1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/blk-integrity.h>
45 #include <linux/badblocks.h>
46 #include <linux/sysctl.h>
47 #include <linux/seq_file.h>
49 #include <linux/poll.h>
50 #include <linux/ctype.h>
51 #include <linux/string.h>
52 #include <linux/hdreg.h>
53 #include <linux/proc_fs.h>
54 #include <linux/random.h>
55 #include <linux/major.h>
56 #include <linux/module.h>
57 #include <linux/reboot.h>
58 #include <linux/file.h>
59 #include <linux/compat.h>
60 #include <linux/delay.h>
61 #include <linux/raid/md_p.h>
62 #include <linux/raid/md_u.h>
63 #include <linux/raid/detect.h>
64 #include <linux/slab.h>
65 #include <linux/percpu-refcount.h>
66 #include <linux/part_stat.h>
68 #include <trace/events/block.h>
70 #include "md-bitmap.h"
71 #include "md-cluster.h"
73 /* pers_list is a list of registered personalities protected
75 * pers_lock does extra service to protect accesses to
76 * mddev->thread when the mutex cannot be held.
78 static LIST_HEAD(pers_list);
79 static DEFINE_SPINLOCK(pers_lock);
81 static const struct kobj_type md_ktype;
83 struct md_cluster_operations *md_cluster_ops;
84 EXPORT_SYMBOL(md_cluster_ops);
85 static struct module *md_cluster_mod;
87 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
88 static struct workqueue_struct *md_wq;
89 static struct workqueue_struct *md_misc_wq;
90 static struct workqueue_struct *md_rdev_misc_wq;
92 static int remove_and_add_spares(struct mddev *mddev,
93 struct md_rdev *this);
94 static void mddev_detach(struct mddev *mddev);
103 static bool md_is_rdwr(struct mddev *mddev)
105 return (mddev->ro == MD_RDWR);
109 * Default number of read corrections we'll attempt on an rdev
110 * before ejecting it from the array. We divide the read error
111 * count by 2 for every hour elapsed between read errors.
113 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
114 /* Default safemode delay: 200 msec */
115 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
117 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
118 * is 1000 KB/sec, so the extra system load does not show up that much.
119 * Increase it if you want to have more _guaranteed_ speed. Note that
120 * the RAID driver will use the maximum available bandwidth if the IO
121 * subsystem is idle. There is also an 'absolute maximum' reconstruction
122 * speed limit - in case reconstruction slows down your system despite
125 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
126 * or /sys/block/mdX/md/sync_speed_{min,max}
129 static int sysctl_speed_limit_min = 1000;
130 static int sysctl_speed_limit_max = 200000;
131 static inline int speed_min(struct mddev *mddev)
133 return mddev->sync_speed_min ?
134 mddev->sync_speed_min : sysctl_speed_limit_min;
137 static inline int speed_max(struct mddev *mddev)
139 return mddev->sync_speed_max ?
140 mddev->sync_speed_max : sysctl_speed_limit_max;
143 static void rdev_uninit_serial(struct md_rdev *rdev)
145 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
148 kvfree(rdev->serial);
152 static void rdevs_uninit_serial(struct mddev *mddev)
154 struct md_rdev *rdev;
156 rdev_for_each(rdev, mddev)
157 rdev_uninit_serial(rdev);
160 static int rdev_init_serial(struct md_rdev *rdev)
162 /* serial_nums equals with BARRIER_BUCKETS_NR */
163 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
164 struct serial_in_rdev *serial = NULL;
166 if (test_bit(CollisionCheck, &rdev->flags))
169 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
174 for (i = 0; i < serial_nums; i++) {
175 struct serial_in_rdev *serial_tmp = &serial[i];
177 spin_lock_init(&serial_tmp->serial_lock);
178 serial_tmp->serial_rb = RB_ROOT_CACHED;
179 init_waitqueue_head(&serial_tmp->serial_io_wait);
182 rdev->serial = serial;
183 set_bit(CollisionCheck, &rdev->flags);
188 static int rdevs_init_serial(struct mddev *mddev)
190 struct md_rdev *rdev;
193 rdev_for_each(rdev, mddev) {
194 ret = rdev_init_serial(rdev);
199 /* Free all resources if pool is not existed */
200 if (ret && !mddev->serial_info_pool)
201 rdevs_uninit_serial(mddev);
207 * rdev needs to enable serial stuffs if it meets the conditions:
208 * 1. it is multi-queue device flaged with writemostly.
209 * 2. the write-behind mode is enabled.
211 static int rdev_need_serial(struct md_rdev *rdev)
213 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
214 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
215 test_bit(WriteMostly, &rdev->flags));
219 * Init resource for rdev(s), then create serial_info_pool if:
220 * 1. rdev is the first device which return true from rdev_enable_serial.
221 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
223 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
228 if (rdev && !rdev_need_serial(rdev) &&
229 !test_bit(CollisionCheck, &rdev->flags))
233 mddev_suspend(mddev);
236 ret = rdevs_init_serial(mddev);
238 ret = rdev_init_serial(rdev);
242 if (mddev->serial_info_pool == NULL) {
244 * already in memalloc noio context by
247 mddev->serial_info_pool =
248 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
249 sizeof(struct serial_info));
250 if (!mddev->serial_info_pool) {
251 rdevs_uninit_serial(mddev);
252 pr_err("can't alloc memory pool for serialization\n");
262 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
263 * 1. rdev is the last device flaged with CollisionCheck.
264 * 2. when bitmap is destroyed while policy is not enabled.
265 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
267 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
270 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
273 if (mddev->serial_info_pool) {
274 struct md_rdev *temp;
275 int num = 0; /* used to track if other rdevs need the pool */
278 mddev_suspend(mddev);
279 rdev_for_each(temp, mddev) {
281 if (!mddev->serialize_policy ||
282 !rdev_need_serial(temp))
283 rdev_uninit_serial(temp);
286 } else if (temp != rdev &&
287 test_bit(CollisionCheck, &temp->flags))
292 rdev_uninit_serial(rdev);
295 pr_info("The mempool could be used by other devices\n");
297 mempool_destroy(mddev->serial_info_pool);
298 mddev->serial_info_pool = NULL;
305 static struct ctl_table_header *raid_table_header;
307 static struct ctl_table raid_table[] = {
309 .procname = "speed_limit_min",
310 .data = &sysctl_speed_limit_min,
311 .maxlen = sizeof(int),
312 .mode = S_IRUGO|S_IWUSR,
313 .proc_handler = proc_dointvec,
316 .procname = "speed_limit_max",
317 .data = &sysctl_speed_limit_max,
318 .maxlen = sizeof(int),
319 .mode = S_IRUGO|S_IWUSR,
320 .proc_handler = proc_dointvec,
325 static int start_readonly;
328 * The original mechanism for creating an md device is to create
329 * a device node in /dev and to open it. This causes races with device-close.
330 * The preferred method is to write to the "new_array" module parameter.
331 * This can avoid races.
332 * Setting create_on_open to false disables the original mechanism
333 * so all the races disappear.
335 static bool create_on_open = true;
338 * We have a system wide 'event count' that is incremented
339 * on any 'interesting' event, and readers of /proc/mdstat
340 * can use 'poll' or 'select' to find out when the event
344 * start array, stop array, error, add device, remove device,
345 * start build, activate spare
347 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
348 static atomic_t md_event_count;
349 void md_new_event(void)
351 atomic_inc(&md_event_count);
352 wake_up(&md_event_waiters);
354 EXPORT_SYMBOL_GPL(md_new_event);
357 * Enables to iterate over all existing md arrays
358 * all_mddevs_lock protects this list.
360 static LIST_HEAD(all_mddevs);
361 static DEFINE_SPINLOCK(all_mddevs_lock);
363 static bool is_md_suspended(struct mddev *mddev)
365 return percpu_ref_is_dying(&mddev->active_io);
367 /* Rather than calling directly into the personality make_request function,
368 * IO requests come here first so that we can check if the device is
369 * being suspended pending a reconfiguration.
370 * We hold a refcount over the call to ->make_request. By the time that
371 * call has finished, the bio has been linked into some internal structure
372 * and so is visible to ->quiesce(), so we don't need the refcount any more.
374 static bool is_suspended(struct mddev *mddev, struct bio *bio)
376 if (is_md_suspended(mddev))
378 if (bio_data_dir(bio) != WRITE)
380 if (mddev->suspend_lo >= mddev->suspend_hi)
382 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
384 if (bio_end_sector(bio) < mddev->suspend_lo)
389 void md_handle_request(struct mddev *mddev, struct bio *bio)
392 if (is_suspended(mddev, bio)) {
394 /* Bail out if REQ_NOWAIT is set for the bio */
395 if (bio->bi_opf & REQ_NOWAIT) {
396 bio_wouldblock_error(bio);
400 prepare_to_wait(&mddev->sb_wait, &__wait,
401 TASK_UNINTERRUPTIBLE);
402 if (!is_suspended(mddev, bio))
406 finish_wait(&mddev->sb_wait, &__wait);
408 if (!percpu_ref_tryget_live(&mddev->active_io))
409 goto check_suspended;
411 if (!mddev->pers->make_request(mddev, bio)) {
412 percpu_ref_put(&mddev->active_io);
413 goto check_suspended;
416 percpu_ref_put(&mddev->active_io);
418 EXPORT_SYMBOL(md_handle_request);
420 static void md_submit_bio(struct bio *bio)
422 const int rw = bio_data_dir(bio);
423 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
425 if (mddev == NULL || mddev->pers == NULL) {
430 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
435 bio = bio_split_to_limits(bio);
439 if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
440 if (bio_sectors(bio) != 0)
441 bio->bi_status = BLK_STS_IOERR;
446 /* bio could be mergeable after passing to underlayer */
447 bio->bi_opf &= ~REQ_NOMERGE;
449 md_handle_request(mddev, bio);
452 /* mddev_suspend makes sure no new requests are submitted
453 * to the device, and that any requests that have been submitted
454 * are completely handled.
455 * Once mddev_detach() is called and completes, the module will be
458 void mddev_suspend(struct mddev *mddev)
460 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
461 lockdep_assert_held(&mddev->reconfig_mutex);
462 if (mddev->suspended++)
464 wake_up(&mddev->sb_wait);
465 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
466 percpu_ref_kill(&mddev->active_io);
467 wait_event(mddev->sb_wait, percpu_ref_is_zero(&mddev->active_io));
468 mddev->pers->quiesce(mddev, 1);
469 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
470 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
472 del_timer_sync(&mddev->safemode_timer);
473 /* restrict memory reclaim I/O during raid array is suspend */
474 mddev->noio_flag = memalloc_noio_save();
476 EXPORT_SYMBOL_GPL(mddev_suspend);
478 void mddev_resume(struct mddev *mddev)
480 /* entred the memalloc scope from mddev_suspend() */
481 memalloc_noio_restore(mddev->noio_flag);
482 lockdep_assert_held(&mddev->reconfig_mutex);
483 if (--mddev->suspended)
485 percpu_ref_resurrect(&mddev->active_io);
486 wake_up(&mddev->sb_wait);
487 mddev->pers->quiesce(mddev, 0);
489 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
490 md_wakeup_thread(mddev->thread);
491 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
493 EXPORT_SYMBOL_GPL(mddev_resume);
496 * Generic flush handling for md
499 static void md_end_flush(struct bio *bio)
501 struct md_rdev *rdev = bio->bi_private;
502 struct mddev *mddev = rdev->mddev;
506 rdev_dec_pending(rdev, mddev);
508 if (atomic_dec_and_test(&mddev->flush_pending)) {
509 /* The pre-request flush has finished */
510 queue_work(md_wq, &mddev->flush_work);
514 static void md_submit_flush_data(struct work_struct *ws);
516 static void submit_flushes(struct work_struct *ws)
518 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
519 struct md_rdev *rdev;
521 mddev->start_flush = ktime_get_boottime();
522 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
523 atomic_set(&mddev->flush_pending, 1);
525 rdev_for_each_rcu(rdev, mddev)
526 if (rdev->raid_disk >= 0 &&
527 !test_bit(Faulty, &rdev->flags)) {
528 /* Take two references, one is dropped
529 * when request finishes, one after
530 * we reclaim rcu_read_lock
533 atomic_inc(&rdev->nr_pending);
534 atomic_inc(&rdev->nr_pending);
536 bi = bio_alloc_bioset(rdev->bdev, 0,
537 REQ_OP_WRITE | REQ_PREFLUSH,
538 GFP_NOIO, &mddev->bio_set);
539 bi->bi_end_io = md_end_flush;
540 bi->bi_private = rdev;
541 atomic_inc(&mddev->flush_pending);
544 rdev_dec_pending(rdev, mddev);
547 if (atomic_dec_and_test(&mddev->flush_pending))
548 queue_work(md_wq, &mddev->flush_work);
551 static void md_submit_flush_data(struct work_struct *ws)
553 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
554 struct bio *bio = mddev->flush_bio;
557 * must reset flush_bio before calling into md_handle_request to avoid a
558 * deadlock, because other bios passed md_handle_request suspend check
559 * could wait for this and below md_handle_request could wait for those
560 * bios because of suspend check
562 spin_lock_irq(&mddev->lock);
563 mddev->prev_flush_start = mddev->start_flush;
564 mddev->flush_bio = NULL;
565 spin_unlock_irq(&mddev->lock);
566 wake_up(&mddev->sb_wait);
568 if (bio->bi_iter.bi_size == 0) {
569 /* an empty barrier - all done */
572 bio->bi_opf &= ~REQ_PREFLUSH;
573 md_handle_request(mddev, bio);
578 * Manages consolidation of flushes and submitting any flushes needed for
579 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
580 * being finished in another context. Returns false if the flushing is
581 * complete but still needs the I/O portion of the bio to be processed.
583 bool md_flush_request(struct mddev *mddev, struct bio *bio)
585 ktime_t req_start = ktime_get_boottime();
586 spin_lock_irq(&mddev->lock);
587 /* flush requests wait until ongoing flush completes,
588 * hence coalescing all the pending requests.
590 wait_event_lock_irq(mddev->sb_wait,
592 ktime_before(req_start, mddev->prev_flush_start),
594 /* new request after previous flush is completed */
595 if (ktime_after(req_start, mddev->prev_flush_start)) {
596 WARN_ON(mddev->flush_bio);
597 mddev->flush_bio = bio;
600 spin_unlock_irq(&mddev->lock);
603 INIT_WORK(&mddev->flush_work, submit_flushes);
604 queue_work(md_wq, &mddev->flush_work);
606 /* flush was performed for some other bio while we waited. */
607 if (bio->bi_iter.bi_size == 0)
608 /* an empty barrier - all done */
611 bio->bi_opf &= ~REQ_PREFLUSH;
617 EXPORT_SYMBOL(md_flush_request);
619 static inline struct mddev *mddev_get(struct mddev *mddev)
621 lockdep_assert_held(&all_mddevs_lock);
623 if (test_bit(MD_DELETED, &mddev->flags))
625 atomic_inc(&mddev->active);
629 static void mddev_delayed_delete(struct work_struct *ws);
631 void mddev_put(struct mddev *mddev)
633 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
635 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
636 mddev->ctime == 0 && !mddev->hold_active) {
637 /* Array is not configured at all, and not held active,
639 set_bit(MD_DELETED, &mddev->flags);
642 * Call queue_work inside the spinlock so that
643 * flush_workqueue() after mddev_find will succeed in waiting
644 * for the work to be done.
646 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
647 queue_work(md_misc_wq, &mddev->del_work);
649 spin_unlock(&all_mddevs_lock);
652 static void md_safemode_timeout(struct timer_list *t);
654 void mddev_init(struct mddev *mddev)
656 mutex_init(&mddev->open_mutex);
657 mutex_init(&mddev->reconfig_mutex);
658 mutex_init(&mddev->bitmap_info.mutex);
659 INIT_LIST_HEAD(&mddev->disks);
660 INIT_LIST_HEAD(&mddev->all_mddevs);
661 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
662 atomic_set(&mddev->active, 1);
663 atomic_set(&mddev->openers, 0);
664 spin_lock_init(&mddev->lock);
665 atomic_set(&mddev->flush_pending, 0);
666 init_waitqueue_head(&mddev->sb_wait);
667 init_waitqueue_head(&mddev->recovery_wait);
668 mddev->reshape_position = MaxSector;
669 mddev->reshape_backwards = 0;
670 mddev->last_sync_action = "none";
671 mddev->resync_min = 0;
672 mddev->resync_max = MaxSector;
673 mddev->level = LEVEL_NONE;
675 EXPORT_SYMBOL_GPL(mddev_init);
677 static struct mddev *mddev_find_locked(dev_t unit)
681 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
682 if (mddev->unit == unit)
688 /* find an unused unit number */
689 static dev_t mddev_alloc_unit(void)
691 static int next_minor = 512;
692 int start = next_minor;
697 dev = MKDEV(MD_MAJOR, next_minor);
699 if (next_minor > MINORMASK)
701 if (next_minor == start)
702 return 0; /* Oh dear, all in use. */
703 is_free = !mddev_find_locked(dev);
709 static struct mddev *mddev_alloc(dev_t unit)
714 if (unit && MAJOR(unit) != MD_MAJOR)
715 unit &= ~((1 << MdpMinorShift) - 1);
717 new = kzalloc(sizeof(*new), GFP_KERNEL);
719 return ERR_PTR(-ENOMEM);
722 spin_lock(&all_mddevs_lock);
725 if (mddev_find_locked(unit))
728 if (MAJOR(unit) == MD_MAJOR)
729 new->md_minor = MINOR(unit);
731 new->md_minor = MINOR(unit) >> MdpMinorShift;
732 new->hold_active = UNTIL_IOCTL;
735 new->unit = mddev_alloc_unit();
738 new->md_minor = MINOR(new->unit);
739 new->hold_active = UNTIL_STOP;
742 list_add(&new->all_mddevs, &all_mddevs);
743 spin_unlock(&all_mddevs_lock);
746 spin_unlock(&all_mddevs_lock);
748 return ERR_PTR(error);
751 static void mddev_free(struct mddev *mddev)
753 spin_lock(&all_mddevs_lock);
754 list_del(&mddev->all_mddevs);
755 spin_unlock(&all_mddevs_lock);
760 static const struct attribute_group md_redundancy_group;
762 void mddev_unlock(struct mddev *mddev)
764 if (mddev->to_remove) {
765 /* These cannot be removed under reconfig_mutex as
766 * an access to the files will try to take reconfig_mutex
767 * while holding the file unremovable, which leads to
769 * So hold set sysfs_active while the remove in happeing,
770 * and anything else which might set ->to_remove or my
771 * otherwise change the sysfs namespace will fail with
772 * -EBUSY if sysfs_active is still set.
773 * We set sysfs_active under reconfig_mutex and elsewhere
774 * test it under the same mutex to ensure its correct value
777 const struct attribute_group *to_remove = mddev->to_remove;
778 mddev->to_remove = NULL;
779 mddev->sysfs_active = 1;
780 mutex_unlock(&mddev->reconfig_mutex);
782 if (mddev->kobj.sd) {
783 if (to_remove != &md_redundancy_group)
784 sysfs_remove_group(&mddev->kobj, to_remove);
785 if (mddev->pers == NULL ||
786 mddev->pers->sync_request == NULL) {
787 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
788 if (mddev->sysfs_action)
789 sysfs_put(mddev->sysfs_action);
790 if (mddev->sysfs_completed)
791 sysfs_put(mddev->sysfs_completed);
792 if (mddev->sysfs_degraded)
793 sysfs_put(mddev->sysfs_degraded);
794 mddev->sysfs_action = NULL;
795 mddev->sysfs_completed = NULL;
796 mddev->sysfs_degraded = NULL;
799 mddev->sysfs_active = 0;
801 mutex_unlock(&mddev->reconfig_mutex);
803 /* As we've dropped the mutex we need a spinlock to
804 * make sure the thread doesn't disappear
806 spin_lock(&pers_lock);
807 md_wakeup_thread(mddev->thread);
808 wake_up(&mddev->sb_wait);
809 spin_unlock(&pers_lock);
811 EXPORT_SYMBOL_GPL(mddev_unlock);
813 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
815 struct md_rdev *rdev;
817 rdev_for_each_rcu(rdev, mddev)
818 if (rdev->desc_nr == nr)
823 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
825 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
827 struct md_rdev *rdev;
829 rdev_for_each(rdev, mddev)
830 if (rdev->bdev->bd_dev == dev)
836 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
838 struct md_rdev *rdev;
840 rdev_for_each_rcu(rdev, mddev)
841 if (rdev->bdev->bd_dev == dev)
846 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
848 static struct md_personality *find_pers(int level, char *clevel)
850 struct md_personality *pers;
851 list_for_each_entry(pers, &pers_list, list) {
852 if (level != LEVEL_NONE && pers->level == level)
854 if (strcmp(pers->name, clevel)==0)
860 /* return the offset of the super block in 512byte sectors */
861 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
863 return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
866 static int alloc_disk_sb(struct md_rdev *rdev)
868 rdev->sb_page = alloc_page(GFP_KERNEL);
874 void md_rdev_clear(struct md_rdev *rdev)
877 put_page(rdev->sb_page);
879 rdev->sb_page = NULL;
884 put_page(rdev->bb_page);
885 rdev->bb_page = NULL;
887 badblocks_exit(&rdev->badblocks);
889 EXPORT_SYMBOL_GPL(md_rdev_clear);
891 static void super_written(struct bio *bio)
893 struct md_rdev *rdev = bio->bi_private;
894 struct mddev *mddev = rdev->mddev;
896 if (bio->bi_status) {
897 pr_err("md: %s gets error=%d\n", __func__,
898 blk_status_to_errno(bio->bi_status));
899 md_error(mddev, rdev);
900 if (!test_bit(Faulty, &rdev->flags)
901 && (bio->bi_opf & MD_FAILFAST)) {
902 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
903 set_bit(LastDev, &rdev->flags);
906 clear_bit(LastDev, &rdev->flags);
910 rdev_dec_pending(rdev, mddev);
912 if (atomic_dec_and_test(&mddev->pending_writes))
913 wake_up(&mddev->sb_wait);
916 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
917 sector_t sector, int size, struct page *page)
919 /* write first size bytes of page to sector of rdev
920 * Increment mddev->pending_writes before returning
921 * and decrement it on completion, waking up sb_wait
922 * if zero is reached.
923 * If an error occurred, call md_error
930 if (test_bit(Faulty, &rdev->flags))
933 bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
935 REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA,
936 GFP_NOIO, &mddev->sync_set);
938 atomic_inc(&rdev->nr_pending);
940 bio->bi_iter.bi_sector = sector;
941 bio_add_page(bio, page, size, 0);
942 bio->bi_private = rdev;
943 bio->bi_end_io = super_written;
945 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
946 test_bit(FailFast, &rdev->flags) &&
947 !test_bit(LastDev, &rdev->flags))
948 bio->bi_opf |= MD_FAILFAST;
950 atomic_inc(&mddev->pending_writes);
954 int md_super_wait(struct mddev *mddev)
956 /* wait for all superblock writes that were scheduled to complete */
957 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
958 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
963 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
964 struct page *page, blk_opf_t opf, bool metadata_op)
969 if (metadata_op && rdev->meta_bdev)
970 bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
972 bio_init(&bio, rdev->bdev, &bvec, 1, opf);
975 bio.bi_iter.bi_sector = sector + rdev->sb_start;
976 else if (rdev->mddev->reshape_position != MaxSector &&
977 (rdev->mddev->reshape_backwards ==
978 (sector >= rdev->mddev->reshape_position)))
979 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
981 bio.bi_iter.bi_sector = sector + rdev->data_offset;
982 bio_add_page(&bio, page, size, 0);
984 submit_bio_wait(&bio);
986 return !bio.bi_status;
988 EXPORT_SYMBOL_GPL(sync_page_io);
990 static int read_disk_sb(struct md_rdev *rdev, int size)
995 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1001 pr_err("md: disabled device %pg, could not read superblock.\n",
1006 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1008 return sb1->set_uuid0 == sb2->set_uuid0 &&
1009 sb1->set_uuid1 == sb2->set_uuid1 &&
1010 sb1->set_uuid2 == sb2->set_uuid2 &&
1011 sb1->set_uuid3 == sb2->set_uuid3;
1014 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1017 mdp_super_t *tmp1, *tmp2;
1019 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1020 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1022 if (!tmp1 || !tmp2) {
1031 * nr_disks is not constant
1036 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1043 static u32 md_csum_fold(u32 csum)
1045 csum = (csum & 0xffff) + (csum >> 16);
1046 return (csum & 0xffff) + (csum >> 16);
1049 static unsigned int calc_sb_csum(mdp_super_t *sb)
1052 u32 *sb32 = (u32*)sb;
1054 unsigned int disk_csum, csum;
1056 disk_csum = sb->sb_csum;
1059 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1061 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1064 /* This used to use csum_partial, which was wrong for several
1065 * reasons including that different results are returned on
1066 * different architectures. It isn't critical that we get exactly
1067 * the same return value as before (we always csum_fold before
1068 * testing, and that removes any differences). However as we
1069 * know that csum_partial always returned a 16bit value on
1070 * alphas, do a fold to maximise conformity to previous behaviour.
1072 sb->sb_csum = md_csum_fold(disk_csum);
1074 sb->sb_csum = disk_csum;
1080 * Handle superblock details.
1081 * We want to be able to handle multiple superblock formats
1082 * so we have a common interface to them all, and an array of
1083 * different handlers.
1084 * We rely on user-space to write the initial superblock, and support
1085 * reading and updating of superblocks.
1086 * Interface methods are:
1087 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1088 * loads and validates a superblock on dev.
1089 * if refdev != NULL, compare superblocks on both devices
1091 * 0 - dev has a superblock that is compatible with refdev
1092 * 1 - dev has a superblock that is compatible and newer than refdev
1093 * so dev should be used as the refdev in future
1094 * -EINVAL superblock incompatible or invalid
1095 * -othererror e.g. -EIO
1097 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1098 * Verify that dev is acceptable into mddev.
1099 * The first time, mddev->raid_disks will be 0, and data from
1100 * dev should be merged in. Subsequent calls check that dev
1101 * is new enough. Return 0 or -EINVAL
1103 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1104 * Update the superblock for rdev with data in mddev
1105 * This does not write to disc.
1111 struct module *owner;
1112 int (*load_super)(struct md_rdev *rdev,
1113 struct md_rdev *refdev,
1115 int (*validate_super)(struct mddev *mddev,
1116 struct md_rdev *rdev);
1117 void (*sync_super)(struct mddev *mddev,
1118 struct md_rdev *rdev);
1119 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1120 sector_t num_sectors);
1121 int (*allow_new_offset)(struct md_rdev *rdev,
1122 unsigned long long new_offset);
1126 * Check that the given mddev has no bitmap.
1128 * This function is called from the run method of all personalities that do not
1129 * support bitmaps. It prints an error message and returns non-zero if mddev
1130 * has a bitmap. Otherwise, it returns 0.
1133 int md_check_no_bitmap(struct mddev *mddev)
1135 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1137 pr_warn("%s: bitmaps are not supported for %s\n",
1138 mdname(mddev), mddev->pers->name);
1141 EXPORT_SYMBOL(md_check_no_bitmap);
1144 * load_super for 0.90.0
1146 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1150 bool spare_disk = true;
1153 * Calculate the position of the superblock (512byte sectors),
1154 * it's at the end of the disk.
1156 * It also happens to be a multiple of 4Kb.
1158 rdev->sb_start = calc_dev_sboffset(rdev);
1160 ret = read_disk_sb(rdev, MD_SB_BYTES);
1166 sb = page_address(rdev->sb_page);
1168 if (sb->md_magic != MD_SB_MAGIC) {
1169 pr_warn("md: invalid raid superblock magic on %pg\n",
1174 if (sb->major_version != 0 ||
1175 sb->minor_version < 90 ||
1176 sb->minor_version > 91) {
1177 pr_warn("Bad version number %d.%d on %pg\n",
1178 sb->major_version, sb->minor_version, rdev->bdev);
1182 if (sb->raid_disks <= 0)
1185 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1186 pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1190 rdev->preferred_minor = sb->md_minor;
1191 rdev->data_offset = 0;
1192 rdev->new_data_offset = 0;
1193 rdev->sb_size = MD_SB_BYTES;
1194 rdev->badblocks.shift = -1;
1196 if (sb->level == LEVEL_MULTIPATH)
1199 rdev->desc_nr = sb->this_disk.number;
1201 /* not spare disk, or LEVEL_MULTIPATH */
1202 if (sb->level == LEVEL_MULTIPATH ||
1203 (rdev->desc_nr >= 0 &&
1204 rdev->desc_nr < MD_SB_DISKS &&
1205 sb->disks[rdev->desc_nr].state &
1206 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1216 mdp_super_t *refsb = page_address(refdev->sb_page);
1217 if (!md_uuid_equal(refsb, sb)) {
1218 pr_warn("md: %pg has different UUID to %pg\n",
1219 rdev->bdev, refdev->bdev);
1222 if (!md_sb_equal(refsb, sb)) {
1223 pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1224 rdev->bdev, refdev->bdev);
1228 ev2 = md_event(refsb);
1230 if (!spare_disk && ev1 > ev2)
1235 rdev->sectors = rdev->sb_start;
1236 /* Limit to 4TB as metadata cannot record more than that.
