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);
464 bio_end_io_t *orig_bi_end_io;
465 void *orig_bi_private;
466 unsigned long start_time;
467 struct hd_struct *part;
470 static void md_end_io(struct bio *bio)
472 struct md_io *md_io = bio->bi_private;
473 struct mddev *mddev = md_io->mddev;
475 part_end_io_acct(md_io->part, bio, md_io->start_time);
477 bio->bi_end_io = md_io->orig_bi_end_io;
478 bio->bi_private = md_io->orig_bi_private;
480 mempool_free(md_io, &mddev->md_io_pool);
486 static blk_qc_t md_submit_bio(struct bio *bio)
488 const int rw = bio_data_dir(bio);
489 struct mddev *mddev = bio->bi_disk->private_data;
491 if (mddev == NULL || mddev->pers == NULL) {
493 return BLK_QC_T_NONE;
496 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
498 return BLK_QC_T_NONE;
501 blk_queue_split(&bio);
503 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
504 if (bio_sectors(bio) != 0)
505 bio->bi_status = BLK_STS_IOERR;
507 return BLK_QC_T_NONE;
510 if (bio->bi_end_io != md_end_io) {
513 md_io = mempool_alloc(&mddev->md_io_pool, GFP_NOIO);
514 md_io->mddev = mddev;
515 md_io->orig_bi_end_io = bio->bi_end_io;
516 md_io->orig_bi_private = bio->bi_private;
518 bio->bi_end_io = md_end_io;
519 bio->bi_private = md_io;
521 md_io->start_time = part_start_io_acct(mddev->gendisk,
525 /* bio could be mergeable after passing to underlayer */
526 bio->bi_opf &= ~REQ_NOMERGE;
528 md_handle_request(mddev, bio);
530 return BLK_QC_T_NONE;
533 /* mddev_suspend makes sure no new requests are submitted
534 * to the device, and that any requests that have been submitted
535 * are completely handled.
536 * Once mddev_detach() is called and completes, the module will be
539 void mddev_suspend(struct mddev *mddev)
541 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
542 lockdep_assert_held(&mddev->reconfig_mutex);
543 if (mddev->suspended++)
546 wake_up(&mddev->sb_wait);
547 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
548 smp_mb__after_atomic();
549 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
550 mddev->pers->quiesce(mddev, 1);
551 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
552 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
554 del_timer_sync(&mddev->safemode_timer);
555 /* restrict memory reclaim I/O during raid array is suspend */
556 mddev->noio_flag = memalloc_noio_save();
558 EXPORT_SYMBOL_GPL(mddev_suspend);
560 void mddev_resume(struct mddev *mddev)
562 /* entred the memalloc scope from mddev_suspend() */
563 memalloc_noio_restore(mddev->noio_flag);
564 lockdep_assert_held(&mddev->reconfig_mutex);
565 if (--mddev->suspended)
567 wake_up(&mddev->sb_wait);
568 mddev->pers->quiesce(mddev, 0);
570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
571 md_wakeup_thread(mddev->thread);
572 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
574 EXPORT_SYMBOL_GPL(mddev_resume);
577 * Generic flush handling for md
580 static void md_end_flush(struct bio *bio)
582 struct md_rdev *rdev = bio->bi_private;
583 struct mddev *mddev = rdev->mddev;
585 rdev_dec_pending(rdev, mddev);
587 if (atomic_dec_and_test(&mddev->flush_pending)) {
588 /* The pre-request flush has finished */
589 queue_work(md_wq, &mddev->flush_work);
594 static void md_submit_flush_data(struct work_struct *ws);
596 static void submit_flushes(struct work_struct *ws)
598 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
599 struct md_rdev *rdev;
601 mddev->start_flush = ktime_get_boottime();
602 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
603 atomic_set(&mddev->flush_pending, 1);
605 rdev_for_each_rcu(rdev, mddev)
606 if (rdev->raid_disk >= 0 &&
607 !test_bit(Faulty, &rdev->flags)) {
608 /* Take two references, one is dropped
609 * when request finishes, one after
610 * we reclaim rcu_read_lock
613 atomic_inc(&rdev->nr_pending);
614 atomic_inc(&rdev->nr_pending);
616 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
617 bi->bi_end_io = md_end_flush;
618 bi->bi_private = rdev;
619 bio_set_dev(bi, rdev->bdev);
620 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
621 atomic_inc(&mddev->flush_pending);
624 rdev_dec_pending(rdev, mddev);
627 if (atomic_dec_and_test(&mddev->flush_pending))
628 queue_work(md_wq, &mddev->flush_work);
631 static void md_submit_flush_data(struct work_struct *ws)
633 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
634 struct bio *bio = mddev->flush_bio;
637 * must reset flush_bio before calling into md_handle_request to avoid a
638 * deadlock, because other bios passed md_handle_request suspend check
639 * could wait for this and below md_handle_request could wait for those
640 * bios because of suspend check
642 mddev->prev_flush_start = mddev->start_flush;
643 mddev->flush_bio = NULL;
644 wake_up(&mddev->sb_wait);
646 if (bio->bi_iter.bi_size == 0) {
647 /* an empty barrier - all done */
650 bio->bi_opf &= ~REQ_PREFLUSH;
651 md_handle_request(mddev, bio);
656 * Manages consolidation of flushes and submitting any flushes needed for
657 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
658 * being finished in another context. Returns false if the flushing is
659 * complete but still needs the I/O portion of the bio to be processed.
661 bool md_flush_request(struct mddev *mddev, struct bio *bio)
663 ktime_t req_start = ktime_get_boottime();
664 spin_lock_irq(&mddev->lock);
665 /* flush requests wait until ongoing flush completes,
666 * hence coalescing all the pending requests.
668 wait_event_lock_irq(mddev->sb_wait,
670 ktime_before(req_start, mddev->prev_flush_start),
672 /* new request after previous flush is completed */
673 if (ktime_after(req_start, mddev->prev_flush_start)) {
674 WARN_ON(mddev->flush_bio);
675 mddev->flush_bio = bio;
678 spin_unlock_irq(&mddev->lock);
681 INIT_WORK(&mddev->flush_work, submit_flushes);
682 queue_work(md_wq, &mddev->flush_work);
684 /* flush was performed for some other bio while we waited. */
685 if (bio->bi_iter.bi_size == 0)
686 /* an empty barrier - all done */
689 bio->bi_opf &= ~REQ_PREFLUSH;
695 EXPORT_SYMBOL(md_flush_request);
697 static inline struct mddev *mddev_get(struct mddev *mddev)
699 atomic_inc(&mddev->active);
703 static void mddev_delayed_delete(struct work_struct *ws);
705 static void mddev_put(struct mddev *mddev)
707 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
709 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
710 mddev->ctime == 0 && !mddev->hold_active) {
711 /* Array is not configured at all, and not held active,
713 list_del_init(&mddev->all_mddevs);
716 * Call queue_work inside the spinlock so that
717 * flush_workqueue() after mddev_find will succeed in waiting
718 * for the work to be done.
720 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
721 queue_work(md_misc_wq, &mddev->del_work);
723 spin_unlock(&all_mddevs_lock);
726 static void md_safemode_timeout(struct timer_list *t);
728 void mddev_init(struct mddev *mddev)
730 kobject_init(&mddev->kobj, &md_ktype);
731 mutex_init(&mddev->open_mutex);
732 mutex_init(&mddev->reconfig_mutex);
733 mutex_init(&mddev->bitmap_info.mutex);
734 INIT_LIST_HEAD(&mddev->disks);
735 INIT_LIST_HEAD(&mddev->all_mddevs);
736 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
737 atomic_set(&mddev->active, 1);
738 atomic_set(&mddev->openers, 0);
739 atomic_set(&mddev->active_io, 0);
740 spin_lock_init(&mddev->lock);
741 atomic_set(&mddev->flush_pending, 0);
742 init_waitqueue_head(&mddev->sb_wait);
743 init_waitqueue_head(&mddev->recovery_wait);
744 mddev->reshape_position = MaxSector;
745 mddev->reshape_backwards = 0;
746 mddev->last_sync_action = "none";
747 mddev->resync_min = 0;
748 mddev->resync_max = MaxSector;
749 mddev->level = LEVEL_NONE;
751 EXPORT_SYMBOL_GPL(mddev_init);
753 static struct mddev *mddev_find(dev_t unit)
755 struct mddev *mddev, *new = NULL;
757 if (unit && MAJOR(unit) != MD_MAJOR)
758 unit &= ~((1<<MdpMinorShift)-1);
761 spin_lock(&all_mddevs_lock);
764 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
765 if (mddev->unit == unit) {
767 spin_unlock(&all_mddevs_lock);
773 list_add(&new->all_mddevs, &all_mddevs);
774 spin_unlock(&all_mddevs_lock);
775 new->hold_active = UNTIL_IOCTL;
779 /* find an unused unit number */
780 static int next_minor = 512;
781 int start = next_minor;
785 dev = MKDEV(MD_MAJOR, next_minor);
787 if (next_minor > MINORMASK)
789 if (next_minor == start) {
790 /* Oh dear, all in use. */
791 spin_unlock(&all_mddevs_lock);
797 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
798 if (mddev->unit == dev) {
804 new->md_minor = MINOR(dev);
805 new->hold_active = UNTIL_STOP;
806 list_add(&new->all_mddevs, &all_mddevs);
807 spin_unlock(&all_mddevs_lock);
810 spin_unlock(&all_mddevs_lock);
812 new = kzalloc(sizeof(*new), GFP_KERNEL);
817 if (MAJOR(unit) == MD_MAJOR)
818 new->md_minor = MINOR(unit);
820 new->md_minor = MINOR(unit) >> MdpMinorShift;
827 static struct attribute_group md_redundancy_group;
829 void mddev_unlock(struct mddev *mddev)
831 if (mddev->to_remove) {
832 /* These cannot be removed under reconfig_mutex as
833 * an access to the files will try to take reconfig_mutex
834 * while holding the file unremovable, which leads to
836 * So hold set sysfs_active while the remove in happeing,
837 * and anything else which might set ->to_remove or my
838 * otherwise change the sysfs namespace will fail with
839 * -EBUSY if sysfs_active is still set.
840 * We set sysfs_active under reconfig_mutex and elsewhere
841 * test it under the same mutex to ensure its correct value
844 struct attribute_group *to_remove = mddev->to_remove;
845 mddev->to_remove = NULL;
846 mddev->sysfs_active = 1;
847 mutex_unlock(&mddev->reconfig_mutex);
849 if (mddev->kobj.sd) {
850 if (to_remove != &md_redundancy_group)
851 sysfs_remove_group(&mddev->kobj, to_remove);
852 if (mddev->pers == NULL ||
853 mddev->pers->sync_request == NULL) {
854 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
855 if (mddev->sysfs_action)
856 sysfs_put(mddev->sysfs_action);
857 if (mddev->sysfs_completed)
858 sysfs_put(mddev->sysfs_completed);
859 if (mddev->sysfs_degraded)
860 sysfs_put(mddev->sysfs_degraded);
861 mddev->sysfs_action = NULL;
862 mddev->sysfs_completed = NULL;
863 mddev->sysfs_degraded = NULL;
866 mddev->sysfs_active = 0;
868 mutex_unlock(&mddev->reconfig_mutex);
870 /* As we've dropped the mutex we need a spinlock to
871 * make sure the thread doesn't disappear
873 spin_lock(&pers_lock);
874 md_wakeup_thread(mddev->thread);
875 wake_up(&mddev->sb_wait);
876 spin_unlock(&pers_lock);
878 EXPORT_SYMBOL_GPL(mddev_unlock);
880 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
882 struct md_rdev *rdev;
884 rdev_for_each_rcu(rdev, mddev)
885 if (rdev->desc_nr == nr)
890 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
892 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
894 struct md_rdev *rdev;
896 rdev_for_each(rdev, mddev)
897 if (rdev->bdev->bd_dev == dev)
903 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
905 struct md_rdev *rdev;
907 rdev_for_each_rcu(rdev, mddev)
908 if (rdev->bdev->bd_dev == dev)
913 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
915 static struct md_personality *find_pers(int level, char *clevel)
917 struct md_personality *pers;
918 list_for_each_entry(pers, &pers_list, list) {
919 if (level != LEVEL_NONE && pers->level == level)
921 if (strcmp(pers->name, clevel)==0)
927 /* return the offset of the super block in 512byte sectors */
928 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
930 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
931 return MD_NEW_SIZE_SECTORS(num_sectors);
934 static int alloc_disk_sb(struct md_rdev *rdev)
936 rdev->sb_page = alloc_page(GFP_KERNEL);
942 void md_rdev_clear(struct md_rdev *rdev)
945 put_page(rdev->sb_page);
947 rdev->sb_page = NULL;
952 put_page(rdev->bb_page);
953 rdev->bb_page = NULL;
955 badblocks_exit(&rdev->badblocks);
957 EXPORT_SYMBOL_GPL(md_rdev_clear);
959 static void super_written(struct bio *bio)
961 struct md_rdev *rdev = bio->bi_private;
962 struct mddev *mddev = rdev->mddev;
964 if (bio->bi_status) {
965 pr_err("md: %s gets error=%d\n", __func__,
966 blk_status_to_errno(bio->bi_status));
967 md_error(mddev, rdev);
968 if (!test_bit(Faulty, &rdev->flags)
969 && (bio->bi_opf & MD_FAILFAST)) {
970 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
971 set_bit(LastDev, &rdev->flags);
974 clear_bit(LastDev, &rdev->flags);
976 if (atomic_dec_and_test(&mddev->pending_writes))
977 wake_up(&mddev->sb_wait);
978 rdev_dec_pending(rdev, mddev);
982 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
983 sector_t sector, int size, struct page *page)
985 /* write first size bytes of page to sector of rdev
986 * Increment mddev->pending_writes before returning
987 * and decrement it on completion, waking up sb_wait
988 * if zero is reached.
989 * If an error occurred, call md_error
997 if (test_bit(Faulty, &rdev->flags))
1000 bio = md_bio_alloc_sync(mddev);
1002 atomic_inc(&rdev->nr_pending);
1004 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
1005 bio->bi_iter.bi_sector = sector;
1006 bio_add_page(bio, page, size, 0);
1007 bio->bi_private = rdev;
1008 bio->bi_end_io = super_written;
1010 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1011 test_bit(FailFast, &rdev->flags) &&
1012 !test_bit(LastDev, &rdev->flags))
1014 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
1016 atomic_inc(&mddev->pending_writes);
1020 int md_super_wait(struct mddev *mddev)
1022 /* wait for all superblock writes that were scheduled to complete */
1023 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1024 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1029 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1030 struct page *page, int op, int op_flags, bool metadata_op)
1032 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1035 if (metadata_op && rdev->meta_bdev)
1036 bio_set_dev(bio, rdev->meta_bdev);
1038 bio_set_dev(bio, rdev->bdev);
1039 bio_set_op_attrs(bio, op, op_flags);
1041 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1042 else if (rdev->mddev->reshape_position != MaxSector &&
1043 (rdev->mddev->reshape_backwards ==
1044 (sector >= rdev->mddev->reshape_position)))
1045 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
1047 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1048 bio_add_page(bio, page, size, 0);
1050 submit_bio_wait(bio);
1052 ret = !bio->bi_status;
1056 EXPORT_SYMBOL_GPL(sync_page_io);
1058 static int read_disk_sb(struct md_rdev *rdev, int size)
1060 char b[BDEVNAME_SIZE];
1062 if (rdev->sb_loaded)
1065 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1067 rdev->sb_loaded = 1;
1071 pr_err("md: disabled device %s, could not read superblock.\n",
1072 bdevname(rdev->bdev,b));
1076 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1078 return sb1->set_uuid0 == sb2->set_uuid0 &&
1079 sb1->set_uuid1 == sb2->set_uuid1 &&
1080 sb1->set_uuid2 == sb2->set_uuid2 &&
1081 sb1->set_uuid3 == sb2->set_uuid3;
1084 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1087 mdp_super_t *tmp1, *tmp2;
1089 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1090 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1092 if (!tmp1 || !tmp2) {
1101 * nr_disks is not constant
1106 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1113 static u32 md_csum_fold(u32 csum)
1115 csum = (csum & 0xffff) + (csum >> 16);
1116 return (csum & 0xffff) + (csum >> 16);
1119 static unsigned int calc_sb_csum(mdp_super_t *sb)
1122 u32 *sb32 = (u32*)sb;
1124 unsigned int disk_csum, csum;
1126 disk_csum = sb->sb_csum;
1129 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1131 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1134 /* This used to use csum_partial, which was wrong for several
1135 * reasons including that different results are returned on
1136 * different architectures. It isn't critical that we get exactly
1137 * the same return value as before (we always csum_fold before
1138 * testing, and that removes any differences). However as we
1139 * know that csum_partial always returned a 16bit value on
1140 * alphas, do a fold to maximise conformity to previous behaviour.
1142 sb->sb_csum = md_csum_fold(disk_csum);
1144 sb->sb_csum = disk_csum;
1150 * Handle superblock details.
1151 * We want to be able to handle multiple superblock formats
1152 * so we have a common interface to them all, and an array of
1153 * different handlers.
1154 * We rely on user-space to write the initial superblock, and support
1155 * reading and updating of superblocks.
1156 * Interface methods are:
1157 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1158 * loads and validates a superblock on dev.
1159 * if refdev != NULL, compare superblocks on both devices
1161 * 0 - dev has a superblock that is compatible with refdev
1162 * 1 - dev has a superblock that is compatible and newer than refdev
1163 * so dev should be used as the refdev in future
1164 * -EINVAL superblock incompatible or invalid
1165 * -othererror e.g. -EIO
1167 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1168 * Verify that dev is acceptable into mddev.
1169 * The first time, mddev->raid_disks will be 0, and data from
1170 * dev should be merged in. Subsequent calls check that dev
1171 * is new enough. Return 0 or -EINVAL
1173 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1174 * Update the superblock for rdev with data in mddev
1175 * This does not write to disc.
1181 struct module *owner;
1182 int (*load_super)(struct md_rdev *rdev,
1183 struct md_rdev *refdev,
1185 int (*validate_super)(struct mddev *mddev,
1186 struct md_rdev *rdev);
1187 void (*sync_super)(struct mddev *mddev,
1188 struct md_rdev *rdev);
1189 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1190 sector_t num_sectors);
1191 int (*allow_new_offset)(struct md_rdev *rdev,
1192 unsigned long long new_offset);
1196 * Check that the given mddev has no bitmap.
1198 * This function is called from the run method of all personalities that do not
1199 * support bitmaps. It prints an error message and returns non-zero if mddev
1200 * has a bitmap. Otherwise, it returns 0.
1203 int md_check_no_bitmap(struct mddev *mddev)
1205 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1207 pr_warn("%s: bitmaps are not supported for %s\n",
1208 mdname(mddev), mddev->pers->name);
1211 EXPORT_SYMBOL(md_check_no_bitmap);
1214 * load_super for 0.90.0
1216 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1218 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1221 bool spare_disk = true;
1224 * Calculate the position of the superblock (512byte sectors),
1225 * it's at the end of the disk.
1227 * It also happens to be a multiple of 4Kb.
1229 rdev->sb_start = calc_dev_sboffset(rdev);
1231 ret = read_disk_sb(rdev, MD_SB_BYTES);
1237 bdevname(rdev->bdev, b);
1238 sb = page_address(rdev->sb_page);
1240 if (sb->md_magic != MD_SB_MAGIC) {
1241 pr_warn("md: invalid raid superblock magic on %s\n", b);
1245 if (sb->major_version != 0 ||
1246 sb->minor_version < 90 ||
1247 sb->minor_version > 91) {
1248 pr_warn("Bad version number %d.%d on %s\n",
1249 sb->major_version, sb->minor_version, b);
1253 if (sb->raid_disks <= 0)
1256 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1257 pr_warn("md: invalid superblock checksum on %s\n", b);
1261 rdev->preferred_minor = sb->md_minor;
1262 rdev->data_offset = 0;
1263 rdev->new_data_offset = 0;
1264 rdev->sb_size = MD_SB_BYTES;
1265 rdev->badblocks.shift = -1;
1267 if (sb->level == LEVEL_MULTIPATH)
1270 rdev->desc_nr = sb->this_disk.number;
1272 /* not spare disk, or LEVEL_MULTIPATH */
1273 if (sb->level == LEVEL_MULTIPATH ||
1274 (rdev->desc_nr >= 0 &&
1275 rdev->desc_nr < MD_SB_DISKS &&
1276 sb->disks[rdev->desc_nr].state &
1277 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1287 mdp_super_t *refsb = page_address(refdev->sb_page);
1288 if (!md_uuid_equal(refsb, sb)) {
1289 pr_warn("md: %s has different UUID to %s\n",
1290 b, bdevname(refdev->bdev,b2));
1293 if (!md_sb_equal(refsb, sb)) {
1294 pr_warn("md: %s has same UUID but different superblock to %s\n",
1295 b, bdevname(refdev->bdev, b2));
1299 ev2 = md_event(refsb);
1301 if (!spare_disk && ev1 > ev2)
1306 rdev->sectors = rdev->sb_start;
1307 /* Limit to 4TB as metadata cannot record more than that.
1308 * (not needed for Linear and RAID0 as metadata doesn't
1311 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1312 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1314 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1315 /* "this cannot possibly happen" ... */
1323 * validate_super for 0.90.0
1325 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1328 mdp_super_t *sb = page_address(rdev->sb_page);
1329 __u64 ev1 = md_event(sb);
1331 rdev->raid_disk = -1;
1332 clear_bit(Faulty, &rdev->flags);
1333 clear_bit(In_sync, &rdev->flags);
1334 clear_bit(Bitmap_sync, &rdev->flags);
1335 clear_bit(WriteMostly, &rdev->flags);
1337 if (mddev->raid_disks == 0) {
1338 mddev->major_version = 0;
1339 mddev->minor_version = sb->minor_version;
1340 mddev->patch_version = sb->patch_version;
1341 mddev->external = 0;
1342 mddev->chunk_sectors = sb->chunk_size >> 9;
1343 mddev->ctime = sb->ctime;
1344 mddev->utime = sb->utime;
1345 mddev->level = sb->level;
1346 mddev->clevel[0] = 0;
1347 mddev->layout = sb->layout;
1348 mddev->raid_disks = sb->raid_disks;
1349 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1350 mddev->events = ev1;
1351 mddev->bitmap_info.offset = 0;
1352 mddev->bitmap_info.space = 0;
1353 /* bitmap can use 60 K after the 4K superblocks */
1354 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1355 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1356 mddev->reshape_backwards = 0;
1358 if (mddev->minor_version >= 91) {
1359 mddev->reshape_position = sb->reshape_position;
1360 mddev->delta_disks = sb->delta_disks;
1361 mddev->new_level = sb->new_level;
1362 mddev->new_layout = sb->new_layout;
1363 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1364 if (mddev->delta_disks < 0)
1365 mddev->reshape_backwards = 1;
1367 mddev->reshape_position = MaxSector;
1368 mddev->delta_disks = 0;
1369 mddev->new_level = mddev->level;
1370 mddev->new_layout = mddev->layout;
1371 mddev->new_chunk_sectors = mddev->chunk_sectors;
1373 if (mddev->level == 0)
1376 if (sb->state & (1<<MD_SB_CLEAN))
1377 mddev->recovery_cp = MaxSector;
1379 if (sb->events_hi == sb->cp_events_hi &&
1380 sb->events_lo == sb->cp_events_lo) {
1381 mddev->recovery_cp = sb->recovery_cp;
1383 mddev->recovery_cp = 0;
1386 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1387 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1388 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1389 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1391 mddev->max_disks = MD_SB_DISKS;
1393 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1394 mddev->bitmap_info.file == NULL) {
1395 mddev->bitmap_info.offset =
1396 mddev->bitmap_info.default_offset;
1397 mddev->bitmap_info.space =
1398 mddev->bitmap_info.default_space;
1401 } else if (mddev->pers == NULL) {
1402 /* Insist on good event counter while assembling, except
1403 * for spares (which don't need an event count) */
1405 if (sb->disks[rdev->desc_nr].state & (
1406 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1407 if (ev1 < mddev->events)
1409 } else if (mddev->bitmap) {
1410 /* if adding to array with a bitmap, then we can accept an
1411 * older device ... but not too old.
1413 if (ev1 < mddev->bitmap->events_cleared)
1415 if (ev1 < mddev->events)
1416 set_bit(Bitmap_sync, &rdev->flags);
1418 if (ev1 < mddev->events)
1419 /* just a hot-add of a new device, leave raid_disk at -1 */
1423 if (mddev->level != LEVEL_MULTIPATH) {
1424 desc = sb->disks + rdev->desc_nr;
1426 if (desc->state & (1<<MD_DISK_FAULTY))
1427 set_bit(Faulty, &rdev->flags);
1428 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1429 desc->raid_disk < mddev->raid_disks */) {
1430 set_bit(In_sync, &rdev->flags);
1431 rdev->raid_disk = desc->raid_disk;
1432 rdev->saved_raid_disk = desc->raid_disk;
1433 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1434 /* active but not in sync implies recovery up to
1435 * reshape position. We don't know exactly where
1436 * that is, so set to zero for now */
1437 if (mddev->minor_version >= 91) {
1438 rdev->recovery_offset = 0;
1439 rdev->raid_disk = desc->raid_disk;
1442 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1443 set_bit(WriteMostly, &rdev->flags);
1444 if (desc->state & (1<<MD_DISK_FAILFAST))
1445 set_bit(FailFast, &rdev->flags);
1446 } else /* MULTIPATH are always insync */
1447 set_bit(In_sync, &rdev->flags);
1452 * sync_super for 0.90.0
1454 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1457 struct md_rdev *rdev2;
1458 int next_spare = mddev->raid_disks;
1460 /* make rdev->sb match mddev data..
1463 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1464 * 3/ any empty disks < next_spare become removed
1466 * disks[0] gets initialised to REMOVED because
1467 * we cannot be sure from other fields if it has
1468 * been initialised or not.
1471 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1473 rdev->sb_size = MD_SB_BYTES;
1475 sb = page_address(rdev->sb_page);
1477 memset(sb, 0, sizeof(*sb));
1479 sb->md_magic = MD_SB_MAGIC;
1480 sb->major_version = mddev->major_version;
1481 sb->patch_version = mddev->patch_version;
1482 sb->gvalid_words = 0; /* ignored */
1483 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1484 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1485 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1486 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1488 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1489 sb->level = mddev->level;
1490 sb->size = mddev->dev_sectors / 2;
1491 sb->raid_disks = mddev->raid_disks;
1492 sb->md_minor = mddev->md_minor;
1493 sb->not_persistent = 0;
1494 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1496 sb->events_hi = (mddev->events>>32);
1497 sb->events_lo = (u32)mddev->events;
1499 if (mddev->reshape_position == MaxSector)
1500 sb->minor_version = 90;
1502 sb->minor_version = 91;
1503 sb->reshape_position = mddev->reshape_position;
1504 sb->new_level = mddev->new_level;
1505 sb->delta_disks = mddev->delta_disks;
1506 sb->new_layout = mddev->new_layout;
1507 sb->new_chunk = mddev->new_chunk_sectors << 9;
1509 mddev->minor_version = sb->minor_version;
1512 sb->recovery_cp = mddev->recovery_cp;
1513 sb->cp_events_hi = (mddev->events>>32);
1514 sb->cp_events_lo = (u32)mddev->events;
1515 if (mddev->recovery_cp == MaxSector)
1516 sb->state = (1<< MD_SB_CLEAN);
1518 sb->recovery_cp = 0;
1520 sb->layout = mddev->layout;
1521 sb->chunk_size = mddev->chunk_sectors << 9;
1523 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1524 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1526 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1527 rdev_for_each(rdev2, mddev) {
1530 int is_active = test_bit(In_sync, &rdev2->flags);
1532 if (rdev2->raid_disk >= 0 &&
1533 sb->minor_version >= 91)
1534 /* we have nowhere to store the recovery_offset,
1535 * but if it is not below the reshape_position,
1536 * we can piggy-back on that.
