1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/major.h>
55 #include <linux/module.h>
56 #include <linux/reboot.h>
57 #include <linux/file.h>
58 #include <linux/compat.h>
59 #include <linux/delay.h>
60 #include <linux/raid/md_p.h>
61 #include <linux/raid/md_u.h>
62 #include <linux/raid/detect.h>
63 #include <linux/slab.h>
64 #include <linux/percpu-refcount.h>
65 #include <linux/part_stat.h>
67 #include <trace/events/block.h>
69 #include "md-bitmap.h"
70 #include "md-cluster.h"
72 /* pers_list is a list of registered personalities protected
74 * pers_lock does extra service to protect accesses to
75 * mddev->thread when the mutex cannot be held.
77 static LIST_HEAD(pers_list);
78 static DEFINE_SPINLOCK(pers_lock);
80 static struct kobj_type md_ktype;
82 struct md_cluster_operations *md_cluster_ops;
83 EXPORT_SYMBOL(md_cluster_ops);
84 static struct module *md_cluster_mod;
86 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
87 static struct workqueue_struct *md_wq;
88 static struct workqueue_struct *md_misc_wq;
89 static struct workqueue_struct *md_rdev_misc_wq;
91 static int remove_and_add_spares(struct mddev *mddev,
92 struct md_rdev *this);
93 static void mddev_detach(struct mddev *mddev);
96 * Default number of read corrections we'll attempt on an rdev
97 * before ejecting it from the array. We divide the read error
98 * count by 2 for every hour elapsed between read errors.
100 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
101 /* Default safemode delay: 200 msec */
102 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
104 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
105 * is 1000 KB/sec, so the extra system load does not show up that much.
106 * Increase it if you want to have more _guaranteed_ speed. Note that
107 * the RAID driver will use the maximum available bandwidth if the IO
108 * subsystem is idle. There is also an 'absolute maximum' reconstruction
109 * speed limit - in case reconstruction slows down your system despite
112 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
113 * or /sys/block/mdX/md/sync_speed_{min,max}
116 static int sysctl_speed_limit_min = 1000;
117 static int sysctl_speed_limit_max = 200000;
118 static inline int speed_min(struct mddev *mddev)
120 return mddev->sync_speed_min ?
121 mddev->sync_speed_min : sysctl_speed_limit_min;
124 static inline int speed_max(struct mddev *mddev)
126 return mddev->sync_speed_max ?
127 mddev->sync_speed_max : sysctl_speed_limit_max;
130 static void rdev_uninit_serial(struct md_rdev *rdev)
132 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
135 kvfree(rdev->serial);
139 static void rdevs_uninit_serial(struct mddev *mddev)
141 struct md_rdev *rdev;
143 rdev_for_each(rdev, mddev)
144 rdev_uninit_serial(rdev);
147 static int rdev_init_serial(struct md_rdev *rdev)
149 /* serial_nums equals with BARRIER_BUCKETS_NR */
150 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
151 struct serial_in_rdev *serial = NULL;
153 if (test_bit(CollisionCheck, &rdev->flags))
156 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
161 for (i = 0; i < serial_nums; i++) {
162 struct serial_in_rdev *serial_tmp = &serial[i];
164 spin_lock_init(&serial_tmp->serial_lock);
165 serial_tmp->serial_rb = RB_ROOT_CACHED;
166 init_waitqueue_head(&serial_tmp->serial_io_wait);
169 rdev->serial = serial;
170 set_bit(CollisionCheck, &rdev->flags);
175 static int rdevs_init_serial(struct mddev *mddev)
177 struct md_rdev *rdev;
180 rdev_for_each(rdev, mddev) {
181 ret = rdev_init_serial(rdev);
186 /* Free all resources if pool is not existed */
187 if (ret && !mddev->serial_info_pool)
188 rdevs_uninit_serial(mddev);
194 * rdev needs to enable serial stuffs if it meets the conditions:
195 * 1. it is multi-queue device flaged with writemostly.
196 * 2. the write-behind mode is enabled.
198 static int rdev_need_serial(struct md_rdev *rdev)
200 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
201 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
202 test_bit(WriteMostly, &rdev->flags));
206 * Init resource for rdev(s), then create serial_info_pool if:
207 * 1. rdev is the first device which return true from rdev_enable_serial.
208 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
210 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
215 if (rdev && !rdev_need_serial(rdev) &&
216 !test_bit(CollisionCheck, &rdev->flags))
220 mddev_suspend(mddev);
223 ret = rdevs_init_serial(mddev);
225 ret = rdev_init_serial(rdev);
229 if (mddev->serial_info_pool == NULL) {
231 * already in memalloc noio context by
234 mddev->serial_info_pool =
235 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
236 sizeof(struct serial_info));
237 if (!mddev->serial_info_pool) {
238 rdevs_uninit_serial(mddev);
239 pr_err("can't alloc memory pool for serialization\n");
249 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
250 * 1. rdev is the last device flaged with CollisionCheck.
251 * 2. when bitmap is destroyed while policy is not enabled.
252 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
254 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
257 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
260 if (mddev->serial_info_pool) {
261 struct md_rdev *temp;
262 int num = 0; /* used to track if other rdevs need the pool */
265 mddev_suspend(mddev);
266 rdev_for_each(temp, mddev) {
268 if (!mddev->serialize_policy ||
269 !rdev_need_serial(temp))
270 rdev_uninit_serial(temp);
273 } else if (temp != rdev &&
274 test_bit(CollisionCheck, &temp->flags))
279 rdev_uninit_serial(rdev);
282 pr_info("The mempool could be used by other devices\n");
284 mempool_destroy(mddev->serial_info_pool);
285 mddev->serial_info_pool = NULL;
292 static struct ctl_table_header *raid_table_header;
294 static struct ctl_table raid_table[] = {
296 .procname = "speed_limit_min",
297 .data = &sysctl_speed_limit_min,
298 .maxlen = sizeof(int),
299 .mode = S_IRUGO|S_IWUSR,
300 .proc_handler = proc_dointvec,
303 .procname = "speed_limit_max",
304 .data = &sysctl_speed_limit_max,
305 .maxlen = sizeof(int),
306 .mode = S_IRUGO|S_IWUSR,
307 .proc_handler = proc_dointvec,
312 static struct ctl_table raid_dir_table[] = {
316 .mode = S_IRUGO|S_IXUGO,
322 static struct ctl_table raid_root_table[] = {
327 .child = raid_dir_table,
332 static int start_readonly;
335 * The original mechanism for creating an md device is to create
336 * a device node in /dev and to open it. This causes races with device-close.
337 * The preferred method is to write to the "new_array" module parameter.
338 * This can avoid races.
339 * Setting create_on_open to false disables the original mechanism
340 * so all the races disappear.
342 static bool create_on_open = true;
345 * We have a system wide 'event count' that is incremented
346 * on any 'interesting' event, and readers of /proc/mdstat
347 * can use 'poll' or 'select' to find out when the event
351 * start array, stop array, error, add device, remove device,
352 * start build, activate spare
354 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
355 static atomic_t md_event_count;
356 void md_new_event(struct mddev *mddev)
358 atomic_inc(&md_event_count);
359 wake_up(&md_event_waiters);
361 EXPORT_SYMBOL_GPL(md_new_event);
364 * Enables to iterate over all existing md arrays
365 * all_mddevs_lock protects this list.
367 static LIST_HEAD(all_mddevs);
368 static DEFINE_SPINLOCK(all_mddevs_lock);
371 * iterates through all used mddevs in the system.
372 * We take care to grab the all_mddevs_lock whenever navigating
373 * the list, and to always hold a refcount when unlocked.
374 * Any code which breaks out of this loop while own
375 * a reference to the current mddev and must mddev_put it.
377 #define for_each_mddev(_mddev,_tmp) \
379 for (({ spin_lock(&all_mddevs_lock); \
380 _tmp = all_mddevs.next; \
382 ({ if (_tmp != &all_mddevs) \
383 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
384 spin_unlock(&all_mddevs_lock); \
385 if (_mddev) mddev_put(_mddev); \
386 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
387 _tmp != &all_mddevs;}); \
388 ({ spin_lock(&all_mddevs_lock); \
389 _tmp = _tmp->next;}) \
392 /* Rather than calling directly into the personality make_request function,
393 * IO requests come here first so that we can check if the device is
394 * being suspended pending a reconfiguration.
395 * We hold a refcount over the call to ->make_request. By the time that
396 * call has finished, the bio has been linked into some internal structure
397 * and so is visible to ->quiesce(), so we don't need the refcount any more.
399 static bool is_suspended(struct mddev *mddev, struct bio *bio)
401 if (mddev->suspended)
403 if (bio_data_dir(bio) != WRITE)
405 if (mddev->suspend_lo >= mddev->suspend_hi)
407 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
409 if (bio_end_sector(bio) < mddev->suspend_lo)
414 void md_handle_request(struct mddev *mddev, struct bio *bio)
418 if (is_suspended(mddev, bio)) {
421 prepare_to_wait(&mddev->sb_wait, &__wait,
422 TASK_UNINTERRUPTIBLE);
423 if (!is_suspended(mddev, bio))
429 finish_wait(&mddev->sb_wait, &__wait);
431 atomic_inc(&mddev->active_io);
434 if (!mddev->pers->make_request(mddev, bio)) {
435 atomic_dec(&mddev->active_io);
436 wake_up(&mddev->sb_wait);
437 goto check_suspended;
440 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
441 wake_up(&mddev->sb_wait);
443 EXPORT_SYMBOL(md_handle_request);
445 static blk_qc_t md_submit_bio(struct bio *bio)
447 const int rw = bio_data_dir(bio);
448 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
450 if (mddev == NULL || mddev->pers == NULL) {
452 return BLK_QC_T_NONE;
455 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
457 return BLK_QC_T_NONE;
460 blk_queue_split(&bio);
462 return BLK_QC_T_NONE;
464 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
465 if (bio_sectors(bio) != 0)
466 bio->bi_status = BLK_STS_IOERR;
468 return BLK_QC_T_NONE;
471 /* bio could be mergeable after passing to underlayer */
472 bio->bi_opf &= ~REQ_NOMERGE;
474 md_handle_request(mddev, bio);
476 return BLK_QC_T_NONE;
479 /* mddev_suspend makes sure no new requests are submitted
480 * to the device, and that any requests that have been submitted
481 * are completely handled.
482 * Once mddev_detach() is called and completes, the module will be
485 void mddev_suspend(struct mddev *mddev)
487 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
488 lockdep_assert_held(&mddev->reconfig_mutex);
489 if (mddev->suspended++)
492 wake_up(&mddev->sb_wait);
493 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
494 smp_mb__after_atomic();
495 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
496 mddev->pers->quiesce(mddev, 1);
497 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
498 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
500 del_timer_sync(&mddev->safemode_timer);
501 /* restrict memory reclaim I/O during raid array is suspend */
502 mddev->noio_flag = memalloc_noio_save();
504 EXPORT_SYMBOL_GPL(mddev_suspend);
506 void mddev_resume(struct mddev *mddev)
508 /* entred the memalloc scope from mddev_suspend() */
509 memalloc_noio_restore(mddev->noio_flag);
510 lockdep_assert_held(&mddev->reconfig_mutex);
511 if (--mddev->suspended)
513 wake_up(&mddev->sb_wait);
514 mddev->pers->quiesce(mddev, 0);
516 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
517 md_wakeup_thread(mddev->thread);
518 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
520 EXPORT_SYMBOL_GPL(mddev_resume);
523 * Generic flush handling for md
526 static void md_end_flush(struct bio *bio)
528 struct md_rdev *rdev = bio->bi_private;
529 struct mddev *mddev = rdev->mddev;
533 rdev_dec_pending(rdev, mddev);
535 if (atomic_dec_and_test(&mddev->flush_pending)) {
536 /* The pre-request flush has finished */
537 queue_work(md_wq, &mddev->flush_work);
541 static void md_submit_flush_data(struct work_struct *ws);
543 static void submit_flushes(struct work_struct *ws)
545 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
546 struct md_rdev *rdev;
548 mddev->start_flush = ktime_get_boottime();
549 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
550 atomic_set(&mddev->flush_pending, 1);
552 rdev_for_each_rcu(rdev, mddev)
553 if (rdev->raid_disk >= 0 &&
554 !test_bit(Faulty, &rdev->flags)) {
555 /* Take two references, one is dropped
556 * when request finishes, one after
557 * we reclaim rcu_read_lock
560 atomic_inc(&rdev->nr_pending);
561 atomic_inc(&rdev->nr_pending);
563 bi = bio_alloc_bioset(GFP_NOIO, 0, &mddev->bio_set);
564 bi->bi_end_io = md_end_flush;
565 bi->bi_private = rdev;
566 bio_set_dev(bi, rdev->bdev);
567 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
568 atomic_inc(&mddev->flush_pending);
571 rdev_dec_pending(rdev, mddev);
574 if (atomic_dec_and_test(&mddev->flush_pending))
575 queue_work(md_wq, &mddev->flush_work);
578 static void md_submit_flush_data(struct work_struct *ws)
580 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
581 struct bio *bio = mddev->flush_bio;
584 * must reset flush_bio before calling into md_handle_request to avoid a
585 * deadlock, because other bios passed md_handle_request suspend check
586 * could wait for this and below md_handle_request could wait for those
587 * bios because of suspend check
589 spin_lock_irq(&mddev->lock);
590 mddev->prev_flush_start = mddev->start_flush;
591 mddev->flush_bio = NULL;
592 spin_unlock_irq(&mddev->lock);
593 wake_up(&mddev->sb_wait);
595 if (bio->bi_iter.bi_size == 0) {
596 /* an empty barrier - all done */
599 bio->bi_opf &= ~REQ_PREFLUSH;
600 md_handle_request(mddev, bio);
605 * Manages consolidation of flushes and submitting any flushes needed for
606 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
607 * being finished in another context. Returns false if the flushing is
608 * complete but still needs the I/O portion of the bio to be processed.
610 bool md_flush_request(struct mddev *mddev, struct bio *bio)
612 ktime_t req_start = ktime_get_boottime();
613 spin_lock_irq(&mddev->lock);
614 /* flush requests wait until ongoing flush completes,
615 * hence coalescing all the pending requests.
617 wait_event_lock_irq(mddev->sb_wait,
619 ktime_before(req_start, mddev->prev_flush_start),
621 /* new request after previous flush is completed */
622 if (ktime_after(req_start, mddev->prev_flush_start)) {
623 WARN_ON(mddev->flush_bio);
624 mddev->flush_bio = bio;
627 spin_unlock_irq(&mddev->lock);
630 INIT_WORK(&mddev->flush_work, submit_flushes);
631 queue_work(md_wq, &mddev->flush_work);
633 /* flush was performed for some other bio while we waited. */
634 if (bio->bi_iter.bi_size == 0)
635 /* an empty barrier - all done */
638 bio->bi_opf &= ~REQ_PREFLUSH;
644 EXPORT_SYMBOL(md_flush_request);
646 static inline struct mddev *mddev_get(struct mddev *mddev)
648 atomic_inc(&mddev->active);
652 static void mddev_delayed_delete(struct work_struct *ws);
654 static void mddev_put(struct mddev *mddev)
656 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
658 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
659 mddev->ctime == 0 && !mddev->hold_active) {
660 /* Array is not configured at all, and not held active,
662 list_del_init(&mddev->all_mddevs);
665 * Call queue_work inside the spinlock so that
666 * flush_workqueue() after mddev_find will succeed in waiting
667 * for the work to be done.
669 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
670 queue_work(md_misc_wq, &mddev->del_work);
672 spin_unlock(&all_mddevs_lock);
675 static void md_safemode_timeout(struct timer_list *t);
677 void mddev_init(struct mddev *mddev)
679 kobject_init(&mddev->kobj, &md_ktype);
680 mutex_init(&mddev->open_mutex);
681 mutex_init(&mddev->reconfig_mutex);
682 mutex_init(&mddev->bitmap_info.mutex);
683 INIT_LIST_HEAD(&mddev->disks);
684 INIT_LIST_HEAD(&mddev->all_mddevs);
685 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
686 atomic_set(&mddev->active, 1);
687 atomic_set(&mddev->openers, 0);
688 atomic_set(&mddev->active_io, 0);
689 spin_lock_init(&mddev->lock);
690 atomic_set(&mddev->flush_pending, 0);
691 init_waitqueue_head(&mddev->sb_wait);
692 init_waitqueue_head(&mddev->recovery_wait);
693 mddev->reshape_position = MaxSector;
694 mddev->reshape_backwards = 0;
695 mddev->last_sync_action = "none";
696 mddev->resync_min = 0;
697 mddev->resync_max = MaxSector;
698 mddev->level = LEVEL_NONE;
700 EXPORT_SYMBOL_GPL(mddev_init);
702 static struct mddev *mddev_find_locked(dev_t unit)
706 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
707 if (mddev->unit == unit)
713 /* find an unused unit number */
714 static dev_t mddev_alloc_unit(void)
716 static int next_minor = 512;
717 int start = next_minor;
722 dev = MKDEV(MD_MAJOR, next_minor);
724 if (next_minor > MINORMASK)
726 if (next_minor == start)
727 return 0; /* Oh dear, all in use. */
728 is_free = !mddev_find_locked(dev);
734 static struct mddev *mddev_find(dev_t unit)
738 if (MAJOR(unit) != MD_MAJOR)
739 unit &= ~((1 << MdpMinorShift) - 1);
741 spin_lock(&all_mddevs_lock);
742 mddev = mddev_find_locked(unit);
745 spin_unlock(&all_mddevs_lock);
750 static struct mddev *mddev_alloc(dev_t unit)
755 if (unit && MAJOR(unit) != MD_MAJOR)
756 unit &= ~((1 << MdpMinorShift) - 1);
758 new = kzalloc(sizeof(*new), GFP_KERNEL);
760 return ERR_PTR(-ENOMEM);
763 spin_lock(&all_mddevs_lock);
766 if (mddev_find_locked(unit))
769 if (MAJOR(unit) == MD_MAJOR)
770 new->md_minor = MINOR(unit);
772 new->md_minor = MINOR(unit) >> MdpMinorShift;
773 new->hold_active = UNTIL_IOCTL;
776 new->unit = mddev_alloc_unit();
779 new->md_minor = MINOR(new->unit);
780 new->hold_active = UNTIL_STOP;
783 list_add(&new->all_mddevs, &all_mddevs);
784 spin_unlock(&all_mddevs_lock);
787 spin_unlock(&all_mddevs_lock);
789 return ERR_PTR(error);
792 static const struct attribute_group md_redundancy_group;
794 void mddev_unlock(struct mddev *mddev)
796 if (mddev->to_remove) {
797 /* These cannot be removed under reconfig_mutex as
798 * an access to the files will try to take reconfig_mutex
799 * while holding the file unremovable, which leads to
801 * So hold set sysfs_active while the remove in happeing,
802 * and anything else which might set ->to_remove or my
803 * otherwise change the sysfs namespace will fail with
804 * -EBUSY if sysfs_active is still set.
805 * We set sysfs_active under reconfig_mutex and elsewhere
806 * test it under the same mutex to ensure its correct value
809 const struct attribute_group *to_remove = mddev->to_remove;
810 mddev->to_remove = NULL;
811 mddev->sysfs_active = 1;
812 mutex_unlock(&mddev->reconfig_mutex);
814 if (mddev->kobj.sd) {
815 if (to_remove != &md_redundancy_group)
816 sysfs_remove_group(&mddev->kobj, to_remove);
817 if (mddev->pers == NULL ||
818 mddev->pers->sync_request == NULL) {
819 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
820 if (mddev->sysfs_action)
821 sysfs_put(mddev->sysfs_action);
822 if (mddev->sysfs_completed)
823 sysfs_put(mddev->sysfs_completed);
824 if (mddev->sysfs_degraded)
825 sysfs_put(mddev->sysfs_degraded);
826 mddev->sysfs_action = NULL;
827 mddev->sysfs_completed = NULL;
828 mddev->sysfs_degraded = NULL;
831 mddev->sysfs_active = 0;
833 mutex_unlock(&mddev->reconfig_mutex);
835 /* As we've dropped the mutex we need a spinlock to
836 * make sure the thread doesn't disappear
838 spin_lock(&pers_lock);
839 md_wakeup_thread(mddev->thread);
840 wake_up(&mddev->sb_wait);
841 spin_unlock(&pers_lock);
843 EXPORT_SYMBOL_GPL(mddev_unlock);
845 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
847 struct md_rdev *rdev;
849 rdev_for_each_rcu(rdev, mddev)
850 if (rdev->desc_nr == nr)
855 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
857 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
859 struct md_rdev *rdev;
861 rdev_for_each(rdev, mddev)
862 if (rdev->bdev->bd_dev == dev)
868 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
870 struct md_rdev *rdev;
872 rdev_for_each_rcu(rdev, mddev)
873 if (rdev->bdev->bd_dev == dev)
878 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
880 static struct md_personality *find_pers(int level, char *clevel)
882 struct md_personality *pers;
883 list_for_each_entry(pers, &pers_list, list) {
884 if (level != LEVEL_NONE && pers->level == level)
886 if (strcmp(pers->name, clevel)==0)
892 /* return the offset of the super block in 512byte sectors */
893 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
895 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
896 return MD_NEW_SIZE_SECTORS(num_sectors);
899 static int alloc_disk_sb(struct md_rdev *rdev)
901 rdev->sb_page = alloc_page(GFP_KERNEL);
907 void md_rdev_clear(struct md_rdev *rdev)
910 put_page(rdev->sb_page);
912 rdev->sb_page = NULL;
917 put_page(rdev->bb_page);
918 rdev->bb_page = NULL;
920 badblocks_exit(&rdev->badblocks);
922 EXPORT_SYMBOL_GPL(md_rdev_clear);
924 static void super_written(struct bio *bio)
926 struct md_rdev *rdev = bio->bi_private;
927 struct mddev *mddev = rdev->mddev;
929 if (bio->bi_status) {
930 pr_err("md: %s gets error=%d\n", __func__,
931 blk_status_to_errno(bio->bi_status));
932 md_error(mddev, rdev);
933 if (!test_bit(Faulty, &rdev->flags)
934 && (bio->bi_opf & MD_FAILFAST)) {
935 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
936 set_bit(LastDev, &rdev->flags);
939 clear_bit(LastDev, &rdev->flags);
943 rdev_dec_pending(rdev, mddev);
945 if (atomic_dec_and_test(&mddev->pending_writes))
946 wake_up(&mddev->sb_wait);
949 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
950 sector_t sector, int size, struct page *page)
952 /* write first size bytes of page to sector of rdev
953 * Increment mddev->pending_writes before returning
954 * and decrement it on completion, waking up sb_wait
955 * if zero is reached.
956 * If an error occurred, call md_error
964 if (test_bit(Faulty, &rdev->flags))
967 bio = bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
969 atomic_inc(&rdev->nr_pending);
971 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
972 bio->bi_iter.bi_sector = sector;
973 bio_add_page(bio, page, size, 0);
974 bio->bi_private = rdev;
975 bio->bi_end_io = super_written;
977 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
978 test_bit(FailFast, &rdev->flags) &&
979 !test_bit(LastDev, &rdev->flags))
981 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
983 atomic_inc(&mddev->pending_writes);
987 int md_super_wait(struct mddev *mddev)
989 /* wait for all superblock writes that were scheduled to complete */
990 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
991 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
996 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
997 struct page *page, int op, int op_flags, bool metadata_op)
1000 struct bio_vec bvec;
1002 bio_init(&bio, &bvec, 1);
1004 if (metadata_op && rdev->meta_bdev)
1005 bio_set_dev(&bio, rdev->meta_bdev);
1007 bio_set_dev(&bio, rdev->bdev);
1008 bio.bi_opf = op | op_flags;
1010 bio.bi_iter.bi_sector = sector + rdev->sb_start;
1011 else if (rdev->mddev->reshape_position != MaxSector &&
1012 (rdev->mddev->reshape_backwards ==
1013 (sector >= rdev->mddev->reshape_position)))
1014 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1016 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1017 bio_add_page(&bio, page, size, 0);
1019 submit_bio_wait(&bio);
1021 return !bio.bi_status;
1023 EXPORT_SYMBOL_GPL(sync_page_io);
1025 static int read_disk_sb(struct md_rdev *rdev, int size)
1027 char b[BDEVNAME_SIZE];
1029 if (rdev->sb_loaded)
1032 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1034 rdev->sb_loaded = 1;
1038 pr_err("md: disabled device %s, could not read superblock.\n",
1039 bdevname(rdev->bdev,b));
1043 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1045 return sb1->set_uuid0 == sb2->set_uuid0 &&
1046 sb1->set_uuid1 == sb2->set_uuid1 &&
1047 sb1->set_uuid2 == sb2->set_uuid2 &&
1048 sb1->set_uuid3 == sb2->set_uuid3;
1051 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1054 mdp_super_t *tmp1, *tmp2;
1056 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1057 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1059 if (!tmp1 || !tmp2) {
1068 * nr_disks is not constant
1073 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1080 static u32 md_csum_fold(u32 csum)
1082 csum = (csum & 0xffff) + (csum >> 16);
1083 return (csum & 0xffff) + (csum >> 16);
1086 static unsigned int calc_sb_csum(mdp_super_t *sb)
1089 u32 *sb32 = (u32*)sb;
1091 unsigned int disk_csum, csum;
1093 disk_csum = sb->sb_csum;
1096 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1098 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1101 /* This used to use csum_partial, which was wrong for several
1102 * reasons including that different results are returned on
1103 * different architectures. It isn't critical that we get exactly
1104 * the same return value as before (we always csum_fold before
1105 * testing, and that removes any differences). However as we
1106 * know that csum_partial always returned a 16bit value on
1107 * alphas, do a fold to maximise conformity to previous behaviour.
1109 sb->sb_csum = md_csum_fold(disk_csum);
1111 sb->sb_csum = disk_csum;
1117 * Handle superblock details.
1118 * We want to be able to handle multiple superblock formats
1119 * so we have a common interface to them all, and an array of
1120 * different handlers.
1121 * We rely on user-space to write the initial superblock, and support
1122 * reading and updating of superblocks.
1123 * Interface methods are:
1124 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1125 * loads and validates a superblock on dev.
1126 * if refdev != NULL, compare superblocks on both devices
1128 * 0 - dev has a superblock that is compatible with refdev
1129 * 1 - dev has a superblock that is compatible and newer than refdev
1130 * so dev should be used as the refdev in future
1131 * -EINVAL superblock incompatible or invalid
1132 * -othererror e.g. -EIO
1134 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1135 * Verify that dev is acceptable into mddev.
1136 * The first time, mddev->raid_disks will be 0, and data from
1137 * dev should be merged in. Subsequent calls check that dev
1138 * is new enough. Return 0 or -EINVAL
1140 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1141 * Update the superblock for rdev with data in mddev
1142 * This does not write to disc.
1148 struct module *owner;
1149 int (*load_super)(struct md_rdev *rdev,
1150 struct md_rdev *refdev,
1152 int (*validate_super)(struct mddev *mddev,
1153 struct md_rdev *rdev);
1154 void (*sync_super)(struct mddev *mddev,
1155 struct md_rdev *rdev);
1156 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1157 sector_t num_sectors);
1158 int (*allow_new_offset)(struct md_rdev *rdev,
1159 unsigned long long new_offset);
1163 * Check that the given mddev has no bitmap.
1165 * This function is called from the run method of all personalities that do not
1166 * support bitmaps. It prints an error message and returns non-zero if mddev
1167 * has a bitmap. Otherwise, it returns 0.
1170 int md_check_no_bitmap(struct mddev *mddev)
1172 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1174 pr_warn("%s: bitmaps are not supported for %s\n",
1175 mdname(mddev), mddev->pers->name);
1178 EXPORT_SYMBOL(md_check_no_bitmap);
1181 * load_super for 0.90.0
1183 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1185 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1188 bool spare_disk = true;
1191 * Calculate the position of the superblock (512byte sectors),
1192 * it's at the end of the disk.
1194 * It also happens to be a multiple of 4Kb.
1196 rdev->sb_start = calc_dev_sboffset(rdev);
1198 ret = read_disk_sb(rdev, MD_SB_BYTES);
1204 bdevname(rdev->bdev, b);
1205 sb = page_address(rdev->sb_page);
1207 if (sb->md_magic != MD_SB_MAGIC) {
1208 pr_warn("md: invalid raid superblock magic on %s\n", b);
1212 if (sb->major_version != 0 ||
1213 sb->minor_version < 90 ||
1214 sb->minor_version > 91) {
1215 pr_warn("Bad version number %d.%d on %s\n",
1216 sb->major_version, sb->minor_version, b);
1220 if (sb->raid_disks <= 0)
1223 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1224 pr_warn("md: invalid superblock checksum on %s\n", b);
1228 rdev->preferred_minor = sb->md_minor;
1229 rdev->data_offset = 0;
1230 rdev->new_data_offset = 0;
1231 rdev->sb_size = MD_SB_BYTES;
1232 rdev->badblocks.shift = -1;
1234 if (sb->level == LEVEL_MULTIPATH)
1237 rdev->desc_nr = sb->this_disk.number;
1239 /* not spare disk, or LEVEL_MULTIPATH */
1240 if (sb->level == LEVEL_MULTIPATH ||
1241 (rdev->desc_nr >= 0 &&
1242 rdev->desc_nr < MD_SB_DISKS &&
1243 sb->disks[rdev->desc_nr].state &
1244 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1254 mdp_super_t *refsb = page_address(refdev->sb_page);
1255 if (!md_uuid_equal(refsb, sb)) {
1256 pr_warn("md: %s has different UUID to %s\n",
1257 b, bdevname(refdev->bdev,b2));
1260 if (!md_sb_equal(refsb, sb)) {
1261 pr_warn("md: %s has same UUID but different superblock to %s\n",
1262 b, bdevname(refdev->bdev, b2));
1266 ev2 = md_event(refsb);
1268 if (!spare_disk && ev1 > ev2)
1273 rdev->sectors = rdev->sb_start;
1274 /* Limit to 4TB as metadata cannot record more than that.