1237 * (not needed for Linear and RAID0 as metadata doesn't
1240 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1241 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1243 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1244 /* "this cannot possibly happen" ... */
1252 * validate_super for 0.90.0
1254 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1257 mdp_super_t *sb = page_address(rdev->sb_page);
1258 __u64 ev1 = md_event(sb);
1260 rdev->raid_disk = -1;
1261 clear_bit(Faulty, &rdev->flags);
1262 clear_bit(In_sync, &rdev->flags);
1263 clear_bit(Bitmap_sync, &rdev->flags);
1264 clear_bit(WriteMostly, &rdev->flags);
1266 if (mddev->raid_disks == 0) {
1267 mddev->major_version = 0;
1268 mddev->minor_version = sb->minor_version;
1269 mddev->patch_version = sb->patch_version;
1270 mddev->external = 0;
1271 mddev->chunk_sectors = sb->chunk_size >> 9;
1272 mddev->ctime = sb->ctime;
1273 mddev->utime = sb->utime;
1274 mddev->level = sb->level;
1275 mddev->clevel[0] = 0;
1276 mddev->layout = sb->layout;
1277 mddev->raid_disks = sb->raid_disks;
1278 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1279 mddev->events = ev1;
1280 mddev->bitmap_info.offset = 0;
1281 mddev->bitmap_info.space = 0;
1282 /* bitmap can use 60 K after the 4K superblocks */
1283 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1284 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1285 mddev->reshape_backwards = 0;
1287 if (mddev->minor_version >= 91) {
1288 mddev->reshape_position = sb->reshape_position;
1289 mddev->delta_disks = sb->delta_disks;
1290 mddev->new_level = sb->new_level;
1291 mddev->new_layout = sb->new_layout;
1292 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1293 if (mddev->delta_disks < 0)
1294 mddev->reshape_backwards = 1;
1296 mddev->reshape_position = MaxSector;
1297 mddev->delta_disks = 0;
1298 mddev->new_level = mddev->level;
1299 mddev->new_layout = mddev->layout;
1300 mddev->new_chunk_sectors = mddev->chunk_sectors;
1302 if (mddev->level == 0)
1305 if (sb->state & (1<<MD_SB_CLEAN))
1306 mddev->recovery_cp = MaxSector;
1308 if (sb->events_hi == sb->cp_events_hi &&
1309 sb->events_lo == sb->cp_events_lo) {
1310 mddev->recovery_cp = sb->recovery_cp;
1312 mddev->recovery_cp = 0;
1315 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1316 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1317 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1318 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1320 mddev->max_disks = MD_SB_DISKS;
1322 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1323 mddev->bitmap_info.file == NULL) {
1324 mddev->bitmap_info.offset =
1325 mddev->bitmap_info.default_offset;
1326 mddev->bitmap_info.space =
1327 mddev->bitmap_info.default_space;
1330 } else if (mddev->pers == NULL) {
1331 /* Insist on good event counter while assembling, except
1332 * for spares (which don't need an event count) */
1334 if (sb->disks[rdev->desc_nr].state & (
1335 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1336 if (ev1 < mddev->events)
1338 } else if (mddev->bitmap) {
1339 /* if adding to array with a bitmap, then we can accept an
1340 * older device ... but not too old.
1342 if (ev1 < mddev->bitmap->events_cleared)
1344 if (ev1 < mddev->events)
1345 set_bit(Bitmap_sync, &rdev->flags);
1347 if (ev1 < mddev->events)
1348 /* just a hot-add of a new device, leave raid_disk at -1 */
1352 if (mddev->level != LEVEL_MULTIPATH) {
1353 desc = sb->disks + rdev->desc_nr;
1355 if (desc->state & (1<<MD_DISK_FAULTY))
1356 set_bit(Faulty, &rdev->flags);
1357 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1358 desc->raid_disk < mddev->raid_disks */) {
1359 set_bit(In_sync, &rdev->flags);
1360 rdev->raid_disk = desc->raid_disk;
1361 rdev->saved_raid_disk = desc->raid_disk;
1362 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1363 /* active but not in sync implies recovery up to
1364 * reshape position. We don't know exactly where
1365 * that is, so set to zero for now */
1366 if (mddev->minor_version >= 91) {
1367 rdev->recovery_offset = 0;
1368 rdev->raid_disk = desc->raid_disk;
1371 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1372 set_bit(WriteMostly, &rdev->flags);
1373 if (desc->state & (1<<MD_DISK_FAILFAST))
1374 set_bit(FailFast, &rdev->flags);
1375 } else /* MULTIPATH are always insync */
1376 set_bit(In_sync, &rdev->flags);
1381 * sync_super for 0.90.0
1383 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1386 struct md_rdev *rdev2;
1387 int next_spare = mddev->raid_disks;
1389 /* make rdev->sb match mddev data..
1392 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1393 * 3/ any empty disks < next_spare become removed
1395 * disks[0] gets initialised to REMOVED because
1396 * we cannot be sure from other fields if it has
1397 * been initialised or not.
1400 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1402 rdev->sb_size = MD_SB_BYTES;
1404 sb = page_address(rdev->sb_page);
1406 memset(sb, 0, sizeof(*sb));
1408 sb->md_magic = MD_SB_MAGIC;
1409 sb->major_version = mddev->major_version;
1410 sb->patch_version = mddev->patch_version;
1411 sb->gvalid_words = 0; /* ignored */
1412 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1413 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1414 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1415 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1417 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1418 sb->level = mddev->level;
1419 sb->size = mddev->dev_sectors / 2;
1420 sb->raid_disks = mddev->raid_disks;
1421 sb->md_minor = mddev->md_minor;
1422 sb->not_persistent = 0;
1423 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1425 sb->events_hi = (mddev->events>>32);
1426 sb->events_lo = (u32)mddev->events;
1428 if (mddev->reshape_position == MaxSector)
1429 sb->minor_version = 90;
1431 sb->minor_version = 91;
1432 sb->reshape_position = mddev->reshape_position;
1433 sb->new_level = mddev->new_level;
1434 sb->delta_disks = mddev->delta_disks;
1435 sb->new_layout = mddev->new_layout;
1436 sb->new_chunk = mddev->new_chunk_sectors << 9;
1438 mddev->minor_version = sb->minor_version;
1441 sb->recovery_cp = mddev->recovery_cp;
1442 sb->cp_events_hi = (mddev->events>>32);
1443 sb->cp_events_lo = (u32)mddev->events;
1444 if (mddev->recovery_cp == MaxSector)
1445 sb->state = (1<< MD_SB_CLEAN);
1447 sb->recovery_cp = 0;
1449 sb->layout = mddev->layout;
1450 sb->chunk_size = mddev->chunk_sectors << 9;
1452 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1453 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1455 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1456 rdev_for_each(rdev2, mddev) {
1459 int is_active = test_bit(In_sync, &rdev2->flags);
1461 if (rdev2->raid_disk >= 0 &&
1462 sb->minor_version >= 91)
1463 /* we have nowhere to store the recovery_offset,
1464 * but if it is not below the reshape_position,
1465 * we can piggy-back on that.
1468 if (rdev2->raid_disk < 0 ||
1469 test_bit(Faulty, &rdev2->flags))
1472 desc_nr = rdev2->raid_disk;
1474 desc_nr = next_spare++;
1475 rdev2->desc_nr = desc_nr;
1476 d = &sb->disks[rdev2->desc_nr];
1478 d->number = rdev2->desc_nr;
1479 d->major = MAJOR(rdev2->bdev->bd_dev);
1480 d->minor = MINOR(rdev2->bdev->bd_dev);
1482 d->raid_disk = rdev2->raid_disk;
1484 d->raid_disk = rdev2->desc_nr; /* compatibility */
1485 if (test_bit(Faulty, &rdev2->flags))
1486 d->state = (1<<MD_DISK_FAULTY);
1487 else if (is_active) {
1488 d->state = (1<<MD_DISK_ACTIVE);
1489 if (test_bit(In_sync, &rdev2->flags))
1490 d->state |= (1<<MD_DISK_SYNC);
1498 if (test_bit(WriteMostly, &rdev2->flags))
1499 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1500 if (test_bit(FailFast, &rdev2->flags))
1501 d->state |= (1<<MD_DISK_FAILFAST);
1503 /* now set the "removed" and "faulty" bits on any missing devices */
1504 for (i=0 ; i < mddev->raid_disks ; i++) {
1505 mdp_disk_t *d = &sb->disks[i];
1506 if (d->state == 0 && d->number == 0) {
1509 d->state = (1<<MD_DISK_REMOVED);
1510 d->state |= (1<<MD_DISK_FAULTY);
1514 sb->nr_disks = nr_disks;
1515 sb->active_disks = active;
1516 sb->working_disks = working;
1517 sb->failed_disks = failed;
1518 sb->spare_disks = spare;
1520 sb->this_disk = sb->disks[rdev->desc_nr];
1521 sb->sb_csum = calc_sb_csum(sb);
1525 * rdev_size_change for 0.90.0
1527 static unsigned long long
1528 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1530 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1531 return 0; /* component must fit device */
1532 if (rdev->mddev->bitmap_info.offset)
1533 return 0; /* can't move bitmap */
1534 rdev->sb_start = calc_dev_sboffset(rdev);
1535 if (!num_sectors || num_sectors > rdev->sb_start)
1536 num_sectors = rdev->sb_start;
1537 /* Limit to 4TB as metadata cannot record more than that.
1538 * 4TB == 2^32 KB, or 2*2^32 sectors.
1540 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1541 num_sectors = (sector_t)(2ULL << 32) - 2;
1543 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1545 } while (md_super_wait(rdev->mddev) < 0);
1550 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1552 /* non-zero offset changes not possible with v0.90 */
1553 return new_offset == 0;
1557 * version 1 superblock
1560 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1564 unsigned long long newcsum;
1565 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1566 __le32 *isuper = (__le32*)sb;
1568 disk_csum = sb->sb_csum;
1571 for (; size >= 4; size -= 4)
1572 newcsum += le32_to_cpu(*isuper++);
1575 newcsum += le16_to_cpu(*(__le16*) isuper);
1577 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1578 sb->sb_csum = disk_csum;
1579 return cpu_to_le32(csum);
1582 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1584 struct mdp_superblock_1 *sb;
1589 bool spare_disk = true;
1592 * Calculate the position of the superblock in 512byte sectors.
1593 * It is always aligned to a 4K boundary and
1594 * depeding on minor_version, it can be:
1595 * 0: At least 8K, but less than 12K, from end of device
1596 * 1: At start of device
1597 * 2: 4K from start of device.
1599 switch(minor_version) {
1601 sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1602 sb_start &= ~(sector_t)(4*2-1);
1613 rdev->sb_start = sb_start;
1615 /* superblock is rarely larger than 1K, but it can be larger,
1616 * and it is safe to read 4k, so we do that
1618 ret = read_disk_sb(rdev, 4096);
1619 if (ret) return ret;
1621 sb = page_address(rdev->sb_page);
1623 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1624 sb->major_version != cpu_to_le32(1) ||
1625 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1626 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1627 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1630 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1631 pr_warn("md: invalid superblock checksum on %pg\n",
1635 if (le64_to_cpu(sb->data_size) < 10) {
1636 pr_warn("md: data_size too small on %pg\n",
1642 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1643 /* Some padding is non-zero, might be a new feature */
1646 rdev->preferred_minor = 0xffff;
1647 rdev->data_offset = le64_to_cpu(sb->data_offset);
1648 rdev->new_data_offset = rdev->data_offset;
1649 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1650 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1651 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1652 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1654 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1655 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1656 if (rdev->sb_size & bmask)
1657 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1660 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1663 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1666 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1669 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1671 if (!rdev->bb_page) {
1672 rdev->bb_page = alloc_page(GFP_KERNEL);
1676 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1677 rdev->badblocks.count == 0) {
1678 /* need to load the bad block list.
1679 * Currently we limit it to one page.
1685 int sectors = le16_to_cpu(sb->bblog_size);
1686 if (sectors > (PAGE_SIZE / 512))
1688 offset = le32_to_cpu(sb->bblog_offset);
1691 bb_sector = (long long)offset;
1692 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1693 rdev->bb_page, REQ_OP_READ, true))
1695 bbp = (__le64 *)page_address(rdev->bb_page);
1696 rdev->badblocks.shift = sb->bblog_shift;
1697 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1698 u64 bb = le64_to_cpu(*bbp);
1699 int count = bb & (0x3ff);
1700 u64 sector = bb >> 10;
1701 sector <<= sb->bblog_shift;
1702 count <<= sb->bblog_shift;
1705 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1708 } else if (sb->bblog_offset != 0)
1709 rdev->badblocks.shift = 0;
1711 if ((le32_to_cpu(sb->feature_map) &
1712 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1713 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1714 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1715 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1718 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1722 /* not spare disk, or LEVEL_MULTIPATH */
1723 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1724 (rdev->desc_nr >= 0 &&
1725 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1726 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1727 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1737 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1739 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1740 sb->level != refsb->level ||
1741 sb->layout != refsb->layout ||
1742 sb->chunksize != refsb->chunksize) {
1743 pr_warn("md: %pg has strangely different superblock to %pg\n",
1748 ev1 = le64_to_cpu(sb->events);
1749 ev2 = le64_to_cpu(refsb->events);
1751 if (!spare_disk && ev1 > ev2)
1757 sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1759 sectors = rdev->sb_start;
1760 if (sectors < le64_to_cpu(sb->data_size))
1762 rdev->sectors = le64_to_cpu(sb->data_size);
1766 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1768 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1769 __u64 ev1 = le64_to_cpu(sb->events);
1771 rdev->raid_disk = -1;
1772 clear_bit(Faulty, &rdev->flags);
1773 clear_bit(In_sync, &rdev->flags);
1774 clear_bit(Bitmap_sync, &rdev->flags);
1775 clear_bit(WriteMostly, &rdev->flags);
1777 if (mddev->raid_disks == 0) {
1778 mddev->major_version = 1;
1779 mddev->patch_version = 0;
1780 mddev->external = 0;
1781 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1782 mddev->ctime = le64_to_cpu(sb->ctime);
1783 mddev->utime = le64_to_cpu(sb->utime);
1784 mddev->level = le32_to_cpu(sb->level);
1785 mddev->clevel[0] = 0;
1786 mddev->layout = le32_to_cpu(sb->layout);
1787 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1788 mddev->dev_sectors = le64_to_cpu(sb->size);
1789 mddev->events = ev1;
1790 mddev->bitmap_info.offset = 0;
1791 mddev->bitmap_info.space = 0;
1792 /* Default location for bitmap is 1K after superblock
1793 * using 3K - total of 4K
1795 mddev->bitmap_info.default_offset = 1024 >> 9;
1796 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1797 mddev->reshape_backwards = 0;
1799 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1800 memcpy(mddev->uuid, sb->set_uuid, 16);
1802 mddev->max_disks = (4096-256)/2;
1804 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1805 mddev->bitmap_info.file == NULL) {
1806 mddev->bitmap_info.offset =
1807 (__s32)le32_to_cpu(sb->bitmap_offset);
1808 /* Metadata doesn't record how much space is available.
1809 * For 1.0, we assume we can use up to the superblock
1810 * if before, else to 4K beyond superblock.
1811 * For others, assume no change is possible.
1813 if (mddev->minor_version > 0)
1814 mddev->bitmap_info.space = 0;
1815 else if (mddev->bitmap_info.offset > 0)
1816 mddev->bitmap_info.space =
1817 8 - mddev->bitmap_info.offset;
1819 mddev->bitmap_info.space =
1820 -mddev->bitmap_info.offset;
1823 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1824 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1825 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1826 mddev->new_level = le32_to_cpu(sb->new_level);
1827 mddev->new_layout = le32_to_cpu(sb->new_layout);
1828 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1829 if (mddev->delta_disks < 0 ||
1830 (mddev->delta_disks == 0 &&
1831 (le32_to_cpu(sb->feature_map)
1832 & MD_FEATURE_RESHAPE_BACKWARDS)))
1833 mddev->reshape_backwards = 1;
1835 mddev->reshape_position = MaxSector;
1836 mddev->delta_disks = 0;
1837 mddev->new_level = mddev->level;
1838 mddev->new_layout = mddev->layout;
1839 mddev->new_chunk_sectors = mddev->chunk_sectors;
1842 if (mddev->level == 0 &&
1843 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1846 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1847 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1849 if (le32_to_cpu(sb->feature_map) &
1850 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1851 if (le32_to_cpu(sb->feature_map) &
1852 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1854 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1855 (le32_to_cpu(sb->feature_map) &
1856 MD_FEATURE_MULTIPLE_PPLS))
1858 set_bit(MD_HAS_PPL, &mddev->flags);
1860 } else if (mddev->pers == NULL) {
1861 /* Insist of good event counter while assembling, except for
1862 * spares (which don't need an event count) */
1864 if (rdev->desc_nr >= 0 &&
1865 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1866 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1867 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1868 if (ev1 < mddev->events)
1870 } else if (mddev->bitmap) {
1871 /* If adding to array with a bitmap, then we can accept an
1872 * older device, but not too old.
1874 if (ev1 < mddev->bitmap->events_cleared)
1876 if (ev1 < mddev->events)
1877 set_bit(Bitmap_sync, &rdev->flags);
1879 if (ev1 < mddev->events)
1880 /* just a hot-add of a new device, leave raid_disk at -1 */
1883 if (mddev->level != LEVEL_MULTIPATH) {
1885 if (rdev->desc_nr < 0 ||
1886 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1887 role = MD_DISK_ROLE_SPARE;
1890 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1892 case MD_DISK_ROLE_SPARE: /* spare */
1894 case MD_DISK_ROLE_FAULTY: /* faulty */
1895 set_bit(Faulty, &rdev->flags);
1897 case MD_DISK_ROLE_JOURNAL: /* journal device */
1898 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1899 /* journal device without journal feature */
1900 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1903 set_bit(Journal, &rdev->flags);
1904 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1905 rdev->raid_disk = 0;
1908 rdev->saved_raid_disk = role;
1909 if ((le32_to_cpu(sb->feature_map) &
1910 MD_FEATURE_RECOVERY_OFFSET)) {
1911 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1912 if (!(le32_to_cpu(sb->feature_map) &
1913 MD_FEATURE_RECOVERY_BITMAP))
1914 rdev->saved_raid_disk = -1;
1917 * If the array is FROZEN, then the device can't
1918 * be in_sync with rest of array.
1920 if (!test_bit(MD_RECOVERY_FROZEN,
1922 set_bit(In_sync, &rdev->flags);
1924 rdev->raid_disk = role;
1927 if (sb->devflags & WriteMostly1)
1928 set_bit(WriteMostly, &rdev->flags);
1929 if (sb->devflags & FailFast1)
1930 set_bit(FailFast, &rdev->flags);
1931 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1932 set_bit(Replacement, &rdev->flags);
1933 } else /* MULTIPATH are always insync */
1934 set_bit(In_sync, &rdev->flags);
1939 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1941 struct mdp_superblock_1 *sb;
1942 struct md_rdev *rdev2;
1944 /* make rdev->sb match mddev and rdev data. */
1946 sb = page_address(rdev->sb_page);
1948 sb->feature_map = 0;
1950 sb->recovery_offset = cpu_to_le64(0);
1951 memset(sb->pad3, 0, sizeof(sb->pad3));
1953 sb->utime = cpu_to_le64((__u64)mddev->utime);
1954 sb->events = cpu_to_le64(mddev->events);
1956 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1957 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1958 sb->resync_offset = cpu_to_le64(MaxSector);
1960 sb->resync_offset = cpu_to_le64(0);
1962 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1964 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1965 sb->size = cpu_to_le64(mddev->dev_sectors);
1966 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1967 sb->level = cpu_to_le32(mddev->level);
1968 sb->layout = cpu_to_le32(mddev->layout);
1969 if (test_bit(FailFast, &rdev->flags))
1970 sb->devflags |= FailFast1;
1972 sb->devflags &= ~FailFast1;
1974 if (test_bit(WriteMostly, &rdev->flags))
1975 sb->devflags |= WriteMostly1;
1977 sb->devflags &= ~WriteMostly1;
1978 sb->data_offset = cpu_to_le64(rdev->data_offset);
1979 sb->data_size = cpu_to_le64(rdev->sectors);
1981 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1982 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1983 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1986 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1987 !test_bit(In_sync, &rdev->flags)) {
1989 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1990 sb->recovery_offset =
1991 cpu_to_le64(rdev->recovery_offset);
1992 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1994 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1996 /* Note: recovery_offset and journal_tail share space */
1997 if (test_bit(Journal, &rdev->flags))
1998 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1999 if (test_bit(Replacement, &rdev->flags))
2001 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2003 if (mddev->reshape_position != MaxSector) {
2004 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2005 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2006 sb->new_layout = cpu_to_le32(mddev->new_layout);
2007 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2008 sb->new_level = cpu_to_le32(mddev->new_level);
2009 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2010 if (mddev->delta_disks == 0 &&
2011 mddev->reshape_backwards)
2013 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2014 if (rdev->new_data_offset != rdev->data_offset) {
2016 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2017 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2018 - rdev->data_offset));
2022 if (mddev_is_clustered(mddev))
2023 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2025 if (rdev->badblocks.count == 0)
2026 /* Nothing to do for bad blocks*/ ;
2027 else if (sb->bblog_offset == 0)
2028 /* Cannot record bad blocks on this device */
2029 md_error(mddev, rdev);
2031 struct badblocks *bb = &rdev->badblocks;
2032 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2034 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2039 seq = read_seqbegin(&bb->lock);
2041 memset(bbp, 0xff, PAGE_SIZE);
2043 for (i = 0 ; i < bb->count ; i++) {
2044 u64 internal_bb = p[i];
2045 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2046 | BB_LEN(internal_bb));
2047 bbp[i] = cpu_to_le64(store_bb);
2050 if (read_seqretry(&bb->lock, seq))
2053 bb->sector = (rdev->sb_start +
2054 (int)le32_to_cpu(sb->bblog_offset));
2055 bb->size = le16_to_cpu(sb->bblog_size);
2060 rdev_for_each(rdev2, mddev)
2061 if (rdev2->desc_nr+1 > max_dev)
2062 max_dev = rdev2->desc_nr+1;
2064 if (max_dev > le32_to_cpu(sb->max_dev)) {
2066 sb->max_dev = cpu_to_le32(max_dev);
2067 rdev->sb_size = max_dev * 2 + 256;
2068 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2069 if (rdev->sb_size & bmask)
2070 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2072 max_dev = le32_to_cpu(sb->max_dev);
2074 for (i=0; i<max_dev;i++)
2075 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2077 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2078 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2080 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2081 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2083 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2085 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2086 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2087 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2090 rdev_for_each(rdev2, mddev) {
2092 if (test_bit(Faulty, &rdev2->flags))
2093 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2094 else if (test_bit(In_sync, &rdev2->flags))
2095 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2096 else if (test_bit(Journal, &rdev2->flags))
2097 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2098 else if (rdev2->raid_disk >= 0)
2099 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2101 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2104 sb->sb_csum = calc_sb_1_csum(sb);
2107 static sector_t super_1_choose_bm_space(sector_t dev_size)
2111 /* if the device is bigger than 8Gig, save 64k for bitmap
2112 * usage, if bigger than 200Gig, save 128k
2114 if (dev_size < 64*2)
2116 else if (dev_size - 64*2 >= 200*1024*1024*2)
2118 else if (dev_size - 4*2 > 8*1024*1024*2)
2125 static unsigned long long
2126 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2128 struct mdp_superblock_1 *sb;
2129 sector_t max_sectors;
2130 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2131 return 0; /* component must fit device */
2132 if (rdev->data_offset != rdev->new_data_offset)
2133 return 0; /* too confusing */
2134 if (rdev->sb_start < rdev->data_offset) {
2135 /* minor versions 1 and 2; superblock before data */
2136 max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2137 if (!num_sectors || num_sectors > max_sectors)
2138 num_sectors = max_sectors;
2139 } else if (rdev->mddev->bitmap_info.offset) {
2140 /* minor version 0 with bitmap we can't move */
2143 /* minor version 0; superblock after data */
2144 sector_t sb_start, bm_space;
2145 sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2147 /* 8K is for superblock */
2148 sb_start = dev_size - 8*2;
2149 sb_start &= ~(sector_t)(4*2 - 1);
2151 bm_space = super_1_choose_bm_space(dev_size);
2153 /* Space that can be used to store date needs to decrease
2154 * superblock bitmap space and bad block space(4K)
2156 max_sectors = sb_start - bm_space - 4*2;
2158 if (!num_sectors || num_sectors > max_sectors)
2159 num_sectors = max_sectors;
2160 rdev->sb_start = sb_start;
2162 sb = page_address(rdev->sb_page);
2163 sb->data_size = cpu_to_le64(num_sectors);
2164 sb->super_offset = cpu_to_le64(rdev->sb_start);
2165 sb->sb_csum = calc_sb_1_csum(sb);
2167 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2169 } while (md_super_wait(rdev->mddev) < 0);
2175 super_1_allow_new_offset(struct md_rdev *rdev,
2176 unsigned long long new_offset)
2178 /* All necessary checks on new >= old have been done */
2179 struct bitmap *bitmap;
2180 if (new_offset >= rdev->data_offset)
2183 /* with 1.0 metadata, there is no metadata to tread on
2184 * so we can always move back */
2185 if (rdev->mddev->minor_version == 0)
2188 /* otherwise we must be sure not to step on
2189 * any metadata, so stay:
2190 * 36K beyond start of superblock
2191 * beyond end of badblocks
2192 * beyond write-intent bitmap
2194 if (rdev->sb_start + (32+4)*2 > new_offset)
2196 bitmap = rdev->mddev->bitmap;
2197 if (bitmap && !rdev->mddev->bitmap_info.file &&
2198 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2199 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2201 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2207 static struct super_type super_types[] = {
2210 .owner = THIS_MODULE,
2211 .load_super = super_90_load,
2212 .validate_super = super_90_validate,
2213 .sync_super = super_90_sync,
2214 .rdev_size_change = super_90_rdev_size_change,
2215 .allow_new_offset = super_90_allow_new_offset,
2219 .owner = THIS_MODULE,
2220 .load_super = super_1_load,
2221 .validate_super = super_1_validate,
2222 .sync_super = super_1_sync,
2223 .rdev_size_change = super_1_rdev_size_change,
2224 .allow_new_offset = super_1_allow_new_offset,
2228 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2230 if (mddev->sync_super) {
2231 mddev->sync_super(mddev, rdev);
2235 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2237 super_types[mddev->major_version].sync_super(mddev, rdev);
2240 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2242 struct md_rdev *rdev, *rdev2;
2245 rdev_for_each_rcu(rdev, mddev1) {
2246 if (test_bit(Faulty, &rdev->flags) ||
2247 test_bit(Journal, &rdev->flags) ||
2248 rdev->raid_disk == -1)
2250 rdev_for_each_rcu(rdev2, mddev2) {
2251 if (test_bit(Faulty, &rdev2->flags) ||
2252 test_bit(Journal, &rdev2->flags) ||
2253 rdev2->raid_disk == -1)
2255 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2265 static LIST_HEAD(pending_raid_disks);
2268 * Try to register data integrity profile for an mddev
2270 * This is called when an array is started and after a disk has been kicked
2271 * from the array. It only succeeds if all working and active component devices
2272 * are integrity capable with matching profiles.
2274 int md_integrity_register(struct mddev *mddev)
2276 struct md_rdev *rdev, *reference = NULL;
2278 if (list_empty(&mddev->disks))
2279 return 0; /* nothing to do */
2280 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2281 return 0; /* shouldn't register, or already is */
2282 rdev_for_each(rdev, mddev) {
2283 /* skip spares and non-functional disks */
2284 if (test_bit(Faulty, &rdev->flags))
2286 if (rdev->raid_disk < 0)
2289 /* Use the first rdev as the reference */
2293 /* does this rdev's profile match the reference profile? */
2294 if (blk_integrity_compare(reference->bdev->bd_disk,
2295 rdev->bdev->bd_disk) < 0)
2298 if (!reference || !bdev_get_integrity(reference->bdev))
2301 * All component devices are integrity capable and have matching
2302 * profiles, register the common profile for the md device.
