1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/module.h>
55 #include <linux/reboot.h>
56 #include <linux/file.h>
57 #include <linux/compat.h>
58 #include <linux/delay.h>
59 #include <linux/raid/md_p.h>
60 #include <linux/raid/md_u.h>
61 #include <linux/raid/detect.h>
62 #include <linux/slab.h>
63 #include <linux/percpu-refcount.h>
64 #include <linux/part_stat.h>
66 #include <trace/events/block.h>
68 #include "md-bitmap.h"
69 #include "md-cluster.h"
71 /* pers_list is a list of registered personalities protected
73 * pers_lock does extra service to protect accesses to
74 * mddev->thread when the mutex cannot be held.
76 static LIST_HEAD(pers_list);
77 static DEFINE_SPINLOCK(pers_lock);
79 static struct kobj_type md_ktype;
81 struct md_cluster_operations *md_cluster_ops;
82 EXPORT_SYMBOL(md_cluster_ops);
83 static struct module *md_cluster_mod;
85 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
86 static struct workqueue_struct *md_wq;
87 static struct workqueue_struct *md_misc_wq;
88 static struct workqueue_struct *md_rdev_misc_wq;
90 static int remove_and_add_spares(struct mddev *mddev,
91 struct md_rdev *this);
92 static void mddev_detach(struct mddev *mddev);
95 * Default number of read corrections we'll attempt on an rdev
96 * before ejecting it from the array. We divide the read error
97 * count by 2 for every hour elapsed between read errors.
99 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
100 /* Default safemode delay: 200 msec */
101 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
103 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
104 * is 1000 KB/sec, so the extra system load does not show up that much.
105 * Increase it if you want to have more _guaranteed_ speed. Note that
106 * the RAID driver will use the maximum available bandwidth if the IO
107 * subsystem is idle. There is also an 'absolute maximum' reconstruction
108 * speed limit - in case reconstruction slows down your system despite
111 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
112 * or /sys/block/mdX/md/sync_speed_{min,max}
115 static int sysctl_speed_limit_min = 1000;
116 static int sysctl_speed_limit_max = 200000;
117 static inline int speed_min(struct mddev *mddev)
119 return mddev->sync_speed_min ?
120 mddev->sync_speed_min : sysctl_speed_limit_min;
123 static inline int speed_max(struct mddev *mddev)
125 return mddev->sync_speed_max ?
126 mddev->sync_speed_max : sysctl_speed_limit_max;
129 static void rdev_uninit_serial(struct md_rdev *rdev)
131 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
134 kvfree(rdev->serial);
138 static void rdevs_uninit_serial(struct mddev *mddev)
140 struct md_rdev *rdev;
142 rdev_for_each(rdev, mddev)
143 rdev_uninit_serial(rdev);
146 static int rdev_init_serial(struct md_rdev *rdev)
148 /* serial_nums equals with BARRIER_BUCKETS_NR */
149 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
150 struct serial_in_rdev *serial = NULL;
152 if (test_bit(CollisionCheck, &rdev->flags))
155 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
160 for (i = 0; i < serial_nums; i++) {
161 struct serial_in_rdev *serial_tmp = &serial[i];
163 spin_lock_init(&serial_tmp->serial_lock);
164 serial_tmp->serial_rb = RB_ROOT_CACHED;
165 init_waitqueue_head(&serial_tmp->serial_io_wait);
168 rdev->serial = serial;
169 set_bit(CollisionCheck, &rdev->flags);
174 static int rdevs_init_serial(struct mddev *mddev)
176 struct md_rdev *rdev;
179 rdev_for_each(rdev, mddev) {
180 ret = rdev_init_serial(rdev);
185 /* Free all resources if pool is not existed */
186 if (ret && !mddev->serial_info_pool)
187 rdevs_uninit_serial(mddev);
193 * rdev needs to enable serial stuffs if it meets the conditions:
194 * 1. it is multi-queue device flaged with writemostly.
195 * 2. the write-behind mode is enabled.
197 static int rdev_need_serial(struct md_rdev *rdev)
199 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
200 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
201 test_bit(WriteMostly, &rdev->flags));
205 * Init resource for rdev(s), then create serial_info_pool if:
206 * 1. rdev is the first device which return true from rdev_enable_serial.
207 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
209 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
214 if (rdev && !rdev_need_serial(rdev) &&
215 !test_bit(CollisionCheck, &rdev->flags))
219 mddev_suspend(mddev);
222 ret = rdevs_init_serial(mddev);
224 ret = rdev_init_serial(rdev);
228 if (mddev->serial_info_pool == NULL) {
230 * already in memalloc noio context by
233 mddev->serial_info_pool =
234 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
235 sizeof(struct serial_info));
236 if (!mddev->serial_info_pool) {
237 rdevs_uninit_serial(mddev);
238 pr_err("can't alloc memory pool for serialization\n");
248 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
249 * 1. rdev is the last device flaged with CollisionCheck.
250 * 2. when bitmap is destroyed while policy is not enabled.
251 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
253 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
256 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
259 if (mddev->serial_info_pool) {
260 struct md_rdev *temp;
261 int num = 0; /* used to track if other rdevs need the pool */
264 mddev_suspend(mddev);
265 rdev_for_each(temp, mddev) {
267 if (!mddev->serialize_policy ||
268 !rdev_need_serial(temp))
269 rdev_uninit_serial(temp);
272 } else if (temp != rdev &&
273 test_bit(CollisionCheck, &temp->flags))
278 rdev_uninit_serial(rdev);
281 pr_info("The mempool could be used by other devices\n");
283 mempool_destroy(mddev->serial_info_pool);
284 mddev->serial_info_pool = NULL;
291 static struct ctl_table_header *raid_table_header;
293 static struct ctl_table raid_table[] = {
295 .procname = "speed_limit_min",
296 .data = &sysctl_speed_limit_min,
297 .maxlen = sizeof(int),
298 .mode = S_IRUGO|S_IWUSR,
299 .proc_handler = proc_dointvec,
302 .procname = "speed_limit_max",
303 .data = &sysctl_speed_limit_max,
304 .maxlen = sizeof(int),
305 .mode = S_IRUGO|S_IWUSR,
306 .proc_handler = proc_dointvec,
311 static struct ctl_table raid_dir_table[] = {
315 .mode = S_IRUGO|S_IXUGO,
321 static struct ctl_table raid_root_table[] = {
326 .child = raid_dir_table,
331 static int start_readonly;
334 * The original mechanism for creating an md device is to create
335 * a device node in /dev and to open it. This causes races with device-close.
336 * The preferred method is to write to the "new_array" module parameter.
337 * This can avoid races.
338 * Setting create_on_open to false disables the original mechanism
339 * so all the races disappear.
341 static bool create_on_open = true;
344 * We have a system wide 'event count' that is incremented
345 * on any 'interesting' event, and readers of /proc/mdstat
346 * can use 'poll' or 'select' to find out when the event
350 * start array, stop array, error, add device, remove device,
351 * start build, activate spare
353 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
354 static atomic_t md_event_count;
355 void md_new_event(struct mddev *mddev)
357 atomic_inc(&md_event_count);
358 wake_up(&md_event_waiters);
360 EXPORT_SYMBOL_GPL(md_new_event);
363 * Enables to iterate over all existing md arrays
364 * all_mddevs_lock protects this list.
366 static LIST_HEAD(all_mddevs);
367 static DEFINE_SPINLOCK(all_mddevs_lock);
370 * iterates through all used mddevs in the system.
371 * We take care to grab the all_mddevs_lock whenever navigating
372 * the list, and to always hold a refcount when unlocked.
373 * Any code which breaks out of this loop while own
374 * a reference to the current mddev and must mddev_put it.
376 #define for_each_mddev(_mddev,_tmp) \
378 for (({ spin_lock(&all_mddevs_lock); \
379 _tmp = all_mddevs.next; \
381 ({ if (_tmp != &all_mddevs) \
382 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
383 spin_unlock(&all_mddevs_lock); \
384 if (_mddev) mddev_put(_mddev); \
385 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
386 _tmp != &all_mddevs;}); \
387 ({ spin_lock(&all_mddevs_lock); \
388 _tmp = _tmp->next;}) \
391 /* Rather than calling directly into the personality make_request function,
392 * IO requests come here first so that we can check if the device is
393 * being suspended pending a reconfiguration.
394 * We hold a refcount over the call to ->make_request. By the time that
395 * call has finished, the bio has been linked into some internal structure
396 * and so is visible to ->quiesce(), so we don't need the refcount any more.
398 static bool is_suspended(struct mddev *mddev, struct bio *bio)
400 if (mddev->suspended)
402 if (bio_data_dir(bio) != WRITE)
404 if (mddev->suspend_lo >= mddev->suspend_hi)
406 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
408 if (bio_end_sector(bio) < mddev->suspend_lo)
413 void md_handle_request(struct mddev *mddev, struct bio *bio)
417 if (is_suspended(mddev, bio)) {
420 prepare_to_wait(&mddev->sb_wait, &__wait,
421 TASK_UNINTERRUPTIBLE);
422 if (!is_suspended(mddev, bio))
428 finish_wait(&mddev->sb_wait, &__wait);
430 atomic_inc(&mddev->active_io);
433 if (!mddev->pers->make_request(mddev, bio)) {
434 atomic_dec(&mddev->active_io);
435 wake_up(&mddev->sb_wait);
436 goto check_suspended;
439 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
440 wake_up(&mddev->sb_wait);
442 EXPORT_SYMBOL(md_handle_request);
444 static blk_qc_t md_submit_bio(struct bio *bio)
446 const int rw = bio_data_dir(bio);
447 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
449 if (mddev == NULL || mddev->pers == NULL) {
451 return BLK_QC_T_NONE;
454 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
456 return BLK_QC_T_NONE;
459 blk_queue_split(&bio);
461 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
462 if (bio_sectors(bio) != 0)
463 bio->bi_status = BLK_STS_IOERR;
465 return BLK_QC_T_NONE;
468 /* bio could be mergeable after passing to underlayer */
469 bio->bi_opf &= ~REQ_NOMERGE;
471 md_handle_request(mddev, bio);
473 return BLK_QC_T_NONE;
476 /* mddev_suspend makes sure no new requests are submitted
477 * to the device, and that any requests that have been submitted
478 * are completely handled.
479 * Once mddev_detach() is called and completes, the module will be
482 void mddev_suspend(struct mddev *mddev)
484 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
485 lockdep_assert_held(&mddev->reconfig_mutex);
486 if (mddev->suspended++)
489 wake_up(&mddev->sb_wait);
490 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
491 smp_mb__after_atomic();
492 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
493 mddev->pers->quiesce(mddev, 1);
494 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
495 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
497 del_timer_sync(&mddev->safemode_timer);
498 /* restrict memory reclaim I/O during raid array is suspend */
499 mddev->noio_flag = memalloc_noio_save();
501 EXPORT_SYMBOL_GPL(mddev_suspend);
503 void mddev_resume(struct mddev *mddev)
505 /* entred the memalloc scope from mddev_suspend() */
506 memalloc_noio_restore(mddev->noio_flag);
507 lockdep_assert_held(&mddev->reconfig_mutex);
508 if (--mddev->suspended)
510 wake_up(&mddev->sb_wait);
511 mddev->pers->quiesce(mddev, 0);
513 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
514 md_wakeup_thread(mddev->thread);
515 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
517 EXPORT_SYMBOL_GPL(mddev_resume);
520 * Generic flush handling for md
523 static void md_end_flush(struct bio *bio)
525 struct md_rdev *rdev = bio->bi_private;
526 struct mddev *mddev = rdev->mddev;
528 rdev_dec_pending(rdev, mddev);
530 if (atomic_dec_and_test(&mddev->flush_pending)) {
531 /* The pre-request flush has finished */
532 queue_work(md_wq, &mddev->flush_work);
537 static void md_submit_flush_data(struct work_struct *ws);
539 static void submit_flushes(struct work_struct *ws)
541 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
542 struct md_rdev *rdev;
544 mddev->start_flush = ktime_get_boottime();
545 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
546 atomic_set(&mddev->flush_pending, 1);
548 rdev_for_each_rcu(rdev, mddev)
549 if (rdev->raid_disk >= 0 &&
550 !test_bit(Faulty, &rdev->flags)) {
551 /* Take two references, one is dropped
552 * when request finishes, one after
553 * we reclaim rcu_read_lock
556 atomic_inc(&rdev->nr_pending);
557 atomic_inc(&rdev->nr_pending);
559 bi = bio_alloc_bioset(GFP_NOIO, 0, &mddev->bio_set);
560 bi->bi_end_io = md_end_flush;
561 bi->bi_private = rdev;
562 bio_set_dev(bi, rdev->bdev);
563 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
564 atomic_inc(&mddev->flush_pending);
567 rdev_dec_pending(rdev, mddev);
570 if (atomic_dec_and_test(&mddev->flush_pending))
571 queue_work(md_wq, &mddev->flush_work);
574 static void md_submit_flush_data(struct work_struct *ws)
576 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
577 struct bio *bio = mddev->flush_bio;
580 * must reset flush_bio before calling into md_handle_request to avoid a
581 * deadlock, because other bios passed md_handle_request suspend check
582 * could wait for this and below md_handle_request could wait for those
583 * bios because of suspend check
585 spin_lock_irq(&mddev->lock);
586 mddev->prev_flush_start = mddev->start_flush;
587 mddev->flush_bio = NULL;
588 spin_unlock_irq(&mddev->lock);
589 wake_up(&mddev->sb_wait);
591 if (bio->bi_iter.bi_size == 0) {
592 /* an empty barrier - all done */
595 bio->bi_opf &= ~REQ_PREFLUSH;
596 md_handle_request(mddev, bio);
601 * Manages consolidation of flushes and submitting any flushes needed for
602 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
603 * being finished in another context. Returns false if the flushing is
604 * complete but still needs the I/O portion of the bio to be processed.
606 bool md_flush_request(struct mddev *mddev, struct bio *bio)
608 ktime_t req_start = ktime_get_boottime();
609 spin_lock_irq(&mddev->lock);
610 /* flush requests wait until ongoing flush completes,
611 * hence coalescing all the pending requests.
613 wait_event_lock_irq(mddev->sb_wait,
615 ktime_before(req_start, mddev->prev_flush_start),
617 /* new request after previous flush is completed */
618 if (ktime_after(req_start, mddev->prev_flush_start)) {
619 WARN_ON(mddev->flush_bio);
620 mddev->flush_bio = bio;
623 spin_unlock_irq(&mddev->lock);
626 INIT_WORK(&mddev->flush_work, submit_flushes);
627 queue_work(md_wq, &mddev->flush_work);
629 /* flush was performed for some other bio while we waited. */
630 if (bio->bi_iter.bi_size == 0)
631 /* an empty barrier - all done */
634 bio->bi_opf &= ~REQ_PREFLUSH;
640 EXPORT_SYMBOL(md_flush_request);
642 static inline struct mddev *mddev_get(struct mddev *mddev)
644 atomic_inc(&mddev->active);
648 static void mddev_delayed_delete(struct work_struct *ws);
650 static void mddev_put(struct mddev *mddev)
652 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
654 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
655 mddev->ctime == 0 && !mddev->hold_active) {
656 /* Array is not configured at all, and not held active,
658 list_del_init(&mddev->all_mddevs);
661 * Call queue_work inside the spinlock so that
662 * flush_workqueue() after mddev_find will succeed in waiting
663 * for the work to be done.
665 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
666 queue_work(md_misc_wq, &mddev->del_work);
668 spin_unlock(&all_mddevs_lock);
671 static void md_safemode_timeout(struct timer_list *t);
673 void mddev_init(struct mddev *mddev)
675 kobject_init(&mddev->kobj, &md_ktype);
676 mutex_init(&mddev->open_mutex);
677 mutex_init(&mddev->reconfig_mutex);
678 mutex_init(&mddev->bitmap_info.mutex);
679 INIT_LIST_HEAD(&mddev->disks);
680 INIT_LIST_HEAD(&mddev->all_mddevs);
681 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
682 atomic_set(&mddev->active, 1);
683 atomic_set(&mddev->openers, 0);
684 atomic_set(&mddev->active_io, 0);
685 spin_lock_init(&mddev->lock);
686 atomic_set(&mddev->flush_pending, 0);
687 init_waitqueue_head(&mddev->sb_wait);
688 init_waitqueue_head(&mddev->recovery_wait);
689 mddev->reshape_position = MaxSector;
690 mddev->reshape_backwards = 0;
691 mddev->last_sync_action = "none";
692 mddev->resync_min = 0;
693 mddev->resync_max = MaxSector;
694 mddev->level = LEVEL_NONE;
696 EXPORT_SYMBOL_GPL(mddev_init);
698 static struct mddev *mddev_find_locked(dev_t unit)
702 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
703 if (mddev->unit == unit)
709 /* find an unused unit number */
710 static dev_t mddev_alloc_unit(void)
712 static int next_minor = 512;
713 int start = next_minor;
718 dev = MKDEV(MD_MAJOR, next_minor);
720 if (next_minor > MINORMASK)
722 if (next_minor == start)
723 return 0; /* Oh dear, all in use. */
724 is_free = !mddev_find_locked(dev);
730 static struct mddev *mddev_find(dev_t unit)
734 if (MAJOR(unit) != MD_MAJOR)
735 unit &= ~((1 << MdpMinorShift) - 1);
737 spin_lock(&all_mddevs_lock);
738 mddev = mddev_find_locked(unit);
741 spin_unlock(&all_mddevs_lock);
746 static struct mddev *mddev_alloc(dev_t unit)
751 if (unit && MAJOR(unit) != MD_MAJOR)
752 unit &= ~((1 << MdpMinorShift) - 1);
754 new = kzalloc(sizeof(*new), GFP_KERNEL);
756 return ERR_PTR(-ENOMEM);
759 spin_lock(&all_mddevs_lock);
762 if (mddev_find_locked(unit))
765 if (MAJOR(unit) == MD_MAJOR)
766 new->md_minor = MINOR(unit);
768 new->md_minor = MINOR(unit) >> MdpMinorShift;
769 new->hold_active = UNTIL_IOCTL;
772 new->unit = mddev_alloc_unit();
775 new->md_minor = MINOR(new->unit);
776 new->hold_active = UNTIL_STOP;
779 list_add(&new->all_mddevs, &all_mddevs);
780 spin_unlock(&all_mddevs_lock);
783 spin_unlock(&all_mddevs_lock);
785 return ERR_PTR(error);
788 static const struct attribute_group md_redundancy_group;
790 void mddev_unlock(struct mddev *mddev)
792 if (mddev->to_remove) {
793 /* These cannot be removed under reconfig_mutex as
794 * an access to the files will try to take reconfig_mutex
795 * while holding the file unremovable, which leads to
797 * So hold set sysfs_active while the remove in happeing,
798 * and anything else which might set ->to_remove or my
799 * otherwise change the sysfs namespace will fail with
800 * -EBUSY if sysfs_active is still set.
801 * We set sysfs_active under reconfig_mutex and elsewhere
802 * test it under the same mutex to ensure its correct value
805 const struct attribute_group *to_remove = mddev->to_remove;
806 mddev->to_remove = NULL;
807 mddev->sysfs_active = 1;
808 mutex_unlock(&mddev->reconfig_mutex);
810 if (mddev->kobj.sd) {
811 if (to_remove != &md_redundancy_group)
812 sysfs_remove_group(&mddev->kobj, to_remove);
813 if (mddev->pers == NULL ||
814 mddev->pers->sync_request == NULL) {
815 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
816 if (mddev->sysfs_action)
817 sysfs_put(mddev->sysfs_action);
818 if (mddev->sysfs_completed)
819 sysfs_put(mddev->sysfs_completed);
820 if (mddev->sysfs_degraded)
821 sysfs_put(mddev->sysfs_degraded);
822 mddev->sysfs_action = NULL;
823 mddev->sysfs_completed = NULL;
824 mddev->sysfs_degraded = NULL;
827 mddev->sysfs_active = 0;
829 mutex_unlock(&mddev->reconfig_mutex);
831 /* As we've dropped the mutex we need a spinlock to
832 * make sure the thread doesn't disappear
834 spin_lock(&pers_lock);
835 md_wakeup_thread(mddev->thread);
836 wake_up(&mddev->sb_wait);
837 spin_unlock(&pers_lock);
839 EXPORT_SYMBOL_GPL(mddev_unlock);
841 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
843 struct md_rdev *rdev;
845 rdev_for_each_rcu(rdev, mddev)
846 if (rdev->desc_nr == nr)
851 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
853 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
855 struct md_rdev *rdev;
857 rdev_for_each(rdev, mddev)
858 if (rdev->bdev->bd_dev == dev)
864 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
866 struct md_rdev *rdev;
868 rdev_for_each_rcu(rdev, mddev)
869 if (rdev->bdev->bd_dev == dev)
874 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
876 static struct md_personality *find_pers(int level, char *clevel)
878 struct md_personality *pers;
879 list_for_each_entry(pers, &pers_list, list) {
880 if (level != LEVEL_NONE && pers->level == level)
882 if (strcmp(pers->name, clevel)==0)
888 /* return the offset of the super block in 512byte sectors */
889 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
891 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
892 return MD_NEW_SIZE_SECTORS(num_sectors);
895 static int alloc_disk_sb(struct md_rdev *rdev)
897 rdev->sb_page = alloc_page(GFP_KERNEL);
903 void md_rdev_clear(struct md_rdev *rdev)
906 put_page(rdev->sb_page);
908 rdev->sb_page = NULL;
913 put_page(rdev->bb_page);
914 rdev->bb_page = NULL;
916 badblocks_exit(&rdev->badblocks);
918 EXPORT_SYMBOL_GPL(md_rdev_clear);
920 static void super_written(struct bio *bio)
922 struct md_rdev *rdev = bio->bi_private;
923 struct mddev *mddev = rdev->mddev;
925 if (bio->bi_status) {
926 pr_err("md: %s gets error=%d\n", __func__,
927 blk_status_to_errno(bio->bi_status));
928 md_error(mddev, rdev);
929 if (!test_bit(Faulty, &rdev->flags)
930 && (bio->bi_opf & MD_FAILFAST)) {
931 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
932 set_bit(LastDev, &rdev->flags);
935 clear_bit(LastDev, &rdev->flags);
937 if (atomic_dec_and_test(&mddev->pending_writes))
938 wake_up(&mddev->sb_wait);
939 rdev_dec_pending(rdev, mddev);
943 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
944 sector_t sector, int size, struct page *page)
946 /* write first size bytes of page to sector of rdev
947 * Increment mddev->pending_writes before returning
948 * and decrement it on completion, waking up sb_wait
949 * if zero is reached.
950 * If an error occurred, call md_error
958 if (test_bit(Faulty, &rdev->flags))
961 bio = bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
963 atomic_inc(&rdev->nr_pending);
965 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
966 bio->bi_iter.bi_sector = sector;
967 bio_add_page(bio, page, size, 0);
968 bio->bi_private = rdev;
969 bio->bi_end_io = super_written;
971 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
972 test_bit(FailFast, &rdev->flags) &&
973 !test_bit(LastDev, &rdev->flags))
975 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
977 atomic_inc(&mddev->pending_writes);
981 int md_super_wait(struct mddev *mddev)
983 /* wait for all superblock writes that were scheduled to complete */
984 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
985 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
990 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
991 struct page *page, int op, int op_flags, bool metadata_op)
996 bio_init(&bio, &bvec, 1);
998 if (metadata_op && rdev->meta_bdev)
999 bio_set_dev(&bio, rdev->meta_bdev);
1001 bio_set_dev(&bio, rdev->bdev);
1002 bio.bi_opf = op | op_flags;
1004 bio.bi_iter.bi_sector = sector + rdev->sb_start;
1005 else if (rdev->mddev->reshape_position != MaxSector &&
1006 (rdev->mddev->reshape_backwards ==
1007 (sector >= rdev->mddev->reshape_position)))
1008 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1010 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1011 bio_add_page(&bio, page, size, 0);
1013 submit_bio_wait(&bio);
1015 return !bio.bi_status;
1017 EXPORT_SYMBOL_GPL(sync_page_io);
1019 static int read_disk_sb(struct md_rdev *rdev, int size)
1021 char b[BDEVNAME_SIZE];
1023 if (rdev->sb_loaded)
1026 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1028 rdev->sb_loaded = 1;
1032 pr_err("md: disabled device %s, could not read superblock.\n",
1033 bdevname(rdev->bdev,b));
1037 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1039 return sb1->set_uuid0 == sb2->set_uuid0 &&
1040 sb1->set_uuid1 == sb2->set_uuid1 &&
1041 sb1->set_uuid2 == sb2->set_uuid2 &&
1042 sb1->set_uuid3 == sb2->set_uuid3;
1045 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1048 mdp_super_t *tmp1, *tmp2;
1050 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1051 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1053 if (!tmp1 || !tmp2) {
1062 * nr_disks is not constant
1067 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1074 static u32 md_csum_fold(u32 csum)
1076 csum = (csum & 0xffff) + (csum >> 16);
1077 return (csum & 0xffff) + (csum >> 16);
1080 static unsigned int calc_sb_csum(mdp_super_t *sb)
1083 u32 *sb32 = (u32*)sb;
1085 unsigned int disk_csum, csum;
1087 disk_csum = sb->sb_csum;
1090 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1092 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1095 /* This used to use csum_partial, which was wrong for several
1096 * reasons including that different results are returned on
1097 * different architectures. It isn't critical that we get exactly
1098 * the same return value as before (we always csum_fold before
1099 * testing, and that removes any differences). However as we
1100 * know that csum_partial always returned a 16bit value on
1101 * alphas, do a fold to maximise conformity to previous behaviour.
1103 sb->sb_csum = md_csum_fold(disk_csum);
1105 sb->sb_csum = disk_csum;
1111 * Handle superblock details.
1112 * We want to be able to handle multiple superblock formats
1113 * so we have a common interface to them all, and an array of
1114 * different handlers.
1115 * We rely on user-space to write the initial superblock, and support
1116 * reading and updating of superblocks.
1117 * Interface methods are:
1118 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1119 * loads and validates a superblock on dev.
1120 * if refdev != NULL, compare superblocks on both devices
1122 * 0 - dev has a superblock that is compatible with refdev
1123 * 1 - dev has a superblock that is compatible and newer than refdev
1124 * so dev should be used as the refdev in future
1125 * -EINVAL superblock incompatible or invalid
1126 * -othererror e.g. -EIO
1128 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1129 * Verify that dev is acceptable into mddev.
1130 * The first time, mddev->raid_disks will be 0, and data from
1131 * dev should be merged in. Subsequent calls check that dev
1132 * is new enough. Return 0 or -EINVAL
1134 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1135 * Update the superblock for rdev with data in mddev
1136 * This does not write to disc.
1142 struct module *owner;
1143 int (*load_super)(struct md_rdev *rdev,
1144 struct md_rdev *refdev,
1146 int (*validate_super)(struct mddev *mddev,
1147 struct md_rdev *rdev);
1148 void (*sync_super)(struct mddev *mddev,
1149 struct md_rdev *rdev);
1150 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1151 sector_t num_sectors);
1152 int (*allow_new_offset)(struct md_rdev *rdev,
1153 unsigned long long new_offset);
1157 * Check that the given mddev has no bitmap.
1159 * This function is called from the run method of all personalities that do not
1160 * support bitmaps. It prints an error message and returns non-zero if mddev
1161 * has a bitmap. Otherwise, it returns 0.
1164 int md_check_no_bitmap(struct mddev *mddev)
1166 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1168 pr_warn("%s: bitmaps are not supported for %s\n",
1169 mdname(mddev), mddev->pers->name);
1172 EXPORT_SYMBOL(md_check_no_bitmap);
1175 * load_super for 0.90.0
1177 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1179 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1182 bool spare_disk = true;
1185 * Calculate the position of the superblock (512byte sectors),
1186 * it's at the end of the disk.
1188 * It also happens to be a multiple of 4Kb.
1190 rdev->sb_start = calc_dev_sboffset(rdev);
1192 ret = read_disk_sb(rdev, MD_SB_BYTES);
1198 bdevname(rdev->bdev, b);
1199 sb = page_address(rdev->sb_page);
1201 if (sb->md_magic != MD_SB_MAGIC) {
1202 pr_warn("md: invalid raid superblock magic on %s\n", b);
1206 if (sb->major_version != 0 ||
1207 sb->minor_version < 90 ||
1208 sb->minor_version > 91) {
1209 pr_warn("Bad version number %d.%d on %s\n",
1210 sb->major_version, sb->minor_version, b);
1214 if (sb->raid_disks <= 0)
1217 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1218 pr_warn("md: invalid superblock checksum on %s\n", b);
1222 rdev->preferred_minor = sb->md_minor;
1223 rdev->data_offset = 0;
1224 rdev->new_data_offset = 0;
1225 rdev->sb_size = MD_SB_BYTES;
1226 rdev->badblocks.shift = -1;
1228 if (sb->level == LEVEL_MULTIPATH)
1231 rdev->desc_nr = sb->this_disk.number;
1233 /* not spare disk, or LEVEL_MULTIPATH */
1234 if (sb->level == LEVEL_MULTIPATH ||
1235 (rdev->desc_nr >= 0 &&
1236 rdev->desc_nr < MD_SB_DISKS &&
1237 sb->disks[rdev->desc_nr].state &
1238 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1248 mdp_super_t *refsb = page_address(refdev->sb_page);
1249 if (!md_uuid_equal(refsb, sb)) {
1250 pr_warn("md: %s has different UUID to %s\n",
1251 b, bdevname(refdev->bdev,b2));
1254 if (!md_sb_equal(refsb, sb)) {
1255 pr_warn("md: %s has same UUID but different superblock to %s\n",
1256 b, bdevname(refdev->bdev, b2));
1260 ev2 = md_event(refsb);
1262 if (!spare_disk && ev1 > ev2)
1267 rdev->sectors = rdev->sb_start;
1268 /* Limit to 4TB as metadata cannot record more than that.