1539 if (rdev2->raid_disk < 0 ||
1540 test_bit(Faulty, &rdev2->flags))
1543 desc_nr = rdev2->raid_disk;
1545 desc_nr = next_spare++;
1546 rdev2->desc_nr = desc_nr;
1547 d = &sb->disks[rdev2->desc_nr];
1549 d->number = rdev2->desc_nr;
1550 d->major = MAJOR(rdev2->bdev->bd_dev);
1551 d->minor = MINOR(rdev2->bdev->bd_dev);
1553 d->raid_disk = rdev2->raid_disk;
1555 d->raid_disk = rdev2->desc_nr; /* compatibility */
1556 if (test_bit(Faulty, &rdev2->flags))
1557 d->state = (1<<MD_DISK_FAULTY);
1558 else if (is_active) {
1559 d->state = (1<<MD_DISK_ACTIVE);
1560 if (test_bit(In_sync, &rdev2->flags))
1561 d->state |= (1<<MD_DISK_SYNC);
1569 if (test_bit(WriteMostly, &rdev2->flags))
1570 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1571 if (test_bit(FailFast, &rdev2->flags))
1572 d->state |= (1<<MD_DISK_FAILFAST);
1574 /* now set the "removed" and "faulty" bits on any missing devices */
1575 for (i=0 ; i < mddev->raid_disks ; i++) {
1576 mdp_disk_t *d = &sb->disks[i];
1577 if (d->state == 0 && d->number == 0) {
1580 d->state = (1<<MD_DISK_REMOVED);
1581 d->state |= (1<<MD_DISK_FAULTY);
1585 sb->nr_disks = nr_disks;
1586 sb->active_disks = active;
1587 sb->working_disks = working;
1588 sb->failed_disks = failed;
1589 sb->spare_disks = spare;
1591 sb->this_disk = sb->disks[rdev->desc_nr];
1592 sb->sb_csum = calc_sb_csum(sb);
1596 * rdev_size_change for 0.90.0
1598 static unsigned long long
1599 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1601 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1602 return 0; /* component must fit device */
1603 if (rdev->mddev->bitmap_info.offset)
1604 return 0; /* can't move bitmap */
1605 rdev->sb_start = calc_dev_sboffset(rdev);
1606 if (!num_sectors || num_sectors > rdev->sb_start)
1607 num_sectors = rdev->sb_start;
1608 /* Limit to 4TB as metadata cannot record more than that.
1609 * 4TB == 2^32 KB, or 2*2^32 sectors.
1611 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1612 num_sectors = (sector_t)(2ULL << 32) - 2;
1614 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1616 } while (md_super_wait(rdev->mddev) < 0);
1621 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1623 /* non-zero offset changes not possible with v0.90 */
1624 return new_offset == 0;
1628 * version 1 superblock
1631 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1635 unsigned long long newcsum;
1636 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1637 __le32 *isuper = (__le32*)sb;
1639 disk_csum = sb->sb_csum;
1642 for (; size >= 4; size -= 4)
1643 newcsum += le32_to_cpu(*isuper++);
1646 newcsum += le16_to_cpu(*(__le16*) isuper);
1648 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1649 sb->sb_csum = disk_csum;
1650 return cpu_to_le32(csum);
1653 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1655 struct mdp_superblock_1 *sb;
1659 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1661 bool spare_disk = true;
1664 * Calculate the position of the superblock in 512byte sectors.
1665 * It is always aligned to a 4K boundary and
1666 * depeding on minor_version, it can be:
1667 * 0: At least 8K, but less than 12K, from end of device
1668 * 1: At start of device
1669 * 2: 4K from start of device.
1671 switch(minor_version) {
1673 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1675 sb_start &= ~(sector_t)(4*2-1);
1686 rdev->sb_start = sb_start;
1688 /* superblock is rarely larger than 1K, but it can be larger,
1689 * and it is safe to read 4k, so we do that
1691 ret = read_disk_sb(rdev, 4096);
1692 if (ret) return ret;
1694 sb = page_address(rdev->sb_page);
1696 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1697 sb->major_version != cpu_to_le32(1) ||
1698 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1699 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1700 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1703 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1704 pr_warn("md: invalid superblock checksum on %s\n",
1705 bdevname(rdev->bdev,b));
1708 if (le64_to_cpu(sb->data_size) < 10) {
1709 pr_warn("md: data_size too small on %s\n",
1710 bdevname(rdev->bdev,b));
1715 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1716 /* Some padding is non-zero, might be a new feature */
1719 rdev->preferred_minor = 0xffff;
1720 rdev->data_offset = le64_to_cpu(sb->data_offset);
1721 rdev->new_data_offset = rdev->data_offset;
1722 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1723 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1724 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1725 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1727 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1728 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1729 if (rdev->sb_size & bmask)
1730 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1733 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1736 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1739 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1742 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1744 if (!rdev->bb_page) {
1745 rdev->bb_page = alloc_page(GFP_KERNEL);
1749 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1750 rdev->badblocks.count == 0) {
1751 /* need to load the bad block list.
1752 * Currently we limit it to one page.
1758 int sectors = le16_to_cpu(sb->bblog_size);
1759 if (sectors > (PAGE_SIZE / 512))
1761 offset = le32_to_cpu(sb->bblog_offset);
1764 bb_sector = (long long)offset;
1765 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1766 rdev->bb_page, REQ_OP_READ, 0, true))
1768 bbp = (__le64 *)page_address(rdev->bb_page);
1769 rdev->badblocks.shift = sb->bblog_shift;
1770 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1771 u64 bb = le64_to_cpu(*bbp);
1772 int count = bb & (0x3ff);
1773 u64 sector = bb >> 10;
1774 sector <<= sb->bblog_shift;
1775 count <<= sb->bblog_shift;
1778 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1781 } else if (sb->bblog_offset != 0)
1782 rdev->badblocks.shift = 0;
1784 if ((le32_to_cpu(sb->feature_map) &
1785 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1786 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1787 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1788 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1791 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1795 /* not spare disk, or LEVEL_MULTIPATH */
1796 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1797 (rdev->desc_nr >= 0 &&
1798 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1799 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1800 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1810 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1812 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1813 sb->level != refsb->level ||
1814 sb->layout != refsb->layout ||
1815 sb->chunksize != refsb->chunksize) {
1816 pr_warn("md: %s has strangely different superblock to %s\n",
1817 bdevname(rdev->bdev,b),
1818 bdevname(refdev->bdev,b2));
1821 ev1 = le64_to_cpu(sb->events);
1822 ev2 = le64_to_cpu(refsb->events);
1824 if (!spare_disk && ev1 > ev2)
1829 if (minor_version) {
1830 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1831 sectors -= rdev->data_offset;
1833 sectors = rdev->sb_start;
1834 if (sectors < le64_to_cpu(sb->data_size))
1836 rdev->sectors = le64_to_cpu(sb->data_size);
1840 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1842 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1843 __u64 ev1 = le64_to_cpu(sb->events);
1845 rdev->raid_disk = -1;
1846 clear_bit(Faulty, &rdev->flags);
1847 clear_bit(In_sync, &rdev->flags);
1848 clear_bit(Bitmap_sync, &rdev->flags);
1849 clear_bit(WriteMostly, &rdev->flags);
1851 if (mddev->raid_disks == 0) {
1852 mddev->major_version = 1;
1853 mddev->patch_version = 0;
1854 mddev->external = 0;
1855 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1856 mddev->ctime = le64_to_cpu(sb->ctime);
1857 mddev->utime = le64_to_cpu(sb->utime);
1858 mddev->level = le32_to_cpu(sb->level);
1859 mddev->clevel[0] = 0;
1860 mddev->layout = le32_to_cpu(sb->layout);
1861 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1862 mddev->dev_sectors = le64_to_cpu(sb->size);
1863 mddev->events = ev1;
1864 mddev->bitmap_info.offset = 0;
1865 mddev->bitmap_info.space = 0;
1866 /* Default location for bitmap is 1K after superblock
1867 * using 3K - total of 4K
1869 mddev->bitmap_info.default_offset = 1024 >> 9;
1870 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1871 mddev->reshape_backwards = 0;
1873 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1874 memcpy(mddev->uuid, sb->set_uuid, 16);
1876 mddev->max_disks = (4096-256)/2;
1878 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1879 mddev->bitmap_info.file == NULL) {
1880 mddev->bitmap_info.offset =
1881 (__s32)le32_to_cpu(sb->bitmap_offset);
1882 /* Metadata doesn't record how much space is available.
1883 * For 1.0, we assume we can use up to the superblock
1884 * if before, else to 4K beyond superblock.
1885 * For others, assume no change is possible.
1887 if (mddev->minor_version > 0)
1888 mddev->bitmap_info.space = 0;
1889 else if (mddev->bitmap_info.offset > 0)
1890 mddev->bitmap_info.space =
1891 8 - mddev->bitmap_info.offset;
1893 mddev->bitmap_info.space =
1894 -mddev->bitmap_info.offset;
1897 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1898 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1899 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1900 mddev->new_level = le32_to_cpu(sb->new_level);
1901 mddev->new_layout = le32_to_cpu(sb->new_layout);
1902 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1903 if (mddev->delta_disks < 0 ||
1904 (mddev->delta_disks == 0 &&
1905 (le32_to_cpu(sb->feature_map)
1906 & MD_FEATURE_RESHAPE_BACKWARDS)))
1907 mddev->reshape_backwards = 1;
1909 mddev->reshape_position = MaxSector;
1910 mddev->delta_disks = 0;
1911 mddev->new_level = mddev->level;
1912 mddev->new_layout = mddev->layout;
1913 mddev->new_chunk_sectors = mddev->chunk_sectors;
1916 if (mddev->level == 0 &&
1917 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1920 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1921 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1923 if (le32_to_cpu(sb->feature_map) &
1924 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1925 if (le32_to_cpu(sb->feature_map) &
1926 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1928 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1929 (le32_to_cpu(sb->feature_map) &
1930 MD_FEATURE_MULTIPLE_PPLS))
1932 set_bit(MD_HAS_PPL, &mddev->flags);
1934 } else if (mddev->pers == NULL) {
1935 /* Insist of good event counter while assembling, except for
1936 * spares (which don't need an event count) */
1938 if (rdev->desc_nr >= 0 &&
1939 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1940 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1941 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1942 if (ev1 < mddev->events)
1944 } else if (mddev->bitmap) {
1945 /* If adding to array with a bitmap, then we can accept an
1946 * older device, but not too old.
1948 if (ev1 < mddev->bitmap->events_cleared)
1950 if (ev1 < mddev->events)
1951 set_bit(Bitmap_sync, &rdev->flags);
1953 if (ev1 < mddev->events)
1954 /* just a hot-add of a new device, leave raid_disk at -1 */
1957 if (mddev->level != LEVEL_MULTIPATH) {
1959 if (rdev->desc_nr < 0 ||
1960 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1961 role = MD_DISK_ROLE_SPARE;
1964 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1966 case MD_DISK_ROLE_SPARE: /* spare */
1968 case MD_DISK_ROLE_FAULTY: /* faulty */
1969 set_bit(Faulty, &rdev->flags);
1971 case MD_DISK_ROLE_JOURNAL: /* journal device */
1972 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1973 /* journal device without journal feature */
1974 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1977 set_bit(Journal, &rdev->flags);
1978 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1979 rdev->raid_disk = 0;
1982 rdev->saved_raid_disk = role;
1983 if ((le32_to_cpu(sb->feature_map) &
1984 MD_FEATURE_RECOVERY_OFFSET)) {
1985 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1986 if (!(le32_to_cpu(sb->feature_map) &
1987 MD_FEATURE_RECOVERY_BITMAP))
1988 rdev->saved_raid_disk = -1;
1991 * If the array is FROZEN, then the device can't
1992 * be in_sync with rest of array.
1994 if (!test_bit(MD_RECOVERY_FROZEN,
1996 set_bit(In_sync, &rdev->flags);
1998 rdev->raid_disk = role;
2001 if (sb->devflags & WriteMostly1)
2002 set_bit(WriteMostly, &rdev->flags);
2003 if (sb->devflags & FailFast1)
2004 set_bit(FailFast, &rdev->flags);
2005 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2006 set_bit(Replacement, &rdev->flags);
2007 } else /* MULTIPATH are always insync */
2008 set_bit(In_sync, &rdev->flags);
2013 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2015 struct mdp_superblock_1 *sb;
2016 struct md_rdev *rdev2;
2018 /* make rdev->sb match mddev and rdev data. */
2020 sb = page_address(rdev->sb_page);
2022 sb->feature_map = 0;
2024 sb->recovery_offset = cpu_to_le64(0);
2025 memset(sb->pad3, 0, sizeof(sb->pad3));
2027 sb->utime = cpu_to_le64((__u64)mddev->utime);
2028 sb->events = cpu_to_le64(mddev->events);
2030 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2031 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2032 sb->resync_offset = cpu_to_le64(MaxSector);
2034 sb->resync_offset = cpu_to_le64(0);
2036 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2038 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2039 sb->size = cpu_to_le64(mddev->dev_sectors);
2040 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2041 sb->level = cpu_to_le32(mddev->level);
2042 sb->layout = cpu_to_le32(mddev->layout);
2043 if (test_bit(FailFast, &rdev->flags))
2044 sb->devflags |= FailFast1;
2046 sb->devflags &= ~FailFast1;
2048 if (test_bit(WriteMostly, &rdev->flags))
2049 sb->devflags |= WriteMostly1;
2051 sb->devflags &= ~WriteMostly1;
2052 sb->data_offset = cpu_to_le64(rdev->data_offset);
2053 sb->data_size = cpu_to_le64(rdev->sectors);
2055 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2056 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2057 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2060 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2061 !test_bit(In_sync, &rdev->flags)) {
2063 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2064 sb->recovery_offset =
2065 cpu_to_le64(rdev->recovery_offset);
2066 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2068 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2070 /* Note: recovery_offset and journal_tail share space */
2071 if (test_bit(Journal, &rdev->flags))
2072 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2073 if (test_bit(Replacement, &rdev->flags))
2075 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2077 if (mddev->reshape_position != MaxSector) {
2078 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2079 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2080 sb->new_layout = cpu_to_le32(mddev->new_layout);
2081 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2082 sb->new_level = cpu_to_le32(mddev->new_level);
2083 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2084 if (mddev->delta_disks == 0 &&
2085 mddev->reshape_backwards)
2087 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2088 if (rdev->new_data_offset != rdev->data_offset) {
2090 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2091 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2092 - rdev->data_offset));
2096 if (mddev_is_clustered(mddev))
2097 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2099 if (rdev->badblocks.count == 0)
2100 /* Nothing to do for bad blocks*/ ;
2101 else if (sb->bblog_offset == 0)
2102 /* Cannot record bad blocks on this device */
2103 md_error(mddev, rdev);
2105 struct badblocks *bb = &rdev->badblocks;
2106 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2108 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2113 seq = read_seqbegin(&bb->lock);
2115 memset(bbp, 0xff, PAGE_SIZE);
2117 for (i = 0 ; i < bb->count ; i++) {
2118 u64 internal_bb = p[i];
2119 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2120 | BB_LEN(internal_bb));
2121 bbp[i] = cpu_to_le64(store_bb);
2124 if (read_seqretry(&bb->lock, seq))
2127 bb->sector = (rdev->sb_start +
2128 (int)le32_to_cpu(sb->bblog_offset));
2129 bb->size = le16_to_cpu(sb->bblog_size);
2134 rdev_for_each(rdev2, mddev)
2135 if (rdev2->desc_nr+1 > max_dev)
2136 max_dev = rdev2->desc_nr+1;
2138 if (max_dev > le32_to_cpu(sb->max_dev)) {
2140 sb->max_dev = cpu_to_le32(max_dev);
2141 rdev->sb_size = max_dev * 2 + 256;
2142 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2143 if (rdev->sb_size & bmask)
2144 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2146 max_dev = le32_to_cpu(sb->max_dev);
2148 for (i=0; i<max_dev;i++)
2149 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2151 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2152 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2154 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2155 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2157 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2159 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2160 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2161 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2164 rdev_for_each(rdev2, mddev) {
2166 if (test_bit(Faulty, &rdev2->flags))
2167 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2168 else if (test_bit(In_sync, &rdev2->flags))
2169 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2170 else if (test_bit(Journal, &rdev2->flags))
2171 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2172 else if (rdev2->raid_disk >= 0)
2173 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2175 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2178 sb->sb_csum = calc_sb_1_csum(sb);
2181 static sector_t super_1_choose_bm_space(sector_t dev_size)
2185 /* if the device is bigger than 8Gig, save 64k for bitmap
2186 * usage, if bigger than 200Gig, save 128k
2188 if (dev_size < 64*2)
2190 else if (dev_size - 64*2 >= 200*1024*1024*2)
2192 else if (dev_size - 4*2 > 8*1024*1024*2)
2199 static unsigned long long
2200 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2202 struct mdp_superblock_1 *sb;
2203 sector_t max_sectors;
2204 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2205 return 0; /* component must fit device */
2206 if (rdev->data_offset != rdev->new_data_offset)
2207 return 0; /* too confusing */
2208 if (rdev->sb_start < rdev->data_offset) {
2209 /* minor versions 1 and 2; superblock before data */
2210 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2211 max_sectors -= rdev->data_offset;
2212 if (!num_sectors || num_sectors > max_sectors)
2213 num_sectors = max_sectors;
2214 } else if (rdev->mddev->bitmap_info.offset) {
2215 /* minor version 0 with bitmap we can't move */
2218 /* minor version 0; superblock after data */
2219 sector_t sb_start, bm_space;
2220 sector_t dev_size = i_size_read(rdev->bdev->bd_inode) >> 9;
2222 /* 8K is for superblock */
2223 sb_start = dev_size - 8*2;
2224 sb_start &= ~(sector_t)(4*2 - 1);
2226 bm_space = super_1_choose_bm_space(dev_size);
2228 /* Space that can be used to store date needs to decrease
2229 * superblock bitmap space and bad block space(4K)
2231 max_sectors = sb_start - bm_space - 4*2;
2233 if (!num_sectors || num_sectors > max_sectors)
2234 num_sectors = max_sectors;
2236 sb = page_address(rdev->sb_page);
2237 sb->data_size = cpu_to_le64(num_sectors);
2238 sb->super_offset = cpu_to_le64(rdev->sb_start);
2239 sb->sb_csum = calc_sb_1_csum(sb);
2241 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2243 } while (md_super_wait(rdev->mddev) < 0);
2249 super_1_allow_new_offset(struct md_rdev *rdev,
2250 unsigned long long new_offset)
2252 /* All necessary checks on new >= old have been done */
2253 struct bitmap *bitmap;
2254 if (new_offset >= rdev->data_offset)
2257 /* with 1.0 metadata, there is no metadata to tread on
2258 * so we can always move back */
2259 if (rdev->mddev->minor_version == 0)
2262 /* otherwise we must be sure not to step on
2263 * any metadata, so stay:
2264 * 36K beyond start of superblock
2265 * beyond end of badblocks
2266 * beyond write-intent bitmap
2268 if (rdev->sb_start + (32+4)*2 > new_offset)
2270 bitmap = rdev->mddev->bitmap;
2271 if (bitmap && !rdev->mddev->bitmap_info.file &&
2272 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2273 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2275 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2281 static struct super_type super_types[] = {
2284 .owner = THIS_MODULE,
2285 .load_super = super_90_load,
2286 .validate_super = super_90_validate,
2287 .sync_super = super_90_sync,
2288 .rdev_size_change = super_90_rdev_size_change,
2289 .allow_new_offset = super_90_allow_new_offset,
2293 .owner = THIS_MODULE,
2294 .load_super = super_1_load,
2295 .validate_super = super_1_validate,
2296 .sync_super = super_1_sync,
2297 .rdev_size_change = super_1_rdev_size_change,
2298 .allow_new_offset = super_1_allow_new_offset,
2302 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2304 if (mddev->sync_super) {
2305 mddev->sync_super(mddev, rdev);
2309 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2311 super_types[mddev->major_version].sync_super(mddev, rdev);
2314 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2316 struct md_rdev *rdev, *rdev2;
2319 rdev_for_each_rcu(rdev, mddev1) {
2320 if (test_bit(Faulty, &rdev->flags) ||
2321 test_bit(Journal, &rdev->flags) ||
2322 rdev->raid_disk == -1)
2324 rdev_for_each_rcu(rdev2, mddev2) {
2325 if (test_bit(Faulty, &rdev2->flags) ||
2326 test_bit(Journal, &rdev2->flags) ||
2327 rdev2->raid_disk == -1)
2329 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2339 static LIST_HEAD(pending_raid_disks);
2342 * Try to register data integrity profile for an mddev
2344 * This is called when an array is started and after a disk has been kicked
2345 * from the array. It only succeeds if all working and active component devices
2346 * are integrity capable with matching profiles.
2348 int md_integrity_register(struct mddev *mddev)
2350 struct md_rdev *rdev, *reference = NULL;
2352 if (list_empty(&mddev->disks))
2353 return 0; /* nothing to do */
2354 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2355 return 0; /* shouldn't register, or already is */
2356 rdev_for_each(rdev, mddev) {
2357 /* skip spares and non-functional disks */
2358 if (test_bit(Faulty, &rdev->flags))
2360 if (rdev->raid_disk < 0)
2363 /* Use the first rdev as the reference */
2367 /* does this rdev's profile match the reference profile? */
2368 if (blk_integrity_compare(reference->bdev->bd_disk,
2369 rdev->bdev->bd_disk) < 0)
2372 if (!reference || !bdev_get_integrity(reference->bdev))
2375 * All component devices are integrity capable and have matching
2376 * profiles, register the common profile for the md device.
2378 blk_integrity_register(mddev->gendisk,
2379 bdev_get_integrity(reference->bdev));
2381 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2382 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2383 pr_err("md: failed to create integrity pool for %s\n",
2389 EXPORT_SYMBOL(md_integrity_register);
2392 * Attempt to add an rdev, but only if it is consistent with the current
2395 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2397 struct blk_integrity *bi_mddev;
2398 char name[BDEVNAME_SIZE];
2400 if (!mddev->gendisk)
2403 bi_mddev = blk_get_integrity(mddev->gendisk);
2405 if (!bi_mddev) /* nothing to do */
2408 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2409 pr_err("%s: incompatible integrity profile for %s\n",
2410 mdname(mddev), bdevname(rdev->bdev, name));
2416 EXPORT_SYMBOL(md_integrity_add_rdev);
2418 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2420 char b[BDEVNAME_SIZE];
2424 /* prevent duplicates */
2425 if (find_rdev(mddev, rdev->bdev->bd_dev))
2428 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2432 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2433 if (!test_bit(Journal, &rdev->flags) &&
2435 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2437 /* Cannot change size, so fail
2438 * If mddev->level <= 0, then we don't care
2439 * about aligning sizes (e.g. linear)
2441 if (mddev->level > 0)
2444 mddev->dev_sectors = rdev->sectors;
2447 /* Verify rdev->desc_nr is unique.
2448 * If it is -1, assign a free number, else
2449 * check number is not in use
2452 if (rdev->desc_nr < 0) {
2455 choice = mddev->raid_disks;
2456 while (md_find_rdev_nr_rcu(mddev, choice))
2458 rdev->desc_nr = choice;
2460 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2466 if (!test_bit(Journal, &rdev->flags) &&
2467 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2468 pr_warn("md: %s: array is limited to %d devices\n",
2469 mdname(mddev), mddev->max_disks);
2472 bdevname(rdev->bdev,b);
2473 strreplace(b, '/', '!');
2475 rdev->mddev = mddev;
2476 pr_debug("md: bind<%s>\n", b);
2478 if (mddev->raid_disks)
2479 mddev_create_serial_pool(mddev, rdev, false);
2481 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2484 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2485 /* failure here is OK */
2486 err = sysfs_create_link(&rdev->kobj, ko, "block");
2487 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2488 rdev->sysfs_unack_badblocks =
2489 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2490 rdev->sysfs_badblocks =
2491 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2493 list_add_rcu(&rdev->same_set, &mddev->disks);
2494 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2496 /* May as well allow recovery to be retried once */
2497 mddev->recovery_disabled++;
2502 pr_warn("md: failed to register dev-%s for %s\n",
2507 static void rdev_delayed_delete(struct work_struct *ws)
2509 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2510 kobject_del(&rdev->kobj);
2511 kobject_put(&rdev->kobj);
2514 static void unbind_rdev_from_array(struct md_rdev *rdev)
2516 char b[BDEVNAME_SIZE];
2518 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2519 list_del_rcu(&rdev->same_set);
2520 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2521 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2523 sysfs_remove_link(&rdev->kobj, "block");
2524 sysfs_put(rdev->sysfs_state);
2525 sysfs_put(rdev->sysfs_unack_badblocks);
2526 sysfs_put(rdev->sysfs_badblocks);
2527 rdev->sysfs_state = NULL;
2528 rdev->sysfs_unack_badblocks = NULL;
2529 rdev->sysfs_badblocks = NULL;
2530 rdev->badblocks.count = 0;
2531 /* We need to delay this, otherwise we can deadlock when
2532 * writing to 'remove' to "dev/state". We also need
2533 * to delay it due to rcu usage.
2536 INIT_WORK(&rdev->del_work, rdev_delayed_delete);
2537 kobject_get(&rdev->kobj);
2538 queue_work(md_rdev_misc_wq, &rdev->del_work);
2542 * prevent the device from being mounted, repartitioned or
2543 * otherwise reused by a RAID array (or any other kernel
2544 * subsystem), by bd_claiming the device.
2546 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2549 struct block_device *bdev;
2551 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2552 shared ? (struct md_rdev *)lock_rdev : rdev);
2554 pr_warn("md: could not open device unknown-block(%u,%u).\n",
2555 MAJOR(dev), MINOR(dev));
2556 return PTR_ERR(bdev);
2562 static void unlock_rdev(struct md_rdev *rdev)
2564 struct block_device *bdev = rdev->bdev;
2566 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2569 void md_autodetect_dev(dev_t dev);
2571 static void export_rdev(struct md_rdev *rdev)
2573 char b[BDEVNAME_SIZE];
2575 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2576 md_rdev_clear(rdev);
2578 if (test_bit(AutoDetected, &rdev->flags))
2579 md_autodetect_dev(rdev->bdev->bd_dev);
2582 kobject_put(&rdev->kobj);
2585 void md_kick_rdev_from_array(struct md_rdev *rdev)
2587 unbind_rdev_from_array(rdev);
2590 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2592 static void export_array(struct mddev *mddev)
2594 struct md_rdev *rdev;
2596 while (!list_empty(&mddev->disks)) {
2597 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2599 md_kick_rdev_from_array(rdev);
2601 mddev->raid_disks = 0;
2602 mddev->major_version = 0;
2605 static bool set_in_sync(struct mddev *mddev)
2607 lockdep_assert_held(&mddev->lock);
2608 if (!mddev->in_sync) {
2609 mddev->sync_checkers++;
2610 spin_unlock(&mddev->lock);
2611 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2612 spin_lock(&mddev->lock);
2613 if (!mddev->in_sync &&
2614 percpu_ref_is_zero(&mddev->writes_pending)) {
2617 * Ensure ->in_sync is visible before we clear
2621 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2622 sysfs_notify_dirent_safe(mddev->sysfs_state);
2624 if (--mddev->sync_checkers == 0)
2625 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2627 if (mddev->safemode == 1)
2628 mddev->safemode = 0;
2629 return mddev->in_sync;
2632 static void sync_sbs(struct mddev *mddev, int nospares)
2634 /* Update each superblock (in-memory image), but
2635 * if we are allowed to, skip spares which already
2636 * have the right event counter, or have one earlier
2637 * (which would mean they aren't being marked as dirty
2638 * with the rest of the array)
2640 struct md_rdev *rdev;
2641 rdev_for_each(rdev, mddev) {
2642 if (rdev->sb_events == mddev->events ||
2644 rdev->raid_disk < 0 &&
2645 rdev->sb_events+1 == mddev->events)) {
2646 /* Don't update this superblock */
2647 rdev->sb_loaded = 2;
2649 sync_super(mddev, rdev);
2650 rdev->sb_loaded = 1;
2655 static bool does_sb_need_changing(struct mddev *mddev)
2657 struct md_rdev *rdev;
2658 struct mdp_superblock_1 *sb;
2661 /* Find a good rdev */
2662 rdev_for_each(rdev, mddev)
2663 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2666 /* No good device found. */
2670 sb = page_address(rdev->sb_page);
2671 /* Check if a device has become faulty or a spare become active */
2672 rdev_for_each(rdev, mddev) {
2673 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2674 /* Device activated? */
2675 if (role == 0xffff && rdev->raid_disk >=0 &&
2676 !test_bit(Faulty, &rdev->flags))
2678 /* Device turned faulty? */
2679 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2683 /* Check if any mddev parameters have changed */
2684 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2685 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2686 (mddev->layout != le32_to_cpu(sb->layout)) ||
2687 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2688 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2694 void md_update_sb(struct mddev *mddev, int force_change)
2696 struct md_rdev *rdev;
2699 int any_badblocks_changed = 0;
2704 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2709 if (mddev_is_clustered(mddev)) {
2710 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2712 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2714 ret = md_cluster_ops->metadata_update_start(mddev);
2715 /* Has someone else has updated the sb */
2716 if (!does_sb_need_changing(mddev)) {
2718 md_cluster_ops->metadata_update_cancel(mddev);
2719 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2720 BIT(MD_SB_CHANGE_DEVS) |
2721 BIT(MD_SB_CHANGE_CLEAN));
2727 * First make sure individual recovery_offsets are correct
2728 * curr_resync_completed can only be used during recovery.