2304 blk_integrity_register(mddev->gendisk,
2305 bdev_get_integrity(reference->bdev));
2307 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2308 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2309 (mddev->level != 1 && mddev->level != 10 &&
2310 bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2312 * No need to handle the failure of bioset_integrity_create,
2313 * because the function is called by md_run() -> pers->run(),
2314 * md_run calls bioset_exit -> bioset_integrity_free in case
2317 pr_err("md: failed to create integrity pool for %s\n",
2323 EXPORT_SYMBOL(md_integrity_register);
2326 * Attempt to add an rdev, but only if it is consistent with the current
2329 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2331 struct blk_integrity *bi_mddev;
2333 if (!mddev->gendisk)
2336 bi_mddev = blk_get_integrity(mddev->gendisk);
2338 if (!bi_mddev) /* nothing to do */
2341 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2342 pr_err("%s: incompatible integrity profile for %pg\n",
2343 mdname(mddev), rdev->bdev);
2349 EXPORT_SYMBOL(md_integrity_add_rdev);
2351 static bool rdev_read_only(struct md_rdev *rdev)
2353 return bdev_read_only(rdev->bdev) ||
2354 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2357 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2359 char b[BDEVNAME_SIZE];
2362 /* prevent duplicates */
2363 if (find_rdev(mddev, rdev->bdev->bd_dev))
2366 if (rdev_read_only(rdev) && mddev->pers)
2369 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2370 if (!test_bit(Journal, &rdev->flags) &&
2372 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2374 /* Cannot change size, so fail
2375 * If mddev->level <= 0, then we don't care
2376 * about aligning sizes (e.g. linear)
2378 if (mddev->level > 0)
2381 mddev->dev_sectors = rdev->sectors;
2384 /* Verify rdev->desc_nr is unique.
2385 * If it is -1, assign a free number, else
2386 * check number is not in use
2389 if (rdev->desc_nr < 0) {
2392 choice = mddev->raid_disks;
2393 while (md_find_rdev_nr_rcu(mddev, choice))
2395 rdev->desc_nr = choice;
2397 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2403 if (!test_bit(Journal, &rdev->flags) &&
2404 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2405 pr_warn("md: %s: array is limited to %d devices\n",
2406 mdname(mddev), mddev->max_disks);
2409 snprintf(b, sizeof(b), "%pg", rdev->bdev);
2410 strreplace(b, '/', '!');
2412 rdev->mddev = mddev;
2413 pr_debug("md: bind<%s>\n", b);
2415 if (mddev->raid_disks)
2416 mddev_create_serial_pool(mddev, rdev, false);
2418 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2421 /* failure here is OK */
2422 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2423 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2424 rdev->sysfs_unack_badblocks =
2425 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2426 rdev->sysfs_badblocks =
2427 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2429 list_add_rcu(&rdev->same_set, &mddev->disks);
2430 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2432 /* May as well allow recovery to be retried once */
2433 mddev->recovery_disabled++;
2438 pr_warn("md: failed to register dev-%s for %s\n",
2443 static void rdev_delayed_delete(struct work_struct *ws)
2445 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2446 kobject_del(&rdev->kobj);
2447 kobject_put(&rdev->kobj);
2450 void md_autodetect_dev(dev_t dev);
2452 static void export_rdev(struct md_rdev *rdev)
2454 pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2455 md_rdev_clear(rdev);
2457 if (test_bit(AutoDetected, &rdev->flags))
2458 md_autodetect_dev(rdev->bdev->bd_dev);
2460 blkdev_put(rdev->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2462 kobject_put(&rdev->kobj);
2465 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2467 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2468 list_del_rcu(&rdev->same_set);
2469 pr_debug("md: unbind<%pg>\n", rdev->bdev);
2470 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2472 sysfs_remove_link(&rdev->kobj, "block");
2473 sysfs_put(rdev->sysfs_state);
2474 sysfs_put(rdev->sysfs_unack_badblocks);
2475 sysfs_put(rdev->sysfs_badblocks);
2476 rdev->sysfs_state = NULL;
2477 rdev->sysfs_unack_badblocks = NULL;
2478 rdev->sysfs_badblocks = NULL;
2479 rdev->badblocks.count = 0;
2480 /* We need to delay this, otherwise we can deadlock when
2481 * writing to 'remove' to "dev/state". We also need
2482 * to delay it due to rcu usage.
2485 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2486 kobject_get(&rdev->kobj);
2487 queue_work(md_rdev_misc_wq, &rdev->del_work);
2491 static void export_array(struct mddev *mddev)
2493 struct md_rdev *rdev;
2495 while (!list_empty(&mddev->disks)) {
2496 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2498 md_kick_rdev_from_array(rdev);
2500 mddev->raid_disks = 0;
2501 mddev->major_version = 0;
2504 static bool set_in_sync(struct mddev *mddev)
2506 lockdep_assert_held(&mddev->lock);
2507 if (!mddev->in_sync) {
2508 mddev->sync_checkers++;
2509 spin_unlock(&mddev->lock);
2510 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2511 spin_lock(&mddev->lock);
2512 if (!mddev->in_sync &&
2513 percpu_ref_is_zero(&mddev->writes_pending)) {
2516 * Ensure ->in_sync is visible before we clear
2520 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2521 sysfs_notify_dirent_safe(mddev->sysfs_state);
2523 if (--mddev->sync_checkers == 0)
2524 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2526 if (mddev->safemode == 1)
2527 mddev->safemode = 0;
2528 return mddev->in_sync;
2531 static void sync_sbs(struct mddev *mddev, int nospares)
2533 /* Update each superblock (in-memory image), but
2534 * if we are allowed to, skip spares which already
2535 * have the right event counter, or have one earlier
2536 * (which would mean they aren't being marked as dirty
2537 * with the rest of the array)
2539 struct md_rdev *rdev;
2540 rdev_for_each(rdev, mddev) {
2541 if (rdev->sb_events == mddev->events ||
2543 rdev->raid_disk < 0 &&
2544 rdev->sb_events+1 == mddev->events)) {
2545 /* Don't update this superblock */
2546 rdev->sb_loaded = 2;
2548 sync_super(mddev, rdev);
2549 rdev->sb_loaded = 1;
2554 static bool does_sb_need_changing(struct mddev *mddev)
2556 struct md_rdev *rdev = NULL, *iter;
2557 struct mdp_superblock_1 *sb;
2560 /* Find a good rdev */
2561 rdev_for_each(iter, mddev)
2562 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2567 /* No good device found. */
2571 sb = page_address(rdev->sb_page);
2572 /* Check if a device has become faulty or a spare become active */
2573 rdev_for_each(rdev, mddev) {
2574 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2575 /* Device activated? */
2576 if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2577 !test_bit(Faulty, &rdev->flags))
2579 /* Device turned faulty? */
2580 if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2584 /* Check if any mddev parameters have changed */
2585 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2586 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2587 (mddev->layout != le32_to_cpu(sb->layout)) ||
2588 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2589 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2595 void md_update_sb(struct mddev *mddev, int force_change)
2597 struct md_rdev *rdev;
2600 int any_badblocks_changed = 0;
2603 if (!md_is_rdwr(mddev)) {
2605 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2610 if (mddev_is_clustered(mddev)) {
2611 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2613 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2615 ret = md_cluster_ops->metadata_update_start(mddev);
2616 /* Has someone else has updated the sb */
2617 if (!does_sb_need_changing(mddev)) {
2619 md_cluster_ops->metadata_update_cancel(mddev);
2620 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2621 BIT(MD_SB_CHANGE_DEVS) |
2622 BIT(MD_SB_CHANGE_CLEAN));
2628 * First make sure individual recovery_offsets are correct
2629 * curr_resync_completed can only be used during recovery.
2630 * During reshape/resync it might use array-addresses rather
2631 * that device addresses.
2633 rdev_for_each(rdev, mddev) {
2634 if (rdev->raid_disk >= 0 &&
2635 mddev->delta_disks >= 0 &&
2636 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2637 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2638 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2639 !test_bit(Journal, &rdev->flags) &&
2640 !test_bit(In_sync, &rdev->flags) &&
2641 mddev->curr_resync_completed > rdev->recovery_offset)
2642 rdev->recovery_offset = mddev->curr_resync_completed;
2645 if (!mddev->persistent) {
2646 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2647 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2648 if (!mddev->external) {
2649 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2650 rdev_for_each(rdev, mddev) {
2651 if (rdev->badblocks.changed) {
2652 rdev->badblocks.changed = 0;
2653 ack_all_badblocks(&rdev->badblocks);
2654 md_error(mddev, rdev);
2656 clear_bit(Blocked, &rdev->flags);
2657 clear_bit(BlockedBadBlocks, &rdev->flags);
2658 wake_up(&rdev->blocked_wait);
2661 wake_up(&mddev->sb_wait);
2665 spin_lock(&mddev->lock);
2667 mddev->utime = ktime_get_real_seconds();
2669 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2671 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2672 /* just a clean<-> dirty transition, possibly leave spares alone,
2673 * though if events isn't the right even/odd, we will have to do
2679 if (mddev->degraded)
2680 /* If the array is degraded, then skipping spares is both
2681 * dangerous and fairly pointless.
2682 * Dangerous because a device that was removed from the array
2683 * might have a event_count that still looks up-to-date,
2684 * so it can be re-added without a resync.
2685 * Pointless because if there are any spares to skip,
2686 * then a recovery will happen and soon that array won't
2687 * be degraded any more and the spare can go back to sleep then.
2691 sync_req = mddev->in_sync;
2693 /* If this is just a dirty<->clean transition, and the array is clean
2694 * and 'events' is odd, we can roll back to the previous clean state */
2696 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2697 && mddev->can_decrease_events
2698 && mddev->events != 1) {
2700 mddev->can_decrease_events = 0;
2702 /* otherwise we have to go forward and ... */
2704 mddev->can_decrease_events = nospares;
2708 * This 64-bit counter should never wrap.
2709 * Either we are in around ~1 trillion A.C., assuming
2710 * 1 reboot per second, or we have a bug...
2712 WARN_ON(mddev->events == 0);
2714 rdev_for_each(rdev, mddev) {
2715 if (rdev->badblocks.changed)
2716 any_badblocks_changed++;
2717 if (test_bit(Faulty, &rdev->flags))
2718 set_bit(FaultRecorded, &rdev->flags);
2721 sync_sbs(mddev, nospares);
2722 spin_unlock(&mddev->lock);
2724 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2725 mdname(mddev), mddev->in_sync);
2728 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2730 md_bitmap_update_sb(mddev->bitmap);
2731 rdev_for_each(rdev, mddev) {
2732 if (rdev->sb_loaded != 1)
2733 continue; /* no noise on spare devices */
2735 if (!test_bit(Faulty, &rdev->flags)) {
2736 md_super_write(mddev,rdev,
2737 rdev->sb_start, rdev->sb_size,
2739 pr_debug("md: (write) %pg's sb offset: %llu\n",
2741 (unsigned long long)rdev->sb_start);
2742 rdev->sb_events = mddev->events;
2743 if (rdev->badblocks.size) {
2744 md_super_write(mddev, rdev,
2745 rdev->badblocks.sector,
2746 rdev->badblocks.size << 9,
2748 rdev->badblocks.size = 0;
2752 pr_debug("md: %pg (skipping faulty)\n",
2755 if (mddev->level == LEVEL_MULTIPATH)
2756 /* only need to write one superblock... */
2759 if (md_super_wait(mddev) < 0)
2761 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2763 if (mddev_is_clustered(mddev) && ret == 0)
2764 md_cluster_ops->metadata_update_finish(mddev);
2766 if (mddev->in_sync != sync_req ||
2767 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2768 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2769 /* have to write it out again */
2771 wake_up(&mddev->sb_wait);
2772 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2773 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2775 rdev_for_each(rdev, mddev) {
2776 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2777 clear_bit(Blocked, &rdev->flags);
2779 if (any_badblocks_changed)
2780 ack_all_badblocks(&rdev->badblocks);
2781 clear_bit(BlockedBadBlocks, &rdev->flags);
2782 wake_up(&rdev->blocked_wait);
2785 EXPORT_SYMBOL(md_update_sb);
2787 static int add_bound_rdev(struct md_rdev *rdev)
2789 struct mddev *mddev = rdev->mddev;
2791 bool add_journal = test_bit(Journal, &rdev->flags);
2793 if (!mddev->pers->hot_remove_disk || add_journal) {
2794 /* If there is hot_add_disk but no hot_remove_disk
2795 * then added disks for geometry changes,
2796 * and should be added immediately.
2798 super_types[mddev->major_version].
2799 validate_super(mddev, rdev);
2801 mddev_suspend(mddev);
2802 err = mddev->pers->hot_add_disk(mddev, rdev);
2804 mddev_resume(mddev);
2806 md_kick_rdev_from_array(rdev);
2810 sysfs_notify_dirent_safe(rdev->sysfs_state);
2812 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2813 if (mddev->degraded)
2814 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2815 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2817 md_wakeup_thread(mddev->thread);
2821 /* words written to sysfs files may, or may not, be \n terminated.
2822 * We want to accept with case. For this we use cmd_match.
2824 static int cmd_match(const char *cmd, const char *str)
2826 /* See if cmd, written into a sysfs file, matches
2827 * str. They must either be the same, or cmd can
2828 * have a trailing newline
2830 while (*cmd && *str && *cmd == *str) {
2841 struct rdev_sysfs_entry {
2842 struct attribute attr;
2843 ssize_t (*show)(struct md_rdev *, char *);
2844 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2848 state_show(struct md_rdev *rdev, char *page)
2852 unsigned long flags = READ_ONCE(rdev->flags);
2854 if (test_bit(Faulty, &flags) ||
2855 (!test_bit(ExternalBbl, &flags) &&
2856 rdev->badblocks.unacked_exist))
2857 len += sprintf(page+len, "faulty%s", sep);
2858 if (test_bit(In_sync, &flags))
2859 len += sprintf(page+len, "in_sync%s", sep);
2860 if (test_bit(Journal, &flags))
2861 len += sprintf(page+len, "journal%s", sep);
2862 if (test_bit(WriteMostly, &flags))
2863 len += sprintf(page+len, "write_mostly%s", sep);
2864 if (test_bit(Blocked, &flags) ||
2865 (rdev->badblocks.unacked_exist
2866 && !test_bit(Faulty, &flags)))
2867 len += sprintf(page+len, "blocked%s", sep);
2868 if (!test_bit(Faulty, &flags) &&
2869 !test_bit(Journal, &flags) &&
2870 !test_bit(In_sync, &flags))
2871 len += sprintf(page+len, "spare%s", sep);
2872 if (test_bit(WriteErrorSeen, &flags))
2873 len += sprintf(page+len, "write_error%s", sep);
2874 if (test_bit(WantReplacement, &flags))
2875 len += sprintf(page+len, "want_replacement%s", sep);
2876 if (test_bit(Replacement, &flags))
2877 len += sprintf(page+len, "replacement%s", sep);
2878 if (test_bit(ExternalBbl, &flags))
2879 len += sprintf(page+len, "external_bbl%s", sep);
2880 if (test_bit(FailFast, &flags))
2881 len += sprintf(page+len, "failfast%s", sep);
2886 return len+sprintf(page+len, "\n");
2890 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2893 * faulty - simulates an error
2894 * remove - disconnects the device
2895 * writemostly - sets write_mostly
2896 * -writemostly - clears write_mostly
2897 * blocked - sets the Blocked flags
2898 * -blocked - clears the Blocked and possibly simulates an error
2899 * insync - sets Insync providing device isn't active
2900 * -insync - clear Insync for a device with a slot assigned,
2901 * so that it gets rebuilt based on bitmap
2902 * write_error - sets WriteErrorSeen
2903 * -write_error - clears WriteErrorSeen
2904 * {,-}failfast - set/clear FailFast
2907 struct mddev *mddev = rdev->mddev;
2909 bool need_update_sb = false;
2911 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2912 md_error(rdev->mddev, rdev);
2914 if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2918 } else if (cmd_match(buf, "remove")) {
2919 if (rdev->mddev->pers) {
2920 clear_bit(Blocked, &rdev->flags);
2921 remove_and_add_spares(rdev->mddev, rdev);
2923 if (rdev->raid_disk >= 0)
2927 if (mddev_is_clustered(mddev))
2928 err = md_cluster_ops->remove_disk(mddev, rdev);
2931 md_kick_rdev_from_array(rdev);
2933 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2934 md_wakeup_thread(mddev->thread);
2939 } else if (cmd_match(buf, "writemostly")) {
2940 set_bit(WriteMostly, &rdev->flags);
2941 mddev_create_serial_pool(rdev->mddev, rdev, false);
2942 need_update_sb = true;
2944 } else if (cmd_match(buf, "-writemostly")) {
2945 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2946 clear_bit(WriteMostly, &rdev->flags);
2947 need_update_sb = true;
2949 } else if (cmd_match(buf, "blocked")) {
2950 set_bit(Blocked, &rdev->flags);
2952 } else if (cmd_match(buf, "-blocked")) {
2953 if (!test_bit(Faulty, &rdev->flags) &&
2954 !test_bit(ExternalBbl, &rdev->flags) &&
2955 rdev->badblocks.unacked_exist) {
2956 /* metadata handler doesn't understand badblocks,
2957 * so we need to fail the device
2959 md_error(rdev->mddev, rdev);
2961 clear_bit(Blocked, &rdev->flags);
2962 clear_bit(BlockedBadBlocks, &rdev->flags);
2963 wake_up(&rdev->blocked_wait);
2964 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2965 md_wakeup_thread(rdev->mddev->thread);
2968 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2969 set_bit(In_sync, &rdev->flags);
2971 } else if (cmd_match(buf, "failfast")) {
2972 set_bit(FailFast, &rdev->flags);
2973 need_update_sb = true;
2975 } else if (cmd_match(buf, "-failfast")) {
2976 clear_bit(FailFast, &rdev->flags);
2977 need_update_sb = true;
2979 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2980 !test_bit(Journal, &rdev->flags)) {
2981 if (rdev->mddev->pers == NULL) {
2982 clear_bit(In_sync, &rdev->flags);
2983 rdev->saved_raid_disk = rdev->raid_disk;
2984 rdev->raid_disk = -1;
2987 } else if (cmd_match(buf, "write_error")) {
2988 set_bit(WriteErrorSeen, &rdev->flags);
2990 } else if (cmd_match(buf, "-write_error")) {
2991 clear_bit(WriteErrorSeen, &rdev->flags);
2993 } else if (cmd_match(buf, "want_replacement")) {
2994 /* Any non-spare device that is not a replacement can
2995 * become want_replacement at any time, but we then need to
2996 * check if recovery is needed.
2998 if (rdev->raid_disk >= 0 &&
2999 !test_bit(Journal, &rdev->flags) &&
3000 !test_bit(Replacement, &rdev->flags))
3001 set_bit(WantReplacement, &rdev->flags);
3002 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3003 md_wakeup_thread(rdev->mddev->thread);
3005 } else if (cmd_match(buf, "-want_replacement")) {
3006 /* Clearing 'want_replacement' is always allowed.
3007 * Once replacements starts it is too late though.
3010 clear_bit(WantReplacement, &rdev->flags);
3011 } else if (cmd_match(buf, "replacement")) {
3012 /* Can only set a device as a replacement when array has not
3013 * yet been started. Once running, replacement is automatic
3014 * from spares, or by assigning 'slot'.
3016 if (rdev->mddev->pers)
3019 set_bit(Replacement, &rdev->flags);
3022 } else if (cmd_match(buf, "-replacement")) {
3023 /* Similarly, can only clear Replacement before start */
3024 if (rdev->mddev->pers)
3027 clear_bit(Replacement, &rdev->flags);
3030 } else if (cmd_match(buf, "re-add")) {
3031 if (!rdev->mddev->pers)
3033 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3034 rdev->saved_raid_disk >= 0) {
3035 /* clear_bit is performed _after_ all the devices
3036 * have their local Faulty bit cleared. If any writes
3037 * happen in the meantime in the local node, they
3038 * will land in the local bitmap, which will be synced
3039 * by this node eventually
3041 if (!mddev_is_clustered(rdev->mddev) ||
3042 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3043 clear_bit(Faulty, &rdev->flags);
3044 err = add_bound_rdev(rdev);
3048 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3049 set_bit(ExternalBbl, &rdev->flags);
3050 rdev->badblocks.shift = 0;
3052 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3053 clear_bit(ExternalBbl, &rdev->flags);
3057 md_update_sb(mddev, 1);
3059 sysfs_notify_dirent_safe(rdev->sysfs_state);
3060 return err ? err : len;
3062 static struct rdev_sysfs_entry rdev_state =
3063 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3066 errors_show(struct md_rdev *rdev, char *page)
3068 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3072 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3077 rv = kstrtouint(buf, 10, &n);
3080 atomic_set(&rdev->corrected_errors, n);
3083 static struct rdev_sysfs_entry rdev_errors =
3084 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3087 slot_show(struct md_rdev *rdev, char *page)
3089 if (test_bit(Journal, &rdev->flags))
3090 return sprintf(page, "journal\n");
3091 else if (rdev->raid_disk < 0)
3092 return sprintf(page, "none\n");
3094 return sprintf(page, "%d\n", rdev->raid_disk);
3098 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3103 if (test_bit(Journal, &rdev->flags))
3105 if (strncmp(buf, "none", 4)==0)
3108 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3115 if (rdev->mddev->pers && slot == -1) {
3116 /* Setting 'slot' on an active array requires also
3117 * updating the 'rd%d' link, and communicating
3118 * with the personality with ->hot_*_disk.
3119 * For now we only support removing
3120 * failed/spare devices. This normally happens automatically,
3121 * but not when the metadata is externally managed.
3123 if (rdev->raid_disk == -1)
3125 /* personality does all needed checks */
3126 if (rdev->mddev->pers->hot_remove_disk == NULL)
3128 clear_bit(Blocked, &rdev->flags);
3129 remove_and_add_spares(rdev->mddev, rdev);
3130 if (rdev->raid_disk >= 0)
3132 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3133 md_wakeup_thread(rdev->mddev->thread);
3134 } else if (rdev->mddev->pers) {
3135 /* Activating a spare .. or possibly reactivating
3136 * if we ever get bitmaps working here.
3140 if (rdev->raid_disk != -1)
3143 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3146 if (rdev->mddev->pers->hot_add_disk == NULL)
3149 if (slot >= rdev->mddev->raid_disks &&
3150 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3153 rdev->raid_disk = slot;
3154 if (test_bit(In_sync, &rdev->flags))
3155 rdev->saved_raid_disk = slot;
3157 rdev->saved_raid_disk = -1;
3158 clear_bit(In_sync, &rdev->flags);
3159 clear_bit(Bitmap_sync, &rdev->flags);
3160 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3162 rdev->raid_disk = -1;
3165 sysfs_notify_dirent_safe(rdev->sysfs_state);
3166 /* failure here is OK */;
3167 sysfs_link_rdev(rdev->mddev, rdev);
3168 /* don't wakeup anyone, leave that to userspace. */
3170 if (slot >= rdev->mddev->raid_disks &&
3171 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3173 rdev->raid_disk = slot;
3174 /* assume it is working */
3175 clear_bit(Faulty, &rdev->flags);
3176 clear_bit(WriteMostly, &rdev->flags);
3177 set_bit(In_sync, &rdev->flags);
3178 sysfs_notify_dirent_safe(rdev->sysfs_state);
3183 static struct rdev_sysfs_entry rdev_slot =
3184 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3187 offset_show(struct md_rdev *rdev, char *page)
3189 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3193 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3195 unsigned long long offset;
3196 if (kstrtoull(buf, 10, &offset) < 0)
3198 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3200 if (rdev->sectors && rdev->mddev->external)
3201 /* Must set offset before size, so overlap checks
3204 rdev->data_offset = offset;
3205 rdev->new_data_offset = offset;
3209 static struct rdev_sysfs_entry rdev_offset =
3210 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3212 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3214 return sprintf(page, "%llu\n",
3215 (unsigned long long)rdev->new_data_offset);
3218 static ssize_t new_offset_store(struct md_rdev *rdev,
3219 const char *buf, size_t len)
3221 unsigned long long new_offset;
3222 struct mddev *mddev = rdev->mddev;
3224 if (kstrtoull(buf, 10, &new_offset) < 0)
3227 if (mddev->sync_thread ||
3228 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3230 if (new_offset == rdev->data_offset)
3231 /* reset is always permitted */
3233 else if (new_offset > rdev->data_offset) {
3234 /* must not push array size beyond rdev_sectors */
3235 if (new_offset - rdev->data_offset
3236 + mddev->dev_sectors > rdev->sectors)
3239 /* Metadata worries about other space details. */
3241 /* decreasing the offset is inconsistent with a backwards
3244 if (new_offset < rdev->data_offset &&
3245 mddev->reshape_backwards)
3247 /* Increasing offset is inconsistent with forwards
3248 * reshape. reshape_direction should be set to
3249 * 'backwards' first.
3251 if (new_offset > rdev->data_offset &&
3252 !mddev->reshape_backwards)
3255 if (mddev->pers && mddev->persistent &&
3256 !super_types[mddev->major_version]
3257 .allow_new_offset(rdev, new_offset))
3259 rdev->new_data_offset = new_offset;
3260 if (new_offset > rdev->data_offset)
3261 mddev->reshape_backwards = 1;
3262 else if (new_offset < rdev->data_offset)
3263 mddev->reshape_backwards = 0;
3267 static struct rdev_sysfs_entry rdev_new_offset =
3268 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3271 rdev_size_show(struct md_rdev *rdev, char *page)
3273 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3276 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3278 /* check if two start/length pairs overlap */
3279 if (a->data_offset + a->sectors <= b->data_offset)
3281 if (b->data_offset + b->sectors <= a->data_offset)
3286 static bool md_rdev_overlaps(struct md_rdev *rdev)
3288 struct mddev *mddev;
3289 struct md_rdev *rdev2;
3291 spin_lock(&all_mddevs_lock);
3292 list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3293 if (test_bit(MD_DELETED, &mddev->flags))
3295 rdev_for_each(rdev2, mddev) {
3296 if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3297 md_rdevs_overlap(rdev, rdev2)) {
3298 spin_unlock(&all_mddevs_lock);
3303 spin_unlock(&all_mddevs_lock);
3307 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3309 unsigned long long blocks;
3312 if (kstrtoull(buf, 10, &blocks) < 0)
3315 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3316 return -EINVAL; /* sector conversion overflow */
3319 if (new != blocks * 2)
3320 return -EINVAL; /* unsigned long long to sector_t overflow */
3327 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3329 struct mddev *my_mddev = rdev->mddev;
3330 sector_t oldsectors = rdev->sectors;
3333 if (test_bit(Journal, &rdev->flags))
3335 if (strict_blocks_to_sectors(buf, §ors) < 0)
3337 if (rdev->data_offset != rdev->new_data_offset)
3338 return -EINVAL; /* too confusing */
3339 if (my_mddev->pers && rdev->raid_disk >= 0) {
3340 if (my_mddev->persistent) {
3341 sectors = super_types[my_mddev->major_version].
3342 rdev_size_change(rdev, sectors);
3345 } else if (!sectors)
3346 sectors = bdev_nr_sectors(rdev->bdev) -
3348 if (!my_mddev->pers->resize)
3349 /* Cannot change size for RAID0 or Linear etc */
3352 if (sectors < my_mddev->dev_sectors)
3353 return -EINVAL; /* component must fit device */
3355 rdev->sectors = sectors;
3358 * Check that all other rdevs with the same bdev do not overlap. This
3359 * check does not provide a hard guarantee, it just helps avoid
3360 * dangerous mistakes.
3362 if (sectors > oldsectors && my_mddev->external &&
3363 md_rdev_overlaps(rdev)) {
3365 * Someone else could have slipped in a size change here, but
3366 * doing so is just silly. We put oldsectors back because we
3367 * know it is safe, and trust userspace not to race with itself.
3369 rdev->sectors = oldsectors;
3375 static struct rdev_sysfs_entry rdev_size =
3376 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3378 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3380 unsigned long long recovery_start = rdev->recovery_offset;
3382 if (test_bit(In_sync, &rdev->flags) ||
3383 recovery_start == MaxSector)
3384 return sprintf(page, "none\n");
3386 return sprintf(page, "%llu\n", recovery_start);
3389 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3391 unsigned long long recovery_start;
3393 if (cmd_match(buf, "none"))
3394 recovery_start = MaxSector;
3395 else if (kstrtoull(buf, 10, &recovery_start))
3398 if (rdev->mddev->pers &&
3399 rdev->raid_disk >= 0)
3402 rdev->recovery_offset = recovery_start;
3403 if (recovery_start == MaxSector)
3404 set_bit(In_sync, &rdev->flags);
3406 clear_bit(In_sync, &rdev->flags);
3410 static struct rdev_sysfs_entry rdev_recovery_start =
3411 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3413 /* sysfs access to bad-blocks list.
3414 * We present two files.
3415 * 'bad-blocks' lists sector numbers and lengths of ranges that
3416 * are recorded as bad. The list is truncated to fit within
3417 * the one-page limit of sysfs.
3418 * Writing "sector length" to this file adds an acknowledged
3420 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3421 * been acknowledged. Writing to this file adds bad blocks
3422 * without acknowledging them. This is largely for testing.