1269 * (not needed for Linear and RAID0 as metadata doesn't
1272 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1273 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1275 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1276 /* "this cannot possibly happen" ... */
1284 * validate_super for 0.90.0
1286 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1289 mdp_super_t *sb = page_address(rdev->sb_page);
1290 __u64 ev1 = md_event(sb);
1292 rdev->raid_disk = -1;
1293 clear_bit(Faulty, &rdev->flags);
1294 clear_bit(In_sync, &rdev->flags);
1295 clear_bit(Bitmap_sync, &rdev->flags);
1296 clear_bit(WriteMostly, &rdev->flags);
1298 if (mddev->raid_disks == 0) {
1299 mddev->major_version = 0;
1300 mddev->minor_version = sb->minor_version;
1301 mddev->patch_version = sb->patch_version;
1302 mddev->external = 0;
1303 mddev->chunk_sectors = sb->chunk_size >> 9;
1304 mddev->ctime = sb->ctime;
1305 mddev->utime = sb->utime;
1306 mddev->level = sb->level;
1307 mddev->clevel[0] = 0;
1308 mddev->layout = sb->layout;
1309 mddev->raid_disks = sb->raid_disks;
1310 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1311 mddev->events = ev1;
1312 mddev->bitmap_info.offset = 0;
1313 mddev->bitmap_info.space = 0;
1314 /* bitmap can use 60 K after the 4K superblocks */
1315 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1316 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1317 mddev->reshape_backwards = 0;
1319 if (mddev->minor_version >= 91) {
1320 mddev->reshape_position = sb->reshape_position;
1321 mddev->delta_disks = sb->delta_disks;
1322 mddev->new_level = sb->new_level;
1323 mddev->new_layout = sb->new_layout;
1324 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1325 if (mddev->delta_disks < 0)
1326 mddev->reshape_backwards = 1;
1328 mddev->reshape_position = MaxSector;
1329 mddev->delta_disks = 0;
1330 mddev->new_level = mddev->level;
1331 mddev->new_layout = mddev->layout;
1332 mddev->new_chunk_sectors = mddev->chunk_sectors;
1334 if (mddev->level == 0)
1337 if (sb->state & (1<<MD_SB_CLEAN))
1338 mddev->recovery_cp = MaxSector;
1340 if (sb->events_hi == sb->cp_events_hi &&
1341 sb->events_lo == sb->cp_events_lo) {
1342 mddev->recovery_cp = sb->recovery_cp;
1344 mddev->recovery_cp = 0;
1347 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1348 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1349 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1350 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1352 mddev->max_disks = MD_SB_DISKS;
1354 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1355 mddev->bitmap_info.file == NULL) {
1356 mddev->bitmap_info.offset =
1357 mddev->bitmap_info.default_offset;
1358 mddev->bitmap_info.space =
1359 mddev->bitmap_info.default_space;
1362 } else if (mddev->pers == NULL) {
1363 /* Insist on good event counter while assembling, except
1364 * for spares (which don't need an event count) */
1366 if (sb->disks[rdev->desc_nr].state & (
1367 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1368 if (ev1 < mddev->events)
1370 } else if (mddev->bitmap) {
1371 /* if adding to array with a bitmap, then we can accept an
1372 * older device ... but not too old.
1374 if (ev1 < mddev->bitmap->events_cleared)
1376 if (ev1 < mddev->events)
1377 set_bit(Bitmap_sync, &rdev->flags);
1379 if (ev1 < mddev->events)
1380 /* just a hot-add of a new device, leave raid_disk at -1 */
1384 if (mddev->level != LEVEL_MULTIPATH) {
1385 desc = sb->disks + rdev->desc_nr;
1387 if (desc->state & (1<<MD_DISK_FAULTY))
1388 set_bit(Faulty, &rdev->flags);
1389 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1390 desc->raid_disk < mddev->raid_disks */) {
1391 set_bit(In_sync, &rdev->flags);
1392 rdev->raid_disk = desc->raid_disk;
1393 rdev->saved_raid_disk = desc->raid_disk;
1394 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1395 /* active but not in sync implies recovery up to
1396 * reshape position. We don't know exactly where
1397 * that is, so set to zero for now */
1398 if (mddev->minor_version >= 91) {
1399 rdev->recovery_offset = 0;
1400 rdev->raid_disk = desc->raid_disk;
1403 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1404 set_bit(WriteMostly, &rdev->flags);
1405 if (desc->state & (1<<MD_DISK_FAILFAST))
1406 set_bit(FailFast, &rdev->flags);
1407 } else /* MULTIPATH are always insync */
1408 set_bit(In_sync, &rdev->flags);
1413 * sync_super for 0.90.0
1415 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1418 struct md_rdev *rdev2;
1419 int next_spare = mddev->raid_disks;
1421 /* make rdev->sb match mddev data..
1424 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1425 * 3/ any empty disks < next_spare become removed
1427 * disks[0] gets initialised to REMOVED because
1428 * we cannot be sure from other fields if it has
1429 * been initialised or not.
1432 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1434 rdev->sb_size = MD_SB_BYTES;
1436 sb = page_address(rdev->sb_page);
1438 memset(sb, 0, sizeof(*sb));
1440 sb->md_magic = MD_SB_MAGIC;
1441 sb->major_version = mddev->major_version;
1442 sb->patch_version = mddev->patch_version;
1443 sb->gvalid_words = 0; /* ignored */
1444 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1445 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1446 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1447 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1449 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1450 sb->level = mddev->level;
1451 sb->size = mddev->dev_sectors / 2;
1452 sb->raid_disks = mddev->raid_disks;
1453 sb->md_minor = mddev->md_minor;
1454 sb->not_persistent = 0;
1455 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1457 sb->events_hi = (mddev->events>>32);
1458 sb->events_lo = (u32)mddev->events;
1460 if (mddev->reshape_position == MaxSector)
1461 sb->minor_version = 90;
1463 sb->minor_version = 91;
1464 sb->reshape_position = mddev->reshape_position;
1465 sb->new_level = mddev->new_level;
1466 sb->delta_disks = mddev->delta_disks;
1467 sb->new_layout = mddev->new_layout;
1468 sb->new_chunk = mddev->new_chunk_sectors << 9;
1470 mddev->minor_version = sb->minor_version;
1473 sb->recovery_cp = mddev->recovery_cp;
1474 sb->cp_events_hi = (mddev->events>>32);
1475 sb->cp_events_lo = (u32)mddev->events;
1476 if (mddev->recovery_cp == MaxSector)
1477 sb->state = (1<< MD_SB_CLEAN);
1479 sb->recovery_cp = 0;
1481 sb->layout = mddev->layout;
1482 sb->chunk_size = mddev->chunk_sectors << 9;
1484 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1485 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1487 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1488 rdev_for_each(rdev2, mddev) {
1491 int is_active = test_bit(In_sync, &rdev2->flags);
1493 if (rdev2->raid_disk >= 0 &&
1494 sb->minor_version >= 91)
1495 /* we have nowhere to store the recovery_offset,
1496 * but if it is not below the reshape_position,
1497 * we can piggy-back on that.
1500 if (rdev2->raid_disk < 0 ||
1501 test_bit(Faulty, &rdev2->flags))
1504 desc_nr = rdev2->raid_disk;
1506 desc_nr = next_spare++;
1507 rdev2->desc_nr = desc_nr;
1508 d = &sb->disks[rdev2->desc_nr];
1510 d->number = rdev2->desc_nr;
1511 d->major = MAJOR(rdev2->bdev->bd_dev);
1512 d->minor = MINOR(rdev2->bdev->bd_dev);
1514 d->raid_disk = rdev2->raid_disk;
1516 d->raid_disk = rdev2->desc_nr; /* compatibility */
1517 if (test_bit(Faulty, &rdev2->flags))
1518 d->state = (1<<MD_DISK_FAULTY);
1519 else if (is_active) {
1520 d->state = (1<<MD_DISK_ACTIVE);
1521 if (test_bit(In_sync, &rdev2->flags))
1522 d->state |= (1<<MD_DISK_SYNC);
1530 if (test_bit(WriteMostly, &rdev2->flags))
1531 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1532 if (test_bit(FailFast, &rdev2->flags))
1533 d->state |= (1<<MD_DISK_FAILFAST);
1535 /* now set the "removed" and "faulty" bits on any missing devices */
1536 for (i=0 ; i < mddev->raid_disks ; i++) {
1537 mdp_disk_t *d = &sb->disks[i];
1538 if (d->state == 0 && d->number == 0) {
1541 d->state = (1<<MD_DISK_REMOVED);
1542 d->state |= (1<<MD_DISK_FAULTY);
1546 sb->nr_disks = nr_disks;
1547 sb->active_disks = active;
1548 sb->working_disks = working;
1549 sb->failed_disks = failed;
1550 sb->spare_disks = spare;
1552 sb->this_disk = sb->disks[rdev->desc_nr];
1553 sb->sb_csum = calc_sb_csum(sb);
1557 * rdev_size_change for 0.90.0
1559 static unsigned long long
1560 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1562 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1563 return 0; /* component must fit device */
1564 if (rdev->mddev->bitmap_info.offset)
1565 return 0; /* can't move bitmap */
1566 rdev->sb_start = calc_dev_sboffset(rdev);
1567 if (!num_sectors || num_sectors > rdev->sb_start)
1568 num_sectors = rdev->sb_start;
1569 /* Limit to 4TB as metadata cannot record more than that.
1570 * 4TB == 2^32 KB, or 2*2^32 sectors.
1572 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1573 num_sectors = (sector_t)(2ULL << 32) - 2;
1575 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1577 } while (md_super_wait(rdev->mddev) < 0);
1582 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1584 /* non-zero offset changes not possible with v0.90 */
1585 return new_offset == 0;
1589 * version 1 superblock
1592 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1596 unsigned long long newcsum;
1597 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1598 __le32 *isuper = (__le32*)sb;
1600 disk_csum = sb->sb_csum;
1603 for (; size >= 4; size -= 4)
1604 newcsum += le32_to_cpu(*isuper++);
1607 newcsum += le16_to_cpu(*(__le16*) isuper);
1609 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1610 sb->sb_csum = disk_csum;
1611 return cpu_to_le32(csum);
1614 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1616 struct mdp_superblock_1 *sb;
1620 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1622 bool spare_disk = true;
1625 * Calculate the position of the superblock in 512byte sectors.
1626 * It is always aligned to a 4K boundary and
1627 * depeding on minor_version, it can be:
1628 * 0: At least 8K, but less than 12K, from end of device
1629 * 1: At start of device
1630 * 2: 4K from start of device.
1632 switch(minor_version) {
1634 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1636 sb_start &= ~(sector_t)(4*2-1);
1647 rdev->sb_start = sb_start;
1649 /* superblock is rarely larger than 1K, but it can be larger,
1650 * and it is safe to read 4k, so we do that
1652 ret = read_disk_sb(rdev, 4096);
1653 if (ret) return ret;
1655 sb = page_address(rdev->sb_page);
1657 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1658 sb->major_version != cpu_to_le32(1) ||
1659 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1660 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1661 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1664 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1665 pr_warn("md: invalid superblock checksum on %s\n",
1666 bdevname(rdev->bdev,b));
1669 if (le64_to_cpu(sb->data_size) < 10) {
1670 pr_warn("md: data_size too small on %s\n",
1671 bdevname(rdev->bdev,b));
1676 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1677 /* Some padding is non-zero, might be a new feature */
1680 rdev->preferred_minor = 0xffff;
1681 rdev->data_offset = le64_to_cpu(sb->data_offset);
1682 rdev->new_data_offset = rdev->data_offset;
1683 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1684 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1685 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1686 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1688 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1689 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1690 if (rdev->sb_size & bmask)
1691 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1694 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1697 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1700 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1703 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1705 if (!rdev->bb_page) {
1706 rdev->bb_page = alloc_page(GFP_KERNEL);
1710 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1711 rdev->badblocks.count == 0) {
1712 /* need to load the bad block list.
1713 * Currently we limit it to one page.
1719 int sectors = le16_to_cpu(sb->bblog_size);
1720 if (sectors > (PAGE_SIZE / 512))
1722 offset = le32_to_cpu(sb->bblog_offset);
1725 bb_sector = (long long)offset;
1726 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1727 rdev->bb_page, REQ_OP_READ, 0, true))
1729 bbp = (__le64 *)page_address(rdev->bb_page);
1730 rdev->badblocks.shift = sb->bblog_shift;
1731 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1732 u64 bb = le64_to_cpu(*bbp);
1733 int count = bb & (0x3ff);
1734 u64 sector = bb >> 10;
1735 sector <<= sb->bblog_shift;
1736 count <<= sb->bblog_shift;
1739 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1742 } else if (sb->bblog_offset != 0)
1743 rdev->badblocks.shift = 0;
1745 if ((le32_to_cpu(sb->feature_map) &
1746 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1747 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1748 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1749 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1752 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1756 /* not spare disk, or LEVEL_MULTIPATH */
1757 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1758 (rdev->desc_nr >= 0 &&
1759 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1760 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1761 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1771 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1773 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1774 sb->level != refsb->level ||
1775 sb->layout != refsb->layout ||
1776 sb->chunksize != refsb->chunksize) {
1777 pr_warn("md: %s has strangely different superblock to %s\n",
1778 bdevname(rdev->bdev,b),
1779 bdevname(refdev->bdev,b2));
1782 ev1 = le64_to_cpu(sb->events);
1783 ev2 = le64_to_cpu(refsb->events);
1785 if (!spare_disk && ev1 > ev2)
1790 if (minor_version) {
1791 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1792 sectors -= rdev->data_offset;
1794 sectors = rdev->sb_start;
1795 if (sectors < le64_to_cpu(sb->data_size))
1797 rdev->sectors = le64_to_cpu(sb->data_size);
1801 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1803 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1804 __u64 ev1 = le64_to_cpu(sb->events);
1806 rdev->raid_disk = -1;
1807 clear_bit(Faulty, &rdev->flags);
1808 clear_bit(In_sync, &rdev->flags);
1809 clear_bit(Bitmap_sync, &rdev->flags);
1810 clear_bit(WriteMostly, &rdev->flags);
1812 if (mddev->raid_disks == 0) {
1813 mddev->major_version = 1;
1814 mddev->patch_version = 0;
1815 mddev->external = 0;
1816 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1817 mddev->ctime = le64_to_cpu(sb->ctime);
1818 mddev->utime = le64_to_cpu(sb->utime);
1819 mddev->level = le32_to_cpu(sb->level);
1820 mddev->clevel[0] = 0;
1821 mddev->layout = le32_to_cpu(sb->layout);
1822 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1823 mddev->dev_sectors = le64_to_cpu(sb->size);
1824 mddev->events = ev1;
1825 mddev->bitmap_info.offset = 0;
1826 mddev->bitmap_info.space = 0;
1827 /* Default location for bitmap is 1K after superblock
1828 * using 3K - total of 4K
1830 mddev->bitmap_info.default_offset = 1024 >> 9;
1831 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1832 mddev->reshape_backwards = 0;
1834 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1835 memcpy(mddev->uuid, sb->set_uuid, 16);
1837 mddev->max_disks = (4096-256)/2;
1839 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1840 mddev->bitmap_info.file == NULL) {
1841 mddev->bitmap_info.offset =
1842 (__s32)le32_to_cpu(sb->bitmap_offset);
1843 /* Metadata doesn't record how much space is available.
1844 * For 1.0, we assume we can use up to the superblock
1845 * if before, else to 4K beyond superblock.
1846 * For others, assume no change is possible.
1848 if (mddev->minor_version > 0)
1849 mddev->bitmap_info.space = 0;
1850 else if (mddev->bitmap_info.offset > 0)
1851 mddev->bitmap_info.space =
1852 8 - mddev->bitmap_info.offset;
1854 mddev->bitmap_info.space =
1855 -mddev->bitmap_info.offset;
1858 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1859 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1860 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1861 mddev->new_level = le32_to_cpu(sb->new_level);
1862 mddev->new_layout = le32_to_cpu(sb->new_layout);
1863 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1864 if (mddev->delta_disks < 0 ||
1865 (mddev->delta_disks == 0 &&
1866 (le32_to_cpu(sb->feature_map)
1867 & MD_FEATURE_RESHAPE_BACKWARDS)))
1868 mddev->reshape_backwards = 1;
1870 mddev->reshape_position = MaxSector;
1871 mddev->delta_disks = 0;
1872 mddev->new_level = mddev->level;
1873 mddev->new_layout = mddev->layout;
1874 mddev->new_chunk_sectors = mddev->chunk_sectors;
1877 if (mddev->level == 0 &&
1878 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1881 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1882 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1884 if (le32_to_cpu(sb->feature_map) &
1885 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1886 if (le32_to_cpu(sb->feature_map) &
1887 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1889 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1890 (le32_to_cpu(sb->feature_map) &
1891 MD_FEATURE_MULTIPLE_PPLS))
1893 set_bit(MD_HAS_PPL, &mddev->flags);
1895 } else if (mddev->pers == NULL) {
1896 /* Insist of good event counter while assembling, except for
1897 * spares (which don't need an event count) */
1899 if (rdev->desc_nr >= 0 &&
1900 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1901 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1902 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1903 if (ev1 < mddev->events)
1905 } else if (mddev->bitmap) {
1906 /* If adding to array with a bitmap, then we can accept an
1907 * older device, but not too old.
1909 if (ev1 < mddev->bitmap->events_cleared)
1911 if (ev1 < mddev->events)
1912 set_bit(Bitmap_sync, &rdev->flags);
1914 if (ev1 < mddev->events)
1915 /* just a hot-add of a new device, leave raid_disk at -1 */
1918 if (mddev->level != LEVEL_MULTIPATH) {
1920 if (rdev->desc_nr < 0 ||
1921 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1922 role = MD_DISK_ROLE_SPARE;
1925 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1927 case MD_DISK_ROLE_SPARE: /* spare */
1929 case MD_DISK_ROLE_FAULTY: /* faulty */
1930 set_bit(Faulty, &rdev->flags);
1932 case MD_DISK_ROLE_JOURNAL: /* journal device */
1933 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1934 /* journal device without journal feature */
1935 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1938 set_bit(Journal, &rdev->flags);
1939 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1940 rdev->raid_disk = 0;
1943 rdev->saved_raid_disk = role;
1944 if ((le32_to_cpu(sb->feature_map) &
1945 MD_FEATURE_RECOVERY_OFFSET)) {
1946 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1947 if (!(le32_to_cpu(sb->feature_map) &
1948 MD_FEATURE_RECOVERY_BITMAP))
1949 rdev->saved_raid_disk = -1;
1952 * If the array is FROZEN, then the device can't
1953 * be in_sync with rest of array.
1955 if (!test_bit(MD_RECOVERY_FROZEN,
1957 set_bit(In_sync, &rdev->flags);
1959 rdev->raid_disk = role;
1962 if (sb->devflags & WriteMostly1)
1963 set_bit(WriteMostly, &rdev->flags);
1964 if (sb->devflags & FailFast1)
1965 set_bit(FailFast, &rdev->flags);
1966 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1967 set_bit(Replacement, &rdev->flags);
1968 } else /* MULTIPATH are always insync */
1969 set_bit(In_sync, &rdev->flags);
1974 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1976 struct mdp_superblock_1 *sb;
1977 struct md_rdev *rdev2;
1979 /* make rdev->sb match mddev and rdev data. */
1981 sb = page_address(rdev->sb_page);
1983 sb->feature_map = 0;
1985 sb->recovery_offset = cpu_to_le64(0);
1986 memset(sb->pad3, 0, sizeof(sb->pad3));
1988 sb->utime = cpu_to_le64((__u64)mddev->utime);
1989 sb->events = cpu_to_le64(mddev->events);
1991 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1992 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1993 sb->resync_offset = cpu_to_le64(MaxSector);
1995 sb->resync_offset = cpu_to_le64(0);
1997 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1999 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2000 sb->size = cpu_to_le64(mddev->dev_sectors);
2001 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2002 sb->level = cpu_to_le32(mddev->level);
2003 sb->layout = cpu_to_le32(mddev->layout);
2004 if (test_bit(FailFast, &rdev->flags))
2005 sb->devflags |= FailFast1;
2007 sb->devflags &= ~FailFast1;
2009 if (test_bit(WriteMostly, &rdev->flags))
2010 sb->devflags |= WriteMostly1;
2012 sb->devflags &= ~WriteMostly1;
2013 sb->data_offset = cpu_to_le64(rdev->data_offset);
2014 sb->data_size = cpu_to_le64(rdev->sectors);
2016 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2017 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2018 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2021 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2022 !test_bit(In_sync, &rdev->flags)) {
2024 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2025 sb->recovery_offset =
2026 cpu_to_le64(rdev->recovery_offset);
2027 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2029 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2031 /* Note: recovery_offset and journal_tail share space */
2032 if (test_bit(Journal, &rdev->flags))
2033 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2034 if (test_bit(Replacement, &rdev->flags))
2036 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2038 if (mddev->reshape_position != MaxSector) {
2039 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2040 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2041 sb->new_layout = cpu_to_le32(mddev->new_layout);
2042 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2043 sb->new_level = cpu_to_le32(mddev->new_level);
2044 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2045 if (mddev->delta_disks == 0 &&
2046 mddev->reshape_backwards)
2048 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2049 if (rdev->new_data_offset != rdev->data_offset) {
2051 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2052 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2053 - rdev->data_offset));
2057 if (mddev_is_clustered(mddev))
2058 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2060 if (rdev->badblocks.count == 0)
2061 /* Nothing to do for bad blocks*/ ;
2062 else if (sb->bblog_offset == 0)
2063 /* Cannot record bad blocks on this device */
2064 md_error(mddev, rdev);
2066 struct badblocks *bb = &rdev->badblocks;
2067 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2069 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2074 seq = read_seqbegin(&bb->lock);
2076 memset(bbp, 0xff, PAGE_SIZE);
2078 for (i = 0 ; i < bb->count ; i++) {
2079 u64 internal_bb = p[i];
2080 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2081 | BB_LEN(internal_bb));
2082 bbp[i] = cpu_to_le64(store_bb);
2085 if (read_seqretry(&bb->lock, seq))
2088 bb->sector = (rdev->sb_start +
2089 (int)le32_to_cpu(sb->bblog_offset));
2090 bb->size = le16_to_cpu(sb->bblog_size);
2095 rdev_for_each(rdev2, mddev)
2096 if (rdev2->desc_nr+1 > max_dev)
2097 max_dev = rdev2->desc_nr+1;
2099 if (max_dev > le32_to_cpu(sb->max_dev)) {
2101 sb->max_dev = cpu_to_le32(max_dev);
2102 rdev->sb_size = max_dev * 2 + 256;
2103 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2104 if (rdev->sb_size & bmask)
2105 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2107 max_dev = le32_to_cpu(sb->max_dev);
2109 for (i=0; i<max_dev;i++)
2110 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2112 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2113 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2115 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2116 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2118 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2120 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2121 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2122 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2125 rdev_for_each(rdev2, mddev) {
2127 if (test_bit(Faulty, &rdev2->flags))
2128 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2129 else if (test_bit(In_sync, &rdev2->flags))
2130 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2131 else if (test_bit(Journal, &rdev2->flags))
2132 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2133 else if (rdev2->raid_disk >= 0)
2134 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2136 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2139 sb->sb_csum = calc_sb_1_csum(sb);
2142 static sector_t super_1_choose_bm_space(sector_t dev_size)
2146 /* if the device is bigger than 8Gig, save 64k for bitmap
2147 * usage, if bigger than 200Gig, save 128k
2149 if (dev_size < 64*2)
2151 else if (dev_size - 64*2 >= 200*1024*1024*2)
2153 else if (dev_size - 4*2 > 8*1024*1024*2)
2160 static unsigned long long
2161 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2163 struct mdp_superblock_1 *sb;
2164 sector_t max_sectors;
2165 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2166 return 0; /* component must fit device */
2167 if (rdev->data_offset != rdev->new_data_offset)
2168 return 0; /* too confusing */
2169 if (rdev->sb_start < rdev->data_offset) {
2170 /* minor versions 1 and 2; superblock before data */
2171 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2172 max_sectors -= rdev->data_offset;
2173 if (!num_sectors || num_sectors > max_sectors)
2174 num_sectors = max_sectors;
2175 } else if (rdev->mddev->bitmap_info.offset) {
2176 /* minor version 0 with bitmap we can't move */
2179 /* minor version 0; superblock after data */
2180 sector_t sb_start, bm_space;
2181 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2183 /* 8K is for superblock */
2184 sb_start = dev_size - 8*2;
2185 sb_start &= ~(sector_t)(4*2 - 1);
2187 bm_space = super_1_choose_bm_space(dev_size);
2189 /* Space that can be used to store date needs to decrease
2190 * superblock bitmap space and bad block space(4K)
2192 max_sectors = sb_start - bm_space - 4*2;
2194 if (!num_sectors || num_sectors > max_sectors)
2195 num_sectors = max_sectors;
2197 sb = page_address(rdev->sb_page);
2198 sb->data_size = cpu_to_le64(num_sectors);
2199 sb->super_offset = cpu_to_le64(rdev->sb_start);
2200 sb->sb_csum = calc_sb_1_csum(sb);
2202 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2204 } while (md_super_wait(rdev->mddev) < 0);
2210 super_1_allow_new_offset(struct md_rdev *rdev,
2211 unsigned long long new_offset)
2213 /* All necessary checks on new >= old have been done */
2214 struct bitmap *bitmap;
2215 if (new_offset >= rdev->data_offset)
2218 /* with 1.0 metadata, there is no metadata to tread on
2219 * so we can always move back */
2220 if (rdev->mddev->minor_version == 0)
2223 /* otherwise we must be sure not to step on
2224 * any metadata, so stay:
2225 * 36K beyond start of superblock
2226 * beyond end of badblocks
2227 * beyond write-intent bitmap
2229 if (rdev->sb_start + (32+4)*2 > new_offset)
2231 bitmap = rdev->mddev->bitmap;
2232 if (bitmap && !rdev->mddev->bitmap_info.file &&
2233 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2234 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2236 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2242 static struct super_type super_types[] = {
2245 .owner = THIS_MODULE,
2246 .load_super = super_90_load,
2247 .validate_super = super_90_validate,
2248 .sync_super = super_90_sync,
2249 .rdev_size_change = super_90_rdev_size_change,
2250 .allow_new_offset = super_90_allow_new_offset,
2254 .owner = THIS_MODULE,
2255 .load_super = super_1_load,
2256 .validate_super = super_1_validate,
2257 .sync_super = super_1_sync,
2258 .rdev_size_change = super_1_rdev_size_change,
2259 .allow_new_offset = super_1_allow_new_offset,
2263 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2265 if (mddev->sync_super) {
2266 mddev->sync_super(mddev, rdev);
2270 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2272 super_types[mddev->major_version].sync_super(mddev, rdev);
2275 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2277 struct md_rdev *rdev, *rdev2;
2280 rdev_for_each_rcu(rdev, mddev1) {
2281 if (test_bit(Faulty, &rdev->flags) ||
2282 test_bit(Journal, &rdev->flags) ||
2283 rdev->raid_disk == -1)
2285 rdev_for_each_rcu(rdev2, mddev2) {
2286 if (test_bit(Faulty, &rdev2->flags) ||
2287 test_bit(Journal, &rdev2->flags) ||
2288 rdev2->raid_disk == -1)
2290 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2300 static LIST_HEAD(pending_raid_disks);
2303 * Try to register data integrity profile for an mddev
2305 * This is called when an array is started and after a disk has been kicked
2306 * from the array. It only succeeds if all working and active component devices
2307 * are integrity capable with matching profiles.
2309 int md_integrity_register(struct mddev *mddev)
2311 struct md_rdev *rdev, *reference = NULL;
2313 if (list_empty(&mddev->disks))
2314 return 0; /* nothing to do */
2315 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2316 return 0; /* shouldn't register, or already is */
2317 rdev_for_each(rdev, mddev) {
2318 /* skip spares and non-functional disks */
2319 if (test_bit(Faulty, &rdev->flags))
2321 if (rdev->raid_disk < 0)
2324 /* Use the first rdev as the reference */
2328 /* does this rdev's profile match the reference profile? */
2329 if (blk_integrity_compare(reference->bdev->bd_disk,
2330 rdev->bdev->bd_disk) < 0)
2333 if (!reference || !bdev_get_integrity(reference->bdev))
2336 * All component devices are integrity capable and have matching
2337 * profiles, register the common profile for the md device.
