size' not in effect?\n",
6028 pr_warn("md: invalid array_size %llu > default size %llu\n",
6029 (unsigned long long)mddev->array_sectors / 2,
6030 (unsigned long long)pers->size(mddev, 0, 0) / 2);
6033 if (err == 0 && pers->sync_request &&
6034 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6035 struct bitmap *bitmap;
6037 bitmap = md_bitmap_create(mddev, -1);
6038 if (IS_ERR(bitmap)) {
6039 err = PTR_ERR(bitmap);
6040 pr_warn("%s: failed to create bitmap (%d)\n",
6041 mdname(mddev), err);
6043 mddev->bitmap = bitmap;
6049 if (mddev->bitmap_info.max_write_behind > 0) {
6050 bool create_pool = false;
6052 rdev_for_each(rdev, mddev) {
6053 if (test_bit(WriteMostly, &rdev->flags) &&
6054 rdev_init_serial(rdev))
6057 if (create_pool && mddev->serial_info_pool == NULL) {
6058 mddev->serial_info_pool =
6059 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6060 sizeof(struct serial_info));
6061 if (!mddev->serial_info_pool) {
6071 rdev_for_each(rdev, mddev) {
6072 if (rdev->raid_disk >= 0 &&
6073 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6078 if (mddev->degraded)
6081 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6083 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6085 if (pers->sync_request) {
6086 if (mddev->kobj.sd &&
6087 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6088 pr_warn("md: cannot register extra attributes for %s\n",
6090 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6091 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6092 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6093 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6096 atomic_set(&mddev->max_corr_read_errors,
6097 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6098 mddev->safemode = 0;
6099 if (mddev_is_clustered(mddev))
6100 mddev->safemode_delay = 0;
6102 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6105 spin_lock(&mddev->lock);
6107 spin_unlock(&mddev->lock);
6108 rdev_for_each(rdev, mddev)
6109 if (rdev->raid_disk >= 0)
6110 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6112 if (mddev->degraded && !mddev->ro)
6113 /* This ensures that recovering status is reported immediately
6114 * via sysfs - until a lack of spares is confirmed.
6116 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6117 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6119 if (mddev->sb_flags)
6120 md_update_sb(mddev, 0);
6122 md_new_event(mddev);
6126 mddev_detach(mddev);
6128 pers->free(mddev, mddev->private);
6129 mddev->private = NULL;
6130 module_put(pers->owner);
6131 md_bitmap_destroy(mddev);
6133 bioset_exit(&mddev->bio_set);
6134 bioset_exit(&mddev->sync_set);
6137 EXPORT_SYMBOL_GPL(md_run);
6139 int do_md_run(struct mddev *mddev)
6143 set_bit(MD_NOT_READY, &mddev->flags);
6144 err = md_run(mddev);
6147 err = md_bitmap_load(mddev);
6149 md_bitmap_destroy(mddev);
6153 if (mddev_is_clustered(mddev))
6154 md_allow_write(mddev);
6156 /* run start up tasks that require md_thread */
6159 md_wakeup_thread(mddev->thread);
6160 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6162 set_capacity(mddev->gendisk, mddev->array_sectors);
6163 revalidate_disk_size(mddev->gendisk, true);
6164 clear_bit(MD_NOT_READY, &mddev->flags);
6166 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6167 sysfs_notify_dirent_safe(mddev->sysfs_state);
6168 sysfs_notify_dirent_safe(mddev->sysfs_action);
6169 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6171 clear_bit(MD_NOT_READY, &mddev->flags);
6175 int md_start(struct mddev *mddev)
6179 if (mddev->pers->start) {
6180 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6181 md_wakeup_thread(mddev->thread);
6182 ret = mddev->pers->start(mddev);
6183 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6184 md_wakeup_thread(mddev->sync_thread);
6188 EXPORT_SYMBOL_GPL(md_start);
6190 static int restart_array(struct mddev *mddev)
6192 struct gendisk *disk = mddev->gendisk;
6193 struct md_rdev *rdev;
6194 bool has_journal = false;
6195 bool has_readonly = false;
6197 /* Complain if it has no devices */
6198 if (list_empty(&mddev->disks))
6206 rdev_for_each_rcu(rdev, mddev) {
6207 if (test_bit(Journal, &rdev->flags) &&
6208 !test_bit(Faulty, &rdev->flags))
6210 if (bdev_read_only(rdev->bdev))
6211 has_readonly = true;
6214 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6215 /* Don't restart rw with journal missing/faulty */
6220 mddev->safemode = 0;
6222 set_disk_ro(disk, 0);
6223 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6224 /* Kick recovery or resync if necessary */
6225 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6226 md_wakeup_thread(mddev->thread);
6227 md_wakeup_thread(mddev->sync_thread);
6228 sysfs_notify_dirent_safe(mddev->sysfs_state);
6232 static void md_clean(struct mddev *mddev)
6234 mddev->array_sectors = 0;
6235 mddev->external_size = 0;
6236 mddev->dev_sectors = 0;
6237 mddev->raid_disks = 0;
6238 mddev->recovery_cp = 0;
6239 mddev->resync_min = 0;
6240 mddev->resync_max = MaxSector;
6241 mddev->reshape_position = MaxSector;
6242 mddev->external = 0;
6243 mddev->persistent = 0;
6244 mddev->level = LEVEL_NONE;
6245 mddev->clevel[0] = 0;
6247 * Don't clear MD_CLOSING, or mddev can be opened again.
6248 * 'hold_active != 0' means mddev is still in the creation
6249 * process and will be used later.
6251 if (mddev->hold_active)
6254 mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6255 mddev->sb_flags = 0;
6257 mddev->metadata_type[0] = 0;
6258 mddev->chunk_sectors = 0;
6259 mddev->ctime = mddev->utime = 0;
6261 mddev->max_disks = 0;
6263 mddev->can_decrease_events = 0;
6264 mddev->delta_disks = 0;
6265 mddev->reshape_backwards = 0;
6266 mddev->new_level = LEVEL_NONE;
6267 mddev->new_layout = 0;
6268 mddev->new_chunk_sectors = 0;
6269 mddev->curr_resync = 0;
6270 atomic64_set(&mddev->resync_mismatches, 0);
6271 mddev->suspend_lo = mddev->suspend_hi = 0;
6272 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6273 mddev->recovery = 0;
6276 mddev->degraded = 0;
6277 mddev->safemode = 0;
6278 mddev->private = NULL;
6279 mddev->cluster_info = NULL;
6280 mddev->bitmap_info.offset = 0;
6281 mddev->bitmap_info.default_offset = 0;
6282 mddev->bitmap_info.default_space = 0;
6283 mddev->bitmap_info.chunksize = 0;
6284 mddev->bitmap_info.daemon_sleep = 0;
6285 mddev->bitmap_info.max_write_behind = 0;
6286 mddev->bitmap_info.nodes = 0;
6289 static void __md_stop_writes(struct mddev *mddev)
6291 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6292 if (work_pending(&mddev->del_work))
6293 flush_workqueue(md_misc_wq);
6294 if (mddev->sync_thread) {
6295 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6296 md_reap_sync_thread(mddev);
6299 del_timer_sync(&mddev->safemode_timer);
6301 if (mddev->pers && mddev->pers->quiesce) {
6302 mddev->pers->quiesce(mddev, 1);
6303 mddev->pers->quiesce(mddev, 0);
6305 md_bitmap_flush(mddev);
6307 if (mddev->ro == 0 &&
6308 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6310 /* mark array as shutdown cleanly */
6311 if (!mddev_is_clustered(mddev))
6313 md_update_sb(mddev, 1);
6315 /* disable policy to guarantee rdevs free resources for serialization */
6316 mddev->serialize_policy = 0;
6317 mddev_destroy_serial_pool(mddev, NULL, true);
6320 void md_stop_writes(struct mddev *mddev)
6322 mddev_lock_nointr(mddev);
6323 __md_stop_writes(mddev);
6324 mddev_unlock(mddev);
6326 EXPORT_SYMBOL_GPL(md_stop_writes);
6328 static void mddev_detach(struct mddev *mddev)
6330 md_bitmap_wait_behind_writes(mddev);
6331 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6332 mddev->pers->quiesce(mddev, 1);
6333 mddev->pers->quiesce(mddev, 0);
6335 md_unregister_thread(&mddev->thread);
6337 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6340 static void __md_stop(struct mddev *mddev)
6342 struct md_personality *pers = mddev->pers;
6343 md_bitmap_destroy(mddev);
6344 mddev_detach(mddev);
6345 /* Ensure ->event_work is done */
6346 if (mddev->event_work.func)
6347 flush_workqueue(md_misc_wq);
6348 spin_lock(&mddev->lock);
6350 spin_unlock(&mddev->lock);
6351 pers->free(mddev, mddev->private);
6352 mddev->private = NULL;
6353 if (pers->sync_request && mddev->to_remove == NULL)
6354 mddev->to_remove = &md_redundancy_group;
6355 module_put(pers->owner);
6356 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6359 void md_stop(struct mddev *mddev)
6361 lockdep_assert_held(&mddev->reconfig_mutex);
6363 /* stop the array and free an attached data structures.
