1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/module.h>
55 #include <linux/reboot.h>
56 #include <linux/file.h>
57 #include <linux/compat.h>
58 #include <linux/delay.h>
59 #include <linux/raid/md_p.h>
60 #include <linux/raid/md_u.h>
61 #include <linux/raid/detect.h>
62 #include <linux/slab.h>
63 #include <linux/percpu-refcount.h>
64 #include <linux/part_stat.h>
66 #include <trace/events/block.h>
68 #include "md-bitmap.h"
69 #include "md-cluster.h"
71 /* pers_list is a list of registered personalities protected
73 * pers_lock does extra service to protect accesses to
74 * mddev->thread when the mutex cannot be held.
76 static LIST_HEAD(pers_list);
77 static DEFINE_SPINLOCK(pers_lock);
79 static struct kobj_type md_ktype;
81 struct md_cluster_operations *md_cluster_ops;
82 EXPORT_SYMBOL(md_cluster_ops);
83 static struct module *md_cluster_mod;
85 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
86 static struct workqueue_struct *md_wq;
87 static struct workqueue_struct *md_misc_wq;
88 static struct workqueue_struct *md_rdev_misc_wq;
90 static int remove_and_add_spares(struct mddev *mddev,
91 struct md_rdev *this);
92 static void mddev_detach(struct mddev *mddev);
95 * Default number of read corrections we'll attempt on an rdev
96 * before ejecting it from the array. We divide the read error
97 * count by 2 for every hour elapsed between read errors.
99 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
100 /* Default safemode delay: 200 msec */
101 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
103 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
104 * is 1000 KB/sec, so the extra system load does not show up that much.
105 * Increase it if you want to have more _guaranteed_ speed. Note that
106 * the RAID driver will use the maximum available bandwidth if the IO
107 * subsystem is idle. There is also an 'absolute maximum' reconstruction
108 * speed limit - in case reconstruction slows down your system despite
111 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
112 * or /sys/block/mdX/md/sync_speed_{min,max}
115 static int sysctl_speed_limit_min = 1000;
116 static int sysctl_speed_limit_max = 200000;
117 static inline int speed_min(struct mddev *mddev)
119 return mddev->sync_speed_min ?
120 mddev->sync_speed_min : sysctl_speed_limit_min;
123 static inline int speed_max(struct mddev *mddev)
125 return mddev->sync_speed_max ?
126 mddev->sync_speed_max : sysctl_speed_limit_max;
129 static void rdev_uninit_serial(struct md_rdev *rdev)
131 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
134 kvfree(rdev->serial);
138 static void rdevs_uninit_serial(struct mddev *mddev)
140 struct md_rdev *rdev;
142 rdev_for_each(rdev, mddev)
143 rdev_uninit_serial(rdev);
146 static int rdev_init_serial(struct md_rdev *rdev)
148 /* serial_nums equals with BARRIER_BUCKETS_NR */
149 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
150 struct serial_in_rdev *serial = NULL;
152 if (test_bit(CollisionCheck, &rdev->flags))
155 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
160 for (i = 0; i < serial_nums; i++) {
161 struct serial_in_rdev *serial_tmp = &serial[i];
163 spin_lock_init(&serial_tmp->serial_lock);
164 serial_tmp->serial_rb = RB_ROOT_CACHED;
165 init_waitqueue_head(&serial_tmp->serial_io_wait);
168 rdev->serial = serial;
169 set_bit(CollisionCheck, &rdev->flags);
174 static int rdevs_init_serial(struct mddev *mddev)
176 struct md_rdev *rdev;
179 rdev_for_each(rdev, mddev) {
180 ret = rdev_init_serial(rdev);
185 /* Free all resources if pool is not existed */
186 if (ret && !mddev->serial_info_pool)
187 rdevs_uninit_serial(mddev);
193 * rdev needs to enable serial stuffs if it meets the conditions:
194 * 1. it is multi-queue device flaged with writemostly.
195 * 2. the write-behind mode is enabled.
197 static int rdev_need_serial(struct md_rdev *rdev)
199 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
200 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
201 test_bit(WriteMostly, &rdev->flags));
205 * Init resource for rdev(s), then create serial_info_pool if:
206 * 1. rdev is the first device which return true from rdev_enable_serial.
207 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
209 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
214 if (rdev && !rdev_need_serial(rdev) &&
215 !test_bit(CollisionCheck, &rdev->flags))
219 mddev_suspend(mddev);
222 ret = rdevs_init_serial(mddev);
224 ret = rdev_init_serial(rdev);
228 if (mddev->serial_info_pool == NULL) {
230 * already in memalloc noio context by
233 mddev->serial_info_pool =
234 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
235 sizeof(struct serial_info));
236 if (!mddev->serial_info_pool) {
237 rdevs_uninit_serial(mddev);
238 pr_err("can't alloc memory pool for serialization\n");
248 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
249 * 1. rdev is the last device flaged with CollisionCheck.
250 * 2. when bitmap is destroyed while policy is not enabled.
251 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
253 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
256 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
259 if (mddev->serial_info_pool) {
260 struct md_rdev *temp;
261 int num = 0; /* used to track if other rdevs need the pool */
264 mddev_suspend(mddev);
265 rdev_for_each(temp, mddev) {
267 if (!mddev->serialize_policy ||
268 !rdev_need_serial(temp))
269 rdev_uninit_serial(temp);
272 } else if (temp != rdev &&
273 test_bit(CollisionCheck, &temp->flags))
278 rdev_uninit_serial(rdev);
281 pr_info("The mempool could be used by other devices\n");
283 mempool_destroy(mddev->serial_info_pool);
284 mddev->serial_info_pool = NULL;
291 static struct ctl_table_header *raid_table_header;
293 static struct ctl_table raid_table[] = {
295 .procname = "speed_limit_min",
296 .data = &sysctl_speed_limit_min,
297 .maxlen = sizeof(int),
298 .mode = S_IRUGO|S_IWUSR,
299 .proc_handler = proc_dointvec,
302 .procname = "speed_limit_max",
303 .data = &sysctl_speed_limit_max,
304 .maxlen = sizeof(int),
305 .mode = S_IRUGO|S_IWUSR,
306 .proc_handler = proc_dointvec,
311 static struct ctl_table raid_dir_table[] = {
315 .mode = S_IRUGO|S_IXUGO,
321 static struct ctl_table raid_root_table[] = {
326 .child = raid_dir_table,
331 static int start_readonly;
334 * The original mechanism for creating an md device is to create
335 * a device node in /dev and to open it. This causes races with device-close.
336 * The preferred method is to write to the "new_array" module parameter.
337 * This can avoid races.
338 * Setting create_on_open to false disables the original mechanism
339 * so all the races disappear.
341 static bool create_on_open = true;
343 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
346 if (!mddev || !bioset_initialized(&mddev->bio_set))
347 return bio_alloc(gfp_mask, nr_iovecs);
349 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
351 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
353 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
355 if (!mddev || !bioset_initialized(&mddev->sync_set))
356 return bio_alloc(GFP_NOIO, 1);
358 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
362 * We have a system wide 'event count' that is incremented
363 * on any 'interesting' event, and readers of /proc/mdstat
364 * can use 'poll' or 'select' to find out when the event
368 * start array, stop array, error, add device, remove device,
369 * start build, activate spare
371 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
372 static atomic_t md_event_count;
373 void md_new_event(struct mddev *mddev)
375 atomic_inc(&md_event_count);
376 wake_up(&md_event_waiters);
378 EXPORT_SYMBOL_GPL(md_new_event);
381 * Enables to iterate over all existing md arrays
382 * all_mddevs_lock protects this list.
384 static LIST_HEAD(all_mddevs);
385 static DEFINE_SPINLOCK(all_mddevs_lock);
388 * iterates through all used mddevs in the system.
389 * We take care to grab the all_mddevs_lock whenever navigating
390 * the list, and to always hold a refcount when unlocked.
391 * Any code which breaks out of this loop while own
392 * a reference to the current mddev and must mddev_put it.
394 #define for_each_mddev(_mddev,_tmp) \
396 for (({ spin_lock(&all_mddevs_lock); \
397 _tmp = all_mddevs.next; \
399 ({ if (_tmp != &all_mddevs) \
400 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
401 spin_unlock(&all_mddevs_lock); \
402 if (_mddev) mddev_put(_mddev); \
403 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
404 _tmp != &all_mddevs;}); \
405 ({ spin_lock(&all_mddevs_lock); \
406 _tmp = _tmp->next;}) \
409 /* Rather than calling directly into the personality make_request function,
410 * IO requests come here first so that we can check if the device is
411 * being suspended pending a reconfiguration.
412 * We hold a refcount over the call to ->make_request. By the time that
413 * call has finished, the bio has been linked into some internal structure
414 * and so is visible to ->quiesce(), so we don't need the refcount any more.
416 static bool is_suspended(struct mddev *mddev, struct bio *bio)
418 if (mddev->suspended)
420 if (bio_data_dir(bio) != WRITE)
422 if (mddev->suspend_lo >= mddev->suspend_hi)
424 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
426 if (bio_end_sector(bio) < mddev->suspend_lo)
431 void md_handle_request(struct mddev *mddev, struct bio *bio)
435 if (is_suspended(mddev, bio)) {
438 prepare_to_wait(&mddev->sb_wait, &__wait,
439 TASK_UNINTERRUPTIBLE);
440 if (!is_suspended(mddev, bio))
446 finish_wait(&mddev->sb_wait, &__wait);
448 atomic_inc(&mddev->active_io);
451 if (!mddev->pers->make_request(mddev, bio)) {
452 atomic_dec(&mddev->active_io);
453 wake_up(&mddev->sb_wait);
454 goto check_suspended;
457 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
458 wake_up(&mddev->sb_wait);
460 EXPORT_SYMBOL(md_handle_request);
462 static blk_qc_t md_submit_bio(struct bio *bio)
464 const int rw = bio_data_dir(bio);
465 const int sgrp = op_stat_group(bio_op(bio));
466 struct mddev *mddev = bio->bi_disk->private_data;
467 unsigned int sectors;
469 if (mddev == NULL || mddev->pers == NULL) {
471 return BLK_QC_T_NONE;
474 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
476 return BLK_QC_T_NONE;
479 blk_queue_split(&bio);
481 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
482 if (bio_sectors(bio) != 0)
483 bio->bi_status = BLK_STS_IOERR;
485 return BLK_QC_T_NONE;
489 * save the sectors now since our bio can
490 * go away inside make_request
492 sectors = bio_sectors(bio);
493 /* bio could be mergeable after passing to underlayer */
494 bio->bi_opf &= ~REQ_NOMERGE;
496 md_handle_request(mddev, bio);
499 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
500 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
503 return BLK_QC_T_NONE;
506 /* mddev_suspend makes sure no new requests are submitted
507 * to the device, and that any requests that have been submitted
508 * are completely handled.
509 * Once mddev_detach() is called and completes, the module will be
512 void mddev_suspend(struct mddev *mddev)
514 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
515 lockdep_assert_held(&mddev->reconfig_mutex);
516 if (mddev->suspended++)
519 wake_up(&mddev->sb_wait);
520 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
521 smp_mb__after_atomic();
522 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
523 mddev->pers->quiesce(mddev, 1);
524 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
525 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
527 del_timer_sync(&mddev->safemode_timer);
528 /* restrict memory reclaim I/O during raid array is suspend */
529 mddev->noio_flag = memalloc_noio_save();
531 EXPORT_SYMBOL_GPL(mddev_suspend);
533 void mddev_resume(struct mddev *mddev)
535 /* entred the memalloc scope from mddev_suspend() */
536 memalloc_noio_restore(mddev->noio_flag);
537 lockdep_assert_held(&mddev->reconfig_mutex);
538 if (--mddev->suspended)
540 wake_up(&mddev->sb_wait);
541 mddev->pers->quiesce(mddev, 0);
543 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
544 md_wakeup_thread(mddev->thread);
545 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
547 EXPORT_SYMBOL_GPL(mddev_resume);
550 * Generic flush handling for md
553 static void md_end_flush(struct bio *bio)
555 struct md_rdev *rdev = bio->bi_private;
556 struct mddev *mddev = rdev->mddev;
558 rdev_dec_pending(rdev, mddev);
560 if (atomic_dec_and_test(&mddev->flush_pending)) {
561 /* The pre-request flush has finished */
562 queue_work(md_wq, &mddev->flush_work);
567 static void md_submit_flush_data(struct work_struct *ws);
569 static void submit_flushes(struct work_struct *ws)
571 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
572 struct md_rdev *rdev;
574 mddev->start_flush = ktime_get_boottime();
575 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
576 atomic_set(&mddev->flush_pending, 1);
578 rdev_for_each_rcu(rdev, mddev)
579 if (rdev->raid_disk >= 0 &&
580 !test_bit(Faulty, &rdev->flags)) {
581 /* Take two references, one is dropped
582 * when request finishes, one after
583 * we reclaim rcu_read_lock
586 atomic_inc(&rdev->nr_pending);
587 atomic_inc(&rdev->nr_pending);
589 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
590 bi->bi_end_io = md_end_flush;
591 bi->bi_private = rdev;
592 bio_set_dev(bi, rdev->bdev);
593 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
594 atomic_inc(&mddev->flush_pending);
597 rdev_dec_pending(rdev, mddev);
600 if (atomic_dec_and_test(&mddev->flush_pending))
601 queue_work(md_wq, &mddev->flush_work);
604 static void md_submit_flush_data(struct work_struct *ws)
606 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
607 struct bio *bio = mddev->flush_bio;
610 * must reset flush_bio before calling into md_handle_request to avoid a
611 * deadlock, because other bios passed md_handle_request suspend check
612 * could wait for this and below md_handle_request could wait for those
613 * bios because of suspend check
615 spin_lock_irq(&mddev->lock);
616 mddev->last_flush = mddev->start_flush;
617 mddev->flush_bio = NULL;
618 spin_unlock_irq(&mddev->lock);
619 wake_up(&mddev->sb_wait);
621 if (bio->bi_iter.bi_size == 0) {
622 /* an empty barrier - all done */
625 bio->bi_opf &= ~REQ_PREFLUSH;
626 md_handle_request(mddev, bio);
631 * Manages consolidation of flushes and submitting any flushes needed for
632 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
633 * being finished in another context. Returns false if the flushing is
634 * complete but still needs the I/O portion of the bio to be processed.
636 bool md_flush_request(struct mddev *mddev, struct bio *bio)
638 ktime_t start = ktime_get_boottime();
639 spin_lock_irq(&mddev->lock);
640 wait_event_lock_irq(mddev->sb_wait,
642 ktime_after(mddev->last_flush, start),
644 if (!ktime_after(mddev->last_flush, start)) {
645 WARN_ON(mddev->flush_bio);
646 mddev->flush_bio = bio;
649 spin_unlock_irq(&mddev->lock);
652 INIT_WORK(&mddev->flush_work, submit_flushes);
653 queue_work(md_wq, &mddev->flush_work);
655 /* flush was performed for some other bio while we waited. */
656 if (bio->bi_iter.bi_size == 0)
657 /* an empty barrier - all done */
660 bio->bi_opf &= ~REQ_PREFLUSH;
666 EXPORT_SYMBOL(md_flush_request);
668 static inline struct mddev *mddev_get(struct mddev *mddev)
670 atomic_inc(&mddev->active);
674 static void mddev_delayed_delete(struct work_struct *ws);
676 static void mddev_put(struct mddev *mddev)
678 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
680 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
681 mddev->ctime == 0 && !mddev->hold_active) {
682 /* Array is not configured at all, and not held active,
684 list_del_init(&mddev->all_mddevs);
687 * Call queue_work inside the spinlock so that
688 * flush_workqueue() after mddev_find will succeed in waiting
689 * for the work to be done.
691 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
692 queue_work(md_misc_wq, &mddev->del_work);
694 spin_unlock(&all_mddevs_lock);
697 static void md_safemode_timeout(struct timer_list *t);
699 void mddev_init(struct mddev *mddev)
701 kobject_init(&mddev->kobj, &md_ktype);
702 mutex_init(&mddev->open_mutex);
703 mutex_init(&mddev->reconfig_mutex);
704 mutex_init(&mddev->bitmap_info.mutex);
705 INIT_LIST_HEAD(&mddev->disks);
706 INIT_LIST_HEAD(&mddev->all_mddevs);
707 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
708 atomic_set(&mddev->active, 1);
709 atomic_set(&mddev->openers, 0);
710 atomic_set(&mddev->active_io, 0);
711 spin_lock_init(&mddev->lock);
712 atomic_set(&mddev->flush_pending, 0);
713 init_waitqueue_head(&mddev->sb_wait);
714 init_waitqueue_head(&mddev->recovery_wait);
715 mddev->reshape_position = MaxSector;
716 mddev->reshape_backwards = 0;
717 mddev->last_sync_action = "none";
718 mddev->resync_min = 0;
719 mddev->resync_max = MaxSector;
720 mddev->level = LEVEL_NONE;
722 EXPORT_SYMBOL_GPL(mddev_init);
724 static struct mddev *mddev_find_locked(dev_t unit)
728 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
729 if (mddev->unit == unit)
735 static struct mddev *mddev_find(dev_t unit)
739 if (MAJOR(unit) != MD_MAJOR)
740 unit &= ~((1 << MdpMinorShift) - 1);
742 spin_lock(&all_mddevs_lock);
743 mddev = mddev_find_locked(unit);
746 spin_unlock(&all_mddevs_lock);
751 static struct mddev *mddev_find_or_alloc(dev_t unit)
753 struct mddev *mddev, *new = NULL;
755 if (unit && MAJOR(unit) != MD_MAJOR)
756 unit &= ~((1<<MdpMinorShift)-1);
759 spin_lock(&all_mddevs_lock);
762 mddev = mddev_find_locked(unit);
765 spin_unlock(&all_mddevs_lock);
771 list_add(&new->all_mddevs, &all_mddevs);
772 spin_unlock(&all_mddevs_lock);
773 new->hold_active = UNTIL_IOCTL;
777 /* find an unused unit number */
778 static int next_minor = 512;
779 int start = next_minor;
783 dev = MKDEV(MD_MAJOR, next_minor);
785 if (next_minor > MINORMASK)
787 if (next_minor == start) {
788 /* Oh dear, all in use. */
789 spin_unlock(&all_mddevs_lock);
794 is_free = !mddev_find_locked(dev);
797 new->md_minor = MINOR(dev);
798 new->hold_active = UNTIL_STOP;
799 list_add(&new->all_mddevs, &all_mddevs);
800 spin_unlock(&all_mddevs_lock);
803 spin_unlock(&all_mddevs_lock);
805 new = kzalloc(sizeof(*new), GFP_KERNEL);
810 if (MAJOR(unit) == MD_MAJOR)
811 new->md_minor = MINOR(unit);
813 new->md_minor = MINOR(unit) >> MdpMinorShift;
820 static struct attribute_group md_redundancy_group;
822 void mddev_unlock(struct mddev *mddev)
824 if (mddev->to_remove) {
825 /* These cannot be removed under reconfig_mutex as
826 * an access to the files will try to take reconfig_mutex
827 * while holding the file unremovable, which leads to
829 * So hold set sysfs_active while the remove in happeing,
830 * and anything else which might set ->to_remove or my
831 * otherwise change the sysfs namespace will fail with
832 * -EBUSY if sysfs_active is still set.
833 * We set sysfs_active under reconfig_mutex and elsewhere
834 * test it under the same mutex to ensure its correct value
837 struct attribute_group *to_remove = mddev->to_remove;
838 mddev->to_remove = NULL;
839 mddev->sysfs_active = 1;
840 mutex_unlock(&mddev->reconfig_mutex);
842 if (mddev->kobj.sd) {
843 if (to_remove != &md_redundancy_group)
844 sysfs_remove_group(&mddev->kobj, to_remove);
845 if (mddev->pers == NULL ||
846 mddev->pers->sync_request == NULL) {
847 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
848 if (mddev->sysfs_action)
849 sysfs_put(mddev->sysfs_action);
850 if (mddev->sysfs_completed)
851 sysfs_put(mddev->sysfs_completed);
852 if (mddev->sysfs_degraded)
853 sysfs_put(mddev->sysfs_degraded);
854 mddev->sysfs_action = NULL;
855 mddev->sysfs_completed = NULL;
856 mddev->sysfs_degraded = NULL;
859 mddev->sysfs_active = 0;
861 mutex_unlock(&mddev->reconfig_mutex);
863 /* As we've dropped the mutex we need a spinlock to
864 * make sure the thread doesn't disappear
866 spin_lock(&pers_lock);
867 md_wakeup_thread(mddev->thread);
868 wake_up(&mddev->sb_wait);
869 spin_unlock(&pers_lock);
871 EXPORT_SYMBOL_GPL(mddev_unlock);
873 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
875 struct md_rdev *rdev;
877 rdev_for_each_rcu(rdev, mddev)
878 if (rdev->desc_nr == nr)
883 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
885 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
887 struct md_rdev *rdev;
889 rdev_for_each(rdev, mddev)
890 if (rdev->bdev->bd_dev == dev)
896 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
898 struct md_rdev *rdev;
900 rdev_for_each_rcu(rdev, mddev)
901 if (rdev->bdev->bd_dev == dev)
906 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
908 static struct md_personality *find_pers(int level, char *clevel)
910 struct md_personality *pers;
911 list_for_each_entry(pers, &pers_list, list) {
912 if (level != LEVEL_NONE && pers->level == level)
914 if (strcmp(pers->name, clevel)==0)
920 /* return the offset of the super block in 512byte sectors */
921 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
923 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
924 return MD_NEW_SIZE_SECTORS(num_sectors);
927 static int alloc_disk_sb(struct md_rdev *rdev)
929 rdev->sb_page = alloc_page(GFP_KERNEL);
935 void md_rdev_clear(struct md_rdev *rdev)
938 put_page(rdev->sb_page);
940 rdev->sb_page = NULL;
945 put_page(rdev->bb_page);
946 rdev->bb_page = NULL;
948 badblocks_exit(&rdev->badblocks);
950 EXPORT_SYMBOL_GPL(md_rdev_clear);
952 static void super_written(struct bio *bio)
954 struct md_rdev *rdev = bio->bi_private;
955 struct mddev *mddev = rdev->mddev;
957 if (bio->bi_status) {
958 pr_err("md: %s gets error=%d\n", __func__,
959 blk_status_to_errno(bio->bi_status));
960 md_error(mddev, rdev);
961 if (!test_bit(Faulty, &rdev->flags)
962 && (bio->bi_opf & MD_FAILFAST)) {
963 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
964 set_bit(LastDev, &rdev->flags);
967 clear_bit(LastDev, &rdev->flags);
969 if (atomic_dec_and_test(&mddev->pending_writes))
970 wake_up(&mddev->sb_wait);
971 rdev_dec_pending(rdev, mddev);
975 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
976 sector_t sector, int size, struct page *page)
978 /* write first size bytes of page to sector of rdev
979 * Increment mddev->pending_writes before returning
980 * and decrement it on completion, waking up sb_wait
981 * if zero is reached.
982 * If an error occurred, call md_error
990 if (test_bit(Faulty, &rdev->flags))
993 bio = md_bio_alloc_sync(mddev);
995 atomic_inc(&rdev->nr_pending);
997 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
998 bio->bi_iter.bi_sector = sector;
999 bio_add_page(bio, page, size, 0);
1000 bio->bi_private = rdev;
1001 bio->bi_end_io = super_written;
1003 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1004 test_bit(FailFast, &rdev->flags) &&
1005 !test_bit(LastDev, &rdev->flags))
1007 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
1009 atomic_inc(&mddev->pending_writes);
1013 int md_super_wait(struct mddev *mddev)
1015 /* wait for all superblock writes that were scheduled to complete */
1016 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1017 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1022 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1023 struct page *page, int op, int op_flags, bool metadata_op)
1025 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1028 if (metadata_op && rdev->meta_bdev)
1029 bio_set_dev(bio, rdev->meta_bdev);
1031 bio_set_dev(bio, rdev->bdev);
1032 bio_set_op_attrs(bio, op, op_flags);
1034 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1035 else if (rdev->mddev->reshape_position != MaxSector &&
1036 (rdev->mddev->reshape_backwards ==
1037 (sector >= rdev->mddev->reshape_position)))
1038 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
1040 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1041 bio_add_page(bio, page, size, 0);
1043 submit_bio_wait(bio);
1045 ret = !bio->bi_status;
1049 EXPORT_SYMBOL_GPL(sync_page_io);
1051 static int read_disk_sb(struct md_rdev *rdev, int size)
1053 char b[BDEVNAME_SIZE];
1055 if (rdev->sb_loaded)
1058 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1060 rdev->sb_loaded = 1;
1064 pr_err("md: disabled device %s, could not read superblock.\n",
1065 bdevname(rdev->bdev,b));
1069 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1071 return sb1->set_uuid0 == sb2->set_uuid0 &&
1072 sb1->set_uuid1 == sb2->set_uuid1 &&
1073 sb1->set_uuid2 == sb2->set_uuid2 &&
1074 sb1->set_uuid3 == sb2->set_uuid3;
1077 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1080 mdp_super_t *tmp1, *tmp2;
1082 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1083 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1085 if (!tmp1 || !tmp2) {
1094 * nr_disks is not constant
1099 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1106 static u32 md_csum_fold(u32 csum)
1108 csum = (csum & 0xffff) + (csum >> 16);
1109 return (csum & 0xffff) + (csum >> 16);
1112 static unsigned int calc_sb_csum(mdp_super_t *sb)
1115 u32 *sb32 = (u32*)sb;
1117 unsigned int disk_csum, csum;
1119 disk_csum = sb->sb_csum;
1122 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1124 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1127 /* This used to use csum_partial, which was wrong for several
1128 * reasons including that different results are returned on
1129 * different architectures. It isn't critical that we get exactly
1130 * the same return value as before (we always csum_fold before
1131 * testing, and that removes any differences). However as we
1132 * know that csum_partial always returned a 16bit value on
1133 * alphas, do a fold to maximise conformity to previous behaviour.
1135 sb->sb_csum = md_csum_fold(disk_csum);
1137 sb->sb_csum = disk_csum;
1143 * Handle superblock details.
1144 * We want to be able to handle multiple superblock formats
1145 * so we have a common interface to them all, and an array of
1146 * different handlers.
1147 * We rely on user-space to write the initial superblock, and support
1148 * reading and updating of superblocks.
1149 * Interface methods are:
1150 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1151 * loads and validates a superblock on dev.
1152 * if refdev != NULL, compare superblocks on both devices
1154 * 0 - dev has a superblock that is compatible with refdev
1155 * 1 - dev has a superblock that is compatible and newer than refdev
1156 * so dev should be used as the refdev in future
1157 * -EINVAL superblock incompatible or invalid
1158 * -othererror e.g. -EIO
1160 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1161 * Verify that dev is acceptable into mddev.
1162 * The first time, mddev->raid_disks will be 0, and data from
1163 * dev should be merged in. Subsequent calls check that dev
1164 * is new enough. Return 0 or -EINVAL
1166 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1167 * Update the superblock for rdev with data in mddev
1168 * This does not write to disc.
1174 struct module *owner;
1175 int (*load_super)(struct md_rdev *rdev,
1176 struct md_rdev *refdev,
1178 int (*validate_super)(struct mddev *mddev,
1179 struct md_rdev *rdev);
1180 void (*sync_super)(struct mddev *mddev,
1181 struct md_rdev *rdev);
1182 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1183 sector_t num_sectors);
1184 int (*allow_new_offset)(struct md_rdev *rdev,
1185 unsigned long long new_offset);
1189 * Check that the given mddev has no bitmap.
1191 * This function is called from the run method of all personalities that do not
1192 * support bitmaps. It prints an error message and returns non-zero if mddev
1193 * has a bitmap. Otherwise, it returns 0.
1196 int md_check_no_bitmap(struct mddev *mddev)
1198 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1200 pr_warn("%s: bitmaps are not supported for %s\n",
1201 mdname(mddev), mddev->pers->name);
1204 EXPORT_SYMBOL(md_check_no_bitmap);
1207 * load_super for 0.90.0
1209 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1211 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1214 bool spare_disk = true;
1217 * Calculate the position of the superblock (512byte sectors),
1218 * it's at the end of the disk.
1220 * It also happens to be a multiple of 4Kb.