2729 * During reshape/resync it might use array-addresses rather
2730 * that device addresses.
2732 rdev_for_each(rdev, mddev) {
2733 if (rdev->raid_disk >= 0 &&
2734 mddev->delta_disks >= 0 &&
2735 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2736 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2737 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2738 !test_bit(Journal, &rdev->flags) &&
2739 !test_bit(In_sync, &rdev->flags) &&
2740 mddev->curr_resync_completed > rdev->recovery_offset)
2741 rdev->recovery_offset = mddev->curr_resync_completed;
2744 if (!mddev->persistent) {
2745 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2746 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2747 if (!mddev->external) {
2748 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2749 rdev_for_each(rdev, mddev) {
2750 if (rdev->badblocks.changed) {
2751 rdev->badblocks.changed = 0;
2752 ack_all_badblocks(&rdev->badblocks);
2753 md_error(mddev, rdev);
2755 clear_bit(Blocked, &rdev->flags);
2756 clear_bit(BlockedBadBlocks, &rdev->flags);
2757 wake_up(&rdev->blocked_wait);
2760 wake_up(&mddev->sb_wait);
2764 spin_lock(&mddev->lock);
2766 mddev->utime = ktime_get_real_seconds();
2768 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2770 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2771 /* just a clean<-> dirty transition, possibly leave spares alone,
2772 * though if events isn't the right even/odd, we will have to do
2778 if (mddev->degraded)
2779 /* If the array is degraded, then skipping spares is both
2780 * dangerous and fairly pointless.
2781 * Dangerous because a device that was removed from the array
2782 * might have a event_count that still looks up-to-date,
2783 * so it can be re-added without a resync.
2784 * Pointless because if there are any spares to skip,
2785 * then a recovery will happen and soon that array won't
2786 * be degraded any more and the spare can go back to sleep then.
2790 sync_req = mddev->in_sync;
2792 /* If this is just a dirty<->clean transition, and the array is clean
2793 * and 'events' is odd, we can roll back to the previous clean state */
2795 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2796 && mddev->can_decrease_events
2797 && mddev->events != 1) {
2799 mddev->can_decrease_events = 0;
2801 /* otherwise we have to go forward and ... */
2803 mddev->can_decrease_events = nospares;
2807 * This 64-bit counter should never wrap.
2808 * Either we are in around ~1 trillion A.C., assuming
2809 * 1 reboot per second, or we have a bug...
2811 WARN_ON(mddev->events == 0);
2813 rdev_for_each(rdev, mddev) {
2814 if (rdev->badblocks.changed)
2815 any_badblocks_changed++;
2816 if (test_bit(Faulty, &rdev->flags))
2817 set_bit(FaultRecorded, &rdev->flags);
2820 sync_sbs(mddev, nospares);
2821 spin_unlock(&mddev->lock);
2823 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2824 mdname(mddev), mddev->in_sync);
2827 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2829 md_bitmap_update_sb(mddev->bitmap);
2830 rdev_for_each(rdev, mddev) {
2831 char b[BDEVNAME_SIZE];
2833 if (rdev->sb_loaded != 1)
2834 continue; /* no noise on spare devices */
2836 if (!test_bit(Faulty, &rdev->flags)) {
2837 md_super_write(mddev,rdev,
2838 rdev->sb_start, rdev->sb_size,
2840 pr_debug("md: (write) %s's sb offset: %llu\n",
2841 bdevname(rdev->bdev, b),
2842 (unsigned long long)rdev->sb_start);
2843 rdev->sb_events = mddev->events;
2844 if (rdev->badblocks.size) {
2845 md_super_write(mddev, rdev,
2846 rdev->badblocks.sector,
2847 rdev->badblocks.size << 9,
2849 rdev->badblocks.size = 0;
2853 pr_debug("md: %s (skipping faulty)\n",
2854 bdevname(rdev->bdev, b));
2856 if (mddev->level == LEVEL_MULTIPATH)
2857 /* only need to write one superblock... */
2860 if (md_super_wait(mddev) < 0)
2862 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2864 if (mddev_is_clustered(mddev) && ret == 0)
2865 md_cluster_ops->metadata_update_finish(mddev);
2867 if (mddev->in_sync != sync_req ||
2868 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2869 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2870 /* have to write it out again */
2872 wake_up(&mddev->sb_wait);
2873 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2874 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2876 rdev_for_each(rdev, mddev) {
2877 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2878 clear_bit(Blocked, &rdev->flags);
2880 if (any_badblocks_changed)
2881 ack_all_badblocks(&rdev->badblocks);
2882 clear_bit(BlockedBadBlocks, &rdev->flags);
2883 wake_up(&rdev->blocked_wait);
2886 EXPORT_SYMBOL(md_update_sb);
2888 static int add_bound_rdev(struct md_rdev *rdev)
2890 struct mddev *mddev = rdev->mddev;
2892 bool add_journal = test_bit(Journal, &rdev->flags);
2894 if (!mddev->pers->hot_remove_disk || add_journal) {
2895 /* If there is hot_add_disk but no hot_remove_disk
2896 * then added disks for geometry changes,
2897 * and should be added immediately.
2899 super_types[mddev->major_version].
2900 validate_super(mddev, rdev);
2902 mddev_suspend(mddev);
2903 err = mddev->pers->hot_add_disk(mddev, rdev);
2905 mddev_resume(mddev);
2907 md_kick_rdev_from_array(rdev);
2911 sysfs_notify_dirent_safe(rdev->sysfs_state);
2913 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2914 if (mddev->degraded)
2915 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2916 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2917 md_new_event(mddev);
2918 md_wakeup_thread(mddev->thread);
2922 /* words written to sysfs files may, or may not, be \n terminated.
2923 * We want to accept with case. For this we use cmd_match.
2925 static int cmd_match(const char *cmd, const char *str)
2927 /* See if cmd, written into a sysfs file, matches
2928 * str. They must either be the same, or cmd can
2929 * have a trailing newline
2931 while (*cmd && *str && *cmd == *str) {
2942 struct rdev_sysfs_entry {
2943 struct attribute attr;
2944 ssize_t (*show)(struct md_rdev *, char *);
2945 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2949 state_show(struct md_rdev *rdev, char *page)
2953 unsigned long flags = READ_ONCE(rdev->flags);
2955 if (test_bit(Faulty, &flags) ||
2956 (!test_bit(ExternalBbl, &flags) &&
2957 rdev->badblocks.unacked_exist))
2958 len += sprintf(page+len, "faulty%s", sep);
2959 if (test_bit(In_sync, &flags))
2960 len += sprintf(page+len, "in_sync%s", sep);
2961 if (test_bit(Journal, &flags))
2962 len += sprintf(page+len, "journal%s", sep);
2963 if (test_bit(WriteMostly, &flags))
2964 len += sprintf(page+len, "write_mostly%s", sep);
2965 if (test_bit(Blocked, &flags) ||
2966 (rdev->badblocks.unacked_exist
2967 && !test_bit(Faulty, &flags)))
2968 len += sprintf(page+len, "blocked%s", sep);
2969 if (!test_bit(Faulty, &flags) &&
2970 !test_bit(Journal, &flags) &&
2971 !test_bit(In_sync, &flags))
2972 len += sprintf(page+len, "spare%s", sep);
2973 if (test_bit(WriteErrorSeen, &flags))
2974 len += sprintf(page+len, "write_error%s", sep);
2975 if (test_bit(WantReplacement, &flags))
2976 len += sprintf(page+len, "want_replacement%s", sep);
2977 if (test_bit(Replacement, &flags))
2978 len += sprintf(page+len, "replacement%s", sep);
2979 if (test_bit(ExternalBbl, &flags))
2980 len += sprintf(page+len, "external_bbl%s", sep);
2981 if (test_bit(FailFast, &flags))
2982 len += sprintf(page+len, "failfast%s", sep);
2987 return len+sprintf(page+len, "\n");
2991 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2994 * faulty - simulates an error
2995 * remove - disconnects the device
2996 * writemostly - sets write_mostly
2997 * -writemostly - clears write_mostly
2998 * blocked - sets the Blocked flags
2999 * -blocked - clears the Blocked and possibly simulates an error
3000 * insync - sets Insync providing device isn't active
3001 * -insync - clear Insync for a device with a slot assigned,
3002 * so that it gets rebuilt based on bitmap
3003 * write_error - sets WriteErrorSeen
3004 * -write_error - clears WriteErrorSeen
3005 * {,-}failfast - set/clear FailFast
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)
3022 struct mddev *mddev = rdev->mddev;
3024 if (mddev_is_clustered(mddev))
3025 err = md_cluster_ops->remove_disk(mddev, rdev);
3028 md_kick_rdev_from_array(rdev);
3030 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3031 md_wakeup_thread(mddev->thread);
3033 md_new_event(mddev);
3036 } else if (cmd_match(buf, "writemostly")) {
3037 set_bit(WriteMostly, &rdev->flags);
3038 mddev_create_serial_pool(rdev->mddev, rdev, false);
3040 } else if (cmd_match(buf, "-writemostly")) {
3041 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
3042 clear_bit(WriteMostly, &rdev->flags);
3044 } else if (cmd_match(buf, "blocked")) {
3045 set_bit(Blocked, &rdev->flags);
3047 } else if (cmd_match(buf, "-blocked")) {
3048 if (!test_bit(Faulty, &rdev->flags) &&
3049 !test_bit(ExternalBbl, &rdev->flags) &&
3050 rdev->badblocks.unacked_exist) {
3051 /* metadata handler doesn't understand badblocks,
3052 * so we need to fail the device
3054 md_error(rdev->mddev, rdev);
3056 clear_bit(Blocked, &rdev->flags);
3057 clear_bit(BlockedBadBlocks, &rdev->flags);
3058 wake_up(&rdev->blocked_wait);
3059 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3060 md_wakeup_thread(rdev->mddev->thread);
3063 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3064 set_bit(In_sync, &rdev->flags);
3066 } else if (cmd_match(buf, "failfast")) {
3067 set_bit(FailFast, &rdev->flags);
3069 } else if (cmd_match(buf, "-failfast")) {
3070 clear_bit(FailFast, &rdev->flags);
3072 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3073 !test_bit(Journal, &rdev->flags)) {
3074 if (rdev->mddev->pers == NULL) {
3075 clear_bit(In_sync, &rdev->flags);
3076 rdev->saved_raid_disk = rdev->raid_disk;
3077 rdev->raid_disk = -1;
3080 } else if (cmd_match(buf, "write_error")) {
3081 set_bit(WriteErrorSeen, &rdev->flags);
3083 } else if (cmd_match(buf, "-write_error")) {
3084 clear_bit(WriteErrorSeen, &rdev->flags);
3086 } else if (cmd_match(buf, "want_replacement")) {
3087 /* Any non-spare device that is not a replacement can
3088 * become want_replacement at any time, but we then need to
3089 * check if recovery is needed.
3091 if (rdev->raid_disk >= 0 &&
3092 !test_bit(Journal, &rdev->flags) &&
3093 !test_bit(Replacement, &rdev->flags))
3094 set_bit(WantReplacement, &rdev->flags);
3095 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3096 md_wakeup_thread(rdev->mddev->thread);
3098 } else if (cmd_match(buf, "-want_replacement")) {
3099 /* Clearing 'want_replacement' is always allowed.
3100 * Once replacements starts it is too late though.
3103 clear_bit(WantReplacement, &rdev->flags);
3104 } else if (cmd_match(buf, "replacement")) {
3105 /* Can only set a device as a replacement when array has not
3106 * yet been started. Once running, replacement is automatic
3107 * from spares, or by assigning 'slot'.
3109 if (rdev->mddev->pers)
3112 set_bit(Replacement, &rdev->flags);
3115 } else if (cmd_match(buf, "-replacement")) {
3116 /* Similarly, can only clear Replacement before start */
3117 if (rdev->mddev->pers)
3120 clear_bit(Replacement, &rdev->flags);
3123 } else if (cmd_match(buf, "re-add")) {
3124 if (!rdev->mddev->pers)
3126 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3127 rdev->saved_raid_disk >= 0) {
3128 /* clear_bit is performed _after_ all the devices
3129 * have their local Faulty bit cleared. If any writes
3130 * happen in the meantime in the local node, they
3131 * will land in the local bitmap, which will be synced
3132 * by this node eventually
3134 if (!mddev_is_clustered(rdev->mddev) ||
3135 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3136 clear_bit(Faulty, &rdev->flags);
3137 err = add_bound_rdev(rdev);
3141 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3142 set_bit(ExternalBbl, &rdev->flags);
3143 rdev->badblocks.shift = 0;
3145 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3146 clear_bit(ExternalBbl, &rdev->flags);
3150 sysfs_notify_dirent_safe(rdev->sysfs_state);
3151 return err ? err : len;
3153 static struct rdev_sysfs_entry rdev_state =
3154 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3157 errors_show(struct md_rdev *rdev, char *page)
3159 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3163 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3168 rv = kstrtouint(buf, 10, &n);
3171 atomic_set(&rdev->corrected_errors, n);
3174 static struct rdev_sysfs_entry rdev_errors =
3175 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3178 slot_show(struct md_rdev *rdev, char *page)
3180 if (test_bit(Journal, &rdev->flags))
3181 return sprintf(page, "journal\n");
3182 else if (rdev->raid_disk < 0)
3183 return sprintf(page, "none\n");
3185 return sprintf(page, "%d\n", rdev->raid_disk);
3189 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3194 if (test_bit(Journal, &rdev->flags))
3196 if (strncmp(buf, "none", 4)==0)
3199 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3203 if (rdev->mddev->pers && slot == -1) {
3204 /* Setting 'slot' on an active array requires also
3205 * updating the 'rd%d' link, and communicating
3206 * with the personality with ->hot_*_disk.
3207 * For now we only support removing
3208 * failed/spare devices. This normally happens automatically,
3209 * but not when the metadata is externally managed.
3211 if (rdev->raid_disk == -1)
3213 /* personality does all needed checks */
3214 if (rdev->mddev->pers->hot_remove_disk == NULL)
3216 clear_bit(Blocked, &rdev->flags);
3217 remove_and_add_spares(rdev->mddev, rdev);
3218 if (rdev->raid_disk >= 0)
3220 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3221 md_wakeup_thread(rdev->mddev->thread);
3222 } else if (rdev->mddev->pers) {
3223 /* Activating a spare .. or possibly reactivating
3224 * if we ever get bitmaps working here.
3228 if (rdev->raid_disk != -1)
3231 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3234 if (rdev->mddev->pers->hot_add_disk == NULL)
3237 if (slot >= rdev->mddev->raid_disks &&
3238 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3241 rdev->raid_disk = slot;
3242 if (test_bit(In_sync, &rdev->flags))
3243 rdev->saved_raid_disk = slot;
3245 rdev->saved_raid_disk = -1;
3246 clear_bit(In_sync, &rdev->flags);
3247 clear_bit(Bitmap_sync, &rdev->flags);
3248 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3250 rdev->raid_disk = -1;
3253 sysfs_notify_dirent_safe(rdev->sysfs_state);
3254 /* failure here is OK */;
3255 sysfs_link_rdev(rdev->mddev, rdev);
3256 /* don't wakeup anyone, leave that to userspace. */
3258 if (slot >= rdev->mddev->raid_disks &&
3259 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3261 rdev->raid_disk = slot;
3262 /* assume it is working */
3263 clear_bit(Faulty, &rdev->flags);
3264 clear_bit(WriteMostly, &rdev->flags);
3265 set_bit(In_sync, &rdev->flags);
3266 sysfs_notify_dirent_safe(rdev->sysfs_state);
3271 static struct rdev_sysfs_entry rdev_slot =
3272 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3275 offset_show(struct md_rdev *rdev, char *page)
3277 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3281 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3283 unsigned long long offset;
3284 if (kstrtoull(buf, 10, &offset) < 0)
3286 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3288 if (rdev->sectors && rdev->mddev->external)
3289 /* Must set offset before size, so overlap checks
3292 rdev->data_offset = offset;
3293 rdev->new_data_offset = offset;
3297 static struct rdev_sysfs_entry rdev_offset =
3298 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3300 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3302 return sprintf(page, "%llu\n",
3303 (unsigned long long)rdev->new_data_offset);
3306 static ssize_t new_offset_store(struct md_rdev *rdev,
3307 const char *buf, size_t len)
3309 unsigned long long new_offset;
3310 struct mddev *mddev = rdev->mddev;
3312 if (kstrtoull(buf, 10, &new_offset) < 0)
3315 if (mddev->sync_thread ||
3316 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3318 if (new_offset == rdev->data_offset)
3319 /* reset is always permitted */
3321 else if (new_offset > rdev->data_offset) {
3322 /* must not push array size beyond rdev_sectors */
3323 if (new_offset - rdev->data_offset
3324 + mddev->dev_sectors > rdev->sectors)
3327 /* Metadata worries about other space details. */
3329 /* decreasing the offset is inconsistent with a backwards
3332 if (new_offset < rdev->data_offset &&
3333 mddev->reshape_backwards)
3335 /* Increasing offset is inconsistent with forwards
3336 * reshape. reshape_direction should be set to
3337 * 'backwards' first.
3339 if (new_offset > rdev->data_offset &&
3340 !mddev->reshape_backwards)
3343 if (mddev->pers && mddev->persistent &&
3344 !super_types[mddev->major_version]
3345 .allow_new_offset(rdev, new_offset))
3347 rdev->new_data_offset = new_offset;
3348 if (new_offset > rdev->data_offset)
3349 mddev->reshape_backwards = 1;
3350 else if (new_offset < rdev->data_offset)
3351 mddev->reshape_backwards = 0;
3355 static struct rdev_sysfs_entry rdev_new_offset =
3356 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3359 rdev_size_show(struct md_rdev *rdev, char *page)
3361 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3364 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3366 /* check if two start/length pairs overlap */
3374 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3376 unsigned long long blocks;
3379 if (kstrtoull(buf, 10, &blocks) < 0)
3382 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3383 return -EINVAL; /* sector conversion overflow */
3386 if (new != blocks * 2)
3387 return -EINVAL; /* unsigned long long to sector_t overflow */
3394 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3396 struct mddev *my_mddev = rdev->mddev;
3397 sector_t oldsectors = rdev->sectors;
3400 if (test_bit(Journal, &rdev->flags))
3402 if (strict_blocks_to_sectors(buf, §ors) < 0)
3404 if (rdev->data_offset != rdev->new_data_offset)
3405 return -EINVAL; /* too confusing */
3406 if (my_mddev->pers && rdev->raid_disk >= 0) {
3407 if (my_mddev->persistent) {
3408 sectors = super_types[my_mddev->major_version].
3409 rdev_size_change(rdev, sectors);
3412 } else if (!sectors)
3413 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3415 if (!my_mddev->pers->resize)
3416 /* Cannot change size for RAID0 or Linear etc */
3419 if (sectors < my_mddev->dev_sectors)
3420 return -EINVAL; /* component must fit device */
3422 rdev->sectors = sectors;
3423 if (sectors > oldsectors && my_mddev->external) {
3424 /* Need to check that all other rdevs with the same
3425 * ->bdev do not overlap. 'rcu' is sufficient to walk
3426 * the rdev lists safely.
3427 * This check does not provide a hard guarantee, it
3428 * just helps avoid dangerous mistakes.
3430 struct mddev *mddev;
3432 struct list_head *tmp;
3435 for_each_mddev(mddev, tmp) {
3436 struct md_rdev *rdev2;
3438 rdev_for_each(rdev2, mddev)
3439 if (rdev->bdev == rdev2->bdev &&
3441 overlaps(rdev->data_offset, rdev->sectors,
3454 /* Someone else could have slipped in a size
3455 * change here, but doing so is just silly.
3456 * We put oldsectors back because we *know* it is
3457 * safe, and trust userspace not to race with
3460 rdev->sectors = oldsectors;
3467 static struct rdev_sysfs_entry rdev_size =
3468 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3470 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3472 unsigned long long recovery_start = rdev->recovery_offset;
3474 if (test_bit(In_sync, &rdev->flags) ||
3475 recovery_start == MaxSector)
3476 return sprintf(page, "none\n");
3478 return sprintf(page, "%llu\n", recovery_start);
3481 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3483 unsigned long long recovery_start;
3485 if (cmd_match(buf, "none"))
3486 recovery_start = MaxSector;
3487 else if (kstrtoull(buf, 10, &recovery_start))
3490 if (rdev->mddev->pers &&
3491 rdev->raid_disk >= 0)
3494 rdev->recovery_offset = recovery_start;
3495 if (recovery_start == MaxSector)
3496 set_bit(In_sync, &rdev->flags);
3498 clear_bit(In_sync, &rdev->flags);
3502 static struct rdev_sysfs_entry rdev_recovery_start =
3503 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3505 /* sysfs access to bad-blocks list.
3506 * We present two files.
3507 * 'bad-blocks' lists sector numbers and lengths of ranges that
3508 * are recorded as bad. The list is truncated to fit within
3509 * the one-page limit of sysfs.
3510 * Writing "sector length" to this file adds an acknowledged
3512 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3513 * been acknowledged. Writing to this file adds bad blocks
3514 * without acknowledging them. This is largely for testing.
3516 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3518 return badblocks_show(&rdev->badblocks, page, 0);
3520 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3522 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3523 /* Maybe that ack was all we needed */
3524 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3525 wake_up(&rdev->blocked_wait);
3528 static struct rdev_sysfs_entry rdev_bad_blocks =
3529 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3531 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3533 return badblocks_show(&rdev->badblocks, page, 1);
3535 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3537 return badblocks_store(&rdev->badblocks, page, len, 1);
3539 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3540 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3543 ppl_sector_show(struct md_rdev *rdev, char *page)
3545 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3549 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3551 unsigned long long sector;
3553 if (kstrtoull(buf, 10, §or) < 0)
3555 if (sector != (sector_t)sector)
3558 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3559 rdev->raid_disk >= 0)
3562 if (rdev->mddev->persistent) {
3563 if (rdev->mddev->major_version == 0)
3565 if ((sector > rdev->sb_start &&
3566 sector - rdev->sb_start > S16_MAX) ||
3567 (sector < rdev->sb_start &&
3568 rdev->sb_start - sector > -S16_MIN))
3570 rdev->ppl.offset = sector - rdev->sb_start;
3571 } else if (!rdev->mddev->external) {
3574 rdev->ppl.sector = sector;
3578 static struct rdev_sysfs_entry rdev_ppl_sector =
3579 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3582 ppl_size_show(struct md_rdev *rdev, char *page)
3584 return sprintf(page, "%u\n", rdev->ppl.size);
3588 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3592 if (kstrtouint(buf, 10, &size) < 0)
3595 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3596 rdev->raid_disk >= 0)
3599 if (rdev->mddev->persistent) {
3600 if (rdev->mddev->major_version == 0)
3604 } else if (!rdev->mddev->external) {
3607 rdev->ppl.size = size;
3611 static struct rdev_sysfs_entry rdev_ppl_size =
3612 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3614 static struct attribute *rdev_default_attrs[] = {
3619 &rdev_new_offset.attr,
3621 &rdev_recovery_start.attr,
3622 &rdev_bad_blocks.attr,
3623 &rdev_unack_bad_blocks.attr,
3624 &rdev_ppl_sector.attr,
3625 &rdev_ppl_size.attr,
3629 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3631 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3632 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3638 return entry->show(rdev, page);
3642 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3643 const char *page, size_t length)
3645 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3646 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3648 struct mddev *mddev = rdev->mddev;
3652 if (!capable(CAP_SYS_ADMIN))
3654 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3656 if (rdev->mddev == NULL)
3659 rv = entry->store(rdev, page, length);
3660 mddev_unlock(mddev);
3665 static void rdev_free(struct kobject *ko)
3667 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3670 static const struct sysfs_ops rdev_sysfs_ops = {
3671 .show = rdev_attr_show,
3672 .store = rdev_attr_store,
3674 static struct kobj_type rdev_ktype = {
3675 .release = rdev_free,
3676 .sysfs_ops = &rdev_sysfs_ops,
3677 .default_attrs = rdev_default_attrs,
3680 int md_rdev_init(struct md_rdev *rdev)
3683 rdev->saved_raid_disk = -1;
3684 rdev->raid_disk = -1;
3686 rdev->data_offset = 0;
3687 rdev->new_data_offset = 0;
3688 rdev->sb_events = 0;
3689 rdev->last_read_error = 0;
3690 rdev->sb_loaded = 0;
3691 rdev->bb_page = NULL;
3692 atomic_set(&rdev->nr_pending, 0);
3693 atomic_set(&rdev->read_errors, 0);
3694 atomic_set(&rdev->corrected_errors, 0);
3696 INIT_LIST_HEAD(&rdev->same_set);
3697 init_waitqueue_head(&rdev->blocked_wait);
3699 /* Add space to store bad block list.
3700 * This reserves the space even on arrays where it cannot
3701 * be used - I wonder if that matters
3703 return badblocks_init(&rdev->badblocks, 0);
3705 EXPORT_SYMBOL_GPL(md_rdev_init);
3707 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3709 * mark the device faulty if:
3711 * - the device is nonexistent (zero size)
3712 * - the device has no valid superblock
3714 * a faulty rdev _never_ has rdev->sb set.
3716 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3718 char b[BDEVNAME_SIZE];
3720 struct md_rdev *rdev;
3723 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3725 return ERR_PTR(-ENOMEM);
3727 err = md_rdev_init(rdev);
3730 err = alloc_disk_sb(rdev);
3734 err = lock_rdev(rdev, newdev, super_format == -2);
3738 kobject_init(&rdev->kobj, &rdev_ktype);
3740 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3742 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3743 bdevname(rdev->bdev,b));
3748 if (super_format >= 0) {
3749 err = super_types[super_format].
3750 load_super(rdev, NULL, super_minor);
3751 if (err == -EINVAL) {
3752 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3753 bdevname(rdev->bdev,b),
3754 super_format, super_minor);
3758 pr_warn("md: could not read %s's sb, not importing!\n",
3759 bdevname(rdev->bdev,b));
3769 md_rdev_clear(rdev);
3771 return ERR_PTR(err);
3775 * Check a full RAID array for plausibility
3778 static int analyze_sbs(struct mddev *mddev)
3781 struct md_rdev *rdev, *freshest, *tmp;
3782 char b[BDEVNAME_SIZE];
3785 rdev_for_each_safe(rdev, tmp, mddev)
3786 switch (super_types[mddev->major_version].
3787 load_super(rdev, freshest, mddev->minor_version)) {
3794 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3795 bdevname(rdev->bdev,b));
3796 md_kick_rdev_from_array(rdev);
3799 /* Cannot find a valid fresh disk */
3801 pr_warn("md: cannot find a valid disk\n");
3805 super_types[mddev->major_version].