6364 * This is called from dm-raid
6366 __md_stop_writes(mddev);
6368 bioset_exit(&mddev->bio_set);
6369 bioset_exit(&mddev->sync_set);
6372 EXPORT_SYMBOL_GPL(md_stop);
6374 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6379 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6381 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6382 md_wakeup_thread(mddev->thread);
6384 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6385 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6386 if (mddev->sync_thread)
6387 /* Thread might be blocked waiting for metadata update
6388 * which will now never happen */
6389 wake_up_process(mddev->sync_thread->tsk);
6391 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6393 mddev_unlock(mddev);
6394 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6396 wait_event(mddev->sb_wait,
6397 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6398 mddev_lock_nointr(mddev);
6400 mutex_lock(&mddev->open_mutex);
6401 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6402 mddev->sync_thread ||
6403 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6404 pr_warn("md: %s still in use.\n",mdname(mddev));
6406 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6407 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6408 md_wakeup_thread(mddev->thread);
6414 __md_stop_writes(mddev);
6420 set_disk_ro(mddev->gendisk, 1);
6421 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6422 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6423 md_wakeup_thread(mddev->thread);
6424 sysfs_notify_dirent_safe(mddev->sysfs_state);
6428 mutex_unlock(&mddev->open_mutex);
6433 * 0 - completely stop and dis-assemble array
6434 * 2 - stop but do not disassemble array
6436 static int do_md_stop(struct mddev *mddev, int mode,
6437 struct block_device *bdev)
6439 struct gendisk *disk = mddev->gendisk;
6440 struct md_rdev *rdev;
6443 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6445 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6446 md_wakeup_thread(mddev->thread);
6448 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6449 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6450 if (mddev->sync_thread)
6451 /* Thread might be blocked waiting for metadata update
6452 * which will now never happen */
6453 wake_up_process(mddev->sync_thread->tsk);
6455 mddev_unlock(mddev);
6456 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6457 !test_bit(MD_RECOVERY_RUNNING,
6458 &mddev->recovery)));
6459 mddev_lock_nointr(mddev);
6461 mutex_lock(&mddev->open_mutex);
6462 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6463 mddev->sysfs_active ||
6464 mddev->sync_thread ||
6465 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6466 pr_warn("md: %s still in use.\n",mdname(mddev));
6467 mutex_unlock(&mddev->open_mutex);
6469 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6470 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6471 md_wakeup_thread(mddev->thread);
6477 set_disk_ro(disk, 0);
6479 __md_stop_writes(mddev);
6482 /* tell userspace to handle 'inactive' */
6483 sysfs_notify_dirent_safe(mddev->sysfs_state);
6485 rdev_for_each(rdev, mddev)
6486 if (rdev->raid_disk >= 0)
6487 sysfs_unlink_rdev(mddev, rdev);
6489 set_capacity(disk, 0);
6490 mutex_unlock(&mddev->open_mutex);
6492 revalidate_disk_size(disk, true);
6497 mutex_unlock(&mddev->open_mutex);
6499 * Free resources if final stop
6502 pr_info("md: %s stopped.\n", mdname(mddev));
6504 if (mddev->bitmap_info.file) {
6505 struct file *f = mddev->bitmap_info.file;
6506 spin_lock(&mddev->lock);
6507 mddev->bitmap_info.file = NULL;
6508 spin_unlock(&mddev->lock);
6511 mddev->bitmap_info.offset = 0;
6513 export_array(mddev);
6516 if (mddev->hold_active == UNTIL_STOP)
6517 mddev->hold_active = 0;
6519 md_new_event(mddev);
6520 sysfs_notify_dirent_safe(mddev->sysfs_state);
6525 static void autorun_array(struct mddev *mddev)
6527 struct md_rdev *rdev;
6530 if (list_empty(&mddev->disks))
6533 pr_info("md: running: ");
6535 rdev_for_each(rdev, mddev) {
6536 char b[BDEVNAME_SIZE];
6537 pr_cont("<%s>", bdevname(rdev->bdev,b));
6541 err = do_md_run(mddev);
6543 pr_warn("md: do_md_run() returned %d\n", err);
6544 do_md_stop(mddev, 0, NULL);
6549 * lets try to run arrays based on all disks that have arrived
6550 * until now. (those are in pending_raid_disks)
6552 * the method: pick the first pending disk, collect all disks with
6553 * the same UUID, remove all from the pending list and put them into
6554 * the 'same_array' list. Then order this list based on superblock
6555 * update time (freshest comes first), kick out 'old' disks and
6556 * compare superblocks. If everything's fine then run it.
6558 * If "unit" is allocated, then bump its reference count
6560 static void autorun_devices(int part)
6562 struct md_rdev *rdev0, *rdev, *tmp;
6563 struct mddev *mddev;
6564 char b[BDEVNAME_SIZE];
6566 pr_info("md: autorun ...\n");
6567 while (!list_empty(&pending_raid_disks)) {
6570 LIST_HEAD(candidates);
6571 rdev0 = list_entry(pending_raid_disks.next,
6572 struct md_rdev, same_set);
6574 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6575 INIT_LIST_HEAD(&candidates);
6576 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6577 if (super_90_load(rdev, rdev0, 0) >= 0) {
6578 pr_debug("md: adding %s ...\n",
6579 bdevname(rdev->bdev,b));
6580 list_move(&rdev->same_set, &candidates);
6583 * now we have a set of devices, with all of them having
6584 * mostly sane superblocks. It's time to allocate the
6588 dev = MKDEV(mdp_major,
6589 rdev0->preferred_minor << MdpMinorShift);
6590 unit = MINOR(dev) >> MdpMinorShift;
6592 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6595 if (rdev0->preferred_minor != unit) {
6596 pr_warn("md: unit number in %s is bad: %d\n",
6597 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6601 md_probe(dev, NULL, NULL);
6602 mddev = mddev_find(dev);
6606 if (mddev_lock(mddev))
6607 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6608 else if (mddev->raid_disks || mddev->major_version
6609 || !list_empty(&mddev->disks)) {
6610 pr_warn("md: %s already running, cannot run %s\n",
6611 mdname(mddev), bdevname(rdev0->bdev,b));
6612 mddev_unlock(mddev);
6614 pr_debug("md: created %s\n", mdname(mddev));
6615 mddev->persistent = 1;
6616 rdev_for_each_list(rdev, tmp, &candidates) {
6617 list_del_init(&rdev->same_set);
6618 if (bind_rdev_to_array(rdev, mddev))
6621 autorun_array(mddev);
6622 mddev_unlock(mddev);
6624 /* on success, candidates will be empty, on error
6627 rdev_for_each_list(rdev, tmp, &candidates) {
6628 list_del_init(&rdev->same_set);
6633 pr_info("md: ... autorun DONE.\n");
6635 #endif /* !MODULE */
6637 static int get_version(void __user *arg)
6641 ver.major = MD_MAJOR_VERSION;
6642 ver.minor = MD_MINOR_VERSION;
6643 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6645 if (copy_to_user(arg, &ver, sizeof(ver)))
6651 static int get_array_info(struct mddev *mddev, void __user *arg)
6653 mdu_array_info_t info;
6654 int nr,working,insync,failed,spare;
6655 struct md_rdev *rdev;
6657 nr = working = insync = failed = spare = 0;
6659 rdev_for_each_rcu(rdev, mddev) {
6661 if (test_bit(Faulty, &rdev->flags))
6665 if (test_bit(In_sync, &rdev->flags))
6667 else if (test_bit(Journal, &rdev->flags))
6668 /* TODO: add journal count to md_u.h */
6676 info.major_version = mddev->major_version;
6677 info.minor_version = mddev->minor_version;
6678 info.patch_version = MD_PATCHLEVEL_VERSION;
6679 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6680 info.level = mddev->level;
6681 info.size = mddev->dev_sectors / 2;
6682 if (info.size != mddev->dev_sectors / 2) /* overflow */
6685 info.raid_disks = mddev->raid_disks;
6686 info.md_minor = mddev->md_minor;
6687 info.not_persistent= !mddev->persistent;
6689 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6692 info.state = (1<<MD_SB_CLEAN);
6693 if (mddev->bitmap && mddev->bitmap_info.offset)
6694 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6695 if (mddev_is_clustered(mddev))
6696 info.state |= (1<<MD_SB_CLUSTERED);
6697 info.active_disks = insync;
6698 info.working_disks = working;
6699 info.failed_disks = failed;
6700 info.spare_disks = spare;
6702 info.layout = mddev->layout;
6703 info.chunk_size = mddev->chunk_sectors << 9;
6705 if (copy_to_user(arg, &info, sizeof(info)))
6711 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6713 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6717 file = kzalloc(sizeof(*file), GFP_NOIO);
6722 spin_lock(&mddev->lock);
6723 /* bitmap enabled */
6724 if (mddev->bitmap_info.file) {
6725 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6726 sizeof(file->pathname));
6730 memmove(file->pathname, ptr,
6731 sizeof(file->pathname)-(ptr-file->pathname));
6733 spin_unlock(&mddev->lock);
6736 copy_to_user(arg, file, sizeof(*file)))
6743 static int get_disk_info(struct mddev *mddev, void __user * arg)
6745 mdu_disk_info_t info;
6746 struct md_rdev *rdev;
6748 if (copy_from_user(&info, arg, sizeof(info)))
6752 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6754 info.major = MAJOR(rdev->bdev->bd_dev);
6755 info.minor = MINOR(rdev->bdev->bd_dev);
6756 info.raid_disk = rdev->raid_disk;
6758 if (test_bit(Faulty, &rdev->flags))
6759 info.state |= (1<<MD_DISK_FAULTY);
6760 else if (test_bit(In_sync, &rdev->flags)) {
6761 info.state |= (1<<MD_DISK_ACTIVE);
6762 info.state |= (1<<MD_DISK_SYNC);
6764 if (test_bit(Journal, &rdev->flags))
6765 info.state |= (1<<MD_DISK_JOURNAL);
6766 if (test_bit(WriteMostly, &rdev->flags))
6767 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6768 if (test_bit(FailFast, &rdev->flags))
6769 info.state |= (1<<MD_DISK_FAILFAST);
6771 info.major = info.minor = 0;
6772 info.raid_disk = -1;
6773 info.state = (1<<MD_DISK_REMOVED);
6777 if (copy_to_user(arg, &info, sizeof(info)))
6783 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6785 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6786 struct md_rdev *rdev;
6787 dev_t dev = MKDEV(info->major,info->minor);
6789 if (mddev_is_clustered(mddev) &&
6790 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6791 pr_warn("%s: Cannot add to clustered mddev.\n",
6796 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6799 if (!mddev->raid_disks) {
6801 /* expecting a device which has a superblock */
6802 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6804 pr_warn("md: md_import_device returned %ld\n",
6806 return PTR_ERR(rdev);
6808 if (!list_empty(&mddev->disks)) {
6809 struct md_rdev *rdev0
6810 = list_entry(mddev->disks.next,
6811 struct md_rdev, same_set);
6812 err = super_types[mddev->major_version]
6813 .load_super(rdev, rdev0, mddev->minor_version);
6815 pr_warn("md: %s has different UUID to %s\n",
6816 bdevname(rdev->bdev,b),
6817 bdevname(rdev0->bdev,b2));
6822 err = bind_rdev_to_array(rdev, mddev);
6829 * md_add_new_disk can be used once the array is assembled
6830 * to add "hot spares". They must already have a superblock
6835 if (!mddev->pers->hot_add_disk) {
6836 pr_warn("%s: personality does not support diskops!\n",
6840 if (mddev->persistent)
6841 rdev = md_import_device(dev, mddev->major_version,
6842 mddev->minor_version);
6844 rdev = md_import_device(dev, -1, -1);
6846 pr_warn("md: md_import_device returned %ld\n",
6848 return PTR_ERR(rdev);
6850 /* set saved_raid_disk if appropriate */
6851 if (!mddev->persistent) {
6852 if (info->state & (1<<MD_DISK_SYNC) &&
6853 info->raid_disk < mddev->raid_disks) {
6854 rdev->raid_disk = info->raid_disk;
6855 set_bit(In_sync, &rdev->flags);
6856 clear_bit(Bitmap_sync, &rdev->flags);
6858 rdev->raid_disk = -1;
6859 rdev->saved_raid_disk = rdev->raid_disk;
6861 super_types[mddev->major_version].