3424 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3426 return badblocks_show(&rdev->badblocks, page, 0);
3428 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3430 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3431 /* Maybe that ack was all we needed */
3432 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3433 wake_up(&rdev->blocked_wait);
3436 static struct rdev_sysfs_entry rdev_bad_blocks =
3437 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3439 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3441 return badblocks_show(&rdev->badblocks, page, 1);
3443 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3445 return badblocks_store(&rdev->badblocks, page, len, 1);
3447 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3448 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3451 ppl_sector_show(struct md_rdev *rdev, char *page)
3453 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3457 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3459 unsigned long long sector;
3461 if (kstrtoull(buf, 10, §or) < 0)
3463 if (sector != (sector_t)sector)
3466 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3467 rdev->raid_disk >= 0)
3470 if (rdev->mddev->persistent) {
3471 if (rdev->mddev->major_version == 0)
3473 if ((sector > rdev->sb_start &&
3474 sector - rdev->sb_start > S16_MAX) ||
3475 (sector < rdev->sb_start &&
3476 rdev->sb_start - sector > -S16_MIN))
3478 rdev->ppl.offset = sector - rdev->sb_start;
3479 } else if (!rdev->mddev->external) {
3482 rdev->ppl.sector = sector;
3486 static struct rdev_sysfs_entry rdev_ppl_sector =
3487 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3490 ppl_size_show(struct md_rdev *rdev, char *page)
3492 return sprintf(page, "%u\n", rdev->ppl.size);
3496 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3500 if (kstrtouint(buf, 10, &size) < 0)
3503 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3504 rdev->raid_disk >= 0)
3507 if (rdev->mddev->persistent) {
3508 if (rdev->mddev->major_version == 0)
3512 } else if (!rdev->mddev->external) {
3515 rdev->ppl.size = size;
3519 static struct rdev_sysfs_entry rdev_ppl_size =
3520 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3522 static struct attribute *rdev_default_attrs[] = {
3527 &rdev_new_offset.attr,
3529 &rdev_recovery_start.attr,
3530 &rdev_bad_blocks.attr,
3531 &rdev_unack_bad_blocks.attr,
3532 &rdev_ppl_sector.attr,
3533 &rdev_ppl_size.attr,
3536 ATTRIBUTE_GROUPS(rdev_default);
3538 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3540 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3541 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3547 return entry->show(rdev, page);
3551 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3552 const char *page, size_t length)
3554 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3555 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3557 struct mddev *mddev = rdev->mddev;
3561 if (!capable(CAP_SYS_ADMIN))
3563 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3565 if (rdev->mddev == NULL)
3568 rv = entry->store(rdev, page, length);
3569 mddev_unlock(mddev);
3574 static void rdev_free(struct kobject *ko)
3576 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3579 static const struct sysfs_ops rdev_sysfs_ops = {
3580 .show = rdev_attr_show,
3581 .store = rdev_attr_store,
3583 static const struct kobj_type rdev_ktype = {
3584 .release = rdev_free,
3585 .sysfs_ops = &rdev_sysfs_ops,
3586 .default_groups = rdev_default_groups,
3589 int md_rdev_init(struct md_rdev *rdev)
3592 rdev->saved_raid_disk = -1;
3593 rdev->raid_disk = -1;
3595 rdev->data_offset = 0;
3596 rdev->new_data_offset = 0;
3597 rdev->sb_events = 0;
3598 rdev->last_read_error = 0;
3599 rdev->sb_loaded = 0;
3600 rdev->bb_page = NULL;
3601 atomic_set(&rdev->nr_pending, 0);
3602 atomic_set(&rdev->read_errors, 0);
3603 atomic_set(&rdev->corrected_errors, 0);
3605 INIT_LIST_HEAD(&rdev->same_set);
3606 init_waitqueue_head(&rdev->blocked_wait);
3608 /* Add space to store bad block list.
3609 * This reserves the space even on arrays where it cannot
3610 * be used - I wonder if that matters
3612 return badblocks_init(&rdev->badblocks, 0);
3614 EXPORT_SYMBOL_GPL(md_rdev_init);
3616 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3618 * mark the device faulty if:
3620 * - the device is nonexistent (zero size)
3621 * - the device has no valid superblock
3623 * a faulty rdev _never_ has rdev->sb set.
3625 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3627 static struct md_rdev claim_rdev; /* just for claiming the bdev */
3628 struct md_rdev *rdev;
3632 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3634 return ERR_PTR(-ENOMEM);
3636 err = md_rdev_init(rdev);
3639 err = alloc_disk_sb(rdev);
3641 goto out_clear_rdev;
3643 rdev->bdev = blkdev_get_by_dev(newdev,
3644 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
3645 super_format == -2 ? &claim_rdev : rdev);
3646 if (IS_ERR(rdev->bdev)) {
3647 pr_warn("md: could not open device unknown-block(%u,%u).\n",
3648 MAJOR(newdev), MINOR(newdev));
3649 err = PTR_ERR(rdev->bdev);
3650 goto out_clear_rdev;
3653 kobject_init(&rdev->kobj, &rdev_ktype);
3655 size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3657 pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3660 goto out_blkdev_put;
3663 if (super_format >= 0) {
3664 err = super_types[super_format].
3665 load_super(rdev, NULL, super_minor);
3666 if (err == -EINVAL) {
3667 pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3669 super_format, super_minor);
3670 goto out_blkdev_put;
3673 pr_warn("md: could not read %pg's sb, not importing!\n",
3675 goto out_blkdev_put;
3682 blkdev_put(rdev->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3684 md_rdev_clear(rdev);
3687 return ERR_PTR(err);
3691 * Check a full RAID array for plausibility
3694 static int analyze_sbs(struct mddev *mddev)
3697 struct md_rdev *rdev, *freshest, *tmp;
3700 rdev_for_each_safe(rdev, tmp, mddev)
3701 switch (super_types[mddev->major_version].
3702 load_super(rdev, freshest, mddev->minor_version)) {
3709 pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3711 md_kick_rdev_from_array(rdev);
3714 /* Cannot find a valid fresh disk */
3716 pr_warn("md: cannot find a valid disk\n");
3720 super_types[mddev->major_version].
3721 validate_super(mddev, freshest);
3724 rdev_for_each_safe(rdev, tmp, mddev) {
3725 if (mddev->max_disks &&
3726 (rdev->desc_nr >= mddev->max_disks ||
3727 i > mddev->max_disks)) {
3728 pr_warn("md: %s: %pg: only %d devices permitted\n",
3729 mdname(mddev), rdev->bdev,
3731 md_kick_rdev_from_array(rdev);
3734 if (rdev != freshest) {
3735 if (super_types[mddev->major_version].
3736 validate_super(mddev, rdev)) {
3737 pr_warn("md: kicking non-fresh %pg from array!\n",
3739 md_kick_rdev_from_array(rdev);
3743 if (mddev->level == LEVEL_MULTIPATH) {
3744 rdev->desc_nr = i++;
3745 rdev->raid_disk = rdev->desc_nr;
3746 set_bit(In_sync, &rdev->flags);
3747 } else if (rdev->raid_disk >=
3748 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3749 !test_bit(Journal, &rdev->flags)) {
3750 rdev->raid_disk = -1;
3751 clear_bit(In_sync, &rdev->flags);
3758 /* Read a fixed-point number.
3759 * Numbers in sysfs attributes should be in "standard" units where
3760 * possible, so time should be in seconds.
3761 * However we internally use a a much smaller unit such as
3762 * milliseconds or jiffies.
3763 * This function takes a decimal number with a possible fractional
3764 * component, and produces an integer which is the result of
3765 * multiplying that number by 10^'scale'.
3766 * all without any floating-point arithmetic.
3768 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3770 unsigned long result = 0;
3772 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3775 else if (decimals < scale) {
3778 result = result * 10 + value;
3790 *res = result * int_pow(10, scale - decimals);
3795 safe_delay_show(struct mddev *mddev, char *page)
3797 int msec = (mddev->safemode_delay*1000)/HZ;
3798 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3801 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3805 if (mddev_is_clustered(mddev)) {
3806 pr_warn("md: Safemode is disabled for clustered mode\n");
3810 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3813 mddev->safemode_delay = 0;
3815 unsigned long old_delay = mddev->safemode_delay;
3816 unsigned long new_delay = (msec*HZ)/1000;
3820 mddev->safemode_delay = new_delay;
3821 if (new_delay < old_delay || old_delay == 0)
3822 mod_timer(&mddev->safemode_timer, jiffies+1);
3826 static struct md_sysfs_entry md_safe_delay =
3827 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3830 level_show(struct mddev *mddev, char *page)
3832 struct md_personality *p;
3834 spin_lock(&mddev->lock);
3837 ret = sprintf(page, "%s\n", p->name);
3838 else if (mddev->clevel[0])
3839 ret = sprintf(page, "%s\n", mddev->clevel);
3840 else if (mddev->level != LEVEL_NONE)
3841 ret = sprintf(page, "%d\n", mddev->level);
3844 spin_unlock(&mddev->lock);
3849 level_store(struct mddev *mddev, const char *buf, size_t len)
3854 struct md_personality *pers, *oldpers;
3856 void *priv, *oldpriv;
3857 struct md_rdev *rdev;
3859 if (slen == 0 || slen >= sizeof(clevel))
3862 rv = mddev_lock(mddev);
3866 if (mddev->pers == NULL) {
3867 strncpy(mddev->clevel, buf, slen);
3868 if (mddev->clevel[slen-1] == '\n')
3870 mddev->clevel[slen] = 0;
3871 mddev->level = LEVEL_NONE;
3876 if (!md_is_rdwr(mddev))
3879 /* request to change the personality. Need to ensure:
3880 * - array is not engaged in resync/recovery/reshape
3881 * - old personality can be suspended
3882 * - new personality will access other array.
3886 if (mddev->sync_thread ||
3887 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3888 mddev->reshape_position != MaxSector ||
3889 mddev->sysfs_active)
3893 if (!mddev->pers->quiesce) {
3894 pr_warn("md: %s: %s does not support online personality change\n",
3895 mdname(mddev), mddev->pers->name);
3899 /* Now find the new personality */
3900 strncpy(clevel, buf, slen);
3901 if (clevel[slen-1] == '\n')
3904 if (kstrtol(clevel, 10, &level))
3907 if (request_module("md-%s", clevel) != 0)
3908 request_module("md-level-%s", clevel);
3909 spin_lock(&pers_lock);
3910 pers = find_pers(level, clevel);
3911 if (!pers || !try_module_get(pers->owner)) {
3912 spin_unlock(&pers_lock);
3913 pr_warn("md: personality %s not loaded\n", clevel);
3917 spin_unlock(&pers_lock);
3919 if (pers == mddev->pers) {
3920 /* Nothing to do! */
3921 module_put(pers->owner);
3925 if (!pers->takeover) {
3926 module_put(pers->owner);
3927 pr_warn("md: %s: %s does not support personality takeover\n",
3928 mdname(mddev), clevel);
3933 rdev_for_each(rdev, mddev)
3934 rdev->new_raid_disk = rdev->raid_disk;
3936 /* ->takeover must set new_* and/or delta_disks
3937 * if it succeeds, and may set them when it fails.
3939 priv = pers->takeover(mddev);
3941 mddev->new_level = mddev->level;
3942 mddev->new_layout = mddev->layout;
3943 mddev->new_chunk_sectors = mddev->chunk_sectors;
3944 mddev->raid_disks -= mddev->delta_disks;
3945 mddev->delta_disks = 0;
3946 mddev->reshape_backwards = 0;
3947 module_put(pers->owner);
3948 pr_warn("md: %s: %s would not accept array\n",
3949 mdname(mddev), clevel);
3954 /* Looks like we have a winner */
3955 mddev_suspend(mddev);
3956 mddev_detach(mddev);
3958 spin_lock(&mddev->lock);
3959 oldpers = mddev->pers;
3960 oldpriv = mddev->private;
3962 mddev->private = priv;
3963 strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3964 mddev->level = mddev->new_level;
3965 mddev->layout = mddev->new_layout;
3966 mddev->chunk_sectors = mddev->new_chunk_sectors;
3967 mddev->delta_disks = 0;
3968 mddev->reshape_backwards = 0;
3969 mddev->degraded = 0;
3970 spin_unlock(&mddev->lock);
3972 if (oldpers->sync_request == NULL &&
3974 /* We are converting from a no-redundancy array
3975 * to a redundancy array and metadata is managed
3976 * externally so we need to be sure that writes
3977 * won't block due to a need to transition
3979 * until external management is started.
3982 mddev->safemode_delay = 0;
3983 mddev->safemode = 0;
3986 oldpers->free(mddev, oldpriv);
3988 if (oldpers->sync_request == NULL &&
3989 pers->sync_request != NULL) {
3990 /* need to add the md_redundancy_group */
3991 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3992 pr_warn("md: cannot register extra attributes for %s\n",
3994 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3995 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
3996 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
3998 if (oldpers->sync_request != NULL &&
3999 pers->sync_request == NULL) {
4000 /* need to remove the md_redundancy_group */
4001 if (mddev->to_remove == NULL)
4002 mddev->to_remove = &md_redundancy_group;
4005 module_put(oldpers->owner);
4007 rdev_for_each(rdev, mddev) {
4008 if (rdev->raid_disk < 0)
4010 if (rdev->new_raid_disk >= mddev->raid_disks)
4011 rdev->new_raid_disk = -1;
4012 if (rdev->new_raid_disk == rdev->raid_disk)
4014 sysfs_unlink_rdev(mddev, rdev);
4016 rdev_for_each(rdev, mddev) {
4017 if (rdev->raid_disk < 0)
4019 if (rdev->new_raid_disk == rdev->raid_disk)
4021 rdev->raid_disk = rdev->new_raid_disk;
4022 if (rdev->raid_disk < 0)
4023 clear_bit(In_sync, &rdev->flags);
4025 if (sysfs_link_rdev(mddev, rdev))
4026 pr_warn("md: cannot register rd%d for %s after level change\n",
4027 rdev->raid_disk, mdname(mddev));
4031 if (pers->sync_request == NULL) {
4032 /* this is now an array without redundancy, so
4033 * it must always be in_sync
4036 del_timer_sync(&mddev->safemode_timer);
4038 blk_set_stacking_limits(&mddev->queue->limits);
4040 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4041 mddev_resume(mddev);
4043 md_update_sb(mddev, 1);
4044 sysfs_notify_dirent_safe(mddev->sysfs_level);
4048 mddev_unlock(mddev);
4052 static struct md_sysfs_entry md_level =
4053 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4056 layout_show(struct mddev *mddev, char *page)
4058 /* just a number, not meaningful for all levels */
4059 if (mddev->reshape_position != MaxSector &&
4060 mddev->layout != mddev->new_layout)
4061 return sprintf(page, "%d (%d)\n",
4062 mddev->new_layout, mddev->layout);
4063 return sprintf(page, "%d\n", mddev->layout);
4067 layout_store(struct mddev *mddev, const char *buf, size_t len)
4072 err = kstrtouint(buf, 10, &n);
4075 err = mddev_lock(mddev);
4080 if (mddev->pers->check_reshape == NULL)
4082 else if (!md_is_rdwr(mddev))
4085 mddev->new_layout = n;
4086 err = mddev->pers->check_reshape(mddev);
4088 mddev->new_layout = mddev->layout;
4091 mddev->new_layout = n;
4092 if (mddev->reshape_position == MaxSector)
4095 mddev_unlock(mddev);
4098 static struct md_sysfs_entry md_layout =
4099 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4102 raid_disks_show(struct mddev *mddev, char *page)
4104 if (mddev->raid_disks == 0)
4106 if (mddev->reshape_position != MaxSector &&
4107 mddev->delta_disks != 0)
4108 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4109 mddev->raid_disks - mddev->delta_disks);
4110 return sprintf(page, "%d\n", mddev->raid_disks);
4113 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4116 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4121 err = kstrtouint(buf, 10, &n);
4125 err = mddev_lock(mddev);
4129 err = update_raid_disks(mddev, n);
4130 else if (mddev->reshape_position != MaxSector) {
4131 struct md_rdev *rdev;
4132 int olddisks = mddev->raid_disks - mddev->delta_disks;
4135 rdev_for_each(rdev, mddev) {
4137 rdev->data_offset < rdev->new_data_offset)
4140 rdev->data_offset > rdev->new_data_offset)
4144 mddev->delta_disks = n - olddisks;
4145 mddev->raid_disks = n;
4146 mddev->reshape_backwards = (mddev->delta_disks < 0);
4148 mddev->raid_disks = n;
4150 mddev_unlock(mddev);
4151 return err ? err : len;
4153 static struct md_sysfs_entry md_raid_disks =
4154 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4157 uuid_show(struct mddev *mddev, char *page)
4159 return sprintf(page, "%pU\n", mddev->uuid);
4161 static struct md_sysfs_entry md_uuid =
4162 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4165 chunk_size_show(struct mddev *mddev, char *page)
4167 if (mddev->reshape_position != MaxSector &&
4168 mddev->chunk_sectors != mddev->new_chunk_sectors)
4169 return sprintf(page, "%d (%d)\n",
4170 mddev->new_chunk_sectors << 9,
4171 mddev->chunk_sectors << 9);
4172 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4176 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4181 err = kstrtoul(buf, 10, &n);
4185 err = mddev_lock(mddev);
4189 if (mddev->pers->check_reshape == NULL)
4191 else if (!md_is_rdwr(mddev))
4194 mddev->new_chunk_sectors = n >> 9;
4195 err = mddev->pers->check_reshape(mddev);
4197 mddev->new_chunk_sectors = mddev->chunk_sectors;
4200 mddev->new_chunk_sectors = n >> 9;
4201 if (mddev->reshape_position == MaxSector)
4202 mddev->chunk_sectors = n >> 9;
4204 mddev_unlock(mddev);
4207 static struct md_sysfs_entry md_chunk_size =
4208 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4211 resync_start_show(struct mddev *mddev, char *page)
4213 if (mddev->recovery_cp == MaxSector)
4214 return sprintf(page, "none\n");
4215 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4219 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4221 unsigned long long n;
4224 if (cmd_match(buf, "none"))
4227 err = kstrtoull(buf, 10, &n);
4230 if (n != (sector_t)n)
4234 err = mddev_lock(mddev);
4237 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4241 mddev->recovery_cp = n;
4243 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4245 mddev_unlock(mddev);
4248 static struct md_sysfs_entry md_resync_start =
4249 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4250 resync_start_show, resync_start_store);
4253 * The array state can be:
4256 * No devices, no size, no level
4257 * Equivalent to STOP_ARRAY ioctl
4259 * May have some settings, but array is not active
4260 * all IO results in error
4261 * When written, doesn't tear down array, but just stops it
4262 * suspended (not supported yet)
4263 * All IO requests will block. The array can be reconfigured.
4264 * Writing this, if accepted, will block until array is quiescent
4266 * no resync can happen. no superblocks get written.
4267 * write requests fail
4269 * like readonly, but behaves like 'clean' on a write request.
4271 * clean - no pending writes, but otherwise active.
4272 * When written to inactive array, starts without resync
4273 * If a write request arrives then
4274 * if metadata is known, mark 'dirty' and switch to 'active'.
4275 * if not known, block and switch to write-pending
4276 * If written to an active array that has pending writes, then fails.
4278 * fully active: IO and resync can be happening.
4279 * When written to inactive array, starts with resync
4282 * clean, but writes are blocked waiting for 'active' to be written.
4285 * like active, but no writes have been seen for a while (100msec).
4288 * Array is failed. It's useful because mounted-arrays aren't stopped
4289 * when array is failed, so this state will at least alert the user that
4290 * something is wrong.
4292 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4293 write_pending, active_idle, broken, bad_word};
4294 static char *array_states[] = {
4295 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4296 "write-pending", "active-idle", "broken", NULL };
4298 static int match_word(const char *word, char **list)
4301 for (n=0; list[n]; n++)
4302 if (cmd_match(word, list[n]))
4308 array_state_show(struct mddev *mddev, char *page)
4310 enum array_state st = inactive;
4312 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4321 spin_lock(&mddev->lock);
4322 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4324 else if (mddev->in_sync)
4326 else if (mddev->safemode)
4330 spin_unlock(&mddev->lock);
4333 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4336 if (list_empty(&mddev->disks) &&
4337 mddev->raid_disks == 0 &&
4338 mddev->dev_sectors == 0)
4343 return sprintf(page, "%s\n", array_states[st]);
4346 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4347 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4348 static int restart_array(struct mddev *mddev);
4351 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4354 enum array_state st = match_word(buf, array_states);
4356 if (mddev->pers && (st == active || st == clean) &&
4357 mddev->ro != MD_RDONLY) {
4358 /* don't take reconfig_mutex when toggling between
4361 spin_lock(&mddev->lock);
4363 restart_array(mddev);
4364 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4365 md_wakeup_thread(mddev->thread);
4366 wake_up(&mddev->sb_wait);
4367 } else /* st == clean */ {
4368 restart_array(mddev);
4369 if (!set_in_sync(mddev))
4373 sysfs_notify_dirent_safe(mddev->sysfs_state);
4374 spin_unlock(&mddev->lock);
4377 err = mddev_lock(mddev);
4385 /* stopping an active array */
4386 err = do_md_stop(mddev, 0, NULL);
4389 /* stopping an active array */
4391 err = do_md_stop(mddev, 2, NULL);
4393 err = 0; /* already inactive */
4396 break; /* not supported yet */
4399 err = md_set_readonly(mddev, NULL);
4401 mddev->ro = MD_RDONLY;
4402 set_disk_ro(mddev->gendisk, 1);
4403 err = do_md_run(mddev);
4408 if (md_is_rdwr(mddev))
4409 err = md_set_readonly(mddev, NULL);
4410 else if (mddev->ro == MD_RDONLY)
4411 err = restart_array(mddev);
4413 mddev->ro = MD_AUTO_READ;
4414 set_disk_ro(mddev->gendisk, 0);
4417 mddev->ro = MD_AUTO_READ;
4418 err = do_md_run(mddev);
4423 err = restart_array(mddev);
4426 spin_lock(&mddev->lock);
4427 if (!set_in_sync(mddev))
4429 spin_unlock(&mddev->lock);
4435 err = restart_array(mddev);
4438 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4439 wake_up(&mddev->sb_wait);
4442 mddev->ro = MD_RDWR;
4443 set_disk_ro(mddev->gendisk, 0);
4444 err = do_md_run(mddev);
4450 /* these cannot be set */
4455 if (mddev->hold_active == UNTIL_IOCTL)
4456 mddev->hold_active = 0;
4457 sysfs_notify_dirent_safe(mddev->sysfs_state);
4459 mddev_unlock(mddev);
4462 static struct md_sysfs_entry md_array_state =
4463 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4466 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4467 return sprintf(page, "%d\n",
4468 atomic_read(&mddev->max_corr_read_errors));
4472 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4477 rv = kstrtouint(buf, 10, &n);
4480 atomic_set(&mddev->max_corr_read_errors, n);
4484 static struct md_sysfs_entry max_corr_read_errors =
4485 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4486 max_corrected_read_errors_store);
4489 null_show(struct mddev *mddev, char *page)
4494 /* need to ensure rdev_delayed_delete() has completed */
4495 static void flush_rdev_wq(struct mddev *mddev)
4497 struct md_rdev *rdev;
4500 rdev_for_each_rcu(rdev, mddev)
4501 if (work_pending(&rdev->del_work)) {
4502 flush_workqueue(md_rdev_misc_wq);
4509 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4511 /* buf must be %d:%d\n? giving major and minor numbers */
4512 /* The new device is added to the array.
4513 * If the array has a persistent superblock, we read the
4514 * superblock to initialise info and check validity.
4515 * Otherwise, only checking done is that in bind_rdev_to_array,
4516 * which mainly checks size.
4519 int major = simple_strtoul(buf, &e, 10);
4522 struct md_rdev *rdev;
4525 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4527 minor = simple_strtoul(e+1, &e, 10);
4528 if (*e && *e != '\n')
4530 dev = MKDEV(major, minor);
4531 if (major != MAJOR(dev) ||
4532 minor != MINOR(dev))
4535 flush_rdev_wq(mddev);
4536 err = mddev_lock(mddev);
4539 if (mddev->persistent) {
4540 rdev = md_import_device(dev, mddev->major_version,
4541 mddev->minor_version);
4542 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4543 struct md_rdev *rdev0
4544 = list_entry(mddev->disks.next,
4545 struct md_rdev, same_set);
4546 err = super_types[mddev->major_version]
4547 .load_super(rdev, rdev0, mddev->minor_version);
4551 } else if (mddev->external)
4552 rdev = md_import_device(dev, -2, -1);
4554 rdev = md_import_device(dev, -1, -1);
4557 mddev_unlock(mddev);
4558 return PTR_ERR(rdev);
4560 err = bind_rdev_to_array(rdev, mddev);
4564 mddev_unlock(mddev);
4567 return err ? err : len;
4570 static struct md_sysfs_entry md_new_device =
4571 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4574 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4577 unsigned long chunk, end_chunk;
4580 err = mddev_lock(mddev);
4585 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4587 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4588 if (buf == end) break;
4589 if (*end == '-') { /* range */
4591 end_chunk = simple_strtoul(buf, &end, 0);
4592 if (buf == end) break;
4594 if (*end && !isspace(*end)) break;
4595 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4596 buf = skip_spaces(end);
4598 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4600 mddev_unlock(mddev);
4604 static struct md_sysfs_entry md_bitmap =
4605 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4608 size_show(struct mddev *mddev, char *page)
4610 return sprintf(page, "%llu\n",
4611 (unsigned long long)mddev->dev_sectors / 2);
4614 static int update_size(struct mddev *mddev, sector_t num_sectors);
4617 size_store(struct mddev *mddev, const char *buf, size_t len)
4619 /* If array is inactive, we can reduce the component size, but
4620 * not increase it (except from 0).
4621 * If array is active, we can try an on-line resize
4624 int err = strict_blocks_to_sectors(buf, §ors);
4628 err = mddev_lock(mddev);
4632 err = update_size(mddev, sectors);
4634 md_update_sb(mddev, 1);
4636 if (mddev->dev_sectors == 0 ||
4637 mddev->dev_sectors > sectors)
4638 mddev->dev_sectors = sectors;
4642 mddev_unlock(mddev);
4643 return err ? err : len;
4646 static struct md_sysfs_entry md_size =
4647 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4649 /* Metadata version.
4651 * 'none' for arrays with no metadata (good luck...)
4652 * 'external' for arrays with externally managed metadata,
4653 * or N.M for internally known formats
4656 metadata_show(struct mddev *mddev, char *page)
4658 if (mddev->persistent)
4659 return sprintf(page, "%d.%d\n",
4660 mddev->major_version, mddev->minor_version);
4661 else if (mddev->external)
4662 return sprintf(page, "external:%s\n", mddev->metadata_type);
4664 return sprintf(page, "none\n");
4668 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4673 /* Changing the details of 'external' metadata is
4674 * always permitted. Otherwise there must be
4675 * no devices attached to the array.
4678 err = mddev_lock(mddev);
4682 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4684 else if (!list_empty(&mddev->disks))
4688 if (cmd_match(buf, "none")) {
4689 mddev->persistent = 0;
4690 mddev->external = 0;
4691 mddev->major_version = 0;
4692 mddev->minor_version = 90;
4695 if (strncmp(buf, "external:", 9) == 0) {
4696 size_t namelen = len-9;
4697 if (namelen >= sizeof(mddev->metadata_type))
4698 namelen = sizeof(mddev->metadata_type)-1;
4699 strncpy(mddev->metadata_type, buf+9, namelen);
4700 mddev->metadata_type[namelen] = 0;
4701 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4702 mddev->metadata_type[--namelen] = 0;
4703 mddev->persistent = 0;
4704 mddev->external = 1;
4705 mddev->major_version = 0;
4706 mddev->minor_version = 90;
4709 major = simple_strtoul(buf, &e, 10);
4711 if (e==buf || *e != '.')
4714 minor = simple_strtoul(buf, &e, 10);
4715 if (e==buf || (*e && *e != '\n') )
4718 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4720 mddev->major_version = major;
4721 mddev->minor_version = minor;
4722 mddev->persistent = 1;
4723 mddev->external = 0;
4726 mddev_unlock(mddev);
4730 static struct md_sysfs_entry md_metadata =
4731 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4734 action_show(struct mddev *mddev, char *page)
4736 char *type = "idle";
4737 unsigned long recovery = mddev->recovery;
4738 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4740 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4741 (md_is_rdwr(mddev) && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4742 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4744 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4745 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4747 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4751 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4753 else if (mddev->reshape_position != MaxSector)
4756 return sprintf(page, "%s\n", type);
4760 action_store(struct mddev *mddev, const char *page, size_t len)
4762 if (!mddev->pers || !mddev->pers->sync_request)
4766 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4767 if (cmd_match(page, "frozen"))
4768 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4770 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4771 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4772 mddev_lock(mddev) == 0) {
4773 if (work_pending(&mddev->del_work))
4774 flush_workqueue(md_misc_wq);
4775 if (mddev->sync_thread) {
4776 sector_t save_rp = mddev->reshape_position;
4778 mddev_unlock(mddev);
4779 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4780 md_unregister_thread(&mddev->sync_thread);
4781 mddev_lock_nointr(mddev);
4783 * set RECOVERY_INTR again and restore reshape
4784 * position in case others changed them after
4785 * got lock, eg, reshape_position_store and
4786 * md_check_recovery.