3806 validate_super(mddev, freshest);
3809 rdev_for_each_safe(rdev, tmp, mddev) {
3810 if (mddev->max_disks &&
3811 (rdev->desc_nr >= mddev->max_disks ||
3812 i > mddev->max_disks)) {
3813 pr_warn("md: %s: %s: only %d devices permitted\n",
3814 mdname(mddev), bdevname(rdev->bdev, b),
3816 md_kick_rdev_from_array(rdev);
3819 if (rdev != freshest) {
3820 if (super_types[mddev->major_version].
3821 validate_super(mddev, rdev)) {
3822 pr_warn("md: kicking non-fresh %s from array!\n",
3823 bdevname(rdev->bdev,b));
3824 md_kick_rdev_from_array(rdev);
3828 if (mddev->level == LEVEL_MULTIPATH) {
3829 rdev->desc_nr = i++;
3830 rdev->raid_disk = rdev->desc_nr;
3831 set_bit(In_sync, &rdev->flags);
3832 } else if (rdev->raid_disk >=
3833 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3834 !test_bit(Journal, &rdev->flags)) {
3835 rdev->raid_disk = -1;
3836 clear_bit(In_sync, &rdev->flags);
3843 /* Read a fixed-point number.
3844 * Numbers in sysfs attributes should be in "standard" units where
3845 * possible, so time should be in seconds.
3846 * However we internally use a a much smaller unit such as
3847 * milliseconds or jiffies.
3848 * This function takes a decimal number with a possible fractional
3849 * component, and produces an integer which is the result of
3850 * multiplying that number by 10^'scale'.
3851 * all without any floating-point arithmetic.
3853 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3855 unsigned long result = 0;
3857 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3860 else if (decimals < scale) {
3863 result = result * 10 + value;
3875 *res = result * int_pow(10, scale - decimals);
3880 safe_delay_show(struct mddev *mddev, char *page)
3882 int msec = (mddev->safemode_delay*1000)/HZ;
3883 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3886 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3890 if (mddev_is_clustered(mddev)) {
3891 pr_warn("md: Safemode is disabled for clustered mode\n");
3895 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3898 mddev->safemode_delay = 0;
3900 unsigned long old_delay = mddev->safemode_delay;
3901 unsigned long new_delay = (msec*HZ)/1000;
3905 mddev->safemode_delay = new_delay;
3906 if (new_delay < old_delay || old_delay == 0)
3907 mod_timer(&mddev->safemode_timer, jiffies+1);
3911 static struct md_sysfs_entry md_safe_delay =
3912 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3915 level_show(struct mddev *mddev, char *page)
3917 struct md_personality *p;
3919 spin_lock(&mddev->lock);
3922 ret = sprintf(page, "%s\n", p->name);
3923 else if (mddev->clevel[0])
3924 ret = sprintf(page, "%s\n", mddev->clevel);
3925 else if (mddev->level != LEVEL_NONE)
3926 ret = sprintf(page, "%d\n", mddev->level);
3929 spin_unlock(&mddev->lock);
3934 level_store(struct mddev *mddev, const char *buf, size_t len)
3939 struct md_personality *pers, *oldpers;
3941 void *priv, *oldpriv;
3942 struct md_rdev *rdev;
3944 if (slen == 0 || slen >= sizeof(clevel))
3947 rv = mddev_lock(mddev);
3951 if (mddev->pers == NULL) {
3952 strncpy(mddev->clevel, buf, slen);
3953 if (mddev->clevel[slen-1] == '\n')
3955 mddev->clevel[slen] = 0;
3956 mddev->level = LEVEL_NONE;
3964 /* request to change the personality. Need to ensure:
3965 * - array is not engaged in resync/recovery/reshape
3966 * - old personality can be suspended
3967 * - new personality will access other array.
3971 if (mddev->sync_thread ||
3972 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3973 mddev->reshape_position != MaxSector ||
3974 mddev->sysfs_active)
3978 if (!mddev->pers->quiesce) {
3979 pr_warn("md: %s: %s does not support online personality change\n",
3980 mdname(mddev), mddev->pers->name);
3984 /* Now find the new personality */
3985 strncpy(clevel, buf, slen);
3986 if (clevel[slen-1] == '\n')
3989 if (kstrtol(clevel, 10, &level))
3992 if (request_module("md-%s", clevel) != 0)
3993 request_module("md-level-%s", clevel);
3994 spin_lock(&pers_lock);
3995 pers = find_pers(level, clevel);
3996 if (!pers || !try_module_get(pers->owner)) {
3997 spin_unlock(&pers_lock);
3998 pr_warn("md: personality %s not loaded\n", clevel);
4002 spin_unlock(&pers_lock);
4004 if (pers == mddev->pers) {
4005 /* Nothing to do! */
4006 module_put(pers->owner);
4010 if (!pers->takeover) {
4011 module_put(pers->owner);
4012 pr_warn("md: %s: %s does not support personality takeover\n",
4013 mdname(mddev), clevel);
4018 rdev_for_each(rdev, mddev)
4019 rdev->new_raid_disk = rdev->raid_disk;
4021 /* ->takeover must set new_* and/or delta_disks
4022 * if it succeeds, and may set them when it fails.
4024 priv = pers->takeover(mddev);
4026 mddev->new_level = mddev->level;
4027 mddev->new_layout = mddev->layout;
4028 mddev->new_chunk_sectors = mddev->chunk_sectors;
4029 mddev->raid_disks -= mddev->delta_disks;
4030 mddev->delta_disks = 0;
4031 mddev->reshape_backwards = 0;
4032 module_put(pers->owner);
4033 pr_warn("md: %s: %s would not accept array\n",
4034 mdname(mddev), clevel);
4039 /* Looks like we have a winner */
4040 mddev_suspend(mddev);
4041 mddev_detach(mddev);
4043 spin_lock(&mddev->lock);
4044 oldpers = mddev->pers;
4045 oldpriv = mddev->private;
4047 mddev->private = priv;
4048 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4049 mddev->level = mddev->new_level;
4050 mddev->layout = mddev->new_layout;
4051 mddev->chunk_sectors = mddev->new_chunk_sectors;
4052 mddev->delta_disks = 0;
4053 mddev->reshape_backwards = 0;
4054 mddev->degraded = 0;
4055 spin_unlock(&mddev->lock);
4057 if (oldpers->sync_request == NULL &&
4059 /* We are converting from a no-redundancy array
4060 * to a redundancy array and metadata is managed
4061 * externally so we need to be sure that writes
4062 * won't block due to a need to transition
4064 * until external management is started.
4067 mddev->safemode_delay = 0;
4068 mddev->safemode = 0;
4071 oldpers->free(mddev, oldpriv);
4073 if (oldpers->sync_request == NULL &&
4074 pers->sync_request != NULL) {
4075 /* need to add the md_redundancy_group */
4076 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4077 pr_warn("md: cannot register extra attributes for %s\n",
4079 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4080 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4081 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4083 if (oldpers->sync_request != NULL &&
4084 pers->sync_request == NULL) {
4085 /* need to remove the md_redundancy_group */
4086 if (mddev->to_remove == NULL)
4087 mddev->to_remove = &md_redundancy_group;
4090 module_put(oldpers->owner);
4092 rdev_for_each(rdev, mddev) {
4093 if (rdev->raid_disk < 0)
4095 if (rdev->new_raid_disk >= mddev->raid_disks)
4096 rdev->new_raid_disk = -1;
4097 if (rdev->new_raid_disk == rdev->raid_disk)
4099 sysfs_unlink_rdev(mddev, rdev);
4101 rdev_for_each(rdev, mddev) {
4102 if (rdev->raid_disk < 0)
4104 if (rdev->new_raid_disk == rdev->raid_disk)
4106 rdev->raid_disk = rdev->new_raid_disk;
4107 if (rdev->raid_disk < 0)
4108 clear_bit(In_sync, &rdev->flags);
4110 if (sysfs_link_rdev(mddev, rdev))
4111 pr_warn("md: cannot register rd%d for %s after level change\n",
4112 rdev->raid_disk, mdname(mddev));
4116 if (pers->sync_request == NULL) {
4117 /* this is now an array without redundancy, so
4118 * it must always be in_sync
4121 del_timer_sync(&mddev->safemode_timer);
4123 blk_set_stacking_limits(&mddev->queue->limits);
4125 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4126 mddev_resume(mddev);
4128 md_update_sb(mddev, 1);
4129 sysfs_notify_dirent_safe(mddev->sysfs_level);
4130 md_new_event(mddev);
4133 mddev_unlock(mddev);
4137 static struct md_sysfs_entry md_level =
4138 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4141 layout_show(struct mddev *mddev, char *page)
4143 /* just a number, not meaningful for all levels */
4144 if (mddev->reshape_position != MaxSector &&
4145 mddev->layout != mddev->new_layout)
4146 return sprintf(page, "%d (%d)\n",
4147 mddev->new_layout, mddev->layout);
4148 return sprintf(page, "%d\n", mddev->layout);
4152 layout_store(struct mddev *mddev, const char *buf, size_t len)
4157 err = kstrtouint(buf, 10, &n);
4160 err = mddev_lock(mddev);
4165 if (mddev->pers->check_reshape == NULL)
4170 mddev->new_layout = n;
4171 err = mddev->pers->check_reshape(mddev);
4173 mddev->new_layout = mddev->layout;
4176 mddev->new_layout = n;
4177 if (mddev->reshape_position == MaxSector)
4180 mddev_unlock(mddev);
4183 static struct md_sysfs_entry md_layout =
4184 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4187 raid_disks_show(struct mddev *mddev, char *page)
4189 if (mddev->raid_disks == 0)
4191 if (mddev->reshape_position != MaxSector &&
4192 mddev->delta_disks != 0)
4193 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4194 mddev->raid_disks - mddev->delta_disks);
4195 return sprintf(page, "%d\n", mddev->raid_disks);
4198 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4201 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4206 err = kstrtouint(buf, 10, &n);
4210 err = mddev_lock(mddev);
4214 err = update_raid_disks(mddev, n);
4215 else if (mddev->reshape_position != MaxSector) {
4216 struct md_rdev *rdev;
4217 int olddisks = mddev->raid_disks - mddev->delta_disks;
4220 rdev_for_each(rdev, mddev) {
4222 rdev->data_offset < rdev->new_data_offset)
4225 rdev->data_offset > rdev->new_data_offset)
4229 mddev->delta_disks = n - olddisks;
4230 mddev->raid_disks = n;
4231 mddev->reshape_backwards = (mddev->delta_disks < 0);
4233 mddev->raid_disks = n;
4235 mddev_unlock(mddev);
4236 return err ? err : len;
4238 static struct md_sysfs_entry md_raid_disks =
4239 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4242 uuid_show(struct mddev *mddev, char *page)
4244 return sprintf(page, "%pU\n", mddev->uuid);
4246 static struct md_sysfs_entry md_uuid =
4247 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4250 chunk_size_show(struct mddev *mddev, char *page)
4252 if (mddev->reshape_position != MaxSector &&
4253 mddev->chunk_sectors != mddev->new_chunk_sectors)
4254 return sprintf(page, "%d (%d)\n",
4255 mddev->new_chunk_sectors << 9,
4256 mddev->chunk_sectors << 9);
4257 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4261 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4266 err = kstrtoul(buf, 10, &n);
4270 err = mddev_lock(mddev);
4274 if (mddev->pers->check_reshape == NULL)
4279 mddev->new_chunk_sectors = n >> 9;
4280 err = mddev->pers->check_reshape(mddev);
4282 mddev->new_chunk_sectors = mddev->chunk_sectors;
4285 mddev->new_chunk_sectors = n >> 9;
4286 if (mddev->reshape_position == MaxSector)
4287 mddev->chunk_sectors = n >> 9;
4289 mddev_unlock(mddev);
4292 static struct md_sysfs_entry md_chunk_size =
4293 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4296 resync_start_show(struct mddev *mddev, char *page)
4298 if (mddev->recovery_cp == MaxSector)
4299 return sprintf(page, "none\n");
4300 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4304 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4306 unsigned long long n;
4309 if (cmd_match(buf, "none"))
4312 err = kstrtoull(buf, 10, &n);
4315 if (n != (sector_t)n)
4319 err = mddev_lock(mddev);
4322 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4326 mddev->recovery_cp = n;
4328 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4330 mddev_unlock(mddev);
4333 static struct md_sysfs_entry md_resync_start =
4334 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4335 resync_start_show, resync_start_store);
4338 * The array state can be:
4341 * No devices, no size, no level
4342 * Equivalent to STOP_ARRAY ioctl
4344 * May have some settings, but array is not active
4345 * all IO results in error
4346 * When written, doesn't tear down array, but just stops it
4347 * suspended (not supported yet)
4348 * All IO requests will block. The array can be reconfigured.
4349 * Writing this, if accepted, will block until array is quiescent
4351 * no resync can happen. no superblocks get written.
4352 * write requests fail
4354 * like readonly, but behaves like 'clean' on a write request.
4356 * clean - no pending writes, but otherwise active.
4357 * When written to inactive array, starts without resync
4358 * If a write request arrives then
4359 * if metadata is known, mark 'dirty' and switch to 'active'.
4360 * if not known, block and switch to write-pending
4361 * If written to an active array that has pending writes, then fails.
4363 * fully active: IO and resync can be happening.
4364 * When written to inactive array, starts with resync
4367 * clean, but writes are blocked waiting for 'active' to be written.
4370 * like active, but no writes have been seen for a while (100msec).
4373 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4374 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4375 * when a member is gone, so this state will at least alert the
4376 * user that something is wrong.
4378 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4379 write_pending, active_idle, broken, bad_word};
4380 static char *array_states[] = {
4381 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4382 "write-pending", "active-idle", "broken", NULL };
4384 static int match_word(const char *word, char **list)
4387 for (n=0; list[n]; n++)
4388 if (cmd_match(word, list[n]))
4394 array_state_show(struct mddev *mddev, char *page)
4396 enum array_state st = inactive;
4398 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4407 spin_lock(&mddev->lock);
4408 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4410 else if (mddev->in_sync)
4412 else if (mddev->safemode)
4416 spin_unlock(&mddev->lock);
4419 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4422 if (list_empty(&mddev->disks) &&
4423 mddev->raid_disks == 0 &&
4424 mddev->dev_sectors == 0)
4429 return sprintf(page, "%s\n", array_states[st]);
4432 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4433 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4434 static int restart_array(struct mddev *mddev);
4437 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4440 enum array_state st = match_word(buf, array_states);
4442 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4443 /* don't take reconfig_mutex when toggling between
4446 spin_lock(&mddev->lock);
4448 restart_array(mddev);
4449 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4450 md_wakeup_thread(mddev->thread);
4451 wake_up(&mddev->sb_wait);
4452 } else /* st == clean */ {
4453 restart_array(mddev);
4454 if (!set_in_sync(mddev))
4458 sysfs_notify_dirent_safe(mddev->sysfs_state);
4459 spin_unlock(&mddev->lock);
4462 err = mddev_lock(mddev);
4470 /* stopping an active array */
4471 err = do_md_stop(mddev, 0, NULL);
4474 /* stopping an active array */
4476 err = do_md_stop(mddev, 2, NULL);
4478 err = 0; /* already inactive */
4481 break; /* not supported yet */
4484 err = md_set_readonly(mddev, NULL);
4487 set_disk_ro(mddev->gendisk, 1);
4488 err = do_md_run(mddev);
4494 err = md_set_readonly(mddev, NULL);
4495 else if (mddev->ro == 1)
4496 err = restart_array(mddev);
4499 set_disk_ro(mddev->gendisk, 0);
4503 err = do_md_run(mddev);
4508 err = restart_array(mddev);
4511 spin_lock(&mddev->lock);
4512 if (!set_in_sync(mddev))
4514 spin_unlock(&mddev->lock);
4520 err = restart_array(mddev);
4523 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4524 wake_up(&mddev->sb_wait);
4528 set_disk_ro(mddev->gendisk, 0);
4529 err = do_md_run(mddev);
4535 /* these cannot be set */
4540 if (mddev->hold_active == UNTIL_IOCTL)
4541 mddev->hold_active = 0;
4542 sysfs_notify_dirent_safe(mddev->sysfs_state);
4544 mddev_unlock(mddev);
4547 static struct md_sysfs_entry md_array_state =
4548 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4551 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4552 return sprintf(page, "%d\n",
4553 atomic_read(&mddev->max_corr_read_errors));
4557 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4562 rv = kstrtouint(buf, 10, &n);
4565 atomic_set(&mddev->max_corr_read_errors, n);
4569 static struct md_sysfs_entry max_corr_read_errors =
4570 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4571 max_corrected_read_errors_store);
4574 null_show(struct mddev *mddev, char *page)
4579 /* need to ensure rdev_delayed_delete() has completed */
4580 static void flush_rdev_wq(struct mddev *mddev)
4582 struct md_rdev *rdev;
4585 rdev_for_each_rcu(rdev, mddev)
4586 if (work_pending(&rdev->del_work)) {
4587 flush_workqueue(md_rdev_misc_wq);
4594 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4596 /* buf must be %d:%d\n? giving major and minor numbers */
4597 /* The new device is added to the array.
4598 * If the array has a persistent superblock, we read the
4599 * superblock to initialise info and check validity.
4600 * Otherwise, only checking done is that in bind_rdev_to_array,
4601 * which mainly checks size.
4604 int major = simple_strtoul(buf, &e, 10);
4607 struct md_rdev *rdev;
4610 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4612 minor = simple_strtoul(e+1, &e, 10);
4613 if (*e && *e != '\n')
4615 dev = MKDEV(major, minor);
4616 if (major != MAJOR(dev) ||
4617 minor != MINOR(dev))
4620 flush_rdev_wq(mddev);
4621 err = mddev_lock(mddev);
4624 if (mddev->persistent) {
4625 rdev = md_import_device(dev, mddev->major_version,
4626 mddev->minor_version);
4627 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4628 struct md_rdev *rdev0
4629 = list_entry(mddev->disks.next,
4630 struct md_rdev, same_set);
4631 err = super_types[mddev->major_version]
4632 .load_super(rdev, rdev0, mddev->minor_version);
4636 } else if (mddev->external)
4637 rdev = md_import_device(dev, -2, -1);
4639 rdev = md_import_device(dev, -1, -1);
4642 mddev_unlock(mddev);
4643 return PTR_ERR(rdev);
4645 err = bind_rdev_to_array(rdev, mddev);
4649 mddev_unlock(mddev);
4651 md_new_event(mddev);
4652 return err ? err : len;
4655 static struct md_sysfs_entry md_new_device =
4656 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4659 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4662 unsigned long chunk, end_chunk;
4665 err = mddev_lock(mddev);
4670 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4672 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4673 if (buf == end) break;
4674 if (*end == '-') { /* range */
4676 end_chunk = simple_strtoul(buf, &end, 0);
4677 if (buf == end) break;
4679 if (*end && !isspace(*end)) break;
4680 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4681 buf = skip_spaces(end);
4683 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4685 mddev_unlock(mddev);
4689 static struct md_sysfs_entry md_bitmap =
4690 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4693 size_show(struct mddev *mddev, char *page)
4695 return sprintf(page, "%llu\n",
4696 (unsigned long long)mddev->dev_sectors / 2);
4699 static int update_size(struct mddev *mddev, sector_t num_sectors);
4702 size_store(struct mddev *mddev, const char *buf, size_t len)
4704 /* If array is inactive, we can reduce the component size, but
4705 * not increase it (except from 0).
4706 * If array is active, we can try an on-line resize
4709 int err = strict_blocks_to_sectors(buf, §ors);
4713 err = mddev_lock(mddev);
4717 err = update_size(mddev, sectors);
4719 md_update_sb(mddev, 1);
4721 if (mddev->dev_sectors == 0 ||
4722 mddev->dev_sectors > sectors)
4723 mddev->dev_sectors = sectors;
4727 mddev_unlock(mddev);
4728 return err ? err : len;
4731 static struct md_sysfs_entry md_size =
4732 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4734 /* Metadata version.
4736 * 'none' for arrays with no metadata (good luck...)
4737 * 'external' for arrays with externally managed metadata,
4738 * or N.M for internally known formats
4741 metadata_show(struct mddev *mddev, char *page)
4743 if (mddev->persistent)
4744 return sprintf(page, "%d.%d\n",
4745 mddev->major_version, mddev->minor_version);
4746 else if (mddev->external)
4747 return sprintf(page, "external:%s\n", mddev->metadata_type);
4749 return sprintf(page, "none\n");
4753 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4758 /* Changing the details of 'external' metadata is
4759 * always permitted. Otherwise there must be
4760 * no devices attached to the array.
4763 err = mddev_lock(mddev);
4767 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4769 else if (!list_empty(&mddev->disks))
4773 if (cmd_match(buf, "none")) {
4774 mddev->persistent = 0;
4775 mddev->external = 0;
4776 mddev->major_version = 0;
4777 mddev->minor_version = 90;
4780 if (strncmp(buf, "external:", 9) == 0) {
4781 size_t namelen = len-9;
4782 if (namelen >= sizeof(mddev->metadata_type))
4783 namelen = sizeof(mddev->metadata_type)-1;
4784 strncpy(mddev->metadata_type, buf+9, namelen);
4785 mddev->metadata_type[namelen] = 0;
4786 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4787 mddev->metadata_type[--namelen] = 0;
4788 mddev->persistent = 0;
4789 mddev->external = 1;
4790 mddev->major_version = 0;
4791 mddev->minor_version = 90;
4794 major = simple_strtoul(buf, &e, 10);
4796 if (e==buf || *e != '.')
4799 minor = simple_strtoul(buf, &e, 10);
4800 if (e==buf || (*e && *e != '\n') )
4803 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4805 mddev->major_version = major;
4806 mddev->minor_version = minor;
4807 mddev->persistent = 1;
4808 mddev->external = 0;
4811 mddev_unlock(mddev);
4815 static struct md_sysfs_entry md_metadata =
4816 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4819 action_show(struct mddev *mddev, char *page)
4821 char *type = "idle";
4822 unsigned long recovery = mddev->recovery;
4823 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4825 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4826 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4827 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4829 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4830 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4832 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4836 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4838 else if (mddev->reshape_position != MaxSector)
4841 return sprintf(page, "%s\n", type);
4845 action_store(struct mddev *mddev, const char *page, size_t len)
4847 if (!mddev->pers || !mddev->pers->sync_request)
4851 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4852 if (cmd_match(page, "frozen"))
4853 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4855 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4856 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4857 mddev_lock(mddev) == 0) {
4858 if (work_pending(&mddev->del_work))
4859 flush_workqueue(md_misc_wq);
4860 if (mddev->sync_thread) {
4861 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4862 md_reap_sync_thread(mddev);
4864 mddev_unlock(mddev);
4866 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4868 else if (cmd_match(page, "resync"))
4869 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4870 else if (cmd_match(page, "recover")) {
4871 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4872 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4873 } else if (cmd_match(page, "reshape")) {
4875 if (mddev->pers->start_reshape == NULL)
4877 err = mddev_lock(mddev);
4879 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4882 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4883 err = mddev->pers->start_reshape(mddev);
4885 mddev_unlock(mddev);
4889 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4891 if (cmd_match(page, "check"))
4892 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4893 else if (!cmd_match(page, "repair"))
4895 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4896 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4897 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4899 if (mddev->ro == 2) {
4900 /* A write to sync_action is enough to justify
4901 * canceling read-auto mode
4904 md_wakeup_thread(mddev->sync_thread);
4906 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4907 md_wakeup_thread(mddev->thread);
4908 sysfs_notify_dirent_safe(mddev->sysfs_action);
4912 static struct md_sysfs_entry md_scan_mode =
4913 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4916 last_sync_action_show(struct mddev *mddev, char *page)
4918 return sprintf(page, "%s\n", mddev->last_sync_action);
4921 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4924 mismatch_cnt_show(struct mddev *mddev, char *page)
4926 return sprintf(page, "%llu\n",
4927 (unsigned long long)
4928 atomic64_read(&mddev->resync_mismatches));
4931 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4934 sync_min_show(struct mddev *mddev, char *page)
4936 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4937 mddev->sync_speed_min ? "local": "system");
4941 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4946 if (strncmp(buf, "system", 6)==0) {
4949 rv = kstrtouint(buf, 10, &min);
4955 mddev->sync_speed_min = min;
4959 static struct md_sysfs_entry md_sync_min =
4960 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4963 sync_max_show(struct mddev *mddev, char *page)
4965 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4966 mddev->sync_speed_max ? "local": "system");
4970 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4975 if (strncmp(buf, "system", 6)==0) {
4978 rv = kstrtouint(buf, 10, &max);
4984 mddev->sync_speed_max = max;
4988 static struct md_sysfs_entry md_sync_max =
4989 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4992 degraded_show(struct mddev *mddev, char *page)
4994 return sprintf(page, "%d\n", mddev->degraded);
4996 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4999 sync_force_parallel_show(struct mddev *mddev, char *page)
5001 return sprintf(page, "%d\n", mddev->parallel_resync);
5005 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5009 if (kstrtol(buf, 10, &n))
5012 if (n != 0 && n != 1)
5015 mddev->parallel_resync = n;
5017 if (mddev->sync_thread)
5018 wake_up(&resync_wait);
5023 /* force parallel resync, even with shared block devices */
5024 static struct md_sysfs_entry md_sync_force_parallel =
5025 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5026 sync_force_parallel_show, sync_force_parallel_store);
5029 sync_speed_show(struct mddev *mddev, char *page)
5031 unsigned long resync, dt, db;
5032 if (mddev->curr_resync == 0)
5033 return sprintf(page, "none\n");
5034 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5035 dt = (jiffies - mddev->resync_mark) / HZ;
5037 db = resync - mddev->resync_mark_cnt;
5038 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5041 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5044 sync_completed_show(struct mddev *mddev, char *page)
5046 unsigned long long max_sectors, resync;
5048 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5049 return sprintf(page, "none\n");
5051 if (mddev->curr_resync == 1 ||
5052 mddev->curr_resync == 2)
5053 return sprintf(page, "delayed\n");
5055 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5056 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5057 max_sectors = mddev->resync_max_sectors;
5059 max_sectors = mddev->dev_sectors;
5061 resync = mddev->curr_resync_completed;
5062 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5065 static struct md_sysfs_entry md_sync_completed =
5066 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5069 min_sync_show(struct mddev *mddev, char *page)
5071 return sprintf(page, "%llu\n",
5072 (unsigned long long)mddev->resync_min);
5075 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5077 unsigned long long min;
5080 if (kstrtoull(buf, 10, &min))
5083 spin_lock(&mddev->lock);
5085 if (min > mddev->resync_max)
5089 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5092 /* Round down to multiple of 4K for safety */
5093 mddev->resync_min = round_down(min, 8);
5097 spin_unlock(&mddev->lock);
5101 static struct md_sysfs_entry md_min_sync =
5102 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5105 max_sync_show(struct mddev *mddev, char *page)
5107 if (mddev->resync_max == MaxSector)
5108 return sprintf(page, "max\n");
5110 return sprintf(page, "%llu\n",
5111 (unsigned long long)mddev->resync_max);
5114 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5117 spin_lock(&mddev->lock);
5118 if (strncmp(buf, "max", 3) == 0)
5119 mddev->resync_max = MaxSector;
5121 unsigned long long max;
5125 if (kstrtoull(buf, 10, &max))
5127 if (max < mddev->resync_min)
5131 if (max < mddev->resync_max &&
5133 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5136 /* Must be a multiple of chunk_size */
5137 chunk = mddev->chunk_sectors;
5139 sector_t temp = max;
5142 if (sector_div(temp, chunk))
5145 mddev->resync_max = max;
5147 wake_up(&mddev->recovery_wait);
5150 spin_unlock(&mddev->lock);
5154 static struct md_sysfs_entry md_max_sync =
5155 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5158 suspend_lo_show(struct mddev *mddev, char *page)
5160 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5164 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5166 unsigned long long new;
5169 err = kstrtoull(buf, 10, &new);
5172 if (new != (sector_t)new)
5175 err = mddev_lock(mddev);
5179 if (mddev->pers == NULL ||
5180 mddev->pers->quiesce == NULL)
5182 mddev_suspend(mddev);
5183 mddev->suspend_lo = new;
5184 mddev_resume(mddev);
5188 mddev_unlock(mddev);
5191 static struct md_sysfs_entry md_suspend_lo =
5192 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5195 suspend_hi_show(struct mddev *mddev, char *page)
5197 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5201 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5203 unsigned long long new;
5206 err = kstrtoull(buf, 10, &new);
5209 if (new != (sector_t)new)
5212 err = mddev_lock(mddev);
5216 if (mddev->pers == NULL)
5219 mddev_suspend(mddev);
5220 mddev->suspend_hi = new;
5221 mddev_resume(mddev);
5225 mddev_unlock(mddev);
5228 static struct md_sysfs_entry md_suspend_hi =
5229 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5232 reshape_position_show(struct mddev *mddev, char *page)
5234 if (mddev->reshape_position != MaxSector)
5235 return sprintf(page, "%llu\n",
5236 (unsigned long long)mddev->reshape_position);
5237 strcpy(page, "none\n");
5242 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5244 struct md_rdev *rdev;
5245 unsigned long long new;
5248 err = kstrtoull(buf, 10, &new);
5251 if (new != (sector_t)new)
5253 err = mddev_lock(mddev);
5259 mddev->reshape_position = new;
5260 mddev->delta_disks = 0;
5261 mddev->reshape_backwards = 0;
5262 mddev->new_level = mddev->level;
5263 mddev->new_layout = mddev->layout;
5264 mddev->new_chunk_sectors = mddev->chunk_sectors;
5265 rdev_for_each(rdev, mddev)
5266 rdev->new_data_offset = rdev->data_offset;
5269 mddev_unlock(mddev);
5273 static struct md_sysfs_entry md_reshape_position =
5274 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5275 reshape_position_store);
5278 reshape_direction_show(struct mddev *mddev, char *page)
5280 return sprintf(page, "%s\n",
5281 mddev->reshape_backwards ? "backwards" : "forwards");
5285 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5290 if (cmd_match(buf, "forwards"))
5292 else if (cmd_match(buf, "backwards"))
5296 if (mddev->reshape_backwards == backwards)
5299 err = mddev_lock(mddev);
5302 /* check if we are allowed to change */
5303 if (mddev->delta_disks)
5305 else if (mddev->persistent &&
5306 mddev->major_version == 0)
5309 mddev->reshape_backwards = backwards;
5310 mddev_unlock(mddev);
5314 static struct md_sysfs_entry md_reshape_direction =
5315 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5316 reshape_direction_store);
5319 array_size_show(struct mddev *mddev, char *page)
5321 if (mddev->external_size)
5322 return sprintf(page, "%llu\n",
5323 (unsigned long long)mddev->array_sectors/2);
5325 return sprintf(page, "default\n");
5329 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5334 err = mddev_lock(mddev);
5338 /* cluster raid doesn't support change array_sectors */
5339 if (mddev_is_clustered(mddev)) {
5340 mddev_unlock(mddev);
5344 if (strncmp(buf, "default", 7) == 0) {
5346 sectors = mddev->pers->size(mddev, 0, 0);
5348 sectors = mddev->array_sectors;
5350 mddev->external_size = 0;
5352 if (strict_blocks_to_sectors(buf, §ors) < 0)
5354 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5357 mddev->external_size = 1;
5361 mddev->array_sectors = sectors;
5363 set_capacity(mddev->gendisk, mddev->array_sectors);
5364 revalidate_disk_size(mddev->gendisk, true);
5367 mddev_unlock(mddev);
5371 static struct md_sysfs_entry md_array_size =
5372 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5376 consistency_policy_show(struct mddev *mddev, char *page)
5380 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5381 ret = sprintf(page, "journal\n");
5382 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5383 ret = sprintf(page, "ppl\n");
5384 } else if (mddev->bitmap) {
5385 ret = sprintf(page, "bitmap\n");
5386 } else if (mddev->pers) {
5387 if (mddev->pers->sync_request)
5388 ret = sprintf(page, "resync\n");
5390 ret = sprintf(page, "none\n");
5392 ret = sprintf(page, "unknown\n");
5399 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5404 if (mddev->pers->change_consistency_policy)
5405 err = mddev->pers->change_consistency_policy(mddev, buf);
5408 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5409 set_bit(MD_HAS_PPL, &mddev->flags);
5414 return err ? err : len;
5417 static struct md_sysfs_entry md_consistency_policy =
5418 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5419 consistency_policy_store);
5421 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5423 return sprintf(page, "%d\n", mddev->fail_last_dev);
5427 * Setting fail_last_dev to true to allow last device to be forcibly removed
5428 * from RAID1/RAID10.
