2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 Errors, Warnings, etc.
36 pr_crit() for error conditions that risk data loss
37 pr_err() for error conditions that are unexpected, like an IO error
38 or internal inconsistency
39 pr_warn() for error conditions that could have been predicated, like
40 adding a device to an array when it has incompatible metadata
41 pr_info() for every interesting, very rare events, like an array starting
42 or stopping, or resync starting or stopping
43 pr_debug() for everything else.
47 #include <linux/sched/signal.h>
48 #include <linux/kthread.h>
49 #include <linux/blkdev.h>
50 #include <linux/badblocks.h>
51 #include <linux/sysctl.h>
52 #include <linux/seq_file.h>
54 #include <linux/poll.h>
55 #include <linux/ctype.h>
56 #include <linux/string.h>
57 #include <linux/hdreg.h>
58 #include <linux/proc_fs.h>
59 #include <linux/random.h>
60 #include <linux/module.h>
61 #include <linux/reboot.h>
62 #include <linux/file.h>
63 #include <linux/compat.h>
64 #include <linux/delay.h>
65 #include <linux/raid/md_p.h>
66 #include <linux/raid/md_u.h>
67 #include <linux/slab.h>
68 #include <linux/percpu-refcount.h>
70 #include <trace/events/block.h>
72 #include "md-bitmap.h"
73 #include "md-cluster.h"
76 static void autostart_arrays(int part);
79 /* pers_list is a list of registered personalities protected
81 * pers_lock does extra service to protect accesses to
82 * mddev->thread when the mutex cannot be held.
84 static LIST_HEAD(pers_list);
85 static DEFINE_SPINLOCK(pers_lock);
87 static struct kobj_type md_ktype;
89 struct md_cluster_operations *md_cluster_ops;
90 EXPORT_SYMBOL(md_cluster_ops);
91 struct module *md_cluster_mod;
92 EXPORT_SYMBOL(md_cluster_mod);
94 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
95 static struct workqueue_struct *md_wq;
96 static struct workqueue_struct *md_misc_wq;
98 static int remove_and_add_spares(struct mddev *mddev,
99 struct md_rdev *this);
100 static void mddev_detach(struct mddev *mddev);
103 * Default number of read corrections we'll attempt on an rdev
104 * before ejecting it from the array. We divide the read error
105 * count by 2 for every hour elapsed between read errors.
107 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
109 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
110 * is 1000 KB/sec, so the extra system load does not show up that much.
111 * Increase it if you want to have more _guaranteed_ speed. Note that
112 * the RAID driver will use the maximum available bandwidth if the IO
113 * subsystem is idle. There is also an 'absolute maximum' reconstruction
114 * speed limit - in case reconstruction slows down your system despite
117 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
118 * or /sys/block/mdX/md/sync_speed_{min,max}
121 static int sysctl_speed_limit_min = 1000;
122 static int sysctl_speed_limit_max = 200000;
123 static inline int speed_min(struct mddev *mddev)
125 return mddev->sync_speed_min ?
126 mddev->sync_speed_min : sysctl_speed_limit_min;
129 static inline int speed_max(struct mddev *mddev)
131 return mddev->sync_speed_max ?
132 mddev->sync_speed_max : sysctl_speed_limit_max;
135 static struct ctl_table_header *raid_table_header;
137 static struct ctl_table raid_table[] = {
139 .procname = "speed_limit_min",
140 .data = &sysctl_speed_limit_min,
141 .maxlen = sizeof(int),
142 .mode = S_IRUGO|S_IWUSR,
143 .proc_handler = proc_dointvec,
146 .procname = "speed_limit_max",
147 .data = &sysctl_speed_limit_max,
148 .maxlen = sizeof(int),
149 .mode = S_IRUGO|S_IWUSR,
150 .proc_handler = proc_dointvec,
155 static struct ctl_table raid_dir_table[] = {
159 .mode = S_IRUGO|S_IXUGO,
165 static struct ctl_table raid_root_table[] = {
170 .child = raid_dir_table,
175 static const struct block_device_operations md_fops;
177 static int start_readonly;
180 * The original mechanism for creating an md device is to create
181 * a device node in /dev and to open it. This causes races with device-close.
182 * The preferred method is to write to the "new_array" module parameter.
183 * This can avoid races.
184 * Setting create_on_open to false disables the original mechanism
185 * so all the races disappear.
187 static bool create_on_open = true;
189 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
194 if (!mddev || !bioset_initialized(&mddev->bio_set))
195 return bio_alloc(gfp_mask, nr_iovecs);
197 b = bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
202 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
204 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
206 if (!mddev || !bioset_initialized(&mddev->sync_set))
207 return bio_alloc(GFP_NOIO, 1);
209 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
213 * We have a system wide 'event count' that is incremented
214 * on any 'interesting' event, and readers of /proc/mdstat
215 * can use 'poll' or 'select' to find out when the event
219 * start array, stop array, error, add device, remove device,
220 * start build, activate spare
222 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
223 static atomic_t md_event_count;
224 void md_new_event(struct mddev *mddev)
226 atomic_inc(&md_event_count);
227 wake_up(&md_event_waiters);
229 EXPORT_SYMBOL_GPL(md_new_event);
232 * Enables to iterate over all existing md arrays
233 * all_mddevs_lock protects this list.
235 static LIST_HEAD(all_mddevs);
236 static DEFINE_SPINLOCK(all_mddevs_lock);
239 * iterates through all used mddevs in the system.
240 * We take care to grab the all_mddevs_lock whenever navigating
241 * the list, and to always hold a refcount when unlocked.
242 * Any code which breaks out of this loop while own
243 * a reference to the current mddev and must mddev_put it.
245 #define for_each_mddev(_mddev,_tmp) \
247 for (({ spin_lock(&all_mddevs_lock); \
248 _tmp = all_mddevs.next; \
250 ({ if (_tmp != &all_mddevs) \
251 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
252 spin_unlock(&all_mddevs_lock); \
253 if (_mddev) mddev_put(_mddev); \
254 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
255 _tmp != &all_mddevs;}); \
256 ({ spin_lock(&all_mddevs_lock); \
257 _tmp = _tmp->next;}) \
260 /* Rather than calling directly into the personality make_request function,
261 * IO requests come here first so that we can check if the device is
262 * being suspended pending a reconfiguration.
263 * We hold a refcount over the call to ->make_request. By the time that
264 * call has finished, the bio has been linked into some internal structure
265 * and so is visible to ->quiesce(), so we don't need the refcount any more.
267 static bool is_suspended(struct mddev *mddev, struct bio *bio)
269 if (mddev->suspended)
271 if (bio_data_dir(bio) != WRITE)
273 if (mddev->suspend_lo >= mddev->suspend_hi)
275 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
277 if (bio_end_sector(bio) < mddev->suspend_lo)
282 void md_handle_request(struct mddev *mddev, struct bio *bio)
286 if (is_suspended(mddev, bio)) {
289 prepare_to_wait(&mddev->sb_wait, &__wait,
290 TASK_UNINTERRUPTIBLE);
291 if (!is_suspended(mddev, bio))
297 finish_wait(&mddev->sb_wait, &__wait);
299 atomic_inc(&mddev->active_io);
302 if (!mddev->pers->make_request(mddev, bio)) {
303 atomic_dec(&mddev->active_io);
304 wake_up(&mddev->sb_wait);
305 goto check_suspended;
308 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
309 wake_up(&mddev->sb_wait);
311 EXPORT_SYMBOL(md_handle_request);
313 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
315 const int rw = bio_data_dir(bio);
316 const int sgrp = op_stat_group(bio_op(bio));
317 struct mddev *mddev = q->queuedata;
318 unsigned int sectors;
321 blk_queue_split(q, &bio);
323 if (mddev == NULL || mddev->pers == NULL) {
325 return BLK_QC_T_NONE;
327 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
328 if (bio_sectors(bio) != 0)
329 bio->bi_status = BLK_STS_IOERR;
331 return BLK_QC_T_NONE;
335 * save the sectors now since our bio can
336 * go away inside make_request
338 sectors = bio_sectors(bio);
339 /* bio could be mergeable after passing to underlayer */
340 bio->bi_opf &= ~REQ_NOMERGE;
342 md_handle_request(mddev, bio);
344 cpu = part_stat_lock();
345 part_stat_inc(cpu, &mddev->gendisk->part0, ios[sgrp]);
346 part_stat_add(cpu, &mddev->gendisk->part0, sectors[sgrp], sectors);
349 return BLK_QC_T_NONE;
352 /* mddev_suspend makes sure no new requests are submitted
353 * to the device, and that any requests that have been submitted
354 * are completely handled.
355 * Once mddev_detach() is called and completes, the module will be
358 void mddev_suspend(struct mddev *mddev)
360 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
361 lockdep_assert_held(&mddev->reconfig_mutex);
362 if (mddev->suspended++)
365 wake_up(&mddev->sb_wait);
366 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
367 smp_mb__after_atomic();
368 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
369 mddev->pers->quiesce(mddev, 1);
370 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
371 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
373 del_timer_sync(&mddev->safemode_timer);
375 EXPORT_SYMBOL_GPL(mddev_suspend);
377 void mddev_resume(struct mddev *mddev)
379 lockdep_assert_held(&mddev->reconfig_mutex);
380 if (--mddev->suspended)
382 wake_up(&mddev->sb_wait);
383 mddev->pers->quiesce(mddev, 0);
385 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
386 md_wakeup_thread(mddev->thread);
387 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
389 EXPORT_SYMBOL_GPL(mddev_resume);
391 int mddev_congested(struct mddev *mddev, int bits)
393 struct md_personality *pers = mddev->pers;
397 if (mddev->suspended)
399 else if (pers && pers->congested)
400 ret = pers->congested(mddev, bits);
404 EXPORT_SYMBOL_GPL(mddev_congested);
405 static int md_congested(void *data, int bits)
407 struct mddev *mddev = data;
408 return mddev_congested(mddev, bits);
412 * Generic flush handling for md
415 static void md_end_flush(struct bio *bio)
417 struct md_rdev *rdev = bio->bi_private;
418 struct mddev *mddev = rdev->mddev;
422 rdev_dec_pending(rdev, mddev);
424 if (atomic_dec_and_test(&mddev->flush_pending)) {
425 /* The pre-request flush has finished */
426 queue_work(md_wq, &mddev->flush_work);
430 static void md_submit_flush_data(struct work_struct *ws);
432 static void submit_flushes(struct work_struct *ws)
434 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
435 struct md_rdev *rdev;
437 mddev->start_flush = ktime_get_boottime();
438 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
439 atomic_set(&mddev->flush_pending, 1);
441 rdev_for_each_rcu(rdev, mddev)
442 if (rdev->raid_disk >= 0 &&
443 !test_bit(Faulty, &rdev->flags)) {
444 /* Take two references, one is dropped
445 * when request finishes, one after
446 * we reclaim rcu_read_lock
449 atomic_inc(&rdev->nr_pending);
450 atomic_inc(&rdev->nr_pending);
452 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
453 bi->bi_end_io = md_end_flush;
454 bi->bi_private = rdev;
455 bio_set_dev(bi, rdev->bdev);
456 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
457 atomic_inc(&mddev->flush_pending);
460 rdev_dec_pending(rdev, mddev);
463 if (atomic_dec_and_test(&mddev->flush_pending))
464 queue_work(md_wq, &mddev->flush_work);
467 static void md_submit_flush_data(struct work_struct *ws)
469 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
470 struct bio *bio = mddev->flush_bio;
473 * must reset flush_bio before calling into md_handle_request to avoid a
474 * deadlock, because other bios passed md_handle_request suspend check
475 * could wait for this and below md_handle_request could wait for those
476 * bios because of suspend check
478 spin_lock_irq(&mddev->lock);
479 mddev->last_flush = mddev->start_flush;
480 mddev->flush_bio = NULL;
481 spin_unlock_irq(&mddev->lock);
482 wake_up(&mddev->sb_wait);
484 if (bio->bi_iter.bi_size == 0) {
485 /* an empty barrier - all done */
488 bio->bi_opf &= ~REQ_PREFLUSH;
489 md_handle_request(mddev, bio);
494 * Manages consolidation of flushes and submitting any flushes needed for
495 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
496 * being finished in another context. Returns false if the flushing is
497 * complete but still needs the I/O portion of the bio to be processed.
499 bool md_flush_request(struct mddev *mddev, struct bio *bio)
501 ktime_t start = ktime_get_boottime();
502 spin_lock_irq(&mddev->lock);
503 wait_event_lock_irq(mddev->sb_wait,
505 ktime_after(mddev->last_flush, start),
507 if (!ktime_after(mddev->last_flush, start)) {
508 WARN_ON(mddev->flush_bio);
509 mddev->flush_bio = bio;
512 spin_unlock_irq(&mddev->lock);
515 INIT_WORK(&mddev->flush_work, submit_flushes);
516 queue_work(md_wq, &mddev->flush_work);
518 /* flush was performed for some other bio while we waited. */
519 if (bio->bi_iter.bi_size == 0)
520 /* an empty barrier - all done */
523 bio->bi_opf &= ~REQ_PREFLUSH;
529 EXPORT_SYMBOL(md_flush_request);
531 static inline struct mddev *mddev_get(struct mddev *mddev)
533 atomic_inc(&mddev->active);
537 static void mddev_delayed_delete(struct work_struct *ws);
539 static void mddev_put(struct mddev *mddev)
541 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
543 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
544 mddev->ctime == 0 && !mddev->hold_active) {
545 /* Array is not configured at all, and not held active,
547 list_del_init(&mddev->all_mddevs);
550 * Call queue_work inside the spinlock so that
551 * flush_workqueue() after mddev_find will succeed in waiting
552 * for the work to be done.
554 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
555 queue_work(md_misc_wq, &mddev->del_work);
557 spin_unlock(&all_mddevs_lock);
560 static void md_safemode_timeout(struct timer_list *t);
562 void mddev_init(struct mddev *mddev)
564 kobject_init(&mddev->kobj, &md_ktype);
565 mutex_init(&mddev->open_mutex);
566 mutex_init(&mddev->reconfig_mutex);
567 mutex_init(&mddev->bitmap_info.mutex);
568 INIT_LIST_HEAD(&mddev->disks);
569 INIT_LIST_HEAD(&mddev->all_mddevs);
570 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
571 atomic_set(&mddev->active, 1);
572 atomic_set(&mddev->openers, 0);
573 atomic_set(&mddev->active_io, 0);
574 spin_lock_init(&mddev->lock);
575 atomic_set(&mddev->flush_pending, 0);
576 init_waitqueue_head(&mddev->sb_wait);
577 init_waitqueue_head(&mddev->recovery_wait);
578 mddev->reshape_position = MaxSector;
579 mddev->reshape_backwards = 0;
580 mddev->last_sync_action = "none";
581 mddev->resync_min = 0;
582 mddev->resync_max = MaxSector;
583 mddev->level = LEVEL_NONE;
585 EXPORT_SYMBOL_GPL(mddev_init);
587 static struct mddev *mddev_find_locked(dev_t unit)
591 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
592 if (mddev->unit == unit)
598 static struct mddev *mddev_find(dev_t unit)
602 if (MAJOR(unit) != MD_MAJOR)
603 unit &= ~((1 << MdpMinorShift) - 1);
605 spin_lock(&all_mddevs_lock);
606 mddev = mddev_find_locked(unit);
609 spin_unlock(&all_mddevs_lock);
614 static struct mddev *mddev_find_or_alloc(dev_t unit)
616 struct mddev *mddev, *new = NULL;
618 if (unit && MAJOR(unit) != MD_MAJOR)
619 unit &= ~((1<<MdpMinorShift)-1);
622 spin_lock(&all_mddevs_lock);
625 mddev = mddev_find_locked(unit);
628 spin_unlock(&all_mddevs_lock);
634 list_add(&new->all_mddevs, &all_mddevs);
635 spin_unlock(&all_mddevs_lock);
636 new->hold_active = UNTIL_IOCTL;
640 /* find an unused unit number */
641 static int next_minor = 512;
642 int start = next_minor;
646 dev = MKDEV(MD_MAJOR, next_minor);
648 if (next_minor > MINORMASK)
650 if (next_minor == start) {
651 /* Oh dear, all in use. */
652 spin_unlock(&all_mddevs_lock);
657 is_free = !mddev_find_locked(dev);
660 new->md_minor = MINOR(dev);
661 new->hold_active = UNTIL_STOP;
662 list_add(&new->all_mddevs, &all_mddevs);
663 spin_unlock(&all_mddevs_lock);
666 spin_unlock(&all_mddevs_lock);
668 new = kzalloc(sizeof(*new), GFP_KERNEL);
673 if (MAJOR(unit) == MD_MAJOR)
674 new->md_minor = MINOR(unit);
676 new->md_minor = MINOR(unit) >> MdpMinorShift;
683 static struct attribute_group md_redundancy_group;
685 void mddev_unlock(struct mddev *mddev)
687 if (mddev->to_remove) {
688 /* These cannot be removed under reconfig_mutex as
689 * an access to the files will try to take reconfig_mutex
690 * while holding the file unremovable, which leads to
692 * So hold set sysfs_active while the remove in happeing,
693 * and anything else which might set ->to_remove or my
694 * otherwise change the sysfs namespace will fail with
695 * -EBUSY if sysfs_active is still set.
696 * We set sysfs_active under reconfig_mutex and elsewhere
697 * test it under the same mutex to ensure its correct value
700 struct attribute_group *to_remove = mddev->to_remove;
701 mddev->to_remove = NULL;
702 mddev->sysfs_active = 1;
703 mutex_unlock(&mddev->reconfig_mutex);
705 if (mddev->kobj.sd) {
706 if (to_remove != &md_redundancy_group)
707 sysfs_remove_group(&mddev->kobj, to_remove);
708 if (mddev->pers == NULL ||
709 mddev->pers->sync_request == NULL) {
710 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
711 if (mddev->sysfs_action)
712 sysfs_put(mddev->sysfs_action);
713 mddev->sysfs_action = NULL;
716 mddev->sysfs_active = 0;
718 mutex_unlock(&mddev->reconfig_mutex);
720 /* As we've dropped the mutex we need a spinlock to
721 * make sure the thread doesn't disappear
723 spin_lock(&pers_lock);
724 md_wakeup_thread(mddev->thread);
725 wake_up(&mddev->sb_wait);
726 spin_unlock(&pers_lock);
728 EXPORT_SYMBOL_GPL(mddev_unlock);
730 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
732 struct md_rdev *rdev;
734 rdev_for_each_rcu(rdev, mddev)
735 if (rdev->desc_nr == nr)
740 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
742 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
744 struct md_rdev *rdev;
746 rdev_for_each(rdev, mddev)
747 if (rdev->bdev->bd_dev == dev)
753 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
755 struct md_rdev *rdev;
757 rdev_for_each_rcu(rdev, mddev)
758 if (rdev->bdev->bd_dev == dev)
763 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
765 static struct md_personality *find_pers(int level, char *clevel)
767 struct md_personality *pers;
768 list_for_each_entry(pers, &pers_list, list) {
769 if (level != LEVEL_NONE && pers->level == level)
771 if (strcmp(pers->name, clevel)==0)
777 /* return the offset of the super block in 512byte sectors */
778 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
780 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
781 return MD_NEW_SIZE_SECTORS(num_sectors);
784 static int alloc_disk_sb(struct md_rdev *rdev)
786 rdev->sb_page = alloc_page(GFP_KERNEL);
792 void md_rdev_clear(struct md_rdev *rdev)
795 put_page(rdev->sb_page);
797 rdev->sb_page = NULL;
802 put_page(rdev->bb_page);
803 rdev->bb_page = NULL;
805 badblocks_exit(&rdev->badblocks);
807 EXPORT_SYMBOL_GPL(md_rdev_clear);
809 static void super_written(struct bio *bio)
811 struct md_rdev *rdev = bio->bi_private;
812 struct mddev *mddev = rdev->mddev;
814 if (bio->bi_status) {
815 pr_err("md: super_written gets error=%d\n", bio->bi_status);
816 md_error(mddev, rdev);
817 if (!test_bit(Faulty, &rdev->flags)
818 && (bio->bi_opf & MD_FAILFAST)) {
819 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
820 set_bit(LastDev, &rdev->flags);
823 clear_bit(LastDev, &rdev->flags);
827 rdev_dec_pending(rdev, mddev);
829 if (atomic_dec_and_test(&mddev->pending_writes))
830 wake_up(&mddev->sb_wait);
833 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
834 sector_t sector, int size, struct page *page)
836 /* write first size bytes of page to sector of rdev
837 * Increment mddev->pending_writes before returning
838 * and decrement it on completion, waking up sb_wait
839 * if zero is reached.
840 * If an error occurred, call md_error
848 if (test_bit(Faulty, &rdev->flags))
851 bio = md_bio_alloc_sync(mddev);
853 atomic_inc(&rdev->nr_pending);
855 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
856 bio->bi_iter.bi_sector = sector;
857 bio_add_page(bio, page, size, 0);
858 bio->bi_private = rdev;
859 bio->bi_end_io = super_written;
861 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
862 test_bit(FailFast, &rdev->flags) &&
863 !test_bit(LastDev, &rdev->flags))
865 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
867 atomic_inc(&mddev->pending_writes);
871 int md_super_wait(struct mddev *mddev)
873 /* wait for all superblock writes that were scheduled to complete */
874 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
875 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
880 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
881 struct page *page, int op, int op_flags, bool metadata_op)
883 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
886 if (metadata_op && rdev->meta_bdev)
887 bio_set_dev(bio, rdev->meta_bdev);
889 bio_set_dev(bio, rdev->bdev);
890 bio_set_op_attrs(bio, op, op_flags);
892 bio->bi_iter.bi_sector = sector + rdev->sb_start;
893 else if (rdev->mddev->reshape_position != MaxSector &&
894 (rdev->mddev->reshape_backwards ==
895 (sector >= rdev->mddev->reshape_position)))
896 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
898 bio->bi_iter.bi_sector = sector + rdev->data_offset;
899 bio_add_page(bio, page, size, 0);
901 submit_bio_wait(bio);
903 ret = !bio->bi_status;
907 EXPORT_SYMBOL_GPL(sync_page_io);
909 static int read_disk_sb(struct md_rdev *rdev, int size)
911 char b[BDEVNAME_SIZE];
916 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
922 pr_err("md: disabled device %s, could not read superblock.\n",
923 bdevname(rdev->bdev,b));
927 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
929 return sb1->set_uuid0 == sb2->set_uuid0 &&
930 sb1->set_uuid1 == sb2->set_uuid1 &&
931 sb1->set_uuid2 == sb2->set_uuid2 &&
932 sb1->set_uuid3 == sb2->set_uuid3;
935 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
938 mdp_super_t *tmp1, *tmp2;
940 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
941 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
943 if (!tmp1 || !tmp2) {
952 * nr_disks is not constant
957 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
964 static u32 md_csum_fold(u32 csum)
966 csum = (csum & 0xffff) + (csum >> 16);
967 return (csum & 0xffff) + (csum >> 16);
970 static unsigned int calc_sb_csum(mdp_super_t *sb)
973 u32 *sb32 = (u32*)sb;
975 unsigned int disk_csum, csum;
977 disk_csum = sb->sb_csum;
980 for (i = 0; i < MD_SB_BYTES/4 ; i++)
982 csum = (newcsum & 0xffffffff) + (newcsum>>32);
985 /* This used to use csum_partial, which was wrong for several
986 * reasons including that different results are returned on
987 * different architectures. It isn't critical that we get exactly
988 * the same return value as before (we always csum_fold before
989 * testing, and that removes any differences). However as we
990 * know that csum_partial always returned a 16bit value on
991 * alphas, do a fold to maximise conformity to previous behaviour.
993 sb->sb_csum = md_csum_fold(disk_csum);
995 sb->sb_csum = disk_csum;
1001 * Handle superblock details.
1002 * We want to be able to handle multiple superblock formats
1003 * so we have a common interface to them all, and an array of
1004 * different handlers.
1005 * We rely on user-space to write the initial superblock, and support
1006 * reading and updating of superblocks.
1007 * Interface methods are:
1008 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1009 * loads and validates a superblock on dev.
1010 * if refdev != NULL, compare superblocks on both devices
1012 * 0 - dev has a superblock that is compatible with refdev
1013 * 1 - dev has a superblock that is compatible and newer than refdev
1014 * so dev should be used as the refdev in future
1015 * -EINVAL superblock incompatible or invalid
1016 * -othererror e.g. -EIO
1018 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1019 * Verify that dev is acceptable into mddev.
1020 * The first time, mddev->raid_disks will be 0, and data from
1021 * dev should be merged in. Subsequent calls check that dev
1022 * is new enough. Return 0 or -EINVAL
1024 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1025 * Update the superblock for rdev with data in mddev
1026 * This does not write to disc.
1032 struct module *owner;
1033 int (*load_super)(struct md_rdev *rdev,
1034 struct md_rdev *refdev,
1036 int (*validate_super)(struct mddev *mddev,
1037 struct md_rdev *freshest,
1038 struct md_rdev *rdev);
1039 void (*sync_super)(struct mddev *mddev,
1040 struct md_rdev *rdev);
1041 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1042 sector_t num_sectors);
1043 int (*allow_new_offset)(struct md_rdev *rdev,
1044 unsigned long long new_offset);
1048 * Check that the given mddev has no bitmap.
1050 * This function is called from the run method of all personalities that do not
1051 * support bitmaps. It prints an error message and returns non-zero if mddev
1052 * has a bitmap. Otherwise, it returns 0.
1055 int md_check_no_bitmap(struct mddev *mddev)
1057 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1059 pr_warn("%s: bitmaps are not supported for %s\n",
1060 mdname(mddev), mddev->pers->name);
1063 EXPORT_SYMBOL(md_check_no_bitmap);
1066 * load_super for 0.90.0
1068 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1070 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1075 * Calculate the position of the superblock (512byte sectors),
1076 * it's at the end of the disk.
1078 * It also happens to be a multiple of 4Kb.
1080 rdev->sb_start = calc_dev_sboffset(rdev);
1082 ret = read_disk_sb(rdev, MD_SB_BYTES);
1088 bdevname(rdev->bdev, b);
1089 sb = page_address(rdev->sb_page);
1091 if (sb->md_magic != MD_SB_MAGIC) {
1092 pr_warn("md: invalid raid superblock magic on %s\n", b);
1096 if (sb->major_version != 0 ||
1097 sb->minor_version < 90 ||
1098 sb->minor_version > 91) {
1099 pr_warn("Bad version number %d.%d on %s\n",
1100 sb->major_version, sb->minor_version, b);
1104 if (sb->raid_disks <= 0)
1107 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1108 pr_warn("md: invalid superblock checksum on %s\n", b);
1112 rdev->preferred_minor = sb->md_minor;
1113 rdev->data_offset = 0;
1114 rdev->new_data_offset = 0;
1115 rdev->sb_size = MD_SB_BYTES;
1116 rdev->badblocks.shift = -1;
1118 if (sb->level == LEVEL_MULTIPATH)
1121 rdev->desc_nr = sb->this_disk.number;
1127 mdp_super_t *refsb = page_address(refdev->sb_page);
1128 if (!md_uuid_equal(refsb, sb)) {
1129 pr_warn("md: %s has different UUID to %s\n",
1130 b, bdevname(refdev->bdev,b2));
1133 if (!md_sb_equal(refsb, sb)) {
1134 pr_warn("md: %s has same UUID but different superblock to %s\n",
1135 b, bdevname(refdev->bdev, b2));
1139 ev2 = md_event(refsb);
1145 rdev->sectors = rdev->sb_start;
1146 /* Limit to 4TB as metadata cannot record more than that.
