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 static struct module *md_cluster_mod;
93 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
94 static struct workqueue_struct *md_wq;
95 static struct workqueue_struct *md_misc_wq;
97 static int remove_and_add_spares(struct mddev *mddev,
98 struct md_rdev *this);
99 static void mddev_detach(struct mddev *mddev);
102 * Default number of read corrections we'll attempt on an rdev
103 * before ejecting it from the array. We divide the read error
104 * count by 2 for every hour elapsed between read errors.
106 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
108 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
109 * is 1000 KB/sec, so the extra system load does not show up that much.
110 * Increase it if you want to have more _guaranteed_ speed. Note that
111 * the RAID driver will use the maximum available bandwidth if the IO
112 * subsystem is idle. There is also an 'absolute maximum' reconstruction
113 * speed limit - in case reconstruction slows down your system despite
116 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
117 * or /sys/block/mdX/md/sync_speed_{min,max}
120 static int sysctl_speed_limit_min = 1000;
121 static int sysctl_speed_limit_max = 200000;
122 static inline int speed_min(struct mddev *mddev)
124 return mddev->sync_speed_min ?
125 mddev->sync_speed_min : sysctl_speed_limit_min;
128 static inline int speed_max(struct mddev *mddev)
130 return mddev->sync_speed_max ?
131 mddev->sync_speed_max : sysctl_speed_limit_max;
134 static struct ctl_table_header *raid_table_header;
136 static struct ctl_table raid_table[] = {
138 .procname = "speed_limit_min",
139 .data = &sysctl_speed_limit_min,
140 .maxlen = sizeof(int),
141 .mode = S_IRUGO|S_IWUSR,
142 .proc_handler = proc_dointvec,
145 .procname = "speed_limit_max",
146 .data = &sysctl_speed_limit_max,
147 .maxlen = sizeof(int),
148 .mode = S_IRUGO|S_IWUSR,
149 .proc_handler = proc_dointvec,
154 static struct ctl_table raid_dir_table[] = {
158 .mode = S_IRUGO|S_IXUGO,
164 static struct ctl_table raid_root_table[] = {
169 .child = raid_dir_table,
174 static const struct block_device_operations md_fops;
176 static int start_readonly;
179 * The original mechanism for creating an md device is to create
180 * a device node in /dev and to open it. This causes races with device-close.
181 * The preferred method is to write to the "new_array" module parameter.
182 * This can avoid races.
183 * Setting create_on_open to false disables the original mechanism
184 * so all the races disappear.
186 static bool create_on_open = true;
188 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
191 if (!mddev || !bioset_initialized(&mddev->bio_set))
192 return bio_alloc(gfp_mask, nr_iovecs);
194 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
196 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
198 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
200 if (!mddev || !bioset_initialized(&mddev->sync_set))
201 return bio_alloc(GFP_NOIO, 1);
203 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
207 * We have a system wide 'event count' that is incremented
208 * on any 'interesting' event, and readers of /proc/mdstat
209 * can use 'poll' or 'select' to find out when the event
213 * start array, stop array, error, add device, remove device,
214 * start build, activate spare
216 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
217 static atomic_t md_event_count;
218 void md_new_event(struct mddev *mddev)
220 atomic_inc(&md_event_count);
221 wake_up(&md_event_waiters);
223 EXPORT_SYMBOL_GPL(md_new_event);
226 * Enables to iterate over all existing md arrays
227 * all_mddevs_lock protects this list.
229 static LIST_HEAD(all_mddevs);
230 static DEFINE_SPINLOCK(all_mddevs_lock);
233 * iterates through all used mddevs in the system.
234 * We take care to grab the all_mddevs_lock whenever navigating
235 * the list, and to always hold a refcount when unlocked.
236 * Any code which breaks out of this loop while own
237 * a reference to the current mddev and must mddev_put it.
239 #define for_each_mddev(_mddev,_tmp) \
241 for (({ spin_lock(&all_mddevs_lock); \
242 _tmp = all_mddevs.next; \
244 ({ if (_tmp != &all_mddevs) \
245 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
246 spin_unlock(&all_mddevs_lock); \
247 if (_mddev) mddev_put(_mddev); \
248 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
249 _tmp != &all_mddevs;}); \
250 ({ spin_lock(&all_mddevs_lock); \
251 _tmp = _tmp->next;}) \
254 /* Rather than calling directly into the personality make_request function,
255 * IO requests come here first so that we can check if the device is
256 * being suspended pending a reconfiguration.
257 * We hold a refcount over the call to ->make_request. By the time that
258 * call has finished, the bio has been linked into some internal structure
259 * and so is visible to ->quiesce(), so we don't need the refcount any more.
261 static bool is_suspended(struct mddev *mddev, struct bio *bio)
263 if (mddev->suspended)
265 if (bio_data_dir(bio) != WRITE)
267 if (mddev->suspend_lo >= mddev->suspend_hi)
269 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
271 if (bio_end_sector(bio) < mddev->suspend_lo)
276 void md_handle_request(struct mddev *mddev, struct bio *bio)
280 if (is_suspended(mddev, bio)) {
283 prepare_to_wait(&mddev->sb_wait, &__wait,
284 TASK_UNINTERRUPTIBLE);
285 if (!is_suspended(mddev, bio))
291 finish_wait(&mddev->sb_wait, &__wait);
293 atomic_inc(&mddev->active_io);
296 if (!mddev->pers->make_request(mddev, bio)) {
297 atomic_dec(&mddev->active_io);
298 wake_up(&mddev->sb_wait);
299 goto check_suspended;
302 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
303 wake_up(&mddev->sb_wait);
305 EXPORT_SYMBOL(md_handle_request);
307 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
309 const int rw = bio_data_dir(bio);
310 const int sgrp = op_stat_group(bio_op(bio));
311 struct mddev *mddev = q->queuedata;
312 unsigned int sectors;
314 blk_queue_split(q, &bio);
316 if (mddev == NULL || mddev->pers == NULL) {
318 return BLK_QC_T_NONE;
320 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
321 if (bio_sectors(bio) != 0)
322 bio->bi_status = BLK_STS_IOERR;
324 return BLK_QC_T_NONE;
328 * save the sectors now since our bio can
329 * go away inside make_request
331 sectors = bio_sectors(bio);
332 /* bio could be mergeable after passing to underlayer */
333 bio->bi_opf &= ~REQ_NOMERGE;
335 md_handle_request(mddev, bio);
338 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
339 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
342 return BLK_QC_T_NONE;
345 /* mddev_suspend makes sure no new requests are submitted
346 * to the device, and that any requests that have been submitted
347 * are completely handled.
348 * Once mddev_detach() is called and completes, the module will be
351 void mddev_suspend(struct mddev *mddev)
353 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
354 lockdep_assert_held(&mddev->reconfig_mutex);
355 if (mddev->suspended++)
358 wake_up(&mddev->sb_wait);
359 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
360 smp_mb__after_atomic();
361 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
362 mddev->pers->quiesce(mddev, 1);
363 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
364 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
366 del_timer_sync(&mddev->safemode_timer);
368 EXPORT_SYMBOL_GPL(mddev_suspend);
370 void mddev_resume(struct mddev *mddev)
372 lockdep_assert_held(&mddev->reconfig_mutex);
373 if (--mddev->suspended)
375 wake_up(&mddev->sb_wait);
376 mddev->pers->quiesce(mddev, 0);
378 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
379 md_wakeup_thread(mddev->thread);
380 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
382 EXPORT_SYMBOL_GPL(mddev_resume);
384 int mddev_congested(struct mddev *mddev, int bits)
386 struct md_personality *pers = mddev->pers;
390 if (mddev->suspended)
392 else if (pers && pers->congested)
393 ret = pers->congested(mddev, bits);
397 EXPORT_SYMBOL_GPL(mddev_congested);
398 static int md_congested(void *data, int bits)
400 struct mddev *mddev = data;
401 return mddev_congested(mddev, bits);
405 * Generic flush handling for md
408 static void md_end_flush(struct bio *bio)
410 struct md_rdev *rdev = bio->bi_private;
411 struct mddev *mddev = rdev->mddev;
413 rdev_dec_pending(rdev, mddev);
415 if (atomic_dec_and_test(&mddev->flush_pending)) {
416 /* The pre-request flush has finished */
417 queue_work(md_wq, &mddev->flush_work);
422 static void md_submit_flush_data(struct work_struct *ws);
424 static void submit_flushes(struct work_struct *ws)
426 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
427 struct md_rdev *rdev;
429 mddev->start_flush = ktime_get_boottime();
430 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
431 atomic_set(&mddev->flush_pending, 1);
433 rdev_for_each_rcu(rdev, mddev)
434 if (rdev->raid_disk >= 0 &&
435 !test_bit(Faulty, &rdev->flags)) {
436 /* Take two references, one is dropped
437 * when request finishes, one after
438 * we reclaim rcu_read_lock
441 atomic_inc(&rdev->nr_pending);
442 atomic_inc(&rdev->nr_pending);
444 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
445 bi->bi_end_io = md_end_flush;
446 bi->bi_private = rdev;
447 bio_set_dev(bi, rdev->bdev);
448 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
449 atomic_inc(&mddev->flush_pending);
452 rdev_dec_pending(rdev, mddev);
455 if (atomic_dec_and_test(&mddev->flush_pending))
456 queue_work(md_wq, &mddev->flush_work);
459 static void md_submit_flush_data(struct work_struct *ws)
461 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
462 struct bio *bio = mddev->flush_bio;
465 * must reset flush_bio before calling into md_handle_request to avoid a
466 * deadlock, because other bios passed md_handle_request suspend check
467 * could wait for this and below md_handle_request could wait for those
468 * bios because of suspend check
470 mddev->last_flush = mddev->start_flush;
471 mddev->flush_bio = NULL;
472 wake_up(&mddev->sb_wait);
474 if (bio->bi_iter.bi_size == 0) {
475 /* an empty barrier - all done */
478 bio->bi_opf &= ~REQ_PREFLUSH;
479 md_handle_request(mddev, bio);
483 void md_flush_request(struct mddev *mddev, struct bio *bio)
485 ktime_t start = ktime_get_boottime();
486 spin_lock_irq(&mddev->lock);
487 wait_event_lock_irq(mddev->sb_wait,
489 ktime_after(mddev->last_flush, start),
491 if (!ktime_after(mddev->last_flush, start)) {
492 WARN_ON(mddev->flush_bio);
493 mddev->flush_bio = bio;
496 spin_unlock_irq(&mddev->lock);
499 INIT_WORK(&mddev->flush_work, submit_flushes);
500 queue_work(md_wq, &mddev->flush_work);
502 /* flush was performed for some other bio while we waited. */
503 if (bio->bi_iter.bi_size == 0)
504 /* an empty barrier - all done */
507 bio->bi_opf &= ~REQ_PREFLUSH;
508 mddev->pers->make_request(mddev, bio);
512 EXPORT_SYMBOL(md_flush_request);
514 static inline struct mddev *mddev_get(struct mddev *mddev)
516 atomic_inc(&mddev->active);
520 static void mddev_delayed_delete(struct work_struct *ws);
522 static void mddev_put(struct mddev *mddev)
524 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
526 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
527 mddev->ctime == 0 && !mddev->hold_active) {
528 /* Array is not configured at all, and not held active,
530 list_del_init(&mddev->all_mddevs);
533 * Call queue_work inside the spinlock so that
534 * flush_workqueue() after mddev_find will succeed in waiting
535 * for the work to be done.
537 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
538 queue_work(md_misc_wq, &mddev->del_work);
540 spin_unlock(&all_mddevs_lock);
543 static void md_safemode_timeout(struct timer_list *t);
545 void mddev_init(struct mddev *mddev)
547 kobject_init(&mddev->kobj, &md_ktype);
548 mutex_init(&mddev->open_mutex);
549 mutex_init(&mddev->reconfig_mutex);
550 mutex_init(&mddev->bitmap_info.mutex);
551 INIT_LIST_HEAD(&mddev->disks);
552 INIT_LIST_HEAD(&mddev->all_mddevs);
553 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
554 atomic_set(&mddev->active, 1);
555 atomic_set(&mddev->openers, 0);
556 atomic_set(&mddev->active_io, 0);
557 spin_lock_init(&mddev->lock);
558 atomic_set(&mddev->flush_pending, 0);
559 init_waitqueue_head(&mddev->sb_wait);
560 init_waitqueue_head(&mddev->recovery_wait);
561 mddev->reshape_position = MaxSector;
562 mddev->reshape_backwards = 0;
563 mddev->last_sync_action = "none";
564 mddev->resync_min = 0;
565 mddev->resync_max = MaxSector;
566 mddev->level = LEVEL_NONE;
568 EXPORT_SYMBOL_GPL(mddev_init);
570 static struct mddev *mddev_find(dev_t unit)
572 struct mddev *mddev, *new = NULL;
574 if (unit && MAJOR(unit) != MD_MAJOR)
575 unit &= ~((1<<MdpMinorShift)-1);
578 spin_lock(&all_mddevs_lock);
581 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
582 if (mddev->unit == unit) {
584 spin_unlock(&all_mddevs_lock);
590 list_add(&new->all_mddevs, &all_mddevs);
591 spin_unlock(&all_mddevs_lock);
592 new->hold_active = UNTIL_IOCTL;
596 /* find an unused unit number */
597 static int next_minor = 512;
598 int start = next_minor;
602 dev = MKDEV(MD_MAJOR, next_minor);
604 if (next_minor > MINORMASK)
606 if (next_minor == start) {
607 /* Oh dear, all in use. */
608 spin_unlock(&all_mddevs_lock);
614 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
615 if (mddev->unit == dev) {
621 new->md_minor = MINOR(dev);
622 new->hold_active = UNTIL_STOP;
623 list_add(&new->all_mddevs, &all_mddevs);
624 spin_unlock(&all_mddevs_lock);
627 spin_unlock(&all_mddevs_lock);
629 new = kzalloc(sizeof(*new), GFP_KERNEL);
634 if (MAJOR(unit) == MD_MAJOR)
635 new->md_minor = MINOR(unit);
637 new->md_minor = MINOR(unit) >> MdpMinorShift;
644 static struct attribute_group md_redundancy_group;
646 void mddev_unlock(struct mddev *mddev)
648 if (mddev->to_remove) {
649 /* These cannot be removed under reconfig_mutex as
650 * an access to the files will try to take reconfig_mutex
651 * while holding the file unremovable, which leads to
653 * So hold set sysfs_active while the remove in happeing,
654 * and anything else which might set ->to_remove or my
655 * otherwise change the sysfs namespace will fail with
656 * -EBUSY if sysfs_active is still set.
657 * We set sysfs_active under reconfig_mutex and elsewhere
658 * test it under the same mutex to ensure its correct value
661 struct attribute_group *to_remove = mddev->to_remove;
662 mddev->to_remove = NULL;
663 mddev->sysfs_active = 1;
664 mutex_unlock(&mddev->reconfig_mutex);
666 if (mddev->kobj.sd) {
667 if (to_remove != &md_redundancy_group)
668 sysfs_remove_group(&mddev->kobj, to_remove);
669 if (mddev->pers == NULL ||
670 mddev->pers->sync_request == NULL) {
671 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
672 if (mddev->sysfs_action)
673 sysfs_put(mddev->sysfs_action);
674 mddev->sysfs_action = NULL;
677 mddev->sysfs_active = 0;
679 mutex_unlock(&mddev->reconfig_mutex);
681 /* As we've dropped the mutex we need a spinlock to
682 * make sure the thread doesn't disappear
684 spin_lock(&pers_lock);
685 md_wakeup_thread(mddev->thread);
686 wake_up(&mddev->sb_wait);
687 spin_unlock(&pers_lock);
689 EXPORT_SYMBOL_GPL(mddev_unlock);
691 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
693 struct md_rdev *rdev;
695 rdev_for_each_rcu(rdev, mddev)
696 if (rdev->desc_nr == nr)
701 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
703 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
705 struct md_rdev *rdev;
707 rdev_for_each(rdev, mddev)
708 if (rdev->bdev->bd_dev == dev)
714 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
716 struct md_rdev *rdev;
718 rdev_for_each_rcu(rdev, mddev)
719 if (rdev->bdev->bd_dev == dev)
724 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
726 static struct md_personality *find_pers(int level, char *clevel)
728 struct md_personality *pers;
729 list_for_each_entry(pers, &pers_list, list) {
730 if (level != LEVEL_NONE && pers->level == level)
732 if (strcmp(pers->name, clevel)==0)
738 /* return the offset of the super block in 512byte sectors */
739 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
741 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
742 return MD_NEW_SIZE_SECTORS(num_sectors);
745 static int alloc_disk_sb(struct md_rdev *rdev)
747 rdev->sb_page = alloc_page(GFP_KERNEL);
753 void md_rdev_clear(struct md_rdev *rdev)
756 put_page(rdev->sb_page);
758 rdev->sb_page = NULL;
763 put_page(rdev->bb_page);
764 rdev->bb_page = NULL;
766 badblocks_exit(&rdev->badblocks);
768 EXPORT_SYMBOL_GPL(md_rdev_clear);
770 static void super_written(struct bio *bio)
772 struct md_rdev *rdev = bio->bi_private;
773 struct mddev *mddev = rdev->mddev;
775 if (bio->bi_status) {
776 pr_err("md: super_written gets error=%d\n", bio->bi_status);
777 md_error(mddev, rdev);
778 if (!test_bit(Faulty, &rdev->flags)
779 && (bio->bi_opf & MD_FAILFAST)) {
780 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
781 set_bit(LastDev, &rdev->flags);
784 clear_bit(LastDev, &rdev->flags);
786 if (atomic_dec_and_test(&mddev->pending_writes))
787 wake_up(&mddev->sb_wait);
788 rdev_dec_pending(rdev, mddev);
792 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
793 sector_t sector, int size, struct page *page)
795 /* write first size bytes of page to sector of rdev
796 * Increment mddev->pending_writes before returning
797 * and decrement it on completion, waking up sb_wait
798 * if zero is reached.
799 * If an error occurred, call md_error
807 if (test_bit(Faulty, &rdev->flags))
810 bio = md_bio_alloc_sync(mddev);
812 atomic_inc(&rdev->nr_pending);
814 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
815 bio->bi_iter.bi_sector = sector;
816 bio_add_page(bio, page, size, 0);
817 bio->bi_private = rdev;
818 bio->bi_end_io = super_written;
820 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
821 test_bit(FailFast, &rdev->flags) &&
822 !test_bit(LastDev, &rdev->flags))
824 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
826 atomic_inc(&mddev->pending_writes);
830 int md_super_wait(struct mddev *mddev)
832 /* wait for all superblock writes that were scheduled to complete */
833 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
834 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
839 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
840 struct page *page, int op, int op_flags, bool metadata_op)
842 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
845 if (metadata_op && rdev->meta_bdev)
846 bio_set_dev(bio, rdev->meta_bdev);
848 bio_set_dev(bio, rdev->bdev);
849 bio_set_op_attrs(bio, op, op_flags);
851 bio->bi_iter.bi_sector = sector + rdev->sb_start;
852 else if (rdev->mddev->reshape_position != MaxSector &&
853 (rdev->mddev->reshape_backwards ==
854 (sector >= rdev->mddev->reshape_position)))
855 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
857 bio->bi_iter.bi_sector = sector + rdev->data_offset;
858 bio_add_page(bio, page, size, 0);
860 submit_bio_wait(bio);
862 ret = !bio->bi_status;
866 EXPORT_SYMBOL_GPL(sync_page_io);
868 static int read_disk_sb(struct md_rdev *rdev, int size)
870 char b[BDEVNAME_SIZE];
875 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
881 pr_err("md: disabled device %s, could not read superblock.\n",
882 bdevname(rdev->bdev,b));
886 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
888 return sb1->set_uuid0 == sb2->set_uuid0 &&
889 sb1->set_uuid1 == sb2->set_uuid1 &&
890 sb1->set_uuid2 == sb2->set_uuid2 &&
891 sb1->set_uuid3 == sb2->set_uuid3;
894 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
897 mdp_super_t *tmp1, *tmp2;
899 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
900 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
902 if (!tmp1 || !tmp2) {
911 * nr_disks is not constant
916 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
923 static u32 md_csum_fold(u32 csum)
925 csum = (csum & 0xffff) + (csum >> 16);
926 return (csum & 0xffff) + (csum >> 16);
929 static unsigned int calc_sb_csum(mdp_super_t *sb)
932 u32 *sb32 = (u32*)sb;
934 unsigned int disk_csum, csum;
936 disk_csum = sb->sb_csum;
939 for (i = 0; i < MD_SB_BYTES/4 ; i++)
941 csum = (newcsum & 0xffffffff) + (newcsum>>32);
944 /* This used to use csum_partial, which was wrong for several
945 * reasons including that different results are returned on
946 * different architectures. It isn't critical that we get exactly
947 * the same return value as before (we always csum_fold before
948 * testing, and that removes any differences). However as we
949 * know that csum_partial always returned a 16bit value on
950 * alphas, do a fold to maximise conformity to previous behaviour.
952 sb->sb_csum = md_csum_fold(disk_csum);
954 sb->sb_csum = disk_csum;
960 * Handle superblock details.
961 * We want to be able to handle multiple superblock formats
962 * so we have a common interface to them all, and an array of
963 * different handlers.
964 * We rely on user-space to write the initial superblock, and support
965 * reading and updating of superblocks.
966 * Interface methods are:
967 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
968 * loads and validates a superblock on dev.
969 * if refdev != NULL, compare superblocks on both devices
971 * 0 - dev has a superblock that is compatible with refdev
972 * 1 - dev has a superblock that is compatible and newer than refdev
973 * so dev should be used as the refdev in future
974 * -EINVAL superblock incompatible or invalid
975 * -othererror e.g. -EIO
977 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
978 * Verify that dev is acceptable into mddev.
979 * The first time, mddev->raid_disks will be 0, and data from
980 * dev should be merged in. Subsequent calls check that dev
981 * is new enough. Return 0 or -EINVAL
983 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
984 * Update the superblock for rdev with data in mddev
985 * This does not write to disc.
991 struct module *owner;
992 int (*load_super)(struct md_rdev *rdev,
993 struct md_rdev *refdev,
995 int (*validate_super)(struct mddev *mddev,
996 struct md_rdev *rdev);
997 void (*sync_super)(struct mddev *mddev,
998 struct md_rdev *rdev);
999 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1000 sector_t num_sectors);
1001 int (*allow_new_offset)(struct md_rdev *rdev,
1002 unsigned long long new_offset);
1006 * Check that the given mddev has no bitmap.
1008 * This function is called from the run method of all personalities that do not
1009 * support bitmaps. It prints an error message and returns non-zero if mddev
1010 * has a bitmap. Otherwise, it returns 0.
1013 int md_check_no_bitmap(struct mddev *mddev)
1015 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1017 pr_warn("%s: bitmaps are not supported for %s\n",
1018 mdname(mddev), mddev->pers->name);
1021 EXPORT_SYMBOL(md_check_no_bitmap);
1024 * load_super for 0.90.0
1026 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1028 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1033 * Calculate the position of the superblock (512byte sectors),
1034 * it's at the end of the disk.
1036 * It also happens to be a multiple of 4Kb.
1038 rdev->sb_start = calc_dev_sboffset(rdev);
1040 ret = read_disk_sb(rdev, MD_SB_BYTES);
1046 bdevname(rdev->bdev, b);
1047 sb = page_address(rdev->sb_page);
1049 if (sb->md_magic != MD_SB_MAGIC) {
1050 pr_warn("md: invalid raid superblock magic on %s\n", b);
1054 if (sb->major_version != 0 ||
1055 sb->minor_version < 90 ||
1056 sb->minor_version > 91) {
1057 pr_warn("Bad version number %d.%d on %s\n",
1058 sb->major_version, sb->minor_version, b);
1062 if (sb->raid_disks <= 0)
1065 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1066 pr_warn("md: invalid superblock checksum on %s\n", b);
1070 rdev->preferred_minor = sb->md_minor;
1071 rdev->data_offset = 0;
1072 rdev->new_data_offset = 0;
1073 rdev->sb_size = MD_SB_BYTES;
1074 rdev->badblocks.shift = -1;
1076 if (sb->level == LEVEL_MULTIPATH)
1079 rdev->desc_nr = sb->this_disk.number;
1085 mdp_super_t *refsb = page_address(refdev->sb_page);
1086 if (!md_uuid_equal(refsb, sb)) {
1087 pr_warn("md: %s has different UUID to %s\n",
1088 b, bdevname(refdev->bdev,b2));
1091 if (!md_sb_equal(refsb, sb)) {
1092 pr_warn("md: %s has same UUID but different superblock to %s\n",
1093 b, bdevname(refdev->bdev, b2));
1097 ev2 = md_event(refsb);
1103 rdev->sectors = rdev->sb_start;
1104 /* Limit to 4TB as metadata cannot record more than that.
1105 * (not needed for Linear and RAID0 as metadata doesn't
1108 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1109 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1111 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1112 /* "this cannot possibly happen" ... */
1120 * validate_super for 0.90.0
1122 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1125 mdp_super_t *sb = page_address(rdev->sb_page);
1126 __u64 ev1 = md_event(sb);
1128 rdev->raid_disk = -1;
1129 clear_bit(Faulty, &rdev->flags);
1130 clear_bit(In_sync, &rdev->flags);
1131 clear_bit(Bitmap_sync, &rdev->flags);
1132 clear_bit(WriteMostly, &rdev->flags);
1134 if (mddev->raid_disks == 0) {
1135 mddev->major_version = 0;
1136 mddev->minor_version = sb->minor_version;
1137 mddev->patch_version = sb->patch_version;
1138 mddev->external = 0;
1139 mddev->chunk_sectors = sb->chunk_size >> 9;
1140 mddev->ctime = sb->ctime;
1141 mddev->utime = sb->utime;
1142 mddev->level = sb->level;
1143 mddev->clevel[0] = 0;
1144 mddev->layout = sb->layout;
1145 mddev->raid_disks = sb->raid_disks;
1146 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1147 mddev->events = ev1;
1148 mddev->bitmap_info.offset = 0;
1149 mddev->bitmap_info.space = 0;
1150 /* bitmap can use 60 K after the 4K superblocks */
1151 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1152 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1153 mddev->reshape_backwards = 0;
1155 if (mddev->minor_version >= 91) {
1156 mddev->reshape_position = sb->reshape_position;
1157 mddev->delta_disks = sb->delta_disks;
1158 mddev->new_level = sb->new_level;
1159 mddev->new_layout = sb->new_layout;
1160 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1161 if (mddev->delta_disks < 0)
1162 mddev->reshape_backwards = 1;
1164 mddev->reshape_position = MaxSector;
1165 mddev->delta_disks = 0;
1166 mddev->new_level = mddev->level;
1167 mddev->new_layout = mddev->layout;
1168 mddev->new_chunk_sectors = mddev->chunk_sectors;
1171 if (sb->state & (1<<MD_SB_CLEAN))
1172 mddev->recovery_cp = MaxSector;
1174 if (sb->events_hi == sb->cp_events_hi &&
1175 sb->events_lo == sb->cp_events_lo) {
1176 mddev->recovery_cp = sb->recovery_cp;
1178 mddev->recovery_cp = 0;
1181 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1182 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1183 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1184 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1186 mddev->max_disks = MD_SB_DISKS;
1188 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1189 mddev->bitmap_info.file == NULL) {
1190 mddev->bitmap_info.offset =
1191 mddev->bitmap_info.default_offset;
1192 mddev->bitmap_info.space =
1193 mddev->bitmap_info.default_space;
1196 } else if (mddev->pers == NULL) {
1197 /* Insist on good event counter while assembling, except
1198 * for spares (which don't need an event count) */
1200 if (sb->disks[rdev->desc_nr].state & (
1201 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1202 if (ev1 < mddev->events)
1204 } else if (mddev->bitmap) {
1205 /* if adding to array with a bitmap, then we can accept an
1206 * older device ... but not too old.