1275 * (not needed for Linear and RAID0 as metadata doesn't
1278 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1279 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1281 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1282 /* "this cannot possibly happen" ... */
1290 * validate_super for 0.90.0
1292 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1295 mdp_super_t *sb = page_address(rdev->sb_page);
1296 __u64 ev1 = md_event(sb);
1298 rdev->raid_disk = -1;
1299 clear_bit(Faulty, &rdev->flags);
1300 clear_bit(In_sync, &rdev->flags);
1301 clear_bit(Bitmap_sync, &rdev->flags);
1302 clear_bit(WriteMostly, &rdev->flags);
1304 if (mddev->raid_disks == 0) {
1305 mddev->major_version = 0;
1306 mddev->minor_version = sb->minor_version;
1307 mddev->patch_version = sb->patch_version;
1308 mddev->external = 0;
1309 mddev->chunk_sectors = sb->chunk_size >> 9;
1310 mddev->ctime = sb->ctime;
1311 mddev->utime = sb->utime;
1312 mddev->level = sb->level;
1313 mddev->clevel[0] = 0;
1314 mddev->layout = sb->layout;
1315 mddev->raid_disks = sb->raid_disks;
1316 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1317 mddev->events = ev1;
1318 mddev->bitmap_info.offset = 0;
1319 mddev->bitmap_info.space = 0;
1320 /* bitmap can use 60 K after the 4K superblocks */
1321 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1322 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1323 mddev->reshape_backwards = 0;
1325 if (mddev->minor_version >= 91) {
1326 mddev->reshape_position = sb->reshape_position;
1327 mddev->delta_disks = sb->delta_disks;
1328 mddev->new_level = sb->new_level;
1329 mddev->new_layout = sb->new_layout;
1330 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1331 if (mddev->delta_disks < 0)
1332 mddev->reshape_backwards = 1;
1334 mddev->reshape_position = MaxSector;
1335 mddev->delta_disks = 0;
1336 mddev->new_level = mddev->level;
1337 mddev->new_layout = mddev->layout;
1338 mddev->new_chunk_sectors = mddev->chunk_sectors;
1340 if (mddev->level == 0)
1343 if (sb->state & (1<<MD_SB_CLEAN))
1344 mddev->recovery_cp = MaxSector;
1346 if (sb->events_hi == sb->cp_events_hi &&
1347 sb->events_lo == sb->cp_events_lo) {
1348 mddev->recovery_cp = sb->recovery_cp;
1350 mddev->recovery_cp = 0;
1353 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1354 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1355 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1356 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1358 mddev->max_disks = MD_SB_DISKS;
1360 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1361 mddev->bitmap_info.file == NULL) {
1362 mddev->bitmap_info.offset =
1363 mddev->bitmap_info.default_offset;
1364 mddev->bitmap_info.space =
1365 mddev->bitmap_info.default_space;
1368 } else if (mddev->pers == NULL) {
1369 /* Insist on good event counter while assembling, except
1370 * for spares (which don't need an event count) */
1372 if (sb->disks[rdev->desc_nr].state & (
1373 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1374 if (ev1 < mddev->events)
1376 } else if (mddev->bitmap) {
1377 /* if adding to array with a bitmap, then we can accept an
1378 * older device ... but not too old.
1380 if (ev1 < mddev->bitmap->events_cleared)
1382 if (ev1 < mddev->events)
1383 set_bit(Bitmap_sync, &rdev->flags);
1385 if (ev1 < mddev->events)
1386 /* just a hot-add of a new device, leave raid_disk at -1 */
1390 if (mddev->level != LEVEL_MULTIPATH) {
1391 desc = sb->disks + rdev->desc_nr;
1393 if (desc->state & (1<<MD_DISK_FAULTY))
1394 set_bit(Faulty, &rdev->flags);
1395 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1396 desc->raid_disk < mddev->raid_disks */) {
1397 set_bit(In_sync, &rdev->flags);
1398 rdev->raid_disk = desc->raid_disk;
1399 rdev->saved_raid_disk = desc->raid_disk;
1400 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1401 /* active but not in sync implies recovery up to
1402 * reshape position. We don't know exactly where
1403 * that is, so set to zero for now */
1404 if (mddev->minor_version >= 91) {
1405 rdev->recovery_offset = 0;
1406 rdev->raid_disk = desc->raid_disk;
1409 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1410 set_bit(WriteMostly, &rdev->flags);
1411 if (desc->state & (1<<MD_DISK_FAILFAST))
1412 set_bit(FailFast, &rdev->flags);
1413 } else /* MULTIPATH are always insync */
1414 set_bit(In_sync, &rdev->flags);
1419 * sync_super for 0.90.0
1421 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1424 struct md_rdev *rdev2;
1425 int next_spare = mddev->raid_disks;
1427 /* make rdev->sb match mddev data..
1430 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1431 * 3/ any empty disks < next_spare become removed
1433 * disks[0] gets initialised to REMOVED because
1434 * we cannot be sure from other fields if it has
1435 * been initialised or not.
1438 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1440 rdev->sb_size = MD_SB_BYTES;
1442 sb = page_address(rdev->sb_page);
1444 memset(sb, 0, sizeof(*sb));
1446 sb->md_magic = MD_SB_MAGIC;
1447 sb->major_version = mddev->major_version;
1448 sb->patch_version = mddev->patch_version;
1449 sb->gvalid_words = 0; /* ignored */
1450 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1451 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1452 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1453 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1455 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1456 sb->level = mddev->level;
1457 sb->size = mddev->dev_sectors / 2;
1458 sb->raid_disks = mddev->raid_disks;
1459 sb->md_minor = mddev->md_minor;
1460 sb->not_persistent = 0;
1461 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1463 sb->events_hi = (mddev->events>>32);
1464 sb->events_lo = (u32)mddev->events;
1466 if (mddev->reshape_position == MaxSector)
1467 sb->minor_version = 90;
1469 sb->minor_version = 91;
1470 sb->reshape_position = mddev->reshape_position;
1471 sb->new_level = mddev->new_level;
1472 sb->delta_disks = mddev->delta_disks;
1473 sb->new_layout = mddev->new_layout;
1474 sb->new_chunk = mddev->new_chunk_sectors << 9;
1476 mddev->minor_version = sb->minor_version;
1479 sb->recovery_cp = mddev->recovery_cp;
1480 sb->cp_events_hi = (mddev->events>>32);
1481 sb->cp_events_lo = (u32)mddev->events;
1482 if (mddev->recovery_cp == MaxSector)
1483 sb->state = (1<< MD_SB_CLEAN);
1485 sb->recovery_cp = 0;
1487 sb->layout = mddev->layout;
1488 sb->chunk_size = mddev->chunk_sectors << 9;
1490 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1491 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1493 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1494 rdev_for_each(rdev2, mddev) {
1497 int is_active = test_bit(In_sync, &rdev2->flags);
1499 if (rdev2->raid_disk >= 0 &&
1500 sb->minor_version >= 91)
1501 /* we have nowhere to store the recovery_offset,
1502 * but if it is not below the reshape_position,
1503 * we can piggy-back on that.
1506 if (rdev2->raid_disk < 0 ||
1507 test_bit(Faulty, &rdev2->flags))
1510 desc_nr = rdev2->raid_disk;
1512 desc_nr = next_spare++;
1513 rdev2->desc_nr = desc_nr;
1514 d = &sb->disks[rdev2->desc_nr];
1516 d->number = rdev2->desc_nr;
1517 d->major = MAJOR(rdev2->bdev->bd_dev);
1518 d->minor = MINOR(rdev2->bdev->bd_dev);
1520 d->raid_disk = rdev2->raid_disk;
1522 d->raid_disk = rdev2->desc_nr; /* compatibility */
1523 if (test_bit(Faulty, &rdev2->flags))
1524 d->state = (1<<MD_DISK_FAULTY);
1525 else if (is_active) {
1526 d->state = (1<<MD_DISK_ACTIVE);
1527 if (test_bit(In_sync, &rdev2->flags))
1528 d->state |= (1<<MD_DISK_SYNC);
1536 if (test_bit(WriteMostly, &rdev2->flags))
1537 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1538 if (test_bit(FailFast, &rdev2->flags))
1539 d->state |= (1<<MD_DISK_FAILFAST);
1541 /* now set the "removed" and "faulty" bits on any missing devices */
1542 for (i=0 ; i < mddev->raid_disks ; i++) {
1543 mdp_disk_t *d = &sb->disks[i];
1544 if (d->state == 0 && d->number == 0) {
1547 d->state = (1<<MD_DISK_REMOVED);
1548 d->state |= (1<<MD_DISK_FAULTY);
1552 sb->nr_disks = nr_disks;
1553 sb->active_disks = active;
1554 sb->working_disks = working;
1555 sb->failed_disks = failed;
1556 sb->spare_disks = spare;
1558 sb->this_disk = sb->disks[rdev->desc_nr];
1559 sb->sb_csum = calc_sb_csum(sb);
1563 * rdev_size_change for 0.90.0
1565 static unsigned long long
1566 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1568 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1569 return 0; /* component must fit device */
1570 if (rdev->mddev->bitmap_info.offset)
1571 return 0; /* can't move bitmap */
1572 rdev->sb_start = calc_dev_sboffset(rdev);
1573 if (!num_sectors || num_sectors > rdev->sb_start)
1574 num_sectors = rdev->sb_start;
1575 /* Limit to 4TB as metadata cannot record more than that.
1576 * 4TB == 2^32 KB, or 2*2^32 sectors.
1578 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1579 num_sectors = (sector_t)(2ULL << 32) - 2;
1581 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1583 } while (md_super_wait(rdev->mddev) < 0);
1588 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1590 /* non-zero offset changes not possible with v0.90 */
1591 return new_offset == 0;
1595 * version 1 superblock
1598 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1602 unsigned long long newcsum;
1603 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1604 __le32 *isuper = (__le32*)sb;
1606 disk_csum = sb->sb_csum;
1609 for (; size >= 4; size -= 4)
1610 newcsum += le32_to_cpu(*isuper++);
1613 newcsum += le16_to_cpu(*(__le16*) isuper);
1615 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1616 sb->sb_csum = disk_csum;
1617 return cpu_to_le32(csum);
1620 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1622 struct mdp_superblock_1 *sb;
1626 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1628 bool spare_disk = true;
1631 * Calculate the position of the superblock in 512byte sectors.
1632 * It is always aligned to a 4K boundary and
1633 * depeding on minor_version, it can be:
1634 * 0: At least 8K, but less than 12K, from end of device
1635 * 1: At start of device
1636 * 2: 4K from start of device.
1638 switch(minor_version) {
1640 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1642 sb_start &= ~(sector_t)(4*2-1);
1653 rdev->sb_start = sb_start;
1655 /* superblock is rarely larger than 1K, but it can be larger,
1656 * and it is safe to read 4k, so we do that
1658 ret = read_disk_sb(rdev, 4096);
1659 if (ret) return ret;
1661 sb = page_address(rdev->sb_page);
1663 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1664 sb->major_version != cpu_to_le32(1) ||
1665 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1666 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1667 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1670 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1671 pr_warn("md: invalid superblock checksum on %s\n",
1672 bdevname(rdev->bdev,b));
1675 if (le64_to_cpu(sb->data_size) < 10) {
1676 pr_warn("md: data_size too small on %s\n",
1677 bdevname(rdev->bdev,b));
1682 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1683 /* Some padding is non-zero, might be a new feature */
1686 rdev->preferred_minor = 0xffff;
1687 rdev->data_offset = le64_to_cpu(sb->data_offset);
1688 rdev->new_data_offset = rdev->data_offset;
1689 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1690 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1691 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1692 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1694 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1695 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1696 if (rdev->sb_size & bmask)
1697 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1700 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1703 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1706 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1709 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1711 if (!rdev->bb_page) {
1712 rdev->bb_page = alloc_page(GFP_KERNEL);
1716 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1717 rdev->badblocks.count == 0) {
1718 /* need to load the bad block list.
1719 * Currently we limit it to one page.
1725 int sectors = le16_to_cpu(sb->bblog_size);
1726 if (sectors > (PAGE_SIZE / 512))
1728 offset = le32_to_cpu(sb->bblog_offset);
1731 bb_sector = (long long)offset;
1732 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1733 rdev->bb_page, REQ_OP_READ, 0, true))
1735 bbp = (__le64 *)page_address(rdev->bb_page);
1736 rdev->badblocks.shift = sb->bblog_shift;
1737 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1738 u64 bb = le64_to_cpu(*bbp);
1739 int count = bb & (0x3ff);
1740 u64 sector = bb >> 10;
1741 sector <<= sb->bblog_shift;
1742 count <<= sb->bblog_shift;
1745 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1748 } else if (sb->bblog_offset != 0)
1749 rdev->badblocks.shift = 0;
1751 if ((le32_to_cpu(sb->feature_map) &
1752 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1753 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1754 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1755 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1758 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1762 /* not spare disk, or LEVEL_MULTIPATH */
1763 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1764 (rdev->desc_nr >= 0 &&
1765 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1766 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1767 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1777 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1779 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1780 sb->level != refsb->level ||
1781 sb->layout != refsb->layout ||
1782 sb->chunksize != refsb->chunksize) {
1783 pr_warn("md: %s has strangely different superblock to %s\n",
1784 bdevname(rdev->bdev,b),
1785 bdevname(refdev->bdev,b2));
1788 ev1 = le64_to_cpu(sb->events);
1789 ev2 = le64_to_cpu(refsb->events);
1791 if (!spare_disk && ev1 > ev2)
1796 if (minor_version) {
1797 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1798 sectors -= rdev->data_offset;
1800 sectors = rdev->sb_start;
1801 if (sectors < le64_to_cpu(sb->data_size))
1803 rdev->sectors = le64_to_cpu(sb->data_size);
1807 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1809 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1810 __u64 ev1 = le64_to_cpu(sb->events);
1812 rdev->raid_disk = -1;
1813 clear_bit(Faulty, &rdev->flags);
1814 clear_bit(In_sync, &rdev->flags);
1815 clear_bit(Bitmap_sync, &rdev->flags);
1816 clear_bit(WriteMostly, &rdev->flags);
1818 if (mddev->raid_disks == 0) {
1819 mddev->major_version = 1;
1820 mddev->patch_version = 0;
1821 mddev->external = 0;
1822 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1823 mddev->ctime = le64_to_cpu(sb->ctime);
1824 mddev->utime = le64_to_cpu(sb->utime);
1825 mddev->level = le32_to_cpu(sb->level);
1826 mddev->clevel[0] = 0;
1827 mddev->layout = le32_to_cpu(sb->layout);
1828 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1829 mddev->dev_sectors = le64_to_cpu(sb->size);
1830 mddev->events = ev1;
1831 mddev->bitmap_info.offset = 0;
1832 mddev->bitmap_info.space = 0;
1833 /* Default location for bitmap is 1K after superblock
1834 * using 3K - total of 4K
1836 mddev->bitmap_info.default_offset = 1024 >> 9;
1837 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1838 mddev->reshape_backwards = 0;
1840 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1841 memcpy(mddev->uuid, sb->set_uuid, 16);
1843 mddev->max_disks = (4096-256)/2;
1845 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1846 mddev->bitmap_info.file == NULL) {
1847 mddev->bitmap_info.offset =
1848 (__s32)le32_to_cpu(sb->bitmap_offset);
1849 /* Metadata doesn't record how much space is available.
1850 * For 1.0, we assume we can use up to the superblock
1851 * if before, else to 4K beyond superblock.
1852 * For others, assume no change is possible.
1854 if (mddev->minor_version > 0)
1855 mddev->bitmap_info.space = 0;
1856 else if (mddev->bitmap_info.offset > 0)
1857 mddev->bitmap_info.space =
1858 8 - mddev->bitmap_info.offset;
1860 mddev->bitmap_info.space =
1861 -mddev->bitmap_info.offset;
1864 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1865 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1866 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1867 mddev->new_level = le32_to_cpu(sb->new_level);
1868 mddev->new_layout = le32_to_cpu(sb->new_layout);
1869 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1870 if (mddev->delta_disks < 0 ||
1871 (mddev->delta_disks == 0 &&
1872 (le32_to_cpu(sb->feature_map)
1873 & MD_FEATURE_RESHAPE_BACKWARDS)))
1874 mddev->reshape_backwards = 1;
1876 mddev->reshape_position = MaxSector;
1877 mddev->delta_disks = 0;
1878 mddev->new_level = mddev->level;
1879 mddev->new_layout = mddev->layout;
1880 mddev->new_chunk_sectors = mddev->chunk_sectors;
1883 if (mddev->level == 0 &&
1884 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1887 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1888 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1890 if (le32_to_cpu(sb->feature_map) &
1891 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1892 if (le32_to_cpu(sb->feature_map) &
1893 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1895 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1896 (le32_to_cpu(sb->feature_map) &
1897 MD_FEATURE_MULTIPLE_PPLS))
1899 set_bit(MD_HAS_PPL, &mddev->flags);
1901 } else if (mddev->pers == NULL) {
1902 /* Insist of good event counter while assembling, except for
1903 * spares (which don't need an event count) */
1905 if (rdev->desc_nr >= 0 &&
1906 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1907 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1908 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1909 if (ev1 < mddev->events)
1911 } else if (mddev->bitmap) {
1912 /* If adding to array with a bitmap, then we can accept an
1913 * older device, but not too old.
1915 if (ev1 < mddev->bitmap->events_cleared)
1917 if (ev1 < mddev->events)
1918 set_bit(Bitmap_sync, &rdev->flags);
1920 if (ev1 < mddev->events)
1921 /* just a hot-add of a new device, leave raid_disk at -1 */
1924 if (mddev->level != LEVEL_MULTIPATH) {
1926 if (rdev->desc_nr < 0 ||
1927 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1928 role = MD_DISK_ROLE_SPARE;
1931 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1933 case MD_DISK_ROLE_SPARE: /* spare */
1935 case MD_DISK_ROLE_FAULTY: /* faulty */
1936 set_bit(Faulty, &rdev->flags);
1938 case MD_DISK_ROLE_JOURNAL: /* journal device */
1939 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1940 /* journal device without journal feature */
1941 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1944 set_bit(Journal, &rdev->flags);
1945 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1946 rdev->raid_disk = 0;
1949 rdev->saved_raid_disk = role;
1950 if ((le32_to_cpu(sb->feature_map) &
1951 MD_FEATURE_RECOVERY_OFFSET)) {
1952 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1953 if (!(le32_to_cpu(sb->feature_map) &
1954 MD_FEATURE_RECOVERY_BITMAP))
1955 rdev->saved_raid_disk = -1;
1958 * If the array is FROZEN, then the device can't
1959 * be in_sync with rest of array.
1961 if (!test_bit(MD_RECOVERY_FROZEN,
1963 set_bit(In_sync, &rdev->flags);
1965 rdev->raid_disk = role;
1968 if (sb->devflags & WriteMostly1)
1969 set_bit(WriteMostly, &rdev->flags);
1970 if (sb->devflags & FailFast1)
1971 set_bit(FailFast, &rdev->flags);
1972 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1973 set_bit(Replacement, &rdev->flags);
1974 } else /* MULTIPATH are always insync */
1975 set_bit(In_sync, &rdev->flags);
1980 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1982 struct mdp_superblock_1 *sb;
1983 struct md_rdev *rdev2;
1985 /* make rdev->sb match mddev and rdev data. */
1987 sb = page_address(rdev->sb_page);
1989 sb->feature_map = 0;
1991 sb->recovery_offset = cpu_to_le64(0);
1992 memset(sb->pad3, 0, sizeof(sb->pad3));
1994 sb->utime = cpu_to_le64((__u64)mddev->utime);
1995 sb->events = cpu_to_le64(mddev->events);
1997 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1998 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1999 sb->resync_offset = cpu_to_le64(MaxSector);
2001 sb->resync_offset = cpu_to_le64(0);
2003 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2005 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2006 sb->size = cpu_to_le64(mddev->dev_sectors);
2007 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2008 sb->level = cpu_to_le32(mddev->level);
2009 sb->layout = cpu_to_le32(mddev->layout);
2010 if (test_bit(FailFast, &rdev->flags))
2011 sb->devflags |= FailFast1;
2013 sb->devflags &= ~FailFast1;
2015 if (test_bit(WriteMostly, &rdev->flags))
2016 sb->devflags |= WriteMostly1;
2018 sb->devflags &= ~WriteMostly1;
2019 sb->data_offset = cpu_to_le64(rdev->data_offset);
2020 sb->data_size = cpu_to_le64(rdev->sectors);
2022 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2023 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2024 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2027 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2028 !test_bit(In_sync, &rdev->flags)) {
2030 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2031 sb->recovery_offset =
2032 cpu_to_le64(rdev->recovery_offset);
2033 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2035 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2037 /* Note: recovery_offset and journal_tail share space */
2038 if (test_bit(Journal, &rdev->flags))
2039 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2040 if (test_bit(Replacement, &rdev->flags))
2042 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2044 if (mddev->reshape_position != MaxSector) {
2045 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2046 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2047 sb->new_layout = cpu_to_le32(mddev->new_layout);
2048 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2049 sb->new_level = cpu_to_le32(mddev->new_level);
2050 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2051 if (mddev->delta_disks == 0 &&
2052 mddev->reshape_backwards)
2054 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2055 if (rdev->new_data_offset != rdev->data_offset) {
2057 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2058 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2059 - rdev->data_offset));
2063 if (mddev_is_clustered(mddev))
2064 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2066 if (rdev->badblocks.count == 0)
2067 /* Nothing to do for bad blocks*/ ;
2068 else if (sb->bblog_offset == 0)
2069 /* Cannot record bad blocks on this device */
2070 md_error(mddev, rdev);
2072 struct badblocks *bb = &rdev->badblocks;
2073 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2075 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2080 seq = read_seqbegin(&bb->lock);
2082 memset(bbp, 0xff, PAGE_SIZE);
2084 for (i = 0 ; i < bb->count ; i++) {
2085 u64 internal_bb = p[i];
2086 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2087 | BB_LEN(internal_bb));
2088 bbp[i] = cpu_to_le64(store_bb);
2091 if (read_seqretry(&bb->lock, seq))
2094 bb->sector = (rdev->sb_start +
2095 (int)le32_to_cpu(sb->bblog_offset));
2096 bb->size = le16_to_cpu(sb->bblog_size);
2101 rdev_for_each(rdev2, mddev)
2102 if (rdev2->desc_nr+1 > max_dev)
2103 max_dev = rdev2->desc_nr+1;
2105 if (max_dev > le32_to_cpu(sb->max_dev)) {
2107 sb->max_dev = cpu_to_le32(max_dev);
2108 rdev->sb_size = max_dev * 2 + 256;
2109 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2110 if (rdev->sb_size & bmask)
2111 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2113 max_dev = le32_to_cpu(sb->max_dev);
2115 for (i=0; i<max_dev;i++)
2116 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2118 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2119 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2121 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2122 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2124 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2126 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2127 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2128 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2131 rdev_for_each(rdev2, mddev) {
2133 if (test_bit(Faulty, &rdev2->flags))
2134 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2135 else if (test_bit(In_sync, &rdev2->flags))
2136 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2137 else if (test_bit(Journal, &rdev2->flags))
2138 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2139 else if (rdev2->raid_disk >= 0)
2140 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2142 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2145 sb->sb_csum = calc_sb_1_csum(sb);
2148 static sector_t super_1_choose_bm_space(sector_t dev_size)
2152 /* if the device is bigger than 8Gig, save 64k for bitmap
2153 * usage, if bigger than 200Gig, save 128k
2155 if (dev_size < 64*2)
2157 else if (dev_size - 64*2 >= 200*1024*1024*2)
2159 else if (dev_size - 4*2 > 8*1024*1024*2)
2166 static unsigned long long
2167 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2169 struct mdp_superblock_1 *sb;
2170 sector_t max_sectors;
2171 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2172 return 0; /* component must fit device */
2173 if (rdev->data_offset != rdev->new_data_offset)
2174 return 0; /* too confusing */
2175 if (rdev->sb_start < rdev->data_offset) {
2176 /* minor versions 1 and 2; superblock before data */
2177 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2178 max_sectors -= rdev->data_offset;
2179 if (!num_sectors || num_sectors > max_sectors)
2180 num_sectors = max_sectors;
2181 } else if (rdev->mddev->bitmap_info.offset) {
2182 /* minor version 0 with bitmap we can't move */
2185 /* minor version 0; superblock after data */
2186 sector_t sb_start, bm_space;
2187 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2189 /* 8K is for superblock */
2190 sb_start = dev_size - 8*2;
2191 sb_start &= ~(sector_t)(4*2 - 1);
2193 bm_space = super_1_choose_bm_space(dev_size);
2195 /* Space that can be used to store date needs to decrease
2196 * superblock bitmap space and bad block space(4K)
2198 max_sectors = sb_start - bm_space - 4*2;
2200 if (!num_sectors || num_sectors > max_sectors)
2201 num_sectors = max_sectors;
2202 rdev->sb_start = sb_start;
2204 sb = page_address(rdev->sb_page);
2205 sb->data_size = cpu_to_le64(num_sectors);
2206 sb->super_offset = cpu_to_le64(rdev->sb_start);
2207 sb->sb_csum = calc_sb_1_csum(sb);
2209 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2211 } while (md_super_wait(rdev->mddev) < 0);
2217 super_1_allow_new_offset(struct md_rdev *rdev,
2218 unsigned long long new_offset)
2220 /* All necessary checks on new >= old have been done */
2221 struct bitmap *bitmap;
2222 if (new_offset >= rdev->data_offset)
2225 /* with 1.0 metadata, there is no metadata to tread on
2226 * so we can always move back */
2227 if (rdev->mddev->minor_version == 0)
2230 /* otherwise we must be sure not to step on
2231 * any metadata, so stay:
2232 * 36K beyond start of superblock
2233 * beyond end of badblocks
2234 * beyond write-intent bitmap
2236 if (rdev->sb_start + (32+4)*2 > new_offset)
2238 bitmap = rdev->mddev->bitmap;
2239 if (bitmap && !rdev->mddev->bitmap_info.file &&
2240 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2241 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2243 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2249 static struct super_type super_types[] = {
2252 .owner = THIS_MODULE,
2253 .load_super = super_90_load,
2254 .validate_super = super_90_validate,
2255 .sync_super = super_90_sync,
2256 .rdev_size_change = super_90_rdev_size_change,
2257 .allow_new_offset = super_90_allow_new_offset,
2261 .owner = THIS_MODULE,
2262 .load_super = super_1_load,
2263 .validate_super = super_1_validate,
2264 .sync_super = super_1_sync,
2265 .rdev_size_change = super_1_rdev_size_change,
2266 .allow_new_offset = super_1_allow_new_offset,
2270 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2272 if (mddev->sync_super) {
2273 mddev->sync_super(mddev, rdev);
2277 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2279 super_types[mddev->major_version].sync_super(mddev, rdev);
2282 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2284 struct md_rdev *rdev, *rdev2;
2287 rdev_for_each_rcu(rdev, mddev1) {
2288 if (test_bit(Faulty, &rdev->flags) ||
2289 test_bit(Journal, &rdev->flags) ||
2290 rdev->raid_disk == -1)
2292 rdev_for_each_rcu(rdev2, mddev2) {
2293 if (test_bit(Faulty, &rdev2->flags) ||
2294 test_bit(Journal, &rdev2->flags) ||
2295 rdev2->raid_disk == -1)
2297 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2307 static LIST_HEAD(pending_raid_disks);
2310 * Try to register data integrity profile for an mddev
2312 * This is called when an array is started and after a disk has been kicked
2313 * from the array. It only succeeds if all working and active component devices
2314 * are integrity capable with matching profiles.
2316 int md_integrity_register(struct mddev *mddev)
2318 struct md_rdev *rdev, *reference = NULL;
2320 if (list_empty(&mddev->disks))
2321 return 0; /* nothing to do */
2322 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2323 return 0; /* shouldn't register, or already is */
2324 rdev_for_each(rdev, mddev) {
2325 /* skip spares and non-functional disks */
2326 if (test_bit(Faulty, &rdev->flags))
2328 if (rdev->raid_disk < 0)
2331 /* Use the first rdev as the reference */
2335 /* does this rdev's profile match the reference profile? */
2336 if (blk_integrity_compare(reference->bdev->bd_disk,
2337 rdev->bdev->bd_disk) < 0)
2340 if (!reference || !bdev_get_integrity(reference->bdev))
2343 * All component devices are integrity capable and have matching
2344 * profiles, register the common profile for the md device.