1222 rdev->sb_start = calc_dev_sboffset(rdev);
1224 ret = read_disk_sb(rdev, MD_SB_BYTES);
1230 bdevname(rdev->bdev, b);
1231 sb = page_address(rdev->sb_page);
1233 if (sb->md_magic != MD_SB_MAGIC) {
1234 pr_warn("md: invalid raid superblock magic on %s\n", b);
1238 if (sb->major_version != 0 ||
1239 sb->minor_version < 90 ||
1240 sb->minor_version > 91) {
1241 pr_warn("Bad version number %d.%d on %s\n",
1242 sb->major_version, sb->minor_version, b);
1246 if (sb->raid_disks <= 0)
1249 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1250 pr_warn("md: invalid superblock checksum on %s\n", b);
1254 rdev->preferred_minor = sb->md_minor;
1255 rdev->data_offset = 0;
1256 rdev->new_data_offset = 0;
1257 rdev->sb_size = MD_SB_BYTES;
1258 rdev->badblocks.shift = -1;
1260 if (sb->level == LEVEL_MULTIPATH)
1263 rdev->desc_nr = sb->this_disk.number;
1265 /* not spare disk, or LEVEL_MULTIPATH */
1266 if (sb->level == LEVEL_MULTIPATH ||
1267 (rdev->desc_nr >= 0 &&
1268 rdev->desc_nr < MD_SB_DISKS &&
1269 sb->disks[rdev->desc_nr].state &
1270 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1280 mdp_super_t *refsb = page_address(refdev->sb_page);
1281 if (!md_uuid_equal(refsb, sb)) {
1282 pr_warn("md: %s has different UUID to %s\n",
1283 b, bdevname(refdev->bdev,b2));
1286 if (!md_sb_equal(refsb, sb)) {
1287 pr_warn("md: %s has same UUID but different superblock to %s\n",
1288 b, bdevname(refdev->bdev, b2));
1292 ev2 = md_event(refsb);
1294 if (!spare_disk && ev1 > ev2)
1299 rdev->sectors = rdev->sb_start;
1300 /* Limit to 4TB as metadata cannot record more than that.
1301 * (not needed for Linear and RAID0 as metadata doesn't
1304 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1305 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1307 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1308 /* "this cannot possibly happen" ... */
1316 * validate_super for 0.90.0
1318 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1321 mdp_super_t *sb = page_address(rdev->sb_page);
1322 __u64 ev1 = md_event(sb);
1324 rdev->raid_disk = -1;
1325 clear_bit(Faulty, &rdev->flags);
1326 clear_bit(In_sync, &rdev->flags);
1327 clear_bit(Bitmap_sync, &rdev->flags);
1328 clear_bit(WriteMostly, &rdev->flags);
1330 if (mddev->raid_disks == 0) {
1331 mddev->major_version = 0;
1332 mddev->minor_version = sb->minor_version;
1333 mddev->patch_version = sb->patch_version;
1334 mddev->external = 0;
1335 mddev->chunk_sectors = sb->chunk_size >> 9;
1336 mddev->ctime = sb->ctime;
1337 mddev->utime = sb->utime;
1338 mddev->level = sb->level;
1339 mddev->clevel[0] = 0;
1340 mddev->layout = sb->layout;
1341 mddev->raid_disks = sb->raid_disks;
1342 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1343 mddev->events = ev1;
1344 mddev->bitmap_info.offset = 0;
1345 mddev->bitmap_info.space = 0;
1346 /* bitmap can use 60 K after the 4K superblocks */
1347 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1348 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1349 mddev->reshape_backwards = 0;
1351 if (mddev->minor_version >= 91) {
1352 mddev->reshape_position = sb->reshape_position;
1353 mddev->delta_disks = sb->delta_disks;
1354 mddev->new_level = sb->new_level;
1355 mddev->new_layout = sb->new_layout;
1356 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1357 if (mddev->delta_disks < 0)
1358 mddev->reshape_backwards = 1;
1360 mddev->reshape_position = MaxSector;
1361 mddev->delta_disks = 0;
1362 mddev->new_level = mddev->level;
1363 mddev->new_layout = mddev->layout;
1364 mddev->new_chunk_sectors = mddev->chunk_sectors;
1366 if (mddev->level == 0)
1369 if (sb->state & (1<<MD_SB_CLEAN))
1370 mddev->recovery_cp = MaxSector;
1372 if (sb->events_hi == sb->cp_events_hi &&
1373 sb->events_lo == sb->cp_events_lo) {
1374 mddev->recovery_cp = sb->recovery_cp;
1376 mddev->recovery_cp = 0;
1379 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1380 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1381 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1382 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1384 mddev->max_disks = MD_SB_DISKS;
1386 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1387 mddev->bitmap_info.file == NULL) {
1388 mddev->bitmap_info.offset =
1389 mddev->bitmap_info.default_offset;
1390 mddev->bitmap_info.space =
1391 mddev->bitmap_info.default_space;
1394 } else if (mddev->pers == NULL) {
1395 /* Insist on good event counter while assembling, except
1396 * for spares (which don't need an event count) */
1398 if (sb->disks[rdev->desc_nr].state & (
1399 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1400 if (ev1 < mddev->events)
1402 } else if (mddev->bitmap) {
1403 /* if adding to array with a bitmap, then we can accept an
1404 * older device ... but not too old.
1406 if (ev1 < mddev->bitmap->events_cleared)
1408 if (ev1 < mddev->events)
1409 set_bit(Bitmap_sync, &rdev->flags);
1411 if (ev1 < mddev->events)
1412 /* just a hot-add of a new device, leave raid_disk at -1 */
1416 if (mddev->level != LEVEL_MULTIPATH) {
1417 desc = sb->disks + rdev->desc_nr;
1419 if (desc->state & (1<<MD_DISK_FAULTY))
1420 set_bit(Faulty, &rdev->flags);
1421 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1422 desc->raid_disk < mddev->raid_disks */) {
1423 set_bit(In_sync, &rdev->flags);
1424 rdev->raid_disk = desc->raid_disk;
1425 rdev->saved_raid_disk = desc->raid_disk;
1426 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1427 /* active but not in sync implies recovery up to
1428 * reshape position. We don't know exactly where
1429 * that is, so set to zero for now */
1430 if (mddev->minor_version >= 91) {
1431 rdev->recovery_offset = 0;
1432 rdev->raid_disk = desc->raid_disk;
1435 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1436 set_bit(WriteMostly, &rdev->flags);
1437 if (desc->state & (1<<MD_DISK_FAILFAST))
1438 set_bit(FailFast, &rdev->flags);
1439 } else /* MULTIPATH are always insync */
1440 set_bit(In_sync, &rdev->flags);
1445 * sync_super for 0.90.0
1447 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1450 struct md_rdev *rdev2;
1451 int next_spare = mddev->raid_disks;
1453 /* make rdev->sb match mddev data..
1456 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1457 * 3/ any empty disks < next_spare become removed
1459 * disks[0] gets initialised to REMOVED because
1460 * we cannot be sure from other fields if it has
1461 * been initialised or not.
1464 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1466 rdev->sb_size = MD_SB_BYTES;
1468 sb = page_address(rdev->sb_page);
1470 memset(sb, 0, sizeof(*sb));
1472 sb->md_magic = MD_SB_MAGIC;
1473 sb->major_version = mddev->major_version;
1474 sb->patch_version = mddev->patch_version;
1475 sb->gvalid_words = 0; /* ignored */
1476 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1477 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1478 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1479 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1481 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1482 sb->level = mddev->level;
1483 sb->size = mddev->dev_sectors / 2;
1484 sb->raid_disks = mddev->raid_disks;
1485 sb->md_minor = mddev->md_minor;
1486 sb->not_persistent = 0;
1487 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1489 sb->events_hi = (mddev->events>>32);
1490 sb->events_lo = (u32)mddev->events;
1492 if (mddev->reshape_position == MaxSector)
1493 sb->minor_version = 90;
1495 sb->minor_version = 91;
1496 sb->reshape_position = mddev->reshape_position;
1497 sb->new_level = mddev->new_level;
1498 sb->delta_disks = mddev->delta_disks;
1499 sb->new_layout = mddev->new_layout;
1500 sb->new_chunk = mddev->new_chunk_sectors << 9;
1502 mddev->minor_version = sb->minor_version;
1505 sb->recovery_cp = mddev->recovery_cp;
1506 sb->cp_events_hi = (mddev->events>>32);
1507 sb->cp_events_lo = (u32)mddev->events;
1508 if (mddev->recovery_cp == MaxSector)
1509 sb->state = (1<< MD_SB_CLEAN);
1511 sb->recovery_cp = 0;
1513 sb->layout = mddev->layout;
1514 sb->chunk_size = mddev->chunk_sectors << 9;
1516 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1517 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1519 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1520 rdev_for_each(rdev2, mddev) {
1523 int is_active = test_bit(In_sync, &rdev2->flags);
1525 if (rdev2->raid_disk >= 0 &&
1526 sb->minor_version >= 91)
1527 /* we have nowhere to store the recovery_offset,
1528 * but if it is not below the reshape_position,
1529 * we can piggy-back on that.
1532 if (rdev2->raid_disk < 0 ||
1533 test_bit(Faulty, &rdev2->flags))
1536 desc_nr = rdev2->raid_disk;
1538 desc_nr = next_spare++;
1539 rdev2->desc_nr = desc_nr;
1540 d = &sb->disks[rdev2->desc_nr];
1542 d->number = rdev2->desc_nr;
1543 d->major = MAJOR(rdev2->bdev->bd_dev);
1544 d->minor = MINOR(rdev2->bdev->bd_dev);
1546 d->raid_disk = rdev2->raid_disk;
1548 d->raid_disk = rdev2->desc_nr; /* compatibility */
1549 if (test_bit(Faulty, &rdev2->flags))
1550 d->state = (1<<MD_DISK_FAULTY);
1551 else if (is_active) {
1552 d->state = (1<<MD_DISK_ACTIVE);
1553 if (test_bit(In_sync, &rdev2->flags))
1554 d->state |= (1<<MD_DISK_SYNC);
1562 if (test_bit(WriteMostly, &rdev2->flags))
1563 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1564 if (test_bit(FailFast, &rdev2->flags))
1565 d->state |= (1<<MD_DISK_FAILFAST);
1567 /* now set the "removed" and "faulty" bits on any missing devices */
1568 for (i=0 ; i < mddev->raid_disks ; i++) {
1569 mdp_disk_t *d = &sb->disks[i];
1570 if (d->state == 0 && d->number == 0) {
1573 d->state = (1<<MD_DISK_REMOVED);
1574 d->state |= (1<<MD_DISK_FAULTY);
1578 sb->nr_disks = nr_disks;
1579 sb->active_disks = active;
1580 sb->working_disks = working;
1581 sb->failed_disks = failed;
1582 sb->spare_disks = spare;
1584 sb->this_disk = sb->disks[rdev->desc_nr];
1585 sb->sb_csum = calc_sb_csum(sb);
1589 * rdev_size_change for 0.90.0
1591 static unsigned long long
1592 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1594 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1595 return 0; /* component must fit device */
1596 if (rdev->mddev->bitmap_info.offset)
1597 return 0; /* can't move bitmap */
1598 rdev->sb_start = calc_dev_sboffset(rdev);
1599 if (!num_sectors || num_sectors > rdev->sb_start)
1600 num_sectors = rdev->sb_start;
1601 /* Limit to 4TB as metadata cannot record more than that.
1602 * 4TB == 2^32 KB, or 2*2^32 sectors.
1604 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1605 num_sectors = (sector_t)(2ULL << 32) - 2;
1607 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1609 } while (md_super_wait(rdev->mddev) < 0);
1614 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1616 /* non-zero offset changes not possible with v0.90 */
1617 return new_offset == 0;
1621 * version 1 superblock
1624 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1628 unsigned long long newcsum;
1629 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1630 __le32 *isuper = (__le32*)sb;
1632 disk_csum = sb->sb_csum;
1635 for (; size >= 4; size -= 4)
1636 newcsum += le32_to_cpu(*isuper++);
1639 newcsum += le16_to_cpu(*(__le16*) isuper);
1641 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1642 sb->sb_csum = disk_csum;
1643 return cpu_to_le32(csum);
1646 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1648 struct mdp_superblock_1 *sb;
1652 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1654 bool spare_disk = true;
1657 * Calculate the position of the superblock in 512byte sectors.
1658 * It is always aligned to a 4K boundary and
1659 * depeding on minor_version, it can be:
1660 * 0: At least 8K, but less than 12K, from end of device
1661 * 1: At start of device
1662 * 2: 4K from start of device.
1664 switch(minor_version) {
1666 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1668 sb_start &= ~(sector_t)(4*2-1);
1679 rdev->sb_start = sb_start;
1681 /* superblock is rarely larger than 1K, but it can be larger,
1682 * and it is safe to read 4k, so we do that
1684 ret = read_disk_sb(rdev, 4096);
1685 if (ret) return ret;
1687 sb = page_address(rdev->sb_page);
1689 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1690 sb->major_version != cpu_to_le32(1) ||
1691 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1692 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1693 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1696 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1697 pr_warn("md: invalid superblock checksum on %s\n",
1698 bdevname(rdev->bdev,b));
1701 if (le64_to_cpu(sb->data_size) < 10) {
1702 pr_warn("md: data_size too small on %s\n",
1703 bdevname(rdev->bdev,b));
1708 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1709 /* Some padding is non-zero, might be a new feature */
1712 rdev->preferred_minor = 0xffff;
1713 rdev->data_offset = le64_to_cpu(sb->data_offset);
1714 rdev->new_data_offset = rdev->data_offset;
1715 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1716 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1717 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1718 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1720 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1721 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1722 if (rdev->sb_size & bmask)
1723 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1726 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1729 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1732 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1735 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1737 if (!rdev->bb_page) {
1738 rdev->bb_page = alloc_page(GFP_KERNEL);
1742 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1743 rdev->badblocks.count == 0) {
1744 /* need to load the bad block list.
1745 * Currently we limit it to one page.
1751 int sectors = le16_to_cpu(sb->bblog_size);
1752 if (sectors > (PAGE_SIZE / 512))
1754 offset = le32_to_cpu(sb->bblog_offset);
1757 bb_sector = (long long)offset;
1758 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1759 rdev->bb_page, REQ_OP_READ, 0, true))
1761 bbp = (__le64 *)page_address(rdev->bb_page);
1762 rdev->badblocks.shift = sb->bblog_shift;
1763 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1764 u64 bb = le64_to_cpu(*bbp);
1765 int count = bb & (0x3ff);
1766 u64 sector = bb >> 10;
1767 sector <<= sb->bblog_shift;
1768 count <<= sb->bblog_shift;
1771 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1774 } else if (sb->bblog_offset != 0)
1775 rdev->badblocks.shift = 0;
1777 if ((le32_to_cpu(sb->feature_map) &
1778 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1779 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1780 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1781 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1784 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1788 /* not spare disk, or LEVEL_MULTIPATH */
1789 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1790 (rdev->desc_nr >= 0 &&
1791 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1792 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1793 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1803 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1805 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1806 sb->level != refsb->level ||
1807 sb->layout != refsb->layout ||
1808 sb->chunksize != refsb->chunksize) {
1809 pr_warn("md: %s has strangely different superblock to %s\n",
1810 bdevname(rdev->bdev,b),
1811 bdevname(refdev->bdev,b2));
1814 ev1 = le64_to_cpu(sb->events);
1815 ev2 = le64_to_cpu(refsb->events);
1817 if (!spare_disk && ev1 > ev2)
1822 if (minor_version) {
1823 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1824 sectors -= rdev->data_offset;
1826 sectors = rdev->sb_start;
1827 if (sectors < le64_to_cpu(sb->data_size))
1829 rdev->sectors = le64_to_cpu(sb->data_size);
1833 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1835 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1836 __u64 ev1 = le64_to_cpu(sb->events);
1838 rdev->raid_disk = -1;
1839 clear_bit(Faulty, &rdev->flags);
1840 clear_bit(In_sync, &rdev->flags);
1841 clear_bit(Bitmap_sync, &rdev->flags);
1842 clear_bit(WriteMostly, &rdev->flags);
1844 if (mddev->raid_disks == 0) {
1845 mddev->major_version = 1;
1846 mddev->patch_version = 0;
1847 mddev->external = 0;
1848 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1849 mddev->ctime = le64_to_cpu(sb->ctime);
1850 mddev->utime = le64_to_cpu(sb->utime);
1851 mddev->level = le32_to_cpu(sb->level);
1852 mddev->clevel[0] = 0;
1853 mddev->layout = le32_to_cpu(sb->layout);
1854 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1855 mddev->dev_sectors = le64_to_cpu(sb->size);
1856 mddev->events = ev1;
1857 mddev->bitmap_info.offset = 0;
1858 mddev->bitmap_info.space = 0;
1859 /* Default location for bitmap is 1K after superblock
1860 * using 3K - total of 4K
1862 mddev->bitmap_info.default_offset = 1024 >> 9;
1863 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1864 mddev->reshape_backwards = 0;
1866 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1867 memcpy(mddev->uuid, sb->set_uuid, 16);
1869 mddev->max_disks = (4096-256)/2;
1871 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1872 mddev->bitmap_info.file == NULL) {
1873 mddev->bitmap_info.offset =
1874 (__s32)le32_to_cpu(sb->bitmap_offset);
1875 /* Metadata doesn't record how much space is available.
1876 * For 1.0, we assume we can use up to the superblock
1877 * if before, else to 4K beyond superblock.
1878 * For others, assume no change is possible.
1880 if (mddev->minor_version > 0)
1881 mddev->bitmap_info.space = 0;
1882 else if (mddev->bitmap_info.offset > 0)
1883 mddev->bitmap_info.space =
1884 8 - mddev->bitmap_info.offset;
1886 mddev->bitmap_info.space =
1887 -mddev->bitmap_info.offset;
1890 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1891 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1892 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1893 mddev->new_level = le32_to_cpu(sb->new_level);
1894 mddev->new_layout = le32_to_cpu(sb->new_layout);
1895 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1896 if (mddev->delta_disks < 0 ||
1897 (mddev->delta_disks == 0 &&
1898 (le32_to_cpu(sb->feature_map)
1899 & MD_FEATURE_RESHAPE_BACKWARDS)))
1900 mddev->reshape_backwards = 1;
1902 mddev->reshape_position = MaxSector;
1903 mddev->delta_disks = 0;
1904 mddev->new_level = mddev->level;
1905 mddev->new_layout = mddev->layout;
1906 mddev->new_chunk_sectors = mddev->chunk_sectors;
1909 if (mddev->level == 0 &&
1910 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1913 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1914 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1916 if (le32_to_cpu(sb->feature_map) &
1917 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1918 if (le32_to_cpu(sb->feature_map) &
1919 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1921 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1922 (le32_to_cpu(sb->feature_map) &
1923 MD_FEATURE_MULTIPLE_PPLS))
1925 set_bit(MD_HAS_PPL, &mddev->flags);
1927 } else if (mddev->pers == NULL) {
1928 /* Insist of good event counter while assembling, except for
1929 * spares (which don't need an event count) */
1931 if (rdev->desc_nr >= 0 &&
1932 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1933 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1934 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1935 if (ev1 < mddev->events)
1937 } else if (mddev->bitmap) {
1938 /* If adding to array with a bitmap, then we can accept an
1939 * older device, but not too old.
1941 if (ev1 < mddev->bitmap->events_cleared)
1943 if (ev1 < mddev->events)
1944 set_bit(Bitmap_sync, &rdev->flags);
1946 if (ev1 < mddev->events)
1947 /* just a hot-add of a new device, leave raid_disk at -1 */
1950 if (mddev->level != LEVEL_MULTIPATH) {
1952 if (rdev->desc_nr < 0 ||
1953 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1954 role = MD_DISK_ROLE_SPARE;
1957 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1959 case MD_DISK_ROLE_SPARE: /* spare */
1961 case MD_DISK_ROLE_FAULTY: /* faulty */
1962 set_bit(Faulty, &rdev->flags);
1964 case MD_DISK_ROLE_JOURNAL: /* journal device */
1965 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1966 /* journal device without journal feature */
1967 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1970 set_bit(Journal, &rdev->flags);
1971 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1972 rdev->raid_disk = 0;
1975 rdev->saved_raid_disk = role;
1976 if ((le32_to_cpu(sb->feature_map) &
1977 MD_FEATURE_RECOVERY_OFFSET)) {
1978 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1979 if (!(le32_to_cpu(sb->feature_map) &
1980 MD_FEATURE_RECOVERY_BITMAP))
1981 rdev->saved_raid_disk = -1;
1984 * If the array is FROZEN, then the device can't
1985 * be in_sync with rest of array.
1987 if (!test_bit(MD_RECOVERY_FROZEN,
1989 set_bit(In_sync, &rdev->flags);
1991 rdev->raid_disk = role;
1994 if (sb->devflags & WriteMostly1)
1995 set_bit(WriteMostly, &rdev->flags);
1996 if (sb->devflags & FailFast1)
1997 set_bit(FailFast, &rdev->flags);
1998 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1999 set_bit(Replacement, &rdev->flags);
2000 } else /* MULTIPATH are always insync */
2001 set_bit(In_sync, &rdev->flags);
2006 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2008 struct mdp_superblock_1 *sb;
2009 struct md_rdev *rdev2;
2011 /* make rdev->sb match mddev and rdev data. */
2013 sb = page_address(rdev->sb_page);
2015 sb->feature_map = 0;
2017 sb->recovery_offset = cpu_to_le64(0);
2018 memset(sb->pad3, 0, sizeof(sb->pad3));
2020 sb->utime = cpu_to_le64((__u64)mddev->utime);
2021 sb->events = cpu_to_le64(mddev->events);
2023 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2024 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2025 sb->resync_offset = cpu_to_le64(MaxSector);
2027 sb->resync_offset = cpu_to_le64(0);
2029 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2031 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2032 sb->size = cpu_to_le64(mddev->dev_sectors);
2033 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2034 sb->level = cpu_to_le32(mddev->level);
2035 sb->layout = cpu_to_le32(mddev->layout);
2036 if (test_bit(FailFast, &rdev->flags))
2037 sb->devflags |= FailFast1;
2039 sb->devflags &= ~FailFast1;
2041 if (test_bit(WriteMostly, &rdev->flags))
2042 sb->devflags |= WriteMostly1;
2044 sb->devflags &= ~WriteMostly1;
2045 sb->data_offset = cpu_to_le64(rdev->data_offset);
2046 sb->data_size = cpu_to_le64(rdev->sectors);
2048 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2049 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2050 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2053 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2054 !test_bit(In_sync, &rdev->flags)) {
2056 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2057 sb->recovery_offset =
2058 cpu_to_le64(rdev->recovery_offset);
2059 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2061 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2063 /* Note: recovery_offset and journal_tail share space */
2064 if (test_bit(Journal, &rdev->flags))
2065 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2066 if (test_bit(Replacement, &rdev->flags))
2068 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2070 if (mddev->reshape_position != MaxSector) {
2071 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2072 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2073 sb->new_layout = cpu_to_le32(mddev->new_layout);
2074 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2075 sb->new_level = cpu_to_le32(mddev->new_level);
2076 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2077 if (mddev->delta_disks == 0 &&
2078 mddev->reshape_backwards)
2080 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2081 if (rdev->new_data_offset != rdev->data_offset) {
2083 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2084 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2085 - rdev->data_offset));
2089 if (mddev_is_clustered(mddev))
2090 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2092 if (rdev->badblocks.count == 0)
2093 /* Nothing to do for bad blocks*/ ;
2094 else if (sb->bblog_offset == 0)
2095 /* Cannot record bad blocks on this device */
2096 md_error(mddev, rdev);
2098 struct badblocks *bb = &rdev->badblocks;
2099 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2101 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2106 seq = read_seqbegin(&bb->lock);
2108 memset(bbp, 0xff, PAGE_SIZE);
2110 for (i = 0 ; i < bb->count ; i++) {
2111 u64 internal_bb = p[i];
2112 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2113 | BB_LEN(internal_bb));
2114 bbp[i] = cpu_to_le64(store_bb);
2117 if (read_seqretry(&bb->lock, seq))
2120 bb->sector = (rdev->sb_start +
2121 (int)le32_to_cpu(sb->bblog_offset));
2122 bb->size = le16_to_cpu(sb->bblog_size);
2127 rdev_for_each(rdev2, mddev)
2128 if (rdev2->desc_nr+1 > max_dev)
2129 max_dev = rdev2->desc_nr+1;
2131 if (max_dev > le32_to_cpu(sb->max_dev)) {
2133 sb->max_dev = cpu_to_le32(max_dev);
2134 rdev->sb_size = max_dev * 2 + 256;
2135 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2136 if (rdev->sb_size & bmask)
2137 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2139 max_dev = le32_to_cpu(sb->max_dev);
2141 for (i=0; i<max_dev;i++)
2142 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2144 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2145 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2147 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2148 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2150 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2152 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2153 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2154 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2157 rdev_for_each(rdev2, mddev) {
2159 if (test_bit(Faulty, &rdev2->flags))
2160 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2161 else if (test_bit(In_sync, &rdev2->flags))
2162 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2163 else if (test_bit(Journal, &rdev2->flags))
2164 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2165 else if (rdev2->raid_disk >= 0)
2166 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2168 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2171 sb->sb_csum = calc_sb_1_csum(sb);
2174 static sector_t super_1_choose_bm_space(sector_t dev_size)
2178 /* if the device is bigger than 8Gig, save 64k for bitmap
2179 * usage, if bigger than 200Gig, save 128k
2181 if (dev_size < 64*2)
2183 else if (dev_size - 64*2 >= 200*1024*1024*2)
2185 else if (dev_size - 4*2 > 8*1024*1024*2)
2192 static unsigned long long
2193 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2195 struct mdp_superblock_1 *sb;
2196 sector_t max_sectors;
2197 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2198 return 0; /* component must fit device */
2199 if (rdev->data_offset != rdev->new_data_offset)
2200 return 0; /* too confusing */
2201 if (rdev->sb_start < rdev->data_offset) {
2202 /* minor versions 1 and 2; superblock before data */
2203 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2204 max_sectors -= rdev->data_offset;
2205 if (!num_sectors || num_sectors > max_sectors)
2206 num_sectors = max_sectors;
2207 } else if (rdev->mddev->bitmap_info.offset) {
2208 /* minor version 0 with bitmap we can't move */
2211 /* minor version 0; superblock after data */
2212 sector_t sb_start, bm_space;
2213 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2215 /* 8K is for superblock */
2216 sb_start = dev_size - 8*2;
2217 sb_start &= ~(sector_t)(4*2 - 1);
2219 bm_space = super_1_choose_bm_space(dev_size);
2221 /* Space that can be used to store date needs to decrease
2222 * superblock bitmap space and bad block space(4K)
2224 max_sectors = sb_start - bm_space - 4*2;
2226 if (!num_sectors || num_sectors > max_sectors)
2227 num_sectors = max_sectors;
2228 rdev->sb_start = sb_start;
2230 sb = page_address(rdev->sb_page);
2231 sb->data_size = cpu_to_le64(num_sectors);
2232 sb->super_offset = cpu_to_le64(rdev->sb_start);
2233 sb->sb_csum = calc_sb_1_csum(sb);
2235 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2237 } while (md_super_wait(rdev->mddev) < 0);
2243 super_1_allow_new_offset(struct md_rdev *rdev,
2244 unsigned long long new_offset)
2246 /* All necessary checks on new >= old have been done */
2247 struct bitmap *bitmap;
2248 if (new_offset >= rdev->data_offset)
2251 /* with 1.0 metadata, there is no metadata to tread on
2252 * so we can always move back */
2253 if (rdev->mddev->minor_version == 0)
2256 /* otherwise we must be sure not to step on
2257 * any metadata, so stay:
2258 * 36K beyond start of superblock
2259 * beyond end of badblocks
2260 * beyond write-intent bitmap
2262 if (rdev->sb_start + (32+4)*2 > new_offset)
2264 bitmap = rdev->mddev->bitmap;
2265 if (bitmap && !rdev->mddev->bitmap_info.file &&
2266 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2267 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2269 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2275 static struct super_type super_types[] = {
2278 .owner = THIS_MODULE,
2279 .load_super = super_90_load,
2280 .validate_super = super_90_validate,
2281 .sync_super = super_90_sync,
2282 .rdev_size_change = super_90_rdev_size_change,
2283 .allow_new_offset = super_90_allow_new_offset,
2287 .owner = THIS_MODULE,
2288 .load_super = super_1_load,
2289 .validate_super = super_1_validate,
2290 .sync_super = super_1_sync,
2291 .rdev_size_change = super_1_rdev_size_change,
2292 .allow_new_offset = super_1_allow_new_offset,
2296 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2298 if (mddev->sync_super) {
2299 mddev->sync_super(mddev, rdev);
2303 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2305 super_types[mddev->major_version].sync_super(mddev, rdev);
2308 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2310 struct md_rdev *rdev, *rdev2;
2313 rdev_for_each_rcu(rdev, mddev1) {
2314 if (test_bit(Faulty, &rdev->flags) ||
2315 test_bit(Journal, &rdev->flags) ||
2316 rdev->raid_disk == -1)
2318 rdev_for_each_rcu(rdev2, mddev2) {
2319 if (test_bit(Faulty, &rdev2->flags) ||
2320 test_bit(Journal, &rdev2->flags) ||
2321 rdev2->raid_disk == -1)
2323 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2333 static LIST_HEAD(pending_raid_disks);
2336 * Try to register data integrity profile for an mddev
2338 * This is called when an array is started and after a disk has been kicked
2339 * from the array. It only succeeds if all working and active component devices
2340 * are integrity capable with matching profiles.