1208 if (ev1 < mddev->bitmap->events_cleared)
1210 if (ev1 < mddev->events)
1211 set_bit(Bitmap_sync, &rdev->flags);
1213 if (ev1 < mddev->events)
1214 /* just a hot-add of a new device, leave raid_disk at -1 */
1218 if (mddev->level != LEVEL_MULTIPATH) {
1219 desc = sb->disks + rdev->desc_nr;
1221 if (desc->state & (1<<MD_DISK_FAULTY))
1222 set_bit(Faulty, &rdev->flags);
1223 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1224 desc->raid_disk < mddev->raid_disks */) {
1225 set_bit(In_sync, &rdev->flags);
1226 rdev->raid_disk = desc->raid_disk;
1227 rdev->saved_raid_disk = desc->raid_disk;
1228 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1229 /* active but not in sync implies recovery up to
1230 * reshape position. We don't know exactly where
1231 * that is, so set to zero for now */
1232 if (mddev->minor_version >= 91) {
1233 rdev->recovery_offset = 0;
1234 rdev->raid_disk = desc->raid_disk;
1237 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1238 set_bit(WriteMostly, &rdev->flags);
1239 if (desc->state & (1<<MD_DISK_FAILFAST))
1240 set_bit(FailFast, &rdev->flags);
1241 } else /* MULTIPATH are always insync */
1242 set_bit(In_sync, &rdev->flags);
1247 * sync_super for 0.90.0
1249 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1252 struct md_rdev *rdev2;
1253 int next_spare = mddev->raid_disks;
1255 /* make rdev->sb match mddev data..
1258 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1259 * 3/ any empty disks < next_spare become removed
1261 * disks[0] gets initialised to REMOVED because
1262 * we cannot be sure from other fields if it has
1263 * been initialised or not.
1266 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1268 rdev->sb_size = MD_SB_BYTES;
1270 sb = page_address(rdev->sb_page);
1272 memset(sb, 0, sizeof(*sb));
1274 sb->md_magic = MD_SB_MAGIC;
1275 sb->major_version = mddev->major_version;
1276 sb->patch_version = mddev->patch_version;
1277 sb->gvalid_words = 0; /* ignored */
1278 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1279 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1280 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1281 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1283 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1284 sb->level = mddev->level;
1285 sb->size = mddev->dev_sectors / 2;
1286 sb->raid_disks = mddev->raid_disks;
1287 sb->md_minor = mddev->md_minor;
1288 sb->not_persistent = 0;
1289 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1291 sb->events_hi = (mddev->events>>32);
1292 sb->events_lo = (u32)mddev->events;
1294 if (mddev->reshape_position == MaxSector)
1295 sb->minor_version = 90;
1297 sb->minor_version = 91;
1298 sb->reshape_position = mddev->reshape_position;
1299 sb->new_level = mddev->new_level;
1300 sb->delta_disks = mddev->delta_disks;
1301 sb->new_layout = mddev->new_layout;
1302 sb->new_chunk = mddev->new_chunk_sectors << 9;
1304 mddev->minor_version = sb->minor_version;
1307 sb->recovery_cp = mddev->recovery_cp;
1308 sb->cp_events_hi = (mddev->events>>32);
1309 sb->cp_events_lo = (u32)mddev->events;
1310 if (mddev->recovery_cp == MaxSector)
1311 sb->state = (1<< MD_SB_CLEAN);
1313 sb->recovery_cp = 0;
1315 sb->layout = mddev->layout;
1316 sb->chunk_size = mddev->chunk_sectors << 9;
1318 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1319 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1321 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1322 rdev_for_each(rdev2, mddev) {
1325 int is_active = test_bit(In_sync, &rdev2->flags);
1327 if (rdev2->raid_disk >= 0 &&
1328 sb->minor_version >= 91)
1329 /* we have nowhere to store the recovery_offset,
1330 * but if it is not below the reshape_position,
1331 * we can piggy-back on that.
1334 if (rdev2->raid_disk < 0 ||
1335 test_bit(Faulty, &rdev2->flags))
1338 desc_nr = rdev2->raid_disk;
1340 desc_nr = next_spare++;
1341 rdev2->desc_nr = desc_nr;
1342 d = &sb->disks[rdev2->desc_nr];
1344 d->number = rdev2->desc_nr;
1345 d->major = MAJOR(rdev2->bdev->bd_dev);
1346 d->minor = MINOR(rdev2->bdev->bd_dev);
1348 d->raid_disk = rdev2->raid_disk;
1350 d->raid_disk = rdev2->desc_nr; /* compatibility */
1351 if (test_bit(Faulty, &rdev2->flags))
1352 d->state = (1<<MD_DISK_FAULTY);
1353 else if (is_active) {
1354 d->state = (1<<MD_DISK_ACTIVE);
1355 if (test_bit(In_sync, &rdev2->flags))
1356 d->state |= (1<<MD_DISK_SYNC);
1364 if (test_bit(WriteMostly, &rdev2->flags))
1365 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1366 if (test_bit(FailFast, &rdev2->flags))
1367 d->state |= (1<<MD_DISK_FAILFAST);
1369 /* now set the "removed" and "faulty" bits on any missing devices */
1370 for (i=0 ; i < mddev->raid_disks ; i++) {
1371 mdp_disk_t *d = &sb->disks[i];
1372 if (d->state == 0 && d->number == 0) {
1375 d->state = (1<<MD_DISK_REMOVED);
1376 d->state |= (1<<MD_DISK_FAULTY);
1380 sb->nr_disks = nr_disks;
1381 sb->active_disks = active;
1382 sb->working_disks = working;
1383 sb->failed_disks = failed;
1384 sb->spare_disks = spare;
1386 sb->this_disk = sb->disks[rdev->desc_nr];
1387 sb->sb_csum = calc_sb_csum(sb);
1391 * rdev_size_change for 0.90.0
1393 static unsigned long long
1394 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1396 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1397 return 0; /* component must fit device */
1398 if (rdev->mddev->bitmap_info.offset)
1399 return 0; /* can't move bitmap */
1400 rdev->sb_start = calc_dev_sboffset(rdev);
1401 if (!num_sectors || num_sectors > rdev->sb_start)
1402 num_sectors = rdev->sb_start;
1403 /* Limit to 4TB as metadata cannot record more than that.
1404 * 4TB == 2^32 KB, or 2*2^32 sectors.
1406 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1407 num_sectors = (sector_t)(2ULL << 32) - 2;
1409 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1411 } while (md_super_wait(rdev->mddev) < 0);
1416 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1418 /* non-zero offset changes not possible with v0.90 */
1419 return new_offset == 0;
1423 * version 1 superblock
1426 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1430 unsigned long long newcsum;
1431 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1432 __le32 *isuper = (__le32*)sb;
1434 disk_csum = sb->sb_csum;
1437 for (; size >= 4; size -= 4)
1438 newcsum += le32_to_cpu(*isuper++);
1441 newcsum += le16_to_cpu(*(__le16*) isuper);
1443 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1444 sb->sb_csum = disk_csum;
1445 return cpu_to_le32(csum);
1448 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1450 struct mdp_superblock_1 *sb;
1454 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1458 * Calculate the position of the superblock in 512byte sectors.
1459 * It is always aligned to a 4K boundary and
1460 * depeding on minor_version, it can be:
1461 * 0: At least 8K, but less than 12K, from end of device
1462 * 1: At start of device
1463 * 2: 4K from start of device.
1465 switch(minor_version) {
1467 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1469 sb_start &= ~(sector_t)(4*2-1);
1480 rdev->sb_start = sb_start;
1482 /* superblock is rarely larger than 1K, but it can be larger,
1483 * and it is safe to read 4k, so we do that
1485 ret = read_disk_sb(rdev, 4096);
1486 if (ret) return ret;
1488 sb = page_address(rdev->sb_page);
1490 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1491 sb->major_version != cpu_to_le32(1) ||
1492 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1493 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1494 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1497 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1498 pr_warn("md: invalid superblock checksum on %s\n",
1499 bdevname(rdev->bdev,b));
1502 if (le64_to_cpu(sb->data_size) < 10) {
1503 pr_warn("md: data_size too small on %s\n",
1504 bdevname(rdev->bdev,b));
1509 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1510 /* Some padding is non-zero, might be a new feature */
1513 rdev->preferred_minor = 0xffff;
1514 rdev->data_offset = le64_to_cpu(sb->data_offset);
1515 rdev->new_data_offset = rdev->data_offset;
1516 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1517 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1518 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1519 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1521 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1522 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1523 if (rdev->sb_size & bmask)
1524 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1527 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1530 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1533 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1536 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1538 if (!rdev->bb_page) {
1539 rdev->bb_page = alloc_page(GFP_KERNEL);
1543 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1544 rdev->badblocks.count == 0) {
1545 /* need to load the bad block list.
1546 * Currently we limit it to one page.
1552 int sectors = le16_to_cpu(sb->bblog_size);
1553 if (sectors > (PAGE_SIZE / 512))
1555 offset = le32_to_cpu(sb->bblog_offset);
1558 bb_sector = (long long)offset;
1559 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1560 rdev->bb_page, REQ_OP_READ, 0, true))
1562 bbp = (__le64 *)page_address(rdev->bb_page);
1563 rdev->badblocks.shift = sb->bblog_shift;
1564 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1565 u64 bb = le64_to_cpu(*bbp);
1566 int count = bb & (0x3ff);
1567 u64 sector = bb >> 10;
1568 sector <<= sb->bblog_shift;
1569 count <<= sb->bblog_shift;
1572 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1575 } else if (sb->bblog_offset != 0)
1576 rdev->badblocks.shift = 0;
1578 if ((le32_to_cpu(sb->feature_map) &
1579 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1580 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1581 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1582 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1589 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1591 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1592 sb->level != refsb->level ||
1593 sb->layout != refsb->layout ||
1594 sb->chunksize != refsb->chunksize) {
1595 pr_warn("md: %s has strangely different superblock to %s\n",
1596 bdevname(rdev->bdev,b),
1597 bdevname(refdev->bdev,b2));
1600 ev1 = le64_to_cpu(sb->events);
1601 ev2 = le64_to_cpu(refsb->events);
1608 if (minor_version) {
1609 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1610 sectors -= rdev->data_offset;
1612 sectors = rdev->sb_start;
1613 if (sectors < le64_to_cpu(sb->data_size))
1615 rdev->sectors = le64_to_cpu(sb->data_size);
1619 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1621 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1622 __u64 ev1 = le64_to_cpu(sb->events);
1624 rdev->raid_disk = -1;
1625 clear_bit(Faulty, &rdev->flags);
1626 clear_bit(In_sync, &rdev->flags);
1627 clear_bit(Bitmap_sync, &rdev->flags);
1628 clear_bit(WriteMostly, &rdev->flags);
1630 if (mddev->raid_disks == 0) {
1631 mddev->major_version = 1;
1632 mddev->patch_version = 0;
1633 mddev->external = 0;
1634 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1635 mddev->ctime = le64_to_cpu(sb->ctime);
1636 mddev->utime = le64_to_cpu(sb->utime);
1637 mddev->level = le32_to_cpu(sb->level);
1638 mddev->clevel[0] = 0;
1639 mddev->layout = le32_to_cpu(sb->layout);
1640 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1641 mddev->dev_sectors = le64_to_cpu(sb->size);
1642 mddev->events = ev1;
1643 mddev->bitmap_info.offset = 0;
1644 mddev->bitmap_info.space = 0;
1645 /* Default location for bitmap is 1K after superblock
1646 * using 3K - total of 4K
1648 mddev->bitmap_info.default_offset = 1024 >> 9;
1649 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1650 mddev->reshape_backwards = 0;
1652 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1653 memcpy(mddev->uuid, sb->set_uuid, 16);
1655 mddev->max_disks = (4096-256)/2;
1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1658 mddev->bitmap_info.file == NULL) {
1659 mddev->bitmap_info.offset =
1660 (__s32)le32_to_cpu(sb->bitmap_offset);
1661 /* Metadata doesn't record how much space is available.
1662 * For 1.0, we assume we can use up to the superblock
1663 * if before, else to 4K beyond superblock.
1664 * For others, assume no change is possible.
1666 if (mddev->minor_version > 0)
1667 mddev->bitmap_info.space = 0;
1668 else if (mddev->bitmap_info.offset > 0)
1669 mddev->bitmap_info.space =
1670 8 - mddev->bitmap_info.offset;
1672 mddev->bitmap_info.space =
1673 -mddev->bitmap_info.offset;
1676 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1677 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1678 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1679 mddev->new_level = le32_to_cpu(sb->new_level);
1680 mddev->new_layout = le32_to_cpu(sb->new_layout);
1681 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1682 if (mddev->delta_disks < 0 ||
1683 (mddev->delta_disks == 0 &&
1684 (le32_to_cpu(sb->feature_map)
1685 & MD_FEATURE_RESHAPE_BACKWARDS)))
1686 mddev->reshape_backwards = 1;
1688 mddev->reshape_position = MaxSector;
1689 mddev->delta_disks = 0;
1690 mddev->new_level = mddev->level;
1691 mddev->new_layout = mddev->layout;
1692 mddev->new_chunk_sectors = mddev->chunk_sectors;
1695 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1696 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1698 if (le32_to_cpu(sb->feature_map) &
1699 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1700 if (le32_to_cpu(sb->feature_map) &
1701 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1703 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1704 (le32_to_cpu(sb->feature_map) &
1705 MD_FEATURE_MULTIPLE_PPLS))
1707 set_bit(MD_HAS_PPL, &mddev->flags);
1709 } else if (mddev->pers == NULL) {
1710 /* Insist of good event counter while assembling, except for
1711 * spares (which don't need an event count) */
1713 if (rdev->desc_nr >= 0 &&
1714 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1715 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1716 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1717 if (ev1 < mddev->events)
1719 } else if (mddev->bitmap) {
1720 /* If adding to array with a bitmap, then we can accept an
1721 * older device, but not too old.
1723 if (ev1 < mddev->bitmap->events_cleared)
1725 if (ev1 < mddev->events)
1726 set_bit(Bitmap_sync, &rdev->flags);
1728 if (ev1 < mddev->events)
1729 /* just a hot-add of a new device, leave raid_disk at -1 */
1732 if (mddev->level != LEVEL_MULTIPATH) {
1734 if (rdev->desc_nr < 0 ||
1735 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1736 role = MD_DISK_ROLE_SPARE;
1739 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1741 case MD_DISK_ROLE_SPARE: /* spare */
1743 case MD_DISK_ROLE_FAULTY: /* faulty */
1744 set_bit(Faulty, &rdev->flags);
1746 case MD_DISK_ROLE_JOURNAL: /* journal device */
1747 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1748 /* journal device without journal feature */
1749 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1752 set_bit(Journal, &rdev->flags);
1753 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1754 rdev->raid_disk = 0;
1757 rdev->saved_raid_disk = role;
1758 if ((le32_to_cpu(sb->feature_map) &
1759 MD_FEATURE_RECOVERY_OFFSET)) {
1760 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1761 if (!(le32_to_cpu(sb->feature_map) &
1762 MD_FEATURE_RECOVERY_BITMAP))
1763 rdev->saved_raid_disk = -1;
1765 set_bit(In_sync, &rdev->flags);
1766 rdev->raid_disk = role;
1769 if (sb->devflags & WriteMostly1)
1770 set_bit(WriteMostly, &rdev->flags);
1771 if (sb->devflags & FailFast1)
1772 set_bit(FailFast, &rdev->flags);
1773 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1774 set_bit(Replacement, &rdev->flags);
1775 } else /* MULTIPATH are always insync */
1776 set_bit(In_sync, &rdev->flags);
1781 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1783 struct mdp_superblock_1 *sb;
1784 struct md_rdev *rdev2;
1786 /* make rdev->sb match mddev and rdev data. */
1788 sb = page_address(rdev->sb_page);
1790 sb->feature_map = 0;
1792 sb->recovery_offset = cpu_to_le64(0);
1793 memset(sb->pad3, 0, sizeof(sb->pad3));
1795 sb->utime = cpu_to_le64((__u64)mddev->utime);
1796 sb->events = cpu_to_le64(mddev->events);
1798 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1799 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1800 sb->resync_offset = cpu_to_le64(MaxSector);
1802 sb->resync_offset = cpu_to_le64(0);
1804 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1806 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1807 sb->size = cpu_to_le64(mddev->dev_sectors);
1808 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1809 sb->level = cpu_to_le32(mddev->level);
1810 sb->layout = cpu_to_le32(mddev->layout);
1811 if (test_bit(FailFast, &rdev->flags))
1812 sb->devflags |= FailFast1;
1814 sb->devflags &= ~FailFast1;
1816 if (test_bit(WriteMostly, &rdev->flags))
1817 sb->devflags |= WriteMostly1;
1819 sb->devflags &= ~WriteMostly1;
1820 sb->data_offset = cpu_to_le64(rdev->data_offset);
1821 sb->data_size = cpu_to_le64(rdev->sectors);
1823 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1824 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1825 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1828 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1829 !test_bit(In_sync, &rdev->flags)) {
1831 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1832 sb->recovery_offset =
1833 cpu_to_le64(rdev->recovery_offset);
1834 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1836 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1838 /* Note: recovery_offset and journal_tail share space */
1839 if (test_bit(Journal, &rdev->flags))
1840 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1841 if (test_bit(Replacement, &rdev->flags))
1843 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1845 if (mddev->reshape_position != MaxSector) {
1846 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1847 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1848 sb->new_layout = cpu_to_le32(mddev->new_layout);
1849 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1850 sb->new_level = cpu_to_le32(mddev->new_level);
1851 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1852 if (mddev->delta_disks == 0 &&
1853 mddev->reshape_backwards)
1855 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1856 if (rdev->new_data_offset != rdev->data_offset) {
1858 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1859 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1860 - rdev->data_offset));
1864 if (mddev_is_clustered(mddev))
1865 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1867 if (rdev->badblocks.count == 0)
1868 /* Nothing to do for bad blocks*/ ;
1869 else if (sb->bblog_offset == 0)
1870 /* Cannot record bad blocks on this device */
1871 md_error(mddev, rdev);
1873 struct badblocks *bb = &rdev->badblocks;
1874 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
1876 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1881 seq = read_seqbegin(&bb->lock);
1883 memset(bbp, 0xff, PAGE_SIZE);
1885 for (i = 0 ; i < bb->count ; i++) {
1886 u64 internal_bb = p[i];
1887 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1888 | BB_LEN(internal_bb));
1889 bbp[i] = cpu_to_le64(store_bb);
1892 if (read_seqretry(&bb->lock, seq))
1895 bb->sector = (rdev->sb_start +
1896 (int)le32_to_cpu(sb->bblog_offset));
1897 bb->size = le16_to_cpu(sb->bblog_size);
1902 rdev_for_each(rdev2, mddev)
1903 if (rdev2->desc_nr+1 > max_dev)
1904 max_dev = rdev2->desc_nr+1;
1906 if (max_dev > le32_to_cpu(sb->max_dev)) {
1908 sb->max_dev = cpu_to_le32(max_dev);
1909 rdev->sb_size = max_dev * 2 + 256;
1910 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1911 if (rdev->sb_size & bmask)
1912 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1914 max_dev = le32_to_cpu(sb->max_dev);
1916 for (i=0; i<max_dev;i++)
1917 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1919 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1920 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1922 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
1923 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
1925 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
1927 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
1928 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
1929 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
1932 rdev_for_each(rdev2, mddev) {
1934 if (test_bit(Faulty, &rdev2->flags))
1935 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1936 else if (test_bit(In_sync, &rdev2->flags))
1937 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1938 else if (test_bit(Journal, &rdev2->flags))
1939 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1940 else if (rdev2->raid_disk >= 0)
1941 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1943 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1946 sb->sb_csum = calc_sb_1_csum(sb);
1949 static unsigned long long
1950 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1952 struct mdp_superblock_1 *sb;
1953 sector_t max_sectors;
1954 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1955 return 0; /* component must fit device */
1956 if (rdev->data_offset != rdev->new_data_offset)
1957 return 0; /* too confusing */
1958 if (rdev->sb_start < rdev->data_offset) {
1959 /* minor versions 1 and 2; superblock before data */
1960 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1961 max_sectors -= rdev->data_offset;
1962 if (!num_sectors || num_sectors > max_sectors)
1963 num_sectors = max_sectors;
1964 } else if (rdev->mddev->bitmap_info.offset) {
1965 /* minor version 0 with bitmap we can't move */
1968 /* minor version 0; superblock after data */
1970 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1971 sb_start &= ~(sector_t)(4*2 - 1);
1972 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1973 if (!num_sectors || num_sectors > max_sectors)
1974 num_sectors = max_sectors;
1975 rdev->sb_start = sb_start;
1977 sb = page_address(rdev->sb_page);
1978 sb->data_size = cpu_to_le64(num_sectors);
1979 sb->super_offset = cpu_to_le64(rdev->sb_start);
1980 sb->sb_csum = calc_sb_1_csum(sb);
1982 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1984 } while (md_super_wait(rdev->mddev) < 0);
1990 super_1_allow_new_offset(struct md_rdev *rdev,
1991 unsigned long long new_offset)
1993 /* All necessary checks on new >= old have been done */
1994 struct bitmap *bitmap;
1995 if (new_offset >= rdev->data_offset)
1998 /* with 1.0 metadata, there is no metadata to tread on
1999 * so we can always move back */
2000 if (rdev->mddev->minor_version == 0)
2003 /* otherwise we must be sure not to step on
2004 * any metadata, so stay:
2005 * 36K beyond start of superblock
2006 * beyond end of badblocks
2007 * beyond write-intent bitmap
2009 if (rdev->sb_start + (32+4)*2 > new_offset)
2011 bitmap = rdev->mddev->bitmap;
2012 if (bitmap && !rdev->mddev->bitmap_info.file &&
2013 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2014 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2016 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2022 static struct super_type super_types[] = {
2025 .owner = THIS_MODULE,
2026 .load_super = super_90_load,
2027 .validate_super = super_90_validate,
2028 .sync_super = super_90_sync,
2029 .rdev_size_change = super_90_rdev_size_change,
2030 .allow_new_offset = super_90_allow_new_offset,
2034 .owner = THIS_MODULE,
2035 .load_super = super_1_load,
2036 .validate_super = super_1_validate,
2037 .sync_super = super_1_sync,
2038 .rdev_size_change = super_1_rdev_size_change,
2039 .allow_new_offset = super_1_allow_new_offset,
2043 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2045 if (mddev->sync_super) {
2046 mddev->sync_super(mddev, rdev);
2050 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2052 super_types[mddev->major_version].sync_super(mddev, rdev);
2055 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2057 struct md_rdev *rdev, *rdev2;
2060 rdev_for_each_rcu(rdev, mddev1) {
2061 if (test_bit(Faulty, &rdev->flags) ||
2062 test_bit(Journal, &rdev->flags) ||
2063 rdev->raid_disk == -1)
2065 rdev_for_each_rcu(rdev2, mddev2) {
2066 if (test_bit(Faulty, &rdev2->flags) ||
2067 test_bit(Journal, &rdev2->flags) ||
2068 rdev2->raid_disk == -1)
2070 if (rdev->bdev->bd_contains ==
2071 rdev2->bdev->bd_contains) {
2081 static LIST_HEAD(pending_raid_disks);
2084 * Try to register data integrity profile for an mddev
2086 * This is called when an array is started and after a disk has been kicked
2087 * from the array. It only succeeds if all working and active component devices
2088 * are integrity capable with matching profiles.
2090 int md_integrity_register(struct mddev *mddev)
2092 struct md_rdev *rdev, *reference = NULL;
2094 if (list_empty(&mddev->disks))
2095 return 0; /* nothing to do */
2096 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2097 return 0; /* shouldn't register, or already is */
2098 rdev_for_each(rdev, mddev) {
2099 /* skip spares and non-functional disks */
2100 if (test_bit(Faulty, &rdev->flags))
2102 if (rdev->raid_disk < 0)
2105 /* Use the first rdev as the reference */
2109 /* does this rdev's profile match the reference profile? */
2110 if (blk_integrity_compare(reference->bdev->bd_disk,
2111 rdev->bdev->bd_disk) < 0)
2114 if (!reference || !bdev_get_integrity(reference->bdev))
2117 * All component devices are integrity capable and have matching
2118 * profiles, register the common profile for the md device.
2120 blk_integrity_register(mddev->gendisk,
2121 bdev_get_integrity(reference->bdev));
2123 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2124 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2125 pr_err("md: failed to create integrity pool for %s\n",
2131 EXPORT_SYMBOL(md_integrity_register);
2134 * Attempt to add an rdev, but only if it is consistent with the current
2137 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2139 struct blk_integrity *bi_mddev;
2140 char name[BDEVNAME_SIZE];
2142 if (!mddev->gendisk)
2145 bi_mddev = blk_get_integrity(mddev->gendisk);
2147 if (!bi_mddev) /* nothing to do */
2150 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2151 pr_err("%s: incompatible integrity profile for %s\n",
2152 mdname(mddev), bdevname(rdev->bdev, name));
2158 EXPORT_SYMBOL(md_integrity_add_rdev);
2160 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2162 char b[BDEVNAME_SIZE];
2166 /* prevent duplicates */
2167 if (find_rdev(mddev, rdev->bdev->bd_dev))
2170 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2174 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2175 if (!test_bit(Journal, &rdev->flags) &&
2177 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2179 /* Cannot change size, so fail
2180 * If mddev->level <= 0, then we don't care
2181 * about aligning sizes (e.g. linear)
2183 if (mddev->level > 0)
2186 mddev->dev_sectors = rdev->sectors;
2189 /* Verify rdev->desc_nr is unique.
2190 * If it is -1, assign a free number, else
2191 * check number is not in use
2194 if (rdev->desc_nr < 0) {
2197 choice = mddev->raid_disks;
2198 while (md_find_rdev_nr_rcu(mddev, choice))
2200 rdev->desc_nr = choice;
2202 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2208 if (!test_bit(Journal, &rdev->flags) &&
2209 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2210 pr_warn("md: %s: array is limited to %d devices\n",
2211 mdname(mddev), mddev->max_disks);
2214 bdevname(rdev->bdev,b);
2215 strreplace(b, '/', '!');
2217 rdev->mddev = mddev;
2218 pr_debug("md: bind<%s>\n", b);
2220 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2223 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2224 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2225 /* failure here is OK */;
2226 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2228 list_add_rcu(&rdev->same_set, &mddev->disks);
2229 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2231 /* May as well allow recovery to be retried once */
2232 mddev->recovery_disabled++;
2237 pr_warn("md: failed to register dev-%s for %s\n",
2242 static void md_delayed_delete(struct work_struct *ws)
2244 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2245 kobject_del(&rdev->kobj);
2246 kobject_put(&rdev->kobj);
2249 static void unbind_rdev_from_array(struct md_rdev *rdev)
2251 char b[BDEVNAME_SIZE];
2253 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2254 list_del_rcu(&rdev->same_set);
2255 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2257 sysfs_remove_link(&rdev->kobj, "block");
2258 sysfs_put(rdev->sysfs_state);
2259 rdev->sysfs_state = NULL;
2260 rdev->badblocks.count = 0;
2261 /* We need to delay this, otherwise we can deadlock when
2262 * writing to 'remove' to "dev/state". We also need
2263 * to delay it due to rcu usage.
2266 INIT_WORK(&rdev->del_work, md_delayed_delete);
2267 kobject_get(&rdev->kobj);
2268 queue_work(md_misc_wq, &rdev->del_work);
2272 * prevent the device from being mounted, repartitioned or
2273 * otherwise reused by a RAID array (or any other kernel
2274 * subsystem), by bd_claiming the device.