2339 blk_integrity_register(mddev->gendisk,
2340 bdev_get_integrity(reference->bdev));
2342 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2343 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2344 (mddev->level != 1 && mddev->level != 10 &&
2345 bioset_integrity_create(&mddev->io_acct_set, BIO_POOL_SIZE))) {
2347 * No need to handle the failure of bioset_integrity_create,
2348 * because the function is called by md_run() -> pers->run(),
2349 * md_run calls bioset_exit -> bioset_integrity_free in case
2352 pr_err("md: failed to create integrity pool for %s\n",
2358 EXPORT_SYMBOL(md_integrity_register);
2361 * Attempt to add an rdev, but only if it is consistent with the current
2364 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2366 struct blk_integrity *bi_mddev;
2367 char name[BDEVNAME_SIZE];
2369 if (!mddev->gendisk)
2372 bi_mddev = blk_get_integrity(mddev->gendisk);
2374 if (!bi_mddev) /* nothing to do */
2377 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2378 pr_err("%s: incompatible integrity profile for %s\n",
2379 mdname(mddev), bdevname(rdev->bdev, name));
2385 EXPORT_SYMBOL(md_integrity_add_rdev);
2387 static bool rdev_read_only(struct md_rdev *rdev)
2389 return bdev_read_only(rdev->bdev) ||
2390 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2393 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2395 char b[BDEVNAME_SIZE];
2398 /* prevent duplicates */
2399 if (find_rdev(mddev, rdev->bdev->bd_dev))
2402 if (rdev_read_only(rdev) && mddev->pers)
2405 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2406 if (!test_bit(Journal, &rdev->flags) &&
2408 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2410 /* Cannot change size, so fail
2411 * If mddev->level <= 0, then we don't care
2412 * about aligning sizes (e.g. linear)
2414 if (mddev->level > 0)
2417 mddev->dev_sectors = rdev->sectors;
2420 /* Verify rdev->desc_nr is unique.
2421 * If it is -1, assign a free number, else
2422 * check number is not in use
2425 if (rdev->desc_nr < 0) {
2428 choice = mddev->raid_disks;
2429 while (md_find_rdev_nr_rcu(mddev, choice))
2431 rdev->desc_nr = choice;
2433 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2439 if (!test_bit(Journal, &rdev->flags) &&
2440 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2441 pr_warn("md: %s: array is limited to %d devices\n",
2442 mdname(mddev), mddev->max_disks);
2445 bdevname(rdev->bdev,b);
2446 strreplace(b, '/', '!');
2448 rdev->mddev = mddev;
2449 pr_debug("md: bind<%s>\n", b);
2451 if (mddev->raid_disks)
2452 mddev_create_serial_pool(mddev, rdev, false);
2454 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2457 /* failure here is OK */
2458 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2459 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2460 rdev->sysfs_unack_badblocks =
2461 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2462 rdev->sysfs_badblocks =
2463 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2465 list_add_rcu(&rdev->same_set, &mddev->disks);
2466 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2468 /* May as well allow recovery to be retried once */
2469 mddev->recovery_disabled++;
2474 pr_warn("md: failed to register dev-%s for %s\n",
2479 static void rdev_delayed_delete(struct work_struct *ws)
2481 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2482 kobject_del(&rdev->kobj);
2483 kobject_put(&rdev->kobj);
2486 static void unbind_rdev_from_array(struct md_rdev *rdev)
2488 char b[BDEVNAME_SIZE];
2490 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2491 list_del_rcu(&rdev->same_set);
2492 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2493 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2495 sysfs_remove_link(&rdev->kobj, "block");
2496 sysfs_put(rdev->sysfs_state);
2497 sysfs_put(rdev->sysfs_unack_badblocks);
2498 sysfs_put(rdev->sysfs_badblocks);
2499 rdev->sysfs_state = NULL;
2500 rdev->sysfs_unack_badblocks = NULL;
2501 rdev->sysfs_badblocks = NULL;
2502 rdev->badblocks.count = 0;
2503 /* We need to delay this, otherwise we can deadlock when
2504 * writing to 'remove' to "dev/state". We also need
2505 * to delay it due to rcu usage.
2508 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2509 kobject_get(&rdev->kobj);
2510 queue_work(md_rdev_misc_wq, &rdev->del_work);
2514 * prevent the device from being mounted, repartitioned or
2515 * otherwise reused by a RAID array (or any other kernel
2516 * subsystem), by bd_claiming the device.
2518 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2521 struct block_device *bdev;
2523 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2524 shared ? (struct md_rdev *)lock_rdev : rdev);
2526 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2527 MAJOR(dev), MINOR(dev));
2528 return PTR_ERR(bdev);
2534 static void unlock_rdev(struct md_rdev *rdev)
2536 struct block_device *bdev = rdev->bdev;
2538 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2541 void md_autodetect_dev(dev_t dev);
2543 static void export_rdev(struct md_rdev *rdev)
2545 char b[BDEVNAME_SIZE];
2547 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2548 md_rdev_clear(rdev);
2550 if (test_bit(AutoDetected, &rdev->flags))
2551 md_autodetect_dev(rdev->bdev->bd_dev);
2554 kobject_put(&rdev->kobj);
2557 void md_kick_rdev_from_array(struct md_rdev *rdev)
2559 unbind_rdev_from_array(rdev);
2562 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2564 static void export_array(struct mddev *mddev)
2566 struct md_rdev *rdev;
2568 while (!list_empty(&mddev->disks)) {
2569 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2571 md_kick_rdev_from_array(rdev);
2573 mddev->raid_disks = 0;
2574 mddev->major_version = 0;
2577 static bool set_in_sync(struct mddev *mddev)
2579 lockdep_assert_held(&mddev->lock);
2580 if (!mddev->in_sync) {
2581 mddev->sync_checkers++;
2582 spin_unlock(&mddev->lock);
2583 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2584 spin_lock(&mddev->lock);
2585 if (!mddev->in_sync &&
2586 percpu_ref_is_zero(&mddev->writes_pending)) {
2589 * Ensure ->in_sync is visible before we clear
2593 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2594 sysfs_notify_dirent_safe(mddev->sysfs_state);
2596 if (--mddev->sync_checkers == 0)
2597 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2599 if (mddev->safemode == 1)
2600 mddev->safemode = 0;
2601 return mddev->in_sync;
2604 static void sync_sbs(struct mddev *mddev, int nospares)
2606 /* Update each superblock (in-memory image), but
2607 * if we are allowed to, skip spares which already
2608 * have the right event counter, or have one earlier
2609 * (which would mean they aren't being marked as dirty
2610 * with the rest of the array)
2612 struct md_rdev *rdev;
2613 rdev_for_each(rdev, mddev) {
2614 if (rdev->sb_events == mddev->events ||
2616 rdev->raid_disk < 0 &&
2617 rdev->sb_events+1 == mddev->events)) {
2618 /* Don't update this superblock */
2619 rdev->sb_loaded = 2;
2621 sync_super(mddev, rdev);
2622 rdev->sb_loaded = 1;
2627 static bool does_sb_need_changing(struct mddev *mddev)
2629 struct md_rdev *rdev;
2630 struct mdp_superblock_1 *sb;
2633 /* Find a good rdev */
2634 rdev_for_each(rdev, mddev)
2635 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2638 /* No good device found. */
2642 sb = page_address(rdev->sb_page);
2643 /* Check if a device has become faulty or a spare become active */
2644 rdev_for_each(rdev, mddev) {
2645 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2646 /* Device activated? */
2647 if (role == 0xffff && rdev->raid_disk >=0 &&
2648 !test_bit(Faulty, &rdev->flags))
2650 /* Device turned faulty? */
2651 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2655 /* Check if any mddev parameters have changed */
2656 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2657 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2658 (mddev->layout != le32_to_cpu(sb->layout)) ||
2659 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2660 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2666 void md_update_sb(struct mddev *mddev, int force_change)
2668 struct md_rdev *rdev;
2671 int any_badblocks_changed = 0;
2676 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2681 if (mddev_is_clustered(mddev)) {
2682 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2684 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2686 ret = md_cluster_ops->metadata_update_start(mddev);
2687 /* Has someone else has updated the sb */
2688 if (!does_sb_need_changing(mddev)) {
2690 md_cluster_ops->metadata_update_cancel(mddev);
2691 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2692 BIT(MD_SB_CHANGE_DEVS) |
2693 BIT(MD_SB_CHANGE_CLEAN));
2699 * First make sure individual recovery_offsets are correct
2700 * curr_resync_completed can only be used during recovery.
2701 * During reshape/resync it might use array-addresses rather
2702 * that device addresses.
2704 rdev_for_each(rdev, mddev) {
2705 if (rdev->raid_disk >= 0 &&
2706 mddev->delta_disks >= 0 &&
2707 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2708 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2709 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2710 !test_bit(Journal, &rdev->flags) &&
2711 !test_bit(In_sync, &rdev->flags) &&
2712 mddev->curr_resync_completed > rdev->recovery_offset)
2713 rdev->recovery_offset = mddev->curr_resync_completed;
2716 if (!mddev->persistent) {
2717 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2718 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2719 if (!mddev->external) {
2720 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2721 rdev_for_each(rdev, mddev) {
2722 if (rdev->badblocks.changed) {
2723 rdev->badblocks.changed = 0;
2724 ack_all_badblocks(&rdev->badblocks);
2725 md_error(mddev, rdev);
2727 clear_bit(Blocked, &rdev->flags);
2728 clear_bit(BlockedBadBlocks, &rdev->flags);
2729 wake_up(&rdev->blocked_wait);
2732 wake_up(&mddev->sb_wait);
2736 spin_lock(&mddev->lock);
2738 mddev->utime = ktime_get_real_seconds();
2740 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2742 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2743 /* just a clean<-> dirty transition, possibly leave spares alone,
2744 * though if events isn't the right even/odd, we will have to do
2750 if (mddev->degraded)
2751 /* If the array is degraded, then skipping spares is both
2752 * dangerous and fairly pointless.
2753 * Dangerous because a device that was removed from the array
2754 * might have a event_count that still looks up-to-date,
2755 * so it can be re-added without a resync.
2756 * Pointless because if there are any spares to skip,
2757 * then a recovery will happen and soon that array won't
2758 * be degraded any more and the spare can go back to sleep then.
2762 sync_req = mddev->in_sync;
2764 /* If this is just a dirty<->clean transition, and the array is clean
2765 * and 'events' is odd, we can roll back to the previous clean state */
2767 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2768 && mddev->can_decrease_events
2769 && mddev->events != 1) {
2771 mddev->can_decrease_events = 0;
2773 /* otherwise we have to go forward and ... */
2775 mddev->can_decrease_events = nospares;
2779 * This 64-bit counter should never wrap.
2780 * Either we are in around ~1 trillion A.C., assuming
2781 * 1 reboot per second, or we have a bug...
2783 WARN_ON(mddev->events == 0);
2785 rdev_for_each(rdev, mddev) {
2786 if (rdev->badblocks.changed)
2787 any_badblocks_changed++;
2788 if (test_bit(Faulty, &rdev->flags))
2789 set_bit(FaultRecorded, &rdev->flags);
2792 sync_sbs(mddev, nospares);
2793 spin_unlock(&mddev->lock);
2795 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2796 mdname(mddev), mddev->in_sync);
2799 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2801 md_bitmap_update_sb(mddev->bitmap);
2802 rdev_for_each(rdev, mddev) {
2803 char b[BDEVNAME_SIZE];
2805 if (rdev->sb_loaded != 1)
2806 continue; /* no noise on spare devices */
2808 if (!test_bit(Faulty, &rdev->flags)) {
2809 md_super_write(mddev,rdev,
2810 rdev->sb_start, rdev->sb_size,
2812 pr_debug("md: (write) %s's sb offset: %llu\n",
2813 bdevname(rdev->bdev, b),
2814 (unsigned long long)rdev->sb_start);
2815 rdev->sb_events = mddev->events;
2816 if (rdev->badblocks.size) {
2817 md_super_write(mddev, rdev,
2818 rdev->badblocks.sector,
2819 rdev->badblocks.size << 9,
2821 rdev->badblocks.size = 0;
2825 pr_debug("md: %s (skipping faulty)\n",
2826 bdevname(rdev->bdev, b));
2828 if (mddev->level == LEVEL_MULTIPATH)
2829 /* only need to write one superblock... */
2832 if (md_super_wait(mddev) < 0)
2834 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2836 if (mddev_is_clustered(mddev) && ret == 0)
2837 md_cluster_ops->metadata_update_finish(mddev);
2839 if (mddev->in_sync != sync_req ||
2840 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2841 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2842 /* have to write it out again */
2844 wake_up(&mddev->sb_wait);
2845 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2846 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2848 rdev_for_each(rdev, mddev) {
2849 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2850 clear_bit(Blocked, &rdev->flags);
2852 if (any_badblocks_changed)
2853 ack_all_badblocks(&rdev->badblocks);
2854 clear_bit(BlockedBadBlocks, &rdev->flags);
2855 wake_up(&rdev->blocked_wait);
2858 EXPORT_SYMBOL(md_update_sb);
2860 static int add_bound_rdev(struct md_rdev *rdev)
2862 struct mddev *mddev = rdev->mddev;
2864 bool add_journal = test_bit(Journal, &rdev->flags);
2866 if (!mddev->pers->hot_remove_disk || add_journal) {
2867 /* If there is hot_add_disk but no hot_remove_disk
2868 * then added disks for geometry changes,
2869 * and should be added immediately.
2871 super_types[mddev->major_version].
2872 validate_super(mddev, rdev);
2874 mddev_suspend(mddev);
2875 err = mddev->pers->hot_add_disk(mddev, rdev);
2877 mddev_resume(mddev);
2879 md_kick_rdev_from_array(rdev);
2883 sysfs_notify_dirent_safe(rdev->sysfs_state);
2885 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2886 if (mddev->degraded)
2887 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2888 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2889 md_new_event(mddev);
2890 md_wakeup_thread(mddev->thread);
2894 /* words written to sysfs files may, or may not, be \n terminated.
2895 * We want to accept with case. For this we use cmd_match.
2897 static int cmd_match(const char *cmd, const char *str)
2899 /* See if cmd, written into a sysfs file, matches
2900 * str. They must either be the same, or cmd can
2901 * have a trailing newline
2903 while (*cmd && *str && *cmd == *str) {
2914 struct rdev_sysfs_entry {
2915 struct attribute attr;
2916 ssize_t (*show)(struct md_rdev *, char *);
2917 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2921 state_show(struct md_rdev *rdev, char *page)
2925 unsigned long flags = READ_ONCE(rdev->flags);
2927 if (test_bit(Faulty, &flags) ||
2928 (!test_bit(ExternalBbl, &flags) &&
2929 rdev->badblocks.unacked_exist))
2930 len += sprintf(page+len, "faulty%s", sep);
2931 if (test_bit(In_sync, &flags))
2932 len += sprintf(page+len, "in_sync%s", sep);
2933 if (test_bit(Journal, &flags))
2934 len += sprintf(page+len, "journal%s", sep);
2935 if (test_bit(WriteMostly, &flags))
2936 len += sprintf(page+len, "write_mostly%s", sep);
2937 if (test_bit(Blocked, &flags) ||
2938 (rdev->badblocks.unacked_exist
2939 && !test_bit(Faulty, &flags)))
2940 len += sprintf(page+len, "blocked%s", sep);
2941 if (!test_bit(Faulty, &flags) &&
2942 !test_bit(Journal, &flags) &&
2943 !test_bit(In_sync, &flags))
2944 len += sprintf(page+len, "spare%s", sep);
2945 if (test_bit(WriteErrorSeen, &flags))
2946 len += sprintf(page+len, "write_error%s", sep);
2947 if (test_bit(WantReplacement, &flags))
2948 len += sprintf(page+len, "want_replacement%s", sep);
2949 if (test_bit(Replacement, &flags))
2950 len += sprintf(page+len, "replacement%s", sep);
2951 if (test_bit(ExternalBbl, &flags))
2952 len += sprintf(page+len, "external_bbl%s", sep);
2953 if (test_bit(FailFast, &flags))
2954 len += sprintf(page+len, "failfast%s", sep);
2959 return len+sprintf(page+len, "\n");
2963 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2966 * faulty - simulates an error
2967 * remove - disconnects the device
2968 * writemostly - sets write_mostly
2969 * -writemostly - clears write_mostly
2970 * blocked - sets the Blocked flags
2971 * -blocked - clears the Blocked and possibly simulates an error
2972 * insync - sets Insync providing device isn't active
2973 * -insync - clear Insync for a device with a slot assigned,
2974 * so that it gets rebuilt based on bitmap
2975 * write_error - sets WriteErrorSeen
2976 * -write_error - clears WriteErrorSeen
2977 * {,-}failfast - set/clear FailFast
2980 struct mddev *mddev = rdev->mddev;
2982 bool need_update_sb = false;
2984 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2985 md_error(rdev->mddev, rdev);
2986 if (test_bit(Faulty, &rdev->flags))
2990 } else if (cmd_match(buf, "remove")) {
2991 if (rdev->mddev->pers) {
2992 clear_bit(Blocked, &rdev->flags);
2993 remove_and_add_spares(rdev->mddev, rdev);
2995 if (rdev->raid_disk >= 0)
2999 if (mddev_is_clustered(mddev))
3000 err = md_cluster_ops->remove_disk(mddev, rdev);
3003 md_kick_rdev_from_array(rdev);
3005 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3006 md_wakeup_thread(mddev->thread);
3008 md_new_event(mddev);
3011 } else if (cmd_match(buf, "writemostly")) {
3012 set_bit(WriteMostly, &rdev->flags);
3013 mddev_create_serial_pool(rdev->mddev, rdev, false);
3014 need_update_sb = true;
3016 } else if (cmd_match(buf, "-writemostly")) {
3017 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3018 clear_bit(WriteMostly, &rdev->flags);
3019 need_update_sb = true;
3021 } else if (cmd_match(buf, "blocked")) {
3022 set_bit(Blocked, &rdev->flags);
3024 } else if (cmd_match(buf, "-blocked")) {
3025 if (!test_bit(Faulty, &rdev->flags) &&
3026 !test_bit(ExternalBbl, &rdev->flags) &&
3027 rdev->badblocks.unacked_exist) {
3028 /* metadata handler doesn't understand badblocks,
3029 * so we need to fail the device
3031 md_error(rdev->mddev, rdev);
3033 clear_bit(Blocked, &rdev->flags);
3034 clear_bit(BlockedBadBlocks, &rdev->flags);
3035 wake_up(&rdev->blocked_wait);
3036 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3037 md_wakeup_thread(rdev->mddev->thread);
3040 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3041 set_bit(In_sync, &rdev->flags);
3043 } else if (cmd_match(buf, "failfast")) {
3044 set_bit(FailFast, &rdev->flags);
3045 need_update_sb = true;
3047 } else if (cmd_match(buf, "-failfast")) {
3048 clear_bit(FailFast, &rdev->flags);
3049 need_update_sb = true;
3051 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3052 !test_bit(Journal, &rdev->flags)) {
3053 if (rdev->mddev->pers == NULL) {
3054 clear_bit(In_sync, &rdev->flags);
3055 rdev->saved_raid_disk = rdev->raid_disk;
3056 rdev->raid_disk = -1;
3059 } else if (cmd_match(buf, "write_error")) {
3060 set_bit(WriteErrorSeen, &rdev->flags);
3062 } else if (cmd_match(buf, "-write_error")) {
3063 clear_bit(WriteErrorSeen, &rdev->flags);
3065 } else if (cmd_match(buf, "want_replacement")) {
3066 /* Any non-spare device that is not a replacement can
3067 * become want_replacement at any time, but we then need to
3068 * check if recovery is needed.
3070 if (rdev->raid_disk >= 0 &&
3071 !test_bit(Journal, &rdev->flags) &&
3072 !test_bit(Replacement, &rdev->flags))
3073 set_bit(WantReplacement, &rdev->flags);
3074 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3075 md_wakeup_thread(rdev->mddev->thread);
3077 } else if (cmd_match(buf, "-want_replacement")) {
3078 /* Clearing 'want_replacement' is always allowed.
3079 * Once replacements starts it is too late though.
3082 clear_bit(WantReplacement, &rdev->flags);
3083 } else if (cmd_match(buf, "replacement")) {
3084 /* Can only set a device as a replacement when array has not
3085 * yet been started. Once running, replacement is automatic
3086 * from spares, or by assigning 'slot'.
3088 if (rdev->mddev->pers)
3091 set_bit(Replacement, &rdev->flags);
3094 } else if (cmd_match(buf, "-replacement")) {
3095 /* Similarly, can only clear Replacement before start */
3096 if (rdev->mddev->pers)
3099 clear_bit(Replacement, &rdev->flags);
3102 } else if (cmd_match(buf, "re-add")) {
3103 if (!rdev->mddev->pers)
3105 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3106 rdev->saved_raid_disk >= 0) {
3107 /* clear_bit is performed _after_ all the devices
3108 * have their local Faulty bit cleared. If any writes
3109 * happen in the meantime in the local node, they
3110 * will land in the local bitmap, which will be synced
3111 * by this node eventually
3113 if (!mddev_is_clustered(rdev->mddev) ||
3114 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3115 clear_bit(Faulty, &rdev->flags);
3116 err = add_bound_rdev(rdev);
3120 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3121 set_bit(ExternalBbl, &rdev->flags);
3122 rdev->badblocks.shift = 0;
3124 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3125 clear_bit(ExternalBbl, &rdev->flags);
3129 md_update_sb(mddev, 1);
3131 sysfs_notify_dirent_safe(rdev->sysfs_state);
3132 return err ? err : len;
3134 static struct rdev_sysfs_entry rdev_state =
3135 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3138 errors_show(struct md_rdev *rdev, char *page)
3140 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3144 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3149 rv = kstrtouint(buf, 10, &n);
3152 atomic_set(&rdev->corrected_errors, n);
3155 static struct rdev_sysfs_entry rdev_errors =
3156 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3159 slot_show(struct md_rdev *rdev, char *page)
3161 if (test_bit(Journal, &rdev->flags))
3162 return sprintf(page, "journal\n");
3163 else if (rdev->raid_disk < 0)
3164 return sprintf(page, "none\n");
3166 return sprintf(page, "%d\n", rdev->raid_disk);
3170 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3175 if (test_bit(Journal, &rdev->flags))
3177 if (strncmp(buf, "none", 4)==0)
3180 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3184 if (rdev->mddev->pers && slot == -1) {
3185 /* Setting 'slot' on an active array requires also
3186 * updating the 'rd%d' link, and communicating
3187 * with the personality with ->hot_*_disk.
3188 * For now we only support removing
3189 * failed/spare devices. This normally happens automatically,
3190 * but not when the metadata is externally managed.
3192 if (rdev->raid_disk == -1)
3194 /* personality does all needed checks */
3195 if (rdev->mddev->pers->hot_remove_disk == NULL)
3197 clear_bit(Blocked, &rdev->flags);
3198 remove_and_add_spares(rdev->mddev, rdev);
3199 if (rdev->raid_disk >= 0)
3201 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3202 md_wakeup_thread(rdev->mddev->thread);
3203 } else if (rdev->mddev->pers) {
3204 /* Activating a spare .. or possibly reactivating
3205 * if we ever get bitmaps working here.
3209 if (rdev->raid_disk != -1)
3212 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3215 if (rdev->mddev->pers->hot_add_disk == NULL)
3218 if (slot >= rdev->mddev->raid_disks &&
3219 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3222 rdev->raid_disk = slot;
3223 if (test_bit(In_sync, &rdev->flags))
3224 rdev->saved_raid_disk = slot;
3226 rdev->saved_raid_disk = -1;
3227 clear_bit(In_sync, &rdev->flags);
3228 clear_bit(Bitmap_sync, &rdev->flags);
3229 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3231 rdev->raid_disk = -1;
3234 sysfs_notify_dirent_safe(rdev->sysfs_state);
3235 /* failure here is OK */;
3236 sysfs_link_rdev(rdev->mddev, rdev);
3237 /* don't wakeup anyone, leave that to userspace. */
3239 if (slot >= rdev->mddev->raid_disks &&
3240 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3242 rdev->raid_disk = slot;
3243 /* assume it is working */
3244 clear_bit(Faulty, &rdev->flags);
3245 clear_bit(WriteMostly, &rdev->flags);
3246 set_bit(In_sync, &rdev->flags);
3247 sysfs_notify_dirent_safe(rdev->sysfs_state);
3252 static struct rdev_sysfs_entry rdev_slot =
3253 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3256 offset_show(struct md_rdev *rdev, char *page)
3258 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3262 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3264 unsigned long long offset;
3265 if (kstrtoull(buf, 10, &offset) < 0)
3267 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3269 if (rdev->sectors && rdev->mddev->external)
3270 /* Must set offset before size, so overlap checks
3273 rdev->data_offset = offset;
3274 rdev->new_data_offset = offset;
3278 static struct rdev_sysfs_entry rdev_offset =
3279 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3281 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3283 return sprintf(page, "%llu\n",
3284 (unsigned long long)rdev->new_data_offset);
3287 static ssize_t new_offset_store(struct md_rdev *rdev,
3288 const char *buf, size_t len)
3290 unsigned long long new_offset;
3291 struct mddev *mddev = rdev->mddev;
3293 if (kstrtoull(buf, 10, &new_offset) < 0)
3296 if (mddev->sync_thread ||
3297 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3299 if (new_offset == rdev->data_offset)
3300 /* reset is always permitted */
3302 else if (new_offset > rdev->data_offset) {
3303 /* must not push array size beyond rdev_sectors */
3304 if (new_offset - rdev->data_offset
3305 + mddev->dev_sectors > rdev->sectors)
3308 /* Metadata worries about other space details. */
3310 /* decreasing the offset is inconsistent with a backwards
3313 if (new_offset < rdev->data_offset &&
3314 mddev->reshape_backwards)
3316 /* Increasing offset is inconsistent with forwards
3317 * reshape. reshape_direction should be set to
3318 * 'backwards' first.
3320 if (new_offset > rdev->data_offset &&
3321 !mddev->reshape_backwards)
3324 if (mddev->pers && mddev->persistent &&
3325 !super_types[mddev->major_version]
3326 .allow_new_offset(rdev, new_offset))
3328 rdev->new_data_offset = new_offset;
3329 if (new_offset > rdev->data_offset)
3330 mddev->reshape_backwards = 1;
3331 else if (new_offset < rdev->data_offset)
3332 mddev->reshape_backwards = 0;
3336 static struct rdev_sysfs_entry rdev_new_offset =
3337 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3340 rdev_size_show(struct md_rdev *rdev, char *page)
3342 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3345 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3347 /* check if two start/length pairs overlap */
3355 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3357 unsigned long long blocks;
3360 if (kstrtoull(buf, 10, &blocks) < 0)
3363 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3364 return -EINVAL; /* sector conversion overflow */
3367 if (new != blocks * 2)
3368 return -EINVAL; /* unsigned long long to sector_t overflow */
3375 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3377 struct mddev *my_mddev = rdev->mddev;
3378 sector_t oldsectors = rdev->sectors;
3381 if (test_bit(Journal, &rdev->flags))
3383 if (strict_blocks_to_sectors(buf, §ors) < 0)
3385 if (rdev->data_offset != rdev->new_data_offset)
3386 return -EINVAL; /* too confusing */
3387 if (my_mddev->pers && rdev->raid_disk >= 0) {
3388 if (my_mddev->persistent) {
3389 sectors = super_types[my_mddev->major_version].
3390 rdev_size_change(rdev, sectors);
3393 } else if (!sectors)
3394 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3396 if (!my_mddev->pers->resize)
3397 /* Cannot change size for RAID0 or Linear etc */
3400 if (sectors < my_mddev->dev_sectors)
3401 return -EINVAL; /* component must fit device */
3403 rdev->sectors = sectors;
3404 if (sectors > oldsectors && my_mddev->external) {
3405 /* Need to check that all other rdevs with the same
3406 * ->bdev do not overlap. 'rcu' is sufficient to walk
3407 * the rdev lists safely.
3408 * This check does not provide a hard guarantee, it
3409 * just helps avoid dangerous mistakes.
3411 struct mddev *mddev;
3413 struct list_head *tmp;
3416 for_each_mddev(mddev, tmp) {
3417 struct md_rdev *rdev2;
3419 rdev_for_each(rdev2, mddev)
3420 if (rdev->bdev == rdev2->bdev &&
3422 overlaps(rdev->data_offset, rdev->sectors,
3435 /* Someone else could have slipped in a size
3436 * change here, but doing so is just silly.
3437 * We put oldsectors back because we *know* it is
3438 * safe, and trust userspace not to race with
3441 rdev->sectors = oldsectors;
3448 static struct rdev_sysfs_entry rdev_size =
3449 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3451 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3453 unsigned long long recovery_start = rdev->recovery_offset;
3455 if (test_bit(In_sync, &rdev->flags) ||
3456 recovery_start == MaxSector)
3457 return sprintf(page, "none\n");
3459 return sprintf(page, "%llu\n", recovery_start);
3462 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3464 unsigned long long recovery_start;
3466 if (cmd_match(buf, "none"))
3467 recovery_start = MaxSector;
3468 else if (kstrtoull(buf, 10, &recovery_start))
3471 if (rdev->mddev->pers &&
3472 rdev->raid_disk >= 0)
3475 rdev->recovery_offset = recovery_start;
3476 if (recovery_start == MaxSector)
3477 set_bit(In_sync, &rdev->flags);
3479 clear_bit(In_sync, &rdev->flags);
3483 static struct rdev_sysfs_entry rdev_recovery_start =
3484 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3486 /* sysfs access to bad-blocks list.
3487 * We present two files.
3488 * 'bad-blocks' lists sector numbers and lengths of ranges that
3489 * are recorded as bad. The list is truncated to fit within
3490 * the one-page limit of sysfs.