2342 int md_integrity_register(struct mddev *mddev)
2344 struct md_rdev *rdev, *reference = NULL;
2346 if (list_empty(&mddev->disks))
2347 return 0; /* nothing to do */
2348 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2349 return 0; /* shouldn't register, or already is */
2350 rdev_for_each(rdev, mddev) {
2351 /* skip spares and non-functional disks */
2352 if (test_bit(Faulty, &rdev->flags))
2354 if (rdev->raid_disk < 0)
2357 /* Use the first rdev as the reference */
2361 /* does this rdev's profile match the reference profile? */
2362 if (blk_integrity_compare(reference->bdev->bd_disk,
2363 rdev->bdev->bd_disk) < 0)
2366 if (!reference || !bdev_get_integrity(reference->bdev))
2369 * All component devices are integrity capable and have matching
2370 * profiles, register the common profile for the md device.
2372 blk_integrity_register(mddev->gendisk,
2373 bdev_get_integrity(reference->bdev));
2375 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2376 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2377 pr_err("md: failed to create integrity pool for %s\n",
2383 EXPORT_SYMBOL(md_integrity_register);
2386 * Attempt to add an rdev, but only if it is consistent with the current
2389 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2391 struct blk_integrity *bi_mddev;
2392 char name[BDEVNAME_SIZE];
2394 if (!mddev->gendisk)
2397 bi_mddev = blk_get_integrity(mddev->gendisk);
2399 if (!bi_mddev) /* nothing to do */
2402 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2403 pr_err("%s: incompatible integrity profile for %s\n",
2404 mdname(mddev), bdevname(rdev->bdev, name));
2410 EXPORT_SYMBOL(md_integrity_add_rdev);
2412 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2414 char b[BDEVNAME_SIZE];
2418 /* prevent duplicates */
2419 if (find_rdev(mddev, rdev->bdev->bd_dev))
2422 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2426 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2427 if (!test_bit(Journal, &rdev->flags) &&
2429 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2431 /* Cannot change size, so fail
2432 * If mddev->level <= 0, then we don't care
2433 * about aligning sizes (e.g. linear)
2435 if (mddev->level > 0)
2438 mddev->dev_sectors = rdev->sectors;
2441 /* Verify rdev->desc_nr is unique.
2442 * If it is -1, assign a free number, else
2443 * check number is not in use
2446 if (rdev->desc_nr < 0) {
2449 choice = mddev->raid_disks;
2450 while (md_find_rdev_nr_rcu(mddev, choice))
2452 rdev->desc_nr = choice;
2454 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2460 if (!test_bit(Journal, &rdev->flags) &&
2461 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2462 pr_warn("md: %s: array is limited to %d devices\n",
2463 mdname(mddev), mddev->max_disks);
2466 bdevname(rdev->bdev,b);
2467 strreplace(b, '/', '!');
2469 rdev->mddev = mddev;
2470 pr_debug("md: bind<%s>\n", b);
2472 if (mddev->raid_disks)
2473 mddev_create_serial_pool(mddev, rdev, false);
2475 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2478 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2479 /* failure here is OK */
2480 err = sysfs_create_link(&rdev->kobj, ko, "block");
2481 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2482 rdev->sysfs_unack_badblocks =
2483 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2484 rdev->sysfs_badblocks =
2485 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2487 list_add_rcu(&rdev->same_set, &mddev->disks);
2488 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2490 /* May as well allow recovery to be retried once */
2491 mddev->recovery_disabled++;
2496 pr_warn("md: failed to register dev-%s for %s\n",
2501 static void rdev_delayed_delete(struct work_struct *ws)
2503 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2504 kobject_del(&rdev->kobj);
2505 kobject_put(&rdev->kobj);
2508 static void unbind_rdev_from_array(struct md_rdev *rdev)
2510 char b[BDEVNAME_SIZE];
2512 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2513 list_del_rcu(&rdev->same_set);
2514 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2515 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2517 sysfs_remove_link(&rdev->kobj, "block");
2518 sysfs_put(rdev->sysfs_state);
2519 sysfs_put(rdev->sysfs_unack_badblocks);
2520 sysfs_put(rdev->sysfs_badblocks);
2521 rdev->sysfs_state = NULL;
2522 rdev->sysfs_unack_badblocks = NULL;
2523 rdev->sysfs_badblocks = NULL;
2524 rdev->badblocks.count = 0;
2525 /* We need to delay this, otherwise we can deadlock when
2526 * writing to 'remove' to "dev/state". We also need
2527 * to delay it due to rcu usage.
2530 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2531 kobject_get(&rdev->kobj);
2532 queue_work(md_rdev_misc_wq, &rdev->del_work);
2536 * prevent the device from being mounted, repartitioned or
2537 * otherwise reused by a RAID array (or any other kernel
2538 * subsystem), by bd_claiming the device.
2540 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2543 struct block_device *bdev;
2545 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2546 shared ? (struct md_rdev *)lock_rdev : rdev);
2548 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2549 MAJOR(dev), MINOR(dev));
2550 return PTR_ERR(bdev);
2556 static void unlock_rdev(struct md_rdev *rdev)
2558 struct block_device *bdev = rdev->bdev;
2560 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2563 void md_autodetect_dev(dev_t dev);
2565 static void export_rdev(struct md_rdev *rdev)
2567 char b[BDEVNAME_SIZE];
2569 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2570 md_rdev_clear(rdev);
2572 if (test_bit(AutoDetected, &rdev->flags))
2573 md_autodetect_dev(rdev->bdev->bd_dev);
2576 kobject_put(&rdev->kobj);
2579 void md_kick_rdev_from_array(struct md_rdev *rdev)
2581 unbind_rdev_from_array(rdev);
2584 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2586 static void export_array(struct mddev *mddev)
2588 struct md_rdev *rdev;
2590 while (!list_empty(&mddev->disks)) {
2591 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2593 md_kick_rdev_from_array(rdev);
2595 mddev->raid_disks = 0;
2596 mddev->major_version = 0;
2599 static bool set_in_sync(struct mddev *mddev)
2601 lockdep_assert_held(&mddev->lock);
2602 if (!mddev->in_sync) {
2603 mddev->sync_checkers++;
2604 spin_unlock(&mddev->lock);
2605 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2606 spin_lock(&mddev->lock);
2607 if (!mddev->in_sync &&
2608 percpu_ref_is_zero(&mddev->writes_pending)) {
2611 * Ensure ->in_sync is visible before we clear
2615 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2616 sysfs_notify_dirent_safe(mddev->sysfs_state);
2618 if (--mddev->sync_checkers == 0)
2619 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2621 if (mddev->safemode == 1)
2622 mddev->safemode = 0;
2623 return mddev->in_sync;
2626 static void sync_sbs(struct mddev *mddev, int nospares)
2628 /* Update each superblock (in-memory image), but
2629 * if we are allowed to, skip spares which already
2630 * have the right event counter, or have one earlier
2631 * (which would mean they aren't being marked as dirty
2632 * with the rest of the array)
2634 struct md_rdev *rdev;
2635 rdev_for_each(rdev, mddev) {
2636 if (rdev->sb_events == mddev->events ||
2638 rdev->raid_disk < 0 &&
2639 rdev->sb_events+1 == mddev->events)) {
2640 /* Don't update this superblock */
2641 rdev->sb_loaded = 2;
2643 sync_super(mddev, rdev);
2644 rdev->sb_loaded = 1;
2649 static bool does_sb_need_changing(struct mddev *mddev)
2651 struct md_rdev *rdev = NULL, *iter;
2652 struct mdp_superblock_1 *sb;
2655 /* Find a good rdev */
2656 rdev_for_each(iter, mddev)
2657 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2662 /* No good device found. */
2666 sb = page_address(rdev->sb_page);
2667 /* Check if a device has become faulty or a spare become active */
2668 rdev_for_each(rdev, mddev) {
2669 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2670 /* Device activated? */
2671 if (role == 0xffff && rdev->raid_disk >=0 &&
2672 !test_bit(Faulty, &rdev->flags))
2674 /* Device turned faulty? */
2675 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2679 /* Check if any mddev parameters have changed */
2680 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2681 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2682 (mddev->layout != le32_to_cpu(sb->layout)) ||
2683 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2684 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2690 void md_update_sb(struct mddev *mddev, int force_change)
2692 struct md_rdev *rdev;
2695 int any_badblocks_changed = 0;
2700 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2705 if (mddev_is_clustered(mddev)) {
2706 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2708 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2710 ret = md_cluster_ops->metadata_update_start(mddev);
2711 /* Has someone else has updated the sb */
2712 if (!does_sb_need_changing(mddev)) {
2714 md_cluster_ops->metadata_update_cancel(mddev);
2715 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2716 BIT(MD_SB_CHANGE_DEVS) |
2717 BIT(MD_SB_CHANGE_CLEAN));
2723 * First make sure individual recovery_offsets are correct
2724 * curr_resync_completed can only be used during recovery.
2725 * During reshape/resync it might use array-addresses rather
2726 * that device addresses.
2728 rdev_for_each(rdev, mddev) {
2729 if (rdev->raid_disk >= 0 &&
2730 mddev->delta_disks >= 0 &&
2731 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2732 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2733 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2734 !test_bit(Journal, &rdev->flags) &&
2735 !test_bit(In_sync, &rdev->flags) &&
2736 mddev->curr_resync_completed > rdev->recovery_offset)
2737 rdev->recovery_offset = mddev->curr_resync_completed;
2740 if (!mddev->persistent) {
2741 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2742 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2743 if (!mddev->external) {
2744 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2745 rdev_for_each(rdev, mddev) {
2746 if (rdev->badblocks.changed) {
2747 rdev->badblocks.changed = 0;
2748 ack_all_badblocks(&rdev->badblocks);
2749 md_error(mddev, rdev);
2751 clear_bit(Blocked, &rdev->flags);
2752 clear_bit(BlockedBadBlocks, &rdev->flags);
2753 wake_up(&rdev->blocked_wait);
2756 wake_up(&mddev->sb_wait);
2760 spin_lock(&mddev->lock);
2762 mddev->utime = ktime_get_real_seconds();
2764 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2766 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2767 /* just a clean<-> dirty transition, possibly leave spares alone,
2768 * though if events isn't the right even/odd, we will have to do
2774 if (mddev->degraded)
2775 /* If the array is degraded, then skipping spares is both
2776 * dangerous and fairly pointless.
2777 * Dangerous because a device that was removed from the array
2778 * might have a event_count that still looks up-to-date,
2779 * so it can be re-added without a resync.
2780 * Pointless because if there are any spares to skip,
2781 * then a recovery will happen and soon that array won't
2782 * be degraded any more and the spare can go back to sleep then.
2786 sync_req = mddev->in_sync;
2788 /* If this is just a dirty<->clean transition, and the array is clean
2789 * and 'events' is odd, we can roll back to the previous clean state */
2791 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2792 && mddev->can_decrease_events
2793 && mddev->events != 1) {
2795 mddev->can_decrease_events = 0;
2797 /* otherwise we have to go forward and ... */
2799 mddev->can_decrease_events = nospares;
2803 * This 64-bit counter should never wrap.
2804 * Either we are in around ~1 trillion A.C., assuming
2805 * 1 reboot per second, or we have a bug...
2807 WARN_ON(mddev->events == 0);
2809 rdev_for_each(rdev, mddev) {
2810 if (rdev->badblocks.changed)
2811 any_badblocks_changed++;
2812 if (test_bit(Faulty, &rdev->flags))
2813 set_bit(FaultRecorded, &rdev->flags);
2816 sync_sbs(mddev, nospares);
2817 spin_unlock(&mddev->lock);
2819 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2820 mdname(mddev), mddev->in_sync);
2823 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2825 md_bitmap_update_sb(mddev->bitmap);
2826 rdev_for_each(rdev, mddev) {
2827 char b[BDEVNAME_SIZE];
2829 if (rdev->sb_loaded != 1)
2830 continue; /* no noise on spare devices */
2832 if (!test_bit(Faulty, &rdev->flags)) {
2833 md_super_write(mddev,rdev,
2834 rdev->sb_start, rdev->sb_size,
2836 pr_debug("md: (write) %s's sb offset: %llu\n",
2837 bdevname(rdev->bdev, b),
2838 (unsigned long long)rdev->sb_start);
2839 rdev->sb_events = mddev->events;
2840 if (rdev->badblocks.size) {
2841 md_super_write(mddev, rdev,
2842 rdev->badblocks.sector,
2843 rdev->badblocks.size << 9,
2845 rdev->badblocks.size = 0;
2849 pr_debug("md: %s (skipping faulty)\n",
2850 bdevname(rdev->bdev, b));
2852 if (mddev->level == LEVEL_MULTIPATH)
2853 /* only need to write one superblock... */
2856 if (md_super_wait(mddev) < 0)
2858 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2860 if (mddev_is_clustered(mddev) && ret == 0)
2861 md_cluster_ops->metadata_update_finish(mddev);
2863 if (mddev->in_sync != sync_req ||
2864 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2865 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2866 /* have to write it out again */
2868 wake_up(&mddev->sb_wait);
2869 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2870 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2872 rdev_for_each(rdev, mddev) {
2873 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2874 clear_bit(Blocked, &rdev->flags);
2876 if (any_badblocks_changed)
2877 ack_all_badblocks(&rdev->badblocks);
2878 clear_bit(BlockedBadBlocks, &rdev->flags);
2879 wake_up(&rdev->blocked_wait);
2882 EXPORT_SYMBOL(md_update_sb);
2884 static int add_bound_rdev(struct md_rdev *rdev)
2886 struct mddev *mddev = rdev->mddev;
2888 bool add_journal = test_bit(Journal, &rdev->flags);
2890 if (!mddev->pers->hot_remove_disk || add_journal) {
2891 /* If there is hot_add_disk but no hot_remove_disk
2892 * then added disks for geometry changes,
2893 * and should be added immediately.
2895 super_types[mddev->major_version].
2896 validate_super(mddev, rdev);
2898 mddev_suspend(mddev);
2899 err = mddev->pers->hot_add_disk(mddev, rdev);
2901 mddev_resume(mddev);
2903 md_kick_rdev_from_array(rdev);
2907 sysfs_notify_dirent_safe(rdev->sysfs_state);
2909 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2910 if (mddev->degraded)
2911 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2912 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2913 md_new_event(mddev);
2914 md_wakeup_thread(mddev->thread);
2918 /* words written to sysfs files may, or may not, be \n terminated.
2919 * We want to accept with case. For this we use cmd_match.
2921 static int cmd_match(const char *cmd, const char *str)
2923 /* See if cmd, written into a sysfs file, matches
2924 * str. They must either be the same, or cmd can
2925 * have a trailing newline
2927 while (*cmd && *str && *cmd == *str) {
2938 struct rdev_sysfs_entry {
2939 struct attribute attr;
2940 ssize_t (*show)(struct md_rdev *, char *);
2941 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2945 state_show(struct md_rdev *rdev, char *page)
2949 unsigned long flags = READ_ONCE(rdev->flags);
2951 if (test_bit(Faulty, &flags) ||
2952 (!test_bit(ExternalBbl, &flags) &&
2953 rdev->badblocks.unacked_exist))
2954 len += sprintf(page+len, "faulty%s", sep);
2955 if (test_bit(In_sync, &flags))
2956 len += sprintf(page+len, "in_sync%s", sep);
2957 if (test_bit(Journal, &flags))
2958 len += sprintf(page+len, "journal%s", sep);
2959 if (test_bit(WriteMostly, &flags))
2960 len += sprintf(page+len, "write_mostly%s", sep);
2961 if (test_bit(Blocked, &flags) ||
2962 (rdev->badblocks.unacked_exist
2963 && !test_bit(Faulty, &flags)))
2964 len += sprintf(page+len, "blocked%s", sep);
2965 if (!test_bit(Faulty, &flags) &&
2966 !test_bit(Journal, &flags) &&
2967 !test_bit(In_sync, &flags))
2968 len += sprintf(page+len, "spare%s", sep);
2969 if (test_bit(WriteErrorSeen, &flags))
2970 len += sprintf(page+len, "write_error%s", sep);
2971 if (test_bit(WantReplacement, &flags))
2972 len += sprintf(page+len, "want_replacement%s", sep);
2973 if (test_bit(Replacement, &flags))
2974 len += sprintf(page+len, "replacement%s", sep);
2975 if (test_bit(ExternalBbl, &flags))
2976 len += sprintf(page+len, "external_bbl%s", sep);
2977 if (test_bit(FailFast, &flags))
2978 len += sprintf(page+len, "failfast%s", sep);
2983 return len+sprintf(page+len, "\n");
2987 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2990 * faulty - simulates an error
2991 * remove - disconnects the device
2992 * writemostly - sets write_mostly
2993 * -writemostly - clears write_mostly
2994 * blocked - sets the Blocked flags
2995 * -blocked - clears the Blocked and possibly simulates an error
2996 * insync - sets Insync providing device isn't active
2997 * -insync - clear Insync for a device with a slot assigned,
2998 * so that it gets rebuilt based on bitmap
2999 * write_error - sets WriteErrorSeen
3000 * -write_error - clears WriteErrorSeen
3001 * {,-}failfast - set/clear FailFast
3004 struct mddev *mddev = rdev->mddev;
3006 bool need_update_sb = false;
3008 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
3009 md_error(rdev->mddev, rdev);
3010 if (test_bit(Faulty, &rdev->flags))
3014 } else if (cmd_match(buf, "remove")) {
3015 if (rdev->mddev->pers) {
3016 clear_bit(Blocked, &rdev->flags);
3017 remove_and_add_spares(rdev->mddev, rdev);
3019 if (rdev->raid_disk >= 0)
3023 if (mddev_is_clustered(mddev))
3024 err = md_cluster_ops->remove_disk(mddev, rdev);
3027 md_kick_rdev_from_array(rdev);
3029 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3030 md_wakeup_thread(mddev->thread);
3032 md_new_event(mddev);
3035 } else if (cmd_match(buf, "writemostly")) {
3036 set_bit(WriteMostly, &rdev->flags);
3037 mddev_create_serial_pool(rdev->mddev, rdev, false);
3038 need_update_sb = true;
3040 } else if (cmd_match(buf, "-writemostly")) {
3041 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3042 clear_bit(WriteMostly, &rdev->flags);
3043 need_update_sb = true;
3045 } else if (cmd_match(buf, "blocked")) {
3046 set_bit(Blocked, &rdev->flags);
3048 } else if (cmd_match(buf, "-blocked")) {
3049 if (!test_bit(Faulty, &rdev->flags) &&
3050 !test_bit(ExternalBbl, &rdev->flags) &&
3051 rdev->badblocks.unacked_exist) {
3052 /* metadata handler doesn't understand badblocks,
3053 * so we need to fail the device
3055 md_error(rdev->mddev, rdev);
3057 clear_bit(Blocked, &rdev->flags);
3058 clear_bit(BlockedBadBlocks, &rdev->flags);
3059 wake_up(&rdev->blocked_wait);
3060 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3061 md_wakeup_thread(rdev->mddev->thread);
3064 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3065 set_bit(In_sync, &rdev->flags);
3067 } else if (cmd_match(buf, "failfast")) {
3068 set_bit(FailFast, &rdev->flags);
3069 need_update_sb = true;
3071 } else if (cmd_match(buf, "-failfast")) {
3072 clear_bit(FailFast, &rdev->flags);
3073 need_update_sb = true;
3075 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3076 !test_bit(Journal, &rdev->flags)) {
3077 if (rdev->mddev->pers == NULL) {
3078 clear_bit(In_sync, &rdev->flags);
3079 rdev->saved_raid_disk = rdev->raid_disk;
3080 rdev->raid_disk = -1;
3083 } else if (cmd_match(buf, "write_error")) {
3084 set_bit(WriteErrorSeen, &rdev->flags);
3086 } else if (cmd_match(buf, "-write_error")) {
3087 clear_bit(WriteErrorSeen, &rdev->flags);
3089 } else if (cmd_match(buf, "want_replacement")) {
3090 /* Any non-spare device that is not a replacement can
3091 * become want_replacement at any time, but we then need to
3092 * check if recovery is needed.
3094 if (rdev->raid_disk >= 0 &&
3095 !test_bit(Journal, &rdev->flags) &&
3096 !test_bit(Replacement, &rdev->flags))
3097 set_bit(WantReplacement, &rdev->flags);
3098 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3099 md_wakeup_thread(rdev->mddev->thread);
3101 } else if (cmd_match(buf, "-want_replacement")) {
3102 /* Clearing 'want_replacement' is always allowed.
3103 * Once replacements starts it is too late though.
3106 clear_bit(WantReplacement, &rdev->flags);
3107 } else if (cmd_match(buf, "replacement")) {
3108 /* Can only set a device as a replacement when array has not
3109 * yet been started. Once running, replacement is automatic
3110 * from spares, or by assigning 'slot'.
3112 if (rdev->mddev->pers)
3115 set_bit(Replacement, &rdev->flags);
3118 } else if (cmd_match(buf, "-replacement")) {
3119 /* Similarly, can only clear Replacement before start */
3120 if (rdev->mddev->pers)
3123 clear_bit(Replacement, &rdev->flags);
3126 } else if (cmd_match(buf, "re-add")) {
3127 if (!rdev->mddev->pers)
3129 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3130 rdev->saved_raid_disk >= 0) {
3131 /* clear_bit is performed _after_ all the devices
3132 * have their local Faulty bit cleared. If any writes
3133 * happen in the meantime in the local node, they
3134 * will land in the local bitmap, which will be synced
3135 * by this node eventually
3137 if (!mddev_is_clustered(rdev->mddev) ||
3138 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3139 clear_bit(Faulty, &rdev->flags);
3140 err = add_bound_rdev(rdev);
3144 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3145 set_bit(ExternalBbl, &rdev->flags);
3146 rdev->badblocks.shift = 0;
3148 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3149 clear_bit(ExternalBbl, &rdev->flags);
3153 md_update_sb(mddev, 1);
3155 sysfs_notify_dirent_safe(rdev->sysfs_state);
3156 return err ? err : len;
3158 static struct rdev_sysfs_entry rdev_state =
3159 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3162 errors_show(struct md_rdev *rdev, char *page)
3164 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3168 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3173 rv = kstrtouint(buf, 10, &n);
3176 atomic_set(&rdev->corrected_errors, n);
3179 static struct rdev_sysfs_entry rdev_errors =
3180 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3183 slot_show(struct md_rdev *rdev, char *page)
3185 if (test_bit(Journal, &rdev->flags))
3186 return sprintf(page, "journal\n");
3187 else if (rdev->raid_disk < 0)
3188 return sprintf(page, "none\n");
3190 return sprintf(page, "%d\n", rdev->raid_disk);
3194 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3199 if (test_bit(Journal, &rdev->flags))
3201 if (strncmp(buf, "none", 4)==0)
3204 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3208 if (rdev->mddev->pers && slot == -1) {
3209 /* Setting 'slot' on an active array requires also
3210 * updating the 'rd%d' link, and communicating
3211 * with the personality with ->hot_*_disk.
3212 * For now we only support removing
3213 * failed/spare devices. This normally happens automatically,
3214 * but not when the metadata is externally managed.
3216 if (rdev->raid_disk == -1)
3218 /* personality does all needed checks */
3219 if (rdev->mddev->pers->hot_remove_disk == NULL)
3221 clear_bit(Blocked, &rdev->flags);
3222 remove_and_add_spares(rdev->mddev, rdev);
3223 if (rdev->raid_disk >= 0)
3225 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3226 md_wakeup_thread(rdev->mddev->thread);
3227 } else if (rdev->mddev->pers) {
3228 /* Activating a spare .. or possibly reactivating
3229 * if we ever get bitmaps working here.
3233 if (rdev->raid_disk != -1)
3236 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3239 if (rdev->mddev->pers->hot_add_disk == NULL)
3242 if (slot >= rdev->mddev->raid_disks &&
3243 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3246 rdev->raid_disk = slot;
3247 if (test_bit(In_sync, &rdev->flags))
3248 rdev->saved_raid_disk = slot;
3250 rdev->saved_raid_disk = -1;
3251 clear_bit(In_sync, &rdev->flags);
3252 clear_bit(Bitmap_sync, &rdev->flags);
3253 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3255 rdev->raid_disk = -1;
3258 sysfs_notify_dirent_safe(rdev->sysfs_state);
3259 /* failure here is OK */;
3260 sysfs_link_rdev(rdev->mddev, rdev);
3261 /* don't wakeup anyone, leave that to userspace. */
3263 if (slot >= rdev->mddev->raid_disks &&
3264 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3266 rdev->raid_disk = slot;
3267 /* assume it is working */
3268 clear_bit(Faulty, &rdev->flags);
3269 clear_bit(WriteMostly, &rdev->flags);
3270 set_bit(In_sync, &rdev->flags);
3271 sysfs_notify_dirent_safe(rdev->sysfs_state);
3276 static struct rdev_sysfs_entry rdev_slot =
3277 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3280 offset_show(struct md_rdev *rdev, char *page)
3282 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3286 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3288 unsigned long long offset;
3289 if (kstrtoull(buf, 10, &offset) < 0)
3291 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3293 if (rdev->sectors && rdev->mddev->external)
3294 /* Must set offset before size, so overlap checks
3297 rdev->data_offset = offset;
3298 rdev->new_data_offset = offset;
3302 static struct rdev_sysfs_entry rdev_offset =
3303 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3305 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3307 return sprintf(page, "%llu\n",
3308 (unsigned long long)rdev->new_data_offset);
3311 static ssize_t new_offset_store(struct md_rdev *rdev,
3312 const char *buf, size_t len)
3314 unsigned long long new_offset;
3315 struct mddev *mddev = rdev->mddev;
3317 if (kstrtoull(buf, 10, &new_offset) < 0)
3320 if (mddev->sync_thread ||
3321 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3323 if (new_offset == rdev->data_offset)
3324 /* reset is always permitted */
3326 else if (new_offset > rdev->data_offset) {
3327 /* must not push array size beyond rdev_sectors */
3328 if (new_offset - rdev->data_offset
3329 + mddev->dev_sectors > rdev->sectors)
3332 /* Metadata worries about other space details. */
3334 /* decreasing the offset is inconsistent with a backwards
3337 if (new_offset < rdev->data_offset &&
3338 mddev->reshape_backwards)
3340 /* Increasing offset is inconsistent with forwards
3341 * reshape. reshape_direction should be set to
3342 * 'backwards' first.
3344 if (new_offset > rdev->data_offset &&
3345 !mddev->reshape_backwards)
3348 if (mddev->pers && mddev->persistent &&
3349 !super_types[mddev->major_version]
3350 .allow_new_offset(rdev, new_offset))
3352 rdev->new_data_offset = new_offset;
3353 if (new_offset > rdev->data_offset)
3354 mddev->reshape_backwards = 1;
3355 else if (new_offset < rdev->data_offset)
3356 mddev->reshape_backwards = 0;
3360 static struct rdev_sysfs_entry rdev_new_offset =
3361 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3364 rdev_size_show(struct md_rdev *rdev, char *page)
3366 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3369 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3371 /* check if two start/length pairs overlap */
3379 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3381 unsigned long long blocks;
3384 if (kstrtoull(buf, 10, &blocks) < 0)
3387 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3388 return -EINVAL; /* sector conversion overflow */
3391 if (new != blocks * 2)
3392 return -EINVAL; /* unsigned long long to sector_t overflow */
3399 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3401 struct mddev *my_mddev = rdev->mddev;
3402 sector_t oldsectors = rdev->sectors;
3405 if (test_bit(Journal, &rdev->flags))
3407 if (strict_blocks_to_sectors(buf, §ors) < 0)
3409 if (rdev->data_offset != rdev->new_data_offset)
3410 return -EINVAL; /* too confusing */
3411 if (my_mddev->pers && rdev->raid_disk >= 0) {
3412 if (my_mddev->persistent) {
3413 sectors = super_types[my_mddev->major_version].
3414 rdev_size_change(rdev, sectors);
3417 } else if (!sectors)
3418 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3420 if (!my_mddev->pers->resize)
3421 /* Cannot change size for RAID0 or Linear etc */
3424 if (sectors < my_mddev->dev_sectors)
3425 return -EINVAL; /* component must fit device */
3427 rdev->sectors = sectors;
3428 if (sectors > oldsectors && my_mddev->external) {
3429 /* Need to check that all other rdevs with the same
3430 * ->bdev do not overlap. 'rcu' is sufficient to walk
3431 * the rdev lists safely.
3432 * This check does not provide a hard guarantee, it
3433 * just helps avoid dangerous mistakes.
3435 struct mddev *mddev;
3437 struct list_head *tmp;
3440 for_each_mddev(mddev, tmp) {
3441 struct md_rdev *rdev2;
3443 rdev_for_each(rdev2, mddev)
3444 if (rdev->bdev == rdev2->bdev &&
3446 overlaps(rdev->data_offset, rdev->sectors,
3459 /* Someone else could have slipped in a size
3460 * change here, but doing so is just silly.