2276 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2279 struct block_device *bdev;
2280 char b[BDEVNAME_SIZE];
2282 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2283 shared ? (struct md_rdev *)lock_rdev : rdev);
2285 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2286 return PTR_ERR(bdev);
2292 static void unlock_rdev(struct md_rdev *rdev)
2294 struct block_device *bdev = rdev->bdev;
2296 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2299 void md_autodetect_dev(dev_t dev);
2301 static void export_rdev(struct md_rdev *rdev)
2303 char b[BDEVNAME_SIZE];
2305 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2306 md_rdev_clear(rdev);
2308 if (test_bit(AutoDetected, &rdev->flags))
2309 md_autodetect_dev(rdev->bdev->bd_dev);
2312 kobject_put(&rdev->kobj);
2315 void md_kick_rdev_from_array(struct md_rdev *rdev)
2317 unbind_rdev_from_array(rdev);
2320 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2322 static void export_array(struct mddev *mddev)
2324 struct md_rdev *rdev;
2326 while (!list_empty(&mddev->disks)) {
2327 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2329 md_kick_rdev_from_array(rdev);
2331 mddev->raid_disks = 0;
2332 mddev->major_version = 0;
2335 static bool set_in_sync(struct mddev *mddev)
2337 lockdep_assert_held(&mddev->lock);
2338 if (!mddev->in_sync) {
2339 mddev->sync_checkers++;
2340 spin_unlock(&mddev->lock);
2341 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2342 spin_lock(&mddev->lock);
2343 if (!mddev->in_sync &&
2344 percpu_ref_is_zero(&mddev->writes_pending)) {
2347 * Ensure ->in_sync is visible before we clear
2351 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2352 sysfs_notify_dirent_safe(mddev->sysfs_state);
2354 if (--mddev->sync_checkers == 0)
2355 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2357 if (mddev->safemode == 1)
2358 mddev->safemode = 0;
2359 return mddev->in_sync;
2362 static void sync_sbs(struct mddev *mddev, int nospares)
2364 /* Update each superblock (in-memory image), but
2365 * if we are allowed to, skip spares which already
2366 * have the right event counter, or have one earlier
2367 * (which would mean they aren't being marked as dirty
2368 * with the rest of the array)
2370 struct md_rdev *rdev;
2371 rdev_for_each(rdev, mddev) {
2372 if (rdev->sb_events == mddev->events ||
2374 rdev->raid_disk < 0 &&
2375 rdev->sb_events+1 == mddev->events)) {
2376 /* Don't update this superblock */
2377 rdev->sb_loaded = 2;
2379 sync_super(mddev, rdev);
2380 rdev->sb_loaded = 1;
2385 static bool does_sb_need_changing(struct mddev *mddev)
2387 struct md_rdev *rdev;
2388 struct mdp_superblock_1 *sb;
2391 /* Find a good rdev */
2392 rdev_for_each(rdev, mddev)
2393 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2396 /* No good device found. */
2400 sb = page_address(rdev->sb_page);
2401 /* Check if a device has become faulty or a spare become active */
2402 rdev_for_each(rdev, mddev) {
2403 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2404 /* Device activated? */
2405 if (role == 0xffff && rdev->raid_disk >=0 &&
2406 !test_bit(Faulty, &rdev->flags))
2408 /* Device turned faulty? */
2409 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2413 /* Check if any mddev parameters have changed */
2414 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2415 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2416 (mddev->layout != le32_to_cpu(sb->layout)) ||
2417 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2418 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2424 void md_update_sb(struct mddev *mddev, int force_change)
2426 struct md_rdev *rdev;
2429 int any_badblocks_changed = 0;
2434 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2439 if (mddev_is_clustered(mddev)) {
2440 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2442 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2444 ret = md_cluster_ops->metadata_update_start(mddev);
2445 /* Has someone else has updated the sb */
2446 if (!does_sb_need_changing(mddev)) {
2448 md_cluster_ops->metadata_update_cancel(mddev);
2449 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2450 BIT(MD_SB_CHANGE_DEVS) |
2451 BIT(MD_SB_CHANGE_CLEAN));
2457 * First make sure individual recovery_offsets are correct
2458 * curr_resync_completed can only be used during recovery.
2459 * During reshape/resync it might use array-addresses rather
2460 * that device addresses.
2462 rdev_for_each(rdev, mddev) {
2463 if (rdev->raid_disk >= 0 &&
2464 mddev->delta_disks >= 0 &&
2465 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2466 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2467 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2468 !test_bit(Journal, &rdev->flags) &&
2469 !test_bit(In_sync, &rdev->flags) &&
2470 mddev->curr_resync_completed > rdev->recovery_offset)
2471 rdev->recovery_offset = mddev->curr_resync_completed;
2474 if (!mddev->persistent) {
2475 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2476 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2477 if (!mddev->external) {
2478 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2479 rdev_for_each(rdev, mddev) {
2480 if (rdev->badblocks.changed) {
2481 rdev->badblocks.changed = 0;
2482 ack_all_badblocks(&rdev->badblocks);
2483 md_error(mddev, rdev);
2485 clear_bit(Blocked, &rdev->flags);
2486 clear_bit(BlockedBadBlocks, &rdev->flags);
2487 wake_up(&rdev->blocked_wait);
2490 wake_up(&mddev->sb_wait);
2494 spin_lock(&mddev->lock);
2496 mddev->utime = ktime_get_real_seconds();
2498 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2500 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2501 /* just a clean<-> dirty transition, possibly leave spares alone,
2502 * though if events isn't the right even/odd, we will have to do
2508 if (mddev->degraded)
2509 /* If the array is degraded, then skipping spares is both
2510 * dangerous and fairly pointless.
2511 * Dangerous because a device that was removed from the array
2512 * might have a event_count that still looks up-to-date,
2513 * so it can be re-added without a resync.
2514 * Pointless because if there are any spares to skip,
2515 * then a recovery will happen and soon that array won't
2516 * be degraded any more and the spare can go back to sleep then.
2520 sync_req = mddev->in_sync;
2522 /* If this is just a dirty<->clean transition, and the array is clean
2523 * and 'events' is odd, we can roll back to the previous clean state */
2525 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2526 && mddev->can_decrease_events
2527 && mddev->events != 1) {
2529 mddev->can_decrease_events = 0;
2531 /* otherwise we have to go forward and ... */
2533 mddev->can_decrease_events = nospares;
2537 * This 64-bit counter should never wrap.
2538 * Either we are in around ~1 trillion A.C., assuming
2539 * 1 reboot per second, or we have a bug...
2541 WARN_ON(mddev->events == 0);
2543 rdev_for_each(rdev, mddev) {
2544 if (rdev->badblocks.changed)
2545 any_badblocks_changed++;
2546 if (test_bit(Faulty, &rdev->flags))
2547 set_bit(FaultRecorded, &rdev->flags);
2550 sync_sbs(mddev, nospares);
2551 spin_unlock(&mddev->lock);
2553 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2554 mdname(mddev), mddev->in_sync);
2557 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2559 md_bitmap_update_sb(mddev->bitmap);
2560 rdev_for_each(rdev, mddev) {
2561 char b[BDEVNAME_SIZE];
2563 if (rdev->sb_loaded != 1)
2564 continue; /* no noise on spare devices */
2566 if (!test_bit(Faulty, &rdev->flags)) {
2567 md_super_write(mddev,rdev,
2568 rdev->sb_start, rdev->sb_size,
2570 pr_debug("md: (write) %s's sb offset: %llu\n",
2571 bdevname(rdev->bdev, b),
2572 (unsigned long long)rdev->sb_start);
2573 rdev->sb_events = mddev->events;
2574 if (rdev->badblocks.size) {
2575 md_super_write(mddev, rdev,
2576 rdev->badblocks.sector,
2577 rdev->badblocks.size << 9,
2579 rdev->badblocks.size = 0;
2583 pr_debug("md: %s (skipping faulty)\n",
2584 bdevname(rdev->bdev, b));
2586 if (mddev->level == LEVEL_MULTIPATH)
2587 /* only need to write one superblock... */
2590 if (md_super_wait(mddev) < 0)
2592 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2594 if (mddev_is_clustered(mddev) && ret == 0)
2595 md_cluster_ops->metadata_update_finish(mddev);
2597 if (mddev->in_sync != sync_req ||
2598 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2599 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2600 /* have to write it out again */
2602 wake_up(&mddev->sb_wait);
2603 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2604 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2606 rdev_for_each(rdev, mddev) {
2607 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2608 clear_bit(Blocked, &rdev->flags);
2610 if (any_badblocks_changed)
2611 ack_all_badblocks(&rdev->badblocks);
2612 clear_bit(BlockedBadBlocks, &rdev->flags);
2613 wake_up(&rdev->blocked_wait);
2616 EXPORT_SYMBOL(md_update_sb);
2618 static int add_bound_rdev(struct md_rdev *rdev)
2620 struct mddev *mddev = rdev->mddev;
2622 bool add_journal = test_bit(Journal, &rdev->flags);
2624 if (!mddev->pers->hot_remove_disk || add_journal) {
2625 /* If there is hot_add_disk but no hot_remove_disk
2626 * then added disks for geometry changes,
2627 * and should be added immediately.
2629 super_types[mddev->major_version].
2630 validate_super(mddev, rdev);
2632 mddev_suspend(mddev);
2633 err = mddev->pers->hot_add_disk(mddev, rdev);
2635 mddev_resume(mddev);
2637 md_kick_rdev_from_array(rdev);
2641 sysfs_notify_dirent_safe(rdev->sysfs_state);
2643 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2644 if (mddev->degraded)
2645 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2646 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2647 md_new_event(mddev);
2648 md_wakeup_thread(mddev->thread);
2652 /* words written to sysfs files may, or may not, be \n terminated.
2653 * We want to accept with case. For this we use cmd_match.
2655 static int cmd_match(const char *cmd, const char *str)
2657 /* See if cmd, written into a sysfs file, matches
2658 * str. They must either be the same, or cmd can
2659 * have a trailing newline
2661 while (*cmd && *str && *cmd == *str) {
2672 struct rdev_sysfs_entry {
2673 struct attribute attr;
2674 ssize_t (*show)(struct md_rdev *, char *);
2675 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2679 state_show(struct md_rdev *rdev, char *page)
2683 unsigned long flags = READ_ONCE(rdev->flags);
2685 if (test_bit(Faulty, &flags) ||
2686 (!test_bit(ExternalBbl, &flags) &&
2687 rdev->badblocks.unacked_exist))
2688 len += sprintf(page+len, "faulty%s", sep);
2689 if (test_bit(In_sync, &flags))
2690 len += sprintf(page+len, "in_sync%s", sep);
2691 if (test_bit(Journal, &flags))
2692 len += sprintf(page+len, "journal%s", sep);
2693 if (test_bit(WriteMostly, &flags))
2694 len += sprintf(page+len, "write_mostly%s", sep);
2695 if (test_bit(Blocked, &flags) ||
2696 (rdev->badblocks.unacked_exist
2697 && !test_bit(Faulty, &flags)))
2698 len += sprintf(page+len, "blocked%s", sep);
2699 if (!test_bit(Faulty, &flags) &&
2700 !test_bit(Journal, &flags) &&
2701 !test_bit(In_sync, &flags))
2702 len += sprintf(page+len, "spare%s", sep);
2703 if (test_bit(WriteErrorSeen, &flags))
2704 len += sprintf(page+len, "write_error%s", sep);
2705 if (test_bit(WantReplacement, &flags))
2706 len += sprintf(page+len, "want_replacement%s", sep);
2707 if (test_bit(Replacement, &flags))
2708 len += sprintf(page+len, "replacement%s", sep);
2709 if (test_bit(ExternalBbl, &flags))
2710 len += sprintf(page+len, "external_bbl%s", sep);
2711 if (test_bit(FailFast, &flags))
2712 len += sprintf(page+len, "failfast%s", sep);
2717 return len+sprintf(page+len, "\n");
2721 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2724 * faulty - simulates an error
2725 * remove - disconnects the device
2726 * writemostly - sets write_mostly
2727 * -writemostly - clears write_mostly
2728 * blocked - sets the Blocked flags
2729 * -blocked - clears the Blocked and possibly simulates an error
2730 * insync - sets Insync providing device isn't active
2731 * -insync - clear Insync for a device with a slot assigned,
2732 * so that it gets rebuilt based on bitmap
2733 * write_error - sets WriteErrorSeen
2734 * -write_error - clears WriteErrorSeen
2735 * {,-}failfast - set/clear FailFast
2738 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2739 md_error(rdev->mddev, rdev);
2740 if (test_bit(Faulty, &rdev->flags))
2744 } else if (cmd_match(buf, "remove")) {
2745 if (rdev->mddev->pers) {
2746 clear_bit(Blocked, &rdev->flags);
2747 remove_and_add_spares(rdev->mddev, rdev);
2749 if (rdev->raid_disk >= 0)
2752 struct mddev *mddev = rdev->mddev;
2754 if (mddev_is_clustered(mddev))
2755 err = md_cluster_ops->remove_disk(mddev, rdev);
2758 md_kick_rdev_from_array(rdev);
2760 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2761 md_wakeup_thread(mddev->thread);
2763 md_new_event(mddev);
2766 } else if (cmd_match(buf, "writemostly")) {
2767 set_bit(WriteMostly, &rdev->flags);
2769 } else if (cmd_match(buf, "-writemostly")) {
2770 clear_bit(WriteMostly, &rdev->flags);
2772 } else if (cmd_match(buf, "blocked")) {
2773 set_bit(Blocked, &rdev->flags);
2775 } else if (cmd_match(buf, "-blocked")) {
2776 if (!test_bit(Faulty, &rdev->flags) &&
2777 !test_bit(ExternalBbl, &rdev->flags) &&
2778 rdev->badblocks.unacked_exist) {
2779 /* metadata handler doesn't understand badblocks,
2780 * so we need to fail the device
2782 md_error(rdev->mddev, rdev);
2784 clear_bit(Blocked, &rdev->flags);
2785 clear_bit(BlockedBadBlocks, &rdev->flags);
2786 wake_up(&rdev->blocked_wait);
2787 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2788 md_wakeup_thread(rdev->mddev->thread);
2791 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2792 set_bit(In_sync, &rdev->flags);
2794 } else if (cmd_match(buf, "failfast")) {
2795 set_bit(FailFast, &rdev->flags);
2797 } else if (cmd_match(buf, "-failfast")) {
2798 clear_bit(FailFast, &rdev->flags);
2800 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2801 !test_bit(Journal, &rdev->flags)) {
2802 if (rdev->mddev->pers == NULL) {
2803 clear_bit(In_sync, &rdev->flags);
2804 rdev->saved_raid_disk = rdev->raid_disk;
2805 rdev->raid_disk = -1;
2808 } else if (cmd_match(buf, "write_error")) {
2809 set_bit(WriteErrorSeen, &rdev->flags);
2811 } else if (cmd_match(buf, "-write_error")) {
2812 clear_bit(WriteErrorSeen, &rdev->flags);
2814 } else if (cmd_match(buf, "want_replacement")) {
2815 /* Any non-spare device that is not a replacement can
2816 * become want_replacement at any time, but we then need to
2817 * check if recovery is needed.
2819 if (rdev->raid_disk >= 0 &&
2820 !test_bit(Journal, &rdev->flags) &&
2821 !test_bit(Replacement, &rdev->flags))
2822 set_bit(WantReplacement, &rdev->flags);
2823 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2824 md_wakeup_thread(rdev->mddev->thread);
2826 } else if (cmd_match(buf, "-want_replacement")) {
2827 /* Clearing 'want_replacement' is always allowed.
2828 * Once replacements starts it is too late though.
2831 clear_bit(WantReplacement, &rdev->flags);
2832 } else if (cmd_match(buf, "replacement")) {
2833 /* Can only set a device as a replacement when array has not
2834 * yet been started. Once running, replacement is automatic
2835 * from spares, or by assigning 'slot'.
2837 if (rdev->mddev->pers)
2840 set_bit(Replacement, &rdev->flags);
2843 } else if (cmd_match(buf, "-replacement")) {
2844 /* Similarly, can only clear Replacement before start */
2845 if (rdev->mddev->pers)
2848 clear_bit(Replacement, &rdev->flags);
2851 } else if (cmd_match(buf, "re-add")) {
2852 if (!rdev->mddev->pers)
2854 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2855 rdev->saved_raid_disk >= 0) {
2856 /* clear_bit is performed _after_ all the devices
2857 * have their local Faulty bit cleared. If any writes
2858 * happen in the meantime in the local node, they
2859 * will land in the local bitmap, which will be synced
2860 * by this node eventually
2862 if (!mddev_is_clustered(rdev->mddev) ||
2863 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2864 clear_bit(Faulty, &rdev->flags);
2865 err = add_bound_rdev(rdev);
2869 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2870 set_bit(ExternalBbl, &rdev->flags);
2871 rdev->badblocks.shift = 0;
2873 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2874 clear_bit(ExternalBbl, &rdev->flags);
2878 sysfs_notify_dirent_safe(rdev->sysfs_state);
2879 return err ? err : len;
2881 static struct rdev_sysfs_entry rdev_state =
2882 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2885 errors_show(struct md_rdev *rdev, char *page)
2887 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2891 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2896 rv = kstrtouint(buf, 10, &n);
2899 atomic_set(&rdev->corrected_errors, n);
2902 static struct rdev_sysfs_entry rdev_errors =
2903 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2906 slot_show(struct md_rdev *rdev, char *page)
2908 if (test_bit(Journal, &rdev->flags))
2909 return sprintf(page, "journal\n");
2910 else if (rdev->raid_disk < 0)
2911 return sprintf(page, "none\n");
2913 return sprintf(page, "%d\n", rdev->raid_disk);
2917 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2922 if (test_bit(Journal, &rdev->flags))
2924 if (strncmp(buf, "none", 4)==0)
2927 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2931 if (rdev->mddev->pers && slot == -1) {
2932 /* Setting 'slot' on an active array requires also
2933 * updating the 'rd%d' link, and communicating
2934 * with the personality with ->hot_*_disk.
2935 * For now we only support removing
2936 * failed/spare devices. This normally happens automatically,
2937 * but not when the metadata is externally managed.
2939 if (rdev->raid_disk == -1)
2941 /* personality does all needed checks */
2942 if (rdev->mddev->pers->hot_remove_disk == NULL)
2944 clear_bit(Blocked, &rdev->flags);
2945 remove_and_add_spares(rdev->mddev, rdev);
2946 if (rdev->raid_disk >= 0)
2948 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2949 md_wakeup_thread(rdev->mddev->thread);
2950 } else if (rdev->mddev->pers) {
2951 /* Activating a spare .. or possibly reactivating
2952 * if we ever get bitmaps working here.
2956 if (rdev->raid_disk != -1)
2959 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2962 if (rdev->mddev->pers->hot_add_disk == NULL)
2965 if (slot >= rdev->mddev->raid_disks &&
2966 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2969 rdev->raid_disk = slot;
2970 if (test_bit(In_sync, &rdev->flags))
2971 rdev->saved_raid_disk = slot;
2973 rdev->saved_raid_disk = -1;
2974 clear_bit(In_sync, &rdev->flags);
2975 clear_bit(Bitmap_sync, &rdev->flags);
2976 err = rdev->mddev->pers->
2977 hot_add_disk(rdev->mddev, rdev);
2979 rdev->raid_disk = -1;
2982 sysfs_notify_dirent_safe(rdev->sysfs_state);
2983 if (sysfs_link_rdev(rdev->mddev, rdev))
2984 /* failure here is OK */;
2985 /* don't wakeup anyone, leave that to userspace. */
2987 if (slot >= rdev->mddev->raid_disks &&
2988 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2990 rdev->raid_disk = slot;
2991 /* assume it is working */
2992 clear_bit(Faulty, &rdev->flags);
2993 clear_bit(WriteMostly, &rdev->flags);
2994 set_bit(In_sync, &rdev->flags);
2995 sysfs_notify_dirent_safe(rdev->sysfs_state);
3000 static struct rdev_sysfs_entry rdev_slot =
3001 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3004 offset_show(struct md_rdev *rdev, char *page)
3006 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3010 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3012 unsigned long long offset;
3013 if (kstrtoull(buf, 10, &offset) < 0)
3015 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3017 if (rdev->sectors && rdev->mddev->external)
3018 /* Must set offset before size, so overlap checks
3021 rdev->data_offset = offset;
3022 rdev->new_data_offset = offset;
3026 static struct rdev_sysfs_entry rdev_offset =
3027 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3029 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3031 return sprintf(page, "%llu\n",
3032 (unsigned long long)rdev->new_data_offset);
3035 static ssize_t new_offset_store(struct md_rdev *rdev,
3036 const char *buf, size_t len)
3038 unsigned long long new_offset;
3039 struct mddev *mddev = rdev->mddev;
3041 if (kstrtoull(buf, 10, &new_offset) < 0)
3044 if (mddev->sync_thread ||
3045 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3047 if (new_offset == rdev->data_offset)
3048 /* reset is always permitted */
3050 else if (new_offset > rdev->data_offset) {
3051 /* must not push array size beyond rdev_sectors */
3052 if (new_offset - rdev->data_offset
3053 + mddev->dev_sectors > rdev->sectors)
3056 /* Metadata worries about other space details. */
3058 /* decreasing the offset is inconsistent with a backwards
3061 if (new_offset < rdev->data_offset &&
3062 mddev->reshape_backwards)
3064 /* Increasing offset is inconsistent with forwards
3065 * reshape. reshape_direction should be set to
3066 * 'backwards' first.
3068 if (new_offset > rdev->data_offset &&
3069 !mddev->reshape_backwards)
3072 if (mddev->pers && mddev->persistent &&
3073 !super_types[mddev->major_version]
3074 .allow_new_offset(rdev, new_offset))
3076 rdev->new_data_offset = new_offset;
3077 if (new_offset > rdev->data_offset)
3078 mddev->reshape_backwards = 1;
3079 else if (new_offset < rdev->data_offset)
3080 mddev->reshape_backwards = 0;
3084 static struct rdev_sysfs_entry rdev_new_offset =
3085 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3088 rdev_size_show(struct md_rdev *rdev, char *page)
3090 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3093 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3095 /* check if two start/length pairs overlap */
3103 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3105 unsigned long long blocks;
3108 if (kstrtoull(buf, 10, &blocks) < 0)
3111 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3112 return -EINVAL; /* sector conversion overflow */
3115 if (new != blocks * 2)
3116 return -EINVAL; /* unsigned long long to sector_t overflow */
3123 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3125 struct mddev *my_mddev = rdev->mddev;
3126 sector_t oldsectors = rdev->sectors;
3129 if (test_bit(Journal, &rdev->flags))
3131 if (strict_blocks_to_sectors(buf, §ors) < 0)
3133 if (rdev->data_offset != rdev->new_data_offset)
3134 return -EINVAL; /* too confusing */
3135 if (my_mddev->pers && rdev->raid_disk >= 0) {
3136 if (my_mddev->persistent) {
3137 sectors = super_types[my_mddev->major_version].
3138 rdev_size_change(rdev, sectors);
3141 } else if (!sectors)
3142 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3144 if (!my_mddev->pers->resize)
3145 /* Cannot change size for RAID0 or Linear etc */
3148 if (sectors < my_mddev->dev_sectors)
3149 return -EINVAL; /* component must fit device */
3151 rdev->sectors = sectors;
3152 if (sectors > oldsectors && my_mddev->external) {
3153 /* Need to check that all other rdevs with the same
3154 * ->bdev do not overlap. 'rcu' is sufficient to walk
3155 * the rdev lists safely.
3156 * This check does not provide a hard guarantee, it
3157 * just helps avoid dangerous mistakes.
3159 struct mddev *mddev;
3161 struct list_head *tmp;
3164 for_each_mddev(mddev, tmp) {
3165 struct md_rdev *rdev2;
3167 rdev_for_each(rdev2, mddev)
3168 if (rdev->bdev == rdev2->bdev &&
3170 overlaps(rdev->data_offset, rdev->sectors,
3183 /* Someone else could have slipped in a size
3184 * change here, but doing so is just silly.
3185 * We put oldsectors back because we *know* it is
3186 * safe, and trust userspace not to race with
3189 rdev->sectors = oldsectors;
3196 static struct rdev_sysfs_entry rdev_size =
3197 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3199 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3201 unsigned long long recovery_start = rdev->recovery_offset;
3203 if (test_bit(In_sync, &rdev->flags) ||
3204 recovery_start == MaxSector)
3205 return sprintf(page, "none\n");
3207 return sprintf(page, "%llu\n", recovery_start);
3210 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3212 unsigned long long recovery_start;
3214 if (cmd_match(buf, "none"))
3215 recovery_start = MaxSector;
3216 else if (kstrtoull(buf, 10, &recovery_start))
3219 if (rdev->mddev->pers &&
3220 rdev->raid_disk >= 0)
3223 rdev->recovery_offset = recovery_start;
3224 if (recovery_start == MaxSector)
3225 set_bit(In_sync, &rdev->flags);
3227 clear_bit(In_sync, &rdev->flags);
3231 static struct rdev_sysfs_entry rdev_recovery_start =
3232 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3234 /* sysfs access to bad-blocks list.
3235 * We present two files.
3236 * 'bad-blocks' lists sector numbers and lengths of ranges that
3237 * are recorded as bad. The list is truncated to fit within
3238 * the one-page limit of sysfs.
3239 * Writing "sector length" to this file adds an acknowledged
3241 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3242 * been acknowledged. Writing to this file adds bad blocks
3243 * without acknowledging them. This is largely for testing.
3245 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3247 return badblocks_show(&rdev->badblocks, page, 0);
3249 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3251 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3252 /* Maybe that ack was all we needed */
3253 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3254 wake_up(&rdev->blocked_wait);
3257 static struct rdev_sysfs_entry rdev_bad_blocks =
3258 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3260 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3262 return badblocks_show(&rdev->badblocks, page, 1);
3264 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3266 return badblocks_store(&rdev->badblocks, page, len, 1);
3268 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3269 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3272 ppl_sector_show(struct md_rdev *rdev, char *page)
3274 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3278 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3280 unsigned long long sector;
3282 if (kstrtoull(buf, 10, §or) < 0)
3284 if (sector != (sector_t)sector)
3287 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3288 rdev->raid_disk >= 0)
3291 if (rdev->mddev->persistent) {
3292 if (rdev->mddev->major_version == 0)
3294 if ((sector > rdev->sb_start &&
3295 sector - rdev->sb_start > S16_MAX) ||
3296 (sector < rdev->sb_start &&
3297 rdev->sb_start - sector > -S16_MIN))
3299 rdev->ppl.offset = sector - rdev->sb_start;
3300 } else if (!rdev->mddev->external) {
3303 rdev->ppl.sector = sector;
3307 static struct rdev_sysfs_entry rdev_ppl_sector =
3308 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3311 ppl_size_show(struct md_rdev *rdev, char *page)
3313 return sprintf(page, "%u\n", rdev->ppl.size);
3317 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3321 if (kstrtouint(buf, 10, &size) < 0)
3324 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3325 rdev->raid_disk >= 0)
3328 if (rdev->mddev->persistent) {
3329 if (rdev->mddev->major_version == 0)
3333 } else if (!rdev->mddev->external) {
3336 rdev->ppl.size = size;
3340 static struct rdev_sysfs_entry rdev_ppl_size =
3341 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3343 static struct attribute *rdev_default_attrs[] = {
3348 &rdev_new_offset.attr,
3350 &rdev_recovery_start.attr,
3351 &rdev_bad_blocks.attr,
3352 &rdev_unack_bad_blocks.attr,
3353 &rdev_ppl_sector.attr,
3354 &rdev_ppl_size.attr,
3358 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3360 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3361 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3367 return entry->show(rdev, page);
3371 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3372 const char *page, size_t length)
3374 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3375 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3377 struct mddev *mddev = rdev->mddev;
3381 if (!capable(CAP_SYS_ADMIN))
3383 rv = mddev ? mddev_lock(mddev): -EBUSY;
3385 if (rdev->mddev == NULL)
3388 rv = entry->store(rdev, page, length);
3389 mddev_unlock(mddev);
3394 static void rdev_free(struct kobject *ko)
3396 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3399 static const struct sysfs_ops rdev_sysfs_ops = {
3400 .show = rdev_attr_show,
3401 .store = rdev_attr_store,
3403 static struct kobj_type rdev_ktype = {
3404 .release = rdev_free,
3405 .sysfs_ops = &rdev_sysfs_ops,
3406 .default_attrs = rdev_default_attrs,
3409 int md_rdev_init(struct md_rdev *rdev)
3412 rdev->saved_raid_disk = -1;
3413 rdev->raid_disk = -1;
3415 rdev->data_offset = 0;
3416 rdev->new_data_offset = 0;
3417 rdev->sb_events = 0;
3418 rdev->last_read_error = 0;
3419 rdev->sb_loaded = 0;
3420 rdev->bb_page = NULL;
3421 atomic_set(&rdev->nr_pending, 0);
3422 atomic_set(&rdev->read_errors, 0);
3423 atomic_set(&rdev->corrected_errors, 0);
3425 INIT_LIST_HEAD(&rdev->same_set);
3426 init_waitqueue_head(&rdev->blocked_wait);
3428 /* Add space to store bad block list.