6862 validate_super(mddev, NULL/*freshest*/, rdev);
6863 if ((info->state & (1<<MD_DISK_SYNC)) &&
6864 rdev->raid_disk != info->raid_disk) {
6865 /* This was a hot-add request, but events doesn't
6866 * match, so reject it.
6872 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6873 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6874 set_bit(WriteMostly, &rdev->flags);
6876 clear_bit(WriteMostly, &rdev->flags);
6877 if (info->state & (1<<MD_DISK_FAILFAST))
6878 set_bit(FailFast, &rdev->flags);
6880 clear_bit(FailFast, &rdev->flags);
6882 if (info->state & (1<<MD_DISK_JOURNAL)) {
6883 struct md_rdev *rdev2;
6884 bool has_journal = false;
6886 /* make sure no existing journal disk */
6887 rdev_for_each(rdev2, mddev) {
6888 if (test_bit(Journal, &rdev2->flags)) {
6893 if (has_journal || mddev->bitmap) {
6897 set_bit(Journal, &rdev->flags);
6900 * check whether the device shows up in other nodes
6902 if (mddev_is_clustered(mddev)) {
6903 if (info->state & (1 << MD_DISK_CANDIDATE))
6904 set_bit(Candidate, &rdev->flags);
6905 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6906 /* --add initiated by this node */
6907 err = md_cluster_ops->add_new_disk(mddev, rdev);
6915 rdev->raid_disk = -1;
6916 err = bind_rdev_to_array(rdev, mddev);
6921 if (mddev_is_clustered(mddev)) {
6922 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6924 err = md_cluster_ops->new_disk_ack(mddev,
6927 md_kick_rdev_from_array(rdev);
6931 md_cluster_ops->add_new_disk_cancel(mddev);
6933 err = add_bound_rdev(rdev);
6937 err = add_bound_rdev(rdev);
6942 /* otherwise, md_add_new_disk is only allowed
6943 * for major_version==0 superblocks
6945 if (mddev->major_version != 0) {
6946 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6950 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6952 rdev = md_import_device(dev, -1, 0);
6954 pr_warn("md: error, md_import_device() returned %ld\n",
6956 return PTR_ERR(rdev);
6958 rdev->desc_nr = info->number;
6959 if (info->raid_disk < mddev->raid_disks)
6960 rdev->raid_disk = info->raid_disk;
6962 rdev->raid_disk = -1;
6964 if (rdev->raid_disk < mddev->raid_disks)
6965 if (info->state & (1<<MD_DISK_SYNC))
6966 set_bit(In_sync, &rdev->flags);
6968 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6969 set_bit(WriteMostly, &rdev->flags);
6970 if (info->state & (1<<MD_DISK_FAILFAST))
6971 set_bit(FailFast, &rdev->flags);
6973 if (!mddev->persistent) {
6974 pr_debug("md: nonpersistent superblock ...\n");
6975 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6977 rdev->sb_start = calc_dev_sboffset(rdev);
6978 rdev->sectors = rdev->sb_start;
6980 err = bind_rdev_to_array(rdev, mddev);
6990 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6992 char b[BDEVNAME_SIZE];
6993 struct md_rdev *rdev;
6998 rdev = find_rdev(mddev, dev);
7002 if (rdev->raid_disk < 0)
7005 clear_bit(Blocked, &rdev->flags);
7006 remove_and_add_spares(mddev, rdev);
7008 if (rdev->raid_disk >= 0)
7012 if (mddev_is_clustered(mddev)) {
7013 if (md_cluster_ops->remove_disk(mddev, rdev))
7017 md_kick_rdev_from_array(rdev);
7018 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7020 md_wakeup_thread(mddev->thread);
7022 md_update_sb(mddev, 1);
7023 md_new_event(mddev);
7027 pr_debug("md: cannot remove active disk %s from %s ...\n",
7028 bdevname(rdev->bdev,b), mdname(mddev));
7032 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7034 char b[BDEVNAME_SIZE];
7036 struct md_rdev *rdev;
7041 if (mddev->major_version != 0) {
7042 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7046 if (!mddev->pers->hot_add_disk) {
7047 pr_warn("%s: personality does not support diskops!\n",
7052 rdev = md_import_device(dev, -1, 0);
7054 pr_warn("md: error, md_import_device() returned %ld\n",
7059 if (mddev->persistent)
7060 rdev->sb_start = calc_dev_sboffset(rdev);
7062 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
7064 rdev->sectors = rdev->sb_start;
7066 if (test_bit(Faulty, &rdev->flags)) {
7067 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
7068 bdevname(rdev->bdev,b), mdname(mddev));
7073 clear_bit(In_sync, &rdev->flags);
7075 rdev->saved_raid_disk = -1;
7076 err = bind_rdev_to_array(rdev, mddev);
7081 * The rest should better be atomic, we can have disk failures
7082 * noticed in interrupt contexts ...
7085 rdev->raid_disk = -1;
7087 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7089 md_update_sb(mddev, 1);
7091 * Kick recovery, maybe this spare has to be added to the
7092 * array immediately.
7094 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7095 md_wakeup_thread(mddev->thread);
7096 md_new_event(mddev);
7104 static int set_bitmap_file(struct mddev *mddev, int fd)
7109 if (!mddev->pers->quiesce || !mddev->thread)
7111 if (mddev->recovery || mddev->sync_thread)
7113 /* we should be able to change the bitmap.. */
7117 struct inode *inode;
7120 if (mddev->bitmap || mddev->bitmap_info.file)
7121 return -EEXIST; /* cannot add when bitmap is present */
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) ||
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(mddev->gendisk, mddev->array_sectors);
7324 revalidate_disk_size(mddev->gendisk, true);
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)
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(Faulty, &rdev->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_ioctl(struct block_device *bdev, fmode_t mode,
7568 unsigned int cmd, unsigned long arg)
7571 void __user *argp = (void __user *)arg;
7572 struct mddev *mddev = NULL;
7574 if (!md_ioctl_valid(cmd))
7579 case GET_ARRAY_INFO:
7583 if (!capable(CAP_SYS_ADMIN))
7588 * Commands dealing with the RAID driver but not any
7593 err = get_version(argp);
7599 * Commands creating/starting a new array:
7602 mddev = bdev->bd_disk->private_data;
7609 /* Some actions do not requires the mutex */
7611 case GET_ARRAY_INFO:
7612 if (!mddev->raid_disks && !mddev->external)
7615 err = get_array_info(mddev, argp);
7619 if (!mddev->raid_disks && !mddev->external)
7622 err = get_disk_info(mddev, argp);
7625 case SET_DISK_FAULTY:
7626 err = set_disk_faulty(mddev, new_decode_dev(arg));
7629 case GET_BITMAP_FILE:
7630 err = get_bitmap_file(mddev, argp);
7635 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7636 flush_rdev_wq(mddev);
7638 if (cmd == HOT_REMOVE_DISK)
7639 /* need to ensure recovery thread has run */
7640 wait_event_interruptible_timeout(mddev->sb_wait,
7641 !test_bit(MD_RECOVERY_NEEDED,
7643 msecs_to_jiffies(5000));
7644 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7645 /* Need to flush page cache, and ensure no-one else opens
7648 mutex_lock(&mddev->open_mutex);
7649 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7650 mutex_unlock(&mddev->open_mutex);
7654 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7655 mutex_unlock(&mddev->open_mutex);
7659 mutex_unlock(&mddev->open_mutex);
7660 sync_blockdev(bdev);
7662 err = mddev_lock(mddev);
7664 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7669 if (cmd == SET_ARRAY_INFO) {
7670 mdu_array_info_t info;
7672 memset(&info, 0, sizeof(info));
7673 else if (copy_from_user(&info, argp, sizeof(info))) {
7678 err = update_array_info(mddev, &info);
7680 pr_warn("md: couldn't update array info. %d\n", err);
7685 if (!list_empty(&mddev->disks)) {
7686 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7690 if (mddev->raid_disks) {
7691 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7695 err = md_set_array_info(mddev, &info);
7697 pr_warn("md: couldn't set array info. %d\n", err);
7704 * Commands querying/configuring an existing array:
7706 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7707 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7708 if ((!mddev->raid_disks && !mddev->external)
7709 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7710 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7711 && cmd != GET_BITMAP_FILE) {
7717 * Commands even a read-only array can execute:
7720 case RESTART_ARRAY_RW:
7721 err = restart_array(mddev);
7725 err = do_md_stop(mddev, 0, bdev);
7729 err = md_set_readonly(mddev, bdev);
7732 case HOT_REMOVE_DISK:
7733 err = hot_remove_disk(mddev, new_decode_dev(arg));
7737 /* We can support ADD_NEW_DISK on read-only arrays
7738 * only if we are re-adding a preexisting device.
7739 * So require mddev->pers and MD_DISK_SYNC.
7742 mdu_disk_info_t info;
7743 if (copy_from_user(&info, argp, sizeof(info)))
7745 else if (!(info.state & (1<<MD_DISK_SYNC)))
7746 /* Need to clear read-only for this */
7749 err = md_add_new_disk(mddev, &info);
7756 * The remaining ioctls are changing the state of the
7757 * superblock, so we do not allow them on read-only arrays.
7759 if (mddev->ro && mddev->pers) {
7760 if (mddev->ro == 2) {
7762 sysfs_notify_dirent_safe(mddev->sysfs_state);
7763 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7764 /* mddev_unlock will wake thread */
7765 /* If a device failed while we were read-only, we
7766 * need to make sure the metadata is updated now.