2346 blk_integrity_register(mddev->gendisk,
2347 bdev_get_integrity(reference->bdev));
2349 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2350 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2351 (mddev->level != 1 && mddev->level != 10 &&
2352 bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2354 * No need to handle the failure of bioset_integrity_create,
2355 * because the function is called by md_run() -> pers->run(),
2356 * md_run calls bioset_exit -> bioset_integrity_free in case
2359 pr_err("md: failed to create integrity pool for %s\n",
2365 EXPORT_SYMBOL(md_integrity_register);
2368 * Attempt to add an rdev, but only if it is consistent with the current
2371 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2373 struct blk_integrity *bi_mddev;
2374 char name[BDEVNAME_SIZE];
2376 if (!mddev->gendisk)
2379 bi_mddev = blk_get_integrity(mddev->gendisk);
2381 if (!bi_mddev) /* nothing to do */
2384 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2385 pr_err("%s: incompatible integrity profile for %s\n",
2386 mdname(mddev), bdevname(rdev->bdev, name));
2392 EXPORT_SYMBOL(md_integrity_add_rdev);
2394 static bool rdev_read_only(struct md_rdev *rdev)
2396 return bdev_read_only(rdev->bdev) ||
2397 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2400 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2402 char b[BDEVNAME_SIZE];
2405 /* prevent duplicates */
2406 if (find_rdev(mddev, rdev->bdev->bd_dev))
2409 if (rdev_read_only(rdev) && mddev->pers)
2412 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2413 if (!test_bit(Journal, &rdev->flags) &&
2415 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2417 /* Cannot change size, so fail
2418 * If mddev->level <= 0, then we don't care
2419 * about aligning sizes (e.g. linear)
2421 if (mddev->level > 0)
2424 mddev->dev_sectors = rdev->sectors;
2427 /* Verify rdev->desc_nr is unique.
2428 * If it is -1, assign a free number, else
2429 * check number is not in use
2432 if (rdev->desc_nr < 0) {
2435 choice = mddev->raid_disks;
2436 while (md_find_rdev_nr_rcu(mddev, choice))
2438 rdev->desc_nr = choice;
2440 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2446 if (!test_bit(Journal, &rdev->flags) &&
2447 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2448 pr_warn("md: %s: array is limited to %d devices\n",
2449 mdname(mddev), mddev->max_disks);
2452 bdevname(rdev->bdev,b);
2453 strreplace(b, '/', '!');
2455 rdev->mddev = mddev;
2456 pr_debug("md: bind<%s>\n", b);
2458 if (mddev->raid_disks)
2459 mddev_create_serial_pool(mddev, rdev, false);
2461 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2464 /* failure here is OK */
2465 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2466 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2467 rdev->sysfs_unack_badblocks =
2468 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2469 rdev->sysfs_badblocks =
2470 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2472 list_add_rcu(&rdev->same_set, &mddev->disks);
2473 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2475 /* May as well allow recovery to be retried once */
2476 mddev->recovery_disabled++;
2481 pr_warn("md: failed to register dev-%s for %s\n",
2486 static void rdev_delayed_delete(struct work_struct *ws)
2488 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2489 kobject_del(&rdev->kobj);
2490 kobject_put(&rdev->kobj);
2493 static void unbind_rdev_from_array(struct md_rdev *rdev)
2495 char b[BDEVNAME_SIZE];
2497 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2498 list_del_rcu(&rdev->same_set);
2499 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2500 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2502 sysfs_remove_link(&rdev->kobj, "block");
2503 sysfs_put(rdev->sysfs_state);
2504 sysfs_put(rdev->sysfs_unack_badblocks);
2505 sysfs_put(rdev->sysfs_badblocks);
2506 rdev->sysfs_state = NULL;
2507 rdev->sysfs_unack_badblocks = NULL;
2508 rdev->sysfs_badblocks = NULL;
2509 rdev->badblocks.count = 0;
2510 /* We need to delay this, otherwise we can deadlock when
2511 * writing to 'remove' to "dev/state". We also need
2512 * to delay it due to rcu usage.
2515 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2516 kobject_get(&rdev->kobj);
2517 queue_work(md_rdev_misc_wq, &rdev->del_work);
2521 * prevent the device from being mounted, repartitioned or
2522 * otherwise reused by a RAID array (or any other kernel
2523 * subsystem), by bd_claiming the device.
2525 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2528 struct block_device *bdev;
2530 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2531 shared ? (struct md_rdev *)lock_rdev : rdev);
2533 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2534 MAJOR(dev), MINOR(dev));
2535 return PTR_ERR(bdev);
2541 static void unlock_rdev(struct md_rdev *rdev)
2543 struct block_device *bdev = rdev->bdev;
2545 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2548 void md_autodetect_dev(dev_t dev);
2550 static void export_rdev(struct md_rdev *rdev)
2552 char b[BDEVNAME_SIZE];
2554 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2555 md_rdev_clear(rdev);
2557 if (test_bit(AutoDetected, &rdev->flags))
2558 md_autodetect_dev(rdev->bdev->bd_dev);
2561 kobject_put(&rdev->kobj);
2564 void md_kick_rdev_from_array(struct md_rdev *rdev)
2566 unbind_rdev_from_array(rdev);
2569 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2571 static void export_array(struct mddev *mddev)
2573 struct md_rdev *rdev;
2575 while (!list_empty(&mddev->disks)) {
2576 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2578 md_kick_rdev_from_array(rdev);
2580 mddev->raid_disks = 0;
2581 mddev->major_version = 0;
2584 static bool set_in_sync(struct mddev *mddev)
2586 lockdep_assert_held(&mddev->lock);
2587 if (!mddev->in_sync) {
2588 mddev->sync_checkers++;
2589 spin_unlock(&mddev->lock);
2590 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2591 spin_lock(&mddev->lock);
2592 if (!mddev->in_sync &&
2593 percpu_ref_is_zero(&mddev->writes_pending)) {
2596 * Ensure ->in_sync is visible before we clear
2600 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2601 sysfs_notify_dirent_safe(mddev->sysfs_state);
2603 if (--mddev->sync_checkers == 0)
2604 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2606 if (mddev->safemode == 1)
2607 mddev->safemode = 0;
2608 return mddev->in_sync;
2611 static void sync_sbs(struct mddev *mddev, int nospares)
2613 /* Update each superblock (in-memory image), but
2614 * if we are allowed to, skip spares which already
2615 * have the right event counter, or have one earlier
2616 * (which would mean they aren't being marked as dirty
2617 * with the rest of the array)
2619 struct md_rdev *rdev;
2620 rdev_for_each(rdev, mddev) {
2621 if (rdev->sb_events == mddev->events ||
2623 rdev->raid_disk < 0 &&
2624 rdev->sb_events+1 == mddev->events)) {
2625 /* Don't update this superblock */
2626 rdev->sb_loaded = 2;
2628 sync_super(mddev, rdev);
2629 rdev->sb_loaded = 1;
2634 static bool does_sb_need_changing(struct mddev *mddev)
2636 struct md_rdev *rdev = NULL, *iter;
2637 struct mdp_superblock_1 *sb;
2640 /* Find a good rdev */
2641 rdev_for_each(iter, mddev)
2642 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2647 /* No good device found. */
2651 sb = page_address(rdev->sb_page);
2652 /* Check if a device has become faulty or a spare become active */
2653 rdev_for_each(rdev, mddev) {
2654 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2655 /* Device activated? */
2656 if (role == 0xffff && rdev->raid_disk >=0 &&
2657 !test_bit(Faulty, &rdev->flags))
2659 /* Device turned faulty? */
2660 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2664 /* Check if any mddev parameters have changed */
2665 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2666 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2667 (mddev->layout != le32_to_cpu(sb->layout)) ||
2668 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2669 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2675 void md_update_sb(struct mddev *mddev, int force_change)
2677 struct md_rdev *rdev;
2680 int any_badblocks_changed = 0;
2685 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2690 if (mddev_is_clustered(mddev)) {
2691 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2693 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2695 ret = md_cluster_ops->metadata_update_start(mddev);
2696 /* Has someone else has updated the sb */
2697 if (!does_sb_need_changing(mddev)) {
2699 md_cluster_ops->metadata_update_cancel(mddev);
2700 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2701 BIT(MD_SB_CHANGE_DEVS) |
2702 BIT(MD_SB_CHANGE_CLEAN));
2708 * First make sure individual recovery_offsets are correct
2709 * curr_resync_completed can only be used during recovery.
2710 * During reshape/resync it might use array-addresses rather
2711 * that device addresses.
2713 rdev_for_each(rdev, mddev) {
2714 if (rdev->raid_disk >= 0 &&
2715 mddev->delta_disks >= 0 &&
2716 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2717 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2718 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2719 !test_bit(Journal, &rdev->flags) &&
2720 !test_bit(In_sync, &rdev->flags) &&
2721 mddev->curr_resync_completed > rdev->recovery_offset)
2722 rdev->recovery_offset = mddev->curr_resync_completed;
2725 if (!mddev->persistent) {
2726 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2727 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2728 if (!mddev->external) {
2729 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2730 rdev_for_each(rdev, mddev) {
2731 if (rdev->badblocks.changed) {
2732 rdev->badblocks.changed = 0;
2733 ack_all_badblocks(&rdev->badblocks);
2734 md_error(mddev, rdev);
2736 clear_bit(Blocked, &rdev->flags);
2737 clear_bit(BlockedBadBlocks, &rdev->flags);
2738 wake_up(&rdev->blocked_wait);
2741 wake_up(&mddev->sb_wait);
2745 spin_lock(&mddev->lock);
2747 mddev->utime = ktime_get_real_seconds();
2749 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2751 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2752 /* just a clean<-> dirty transition, possibly leave spares alone,
2753 * though if events isn't the right even/odd, we will have to do
2759 if (mddev->degraded)
2760 /* If the array is degraded, then skipping spares is both
2761 * dangerous and fairly pointless.
2762 * Dangerous because a device that was removed from the array
2763 * might have a event_count that still looks up-to-date,
2764 * so it can be re-added without a resync.
2765 * Pointless because if there are any spares to skip,
2766 * then a recovery will happen and soon that array won't
2767 * be degraded any more and the spare can go back to sleep then.
2771 sync_req = mddev->in_sync;
2773 /* If this is just a dirty<->clean transition, and the array is clean
2774 * and 'events' is odd, we can roll back to the previous clean state */
2776 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2777 && mddev->can_decrease_events
2778 && mddev->events != 1) {
2780 mddev->can_decrease_events = 0;
2782 /* otherwise we have to go forward and ... */
2784 mddev->can_decrease_events = nospares;
2788 * This 64-bit counter should never wrap.
2789 * Either we are in around ~1 trillion A.C., assuming
2790 * 1 reboot per second, or we have a bug...
2792 WARN_ON(mddev->events == 0);
2794 rdev_for_each(rdev, mddev) {
2795 if (rdev->badblocks.changed)
2796 any_badblocks_changed++;
2797 if (test_bit(Faulty, &rdev->flags))
2798 set_bit(FaultRecorded, &rdev->flags);
2801 sync_sbs(mddev, nospares);
2802 spin_unlock(&mddev->lock);
2804 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2805 mdname(mddev), mddev->in_sync);
2808 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2810 md_bitmap_update_sb(mddev->bitmap);
2811 rdev_for_each(rdev, mddev) {
2812 char b[BDEVNAME_SIZE];
2814 if (rdev->sb_loaded != 1)
2815 continue; /* no noise on spare devices */
2817 if (!test_bit(Faulty, &rdev->flags)) {
2818 md_super_write(mddev,rdev,
2819 rdev->sb_start, rdev->sb_size,
2821 pr_debug("md: (write) %s's sb offset: %llu\n",
2822 bdevname(rdev->bdev, b),
2823 (unsigned long long)rdev->sb_start);
2824 rdev->sb_events = mddev->events;
2825 if (rdev->badblocks.size) {
2826 md_super_write(mddev, rdev,
2827 rdev->badblocks.sector,
2828 rdev->badblocks.size << 9,
2830 rdev->badblocks.size = 0;
2834 pr_debug("md: %s (skipping faulty)\n",
2835 bdevname(rdev->bdev, b));
2837 if (mddev->level == LEVEL_MULTIPATH)
2838 /* only need to write one superblock... */
2841 if (md_super_wait(mddev) < 0)
2843 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2845 if (mddev_is_clustered(mddev) && ret == 0)
2846 md_cluster_ops->metadata_update_finish(mddev);
2848 if (mddev->in_sync != sync_req ||
2849 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2850 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2851 /* have to write it out again */
2853 wake_up(&mddev->sb_wait);
2854 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2855 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2857 rdev_for_each(rdev, mddev) {
2858 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2859 clear_bit(Blocked, &rdev->flags);
2861 if (any_badblocks_changed)
2862 ack_all_badblocks(&rdev->badblocks);
2863 clear_bit(BlockedBadBlocks, &rdev->flags);
2864 wake_up(&rdev->blocked_wait);
2867 EXPORT_SYMBOL(md_update_sb);
2869 static int add_bound_rdev(struct md_rdev *rdev)
2871 struct mddev *mddev = rdev->mddev;
2873 bool add_journal = test_bit(Journal, &rdev->flags);
2875 if (!mddev->pers->hot_remove_disk || add_journal) {
2876 /* If there is hot_add_disk but no hot_remove_disk
2877 * then added disks for geometry changes,
2878 * and should be added immediately.
2880 super_types[mddev->major_version].
2881 validate_super(mddev, rdev);
2883 mddev_suspend(mddev);
2884 err = mddev->pers->hot_add_disk(mddev, rdev);
2886 mddev_resume(mddev);
2888 md_kick_rdev_from_array(rdev);
2892 sysfs_notify_dirent_safe(rdev->sysfs_state);
2894 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2895 if (mddev->degraded)
2896 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2897 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2898 md_new_event(mddev);
2899 md_wakeup_thread(mddev->thread);
2903 /* words written to sysfs files may, or may not, be \n terminated.
2904 * We want to accept with case. For this we use cmd_match.
2906 static int cmd_match(const char *cmd, const char *str)
2908 /* See if cmd, written into a sysfs file, matches
2909 * str. They must either be the same, or cmd can
2910 * have a trailing newline
2912 while (*cmd && *str && *cmd == *str) {
2923 struct rdev_sysfs_entry {
2924 struct attribute attr;
2925 ssize_t (*show)(struct md_rdev *, char *);
2926 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2930 state_show(struct md_rdev *rdev, char *page)
2934 unsigned long flags = READ_ONCE(rdev->flags);
2936 if (test_bit(Faulty, &flags) ||
2937 (!test_bit(ExternalBbl, &flags) &&
2938 rdev->badblocks.unacked_exist))
2939 len += sprintf(page+len, "faulty%s", sep);
2940 if (test_bit(In_sync, &flags))
2941 len += sprintf(page+len, "in_sync%s", sep);
2942 if (test_bit(Journal, &flags))
2943 len += sprintf(page+len, "journal%s", sep);
2944 if (test_bit(WriteMostly, &flags))
2945 len += sprintf(page+len, "write_mostly%s", sep);
2946 if (test_bit(Blocked, &flags) ||
2947 (rdev->badblocks.unacked_exist
2948 && !test_bit(Faulty, &flags)))
2949 len += sprintf(page+len, "blocked%s", sep);
2950 if (!test_bit(Faulty, &flags) &&
2951 !test_bit(Journal, &flags) &&
2952 !test_bit(In_sync, &flags))
2953 len += sprintf(page+len, "spare%s", sep);
2954 if (test_bit(WriteErrorSeen, &flags))
2955 len += sprintf(page+len, "write_error%s", sep);
2956 if (test_bit(WantReplacement, &flags))
2957 len += sprintf(page+len, "want_replacement%s", sep);
2958 if (test_bit(Replacement, &flags))
2959 len += sprintf(page+len, "replacement%s", sep);
2960 if (test_bit(ExternalBbl, &flags))
2961 len += sprintf(page+len, "external_bbl%s", sep);
2962 if (test_bit(FailFast, &flags))
2963 len += sprintf(page+len, "failfast%s", sep);
2968 return len+sprintf(page+len, "\n");
2972 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2975 * faulty - simulates an error
2976 * remove - disconnects the device
2977 * writemostly - sets write_mostly
2978 * -writemostly - clears write_mostly
2979 * blocked - sets the Blocked flags
2980 * -blocked - clears the Blocked and possibly simulates an error
2981 * insync - sets Insync providing device isn't active
2982 * -insync - clear Insync for a device with a slot assigned,
2983 * so that it gets rebuilt based on bitmap
2984 * write_error - sets WriteErrorSeen
2985 * -write_error - clears WriteErrorSeen
2986 * {,-}failfast - set/clear FailFast
2989 struct mddev *mddev = rdev->mddev;
2991 bool need_update_sb = false;
2993 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2994 md_error(rdev->mddev, rdev);
2995 if (test_bit(Faulty, &rdev->flags))
2999 } else if (cmd_match(buf, "remove")) {
3000 if (rdev->mddev->pers) {
3001 clear_bit(Blocked, &rdev->flags);
3002 remove_and_add_spares(rdev->mddev, rdev);
3004 if (rdev->raid_disk >= 0)
3008 if (mddev_is_clustered(mddev))
3009 err = md_cluster_ops->remove_disk(mddev, rdev);
3012 md_kick_rdev_from_array(rdev);
3014 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3015 md_wakeup_thread(mddev->thread);
3017 md_new_event(mddev);
3020 } else if (cmd_match(buf, "writemostly")) {
3021 set_bit(WriteMostly, &rdev->flags);
3022 mddev_create_serial_pool(rdev->mddev, rdev, false);
3023 need_update_sb = true;
3025 } else if (cmd_match(buf, "-writemostly")) {
3026 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3027 clear_bit(WriteMostly, &rdev->flags);
3028 need_update_sb = true;
3030 } else if (cmd_match(buf, "blocked")) {
3031 set_bit(Blocked, &rdev->flags);
3033 } else if (cmd_match(buf, "-blocked")) {
3034 if (!test_bit(Faulty, &rdev->flags) &&
3035 !test_bit(ExternalBbl, &rdev->flags) &&
3036 rdev->badblocks.unacked_exist) {
3037 /* metadata handler doesn't understand badblocks,
3038 * so we need to fail the device
3040 md_error(rdev->mddev, rdev);
3042 clear_bit(Blocked, &rdev->flags);
3043 clear_bit(BlockedBadBlocks, &rdev->flags);
3044 wake_up(&rdev->blocked_wait);
3045 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3046 md_wakeup_thread(rdev->mddev->thread);
3049 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3050 set_bit(In_sync, &rdev->flags);
3052 } else if (cmd_match(buf, "failfast")) {
3053 set_bit(FailFast, &rdev->flags);
3054 need_update_sb = true;
3056 } else if (cmd_match(buf, "-failfast")) {
3057 clear_bit(FailFast, &rdev->flags);
3058 need_update_sb = true;
3060 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3061 !test_bit(Journal, &rdev->flags)) {
3062 if (rdev->mddev->pers == NULL) {
3063 clear_bit(In_sync, &rdev->flags);
3064 rdev->saved_raid_disk = rdev->raid_disk;
3065 rdev->raid_disk = -1;
3068 } else if (cmd_match(buf, "write_error")) {
3069 set_bit(WriteErrorSeen, &rdev->flags);
3071 } else if (cmd_match(buf, "-write_error")) {
3072 clear_bit(WriteErrorSeen, &rdev->flags);
3074 } else if (cmd_match(buf, "want_replacement")) {
3075 /* Any non-spare device that is not a replacement can
3076 * become want_replacement at any time, but we then need to
3077 * check if recovery is needed.
3079 if (rdev->raid_disk >= 0 &&
3080 !test_bit(Journal, &rdev->flags) &&
3081 !test_bit(Replacement, &rdev->flags))
3082 set_bit(WantReplacement, &rdev->flags);
3083 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3084 md_wakeup_thread(rdev->mddev->thread);
3086 } else if (cmd_match(buf, "-want_replacement")) {
3087 /* Clearing 'want_replacement' is always allowed.
3088 * Once replacements starts it is too late though.
3091 clear_bit(WantReplacement, &rdev->flags);
3092 } else if (cmd_match(buf, "replacement")) {
3093 /* Can only set a device as a replacement when array has not
3094 * yet been started. Once running, replacement is automatic
3095 * from spares, or by assigning 'slot'.
3097 if (rdev->mddev->pers)
3100 set_bit(Replacement, &rdev->flags);
3103 } else if (cmd_match(buf, "-replacement")) {
3104 /* Similarly, can only clear Replacement before start */
3105 if (rdev->mddev->pers)
3108 clear_bit(Replacement, &rdev->flags);
3111 } else if (cmd_match(buf, "re-add")) {
3112 if (!rdev->mddev->pers)
3114 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3115 rdev->saved_raid_disk >= 0) {
3116 /* clear_bit is performed _after_ all the devices
3117 * have their local Faulty bit cleared. If any writes
3118 * happen in the meantime in the local node, they
3119 * will land in the local bitmap, which will be synced
3120 * by this node eventually
3122 if (!mddev_is_clustered(rdev->mddev) ||
3123 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3124 clear_bit(Faulty, &rdev->flags);
3125 err = add_bound_rdev(rdev);
3129 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3130 set_bit(ExternalBbl, &rdev->flags);
3131 rdev->badblocks.shift = 0;
3133 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3134 clear_bit(ExternalBbl, &rdev->flags);
3138 md_update_sb(mddev, 1);
3140 sysfs_notify_dirent_safe(rdev->sysfs_state);
3141 return err ? err : len;
3143 static struct rdev_sysfs_entry rdev_state =
3144 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3147 errors_show(struct md_rdev *rdev, char *page)
3149 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3153 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3158 rv = kstrtouint(buf, 10, &n);
3161 atomic_set(&rdev->corrected_errors, n);
3164 static struct rdev_sysfs_entry rdev_errors =
3165 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3168 slot_show(struct md_rdev *rdev, char *page)
3170 if (test_bit(Journal, &rdev->flags))
3171 return sprintf(page, "journal\n");
3172 else if (rdev->raid_disk < 0)
3173 return sprintf(page, "none\n");
3175 return sprintf(page, "%d\n", rdev->raid_disk);
3179 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3184 if (test_bit(Journal, &rdev->flags))
3186 if (strncmp(buf, "none", 4)==0)
3189 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3193 if (rdev->mddev->pers && slot == -1) {
3194 /* Setting 'slot' on an active array requires also
3195 * updating the 'rd%d' link, and communicating
3196 * with the personality with ->hot_*_disk.
3197 * For now we only support removing
3198 * failed/spare devices. This normally happens automatically,
3199 * but not when the metadata is externally managed.
3201 if (rdev->raid_disk == -1)
3203 /* personality does all needed checks */
3204 if (rdev->mddev->pers->hot_remove_disk == NULL)
3206 clear_bit(Blocked, &rdev->flags);
3207 remove_and_add_spares(rdev->mddev, rdev);
3208 if (rdev->raid_disk >= 0)
3210 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3211 md_wakeup_thread(rdev->mddev->thread);
3212 } else if (rdev->mddev->pers) {
3213 /* Activating a spare .. or possibly reactivating
3214 * if we ever get bitmaps working here.
3218 if (rdev->raid_disk != -1)
3221 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3224 if (rdev->mddev->pers->hot_add_disk == NULL)
3227 if (slot >= rdev->mddev->raid_disks &&
3228 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3231 rdev->raid_disk = slot;
3232 if (test_bit(In_sync, &rdev->flags))
3233 rdev->saved_raid_disk = slot;
3235 rdev->saved_raid_disk = -1;
3236 clear_bit(In_sync, &rdev->flags);
3237 clear_bit(Bitmap_sync, &rdev->flags);
3238 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3240 rdev->raid_disk = -1;
3243 sysfs_notify_dirent_safe(rdev->sysfs_state);
3244 /* failure here is OK */;
3245 sysfs_link_rdev(rdev->mddev, rdev);
3246 /* don't wakeup anyone, leave that to userspace. */
3248 if (slot >= rdev->mddev->raid_disks &&
3249 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3251 rdev->raid_disk = slot;
3252 /* assume it is working */
3253 clear_bit(Faulty, &rdev->flags);
3254 clear_bit(WriteMostly, &rdev->flags);
3255 set_bit(In_sync, &rdev->flags);
3256 sysfs_notify_dirent_safe(rdev->sysfs_state);
3261 static struct rdev_sysfs_entry rdev_slot =
3262 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3265 offset_show(struct md_rdev *rdev, char *page)
3267 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3271 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3273 unsigned long long offset;
3274 if (kstrtoull(buf, 10, &offset) < 0)
3276 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3278 if (rdev->sectors && rdev->mddev->external)
3279 /* Must set offset before size, so overlap checks
3282 rdev->data_offset = offset;
3283 rdev->new_data_offset = offset;
3287 static struct rdev_sysfs_entry rdev_offset =
3288 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3290 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3292 return sprintf(page, "%llu\n",
3293 (unsigned long long)rdev->new_data_offset);
3296 static ssize_t new_offset_store(struct md_rdev *rdev,
3297 const char *buf, size_t len)
3299 unsigned long long new_offset;
3300 struct mddev *mddev = rdev->mddev;
3302 if (kstrtoull(buf, 10, &new_offset) < 0)
3305 if (mddev->sync_thread ||
3306 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3308 if (new_offset == rdev->data_offset)
3309 /* reset is always permitted */
3311 else if (new_offset > rdev->data_offset) {
3312 /* must not push array size beyond rdev_sectors */
3313 if (new_offset - rdev->data_offset
3314 + mddev->dev_sectors > rdev->sectors)
3317 /* Metadata worries about other space details. */
3319 /* decreasing the offset is inconsistent with a backwards
3322 if (new_offset < rdev->data_offset &&
3323 mddev->reshape_backwards)
3325 /* Increasing offset is inconsistent with forwards
3326 * reshape. reshape_direction should be set to
3327 * 'backwards' first.
3329 if (new_offset > rdev->data_offset &&
3330 !mddev->reshape_backwards)
3333 if (mddev->pers && mddev->persistent &&
3334 !super_types[mddev->major_version]
3335 .allow_new_offset(rdev, new_offset))
3337 rdev->new_data_offset = new_offset;
3338 if (new_offset > rdev->data_offset)
3339 mddev->reshape_backwards = 1;
3340 else if (new_offset < rdev->data_offset)
3341 mddev->reshape_backwards = 0;
3345 static struct rdev_sysfs_entry rdev_new_offset =
3346 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3349 rdev_size_show(struct md_rdev *rdev, char *page)
3351 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3354 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3356 /* check if two start/length pairs overlap */
3364 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3366 unsigned long long blocks;
3369 if (kstrtoull(buf, 10, &blocks) < 0)
3372 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3373 return -EINVAL; /* sector conversion overflow */
3376 if (new != blocks * 2)
3377 return -EINVAL; /* unsigned long long to sector_t overflow */
3384 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3386 struct mddev *my_mddev = rdev->mddev;
3387 sector_t oldsectors = rdev->sectors;
3390 if (test_bit(Journal, &rdev->flags))
3392 if (strict_blocks_to_sectors(buf, §ors) < 0)
3394 if (rdev->data_offset != rdev->new_data_offset)
3395 return -EINVAL; /* too confusing */
3396 if (my_mddev->pers && rdev->raid_disk >= 0) {
3397 if (my_mddev->persistent) {
3398 sectors = super_types[my_mddev->major_version].
3399 rdev_size_change(rdev, sectors);
3402 } else if (!sectors)
3403 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3405 if (!my_mddev->pers->resize)
3406 /* Cannot change size for RAID0 or Linear etc */
3409 if (sectors < my_mddev->dev_sectors)
3410 return -EINVAL; /* component must fit device */
3412 rdev->sectors = sectors;
3413 if (sectors > oldsectors && my_mddev->external) {
3414 /* Need to check that all other rdevs with the same
3415 * ->bdev do not overlap. 'rcu' is sufficient to walk
3416 * the rdev lists safely.
3417 * This check does not provide a hard guarantee, it
3418 * just helps avoid dangerous mistakes.
3420 struct mddev *mddev;
3422 struct list_head *tmp;
3425 for_each_mddev(mddev, tmp) {
3426 struct md_rdev *rdev2;
3428 rdev_for_each(rdev2, mddev)
3429 if (rdev->bdev == rdev2->bdev &&
3431 overlaps(rdev->data_offset, rdev->sectors,
3444 /* Someone else could have slipped in a size
3445 * change here, but doing so is just silly.
3446 * We put oldsectors back because we *know* it is
3447 * safe, and trust userspace not to race with
3450 rdev->sectors = oldsectors;
3457 static struct rdev_sysfs_entry rdev_size =
3458 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3460 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3462 unsigned long long recovery_start = rdev->recovery_offset;
3464 if (test_bit(In_sync, &rdev->flags) ||
3465 recovery_start == MaxSector)
3466 return sprintf(page, "none\n");
3468 return sprintf(page, "%llu\n", recovery_start);
3471 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3473 unsigned long long recovery_start;
3475 if (cmd_match(buf, "none"))
3476 recovery_start = MaxSector;
3477 else if (kstrtoull(buf, 10, &recovery_start))
3480 if (rdev->mddev->pers &&
3481 rdev->raid_disk >= 0)
3484 rdev->recovery_offset = recovery_start;
3485 if (recovery_start == MaxSector)
3486 set_bit(In_sync, &rdev->flags);
3488 clear_bit(In_sync, &rdev->flags);
3492 static struct rdev_sysfs_entry rdev_recovery_start =
3493 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3495 /* sysfs access to bad-blocks list.