3429 * This reserves the space even on arrays where it cannot
3430 * be used - I wonder if that matters
3432 return badblocks_init(&rdev->badblocks, 0);
3434 EXPORT_SYMBOL_GPL(md_rdev_init);
3436 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3438 * mark the device faulty if:
3440 * - the device is nonexistent (zero size)
3441 * - the device has no valid superblock
3443 * a faulty rdev _never_ has rdev->sb set.
3445 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3447 char b[BDEVNAME_SIZE];
3449 struct md_rdev *rdev;
3452 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3454 return ERR_PTR(-ENOMEM);
3456 err = md_rdev_init(rdev);
3459 err = alloc_disk_sb(rdev);
3463 err = lock_rdev(rdev, newdev, super_format == -2);
3467 kobject_init(&rdev->kobj, &rdev_ktype);
3469 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3471 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3472 bdevname(rdev->bdev,b));
3477 if (super_format >= 0) {
3478 err = super_types[super_format].
3479 load_super(rdev, NULL, super_minor);
3480 if (err == -EINVAL) {
3481 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3482 bdevname(rdev->bdev,b),
3483 super_format, super_minor);
3487 pr_warn("md: could not read %s's sb, not importing!\n",
3488 bdevname(rdev->bdev,b));
3498 md_rdev_clear(rdev);
3500 return ERR_PTR(err);
3504 * Check a full RAID array for plausibility
3507 static void analyze_sbs(struct mddev *mddev)
3510 struct md_rdev *rdev, *freshest, *tmp;
3511 char b[BDEVNAME_SIZE];
3514 rdev_for_each_safe(rdev, tmp, mddev)
3515 switch (super_types[mddev->major_version].
3516 load_super(rdev, freshest, mddev->minor_version)) {
3523 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3524 bdevname(rdev->bdev,b));
3525 md_kick_rdev_from_array(rdev);
3528 super_types[mddev->major_version].
3529 validate_super(mddev, freshest);
3532 rdev_for_each_safe(rdev, tmp, mddev) {
3533 if (mddev->max_disks &&
3534 (rdev->desc_nr >= mddev->max_disks ||
3535 i > mddev->max_disks)) {
3536 pr_warn("md: %s: %s: only %d devices permitted\n",
3537 mdname(mddev), bdevname(rdev->bdev, b),
3539 md_kick_rdev_from_array(rdev);
3542 if (rdev != freshest) {
3543 if (super_types[mddev->major_version].
3544 validate_super(mddev, rdev)) {
3545 pr_warn("md: kicking non-fresh %s from array!\n",
3546 bdevname(rdev->bdev,b));
3547 md_kick_rdev_from_array(rdev);
3551 if (mddev->level == LEVEL_MULTIPATH) {
3552 rdev->desc_nr = i++;
3553 rdev->raid_disk = rdev->desc_nr;
3554 set_bit(In_sync, &rdev->flags);
3555 } else if (rdev->raid_disk >=
3556 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3557 !test_bit(Journal, &rdev->flags)) {
3558 rdev->raid_disk = -1;
3559 clear_bit(In_sync, &rdev->flags);
3564 /* Read a fixed-point number.
3565 * Numbers in sysfs attributes should be in "standard" units where
3566 * possible, so time should be in seconds.
3567 * However we internally use a a much smaller unit such as
3568 * milliseconds or jiffies.
3569 * This function takes a decimal number with a possible fractional
3570 * component, and produces an integer which is the result of
3571 * multiplying that number by 10^'scale'.
3572 * all without any floating-point arithmetic.
3574 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3576 unsigned long result = 0;
3578 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3581 else if (decimals < scale) {
3584 result = result * 10 + value;
3596 while (decimals < scale) {
3605 safe_delay_show(struct mddev *mddev, char *page)
3607 int msec = (mddev->safemode_delay*1000)/HZ;
3608 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3611 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3615 if (mddev_is_clustered(mddev)) {
3616 pr_warn("md: Safemode is disabled for clustered mode\n");
3620 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3623 mddev->safemode_delay = 0;
3625 unsigned long old_delay = mddev->safemode_delay;
3626 unsigned long new_delay = (msec*HZ)/1000;
3630 mddev->safemode_delay = new_delay;
3631 if (new_delay < old_delay || old_delay == 0)
3632 mod_timer(&mddev->safemode_timer, jiffies+1);
3636 static struct md_sysfs_entry md_safe_delay =
3637 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3640 level_show(struct mddev *mddev, char *page)
3642 struct md_personality *p;
3644 spin_lock(&mddev->lock);
3647 ret = sprintf(page, "%s\n", p->name);
3648 else if (mddev->clevel[0])
3649 ret = sprintf(page, "%s\n", mddev->clevel);
3650 else if (mddev->level != LEVEL_NONE)
3651 ret = sprintf(page, "%d\n", mddev->level);
3654 spin_unlock(&mddev->lock);
3659 level_store(struct mddev *mddev, const char *buf, size_t len)
3664 struct md_personality *pers, *oldpers;
3666 void *priv, *oldpriv;
3667 struct md_rdev *rdev;
3669 if (slen == 0 || slen >= sizeof(clevel))
3672 rv = mddev_lock(mddev);
3676 if (mddev->pers == NULL) {
3677 strncpy(mddev->clevel, buf, slen);
3678 if (mddev->clevel[slen-1] == '\n')
3680 mddev->clevel[slen] = 0;
3681 mddev->level = LEVEL_NONE;
3689 /* request to change the personality. Need to ensure:
3690 * - array is not engaged in resync/recovery/reshape
3691 * - old personality can be suspended
3692 * - new personality will access other array.
3696 if (mddev->sync_thread ||
3697 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3698 mddev->reshape_position != MaxSector ||
3699 mddev->sysfs_active)
3703 if (!mddev->pers->quiesce) {
3704 pr_warn("md: %s: %s does not support online personality change\n",
3705 mdname(mddev), mddev->pers->name);
3709 /* Now find the new personality */
3710 strncpy(clevel, buf, slen);
3711 if (clevel[slen-1] == '\n')
3714 if (kstrtol(clevel, 10, &level))
3717 if (request_module("md-%s", clevel) != 0)
3718 request_module("md-level-%s", clevel);
3719 spin_lock(&pers_lock);
3720 pers = find_pers(level, clevel);
3721 if (!pers || !try_module_get(pers->owner)) {
3722 spin_unlock(&pers_lock);
3723 pr_warn("md: personality %s not loaded\n", clevel);
3727 spin_unlock(&pers_lock);
3729 if (pers == mddev->pers) {
3730 /* Nothing to do! */
3731 module_put(pers->owner);
3735 if (!pers->takeover) {
3736 module_put(pers->owner);
3737 pr_warn("md: %s: %s does not support personality takeover\n",
3738 mdname(mddev), clevel);
3743 rdev_for_each(rdev, mddev)
3744 rdev->new_raid_disk = rdev->raid_disk;
3746 /* ->takeover must set new_* and/or delta_disks
3747 * if it succeeds, and may set them when it fails.
3749 priv = pers->takeover(mddev);
3751 mddev->new_level = mddev->level;
3752 mddev->new_layout = mddev->layout;
3753 mddev->new_chunk_sectors = mddev->chunk_sectors;
3754 mddev->raid_disks -= mddev->delta_disks;
3755 mddev->delta_disks = 0;
3756 mddev->reshape_backwards = 0;
3757 module_put(pers->owner);
3758 pr_warn("md: %s: %s would not accept array\n",
3759 mdname(mddev), clevel);
3764 /* Looks like we have a winner */
3765 mddev_suspend(mddev);
3766 mddev_detach(mddev);
3768 spin_lock(&mddev->lock);
3769 oldpers = mddev->pers;
3770 oldpriv = mddev->private;
3772 mddev->private = priv;
3773 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3774 mddev->level = mddev->new_level;
3775 mddev->layout = mddev->new_layout;
3776 mddev->chunk_sectors = mddev->new_chunk_sectors;
3777 mddev->delta_disks = 0;
3778 mddev->reshape_backwards = 0;
3779 mddev->degraded = 0;
3780 spin_unlock(&mddev->lock);
3782 if (oldpers->sync_request == NULL &&
3784 /* We are converting from a no-redundancy array
3785 * to a redundancy array and metadata is managed
3786 * externally so we need to be sure that writes
3787 * won't block due to a need to transition
3789 * until external management is started.
3792 mddev->safemode_delay = 0;
3793 mddev->safemode = 0;
3796 oldpers->free(mddev, oldpriv);
3798 if (oldpers->sync_request == NULL &&
3799 pers->sync_request != NULL) {
3800 /* need to add the md_redundancy_group */
3801 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3802 pr_warn("md: cannot register extra attributes for %s\n",
3804 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3806 if (oldpers->sync_request != NULL &&
3807 pers->sync_request == NULL) {
3808 /* need to remove the md_redundancy_group */
3809 if (mddev->to_remove == NULL)
3810 mddev->to_remove = &md_redundancy_group;
3813 module_put(oldpers->owner);
3815 rdev_for_each(rdev, mddev) {
3816 if (rdev->raid_disk < 0)
3818 if (rdev->new_raid_disk >= mddev->raid_disks)
3819 rdev->new_raid_disk = -1;
3820 if (rdev->new_raid_disk == rdev->raid_disk)
3822 sysfs_unlink_rdev(mddev, rdev);
3824 rdev_for_each(rdev, mddev) {
3825 if (rdev->raid_disk < 0)
3827 if (rdev->new_raid_disk == rdev->raid_disk)
3829 rdev->raid_disk = rdev->new_raid_disk;
3830 if (rdev->raid_disk < 0)
3831 clear_bit(In_sync, &rdev->flags);
3833 if (sysfs_link_rdev(mddev, rdev))
3834 pr_warn("md: cannot register rd%d for %s after level change\n",
3835 rdev->raid_disk, mdname(mddev));
3839 if (pers->sync_request == NULL) {
3840 /* this is now an array without redundancy, so
3841 * it must always be in_sync
3844 del_timer_sync(&mddev->safemode_timer);
3846 blk_set_stacking_limits(&mddev->queue->limits);
3848 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3849 mddev_resume(mddev);
3851 md_update_sb(mddev, 1);
3852 sysfs_notify(&mddev->kobj, NULL, "level");
3853 md_new_event(mddev);
3856 mddev_unlock(mddev);
3860 static struct md_sysfs_entry md_level =
3861 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3864 layout_show(struct mddev *mddev, char *page)
3866 /* just a number, not meaningful for all levels */
3867 if (mddev->reshape_position != MaxSector &&
3868 mddev->layout != mddev->new_layout)
3869 return sprintf(page, "%d (%d)\n",
3870 mddev->new_layout, mddev->layout);
3871 return sprintf(page, "%d\n", mddev->layout);
3875 layout_store(struct mddev *mddev, const char *buf, size_t len)
3880 err = kstrtouint(buf, 10, &n);
3883 err = mddev_lock(mddev);
3888 if (mddev->pers->check_reshape == NULL)
3893 mddev->new_layout = n;
3894 err = mddev->pers->check_reshape(mddev);
3896 mddev->new_layout = mddev->layout;
3899 mddev->new_layout = n;
3900 if (mddev->reshape_position == MaxSector)
3903 mddev_unlock(mddev);
3906 static struct md_sysfs_entry md_layout =
3907 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3910 raid_disks_show(struct mddev *mddev, char *page)
3912 if (mddev->raid_disks == 0)
3914 if (mddev->reshape_position != MaxSector &&
3915 mddev->delta_disks != 0)
3916 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3917 mddev->raid_disks - mddev->delta_disks);
3918 return sprintf(page, "%d\n", mddev->raid_disks);
3921 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3924 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3929 err = kstrtouint(buf, 10, &n);
3933 err = mddev_lock(mddev);
3937 err = update_raid_disks(mddev, n);
3938 else if (mddev->reshape_position != MaxSector) {
3939 struct md_rdev *rdev;
3940 int olddisks = mddev->raid_disks - mddev->delta_disks;
3943 rdev_for_each(rdev, mddev) {
3945 rdev->data_offset < rdev->new_data_offset)
3948 rdev->data_offset > rdev->new_data_offset)
3952 mddev->delta_disks = n - olddisks;
3953 mddev->raid_disks = n;
3954 mddev->reshape_backwards = (mddev->delta_disks < 0);
3956 mddev->raid_disks = n;
3958 mddev_unlock(mddev);
3959 return err ? err : len;
3961 static struct md_sysfs_entry md_raid_disks =
3962 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3965 chunk_size_show(struct mddev *mddev, char *page)
3967 if (mddev->reshape_position != MaxSector &&
3968 mddev->chunk_sectors != mddev->new_chunk_sectors)
3969 return sprintf(page, "%d (%d)\n",
3970 mddev->new_chunk_sectors << 9,
3971 mddev->chunk_sectors << 9);
3972 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3976 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3981 err = kstrtoul(buf, 10, &n);
3985 err = mddev_lock(mddev);
3989 if (mddev->pers->check_reshape == NULL)
3994 mddev->new_chunk_sectors = n >> 9;
3995 err = mddev->pers->check_reshape(mddev);
3997 mddev->new_chunk_sectors = mddev->chunk_sectors;
4000 mddev->new_chunk_sectors = n >> 9;
4001 if (mddev->reshape_position == MaxSector)
4002 mddev->chunk_sectors = n >> 9;
4004 mddev_unlock(mddev);
4007 static struct md_sysfs_entry md_chunk_size =
4008 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4011 resync_start_show(struct mddev *mddev, char *page)
4013 if (mddev->recovery_cp == MaxSector)
4014 return sprintf(page, "none\n");
4015 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4019 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4021 unsigned long long n;
4024 if (cmd_match(buf, "none"))
4027 err = kstrtoull(buf, 10, &n);
4030 if (n != (sector_t)n)
4034 err = mddev_lock(mddev);
4037 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4041 mddev->recovery_cp = n;
4043 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4045 mddev_unlock(mddev);
4048 static struct md_sysfs_entry md_resync_start =
4049 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4050 resync_start_show, resync_start_store);
4053 * The array state can be:
4056 * No devices, no size, no level
4057 * Equivalent to STOP_ARRAY ioctl
4059 * May have some settings, but array is not active
4060 * all IO results in error
4061 * When written, doesn't tear down array, but just stops it
4062 * suspended (not supported yet)
4063 * All IO requests will block. The array can be reconfigured.
4064 * Writing this, if accepted, will block until array is quiescent
4066 * no resync can happen. no superblocks get written.
4067 * write requests fail
4069 * like readonly, but behaves like 'clean' on a write request.
4071 * clean - no pending writes, but otherwise active.
4072 * When written to inactive array, starts without resync
4073 * If a write request arrives then
4074 * if metadata is known, mark 'dirty' and switch to 'active'.
4075 * if not known, block and switch to write-pending
4076 * If written to an active array that has pending writes, then fails.
4078 * fully active: IO and resync can be happening.
4079 * When written to inactive array, starts with resync
4082 * clean, but writes are blocked waiting for 'active' to be written.
4085 * like active, but no writes have been seen for a while (100msec).
4088 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4089 write_pending, active_idle, bad_word};
4090 static char *array_states[] = {
4091 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4092 "write-pending", "active-idle", NULL };
4094 static int match_word(const char *word, char **list)
4097 for (n=0; list[n]; n++)
4098 if (cmd_match(word, list[n]))
4104 array_state_show(struct mddev *mddev, char *page)
4106 enum array_state st = inactive;
4117 spin_lock(&mddev->lock);
4118 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4120 else if (mddev->in_sync)
4122 else if (mddev->safemode)
4126 spin_unlock(&mddev->lock);
4129 if (list_empty(&mddev->disks) &&
4130 mddev->raid_disks == 0 &&
4131 mddev->dev_sectors == 0)
4136 return sprintf(page, "%s\n", array_states[st]);
4139 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4140 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4141 static int do_md_run(struct mddev *mddev);
4142 static int restart_array(struct mddev *mddev);
4145 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4148 enum array_state st = match_word(buf, array_states);
4150 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4151 /* don't take reconfig_mutex when toggling between
4154 spin_lock(&mddev->lock);
4156 restart_array(mddev);
4157 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4158 md_wakeup_thread(mddev->thread);
4159 wake_up(&mddev->sb_wait);
4160 } else /* st == clean */ {
4161 restart_array(mddev);
4162 if (!set_in_sync(mddev))
4166 sysfs_notify_dirent_safe(mddev->sysfs_state);
4167 spin_unlock(&mddev->lock);
4170 err = mddev_lock(mddev);
4178 /* stopping an active array */
4179 err = do_md_stop(mddev, 0, NULL);
4182 /* stopping an active array */
4184 err = do_md_stop(mddev, 2, NULL);
4186 err = 0; /* already inactive */
4189 break; /* not supported yet */
4192 err = md_set_readonly(mddev, NULL);
4195 set_disk_ro(mddev->gendisk, 1);
4196 err = do_md_run(mddev);
4202 err = md_set_readonly(mddev, NULL);
4203 else if (mddev->ro == 1)
4204 err = restart_array(mddev);
4207 set_disk_ro(mddev->gendisk, 0);
4211 err = do_md_run(mddev);
4216 err = restart_array(mddev);
4219 spin_lock(&mddev->lock);
4220 if (!set_in_sync(mddev))
4222 spin_unlock(&mddev->lock);
4228 err = restart_array(mddev);
4231 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4232 wake_up(&mddev->sb_wait);
4236 set_disk_ro(mddev->gendisk, 0);
4237 err = do_md_run(mddev);
4242 /* these cannot be set */
4247 if (mddev->hold_active == UNTIL_IOCTL)
4248 mddev->hold_active = 0;
4249 sysfs_notify_dirent_safe(mddev->sysfs_state);
4251 mddev_unlock(mddev);
4254 static struct md_sysfs_entry md_array_state =
4255 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4258 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4259 return sprintf(page, "%d\n",
4260 atomic_read(&mddev->max_corr_read_errors));
4264 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4269 rv = kstrtouint(buf, 10, &n);
4272 atomic_set(&mddev->max_corr_read_errors, n);
4276 static struct md_sysfs_entry max_corr_read_errors =
4277 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4278 max_corrected_read_errors_store);
4281 null_show(struct mddev *mddev, char *page)
4287 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4289 /* buf must be %d:%d\n? giving major and minor numbers */
4290 /* The new device is added to the array.
4291 * If the array has a persistent superblock, we read the
4292 * superblock to initialise info and check validity.
4293 * Otherwise, only checking done is that in bind_rdev_to_array,
4294 * which mainly checks size.
4297 int major = simple_strtoul(buf, &e, 10);
4300 struct md_rdev *rdev;
4303 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4305 minor = simple_strtoul(e+1, &e, 10);
4306 if (*e && *e != '\n')
4308 dev = MKDEV(major, minor);
4309 if (major != MAJOR(dev) ||
4310 minor != MINOR(dev))
4313 flush_workqueue(md_misc_wq);
4315 err = mddev_lock(mddev);
4318 if (mddev->persistent) {
4319 rdev = md_import_device(dev, mddev->major_version,
4320 mddev->minor_version);
4321 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4322 struct md_rdev *rdev0
4323 = list_entry(mddev->disks.next,
4324 struct md_rdev, same_set);
4325 err = super_types[mddev->major_version]
4326 .load_super(rdev, rdev0, mddev->minor_version);
4330 } else if (mddev->external)
4331 rdev = md_import_device(dev, -2, -1);
4333 rdev = md_import_device(dev, -1, -1);
4336 mddev_unlock(mddev);
4337 return PTR_ERR(rdev);
4339 err = bind_rdev_to_array(rdev, mddev);
4343 mddev_unlock(mddev);
4345 md_new_event(mddev);
4346 return err ? err : len;
4349 static struct md_sysfs_entry md_new_device =
4350 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4353 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4356 unsigned long chunk, end_chunk;
4359 err = mddev_lock(mddev);
4364 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4366 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4367 if (buf == end) break;
4368 if (*end == '-') { /* range */
4370 end_chunk = simple_strtoul(buf, &end, 0);
4371 if (buf == end) break;
4373 if (*end && !isspace(*end)) break;
4374 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4375 buf = skip_spaces(end);
4377 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4379 mddev_unlock(mddev);
4383 static struct md_sysfs_entry md_bitmap =
4384 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4387 size_show(struct mddev *mddev, char *page)
4389 return sprintf(page, "%llu\n",
4390 (unsigned long long)mddev->dev_sectors / 2);
4393 static int update_size(struct mddev *mddev, sector_t num_sectors);
4396 size_store(struct mddev *mddev, const char *buf, size_t len)
4398 /* If array is inactive, we can reduce the component size, but
4399 * not increase it (except from 0).
4400 * If array is active, we can try an on-line resize
4403 int err = strict_blocks_to_sectors(buf, §ors);
4407 err = mddev_lock(mddev);
4411 err = update_size(mddev, sectors);
4413 md_update_sb(mddev, 1);
4415 if (mddev->dev_sectors == 0 ||
4416 mddev->dev_sectors > sectors)
4417 mddev->dev_sectors = sectors;
4421 mddev_unlock(mddev);
4422 return err ? err : len;
4425 static struct md_sysfs_entry md_size =
4426 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4428 /* Metadata version.
4430 * 'none' for arrays with no metadata (good luck...)
4431 * 'external' for arrays with externally managed metadata,
4432 * or N.M for internally known formats
4435 metadata_show(struct mddev *mddev, char *page)
4437 if (mddev->persistent)
4438 return sprintf(page, "%d.%d\n",
4439 mddev->major_version, mddev->minor_version);
4440 else if (mddev->external)
4441 return sprintf(page, "external:%s\n", mddev->metadata_type);
4443 return sprintf(page, "none\n");
4447 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4452 /* Changing the details of 'external' metadata is
4453 * always permitted. Otherwise there must be
4454 * no devices attached to the array.
4457 err = mddev_lock(mddev);
4461 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4463 else if (!list_empty(&mddev->disks))
4467 if (cmd_match(buf, "none")) {
4468 mddev->persistent = 0;
4469 mddev->external = 0;
4470 mddev->major_version = 0;
4471 mddev->minor_version = 90;
4474 if (strncmp(buf, "external:", 9) == 0) {
4475 size_t namelen = len-9;
4476 if (namelen >= sizeof(mddev->metadata_type))
4477 namelen = sizeof(mddev->metadata_type)-1;
4478 strncpy(mddev->metadata_type, buf+9, namelen);
4479 mddev->metadata_type[namelen] = 0;
4480 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4481 mddev->metadata_type[--namelen] = 0;
4482 mddev->persistent = 0;
4483 mddev->external = 1;
4484 mddev->major_version = 0;
4485 mddev->minor_version = 90;
4488 major = simple_strtoul(buf, &e, 10);
4490 if (e==buf || *e != '.')