1147 * (not needed for Linear and RAID0 as metadata doesn't
1150 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1152 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1154 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1155 /* "this cannot possibly happen" ... */
1163 * validate_super for 0.90.0
1164 * note: we are not using "freshest" for 0.9 superblock
1166 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1169 mdp_super_t *sb = page_address(rdev->sb_page);
1170 __u64 ev1 = md_event(sb);
1172 rdev->raid_disk = -1;
1173 clear_bit(Faulty, &rdev->flags);
1174 clear_bit(In_sync, &rdev->flags);
1175 clear_bit(Bitmap_sync, &rdev->flags);
1176 clear_bit(WriteMostly, &rdev->flags);
1178 if (mddev->raid_disks == 0) {
1179 mddev->major_version = 0;
1180 mddev->minor_version = sb->minor_version;
1181 mddev->patch_version = sb->patch_version;
1182 mddev->external = 0;
1183 mddev->chunk_sectors = sb->chunk_size >> 9;
1184 mddev->ctime = sb->ctime;
1185 mddev->utime = sb->utime;
1186 mddev->level = sb->level;
1187 mddev->clevel[0] = 0;
1188 mddev->layout = sb->layout;
1189 mddev->raid_disks = sb->raid_disks;
1190 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1191 mddev->events = ev1;
1192 mddev->bitmap_info.offset = 0;
1193 mddev->bitmap_info.space = 0;
1194 /* bitmap can use 60 K after the 4K superblocks */
1195 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1196 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1197 mddev->reshape_backwards = 0;
1199 if (mddev->minor_version >= 91) {
1200 mddev->reshape_position = sb->reshape_position;
1201 mddev->delta_disks = sb->delta_disks;
1202 mddev->new_level = sb->new_level;
1203 mddev->new_layout = sb->new_layout;
1204 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1205 if (mddev->delta_disks < 0)
1206 mddev->reshape_backwards = 1;
1208 mddev->reshape_position = MaxSector;
1209 mddev->delta_disks = 0;
1210 mddev->new_level = mddev->level;
1211 mddev->new_layout = mddev->layout;
1212 mddev->new_chunk_sectors = mddev->chunk_sectors;
1214 if (mddev->level == 0)
1217 if (sb->state & (1<<MD_SB_CLEAN))
1218 mddev->recovery_cp = MaxSector;
1220 if (sb->events_hi == sb->cp_events_hi &&
1221 sb->events_lo == sb->cp_events_lo) {
1222 mddev->recovery_cp = sb->recovery_cp;
1224 mddev->recovery_cp = 0;
1227 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1228 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1229 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1230 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1232 mddev->max_disks = MD_SB_DISKS;
1234 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1235 mddev->bitmap_info.file == NULL) {
1236 mddev->bitmap_info.offset =
1237 mddev->bitmap_info.default_offset;
1238 mddev->bitmap_info.space =
1239 mddev->bitmap_info.default_space;
1242 } else if (mddev->pers == NULL) {
1243 /* Insist on good event counter while assembling, except
1244 * for spares (which don't need an event count) */
1246 if (sb->disks[rdev->desc_nr].state & (
1247 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1248 if (ev1 < mddev->events)
1250 } else if (mddev->bitmap) {
1251 /* if adding to array with a bitmap, then we can accept an
1252 * older device ... but not too old.
1254 if (ev1 < mddev->bitmap->events_cleared)
1256 if (ev1 < mddev->events)
1257 set_bit(Bitmap_sync, &rdev->flags);
1259 if (ev1 < mddev->events)
1260 /* just a hot-add of a new device, leave raid_disk at -1 */
1264 if (mddev->level != LEVEL_MULTIPATH) {
1265 desc = sb->disks + rdev->desc_nr;
1267 if (desc->state & (1<<MD_DISK_FAULTY))
1268 set_bit(Faulty, &rdev->flags);
1269 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1270 desc->raid_disk < mddev->raid_disks */) {
1271 set_bit(In_sync, &rdev->flags);
1272 rdev->raid_disk = desc->raid_disk;
1273 rdev->saved_raid_disk = desc->raid_disk;
1274 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1275 /* active but not in sync implies recovery up to
1276 * reshape position. We don't know exactly where
1277 * that is, so set to zero for now */
1278 if (mddev->minor_version >= 91) {
1279 rdev->recovery_offset = 0;
1280 rdev->raid_disk = desc->raid_disk;
1283 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1284 set_bit(WriteMostly, &rdev->flags);
1285 if (desc->state & (1<<MD_DISK_FAILFAST))
1286 set_bit(FailFast, &rdev->flags);
1287 } else /* MULTIPATH are always insync */
1288 set_bit(In_sync, &rdev->flags);
1293 * sync_super for 0.90.0
1295 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1298 struct md_rdev *rdev2;
1299 int next_spare = mddev->raid_disks;
1301 /* make rdev->sb match mddev data..
1304 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1305 * 3/ any empty disks < next_spare become removed
1307 * disks[0] gets initialised to REMOVED because
1308 * we cannot be sure from other fields if it has
1309 * been initialised or not.
1312 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1314 rdev->sb_size = MD_SB_BYTES;
1316 sb = page_address(rdev->sb_page);
1318 memset(sb, 0, sizeof(*sb));
1320 sb->md_magic = MD_SB_MAGIC;
1321 sb->major_version = mddev->major_version;
1322 sb->patch_version = mddev->patch_version;
1323 sb->gvalid_words = 0; /* ignored */
1324 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1325 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1326 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1327 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1329 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1330 sb->level = mddev->level;
1331 sb->size = mddev->dev_sectors / 2;
1332 sb->raid_disks = mddev->raid_disks;
1333 sb->md_minor = mddev->md_minor;
1334 sb->not_persistent = 0;
1335 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1337 sb->events_hi = (mddev->events>>32);
1338 sb->events_lo = (u32)mddev->events;
1340 if (mddev->reshape_position == MaxSector)
1341 sb->minor_version = 90;
1343 sb->minor_version = 91;
1344 sb->reshape_position = mddev->reshape_position;
1345 sb->new_level = mddev->new_level;
1346 sb->delta_disks = mddev->delta_disks;
1347 sb->new_layout = mddev->new_layout;
1348 sb->new_chunk = mddev->new_chunk_sectors << 9;
1350 mddev->minor_version = sb->minor_version;
1353 sb->recovery_cp = mddev->recovery_cp;
1354 sb->cp_events_hi = (mddev->events>>32);
1355 sb->cp_events_lo = (u32)mddev->events;
1356 if (mddev->recovery_cp == MaxSector)
1357 sb->state = (1<< MD_SB_CLEAN);
1359 sb->recovery_cp = 0;
1361 sb->layout = mddev->layout;
1362 sb->chunk_size = mddev->chunk_sectors << 9;
1364 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1365 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1367 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1368 rdev_for_each(rdev2, mddev) {
1371 int is_active = test_bit(In_sync, &rdev2->flags);
1373 if (rdev2->raid_disk >= 0 &&
1374 sb->minor_version >= 91)
1375 /* we have nowhere to store the recovery_offset,
1376 * but if it is not below the reshape_position,
1377 * we can piggy-back on that.
1380 if (rdev2->raid_disk < 0 ||
1381 test_bit(Faulty, &rdev2->flags))
1384 desc_nr = rdev2->raid_disk;
1386 desc_nr = next_spare++;
1387 rdev2->desc_nr = desc_nr;
1388 d = &sb->disks[rdev2->desc_nr];
1390 d->number = rdev2->desc_nr;
1391 d->major = MAJOR(rdev2->bdev->bd_dev);
1392 d->minor = MINOR(rdev2->bdev->bd_dev);
1394 d->raid_disk = rdev2->raid_disk;
1396 d->raid_disk = rdev2->desc_nr; /* compatibility */
1397 if (test_bit(Faulty, &rdev2->flags))
1398 d->state = (1<<MD_DISK_FAULTY);
1399 else if (is_active) {
1400 d->state = (1<<MD_DISK_ACTIVE);
1401 if (test_bit(In_sync, &rdev2->flags))
1402 d->state |= (1<<MD_DISK_SYNC);
1410 if (test_bit(WriteMostly, &rdev2->flags))
1411 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1412 if (test_bit(FailFast, &rdev2->flags))
1413 d->state |= (1<<MD_DISK_FAILFAST);
1415 /* now set the "removed" and "faulty" bits on any missing devices */
1416 for (i=0 ; i < mddev->raid_disks ; i++) {
1417 mdp_disk_t *d = &sb->disks[i];
1418 if (d->state == 0 && d->number == 0) {
1421 d->state = (1<<MD_DISK_REMOVED);
1422 d->state |= (1<<MD_DISK_FAULTY);
1426 sb->nr_disks = nr_disks;
1427 sb->active_disks = active;
1428 sb->working_disks = working;
1429 sb->failed_disks = failed;
1430 sb->spare_disks = spare;
1432 sb->this_disk = sb->disks[rdev->desc_nr];
1433 sb->sb_csum = calc_sb_csum(sb);
1437 * rdev_size_change for 0.90.0
1439 static unsigned long long
1440 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1442 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1443 return 0; /* component must fit device */
1444 if (rdev->mddev->bitmap_info.offset)
1445 return 0; /* can't move bitmap */
1446 rdev->sb_start = calc_dev_sboffset(rdev);
1447 if (!num_sectors || num_sectors > rdev->sb_start)
1448 num_sectors = rdev->sb_start;
1449 /* Limit to 4TB as metadata cannot record more than that.
1450 * 4TB == 2^32 KB, or 2*2^32 sectors.
1452 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1453 rdev->mddev->level >= 1)
1454 num_sectors = (sector_t)(2ULL << 32) - 2;
1456 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1458 } while (md_super_wait(rdev->mddev) < 0);
1463 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1465 /* non-zero offset changes not possible with v0.90 */
1466 return new_offset == 0;
1470 * version 1 superblock
1473 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1477 unsigned long long newcsum;
1478 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1479 __le32 *isuper = (__le32*)sb;
1481 disk_csum = sb->sb_csum;
1484 for (; size >= 4; size -= 4)
1485 newcsum += le32_to_cpu(*isuper++);
1488 newcsum += le16_to_cpu(*(__le16*) isuper);
1490 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1491 sb->sb_csum = disk_csum;
1492 return cpu_to_le32(csum);
1495 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1497 struct mdp_superblock_1 *sb;
1501 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1505 * Calculate the position of the superblock in 512byte sectors.
1506 * It is always aligned to a 4K boundary and
1507 * depeding on minor_version, it can be:
1508 * 0: At least 8K, but less than 12K, from end of device
1509 * 1: At start of device
1510 * 2: 4K from start of device.
1512 switch(minor_version) {
1514 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1516 sb_start &= ~(sector_t)(4*2-1);
1527 rdev->sb_start = sb_start;
1529 /* superblock is rarely larger than 1K, but it can be larger,
1530 * and it is safe to read 4k, so we do that
1532 ret = read_disk_sb(rdev, 4096);
1533 if (ret) return ret;
1535 sb = page_address(rdev->sb_page);
1537 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1538 sb->major_version != cpu_to_le32(1) ||
1539 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1540 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1541 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1544 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1545 pr_warn("md: invalid superblock checksum on %s\n",
1546 bdevname(rdev->bdev,b));
1549 if (le64_to_cpu(sb->data_size) < 10) {
1550 pr_warn("md: data_size too small on %s\n",
1551 bdevname(rdev->bdev,b));
1556 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1557 /* Some padding is non-zero, might be a new feature */
1560 rdev->preferred_minor = 0xffff;
1561 rdev->data_offset = le64_to_cpu(sb->data_offset);
1562 rdev->new_data_offset = rdev->data_offset;
1563 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1564 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1565 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1566 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1568 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1569 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1570 if (rdev->sb_size & bmask)
1571 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1574 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1577 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1580 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1583 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1585 if (!rdev->bb_page) {
1586 rdev->bb_page = alloc_page(GFP_KERNEL);
1590 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1591 rdev->badblocks.count == 0) {
1592 /* need to load the bad block list.
1593 * Currently we limit it to one page.
1599 int sectors = le16_to_cpu(sb->bblog_size);
1600 if (sectors > (PAGE_SIZE / 512))
1602 offset = le32_to_cpu(sb->bblog_offset);
1605 bb_sector = (long long)offset;
1606 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1607 rdev->bb_page, REQ_OP_READ, 0, true))
1609 bbp = (u64 *)page_address(rdev->bb_page);
1610 rdev->badblocks.shift = sb->bblog_shift;
1611 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1612 u64 bb = le64_to_cpu(*bbp);
1613 int count = bb & (0x3ff);
1614 u64 sector = bb >> 10;
1615 sector <<= sb->bblog_shift;
1616 count <<= sb->bblog_shift;
1619 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1622 } else if (sb->bblog_offset != 0)
1623 rdev->badblocks.shift = 0;
1625 if ((le32_to_cpu(sb->feature_map) &
1626 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1627 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1628 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1629 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1632 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1640 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1642 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1643 sb->level != refsb->level ||
1644 sb->layout != refsb->layout ||
1645 sb->chunksize != refsb->chunksize) {
1646 pr_warn("md: %s has strangely different superblock to %s\n",
1647 bdevname(rdev->bdev,b),
1648 bdevname(refdev->bdev,b2));
1651 ev1 = le64_to_cpu(sb->events);
1652 ev2 = le64_to_cpu(refsb->events);
1659 if (minor_version) {
1660 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1661 sectors -= rdev->data_offset;
1663 sectors = rdev->sb_start;
1664 if (sectors < le64_to_cpu(sb->data_size))
1666 rdev->sectors = le64_to_cpu(sb->data_size);
1670 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1672 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1673 __u64 ev1 = le64_to_cpu(sb->events);
1675 rdev->raid_disk = -1;
1676 clear_bit(Faulty, &rdev->flags);
1677 clear_bit(In_sync, &rdev->flags);
1678 clear_bit(Bitmap_sync, &rdev->flags);
1679 clear_bit(WriteMostly, &rdev->flags);
1681 if (mddev->raid_disks == 0) {
1682 mddev->major_version = 1;
1683 mddev->patch_version = 0;
1684 mddev->external = 0;
1685 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1686 mddev->ctime = le64_to_cpu(sb->ctime);
1687 mddev->utime = le64_to_cpu(sb->utime);
1688 mddev->level = le32_to_cpu(sb->level);
1689 mddev->clevel[0] = 0;
1690 mddev->layout = le32_to_cpu(sb->layout);
1691 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1692 mddev->dev_sectors = le64_to_cpu(sb->size);
1693 mddev->events = ev1;
1694 mddev->bitmap_info.offset = 0;
1695 mddev->bitmap_info.space = 0;
1696 /* Default location for bitmap is 1K after superblock
1697 * using 3K - total of 4K
1699 mddev->bitmap_info.default_offset = 1024 >> 9;
1700 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1701 mddev->reshape_backwards = 0;
1703 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1704 memcpy(mddev->uuid, sb->set_uuid, 16);
1706 mddev->max_disks = (4096-256)/2;
1708 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1709 mddev->bitmap_info.file == NULL) {
1710 mddev->bitmap_info.offset =
1711 (__s32)le32_to_cpu(sb->bitmap_offset);
1712 /* Metadata doesn't record how much space is available.
1713 * For 1.0, we assume we can use up to the superblock
1714 * if before, else to 4K beyond superblock.
1715 * For others, assume no change is possible.
1717 if (mddev->minor_version > 0)
1718 mddev->bitmap_info.space = 0;
1719 else if (mddev->bitmap_info.offset > 0)
1720 mddev->bitmap_info.space =
1721 8 - mddev->bitmap_info.offset;
1723 mddev->bitmap_info.space =
1724 -mddev->bitmap_info.offset;
1727 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1728 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1729 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1730 mddev->new_level = le32_to_cpu(sb->new_level);
1731 mddev->new_layout = le32_to_cpu(sb->new_layout);
1732 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1733 if (mddev->delta_disks < 0 ||
1734 (mddev->delta_disks == 0 &&
1735 (le32_to_cpu(sb->feature_map)
1736 & MD_FEATURE_RESHAPE_BACKWARDS)))
1737 mddev->reshape_backwards = 1;
1739 mddev->reshape_position = MaxSector;
1740 mddev->delta_disks = 0;
1741 mddev->new_level = mddev->level;
1742 mddev->new_layout = mddev->layout;
1743 mddev->new_chunk_sectors = mddev->chunk_sectors;
1746 if (mddev->level == 0 &&
1747 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1750 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1751 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1753 if (le32_to_cpu(sb->feature_map) &
1754 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1755 if (le32_to_cpu(sb->feature_map) &
1756 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1758 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1759 (le32_to_cpu(sb->feature_map) &
1760 MD_FEATURE_MULTIPLE_PPLS))
1762 set_bit(MD_HAS_PPL, &mddev->flags);
1764 } else if (mddev->pers == NULL) {
1765 /* Insist of good event counter while assembling, except for
1766 * spares (which don't need an event count).
1767 * Similar to mdadm, we allow event counter difference of 1
1768 * from the freshest device.
1770 if (rdev->desc_nr >= 0 &&
1771 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1772 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1773 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1774 if (ev1 + 1 < mddev->events)
1776 } else if (mddev->bitmap) {
1777 /* If adding to array with a bitmap, then we can accept an
1778 * older device, but not too old.
1780 if (ev1 < mddev->bitmap->events_cleared)
1782 if (ev1 < mddev->events)
1783 set_bit(Bitmap_sync, &rdev->flags);
1785 if (ev1 < mddev->events)
1786 /* just a hot-add of a new device, leave raid_disk at -1 */
1789 if (mddev->level != LEVEL_MULTIPATH) {
1791 if (rdev->desc_nr < 0 ||
1792 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1793 role = MD_DISK_ROLE_SPARE;
1795 } else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1797 * If we are assembling, and our event counter is smaller than the
1798 * highest event counter, we cannot trust our superblock about the role.
1799 * It could happen that our rdev was marked as Faulty, and all other
1800 * superblocks were updated with +1 event counter.
1801 * Then, before the next superblock update, which typically happens when
1802 * remove_and_add_spares() removes the device from the array, there was
1803 * a crash or reboot.
1804 * If we allow current rdev without consulting the freshest superblock,
1805 * we could cause data corruption.
1806 * Note that in this case our event counter is smaller by 1 than the
1807 * highest, otherwise, this rdev would not be allowed into array;
1808 * both kernel and mdadm allow event counter difference of 1.
1810 struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1811 u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1813 if (rdev->desc_nr >= freshest_max_dev) {
1814 /* this is unexpected, better not proceed */
1815 pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1816 mdname(mddev), rdev->bdev, rdev->desc_nr,
1817 freshest->bdev, freshest_max_dev);
1821 role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1822 pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1823 mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1825 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1828 case MD_DISK_ROLE_SPARE: /* spare */
1830 case MD_DISK_ROLE_FAULTY: /* faulty */
1831 set_bit(Faulty, &rdev->flags);
1833 case MD_DISK_ROLE_JOURNAL: /* journal device */
1834 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1835 /* journal device without journal feature */
1836 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1839 set_bit(Journal, &rdev->flags);
1840 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1841 rdev->raid_disk = 0;
1844 rdev->saved_raid_disk = role;
1845 if ((le32_to_cpu(sb->feature_map) &
1846 MD_FEATURE_RECOVERY_OFFSET)) {
1847 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1848 if (!(le32_to_cpu(sb->feature_map) &
1849 MD_FEATURE_RECOVERY_BITMAP))
1850 rdev->saved_raid_disk = -1;
1853 * If the array is FROZEN, then the device can't
1854 * be in_sync with rest of array.
1856 if (!test_bit(MD_RECOVERY_FROZEN,
1858 set_bit(In_sync, &rdev->flags);
1860 rdev->raid_disk = role;
1863 if (sb->devflags & WriteMostly1)
1864 set_bit(WriteMostly, &rdev->flags);
1865 if (sb->devflags & FailFast1)
1866 set_bit(FailFast, &rdev->flags);
1867 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1868 set_bit(Replacement, &rdev->flags);
1869 } else /* MULTIPATH are always insync */
1870 set_bit(In_sync, &rdev->flags);
1875 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1877 struct mdp_superblock_1 *sb;
1878 struct md_rdev *rdev2;
1880 /* make rdev->sb match mddev and rdev data. */
1882 sb = page_address(rdev->sb_page);
1884 sb->feature_map = 0;
1886 sb->recovery_offset = cpu_to_le64(0);
1887 memset(sb->pad3, 0, sizeof(sb->pad3));
1889 sb->utime = cpu_to_le64((__u64)mddev->utime);
1890 sb->events = cpu_to_le64(mddev->events);
1892 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1893 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1894 sb->resync_offset = cpu_to_le64(MaxSector);
1896 sb->resync_offset = cpu_to_le64(0);
1898 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1900 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1901 sb->size = cpu_to_le64(mddev->dev_sectors);
1902 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1903 sb->level = cpu_to_le32(mddev->level);
1904 sb->layout = cpu_to_le32(mddev->layout);
1905 if (test_bit(FailFast, &rdev->flags))
1906 sb->devflags |= FailFast1;
1908 sb->devflags &= ~FailFast1;
1910 if (test_bit(WriteMostly, &rdev->flags))
1911 sb->devflags |= WriteMostly1;
1913 sb->devflags &= ~WriteMostly1;
1914 sb->data_offset = cpu_to_le64(rdev->data_offset);
1915 sb->data_size = cpu_to_le64(rdev->sectors);
1917 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1918 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1919 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1922 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1923 !test_bit(In_sync, &rdev->flags)) {
1925 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1926 sb->recovery_offset =
1927 cpu_to_le64(rdev->recovery_offset);
1928 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1930 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1932 /* Note: recovery_offset and journal_tail share space */
1933 if (test_bit(Journal, &rdev->flags))
1934 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1935 if (test_bit(Replacement, &rdev->flags))
1937 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1939 if (mddev->reshape_position != MaxSector) {
1940 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1941 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1942 sb->new_layout = cpu_to_le32(mddev->new_layout);
1943 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1944 sb->new_level = cpu_to_le32(mddev->new_level);
1945 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1946 if (mddev->delta_disks == 0 &&
1947 mddev->reshape_backwards)
1949 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1950 if (rdev->new_data_offset != rdev->data_offset) {
1952 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1953 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1954 - rdev->data_offset));
1958 if (mddev_is_clustered(mddev))
1959 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1961 if (rdev->badblocks.count == 0)
1962 /* Nothing to do for bad blocks*/ ;
1963 else if (sb->bblog_offset == 0)
1964 /* Cannot record bad blocks on this device */
1965 md_error(mddev, rdev);
1967 struct badblocks *bb = &rdev->badblocks;
1968 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1970 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1975 seq = read_seqbegin(&bb->lock);
1977 memset(bbp, 0xff, PAGE_SIZE);
1979 for (i = 0 ; i < bb->count ; i++) {
1980 u64 internal_bb = p[i];
1981 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1982 | BB_LEN(internal_bb));
1983 bbp[i] = cpu_to_le64(store_bb);
1986 if (read_seqretry(&bb->lock, seq))
1989 bb->sector = (rdev->sb_start +
1990 (int)le32_to_cpu(sb->bblog_offset));
1991 bb->size = le16_to_cpu(sb->bblog_size);
1996 rdev_for_each(rdev2, mddev)
1997 if (rdev2->desc_nr+1 > max_dev)
1998 max_dev = rdev2->desc_nr+1;
2000 if (max_dev > le32_to_cpu(sb->max_dev)) {
2002 sb->max_dev = cpu_to_le32(max_dev);
2003 rdev->sb_size = max_dev * 2 + 256;
2004 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2005 if (rdev->sb_size & bmask)
2006 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2008 max_dev = le32_to_cpu(sb->max_dev);
2010 for (i=0; i<max_dev;i++)
2011 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2013 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2014 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2016 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2017 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2019 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2021 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2022 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2023 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2026 rdev_for_each(rdev2, mddev) {
2028 if (test_bit(Faulty, &rdev2->flags))
2029 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2030 else if (test_bit(In_sync, &rdev2->flags))
2031 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2032 else if (test_bit(Journal, &rdev2->flags))
2033 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2034 else if (rdev2->raid_disk >= 0)
2035 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2037 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2040 sb->sb_csum = calc_sb_1_csum(sb);
2043 static unsigned long long
2044 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2046 struct mdp_superblock_1 *sb;
2047 sector_t max_sectors;
2048 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2049 return 0; /* component must fit device */
2050 if (rdev->data_offset != rdev->new_data_offset)
2051 return 0; /* too confusing */
2052 if (rdev->sb_start < rdev->data_offset) {
2053 /* minor versions 1 and 2; superblock before data */
2054 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2055 max_sectors -= rdev->data_offset;
2056 if (!num_sectors || num_sectors > max_sectors)
2057 num_sectors = max_sectors;
2058 } else if (rdev->mddev->bitmap_info.offset) {
2059 /* minor version 0 with bitmap we can't move */
2062 /* minor version 0; superblock after data */
2064 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
2065 sb_start &= ~(sector_t)(4*2 - 1);
2066 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
2067 if (!num_sectors || num_sectors > max_sectors)
2068 num_sectors = max_sectors;
2069 rdev->sb_start = sb_start;
2071 sb = page_address(rdev->sb_page);
2072 sb->data_size = cpu_to_le64(num_sectors);
2073 sb->super_offset = cpu_to_le64(rdev->sb_start);
2074 sb->sb_csum = calc_sb_1_csum(sb);
2076 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2078 } while (md_super_wait(rdev->mddev) < 0);
2084 super_1_allow_new_offset(struct md_rdev *rdev,
2085 unsigned long long new_offset)
2087 /* All necessary checks on new >= old have been done */
2088 struct bitmap *bitmap;
2089 if (new_offset >= rdev->data_offset)
2092 /* with 1.0 metadata, there is no metadata to tread on
2093 * so we can always move back */
2094 if (rdev->mddev->minor_version == 0)
2097 /* otherwise we must be sure not to step on
2098 * any metadata, so stay:
2099 * 36K beyond start of superblock
2100 * beyond end of badblocks
2101 * beyond write-intent bitmap
2103 if (rdev->sb_start + (32+4)*2 > new_offset)
2105 bitmap = rdev->mddev->bitmap;
2106 if (bitmap && !rdev->mddev->bitmap_info.file &&
2107 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2108 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2110 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2116 static struct super_type super_types[] = {
2119 .owner = THIS_MODULE,
2120 .load_super = super_90_load,
2121 .validate_super = super_90_validate,
2122 .sync_super = super_90_sync,
2123 .rdev_size_change = super_90_rdev_size_change,
2124 .allow_new_offset = super_90_allow_new_offset,
2128 .owner = THIS_MODULE,
2129 .load_super = super_1_load,
2130 .validate_super = super_1_validate,
2131 .sync_super = super_1_sync,
2132 .rdev_size_change = super_1_rdev_size_change,
2133 .allow_new_offset = super_1_allow_new_offset,
2137 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2139 if (mddev->sync_super) {
2140 mddev->sync_super(mddev, rdev);
2144 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2146 super_types[mddev->major_version].sync_super(mddev, rdev);
2149 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2151 struct md_rdev *rdev, *rdev2;
2154 rdev_for_each_rcu(rdev, mddev1) {
2155 if (test_bit(Faulty, &rdev->flags) ||
2156 test_bit(Journal, &rdev->flags) ||
2157 rdev->raid_disk == -1)
2159 rdev_for_each_rcu(rdev2, mddev2) {
2160 if (test_bit(Faulty, &rdev2->flags) ||
2161 test_bit(Journal, &rdev2->flags) ||
2162 rdev2->raid_disk == -1)
2164 if (rdev->bdev->bd_contains ==
2165 rdev2->bdev->bd_contains) {
2175 static LIST_HEAD(pending_raid_disks);
2178 * Try to register data integrity profile for an mddev
2180 * This is called when an array is started and after a disk has been kicked
2181 * from the array. It only succeeds if all working and active component devices
2182 * are integrity capable with matching profiles.
2184 int md_integrity_register(struct mddev *mddev)
2186 struct md_rdev *rdev, *reference = NULL;
2188 if (list_empty(&mddev->disks))
2189 return 0; /* nothing to do */
2190 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2191 return 0; /* shouldn't register, or already is */
2192 rdev_for_each(rdev, mddev) {
2193 /* skip spares and non-functional disks */
2194 if (test_bit(Faulty, &rdev->flags))
2196 if (rdev->raid_disk < 0)
2199 /* Use the first rdev as the reference */
2203 /* does this rdev's profile match the reference profile? */
2204 if (blk_integrity_compare(reference->bdev->bd_disk,
2205 rdev->bdev->bd_disk) < 0)
2208 if (!reference || !bdev_get_integrity(reference->bdev))
2211 * All component devices are integrity capable and have matching
2212 * profiles, register the common profile for the md device.