7768 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7769 mddev_unlock(mddev);
7770 wait_event(mddev->sb_wait,
7771 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7772 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7773 mddev_lock_nointr(mddev);
7784 mdu_disk_info_t info;
7785 if (copy_from_user(&info, argp, sizeof(info)))
7788 err = md_add_new_disk(mddev, &info);
7792 case CLUSTERED_DISK_NACK:
7793 if (mddev_is_clustered(mddev))
7794 md_cluster_ops->new_disk_ack(mddev, false);
7800 err = hot_add_disk(mddev, new_decode_dev(arg));
7804 err = do_md_run(mddev);
7807 case SET_BITMAP_FILE:
7808 err = set_bitmap_file(mddev, (int)arg);
7817 if (mddev->hold_active == UNTIL_IOCTL &&
7819 mddev->hold_active = 0;
7820 mddev_unlock(mddev);
7822 if (cmd == STOP_ARRAY_RO || (err && cmd == STOP_ARRAY))
7823 clear_bit(MD_CLOSING, &mddev->flags);
7826 #ifdef CONFIG_COMPAT
7827 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7828 unsigned int cmd, unsigned long arg)
7831 case HOT_REMOVE_DISK:
7833 case SET_DISK_FAULTY:
7834 case SET_BITMAP_FILE:
7835 /* These take in integer arg, do not convert */
7838 arg = (unsigned long)compat_ptr(arg);
7842 return md_ioctl(bdev, mode, cmd, arg);
7844 #endif /* CONFIG_COMPAT */
7846 static int md_set_read_only(struct block_device *bdev, bool ro)
7848 struct mddev *mddev = bdev->bd_disk->private_data;
7851 err = mddev_lock(mddev);
7855 if (!mddev->raid_disks && !mddev->external) {
7861 * Transitioning to read-auto need only happen for arrays that call
7862 * md_write_start and which are not ready for writes yet.
7864 if (!ro && mddev->ro == 1 && mddev->pers) {
7865 err = restart_array(mddev);
7872 mddev_unlock(mddev);
7876 static int md_open(struct block_device *bdev, fmode_t mode)
7879 * Succeed if we can lock the mddev, which confirms that
7880 * it isn't being stopped right now.
7882 struct mddev *mddev = mddev_find(bdev->bd_dev);
7888 if (mddev->gendisk != bdev->bd_disk) {
7889 /* we are racing with mddev_put which is discarding this
7893 /* Wait until bdev->bd_disk is definitely gone */
7894 if (work_pending(&mddev->del_work))
7895 flush_workqueue(md_misc_wq);
7898 BUG_ON(mddev != bdev->bd_disk->private_data);
7900 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7903 if (test_bit(MD_CLOSING, &mddev->flags)) {
7904 mutex_unlock(&mddev->open_mutex);
7910 atomic_inc(&mddev->openers);
7911 mutex_unlock(&mddev->open_mutex);
7913 bdev_check_media_change(bdev);
7920 static void md_release(struct gendisk *disk, fmode_t mode)
7922 struct mddev *mddev = disk->private_data;
7925 atomic_dec(&mddev->openers);
7929 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7931 struct mddev *mddev = disk->private_data;
7932 unsigned int ret = 0;
7935 ret = DISK_EVENT_MEDIA_CHANGE;
7940 const struct block_device_operations md_fops =
7942 .owner = THIS_MODULE,
7943 .submit_bio = md_submit_bio,
7945 .release = md_release,
7947 #ifdef CONFIG_COMPAT
7948 .compat_ioctl = md_compat_ioctl,
7950 .getgeo = md_getgeo,
7951 .check_events = md_check_events,
7952 .set_read_only = md_set_read_only,
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 void md_wakeup_thread(struct md_thread *thread)
8001 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
8002 set_bit(THREAD_WAKEUP, &thread->flags);
8003 wake_up(&thread->wqueue);
8006 EXPORT_SYMBOL(md_wakeup_thread);
8008 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
8009 struct mddev *mddev, const char *name)
8011 struct md_thread *thread;
8013 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
8017 init_waitqueue_head(&thread->wqueue);
8020 thread->mddev = mddev;
8021 thread->timeout = MAX_SCHEDULE_TIMEOUT;
8022 thread->tsk = kthread_run(md_thread, thread,
8024 mdname(thread->mddev),
8026 if (IS_ERR(thread->tsk)) {
8032 EXPORT_SYMBOL(md_register_thread);
8034 void md_unregister_thread(struct md_thread **threadp)
8036 struct md_thread *thread;
8039 * Locking ensures that mddev_unlock does not wake_up a
8040 * non-existent thread
8042 spin_lock(&pers_lock);
8045 spin_unlock(&pers_lock);
8049 spin_unlock(&pers_lock);
8051 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8052 kthread_stop(thread->tsk);
8055 EXPORT_SYMBOL(md_unregister_thread);
8057 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8059 if (!rdev || test_bit(Faulty, &rdev->flags))
8062 if (!mddev->pers || !mddev->pers->error_handler)
8064 mddev->pers->error_handler(mddev,rdev);
8065 if (mddev->degraded)
8066 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8067 sysfs_notify_dirent_safe(rdev->sysfs_state);
8068 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8069 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8070 md_wakeup_thread(mddev->thread);
8071 if (mddev->event_work.func)
8072 queue_work(md_misc_wq, &mddev->event_work);
8073 md_new_event(mddev);
8075 EXPORT_SYMBOL(md_error);
8077 /* seq_file implementation /proc/mdstat */
8079 static void status_unused(struct seq_file *seq)
8082 struct md_rdev *rdev;
8084 seq_printf(seq, "unused devices: ");
8086 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8087 char b[BDEVNAME_SIZE];
8089 seq_printf(seq, "%s ",
8090 bdevname(rdev->bdev,b));
8093 seq_printf(seq, "<none>");
8095 seq_printf(seq, "\n");
8098 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8100 sector_t max_sectors, resync, res;
8101 unsigned long dt, db = 0;
8102 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8103 int scale, recovery_active;
8104 unsigned int per_milli;
8106 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8107 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8108 max_sectors = mddev->resync_max_sectors;
8110 max_sectors = mddev->dev_sectors;
8112 resync = mddev->curr_resync;
8114 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8115 /* Still cleaning up */
8116 resync = max_sectors;
8117 } else if (resync > max_sectors)
8118 resync = max_sectors;
8120 resync -= atomic_read(&mddev->recovery_active);
8123 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8124 struct md_rdev *rdev;
8126 rdev_for_each(rdev, mddev)
8127 if (rdev->raid_disk >= 0 &&
8128 !test_bit(Faulty, &rdev->flags) &&
8129 rdev->recovery_offset != MaxSector &&
8130 rdev->recovery_offset) {
8131 seq_printf(seq, "\trecover=REMOTE");
8134 if (mddev->reshape_position != MaxSector)
8135 seq_printf(seq, "\treshape=REMOTE");
8137 seq_printf(seq, "\tresync=REMOTE");
8140 if (mddev->recovery_cp < MaxSector) {
8141 seq_printf(seq, "\tresync=PENDING");
8147 seq_printf(seq, "\tresync=DELAYED");
8151 WARN_ON(max_sectors == 0);
8152 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8153 * in a sector_t, and (max_sectors>>scale) will fit in a
8154 * u32, as those are the requirements for sector_div.
8155 * Thus 'scale' must be at least 10
8158 if (sizeof(sector_t) > sizeof(unsigned long)) {
8159 while ( max_sectors/2 > (1ULL<<(scale+32)))
8162 res = (resync>>scale)*1000;
8163 sector_div(res, (u32)((max_sectors>>scale)+1));
8167 int i, x = per_milli/50, y = 20-x;
8168 seq_printf(seq, "[");
8169 for (i = 0; i < x; i++)
8170 seq_printf(seq, "=");
8171 seq_printf(seq, ">");
8172 for (i = 0; i < y; i++)
8173 seq_printf(seq, ".");
8174 seq_printf(seq, "] ");
8176 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8177 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8179 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8181 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8182 "resync" : "recovery"))),
8183 per_milli/10, per_milli % 10,
8184 (unsigned long long) resync/2,
8185 (unsigned long long) max_sectors/2);
8188 * dt: time from mark until now
8189 * db: blocks written from mark until now
8190 * rt: remaining time
8192 * rt is a sector_t, which is always 64bit now. We are keeping
8193 * the original algorithm, but it is not really necessary.
8195 * Original algorithm:
8196 * So we divide before multiply in case it is 32bit and close
8198 * We scale the divisor (db) by 32 to avoid losing precision
8199 * near the end of resync when the number of remaining sectors
8201 * We then divide rt by 32 after multiplying by db to compensate.
8202 * The '+1' avoids division by zero if db is very small.