5431 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5436 ret = kstrtobool(buf, &value);
5440 if (value != mddev->fail_last_dev)
5441 mddev->fail_last_dev = value;
5445 static struct md_sysfs_entry md_fail_last_dev =
5446 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5447 fail_last_dev_store);
5449 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5451 if (mddev->pers == NULL || (mddev->pers->level != 1))
5452 return sprintf(page, "n/a\n");
5454 return sprintf(page, "%d\n", mddev->serialize_policy);
5458 * Setting serialize_policy to true to enforce write IO is not reordered
5462 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5467 err = kstrtobool(buf, &value);
5471 if (value == mddev->serialize_policy)
5474 err = mddev_lock(mddev);
5477 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5478 pr_err("md: serialize_policy is only effective for raid1\n");
5483 mddev_suspend(mddev);
5485 mddev_create_serial_pool(mddev, NULL, true);
5487 mddev_destroy_serial_pool(mddev, NULL, true);
5488 mddev->serialize_policy = value;
5489 mddev_resume(mddev);
5491 mddev_unlock(mddev);
5495 static struct md_sysfs_entry md_serialize_policy =
5496 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5497 serialize_policy_store);
5500 static struct attribute *md_default_attrs[] = {
5503 &md_raid_disks.attr,
5505 &md_chunk_size.attr,
5507 &md_resync_start.attr,
5509 &md_new_device.attr,
5510 &md_safe_delay.attr,
5511 &md_array_state.attr,
5512 &md_reshape_position.attr,
5513 &md_reshape_direction.attr,
5514 &md_array_size.attr,
5515 &max_corr_read_errors.attr,
5516 &md_consistency_policy.attr,
5517 &md_fail_last_dev.attr,
5518 &md_serialize_policy.attr,
5522 static struct attribute *md_redundancy_attrs[] = {
5524 &md_last_scan_mode.attr,
5525 &md_mismatches.attr,
5528 &md_sync_speed.attr,
5529 &md_sync_force_parallel.attr,
5530 &md_sync_completed.attr,
5533 &md_suspend_lo.attr,
5534 &md_suspend_hi.attr,
5539 static struct attribute_group md_redundancy_group = {
5541 .attrs = md_redundancy_attrs,
5545 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5547 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5548 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5553 spin_lock(&all_mddevs_lock);
5554 if (list_empty(&mddev->all_mddevs)) {
5555 spin_unlock(&all_mddevs_lock);
5559 spin_unlock(&all_mddevs_lock);
5561 rv = entry->show(mddev, page);
5567 md_attr_store(struct kobject *kobj, struct attribute *attr,
5568 const char *page, size_t length)
5570 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5571 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5576 if (!capable(CAP_SYS_ADMIN))
5578 spin_lock(&all_mddevs_lock);
5579 if (list_empty(&mddev->all_mddevs)) {
5580 spin_unlock(&all_mddevs_lock);
5584 spin_unlock(&all_mddevs_lock);
5585 rv = entry->store(mddev, page, length);
5590 static void md_free(struct kobject *ko)
5592 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5594 if (mddev->sysfs_state)
5595 sysfs_put(mddev->sysfs_state);
5596 if (mddev->sysfs_level)
5597 sysfs_put(mddev->sysfs_level);
5600 del_gendisk(mddev->gendisk);
5602 blk_cleanup_queue(mddev->queue);
5604 put_disk(mddev->gendisk);
5605 percpu_ref_exit(&mddev->writes_pending);
5607 bioset_exit(&mddev->bio_set);
5608 bioset_exit(&mddev->sync_set);
5609 mempool_exit(&mddev->md_io_pool);
5613 static const struct sysfs_ops md_sysfs_ops = {
5614 .show = md_attr_show,
5615 .store = md_attr_store,
5617 static struct kobj_type md_ktype = {
5619 .sysfs_ops = &md_sysfs_ops,
5620 .default_attrs = md_default_attrs,
5625 static void mddev_delayed_delete(struct work_struct *ws)
5627 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5629 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5630 kobject_del(&mddev->kobj);
5631 kobject_put(&mddev->kobj);
5634 static void no_op(struct percpu_ref *r) {}
5636 int mddev_init_writes_pending(struct mddev *mddev)
5638 if (mddev->writes_pending.percpu_count_ptr)
5640 if (percpu_ref_init(&mddev->writes_pending, no_op,
5641 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5643 /* We want to start with the refcount at zero */
5644 percpu_ref_put(&mddev->writes_pending);
5647 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5649 static int md_alloc(dev_t dev, char *name)
5652 * If dev is zero, name is the name of a device to allocate with
5653 * an arbitrary minor number. It will be "md_???"
5654 * If dev is non-zero it must be a device number with a MAJOR of
5655 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5656 * the device is being created by opening a node in /dev.
5657 * If "name" is not NULL, the device is being created by
5658 * writing to /sys/module/md_mod/parameters/new_array.
5660 static DEFINE_MUTEX(disks_mutex);
5661 struct mddev *mddev = mddev_find(dev);
5662 struct gendisk *disk;
5671 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5672 shift = partitioned ? MdpMinorShift : 0;
5673 unit = MINOR(mddev->unit) >> shift;
5675 /* wait for any previous instance of this device to be
5676 * completely removed (mddev_delayed_delete).
5678 flush_workqueue(md_misc_wq);
5680 mutex_lock(&disks_mutex);
5686 /* Need to ensure that 'name' is not a duplicate.
5688 struct mddev *mddev2;
5689 spin_lock(&all_mddevs_lock);
5691 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5692 if (mddev2->gendisk &&
5693 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5694 spin_unlock(&all_mddevs_lock);
5697 spin_unlock(&all_mddevs_lock);
5701 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5703 mddev->hold_active = UNTIL_STOP;
5705 error = mempool_init_kmalloc_pool(&mddev->md_io_pool, BIO_POOL_SIZE,
5706 sizeof(struct md_io));
5711 mddev->queue = blk_alloc_queue(NUMA_NO_NODE);
5715 blk_set_stacking_limits(&mddev->queue->limits);
5717 disk = alloc_disk(1 << shift);
5719 blk_cleanup_queue(mddev->queue);
5720 mddev->queue = NULL;
5723 disk->major = MAJOR(mddev->unit);
5724 disk->first_minor = unit << shift;
5726 strcpy(disk->disk_name, name);
5727 else if (partitioned)
5728 sprintf(disk->disk_name, "md_d%d", unit);
5730 sprintf(disk->disk_name, "md%d", unit);
5731 disk->fops = &md_fops;
5732 disk->private_data = mddev;
5733 disk->queue = mddev->queue;
5734 blk_queue_write_cache(mddev->queue, true, true);
5735 /* Allow extended partitions. This makes the
5736 * 'mdp' device redundant, but we can't really
5739 disk->flags |= GENHD_FL_EXT_DEVT;
5740 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5741 mddev->gendisk = disk;
5742 /* As soon as we call add_disk(), another thread could get
5743 * through to md_open, so make sure it doesn't get too far
5745 mutex_lock(&mddev->open_mutex);
5748 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5750 /* This isn't possible, but as kobject_init_and_add is marked
5751 * __must_check, we must do something with the result
5753 pr_debug("md: cannot register %s/md - name in use\n",
5757 if (mddev->kobj.sd &&
5758 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5759 pr_debug("pointless warning\n");
5760 mutex_unlock(&mddev->open_mutex);
5762 mutex_unlock(&disks_mutex);
5763 if (!error && mddev->kobj.sd) {
5764 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5765 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5766 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5772 static void md_probe(dev_t dev)
5774 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5777 md_alloc(dev, NULL);
5780 static int add_named_array(const char *val, const struct kernel_param *kp)
5783 * val must be "md_*" or "mdNNN".
5784 * For "md_*" we allocate an array with a large free minor number, and
5785 * set the name to val. val must not already be an active name.
5786 * For "mdNNN" we allocate an array with the minor number NNN
5787 * which must not already be in use.
5789 int len = strlen(val);
5790 char buf[DISK_NAME_LEN];
5791 unsigned long devnum;
5793 while (len && val[len-1] == '\n')
5795 if (len >= DISK_NAME_LEN)
5797 strlcpy(buf, val, len+1);
5798 if (strncmp(buf, "md_", 3) == 0)
5799 return md_alloc(0, buf);
5800 if (strncmp(buf, "md", 2) == 0 &&
5802 kstrtoul(buf+2, 10, &devnum) == 0 &&
5803 devnum <= MINORMASK)
5804 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5809 static void md_safemode_timeout(struct timer_list *t)
5811 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5813 mddev->safemode = 1;
5814 if (mddev->external)
5815 sysfs_notify_dirent_safe(mddev->sysfs_state);
5817 md_wakeup_thread(mddev->thread);
5820 static int start_dirty_degraded;
5822 int md_run(struct mddev *mddev)
5825 struct md_rdev *rdev;
5826 struct md_personality *pers;
5828 if (list_empty(&mddev->disks))
5829 /* cannot run an array with no devices.. */
5834 /* Cannot run until previous stop completes properly */
5835 if (mddev->sysfs_active)
5839 * Analyze all RAID superblock(s)
5841 if (!mddev->raid_disks) {
5842 if (!mddev->persistent)
5844 err = analyze_sbs(mddev);
5849 if (mddev->level != LEVEL_NONE)
5850 request_module("md-level-%d", mddev->level);
5851 else if (mddev->clevel[0])
5852 request_module("md-%s", mddev->clevel);
5855 * Drop all container device buffers, from now on
5856 * the only valid external interface is through the md
5859 mddev->has_superblocks = false;
5860 rdev_for_each(rdev, mddev) {
5861 if (test_bit(Faulty, &rdev->flags))
5863 sync_blockdev(rdev->bdev);
5864 invalidate_bdev(rdev->bdev);
5865 if (mddev->ro != 1 &&
5866 (bdev_read_only(rdev->bdev) ||
5867 bdev_read_only(rdev->meta_bdev))) {
5870 set_disk_ro(mddev->gendisk, 1);
5874 mddev->has_superblocks = true;
5876 /* perform some consistency tests on the device.
5877 * We don't want the data to overlap the metadata,
5878 * Internal Bitmap issues have been handled elsewhere.
5880 if (rdev->meta_bdev) {
5881 /* Nothing to check */;
5882 } else if (rdev->data_offset < rdev->sb_start) {
5883 if (mddev->dev_sectors &&
5884 rdev->data_offset + mddev->dev_sectors
5886 pr_warn("md: %s: data overlaps metadata\n",
5891 if (rdev->sb_start + rdev->sb_size/512
5892 > rdev->data_offset) {
5893 pr_warn("md: %s: metadata overlaps data\n",
5898 sysfs_notify_dirent_safe(rdev->sysfs_state);
5901 if (!bioset_initialized(&mddev->bio_set)) {
5902 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5906 if (!bioset_initialized(&mddev->sync_set)) {
5907 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5912 spin_lock(&pers_lock);
5913 pers = find_pers(mddev->level, mddev->clevel);
5914 if (!pers || !try_module_get(pers->owner)) {
5915 spin_unlock(&pers_lock);
5916 if (mddev->level != LEVEL_NONE)
5917 pr_warn("md: personality for level %d is not loaded!\n",
5920 pr_warn("md: personality for level %s is not loaded!\n",
5925 spin_unlock(&pers_lock);
5926 if (mddev->level != pers->level) {
5927 mddev->level = pers->level;
5928 mddev->new_level = pers->level;
5930 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5932 if (mddev->reshape_position != MaxSector &&
5933 pers->start_reshape == NULL) {
5934 /* This personality cannot handle reshaping... */
5935 module_put(pers->owner);
5940 if (pers->sync_request) {
5941 /* Warn if this is a potentially silly
5944 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5945 struct md_rdev *rdev2;
5948 rdev_for_each(rdev, mddev)
5949 rdev_for_each(rdev2, mddev) {
5951 rdev->bdev->bd_disk ==
5952 rdev2->bdev->bd_disk) {
5953 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5955 bdevname(rdev->bdev,b),
5956 bdevname(rdev2->bdev,b2));
5962 pr_warn("True protection against single-disk failure might be compromised.\n");
5965 mddev->recovery = 0;
5966 /* may be over-ridden by personality */
5967 mddev->resync_max_sectors = mddev->dev_sectors;
5969 mddev->ok_start_degraded = start_dirty_degraded;
5971 if (start_readonly && mddev->ro == 0)
5972 mddev->ro = 2; /* read-only, but switch on first write */
5974 err = pers->run(mddev);
5976 pr_warn("md: pers->run() failed ...\n");
5977 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5978 WARN_ONCE(!mddev->external_size,
5979 "%s: default size too small, but 'external_size' not in effect?\n",
5981 pr_warn("md: invalid array_size %llu > default size %llu\n",
5982 (unsigned long long)mddev->array_sectors / 2,
5983 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5986 if (err == 0 && pers->sync_request &&
5987 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5988 struct bitmap *bitmap;
5990 bitmap = md_bitmap_create(mddev, -1);
5991 if (IS_ERR(bitmap)) {
5992 err = PTR_ERR(bitmap);
5993 pr_warn("%s: failed to create bitmap (%d)\n",
5994 mdname(mddev), err);
5996 mddev->bitmap = bitmap;
6002 if (mddev->bitmap_info.max_write_behind > 0) {
6003 bool create_pool = false;
6005 rdev_for_each(rdev, mddev) {
6006 if (test_bit(WriteMostly, &rdev->flags) &&
6007 rdev_init_serial(rdev))
6010 if (create_pool && mddev->serial_info_pool == NULL) {
6011 mddev->serial_info_pool =
6012 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6013 sizeof(struct serial_info));
6014 if (!mddev->serial_info_pool) {
6024 rdev_for_each(rdev, mddev) {
6025 if (rdev->raid_disk >= 0 &&
6026 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
6031 if (mddev->degraded)
6034 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
6036 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
6038 if (pers->sync_request) {
6039 if (mddev->kobj.sd &&
6040 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6041 pr_warn("md: cannot register extra attributes for %s\n",
6043 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6044 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6045 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6046 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
6049 atomic_set(&mddev->max_corr_read_errors,
6050 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6051 mddev->safemode = 0;
6052 if (mddev_is_clustered(mddev))
6053 mddev->safemode_delay = 0;
6055 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6058 spin_lock(&mddev->lock);
6060 spin_unlock(&mddev->lock);
6061 rdev_for_each(rdev, mddev)
6062 if (rdev->raid_disk >= 0)
6063 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6065 if (mddev->degraded && !mddev->ro)
6066 /* This ensures that recovering status is reported immediately
6067 * via sysfs - until a lack of spares is confirmed.
6069 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6072 if (mddev->sb_flags)
6073 md_update_sb(mddev, 0);
6075 md_new_event(mddev);
6079 mddev_detach(mddev);
6081 pers->free(mddev, mddev->private);
6082 mddev->private = NULL;
6083 module_put(pers->owner);
6084 md_bitmap_destroy(mddev);
6086 bioset_exit(&mddev->bio_set);
6087 bioset_exit(&mddev->sync_set);
6090 EXPORT_SYMBOL_GPL(md_run);
6092 int do_md_run(struct mddev *mddev)
6096 set_bit(MD_NOT_READY, &mddev->flags);
6097 err = md_run(mddev);
6100 err = md_bitmap_load(mddev);
6102 md_bitmap_destroy(mddev);
6106 if (mddev_is_clustered(mddev))
6107 md_allow_write(mddev);
6109 /* run start up tasks that require md_thread */
6112 md_wakeup_thread(mddev->thread);
6113 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6115 set_capacity(mddev->gendisk, mddev->array_sectors);
6116 revalidate_disk_size(mddev->gendisk, true);
6117 clear_bit(MD_NOT_READY, &mddev->flags);
6119 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6120 sysfs_notify_dirent_safe(mddev->sysfs_state);
6121 sysfs_notify_dirent_safe(mddev->sysfs_action);
6122 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6124 clear_bit(MD_NOT_READY, &mddev->flags);
6128 int md_start(struct mddev *mddev)
6132 if (mddev->pers->start) {
6133 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6134 md_wakeup_thread(mddev->thread);
6135 ret = mddev->pers->start(mddev);
6136 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6137 md_wakeup_thread(mddev->sync_thread);
6141 EXPORT_SYMBOL_GPL(md_start);
6143 static int restart_array(struct mddev *mddev)
6145 struct gendisk *disk = mddev->gendisk;
6146 struct md_rdev *rdev;
6147 bool has_journal = false;
6148 bool has_readonly = false;
6150 /* Complain if it has no devices */
6151 if (list_empty(&mddev->disks))
6159 rdev_for_each_rcu(rdev, mddev) {
6160 if (test_bit(Journal, &rdev->flags) &&
6161 !test_bit(Faulty, &rdev->flags))
6163 if (bdev_read_only(rdev->bdev))
6164 has_readonly = true;
6167 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6168 /* Don't restart rw with journal missing/faulty */
6173 mddev->safemode = 0;
6175 set_disk_ro(disk, 0);
6176 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6177 /* Kick recovery or resync if necessary */
6178 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6179 md_wakeup_thread(mddev->thread);
6180 md_wakeup_thread(mddev->sync_thread);
6181 sysfs_notify_dirent_safe(mddev->sysfs_state);
6185 static void md_clean(struct mddev *mddev)
6187 mddev->array_sectors = 0;
6188 mddev->external_size = 0;
6189 mddev->dev_sectors = 0;
6190 mddev->raid_disks = 0;
6191 mddev->recovery_cp = 0;
6192 mddev->resync_min = 0;
6193 mddev->resync_max = MaxSector;
6194 mddev->reshape_position = MaxSector;
6195 mddev->external = 0;
6196 mddev->persistent = 0;
6197 mddev->level = LEVEL_NONE;
6198 mddev->clevel[0] = 0;
6200 mddev->sb_flags = 0;
6202 mddev->metadata_type[0] = 0;
6203 mddev->chunk_sectors = 0;
6204 mddev->ctime = mddev->utime = 0;
6206 mddev->max_disks = 0;
6208 mddev->can_decrease_events = 0;
6209 mddev->delta_disks = 0;
6210 mddev->reshape_backwards = 0;
6211 mddev->new_level = LEVEL_NONE;
6212 mddev->new_layout = 0;
6213 mddev->new_chunk_sectors = 0;
6214 mddev->curr_resync = 0;
6215 atomic64_set(&mddev->resync_mismatches, 0);
6216 mddev->suspend_lo = mddev->suspend_hi = 0;
6217 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6218 mddev->recovery = 0;
6221 mddev->degraded = 0;
6222 mddev->safemode = 0;
6223 mddev->private = NULL;
6224 mddev->cluster_info = NULL;
6225 mddev->bitmap_info.offset = 0;
6226 mddev->bitmap_info.default_offset = 0;
6227 mddev->bitmap_info.default_space = 0;
6228 mddev->bitmap_info.chunksize = 0;
6229 mddev->bitmap_info.daemon_sleep = 0;
6230 mddev->bitmap_info.max_write_behind = 0;
6231 mddev->bitmap_info.nodes = 0;
6234 static void __md_stop_writes(struct mddev *mddev)
6236 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6237 if (work_pending(&mddev->del_work))
6238 flush_workqueue(md_misc_wq);
6239 if (mddev->sync_thread) {
6240 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6241 md_reap_sync_thread(mddev);
6244 del_timer_sync(&mddev->safemode_timer);
6246 if (mddev->pers && mddev->pers->quiesce) {
6247 mddev->pers->quiesce(mddev, 1);
6248 mddev->pers->quiesce(mddev, 0);
6250 md_bitmap_flush(mddev);
6252 if (mddev->ro == 0 &&
6253 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6255 /* mark array as shutdown cleanly */
6256 if (!mddev_is_clustered(mddev))
6258 md_update_sb(mddev, 1);
6260 /* disable policy to guarantee rdevs free resources for serialization */
6261 mddev->serialize_policy = 0;
6262 mddev_destroy_serial_pool(mddev, NULL, true);
6265 void md_stop_writes(struct mddev *mddev)
6267 mddev_lock_nointr(mddev);
6268 __md_stop_writes(mddev);
6269 mddev_unlock(mddev);
6271 EXPORT_SYMBOL_GPL(md_stop_writes);
6273 static void mddev_detach(struct mddev *mddev)
6275 md_bitmap_wait_behind_writes(mddev);
6276 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
6277 mddev->pers->quiesce(mddev, 1);
6278 mddev->pers->quiesce(mddev, 0);
6280 md_unregister_thread(&mddev->thread);
6282 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6285 static void __md_stop(struct mddev *mddev)
6287 struct md_personality *pers = mddev->pers;
6288 md_bitmap_destroy(mddev);
6289 mddev_detach(mddev);
6290 /* Ensure ->event_work is done */
6291 if (mddev->event_work.func)
6292 flush_workqueue(md_misc_wq);
6293 spin_lock(&mddev->lock);
6295 spin_unlock(&mddev->lock);
6296 pers->free(mddev, mddev->private);
6297 mddev->private = NULL;
6298 if (pers->sync_request && mddev->to_remove == NULL)
6299 mddev->to_remove = &md_redundancy_group;
6300 module_put(pers->owner);
6301 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6304 void md_stop(struct mddev *mddev)
6306 /* stop the array and free an attached data structures.
6307 * This is called from dm-raid
6310 bioset_exit(&mddev->bio_set);
6311 bioset_exit(&mddev->sync_set);
6314 EXPORT_SYMBOL_GPL(md_stop);
6316 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6321 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6323 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6324 md_wakeup_thread(mddev->thread);
6326 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6327 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6328 if (mddev->sync_thread)
6329 /* Thread might be blocked waiting for metadata update
6330 * which will now never happen */
6331 wake_up_process(mddev->sync_thread->tsk);
6333 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6335 mddev_unlock(mddev);
6336 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6338 wait_event(mddev->sb_wait,
6339 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6340 mddev_lock_nointr(mddev);
6342 mutex_lock(&mddev->open_mutex);
6343 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6344 mddev->sync_thread ||
6345 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6346 pr_warn("md: %s still in use.\n",mdname(mddev));
6348 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6349 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6350 md_wakeup_thread(mddev->thread);
6356 __md_stop_writes(mddev);
6362 set_disk_ro(mddev->gendisk, 1);
6363 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6364 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6365 md_wakeup_thread(mddev->thread);
6366 sysfs_notify_dirent_safe(mddev->sysfs_state);
6370 mutex_unlock(&mddev->open_mutex);
6375 * 0 - completely stop and dis-assemble array
6376 * 2 - stop but do not disassemble array
6378 static int do_md_stop(struct mddev *mddev, int mode,
6379 struct block_device *bdev)
6381 struct gendisk *disk = mddev->gendisk;
6382 struct md_rdev *rdev;
6385 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6387 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6388 md_wakeup_thread(mddev->thread);
6390 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6391 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6392 if (mddev->sync_thread)
6393 /* Thread might be blocked waiting for metadata update
6394 * which will now never happen */
6395 wake_up_process(mddev->sync_thread->tsk);
6397 mddev_unlock(mddev);
6398 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6399 !test_bit(MD_RECOVERY_RUNNING,
6400 &mddev->recovery)));
6401 mddev_lock_nointr(mddev);
6403 mutex_lock(&mddev->open_mutex);
6404 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6405 mddev->sysfs_active ||
6406 mddev->sync_thread ||
6407 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6408 pr_warn("md: %s still in use.\n",mdname(mddev));
6409 mutex_unlock(&mddev->open_mutex);
6411 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6412 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6413 md_wakeup_thread(mddev->thread);
6419 set_disk_ro(disk, 0);
6421 __md_stop_writes(mddev);
6424 /* tell userspace to handle 'inactive' */
6425 sysfs_notify_dirent_safe(mddev->sysfs_state);
6427 rdev_for_each(rdev, mddev)
6428 if (rdev->raid_disk >= 0)
6429 sysfs_unlink_rdev(mddev, rdev);
6431 set_capacity(disk, 0);
6432 mutex_unlock(&mddev->open_mutex);
6434 revalidate_disk_size(disk, true);
6439 mutex_unlock(&mddev->open_mutex);
6441 * Free resources if final stop
6444 pr_info("md: %s stopped.\n", mdname(mddev));
6446 if (mddev->bitmap_info.file) {
6447 struct file *f = mddev->bitmap_info.file;
6448 spin_lock(&mddev->lock);
6449 mddev->bitmap_info.file = NULL;
6450 spin_unlock(&mddev->lock);
6453 mddev->bitmap_info.offset = 0;
6455 export_array(mddev);
6458 if (mddev->hold_active == UNTIL_STOP)
6459 mddev->hold_active = 0;
6461 md_new_event(mddev);
6462 sysfs_notify_dirent_safe(mddev->sysfs_state);
6467 static void autorun_array(struct mddev *mddev)
6469 struct md_rdev *rdev;
6472 if (list_empty(&mddev->disks))
6475 pr_info("md: running: ");
6477 rdev_for_each(rdev, mddev) {
6478 char b[BDEVNAME_SIZE];
6479 pr_cont("<%s>", bdevname(rdev->bdev,b));
6483 err = do_md_run(mddev);
6485 pr_warn("md: do_md_run() returned %d\n", err);
6486 do_md_stop(mddev, 0, NULL);
6491 * lets try to run arrays based on all disks that have arrived
6492 * until now. (those are in pending_raid_disks)
6494 * the method: pick the first pending disk, collect all disks with
6495 * the same UUID, remove all from the pending list and put them into
6496 * the 'same_array' list. Then order this list based on superblock
6497 * update time (freshest comes first), kick out 'old' disks and
6498 * compare superblocks. If everything's fine then run it.