3461 * We put oldsectors back because we *know* it is
3462 * safe, and trust userspace not to race with
3465 rdev->sectors = oldsectors;
3472 static struct rdev_sysfs_entry rdev_size =
3473 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3475 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3477 unsigned long long recovery_start = rdev->recovery_offset;
3479 if (test_bit(In_sync, &rdev->flags) ||
3480 recovery_start == MaxSector)
3481 return sprintf(page, "none\n");
3483 return sprintf(page, "%llu\n", recovery_start);
3486 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3488 unsigned long long recovery_start;
3490 if (cmd_match(buf, "none"))
3491 recovery_start = MaxSector;
3492 else if (kstrtoull(buf, 10, &recovery_start))
3495 if (rdev->mddev->pers &&
3496 rdev->raid_disk >= 0)
3499 rdev->recovery_offset = recovery_start;
3500 if (recovery_start == MaxSector)
3501 set_bit(In_sync, &rdev->flags);
3503 clear_bit(In_sync, &rdev->flags);
3507 static struct rdev_sysfs_entry rdev_recovery_start =
3508 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3510 /* sysfs access to bad-blocks list.
3511 * We present two files.
3512 * 'bad-blocks' lists sector numbers and lengths of ranges that
3513 * are recorded as bad. The list is truncated to fit within
3514 * the one-page limit of sysfs.
3515 * Writing "sector length" to this file adds an acknowledged
3517 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3518 * been acknowledged. Writing to this file adds bad blocks
3519 * without acknowledging them. This is largely for testing.
3521 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3523 return badblocks_show(&rdev->badblocks, page, 0);
3525 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3527 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3528 /* Maybe that ack was all we needed */
3529 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3530 wake_up(&rdev->blocked_wait);
3533 static struct rdev_sysfs_entry rdev_bad_blocks =
3534 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3536 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3538 return badblocks_show(&rdev->badblocks, page, 1);
3540 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3542 return badblocks_store(&rdev->badblocks, page, len, 1);
3544 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3545 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3548 ppl_sector_show(struct md_rdev *rdev, char *page)
3550 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3554 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3556 unsigned long long sector;
3558 if (kstrtoull(buf, 10, §or) < 0)
3560 if (sector != (sector_t)sector)
3563 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3564 rdev->raid_disk >= 0)
3567 if (rdev->mddev->persistent) {
3568 if (rdev->mddev->major_version == 0)
3570 if ((sector > rdev->sb_start &&
3571 sector - rdev->sb_start > S16_MAX) ||
3572 (sector < rdev->sb_start &&
3573 rdev->sb_start - sector > -S16_MIN))
3575 rdev->ppl.offset = sector - rdev->sb_start;
3576 } else if (!rdev->mddev->external) {
3579 rdev->ppl.sector = sector;
3583 static struct rdev_sysfs_entry rdev_ppl_sector =
3584 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3587 ppl_size_show(struct md_rdev *rdev, char *page)
3589 return sprintf(page, "%u\n", rdev->ppl.size);
3593 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3597 if (kstrtouint(buf, 10, &size) < 0)
3600 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3601 rdev->raid_disk >= 0)
3604 if (rdev->mddev->persistent) {
3605 if (rdev->mddev->major_version == 0)
3609 } else if (!rdev->mddev->external) {
3612 rdev->ppl.size = size;
3616 static struct rdev_sysfs_entry rdev_ppl_size =
3617 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3619 static struct attribute *rdev_default_attrs[] = {
3624 &rdev_new_offset.attr,
3626 &rdev_recovery_start.attr,
3627 &rdev_bad_blocks.attr,
3628 &rdev_unack_bad_blocks.attr,
3629 &rdev_ppl_sector.attr,
3630 &rdev_ppl_size.attr,
3634 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3636 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3637 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3643 return entry->show(rdev, page);
3647 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3648 const char *page, size_t length)
3650 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3651 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3653 struct mddev *mddev = rdev->mddev;
3657 if (!capable(CAP_SYS_ADMIN))
3659 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3661 if (rdev->mddev == NULL)
3664 rv = entry->store(rdev, page, length);
3665 mddev_unlock(mddev);
3670 static void rdev_free(struct kobject *ko)
3672 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3675 static const struct sysfs_ops rdev_sysfs_ops = {
3676 .show = rdev_attr_show,
3677 .store = rdev_attr_store,
3679 static struct kobj_type rdev_ktype = {
3680 .release = rdev_free,
3681 .sysfs_ops = &rdev_sysfs_ops,
3682 .default_attrs = rdev_default_attrs,
3685 int md_rdev_init(struct md_rdev *rdev)
3688 rdev->saved_raid_disk = -1;
3689 rdev->raid_disk = -1;
3691 rdev->data_offset = 0;
3692 rdev->new_data_offset = 0;
3693 rdev->sb_events = 0;
3694 rdev->last_read_error = 0;
3695 rdev->sb_loaded = 0;
3696 rdev->bb_page = NULL;
3697 atomic_set(&rdev->nr_pending, 0);
3698 atomic_set(&rdev->read_errors, 0);
3699 atomic_set(&rdev->corrected_errors, 0);
3701 INIT_LIST_HEAD(&rdev->same_set);
3702 init_waitqueue_head(&rdev->blocked_wait);
3704 /* Add space to store bad block list.
3705 * This reserves the space even on arrays where it cannot
3706 * be used - I wonder if that matters
3708 return badblocks_init(&rdev->badblocks, 0);
3710 EXPORT_SYMBOL_GPL(md_rdev_init);
3712 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3714 * mark the device faulty if:
3716 * - the device is nonexistent (zero size)
3717 * - the device has no valid superblock
3719 * a faulty rdev _never_ has rdev->sb set.
3721 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3723 char b[BDEVNAME_SIZE];
3725 struct md_rdev *rdev;
3728 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3730 return ERR_PTR(-ENOMEM);
3732 err = md_rdev_init(rdev);
3735 err = alloc_disk_sb(rdev);
3739 err = lock_rdev(rdev, newdev, super_format == -2);
3743 kobject_init(&rdev->kobj, &rdev_ktype);
3745 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3747 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3748 bdevname(rdev->bdev,b));
3753 if (super_format >= 0) {
3754 err = super_types[super_format].
3755 load_super(rdev, NULL, super_minor);
3756 if (err == -EINVAL) {
3757 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3758 bdevname(rdev->bdev,b),
3759 super_format, super_minor);
3763 pr_warn("md: could not read %s's sb, not importing!\n",
3764 bdevname(rdev->bdev,b));
3774 md_rdev_clear(rdev);
3776 return ERR_PTR(err);
3780 * Check a full RAID array for plausibility
3783 static int analyze_sbs(struct mddev *mddev)
3786 struct md_rdev *rdev, *freshest, *tmp;
3787 char b[BDEVNAME_SIZE];
3790 rdev_for_each_safe(rdev, tmp, mddev)
3791 switch (super_types[mddev->major_version].
3792 load_super(rdev, freshest, mddev->minor_version)) {
3799 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3800 bdevname(rdev->bdev,b));
3801 md_kick_rdev_from_array(rdev);
3804 /* Cannot find a valid fresh disk */
3806 pr_warn("md: cannot find a valid disk\n");
3810 super_types[mddev->major_version].
3811 validate_super(mddev, freshest);
3814 rdev_for_each_safe(rdev, tmp, mddev) {
3815 if (mddev->max_disks &&
3816 (rdev->desc_nr >= mddev->max_disks ||
3817 i > mddev->max_disks)) {
3818 pr_warn("md: %s: %s: only %d devices permitted\n",
3819 mdname(mddev), bdevname(rdev->bdev, b),
3821 md_kick_rdev_from_array(rdev);
3824 if (rdev != freshest) {
3825 if (super_types[mddev->major_version].
3826 validate_super(mddev, rdev)) {
3827 pr_warn("md: kicking non-fresh %s from array!\n",
3828 bdevname(rdev->bdev,b));
3829 md_kick_rdev_from_array(rdev);
3833 if (mddev->level == LEVEL_MULTIPATH) {
3834 rdev->desc_nr = i++;
3835 rdev->raid_disk = rdev->desc_nr;
3836 set_bit(In_sync, &rdev->flags);
3837 } else if (rdev->raid_disk >=
3838 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3839 !test_bit(Journal, &rdev->flags)) {
3840 rdev->raid_disk = -1;
3841 clear_bit(In_sync, &rdev->flags);
3848 /* Read a fixed-point number.
3849 * Numbers in sysfs attributes should be in "standard" units where
3850 * possible, so time should be in seconds.
3851 * However we internally use a a much smaller unit such as
3852 * milliseconds or jiffies.
3853 * This function takes a decimal number with a possible fractional
3854 * component, and produces an integer which is the result of
3855 * multiplying that number by 10^'scale'.
3856 * all without any floating-point arithmetic.
3858 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3860 unsigned long result = 0;
3862 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3865 else if (decimals < scale) {
3868 result = result * 10 + value;
3880 *res = result * int_pow(10, scale - decimals);
3885 safe_delay_show(struct mddev *mddev, char *page)
3887 int msec = (mddev->safemode_delay*1000)/HZ;
3888 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3891 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3895 if (mddev_is_clustered(mddev)) {
3896 pr_warn("md: Safemode is disabled for clustered mode\n");
3900 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3903 mddev->safemode_delay = 0;
3905 unsigned long old_delay = mddev->safemode_delay;
3906 unsigned long new_delay = (msec*HZ)/1000;
3910 mddev->safemode_delay = new_delay;
3911 if (new_delay < old_delay || old_delay == 0)
3912 mod_timer(&mddev->safemode_timer, jiffies+1);
3916 static struct md_sysfs_entry md_safe_delay =
3917 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3920 level_show(struct mddev *mddev, char *page)
3922 struct md_personality *p;
3924 spin_lock(&mddev->lock);
3927 ret = sprintf(page, "%s\n", p->name);
3928 else if (mddev->clevel[0])
3929 ret = sprintf(page, "%s\n", mddev->clevel);
3930 else if (mddev->level != LEVEL_NONE)
3931 ret = sprintf(page, "%d\n", mddev->level);
3934 spin_unlock(&mddev->lock);
3939 level_store(struct mddev *mddev, const char *buf, size_t len)
3944 struct md_personality *pers, *oldpers;
3946 void *priv, *oldpriv;
3947 struct md_rdev *rdev;
3949 if (slen == 0 || slen >= sizeof(clevel))
3952 rv = mddev_lock(mddev);
3956 if (mddev->pers == NULL) {
3957 strncpy(mddev->clevel, buf, slen);
3958 if (mddev->clevel[slen-1] == '\n')
3960 mddev->clevel[slen] = 0;
3961 mddev->level = LEVEL_NONE;
3969 /* request to change the personality. Need to ensure:
3970 * - array is not engaged in resync/recovery/reshape
3971 * - old personality can be suspended
3972 * - new personality will access other array.
3976 if (mddev->sync_thread ||
3977 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3978 mddev->reshape_position != MaxSector ||
3979 mddev->sysfs_active)
3983 if (!mddev->pers->quiesce) {
3984 pr_warn("md: %s: %s does not support online personality change\n",
3985 mdname(mddev), mddev->pers->name);
3989 /* Now find the new personality */
3990 strncpy(clevel, buf, slen);
3991 if (clevel[slen-1] == '\n')
3994 if (kstrtol(clevel, 10, &level))
3997 if (request_module("md-%s", clevel) != 0)
3998 request_module("md-level-%s", clevel);
3999 spin_lock(&pers_lock);
4000 pers = find_pers(level, clevel);
4001 if (!pers || !try_module_get(pers->owner)) {
4002 spin_unlock(&pers_lock);
4003 pr_warn("md: personality %s not loaded\n", clevel);
4007 spin_unlock(&pers_lock);
4009 if (pers == mddev->pers) {
4010 /* Nothing to do! */
4011 module_put(pers->owner);
4015 if (!pers->takeover) {
4016 module_put(pers->owner);
4017 pr_warn("md: %s: %s does not support personality takeover\n",
4018 mdname(mddev), clevel);
4023 rdev_for_each(rdev, mddev)
4024 rdev->new_raid_disk = rdev->raid_disk;
4026 /* ->takeover must set new_* and/or delta_disks
4027 * if it succeeds, and may set them when it fails.
4029 priv = pers->takeover(mddev);
4031 mddev->new_level = mddev->level;
4032 mddev->new_layout = mddev->layout;
4033 mddev->new_chunk_sectors = mddev->chunk_sectors;
4034 mddev->raid_disks -= mddev->delta_disks;
4035 mddev->delta_disks = 0;
4036 mddev->reshape_backwards = 0;
4037 module_put(pers->owner);
4038 pr_warn("md: %s: %s would not accept array\n",
4039 mdname(mddev), clevel);
4044 /* Looks like we have a winner */
4045 mddev_suspend(mddev);
4046 mddev_detach(mddev);
4048 spin_lock(&mddev->lock);
4049 oldpers = mddev->pers;
4050 oldpriv = mddev->private;
4052 mddev->private = priv;
4053 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4054 mddev->level = mddev->new_level;
4055 mddev->layout = mddev->new_layout;
4056 mddev->chunk_sectors = mddev->new_chunk_sectors;
4057 mddev->delta_disks = 0;
4058 mddev->reshape_backwards = 0;
4059 mddev->degraded = 0;
4060 spin_unlock(&mddev->lock);
4062 if (oldpers->sync_request == NULL &&
4064 /* We are converting from a no-redundancy array
4065 * to a redundancy array and metadata is managed
4066 * externally so we need to be sure that writes
4067 * won't block due to a need to transition
4069 * until external management is started.
4072 mddev->safemode_delay = 0;
4073 mddev->safemode = 0;
4076 oldpers->free(mddev, oldpriv);
4078 if (oldpers->sync_request == NULL &&
4079 pers->sync_request != NULL) {
4080 /* need to add the md_redundancy_group */
4081 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4082 pr_warn("md: cannot register extra attributes for %s\n",
4084 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4085 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4086 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4088 if (oldpers->sync_request != NULL &&
4089 pers->sync_request == NULL) {
4090 /* need to remove the md_redundancy_group */
4091 if (mddev->to_remove == NULL)
4092 mddev->to_remove = &md_redundancy_group;
4095 module_put(oldpers->owner);
4097 rdev_for_each(rdev, mddev) {
4098 if (rdev->raid_disk < 0)
4100 if (rdev->new_raid_disk >= mddev->raid_disks)
4101 rdev->new_raid_disk = -1;
4102 if (rdev->new_raid_disk == rdev->raid_disk)
4104 sysfs_unlink_rdev(mddev, rdev);
4106 rdev_for_each(rdev, mddev) {
4107 if (rdev->raid_disk < 0)
4109 if (rdev->new_raid_disk == rdev->raid_disk)
4111 rdev->raid_disk = rdev->new_raid_disk;
4112 if (rdev->raid_disk < 0)
4113 clear_bit(In_sync, &rdev->flags);
4115 if (sysfs_link_rdev(mddev, rdev))
4116 pr_warn("md: cannot register rd%d for %s after level change\n",
4117 rdev->raid_disk, mdname(mddev));
4121 if (pers->sync_request == NULL) {
4122 /* this is now an array without redundancy, so
4123 * it must always be in_sync
4126 del_timer_sync(&mddev->safemode_timer);
4128 blk_set_stacking_limits(&mddev->queue->limits);
4130 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4131 mddev_resume(mddev);
4133 md_update_sb(mddev, 1);
4134 sysfs_notify_dirent_safe(mddev->sysfs_level);
4135 md_new_event(mddev);
4138 mddev_unlock(mddev);
4142 static struct md_sysfs_entry md_level =
4143 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4146 layout_show(struct mddev *mddev, char *page)
4148 /* just a number, not meaningful for all levels */
4149 if (mddev->reshape_position != MaxSector &&
4150 mddev->layout != mddev->new_layout)
4151 return sprintf(page, "%d (%d)\n",
4152 mddev->new_layout, mddev->layout);
4153 return sprintf(page, "%d\n", mddev->layout);
4157 layout_store(struct mddev *mddev, const char *buf, size_t len)
4162 err = kstrtouint(buf, 10, &n);
4165 err = mddev_lock(mddev);
4170 if (mddev->pers->check_reshape == NULL)
4175 mddev->new_layout = n;
4176 err = mddev->pers->check_reshape(mddev);
4178 mddev->new_layout = mddev->layout;
4181 mddev->new_layout = n;
4182 if (mddev->reshape_position == MaxSector)
4185 mddev_unlock(mddev);
4188 static struct md_sysfs_entry md_layout =
4189 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4192 raid_disks_show(struct mddev *mddev, char *page)
4194 if (mddev->raid_disks == 0)
4196 if (mddev->reshape_position != MaxSector &&
4197 mddev->delta_disks != 0)
4198 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4199 mddev->raid_disks - mddev->delta_disks);
4200 return sprintf(page, "%d\n", mddev->raid_disks);
4203 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4206 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4211 err = kstrtouint(buf, 10, &n);
4215 err = mddev_lock(mddev);
4219 err = update_raid_disks(mddev, n);
4220 else if (mddev->reshape_position != MaxSector) {
4221 struct md_rdev *rdev;
4222 int olddisks = mddev->raid_disks - mddev->delta_disks;
4225 rdev_for_each(rdev, mddev) {
4227 rdev->data_offset < rdev->new_data_offset)
4230 rdev->data_offset > rdev->new_data_offset)
4234 mddev->delta_disks = n - olddisks;
4235 mddev->raid_disks = n;
4236 mddev->reshape_backwards = (mddev->delta_disks < 0);
4238 mddev->raid_disks = n;
4240 mddev_unlock(mddev);
4241 return err ? err : len;
4243 static struct md_sysfs_entry md_raid_disks =
4244 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4247 uuid_show(struct mddev *mddev, char *page)
4249 return sprintf(page, "%pU\n", mddev->uuid);
4251 static struct md_sysfs_entry md_uuid =
4252 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4255 chunk_size_show(struct mddev *mddev, char *page)
4257 if (mddev->reshape_position != MaxSector &&
4258 mddev->chunk_sectors != mddev->new_chunk_sectors)
4259 return sprintf(page, "%d (%d)\n",
4260 mddev->new_chunk_sectors << 9,
4261 mddev->chunk_sectors << 9);
4262 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4266 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4271 err = kstrtoul(buf, 10, &n);
4275 err = mddev_lock(mddev);
4279 if (mddev->pers->check_reshape == NULL)
4284 mddev->new_chunk_sectors = n >> 9;
4285 err = mddev->pers->check_reshape(mddev);
4287 mddev->new_chunk_sectors = mddev->chunk_sectors;
4290 mddev->new_chunk_sectors = n >> 9;
4291 if (mddev->reshape_position == MaxSector)
4292 mddev->chunk_sectors = n >> 9;
4294 mddev_unlock(mddev);
4297 static struct md_sysfs_entry md_chunk_size =
4298 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4301 resync_start_show(struct mddev *mddev, char *page)
4303 if (mddev->recovery_cp == MaxSector)
4304 return sprintf(page, "none\n");
4305 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4309 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4311 unsigned long long n;
4314 if (cmd_match(buf, "none"))
4317 err = kstrtoull(buf, 10, &n);
4320 if (n != (sector_t)n)
4324 err = mddev_lock(mddev);
4327 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4331 mddev->recovery_cp = n;
4333 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4335 mddev_unlock(mddev);
4338 static struct md_sysfs_entry md_resync_start =
4339 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4340 resync_start_show, resync_start_store);
4343 * The array state can be:
4346 * No devices, no size, no level
4347 * Equivalent to STOP_ARRAY ioctl
4349 * May have some settings, but array is not active
4350 * all IO results in error
4351 * When written, doesn't tear down array, but just stops it
4352 * suspended (not supported yet)
4353 * All IO requests will block. The array can be reconfigured.
4354 * Writing this, if accepted, will block until array is quiescent
4356 * no resync can happen. no superblocks get written.
4357 * write requests fail
4359 * like readonly, but behaves like 'clean' on a write request.
4361 * clean - no pending writes, but otherwise active.
4362 * When written to inactive array, starts without resync
4363 * If a write request arrives then
4364 * if metadata is known, mark 'dirty' and switch to 'active'.
4365 * if not known, block and switch to write-pending
4366 * If written to an active array that has pending writes, then fails.
4368 * fully active: IO and resync can be happening.
4369 * When written to inactive array, starts with resync
4372 * clean, but writes are blocked waiting for 'active' to be written.
4375 * like active, but no writes have been seen for a while (100msec).
4378 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4379 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4380 * when a member is gone, so this state will at least alert the
4381 * user that something is wrong.
4383 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4384 write_pending, active_idle, broken, bad_word};
4385 static char *array_states[] = {
4386 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4387 "write-pending", "active-idle", "broken", NULL };
4389 static int match_word(const char *word, char **list)
4392 for (n=0; list[n]; n++)
4393 if (cmd_match(word, list[n]))
4399 array_state_show(struct mddev *mddev, char *page)
4401 enum array_state st = inactive;
4403 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4412 spin_lock(&mddev->lock);
4413 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4415 else if (mddev->in_sync)
4417 else if (mddev->safemode)
4421 spin_unlock(&mddev->lock);
4424 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4427 if (list_empty(&mddev->disks) &&
4428 mddev->raid_disks == 0 &&
4429 mddev->dev_sectors == 0)
4434 return sprintf(page, "%s\n", array_states[st]);
4437 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4438 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4439 static int restart_array(struct mddev *mddev);
4442 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4445 enum array_state st = match_word(buf, array_states);
4447 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4448 /* don't take reconfig_mutex when toggling between
4451 spin_lock(&mddev->lock);
4453 restart_array(mddev);
4454 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4455 md_wakeup_thread(mddev->thread);
4456 wake_up(&mddev->sb_wait);
4457 } else /* st == clean */ {
4458 restart_array(mddev);
4459 if (!set_in_sync(mddev))
4463 sysfs_notify_dirent_safe(mddev->sysfs_state);
4464 spin_unlock(&mddev->lock);
4467 err = mddev_lock(mddev);
4475 /* stopping an active array */
4476 err = do_md_stop(mddev, 0, NULL);
4479 /* stopping an active array */
4481 err = do_md_stop(mddev, 2, NULL);
4483 err = 0; /* already inactive */
4486 break; /* not supported yet */
4489 err = md_set_readonly(mddev, NULL);
4492 set_disk_ro(mddev->gendisk, 1);
4493 err = do_md_run(mddev);
4499 err = md_set_readonly(mddev, NULL);
4500 else if (mddev->ro == 1)
4501 err = restart_array(mddev);
4504 set_disk_ro(mddev->gendisk, 0);
4508 err = do_md_run(mddev);
4513 err = restart_array(mddev);
4516 spin_lock(&mddev->lock);
4517 if (!set_in_sync(mddev))
4519 spin_unlock(&mddev->lock);
4525 err = restart_array(mddev);
4528 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4529 wake_up(&mddev->sb_wait);
4533 set_disk_ro(mddev->gendisk, 0);
4534 err = do_md_run(mddev);
4540 /* these cannot be set */
4545 if (mddev->hold_active == UNTIL_IOCTL)
4546 mddev->hold_active = 0;
4547 sysfs_notify_dirent_safe(mddev->sysfs_state);
4549 mddev_unlock(mddev);
4552 static struct md_sysfs_entry md_array_state =
4553 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4556 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4557 return sprintf(page, "%d\n",
4558 atomic_read(&mddev->max_corr_read_errors));
4562 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4567 rv = kstrtouint(buf, 10, &n);
4570 atomic_set(&mddev->max_corr_read_errors, n);
4574 static struct md_sysfs_entry max_corr_read_errors =
4575 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4576 max_corrected_read_errors_store);
4579 null_show(struct mddev *mddev, char *page)
4584 /* need to ensure rdev_delayed_delete() has completed */
4585 static void flush_rdev_wq(struct mddev *mddev)
4587 struct md_rdev *rdev;
4590 rdev_for_each_rcu(rdev, mddev)
4591 if (work_pending(&rdev->del_work)) {
4592 flush_workqueue(md_rdev_misc_wq);
4599 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4601 /* buf must be %d:%d\n? giving major and minor numbers */
4602 /* The new device is added to the array.
4603 * If the array has a persistent superblock, we read the
4604 * superblock to initialise info and check validity.
4605 * Otherwise, only checking done is that in bind_rdev_to_array,
4606 * which mainly checks size.
4609 int major = simple_strtoul(buf, &e, 10);
4612 struct md_rdev *rdev;
4615 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4617 minor = simple_strtoul(e+1, &e, 10);
4618 if (*e && *e != '\n')
4620 dev = MKDEV(major, minor);
4621 if (major != MAJOR(dev) ||
4622 minor != MINOR(dev))
4625 flush_rdev_wq(mddev);
4626 err = mddev_lock(mddev);
4629 if (mddev->persistent) {
4630 rdev = md_import_device(dev, mddev->major_version,
4631 mddev->minor_version);
4632 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4633 struct md_rdev *rdev0
4634 = list_entry(mddev->disks.next,
4635 struct md_rdev, same_set);
4636 err = super_types[mddev->major_version]
4637 .load_super(rdev, rdev0, mddev->minor_version);
4641 } else if (mddev->external)
4642 rdev = md_import_device(dev, -2, -1);
4644 rdev = md_import_device(dev, -1, -1);
4647 mddev_unlock(mddev);
4648 return PTR_ERR(rdev);
4650 err = bind_rdev_to_array(rdev, mddev);
4654 mddev_unlock(mddev);
4656 md_new_event(mddev);
4657 return err ? err : len;
4660 static struct md_sysfs_entry md_new_device =
4661 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4664 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4667 unsigned long chunk, end_chunk;
4670 err = mddev_lock(mddev);
4675 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4677 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4678 if (buf == end) break;
4679 if (*end == '-') { /* range */
4681 end_chunk = simple_strtoul(buf, &end, 0);
4682 if (buf == end) break;
4684 if (*end && !isspace(*end)) break;
4685 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4686 buf = skip_spaces(end);
4688 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4690 mddev_unlock(mddev);
4694 static struct md_sysfs_entry md_bitmap =
4695 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4698 size_show(struct mddev *mddev, char *page)
4700 return sprintf(page, "%llu\n",
4701 (unsigned long long)mddev->dev_sectors / 2);
4704 static int update_size(struct mddev *mddev, sector_t num_sectors);
4707 size_store(struct mddev *mddev, const char *buf, size_t len)
4709 /* If array is inactive, we can reduce the component size, but
4710 * not increase it (except from 0).
4711 * If array is active, we can try an on-line resize
4714 int err = strict_blocks_to_sectors(buf, §ors);
4718 err = mddev_lock(mddev);
4722 err = update_size(mddev, sectors);
4724 md_update_sb(mddev, 1);
4726 if (mddev->dev_sectors == 0 ||
4727 mddev->dev_sectors > sectors)
4728 mddev->dev_sectors = sectors;
4732 mddev_unlock(mddev);
4733 return err ? err : len;
4736 static struct md_sysfs_entry md_size =
4737 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4739 /* Metadata version.
4741 * 'none' for arrays with no metadata (good luck...)
4742 * 'external' for arrays with externally managed metadata,
4743 * or N.M for internally known formats
4746 metadata_show(struct mddev *mddev, char *page)
4748 if (mddev->persistent)
4749 return sprintf(page, "%d.%d\n",
4750 mddev->major_version, mddev->minor_version);
4751 else if (mddev->external)
4752 return sprintf(page, "external:%s\n", mddev->metadata_type);
4754 return sprintf(page, "none\n");
4758 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4763 /* Changing the details of 'external' metadata is
4764 * always permitted. Otherwise there must be
4765 * no devices attached to the array.
4768 err = mddev_lock(mddev);
4772 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4774 else if (!list_empty(&mddev->disks))
4778 if (cmd_match(buf, "none")) {
4779 mddev->persistent = 0;
4780 mddev->external = 0;
4781 mddev->major_version = 0;
4782 mddev->minor_version = 90;
4785 if (strncmp(buf, "external:", 9) == 0) {
4786 size_t namelen = len-9;
4787 if (namelen >= sizeof(mddev->metadata_type))
4788 namelen = sizeof(mddev->metadata_type)-1;
4789 strncpy(mddev->metadata_type, buf+9, namelen);
4790 mddev->metadata_type[namelen] = 0;
4791 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4792 mddev->metadata_type[--namelen] = 0;
4793 mddev->persistent = 0;
4794 mddev->external = 1;
4795 mddev->major_version = 0;
4796 mddev->minor_version = 90;
4799 major = simple_strtoul(buf, &e, 10);
4801 if (e==buf || *e != '.')