4788 mddev->reshape_position = save_rp;
4789 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4790 md_reap_sync_thread(mddev);
4792 mddev_unlock(mddev);
4794 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4796 else if (cmd_match(page, "resync"))
4797 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4798 else if (cmd_match(page, "recover")) {
4799 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4800 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4801 } else if (cmd_match(page, "reshape")) {
4803 if (mddev->pers->start_reshape == NULL)
4805 err = mddev_lock(mddev);
4807 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4810 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4811 err = mddev->pers->start_reshape(mddev);
4813 mddev_unlock(mddev);
4817 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4819 if (cmd_match(page, "check"))
4820 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4821 else if (!cmd_match(page, "repair"))
4823 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4824 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4825 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4827 if (mddev->ro == MD_AUTO_READ) {
4828 /* A write to sync_action is enough to justify
4829 * canceling read-auto mode
4831 mddev->ro = MD_RDWR;
4832 md_wakeup_thread(mddev->sync_thread);
4834 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4835 md_wakeup_thread(mddev->thread);
4836 sysfs_notify_dirent_safe(mddev->sysfs_action);
4840 static struct md_sysfs_entry md_scan_mode =
4841 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4844 last_sync_action_show(struct mddev *mddev, char *page)
4846 return sprintf(page, "%s\n", mddev->last_sync_action);
4849 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4852 mismatch_cnt_show(struct mddev *mddev, char *page)
4854 return sprintf(page, "%llu\n",
4855 (unsigned long long)
4856 atomic64_read(&mddev->resync_mismatches));
4859 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4862 sync_min_show(struct mddev *mddev, char *page)
4864 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4865 mddev->sync_speed_min ? "local": "system");
4869 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4874 if (strncmp(buf, "system", 6)==0) {
4877 rv = kstrtouint(buf, 10, &min);
4883 mddev->sync_speed_min = min;
4887 static struct md_sysfs_entry md_sync_min =
4888 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4891 sync_max_show(struct mddev *mddev, char *page)
4893 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4894 mddev->sync_speed_max ? "local": "system");
4898 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4903 if (strncmp(buf, "system", 6)==0) {
4906 rv = kstrtouint(buf, 10, &max);
4912 mddev->sync_speed_max = max;
4916 static struct md_sysfs_entry md_sync_max =
4917 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4920 degraded_show(struct mddev *mddev, char *page)
4922 return sprintf(page, "%d\n", mddev->degraded);
4924 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4927 sync_force_parallel_show(struct mddev *mddev, char *page)
4929 return sprintf(page, "%d\n", mddev->parallel_resync);
4933 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4937 if (kstrtol(buf, 10, &n))
4940 if (n != 0 && n != 1)
4943 mddev->parallel_resync = n;
4945 if (mddev->sync_thread)
4946 wake_up(&resync_wait);
4951 /* force parallel resync, even with shared block devices */
4952 static struct md_sysfs_entry md_sync_force_parallel =
4953 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4954 sync_force_parallel_show, sync_force_parallel_store);
4957 sync_speed_show(struct mddev *mddev, char *page)
4959 unsigned long resync, dt, db;
4960 if (mddev->curr_resync == MD_RESYNC_NONE)
4961 return sprintf(page, "none\n");
4962 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4963 dt = (jiffies - mddev->resync_mark) / HZ;
4965 db = resync - mddev->resync_mark_cnt;
4966 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4969 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4972 sync_completed_show(struct mddev *mddev, char *page)
4974 unsigned long long max_sectors, resync;
4976 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4977 return sprintf(page, "none\n");
4979 if (mddev->curr_resync == MD_RESYNC_YIELDED ||
4980 mddev->curr_resync == MD_RESYNC_DELAYED)
4981 return sprintf(page, "delayed\n");
4983 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4984 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4985 max_sectors = mddev->resync_max_sectors;
4987 max_sectors = mddev->dev_sectors;
4989 resync = mddev->curr_resync_completed;
4990 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4993 static struct md_sysfs_entry md_sync_completed =
4994 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4997 min_sync_show(struct mddev *mddev, char *page)
4999 return sprintf(page, "%llu\n",
5000 (unsigned long long)mddev->resync_min);
5003 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5005 unsigned long long min;
5008 if (kstrtoull(buf, 10, &min))
5011 spin_lock(&mddev->lock);
5013 if (min > mddev->resync_max)
5017 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5020 /* Round down to multiple of 4K for safety */
5021 mddev->resync_min = round_down(min, 8);
5025 spin_unlock(&mddev->lock);
5029 static struct md_sysfs_entry md_min_sync =
5030 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5033 max_sync_show(struct mddev *mddev, char *page)
5035 if (mddev->resync_max == MaxSector)
5036 return sprintf(page, "max\n");
5038 return sprintf(page, "%llu\n",
5039 (unsigned long long)mddev->resync_max);
5042 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5045 spin_lock(&mddev->lock);
5046 if (strncmp(buf, "max", 3) == 0)
5047 mddev->resync_max = MaxSector;
5049 unsigned long long max;
5053 if (kstrtoull(buf, 10, &max))
5055 if (max < mddev->resync_min)
5059 if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5060 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5063 /* Must be a multiple of chunk_size */
5064 chunk = mddev->chunk_sectors;
5066 sector_t temp = max;
5069 if (sector_div(temp, chunk))
5072 mddev->resync_max = max;
5074 wake_up(&mddev->recovery_wait);
5077 spin_unlock(&mddev->lock);
5081 static struct md_sysfs_entry md_max_sync =
5082 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5085 suspend_lo_show(struct mddev *mddev, char *page)
5087 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5091 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5093 unsigned long long new;
5096 err = kstrtoull(buf, 10, &new);
5099 if (new != (sector_t)new)
5102 err = mddev_lock(mddev);
5106 if (mddev->pers == NULL ||
5107 mddev->pers->quiesce == NULL)
5109 mddev_suspend(mddev);
5110 mddev->suspend_lo = new;
5111 mddev_resume(mddev);
5115 mddev_unlock(mddev);
5118 static struct md_sysfs_entry md_suspend_lo =
5119 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5122 suspend_hi_show(struct mddev *mddev, char *page)
5124 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5128 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5130 unsigned long long new;
5133 err = kstrtoull(buf, 10, &new);
5136 if (new != (sector_t)new)
5139 err = mddev_lock(mddev);
5143 if (mddev->pers == NULL)
5146 mddev_suspend(mddev);
5147 mddev->suspend_hi = new;
5148 mddev_resume(mddev);
5152 mddev_unlock(mddev);
5155 static struct md_sysfs_entry md_suspend_hi =
5156 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5159 reshape_position_show(struct mddev *mddev, char *page)
5161 if (mddev->reshape_position != MaxSector)
5162 return sprintf(page, "%llu\n",
5163 (unsigned long long)mddev->reshape_position);
5164 strcpy(page, "none\n");
5169 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5171 struct md_rdev *rdev;
5172 unsigned long long new;
5175 err = kstrtoull(buf, 10, &new);
5178 if (new != (sector_t)new)
5180 err = mddev_lock(mddev);
5186 mddev->reshape_position = new;
5187 mddev->delta_disks = 0;
5188 mddev->reshape_backwards = 0;
5189 mddev->new_level = mddev->level;
5190 mddev->new_layout = mddev->layout;
5191 mddev->new_chunk_sectors = mddev->chunk_sectors;
5192 rdev_for_each(rdev, mddev)
5193 rdev->new_data_offset = rdev->data_offset;
5196 mddev_unlock(mddev);
5200 static struct md_sysfs_entry md_reshape_position =
5201 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5202 reshape_position_store);
5205 reshape_direction_show(struct mddev *mddev, char *page)
5207 return sprintf(page, "%s\n",
5208 mddev->reshape_backwards ? "backwards" : "forwards");
5212 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5217 if (cmd_match(buf, "forwards"))
5219 else if (cmd_match(buf, "backwards"))
5223 if (mddev->reshape_backwards == backwards)
5226 err = mddev_lock(mddev);
5229 /* check if we are allowed to change */
5230 if (mddev->delta_disks)
5232 else if (mddev->persistent &&
5233 mddev->major_version == 0)
5236 mddev->reshape_backwards = backwards;
5237 mddev_unlock(mddev);
5241 static struct md_sysfs_entry md_reshape_direction =
5242 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5243 reshape_direction_store);
5246 array_size_show(struct mddev *mddev, char *page)
5248 if (mddev->external_size)
5249 return sprintf(page, "%llu\n",
5250 (unsigned long long)mddev->array_sectors/2);
5252 return sprintf(page, "default\n");
5256 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5261 err = mddev_lock(mddev);
5265 /* cluster raid doesn't support change array_sectors */
5266 if (mddev_is_clustered(mddev)) {
5267 mddev_unlock(mddev);
5271 if (strncmp(buf, "default", 7) == 0) {
5273 sectors = mddev->pers->size(mddev, 0, 0);
5275 sectors = mddev->array_sectors;
5277 mddev->external_size = 0;
5279 if (strict_blocks_to_sectors(buf, §ors) < 0)
5281 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5284 mddev->external_size = 1;
5288 mddev->array_sectors = sectors;
5290 set_capacity_and_notify(mddev->gendisk,
5291 mddev->array_sectors);
5293 mddev_unlock(mddev);
5297 static struct md_sysfs_entry md_array_size =
5298 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5302 consistency_policy_show(struct mddev *mddev, char *page)
5306 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5307 ret = sprintf(page, "journal\n");
5308 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5309 ret = sprintf(page, "ppl\n");
5310 } else if (mddev->bitmap) {
5311 ret = sprintf(page, "bitmap\n");
5312 } else if (mddev->pers) {
5313 if (mddev->pers->sync_request)
5314 ret = sprintf(page, "resync\n");
5316 ret = sprintf(page, "none\n");
5318 ret = sprintf(page, "unknown\n");
5325 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5330 if (mddev->pers->change_consistency_policy)
5331 err = mddev->pers->change_consistency_policy(mddev, buf);
5334 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5335 set_bit(MD_HAS_PPL, &mddev->flags);
5340 return err ? err : len;
5343 static struct md_sysfs_entry md_consistency_policy =
5344 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5345 consistency_policy_store);
5347 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5349 return sprintf(page, "%d\n", mddev->fail_last_dev);
5353 * Setting fail_last_dev to true to allow last device to be forcibly removed
5354 * from RAID1/RAID10.
5357 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5362 ret = kstrtobool(buf, &value);
5366 if (value != mddev->fail_last_dev)
5367 mddev->fail_last_dev = value;
5371 static struct md_sysfs_entry md_fail_last_dev =
5372 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5373 fail_last_dev_store);
5375 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5377 if (mddev->pers == NULL || (mddev->pers->level != 1))
5378 return sprintf(page, "n/a\n");
5380 return sprintf(page, "%d\n", mddev->serialize_policy);
5384 * Setting serialize_policy to true to enforce write IO is not reordered
5388 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5393 err = kstrtobool(buf, &value);
5397 if (value == mddev->serialize_policy)
5400 err = mddev_lock(mddev);
5403 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5404 pr_err("md: serialize_policy is only effective for raid1\n");
5409 mddev_suspend(mddev);
5411 mddev_create_serial_pool(mddev, NULL, true);
5413 mddev_destroy_serial_pool(mddev, NULL, true);
5414 mddev->serialize_policy = value;
5415 mddev_resume(mddev);
5417 mddev_unlock(mddev);
5421 static struct md_sysfs_entry md_serialize_policy =
5422 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5423 serialize_policy_store);
5426 static struct attribute *md_default_attrs[] = {
5429 &md_raid_disks.attr,
5431 &md_chunk_size.attr,
5433 &md_resync_start.attr,
5435 &md_new_device.attr,
5436 &md_safe_delay.attr,
5437 &md_array_state.attr,
5438 &md_reshape_position.attr,
5439 &md_reshape_direction.attr,
5440 &md_array_size.attr,
5441 &max_corr_read_errors.attr,
5442 &md_consistency_policy.attr,
5443 &md_fail_last_dev.attr,
5444 &md_serialize_policy.attr,
5448 static const struct attribute_group md_default_group = {
5449 .attrs = md_default_attrs,
5452 static struct attribute *md_redundancy_attrs[] = {
5454 &md_last_scan_mode.attr,
5455 &md_mismatches.attr,
5458 &md_sync_speed.attr,
5459 &md_sync_force_parallel.attr,
5460 &md_sync_completed.attr,
5463 &md_suspend_lo.attr,
5464 &md_suspend_hi.attr,
5469 static const struct attribute_group md_redundancy_group = {
5471 .attrs = md_redundancy_attrs,
5474 static const struct attribute_group *md_attr_groups[] = {
5481 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5483 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5484 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5489 spin_lock(&all_mddevs_lock);
5490 if (!mddev_get(mddev)) {
5491 spin_unlock(&all_mddevs_lock);
5494 spin_unlock(&all_mddevs_lock);
5496 rv = entry->show(mddev, page);
5502 md_attr_store(struct kobject *kobj, struct attribute *attr,
5503 const char *page, size_t length)
5505 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5506 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5511 if (!capable(CAP_SYS_ADMIN))
5513 spin_lock(&all_mddevs_lock);
5514 if (!mddev_get(mddev)) {
5515 spin_unlock(&all_mddevs_lock);
5518 spin_unlock(&all_mddevs_lock);
5519 rv = entry->store(mddev, page, length);
5524 static void md_kobj_release(struct kobject *ko)
5526 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5528 if (mddev->sysfs_state)
5529 sysfs_put(mddev->sysfs_state);
5530 if (mddev->sysfs_level)
5531 sysfs_put(mddev->sysfs_level);
5533 del_gendisk(mddev->gendisk);
5534 put_disk(mddev->gendisk);
5537 static const struct sysfs_ops md_sysfs_ops = {
5538 .show = md_attr_show,
5539 .store = md_attr_store,
5541 static const struct kobj_type md_ktype = {
5542 .release = md_kobj_release,
5543 .sysfs_ops = &md_sysfs_ops,
5544 .default_groups = md_attr_groups,
5549 static void mddev_delayed_delete(struct work_struct *ws)
5551 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5553 kobject_put(&mddev->kobj);
5556 static void no_op(struct percpu_ref *r) {}
5558 int mddev_init_writes_pending(struct mddev *mddev)
5560 if (mddev->writes_pending.percpu_count_ptr)
5562 if (percpu_ref_init(&mddev->writes_pending, no_op,
5563 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5565 /* We want to start with the refcount at zero */
5566 percpu_ref_put(&mddev->writes_pending);
5569 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5571 struct mddev *md_alloc(dev_t dev, char *name)
5574 * If dev is zero, name is the name of a device to allocate with
5575 * an arbitrary minor number. It will be "md_???"
5576 * If dev is non-zero it must be a device number with a MAJOR of
5577 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5578 * the device is being created by opening a node in /dev.
5579 * If "name" is not NULL, the device is being created by
5580 * writing to /sys/module/md_mod/parameters/new_array.
5582 static DEFINE_MUTEX(disks_mutex);
5583 struct mddev *mddev;
5584 struct gendisk *disk;
5591 * Wait for any previous instance of this device to be completely
5592 * removed (mddev_delayed_delete).
5594 flush_workqueue(md_misc_wq);
5595 flush_workqueue(md_rdev_misc_wq);
5597 mutex_lock(&disks_mutex);
5598 mddev = mddev_alloc(dev);
5599 if (IS_ERR(mddev)) {
5600 error = PTR_ERR(mddev);
5604 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5605 shift = partitioned ? MdpMinorShift : 0;
5606 unit = MINOR(mddev->unit) >> shift;
5609 /* Need to ensure that 'name' is not a duplicate.
5611 struct mddev *mddev2;
5612 spin_lock(&all_mddevs_lock);
5614 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5615 if (mddev2->gendisk &&
5616 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5617 spin_unlock(&all_mddevs_lock);
5619 goto out_free_mddev;
5621 spin_unlock(&all_mddevs_lock);
5625 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5627 mddev->hold_active = UNTIL_STOP;
5630 disk = blk_alloc_disk(NUMA_NO_NODE);
5632 goto out_free_mddev;
5634 disk->major = MAJOR(mddev->unit);
5635 disk->first_minor = unit << shift;
5636 disk->minors = 1 << shift;
5638 strcpy(disk->disk_name, name);
5639 else if (partitioned)
5640 sprintf(disk->disk_name, "md_d%d", unit);
5642 sprintf(disk->disk_name, "md%d", unit);
5643 disk->fops = &md_fops;
5644 disk->private_data = mddev;
5646 mddev->queue = disk->queue;
5647 blk_set_stacking_limits(&mddev->queue->limits);
5648 blk_queue_write_cache(mddev->queue, true, true);
5649 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5650 mddev->gendisk = disk;
5651 error = add_disk(disk);
5655 kobject_init(&mddev->kobj, &md_ktype);
5656 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5659 * The disk is already live at this point. Clear the hold flag
5660 * and let mddev_put take care of the deletion, as it isn't any
5661 * different from a normal close on last release now.
5663 mddev->hold_active = 0;
5664 mutex_unlock(&disks_mutex);
5666 return ERR_PTR(error);
5669 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5670 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5671 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5672 mutex_unlock(&disks_mutex);
5680 mutex_unlock(&disks_mutex);
5681 return ERR_PTR(error);
5684 static int md_alloc_and_put(dev_t dev, char *name)
5686 struct mddev *mddev = md_alloc(dev, name);
5689 return PTR_ERR(mddev);
5694 static void md_probe(dev_t dev)
5696 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5699 md_alloc_and_put(dev, NULL);
5702 static int add_named_array(const char *val, const struct kernel_param *kp)
5705 * val must be "md_*" or "mdNNN".
5706 * For "md_*" we allocate an array with a large free minor number, and
5707 * set the name to val. val must not already be an active name.
5708 * For "mdNNN" we allocate an array with the minor number NNN
5709 * which must not already be in use.
5711 int len = strlen(val);
5712 char buf[DISK_NAME_LEN];
5713 unsigned long devnum;
5715 while (len && val[len-1] == '\n')
5717 if (len >= DISK_NAME_LEN)
5719 strscpy(buf, val, len+1);
5720 if (strncmp(buf, "md_", 3) == 0)
5721 return md_alloc_and_put(0, buf);
5722 if (strncmp(buf, "md", 2) == 0 &&
5724 kstrtoul(buf+2, 10, &devnum) == 0 &&
5725 devnum <= MINORMASK)
5726 return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5731 static void md_safemode_timeout(struct timer_list *t)
5733 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5735 mddev->safemode = 1;
5736 if (mddev->external)
5737 sysfs_notify_dirent_safe(mddev->sysfs_state);
5739 md_wakeup_thread(mddev->thread);
5742 static int start_dirty_degraded;
5743 static void active_io_release(struct percpu_ref *ref)
5745 struct mddev *mddev = container_of(ref, struct mddev, active_io);
5747 wake_up(&mddev->sb_wait);
5750 int md_run(struct mddev *mddev)
5753 struct md_rdev *rdev;
5754 struct md_personality *pers;
5757 if (list_empty(&mddev->disks))
5758 /* cannot run an array with no devices.. */
5763 /* Cannot run until previous stop completes properly */
5764 if (mddev->sysfs_active)
5768 * Analyze all RAID superblock(s)
5770 if (!mddev->raid_disks) {
5771 if (!mddev->persistent)
5773 err = analyze_sbs(mddev);
5778 if (mddev->level != LEVEL_NONE)
5779 request_module("md-level-%d", mddev->level);
5780 else if (mddev->clevel[0])
5781 request_module("md-%s", mddev->clevel);
5784 * Drop all container device buffers, from now on
5785 * the only valid external interface is through the md
5788 mddev->has_superblocks = false;
5789 rdev_for_each(rdev, mddev) {
5790 if (test_bit(Faulty, &rdev->flags))
5792 sync_blockdev(rdev->bdev);
5793 invalidate_bdev(rdev->bdev);
5794 if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
5795 mddev->ro = MD_RDONLY;
5797 set_disk_ro(mddev->gendisk, 1);
5801 mddev->has_superblocks = true;
5803 /* perform some consistency tests on the device.
5804 * We don't want the data to overlap the metadata,
5805 * Internal Bitmap issues have been handled elsewhere.
5807 if (rdev->meta_bdev) {
5808 /* Nothing to check */;
5809 } else if (rdev->data_offset < rdev->sb_start) {
5810 if (mddev->dev_sectors &&
5811 rdev->data_offset + mddev->dev_sectors
5813 pr_warn("md: %s: data overlaps metadata\n",
5818 if (rdev->sb_start + rdev->sb_size/512
5819 > rdev->data_offset) {
5820 pr_warn("md: %s: metadata overlaps data\n",
5825 sysfs_notify_dirent_safe(rdev->sysfs_state);
5826 nowait = nowait && bdev_nowait(rdev->bdev);
5829 err = percpu_ref_init(&mddev->active_io, active_io_release,
5830 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL);
5834 if (!bioset_initialized(&mddev->bio_set)) {
5835 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5837 goto exit_active_io;
5839 if (!bioset_initialized(&mddev->sync_set)) {
5840 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5845 spin_lock(&pers_lock);
5846 pers = find_pers(mddev->level, mddev->clevel);
5847 if (!pers || !try_module_get(pers->owner)) {
5848 spin_unlock(&pers_lock);
5849 if (mddev->level != LEVEL_NONE)
5850 pr_warn("md: personality for level %d is not loaded!\n",
5853 pr_warn("md: personality for level %s is not loaded!\n",
5858 spin_unlock(&pers_lock);
5859 if (mddev->level != pers->level) {
5860 mddev->level = pers->level;
5861 mddev->new_level = pers->level;
5863 strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5865 if (mddev->reshape_position != MaxSector &&
5866 pers->start_reshape == NULL) {
5867 /* This personality cannot handle reshaping... */
5868 module_put(pers->owner);
5873 if (pers->sync_request) {
5874 /* Warn if this is a potentially silly
5877 struct md_rdev *rdev2;
5880 rdev_for_each(rdev, mddev)
5881 rdev_for_each(rdev2, mddev) {
5883 rdev->bdev->bd_disk ==
5884 rdev2->bdev->bd_disk) {
5885 pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
5894 pr_warn("True protection against single-disk failure might be compromised.\n");
5897 mddev->recovery = 0;
5898 /* may be over-ridden by personality */
5899 mddev->resync_max_sectors = mddev->dev_sectors;
5901 mddev->ok_start_degraded = start_dirty_degraded;
5903 if (start_readonly && md_is_rdwr(mddev))
5904 mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
5906 err = pers->run(mddev);
5908 pr_warn("md: pers->run() failed ...\n");
5909 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5910 WARN_ONCE(!mddev->external_size,
5911 "%s: default size too small, but 'external_size' not in effect?\n",
5913 pr_warn("md: invalid array_size %llu > default size %llu\n",
5914 (unsigned long long)mddev->array_sectors / 2,
5915 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5918 if (err == 0 && pers->sync_request &&
5919 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5920 struct bitmap *bitmap;
5922 bitmap = md_bitmap_create(mddev, -1);
5923 if (IS_ERR(bitmap)) {
5924 err = PTR_ERR(bitmap);
5925 pr_warn("%s: failed to create bitmap (%d)\n",
5926 mdname(mddev), err);
5928 mddev->bitmap = bitmap;
5934 if (mddev->bitmap_info.max_write_behind > 0) {
5935 bool create_pool = false;
5937 rdev_for_each(rdev, mddev) {
5938 if (test_bit(WriteMostly, &rdev->flags) &&
5939 rdev_init_serial(rdev))
5942 if (create_pool && mddev->serial_info_pool == NULL) {
5943 mddev->serial_info_pool =
5944 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5945 sizeof(struct serial_info));
5946 if (!mddev->serial_info_pool) {
5956 rdev_for_each(rdev, mddev) {
5957 if (rdev->raid_disk >= 0 && !bdev_nonrot(rdev->bdev)) {
5962 if (mddev->degraded)
5965 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5967 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5968 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
5970 /* Set the NOWAIT flags if all underlying devices support it */
5972 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, mddev->queue);
5974 if (pers->sync_request) {
5975 if (mddev->kobj.sd &&
5976 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5977 pr_warn("md: cannot register extra attributes for %s\n",
5979 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5980 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
5981 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
5982 } else if (mddev->ro == MD_AUTO_READ)
5983 mddev->ro = MD_RDWR;
5985 atomic_set(&mddev->max_corr_read_errors,
5986 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5987 mddev->safemode = 0;
5988 if (mddev_is_clustered(mddev))
5989 mddev->safemode_delay = 0;
5991 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
5994 spin_lock(&mddev->lock);
5996 spin_unlock(&mddev->lock);
5997 rdev_for_each(rdev, mddev)
5998 if (rdev->raid_disk >= 0)
5999 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6001 if (mddev->degraded && md_is_rdwr(mddev))
6002 /* This ensures that recovering status is reported immediately
6003 * via sysfs - until a lack of spares is confirmed.
6005 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6006 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6008 if (mddev->sb_flags)
6009 md_update_sb(mddev, 0);
6015 mddev_detach(mddev);
6017 pers->free(mddev, mddev->private);
6018 mddev->private = NULL;
6019 module_put(pers->owner);
6020 md_bitmap_destroy(mddev);
6022 bioset_exit(&mddev->sync_set);
6024 bioset_exit(&mddev->bio_set);
6026 percpu_ref_exit(&mddev->active_io);
6029 EXPORT_SYMBOL_GPL(md_run);
6031 int do_md_run(struct mddev *mddev)
6035 set_bit(MD_NOT_READY, &mddev->flags);
6036 err = md_run(mddev);
6039 err = md_bitmap_load(mddev);
6041 md_bitmap_destroy(mddev);
6045 if (mddev_is_clustered(mddev))
6046 md_allow_write(mddev);
6048 /* run start up tasks that require md_thread */
6051 md_wakeup_thread(mddev->thread);
6052 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6054 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6055 clear_bit(MD_NOT_READY, &mddev->flags);
6057 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6058 sysfs_notify_dirent_safe(mddev->sysfs_state);
6059 sysfs_notify_dirent_safe(mddev->sysfs_action);
6060 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6062 clear_bit(MD_NOT_READY, &mddev->flags);
6066 int md_start(struct mddev *mddev)
6070 if (mddev->pers->start) {
6071 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6072 md_wakeup_thread(mddev->thread);
6073 ret = mddev->pers->start(mddev);
6074 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6075 md_wakeup_thread(mddev->sync_thread);
6079 EXPORT_SYMBOL_GPL(md_start);
6081 static int restart_array(struct mddev *mddev)
6083 struct gendisk *disk = mddev->gendisk;
6084 struct md_rdev *rdev;
6085 bool has_journal = false;
6086 bool has_readonly = false;
6088 /* Complain if it has no devices */
6089 if (list_empty(&mddev->disks))
6093 if (md_is_rdwr(mddev))
6097 rdev_for_each_rcu(rdev, mddev) {
6098 if (test_bit(Journal, &rdev->flags) &&
6099 !test_bit(Faulty, &rdev->flags))
6101 if (rdev_read_only(rdev))
6102 has_readonly = true;
6105 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6106 /* Don't restart rw with journal missing/faulty */
6111 mddev->safemode = 0;
6112 mddev->ro = MD_RDWR;
6113 set_disk_ro(disk, 0);
6114 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6115 /* Kick recovery or resync if necessary */
6116 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6117 md_wakeup_thread(mddev->thread);
6118 md_wakeup_thread(mddev->sync_thread);
6119 sysfs_notify_dirent_safe(mddev->sysfs_state);
6123 static void md_clean(struct mddev *mddev)
6125 mddev->array_sectors = 0;
6126 mddev->external_size = 0;
6127 mddev->dev_sectors = 0;
6128 mddev->raid_disks = 0;
6129 mddev->recovery_cp = 0;
6130 mddev->resync_min = 0;
6131 mddev->resync_max = MaxSector;
6132 mddev->reshape_position = MaxSector;
6133 mddev->external = 0;
6134 mddev->persistent = 0;
6135 mddev->level = LEVEL_NONE;
6136 mddev->clevel[0] = 0;
6138 mddev->sb_flags = 0;
6139 mddev->ro = MD_RDWR;
6140 mddev->metadata_type[0] = 0;
6141 mddev->chunk_sectors = 0;
6142 mddev->ctime = mddev->utime = 0;
6144 mddev->max_disks = 0;
6146 mddev->can_decrease_events = 0;
6147 mddev->delta_disks = 0;
6148 mddev->reshape_backwards = 0;
6149 mddev->new_level = LEVEL_NONE;
6150 mddev->new_layout = 0;
6151 mddev->new_chunk_sectors = 0;
6152 mddev->curr_resync = MD_RESYNC_NONE;
6153 atomic64_set(&mddev->resync_mismatches, 0);
6154 mddev->suspend_lo = mddev->suspend_hi = 0;
6155 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6156 mddev->recovery = 0;
6159 mddev->degraded = 0;
6160 mddev->safemode = 0;
6161 mddev->private = NULL;
6162 mddev->cluster_info = NULL;
6163 mddev->bitmap_info.offset = 0;
6164 mddev->bitmap_info.default_offset = 0;
6165 mddev->bitmap_info.default_space = 0;
6166 mddev->bitmap_info.chunksize = 0;
6167 mddev->bitmap_info.daemon_sleep = 0;
6168 mddev->bitmap_info.max_write_behind = 0;
6169 mddev->bitmap_info.nodes = 0;
6172 static void __md_stop_writes(struct mddev *mddev)
6174 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6175 if (work_pending(&mddev->del_work))
6176 flush_workqueue(md_misc_wq);
6177 if (mddev->sync_thread) {
6178 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6179 md_unregister_thread(&mddev->sync_thread);
6180 md_reap_sync_thread(mddev);
6183 del_timer_sync(&mddev->safemode_timer);
6185 if (mddev->pers && mddev->pers->quiesce) {
6186 mddev->pers->quiesce(mddev, 1);
6187 mddev->pers->quiesce(mddev, 0);
6189 md_bitmap_flush(mddev);
6191 if (md_is_rdwr(mddev) &&
6192 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6194 /* mark array as shutdown cleanly */
6195 if (!mddev_is_clustered(mddev))
6197 md_update_sb(mddev, 1);
6199 /* disable policy to guarantee rdevs free resources for serialization */
6200 mddev->serialize_policy = 0;
6201 mddev_destroy_serial_pool(mddev, NULL, true);
6204 void md_stop_writes(struct mddev *mddev)
6206 mddev_lock_nointr(mddev);
6207 __md_stop_writes(mddev);
6208 mddev_unlock(mddev);
6210 EXPORT_SYMBOL_GPL(md_stop_writes);
6212 static void mddev_detach(struct mddev *mddev)
6214 md_bitmap_wait_behind_writes(mddev);
6215 if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6216 mddev->pers->quiesce(mddev, 1);
6217 mddev->pers->quiesce(mddev, 0);
6219 md_unregister_thread(&mddev->thread);
6221 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6224 static void __md_stop(struct mddev *mddev)
6226 struct md_personality *pers = mddev->pers;
6227 md_bitmap_destroy(mddev);
6228 mddev_detach(mddev);
6229 /* Ensure ->event_work is done */
6230 if (mddev->event_work.func)
6231 flush_workqueue(md_misc_wq);
6232 spin_lock(&mddev->lock);
6234 spin_unlock(&mddev->lock);
6236 pers->free(mddev, mddev->private);
6237 mddev->private = NULL;
6238 if (pers->sync_request && mddev->to_remove == NULL)
6239 mddev->to_remove = &md_redundancy_group;
6240 module_put(pers->owner);
6241 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6243 percpu_ref_exit(&mddev->active_io);
6244 bioset_exit(&mddev->bio_set);
6245 bioset_exit(&mddev->sync_set);
6248 void md_stop(struct mddev *mddev)
6250 /* stop the array and free an attached data structures.