6500 * If "unit" is allocated, then bump its reference count
6502 static void autorun_devices(int part)
6504 struct md_rdev *rdev0, *rdev, *tmp;
6505 struct mddev *mddev;
6506 char b[BDEVNAME_SIZE];
6508 pr_info("md: autorun ...\n");
6509 while (!list_empty(&pending_raid_disks)) {
6512 LIST_HEAD(candidates);
6513 rdev0 = list_entry(pending_raid_disks.next,
6514 struct md_rdev, same_set);
6516 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6517 INIT_LIST_HEAD(&candidates);
6518 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6519 if (super_90_load(rdev, rdev0, 0) >= 0) {
6520 pr_debug("md: adding %s ...\n",
6521 bdevname(rdev->bdev,b));
6522 list_move(&rdev->same_set, &candidates);
6525 * now we have a set of devices, with all of them having
6526 * mostly sane superblocks. It's time to allocate the
6530 dev = MKDEV(mdp_major,
6531 rdev0->preferred_minor << MdpMinorShift);
6532 unit = MINOR(dev) >> MdpMinorShift;
6534 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6537 if (rdev0->preferred_minor != unit) {
6538 pr_warn("md: unit number in %s is bad: %d\n",
6539 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6544 mddev = mddev_find(dev);
6545 if (!mddev || !mddev->gendisk) {
6550 if (mddev_lock(mddev))
6551 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6552 else if (mddev->raid_disks || mddev->major_version
6553 || !list_empty(&mddev->disks)) {
6554 pr_warn("md: %s already running, cannot run %s\n",
6555 mdname(mddev), bdevname(rdev0->bdev,b));
6556 mddev_unlock(mddev);
6558 pr_debug("md: created %s\n", mdname(mddev));
6559 mddev->persistent = 1;
6560 rdev_for_each_list(rdev, tmp, &candidates) {
6561 list_del_init(&rdev->same_set);
6562 if (bind_rdev_to_array(rdev, mddev))
6565 autorun_array(mddev);
6566 mddev_unlock(mddev);
6568 /* on success, candidates will be empty, on error
6571 rdev_for_each_list(rdev, tmp, &candidates) {
6572 list_del_init(&rdev->same_set);
6577 pr_info("md: ... autorun DONE.\n");
6579 #endif /* !MODULE */
6581 static int get_version(void __user *arg)
6585 ver.major = MD_MAJOR_VERSION;
6586 ver.minor = MD_MINOR_VERSION;
6587 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6589 if (copy_to_user(arg, &ver, sizeof(ver)))
6595 static int get_array_info(struct mddev *mddev, void __user *arg)
6597 mdu_array_info_t info;
6598 int nr,working,insync,failed,spare;
6599 struct md_rdev *rdev;
6601 nr = working = insync = failed = spare = 0;
6603 rdev_for_each_rcu(rdev, mddev) {
6605 if (test_bit(Faulty, &rdev->flags))
6609 if (test_bit(In_sync, &rdev->flags))
6611 else if (test_bit(Journal, &rdev->flags))
6612 /* TODO: add journal count to md_u.h */
6620 info.major_version = mddev->major_version;
6621 info.minor_version = mddev->minor_version;
6622 info.patch_version = MD_PATCHLEVEL_VERSION;
6623 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6624 info.level = mddev->level;
6625 info.size = mddev->dev_sectors / 2;
6626 if (info.size != mddev->dev_sectors / 2) /* overflow */
6629 info.raid_disks = mddev->raid_disks;
6630 info.md_minor = mddev->md_minor;
6631 info.not_persistent= !mddev->persistent;
6633 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6636 info.state = (1<<MD_SB_CLEAN);
6637 if (mddev->bitmap && mddev->bitmap_info.offset)
6638 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6639 if (mddev_is_clustered(mddev))
6640 info.state |= (1<<MD_SB_CLUSTERED);
6641 info.active_disks = insync;
6642 info.working_disks = working;
6643 info.failed_disks = failed;
6644 info.spare_disks = spare;
6646 info.layout = mddev->layout;
6647 info.chunk_size = mddev->chunk_sectors << 9;
6649 if (copy_to_user(arg, &info, sizeof(info)))
6655 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6657 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6661 file = kzalloc(sizeof(*file), GFP_NOIO);
6666 spin_lock(&mddev->lock);
6667 /* bitmap enabled */
6668 if (mddev->bitmap_info.file) {
6669 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6670 sizeof(file->pathname));
6674 memmove(file->pathname, ptr,
6675 sizeof(file->pathname)-(ptr-file->pathname));
6677 spin_unlock(&mddev->lock);
6680 copy_to_user(arg, file, sizeof(*file)))
6687 static int get_disk_info(struct mddev *mddev, void __user * arg)
6689 mdu_disk_info_t info;
6690 struct md_rdev *rdev;
6692 if (copy_from_user(&info, arg, sizeof(info)))
6696 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6698 info.major = MAJOR(rdev->bdev->bd_dev);
6699 info.minor = MINOR(rdev->bdev->bd_dev);
6700 info.raid_disk = rdev->raid_disk;
6702 if (test_bit(Faulty, &rdev->flags))
6703 info.state |= (1<<MD_DISK_FAULTY);
6704 else if (test_bit(In_sync, &rdev->flags)) {
6705 info.state |= (1<<MD_DISK_ACTIVE);
6706 info.state |= (1<<MD_DISK_SYNC);
6708 if (test_bit(Journal, &rdev->flags))
6709 info.state |= (1<<MD_DISK_JOURNAL);
6710 if (test_bit(WriteMostly, &rdev->flags))
6711 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6712 if (test_bit(FailFast, &rdev->flags))
6713 info.state |= (1<<MD_DISK_FAILFAST);
6715 info.major = info.minor = 0;
6716 info.raid_disk = -1;
6717 info.state = (1<<MD_DISK_REMOVED);
6721 if (copy_to_user(arg, &info, sizeof(info)))
6727 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6729 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6730 struct md_rdev *rdev;
6731 dev_t dev = MKDEV(info->major,info->minor);
6733 if (mddev_is_clustered(mddev) &&
6734 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6735 pr_warn("%s: Cannot add to clustered mddev.\n",
6740 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6743 if (!mddev->raid_disks) {
6745 /* expecting a device which has a superblock */
6746 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6748 pr_warn("md: md_import_device returned %ld\n",
6750 return PTR_ERR(rdev);
6752 if (!list_empty(&mddev->disks)) {
6753 struct md_rdev *rdev0
6754 = list_entry(mddev->disks.next,
6755 struct md_rdev, same_set);
6756 err = super_types[mddev->major_version]
6757 .load_super(rdev, rdev0, mddev->minor_version);
6759 pr_warn("md: %s has different UUID to %s\n",
6760 bdevname(rdev->bdev,b),
6761 bdevname(rdev0->bdev,b2));
6766 err = bind_rdev_to_array(rdev, mddev);
6773 * md_add_new_disk can be used once the array is assembled
6774 * to add "hot spares". They must already have a superblock
6779 if (!mddev->pers->hot_add_disk) {
6780 pr_warn("%s: personality does not support diskops!\n",
6784 if (mddev->persistent)
6785 rdev = md_import_device(dev, mddev->major_version,
6786 mddev->minor_version);
6788 rdev = md_import_device(dev, -1, -1);
6790 pr_warn("md: md_import_device returned %ld\n",
6792 return PTR_ERR(rdev);
6794 /* set saved_raid_disk if appropriate */
6795 if (!mddev->persistent) {
6796 if (info->state & (1<<MD_DISK_SYNC) &&
6797 info->raid_disk < mddev->raid_disks) {
6798 rdev->raid_disk = info->raid_disk;
6799 set_bit(In_sync, &rdev->flags);
6800 clear_bit(Bitmap_sync, &rdev->flags);
6802 rdev->raid_disk = -1;
6803 rdev->saved_raid_disk = rdev->raid_disk;
6805 super_types[mddev->major_version].
6806 validate_super(mddev, rdev);
6807 if ((info->state & (1<<MD_DISK_SYNC)) &&
6808 rdev->raid_disk != info->raid_disk) {
6809 /* This was a hot-add request, but events doesn't
6810 * match, so reject it.
6816 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6817 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6818 set_bit(WriteMostly, &rdev->flags);
6820 clear_bit(WriteMostly, &rdev->flags);
6821 if (info->state & (1<<MD_DISK_FAILFAST))
6822 set_bit(FailFast, &rdev->flags);
6824 clear_bit(FailFast, &rdev->flags);
6826 if (info->state & (1<<MD_DISK_JOURNAL)) {
6827 struct md_rdev *rdev2;
6828 bool has_journal = false;
6830 /* make sure no existing journal disk */
6831 rdev_for_each(rdev2, mddev) {
6832 if (test_bit(Journal, &rdev2->flags)) {
6837 if (has_journal || mddev->bitmap) {
6841 set_bit(Journal, &rdev->flags);
6844 * check whether the device shows up in other nodes
6846 if (mddev_is_clustered(mddev)) {
6847 if (info->state & (1 << MD_DISK_CANDIDATE))
6848 set_bit(Candidate, &rdev->flags);
6849 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6850 /* --add initiated by this node */
6851 err = md_cluster_ops->add_new_disk(mddev, rdev);
6859 rdev->raid_disk = -1;
6860 err = bind_rdev_to_array(rdev, mddev);
6865 if (mddev_is_clustered(mddev)) {
6866 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6868 err = md_cluster_ops->new_disk_ack(mddev,
6871 md_kick_rdev_from_array(rdev);
6875 md_cluster_ops->add_new_disk_cancel(mddev);
6877 err = add_bound_rdev(rdev);
6881 err = add_bound_rdev(rdev);
6886 /* otherwise, md_add_new_disk is only allowed
6887 * for major_version==0 superblocks
6889 if (mddev->major_version != 0) {
6890 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6894 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6896 rdev = md_import_device(dev, -1, 0);
6898 pr_warn("md: error, md_import_device() returned %ld\n",
6900 return PTR_ERR(rdev);
6902 rdev->desc_nr = info->number;
6903 if (info->raid_disk < mddev->raid_disks)
6904 rdev->raid_disk = info->raid_disk;
6906 rdev->raid_disk = -1;
6908 if (rdev->raid_disk < mddev->raid_disks)
6909 if (info->state & (1<<MD_DISK_SYNC))
6910 set_bit(In_sync, &rdev->flags);
6912 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6913 set_bit(WriteMostly, &rdev->flags);
6914 if (info->state & (1<<MD_DISK_FAILFAST))
6915 set_bit(FailFast, &rdev->flags);
6917 if (!mddev->persistent) {
6918 pr_debug("md: nonpersistent superblock ...\n");
6919 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6921 rdev->sb_start = calc_dev_sboffset(rdev);
6922 rdev->sectors = rdev->sb_start;
6924 err = bind_rdev_to_array(rdev, mddev);
6934 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6936 char b[BDEVNAME_SIZE];
6937 struct md_rdev *rdev;
6942 rdev = find_rdev(mddev, dev);
6946 if (rdev->raid_disk < 0)
6949 clear_bit(Blocked, &rdev->flags);
6950 remove_and_add_spares(mddev, rdev);
6952 if (rdev->raid_disk >= 0)
6956 if (mddev_is_clustered(mddev))
6957 md_cluster_ops->remove_disk(mddev, rdev);
6959 md_kick_rdev_from_array(rdev);
6960 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6962 md_wakeup_thread(mddev->thread);
6964 md_update_sb(mddev, 1);
6965 md_new_event(mddev);
6969 pr_debug("md: cannot remove active disk %s from %s ...\n",
6970 bdevname(rdev->bdev,b), mdname(mddev));
6974 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6976 char b[BDEVNAME_SIZE];
6978 struct md_rdev *rdev;
6983 if (mddev->major_version != 0) {
6984 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6988 if (!mddev->pers->hot_add_disk) {
6989 pr_warn("%s: personality does not support diskops!\n",
6994 rdev = md_import_device(dev, -1, 0);
6996 pr_warn("md: error, md_import_device() returned %ld\n",
7001 if (mddev->persistent)
7002 rdev->sb_start = calc_dev_sboffset(rdev);
7004 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
7006 rdev->sectors = rdev->sb_start;
7008 if (test_bit(Faulty, &rdev->flags)) {
7009 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
7010 bdevname(rdev->bdev,b), mdname(mddev));
7015 clear_bit(In_sync, &rdev->flags);
7017 rdev->saved_raid_disk = -1;
7018 err = bind_rdev_to_array(rdev, mddev);
7023 * The rest should better be atomic, we can have disk failures
7024 * noticed in interrupt contexts ...
7027 rdev->raid_disk = -1;
7029 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7031 md_update_sb(mddev, 1);
7033 * Kick recovery, maybe this spare has to be added to the
7034 * array immediately.
7036 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7037 md_wakeup_thread(mddev->thread);
7038 md_new_event(mddev);
7046 static int set_bitmap_file(struct mddev *mddev, int fd)
7051 if (!mddev->pers->quiesce || !mddev->thread)
7053 if (mddev->recovery || mddev->sync_thread)
7055 /* we should be able to change the bitmap.. */
7059 struct inode *inode;
7062 if (mddev->bitmap || mddev->bitmap_info.file)
7063 return -EEXIST; /* cannot add when bitmap is present */
7067 pr_warn("%s: error: failed to get bitmap file\n",
7072 inode = f->f_mapping->host;
7073 if (!S_ISREG(inode->i_mode)) {
7074 pr_warn("%s: error: bitmap file must be a regular file\n",
7077 } else if (!(f->f_mode & FMODE_WRITE)) {
7078 pr_warn("%s: error: bitmap file must open for write\n",
7081 } else if (atomic_read(&inode->i_writecount) != 1) {
7082 pr_warn("%s: error: bitmap file is already in use\n",
7090 mddev->bitmap_info.file = f;
7091 mddev->bitmap_info.offset = 0; /* file overrides offset */
7092 } else if (mddev->bitmap == NULL)
7093 return -ENOENT; /* cannot remove what isn't there */
7097 struct bitmap *bitmap;
7099 bitmap = md_bitmap_create(mddev, -1);
7100 mddev_suspend(mddev);
7101 if (!IS_ERR(bitmap)) {
7102 mddev->bitmap = bitmap;
7103 err = md_bitmap_load(mddev);
7105 err = PTR_ERR(bitmap);
7107 md_bitmap_destroy(mddev);
7110 mddev_resume(mddev);
7111 } else if (fd < 0) {
7112 mddev_suspend(mddev);
7113 md_bitmap_destroy(mddev);
7114 mddev_resume(mddev);
7118 struct file *f = mddev->bitmap_info.file;
7120 spin_lock(&mddev->lock);
7121 mddev->bitmap_info.file = NULL;
7122 spin_unlock(&mddev->lock);
7131 * md_set_array_info is used two different ways
7132 * The original usage is when creating a new array.
7133 * In this usage, raid_disks is > 0 and it together with
7134 * level, size, not_persistent,layout,chunksize determine the
7135 * shape of the array.
7136 * This will always create an array with a type-0.90.0 superblock.
7137 * The newer usage is when assembling an array.
7138 * In this case raid_disks will be 0, and the major_version field is
7139 * use to determine which style super-blocks are to be found on the devices.
7140 * The minor and patch _version numbers are also kept incase the
7141 * super_block handler wishes to interpret them.
7143 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7145 if (info->raid_disks == 0) {
7146 /* just setting version number for superblock loading */
7147 if (info->major_version < 0 ||
7148 info->major_version >= ARRAY_SIZE(super_types) ||
7149 super_types[info->major_version].name == NULL) {
7150 /* maybe try to auto-load a module? */
7151 pr_warn("md: superblock version %d not known\n",
7152 info->major_version);
7155 mddev->major_version = info->major_version;
7156 mddev->minor_version = info->minor_version;
7157 mddev->patch_version = info->patch_version;
7158 mddev->persistent = !info->not_persistent;
7159 /* ensure mddev_put doesn't delete this now that there
7160 * is some minimal configuration.
7162 mddev->ctime = ktime_get_real_seconds();
7165 mddev->major_version = MD_MAJOR_VERSION;
7166 mddev->minor_version = MD_MINOR_VERSION;
7167 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7168 mddev->ctime = ktime_get_real_seconds();
7170 mddev->level = info->level;
7171 mddev->clevel[0] = 0;
7172 mddev->dev_sectors = 2 * (sector_t)info->size;
7173 mddev->raid_disks = info->raid_disks;
7174 /* don't set md_minor, it is determined by which /dev/md* was
7177 if (info->state & (1<<MD_SB_CLEAN))
7178 mddev->recovery_cp = MaxSector;
7180 mddev->recovery_cp = 0;
7181 mddev->persistent = ! info->not_persistent;
7182 mddev->external = 0;
7184 mddev->layout = info->layout;
7185 if (mddev->level == 0)
7186 /* Cannot trust RAID0 layout info here */
7188 mddev->chunk_sectors = info->chunk_size >> 9;
7190 if (mddev->persistent) {
7191 mddev->max_disks = MD_SB_DISKS;
7193 mddev->sb_flags = 0;
7195 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7197 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7198 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7199 mddev->bitmap_info.offset = 0;
7201 mddev->reshape_position = MaxSector;
7204 * Generate a 128 bit UUID
7206 get_random_bytes(mddev->uuid, 16);
7208 mddev->new_level = mddev->level;
7209 mddev->new_chunk_sectors = mddev->chunk_sectors;
7210 mddev->new_layout = mddev->layout;
7211 mddev->delta_disks = 0;
7212 mddev->reshape_backwards = 0;
7217 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7219 lockdep_assert_held(&mddev->reconfig_mutex);
7221 if (mddev->external_size)
7224 mddev->array_sectors = array_sectors;
7226 EXPORT_SYMBOL(md_set_array_sectors);
7228 static int update_size(struct mddev *mddev, sector_t num_sectors)
7230 struct md_rdev *rdev;
7232 int fit = (num_sectors == 0);
7233 sector_t old_dev_sectors = mddev->dev_sectors;
7235 if (mddev->pers->resize == NULL)
7237 /* The "num_sectors" is the number of sectors of each device that
7238 * is used. This can only make sense for arrays with redundancy.
7239 * linear and raid0 always use whatever space is available. We can only
7240 * consider changing this number if no resync or reconstruction is
7241 * happening, and if the new size is acceptable. It must fit before the
7242 * sb_start or, if that is <data_offset, it must fit before the size
7243 * of each device. If num_sectors is zero, we find the largest size
7246 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7252 rdev_for_each(rdev, mddev) {
7253 sector_t avail = rdev->sectors;
7255 if (fit && (num_sectors == 0 || num_sectors > avail))
7256 num_sectors = avail;
7257 if (avail < num_sectors)
7260 rv = mddev->pers->resize(mddev, num_sectors);
7262 if (mddev_is_clustered(mddev))
7263 md_cluster_ops->update_size(mddev, old_dev_sectors);
7264 else if (mddev->queue) {
7265 set_capacity(mddev->gendisk, mddev->array_sectors);
7266 revalidate_disk_size(mddev->gendisk, true);
7272 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7275 struct md_rdev *rdev;
7276 /* change the number of raid disks */
7277 if (mddev->pers->check_reshape == NULL)
7281 if (raid_disks <= 0 ||
7282 (mddev->max_disks && raid_disks >= mddev->max_disks))
7284 if (mddev->sync_thread ||
7285 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7286 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7287 mddev->reshape_position != MaxSector)
7290 rdev_for_each(rdev, mddev) {
7291 if (mddev->raid_disks < raid_disks &&
7292 rdev->data_offset < rdev->new_data_offset)
7294 if (mddev->raid_disks > raid_disks &&
7295 rdev->data_offset > rdev->new_data_offset)
7299 mddev->delta_disks = raid_disks - mddev->raid_disks;
7300 if (mddev->delta_disks < 0)
7301 mddev->reshape_backwards = 1;
7302 else if (mddev->delta_disks > 0)
7303 mddev->reshape_backwards = 0;
7305 rv = mddev->pers->check_reshape(mddev);
7307 mddev->delta_disks = 0;
7308 mddev->reshape_backwards = 0;
7314 * update_array_info is used to change the configuration of an
7316 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7317 * fields in the info are checked against the array.
7318 * Any differences that cannot be handled will cause an error.
7319 * Normally, only one change can be managed at a time.
7321 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7327 /* calculate expected state,ignoring low bits */
7328 if (mddev->bitmap && mddev->bitmap_info.offset)
7329 state |= (1 << MD_SB_BITMAP_PRESENT);
7331 if (mddev->major_version != info->major_version ||
7332 mddev->minor_version != info->minor_version ||
7333 /* mddev->patch_version != info->patch_version || */
7334 mddev->ctime != info->ctime ||
7335 mddev->level != info->level ||
7336 /* mddev->layout != info->layout || */
7337 mddev->persistent != !info->not_persistent ||
7338 mddev->chunk_sectors != info->chunk_size >> 9 ||
7339 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7340 ((state^info->state) & 0xfffffe00)
7343 /* Check there is only one change */
7344 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7346 if (mddev->raid_disks != info->raid_disks)
7348 if (mddev->layout != info->layout)
7350 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7357 if (mddev->layout != info->layout) {
7359 * we don't need to do anything at the md level, the
7360 * personality will take care of it all.
7362 if (mddev->pers->check_reshape == NULL)
7365 mddev->new_layout = info->layout;
7366 rv = mddev->pers->check_reshape(mddev);
7368 mddev->new_layout = mddev->layout;
7372 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7373 rv = update_size(mddev, (sector_t)info->size * 2);
7375 if (mddev->raid_disks != info->raid_disks)
7376 rv = update_raid_disks(mddev, info->raid_disks);
7378 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7379 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7383 if (mddev->recovery || mddev->sync_thread) {
7387 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7388 struct bitmap *bitmap;
7389 /* add the bitmap */
7390 if (mddev->bitmap) {
7394 if (mddev->bitmap_info.default_offset == 0) {
7398 mddev->bitmap_info.offset =
7399 mddev->bitmap_info.default_offset;
7400 mddev->bitmap_info.space =
7401 mddev->bitmap_info.default_space;
7402 bitmap = md_bitmap_create(mddev, -1);
7403 mddev_suspend(mddev);
7404 if (!IS_ERR(bitmap)) {
7405 mddev->bitmap = bitmap;
7406 rv = md_bitmap_load(mddev);
7408 rv = PTR_ERR(bitmap);
7410 md_bitmap_destroy(mddev);
7411 mddev_resume(mddev);
7413 /* remove the bitmap */
7414 if (!mddev->bitmap) {
7418 if (mddev->bitmap->storage.file) {
7422 if (mddev->bitmap_info.nodes) {
7423 /* hold PW on all the bitmap lock */
7424 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7425 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7427 md_cluster_ops->unlock_all_bitmaps(mddev);
7431 mddev->bitmap_info.nodes = 0;
7432 md_cluster_ops->leave(mddev);
7433 module_put(md_cluster_mod);
7434 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7436 mddev_suspend(mddev);
7437 md_bitmap_destroy(mddev);
7438 mddev_resume(mddev);
7439 mddev->bitmap_info.offset = 0;
7442 md_update_sb(mddev, 1);
7448 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7450 struct md_rdev *rdev;
7453 if (mddev->pers == NULL)
7457 rdev = md_find_rdev_rcu(mddev, dev);
7461 md_error(mddev, rdev);
7462 if (!test_bit(Faulty, &rdev->flags))
7470 * We have a problem here : there is no easy way to give a CHS
7471 * virtual geometry. We currently pretend that we have a 2 heads
7472 * 4 sectors (with a BIG number of cylinders...). This drives
7473 * dosfs just mad... ;-)
7475 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7477 struct mddev *mddev = bdev->bd_disk->private_data;
7481 geo->cylinders = mddev->array_sectors / 8;
7485 static inline bool md_ioctl_valid(unsigned int cmd)
7489 case GET_ARRAY_INFO:
7490 case GET_BITMAP_FILE:
7493 case HOT_REMOVE_DISK:
7495 case RESTART_ARRAY_RW:
7497 case SET_ARRAY_INFO:
7498 case SET_BITMAP_FILE:
7499 case SET_DISK_FAULTY:
7502 case CLUSTERED_DISK_NACK:
7509 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7510 unsigned int cmd, unsigned long arg)
7513 void __user *argp = (void __user *)arg;
7514 struct mddev *mddev = NULL;
7515 bool did_set_md_closing = false;
7517 if (!md_ioctl_valid(cmd))
7522 case GET_ARRAY_INFO:
7526 if (!capable(CAP_SYS_ADMIN))
7531 * Commands dealing with the RAID driver but not any
7536 err = get_version(argp);
7542 * Commands creating/starting a new array:
7545 mddev = bdev->bd_disk->private_data;
7552 /* Some actions do not requires the mutex */
7554 case GET_ARRAY_INFO:
7555 if (!mddev->raid_disks && !mddev->external)
7558 err = get_array_info(mddev, argp);
7562 if (!mddev->raid_disks && !mddev->external)
7565 err = get_disk_info(mddev, argp);
7568 case SET_DISK_FAULTY:
7569 err = set_disk_faulty(mddev, new_decode_dev(arg));
7572 case GET_BITMAP_FILE:
7573 err = get_bitmap_file(mddev, argp);
7578 if (cmd == ADD_NEW_DISK || cmd == HOT_ADD_DISK)
7579 flush_rdev_wq(mddev);
7581 if (cmd == HOT_REMOVE_DISK)
7582 /* need to ensure recovery thread has run */
7583 wait_event_interruptible_timeout(mddev->sb_wait,
7584 !test_bit(MD_RECOVERY_NEEDED,
7586 msecs_to_jiffies(5000));
7587 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7588 /* Need to flush page cache, and ensure no-one else opens
7591 mutex_lock(&mddev->open_mutex);
7592 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7593 mutex_unlock(&mddev->open_mutex);
7597 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7598 mutex_unlock(&mddev->open_mutex);
7602 did_set_md_closing = true;
7603 mutex_unlock(&mddev->open_mutex);
7604 sync_blockdev(bdev);
7606 err = mddev_lock(mddev);
7608 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7613 if (cmd == SET_ARRAY_INFO) {
7614 mdu_array_info_t info;
7616 memset(&info, 0, sizeof(info));
7617 else if (copy_from_user(&info, argp, sizeof(info))) {
7622 err = update_array_info(mddev, &info);
7624 pr_warn("md: couldn't update array info. %d\n", err);
7629 if (!list_empty(&mddev->disks)) {
7630 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7634 if (mddev->raid_disks) {
7635 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7639 err = md_set_array_info(mddev, &info);
7641 pr_warn("md: couldn't set array info. %d\n", err);
7648 * Commands querying/configuring an existing array:
7650 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7651 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7652 if ((!mddev->raid_disks && !mddev->external)
7653 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7654 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7655 && cmd != GET_BITMAP_FILE) {
7661 * Commands even a read-only array can execute:
7664 case RESTART_ARRAY_RW:
7665 err = restart_array(mddev);
7669 err = do_md_stop(mddev, 0, bdev);
7673 err = md_set_readonly(mddev, bdev);
7676 case HOT_REMOVE_DISK:
7677 err = hot_remove_disk(mddev, new_decode_dev(arg));
7681 /* We can support ADD_NEW_DISK on read-only arrays
7682 * only if we are re-adding a preexisting device.