3491 * Writing "sector length" to this file adds an acknowledged
3493 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3494 * been acknowledged. Writing to this file adds bad blocks
3495 * without acknowledging them. This is largely for testing.
3497 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3499 return badblocks_show(&rdev->badblocks, page, 0);
3501 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3503 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3504 /* Maybe that ack was all we needed */
3505 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3506 wake_up(&rdev->blocked_wait);
3509 static struct rdev_sysfs_entry rdev_bad_blocks =
3510 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3512 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3514 return badblocks_show(&rdev->badblocks, page, 1);
3516 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3518 return badblocks_store(&rdev->badblocks, page, len, 1);
3520 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3521 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3524 ppl_sector_show(struct md_rdev *rdev, char *page)
3526 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3530 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3532 unsigned long long sector;
3534 if (kstrtoull(buf, 10, §or) < 0)
3536 if (sector != (sector_t)sector)
3539 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3540 rdev->raid_disk >= 0)
3543 if (rdev->mddev->persistent) {
3544 if (rdev->mddev->major_version == 0)
3546 if ((sector > rdev->sb_start &&
3547 sector - rdev->sb_start > S16_MAX) ||
3548 (sector < rdev->sb_start &&
3549 rdev->sb_start - sector > -S16_MIN))
3551 rdev->ppl.offset = sector - rdev->sb_start;
3552 } else if (!rdev->mddev->external) {
3555 rdev->ppl.sector = sector;
3559 static struct rdev_sysfs_entry rdev_ppl_sector =
3560 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3563 ppl_size_show(struct md_rdev *rdev, char *page)
3565 return sprintf(page, "%u\n", rdev->ppl.size);
3569 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3573 if (kstrtouint(buf, 10, &size) < 0)
3576 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3577 rdev->raid_disk >= 0)
3580 if (rdev->mddev->persistent) {
3581 if (rdev->mddev->major_version == 0)
3585 } else if (!rdev->mddev->external) {
3588 rdev->ppl.size = size;
3592 static struct rdev_sysfs_entry rdev_ppl_size =
3593 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3595 static struct attribute *rdev_default_attrs[] = {
3600 &rdev_new_offset.attr,
3602 &rdev_recovery_start.attr,
3603 &rdev_bad_blocks.attr,
3604 &rdev_unack_bad_blocks.attr,
3605 &rdev_ppl_sector.attr,
3606 &rdev_ppl_size.attr,
3610 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3612 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3613 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3619 return entry->show(rdev, page);
3623 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3624 const char *page, size_t length)
3626 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3627 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3629 struct mddev *mddev = rdev->mddev;
3633 if (!capable(CAP_SYS_ADMIN))
3635 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3637 if (rdev->mddev == NULL)
3640 rv = entry->store(rdev, page, length);
3641 mddev_unlock(mddev);
3646 static void rdev_free(struct kobject *ko)
3648 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3651 static const struct sysfs_ops rdev_sysfs_ops = {
3652 .show = rdev_attr_show,
3653 .store = rdev_attr_store,
3655 static struct kobj_type rdev_ktype = {
3656 .release = rdev_free,
3657 .sysfs_ops = &rdev_sysfs_ops,
3658 .default_attrs = rdev_default_attrs,
3661 int md_rdev_init(struct md_rdev *rdev)
3664 rdev->saved_raid_disk = -1;
3665 rdev->raid_disk = -1;
3667 rdev->data_offset = 0;
3668 rdev->new_data_offset = 0;
3669 rdev->sb_events = 0;
3670 rdev->last_read_error = 0;
3671 rdev->sb_loaded = 0;
3672 rdev->bb_page = NULL;
3673 atomic_set(&rdev->nr_pending, 0);
3674 atomic_set(&rdev->read_errors, 0);
3675 atomic_set(&rdev->corrected_errors, 0);
3677 INIT_LIST_HEAD(&rdev->same_set);
3678 init_waitqueue_head(&rdev->blocked_wait);
3680 /* Add space to store bad block list.
3681 * This reserves the space even on arrays where it cannot
3682 * be used - I wonder if that matters
3684 return badblocks_init(&rdev->badblocks, 0);
3686 EXPORT_SYMBOL_GPL(md_rdev_init);
3688 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3690 * mark the device faulty if:
3692 * - the device is nonexistent (zero size)
3693 * - the device has no valid superblock
3695 * a faulty rdev _never_ has rdev->sb set.
3697 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3699 char b[BDEVNAME_SIZE];
3701 struct md_rdev *rdev;
3704 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3706 return ERR_PTR(-ENOMEM);
3708 err = md_rdev_init(rdev);
3711 err = alloc_disk_sb(rdev);
3715 err = lock_rdev(rdev, newdev, super_format == -2);
3719 kobject_init(&rdev->kobj, &rdev_ktype);
3721 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3723 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3724 bdevname(rdev->bdev,b));
3729 if (super_format >= 0) {
3730 err = super_types[super_format].
3731 load_super(rdev, NULL, super_minor);
3732 if (err == -EINVAL) {
3733 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3734 bdevname(rdev->bdev,b),
3735 super_format, super_minor);
3739 pr_warn("md: could not read %s's sb, not importing!\n",
3740 bdevname(rdev->bdev,b));
3750 md_rdev_clear(rdev);
3752 return ERR_PTR(err);
3756 * Check a full RAID array for plausibility
3759 static int analyze_sbs(struct mddev *mddev)
3762 struct md_rdev *rdev, *freshest, *tmp;
3763 char b[BDEVNAME_SIZE];
3766 rdev_for_each_safe(rdev, tmp, mddev)
3767 switch (super_types[mddev->major_version].
3768 load_super(rdev, freshest, mddev->minor_version)) {
3775 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3776 bdevname(rdev->bdev,b));
3777 md_kick_rdev_from_array(rdev);
3780 /* Cannot find a valid fresh disk */
3782 pr_warn("md: cannot find a valid disk\n");
3786 super_types[mddev->major_version].
3787 validate_super(mddev, freshest);
3790 rdev_for_each_safe(rdev, tmp, mddev) {
3791 if (mddev->max_disks &&
3792 (rdev->desc_nr >= mddev->max_disks ||
3793 i > mddev->max_disks)) {
3794 pr_warn("md: %s: %s: only %d devices permitted\n",
3795 mdname(mddev), bdevname(rdev->bdev, b),
3797 md_kick_rdev_from_array(rdev);
3800 if (rdev != freshest) {
3801 if (super_types[mddev->major_version].
3802 validate_super(mddev, rdev)) {
3803 pr_warn("md: kicking non-fresh %s from array!\n",
3804 bdevname(rdev->bdev,b));
3805 md_kick_rdev_from_array(rdev);
3809 if (mddev->level == LEVEL_MULTIPATH) {
3810 rdev->desc_nr = i++;
3811 rdev->raid_disk = rdev->desc_nr;
3812 set_bit(In_sync, &rdev->flags);
3813 } else if (rdev->raid_disk >=
3814 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3815 !test_bit(Journal, &rdev->flags)) {
3816 rdev->raid_disk = -1;
3817 clear_bit(In_sync, &rdev->flags);
3824 /* Read a fixed-point number.
3825 * Numbers in sysfs attributes should be in "standard" units where
3826 * possible, so time should be in seconds.
3827 * However we internally use a a much smaller unit such as
3828 * milliseconds or jiffies.
3829 * This function takes a decimal number with a possible fractional
3830 * component, and produces an integer which is the result of
3831 * multiplying that number by 10^'scale'.
3832 * all without any floating-point arithmetic.
3834 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3836 unsigned long result = 0;
3838 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3841 else if (decimals < scale) {
3844 result = result * 10 + value;
3856 *res = result * int_pow(10, scale - decimals);
3861 safe_delay_show(struct mddev *mddev, char *page)
3863 int msec = (mddev->safemode_delay*1000)/HZ;
3864 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3867 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3871 if (mddev_is_clustered(mddev)) {
3872 pr_warn("md: Safemode is disabled for clustered mode\n");
3876 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3879 mddev->safemode_delay = 0;
3881 unsigned long old_delay = mddev->safemode_delay;
3882 unsigned long new_delay = (msec*HZ)/1000;
3886 mddev->safemode_delay = new_delay;
3887 if (new_delay < old_delay || old_delay == 0)
3888 mod_timer(&mddev->safemode_timer, jiffies+1);
3892 static struct md_sysfs_entry md_safe_delay =
3893 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3896 level_show(struct mddev *mddev, char *page)
3898 struct md_personality *p;
3900 spin_lock(&mddev->lock);
3903 ret = sprintf(page, "%s\n", p->name);
3904 else if (mddev->clevel[0])
3905 ret = sprintf(page, "%s\n", mddev->clevel);
3906 else if (mddev->level != LEVEL_NONE)
3907 ret = sprintf(page, "%d\n", mddev->level);
3910 spin_unlock(&mddev->lock);
3915 level_store(struct mddev *mddev, const char *buf, size_t len)
3920 struct md_personality *pers, *oldpers;
3922 void *priv, *oldpriv;
3923 struct md_rdev *rdev;
3925 if (slen == 0 || slen >= sizeof(clevel))
3928 rv = mddev_lock(mddev);
3932 if (mddev->pers == NULL) {
3933 strncpy(mddev->clevel, buf, slen);
3934 if (mddev->clevel[slen-1] == '\n')
3936 mddev->clevel[slen] = 0;
3937 mddev->level = LEVEL_NONE;
3945 /* request to change the personality. Need to ensure:
3946 * - array is not engaged in resync/recovery/reshape
3947 * - old personality can be suspended
3948 * - new personality will access other array.
3952 if (mddev->sync_thread ||
3953 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3954 mddev->reshape_position != MaxSector ||
3955 mddev->sysfs_active)
3959 if (!mddev->pers->quiesce) {
3960 pr_warn("md: %s: %s does not support online personality change\n",
3961 mdname(mddev), mddev->pers->name);
3965 /* Now find the new personality */
3966 strncpy(clevel, buf, slen);
3967 if (clevel[slen-1] == '\n')
3970 if (kstrtol(clevel, 10, &level))
3973 if (request_module("md-%s", clevel) != 0)
3974 request_module("md-level-%s", clevel);
3975 spin_lock(&pers_lock);
3976 pers = find_pers(level, clevel);
3977 if (!pers || !try_module_get(pers->owner)) {
3978 spin_unlock(&pers_lock);
3979 pr_warn("md: personality %s not loaded\n", clevel);
3983 spin_unlock(&pers_lock);
3985 if (pers == mddev->pers) {
3986 /* Nothing to do! */
3987 module_put(pers->owner);
3991 if (!pers->takeover) {
3992 module_put(pers->owner);
3993 pr_warn("md: %s: %s does not support personality takeover\n",
3994 mdname(mddev), clevel);
3999 rdev_for_each(rdev, mddev)
4000 rdev->new_raid_disk = rdev->raid_disk;
4002 /* ->takeover must set new_* and/or delta_disks
4003 * if it succeeds, and may set them when it fails.
4005 priv = pers->takeover(mddev);
4007 mddev->new_level = mddev->level;
4008 mddev->new_layout = mddev->layout;
4009 mddev->new_chunk_sectors = mddev->chunk_sectors;
4010 mddev->raid_disks -= mddev->delta_disks;
4011 mddev->delta_disks = 0;
4012 mddev->reshape_backwards = 0;
4013 module_put(pers->owner);
4014 pr_warn("md: %s: %s would not accept array\n",
4015 mdname(mddev), clevel);
4020 /* Looks like we have a winner */
4021 mddev_suspend(mddev);
4022 mddev_detach(mddev);
4024 spin_lock(&mddev->lock);
4025 oldpers = mddev->pers;
4026 oldpriv = mddev->private;
4028 mddev->private = priv;
4029 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4030 mddev->level = mddev->new_level;
4031 mddev->layout = mddev->new_layout;
4032 mddev->chunk_sectors = mddev->new_chunk_sectors;
4033 mddev->delta_disks = 0;
4034 mddev->reshape_backwards = 0;
4035 mddev->degraded = 0;
4036 spin_unlock(&mddev->lock);
4038 if (oldpers->sync_request == NULL &&
4040 /* We are converting from a no-redundancy array
4041 * to a redundancy array and metadata is managed
4042 * externally so we need to be sure that writes
4043 * won't block due to a need to transition
4045 * until external management is started.
4048 mddev->safemode_delay = 0;
4049 mddev->safemode = 0;
4052 oldpers->free(mddev, oldpriv);
4054 if (oldpers->sync_request == NULL &&
4055 pers->sync_request != NULL) {
4056 /* need to add the md_redundancy_group */
4057 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4058 pr_warn("md: cannot register extra attributes for %s\n",
4060 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4061 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4062 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4064 if (oldpers->sync_request != NULL &&
4065 pers->sync_request == NULL) {
4066 /* need to remove the md_redundancy_group */
4067 if (mddev->to_remove == NULL)
4068 mddev->to_remove = &md_redundancy_group;
4071 module_put(oldpers->owner);
4073 rdev_for_each(rdev, mddev) {
4074 if (rdev->raid_disk < 0)
4076 if (rdev->new_raid_disk >= mddev->raid_disks)
4077 rdev->new_raid_disk = -1;
4078 if (rdev->new_raid_disk == rdev->raid_disk)
4080 sysfs_unlink_rdev(mddev, rdev);
4082 rdev_for_each(rdev, mddev) {
4083 if (rdev->raid_disk < 0)
4085 if (rdev->new_raid_disk == rdev->raid_disk)
4087 rdev->raid_disk = rdev->new_raid_disk;
4088 if (rdev->raid_disk < 0)
4089 clear_bit(In_sync, &rdev->flags);
4091 if (sysfs_link_rdev(mddev, rdev))
4092 pr_warn("md: cannot register rd%d for %s after level change\n",
4093 rdev->raid_disk, mdname(mddev));
4097 if (pers->sync_request == NULL) {
4098 /* this is now an array without redundancy, so
4099 * it must always be in_sync
4102 del_timer_sync(&mddev->safemode_timer);
4104 blk_set_stacking_limits(&mddev->queue->limits);
4106 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4107 mddev_resume(mddev);
4109 md_update_sb(mddev, 1);
4110 sysfs_notify_dirent_safe(mddev->sysfs_level);
4111 md_new_event(mddev);
4114 mddev_unlock(mddev);
4118 static struct md_sysfs_entry md_level =
4119 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4122 layout_show(struct mddev *mddev, char *page)
4124 /* just a number, not meaningful for all levels */
4125 if (mddev->reshape_position != MaxSector &&
4126 mddev->layout != mddev->new_layout)
4127 return sprintf(page, "%d (%d)\n",
4128 mddev->new_layout, mddev->layout);
4129 return sprintf(page, "%d\n", mddev->layout);
4133 layout_store(struct mddev *mddev, const char *buf, size_t len)
4138 err = kstrtouint(buf, 10, &n);
4141 err = mddev_lock(mddev);
4146 if (mddev->pers->check_reshape == NULL)
4151 mddev->new_layout = n;
4152 err = mddev->pers->check_reshape(mddev);
4154 mddev->new_layout = mddev->layout;
4157 mddev->new_layout = n;
4158 if (mddev->reshape_position == MaxSector)
4161 mddev_unlock(mddev);
4164 static struct md_sysfs_entry md_layout =
4165 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4168 raid_disks_show(struct mddev *mddev, char *page)
4170 if (mddev->raid_disks == 0)
4172 if (mddev->reshape_position != MaxSector &&
4173 mddev->delta_disks != 0)
4174 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4175 mddev->raid_disks - mddev->delta_disks);
4176 return sprintf(page, "%d\n", mddev->raid_disks);
4179 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4182 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4187 err = kstrtouint(buf, 10, &n);
4191 err = mddev_lock(mddev);
4195 err = update_raid_disks(mddev, n);
4196 else if (mddev->reshape_position != MaxSector) {
4197 struct md_rdev *rdev;
4198 int olddisks = mddev->raid_disks - mddev->delta_disks;
4201 rdev_for_each(rdev, mddev) {
4203 rdev->data_offset < rdev->new_data_offset)
4206 rdev->data_offset > rdev->new_data_offset)
4210 mddev->delta_disks = n - olddisks;
4211 mddev->raid_disks = n;
4212 mddev->reshape_backwards = (mddev->delta_disks < 0);
4214 mddev->raid_disks = n;
4216 mddev_unlock(mddev);
4217 return err ? err : len;
4219 static struct md_sysfs_entry md_raid_disks =
4220 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4223 uuid_show(struct mddev *mddev, char *page)
4225 return sprintf(page, "%pU\n", mddev->uuid);
4227 static struct md_sysfs_entry md_uuid =
4228 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4231 chunk_size_show(struct mddev *mddev, char *page)
4233 if (mddev->reshape_position != MaxSector &&
4234 mddev->chunk_sectors != mddev->new_chunk_sectors)
4235 return sprintf(page, "%d (%d)\n",
4236 mddev->new_chunk_sectors << 9,
4237 mddev->chunk_sectors << 9);
4238 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4242 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4247 err = kstrtoul(buf, 10, &n);
4251 err = mddev_lock(mddev);
4255 if (mddev->pers->check_reshape == NULL)
4260 mddev->new_chunk_sectors = n >> 9;
4261 err = mddev->pers->check_reshape(mddev);
4263 mddev->new_chunk_sectors = mddev->chunk_sectors;
4266 mddev->new_chunk_sectors = n >> 9;
4267 if (mddev->reshape_position == MaxSector)
4268 mddev->chunk_sectors = n >> 9;
4270 mddev_unlock(mddev);
4273 static struct md_sysfs_entry md_chunk_size =
4274 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4277 resync_start_show(struct mddev *mddev, char *page)
4279 if (mddev->recovery_cp == MaxSector)
4280 return sprintf(page, "none\n");
4281 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4285 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4287 unsigned long long n;
4290 if (cmd_match(buf, "none"))
4293 err = kstrtoull(buf, 10, &n);
4296 if (n != (sector_t)n)
4300 err = mddev_lock(mddev);
4303 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4307 mddev->recovery_cp = n;
4309 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4311 mddev_unlock(mddev);
4314 static struct md_sysfs_entry md_resync_start =
4315 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4316 resync_start_show, resync_start_store);
4319 * The array state can be:
4322 * No devices, no size, no level
4323 * Equivalent to STOP_ARRAY ioctl
4325 * May have some settings, but array is not active
4326 * all IO results in error
4327 * When written, doesn't tear down array, but just stops it
4328 * suspended (not supported yet)
4329 * All IO requests will block. The array can be reconfigured.
4330 * Writing this, if accepted, will block until array is quiescent
4332 * no resync can happen. no superblocks get written.
4333 * write requests fail
4335 * like readonly, but behaves like 'clean' on a write request.
4337 * clean - no pending writes, but otherwise active.
4338 * When written to inactive array, starts without resync
4339 * If a write request arrives then
4340 * if metadata is known, mark 'dirty' and switch to 'active'.
4341 * if not known, block and switch to write-pending
4342 * If written to an active array that has pending writes, then fails.
4344 * fully active: IO and resync can be happening.
4345 * When written to inactive array, starts with resync
4348 * clean, but writes are blocked waiting for 'active' to be written.
4351 * like active, but no writes have been seen for a while (100msec).
4354 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4355 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4356 * when a member is gone, so this state will at least alert the
4357 * user that something is wrong.
4359 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4360 write_pending, active_idle, broken, bad_word};
4361 static char *array_states[] = {
4362 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4363 "write-pending", "active-idle", "broken", NULL };
4365 static int match_word(const char *word, char **list)
4368 for (n=0; list[n]; n++)
4369 if (cmd_match(word, list[n]))
4375 array_state_show(struct mddev *mddev, char *page)
4377 enum array_state st = inactive;
4379 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4388 spin_lock(&mddev->lock);
4389 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4391 else if (mddev->in_sync)
4393 else if (mddev->safemode)
4397 spin_unlock(&mddev->lock);
4400 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4403 if (list_empty(&mddev->disks) &&
4404 mddev->raid_disks == 0 &&
4405 mddev->dev_sectors == 0)
4410 return sprintf(page, "%s\n", array_states[st]);
4413 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4414 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4415 static int restart_array(struct mddev *mddev);
4418 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4421 enum array_state st = match_word(buf, array_states);
4423 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4424 /* don't take reconfig_mutex when toggling between
4427 spin_lock(&mddev->lock);
4429 restart_array(mddev);
4430 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4431 md_wakeup_thread(mddev->thread);
4432 wake_up(&mddev->sb_wait);
4433 } else /* st == clean */ {
4434 restart_array(mddev);
4435 if (!set_in_sync(mddev))
4439 sysfs_notify_dirent_safe(mddev->sysfs_state);
4440 spin_unlock(&mddev->lock);
4443 err = mddev_lock(mddev);
4451 /* stopping an active array */
4452 err = do_md_stop(mddev, 0, NULL);
4455 /* stopping an active array */
4457 err = do_md_stop(mddev, 2, NULL);
4459 err = 0; /* already inactive */
4462 break; /* not supported yet */
4465 err = md_set_readonly(mddev, NULL);
4468 set_disk_ro(mddev->gendisk, 1);
4469 err = do_md_run(mddev);
4475 err = md_set_readonly(mddev, NULL);
4476 else if (mddev->ro == 1)
4477 err = restart_array(mddev);
4480 set_disk_ro(mddev->gendisk, 0);
4484 err = do_md_run(mddev);
4489 err = restart_array(mddev);
4492 spin_lock(&mddev->lock);
4493 if (!set_in_sync(mddev))
4495 spin_unlock(&mddev->lock);
4501 err = restart_array(mddev);
4504 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4505 wake_up(&mddev->sb_wait);
4509 set_disk_ro(mddev->gendisk, 0);
4510 err = do_md_run(mddev);
4516 /* these cannot be set */
4521 if (mddev->hold_active == UNTIL_IOCTL)
4522 mddev->hold_active = 0;
4523 sysfs_notify_dirent_safe(mddev->sysfs_state);
4525 mddev_unlock(mddev);
4528 static struct md_sysfs_entry md_array_state =
4529 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4532 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4533 return sprintf(page, "%d\n",
4534 atomic_read(&mddev->max_corr_read_errors));
4538 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4543 rv = kstrtouint(buf, 10, &n);
4546 atomic_set(&mddev->max_corr_read_errors, n);
4550 static struct md_sysfs_entry max_corr_read_errors =
4551 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4552 max_corrected_read_errors_store);
4555 null_show(struct mddev *mddev, char *page)
4560 /* need to ensure rdev_delayed_delete() has completed */
4561 static void flush_rdev_wq(struct mddev *mddev)
4563 struct md_rdev *rdev;
4566 rdev_for_each_rcu(rdev, mddev)
4567 if (work_pending(&rdev->del_work)) {
4568 flush_workqueue(md_rdev_misc_wq);
4575 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4577 /* buf must be %d:%d\n? giving major and minor numbers */
4578 /* The new device is added to the array.
4579 * If the array has a persistent superblock, we read the
4580 * superblock to initialise info and check validity.
4581 * Otherwise, only checking done is that in bind_rdev_to_array,
4582 * which mainly checks size.
4585 int major = simple_strtoul(buf, &e, 10);
4588 struct md_rdev *rdev;
4591 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4593 minor = simple_strtoul(e+1, &e, 10);
4594 if (*e && *e != '\n')
4596 dev = MKDEV(major, minor);
4597 if (major != MAJOR(dev) ||
4598 minor != MINOR(dev))
4601 flush_rdev_wq(mddev);
4602 err = mddev_lock(mddev);
4605 if (mddev->persistent) {
4606 rdev = md_import_device(dev, mddev->major_version,
4607 mddev->minor_version);
4608 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4609 struct md_rdev *rdev0
4610 = list_entry(mddev->disks.next,
4611 struct md_rdev, same_set);
4612 err = super_types[mddev->major_version]
4613 .load_super(rdev, rdev0, mddev->minor_version);
4617 } else if (mddev->external)
4618 rdev = md_import_device(dev, -2, -1);
4620 rdev = md_import_device(dev, -1, -1);
4623 mddev_unlock(mddev);
4624 return PTR_ERR(rdev);
4626 err = bind_rdev_to_array(rdev, mddev);
4630 mddev_unlock(mddev);
4632 md_new_event(mddev);
4633 return err ? err : len;
4636 static struct md_sysfs_entry md_new_device =
4637 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4640 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4643 unsigned long chunk, end_chunk;
4646 err = mddev_lock(mddev);
4651 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4653 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4654 if (buf == end) break;
4655 if (*end == '-') { /* range */
4657 end_chunk = simple_strtoul(buf, &end, 0);
4658 if (buf == end) break;
4660 if (*end && !isspace(*end)) break;
4661 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4662 buf = skip_spaces(end);
4664 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4666 mddev_unlock(mddev);
4670 static struct md_sysfs_entry md_bitmap =
4671 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4674 size_show(struct mddev *mddev, char *page)
4676 return sprintf(page, "%llu\n",
4677 (unsigned long long)mddev->dev_sectors / 2);
4680 static int update_size(struct mddev *mddev, sector_t num_sectors);
4683 size_store(struct mddev *mddev, const char *buf, size_t len)
4685 /* If array is inactive, we can reduce the component size, but
4686 * not increase it (except from 0).
4687 * If array is active, we can try an on-line resize
4690 int err = strict_blocks_to_sectors(buf, §ors);
4694 err = mddev_lock(mddev);
4698 err = update_size(mddev, sectors);
4700 md_update_sb(mddev, 1);
4702 if (mddev->dev_sectors == 0 ||
4703 mddev->dev_sectors > sectors)
4704 mddev->dev_sectors = sectors;
4708 mddev_unlock(mddev);
4709 return err ? err : len;
4712 static struct md_sysfs_entry md_size =
4713 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4715 /* Metadata version.
4717 * 'none' for arrays with no metadata (good luck...)
4718 * 'external' for arrays with externally managed metadata,
4719 * or N.M for internally known formats
4722 metadata_show(struct mddev *mddev, char *page)
4724 if (mddev->persistent)
4725 return sprintf(page, "%d.%d\n",
4726 mddev->major_version, mddev->minor_version);
4727 else if (mddev->external)
4728 return sprintf(page, "external:%s\n", mddev->metadata_type);
4730 return sprintf(page, "none\n");
4734 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4739 /* Changing the details of 'external' metadata is
4740 * always permitted. Otherwise there must be
4741 * no devices attached to the array.
4744 err = mddev_lock(mddev);
4748 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4750 else if (!list_empty(&mddev->disks))
4754 if (cmd_match(buf, "none")) {
4755 mddev->persistent = 0;
4756 mddev->external = 0;
4757 mddev->major_version = 0;
4758 mddev->minor_version = 90;
4761 if (strncmp(buf, "external:", 9) == 0) {
4762 size_t namelen = len-9;
4763 if (namelen >= sizeof(mddev->metadata_type))
4764 namelen = sizeof(mddev->metadata_type)-1;
4765 strncpy(mddev->metadata_type, buf+9, namelen);
4766 mddev->metadata_type[namelen] = 0;
4767 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4768 mddev->metadata_type[--namelen] = 0;
4769 mddev->persistent = 0;
4770 mddev->external = 1;
4771 mddev->major_version = 0;
4772 mddev->minor_version = 90;
4775 major = simple_strtoul(buf, &e, 10);
4777 if (e==buf || *e != '.')