8204 dt = ((jiffies - mddev->resync_mark) / HZ);
8207 curr_mark_cnt = mddev->curr_mark_cnt;
8208 recovery_active = atomic_read(&mddev->recovery_active);
8209 resync_mark_cnt = mddev->resync_mark_cnt;
8211 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8212 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8214 rt = max_sectors - resync; /* number of remaining sectors */
8215 rt = div64_u64(rt, db/32+1);
8219 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8220 ((unsigned long)rt % 60)/6);
8222 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8226 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8228 struct list_head *tmp;
8230 struct mddev *mddev;
8242 spin_lock(&all_mddevs_lock);
8243 list_for_each(tmp,&all_mddevs)
8245 mddev = list_entry(tmp, struct mddev, all_mddevs);
8247 spin_unlock(&all_mddevs_lock);
8250 spin_unlock(&all_mddevs_lock);
8252 return (void*)2;/* tail */
8256 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8258 struct list_head *tmp;
8259 struct mddev *next_mddev, *mddev = v;
8265 spin_lock(&all_mddevs_lock);
8267 tmp = all_mddevs.next;
8269 tmp = mddev->all_mddevs.next;
8270 if (tmp != &all_mddevs)
8271 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8273 next_mddev = (void*)2;
8276 spin_unlock(&all_mddevs_lock);
8284 static void md_seq_stop(struct seq_file *seq, void *v)
8286 struct mddev *mddev = v;
8288 if (mddev && v != (void*)1 && v != (void*)2)
8292 static int md_seq_show(struct seq_file *seq, void *v)
8294 struct mddev *mddev = v;
8296 struct md_rdev *rdev;
8298 if (v == (void*)1) {
8299 struct md_personality *pers;
8300 seq_printf(seq, "Personalities : ");
8301 spin_lock(&pers_lock);
8302 list_for_each_entry(pers, &pers_list, list)
8303 seq_printf(seq, "[%s] ", pers->name);
8305 spin_unlock(&pers_lock);
8306 seq_printf(seq, "\n");
8307 seq->poll_event = atomic_read(&md_event_count);
8310 if (v == (void*)2) {
8315 spin_lock(&mddev->lock);
8316 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8317 seq_printf(seq, "%s : %sactive", mdname(mddev),
8318 mddev->pers ? "" : "in");
8321 seq_printf(seq, " (read-only)");
8323 seq_printf(seq, " (auto-read-only)");
8324 seq_printf(seq, " %s", mddev->pers->name);
8329 rdev_for_each_rcu(rdev, mddev) {
8330 char b[BDEVNAME_SIZE];
8331 seq_printf(seq, " %s[%d]",
8332 bdevname(rdev->bdev,b), rdev->desc_nr);
8333 if (test_bit(WriteMostly, &rdev->flags))
8334 seq_printf(seq, "(W)");
8335 if (test_bit(Journal, &rdev->flags))
8336 seq_printf(seq, "(J)");
8337 if (test_bit(Faulty, &rdev->flags)) {
8338 seq_printf(seq, "(F)");
8341 if (rdev->raid_disk < 0)
8342 seq_printf(seq, "(S)"); /* spare */
8343 if (test_bit(Replacement, &rdev->flags))
8344 seq_printf(seq, "(R)");
8345 sectors += rdev->sectors;
8349 if (!list_empty(&mddev->disks)) {
8351 seq_printf(seq, "\n %llu blocks",
8352 (unsigned long long)
8353 mddev->array_sectors / 2);
8355 seq_printf(seq, "\n %llu blocks",
8356 (unsigned long long)sectors / 2);
8358 if (mddev->persistent) {
8359 if (mddev->major_version != 0 ||
8360 mddev->minor_version != 90) {
8361 seq_printf(seq," super %d.%d",
8362 mddev->major_version,
8363 mddev->minor_version);
8365 } else if (mddev->external)
8366 seq_printf(seq, " super external:%s",
8367 mddev->metadata_type);
8369 seq_printf(seq, " super non-persistent");
8372 mddev->pers->status(seq, mddev);
8373 seq_printf(seq, "\n ");
8374 if (mddev->pers->sync_request) {
8375 if (status_resync(seq, mddev))
8376 seq_printf(seq, "\n ");
8379 seq_printf(seq, "\n ");
8381 md_bitmap_status(seq, mddev->bitmap);
8383 seq_printf(seq, "\n");
8385 spin_unlock(&mddev->lock);
8390 static const struct seq_operations md_seq_ops = {
8391 .start = md_seq_start,
8392 .next = md_seq_next,
8393 .stop = md_seq_stop,
8394 .show = md_seq_show,
8397 static int md_seq_open(struct inode *inode, struct file *file)
8399 struct seq_file *seq;
8402 error = seq_open(file, &md_seq_ops);
8406 seq = file->private_data;
8407 seq->poll_event = atomic_read(&md_event_count);
8411 static int md_unloading;
8412 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8414 struct seq_file *seq = filp->private_data;
8418 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8419 poll_wait(filp, &md_event_waiters, wait);
8421 /* always allow read */
8422 mask = EPOLLIN | EPOLLRDNORM;
8424 if (seq->poll_event != atomic_read(&md_event_count))
8425 mask |= EPOLLERR | EPOLLPRI;
8429 static const struct proc_ops mdstat_proc_ops = {
8430 .proc_open = md_seq_open,
8431 .proc_read = seq_read,
8432 .proc_lseek = seq_lseek,
8433 .proc_release = seq_release,
8434 .proc_poll = mdstat_poll,
8437 int register_md_personality(struct md_personality *p)
8439 pr_debug("md: %s personality registered for level %d\n",
8441 spin_lock(&pers_lock);
8442 list_add_tail(&p->list, &pers_list);
8443 spin_unlock(&pers_lock);
8446 EXPORT_SYMBOL(register_md_personality);
8448 int unregister_md_personality(struct md_personality *p)
8450 pr_debug("md: %s personality unregistered\n", p->name);
8451 spin_lock(&pers_lock);
8452 list_del_init(&p->list);
8453 spin_unlock(&pers_lock);
8456 EXPORT_SYMBOL(unregister_md_personality);
8458 int register_md_cluster_operations(struct md_cluster_operations *ops,
8459 struct module *module)
8462 spin_lock(&pers_lock);
8463 if (md_cluster_ops != NULL)
8466 md_cluster_ops = ops;
8467 md_cluster_mod = module;
8469 spin_unlock(&pers_lock);
8472 EXPORT_SYMBOL(register_md_cluster_operations);
8474 int unregister_md_cluster_operations(void)
8476 spin_lock(&pers_lock);
8477 md_cluster_ops = NULL;
8478 spin_unlock(&pers_lock);
8481 EXPORT_SYMBOL(unregister_md_cluster_operations);
8483 int md_setup_cluster(struct mddev *mddev, int nodes)
8486 if (!md_cluster_ops)
8487 request_module("md-cluster");
8488 spin_lock(&pers_lock);
8489 /* ensure module won't be unloaded */
8490 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8491 pr_warn("can't find md-cluster module or get it's reference.\n");
8492 spin_unlock(&pers_lock);
8495 spin_unlock(&pers_lock);
8497 ret = md_cluster_ops->join(mddev, nodes);
8499 mddev->safemode_delay = 0;
8503 void md_cluster_stop(struct mddev *mddev)
8505 if (!md_cluster_ops)
8507 md_cluster_ops->leave(mddev);
8508 module_put(md_cluster_mod);
8511 static int is_mddev_idle(struct mddev *mddev, int init)
8513 struct md_rdev *rdev;
8519 rdev_for_each_rcu(rdev, mddev) {
8520 struct gendisk *disk = rdev->bdev->bd_disk;
8521 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8522 atomic_read(&disk->sync_io);
8523 /* sync IO will cause sync_io to increase before the disk_stats
8524 * as sync_io is counted when a request starts, and
8525 * disk_stats is counted when it completes.
8526 * So resync activity will cause curr_events to be smaller than
8527 * when there was no such activity.
8528 * non-sync IO will cause disk_stat to increase without
8529 * increasing sync_io so curr_events will (eventually)
8530 * be larger than it was before. Once it becomes
8531 * substantially larger, the test below will cause
8532 * the array to appear non-idle, and resync will slow
8534 * If there is a lot of outstanding resync activity when
8535 * we set last_event to curr_events, then all that activity
8536 * completing might cause the array to appear non-idle
8537 * and resync will be slowed down even though there might
8538 * not have been non-resync activity. This will only
8539 * happen once though. 'last_events' will soon reflect
8540 * the state where there is little or no outstanding
8541 * resync requests, and further resync activity will
8542 * always make curr_events less than last_events.
8545 if (init || curr_events - rdev->last_events > 64) {
8546 rdev->last_events = curr_events;
8554 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8556 /* another "blocks" (512byte) blocks have been synced */
8557 atomic_sub(blocks, &mddev->recovery_active);
8558 wake_up(&mddev->recovery_wait);
8560 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8561 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8562 md_wakeup_thread(mddev->thread);
8563 // stop recovery, signal do_sync ....
8566 EXPORT_SYMBOL(md_done_sync);
8568 /* md_write_start(mddev, bi)
8569 * If we need to update some array metadata (e.g. 'active' flag
8570 * in superblock) before writing, schedule a superblock update
8571 * and wait for it to complete.
8572 * A return value of 'false' means that the write wasn't recorded
8573 * and cannot proceed as the array is being suspend.
8575 bool md_write_start(struct mddev *mddev, struct bio *bi)
8579 if (bio_data_dir(bi) != WRITE)
8582 BUG_ON(mddev->ro == 1);
8583 if (mddev->ro == 2) {
8584 /* need to switch to read/write */
8586 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8587 md_wakeup_thread(mddev->thread);
8588 md_wakeup_thread(mddev->sync_thread);
8592 percpu_ref_get(&mddev->writes_pending);
8593 smp_mb(); /* Match smp_mb in set_in_sync() */
8594 if (mddev->safemode == 1)
8595 mddev->safemode = 0;
8596 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8597 if (mddev->in_sync || mddev->sync_checkers) {
8598 spin_lock(&mddev->lock);
8599 if (mddev->in_sync) {
8601 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8602 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8603 md_wakeup_thread(mddev->thread);
8606 spin_unlock(&mddev->lock);
8610 sysfs_notify_dirent_safe(mddev->sysfs_state);
8611 if (!mddev->has_superblocks)
8613 wait_event(mddev->sb_wait,
8614 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8616 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8617 percpu_ref_put(&mddev->writes_pending);
8622 EXPORT_SYMBOL(md_write_start);
8624 /* md_write_inc can only be called when md_write_start() has
8625 * already been called at least once of the current request.
8626 * It increments the counter and is useful when a single request
8627 * is split into several parts. Each part causes an increment and
8628 * so needs a matching md_write_end().
8629 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8630 * a spinlocked region.
8632 void md_write_inc(struct mddev *mddev, struct bio *bi)
8634 if (bio_data_dir(bi) != WRITE)
8636 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8637 percpu_ref_get(&mddev->writes_pending);
8639 EXPORT_SYMBOL(md_write_inc);
8641 void md_write_end(struct mddev *mddev)
8643 percpu_ref_put(&mddev->writes_pending);
8645 if (mddev->safemode == 2)
8646 md_wakeup_thread(mddev->thread);
8647 else if (mddev->safemode_delay)
8648 /* The roundup() ensures this only performs locking once
8649 * every ->safemode_delay jiffies
8651 mod_timer(&mddev->safemode_timer,
8652 roundup(jiffies, mddev->safemode_delay) +
8653 mddev->safemode_delay);
8656 EXPORT_SYMBOL(md_write_end);
8658 /* md_allow_write(mddev)
8659 * Calling this ensures that the array is marked 'active' so that writes
8660 * may proceed without blocking. It is important to call this before
8661 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8662 * Must be called with mddev_lock held.
8664 void md_allow_write(struct mddev *mddev)
8670 if (!mddev->pers->sync_request)
8673 spin_lock(&mddev->lock);
8674 if (mddev->in_sync) {
8676 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8677 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8678 if (mddev->safemode_delay &&
8679 mddev->safemode == 0)
8680 mddev->safemode = 1;
8681 spin_unlock(&mddev->lock);
8682 md_update_sb(mddev, 0);
8683 sysfs_notify_dirent_safe(mddev->sysfs_state);
8684 /* wait for the dirty state to be recorded in the metadata */
8685 wait_event(mddev->sb_wait,
8686 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8688 spin_unlock(&mddev->lock);
8690 EXPORT_SYMBOL_GPL(md_allow_write);
8692 #define SYNC_MARKS 10
8693 #define SYNC_MARK_STEP (3*HZ)
8694 #define UPDATE_FREQUENCY (5*60*HZ)
8695 void md_do_sync(struct md_thread *thread)
8697 struct mddev *mddev = thread->mddev;
8698 struct mddev *mddev2;
8699 unsigned int currspeed = 0, window;
8700 sector_t max_sectors,j, io_sectors, recovery_done;
8701 unsigned long mark[SYNC_MARKS];
8702 unsigned long update_time;
8703 sector_t mark_cnt[SYNC_MARKS];
8705 struct list_head *tmp;
8706 sector_t last_check;
8708 struct md_rdev *rdev;
8709 char *desc, *action = NULL;
8710 struct blk_plug plug;
8713 /* just incase thread restarts... */
8714 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8715 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8717 if (mddev->ro) {/* never try to sync a read-only array */
8718 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8722 if (mddev_is_clustered(mddev)) {
8723 ret = md_cluster_ops->resync_start(mddev);
8727 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8728 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8729 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8730 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8731 && ((unsigned long long)mddev->curr_resync_completed
8732 < (unsigned long long)mddev->resync_max_sectors))
8736 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8737 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8738 desc = "data-check";
8740 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8741 desc = "requested-resync";
8745 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8750 mddev->last_sync_action = action ?: desc;
8752 /* we overload curr_resync somewhat here.