7683 * So require mddev->pers and MD_DISK_SYNC.
7686 mdu_disk_info_t info;
7687 if (copy_from_user(&info, argp, sizeof(info)))
7689 else if (!(info.state & (1<<MD_DISK_SYNC)))
7690 /* Need to clear read-only for this */
7693 err = md_add_new_disk(mddev, &info);
7700 * The remaining ioctls are changing the state of the
7701 * superblock, so we do not allow them on read-only arrays.
7703 if (mddev->ro && mddev->pers) {
7704 if (mddev->ro == 2) {
7706 sysfs_notify_dirent_safe(mddev->sysfs_state);
7707 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7708 /* mddev_unlock will wake thread */
7709 /* If a device failed while we were read-only, we
7710 * need to make sure the metadata is updated now.
7712 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7713 mddev_unlock(mddev);
7714 wait_event(mddev->sb_wait,
7715 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7716 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7717 mddev_lock_nointr(mddev);
7728 mdu_disk_info_t info;
7729 if (copy_from_user(&info, argp, sizeof(info)))
7732 err = md_add_new_disk(mddev, &info);
7736 case CLUSTERED_DISK_NACK:
7737 if (mddev_is_clustered(mddev))
7738 md_cluster_ops->new_disk_ack(mddev, false);
7744 err = hot_add_disk(mddev, new_decode_dev(arg));
7748 err = do_md_run(mddev);
7751 case SET_BITMAP_FILE:
7752 err = set_bitmap_file(mddev, (int)arg);
7761 if (mddev->hold_active == UNTIL_IOCTL &&
7763 mddev->hold_active = 0;
7764 mddev_unlock(mddev);
7766 if(did_set_md_closing)
7767 clear_bit(MD_CLOSING, &mddev->flags);
7770 #ifdef CONFIG_COMPAT
7771 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7772 unsigned int cmd, unsigned long arg)
7775 case HOT_REMOVE_DISK:
7777 case SET_DISK_FAULTY:
7778 case SET_BITMAP_FILE:
7779 /* These take in integer arg, do not convert */
7782 arg = (unsigned long)compat_ptr(arg);
7786 return md_ioctl(bdev, mode, cmd, arg);
7788 #endif /* CONFIG_COMPAT */
7790 static int md_set_read_only(struct block_device *bdev, bool ro)
7792 struct mddev *mddev = bdev->bd_disk->private_data;
7795 err = mddev_lock(mddev);
7799 if (!mddev->raid_disks && !mddev->external) {
7805 * Transitioning to read-auto need only happen for arrays that call
7806 * md_write_start and which are not ready for writes yet.
7808 if (!ro && mddev->ro == 1 && mddev->pers) {
7809 err = restart_array(mddev);
7816 mddev_unlock(mddev);
7820 static int md_open(struct block_device *bdev, fmode_t mode)
7823 * Succeed if we can lock the mddev, which confirms that
7824 * it isn't being stopped right now.
7826 struct mddev *mddev = mddev_find(bdev->bd_dev);
7832 if (mddev->gendisk != bdev->bd_disk) {
7833 /* we are racing with mddev_put which is discarding this
7837 /* Wait until bdev->bd_disk is definitely gone */
7838 if (work_pending(&mddev->del_work))
7839 flush_workqueue(md_misc_wq);
7840 /* Then retry the open from the top */
7841 return -ERESTARTSYS;
7843 BUG_ON(mddev != bdev->bd_disk->private_data);
7845 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7848 if (test_bit(MD_CLOSING, &mddev->flags)) {
7849 mutex_unlock(&mddev->open_mutex);
7855 atomic_inc(&mddev->openers);
7856 mutex_unlock(&mddev->open_mutex);
7858 bdev_check_media_change(bdev);
7865 static void md_release(struct gendisk *disk, fmode_t mode)
7867 struct mddev *mddev = disk->private_data;
7870 atomic_dec(&mddev->openers);
7874 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7876 struct mddev *mddev = disk->private_data;
7877 unsigned int ret = 0;
7880 ret = DISK_EVENT_MEDIA_CHANGE;
7885 const struct block_device_operations md_fops =
7887 .owner = THIS_MODULE,
7888 .submit_bio = md_submit_bio,
7890 .release = md_release,
7892 #ifdef CONFIG_COMPAT
7893 .compat_ioctl = md_compat_ioctl,
7895 .getgeo = md_getgeo,
7896 .check_events = md_check_events,
7897 .set_read_only = md_set_read_only,
7900 static int md_thread(void *arg)
7902 struct md_thread *thread = arg;
7905 * md_thread is a 'system-thread', it's priority should be very
7906 * high. We avoid resource deadlocks individually in each
7907 * raid personality. (RAID5 does preallocation) We also use RR and
7908 * the very same RT priority as kswapd, thus we will never get
7909 * into a priority inversion deadlock.
7911 * we definitely have to have equal or higher priority than
7912 * bdflush, otherwise bdflush will deadlock if there are too
7913 * many dirty RAID5 blocks.
7916 allow_signal(SIGKILL);
7917 while (!kthread_should_stop()) {
7919 /* We need to wait INTERRUPTIBLE so that
7920 * we don't add to the load-average.
7921 * That means we need to be sure no signals are
7924 if (signal_pending(current))
7925 flush_signals(current);
7927 wait_event_interruptible_timeout
7929 test_bit(THREAD_WAKEUP, &thread->flags)
7930 || kthread_should_stop() || kthread_should_park(),
7933 clear_bit(THREAD_WAKEUP, &thread->flags);
7934 if (kthread_should_park())
7936 if (!kthread_should_stop())
7937 thread->run(thread);
7943 void md_wakeup_thread(struct md_thread *thread)
7946 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7947 set_bit(THREAD_WAKEUP, &thread->flags);
7948 wake_up(&thread->wqueue);
7951 EXPORT_SYMBOL(md_wakeup_thread);
7953 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7954 struct mddev *mddev, const char *name)
7956 struct md_thread *thread;
7958 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7962 init_waitqueue_head(&thread->wqueue);
7965 thread->mddev = mddev;
7966 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7967 thread->tsk = kthread_run(md_thread, thread,
7969 mdname(thread->mddev),
7971 if (IS_ERR(thread->tsk)) {
7977 EXPORT_SYMBOL(md_register_thread);
7979 void md_unregister_thread(struct md_thread **threadp)
7981 struct md_thread *thread = *threadp;
7984 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7985 /* Locking ensures that mddev_unlock does not wake_up a
7986 * non-existent thread
7988 spin_lock(&pers_lock);
7990 spin_unlock(&pers_lock);
7992 kthread_stop(thread->tsk);
7995 EXPORT_SYMBOL(md_unregister_thread);
7997 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7999 if (!rdev || test_bit(Faulty, &rdev->flags))
8002 if (!mddev->pers || !mddev->pers->error_handler)
8004 mddev->pers->error_handler(mddev,rdev);
8005 if (mddev->degraded)
8006 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8007 sysfs_notify_dirent_safe(rdev->sysfs_state);
8008 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8009 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8010 md_wakeup_thread(mddev->thread);
8011 if (mddev->event_work.func)
8012 queue_work(md_misc_wq, &mddev->event_work);
8013 md_new_event(mddev);
8015 EXPORT_SYMBOL(md_error);
8017 /* seq_file implementation /proc/mdstat */
8019 static void status_unused(struct seq_file *seq)
8022 struct md_rdev *rdev;
8024 seq_printf(seq, "unused devices: ");
8026 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8027 char b[BDEVNAME_SIZE];
8029 seq_printf(seq, "%s ",
8030 bdevname(rdev->bdev,b));
8033 seq_printf(seq, "<none>");
8035 seq_printf(seq, "\n");
8038 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8040 sector_t max_sectors, resync, res;
8041 unsigned long dt, db = 0;
8042 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8043 int scale, recovery_active;
8044 unsigned int per_milli;
8046 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8047 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8048 max_sectors = mddev->resync_max_sectors;
8050 max_sectors = mddev->dev_sectors;
8052 resync = mddev->curr_resync;
8054 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8055 /* Still cleaning up */
8056 resync = max_sectors;
8057 } else if (resync > max_sectors)
8058 resync = max_sectors;
8060 resync -= atomic_read(&mddev->recovery_active);
8063 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8064 struct md_rdev *rdev;
8066 rdev_for_each(rdev, mddev)
8067 if (rdev->raid_disk >= 0 &&
8068 !test_bit(Faulty, &rdev->flags) &&
8069 rdev->recovery_offset != MaxSector &&
8070 rdev->recovery_offset) {
8071 seq_printf(seq, "\trecover=REMOTE");
8074 if (mddev->reshape_position != MaxSector)
8075 seq_printf(seq, "\treshape=REMOTE");
8077 seq_printf(seq, "\tresync=REMOTE");
8080 if (mddev->recovery_cp < MaxSector) {
8081 seq_printf(seq, "\tresync=PENDING");
8087 seq_printf(seq, "\tresync=DELAYED");
8091 WARN_ON(max_sectors == 0);
8092 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8093 * in a sector_t, and (max_sectors>>scale) will fit in a
8094 * u32, as those are the requirements for sector_div.
8095 * Thus 'scale' must be at least 10
8098 if (sizeof(sector_t) > sizeof(unsigned long)) {
8099 while ( max_sectors/2 > (1ULL<<(scale+32)))
8102 res = (resync>>scale)*1000;
8103 sector_div(res, (u32)((max_sectors>>scale)+1));
8107 int i, x = per_milli/50, y = 20-x;
8108 seq_printf(seq, "[");
8109 for (i = 0; i < x; i++)
8110 seq_printf(seq, "=");
8111 seq_printf(seq, ">");
8112 for (i = 0; i < y; i++)
8113 seq_printf(seq, ".");
8114 seq_printf(seq, "] ");
8116 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8117 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8119 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8121 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8122 "resync" : "recovery"))),
8123 per_milli/10, per_milli % 10,
8124 (unsigned long long) resync/2,
8125 (unsigned long long) max_sectors/2);
8128 * dt: time from mark until now
8129 * db: blocks written from mark until now
8130 * rt: remaining time
8132 * rt is a sector_t, which is always 64bit now. We are keeping
8133 * the original algorithm, but it is not really necessary.
8135 * Original algorithm:
8136 * So we divide before multiply in case it is 32bit and close
8138 * We scale the divisor (db) by 32 to avoid losing precision
8139 * near the end of resync when the number of remaining sectors
8141 * We then divide rt by 32 after multiplying by db to compensate.
8142 * The '+1' avoids division by zero if db is very small.
8144 dt = ((jiffies - mddev->resync_mark) / HZ);
8147 curr_mark_cnt = mddev->curr_mark_cnt;
8148 recovery_active = atomic_read(&mddev->recovery_active);
8149 resync_mark_cnt = mddev->resync_mark_cnt;
8151 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8152 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8154 rt = max_sectors - resync; /* number of remaining sectors */
8155 rt = div64_u64(rt, db/32+1);
8159 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8160 ((unsigned long)rt % 60)/6);
8162 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8166 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8168 struct list_head *tmp;
8170 struct mddev *mddev;
8178 spin_lock(&all_mddevs_lock);
8179 list_for_each(tmp,&all_mddevs)
8181 mddev = list_entry(tmp, struct mddev, all_mddevs);
8183 spin_unlock(&all_mddevs_lock);
8186 spin_unlock(&all_mddevs_lock);
8188 return (void*)2;/* tail */
8192 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8194 struct list_head *tmp;
8195 struct mddev *next_mddev, *mddev = v;
8201 spin_lock(&all_mddevs_lock);
8203 tmp = all_mddevs.next;
8205 tmp = mddev->all_mddevs.next;
8206 if (tmp != &all_mddevs)
8207 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8209 next_mddev = (void*)2;
8212 spin_unlock(&all_mddevs_lock);
8220 static void md_seq_stop(struct seq_file *seq, void *v)
8222 struct mddev *mddev = v;
8224 if (mddev && v != (void*)1 && v != (void*)2)
8228 static int md_seq_show(struct seq_file *seq, void *v)
8230 struct mddev *mddev = v;
8232 struct md_rdev *rdev;
8234 if (v == (void*)1) {
8235 struct md_personality *pers;
8236 seq_printf(seq, "Personalities : ");
8237 spin_lock(&pers_lock);
8238 list_for_each_entry(pers, &pers_list, list)
8239 seq_printf(seq, "[%s] ", pers->name);
8241 spin_unlock(&pers_lock);
8242 seq_printf(seq, "\n");
8243 seq->poll_event = atomic_read(&md_event_count);
8246 if (v == (void*)2) {
8251 spin_lock(&mddev->lock);
8252 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8253 seq_printf(seq, "%s : %sactive", mdname(mddev),
8254 mddev->pers ? "" : "in");
8257 seq_printf(seq, " (read-only)");
8259 seq_printf(seq, " (auto-read-only)");
8260 seq_printf(seq, " %s", mddev->pers->name);
8265 rdev_for_each_rcu(rdev, mddev) {
8266 char b[BDEVNAME_SIZE];
8267 seq_printf(seq, " %s[%d]",
8268 bdevname(rdev->bdev,b), rdev->desc_nr);
8269 if (test_bit(WriteMostly, &rdev->flags))
8270 seq_printf(seq, "(W)");
8271 if (test_bit(Journal, &rdev->flags))
8272 seq_printf(seq, "(J)");
8273 if (test_bit(Faulty, &rdev->flags)) {
8274 seq_printf(seq, "(F)");
8277 if (rdev->raid_disk < 0)
8278 seq_printf(seq, "(S)"); /* spare */
8279 if (test_bit(Replacement, &rdev->flags))
8280 seq_printf(seq, "(R)");
8281 sectors += rdev->sectors;
8285 if (!list_empty(&mddev->disks)) {
8287 seq_printf(seq, "\n %llu blocks",
8288 (unsigned long long)
8289 mddev->array_sectors / 2);
8291 seq_printf(seq, "\n %llu blocks",
8292 (unsigned long long)sectors / 2);
8294 if (mddev->persistent) {
8295 if (mddev->major_version != 0 ||
8296 mddev->minor_version != 90) {
8297 seq_printf(seq," super %d.%d",
8298 mddev->major_version,
8299 mddev->minor_version);
8301 } else if (mddev->external)
8302 seq_printf(seq, " super external:%s",
8303 mddev->metadata_type);
8305 seq_printf(seq, " super non-persistent");
8308 mddev->pers->status(seq, mddev);
8309 seq_printf(seq, "\n ");
8310 if (mddev->pers->sync_request) {
8311 if (status_resync(seq, mddev))
8312 seq_printf(seq, "\n ");
8315 seq_printf(seq, "\n ");
8317 md_bitmap_status(seq, mddev->bitmap);
8319 seq_printf(seq, "\n");
8321 spin_unlock(&mddev->lock);
8326 static const struct seq_operations md_seq_ops = {
8327 .start = md_seq_start,
8328 .next = md_seq_next,
8329 .stop = md_seq_stop,
8330 .show = md_seq_show,
8333 static int md_seq_open(struct inode *inode, struct file *file)
8335 struct seq_file *seq;
8338 error = seq_open(file, &md_seq_ops);
8342 seq = file->private_data;
8343 seq->poll_event = atomic_read(&md_event_count);
8347 static int md_unloading;
8348 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8350 struct seq_file *seq = filp->private_data;
8354 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8355 poll_wait(filp, &md_event_waiters, wait);
8357 /* always allow read */
8358 mask = EPOLLIN | EPOLLRDNORM;
8360 if (seq->poll_event != atomic_read(&md_event_count))
8361 mask |= EPOLLERR | EPOLLPRI;
8365 static const struct proc_ops mdstat_proc_ops = {
8366 .proc_open = md_seq_open,
8367 .proc_read = seq_read,
8368 .proc_lseek = seq_lseek,
8369 .proc_release = seq_release,
8370 .proc_poll = mdstat_poll,
8373 int register_md_personality(struct md_personality *p)
8375 pr_debug("md: %s personality registered for level %d\n",
8377 spin_lock(&pers_lock);
8378 list_add_tail(&p->list, &pers_list);
8379 spin_unlock(&pers_lock);
8382 EXPORT_SYMBOL(register_md_personality);
8384 int unregister_md_personality(struct md_personality *p)
8386 pr_debug("md: %s personality unregistered\n", p->name);
8387 spin_lock(&pers_lock);
8388 list_del_init(&p->list);
8389 spin_unlock(&pers_lock);
8392 EXPORT_SYMBOL(unregister_md_personality);
8394 int register_md_cluster_operations(struct md_cluster_operations *ops,
8395 struct module *module)
8398 spin_lock(&pers_lock);
8399 if (md_cluster_ops != NULL)
8402 md_cluster_ops = ops;
8403 md_cluster_mod = module;
8405 spin_unlock(&pers_lock);
8408 EXPORT_SYMBOL(register_md_cluster_operations);
8410 int unregister_md_cluster_operations(void)
8412 spin_lock(&pers_lock);
8413 md_cluster_ops = NULL;
8414 spin_unlock(&pers_lock);
8417 EXPORT_SYMBOL(unregister_md_cluster_operations);
8419 int md_setup_cluster(struct mddev *mddev, int nodes)
8422 if (!md_cluster_ops)
8423 request_module("md-cluster");
8424 spin_lock(&pers_lock);
8425 /* ensure module won't be unloaded */
8426 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8427 pr_warn("can't find md-cluster module or get it's reference.\n");
8428 spin_unlock(&pers_lock);
8431 spin_unlock(&pers_lock);
8433 ret = md_cluster_ops->join(mddev, nodes);
8435 mddev->safemode_delay = 0;
8439 void md_cluster_stop(struct mddev *mddev)
8441 if (!md_cluster_ops)
8443 md_cluster_ops->leave(mddev);
8444 module_put(md_cluster_mod);
8447 static int is_mddev_idle(struct mddev *mddev, int init)
8449 struct md_rdev *rdev;
8455 rdev_for_each_rcu(rdev, mddev) {
8456 struct gendisk *disk = rdev->bdev->bd_disk;
8457 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8458 atomic_read(&disk->sync_io);
8459 /* sync IO will cause sync_io to increase before the disk_stats
8460 * as sync_io is counted when a request starts, and
8461 * disk_stats is counted when it completes.
8462 * So resync activity will cause curr_events to be smaller than
8463 * when there was no such activity.
8464 * non-sync IO will cause disk_stat to increase without
8465 * increasing sync_io so curr_events will (eventually)
8466 * be larger than it was before. Once it becomes
8467 * substantially larger, the test below will cause
8468 * the array to appear non-idle, and resync will slow
8470 * If there is a lot of outstanding resync activity when
8471 * we set last_event to curr_events, then all that activity
8472 * completing might cause the array to appear non-idle
8473 * and resync will be slowed down even though there might
8474 * not have been non-resync activity. This will only
8475 * happen once though. 'last_events' will soon reflect
8476 * the state where there is little or no outstanding
8477 * resync requests, and further resync activity will
8478 * always make curr_events less than last_events.
8481 if (init || curr_events - rdev->last_events > 64) {
8482 rdev->last_events = curr_events;
8490 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8492 /* another "blocks" (512byte) blocks have been synced */
8493 atomic_sub(blocks, &mddev->recovery_active);
8494 wake_up(&mddev->recovery_wait);
8496 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8497 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8498 md_wakeup_thread(mddev->thread);
8499 // stop recovery, signal do_sync ....
8502 EXPORT_SYMBOL(md_done_sync);
8504 /* md_write_start(mddev, bi)
8505 * If we need to update some array metadata (e.g. 'active' flag
8506 * in superblock) before writing, schedule a superblock update
8507 * and wait for it to complete.
8508 * A return value of 'false' means that the write wasn't recorded
8509 * and cannot proceed as the array is being suspend.
8511 bool md_write_start(struct mddev *mddev, struct bio *bi)
8515 if (bio_data_dir(bi) != WRITE)
8518 BUG_ON(mddev->ro == 1);
8519 if (mddev->ro == 2) {
8520 /* need to switch to read/write */
8522 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8523 md_wakeup_thread(mddev->thread);
8524 md_wakeup_thread(mddev->sync_thread);
8528 percpu_ref_get(&mddev->writes_pending);
8529 smp_mb(); /* Match smp_mb in set_in_sync() */
8530 if (mddev->safemode == 1)
8531 mddev->safemode = 0;
8532 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8533 if (mddev->in_sync || mddev->sync_checkers) {
8534 spin_lock(&mddev->lock);
8535 if (mddev->in_sync) {
8537 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8538 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8539 md_wakeup_thread(mddev->thread);
8542 spin_unlock(&mddev->lock);
8546 sysfs_notify_dirent_safe(mddev->sysfs_state);
8547 if (!mddev->has_superblocks)
8549 wait_event(mddev->sb_wait,
8550 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8552 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8553 percpu_ref_put(&mddev->writes_pending);
8558 EXPORT_SYMBOL(md_write_start);
8560 /* md_write_inc can only be called when md_write_start() has
8561 * already been called at least once of the current request.
8562 * It increments the counter and is useful when a single request
8563 * is split into several parts. Each part causes an increment and
8564 * so needs a matching md_write_end().
8565 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8566 * a spinlocked region.
8568 void md_write_inc(struct mddev *mddev, struct bio *bi)
8570 if (bio_data_dir(bi) != WRITE)
8572 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8573 percpu_ref_get(&mddev->writes_pending);
8575 EXPORT_SYMBOL(md_write_inc);
8577 void md_write_end(struct mddev *mddev)
8579 percpu_ref_put(&mddev->writes_pending);
8581 if (mddev->safemode == 2)
8582 md_wakeup_thread(mddev->thread);
8583 else if (mddev->safemode_delay)
8584 /* The roundup() ensures this only performs locking once
8585 * every ->safemode_delay jiffies
8587 mod_timer(&mddev->safemode_timer,
8588 roundup(jiffies, mddev->safemode_delay) +
8589 mddev->safemode_delay);
8592 EXPORT_SYMBOL(md_write_end);
8594 /* This is used by raid0 and raid10 */
8595 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8596 struct bio *bio, sector_t start, sector_t size)
8598 struct bio *discard_bio = NULL;
8600 if (__blkdev_issue_discard(rdev->bdev, start, size,
8601 GFP_NOIO, 0, &discard_bio) || !discard_bio)
8604 bio_chain(discard_bio, bio);
8605 bio_clone_blkg_association(discard_bio, bio);
8607 trace_block_bio_remap(bdev_get_queue(rdev->bdev),
8608 discard_bio, disk_devt(mddev->gendisk),
8609 bio->bi_iter.bi_sector);
8610 submit_bio_noacct(discard_bio);
8612 EXPORT_SYMBOL(md_submit_discard_bio);
8614 /* md_allow_write(mddev)
8615 * Calling this ensures that the array is marked 'active' so that writes
8616 * may proceed without blocking. It is important to call this before
8617 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8618 * Must be called with mddev_lock held.
8620 void md_allow_write(struct mddev *mddev)
8626 if (!mddev->pers->sync_request)
8629 spin_lock(&mddev->lock);
8630 if (mddev->in_sync) {
8632 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8633 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8634 if (mddev->safemode_delay &&
8635 mddev->safemode == 0)
8636 mddev->safemode = 1;
8637 spin_unlock(&mddev->lock);
8638 md_update_sb(mddev, 0);
8639 sysfs_notify_dirent_safe(mddev->sysfs_state);
8640 /* wait for the dirty state to be recorded in the metadata */
8641 wait_event(mddev->sb_wait,
8642 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8644 spin_unlock(&mddev->lock);
8646 EXPORT_SYMBOL_GPL(md_allow_write);
8648 #define SYNC_MARKS 10
8649 #define SYNC_MARK_STEP (3*HZ)
8650 #define UPDATE_FREQUENCY (5*60*HZ)
8651 void md_do_sync(struct md_thread *thread)
8653 struct mddev *mddev = thread->mddev;
8654 struct mddev *mddev2;
8655 unsigned int currspeed = 0, window;
8656 sector_t max_sectors,j, io_sectors, recovery_done;
8657 unsigned long mark[SYNC_MARKS];
8658 unsigned long update_time;
8659 sector_t mark_cnt[SYNC_MARKS];
8661 struct list_head *tmp;
8662 sector_t last_check;
8664 struct md_rdev *rdev;
8665 char *desc, *action = NULL;
8666 struct blk_plug plug;
8669 /* just incase thread restarts... */
8670 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8671 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8673 if (mddev->ro) {/* never try to sync a read-only array */
8674 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8678 if (mddev_is_clustered(mddev)) {
8679 ret = md_cluster_ops->resync_start(mddev);
8683 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8684 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8685 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8686 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8687 && ((unsigned long long)mddev->curr_resync_completed
8688 < (unsigned long long)mddev->resync_max_sectors))
8692 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8693 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8694 desc = "data-check";
8696 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8697 desc = "requested-resync";
8701 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8706 mddev->last_sync_action = action ?: desc;
8708 /* we overload curr_resync somewhat here.
8709 * 0 == not engaged in resync at all
8710 * 2 == checking that there is no conflict with another sync
8711 * 1 == like 2, but have yielded to allow conflicting resync to
8713 * other == active in resync - this many blocks
8715 * Before starting a resync we must have set curr_resync to
8716 * 2, and then checked that every "conflicting" array has curr_resync
8717 * less than ours. When we find one that is the same or higher
8718 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8719 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8720 * This will mean we have to start checking from the beginning again.
8725 int mddev2_minor = -1;
8726 mddev->curr_resync = 2;
8729 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8731 for_each_mddev(mddev2, tmp) {
8732 if (mddev2 == mddev)
8734 if (!mddev->parallel_resync
8735 && mddev2->curr_resync
8736 && match_mddev_units(mddev, mddev2)) {
8738 if (mddev < mddev2 && mddev->curr_resync == 2) {
8739 /* arbitrarily yield */
8740 mddev->curr_resync = 1;
8741 wake_up(&resync_wait);
8743 if (mddev > mddev2 && mddev->curr_resync == 1)
8744 /* no need to wait here, we can wait the next
8745 * time 'round when curr_resync == 2
8748 /* We need to wait 'interruptible' so as not to
8749 * contribute to the load average, and not to
8750 * be caught by 'softlockup'
8752 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8753 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8754 mddev2->curr_resync >= mddev->curr_resync) {
8755 if (mddev2_minor != mddev2->md_minor) {
8756 mddev2_minor = mddev2->md_minor;
8757 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8758 desc, mdname(mddev),
8762 if (signal_pending(current))
8763 flush_signals(current);
8765 finish_wait(&resync_wait, &wq);
8768 finish_wait(&resync_wait, &wq);
8771 } while (mddev->curr_resync < 2);
8774 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8775 /* resync follows the size requested by the personality,
8776 * which defaults to physical size, but can be virtual size
8778 max_sectors = mddev->resync_max_sectors;
8779 atomic64_set(&mddev->resync_mismatches, 0);
8780 /* we don't use the checkpoint if there's a bitmap */
8781 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8782 j = mddev->resync_min;
8783 else if (!mddev->bitmap)
8784 j = mddev->recovery_cp;
8786 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8787 max_sectors = mddev->resync_max_sectors;
8789 * If the original node aborts reshaping then we continue the
8790 * reshaping, so set j again to avoid restart reshape from the
8793 if (mddev_is_clustered(mddev) &&
8794 mddev->reshape_position != MaxSector)
8795 j = mddev->reshape_position;
8797 /* recovery follows the physical size of devices */
8798 max_sectors = mddev->dev_sectors;
8801 rdev_for_each_rcu(rdev, mddev)
8802 if (rdev->raid_disk >= 0 &&
8803 !test_bit(Journal, &rdev->flags) &&
8804 !test_bit(Faulty, &rdev->flags) &&
8805 !test_bit(In_sync, &rdev->flags) &&
8806 rdev->recovery_offset < j)
8807 j = rdev->recovery_offset;
8810 /* If there is a bitmap, we need to make sure all
8811 * writes that started before we added a spare
8812 * complete before we start doing a recovery.