4780 minor = simple_strtoul(buf, &e, 10);
4781 if (e==buf || (*e && *e != '\n') )
4784 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4786 mddev->major_version = major;
4787 mddev->minor_version = minor;
4788 mddev->persistent = 1;
4789 mddev->external = 0;
4792 mddev_unlock(mddev);
4796 static struct md_sysfs_entry md_metadata =
4797 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4800 action_show(struct mddev *mddev, char *page)
4802 char *type = "idle";
4803 unsigned long recovery = mddev->recovery;
4804 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4806 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4807 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4808 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4810 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4811 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4813 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4817 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4819 else if (mddev->reshape_position != MaxSector)
4822 return sprintf(page, "%s\n", type);
4826 action_store(struct mddev *mddev, const char *page, size_t len)
4828 if (!mddev->pers || !mddev->pers->sync_request)
4832 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4833 if (cmd_match(page, "frozen"))
4834 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4836 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4837 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4838 mddev_lock(mddev) == 0) {
4839 if (work_pending(&mddev->del_work))
4840 flush_workqueue(md_misc_wq);
4841 if (mddev->sync_thread) {
4842 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4843 md_reap_sync_thread(mddev);
4845 mddev_unlock(mddev);
4847 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4849 else if (cmd_match(page, "resync"))
4850 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4851 else if (cmd_match(page, "recover")) {
4852 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4853 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4854 } else if (cmd_match(page, "reshape")) {
4856 if (mddev->pers->start_reshape == NULL)
4858 err = mddev_lock(mddev);
4860 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4863 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4864 err = mddev->pers->start_reshape(mddev);
4866 mddev_unlock(mddev);
4870 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4872 if (cmd_match(page, "check"))
4873 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4874 else if (!cmd_match(page, "repair"))
4876 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4877 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4878 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4880 if (mddev->ro == 2) {
4881 /* A write to sync_action is enough to justify
4882 * canceling read-auto mode
4885 md_wakeup_thread(mddev->sync_thread);
4887 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4888 md_wakeup_thread(mddev->thread);
4889 sysfs_notify_dirent_safe(mddev->sysfs_action);
4893 static struct md_sysfs_entry md_scan_mode =
4894 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4897 last_sync_action_show(struct mddev *mddev, char *page)
4899 return sprintf(page, "%s\n", mddev->last_sync_action);
4902 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4905 mismatch_cnt_show(struct mddev *mddev, char *page)
4907 return sprintf(page, "%llu\n",
4908 (unsigned long long)
4909 atomic64_read(&mddev->resync_mismatches));
4912 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4915 sync_min_show(struct mddev *mddev, char *page)
4917 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4918 mddev->sync_speed_min ? "local": "system");
4922 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4927 if (strncmp(buf, "system", 6)==0) {
4930 rv = kstrtouint(buf, 10, &min);
4936 mddev->sync_speed_min = min;
4940 static struct md_sysfs_entry md_sync_min =
4941 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4944 sync_max_show(struct mddev *mddev, char *page)
4946 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4947 mddev->sync_speed_max ? "local": "system");
4951 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4956 if (strncmp(buf, "system", 6)==0) {
4959 rv = kstrtouint(buf, 10, &max);
4965 mddev->sync_speed_max = max;
4969 static struct md_sysfs_entry md_sync_max =
4970 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4973 degraded_show(struct mddev *mddev, char *page)
4975 return sprintf(page, "%d\n", mddev->degraded);
4977 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4980 sync_force_parallel_show(struct mddev *mddev, char *page)
4982 return sprintf(page, "%d\n", mddev->parallel_resync);
4986 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4990 if (kstrtol(buf, 10, &n))
4993 if (n != 0 && n != 1)
4996 mddev->parallel_resync = n;
4998 if (mddev->sync_thread)
4999 wake_up(&resync_wait);
5004 /* force parallel resync, even with shared block devices */
5005 static struct md_sysfs_entry md_sync_force_parallel =
5006 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5007 sync_force_parallel_show, sync_force_parallel_store);
5010 sync_speed_show(struct mddev *mddev, char *page)
5012 unsigned long resync, dt, db;
5013 if (mddev->curr_resync == 0)
5014 return sprintf(page, "none\n");
5015 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5016 dt = (jiffies - mddev->resync_mark) / HZ;
5018 db = resync - mddev->resync_mark_cnt;
5019 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5022 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5025 sync_completed_show(struct mddev *mddev, char *page)
5027 unsigned long long max_sectors, resync;
5029 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5030 return sprintf(page, "none\n");
5032 if (mddev->curr_resync == 1 ||
5033 mddev->curr_resync == 2)
5034 return sprintf(page, "delayed\n");
5036 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5037 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5038 max_sectors = mddev->resync_max_sectors;
5040 max_sectors = mddev->dev_sectors;
5042 resync = mddev->curr_resync_completed;
5043 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5046 static struct md_sysfs_entry md_sync_completed =
5047 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5050 min_sync_show(struct mddev *mddev, char *page)
5052 return sprintf(page, "%llu\n",
5053 (unsigned long long)mddev->resync_min);
5056 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5058 unsigned long long min;
5061 if (kstrtoull(buf, 10, &min))
5064 spin_lock(&mddev->lock);
5066 if (min > mddev->resync_max)
5070 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5073 /* Round down to multiple of 4K for safety */
5074 mddev->resync_min = round_down(min, 8);
5078 spin_unlock(&mddev->lock);
5082 static struct md_sysfs_entry md_min_sync =
5083 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5086 max_sync_show(struct mddev *mddev, char *page)
5088 if (mddev->resync_max == MaxSector)
5089 return sprintf(page, "max\n");
5091 return sprintf(page, "%llu\n",
5092 (unsigned long long)mddev->resync_max);
5095 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5098 spin_lock(&mddev->lock);
5099 if (strncmp(buf, "max", 3) == 0)
5100 mddev->resync_max = MaxSector;
5102 unsigned long long max;
5106 if (kstrtoull(buf, 10, &max))
5108 if (max < mddev->resync_min)
5112 if (max < mddev->resync_max &&
5114 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5117 /* Must be a multiple of chunk_size */
5118 chunk = mddev->chunk_sectors;
5120 sector_t temp = max;
5123 if (sector_div(temp, chunk))
5126 mddev->resync_max = max;
5128 wake_up(&mddev->recovery_wait);
5131 spin_unlock(&mddev->lock);
5135 static struct md_sysfs_entry md_max_sync =
5136 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5139 suspend_lo_show(struct mddev *mddev, char *page)
5141 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5145 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5147 unsigned long long new;
5150 err = kstrtoull(buf, 10, &new);
5153 if (new != (sector_t)new)
5156 err = mddev_lock(mddev);
5160 if (mddev->pers == NULL ||
5161 mddev->pers->quiesce == NULL)
5163 mddev_suspend(mddev);
5164 mddev->suspend_lo = new;
5165 mddev_resume(mddev);
5169 mddev_unlock(mddev);
5172 static struct md_sysfs_entry md_suspend_lo =
5173 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5176 suspend_hi_show(struct mddev *mddev, char *page)
5178 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5182 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5184 unsigned long long new;
5187 err = kstrtoull(buf, 10, &new);
5190 if (new != (sector_t)new)
5193 err = mddev_lock(mddev);
5197 if (mddev->pers == NULL)
5200 mddev_suspend(mddev);
5201 mddev->suspend_hi = new;
5202 mddev_resume(mddev);
5206 mddev_unlock(mddev);
5209 static struct md_sysfs_entry md_suspend_hi =
5210 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5213 reshape_position_show(struct mddev *mddev, char *page)
5215 if (mddev->reshape_position != MaxSector)
5216 return sprintf(page, "%llu\n",
5217 (unsigned long long)mddev->reshape_position);
5218 strcpy(page, "none\n");
5223 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5225 struct md_rdev *rdev;
5226 unsigned long long new;
5229 err = kstrtoull(buf, 10, &new);
5232 if (new != (sector_t)new)
5234 err = mddev_lock(mddev);
5240 mddev->reshape_position = new;
5241 mddev->delta_disks = 0;
5242 mddev->reshape_backwards = 0;
5243 mddev->new_level = mddev->level;
5244 mddev->new_layout = mddev->layout;
5245 mddev->new_chunk_sectors = mddev->chunk_sectors;
5246 rdev_for_each(rdev, mddev)
5247 rdev->new_data_offset = rdev->data_offset;
5250 mddev_unlock(mddev);
5254 static struct md_sysfs_entry md_reshape_position =
5255 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5256 reshape_position_store);
5259 reshape_direction_show(struct mddev *mddev, char *page)
5261 return sprintf(page, "%s\n",
5262 mddev->reshape_backwards ? "backwards" : "forwards");
5266 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5271 if (cmd_match(buf, "forwards"))
5273 else if (cmd_match(buf, "backwards"))
5277 if (mddev->reshape_backwards == backwards)
5280 err = mddev_lock(mddev);
5283 /* check if we are allowed to change */
5284 if (mddev->delta_disks)
5286 else if (mddev->persistent &&
5287 mddev->major_version == 0)
5290 mddev->reshape_backwards = backwards;
5291 mddev_unlock(mddev);
5295 static struct md_sysfs_entry md_reshape_direction =
5296 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5297 reshape_direction_store);
5300 array_size_show(struct mddev *mddev, char *page)
5302 if (mddev->external_size)
5303 return sprintf(page, "%llu\n",
5304 (unsigned long long)mddev->array_sectors/2);
5306 return sprintf(page, "default\n");
5310 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5315 err = mddev_lock(mddev);
5319 /* cluster raid doesn't support change array_sectors */
5320 if (mddev_is_clustered(mddev)) {
5321 mddev_unlock(mddev);
5325 if (strncmp(buf, "default", 7) == 0) {
5327 sectors = mddev->pers->size(mddev, 0, 0);
5329 sectors = mddev->array_sectors;
5331 mddev->external_size = 0;
5333 if (strict_blocks_to_sectors(buf, §ors) < 0)
5335 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5338 mddev->external_size = 1;
5342 mddev->array_sectors = sectors;
5344 set_capacity_and_notify(mddev->gendisk,
5345 mddev->array_sectors);
5347 mddev_unlock(mddev);
5351 static struct md_sysfs_entry md_array_size =
5352 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5356 consistency_policy_show(struct mddev *mddev, char *page)
5360 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5361 ret = sprintf(page, "journal\n");
5362 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5363 ret = sprintf(page, "ppl\n");
5364 } else if (mddev->bitmap) {
5365 ret = sprintf(page, "bitmap\n");
5366 } else if (mddev->pers) {
5367 if (mddev->pers->sync_request)
5368 ret = sprintf(page, "resync\n");
5370 ret = sprintf(page, "none\n");
5372 ret = sprintf(page, "unknown\n");
5379 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5384 if (mddev->pers->change_consistency_policy)
5385 err = mddev->pers->change_consistency_policy(mddev, buf);
5388 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5389 set_bit(MD_HAS_PPL, &mddev->flags);
5394 return err ? err : len;
5397 static struct md_sysfs_entry md_consistency_policy =
5398 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5399 consistency_policy_store);
5401 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5403 return sprintf(page, "%d\n", mddev->fail_last_dev);
5407 * Setting fail_last_dev to true to allow last device to be forcibly removed
5408 * from RAID1/RAID10.
5411 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5416 ret = kstrtobool(buf, &value);
5420 if (value != mddev->fail_last_dev)
5421 mddev->fail_last_dev = value;
5425 static struct md_sysfs_entry md_fail_last_dev =
5426 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5427 fail_last_dev_store);
5429 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5431 if (mddev->pers == NULL || (mddev->pers->level != 1))
5432 return sprintf(page, "n/a\n");
5434 return sprintf(page, "%d\n", mddev->serialize_policy);
5438 * Setting serialize_policy to true to enforce write IO is not reordered
5442 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5447 err = kstrtobool(buf, &value);
5451 if (value == mddev->serialize_policy)
5454 err = mddev_lock(mddev);
5457 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5458 pr_err("md: serialize_policy is only effective for raid1\n");
5463 mddev_suspend(mddev);
5465 mddev_create_serial_pool(mddev, NULL, true);
5467 mddev_destroy_serial_pool(mddev, NULL, true);
5468 mddev->serialize_policy = value;
5469 mddev_resume(mddev);
5471 mddev_unlock(mddev);
5475 static struct md_sysfs_entry md_serialize_policy =
5476 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5477 serialize_policy_store);
5480 static struct attribute *md_default_attrs[] = {
5483 &md_raid_disks.attr,
5485 &md_chunk_size.attr,
5487 &md_resync_start.attr,
5489 &md_new_device.attr,
5490 &md_safe_delay.attr,
5491 &md_array_state.attr,
5492 &md_reshape_position.attr,
5493 &md_reshape_direction.attr,
5494 &md_array_size.attr,
5495 &max_corr_read_errors.attr,
5496 &md_consistency_policy.attr,
5497 &md_fail_last_dev.attr,
5498 &md_serialize_policy.attr,
5502 static struct attribute *md_redundancy_attrs[] = {
5504 &md_last_scan_mode.attr,
5505 &md_mismatches.attr,
5508 &md_sync_speed.attr,
5509 &md_sync_force_parallel.attr,
5510 &md_sync_completed.attr,
5513 &md_suspend_lo.attr,
5514 &md_suspend_hi.attr,
5519 static const struct attribute_group md_redundancy_group = {
5521 .attrs = md_redundancy_attrs,
5525 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5527 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5528 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5533 spin_lock(&all_mddevs_lock);
5534 if (list_empty(&mddev->all_mddevs)) {
5535 spin_unlock(&all_mddevs_lock);
5539 spin_unlock(&all_mddevs_lock);
5541 rv = entry->show(mddev, page);
5547 md_attr_store(struct kobject *kobj, struct attribute *attr,
5548 const char *page, size_t length)
5550 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5551 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5556 if (!capable(CAP_SYS_ADMIN))
5558 spin_lock(&all_mddevs_lock);
5559 if (list_empty(&mddev->all_mddevs)) {
5560 spin_unlock(&all_mddevs_lock);
5564 spin_unlock(&all_mddevs_lock);
5565 rv = entry->store(mddev, page, length);
5570 static void md_free(struct kobject *ko)
5572 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5574 if (mddev->sysfs_state)
5575 sysfs_put(mddev->sysfs_state);
5576 if (mddev->sysfs_level)
5577 sysfs_put(mddev->sysfs_level);
5579 if (mddev->gendisk) {
5580 del_gendisk(mddev->gendisk);
5581 blk_cleanup_disk(mddev->gendisk);
5583 percpu_ref_exit(&mddev->writes_pending);
5585 bioset_exit(&mddev->bio_set);
5586 bioset_exit(&mddev->sync_set);
5587 if (mddev->level != 1 && mddev->level != 10)
5588 bioset_exit(&mddev->io_acct_set);
5592 static const struct sysfs_ops md_sysfs_ops = {
5593 .show = md_attr_show,
5594 .store = md_attr_store,
5596 static struct kobj_type md_ktype = {
5598 .sysfs_ops = &md_sysfs_ops,
5599 .default_attrs = md_default_attrs,
5604 static void mddev_delayed_delete(struct work_struct *ws)
5606 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5608 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5609 kobject_del(&mddev->kobj);
5610 kobject_put(&mddev->kobj);
5613 static void no_op(struct percpu_ref *r) {}
5615 int mddev_init_writes_pending(struct mddev *mddev)
5617 if (mddev->writes_pending.percpu_count_ptr)
5619 if (percpu_ref_init(&mddev->writes_pending, no_op,
5620 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5622 /* We want to start with the refcount at zero */
5623 percpu_ref_put(&mddev->writes_pending);
5626 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5628 static int md_alloc(dev_t dev, char *name)
5631 * If dev is zero, name is the name of a device to allocate with
5632 * an arbitrary minor number. It will be "md_???"
5633 * If dev is non-zero it must be a device number with a MAJOR of
5634 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5635 * the device is being created by opening a node in /dev.
5636 * If "name" is not NULL, the device is being created by
5637 * writing to /sys/module/md_mod/parameters/new_array.
5639 static DEFINE_MUTEX(disks_mutex);
5640 struct mddev *mddev;
5641 struct gendisk *disk;
5648 * Wait for any previous instance of this device to be completely
5649 * removed (mddev_delayed_delete).
5651 flush_workqueue(md_misc_wq);
5653 mutex_lock(&disks_mutex);
5654 mddev = mddev_alloc(dev);
5655 if (IS_ERR(mddev)) {
5656 mutex_unlock(&disks_mutex);
5657 return PTR_ERR(mddev);
5660 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5661 shift = partitioned ? MdpMinorShift : 0;
5662 unit = MINOR(mddev->unit) >> shift;
5665 /* Need to ensure that 'name' is not a duplicate.
5667 struct mddev *mddev2;
5668 spin_lock(&all_mddevs_lock);
5670 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5671 if (mddev2->gendisk &&
5672 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5673 spin_unlock(&all_mddevs_lock);
5677 spin_unlock(&all_mddevs_lock);
5681 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5683 mddev->hold_active = UNTIL_STOP;
5686 disk = blk_alloc_disk(NUMA_NO_NODE);
5690 disk->major = MAJOR(mddev->unit);
5691 disk->first_minor = unit << shift;
5692 disk->minors = 1 << shift;
5694 strcpy(disk->disk_name, name);
5695 else if (partitioned)
5696 sprintf(disk->disk_name, "md_d%d", unit);
5698 sprintf(disk->disk_name, "md%d", unit);
5699 disk->fops = &md_fops;
5700 disk->private_data = mddev;
5702 mddev->queue = disk->queue;
5703 blk_set_stacking_limits(&mddev->queue->limits);
5704 blk_queue_write_cache(mddev->queue, true, true);
5705 /* Allow extended partitions. This makes the
5706 * 'mdp' device redundant, but we can't really
5709 disk->flags |= GENHD_FL_EXT_DEVT;
5710 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5711 mddev->gendisk = disk;
5714 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5716 /* This isn't possible, but as kobject_init_and_add is marked
5717 * __must_check, we must do something with the result
5719 pr_debug("md: cannot register %s/md - name in use\n",
5723 if (mddev->kobj.sd &&
5724 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5725 pr_debug("pointless warning\n");
5727 mutex_unlock(&disks_mutex);
5728 if (!error && mddev->kobj.sd) {
5729 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5730 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5731 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5737 static void md_probe(dev_t dev)
5739 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5742 md_alloc(dev, NULL);
5745 static int add_named_array(const char *val, const struct kernel_param *kp)
5748 * val must be "md_*" or "mdNNN".
5749 * For "md_*" we allocate an array with a large free minor number, and
5750 * set the name to val. val must not already be an active name.
5751 * For "mdNNN" we allocate an array with the minor number NNN
5752 * which must not already be in use.
5754 int len = strlen(val);
5755 char buf[DISK_NAME_LEN];
5756 unsigned long devnum;
5758 while (len && val[len-1] == '\n')
5760 if (len >= DISK_NAME_LEN)
5762 strlcpy(buf, val, len+1);
5763 if (strncmp(buf, "md_", 3) == 0)
5764 return md_alloc(0, buf);
5765 if (strncmp(buf, "md", 2) == 0 &&
5767 kstrtoul(buf+2, 10, &devnum) == 0 &&
5768 devnum <= MINORMASK)
5769 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5774 static void md_safemode_timeout(struct timer_list *t)
5776 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5778 mddev->safemode = 1;
5779 if (mddev->external)
5780 sysfs_notify_dirent_safe(mddev->sysfs_state);
5782 md_wakeup_thread(mddev->thread);
5785 static int start_dirty_degraded;
5787 int md_run(struct mddev *mddev)
5790 struct md_rdev *rdev;
5791 struct md_personality *pers;
5793 if (list_empty(&mddev->disks))
5794 /* cannot run an array with no devices.. */
5799 /* Cannot run until previous stop completes properly */
5800 if (mddev->sysfs_active)
5804 * Analyze all RAID superblock(s)
5806 if (!mddev->raid_disks) {
5807 if (!mddev->persistent)
5809 err = analyze_sbs(mddev);
5814 if (mddev->level != LEVEL_NONE)
5815 request_module("md-level-%d", mddev->level);
5816 else if (mddev->clevel[0])
5817 request_module("md-%s", mddev->clevel);
5820 * Drop all container device buffers, from now on
5821 * the only valid external interface is through the md
5824 mddev->has_superblocks = false;
5825 rdev_for_each(rdev, mddev) {
5826 if (test_bit(Faulty, &rdev->flags))
5828 sync_blockdev(rdev->bdev);
5829 invalidate_bdev(rdev->bdev);
5830 if (mddev->ro != 1 && rdev_read_only(rdev)) {
5833 set_disk_ro(mddev->gendisk, 1);
5837 mddev->has_superblocks = true;
5839 /* perform some consistency tests on the device.
5840 * We don't want the data to overlap the metadata,
5841 * Internal Bitmap issues have been handled elsewhere.
5843 if (rdev->meta_bdev) {
5844 /* Nothing to check */;
5845 } else if (rdev->data_offset < rdev->sb_start) {
5846 if (mddev->dev_sectors &&
5847 rdev->data_offset + mddev->dev_sectors
5849 pr_warn("md: %s: data overlaps metadata\n",
5854 if (rdev->sb_start + rdev->sb_size/512
5855 > rdev->data_offset) {
5856 pr_warn("md: %s: metadata overlaps data\n",
5861 sysfs_notify_dirent_safe(rdev->sysfs_state);
5864 if (!bioset_initialized(&mddev->bio_set)) {
5865 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5869 if (!bioset_initialized(&mddev->sync_set)) {
5870 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5874 if (mddev->level != 1 && mddev->level != 10 &&
5875 !bioset_initialized(&mddev->io_acct_set)) {
5876 err = bioset_init(&mddev->io_acct_set, BIO_POOL_SIZE,
5877 offsetof(struct md_io_acct, bio_clone), 0);
5882 spin_lock(&pers_lock);
5883 pers = find_pers(mddev->level, mddev->clevel);
5884 if (!pers || !try_module_get(pers->owner)) {
5885 spin_unlock(&pers_lock);
5886 if (mddev->level != LEVEL_NONE)
5887 pr_warn("md: personality for level %d is not loaded!\n",
5890 pr_warn("md: personality for level %s is not loaded!\n",
5895 spin_unlock(&pers_lock);
5896 if (mddev->level != pers->level) {
5897 mddev->level = pers->level;
5898 mddev->new_level = pers->level;
5900 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5902 if (mddev->reshape_position != MaxSector &&
5903 pers->start_reshape == NULL) {
5904 /* This personality cannot handle reshaping... */
5905 module_put(pers->owner);
5910 if (pers->sync_request) {
5911 /* Warn if this is a potentially silly
5914 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5915 struct md_rdev *rdev2;
5918 rdev_for_each(rdev, mddev)
5919 rdev_for_each(rdev2, mddev) {
5921 rdev->bdev->bd_disk ==
5922 rdev2->bdev->bd_disk) {
5923 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5925 bdevname(rdev->bdev,b),
5926 bdevname(rdev2->bdev,b2));
5932 pr_warn("True protection against single-disk failure might be compromised.\n");
5935 mddev->recovery = 0;
5936 /* may be over-ridden by personality */
5937 mddev->resync_max_sectors = mddev->dev_sectors;
5939 mddev->ok_start_degraded = start_dirty_degraded;
5941 if (start_readonly && mddev->ro == 0)
5942 mddev->ro = 2; /* read-only, but switch on first write */
5944 err = pers->run(mddev);
5946 pr_warn("md: pers->run() failed ...\n");
5947 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5948 WARN_ONCE(!mddev->external_size,
5949 "%s: default size too small, but 'external_size' not in effect?\n",
5951 pr_warn("md: invalid array_size %llu > default size %llu\n",
5952 (unsigned long long)mddev->array_sectors / 2,
5953 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5956 if (err == 0 && pers->sync_request &&
5957 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5958 struct bitmap *bitmap;
5960 bitmap = md_bitmap_create(mddev, -1);
5961 if (IS_ERR(bitmap)) {
5962 err = PTR_ERR(bitmap);
5963 pr_warn("%s: failed to create bitmap (%d)\n",
5964 mdname(mddev), err);
5966 mddev->bitmap = bitmap;
5972 if (mddev->bitmap_info.max_write_behind > 0) {
5973 bool create_pool = false;
5975 rdev_for_each(rdev, mddev) {
5976 if (test_bit(WriteMostly, &rdev->flags) &&
5977 rdev_init_serial(rdev))
5980 if (create_pool && mddev->serial_info_pool == NULL) {
5981 mddev->serial_info_pool =
5982 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5983 sizeof(struct serial_info));
5984 if (!mddev->serial_info_pool) {
5994 rdev_for_each(rdev, mddev) {
5995 if (rdev->raid_disk >= 0 &&
5996 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6001 if (mddev->degraded)
6004 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6006 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6007 blk_queue_flag_set(QUEUE_FLAG_IO_STAT, mddev->queue);
6009 if (pers->sync_request) {
6010 if (mddev->kobj.sd &&
6011 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6012 pr_warn("md: cannot register extra attributes for %s\n",
6014 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6015 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6016 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6017 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6020 atomic_set(&mddev->max_corr_read_errors,
6021 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6022 mddev->safemode = 0;
6023 if (mddev_is_clustered(mddev))
6024 mddev->safemode_delay = 0;
6026 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6029 spin_lock(&mddev->lock);
6031 spin_unlock(&mddev->lock);
6032 rdev_for_each(rdev, mddev)
6033 if (rdev->raid_disk >= 0)
6034 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6036 if (mddev->degraded && !mddev->ro)
6037 /* This ensures that recovering status is reported immediately
6038 * via sysfs - until a lack of spares is confirmed.
6040 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6041 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6043 if (mddev->sb_flags)
6044 md_update_sb(mddev, 0);
6046 md_new_event(mddev);
6050 mddev_detach(mddev);
6052 pers->free(mddev, mddev->private);
6053 mddev->private = NULL;
6054 module_put(pers->owner);
6055 md_bitmap_destroy(mddev);
6057 if (mddev->level != 1 && mddev->level != 10)
6058 bioset_exit(&mddev->io_acct_set);
6060 bioset_exit(&mddev->sync_set);
6062 bioset_exit(&mddev->bio_set);
6065 EXPORT_SYMBOL_GPL(md_run);
6067 int do_md_run(struct mddev *mddev)
6071 set_bit(MD_NOT_READY, &mddev->flags);
6072 err = md_run(mddev);
6075 err = md_bitmap_load(mddev);
6077 md_bitmap_destroy(mddev);
6081 if (mddev_is_clustered(mddev))
6082 md_allow_write(mddev);
6084 /* run start up tasks that require md_thread */
6087 md_wakeup_thread(mddev->thread);
6088 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6090 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6091 clear_bit(MD_NOT_READY, &mddev->flags);
6093 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6094 sysfs_notify_dirent_safe(mddev->sysfs_state);
6095 sysfs_notify_dirent_safe(mddev->sysfs_action);
6096 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6098 clear_bit(MD_NOT_READY, &mddev->flags);
6102 int md_start(struct mddev *mddev)
6106 if (mddev->pers->start) {
6107 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6108 md_wakeup_thread(mddev->thread);
6109 ret = mddev->pers->start(mddev);
6110 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6111 md_wakeup_thread(mddev->sync_thread);
6115 EXPORT_SYMBOL_GPL(md_start);
6117 static int restart_array(struct mddev *mddev)
6119 struct gendisk *disk = mddev->gendisk;
6120 struct md_rdev *rdev;
6121 bool has_journal = false;
6122 bool has_readonly = false;
6124 /* Complain if it has no devices */
6125 if (list_empty(&mddev->disks))
6133 rdev_for_each_rcu(rdev, mddev) {
6134 if (test_bit(Journal, &rdev->flags) &&
6135 !test_bit(Faulty, &rdev->flags))
6137 if (rdev_read_only(rdev))
6138 has_readonly = true;
6141 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6142 /* Don't restart rw with journal missing/faulty */
6147 mddev->safemode = 0;
6149 set_disk_ro(disk, 0);
6150 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6151 /* Kick recovery or resync if necessary */
6152 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6153 md_wakeup_thread(mddev->thread);
6154 md_wakeup_thread(mddev->sync_thread);
6155 sysfs_notify_dirent_safe(mddev->sysfs_state);
6159 static void md_clean(struct mddev *mddev)
6161 mddev->array_sectors = 0;
6162 mddev->external_size = 0;
6163 mddev->dev_sectors = 0;
6164 mddev->raid_disks = 0;
6165 mddev->recovery_cp = 0;
6166 mddev->resync_min = 0;
6167 mddev->resync_max = MaxSector;
6168 mddev->reshape_position = MaxSector;
6169 mddev->external = 0;
6170 mddev->persistent = 0;
6171 mddev->level = LEVEL_NONE;
6172 mddev->clevel[0] = 0;
6174 mddev->sb_flags = 0;
6176 mddev->metadata_type[0] = 0;
6177 mddev->chunk_sectors = 0;
6178 mddev->ctime = mddev->utime = 0;
6180 mddev->max_disks = 0;
6182 mddev->can_decrease_events = 0;
6183 mddev->delta_disks = 0;
6184 mddev->reshape_backwards = 0;
6185 mddev->new_level = LEVEL_NONE;
6186 mddev->new_layout = 0;
6187 mddev->new_chunk_sectors = 0;
6188 mddev->curr_resync = 0;
6189 atomic64_set(&mddev->resync_mismatches, 0);
6190 mddev->suspend_lo = mddev->suspend_hi = 0;
6191 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6192 mddev->recovery = 0;
6195 mddev->degraded = 0;
6196 mddev->safemode = 0;
6197 mddev->private = NULL;
6198 mddev->cluster_info = NULL;
6199 mddev->bitmap_info.offset = 0;
6200 mddev->bitmap_info.default_offset = 0;
6201 mddev->bitmap_info.default_space = 0;
6202 mddev->bitmap_info.chunksize = 0;
6203 mddev->bitmap_info.daemon_sleep = 0;
6204 mddev->bitmap_info.max_write_behind = 0;
6205 mddev->bitmap_info.nodes = 0;
6208 static void __md_stop_writes(struct mddev *mddev)
6210 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6211 if (work_pending(&mddev->del_work))
6212 flush_workqueue(md_misc_wq);
6213 if (mddev->sync_thread) {
6214 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6215 md_reap_sync_thread(mddev);
6218 del_timer_sync(&mddev->safemode_timer);
6220 if (mddev->pers && mddev->pers->quiesce) {
6221 mddev->pers->quiesce(mddev, 1);
6222 mddev->pers->quiesce(mddev, 0);
6224 md_bitmap_flush(mddev);
6226 if (mddev->ro == 0 &&
6227 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6229 /* mark array as shutdown cleanly */
6230 if (!mddev_is_clustered(mddev))
6232 md_update_sb(mddev, 1);
6234 /* disable policy to guarantee rdevs free resources for serialization */
6235 mddev->serialize_policy = 0;
6236 mddev_destroy_serial_pool(mddev, NULL, true);
6239 void md_stop_writes(struct mddev *mddev)
6241 mddev_lock_nointr(mddev);
6242 __md_stop_writes(mddev);
6243 mddev_unlock(mddev);
6245 EXPORT_SYMBOL_GPL(md_stop_writes);
6247 static void mddev_detach(struct mddev *mddev)
6249 md_bitmap_wait_behind_writes(mddev);
6250 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6251 mddev->pers->quiesce(mddev, 1);
6252 mddev->pers->quiesce(mddev, 0);
6254 md_unregister_thread(&mddev->thread);
6256 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6259 static void __md_stop(struct mddev *mddev)
6261 struct md_personality *pers = mddev->pers;
6262 md_bitmap_destroy(mddev);
6263 mddev_detach(mddev);
6264 /* Ensure ->event_work is done */
6265 if (mddev->event_work.func)
6266 flush_workqueue(md_misc_wq);
6267 spin_lock(&mddev->lock);
6269 spin_unlock(&mddev->lock);
6270 pers->free(mddev, mddev->private);
6271 mddev->private = NULL;
6272 if (pers->sync_request && mddev->to_remove == NULL)
6273 mddev->to_remove = &md_redundancy_group;
6274 module_put(pers->owner);
6275 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6278 void md_stop(struct mddev *mddev)
6280 /* stop the array and free an attached data structures.