2214 blk_integrity_register(mddev->gendisk,
2215 bdev_get_integrity(reference->bdev));
2217 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2218 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2219 pr_err("md: failed to create integrity pool for %s\n",
2225 EXPORT_SYMBOL(md_integrity_register);
2228 * Attempt to add an rdev, but only if it is consistent with the current
2231 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2233 struct blk_integrity *bi_rdev;
2234 struct blk_integrity *bi_mddev;
2235 char name[BDEVNAME_SIZE];
2237 if (!mddev->gendisk)
2240 bi_rdev = bdev_get_integrity(rdev->bdev);
2241 bi_mddev = blk_get_integrity(mddev->gendisk);
2243 if (!bi_mddev) /* nothing to do */
2246 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2247 pr_err("%s: incompatible integrity profile for %s\n",
2248 mdname(mddev), bdevname(rdev->bdev, name));
2254 EXPORT_SYMBOL(md_integrity_add_rdev);
2256 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2258 char b[BDEVNAME_SIZE];
2262 /* prevent duplicates */
2263 if (find_rdev(mddev, rdev->bdev->bd_dev))
2266 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2270 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2271 if (!test_bit(Journal, &rdev->flags) &&
2273 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2275 /* Cannot change size, so fail
2276 * If mddev->level <= 0, then we don't care
2277 * about aligning sizes (e.g. linear)
2279 if (mddev->level > 0)
2282 mddev->dev_sectors = rdev->sectors;
2285 /* Verify rdev->desc_nr is unique.
2286 * If it is -1, assign a free number, else
2287 * check number is not in use
2290 if (rdev->desc_nr < 0) {
2293 choice = mddev->raid_disks;
2294 while (md_find_rdev_nr_rcu(mddev, choice))
2296 rdev->desc_nr = choice;
2298 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2304 if (!test_bit(Journal, &rdev->flags) &&
2305 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2306 pr_warn("md: %s: array is limited to %d devices\n",
2307 mdname(mddev), mddev->max_disks);
2310 bdevname(rdev->bdev,b);
2311 strreplace(b, '/', '!');
2313 rdev->mddev = mddev;
2314 pr_debug("md: bind<%s>\n", b);
2316 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2319 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2320 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2321 /* failure here is OK */;
2322 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2324 list_add_rcu(&rdev->same_set, &mddev->disks);
2325 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2327 /* May as well allow recovery to be retried once */
2328 mddev->recovery_disabled++;
2333 pr_warn("md: failed to register dev-%s for %s\n",
2338 static void md_delayed_delete(struct work_struct *ws)
2340 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2341 kobject_del(&rdev->kobj);
2342 kobject_put(&rdev->kobj);
2345 static void unbind_rdev_from_array(struct md_rdev *rdev)
2347 char b[BDEVNAME_SIZE];
2349 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2350 list_del_rcu(&rdev->same_set);
2351 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2353 sysfs_remove_link(&rdev->kobj, "block");
2354 sysfs_put(rdev->sysfs_state);
2355 rdev->sysfs_state = NULL;
2356 rdev->badblocks.count = 0;
2357 /* We need to delay this, otherwise we can deadlock when
2358 * writing to 'remove' to "dev/state". We also need
2359 * to delay it due to rcu usage.
2362 INIT_WORK(&rdev->del_work, md_delayed_delete);
2363 kobject_get(&rdev->kobj);
2364 queue_work(md_misc_wq, &rdev->del_work);
2368 * prevent the device from being mounted, repartitioned or
2369 * otherwise reused by a RAID array (or any other kernel
2370 * subsystem), by bd_claiming the device.
2372 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2375 struct block_device *bdev;
2376 char b[BDEVNAME_SIZE];
2378 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2379 shared ? (struct md_rdev *)lock_rdev : rdev);
2381 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2382 return PTR_ERR(bdev);
2388 static void unlock_rdev(struct md_rdev *rdev)
2390 struct block_device *bdev = rdev->bdev;
2392 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2395 void md_autodetect_dev(dev_t dev);
2397 static void export_rdev(struct md_rdev *rdev)
2399 char b[BDEVNAME_SIZE];
2401 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2402 md_rdev_clear(rdev);
2404 if (test_bit(AutoDetected, &rdev->flags))
2405 md_autodetect_dev(rdev->bdev->bd_dev);
2408 kobject_put(&rdev->kobj);
2411 void md_kick_rdev_from_array(struct md_rdev *rdev)
2413 unbind_rdev_from_array(rdev);
2416 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2418 static void export_array(struct mddev *mddev)
2420 struct md_rdev *rdev;
2422 while (!list_empty(&mddev->disks)) {
2423 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2425 md_kick_rdev_from_array(rdev);
2427 mddev->raid_disks = 0;
2428 mddev->major_version = 0;
2431 static bool set_in_sync(struct mddev *mddev)
2433 lockdep_assert_held(&mddev->lock);
2434 if (!mddev->in_sync) {
2435 mddev->sync_checkers++;
2436 spin_unlock(&mddev->lock);
2437 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2438 spin_lock(&mddev->lock);
2439 if (!mddev->in_sync &&
2440 percpu_ref_is_zero(&mddev->writes_pending)) {
2443 * Ensure ->in_sync is visible before we clear
2447 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2448 sysfs_notify_dirent_safe(mddev->sysfs_state);
2450 if (--mddev->sync_checkers == 0)
2451 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2453 if (mddev->safemode == 1)
2454 mddev->safemode = 0;
2455 return mddev->in_sync;
2458 static void sync_sbs(struct mddev *mddev, int nospares)
2460 /* Update each superblock (in-memory image), but
2461 * if we are allowed to, skip spares which already
2462 * have the right event counter, or have one earlier
2463 * (which would mean they aren't being marked as dirty
2464 * with the rest of the array)
2466 struct md_rdev *rdev;
2467 rdev_for_each(rdev, mddev) {
2468 if (rdev->sb_events == mddev->events ||
2470 rdev->raid_disk < 0 &&
2471 rdev->sb_events+1 == mddev->events)) {
2472 /* Don't update this superblock */
2473 rdev->sb_loaded = 2;
2475 sync_super(mddev, rdev);
2476 rdev->sb_loaded = 1;
2481 static bool does_sb_need_changing(struct mddev *mddev)
2483 struct md_rdev *rdev = NULL, *iter;
2484 struct mdp_superblock_1 *sb;
2487 /* Find a good rdev */
2488 rdev_for_each(iter, mddev)
2489 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2494 /* No good device found. */
2498 sb = page_address(rdev->sb_page);
2499 /* Check if a device has become faulty or a spare become active */
2500 rdev_for_each(rdev, mddev) {
2501 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2502 /* Device activated? */
2503 if (role == 0xffff && rdev->raid_disk >=0 &&
2504 !test_bit(Faulty, &rdev->flags))
2506 /* Device turned faulty? */
2507 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2511 /* Check if any mddev parameters have changed */
2512 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2513 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2514 (mddev->layout != le32_to_cpu(sb->layout)) ||
2515 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2516 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2522 void md_update_sb(struct mddev *mddev, int force_change)
2524 struct md_rdev *rdev;
2527 int any_badblocks_changed = 0;
2532 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2537 if (mddev_is_clustered(mddev)) {
2538 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2540 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2542 ret = md_cluster_ops->metadata_update_start(mddev);
2543 /* Has someone else has updated the sb */
2544 if (!does_sb_need_changing(mddev)) {
2546 md_cluster_ops->metadata_update_cancel(mddev);
2547 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2548 BIT(MD_SB_CHANGE_DEVS) |
2549 BIT(MD_SB_CHANGE_CLEAN));
2555 * First make sure individual recovery_offsets are correct
2556 * curr_resync_completed can only be used during recovery.
2557 * During reshape/resync it might use array-addresses rather
2558 * that device addresses.
2560 rdev_for_each(rdev, mddev) {
2561 if (rdev->raid_disk >= 0 &&
2562 mddev->delta_disks >= 0 &&
2563 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2564 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2565 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2566 !test_bit(Journal, &rdev->flags) &&
2567 !test_bit(In_sync, &rdev->flags) &&
2568 mddev->curr_resync_completed > rdev->recovery_offset)
2569 rdev->recovery_offset = mddev->curr_resync_completed;
2572 if (!mddev->persistent) {
2573 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2574 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2575 if (!mddev->external) {
2576 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2577 rdev_for_each(rdev, mddev) {
2578 if (rdev->badblocks.changed) {
2579 rdev->badblocks.changed = 0;
2580 ack_all_badblocks(&rdev->badblocks);
2581 md_error(mddev, rdev);
2583 clear_bit(Blocked, &rdev->flags);
2584 clear_bit(BlockedBadBlocks, &rdev->flags);
2585 wake_up(&rdev->blocked_wait);
2588 wake_up(&mddev->sb_wait);
2592 spin_lock(&mddev->lock);
2594 mddev->utime = ktime_get_real_seconds();
2596 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2598 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2599 /* just a clean<-> dirty transition, possibly leave spares alone,
2600 * though if events isn't the right even/odd, we will have to do
2606 if (mddev->degraded)
2607 /* If the array is degraded, then skipping spares is both
2608 * dangerous and fairly pointless.
2609 * Dangerous because a device that was removed from the array
2610 * might have a event_count that still looks up-to-date,
2611 * so it can be re-added without a resync.
2612 * Pointless because if there are any spares to skip,
2613 * then a recovery will happen and soon that array won't
2614 * be degraded any more and the spare can go back to sleep then.
2618 sync_req = mddev->in_sync;
2620 /* If this is just a dirty<->clean transition, and the array is clean
2621 * and 'events' is odd, we can roll back to the previous clean state */
2623 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2624 && mddev->can_decrease_events
2625 && mddev->events != 1) {
2627 mddev->can_decrease_events = 0;
2629 /* otherwise we have to go forward and ... */
2631 mddev->can_decrease_events = nospares;
2635 * This 64-bit counter should never wrap.
2636 * Either we are in around ~1 trillion A.C., assuming
2637 * 1 reboot per second, or we have a bug...
2639 WARN_ON(mddev->events == 0);
2641 rdev_for_each(rdev, mddev) {
2642 if (rdev->badblocks.changed)
2643 any_badblocks_changed++;
2644 if (test_bit(Faulty, &rdev->flags))
2645 set_bit(FaultRecorded, &rdev->flags);
2648 sync_sbs(mddev, nospares);
2649 spin_unlock(&mddev->lock);
2651 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2652 mdname(mddev), mddev->in_sync);
2655 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2657 md_bitmap_update_sb(mddev->bitmap);
2658 rdev_for_each(rdev, mddev) {
2659 char b[BDEVNAME_SIZE];
2661 if (rdev->sb_loaded != 1)
2662 continue; /* no noise on spare devices */
2664 if (!test_bit(Faulty, &rdev->flags)) {
2665 md_super_write(mddev,rdev,
2666 rdev->sb_start, rdev->sb_size,
2668 pr_debug("md: (write) %s's sb offset: %llu\n",
2669 bdevname(rdev->bdev, b),
2670 (unsigned long long)rdev->sb_start);
2671 rdev->sb_events = mddev->events;
2672 if (rdev->badblocks.size) {
2673 md_super_write(mddev, rdev,
2674 rdev->badblocks.sector,
2675 rdev->badblocks.size << 9,
2677 rdev->badblocks.size = 0;
2681 pr_debug("md: %s (skipping faulty)\n",
2682 bdevname(rdev->bdev, b));
2684 if (mddev->level == LEVEL_MULTIPATH)
2685 /* only need to write one superblock... */
2688 if (md_super_wait(mddev) < 0)
2690 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2692 if (mddev_is_clustered(mddev) && ret == 0)
2693 md_cluster_ops->metadata_update_finish(mddev);
2695 if (mddev->in_sync != sync_req ||
2696 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2697 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2698 /* have to write it out again */
2700 wake_up(&mddev->sb_wait);
2701 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2702 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2704 rdev_for_each(rdev, mddev) {
2705 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2706 clear_bit(Blocked, &rdev->flags);
2708 if (any_badblocks_changed)
2709 ack_all_badblocks(&rdev->badblocks);
2710 clear_bit(BlockedBadBlocks, &rdev->flags);
2711 wake_up(&rdev->blocked_wait);
2714 EXPORT_SYMBOL(md_update_sb);
2716 static int add_bound_rdev(struct md_rdev *rdev)
2718 struct mddev *mddev = rdev->mddev;
2720 bool add_journal = test_bit(Journal, &rdev->flags);
2722 if (!mddev->pers->hot_remove_disk || add_journal) {
2723 /* If there is hot_add_disk but no hot_remove_disk
2724 * then added disks for geometry changes,
2725 * and should be added immediately.
2727 super_types[mddev->major_version].
2728 validate_super(mddev, NULL/*freshest*/, rdev);
2730 mddev_suspend(mddev);
2731 err = mddev->pers->hot_add_disk(mddev, rdev);
2733 mddev_resume(mddev);
2735 md_kick_rdev_from_array(rdev);
2739 sysfs_notify_dirent_safe(rdev->sysfs_state);
2741 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2742 if (mddev->degraded)
2743 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2744 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2745 md_new_event(mddev);
2746 md_wakeup_thread(mddev->thread);
2750 /* words written to sysfs files may, or may not, be \n terminated.
2751 * We want to accept with case. For this we use cmd_match.
2753 static int cmd_match(const char *cmd, const char *str)
2755 /* See if cmd, written into a sysfs file, matches
2756 * str. They must either be the same, or cmd can
2757 * have a trailing newline
2759 while (*cmd && *str && *cmd == *str) {
2770 struct rdev_sysfs_entry {
2771 struct attribute attr;
2772 ssize_t (*show)(struct md_rdev *, char *);
2773 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2777 state_show(struct md_rdev *rdev, char *page)
2781 unsigned long flags = READ_ONCE(rdev->flags);
2783 if (test_bit(Faulty, &flags) ||
2784 (!test_bit(ExternalBbl, &flags) &&
2785 rdev->badblocks.unacked_exist))
2786 len += sprintf(page+len, "faulty%s", sep);
2787 if (test_bit(In_sync, &flags))
2788 len += sprintf(page+len, "in_sync%s", sep);
2789 if (test_bit(Journal, &flags))
2790 len += sprintf(page+len, "journal%s", sep);
2791 if (test_bit(WriteMostly, &flags))
2792 len += sprintf(page+len, "write_mostly%s", sep);
2793 if (test_bit(Blocked, &flags) ||
2794 (rdev->badblocks.unacked_exist
2795 && !test_bit(Faulty, &flags)))
2796 len += sprintf(page+len, "blocked%s", sep);
2797 if (!test_bit(Faulty, &flags) &&
2798 !test_bit(Journal, &flags) &&
2799 !test_bit(In_sync, &flags))
2800 len += sprintf(page+len, "spare%s", sep);
2801 if (test_bit(WriteErrorSeen, &flags))
2802 len += sprintf(page+len, "write_error%s", sep);
2803 if (test_bit(WantReplacement, &flags))
2804 len += sprintf(page+len, "want_replacement%s", sep);
2805 if (test_bit(Replacement, &flags))
2806 len += sprintf(page+len, "replacement%s", sep);
2807 if (test_bit(ExternalBbl, &flags))
2808 len += sprintf(page+len, "external_bbl%s", sep);
2809 if (test_bit(FailFast, &flags))
2810 len += sprintf(page+len, "failfast%s", sep);
2815 return len+sprintf(page+len, "\n");
2819 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2822 * faulty - simulates an error
2823 * remove - disconnects the device
2824 * writemostly - sets write_mostly
2825 * -writemostly - clears write_mostly
2826 * blocked - sets the Blocked flags
2827 * -blocked - clears the Blocked and possibly simulates an error
2828 * insync - sets Insync providing device isn't active
2829 * -insync - clear Insync for a device with a slot assigned,
2830 * so that it gets rebuilt based on bitmap
2831 * write_error - sets WriteErrorSeen
2832 * -write_error - clears WriteErrorSeen
2833 * {,-}failfast - set/clear FailFast
2836 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2837 md_error(rdev->mddev, rdev);
2838 if (test_bit(Faulty, &rdev->flags))
2842 } else if (cmd_match(buf, "remove")) {
2843 if (rdev->mddev->pers) {
2844 clear_bit(Blocked, &rdev->flags);
2845 remove_and_add_spares(rdev->mddev, rdev);
2847 if (rdev->raid_disk >= 0)
2850 struct mddev *mddev = rdev->mddev;
2852 if (mddev_is_clustered(mddev))
2853 err = md_cluster_ops->remove_disk(mddev, rdev);
2856 md_kick_rdev_from_array(rdev);
2858 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2859 md_wakeup_thread(mddev->thread);
2861 md_new_event(mddev);
2864 } else if (cmd_match(buf, "writemostly")) {
2865 set_bit(WriteMostly, &rdev->flags);
2867 } else if (cmd_match(buf, "-writemostly")) {
2868 clear_bit(WriteMostly, &rdev->flags);
2870 } else if (cmd_match(buf, "blocked")) {
2871 set_bit(Blocked, &rdev->flags);
2873 } else if (cmd_match(buf, "-blocked")) {
2874 if (!test_bit(Faulty, &rdev->flags) &&
2875 !test_bit(ExternalBbl, &rdev->flags) &&
2876 rdev->badblocks.unacked_exist) {
2877 /* metadata handler doesn't understand badblocks,
2878 * so we need to fail the device
2880 md_error(rdev->mddev, rdev);
2882 clear_bit(Blocked, &rdev->flags);
2883 clear_bit(BlockedBadBlocks, &rdev->flags);
2884 wake_up(&rdev->blocked_wait);
2885 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2886 md_wakeup_thread(rdev->mddev->thread);
2889 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2890 set_bit(In_sync, &rdev->flags);
2892 } else if (cmd_match(buf, "failfast")) {
2893 set_bit(FailFast, &rdev->flags);
2895 } else if (cmd_match(buf, "-failfast")) {
2896 clear_bit(FailFast, &rdev->flags);
2898 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2899 !test_bit(Journal, &rdev->flags)) {
2900 if (rdev->mddev->pers == NULL) {
2901 clear_bit(In_sync, &rdev->flags);
2902 rdev->saved_raid_disk = rdev->raid_disk;
2903 rdev->raid_disk = -1;
2906 } else if (cmd_match(buf, "write_error")) {
2907 set_bit(WriteErrorSeen, &rdev->flags);
2909 } else if (cmd_match(buf, "-write_error")) {
2910 clear_bit(WriteErrorSeen, &rdev->flags);
2912 } else if (cmd_match(buf, "want_replacement")) {
2913 /* Any non-spare device that is not a replacement can
2914 * become want_replacement at any time, but we then need to
2915 * check if recovery is needed.
2917 if (rdev->raid_disk >= 0 &&
2918 !test_bit(Journal, &rdev->flags) &&
2919 !test_bit(Replacement, &rdev->flags))
2920 set_bit(WantReplacement, &rdev->flags);
2921 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2922 md_wakeup_thread(rdev->mddev->thread);
2924 } else if (cmd_match(buf, "-want_replacement")) {
2925 /* Clearing 'want_replacement' is always allowed.
2926 * Once replacements starts it is too late though.
2929 clear_bit(WantReplacement, &rdev->flags);
2930 } else if (cmd_match(buf, "replacement")) {
2931 /* Can only set a device as a replacement when array has not
2932 * yet been started. Once running, replacement is automatic
2933 * from spares, or by assigning 'slot'.
2935 if (rdev->mddev->pers)
2938 set_bit(Replacement, &rdev->flags);
2941 } else if (cmd_match(buf, "-replacement")) {
2942 /* Similarly, can only clear Replacement before start */
2943 if (rdev->mddev->pers)
2946 clear_bit(Replacement, &rdev->flags);
2949 } else if (cmd_match(buf, "re-add")) {
2950 if (!rdev->mddev->pers)
2952 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2953 rdev->saved_raid_disk >= 0) {
2954 /* clear_bit is performed _after_ all the devices
2955 * have their local Faulty bit cleared. If any writes
2956 * happen in the meantime in the local node, they
2957 * will land in the local bitmap, which will be synced
2958 * by this node eventually
2960 if (!mddev_is_clustered(rdev->mddev) ||
2961 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2962 clear_bit(Faulty, &rdev->flags);
2963 err = add_bound_rdev(rdev);
2967 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2968 set_bit(ExternalBbl, &rdev->flags);
2969 rdev->badblocks.shift = 0;
2971 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2972 clear_bit(ExternalBbl, &rdev->flags);
2976 sysfs_notify_dirent_safe(rdev->sysfs_state);
2977 return err ? err : len;
2979 static struct rdev_sysfs_entry rdev_state =
2980 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2983 errors_show(struct md_rdev *rdev, char *page)
2985 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2989 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2994 rv = kstrtouint(buf, 10, &n);
2997 atomic_set(&rdev->corrected_errors, n);
3000 static struct rdev_sysfs_entry rdev_errors =
3001 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3004 slot_show(struct md_rdev *rdev, char *page)
3006 if (test_bit(Journal, &rdev->flags))
3007 return sprintf(page, "journal\n");
3008 else if (rdev->raid_disk < 0)
3009 return sprintf(page, "none\n");
3011 return sprintf(page, "%d\n", rdev->raid_disk);
3015 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3020 if (test_bit(Journal, &rdev->flags))
3022 if (strncmp(buf, "none", 4)==0)
3025 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3032 if (rdev->mddev->pers && slot == -1) {
3033 /* Setting 'slot' on an active array requires also
3034 * updating the 'rd%d' link, and communicating
3035 * with the personality with ->hot_*_disk.
3036 * For now we only support removing
3037 * failed/spare devices. This normally happens automatically,
3038 * but not when the metadata is externally managed.
3040 if (rdev->raid_disk == -1)
3042 /* personality does all needed checks */
3043 if (rdev->mddev->pers->hot_remove_disk == NULL)
3045 clear_bit(Blocked, &rdev->flags);
3046 remove_and_add_spares(rdev->mddev, rdev);
3047 if (rdev->raid_disk >= 0)
3049 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3050 md_wakeup_thread(rdev->mddev->thread);
3051 } else if (rdev->mddev->pers) {
3052 /* Activating a spare .. or possibly reactivating
3053 * if we ever get bitmaps working here.
3057 if (rdev->raid_disk != -1)
3060 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3063 if (rdev->mddev->pers->hot_add_disk == NULL)
3066 if (slot >= rdev->mddev->raid_disks &&
3067 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3070 rdev->raid_disk = slot;
3071 if (test_bit(In_sync, &rdev->flags))
3072 rdev->saved_raid_disk = slot;
3074 rdev->saved_raid_disk = -1;
3075 clear_bit(In_sync, &rdev->flags);
3076 clear_bit(Bitmap_sync, &rdev->flags);
3077 err = rdev->mddev->pers->
3078 hot_add_disk(rdev->mddev, rdev);
3080 rdev->raid_disk = -1;
3083 sysfs_notify_dirent_safe(rdev->sysfs_state);
3084 if (sysfs_link_rdev(rdev->mddev, rdev))
3085 /* failure here is OK */;
3086 /* don't wakeup anyone, leave that to userspace. */
3088 if (slot >= rdev->mddev->raid_disks &&
3089 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3091 rdev->raid_disk = slot;
3092 /* assume it is working */
3093 clear_bit(Faulty, &rdev->flags);
3094 clear_bit(WriteMostly, &rdev->flags);
3095 set_bit(In_sync, &rdev->flags);
3096 sysfs_notify_dirent_safe(rdev->sysfs_state);
3101 static struct rdev_sysfs_entry rdev_slot =
3102 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3105 offset_show(struct md_rdev *rdev, char *page)
3107 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3111 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3113 unsigned long long offset;
3114 if (kstrtoull(buf, 10, &offset) < 0)
3116 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3118 if (rdev->sectors && rdev->mddev->external)
3119 /* Must set offset before size, so overlap checks
3122 rdev->data_offset = offset;
3123 rdev->new_data_offset = offset;
3127 static struct rdev_sysfs_entry rdev_offset =
3128 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3130 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3132 return sprintf(page, "%llu\n",
3133 (unsigned long long)rdev->new_data_offset);
3136 static ssize_t new_offset_store(struct md_rdev *rdev,
3137 const char *buf, size_t len)
3139 unsigned long long new_offset;
3140 struct mddev *mddev = rdev->mddev;
3142 if (kstrtoull(buf, 10, &new_offset) < 0)
3145 if (mddev->sync_thread ||
3146 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3148 if (new_offset == rdev->data_offset)
3149 /* reset is always permitted */
3151 else if (new_offset > rdev->data_offset) {
3152 /* must not push array size beyond rdev_sectors */
3153 if (new_offset - rdev->data_offset
3154 + mddev->dev_sectors > rdev->sectors)
3157 /* Metadata worries about other space details. */
3159 /* decreasing the offset is inconsistent with a backwards
3162 if (new_offset < rdev->data_offset &&
3163 mddev->reshape_backwards)
3165 /* Increasing offset is inconsistent with forwards
3166 * reshape. reshape_direction should be set to
3167 * 'backwards' first.
3169 if (new_offset > rdev->data_offset &&
3170 !mddev->reshape_backwards)
3173 if (mddev->pers && mddev->persistent &&
3174 !super_types[mddev->major_version]
3175 .allow_new_offset(rdev, new_offset))
3177 rdev->new_data_offset = new_offset;
3178 if (new_offset > rdev->data_offset)
3179 mddev->reshape_backwards = 1;
3180 else if (new_offset < rdev->data_offset)
3181 mddev->reshape_backwards = 0;
3185 static struct rdev_sysfs_entry rdev_new_offset =
3186 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3189 rdev_size_show(struct md_rdev *rdev, char *page)
3191 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3194 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3196 /* check if two start/length pairs overlap */
3204 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3206 unsigned long long blocks;
3209 if (kstrtoull(buf, 10, &blocks) < 0)
3212 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3213 return -EINVAL; /* sector conversion overflow */
3216 if (new != blocks * 2)
3217 return -EINVAL; /* unsigned long long to sector_t overflow */
3224 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3226 struct mddev *my_mddev = rdev->mddev;
3227 sector_t oldsectors = rdev->sectors;
3230 if (test_bit(Journal, &rdev->flags))
3232 if (strict_blocks_to_sectors(buf, §ors) < 0)
3234 if (rdev->data_offset != rdev->new_data_offset)
3235 return -EINVAL; /* too confusing */
3236 if (my_mddev->pers && rdev->raid_disk >= 0) {
3237 if (my_mddev->persistent) {
3238 sectors = super_types[my_mddev->major_version].
3239 rdev_size_change(rdev, sectors);
3242 } else if (!sectors)
3243 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3245 if (!my_mddev->pers->resize)
3246 /* Cannot change size for RAID0 or Linear etc */
3249 if (sectors < my_mddev->dev_sectors)
3250 return -EINVAL; /* component must fit device */
3252 rdev->sectors = sectors;
3253 if (sectors > oldsectors && my_mddev->external) {
3254 /* Need to check that all other rdevs with the same
3255 * ->bdev do not overlap. 'rcu' is sufficient to walk
3256 * the rdev lists safely.
3257 * This check does not provide a hard guarantee, it
3258 * just helps avoid dangerous mistakes.
3260 struct mddev *mddev;
3262 struct list_head *tmp;
3265 for_each_mddev(mddev, tmp) {
3266 struct md_rdev *rdev2;
3268 rdev_for_each(rdev2, mddev)
3269 if (rdev->bdev == rdev2->bdev &&
3271 overlaps(rdev->data_offset, rdev->sectors,
3284 /* Someone else could have slipped in a size
3285 * change here, but doing so is just silly.
3286 * We put oldsectors back because we *know* it is
3287 * safe, and trust userspace not to race with
3290 rdev->sectors = oldsectors;
3297 static struct rdev_sysfs_entry rdev_size =
3298 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3300 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3302 unsigned long long recovery_start = rdev->recovery_offset;
3304 if (test_bit(In_sync, &rdev->flags) ||
3305 recovery_start == MaxSector)
3306 return sprintf(page, "none\n");
3308 return sprintf(page, "%llu\n", recovery_start);
3311 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3313 unsigned long long recovery_start;
3315 if (cmd_match(buf, "none"))
3316 recovery_start = MaxSector;
3317 else if (kstrtoull(buf, 10, &recovery_start))
3320 if (rdev->mddev->pers &&
3321 rdev->raid_disk >= 0)
3324 rdev->recovery_offset = recovery_start;
3325 if (recovery_start == MaxSector)
3326 set_bit(In_sync, &rdev->flags);
3328 clear_bit(In_sync, &rdev->flags);
3332 static struct rdev_sysfs_entry rdev_recovery_start =
3333 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3335 /* sysfs access to bad-blocks list.