6251 * This is called from dm-raid
6253 __md_stop_writes(mddev);
6255 percpu_ref_exit(&mddev->writes_pending);
6258 EXPORT_SYMBOL_GPL(md_stop);
6260 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6265 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6267 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6268 md_wakeup_thread(mddev->thread);
6270 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6271 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6272 if (mddev->sync_thread)
6273 /* Thread might be blocked waiting for metadata update
6274 * which will now never happen */
6275 wake_up_process(mddev->sync_thread->tsk);
6277 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6279 mddev_unlock(mddev);
6280 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6282 wait_event(mddev->sb_wait,
6283 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6284 mddev_lock_nointr(mddev);
6286 mutex_lock(&mddev->open_mutex);
6287 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6288 mddev->sync_thread ||
6289 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6290 pr_warn("md: %s still in use.\n",mdname(mddev));
6292 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6293 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6294 md_wakeup_thread(mddev->thread);
6300 __md_stop_writes(mddev);
6303 if (mddev->ro == MD_RDONLY)
6305 mddev->ro = MD_RDONLY;
6306 set_disk_ro(mddev->gendisk, 1);
6307 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6308 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6309 md_wakeup_thread(mddev->thread);
6310 sysfs_notify_dirent_safe(mddev->sysfs_state);
6314 mutex_unlock(&mddev->open_mutex);
6319 * 0 - completely stop and dis-assemble array
6320 * 2 - stop but do not disassemble array
6322 static int do_md_stop(struct mddev *mddev, int mode,
6323 struct block_device *bdev)
6325 struct gendisk *disk = mddev->gendisk;
6326 struct md_rdev *rdev;
6329 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6331 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6332 md_wakeup_thread(mddev->thread);
6334 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6335 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6336 if (mddev->sync_thread)
6337 /* Thread might be blocked waiting for metadata update
6338 * which will now never happen */
6339 wake_up_process(mddev->sync_thread->tsk);
6341 mddev_unlock(mddev);
6342 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6343 !test_bit(MD_RECOVERY_RUNNING,
6344 &mddev->recovery)));
6345 mddev_lock_nointr(mddev);
6347 mutex_lock(&mddev->open_mutex);
6348 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6349 mddev->sysfs_active ||
6350 mddev->sync_thread ||
6351 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6352 pr_warn("md: %s still in use.\n",mdname(mddev));
6353 mutex_unlock(&mddev->open_mutex);
6355 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6356 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6357 md_wakeup_thread(mddev->thread);
6362 if (!md_is_rdwr(mddev))
6363 set_disk_ro(disk, 0);
6365 __md_stop_writes(mddev);
6368 /* tell userspace to handle 'inactive' */
6369 sysfs_notify_dirent_safe(mddev->sysfs_state);
6371 rdev_for_each(rdev, mddev)
6372 if (rdev->raid_disk >= 0)
6373 sysfs_unlink_rdev(mddev, rdev);
6375 set_capacity_and_notify(disk, 0);
6376 mutex_unlock(&mddev->open_mutex);
6379 if (!md_is_rdwr(mddev))
6380 mddev->ro = MD_RDWR;
6382 mutex_unlock(&mddev->open_mutex);
6384 * Free resources if final stop
6387 pr_info("md: %s stopped.\n", mdname(mddev));
6389 if (mddev->bitmap_info.file) {
6390 struct file *f = mddev->bitmap_info.file;
6391 spin_lock(&mddev->lock);
6392 mddev->bitmap_info.file = NULL;
6393 spin_unlock(&mddev->lock);
6396 mddev->bitmap_info.offset = 0;
6398 export_array(mddev);
6401 if (mddev->hold_active == UNTIL_STOP)
6402 mddev->hold_active = 0;
6405 sysfs_notify_dirent_safe(mddev->sysfs_state);
6410 static void autorun_array(struct mddev *mddev)
6412 struct md_rdev *rdev;
6415 if (list_empty(&mddev->disks))
6418 pr_info("md: running: ");
6420 rdev_for_each(rdev, mddev) {
6421 pr_cont("<%pg>", rdev->bdev);
6425 err = do_md_run(mddev);
6427 pr_warn("md: do_md_run() returned %d\n", err);
6428 do_md_stop(mddev, 0, NULL);
6433 * lets try to run arrays based on all disks that have arrived
6434 * until now. (those are in pending_raid_disks)
6436 * the method: pick the first pending disk, collect all disks with
6437 * the same UUID, remove all from the pending list and put them into
6438 * the 'same_array' list. Then order this list based on superblock
6439 * update time (freshest comes first), kick out 'old' disks and
6440 * compare superblocks. If everything's fine then run it.
6442 * If "unit" is allocated, then bump its reference count
6444 static void autorun_devices(int part)
6446 struct md_rdev *rdev0, *rdev, *tmp;
6447 struct mddev *mddev;
6449 pr_info("md: autorun ...\n");
6450 while (!list_empty(&pending_raid_disks)) {
6453 LIST_HEAD(candidates);
6454 rdev0 = list_entry(pending_raid_disks.next,
6455 struct md_rdev, same_set);
6457 pr_debug("md: considering %pg ...\n", rdev0->bdev);
6458 INIT_LIST_HEAD(&candidates);
6459 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6460 if (super_90_load(rdev, rdev0, 0) >= 0) {
6461 pr_debug("md: adding %pg ...\n",
6463 list_move(&rdev->same_set, &candidates);
6466 * now we have a set of devices, with all of them having
6467 * mostly sane superblocks. It's time to allocate the
6471 dev = MKDEV(mdp_major,
6472 rdev0->preferred_minor << MdpMinorShift);
6473 unit = MINOR(dev) >> MdpMinorShift;
6475 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6478 if (rdev0->preferred_minor != unit) {
6479 pr_warn("md: unit number in %pg is bad: %d\n",
6480 rdev0->bdev, rdev0->preferred_minor);
6484 mddev = md_alloc(dev, NULL);
6488 if (mddev_lock(mddev))
6489 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6490 else if (mddev->raid_disks || mddev->major_version
6491 || !list_empty(&mddev->disks)) {
6492 pr_warn("md: %s already running, cannot run %pg\n",
6493 mdname(mddev), rdev0->bdev);
6494 mddev_unlock(mddev);
6496 pr_debug("md: created %s\n", mdname(mddev));
6497 mddev->persistent = 1;
6498 rdev_for_each_list(rdev, tmp, &candidates) {
6499 list_del_init(&rdev->same_set);
6500 if (bind_rdev_to_array(rdev, mddev))
6503 autorun_array(mddev);
6504 mddev_unlock(mddev);
6506 /* on success, candidates will be empty, on error
6509 rdev_for_each_list(rdev, tmp, &candidates) {
6510 list_del_init(&rdev->same_set);
6515 pr_info("md: ... autorun DONE.\n");
6517 #endif /* !MODULE */
6519 static int get_version(void __user *arg)
6523 ver.major = MD_MAJOR_VERSION;
6524 ver.minor = MD_MINOR_VERSION;
6525 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6527 if (copy_to_user(arg, &ver, sizeof(ver)))
6533 static int get_array_info(struct mddev *mddev, void __user *arg)
6535 mdu_array_info_t info;
6536 int nr,working,insync,failed,spare;
6537 struct md_rdev *rdev;
6539 nr = working = insync = failed = spare = 0;
6541 rdev_for_each_rcu(rdev, mddev) {
6543 if (test_bit(Faulty, &rdev->flags))
6547 if (test_bit(In_sync, &rdev->flags))
6549 else if (test_bit(Journal, &rdev->flags))
6550 /* TODO: add journal count to md_u.h */
6558 info.major_version = mddev->major_version;
6559 info.minor_version = mddev->minor_version;
6560 info.patch_version = MD_PATCHLEVEL_VERSION;
6561 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6562 info.level = mddev->level;
6563 info.size = mddev->dev_sectors / 2;
6564 if (info.size != mddev->dev_sectors / 2) /* overflow */
6567 info.raid_disks = mddev->raid_disks;
6568 info.md_minor = mddev->md_minor;
6569 info.not_persistent= !mddev->persistent;
6571 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6574 info.state = (1<<MD_SB_CLEAN);
6575 if (mddev->bitmap && mddev->bitmap_info.offset)
6576 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6577 if (mddev_is_clustered(mddev))
6578 info.state |= (1<<MD_SB_CLUSTERED);
6579 info.active_disks = insync;
6580 info.working_disks = working;
6581 info.failed_disks = failed;
6582 info.spare_disks = spare;
6584 info.layout = mddev->layout;
6585 info.chunk_size = mddev->chunk_sectors << 9;
6587 if (copy_to_user(arg, &info, sizeof(info)))
6593 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6595 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6599 file = kzalloc(sizeof(*file), GFP_NOIO);
6604 spin_lock(&mddev->lock);
6605 /* bitmap enabled */
6606 if (mddev->bitmap_info.file) {
6607 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6608 sizeof(file->pathname));
6612 memmove(file->pathname, ptr,
6613 sizeof(file->pathname)-(ptr-file->pathname));
6615 spin_unlock(&mddev->lock);
6618 copy_to_user(arg, file, sizeof(*file)))
6625 static int get_disk_info(struct mddev *mddev, void __user * arg)
6627 mdu_disk_info_t info;
6628 struct md_rdev *rdev;
6630 if (copy_from_user(&info, arg, sizeof(info)))
6634 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6636 info.major = MAJOR(rdev->bdev->bd_dev);
6637 info.minor = MINOR(rdev->bdev->bd_dev);
6638 info.raid_disk = rdev->raid_disk;
6640 if (test_bit(Faulty, &rdev->flags))
6641 info.state |= (1<<MD_DISK_FAULTY);
6642 else if (test_bit(In_sync, &rdev->flags)) {
6643 info.state |= (1<<MD_DISK_ACTIVE);
6644 info.state |= (1<<MD_DISK_SYNC);
6646 if (test_bit(Journal, &rdev->flags))
6647 info.state |= (1<<MD_DISK_JOURNAL);
6648 if (test_bit(WriteMostly, &rdev->flags))
6649 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6650 if (test_bit(FailFast, &rdev->flags))
6651 info.state |= (1<<MD_DISK_FAILFAST);
6653 info.major = info.minor = 0;
6654 info.raid_disk = -1;
6655 info.state = (1<<MD_DISK_REMOVED);
6659 if (copy_to_user(arg, &info, sizeof(info)))
6665 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6667 struct md_rdev *rdev;
6668 dev_t dev = MKDEV(info->major,info->minor);
6670 if (mddev_is_clustered(mddev) &&
6671 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6672 pr_warn("%s: Cannot add to clustered mddev.\n",
6677 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6680 if (!mddev->raid_disks) {
6682 /* expecting a device which has a superblock */
6683 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6685 pr_warn("md: md_import_device returned %ld\n",
6687 return PTR_ERR(rdev);
6689 if (!list_empty(&mddev->disks)) {
6690 struct md_rdev *rdev0
6691 = list_entry(mddev->disks.next,
6692 struct md_rdev, same_set);
6693 err = super_types[mddev->major_version]
6694 .load_super(rdev, rdev0, mddev->minor_version);
6696 pr_warn("md: %pg has different UUID to %pg\n",
6703 err = bind_rdev_to_array(rdev, mddev);
6710 * md_add_new_disk can be used once the array is assembled
6711 * to add "hot spares". They must already have a superblock
6716 if (!mddev->pers->hot_add_disk) {
6717 pr_warn("%s: personality does not support diskops!\n",
6721 if (mddev->persistent)
6722 rdev = md_import_device(dev, mddev->major_version,
6723 mddev->minor_version);
6725 rdev = md_import_device(dev, -1, -1);
6727 pr_warn("md: md_import_device returned %ld\n",
6729 return PTR_ERR(rdev);
6731 /* set saved_raid_disk if appropriate */
6732 if (!mddev->persistent) {
6733 if (info->state & (1<<MD_DISK_SYNC) &&
6734 info->raid_disk < mddev->raid_disks) {
6735 rdev->raid_disk = info->raid_disk;
6736 set_bit(In_sync, &rdev->flags);
6737 clear_bit(Bitmap_sync, &rdev->flags);
6739 rdev->raid_disk = -1;
6740 rdev->saved_raid_disk = rdev->raid_disk;
6742 super_types[mddev->major_version].
6743 validate_super(mddev, rdev);
6744 if ((info->state & (1<<MD_DISK_SYNC)) &&
6745 rdev->raid_disk != info->raid_disk) {
6746 /* This was a hot-add request, but events doesn't
6747 * match, so reject it.
6753 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6754 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6755 set_bit(WriteMostly, &rdev->flags);
6757 clear_bit(WriteMostly, &rdev->flags);
6758 if (info->state & (1<<MD_DISK_FAILFAST))
6759 set_bit(FailFast, &rdev->flags);
6761 clear_bit(FailFast, &rdev->flags);
6763 if (info->state & (1<<MD_DISK_JOURNAL)) {
6764 struct md_rdev *rdev2;
6765 bool has_journal = false;
6767 /* make sure no existing journal disk */
6768 rdev_for_each(rdev2, mddev) {
6769 if (test_bit(Journal, &rdev2->flags)) {
6774 if (has_journal || mddev->bitmap) {
6778 set_bit(Journal, &rdev->flags);
6781 * check whether the device shows up in other nodes
6783 if (mddev_is_clustered(mddev)) {
6784 if (info->state & (1 << MD_DISK_CANDIDATE))
6785 set_bit(Candidate, &rdev->flags);
6786 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6787 /* --add initiated by this node */
6788 err = md_cluster_ops->add_new_disk(mddev, rdev);
6796 rdev->raid_disk = -1;
6797 err = bind_rdev_to_array(rdev, mddev);
6802 if (mddev_is_clustered(mddev)) {
6803 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6805 err = md_cluster_ops->new_disk_ack(mddev,
6808 md_kick_rdev_from_array(rdev);
6812 md_cluster_ops->add_new_disk_cancel(mddev);
6814 err = add_bound_rdev(rdev);
6818 err = add_bound_rdev(rdev);
6823 /* otherwise, md_add_new_disk is only allowed
6824 * for major_version==0 superblocks
6826 if (mddev->major_version != 0) {
6827 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6831 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6833 rdev = md_import_device(dev, -1, 0);
6835 pr_warn("md: error, md_import_device() returned %ld\n",
6837 return PTR_ERR(rdev);
6839 rdev->desc_nr = info->number;
6840 if (info->raid_disk < mddev->raid_disks)
6841 rdev->raid_disk = info->raid_disk;
6843 rdev->raid_disk = -1;
6845 if (rdev->raid_disk < mddev->raid_disks)
6846 if (info->state & (1<<MD_DISK_SYNC))
6847 set_bit(In_sync, &rdev->flags);
6849 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6850 set_bit(WriteMostly, &rdev->flags);
6851 if (info->state & (1<<MD_DISK_FAILFAST))
6852 set_bit(FailFast, &rdev->flags);
6854 if (!mddev->persistent) {
6855 pr_debug("md: nonpersistent superblock ...\n");
6856 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6858 rdev->sb_start = calc_dev_sboffset(rdev);
6859 rdev->sectors = rdev->sb_start;
6861 err = bind_rdev_to_array(rdev, mddev);
6871 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6873 struct md_rdev *rdev;
6878 rdev = find_rdev(mddev, dev);
6882 if (rdev->raid_disk < 0)
6885 clear_bit(Blocked, &rdev->flags);
6886 remove_and_add_spares(mddev, rdev);
6888 if (rdev->raid_disk >= 0)
6892 if (mddev_is_clustered(mddev)) {
6893 if (md_cluster_ops->remove_disk(mddev, rdev))
6897 md_kick_rdev_from_array(rdev);
6898 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6900 md_wakeup_thread(mddev->thread);
6902 md_update_sb(mddev, 1);
6907 pr_debug("md: cannot remove active disk %pg from %s ...\n",
6908 rdev->bdev, mdname(mddev));
6912 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6915 struct md_rdev *rdev;
6920 if (mddev->major_version != 0) {
6921 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6925 if (!mddev->pers->hot_add_disk) {
6926 pr_warn("%s: personality does not support diskops!\n",
6931 rdev = md_import_device(dev, -1, 0);
6933 pr_warn("md: error, md_import_device() returned %ld\n",
6938 if (mddev->persistent)
6939 rdev->sb_start = calc_dev_sboffset(rdev);
6941 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
6943 rdev->sectors = rdev->sb_start;
6945 if (test_bit(Faulty, &rdev->flags)) {
6946 pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
6947 rdev->bdev, mdname(mddev));
6952 clear_bit(In_sync, &rdev->flags);
6954 rdev->saved_raid_disk = -1;
6955 err = bind_rdev_to_array(rdev, mddev);
6960 * The rest should better be atomic, we can have disk failures
6961 * noticed in interrupt contexts ...
6964 rdev->raid_disk = -1;
6966 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6968 md_update_sb(mddev, 1);
6970 * If the new disk does not support REQ_NOWAIT,
6971 * disable on the whole MD.
6973 if (!bdev_nowait(rdev->bdev)) {
6974 pr_info("%s: Disabling nowait because %pg does not support nowait\n",
6975 mdname(mddev), rdev->bdev);
6976 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, mddev->queue);
6979 * Kick recovery, maybe this spare has to be added to the
6980 * array immediately.
6982 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6983 md_wakeup_thread(mddev->thread);
6992 static int set_bitmap_file(struct mddev *mddev, int fd)
6997 if (!mddev->pers->quiesce || !mddev->thread)
6999 if (mddev->recovery || mddev->sync_thread)
7001 /* we should be able to change the bitmap.. */
7005 struct inode *inode;
7008 if (mddev->bitmap || mddev->bitmap_info.file)
7009 return -EEXIST; /* cannot add when bitmap is present */
7013 pr_warn("%s: error: failed to get bitmap file\n",
7018 inode = f->f_mapping->host;
7019 if (!S_ISREG(inode->i_mode)) {
7020 pr_warn("%s: error: bitmap file must be a regular file\n",
7023 } else if (!(f->f_mode & FMODE_WRITE)) {
7024 pr_warn("%s: error: bitmap file must open for write\n",
7027 } else if (atomic_read(&inode->i_writecount) != 1) {
7028 pr_warn("%s: error: bitmap file is already in use\n",
7036 mddev->bitmap_info.file = f;
7037 mddev->bitmap_info.offset = 0; /* file overrides offset */
7038 } else if (mddev->bitmap == NULL)
7039 return -ENOENT; /* cannot remove what isn't there */
7043 struct bitmap *bitmap;
7045 bitmap = md_bitmap_create(mddev, -1);
7046 mddev_suspend(mddev);
7047 if (!IS_ERR(bitmap)) {
7048 mddev->bitmap = bitmap;
7049 err = md_bitmap_load(mddev);
7051 err = PTR_ERR(bitmap);
7053 md_bitmap_destroy(mddev);
7056 mddev_resume(mddev);
7057 } else if (fd < 0) {
7058 mddev_suspend(mddev);
7059 md_bitmap_destroy(mddev);
7060 mddev_resume(mddev);
7064 struct file *f = mddev->bitmap_info.file;
7066 spin_lock(&mddev->lock);
7067 mddev->bitmap_info.file = NULL;
7068 spin_unlock(&mddev->lock);
7077 * md_set_array_info is used two different ways
7078 * The original usage is when creating a new array.
7079 * In this usage, raid_disks is > 0 and it together with
7080 * level, size, not_persistent,layout,chunksize determine the
7081 * shape of the array.
7082 * This will always create an array with a type-0.90.0 superblock.
7083 * The newer usage is when assembling an array.
7084 * In this case raid_disks will be 0, and the major_version field is
7085 * use to determine which style super-blocks are to be found on the devices.
7086 * The minor and patch _version numbers are also kept incase the
7087 * super_block handler wishes to interpret them.
7089 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7091 if (info->raid_disks == 0) {
7092 /* just setting version number for superblock loading */
7093 if (info->major_version < 0 ||
7094 info->major_version >= ARRAY_SIZE(super_types) ||
7095 super_types[info->major_version].name == NULL) {
7096 /* maybe try to auto-load a module? */
7097 pr_warn("md: superblock version %d not known\n",
7098 info->major_version);
7101 mddev->major_version = info->major_version;
7102 mddev->minor_version = info->minor_version;
7103 mddev->patch_version = info->patch_version;
7104 mddev->persistent = !info->not_persistent;
7105 /* ensure mddev_put doesn't delete this now that there
7106 * is some minimal configuration.
7108 mddev->ctime = ktime_get_real_seconds();
7111 mddev->major_version = MD_MAJOR_VERSION;
7112 mddev->minor_version = MD_MINOR_VERSION;
7113 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7114 mddev->ctime = ktime_get_real_seconds();
7116 mddev->level = info->level;
7117 mddev->clevel[0] = 0;
7118 mddev->dev_sectors = 2 * (sector_t)info->size;
7119 mddev->raid_disks = info->raid_disks;
7120 /* don't set md_minor, it is determined by which /dev/md* was
7123 if (info->state & (1<<MD_SB_CLEAN))
7124 mddev->recovery_cp = MaxSector;
7126 mddev->recovery_cp = 0;
7127 mddev->persistent = ! info->not_persistent;
7128 mddev->external = 0;
7130 mddev->layout = info->layout;
7131 if (mddev->level == 0)
7132 /* Cannot trust RAID0 layout info here */
7134 mddev->chunk_sectors = info->chunk_size >> 9;
7136 if (mddev->persistent) {
7137 mddev->max_disks = MD_SB_DISKS;
7139 mddev->sb_flags = 0;
7141 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7143 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7144 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7145 mddev->bitmap_info.offset = 0;
7147 mddev->reshape_position = MaxSector;
7150 * Generate a 128 bit UUID
7152 get_random_bytes(mddev->uuid, 16);
7154 mddev->new_level = mddev->level;
7155 mddev->new_chunk_sectors = mddev->chunk_sectors;
7156 mddev->new_layout = mddev->layout;
7157 mddev->delta_disks = 0;
7158 mddev->reshape_backwards = 0;
7163 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7165 lockdep_assert_held(&mddev->reconfig_mutex);
7167 if (mddev->external_size)
7170 mddev->array_sectors = array_sectors;
7172 EXPORT_SYMBOL(md_set_array_sectors);
7174 static int update_size(struct mddev *mddev, sector_t num_sectors)
7176 struct md_rdev *rdev;
7178 int fit = (num_sectors == 0);
7179 sector_t old_dev_sectors = mddev->dev_sectors;
7181 if (mddev->pers->resize == NULL)
7183 /* The "num_sectors" is the number of sectors of each device that
7184 * is used. This can only make sense for arrays with redundancy.
7185 * linear and raid0 always use whatever space is available. We can only
7186 * consider changing this number if no resync or reconstruction is
7187 * happening, and if the new size is acceptable. It must fit before the
7188 * sb_start or, if that is <data_offset, it must fit before the size
7189 * of each device. If num_sectors is zero, we find the largest size
7192 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7195 if (!md_is_rdwr(mddev))
7198 rdev_for_each(rdev, mddev) {
7199 sector_t avail = rdev->sectors;
7201 if (fit && (num_sectors == 0 || num_sectors > avail))
7202 num_sectors = avail;
7203 if (avail < num_sectors)
7206 rv = mddev->pers->resize(mddev, num_sectors);
7208 if (mddev_is_clustered(mddev))
7209 md_cluster_ops->update_size(mddev, old_dev_sectors);
7210 else if (mddev->queue) {
7211 set_capacity_and_notify(mddev->gendisk,
7212 mddev->array_sectors);
7218 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7221 struct md_rdev *rdev;
7222 /* change the number of raid disks */
7223 if (mddev->pers->check_reshape == NULL)
7225 if (!md_is_rdwr(mddev))
7227 if (raid_disks <= 0 ||
7228 (mddev->max_disks && raid_disks >= mddev->max_disks))
7230 if (mddev->sync_thread ||
7231 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7232 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7233 mddev->reshape_position != MaxSector)
7236 rdev_for_each(rdev, mddev) {
7237 if (mddev->raid_disks < raid_disks &&
7238 rdev->data_offset < rdev->new_data_offset)
7240 if (mddev->raid_disks > raid_disks &&
7241 rdev->data_offset > rdev->new_data_offset)
7245 mddev->delta_disks = raid_disks - mddev->raid_disks;
7246 if (mddev->delta_disks < 0)
7247 mddev->reshape_backwards = 1;
7248 else if (mddev->delta_disks > 0)
7249 mddev->reshape_backwards = 0;
7251 rv = mddev->pers->check_reshape(mddev);
7253 mddev->delta_disks = 0;
7254 mddev->reshape_backwards = 0;
7260 * update_array_info is used to change the configuration of an
7262 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7263 * fields in the info are checked against the array.
7264 * Any differences that cannot be handled will cause an error.
7265 * Normally, only one change can be managed at a time.
7267 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7273 /* calculate expected state,ignoring low bits */
7274 if (mddev->bitmap && mddev->bitmap_info.offset)
7275 state |= (1 << MD_SB_BITMAP_PRESENT);
7277 if (mddev->major_version != info->major_version ||
7278 mddev->minor_version != info->minor_version ||
7279 /* mddev->patch_version != info->patch_version || */
7280 mddev->ctime != info->ctime ||
7281 mddev->level != info->level ||
7282 /* mddev->layout != info->layout || */
7283 mddev->persistent != !info->not_persistent ||
7284 mddev->chunk_sectors != info->chunk_size >> 9 ||
7285 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7286 ((state^info->state) & 0xfffffe00)
7289 /* Check there is only one change */
7290 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7292 if (mddev->raid_disks != info->raid_disks)
7294 if (mddev->layout != info->layout)
7296 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7303 if (mddev->layout != info->layout) {
7305 * we don't need to do anything at the md level, the
7306 * personality will take care of it all.