4493 minor = simple_strtoul(buf, &e, 10);
4494 if (e==buf || (*e && *e != '\n') )
4497 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4499 mddev->major_version = major;
4500 mddev->minor_version = minor;
4501 mddev->persistent = 1;
4502 mddev->external = 0;
4505 mddev_unlock(mddev);
4509 static struct md_sysfs_entry md_metadata =
4510 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4513 action_show(struct mddev *mddev, char *page)
4515 char *type = "idle";
4516 unsigned long recovery = mddev->recovery;
4517 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4519 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4520 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4521 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4523 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4524 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4526 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4530 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4532 else if (mddev->reshape_position != MaxSector)
4535 return sprintf(page, "%s\n", type);
4539 action_store(struct mddev *mddev, const char *page, size_t len)
4541 if (!mddev->pers || !mddev->pers->sync_request)
4545 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4546 if (cmd_match(page, "frozen"))
4547 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4549 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4550 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4551 mddev_lock(mddev) == 0) {
4552 flush_workqueue(md_misc_wq);
4553 if (mddev->sync_thread) {
4554 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4555 md_reap_sync_thread(mddev);
4557 mddev_unlock(mddev);
4559 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4561 else if (cmd_match(page, "resync"))
4562 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4563 else if (cmd_match(page, "recover")) {
4564 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4565 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4566 } else if (cmd_match(page, "reshape")) {
4568 if (mddev->pers->start_reshape == NULL)
4570 err = mddev_lock(mddev);
4572 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4575 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4576 err = mddev->pers->start_reshape(mddev);
4578 mddev_unlock(mddev);
4582 sysfs_notify(&mddev->kobj, NULL, "degraded");
4584 if (cmd_match(page, "check"))
4585 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4586 else if (!cmd_match(page, "repair"))
4588 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4589 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4590 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4592 if (mddev->ro == 2) {
4593 /* A write to sync_action is enough to justify
4594 * canceling read-auto mode
4597 md_wakeup_thread(mddev->sync_thread);
4599 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4600 md_wakeup_thread(mddev->thread);
4601 sysfs_notify_dirent_safe(mddev->sysfs_action);
4605 static struct md_sysfs_entry md_scan_mode =
4606 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4609 last_sync_action_show(struct mddev *mddev, char *page)
4611 return sprintf(page, "%s\n", mddev->last_sync_action);
4614 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4617 mismatch_cnt_show(struct mddev *mddev, char *page)
4619 return sprintf(page, "%llu\n",
4620 (unsigned long long)
4621 atomic64_read(&mddev->resync_mismatches));
4624 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4627 sync_min_show(struct mddev *mddev, char *page)
4629 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4630 mddev->sync_speed_min ? "local": "system");
4634 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4639 if (strncmp(buf, "system", 6)==0) {
4642 rv = kstrtouint(buf, 10, &min);
4648 mddev->sync_speed_min = min;
4652 static struct md_sysfs_entry md_sync_min =
4653 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4656 sync_max_show(struct mddev *mddev, char *page)
4658 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4659 mddev->sync_speed_max ? "local": "system");
4663 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4668 if (strncmp(buf, "system", 6)==0) {
4671 rv = kstrtouint(buf, 10, &max);
4677 mddev->sync_speed_max = max;
4681 static struct md_sysfs_entry md_sync_max =
4682 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4685 degraded_show(struct mddev *mddev, char *page)
4687 return sprintf(page, "%d\n", mddev->degraded);
4689 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4692 sync_force_parallel_show(struct mddev *mddev, char *page)
4694 return sprintf(page, "%d\n", mddev->parallel_resync);
4698 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4702 if (kstrtol(buf, 10, &n))
4705 if (n != 0 && n != 1)
4708 mddev->parallel_resync = n;
4710 if (mddev->sync_thread)
4711 wake_up(&resync_wait);
4716 /* force parallel resync, even with shared block devices */
4717 static struct md_sysfs_entry md_sync_force_parallel =
4718 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4719 sync_force_parallel_show, sync_force_parallel_store);
4722 sync_speed_show(struct mddev *mddev, char *page)
4724 unsigned long resync, dt, db;
4725 if (mddev->curr_resync == 0)
4726 return sprintf(page, "none\n");
4727 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4728 dt = (jiffies - mddev->resync_mark) / HZ;
4730 db = resync - mddev->resync_mark_cnt;
4731 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4734 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4737 sync_completed_show(struct mddev *mddev, char *page)
4739 unsigned long long max_sectors, resync;
4741 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4742 return sprintf(page, "none\n");
4744 if (mddev->curr_resync == 1 ||
4745 mddev->curr_resync == 2)
4746 return sprintf(page, "delayed\n");
4748 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4749 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4750 max_sectors = mddev->resync_max_sectors;
4752 max_sectors = mddev->dev_sectors;
4754 resync = mddev->curr_resync_completed;
4755 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4758 static struct md_sysfs_entry md_sync_completed =
4759 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4762 min_sync_show(struct mddev *mddev, char *page)
4764 return sprintf(page, "%llu\n",
4765 (unsigned long long)mddev->resync_min);
4768 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4770 unsigned long long min;
4773 if (kstrtoull(buf, 10, &min))
4776 spin_lock(&mddev->lock);
4778 if (min > mddev->resync_max)
4782 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4785 /* Round down to multiple of 4K for safety */
4786 mddev->resync_min = round_down(min, 8);
4790 spin_unlock(&mddev->lock);
4794 static struct md_sysfs_entry md_min_sync =
4795 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4798 max_sync_show(struct mddev *mddev, char *page)
4800 if (mddev->resync_max == MaxSector)
4801 return sprintf(page, "max\n");
4803 return sprintf(page, "%llu\n",
4804 (unsigned long long)mddev->resync_max);
4807 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4810 spin_lock(&mddev->lock);
4811 if (strncmp(buf, "max", 3) == 0)
4812 mddev->resync_max = MaxSector;
4814 unsigned long long max;
4818 if (kstrtoull(buf, 10, &max))
4820 if (max < mddev->resync_min)
4824 if (max < mddev->resync_max &&
4826 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4829 /* Must be a multiple of chunk_size */
4830 chunk = mddev->chunk_sectors;
4832 sector_t temp = max;
4835 if (sector_div(temp, chunk))
4838 mddev->resync_max = max;
4840 wake_up(&mddev->recovery_wait);
4843 spin_unlock(&mddev->lock);
4847 static struct md_sysfs_entry md_max_sync =
4848 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4851 suspend_lo_show(struct mddev *mddev, char *page)
4853 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4857 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4859 unsigned long long new;
4862 err = kstrtoull(buf, 10, &new);
4865 if (new != (sector_t)new)
4868 err = mddev_lock(mddev);
4872 if (mddev->pers == NULL ||
4873 mddev->pers->quiesce == NULL)
4875 mddev_suspend(mddev);
4876 mddev->suspend_lo = new;
4877 mddev_resume(mddev);
4881 mddev_unlock(mddev);
4884 static struct md_sysfs_entry md_suspend_lo =
4885 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4888 suspend_hi_show(struct mddev *mddev, char *page)
4890 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4894 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4896 unsigned long long new;
4899 err = kstrtoull(buf, 10, &new);
4902 if (new != (sector_t)new)
4905 err = mddev_lock(mddev);
4909 if (mddev->pers == NULL)
4912 mddev_suspend(mddev);
4913 mddev->suspend_hi = new;
4914 mddev_resume(mddev);
4918 mddev_unlock(mddev);
4921 static struct md_sysfs_entry md_suspend_hi =
4922 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4925 reshape_position_show(struct mddev *mddev, char *page)
4927 if (mddev->reshape_position != MaxSector)
4928 return sprintf(page, "%llu\n",
4929 (unsigned long long)mddev->reshape_position);
4930 strcpy(page, "none\n");
4935 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4937 struct md_rdev *rdev;
4938 unsigned long long new;
4941 err = kstrtoull(buf, 10, &new);
4944 if (new != (sector_t)new)
4946 err = mddev_lock(mddev);
4952 mddev->reshape_position = new;
4953 mddev->delta_disks = 0;
4954 mddev->reshape_backwards = 0;
4955 mddev->new_level = mddev->level;
4956 mddev->new_layout = mddev->layout;
4957 mddev->new_chunk_sectors = mddev->chunk_sectors;
4958 rdev_for_each(rdev, mddev)
4959 rdev->new_data_offset = rdev->data_offset;
4962 mddev_unlock(mddev);
4966 static struct md_sysfs_entry md_reshape_position =
4967 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4968 reshape_position_store);
4971 reshape_direction_show(struct mddev *mddev, char *page)
4973 return sprintf(page, "%s\n",
4974 mddev->reshape_backwards ? "backwards" : "forwards");
4978 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4983 if (cmd_match(buf, "forwards"))
4985 else if (cmd_match(buf, "backwards"))
4989 if (mddev->reshape_backwards == backwards)
4992 err = mddev_lock(mddev);
4995 /* check if we are allowed to change */
4996 if (mddev->delta_disks)
4998 else if (mddev->persistent &&
4999 mddev->major_version == 0)
5002 mddev->reshape_backwards = backwards;
5003 mddev_unlock(mddev);
5007 static struct md_sysfs_entry md_reshape_direction =
5008 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5009 reshape_direction_store);
5012 array_size_show(struct mddev *mddev, char *page)
5014 if (mddev->external_size)
5015 return sprintf(page, "%llu\n",
5016 (unsigned long long)mddev->array_sectors/2);
5018 return sprintf(page, "default\n");
5022 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5027 err = mddev_lock(mddev);
5031 /* cluster raid doesn't support change array_sectors */
5032 if (mddev_is_clustered(mddev)) {
5033 mddev_unlock(mddev);
5037 if (strncmp(buf, "default", 7) == 0) {
5039 sectors = mddev->pers->size(mddev, 0, 0);
5041 sectors = mddev->array_sectors;
5043 mddev->external_size = 0;
5045 if (strict_blocks_to_sectors(buf, §ors) < 0)
5047 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5050 mddev->external_size = 1;
5054 mddev->array_sectors = sectors;
5056 set_capacity(mddev->gendisk, mddev->array_sectors);
5057 revalidate_disk(mddev->gendisk);
5060 mddev_unlock(mddev);
5064 static struct md_sysfs_entry md_array_size =
5065 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5069 consistency_policy_show(struct mddev *mddev, char *page)
5073 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5074 ret = sprintf(page, "journal\n");
5075 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5076 ret = sprintf(page, "ppl\n");
5077 } else if (mddev->bitmap) {
5078 ret = sprintf(page, "bitmap\n");
5079 } else if (mddev->pers) {
5080 if (mddev->pers->sync_request)
5081 ret = sprintf(page, "resync\n");
5083 ret = sprintf(page, "none\n");
5085 ret = sprintf(page, "unknown\n");
5092 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5097 if (mddev->pers->change_consistency_policy)
5098 err = mddev->pers->change_consistency_policy(mddev, buf);
5101 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5102 set_bit(MD_HAS_PPL, &mddev->flags);
5107 return err ? err : len;
5110 static struct md_sysfs_entry md_consistency_policy =
5111 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5112 consistency_policy_store);
5114 static struct attribute *md_default_attrs[] = {
5117 &md_raid_disks.attr,
5118 &md_chunk_size.attr,
5120 &md_resync_start.attr,
5122 &md_new_device.attr,
5123 &md_safe_delay.attr,
5124 &md_array_state.attr,
5125 &md_reshape_position.attr,
5126 &md_reshape_direction.attr,
5127 &md_array_size.attr,
5128 &max_corr_read_errors.attr,
5129 &md_consistency_policy.attr,
5133 static struct attribute *md_redundancy_attrs[] = {
5135 &md_last_scan_mode.attr,
5136 &md_mismatches.attr,
5139 &md_sync_speed.attr,
5140 &md_sync_force_parallel.attr,
5141 &md_sync_completed.attr,
5144 &md_suspend_lo.attr,
5145 &md_suspend_hi.attr,
5150 static struct attribute_group md_redundancy_group = {
5152 .attrs = md_redundancy_attrs,
5156 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5158 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5159 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5164 spin_lock(&all_mddevs_lock);
5165 if (list_empty(&mddev->all_mddevs)) {
5166 spin_unlock(&all_mddevs_lock);
5170 spin_unlock(&all_mddevs_lock);
5172 rv = entry->show(mddev, page);
5178 md_attr_store(struct kobject *kobj, struct attribute *attr,
5179 const char *page, size_t length)
5181 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5182 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5187 if (!capable(CAP_SYS_ADMIN))
5189 spin_lock(&all_mddevs_lock);
5190 if (list_empty(&mddev->all_mddevs)) {
5191 spin_unlock(&all_mddevs_lock);
5195 spin_unlock(&all_mddevs_lock);
5196 rv = entry->store(mddev, page, length);
5201 static void md_free(struct kobject *ko)
5203 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5205 if (mddev->sysfs_state)
5206 sysfs_put(mddev->sysfs_state);
5209 del_gendisk(mddev->gendisk);
5211 blk_cleanup_queue(mddev->queue);
5213 put_disk(mddev->gendisk);
5214 percpu_ref_exit(&mddev->writes_pending);
5216 bioset_exit(&mddev->bio_set);
5217 bioset_exit(&mddev->sync_set);
5221 static const struct sysfs_ops md_sysfs_ops = {
5222 .show = md_attr_show,
5223 .store = md_attr_store,
5225 static struct kobj_type md_ktype = {
5227 .sysfs_ops = &md_sysfs_ops,
5228 .default_attrs = md_default_attrs,
5233 static void mddev_delayed_delete(struct work_struct *ws)
5235 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5237 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5238 kobject_del(&mddev->kobj);
5239 kobject_put(&mddev->kobj);
5242 static void no_op(struct percpu_ref *r) {}
5244 int mddev_init_writes_pending(struct mddev *mddev)
5246 if (mddev->writes_pending.percpu_count_ptr)
5248 if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0)
5250 /* We want to start with the refcount at zero */
5251 percpu_ref_put(&mddev->writes_pending);
5254 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5256 static int md_alloc(dev_t dev, char *name)
5259 * If dev is zero, name is the name of a device to allocate with
5260 * an arbitrary minor number. It will be "md_???"
5261 * If dev is non-zero it must be a device number with a MAJOR of
5262 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5263 * the device is being created by opening a node in /dev.
5264 * If "name" is not NULL, the device is being created by
5265 * writing to /sys/module/md_mod/parameters/new_array.
5267 static DEFINE_MUTEX(disks_mutex);
5268 struct mddev *mddev = mddev_find(dev);
5269 struct gendisk *disk;
5278 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5279 shift = partitioned ? MdpMinorShift : 0;
5280 unit = MINOR(mddev->unit) >> shift;
5282 /* wait for any previous instance of this device to be
5283 * completely removed (mddev_delayed_delete).
5285 flush_workqueue(md_misc_wq);
5287 mutex_lock(&disks_mutex);
5293 /* Need to ensure that 'name' is not a duplicate.
5295 struct mddev *mddev2;
5296 spin_lock(&all_mddevs_lock);
5298 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5299 if (mddev2->gendisk &&
5300 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5301 spin_unlock(&all_mddevs_lock);
5304 spin_unlock(&all_mddevs_lock);
5308 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5310 mddev->hold_active = UNTIL_STOP;
5313 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5316 mddev->queue->queuedata = mddev;
5318 blk_queue_make_request(mddev->queue, md_make_request);
5319 blk_set_stacking_limits(&mddev->queue->limits);
5321 disk = alloc_disk(1 << shift);
5323 blk_cleanup_queue(mddev->queue);
5324 mddev->queue = NULL;
5327 disk->major = MAJOR(mddev->unit);
5328 disk->first_minor = unit << shift;
5330 strcpy(disk->disk_name, name);
5331 else if (partitioned)
5332 sprintf(disk->disk_name, "md_d%d", unit);
5334 sprintf(disk->disk_name, "md%d", unit);
5335 disk->fops = &md_fops;
5336 disk->private_data = mddev;
5337 disk->queue = mddev->queue;
5338 blk_queue_write_cache(mddev->queue, true, true);
5339 /* Allow extended partitions. This makes the
5340 * 'mdp' device redundant, but we can't really
5343 disk->flags |= GENHD_FL_EXT_DEVT;
5344 mddev->gendisk = disk;
5345 /* As soon as we call add_disk(), another thread could get
5346 * through to md_open, so make sure it doesn't get too far
5348 mutex_lock(&mddev->open_mutex);
5351 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5353 /* This isn't possible, but as kobject_init_and_add is marked
5354 * __must_check, we must do something with the result
5356 pr_debug("md: cannot register %s/md - name in use\n",
5360 if (mddev->kobj.sd &&
5361 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5362 pr_debug("pointless warning\n");
5363 mutex_unlock(&mddev->open_mutex);
5365 mutex_unlock(&disks_mutex);
5366 if (!error && mddev->kobj.sd) {
5367 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5368 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5374 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5377 md_alloc(dev, NULL);
5381 static int add_named_array(const char *val, const struct kernel_param *kp)
5384 * val must be "md_*" or "mdNNN".
5385 * For "md_*" we allocate an array with a large free minor number, and
5386 * set the name to val. val must not already be an active name.
5387 * For "mdNNN" we allocate an array with the minor number NNN
5388 * which must not already be in use.
5390 int len = strlen(val);
5391 char buf[DISK_NAME_LEN];
5392 unsigned long devnum;
5394 while (len && val[len-1] == '\n')
5396 if (len >= DISK_NAME_LEN)
5398 strlcpy(buf, val, len+1);
5399 if (strncmp(buf, "md_", 3) == 0)
5400 return md_alloc(0, buf);
5401 if (strncmp(buf, "md", 2) == 0 &&
5403 kstrtoul(buf+2, 10, &devnum) == 0 &&
5404 devnum <= MINORMASK)
5405 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5410 static void md_safemode_timeout(struct timer_list *t)
5412 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5414 mddev->safemode = 1;
5415 if (mddev->external)
5416 sysfs_notify_dirent_safe(mddev->sysfs_state);
5418 md_wakeup_thread(mddev->thread);
5421 static int start_dirty_degraded;
5423 int md_run(struct mddev *mddev)
5426 struct md_rdev *rdev;
5427 struct md_personality *pers;
5429 if (list_empty(&mddev->disks))
5430 /* cannot run an array with no devices.. */
5435 /* Cannot run until previous stop completes properly */
5436 if (mddev->sysfs_active)
5440 * Analyze all RAID superblock(s)
5442 if (!mddev->raid_disks) {
5443 if (!mddev->persistent)
5448 if (mddev->level != LEVEL_NONE)
5449 request_module("md-level-%d", mddev->level);
5450 else if (mddev->clevel[0])
5451 request_module("md-%s", mddev->clevel);
5454 * Drop all container device buffers, from now on
5455 * the only valid external interface is through the md
5458 mddev->has_superblocks = false;
5459 rdev_for_each(rdev, mddev) {
5460 if (test_bit(Faulty, &rdev->flags))
5462 sync_blockdev(rdev->bdev);
5463 invalidate_bdev(rdev->bdev);
5464 if (mddev->ro != 1 &&
5465 (bdev_read_only(rdev->bdev) ||
5466 bdev_read_only(rdev->meta_bdev))) {
5469 set_disk_ro(mddev->gendisk, 1);
5473 mddev->has_superblocks = true;
5475 /* perform some consistency tests on the device.
5476 * We don't want the data to overlap the metadata,
5477 * Internal Bitmap issues have been handled elsewhere.
5479 if (rdev->meta_bdev) {
5480 /* Nothing to check */;
5481 } else if (rdev->data_offset < rdev->sb_start) {
5482 if (mddev->dev_sectors &&
5483 rdev->data_offset + mddev->dev_sectors
5485 pr_warn("md: %s: data overlaps metadata\n",
5490 if (rdev->sb_start + rdev->sb_size/512
5491 > rdev->data_offset) {
5492 pr_warn("md: %s: metadata overlaps data\n",
5497 sysfs_notify_dirent_safe(rdev->sysfs_state);
5500 if (!bioset_initialized(&mddev->bio_set)) {
5501 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5505 if (!bioset_initialized(&mddev->sync_set)) {
5506 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5511 spin_lock(&pers_lock);
5512 pers = find_pers(mddev->level, mddev->clevel);
5513 if (!pers || !try_module_get(pers->owner)) {
5514 spin_unlock(&pers_lock);
5515 if (mddev->level != LEVEL_NONE)
5516 pr_warn("md: personality for level %d is not loaded!\n",
5519 pr_warn("md: personality for level %s is not loaded!\n",
5524 spin_unlock(&pers_lock);
5525 if (mddev->level != pers->level) {
5526 mddev->level = pers->level;
5527 mddev->new_level = pers->level;
5529 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5531 if (mddev->reshape_position != MaxSector &&
5532 pers->start_reshape == NULL) {
5533 /* This personality cannot handle reshaping... */
5534 module_put(pers->owner);
5539 if (pers->sync_request) {
5540 /* Warn if this is a potentially silly
5543 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5544 struct md_rdev *rdev2;
5547 rdev_for_each(rdev, mddev)
5548 rdev_for_each(rdev2, mddev) {
5550 rdev->bdev->bd_contains ==
5551 rdev2->bdev->bd_contains) {
5552 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5554 bdevname(rdev->bdev,b),
5555 bdevname(rdev2->bdev,b2));
5561 pr_warn("True protection against single-disk failure might be compromised.\n");
5564 mddev->recovery = 0;
5565 /* may be over-ridden by personality */
5566 mddev->resync_max_sectors = mddev->dev_sectors;
5568 mddev->ok_start_degraded = start_dirty_degraded;
5570 if (start_readonly && mddev->ro == 0)
5571 mddev->ro = 2; /* read-only, but switch on first write */
5573 err = pers->run(mddev);
5575 pr_warn("md: pers->run() failed ...\n");
5576 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5577 WARN_ONCE(!mddev->external_size,
5578 "%s: default size too small, but 'external_size' not in effect?\n",
5580 pr_warn("md: invalid array_size %llu > default size %llu\n",
5581 (unsigned long long)mddev->array_sectors / 2,
5582 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5585 if (err == 0 && pers->sync_request &&
5586 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5587 struct bitmap *bitmap;
5589 bitmap = md_bitmap_create(mddev, -1);
5590 if (IS_ERR(bitmap)) {
5591 err = PTR_ERR(bitmap);
5592 pr_warn("%s: failed to create bitmap (%d)\n",
5593 mdname(mddev), err);
5595 mddev->bitmap = bitmap;
5599 mddev_detach(mddev);
5601 pers->free(mddev, mddev->private);
5602 mddev->private = NULL;
5603 module_put(pers->owner);
5604 md_bitmap_destroy(mddev);
5610 rdev_for_each(rdev, mddev) {
5611 if (rdev->raid_disk >= 0 &&
5612 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5617 if (mddev->degraded)
5620 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5622 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5623 mddev->queue->backing_dev_info->congested_data = mddev;
5624 mddev->queue->backing_dev_info->congested_fn = md_congested;
5626 if (pers->sync_request) {
5627 if (mddev->kobj.sd &&
5628 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5629 pr_warn("md: cannot register extra attributes for %s\n",
5631 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5632 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5635 atomic_set(&mddev->max_corr_read_errors,
5636 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5637 mddev->safemode = 0;
5638 if (mddev_is_clustered(mddev))
5639 mddev->safemode_delay = 0;
5641 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5644 spin_lock(&mddev->lock);
5646 spin_unlock(&mddev->lock);
5647 rdev_for_each(rdev, mddev)
5648 if (rdev->raid_disk >= 0)
5649 if (sysfs_link_rdev(mddev, rdev))
5650 /* failure here is OK */;
5652 if (mddev->degraded && !mddev->ro)
5653 /* This ensures that recovering status is reported immediately
5654 * via sysfs - until a lack of spares is confirmed.
5656 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5657 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5659 if (mddev->sb_flags)
5660 md_update_sb(mddev, 0);
5662 md_new_event(mddev);
5663 sysfs_notify_dirent_safe(mddev->sysfs_state);
5664 sysfs_notify_dirent_safe(mddev->sysfs_action);
5665 sysfs_notify(&mddev->kobj, NULL, "degraded");
5669 bioset_exit(&mddev->bio_set);
5670 bioset_exit(&mddev->sync_set);
5673 EXPORT_SYMBOL_GPL(md_run);
5675 static int do_md_run(struct mddev *mddev)
5679 err = md_run(mddev);
5682 err = md_bitmap_load(mddev);
5684 md_bitmap_destroy(mddev);
5688 if (mddev_is_clustered(mddev))
5689 md_allow_write(mddev);
5691 /* run start up tasks that require md_thread */
5694 md_wakeup_thread(mddev->thread);
5695 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5697 set_capacity(mddev->gendisk, mddev->array_sectors);
5698 revalidate_disk(mddev->gendisk);
5700 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5705 int md_start(struct mddev *mddev)
5709 if (mddev->pers->start) {
5710 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5711 md_wakeup_thread(mddev->thread);
5712 ret = mddev->pers->start(mddev);
5713 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5714 md_wakeup_thread(mddev->sync_thread);
5718 EXPORT_SYMBOL_GPL(md_start);
5720 static int restart_array(struct mddev *mddev)
5722 struct gendisk *disk = mddev->gendisk;
5723 struct md_rdev *rdev;
5724 bool has_journal = false;
5725 bool has_readonly = false;
5727 /* Complain if it has no devices */
5728 if (list_empty(&mddev->disks))
5736 rdev_for_each_rcu(rdev, mddev) {
5737 if (test_bit(Journal, &rdev->flags) &&
5738 !test_bit(Faulty, &rdev->flags))
5740 if (bdev_read_only(rdev->bdev))
5741 has_readonly = true;
5744 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5745 /* Don't restart rw with journal missing/faulty */
5750 mddev->safemode = 0;
5752 set_disk_ro(disk, 0);
5753 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5754 /* Kick recovery or resync if necessary */
5755 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5756 md_wakeup_thread(mddev->thread);
5757 md_wakeup_thread(mddev->sync_thread);
5758 sysfs_notify_dirent_safe(mddev->sysfs_state);
5762 static void md_clean(struct mddev *mddev)
5764 mddev->array_sectors = 0;
5765 mddev->external_size = 0;
5766 mddev->dev_sectors = 0;
5767 mddev->raid_disks = 0;
5768 mddev->recovery_cp = 0;
5769 mddev->resync_min = 0;
5770 mddev->resync_max = MaxSector;
5771 mddev->reshape_position = MaxSector;
5772 mddev->external = 0;
5773 mddev->persistent = 0;
5774 mddev->level = LEVEL_NONE;
5775 mddev->clevel[0] = 0;
5777 mddev->sb_flags = 0;
5779 mddev->metadata_type[0] = 0;
5780 mddev->chunk_sectors = 0;
5781 mddev->ctime = mddev->utime = 0;
5783 mddev->max_disks = 0;
5785 mddev->can_decrease_events = 0;
5786 mddev->delta_disks = 0;
5787 mddev->reshape_backwards = 0;
5788 mddev->new_level = LEVEL_NONE;
5789 mddev->new_layout = 0;
5790 mddev->new_chunk_sectors = 0;
5791 mddev->curr_resync = 0;
5792 atomic64_set(&mddev->resync_mismatches, 0);
5793 mddev->suspend_lo = mddev->suspend_hi = 0;
5794 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5795 mddev->recovery = 0;
5798 mddev->degraded = 0;
5799 mddev->safemode = 0;
5800 mddev->private = NULL;
5801 mddev->cluster_info = NULL;
5802 mddev->bitmap_info.offset = 0;
5803 mddev->bitmap_info.default_offset = 0;
5804 mddev->bitmap_info.default_space = 0;
5805 mddev->bitmap_info.chunksize = 0;
5806 mddev->bitmap_info.daemon_sleep = 0;
5807 mddev->bitmap_info.max_write_behind = 0;
5808 mddev->bitmap_info.nodes = 0;
5811 static void __md_stop_writes(struct mddev *mddev)
5813 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5814 flush_workqueue(md_misc_wq);
5815 if (mddev->sync_thread) {
5816 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5817 md_reap_sync_thread(mddev);
5820 del_timer_sync(&mddev->safemode_timer);
5822 if (mddev->pers && mddev->pers->quiesce) {
5823 mddev->pers->quiesce(mddev, 1);
5824 mddev->pers->quiesce(mddev, 0);
5826 md_bitmap_flush(mddev);
5828 if (mddev->ro == 0 &&
5829 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5831 /* mark array as shutdown cleanly */
5832 if (!mddev_is_clustered(mddev))
5834 md_update_sb(mddev, 1);
5838 void md_stop_writes(struct mddev *mddev)
5840 mddev_lock_nointr(mddev);
5841 __md_stop_writes(mddev);
5842 mddev_unlock(mddev);
5844 EXPORT_SYMBOL_GPL(md_stop_writes);
5846 static void mddev_detach(struct mddev *mddev)
5848 md_bitmap_wait_behind_writes(mddev);
5849 if (mddev->pers && mddev->pers->quiesce) {
5850 mddev->pers->quiesce(mddev, 1);
5851 mddev->pers->quiesce(mddev, 0);
5853 md_unregister_thread(&mddev->thread);
5855 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5858 static void __md_stop(struct mddev *mddev)
5860 struct md_personality *pers = mddev->pers;
5861 md_bitmap_destroy(mddev);
5862 mddev_detach(mddev);
5863 /* Ensure ->event_work is done */
5864 flush_workqueue(md_misc_wq);
5865 spin_lock(&mddev->lock);
5867 spin_unlock(&mddev->lock);
5868 pers->free(mddev, mddev->private);
5869 mddev->private = NULL;
5870 if (pers->sync_request && mddev->to_remove == NULL)
5871 mddev->to_remove = &md_redundancy_group;
5872 module_put(pers->owner);
5873 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5876 void md_stop(struct mddev *mddev)
5878 /* stop the array and free an attached data structures.