3496 * We present two files.
3497 * 'bad-blocks' lists sector numbers and lengths of ranges that
3498 * are recorded as bad. The list is truncated to fit within
3499 * the one-page limit of sysfs.
3500 * Writing "sector length" to this file adds an acknowledged
3502 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3503 * been acknowledged. Writing to this file adds bad blocks
3504 * without acknowledging them. This is largely for testing.
3506 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3508 return badblocks_show(&rdev->badblocks, page, 0);
3510 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3512 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3513 /* Maybe that ack was all we needed */
3514 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3515 wake_up(&rdev->blocked_wait);
3518 static struct rdev_sysfs_entry rdev_bad_blocks =
3519 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3521 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3523 return badblocks_show(&rdev->badblocks, page, 1);
3525 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3527 return badblocks_store(&rdev->badblocks, page, len, 1);
3529 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3530 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3533 ppl_sector_show(struct md_rdev *rdev, char *page)
3535 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3539 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3541 unsigned long long sector;
3543 if (kstrtoull(buf, 10, §or) < 0)
3545 if (sector != (sector_t)sector)
3548 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3549 rdev->raid_disk >= 0)
3552 if (rdev->mddev->persistent) {
3553 if (rdev->mddev->major_version == 0)
3555 if ((sector > rdev->sb_start &&
3556 sector - rdev->sb_start > S16_MAX) ||
3557 (sector < rdev->sb_start &&
3558 rdev->sb_start - sector > -S16_MIN))
3560 rdev->ppl.offset = sector - rdev->sb_start;
3561 } else if (!rdev->mddev->external) {
3564 rdev->ppl.sector = sector;
3568 static struct rdev_sysfs_entry rdev_ppl_sector =
3569 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3572 ppl_size_show(struct md_rdev *rdev, char *page)
3574 return sprintf(page, "%u\n", rdev->ppl.size);
3578 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3582 if (kstrtouint(buf, 10, &size) < 0)
3585 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3586 rdev->raid_disk >= 0)
3589 if (rdev->mddev->persistent) {
3590 if (rdev->mddev->major_version == 0)
3594 } else if (!rdev->mddev->external) {
3597 rdev->ppl.size = size;
3601 static struct rdev_sysfs_entry rdev_ppl_size =
3602 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3604 static struct attribute *rdev_default_attrs[] = {
3609 &rdev_new_offset.attr,
3611 &rdev_recovery_start.attr,
3612 &rdev_bad_blocks.attr,
3613 &rdev_unack_bad_blocks.attr,
3614 &rdev_ppl_sector.attr,
3615 &rdev_ppl_size.attr,
3619 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3621 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3622 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3628 return entry->show(rdev, page);
3632 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3633 const char *page, size_t length)
3635 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3636 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3638 struct mddev *mddev = rdev->mddev;
3642 if (!capable(CAP_SYS_ADMIN))
3644 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3646 if (rdev->mddev == NULL)
3649 rv = entry->store(rdev, page, length);
3650 mddev_unlock(mddev);
3655 static void rdev_free(struct kobject *ko)
3657 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3660 static const struct sysfs_ops rdev_sysfs_ops = {
3661 .show = rdev_attr_show,
3662 .store = rdev_attr_store,
3664 static struct kobj_type rdev_ktype = {
3665 .release = rdev_free,
3666 .sysfs_ops = &rdev_sysfs_ops,
3667 .default_attrs = rdev_default_attrs,
3670 int md_rdev_init(struct md_rdev *rdev)
3673 rdev->saved_raid_disk = -1;
3674 rdev->raid_disk = -1;
3676 rdev->data_offset = 0;
3677 rdev->new_data_offset = 0;
3678 rdev->sb_events = 0;
3679 rdev->last_read_error = 0;
3680 rdev->sb_loaded = 0;
3681 rdev->bb_page = NULL;
3682 atomic_set(&rdev->nr_pending, 0);
3683 atomic_set(&rdev->read_errors, 0);
3684 atomic_set(&rdev->corrected_errors, 0);
3686 INIT_LIST_HEAD(&rdev->same_set);
3687 init_waitqueue_head(&rdev->blocked_wait);
3689 /* Add space to store bad block list.
3690 * This reserves the space even on arrays where it cannot
3691 * be used - I wonder if that matters
3693 return badblocks_init(&rdev->badblocks, 0);
3695 EXPORT_SYMBOL_GPL(md_rdev_init);
3697 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3699 * mark the device faulty if:
3701 * - the device is nonexistent (zero size)
3702 * - the device has no valid superblock
3704 * a faulty rdev _never_ has rdev->sb set.
3706 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3708 char b[BDEVNAME_SIZE];
3710 struct md_rdev *rdev;
3713 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3715 return ERR_PTR(-ENOMEM);
3717 err = md_rdev_init(rdev);
3720 err = alloc_disk_sb(rdev);
3724 err = lock_rdev(rdev, newdev, super_format == -2);
3728 kobject_init(&rdev->kobj, &rdev_ktype);
3730 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3732 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3733 bdevname(rdev->bdev,b));
3738 if (super_format >= 0) {
3739 err = super_types[super_format].
3740 load_super(rdev, NULL, super_minor);
3741 if (err == -EINVAL) {
3742 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3743 bdevname(rdev->bdev,b),
3744 super_format, super_minor);
3748 pr_warn("md: could not read %s's sb, not importing!\n",
3749 bdevname(rdev->bdev,b));
3759 md_rdev_clear(rdev);
3761 return ERR_PTR(err);
3765 * Check a full RAID array for plausibility
3768 static int analyze_sbs(struct mddev *mddev)
3771 struct md_rdev *rdev, *freshest, *tmp;
3772 char b[BDEVNAME_SIZE];
3775 rdev_for_each_safe(rdev, tmp, mddev)
3776 switch (super_types[mddev->major_version].
3777 load_super(rdev, freshest, mddev->minor_version)) {
3784 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3785 bdevname(rdev->bdev,b));
3786 md_kick_rdev_from_array(rdev);
3789 /* Cannot find a valid fresh disk */
3791 pr_warn("md: cannot find a valid disk\n");
3795 super_types[mddev->major_version].
3796 validate_super(mddev, freshest);
3799 rdev_for_each_safe(rdev, tmp, mddev) {
3800 if (mddev->max_disks &&
3801 (rdev->desc_nr >= mddev->max_disks ||
3802 i > mddev->max_disks)) {
3803 pr_warn("md: %s: %s: only %d devices permitted\n",
3804 mdname(mddev), bdevname(rdev->bdev, b),
3806 md_kick_rdev_from_array(rdev);
3809 if (rdev != freshest) {
3810 if (super_types[mddev->major_version].
3811 validate_super(mddev, rdev)) {
3812 pr_warn("md: kicking non-fresh %s from array!\n",
3813 bdevname(rdev->bdev,b));
3814 md_kick_rdev_from_array(rdev);
3818 if (mddev->level == LEVEL_MULTIPATH) {
3819 rdev->desc_nr = i++;
3820 rdev->raid_disk = rdev->desc_nr;
3821 set_bit(In_sync, &rdev->flags);
3822 } else if (rdev->raid_disk >=
3823 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3824 !test_bit(Journal, &rdev->flags)) {
3825 rdev->raid_disk = -1;
3826 clear_bit(In_sync, &rdev->flags);
3833 /* Read a fixed-point number.
3834 * Numbers in sysfs attributes should be in "standard" units where
3835 * possible, so time should be in seconds.
3836 * However we internally use a a much smaller unit such as
3837 * milliseconds or jiffies.
3838 * This function takes a decimal number with a possible fractional
3839 * component, and produces an integer which is the result of
3840 * multiplying that number by 10^'scale'.
3841 * all without any floating-point arithmetic.
3843 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3845 unsigned long result = 0;
3847 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3850 else if (decimals < scale) {
3853 result = result * 10 + value;
3865 *res = result * int_pow(10, scale - decimals);
3870 safe_delay_show(struct mddev *mddev, char *page)
3872 int msec = (mddev->safemode_delay*1000)/HZ;
3873 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3876 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3880 if (mddev_is_clustered(mddev)) {
3881 pr_warn("md: Safemode is disabled for clustered mode\n");
3885 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3888 mddev->safemode_delay = 0;
3890 unsigned long old_delay = mddev->safemode_delay;
3891 unsigned long new_delay = (msec*HZ)/1000;
3895 mddev->safemode_delay = new_delay;
3896 if (new_delay < old_delay || old_delay == 0)
3897 mod_timer(&mddev->safemode_timer, jiffies+1);
3901 static struct md_sysfs_entry md_safe_delay =
3902 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3905 level_show(struct mddev *mddev, char *page)
3907 struct md_personality *p;
3909 spin_lock(&mddev->lock);
3912 ret = sprintf(page, "%s\n", p->name);
3913 else if (mddev->clevel[0])
3914 ret = sprintf(page, "%s\n", mddev->clevel);
3915 else if (mddev->level != LEVEL_NONE)
3916 ret = sprintf(page, "%d\n", mddev->level);
3919 spin_unlock(&mddev->lock);
3924 level_store(struct mddev *mddev, const char *buf, size_t len)
3929 struct md_personality *pers, *oldpers;
3931 void *priv, *oldpriv;
3932 struct md_rdev *rdev;
3934 if (slen == 0 || slen >= sizeof(clevel))
3937 rv = mddev_lock(mddev);
3941 if (mddev->pers == NULL) {
3942 strncpy(mddev->clevel, buf, slen);
3943 if (mddev->clevel[slen-1] == '\n')
3945 mddev->clevel[slen] = 0;
3946 mddev->level = LEVEL_NONE;
3954 /* request to change the personality. Need to ensure:
3955 * - array is not engaged in resync/recovery/reshape
3956 * - old personality can be suspended
3957 * - new personality will access other array.
3961 if (mddev->sync_thread ||
3962 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3963 mddev->reshape_position != MaxSector ||
3964 mddev->sysfs_active)
3968 if (!mddev->pers->quiesce) {
3969 pr_warn("md: %s: %s does not support online personality change\n",
3970 mdname(mddev), mddev->pers->name);
3974 /* Now find the new personality */
3975 strncpy(clevel, buf, slen);
3976 if (clevel[slen-1] == '\n')
3979 if (kstrtol(clevel, 10, &level))
3982 if (request_module("md-%s", clevel) != 0)
3983 request_module("md-level-%s", clevel);
3984 spin_lock(&pers_lock);
3985 pers = find_pers(level, clevel);
3986 if (!pers || !try_module_get(pers->owner)) {
3987 spin_unlock(&pers_lock);
3988 pr_warn("md: personality %s not loaded\n", clevel);
3992 spin_unlock(&pers_lock);
3994 if (pers == mddev->pers) {
3995 /* Nothing to do! */
3996 module_put(pers->owner);
4000 if (!pers->takeover) {
4001 module_put(pers->owner);
4002 pr_warn("md: %s: %s does not support personality takeover\n",
4003 mdname(mddev), clevel);
4008 rdev_for_each(rdev, mddev)
4009 rdev->new_raid_disk = rdev->raid_disk;
4011 /* ->takeover must set new_* and/or delta_disks
4012 * if it succeeds, and may set them when it fails.
4014 priv = pers->takeover(mddev);
4016 mddev->new_level = mddev->level;
4017 mddev->new_layout = mddev->layout;
4018 mddev->new_chunk_sectors = mddev->chunk_sectors;
4019 mddev->raid_disks -= mddev->delta_disks;
4020 mddev->delta_disks = 0;
4021 mddev->reshape_backwards = 0;
4022 module_put(pers->owner);
4023 pr_warn("md: %s: %s would not accept array\n",
4024 mdname(mddev), clevel);
4029 /* Looks like we have a winner */
4030 mddev_suspend(mddev);
4031 mddev_detach(mddev);
4033 spin_lock(&mddev->lock);
4034 oldpers = mddev->pers;
4035 oldpriv = mddev->private;
4037 mddev->private = priv;
4038 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4039 mddev->level = mddev->new_level;
4040 mddev->layout = mddev->new_layout;
4041 mddev->chunk_sectors = mddev->new_chunk_sectors;
4042 mddev->delta_disks = 0;
4043 mddev->reshape_backwards = 0;
4044 mddev->degraded = 0;
4045 spin_unlock(&mddev->lock);
4047 if (oldpers->sync_request == NULL &&
4049 /* We are converting from a no-redundancy array
4050 * to a redundancy array and metadata is managed
4051 * externally so we need to be sure that writes
4052 * won't block due to a need to transition
4054 * until external management is started.
4057 mddev->safemode_delay = 0;
4058 mddev->safemode = 0;
4061 oldpers->free(mddev, oldpriv);
4063 if (oldpers->sync_request == NULL &&
4064 pers->sync_request != NULL) {
4065 /* need to add the md_redundancy_group */
4066 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4067 pr_warn("md: cannot register extra attributes for %s\n",
4069 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4070 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4071 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4073 if (oldpers->sync_request != NULL &&
4074 pers->sync_request == NULL) {
4075 /* need to remove the md_redundancy_group */
4076 if (mddev->to_remove == NULL)
4077 mddev->to_remove = &md_redundancy_group;
4080 module_put(oldpers->owner);
4082 rdev_for_each(rdev, mddev) {
4083 if (rdev->raid_disk < 0)
4085 if (rdev->new_raid_disk >= mddev->raid_disks)
4086 rdev->new_raid_disk = -1;
4087 if (rdev->new_raid_disk == rdev->raid_disk)
4089 sysfs_unlink_rdev(mddev, rdev);
4091 rdev_for_each(rdev, mddev) {
4092 if (rdev->raid_disk < 0)
4094 if (rdev->new_raid_disk == rdev->raid_disk)
4096 rdev->raid_disk = rdev->new_raid_disk;
4097 if (rdev->raid_disk < 0)
4098 clear_bit(In_sync, &rdev->flags);
4100 if (sysfs_link_rdev(mddev, rdev))
4101 pr_warn("md: cannot register rd%d for %s after level change\n",
4102 rdev->raid_disk, mdname(mddev));
4106 if (pers->sync_request == NULL) {
4107 /* this is now an array without redundancy, so
4108 * it must always be in_sync
4111 del_timer_sync(&mddev->safemode_timer);
4113 blk_set_stacking_limits(&mddev->queue->limits);
4115 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4116 mddev_resume(mddev);
4118 md_update_sb(mddev, 1);
4119 sysfs_notify_dirent_safe(mddev->sysfs_level);
4120 md_new_event(mddev);
4123 mddev_unlock(mddev);
4127 static struct md_sysfs_entry md_level =
4128 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4131 layout_show(struct mddev *mddev, char *page)
4133 /* just a number, not meaningful for all levels */
4134 if (mddev->reshape_position != MaxSector &&
4135 mddev->layout != mddev->new_layout)
4136 return sprintf(page, "%d (%d)\n",
4137 mddev->new_layout, mddev->layout);
4138 return sprintf(page, "%d\n", mddev->layout);
4142 layout_store(struct mddev *mddev, const char *buf, size_t len)
4147 err = kstrtouint(buf, 10, &n);
4150 err = mddev_lock(mddev);
4155 if (mddev->pers->check_reshape == NULL)
4160 mddev->new_layout = n;
4161 err = mddev->pers->check_reshape(mddev);
4163 mddev->new_layout = mddev->layout;
4166 mddev->new_layout = n;
4167 if (mddev->reshape_position == MaxSector)
4170 mddev_unlock(mddev);
4173 static struct md_sysfs_entry md_layout =
4174 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4177 raid_disks_show(struct mddev *mddev, char *page)
4179 if (mddev->raid_disks == 0)
4181 if (mddev->reshape_position != MaxSector &&
4182 mddev->delta_disks != 0)
4183 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4184 mddev->raid_disks - mddev->delta_disks);
4185 return sprintf(page, "%d\n", mddev->raid_disks);
4188 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4191 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4196 err = kstrtouint(buf, 10, &n);
4200 err = mddev_lock(mddev);
4204 err = update_raid_disks(mddev, n);
4205 else if (mddev->reshape_position != MaxSector) {
4206 struct md_rdev *rdev;
4207 int olddisks = mddev->raid_disks - mddev->delta_disks;
4210 rdev_for_each(rdev, mddev) {
4212 rdev->data_offset < rdev->new_data_offset)
4215 rdev->data_offset > rdev->new_data_offset)
4219 mddev->delta_disks = n - olddisks;
4220 mddev->raid_disks = n;
4221 mddev->reshape_backwards = (mddev->delta_disks < 0);
4223 mddev->raid_disks = n;
4225 mddev_unlock(mddev);
4226 return err ? err : len;
4228 static struct md_sysfs_entry md_raid_disks =
4229 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4232 uuid_show(struct mddev *mddev, char *page)
4234 return sprintf(page, "%pU\n", mddev->uuid);
4236 static struct md_sysfs_entry md_uuid =
4237 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4240 chunk_size_show(struct mddev *mddev, char *page)
4242 if (mddev->reshape_position != MaxSector &&
4243 mddev->chunk_sectors != mddev->new_chunk_sectors)
4244 return sprintf(page, "%d (%d)\n",
4245 mddev->new_chunk_sectors << 9,
4246 mddev->chunk_sectors << 9);
4247 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4251 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4256 err = kstrtoul(buf, 10, &n);
4260 err = mddev_lock(mddev);
4264 if (mddev->pers->check_reshape == NULL)
4269 mddev->new_chunk_sectors = n >> 9;
4270 err = mddev->pers->check_reshape(mddev);
4272 mddev->new_chunk_sectors = mddev->chunk_sectors;
4275 mddev->new_chunk_sectors = n >> 9;
4276 if (mddev->reshape_position == MaxSector)
4277 mddev->chunk_sectors = n >> 9;
4279 mddev_unlock(mddev);
4282 static struct md_sysfs_entry md_chunk_size =
4283 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4286 resync_start_show(struct mddev *mddev, char *page)
4288 if (mddev->recovery_cp == MaxSector)
4289 return sprintf(page, "none\n");
4290 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4294 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4296 unsigned long long n;
4299 if (cmd_match(buf, "none"))
4302 err = kstrtoull(buf, 10, &n);
4305 if (n != (sector_t)n)
4309 err = mddev_lock(mddev);
4312 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4316 mddev->recovery_cp = n;
4318 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4320 mddev_unlock(mddev);
4323 static struct md_sysfs_entry md_resync_start =
4324 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4325 resync_start_show, resync_start_store);
4328 * The array state can be:
4331 * No devices, no size, no level
4332 * Equivalent to STOP_ARRAY ioctl
4334 * May have some settings, but array is not active
4335 * all IO results in error
4336 * When written, doesn't tear down array, but just stops it
4337 * suspended (not supported yet)
4338 * All IO requests will block. The array can be reconfigured.
4339 * Writing this, if accepted, will block until array is quiescent
4341 * no resync can happen. no superblocks get written.
4342 * write requests fail
4344 * like readonly, but behaves like 'clean' on a write request.
4346 * clean - no pending writes, but otherwise active.
4347 * When written to inactive array, starts without resync
4348 * If a write request arrives then
4349 * if metadata is known, mark 'dirty' and switch to 'active'.
4350 * if not known, block and switch to write-pending
4351 * If written to an active array that has pending writes, then fails.
4353 * fully active: IO and resync can be happening.
4354 * When written to inactive array, starts with resync
4357 * clean, but writes are blocked waiting for 'active' to be written.
4360 * like active, but no writes have been seen for a while (100msec).
4363 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4364 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4365 * when a member is gone, so this state will at least alert the
4366 * user that something is wrong.
4368 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4369 write_pending, active_idle, broken, bad_word};
4370 static char *array_states[] = {
4371 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4372 "write-pending", "active-idle", "broken", NULL };
4374 static int match_word(const char *word, char **list)
4377 for (n=0; list[n]; n++)
4378 if (cmd_match(word, list[n]))
4384 array_state_show(struct mddev *mddev, char *page)
4386 enum array_state st = inactive;
4388 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4397 spin_lock(&mddev->lock);
4398 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4400 else if (mddev->in_sync)
4402 else if (mddev->safemode)
4406 spin_unlock(&mddev->lock);
4409 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4412 if (list_empty(&mddev->disks) &&
4413 mddev->raid_disks == 0 &&
4414 mddev->dev_sectors == 0)
4419 return sprintf(page, "%s\n", array_states[st]);
4422 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4423 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4424 static int restart_array(struct mddev *mddev);
4427 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4430 enum array_state st = match_word(buf, array_states);
4432 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4433 /* don't take reconfig_mutex when toggling between
4436 spin_lock(&mddev->lock);
4438 restart_array(mddev);
4439 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4440 md_wakeup_thread(mddev->thread);
4441 wake_up(&mddev->sb_wait);
4442 } else /* st == clean */ {
4443 restart_array(mddev);
4444 if (!set_in_sync(mddev))
4448 sysfs_notify_dirent_safe(mddev->sysfs_state);
4449 spin_unlock(&mddev->lock);
4452 err = mddev_lock(mddev);
4460 /* stopping an active array */
4461 err = do_md_stop(mddev, 0, NULL);
4464 /* stopping an active array */
4466 err = do_md_stop(mddev, 2, NULL);
4468 err = 0; /* already inactive */
4471 break; /* not supported yet */
4474 err = md_set_readonly(mddev, NULL);
4477 set_disk_ro(mddev->gendisk, 1);
4478 err = do_md_run(mddev);
4484 err = md_set_readonly(mddev, NULL);
4485 else if (mddev->ro == 1)
4486 err = restart_array(mddev);
4489 set_disk_ro(mddev->gendisk, 0);
4493 err = do_md_run(mddev);
4498 err = restart_array(mddev);
4501 spin_lock(&mddev->lock);
4502 if (!set_in_sync(mddev))
4504 spin_unlock(&mddev->lock);
4510 err = restart_array(mddev);
4513 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4514 wake_up(&mddev->sb_wait);
4518 set_disk_ro(mddev->gendisk, 0);
4519 err = do_md_run(mddev);
4525 /* these cannot be set */
4530 if (mddev->hold_active == UNTIL_IOCTL)
4531 mddev->hold_active = 0;
4532 sysfs_notify_dirent_safe(mddev->sysfs_state);
4534 mddev_unlock(mddev);
4537 static struct md_sysfs_entry md_array_state =
4538 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4541 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4542 return sprintf(page, "%d\n",
4543 atomic_read(&mddev->max_corr_read_errors));
4547 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4552 rv = kstrtouint(buf, 10, &n);
4555 atomic_set(&mddev->max_corr_read_errors, n);
4559 static struct md_sysfs_entry max_corr_read_errors =
4560 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4561 max_corrected_read_errors_store);
4564 null_show(struct mddev *mddev, char *page)
4569 /* need to ensure rdev_delayed_delete() has completed */
4570 static void flush_rdev_wq(struct mddev *mddev)
4572 struct md_rdev *rdev;
4575 rdev_for_each_rcu(rdev, mddev)
4576 if (work_pending(&rdev->del_work)) {
4577 flush_workqueue(md_rdev_misc_wq);
4584 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4586 /* buf must be %d:%d\n? giving major and minor numbers */
4587 /* The new device is added to the array.
4588 * If the array has a persistent superblock, we read the
4589 * superblock to initialise info and check validity.
4590 * Otherwise, only checking done is that in bind_rdev_to_array,
4591 * which mainly checks size.
4594 int major = simple_strtoul(buf, &e, 10);
4597 struct md_rdev *rdev;
4600 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4602 minor = simple_strtoul(e+1, &e, 10);
4603 if (*e && *e != '\n')
4605 dev = MKDEV(major, minor);
4606 if (major != MAJOR(dev) ||
4607 minor != MINOR(dev))
4610 flush_rdev_wq(mddev);
4611 err = mddev_lock(mddev);
4614 if (mddev->persistent) {
4615 rdev = md_import_device(dev, mddev->major_version,
4616 mddev->minor_version);
4617 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4618 struct md_rdev *rdev0
4619 = list_entry(mddev->disks.next,
4620 struct md_rdev, same_set);
4621 err = super_types[mddev->major_version]
4622 .load_super(rdev, rdev0, mddev->minor_version);
4626 } else if (mddev->external)
4627 rdev = md_import_device(dev, -2, -1);
4629 rdev = md_import_device(dev, -1, -1);
4632 mddev_unlock(mddev);
4633 return PTR_ERR(rdev);
4635 err = bind_rdev_to_array(rdev, mddev);
4639 mddev_unlock(mddev);
4641 md_new_event(mddev);
4642 return err ? err : len;
4645 static struct md_sysfs_entry md_new_device =
4646 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4649 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4652 unsigned long chunk, end_chunk;
4655 err = mddev_lock(mddev);
4660 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4662 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4663 if (buf == end) break;
4664 if (*end == '-') { /* range */
4666 end_chunk = simple_strtoul(buf, &end, 0);
4667 if (buf == end) break;
4669 if (*end && !isspace(*end)) break;
4670 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4671 buf = skip_spaces(end);
4673 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4675 mddev_unlock(mddev);
4679 static struct md_sysfs_entry md_bitmap =
4680 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4683 size_show(struct mddev *mddev, char *page)
4685 return sprintf(page, "%llu\n",
4686 (unsigned long long)mddev->dev_sectors / 2);
4689 static int update_size(struct mddev *mddev, sector_t num_sectors);
4692 size_store(struct mddev *mddev, const char *buf, size_t len)
4694 /* If array is inactive, we can reduce the component size, but
4695 * not increase it (except from 0).
4696 * If array is active, we can try an on-line resize
4699 int err = strict_blocks_to_sectors(buf, §ors);
4703 err = mddev_lock(mddev);
4707 err = update_size(mddev, sectors);
4709 md_update_sb(mddev, 1);
4711 if (mddev->dev_sectors == 0 ||
4712 mddev->dev_sectors > sectors)
4713 mddev->dev_sectors = sectors;
4717 mddev_unlock(mddev);
4718 return err ? err : len;
4721 static struct md_sysfs_entry md_size =
4722 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4724 /* Metadata version.
4726 * 'none' for arrays with no metadata (good luck...)
4727 * 'external' for arrays with externally managed metadata,
4728 * or N.M for internally known formats
4731 metadata_show(struct mddev *mddev, char *page)
4733 if (mddev->persistent)
4734 return sprintf(page, "%d.%d\n",
4735 mddev->major_version, mddev->minor_version);
4736 else if (mddev->external)
4737 return sprintf(page, "external:%s\n", mddev->metadata_type);
4739 return sprintf(page, "none\n");
4743 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4748 /* Changing the details of 'external' metadata is
4749 * always permitted. Otherwise there must be
4750 * no devices attached to the array.
4753 err = mddev_lock(mddev);
4757 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4759 else if (!list_empty(&mddev->disks))
4763 if (cmd_match(buf, "none")) {
4764 mddev->persistent = 0;
4765 mddev->external = 0;
4766 mddev->major_version = 0;
4767 mddev->minor_version = 90;
4770 if (strncmp(buf, "external:", 9) == 0) {
4771 size_t namelen = len-9;
4772 if (namelen >= sizeof(mddev->metadata_type))
4773 namelen = sizeof(mddev->metadata_type)-1;
4774 strncpy(mddev->metadata_type, buf+9, namelen);
4775 mddev->metadata_type[namelen] = 0;
4776 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4777 mddev->metadata_type[--namelen] = 0;
4778 mddev->persistent = 0;
4779 mddev->external = 1;
4780 mddev->major_version = 0;
4781 mddev->minor_version = 90;
4784 major = simple_strtoul(buf, &e, 10);
4786 if (e==buf || *e != '.')