8753 * 0 == not engaged in resync at all
8754 * 2 == checking that there is no conflict with another sync
8755 * 1 == like 2, but have yielded to allow conflicting resync to
8757 * other == active in resync - this many blocks
8759 * Before starting a resync we must have set curr_resync to
8760 * 2, and then checked that every "conflicting" array has curr_resync
8761 * less than ours. When we find one that is the same or higher
8762 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8763 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8764 * This will mean we have to start checking from the beginning again.
8769 int mddev2_minor = -1;
8770 mddev->curr_resync = 2;
8773 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8775 for_each_mddev(mddev2, tmp) {
8776 if (mddev2 == mddev)
8778 if (!mddev->parallel_resync
8779 && mddev2->curr_resync
8780 && match_mddev_units(mddev, mddev2)) {
8782 if (mddev < mddev2 && mddev->curr_resync == 2) {
8783 /* arbitrarily yield */
8784 mddev->curr_resync = 1;
8785 wake_up(&resync_wait);
8787 if (mddev > mddev2 && mddev->curr_resync == 1)
8788 /* no need to wait here, we can wait the next
8789 * time 'round when curr_resync == 2
8792 /* We need to wait 'interruptible' so as not to
8793 * contribute to the load average, and not to
8794 * be caught by 'softlockup'
8796 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8797 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8798 mddev2->curr_resync >= mddev->curr_resync) {
8799 if (mddev2_minor != mddev2->md_minor) {
8800 mddev2_minor = mddev2->md_minor;
8801 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8802 desc, mdname(mddev),
8806 if (signal_pending(current))
8807 flush_signals(current);
8809 finish_wait(&resync_wait, &wq);
8812 finish_wait(&resync_wait, &wq);
8815 } while (mddev->curr_resync < 2);
8818 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8819 /* resync follows the size requested by the personality,
8820 * which defaults to physical size, but can be virtual size
8822 max_sectors = mddev->resync_max_sectors;
8823 atomic64_set(&mddev->resync_mismatches, 0);
8824 /* we don't use the checkpoint if there's a bitmap */
8825 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8826 j = mddev->resync_min;
8827 else if (!mddev->bitmap)
8828 j = mddev->recovery_cp;
8830 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8831 max_sectors = mddev->resync_max_sectors;
8833 * If the original node aborts reshaping then we continue the
8834 * reshaping, so set j again to avoid restart reshape from the
8837 if (mddev_is_clustered(mddev) &&
8838 mddev->reshape_position != MaxSector)
8839 j = mddev->reshape_position;
8841 /* recovery follows the physical size of devices */
8842 max_sectors = mddev->dev_sectors;
8845 rdev_for_each_rcu(rdev, mddev)
8846 if (rdev->raid_disk >= 0 &&
8847 !test_bit(Journal, &rdev->flags) &&
8848 !test_bit(Faulty, &rdev->flags) &&
8849 !test_bit(In_sync, &rdev->flags) &&
8850 rdev->recovery_offset < j)
8851 j = rdev->recovery_offset;
8854 /* If there is a bitmap, we need to make sure all
8855 * writes that started before we added a spare
8856 * complete before we start doing a recovery.
8857 * Otherwise the write might complete and (via
8858 * bitmap_endwrite) set a bit in the bitmap after the
8859 * recovery has checked that bit and skipped that
8862 if (mddev->bitmap) {
8863 mddev->pers->quiesce(mddev, 1);
8864 mddev->pers->quiesce(mddev, 0);
8868 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8869 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8870 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8871 speed_max(mddev), desc);
8873 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8876 for (m = 0; m < SYNC_MARKS; m++) {
8878 mark_cnt[m] = io_sectors;
8881 mddev->resync_mark = mark[last_mark];
8882 mddev->resync_mark_cnt = mark_cnt[last_mark];
8885 * Tune reconstruction:
8887 window = 32 * (PAGE_SIZE / 512);
8888 pr_debug("md: using %dk window, over a total of %lluk.\n",
8889 window/2, (unsigned long long)max_sectors/2);
8891 atomic_set(&mddev->recovery_active, 0);
8895 pr_debug("md: resuming %s of %s from checkpoint.\n",
8896 desc, mdname(mddev));
8897 mddev->curr_resync = j;
8899 mddev->curr_resync = 3; /* no longer delayed */
8900 mddev->curr_resync_completed = j;
8901 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8902 md_new_event(mddev);
8903 update_time = jiffies;
8905 blk_start_plug(&plug);
8906 while (j < max_sectors) {
8911 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8912 ((mddev->curr_resync > mddev->curr_resync_completed &&
8913 (mddev->curr_resync - mddev->curr_resync_completed)
8914 > (max_sectors >> 4)) ||
8915 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8916 (j - mddev->curr_resync_completed)*2
8917 >= mddev->resync_max - mddev->curr_resync_completed ||
8918 mddev->curr_resync_completed > mddev->resync_max
8920 /* time to update curr_resync_completed */
8921 wait_event(mddev->recovery_wait,
8922 atomic_read(&mddev->recovery_active) == 0);
8923 mddev->curr_resync_completed = j;
8924 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8925 j > mddev->recovery_cp)
8926 mddev->recovery_cp = j;
8927 update_time = jiffies;
8928 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8929 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8932 while (j >= mddev->resync_max &&
8933 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8934 /* As this condition is controlled by user-space,
8935 * we can block indefinitely, so use '_interruptible'
8936 * to avoid triggering warnings.
8938 flush_signals(current); /* just in case */
8939 wait_event_interruptible(mddev->recovery_wait,
8940 mddev->resync_max > j
8941 || test_bit(MD_RECOVERY_INTR,
8945 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8948 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8950 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8954 if (!skipped) { /* actual IO requested */
8955 io_sectors += sectors;
8956 atomic_add(sectors, &mddev->recovery_active);
8959 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8963 if (j > max_sectors)
8964 /* when skipping, extra large numbers can be returned. */
8967 mddev->curr_resync = j;
8968 mddev->curr_mark_cnt = io_sectors;
8969 if (last_check == 0)
8970 /* this is the earliest that rebuild will be
8971 * visible in /proc/mdstat
8973 md_new_event(mddev);
8975 if (last_check + window > io_sectors || j == max_sectors)
8978 last_check = io_sectors;
8980 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8982 int next = (last_mark+1) % SYNC_MARKS;
8984 mddev->resync_mark = mark[next];
8985 mddev->resync_mark_cnt = mark_cnt[next];
8986 mark[next] = jiffies;
8987 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8991 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8995 * this loop exits only if either when we are slower than
8996 * the 'hard' speed limit, or the system was IO-idle for
8998 * the system might be non-idle CPU-wise, but we only care
8999 * about not overloading the IO subsystem. (things like an
9000 * e2fsck being done on the RAID array should execute fast)
9004 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9005 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9006 /((jiffies-mddev->resync_mark)/HZ +1) +1;
9008 if (currspeed > speed_min(mddev)) {
9009 if (currspeed > speed_max(mddev)) {
9013 if (!is_mddev_idle(mddev, 0)) {
9015 * Give other IO more of a chance.
9016 * The faster the devices, the less we wait.
9018 wait_event(mddev->recovery_wait,
9019 !atomic_read(&mddev->recovery_active));
9023 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9024 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9025 ? "interrupted" : "done");
9027 * this also signals 'finished resyncing' to md_stop
9029 blk_finish_plug(&plug);
9030 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9032 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9033 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9034 mddev->curr_resync > 3) {
9035 mddev->curr_resync_completed = mddev->curr_resync;
9036 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9038 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9040 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9041 mddev->curr_resync > 3) {
9042 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9043 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9044 if (mddev->curr_resync >= mddev->recovery_cp) {
9045 pr_debug("md: checkpointing %s of %s.\n",
9046 desc, mdname(mddev));
9047 if (test_bit(MD_RECOVERY_ERROR,
9049 mddev->recovery_cp =
9050 mddev->curr_resync_completed;
9052 mddev->recovery_cp =
9056 mddev->recovery_cp = MaxSector;
9058 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9059 mddev->curr_resync = MaxSector;
9060 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9061 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9063 rdev_for_each_rcu(rdev, mddev)
9064 if (rdev->raid_disk >= 0 &&
9065 mddev->delta_disks >= 0 &&
9066 !test_bit(Journal, &rdev->flags) &&
9067 !test_bit(Faulty, &rdev->flags) &&
9068 !test_bit(In_sync, &rdev->flags) &&
9069 rdev->recovery_offset < mddev->curr_resync)
9070 rdev->recovery_offset = mddev->curr_resync;
9076 /* set CHANGE_PENDING here since maybe another update is needed,
9077 * so other nodes are informed. It should be harmless for normal
9079 set_mask_bits(&mddev->sb_flags, 0,
9080 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9082 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9083 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9084 mddev->delta_disks > 0 &&
9085 mddev->pers->finish_reshape &&
9086 mddev->pers->size &&
9088 mddev_lock_nointr(mddev);
9089 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9090 mddev_unlock(mddev);
9091 if (!mddev_is_clustered(mddev)) {
9092 set_capacity(mddev->gendisk, mddev->array_sectors);
9093 revalidate_disk_size(mddev->gendisk, true);
9097 spin_lock(&mddev->lock);
9098 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9099 /* We completed so min/max setting can be forgotten if used. */
9100 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9101 mddev->resync_min = 0;
9102 mddev->resync_max = MaxSector;
9103 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9104 mddev->resync_min = mddev->curr_resync_completed;
9105 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9106 mddev->curr_resync = 0;
9107 spin_unlock(&mddev->lock);
9109 wake_up(&resync_wait);
9110 md_wakeup_thread(mddev->thread);
9113 EXPORT_SYMBOL_GPL(md_do_sync);
9115 static int remove_and_add_spares(struct mddev *mddev,
9116 struct md_rdev *this)
9118 struct md_rdev *rdev;
9121 bool remove_some = false;
9123 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9124 /* Mustn't remove devices when resync thread is running */
9127 rdev_for_each(rdev, mddev) {
9128 if ((this == NULL || rdev == this) &&
9129 rdev->raid_disk >= 0 &&
9130 !test_bit(Blocked, &rdev->flags) &&
9131 test_bit(Faulty, &rdev->flags) &&
9132 atomic_read(&rdev->nr_pending)==0) {
9133 /* Faulty non-Blocked devices with nr_pending == 0
9134 * never get nr_pending incremented,
9135 * never get Faulty cleared, and never get Blocked set.