5879 * This is called from dm-raid
5882 bioset_exit(&mddev->bio_set);
5883 bioset_exit(&mddev->sync_set);
5886 EXPORT_SYMBOL_GPL(md_stop);
5888 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5893 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5895 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5896 md_wakeup_thread(mddev->thread);
5898 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5899 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5900 if (mddev->sync_thread)
5901 /* Thread might be blocked waiting for metadata update
5902 * which will now never happen */
5903 wake_up_process(mddev->sync_thread->tsk);
5905 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5907 mddev_unlock(mddev);
5908 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5910 wait_event(mddev->sb_wait,
5911 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5912 mddev_lock_nointr(mddev);
5914 mutex_lock(&mddev->open_mutex);
5915 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5916 mddev->sync_thread ||
5917 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5918 pr_warn("md: %s still in use.\n",mdname(mddev));
5920 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5921 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5922 md_wakeup_thread(mddev->thread);
5928 __md_stop_writes(mddev);
5934 set_disk_ro(mddev->gendisk, 1);
5935 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5936 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5937 md_wakeup_thread(mddev->thread);
5938 sysfs_notify_dirent_safe(mddev->sysfs_state);
5942 mutex_unlock(&mddev->open_mutex);
5947 * 0 - completely stop and dis-assemble array
5948 * 2 - stop but do not disassemble array
5950 static int do_md_stop(struct mddev *mddev, int mode,
5951 struct block_device *bdev)
5953 struct gendisk *disk = mddev->gendisk;
5954 struct md_rdev *rdev;
5957 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5959 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5960 md_wakeup_thread(mddev->thread);
5962 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5963 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5964 if (mddev->sync_thread)
5965 /* Thread might be blocked waiting for metadata update
5966 * which will now never happen */
5967 wake_up_process(mddev->sync_thread->tsk);
5969 mddev_unlock(mddev);
5970 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5971 !test_bit(MD_RECOVERY_RUNNING,
5972 &mddev->recovery)));
5973 mddev_lock_nointr(mddev);
5975 mutex_lock(&mddev->open_mutex);
5976 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5977 mddev->sysfs_active ||
5978 mddev->sync_thread ||
5979 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5980 pr_warn("md: %s still in use.\n",mdname(mddev));
5981 mutex_unlock(&mddev->open_mutex);
5983 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5984 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5985 md_wakeup_thread(mddev->thread);
5991 set_disk_ro(disk, 0);
5993 __md_stop_writes(mddev);
5995 mddev->queue->backing_dev_info->congested_fn = NULL;
5997 /* tell userspace to handle 'inactive' */
5998 sysfs_notify_dirent_safe(mddev->sysfs_state);
6000 rdev_for_each(rdev, mddev)
6001 if (rdev->raid_disk >= 0)
6002 sysfs_unlink_rdev(mddev, rdev);
6004 set_capacity(disk, 0);
6005 mutex_unlock(&mddev->open_mutex);
6007 revalidate_disk(disk);
6012 mutex_unlock(&mddev->open_mutex);
6014 * Free resources if final stop
6017 pr_info("md: %s stopped.\n", mdname(mddev));
6019 if (mddev->bitmap_info.file) {
6020 struct file *f = mddev->bitmap_info.file;
6021 spin_lock(&mddev->lock);
6022 mddev->bitmap_info.file = NULL;
6023 spin_unlock(&mddev->lock);
6026 mddev->bitmap_info.offset = 0;
6028 export_array(mddev);
6031 if (mddev->hold_active == UNTIL_STOP)
6032 mddev->hold_active = 0;
6034 md_new_event(mddev);
6035 sysfs_notify_dirent_safe(mddev->sysfs_state);
6040 static void autorun_array(struct mddev *mddev)
6042 struct md_rdev *rdev;
6045 if (list_empty(&mddev->disks))
6048 pr_info("md: running: ");
6050 rdev_for_each(rdev, mddev) {
6051 char b[BDEVNAME_SIZE];
6052 pr_cont("<%s>", bdevname(rdev->bdev,b));
6056 err = do_md_run(mddev);
6058 pr_warn("md: do_md_run() returned %d\n", err);
6059 do_md_stop(mddev, 0, NULL);
6064 * lets try to run arrays based on all disks that have arrived
6065 * until now. (those are in pending_raid_disks)
6067 * the method: pick the first pending disk, collect all disks with
6068 * the same UUID, remove all from the pending list and put them into
6069 * the 'same_array' list. Then order this list based on superblock
6070 * update time (freshest comes first), kick out 'old' disks and
6071 * compare superblocks. If everything's fine then run it.
6073 * If "unit" is allocated, then bump its reference count
6075 static void autorun_devices(int part)
6077 struct md_rdev *rdev0, *rdev, *tmp;
6078 struct mddev *mddev;
6079 char b[BDEVNAME_SIZE];
6081 pr_info("md: autorun ...\n");
6082 while (!list_empty(&pending_raid_disks)) {
6085 LIST_HEAD(candidates);
6086 rdev0 = list_entry(pending_raid_disks.next,
6087 struct md_rdev, same_set);
6089 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6090 INIT_LIST_HEAD(&candidates);
6091 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6092 if (super_90_load(rdev, rdev0, 0) >= 0) {
6093 pr_debug("md: adding %s ...\n",
6094 bdevname(rdev->bdev,b));
6095 list_move(&rdev->same_set, &candidates);
6098 * now we have a set of devices, with all of them having
6099 * mostly sane superblocks. It's time to allocate the
6103 dev = MKDEV(mdp_major,
6104 rdev0->preferred_minor << MdpMinorShift);
6105 unit = MINOR(dev) >> MdpMinorShift;
6107 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6110 if (rdev0->preferred_minor != unit) {
6111 pr_warn("md: unit number in %s is bad: %d\n",
6112 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6116 md_probe(dev, NULL, NULL);
6117 mddev = mddev_find(dev);
6118 if (!mddev || !mddev->gendisk) {
6123 if (mddev_lock(mddev))
6124 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6125 else if (mddev->raid_disks || mddev->major_version
6126 || !list_empty(&mddev->disks)) {
6127 pr_warn("md: %s already running, cannot run %s\n",
6128 mdname(mddev), bdevname(rdev0->bdev,b));
6129 mddev_unlock(mddev);
6131 pr_debug("md: created %s\n", mdname(mddev));
6132 mddev->persistent = 1;
6133 rdev_for_each_list(rdev, tmp, &candidates) {
6134 list_del_init(&rdev->same_set);
6135 if (bind_rdev_to_array(rdev, mddev))
6138 autorun_array(mddev);
6139 mddev_unlock(mddev);
6141 /* on success, candidates will be empty, on error
6144 rdev_for_each_list(rdev, tmp, &candidates) {
6145 list_del_init(&rdev->same_set);
6150 pr_info("md: ... autorun DONE.\n");
6152 #endif /* !MODULE */
6154 static int get_version(void __user *arg)
6158 ver.major = MD_MAJOR_VERSION;
6159 ver.minor = MD_MINOR_VERSION;
6160 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6162 if (copy_to_user(arg, &ver, sizeof(ver)))
6168 static int get_array_info(struct mddev *mddev, void __user *arg)
6170 mdu_array_info_t info;
6171 int nr,working,insync,failed,spare;
6172 struct md_rdev *rdev;
6174 nr = working = insync = failed = spare = 0;
6176 rdev_for_each_rcu(rdev, mddev) {
6178 if (test_bit(Faulty, &rdev->flags))
6182 if (test_bit(In_sync, &rdev->flags))
6184 else if (test_bit(Journal, &rdev->flags))
6185 /* TODO: add journal count to md_u.h */
6193 info.major_version = mddev->major_version;
6194 info.minor_version = mddev->minor_version;
6195 info.patch_version = MD_PATCHLEVEL_VERSION;
6196 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6197 info.level = mddev->level;
6198 info.size = mddev->dev_sectors / 2;
6199 if (info.size != mddev->dev_sectors / 2) /* overflow */
6202 info.raid_disks = mddev->raid_disks;
6203 info.md_minor = mddev->md_minor;
6204 info.not_persistent= !mddev->persistent;
6206 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6209 info.state = (1<<MD_SB_CLEAN);
6210 if (mddev->bitmap && mddev->bitmap_info.offset)
6211 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6212 if (mddev_is_clustered(mddev))
6213 info.state |= (1<<MD_SB_CLUSTERED);
6214 info.active_disks = insync;
6215 info.working_disks = working;
6216 info.failed_disks = failed;
6217 info.spare_disks = spare;
6219 info.layout = mddev->layout;
6220 info.chunk_size = mddev->chunk_sectors << 9;
6222 if (copy_to_user(arg, &info, sizeof(info)))
6228 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6230 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6234 file = kzalloc(sizeof(*file), GFP_NOIO);
6239 spin_lock(&mddev->lock);
6240 /* bitmap enabled */
6241 if (mddev->bitmap_info.file) {
6242 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6243 sizeof(file->pathname));
6247 memmove(file->pathname, ptr,
6248 sizeof(file->pathname)-(ptr-file->pathname));
6250 spin_unlock(&mddev->lock);
6253 copy_to_user(arg, file, sizeof(*file)))
6260 static int get_disk_info(struct mddev *mddev, void __user * arg)
6262 mdu_disk_info_t info;
6263 struct md_rdev *rdev;
6265 if (copy_from_user(&info, arg, sizeof(info)))
6269 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6271 info.major = MAJOR(rdev->bdev->bd_dev);
6272 info.minor = MINOR(rdev->bdev->bd_dev);
6273 info.raid_disk = rdev->raid_disk;
6275 if (test_bit(Faulty, &rdev->flags))
6276 info.state |= (1<<MD_DISK_FAULTY);
6277 else if (test_bit(In_sync, &rdev->flags)) {
6278 info.state |= (1<<MD_DISK_ACTIVE);
6279 info.state |= (1<<MD_DISK_SYNC);
6281 if (test_bit(Journal, &rdev->flags))
6282 info.state |= (1<<MD_DISK_JOURNAL);
6283 if (test_bit(WriteMostly, &rdev->flags))
6284 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6285 if (test_bit(FailFast, &rdev->flags))
6286 info.state |= (1<<MD_DISK_FAILFAST);
6288 info.major = info.minor = 0;
6289 info.raid_disk = -1;
6290 info.state = (1<<MD_DISK_REMOVED);
6294 if (copy_to_user(arg, &info, sizeof(info)))
6300 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6302 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6303 struct md_rdev *rdev;
6304 dev_t dev = MKDEV(info->major,info->minor);
6306 if (mddev_is_clustered(mddev) &&
6307 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6308 pr_warn("%s: Cannot add to clustered mddev.\n",
6313 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6316 if (!mddev->raid_disks) {
6318 /* expecting a device which has a superblock */
6319 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6321 pr_warn("md: md_import_device returned %ld\n",
6323 return PTR_ERR(rdev);
6325 if (!list_empty(&mddev->disks)) {
6326 struct md_rdev *rdev0
6327 = list_entry(mddev->disks.next,
6328 struct md_rdev, same_set);
6329 err = super_types[mddev->major_version]
6330 .load_super(rdev, rdev0, mddev->minor_version);
6332 pr_warn("md: %s has different UUID to %s\n",
6333 bdevname(rdev->bdev,b),
6334 bdevname(rdev0->bdev,b2));
6339 err = bind_rdev_to_array(rdev, mddev);
6346 * add_new_disk can be used once the array is assembled
6347 * to add "hot spares". They must already have a superblock
6352 if (!mddev->pers->hot_add_disk) {
6353 pr_warn("%s: personality does not support diskops!\n",
6357 if (mddev->persistent)
6358 rdev = md_import_device(dev, mddev->major_version,
6359 mddev->minor_version);
6361 rdev = md_import_device(dev, -1, -1);
6363 pr_warn("md: md_import_device returned %ld\n",
6365 return PTR_ERR(rdev);
6367 /* set saved_raid_disk if appropriate */
6368 if (!mddev->persistent) {
6369 if (info->state & (1<<MD_DISK_SYNC) &&
6370 info->raid_disk < mddev->raid_disks) {
6371 rdev->raid_disk = info->raid_disk;
6372 set_bit(In_sync, &rdev->flags);
6373 clear_bit(Bitmap_sync, &rdev->flags);
6375 rdev->raid_disk = -1;
6376 rdev->saved_raid_disk = rdev->raid_disk;
6378 super_types[mddev->major_version].
6379 validate_super(mddev, rdev);
6380 if ((info->state & (1<<MD_DISK_SYNC)) &&
6381 rdev->raid_disk != info->raid_disk) {
6382 /* This was a hot-add request, but events doesn't
6383 * match, so reject it.
6389 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6390 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6391 set_bit(WriteMostly, &rdev->flags);
6393 clear_bit(WriteMostly, &rdev->flags);
6394 if (info->state & (1<<MD_DISK_FAILFAST))
6395 set_bit(FailFast, &rdev->flags);
6397 clear_bit(FailFast, &rdev->flags);
6399 if (info->state & (1<<MD_DISK_JOURNAL)) {
6400 struct md_rdev *rdev2;
6401 bool has_journal = false;
6403 /* make sure no existing journal disk */
6404 rdev_for_each(rdev2, mddev) {
6405 if (test_bit(Journal, &rdev2->flags)) {
6410 if (has_journal || mddev->bitmap) {
6414 set_bit(Journal, &rdev->flags);
6417 * check whether the device shows up in other nodes
6419 if (mddev_is_clustered(mddev)) {
6420 if (info->state & (1 << MD_DISK_CANDIDATE))
6421 set_bit(Candidate, &rdev->flags);
6422 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6423 /* --add initiated by this node */
6424 err = md_cluster_ops->add_new_disk(mddev, rdev);
6432 rdev->raid_disk = -1;
6433 err = bind_rdev_to_array(rdev, mddev);
6438 if (mddev_is_clustered(mddev)) {
6439 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6441 err = md_cluster_ops->new_disk_ack(mddev,
6444 md_kick_rdev_from_array(rdev);
6448 md_cluster_ops->add_new_disk_cancel(mddev);
6450 err = add_bound_rdev(rdev);
6454 err = add_bound_rdev(rdev);
6459 /* otherwise, add_new_disk is only allowed
6460 * for major_version==0 superblocks
6462 if (mddev->major_version != 0) {
6463 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6467 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6469 rdev = md_import_device(dev, -1, 0);
6471 pr_warn("md: error, md_import_device() returned %ld\n",
6473 return PTR_ERR(rdev);
6475 rdev->desc_nr = info->number;
6476 if (info->raid_disk < mddev->raid_disks)
6477 rdev->raid_disk = info->raid_disk;
6479 rdev->raid_disk = -1;
6481 if (rdev->raid_disk < mddev->raid_disks)
6482 if (info->state & (1<<MD_DISK_SYNC))
6483 set_bit(In_sync, &rdev->flags);
6485 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6486 set_bit(WriteMostly, &rdev->flags);
6487 if (info->state & (1<<MD_DISK_FAILFAST))
6488 set_bit(FailFast, &rdev->flags);
6490 if (!mddev->persistent) {
6491 pr_debug("md: nonpersistent superblock ...\n");
6492 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6494 rdev->sb_start = calc_dev_sboffset(rdev);
6495 rdev->sectors = rdev->sb_start;
6497 err = bind_rdev_to_array(rdev, mddev);
6507 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6509 char b[BDEVNAME_SIZE];
6510 struct md_rdev *rdev;
6515 rdev = find_rdev(mddev, dev);
6519 if (rdev->raid_disk < 0)
6522 clear_bit(Blocked, &rdev->flags);
6523 remove_and_add_spares(mddev, rdev);
6525 if (rdev->raid_disk >= 0)
6529 if (mddev_is_clustered(mddev))
6530 md_cluster_ops->remove_disk(mddev, rdev);
6532 md_kick_rdev_from_array(rdev);
6533 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6535 md_wakeup_thread(mddev->thread);
6537 md_update_sb(mddev, 1);
6538 md_new_event(mddev);
6542 pr_debug("md: cannot remove active disk %s from %s ...\n",
6543 bdevname(rdev->bdev,b), mdname(mddev));
6547 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6549 char b[BDEVNAME_SIZE];
6551 struct md_rdev *rdev;
6556 if (mddev->major_version != 0) {
6557 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6561 if (!mddev->pers->hot_add_disk) {
6562 pr_warn("%s: personality does not support diskops!\n",
6567 rdev = md_import_device(dev, -1, 0);
6569 pr_warn("md: error, md_import_device() returned %ld\n",
6574 if (mddev->persistent)
6575 rdev->sb_start = calc_dev_sboffset(rdev);
6577 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6579 rdev->sectors = rdev->sb_start;
6581 if (test_bit(Faulty, &rdev->flags)) {
6582 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6583 bdevname(rdev->bdev,b), mdname(mddev));
6588 clear_bit(In_sync, &rdev->flags);
6590 rdev->saved_raid_disk = -1;
6591 err = bind_rdev_to_array(rdev, mddev);
6596 * The rest should better be atomic, we can have disk failures
6597 * noticed in interrupt contexts ...
6600 rdev->raid_disk = -1;
6602 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6604 md_update_sb(mddev, 1);
6606 * Kick recovery, maybe this spare has to be added to the
6607 * array immediately.
6609 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6610 md_wakeup_thread(mddev->thread);
6611 md_new_event(mddev);
6619 static int set_bitmap_file(struct mddev *mddev, int fd)
6624 if (!mddev->pers->quiesce || !mddev->thread)
6626 if (mddev->recovery || mddev->sync_thread)
6628 /* we should be able to change the bitmap.. */
6632 struct inode *inode;
6635 if (mddev->bitmap || mddev->bitmap_info.file)
6636 return -EEXIST; /* cannot add when bitmap is present */
6640 pr_warn("%s: error: failed to get bitmap file\n",
6645 inode = f->f_mapping->host;
6646 if (!S_ISREG(inode->i_mode)) {
6647 pr_warn("%s: error: bitmap file must be a regular file\n",
6650 } else if (!(f->f_mode & FMODE_WRITE)) {
6651 pr_warn("%s: error: bitmap file must open for write\n",
6654 } else if (atomic_read(&inode->i_writecount) != 1) {
6655 pr_warn("%s: error: bitmap file is already in use\n",
6663 mddev->bitmap_info.file = f;
6664 mddev->bitmap_info.offset = 0; /* file overrides offset */
6665 } else if (mddev->bitmap == NULL)
6666 return -ENOENT; /* cannot remove what isn't there */
6670 struct bitmap *bitmap;
6672 bitmap = md_bitmap_create(mddev, -1);
6673 mddev_suspend(mddev);
6674 if (!IS_ERR(bitmap)) {
6675 mddev->bitmap = bitmap;
6676 err = md_bitmap_load(mddev);
6678 err = PTR_ERR(bitmap);
6680 md_bitmap_destroy(mddev);
6683 mddev_resume(mddev);
6684 } else if (fd < 0) {
6685 mddev_suspend(mddev);
6686 md_bitmap_destroy(mddev);
6687 mddev_resume(mddev);
6691 struct file *f = mddev->bitmap_info.file;
6693 spin_lock(&mddev->lock);
6694 mddev->bitmap_info.file = NULL;
6695 spin_unlock(&mddev->lock);
6704 * set_array_info is used two different ways
6705 * The original usage is when creating a new array.
6706 * In this usage, raid_disks is > 0 and it together with
6707 * level, size, not_persistent,layout,chunksize determine the
6708 * shape of the array.
6709 * This will always create an array with a type-0.90.0 superblock.
6710 * The newer usage is when assembling an array.
6711 * In this case raid_disks will be 0, and the major_version field is
6712 * use to determine which style super-blocks are to be found on the devices.
6713 * The minor and patch _version numbers are also kept incase the
6714 * super_block handler wishes to interpret them.
6716 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6719 if (info->raid_disks == 0) {
6720 /* just setting version number for superblock loading */
6721 if (info->major_version < 0 ||
6722 info->major_version >= ARRAY_SIZE(super_types) ||
6723 super_types[info->major_version].name == NULL) {
6724 /* maybe try to auto-load a module? */
6725 pr_warn("md: superblock version %d not known\n",
6726 info->major_version);
6729 mddev->major_version = info->major_version;
6730 mddev->minor_version = info->minor_version;
6731 mddev->patch_version = info->patch_version;
6732 mddev->persistent = !info->not_persistent;
6733 /* ensure mddev_put doesn't delete this now that there
6734 * is some minimal configuration.
6736 mddev->ctime = ktime_get_real_seconds();
6739 mddev->major_version = MD_MAJOR_VERSION;
6740 mddev->minor_version = MD_MINOR_VERSION;
6741 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6742 mddev->ctime = ktime_get_real_seconds();
6744 mddev->level = info->level;
6745 mddev->clevel[0] = 0;
6746 mddev->dev_sectors = 2 * (sector_t)info->size;
6747 mddev->raid_disks = info->raid_disks;
6748 /* don't set md_minor, it is determined by which /dev/md* was
6751 if (info->state & (1<<MD_SB_CLEAN))
6752 mddev->recovery_cp = MaxSector;
6754 mddev->recovery_cp = 0;
6755 mddev->persistent = ! info->not_persistent;
6756 mddev->external = 0;
6758 mddev->layout = info->layout;
6759 mddev->chunk_sectors = info->chunk_size >> 9;
6761 if (mddev->persistent) {
6762 mddev->max_disks = MD_SB_DISKS;
6764 mddev->sb_flags = 0;
6766 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6768 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6769 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6770 mddev->bitmap_info.offset = 0;
6772 mddev->reshape_position = MaxSector;
6775 * Generate a 128 bit UUID
6777 get_random_bytes(mddev->uuid, 16);
6779 mddev->new_level = mddev->level;
6780 mddev->new_chunk_sectors = mddev->chunk_sectors;
6781 mddev->new_layout = mddev->layout;
6782 mddev->delta_disks = 0;
6783 mddev->reshape_backwards = 0;
6788 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6790 lockdep_assert_held(&mddev->reconfig_mutex);
6792 if (mddev->external_size)
6795 mddev->array_sectors = array_sectors;
6797 EXPORT_SYMBOL(md_set_array_sectors);
6799 static int update_size(struct mddev *mddev, sector_t num_sectors)
6801 struct md_rdev *rdev;
6803 int fit = (num_sectors == 0);
6804 sector_t old_dev_sectors = mddev->dev_sectors;
6806 if (mddev->pers->resize == NULL)
6808 /* The "num_sectors" is the number of sectors of each device that
6809 * is used. This can only make sense for arrays with redundancy.
6810 * linear and raid0 always use whatever space is available. We can only
6811 * consider changing this number if no resync or reconstruction is
6812 * happening, and if the new size is acceptable. It must fit before the
6813 * sb_start or, if that is <data_offset, it must fit before the size
6814 * of each device. If num_sectors is zero, we find the largest size
6817 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6823 rdev_for_each(rdev, mddev) {
6824 sector_t avail = rdev->sectors;
6826 if (fit && (num_sectors == 0 || num_sectors > avail))
6827 num_sectors = avail;
6828 if (avail < num_sectors)
6831 rv = mddev->pers->resize(mddev, num_sectors);
6833 if (mddev_is_clustered(mddev))
6834 md_cluster_ops->update_size(mddev, old_dev_sectors);
6835 else if (mddev->queue) {
6836 set_capacity(mddev->gendisk, mddev->array_sectors);
6837 revalidate_disk(mddev->gendisk);
6843 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6846 struct md_rdev *rdev;
6847 /* change the number of raid disks */
6848 if (mddev->pers->check_reshape == NULL)
6852 if (raid_disks <= 0 ||
6853 (mddev->max_disks && raid_disks >= mddev->max_disks))
6855 if (mddev->sync_thread ||
6856 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6857 mddev->reshape_position != MaxSector)
6860 rdev_for_each(rdev, mddev) {
6861 if (mddev->raid_disks < raid_disks &&
6862 rdev->data_offset < rdev->new_data_offset)
6864 if (mddev->raid_disks > raid_disks &&
6865 rdev->data_offset > rdev->new_data_offset)
6869 mddev->delta_disks = raid_disks - mddev->raid_disks;
6870 if (mddev->delta_disks < 0)
6871 mddev->reshape_backwards = 1;
6872 else if (mddev->delta_disks > 0)
6873 mddev->reshape_backwards = 0;
6875 rv = mddev->pers->check_reshape(mddev);
6877 mddev->delta_disks = 0;
6878 mddev->reshape_backwards = 0;
6884 * update_array_info is used to change the configuration of an
6886 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6887 * fields in the info are checked against the array.
6888 * Any differences that cannot be handled will cause an error.
6889 * Normally, only one change can be managed at a time.
6891 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6897 /* calculate expected state,ignoring low bits */
6898 if (mddev->bitmap && mddev->bitmap_info.offset)
6899 state |= (1 << MD_SB_BITMAP_PRESENT);
6901 if (mddev->major_version != info->major_version ||
6902 mddev->minor_version != info->minor_version ||
6903 /* mddev->patch_version != info->patch_version || */
6904 mddev->ctime != info->ctime ||
6905 mddev->level != info->level ||
6906 /* mddev->layout != info->layout || */
6907 mddev->persistent != !info->not_persistent ||
6908 mddev->chunk_sectors != info->chunk_size >> 9 ||
6909 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6910 ((state^info->state) & 0xfffffe00)
6913 /* Check there is only one change */
6914 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6916 if (mddev->raid_disks != info->raid_disks)
6918 if (mddev->layout != info->layout)
6920 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6927 if (mddev->layout != info->layout) {
6929 * we don't need to do anything at the md level, the
6930 * personality will take care of it all.
6932 if (mddev->pers->check_reshape == NULL)
6935 mddev->new_layout = info->layout;
6936 rv = mddev->pers->check_reshape(mddev);
6938 mddev->new_layout = mddev->layout;
6942 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6943 rv = update_size(mddev, (sector_t)info->size * 2);
6945 if (mddev->raid_disks != info->raid_disks)
6946 rv = update_raid_disks(mddev, info->raid_disks);
6948 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6949 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6953 if (mddev->recovery || mddev->sync_thread) {
6957 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6958 struct bitmap *bitmap;
6959 /* add the bitmap */
6960 if (mddev->bitmap) {
6964 if (mddev->bitmap_info.default_offset == 0) {
6968 mddev->bitmap_info.offset =
6969 mddev->bitmap_info.default_offset;
6970 mddev->bitmap_info.space =
6971 mddev->bitmap_info.default_space;
6972 bitmap = md_bitmap_create(mddev, -1);
6973 mddev_suspend(mddev);
6974 if (!IS_ERR(bitmap)) {
6975 mddev->bitmap = bitmap;
6976 rv = md_bitmap_load(mddev);
6978 rv = PTR_ERR(bitmap);
6980 md_bitmap_destroy(mddev);
6981 mddev_resume(mddev);
6983 /* remove the bitmap */
6984 if (!mddev->bitmap) {
6988 if (mddev->bitmap->storage.file) {
6992 if (mddev->bitmap_info.nodes) {
6993 /* hold PW on all the bitmap lock */
6994 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6995 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6997 md_cluster_ops->unlock_all_bitmaps(mddev);
7001 mddev->bitmap_info.nodes = 0;
7002 md_cluster_ops->leave(mddev);
7004 mddev_suspend(mddev);
7005 md_bitmap_destroy(mddev);
7006 mddev_resume(mddev);
7007 mddev->bitmap_info.offset = 0;
7010 md_update_sb(mddev, 1);
7016 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7018 struct md_rdev *rdev;
7021 if (mddev->pers == NULL)
7025 rdev = md_find_rdev_rcu(mddev, dev);
7029 md_error(mddev, rdev);
7030 if (!test_bit(Faulty, &rdev->flags))
7038 * We have a problem here : there is no easy way to give a CHS
7039 * virtual geometry. We currently pretend that we have a 2 heads
7040 * 4 sectors (with a BIG number of cylinders...). This drives
7041 * dosfs just mad... ;-)
7043 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7045 struct mddev *mddev = bdev->bd_disk->private_data;
7049 geo->cylinders = mddev->array_sectors / 8;
7053 static inline bool md_ioctl_valid(unsigned int cmd)
7058 case GET_ARRAY_INFO:
7059 case GET_BITMAP_FILE:
7062 case HOT_REMOVE_DISK:
7065 case RESTART_ARRAY_RW:
7067 case SET_ARRAY_INFO:
7068 case SET_BITMAP_FILE:
7069 case SET_DISK_FAULTY:
7072 case CLUSTERED_DISK_NACK:
7079 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7080 unsigned int cmd, unsigned long arg)
7083 void __user *argp = (void __user *)arg;
7084 struct mddev *mddev = NULL;
7086 bool did_set_md_closing = false;
7088 if (!md_ioctl_valid(cmd))
7093 case GET_ARRAY_INFO:
7097 if (!capable(CAP_SYS_ADMIN))
7102 * Commands dealing with the RAID driver but not any
7107 err = get_version(argp);
7113 autostart_arrays(arg);
7120 * Commands creating/starting a new array:
7123 mddev = bdev->bd_disk->private_data;
7130 /* Some actions do not requires the mutex */
7132 case GET_ARRAY_INFO:
7133 if (!mddev->raid_disks && !mddev->external)
7136 err = get_array_info(mddev, argp);
7140 if (!mddev->raid_disks && !mddev->external)
7143 err = get_disk_info(mddev, argp);
7146 case SET_DISK_FAULTY:
7147 err = set_disk_faulty(mddev, new_decode_dev(arg));
7150 case GET_BITMAP_FILE:
7151 err = get_bitmap_file(mddev, argp);
7156 if (cmd == ADD_NEW_DISK)
7157 /* need to ensure md_delayed_delete() has completed */
7158 flush_workqueue(md_misc_wq);
7160 if (cmd == HOT_REMOVE_DISK)
7161 /* need to ensure recovery thread has run */
7162 wait_event_interruptible_timeout(mddev->sb_wait,
7163 !test_bit(MD_RECOVERY_NEEDED,
7165 msecs_to_jiffies(5000));
7166 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7167 /* Need to flush page cache, and ensure no-one else opens
7170 mutex_lock(&mddev->open_mutex);
7171 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7172 mutex_unlock(&mddev->open_mutex);
7176 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7177 set_bit(MD_CLOSING, &mddev->flags);
7178 did_set_md_closing = true;
7179 mutex_unlock(&mddev->open_mutex);
7180 sync_blockdev(bdev);
7182 err = mddev_lock(mddev);
7184 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7189 if (cmd == SET_ARRAY_INFO) {
7190 mdu_array_info_t info;
7192 memset(&info, 0, sizeof(info));
7193 else if (copy_from_user(&info, argp, sizeof(info))) {
7198 err = update_array_info(mddev, &info);
7200 pr_warn("md: couldn't update array info. %d\n", err);
7205 if (!list_empty(&mddev->disks)) {
7206 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7210 if (mddev->raid_disks) {
7211 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7215 err = set_array_info(mddev, &info);
7217 pr_warn("md: couldn't set array info. %d\n", err);
7224 * Commands querying/configuring an existing array:
7226 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7227 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7228 if ((!mddev->raid_disks && !mddev->external)
7229 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7230 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7231 && cmd != GET_BITMAP_FILE) {
7237 * Commands even a read-only array can execute:
7240 case RESTART_ARRAY_RW:
7241 err = restart_array(mddev);
7245 err = do_md_stop(mddev, 0, bdev);
7249 err = md_set_readonly(mddev, bdev);
7252 case HOT_REMOVE_DISK:
7253 err = hot_remove_disk(mddev, new_decode_dev(arg));
7257 /* We can support ADD_NEW_DISK on read-only arrays
7258 * only if we are re-adding a preexisting device.