4789 minor = simple_strtoul(buf, &e, 10);
4790 if (e==buf || (*e && *e != '\n') )
4793 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4795 mddev->major_version = major;
4796 mddev->minor_version = minor;
4797 mddev->persistent = 1;
4798 mddev->external = 0;
4801 mddev_unlock(mddev);
4805 static struct md_sysfs_entry md_metadata =
4806 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4809 action_show(struct mddev *mddev, char *page)
4811 char *type = "idle";
4812 unsigned long recovery = mddev->recovery;
4813 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4815 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4816 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4817 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4819 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4820 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4822 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4826 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4828 else if (mddev->reshape_position != MaxSector)
4831 return sprintf(page, "%s\n", type);
4835 action_store(struct mddev *mddev, const char *page, size_t len)
4837 if (!mddev->pers || !mddev->pers->sync_request)
4841 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4842 if (cmd_match(page, "frozen"))
4843 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4845 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4846 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4847 mddev_lock(mddev) == 0) {
4848 if (work_pending(&mddev->del_work))
4849 flush_workqueue(md_misc_wq);
4850 if (mddev->sync_thread) {
4851 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4852 md_reap_sync_thread(mddev);
4854 mddev_unlock(mddev);
4856 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4858 else if (cmd_match(page, "resync"))
4859 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4860 else if (cmd_match(page, "recover")) {
4861 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4862 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4863 } else if (cmd_match(page, "reshape")) {
4865 if (mddev->pers->start_reshape == NULL)
4867 err = mddev_lock(mddev);
4869 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4872 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4873 err = mddev->pers->start_reshape(mddev);
4875 mddev_unlock(mddev);
4879 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4881 if (cmd_match(page, "check"))
4882 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4883 else if (!cmd_match(page, "repair"))
4885 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4886 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4887 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4889 if (mddev->ro == 2) {
4890 /* A write to sync_action is enough to justify
4891 * canceling read-auto mode
4894 md_wakeup_thread(mddev->sync_thread);
4896 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4897 md_wakeup_thread(mddev->thread);
4898 sysfs_notify_dirent_safe(mddev->sysfs_action);
4902 static struct md_sysfs_entry md_scan_mode =
4903 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4906 last_sync_action_show(struct mddev *mddev, char *page)
4908 return sprintf(page, "%s\n", mddev->last_sync_action);
4911 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4914 mismatch_cnt_show(struct mddev *mddev, char *page)
4916 return sprintf(page, "%llu\n",
4917 (unsigned long long)
4918 atomic64_read(&mddev->resync_mismatches));
4921 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4924 sync_min_show(struct mddev *mddev, char *page)
4926 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4927 mddev->sync_speed_min ? "local": "system");
4931 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4936 if (strncmp(buf, "system", 6)==0) {
4939 rv = kstrtouint(buf, 10, &min);
4945 mddev->sync_speed_min = min;
4949 static struct md_sysfs_entry md_sync_min =
4950 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4953 sync_max_show(struct mddev *mddev, char *page)
4955 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4956 mddev->sync_speed_max ? "local": "system");
4960 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4965 if (strncmp(buf, "system", 6)==0) {
4968 rv = kstrtouint(buf, 10, &max);
4974 mddev->sync_speed_max = max;
4978 static struct md_sysfs_entry md_sync_max =
4979 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4982 degraded_show(struct mddev *mddev, char *page)
4984 return sprintf(page, "%d\n", mddev->degraded);
4986 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4989 sync_force_parallel_show(struct mddev *mddev, char *page)
4991 return sprintf(page, "%d\n", mddev->parallel_resync);
4995 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4999 if (kstrtol(buf, 10, &n))
5002 if (n != 0 && n != 1)
5005 mddev->parallel_resync = n;
5007 if (mddev->sync_thread)
5008 wake_up(&resync_wait);
5013 /* force parallel resync, even with shared block devices */
5014 static struct md_sysfs_entry md_sync_force_parallel =
5015 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5016 sync_force_parallel_show, sync_force_parallel_store);
5019 sync_speed_show(struct mddev *mddev, char *page)
5021 unsigned long resync, dt, db;
5022 if (mddev->curr_resync == 0)
5023 return sprintf(page, "none\n");
5024 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5025 dt = (jiffies - mddev->resync_mark) / HZ;
5027 db = resync - mddev->resync_mark_cnt;
5028 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5031 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5034 sync_completed_show(struct mddev *mddev, char *page)
5036 unsigned long long max_sectors, resync;
5038 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5039 return sprintf(page, "none\n");
5041 if (mddev->curr_resync == 1 ||
5042 mddev->curr_resync == 2)
5043 return sprintf(page, "delayed\n");
5045 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5046 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5047 max_sectors = mddev->resync_max_sectors;
5049 max_sectors = mddev->dev_sectors;
5051 resync = mddev->curr_resync_completed;
5052 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5055 static struct md_sysfs_entry md_sync_completed =
5056 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5059 min_sync_show(struct mddev *mddev, char *page)
5061 return sprintf(page, "%llu\n",
5062 (unsigned long long)mddev->resync_min);
5065 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5067 unsigned long long min;
5070 if (kstrtoull(buf, 10, &min))
5073 spin_lock(&mddev->lock);
5075 if (min > mddev->resync_max)
5079 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5082 /* Round down to multiple of 4K for safety */
5083 mddev->resync_min = round_down(min, 8);
5087 spin_unlock(&mddev->lock);
5091 static struct md_sysfs_entry md_min_sync =
5092 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5095 max_sync_show(struct mddev *mddev, char *page)
5097 if (mddev->resync_max == MaxSector)
5098 return sprintf(page, "max\n");
5100 return sprintf(page, "%llu\n",
5101 (unsigned long long)mddev->resync_max);
5104 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5107 spin_lock(&mddev->lock);
5108 if (strncmp(buf, "max", 3) == 0)
5109 mddev->resync_max = MaxSector;
5111 unsigned long long max;
5115 if (kstrtoull(buf, 10, &max))
5117 if (max < mddev->resync_min)
5121 if (max < mddev->resync_max &&
5123 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5126 /* Must be a multiple of chunk_size */
5127 chunk = mddev->chunk_sectors;
5129 sector_t temp = max;
5132 if (sector_div(temp, chunk))
5135 mddev->resync_max = max;
5137 wake_up(&mddev->recovery_wait);
5140 spin_unlock(&mddev->lock);
5144 static struct md_sysfs_entry md_max_sync =
5145 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5148 suspend_lo_show(struct mddev *mddev, char *page)
5150 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5154 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5156 unsigned long long new;
5159 err = kstrtoull(buf, 10, &new);
5162 if (new != (sector_t)new)
5165 err = mddev_lock(mddev);
5169 if (mddev->pers == NULL ||
5170 mddev->pers->quiesce == NULL)
5172 mddev_suspend(mddev);
5173 mddev->suspend_lo = new;
5174 mddev_resume(mddev);
5178 mddev_unlock(mddev);
5181 static struct md_sysfs_entry md_suspend_lo =
5182 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5185 suspend_hi_show(struct mddev *mddev, char *page)
5187 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5191 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5193 unsigned long long new;
5196 err = kstrtoull(buf, 10, &new);
5199 if (new != (sector_t)new)
5202 err = mddev_lock(mddev);
5206 if (mddev->pers == NULL)
5209 mddev_suspend(mddev);
5210 mddev->suspend_hi = new;
5211 mddev_resume(mddev);
5215 mddev_unlock(mddev);
5218 static struct md_sysfs_entry md_suspend_hi =
5219 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5222 reshape_position_show(struct mddev *mddev, char *page)
5224 if (mddev->reshape_position != MaxSector)
5225 return sprintf(page, "%llu\n",
5226 (unsigned long long)mddev->reshape_position);
5227 strcpy(page, "none\n");
5232 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5234 struct md_rdev *rdev;
5235 unsigned long long new;
5238 err = kstrtoull(buf, 10, &new);
5241 if (new != (sector_t)new)
5243 err = mddev_lock(mddev);
5249 mddev->reshape_position = new;
5250 mddev->delta_disks = 0;
5251 mddev->reshape_backwards = 0;
5252 mddev->new_level = mddev->level;
5253 mddev->new_layout = mddev->layout;
5254 mddev->new_chunk_sectors = mddev->chunk_sectors;
5255 rdev_for_each(rdev, mddev)
5256 rdev->new_data_offset = rdev->data_offset;
5259 mddev_unlock(mddev);
5263 static struct md_sysfs_entry md_reshape_position =
5264 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5265 reshape_position_store);
5268 reshape_direction_show(struct mddev *mddev, char *page)
5270 return sprintf(page, "%s\n",
5271 mddev->reshape_backwards ? "backwards" : "forwards");
5275 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5280 if (cmd_match(buf, "forwards"))
5282 else if (cmd_match(buf, "backwards"))
5286 if (mddev->reshape_backwards == backwards)
5289 err = mddev_lock(mddev);
5292 /* check if we are allowed to change */
5293 if (mddev->delta_disks)
5295 else if (mddev->persistent &&
5296 mddev->major_version == 0)
5299 mddev->reshape_backwards = backwards;
5300 mddev_unlock(mddev);
5304 static struct md_sysfs_entry md_reshape_direction =
5305 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5306 reshape_direction_store);
5309 array_size_show(struct mddev *mddev, char *page)
5311 if (mddev->external_size)
5312 return sprintf(page, "%llu\n",
5313 (unsigned long long)mddev->array_sectors/2);
5315 return sprintf(page, "default\n");
5319 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5324 err = mddev_lock(mddev);
5328 /* cluster raid doesn't support change array_sectors */
5329 if (mddev_is_clustered(mddev)) {
5330 mddev_unlock(mddev);
5334 if (strncmp(buf, "default", 7) == 0) {
5336 sectors = mddev->pers->size(mddev, 0, 0);
5338 sectors = mddev->array_sectors;
5340 mddev->external_size = 0;
5342 if (strict_blocks_to_sectors(buf, §ors) < 0)
5344 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5347 mddev->external_size = 1;
5351 mddev->array_sectors = sectors;
5353 set_capacity_and_notify(mddev->gendisk,
5354 mddev->array_sectors);
5356 mddev_unlock(mddev);
5360 static struct md_sysfs_entry md_array_size =
5361 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5365 consistency_policy_show(struct mddev *mddev, char *page)
5369 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5370 ret = sprintf(page, "journal\n");
5371 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5372 ret = sprintf(page, "ppl\n");
5373 } else if (mddev->bitmap) {
5374 ret = sprintf(page, "bitmap\n");
5375 } else if (mddev->pers) {
5376 if (mddev->pers->sync_request)
5377 ret = sprintf(page, "resync\n");
5379 ret = sprintf(page, "none\n");
5381 ret = sprintf(page, "unknown\n");
5388 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5393 if (mddev->pers->change_consistency_policy)
5394 err = mddev->pers->change_consistency_policy(mddev, buf);
5397 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5398 set_bit(MD_HAS_PPL, &mddev->flags);
5403 return err ? err : len;
5406 static struct md_sysfs_entry md_consistency_policy =
5407 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5408 consistency_policy_store);
5410 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5412 return sprintf(page, "%d\n", mddev->fail_last_dev);
5416 * Setting fail_last_dev to true to allow last device to be forcibly removed
5417 * from RAID1/RAID10.
5420 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5425 ret = kstrtobool(buf, &value);
5429 if (value != mddev->fail_last_dev)
5430 mddev->fail_last_dev = value;
5434 static struct md_sysfs_entry md_fail_last_dev =
5435 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5436 fail_last_dev_store);
5438 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5440 if (mddev->pers == NULL || (mddev->pers->level != 1))
5441 return sprintf(page, "n/a\n");
5443 return sprintf(page, "%d\n", mddev->serialize_policy);
5447 * Setting serialize_policy to true to enforce write IO is not reordered
5451 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5456 err = kstrtobool(buf, &value);
5460 if (value == mddev->serialize_policy)
5463 err = mddev_lock(mddev);
5466 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5467 pr_err("md: serialize_policy is only effective for raid1\n");
5472 mddev_suspend(mddev);
5474 mddev_create_serial_pool(mddev, NULL, true);
5476 mddev_destroy_serial_pool(mddev, NULL, true);
5477 mddev->serialize_policy = value;
5478 mddev_resume(mddev);
5480 mddev_unlock(mddev);
5484 static struct md_sysfs_entry md_serialize_policy =
5485 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5486 serialize_policy_store);
5489 static struct attribute *md_default_attrs[] = {
5492 &md_raid_disks.attr,
5494 &md_chunk_size.attr,
5496 &md_resync_start.attr,
5498 &md_new_device.attr,
5499 &md_safe_delay.attr,
5500 &md_array_state.attr,
5501 &md_reshape_position.attr,
5502 &md_reshape_direction.attr,
5503 &md_array_size.attr,
5504 &max_corr_read_errors.attr,
5505 &md_consistency_policy.attr,
5506 &md_fail_last_dev.attr,
5507 &md_serialize_policy.attr,
5511 static struct attribute *md_redundancy_attrs[] = {
5513 &md_last_scan_mode.attr,
5514 &md_mismatches.attr,
5517 &md_sync_speed.attr,
5518 &md_sync_force_parallel.attr,
5519 &md_sync_completed.attr,
5522 &md_suspend_lo.attr,
5523 &md_suspend_hi.attr,
5528 static const struct attribute_group md_redundancy_group = {
5530 .attrs = md_redundancy_attrs,
5534 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5536 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5537 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5542 spin_lock(&all_mddevs_lock);
5543 if (list_empty(&mddev->all_mddevs)) {
5544 spin_unlock(&all_mddevs_lock);
5548 spin_unlock(&all_mddevs_lock);
5550 rv = entry->show(mddev, page);
5556 md_attr_store(struct kobject *kobj, struct attribute *attr,
5557 const char *page, size_t length)
5559 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5560 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5565 if (!capable(CAP_SYS_ADMIN))
5567 spin_lock(&all_mddevs_lock);
5568 if (list_empty(&mddev->all_mddevs)) {
5569 spin_unlock(&all_mddevs_lock);
5573 spin_unlock(&all_mddevs_lock);
5574 rv = entry->store(mddev, page, length);
5579 static void md_free(struct kobject *ko)
5581 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5583 if (mddev->sysfs_state)
5584 sysfs_put(mddev->sysfs_state);
5585 if (mddev->sysfs_level)
5586 sysfs_put(mddev->sysfs_level);
5588 if (mddev->gendisk) {
5589 del_gendisk(mddev->gendisk);
5590 blk_cleanup_disk(mddev->gendisk);
5592 percpu_ref_exit(&mddev->writes_pending);
5594 bioset_exit(&mddev->bio_set);
5595 bioset_exit(&mddev->sync_set);
5599 static const struct sysfs_ops md_sysfs_ops = {
5600 .show = md_attr_show,
5601 .store = md_attr_store,
5603 static struct kobj_type md_ktype = {
5605 .sysfs_ops = &md_sysfs_ops,
5606 .default_attrs = md_default_attrs,
5611 static void mddev_delayed_delete(struct work_struct *ws)
5613 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5615 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5616 kobject_del(&mddev->kobj);
5617 kobject_put(&mddev->kobj);
5620 static void no_op(struct percpu_ref *r) {}
5622 int mddev_init_writes_pending(struct mddev *mddev)
5624 if (mddev->writes_pending.percpu_count_ptr)
5626 if (percpu_ref_init(&mddev->writes_pending, no_op,
5627 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5629 /* We want to start with the refcount at zero */
5630 percpu_ref_put(&mddev->writes_pending);
5633 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5635 static int md_alloc(dev_t dev, char *name)
5638 * If dev is zero, name is the name of a device to allocate with
5639 * an arbitrary minor number. It will be "md_???"
5640 * If dev is non-zero it must be a device number with a MAJOR of
5641 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5642 * the device is being created by opening a node in /dev.
5643 * If "name" is not NULL, the device is being created by
5644 * writing to /sys/module/md_mod/parameters/new_array.
5646 static DEFINE_MUTEX(disks_mutex);
5647 struct mddev *mddev;
5648 struct gendisk *disk;
5655 * Wait for any previous instance of this device to be completely
5656 * removed (mddev_delayed_delete).
5658 flush_workqueue(md_misc_wq);
5659 flush_workqueue(md_rdev_misc_wq);
5661 mutex_lock(&disks_mutex);
5662 mddev = mddev_alloc(dev);
5663 if (IS_ERR(mddev)) {
5664 mutex_unlock(&disks_mutex);
5665 return PTR_ERR(mddev);
5668 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5669 shift = partitioned ? MdpMinorShift : 0;
5670 unit = MINOR(mddev->unit) >> shift;
5673 /* Need to ensure that 'name' is not a duplicate.
5675 struct mddev *mddev2;
5676 spin_lock(&all_mddevs_lock);
5678 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5679 if (mddev2->gendisk &&
5680 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5681 spin_unlock(&all_mddevs_lock);
5685 spin_unlock(&all_mddevs_lock);
5689 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5691 mddev->hold_active = UNTIL_STOP;
5694 disk = blk_alloc_disk(NUMA_NO_NODE);
5698 disk->major = MAJOR(mddev->unit);
5699 disk->first_minor = unit << shift;
5700 disk->minors = 1 << shift;
5702 strcpy(disk->disk_name, name);
5703 else if (partitioned)
5704 sprintf(disk->disk_name, "md_d%d", unit);
5706 sprintf(disk->disk_name, "md%d", unit);
5707 disk->fops = &md_fops;
5708 disk->private_data = mddev;
5710 mddev->queue = disk->queue;
5711 blk_set_stacking_limits(&mddev->queue->limits);
5712 blk_queue_write_cache(mddev->queue, true, true);
5713 /* Allow extended partitions. This makes the
5714 * 'mdp' device redundant, but we can't really
5717 disk->flags |= GENHD_FL_EXT_DEVT;
5718 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5719 mddev->gendisk = disk;
5722 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5724 /* This isn't possible, but as kobject_init_and_add is marked
5725 * __must_check, we must do something with the result
5727 pr_debug("md: cannot register %s/md - name in use\n",
5731 if (mddev->kobj.sd &&
5732 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5733 pr_debug("pointless warning\n");
5735 mutex_unlock(&disks_mutex);
5736 if (!error && mddev->kobj.sd) {
5737 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5738 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5739 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5745 static void md_probe(dev_t dev)
5747 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5750 md_alloc(dev, NULL);
5753 static int add_named_array(const char *val, const struct kernel_param *kp)
5756 * val must be "md_*" or "mdNNN".
5757 * For "md_*" we allocate an array with a large free minor number, and
5758 * set the name to val. val must not already be an active name.
5759 * For "mdNNN" we allocate an array with the minor number NNN
5760 * which must not already be in use.
5762 int len = strlen(val);
5763 char buf[DISK_NAME_LEN];
5764 unsigned long devnum;
5766 while (len && val[len-1] == '\n')
5768 if (len >= DISK_NAME_LEN)
5770 strlcpy(buf, val, len+1);
5771 if (strncmp(buf, "md_", 3) == 0)
5772 return md_alloc(0, buf);
5773 if (strncmp(buf, "md", 2) == 0 &&
5775 kstrtoul(buf+2, 10, &devnum) == 0 &&
5776 devnum <= MINORMASK)
5777 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5782 static void md_safemode_timeout(struct timer_list *t)
5784 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5786 mddev->safemode = 1;
5787 if (mddev->external)
5788 sysfs_notify_dirent_safe(mddev->sysfs_state);
5790 md_wakeup_thread(mddev->thread);
5793 static int start_dirty_degraded;
5795 int md_run(struct mddev *mddev)
5798 struct md_rdev *rdev;
5799 struct md_personality *pers;
5801 if (list_empty(&mddev->disks))
5802 /* cannot run an array with no devices.. */
5807 /* Cannot run until previous stop completes properly */
5808 if (mddev->sysfs_active)
5812 * Analyze all RAID superblock(s)
5814 if (!mddev->raid_disks) {
5815 if (!mddev->persistent)
5817 err = analyze_sbs(mddev);
5822 if (mddev->level != LEVEL_NONE)
5823 request_module("md-level-%d", mddev->level);
5824 else if (mddev->clevel[0])
5825 request_module("md-%s", mddev->clevel);
5828 * Drop all container device buffers, from now on
5829 * the only valid external interface is through the md
5832 mddev->has_superblocks = false;
5833 rdev_for_each(rdev, mddev) {
5834 if (test_bit(Faulty, &rdev->flags))
5836 sync_blockdev(rdev->bdev);
5837 invalidate_bdev(rdev->bdev);
5838 if (mddev->ro != 1 && rdev_read_only(rdev)) {
5841 set_disk_ro(mddev->gendisk, 1);
5845 mddev->has_superblocks = true;
5847 /* perform some consistency tests on the device.
5848 * We don't want the data to overlap the metadata,
5849 * Internal Bitmap issues have been handled elsewhere.
5851 if (rdev->meta_bdev) {
5852 /* Nothing to check */;
5853 } else if (rdev->data_offset < rdev->sb_start) {
5854 if (mddev->dev_sectors &&
5855 rdev->data_offset + mddev->dev_sectors
5857 pr_warn("md: %s: data overlaps metadata\n",
5862 if (rdev->sb_start + rdev->sb_size/512
5863 > rdev->data_offset) {
5864 pr_warn("md: %s: metadata overlaps data\n",
5869 sysfs_notify_dirent_safe(rdev->sysfs_state);
5872 if (!bioset_initialized(&mddev->bio_set)) {
5873 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5877 if (!bioset_initialized(&mddev->sync_set)) {
5878 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5883 spin_lock(&pers_lock);
5884 pers = find_pers(mddev->level, mddev->clevel);
5885 if (!pers || !try_module_get(pers->owner)) {
5886 spin_unlock(&pers_lock);
5887 if (mddev->level != LEVEL_NONE)
5888 pr_warn("md: personality for level %d is not loaded!\n",
5891 pr_warn("md: personality for level %s is not loaded!\n",
5896 spin_unlock(&pers_lock);
5897 if (mddev->level != pers->level) {
5898 mddev->level = pers->level;
5899 mddev->new_level = pers->level;
5901 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5903 if (mddev->reshape_position != MaxSector &&
5904 pers->start_reshape == NULL) {
5905 /* This personality cannot handle reshaping... */
5906 module_put(pers->owner);
5911 if (pers->sync_request) {
5912 /* Warn if this is a potentially silly
5915 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5916 struct md_rdev *rdev2;
5919 rdev_for_each(rdev, mddev)
5920 rdev_for_each(rdev2, mddev) {
5922 rdev->bdev->bd_disk ==
5923 rdev2->bdev->bd_disk) {
5924 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5926 bdevname(rdev->bdev,b),
5927 bdevname(rdev2->bdev,b2));
5933 pr_warn("True protection against single-disk failure might be compromised.\n");
5936 mddev->recovery = 0;
5937 /* may be over-ridden by personality */
5938 mddev->resync_max_sectors = mddev->dev_sectors;
5940 mddev->ok_start_degraded = start_dirty_degraded;
5942 if (start_readonly && mddev->ro == 0)
5943 mddev->ro = 2; /* read-only, but switch on first write */
5945 err = pers->run(mddev);
5947 pr_warn("md: pers->run() failed ...\n");
5948 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5949 WARN_ONCE(!mddev->external_size,
5950 "%s: default size too small, but 'external_size' not in effect?\n",
5952 pr_warn("md: invalid array_size %llu > default size %llu\n",
5953 (unsigned long long)mddev->array_sectors / 2,
5954 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5957 if (err == 0 && pers->sync_request &&
5958 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5959 struct bitmap *bitmap;
5961 bitmap = md_bitmap_create(mddev, -1);
5962 if (IS_ERR(bitmap)) {
5963 err = PTR_ERR(bitmap);
5964 pr_warn("%s: failed to create bitmap (%d)\n",
5965 mdname(mddev), err);
5967 mddev->bitmap = bitmap;
5973 if (mddev->bitmap_info.max_write_behind > 0) {
5974 bool create_pool = false;
5976 rdev_for_each(rdev, mddev) {
5977 if (test_bit(WriteMostly, &rdev->flags) &&
5978 rdev_init_serial(rdev))
5981 if (create_pool && mddev->serial_info_pool == NULL) {
5982 mddev->serial_info_pool =
5983 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5984 sizeof(struct serial_info));
5985 if (!mddev->serial_info_pool) {
5995 rdev_for_each(rdev, mddev) {
5996 if (rdev->raid_disk >= 0 &&
5997 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6002 if (mddev->degraded)
6005 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6007 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6008 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6010 if (pers->sync_request) {
6011 if (mddev->kobj.sd &&
6012 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6013 pr_warn("md: cannot register extra attributes for %s\n",
6015 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6016 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6017 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6018 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6021 atomic_set(&mddev->max_corr_read_errors,
6022 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6023 mddev->safemode = 0;
6024 if (mddev_is_clustered(mddev))
6025 mddev->safemode_delay = 0;
6027 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6030 spin_lock(&mddev->lock);
6032 spin_unlock(&mddev->lock);
6033 rdev_for_each(rdev, mddev)
6034 if (rdev->raid_disk >= 0)
6035 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6037 if (mddev->degraded && !mddev->ro)
6038 /* This ensures that recovering status is reported immediately
6039 * via sysfs - until a lack of spares is confirmed.
6041 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6042 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6044 if (mddev->sb_flags)
6045 md_update_sb(mddev, 0);
6047 md_new_event(mddev);
6051 mddev_detach(mddev);
6053 pers->free(mddev, mddev->private);
6054 mddev->private = NULL;
6055 module_put(pers->owner);
6056 md_bitmap_destroy(mddev);
6058 bioset_exit(&mddev->sync_set);
6060 bioset_exit(&mddev->bio_set);
6063 EXPORT_SYMBOL_GPL(md_run);
6065 int do_md_run(struct mddev *mddev)
6069 set_bit(MD_NOT_READY, &mddev->flags);
6070 err = md_run(mddev);
6073 err = md_bitmap_load(mddev);
6075 md_bitmap_destroy(mddev);
6079 if (mddev_is_clustered(mddev))
6080 md_allow_write(mddev);
6082 /* run start up tasks that require md_thread */
6085 md_wakeup_thread(mddev->thread);
6086 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6088 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6089 clear_bit(MD_NOT_READY, &mddev->flags);
6091 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6092 sysfs_notify_dirent_safe(mddev->sysfs_state);
6093 sysfs_notify_dirent_safe(mddev->sysfs_action);
6094 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6096 clear_bit(MD_NOT_READY, &mddev->flags);
6100 int md_start(struct mddev *mddev)
6104 if (mddev->pers->start) {
6105 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6106 md_wakeup_thread(mddev->thread);
6107 ret = mddev->pers->start(mddev);
6108 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6109 md_wakeup_thread(mddev->sync_thread);
6113 EXPORT_SYMBOL_GPL(md_start);
6115 static int restart_array(struct mddev *mddev)
6117 struct gendisk *disk = mddev->gendisk;
6118 struct md_rdev *rdev;
6119 bool has_journal = false;
6120 bool has_readonly = false;
6122 /* Complain if it has no devices */
6123 if (list_empty(&mddev->disks))
6131 rdev_for_each_rcu(rdev, mddev) {
6132 if (test_bit(Journal, &rdev->flags) &&
6133 !test_bit(Faulty, &rdev->flags))
6135 if (rdev_read_only(rdev))
6136 has_readonly = true;
6139 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6140 /* Don't restart rw with journal missing/faulty */
6145 mddev->safemode = 0;
6147 set_disk_ro(disk, 0);
6148 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6149 /* Kick recovery or resync if necessary */
6150 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6151 md_wakeup_thread(mddev->thread);
6152 md_wakeup_thread(mddev->sync_thread);
6153 sysfs_notify_dirent_safe(mddev->sysfs_state);
6157 static void md_clean(struct mddev *mddev)
6159 mddev->array_sectors = 0;
6160 mddev->external_size = 0;
6161 mddev->dev_sectors = 0;
6162 mddev->raid_disks = 0;
6163 mddev->recovery_cp = 0;
6164 mddev->resync_min = 0;
6165 mddev->resync_max = MaxSector;
6166 mddev->reshape_position = MaxSector;
6167 mddev->external = 0;
6168 mddev->persistent = 0;
6169 mddev->level = LEVEL_NONE;
6170 mddev->clevel[0] = 0;
6172 mddev->sb_flags = 0;
6174 mddev->metadata_type[0] = 0;
6175 mddev->chunk_sectors = 0;
6176 mddev->ctime = mddev->utime = 0;
6178 mddev->max_disks = 0;
6180 mddev->can_decrease_events = 0;
6181 mddev->delta_disks = 0;
6182 mddev->reshape_backwards = 0;
6183 mddev->new_level = LEVEL_NONE;
6184 mddev->new_layout = 0;
6185 mddev->new_chunk_sectors = 0;
6186 mddev->curr_resync = 0;
6187 atomic64_set(&mddev->resync_mismatches, 0);
6188 mddev->suspend_lo = mddev->suspend_hi = 0;
6189 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6190 mddev->recovery = 0;
6193 mddev->degraded = 0;
6194 mddev->safemode = 0;
6195 mddev->private = NULL;
6196 mddev->cluster_info = NULL;
6197 mddev->bitmap_info.offset = 0;
6198 mddev->bitmap_info.default_offset = 0;
6199 mddev->bitmap_info.default_space = 0;
6200 mddev->bitmap_info.chunksize = 0;
6201 mddev->bitmap_info.daemon_sleep = 0;
6202 mddev->bitmap_info.max_write_behind = 0;
6203 mddev->bitmap_info.nodes = 0;
6206 static void __md_stop_writes(struct mddev *mddev)
6208 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6209 if (work_pending(&mddev->del_work))
6210 flush_workqueue(md_misc_wq);
6211 if (mddev->sync_thread) {
6212 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6213 md_reap_sync_thread(mddev);
6216 del_timer_sync(&mddev->safemode_timer);
6218 if (mddev->pers && mddev->pers->quiesce) {
6219 mddev->pers->quiesce(mddev, 1);
6220 mddev->pers->quiesce(mddev, 0);
6222 md_bitmap_flush(mddev);
6224 if (mddev->ro == 0 &&
6225 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6227 /* mark array as shutdown cleanly */
6228 if (!mddev_is_clustered(mddev))
6230 md_update_sb(mddev, 1);
6232 /* disable policy to guarantee rdevs free resources for serialization */
6233 mddev->serialize_policy = 0;
6234 mddev_destroy_serial_pool(mddev, NULL, true);
6237 void md_stop_writes(struct mddev *mddev)
6239 mddev_lock_nointr(mddev);
6240 __md_stop_writes(mddev);
6241 mddev_unlock(mddev);
6243 EXPORT_SYMBOL_GPL(md_stop_writes);
6245 static void mddev_detach(struct mddev *mddev)
6247 md_bitmap_wait_behind_writes(mddev);
6248 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6249 mddev->pers->quiesce(mddev, 1);
6250 mddev->pers->quiesce(mddev, 0);
6252 md_unregister_thread(&mddev->thread);
6254 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6257 static void __md_stop(struct mddev *mddev)
6259 struct md_personality *pers = mddev->pers;
6260 md_bitmap_destroy(mddev);
6261 mddev_detach(mddev);
6262 /* Ensure ->event_work is done */
6263 if (mddev->event_work.func)
6264 flush_workqueue(md_misc_wq);
6265 spin_lock(&mddev->lock);
6267 spin_unlock(&mddev->lock);
6268 pers->free(mddev, mddev->private);
6269 mddev->private = NULL;
6270 if (pers->sync_request && mddev->to_remove == NULL)
6271 mddev->to_remove = &md_redundancy_group;
6272 module_put(pers->owner);
6273 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6276 void md_stop(struct mddev *mddev)
6278 /* stop the array and free an attached data structures.