9136 * So we can synchronize_rcu now rather than once per device
9139 set_bit(RemoveSynchronized, &rdev->flags);
9145 rdev_for_each(rdev, mddev) {
9146 if ((this == NULL || rdev == this) &&
9147 rdev->raid_disk >= 0 &&
9148 !test_bit(Blocked, &rdev->flags) &&
9149 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9150 (!test_bit(In_sync, &rdev->flags) &&
9151 !test_bit(Journal, &rdev->flags))) &&
9152 atomic_read(&rdev->nr_pending)==0)) {
9153 if (mddev->pers->hot_remove_disk(
9154 mddev, rdev) == 0) {
9155 sysfs_unlink_rdev(mddev, rdev);
9156 rdev->saved_raid_disk = rdev->raid_disk;
9157 rdev->raid_disk = -1;
9161 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9162 clear_bit(RemoveSynchronized, &rdev->flags);
9165 if (removed && mddev->kobj.sd)
9166 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9168 if (this && removed)
9171 rdev_for_each(rdev, mddev) {
9172 if (this && this != rdev)
9174 if (test_bit(Candidate, &rdev->flags))
9176 if (rdev->raid_disk >= 0 &&
9177 !test_bit(In_sync, &rdev->flags) &&
9178 !test_bit(Journal, &rdev->flags) &&
9179 !test_bit(Faulty, &rdev->flags))
9181 if (rdev->raid_disk >= 0)
9183 if (test_bit(Faulty, &rdev->flags))
9185 if (!test_bit(Journal, &rdev->flags)) {
9187 ! (rdev->saved_raid_disk >= 0 &&
9188 !test_bit(Bitmap_sync, &rdev->flags)))
9191 rdev->recovery_offset = 0;
9193 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9194 /* failure here is OK */
9195 sysfs_link_rdev(mddev, rdev);
9196 if (!test_bit(Journal, &rdev->flags))
9198 md_new_event(mddev);
9199 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9204 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9208 static void md_start_sync(struct work_struct *ws)
9210 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9212 mddev->sync_thread = md_register_thread(md_do_sync,
9215 if (!mddev->sync_thread) {
9216 pr_warn("%s: could not start resync thread...\n",
9218 /* leave the spares where they are, it shouldn't hurt */
9219 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9220 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9221 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9222 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9223 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9224 wake_up(&resync_wait);
9225 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9227 if (mddev->sysfs_action)
9228 sysfs_notify_dirent_safe(mddev->sysfs_action);
9230 md_wakeup_thread(mddev->sync_thread);
9231 sysfs_notify_dirent_safe(mddev->sysfs_action);
9232 md_new_event(mddev);
9236 * This routine is regularly called by all per-raid-array threads to
9237 * deal with generic issues like resync and super-block update.
9238 * Raid personalities that don't have a thread (linear/raid0) do not
9239 * need this as they never do any recovery or update the superblock.
9241 * It does not do any resync itself, but rather "forks" off other threads
9242 * to do that as needed.
9243 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9244 * "->recovery" and create a thread at ->sync_thread.
9245 * When the thread finishes it sets MD_RECOVERY_DONE
9246 * and wakeups up this thread which will reap the thread and finish up.
9247 * This thread also removes any faulty devices (with nr_pending == 0).
9249 * The overall approach is:
9250 * 1/ if the superblock needs updating, update it.
9251 * 2/ If a recovery thread is running, don't do anything else.
9252 * 3/ If recovery has finished, clean up, possibly marking spares active.
9253 * 4/ If there are any faulty devices, remove them.
9254 * 5/ If array is degraded, try to add spares devices
9255 * 6/ If array has spares or is not in-sync, start a resync thread.
9257 void md_check_recovery(struct mddev *mddev)
9259 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9260 /* Write superblock - thread that called mddev_suspend()
9261 * holds reconfig_mutex for us.
9263 set_bit(MD_UPDATING_SB, &mddev->flags);
9264 smp_mb__after_atomic();
9265 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9266 md_update_sb(mddev, 0);
9267 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9268 wake_up(&mddev->sb_wait);
9271 if (mddev->suspended)
9275 md_bitmap_daemon_work(mddev);
9277 if (signal_pending(current)) {
9278 if (mddev->pers->sync_request && !mddev->external) {
9279 pr_debug("md: %s in immediate safe mode\n",
9281 mddev->safemode = 2;
9283 flush_signals(current);
9286 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9289 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9290 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9291 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9292 (mddev->external == 0 && mddev->safemode == 1) ||
9293 (mddev->safemode == 2
9294 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9298 if (mddev_trylock(mddev)) {
9300 bool try_set_sync = mddev->safemode != 0;
9302 if (!mddev->external && mddev->safemode == 1)
9303 mddev->safemode = 0;
9306 struct md_rdev *rdev;
9307 if (!mddev->external && mddev->in_sync)
9308 /* 'Blocked' flag not needed as failed devices
9309 * will be recorded if array switched to read/write.
9310 * Leaving it set will prevent the device
9311 * from being removed.
9313 rdev_for_each(rdev, mddev)
9314 clear_bit(Blocked, &rdev->flags);
9315 /* On a read-only array we can:
9316 * - remove failed devices
9317 * - add already-in_sync devices if the array itself
9319 * As we only add devices that are already in-sync,
9320 * we can activate the spares immediately.
9322 remove_and_add_spares(mddev, NULL);
9323 /* There is no thread, but we need to call
9324 * ->spare_active and clear saved_raid_disk
9326 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9327 md_reap_sync_thread(mddev);
9328 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9329 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9330 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9334 if (mddev_is_clustered(mddev)) {
9335 struct md_rdev *rdev, *tmp;
9336 /* kick the device if another node issued a
9339 rdev_for_each_safe(rdev, tmp, mddev) {
9340 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9341 rdev->raid_disk < 0)
9342 md_kick_rdev_from_array(rdev);
9346 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9347 spin_lock(&mddev->lock);
9349 spin_unlock(&mddev->lock);
9352 if (mddev->sb_flags)
9353 md_update_sb(mddev, 0);
9355 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9356 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9357 /* resync/recovery still happening */
9358 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9361 if (mddev->sync_thread) {
9362 md_reap_sync_thread(mddev);
9365 /* Set RUNNING before clearing NEEDED to avoid
9366 * any transients in the value of "sync_action".
9368 mddev->curr_resync_completed = 0;
9369 spin_lock(&mddev->lock);
9370 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9371 spin_unlock(&mddev->lock);
9372 /* Clear some bits that don't mean anything, but
9375 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9376 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9378 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9379 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9381 /* no recovery is running.
9382 * remove any failed drives, then
9383 * add spares if possible.
9384 * Spares are also removed and re-added, to allow
9385 * the personality to fail the re-add.
9388 if (mddev->reshape_position != MaxSector) {
9389 if (mddev->pers->check_reshape == NULL ||
9390 mddev->pers->check_reshape(mddev) != 0)
9391 /* Cannot proceed */
9393 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9394 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9395 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9396 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9397 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9398 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9399 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9400 } else if (mddev->recovery_cp < MaxSector) {
9401 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9402 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9403 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9404 /* nothing to be done ... */
9407 if (mddev->pers->sync_request) {
9409 /* We are adding a device or devices to an array
9410 * which has the bitmap stored on all devices.
9411 * So make sure all bitmap pages get written
9413 md_bitmap_write_all(mddev->bitmap);
9415 INIT_WORK(&mddev->del_work, md_start_sync);
9416 queue_work(md_misc_wq, &mddev->del_work);
9420 if (!mddev->sync_thread) {
9421 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9422 wake_up(&resync_wait);
9423 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9425 if (mddev->sysfs_action)
9426 sysfs_notify_dirent_safe(mddev->sysfs_action);
9429 wake_up(&mddev->sb_wait);
9430 mddev_unlock(mddev);
9433 EXPORT_SYMBOL(md_check_recovery);
9435 void md_reap_sync_thread(struct mddev *mddev)
9437 struct md_rdev *rdev;
9438 sector_t old_dev_sectors = mddev->dev_sectors;
9439 bool is_reshaped = false;
9441 /* resync has finished, collect result */
9442 md_unregister_thread(&mddev->sync_thread);
9443 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9444 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9445 mddev->degraded != mddev->raid_disks) {
9447 /* activate any spares */
9448 if (mddev->pers->spare_active(mddev)) {
9449 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9450 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9453 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9454 mddev->pers->finish_reshape) {
9455 mddev->pers->finish_reshape(mddev);
9456 if (mddev_is_clustered(mddev))
9460 /* If array is no-longer degraded, then any saved_raid_disk
9461 * information must be scrapped.
9463 if (!mddev->degraded)
9464 rdev_for_each(rdev, mddev)
9465 rdev->saved_raid_disk = -1;
9467 md_update_sb(mddev, 1);
9468 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9469 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9471 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9472 md_cluster_ops->resync_finish(mddev);
9473 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9474 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9475 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9476 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9477 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9478 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9480 * We call md_cluster_ops->update_size here because sync_size could
9481 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9482 * so it is time to update size across cluster.