7308 if (mddev->pers->check_reshape == NULL)
7311 mddev->new_layout = info->layout;
7312 rv = mddev->pers->check_reshape(mddev);
7314 mddev->new_layout = mddev->layout;
7318 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7319 rv = update_size(mddev, (sector_t)info->size * 2);
7321 if (mddev->raid_disks != info->raid_disks)
7322 rv = update_raid_disks(mddev, info->raid_disks);
7324 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7325 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7329 if (mddev->recovery || mddev->sync_thread) {
7333 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7334 struct bitmap *bitmap;
7335 /* add the bitmap */
7336 if (mddev->bitmap) {
7340 if (mddev->bitmap_info.default_offset == 0) {
7344 mddev->bitmap_info.offset =
7345 mddev->bitmap_info.default_offset;
7346 mddev->bitmap_info.space =
7347 mddev->bitmap_info.default_space;
7348 bitmap = md_bitmap_create(mddev, -1);
7349 mddev_suspend(mddev);
7350 if (!IS_ERR(bitmap)) {
7351 mddev->bitmap = bitmap;
7352 rv = md_bitmap_load(mddev);
7354 rv = PTR_ERR(bitmap);
7356 md_bitmap_destroy(mddev);
7357 mddev_resume(mddev);
7359 /* remove the bitmap */
7360 if (!mddev->bitmap) {
7364 if (mddev->bitmap->storage.file) {
7368 if (mddev->bitmap_info.nodes) {
7369 /* hold PW on all the bitmap lock */
7370 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7371 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7373 md_cluster_ops->unlock_all_bitmaps(mddev);
7377 mddev->bitmap_info.nodes = 0;
7378 md_cluster_ops->leave(mddev);
7379 module_put(md_cluster_mod);
7380 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7382 mddev_suspend(mddev);
7383 md_bitmap_destroy(mddev);
7384 mddev_resume(mddev);
7385 mddev->bitmap_info.offset = 0;
7388 md_update_sb(mddev, 1);
7394 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7396 struct md_rdev *rdev;
7399 if (mddev->pers == NULL)
7403 rdev = md_find_rdev_rcu(mddev, dev);
7407 md_error(mddev, rdev);
7408 if (test_bit(MD_BROKEN, &mddev->flags))
7416 * We have a problem here : there is no easy way to give a CHS
7417 * virtual geometry. We currently pretend that we have a 2 heads
7418 * 4 sectors (with a BIG number of cylinders...). This drives
7419 * dosfs just mad... ;-)
7421 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7423 struct mddev *mddev = bdev->bd_disk->private_data;
7427 geo->cylinders = mddev->array_sectors / 8;
7431 static inline bool md_ioctl_valid(unsigned int cmd)
7435 case GET_ARRAY_INFO:
7436 case GET_BITMAP_FILE:
7439 case HOT_REMOVE_DISK:
7441 case RESTART_ARRAY_RW:
7443 case SET_ARRAY_INFO:
7444 case SET_BITMAP_FILE:
7445 case SET_DISK_FAULTY:
7448 case CLUSTERED_DISK_NACK:
7455 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7457 mdu_array_info_t info;
7461 memset(&info, 0, sizeof(info));
7462 else if (copy_from_user(&info, argp, sizeof(info)))
7466 err = update_array_info(mddev, &info);
7468 pr_warn("md: couldn't update array info. %d\n", err);
7472 if (!list_empty(&mddev->disks)) {
7473 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7477 if (mddev->raid_disks) {
7478 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7482 err = md_set_array_info(mddev, &info);
7484 pr_warn("md: couldn't set array info. %d\n", err);
7489 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7490 unsigned int cmd, unsigned long arg)
7493 void __user *argp = (void __user *)arg;
7494 struct mddev *mddev = NULL;
7495 bool did_set_md_closing = false;
7497 if (!md_ioctl_valid(cmd))
7502 case GET_ARRAY_INFO:
7506 if (!capable(CAP_SYS_ADMIN))
7511 * Commands dealing with the RAID driver but not any
7516 err = get_version(argp);
7522 * Commands creating/starting a new array:
7525 mddev = bdev->bd_disk->private_data;
7532 /* Some actions do not requires the mutex */
7534 case GET_ARRAY_INFO:
7535 if (!mddev->raid_disks && !mddev->external)
7538 err = get_array_info(mddev, argp);
7542 if (!mddev->raid_disks && !mddev->external)
7545 err = get_disk_info(mddev, argp);
7548 case SET_DISK_FAULTY:
7549 err = set_disk_faulty(mddev, new_decode_dev(arg));
7552 case GET_BITMAP_FILE:
7553 err = get_bitmap_file(mddev, argp);
7558 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7559 flush_rdev_wq(mddev);
7561 if (cmd == HOT_REMOVE_DISK)
7562 /* need to ensure recovery thread has run */
7563 wait_event_interruptible_timeout(mddev->sb_wait,
7564 !test_bit(MD_RECOVERY_NEEDED,
7566 msecs_to_jiffies(5000));
7567 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7568 /* Need to flush page cache, and ensure no-one else opens
7571 mutex_lock(&mddev->open_mutex);
7572 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7573 mutex_unlock(&mddev->open_mutex);
7577 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7578 mutex_unlock(&mddev->open_mutex);
7582 did_set_md_closing = true;
7583 mutex_unlock(&mddev->open_mutex);
7584 sync_blockdev(bdev);
7586 err = mddev_lock(mddev);
7588 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7593 if (cmd == SET_ARRAY_INFO) {
7594 err = __md_set_array_info(mddev, argp);
7599 * Commands querying/configuring an existing array:
7601 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7602 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7603 if ((!mddev->raid_disks && !mddev->external)
7604 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7605 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7606 && cmd != GET_BITMAP_FILE) {
7612 * Commands even a read-only array can execute:
7615 case RESTART_ARRAY_RW:
7616 err = restart_array(mddev);
7620 err = do_md_stop(mddev, 0, bdev);
7624 err = md_set_readonly(mddev, bdev);
7627 case HOT_REMOVE_DISK:
7628 err = hot_remove_disk(mddev, new_decode_dev(arg));
7632 /* We can support ADD_NEW_DISK on read-only arrays
7633 * only if we are re-adding a preexisting device.
7634 * So require mddev->pers and MD_DISK_SYNC.
7637 mdu_disk_info_t info;
7638 if (copy_from_user(&info, argp, sizeof(info)))
7640 else if (!(info.state & (1<<MD_DISK_SYNC)))
7641 /* Need to clear read-only for this */
7644 err = md_add_new_disk(mddev, &info);
7651 * The remaining ioctls are changing the state of the
7652 * superblock, so we do not allow them on read-only arrays.
7654 if (!md_is_rdwr(mddev) && mddev->pers) {
7655 if (mddev->ro != MD_AUTO_READ) {
7659 mddev->ro = MD_RDWR;
7660 sysfs_notify_dirent_safe(mddev->sysfs_state);
7661 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7662 /* mddev_unlock will wake thread */
7663 /* If a device failed while we were read-only, we
7664 * need to make sure the metadata is updated now.
7666 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7667 mddev_unlock(mddev);
7668 wait_event(mddev->sb_wait,
7669 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7670 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7671 mddev_lock_nointr(mddev);
7678 mdu_disk_info_t info;
7679 if (copy_from_user(&info, argp, sizeof(info)))
7682 err = md_add_new_disk(mddev, &info);
7686 case CLUSTERED_DISK_NACK:
7687 if (mddev_is_clustered(mddev))
7688 md_cluster_ops->new_disk_ack(mddev, false);
7694 err = hot_add_disk(mddev, new_decode_dev(arg));
7698 err = do_md_run(mddev);
7701 case SET_BITMAP_FILE:
7702 err = set_bitmap_file(mddev, (int)arg);
7711 if (mddev->hold_active == UNTIL_IOCTL &&
7713 mddev->hold_active = 0;
7714 mddev_unlock(mddev);
7716 if(did_set_md_closing)
7717 clear_bit(MD_CLOSING, &mddev->flags);
7720 #ifdef CONFIG_COMPAT
7721 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7722 unsigned int cmd, unsigned long arg)
7725 case HOT_REMOVE_DISK:
7727 case SET_DISK_FAULTY:
7728 case SET_BITMAP_FILE:
7729 /* These take in integer arg, do not convert */
7732 arg = (unsigned long)compat_ptr(arg);
7736 return md_ioctl(bdev, mode, cmd, arg);
7738 #endif /* CONFIG_COMPAT */
7740 static int md_set_read_only(struct block_device *bdev, bool ro)
7742 struct mddev *mddev = bdev->bd_disk->private_data;
7745 err = mddev_lock(mddev);
7749 if (!mddev->raid_disks && !mddev->external) {
7755 * Transitioning to read-auto need only happen for arrays that call
7756 * md_write_start and which are not ready for writes yet.
7758 if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7759 err = restart_array(mddev);
7762 mddev->ro = MD_AUTO_READ;
7766 mddev_unlock(mddev);
7770 static int md_open(struct block_device *bdev, fmode_t mode)
7772 struct mddev *mddev;
7775 spin_lock(&all_mddevs_lock);
7776 mddev = mddev_get(bdev->bd_disk->private_data);
7777 spin_unlock(&all_mddevs_lock);
7781 err = mutex_lock_interruptible(&mddev->open_mutex);
7786 if (test_bit(MD_CLOSING, &mddev->flags))
7789 atomic_inc(&mddev->openers);
7790 mutex_unlock(&mddev->open_mutex);
7792 bdev_check_media_change(bdev);
7796 mutex_unlock(&mddev->open_mutex);
7802 static void md_release(struct gendisk *disk, fmode_t mode)
7804 struct mddev *mddev = disk->private_data;
7807 atomic_dec(&mddev->openers);
7811 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7813 struct mddev *mddev = disk->private_data;
7814 unsigned int ret = 0;
7817 ret = DISK_EVENT_MEDIA_CHANGE;
7822 static void md_free_disk(struct gendisk *disk)
7824 struct mddev *mddev = disk->private_data;
7826 percpu_ref_exit(&mddev->writes_pending);
7830 const struct block_device_operations md_fops =
7832 .owner = THIS_MODULE,
7833 .submit_bio = md_submit_bio,
7835 .release = md_release,
7837 #ifdef CONFIG_COMPAT
7838 .compat_ioctl = md_compat_ioctl,
7840 .getgeo = md_getgeo,
7841 .check_events = md_check_events,
7842 .set_read_only = md_set_read_only,
7843 .free_disk = md_free_disk,
7846 static int md_thread(void *arg)
7848 struct md_thread *thread = arg;
7851 * md_thread is a 'system-thread', it's priority should be very
7852 * high. We avoid resource deadlocks individually in each
7853 * raid personality. (RAID5 does preallocation) We also use RR and
7854 * the very same RT priority as kswapd, thus we will never get
7855 * into a priority inversion deadlock.
7857 * we definitely have to have equal or higher priority than
7858 * bdflush, otherwise bdflush will deadlock if there are too
7859 * many dirty RAID5 blocks.
7862 allow_signal(SIGKILL);
7863 while (!kthread_should_stop()) {
7865 /* We need to wait INTERRUPTIBLE so that
7866 * we don't add to the load-average.
7867 * That means we need to be sure no signals are
7870 if (signal_pending(current))
7871 flush_signals(current);
7873 wait_event_interruptible_timeout
7875 test_bit(THREAD_WAKEUP, &thread->flags)
7876 || kthread_should_stop() || kthread_should_park(),
7879 clear_bit(THREAD_WAKEUP, &thread->flags);
7880 if (kthread_should_park())
7882 if (!kthread_should_stop())
7883 thread->run(thread);
7889 void md_wakeup_thread(struct md_thread *thread)
7892 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7893 set_bit(THREAD_WAKEUP, &thread->flags);
7894 wake_up(&thread->wqueue);
7897 EXPORT_SYMBOL(md_wakeup_thread);
7899 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7900 struct mddev *mddev, const char *name)
7902 struct md_thread *thread;
7904 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7908 init_waitqueue_head(&thread->wqueue);
7911 thread->mddev = mddev;
7912 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7913 thread->tsk = kthread_run(md_thread, thread,
7915 mdname(thread->mddev),
7917 if (IS_ERR(thread->tsk)) {
7923 EXPORT_SYMBOL(md_register_thread);
7925 void md_unregister_thread(struct md_thread **threadp)
7927 struct md_thread *thread;
7930 * Locking ensures that mddev_unlock does not wake_up a
7931 * non-existent thread
7933 spin_lock(&pers_lock);
7936 spin_unlock(&pers_lock);
7940 spin_unlock(&pers_lock);
7942 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7943 kthread_stop(thread->tsk);
7946 EXPORT_SYMBOL(md_unregister_thread);
7948 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7950 if (!rdev || test_bit(Faulty, &rdev->flags))
7953 if (!mddev->pers || !mddev->pers->error_handler)
7955 mddev->pers->error_handler(mddev, rdev);
7957 if (mddev->pers->level == 0 || mddev->pers->level == LEVEL_LINEAR)
7960 if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
7961 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7962 sysfs_notify_dirent_safe(rdev->sysfs_state);
7963 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7964 if (!test_bit(MD_BROKEN, &mddev->flags)) {
7965 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7966 md_wakeup_thread(mddev->thread);
7968 if (mddev->event_work.func)
7969 queue_work(md_misc_wq, &mddev->event_work);
7972 EXPORT_SYMBOL(md_error);
7974 /* seq_file implementation /proc/mdstat */
7976 static void status_unused(struct seq_file *seq)
7979 struct md_rdev *rdev;
7981 seq_printf(seq, "unused devices: ");
7983 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7985 seq_printf(seq, "%pg ", rdev->bdev);
7988 seq_printf(seq, "<none>");
7990 seq_printf(seq, "\n");
7993 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7995 sector_t max_sectors, resync, res;
7996 unsigned long dt, db = 0;
7997 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7998 int scale, recovery_active;
7999 unsigned int per_milli;
8001 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8002 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8003 max_sectors = mddev->resync_max_sectors;
8005 max_sectors = mddev->dev_sectors;
8007 resync = mddev->curr_resync;
8008 if (resync < MD_RESYNC_ACTIVE) {
8009 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8010 /* Still cleaning up */
8011 resync = max_sectors;
8012 } else if (resync > max_sectors) {
8013 resync = max_sectors;
8015 res = atomic_read(&mddev->recovery_active);
8017 * Resync has started, but the subtraction has overflowed or
8018 * yielded one of the special values. Force it to active to
8019 * ensure the status reports an active resync.
8021 if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8022 resync = MD_RESYNC_ACTIVE;
8027 if (resync == MD_RESYNC_NONE) {
8028 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8029 struct md_rdev *rdev;
8031 rdev_for_each(rdev, mddev)
8032 if (rdev->raid_disk >= 0 &&
8033 !test_bit(Faulty, &rdev->flags) &&
8034 rdev->recovery_offset != MaxSector &&
8035 rdev->recovery_offset) {
8036 seq_printf(seq, "\trecover=REMOTE");
8039 if (mddev->reshape_position != MaxSector)
8040 seq_printf(seq, "\treshape=REMOTE");
8042 seq_printf(seq, "\tresync=REMOTE");
8045 if (mddev->recovery_cp < MaxSector) {
8046 seq_printf(seq, "\tresync=PENDING");
8051 if (resync < MD_RESYNC_ACTIVE) {
8052 seq_printf(seq, "\tresync=DELAYED");
8056 WARN_ON(max_sectors == 0);
8057 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8058 * in a sector_t, and (max_sectors>>scale) will fit in a
8059 * u32, as those are the requirements for sector_div.
8060 * Thus 'scale' must be at least 10
8063 if (sizeof(sector_t) > sizeof(unsigned long)) {
8064 while ( max_sectors/2 > (1ULL<<(scale+32)))
8067 res = (resync>>scale)*1000;
8068 sector_div(res, (u32)((max_sectors>>scale)+1));
8072 int i, x = per_milli/50, y = 20-x;
8073 seq_printf(seq, "[");
8074 for (i = 0; i < x; i++)
8075 seq_printf(seq, "=");
8076 seq_printf(seq, ">");
8077 for (i = 0; i < y; i++)
8078 seq_printf(seq, ".");
8079 seq_printf(seq, "] ");
8081 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8082 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8084 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8086 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8087 "resync" : "recovery"))),
8088 per_milli/10, per_milli % 10,
8089 (unsigned long long) resync/2,
8090 (unsigned long long) max_sectors/2);
8093 * dt: time from mark until now
8094 * db: blocks written from mark until now
8095 * rt: remaining time
8097 * rt is a sector_t, which is always 64bit now. We are keeping
8098 * the original algorithm, but it is not really necessary.
8100 * Original algorithm:
8101 * So we divide before multiply in case it is 32bit and close
8103 * We scale the divisor (db) by 32 to avoid losing precision
8104 * near the end of resync when the number of remaining sectors
8106 * We then divide rt by 32 after multiplying by db to compensate.
8107 * The '+1' avoids division by zero if db is very small.
8109 dt = ((jiffies - mddev->resync_mark) / HZ);
8112 curr_mark_cnt = mddev->curr_mark_cnt;
8113 recovery_active = atomic_read(&mddev->recovery_active);
8114 resync_mark_cnt = mddev->resync_mark_cnt;
8116 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8117 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8119 rt = max_sectors - resync; /* number of remaining sectors */
8120 rt = div64_u64(rt, db/32+1);
8124 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8125 ((unsigned long)rt % 60)/6);
8127 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8131 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8133 struct list_head *tmp;
8135 struct mddev *mddev;
8147 spin_lock(&all_mddevs_lock);
8148 list_for_each(tmp,&all_mddevs)
8150 mddev = list_entry(tmp, struct mddev, all_mddevs);
8151 if (!mddev_get(mddev))
8153 spin_unlock(&all_mddevs_lock);
8156 spin_unlock(&all_mddevs_lock);
8158 return (void*)2;/* tail */
8162 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8164 struct list_head *tmp;
8165 struct mddev *next_mddev, *mddev = v;
8166 struct mddev *to_put = NULL;
8172 spin_lock(&all_mddevs_lock);
8173 if (v == (void*)1) {
8174 tmp = all_mddevs.next;
8177 tmp = mddev->all_mddevs.next;
8181 if (tmp == &all_mddevs) {
8182 next_mddev = (void*)2;
8186 next_mddev = list_entry(tmp, struct mddev, all_mddevs);
8187 if (mddev_get(next_mddev))
8190 tmp = mddev->all_mddevs.next;
8192 spin_unlock(&all_mddevs_lock);
8200 static void md_seq_stop(struct seq_file *seq, void *v)
8202 struct mddev *mddev = v;
8204 if (mddev && v != (void*)1 && v != (void*)2)
8208 static int md_seq_show(struct seq_file *seq, void *v)
8210 struct mddev *mddev = v;
8212 struct md_rdev *rdev;
8214 if (v == (void*)1) {
8215 struct md_personality *pers;
8216 seq_printf(seq, "Personalities : ");
8217 spin_lock(&pers_lock);
8218 list_for_each_entry(pers, &pers_list, list)
8219 seq_printf(seq, "[%s] ", pers->name);
8221 spin_unlock(&pers_lock);
8222 seq_printf(seq, "\n");
8223 seq->poll_event = atomic_read(&md_event_count);
8226 if (v == (void*)2) {
8231 spin_lock(&mddev->lock);
8232 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8233 seq_printf(seq, "%s : %sactive", mdname(mddev),
8234 mddev->pers ? "" : "in");
8236 if (mddev->ro == MD_RDONLY)
8237 seq_printf(seq, " (read-only)");
8238 if (mddev->ro == MD_AUTO_READ)
8239 seq_printf(seq, " (auto-read-only)");
8240 seq_printf(seq, " %s", mddev->pers->name);
8245 rdev_for_each_rcu(rdev, mddev) {
8246 seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8248 if (test_bit(WriteMostly, &rdev->flags))
8249 seq_printf(seq, "(W)");
8250 if (test_bit(Journal, &rdev->flags))
8251 seq_printf(seq, "(J)");
8252 if (test_bit(Faulty, &rdev->flags)) {
8253 seq_printf(seq, "(F)");
8256 if (rdev->raid_disk < 0)
8257 seq_printf(seq, "(S)"); /* spare */
8258 if (test_bit(Replacement, &rdev->flags))
8259 seq_printf(seq, "(R)");
8260 sectors += rdev->sectors;
8264 if (!list_empty(&mddev->disks)) {
8266 seq_printf(seq, "\n %llu blocks",
8267 (unsigned long long)
8268 mddev->array_sectors / 2);
8270 seq_printf(seq, "\n %llu blocks",
8271 (unsigned long long)sectors / 2);
8273 if (mddev->persistent) {
8274 if (mddev->major_version != 0 ||
8275 mddev->minor_version != 90) {
8276 seq_printf(seq," super %d.%d",
8277 mddev->major_version,
8278 mddev->minor_version);
8280 } else if (mddev->external)
8281 seq_printf(seq, " super external:%s",
8282 mddev->metadata_type);
8284 seq_printf(seq, " super non-persistent");
8287 mddev->pers->status(seq, mddev);
8288 seq_printf(seq, "\n ");
8289 if (mddev->pers->sync_request) {
8290 if (status_resync(seq, mddev))
8291 seq_printf(seq, "\n ");
8294 seq_printf(seq, "\n ");
8296 md_bitmap_status(seq, mddev->bitmap);
8298 seq_printf(seq, "\n");
8300 spin_unlock(&mddev->lock);
8305 static const struct seq_operations md_seq_ops = {
8306 .start = md_seq_start,
8307 .next = md_seq_next,
8308 .stop = md_seq_stop,
8309 .show = md_seq_show,
8312 static int md_seq_open(struct inode *inode, struct file *file)
8314 struct seq_file *seq;
8317 error = seq_open(file, &md_seq_ops);
8321 seq = file->private_data;
8322 seq->poll_event = atomic_read(&md_event_count);
8326 static int md_unloading;
8327 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8329 struct seq_file *seq = filp->private_data;
8333 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8334 poll_wait(filp, &md_event_waiters, wait);
8336 /* always allow read */
8337 mask = EPOLLIN | EPOLLRDNORM;
8339 if (seq->poll_event != atomic_read(&md_event_count))
8340 mask |= EPOLLERR | EPOLLPRI;
8344 static const struct proc_ops mdstat_proc_ops = {
8345 .proc_open = md_seq_open,
8346 .proc_read = seq_read,
8347 .proc_lseek = seq_lseek,
8348 .proc_release = seq_release,
8349 .proc_poll = mdstat_poll,
8352 int register_md_personality(struct md_personality *p)
8354 pr_debug("md: %s personality registered for level %d\n",
8356 spin_lock(&pers_lock);
8357 list_add_tail(&p->list, &pers_list);
8358 spin_unlock(&pers_lock);
8361 EXPORT_SYMBOL(register_md_personality);
8363 int unregister_md_personality(struct md_personality *p)
8365 pr_debug("md: %s personality unregistered\n", p->name);
8366 spin_lock(&pers_lock);
8367 list_del_init(&p->list);
8368 spin_unlock(&pers_lock);
8371 EXPORT_SYMBOL(unregister_md_personality);
8373 int register_md_cluster_operations(struct md_cluster_operations *ops,
8374 struct module *module)
8377 spin_lock(&pers_lock);
8378 if (md_cluster_ops != NULL)
8381 md_cluster_ops = ops;
8382 md_cluster_mod = module;
8384 spin_unlock(&pers_lock);
8387 EXPORT_SYMBOL(register_md_cluster_operations);
8389 int unregister_md_cluster_operations(void)
8391 spin_lock(&pers_lock);
8392 md_cluster_ops = NULL;
8393 spin_unlock(&pers_lock);
8396 EXPORT_SYMBOL(unregister_md_cluster_operations);
8398 int md_setup_cluster(struct mddev *mddev, int nodes)
8401 if (!md_cluster_ops)
8402 request_module("md-cluster");
8403 spin_lock(&pers_lock);
8404 /* ensure module won't be unloaded */
8405 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8406 pr_warn("can't find md-cluster module or get its reference.\n");
8407 spin_unlock(&pers_lock);
8410 spin_unlock(&pers_lock);
8412 ret = md_cluster_ops->join(mddev, nodes);
8414 mddev->safemode_delay = 0;
8418 void md_cluster_stop(struct mddev *mddev)
8420 if (!md_cluster_ops)
8422 md_cluster_ops->leave(mddev);
8423 module_put(md_cluster_mod);
8426 static int is_mddev_idle(struct mddev *mddev, int init)
8428 struct md_rdev *rdev;
8434 rdev_for_each_rcu(rdev, mddev) {
8435 struct gendisk *disk = rdev->bdev->bd_disk;
8436 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8437 atomic_read(&disk->sync_io);
8438 /* sync IO will cause sync_io to increase before the disk_stats
8439 * as sync_io is counted when a request starts, and
8440 * disk_stats is counted when it completes.
8441 * So resync activity will cause curr_events to be smaller than
8442 * when there was no such activity.
8443 * non-sync IO will cause disk_stat to increase without
8444 * increasing sync_io so curr_events will (eventually)
8445 * be larger than it was before. Once it becomes
8446 * substantially larger, the test below will cause
8447 * the array to appear non-idle, and resync will slow
8449 * If there is a lot of outstanding resync activity when
8450 * we set last_event to curr_events, then all that activity
8451 * completing might cause the array to appear non-idle
8452 * and resync will be slowed down even though there might
8453 * not have been non-resync activity. This will only
8454 * happen once though. 'last_events' will soon reflect
8455 * the state where there is little or no outstanding
8456 * resync requests, and further resync activity will
8457 * always make curr_events less than last_events.
8460 if (init || curr_events - rdev->last_events > 64) {
8461 rdev->last_events = curr_events;
8469 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8471 /* another "blocks" (512byte) blocks have been synced */
8472 atomic_sub(blocks, &mddev->recovery_active);
8473 wake_up(&mddev->recovery_wait);
8475 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8476 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8477 md_wakeup_thread(mddev->thread);
8478 // stop recovery, signal do_sync ....
8481 EXPORT_SYMBOL(md_done_sync);
8483 /* md_write_start(mddev, bi)
8484 * If we need to update some array metadata (e.g. 'active' flag
8485 * in superblock) before writing, schedule a superblock update
8486 * and wait for it to complete.
8487 * A return value of 'false' means that the write wasn't recorded
8488 * and cannot proceed as the array is being suspend.
8490 bool md_write_start(struct mddev *mddev, struct bio *bi)
8494 if (bio_data_dir(bi) != WRITE)
8497 BUG_ON(mddev->ro == MD_RDONLY);
8498 if (mddev->ro == MD_AUTO_READ) {
8499 /* need to switch to read/write */
8500 mddev->ro = MD_RDWR;
8501 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8502 md_wakeup_thread(mddev->thread);
8503 md_wakeup_thread(mddev->sync_thread);
8507 percpu_ref_get(&mddev->writes_pending);
8508 smp_mb(); /* Match smp_mb in set_in_sync() */
8509 if (mddev->safemode == 1)
8510 mddev->safemode = 0;
8511 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8512 if (mddev->in_sync || mddev->sync_checkers) {
8513 spin_lock(&mddev->lock);
8514 if (mddev->in_sync) {
8516 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8517 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8518 md_wakeup_thread(mddev->thread);
8521 spin_unlock(&mddev->lock);
8525 sysfs_notify_dirent_safe(mddev->sysfs_state);
8526 if (!mddev->has_superblocks)
8528 wait_event(mddev->sb_wait,
8529 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8530 is_md_suspended(mddev));
8531 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8532 percpu_ref_put(&mddev->writes_pending);
8537 EXPORT_SYMBOL(md_write_start);
8539 /* md_write_inc can only be called when md_write_start() has
8540 * already been called at least once of the current request.
8541 * It increments the counter and is useful when a single request
8542 * is split into several parts. Each part causes an increment and
8543 * so needs a matching md_write_end().
8544 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8545 * a spinlocked region.
8547 void md_write_inc(struct mddev *mddev, struct bio *bi)
8549 if (bio_data_dir(bi) != WRITE)
8551 WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8552 percpu_ref_get(&mddev->writes_pending);
8554 EXPORT_SYMBOL(md_write_inc);
8556 void md_write_end(struct mddev *mddev)
8558 percpu_ref_put(&mddev->writes_pending);
8560 if (mddev->safemode == 2)
8561 md_wakeup_thread(mddev->thread);
8562 else if (mddev->safemode_delay)
8563 /* The roundup() ensures this only performs locking once
8564 * every ->safemode_delay jiffies
8566 mod_timer(&mddev->safemode_timer,
8567 roundup(jiffies, mddev->safemode_delay) +
8568 mddev->safemode_delay);
8571 EXPORT_SYMBOL(md_write_end);
8573 /* This is used by raid0 and raid10 */
8574 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8575 struct bio *bio, sector_t start, sector_t size)
8577 struct bio *discard_bio = NULL;
8579 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8580 &discard_bio) || !discard_bio)
8583 bio_chain(discard_bio, bio);
8584 bio_clone_blkg_association(discard_bio, bio);
8586 trace_block_bio_remap(discard_bio,
8587 disk_devt(mddev->gendisk),
8588 bio->bi_iter.bi_sector);
8589 submit_bio_noacct(discard_bio);
8591 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8593 int acct_bioset_init(struct mddev *mddev)
8597 if (!bioset_initialized(&mddev->io_acct_set))
8598 err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
8599 offsetof(struct md_io_acct, bio_clone), 0);
8602 EXPORT_SYMBOL_GPL(acct_bioset_init);
8604 void acct_bioset_exit(struct mddev *mddev)
8606 bioset_exit(&mddev->io_acct_set);
8608 EXPORT_SYMBOL_GPL(acct_bioset_exit);
8610 static void md_end_io_acct(struct bio *bio)
8612 struct md_io_acct *md_io_acct = bio->bi_private;
8613 struct bio *orig_bio = md_io_acct->orig_bio;
8614 struct mddev *mddev = md_io_acct->mddev;
8616 orig_bio->bi_status = bio->bi_status;
8618 bio_end_io_acct(orig_bio, md_io_acct->start_time);
8620 bio_endio(orig_bio);
8622 percpu_ref_put(&mddev->active_io);
8626 * Used by personalities that don't already clone the bio and thus can't
8627 * easily add the timestamp to their extended bio structure.