4804 minor = simple_strtoul(buf, &e, 10);
4805 if (e==buf || (*e && *e != '\n') )
4808 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4810 mddev->major_version = major;
4811 mddev->minor_version = minor;
4812 mddev->persistent = 1;
4813 mddev->external = 0;
4816 mddev_unlock(mddev);
4820 static struct md_sysfs_entry md_metadata =
4821 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4824 action_show(struct mddev *mddev, char *page)
4826 char *type = "idle";
4827 unsigned long recovery = mddev->recovery;
4828 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4830 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4831 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4832 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4834 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4835 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4837 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4841 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4843 else if (mddev->reshape_position != MaxSector)
4846 return sprintf(page, "%s\n", type);
4850 action_store(struct mddev *mddev, const char *page, size_t len)
4852 if (!mddev->pers || !mddev->pers->sync_request)
4856 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4857 if (cmd_match(page, "frozen"))
4858 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4860 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4861 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4862 mddev_lock(mddev) == 0) {
4863 if (work_pending(&mddev->del_work))
4864 flush_workqueue(md_misc_wq);
4865 if (mddev->sync_thread) {
4866 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4867 md_reap_sync_thread(mddev);
4869 mddev_unlock(mddev);
4871 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4873 else if (cmd_match(page, "resync"))
4874 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4875 else if (cmd_match(page, "recover")) {
4876 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4877 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4878 } else if (cmd_match(page, "reshape")) {
4880 if (mddev->pers->start_reshape == NULL)
4882 err = mddev_lock(mddev);
4884 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4887 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4888 err = mddev->pers->start_reshape(mddev);
4890 mddev_unlock(mddev);
4894 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4896 if (cmd_match(page, "check"))
4897 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4898 else if (!cmd_match(page, "repair"))
4900 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4901 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4902 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4904 if (mddev->ro == 2) {
4905 /* A write to sync_action is enough to justify
4906 * canceling read-auto mode
4909 md_wakeup_thread(mddev->sync_thread);
4911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4912 md_wakeup_thread(mddev->thread);
4913 sysfs_notify_dirent_safe(mddev->sysfs_action);
4917 static struct md_sysfs_entry md_scan_mode =
4918 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4921 last_sync_action_show(struct mddev *mddev, char *page)
4923 return sprintf(page, "%s\n", mddev->last_sync_action);
4926 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4929 mismatch_cnt_show(struct mddev *mddev, char *page)
4931 return sprintf(page, "%llu\n",
4932 (unsigned long long)
4933 atomic64_read(&mddev->resync_mismatches));
4936 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4939 sync_min_show(struct mddev *mddev, char *page)
4941 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4942 mddev->sync_speed_min ? "local": "system");
4946 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4951 if (strncmp(buf, "system", 6)==0) {
4954 rv = kstrtouint(buf, 10, &min);
4960 mddev->sync_speed_min = min;
4964 static struct md_sysfs_entry md_sync_min =
4965 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4968 sync_max_show(struct mddev *mddev, char *page)
4970 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4971 mddev->sync_speed_max ? "local": "system");
4975 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4980 if (strncmp(buf, "system", 6)==0) {
4983 rv = kstrtouint(buf, 10, &max);
4989 mddev->sync_speed_max = max;
4993 static struct md_sysfs_entry md_sync_max =
4994 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4997 degraded_show(struct mddev *mddev, char *page)
4999 return sprintf(page, "%d\n", mddev->degraded);
5001 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5004 sync_force_parallel_show(struct mddev *mddev, char *page)
5006 return sprintf(page, "%d\n", mddev->parallel_resync);
5010 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5014 if (kstrtol(buf, 10, &n))
5017 if (n != 0 && n != 1)
5020 mddev->parallel_resync = n;
5022 if (mddev->sync_thread)
5023 wake_up(&resync_wait);
5028 /* force parallel resync, even with shared block devices */
5029 static struct md_sysfs_entry md_sync_force_parallel =
5030 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5031 sync_force_parallel_show, sync_force_parallel_store);
5034 sync_speed_show(struct mddev *mddev, char *page)
5036 unsigned long resync, dt, db;
5037 if (mddev->curr_resync == 0)
5038 return sprintf(page, "none\n");
5039 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5040 dt = (jiffies - mddev->resync_mark) / HZ;
5042 db = resync - mddev->resync_mark_cnt;
5043 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5046 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5049 sync_completed_show(struct mddev *mddev, char *page)
5051 unsigned long long max_sectors, resync;
5053 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5054 return sprintf(page, "none\n");
5056 if (mddev->curr_resync == 1 ||
5057 mddev->curr_resync == 2)
5058 return sprintf(page, "delayed\n");
5060 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5061 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5062 max_sectors = mddev->resync_max_sectors;
5064 max_sectors = mddev->dev_sectors;
5066 resync = mddev->curr_resync_completed;
5067 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5070 static struct md_sysfs_entry md_sync_completed =
5071 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5074 min_sync_show(struct mddev *mddev, char *page)
5076 return sprintf(page, "%llu\n",
5077 (unsigned long long)mddev->resync_min);
5080 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5082 unsigned long long min;
5085 if (kstrtoull(buf, 10, &min))
5088 spin_lock(&mddev->lock);
5090 if (min > mddev->resync_max)
5094 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5097 /* Round down to multiple of 4K for safety */
5098 mddev->resync_min = round_down(min, 8);
5102 spin_unlock(&mddev->lock);
5106 static struct md_sysfs_entry md_min_sync =
5107 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5110 max_sync_show(struct mddev *mddev, char *page)
5112 if (mddev->resync_max == MaxSector)
5113 return sprintf(page, "max\n");
5115 return sprintf(page, "%llu\n",
5116 (unsigned long long)mddev->resync_max);
5119 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5122 spin_lock(&mddev->lock);
5123 if (strncmp(buf, "max", 3) == 0)
5124 mddev->resync_max = MaxSector;
5126 unsigned long long max;
5130 if (kstrtoull(buf, 10, &max))
5132 if (max < mddev->resync_min)
5136 if (max < mddev->resync_max &&
5138 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5141 /* Must be a multiple of chunk_size */
5142 chunk = mddev->chunk_sectors;
5144 sector_t temp = max;
5147 if (sector_div(temp, chunk))
5150 mddev->resync_max = max;
5152 wake_up(&mddev->recovery_wait);
5155 spin_unlock(&mddev->lock);
5159 static struct md_sysfs_entry md_max_sync =
5160 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5163 suspend_lo_show(struct mddev *mddev, char *page)
5165 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5169 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5171 unsigned long long new;
5174 err = kstrtoull(buf, 10, &new);
5177 if (new != (sector_t)new)
5180 err = mddev_lock(mddev);
5184 if (mddev->pers == NULL ||
5185 mddev->pers->quiesce == NULL)
5187 mddev_suspend(mddev);
5188 mddev->suspend_lo = new;
5189 mddev_resume(mddev);
5193 mddev_unlock(mddev);
5196 static struct md_sysfs_entry md_suspend_lo =
5197 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5200 suspend_hi_show(struct mddev *mddev, char *page)
5202 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5206 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5208 unsigned long long new;
5211 err = kstrtoull(buf, 10, &new);
5214 if (new != (sector_t)new)
5217 err = mddev_lock(mddev);
5221 if (mddev->pers == NULL)
5224 mddev_suspend(mddev);
5225 mddev->suspend_hi = new;
5226 mddev_resume(mddev);
5230 mddev_unlock(mddev);
5233 static struct md_sysfs_entry md_suspend_hi =
5234 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5237 reshape_position_show(struct mddev *mddev, char *page)
5239 if (mddev->reshape_position != MaxSector)
5240 return sprintf(page, "%llu\n",
5241 (unsigned long long)mddev->reshape_position);
5242 strcpy(page, "none\n");
5247 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5249 struct md_rdev *rdev;
5250 unsigned long long new;
5253 err = kstrtoull(buf, 10, &new);
5256 if (new != (sector_t)new)
5258 err = mddev_lock(mddev);
5264 mddev->reshape_position = new;
5265 mddev->delta_disks = 0;
5266 mddev->reshape_backwards = 0;
5267 mddev->new_level = mddev->level;
5268 mddev->new_layout = mddev->layout;
5269 mddev->new_chunk_sectors = mddev->chunk_sectors;
5270 rdev_for_each(rdev, mddev)
5271 rdev->new_data_offset = rdev->data_offset;
5274 mddev_unlock(mddev);
5278 static struct md_sysfs_entry md_reshape_position =
5279 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5280 reshape_position_store);
5283 reshape_direction_show(struct mddev *mddev, char *page)
5285 return sprintf(page, "%s\n",
5286 mddev->reshape_backwards ? "backwards" : "forwards");
5290 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5295 if (cmd_match(buf, "forwards"))
5297 else if (cmd_match(buf, "backwards"))
5301 if (mddev->reshape_backwards == backwards)
5304 err = mddev_lock(mddev);
5307 /* check if we are allowed to change */
5308 if (mddev->delta_disks)
5310 else if (mddev->persistent &&
5311 mddev->major_version == 0)
5314 mddev->reshape_backwards = backwards;
5315 mddev_unlock(mddev);
5319 static struct md_sysfs_entry md_reshape_direction =
5320 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5321 reshape_direction_store);
5324 array_size_show(struct mddev *mddev, char *page)
5326 if (mddev->external_size)
5327 return sprintf(page, "%llu\n",
5328 (unsigned long long)mddev->array_sectors/2);
5330 return sprintf(page, "default\n");
5334 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5339 err = mddev_lock(mddev);
5343 /* cluster raid doesn't support change array_sectors */
5344 if (mddev_is_clustered(mddev)) {
5345 mddev_unlock(mddev);
5349 if (strncmp(buf, "default", 7) == 0) {
5351 sectors = mddev->pers->size(mddev, 0, 0);
5353 sectors = mddev->array_sectors;
5355 mddev->external_size = 0;
5357 if (strict_blocks_to_sectors(buf, §ors) < 0)
5359 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5362 mddev->external_size = 1;
5366 mddev->array_sectors = sectors;
5368 set_capacity(mddev->gendisk, mddev->array_sectors);
5369 revalidate_disk_size(mddev->gendisk, true);
5372 mddev_unlock(mddev);
5376 static struct md_sysfs_entry md_array_size =
5377 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5381 consistency_policy_show(struct mddev *mddev, char *page)
5385 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5386 ret = sprintf(page, "journal\n");
5387 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5388 ret = sprintf(page, "ppl\n");
5389 } else if (mddev->bitmap) {
5390 ret = sprintf(page, "bitmap\n");
5391 } else if (mddev->pers) {
5392 if (mddev->pers->sync_request)
5393 ret = sprintf(page, "resync\n");
5395 ret = sprintf(page, "none\n");
5397 ret = sprintf(page, "unknown\n");
5404 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5409 if (mddev->pers->change_consistency_policy)
5410 err = mddev->pers->change_consistency_policy(mddev, buf);
5413 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5414 set_bit(MD_HAS_PPL, &mddev->flags);
5419 return err ? err : len;
5422 static struct md_sysfs_entry md_consistency_policy =
5423 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5424 consistency_policy_store);
5426 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5428 return sprintf(page, "%d\n", mddev->fail_last_dev);
5432 * Setting fail_last_dev to true to allow last device to be forcibly removed
5433 * from RAID1/RAID10.
5436 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5441 ret = kstrtobool(buf, &value);
5445 if (value != mddev->fail_last_dev)
5446 mddev->fail_last_dev = value;
5450 static struct md_sysfs_entry md_fail_last_dev =
5451 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5452 fail_last_dev_store);
5454 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5456 if (mddev->pers == NULL || (mddev->pers->level != 1))
5457 return sprintf(page, "n/a\n");
5459 return sprintf(page, "%d\n", mddev->serialize_policy);
5463 * Setting serialize_policy to true to enforce write IO is not reordered
5467 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5472 err = kstrtobool(buf, &value);
5476 if (value == mddev->serialize_policy)
5479 err = mddev_lock(mddev);
5482 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5483 pr_err("md: serialize_policy is only effective for raid1\n");
5488 mddev_suspend(mddev);
5490 mddev_create_serial_pool(mddev, NULL, true);
5492 mddev_destroy_serial_pool(mddev, NULL, true);
5493 mddev->serialize_policy = value;
5494 mddev_resume(mddev);
5496 mddev_unlock(mddev);
5500 static struct md_sysfs_entry md_serialize_policy =
5501 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5502 serialize_policy_store);
5505 static struct attribute *md_default_attrs[] = {
5508 &md_raid_disks.attr,
5510 &md_chunk_size.attr,
5512 &md_resync_start.attr,
5514 &md_new_device.attr,
5515 &md_safe_delay.attr,
5516 &md_array_state.attr,
5517 &md_reshape_position.attr,
5518 &md_reshape_direction.attr,
5519 &md_array_size.attr,
5520 &max_corr_read_errors.attr,
5521 &md_consistency_policy.attr,
5522 &md_fail_last_dev.attr,
5523 &md_serialize_policy.attr,
5527 static struct attribute *md_redundancy_attrs[] = {
5529 &md_last_scan_mode.attr,
5530 &md_mismatches.attr,
5533 &md_sync_speed.attr,
5534 &md_sync_force_parallel.attr,
5535 &md_sync_completed.attr,
5538 &md_suspend_lo.attr,
5539 &md_suspend_hi.attr,
5544 static struct attribute_group md_redundancy_group = {
5546 .attrs = md_redundancy_attrs,
5550 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5552 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5553 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
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);
5566 rv = entry->show(mddev, page);
5572 md_attr_store(struct kobject *kobj, struct attribute *attr,
5573 const char *page, size_t length)
5575 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5576 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5581 if (!capable(CAP_SYS_ADMIN))
5583 spin_lock(&all_mddevs_lock);
5584 if (list_empty(&mddev->all_mddevs)) {
5585 spin_unlock(&all_mddevs_lock);
5589 spin_unlock(&all_mddevs_lock);
5590 rv = entry->store(mddev, page, length);
5595 static void md_free(struct kobject *ko)
5597 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5599 if (mddev->sysfs_state)
5600 sysfs_put(mddev->sysfs_state);
5601 if (mddev->sysfs_level)
5602 sysfs_put(mddev->sysfs_level);
5605 del_gendisk(mddev->gendisk);
5607 blk_cleanup_queue(mddev->queue);
5609 put_disk(mddev->gendisk);
5610 percpu_ref_exit(&mddev->writes_pending);
5612 bioset_exit(&mddev->bio_set);
5613 bioset_exit(&mddev->sync_set);
5617 static const struct sysfs_ops md_sysfs_ops = {
5618 .show = md_attr_show,
5619 .store = md_attr_store,
5621 static struct kobj_type md_ktype = {
5623 .sysfs_ops = &md_sysfs_ops,
5624 .default_attrs = md_default_attrs,
5629 static void mddev_delayed_delete(struct work_struct *ws)
5631 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5633 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5634 kobject_del(&mddev->kobj);
5635 kobject_put(&mddev->kobj);
5638 static void no_op(struct percpu_ref *r) {}
5640 int mddev_init_writes_pending(struct mddev *mddev)
5642 if (mddev->writes_pending.percpu_count_ptr)
5644 if (percpu_ref_init(&mddev->writes_pending, no_op,
5645 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5647 /* We want to start with the refcount at zero */
5648 percpu_ref_put(&mddev->writes_pending);
5651 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5653 static int md_alloc(dev_t dev, char *name)
5656 * If dev is zero, name is the name of a device to allocate with
5657 * an arbitrary minor number. It will be "md_???"
5658 * If dev is non-zero it must be a device number with a MAJOR of
5659 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5660 * the device is being created by opening a node in /dev.
5661 * If "name" is not NULL, the device is being created by
5662 * writing to /sys/module/md_mod/parameters/new_array.
5664 static DEFINE_MUTEX(disks_mutex);
5665 struct mddev *mddev = mddev_find_or_alloc(dev);
5666 struct gendisk *disk;
5675 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5676 shift = partitioned ? MdpMinorShift : 0;
5677 unit = MINOR(mddev->unit) >> shift;
5679 /* wait for any previous instance of this device to be
5680 * completely removed (mddev_delayed_delete).
5682 flush_workqueue(md_misc_wq);
5684 mutex_lock(&disks_mutex);
5690 /* Need to ensure that 'name' is not a duplicate.
5692 struct mddev *mddev2;
5693 spin_lock(&all_mddevs_lock);
5695 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5696 if (mddev2->gendisk &&
5697 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5698 spin_unlock(&all_mddevs_lock);
5701 spin_unlock(&all_mddevs_lock);
5705 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5707 mddev->hold_active = UNTIL_STOP;
5710 mddev->queue = blk_alloc_queue(NUMA_NO_NODE);
5714 blk_set_stacking_limits(&mddev->queue->limits);
5716 disk = alloc_disk(1 << shift);
5718 blk_cleanup_queue(mddev->queue);
5719 mddev->queue = NULL;
5722 disk->major = MAJOR(mddev->unit);
5723 disk->first_minor = unit << shift;
5725 strcpy(disk->disk_name, name);
5726 else if (partitioned)
5727 sprintf(disk->disk_name, "md_d%d", unit);
5729 sprintf(disk->disk_name, "md%d", unit);
5730 disk->fops = &md_fops;
5731 disk->private_data = mddev;
5732 disk->queue = mddev->queue;
5733 blk_queue_write_cache(mddev->queue, true, true);
5734 /* Allow extended partitions. This makes the
5735 * 'mdp' device redundant, but we can't really
5738 disk->flags |= GENHD_FL_EXT_DEVT;
5739 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5740 mddev->gendisk = disk;
5743 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5745 /* This isn't possible, but as kobject_init_and_add is marked
5746 * __must_check, we must do something with the result
5748 pr_debug("md: cannot register %s/md - name in use\n",
5752 if (mddev->kobj.sd &&
5753 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5754 pr_debug("pointless warning\n");
5756 mutex_unlock(&disks_mutex);
5757 if (!error && mddev->kobj.sd) {
5758 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5759 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5760 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5766 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5769 md_alloc(dev, NULL);
5773 static int add_named_array(const char *val, const struct kernel_param *kp)
5776 * val must be "md_*" or "mdNNN".
5777 * For "md_*" we allocate an array with a large free minor number, and
5778 * set the name to val. val must not already be an active name.
5779 * For "mdNNN" we allocate an array with the minor number NNN
5780 * which must not already be in use.
5782 int len = strlen(val);
5783 char buf[DISK_NAME_LEN];
5784 unsigned long devnum;
5786 while (len && val[len-1] == '\n')
5788 if (len >= DISK_NAME_LEN)
5790 strlcpy(buf, val, len+1);
5791 if (strncmp(buf, "md_", 3) == 0)
5792 return md_alloc(0, buf);
5793 if (strncmp(buf, "md", 2) == 0 &&
5795 kstrtoul(buf+2, 10, &devnum) == 0 &&
5796 devnum <= MINORMASK)
5797 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5802 static void md_safemode_timeout(struct timer_list *t)
5804 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5806 mddev->safemode = 1;
5807 if (mddev->external)
5808 sysfs_notify_dirent_safe(mddev->sysfs_state);
5810 md_wakeup_thread(mddev->thread);
5813 static int start_dirty_degraded;
5815 int md_run(struct mddev *mddev)
5818 struct md_rdev *rdev;
5819 struct md_personality *pers;
5821 if (list_empty(&mddev->disks))
5822 /* cannot run an array with no devices.. */
5827 /* Cannot run until previous stop completes properly */
5828 if (mddev->sysfs_active)
5832 * Analyze all RAID superblock(s)
5834 if (!mddev->raid_disks) {
5835 if (!mddev->persistent)
5837 err = analyze_sbs(mddev);
5842 if (mddev->level != LEVEL_NONE)
5843 request_module("md-level-%d", mddev->level);
5844 else if (mddev->clevel[0])
5845 request_module("md-%s", mddev->clevel);
5848 * Drop all container device buffers, from now on
5849 * the only valid external interface is through the md
5852 mddev->has_superblocks = false;
5853 rdev_for_each(rdev, mddev) {
5854 if (test_bit(Faulty, &rdev->flags))
5856 sync_blockdev(rdev->bdev);
5857 invalidate_bdev(rdev->bdev);
5858 if (mddev->ro != 1 &&
5859 (bdev_read_only(rdev->bdev) ||
5860 bdev_read_only(rdev->meta_bdev))) {
5863 set_disk_ro(mddev->gendisk, 1);
5867 mddev->has_superblocks = true;
5869 /* perform some consistency tests on the device.
5870 * We don't want the data to overlap the metadata,
5871 * Internal Bitmap issues have been handled elsewhere.
5873 if (rdev->meta_bdev) {
5874 /* Nothing to check */;
5875 } else if (rdev->data_offset < rdev->sb_start) {
5876 if (mddev->dev_sectors &&
5877 rdev->data_offset + mddev->dev_sectors
5879 pr_warn("md: %s: data overlaps metadata\n",
5884 if (rdev->sb_start + rdev->sb_size/512
5885 > rdev->data_offset) {
5886 pr_warn("md: %s: metadata overlaps data\n",
5891 sysfs_notify_dirent_safe(rdev->sysfs_state);
5894 if (!bioset_initialized(&mddev->bio_set)) {
5895 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5899 if (!bioset_initialized(&mddev->sync_set)) {
5900 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5905 spin_lock(&pers_lock);
5906 pers = find_pers(mddev->level, mddev->clevel);
5907 if (!pers || !try_module_get(pers->owner)) {
5908 spin_unlock(&pers_lock);
5909 if (mddev->level != LEVEL_NONE)
5910 pr_warn("md: personality for level %d is not loaded!\n",
5913 pr_warn("md: personality for level %s is not loaded!\n",
5918 spin_unlock(&pers_lock);
5919 if (mddev->level != pers->level) {
5920 mddev->level = pers->level;
5921 mddev->new_level = pers->level;
5923 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5925 if (mddev->reshape_position != MaxSector &&
5926 pers->start_reshape == NULL) {
5927 /* This personality cannot handle reshaping... */
5928 module_put(pers->owner);
5933 if (pers->sync_request) {
5934 /* Warn if this is a potentially silly
5937 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5938 struct md_rdev *rdev2;
5941 rdev_for_each(rdev, mddev)
5942 rdev_for_each(rdev2, mddev) {
5944 rdev->bdev->bd_disk ==
5945 rdev2->bdev->bd_disk) {
5946 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5948 bdevname(rdev->bdev,b),
5949 bdevname(rdev2->bdev,b2));
5955 pr_warn("True protection against single-disk failure might be compromised.\n");
5958 mddev->recovery = 0;
5959 /* may be over-ridden by personality */
5960 mddev->resync_max_sectors = mddev->dev_sectors;
5962 mddev->ok_start_degraded = start_dirty_degraded;
5964 if (start_readonly && mddev->ro == 0)
5965 mddev->ro = 2; /* read-only, but switch on first write */
5967 err = pers->run(mddev);
5969 pr_warn("md: pers->run() failed ...\n");
5970 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5971 WARN_ONCE(!mddev->external_size,
5972 "%s: default size too small, but 'external_size' not in effect?\n",
5974 pr_warn("md: invalid array_size %llu > default size %llu\n",
5975 (unsigned long long)mddev->array_sectors / 2,
5976 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5979 if (err == 0 && pers->sync_request &&
5980 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5981 struct bitmap *bitmap;
5983 bitmap = md_bitmap_create(mddev, -1);
5984 if (IS_ERR(bitmap)) {
5985 err = PTR_ERR(bitmap);
5986 pr_warn("%s: failed to create bitmap (%d)\n",
5987 mdname(mddev), err);
5989 mddev->bitmap = bitmap;
5995 if (mddev->bitmap_info.max_write_behind > 0) {
5996 bool create_pool = false;
5998 rdev_for_each(rdev, mddev) {
5999 if (test_bit(WriteMostly, &rdev->flags) &&
6000 rdev_init_serial(rdev))
6003 if (create_pool && mddev->serial_info_pool == NULL) {
6004 mddev->serial_info_pool =
6005 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6006 sizeof(struct serial_info));
6007 if (!mddev->serial_info_pool) {
6017 rdev_for_each(rdev, mddev) {
6018 if (rdev->raid_disk >= 0 &&
6019 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6024 if (mddev->degraded)
6027 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6029 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6031 if (pers->sync_request) {
6032 if (mddev->kobj.sd &&
6033 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6034 pr_warn("md: cannot register extra attributes for %s\n",
6036 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6037 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6038 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6039 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6042 atomic_set(&mddev->max_corr_read_errors,
6043 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6044 mddev->safemode = 0;
6045 if (mddev_is_clustered(mddev))
6046 mddev->safemode_delay = 0;
6048 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6051 spin_lock(&mddev->lock);
6053 spin_unlock(&mddev->lock);
6054 rdev_for_each(rdev, mddev)
6055 if (rdev->raid_disk >= 0)
6056 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6058 if (mddev->degraded && !mddev->ro)
6059 /* This ensures that recovering status is reported immediately
6060 * via sysfs - until a lack of spares is confirmed.
6062 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6063 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6065 if (mddev->sb_flags)
6066 md_update_sb(mddev, 0);
6068 md_new_event(mddev);
6072 mddev_detach(mddev);
6074 pers->free(mddev, mddev->private);
6075 mddev->private = NULL;
6076 module_put(pers->owner);
6077 md_bitmap_destroy(mddev);
6079 bioset_exit(&mddev->bio_set);
6080 bioset_exit(&mddev->sync_set);
6083 EXPORT_SYMBOL_GPL(md_run);
6085 int do_md_run(struct mddev *mddev)
6089 set_bit(MD_NOT_READY, &mddev->flags);
6090 err = md_run(mddev);
6093 err = md_bitmap_load(mddev);
6095 md_bitmap_destroy(mddev);
6099 if (mddev_is_clustered(mddev))
6100 md_allow_write(mddev);
6102 /* run start up tasks that require md_thread */
6105 md_wakeup_thread(mddev->thread);
6106 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6108 set_capacity(mddev->gendisk, mddev->array_sectors);
6109 revalidate_disk_size(mddev->gendisk, true);
6110 clear_bit(MD_NOT_READY, &mddev->flags);
6112 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6113 sysfs_notify_dirent_safe(mddev->sysfs_state);
6114 sysfs_notify_dirent_safe(mddev->sysfs_action);
6115 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6117 clear_bit(MD_NOT_READY, &mddev->flags);
6121 int md_start(struct mddev *mddev)
6125 if (mddev->pers->start) {
6126 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6127 md_wakeup_thread(mddev->thread);
6128 ret = mddev->pers->start(mddev);
6129 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6130 md_wakeup_thread(mddev->sync_thread);
6134 EXPORT_SYMBOL_GPL(md_start);
6136 static int restart_array(struct mddev *mddev)
6138 struct gendisk *disk = mddev->gendisk;
6139 struct md_rdev *rdev;
6140 bool has_journal = false;
6141 bool has_readonly = false;
6143 /* Complain if it has no devices */
6144 if (list_empty(&mddev->disks))
6152 rdev_for_each_rcu(rdev, mddev) {
6153 if (test_bit(Journal, &rdev->flags) &&
6154 !test_bit(Faulty, &rdev->flags))
6156 if (bdev_read_only(rdev->bdev))
6157 has_readonly = true;
6160 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6161 /* Don't restart rw with journal missing/faulty */
6166 mddev->safemode = 0;
6168 set_disk_ro(disk, 0);
6169 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6170 /* Kick recovery or resync if necessary */
6171 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6172 md_wakeup_thread(mddev->thread);
6173 md_wakeup_thread(mddev->sync_thread);
6174 sysfs_notify_dirent_safe(mddev->sysfs_state);
6178 static void md_clean(struct mddev *mddev)
6180 mddev->array_sectors = 0;
6181 mddev->external_size = 0;
6182 mddev->dev_sectors = 0;
6183 mddev->raid_disks = 0;
6184 mddev->recovery_cp = 0;
6185 mddev->resync_min = 0;
6186 mddev->resync_max = MaxSector;
6187 mddev->reshape_position = MaxSector;
6188 mddev->external = 0;
6189 mddev->persistent = 0;
6190 mddev->level = LEVEL_NONE;
6191 mddev->clevel[0] = 0;
6193 mddev->sb_flags = 0;
6195 mddev->metadata_type[0] = 0;
6196 mddev->chunk_sectors = 0;
6197 mddev->ctime = mddev->utime = 0;
6199 mddev->max_disks = 0;
6201 mddev->can_decrease_events = 0;
6202 mddev->delta_disks = 0;
6203 mddev->reshape_backwards = 0;
6204 mddev->new_level = LEVEL_NONE;
6205 mddev->new_layout = 0;
6206 mddev->new_chunk_sectors = 0;
6207 mddev->curr_resync = 0;
6208 atomic64_set(&mddev->resync_mismatches, 0);
6209 mddev->suspend_lo = mddev->suspend_hi = 0;
6210 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6211 mddev->recovery = 0;
6214 mddev->degraded = 0;
6215 mddev->safemode = 0;
6216 mddev->private = NULL;
6217 mddev->cluster_info = NULL;
6218 mddev->bitmap_info.offset = 0;
6219 mddev->bitmap_info.default_offset = 0;
6220 mddev->bitmap_info.default_space = 0;
6221 mddev->bitmap_info.chunksize = 0;
6222 mddev->bitmap_info.daemon_sleep = 0;
6223 mddev->bitmap_info.max_write_behind = 0;
6224 mddev->bitmap_info.nodes = 0;
6227 static void __md_stop_writes(struct mddev *mddev)
6229 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6230 if (work_pending(&mddev->del_work))
6231 flush_workqueue(md_misc_wq);
6232 if (mddev->sync_thread) {
6233 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6234 md_reap_sync_thread(mddev);
6237 del_timer_sync(&mddev->safemode_timer);
6239 if (mddev->pers && mddev->pers->quiesce) {
6240 mddev->pers->quiesce(mddev, 1);
6241 mddev->pers->quiesce(mddev, 0);
6243 md_bitmap_flush(mddev);
6245 if (mddev->ro == 0 &&
6246 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6248 /* mark array as shutdown cleanly */
6249 if (!mddev_is_clustered(mddev))
6251 md_update_sb(mddev, 1);
6253 /* disable policy to guarantee rdevs free resources for serialization */
6254 mddev->serialize_policy = 0;
6255 mddev_destroy_serial_pool(mddev, NULL, true);
6258 void md_stop_writes(struct mddev *mddev)
6260 mddev_lock_nointr(mddev);
6261 __md_stop_writes(mddev);
6262 mddev_unlock(mddev);
6264 EXPORT_SYMBOL_GPL(md_stop_writes);
6266 static void mddev_detach(struct mddev *mddev)
6268 md_bitmap_wait_behind_writes(mddev);
6269 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6270 mddev->pers->quiesce(mddev, 1);
6271 mddev->pers->quiesce(mddev, 0);
6273 md_unregister_thread(&mddev->thread);
6275 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6278 static void __md_stop(struct mddev *mddev)
6280 struct md_personality *pers = mddev->pers;
6281 md_bitmap_destroy(mddev);
6282 mddev_detach(mddev);
6283 /* Ensure ->event_work is done */
6284 if (mddev->event_work.func)
6285 flush_workqueue(md_misc_wq);
6286 spin_lock(&mddev->lock);
6288 spin_unlock(&mddev->lock);
6289 pers->free(mddev, mddev->private);
6290 mddev->private = NULL;
6291 if (pers->sync_request && mddev->to_remove == NULL)
6292 mddev->to_remove = &md_redundancy_group;
6293 module_put(pers->owner);
6294 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6297 void md_stop(struct mddev *mddev)
6299 /* stop the array and free an attached data structures.