3336 * We present two files.
3337 * 'bad-blocks' lists sector numbers and lengths of ranges that
3338 * are recorded as bad. The list is truncated to fit within
3339 * the one-page limit of sysfs.
3340 * Writing "sector length" to this file adds an acknowledged
3342 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3343 * been acknowledged. Writing to this file adds bad blocks
3344 * without acknowledging them. This is largely for testing.
3346 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3348 return badblocks_show(&rdev->badblocks, page, 0);
3350 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3352 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3353 /* Maybe that ack was all we needed */
3354 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3355 wake_up(&rdev->blocked_wait);
3358 static struct rdev_sysfs_entry rdev_bad_blocks =
3359 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3361 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3363 return badblocks_show(&rdev->badblocks, page, 1);
3365 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3367 return badblocks_store(&rdev->badblocks, page, len, 1);
3369 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3370 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3373 ppl_sector_show(struct md_rdev *rdev, char *page)
3375 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3379 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3381 unsigned long long sector;
3383 if (kstrtoull(buf, 10, §or) < 0)
3385 if (sector != (sector_t)sector)
3388 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3389 rdev->raid_disk >= 0)
3392 if (rdev->mddev->persistent) {
3393 if (rdev->mddev->major_version == 0)
3395 if ((sector > rdev->sb_start &&
3396 sector - rdev->sb_start > S16_MAX) ||
3397 (sector < rdev->sb_start &&
3398 rdev->sb_start - sector > -S16_MIN))
3400 rdev->ppl.offset = sector - rdev->sb_start;
3401 } else if (!rdev->mddev->external) {
3404 rdev->ppl.sector = sector;
3408 static struct rdev_sysfs_entry rdev_ppl_sector =
3409 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3412 ppl_size_show(struct md_rdev *rdev, char *page)
3414 return sprintf(page, "%u\n", rdev->ppl.size);
3418 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3422 if (kstrtouint(buf, 10, &size) < 0)
3425 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3426 rdev->raid_disk >= 0)
3429 if (rdev->mddev->persistent) {
3430 if (rdev->mddev->major_version == 0)
3434 } else if (!rdev->mddev->external) {
3437 rdev->ppl.size = size;
3441 static struct rdev_sysfs_entry rdev_ppl_size =
3442 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3444 static struct attribute *rdev_default_attrs[] = {
3449 &rdev_new_offset.attr,
3451 &rdev_recovery_start.attr,
3452 &rdev_bad_blocks.attr,
3453 &rdev_unack_bad_blocks.attr,
3454 &rdev_ppl_sector.attr,
3455 &rdev_ppl_size.attr,
3459 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3461 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3462 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3468 return entry->show(rdev, page);
3472 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3473 const char *page, size_t length)
3475 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3476 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3478 struct mddev *mddev = rdev->mddev;
3482 if (!capable(CAP_SYS_ADMIN))
3484 rv = mddev ? mddev_lock(mddev): -EBUSY;
3486 if (rdev->mddev == NULL)
3489 rv = entry->store(rdev, page, length);
3490 mddev_unlock(mddev);
3495 static void rdev_free(struct kobject *ko)
3497 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3500 static const struct sysfs_ops rdev_sysfs_ops = {
3501 .show = rdev_attr_show,
3502 .store = rdev_attr_store,
3504 static struct kobj_type rdev_ktype = {
3505 .release = rdev_free,
3506 .sysfs_ops = &rdev_sysfs_ops,
3507 .default_attrs = rdev_default_attrs,
3510 int md_rdev_init(struct md_rdev *rdev)
3513 rdev->saved_raid_disk = -1;
3514 rdev->raid_disk = -1;
3516 rdev->data_offset = 0;
3517 rdev->new_data_offset = 0;
3518 rdev->sb_events = 0;
3519 rdev->last_read_error = 0;
3520 rdev->sb_loaded = 0;
3521 rdev->bb_page = NULL;
3522 atomic_set(&rdev->nr_pending, 0);
3523 atomic_set(&rdev->read_errors, 0);
3524 atomic_set(&rdev->corrected_errors, 0);
3526 INIT_LIST_HEAD(&rdev->same_set);
3527 init_waitqueue_head(&rdev->blocked_wait);
3529 /* Add space to store bad block list.
3530 * This reserves the space even on arrays where it cannot
3531 * be used - I wonder if that matters
3533 return badblocks_init(&rdev->badblocks, 0);
3535 EXPORT_SYMBOL_GPL(md_rdev_init);
3537 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3539 * mark the device faulty if:
3541 * - the device is nonexistent (zero size)
3542 * - the device has no valid superblock
3544 * a faulty rdev _never_ has rdev->sb set.
3546 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3548 char b[BDEVNAME_SIZE];
3550 struct md_rdev *rdev;
3553 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3555 return ERR_PTR(-ENOMEM);
3557 err = md_rdev_init(rdev);
3560 err = alloc_disk_sb(rdev);
3564 err = lock_rdev(rdev, newdev, super_format == -2);
3568 kobject_init(&rdev->kobj, &rdev_ktype);
3570 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3572 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3573 bdevname(rdev->bdev,b));
3578 if (super_format >= 0) {
3579 err = super_types[super_format].
3580 load_super(rdev, NULL, super_minor);
3581 if (err == -EINVAL) {
3582 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3583 bdevname(rdev->bdev,b),
3584 super_format, super_minor);
3588 pr_warn("md: could not read %s's sb, not importing!\n",
3589 bdevname(rdev->bdev,b));
3599 md_rdev_clear(rdev);
3601 return ERR_PTR(err);
3605 * Check a full RAID array for plausibility
3608 static void analyze_sbs(struct mddev *mddev)
3611 struct md_rdev *rdev, *freshest, *tmp;
3612 char b[BDEVNAME_SIZE];
3615 rdev_for_each_safe(rdev, tmp, mddev)
3616 switch (super_types[mddev->major_version].
3617 load_super(rdev, freshest, mddev->minor_version)) {
3624 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3625 bdevname(rdev->bdev,b));
3626 md_kick_rdev_from_array(rdev);
3629 super_types[mddev->major_version].
3630 validate_super(mddev, NULL/*freshest*/, freshest);
3633 rdev_for_each_safe(rdev, tmp, mddev) {
3634 if (mddev->max_disks &&
3635 (rdev->desc_nr >= mddev->max_disks ||
3636 i > mddev->max_disks)) {
3637 pr_warn("md: %s: %s: only %d devices permitted\n",
3638 mdname(mddev), bdevname(rdev->bdev, b),
3640 md_kick_rdev_from_array(rdev);
3643 if (rdev != freshest) {
3644 if (super_types[mddev->major_version].
3645 validate_super(mddev, freshest, rdev)) {
3646 pr_warn("md: kicking non-fresh %s from array!\n",
3647 bdevname(rdev->bdev,b));
3648 md_kick_rdev_from_array(rdev);
3652 if (mddev->level == LEVEL_MULTIPATH) {
3653 rdev->desc_nr = i++;
3654 rdev->raid_disk = rdev->desc_nr;
3655 set_bit(In_sync, &rdev->flags);
3656 } else if (rdev->raid_disk >=
3657 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3658 !test_bit(Journal, &rdev->flags)) {
3659 rdev->raid_disk = -1;
3660 clear_bit(In_sync, &rdev->flags);
3665 /* Read a fixed-point number.
3666 * Numbers in sysfs attributes should be in "standard" units where
3667 * possible, so time should be in seconds.
3668 * However we internally use a a much smaller unit such as
3669 * milliseconds or jiffies.
3670 * This function takes a decimal number with a possible fractional
3671 * component, and produces an integer which is the result of
3672 * multiplying that number by 10^'scale'.
3673 * all without any floating-point arithmetic.
3675 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3677 unsigned long result = 0;
3679 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3682 else if (decimals < scale) {
3685 result = result * 10 + value;
3697 while (decimals < scale) {
3706 safe_delay_show(struct mddev *mddev, char *page)
3708 unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3710 return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3713 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3717 if (mddev_is_clustered(mddev)) {
3718 pr_warn("md: Safemode is disabled for clustered mode\n");
3722 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3725 mddev->safemode_delay = 0;
3727 unsigned long old_delay = mddev->safemode_delay;
3728 unsigned long new_delay = (msec*HZ)/1000;
3732 mddev->safemode_delay = new_delay;
3733 if (new_delay < old_delay || old_delay == 0)
3734 mod_timer(&mddev->safemode_timer, jiffies+1);
3738 static struct md_sysfs_entry md_safe_delay =
3739 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3742 level_show(struct mddev *mddev, char *page)
3744 struct md_personality *p;
3746 spin_lock(&mddev->lock);
3749 ret = sprintf(page, "%s\n", p->name);
3750 else if (mddev->clevel[0])
3751 ret = sprintf(page, "%s\n", mddev->clevel);
3752 else if (mddev->level != LEVEL_NONE)
3753 ret = sprintf(page, "%d\n", mddev->level);
3756 spin_unlock(&mddev->lock);
3761 level_store(struct mddev *mddev, const char *buf, size_t len)
3766 struct md_personality *pers, *oldpers;
3768 void *priv, *oldpriv;
3769 struct md_rdev *rdev;
3771 if (slen == 0 || slen >= sizeof(clevel))
3774 rv = mddev_lock(mddev);
3778 if (mddev->pers == NULL) {
3779 strncpy(mddev->clevel, buf, slen);
3780 if (mddev->clevel[slen-1] == '\n')
3782 mddev->clevel[slen] = 0;
3783 mddev->level = LEVEL_NONE;
3791 /* request to change the personality. Need to ensure:
3792 * - array is not engaged in resync/recovery/reshape
3793 * - old personality can be suspended
3794 * - new personality will access other array.
3798 if (mddev->sync_thread ||
3799 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3800 mddev->reshape_position != MaxSector ||
3801 mddev->sysfs_active)
3805 if (!mddev->pers->quiesce) {
3806 pr_warn("md: %s: %s does not support online personality change\n",
3807 mdname(mddev), mddev->pers->name);
3811 /* Now find the new personality */
3812 strncpy(clevel, buf, slen);
3813 if (clevel[slen-1] == '\n')
3816 if (kstrtol(clevel, 10, &level))
3819 if (request_module("md-%s", clevel) != 0)
3820 request_module("md-level-%s", clevel);
3821 spin_lock(&pers_lock);
3822 pers = find_pers(level, clevel);
3823 if (!pers || !try_module_get(pers->owner)) {
3824 spin_unlock(&pers_lock);
3825 pr_warn("md: personality %s not loaded\n", clevel);
3829 spin_unlock(&pers_lock);
3831 if (pers == mddev->pers) {
3832 /* Nothing to do! */
3833 module_put(pers->owner);
3837 if (!pers->takeover) {
3838 module_put(pers->owner);
3839 pr_warn("md: %s: %s does not support personality takeover\n",
3840 mdname(mddev), clevel);
3845 rdev_for_each(rdev, mddev)
3846 rdev->new_raid_disk = rdev->raid_disk;
3848 /* ->takeover must set new_* and/or delta_disks
3849 * if it succeeds, and may set them when it fails.
3851 priv = pers->takeover(mddev);
3853 mddev->new_level = mddev->level;
3854 mddev->new_layout = mddev->layout;
3855 mddev->new_chunk_sectors = mddev->chunk_sectors;
3856 mddev->raid_disks -= mddev->delta_disks;
3857 mddev->delta_disks = 0;
3858 mddev->reshape_backwards = 0;
3859 module_put(pers->owner);
3860 pr_warn("md: %s: %s would not accept array\n",
3861 mdname(mddev), clevel);
3866 /* Looks like we have a winner */
3867 mddev_suspend(mddev);
3868 mddev_detach(mddev);
3870 spin_lock(&mddev->lock);
3871 oldpers = mddev->pers;
3872 oldpriv = mddev->private;
3874 mddev->private = priv;
3875 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3876 mddev->level = mddev->new_level;
3877 mddev->layout = mddev->new_layout;
3878 mddev->chunk_sectors = mddev->new_chunk_sectors;
3879 mddev->delta_disks = 0;
3880 mddev->reshape_backwards = 0;
3881 mddev->degraded = 0;
3882 spin_unlock(&mddev->lock);
3884 if (oldpers->sync_request == NULL &&
3886 /* We are converting from a no-redundancy array
3887 * to a redundancy array and metadata is managed
3888 * externally so we need to be sure that writes
3889 * won't block due to a need to transition
3891 * until external management is started.
3894 mddev->safemode_delay = 0;
3895 mddev->safemode = 0;
3898 oldpers->free(mddev, oldpriv);
3900 if (oldpers->sync_request == NULL &&
3901 pers->sync_request != NULL) {
3902 /* need to add the md_redundancy_group */
3903 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3904 pr_warn("md: cannot register extra attributes for %s\n",
3906 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3908 if (oldpers->sync_request != NULL &&
3909 pers->sync_request == NULL) {
3910 /* need to remove the md_redundancy_group */
3911 if (mddev->to_remove == NULL)
3912 mddev->to_remove = &md_redundancy_group;
3915 module_put(oldpers->owner);
3917 rdev_for_each(rdev, mddev) {
3918 if (rdev->raid_disk < 0)
3920 if (rdev->new_raid_disk >= mddev->raid_disks)
3921 rdev->new_raid_disk = -1;
3922 if (rdev->new_raid_disk == rdev->raid_disk)
3924 sysfs_unlink_rdev(mddev, rdev);
3926 rdev_for_each(rdev, mddev) {
3927 if (rdev->raid_disk < 0)
3929 if (rdev->new_raid_disk == rdev->raid_disk)
3931 rdev->raid_disk = rdev->new_raid_disk;
3932 if (rdev->raid_disk < 0)
3933 clear_bit(In_sync, &rdev->flags);
3935 if (sysfs_link_rdev(mddev, rdev))
3936 pr_warn("md: cannot register rd%d for %s after level change\n",
3937 rdev->raid_disk, mdname(mddev));
3941 if (pers->sync_request == NULL) {
3942 /* this is now an array without redundancy, so
3943 * it must always be in_sync
3946 del_timer_sync(&mddev->safemode_timer);
3948 blk_set_stacking_limits(&mddev->queue->limits);
3950 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3951 mddev_resume(mddev);
3953 md_update_sb(mddev, 1);
3954 sysfs_notify(&mddev->kobj, NULL, "level");
3955 md_new_event(mddev);
3958 mddev_unlock(mddev);
3962 static struct md_sysfs_entry md_level =
3963 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3966 layout_show(struct mddev *mddev, char *page)
3968 /* just a number, not meaningful for all levels */
3969 if (mddev->reshape_position != MaxSector &&
3970 mddev->layout != mddev->new_layout)
3971 return sprintf(page, "%d (%d)\n",
3972 mddev->new_layout, mddev->layout);
3973 return sprintf(page, "%d\n", mddev->layout);
3977 layout_store(struct mddev *mddev, const char *buf, size_t len)
3982 err = kstrtouint(buf, 10, &n);
3985 err = mddev_lock(mddev);
3990 if (mddev->pers->check_reshape == NULL)
3995 mddev->new_layout = n;
3996 err = mddev->pers->check_reshape(mddev);
3998 mddev->new_layout = mddev->layout;
4001 mddev->new_layout = n;
4002 if (mddev->reshape_position == MaxSector)
4005 mddev_unlock(mddev);
4008 static struct md_sysfs_entry md_layout =
4009 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4012 raid_disks_show(struct mddev *mddev, char *page)
4014 if (mddev->raid_disks == 0)
4016 if (mddev->reshape_position != MaxSector &&
4017 mddev->delta_disks != 0)
4018 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4019 mddev->raid_disks - mddev->delta_disks);
4020 return sprintf(page, "%d\n", mddev->raid_disks);
4023 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4026 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4031 err = kstrtouint(buf, 10, &n);
4035 err = mddev_lock(mddev);
4039 err = update_raid_disks(mddev, n);
4040 else if (mddev->reshape_position != MaxSector) {
4041 struct md_rdev *rdev;
4042 int olddisks = mddev->raid_disks - mddev->delta_disks;
4045 rdev_for_each(rdev, mddev) {
4047 rdev->data_offset < rdev->new_data_offset)
4050 rdev->data_offset > rdev->new_data_offset)
4054 mddev->delta_disks = n - olddisks;
4055 mddev->raid_disks = n;
4056 mddev->reshape_backwards = (mddev->delta_disks < 0);
4058 mddev->raid_disks = n;
4060 mddev_unlock(mddev);
4061 return err ? err : len;
4063 static struct md_sysfs_entry md_raid_disks =
4064 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4067 chunk_size_show(struct mddev *mddev, char *page)
4069 if (mddev->reshape_position != MaxSector &&
4070 mddev->chunk_sectors != mddev->new_chunk_sectors)
4071 return sprintf(page, "%d (%d)\n",
4072 mddev->new_chunk_sectors << 9,
4073 mddev->chunk_sectors << 9);
4074 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4078 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4083 err = kstrtoul(buf, 10, &n);
4087 err = mddev_lock(mddev);
4091 if (mddev->pers->check_reshape == NULL)
4096 mddev->new_chunk_sectors = n >> 9;
4097 err = mddev->pers->check_reshape(mddev);
4099 mddev->new_chunk_sectors = mddev->chunk_sectors;
4102 mddev->new_chunk_sectors = n >> 9;
4103 if (mddev->reshape_position == MaxSector)
4104 mddev->chunk_sectors = n >> 9;
4106 mddev_unlock(mddev);
4109 static struct md_sysfs_entry md_chunk_size =
4110 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4113 resync_start_show(struct mddev *mddev, char *page)
4115 if (mddev->recovery_cp == MaxSector)
4116 return sprintf(page, "none\n");
4117 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4121 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4123 unsigned long long n;
4126 if (cmd_match(buf, "none"))
4129 err = kstrtoull(buf, 10, &n);
4132 if (n != (sector_t)n)
4136 err = mddev_lock(mddev);
4139 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4143 mddev->recovery_cp = n;
4145 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4147 mddev_unlock(mddev);
4150 static struct md_sysfs_entry md_resync_start =
4151 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4152 resync_start_show, resync_start_store);
4155 * The array state can be:
4158 * No devices, no size, no level
4159 * Equivalent to STOP_ARRAY ioctl
4161 * May have some settings, but array is not active
4162 * all IO results in error
4163 * When written, doesn't tear down array, but just stops it
4164 * suspended (not supported yet)
4165 * All IO requests will block. The array can be reconfigured.
4166 * Writing this, if accepted, will block until array is quiescent
4168 * no resync can happen. no superblocks get written.
4169 * write requests fail
4171 * like readonly, but behaves like 'clean' on a write request.
4173 * clean - no pending writes, but otherwise active.
4174 * When written to inactive array, starts without resync
4175 * If a write request arrives then
4176 * if metadata is known, mark 'dirty' and switch to 'active'.
4177 * if not known, block and switch to write-pending
4178 * If written to an active array that has pending writes, then fails.
4180 * fully active: IO and resync can be happening.
4181 * When written to inactive array, starts with resync
4184 * clean, but writes are blocked waiting for 'active' to be written.
4187 * like active, but no writes have been seen for a while (100msec).
4190 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4191 write_pending, active_idle, bad_word};
4192 static char *array_states[] = {
4193 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4194 "write-pending", "active-idle", NULL };
4196 static int match_word(const char *word, char **list)
4199 for (n=0; list[n]; n++)
4200 if (cmd_match(word, list[n]))
4206 array_state_show(struct mddev *mddev, char *page)
4208 enum array_state st = inactive;
4210 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags))
4219 spin_lock(&mddev->lock);
4220 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4222 else if (mddev->in_sync)
4224 else if (mddev->safemode)
4228 spin_unlock(&mddev->lock);
4231 if (list_empty(&mddev->disks) &&
4232 mddev->raid_disks == 0 &&
4233 mddev->dev_sectors == 0)
4238 return sprintf(page, "%s\n", array_states[st]);
4241 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4242 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4243 static int do_md_run(struct mddev *mddev);
4244 static int restart_array(struct mddev *mddev);
4247 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4250 enum array_state st = match_word(buf, array_states);
4252 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4253 /* don't take reconfig_mutex when toggling between
4256 spin_lock(&mddev->lock);
4258 restart_array(mddev);
4259 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4260 md_wakeup_thread(mddev->thread);
4261 wake_up(&mddev->sb_wait);
4262 } else /* st == clean */ {
4263 restart_array(mddev);
4264 if (!set_in_sync(mddev))
4268 sysfs_notify_dirent_safe(mddev->sysfs_state);
4269 spin_unlock(&mddev->lock);
4272 err = mddev_lock(mddev);
4280 /* stopping an active array */
4281 err = do_md_stop(mddev, 0, NULL);
4284 /* stopping an active array */
4286 err = do_md_stop(mddev, 2, NULL);
4288 err = 0; /* already inactive */
4291 break; /* not supported yet */
4294 err = md_set_readonly(mddev, NULL);
4297 set_disk_ro(mddev->gendisk, 1);
4298 err = do_md_run(mddev);
4304 err = md_set_readonly(mddev, NULL);
4305 else if (mddev->ro == 1)
4306 err = restart_array(mddev);
4309 set_disk_ro(mddev->gendisk, 0);
4313 err = do_md_run(mddev);
4318 err = restart_array(mddev);
4321 spin_lock(&mddev->lock);
4322 if (!set_in_sync(mddev))
4324 spin_unlock(&mddev->lock);
4330 err = restart_array(mddev);
4333 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4334 wake_up(&mddev->sb_wait);
4338 set_disk_ro(mddev->gendisk, 0);
4339 err = do_md_run(mddev);
4344 /* these cannot be set */
4349 if (mddev->hold_active == UNTIL_IOCTL)
4350 mddev->hold_active = 0;
4351 sysfs_notify_dirent_safe(mddev->sysfs_state);
4353 mddev_unlock(mddev);
4356 static struct md_sysfs_entry md_array_state =
4357 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4360 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4361 return sprintf(page, "%d\n",
4362 atomic_read(&mddev->max_corr_read_errors));
4366 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4371 rv = kstrtouint(buf, 10, &n);
4376 atomic_set(&mddev->max_corr_read_errors, n);
4380 static struct md_sysfs_entry max_corr_read_errors =
4381 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4382 max_corrected_read_errors_store);
4385 null_show(struct mddev *mddev, char *page)
4391 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4393 /* buf must be %d:%d\n? giving major and minor numbers */
4394 /* The new device is added to the array.
4395 * If the array has a persistent superblock, we read the
4396 * superblock to initialise info and check validity.
4397 * Otherwise, only checking done is that in bind_rdev_to_array,
4398 * which mainly checks size.
4401 int major = simple_strtoul(buf, &e, 10);
4404 struct md_rdev *rdev;
4407 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4409 minor = simple_strtoul(e+1, &e, 10);
4410 if (*e && *e != '\n')
4412 dev = MKDEV(major, minor);
4413 if (major != MAJOR(dev) ||
4414 minor != MINOR(dev))
4417 flush_workqueue(md_misc_wq);
4419 err = mddev_lock(mddev);
4422 if (mddev->persistent) {
4423 rdev = md_import_device(dev, mddev->major_version,
4424 mddev->minor_version);
4425 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4426 struct md_rdev *rdev0
4427 = list_entry(mddev->disks.next,
4428 struct md_rdev, same_set);
4429 err = super_types[mddev->major_version]
4430 .load_super(rdev, rdev0, mddev->minor_version);
4434 } else if (mddev->external)
4435 rdev = md_import_device(dev, -2, -1);
4437 rdev = md_import_device(dev, -1, -1);
4440 mddev_unlock(mddev);
4441 return PTR_ERR(rdev);
4443 err = bind_rdev_to_array(rdev, mddev);
4447 mddev_unlock(mddev);
4449 md_new_event(mddev);
4450 return err ? err : len;
4453 static struct md_sysfs_entry md_new_device =
4454 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4457 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4460 unsigned long chunk, end_chunk;
4463 err = mddev_lock(mddev);
4468 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4470 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4471 if (buf == end) break;
4472 if (*end == '-') { /* range */
4474 end_chunk = simple_strtoul(buf, &end, 0);
4475 if (buf == end) break;
4477 if (*end && !isspace(*end)) break;
4478 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4479 buf = skip_spaces(end);
4481 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4483 mddev_unlock(mddev);
4487 static struct md_sysfs_entry md_bitmap =
4488 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4491 size_show(struct mddev *mddev, char *page)
4493 return sprintf(page, "%llu\n",
4494 (unsigned long long)mddev->dev_sectors / 2);
4497 static int update_size(struct mddev *mddev, sector_t num_sectors);
4500 size_store(struct mddev *mddev, const char *buf, size_t len)
4502 /* If array is inactive, we can reduce the component size, but
4503 * not increase it (except from 0).
4504 * If array is active, we can try an on-line resize
4507 int err = strict_blocks_to_sectors(buf, §ors);
4511 err = mddev_lock(mddev);
4515 err = update_size(mddev, sectors);
4517 md_update_sb(mddev, 1);
4519 if (mddev->dev_sectors == 0 ||
4520 mddev->dev_sectors > sectors)
4521 mddev->dev_sectors = sectors;
4525 mddev_unlock(mddev);
4526 return err ? err : len;
4529 static struct md_sysfs_entry md_size =
4530 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4532 /* Metadata version.
4534 * 'none' for arrays with no metadata (good luck...)
4535 * 'external' for arrays with externally managed metadata,
4536 * or N.M for internally known formats
4539 metadata_show(struct mddev *mddev, char *page)
4541 if (mddev->persistent)
4542 return sprintf(page, "%d.%d\n",
4543 mddev->major_version, mddev->minor_version);
4544 else if (mddev->external)
4545 return sprintf(page, "external:%s\n", mddev->metadata_type);
4547 return sprintf(page, "none\n");
4551 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4556 /* Changing the details of 'external' metadata is
4557 * always permitted. Otherwise there must be
4558 * no devices attached to the array.
4561 err = mddev_lock(mddev);
4565 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4567 else if (!list_empty(&mddev->disks))
4571 if (cmd_match(buf, "none")) {
4572 mddev->persistent = 0;
4573 mddev->external = 0;
4574 mddev->major_version = 0;
4575 mddev->minor_version = 90;
4578 if (strncmp(buf, "external:", 9) == 0) {
4579 size_t namelen = len-9;
4580 if (namelen >= sizeof(mddev->metadata_type))
4581 namelen = sizeof(mddev->metadata_type)-1;
4582 strncpy(mddev->metadata_type, buf+9, namelen);
4583 mddev->metadata_type[namelen] = 0;
4584 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4585 mddev->metadata_type[--namelen] = 0;
4586 mddev->persistent = 0;
4587 mddev->external = 1;
4588 mddev->major_version = 0;
4589 mddev->minor_version = 90;
4592 major = simple_strtoul(buf, &e, 10);
4594 if (e==buf || *e != '.')
4597 minor = simple_strtoul(buf, &e, 10);
4598 if (e==buf || (*e && *e != '\n') )
4601 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4603 mddev->major_version = major;
4604 mddev->minor_version = minor;
4605 mddev->persistent = 1;
4606 mddev->external = 0;
4609 mddev_unlock(mddev);
4613 static struct md_sysfs_entry md_metadata =
4614 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4617 action_show(struct mddev *mddev, char *page)
4619 char *type = "idle";
4620 unsigned long recovery = mddev->recovery;
4621 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4623 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4624 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4625 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4627 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4628 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4630 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4634 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4636 else if (mddev->reshape_position != MaxSector)
4639 return sprintf(page, "%s\n", type);
4643 action_store(struct mddev *mddev, const char *page, size_t len)
4645 if (!mddev->pers || !mddev->pers->sync_request)
4649 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4650 if (cmd_match(page, "frozen"))
4651 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4653 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4654 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4655 mddev_lock(mddev) == 0) {
4656 flush_workqueue(md_misc_wq);
4657 if (mddev->sync_thread) {
4658 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4659 md_reap_sync_thread(mddev);
4661 mddev_unlock(mddev);
4663 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4665 else if (cmd_match(page, "resync"))
4666 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4667 else if (cmd_match(page, "recover")) {
4668 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4669 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4670 } else if (cmd_match(page, "reshape")) {
4672 if (mddev->pers->start_reshape == NULL)
4674 err = mddev_lock(mddev);
4676 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4678 } else if (mddev->reshape_position == MaxSector ||
4679 mddev->pers->check_reshape == NULL ||
4680 mddev->pers->check_reshape(mddev)) {
4681 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4682 err = mddev->pers->start_reshape(mddev);
4685 * If reshape is still in progress, and
4686 * md_check_recovery() can continue to reshape,
4687 * don't restart reshape because data can be
4688 * corrupted for raid456.