6279 * This is called from dm-raid
6281 __md_stop_writes(mddev);
6283 bioset_exit(&mddev->bio_set);
6284 bioset_exit(&mddev->sync_set);
6287 EXPORT_SYMBOL_GPL(md_stop);
6289 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6294 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6296 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6297 md_wakeup_thread(mddev->thread);
6299 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6300 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6301 if (mddev->sync_thread)
6302 /* Thread might be blocked waiting for metadata update
6303 * which will now never happen */
6304 wake_up_process(mddev->sync_thread->tsk);
6306 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6308 mddev_unlock(mddev);
6309 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6311 wait_event(mddev->sb_wait,
6312 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6313 mddev_lock_nointr(mddev);
6315 mutex_lock(&mddev->open_mutex);
6316 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6317 mddev->sync_thread ||
6318 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6319 pr_warn("md: %s still in use.\n",mdname(mddev));
6321 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6322 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6323 md_wakeup_thread(mddev->thread);
6329 __md_stop_writes(mddev);
6335 set_disk_ro(mddev->gendisk, 1);
6336 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6337 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6338 md_wakeup_thread(mddev->thread);
6339 sysfs_notify_dirent_safe(mddev->sysfs_state);
6343 mutex_unlock(&mddev->open_mutex);
6348 * 0 - completely stop and dis-assemble array
6349 * 2 - stop but do not disassemble array
6351 static int do_md_stop(struct mddev *mddev, int mode,
6352 struct block_device *bdev)
6354 struct gendisk *disk = mddev->gendisk;
6355 struct md_rdev *rdev;
6358 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6360 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6361 md_wakeup_thread(mddev->thread);
6363 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6364 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6365 if (mddev->sync_thread)
6366 /* Thread might be blocked waiting for metadata update
6367 * which will now never happen */
6368 wake_up_process(mddev->sync_thread->tsk);
6370 mddev_unlock(mddev);
6371 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6372 !test_bit(MD_RECOVERY_RUNNING,
6373 &mddev->recovery)));
6374 mddev_lock_nointr(mddev);
6376 mutex_lock(&mddev->open_mutex);
6377 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6378 mddev->sysfs_active ||
6379 mddev->sync_thread ||
6380 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6381 pr_warn("md: %s still in use.\n",mdname(mddev));
6382 mutex_unlock(&mddev->open_mutex);
6384 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6385 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6386 md_wakeup_thread(mddev->thread);
6392 set_disk_ro(disk, 0);
6394 __md_stop_writes(mddev);
6397 /* tell userspace to handle 'inactive' */
6398 sysfs_notify_dirent_safe(mddev->sysfs_state);
6400 rdev_for_each(rdev, mddev)
6401 if (rdev->raid_disk >= 0)
6402 sysfs_unlink_rdev(mddev, rdev);
6404 set_capacity_and_notify(disk, 0);
6405 mutex_unlock(&mddev->open_mutex);
6411 mutex_unlock(&mddev->open_mutex);
6413 * Free resources if final stop
6416 pr_info("md: %s stopped.\n", mdname(mddev));
6418 if (mddev->bitmap_info.file) {
6419 struct file *f = mddev->bitmap_info.file;
6420 spin_lock(&mddev->lock);
6421 mddev->bitmap_info.file = NULL;
6422 spin_unlock(&mddev->lock);
6425 mddev->bitmap_info.offset = 0;
6427 export_array(mddev);
6430 if (mddev->hold_active == UNTIL_STOP)
6431 mddev->hold_active = 0;
6433 md_new_event(mddev);
6434 sysfs_notify_dirent_safe(mddev->sysfs_state);
6439 static void autorun_array(struct mddev *mddev)
6441 struct md_rdev *rdev;
6444 if (list_empty(&mddev->disks))
6447 pr_info("md: running: ");
6449 rdev_for_each(rdev, mddev) {
6450 char b[BDEVNAME_SIZE];
6451 pr_cont("<%s>", bdevname(rdev->bdev,b));
6455 err = do_md_run(mddev);
6457 pr_warn("md: do_md_run() returned %d\n", err);
6458 do_md_stop(mddev, 0, NULL);
6463 * lets try to run arrays based on all disks that have arrived
6464 * until now. (those are in pending_raid_disks)
6466 * the method: pick the first pending disk, collect all disks with
6467 * the same UUID, remove all from the pending list and put them into
6468 * the 'same_array' list. Then order this list based on superblock
6469 * update time (freshest comes first), kick out 'old' disks and
6470 * compare superblocks. If everything's fine then run it.
6472 * If "unit" is allocated, then bump its reference count
6474 static void autorun_devices(int part)
6476 struct md_rdev *rdev0, *rdev, *tmp;
6477 struct mddev *mddev;
6478 char b[BDEVNAME_SIZE];
6480 pr_info("md: autorun ...\n");
6481 while (!list_empty(&pending_raid_disks)) {
6484 LIST_HEAD(candidates);
6485 rdev0 = list_entry(pending_raid_disks.next,
6486 struct md_rdev, same_set);
6488 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6489 INIT_LIST_HEAD(&candidates);
6490 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6491 if (super_90_load(rdev, rdev0, 0) >= 0) {
6492 pr_debug("md: adding %s ...\n",
6493 bdevname(rdev->bdev,b));
6494 list_move(&rdev->same_set, &candidates);
6497 * now we have a set of devices, with all of them having
6498 * mostly sane superblocks. It's time to allocate the
6502 dev = MKDEV(mdp_major,
6503 rdev0->preferred_minor << MdpMinorShift);
6504 unit = MINOR(dev) >> MdpMinorShift;
6506 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6509 if (rdev0->preferred_minor != unit) {
6510 pr_warn("md: unit number in %s is bad: %d\n",
6511 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6516 mddev = mddev_find(dev);
6520 if (mddev_lock(mddev))
6521 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6522 else if (mddev->raid_disks || mddev->major_version
6523 || !list_empty(&mddev->disks)) {
6524 pr_warn("md: %s already running, cannot run %s\n",
6525 mdname(mddev), bdevname(rdev0->bdev,b));
6526 mddev_unlock(mddev);
6528 pr_debug("md: created %s\n", mdname(mddev));
6529 mddev->persistent = 1;
6530 rdev_for_each_list(rdev, tmp, &candidates) {
6531 list_del_init(&rdev->same_set);
6532 if (bind_rdev_to_array(rdev, mddev))
6535 autorun_array(mddev);
6536 mddev_unlock(mddev);
6538 /* on success, candidates will be empty, on error
6541 rdev_for_each_list(rdev, tmp, &candidates) {
6542 list_del_init(&rdev->same_set);
6547 pr_info("md: ... autorun DONE.\n");
6549 #endif /* !MODULE */
6551 static int get_version(void __user *arg)
6555 ver.major = MD_MAJOR_VERSION;
6556 ver.minor = MD_MINOR_VERSION;
6557 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6559 if (copy_to_user(arg, &ver, sizeof(ver)))
6565 static int get_array_info(struct mddev *mddev, void __user *arg)
6567 mdu_array_info_t info;
6568 int nr,working,insync,failed,spare;
6569 struct md_rdev *rdev;
6571 nr = working = insync = failed = spare = 0;
6573 rdev_for_each_rcu(rdev, mddev) {
6575 if (test_bit(Faulty, &rdev->flags))
6579 if (test_bit(In_sync, &rdev->flags))
6581 else if (test_bit(Journal, &rdev->flags))
6582 /* TODO: add journal count to md_u.h */
6590 info.major_version = mddev->major_version;
6591 info.minor_version = mddev->minor_version;
6592 info.patch_version = MD_PATCHLEVEL_VERSION;
6593 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6594 info.level = mddev->level;
6595 info.size = mddev->dev_sectors / 2;
6596 if (info.size != mddev->dev_sectors / 2) /* overflow */
6599 info.raid_disks = mddev->raid_disks;
6600 info.md_minor = mddev->md_minor;
6601 info.not_persistent= !mddev->persistent;
6603 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6606 info.state = (1<<MD_SB_CLEAN);
6607 if (mddev->bitmap && mddev->bitmap_info.offset)
6608 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6609 if (mddev_is_clustered(mddev))
6610 info.state |= (1<<MD_SB_CLUSTERED);
6611 info.active_disks = insync;
6612 info.working_disks = working;
6613 info.failed_disks = failed;
6614 info.spare_disks = spare;
6616 info.layout = mddev->layout;
6617 info.chunk_size = mddev->chunk_sectors << 9;
6619 if (copy_to_user(arg, &info, sizeof(info)))
6625 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6627 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6631 file = kzalloc(sizeof(*file), GFP_NOIO);
6636 spin_lock(&mddev->lock);
6637 /* bitmap enabled */
6638 if (mddev->bitmap_info.file) {
6639 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6640 sizeof(file->pathname));
6644 memmove(file->pathname, ptr,
6645 sizeof(file->pathname)-(ptr-file->pathname));
6647 spin_unlock(&mddev->lock);
6650 copy_to_user(arg, file, sizeof(*file)))
6657 static int get_disk_info(struct mddev *mddev, void __user * arg)
6659 mdu_disk_info_t info;
6660 struct md_rdev *rdev;
6662 if (copy_from_user(&info, arg, sizeof(info)))
6666 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6668 info.major = MAJOR(rdev->bdev->bd_dev);
6669 info.minor = MINOR(rdev->bdev->bd_dev);
6670 info.raid_disk = rdev->raid_disk;
6672 if (test_bit(Faulty, &rdev->flags))
6673 info.state |= (1<<MD_DISK_FAULTY);
6674 else if (test_bit(In_sync, &rdev->flags)) {
6675 info.state |= (1<<MD_DISK_ACTIVE);
6676 info.state |= (1<<MD_DISK_SYNC);
6678 if (test_bit(Journal, &rdev->flags))
6679 info.state |= (1<<MD_DISK_JOURNAL);
6680 if (test_bit(WriteMostly, &rdev->flags))
6681 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6682 if (test_bit(FailFast, &rdev->flags))
6683 info.state |= (1<<MD_DISK_FAILFAST);
6685 info.major = info.minor = 0;
6686 info.raid_disk = -1;
6687 info.state = (1<<MD_DISK_REMOVED);
6691 if (copy_to_user(arg, &info, sizeof(info)))
6697 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6699 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6700 struct md_rdev *rdev;
6701 dev_t dev = MKDEV(info->major,info->minor);
6703 if (mddev_is_clustered(mddev) &&
6704 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6705 pr_warn("%s: Cannot add to clustered mddev.\n",
6710 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6713 if (!mddev->raid_disks) {
6715 /* expecting a device which has a superblock */
6716 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6718 pr_warn("md: md_import_device returned %ld\n",
6720 return PTR_ERR(rdev);
6722 if (!list_empty(&mddev->disks)) {
6723 struct md_rdev *rdev0
6724 = list_entry(mddev->disks.next,
6725 struct md_rdev, same_set);
6726 err = super_types[mddev->major_version]
6727 .load_super(rdev, rdev0, mddev->minor_version);
6729 pr_warn("md: %s has different UUID to %s\n",
6730 bdevname(rdev->bdev,b),
6731 bdevname(rdev0->bdev,b2));
6736 err = bind_rdev_to_array(rdev, mddev);
6743 * md_add_new_disk can be used once the array is assembled
6744 * to add "hot spares". They must already have a superblock
6749 if (!mddev->pers->hot_add_disk) {
6750 pr_warn("%s: personality does not support diskops!\n",
6754 if (mddev->persistent)
6755 rdev = md_import_device(dev, mddev->major_version,
6756 mddev->minor_version);
6758 rdev = md_import_device(dev, -1, -1);
6760 pr_warn("md: md_import_device returned %ld\n",
6762 return PTR_ERR(rdev);
6764 /* set saved_raid_disk if appropriate */
6765 if (!mddev->persistent) {
6766 if (info->state & (1<<MD_DISK_SYNC) &&
6767 info->raid_disk < mddev->raid_disks) {
6768 rdev->raid_disk = info->raid_disk;
6769 set_bit(In_sync, &rdev->flags);
6770 clear_bit(Bitmap_sync, &rdev->flags);
6772 rdev->raid_disk = -1;
6773 rdev->saved_raid_disk = rdev->raid_disk;
6775 super_types[mddev->major_version].
6776 validate_super(mddev, rdev);
6777 if ((info->state & (1<<MD_DISK_SYNC)) &&
6778 rdev->raid_disk != info->raid_disk) {
6779 /* This was a hot-add request, but events doesn't
6780 * match, so reject it.
6786 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6787 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6788 set_bit(WriteMostly, &rdev->flags);
6790 clear_bit(WriteMostly, &rdev->flags);
6791 if (info->state & (1<<MD_DISK_FAILFAST))
6792 set_bit(FailFast, &rdev->flags);
6794 clear_bit(FailFast, &rdev->flags);
6796 if (info->state & (1<<MD_DISK_JOURNAL)) {
6797 struct md_rdev *rdev2;
6798 bool has_journal = false;
6800 /* make sure no existing journal disk */
6801 rdev_for_each(rdev2, mddev) {
6802 if (test_bit(Journal, &rdev2->flags)) {
6807 if (has_journal || mddev->bitmap) {
6811 set_bit(Journal, &rdev->flags);
6814 * check whether the device shows up in other nodes
6816 if (mddev_is_clustered(mddev)) {
6817 if (info->state & (1 << MD_DISK_CANDIDATE))
6818 set_bit(Candidate, &rdev->flags);
6819 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6820 /* --add initiated by this node */
6821 err = md_cluster_ops->add_new_disk(mddev, rdev);
6829 rdev->raid_disk = -1;
6830 err = bind_rdev_to_array(rdev, mddev);
6835 if (mddev_is_clustered(mddev)) {
6836 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6838 err = md_cluster_ops->new_disk_ack(mddev,
6841 md_kick_rdev_from_array(rdev);
6845 md_cluster_ops->add_new_disk_cancel(mddev);
6847 err = add_bound_rdev(rdev);
6851 err = add_bound_rdev(rdev);
6856 /* otherwise, md_add_new_disk is only allowed
6857 * for major_version==0 superblocks
6859 if (mddev->major_version != 0) {
6860 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6864 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6866 rdev = md_import_device(dev, -1, 0);
6868 pr_warn("md: error, md_import_device() returned %ld\n",
6870 return PTR_ERR(rdev);
6872 rdev->desc_nr = info->number;
6873 if (info->raid_disk < mddev->raid_disks)
6874 rdev->raid_disk = info->raid_disk;
6876 rdev->raid_disk = -1;
6878 if (rdev->raid_disk < mddev->raid_disks)
6879 if (info->state & (1<<MD_DISK_SYNC))
6880 set_bit(In_sync, &rdev->flags);
6882 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6883 set_bit(WriteMostly, &rdev->flags);
6884 if (info->state & (1<<MD_DISK_FAILFAST))
6885 set_bit(FailFast, &rdev->flags);
6887 if (!mddev->persistent) {
6888 pr_debug("md: nonpersistent superblock ...\n");
6889 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6891 rdev->sb_start = calc_dev_sboffset(rdev);
6892 rdev->sectors = rdev->sb_start;
6894 err = bind_rdev_to_array(rdev, mddev);
6904 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6906 char b[BDEVNAME_SIZE];
6907 struct md_rdev *rdev;
6912 rdev = find_rdev(mddev, dev);
6916 if (rdev->raid_disk < 0)
6919 clear_bit(Blocked, &rdev->flags);
6920 remove_and_add_spares(mddev, rdev);
6922 if (rdev->raid_disk >= 0)
6926 if (mddev_is_clustered(mddev)) {
6927 if (md_cluster_ops->remove_disk(mddev, rdev))
6931 md_kick_rdev_from_array(rdev);
6932 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6934 md_wakeup_thread(mddev->thread);
6936 md_update_sb(mddev, 1);
6937 md_new_event(mddev);
6941 pr_debug("md: cannot remove active disk %s from %s ...\n",
6942 bdevname(rdev->bdev,b), mdname(mddev));
6946 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6948 char b[BDEVNAME_SIZE];
6950 struct md_rdev *rdev;
6955 if (mddev->major_version != 0) {
6956 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6960 if (!mddev->pers->hot_add_disk) {
6961 pr_warn("%s: personality does not support diskops!\n",
6966 rdev = md_import_device(dev, -1, 0);
6968 pr_warn("md: error, md_import_device() returned %ld\n",
6973 if (mddev->persistent)
6974 rdev->sb_start = calc_dev_sboffset(rdev);
6976 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6978 rdev->sectors = rdev->sb_start;
6980 if (test_bit(Faulty, &rdev->flags)) {
6981 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6982 bdevname(rdev->bdev,b), mdname(mddev));
6987 clear_bit(In_sync, &rdev->flags);
6989 rdev->saved_raid_disk = -1;
6990 err = bind_rdev_to_array(rdev, mddev);
6995 * The rest should better be atomic, we can have disk failures
6996 * noticed in interrupt contexts ...
6999 rdev->raid_disk = -1;
7001 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7003 md_update_sb(mddev, 1);
7005 * Kick recovery, maybe this spare has to be added to the
7006 * array immediately.
7008 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7009 md_wakeup_thread(mddev->thread);
7010 md_new_event(mddev);
7018 static int set_bitmap_file(struct mddev *mddev, int fd)
7023 if (!mddev->pers->quiesce || !mddev->thread)
7025 if (mddev->recovery || mddev->sync_thread)
7027 /* we should be able to change the bitmap.. */
7031 struct inode *inode;
7034 if (mddev->bitmap || mddev->bitmap_info.file)
7035 return -EEXIST; /* cannot add when bitmap is present */
7039 pr_warn("%s: error: failed to get bitmap file\n",
7044 inode = f->f_mapping->host;
7045 if (!S_ISREG(inode->i_mode)) {
7046 pr_warn("%s: error: bitmap file must be a regular file\n",
7049 } else if (!(f->f_mode & FMODE_WRITE)) {
7050 pr_warn("%s: error: bitmap file must open for write\n",
7053 } else if (atomic_read(&inode->i_writecount) != 1) {
7054 pr_warn("%s: error: bitmap file is already in use\n",
7062 mddev->bitmap_info.file = f;
7063 mddev->bitmap_info.offset = 0; /* file overrides offset */
7064 } else if (mddev->bitmap == NULL)
7065 return -ENOENT; /* cannot remove what isn't there */
7069 struct bitmap *bitmap;
7071 bitmap = md_bitmap_create(mddev, -1);
7072 mddev_suspend(mddev);
7073 if (!IS_ERR(bitmap)) {
7074 mddev->bitmap = bitmap;
7075 err = md_bitmap_load(mddev);
7077 err = PTR_ERR(bitmap);
7079 md_bitmap_destroy(mddev);
7082 mddev_resume(mddev);
7083 } else if (fd < 0) {
7084 mddev_suspend(mddev);
7085 md_bitmap_destroy(mddev);
7086 mddev_resume(mddev);
7090 struct file *f = mddev->bitmap_info.file;
7092 spin_lock(&mddev->lock);
7093 mddev->bitmap_info.file = NULL;
7094 spin_unlock(&mddev->lock);
7103 * md_set_array_info is used two different ways
7104 * The original usage is when creating a new array.
7105 * In this usage, raid_disks is > 0 and it together with
7106 * level, size, not_persistent,layout,chunksize determine the
7107 * shape of the array.
7108 * This will always create an array with a type-0.90.0 superblock.
7109 * The newer usage is when assembling an array.
7110 * In this case raid_disks will be 0, and the major_version field is
7111 * use to determine which style super-blocks are to be found on the devices.
7112 * The minor and patch _version numbers are also kept incase the
7113 * super_block handler wishes to interpret them.
7115 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7117 if (info->raid_disks == 0) {
7118 /* just setting version number for superblock loading */
7119 if (info->major_version < 0 ||
7120 info->major_version >= ARRAY_SIZE(super_types) ||
7121 super_types[info->major_version].name == NULL) {
7122 /* maybe try to auto-load a module? */
7123 pr_warn("md: superblock version %d not known\n",
7124 info->major_version);
7127 mddev->major_version = info->major_version;
7128 mddev->minor_version = info->minor_version;
7129 mddev->patch_version = info->patch_version;
7130 mddev->persistent = !info->not_persistent;
7131 /* ensure mddev_put doesn't delete this now that there
7132 * is some minimal configuration.
7134 mddev->ctime = ktime_get_real_seconds();
7137 mddev->major_version = MD_MAJOR_VERSION;
7138 mddev->minor_version = MD_MINOR_VERSION;
7139 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7140 mddev->ctime = ktime_get_real_seconds();
7142 mddev->level = info->level;
7143 mddev->clevel[0] = 0;
7144 mddev->dev_sectors = 2 * (sector_t)info->size;
7145 mddev->raid_disks = info->raid_disks;
7146 /* don't set md_minor, it is determined by which /dev/md* was
7149 if (info->state & (1<<MD_SB_CLEAN))
7150 mddev->recovery_cp = MaxSector;
7152 mddev->recovery_cp = 0;
7153 mddev->persistent = ! info->not_persistent;
7154 mddev->external = 0;
7156 mddev->layout = info->layout;
7157 if (mddev->level == 0)
7158 /* Cannot trust RAID0 layout info here */
7160 mddev->chunk_sectors = info->chunk_size >> 9;
7162 if (mddev->persistent) {
7163 mddev->max_disks = MD_SB_DISKS;
7165 mddev->sb_flags = 0;
7167 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7169 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7170 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7171 mddev->bitmap_info.offset = 0;
7173 mddev->reshape_position = MaxSector;
7176 * Generate a 128 bit UUID
7178 get_random_bytes(mddev->uuid, 16);
7180 mddev->new_level = mddev->level;
7181 mddev->new_chunk_sectors = mddev->chunk_sectors;
7182 mddev->new_layout = mddev->layout;
7183 mddev->delta_disks = 0;
7184 mddev->reshape_backwards = 0;
7189 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7191 lockdep_assert_held(&mddev->reconfig_mutex);
7193 if (mddev->external_size)
7196 mddev->array_sectors = array_sectors;
7198 EXPORT_SYMBOL(md_set_array_sectors);
7200 static int update_size(struct mddev *mddev, sector_t num_sectors)
7202 struct md_rdev *rdev;
7204 int fit = (num_sectors == 0);
7205 sector_t old_dev_sectors = mddev->dev_sectors;
7207 if (mddev->pers->resize == NULL)
7209 /* The "num_sectors" is the number of sectors of each device that
7210 * is used. This can only make sense for arrays with redundancy.
7211 * linear and raid0 always use whatever space is available. We can only
7212 * consider changing this number if no resync or reconstruction is
7213 * happening, and if the new size is acceptable. It must fit before the
7214 * sb_start or, if that is <data_offset, it must fit before the size
7215 * of each device. If num_sectors is zero, we find the largest size
7218 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7224 rdev_for_each(rdev, mddev) {
7225 sector_t avail = rdev->sectors;
7227 if (fit && (num_sectors == 0 || num_sectors > avail))
7228 num_sectors = avail;
7229 if (avail < num_sectors)
7232 rv = mddev->pers->resize(mddev, num_sectors);
7234 if (mddev_is_clustered(mddev))
7235 md_cluster_ops->update_size(mddev, old_dev_sectors);
7236 else if (mddev->queue) {
7237 set_capacity_and_notify(mddev->gendisk,
7238 mddev->array_sectors);
7244 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7247 struct md_rdev *rdev;
7248 /* change the number of raid disks */
7249 if (mddev->pers->check_reshape == NULL)
7253 if (raid_disks <= 0 ||
7254 (mddev->max_disks && raid_disks >= mddev->max_disks))
7256 if (mddev->sync_thread ||
7257 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7258 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7259 mddev->reshape_position != MaxSector)
7262 rdev_for_each(rdev, mddev) {
7263 if (mddev->raid_disks < raid_disks &&
7264 rdev->data_offset < rdev->new_data_offset)
7266 if (mddev->raid_disks > raid_disks &&
7267 rdev->data_offset > rdev->new_data_offset)
7271 mddev->delta_disks = raid_disks - mddev->raid_disks;
7272 if (mddev->delta_disks < 0)
7273 mddev->reshape_backwards = 1;
7274 else if (mddev->delta_disks > 0)
7275 mddev->reshape_backwards = 0;
7277 rv = mddev->pers->check_reshape(mddev);
7279 mddev->delta_disks = 0;
7280 mddev->reshape_backwards = 0;
7286 * update_array_info is used to change the configuration of an
7288 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7289 * fields in the info are checked against the array.
7290 * Any differences that cannot be handled will cause an error.
7291 * Normally, only one change can be managed at a time.
7293 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7299 /* calculate expected state,ignoring low bits */
7300 if (mddev->bitmap && mddev->bitmap_info.offset)
7301 state |= (1 << MD_SB_BITMAP_PRESENT);
7303 if (mddev->major_version != info->major_version ||
7304 mddev->minor_version != info->minor_version ||
7305 /* mddev->patch_version != info->patch_version || */
7306 mddev->ctime != info->ctime ||
7307 mddev->level != info->level ||
7308 /* mddev->layout != info->layout || */
7309 mddev->persistent != !info->not_persistent ||
7310 mddev->chunk_sectors != info->chunk_size >> 9 ||
7311 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7312 ((state^info->state) & 0xfffffe00)
7315 /* Check there is only one change */
7316 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7318 if (mddev->raid_disks != info->raid_disks)
7320 if (mddev->layout != info->layout)
7322 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7329 if (mddev->layout != info->layout) {
7331 * we don't need to do anything at the md level, the
7332 * personality will take care of it all.