6300 * This is called from dm-raid
6302 __md_stop_writes(mddev);
6304 bioset_exit(&mddev->bio_set);
6305 bioset_exit(&mddev->sync_set);
6308 EXPORT_SYMBOL_GPL(md_stop);
6310 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6315 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6318 md_wakeup_thread(mddev->thread);
6320 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6321 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6322 if (mddev->sync_thread)
6323 /* Thread might be blocked waiting for metadata update
6324 * which will now never happen */
6325 wake_up_process(mddev->sync_thread->tsk);
6327 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6329 mddev_unlock(mddev);
6330 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6332 wait_event(mddev->sb_wait,
6333 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6334 mddev_lock_nointr(mddev);
6336 mutex_lock(&mddev->open_mutex);
6337 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6338 mddev->sync_thread ||
6339 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6340 pr_warn("md: %s still in use.\n",mdname(mddev));
6342 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6343 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6344 md_wakeup_thread(mddev->thread);
6350 __md_stop_writes(mddev);
6356 set_disk_ro(mddev->gendisk, 1);
6357 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6358 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6359 md_wakeup_thread(mddev->thread);
6360 sysfs_notify_dirent_safe(mddev->sysfs_state);
6364 mutex_unlock(&mddev->open_mutex);
6369 * 0 - completely stop and dis-assemble array
6370 * 2 - stop but do not disassemble array
6372 static int do_md_stop(struct mddev *mddev, int mode,
6373 struct block_device *bdev)
6375 struct gendisk *disk = mddev->gendisk;
6376 struct md_rdev *rdev;
6379 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6381 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6382 md_wakeup_thread(mddev->thread);
6384 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6385 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6386 if (mddev->sync_thread)
6387 /* Thread might be blocked waiting for metadata update
6388 * which will now never happen */
6389 wake_up_process(mddev->sync_thread->tsk);
6391 mddev_unlock(mddev);
6392 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6393 !test_bit(MD_RECOVERY_RUNNING,
6394 &mddev->recovery)));
6395 mddev_lock_nointr(mddev);
6397 mutex_lock(&mddev->open_mutex);
6398 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6399 mddev->sysfs_active ||
6400 mddev->sync_thread ||
6401 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6402 pr_warn("md: %s still in use.\n",mdname(mddev));
6403 mutex_unlock(&mddev->open_mutex);
6405 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6407 md_wakeup_thread(mddev->thread);
6413 set_disk_ro(disk, 0);
6415 __md_stop_writes(mddev);
6418 /* tell userspace to handle 'inactive' */
6419 sysfs_notify_dirent_safe(mddev->sysfs_state);
6421 rdev_for_each(rdev, mddev)
6422 if (rdev->raid_disk >= 0)
6423 sysfs_unlink_rdev(mddev, rdev);
6425 set_capacity(disk, 0);
6426 mutex_unlock(&mddev->open_mutex);
6428 revalidate_disk_size(disk, true);
6433 mutex_unlock(&mddev->open_mutex);
6435 * Free resources if final stop
6438 pr_info("md: %s stopped.\n", mdname(mddev));
6440 if (mddev->bitmap_info.file) {
6441 struct file *f = mddev->bitmap_info.file;
6442 spin_lock(&mddev->lock);
6443 mddev->bitmap_info.file = NULL;
6444 spin_unlock(&mddev->lock);
6447 mddev->bitmap_info.offset = 0;
6449 export_array(mddev);
6452 if (mddev->hold_active == UNTIL_STOP)
6453 mddev->hold_active = 0;
6455 md_new_event(mddev);
6456 sysfs_notify_dirent_safe(mddev->sysfs_state);
6461 static void autorun_array(struct mddev *mddev)
6463 struct md_rdev *rdev;
6466 if (list_empty(&mddev->disks))
6469 pr_info("md: running: ");
6471 rdev_for_each(rdev, mddev) {
6472 char b[BDEVNAME_SIZE];
6473 pr_cont("<%s>", bdevname(rdev->bdev,b));
6477 err = do_md_run(mddev);
6479 pr_warn("md: do_md_run() returned %d\n", err);
6480 do_md_stop(mddev, 0, NULL);
6485 * lets try to run arrays based on all disks that have arrived
6486 * until now. (those are in pending_raid_disks)
6488 * the method: pick the first pending disk, collect all disks with
6489 * the same UUID, remove all from the pending list and put them into
6490 * the 'same_array' list. Then order this list based on superblock
6491 * update time (freshest comes first), kick out 'old' disks and
6492 * compare superblocks. If everything's fine then run it.
6494 * If "unit" is allocated, then bump its reference count
6496 static void autorun_devices(int part)
6498 struct md_rdev *rdev0, *rdev, *tmp;
6499 struct mddev *mddev;
6500 char b[BDEVNAME_SIZE];
6502 pr_info("md: autorun ...\n");
6503 while (!list_empty(&pending_raid_disks)) {
6506 LIST_HEAD(candidates);
6507 rdev0 = list_entry(pending_raid_disks.next,
6508 struct md_rdev, same_set);
6510 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6511 INIT_LIST_HEAD(&candidates);
6512 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6513 if (super_90_load(rdev, rdev0, 0) >= 0) {
6514 pr_debug("md: adding %s ...\n",
6515 bdevname(rdev->bdev,b));
6516 list_move(&rdev->same_set, &candidates);
6519 * now we have a set of devices, with all of them having
6520 * mostly sane superblocks. It's time to allocate the
6524 dev = MKDEV(mdp_major,
6525 rdev0->preferred_minor << MdpMinorShift);
6526 unit = MINOR(dev) >> MdpMinorShift;
6528 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6531 if (rdev0->preferred_minor != unit) {
6532 pr_warn("md: unit number in %s is bad: %d\n",
6533 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6537 md_probe(dev, NULL, NULL);
6538 mddev = mddev_find(dev);
6542 if (mddev_lock(mddev))
6543 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6544 else if (mddev->raid_disks || mddev->major_version
6545 || !list_empty(&mddev->disks)) {
6546 pr_warn("md: %s already running, cannot run %s\n",
6547 mdname(mddev), bdevname(rdev0->bdev,b));
6548 mddev_unlock(mddev);
6550 pr_debug("md: created %s\n", mdname(mddev));
6551 mddev->persistent = 1;
6552 rdev_for_each_list(rdev, tmp, &candidates) {
6553 list_del_init(&rdev->same_set);
6554 if (bind_rdev_to_array(rdev, mddev))
6557 autorun_array(mddev);
6558 mddev_unlock(mddev);
6560 /* on success, candidates will be empty, on error
6563 rdev_for_each_list(rdev, tmp, &candidates) {
6564 list_del_init(&rdev->same_set);
6569 pr_info("md: ... autorun DONE.\n");
6571 #endif /* !MODULE */
6573 static int get_version(void __user *arg)
6577 ver.major = MD_MAJOR_VERSION;
6578 ver.minor = MD_MINOR_VERSION;
6579 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6581 if (copy_to_user(arg, &ver, sizeof(ver)))
6587 static int get_array_info(struct mddev *mddev, void __user *arg)
6589 mdu_array_info_t info;
6590 int nr,working,insync,failed,spare;
6591 struct md_rdev *rdev;
6593 nr = working = insync = failed = spare = 0;
6595 rdev_for_each_rcu(rdev, mddev) {
6597 if (test_bit(Faulty, &rdev->flags))
6601 if (test_bit(In_sync, &rdev->flags))
6603 else if (test_bit(Journal, &rdev->flags))
6604 /* TODO: add journal count to md_u.h */
6612 info.major_version = mddev->major_version;
6613 info.minor_version = mddev->minor_version;
6614 info.patch_version = MD_PATCHLEVEL_VERSION;
6615 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6616 info.level = mddev->level;
6617 info.size = mddev->dev_sectors / 2;
6618 if (info.size != mddev->dev_sectors / 2) /* overflow */
6621 info.raid_disks = mddev->raid_disks;
6622 info.md_minor = mddev->md_minor;
6623 info.not_persistent= !mddev->persistent;
6625 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6628 info.state = (1<<MD_SB_CLEAN);
6629 if (mddev->bitmap && mddev->bitmap_info.offset)
6630 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6631 if (mddev_is_clustered(mddev))
6632 info.state |= (1<<MD_SB_CLUSTERED);
6633 info.active_disks = insync;
6634 info.working_disks = working;
6635 info.failed_disks = failed;
6636 info.spare_disks = spare;
6638 info.layout = mddev->layout;
6639 info.chunk_size = mddev->chunk_sectors << 9;
6641 if (copy_to_user(arg, &info, sizeof(info)))
6647 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6649 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6653 file = kzalloc(sizeof(*file), GFP_NOIO);
6658 spin_lock(&mddev->lock);
6659 /* bitmap enabled */
6660 if (mddev->bitmap_info.file) {
6661 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6662 sizeof(file->pathname));
6666 memmove(file->pathname, ptr,
6667 sizeof(file->pathname)-(ptr-file->pathname));
6669 spin_unlock(&mddev->lock);
6672 copy_to_user(arg, file, sizeof(*file)))
6679 static int get_disk_info(struct mddev *mddev, void __user * arg)
6681 mdu_disk_info_t info;
6682 struct md_rdev *rdev;
6684 if (copy_from_user(&info, arg, sizeof(info)))
6688 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6690 info.major = MAJOR(rdev->bdev->bd_dev);
6691 info.minor = MINOR(rdev->bdev->bd_dev);
6692 info.raid_disk = rdev->raid_disk;
6694 if (test_bit(Faulty, &rdev->flags))
6695 info.state |= (1<<MD_DISK_FAULTY);
6696 else if (test_bit(In_sync, &rdev->flags)) {
6697 info.state |= (1<<MD_DISK_ACTIVE);
6698 info.state |= (1<<MD_DISK_SYNC);
6700 if (test_bit(Journal, &rdev->flags))
6701 info.state |= (1<<MD_DISK_JOURNAL);
6702 if (test_bit(WriteMostly, &rdev->flags))
6703 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6704 if (test_bit(FailFast, &rdev->flags))
6705 info.state |= (1<<MD_DISK_FAILFAST);
6707 info.major = info.minor = 0;
6708 info.raid_disk = -1;
6709 info.state = (1<<MD_DISK_REMOVED);
6713 if (copy_to_user(arg, &info, sizeof(info)))
6719 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6721 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6722 struct md_rdev *rdev;
6723 dev_t dev = MKDEV(info->major,info->minor);
6725 if (mddev_is_clustered(mddev) &&
6726 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6727 pr_warn("%s: Cannot add to clustered mddev.\n",
6732 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6735 if (!mddev->raid_disks) {
6737 /* expecting a device which has a superblock */
6738 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6740 pr_warn("md: md_import_device returned %ld\n",
6742 return PTR_ERR(rdev);
6744 if (!list_empty(&mddev->disks)) {
6745 struct md_rdev *rdev0
6746 = list_entry(mddev->disks.next,
6747 struct md_rdev, same_set);
6748 err = super_types[mddev->major_version]
6749 .load_super(rdev, rdev0, mddev->minor_version);
6751 pr_warn("md: %s has different UUID to %s\n",
6752 bdevname(rdev->bdev,b),
6753 bdevname(rdev0->bdev,b2));
6758 err = bind_rdev_to_array(rdev, mddev);
6765 * md_add_new_disk can be used once the array is assembled
6766 * to add "hot spares". They must already have a superblock
6771 if (!mddev->pers->hot_add_disk) {
6772 pr_warn("%s: personality does not support diskops!\n",
6776 if (mddev->persistent)
6777 rdev = md_import_device(dev, mddev->major_version,
6778 mddev->minor_version);
6780 rdev = md_import_device(dev, -1, -1);
6782 pr_warn("md: md_import_device returned %ld\n",
6784 return PTR_ERR(rdev);
6786 /* set saved_raid_disk if appropriate */
6787 if (!mddev->persistent) {
6788 if (info->state & (1<<MD_DISK_SYNC) &&
6789 info->raid_disk < mddev->raid_disks) {
6790 rdev->raid_disk = info->raid_disk;
6791 set_bit(In_sync, &rdev->flags);
6792 clear_bit(Bitmap_sync, &rdev->flags);
6794 rdev->raid_disk = -1;
6795 rdev->saved_raid_disk = rdev->raid_disk;
6797 super_types[mddev->major_version].
6798 validate_super(mddev, rdev);
6799 if ((info->state & (1<<MD_DISK_SYNC)) &&
6800 rdev->raid_disk != info->raid_disk) {
6801 /* This was a hot-add request, but events doesn't
6802 * match, so reject it.
6808 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6809 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6810 set_bit(WriteMostly, &rdev->flags);
6812 clear_bit(WriteMostly, &rdev->flags);
6813 if (info->state & (1<<MD_DISK_FAILFAST))
6814 set_bit(FailFast, &rdev->flags);
6816 clear_bit(FailFast, &rdev->flags);
6818 if (info->state & (1<<MD_DISK_JOURNAL)) {
6819 struct md_rdev *rdev2;
6820 bool has_journal = false;
6822 /* make sure no existing journal disk */
6823 rdev_for_each(rdev2, mddev) {
6824 if (test_bit(Journal, &rdev2->flags)) {
6829 if (has_journal || mddev->bitmap) {
6833 set_bit(Journal, &rdev->flags);
6836 * check whether the device shows up in other nodes
6838 if (mddev_is_clustered(mddev)) {
6839 if (info->state & (1 << MD_DISK_CANDIDATE))
6840 set_bit(Candidate, &rdev->flags);
6841 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6842 /* --add initiated by this node */
6843 err = md_cluster_ops->add_new_disk(mddev, rdev);
6851 rdev->raid_disk = -1;
6852 err = bind_rdev_to_array(rdev, mddev);
6857 if (mddev_is_clustered(mddev)) {
6858 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6860 err = md_cluster_ops->new_disk_ack(mddev,
6863 md_kick_rdev_from_array(rdev);
6867 md_cluster_ops->add_new_disk_cancel(mddev);
6869 err = add_bound_rdev(rdev);
6873 err = add_bound_rdev(rdev);
6878 /* otherwise, md_add_new_disk is only allowed
6879 * for major_version==0 superblocks
6881 if (mddev->major_version != 0) {
6882 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6886 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6888 rdev = md_import_device(dev, -1, 0);
6890 pr_warn("md: error, md_import_device() returned %ld\n",
6892 return PTR_ERR(rdev);
6894 rdev->desc_nr = info->number;
6895 if (info->raid_disk < mddev->raid_disks)
6896 rdev->raid_disk = info->raid_disk;
6898 rdev->raid_disk = -1;
6900 if (rdev->raid_disk < mddev->raid_disks)
6901 if (info->state & (1<<MD_DISK_SYNC))
6902 set_bit(In_sync, &rdev->flags);
6904 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6905 set_bit(WriteMostly, &rdev->flags);
6906 if (info->state & (1<<MD_DISK_FAILFAST))
6907 set_bit(FailFast, &rdev->flags);
6909 if (!mddev->persistent) {
6910 pr_debug("md: nonpersistent superblock ...\n");
6911 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6913 rdev->sb_start = calc_dev_sboffset(rdev);
6914 rdev->sectors = rdev->sb_start;
6916 err = bind_rdev_to_array(rdev, mddev);
6926 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6928 char b[BDEVNAME_SIZE];
6929 struct md_rdev *rdev;
6934 rdev = find_rdev(mddev, dev);
6938 if (rdev->raid_disk < 0)
6941 clear_bit(Blocked, &rdev->flags);
6942 remove_and_add_spares(mddev, rdev);
6944 if (rdev->raid_disk >= 0)
6948 if (mddev_is_clustered(mddev)) {
6949 if (md_cluster_ops->remove_disk(mddev, rdev))
6953 md_kick_rdev_from_array(rdev);
6954 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6956 md_wakeup_thread(mddev->thread);
6958 md_update_sb(mddev, 1);
6959 md_new_event(mddev);
6963 pr_debug("md: cannot remove active disk %s from %s ...\n",
6964 bdevname(rdev->bdev,b), mdname(mddev));
6968 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6970 char b[BDEVNAME_SIZE];
6972 struct md_rdev *rdev;
6977 if (mddev->major_version != 0) {
6978 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6982 if (!mddev->pers->hot_add_disk) {
6983 pr_warn("%s: personality does not support diskops!\n",
6988 rdev = md_import_device(dev, -1, 0);
6990 pr_warn("md: error, md_import_device() returned %ld\n",
6995 if (mddev->persistent)
6996 rdev->sb_start = calc_dev_sboffset(rdev);
6998 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
7000 rdev->sectors = rdev->sb_start;
7002 if (test_bit(Faulty, &rdev->flags)) {
7003 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
7004 bdevname(rdev->bdev,b), mdname(mddev));
7009 clear_bit(In_sync, &rdev->flags);
7011 rdev->saved_raid_disk = -1;
7012 err = bind_rdev_to_array(rdev, mddev);
7017 * The rest should better be atomic, we can have disk failures
7018 * noticed in interrupt contexts ...
7021 rdev->raid_disk = -1;
7023 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7025 md_update_sb(mddev, 1);
7027 * Kick recovery, maybe this spare has to be added to the
7028 * array immediately.
7030 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7031 md_wakeup_thread(mddev->thread);
7032 md_new_event(mddev);
7040 static int set_bitmap_file(struct mddev *mddev, int fd)
7045 if (!mddev->pers->quiesce || !mddev->thread)
7047 if (mddev->recovery || mddev->sync_thread)
7049 /* we should be able to change the bitmap.. */
7053 struct inode *inode;
7056 if (mddev->bitmap || mddev->bitmap_info.file)
7057 return -EEXIST; /* cannot add when bitmap is present */
7061 pr_warn("%s: error: failed to get bitmap file\n",
7066 inode = f->f_mapping->host;
7067 if (!S_ISREG(inode->i_mode)) {
7068 pr_warn("%s: error: bitmap file must be a regular file\n",
7071 } else if (!(f->f_mode & FMODE_WRITE)) {
7072 pr_warn("%s: error: bitmap file must open for write\n",
7075 } else if (atomic_read(&inode->i_writecount) != 1) {
7076 pr_warn("%s: error: bitmap file is already in use\n",
7084 mddev->bitmap_info.file = f;
7085 mddev->bitmap_info.offset = 0; /* file overrides offset */
7086 } else if (mddev->bitmap == NULL)
7087 return -ENOENT; /* cannot remove what isn't there */
7091 struct bitmap *bitmap;
7093 bitmap = md_bitmap_create(mddev, -1);
7094 mddev_suspend(mddev);
7095 if (!IS_ERR(bitmap)) {
7096 mddev->bitmap = bitmap;
7097 err = md_bitmap_load(mddev);
7099 err = PTR_ERR(bitmap);
7101 md_bitmap_destroy(mddev);
7104 mddev_resume(mddev);
7105 } else if (fd < 0) {
7106 mddev_suspend(mddev);
7107 md_bitmap_destroy(mddev);
7108 mddev_resume(mddev);
7112 struct file *f = mddev->bitmap_info.file;
7114 spin_lock(&mddev->lock);
7115 mddev->bitmap_info.file = NULL;
7116 spin_unlock(&mddev->lock);
7125 * md_set_array_info is used two different ways
7126 * The original usage is when creating a new array.
7127 * In this usage, raid_disks is > 0 and it together with
7128 * level, size, not_persistent,layout,chunksize determine the
7129 * shape of the array.
7130 * This will always create an array with a type-0.90.0 superblock.
7131 * The newer usage is when assembling an array.
7132 * In this case raid_disks will be 0, and the major_version field is
7133 * use to determine which style super-blocks are to be found on the devices.
7134 * The minor and patch _version numbers are also kept incase the
7135 * super_block handler wishes to interpret them.
7137 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7139 if (info->raid_disks == 0) {
7140 /* just setting version number for superblock loading */
7141 if (info->major_version < 0 ||
7142 info->major_version >= ARRAY_SIZE(super_types) ||
7143 super_types[info->major_version].name == NULL) {
7144 /* maybe try to auto-load a module? */
7145 pr_warn("md: superblock version %d not known\n",
7146 info->major_version);
7149 mddev->major_version = info->major_version;
7150 mddev->minor_version = info->minor_version;
7151 mddev->patch_version = info->patch_version;
7152 mddev->persistent = !info->not_persistent;
7153 /* ensure mddev_put doesn't delete this now that there
7154 * is some minimal configuration.
7156 mddev->ctime = ktime_get_real_seconds();
7159 mddev->major_version = MD_MAJOR_VERSION;
7160 mddev->minor_version = MD_MINOR_VERSION;
7161 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7162 mddev->ctime = ktime_get_real_seconds();
7164 mddev->level = info->level;
7165 mddev->clevel[0] = 0;
7166 mddev->dev_sectors = 2 * (sector_t)info->size;
7167 mddev->raid_disks = info->raid_disks;
7168 /* don't set md_minor, it is determined by which /dev/md* was
7171 if (info->state & (1<<MD_SB_CLEAN))
7172 mddev->recovery_cp = MaxSector;
7174 mddev->recovery_cp = 0;
7175 mddev->persistent = ! info->not_persistent;
7176 mddev->external = 0;
7178 mddev->layout = info->layout;
7179 if (mddev->level == 0)
7180 /* Cannot trust RAID0 layout info here */
7182 mddev->chunk_sectors = info->chunk_size >> 9;
7184 if (mddev->persistent) {
7185 mddev->max_disks = MD_SB_DISKS;
7187 mddev->sb_flags = 0;
7189 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7191 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7192 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7193 mddev->bitmap_info.offset = 0;
7195 mddev->reshape_position = MaxSector;
7198 * Generate a 128 bit UUID
7200 get_random_bytes(mddev->uuid, 16);
7202 mddev->new_level = mddev->level;
7203 mddev->new_chunk_sectors = mddev->chunk_sectors;
7204 mddev->new_layout = mddev->layout;
7205 mddev->delta_disks = 0;
7206 mddev->reshape_backwards = 0;
7211 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7213 lockdep_assert_held(&mddev->reconfig_mutex);
7215 if (mddev->external_size)
7218 mddev->array_sectors = array_sectors;
7220 EXPORT_SYMBOL(md_set_array_sectors);
7222 static int update_size(struct mddev *mddev, sector_t num_sectors)
7224 struct md_rdev *rdev;
7226 int fit = (num_sectors == 0);
7227 sector_t old_dev_sectors = mddev->dev_sectors;
7229 if (mddev->pers->resize == NULL)
7231 /* The "num_sectors" is the number of sectors of each device that
7232 * is used. This can only make sense for arrays with redundancy.
7233 * linear and raid0 always use whatever space is available. We can only
7234 * consider changing this number if no resync or reconstruction is
7235 * happening, and if the new size is acceptable. It must fit before the
7236 * sb_start or, if that is <data_offset, it must fit before the size
7237 * of each device. If num_sectors is zero, we find the largest size
7240 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7246 rdev_for_each(rdev, mddev) {
7247 sector_t avail = rdev->sectors;
7249 if (fit && (num_sectors == 0 || num_sectors > avail))
7250 num_sectors = avail;
7251 if (avail < num_sectors)
7254 rv = mddev->pers->resize(mddev, num_sectors);
7256 if (mddev_is_clustered(mddev))
7257 md_cluster_ops->update_size(mddev, old_dev_sectors);
7258 else if (mddev->queue) {
7259 set_capacity(mddev->gendisk, mddev->array_sectors);
7260 revalidate_disk_size(mddev->gendisk, true);
7266 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7269 struct md_rdev *rdev;
7270 /* change the number of raid disks */
7271 if (mddev->pers->check_reshape == NULL)
7275 if (raid_disks <= 0 ||
7276 (mddev->max_disks && raid_disks >= mddev->max_disks))
7278 if (mddev->sync_thread ||
7279 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7280 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7281 mddev->reshape_position != MaxSector)
7284 rdev_for_each(rdev, mddev) {
7285 if (mddev->raid_disks < raid_disks &&
7286 rdev->data_offset < rdev->new_data_offset)
7288 if (mddev->raid_disks > raid_disks &&
7289 rdev->data_offset > rdev->new_data_offset)
7293 mddev->delta_disks = raid_disks - mddev->raid_disks;
7294 if (mddev->delta_disks < 0)
7295 mddev->reshape_backwards = 1;
7296 else if (mddev->delta_disks > 0)
7297 mddev->reshape_backwards = 0;
7299 rv = mddev->pers->check_reshape(mddev);
7301 mddev->delta_disks = 0;
7302 mddev->reshape_backwards = 0;
7308 * update_array_info is used to change the configuration of an
7310 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7311 * fields in the info are checked against the array.
7312 * Any differences that cannot be handled will cause an error.
7313 * Normally, only one change can be managed at a time.
7315 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7321 /* calculate expected state,ignoring low bits */
7322 if (mddev->bitmap && mddev->bitmap_info.offset)
7323 state |= (1 << MD_SB_BITMAP_PRESENT);
7325 if (mddev->major_version != info->major_version ||
7326 mddev->minor_version != info->minor_version ||
7327 /* mddev->patch_version != info->patch_version || */
7328 mddev->ctime != info->ctime ||
7329 mddev->level != info->level ||
7330 /* mddev->layout != info->layout || */
7331 mddev->persistent != !info->not_persistent ||
7332 mddev->chunk_sectors != info->chunk_size >> 9 ||
7333 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7334 ((state^info->state) & 0xfffffe00)
7337 /* Check there is only one change */
7338 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7340 if (mddev->raid_disks != info->raid_disks)
7342 if (mddev->layout != info->layout)
7344 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7351 if (mddev->layout != info->layout) {
7353 * we don't need to do anything at the md level, the
7354 * personality will take care of it all.