8813 * Otherwise the write might complete and (via
8814 * bitmap_endwrite) set a bit in the bitmap after the
8815 * recovery has checked that bit and skipped that
8818 if (mddev->bitmap) {
8819 mddev->pers->quiesce(mddev, 1);
8820 mddev->pers->quiesce(mddev, 0);
8824 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8825 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8826 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8827 speed_max(mddev), desc);
8829 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8832 for (m = 0; m < SYNC_MARKS; m++) {
8834 mark_cnt[m] = io_sectors;
8837 mddev->resync_mark = mark[last_mark];
8838 mddev->resync_mark_cnt = mark_cnt[last_mark];
8841 * Tune reconstruction:
8843 window = 32 * (PAGE_SIZE / 512);
8844 pr_debug("md: using %dk window, over a total of %lluk.\n",
8845 window/2, (unsigned long long)max_sectors/2);
8847 atomic_set(&mddev->recovery_active, 0);
8851 pr_debug("md: resuming %s of %s from checkpoint.\n",
8852 desc, mdname(mddev));
8853 mddev->curr_resync = j;
8855 mddev->curr_resync = 3; /* no longer delayed */
8856 mddev->curr_resync_completed = j;
8857 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8858 md_new_event(mddev);
8859 update_time = jiffies;
8861 blk_start_plug(&plug);
8862 while (j < max_sectors) {
8867 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8868 ((mddev->curr_resync > mddev->curr_resync_completed &&
8869 (mddev->curr_resync - mddev->curr_resync_completed)
8870 > (max_sectors >> 4)) ||
8871 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8872 (j - mddev->curr_resync_completed)*2
8873 >= mddev->resync_max - mddev->curr_resync_completed ||
8874 mddev->curr_resync_completed > mddev->resync_max
8876 /* time to update curr_resync_completed */
8877 wait_event(mddev->recovery_wait,
8878 atomic_read(&mddev->recovery_active) == 0);
8879 mddev->curr_resync_completed = j;
8880 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8881 j > mddev->recovery_cp)
8882 mddev->recovery_cp = j;
8883 update_time = jiffies;
8884 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8885 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8888 while (j >= mddev->resync_max &&
8889 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8890 /* As this condition is controlled by user-space,
8891 * we can block indefinitely, so use '_interruptible'
8892 * to avoid triggering warnings.
8894 flush_signals(current); /* just in case */
8895 wait_event_interruptible(mddev->recovery_wait,
8896 mddev->resync_max > j
8897 || test_bit(MD_RECOVERY_INTR,
8901 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8904 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8906 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8910 if (!skipped) { /* actual IO requested */
8911 io_sectors += sectors;
8912 atomic_add(sectors, &mddev->recovery_active);
8915 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8919 if (j > max_sectors)
8920 /* when skipping, extra large numbers can be returned. */
8923 mddev->curr_resync = j;
8924 mddev->curr_mark_cnt = io_sectors;
8925 if (last_check == 0)
8926 /* this is the earliest that rebuild will be
8927 * visible in /proc/mdstat
8929 md_new_event(mddev);
8931 if (last_check + window > io_sectors || j == max_sectors)
8934 last_check = io_sectors;
8936 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8938 int next = (last_mark+1) % SYNC_MARKS;
8940 mddev->resync_mark = mark[next];
8941 mddev->resync_mark_cnt = mark_cnt[next];
8942 mark[next] = jiffies;
8943 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8947 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8951 * this loop exits only if either when we are slower than
8952 * the 'hard' speed limit, or the system was IO-idle for
8954 * the system might be non-idle CPU-wise, but we only care
8955 * about not overloading the IO subsystem. (things like an
8956 * e2fsck being done on the RAID array should execute fast)
8960 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8961 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8962 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8964 if (currspeed > speed_min(mddev)) {
8965 if (currspeed > speed_max(mddev)) {
8969 if (!is_mddev_idle(mddev, 0)) {
8971 * Give other IO more of a chance.
8972 * The faster the devices, the less we wait.
8974 wait_event(mddev->recovery_wait,
8975 !atomic_read(&mddev->recovery_active));
8979 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8980 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8981 ? "interrupted" : "done");
8983 * this also signals 'finished resyncing' to md_stop
8985 blk_finish_plug(&plug);
8986 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8988 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8989 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8990 mddev->curr_resync > 3) {
8991 mddev->curr_resync_completed = mddev->curr_resync;
8992 sysfs_notify_dirent_safe(mddev->sysfs_completed);
8994 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8996 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8997 mddev->curr_resync > 3) {
8998 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8999 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9000 if (mddev->curr_resync >= mddev->recovery_cp) {
9001 pr_debug("md: checkpointing %s of %s.\n",
9002 desc, mdname(mddev));
9003 if (test_bit(MD_RECOVERY_ERROR,
9005 mddev->recovery_cp =
9006 mddev->curr_resync_completed;
9008 mddev->recovery_cp =
9012 mddev->recovery_cp = MaxSector;
9014 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9015 mddev->curr_resync = MaxSector;
9016 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9017 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9019 rdev_for_each_rcu(rdev, mddev)
9020 if (rdev->raid_disk >= 0 &&
9021 mddev->delta_disks >= 0 &&
9022 !test_bit(Journal, &rdev->flags) &&
9023 !test_bit(Faulty, &rdev->flags) &&
9024 !test_bit(In_sync, &rdev->flags) &&
9025 rdev->recovery_offset < mddev->curr_resync)
9026 rdev->recovery_offset = mddev->curr_resync;
9032 /* set CHANGE_PENDING here since maybe another update is needed,
9033 * so other nodes are informed. It should be harmless for normal
9035 set_mask_bits(&mddev->sb_flags, 0,
9036 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9038 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9039 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9040 mddev->delta_disks > 0 &&
9041 mddev->pers->finish_reshape &&
9042 mddev->pers->size &&
9044 mddev_lock_nointr(mddev);
9045 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9046 mddev_unlock(mddev);
9047 if (!mddev_is_clustered(mddev)) {
9048 set_capacity(mddev->gendisk, mddev->array_sectors);
9049 revalidate_disk_size(mddev->gendisk, true);
9053 spin_lock(&mddev->lock);
9054 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9055 /* We completed so min/max setting can be forgotten if used. */
9056 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9057 mddev->resync_min = 0;
9058 mddev->resync_max = MaxSector;
9059 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9060 mddev->resync_min = mddev->curr_resync_completed;
9061 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9062 mddev->curr_resync = 0;
9063 spin_unlock(&mddev->lock);
9065 wake_up(&resync_wait);
9066 md_wakeup_thread(mddev->thread);
9069 EXPORT_SYMBOL_GPL(md_do_sync);
9071 static int remove_and_add_spares(struct mddev *mddev,
9072 struct md_rdev *this)
9074 struct md_rdev *rdev;
9077 bool remove_some = false;
9079 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9080 /* Mustn't remove devices when resync thread is running */
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(Faulty, &rdev->flags) &&
9088 atomic_read(&rdev->nr_pending)==0) {
9089 /* Faulty non-Blocked devices with nr_pending == 0
9090 * never get nr_pending incremented,
9091 * never get Faulty cleared, and never get Blocked set.
9092 * So we can synchronize_rcu now rather than once per device
9095 set_bit(RemoveSynchronized, &rdev->flags);
9101 rdev_for_each(rdev, mddev) {
9102 if ((this == NULL || rdev == this) &&
9103 rdev->raid_disk >= 0 &&
9104 !test_bit(Blocked, &rdev->flags) &&
9105 ((test_bit(RemoveSynchronized, &rdev->flags) ||
9106 (!test_bit(In_sync, &rdev->flags) &&
9107 !test_bit(Journal, &rdev->flags))) &&
9108 atomic_read(&rdev->nr_pending)==0)) {
9109 if (mddev->pers->hot_remove_disk(
9110 mddev, rdev) == 0) {
9111 sysfs_unlink_rdev(mddev, rdev);
9112 rdev->saved_raid_disk = rdev->raid_disk;
9113 rdev->raid_disk = -1;
9117 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9118 clear_bit(RemoveSynchronized, &rdev->flags);
9121 if (removed && mddev->kobj.sd)
9122 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9124 if (this && removed)
9127 rdev_for_each(rdev, mddev) {
9128 if (this && this != rdev)
9130 if (test_bit(Candidate, &rdev->flags))
9132 if (rdev->raid_disk >= 0 &&
9133 !test_bit(In_sync, &rdev->flags) &&
9134 !test_bit(Journal, &rdev->flags) &&
9135 !test_bit(Faulty, &rdev->flags))
9137 if (rdev->raid_disk >= 0)
9139 if (test_bit(Faulty, &rdev->flags))
9141 if (!test_bit(Journal, &rdev->flags)) {
9143 ! (rdev->saved_raid_disk >= 0 &&
9144 !test_bit(Bitmap_sync, &rdev->flags)))
9147 rdev->recovery_offset = 0;
9149 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9150 /* failure here is OK */
9151 sysfs_link_rdev(mddev, rdev);
9152 if (!test_bit(Journal, &rdev->flags))
9154 md_new_event(mddev);
9155 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9160 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9164 static void md_start_sync(struct work_struct *ws)
9166 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9168 mddev->sync_thread = md_register_thread(md_do_sync,
9171 if (!mddev->sync_thread) {
9172 pr_warn("%s: could not start resync thread...\n",
9174 /* leave the spares where they are, it shouldn't hurt */
9175 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9176 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9177 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9178 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9179 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9180 wake_up(&resync_wait);
9181 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9183 if (mddev->sysfs_action)
9184 sysfs_notify_dirent_safe(mddev->sysfs_action);
9186 md_wakeup_thread(mddev->sync_thread);
9187 sysfs_notify_dirent_safe(mddev->sysfs_action);
9188 md_new_event(mddev);
9192 * This routine is regularly called by all per-raid-array threads to
9193 * deal with generic issues like resync and super-block update.
9194 * Raid personalities that don't have a thread (linear/raid0) do not
9195 * need this as they never do any recovery or update the superblock.
9197 * It does not do any resync itself, but rather "forks" off other threads
9198 * to do that as needed.
9199 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9200 * "->recovery" and create a thread at ->sync_thread.
9201 * When the thread finishes it sets MD_RECOVERY_DONE
9202 * and wakeups up this thread which will reap the thread and finish up.
9203 * This thread also removes any faulty devices (with nr_pending == 0).
9205 * The overall approach is:
9206 * 1/ if the superblock needs updating, update it.
9207 * 2/ If a recovery thread is running, don't do anything else.
9208 * 3/ If recovery has finished, clean up, possibly marking spares active.
9209 * 4/ If there are any faulty devices, remove them.
9210 * 5/ If array is degraded, try to add spares devices
9211 * 6/ If array has spares or is not in-sync, start a resync thread.
9213 void md_check_recovery(struct mddev *mddev)
9215 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9216 /* Write superblock - thread that called mddev_suspend()
9217 * holds reconfig_mutex for us.
9219 set_bit(MD_UPDATING_SB, &mddev->flags);
9220 smp_mb__after_atomic();
9221 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9222 md_update_sb(mddev, 0);
9223 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9224 wake_up(&mddev->sb_wait);
9227 if (mddev->suspended)
9231 md_bitmap_daemon_work(mddev);
9233 if (signal_pending(current)) {
9234 if (mddev->pers->sync_request && !mddev->external) {
9235 pr_debug("md: %s in immediate safe mode\n",
9237 mddev->safemode = 2;
9239 flush_signals(current);
9242 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9245 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9246 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9247 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9248 (mddev->external == 0 && mddev->safemode == 1) ||
9249 (mddev->safemode == 2
9250 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9254 if (mddev_trylock(mddev)) {
9256 bool try_set_sync = mddev->safemode != 0;
9258 if (!mddev->external && mddev->safemode == 1)
9259 mddev->safemode = 0;
9262 struct md_rdev *rdev;
9263 if (!mddev->external && mddev->in_sync)
9264 /* 'Blocked' flag not needed as failed devices
9265 * will be recorded if array switched to read/write.
9266 * Leaving it set will prevent the device
9267 * from being removed.
9269 rdev_for_each(rdev, mddev)
9270 clear_bit(Blocked, &rdev->flags);
9271 /* On a read-only array we can:
9272 * - remove failed devices
9273 * - add already-in_sync devices if the array itself
9275 * As we only add devices that are already in-sync,
9276 * we can activate the spares immediately.
9278 remove_and_add_spares(mddev, NULL);
9279 /* There is no thread, but we need to call
9280 * ->spare_active and clear saved_raid_disk
9282 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9283 md_reap_sync_thread(mddev);
9284 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9285 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9286 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9290 if (mddev_is_clustered(mddev)) {
9291 struct md_rdev *rdev;
9292 /* kick the device if another node issued a
9295 rdev_for_each(rdev, mddev) {
9296 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9297 rdev->raid_disk < 0)
9298 md_kick_rdev_from_array(rdev);
9302 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9303 spin_lock(&mddev->lock);
9305 spin_unlock(&mddev->lock);
9308 if (mddev->sb_flags)
9309 md_update_sb(mddev, 0);
9311 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9312 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9313 /* resync/recovery still happening */
9314 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9317 if (mddev->sync_thread) {
9318 md_reap_sync_thread(mddev);
9321 /* Set RUNNING before clearing NEEDED to avoid
9322 * any transients in the value of "sync_action".
9324 mddev->curr_resync_completed = 0;
9325 spin_lock(&mddev->lock);
9326 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9327 spin_unlock(&mddev->lock);
9328 /* Clear some bits that don't mean anything, but
9331 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9332 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9334 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9335 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9337 /* no recovery is running.
9338 * remove any failed drives, then
9339 * add spares if possible.
9340 * Spares are also removed and re-added, to allow
9341 * the personality to fail the re-add.
9344 if (mddev->reshape_position != MaxSector) {
9345 if (mddev->pers->check_reshape == NULL ||
9346 mddev->pers->check_reshape(mddev) != 0)
9347 /* Cannot proceed */
9349 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9350 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9351 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9352 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9353 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9354 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9355 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9356 } else if (mddev->recovery_cp < MaxSector) {
9357 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9358 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9359 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9360 /* nothing to be done ... */
9363 if (mddev->pers->sync_request) {
9365 /* We are adding a device or devices to an array
9366 * which has the bitmap stored on all devices.
9367 * So make sure all bitmap pages get written
9369 md_bitmap_write_all(mddev->bitmap);
9371 INIT_WORK(&mddev->del_work, md_start_sync);
9372 queue_work(md_misc_wq, &mddev->del_work);
9376 if (!mddev->sync_thread) {
9377 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9378 wake_up(&resync_wait);
9379 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9381 if (mddev->sysfs_action)
9382 sysfs_notify_dirent_safe(mddev->sysfs_action);
9385 wake_up(&mddev->sb_wait);
9386 mddev_unlock(mddev);
9389 EXPORT_SYMBOL(md_check_recovery);
9391 void md_reap_sync_thread(struct mddev *mddev)
9393 struct md_rdev *rdev;
9394 sector_t old_dev_sectors = mddev->dev_sectors;
9395 bool is_reshaped = false;
9397 /* resync has finished, collect result */
9398 md_unregister_thread(&mddev->sync_thread);
9399 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9400 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9401 mddev->degraded != mddev->raid_disks) {
9403 /* activate any spares */
9404 if (mddev->pers->spare_active(mddev)) {
9405 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9406 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9409 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9410 mddev->pers->finish_reshape) {
9411 mddev->pers->finish_reshape(mddev);
9412 if (mddev_is_clustered(mddev))
9416 /* If array is no-longer degraded, then any saved_raid_disk
9417 * information must be scrapped.
9419 if (!mddev->degraded)
9420 rdev_for_each(rdev, mddev)
9421 rdev->saved_raid_disk = -1;
9423 md_update_sb(mddev, 1);
9424 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9425 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9427 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9428 md_cluster_ops->resync_finish(mddev);
9429 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9430 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9431 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9432 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9433 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9434 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9436 * We call md_cluster_ops->update_size here because sync_size could
9437 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9438 * so it is time to update size across cluster.
9440 if (mddev_is_clustered(mddev) && is_reshaped
9441 && !test_bit(MD_CLOSING, &mddev->flags))
9442 md_cluster_ops->update_size(mddev, old_dev_sectors);
9443 wake_up(&resync_wait);
9444 /* flag recovery needed just to double check */
9445 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9446 sysfs_notify_dirent_safe(mddev->sysfs_action);
9447 md_new_event(mddev);
9448 if (mddev->event_work.func)
9449 queue_work(md_misc_wq, &mddev->event_work);
9451 EXPORT_SYMBOL(md_reap_sync_thread);
9453 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9455 sysfs_notify_dirent_safe(rdev->sysfs_state);
9456 wait_event_timeout(rdev->blocked_wait,
9457 !test_bit(Blocked, &rdev->flags) &&
9458 !test_bit(BlockedBadBlocks, &rdev->flags),
9459 msecs_to_jiffies(5000));
9460 rdev_dec_pending(rdev, mddev);
9462 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9464 void md_finish_reshape(struct mddev *mddev)
9466 /* called be personality module when reshape completes. */
9467 struct md_rdev *rdev;
9469 rdev_for_each(rdev, mddev) {
9470 if (rdev->data_offset > rdev->new_data_offset)
9471 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9473 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9474 rdev->data_offset = rdev->new_data_offset;
9477 EXPORT_SYMBOL(md_finish_reshape);
9479 /* Bad block management */
9481 /* Returns 1 on success, 0 on failure */
9482 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9485 struct mddev *mddev = rdev->mddev;
9488 s += rdev->new_data_offset;
9490 s += rdev->data_offset;
9491 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9493 /* Make sure they get written out promptly */
9494 if (test_bit(ExternalBbl, &rdev->flags))
9495 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9496 sysfs_notify_dirent_safe(rdev->sysfs_state);
9497 set_mask_bits(&mddev->sb_flags, 0,
9498 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9499 md_wakeup_thread(rdev->mddev->thread);
9504 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9506 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9511 s += rdev->new_data_offset;
9513 s += rdev->data_offset;
9514 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9515 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9516 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9519 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9521 static int md_notify_reboot(struct notifier_block *this,
9522 unsigned long code, void *x)
9524 struct list_head *tmp;
9525 struct mddev *mddev;
9528 for_each_mddev(mddev, tmp) {
9529 if (mddev_trylock(mddev)) {
9531 __md_stop_writes(mddev);
9532 if (mddev->persistent)
9533 mddev->safemode = 2;
9534 mddev_unlock(mddev);
9539 * certain more exotic SCSI devices are known to be
9540 * volatile wrt too early system reboots. While the
9541 * right place to handle this issue is the given
9542 * driver, we do want to have a safe RAID driver ...
9550 static struct notifier_block md_notifier = {
9551 .notifier_call = md_notify_reboot,
9553 .priority = INT_MAX, /* before any real devices */
9556 static void md_geninit(void)
9558 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9560 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9563 static int __init md_init(void)
9567 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9571 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9575 md_rdev_misc_wq = alloc_workqueue("md_rdev_misc", 0, 0);
9576 if (!md_rdev_misc_wq)
9577 goto err_rdev_misc_wq;
9579 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9583 ret = __register_blkdev(0, "mdp", md_probe);
9588 register_reboot_notifier(&md_notifier);
9589 raid_table_header = register_sysctl_table(raid_root_table);
9595 unregister_blkdev(MD_MAJOR, "md");
9597 destroy_workqueue(md_rdev_misc_wq);
9599 destroy_workqueue(md_misc_wq);
9601 destroy_workqueue(md_wq);
9606 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9608 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9609 struct md_rdev *rdev2;
9611 char b[BDEVNAME_SIZE];
9614 * If size is changed in another node then we need to
9615 * do resize as well.
9617 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9618 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9620 pr_info("md-cluster: resize failed\n");
9622 md_bitmap_update_sb(mddev->bitmap);
9625 /* Check for change of roles in the active devices */
9626 rdev_for_each(rdev2, mddev) {
9627 if (test_bit(Faulty, &rdev2->flags))
9630 /* Check if the roles changed */
9631 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9633 if (test_bit(Candidate, &rdev2->flags)) {
9634 if (role == 0xfffe) {
9635 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9636 md_kick_rdev_from_array(rdev2);
9640 clear_bit(Candidate, &rdev2->flags);
9643 if (role != rdev2->raid_disk) {
9645 * got activated except reshape is happening.
9647 if (rdev2->raid_disk == -1 && role != 0xffff &&
9648 !(le32_to_cpu(sb->feature_map) &
9649 MD_FEATURE_RESHAPE_ACTIVE)) {
9650 rdev2->saved_raid_disk = role;
9651 ret = remove_and_add_spares(mddev, rdev2);
9652 pr_info("Activated spare: %s\n",
9653 bdevname(rdev2->bdev,b));
9654 /* wakeup mddev->thread here, so array could
9655 * perform resync with the new activated disk */
9656 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9657 md_wakeup_thread(mddev->thread);
9660 * We just want to do the minimum to mark the disk
9661 * as faulty. The recovery is performed by the
9662 * one who initiated the error.
9664 if ((role == 0xfffe) || (role == 0xfffd)) {
9665 md_error(mddev, rdev2);
9666 clear_bit(Blocked, &rdev2->flags);
9671 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9672 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9674 pr_warn("md: updating array disks failed. %d\n", ret);
9678 * Since mddev->delta_disks has already updated in update_raid_disks,
9679 * so it is time to check reshape.
9681 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9682 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9684 * reshape is happening in the remote node, we need to
9685 * update reshape_position and call start_reshape.
9687 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9688 if (mddev->pers->update_reshape_pos)
9689 mddev->pers->update_reshape_pos(mddev);
9690 if (mddev->pers->start_reshape)
9691 mddev->pers->start_reshape(mddev);
9692 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9693 mddev->reshape_position != MaxSector &&
9694 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9695 /* reshape is just done in another node. */
9696 mddev->reshape_position = MaxSector;
9697 if (mddev->pers->update_reshape_pos)
9698 mddev->pers->update_reshape_pos(mddev);
9701 /* Finally set the event to be up to date */
9702 mddev->events = le64_to_cpu(sb->events);
9705 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9708 struct page *swapout = rdev->sb_page;
9709 struct mdp_superblock_1 *sb;
9711 /* Store the sb page of the rdev in the swapout temporary
9712 * variable in case we err in the future
9714 rdev->sb_page = NULL;
9715 err = alloc_disk_sb(rdev);
9717 ClearPageUptodate(rdev->sb_page);
9718 rdev->sb_loaded = 0;
9719 err = super_types[mddev->major_version].
9720 load_super(rdev, NULL, mddev->minor_version);
9723 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9724 __func__, __LINE__, rdev->desc_nr, err);
9726 put_page(rdev->sb_page);
9727 rdev->sb_page = swapout;
9728 rdev->sb_loaded = 1;
9732 sb = page_address(rdev->sb_page);
9733 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9737 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9738 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9740 /* The other node finished recovery, call spare_active to set
9741 * device In_sync and mddev->degraded
9743 if (rdev->recovery_offset == MaxSector &&
9744 !test_bit(In_sync, &rdev->flags) &&
9745 mddev->pers->spare_active(mddev))
9746 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9752 void md_reload_sb(struct mddev *mddev, int nr)
9754 struct md_rdev *rdev;
9758 rdev_for_each_rcu(rdev, mddev) {
9759 if (rdev->desc_nr == nr)
9763 if (!rdev || rdev->desc_nr != nr) {
9764 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9768 err = read_rdev(mddev, rdev);
9772 check_sb_changes(mddev, rdev);
9774 /* Read all rdev's to update recovery_offset */
9775 rdev_for_each_rcu(rdev, mddev) {
9776 if (!test_bit(Faulty, &rdev->flags))
9777 read_rdev(mddev, rdev);
9780 EXPORT_SYMBOL(md_reload_sb);
9785 * Searches all registered partitions for autorun RAID arrays
9789 static DEFINE_MUTEX(detected_devices_mutex);
9790 static LIST_HEAD(all_detected_devices);
9791 struct detected_devices_node {
9792 struct list_head list;
9796 void md_autodetect_dev(dev_t dev)
9798 struct detected_devices_node *node_detected_dev;
9800 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9801 if (node_detected_dev) {
9802 node_detected_dev->dev = dev;
9803 mutex_lock(&detected_devices_mutex);
9804 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9805 mutex_unlock(&detected_devices_mutex);
9809 void md_autostart_arrays(int part)
9811 struct md_rdev *rdev;
9812 struct detected_devices_node *node_detected_dev;
9814 int i_scanned, i_passed;
9819 pr_info("md: Autodetecting RAID arrays.\n");
9821 mutex_lock(&detected_devices_mutex);
9822 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9824 node_detected_dev = list_entry(all_detected_devices.next,
9825 struct detected_devices_node, list);
9826 list_del(&node_detected_dev->list);
9827 dev = node_detected_dev->dev;
9828 kfree(node_detected_dev);
9829 mutex_unlock(&detected_devices_mutex);
9830 rdev = md_import_device(dev,0, 90);
9831 mutex_lock(&detected_devices_mutex);
9835 if (test_bit(Faulty, &rdev->flags))
9838 set_bit(AutoDetected, &rdev->flags);
9839 list_add(&rdev->same_set, &pending_raid_disks);
9842 mutex_unlock(&detected_devices_mutex);
9844 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9846 autorun_devices(part);
9849 #endif /* !MODULE */
9851 static __exit void md_exit(void)
9853 struct mddev *mddev;
9854 struct list_head *tmp;
9857 unregister_blkdev(MD_MAJOR,"md");
9858 unregister_blkdev(mdp_major, "mdp");
9859 unregister_reboot_notifier(&md_notifier);
9860 unregister_sysctl_table(raid_table_header);
9862 /* We cannot unload the modules while some process is
9863 * waiting for us in select() or poll() - wake them up
9866 while (waitqueue_active(&md_event_waiters)) {
9867 /* not safe to leave yet */
9868 wake_up(&md_event_waiters);
9872 remove_proc_entry("mdstat", NULL);
9874 for_each_mddev(mddev, tmp) {
9875 export_array(mddev);
9877 mddev->hold_active = 0;
9879 * for_each_mddev() will call mddev_put() at the end of each
9880 * iteration. As the mddev is now fully clear, this will
9881 * schedule the mddev for destruction by a workqueue, and the
9882 * destroy_workqueue() below will wait for that to complete.
9885 destroy_workqueue(md_rdev_misc_wq);
9886 destroy_workqueue(md_misc_wq);
9887 destroy_workqueue(md_wq);
9890 subsys_initcall(md_init);
9891 module_exit(md_exit)
9893 static int get_ro(char *buffer, const struct kernel_param *kp)
9895 return sprintf(buffer, "%d\n", start_readonly);
9897 static int set_ro(const char *val, const struct kernel_param *kp)
9899 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9902 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9903 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9904 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9905 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9907 MODULE_LICENSE("GPL");
9908 MODULE_DESCRIPTION("MD RAID framework");
9910 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);