4690 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4692 mddev_unlock(mddev);
4696 sysfs_notify(&mddev->kobj, NULL, "degraded");
4698 if (cmd_match(page, "check"))
4699 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4700 else if (!cmd_match(page, "repair"))
4702 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4703 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4704 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4706 if (mddev->ro == 2) {
4707 /* A write to sync_action is enough to justify
4708 * canceling read-auto mode
4711 md_wakeup_thread(mddev->sync_thread);
4713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4714 md_wakeup_thread(mddev->thread);
4715 sysfs_notify_dirent_safe(mddev->sysfs_action);
4719 static struct md_sysfs_entry md_scan_mode =
4720 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4723 last_sync_action_show(struct mddev *mddev, char *page)
4725 return sprintf(page, "%s\n", mddev->last_sync_action);
4728 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4731 mismatch_cnt_show(struct mddev *mddev, char *page)
4733 return sprintf(page, "%llu\n",
4734 (unsigned long long)
4735 atomic64_read(&mddev->resync_mismatches));
4738 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4741 sync_min_show(struct mddev *mddev, char *page)
4743 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4744 mddev->sync_speed_min ? "local": "system");
4748 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4753 if (strncmp(buf, "system", 6)==0) {
4756 rv = kstrtouint(buf, 10, &min);
4762 mddev->sync_speed_min = min;
4766 static struct md_sysfs_entry md_sync_min =
4767 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4770 sync_max_show(struct mddev *mddev, char *page)
4772 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4773 mddev->sync_speed_max ? "local": "system");
4777 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4782 if (strncmp(buf, "system", 6)==0) {
4785 rv = kstrtouint(buf, 10, &max);
4791 mddev->sync_speed_max = max;
4795 static struct md_sysfs_entry md_sync_max =
4796 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4799 degraded_show(struct mddev *mddev, char *page)
4801 return sprintf(page, "%d\n", mddev->degraded);
4803 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4806 sync_force_parallel_show(struct mddev *mddev, char *page)
4808 return sprintf(page, "%d\n", mddev->parallel_resync);
4812 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4816 if (kstrtol(buf, 10, &n))
4819 if (n != 0 && n != 1)
4822 mddev->parallel_resync = n;
4824 if (mddev->sync_thread)
4825 wake_up(&resync_wait);
4830 /* force parallel resync, even with shared block devices */
4831 static struct md_sysfs_entry md_sync_force_parallel =
4832 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4833 sync_force_parallel_show, sync_force_parallel_store);
4836 sync_speed_show(struct mddev *mddev, char *page)
4838 unsigned long resync, dt, db;
4839 if (mddev->curr_resync == 0)
4840 return sprintf(page, "none\n");
4841 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4842 dt = (jiffies - mddev->resync_mark) / HZ;
4844 db = resync - mddev->resync_mark_cnt;
4845 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4848 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4851 sync_completed_show(struct mddev *mddev, char *page)
4853 unsigned long long max_sectors, resync;
4855 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4856 return sprintf(page, "none\n");
4858 if (mddev->curr_resync == 1 ||
4859 mddev->curr_resync == 2)
4860 return sprintf(page, "delayed\n");
4862 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4863 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4864 max_sectors = mddev->resync_max_sectors;
4866 max_sectors = mddev->dev_sectors;
4868 resync = mddev->curr_resync_completed;
4869 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4872 static struct md_sysfs_entry md_sync_completed =
4873 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4876 min_sync_show(struct mddev *mddev, char *page)
4878 return sprintf(page, "%llu\n",
4879 (unsigned long long)mddev->resync_min);
4882 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4884 unsigned long long min;
4887 if (kstrtoull(buf, 10, &min))
4890 spin_lock(&mddev->lock);
4892 if (min > mddev->resync_max)
4896 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4899 /* Round down to multiple of 4K for safety */
4900 mddev->resync_min = round_down(min, 8);
4904 spin_unlock(&mddev->lock);
4908 static struct md_sysfs_entry md_min_sync =
4909 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4912 max_sync_show(struct mddev *mddev, char *page)
4914 if (mddev->resync_max == MaxSector)
4915 return sprintf(page, "max\n");
4917 return sprintf(page, "%llu\n",
4918 (unsigned long long)mddev->resync_max);
4921 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4924 spin_lock(&mddev->lock);
4925 if (strncmp(buf, "max", 3) == 0)
4926 mddev->resync_max = MaxSector;
4928 unsigned long long max;
4932 if (kstrtoull(buf, 10, &max))
4934 if (max < mddev->resync_min)
4938 if (max < mddev->resync_max &&
4940 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4943 /* Must be a multiple of chunk_size */
4944 chunk = mddev->chunk_sectors;
4946 sector_t temp = max;
4949 if (sector_div(temp, chunk))
4952 mddev->resync_max = max;
4954 wake_up(&mddev->recovery_wait);
4957 spin_unlock(&mddev->lock);
4961 static struct md_sysfs_entry md_max_sync =
4962 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4965 suspend_lo_show(struct mddev *mddev, char *page)
4967 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4971 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4973 unsigned long long new;
4976 err = kstrtoull(buf, 10, &new);
4979 if (new != (sector_t)new)
4982 err = mddev_lock(mddev);
4986 if (mddev->pers == NULL ||
4987 mddev->pers->quiesce == NULL)
4989 mddev_suspend(mddev);
4990 mddev->suspend_lo = new;
4991 mddev_resume(mddev);
4995 mddev_unlock(mddev);
4998 static struct md_sysfs_entry md_suspend_lo =
4999 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5002 suspend_hi_show(struct mddev *mddev, char *page)
5004 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5008 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5010 unsigned long long new;
5013 err = kstrtoull(buf, 10, &new);
5016 if (new != (sector_t)new)
5019 err = mddev_lock(mddev);
5023 if (mddev->pers == NULL)
5026 mddev_suspend(mddev);
5027 mddev->suspend_hi = new;
5028 mddev_resume(mddev);
5032 mddev_unlock(mddev);
5035 static struct md_sysfs_entry md_suspend_hi =
5036 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5039 reshape_position_show(struct mddev *mddev, char *page)
5041 if (mddev->reshape_position != MaxSector)
5042 return sprintf(page, "%llu\n",
5043 (unsigned long long)mddev->reshape_position);
5044 strcpy(page, "none\n");
5049 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5051 struct md_rdev *rdev;
5052 unsigned long long new;
5055 err = kstrtoull(buf, 10, &new);
5058 if (new != (sector_t)new)
5060 err = mddev_lock(mddev);
5066 mddev->reshape_position = new;
5067 mddev->delta_disks = 0;
5068 mddev->reshape_backwards = 0;
5069 mddev->new_level = mddev->level;
5070 mddev->new_layout = mddev->layout;
5071 mddev->new_chunk_sectors = mddev->chunk_sectors;
5072 rdev_for_each(rdev, mddev)
5073 rdev->new_data_offset = rdev->data_offset;
5076 mddev_unlock(mddev);
5080 static struct md_sysfs_entry md_reshape_position =
5081 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5082 reshape_position_store);
5085 reshape_direction_show(struct mddev *mddev, char *page)
5087 return sprintf(page, "%s\n",
5088 mddev->reshape_backwards ? "backwards" : "forwards");
5092 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5097 if (cmd_match(buf, "forwards"))
5099 else if (cmd_match(buf, "backwards"))
5103 if (mddev->reshape_backwards == backwards)
5106 err = mddev_lock(mddev);
5109 /* check if we are allowed to change */
5110 if (mddev->delta_disks)
5112 else if (mddev->persistent &&
5113 mddev->major_version == 0)
5116 mddev->reshape_backwards = backwards;
5117 mddev_unlock(mddev);
5121 static struct md_sysfs_entry md_reshape_direction =
5122 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5123 reshape_direction_store);
5126 array_size_show(struct mddev *mddev, char *page)
5128 if (mddev->external_size)
5129 return sprintf(page, "%llu\n",
5130 (unsigned long long)mddev->array_sectors/2);
5132 return sprintf(page, "default\n");
5136 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5141 err = mddev_lock(mddev);
5145 /* cluster raid doesn't support change array_sectors */
5146 if (mddev_is_clustered(mddev)) {
5147 mddev_unlock(mddev);
5151 if (strncmp(buf, "default", 7) == 0) {
5153 sectors = mddev->pers->size(mddev, 0, 0);
5155 sectors = mddev->array_sectors;
5157 mddev->external_size = 0;
5159 if (strict_blocks_to_sectors(buf, §ors) < 0)
5161 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5164 mddev->external_size = 1;
5168 mddev->array_sectors = sectors;
5170 set_capacity(mddev->gendisk, mddev->array_sectors);
5171 revalidate_disk(mddev->gendisk);
5174 mddev_unlock(mddev);
5178 static struct md_sysfs_entry md_array_size =
5179 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5183 consistency_policy_show(struct mddev *mddev, char *page)
5187 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5188 ret = sprintf(page, "journal\n");
5189 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5190 ret = sprintf(page, "ppl\n");
5191 } else if (mddev->bitmap) {
5192 ret = sprintf(page, "bitmap\n");
5193 } else if (mddev->pers) {
5194 if (mddev->pers->sync_request)
5195 ret = sprintf(page, "resync\n");
5197 ret = sprintf(page, "none\n");
5199 ret = sprintf(page, "unknown\n");
5206 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5211 if (mddev->pers->change_consistency_policy)
5212 err = mddev->pers->change_consistency_policy(mddev, buf);
5215 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5216 set_bit(MD_HAS_PPL, &mddev->flags);
5221 return err ? err : len;
5224 static struct md_sysfs_entry md_consistency_policy =
5225 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5226 consistency_policy_store);
5228 static struct attribute *md_default_attrs[] = {
5231 &md_raid_disks.attr,
5232 &md_chunk_size.attr,
5234 &md_resync_start.attr,
5236 &md_new_device.attr,
5237 &md_safe_delay.attr,
5238 &md_array_state.attr,
5239 &md_reshape_position.attr,
5240 &md_reshape_direction.attr,
5241 &md_array_size.attr,
5242 &max_corr_read_errors.attr,
5243 &md_consistency_policy.attr,
5247 static struct attribute *md_redundancy_attrs[] = {
5249 &md_last_scan_mode.attr,
5250 &md_mismatches.attr,
5253 &md_sync_speed.attr,
5254 &md_sync_force_parallel.attr,
5255 &md_sync_completed.attr,
5258 &md_suspend_lo.attr,
5259 &md_suspend_hi.attr,
5264 static struct attribute_group md_redundancy_group = {
5266 .attrs = md_redundancy_attrs,
5270 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5272 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5273 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5278 spin_lock(&all_mddevs_lock);
5279 if (list_empty(&mddev->all_mddevs)) {
5280 spin_unlock(&all_mddevs_lock);
5284 spin_unlock(&all_mddevs_lock);
5286 rv = entry->show(mddev, page);
5292 md_attr_store(struct kobject *kobj, struct attribute *attr,
5293 const char *page, size_t length)
5295 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5296 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5301 if (!capable(CAP_SYS_ADMIN))
5303 spin_lock(&all_mddevs_lock);
5304 if (list_empty(&mddev->all_mddevs)) {
5305 spin_unlock(&all_mddevs_lock);
5309 spin_unlock(&all_mddevs_lock);
5310 rv = entry->store(mddev, page, length);
5315 static void md_free(struct kobject *ko)
5317 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5319 if (mddev->sysfs_state)
5320 sysfs_put(mddev->sysfs_state);
5323 del_gendisk(mddev->gendisk);
5325 blk_cleanup_queue(mddev->queue);
5327 put_disk(mddev->gendisk);
5328 percpu_ref_exit(&mddev->writes_pending);
5330 bioset_exit(&mddev->bio_set);
5331 bioset_exit(&mddev->sync_set);
5335 static const struct sysfs_ops md_sysfs_ops = {
5336 .show = md_attr_show,
5337 .store = md_attr_store,
5339 static struct kobj_type md_ktype = {
5341 .sysfs_ops = &md_sysfs_ops,
5342 .default_attrs = md_default_attrs,
5347 static void mddev_delayed_delete(struct work_struct *ws)
5349 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5351 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5352 kobject_del(&mddev->kobj);
5353 kobject_put(&mddev->kobj);
5356 static void no_op(struct percpu_ref *r) {}
5358 int mddev_init_writes_pending(struct mddev *mddev)
5360 if (mddev->writes_pending.percpu_count_ptr)
5362 if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0)
5364 /* We want to start with the refcount at zero */
5365 percpu_ref_put(&mddev->writes_pending);
5368 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5370 static int md_alloc(dev_t dev, char *name)
5373 * If dev is zero, name is the name of a device to allocate with
5374 * an arbitrary minor number. It will be "md_???"
5375 * If dev is non-zero it must be a device number with a MAJOR of
5376 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5377 * the device is being created by opening a node in /dev.
5378 * If "name" is not NULL, the device is being created by
5379 * writing to /sys/module/md_mod/parameters/new_array.
5381 static DEFINE_MUTEX(disks_mutex);
5382 struct mddev *mddev = mddev_find_or_alloc(dev);
5383 struct gendisk *disk;
5392 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5393 shift = partitioned ? MdpMinorShift : 0;
5394 unit = MINOR(mddev->unit) >> shift;
5396 /* wait for any previous instance of this device to be
5397 * completely removed (mddev_delayed_delete).
5399 flush_workqueue(md_misc_wq);
5401 mutex_lock(&disks_mutex);
5407 /* Need to ensure that 'name' is not a duplicate.
5409 struct mddev *mddev2;
5410 spin_lock(&all_mddevs_lock);
5412 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5413 if (mddev2->gendisk &&
5414 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5415 spin_unlock(&all_mddevs_lock);
5418 spin_unlock(&all_mddevs_lock);
5422 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5424 mddev->hold_active = UNTIL_STOP;
5427 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5430 mddev->queue->queuedata = mddev;
5432 blk_queue_make_request(mddev->queue, md_make_request);
5433 blk_set_stacking_limits(&mddev->queue->limits);
5435 disk = alloc_disk(1 << shift);
5437 blk_cleanup_queue(mddev->queue);
5438 mddev->queue = NULL;
5441 disk->major = MAJOR(mddev->unit);
5442 disk->first_minor = unit << shift;
5444 strcpy(disk->disk_name, name);
5445 else if (partitioned)
5446 sprintf(disk->disk_name, "md_d%d", unit);
5448 sprintf(disk->disk_name, "md%d", unit);
5449 disk->fops = &md_fops;
5450 disk->private_data = mddev;
5451 disk->queue = mddev->queue;
5452 blk_queue_write_cache(mddev->queue, true, true);
5453 /* Allow extended partitions. This makes the
5454 * 'mdp' device redundant, but we can't really
5457 disk->flags |= GENHD_FL_EXT_DEVT;
5458 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5459 mddev->gendisk = disk;
5462 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5464 /* This isn't possible, but as kobject_init_and_add is marked
5465 * __must_check, we must do something with the result
5467 pr_debug("md: cannot register %s/md - name in use\n",
5471 if (mddev->kobj.sd &&
5472 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5473 pr_debug("pointless warning\n");
5475 mutex_unlock(&disks_mutex);
5476 if (!error && mddev->kobj.sd) {
5477 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5478 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5484 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5487 md_alloc(dev, NULL);
5491 static int add_named_array(const char *val, const struct kernel_param *kp)
5494 * val must be "md_*" or "mdNNN".
5495 * For "md_*" we allocate an array with a large free minor number, and
5496 * set the name to val. val must not already be an active name.
5497 * For "mdNNN" we allocate an array with the minor number NNN
5498 * which must not already be in use.
5500 int len = strlen(val);
5501 char buf[DISK_NAME_LEN];
5502 unsigned long devnum;
5504 while (len && val[len-1] == '\n')
5506 if (len >= DISK_NAME_LEN)
5508 strlcpy(buf, val, len+1);
5509 if (strncmp(buf, "md_", 3) == 0)
5510 return md_alloc(0, buf);
5511 if (strncmp(buf, "md", 2) == 0 &&
5513 kstrtoul(buf+2, 10, &devnum) == 0 &&
5514 devnum <= MINORMASK)
5515 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5520 static void md_safemode_timeout(struct timer_list *t)
5522 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5524 mddev->safemode = 1;
5525 if (mddev->external)
5526 sysfs_notify_dirent_safe(mddev->sysfs_state);
5528 md_wakeup_thread(mddev->thread);
5531 static int start_dirty_degraded;
5533 int md_run(struct mddev *mddev)
5536 struct md_rdev *rdev;
5537 struct md_personality *pers;
5539 if (list_empty(&mddev->disks))
5540 /* cannot run an array with no devices.. */
5545 /* Cannot run until previous stop completes properly */
5546 if (mddev->sysfs_active)
5550 * Analyze all RAID superblock(s)
5552 if (!mddev->raid_disks) {
5553 if (!mddev->persistent)
5558 if (mddev->level != LEVEL_NONE)
5559 request_module("md-level-%d", mddev->level);
5560 else if (mddev->clevel[0])
5561 request_module("md-%s", mddev->clevel);
5564 * Drop all container device buffers, from now on
5565 * the only valid external interface is through the md
5568 mddev->has_superblocks = false;
5569 rdev_for_each(rdev, mddev) {
5570 if (test_bit(Faulty, &rdev->flags))
5572 sync_blockdev(rdev->bdev);
5573 invalidate_bdev(rdev->bdev);
5574 if (mddev->ro != 1 &&
5575 (bdev_read_only(rdev->bdev) ||
5576 bdev_read_only(rdev->meta_bdev))) {
5579 set_disk_ro(mddev->gendisk, 1);
5583 mddev->has_superblocks = true;
5585 /* perform some consistency tests on the device.
5586 * We don't want the data to overlap the metadata,
5587 * Internal Bitmap issues have been handled elsewhere.
5589 if (rdev->meta_bdev) {
5590 /* Nothing to check */;
5591 } else if (rdev->data_offset < rdev->sb_start) {
5592 if (mddev->dev_sectors &&
5593 rdev->data_offset + mddev->dev_sectors
5595 pr_warn("md: %s: data overlaps metadata\n",
5600 if (rdev->sb_start + rdev->sb_size/512
5601 > rdev->data_offset) {
5602 pr_warn("md: %s: metadata overlaps data\n",
5607 sysfs_notify_dirent_safe(rdev->sysfs_state);
5610 if (!bioset_initialized(&mddev->bio_set)) {
5611 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5615 if (!bioset_initialized(&mddev->sync_set)) {
5616 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5621 spin_lock(&pers_lock);
5622 pers = find_pers(mddev->level, mddev->clevel);
5623 if (!pers || !try_module_get(pers->owner)) {
5624 spin_unlock(&pers_lock);
5625 if (mddev->level != LEVEL_NONE)
5626 pr_warn("md: personality for level %d is not loaded!\n",
5629 pr_warn("md: personality for level %s is not loaded!\n",
5634 spin_unlock(&pers_lock);
5635 if (mddev->level != pers->level) {
5636 mddev->level = pers->level;
5637 mddev->new_level = pers->level;
5639 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5641 if (mddev->reshape_position != MaxSector &&
5642 pers->start_reshape == NULL) {
5643 /* This personality cannot handle reshaping... */
5644 module_put(pers->owner);
5649 if (pers->sync_request) {
5650 /* Warn if this is a potentially silly
5653 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5654 struct md_rdev *rdev2;
5657 rdev_for_each(rdev, mddev)
5658 rdev_for_each(rdev2, mddev) {
5660 rdev->bdev->bd_contains ==
5661 rdev2->bdev->bd_contains) {
5662 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5664 bdevname(rdev->bdev,b),
5665 bdevname(rdev2->bdev,b2));
5671 pr_warn("True protection against single-disk failure might be compromised.\n");
5674 mddev->recovery = 0;
5675 /* may be over-ridden by personality */
5676 mddev->resync_max_sectors = mddev->dev_sectors;
5678 mddev->ok_start_degraded = start_dirty_degraded;
5680 if (start_readonly && mddev->ro == 0)
5681 mddev->ro = 2; /* read-only, but switch on first write */
5683 err = pers->run(mddev);
5685 pr_warn("md: pers->run() failed ...\n");
5686 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5687 WARN_ONCE(!mddev->external_size,
5688 "%s: default size too small, but 'external_size' not in effect?\n",
5690 pr_warn("md: invalid array_size %llu > default size %llu\n",
5691 (unsigned long long)mddev->array_sectors / 2,
5692 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5695 if (err == 0 && pers->sync_request &&
5696 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5697 struct bitmap *bitmap;
5699 bitmap = md_bitmap_create(mddev, -1);
5700 if (IS_ERR(bitmap)) {
5701 err = PTR_ERR(bitmap);
5702 pr_warn("%s: failed to create bitmap (%d)\n",
5703 mdname(mddev), err);
5705 mddev->bitmap = bitmap;
5709 mddev_detach(mddev);
5711 pers->free(mddev, mddev->private);
5712 mddev->private = NULL;
5713 module_put(pers->owner);
5714 md_bitmap_destroy(mddev);
5720 rdev_for_each(rdev, mddev) {
5721 if (rdev->raid_disk >= 0 &&
5722 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5727 if (mddev->degraded)
5730 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5732 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5733 mddev->queue->backing_dev_info->congested_data = mddev;
5734 mddev->queue->backing_dev_info->congested_fn = md_congested;
5736 if (pers->sync_request) {
5737 if (mddev->kobj.sd &&
5738 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5739 pr_warn("md: cannot register extra attributes for %s\n",
5741 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5742 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5745 atomic_set(&mddev->max_corr_read_errors,
5746 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5747 mddev->safemode = 0;
5748 if (mddev_is_clustered(mddev))
5749 mddev->safemode_delay = 0;
5751 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5754 spin_lock(&mddev->lock);
5756 spin_unlock(&mddev->lock);
5757 rdev_for_each(rdev, mddev)
5758 if (rdev->raid_disk >= 0)
5759 if (sysfs_link_rdev(mddev, rdev))
5760 /* failure here is OK */;
5762 if (mddev->degraded && !mddev->ro)
5763 /* This ensures that recovering status is reported immediately
5764 * via sysfs - until a lack of spares is confirmed.
5766 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5767 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5769 if (mddev->sb_flags)
5770 md_update_sb(mddev, 0);
5772 md_new_event(mddev);
5776 bioset_exit(&mddev->bio_set);
5777 bioset_exit(&mddev->sync_set);
5780 EXPORT_SYMBOL_GPL(md_run);
5782 static int do_md_run(struct mddev *mddev)
5786 set_bit(MD_NOT_READY, &mddev->flags);
5787 err = md_run(mddev);
5790 err = md_bitmap_load(mddev);
5792 md_bitmap_destroy(mddev);
5796 if (mddev_is_clustered(mddev))
5797 md_allow_write(mddev);
5799 /* run start up tasks that require md_thread */
5802 md_wakeup_thread(mddev->thread);
5803 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5805 set_capacity(mddev->gendisk, mddev->array_sectors);
5806 revalidate_disk(mddev->gendisk);
5807 clear_bit(MD_NOT_READY, &mddev->flags);
5809 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5810 sysfs_notify_dirent_safe(mddev->sysfs_state);
5811 sysfs_notify_dirent_safe(mddev->sysfs_action);
5812 sysfs_notify(&mddev->kobj, NULL, "degraded");
5814 clear_bit(MD_NOT_READY, &mddev->flags);
5818 int md_start(struct mddev *mddev)
5822 if (mddev->pers->start) {
5823 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5824 md_wakeup_thread(mddev->thread);
5825 ret = mddev->pers->start(mddev);
5826 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5827 md_wakeup_thread(mddev->sync_thread);
5831 EXPORT_SYMBOL_GPL(md_start);
5833 static int restart_array(struct mddev *mddev)
5835 struct gendisk *disk = mddev->gendisk;
5836 struct md_rdev *rdev;
5837 bool has_journal = false;
5838 bool has_readonly = false;
5840 /* Complain if it has no devices */
5841 if (list_empty(&mddev->disks))
5849 rdev_for_each_rcu(rdev, mddev) {
5850 if (test_bit(Journal, &rdev->flags) &&
5851 !test_bit(Faulty, &rdev->flags))
5853 if (bdev_read_only(rdev->bdev))
5854 has_readonly = true;
5857 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5858 /* Don't restart rw with journal missing/faulty */
5863 mddev->safemode = 0;
5865 set_disk_ro(disk, 0);
5866 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5867 /* Kick recovery or resync if necessary */
5868 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5869 md_wakeup_thread(mddev->thread);
5870 md_wakeup_thread(mddev->sync_thread);
5871 sysfs_notify_dirent_safe(mddev->sysfs_state);
5875 static void md_clean(struct mddev *mddev)
5877 mddev->array_sectors = 0;
5878 mddev->external_size = 0;
5879 mddev->dev_sectors = 0;
5880 mddev->raid_disks = 0;
5881 mddev->recovery_cp = 0;
5882 mddev->resync_min = 0;
5883 mddev->resync_max = MaxSector;
5884 mddev->reshape_position = MaxSector;
5885 mddev->external = 0;
5886 mddev->persistent = 0;
5887 mddev->level = LEVEL_NONE;
5888 mddev->clevel[0] = 0;
5890 * Don't clear MD_CLOSING, or mddev can be opened again.
5891 * 'hold_active != 0' means mddev is still in the creation
5892 * process and will be used later.
5894 if (mddev->hold_active)
5897 mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
5898 mddev->sb_flags = 0;
5900 mddev->metadata_type[0] = 0;
5901 mddev->chunk_sectors = 0;
5902 mddev->ctime = mddev->utime = 0;
5904 mddev->max_disks = 0;
5906 mddev->can_decrease_events = 0;
5907 mddev->delta_disks = 0;
5908 mddev->reshape_backwards = 0;
5909 mddev->new_level = LEVEL_NONE;
5910 mddev->new_layout = 0;
5911 mddev->new_chunk_sectors = 0;
5912 mddev->curr_resync = 0;
5913 atomic64_set(&mddev->resync_mismatches, 0);
5914 mddev->suspend_lo = mddev->suspend_hi = 0;
5915 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5916 mddev->recovery = 0;
5919 mddev->degraded = 0;
5920 mddev->safemode = 0;
5921 mddev->private = NULL;
5922 mddev->cluster_info = NULL;
5923 mddev->bitmap_info.offset = 0;
5924 mddev->bitmap_info.default_offset = 0;
5925 mddev->bitmap_info.default_space = 0;
5926 mddev->bitmap_info.chunksize = 0;
5927 mddev->bitmap_info.daemon_sleep = 0;
5928 mddev->bitmap_info.max_write_behind = 0;
5929 mddev->bitmap_info.nodes = 0;
5932 static void __md_stop_writes(struct mddev *mddev)
5934 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5935 flush_workqueue(md_misc_wq);
5936 if (mddev->sync_thread) {
5937 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5938 md_reap_sync_thread(mddev);
5941 del_timer_sync(&mddev->safemode_timer);
5943 if (mddev->pers && mddev->pers->quiesce) {
5944 mddev->pers->quiesce(mddev, 1);
5945 mddev->pers->quiesce(mddev, 0);
5947 md_bitmap_flush(mddev);
5949 if (mddev->ro == 0 &&
5950 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5952 /* mark array as shutdown cleanly */
5953 if (!mddev_is_clustered(mddev))
5955 md_update_sb(mddev, 1);
5959 void md_stop_writes(struct mddev *mddev)
5961 mddev_lock_nointr(mddev);
5962 __md_stop_writes(mddev);
5963 mddev_unlock(mddev);
5965 EXPORT_SYMBOL_GPL(md_stop_writes);
5967 static void mddev_detach(struct mddev *mddev)
5969 md_bitmap_wait_behind_writes(mddev);
5970 if (mddev->pers && mddev->pers->quiesce && !mddev->suspended) {
5971 mddev->pers->quiesce(mddev, 1);
5972 mddev->pers->quiesce(mddev, 0);
5974 md_unregister_thread(&mddev->thread);
5976 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5979 static void __md_stop(struct mddev *mddev)
5981 struct md_personality *pers = mddev->pers;
5982 md_bitmap_destroy(mddev);
5983 mddev_detach(mddev);
5984 /* Ensure ->event_work is done */
5985 flush_workqueue(md_misc_wq);
5986 spin_lock(&mddev->lock);
5988 spin_unlock(&mddev->lock);
5989 pers->free(mddev, mddev->private);
5990 mddev->private = NULL;
5991 if (pers->sync_request && mddev->to_remove == NULL)
5992 mddev->to_remove = &md_redundancy_group;
5993 module_put(pers->owner);
5994 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5997 void md_stop(struct mddev *mddev)
5999 /* stop the array and free an attached data structures.