7356 if (mddev->pers->check_reshape == NULL)
7359 mddev->new_layout = info->layout;
7360 rv = mddev->pers->check_reshape(mddev);
7362 mddev->new_layout = mddev->layout;
7366 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7367 rv = update_size(mddev, (sector_t)info->size * 2);
7369 if (mddev->raid_disks != info->raid_disks)
7370 rv = update_raid_disks(mddev, info->raid_disks);
7372 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7373 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7377 if (mddev->recovery || mddev->sync_thread) {
7381 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7382 struct bitmap *bitmap;
7383 /* add the bitmap */
7384 if (mddev->bitmap) {
7388 if (mddev->bitmap_info.default_offset == 0) {
7392 mddev->bitmap_info.offset =
7393 mddev->bitmap_info.default_offset;
7394 mddev->bitmap_info.space =
7395 mddev->bitmap_info.default_space;
7396 bitmap = md_bitmap_create(mddev, -1);
7397 mddev_suspend(mddev);
7398 if (!IS_ERR(bitmap)) {
7399 mddev->bitmap = bitmap;
7400 rv = md_bitmap_load(mddev);
7402 rv = PTR_ERR(bitmap);
7404 md_bitmap_destroy(mddev);
7405 mddev_resume(mddev);
7407 /* remove the bitmap */
7408 if (!mddev->bitmap) {
7412 if (mddev->bitmap->storage.file) {
7416 if (mddev->bitmap_info.nodes) {
7417 /* hold PW on all the bitmap lock */
7418 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7419 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7421 md_cluster_ops->unlock_all_bitmaps(mddev);
7425 mddev->bitmap_info.nodes = 0;
7426 md_cluster_ops->leave(mddev);
7427 module_put(md_cluster_mod);
7428 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7430 mddev_suspend(mddev);
7431 md_bitmap_destroy(mddev);
7432 mddev_resume(mddev);
7433 mddev->bitmap_info.offset = 0;
7436 md_update_sb(mddev, 1);
7442 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7444 struct md_rdev *rdev;
7447 if (mddev->pers == NULL)
7451 rdev = md_find_rdev_rcu(mddev, dev);
7455 md_error(mddev, rdev);
7456 if (!test_bit(Faulty, &rdev->flags))
7464 * We have a problem here : there is no easy way to give a CHS
7465 * virtual geometry. We currently pretend that we have a 2 heads
7466 * 4 sectors (with a BIG number of cylinders...). This drives
7467 * dosfs just mad... ;-)
7469 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7471 struct mddev *mddev = bdev->bd_disk->private_data;
7475 geo->cylinders = mddev->array_sectors / 8;
7479 static inline bool md_ioctl_valid(unsigned int cmd)
7484 case GET_ARRAY_INFO:
7485 case GET_BITMAP_FILE:
7488 case HOT_REMOVE_DISK:
7490 case RESTART_ARRAY_RW:
7492 case SET_ARRAY_INFO:
7493 case SET_BITMAP_FILE:
7494 case SET_DISK_FAULTY:
7497 case CLUSTERED_DISK_NACK:
7504 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7505 unsigned int cmd, unsigned long arg)
7508 void __user *argp = (void __user *)arg;
7509 struct mddev *mddev = NULL;
7511 bool did_set_md_closing = false;
7513 if (!md_ioctl_valid(cmd))
7518 case GET_ARRAY_INFO:
7522 if (!capable(CAP_SYS_ADMIN))
7527 * Commands dealing with the RAID driver but not any
7532 err = get_version(argp);
7538 * Commands creating/starting a new array:
7541 mddev = bdev->bd_disk->private_data;
7548 /* Some actions do not requires the mutex */
7550 case GET_ARRAY_INFO:
7551 if (!mddev->raid_disks && !mddev->external)
7554 err = get_array_info(mddev, argp);
7558 if (!mddev->raid_disks && !mddev->external)
7561 err = get_disk_info(mddev, argp);
7564 case SET_DISK_FAULTY:
7565 err = set_disk_faulty(mddev, new_decode_dev(arg));
7568 case GET_BITMAP_FILE:
7569 err = get_bitmap_file(mddev, argp);
7574 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7575 flush_rdev_wq(mddev);
7577 if (cmd == HOT_REMOVE_DISK)
7578 /* need to ensure recovery thread has run */
7579 wait_event_interruptible_timeout(mddev->sb_wait,
7580 !test_bit(MD_RECOVERY_NEEDED,
7582 msecs_to_jiffies(5000));
7583 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7584 /* Need to flush page cache, and ensure no-one else opens
7587 mutex_lock(&mddev->open_mutex);
7588 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7589 mutex_unlock(&mddev->open_mutex);
7593 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7594 mutex_unlock(&mddev->open_mutex);
7598 did_set_md_closing = true;
7599 mutex_unlock(&mddev->open_mutex);
7600 sync_blockdev(bdev);
7602 err = mddev_lock(mddev);
7604 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7609 if (cmd == SET_ARRAY_INFO) {
7610 mdu_array_info_t info;
7612 memset(&info, 0, sizeof(info));
7613 else if (copy_from_user(&info, argp, sizeof(info))) {
7618 err = update_array_info(mddev, &info);
7620 pr_warn("md: couldn't update array info. %d\n", err);
7625 if (!list_empty(&mddev->disks)) {
7626 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7630 if (mddev->raid_disks) {
7631 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7635 err = md_set_array_info(mddev, &info);
7637 pr_warn("md: couldn't set array info. %d\n", err);
7644 * Commands querying/configuring an existing array:
7646 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7647 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7648 if ((!mddev->raid_disks && !mddev->external)
7649 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7650 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7651 && cmd != GET_BITMAP_FILE) {
7657 * Commands even a read-only array can execute:
7660 case RESTART_ARRAY_RW:
7661 err = restart_array(mddev);
7665 err = do_md_stop(mddev, 0, bdev);
7669 err = md_set_readonly(mddev, bdev);
7672 case HOT_REMOVE_DISK:
7673 err = hot_remove_disk(mddev, new_decode_dev(arg));
7677 /* We can support ADD_NEW_DISK on read-only arrays
7678 * only if we are re-adding a preexisting device.
7679 * So require mddev->pers and MD_DISK_SYNC.
7682 mdu_disk_info_t info;
7683 if (copy_from_user(&info, argp, sizeof(info)))
7685 else if (!(info.state & (1<<MD_DISK_SYNC)))
7686 /* Need to clear read-only for this */
7689 err = md_add_new_disk(mddev, &info);
7695 if (get_user(ro, (int __user *)(arg))) {
7701 /* if the bdev is going readonly the value of mddev->ro
7702 * does not matter, no writes are coming
7707 /* are we are already prepared for writes? */
7711 /* transitioning to readauto need only happen for
7712 * arrays that call md_write_start
7715 err = restart_array(mddev);
7718 set_disk_ro(mddev->gendisk, 0);
7725 * The remaining ioctls are changing the state of the
7726 * superblock, so we do not allow them on read-only arrays.
7728 if (mddev->ro && mddev->pers) {
7729 if (mddev->ro == 2) {
7731 sysfs_notify_dirent_safe(mddev->sysfs_state);
7732 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7733 /* mddev_unlock will wake thread */
7734 /* If a device failed while we were read-only, we
7735 * need to make sure the metadata is updated now.
7737 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7738 mddev_unlock(mddev);
7739 wait_event(mddev->sb_wait,
7740 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7741 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7742 mddev_lock_nointr(mddev);
7753 mdu_disk_info_t info;
7754 if (copy_from_user(&info, argp, sizeof(info)))
7757 err = md_add_new_disk(mddev, &info);
7761 case CLUSTERED_DISK_NACK:
7762 if (mddev_is_clustered(mddev))
7763 md_cluster_ops->new_disk_ack(mddev, false);
7769 err = hot_add_disk(mddev, new_decode_dev(arg));
7773 err = do_md_run(mddev);
7776 case SET_BITMAP_FILE:
7777 err = set_bitmap_file(mddev, (int)arg);
7786 if (mddev->hold_active == UNTIL_IOCTL &&
7788 mddev->hold_active = 0;
7789 mddev_unlock(mddev);
7791 if(did_set_md_closing)
7792 clear_bit(MD_CLOSING, &mddev->flags);
7795 #ifdef CONFIG_COMPAT
7796 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7797 unsigned int cmd, unsigned long arg)
7800 case HOT_REMOVE_DISK:
7802 case SET_DISK_FAULTY:
7803 case SET_BITMAP_FILE:
7804 /* These take in integer arg, do not convert */
7807 arg = (unsigned long)compat_ptr(arg);
7811 return md_ioctl(bdev, mode, cmd, arg);
7813 #endif /* CONFIG_COMPAT */
7815 static int md_open(struct block_device *bdev, fmode_t mode)
7818 * Succeed if we can lock the mddev, which confirms that
7819 * it isn't being stopped right now.
7821 struct mddev *mddev = mddev_find(bdev->bd_dev);
7827 if (mddev->gendisk != bdev->bd_disk) {
7828 /* we are racing with mddev_put which is discarding this
7832 /* Wait until bdev->bd_disk is definitely gone */
7833 if (work_pending(&mddev->del_work))
7834 flush_workqueue(md_misc_wq);
7837 BUG_ON(mddev != bdev->bd_disk->private_data);
7839 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7842 if (test_bit(MD_CLOSING, &mddev->flags)) {
7843 mutex_unlock(&mddev->open_mutex);
7849 atomic_inc(&mddev->openers);
7850 mutex_unlock(&mddev->open_mutex);
7852 bdev_check_media_change(bdev);
7859 static void md_release(struct gendisk *disk, fmode_t mode)
7861 struct mddev *mddev = disk->private_data;
7864 atomic_dec(&mddev->openers);
7868 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7870 struct mddev *mddev = disk->private_data;
7871 unsigned int ret = 0;
7874 ret = DISK_EVENT_MEDIA_CHANGE;
7879 const struct block_device_operations md_fops =
7881 .owner = THIS_MODULE,
7882 .submit_bio = md_submit_bio,
7884 .release = md_release,
7886 #ifdef CONFIG_COMPAT
7887 .compat_ioctl = md_compat_ioctl,
7889 .getgeo = md_getgeo,
7890 .check_events = md_check_events,
7893 static int md_thread(void *arg)
7895 struct md_thread *thread = arg;
7898 * md_thread is a 'system-thread', it's priority should be very
7899 * high. We avoid resource deadlocks individually in each
7900 * raid personality. (RAID5 does preallocation) We also use RR and
7901 * the very same RT priority as kswapd, thus we will never get
7902 * into a priority inversion deadlock.
7904 * we definitely have to have equal or higher priority than
7905 * bdflush, otherwise bdflush will deadlock if there are too
7906 * many dirty RAID5 blocks.
7909 allow_signal(SIGKILL);
7910 while (!kthread_should_stop()) {
7912 /* We need to wait INTERRUPTIBLE so that
7913 * we don't add to the load-average.
7914 * That means we need to be sure no signals are
7917 if (signal_pending(current))
7918 flush_signals(current);
7920 wait_event_interruptible_timeout
7922 test_bit(THREAD_WAKEUP, &thread->flags)
7923 || kthread_should_stop() || kthread_should_park(),
7926 clear_bit(THREAD_WAKEUP, &thread->flags);
7927 if (kthread_should_park())
7929 if (!kthread_should_stop())
7930 thread->run(thread);
7936 void md_wakeup_thread(struct md_thread *thread)
7939 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7940 set_bit(THREAD_WAKEUP, &thread->flags);
7941 wake_up(&thread->wqueue);
7944 EXPORT_SYMBOL(md_wakeup_thread);
7946 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7947 struct mddev *mddev, const char *name)
7949 struct md_thread *thread;
7951 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7955 init_waitqueue_head(&thread->wqueue);
7958 thread->mddev = mddev;
7959 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7960 thread->tsk = kthread_run(md_thread, thread,
7962 mdname(thread->mddev),
7964 if (IS_ERR(thread->tsk)) {
7970 EXPORT_SYMBOL(md_register_thread);
7972 void md_unregister_thread(struct md_thread **threadp)
7974 struct md_thread *thread;
7977 * Locking ensures that mddev_unlock does not wake_up a
7978 * non-existent thread
7980 spin_lock(&pers_lock);
7983 spin_unlock(&pers_lock);
7987 spin_unlock(&pers_lock);
7989 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7990 kthread_stop(thread->tsk);
7993 EXPORT_SYMBOL(md_unregister_thread);
7995 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7997 if (!rdev || test_bit(Faulty, &rdev->flags))
8000 if (!mddev->pers || !mddev->pers->error_handler)
8002 mddev->pers->error_handler(mddev,rdev);
8003 if (mddev->degraded)
8004 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8005 sysfs_notify_dirent_safe(rdev->sysfs_state);
8006 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8007 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8008 md_wakeup_thread(mddev->thread);
8009 if (mddev->event_work.func)
8010 queue_work(md_misc_wq, &mddev->event_work);
8011 md_new_event(mddev);
8013 EXPORT_SYMBOL(md_error);
8015 /* seq_file implementation /proc/mdstat */
8017 static void status_unused(struct seq_file *seq)
8020 struct md_rdev *rdev;
8022 seq_printf(seq, "unused devices: ");
8024 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8025 char b[BDEVNAME_SIZE];
8027 seq_printf(seq, "%s ",
8028 bdevname(rdev->bdev,b));
8031 seq_printf(seq, "<none>");
8033 seq_printf(seq, "\n");
8036 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8038 sector_t max_sectors, resync, res;
8039 unsigned long dt, db = 0;
8040 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8041 int scale, recovery_active;
8042 unsigned int per_milli;
8044 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8045 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8046 max_sectors = mddev->resync_max_sectors;
8048 max_sectors = mddev->dev_sectors;
8050 resync = mddev->curr_resync;
8052 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8053 /* Still cleaning up */
8054 resync = max_sectors;
8055 } else if (resync > max_sectors)
8056 resync = max_sectors;
8058 resync -= atomic_read(&mddev->recovery_active);
8061 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8062 struct md_rdev *rdev;
8064 rdev_for_each(rdev, mddev)
8065 if (rdev->raid_disk >= 0 &&
8066 !test_bit(Faulty, &rdev->flags) &&
8067 rdev->recovery_offset != MaxSector &&
8068 rdev->recovery_offset) {
8069 seq_printf(seq, "\trecover=REMOTE");
8072 if (mddev->reshape_position != MaxSector)
8073 seq_printf(seq, "\treshape=REMOTE");
8075 seq_printf(seq, "\tresync=REMOTE");
8078 if (mddev->recovery_cp < MaxSector) {
8079 seq_printf(seq, "\tresync=PENDING");
8085 seq_printf(seq, "\tresync=DELAYED");
8089 WARN_ON(max_sectors == 0);
8090 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8091 * in a sector_t, and (max_sectors>>scale) will fit in a
8092 * u32, as those are the requirements for sector_div.
8093 * Thus 'scale' must be at least 10
8096 if (sizeof(sector_t) > sizeof(unsigned long)) {
8097 while ( max_sectors/2 > (1ULL<<(scale+32)))
8100 res = (resync>>scale)*1000;
8101 sector_div(res, (u32)((max_sectors>>scale)+1));
8105 int i, x = per_milli/50, y = 20-x;
8106 seq_printf(seq, "[");
8107 for (i = 0; i < x; i++)
8108 seq_printf(seq, "=");
8109 seq_printf(seq, ">");
8110 for (i = 0; i < y; i++)
8111 seq_printf(seq, ".");
8112 seq_printf(seq, "] ");
8114 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8115 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8117 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8119 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8120 "resync" : "recovery"))),
8121 per_milli/10, per_milli % 10,
8122 (unsigned long long) resync/2,
8123 (unsigned long long) max_sectors/2);
8126 * dt: time from mark until now
8127 * db: blocks written from mark until now
8128 * rt: remaining time
8130 * rt is a sector_t, which is always 64bit now. We are keeping
8131 * the original algorithm, but it is not really necessary.
8133 * Original algorithm:
8134 * So we divide before multiply in case it is 32bit and close
8136 * We scale the divisor (db) by 32 to avoid losing precision
8137 * near the end of resync when the number of remaining sectors
8139 * We then divide rt by 32 after multiplying by db to compensate.
8140 * The '+1' avoids division by zero if db is very small.
8142 dt = ((jiffies - mddev->resync_mark) / HZ);
8145 curr_mark_cnt = mddev->curr_mark_cnt;
8146 recovery_active = atomic_read(&mddev->recovery_active);
8147 resync_mark_cnt = mddev->resync_mark_cnt;
8149 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8150 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8152 rt = max_sectors - resync; /* number of remaining sectors */
8153 rt = div64_u64(rt, db/32+1);
8157 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8158 ((unsigned long)rt % 60)/6);
8160 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8164 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8166 struct list_head *tmp;
8168 struct mddev *mddev;
8180 spin_lock(&all_mddevs_lock);
8181 list_for_each(tmp,&all_mddevs)
8183 mddev = list_entry(tmp, struct mddev, all_mddevs);
8185 spin_unlock(&all_mddevs_lock);
8188 spin_unlock(&all_mddevs_lock);
8190 return (void*)2;/* tail */
8194 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8196 struct list_head *tmp;
8197 struct mddev *next_mddev, *mddev = v;
8203 spin_lock(&all_mddevs_lock);
8205 tmp = all_mddevs.next;
8207 tmp = mddev->all_mddevs.next;
8208 if (tmp != &all_mddevs)
8209 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8211 next_mddev = (void*)2;
8214 spin_unlock(&all_mddevs_lock);
8222 static void md_seq_stop(struct seq_file *seq, void *v)
8224 struct mddev *mddev = v;
8226 if (mddev && v != (void*)1 && v != (void*)2)
8230 static int md_seq_show(struct seq_file *seq, void *v)
8232 struct mddev *mddev = v;
8234 struct md_rdev *rdev;
8236 if (v == (void*)1) {
8237 struct md_personality *pers;
8238 seq_printf(seq, "Personalities : ");
8239 spin_lock(&pers_lock);
8240 list_for_each_entry(pers, &pers_list, list)
8241 seq_printf(seq, "[%s] ", pers->name);
8243 spin_unlock(&pers_lock);
8244 seq_printf(seq, "\n");
8245 seq->poll_event = atomic_read(&md_event_count);
8248 if (v == (void*)2) {
8253 spin_lock(&mddev->lock);
8254 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8255 seq_printf(seq, "%s : %sactive", mdname(mddev),
8256 mddev->pers ? "" : "in");
8259 seq_printf(seq, " (read-only)");
8261 seq_printf(seq, " (auto-read-only)");
8262 seq_printf(seq, " %s", mddev->pers->name);
8267 rdev_for_each_rcu(rdev, mddev) {
8268 char b[BDEVNAME_SIZE];
8269 seq_printf(seq, " %s[%d]",
8270 bdevname(rdev->bdev,b), rdev->desc_nr);
8271 if (test_bit(WriteMostly, &rdev->flags))
8272 seq_printf(seq, "(W)");
8273 if (test_bit(Journal, &rdev->flags))
8274 seq_printf(seq, "(J)");
8275 if (test_bit(Faulty, &rdev->flags)) {
8276 seq_printf(seq, "(F)");
8279 if (rdev->raid_disk < 0)
8280 seq_printf(seq, "(S)"); /* spare */
8281 if (test_bit(Replacement, &rdev->flags))
8282 seq_printf(seq, "(R)");
8283 sectors += rdev->sectors;
8287 if (!list_empty(&mddev->disks)) {
8289 seq_printf(seq, "\n %llu blocks",
8290 (unsigned long long)
8291 mddev->array_sectors / 2);
8293 seq_printf(seq, "\n %llu blocks",
8294 (unsigned long long)sectors / 2);
8296 if (mddev->persistent) {
8297 if (mddev->major_version != 0 ||
8298 mddev->minor_version != 90) {
8299 seq_printf(seq," super %d.%d",
8300 mddev->major_version,
8301 mddev->minor_version);
8303 } else if (mddev->external)
8304 seq_printf(seq, " super external:%s",
8305 mddev->metadata_type);
8307 seq_printf(seq, " super non-persistent");
8310 mddev->pers->status(seq, mddev);
8311 seq_printf(seq, "\n ");
8312 if (mddev->pers->sync_request) {
8313 if (status_resync(seq, mddev))
8314 seq_printf(seq, "\n ");
8317 seq_printf(seq, "\n ");
8319 md_bitmap_status(seq, mddev->bitmap);
8321 seq_printf(seq, "\n");
8323 spin_unlock(&mddev->lock);
8328 static const struct seq_operations md_seq_ops = {
8329 .start = md_seq_start,
8330 .next = md_seq_next,
8331 .stop = md_seq_stop,
8332 .show = md_seq_show,
8335 static int md_seq_open(struct inode *inode, struct file *file)
8337 struct seq_file *seq;
8340 error = seq_open(file, &md_seq_ops);
8344 seq = file->private_data;
8345 seq->poll_event = atomic_read(&md_event_count);
8349 static int md_unloading;
8350 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8352 struct seq_file *seq = filp->private_data;
8356 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8357 poll_wait(filp, &md_event_waiters, wait);
8359 /* always allow read */
8360 mask = EPOLLIN | EPOLLRDNORM;
8362 if (seq->poll_event != atomic_read(&md_event_count))
8363 mask |= EPOLLERR | EPOLLPRI;
8367 static const struct proc_ops mdstat_proc_ops = {
8368 .proc_open = md_seq_open,
8369 .proc_read = seq_read,
8370 .proc_lseek = seq_lseek,
8371 .proc_release = seq_release,
8372 .proc_poll = mdstat_poll,
8375 int register_md_personality(struct md_personality *p)
8377 pr_debug("md: %s personality registered for level %d\n",
8379 spin_lock(&pers_lock);
8380 list_add_tail(&p->list, &pers_list);
8381 spin_unlock(&pers_lock);
8384 EXPORT_SYMBOL(register_md_personality);
8386 int unregister_md_personality(struct md_personality *p)
8388 pr_debug("md: %s personality unregistered\n", p->name);
8389 spin_lock(&pers_lock);
8390 list_del_init(&p->list);
8391 spin_unlock(&pers_lock);
8394 EXPORT_SYMBOL(unregister_md_personality);
8396 int register_md_cluster_operations(struct md_cluster_operations *ops,
8397 struct module *module)
8400 spin_lock(&pers_lock);
8401 if (md_cluster_ops != NULL)
8404 md_cluster_ops = ops;
8405 md_cluster_mod = module;
8407 spin_unlock(&pers_lock);
8410 EXPORT_SYMBOL(register_md_cluster_operations);
8412 int unregister_md_cluster_operations(void)
8414 spin_lock(&pers_lock);
8415 md_cluster_ops = NULL;
8416 spin_unlock(&pers_lock);
8419 EXPORT_SYMBOL(unregister_md_cluster_operations);
8421 int md_setup_cluster(struct mddev *mddev, int nodes)
8424 if (!md_cluster_ops)
8425 request_module("md-cluster");
8426 spin_lock(&pers_lock);
8427 /* ensure module won't be unloaded */
8428 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8429 pr_warn("can't find md-cluster module or get it's reference.\n");
8430 spin_unlock(&pers_lock);
8433 spin_unlock(&pers_lock);
8435 ret = md_cluster_ops->join(mddev, nodes);
8437 mddev->safemode_delay = 0;
8441 void md_cluster_stop(struct mddev *mddev)
8443 if (!md_cluster_ops)
8445 md_cluster_ops->leave(mddev);
8446 module_put(md_cluster_mod);
8449 static int is_mddev_idle(struct mddev *mddev, int init)
8451 struct md_rdev *rdev;
8457 rdev_for_each_rcu(rdev, mddev) {
8458 struct gendisk *disk = rdev->bdev->bd_disk;
8459 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8460 atomic_read(&disk->sync_io);
8461 /* sync IO will cause sync_io to increase before the disk_stats
8462 * as sync_io is counted when a request starts, and
8463 * disk_stats is counted when it completes.
8464 * So resync activity will cause curr_events to be smaller than
8465 * when there was no such activity.
8466 * non-sync IO will cause disk_stat to increase without
8467 * increasing sync_io so curr_events will (eventually)
8468 * be larger than it was before. Once it becomes
8469 * substantially larger, the test below will cause
8470 * the array to appear non-idle, and resync will slow
8472 * If there is a lot of outstanding resync activity when
8473 * we set last_event to curr_events, then all that activity
8474 * completing might cause the array to appear non-idle
8475 * and resync will be slowed down even though there might
8476 * not have been non-resync activity. This will only
8477 * happen once though. 'last_events' will soon reflect
8478 * the state where there is little or no outstanding
8479 * resync requests, and further resync activity will
8480 * always make curr_events less than last_events.
8483 if (init || curr_events - rdev->last_events > 64) {
8484 rdev->last_events = curr_events;
8492 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8494 /* another "blocks" (512byte) blocks have been synced */
8495 atomic_sub(blocks, &mddev->recovery_active);
8496 wake_up(&mddev->recovery_wait);
8498 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8499 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8500 md_wakeup_thread(mddev->thread);
8501 // stop recovery, signal do_sync ....
8504 EXPORT_SYMBOL(md_done_sync);
8506 /* md_write_start(mddev, bi)
8507 * If we need to update some array metadata (e.g. 'active' flag
8508 * in superblock) before writing, schedule a superblock update
8509 * and wait for it to complete.
8510 * A return value of 'false' means that the write wasn't recorded
8511 * and cannot proceed as the array is being suspend.
8513 bool md_write_start(struct mddev *mddev, struct bio *bi)
8517 if (bio_data_dir(bi) != WRITE)
8520 BUG_ON(mddev->ro == 1);
8521 if (mddev->ro == 2) {
8522 /* need to switch to read/write */
8524 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8525 md_wakeup_thread(mddev->thread);
8526 md_wakeup_thread(mddev->sync_thread);
8530 percpu_ref_get(&mddev->writes_pending);
8531 smp_mb(); /* Match smp_mb in set_in_sync() */
8532 if (mddev->safemode == 1)
8533 mddev->safemode = 0;
8534 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8535 if (mddev->in_sync || mddev->sync_checkers) {
8536 spin_lock(&mddev->lock);
8537 if (mddev->in_sync) {
8539 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8540 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8541 md_wakeup_thread(mddev->thread);
8544 spin_unlock(&mddev->lock);
8548 sysfs_notify_dirent_safe(mddev->sysfs_state);
8549 if (!mddev->has_superblocks)
8551 wait_event(mddev->sb_wait,
8552 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8554 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8555 percpu_ref_put(&mddev->writes_pending);
8560 EXPORT_SYMBOL(md_write_start);
8562 /* md_write_inc can only be called when md_write_start() has
8563 * already been called at least once of the current request.
8564 * It increments the counter and is useful when a single request
8565 * is split into several parts. Each part causes an increment and
8566 * so needs a matching md_write_end().
8567 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8568 * a spinlocked region.
8570 void md_write_inc(struct mddev *mddev, struct bio *bi)
8572 if (bio_data_dir(bi) != WRITE)
8574 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8575 percpu_ref_get(&mddev->writes_pending);
8577 EXPORT_SYMBOL(md_write_inc);
8579 void md_write_end(struct mddev *mddev)
8581 percpu_ref_put(&mddev->writes_pending);
8583 if (mddev->safemode == 2)
8584 md_wakeup_thread(mddev->thread);
8585 else if (mddev->safemode_delay)
8586 /* The roundup() ensures this only performs locking once
8587 * every ->safemode_delay jiffies
8589 mod_timer(&mddev->safemode_timer,
8590 roundup(jiffies, mddev->safemode_delay) +
8591 mddev->safemode_delay);
8594 EXPORT_SYMBOL(md_write_end);
8596 /* md_allow_write(mddev)
8597 * Calling this ensures that the array is marked 'active' so that writes
8598 * may proceed without blocking. It is important to call this before
8599 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8600 * Must be called with mddev_lock held.
8602 void md_allow_write(struct mddev *mddev)
8608 if (!mddev->pers->sync_request)
8611 spin_lock(&mddev->lock);
8612 if (mddev->in_sync) {
8614 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8615 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8616 if (mddev->safemode_delay &&
8617 mddev->safemode == 0)
8618 mddev->safemode = 1;
8619 spin_unlock(&mddev->lock);
8620 md_update_sb(mddev, 0);
8621 sysfs_notify_dirent_safe(mddev->sysfs_state);
8622 /* wait for the dirty state to be recorded in the metadata */
8623 wait_event(mddev->sb_wait,
8624 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8626 spin_unlock(&mddev->lock);
8628 EXPORT_SYMBOL_GPL(md_allow_write);
8630 #define SYNC_MARKS 10
8631 #define SYNC_MARK_STEP (3*HZ)
8632 #define UPDATE_FREQUENCY (5*60*HZ)
8633 void md_do_sync(struct md_thread *thread)
8635 struct mddev *mddev = thread->mddev;
8636 struct mddev *mddev2;
8637 unsigned int currspeed = 0, window;
8638 sector_t max_sectors,j, io_sectors, recovery_done;
8639 unsigned long mark[SYNC_MARKS];
8640 unsigned long update_time;
8641 sector_t mark_cnt[SYNC_MARKS];
8643 struct list_head *tmp;
8644 sector_t last_check;
8646 struct md_rdev *rdev;
8647 char *desc, *action = NULL;
8648 struct blk_plug plug;
8651 /* just incase thread restarts... */
8652 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8653 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8655 if (mddev->ro) {/* never try to sync a read-only array */
8656 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8660 if (mddev_is_clustered(mddev)) {
8661 ret = md_cluster_ops->resync_start(mddev);
8665 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8666 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8667 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8668 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8669 && ((unsigned long long)mddev->curr_resync_completed
8670 < (unsigned long long)mddev->resync_max_sectors))
8674 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8675 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8676 desc = "data-check";
8678 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8679 desc = "requested-resync";
8683 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8688 mddev->last_sync_action = action ?: desc;
8690 /* we overload curr_resync somewhat here.
8691 * 0 == not engaged in resync at all
8692 * 2 == checking that there is no conflict with another sync
8693 * 1 == like 2, but have yielded to allow conflicting resync to
8695 * other == active in resync - this many blocks
8697 * Before starting a resync we must have set curr_resync to
8698 * 2, and then checked that every "conflicting" array has curr_resync
8699 * less than ours. When we find one that is the same or higher
8700 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8701 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8702 * This will mean we have to start checking from the beginning again.