8629 void md_account_bio(struct mddev *mddev, struct bio **bio)
8631 struct block_device *bdev = (*bio)->bi_bdev;
8632 struct md_io_acct *md_io_acct;
8635 if (!blk_queue_io_stat(bdev->bd_disk->queue))
8638 percpu_ref_get(&mddev->active_io);
8640 clone = bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_acct_set);
8641 md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8642 md_io_acct->orig_bio = *bio;
8643 md_io_acct->start_time = bio_start_io_acct(*bio);
8644 md_io_acct->mddev = mddev;
8646 clone->bi_end_io = md_end_io_acct;
8647 clone->bi_private = md_io_acct;
8650 EXPORT_SYMBOL_GPL(md_account_bio);
8652 /* md_allow_write(mddev)
8653 * Calling this ensures that the array is marked 'active' so that writes
8654 * may proceed without blocking. It is important to call this before
8655 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8656 * Must be called with mddev_lock held.
8658 void md_allow_write(struct mddev *mddev)
8662 if (!md_is_rdwr(mddev))
8664 if (!mddev->pers->sync_request)
8667 spin_lock(&mddev->lock);
8668 if (mddev->in_sync) {
8670 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8671 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8672 if (mddev->safemode_delay &&
8673 mddev->safemode == 0)
8674 mddev->safemode = 1;
8675 spin_unlock(&mddev->lock);
8676 md_update_sb(mddev, 0);
8677 sysfs_notify_dirent_safe(mddev->sysfs_state);
8678 /* wait for the dirty state to be recorded in the metadata */
8679 wait_event(mddev->sb_wait,
8680 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8682 spin_unlock(&mddev->lock);
8684 EXPORT_SYMBOL_GPL(md_allow_write);
8686 #define SYNC_MARKS 10
8687 #define SYNC_MARK_STEP (3*HZ)
8688 #define UPDATE_FREQUENCY (5*60*HZ)
8689 void md_do_sync(struct md_thread *thread)
8691 struct mddev *mddev = thread->mddev;
8692 struct mddev *mddev2;
8693 unsigned int currspeed = 0, window;
8694 sector_t max_sectors,j, io_sectors, recovery_done;
8695 unsigned long mark[SYNC_MARKS];
8696 unsigned long update_time;
8697 sector_t mark_cnt[SYNC_MARKS];
8699 sector_t last_check;
8701 struct md_rdev *rdev;
8702 char *desc, *action = NULL;
8703 struct blk_plug plug;
8706 /* just incase thread restarts... */
8707 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8708 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8710 if (!md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8711 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8715 if (mddev_is_clustered(mddev)) {
8716 ret = md_cluster_ops->resync_start(mddev);
8720 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8721 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8722 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8723 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8724 && ((unsigned long long)mddev->curr_resync_completed
8725 < (unsigned long long)mddev->resync_max_sectors))
8729 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8730 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8731 desc = "data-check";
8733 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8734 desc = "requested-resync";
8738 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8743 mddev->last_sync_action = action ?: desc;
8746 * Before starting a resync we must have set curr_resync to
8747 * 2, and then checked that every "conflicting" array has curr_resync
8748 * less than ours. When we find one that is the same or higher
8749 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8750 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8751 * This will mean we have to start checking from the beginning again.
8756 int mddev2_minor = -1;
8757 mddev->curr_resync = MD_RESYNC_DELAYED;
8760 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8762 spin_lock(&all_mddevs_lock);
8763 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8764 if (test_bit(MD_DELETED, &mddev2->flags))
8766 if (mddev2 == mddev)
8768 if (!mddev->parallel_resync
8769 && mddev2->curr_resync
8770 && match_mddev_units(mddev, mddev2)) {
8772 if (mddev < mddev2 &&
8773 mddev->curr_resync == MD_RESYNC_DELAYED) {
8774 /* arbitrarily yield */
8775 mddev->curr_resync = MD_RESYNC_YIELDED;
8776 wake_up(&resync_wait);
8778 if (mddev > mddev2 &&
8779 mddev->curr_resync == MD_RESYNC_YIELDED)
8780 /* no need to wait here, we can wait the next
8781 * time 'round when curr_resync == 2
8784 /* We need to wait 'interruptible' so as not to
8785 * contribute to the load average, and not to
8786 * be caught by 'softlockup'
8788 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8789 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8790 mddev2->curr_resync >= mddev->curr_resync) {
8791 if (mddev2_minor != mddev2->md_minor) {
8792 mddev2_minor = mddev2->md_minor;
8793 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8794 desc, mdname(mddev),
8797 spin_unlock(&all_mddevs_lock);
8799 if (signal_pending(current))
8800 flush_signals(current);
8802 finish_wait(&resync_wait, &wq);
8805 finish_wait(&resync_wait, &wq);
8808 spin_unlock(&all_mddevs_lock);
8809 } while (mddev->curr_resync < MD_RESYNC_DELAYED);
8812 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8813 /* resync follows the size requested by the personality,
8814 * which defaults to physical size, but can be virtual size
8816 max_sectors = mddev->resync_max_sectors;
8817 atomic64_set(&mddev->resync_mismatches, 0);
8818 /* we don't use the checkpoint if there's a bitmap */
8819 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8820 j = mddev->resync_min;
8821 else if (!mddev->bitmap)
8822 j = mddev->recovery_cp;
8824 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8825 max_sectors = mddev->resync_max_sectors;
8827 * If the original node aborts reshaping then we continue the
8828 * reshaping, so set j again to avoid restart reshape from the
8831 if (mddev_is_clustered(mddev) &&
8832 mddev->reshape_position != MaxSector)
8833 j = mddev->reshape_position;
8835 /* recovery follows the physical size of devices */
8836 max_sectors = mddev->dev_sectors;
8839 rdev_for_each_rcu(rdev, mddev)
8840 if (rdev->raid_disk >= 0 &&
8841 !test_bit(Journal, &rdev->flags) &&
8842 !test_bit(Faulty, &rdev->flags) &&
8843 !test_bit(In_sync, &rdev->flags) &&
8844 rdev->recovery_offset < j)
8845 j = rdev->recovery_offset;
8848 /* If there is a bitmap, we need to make sure all
8849 * writes that started before we added a spare
8850 * complete before we start doing a recovery.
8851 * Otherwise the write might complete and (via
8852 * bitmap_endwrite) set a bit in the bitmap after the
8853 * recovery has checked that bit and skipped that
8856 if (mddev->bitmap) {
8857 mddev->pers->quiesce(mddev, 1);
8858 mddev->pers->quiesce(mddev, 0);
8862 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8863 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8864 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8865 speed_max(mddev), desc);
8867 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8870 for (m = 0; m < SYNC_MARKS; m++) {
8872 mark_cnt[m] = io_sectors;
8875 mddev->resync_mark = mark[last_mark];
8876 mddev->resync_mark_cnt = mark_cnt[last_mark];
8879 * Tune reconstruction:
8881 window = 32 * (PAGE_SIZE / 512);
8882 pr_debug("md: using %dk window, over a total of %lluk.\n",
8883 window/2, (unsigned long long)max_sectors/2);
8885 atomic_set(&mddev->recovery_active, 0);
8888 if (j >= MD_RESYNC_ACTIVE) {
8889 pr_debug("md: resuming %s of %s from checkpoint.\n",
8890 desc, mdname(mddev));
8891 mddev->curr_resync = j;
8893 mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
8894 mddev->curr_resync_completed = j;
8895 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8897 update_time = jiffies;
8899 blk_start_plug(&plug);
8900 while (j < max_sectors) {
8905 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8906 ((mddev->curr_resync > mddev->curr_resync_completed &&
8907 (mddev->curr_resync - mddev->curr_resync_completed)
8908 > (max_sectors >> 4)) ||
8909 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8910 (j - mddev->curr_resync_completed)*2
8911 >= mddev->resync_max - mddev->curr_resync_completed ||
8912 mddev->curr_resync_completed > mddev->resync_max
8914 /* time to update curr_resync_completed */
8915 wait_event(mddev->recovery_wait,
8916 atomic_read(&mddev->recovery_active) == 0);
8917 mddev->curr_resync_completed = j;
8918 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8919 j > mddev->recovery_cp)
8920 mddev->recovery_cp = j;
8921 update_time = jiffies;
8922 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8923 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8926 while (j >= mddev->resync_max &&
8927 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8928 /* As this condition is controlled by user-space,
8929 * we can block indefinitely, so use '_interruptible'
8930 * to avoid triggering warnings.
8932 flush_signals(current); /* just in case */
8933 wait_event_interruptible(mddev->recovery_wait,
8934 mddev->resync_max > j
8935 || test_bit(MD_RECOVERY_INTR,
8939 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8942 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8944 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8948 if (!skipped) { /* actual IO requested */
8949 io_sectors += sectors;
8950 atomic_add(sectors, &mddev->recovery_active);
8953 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8957 if (j > max_sectors)
8958 /* when skipping, extra large numbers can be returned. */
8960 if (j >= MD_RESYNC_ACTIVE)
8961 mddev->curr_resync = j;
8962 mddev->curr_mark_cnt = io_sectors;
8963 if (last_check == 0)
8964 /* this is the earliest that rebuild will be
8965 * visible in /proc/mdstat
8969 if (last_check + window > io_sectors || j == max_sectors)
8972 last_check = io_sectors;
8974 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8976 int next = (last_mark+1) % SYNC_MARKS;
8978 mddev->resync_mark = mark[next];
8979 mddev->resync_mark_cnt = mark_cnt[next];
8980 mark[next] = jiffies;
8981 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8985 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8989 * this loop exits only if either when we are slower than
8990 * the 'hard' speed limit, or the system was IO-idle for
8992 * the system might be non-idle CPU-wise, but we only care
8993 * about not overloading the IO subsystem. (things like an
8994 * e2fsck being done on the RAID array should execute fast)
8998 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8999 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9000 /((jiffies-mddev->resync_mark)/HZ +1) +1;
9002 if (currspeed > speed_min(mddev)) {
9003 if (currspeed > speed_max(mddev)) {
9007 if (!is_mddev_idle(mddev, 0)) {
9009 * Give other IO more of a chance.
9010 * The faster the devices, the less we wait.
9012 wait_event(mddev->recovery_wait,
9013 !atomic_read(&mddev->recovery_active));
9017 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9018 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9019 ? "interrupted" : "done");
9021 * this also signals 'finished resyncing' to md_stop
9023 blk_finish_plug(&plug);
9024 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9026 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9027 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9028 mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9029 mddev->curr_resync_completed = mddev->curr_resync;
9030 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9032 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9034 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9035 mddev->curr_resync > MD_RESYNC_ACTIVE) {
9036 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9037 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9038 if (mddev->curr_resync >= mddev->recovery_cp) {
9039 pr_debug("md: checkpointing %s of %s.\n",
9040 desc, mdname(mddev));
9041 if (test_bit(MD_RECOVERY_ERROR,
9043 mddev->recovery_cp =
9044 mddev->curr_resync_completed;
9046 mddev->recovery_cp =
9050 mddev->recovery_cp = MaxSector;
9052 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9053 mddev->curr_resync = MaxSector;
9054 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9055 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9057 rdev_for_each_rcu(rdev, mddev)
9058 if (rdev->raid_disk >= 0 &&
9059 mddev->delta_disks >= 0 &&
9060 !test_bit(Journal, &rdev->flags) &&
9061 !test_bit(Faulty, &rdev->flags) &&
9062 !test_bit(In_sync, &rdev->flags) &&
9063 rdev->recovery_offset < mddev->curr_resync)
9064 rdev->recovery_offset = mddev->curr_resync;
9070 /* set CHANGE_PENDING here since maybe another update is needed,
9071 * so other nodes are informed. It should be harmless for normal
9073 set_mask_bits(&mddev->sb_flags, 0,
9074 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9076 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9077 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9078 mddev->delta_disks > 0 &&
9079 mddev->pers->finish_reshape &&
9080 mddev->pers->size &&
9082 mddev_lock_nointr(mddev);
9083 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9084 mddev_unlock(mddev);
9085 if (!mddev_is_clustered(mddev))
9086 set_capacity_and_notify(mddev->gendisk,
9087 mddev->array_sectors);
9090 spin_lock(&mddev->lock);
9091 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9092 /* We completed so min/max setting can be forgotten if used. */
9093 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9094 mddev->resync_min = 0;
9095 mddev->resync_max = MaxSector;
9096 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9097 mddev->resync_min = mddev->curr_resync_completed;
9098 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9099 mddev->curr_resync = MD_RESYNC_NONE;
9100 spin_unlock(&mddev->lock);
9102 wake_up(&resync_wait);
9103 md_wakeup_thread(mddev->thread);
9106 EXPORT_SYMBOL_GPL(md_do_sync);
9108 static int remove_and_add_spares(struct mddev *mddev,
9109 struct md_rdev *this)
9111 struct md_rdev *rdev;
9114 bool remove_some = false;
9116 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9117 /* Mustn't remove devices when resync thread is running */
9120 rdev_for_each(rdev, mddev) {
9121 if ((this == NULL || rdev == this) &&
9122 rdev->raid_disk >= 0 &&
9123 !test_bit(Blocked, &rdev->flags) &&
9124 test_bit(Faulty, &rdev->flags) &&
9125 atomic_read(&rdev->nr_pending)==0) {
9126 /* Faulty non-Blocked devices with nr_pending == 0
9127 * never get nr_pending incremented,
9128 * never get Faulty cleared, and never get Blocked set.
9129 * So we can synchronize_rcu now rather than once per device
9132 set_bit(RemoveSynchronized, &rdev->flags);
9138 rdev_for_each(rdev, mddev) {
9139 if ((this == NULL || rdev == this) &&
9140 rdev->raid_disk >= 0 &&
9141 !test_bit(Blocked, &rdev->flags) &&
9142 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9143 (!test_bit(In_sync, &rdev->flags) &&
9144 !test_bit(Journal, &rdev->flags))) &&
9145 atomic_read(&rdev->nr_pending)==0)) {
9146 if (mddev->pers->hot_remove_disk(
9147 mddev, rdev) == 0) {
9148 sysfs_unlink_rdev(mddev, rdev);
9149 rdev->saved_raid_disk = rdev->raid_disk;
9150 rdev->raid_disk = -1;
9154 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9155 clear_bit(RemoveSynchronized, &rdev->flags);
9158 if (removed && mddev->kobj.sd)
9159 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9161 if (this && removed)
9164 rdev_for_each(rdev, mddev) {
9165 if (this && this != rdev)
9167 if (test_bit(Candidate, &rdev->flags))
9169 if (rdev->raid_disk >= 0 &&
9170 !test_bit(In_sync, &rdev->flags) &&
9171 !test_bit(Journal, &rdev->flags) &&
9172 !test_bit(Faulty, &rdev->flags))
9174 if (rdev->raid_disk >= 0)
9176 if (test_bit(Faulty, &rdev->flags))
9178 if (!test_bit(Journal, &rdev->flags)) {
9179 if (!md_is_rdwr(mddev) &&
9180 !(rdev->saved_raid_disk >= 0 &&
9181 !test_bit(Bitmap_sync, &rdev->flags)))
9184 rdev->recovery_offset = 0;
9186 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9187 /* failure here is OK */
9188 sysfs_link_rdev(mddev, rdev);
9189 if (!test_bit(Journal, &rdev->flags))
9192 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9197 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9201 static void md_start_sync(struct work_struct *ws)
9203 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9205 mddev->sync_thread = md_register_thread(md_do_sync,
9208 if (!mddev->sync_thread) {
9209 pr_warn("%s: could not start resync thread...\n",
9211 /* leave the spares where they are, it shouldn't hurt */
9212 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9213 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9214 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9215 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9216 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9217 wake_up(&resync_wait);
9218 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9220 if (mddev->sysfs_action)
9221 sysfs_notify_dirent_safe(mddev->sysfs_action);
9223 md_wakeup_thread(mddev->sync_thread);
9224 sysfs_notify_dirent_safe(mddev->sysfs_action);
9229 * This routine is regularly called by all per-raid-array threads to
9230 * deal with generic issues like resync and super-block update.
9231 * Raid personalities that don't have a thread (linear/raid0) do not
9232 * need this as they never do any recovery or update the superblock.
9234 * It does not do any resync itself, but rather "forks" off other threads
9235 * to do that as needed.
9236 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9237 * "->recovery" and create a thread at ->sync_thread.
9238 * When the thread finishes it sets MD_RECOVERY_DONE
9239 * and wakeups up this thread which will reap the thread and finish up.
9240 * This thread also removes any faulty devices (with nr_pending == 0).
9242 * The overall approach is:
9243 * 1/ if the superblock needs updating, update it.
9244 * 2/ If a recovery thread is running, don't do anything else.
9245 * 3/ If recovery has finished, clean up, possibly marking spares active.
9246 * 4/ If there are any faulty devices, remove them.
9247 * 5/ If array is degraded, try to add spares devices
9248 * 6/ If array has spares or is not in-sync, start a resync thread.
9250 void md_check_recovery(struct mddev *mddev)
9252 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9253 /* Write superblock - thread that called mddev_suspend()
9254 * holds reconfig_mutex for us.
9256 set_bit(MD_UPDATING_SB, &mddev->flags);
9257 smp_mb__after_atomic();
9258 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9259 md_update_sb(mddev, 0);
9260 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9261 wake_up(&mddev->sb_wait);
9264 if (is_md_suspended(mddev))
9268 md_bitmap_daemon_work(mddev);
9270 if (signal_pending(current)) {
9271 if (mddev->pers->sync_request && !mddev->external) {
9272 pr_debug("md: %s in immediate safe mode\n",
9274 mddev->safemode = 2;
9276 flush_signals(current);
9279 if (!md_is_rdwr(mddev) &&
9280 !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9283 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9284 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9285 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9286 (mddev->external == 0 && mddev->safemode == 1) ||
9287 (mddev->safemode == 2
9288 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9292 if (mddev_trylock(mddev)) {
9294 bool try_set_sync = mddev->safemode != 0;
9296 if (!mddev->external && mddev->safemode == 1)
9297 mddev->safemode = 0;
9299 if (!md_is_rdwr(mddev)) {
9300 struct md_rdev *rdev;
9301 if (!mddev->external && mddev->in_sync)
9302 /* 'Blocked' flag not needed as failed devices
9303 * will be recorded if array switched to read/write.
9304 * Leaving it set will prevent the device
9305 * from being removed.
9307 rdev_for_each(rdev, mddev)
9308 clear_bit(Blocked, &rdev->flags);
9309 /* On a read-only array we can:
9310 * - remove failed devices
9311 * - add already-in_sync devices if the array itself
9313 * As we only add devices that are already in-sync,
9314 * we can activate the spares immediately.
9316 remove_and_add_spares(mddev, NULL);
9317 /* There is no thread, but we need to call
9318 * ->spare_active and clear saved_raid_disk
9320 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9321 md_unregister_thread(&mddev->sync_thread);
9322 md_reap_sync_thread(mddev);
9323 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9324 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9325 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9329 if (mddev_is_clustered(mddev)) {
9330 struct md_rdev *rdev, *tmp;
9331 /* kick the device if another node issued a
9334 rdev_for_each_safe(rdev, tmp, mddev) {
9335 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9336 rdev->raid_disk < 0)
9337 md_kick_rdev_from_array(rdev);
9341 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9342 spin_lock(&mddev->lock);
9344 spin_unlock(&mddev->lock);
9347 if (mddev->sb_flags)
9348 md_update_sb(mddev, 0);
9350 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9351 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9352 /* resync/recovery still happening */
9353 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9356 if (mddev->sync_thread) {
9357 md_unregister_thread(&mddev->sync_thread);
9358 md_reap_sync_thread(mddev);
9361 /* Set RUNNING before clearing NEEDED to avoid
9362 * any transients in the value of "sync_action".
9364 mddev->curr_resync_completed = 0;
9365 spin_lock(&mddev->lock);
9366 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9367 spin_unlock(&mddev->lock);
9368 /* Clear some bits that don't mean anything, but
9371 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9372 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9374 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9375 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9377 /* no recovery is running.
9378 * remove any failed drives, then
9379 * add spares if possible.
9380 * Spares are also removed and re-added, to allow
9381 * the personality to fail the re-add.
9384 if (mddev->reshape_position != MaxSector) {
9385 if (mddev->pers->check_reshape == NULL ||
9386 mddev->pers->check_reshape(mddev) != 0)
9387 /* Cannot proceed */
9389 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9390 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9391 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9392 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9393 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9394 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9395 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9396 } else if (mddev->recovery_cp < MaxSector) {
9397 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9398 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9399 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9400 /* nothing to be done ... */
9403 if (mddev->pers->sync_request) {
9405 /* We are adding a device or devices to an array
9406 * which has the bitmap stored on all devices.
9407 * So make sure all bitmap pages get written
9409 md_bitmap_write_all(mddev->bitmap);
9411 INIT_WORK(&mddev->del_work, md_start_sync);
9412 queue_work(md_misc_wq, &mddev->del_work);
9416 if (!mddev->sync_thread) {
9417 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9418 wake_up(&resync_wait);
9419 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9421 if (mddev->sysfs_action)
9422 sysfs_notify_dirent_safe(mddev->sysfs_action);
9425 wake_up(&mddev->sb_wait);
9426 mddev_unlock(mddev);
9429 EXPORT_SYMBOL(md_check_recovery);
9431 void md_reap_sync_thread(struct mddev *mddev)
9433 struct md_rdev *rdev;
9434 sector_t old_dev_sectors = mddev->dev_sectors;
9435 bool is_reshaped = false;
9437 /* sync_thread should be unregistered, collect result */
9438 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9439 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9440 mddev->degraded != mddev->raid_disks) {
9442 /* activate any spares */
9443 if (mddev->pers->spare_active(mddev)) {
9444 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9445 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9448 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9449 mddev->pers->finish_reshape) {
9450 mddev->pers->finish_reshape(mddev);
9451 if (mddev_is_clustered(mddev))
9455 /* If array is no-longer degraded, then any saved_raid_disk
9456 * information must be scrapped.
9458 if (!mddev->degraded)
9459 rdev_for_each(rdev, mddev)
9460 rdev->saved_raid_disk = -1;
9462 md_update_sb(mddev, 1);
9463 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9464 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9466 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9467 md_cluster_ops->resync_finish(mddev);
9468 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9469 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9470 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9471 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9472 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9473 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9475 * We call md_cluster_ops->update_size here because sync_size could
9476 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9477 * so it is time to update size across cluster.
9479 if (mddev_is_clustered(mddev) && is_reshaped
9480 && !test_bit(MD_CLOSING, &mddev->flags))
9481 md_cluster_ops->update_size(mddev, old_dev_sectors);
9482 wake_up(&resync_wait);
9483 /* flag recovery needed just to double check */
9484 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9485 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9486 sysfs_notify_dirent_safe(mddev->sysfs_action);
9488 if (mddev->event_work.func)
9489 queue_work(md_misc_wq, &mddev->event_work);
9491 EXPORT_SYMBOL(md_reap_sync_thread);
9493 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9495 sysfs_notify_dirent_safe(rdev->sysfs_state);
9496 wait_event_timeout(rdev->blocked_wait,
9497 !test_bit(Blocked, &rdev->flags) &&
9498 !test_bit(BlockedBadBlocks, &rdev->flags),
9499 msecs_to_jiffies(5000));
9500 rdev_dec_pending(rdev, mddev);
9502 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9504 void md_finish_reshape(struct mddev *mddev)
9506 /* called be personality module when reshape completes. */
9507 struct md_rdev *rdev;
9509 rdev_for_each(rdev, mddev) {
9510 if (rdev->data_offset > rdev->new_data_offset)
9511 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9513 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9514 rdev->data_offset = rdev->new_data_offset;
9517 EXPORT_SYMBOL(md_finish_reshape);
9519 /* Bad block management */
9521 /* Returns 1 on success, 0 on failure */
9522 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9525 struct mddev *mddev = rdev->mddev;
9528 s += rdev->new_data_offset;
9530 s += rdev->data_offset;
9531 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9533 /* Make sure they get written out promptly */
9534 if (test_bit(ExternalBbl, &rdev->flags))
9535 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9536 sysfs_notify_dirent_safe(rdev->sysfs_state);
9537 set_mask_bits(&mddev->sb_flags, 0,
9538 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9539 md_wakeup_thread(rdev->mddev->thread);
9544 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9546 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9551 s += rdev->new_data_offset;
9553 s += rdev->data_offset;
9554 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9555 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9556 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9559 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9561 static int md_notify_reboot(struct notifier_block *this,
9562 unsigned long code, void *x)
9564 struct mddev *mddev, *n;
9567 spin_lock(&all_mddevs_lock);
9568 list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9569 if (!mddev_get(mddev))
9571 spin_unlock(&all_mddevs_lock);
9572 if (mddev_trylock(mddev)) {
9574 __md_stop_writes(mddev);
9575 if (mddev->persistent)
9576 mddev->safemode = 2;
9577 mddev_unlock(mddev);
9581 spin_lock(&all_mddevs_lock);
9583 spin_unlock(&all_mddevs_lock);
9586 * certain more exotic SCSI devices are known to be
9587 * volatile wrt too early system reboots. While the
9588 * right place to handle this issue is the given
9589 * driver, we do want to have a safe RAID driver ...
9597 static struct notifier_block md_notifier = {
9598 .notifier_call = md_notify_reboot,
9600 .priority = INT_MAX, /* before any real devices */
9603 static void md_geninit(void)
9605 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9607 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9610 static int __init md_init(void)
9614 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9618 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9622 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9623 if (!md_rdev_misc_wq)
9624 goto err_rdev_misc_wq;
9626 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9630 ret = __register_blkdev(0, "mdp", md_probe);
9635 register_reboot_notifier(&md_notifier);
9636 raid_table_header = register_sysctl("dev/raid", raid_table);
9642 unregister_blkdev(MD_MAJOR, "md");
9644 destroy_workqueue(md_rdev_misc_wq);
9646 destroy_workqueue(md_misc_wq);
9648 destroy_workqueue(md_wq);
9653 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9655 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9656 struct md_rdev *rdev2, *tmp;
9660 * If size is changed in another node then we need to
9661 * do resize as well.
9663 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9664 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9666 pr_info("md-cluster: resize failed\n");
9668 md_bitmap_update_sb(mddev->bitmap);
9671 /* Check for change of roles in the active devices */
9672 rdev_for_each_safe(rdev2, tmp, mddev) {
9673 if (test_bit(Faulty, &rdev2->flags))
9676 /* Check if the roles changed */
9677 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9679 if (test_bit(Candidate, &rdev2->flags)) {
9680 if (role == MD_DISK_ROLE_FAULTY) {
9681 pr_info("md: Removing Candidate device %pg because add failed\n",
9683 md_kick_rdev_from_array(rdev2);
9687 clear_bit(Candidate, &rdev2->flags);
9690 if (role != rdev2->raid_disk) {
9692 * got activated except reshape is happening.
9694 if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9695 !(le32_to_cpu(sb->feature_map) &
9696 MD_FEATURE_RESHAPE_ACTIVE)) {
9697 rdev2->saved_raid_disk = role;
9698 ret = remove_and_add_spares(mddev, rdev2);
9699 pr_info("Activated spare: %pg\n",
9701 /* wakeup mddev->thread here, so array could
9702 * perform resync with the new activated disk */
9703 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9704 md_wakeup_thread(mddev->thread);
9707 * We just want to do the minimum to mark the disk
9708 * as faulty. The recovery is performed by the
9709 * one who initiated the error.
9711 if (role == MD_DISK_ROLE_FAULTY ||
9712 role == MD_DISK_ROLE_JOURNAL) {
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, *n;
9906 unregister_blkdev(MD_MAJOR,"md");
9907 unregister_blkdev(mdp_major, "mdp");
9908 unregister_reboot_notifier(&md_notifier);
9909 unregister_sysctl_table(raid_table_header);
9911 /* We cannot unload the modules while some process is
9912 * waiting for us in select() or poll() - wake them up
9915 while (waitqueue_active(&md_event_waiters)) {
9916 /* not safe to leave yet */
9917 wake_up(&md_event_waiters);
9921 remove_proc_entry("mdstat", NULL);
9923 spin_lock(&all_mddevs_lock);
9924 list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9925 if (!mddev_get(mddev))
9927 spin_unlock(&all_mddevs_lock);
9928 export_array(mddev);
9930 mddev->hold_active = 0;
9932 * As the mddev is now fully clear, mddev_put will schedule
9933 * the mddev for destruction by a workqueue, and the
9934 * destroy_workqueue() below will wait for that to complete.
9937 spin_lock(&all_mddevs_lock);
9939 spin_unlock(&all_mddevs_lock);
9941 destroy_workqueue(md_rdev_misc_wq);
9942 destroy_workqueue(md_misc_wq);
9943 destroy_workqueue(md_wq);
9946 subsys_initcall(md_init);
9947 module_exit(md_exit)
9949 static int get_ro(char *buffer, const struct kernel_param *kp)
9951 return sprintf(buffer, "%d\n", start_readonly);
9953 static int set_ro(const char *val, const struct kernel_param *kp)
9955 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9958 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9959 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9960 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9961 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9963 MODULE_LICENSE("GPL");
9964 MODULE_DESCRIPTION("MD RAID framework");
9966 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);