6000 * This is called from dm-raid
6002 __md_stop_writes(mddev);
6004 bioset_exit(&mddev->bio_set);
6005 bioset_exit(&mddev->sync_set);
6008 EXPORT_SYMBOL_GPL(md_stop);
6010 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6015 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6017 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6018 md_wakeup_thread(mddev->thread);
6020 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6021 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6022 if (mddev->sync_thread)
6023 /* Thread might be blocked waiting for metadata update
6024 * which will now never happen */
6025 wake_up_process(mddev->sync_thread->tsk);
6027 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6029 mddev_unlock(mddev);
6030 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6032 wait_event(mddev->sb_wait,
6033 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6034 mddev_lock_nointr(mddev);
6036 mutex_lock(&mddev->open_mutex);
6037 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6038 mddev->sync_thread ||
6039 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6040 pr_warn("md: %s still in use.\n",mdname(mddev));
6042 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6043 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6044 md_wakeup_thread(mddev->thread);
6050 __md_stop_writes(mddev);
6056 set_disk_ro(mddev->gendisk, 1);
6057 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6058 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6059 md_wakeup_thread(mddev->thread);
6060 sysfs_notify_dirent_safe(mddev->sysfs_state);
6064 mutex_unlock(&mddev->open_mutex);
6069 * 0 - completely stop and dis-assemble array
6070 * 2 - stop but do not disassemble array
6072 static int do_md_stop(struct mddev *mddev, int mode,
6073 struct block_device *bdev)
6075 struct gendisk *disk = mddev->gendisk;
6076 struct md_rdev *rdev;
6079 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6081 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6082 md_wakeup_thread(mddev->thread);
6084 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6085 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6086 if (mddev->sync_thread)
6087 /* Thread might be blocked waiting for metadata update
6088 * which will now never happen */
6089 wake_up_process(mddev->sync_thread->tsk);
6091 mddev_unlock(mddev);
6092 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6093 !test_bit(MD_RECOVERY_RUNNING,
6094 &mddev->recovery)));
6095 mddev_lock_nointr(mddev);
6097 mutex_lock(&mddev->open_mutex);
6098 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6099 mddev->sysfs_active ||
6100 mddev->sync_thread ||
6101 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6102 pr_warn("md: %s still in use.\n",mdname(mddev));
6103 mutex_unlock(&mddev->open_mutex);
6105 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6106 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6107 md_wakeup_thread(mddev->thread);
6113 set_disk_ro(disk, 0);
6115 __md_stop_writes(mddev);
6117 mddev->queue->backing_dev_info->congested_fn = NULL;
6119 /* tell userspace to handle 'inactive' */
6120 sysfs_notify_dirent_safe(mddev->sysfs_state);
6122 rdev_for_each(rdev, mddev)
6123 if (rdev->raid_disk >= 0)
6124 sysfs_unlink_rdev(mddev, rdev);
6126 set_capacity(disk, 0);
6127 mutex_unlock(&mddev->open_mutex);
6129 revalidate_disk(disk);
6134 mutex_unlock(&mddev->open_mutex);
6136 * Free resources if final stop
6139 pr_info("md: %s stopped.\n", mdname(mddev));
6141 if (mddev->bitmap_info.file) {
6142 struct file *f = mddev->bitmap_info.file;
6143 spin_lock(&mddev->lock);
6144 mddev->bitmap_info.file = NULL;
6145 spin_unlock(&mddev->lock);
6148 mddev->bitmap_info.offset = 0;
6150 export_array(mddev);
6153 if (mddev->hold_active == UNTIL_STOP)
6154 mddev->hold_active = 0;
6156 md_new_event(mddev);
6157 sysfs_notify_dirent_safe(mddev->sysfs_state);
6162 static void autorun_array(struct mddev *mddev)
6164 struct md_rdev *rdev;
6167 if (list_empty(&mddev->disks))
6170 pr_info("md: running: ");
6172 rdev_for_each(rdev, mddev) {
6173 char b[BDEVNAME_SIZE];
6174 pr_cont("<%s>", bdevname(rdev->bdev,b));
6178 err = do_md_run(mddev);
6180 pr_warn("md: do_md_run() returned %d\n", err);
6181 do_md_stop(mddev, 0, NULL);
6186 * lets try to run arrays based on all disks that have arrived
6187 * until now. (those are in pending_raid_disks)
6189 * the method: pick the first pending disk, collect all disks with
6190 * the same UUID, remove all from the pending list and put them into
6191 * the 'same_array' list. Then order this list based on superblock
6192 * update time (freshest comes first), kick out 'old' disks and
6193 * compare superblocks. If everything's fine then run it.
6195 * If "unit" is allocated, then bump its reference count
6197 static void autorun_devices(int part)
6199 struct md_rdev *rdev0, *rdev, *tmp;
6200 struct mddev *mddev;
6201 char b[BDEVNAME_SIZE];
6203 pr_info("md: autorun ...\n");
6204 while (!list_empty(&pending_raid_disks)) {
6207 LIST_HEAD(candidates);
6208 rdev0 = list_entry(pending_raid_disks.next,
6209 struct md_rdev, same_set);
6211 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6212 INIT_LIST_HEAD(&candidates);
6213 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6214 if (super_90_load(rdev, rdev0, 0) >= 0) {
6215 pr_debug("md: adding %s ...\n",
6216 bdevname(rdev->bdev,b));
6217 list_move(&rdev->same_set, &candidates);
6220 * now we have a set of devices, with all of them having
6221 * mostly sane superblocks. It's time to allocate the
6225 dev = MKDEV(mdp_major,
6226 rdev0->preferred_minor << MdpMinorShift);
6227 unit = MINOR(dev) >> MdpMinorShift;
6229 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6232 if (rdev0->preferred_minor != unit) {
6233 pr_warn("md: unit number in %s is bad: %d\n",
6234 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6238 md_probe(dev, NULL, NULL);
6239 mddev = mddev_find(dev);
6243 if (mddev_lock(mddev))
6244 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6245 else if (mddev->raid_disks || mddev->major_version
6246 || !list_empty(&mddev->disks)) {
6247 pr_warn("md: %s already running, cannot run %s\n",
6248 mdname(mddev), bdevname(rdev0->bdev,b));
6249 mddev_unlock(mddev);
6251 pr_debug("md: created %s\n", mdname(mddev));
6252 mddev->persistent = 1;
6253 rdev_for_each_list(rdev, tmp, &candidates) {
6254 list_del_init(&rdev->same_set);
6255 if (bind_rdev_to_array(rdev, mddev))
6258 autorun_array(mddev);
6259 mddev_unlock(mddev);
6261 /* on success, candidates will be empty, on error
6264 rdev_for_each_list(rdev, tmp, &candidates) {
6265 list_del_init(&rdev->same_set);
6270 pr_info("md: ... autorun DONE.\n");
6272 #endif /* !MODULE */
6274 static int get_version(void __user *arg)
6278 ver.major = MD_MAJOR_VERSION;
6279 ver.minor = MD_MINOR_VERSION;
6280 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6282 if (copy_to_user(arg, &ver, sizeof(ver)))
6288 static int get_array_info(struct mddev *mddev, void __user *arg)
6290 mdu_array_info_t info;
6291 int nr,working,insync,failed,spare;
6292 struct md_rdev *rdev;
6294 nr = working = insync = failed = spare = 0;
6296 rdev_for_each_rcu(rdev, mddev) {
6298 if (test_bit(Faulty, &rdev->flags))
6302 if (test_bit(In_sync, &rdev->flags))
6304 else if (test_bit(Journal, &rdev->flags))
6305 /* TODO: add journal count to md_u.h */
6313 info.major_version = mddev->major_version;
6314 info.minor_version = mddev->minor_version;
6315 info.patch_version = MD_PATCHLEVEL_VERSION;
6316 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6317 info.level = mddev->level;
6318 info.size = mddev->dev_sectors / 2;
6319 if (info.size != mddev->dev_sectors / 2) /* overflow */
6322 info.raid_disks = mddev->raid_disks;
6323 info.md_minor = mddev->md_minor;
6324 info.not_persistent= !mddev->persistent;
6326 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6329 info.state = (1<<MD_SB_CLEAN);
6330 if (mddev->bitmap && mddev->bitmap_info.offset)
6331 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6332 if (mddev_is_clustered(mddev))
6333 info.state |= (1<<MD_SB_CLUSTERED);
6334 info.active_disks = insync;
6335 info.working_disks = working;
6336 info.failed_disks = failed;
6337 info.spare_disks = spare;
6339 info.layout = mddev->layout;
6340 info.chunk_size = mddev->chunk_sectors << 9;
6342 if (copy_to_user(arg, &info, sizeof(info)))
6348 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6350 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6354 file = kzalloc(sizeof(*file), GFP_NOIO);
6359 spin_lock(&mddev->lock);
6360 /* bitmap enabled */
6361 if (mddev->bitmap_info.file) {
6362 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6363 sizeof(file->pathname));
6367 memmove(file->pathname, ptr,
6368 sizeof(file->pathname)-(ptr-file->pathname));
6370 spin_unlock(&mddev->lock);
6373 copy_to_user(arg, file, sizeof(*file)))
6380 static int get_disk_info(struct mddev *mddev, void __user * arg)
6382 mdu_disk_info_t info;
6383 struct md_rdev *rdev;
6385 if (copy_from_user(&info, arg, sizeof(info)))
6389 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6391 info.major = MAJOR(rdev->bdev->bd_dev);
6392 info.minor = MINOR(rdev->bdev->bd_dev);
6393 info.raid_disk = rdev->raid_disk;
6395 if (test_bit(Faulty, &rdev->flags))
6396 info.state |= (1<<MD_DISK_FAULTY);
6397 else if (test_bit(In_sync, &rdev->flags)) {
6398 info.state |= (1<<MD_DISK_ACTIVE);
6399 info.state |= (1<<MD_DISK_SYNC);
6401 if (test_bit(Journal, &rdev->flags))
6402 info.state |= (1<<MD_DISK_JOURNAL);
6403 if (test_bit(WriteMostly, &rdev->flags))
6404 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6405 if (test_bit(FailFast, &rdev->flags))
6406 info.state |= (1<<MD_DISK_FAILFAST);
6408 info.major = info.minor = 0;
6409 info.raid_disk = -1;
6410 info.state = (1<<MD_DISK_REMOVED);
6414 if (copy_to_user(arg, &info, sizeof(info)))
6420 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6422 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6423 struct md_rdev *rdev;
6424 dev_t dev = MKDEV(info->major,info->minor);
6426 if (mddev_is_clustered(mddev) &&
6427 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6428 pr_warn("%s: Cannot add to clustered mddev.\n",
6433 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6436 if (!mddev->raid_disks) {
6438 /* expecting a device which has a superblock */
6439 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6441 pr_warn("md: md_import_device returned %ld\n",
6443 return PTR_ERR(rdev);
6445 if (!list_empty(&mddev->disks)) {
6446 struct md_rdev *rdev0
6447 = list_entry(mddev->disks.next,
6448 struct md_rdev, same_set);
6449 err = super_types[mddev->major_version]
6450 .load_super(rdev, rdev0, mddev->minor_version);
6452 pr_warn("md: %s has different UUID to %s\n",
6453 bdevname(rdev->bdev,b),
6454 bdevname(rdev0->bdev,b2));
6459 err = bind_rdev_to_array(rdev, mddev);
6466 * add_new_disk can be used once the array is assembled
6467 * to add "hot spares". They must already have a superblock
6472 if (!mddev->pers->hot_add_disk) {
6473 pr_warn("%s: personality does not support diskops!\n",
6477 if (mddev->persistent)
6478 rdev = md_import_device(dev, mddev->major_version,
6479 mddev->minor_version);
6481 rdev = md_import_device(dev, -1, -1);
6483 pr_warn("md: md_import_device returned %ld\n",
6485 return PTR_ERR(rdev);
6487 /* set saved_raid_disk if appropriate */
6488 if (!mddev->persistent) {
6489 if (info->state & (1<<MD_DISK_SYNC) &&
6490 info->raid_disk < mddev->raid_disks) {
6491 rdev->raid_disk = info->raid_disk;
6492 set_bit(In_sync, &rdev->flags);
6493 clear_bit(Bitmap_sync, &rdev->flags);
6495 rdev->raid_disk = -1;
6496 rdev->saved_raid_disk = rdev->raid_disk;
6498 super_types[mddev->major_version].
6499 validate_super(mddev, NULL/*freshest*/, rdev);
6500 if ((info->state & (1<<MD_DISK_SYNC)) &&
6501 rdev->raid_disk != info->raid_disk) {
6502 /* This was a hot-add request, but events doesn't
6503 * match, so reject it.
6509 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6510 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6511 set_bit(WriteMostly, &rdev->flags);
6513 clear_bit(WriteMostly, &rdev->flags);
6514 if (info->state & (1<<MD_DISK_FAILFAST))
6515 set_bit(FailFast, &rdev->flags);
6517 clear_bit(FailFast, &rdev->flags);
6519 if (info->state & (1<<MD_DISK_JOURNAL)) {
6520 struct md_rdev *rdev2;
6521 bool has_journal = false;
6523 /* make sure no existing journal disk */
6524 rdev_for_each(rdev2, mddev) {
6525 if (test_bit(Journal, &rdev2->flags)) {
6530 if (has_journal || mddev->bitmap) {
6534 set_bit(Journal, &rdev->flags);
6537 * check whether the device shows up in other nodes
6539 if (mddev_is_clustered(mddev)) {
6540 if (info->state & (1 << MD_DISK_CANDIDATE))
6541 set_bit(Candidate, &rdev->flags);
6542 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6543 /* --add initiated by this node */
6544 err = md_cluster_ops->add_new_disk(mddev, rdev);
6552 rdev->raid_disk = -1;
6553 err = bind_rdev_to_array(rdev, mddev);
6558 if (mddev_is_clustered(mddev)) {
6559 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6561 err = md_cluster_ops->new_disk_ack(mddev,
6564 md_kick_rdev_from_array(rdev);
6568 md_cluster_ops->add_new_disk_cancel(mddev);
6570 err = add_bound_rdev(rdev);
6574 err = add_bound_rdev(rdev);
6579 /* otherwise, add_new_disk is only allowed
6580 * for major_version==0 superblocks
6582 if (mddev->major_version != 0) {
6583 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6587 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6589 rdev = md_import_device(dev, -1, 0);
6591 pr_warn("md: error, md_import_device() returned %ld\n",
6593 return PTR_ERR(rdev);
6595 rdev->desc_nr = info->number;
6596 if (info->raid_disk < mddev->raid_disks)
6597 rdev->raid_disk = info->raid_disk;
6599 rdev->raid_disk = -1;
6601 if (rdev->raid_disk < mddev->raid_disks)
6602 if (info->state & (1<<MD_DISK_SYNC))
6603 set_bit(In_sync, &rdev->flags);
6605 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6606 set_bit(WriteMostly, &rdev->flags);
6607 if (info->state & (1<<MD_DISK_FAILFAST))
6608 set_bit(FailFast, &rdev->flags);
6610 if (!mddev->persistent) {
6611 pr_debug("md: nonpersistent superblock ...\n");
6612 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6614 rdev->sb_start = calc_dev_sboffset(rdev);
6615 rdev->sectors = rdev->sb_start;
6617 err = bind_rdev_to_array(rdev, mddev);
6627 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6629 char b[BDEVNAME_SIZE];
6630 struct md_rdev *rdev;
6635 rdev = find_rdev(mddev, dev);
6639 if (rdev->raid_disk < 0)
6642 clear_bit(Blocked, &rdev->flags);
6643 remove_and_add_spares(mddev, rdev);
6645 if (rdev->raid_disk >= 0)
6649 if (mddev_is_clustered(mddev)) {
6650 if (md_cluster_ops->remove_disk(mddev, rdev))
6654 md_kick_rdev_from_array(rdev);
6655 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6657 md_wakeup_thread(mddev->thread);
6659 md_update_sb(mddev, 1);
6660 md_new_event(mddev);
6664 pr_debug("md: cannot remove active disk %s from %s ...\n",
6665 bdevname(rdev->bdev,b), mdname(mddev));
6669 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6671 char b[BDEVNAME_SIZE];
6673 struct md_rdev *rdev;
6678 if (mddev->major_version != 0) {
6679 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6683 if (!mddev->pers->hot_add_disk) {
6684 pr_warn("%s: personality does not support diskops!\n",
6689 rdev = md_import_device(dev, -1, 0);
6691 pr_warn("md: error, md_import_device() returned %ld\n",
6696 if (mddev->persistent)
6697 rdev->sb_start = calc_dev_sboffset(rdev);
6699 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6701 rdev->sectors = rdev->sb_start;
6703 if (test_bit(Faulty, &rdev->flags)) {
6704 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6705 bdevname(rdev->bdev,b), mdname(mddev));
6710 clear_bit(In_sync, &rdev->flags);
6712 rdev->saved_raid_disk = -1;
6713 err = bind_rdev_to_array(rdev, mddev);
6718 * The rest should better be atomic, we can have disk failures
6719 * noticed in interrupt contexts ...
6722 rdev->raid_disk = -1;
6724 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6726 md_update_sb(mddev, 1);
6728 * Kick recovery, maybe this spare has to be added to the
6729 * array immediately.
6731 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6732 md_wakeup_thread(mddev->thread);
6733 md_new_event(mddev);
6741 static int set_bitmap_file(struct mddev *mddev, int fd)
6746 if (!mddev->pers->quiesce || !mddev->thread)
6748 if (mddev->recovery || mddev->sync_thread)
6750 /* we should be able to change the bitmap.. */
6754 struct inode *inode;
6757 if (mddev->bitmap || mddev->bitmap_info.file)
6758 return -EEXIST; /* cannot add when bitmap is present */
6762 pr_warn("%s: error: failed to get bitmap file\n",
6767 inode = f->f_mapping->host;
6768 if (!S_ISREG(inode->i_mode)) {
6769 pr_warn("%s: error: bitmap file must be a regular file\n",
6772 } else if (!(f->f_mode & FMODE_WRITE)) {
6773 pr_warn("%s: error: bitmap file must open for write\n",
6776 } else if (atomic_read(&inode->i_writecount) != 1) {
6777 pr_warn("%s: error: bitmap file is already in use\n",
6785 mddev->bitmap_info.file = f;
6786 mddev->bitmap_info.offset = 0; /* file overrides offset */
6787 } else if (mddev->bitmap == NULL)
6788 return -ENOENT; /* cannot remove what isn't there */
6792 struct bitmap *bitmap;
6794 bitmap = md_bitmap_create(mddev, -1);
6795 mddev_suspend(mddev);
6796 if (!IS_ERR(bitmap)) {
6797 mddev->bitmap = bitmap;
6798 err = md_bitmap_load(mddev);
6800 err = PTR_ERR(bitmap);
6802 md_bitmap_destroy(mddev);
6805 mddev_resume(mddev);
6806 } else if (fd < 0) {
6807 mddev_suspend(mddev);
6808 md_bitmap_destroy(mddev);
6809 mddev_resume(mddev);
6813 struct file *f = mddev->bitmap_info.file;
6815 spin_lock(&mddev->lock);
6816 mddev->bitmap_info.file = NULL;
6817 spin_unlock(&mddev->lock);
6826 * set_array_info is used two different ways
6827 * The original usage is when creating a new array.
6828 * In this usage, raid_disks is > 0 and it together with
6829 * level, size, not_persistent,layout,chunksize determine the
6830 * shape of the array.
6831 * This will always create an array with a type-0.90.0 superblock.
6832 * The newer usage is when assembling an array.
6833 * In this case raid_disks will be 0, and the major_version field is
6834 * use to determine which style super-blocks are to be found on the devices.
6835 * The minor and patch _version numbers are also kept incase the
6836 * super_block handler wishes to interpret them.
6838 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6841 if (info->raid_disks == 0) {
6842 /* just setting version number for superblock loading */
6843 if (info->major_version < 0 ||
6844 info->major_version >= ARRAY_SIZE(super_types) ||
6845 super_types[info->major_version].name == NULL) {
6846 /* maybe try to auto-load a module? */
6847 pr_warn("md: superblock version %d not known\n",
6848 info->major_version);
6851 mddev->major_version = info->major_version;
6852 mddev->minor_version = info->minor_version;
6853 mddev->patch_version = info->patch_version;
6854 mddev->persistent = !info->not_persistent;
6855 /* ensure mddev_put doesn't delete this now that there
6856 * is some minimal configuration.
6858 mddev->ctime = ktime_get_real_seconds();
6861 mddev->major_version = MD_MAJOR_VERSION;
6862 mddev->minor_version = MD_MINOR_VERSION;
6863 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6864 mddev->ctime = ktime_get_real_seconds();
6866 mddev->level = info->level;
6867 mddev->clevel[0] = 0;
6868 mddev->dev_sectors = 2 * (sector_t)info->size;
6869 mddev->raid_disks = info->raid_disks;
6870 /* don't set md_minor, it is determined by which /dev/md* was
6873 if (info->state & (1<<MD_SB_CLEAN))
6874 mddev->recovery_cp = MaxSector;
6876 mddev->recovery_cp = 0;
6877 mddev->persistent = ! info->not_persistent;
6878 mddev->external = 0;
6880 mddev->layout = info->layout;
6881 if (mddev->level == 0)
6882 /* Cannot trust RAID0 layout info here */
6884 mddev->chunk_sectors = info->chunk_size >> 9;
6886 if (mddev->persistent) {
6887 mddev->max_disks = MD_SB_DISKS;
6889 mddev->sb_flags = 0;
6891 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6893 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6894 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6895 mddev->bitmap_info.offset = 0;
6897 mddev->reshape_position = MaxSector;
6900 * Generate a 128 bit UUID
6902 get_random_bytes(mddev->uuid, 16);
6904 mddev->new_level = mddev->level;
6905 mddev->new_chunk_sectors = mddev->chunk_sectors;
6906 mddev->new_layout = mddev->layout;
6907 mddev->delta_disks = 0;
6908 mddev->reshape_backwards = 0;
6913 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6915 lockdep_assert_held(&mddev->reconfig_mutex);
6917 if (mddev->external_size)
6920 mddev->array_sectors = array_sectors;
6922 EXPORT_SYMBOL(md_set_array_sectors);
6924 static int update_size(struct mddev *mddev, sector_t num_sectors)
6926 struct md_rdev *rdev;
6928 int fit = (num_sectors == 0);
6929 sector_t old_dev_sectors = mddev->dev_sectors;
6931 if (mddev->pers->resize == NULL)
6933 /* The "num_sectors" is the number of sectors of each device that
6934 * is used. This can only make sense for arrays with redundancy.
6935 * linear and raid0 always use whatever space is available. We can only
6936 * consider changing this number if no resync or reconstruction is
6937 * happening, and if the new size is acceptable. It must fit before the
6938 * sb_start or, if that is <data_offset, it must fit before the size
6939 * of each device. If num_sectors is zero, we find the largest size
6942 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6948 rdev_for_each(rdev, mddev) {
6949 sector_t avail = rdev->sectors;
6951 if (fit && (num_sectors == 0 || num_sectors > avail))
6952 num_sectors = avail;
6953 if (avail < num_sectors)
6956 rv = mddev->pers->resize(mddev, num_sectors);
6958 if (mddev_is_clustered(mddev))
6959 md_cluster_ops->update_size(mddev, old_dev_sectors);
6960 else if (mddev->queue) {
6961 set_capacity(mddev->gendisk, mddev->array_sectors);
6962 revalidate_disk(mddev->gendisk);
6968 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6971 struct md_rdev *rdev;
6972 /* change the number of raid disks */
6973 if (mddev->pers->check_reshape == NULL)
6977 if (raid_disks <= 0 ||
6978 (mddev->max_disks && raid_disks >= mddev->max_disks))
6980 if (mddev->sync_thread ||
6981 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6982 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
6983 mddev->reshape_position != MaxSector)
6986 rdev_for_each(rdev, mddev) {
6987 if (mddev->raid_disks < raid_disks &&
6988 rdev->data_offset < rdev->new_data_offset)
6990 if (mddev->raid_disks > raid_disks &&
6991 rdev->data_offset > rdev->new_data_offset)
6995 mddev->delta_disks = raid_disks - mddev->raid_disks;
6996 if (mddev->delta_disks < 0)
6997 mddev->reshape_backwards = 1;
6998 else if (mddev->delta_disks > 0)
6999 mddev->reshape_backwards = 0;
7001 rv = mddev->pers->check_reshape(mddev);
7003 mddev->delta_disks = 0;
7004 mddev->reshape_backwards = 0;
7010 * update_array_info is used to change the configuration of an
7012 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7013 * fields in the info are checked against the array.
7014 * Any differences that cannot be handled will cause an error.
7015 * Normally, only one change can be managed at a time.
7017 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7023 /* calculate expected state,ignoring low bits */
7024 if (mddev->bitmap && mddev->bitmap_info.offset)
7025 state |= (1 << MD_SB_BITMAP_PRESENT);
7027 if (mddev->major_version != info->major_version ||
7028 mddev->minor_version != info->minor_version ||
7029 /* mddev->patch_version != info->patch_version || */
7030 mddev->ctime != info->ctime ||
7031 mddev->level != info->level ||
7032 /* mddev->layout != info->layout || */
7033 mddev->persistent != !info->not_persistent ||
7034 mddev->chunk_sectors != info->chunk_size >> 9 ||
7035 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7036 ((state^info->state) & 0xfffffe00)
7039 /* Check there is only one change */
7040 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7042 if (mddev->raid_disks != info->raid_disks)
7044 if (mddev->layout != info->layout)
7046 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7053 if (mddev->layout != info->layout) {
7055 * we don't need to do anything at the md level, the
7056 * personality will take care of it all.