6281 * This is called from dm-raid
6284 bioset_exit(&mddev->bio_set);
6285 bioset_exit(&mddev->sync_set);
6286 if (mddev->level != 1 && mddev->level != 10)
6287 bioset_exit(&mddev->io_acct_set);
6290 EXPORT_SYMBOL_GPL(md_stop);
6292 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6297 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6299 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6300 md_wakeup_thread(mddev->thread);
6302 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6303 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6304 if (mddev->sync_thread)
6305 /* Thread might be blocked waiting for metadata update
6306 * which will now never happen */
6307 wake_up_process(mddev->sync_thread->tsk);
6309 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6311 mddev_unlock(mddev);
6312 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6314 wait_event(mddev->sb_wait,
6315 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6316 mddev_lock_nointr(mddev);
6318 mutex_lock(&mddev->open_mutex);
6319 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6320 mddev->sync_thread ||
6321 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6322 pr_warn("md: %s still in use.\n",mdname(mddev));
6324 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6326 md_wakeup_thread(mddev->thread);
6332 __md_stop_writes(mddev);
6338 set_disk_ro(mddev->gendisk, 1);
6339 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6340 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6341 md_wakeup_thread(mddev->thread);
6342 sysfs_notify_dirent_safe(mddev->sysfs_state);
6346 mutex_unlock(&mddev->open_mutex);
6351 * 0 - completely stop and dis-assemble array
6352 * 2 - stop but do not disassemble array
6354 static int do_md_stop(struct mddev *mddev, int mode,
6355 struct block_device *bdev)
6357 struct gendisk *disk = mddev->gendisk;
6358 struct md_rdev *rdev;
6361 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6363 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6364 md_wakeup_thread(mddev->thread);
6366 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6367 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6368 if (mddev->sync_thread)
6369 /* Thread might be blocked waiting for metadata update
6370 * which will now never happen */
6371 wake_up_process(mddev->sync_thread->tsk);
6373 mddev_unlock(mddev);
6374 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6375 !test_bit(MD_RECOVERY_RUNNING,
6376 &mddev->recovery)));
6377 mddev_lock_nointr(mddev);
6379 mutex_lock(&mddev->open_mutex);
6380 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6381 mddev->sysfs_active ||
6382 mddev->sync_thread ||
6383 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6384 pr_warn("md: %s still in use.\n",mdname(mddev));
6385 mutex_unlock(&mddev->open_mutex);
6387 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6388 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6389 md_wakeup_thread(mddev->thread);
6395 set_disk_ro(disk, 0);
6397 __md_stop_writes(mddev);
6400 /* tell userspace to handle 'inactive' */
6401 sysfs_notify_dirent_safe(mddev->sysfs_state);
6403 rdev_for_each(rdev, mddev)
6404 if (rdev->raid_disk >= 0)
6405 sysfs_unlink_rdev(mddev, rdev);
6407 set_capacity_and_notify(disk, 0);
6408 mutex_unlock(&mddev->open_mutex);
6414 mutex_unlock(&mddev->open_mutex);
6416 * Free resources if final stop
6419 pr_info("md: %s stopped.\n", mdname(mddev));
6421 if (mddev->bitmap_info.file) {
6422 struct file *f = mddev->bitmap_info.file;
6423 spin_lock(&mddev->lock);
6424 mddev->bitmap_info.file = NULL;
6425 spin_unlock(&mddev->lock);
6428 mddev->bitmap_info.offset = 0;
6430 export_array(mddev);
6433 if (mddev->hold_active == UNTIL_STOP)
6434 mddev->hold_active = 0;
6436 md_new_event(mddev);
6437 sysfs_notify_dirent_safe(mddev->sysfs_state);
6442 static void autorun_array(struct mddev *mddev)
6444 struct md_rdev *rdev;
6447 if (list_empty(&mddev->disks))
6450 pr_info("md: running: ");
6452 rdev_for_each(rdev, mddev) {
6453 char b[BDEVNAME_SIZE];
6454 pr_cont("<%s>", bdevname(rdev->bdev,b));
6458 err = do_md_run(mddev);
6460 pr_warn("md: do_md_run() returned %d\n", err);
6461 do_md_stop(mddev, 0, NULL);
6466 * lets try to run arrays based on all disks that have arrived
6467 * until now. (those are in pending_raid_disks)
6469 * the method: pick the first pending disk, collect all disks with
6470 * the same UUID, remove all from the pending list and put them into
6471 * the 'same_array' list. Then order this list based on superblock
6472 * update time (freshest comes first), kick out 'old' disks and
6473 * compare superblocks. If everything's fine then run it.
6475 * If "unit" is allocated, then bump its reference count
6477 static void autorun_devices(int part)
6479 struct md_rdev *rdev0, *rdev, *tmp;
6480 struct mddev *mddev;
6481 char b[BDEVNAME_SIZE];
6483 pr_info("md: autorun ...\n");
6484 while (!list_empty(&pending_raid_disks)) {
6487 LIST_HEAD(candidates);
6488 rdev0 = list_entry(pending_raid_disks.next,
6489 struct md_rdev, same_set);
6491 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6492 INIT_LIST_HEAD(&candidates);
6493 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6494 if (super_90_load(rdev, rdev0, 0) >= 0) {
6495 pr_debug("md: adding %s ...\n",
6496 bdevname(rdev->bdev,b));
6497 list_move(&rdev->same_set, &candidates);
6500 * now we have a set of devices, with all of them having
6501 * mostly sane superblocks. It's time to allocate the
6505 dev = MKDEV(mdp_major,
6506 rdev0->preferred_minor << MdpMinorShift);
6507 unit = MINOR(dev) >> MdpMinorShift;
6509 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6512 if (rdev0->preferred_minor != unit) {
6513 pr_warn("md: unit number in %s is bad: %d\n",
6514 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6519 mddev = mddev_find(dev);
6523 if (mddev_lock(mddev))
6524 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6525 else if (mddev->raid_disks || mddev->major_version
6526 || !list_empty(&mddev->disks)) {
6527 pr_warn("md: %s already running, cannot run %s\n",
6528 mdname(mddev), bdevname(rdev0->bdev,b));
6529 mddev_unlock(mddev);
6531 pr_debug("md: created %s\n", mdname(mddev));
6532 mddev->persistent = 1;
6533 rdev_for_each_list(rdev, tmp, &candidates) {
6534 list_del_init(&rdev->same_set);
6535 if (bind_rdev_to_array(rdev, mddev))
6538 autorun_array(mddev);
6539 mddev_unlock(mddev);
6541 /* on success, candidates will be empty, on error
6544 rdev_for_each_list(rdev, tmp, &candidates) {
6545 list_del_init(&rdev->same_set);
6550 pr_info("md: ... autorun DONE.\n");
6552 #endif /* !MODULE */
6554 static int get_version(void __user *arg)
6558 ver.major = MD_MAJOR_VERSION;
6559 ver.minor = MD_MINOR_VERSION;
6560 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6562 if (copy_to_user(arg, &ver, sizeof(ver)))
6568 static int get_array_info(struct mddev *mddev, void __user *arg)
6570 mdu_array_info_t info;
6571 int nr,working,insync,failed,spare;
6572 struct md_rdev *rdev;
6574 nr = working = insync = failed = spare = 0;
6576 rdev_for_each_rcu(rdev, mddev) {
6578 if (test_bit(Faulty, &rdev->flags))
6582 if (test_bit(In_sync, &rdev->flags))
6584 else if (test_bit(Journal, &rdev->flags))
6585 /* TODO: add journal count to md_u.h */
6593 info.major_version = mddev->major_version;
6594 info.minor_version = mddev->minor_version;
6595 info.patch_version = MD_PATCHLEVEL_VERSION;
6596 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6597 info.level = mddev->level;
6598 info.size = mddev->dev_sectors / 2;
6599 if (info.size != mddev->dev_sectors / 2) /* overflow */
6602 info.raid_disks = mddev->raid_disks;
6603 info.md_minor = mddev->md_minor;
6604 info.not_persistent= !mddev->persistent;
6606 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6609 info.state = (1<<MD_SB_CLEAN);
6610 if (mddev->bitmap && mddev->bitmap_info.offset)
6611 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6612 if (mddev_is_clustered(mddev))
6613 info.state |= (1<<MD_SB_CLUSTERED);
6614 info.active_disks = insync;
6615 info.working_disks = working;
6616 info.failed_disks = failed;
6617 info.spare_disks = spare;
6619 info.layout = mddev->layout;
6620 info.chunk_size = mddev->chunk_sectors << 9;
6622 if (copy_to_user(arg, &info, sizeof(info)))
6628 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6630 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6634 file = kzalloc(sizeof(*file), GFP_NOIO);
6639 spin_lock(&mddev->lock);
6640 /* bitmap enabled */
6641 if (mddev->bitmap_info.file) {
6642 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6643 sizeof(file->pathname));
6647 memmove(file->pathname, ptr,
6648 sizeof(file->pathname)-(ptr-file->pathname));
6650 spin_unlock(&mddev->lock);
6653 copy_to_user(arg, file, sizeof(*file)))
6660 static int get_disk_info(struct mddev *mddev, void __user * arg)
6662 mdu_disk_info_t info;
6663 struct md_rdev *rdev;
6665 if (copy_from_user(&info, arg, sizeof(info)))
6669 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6671 info.major = MAJOR(rdev->bdev->bd_dev);
6672 info.minor = MINOR(rdev->bdev->bd_dev);
6673 info.raid_disk = rdev->raid_disk;
6675 if (test_bit(Faulty, &rdev->flags))
6676 info.state |= (1<<MD_DISK_FAULTY);
6677 else if (test_bit(In_sync, &rdev->flags)) {
6678 info.state |= (1<<MD_DISK_ACTIVE);
6679 info.state |= (1<<MD_DISK_SYNC);
6681 if (test_bit(Journal, &rdev->flags))
6682 info.state |= (1<<MD_DISK_JOURNAL);
6683 if (test_bit(WriteMostly, &rdev->flags))
6684 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6685 if (test_bit(FailFast, &rdev->flags))
6686 info.state |= (1<<MD_DISK_FAILFAST);
6688 info.major = info.minor = 0;
6689 info.raid_disk = -1;
6690 info.state = (1<<MD_DISK_REMOVED);
6694 if (copy_to_user(arg, &info, sizeof(info)))
6700 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6702 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6703 struct md_rdev *rdev;
6704 dev_t dev = MKDEV(info->major,info->minor);
6706 if (mddev_is_clustered(mddev) &&
6707 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6708 pr_warn("%s: Cannot add to clustered mddev.\n",
6713 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6716 if (!mddev->raid_disks) {
6718 /* expecting a device which has a superblock */
6719 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6721 pr_warn("md: md_import_device returned %ld\n",
6723 return PTR_ERR(rdev);
6725 if (!list_empty(&mddev->disks)) {
6726 struct md_rdev *rdev0
6727 = list_entry(mddev->disks.next,
6728 struct md_rdev, same_set);
6729 err = super_types[mddev->major_version]
6730 .load_super(rdev, rdev0, mddev->minor_version);
6732 pr_warn("md: %s has different UUID to %s\n",
6733 bdevname(rdev->bdev,b),
6734 bdevname(rdev0->bdev,b2));
6739 err = bind_rdev_to_array(rdev, mddev);
6746 * md_add_new_disk can be used once the array is assembled
6747 * to add "hot spares". They must already have a superblock
6752 if (!mddev->pers->hot_add_disk) {
6753 pr_warn("%s: personality does not support diskops!\n",
6757 if (mddev->persistent)
6758 rdev = md_import_device(dev, mddev->major_version,
6759 mddev->minor_version);
6761 rdev = md_import_device(dev, -1, -1);
6763 pr_warn("md: md_import_device returned %ld\n",
6765 return PTR_ERR(rdev);
6767 /* set saved_raid_disk if appropriate */
6768 if (!mddev->persistent) {
6769 if (info->state & (1<<MD_DISK_SYNC) &&
6770 info->raid_disk < mddev->raid_disks) {
6771 rdev->raid_disk = info->raid_disk;
6772 set_bit(In_sync, &rdev->flags);
6773 clear_bit(Bitmap_sync, &rdev->flags);
6775 rdev->raid_disk = -1;
6776 rdev->saved_raid_disk = rdev->raid_disk;
6778 super_types[mddev->major_version].
6779 validate_super(mddev, rdev);
6780 if ((info->state & (1<<MD_DISK_SYNC)) &&
6781 rdev->raid_disk != info->raid_disk) {
6782 /* This was a hot-add request, but events doesn't
6783 * match, so reject it.
6789 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6790 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6791 set_bit(WriteMostly, &rdev->flags);
6793 clear_bit(WriteMostly, &rdev->flags);
6794 if (info->state & (1<<MD_DISK_FAILFAST))
6795 set_bit(FailFast, &rdev->flags);
6797 clear_bit(FailFast, &rdev->flags);
6799 if (info->state & (1<<MD_DISK_JOURNAL)) {
6800 struct md_rdev *rdev2;
6801 bool has_journal = false;
6803 /* make sure no existing journal disk */
6804 rdev_for_each(rdev2, mddev) {
6805 if (test_bit(Journal, &rdev2->flags)) {
6810 if (has_journal || mddev->bitmap) {
6814 set_bit(Journal, &rdev->flags);
6817 * check whether the device shows up in other nodes
6819 if (mddev_is_clustered(mddev)) {
6820 if (info->state & (1 << MD_DISK_CANDIDATE))
6821 set_bit(Candidate, &rdev->flags);
6822 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6823 /* --add initiated by this node */
6824 err = md_cluster_ops->add_new_disk(mddev, rdev);
6832 rdev->raid_disk = -1;
6833 err = bind_rdev_to_array(rdev, mddev);
6838 if (mddev_is_clustered(mddev)) {
6839 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6841 err = md_cluster_ops->new_disk_ack(mddev,
6844 md_kick_rdev_from_array(rdev);
6848 md_cluster_ops->add_new_disk_cancel(mddev);
6850 err = add_bound_rdev(rdev);
6854 err = add_bound_rdev(rdev);
6859 /* otherwise, md_add_new_disk is only allowed
6860 * for major_version==0 superblocks
6862 if (mddev->major_version != 0) {
6863 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6867 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6869 rdev = md_import_device(dev, -1, 0);
6871 pr_warn("md: error, md_import_device() returned %ld\n",
6873 return PTR_ERR(rdev);
6875 rdev->desc_nr = info->number;
6876 if (info->raid_disk < mddev->raid_disks)
6877 rdev->raid_disk = info->raid_disk;
6879 rdev->raid_disk = -1;
6881 if (rdev->raid_disk < mddev->raid_disks)
6882 if (info->state & (1<<MD_DISK_SYNC))
6883 set_bit(In_sync, &rdev->flags);
6885 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6886 set_bit(WriteMostly, &rdev->flags);
6887 if (info->state & (1<<MD_DISK_FAILFAST))
6888 set_bit(FailFast, &rdev->flags);
6890 if (!mddev->persistent) {
6891 pr_debug("md: nonpersistent superblock ...\n");
6892 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6894 rdev->sb_start = calc_dev_sboffset(rdev);
6895 rdev->sectors = rdev->sb_start;
6897 err = bind_rdev_to_array(rdev, mddev);
6907 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6909 char b[BDEVNAME_SIZE];
6910 struct md_rdev *rdev;
6915 rdev = find_rdev(mddev, dev);
6919 if (rdev->raid_disk < 0)
6922 clear_bit(Blocked, &rdev->flags);
6923 remove_and_add_spares(mddev, rdev);
6925 if (rdev->raid_disk >= 0)
6929 if (mddev_is_clustered(mddev)) {
6930 if (md_cluster_ops->remove_disk(mddev, rdev))
6934 md_kick_rdev_from_array(rdev);
6935 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6937 md_wakeup_thread(mddev->thread);
6939 md_update_sb(mddev, 1);
6940 md_new_event(mddev);
6944 pr_debug("md: cannot remove active disk %s from %s ...\n",
6945 bdevname(rdev->bdev,b), mdname(mddev));
6949 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6951 char b[BDEVNAME_SIZE];
6953 struct md_rdev *rdev;
6958 if (mddev->major_version != 0) {
6959 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6963 if (!mddev->pers->hot_add_disk) {
6964 pr_warn("%s: personality does not support diskops!\n",
6969 rdev = md_import_device(dev, -1, 0);
6971 pr_warn("md: error, md_import_device() returned %ld\n",
6976 if (mddev->persistent)
6977 rdev->sb_start = calc_dev_sboffset(rdev);
6979 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6981 rdev->sectors = rdev->sb_start;
6983 if (test_bit(Faulty, &rdev->flags)) {
6984 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6985 bdevname(rdev->bdev,b), mdname(mddev));
6990 clear_bit(In_sync, &rdev->flags);
6992 rdev->saved_raid_disk = -1;
6993 err = bind_rdev_to_array(rdev, mddev);
6998 * The rest should better be atomic, we can have disk failures
6999 * noticed in interrupt contexts ...
7002 rdev->raid_disk = -1;
7004 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7006 md_update_sb(mddev, 1);
7008 * Kick recovery, maybe this spare has to be added to the
7009 * array immediately.
7011 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7012 md_wakeup_thread(mddev->thread);
7013 md_new_event(mddev);
7021 static int set_bitmap_file(struct mddev *mddev, int fd)
7026 if (!mddev->pers->quiesce || !mddev->thread)
7028 if (mddev->recovery || mddev->sync_thread)
7030 /* we should be able to change the bitmap.. */
7034 struct inode *inode;
7037 if (mddev->bitmap || mddev->bitmap_info.file)
7038 return -EEXIST; /* cannot add when bitmap is present */
7042 pr_warn("%s: error: failed to get bitmap file\n",
7047 inode = f->f_mapping->host;
7048 if (!S_ISREG(inode->i_mode)) {
7049 pr_warn("%s: error: bitmap file must be a regular file\n",
7052 } else if (!(f->f_mode & FMODE_WRITE)) {
7053 pr_warn("%s: error: bitmap file must open for write\n",
7056 } else if (atomic_read(&inode->i_writecount) != 1) {
7057 pr_warn("%s: error: bitmap file is already in use\n",
7065 mddev->bitmap_info.file = f;
7066 mddev->bitmap_info.offset = 0; /* file overrides offset */
7067 } else if (mddev->bitmap == NULL)
7068 return -ENOENT; /* cannot remove what isn't there */
7072 struct bitmap *bitmap;
7074 bitmap = md_bitmap_create(mddev, -1);
7075 mddev_suspend(mddev);
7076 if (!IS_ERR(bitmap)) {
7077 mddev->bitmap = bitmap;
7078 err = md_bitmap_load(mddev);
7080 err = PTR_ERR(bitmap);
7082 md_bitmap_destroy(mddev);
7085 mddev_resume(mddev);
7086 } else if (fd < 0) {
7087 mddev_suspend(mddev);
7088 md_bitmap_destroy(mddev);
7089 mddev_resume(mddev);
7093 struct file *f = mddev->bitmap_info.file;
7095 spin_lock(&mddev->lock);
7096 mddev->bitmap_info.file = NULL;
7097 spin_unlock(&mddev->lock);
7106 * md_set_array_info is used two different ways
7107 * The original usage is when creating a new array.
7108 * In this usage, raid_disks is > 0 and it together with
7109 * level, size, not_persistent,layout,chunksize determine the
7110 * shape of the array.
7111 * This will always create an array with a type-0.90.0 superblock.
7112 * The newer usage is when assembling an array.
7113 * In this case raid_disks will be 0, and the major_version field is
7114 * use to determine which style super-blocks are to be found on the devices.
7115 * The minor and patch _version numbers are also kept incase the
7116 * super_block handler wishes to interpret them.
7118 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7120 if (info->raid_disks == 0) {
7121 /* just setting version number for superblock loading */
7122 if (info->major_version < 0 ||
7123 info->major_version >= ARRAY_SIZE(super_types) ||
7124 super_types[info->major_version].name == NULL) {
7125 /* maybe try to auto-load a module? */
7126 pr_warn("md: superblock version %d not known\n",
7127 info->major_version);
7130 mddev->major_version = info->major_version;
7131 mddev->minor_version = info->minor_version;
7132 mddev->patch_version = info->patch_version;
7133 mddev->persistent = !info->not_persistent;
7134 /* ensure mddev_put doesn't delete this now that there
7135 * is some minimal configuration.
7137 mddev->ctime = ktime_get_real_seconds();
7140 mddev->major_version = MD_MAJOR_VERSION;
7141 mddev->minor_version = MD_MINOR_VERSION;
7142 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7143 mddev->ctime = ktime_get_real_seconds();
7145 mddev->level = info->level;
7146 mddev->clevel[0] = 0;
7147 mddev->dev_sectors = 2 * (sector_t)info->size;
7148 mddev->raid_disks = info->raid_disks;
7149 /* don't set md_minor, it is determined by which /dev/md* was
7152 if (info->state & (1<<MD_SB_CLEAN))
7153 mddev->recovery_cp = MaxSector;
7155 mddev->recovery_cp = 0;
7156 mddev->persistent = ! info->not_persistent;
7157 mddev->external = 0;
7159 mddev->layout = info->layout;
7160 if (mddev->level == 0)
7161 /* Cannot trust RAID0 layout info here */
7163 mddev->chunk_sectors = info->chunk_size >> 9;
7165 if (mddev->persistent) {
7166 mddev->max_disks = MD_SB_DISKS;
7168 mddev->sb_flags = 0;
7170 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7172 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7173 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7174 mddev->bitmap_info.offset = 0;
7176 mddev->reshape_position = MaxSector;
7179 * Generate a 128 bit UUID
7181 get_random_bytes(mddev->uuid, 16);
7183 mddev->new_level = mddev->level;
7184 mddev->new_chunk_sectors = mddev->chunk_sectors;
7185 mddev->new_layout = mddev->layout;
7186 mddev->delta_disks = 0;
7187 mddev->reshape_backwards = 0;
7192 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7194 lockdep_assert_held(&mddev->reconfig_mutex);
7196 if (mddev->external_size)
7199 mddev->array_sectors = array_sectors;
7201 EXPORT_SYMBOL(md_set_array_sectors);
7203 static int update_size(struct mddev *mddev, sector_t num_sectors)
7205 struct md_rdev *rdev;
7207 int fit = (num_sectors == 0);
7208 sector_t old_dev_sectors = mddev->dev_sectors;
7210 if (mddev->pers->resize == NULL)
7212 /* The "num_sectors" is the number of sectors of each device that
7213 * is used. This can only make sense for arrays with redundancy.
7214 * linear and raid0 always use whatever space is available. We can only
7215 * consider changing this number if no resync or reconstruction is
7216 * happening, and if the new size is acceptable. It must fit before the
7217 * sb_start or, if that is <data_offset, it must fit before the size
7218 * of each device. If num_sectors is zero, we find the largest size
7221 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7227 rdev_for_each(rdev, mddev) {
7228 sector_t avail = rdev->sectors;
7230 if (fit && (num_sectors == 0 || num_sectors > avail))
7231 num_sectors = avail;
7232 if (avail < num_sectors)
7235 rv = mddev->pers->resize(mddev, num_sectors);
7237 if (mddev_is_clustered(mddev))
7238 md_cluster_ops->update_size(mddev, old_dev_sectors);
7239 else if (mddev->queue) {
7240 set_capacity_and_notify(mddev->gendisk,
7241 mddev->array_sectors);
7247 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7250 struct md_rdev *rdev;
7251 /* change the number of raid disks */
7252 if (mddev->pers->check_reshape == NULL)
7256 if (raid_disks <= 0 ||
7257 (mddev->max_disks && raid_disks >= mddev->max_disks))
7259 if (mddev->sync_thread ||
7260 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7261 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7262 mddev->reshape_position != MaxSector)
7265 rdev_for_each(rdev, mddev) {
7266 if (mddev->raid_disks < raid_disks &&
7267 rdev->data_offset < rdev->new_data_offset)
7269 if (mddev->raid_disks > raid_disks &&
7270 rdev->data_offset > rdev->new_data_offset)
7274 mddev->delta_disks = raid_disks - mddev->raid_disks;
7275 if (mddev->delta_disks < 0)
7276 mddev->reshape_backwards = 1;
7277 else if (mddev->delta_disks > 0)
7278 mddev->reshape_backwards = 0;
7280 rv = mddev->pers->check_reshape(mddev);
7282 mddev->delta_disks = 0;
7283 mddev->reshape_backwards = 0;
7289 * update_array_info is used to change the configuration of an
7291 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7292 * fields in the info are checked against the array.
7293 * Any differences that cannot be handled will cause an error.
7294 * Normally, only one change can be managed at a time.
7296 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7302 /* calculate expected state,ignoring low bits */
7303 if (mddev->bitmap && mddev->bitmap_info.offset)
7304 state |= (1 << MD_SB_BITMAP_PRESENT);
7306 if (mddev->major_version != info->major_version ||
7307 mddev->minor_version != info->minor_version ||
7308 /* mddev->patch_version != info->patch_version || */
7309 mddev->ctime != info->ctime ||
7310 mddev->level != info->level ||
7311 /* mddev->layout != info->layout || */
7312 mddev->persistent != !info->not_persistent ||
7313 mddev->chunk_sectors != info->chunk_size >> 9 ||
7314 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7315 ((state^info->state) & 0xfffffe00)
7318 /* Check there is only one change */
7319 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7321 if (mddev->raid_disks != info->raid_disks)
7323 if (mddev->layout != info->layout)
7325 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7332 if (mddev->layout != info->layout) {
7334 * we don't need to do anything at the md level, the
7335 * personality will take care of it all.
7337 if (mddev->pers->check_reshape == NULL)
7340 mddev->new_layout = info->layout;
7341 rv = mddev->pers->check_reshape(mddev);
7343 mddev->new_layout = mddev->layout;
7347 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7348 rv = update_size(mddev, (sector_t)info->size * 2);
7350 if (mddev->raid_disks != info->raid_disks)
7351 rv = update_raid_disks(mddev, info->raid_disks);
7353 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7354 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7358 if (mddev->recovery || mddev->sync_thread) {
7362 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7363 struct bitmap *bitmap;
7364 /* add the bitmap */
7365 if (mddev->bitmap) {
7369 if (mddev->bitmap_info.default_offset == 0) {
7373 mddev->bitmap_info.offset =
7374 mddev->bitmap_info.default_offset;
7375 mddev->bitmap_info.space =
7376 mddev->bitmap_info.default_space;
7377 bitmap = md_bitmap_create(mddev, -1);
7378 mddev_suspend(mddev);
7379 if (!IS_ERR(bitmap)) {
7380 mddev->bitmap = bitmap;
7381 rv = md_bitmap_load(mddev);
7383 rv = PTR_ERR(bitmap);
7385 md_bitmap_destroy(mddev);
7386 mddev_resume(mddev);
7388 /* remove the bitmap */
7389 if (!mddev->bitmap) {
7393 if (mddev->bitmap->storage.file) {
7397 if (mddev->bitmap_info.nodes) {
7398 /* hold PW on all the bitmap lock */
7399 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7400 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7402 md_cluster_ops->unlock_all_bitmaps(mddev);
7406 mddev->bitmap_info.nodes = 0;
7407 md_cluster_ops->leave(mddev);
7408 module_put(md_cluster_mod);
7409 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7411 mddev_suspend(mddev);
7412 md_bitmap_destroy(mddev);
7413 mddev_resume(mddev);
7414 mddev->bitmap_info.offset = 0;
7417 md_update_sb(mddev, 1);
7423 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7425 struct md_rdev *rdev;
7428 if (mddev->pers == NULL)
7432 rdev = md_find_rdev_rcu(mddev, dev);
7436 md_error(mddev, rdev);
7437 if (!test_bit(Faulty, &rdev->flags))
7445 * We have a problem here : there is no easy way to give a CHS
7446 * virtual geometry. We currently pretend that we have a 2 heads
7447 * 4 sectors (with a BIG number of cylinders...). This drives
7448 * dosfs just mad... ;-)
7450 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7452 struct mddev *mddev = bdev->bd_disk->private_data;
7456 geo->cylinders = mddev->array_sectors / 8;
7460 static inline bool md_ioctl_valid(unsigned int cmd)
7464 case GET_ARRAY_INFO:
7465 case GET_BITMAP_FILE:
7468 case HOT_REMOVE_DISK:
7470 case RESTART_ARRAY_RW:
7472 case SET_ARRAY_INFO:
7473 case SET_BITMAP_FILE:
7474 case SET_DISK_FAULTY:
7477 case CLUSTERED_DISK_NACK:
7484 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7485 unsigned int cmd, unsigned long arg)
7488 void __user *argp = (void __user *)arg;
7489 struct mddev *mddev = NULL;
7490 bool did_set_md_closing = false;
7492 if (!md_ioctl_valid(cmd))
7497 case GET_ARRAY_INFO:
7501 if (!capable(CAP_SYS_ADMIN))
7506 * Commands dealing with the RAID driver but not any
7511 err = get_version(argp);
7517 * Commands creating/starting a new array:
7520 mddev = bdev->bd_disk->private_data;
7527 /* Some actions do not requires the mutex */
7529 case GET_ARRAY_INFO:
7530 if (!mddev->raid_disks && !mddev->external)
7533 err = get_array_info(mddev, argp);
7537 if (!mddev->raid_disks && !mddev->external)
7540 err = get_disk_info(mddev, argp);
7543 case SET_DISK_FAULTY:
7544 err = set_disk_faulty(mddev, new_decode_dev(arg));
7547 case GET_BITMAP_FILE:
7548 err = get_bitmap_file(mddev, argp);
7553 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7554 flush_rdev_wq(mddev);
7556 if (cmd == HOT_REMOVE_DISK)
7557 /* need to ensure recovery thread has run */
7558 wait_event_interruptible_timeout(mddev->sb_wait,
7559 !test_bit(MD_RECOVERY_NEEDED,
7561 msecs_to_jiffies(5000));
7562 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7563 /* Need to flush page cache, and ensure no-one else opens
7566 mutex_lock(&mddev->open_mutex);
7567 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7568 mutex_unlock(&mddev->open_mutex);
7572 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7573 mutex_unlock(&mddev->open_mutex);
7577 did_set_md_closing = true;
7578 mutex_unlock(&mddev->open_mutex);
7579 sync_blockdev(bdev);
7581 err = mddev_lock(mddev);
7583 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7588 if (cmd == SET_ARRAY_INFO) {
7589 mdu_array_info_t info;
7591 memset(&info, 0, sizeof(info));
7592 else if (copy_from_user(&info, argp, sizeof(info))) {
7597 err = update_array_info(mddev, &info);
7599 pr_warn("md: couldn't update array info. %d\n", err);
7604 if (!list_empty(&mddev->disks)) {
7605 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7609 if (mddev->raid_disks) {
7610 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7614 err = md_set_array_info(mddev, &info);
7616 pr_warn("md: couldn't set array info. %d\n", err);
7623 * Commands querying/configuring an existing array:
7625 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7626 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7627 if ((!mddev->raid_disks && !mddev->external)
7628 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7629 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7630 && cmd != GET_BITMAP_FILE) {
7636 * Commands even a read-only array can execute:
7639 case RESTART_ARRAY_RW:
7640 err = restart_array(mddev);
7644 err = do_md_stop(mddev, 0, bdev);
7648 err = md_set_readonly(mddev, bdev);
7651 case HOT_REMOVE_DISK:
7652 err = hot_remove_disk(mddev, new_decode_dev(arg));
7656 /* We can support ADD_NEW_DISK on read-only arrays
7657 * only if we are re-adding a preexisting device.