9484 if (mddev_is_clustered(mddev) && is_reshaped
9485 && !test_bit(MD_CLOSING, &mddev->flags))
9486 md_cluster_ops->update_size(mddev, old_dev_sectors);
9487 wake_up(&resync_wait);
9488 /* flag recovery needed just to double check */
9489 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9490 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9491 sysfs_notify_dirent_safe(mddev->sysfs_action);
9492 md_new_event(mddev);
9493 if (mddev->event_work.func)
9494 queue_work(md_misc_wq, &mddev->event_work);
9496 EXPORT_SYMBOL(md_reap_sync_thread);
9498 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9500 sysfs_notify_dirent_safe(rdev->sysfs_state);
9501 wait_event_timeout(rdev->blocked_wait,
9502 !test_bit(Blocked, &rdev->flags) &&
9503 !test_bit(BlockedBadBlocks, &rdev->flags),
9504 msecs_to_jiffies(5000));
9505 rdev_dec_pending(rdev, mddev);
9507 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9509 void md_finish_reshape(struct mddev *mddev)
9511 /* called be personality module when reshape completes. */
9512 struct md_rdev *rdev;
9514 rdev_for_each(rdev, mddev) {
9515 if (rdev->data_offset > rdev->new_data_offset)
9516 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9518 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9519 rdev->data_offset = rdev->new_data_offset;
9522 EXPORT_SYMBOL(md_finish_reshape);
9524 /* Bad block management */
9526 /* Returns 1 on success, 0 on failure */
9527 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9530 struct mddev *mddev = rdev->mddev;
9533 s += rdev->new_data_offset;
9535 s += rdev->data_offset;
9536 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9538 /* Make sure they get written out promptly */
9539 if (test_bit(ExternalBbl, &rdev->flags))
9540 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9541 sysfs_notify_dirent_safe(rdev->sysfs_state);
9542 set_mask_bits(&mddev->sb_flags, 0,
9543 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9544 md_wakeup_thread(rdev->mddev->thread);
9549 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9551 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9556 s += rdev->new_data_offset;
9558 s += rdev->data_offset;
9559 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9560 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9561 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9564 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9566 static int md_notify_reboot(struct notifier_block *this,
9567 unsigned long code, void *x)
9569 struct list_head *tmp;
9570 struct mddev *mddev;
9573 for_each_mddev(mddev, tmp) {
9574 if (mddev_trylock(mddev)) {
9576 __md_stop_writes(mddev);
9577 if (mddev->persistent)
9578 mddev->safemode = 2;
9579 mddev_unlock(mddev);
9584 * certain more exotic SCSI devices are known to be
9585 * volatile wrt too early system reboots. While the
9586 * right place to handle this issue is the given
9587 * driver, we do want to have a safe RAID driver ...
9595 static struct notifier_block md_notifier = {
9596 .notifier_call = md_notify_reboot,
9598 .priority = INT_MAX, /* before any real devices */
9601 static void md_geninit(void)
9603 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9605 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9608 static int __init md_init(void)
9612 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9616 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9620 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9621 if (!md_rdev_misc_wq)
9622 goto err_rdev_misc_wq;
9624 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9627 if ((ret = register_blkdev(0, "mdp")) < 0)
9631 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9632 md_probe, NULL, NULL);
9633 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9634 md_probe, NULL, NULL);
9636 register_reboot_notifier(&md_notifier);
9637 raid_table_header = register_sysctl_table(raid_root_table);
9643 unregister_blkdev(MD_MAJOR, "md");
9645 destroy_workqueue(md_rdev_misc_wq);
9647 destroy_workqueue(md_misc_wq);
9649 destroy_workqueue(md_wq);
9654 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9656 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9657 struct md_rdev *rdev2, *tmp;
9659 char b[BDEVNAME_SIZE];
9662 * If size is changed in another node then we need to
9663 * do resize as well.
9665 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9666 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9668 pr_info("md-cluster: resize failed\n");
9670 md_bitmap_update_sb(mddev->bitmap);
9673 /* Check for change of roles in the active devices */
9674 rdev_for_each_safe(rdev2, tmp, mddev) {
9675 if (test_bit(Faulty, &rdev2->flags))
9678 /* Check if the roles changed */
9679 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9681 if (test_bit(Candidate, &rdev2->flags)) {
9682 if (role == 0xfffe) {
9683 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9684 md_kick_rdev_from_array(rdev2);
9688 clear_bit(Candidate, &rdev2->flags);
9691 if (role != rdev2->raid_disk) {
9693 * got activated except reshape is happening.
9695 if (rdev2->raid_disk == -1 && role != 0xffff &&
9696 !(le32_to_cpu(sb->feature_map) &
9697 MD_FEATURE_RESHAPE_ACTIVE)) {
9698 rdev2->saved_raid_disk = role;
9699 ret = remove_and_add_spares(mddev, rdev2);
9700 pr_info("Activated spare: %s\n",
9701 bdevname(rdev2->bdev,b));
9702 /* wakeup mddev->thread here, so array could
9703 * perform resync with the new activated disk */
9704 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9705 md_wakeup_thread(mddev->thread);
9708 * We just want to do the minimum to mark the disk
9709 * as faulty. The recovery is performed by the
9710 * one who initiated the error.
9712 if ((role == 0xfffe) || (role == 0xfffd)) {
9713 md_error(mddev, rdev2);
9714 clear_bit(Blocked, &rdev2->flags);
9719 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9720 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9722 pr_warn("md: updating array disks failed. %d\n", ret);
9726 * Since mddev->delta_disks has already updated in update_raid_disks,
9727 * so it is time to check reshape.
9729 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9730 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9732 * reshape is happening in the remote node, we need to
9733 * update reshape_position and call start_reshape.
9735 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9736 if (mddev->pers->update_reshape_pos)
9737 mddev->pers->update_reshape_pos(mddev);
9738 if (mddev->pers->start_reshape)
9739 mddev->pers->start_reshape(mddev);
9740 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9741 mddev->reshape_position != MaxSector &&
9742 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9743 /* reshape is just done in another node. */
9744 mddev->reshape_position = MaxSector;
9745 if (mddev->pers->update_reshape_pos)
9746 mddev->pers->update_reshape_pos(mddev);
9749 /* Finally set the event to be up to date */
9750 mddev->events = le64_to_cpu(sb->events);
9753 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9756 struct page *swapout = rdev->sb_page;
9757 struct mdp_superblock_1 *sb;
9759 /* Store the sb page of the rdev in the swapout temporary
9760 * variable in case we err in the future
9762 rdev->sb_page = NULL;
9763 err = alloc_disk_sb(rdev);
9765 ClearPageUptodate(rdev->sb_page);
9766 rdev->sb_loaded = 0;
9767 err = super_types[mddev->major_version].
9768 load_super(rdev, NULL, mddev->minor_version);
9771 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9772 __func__, __LINE__, rdev->desc_nr, err);
9774 put_page(rdev->sb_page);
9775 rdev->sb_page = swapout;
9776 rdev->sb_loaded = 1;
9780 sb = page_address(rdev->sb_page);
9781 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9785 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9786 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9788 /* The other node finished recovery, call spare_active to set
9789 * device In_sync and mddev->degraded
9791 if (rdev->recovery_offset == MaxSector &&
9792 !test_bit(In_sync, &rdev->flags) &&
9793 mddev->pers->spare_active(mddev))
9794 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9800 void md_reload_sb(struct mddev *mddev, int nr)
9802 struct md_rdev *rdev = NULL, *iter;
9806 rdev_for_each_rcu(iter, mddev) {
9807 if (iter->desc_nr == nr) {
9814 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9818 err = read_rdev(mddev, rdev);
9822 check_sb_changes(mddev, rdev);
9824 /* Read all rdev's to update recovery_offset */
9825 rdev_for_each_rcu(rdev, mddev) {
9826 if (!test_bit(Faulty, &rdev->flags))
9827 read_rdev(mddev, rdev);
9830 EXPORT_SYMBOL(md_reload_sb);
9835 * Searches all registered partitions for autorun RAID arrays
9839 static DEFINE_MUTEX(detected_devices_mutex);
9840 static LIST_HEAD(all_detected_devices);
9841 struct detected_devices_node {
9842 struct list_head list;
9846 void md_autodetect_dev(dev_t dev)
9848 struct detected_devices_node *node_detected_dev;
9850 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9851 if (node_detected_dev) {
9852 node_detected_dev->dev = dev;
9853 mutex_lock(&detected_devices_mutex);
9854 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9855 mutex_unlock(&detected_devices_mutex);
9859 void md_autostart_arrays(int part)
9861 struct md_rdev *rdev;
9862 struct detected_devices_node *node_detected_dev;
9864 int i_scanned, i_passed;
9869 pr_info("md: Autodetecting RAID arrays.\n");
9871 mutex_lock(&detected_devices_mutex);
9872 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9874 node_detected_dev = list_entry(all_detected_devices.next,
9875 struct detected_devices_node, list);
9876 list_del(&node_detected_dev->list);
9877 dev = node_detected_dev->dev;
9878 kfree(node_detected_dev);
9879 mutex_unlock(&detected_devices_mutex);
9880 rdev = md_import_device(dev,0, 90);
9881 mutex_lock(&detected_devices_mutex);
9885 if (test_bit(Faulty, &rdev->flags))
9888 set_bit(AutoDetected, &rdev->flags);
9889 list_add(&rdev->same_set, &pending_raid_disks);
9892 mutex_unlock(&detected_devices_mutex);
9894 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9896 autorun_devices(part);
9899 #endif /* !MODULE */
9901 static __exit void md_exit(void)
9903 struct mddev *mddev;
9904 struct list_head *tmp;
9907 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9908 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9910 unregister_blkdev(MD_MAJOR,"md");
9911 unregister_blkdev(mdp_major, "mdp");
9912 unregister_reboot_notifier(&md_notifier);
9913 unregister_sysctl_table(raid_table_header);
9915 /* We cannot unload the modules while some process is
9916 * waiting for us in select() or poll() - wake them up
9919 while (waitqueue_active(&md_event_waiters)) {
9920 /* not safe to leave yet */
9921 wake_up(&md_event_waiters);
9925 remove_proc_entry("mdstat", NULL);
9927 for_each_mddev(mddev, tmp) {
9928 export_array(mddev);
9930 mddev->hold_active = 0;
9932 * for_each_mddev() will call mddev_put() at the end of each
9933 * iteration. As the mddev is now fully clear, this will
9934 * schedule the mddev for destruction by a workqueue, and the
9935 * destroy_workqueue() below will wait for that to complete.
9938 destroy_workqueue(md_rdev_misc_wq);
9939 destroy_workqueue(md_misc_wq);
9940 destroy_workqueue(md_wq);
9943 subsys_initcall(md_init);
9944 module_exit(md_exit)
9946 static int get_ro(char *buffer, const struct kernel_param *kp)
9948 return sprintf(buffer, "%d\n", start_readonly);
9950 static int set_ro(const char *val, const struct kernel_param *kp)
9952 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9955 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9956 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9957 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9958 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9960 MODULE_LICENSE("GPL");
9961 MODULE_DESCRIPTION("MD RAID framework");
9963 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);