7259 * So require mddev->pers and MD_DISK_SYNC.
7262 mdu_disk_info_t info;
7263 if (copy_from_user(&info, argp, sizeof(info)))
7265 else if (!(info.state & (1<<MD_DISK_SYNC)))
7266 /* Need to clear read-only for this */
7269 err = add_new_disk(mddev, &info);
7275 if (get_user(ro, (int __user *)(arg))) {
7281 /* if the bdev is going readonly the value of mddev->ro
7282 * does not matter, no writes are coming
7287 /* are we are already prepared for writes? */
7291 /* transitioning to readauto need only happen for
7292 * arrays that call md_write_start
7295 err = restart_array(mddev);
7298 set_disk_ro(mddev->gendisk, 0);
7305 * The remaining ioctls are changing the state of the
7306 * superblock, so we do not allow them on read-only arrays.
7308 if (mddev->ro && mddev->pers) {
7309 if (mddev->ro == 2) {
7311 sysfs_notify_dirent_safe(mddev->sysfs_state);
7312 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7313 /* mddev_unlock will wake thread */
7314 /* If a device failed while we were read-only, we
7315 * need to make sure the metadata is updated now.
7317 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7318 mddev_unlock(mddev);
7319 wait_event(mddev->sb_wait,
7320 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7321 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7322 mddev_lock_nointr(mddev);
7333 mdu_disk_info_t info;
7334 if (copy_from_user(&info, argp, sizeof(info)))
7337 err = add_new_disk(mddev, &info);
7341 case CLUSTERED_DISK_NACK:
7342 if (mddev_is_clustered(mddev))
7343 md_cluster_ops->new_disk_ack(mddev, false);
7349 err = hot_add_disk(mddev, new_decode_dev(arg));
7353 err = do_md_run(mddev);
7356 case SET_BITMAP_FILE:
7357 err = set_bitmap_file(mddev, (int)arg);
7366 if (mddev->hold_active == UNTIL_IOCTL &&
7368 mddev->hold_active = 0;
7369 mddev_unlock(mddev);
7371 if(did_set_md_closing)
7372 clear_bit(MD_CLOSING, &mddev->flags);
7375 #ifdef CONFIG_COMPAT
7376 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7377 unsigned int cmd, unsigned long arg)
7380 case HOT_REMOVE_DISK:
7382 case SET_DISK_FAULTY:
7383 case SET_BITMAP_FILE:
7384 /* These take in integer arg, do not convert */
7387 arg = (unsigned long)compat_ptr(arg);
7391 return md_ioctl(bdev, mode, cmd, arg);
7393 #endif /* CONFIG_COMPAT */
7395 static int md_open(struct block_device *bdev, fmode_t mode)
7398 * Succeed if we can lock the mddev, which confirms that
7399 * it isn't being stopped right now.
7401 struct mddev *mddev = mddev_find(bdev->bd_dev);
7407 if (mddev->gendisk != bdev->bd_disk) {
7408 /* we are racing with mddev_put which is discarding this
7412 /* Wait until bdev->bd_disk is definitely gone */
7413 flush_workqueue(md_misc_wq);
7414 /* Then retry the open from the top */
7415 return -ERESTARTSYS;
7417 BUG_ON(mddev != bdev->bd_disk->private_data);
7419 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7422 if (test_bit(MD_CLOSING, &mddev->flags)) {
7423 mutex_unlock(&mddev->open_mutex);
7429 atomic_inc(&mddev->openers);
7430 mutex_unlock(&mddev->open_mutex);
7432 check_disk_change(bdev);
7439 static void md_release(struct gendisk *disk, fmode_t mode)
7441 struct mddev *mddev = disk->private_data;
7444 atomic_dec(&mddev->openers);
7448 static int md_media_changed(struct gendisk *disk)
7450 struct mddev *mddev = disk->private_data;
7452 return mddev->changed;
7455 static int md_revalidate(struct gendisk *disk)
7457 struct mddev *mddev = disk->private_data;
7462 static const struct block_device_operations md_fops =
7464 .owner = THIS_MODULE,
7466 .release = md_release,
7468 #ifdef CONFIG_COMPAT
7469 .compat_ioctl = md_compat_ioctl,
7471 .getgeo = md_getgeo,
7472 .media_changed = md_media_changed,
7473 .revalidate_disk= md_revalidate,
7476 static int md_thread(void *arg)
7478 struct md_thread *thread = arg;
7481 * md_thread is a 'system-thread', it's priority should be very
7482 * high. We avoid resource deadlocks individually in each
7483 * raid personality. (RAID5 does preallocation) We also use RR and
7484 * the very same RT priority as kswapd, thus we will never get
7485 * into a priority inversion deadlock.
7487 * we definitely have to have equal or higher priority than
7488 * bdflush, otherwise bdflush will deadlock if there are too
7489 * many dirty RAID5 blocks.
7492 allow_signal(SIGKILL);
7493 while (!kthread_should_stop()) {
7495 /* We need to wait INTERRUPTIBLE so that
7496 * we don't add to the load-average.
7497 * That means we need to be sure no signals are
7500 if (signal_pending(current))
7501 flush_signals(current);
7503 wait_event_interruptible_timeout
7505 test_bit(THREAD_WAKEUP, &thread->flags)
7506 || kthread_should_stop() || kthread_should_park(),
7509 clear_bit(THREAD_WAKEUP, &thread->flags);
7510 if (kthread_should_park())
7512 if (!kthread_should_stop())
7513 thread->run(thread);
7519 void md_wakeup_thread(struct md_thread *thread)
7522 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7523 set_bit(THREAD_WAKEUP, &thread->flags);
7524 wake_up(&thread->wqueue);
7527 EXPORT_SYMBOL(md_wakeup_thread);
7529 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7530 struct mddev *mddev, const char *name)
7532 struct md_thread *thread;
7534 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7538 init_waitqueue_head(&thread->wqueue);
7541 thread->mddev = mddev;
7542 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7543 thread->tsk = kthread_run(md_thread, thread,
7545 mdname(thread->mddev),
7547 if (IS_ERR(thread->tsk)) {
7553 EXPORT_SYMBOL(md_register_thread);
7555 void md_unregister_thread(struct md_thread **threadp)
7557 struct md_thread *thread = *threadp;
7560 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7561 /* Locking ensures that mddev_unlock does not wake_up a
7562 * non-existent thread
7564 spin_lock(&pers_lock);
7566 spin_unlock(&pers_lock);
7568 kthread_stop(thread->tsk);
7571 EXPORT_SYMBOL(md_unregister_thread);
7573 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7575 if (!rdev || test_bit(Faulty, &rdev->flags))
7578 if (!mddev->pers || !mddev->pers->error_handler)
7580 mddev->pers->error_handler(mddev,rdev);
7581 if (mddev->degraded)
7582 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7583 sysfs_notify_dirent_safe(rdev->sysfs_state);
7584 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7585 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7586 md_wakeup_thread(mddev->thread);
7587 if (mddev->event_work.func)
7588 queue_work(md_misc_wq, &mddev->event_work);
7589 md_new_event(mddev);
7591 EXPORT_SYMBOL(md_error);
7593 /* seq_file implementation /proc/mdstat */
7595 static void status_unused(struct seq_file *seq)
7598 struct md_rdev *rdev;
7600 seq_printf(seq, "unused devices: ");
7602 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7603 char b[BDEVNAME_SIZE];
7605 seq_printf(seq, "%s ",
7606 bdevname(rdev->bdev,b));
7609 seq_printf(seq, "<none>");
7611 seq_printf(seq, "\n");
7614 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7616 sector_t max_sectors, resync, res;
7617 unsigned long dt, db;
7620 unsigned int per_milli;
7622 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7623 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7624 max_sectors = mddev->resync_max_sectors;
7626 max_sectors = mddev->dev_sectors;
7628 resync = mddev->curr_resync;
7630 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7631 /* Still cleaning up */
7632 resync = max_sectors;
7633 } else if (resync > max_sectors)
7634 resync = max_sectors;
7636 resync -= atomic_read(&mddev->recovery_active);
7639 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7640 struct md_rdev *rdev;
7642 rdev_for_each(rdev, mddev)
7643 if (rdev->raid_disk >= 0 &&
7644 !test_bit(Faulty, &rdev->flags) &&
7645 rdev->recovery_offset != MaxSector &&
7646 rdev->recovery_offset) {
7647 seq_printf(seq, "\trecover=REMOTE");
7650 if (mddev->reshape_position != MaxSector)
7651 seq_printf(seq, "\treshape=REMOTE");
7653 seq_printf(seq, "\tresync=REMOTE");
7656 if (mddev->recovery_cp < MaxSector) {
7657 seq_printf(seq, "\tresync=PENDING");
7663 seq_printf(seq, "\tresync=DELAYED");
7667 WARN_ON(max_sectors == 0);
7668 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7669 * in a sector_t, and (max_sectors>>scale) will fit in a
7670 * u32, as those are the requirements for sector_div.
7671 * Thus 'scale' must be at least 10
7674 if (sizeof(sector_t) > sizeof(unsigned long)) {
7675 while ( max_sectors/2 > (1ULL<<(scale+32)))
7678 res = (resync>>scale)*1000;
7679 sector_div(res, (u32)((max_sectors>>scale)+1));
7683 int i, x = per_milli/50, y = 20-x;
7684 seq_printf(seq, "[");
7685 for (i = 0; i < x; i++)
7686 seq_printf(seq, "=");
7687 seq_printf(seq, ">");
7688 for (i = 0; i < y; i++)
7689 seq_printf(seq, ".");
7690 seq_printf(seq, "] ");
7692 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7693 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7695 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7697 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7698 "resync" : "recovery"))),
7699 per_milli/10, per_milli % 10,
7700 (unsigned long long) resync/2,
7701 (unsigned long long) max_sectors/2);
7704 * dt: time from mark until now
7705 * db: blocks written from mark until now
7706 * rt: remaining time
7708 * rt is a sector_t, so could be 32bit or 64bit.
7709 * So we divide before multiply in case it is 32bit and close
7711 * We scale the divisor (db) by 32 to avoid losing precision
7712 * near the end of resync when the number of remaining sectors
7714 * We then divide rt by 32 after multiplying by db to compensate.
7715 * The '+1' avoids division by zero if db is very small.
7717 dt = ((jiffies - mddev->resync_mark) / HZ);
7719 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7720 - mddev->resync_mark_cnt;
7722 rt = max_sectors - resync; /* number of remaining sectors */
7723 sector_div(rt, db/32+1);
7727 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7728 ((unsigned long)rt % 60)/6);
7730 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7734 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7736 struct list_head *tmp;
7738 struct mddev *mddev;
7746 spin_lock(&all_mddevs_lock);
7747 list_for_each(tmp,&all_mddevs)
7749 mddev = list_entry(tmp, struct mddev, all_mddevs);
7751 spin_unlock(&all_mddevs_lock);
7754 spin_unlock(&all_mddevs_lock);
7756 return (void*)2;/* tail */
7760 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7762 struct list_head *tmp;
7763 struct mddev *next_mddev, *mddev = v;
7769 spin_lock(&all_mddevs_lock);
7771 tmp = all_mddevs.next;
7773 tmp = mddev->all_mddevs.next;
7774 if (tmp != &all_mddevs)
7775 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7777 next_mddev = (void*)2;
7780 spin_unlock(&all_mddevs_lock);
7788 static void md_seq_stop(struct seq_file *seq, void *v)
7790 struct mddev *mddev = v;
7792 if (mddev && v != (void*)1 && v != (void*)2)
7796 static int md_seq_show(struct seq_file *seq, void *v)
7798 struct mddev *mddev = v;
7800 struct md_rdev *rdev;
7802 if (v == (void*)1) {
7803 struct md_personality *pers;
7804 seq_printf(seq, "Personalities : ");
7805 spin_lock(&pers_lock);
7806 list_for_each_entry(pers, &pers_list, list)
7807 seq_printf(seq, "[%s] ", pers->name);
7809 spin_unlock(&pers_lock);
7810 seq_printf(seq, "\n");
7811 seq->poll_event = atomic_read(&md_event_count);
7814 if (v == (void*)2) {
7819 spin_lock(&mddev->lock);
7820 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7821 seq_printf(seq, "%s : %sactive", mdname(mddev),
7822 mddev->pers ? "" : "in");
7825 seq_printf(seq, " (read-only)");
7827 seq_printf(seq, " (auto-read-only)");
7828 seq_printf(seq, " %s", mddev->pers->name);
7833 rdev_for_each_rcu(rdev, mddev) {
7834 char b[BDEVNAME_SIZE];
7835 seq_printf(seq, " %s[%d]",
7836 bdevname(rdev->bdev,b), rdev->desc_nr);
7837 if (test_bit(WriteMostly, &rdev->flags))
7838 seq_printf(seq, "(W)");
7839 if (test_bit(Journal, &rdev->flags))
7840 seq_printf(seq, "(J)");
7841 if (test_bit(Faulty, &rdev->flags)) {
7842 seq_printf(seq, "(F)");
7845 if (rdev->raid_disk < 0)
7846 seq_printf(seq, "(S)"); /* spare */
7847 if (test_bit(Replacement, &rdev->flags))
7848 seq_printf(seq, "(R)");
7849 sectors += rdev->sectors;
7853 if (!list_empty(&mddev->disks)) {
7855 seq_printf(seq, "\n %llu blocks",
7856 (unsigned long long)
7857 mddev->array_sectors / 2);
7859 seq_printf(seq, "\n %llu blocks",
7860 (unsigned long long)sectors / 2);
7862 if (mddev->persistent) {
7863 if (mddev->major_version != 0 ||
7864 mddev->minor_version != 90) {
7865 seq_printf(seq," super %d.%d",
7866 mddev->major_version,
7867 mddev->minor_version);
7869 } else if (mddev->external)
7870 seq_printf(seq, " super external:%s",
7871 mddev->metadata_type);
7873 seq_printf(seq, " super non-persistent");
7876 mddev->pers->status(seq, mddev);
7877 seq_printf(seq, "\n ");
7878 if (mddev->pers->sync_request) {
7879 if (status_resync(seq, mddev))
7880 seq_printf(seq, "\n ");
7883 seq_printf(seq, "\n ");
7885 md_bitmap_status(seq, mddev->bitmap);
7887 seq_printf(seq, "\n");
7889 spin_unlock(&mddev->lock);
7894 static const struct seq_operations md_seq_ops = {
7895 .start = md_seq_start,
7896 .next = md_seq_next,
7897 .stop = md_seq_stop,
7898 .show = md_seq_show,
7901 static int md_seq_open(struct inode *inode, struct file *file)
7903 struct seq_file *seq;
7906 error = seq_open(file, &md_seq_ops);
7910 seq = file->private_data;
7911 seq->poll_event = atomic_read(&md_event_count);
7915 static int md_unloading;
7916 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
7918 struct seq_file *seq = filp->private_data;
7922 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
7923 poll_wait(filp, &md_event_waiters, wait);
7925 /* always allow read */
7926 mask = EPOLLIN | EPOLLRDNORM;
7928 if (seq->poll_event != atomic_read(&md_event_count))
7929 mask |= EPOLLERR | EPOLLPRI;
7933 static const struct file_operations md_seq_fops = {
7934 .owner = THIS_MODULE,
7935 .open = md_seq_open,
7937 .llseek = seq_lseek,
7938 .release = seq_release,
7939 .poll = mdstat_poll,
7942 int register_md_personality(struct md_personality *p)
7944 pr_debug("md: %s personality registered for level %d\n",
7946 spin_lock(&pers_lock);
7947 list_add_tail(&p->list, &pers_list);
7948 spin_unlock(&pers_lock);
7951 EXPORT_SYMBOL(register_md_personality);
7953 int unregister_md_personality(struct md_personality *p)
7955 pr_debug("md: %s personality unregistered\n", p->name);
7956 spin_lock(&pers_lock);
7957 list_del_init(&p->list);
7958 spin_unlock(&pers_lock);
7961 EXPORT_SYMBOL(unregister_md_personality);
7963 int register_md_cluster_operations(struct md_cluster_operations *ops,
7964 struct module *module)
7967 spin_lock(&pers_lock);
7968 if (md_cluster_ops != NULL)
7971 md_cluster_ops = ops;
7972 md_cluster_mod = module;
7974 spin_unlock(&pers_lock);
7977 EXPORT_SYMBOL(register_md_cluster_operations);
7979 int unregister_md_cluster_operations(void)
7981 spin_lock(&pers_lock);
7982 md_cluster_ops = NULL;
7983 spin_unlock(&pers_lock);
7986 EXPORT_SYMBOL(unregister_md_cluster_operations);
7988 int md_setup_cluster(struct mddev *mddev, int nodes)
7990 if (!md_cluster_ops)
7991 request_module("md-cluster");
7992 spin_lock(&pers_lock);
7993 /* ensure module won't be unloaded */
7994 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7995 pr_warn("can't find md-cluster module or get it's reference.\n");
7996 spin_unlock(&pers_lock);
7999 spin_unlock(&pers_lock);
8001 return md_cluster_ops->join(mddev, nodes);
8004 void md_cluster_stop(struct mddev *mddev)
8006 if (!md_cluster_ops)
8008 md_cluster_ops->leave(mddev);
8009 module_put(md_cluster_mod);
8012 static int is_mddev_idle(struct mddev *mddev, int init)
8014 struct md_rdev *rdev;
8020 rdev_for_each_rcu(rdev, mddev) {
8021 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8022 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8023 atomic_read(&disk->sync_io);
8024 /* sync IO will cause sync_io to increase before the disk_stats
8025 * as sync_io is counted when a request starts, and
8026 * disk_stats is counted when it completes.
8027 * So resync activity will cause curr_events to be smaller than
8028 * when there was no such activity.
8029 * non-sync IO will cause disk_stat to increase without
8030 * increasing sync_io so curr_events will (eventually)
8031 * be larger than it was before. Once it becomes
8032 * substantially larger, the test below will cause
8033 * the array to appear non-idle, and resync will slow
8035 * If there is a lot of outstanding resync activity when
8036 * we set last_event to curr_events, then all that activity
8037 * completing might cause the array to appear non-idle
8038 * and resync will be slowed down even though there might
8039 * not have been non-resync activity. This will only
8040 * happen once though. 'last_events' will soon reflect
8041 * the state where there is little or no outstanding
8042 * resync requests, and further resync activity will
8043 * always make curr_events less than last_events.
8046 if (init || curr_events - rdev->last_events > 64) {
8047 rdev->last_events = curr_events;
8055 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8057 /* another "blocks" (512byte) blocks have been synced */
8058 atomic_sub(blocks, &mddev->recovery_active);
8059 wake_up(&mddev->recovery_wait);
8061 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8062 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8063 md_wakeup_thread(mddev->thread);
8064 // stop recovery, signal do_sync ....
8067 EXPORT_SYMBOL(md_done_sync);
8069 /* md_write_start(mddev, bi)
8070 * If we need to update some array metadata (e.g. 'active' flag
8071 * in superblock) before writing, schedule a superblock update
8072 * and wait for it to complete.
8073 * A return value of 'false' means that the write wasn't recorded
8074 * and cannot proceed as the array is being suspend.
8076 bool md_write_start(struct mddev *mddev, struct bio *bi)
8080 if (bio_data_dir(bi) != WRITE)
8083 BUG_ON(mddev->ro == 1);
8084 if (mddev->ro == 2) {
8085 /* need to switch to read/write */
8087 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8088 md_wakeup_thread(mddev->thread);
8089 md_wakeup_thread(mddev->sync_thread);
8093 percpu_ref_get(&mddev->writes_pending);
8094 smp_mb(); /* Match smp_mb in set_in_sync() */
8095 if (mddev->safemode == 1)
8096 mddev->safemode = 0;
8097 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8098 if (mddev->in_sync || mddev->sync_checkers) {
8099 spin_lock(&mddev->lock);
8100 if (mddev->in_sync) {
8102 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8103 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8104 md_wakeup_thread(mddev->thread);
8107 spin_unlock(&mddev->lock);
8111 sysfs_notify_dirent_safe(mddev->sysfs_state);
8112 if (!mddev->has_superblocks)
8114 wait_event(mddev->sb_wait,
8115 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8117 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8118 percpu_ref_put(&mddev->writes_pending);
8123 EXPORT_SYMBOL(md_write_start);
8125 /* md_write_inc can only be called when md_write_start() has
8126 * already been called at least once of the current request.
8127 * It increments the counter and is useful when a single request
8128 * is split into several parts. Each part causes an increment and
8129 * so needs a matching md_write_end().
8130 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8131 * a spinlocked region.
8133 void md_write_inc(struct mddev *mddev, struct bio *bi)
8135 if (bio_data_dir(bi) != WRITE)
8137 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8138 percpu_ref_get(&mddev->writes_pending);
8140 EXPORT_SYMBOL(md_write_inc);
8142 void md_write_end(struct mddev *mddev)
8144 percpu_ref_put(&mddev->writes_pending);
8146 if (mddev->safemode == 2)
8147 md_wakeup_thread(mddev->thread);
8148 else if (mddev->safemode_delay)
8149 /* The roundup() ensures this only performs locking once
8150 * every ->safemode_delay jiffies
8152 mod_timer(&mddev->safemode_timer,
8153 roundup(jiffies, mddev->safemode_delay) +
8154 mddev->safemode_delay);
8157 EXPORT_SYMBOL(md_write_end);
8159 /* md_allow_write(mddev)
8160 * Calling this ensures that the array is marked 'active' so that writes
8161 * may proceed without blocking. It is important to call this before
8162 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8163 * Must be called with mddev_lock held.
8165 void md_allow_write(struct mddev *mddev)
8171 if (!mddev->pers->sync_request)
8174 spin_lock(&mddev->lock);
8175 if (mddev->in_sync) {
8177 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8178 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8179 if (mddev->safemode_delay &&
8180 mddev->safemode == 0)
8181 mddev->safemode = 1;
8182 spin_unlock(&mddev->lock);
8183 md_update_sb(mddev, 0);
8184 sysfs_notify_dirent_safe(mddev->sysfs_state);
8185 /* wait for the dirty state to be recorded in the metadata */
8186 wait_event(mddev->sb_wait,
8187 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8189 spin_unlock(&mddev->lock);
8191 EXPORT_SYMBOL_GPL(md_allow_write);
8193 #define SYNC_MARKS 10
8194 #define SYNC_MARK_STEP (3*HZ)
8195 #define UPDATE_FREQUENCY (5*60*HZ)
8196 void md_do_sync(struct md_thread *thread)
8198 struct mddev *mddev = thread->mddev;
8199 struct mddev *mddev2;
8200 unsigned int currspeed = 0,
8202 sector_t max_sectors,j, io_sectors, recovery_done;
8203 unsigned long mark[SYNC_MARKS];
8204 unsigned long update_time;
8205 sector_t mark_cnt[SYNC_MARKS];
8207 struct list_head *tmp;
8208 sector_t last_check;
8210 struct md_rdev *rdev;
8211 char *desc, *action = NULL;
8212 struct blk_plug plug;
8215 /* just incase thread restarts... */
8216 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8217 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8219 if (mddev->ro) {/* never try to sync a read-only array */
8220 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8224 if (mddev_is_clustered(mddev)) {
8225 ret = md_cluster_ops->resync_start(mddev);
8229 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8230 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8231 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8232 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8233 && ((unsigned long long)mddev->curr_resync_completed
8234 < (unsigned long long)mddev->resync_max_sectors))
8238 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8239 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8240 desc = "data-check";
8242 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8243 desc = "requested-resync";
8247 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8252 mddev->last_sync_action = action ?: desc;
8254 /* we overload curr_resync somewhat here.
8255 * 0 == not engaged in resync at all
8256 * 2 == checking that there is no conflict with another sync
8257 * 1 == like 2, but have yielded to allow conflicting resync to
8259 * other == active in resync - this many blocks
8261 * Before starting a resync we must have set curr_resync to
8262 * 2, and then checked that every "conflicting" array has curr_resync
8263 * less than ours. When we find one that is the same or higher
8264 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8265 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8266 * This will mean we have to start checking from the beginning again.
8271 int mddev2_minor = -1;
8272 mddev->curr_resync = 2;
8275 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8277 for_each_mddev(mddev2, tmp) {
8278 if (mddev2 == mddev)
8280 if (!mddev->parallel_resync
8281 && mddev2->curr_resync
8282 && match_mddev_units(mddev, mddev2)) {
8284 if (mddev < mddev2 && mddev->curr_resync == 2) {
8285 /* arbitrarily yield */
8286 mddev->curr_resync = 1;
8287 wake_up(&resync_wait);
8289 if (mddev > mddev2 && mddev->curr_resync == 1)
8290 /* no need to wait here, we can wait the next
8291 * time 'round when curr_resync == 2
8294 /* We need to wait 'interruptible' so as not to
8295 * contribute to the load average, and not to
8296 * be caught by 'softlockup'
8298 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8299 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8300 mddev2->curr_resync >= mddev->curr_resync) {
8301 if (mddev2_minor != mddev2->md_minor) {
8302 mddev2_minor = mddev2->md_minor;
8303 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8304 desc, mdname(mddev),
8308 if (signal_pending(current))
8309 flush_signals(current);
8311 finish_wait(&resync_wait, &wq);
8314 finish_wait(&resync_wait, &wq);
8317 } while (mddev->curr_resync < 2);
8320 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8321 /* resync follows the size requested by the personality,
8322 * which defaults to physical size, but can be virtual size
8324 max_sectors = mddev->resync_max_sectors;
8325 atomic64_set(&mddev->resync_mismatches, 0);
8326 /* we don't use the checkpoint if there's a bitmap */
8327 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8328 j = mddev->resync_min;
8329 else if (!mddev->bitmap)
8330 j = mddev->recovery_cp;
8332 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8333 max_sectors = mddev->resync_max_sectors;
8335 * If the original node aborts reshaping then we continue the
8336 * reshaping, so set j again to avoid restart reshape from the
8339 if (mddev_is_clustered(mddev) &&
8340 mddev->reshape_position != MaxSector)
8341 j = mddev->reshape_position;
8343 /* recovery follows the physical size of devices */
8344 max_sectors = mddev->dev_sectors;
8347 rdev_for_each_rcu(rdev, mddev)
8348 if (rdev->raid_disk >= 0 &&
8349 !test_bit(Journal, &rdev->flags) &&
8350 !test_bit(Faulty, &rdev->flags) &&
8351 !test_bit(In_sync, &rdev->flags) &&
8352 rdev->recovery_offset < j)
8353 j = rdev->recovery_offset;
8356 /* If there is a bitmap, we need to make sure all
8357 * writes that started before we added a spare
8358 * complete before we start doing a recovery.