7058 if (mddev->pers->check_reshape == NULL)
7061 mddev->new_layout = info->layout;
7062 rv = mddev->pers->check_reshape(mddev);
7064 mddev->new_layout = mddev->layout;
7068 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7069 rv = update_size(mddev, (sector_t)info->size * 2);
7071 if (mddev->raid_disks != info->raid_disks)
7072 rv = update_raid_disks(mddev, info->raid_disks);
7074 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7075 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7079 if (mddev->recovery || mddev->sync_thread) {
7083 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7084 struct bitmap *bitmap;
7085 /* add the bitmap */
7086 if (mddev->bitmap) {
7090 if (mddev->bitmap_info.default_offset == 0) {
7094 mddev->bitmap_info.offset =
7095 mddev->bitmap_info.default_offset;
7096 mddev->bitmap_info.space =
7097 mddev->bitmap_info.default_space;
7098 bitmap = md_bitmap_create(mddev, -1);
7099 mddev_suspend(mddev);
7100 if (!IS_ERR(bitmap)) {
7101 mddev->bitmap = bitmap;
7102 rv = md_bitmap_load(mddev);
7104 rv = PTR_ERR(bitmap);
7106 md_bitmap_destroy(mddev);
7107 mddev_resume(mddev);
7109 /* remove the bitmap */
7110 if (!mddev->bitmap) {
7114 if (mddev->bitmap->storage.file) {
7118 if (mddev->bitmap_info.nodes) {
7119 /* hold PW on all the bitmap lock */
7120 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7121 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7123 md_cluster_ops->unlock_all_bitmaps(mddev);
7127 mddev->bitmap_info.nodes = 0;
7128 md_cluster_ops->leave(mddev);
7130 mddev_suspend(mddev);
7131 md_bitmap_destroy(mddev);
7132 mddev_resume(mddev);
7133 mddev->bitmap_info.offset = 0;
7136 md_update_sb(mddev, 1);
7142 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7144 struct md_rdev *rdev;
7147 if (mddev->pers == NULL)
7151 rdev = md_find_rdev_rcu(mddev, dev);
7155 md_error(mddev, rdev);
7156 if (!test_bit(Faulty, &rdev->flags))
7164 * We have a problem here : there is no easy way to give a CHS
7165 * virtual geometry. We currently pretend that we have a 2 heads
7166 * 4 sectors (with a BIG number of cylinders...). This drives
7167 * dosfs just mad... ;-)
7169 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7171 struct mddev *mddev = bdev->bd_disk->private_data;
7175 geo->cylinders = mddev->array_sectors / 8;
7179 static inline bool md_ioctl_valid(unsigned int cmd)
7183 case GET_ARRAY_INFO:
7184 case GET_BITMAP_FILE:
7187 case HOT_REMOVE_DISK:
7190 case RESTART_ARRAY_RW:
7192 case SET_ARRAY_INFO:
7193 case SET_BITMAP_FILE:
7194 case SET_DISK_FAULTY:
7197 case CLUSTERED_DISK_NACK:
7204 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7205 unsigned int cmd, unsigned long arg)
7208 void __user *argp = (void __user *)arg;
7209 struct mddev *mddev = NULL;
7211 if (!md_ioctl_valid(cmd))
7216 case GET_ARRAY_INFO:
7220 if (!capable(CAP_SYS_ADMIN))
7225 * Commands dealing with the RAID driver but not any
7230 err = get_version(argp);
7236 autostart_arrays(arg);
7243 * Commands creating/starting a new array:
7246 mddev = bdev->bd_disk->private_data;
7253 /* Some actions do not requires the mutex */
7255 case GET_ARRAY_INFO:
7256 if (!mddev->raid_disks && !mddev->external)
7259 err = get_array_info(mddev, argp);
7263 if (!mddev->raid_disks && !mddev->external)
7266 err = get_disk_info(mddev, argp);
7269 case SET_DISK_FAULTY:
7270 err = set_disk_faulty(mddev, new_decode_dev(arg));
7273 case GET_BITMAP_FILE:
7274 err = get_bitmap_file(mddev, argp);
7279 if (cmd == ADD_NEW_DISK)
7280 /* need to ensure md_delayed_delete() has completed */
7281 flush_workqueue(md_misc_wq);
7283 if (cmd == HOT_REMOVE_DISK)
7284 /* need to ensure recovery thread has run */
7285 wait_event_interruptible_timeout(mddev->sb_wait,
7286 !test_bit(MD_RECOVERY_NEEDED,
7288 msecs_to_jiffies(5000));
7289 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7290 /* Need to flush page cache, and ensure no-one else opens
7293 mutex_lock(&mddev->open_mutex);
7294 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7295 mutex_unlock(&mddev->open_mutex);
7299 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
7300 mutex_unlock(&mddev->open_mutex);
7304 mutex_unlock(&mddev->open_mutex);
7305 sync_blockdev(bdev);
7307 err = mddev_lock(mddev);
7309 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7314 if (cmd == SET_ARRAY_INFO) {
7315 mdu_array_info_t info;
7317 memset(&info, 0, sizeof(info));
7318 else if (copy_from_user(&info, argp, sizeof(info))) {
7323 err = update_array_info(mddev, &info);
7325 pr_warn("md: couldn't update array info. %d\n", err);
7330 if (!list_empty(&mddev->disks)) {
7331 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7335 if (mddev->raid_disks) {
7336 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7340 err = set_array_info(mddev, &info);
7342 pr_warn("md: couldn't set array info. %d\n", err);
7349 * Commands querying/configuring an existing array:
7351 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7352 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7353 if ((!mddev->raid_disks && !mddev->external)
7354 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7355 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7356 && cmd != GET_BITMAP_FILE) {
7362 * Commands even a read-only array can execute:
7365 case RESTART_ARRAY_RW:
7366 err = restart_array(mddev);
7370 err = do_md_stop(mddev, 0, bdev);
7374 err = md_set_readonly(mddev, bdev);
7377 case HOT_REMOVE_DISK:
7378 err = hot_remove_disk(mddev, new_decode_dev(arg));
7382 /* We can support ADD_NEW_DISK on read-only arrays
7383 * only if we are re-adding a preexisting device.
7384 * So require mddev->pers and MD_DISK_SYNC.
7387 mdu_disk_info_t info;
7388 if (copy_from_user(&info, argp, sizeof(info)))
7390 else if (!(info.state & (1<<MD_DISK_SYNC)))
7391 /* Need to clear read-only for this */
7394 err = add_new_disk(mddev, &info);
7401 * The remaining ioctls are changing the state of the
7402 * superblock, so we do not allow them on read-only arrays.
7404 if (mddev->ro && mddev->pers) {
7405 if (mddev->ro == 2) {
7407 sysfs_notify_dirent_safe(mddev->sysfs_state);
7408 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7409 /* mddev_unlock will wake thread */
7410 /* If a device failed while we were read-only, we
7411 * need to make sure the metadata is updated now.
7413 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7414 mddev_unlock(mddev);
7415 wait_event(mddev->sb_wait,
7416 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7417 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7418 mddev_lock_nointr(mddev);
7429 mdu_disk_info_t info;
7430 if (copy_from_user(&info, argp, sizeof(info)))
7433 err = add_new_disk(mddev, &info);
7437 case CLUSTERED_DISK_NACK:
7438 if (mddev_is_clustered(mddev))
7439 md_cluster_ops->new_disk_ack(mddev, false);
7445 err = hot_add_disk(mddev, new_decode_dev(arg));
7449 err = do_md_run(mddev);
7452 case SET_BITMAP_FILE:
7453 err = set_bitmap_file(mddev, (int)arg);
7462 if (mddev->hold_active == UNTIL_IOCTL &&
7464 mddev->hold_active = 0;
7465 mddev_unlock(mddev);
7467 if (cmd == STOP_ARRAY_RO || (err && cmd == STOP_ARRAY))
7468 clear_bit(MD_CLOSING, &mddev->flags);
7471 #ifdef CONFIG_COMPAT
7472 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7473 unsigned int cmd, unsigned long arg)
7476 case HOT_REMOVE_DISK:
7478 case SET_DISK_FAULTY:
7479 case SET_BITMAP_FILE:
7480 /* These take in integer arg, do not convert */
7483 arg = (unsigned long)compat_ptr(arg);
7487 return md_ioctl(bdev, mode, cmd, arg);
7489 #endif /* CONFIG_COMPAT */
7491 static int md_set_read_only(struct block_device *bdev, bool ro)
7493 struct mddev *mddev = bdev->bd_disk->private_data;
7496 err = mddev_lock(mddev);
7500 if (!mddev->raid_disks && !mddev->external) {
7506 * Transitioning to read-auto need only happen for arrays that call
7507 * md_write_start and which are not ready for writes yet.
7509 if (!ro && mddev->ro == 1 && mddev->pers) {
7510 err = restart_array(mddev);
7517 mddev_unlock(mddev);
7521 static int md_open(struct block_device *bdev, fmode_t mode)
7524 * Succeed if we can lock the mddev, which confirms that
7525 * it isn't being stopped right now.
7527 struct mddev *mddev = mddev_find(bdev->bd_dev);
7533 if (mddev->gendisk != bdev->bd_disk) {
7534 /* we are racing with mddev_put which is discarding this
7538 /* Wait until bdev->bd_disk is definitely gone */
7539 if (work_pending(&mddev->del_work))
7540 flush_workqueue(md_misc_wq);
7543 BUG_ON(mddev != bdev->bd_disk->private_data);
7545 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7548 if (test_bit(MD_CLOSING, &mddev->flags)) {
7549 mutex_unlock(&mddev->open_mutex);
7555 atomic_inc(&mddev->openers);
7556 mutex_unlock(&mddev->open_mutex);
7558 check_disk_change(bdev);
7565 static void md_release(struct gendisk *disk, fmode_t mode)
7567 struct mddev *mddev = disk->private_data;
7570 atomic_dec(&mddev->openers);
7574 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7576 struct mddev *mddev = disk->private_data;
7577 unsigned int ret = 0;
7580 ret = DISK_EVENT_MEDIA_CHANGE;
7585 static const struct block_device_operations md_fops =
7587 .owner = THIS_MODULE,
7589 .release = md_release,
7591 #ifdef CONFIG_COMPAT
7592 .compat_ioctl = md_compat_ioctl,
7594 .getgeo = md_getgeo,
7595 .check_events = md_check_events,
7596 .set_read_only = md_set_read_only,
7599 static int md_thread(void *arg)
7601 struct md_thread *thread = arg;
7604 * md_thread is a 'system-thread', it's priority should be very
7605 * high. We avoid resource deadlocks individually in each
7606 * raid personality. (RAID5 does preallocation) We also use RR and
7607 * the very same RT priority as kswapd, thus we will never get
7608 * into a priority inversion deadlock.
7610 * we definitely have to have equal or higher priority than
7611 * bdflush, otherwise bdflush will deadlock if there are too
7612 * many dirty RAID5 blocks.
7615 allow_signal(SIGKILL);
7616 while (!kthread_should_stop()) {
7618 /* We need to wait INTERRUPTIBLE so that
7619 * we don't add to the load-average.
7620 * That means we need to be sure no signals are
7623 if (signal_pending(current))
7624 flush_signals(current);
7626 wait_event_interruptible_timeout
7628 test_bit(THREAD_WAKEUP, &thread->flags)
7629 || kthread_should_stop() || kthread_should_park(),
7632 clear_bit(THREAD_WAKEUP, &thread->flags);
7633 if (kthread_should_park())
7635 if (!kthread_should_stop())
7636 thread->run(thread);
7642 void md_wakeup_thread(struct md_thread *thread)
7645 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7646 set_bit(THREAD_WAKEUP, &thread->flags);
7647 wake_up(&thread->wqueue);
7650 EXPORT_SYMBOL(md_wakeup_thread);
7652 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7653 struct mddev *mddev, const char *name)
7655 struct md_thread *thread;
7657 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7661 init_waitqueue_head(&thread->wqueue);
7664 thread->mddev = mddev;
7665 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7666 thread->tsk = kthread_run(md_thread, thread,
7668 mdname(thread->mddev),
7670 if (IS_ERR(thread->tsk)) {
7676 EXPORT_SYMBOL(md_register_thread);
7678 void md_unregister_thread(struct md_thread **threadp)
7680 struct md_thread *thread;
7683 * Locking ensures that mddev_unlock does not wake_up a
7684 * non-existent thread
7686 spin_lock(&pers_lock);
7689 spin_unlock(&pers_lock);
7693 spin_unlock(&pers_lock);
7695 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7696 kthread_stop(thread->tsk);
7699 EXPORT_SYMBOL(md_unregister_thread);
7701 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7703 if (!rdev || test_bit(Faulty, &rdev->flags))
7706 if (!mddev->pers || !mddev->pers->error_handler)
7708 mddev->pers->error_handler(mddev,rdev);
7709 if (mddev->degraded)
7710 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7711 sysfs_notify_dirent_safe(rdev->sysfs_state);
7712 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7714 md_wakeup_thread(mddev->thread);
7715 if (mddev->event_work.func)
7716 queue_work(md_misc_wq, &mddev->event_work);
7717 md_new_event(mddev);
7719 EXPORT_SYMBOL(md_error);
7721 /* seq_file implementation /proc/mdstat */
7723 static void status_unused(struct seq_file *seq)
7726 struct md_rdev *rdev;
7728 seq_printf(seq, "unused devices: ");
7730 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7731 char b[BDEVNAME_SIZE];
7733 seq_printf(seq, "%s ",
7734 bdevname(rdev->bdev,b));
7737 seq_printf(seq, "<none>");
7739 seq_printf(seq, "\n");
7742 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7744 sector_t max_sectors, resync, res;
7745 unsigned long dt, db = 0;
7746 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7747 int scale, recovery_active;
7748 unsigned int per_milli;
7750 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7751 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7752 max_sectors = mddev->resync_max_sectors;
7754 max_sectors = mddev->dev_sectors;
7756 resync = mddev->curr_resync;
7758 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7759 /* Still cleaning up */
7760 resync = max_sectors;
7761 } else if (resync > max_sectors)
7762 resync = max_sectors;
7764 resync -= atomic_read(&mddev->recovery_active);
7767 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7768 struct md_rdev *rdev;
7770 rdev_for_each(rdev, mddev)
7771 if (rdev->raid_disk >= 0 &&
7772 !test_bit(Faulty, &rdev->flags) &&
7773 rdev->recovery_offset != MaxSector &&
7774 rdev->recovery_offset) {
7775 seq_printf(seq, "\trecover=REMOTE");
7778 if (mddev->reshape_position != MaxSector)
7779 seq_printf(seq, "\treshape=REMOTE");
7781 seq_printf(seq, "\tresync=REMOTE");
7784 if (mddev->recovery_cp < MaxSector) {
7785 seq_printf(seq, "\tresync=PENDING");
7791 seq_printf(seq, "\tresync=DELAYED");
7795 WARN_ON(max_sectors == 0);
7796 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7797 * in a sector_t, and (max_sectors>>scale) will fit in a
7798 * u32, as those are the requirements for sector_div.
7799 * Thus 'scale' must be at least 10
7802 if (sizeof(sector_t) > sizeof(unsigned long)) {
7803 while ( max_sectors/2 > (1ULL<<(scale+32)))
7806 res = (resync>>scale)*1000;
7807 sector_div(res, (u32)((max_sectors>>scale)+1));
7811 int i, x = per_milli/50, y = 20-x;
7812 seq_printf(seq, "[");
7813 for (i = 0; i < x; i++)
7814 seq_printf(seq, "=");
7815 seq_printf(seq, ">");
7816 for (i = 0; i < y; i++)
7817 seq_printf(seq, ".");
7818 seq_printf(seq, "] ");
7820 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7821 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7823 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7825 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7826 "resync" : "recovery"))),
7827 per_milli/10, per_milli % 10,
7828 (unsigned long long) resync/2,
7829 (unsigned long long) max_sectors/2);
7832 * dt: time from mark until now
7833 * db: blocks written from mark until now
7834 * rt: remaining time
7836 * rt is a sector_t, which is always 64bit now. We are keeping
7837 * the original algorithm, but it is not really necessary.
7839 * Original algorithm:
7840 * So we divide before multiply in case it is 32bit and close
7842 * We scale the divisor (db) by 32 to avoid losing precision
7843 * near the end of resync when the number of remaining sectors
7845 * We then divide rt by 32 after multiplying by db to compensate.
7846 * The '+1' avoids division by zero if db is very small.
7848 dt = ((jiffies - mddev->resync_mark) / HZ);
7851 curr_mark_cnt = mddev->curr_mark_cnt;
7852 recovery_active = atomic_read(&mddev->recovery_active);
7853 resync_mark_cnt = mddev->resync_mark_cnt;
7855 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
7856 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
7858 rt = max_sectors - resync; /* number of remaining sectors */
7859 rt = div64_u64(rt, db/32+1);
7863 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7864 ((unsigned long)rt % 60)/6);
7866 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7870 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7872 struct list_head *tmp;
7874 struct mddev *mddev;
7886 spin_lock(&all_mddevs_lock);
7887 list_for_each(tmp,&all_mddevs)
7889 mddev = list_entry(tmp, struct mddev, all_mddevs);
7891 spin_unlock(&all_mddevs_lock);
7894 spin_unlock(&all_mddevs_lock);
7896 return (void*)2;/* tail */
7900 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7902 struct list_head *tmp;
7903 struct mddev *next_mddev, *mddev = v;
7909 spin_lock(&all_mddevs_lock);
7911 tmp = all_mddevs.next;
7913 tmp = mddev->all_mddevs.next;
7914 if (tmp != &all_mddevs)
7915 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7917 next_mddev = (void*)2;
7920 spin_unlock(&all_mddevs_lock);
7928 static void md_seq_stop(struct seq_file *seq, void *v)
7930 struct mddev *mddev = v;
7932 if (mddev && v != (void*)1 && v != (void*)2)
7936 static int md_seq_show(struct seq_file *seq, void *v)
7938 struct mddev *mddev = v;
7940 struct md_rdev *rdev;
7942 if (v == (void*)1) {
7943 struct md_personality *pers;
7944 seq_printf(seq, "Personalities : ");
7945 spin_lock(&pers_lock);
7946 list_for_each_entry(pers, &pers_list, list)
7947 seq_printf(seq, "[%s] ", pers->name);
7949 spin_unlock(&pers_lock);
7950 seq_printf(seq, "\n");
7951 seq->poll_event = atomic_read(&md_event_count);
7954 if (v == (void*)2) {
7959 spin_lock(&mddev->lock);
7960 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7961 seq_printf(seq, "%s : %sactive", mdname(mddev),
7962 mddev->pers ? "" : "in");
7965 seq_printf(seq, " (read-only)");
7967 seq_printf(seq, " (auto-read-only)");
7968 seq_printf(seq, " %s", mddev->pers->name);
7973 rdev_for_each_rcu(rdev, mddev) {
7974 char b[BDEVNAME_SIZE];
7975 seq_printf(seq, " %s[%d]",
7976 bdevname(rdev->bdev,b), rdev->desc_nr);
7977 if (test_bit(WriteMostly, &rdev->flags))
7978 seq_printf(seq, "(W)");
7979 if (test_bit(Journal, &rdev->flags))
7980 seq_printf(seq, "(J)");
7981 if (test_bit(Faulty, &rdev->flags)) {
7982 seq_printf(seq, "(F)");
7985 if (rdev->raid_disk < 0)
7986 seq_printf(seq, "(S)"); /* spare */
7987 if (test_bit(Replacement, &rdev->flags))
7988 seq_printf(seq, "(R)");
7989 sectors += rdev->sectors;
7993 if (!list_empty(&mddev->disks)) {
7995 seq_printf(seq, "\n %llu blocks",
7996 (unsigned long long)
7997 mddev->array_sectors / 2);
7999 seq_printf(seq, "\n %llu blocks",
8000 (unsigned long long)sectors / 2);
8002 if (mddev->persistent) {
8003 if (mddev->major_version != 0 ||
8004 mddev->minor_version != 90) {
8005 seq_printf(seq," super %d.%d",
8006 mddev->major_version,
8007 mddev->minor_version);
8009 } else if (mddev->external)
8010 seq_printf(seq, " super external:%s",
8011 mddev->metadata_type);
8013 seq_printf(seq, " super non-persistent");
8016 mddev->pers->status(seq, mddev);
8017 seq_printf(seq, "\n ");
8018 if (mddev->pers->sync_request) {
8019 if (status_resync(seq, mddev))
8020 seq_printf(seq, "\n ");
8023 seq_printf(seq, "\n ");
8025 md_bitmap_status(seq, mddev->bitmap);
8027 seq_printf(seq, "\n");
8029 spin_unlock(&mddev->lock);
8034 static const struct seq_operations md_seq_ops = {
8035 .start = md_seq_start,
8036 .next = md_seq_next,
8037 .stop = md_seq_stop,
8038 .show = md_seq_show,
8041 static int md_seq_open(struct inode *inode, struct file *file)
8043 struct seq_file *seq;
8046 error = seq_open(file, &md_seq_ops);
8050 seq = file->private_data;
8051 seq->poll_event = atomic_read(&md_event_count);
8055 static int md_unloading;
8056 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8058 struct seq_file *seq = filp->private_data;
8062 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8063 poll_wait(filp, &md_event_waiters, wait);
8065 /* always allow read */
8066 mask = EPOLLIN | EPOLLRDNORM;
8068 if (seq->poll_event != atomic_read(&md_event_count))
8069 mask |= EPOLLERR | EPOLLPRI;
8073 static const struct file_operations md_seq_fops = {
8074 .owner = THIS_MODULE,
8075 .open = md_seq_open,
8077 .llseek = seq_lseek,
8078 .release = seq_release,
8079 .poll = mdstat_poll,
8082 int register_md_personality(struct md_personality *p)
8084 pr_debug("md: %s personality registered for level %d\n",
8086 spin_lock(&pers_lock);
8087 list_add_tail(&p->list, &pers_list);
8088 spin_unlock(&pers_lock);
8091 EXPORT_SYMBOL(register_md_personality);
8093 int unregister_md_personality(struct md_personality *p)
8095 pr_debug("md: %s personality unregistered\n", p->name);
8096 spin_lock(&pers_lock);
8097 list_del_init(&p->list);
8098 spin_unlock(&pers_lock);
8101 EXPORT_SYMBOL(unregister_md_personality);
8103 int register_md_cluster_operations(struct md_cluster_operations *ops,
8104 struct module *module)
8107 spin_lock(&pers_lock);
8108 if (md_cluster_ops != NULL)
8111 md_cluster_ops = ops;
8112 md_cluster_mod = module;
8114 spin_unlock(&pers_lock);
8117 EXPORT_SYMBOL(register_md_cluster_operations);
8119 int unregister_md_cluster_operations(void)
8121 spin_lock(&pers_lock);
8122 md_cluster_ops = NULL;
8123 spin_unlock(&pers_lock);
8126 EXPORT_SYMBOL(unregister_md_cluster_operations);
8128 int md_setup_cluster(struct mddev *mddev, int nodes)
8130 if (!md_cluster_ops)
8131 request_module("md-cluster");
8132 spin_lock(&pers_lock);
8133 /* ensure module won't be unloaded */
8134 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8135 pr_warn("can't find md-cluster module or get it's reference.\n");
8136 spin_unlock(&pers_lock);
8139 spin_unlock(&pers_lock);
8141 return md_cluster_ops->join(mddev, nodes);
8144 void md_cluster_stop(struct mddev *mddev)
8146 if (!md_cluster_ops)
8148 md_cluster_ops->leave(mddev);
8149 module_put(md_cluster_mod);
8152 static int is_mddev_idle(struct mddev *mddev, int init)
8154 struct md_rdev *rdev;
8160 rdev_for_each_rcu(rdev, mddev) {
8161 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8162 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8163 atomic_read(&disk->sync_io);
8164 /* sync IO will cause sync_io to increase before the disk_stats
8165 * as sync_io is counted when a request starts, and
8166 * disk_stats is counted when it completes.
8167 * So resync activity will cause curr_events to be smaller than
8168 * when there was no such activity.
8169 * non-sync IO will cause disk_stat to increase without
8170 * increasing sync_io so curr_events will (eventually)
8171 * be larger than it was before. Once it becomes
8172 * substantially larger, the test below will cause
8173 * the array to appear non-idle, and resync will slow
8175 * If there is a lot of outstanding resync activity when
8176 * we set last_event to curr_events, then all that activity
8177 * completing might cause the array to appear non-idle
8178 * and resync will be slowed down even though there might
8179 * not have been non-resync activity. This will only
8180 * happen once though. 'last_events' will soon reflect
8181 * the state where there is little or no outstanding
8182 * resync requests, and further resync activity will
8183 * always make curr_events less than last_events.
8186 if (init || curr_events - rdev->last_events > 64) {
8187 rdev->last_events = curr_events;
8195 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8197 /* another "blocks" (512byte) blocks have been synced */
8198 atomic_sub(blocks, &mddev->recovery_active);
8199 wake_up(&mddev->recovery_wait);
8201 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8202 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8203 md_wakeup_thread(mddev->thread);
8204 // stop recovery, signal do_sync ....
8207 EXPORT_SYMBOL(md_done_sync);
8209 /* md_write_start(mddev, bi)
8210 * If we need to update some array metadata (e.g. 'active' flag
8211 * in superblock) before writing, schedule a superblock update
8212 * and wait for it to complete.
8213 * A return value of 'false' means that the write wasn't recorded
8214 * and cannot proceed as the array is being suspend.
8216 bool md_write_start(struct mddev *mddev, struct bio *bi)
8220 if (bio_data_dir(bi) != WRITE)
8223 BUG_ON(mddev->ro == 1);
8224 if (mddev->ro == 2) {
8225 /* need to switch to read/write */
8227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8228 md_wakeup_thread(mddev->thread);
8229 md_wakeup_thread(mddev->sync_thread);
8233 percpu_ref_get(&mddev->writes_pending);
8234 smp_mb(); /* Match smp_mb in set_in_sync() */
8235 if (mddev->safemode == 1)
8236 mddev->safemode = 0;
8237 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8238 if (mddev->in_sync || mddev->sync_checkers) {
8239 spin_lock(&mddev->lock);
8240 if (mddev->in_sync) {
8242 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8243 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8244 md_wakeup_thread(mddev->thread);
8247 spin_unlock(&mddev->lock);
8251 sysfs_notify_dirent_safe(mddev->sysfs_state);
8252 if (!mddev->has_superblocks)
8254 wait_event(mddev->sb_wait,
8255 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8257 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8258 percpu_ref_put(&mddev->writes_pending);
8263 EXPORT_SYMBOL(md_write_start);
8265 /* md_write_inc can only be called when md_write_start() has
8266 * already been called at least once of the current request.
8267 * It increments the counter and is useful when a single request
8268 * is split into several parts. Each part causes an increment and
8269 * so needs a matching md_write_end().
8270 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8271 * a spinlocked region.
8273 void md_write_inc(struct mddev *mddev, struct bio *bi)
8275 if (bio_data_dir(bi) != WRITE)
8277 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8278 percpu_ref_get(&mddev->writes_pending);
8280 EXPORT_SYMBOL(md_write_inc);
8282 void md_write_end(struct mddev *mddev)
8284 percpu_ref_put(&mddev->writes_pending);
8286 if (mddev->safemode == 2)
8287 md_wakeup_thread(mddev->thread);
8288 else if (mddev->safemode_delay)
8289 /* The roundup() ensures this only performs locking once
8290 * every ->safemode_delay jiffies
8292 mod_timer(&mddev->safemode_timer,
8293 roundup(jiffies, mddev->safemode_delay) +
8294 mddev->safemode_delay);
8297 EXPORT_SYMBOL(md_write_end);
8299 /* md_allow_write(mddev)
8300 * Calling this ensures that the array is marked 'active' so that writes
8301 * may proceed without blocking. It is important to call this before
8302 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8303 * Must be called with mddev_lock held.
8305 void md_allow_write(struct mddev *mddev)
8311 if (!mddev->pers->sync_request)
8314 spin_lock(&mddev->lock);
8315 if (mddev->in_sync) {
8317 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8318 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8319 if (mddev->safemode_delay &&
8320 mddev->safemode == 0)
8321 mddev->safemode = 1;
8322 spin_unlock(&mddev->lock);
8323 md_update_sb(mddev, 0);
8324 sysfs_notify_dirent_safe(mddev->sysfs_state);
8325 /* wait for the dirty state to be recorded in the metadata */
8326 wait_event(mddev->sb_wait,
8327 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8329 spin_unlock(&mddev->lock);
8331 EXPORT_SYMBOL_GPL(md_allow_write);
8333 #define SYNC_MARKS 10
8334 #define SYNC_MARK_STEP (3*HZ)
8335 #define UPDATE_FREQUENCY (5*60*HZ)
8336 void md_do_sync(struct md_thread *thread)
8338 struct mddev *mddev = thread->mddev;
8339 struct mddev *mddev2;
8340 unsigned int currspeed = 0,
8342 sector_t max_sectors,j, io_sectors, recovery_done;
8343 unsigned long mark[SYNC_MARKS];
8344 unsigned long update_time;
8345 sector_t mark_cnt[SYNC_MARKS];
8347 struct list_head *tmp;
8348 sector_t last_check;
8350 struct md_rdev *rdev;
8351 char *desc, *action = NULL;
8352 struct blk_plug plug;
8355 /* just incase thread restarts... */
8356 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8357 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8359 if (mddev->ro) {/* never try to sync a read-only array */
8360 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8364 if (mddev_is_clustered(mddev)) {
8365 ret = md_cluster_ops->resync_start(mddev);
8369 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8370 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8371 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8372 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8373 && ((unsigned long long)mddev->curr_resync_completed
8374 < (unsigned long long)mddev->resync_max_sectors))
8378 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8379 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8380 desc = "data-check";
8382 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8383 desc = "requested-resync";
8387 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8392 mddev->last_sync_action = action ?: desc;
8394 /* we overload curr_resync somewhat here.
8395 * 0 == not engaged in resync at all
8396 * 2 == checking that there is no conflict with another sync
8397 * 1 == like 2, but have yielded to allow conflicting resync to
8399 * other == active in resync - this many blocks
8401 * Before starting a resync we must have set curr_resync to
8402 * 2, and then checked that every "conflicting" array has curr_resync
8403 * less than ours. When we find one that is the same or higher
8404 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8405 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8406 * This will mean we have to start checking from the beginning again.