7658 * So require mddev->pers and MD_DISK_SYNC.
7661 mdu_disk_info_t info;
7662 if (copy_from_user(&info, argp, sizeof(info)))
7664 else if (!(info.state & (1<<MD_DISK_SYNC)))
7665 /* Need to clear read-only for this */
7668 err = md_add_new_disk(mddev, &info);
7675 * The remaining ioctls are changing the state of the
7676 * superblock, so we do not allow them on read-only arrays.
7678 if (mddev->ro && mddev->pers) {
7679 if (mddev->ro == 2) {
7681 sysfs_notify_dirent_safe(mddev->sysfs_state);
7682 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7683 /* mddev_unlock will wake thread */
7684 /* If a device failed while we were read-only, we
7685 * need to make sure the metadata is updated now.
7687 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7688 mddev_unlock(mddev);
7689 wait_event(mddev->sb_wait,
7690 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7691 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7692 mddev_lock_nointr(mddev);
7703 mdu_disk_info_t info;
7704 if (copy_from_user(&info, argp, sizeof(info)))
7707 err = md_add_new_disk(mddev, &info);
7711 case CLUSTERED_DISK_NACK:
7712 if (mddev_is_clustered(mddev))
7713 md_cluster_ops->new_disk_ack(mddev, false);
7719 err = hot_add_disk(mddev, new_decode_dev(arg));
7723 err = do_md_run(mddev);
7726 case SET_BITMAP_FILE:
7727 err = set_bitmap_file(mddev, (int)arg);
7736 if (mddev->hold_active == UNTIL_IOCTL &&
7738 mddev->hold_active = 0;
7739 mddev_unlock(mddev);
7741 if(did_set_md_closing)
7742 clear_bit(MD_CLOSING, &mddev->flags);
7745 #ifdef CONFIG_COMPAT
7746 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7747 unsigned int cmd, unsigned long arg)
7750 case HOT_REMOVE_DISK:
7752 case SET_DISK_FAULTY:
7753 case SET_BITMAP_FILE:
7754 /* These take in integer arg, do not convert */
7757 arg = (unsigned long)compat_ptr(arg);
7761 return md_ioctl(bdev, mode, cmd, arg);
7763 #endif /* CONFIG_COMPAT */
7765 static int md_set_read_only(struct block_device *bdev, bool ro)
7767 struct mddev *mddev = bdev->bd_disk->private_data;
7770 err = mddev_lock(mddev);
7774 if (!mddev->raid_disks && !mddev->external) {
7780 * Transitioning to read-auto need only happen for arrays that call
7781 * md_write_start and which are not ready for writes yet.
7783 if (!ro && mddev->ro == 1 && mddev->pers) {
7784 err = restart_array(mddev);
7791 mddev_unlock(mddev);
7795 static int md_open(struct block_device *bdev, fmode_t mode)
7798 * Succeed if we can lock the mddev, which confirms that
7799 * it isn't being stopped right now.
7801 struct mddev *mddev = mddev_find(bdev->bd_dev);
7807 if (mddev->gendisk != bdev->bd_disk) {
7808 /* we are racing with mddev_put which is discarding this
7812 /* Wait until bdev->bd_disk is definitely gone */
7813 if (work_pending(&mddev->del_work))
7814 flush_workqueue(md_misc_wq);
7817 BUG_ON(mddev != bdev->bd_disk->private_data);
7819 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7822 if (test_bit(MD_CLOSING, &mddev->flags)) {
7823 mutex_unlock(&mddev->open_mutex);
7829 atomic_inc(&mddev->openers);
7830 mutex_unlock(&mddev->open_mutex);
7832 bdev_check_media_change(bdev);
7839 static void md_release(struct gendisk *disk, fmode_t mode)
7841 struct mddev *mddev = disk->private_data;
7844 atomic_dec(&mddev->openers);
7848 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7850 struct mddev *mddev = disk->private_data;
7851 unsigned int ret = 0;
7854 ret = DISK_EVENT_MEDIA_CHANGE;
7859 const struct block_device_operations md_fops =
7861 .owner = THIS_MODULE,
7862 .submit_bio = md_submit_bio,
7864 .release = md_release,
7866 #ifdef CONFIG_COMPAT
7867 .compat_ioctl = md_compat_ioctl,
7869 .getgeo = md_getgeo,
7870 .check_events = md_check_events,
7871 .set_read_only = md_set_read_only,
7874 static int md_thread(void *arg)
7876 struct md_thread *thread = arg;
7879 * md_thread is a 'system-thread', it's priority should be very
7880 * high. We avoid resource deadlocks individually in each
7881 * raid personality. (RAID5 does preallocation) We also use RR and
7882 * the very same RT priority as kswapd, thus we will never get
7883 * into a priority inversion deadlock.
7885 * we definitely have to have equal or higher priority than
7886 * bdflush, otherwise bdflush will deadlock if there are too
7887 * many dirty RAID5 blocks.
7890 allow_signal(SIGKILL);
7891 while (!kthread_should_stop()) {
7893 /* We need to wait INTERRUPTIBLE so that
7894 * we don't add to the load-average.
7895 * That means we need to be sure no signals are
7898 if (signal_pending(current))
7899 flush_signals(current);
7901 wait_event_interruptible_timeout
7903 test_bit(THREAD_WAKEUP, &thread->flags)
7904 || kthread_should_stop() || kthread_should_park(),
7907 clear_bit(THREAD_WAKEUP, &thread->flags);
7908 if (kthread_should_park())
7910 if (!kthread_should_stop())
7911 thread->run(thread);
7917 void md_wakeup_thread(struct md_thread *thread)
7920 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7921 set_bit(THREAD_WAKEUP, &thread->flags);
7922 wake_up(&thread->wqueue);
7925 EXPORT_SYMBOL(md_wakeup_thread);
7927 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7928 struct mddev *mddev, const char *name)
7930 struct md_thread *thread;
7932 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7936 init_waitqueue_head(&thread->wqueue);
7939 thread->mddev = mddev;
7940 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7941 thread->tsk = kthread_run(md_thread, thread,
7943 mdname(thread->mddev),
7945 if (IS_ERR(thread->tsk)) {
7951 EXPORT_SYMBOL(md_register_thread);
7953 void md_unregister_thread(struct md_thread **threadp)
7955 struct md_thread *thread = *threadp;
7958 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7959 /* Locking ensures that mddev_unlock does not wake_up a
7960 * non-existent thread
7962 spin_lock(&pers_lock);
7964 spin_unlock(&pers_lock);
7966 kthread_stop(thread->tsk);
7969 EXPORT_SYMBOL(md_unregister_thread);
7971 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7973 if (!rdev || test_bit(Faulty, &rdev->flags))
7976 if (!mddev->pers || !mddev->pers->error_handler)
7978 mddev->pers->error_handler(mddev,rdev);
7979 if (mddev->degraded)
7980 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7981 sysfs_notify_dirent_safe(rdev->sysfs_state);
7982 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7983 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7984 md_wakeup_thread(mddev->thread);
7985 if (mddev->event_work.func)
7986 queue_work(md_misc_wq, &mddev->event_work);
7987 md_new_event(mddev);
7989 EXPORT_SYMBOL(md_error);
7991 /* seq_file implementation /proc/mdstat */
7993 static void status_unused(struct seq_file *seq)
7996 struct md_rdev *rdev;
7998 seq_printf(seq, "unused devices: ");
8000 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8001 char b[BDEVNAME_SIZE];
8003 seq_printf(seq, "%s ",
8004 bdevname(rdev->bdev,b));
8007 seq_printf(seq, "<none>");
8009 seq_printf(seq, "\n");
8012 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8014 sector_t max_sectors, resync, res;
8015 unsigned long dt, db = 0;
8016 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8017 int scale, recovery_active;
8018 unsigned int per_milli;
8020 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8021 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8022 max_sectors = mddev->resync_max_sectors;
8024 max_sectors = mddev->dev_sectors;
8026 resync = mddev->curr_resync;
8028 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8029 /* Still cleaning up */
8030 resync = max_sectors;
8031 } else if (resync > max_sectors)
8032 resync = max_sectors;
8034 resync -= atomic_read(&mddev->recovery_active);
8037 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8038 struct md_rdev *rdev;
8040 rdev_for_each(rdev, mddev)
8041 if (rdev->raid_disk >= 0 &&
8042 !test_bit(Faulty, &rdev->flags) &&
8043 rdev->recovery_offset != MaxSector &&
8044 rdev->recovery_offset) {
8045 seq_printf(seq, "\trecover=REMOTE");
8048 if (mddev->reshape_position != MaxSector)
8049 seq_printf(seq, "\treshape=REMOTE");
8051 seq_printf(seq, "\tresync=REMOTE");
8054 if (mddev->recovery_cp < MaxSector) {
8055 seq_printf(seq, "\tresync=PENDING");
8061 seq_printf(seq, "\tresync=DELAYED");
8065 WARN_ON(max_sectors == 0);
8066 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8067 * in a sector_t, and (max_sectors>>scale) will fit in a
8068 * u32, as those are the requirements for sector_div.
8069 * Thus 'scale' must be at least 10
8072 if (sizeof(sector_t) > sizeof(unsigned long)) {
8073 while ( max_sectors/2 > (1ULL<<(scale+32)))
8076 res = (resync>>scale)*1000;
8077 sector_div(res, (u32)((max_sectors>>scale)+1));
8081 int i, x = per_milli/50, y = 20-x;
8082 seq_printf(seq, "[");
8083 for (i = 0; i < x; i++)
8084 seq_printf(seq, "=");
8085 seq_printf(seq, ">");
8086 for (i = 0; i < y; i++)
8087 seq_printf(seq, ".");
8088 seq_printf(seq, "] ");
8090 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8091 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8093 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8095 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8096 "resync" : "recovery"))),
8097 per_milli/10, per_milli % 10,
8098 (unsigned long long) resync/2,
8099 (unsigned long long) max_sectors/2);
8102 * dt: time from mark until now
8103 * db: blocks written from mark until now
8104 * rt: remaining time
8106 * rt is a sector_t, which is always 64bit now. We are keeping
8107 * the original algorithm, but it is not really necessary.
8109 * Original algorithm:
8110 * So we divide before multiply in case it is 32bit and close
8112 * We scale the divisor (db) by 32 to avoid losing precision
8113 * near the end of resync when the number of remaining sectors
8115 * We then divide rt by 32 after multiplying by db to compensate.
8116 * The '+1' avoids division by zero if db is very small.
8118 dt = ((jiffies - mddev->resync_mark) / HZ);
8121 curr_mark_cnt = mddev->curr_mark_cnt;
8122 recovery_active = atomic_read(&mddev->recovery_active);
8123 resync_mark_cnt = mddev->resync_mark_cnt;
8125 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8126 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8128 rt = max_sectors - resync; /* number of remaining sectors */
8129 rt = div64_u64(rt, db/32+1);
8133 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8134 ((unsigned long)rt % 60)/6);
8136 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8140 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8142 struct list_head *tmp;
8144 struct mddev *mddev;
8156 spin_lock(&all_mddevs_lock);
8157 list_for_each(tmp,&all_mddevs)
8159 mddev = list_entry(tmp, struct mddev, all_mddevs);
8161 spin_unlock(&all_mddevs_lock);
8164 spin_unlock(&all_mddevs_lock);
8166 return (void*)2;/* tail */
8170 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8172 struct list_head *tmp;
8173 struct mddev *next_mddev, *mddev = v;
8179 spin_lock(&all_mddevs_lock);
8181 tmp = all_mddevs.next;
8183 tmp = mddev->all_mddevs.next;
8184 if (tmp != &all_mddevs)
8185 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8187 next_mddev = (void*)2;
8190 spin_unlock(&all_mddevs_lock);
8198 static void md_seq_stop(struct seq_file *seq, void *v)
8200 struct mddev *mddev = v;
8202 if (mddev && v != (void*)1 && v != (void*)2)
8206 static int md_seq_show(struct seq_file *seq, void *v)
8208 struct mddev *mddev = v;
8210 struct md_rdev *rdev;
8212 if (v == (void*)1) {
8213 struct md_personality *pers;
8214 seq_printf(seq, "Personalities : ");
8215 spin_lock(&pers_lock);
8216 list_for_each_entry(pers, &pers_list, list)
8217 seq_printf(seq, "[%s] ", pers->name);
8219 spin_unlock(&pers_lock);
8220 seq_printf(seq, "\n");
8221 seq->poll_event = atomic_read(&md_event_count);
8224 if (v == (void*)2) {
8229 spin_lock(&mddev->lock);
8230 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8231 seq_printf(seq, "%s : %sactive", mdname(mddev),
8232 mddev->pers ? "" : "in");
8235 seq_printf(seq, " (read-only)");
8237 seq_printf(seq, " (auto-read-only)");
8238 seq_printf(seq, " %s", mddev->pers->name);
8243 rdev_for_each_rcu(rdev, mddev) {
8244 char b[BDEVNAME_SIZE];
8245 seq_printf(seq, " %s[%d]",
8246 bdevname(rdev->bdev,b), rdev->desc_nr);
8247 if (test_bit(WriteMostly, &rdev->flags))
8248 seq_printf(seq, "(W)");
8249 if (test_bit(Journal, &rdev->flags))
8250 seq_printf(seq, "(J)");
8251 if (test_bit(Faulty, &rdev->flags)) {
8252 seq_printf(seq, "(F)");
8255 if (rdev->raid_disk < 0)
8256 seq_printf(seq, "(S)"); /* spare */
8257 if (test_bit(Replacement, &rdev->flags))
8258 seq_printf(seq, "(R)");
8259 sectors += rdev->sectors;
8263 if (!list_empty(&mddev->disks)) {
8265 seq_printf(seq, "\n %llu blocks",
8266 (unsigned long long)
8267 mddev->array_sectors / 2);
8269 seq_printf(seq, "\n %llu blocks",
8270 (unsigned long long)sectors / 2);
8272 if (mddev->persistent) {
8273 if (mddev->major_version != 0 ||
8274 mddev->minor_version != 90) {
8275 seq_printf(seq," super %d.%d",
8276 mddev->major_version,
8277 mddev->minor_version);
8279 } else if (mddev->external)
8280 seq_printf(seq, " super external:%s",
8281 mddev->metadata_type);
8283 seq_printf(seq, " super non-persistent");
8286 mddev->pers->status(seq, mddev);
8287 seq_printf(seq, "\n ");
8288 if (mddev->pers->sync_request) {
8289 if (status_resync(seq, mddev))
8290 seq_printf(seq, "\n ");
8293 seq_printf(seq, "\n ");
8295 md_bitmap_status(seq, mddev->bitmap);
8297 seq_printf(seq, "\n");
8299 spin_unlock(&mddev->lock);
8304 static const struct seq_operations md_seq_ops = {
8305 .start = md_seq_start,
8306 .next = md_seq_next,
8307 .stop = md_seq_stop,
8308 .show = md_seq_show,
8311 static int md_seq_open(struct inode *inode, struct file *file)
8313 struct seq_file *seq;
8316 error = seq_open(file, &md_seq_ops);
8320 seq = file->private_data;
8321 seq->poll_event = atomic_read(&md_event_count);
8325 static int md_unloading;
8326 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8328 struct seq_file *seq = filp->private_data;
8332 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8333 poll_wait(filp, &md_event_waiters, wait);
8335 /* always allow read */
8336 mask = EPOLLIN | EPOLLRDNORM;
8338 if (seq->poll_event != atomic_read(&md_event_count))
8339 mask |= EPOLLERR | EPOLLPRI;
8343 static const struct proc_ops mdstat_proc_ops = {
8344 .proc_open = md_seq_open,
8345 .proc_read = seq_read,
8346 .proc_lseek = seq_lseek,
8347 .proc_release = seq_release,
8348 .proc_poll = mdstat_poll,
8351 int register_md_personality(struct md_personality *p)
8353 pr_debug("md: %s personality registered for level %d\n",
8355 spin_lock(&pers_lock);
8356 list_add_tail(&p->list, &pers_list);
8357 spin_unlock(&pers_lock);
8360 EXPORT_SYMBOL(register_md_personality);
8362 int unregister_md_personality(struct md_personality *p)
8364 pr_debug("md: %s personality unregistered\n", p->name);
8365 spin_lock(&pers_lock);
8366 list_del_init(&p->list);
8367 spin_unlock(&pers_lock);
8370 EXPORT_SYMBOL(unregister_md_personality);
8372 int register_md_cluster_operations(struct md_cluster_operations *ops,
8373 struct module *module)
8376 spin_lock(&pers_lock);
8377 if (md_cluster_ops != NULL)
8380 md_cluster_ops = ops;
8381 md_cluster_mod = module;
8383 spin_unlock(&pers_lock);
8386 EXPORT_SYMBOL(register_md_cluster_operations);
8388 int unregister_md_cluster_operations(void)
8390 spin_lock(&pers_lock);
8391 md_cluster_ops = NULL;
8392 spin_unlock(&pers_lock);
8395 EXPORT_SYMBOL(unregister_md_cluster_operations);
8397 int md_setup_cluster(struct mddev *mddev, int nodes)
8400 if (!md_cluster_ops)
8401 request_module("md-cluster");
8402 spin_lock(&pers_lock);
8403 /* ensure module won't be unloaded */
8404 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8405 pr_warn("can't find md-cluster module or get it's reference.\n");
8406 spin_unlock(&pers_lock);
8409 spin_unlock(&pers_lock);
8411 ret = md_cluster_ops->join(mddev, nodes);
8413 mddev->safemode_delay = 0;
8417 void md_cluster_stop(struct mddev *mddev)
8419 if (!md_cluster_ops)
8421 md_cluster_ops->leave(mddev);
8422 module_put(md_cluster_mod);
8425 static int is_mddev_idle(struct mddev *mddev, int init)
8427 struct md_rdev *rdev;
8433 rdev_for_each_rcu(rdev, mddev) {
8434 struct gendisk *disk = rdev->bdev->bd_disk;
8435 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8436 atomic_read(&disk->sync_io);
8437 /* sync IO will cause sync_io to increase before the disk_stats
8438 * as sync_io is counted when a request starts, and
8439 * disk_stats is counted when it completes.
8440 * So resync activity will cause curr_events to be smaller than
8441 * when there was no such activity.
8442 * non-sync IO will cause disk_stat to increase without
8443 * increasing sync_io so curr_events will (eventually)
8444 * be larger than it was before. Once it becomes
8445 * substantially larger, the test below will cause
8446 * the array to appear non-idle, and resync will slow
8448 * If there is a lot of outstanding resync activity when
8449 * we set last_event to curr_events, then all that activity
8450 * completing might cause the array to appear non-idle
8451 * and resync will be slowed down even though there might
8452 * not have been non-resync activity. This will only
8453 * happen once though. 'last_events' will soon reflect
8454 * the state where there is little or no outstanding
8455 * resync requests, and further resync activity will
8456 * always make curr_events less than last_events.
8459 if (init || curr_events - rdev->last_events > 64) {
8460 rdev->last_events = curr_events;
8468 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8470 /* another "blocks" (512byte) blocks have been synced */
8471 atomic_sub(blocks, &mddev->recovery_active);
8472 wake_up(&mddev->recovery_wait);
8474 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8475 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8476 md_wakeup_thread(mddev->thread);
8477 // stop recovery, signal do_sync ....
8480 EXPORT_SYMBOL(md_done_sync);
8482 /* md_write_start(mddev, bi)
8483 * If we need to update some array metadata (e.g. 'active' flag
8484 * in superblock) before writing, schedule a superblock update
8485 * and wait for it to complete.
8486 * A return value of 'false' means that the write wasn't recorded
8487 * and cannot proceed as the array is being suspend.
8489 bool md_write_start(struct mddev *mddev, struct bio *bi)
8493 if (bio_data_dir(bi) != WRITE)
8496 BUG_ON(mddev->ro == 1);
8497 if (mddev->ro == 2) {
8498 /* need to switch to read/write */
8500 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8501 md_wakeup_thread(mddev->thread);
8502 md_wakeup_thread(mddev->sync_thread);
8506 percpu_ref_get(&mddev->writes_pending);
8507 smp_mb(); /* Match smp_mb in set_in_sync() */
8508 if (mddev->safemode == 1)
8509 mddev->safemode = 0;
8510 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8511 if (mddev->in_sync || mddev->sync_checkers) {
8512 spin_lock(&mddev->lock);
8513 if (mddev->in_sync) {
8515 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8516 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8517 md_wakeup_thread(mddev->thread);
8520 spin_unlock(&mddev->lock);
8524 sysfs_notify_dirent_safe(mddev->sysfs_state);
8525 if (!mddev->has_superblocks)
8527 wait_event(mddev->sb_wait,
8528 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8530 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8531 percpu_ref_put(&mddev->writes_pending);
8536 EXPORT_SYMBOL(md_write_start);
8538 /* md_write_inc can only be called when md_write_start() has
8539 * already been called at least once of the current request.
8540 * It increments the counter and is useful when a single request
8541 * is split into several parts. Each part causes an increment and
8542 * so needs a matching md_write_end().
8543 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8544 * a spinlocked region.
8546 void md_write_inc(struct mddev *mddev, struct bio *bi)
8548 if (bio_data_dir(bi) != WRITE)
8550 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8551 percpu_ref_get(&mddev->writes_pending);
8553 EXPORT_SYMBOL(md_write_inc);
8555 void md_write_end(struct mddev *mddev)
8557 percpu_ref_put(&mddev->writes_pending);
8559 if (mddev->safemode == 2)
8560 md_wakeup_thread(mddev->thread);
8561 else if (mddev->safemode_delay)
8562 /* The roundup() ensures this only performs locking once
8563 * every ->safemode_delay jiffies
8565 mod_timer(&mddev->safemode_timer,
8566 roundup(jiffies, mddev->safemode_delay) +
8567 mddev->safemode_delay);
8570 EXPORT_SYMBOL(md_write_end);
8572 /* This is used by raid0 and raid10 */
8573 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8574 struct bio *bio, sector_t start, sector_t size)
8576 struct bio *discard_bio = NULL;
8578 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO, 0,
8579 &discard_bio) || !discard_bio)
8582 bio_chain(discard_bio, bio);
8583 bio_clone_blkg_association(discard_bio, bio);
8585 trace_block_bio_remap(discard_bio,
8586 disk_devt(mddev->gendisk),
8587 bio->bi_iter.bi_sector);
8588 submit_bio_noacct(discard_bio);
8590 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8592 static void md_end_io_acct(struct bio *bio)
8594 struct md_io_acct *md_io_acct = bio->bi_private;
8595 struct bio *orig_bio = md_io_acct->orig_bio;
8597 orig_bio->bi_status = bio->bi_status;
8599 bio_end_io_acct(orig_bio, md_io_acct->start_time);
8601 bio_endio(orig_bio);
8605 * Used by personalities that don't already clone the bio and thus can't
8606 * easily add the timestamp to their extended bio structure.
8608 void md_account_bio(struct mddev *mddev, struct bio **bio)
8610 struct md_io_acct *md_io_acct;
8613 if (!blk_queue_io_stat((*bio)->bi_bdev->bd_disk->queue))
8616 clone = bio_clone_fast(*bio, GFP_NOIO, &mddev->io_acct_set);
8617 md_io_acct = container_of(clone, struct md_io_acct, bio_clone);
8618 md_io_acct->orig_bio = *bio;
8619 md_io_acct->start_time = bio_start_io_acct(*bio);
8621 clone->bi_end_io = md_end_io_acct;
8622 clone->bi_private = md_io_acct;
8625 EXPORT_SYMBOL_GPL(md_account_bio);
8627 /* md_allow_write(mddev)
8628 * Calling this ensures that the array is marked 'active' so that writes
8629 * may proceed without blocking. It is important to call this before
8630 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8631 * Must be called with mddev_lock held.
8633 void md_allow_write(struct mddev *mddev)
8639 if (!mddev->pers->sync_request)
8642 spin_lock(&mddev->lock);
8643 if (mddev->in_sync) {
8645 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8646 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8647 if (mddev->safemode_delay &&
8648 mddev->safemode == 0)
8649 mddev->safemode = 1;
8650 spin_unlock(&mddev->lock);
8651 md_update_sb(mddev, 0);
8652 sysfs_notify_dirent_safe(mddev->sysfs_state);
8653 /* wait for the dirty state to be recorded in the metadata */
8654 wait_event(mddev->sb_wait,
8655 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8657 spin_unlock(&mddev->lock);
8659 EXPORT_SYMBOL_GPL(md_allow_write);
8661 #define SYNC_MARKS 10
8662 #define SYNC_MARK_STEP (3*HZ)
8663 #define UPDATE_FREQUENCY (5*60*HZ)
8664 void md_do_sync(struct md_thread *thread)
8666 struct mddev *mddev = thread->mddev;
8667 struct mddev *mddev2;
8668 unsigned int currspeed = 0, window;
8669 sector_t max_sectors,j, io_sectors, recovery_done;
8670 unsigned long mark[SYNC_MARKS];
8671 unsigned long update_time;
8672 sector_t mark_cnt[SYNC_MARKS];
8674 struct list_head *tmp;
8675 sector_t last_check;
8677 struct md_rdev *rdev;
8678 char *desc, *action = NULL;
8679 struct blk_plug plug;
8682 /* just incase thread restarts... */
8683 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8684 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8686 if (mddev->ro) {/* never try to sync a read-only array */
8687 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8691 if (mddev_is_clustered(mddev)) {
8692 ret = md_cluster_ops->resync_start(mddev);
8696 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8697 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8698 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8699 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8700 && ((unsigned long long)mddev->curr_resync_completed
8701 < (unsigned long long)mddev->resync_max_sectors))
8705 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8706 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8707 desc = "data-check";
8709 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8710 desc = "requested-resync";
8714 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8719 mddev->last_sync_action = action ?: desc;
8721 /* we overload curr_resync somewhat here.
8722 * 0 == not engaged in resync at all
8723 * 2 == checking that there is no conflict with another sync
8724 * 1 == like 2, but have yielded to allow conflicting resync to
8726 * other == active in resync - this many blocks
8728 * Before starting a resync we must have set curr_resync to
8729 * 2, and then checked that every "conflicting" array has curr_resync
8730 * less than ours. When we find one that is the same or higher
8731 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8732 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8733 * This will mean we have to start checking from the beginning again.