7334 if (mddev->pers->check_reshape == NULL)
7337 mddev->new_layout = info->layout;
7338 rv = mddev->pers->check_reshape(mddev);
7340 mddev->new_layout = mddev->layout;
7344 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7345 rv = update_size(mddev, (sector_t)info->size * 2);
7347 if (mddev->raid_disks != info->raid_disks)
7348 rv = update_raid_disks(mddev, info->raid_disks);
7350 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7351 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7355 if (mddev->recovery || mddev->sync_thread) {
7359 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7360 struct bitmap *bitmap;
7361 /* add the bitmap */
7362 if (mddev->bitmap) {
7366 if (mddev->bitmap_info.default_offset == 0) {
7370 mddev->bitmap_info.offset =
7371 mddev->bitmap_info.default_offset;
7372 mddev->bitmap_info.space =
7373 mddev->bitmap_info.default_space;
7374 bitmap = md_bitmap_create(mddev, -1);
7375 mddev_suspend(mddev);
7376 if (!IS_ERR(bitmap)) {
7377 mddev->bitmap = bitmap;
7378 rv = md_bitmap_load(mddev);
7380 rv = PTR_ERR(bitmap);
7382 md_bitmap_destroy(mddev);
7383 mddev_resume(mddev);
7385 /* remove the bitmap */
7386 if (!mddev->bitmap) {
7390 if (mddev->bitmap->storage.file) {
7394 if (mddev->bitmap_info.nodes) {
7395 /* hold PW on all the bitmap lock */
7396 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7397 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7399 md_cluster_ops->unlock_all_bitmaps(mddev);
7403 mddev->bitmap_info.nodes = 0;
7404 md_cluster_ops->leave(mddev);
7405 module_put(md_cluster_mod);
7406 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7408 mddev_suspend(mddev);
7409 md_bitmap_destroy(mddev);
7410 mddev_resume(mddev);
7411 mddev->bitmap_info.offset = 0;
7414 md_update_sb(mddev, 1);
7420 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7422 struct md_rdev *rdev;
7425 if (mddev->pers == NULL)
7429 rdev = md_find_rdev_rcu(mddev, dev);
7433 md_error(mddev, rdev);
7434 if (!test_bit(Faulty, &rdev->flags))
7442 * We have a problem here : there is no easy way to give a CHS
7443 * virtual geometry. We currently pretend that we have a 2 heads
7444 * 4 sectors (with a BIG number of cylinders...). This drives
7445 * dosfs just mad... ;-)
7447 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7449 struct mddev *mddev = bdev->bd_disk->private_data;
7453 geo->cylinders = mddev->array_sectors / 8;
7457 static inline bool md_ioctl_valid(unsigned int cmd)
7461 case GET_ARRAY_INFO:
7462 case GET_BITMAP_FILE:
7465 case HOT_REMOVE_DISK:
7467 case RESTART_ARRAY_RW:
7469 case SET_ARRAY_INFO:
7470 case SET_BITMAP_FILE:
7471 case SET_DISK_FAULTY:
7474 case CLUSTERED_DISK_NACK:
7481 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7482 unsigned int cmd, unsigned long arg)
7485 void __user *argp = (void __user *)arg;
7486 struct mddev *mddev = NULL;
7487 bool did_set_md_closing = false;
7489 if (!md_ioctl_valid(cmd))
7494 case GET_ARRAY_INFO:
7498 if (!capable(CAP_SYS_ADMIN))
7503 * Commands dealing with the RAID driver but not any
7508 err = get_version(argp);
7514 * Commands creating/starting a new array:
7517 mddev = bdev->bd_disk->private_data;
7524 /* Some actions do not requires the mutex */
7526 case GET_ARRAY_INFO:
7527 if (!mddev->raid_disks && !mddev->external)
7530 err = get_array_info(mddev, argp);
7534 if (!mddev->raid_disks && !mddev->external)
7537 err = get_disk_info(mddev, argp);
7540 case SET_DISK_FAULTY:
7541 err = set_disk_faulty(mddev, new_decode_dev(arg));
7544 case GET_BITMAP_FILE:
7545 err = get_bitmap_file(mddev, argp);
7550 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7551 flush_rdev_wq(mddev);
7553 if (cmd == HOT_REMOVE_DISK)
7554 /* need to ensure recovery thread has run */
7555 wait_event_interruptible_timeout(mddev->sb_wait,
7556 !test_bit(MD_RECOVERY_NEEDED,
7558 msecs_to_jiffies(5000));
7559 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7560 /* Need to flush page cache, and ensure no-one else opens
7563 mutex_lock(&mddev->open_mutex);
7564 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7565 mutex_unlock(&mddev->open_mutex);
7569 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7570 mutex_unlock(&mddev->open_mutex);
7574 did_set_md_closing = true;
7575 mutex_unlock(&mddev->open_mutex);
7576 sync_blockdev(bdev);
7578 err = mddev_lock(mddev);
7580 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7585 if (cmd == SET_ARRAY_INFO) {
7586 mdu_array_info_t info;
7588 memset(&info, 0, sizeof(info));
7589 else if (copy_from_user(&info, argp, sizeof(info))) {
7594 err = update_array_info(mddev, &info);
7596 pr_warn("md: couldn't update array info. %d\n", err);
7601 if (!list_empty(&mddev->disks)) {
7602 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7606 if (mddev->raid_disks) {
7607 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7611 err = md_set_array_info(mddev, &info);
7613 pr_warn("md: couldn't set array info. %d\n", err);
7620 * Commands querying/configuring an existing array:
7622 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7623 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7624 if ((!mddev->raid_disks && !mddev->external)
7625 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7626 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7627 && cmd != GET_BITMAP_FILE) {
7633 * Commands even a read-only array can execute:
7636 case RESTART_ARRAY_RW:
7637 err = restart_array(mddev);
7641 err = do_md_stop(mddev, 0, bdev);
7645 err = md_set_readonly(mddev, bdev);
7648 case HOT_REMOVE_DISK:
7649 err = hot_remove_disk(mddev, new_decode_dev(arg));
7653 /* We can support ADD_NEW_DISK on read-only arrays
7654 * only if we are re-adding a preexisting device.
7655 * So require mddev->pers and MD_DISK_SYNC.
7658 mdu_disk_info_t info;
7659 if (copy_from_user(&info, argp, sizeof(info)))
7661 else if (!(info.state & (1<<MD_DISK_SYNC)))
7662 /* Need to clear read-only for this */
7665 err = md_add_new_disk(mddev, &info);
7672 * The remaining ioctls are changing the state of the
7673 * superblock, so we do not allow them on read-only arrays.
7675 if (mddev->ro && mddev->pers) {
7676 if (mddev->ro == 2) {
7678 sysfs_notify_dirent_safe(mddev->sysfs_state);
7679 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7680 /* mddev_unlock will wake thread */
7681 /* If a device failed while we were read-only, we
7682 * need to make sure the metadata is updated now.
7684 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7685 mddev_unlock(mddev);
7686 wait_event(mddev->sb_wait,
7687 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7688 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7689 mddev_lock_nointr(mddev);
7700 mdu_disk_info_t info;
7701 if (copy_from_user(&info, argp, sizeof(info)))
7704 err = md_add_new_disk(mddev, &info);
7708 case CLUSTERED_DISK_NACK:
7709 if (mddev_is_clustered(mddev))
7710 md_cluster_ops->new_disk_ack(mddev, false);
7716 err = hot_add_disk(mddev, new_decode_dev(arg));
7720 err = do_md_run(mddev);
7723 case SET_BITMAP_FILE:
7724 err = set_bitmap_file(mddev, (int)arg);
7733 if (mddev->hold_active == UNTIL_IOCTL &&
7735 mddev->hold_active = 0;
7736 mddev_unlock(mddev);
7738 if(did_set_md_closing)
7739 clear_bit(MD_CLOSING, &mddev->flags);
7742 #ifdef CONFIG_COMPAT
7743 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7744 unsigned int cmd, unsigned long arg)
7747 case HOT_REMOVE_DISK:
7749 case SET_DISK_FAULTY:
7750 case SET_BITMAP_FILE:
7751 /* These take in integer arg, do not convert */
7754 arg = (unsigned long)compat_ptr(arg);
7758 return md_ioctl(bdev, mode, cmd, arg);
7760 #endif /* CONFIG_COMPAT */
7762 static int md_set_read_only(struct block_device *bdev, bool ro)
7764 struct mddev *mddev = bdev->bd_disk->private_data;
7767 err = mddev_lock(mddev);
7771 if (!mddev->raid_disks && !mddev->external) {
7777 * Transitioning to read-auto need only happen for arrays that call
7778 * md_write_start and which are not ready for writes yet.
7780 if (!ro && mddev->ro == 1 && mddev->pers) {
7781 err = restart_array(mddev);
7788 mddev_unlock(mddev);
7792 static int md_open(struct block_device *bdev, fmode_t mode)
7795 * Succeed if we can lock the mddev, which confirms that
7796 * it isn't being stopped right now.
7798 struct mddev *mddev = mddev_find(bdev->bd_dev);
7804 if (mddev->gendisk != bdev->bd_disk) {
7805 /* we are racing with mddev_put which is discarding this
7809 /* Wait until bdev->bd_disk is definitely gone */
7810 if (work_pending(&mddev->del_work))
7811 flush_workqueue(md_misc_wq);
7814 BUG_ON(mddev != bdev->bd_disk->private_data);
7816 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7819 if (test_bit(MD_CLOSING, &mddev->flags)) {
7820 mutex_unlock(&mddev->open_mutex);
7826 atomic_inc(&mddev->openers);
7827 mutex_unlock(&mddev->open_mutex);
7829 bdev_check_media_change(bdev);
7836 static void md_release(struct gendisk *disk, fmode_t mode)
7838 struct mddev *mddev = disk->private_data;
7841 atomic_dec(&mddev->openers);
7845 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7847 struct mddev *mddev = disk->private_data;
7848 unsigned int ret = 0;
7851 ret = DISK_EVENT_MEDIA_CHANGE;
7856 const struct block_device_operations md_fops =
7858 .owner = THIS_MODULE,
7859 .submit_bio = md_submit_bio,
7861 .release = md_release,
7863 #ifdef CONFIG_COMPAT
7864 .compat_ioctl = md_compat_ioctl,
7866 .getgeo = md_getgeo,
7867 .check_events = md_check_events,
7868 .set_read_only = md_set_read_only,
7871 static int md_thread(void *arg)
7873 struct md_thread *thread = arg;
7876 * md_thread is a 'system-thread', it's priority should be very
7877 * high. We avoid resource deadlocks individually in each
7878 * raid personality. (RAID5 does preallocation) We also use RR and
7879 * the very same RT priority as kswapd, thus we will never get
7880 * into a priority inversion deadlock.
7882 * we definitely have to have equal or higher priority than
7883 * bdflush, otherwise bdflush will deadlock if there are too
7884 * many dirty RAID5 blocks.
7887 allow_signal(SIGKILL);
7888 while (!kthread_should_stop()) {
7890 /* We need to wait INTERRUPTIBLE so that
7891 * we don't add to the load-average.
7892 * That means we need to be sure no signals are
7895 if (signal_pending(current))
7896 flush_signals(current);
7898 wait_event_interruptible_timeout
7900 test_bit(THREAD_WAKEUP, &thread->flags)
7901 || kthread_should_stop() || kthread_should_park(),
7904 clear_bit(THREAD_WAKEUP, &thread->flags);
7905 if (kthread_should_park())
7907 if (!kthread_should_stop())
7908 thread->run(thread);
7914 void md_wakeup_thread(struct md_thread *thread)
7917 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7918 set_bit(THREAD_WAKEUP, &thread->flags);
7919 wake_up(&thread->wqueue);
7922 EXPORT_SYMBOL(md_wakeup_thread);
7924 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7925 struct mddev *mddev, const char *name)
7927 struct md_thread *thread;
7929 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7933 init_waitqueue_head(&thread->wqueue);
7936 thread->mddev = mddev;
7937 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7938 thread->tsk = kthread_run(md_thread, thread,
7940 mdname(thread->mddev),
7942 if (IS_ERR(thread->tsk)) {
7948 EXPORT_SYMBOL(md_register_thread);
7950 void md_unregister_thread(struct md_thread **threadp)
7952 struct md_thread *thread;
7955 * Locking ensures that mddev_unlock does not wake_up a
7956 * non-existent thread
7958 spin_lock(&pers_lock);
7961 spin_unlock(&pers_lock);
7965 spin_unlock(&pers_lock);
7967 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7968 kthread_stop(thread->tsk);
7971 EXPORT_SYMBOL(md_unregister_thread);
7973 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7975 if (!rdev || test_bit(Faulty, &rdev->flags))
7978 if (!mddev->pers || !mddev->pers->error_handler)
7980 mddev->pers->error_handler(mddev,rdev);
7981 if (mddev->degraded)
7982 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7983 sysfs_notify_dirent_safe(rdev->sysfs_state);
7984 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7985 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7986 md_wakeup_thread(mddev->thread);
7987 if (mddev->event_work.func)
7988 queue_work(md_misc_wq, &mddev->event_work);
7989 md_new_event(mddev);
7991 EXPORT_SYMBOL(md_error);
7993 /* seq_file implementation /proc/mdstat */
7995 static void status_unused(struct seq_file *seq)
7998 struct md_rdev *rdev;
8000 seq_printf(seq, "unused devices: ");
8002 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8003 char b[BDEVNAME_SIZE];
8005 seq_printf(seq, "%s ",
8006 bdevname(rdev->bdev,b));
8009 seq_printf(seq, "<none>");
8011 seq_printf(seq, "\n");
8014 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8016 sector_t max_sectors, resync, res;
8017 unsigned long dt, db = 0;
8018 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8019 int scale, recovery_active;
8020 unsigned int per_milli;
8022 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8023 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8024 max_sectors = mddev->resync_max_sectors;
8026 max_sectors = mddev->dev_sectors;
8028 resync = mddev->curr_resync;
8030 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8031 /* Still cleaning up */
8032 resync = max_sectors;
8033 } else if (resync > max_sectors)
8034 resync = max_sectors;
8036 resync -= atomic_read(&mddev->recovery_active);
8039 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8040 struct md_rdev *rdev;
8042 rdev_for_each(rdev, mddev)
8043 if (rdev->raid_disk >= 0 &&
8044 !test_bit(Faulty, &rdev->flags) &&
8045 rdev->recovery_offset != MaxSector &&
8046 rdev->recovery_offset) {
8047 seq_printf(seq, "\trecover=REMOTE");
8050 if (mddev->reshape_position != MaxSector)
8051 seq_printf(seq, "\treshape=REMOTE");
8053 seq_printf(seq, "\tresync=REMOTE");
8056 if (mddev->recovery_cp < MaxSector) {
8057 seq_printf(seq, "\tresync=PENDING");
8063 seq_printf(seq, "\tresync=DELAYED");
8067 WARN_ON(max_sectors == 0);
8068 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8069 * in a sector_t, and (max_sectors>>scale) will fit in a
8070 * u32, as those are the requirements for sector_div.
8071 * Thus 'scale' must be at least 10
8074 if (sizeof(sector_t) > sizeof(unsigned long)) {
8075 while ( max_sectors/2 > (1ULL<<(scale+32)))
8078 res = (resync>>scale)*1000;
8079 sector_div(res, (u32)((max_sectors>>scale)+1));
8083 int i, x = per_milli/50, y = 20-x;
8084 seq_printf(seq, "[");
8085 for (i = 0; i < x; i++)
8086 seq_printf(seq, "=");
8087 seq_printf(seq, ">");
8088 for (i = 0; i < y; i++)
8089 seq_printf(seq, ".");
8090 seq_printf(seq, "] ");
8092 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8093 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8095 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8097 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8098 "resync" : "recovery"))),
8099 per_milli/10, per_milli % 10,
8100 (unsigned long long) resync/2,
8101 (unsigned long long) max_sectors/2);
8104 * dt: time from mark until now
8105 * db: blocks written from mark until now
8106 * rt: remaining time
8108 * rt is a sector_t, which is always 64bit now. We are keeping
8109 * the original algorithm, but it is not really necessary.
8111 * Original algorithm:
8112 * So we divide before multiply in case it is 32bit and close
8114 * We scale the divisor (db) by 32 to avoid losing precision
8115 * near the end of resync when the number of remaining sectors
8117 * We then divide rt by 32 after multiplying by db to compensate.
8118 * The '+1' avoids division by zero if db is very small.
8120 dt = ((jiffies - mddev->resync_mark) / HZ);
8123 curr_mark_cnt = mddev->curr_mark_cnt;
8124 recovery_active = atomic_read(&mddev->recovery_active);
8125 resync_mark_cnt = mddev->resync_mark_cnt;
8127 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8128 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8130 rt = max_sectors - resync; /* number of remaining sectors */
8131 rt = div64_u64(rt, db/32+1);
8135 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8136 ((unsigned long)rt % 60)/6);
8138 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8142 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8144 struct list_head *tmp;
8146 struct mddev *mddev;
8158 spin_lock(&all_mddevs_lock);
8159 list_for_each(tmp,&all_mddevs)
8161 mddev = list_entry(tmp, struct mddev, all_mddevs);
8163 spin_unlock(&all_mddevs_lock);
8166 spin_unlock(&all_mddevs_lock);
8168 return (void*)2;/* tail */
8172 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8174 struct list_head *tmp;
8175 struct mddev *next_mddev, *mddev = v;
8181 spin_lock(&all_mddevs_lock);
8183 tmp = all_mddevs.next;
8185 tmp = mddev->all_mddevs.next;
8186 if (tmp != &all_mddevs)
8187 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8189 next_mddev = (void*)2;
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");
8237 seq_printf(seq, " (read-only)");
8239 seq_printf(seq, " (auto-read-only)");
8240 seq_printf(seq, " %s", mddev->pers->name);
8245 rdev_for_each_rcu(rdev, mddev) {
8246 char b[BDEVNAME_SIZE];
8247 seq_printf(seq, " %s[%d]",
8248 bdevname(rdev->bdev,b), rdev->desc_nr);
8249 if (test_bit(WriteMostly, &rdev->flags))
8250 seq_printf(seq, "(W)");
8251 if (test_bit(Journal, &rdev->flags))
8252 seq_printf(seq, "(J)");
8253 if (test_bit(Faulty, &rdev->flags)) {
8254 seq_printf(seq, "(F)");
8257 if (rdev->raid_disk < 0)
8258 seq_printf(seq, "(S)"); /* spare */
8259 if (test_bit(Replacement, &rdev->flags))
8260 seq_printf(seq, "(R)");
8261 sectors += rdev->sectors;
8265 if (!list_empty(&mddev->disks)) {
8267 seq_printf(seq, "\n %llu blocks",
8268 (unsigned long long)
8269 mddev->array_sectors / 2);
8271 seq_printf(seq, "\n %llu blocks",
8272 (unsigned long long)sectors / 2);
8274 if (mddev->persistent) {
8275 if (mddev->major_version != 0 ||
8276 mddev->minor_version != 90) {
8277 seq_printf(seq," super %d.%d",
8278 mddev->major_version,
8279 mddev->minor_version);
8281 } else if (mddev->external)
8282 seq_printf(seq, " super external:%s",
8283 mddev->metadata_type);
8285 seq_printf(seq, " super non-persistent");
8288 mddev->pers->status(seq, mddev);
8289 seq_printf(seq, "\n ");
8290 if (mddev->pers->sync_request) {
8291 if (status_resync(seq, mddev))
8292 seq_printf(seq, "\n ");
8295 seq_printf(seq, "\n ");
8297 md_bitmap_status(seq, mddev->bitmap);
8299 seq_printf(seq, "\n");
8301 spin_unlock(&mddev->lock);
8306 static const struct seq_operations md_seq_ops = {
8307 .start = md_seq_start,
8308 .next = md_seq_next,
8309 .stop = md_seq_stop,
8310 .show = md_seq_show,
8313 static int md_seq_open(struct inode *inode, struct file *file)
8315 struct seq_file *seq;
8318 error = seq_open(file, &md_seq_ops);
8322 seq = file->private_data;
8323 seq->poll_event = atomic_read(&md_event_count);
8327 static int md_unloading;
8328 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8330 struct seq_file *seq = filp->private_data;
8334 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8335 poll_wait(filp, &md_event_waiters, wait);
8337 /* always allow read */
8338 mask = EPOLLIN | EPOLLRDNORM;
8340 if (seq->poll_event != atomic_read(&md_event_count))
8341 mask |= EPOLLERR | EPOLLPRI;
8345 static const struct proc_ops mdstat_proc_ops = {
8346 .proc_open = md_seq_open,
8347 .proc_read = seq_read,
8348 .proc_lseek = seq_lseek,
8349 .proc_release = seq_release,
8350 .proc_poll = mdstat_poll,
8353 int register_md_personality(struct md_personality *p)
8355 pr_debug("md: %s personality registered for level %d\n",
8357 spin_lock(&pers_lock);
8358 list_add_tail(&p->list, &pers_list);
8359 spin_unlock(&pers_lock);
8362 EXPORT_SYMBOL(register_md_personality);
8364 int unregister_md_personality(struct md_personality *p)
8366 pr_debug("md: %s personality unregistered\n", p->name);
8367 spin_lock(&pers_lock);
8368 list_del_init(&p->list);
8369 spin_unlock(&pers_lock);
8372 EXPORT_SYMBOL(unregister_md_personality);
8374 int register_md_cluster_operations(struct md_cluster_operations *ops,
8375 struct module *module)
8378 spin_lock(&pers_lock);
8379 if (md_cluster_ops != NULL)
8382 md_cluster_ops = ops;
8383 md_cluster_mod = module;
8385 spin_unlock(&pers_lock);
8388 EXPORT_SYMBOL(register_md_cluster_operations);
8390 int unregister_md_cluster_operations(void)
8392 spin_lock(&pers_lock);
8393 md_cluster_ops = NULL;
8394 spin_unlock(&pers_lock);
8397 EXPORT_SYMBOL(unregister_md_cluster_operations);
8399 int md_setup_cluster(struct mddev *mddev, int nodes)
8402 if (!md_cluster_ops)
8403 request_module("md-cluster");
8404 spin_lock(&pers_lock);
8405 /* ensure module won't be unloaded */
8406 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8407 pr_warn("can't find md-cluster module or get it's reference.\n");
8408 spin_unlock(&pers_lock);
8411 spin_unlock(&pers_lock);
8413 ret = md_cluster_ops->join(mddev, nodes);
8415 mddev->safemode_delay = 0;
8419 void md_cluster_stop(struct mddev *mddev)
8421 if (!md_cluster_ops)
8423 md_cluster_ops->leave(mddev);
8424 module_put(md_cluster_mod);
8427 static int is_mddev_idle(struct mddev *mddev, int init)
8429 struct md_rdev *rdev;
8435 rdev_for_each_rcu(rdev, mddev) {
8436 struct gendisk *disk = rdev->bdev->bd_disk;
8437 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8438 atomic_read(&disk->sync_io);
8439 /* sync IO will cause sync_io to increase before the disk_stats
8440 * as sync_io is counted when a request starts, and
8441 * disk_stats is counted when it completes.
8442 * So resync activity will cause curr_events to be smaller than
8443 * when there was no such activity.
8444 * non-sync IO will cause disk_stat to increase without
8445 * increasing sync_io so curr_events will (eventually)
8446 * be larger than it was before. Once it becomes
8447 * substantially larger, the test below will cause
8448 * the array to appear non-idle, and resync will slow
8450 * If there is a lot of outstanding resync activity when
8451 * we set last_event to curr_events, then all that activity
8452 * completing might cause the array to appear non-idle
8453 * and resync will be slowed down even though there might
8454 * not have been non-resync activity. This will only
8455 * happen once though. 'last_events' will soon reflect
8456 * the state where there is little or no outstanding
8457 * resync requests, and further resync activity will
8458 * always make curr_events less than last_events.
8461 if (init || curr_events - rdev->last_events > 64) {
8462 rdev->last_events = curr_events;
8470 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8472 /* another "blocks" (512byte) blocks have been synced */
8473 atomic_sub(blocks, &mddev->recovery_active);
8474 wake_up(&mddev->recovery_wait);
8476 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8477 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8478 md_wakeup_thread(mddev->thread);
8479 // stop recovery, signal do_sync ....
8482 EXPORT_SYMBOL(md_done_sync);
8484 /* md_write_start(mddev, bi)
8485 * If we need to update some array metadata (e.g. 'active' flag
8486 * in superblock) before writing, schedule a superblock update
8487 * and wait for it to complete.
8488 * A return value of 'false' means that the write wasn't recorded
8489 * and cannot proceed as the array is being suspend.
8491 bool md_write_start(struct mddev *mddev, struct bio *bi)
8495 if (bio_data_dir(bi) != WRITE)
8498 BUG_ON(mddev->ro == 1);
8499 if (mddev->ro == 2) {
8500 /* need to switch to read/write */
8502 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8503 md_wakeup_thread(mddev->thread);
8504 md_wakeup_thread(mddev->sync_thread);
8508 percpu_ref_get(&mddev->writes_pending);
8509 smp_mb(); /* Match smp_mb in set_in_sync() */
8510 if (mddev->safemode == 1)
8511 mddev->safemode = 0;
8512 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8513 if (mddev->in_sync || mddev->sync_checkers) {
8514 spin_lock(&mddev->lock);
8515 if (mddev->in_sync) {
8517 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8518 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8519 md_wakeup_thread(mddev->thread);
8522 spin_unlock(&mddev->lock);
8526 sysfs_notify_dirent_safe(mddev->sysfs_state);
8527 if (!mddev->has_superblocks)
8529 wait_event(mddev->sb_wait,
8530 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8532 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8533 percpu_ref_put(&mddev->writes_pending);
8538 EXPORT_SYMBOL(md_write_start);
8540 /* md_write_inc can only be called when md_write_start() has
8541 * already been called at least once of the current request.
8542 * It increments the counter and is useful when a single request
8543 * is split into several parts. Each part causes an increment and
8544 * so needs a matching md_write_end().
8545 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8546 * a spinlocked region.
8548 void md_write_inc(struct mddev *mddev, struct bio *bi)
8550 if (bio_data_dir(bi) != WRITE)
8552 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8553 percpu_ref_get(&mddev->writes_pending);
8555 EXPORT_SYMBOL(md_write_inc);
8557 void md_write_end(struct mddev *mddev)
8559 percpu_ref_put(&mddev->writes_pending);
8561 if (mddev->safemode == 2)
8562 md_wakeup_thread(mddev->thread);
8563 else if (mddev->safemode_delay)
8564 /* The roundup() ensures this only performs locking once
8565 * every ->safemode_delay jiffies
8567 mod_timer(&mddev->safemode_timer,
8568 roundup(jiffies, mddev->safemode_delay) +
8569 mddev->safemode_delay);
8572 EXPORT_SYMBOL(md_write_end);
8574 /* This is used by raid0 and raid10 */
8575 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8576 struct bio *bio, sector_t start, sector_t size)
8578 struct bio *discard_bio = NULL;
8580 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO, 0,
8581 &discard_bio) || !discard_bio)
8584 bio_chain(discard_bio, bio);
8585 bio_clone_blkg_association(discard_bio, bio);
8587 trace_block_bio_remap(discard_bio,
8588 disk_devt(mddev->gendisk),
8589 bio->bi_iter.bi_sector);
8590 submit_bio_noacct(discard_bio);
8592 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8594 int acct_bioset_init(struct mddev *mddev)
8598 if (!bioset_initialized(&mddev->io_acct_set))
8599 err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
8600 offsetof(struct md_io_acct, bio_clone), 0);
8603 EXPORT_SYMBOL_GPL(acct_bioset_init);
8605 void acct_bioset_exit(struct mddev *mddev)
8607 bioset_exit(&mddev->io_acct_set);
8609 EXPORT_SYMBOL_GPL(acct_bioset_exit);
8611 static void md_end_io_acct(struct bio *bio)
8613 struct md_io_acct *md_io_acct = bio->bi_private;
8614 struct bio *orig_bio = md_io_acct->orig_bio;
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);
8624 * Used by personalities that don't already clone the bio and thus can't
8625 * easily add the timestamp to their extended bio structure.
8627 void md_account_bio(struct mddev *mddev, struct bio **bio)
8629 struct md_io_acct *md_io_acct;
8632 if (!blk_queue_io_stat((*bio)->bi_bdev->bd_disk->queue))
8635 clone = bio_clone_fast(*bio, GFP_NOIO, &mddev->io_acct_set);
8636 md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8637 md_io_acct->orig_bio = *bio;
8638 md_io_acct->start_time = bio_start_io_acct(*bio);
8640 clone->bi_end_io = md_end_io_acct;
8641 clone->bi_private = md_io_acct;
8644 EXPORT_SYMBOL_GPL(md_account_bio);
8646 /* md_allow_write(mddev)
8647 * Calling this ensures that the array is marked 'active' so that writes
8648 * may proceed without blocking. It is important to call this before
8649 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8650 * Must be called with mddev_lock held.
8652 void md_allow_write(struct mddev *mddev)
8658 if (!mddev->pers->sync_request)
8661 spin_lock(&mddev->lock);
8662 if (mddev->in_sync) {
8664 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8665 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8666 if (mddev->safemode_delay &&
8667 mddev->safemode == 0)
8668 mddev->safemode = 1;
8669 spin_unlock(&mddev->lock);
8670 md_update_sb(mddev, 0);
8671 sysfs_notify_dirent_safe(mddev->sysfs_state);
8672 /* wait for the dirty state to be recorded in the metadata */
8673 wait_event(mddev->sb_wait,
8674 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8676 spin_unlock(&mddev->lock);
8678 EXPORT_SYMBOL_GPL(md_allow_write);
8680 #define SYNC_MARKS 10
8681 #define SYNC_MARK_STEP (3*HZ)
8682 #define UPDATE_FREQUENCY (5*60*HZ)
8683 void md_do_sync(struct md_thread *thread)
8685 struct mddev *mddev = thread->mddev;
8686 struct mddev *mddev2;
8687 unsigned int currspeed = 0, window;
8688 sector_t max_sectors,j, io_sectors, recovery_done;
8689 unsigned long mark[SYNC_MARKS];
8690 unsigned long update_time;
8691 sector_t mark_cnt[SYNC_MARKS];
8693 struct list_head *tmp;
8694 sector_t last_check;
8696 struct md_rdev *rdev;
8697 char *desc, *action = NULL;
8698 struct blk_plug plug;
8701 /* just incase thread restarts... */
8702 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8703 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8705 if (mddev->ro) {/* never try to sync a read-only array */
8706 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8710 if (mddev_is_clustered(mddev)) {
8711 ret = md_cluster_ops->resync_start(mddev);
8715 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8716 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8717 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8718 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8719 && ((unsigned long long)mddev->curr_resync_completed
8720 < (unsigned long long)mddev->resync_max_sectors))
8724 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8725 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8726 desc = "data-check";
8728 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8729 desc = "requested-resync";
8733 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8738 mddev->last_sync_action = action ?: desc;
8740 /* we overload curr_resync somewhat here.
8741 * 0 == not engaged in resync at all
8742 * 2 == checking that there is no conflict with another sync
8743 * 1 == like 2, but have yielded to allow conflicting resync to
8745 * other == active in resync - this many blocks
8747 * Before starting a resync we must have set curr_resync to
8748 * 2, and then checked that every "conflicting" array has curr_resync
8749 * less than ours. When we find one that is the same or higher
8750 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8751 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8752 * This will mean we have to start checking from the beginning again.