8707 int mddev2_minor = -1;
8708 mddev->curr_resync = 2;
8711 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8713 for_each_mddev(mddev2, tmp) {
8714 if (mddev2 == mddev)
8716 if (!mddev->parallel_resync
8717 && mddev2->curr_resync
8718 && match_mddev_units(mddev, mddev2)) {
8720 if (mddev < mddev2 && mddev->curr_resync == 2) {
8721 /* arbitrarily yield */
8722 mddev->curr_resync = 1;
8723 wake_up(&resync_wait);
8725 if (mddev > mddev2 && mddev->curr_resync == 1)
8726 /* no need to wait here, we can wait the next
8727 * time 'round when curr_resync == 2
8730 /* We need to wait 'interruptible' so as not to
8731 * contribute to the load average, and not to
8732 * be caught by 'softlockup'
8734 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8735 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8736 mddev2->curr_resync >= mddev->curr_resync) {
8737 if (mddev2_minor != mddev2->md_minor) {
8738 mddev2_minor = mddev2->md_minor;
8739 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8740 desc, mdname(mddev),
8744 if (signal_pending(current))
8745 flush_signals(current);
8747 finish_wait(&resync_wait, &wq);
8750 finish_wait(&resync_wait, &wq);
8753 } while (mddev->curr_resync < 2);
8756 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8757 /* resync follows the size requested by the personality,
8758 * which defaults to physical size, but can be virtual size
8760 max_sectors = mddev->resync_max_sectors;
8761 atomic64_set(&mddev->resync_mismatches, 0);
8762 /* we don't use the checkpoint if there's a bitmap */
8763 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8764 j = mddev->resync_min;
8765 else if (!mddev->bitmap)
8766 j = mddev->recovery_cp;
8768 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8769 max_sectors = mddev->resync_max_sectors;
8771 * If the original node aborts reshaping then we continue the
8772 * reshaping, so set j again to avoid restart reshape from the
8775 if (mddev_is_clustered(mddev) &&
8776 mddev->reshape_position != MaxSector)
8777 j = mddev->reshape_position;
8779 /* recovery follows the physical size of devices */
8780 max_sectors = mddev->dev_sectors;
8783 rdev_for_each_rcu(rdev, mddev)
8784 if (rdev->raid_disk >= 0 &&
8785 !test_bit(Journal, &rdev->flags) &&
8786 !test_bit(Faulty, &rdev->flags) &&
8787 !test_bit(In_sync, &rdev->flags) &&
8788 rdev->recovery_offset < j)
8789 j = rdev->recovery_offset;
8792 /* If there is a bitmap, we need to make sure all
8793 * writes that started before we added a spare
8794 * complete before we start doing a recovery.
8795 * Otherwise the write might complete and (via
8796 * bitmap_endwrite) set a bit in the bitmap after the
8797 * recovery has checked that bit and skipped that
8800 if (mddev->bitmap) {
8801 mddev->pers->quiesce(mddev, 1);
8802 mddev->pers->quiesce(mddev, 0);
8806 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8807 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8808 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8809 speed_max(mddev), desc);
8811 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8814 for (m = 0; m < SYNC_MARKS; m++) {
8816 mark_cnt[m] = io_sectors;
8819 mddev->resync_mark = mark[last_mark];
8820 mddev->resync_mark_cnt = mark_cnt[last_mark];
8823 * Tune reconstruction:
8825 window = 32 * (PAGE_SIZE / 512);
8826 pr_debug("md: using %dk window, over a total of %lluk.\n",
8827 window/2, (unsigned long long)max_sectors/2);
8829 atomic_set(&mddev->recovery_active, 0);
8833 pr_debug("md: resuming %s of %s from checkpoint.\n",
8834 desc, mdname(mddev));
8835 mddev->curr_resync = j;
8837 mddev->curr_resync = 3; /* no longer delayed */
8838 mddev->curr_resync_completed = j;
8839 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8840 md_new_event(mddev);
8841 update_time = jiffies;
8843 blk_start_plug(&plug);
8844 while (j < max_sectors) {
8849 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8850 ((mddev->curr_resync > mddev->curr_resync_completed &&
8851 (mddev->curr_resync - mddev->curr_resync_completed)
8852 > (max_sectors >> 4)) ||
8853 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8854 (j - mddev->curr_resync_completed)*2
8855 >= mddev->resync_max - mddev->curr_resync_completed ||
8856 mddev->curr_resync_completed > mddev->resync_max
8858 /* time to update curr_resync_completed */
8859 wait_event(mddev->recovery_wait,
8860 atomic_read(&mddev->recovery_active) == 0);
8861 mddev->curr_resync_completed = j;
8862 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8863 j > mddev->recovery_cp)
8864 mddev->recovery_cp = j;
8865 update_time = jiffies;
8866 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8867 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8870 while (j >= mddev->resync_max &&
8871 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8872 /* As this condition is controlled by user-space,
8873 * we can block indefinitely, so use '_interruptible'
8874 * to avoid triggering warnings.
8876 flush_signals(current); /* just in case */
8877 wait_event_interruptible(mddev->recovery_wait,
8878 mddev->resync_max > j
8879 || test_bit(MD_RECOVERY_INTR,
8883 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8886 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8888 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8892 if (!skipped) { /* actual IO requested */
8893 io_sectors += sectors;
8894 atomic_add(sectors, &mddev->recovery_active);
8897 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8901 if (j > max_sectors)
8902 /* when skipping, extra large numbers can be returned. */
8905 mddev->curr_resync = j;
8906 mddev->curr_mark_cnt = io_sectors;
8907 if (last_check == 0)
8908 /* this is the earliest that rebuild will be
8909 * visible in /proc/mdstat
8911 md_new_event(mddev);
8913 if (last_check + window > io_sectors || j == max_sectors)
8916 last_check = io_sectors;
8918 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8920 int next = (last_mark+1) % SYNC_MARKS;
8922 mddev->resync_mark = mark[next];
8923 mddev->resync_mark_cnt = mark_cnt[next];
8924 mark[next] = jiffies;
8925 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8929 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8933 * this loop exits only if either when we are slower than
8934 * the 'hard' speed limit, or the system was IO-idle for
8936 * the system might be non-idle CPU-wise, but we only care
8937 * about not overloading the IO subsystem. (things like an
8938 * e2fsck being done on the RAID array should execute fast)
8942 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8943 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8944 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8946 if (currspeed > speed_min(mddev)) {
8947 if (currspeed > speed_max(mddev)) {
8951 if (!is_mddev_idle(mddev, 0)) {
8953 * Give other IO more of a chance.
8954 * The faster the devices, the less we wait.
8956 wait_event(mddev->recovery_wait,
8957 !atomic_read(&mddev->recovery_active));
8961 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8962 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8963 ? "interrupted" : "done");
8965 * this also signals 'finished resyncing' to md_stop
8967 blk_finish_plug(&plug);
8968 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8970 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8971 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8972 mddev->curr_resync > 3) {
8973 mddev->curr_resync_completed = mddev->curr_resync;
8974 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8976 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8978 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8979 mddev->curr_resync > 3) {
8980 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8981 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8982 if (mddev->curr_resync >= mddev->recovery_cp) {
8983 pr_debug("md: checkpointing %s of %s.\n",
8984 desc, mdname(mddev));
8985 if (test_bit(MD_RECOVERY_ERROR,
8987 mddev->recovery_cp =
8988 mddev->curr_resync_completed;
8990 mddev->recovery_cp =
8994 mddev->recovery_cp = MaxSector;
8996 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8997 mddev->curr_resync = MaxSector;
8998 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8999 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9001 rdev_for_each_rcu(rdev, mddev)
9002 if (rdev->raid_disk >= 0 &&
9003 mddev->delta_disks >= 0 &&
9004 !test_bit(Journal, &rdev->flags) &&
9005 !test_bit(Faulty, &rdev->flags) &&
9006 !test_bit(In_sync, &rdev->flags) &&
9007 rdev->recovery_offset < mddev->curr_resync)
9008 rdev->recovery_offset = mddev->curr_resync;
9014 /* set CHANGE_PENDING here since maybe another update is needed,
9015 * so other nodes are informed. It should be harmless for normal
9017 set_mask_bits(&mddev->sb_flags, 0,
9018 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9020 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9021 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9022 mddev->delta_disks > 0 &&
9023 mddev->pers->finish_reshape &&
9024 mddev->pers->size &&
9026 mddev_lock_nointr(mddev);
9027 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9028 mddev_unlock(mddev);
9029 if (!mddev_is_clustered(mddev)) {
9030 set_capacity(mddev->gendisk, mddev->array_sectors);
9031 revalidate_disk_size(mddev->gendisk, true);
9035 spin_lock(&mddev->lock);
9036 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9037 /* We completed so min/max setting can be forgotten if used. */
9038 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9039 mddev->resync_min = 0;
9040 mddev->resync_max = MaxSector;
9041 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9042 mddev->resync_min = mddev->curr_resync_completed;
9043 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9044 mddev->curr_resync = 0;
9045 spin_unlock(&mddev->lock);
9047 wake_up(&resync_wait);
9048 md_wakeup_thread(mddev->thread);
9051 EXPORT_SYMBOL_GPL(md_do_sync);
9053 static int remove_and_add_spares(struct mddev *mddev,
9054 struct md_rdev *this)
9056 struct md_rdev *rdev;
9059 bool remove_some = false;
9061 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9062 /* Mustn't remove devices when resync thread is running */
9065 rdev_for_each(rdev, mddev) {
9066 if ((this == NULL || rdev == this) &&
9067 rdev->raid_disk >= 0 &&
9068 !test_bit(Blocked, &rdev->flags) &&
9069 test_bit(Faulty, &rdev->flags) &&
9070 atomic_read(&rdev->nr_pending)==0) {
9071 /* Faulty non-Blocked devices with nr_pending == 0
9072 * never get nr_pending incremented,
9073 * never get Faulty cleared, and never get Blocked set.
9074 * So we can synchronize_rcu now rather than once per device
9077 set_bit(RemoveSynchronized, &rdev->flags);
9083 rdev_for_each(rdev, mddev) {
9084 if ((this == NULL || rdev == this) &&
9085 rdev->raid_disk >= 0 &&
9086 !test_bit(Blocked, &rdev->flags) &&
9087 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9088 (!test_bit(In_sync, &rdev->flags) &&
9089 !test_bit(Journal, &rdev->flags))) &&
9090 atomic_read(&rdev->nr_pending)==0)) {
9091 if (mddev->pers->hot_remove_disk(
9092 mddev, rdev) == 0) {
9093 sysfs_unlink_rdev(mddev, rdev);
9094 rdev->saved_raid_disk = rdev->raid_disk;
9095 rdev->raid_disk = -1;
9099 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9100 clear_bit(RemoveSynchronized, &rdev->flags);
9103 if (removed && mddev->kobj.sd)
9104 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9106 if (this && removed)
9109 rdev_for_each(rdev, mddev) {
9110 if (this && this != rdev)
9112 if (test_bit(Candidate, &rdev->flags))
9114 if (rdev->raid_disk >= 0 &&
9115 !test_bit(In_sync, &rdev->flags) &&
9116 !test_bit(Journal, &rdev->flags) &&
9117 !test_bit(Faulty, &rdev->flags))
9119 if (rdev->raid_disk >= 0)
9121 if (test_bit(Faulty, &rdev->flags))
9123 if (!test_bit(Journal, &rdev->flags)) {
9125 ! (rdev->saved_raid_disk >= 0 &&
9126 !test_bit(Bitmap_sync, &rdev->flags)))
9129 rdev->recovery_offset = 0;
9131 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9132 /* failure here is OK */
9133 sysfs_link_rdev(mddev, rdev);
9134 if (!test_bit(Journal, &rdev->flags))
9136 md_new_event(mddev);
9137 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9142 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9146 static void md_start_sync(struct work_struct *ws)
9148 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9150 mddev->sync_thread = md_register_thread(md_do_sync,
9153 if (!mddev->sync_thread) {
9154 pr_warn("%s: could not start resync thread...\n",
9156 /* leave the spares where they are, it shouldn't hurt */
9157 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9158 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9159 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9160 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9161 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9162 wake_up(&resync_wait);
9163 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9165 if (mddev->sysfs_action)
9166 sysfs_notify_dirent_safe(mddev->sysfs_action);
9168 md_wakeup_thread(mddev->sync_thread);
9169 sysfs_notify_dirent_safe(mddev->sysfs_action);
9170 md_new_event(mddev);
9174 * This routine is regularly called by all per-raid-array threads to
9175 * deal with generic issues like resync and super-block update.
9176 * Raid personalities that don't have a thread (linear/raid0) do not
9177 * need this as they never do any recovery or update the superblock.
9179 * It does not do any resync itself, but rather "forks" off other threads
9180 * to do that as needed.
9181 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9182 * "->recovery" and create a thread at ->sync_thread.
9183 * When the thread finishes it sets MD_RECOVERY_DONE
9184 * and wakeups up this thread which will reap the thread and finish up.
9185 * This thread also removes any faulty devices (with nr_pending == 0).
9187 * The overall approach is:
9188 * 1/ if the superblock needs updating, update it.
9189 * 2/ If a recovery thread is running, don't do anything else.
9190 * 3/ If recovery has finished, clean up, possibly marking spares active.
9191 * 4/ If there are any faulty devices, remove them.
9192 * 5/ If array is degraded, try to add spares devices
9193 * 6/ If array has spares or is not in-sync, start a resync thread.
9195 void md_check_recovery(struct mddev *mddev)
9197 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9198 /* Write superblock - thread that called mddev_suspend()
9199 * holds reconfig_mutex for us.
9201 set_bit(MD_UPDATING_SB, &mddev->flags);
9202 smp_mb__after_atomic();
9203 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9204 md_update_sb(mddev, 0);
9205 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9206 wake_up(&mddev->sb_wait);
9209 if (mddev->suspended)
9213 md_bitmap_daemon_work(mddev);
9215 if (signal_pending(current)) {
9216 if (mddev->pers->sync_request && !mddev->external) {
9217 pr_debug("md: %s in immediate safe mode\n",
9219 mddev->safemode = 2;
9221 flush_signals(current);
9224 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9227 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9228 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9229 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9230 (mddev->external == 0 && mddev->safemode == 1) ||
9231 (mddev->safemode == 2
9232 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9236 if (mddev_trylock(mddev)) {
9238 bool try_set_sync = mddev->safemode != 0;
9240 if (!mddev->external && mddev->safemode == 1)
9241 mddev->safemode = 0;
9244 struct md_rdev *rdev;
9245 if (!mddev->external && mddev->in_sync)
9246 /* 'Blocked' flag not needed as failed devices
9247 * will be recorded if array switched to read/write.
9248 * Leaving it set will prevent the device
9249 * from being removed.
9251 rdev_for_each(rdev, mddev)
9252 clear_bit(Blocked, &rdev->flags);
9253 /* On a read-only array we can:
9254 * - remove failed devices
9255 * - add already-in_sync devices if the array itself
9257 * As we only add devices that are already in-sync,
9258 * we can activate the spares immediately.
9260 remove_and_add_spares(mddev, NULL);
9261 /* There is no thread, but we need to call
9262 * ->spare_active and clear saved_raid_disk
9264 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9265 md_reap_sync_thread(mddev);
9266 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9267 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9268 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9272 if (mddev_is_clustered(mddev)) {
9273 struct md_rdev *rdev, *tmp;
9274 /* kick the device if another node issued a
9277 rdev_for_each_safe(rdev, tmp, mddev) {
9278 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9279 rdev->raid_disk < 0)
9280 md_kick_rdev_from_array(rdev);
9284 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9285 spin_lock(&mddev->lock);
9287 spin_unlock(&mddev->lock);
9290 if (mddev->sb_flags)
9291 md_update_sb(mddev, 0);
9293 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9294 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9295 /* resync/recovery still happening */
9296 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9299 if (mddev->sync_thread) {
9300 md_reap_sync_thread(mddev);
9303 /* Set RUNNING before clearing NEEDED to avoid
9304 * any transients in the value of "sync_action".
9306 mddev->curr_resync_completed = 0;
9307 spin_lock(&mddev->lock);
9308 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9309 spin_unlock(&mddev->lock);
9310 /* Clear some bits that don't mean anything, but
9313 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9314 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9316 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9317 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9319 /* no recovery is running.
9320 * remove any failed drives, then
9321 * add spares if possible.
9322 * Spares are also removed and re-added, to allow
9323 * the personality to fail the re-add.
9326 if (mddev->reshape_position != MaxSector) {
9327 if (mddev->pers->check_reshape == NULL ||
9328 mddev->pers->check_reshape(mddev) != 0)
9329 /* Cannot proceed */
9331 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9332 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9333 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9334 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9335 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9336 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9337 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9338 } else if (mddev->recovery_cp < MaxSector) {
9339 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9340 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9341 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9342 /* nothing to be done ... */
9345 if (mddev->pers->sync_request) {
9347 /* We are adding a device or devices to an array
9348 * which has the bitmap stored on all devices.
9349 * So make sure all bitmap pages get written
9351 md_bitmap_write_all(mddev->bitmap);
9353 INIT_WORK(&mddev->del_work, md_start_sync);
9354 queue_work(md_misc_wq, &mddev->del_work);
9358 if (!mddev->sync_thread) {
9359 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9360 wake_up(&resync_wait);
9361 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9363 if (mddev->sysfs_action)
9364 sysfs_notify_dirent_safe(mddev->sysfs_action);
9367 wake_up(&mddev->sb_wait);
9368 mddev_unlock(mddev);
9371 EXPORT_SYMBOL(md_check_recovery);
9373 void md_reap_sync_thread(struct mddev *mddev)
9375 struct md_rdev *rdev;
9376 sector_t old_dev_sectors = mddev->dev_sectors;
9377 bool is_reshaped = false;
9379 /* resync has finished, collect result */
9380 md_unregister_thread(&mddev->sync_thread);
9381 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9382 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9383 mddev->degraded != mddev->raid_disks) {
9385 /* activate any spares */
9386 if (mddev->pers->spare_active(mddev)) {
9387 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9388 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9391 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9392 mddev->pers->finish_reshape) {
9393 mddev->pers->finish_reshape(mddev);
9394 if (mddev_is_clustered(mddev))
9398 /* If array is no-longer degraded, then any saved_raid_disk
9399 * information must be scrapped.
9401 if (!mddev->degraded)
9402 rdev_for_each(rdev, mddev)
9403 rdev->saved_raid_disk = -1;
9405 md_update_sb(mddev, 1);
9406 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9407 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9409 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9410 md_cluster_ops->resync_finish(mddev);
9411 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9412 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9413 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9414 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9415 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9416 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9418 * We call md_cluster_ops->update_size here because sync_size could
9419 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9420 * so it is time to update size across cluster.
9422 if (mddev_is_clustered(mddev) && is_reshaped
9423 && !test_bit(MD_CLOSING, &mddev->flags))
9424 md_cluster_ops->update_size(mddev, old_dev_sectors);
9425 wake_up(&resync_wait);
9426 /* flag recovery needed just to double check */
9427 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9428 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9429 sysfs_notify_dirent_safe(mddev->sysfs_action);
9430 md_new_event(mddev);
9431 if (mddev->event_work.func)
9432 queue_work(md_misc_wq, &mddev->event_work);
9434 EXPORT_SYMBOL(md_reap_sync_thread);
9436 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9438 sysfs_notify_dirent_safe(rdev->sysfs_state);
9439 wait_event_timeout(rdev->blocked_wait,
9440 !test_bit(Blocked, &rdev->flags) &&
9441 !test_bit(BlockedBadBlocks, &rdev->flags),
9442 msecs_to_jiffies(5000));
9443 rdev_dec_pending(rdev, mddev);
9445 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9447 void md_finish_reshape(struct mddev *mddev)
9449 /* called be personality module when reshape completes. */
9450 struct md_rdev *rdev;
9452 rdev_for_each(rdev, mddev) {
9453 if (rdev->data_offset > rdev->new_data_offset)
9454 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9456 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9457 rdev->data_offset = rdev->new_data_offset;
9460 EXPORT_SYMBOL(md_finish_reshape);
9462 /* Bad block management */
9464 /* Returns 1 on success, 0 on failure */
9465 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9468 struct mddev *mddev = rdev->mddev;
9471 s += rdev->new_data_offset;
9473 s += rdev->data_offset;
9474 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9476 /* Make sure they get written out promptly */
9477 if (test_bit(ExternalBbl, &rdev->flags))
9478 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9479 sysfs_notify_dirent_safe(rdev->sysfs_state);
9480 set_mask_bits(&mddev->sb_flags, 0,
9481 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9482 md_wakeup_thread(rdev->mddev->thread);
9487 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9489 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9494 s += rdev->new_data_offset;
9496 s += rdev->data_offset;
9497 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9498 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9499 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9502 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9504 static int md_notify_reboot(struct notifier_block *this,
9505 unsigned long code, void *x)
9507 struct list_head *tmp;
9508 struct mddev *mddev;
9511 for_each_mddev(mddev, tmp) {
9512 if (mddev_trylock(mddev)) {
9514 __md_stop_writes(mddev);
9515 if (mddev->persistent)
9516 mddev->safemode = 2;
9517 mddev_unlock(mddev);
9522 * certain more exotic SCSI devices are known to be
9523 * volatile wrt too early system reboots. While the
9524 * right place to handle this issue is the given
9525 * driver, we do want to have a safe RAID driver ...
9533 static struct notifier_block md_notifier = {
9534 .notifier_call = md_notify_reboot,
9536 .priority = INT_MAX, /* before any real devices */
9539 static void md_geninit(void)
9541 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9543 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9546 static int __init md_init(void)
9550 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9554 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9558 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9559 if (!md_rdev_misc_wq)
9560 goto err_rdev_misc_wq;
9562 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9565 if ((ret = register_blkdev(0, "mdp")) < 0)
9569 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9570 md_probe, NULL, NULL);
9571 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9572 md_probe, NULL, NULL);
9574 register_reboot_notifier(&md_notifier);
9575 raid_table_header = register_sysctl_table(raid_root_table);
9581 unregister_blkdev(MD_MAJOR, "md");
9583 destroy_workqueue(md_rdev_misc_wq);
9585 destroy_workqueue(md_misc_wq);
9587 destroy_workqueue(md_wq);
9592 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9594 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9595 struct md_rdev *rdev2, *tmp;
9597 char b[BDEVNAME_SIZE];
9600 * If size is changed in another node then we need to
9601 * do resize as well.
9603 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9604 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9606 pr_info("md-cluster: resize failed\n");
9608 md_bitmap_update_sb(mddev->bitmap);
9611 /* Check for change of roles in the active devices */
9612 rdev_for_each_safe(rdev2, tmp, mddev) {
9613 if (test_bit(Faulty, &rdev2->flags))
9616 /* Check if the roles changed */
9617 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9619 if (test_bit(Candidate, &rdev2->flags)) {
9620 if (role == 0xfffe) {
9621 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9622 md_kick_rdev_from_array(rdev2);
9626 clear_bit(Candidate, &rdev2->flags);
9629 if (role != rdev2->raid_disk) {
9631 * got activated except reshape is happening.
9633 if (rdev2->raid_disk == -1 && role != 0xffff &&
9634 !(le32_to_cpu(sb->feature_map) &
9635 MD_FEATURE_RESHAPE_ACTIVE)) {
9636 rdev2->saved_raid_disk = role;
9637 ret = remove_and_add_spares(mddev, rdev2);
9638 pr_info("Activated spare: %s\n",
9639 bdevname(rdev2->bdev,b));
9640 /* wakeup mddev->thread here, so array could
9641 * perform resync with the new activated disk */
9642 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9643 md_wakeup_thread(mddev->thread);
9646 * We just want to do the minimum to mark the disk
9647 * as faulty. The recovery is performed by the
9648 * one who initiated the error.
9650 if ((role == 0xfffe) || (role == 0xfffd)) {
9651 md_error(mddev, rdev2);
9652 clear_bit(Blocked, &rdev2->flags);
9657 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9658 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9660 pr_warn("md: updating array disks failed. %d\n", ret);
9664 * Since mddev->delta_disks has already updated in update_raid_disks,
9665 * so it is time to check reshape.
9667 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9668 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9670 * reshape is happening in the remote node, we need to
9671 * update reshape_position and call start_reshape.
9673 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9674 if (mddev->pers->update_reshape_pos)
9675 mddev->pers->update_reshape_pos(mddev);
9676 if (mddev->pers->start_reshape)
9677 mddev->pers->start_reshape(mddev);
9678 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9679 mddev->reshape_position != MaxSector &&
9680 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9681 /* reshape is just done in another node. */
9682 mddev->reshape_position = MaxSector;
9683 if (mddev->pers->update_reshape_pos)
9684 mddev->pers->update_reshape_pos(mddev);
9687 /* Finally set the event to be up to date */
9688 mddev->events = le64_to_cpu(sb->events);
9691 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9694 struct page *swapout = rdev->sb_page;
9695 struct mdp_superblock_1 *sb;
9697 /* Store the sb page of the rdev in the swapout temporary
9698 * variable in case we err in the future
9700 rdev->sb_page = NULL;
9701 err = alloc_disk_sb(rdev);
9703 ClearPageUptodate(rdev->sb_page);
9704 rdev->sb_loaded = 0;
9705 err = super_types[mddev->major_version].
9706 load_super(rdev, NULL, mddev->minor_version);
9709 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9710 __func__, __LINE__, rdev->desc_nr, err);
9712 put_page(rdev->sb_page);
9713 rdev->sb_page = swapout;
9714 rdev->sb_loaded = 1;
9718 sb = page_address(rdev->sb_page);
9719 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9723 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9724 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9726 /* The other node finished recovery, call spare_active to set
9727 * device In_sync and mddev->degraded
9729 if (rdev->recovery_offset == MaxSector &&
9730 !test_bit(In_sync, &rdev->flags) &&
9731 mddev->pers->spare_active(mddev))
9732 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9738 void md_reload_sb(struct mddev *mddev, int nr)
9740 struct md_rdev *rdev = NULL, *iter;
9744 rdev_for_each_rcu(iter, mddev) {
9745 if (iter->desc_nr == nr) {
9752 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9756 err = read_rdev(mddev, rdev);
9760 check_sb_changes(mddev, rdev);
9762 /* Read all rdev's to update recovery_offset */
9763 rdev_for_each_rcu(rdev, mddev) {
9764 if (!test_bit(Faulty, &rdev->flags))
9765 read_rdev(mddev, rdev);
9768 EXPORT_SYMBOL(md_reload_sb);
9773 * Searches all registered partitions for autorun RAID arrays
9777 static DEFINE_MUTEX(detected_devices_mutex);
9778 static LIST_HEAD(all_detected_devices);
9779 struct detected_devices_node {
9780 struct list_head list;
9784 void md_autodetect_dev(dev_t dev)
9786 struct detected_devices_node *node_detected_dev;
9788 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9789 if (node_detected_dev) {
9790 node_detected_dev->dev = dev;
9791 mutex_lock(&detected_devices_mutex);
9792 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9793 mutex_unlock(&detected_devices_mutex);
9797 void md_autostart_arrays(int part)
9799 struct md_rdev *rdev;
9800 struct detected_devices_node *node_detected_dev;
9802 int i_scanned, i_passed;
9807 pr_info("md: Autodetecting RAID arrays.\n");
9809 mutex_lock(&detected_devices_mutex);
9810 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9812 node_detected_dev = list_entry(all_detected_devices.next,
9813 struct detected_devices_node, list);
9814 list_del(&node_detected_dev->list);
9815 dev = node_detected_dev->dev;
9816 kfree(node_detected_dev);
9817 mutex_unlock(&detected_devices_mutex);
9818 rdev = md_import_device(dev,0, 90);
9819 mutex_lock(&detected_devices_mutex);
9823 if (test_bit(Faulty, &rdev->flags))
9826 set_bit(AutoDetected, &rdev->flags);
9827 list_add(&rdev->same_set, &pending_raid_disks);
9830 mutex_unlock(&detected_devices_mutex);
9832 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9834 autorun_devices(part);
9837 #endif /* !MODULE */
9839 static __exit void md_exit(void)
9841 struct mddev *mddev;
9842 struct list_head *tmp;
9845 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9846 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9848 unregister_blkdev(MD_MAJOR,"md");
9849 unregister_blkdev(mdp_major, "mdp");
9850 unregister_reboot_notifier(&md_notifier);
9851 unregister_sysctl_table(raid_table_header);
9853 /* We cannot unload the modules while some process is
9854 * waiting for us in select() or poll() - wake them up
9857 while (waitqueue_active(&md_event_waiters)) {
9858 /* not safe to leave yet */
9859 wake_up(&md_event_waiters);
9863 remove_proc_entry("mdstat", NULL);
9865 for_each_mddev(mddev, tmp) {
9866 export_array(mddev);
9868 mddev->hold_active = 0;
9870 * for_each_mddev() will call mddev_put() at the end of each
9871 * iteration. As the mddev is now fully clear, this will
9872 * schedule the mddev for destruction by a workqueue, and the
9873 * destroy_workqueue() below will wait for that to complete.
9876 destroy_workqueue(md_rdev_misc_wq);
9877 destroy_workqueue(md_misc_wq);
9878 destroy_workqueue(md_wq);
9881 subsys_initcall(md_init);
9882 module_exit(md_exit)
9884 static int get_ro(char *buffer, const struct kernel_param *kp)
9886 return sprintf(buffer, "%d\n", start_readonly);
9888 static int set_ro(const char *val, const struct kernel_param *kp)
9890 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9893 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9894 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9895 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9896 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9898 MODULE_LICENSE("GPL");
9899 MODULE_DESCRIPTION("MD RAID framework");
9901 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);