8411 int mddev2_minor = -1;
8412 mddev->curr_resync = 2;
8415 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8417 for_each_mddev(mddev2, tmp) {
8418 if (mddev2 == mddev)
8420 if (!mddev->parallel_resync
8421 && mddev2->curr_resync
8422 && match_mddev_units(mddev, mddev2)) {
8424 if (mddev < mddev2 && mddev->curr_resync == 2) {
8425 /* arbitrarily yield */
8426 mddev->curr_resync = 1;
8427 wake_up(&resync_wait);
8429 if (mddev > mddev2 && mddev->curr_resync == 1)
8430 /* no need to wait here, we can wait the next
8431 * time 'round when curr_resync == 2
8434 /* We need to wait 'interruptible' so as not to
8435 * contribute to the load average, and not to
8436 * be caught by 'softlockup'
8438 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8439 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8440 mddev2->curr_resync >= mddev->curr_resync) {
8441 if (mddev2_minor != mddev2->md_minor) {
8442 mddev2_minor = mddev2->md_minor;
8443 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8444 desc, mdname(mddev),
8448 if (signal_pending(current))
8449 flush_signals(current);
8451 finish_wait(&resync_wait, &wq);
8454 finish_wait(&resync_wait, &wq);
8457 } while (mddev->curr_resync < 2);
8460 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8461 /* resync follows the size requested by the personality,
8462 * which defaults to physical size, but can be virtual size
8464 max_sectors = mddev->resync_max_sectors;
8465 atomic64_set(&mddev->resync_mismatches, 0);
8466 /* we don't use the checkpoint if there's a bitmap */
8467 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8468 j = mddev->resync_min;
8469 else if (!mddev->bitmap)
8470 j = mddev->recovery_cp;
8472 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8473 max_sectors = mddev->resync_max_sectors;
8475 /* recovery follows the physical size of devices */
8476 max_sectors = mddev->dev_sectors;
8479 rdev_for_each_rcu(rdev, mddev)
8480 if (rdev->raid_disk >= 0 &&
8481 !test_bit(Journal, &rdev->flags) &&
8482 !test_bit(Faulty, &rdev->flags) &&
8483 !test_bit(In_sync, &rdev->flags) &&
8484 rdev->recovery_offset < j)
8485 j = rdev->recovery_offset;
8488 /* If there is a bitmap, we need to make sure all
8489 * writes that started before we added a spare
8490 * complete before we start doing a recovery.
8491 * Otherwise the write might complete and (via
8492 * bitmap_endwrite) set a bit in the bitmap after the
8493 * recovery has checked that bit and skipped that
8496 if (mddev->bitmap) {
8497 mddev->pers->quiesce(mddev, 1);
8498 mddev->pers->quiesce(mddev, 0);
8502 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8503 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8504 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8505 speed_max(mddev), desc);
8507 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8510 for (m = 0; m < SYNC_MARKS; m++) {
8512 mark_cnt[m] = io_sectors;
8515 mddev->resync_mark = mark[last_mark];
8516 mddev->resync_mark_cnt = mark_cnt[last_mark];
8519 * Tune reconstruction:
8521 window = 32*(PAGE_SIZE/512);
8522 pr_debug("md: using %dk window, over a total of %lluk.\n",
8523 window/2, (unsigned long long)max_sectors/2);
8525 atomic_set(&mddev->recovery_active, 0);
8529 pr_debug("md: resuming %s of %s from checkpoint.\n",
8530 desc, mdname(mddev));
8531 mddev->curr_resync = j;
8533 mddev->curr_resync = 3; /* no longer delayed */
8534 mddev->curr_resync_completed = j;
8535 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8536 md_new_event(mddev);
8537 update_time = jiffies;
8539 blk_start_plug(&plug);
8540 while (j < max_sectors) {
8545 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8546 ((mddev->curr_resync > mddev->curr_resync_completed &&
8547 (mddev->curr_resync - mddev->curr_resync_completed)
8548 > (max_sectors >> 4)) ||
8549 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8550 (j - mddev->curr_resync_completed)*2
8551 >= mddev->resync_max - mddev->curr_resync_completed ||
8552 mddev->curr_resync_completed > mddev->resync_max
8554 /* time to update curr_resync_completed */
8555 wait_event(mddev->recovery_wait,
8556 atomic_read(&mddev->recovery_active) == 0);
8557 mddev->curr_resync_completed = j;
8558 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8559 j > mddev->recovery_cp)
8560 mddev->recovery_cp = j;
8561 update_time = jiffies;
8562 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8563 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8566 while (j >= mddev->resync_max &&
8567 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8568 /* As this condition is controlled by user-space,
8569 * we can block indefinitely, so use '_interruptible'
8570 * to avoid triggering warnings.
8572 flush_signals(current); /* just in case */
8573 wait_event_interruptible(mddev->recovery_wait,
8574 mddev->resync_max > j
8575 || test_bit(MD_RECOVERY_INTR,
8579 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8582 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8584 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8588 if (!skipped) { /* actual IO requested */
8589 io_sectors += sectors;
8590 atomic_add(sectors, &mddev->recovery_active);
8593 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8597 if (j > max_sectors)
8598 /* when skipping, extra large numbers can be returned. */
8601 mddev->curr_resync = j;
8602 mddev->curr_mark_cnt = io_sectors;
8603 if (last_check == 0)
8604 /* this is the earliest that rebuild will be
8605 * visible in /proc/mdstat
8607 md_new_event(mddev);
8609 if (last_check + window > io_sectors || j == max_sectors)
8612 last_check = io_sectors;
8614 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8616 int next = (last_mark+1) % SYNC_MARKS;
8618 mddev->resync_mark = mark[next];
8619 mddev->resync_mark_cnt = mark_cnt[next];
8620 mark[next] = jiffies;
8621 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8625 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8629 * this loop exits only if either when we are slower than
8630 * the 'hard' speed limit, or the system was IO-idle for
8632 * the system might be non-idle CPU-wise, but we only care
8633 * about not overloading the IO subsystem. (things like an
8634 * e2fsck being done on the RAID array should execute fast)
8638 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8639 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8640 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8642 if (currspeed > speed_min(mddev)) {
8643 if (currspeed > speed_max(mddev)) {
8647 if (!is_mddev_idle(mddev, 0)) {
8649 * Give other IO more of a chance.
8650 * The faster the devices, the less we wait.
8652 wait_event(mddev->recovery_wait,
8653 !atomic_read(&mddev->recovery_active));
8657 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8658 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8659 ? "interrupted" : "done");
8661 * this also signals 'finished resyncing' to md_stop
8663 blk_finish_plug(&plug);
8664 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8666 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8667 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8668 mddev->curr_resync > 3) {
8669 mddev->curr_resync_completed = mddev->curr_resync;
8670 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8672 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8674 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8675 mddev->curr_resync > 3) {
8676 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8677 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8678 if (mddev->curr_resync >= mddev->recovery_cp) {
8679 pr_debug("md: checkpointing %s of %s.\n",
8680 desc, mdname(mddev));
8681 if (test_bit(MD_RECOVERY_ERROR,
8683 mddev->recovery_cp =
8684 mddev->curr_resync_completed;
8686 mddev->recovery_cp =
8690 mddev->recovery_cp = MaxSector;
8692 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8693 mddev->curr_resync = MaxSector;
8694 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8695 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8697 rdev_for_each_rcu(rdev, mddev)
8698 if (rdev->raid_disk >= 0 &&
8699 mddev->delta_disks >= 0 &&
8700 !test_bit(Journal, &rdev->flags) &&
8701 !test_bit(Faulty, &rdev->flags) &&
8702 !test_bit(In_sync, &rdev->flags) &&
8703 rdev->recovery_offset < mddev->curr_resync)
8704 rdev->recovery_offset = mddev->curr_resync;
8710 /* set CHANGE_PENDING here since maybe another update is needed,
8711 * so other nodes are informed. It should be harmless for normal
8713 set_mask_bits(&mddev->sb_flags, 0,
8714 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8716 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8717 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8718 mddev->delta_disks > 0 &&
8719 mddev->pers->finish_reshape &&
8720 mddev->pers->size &&
8722 mddev_lock_nointr(mddev);
8723 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8724 mddev_unlock(mddev);
8725 set_capacity(mddev->gendisk, mddev->array_sectors);
8726 revalidate_disk(mddev->gendisk);
8729 spin_lock(&mddev->lock);
8730 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8731 /* We completed so min/max setting can be forgotten if used. */
8732 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8733 mddev->resync_min = 0;
8734 mddev->resync_max = MaxSector;
8735 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8736 mddev->resync_min = mddev->curr_resync_completed;
8737 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8738 mddev->curr_resync = 0;
8739 spin_unlock(&mddev->lock);
8741 wake_up(&resync_wait);
8742 md_wakeup_thread(mddev->thread);
8745 EXPORT_SYMBOL_GPL(md_do_sync);
8747 static int remove_and_add_spares(struct mddev *mddev,
8748 struct md_rdev *this)
8750 struct md_rdev *rdev;
8753 bool remove_some = false;
8755 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8756 /* Mustn't remove devices when resync thread is running */
8759 rdev_for_each(rdev, mddev) {
8760 if ((this == NULL || rdev == this) &&
8761 rdev->raid_disk >= 0 &&
8762 !test_bit(Blocked, &rdev->flags) &&
8763 test_bit(Faulty, &rdev->flags) &&
8764 atomic_read(&rdev->nr_pending)==0) {
8765 /* Faulty non-Blocked devices with nr_pending == 0
8766 * never get nr_pending incremented,
8767 * never get Faulty cleared, and never get Blocked set.
8768 * So we can synchronize_rcu now rather than once per device
8771 set_bit(RemoveSynchronized, &rdev->flags);
8777 rdev_for_each(rdev, mddev) {
8778 if ((this == NULL || rdev == this) &&
8779 rdev->raid_disk >= 0 &&
8780 !test_bit(Blocked, &rdev->flags) &&
8781 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8782 (!test_bit(In_sync, &rdev->flags) &&
8783 !test_bit(Journal, &rdev->flags))) &&
8784 atomic_read(&rdev->nr_pending)==0)) {
8785 if (mddev->pers->hot_remove_disk(
8786 mddev, rdev) == 0) {
8787 sysfs_unlink_rdev(mddev, rdev);
8788 rdev->saved_raid_disk = rdev->raid_disk;
8789 rdev->raid_disk = -1;
8793 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8794 clear_bit(RemoveSynchronized, &rdev->flags);
8797 if (removed && mddev->kobj.sd)
8798 sysfs_notify(&mddev->kobj, NULL, "degraded");
8800 if (this && removed)
8803 rdev_for_each(rdev, mddev) {
8804 if (this && this != rdev)
8806 if (test_bit(Candidate, &rdev->flags))
8808 if (rdev->raid_disk >= 0 &&
8809 !test_bit(In_sync, &rdev->flags) &&
8810 !test_bit(Journal, &rdev->flags) &&
8811 !test_bit(Faulty, &rdev->flags))
8813 if (rdev->raid_disk >= 0)
8815 if (test_bit(Faulty, &rdev->flags))
8817 if (!test_bit(Journal, &rdev->flags)) {
8819 ! (rdev->saved_raid_disk >= 0 &&
8820 !test_bit(Bitmap_sync, &rdev->flags)))
8823 rdev->recovery_offset = 0;
8826 hot_add_disk(mddev, rdev) == 0) {
8827 if (sysfs_link_rdev(mddev, rdev))
8828 /* failure here is OK */;
8829 if (!test_bit(Journal, &rdev->flags))
8831 md_new_event(mddev);
8832 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8837 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8841 static void md_start_sync(struct work_struct *ws)
8843 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8845 mddev->sync_thread = md_register_thread(md_do_sync,
8848 if (!mddev->sync_thread) {
8849 pr_warn("%s: could not start resync thread...\n",
8851 /* leave the spares where they are, it shouldn't hurt */
8852 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8853 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8854 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8855 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8856 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8857 wake_up(&resync_wait);
8858 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8860 if (mddev->sysfs_action)
8861 sysfs_notify_dirent_safe(mddev->sysfs_action);
8863 md_wakeup_thread(mddev->sync_thread);
8864 sysfs_notify_dirent_safe(mddev->sysfs_action);
8865 md_new_event(mddev);
8869 * This routine is regularly called by all per-raid-array threads to
8870 * deal with generic issues like resync and super-block update.
8871 * Raid personalities that don't have a thread (linear/raid0) do not
8872 * need this as they never do any recovery or update the superblock.
8874 * It does not do any resync itself, but rather "forks" off other threads
8875 * to do that as needed.
8876 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8877 * "->recovery" and create a thread at ->sync_thread.
8878 * When the thread finishes it sets MD_RECOVERY_DONE
8879 * and wakeups up this thread which will reap the thread and finish up.
8880 * This thread also removes any faulty devices (with nr_pending == 0).
8882 * The overall approach is:
8883 * 1/ if the superblock needs updating, update it.
8884 * 2/ If a recovery thread is running, don't do anything else.
8885 * 3/ If recovery has finished, clean up, possibly marking spares active.
8886 * 4/ If there are any faulty devices, remove them.
8887 * 5/ If array is degraded, try to add spares devices
8888 * 6/ If array has spares or is not in-sync, start a resync thread.
8890 void md_check_recovery(struct mddev *mddev)
8892 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8893 /* Write superblock - thread that called mddev_suspend()
8894 * holds reconfig_mutex for us.
8896 set_bit(MD_UPDATING_SB, &mddev->flags);
8897 smp_mb__after_atomic();
8898 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8899 md_update_sb(mddev, 0);
8900 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8901 wake_up(&mddev->sb_wait);
8904 if (mddev->suspended)
8908 md_bitmap_daemon_work(mddev);
8910 if (signal_pending(current)) {
8911 if (mddev->pers->sync_request && !mddev->external) {
8912 pr_debug("md: %s in immediate safe mode\n",
8914 mddev->safemode = 2;
8916 flush_signals(current);
8919 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8922 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8923 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8924 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8925 (mddev->external == 0 && mddev->safemode == 1) ||
8926 (mddev->safemode == 2
8927 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8931 if (mddev_trylock(mddev)) {
8933 bool try_set_sync = mddev->safemode != 0;
8935 if (!mddev->external && mddev->safemode == 1)
8936 mddev->safemode = 0;
8939 struct md_rdev *rdev;
8940 if (!mddev->external && mddev->in_sync)
8941 /* 'Blocked' flag not needed as failed devices
8942 * will be recorded if array switched to read/write.
8943 * Leaving it set will prevent the device
8944 * from being removed.
8946 rdev_for_each(rdev, mddev)
8947 clear_bit(Blocked, &rdev->flags);
8948 /* On a read-only array we can:
8949 * - remove failed devices
8950 * - add already-in_sync devices if the array itself
8952 * As we only add devices that are already in-sync,
8953 * we can activate the spares immediately.
8955 remove_and_add_spares(mddev, NULL);
8956 /* There is no thread, but we need to call
8957 * ->spare_active and clear saved_raid_disk
8959 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8960 md_reap_sync_thread(mddev);
8961 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8962 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8963 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8967 if (mddev_is_clustered(mddev)) {
8968 struct md_rdev *rdev, *tmp;
8969 /* kick the device if another node issued a
8972 rdev_for_each_safe(rdev, tmp, mddev) {
8973 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8974 rdev->raid_disk < 0)
8975 md_kick_rdev_from_array(rdev);
8979 if (try_set_sync && !mddev->external && !mddev->in_sync) {
8980 spin_lock(&mddev->lock);
8982 spin_unlock(&mddev->lock);
8985 if (mddev->sb_flags)
8986 md_update_sb(mddev, 0);
8988 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8989 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8990 /* resync/recovery still happening */
8991 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8994 if (mddev->sync_thread) {
8995 md_reap_sync_thread(mddev);
8998 /* Set RUNNING before clearing NEEDED to avoid
8999 * any transients in the value of "sync_action".
9001 mddev->curr_resync_completed = 0;
9002 spin_lock(&mddev->lock);
9003 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9004 spin_unlock(&mddev->lock);
9005 /* Clear some bits that don't mean anything, but
9008 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9009 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9011 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9012 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9014 /* no recovery is running.
9015 * remove any failed drives, then
9016 * add spares if possible.
9017 * Spares are also removed and re-added, to allow
9018 * the personality to fail the re-add.
9021 if (mddev->reshape_position != MaxSector) {
9022 if (mddev->pers->check_reshape == NULL ||
9023 mddev->pers->check_reshape(mddev) != 0)
9024 /* Cannot proceed */
9026 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9027 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9028 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9029 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9030 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9031 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9032 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9033 } else if (mddev->recovery_cp < MaxSector) {
9034 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9035 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9036 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9037 /* nothing to be done ... */
9040 if (mddev->pers->sync_request) {
9042 /* We are adding a device or devices to an array
9043 * which has the bitmap stored on all devices.
9044 * So make sure all bitmap pages get written
9046 md_bitmap_write_all(mddev->bitmap);
9048 INIT_WORK(&mddev->del_work, md_start_sync);
9049 queue_work(md_misc_wq, &mddev->del_work);
9053 if (!mddev->sync_thread) {
9054 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9055 wake_up(&resync_wait);
9056 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9058 if (mddev->sysfs_action)
9059 sysfs_notify_dirent_safe(mddev->sysfs_action);
9062 wake_up(&mddev->sb_wait);
9063 mddev_unlock(mddev);
9066 EXPORT_SYMBOL(md_check_recovery);
9068 void md_reap_sync_thread(struct mddev *mddev)
9070 struct md_rdev *rdev;
9072 /* resync has finished, collect result */
9073 md_unregister_thread(&mddev->sync_thread);
9074 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9075 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9076 mddev->degraded != mddev->raid_disks) {
9078 /* activate any spares */
9079 if (mddev->pers->spare_active(mddev)) {
9080 sysfs_notify(&mddev->kobj, NULL,
9082 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9085 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9086 mddev->pers->finish_reshape)
9087 mddev->pers->finish_reshape(mddev);
9089 /* If array is no-longer degraded, then any saved_raid_disk
9090 * information must be scrapped.
9092 if (!mddev->degraded)
9093 rdev_for_each(rdev, mddev)
9094 rdev->saved_raid_disk = -1;
9096 md_update_sb(mddev, 1);
9097 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9098 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9100 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9101 md_cluster_ops->resync_finish(mddev);
9102 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9103 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9104 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9105 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9106 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9107 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9108 wake_up(&resync_wait);
9109 /* flag recovery needed just to double check */
9110 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9111 sysfs_notify_dirent_safe(mddev->sysfs_action);
9112 md_new_event(mddev);
9113 if (mddev->event_work.func)
9114 queue_work(md_misc_wq, &mddev->event_work);
9116 EXPORT_SYMBOL(md_reap_sync_thread);
9118 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9120 sysfs_notify_dirent_safe(rdev->sysfs_state);
9121 wait_event_timeout(rdev->blocked_wait,
9122 !test_bit(Blocked, &rdev->flags) &&
9123 !test_bit(BlockedBadBlocks, &rdev->flags),
9124 msecs_to_jiffies(5000));
9125 rdev_dec_pending(rdev, mddev);
9127 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9129 void md_finish_reshape(struct mddev *mddev)
9131 /* called be personality module when reshape completes. */
9132 struct md_rdev *rdev;
9134 rdev_for_each(rdev, mddev) {
9135 if (rdev->data_offset > rdev->new_data_offset)
9136 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9138 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9139 rdev->data_offset = rdev->new_data_offset;
9142 EXPORT_SYMBOL(md_finish_reshape);
9144 /* Bad block management */
9146 /* Returns 1 on success, 0 on failure */
9147 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9150 struct mddev *mddev = rdev->mddev;
9153 s += rdev->new_data_offset;
9155 s += rdev->data_offset;
9156 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9158 /* Make sure they get written out promptly */
9159 if (test_bit(ExternalBbl, &rdev->flags))
9160 sysfs_notify(&rdev->kobj, NULL,
9161 "unacknowledged_bad_blocks");
9162 sysfs_notify_dirent_safe(rdev->sysfs_state);
9163 set_mask_bits(&mddev->sb_flags, 0,
9164 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9165 md_wakeup_thread(rdev->mddev->thread);
9170 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9172 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9177 s += rdev->new_data_offset;
9179 s += rdev->data_offset;
9180 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9181 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9182 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9185 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9187 static int md_notify_reboot(struct notifier_block *this,
9188 unsigned long code, void *x)
9190 struct list_head *tmp;
9191 struct mddev *mddev;
9194 for_each_mddev(mddev, tmp) {
9195 if (mddev_trylock(mddev)) {
9197 __md_stop_writes(mddev);
9198 if (mddev->persistent)
9199 mddev->safemode = 2;
9200 mddev_unlock(mddev);
9205 * certain more exotic SCSI devices are known to be
9206 * volatile wrt too early system reboots. While the
9207 * right place to handle this issue is the given
9208 * driver, we do want to have a safe RAID driver ...
9216 static struct notifier_block md_notifier = {
9217 .notifier_call = md_notify_reboot,
9219 .priority = INT_MAX, /* before any real devices */
9222 static void md_geninit(void)
9224 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9226 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9229 static int __init md_init(void)
9233 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9237 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9241 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9244 if ((ret = register_blkdev(0, "mdp")) < 0)
9248 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9249 md_probe, NULL, NULL);
9250 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9251 md_probe, NULL, NULL);
9253 register_reboot_notifier(&md_notifier);
9254 raid_table_header = register_sysctl_table(raid_root_table);
9260 unregister_blkdev(MD_MAJOR, "md");
9262 destroy_workqueue(md_misc_wq);
9264 destroy_workqueue(md_wq);
9269 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9271 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9272 struct md_rdev *rdev2, *tmp;
9274 char b[BDEVNAME_SIZE];
9277 * If size is changed in another node then we need to
9278 * do resize as well.
9280 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9281 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9283 pr_info("md-cluster: resize failed\n");
9285 md_bitmap_update_sb(mddev->bitmap);
9288 /* Check for change of roles in the active devices */
9289 rdev_for_each_safe(rdev2, tmp, mddev) {
9290 if (test_bit(Faulty, &rdev2->flags))
9293 /* Check if the roles changed */
9294 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9296 if (test_bit(Candidate, &rdev2->flags)) {
9297 if (role == 0xfffe) {
9298 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9299 md_kick_rdev_from_array(rdev2);
9303 clear_bit(Candidate, &rdev2->flags);
9306 if (role != rdev2->raid_disk) {
9308 if (rdev2->raid_disk == -1 && role != 0xffff) {
9309 rdev2->saved_raid_disk = role;
9310 ret = remove_and_add_spares(mddev, rdev2);
9311 pr_info("Activated spare: %s\n",
9312 bdevname(rdev2->bdev,b));
9313 /* wakeup mddev->thread here, so array could
9314 * perform resync with the new activated disk */
9315 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9316 md_wakeup_thread(mddev->thread);
9320 * We just want to do the minimum to mark the disk
9321 * as faulty. The recovery is performed by the
9322 * one who initiated the error.
9324 if ((role == 0xfffe) || (role == 0xfffd)) {
9325 md_error(mddev, rdev2);
9326 clear_bit(Blocked, &rdev2->flags);
9331 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
9332 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9334 pr_warn("md: updating array disks failed. %d\n", ret);
9337 /* Finally set the event to be up to date */
9338 mddev->events = le64_to_cpu(sb->events);
9341 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9344 struct page *swapout = rdev->sb_page;
9345 struct mdp_superblock_1 *sb;
9347 /* Store the sb page of the rdev in the swapout temporary
9348 * variable in case we err in the future
9350 rdev->sb_page = NULL;
9351 err = alloc_disk_sb(rdev);
9353 ClearPageUptodate(rdev->sb_page);
9354 rdev->sb_loaded = 0;
9355 err = super_types[mddev->major_version].
9356 load_super(rdev, NULL, mddev->minor_version);
9359 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9360 __func__, __LINE__, rdev->desc_nr, err);
9362 put_page(rdev->sb_page);
9363 rdev->sb_page = swapout;
9364 rdev->sb_loaded = 1;
9368 sb = page_address(rdev->sb_page);
9369 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9373 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9374 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9376 /* The other node finished recovery, call spare_active to set
9377 * device In_sync and mddev->degraded
9379 if (rdev->recovery_offset == MaxSector &&
9380 !test_bit(In_sync, &rdev->flags) &&
9381 mddev->pers->spare_active(mddev))
9382 sysfs_notify(&mddev->kobj, NULL, "degraded");
9388 void md_reload_sb(struct mddev *mddev, int nr)
9390 struct md_rdev *rdev = NULL, *iter;
9394 rdev_for_each_rcu(iter, mddev) {
9395 if (iter->desc_nr == nr) {
9402 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9406 err = read_rdev(mddev, rdev);
9410 check_sb_changes(mddev, rdev);
9412 /* Read all rdev's to update recovery_offset */
9413 rdev_for_each_rcu(rdev, mddev) {
9414 if (!test_bit(Faulty, &rdev->flags))
9415 read_rdev(mddev, rdev);
9418 EXPORT_SYMBOL(md_reload_sb);
9423 * Searches all registered partitions for autorun RAID arrays
9427 static DEFINE_MUTEX(detected_devices_mutex);
9428 static LIST_HEAD(all_detected_devices);
9429 struct detected_devices_node {
9430 struct list_head list;
9434 void md_autodetect_dev(dev_t dev)
9436 struct detected_devices_node *node_detected_dev;
9438 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9439 if (node_detected_dev) {
9440 node_detected_dev->dev = dev;
9441 mutex_lock(&detected_devices_mutex);
9442 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9443 mutex_unlock(&detected_devices_mutex);
9447 static void autostart_arrays(int part)
9449 struct md_rdev *rdev;
9450 struct detected_devices_node *node_detected_dev;
9452 int i_scanned, i_passed;
9457 pr_info("md: Autodetecting RAID arrays.\n");
9459 mutex_lock(&detected_devices_mutex);
9460 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9462 node_detected_dev = list_entry(all_detected_devices.next,
9463 struct detected_devices_node, list);
9464 list_del(&node_detected_dev->list);
9465 dev = node_detected_dev->dev;
9466 kfree(node_detected_dev);
9467 mutex_unlock(&detected_devices_mutex);
9468 rdev = md_import_device(dev,0, 90);
9469 mutex_lock(&detected_devices_mutex);
9473 if (test_bit(Faulty, &rdev->flags))
9476 set_bit(AutoDetected, &rdev->flags);
9477 list_add(&rdev->same_set, &pending_raid_disks);
9480 mutex_unlock(&detected_devices_mutex);
9482 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9484 autorun_devices(part);
9487 #endif /* !MODULE */
9489 static __exit void md_exit(void)
9491 struct mddev *mddev;
9492 struct list_head *tmp;
9495 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9496 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9498 unregister_blkdev(MD_MAJOR,"md");
9499 unregister_blkdev(mdp_major, "mdp");
9500 unregister_reboot_notifier(&md_notifier);
9501 unregister_sysctl_table(raid_table_header);
9503 /* We cannot unload the modules while some process is
9504 * waiting for us in select() or poll() - wake them up
9507 while (waitqueue_active(&md_event_waiters)) {
9508 /* not safe to leave yet */
9509 wake_up(&md_event_waiters);
9513 remove_proc_entry("mdstat", NULL);
9515 for_each_mddev(mddev, tmp) {
9516 export_array(mddev);
9518 mddev->hold_active = 0;
9520 * for_each_mddev() will call mddev_put() at the end of each
9521 * iteration. As the mddev is now fully clear, this will
9522 * schedule the mddev for destruction by a workqueue, and the
9523 * destroy_workqueue() below will wait for that to complete.
9526 destroy_workqueue(md_misc_wq);
9527 destroy_workqueue(md_wq);
9530 subsys_initcall(md_init);
9531 module_exit(md_exit)
9533 static int get_ro(char *buffer, const struct kernel_param *kp)
9535 return sprintf(buffer, "%d", start_readonly);
9537 static int set_ro(const char *val, const struct kernel_param *kp)
9539 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9542 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9543 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9544 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9545 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9547 MODULE_LICENSE("GPL");
9548 MODULE_DESCRIPTION("MD RAID framework");
9550 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);