8359 * Otherwise the write might complete and (via
8360 * bitmap_endwrite) set a bit in the bitmap after the
8361 * recovery has checked that bit and skipped that
8364 if (mddev->bitmap) {
8365 mddev->pers->quiesce(mddev, 1);
8366 mddev->pers->quiesce(mddev, 0);
8370 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8371 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8372 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8373 speed_max(mddev), desc);
8375 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8378 for (m = 0; m < SYNC_MARKS; m++) {
8380 mark_cnt[m] = io_sectors;
8383 mddev->resync_mark = mark[last_mark];
8384 mddev->resync_mark_cnt = mark_cnt[last_mark];
8387 * Tune reconstruction:
8389 window = 32*(PAGE_SIZE/512);
8390 pr_debug("md: using %dk window, over a total of %lluk.\n",
8391 window/2, (unsigned long long)max_sectors/2);
8393 atomic_set(&mddev->recovery_active, 0);
8397 pr_debug("md: resuming %s of %s from checkpoint.\n",
8398 desc, mdname(mddev));
8399 mddev->curr_resync = j;
8401 mddev->curr_resync = 3; /* no longer delayed */
8402 mddev->curr_resync_completed = j;
8403 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8404 md_new_event(mddev);
8405 update_time = jiffies;
8407 blk_start_plug(&plug);
8408 while (j < max_sectors) {
8413 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8414 ((mddev->curr_resync > mddev->curr_resync_completed &&
8415 (mddev->curr_resync - mddev->curr_resync_completed)
8416 > (max_sectors >> 4)) ||
8417 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8418 (j - mddev->curr_resync_completed)*2
8419 >= mddev->resync_max - mddev->curr_resync_completed ||
8420 mddev->curr_resync_completed > mddev->resync_max
8422 /* time to update curr_resync_completed */
8423 wait_event(mddev->recovery_wait,
8424 atomic_read(&mddev->recovery_active) == 0);
8425 mddev->curr_resync_completed = j;
8426 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8427 j > mddev->recovery_cp)
8428 mddev->recovery_cp = j;
8429 update_time = jiffies;
8430 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8431 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8434 while (j >= mddev->resync_max &&
8435 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8436 /* As this condition is controlled by user-space,
8437 * we can block indefinitely, so use '_interruptible'
8438 * to avoid triggering warnings.
8440 flush_signals(current); /* just in case */
8441 wait_event_interruptible(mddev->recovery_wait,
8442 mddev->resync_max > j
8443 || test_bit(MD_RECOVERY_INTR,
8447 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8450 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8452 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8456 if (!skipped) { /* actual IO requested */
8457 io_sectors += sectors;
8458 atomic_add(sectors, &mddev->recovery_active);
8461 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8465 if (j > max_sectors)
8466 /* when skipping, extra large numbers can be returned. */
8469 mddev->curr_resync = j;
8470 mddev->curr_mark_cnt = io_sectors;
8471 if (last_check == 0)
8472 /* this is the earliest that rebuild will be
8473 * visible in /proc/mdstat
8475 md_new_event(mddev);
8477 if (last_check + window > io_sectors || j == max_sectors)
8480 last_check = io_sectors;
8482 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8484 int next = (last_mark+1) % SYNC_MARKS;
8486 mddev->resync_mark = mark[next];
8487 mddev->resync_mark_cnt = mark_cnt[next];
8488 mark[next] = jiffies;
8489 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8493 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8497 * this loop exits only if either when we are slower than
8498 * the 'hard' speed limit, or the system was IO-idle for
8500 * the system might be non-idle CPU-wise, but we only care
8501 * about not overloading the IO subsystem. (things like an
8502 * e2fsck being done on the RAID array should execute fast)
8506 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8507 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8508 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8510 if (currspeed > speed_min(mddev)) {
8511 if (currspeed > speed_max(mddev)) {
8515 if (!is_mddev_idle(mddev, 0)) {
8517 * Give other IO more of a chance.
8518 * The faster the devices, the less we wait.
8520 wait_event(mddev->recovery_wait,
8521 !atomic_read(&mddev->recovery_active));
8525 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8526 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8527 ? "interrupted" : "done");
8529 * this also signals 'finished resyncing' to md_stop
8531 blk_finish_plug(&plug);
8532 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8534 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8535 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8536 mddev->curr_resync > 3) {
8537 mddev->curr_resync_completed = mddev->curr_resync;
8538 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8540 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8542 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8543 mddev->curr_resync > 3) {
8544 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8545 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8546 if (mddev->curr_resync >= mddev->recovery_cp) {
8547 pr_debug("md: checkpointing %s of %s.\n",
8548 desc, mdname(mddev));
8549 if (test_bit(MD_RECOVERY_ERROR,
8551 mddev->recovery_cp =
8552 mddev->curr_resync_completed;
8554 mddev->recovery_cp =
8558 mddev->recovery_cp = MaxSector;
8560 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8561 mddev->curr_resync = MaxSector;
8562 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8563 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8565 rdev_for_each_rcu(rdev, mddev)
8566 if (rdev->raid_disk >= 0 &&
8567 mddev->delta_disks >= 0 &&
8568 !test_bit(Journal, &rdev->flags) &&
8569 !test_bit(Faulty, &rdev->flags) &&
8570 !test_bit(In_sync, &rdev->flags) &&
8571 rdev->recovery_offset < mddev->curr_resync)
8572 rdev->recovery_offset = mddev->curr_resync;
8578 /* set CHANGE_PENDING here since maybe another update is needed,
8579 * so other nodes are informed. It should be harmless for normal
8581 set_mask_bits(&mddev->sb_flags, 0,
8582 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8584 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8585 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8586 mddev->delta_disks > 0 &&
8587 mddev->pers->finish_reshape &&
8588 mddev->pers->size &&
8590 mddev_lock_nointr(mddev);
8591 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8592 mddev_unlock(mddev);
8593 if (!mddev_is_clustered(mddev)) {
8594 set_capacity(mddev->gendisk, mddev->array_sectors);
8595 revalidate_disk(mddev->gendisk);
8599 spin_lock(&mddev->lock);
8600 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8601 /* We completed so min/max setting can be forgotten if used. */
8602 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8603 mddev->resync_min = 0;
8604 mddev->resync_max = MaxSector;
8605 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8606 mddev->resync_min = mddev->curr_resync_completed;
8607 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8608 mddev->curr_resync = 0;
8609 spin_unlock(&mddev->lock);
8611 wake_up(&resync_wait);
8612 md_wakeup_thread(mddev->thread);
8615 EXPORT_SYMBOL_GPL(md_do_sync);
8617 static int remove_and_add_spares(struct mddev *mddev,
8618 struct md_rdev *this)
8620 struct md_rdev *rdev;
8623 bool remove_some = false;
8625 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8626 /* Mustn't remove devices when resync thread is running */
8629 rdev_for_each(rdev, mddev) {
8630 if ((this == NULL || rdev == this) &&
8631 rdev->raid_disk >= 0 &&
8632 !test_bit(Blocked, &rdev->flags) &&
8633 test_bit(Faulty, &rdev->flags) &&
8634 atomic_read(&rdev->nr_pending)==0) {
8635 /* Faulty non-Blocked devices with nr_pending == 0
8636 * never get nr_pending incremented,
8637 * never get Faulty cleared, and never get Blocked set.
8638 * So we can synchronize_rcu now rather than once per device
8641 set_bit(RemoveSynchronized, &rdev->flags);
8647 rdev_for_each(rdev, mddev) {
8648 if ((this == NULL || rdev == this) &&
8649 rdev->raid_disk >= 0 &&
8650 !test_bit(Blocked, &rdev->flags) &&
8651 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8652 (!test_bit(In_sync, &rdev->flags) &&
8653 !test_bit(Journal, &rdev->flags))) &&
8654 atomic_read(&rdev->nr_pending)==0)) {
8655 if (mddev->pers->hot_remove_disk(
8656 mddev, rdev) == 0) {
8657 sysfs_unlink_rdev(mddev, rdev);
8658 rdev->saved_raid_disk = rdev->raid_disk;
8659 rdev->raid_disk = -1;
8663 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8664 clear_bit(RemoveSynchronized, &rdev->flags);
8667 if (removed && mddev->kobj.sd)
8668 sysfs_notify(&mddev->kobj, NULL, "degraded");
8670 if (this && removed)
8673 rdev_for_each(rdev, mddev) {
8674 if (this && this != rdev)
8676 if (test_bit(Candidate, &rdev->flags))
8678 if (rdev->raid_disk >= 0 &&
8679 !test_bit(In_sync, &rdev->flags) &&
8680 !test_bit(Journal, &rdev->flags) &&
8681 !test_bit(Faulty, &rdev->flags))
8683 if (rdev->raid_disk >= 0)
8685 if (test_bit(Faulty, &rdev->flags))
8687 if (!test_bit(Journal, &rdev->flags)) {
8689 ! (rdev->saved_raid_disk >= 0 &&
8690 !test_bit(Bitmap_sync, &rdev->flags)))
8693 rdev->recovery_offset = 0;
8696 hot_add_disk(mddev, rdev) == 0) {
8697 if (sysfs_link_rdev(mddev, rdev))
8698 /* failure here is OK */;
8699 if (!test_bit(Journal, &rdev->flags))
8701 md_new_event(mddev);
8702 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8707 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8711 static void md_start_sync(struct work_struct *ws)
8713 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8715 mddev->sync_thread = md_register_thread(md_do_sync,
8718 if (!mddev->sync_thread) {
8719 pr_warn("%s: could not start resync thread...\n",
8721 /* leave the spares where they are, it shouldn't hurt */
8722 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8723 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8724 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8725 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8726 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8727 wake_up(&resync_wait);
8728 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8730 if (mddev->sysfs_action)
8731 sysfs_notify_dirent_safe(mddev->sysfs_action);
8733 md_wakeup_thread(mddev->sync_thread);
8734 sysfs_notify_dirent_safe(mddev->sysfs_action);
8735 md_new_event(mddev);
8739 * This routine is regularly called by all per-raid-array threads to
8740 * deal with generic issues like resync and super-block update.
8741 * Raid personalities that don't have a thread (linear/raid0) do not
8742 * need this as they never do any recovery or update the superblock.
8744 * It does not do any resync itself, but rather "forks" off other threads
8745 * to do that as needed.
8746 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8747 * "->recovery" and create a thread at ->sync_thread.
8748 * When the thread finishes it sets MD_RECOVERY_DONE
8749 * and wakeups up this thread which will reap the thread and finish up.
8750 * This thread also removes any faulty devices (with nr_pending == 0).
8752 * The overall approach is:
8753 * 1/ if the superblock needs updating, update it.
8754 * 2/ If a recovery thread is running, don't do anything else.
8755 * 3/ If recovery has finished, clean up, possibly marking spares active.
8756 * 4/ If there are any faulty devices, remove them.
8757 * 5/ If array is degraded, try to add spares devices
8758 * 6/ If array has spares or is not in-sync, start a resync thread.
8760 void md_check_recovery(struct mddev *mddev)
8762 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8763 /* Write superblock - thread that called mddev_suspend()
8764 * holds reconfig_mutex for us.
8766 set_bit(MD_UPDATING_SB, &mddev->flags);
8767 smp_mb__after_atomic();
8768 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8769 md_update_sb(mddev, 0);
8770 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8771 wake_up(&mddev->sb_wait);
8774 if (mddev->suspended)
8778 md_bitmap_daemon_work(mddev);
8780 if (signal_pending(current)) {
8781 if (mddev->pers->sync_request && !mddev->external) {
8782 pr_debug("md: %s in immediate safe mode\n",
8784 mddev->safemode = 2;
8786 flush_signals(current);
8789 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8792 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8793 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8794 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8795 (mddev->external == 0 && mddev->safemode == 1) ||
8796 (mddev->safemode == 2
8797 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8801 if (mddev_trylock(mddev)) {
8804 if (!mddev->external && mddev->safemode == 1)
8805 mddev->safemode = 0;
8808 struct md_rdev *rdev;
8809 if (!mddev->external && mddev->in_sync)
8810 /* 'Blocked' flag not needed as failed devices
8811 * will be recorded if array switched to read/write.
8812 * Leaving it set will prevent the device
8813 * from being removed.
8815 rdev_for_each(rdev, mddev)
8816 clear_bit(Blocked, &rdev->flags);
8817 /* On a read-only array we can:
8818 * - remove failed devices
8819 * - add already-in_sync devices if the array itself
8821 * As we only add devices that are already in-sync,
8822 * we can activate the spares immediately.
8824 remove_and_add_spares(mddev, NULL);
8825 /* There is no thread, but we need to call
8826 * ->spare_active and clear saved_raid_disk
8828 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8829 md_reap_sync_thread(mddev);
8830 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8831 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8832 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8836 if (mddev_is_clustered(mddev)) {
8837 struct md_rdev *rdev;
8838 /* kick the device if another node issued a
8841 rdev_for_each(rdev, mddev) {
8842 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8843 rdev->raid_disk < 0)
8844 md_kick_rdev_from_array(rdev);
8848 if (!mddev->external && !mddev->in_sync) {
8849 spin_lock(&mddev->lock);
8851 spin_unlock(&mddev->lock);
8854 if (mddev->sb_flags)
8855 md_update_sb(mddev, 0);
8857 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8858 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8859 /* resync/recovery still happening */
8860 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8863 if (mddev->sync_thread) {
8864 md_reap_sync_thread(mddev);
8867 /* Set RUNNING before clearing NEEDED to avoid
8868 * any transients in the value of "sync_action".
8870 mddev->curr_resync_completed = 0;
8871 spin_lock(&mddev->lock);
8872 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8873 spin_unlock(&mddev->lock);
8874 /* Clear some bits that don't mean anything, but
8877 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8878 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8880 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8881 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8883 /* no recovery is running.
8884 * remove any failed drives, then
8885 * add spares if possible.
8886 * Spares are also removed and re-added, to allow
8887 * the personality to fail the re-add.
8890 if (mddev->reshape_position != MaxSector) {
8891 if (mddev->pers->check_reshape == NULL ||
8892 mddev->pers->check_reshape(mddev) != 0)
8893 /* Cannot proceed */
8895 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8896 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8897 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8898 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8899 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8900 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8901 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8902 } else if (mddev->recovery_cp < MaxSector) {
8903 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8904 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8905 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8906 /* nothing to be done ... */
8909 if (mddev->pers->sync_request) {
8911 /* We are adding a device or devices to an array
8912 * which has the bitmap stored on all devices.
8913 * So make sure all bitmap pages get written
8915 md_bitmap_write_all(mddev->bitmap);
8917 INIT_WORK(&mddev->del_work, md_start_sync);
8918 queue_work(md_misc_wq, &mddev->del_work);
8922 if (!mddev->sync_thread) {
8923 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8924 wake_up(&resync_wait);
8925 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8927 if (mddev->sysfs_action)
8928 sysfs_notify_dirent_safe(mddev->sysfs_action);
8931 wake_up(&mddev->sb_wait);
8932 mddev_unlock(mddev);
8935 EXPORT_SYMBOL(md_check_recovery);
8937 void md_reap_sync_thread(struct mddev *mddev)
8939 struct md_rdev *rdev;
8940 sector_t old_dev_sectors = mddev->dev_sectors;
8941 bool is_reshaped = false;
8943 /* resync has finished, collect result */
8944 md_unregister_thread(&mddev->sync_thread);
8945 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8946 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8948 /* activate any spares */
8949 if (mddev->pers->spare_active(mddev)) {
8950 sysfs_notify(&mddev->kobj, NULL,
8952 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8955 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8956 mddev->pers->finish_reshape) {
8957 mddev->pers->finish_reshape(mddev);
8958 if (mddev_is_clustered(mddev))
8962 /* If array is no-longer degraded, then any saved_raid_disk
8963 * information must be scrapped.
8965 if (!mddev->degraded)
8966 rdev_for_each(rdev, mddev)
8967 rdev->saved_raid_disk = -1;
8969 md_update_sb(mddev, 1);
8970 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8971 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8973 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8974 md_cluster_ops->resync_finish(mddev);
8975 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8976 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8977 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8978 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8979 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8980 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8982 * We call md_cluster_ops->update_size here because sync_size could
8983 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
8984 * so it is time to update size across cluster.
8986 if (mddev_is_clustered(mddev) && is_reshaped
8987 && !test_bit(MD_CLOSING, &mddev->flags))
8988 md_cluster_ops->update_size(mddev, old_dev_sectors);
8989 wake_up(&resync_wait);
8990 /* flag recovery needed just to double check */
8991 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8992 sysfs_notify_dirent_safe(mddev->sysfs_action);
8993 md_new_event(mddev);
8994 if (mddev->event_work.func)
8995 queue_work(md_misc_wq, &mddev->event_work);
8997 EXPORT_SYMBOL(md_reap_sync_thread);
8999 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9001 sysfs_notify_dirent_safe(rdev->sysfs_state);
9002 wait_event_timeout(rdev->blocked_wait,
9003 !test_bit(Blocked, &rdev->flags) &&
9004 !test_bit(BlockedBadBlocks, &rdev->flags),
9005 msecs_to_jiffies(5000));
9006 rdev_dec_pending(rdev, mddev);
9008 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9010 void md_finish_reshape(struct mddev *mddev)
9012 /* called be personality module when reshape completes. */
9013 struct md_rdev *rdev;
9015 rdev_for_each(rdev, mddev) {
9016 if (rdev->data_offset > rdev->new_data_offset)
9017 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9019 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9020 rdev->data_offset = rdev->new_data_offset;
9023 EXPORT_SYMBOL(md_finish_reshape);
9025 /* Bad block management */
9027 /* Returns 1 on success, 0 on failure */
9028 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9031 struct mddev *mddev = rdev->mddev;
9034 s += rdev->new_data_offset;
9036 s += rdev->data_offset;
9037 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9039 /* Make sure they get written out promptly */
9040 if (test_bit(ExternalBbl, &rdev->flags))
9041 sysfs_notify(&rdev->kobj, NULL,
9042 "unacknowledged_bad_blocks");
9043 sysfs_notify_dirent_safe(rdev->sysfs_state);
9044 set_mask_bits(&mddev->sb_flags, 0,
9045 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9046 md_wakeup_thread(rdev->mddev->thread);
9051 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9053 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9058 s += rdev->new_data_offset;
9060 s += rdev->data_offset;
9061 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9062 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9063 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9066 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9068 static int md_notify_reboot(struct notifier_block *this,
9069 unsigned long code, void *x)
9071 struct list_head *tmp;
9072 struct mddev *mddev;
9075 for_each_mddev(mddev, tmp) {
9076 if (mddev_trylock(mddev)) {
9078 __md_stop_writes(mddev);
9079 if (mddev->persistent)
9080 mddev->safemode = 2;
9081 mddev_unlock(mddev);
9086 * certain more exotic SCSI devices are known to be
9087 * volatile wrt too early system reboots. While the
9088 * right place to handle this issue is the given
9089 * driver, we do want to have a safe RAID driver ...
9097 static struct notifier_block md_notifier = {
9098 .notifier_call = md_notify_reboot,
9100 .priority = INT_MAX, /* before any real devices */
9103 static void md_geninit(void)
9105 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9107 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9110 static int __init md_init(void)
9114 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9118 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9122 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9125 if ((ret = register_blkdev(0, "mdp")) < 0)
9129 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9130 md_probe, NULL, NULL);
9131 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9132 md_probe, NULL, NULL);
9134 register_reboot_notifier(&md_notifier);
9135 raid_table_header = register_sysctl_table(raid_root_table);
9141 unregister_blkdev(MD_MAJOR, "md");
9143 destroy_workqueue(md_misc_wq);
9145 destroy_workqueue(md_wq);
9150 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9152 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9153 struct md_rdev *rdev2;
9155 char b[BDEVNAME_SIZE];
9158 * If size is changed in another node then we need to
9159 * do resize as well.
9161 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9162 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9164 pr_info("md-cluster: resize failed\n");
9166 md_bitmap_update_sb(mddev->bitmap);
9169 /* Check for change of roles in the active devices */
9170 rdev_for_each(rdev2, mddev) {
9171 if (test_bit(Faulty, &rdev2->flags))
9174 /* Check if the roles changed */
9175 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9177 if (test_bit(Candidate, &rdev2->flags)) {
9178 if (role == 0xfffe) {
9179 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9180 md_kick_rdev_from_array(rdev2);
9184 clear_bit(Candidate, &rdev2->flags);
9187 if (role != rdev2->raid_disk) {
9189 * got activated except reshape is happening.
9191 if (rdev2->raid_disk == -1 && role != 0xffff &&
9192 !(le32_to_cpu(sb->feature_map) &
9193 MD_FEATURE_RESHAPE_ACTIVE)) {
9194 rdev2->saved_raid_disk = role;
9195 ret = remove_and_add_spares(mddev, rdev2);
9196 pr_info("Activated spare: %s\n",
9197 bdevname(rdev2->bdev,b));
9198 /* wakeup mddev->thread here, so array could
9199 * perform resync with the new activated disk */
9200 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9201 md_wakeup_thread(mddev->thread);
9205 * We just want to do the minimum to mark the disk
9206 * as faulty. The recovery is performed by the
9207 * one who initiated the error.
9209 if ((role == 0xfffe) || (role == 0xfffd)) {
9210 md_error(mddev, rdev2);
9211 clear_bit(Blocked, &rdev2->flags);
9216 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9217 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9220 * Since mddev->delta_disks has already updated in update_raid_disks,
9221 * so it is time to check reshape.
9223 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9224 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9226 * reshape is happening in the remote node, we need to
9227 * update reshape_position and call start_reshape.
9229 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9230 if (mddev->pers->update_reshape_pos)
9231 mddev->pers->update_reshape_pos(mddev);
9232 if (mddev->pers->start_reshape)
9233 mddev->pers->start_reshape(mddev);
9234 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9235 mddev->reshape_position != MaxSector &&
9236 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9237 /* reshape is just done in another node. */
9238 mddev->reshape_position = MaxSector;
9239 if (mddev->pers->update_reshape_pos)
9240 mddev->pers->update_reshape_pos(mddev);
9243 /* Finally set the event to be up to date */
9244 mddev->events = le64_to_cpu(sb->events);
9247 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9250 struct page *swapout = rdev->sb_page;
9251 struct mdp_superblock_1 *sb;
9253 /* Store the sb page of the rdev in the swapout temporary
9254 * variable in case we err in the future
9256 rdev->sb_page = NULL;
9257 err = alloc_disk_sb(rdev);
9259 ClearPageUptodate(rdev->sb_page);
9260 rdev->sb_loaded = 0;
9261 err = super_types[mddev->major_version].
9262 load_super(rdev, NULL, mddev->minor_version);
9265 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9266 __func__, __LINE__, rdev->desc_nr, err);
9268 put_page(rdev->sb_page);
9269 rdev->sb_page = swapout;
9270 rdev->sb_loaded = 1;
9274 sb = page_address(rdev->sb_page);
9275 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9279 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9280 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9282 /* The other node finished recovery, call spare_active to set
9283 * device In_sync and mddev->degraded
9285 if (rdev->recovery_offset == MaxSector &&
9286 !test_bit(In_sync, &rdev->flags) &&
9287 mddev->pers->spare_active(mddev))
9288 sysfs_notify(&mddev->kobj, NULL, "degraded");
9294 void md_reload_sb(struct mddev *mddev, int nr)
9296 struct md_rdev *rdev;
9300 rdev_for_each_rcu(rdev, mddev) {
9301 if (rdev->desc_nr == nr)
9305 if (!rdev || rdev->desc_nr != nr) {
9306 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9310 err = read_rdev(mddev, rdev);
9314 check_sb_changes(mddev, rdev);
9316 /* Read all rdev's to update recovery_offset */
9317 rdev_for_each_rcu(rdev, mddev) {
9318 if (!test_bit(Faulty, &rdev->flags))
9319 read_rdev(mddev, rdev);
9322 EXPORT_SYMBOL(md_reload_sb);
9327 * Searches all registered partitions for autorun RAID arrays
9331 static DEFINE_MUTEX(detected_devices_mutex);
9332 static LIST_HEAD(all_detected_devices);
9333 struct detected_devices_node {
9334 struct list_head list;
9338 void md_autodetect_dev(dev_t dev)
9340 struct detected_devices_node *node_detected_dev;
9342 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9343 if (node_detected_dev) {
9344 node_detected_dev->dev = dev;
9345 mutex_lock(&detected_devices_mutex);
9346 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9347 mutex_unlock(&detected_devices_mutex);
9351 static void autostart_arrays(int part)
9353 struct md_rdev *rdev;
9354 struct detected_devices_node *node_detected_dev;
9356 int i_scanned, i_passed;
9361 pr_info("md: Autodetecting RAID arrays.\n");
9363 mutex_lock(&detected_devices_mutex);
9364 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9366 node_detected_dev = list_entry(all_detected_devices.next,
9367 struct detected_devices_node, list);
9368 list_del(&node_detected_dev->list);
9369 dev = node_detected_dev->dev;
9370 kfree(node_detected_dev);
9371 mutex_unlock(&detected_devices_mutex);
9372 rdev = md_import_device(dev,0, 90);
9373 mutex_lock(&detected_devices_mutex);
9377 if (test_bit(Faulty, &rdev->flags))
9380 set_bit(AutoDetected, &rdev->flags);
9381 list_add(&rdev->same_set, &pending_raid_disks);
9384 mutex_unlock(&detected_devices_mutex);
9386 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9388 autorun_devices(part);
9391 #endif /* !MODULE */
9393 static __exit void md_exit(void)
9395 struct mddev *mddev;
9396 struct list_head *tmp;
9399 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9400 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9402 unregister_blkdev(MD_MAJOR,"md");
9403 unregister_blkdev(mdp_major, "mdp");
9404 unregister_reboot_notifier(&md_notifier);
9405 unregister_sysctl_table(raid_table_header);
9407 /* We cannot unload the modules while some process is
9408 * waiting for us in select() or poll() - wake them up
9411 while (waitqueue_active(&md_event_waiters)) {
9412 /* not safe to leave yet */
9413 wake_up(&md_event_waiters);
9417 remove_proc_entry("mdstat", NULL);
9419 for_each_mddev(mddev, tmp) {
9420 export_array(mddev);
9422 mddev->hold_active = 0;
9424 * for_each_mddev() will call mddev_put() at the end of each
9425 * iteration. As the mddev is now fully clear, this will
9426 * schedule the mddev for destruction by a workqueue, and the
9427 * destroy_workqueue() below will wait for that to complete.
9430 destroy_workqueue(md_misc_wq);
9431 destroy_workqueue(md_wq);
9434 subsys_initcall(md_init);
9435 module_exit(md_exit)
9437 static int get_ro(char *buffer, const struct kernel_param *kp)
9439 return sprintf(buffer, "%d", start_readonly);
9441 static int set_ro(const char *val, const struct kernel_param *kp)
9443 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9446 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9447 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9448 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9449 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9451 MODULE_LICENSE("GPL");
9452 MODULE_DESCRIPTION("MD RAID framework");
9454 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);