8757 int mddev2_minor = -1;
8758 mddev->curr_resync = 2;
8761 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8763 for_each_mddev(mddev2, tmp) {
8764 if (mddev2 == mddev)
8766 if (!mddev->parallel_resync
8767 && mddev2->curr_resync
8768 && match_mddev_units(mddev, mddev2)) {
8770 if (mddev < mddev2 && mddev->curr_resync == 2) {
8771 /* arbitrarily yield */
8772 mddev->curr_resync = 1;
8773 wake_up(&resync_wait);
8775 if (mddev > mddev2 && mddev->curr_resync == 1)
8776 /* no need to wait here, we can wait the next
8777 * time 'round when curr_resync == 2
8780 /* We need to wait 'interruptible' so as not to
8781 * contribute to the load average, and not to
8782 * be caught by 'softlockup'
8784 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8785 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8786 mddev2->curr_resync >= mddev->curr_resync) {
8787 if (mddev2_minor != mddev2->md_minor) {
8788 mddev2_minor = mddev2->md_minor;
8789 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8790 desc, mdname(mddev),
8794 if (signal_pending(current))
8795 flush_signals(current);
8797 finish_wait(&resync_wait, &wq);
8800 finish_wait(&resync_wait, &wq);
8803 } while (mddev->curr_resync < 2);
8806 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8807 /* resync follows the size requested by the personality,
8808 * which defaults to physical size, but can be virtual size
8810 max_sectors = mddev->resync_max_sectors;
8811 atomic64_set(&mddev->resync_mismatches, 0);
8812 /* we don't use the checkpoint if there's a bitmap */
8813 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8814 j = mddev->resync_min;
8815 else if (!mddev->bitmap)
8816 j = mddev->recovery_cp;
8818 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8819 max_sectors = mddev->resync_max_sectors;
8821 * If the original node aborts reshaping then we continue the
8822 * reshaping, so set j again to avoid restart reshape from the
8825 if (mddev_is_clustered(mddev) &&
8826 mddev->reshape_position != MaxSector)
8827 j = mddev->reshape_position;
8829 /* recovery follows the physical size of devices */
8830 max_sectors = mddev->dev_sectors;
8833 rdev_for_each_rcu(rdev, mddev)
8834 if (rdev->raid_disk >= 0 &&
8835 !test_bit(Journal, &rdev->flags) &&
8836 !test_bit(Faulty, &rdev->flags) &&
8837 !test_bit(In_sync, &rdev->flags) &&
8838 rdev->recovery_offset < j)
8839 j = rdev->recovery_offset;
8842 /* If there is a bitmap, we need to make sure all
8843 * writes that started before we added a spare
8844 * complete before we start doing a recovery.
8845 * Otherwise the write might complete and (via
8846 * bitmap_endwrite) set a bit in the bitmap after the
8847 * recovery has checked that bit and skipped that
8850 if (mddev->bitmap) {
8851 mddev->pers->quiesce(mddev, 1);
8852 mddev->pers->quiesce(mddev, 0);
8856 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8857 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8858 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8859 speed_max(mddev), desc);
8861 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8864 for (m = 0; m < SYNC_MARKS; m++) {
8866 mark_cnt[m] = io_sectors;
8869 mddev->resync_mark = mark[last_mark];
8870 mddev->resync_mark_cnt = mark_cnt[last_mark];
8873 * Tune reconstruction:
8875 window = 32 * (PAGE_SIZE / 512);
8876 pr_debug("md: using %dk window, over a total of %lluk.\n",
8877 window/2, (unsigned long long)max_sectors/2);
8879 atomic_set(&mddev->recovery_active, 0);
8883 pr_debug("md: resuming %s of %s from checkpoint.\n",
8884 desc, mdname(mddev));
8885 mddev->curr_resync = j;
8887 mddev->curr_resync = 3; /* no longer delayed */
8888 mddev->curr_resync_completed = j;
8889 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8890 md_new_event(mddev);
8891 update_time = jiffies;
8893 blk_start_plug(&plug);
8894 while (j < max_sectors) {
8899 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8900 ((mddev->curr_resync > mddev->curr_resync_completed &&
8901 (mddev->curr_resync - mddev->curr_resync_completed)
8902 > (max_sectors >> 4)) ||
8903 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8904 (j - mddev->curr_resync_completed)*2
8905 >= mddev->resync_max - mddev->curr_resync_completed ||
8906 mddev->curr_resync_completed > mddev->resync_max
8908 /* time to update curr_resync_completed */
8909 wait_event(mddev->recovery_wait,
8910 atomic_read(&mddev->recovery_active) == 0);
8911 mddev->curr_resync_completed = j;
8912 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8913 j > mddev->recovery_cp)
8914 mddev->recovery_cp = j;
8915 update_time = jiffies;
8916 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8917 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8920 while (j >= mddev->resync_max &&
8921 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8922 /* As this condition is controlled by user-space,
8923 * we can block indefinitely, so use '_interruptible'
8924 * to avoid triggering warnings.
8926 flush_signals(current); /* just in case */
8927 wait_event_interruptible(mddev->recovery_wait,
8928 mddev->resync_max > j
8929 || test_bit(MD_RECOVERY_INTR,
8933 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8936 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8938 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8942 if (!skipped) { /* actual IO requested */
8943 io_sectors += sectors;
8944 atomic_add(sectors, &mddev->recovery_active);
8947 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8951 if (j > max_sectors)
8952 /* when skipping, extra large numbers can be returned. */
8955 mddev->curr_resync = j;
8956 mddev->curr_mark_cnt = io_sectors;
8957 if (last_check == 0)
8958 /* this is the earliest that rebuild will be
8959 * visible in /proc/mdstat
8961 md_new_event(mddev);
8963 if (last_check + window > io_sectors || j == max_sectors)
8966 last_check = io_sectors;
8968 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8970 int next = (last_mark+1) % SYNC_MARKS;
8972 mddev->resync_mark = mark[next];
8973 mddev->resync_mark_cnt = mark_cnt[next];
8974 mark[next] = jiffies;
8975 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8979 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8983 * this loop exits only if either when we are slower than
8984 * the 'hard' speed limit, or the system was IO-idle for
8986 * the system might be non-idle CPU-wise, but we only care
8987 * about not overloading the IO subsystem. (things like an
8988 * e2fsck being done on the RAID array should execute fast)
8992 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8993 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8994 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8996 if (currspeed > speed_min(mddev)) {
8997 if (currspeed > speed_max(mddev)) {
9001 if (!is_mddev_idle(mddev, 0)) {
9003 * Give other IO more of a chance.
9004 * The faster the devices, the less we wait.
9006 wait_event(mddev->recovery_wait,
9007 !atomic_read(&mddev->recovery_active));
9011 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9012 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9013 ? "interrupted" : "done");
9015 * this also signals 'finished resyncing' to md_stop
9017 blk_finish_plug(&plug);
9018 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9020 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9021 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9022 mddev->curr_resync > 3) {
9023 mddev->curr_resync_completed = mddev->curr_resync;
9024 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9026 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9028 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9029 mddev->curr_resync > 3) {
9030 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9031 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9032 if (mddev->curr_resync >= mddev->recovery_cp) {
9033 pr_debug("md: checkpointing %s of %s.\n",
9034 desc, mdname(mddev));
9035 if (test_bit(MD_RECOVERY_ERROR,
9037 mddev->recovery_cp =
9038 mddev->curr_resync_completed;
9040 mddev->recovery_cp =
9044 mddev->recovery_cp = MaxSector;
9046 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9047 mddev->curr_resync = MaxSector;
9048 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9049 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9051 rdev_for_each_rcu(rdev, mddev)
9052 if (rdev->raid_disk >= 0 &&
9053 mddev->delta_disks >= 0 &&
9054 !test_bit(Journal, &rdev->flags) &&
9055 !test_bit(Faulty, &rdev->flags) &&
9056 !test_bit(In_sync, &rdev->flags) &&
9057 rdev->recovery_offset < mddev->curr_resync)
9058 rdev->recovery_offset = mddev->curr_resync;
9064 /* set CHANGE_PENDING here since maybe another update is needed,
9065 * so other nodes are informed. It should be harmless for normal
9067 set_mask_bits(&mddev->sb_flags, 0,
9068 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9070 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9071 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9072 mddev->delta_disks > 0 &&
9073 mddev->pers->finish_reshape &&
9074 mddev->pers->size &&
9076 mddev_lock_nointr(mddev);
9077 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9078 mddev_unlock(mddev);
9079 if (!mddev_is_clustered(mddev))
9080 set_capacity_and_notify(mddev->gendisk,
9081 mddev->array_sectors);
9084 spin_lock(&mddev->lock);
9085 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9086 /* We completed so min/max setting can be forgotten if used. */
9087 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9088 mddev->resync_min = 0;
9089 mddev->resync_max = MaxSector;
9090 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9091 mddev->resync_min = mddev->curr_resync_completed;
9092 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9093 mddev->curr_resync = 0;
9094 spin_unlock(&mddev->lock);
9096 wake_up(&resync_wait);
9097 md_wakeup_thread(mddev->thread);
9100 EXPORT_SYMBOL_GPL(md_do_sync);
9102 static int remove_and_add_spares(struct mddev *mddev,
9103 struct md_rdev *this)
9105 struct md_rdev *rdev;
9108 bool remove_some = false;
9110 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9111 /* Mustn't remove devices when resync thread is running */
9114 rdev_for_each(rdev, mddev) {
9115 if ((this == NULL || rdev == this) &&
9116 rdev->raid_disk >= 0 &&
9117 !test_bit(Blocked, &rdev->flags) &&
9118 test_bit(Faulty, &rdev->flags) &&
9119 atomic_read(&rdev->nr_pending)==0) {
9120 /* Faulty non-Blocked devices with nr_pending == 0
9121 * never get nr_pending incremented,
9122 * never get Faulty cleared, and never get Blocked set.
9123 * So we can synchronize_rcu now rather than once per device
9126 set_bit(RemoveSynchronized, &rdev->flags);
9132 rdev_for_each(rdev, mddev) {
9133 if ((this == NULL || rdev == this) &&
9134 rdev->raid_disk >= 0 &&
9135 !test_bit(Blocked, &rdev->flags) &&
9136 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9137 (!test_bit(In_sync, &rdev->flags) &&
9138 !test_bit(Journal, &rdev->flags))) &&
9139 atomic_read(&rdev->nr_pending)==0)) {
9140 if (mddev->pers->hot_remove_disk(
9141 mddev, rdev) == 0) {
9142 sysfs_unlink_rdev(mddev, rdev);
9143 rdev->saved_raid_disk = rdev->raid_disk;
9144 rdev->raid_disk = -1;
9148 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9149 clear_bit(RemoveSynchronized, &rdev->flags);
9152 if (removed && mddev->kobj.sd)
9153 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9155 if (this && removed)
9158 rdev_for_each(rdev, mddev) {
9159 if (this && this != rdev)
9161 if (test_bit(Candidate, &rdev->flags))
9163 if (rdev->raid_disk >= 0 &&
9164 !test_bit(In_sync, &rdev->flags) &&
9165 !test_bit(Journal, &rdev->flags) &&
9166 !test_bit(Faulty, &rdev->flags))
9168 if (rdev->raid_disk >= 0)
9170 if (test_bit(Faulty, &rdev->flags))
9172 if (!test_bit(Journal, &rdev->flags)) {
9174 ! (rdev->saved_raid_disk >= 0 &&
9175 !test_bit(Bitmap_sync, &rdev->flags)))
9178 rdev->recovery_offset = 0;
9180 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9181 /* failure here is OK */
9182 sysfs_link_rdev(mddev, rdev);
9183 if (!test_bit(Journal, &rdev->flags))
9185 md_new_event(mddev);
9186 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9191 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9195 static void md_start_sync(struct work_struct *ws)
9197 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9199 mddev->sync_thread = md_register_thread(md_do_sync,
9202 if (!mddev->sync_thread) {
9203 pr_warn("%s: could not start resync thread...\n",
9205 /* leave the spares where they are, it shouldn't hurt */
9206 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9207 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9208 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9209 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9210 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9211 wake_up(&resync_wait);
9212 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9214 if (mddev->sysfs_action)
9215 sysfs_notify_dirent_safe(mddev->sysfs_action);
9217 md_wakeup_thread(mddev->sync_thread);
9218 sysfs_notify_dirent_safe(mddev->sysfs_action);
9219 md_new_event(mddev);
9223 * This routine is regularly called by all per-raid-array threads to
9224 * deal with generic issues like resync and super-block update.
9225 * Raid personalities that don't have a thread (linear/raid0) do not
9226 * need this as they never do any recovery or update the superblock.
9228 * It does not do any resync itself, but rather "forks" off other threads
9229 * to do that as needed.
9230 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9231 * "->recovery" and create a thread at ->sync_thread.
9232 * When the thread finishes it sets MD_RECOVERY_DONE
9233 * and wakeups up this thread which will reap the thread and finish up.
9234 * This thread also removes any faulty devices (with nr_pending == 0).
9236 * The overall approach is:
9237 * 1/ if the superblock needs updating, update it.
9238 * 2/ If a recovery thread is running, don't do anything else.
9239 * 3/ If recovery has finished, clean up, possibly marking spares active.
9240 * 4/ If there are any faulty devices, remove them.
9241 * 5/ If array is degraded, try to add spares devices
9242 * 6/ If array has spares or is not in-sync, start a resync thread.
9244 void md_check_recovery(struct mddev *mddev)
9246 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9247 /* Write superblock - thread that called mddev_suspend()
9248 * holds reconfig_mutex for us.
9250 set_bit(MD_UPDATING_SB, &mddev->flags);
9251 smp_mb__after_atomic();
9252 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9253 md_update_sb(mddev, 0);
9254 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9255 wake_up(&mddev->sb_wait);
9258 if (mddev->suspended)
9262 md_bitmap_daemon_work(mddev);
9264 if (signal_pending(current)) {
9265 if (mddev->pers->sync_request && !mddev->external) {
9266 pr_debug("md: %s in immediate safe mode\n",
9268 mddev->safemode = 2;
9270 flush_signals(current);
9273 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9276 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9277 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9278 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9279 (mddev->external == 0 && mddev->safemode == 1) ||
9280 (mddev->safemode == 2
9281 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9285 if (mddev_trylock(mddev)) {
9287 bool try_set_sync = mddev->safemode != 0;
9289 if (!mddev->external && mddev->safemode == 1)
9290 mddev->safemode = 0;
9293 struct md_rdev *rdev;
9294 if (!mddev->external && mddev->in_sync)
9295 /* 'Blocked' flag not needed as failed devices
9296 * will be recorded if array switched to read/write.
9297 * Leaving it set will prevent the device
9298 * from being removed.
9300 rdev_for_each(rdev, mddev)
9301 clear_bit(Blocked, &rdev->flags);
9302 /* On a read-only array we can:
9303 * - remove failed devices
9304 * - add already-in_sync devices if the array itself
9306 * As we only add devices that are already in-sync,
9307 * we can activate the spares immediately.
9309 remove_and_add_spares(mddev, NULL);
9310 /* There is no thread, but we need to call
9311 * ->spare_active and clear saved_raid_disk
9313 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9314 md_reap_sync_thread(mddev);
9315 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9316 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9317 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9321 if (mddev_is_clustered(mddev)) {
9322 struct md_rdev *rdev, *tmp;
9323 /* kick the device if another node issued a
9326 rdev_for_each_safe(rdev, tmp, mddev) {
9327 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9328 rdev->raid_disk < 0)
9329 md_kick_rdev_from_array(rdev);
9333 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9334 spin_lock(&mddev->lock);
9336 spin_unlock(&mddev->lock);
9339 if (mddev->sb_flags)
9340 md_update_sb(mddev, 0);
9342 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9343 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9344 /* resync/recovery still happening */
9345 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9348 if (mddev->sync_thread) {
9349 md_reap_sync_thread(mddev);
9352 /* Set RUNNING before clearing NEEDED to avoid
9353 * any transients in the value of "sync_action".
9355 mddev->curr_resync_completed = 0;
9356 spin_lock(&mddev->lock);
9357 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9358 spin_unlock(&mddev->lock);
9359 /* Clear some bits that don't mean anything, but
9362 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9363 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9365 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9366 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9368 /* no recovery is running.
9369 * remove any failed drives, then
9370 * add spares if possible.
9371 * Spares are also removed and re-added, to allow
9372 * the personality to fail the re-add.
9375 if (mddev->reshape_position != MaxSector) {
9376 if (mddev->pers->check_reshape == NULL ||
9377 mddev->pers->check_reshape(mddev) != 0)
9378 /* Cannot proceed */
9380 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9381 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9382 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9383 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9384 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9385 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9386 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9387 } else if (mddev->recovery_cp < MaxSector) {
9388 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9389 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9390 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9391 /* nothing to be done ... */
9394 if (mddev->pers->sync_request) {
9396 /* We are adding a device or devices to an array
9397 * which has the bitmap stored on all devices.
9398 * So make sure all bitmap pages get written
9400 md_bitmap_write_all(mddev->bitmap);
9402 INIT_WORK(&mddev->del_work, md_start_sync);
9403 queue_work(md_misc_wq, &mddev->del_work);
9407 if (!mddev->sync_thread) {
9408 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9409 wake_up(&resync_wait);
9410 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9412 if (mddev->sysfs_action)
9413 sysfs_notify_dirent_safe(mddev->sysfs_action);
9416 wake_up(&mddev->sb_wait);
9417 mddev_unlock(mddev);
9420 EXPORT_SYMBOL(md_check_recovery);
9422 void md_reap_sync_thread(struct mddev *mddev)
9424 struct md_rdev *rdev;
9425 sector_t old_dev_sectors = mddev->dev_sectors;
9426 bool is_reshaped = false;
9428 /* resync has finished, collect result */
9429 md_unregister_thread(&mddev->sync_thread);
9430 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9431 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9432 mddev->degraded != mddev->raid_disks) {
9434 /* activate any spares */
9435 if (mddev->pers->spare_active(mddev)) {
9436 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9437 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9440 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9441 mddev->pers->finish_reshape) {
9442 mddev->pers->finish_reshape(mddev);
9443 if (mddev_is_clustered(mddev))
9447 /* If array is no-longer degraded, then any saved_raid_disk
9448 * information must be scrapped.
9450 if (!mddev->degraded)
9451 rdev_for_each(rdev, mddev)
9452 rdev->saved_raid_disk = -1;
9454 md_update_sb(mddev, 1);
9455 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9456 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9458 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9459 md_cluster_ops->resync_finish(mddev);
9460 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9461 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9462 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9463 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9464 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9465 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9467 * We call md_cluster_ops->update_size here because sync_size could
9468 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9469 * so it is time to update size across cluster.
9471 if (mddev_is_clustered(mddev) && is_reshaped
9472 && !test_bit(MD_CLOSING, &mddev->flags))
9473 md_cluster_ops->update_size(mddev, old_dev_sectors);
9474 wake_up(&resync_wait);
9475 /* flag recovery needed just to double check */
9476 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9477 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9478 sysfs_notify_dirent_safe(mddev->sysfs_action);
9479 md_new_event(mddev);
9480 if (mddev->event_work.func)
9481 queue_work(md_misc_wq, &mddev->event_work);
9483 EXPORT_SYMBOL(md_reap_sync_thread);
9485 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9487 sysfs_notify_dirent_safe(rdev->sysfs_state);
9488 wait_event_timeout(rdev->blocked_wait,
9489 !test_bit(Blocked, &rdev->flags) &&
9490 !test_bit(BlockedBadBlocks, &rdev->flags),
9491 msecs_to_jiffies(5000));
9492 rdev_dec_pending(rdev, mddev);
9494 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9496 void md_finish_reshape(struct mddev *mddev)
9498 /* called be personality module when reshape completes. */
9499 struct md_rdev *rdev;
9501 rdev_for_each(rdev, mddev) {
9502 if (rdev->data_offset > rdev->new_data_offset)
9503 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9505 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9506 rdev->data_offset = rdev->new_data_offset;
9509 EXPORT_SYMBOL(md_finish_reshape);
9511 /* Bad block management */
9513 /* Returns 1 on success, 0 on failure */
9514 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9517 struct mddev *mddev = rdev->mddev;
9520 s += rdev->new_data_offset;
9522 s += rdev->data_offset;
9523 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9525 /* Make sure they get written out promptly */
9526 if (test_bit(ExternalBbl, &rdev->flags))
9527 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9528 sysfs_notify_dirent_safe(rdev->sysfs_state);
9529 set_mask_bits(&mddev->sb_flags, 0,
9530 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9531 md_wakeup_thread(rdev->mddev->thread);
9536 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9538 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9543 s += rdev->new_data_offset;
9545 s += rdev->data_offset;
9546 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9547 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9548 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9551 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9553 static int md_notify_reboot(struct notifier_block *this,
9554 unsigned long code, void *x)
9556 struct list_head *tmp;
9557 struct mddev *mddev;
9560 for_each_mddev(mddev, tmp) {
9561 if (mddev_trylock(mddev)) {
9563 __md_stop_writes(mddev);
9564 if (mddev->persistent)
9565 mddev->safemode = 2;
9566 mddev_unlock(mddev);
9571 * certain more exotic SCSI devices are known to be
9572 * volatile wrt too early system reboots. While the
9573 * right place to handle this issue is the given
9574 * driver, we do want to have a safe RAID driver ...
9582 static struct notifier_block md_notifier = {
9583 .notifier_call = md_notify_reboot,
9585 .priority = INT_MAX, /* before any real devices */
9588 static void md_geninit(void)
9590 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9592 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9595 static int __init md_init(void)
9599 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9603 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9607 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9608 if (!md_rdev_misc_wq)
9609 goto err_rdev_misc_wq;
9611 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9615 ret = __register_blkdev(0, "mdp", md_probe);
9620 register_reboot_notifier(&md_notifier);
9621 raid_table_header = register_sysctl_table(raid_root_table);
9627 unregister_blkdev(MD_MAJOR, "md");
9629 destroy_workqueue(md_rdev_misc_wq);
9631 destroy_workqueue(md_misc_wq);
9633 destroy_workqueue(md_wq);
9638 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9640 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9641 struct md_rdev *rdev2, *tmp;
9643 char b[BDEVNAME_SIZE];
9646 * If size is changed in another node then we need to
9647 * do resize as well.
9649 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9650 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9652 pr_info("md-cluster: resize failed\n");
9654 md_bitmap_update_sb(mddev->bitmap);
9657 /* Check for change of roles in the active devices */
9658 rdev_for_each_safe(rdev2, tmp, mddev) {
9659 if (test_bit(Faulty, &rdev2->flags))
9662 /* Check if the roles changed */
9663 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9665 if (test_bit(Candidate, &rdev2->flags)) {
9666 if (role == 0xfffe) {
9667 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9668 md_kick_rdev_from_array(rdev2);
9672 clear_bit(Candidate, &rdev2->flags);
9675 if (role != rdev2->raid_disk) {
9677 * got activated except reshape is happening.
9679 if (rdev2->raid_disk == -1 && role != 0xffff &&
9680 !(le32_to_cpu(sb->feature_map) &
9681 MD_FEATURE_RESHAPE_ACTIVE)) {
9682 rdev2->saved_raid_disk = role;
9683 ret = remove_and_add_spares(mddev, rdev2);
9684 pr_info("Activated spare: %s\n",
9685 bdevname(rdev2->bdev,b));
9686 /* wakeup mddev->thread here, so array could
9687 * perform resync with the new activated disk */
9688 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9689 md_wakeup_thread(mddev->thread);
9692 * We just want to do the minimum to mark the disk
9693 * as faulty. The recovery is performed by the
9694 * one who initiated the error.
9696 if ((role == 0xfffe) || (role == 0xfffd)) {
9697 md_error(mddev, rdev2);
9698 clear_bit(Blocked, &rdev2->flags);
9703 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9704 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9706 pr_warn("md: updating array disks failed. %d\n", ret);
9710 * Since mddev->delta_disks has already updated in update_raid_disks,
9711 * so it is time to check reshape.
9713 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9714 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9716 * reshape is happening in the remote node, we need to
9717 * update reshape_position and call start_reshape.
9719 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9720 if (mddev->pers->update_reshape_pos)
9721 mddev->pers->update_reshape_pos(mddev);
9722 if (mddev->pers->start_reshape)
9723 mddev->pers->start_reshape(mddev);
9724 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9725 mddev->reshape_position != MaxSector &&
9726 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9727 /* reshape is just done in another node. */
9728 mddev->reshape_position = MaxSector;
9729 if (mddev->pers->update_reshape_pos)
9730 mddev->pers->update_reshape_pos(mddev);
9733 /* Finally set the event to be up to date */
9734 mddev->events = le64_to_cpu(sb->events);
9737 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9740 struct page *swapout = rdev->sb_page;
9741 struct mdp_superblock_1 *sb;
9743 /* Store the sb page of the rdev in the swapout temporary
9744 * variable in case we err in the future
9746 rdev->sb_page = NULL;
9747 err = alloc_disk_sb(rdev);
9749 ClearPageUptodate(rdev->sb_page);
9750 rdev->sb_loaded = 0;
9751 err = super_types[mddev->major_version].
9752 load_super(rdev, NULL, mddev->minor_version);
9755 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9756 __func__, __LINE__, rdev->desc_nr, err);
9758 put_page(rdev->sb_page);
9759 rdev->sb_page = swapout;
9760 rdev->sb_loaded = 1;
9764 sb = page_address(rdev->sb_page);
9765 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9769 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9770 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9772 /* The other node finished recovery, call spare_active to set
9773 * device In_sync and mddev->degraded
9775 if (rdev->recovery_offset == MaxSector &&
9776 !test_bit(In_sync, &rdev->flags) &&
9777 mddev->pers->spare_active(mddev))
9778 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9784 void md_reload_sb(struct mddev *mddev, int nr)
9786 struct md_rdev *rdev = NULL, *iter;
9790 rdev_for_each_rcu(iter, mddev) {
9791 if (iter->desc_nr == nr) {
9798 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9802 err = read_rdev(mddev, rdev);
9806 check_sb_changes(mddev, rdev);
9808 /* Read all rdev's to update recovery_offset */
9809 rdev_for_each_rcu(rdev, mddev) {
9810 if (!test_bit(Faulty, &rdev->flags))
9811 read_rdev(mddev, rdev);
9814 EXPORT_SYMBOL(md_reload_sb);
9819 * Searches all registered partitions for autorun RAID arrays
9823 static DEFINE_MUTEX(detected_devices_mutex);
9824 static LIST_HEAD(all_detected_devices);
9825 struct detected_devices_node {
9826 struct list_head list;
9830 void md_autodetect_dev(dev_t dev)
9832 struct detected_devices_node *node_detected_dev;
9834 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9835 if (node_detected_dev) {
9836 node_detected_dev->dev = dev;
9837 mutex_lock(&detected_devices_mutex);
9838 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9839 mutex_unlock(&detected_devices_mutex);
9843 void md_autostart_arrays(int part)
9845 struct md_rdev *rdev;
9846 struct detected_devices_node *node_detected_dev;
9848 int i_scanned, i_passed;
9853 pr_info("md: Autodetecting RAID arrays.\n");
9855 mutex_lock(&detected_devices_mutex);
9856 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9858 node_detected_dev = list_entry(all_detected_devices.next,
9859 struct detected_devices_node, list);
9860 list_del(&node_detected_dev->list);
9861 dev = node_detected_dev->dev;
9862 kfree(node_detected_dev);
9863 mutex_unlock(&detected_devices_mutex);
9864 rdev = md_import_device(dev,0, 90);
9865 mutex_lock(&detected_devices_mutex);
9869 if (test_bit(Faulty, &rdev->flags))
9872 set_bit(AutoDetected, &rdev->flags);
9873 list_add(&rdev->same_set, &pending_raid_disks);
9876 mutex_unlock(&detected_devices_mutex);
9878 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9880 autorun_devices(part);
9883 #endif /* !MODULE */
9885 static __exit void md_exit(void)
9887 struct mddev *mddev;
9888 struct list_head *tmp;
9891 unregister_blkdev(MD_MAJOR,"md");
9892 unregister_blkdev(mdp_major, "mdp");
9893 unregister_reboot_notifier(&md_notifier);
9894 unregister_sysctl_table(raid_table_header);
9896 /* We cannot unload the modules while some process is
9897 * waiting for us in select() or poll() - wake them up
9900 while (waitqueue_active(&md_event_waiters)) {
9901 /* not safe to leave yet */
9902 wake_up(&md_event_waiters);
9906 remove_proc_entry("mdstat", NULL);
9908 for_each_mddev(mddev, tmp) {
9909 export_array(mddev);
9911 mddev->hold_active = 0;
9913 * for_each_mddev() will call mddev_put() at the end of each
9914 * iteration. As the mddev is now fully clear, this will
9915 * schedule the mddev for destruction by a workqueue, and the
9916 * destroy_workqueue() below will wait for that to complete.
9919 destroy_workqueue(md_rdev_misc_wq);
9920 destroy_workqueue(md_misc_wq);
9921 destroy_workqueue(md_wq);
9924 subsys_initcall(md_init);
9925 module_exit(md_exit)
9927 static int get_ro(char *buffer, const struct kernel_param *kp)
9929 return sprintf(buffer, "%d\n", start_readonly);
9931 static int set_ro(const char *val, const struct kernel_param *kp)
9933 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9936 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9937 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9938 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9939 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9941 MODULE_LICENSE("GPL");
9942 MODULE_DESCRIPTION("MD RAID framework");
9944 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);