8738 int mddev2_minor = -1;
8739 mddev->curr_resync = 2;
8742 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8744 for_each_mddev(mddev2, tmp) {
8745 if (mddev2 == mddev)
8747 if (!mddev->parallel_resync
8748 && mddev2->curr_resync
8749 && match_mddev_units(mddev, mddev2)) {
8751 if (mddev < mddev2 && mddev->curr_resync == 2) {
8752 /* arbitrarily yield */
8753 mddev->curr_resync = 1;
8754 wake_up(&resync_wait);
8756 if (mddev > mddev2 && mddev->curr_resync == 1)
8757 /* no need to wait here, we can wait the next
8758 * time 'round when curr_resync == 2
8761 /* We need to wait 'interruptible' so as not to
8762 * contribute to the load average, and not to
8763 * be caught by 'softlockup'
8765 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8766 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8767 mddev2->curr_resync >= mddev->curr_resync) {
8768 if (mddev2_minor != mddev2->md_minor) {
8769 mddev2_minor = mddev2->md_minor;
8770 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8771 desc, mdname(mddev),
8775 if (signal_pending(current))
8776 flush_signals(current);
8778 finish_wait(&resync_wait, &wq);
8781 finish_wait(&resync_wait, &wq);
8784 } while (mddev->curr_resync < 2);
8787 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8788 /* resync follows the size requested by the personality,
8789 * which defaults to physical size, but can be virtual size
8791 max_sectors = mddev->resync_max_sectors;
8792 atomic64_set(&mddev->resync_mismatches, 0);
8793 /* we don't use the checkpoint if there's a bitmap */
8794 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8795 j = mddev->resync_min;
8796 else if (!mddev->bitmap)
8797 j = mddev->recovery_cp;
8799 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8800 max_sectors = mddev->resync_max_sectors;
8802 * If the original node aborts reshaping then we continue the
8803 * reshaping, so set j again to avoid restart reshape from the
8806 if (mddev_is_clustered(mddev) &&
8807 mddev->reshape_position != MaxSector)
8808 j = mddev->reshape_position;
8810 /* recovery follows the physical size of devices */
8811 max_sectors = mddev->dev_sectors;
8814 rdev_for_each_rcu(rdev, mddev)
8815 if (rdev->raid_disk >= 0 &&
8816 !test_bit(Journal, &rdev->flags) &&
8817 !test_bit(Faulty, &rdev->flags) &&
8818 !test_bit(In_sync, &rdev->flags) &&
8819 rdev->recovery_offset < j)
8820 j = rdev->recovery_offset;
8823 /* If there is a bitmap, we need to make sure all
8824 * writes that started before we added a spare
8825 * complete before we start doing a recovery.
8826 * Otherwise the write might complete and (via
8827 * bitmap_endwrite) set a bit in the bitmap after the
8828 * recovery has checked that bit and skipped that
8831 if (mddev->bitmap) {
8832 mddev->pers->quiesce(mddev, 1);
8833 mddev->pers->quiesce(mddev, 0);
8837 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8838 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8839 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8840 speed_max(mddev), desc);
8842 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8845 for (m = 0; m < SYNC_MARKS; m++) {
8847 mark_cnt[m] = io_sectors;
8850 mddev->resync_mark = mark[last_mark];
8851 mddev->resync_mark_cnt = mark_cnt[last_mark];
8854 * Tune reconstruction:
8856 window = 32 * (PAGE_SIZE / 512);
8857 pr_debug("md: using %dk window, over a total of %lluk.\n",
8858 window/2, (unsigned long long)max_sectors/2);
8860 atomic_set(&mddev->recovery_active, 0);
8864 pr_debug("md: resuming %s of %s from checkpoint.\n",
8865 desc, mdname(mddev));
8866 mddev->curr_resync = j;
8868 mddev->curr_resync = 3; /* no longer delayed */
8869 mddev->curr_resync_completed = j;
8870 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8871 md_new_event(mddev);
8872 update_time = jiffies;
8874 blk_start_plug(&plug);
8875 while (j < max_sectors) {
8880 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8881 ((mddev->curr_resync > mddev->curr_resync_completed &&
8882 (mddev->curr_resync - mddev->curr_resync_completed)
8883 > (max_sectors >> 4)) ||
8884 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8885 (j - mddev->curr_resync_completed)*2
8886 >= mddev->resync_max - mddev->curr_resync_completed ||
8887 mddev->curr_resync_completed > mddev->resync_max
8889 /* time to update curr_resync_completed */
8890 wait_event(mddev->recovery_wait,
8891 atomic_read(&mddev->recovery_active) == 0);
8892 mddev->curr_resync_completed = j;
8893 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8894 j > mddev->recovery_cp)
8895 mddev->recovery_cp = j;
8896 update_time = jiffies;
8897 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8898 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8901 while (j >= mddev->resync_max &&
8902 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8903 /* As this condition is controlled by user-space,
8904 * we can block indefinitely, so use '_interruptible'
8905 * to avoid triggering warnings.
8907 flush_signals(current); /* just in case */
8908 wait_event_interruptible(mddev->recovery_wait,
8909 mddev->resync_max > j
8910 || test_bit(MD_RECOVERY_INTR,
8914 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8917 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8919 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8923 if (!skipped) { /* actual IO requested */
8924 io_sectors += sectors;
8925 atomic_add(sectors, &mddev->recovery_active);
8928 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8932 if (j > max_sectors)
8933 /* when skipping, extra large numbers can be returned. */
8936 mddev->curr_resync = j;
8937 mddev->curr_mark_cnt = io_sectors;
8938 if (last_check == 0)
8939 /* this is the earliest that rebuild will be
8940 * visible in /proc/mdstat
8942 md_new_event(mddev);
8944 if (last_check + window > io_sectors || j == max_sectors)
8947 last_check = io_sectors;
8949 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8951 int next = (last_mark+1) % SYNC_MARKS;
8953 mddev->resync_mark = mark[next];
8954 mddev->resync_mark_cnt = mark_cnt[next];
8955 mark[next] = jiffies;
8956 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8960 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8964 * this loop exits only if either when we are slower than
8965 * the 'hard' speed limit, or the system was IO-idle for
8967 * the system might be non-idle CPU-wise, but we only care
8968 * about not overloading the IO subsystem. (things like an
8969 * e2fsck being done on the RAID array should execute fast)
8973 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8974 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8975 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8977 if (currspeed > speed_min(mddev)) {
8978 if (currspeed > speed_max(mddev)) {
8982 if (!is_mddev_idle(mddev, 0)) {
8984 * Give other IO more of a chance.
8985 * The faster the devices, the less we wait.
8987 wait_event(mddev->recovery_wait,
8988 !atomic_read(&mddev->recovery_active));
8992 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8993 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8994 ? "interrupted" : "done");
8996 * this also signals 'finished resyncing' to md_stop
8998 blk_finish_plug(&plug);
8999 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9001 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9002 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9003 mddev->curr_resync > 3) {
9004 mddev->curr_resync_completed = mddev->curr_resync;
9005 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9007 mddev->pers->sync_request(mddev, max_sectors, &skipped);
9009 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9010 mddev->curr_resync > 3) {
9011 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9012 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9013 if (mddev->curr_resync >= mddev->recovery_cp) {
9014 pr_debug("md: checkpointing %s of %s.\n",
9015 desc, mdname(mddev));
9016 if (test_bit(MD_RECOVERY_ERROR,
9018 mddev->recovery_cp =
9019 mddev->curr_resync_completed;
9021 mddev->recovery_cp =
9025 mddev->recovery_cp = MaxSector;
9027 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9028 mddev->curr_resync = MaxSector;
9029 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9030 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9032 rdev_for_each_rcu(rdev, mddev)
9033 if (rdev->raid_disk >= 0 &&
9034 mddev->delta_disks >= 0 &&
9035 !test_bit(Journal, &rdev->flags) &&
9036 !test_bit(Faulty, &rdev->flags) &&
9037 !test_bit(In_sync, &rdev->flags) &&
9038 rdev->recovery_offset < mddev->curr_resync)
9039 rdev->recovery_offset = mddev->curr_resync;
9045 /* set CHANGE_PENDING here since maybe another update is needed,
9046 * so other nodes are informed. It should be harmless for normal
9048 set_mask_bits(&mddev->sb_flags, 0,
9049 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9051 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9052 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9053 mddev->delta_disks > 0 &&
9054 mddev->pers->finish_reshape &&
9055 mddev->pers->size &&
9057 mddev_lock_nointr(mddev);
9058 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9059 mddev_unlock(mddev);
9060 if (!mddev_is_clustered(mddev))
9061 set_capacity_and_notify(mddev->gendisk,
9062 mddev->array_sectors);
9065 spin_lock(&mddev->lock);
9066 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9067 /* We completed so min/max setting can be forgotten if used. */
9068 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9069 mddev->resync_min = 0;
9070 mddev->resync_max = MaxSector;
9071 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9072 mddev->resync_min = mddev->curr_resync_completed;
9073 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9074 mddev->curr_resync = 0;
9075 spin_unlock(&mddev->lock);
9077 wake_up(&resync_wait);
9078 md_wakeup_thread(mddev->thread);
9081 EXPORT_SYMBOL_GPL(md_do_sync);
9083 static int remove_and_add_spares(struct mddev *mddev,
9084 struct md_rdev *this)
9086 struct md_rdev *rdev;
9089 bool remove_some = false;
9091 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9092 /* Mustn't remove devices when resync thread is running */
9095 rdev_for_each(rdev, mddev) {
9096 if ((this == NULL || rdev == this) &&
9097 rdev->raid_disk >= 0 &&
9098 !test_bit(Blocked, &rdev->flags) &&
9099 test_bit(Faulty, &rdev->flags) &&
9100 atomic_read(&rdev->nr_pending)==0) {
9101 /* Faulty non-Blocked devices with nr_pending == 0
9102 * never get nr_pending incremented,
9103 * never get Faulty cleared, and never get Blocked set.
9104 * So we can synchronize_rcu now rather than once per device
9107 set_bit(RemoveSynchronized, &rdev->flags);
9113 rdev_for_each(rdev, mddev) {
9114 if ((this == NULL || rdev == this) &&
9115 rdev->raid_disk >= 0 &&
9116 !test_bit(Blocked, &rdev->flags) &&
9117 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9118 (!test_bit(In_sync, &rdev->flags) &&
9119 !test_bit(Journal, &rdev->flags))) &&
9120 atomic_read(&rdev->nr_pending)==0)) {
9121 if (mddev->pers->hot_remove_disk(
9122 mddev, rdev) == 0) {
9123 sysfs_unlink_rdev(mddev, rdev);
9124 rdev->saved_raid_disk = rdev->raid_disk;
9125 rdev->raid_disk = -1;
9129 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9130 clear_bit(RemoveSynchronized, &rdev->flags);
9133 if (removed && mddev->kobj.sd)
9134 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9136 if (this && removed)
9139 rdev_for_each(rdev, mddev) {
9140 if (this && this != rdev)
9142 if (test_bit(Candidate, &rdev->flags))
9144 if (rdev->raid_disk >= 0 &&
9145 !test_bit(In_sync, &rdev->flags) &&
9146 !test_bit(Journal, &rdev->flags) &&
9147 !test_bit(Faulty, &rdev->flags))
9149 if (rdev->raid_disk >= 0)
9151 if (test_bit(Faulty, &rdev->flags))
9153 if (!test_bit(Journal, &rdev->flags)) {
9155 ! (rdev->saved_raid_disk >= 0 &&
9156 !test_bit(Bitmap_sync, &rdev->flags)))
9159 rdev->recovery_offset = 0;
9161 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9162 /* failure here is OK */
9163 sysfs_link_rdev(mddev, rdev);
9164 if (!test_bit(Journal, &rdev->flags))
9166 md_new_event(mddev);
9167 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9172 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9176 static void md_start_sync(struct work_struct *ws)
9178 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9180 mddev->sync_thread = md_register_thread(md_do_sync,
9183 if (!mddev->sync_thread) {
9184 pr_warn("%s: could not start resync thread...\n",
9186 /* leave the spares where they are, it shouldn't hurt */
9187 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9188 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9189 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9190 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9191 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9192 wake_up(&resync_wait);
9193 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9195 if (mddev->sysfs_action)
9196 sysfs_notify_dirent_safe(mddev->sysfs_action);
9198 md_wakeup_thread(mddev->sync_thread);
9199 sysfs_notify_dirent_safe(mddev->sysfs_action);
9200 md_new_event(mddev);
9204 * This routine is regularly called by all per-raid-array threads to
9205 * deal with generic issues like resync and super-block update.
9206 * Raid personalities that don't have a thread (linear/raid0) do not
9207 * need this as they never do any recovery or update the superblock.
9209 * It does not do any resync itself, but rather "forks" off other threads
9210 * to do that as needed.
9211 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9212 * "->recovery" and create a thread at ->sync_thread.
9213 * When the thread finishes it sets MD_RECOVERY_DONE
9214 * and wakeups up this thread which will reap the thread and finish up.
9215 * This thread also removes any faulty devices (with nr_pending == 0).
9217 * The overall approach is:
9218 * 1/ if the superblock needs updating, update it.
9219 * 2/ If a recovery thread is running, don't do anything else.
9220 * 3/ If recovery has finished, clean up, possibly marking spares active.
9221 * 4/ If there are any faulty devices, remove them.
9222 * 5/ If array is degraded, try to add spares devices
9223 * 6/ If array has spares or is not in-sync, start a resync thread.
9225 void md_check_recovery(struct mddev *mddev)
9227 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9228 /* Write superblock - thread that called mddev_suspend()
9229 * holds reconfig_mutex for us.
9231 set_bit(MD_UPDATING_SB, &mddev->flags);
9232 smp_mb__after_atomic();
9233 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9234 md_update_sb(mddev, 0);
9235 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9236 wake_up(&mddev->sb_wait);
9239 if (mddev->suspended)
9243 md_bitmap_daemon_work(mddev);
9245 if (signal_pending(current)) {
9246 if (mddev->pers->sync_request && !mddev->external) {
9247 pr_debug("md: %s in immediate safe mode\n",
9249 mddev->safemode = 2;
9251 flush_signals(current);
9254 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9257 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9258 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9259 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9260 (mddev->external == 0 && mddev->safemode == 1) ||
9261 (mddev->safemode == 2
9262 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9266 if (mddev_trylock(mddev)) {
9268 bool try_set_sync = mddev->safemode != 0;
9270 if (!mddev->external && mddev->safemode == 1)
9271 mddev->safemode = 0;
9274 struct md_rdev *rdev;
9275 if (!mddev->external && mddev->in_sync)
9276 /* 'Blocked' flag not needed as failed devices
9277 * will be recorded if array switched to read/write.
9278 * Leaving it set will prevent the device
9279 * from being removed.
9281 rdev_for_each(rdev, mddev)
9282 clear_bit(Blocked, &rdev->flags);
9283 /* On a read-only array we can:
9284 * - remove failed devices
9285 * - add already-in_sync devices if the array itself
9287 * As we only add devices that are already in-sync,
9288 * we can activate the spares immediately.
9290 remove_and_add_spares(mddev, NULL);
9291 /* There is no thread, but we need to call
9292 * ->spare_active and clear saved_raid_disk
9294 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9295 md_reap_sync_thread(mddev);
9296 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9297 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9298 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9302 if (mddev_is_clustered(mddev)) {
9303 struct md_rdev *rdev, *tmp;
9304 /* kick the device if another node issued a
9307 rdev_for_each_safe(rdev, tmp, mddev) {
9308 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9309 rdev->raid_disk < 0)
9310 md_kick_rdev_from_array(rdev);
9314 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9315 spin_lock(&mddev->lock);
9317 spin_unlock(&mddev->lock);
9320 if (mddev->sb_flags)
9321 md_update_sb(mddev, 0);
9323 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9324 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9325 /* resync/recovery still happening */
9326 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9329 if (mddev->sync_thread) {
9330 md_reap_sync_thread(mddev);
9333 /* Set RUNNING before clearing NEEDED to avoid
9334 * any transients in the value of "sync_action".
9336 mddev->curr_resync_completed = 0;
9337 spin_lock(&mddev->lock);
9338 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9339 spin_unlock(&mddev->lock);
9340 /* Clear some bits that don't mean anything, but
9343 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9344 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9346 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9347 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9349 /* no recovery is running.
9350 * remove any failed drives, then
9351 * add spares if possible.
9352 * Spares are also removed and re-added, to allow
9353 * the personality to fail the re-add.
9356 if (mddev->reshape_position != MaxSector) {
9357 if (mddev->pers->check_reshape == NULL ||
9358 mddev->pers->check_reshape(mddev) != 0)
9359 /* Cannot proceed */
9361 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9362 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9363 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9364 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9365 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9366 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9367 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9368 } else if (mddev->recovery_cp < MaxSector) {
9369 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9370 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9371 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9372 /* nothing to be done ... */
9375 if (mddev->pers->sync_request) {
9377 /* We are adding a device or devices to an array
9378 * which has the bitmap stored on all devices.
9379 * So make sure all bitmap pages get written
9381 md_bitmap_write_all(mddev->bitmap);
9383 INIT_WORK(&mddev->del_work, md_start_sync);
9384 queue_work(md_misc_wq, &mddev->del_work);
9388 if (!mddev->sync_thread) {
9389 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9390 wake_up(&resync_wait);
9391 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9393 if (mddev->sysfs_action)
9394 sysfs_notify_dirent_safe(mddev->sysfs_action);
9397 wake_up(&mddev->sb_wait);
9398 mddev_unlock(mddev);
9401 EXPORT_SYMBOL(md_check_recovery);
9403 void md_reap_sync_thread(struct mddev *mddev)
9405 struct md_rdev *rdev;
9406 sector_t old_dev_sectors = mddev->dev_sectors;
9407 bool is_reshaped = false;
9409 /* resync has finished, collect result */
9410 md_unregister_thread(&mddev->sync_thread);
9411 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9412 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9413 mddev->degraded != mddev->raid_disks) {
9415 /* activate any spares */
9416 if (mddev->pers->spare_active(mddev)) {
9417 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9418 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9421 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9422 mddev->pers->finish_reshape) {
9423 mddev->pers->finish_reshape(mddev);
9424 if (mddev_is_clustered(mddev))
9428 /* If array is no-longer degraded, then any saved_raid_disk
9429 * information must be scrapped.
9431 if (!mddev->degraded)
9432 rdev_for_each(rdev, mddev)
9433 rdev->saved_raid_disk = -1;
9435 md_update_sb(mddev, 1);
9436 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9437 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9439 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9440 md_cluster_ops->resync_finish(mddev);
9441 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9442 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9443 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9444 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9445 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9446 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9448 * We call md_cluster_ops->update_size here because sync_size could
9449 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9450 * so it is time to update size across cluster.
9452 if (mddev_is_clustered(mddev) && is_reshaped
9453 && !test_bit(MD_CLOSING, &mddev->flags))
9454 md_cluster_ops->update_size(mddev, old_dev_sectors);
9455 wake_up(&resync_wait);
9456 /* flag recovery needed just to double check */
9457 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9458 sysfs_notify_dirent_safe(mddev->sysfs_action);
9459 md_new_event(mddev);
9460 if (mddev->event_work.func)
9461 queue_work(md_misc_wq, &mddev->event_work);
9463 EXPORT_SYMBOL(md_reap_sync_thread);
9465 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9467 sysfs_notify_dirent_safe(rdev->sysfs_state);
9468 wait_event_timeout(rdev->blocked_wait,
9469 !test_bit(Blocked, &rdev->flags) &&
9470 !test_bit(BlockedBadBlocks, &rdev->flags),
9471 msecs_to_jiffies(5000));
9472 rdev_dec_pending(rdev, mddev);
9474 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9476 void md_finish_reshape(struct mddev *mddev)
9478 /* called be personality module when reshape completes. */
9479 struct md_rdev *rdev;
9481 rdev_for_each(rdev, mddev) {
9482 if (rdev->data_offset > rdev->new_data_offset)
9483 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9485 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9486 rdev->data_offset = rdev->new_data_offset;
9489 EXPORT_SYMBOL(md_finish_reshape);
9491 /* Bad block management */
9493 /* Returns 1 on success, 0 on failure */
9494 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9497 struct mddev *mddev = rdev->mddev;
9500 s += rdev->new_data_offset;
9502 s += rdev->data_offset;
9503 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9505 /* Make sure they get written out promptly */
9506 if (test_bit(ExternalBbl, &rdev->flags))
9507 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9508 sysfs_notify_dirent_safe(rdev->sysfs_state);
9509 set_mask_bits(&mddev->sb_flags, 0,
9510 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9511 md_wakeup_thread(rdev->mddev->thread);
9516 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9518 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9523 s += rdev->new_data_offset;
9525 s += rdev->data_offset;
9526 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9527 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9528 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9531 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9533 static int md_notify_reboot(struct notifier_block *this,
9534 unsigned long code, void *x)
9536 struct list_head *tmp;
9537 struct mddev *mddev;
9540 for_each_mddev(mddev, tmp) {
9541 if (mddev_trylock(mddev)) {
9543 __md_stop_writes(mddev);
9544 if (mddev->persistent)
9545 mddev->safemode = 2;
9546 mddev_unlock(mddev);
9551 * certain more exotic SCSI devices are known to be
9552 * volatile wrt too early system reboots. While the
9553 * right place to handle this issue is the given
9554 * driver, we do want to have a safe RAID driver ...
9562 static struct notifier_block md_notifier = {
9563 .notifier_call = md_notify_reboot,
9565 .priority = INT_MAX, /* before any real devices */
9568 static void md_geninit(void)
9570 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9572 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9575 static int __init md_init(void)
9579 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9583 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9587 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9588 if (!md_rdev_misc_wq)
9589 goto err_rdev_misc_wq;
9591 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9595 ret = __register_blkdev(0, "mdp", md_probe);
9600 register_reboot_notifier(&md_notifier);
9601 raid_table_header = register_sysctl_table(raid_root_table);
9607 unregister_blkdev(MD_MAJOR, "md");
9609 destroy_workqueue(md_rdev_misc_wq);
9611 destroy_workqueue(md_misc_wq);
9613 destroy_workqueue(md_wq);
9618 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9620 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9621 struct md_rdev *rdev2, *tmp;
9623 char b[BDEVNAME_SIZE];
9626 * If size is changed in another node then we need to
9627 * do resize as well.
9629 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9630 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9632 pr_info("md-cluster: resize failed\n");
9634 md_bitmap_update_sb(mddev->bitmap);
9637 /* Check for change of roles in the active devices */
9638 rdev_for_each_safe(rdev2, tmp, mddev) {
9639 if (test_bit(Faulty, &rdev2->flags))
9642 /* Check if the roles changed */
9643 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9645 if (test_bit(Candidate, &rdev2->flags)) {
9646 if (role == 0xfffe) {
9647 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9648 md_kick_rdev_from_array(rdev2);
9652 clear_bit(Candidate, &rdev2->flags);
9655 if (role != rdev2->raid_disk) {
9657 * got activated except reshape is happening.
9659 if (rdev2->raid_disk == -1 && role != 0xffff &&
9660 !(le32_to_cpu(sb->feature_map) &
9661 MD_FEATURE_RESHAPE_ACTIVE)) {
9662 rdev2->saved_raid_disk = role;
9663 ret = remove_and_add_spares(mddev, rdev2);
9664 pr_info("Activated spare: %s\n",
9665 bdevname(rdev2->bdev,b));
9666 /* wakeup mddev->thread here, so array could
9667 * perform resync with the new activated disk */
9668 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9669 md_wakeup_thread(mddev->thread);
9672 * We just want to do the minimum to mark the disk
9673 * as faulty. The recovery is performed by the
9674 * one who initiated the error.
9676 if ((role == 0xfffe) || (role == 0xfffd)) {
9677 md_error(mddev, rdev2);
9678 clear_bit(Blocked, &rdev2->flags);
9683 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9684 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9686 pr_warn("md: updating array disks failed. %d\n", ret);
9690 * Since mddev->delta_disks has already updated in update_raid_disks,
9691 * so it is time to check reshape.
9693 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9694 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9696 * reshape is happening in the remote node, we need to
9697 * update reshape_position and call start_reshape.
9699 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9700 if (mddev->pers->update_reshape_pos)
9701 mddev->pers->update_reshape_pos(mddev);
9702 if (mddev->pers->start_reshape)
9703 mddev->pers->start_reshape(mddev);
9704 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9705 mddev->reshape_position != MaxSector &&
9706 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9707 /* reshape is just done in another node. */
9708 mddev->reshape_position = MaxSector;
9709 if (mddev->pers->update_reshape_pos)
9710 mddev->pers->update_reshape_pos(mddev);
9713 /* Finally set the event to be up to date */
9714 mddev->events = le64_to_cpu(sb->events);
9717 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9720 struct page *swapout = rdev->sb_page;
9721 struct mdp_superblock_1 *sb;
9723 /* Store the sb page of the rdev in the swapout temporary
9724 * variable in case we err in the future
9726 rdev->sb_page = NULL;
9727 err = alloc_disk_sb(rdev);
9729 ClearPageUptodate(rdev->sb_page);
9730 rdev->sb_loaded = 0;
9731 err = super_types[mddev->major_version].
9732 load_super(rdev, NULL, mddev->minor_version);
9735 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9736 __func__, __LINE__, rdev->desc_nr, err);
9738 put_page(rdev->sb_page);
9739 rdev->sb_page = swapout;
9740 rdev->sb_loaded = 1;
9744 sb = page_address(rdev->sb_page);
9745 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9749 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9750 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9752 /* The other node finished recovery, call spare_active to set
9753 * device In_sync and mddev->degraded
9755 if (rdev->recovery_offset == MaxSector &&
9756 !test_bit(In_sync, &rdev->flags) &&
9757 mddev->pers->spare_active(mddev))
9758 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9764 void md_reload_sb(struct mddev *mddev, int nr)
9766 struct md_rdev *rdev;
9770 rdev_for_each_rcu(rdev, mddev) {
9771 if (rdev->desc_nr == nr)
9775 if (!rdev || rdev->desc_nr != nr) {
9776 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9780 err = read_rdev(mddev, rdev);
9784 check_sb_changes(mddev, rdev);
9786 /* Read all rdev's to update recovery_offset */
9787 rdev_for_each_rcu(rdev, mddev) {
9788 if (!test_bit(Faulty, &rdev->flags))
9789 read_rdev(mddev, rdev);
9792 EXPORT_SYMBOL(md_reload_sb);
9797 * Searches all registered partitions for autorun RAID arrays
9801 static DEFINE_MUTEX(detected_devices_mutex);
9802 static LIST_HEAD(all_detected_devices);
9803 struct detected_devices_node {
9804 struct list_head list;
9808 void md_autodetect_dev(dev_t dev)
9810 struct detected_devices_node *node_detected_dev;
9812 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9813 if (node_detected_dev) {
9814 node_detected_dev->dev = dev;
9815 mutex_lock(&detected_devices_mutex);
9816 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9817 mutex_unlock(&detected_devices_mutex);
9821 void md_autostart_arrays(int part)
9823 struct md_rdev *rdev;
9824 struct detected_devices_node *node_detected_dev;
9826 int i_scanned, i_passed;
9831 pr_info("md: Autodetecting RAID arrays.\n");
9833 mutex_lock(&detected_devices_mutex);
9834 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9836 node_detected_dev = list_entry(all_detected_devices.next,
9837 struct detected_devices_node, list);
9838 list_del(&node_detected_dev->list);
9839 dev = node_detected_dev->dev;
9840 kfree(node_detected_dev);
9841 mutex_unlock(&detected_devices_mutex);
9842 rdev = md_import_device(dev,0, 90);
9843 mutex_lock(&detected_devices_mutex);
9847 if (test_bit(Faulty, &rdev->flags))
9850 set_bit(AutoDetected, &rdev->flags);
9851 list_add(&rdev->same_set, &pending_raid_disks);
9854 mutex_unlock(&detected_devices_mutex);
9856 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9858 autorun_devices(part);
9861 #endif /* !MODULE */
9863 static __exit void md_exit(void)
9865 struct mddev *mddev;
9866 struct list_head *tmp;
9869 unregister_blkdev(MD_MAJOR,"md");
9870 unregister_blkdev(mdp_major, "mdp");
9871 unregister_reboot_notifier(&md_notifier);
9872 unregister_sysctl_table(raid_table_header);
9874 /* We cannot unload the modules while some process is
9875 * waiting for us in select() or poll() - wake them up
9878 while (waitqueue_active(&md_event_waiters)) {
9879 /* not safe to leave yet */
9880 wake_up(&md_event_waiters);
9884 remove_proc_entry("mdstat", NULL);
9886 for_each_mddev(mddev, tmp) {
9887 export_array(mddev);
9889 mddev->hold_active = 0;
9891 * for_each_mddev() will call mddev_put() at the end of each
9892 * iteration. As the mddev is now fully clear, this will
9893 * schedule the mddev for destruction by a workqueue, and the
9894 * destroy_workqueue() below will wait for that to complete.
9897 destroy_workqueue(md_rdev_misc_wq);
9898 destroy_workqueue(md_misc_wq);
9899 destroy_workqueue(md_wq);
9902 subsys_initcall(md_init);
9903 module_exit(md_exit)
9905 static int get_ro(char *buffer, const struct kernel_param *kp)
9907 return sprintf(buffer, "%d\n", start_readonly);
9909 static int set_ro(const char *val, const struct kernel_param *kp)
9911 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9914 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9915 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9916 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9917 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9919 MODULE_LICENSE("GPL");
9920 MODULE_DESCRIPTION("MD RAID framework");
9922 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);